linux/fs/afs/fsclient.c

2317 lines
54 KiB
C
Raw Normal View History

/* AFS File Server client stubs
*
* Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
* Written by David Howells (dhowells@redhat.com)
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#include <linux/init.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
#include <linux/slab.h>
#include <linux/sched.h>
#include <linux/circ_buf.h>
#include <linux/iversion.h>
#include "internal.h"
#include "afs_fs.h"
#include "xdr_fs.h"
static const struct afs_fid afs_zero_fid;
/*
* We need somewhere to discard into in case the server helpfully returns more
* than we asked for in FS.FetchData{,64}.
*/
static u8 afs_discard_buffer[64];
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
static inline void afs_use_fs_server(struct afs_call *call, struct afs_cb_interest *cbi)
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->cbi = afs_get_cb_interest(cbi);
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
}
/*
* decode an AFSFid block
*/
static void xdr_decode_AFSFid(const __be32 **_bp, struct afs_fid *fid)
{
const __be32 *bp = *_bp;
fid->vid = ntohl(*bp++);
fid->vnode = ntohl(*bp++);
fid->unique = ntohl(*bp++);
*_bp = bp;
}
/*
* Dump a bad file status record.
*/
static void xdr_dump_bad(const __be32 *bp)
{
__be32 x[4];
int i;
pr_notice("AFS XDR: Bad status record\n");
for (i = 0; i < 5 * 4 * 4; i += 16) {
memcpy(x, bp, 16);
bp += 4;
pr_notice("%03x: %08x %08x %08x %08x\n",
i, ntohl(x[0]), ntohl(x[1]), ntohl(x[2]), ntohl(x[3]));
}
memcpy(x, bp, 4);
pr_notice("0x50: %08x\n", ntohl(x[0]));
}
/*
* Update the core inode struct from a returned status record.
*/
void afs_update_inode_from_status(struct afs_vnode *vnode,
struct afs_file_status *status,
const afs_dataversion_t *expected_version,
u8 flags)
{
struct timespec t;
umode_t mode;
t.tv_sec = status->mtime_client;
t.tv_nsec = 0;
vnode->vfs_inode.i_ctime = t;
vnode->vfs_inode.i_mtime = t;
vnode->vfs_inode.i_atime = t;
if (flags & (AFS_VNODE_META_CHANGED | AFS_VNODE_NOT_YET_SET)) {
vnode->vfs_inode.i_uid = make_kuid(&init_user_ns, status->owner);
vnode->vfs_inode.i_gid = make_kgid(&init_user_ns, status->group);
set_nlink(&vnode->vfs_inode, status->nlink);
mode = vnode->vfs_inode.i_mode;
mode &= ~S_IALLUGO;
mode |= status->mode;
barrier();
vnode->vfs_inode.i_mode = mode;
}
if (!(flags & AFS_VNODE_NOT_YET_SET)) {
if (expected_version &&
*expected_version != status->data_version) {
_debug("vnode modified %llx on {%x:%u} [exp %llx]",
(unsigned long long) status->data_version,
vnode->fid.vid, vnode->fid.vnode,
(unsigned long long) *expected_version);
vnode->invalid_before = status->data_version;
afs: Fix directory handling AFS directories are structured blobs that are downloaded just like files and then parsed by the lookup and readdir code and, as such, are currently handled in the pagecache like any other file, with the entire directory content being thrown away each time the directory changes. However, since the blob is a known structure and since the data version counter on a directory increases by exactly one for each change committed to that directory, we can actually edit the directory locally rather than fetching it from the server after each locally-induced change. What we can't do, though, is mix data from the server and data from the client since the server is technically at liberty to rearrange or compress a directory if it sees fit, provided it updates the data version number when it does so and breaks the callback (ie. sends a notification). Further, lookup with lookup-ahead, readdir and, when it arrives, local editing are likely want to scan the whole of a directory. So directory handling needs to be improved to maintain the coherency of the directory blob prior to permitting local directory editing. To this end: (1) If any directory page gets discarded, invalidate and reread the entire directory. (2) If readpage notes that if when it fetches a single page that the version number has changed, the entire directory is flagged for invalidation. (3) Read as much of the directory in one go as we can. Note that this removes local caching of directories in fscache for the moment as we can't pass the pages to fscache_read_or_alloc_pages() since page->lru is in use by the LRU. Signed-off-by: David Howells <dhowells@redhat.com>
2018-04-06 13:17:25 +00:00
if (vnode->status.type == AFS_FTYPE_DIR) {
if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags))
afs_stat_v(vnode, n_inval);
} else {
set_bit(AFS_VNODE_ZAP_DATA, &vnode->flags);
}
}
}
if (flags & (AFS_VNODE_DATA_CHANGED | AFS_VNODE_NOT_YET_SET)) {
inode_set_iversion_raw(&vnode->vfs_inode, status->data_version);
i_size_write(&vnode->vfs_inode, status->size);
}
}
/*
* decode an AFSFetchStatus block
*/
static int xdr_decode_AFSFetchStatus(const __be32 **_bp,
struct afs_file_status *status,
struct afs_vnode *vnode,
const afs_dataversion_t *expected_version,
afs: Fix directory handling AFS directories are structured blobs that are downloaded just like files and then parsed by the lookup and readdir code and, as such, are currently handled in the pagecache like any other file, with the entire directory content being thrown away each time the directory changes. However, since the blob is a known structure and since the data version counter on a directory increases by exactly one for each change committed to that directory, we can actually edit the directory locally rather than fetching it from the server after each locally-induced change. What we can't do, though, is mix data from the server and data from the client since the server is technically at liberty to rearrange or compress a directory if it sees fit, provided it updates the data version number when it does so and breaks the callback (ie. sends a notification). Further, lookup with lookup-ahead, readdir and, when it arrives, local editing are likely want to scan the whole of a directory. So directory handling needs to be improved to maintain the coherency of the directory blob prior to permitting local directory editing. To this end: (1) If any directory page gets discarded, invalidate and reread the entire directory. (2) If readpage notes that if when it fetches a single page that the version number has changed, the entire directory is flagged for invalidation. (3) Read as much of the directory in one go as we can. Note that this removes local caching of directories in fscache for the moment as we can't pass the pages to fscache_read_or_alloc_pages() since page->lru is in use by the LRU. Signed-off-by: David Howells <dhowells@redhat.com>
2018-04-06 13:17:25 +00:00
struct afs_read *read_req)
{
const struct afs_xdr_AFSFetchStatus *xdr = (const void *)*_bp;
u64 data_version, size;
u32 type, abort_code;
u8 flags = 0;
int ret;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
if (vnode)
write_seqlock(&vnode->cb_lock);
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
if (xdr->if_version != htonl(AFS_FSTATUS_VERSION)) {
pr_warn("Unknown AFSFetchStatus version %u\n", ntohl(xdr->if_version));
goto bad;
}
type = ntohl(xdr->type);
abort_code = ntohl(xdr->abort_code);
switch (type) {
case AFS_FTYPE_FILE:
case AFS_FTYPE_DIR:
case AFS_FTYPE_SYMLINK:
if (type != status->type &&
vnode &&
!test_bit(AFS_VNODE_UNSET, &vnode->flags)) {
pr_warning("Vnode %x:%x:%x changed type %u to %u\n",
vnode->fid.vid,
vnode->fid.vnode,
vnode->fid.unique,
status->type, type);
goto bad;
}
status->type = type;
break;
case AFS_FTYPE_INVALID:
if (abort_code != 0) {
status->abort_code = abort_code;
goto out;
}
/* Fall through */
default:
goto bad;
}
#define EXTRACT_M(FIELD) \
do { \
u32 x = ntohl(xdr->FIELD); \
if (status->FIELD != x) { \
flags |= AFS_VNODE_META_CHANGED; \
status->FIELD = x; \
} \
} while (0)
EXTRACT_M(nlink);
EXTRACT_M(author);
EXTRACT_M(owner);
EXTRACT_M(caller_access); /* call ticket dependent */
EXTRACT_M(anon_access);
EXTRACT_M(mode);
EXTRACT_M(group);
status->mtime_client = ntohl(xdr->mtime_client);
status->mtime_server = ntohl(xdr->mtime_server);
status->lock_count = ntohl(xdr->lock_count);
size = (u64)ntohl(xdr->size_lo);
size |= (u64)ntohl(xdr->size_hi) << 32;
if (size != status->size) {
status->size = size;
flags |= AFS_VNODE_DATA_CHANGED;
}
data_version = (u64)ntohl(xdr->data_version_lo);
data_version |= (u64)ntohl(xdr->data_version_hi) << 32;
if (data_version != status->data_version) {
status->data_version = data_version;
flags |= AFS_VNODE_DATA_CHANGED;
}
afs: Fix directory handling AFS directories are structured blobs that are downloaded just like files and then parsed by the lookup and readdir code and, as such, are currently handled in the pagecache like any other file, with the entire directory content being thrown away each time the directory changes. However, since the blob is a known structure and since the data version counter on a directory increases by exactly one for each change committed to that directory, we can actually edit the directory locally rather than fetching it from the server after each locally-induced change. What we can't do, though, is mix data from the server and data from the client since the server is technically at liberty to rearrange or compress a directory if it sees fit, provided it updates the data version number when it does so and breaks the callback (ie. sends a notification). Further, lookup with lookup-ahead, readdir and, when it arrives, local editing are likely want to scan the whole of a directory. So directory handling needs to be improved to maintain the coherency of the directory blob prior to permitting local directory editing. To this end: (1) If any directory page gets discarded, invalidate and reread the entire directory. (2) If readpage notes that if when it fetches a single page that the version number has changed, the entire directory is flagged for invalidation. (3) Read as much of the directory in one go as we can. Note that this removes local caching of directories in fscache for the moment as we can't pass the pages to fscache_read_or_alloc_pages() since page->lru is in use by the LRU. Signed-off-by: David Howells <dhowells@redhat.com>
2018-04-06 13:17:25 +00:00
if (read_req) {
read_req->data_version = data_version;
read_req->file_size = size;
}
*_bp = (const void *)*_bp + sizeof(*xdr);
if (vnode) {
if (test_bit(AFS_VNODE_UNSET, &vnode->flags))
flags |= AFS_VNODE_NOT_YET_SET;
afs_update_inode_from_status(vnode, status, expected_version,
flags);
}
ret = 0;
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
out:
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
if (vnode)
write_sequnlock(&vnode->cb_lock);
return ret;
bad:
xdr_dump_bad(*_bp);
ret = -EBADMSG;
goto out;
}
/*
* decode an AFSCallBack block
*/
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
static void xdr_decode_AFSCallBack(struct afs_call *call,
struct afs_vnode *vnode,
const __be32 **_bp)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_cb_interest *old, *cbi = call->cbi;
const __be32 *bp = *_bp;
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
u32 cb_expiry;
write_seqlock(&vnode->cb_lock);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
if (call->cb_break == (vnode->cb_break + cbi->server->cb_s_break)) {
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
vnode->cb_version = ntohl(*bp++);
cb_expiry = ntohl(*bp++);
vnode->cb_type = ntohl(*bp++);
vnode->cb_expires_at = cb_expiry + ktime_get_real_seconds();
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
old = vnode->cb_interest;
if (old != call->cbi) {
vnode->cb_interest = cbi;
cbi = old;
}
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
} else {
bp += 3;
}
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
write_sequnlock(&vnode->cb_lock);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->cbi = cbi;
*_bp = bp;
}
static void xdr_decode_AFSCallBack_raw(const __be32 **_bp,
struct afs_callback *cb)
{
const __be32 *bp = *_bp;
cb->version = ntohl(*bp++);
cb->expiry = ntohl(*bp++);
cb->type = ntohl(*bp++);
*_bp = bp;
}
/*
* decode an AFSVolSync block
*/
static void xdr_decode_AFSVolSync(const __be32 **_bp,
struct afs_volsync *volsync)
{
const __be32 *bp = *_bp;
volsync->creation = ntohl(*bp++);
bp++; /* spare2 */
bp++; /* spare3 */
bp++; /* spare4 */
bp++; /* spare5 */
bp++; /* spare6 */
*_bp = bp;
}
/*
* encode the requested attributes into an AFSStoreStatus block
*/
static void xdr_encode_AFS_StoreStatus(__be32 **_bp, struct iattr *attr)
{
__be32 *bp = *_bp;
u32 mask = 0, mtime = 0, owner = 0, group = 0, mode = 0;
mask = 0;
if (attr->ia_valid & ATTR_MTIME) {
mask |= AFS_SET_MTIME;
mtime = attr->ia_mtime.tv_sec;
}
if (attr->ia_valid & ATTR_UID) {
mask |= AFS_SET_OWNER;
owner = from_kuid(&init_user_ns, attr->ia_uid);
}
if (attr->ia_valid & ATTR_GID) {
mask |= AFS_SET_GROUP;
group = from_kgid(&init_user_ns, attr->ia_gid);
}
if (attr->ia_valid & ATTR_MODE) {
mask |= AFS_SET_MODE;
mode = attr->ia_mode & S_IALLUGO;
}
*bp++ = htonl(mask);
*bp++ = htonl(mtime);
*bp++ = htonl(owner);
*bp++ = htonl(group);
*bp++ = htonl(mode);
*bp++ = 0; /* segment size */
*_bp = bp;
}
/*
* decode an AFSFetchVolumeStatus block
*/
static void xdr_decode_AFSFetchVolumeStatus(const __be32 **_bp,
struct afs_volume_status *vs)
{
const __be32 *bp = *_bp;
vs->vid = ntohl(*bp++);
vs->parent_id = ntohl(*bp++);
vs->online = ntohl(*bp++);
vs->in_service = ntohl(*bp++);
vs->blessed = ntohl(*bp++);
vs->needs_salvage = ntohl(*bp++);
vs->type = ntohl(*bp++);
vs->min_quota = ntohl(*bp++);
vs->max_quota = ntohl(*bp++);
vs->blocks_in_use = ntohl(*bp++);
vs->part_blocks_avail = ntohl(*bp++);
vs->part_max_blocks = ntohl(*bp++);
*_bp = bp;
}
/*
* deliver reply data to an FS.FetchStatus
*/
static int afs_deliver_fs_fetch_status_vnode(struct afs_call *call)
{
struct afs_vnode *vnode = call->reply[0];
const __be32 *bp;
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_transfer_reply(call);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
_enter("{%x:%u}", vnode->fid.vid, vnode->fid.vnode);
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
if (xdr_decode_AFSFetchStatus(&bp, &vnode->status, vnode,
&call->expected_version, NULL) < 0)
return -EBADMSG;
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
xdr_decode_AFSCallBack(call, vnode, &bp);
if (call->reply[1])
xdr_decode_AFSVolSync(&bp, call->reply[1]);
_leave(" = 0 [done]");
return 0;
}
/*
* FS.FetchStatus operation type
*/
static const struct afs_call_type afs_RXFSFetchStatus_vnode = {
.name = "FS.FetchStatus(vnode)",
.op = afs_FS_FetchStatus,
.deliver = afs_deliver_fs_fetch_status_vnode,
.destructor = afs_flat_call_destructor,
};
/*
* fetch the status information for a file
*/
int afs_fs_fetch_file_status(struct afs_fs_cursor *fc, struct afs_volsync *volsync,
bool new_inode)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
_enter(",%x,{%x:%u},,",
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
key_serial(fc->key), vnode->fid.vid, vnode->fid.vnode);
call = afs_alloc_flat_call(net, &afs_RXFSFetchStatus_vnode,
16, (21 + 3 + 6) * 4);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
if (!call) {
fc->ac.error = -ENOMEM;
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
}
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->reply[1] = volsync;
call->expected_version = new_inode ? 1 : vnode->status.data_version;
/* marshall the parameters */
bp = call->request;
bp[0] = htonl(FSFETCHSTATUS);
bp[1] = htonl(vnode->fid.vid);
bp[2] = htonl(vnode->fid.vnode);
bp[3] = htonl(vnode->fid.unique);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->cb_break = fc->cb_break;
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.FetchData
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_fetch_data(struct afs_call *call)
{
struct afs_vnode *vnode = call->reply[0];
struct afs_read *req = call->reply[2];
const __be32 *bp;
unsigned int size;
void *buffer;
int ret;
_enter("{%u,%zu/%u;%llu/%llu}",
call->unmarshall, call->offset, call->count,
req->remain, req->actual_len);
switch (call->unmarshall) {
case 0:
req->actual_len = 0;
call->offset = 0;
call->unmarshall++;
if (call->operation_ID != FSFETCHDATA64) {
call->unmarshall++;
goto no_msw;
}
/* extract the upper part of the returned data length of an
* FSFETCHDATA64 op (which should always be 0 using this
* client) */
case 1:
_debug("extract data length (MSW)");
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, &call->tmp, 4, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
req->actual_len = ntohl(call->tmp);
req->actual_len <<= 32;
call->offset = 0;
call->unmarshall++;
no_msw:
/* extract the returned data length */
case 2:
_debug("extract data length");
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, &call->tmp, 4, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
req->actual_len |= ntohl(call->tmp);
_debug("DATA length: %llu", req->actual_len);
req->remain = req->actual_len;
call->offset = req->pos & (PAGE_SIZE - 1);
req->index = 0;
if (req->actual_len == 0)
goto no_more_data;
call->unmarshall++;
begin_page:
ASSERTCMP(req->index, <, req->nr_pages);
if (req->remain > PAGE_SIZE - call->offset)
size = PAGE_SIZE - call->offset;
else
size = req->remain;
call->count = call->offset + size;
ASSERTCMP(call->count, <=, PAGE_SIZE);
req->remain -= size;
/* extract the returned data */
case 3:
_debug("extract data %llu/%llu %zu/%u",
req->remain, req->actual_len, call->offset, call->count);
buffer = kmap(req->pages[req->index]);
ret = afs_extract_data(call, buffer, call->count, true);
kunmap(req->pages[req->index]);
if (ret < 0)
return ret;
if (call->offset == PAGE_SIZE) {
if (req->page_done)
req->page_done(call, req);
afs: Fix AFS read bug Fix a bug in AFS read whereby the request page afs_read::index isn't incremented after calling ->page_done() if ->remain reaches 0, indicating that the data read is complete. Without this a NULL pointer exception happens when ->page_done() is called twice for the last page because the page clearing loop will call it also and afs_readpages_page_done() clears the current entry in the page list. BUG: unable to handle kernel NULL pointer dereference at (null) IP: afs_readpages_page_done+0x21/0xa4 [kafs] PGD 0 Oops: 0002 [#1] SMP Modules linked in: kafs(E) CPU: 2 PID: 3002 Comm: md5sum Tainted: G E 4.10.0-fscache #485 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 task: ffff8804017d86c0 task.stack: ffff8803fc1d8000 RIP: 0010:afs_readpages_page_done+0x21/0xa4 [kafs] RSP: 0018:ffff8803fc1db978 EFLAGS: 00010282 RAX: ffff880405d39af8 RBX: 0000000000000000 RCX: ffff880407d83ed4 RDX: 0000000000000000 RSI: ffff880405d39a00 RDI: ffff880405c6f400 RBP: ffff8803fc1db988 R08: 0000000000000000 R09: 0000000000000001 R10: ffff8803fc1db820 R11: ffff88040cf56000 R12: ffff8804088f1780 R13: ffff8804017d86c0 R14: ffff8804088f1780 R15: 0000000000003840 FS: 00007f8154469700(0000) GS:ffff88041fb00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000000000000000 CR3: 00000004016ec000 CR4: 00000000001406e0 Call Trace: afs_deliver_fs_fetch_data+0x5b9/0x60e [kafs] ? afs_make_call+0x316/0x4e8 [kafs] ? afs_make_call+0x359/0x4e8 [kafs] afs_deliver_to_call+0x173/0x2e8 [kafs] ? afs_make_call+0x316/0x4e8 [kafs] afs_make_call+0x37a/0x4e8 [kafs] ? wake_up_q+0x4f/0x4f ? __init_waitqueue_head+0x36/0x49 afs_fs_fetch_data+0x21c/0x227 [kafs] ? afs_fs_fetch_data+0x21c/0x227 [kafs] afs_vnode_fetch_data+0xf3/0x1d2 [kafs] afs_readpages+0x314/0x3fd [kafs] __do_page_cache_readahead+0x208/0x2c5 ondemand_readahead+0x3a2/0x3b7 ? ondemand_readahead+0x3a2/0x3b7 page_cache_async_readahead+0x5e/0x67 generic_file_read_iter+0x23b/0x70c ? __inode_security_revalidate+0x2f/0x62 __vfs_read+0xc4/0xe8 vfs_read+0xd1/0x15a SyS_read+0x4c/0x89 do_syscall_64+0x80/0x191 entry_SYSCALL64_slow_path+0x25/0x25 Reported-by: Marc Dionne <marc.dionne@auristor.com> Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Marc Dionne <marc.dionne@auristor.com>
2017-03-16 16:27:46 +00:00
req->index++;
if (req->remain > 0) {
call->offset = 0;
if (req->index >= req->nr_pages) {
call->unmarshall = 4;
goto begin_discard;
}
goto begin_page;
}
}
goto no_more_data;
/* Discard any excess data the server gave us */
begin_discard:
case 4:
size = min_t(loff_t, sizeof(afs_discard_buffer), req->remain);
call->count = size;
_debug("extract discard %llu/%llu %zu/%u",
req->remain, req->actual_len, call->offset, call->count);
call->offset = 0;
ret = afs_extract_data(call, afs_discard_buffer, call->count, true);
req->remain -= call->offset;
if (ret < 0)
return ret;
if (req->remain > 0)
goto begin_discard;
no_more_data:
call->offset = 0;
call->unmarshall = 5;
/* extract the metadata */
case 5:
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, call->buffer,
(21 + 3 + 6) * 4, false);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
bp = call->buffer;
if (xdr_decode_AFSFetchStatus(&bp, &vnode->status, vnode,
afs: Fix directory handling AFS directories are structured blobs that are downloaded just like files and then parsed by the lookup and readdir code and, as such, are currently handled in the pagecache like any other file, with the entire directory content being thrown away each time the directory changes. However, since the blob is a known structure and since the data version counter on a directory increases by exactly one for each change committed to that directory, we can actually edit the directory locally rather than fetching it from the server after each locally-induced change. What we can't do, though, is mix data from the server and data from the client since the server is technically at liberty to rearrange or compress a directory if it sees fit, provided it updates the data version number when it does so and breaks the callback (ie. sends a notification). Further, lookup with lookup-ahead, readdir and, when it arrives, local editing are likely want to scan the whole of a directory. So directory handling needs to be improved to maintain the coherency of the directory blob prior to permitting local directory editing. To this end: (1) If any directory page gets discarded, invalidate and reread the entire directory. (2) If readpage notes that if when it fetches a single page that the version number has changed, the entire directory is flagged for invalidation. (3) Read as much of the directory in one go as we can. Note that this removes local caching of directories in fscache for the moment as we can't pass the pages to fscache_read_or_alloc_pages() since page->lru is in use by the LRU. Signed-off-by: David Howells <dhowells@redhat.com>
2018-04-06 13:17:25 +00:00
&vnode->status.data_version, req) < 0)
return -EBADMSG;
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
xdr_decode_AFSCallBack(call, vnode, &bp);
if (call->reply[1])
xdr_decode_AFSVolSync(&bp, call->reply[1]);
call->offset = 0;
call->unmarshall++;
case 6:
break;
}
for (; req->index < req->nr_pages; req->index++) {
if (call->count < PAGE_SIZE)
zero_user_segment(req->pages[req->index],
call->count, PAGE_SIZE);
if (req->page_done)
req->page_done(call, req);
call->count = 0;
}
_leave(" = 0 [done]");
return 0;
}
static void afs_fetch_data_destructor(struct afs_call *call)
{
struct afs_read *req = call->reply[2];
afs_put_read(req);
afs_flat_call_destructor(call);
}
/*
* FS.FetchData operation type
*/
static const struct afs_call_type afs_RXFSFetchData = {
.name = "FS.FetchData",
.op = afs_FS_FetchData,
.deliver = afs_deliver_fs_fetch_data,
.destructor = afs_fetch_data_destructor,
};
static const struct afs_call_type afs_RXFSFetchData64 = {
.name = "FS.FetchData64",
.op = afs_FS_FetchData64,
.deliver = afs_deliver_fs_fetch_data,
.destructor = afs_fetch_data_destructor,
};
/*
* fetch data from a very large file
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
static int afs_fs_fetch_data64(struct afs_fs_cursor *fc, struct afs_read *req)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
_enter("");
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSFetchData64, 32, (21 + 3 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->reply[1] = NULL; /* volsync */
call->reply[2] = req;
call->expected_version = vnode->status.data_version;
/* marshall the parameters */
bp = call->request;
bp[0] = htonl(FSFETCHDATA64);
bp[1] = htonl(vnode->fid.vid);
bp[2] = htonl(vnode->fid.vnode);
bp[3] = htonl(vnode->fid.unique);
bp[4] = htonl(upper_32_bits(req->pos));
bp[5] = htonl(lower_32_bits(req->pos));
bp[6] = 0;
bp[7] = htonl(lower_32_bits(req->len));
afs: Fix directory handling AFS directories are structured blobs that are downloaded just like files and then parsed by the lookup and readdir code and, as such, are currently handled in the pagecache like any other file, with the entire directory content being thrown away each time the directory changes. However, since the blob is a known structure and since the data version counter on a directory increases by exactly one for each change committed to that directory, we can actually edit the directory locally rather than fetching it from the server after each locally-induced change. What we can't do, though, is mix data from the server and data from the client since the server is technically at liberty to rearrange or compress a directory if it sees fit, provided it updates the data version number when it does so and breaks the callback (ie. sends a notification). Further, lookup with lookup-ahead, readdir and, when it arrives, local editing are likely want to scan the whole of a directory. So directory handling needs to be improved to maintain the coherency of the directory blob prior to permitting local directory editing. To this end: (1) If any directory page gets discarded, invalidate and reread the entire directory. (2) If readpage notes that if when it fetches a single page that the version number has changed, the entire directory is flagged for invalidation. (3) Read as much of the directory in one go as we can. Note that this removes local caching of directories in fscache for the moment as we can't pass the pages to fscache_read_or_alloc_pages() since page->lru is in use by the LRU. Signed-off-by: David Howells <dhowells@redhat.com>
2018-04-06 13:17:25 +00:00
refcount_inc(&req->usage);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->cb_break = fc->cb_break;
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* fetch data from a file
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
int afs_fs_fetch_data(struct afs_fs_cursor *fc, struct afs_read *req)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
if (upper_32_bits(req->pos) ||
upper_32_bits(req->len) ||
upper_32_bits(req->pos + req->len))
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_fs_fetch_data64(fc, req);
_enter("");
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSFetchData, 24, (21 + 3 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->reply[1] = NULL; /* volsync */
call->reply[2] = req;
call->expected_version = vnode->status.data_version;
/* marshall the parameters */
bp = call->request;
bp[0] = htonl(FSFETCHDATA);
bp[1] = htonl(vnode->fid.vid);
bp[2] = htonl(vnode->fid.vnode);
bp[3] = htonl(vnode->fid.unique);
bp[4] = htonl(lower_32_bits(req->pos));
bp[5] = htonl(lower_32_bits(req->len));
afs: Fix directory handling AFS directories are structured blobs that are downloaded just like files and then parsed by the lookup and readdir code and, as such, are currently handled in the pagecache like any other file, with the entire directory content being thrown away each time the directory changes. However, since the blob is a known structure and since the data version counter on a directory increases by exactly one for each change committed to that directory, we can actually edit the directory locally rather than fetching it from the server after each locally-induced change. What we can't do, though, is mix data from the server and data from the client since the server is technically at liberty to rearrange or compress a directory if it sees fit, provided it updates the data version number when it does so and breaks the callback (ie. sends a notification). Further, lookup with lookup-ahead, readdir and, when it arrives, local editing are likely want to scan the whole of a directory. So directory handling needs to be improved to maintain the coherency of the directory blob prior to permitting local directory editing. To this end: (1) If any directory page gets discarded, invalidate and reread the entire directory. (2) If readpage notes that if when it fetches a single page that the version number has changed, the entire directory is flagged for invalidation. (3) Read as much of the directory in one go as we can. Note that this removes local caching of directories in fscache for the moment as we can't pass the pages to fscache_read_or_alloc_pages() since page->lru is in use by the LRU. Signed-off-by: David Howells <dhowells@redhat.com>
2018-04-06 13:17:25 +00:00
refcount_inc(&req->usage);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->cb_break = fc->cb_break;
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.CreateFile or an FS.MakeDir
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_create_vnode(struct afs_call *call)
{
struct afs_vnode *vnode = call->reply[0];
const __be32 *bp;
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
_enter("{%u}", call->unmarshall);
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_transfer_reply(call);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
xdr_decode_AFSFid(&bp, call->reply[1]);
if (xdr_decode_AFSFetchStatus(&bp, call->reply[2], NULL, NULL, NULL) < 0 ||
xdr_decode_AFSFetchStatus(&bp, &vnode->status, vnode,
&call->expected_version, NULL) < 0)
return -EBADMSG;
xdr_decode_AFSCallBack_raw(&bp, call->reply[3]);
/* xdr_decode_AFSVolSync(&bp, call->reply[X]); */
_leave(" = 0 [done]");
return 0;
}
/*
* FS.CreateFile and FS.MakeDir operation type
*/
static const struct afs_call_type afs_RXFSCreateFile = {
.name = "FS.CreateFile",
.op = afs_FS_CreateFile,
.deliver = afs_deliver_fs_create_vnode,
.destructor = afs_flat_call_destructor,
};
static const struct afs_call_type afs_RXFSMakeDir = {
.name = "FS.MakeDir",
.op = afs_FS_MakeDir,
.deliver = afs_deliver_fs_create_vnode,
.destructor = afs_flat_call_destructor,
};
/*
* create a file or make a directory
*/
int afs_fs_create(struct afs_fs_cursor *fc,
const char *name,
umode_t mode,
struct afs_fid *newfid,
struct afs_file_status *newstatus,
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_callback *newcb)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
size_t namesz, reqsz, padsz;
__be32 *bp;
_enter("");
namesz = strlen(name);
padsz = (4 - (namesz & 3)) & 3;
reqsz = (5 * 4) + namesz + padsz + (6 * 4);
call = afs_alloc_flat_call(
net, S_ISDIR(mode) ? &afs_RXFSMakeDir : &afs_RXFSCreateFile,
reqsz, (3 + 21 + 21 + 3 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->reply[1] = newfid;
call->reply[2] = newstatus;
call->reply[3] = newcb;
call->expected_version = vnode->status.data_version;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(S_ISDIR(mode) ? FSMAKEDIR : FSCREATEFILE);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
*bp++ = htonl(namesz);
memcpy(bp, name, namesz);
bp = (void *) bp + namesz;
if (padsz > 0) {
memset(bp, 0, padsz);
bp = (void *) bp + padsz;
}
*bp++ = htonl(AFS_SET_MODE | AFS_SET_MTIME);
*bp++ = htonl(vnode->vfs_inode.i_mtime.tv_sec); /* mtime */
*bp++ = 0; /* owner */
*bp++ = 0; /* group */
*bp++ = htonl(mode & S_IALLUGO); /* unix mode */
*bp++ = 0; /* segment size */
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.RemoveFile or FS.RemoveDir
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_remove(struct afs_call *call)
{
struct afs_vnode *vnode = call->reply[0];
const __be32 *bp;
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
_enter("{%u}", call->unmarshall);
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_transfer_reply(call);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
if (xdr_decode_AFSFetchStatus(&bp, &vnode->status, vnode,
&call->expected_version, NULL) < 0)
return -EBADMSG;
/* xdr_decode_AFSVolSync(&bp, call->reply[X]); */
_leave(" = 0 [done]");
return 0;
}
/*
* FS.RemoveDir/FS.RemoveFile operation type
*/
static const struct afs_call_type afs_RXFSRemoveFile = {
.name = "FS.RemoveFile",
.op = afs_FS_RemoveFile,
.deliver = afs_deliver_fs_remove,
.destructor = afs_flat_call_destructor,
};
static const struct afs_call_type afs_RXFSRemoveDir = {
.name = "FS.RemoveDir",
.op = afs_FS_RemoveDir,
.deliver = afs_deliver_fs_remove,
.destructor = afs_flat_call_destructor,
};
/*
* remove a file or directory
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
int afs_fs_remove(struct afs_fs_cursor *fc, const char *name, bool isdir)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
size_t namesz, reqsz, padsz;
__be32 *bp;
_enter("");
namesz = strlen(name);
padsz = (4 - (namesz & 3)) & 3;
reqsz = (5 * 4) + namesz + padsz;
call = afs_alloc_flat_call(
net, isdir ? &afs_RXFSRemoveDir : &afs_RXFSRemoveFile,
reqsz, (21 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->expected_version = vnode->status.data_version;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(isdir ? FSREMOVEDIR : FSREMOVEFILE);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
*bp++ = htonl(namesz);
memcpy(bp, name, namesz);
bp = (void *) bp + namesz;
if (padsz > 0) {
memset(bp, 0, padsz);
bp = (void *) bp + padsz;
}
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.Link
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_link(struct afs_call *call)
{
struct afs_vnode *dvnode = call->reply[0], *vnode = call->reply[1];
const __be32 *bp;
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
_enter("{%u}", call->unmarshall);
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_transfer_reply(call);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
if (xdr_decode_AFSFetchStatus(&bp, &vnode->status, vnode, NULL, NULL) < 0 ||
xdr_decode_AFSFetchStatus(&bp, &dvnode->status, dvnode,
&call->expected_version, NULL) < 0)
return -EBADMSG;
/* xdr_decode_AFSVolSync(&bp, call->reply[X]); */
_leave(" = 0 [done]");
return 0;
}
/*
* FS.Link operation type
*/
static const struct afs_call_type afs_RXFSLink = {
.name = "FS.Link",
.op = afs_FS_Link,
.deliver = afs_deliver_fs_link,
.destructor = afs_flat_call_destructor,
};
/*
* make a hard link
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
int afs_fs_link(struct afs_fs_cursor *fc, struct afs_vnode *vnode,
const char *name)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *dvnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
size_t namesz, reqsz, padsz;
__be32 *bp;
_enter("");
namesz = strlen(name);
padsz = (4 - (namesz & 3)) & 3;
reqsz = (5 * 4) + namesz + padsz + (3 * 4);
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSLink, reqsz, (21 + 21 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = dvnode;
call->reply[1] = vnode;
call->expected_version = vnode->status.data_version;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSLINK);
*bp++ = htonl(dvnode->fid.vid);
*bp++ = htonl(dvnode->fid.vnode);
*bp++ = htonl(dvnode->fid.unique);
*bp++ = htonl(namesz);
memcpy(bp, name, namesz);
bp = (void *) bp + namesz;
if (padsz > 0) {
memset(bp, 0, padsz);
bp = (void *) bp + padsz;
}
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.Symlink
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_symlink(struct afs_call *call)
{
struct afs_vnode *vnode = call->reply[0];
const __be32 *bp;
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
_enter("{%u}", call->unmarshall);
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_transfer_reply(call);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
xdr_decode_AFSFid(&bp, call->reply[1]);
if (xdr_decode_AFSFetchStatus(&bp, call->reply[2], NULL, NULL, NULL) ||
xdr_decode_AFSFetchStatus(&bp, &vnode->status, vnode,
&call->expected_version, NULL) < 0)
return -EBADMSG;
/* xdr_decode_AFSVolSync(&bp, call->reply[X]); */
_leave(" = 0 [done]");
return 0;
}
/*
* FS.Symlink operation type
*/
static const struct afs_call_type afs_RXFSSymlink = {
.name = "FS.Symlink",
.op = afs_FS_Symlink,
.deliver = afs_deliver_fs_symlink,
.destructor = afs_flat_call_destructor,
};
/*
* create a symbolic link
*/
int afs_fs_symlink(struct afs_fs_cursor *fc,
const char *name,
const char *contents,
struct afs_fid *newfid,
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_file_status *newstatus)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
size_t namesz, reqsz, padsz, c_namesz, c_padsz;
__be32 *bp;
_enter("");
namesz = strlen(name);
padsz = (4 - (namesz & 3)) & 3;
c_namesz = strlen(contents);
c_padsz = (4 - (c_namesz & 3)) & 3;
reqsz = (6 * 4) + namesz + padsz + c_namesz + c_padsz + (6 * 4);
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSSymlink, reqsz,
(3 + 21 + 21 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->reply[1] = newfid;
call->reply[2] = newstatus;
call->expected_version = vnode->status.data_version;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSSYMLINK);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
*bp++ = htonl(namesz);
memcpy(bp, name, namesz);
bp = (void *) bp + namesz;
if (padsz > 0) {
memset(bp, 0, padsz);
bp = (void *) bp + padsz;
}
*bp++ = htonl(c_namesz);
memcpy(bp, contents, c_namesz);
bp = (void *) bp + c_namesz;
if (c_padsz > 0) {
memset(bp, 0, c_padsz);
bp = (void *) bp + c_padsz;
}
*bp++ = htonl(AFS_SET_MODE | AFS_SET_MTIME);
*bp++ = htonl(vnode->vfs_inode.i_mtime.tv_sec); /* mtime */
*bp++ = 0; /* owner */
*bp++ = 0; /* group */
*bp++ = htonl(S_IRWXUGO); /* unix mode */
*bp++ = 0; /* segment size */
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.Rename
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_rename(struct afs_call *call)
{
struct afs_vnode *orig_dvnode = call->reply[0], *new_dvnode = call->reply[1];
const __be32 *bp;
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
_enter("{%u}", call->unmarshall);
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_transfer_reply(call);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
if (xdr_decode_AFSFetchStatus(&bp, &orig_dvnode->status, orig_dvnode,
&call->expected_version, NULL) < 0)
return -EBADMSG;
if (new_dvnode != orig_dvnode &&
xdr_decode_AFSFetchStatus(&bp, &new_dvnode->status, new_dvnode,
&call->expected_version_2, NULL) < 0)
return -EBADMSG;
/* xdr_decode_AFSVolSync(&bp, call->reply[X]); */
_leave(" = 0 [done]");
return 0;
}
/*
* FS.Rename operation type
*/
static const struct afs_call_type afs_RXFSRename = {
.name = "FS.Rename",
.op = afs_FS_Rename,
.deliver = afs_deliver_fs_rename,
.destructor = afs_flat_call_destructor,
};
/*
* create a symbolic link
*/
int afs_fs_rename(struct afs_fs_cursor *fc,
const char *orig_name,
struct afs_vnode *new_dvnode,
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
const char *new_name)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *orig_dvnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(orig_dvnode);
size_t reqsz, o_namesz, o_padsz, n_namesz, n_padsz;
__be32 *bp;
_enter("");
o_namesz = strlen(orig_name);
o_padsz = (4 - (o_namesz & 3)) & 3;
n_namesz = strlen(new_name);
n_padsz = (4 - (n_namesz & 3)) & 3;
reqsz = (4 * 4) +
4 + o_namesz + o_padsz +
(3 * 4) +
4 + n_namesz + n_padsz;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSRename, reqsz, (21 + 21 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = orig_dvnode;
call->reply[1] = new_dvnode;
call->expected_version = orig_dvnode->status.data_version;
call->expected_version_2 = new_dvnode->status.data_version;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSRENAME);
*bp++ = htonl(orig_dvnode->fid.vid);
*bp++ = htonl(orig_dvnode->fid.vnode);
*bp++ = htonl(orig_dvnode->fid.unique);
*bp++ = htonl(o_namesz);
memcpy(bp, orig_name, o_namesz);
bp = (void *) bp + o_namesz;
if (o_padsz > 0) {
memset(bp, 0, o_padsz);
bp = (void *) bp + o_padsz;
}
*bp++ = htonl(new_dvnode->fid.vid);
*bp++ = htonl(new_dvnode->fid.vnode);
*bp++ = htonl(new_dvnode->fid.unique);
*bp++ = htonl(n_namesz);
memcpy(bp, new_name, n_namesz);
bp = (void *) bp + n_namesz;
if (n_padsz > 0) {
memset(bp, 0, n_padsz);
bp = (void *) bp + n_padsz;
}
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &orig_dvnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.StoreData
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_store_data(struct afs_call *call)
{
struct afs_vnode *vnode = call->reply[0];
const __be32 *bp;
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
_enter("");
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_transfer_reply(call);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
if (xdr_decode_AFSFetchStatus(&bp, &vnode->status, vnode,
&call->expected_version, NULL) < 0)
return -EBADMSG;
/* xdr_decode_AFSVolSync(&bp, call->reply[X]); */
afs_pages_written_back(vnode, call);
_leave(" = 0 [done]");
return 0;
}
/*
* FS.StoreData operation type
*/
static const struct afs_call_type afs_RXFSStoreData = {
.name = "FS.StoreData",
.op = afs_FS_StoreData,
.deliver = afs_deliver_fs_store_data,
.destructor = afs_flat_call_destructor,
};
static const struct afs_call_type afs_RXFSStoreData64 = {
.name = "FS.StoreData64",
.op = afs_FS_StoreData64,
.deliver = afs_deliver_fs_store_data,
.destructor = afs_flat_call_destructor,
};
/*
* store a set of pages to a very large file
*/
static int afs_fs_store_data64(struct afs_fs_cursor *fc,
struct address_space *mapping,
pgoff_t first, pgoff_t last,
unsigned offset, unsigned to,
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
loff_t size, loff_t pos, loff_t i_size)
{
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
_enter(",%x,{%x:%u},,",
key_serial(fc->key), vnode->fid.vid, vnode->fid.vnode);
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSStoreData64,
(4 + 6 + 3 * 2) * 4,
(21 + 6) * 4);
if (!call)
return -ENOMEM;
call->key = fc->key;
call->mapping = mapping;
call->reply[0] = vnode;
call->first = first;
call->last = last;
call->first_offset = offset;
call->last_to = to;
call->send_pages = true;
call->expected_version = vnode->status.data_version + 1;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSSTOREDATA64);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
*bp++ = htonl(AFS_SET_MTIME); /* mask */
*bp++ = htonl(vnode->vfs_inode.i_mtime.tv_sec); /* mtime */
*bp++ = 0; /* owner */
*bp++ = 0; /* group */
*bp++ = 0; /* unix mode */
*bp++ = 0; /* segment size */
*bp++ = htonl(pos >> 32);
*bp++ = htonl((u32) pos);
*bp++ = htonl(size >> 32);
*bp++ = htonl((u32) size);
*bp++ = htonl(i_size >> 32);
*bp++ = htonl((u32) i_size);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* store a set of pages
*/
int afs_fs_store_data(struct afs_fs_cursor *fc, struct address_space *mapping,
pgoff_t first, pgoff_t last,
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
unsigned offset, unsigned to)
{
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
loff_t size, pos, i_size;
__be32 *bp;
_enter(",%x,{%x:%u},,",
key_serial(fc->key), vnode->fid.vid, vnode->fid.vnode);
size = (loff_t)to - (loff_t)offset;
if (first != last)
size += (loff_t)(last - first) << PAGE_SHIFT;
pos = (loff_t)first << PAGE_SHIFT;
pos += offset;
i_size = i_size_read(&vnode->vfs_inode);
if (pos + size > i_size)
i_size = size + pos;
_debug("size %llx, at %llx, i_size %llx",
(unsigned long long) size, (unsigned long long) pos,
(unsigned long long) i_size);
if (pos >> 32 || i_size >> 32 || size >> 32 || (pos + size) >> 32)
return afs_fs_store_data64(fc, mapping, first, last, offset, to,
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
size, pos, i_size);
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSStoreData,
(4 + 6 + 3) * 4,
(21 + 6) * 4);
if (!call)
return -ENOMEM;
call->key = fc->key;
call->mapping = mapping;
call->reply[0] = vnode;
call->first = first;
call->last = last;
call->first_offset = offset;
call->last_to = to;
call->send_pages = true;
call->expected_version = vnode->status.data_version + 1;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSSTOREDATA);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
*bp++ = htonl(AFS_SET_MTIME); /* mask */
*bp++ = htonl(vnode->vfs_inode.i_mtime.tv_sec); /* mtime */
*bp++ = 0; /* owner */
*bp++ = 0; /* group */
*bp++ = 0; /* unix mode */
*bp++ = 0; /* segment size */
*bp++ = htonl(pos);
*bp++ = htonl(size);
*bp++ = htonl(i_size);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.StoreStatus
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_store_status(struct afs_call *call)
{
struct afs_vnode *vnode = call->reply[0];
const __be32 *bp;
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
_enter("");
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_transfer_reply(call);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
if (xdr_decode_AFSFetchStatus(&bp, &vnode->status, vnode,
&call->expected_version, NULL) < 0)
return -EBADMSG;
/* xdr_decode_AFSVolSync(&bp, call->reply[X]); */
_leave(" = 0 [done]");
return 0;
}
/*
* FS.StoreStatus operation type
*/
static const struct afs_call_type afs_RXFSStoreStatus = {
.name = "FS.StoreStatus",
.op = afs_FS_StoreStatus,
.deliver = afs_deliver_fs_store_status,
.destructor = afs_flat_call_destructor,
};
static const struct afs_call_type afs_RXFSStoreData_as_Status = {
.name = "FS.StoreData",
.op = afs_FS_StoreData,
.deliver = afs_deliver_fs_store_status,
.destructor = afs_flat_call_destructor,
};
static const struct afs_call_type afs_RXFSStoreData64_as_Status = {
.name = "FS.StoreData64",
.op = afs_FS_StoreData64,
.deliver = afs_deliver_fs_store_status,
.destructor = afs_flat_call_destructor,
};
/*
* set the attributes on a very large file, using FS.StoreData rather than
* FS.StoreStatus so as to alter the file size also
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
static int afs_fs_setattr_size64(struct afs_fs_cursor *fc, struct iattr *attr)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
_enter(",%x,{%x:%u},,",
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
key_serial(fc->key), vnode->fid.vid, vnode->fid.vnode);
ASSERT(attr->ia_valid & ATTR_SIZE);
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSStoreData64_as_Status,
(4 + 6 + 3 * 2) * 4,
(21 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->expected_version = vnode->status.data_version + 1;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSSTOREDATA64);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
xdr_encode_AFS_StoreStatus(&bp, attr);
*bp++ = 0; /* position of start of write */
*bp++ = 0;
*bp++ = 0; /* size of write */
*bp++ = 0;
*bp++ = htonl(attr->ia_size >> 32); /* new file length */
*bp++ = htonl((u32) attr->ia_size);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* set the attributes on a file, using FS.StoreData rather than FS.StoreStatus
* so as to alter the file size also
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
static int afs_fs_setattr_size(struct afs_fs_cursor *fc, struct iattr *attr)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
_enter(",%x,{%x:%u},,",
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
key_serial(fc->key), vnode->fid.vid, vnode->fid.vnode);
ASSERT(attr->ia_valid & ATTR_SIZE);
if (attr->ia_size >> 32)
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_fs_setattr_size64(fc, attr);
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSStoreData_as_Status,
(4 + 6 + 3) * 4,
(21 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->expected_version = vnode->status.data_version + 1;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSSTOREDATA);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
xdr_encode_AFS_StoreStatus(&bp, attr);
*bp++ = 0; /* position of start of write */
*bp++ = 0; /* size of write */
*bp++ = htonl(attr->ia_size); /* new file length */
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* set the attributes on a file, using FS.StoreData if there's a change in file
* size, and FS.StoreStatus otherwise
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
int afs_fs_setattr(struct afs_fs_cursor *fc, struct iattr *attr)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
if (attr->ia_valid & ATTR_SIZE)
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_fs_setattr_size(fc, attr);
_enter(",%x,{%x:%u},,",
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
key_serial(fc->key), vnode->fid.vid, vnode->fid.vnode);
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSStoreStatus,
(4 + 6) * 4,
(21 + 6) * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->expected_version = vnode->status.data_version;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSSTORESTATUS);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
xdr_encode_AFS_StoreStatus(&bp, attr);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.GetVolumeStatus
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_get_volume_status(struct afs_call *call)
{
const __be32 *bp;
char *p;
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
_enter("{%u}", call->unmarshall);
switch (call->unmarshall) {
case 0:
call->offset = 0;
call->unmarshall++;
/* extract the returned status record */
case 1:
_debug("extract status");
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, call->buffer,
12 * 4, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
bp = call->buffer;
xdr_decode_AFSFetchVolumeStatus(&bp, call->reply[1]);
call->offset = 0;
call->unmarshall++;
/* extract the volume name length */
case 2:
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, &call->tmp, 4, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
call->count = ntohl(call->tmp);
_debug("volname length: %u", call->count);
if (call->count >= AFSNAMEMAX)
return -EBADMSG;
call->offset = 0;
call->unmarshall++;
/* extract the volume name */
case 3:
_debug("extract volname");
if (call->count > 0) {
ret = afs_extract_data(call, call->reply[2],
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
call->count, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
}
p = call->reply[2];
p[call->count] = 0;
_debug("volname '%s'", p);
call->offset = 0;
call->unmarshall++;
/* extract the volume name padding */
if ((call->count & 3) == 0) {
call->unmarshall++;
goto no_volname_padding;
}
call->count = 4 - (call->count & 3);
case 4:
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, call->buffer,
call->count, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
call->offset = 0;
call->unmarshall++;
no_volname_padding:
/* extract the offline message length */
case 5:
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, &call->tmp, 4, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
call->count = ntohl(call->tmp);
_debug("offline msg length: %u", call->count);
if (call->count >= AFSNAMEMAX)
return -EBADMSG;
call->offset = 0;
call->unmarshall++;
/* extract the offline message */
case 6:
_debug("extract offline");
if (call->count > 0) {
ret = afs_extract_data(call, call->reply[2],
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
call->count, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
}
p = call->reply[2];
p[call->count] = 0;
_debug("offline '%s'", p);
call->offset = 0;
call->unmarshall++;
/* extract the offline message padding */
if ((call->count & 3) == 0) {
call->unmarshall++;
goto no_offline_padding;
}
call->count = 4 - (call->count & 3);
case 7:
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, call->buffer,
call->count, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
call->offset = 0;
call->unmarshall++;
no_offline_padding:
/* extract the message of the day length */
case 8:
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, &call->tmp, 4, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
call->count = ntohl(call->tmp);
_debug("motd length: %u", call->count);
if (call->count >= AFSNAMEMAX)
return -EBADMSG;
call->offset = 0;
call->unmarshall++;
/* extract the message of the day */
case 9:
_debug("extract motd");
if (call->count > 0) {
ret = afs_extract_data(call, call->reply[2],
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
call->count, true);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
}
p = call->reply[2];
p[call->count] = 0;
_debug("motd '%s'", p);
call->offset = 0;
call->unmarshall++;
/* extract the message of the day padding */
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
call->count = (4 - (call->count & 3)) & 3;
case 10:
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_extract_data(call, call->buffer,
call->count, false);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
call->offset = 0;
call->unmarshall++;
case 11:
break;
}
_leave(" = 0 [done]");
return 0;
}
/*
* destroy an FS.GetVolumeStatus call
*/
static void afs_get_volume_status_call_destructor(struct afs_call *call)
{
kfree(call->reply[2]);
call->reply[2] = NULL;
afs_flat_call_destructor(call);
}
/*
* FS.GetVolumeStatus operation type
*/
static const struct afs_call_type afs_RXFSGetVolumeStatus = {
.name = "FS.GetVolumeStatus",
.op = afs_FS_GetVolumeStatus,
.deliver = afs_deliver_fs_get_volume_status,
.destructor = afs_get_volume_status_call_destructor,
};
/*
* fetch the status of a volume
*/
int afs_fs_get_volume_status(struct afs_fs_cursor *fc,
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_volume_status *vs)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
void *tmpbuf;
_enter("");
tmpbuf = kmalloc(AFSOPAQUEMAX, GFP_KERNEL);
if (!tmpbuf)
return -ENOMEM;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSGetVolumeStatus, 2 * 4, 12 * 4);
if (!call) {
kfree(tmpbuf);
return -ENOMEM;
}
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
call->reply[1] = vs;
call->reply[2] = tmpbuf;
/* marshall the parameters */
bp = call->request;
bp[0] = htonl(FSGETVOLUMESTATUS);
bp[1] = htonl(vnode->fid.vid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* deliver reply data to an FS.SetLock, FS.ExtendLock or FS.ReleaseLock
*/
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
static int afs_deliver_fs_xxxx_lock(struct afs_call *call)
{
const __be32 *bp;
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
int ret;
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
_enter("{%u}", call->unmarshall);
rxrpc: Don't expose skbs to in-kernel users [ver #2] Don't expose skbs to in-kernel users, such as the AFS filesystem, but instead provide a notification hook the indicates that a call needs attention and another that indicates that there's a new call to be collected. This makes the following possibilities more achievable: (1) Call refcounting can be made simpler if skbs don't hold refs to calls. (2) skbs referring to non-data events will be able to be freed much sooner rather than being queued for AFS to pick up as rxrpc_kernel_recv_data will be able to consult the call state. (3) We can shortcut the receive phase when a call is remotely aborted because we don't have to go through all the packets to get to the one cancelling the operation. (4) It makes it easier to do encryption/decryption directly between AFS's buffers and sk_buffs. (5) Encryption/decryption can more easily be done in the AFS's thread contexts - usually that of the userspace process that issued a syscall - rather than in one of rxrpc's background threads on a workqueue. (6) AFS will be able to wait synchronously on a call inside AF_RXRPC. To make this work, the following interface function has been added: int rxrpc_kernel_recv_data( struct socket *sock, struct rxrpc_call *call, void *buffer, size_t bufsize, size_t *_offset, bool want_more, u32 *_abort_code); This is the recvmsg equivalent. It allows the caller to find out about the state of a specific call and to transfer received data into a buffer piecemeal. afs_extract_data() and rxrpc_kernel_recv_data() now do all the extraction logic between them. They don't wait synchronously yet because the socket lock needs to be dealt with. Five interface functions have been removed: rxrpc_kernel_is_data_last() rxrpc_kernel_get_abort_code() rxrpc_kernel_get_error_number() rxrpc_kernel_free_skb() rxrpc_kernel_data_consumed() As a temporary hack, sk_buffs going to an in-kernel call are queued on the rxrpc_call struct (->knlrecv_queue) rather than being handed over to the in-kernel user. To process the queue internally, a temporary function, temp_deliver_data() has been added. This will be replaced with common code between the rxrpc_recvmsg() path and the kernel_rxrpc_recv_data() path in a future patch. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-30 19:42:14 +00:00
ret = afs_transfer_reply(call);
rxrpc: Fix races between skb free, ACK generation and replying Inside the kafs filesystem it is possible to occasionally have a call processed and terminated before we've had a chance to check whether we need to clean up the rx queue for that call because afs_send_simple_reply() ends the call when it is done, but this is done in a workqueue item that might happen to run to completion before afs_deliver_to_call() completes. Further, it is possible for rxrpc_kernel_send_data() to be called to send a reply before the last request-phase data skb is released. The rxrpc skb destructor is where the ACK processing is done and the call state is advanced upon release of the last skb. ACK generation is also deferred to a work item because it's possible that the skb destructor is not called in a context where kernel_sendmsg() can be invoked. To this end, the following changes are made: (1) kernel_rxrpc_data_consumed() is added. This should be called whenever an skb is emptied so as to crank the ACK and call states. This does not release the skb, however. kernel_rxrpc_free_skb() must now be called to achieve that. These together replace rxrpc_kernel_data_delivered(). (2) kernel_rxrpc_data_consumed() is wrapped by afs_data_consumed(). This makes afs_deliver_to_call() easier to work as the skb can simply be discarded unconditionally here without trying to work out what the return value of the ->deliver() function means. The ->deliver() functions can, via afs_data_complete(), afs_transfer_reply() and afs_extract_data() mark that an skb has been consumed (thereby cranking the state) without the need to conditionally free the skb to make sure the state is correct on an incoming call for when the call processor tries to send the reply. (3) rxrpc_recvmsg() now has to call kernel_rxrpc_data_consumed() when it has finished with a packet and MSG_PEEK isn't set. (4) rxrpc_packet_destructor() no longer calls rxrpc_hard_ACK_data(). Because of this, we no longer need to clear the destructor and put the call before we free the skb in cases where we don't want the ACK/call state to be cranked. (5) The ->deliver() call-type callbacks are made to return -EAGAIN rather than 0 if they expect more data (afs_extract_data() returns -EAGAIN to the delivery function already), and the caller is now responsible for producing an abort if that was the last packet. (6) There are many bits of unmarshalling code where: ret = afs_extract_data(call, skb, last, ...); switch (ret) { case 0: break; case -EAGAIN: return 0; default: return ret; } is to be found. As -EAGAIN can now be passed back to the caller, we now just return if ret < 0: ret = afs_extract_data(call, skb, last, ...); if (ret < 0) return ret; (7) Checks for trailing data and empty final data packets has been consolidated as afs_data_complete(). So: if (skb->len > 0) return -EBADMSG; if (!last) return 0; becomes: ret = afs_data_complete(call, skb, last); if (ret < 0) return ret; (8) afs_transfer_reply() now checks the amount of data it has against the amount of data desired and the amount of data in the skb and returns an error to induce an abort if we don't get exactly what we want. Without these changes, the following oops can occasionally be observed, particularly if some printks are inserted into the delivery path: general protection fault: 0000 [#1] SMP Modules linked in: kafs(E) af_rxrpc(E) [last unloaded: af_rxrpc] CPU: 0 PID: 1305 Comm: kworker/u8:3 Tainted: G E 4.7.0-fsdevel+ #1303 Hardware name: ASUS All Series/H97-PLUS, BIOS 2306 10/09/2014 Workqueue: kafsd afs_async_workfn [kafs] task: ffff88040be041c0 ti: ffff88040c070000 task.ti: ffff88040c070000 RIP: 0010:[<ffffffff8108fd3c>] [<ffffffff8108fd3c>] __lock_acquire+0xcf/0x15a1 RSP: 0018:ffff88040c073bc0 EFLAGS: 00010002 RAX: 6b6b6b6b6b6b6b6b RBX: 0000000000000000 RCX: ffff88040d29a710 RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88040d29a710 RBP: ffff88040c073c70 R08: 0000000000000001 R09: 0000000000000001 R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: ffff88040be041c0 R15: ffffffff814c928f FS: 0000000000000000(0000) GS:ffff88041fa00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fa4595f4750 CR3: 0000000001c14000 CR4: 00000000001406f0 Stack: 0000000000000006 000000000be04930 0000000000000000 ffff880400000000 ffff880400000000 ffffffff8108f847 ffff88040be041c0 ffffffff81050446 ffff8803fc08a920 ffff8803fc08a958 ffff88040be041c0 ffff88040c073c38 Call Trace: [<ffffffff8108f847>] ? mark_held_locks+0x5e/0x74 [<ffffffff81050446>] ? __local_bh_enable_ip+0x9b/0xa1 [<ffffffff8108f9ca>] ? trace_hardirqs_on_caller+0x16d/0x189 [<ffffffff810915f4>] lock_acquire+0x122/0x1b6 [<ffffffff810915f4>] ? lock_acquire+0x122/0x1b6 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff81609dbf>] _raw_spin_lock_irqsave+0x35/0x49 [<ffffffff814c928f>] ? skb_dequeue+0x18/0x61 [<ffffffff814c928f>] skb_dequeue+0x18/0x61 [<ffffffffa009aa92>] afs_deliver_to_call+0x344/0x39d [kafs] [<ffffffffa009ab37>] afs_process_async_call+0x4c/0xd5 [kafs] [<ffffffffa0099e9c>] afs_async_workfn+0xe/0x10 [kafs] [<ffffffff81063a3a>] process_one_work+0x29d/0x57c [<ffffffff81064ac2>] worker_thread+0x24a/0x385 [<ffffffff81064878>] ? rescuer_thread+0x2d0/0x2d0 [<ffffffff810696f5>] kthread+0xf3/0xfb [<ffffffff8160a6ff>] ret_from_fork+0x1f/0x40 [<ffffffff81069602>] ? kthread_create_on_node+0x1cf/0x1cf Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-08-03 13:11:40 +00:00
if (ret < 0)
return ret;
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
/* xdr_decode_AFSVolSync(&bp, call->reply[X]); */
_leave(" = 0 [done]");
return 0;
}
/*
* FS.SetLock operation type
*/
static const struct afs_call_type afs_RXFSSetLock = {
.name = "FS.SetLock",
.op = afs_FS_SetLock,
.deliver = afs_deliver_fs_xxxx_lock,
.destructor = afs_flat_call_destructor,
};
/*
* FS.ExtendLock operation type
*/
static const struct afs_call_type afs_RXFSExtendLock = {
.name = "FS.ExtendLock",
.op = afs_FS_ExtendLock,
.deliver = afs_deliver_fs_xxxx_lock,
.destructor = afs_flat_call_destructor,
};
/*
* FS.ReleaseLock operation type
*/
static const struct afs_call_type afs_RXFSReleaseLock = {
.name = "FS.ReleaseLock",
.op = afs_FS_ReleaseLock,
.deliver = afs_deliver_fs_xxxx_lock,
.destructor = afs_flat_call_destructor,
};
/*
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
* Set a lock on a file
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
int afs_fs_set_lock(struct afs_fs_cursor *fc, afs_lock_type_t type)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
_enter("");
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSSetLock, 5 * 4, 6 * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSSETLOCK);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
*bp++ = htonl(type);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* extend a lock on a file
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
int afs_fs_extend_lock(struct afs_fs_cursor *fc)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
_enter("");
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSExtendLock, 4 * 4, 6 * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSEXTENDLOCK);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* release a lock on a file
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
int afs_fs_release_lock(struct afs_fs_cursor *fc)
{
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct afs_vnode *vnode = fc->vnode;
struct afs_call *call;
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
struct afs_net *net = afs_v2net(vnode);
__be32 *bp;
_enter("");
afs: Lay the groundwork for supporting network namespaces Lay the groundwork for supporting network namespaces (netns) to the AFS filesystem by moving various global features to a network-namespace struct (afs_net) and providing an instance of this as a temporary global variable that everything uses via accessor functions for the moment. The following changes have been made: (1) Store the netns in the superblock info. This will be obtained from the mounter's nsproxy on a manual mount and inherited from the parent superblock on an automount. (2) The cell list is made per-netns. It can be viewed through /proc/net/afs/cells and also be modified by writing commands to that file. (3) The local workstation cell is set per-ns in /proc/net/afs/rootcell. This is unset by default. (4) The 'rootcell' module parameter, which sets a cell and VL server list modifies the init net namespace, thereby allowing an AFS root fs to be theoretically used. (5) The volume location lists and the file lock manager are made per-netns. (6) The AF_RXRPC socket and associated I/O bits are made per-ns. The various workqueues remain global for the moment. Changes still to be made: (1) /proc/fs/afs/ should be moved to /proc/net/afs/ and a symlink emplaced from the old name. (2) A per-netns subsys needs to be registered for AFS into which it can store its per-netns data. (3) Rather than the AF_RXRPC socket being opened on module init, it needs to be opened on the creation of a superblock in that netns. (4) The socket needs to be closed when the last superblock using it is destroyed and all outstanding client calls on it have been completed. This prevents a reference loop on the namespace. (5) It is possible that several namespaces will want to use AFS, in which case each one will need its own UDP port. These can either be set through /proc/net/afs/cm_port or the kernel can pick one at random. The init_ns gets 7001 by default. Other issues that need resolving: (1) The DNS keyring needs net-namespacing. (2) Where do upcalls go (eg. DNS request-key upcall)? (3) Need something like open_socket_in_file_ns() syscall so that AFS command line tools attempting to operate on an AFS file/volume have their RPC calls go to the right place. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:45 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSReleaseLock, 4 * 4, 6 * 4);
if (!call)
return -ENOMEM;
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call->key = fc->key;
call->reply[0] = vnode;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSRELEASELOCK);
*bp++ = htonl(vnode->fid.vid);
*bp++ = htonl(vnode->fid.vnode);
*bp++ = htonl(vnode->fid.unique);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &vnode->fid);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
}
/*
* Deliver reply data to an FS.GiveUpAllCallBacks operation.
*/
static int afs_deliver_fs_give_up_all_callbacks(struct afs_call *call)
{
return afs_transfer_reply(call);
}
/*
* FS.GiveUpAllCallBacks operation type
*/
static const struct afs_call_type afs_RXFSGiveUpAllCallBacks = {
.name = "FS.GiveUpAllCallBacks",
.op = afs_FS_GiveUpAllCallBacks,
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
.deliver = afs_deliver_fs_give_up_all_callbacks,
.destructor = afs_flat_call_destructor,
};
/*
* Flush all the callbacks we have on a server.
*/
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
int afs_fs_give_up_all_callbacks(struct afs_net *net,
struct afs_server *server,
struct afs_addr_cursor *ac,
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
struct key *key)
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
{
struct afs_call *call;
__be32 *bp;
_enter("");
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
call = afs_alloc_flat_call(net, &afs_RXFSGiveUpAllCallBacks, 1 * 4, 0);
afs: Overhaul the callback handling Overhaul the AFS callback handling by the following means: (1) Don't give up callback promises on vnodes that we are no longer using, rather let them just expire on the server or let the server break them. This is actually more efficient for the server as the callback lookup is expensive if there are lots of extant callbacks. (2) Only give up the callback promises we have from a server when the server record is destroyed. Then we can just give up *all* the callback promises on it in one go. (3) Servers can end up being shared between cells if cells are aliased, so don't add all the vnodes being backed by a particular server into a big FID-indexed tree on that server as there may be duplicates. Instead have each volume instance (~= superblock) register an interest in a server as it starts to make use of it and use this to allow the processor for callbacks from the server to find the superblock and thence the inode corresponding to the FID being broken by means of ilookup_nowait(). (4) Rather than iterating over the entire callback list when a mass-break comes in from the server, maintain a counter of mass-breaks in afs_server (cb_seq) and make afs_validate() check it against the copy in afs_vnode. It would be nice not to have to take a read_lock whilst doing this, but that's tricky without using RCU. (5) Save a ref on the fileserver we're using for a call in the afs_call struct so that we can access its cb_s_break during call decoding. (6) Write-lock around callback and status storage in a vnode and read-lock around getattr so that we don't see the status mid-update. This has the following consequences: (1) Data invalidation isn't seen until someone calls afs_validate() on a vnode. Unfortunately, we need to use a key to query the server, but getting one from a background thread is tricky without caching loads of keys all over the place. (2) Mass invalidation isn't seen until someone calls afs_validate(). (3) Callback breaking is going to hit the inode_hash_lock quite a bit. Could this be replaced with rcu_read_lock() since inodes are destroyed under RCU conditions. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:49 +00:00
if (!call)
return -ENOMEM;
call->key = key;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSGIVEUPALLCALLBACKS);
/* Can't take a ref on server */
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(ac, call, GFP_NOFS, false);
}
/*
* Deliver reply data to an FS.GetCapabilities operation.
*/
static int afs_deliver_fs_get_capabilities(struct afs_call *call)
{
u32 count;
int ret;
_enter("{%u,%zu/%u}", call->unmarshall, call->offset, call->count);
again:
switch (call->unmarshall) {
case 0:
call->offset = 0;
call->unmarshall++;
/* Extract the capabilities word count */
case 1:
ret = afs_extract_data(call, &call->tmp,
1 * sizeof(__be32),
true);
if (ret < 0)
return ret;
count = ntohl(call->tmp);
call->count = count;
call->count2 = count;
call->offset = 0;
call->unmarshall++;
/* Extract capabilities words */
case 2:
count = min(call->count, 16U);
ret = afs_extract_data(call, call->buffer,
count * sizeof(__be32),
call->count > 16);
if (ret < 0)
return ret;
/* TODO: Examine capabilities */
call->count -= count;
if (call->count > 0)
goto again;
call->offset = 0;
call->unmarshall++;
break;
}
_leave(" = 0 [done]");
return 0;
}
/*
* FS.GetCapabilities operation type
*/
static const struct afs_call_type afs_RXFSGetCapabilities = {
.name = "FS.GetCapabilities",
.op = afs_FS_GetCapabilities,
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
.deliver = afs_deliver_fs_get_capabilities,
.destructor = afs_flat_call_destructor,
};
/*
* Probe a fileserver for the capabilities that it supports. This can
* return up to 196 words.
*/
int afs_fs_get_capabilities(struct afs_net *net,
struct afs_server *server,
struct afs_addr_cursor *ac,
struct key *key)
{
struct afs_call *call;
__be32 *bp;
_enter("");
call = afs_alloc_flat_call(net, &afs_RXFSGetCapabilities, 1 * 4, 16 * 4);
if (!call)
return -ENOMEM;
call->key = key;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSGETCAPABILITIES);
/* Can't take a ref on server */
trace_afs_make_fs_call(call, NULL);
afs: Overhaul volume and server record caching and fileserver rotation The current code assumes that volumes and servers are per-cell and are never shared, but this is not enforced, and, indeed, public cells do exist that are aliases of each other. Further, an organisation can, say, set up a public cell and a private cell with overlapping, but not identical, sets of servers. The difference is purely in the database attached to the VL servers. The current code will malfunction if it sees a server in two cells as it assumes global address -> server record mappings and that each server is in just one cell. Further, each server may have multiple addresses - and may have addresses of different families (IPv4 and IPv6, say). To this end, the following structural changes are made: (1) Server record management is overhauled: (a) Server records are made independent of cell. The namespace keeps track of them, volume records have lists of them and each vnode has a server on which its callback interest currently resides. (b) The cell record no longer keeps a list of servers known to be in that cell. (c) The server records are now kept in a flat list because there's no single address to sort on. (d) Server records are now keyed by their UUID within the namespace. (e) The addresses for a server are obtained with the VL.GetAddrsU rather than with VL.GetEntryByName, using the server's UUID as a parameter. (f) Cached server records are garbage collected after a period of non-use and are counted out of existence before purging is allowed to complete. This protects the work functions against rmmod. (g) The servers list is now in /proc/fs/afs/servers. (2) Volume record management is overhauled: (a) An RCU-replaceable server list is introduced. This tracks both servers and their coresponding callback interests. (b) The superblock is now keyed on cell record and numeric volume ID. (c) The volume record is now tied to the superblock which mounts it, and is activated when mounted and deactivated when unmounted. This makes it easier to handle the cache cookie without causing a double-use in fscache. (d) The volume record is loaded from the VLDB using VL.GetEntryByNameU to get the server UUID list. (e) The volume name is updated if it is seen to have changed when the volume is updated (the update is keyed on the volume ID). (3) The vlocation record is got rid of and VLDB records are no longer cached. Sufficient information is stored in the volume record, though an update to a volume record is now no longer shared between related volumes (volumes come in bundles of three: R/W, R/O and backup). and the following procedural changes are made: (1) The fileserver cursor introduced previously is now fleshed out and used to iterate over fileservers and their addresses. (2) Volume status is checked during iteration, and the server list is replaced if a change is detected. (3) Server status is checked during iteration, and the address list is replaced if a change is detected. (4) The abort code is saved into the address list cursor and -ECONNABORTED returned in afs_make_call() if a remote abort happened rather than translating the abort into an error message. This allows actions to be taken depending on the abort code more easily. (a) If a VMOVED abort is seen then this is handled by rechecking the volume and restarting the iteration. (b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is handled by sleeping for a short period and retrying and/or trying other servers that might serve that volume. A message is also displayed once until the condition has cleared. (c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the moment. (d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to see if it has been deleted; if not, the fileserver is probably indicating that the volume couldn't be attached and needs salvaging. (e) If statfs() sees one of these aborts, it does not sleep, but rather returns an error, so as not to block the umount program. (5) The fileserver iteration functions in vnode.c are now merged into their callers and more heavily macroised around the cursor. vnode.c is removed. (6) Operations on a particular vnode are serialised on that vnode because the server will lock that vnode whilst it operates on it, so a second op sent will just have to wait. (7) Fileservers are probed with FS.GetCapabilities before being used. This is where service upgrade will be done. (8) A callback interest on a fileserver is set up before an FS operation is performed and passed through to afs_make_call() so that it can be set on the vnode if the operation returns a callback. The callback interest is passed through to afs_iget() also so that it can be set there too. In general, record updating is done on an as-needed basis when we try to access servers, volumes or vnodes rather than offloading it to work items and special threads. Notes: (1) Pre AFS-3.4 servers are no longer supported, though this can be added back if necessary (AFS-3.4 was released in 1998). (2) VBUSY is retried forever for the moment at intervals of 1s. (3) /proc/fs/afs/<cell>/servers no longer exists. Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 15:27:50 +00:00
return afs_make_call(ac, call, GFP_NOFS, false);
}
/*
* Deliver reply data to an FS.FetchStatus with no vnode.
*/
static int afs_deliver_fs_fetch_status(struct afs_call *call)
{
struct afs_file_status *status = call->reply[1];
struct afs_callback *callback = call->reply[2];
struct afs_volsync *volsync = call->reply[3];
struct afs_vnode *vnode = call->reply[0];
const __be32 *bp;
int ret;
ret = afs_transfer_reply(call);
if (ret < 0)
return ret;
_enter("{%x:%u}", vnode->fid.vid, vnode->fid.vnode);
/* unmarshall the reply once we've received all of it */
bp = call->buffer;
xdr_decode_AFSFetchStatus(&bp, status, vnode,
&call->expected_version, NULL);
callback[call->count].version = ntohl(bp[0]);
callback[call->count].expiry = ntohl(bp[1]);
callback[call->count].type = ntohl(bp[2]);
if (vnode)
xdr_decode_AFSCallBack(call, vnode, &bp);
else
bp += 3;
if (volsync)
xdr_decode_AFSVolSync(&bp, volsync);
_leave(" = 0 [done]");
return 0;
}
/*
* FS.FetchStatus operation type
*/
static const struct afs_call_type afs_RXFSFetchStatus = {
.name = "FS.FetchStatus",
.op = afs_FS_FetchStatus,
.deliver = afs_deliver_fs_fetch_status,
.destructor = afs_flat_call_destructor,
};
/*
* Fetch the status information for a fid without needing a vnode handle.
*/
int afs_fs_fetch_status(struct afs_fs_cursor *fc,
struct afs_net *net,
struct afs_fid *fid,
struct afs_file_status *status,
struct afs_callback *callback,
struct afs_volsync *volsync)
{
struct afs_call *call;
__be32 *bp;
_enter(",%x,{%x:%u},,",
key_serial(fc->key), fid->vid, fid->vnode);
call = afs_alloc_flat_call(net, &afs_RXFSFetchStatus, 16, (21 + 3 + 6) * 4);
if (!call) {
fc->ac.error = -ENOMEM;
return -ENOMEM;
}
call->key = fc->key;
call->reply[0] = NULL; /* vnode for fid[0] */
call->reply[1] = status;
call->reply[2] = callback;
call->reply[3] = volsync;
call->expected_version = 1; /* vnode->status.data_version */
/* marshall the parameters */
bp = call->request;
bp[0] = htonl(FSFETCHSTATUS);
bp[1] = htonl(fid->vid);
bp[2] = htonl(fid->vnode);
bp[3] = htonl(fid->unique);
call->cb_break = fc->cb_break;
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, fid);
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}
/*
* Deliver reply data to an FS.InlineBulkStatus call
*/
static int afs_deliver_fs_inline_bulk_status(struct afs_call *call)
{
struct afs_file_status *statuses;
struct afs_callback *callbacks;
struct afs_vnode *vnode = call->reply[0];
const __be32 *bp;
u32 tmp;
int ret;
_enter("{%u}", call->unmarshall);
switch (call->unmarshall) {
case 0:
call->offset = 0;
call->unmarshall++;
/* Extract the file status count and array in two steps */
case 1:
_debug("extract status count");
ret = afs_extract_data(call, &call->tmp, 4, true);
if (ret < 0)
return ret;
tmp = ntohl(call->tmp);
_debug("status count: %u/%u", tmp, call->count2);
if (tmp != call->count2)
return -EBADMSG;
call->count = 0;
call->unmarshall++;
more_counts:
call->offset = 0;
case 2:
_debug("extract status array %u", call->count);
ret = afs_extract_data(call, call->buffer, 21 * 4, true);
if (ret < 0)
return ret;
bp = call->buffer;
statuses = call->reply[1];
if (xdr_decode_AFSFetchStatus(&bp, &statuses[call->count],
call->count == 0 ? vnode : NULL,
NULL, NULL) < 0)
return -EBADMSG;
call->count++;
if (call->count < call->count2)
goto more_counts;
call->count = 0;
call->unmarshall++;
call->offset = 0;
/* Extract the callback count and array in two steps */
case 3:
_debug("extract CB count");
ret = afs_extract_data(call, &call->tmp, 4, true);
if (ret < 0)
return ret;
tmp = ntohl(call->tmp);
_debug("CB count: %u", tmp);
if (tmp != call->count2)
return -EBADMSG;
call->count = 0;
call->unmarshall++;
more_cbs:
call->offset = 0;
case 4:
_debug("extract CB array");
ret = afs_extract_data(call, call->buffer, 3 * 4, true);
if (ret < 0)
return ret;
_debug("unmarshall CB array");
bp = call->buffer;
callbacks = call->reply[2];
callbacks[call->count].version = ntohl(bp[0]);
callbacks[call->count].expiry = ntohl(bp[1]);
callbacks[call->count].type = ntohl(bp[2]);
statuses = call->reply[1];
if (call->count == 0 && vnode && statuses[0].abort_code == 0)
xdr_decode_AFSCallBack(call, vnode, &bp);
call->count++;
if (call->count < call->count2)
goto more_cbs;
call->offset = 0;
call->unmarshall++;
case 5:
ret = afs_extract_data(call, call->buffer, 6 * 4, false);
if (ret < 0)
return ret;
bp = call->buffer;
if (call->reply[3])
xdr_decode_AFSVolSync(&bp, call->reply[3]);
call->offset = 0;
call->unmarshall++;
case 6:
break;
}
_leave(" = 0 [done]");
return 0;
}
/*
* FS.InlineBulkStatus operation type
*/
static const struct afs_call_type afs_RXFSInlineBulkStatus = {
.name = "FS.InlineBulkStatus",
.op = afs_FS_InlineBulkStatus,
.deliver = afs_deliver_fs_inline_bulk_status,
.destructor = afs_flat_call_destructor,
};
/*
* Fetch the status information for up to 50 files
*/
int afs_fs_inline_bulk_status(struct afs_fs_cursor *fc,
struct afs_net *net,
struct afs_fid *fids,
struct afs_file_status *statuses,
struct afs_callback *callbacks,
unsigned int nr_fids,
struct afs_volsync *volsync)
{
struct afs_call *call;
__be32 *bp;
int i;
_enter(",%x,{%x:%u},%u",
key_serial(fc->key), fids[0].vid, fids[1].vnode, nr_fids);
call = afs_alloc_flat_call(net, &afs_RXFSInlineBulkStatus,
(2 + nr_fids * 3) * 4,
21 * 4);
if (!call) {
fc->ac.error = -ENOMEM;
return -ENOMEM;
}
call->key = fc->key;
call->reply[0] = NULL; /* vnode for fid[0] */
call->reply[1] = statuses;
call->reply[2] = callbacks;
call->reply[3] = volsync;
call->count2 = nr_fids;
/* marshall the parameters */
bp = call->request;
*bp++ = htonl(FSINLINEBULKSTATUS);
*bp++ = htonl(nr_fids);
for (i = 0; i < nr_fids; i++) {
*bp++ = htonl(fids[i].vid);
*bp++ = htonl(fids[i].vnode);
*bp++ = htonl(fids[i].unique);
}
call->cb_break = fc->cb_break;
afs_use_fs_server(call, fc->cbi);
trace_afs_make_fs_call(call, &fids[0]);
return afs_make_call(&fc->ac, call, GFP_NOFS, false);
}