linux/fs/autofs4/waitq.c

579 lines
14 KiB
C
Raw Normal View History

/* -*- c -*- --------------------------------------------------------------- *
*
* linux/fs/autofs/waitq.c
*
* Copyright 1997-1998 Transmeta Corporation -- All Rights Reserved
* Copyright 2001-2006 Ian Kent <raven@themaw.net>
*
* This file is part of the Linux kernel and is made available under
* the terms of the GNU General Public License, version 2, or at your
* option, any later version, incorporated herein by reference.
*
* ------------------------------------------------------------------------- */
#include <linux/slab.h>
#include <linux/time.h>
#include <linux/signal.h>
#include <linux/file.h>
#include "autofs_i.h"
/* We make this a static variable rather than a part of the superblock; it
is better if we don't reassign numbers easily even across filesystems */
static autofs_wqt_t autofs4_next_wait_queue = 1;
/* These are the signals we allow interrupting a pending mount */
#define SHUTDOWN_SIGS (sigmask(SIGKILL) | sigmask(SIGINT) | sigmask(SIGQUIT))
void autofs4_catatonic_mode(struct autofs_sb_info *sbi)
{
struct autofs_wait_queue *wq, *nwq;
mutex_lock(&sbi->wq_mutex);
if (sbi->catatonic) {
mutex_unlock(&sbi->wq_mutex);
return;
}
DPRINTK("entering catatonic mode");
sbi->catatonic = 1;
wq = sbi->queues;
sbi->queues = NULL; /* Erase all wait queues */
while (wq) {
nwq = wq->next;
wq->status = -ENOENT; /* Magic is gone - report failure */
if (wq->name.name) {
kfree(wq->name.name);
wq->name.name = NULL;
}
wq->wait_ctr--;
wake_up_interruptible(&wq->queue);
wq = nwq;
}
fput(sbi->pipe); /* Close the pipe */
sbi->pipe = NULL;
sbi->pipefd = -1;
mutex_unlock(&sbi->wq_mutex);
}
static int autofs4_write(struct autofs_sb_info *sbi,
struct file *file, const void *addr, int bytes)
{
unsigned long sigpipe, flags;
mm_segment_t fs;
const char *data = (const char *)addr;
ssize_t wr = 0;
sigpipe = sigismember(&current->pending.signal, SIGPIPE);
/* Save pointer to user space and point back to kernel space */
fs = get_fs();
set_fs(KERNEL_DS);
mutex_lock(&sbi->pipe_mutex);
while (bytes &&
(wr = file->f_op->write(file,data,bytes,&file->f_pos)) > 0) {
data += wr;
bytes -= wr;
}
mutex_unlock(&sbi->pipe_mutex);
set_fs(fs);
/* Keep the currently executing process from receiving a
SIGPIPE unless it was already supposed to get one */
if (wr == -EPIPE && !sigpipe) {
spin_lock_irqsave(&current->sighand->siglock, flags);
sigdelset(&current->pending.signal, SIGPIPE);
recalc_sigpending();
spin_unlock_irqrestore(&current->sighand->siglock, flags);
}
return (bytes > 0);
}
autofs: work around unhappy compat problem on x86-64 When the autofs protocol version 5 packet type was added in commit 5c0a32fc2cd0 ("autofs4: add new packet type for v5 communications"), it obvously tried quite hard to be word-size agnostic, and uses explicitly sized fields that are all correctly aligned. However, with the final "char name[NAME_MAX+1]" array at the end, the actual size of the structure ends up being not very well defined: because the struct isn't marked 'packed', doing a "sizeof()" on it will align the size of the struct up to the biggest alignment of the members it has. And despite all the members being the same, the alignment of them is different: a "__u64" has 4-byte alignment on x86-32, but native 8-byte alignment on x86-64. And while 'NAME_MAX+1' ends up being a nice round number (256), the name[] array starts out a 4-byte aligned. End result: the "packed" size of the structure is 300 bytes: 4-byte, but not 8-byte aligned. As a result, despite all the fields being in the same place on all architectures, sizeof() will round up that size to 304 bytes on architectures that have 8-byte alignment for u64. Note that this is *not* a problem for 32-bit compat mode on POWER, since there __u64 is 8-byte aligned even in 32-bit mode. But on x86, 32-bit and 64-bit alignment is different for 64-bit entities, and as a result the structure that has exactly the same layout has different sizes. So on x86-64, but no other architecture, we will just subtract 4 from the size of the structure when running in a compat task. That way we will write the properly sized packet that user mode expects. Not pretty. Sadly, this very subtle, and unnecessary, size difference has been encoded in user space that wants to read packets of *exactly* the right size, and will refuse to touch anything else. Reported-and-tested-by: Thomas Meyer <thomas@m3y3r.de> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-02-22 12:45:44 +00:00
/*
* The autofs_v5 packet was misdesigned.
*
* The packets are identical on x86-32 and x86-64, but have different
* alignment. Which means that 'sizeof()' will give different results.
* Fix it up for the case of running 32-bit user mode on a 64-bit kernel.
*/
static noinline size_t autofs_v5_packet_size(struct autofs_sb_info *sbi)
{
size_t pktsz = sizeof(struct autofs_v5_packet);
#if defined(CONFIG_X86_64) && defined(CONFIG_COMPAT)
if (sbi->compat_daemon > 0)
pktsz -= 4;
#endif
return pktsz;
}
static void autofs4_notify_daemon(struct autofs_sb_info *sbi,
struct autofs_wait_queue *wq,
int type)
{
union {
struct autofs_packet_hdr hdr;
union autofs_packet_union v4_pkt;
union autofs_v5_packet_union v5_pkt;
} pkt;
struct file *pipe = NULL;
size_t pktsz;
DPRINTK("wait id = 0x%08lx, name = %.*s, type=%d",
(unsigned long) wq->wait_queue_token, wq->name.len, wq->name.name, type);
memset(&pkt,0,sizeof pkt); /* For security reasons */
pkt.hdr.proto_version = sbi->version;
pkt.hdr.type = type;
mutex_lock(&sbi->wq_mutex);
/* Check if we have become catatonic */
if (sbi->catatonic) {
mutex_unlock(&sbi->wq_mutex);
return;
}
switch (type) {
/* Kernel protocol v4 missing and expire packets */
case autofs_ptype_missing:
{
struct autofs_packet_missing *mp = &pkt.v4_pkt.missing;
pktsz = sizeof(*mp);
mp->wait_queue_token = wq->wait_queue_token;
mp->len = wq->name.len;
memcpy(mp->name, wq->name.name, wq->name.len);
mp->name[wq->name.len] = '\0';
break;
}
case autofs_ptype_expire_multi:
{
struct autofs_packet_expire_multi *ep = &pkt.v4_pkt.expire_multi;
pktsz = sizeof(*ep);
ep->wait_queue_token = wq->wait_queue_token;
ep->len = wq->name.len;
memcpy(ep->name, wq->name.name, wq->name.len);
ep->name[wq->name.len] = '\0';
break;
}
/*
* Kernel protocol v5 packet for handling indirect and direct
* mount missing and expire requests
*/
case autofs_ptype_missing_indirect:
case autofs_ptype_expire_indirect:
case autofs_ptype_missing_direct:
case autofs_ptype_expire_direct:
{
struct autofs_v5_packet *packet = &pkt.v5_pkt.v5_packet;
autofs: work around unhappy compat problem on x86-64 When the autofs protocol version 5 packet type was added in commit 5c0a32fc2cd0 ("autofs4: add new packet type for v5 communications"), it obvously tried quite hard to be word-size agnostic, and uses explicitly sized fields that are all correctly aligned. However, with the final "char name[NAME_MAX+1]" array at the end, the actual size of the structure ends up being not very well defined: because the struct isn't marked 'packed', doing a "sizeof()" on it will align the size of the struct up to the biggest alignment of the members it has. And despite all the members being the same, the alignment of them is different: a "__u64" has 4-byte alignment on x86-32, but native 8-byte alignment on x86-64. And while 'NAME_MAX+1' ends up being a nice round number (256), the name[] array starts out a 4-byte aligned. End result: the "packed" size of the structure is 300 bytes: 4-byte, but not 8-byte aligned. As a result, despite all the fields being in the same place on all architectures, sizeof() will round up that size to 304 bytes on architectures that have 8-byte alignment for u64. Note that this is *not* a problem for 32-bit compat mode on POWER, since there __u64 is 8-byte aligned even in 32-bit mode. But on x86, 32-bit and 64-bit alignment is different for 64-bit entities, and as a result the structure that has exactly the same layout has different sizes. So on x86-64, but no other architecture, we will just subtract 4 from the size of the structure when running in a compat task. That way we will write the properly sized packet that user mode expects. Not pretty. Sadly, this very subtle, and unnecessary, size difference has been encoded in user space that wants to read packets of *exactly* the right size, and will refuse to touch anything else. Reported-and-tested-by: Thomas Meyer <thomas@m3y3r.de> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-02-22 12:45:44 +00:00
pktsz = autofs_v5_packet_size(sbi);
packet->wait_queue_token = wq->wait_queue_token;
packet->len = wq->name.len;
memcpy(packet->name, wq->name.name, wq->name.len);
packet->name[wq->name.len] = '\0';
packet->dev = wq->dev;
packet->ino = wq->ino;
packet->uid = wq->uid;
packet->gid = wq->gid;
packet->pid = wq->pid;
packet->tgid = wq->tgid;
break;
}
default:
printk("autofs4_notify_daemon: bad type %d!\n", type);
mutex_unlock(&sbi->wq_mutex);
return;
}
pipe = sbi->pipe;
get_file(pipe);
mutex_unlock(&sbi->wq_mutex);
if (autofs4_write(sbi, pipe, &pkt, pktsz))
autofs4_catatonic_mode(sbi);
fput(pipe);
}
static int autofs4_getpath(struct autofs_sb_info *sbi,
struct dentry *dentry, char **name)
{
struct dentry *root = sbi->sb->s_root;
struct dentry *tmp;
char *buf;
char *p;
int len;
unsigned seq;
rename_retry:
buf = *name;
len = 0;
seq = read_seqbegin(&rename_lock);
rcu_read_lock();
spin_lock(&sbi->fs_lock);
for (tmp = dentry ; tmp != root ; tmp = tmp->d_parent)
len += tmp->d_name.len + 1;
if (!len || --len > NAME_MAX) {
spin_unlock(&sbi->fs_lock);
rcu_read_unlock();
if (read_seqretry(&rename_lock, seq))
goto rename_retry;
return 0;
}
*(buf + len) = '\0';
p = buf + len - dentry->d_name.len;
strncpy(p, dentry->d_name.name, dentry->d_name.len);
for (tmp = dentry->d_parent; tmp != root ; tmp = tmp->d_parent) {
*(--p) = '/';
p -= tmp->d_name.len;
strncpy(p, tmp->d_name.name, tmp->d_name.len);
}
spin_unlock(&sbi->fs_lock);
rcu_read_unlock();
if (read_seqretry(&rename_lock, seq))
goto rename_retry;
return len;
}
static struct autofs_wait_queue *
autofs4_find_wait(struct autofs_sb_info *sbi, struct qstr *qstr)
{
struct autofs_wait_queue *wq;
for (wq = sbi->queues; wq; wq = wq->next) {
if (wq->name.hash == qstr->hash &&
wq->name.len == qstr->len &&
wq->name.name &&
!memcmp(wq->name.name, qstr->name, qstr->len))
break;
}
return wq;
}
autofs4: fix pending mount race Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:19 +00:00
/*
* Check if we have a valid request.
* Returns
* 1 if the request should continue.
* In this case we can return an autofs_wait_queue entry if one is
* found or NULL to idicate a new wait needs to be created.
* 0 or a negative errno if the request shouldn't continue.
*/
static int validate_request(struct autofs_wait_queue **wait,
struct autofs_sb_info *sbi,
struct qstr *qstr,
struct dentry*dentry, enum autofs_notify notify)
{
struct autofs_wait_queue *wq;
struct autofs_info *ino;
if (sbi->catatonic)
return -ENOENT;
autofs4: fix pending mount race Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:19 +00:00
/* Wait in progress, continue; */
wq = autofs4_find_wait(sbi, qstr);
if (wq) {
*wait = wq;
return 1;
}
*wait = NULL;
/* If we don't yet have any info this is a new request */
ino = autofs4_dentry_ino(dentry);
if (!ino)
return 1;
/*
* If we've been asked to wait on an existing expire (NFY_NONE)
* but there is no wait in the queue ...
*/
if (notify == NFY_NONE) {
/*
* Either we've betean the pending expire to post it's
* wait or it finished while we waited on the mutex.
* So we need to wait till either, the wait appears
* or the expire finishes.
*/
while (ino->flags & AUTOFS_INF_EXPIRING) {
mutex_unlock(&sbi->wq_mutex);
schedule_timeout_interruptible(HZ/10);
if (mutex_lock_interruptible(&sbi->wq_mutex))
return -EINTR;
if (sbi->catatonic)
return -ENOENT;
autofs4: fix pending mount race Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:19 +00:00
wq = autofs4_find_wait(sbi, qstr);
if (wq) {
*wait = wq;
return 1;
}
}
/*
* Not ideal but the status has already gone. Of the two
* cases where we wait on NFY_NONE neither depend on the
* return status of the wait.
*/
return 0;
}
/*
* If we've been asked to trigger a mount and the request
* completed while we waited on the mutex ...
*/
if (notify == NFY_MOUNT) {
struct dentry *new = NULL;
int valid = 1;
autofs4: fix pending mount race Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:19 +00:00
/*
autofs4: remove hashed check in validate_wait() The recent ->lookup() deadlock correction required the directory inode mutex to be dropped while waiting for expire completion. We were concerned about side effects from this change and one has been identified. I saw several error messages. They cause autofs to become quite confused and don't really point to the actual problem. Things like: handle_packet_missing_direct:1376: can't find map entry for (43,1827932) which is usually totally fatal (although in this case it wouldn't be except that I treat is as such because it normally is). do_mount_direct: direct trigger not valid or already mounted /test/nested/g3c/s1/ss1 which is recoverable, however if this problem is at play it can cause autofs to become quite confused as to the dependencies in the mount tree because mount triggers end up mounted multiple times. It's hard to accurately check for this over mounting case and automount shouldn't need to if the kernel module is doing its job. There was one other message, similar in consequence of this last one but I can't locate a log example just now. When checking if a mount has already completed prior to adding a new mount request to the wait queue we check if the dentry is hashed and, if so, if it is a mount point. But, if a mount successfully completed while we slept on the wait queue mutex the dentry must exist for the mount to have completed so the test is not really needed. Mounts can also be done on top of a global root dentry, so for the above case, where a mount request completes and the wait queue entry has already been removed, the hashed test returning false can cause an incorrect callback to the daemon. Also, d_mountpoint() is not sufficient to check if a mount has completed for the multi-mount case when we don't have a real mount at the base of the tree. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-09 23:26:24 +00:00
* If the dentry was successfully mounted while we slept
* on the wait queue mutex we can return success. If it
* isn't mounted (doesn't have submounts for the case of
* a multi-mount with no mount at it's base) we can
* continue on and create a new request.
*/
if (!IS_ROOT(dentry)) {
if (dentry->d_inode && d_unhashed(dentry)) {
struct dentry *parent = dentry->d_parent;
new = d_lookup(parent, &dentry->d_name);
if (new)
dentry = new;
}
}
autofs4: remove hashed check in validate_wait() The recent ->lookup() deadlock correction required the directory inode mutex to be dropped while waiting for expire completion. We were concerned about side effects from this change and one has been identified. I saw several error messages. They cause autofs to become quite confused and don't really point to the actual problem. Things like: handle_packet_missing_direct:1376: can't find map entry for (43,1827932) which is usually totally fatal (although in this case it wouldn't be except that I treat is as such because it normally is). do_mount_direct: direct trigger not valid or already mounted /test/nested/g3c/s1/ss1 which is recoverable, however if this problem is at play it can cause autofs to become quite confused as to the dependencies in the mount tree because mount triggers end up mounted multiple times. It's hard to accurately check for this over mounting case and automount shouldn't need to if the kernel module is doing its job. There was one other message, similar in consequence of this last one but I can't locate a log example just now. When checking if a mount has already completed prior to adding a new mount request to the wait queue we check if the dentry is hashed and, if so, if it is a mount point. But, if a mount successfully completed while we slept on the wait queue mutex the dentry must exist for the mount to have completed so the test is not really needed. Mounts can also be done on top of a global root dentry, so for the above case, where a mount request completes and the wait queue entry has already been removed, the hashed test returning false can cause an incorrect callback to the daemon. Also, d_mountpoint() is not sufficient to check if a mount has completed for the multi-mount case when we don't have a real mount at the base of the tree. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-09 23:26:24 +00:00
if (have_submounts(dentry))
valid = 0;
if (new)
dput(new);
return valid;
autofs4: fix pending mount race Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:19 +00:00
}
return 1;
}
int autofs4_wait(struct autofs_sb_info *sbi, struct dentry *dentry,
enum autofs_notify notify)
{
struct autofs_wait_queue *wq;
struct qstr qstr;
char *name;
autofs4: fix pending mount race Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:19 +00:00
int status, ret, type;
/* In catatonic mode, we don't wait for nobody */
if (sbi->catatonic)
return -ENOENT;
autofs4: fix pending mount race Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:19 +00:00
autofs4: indirect dentry must almost always be positive We have been seeing mount requests comming to the automount daemon for keys of the form "<map key>/<non key directory>" which are lookups for invalid map keys. But we can check for this in the kernel module and return a fail immediately, without having to send a request to the daemon. It is possible to recognise these requests are invalid based on whether the request dentry is negative and its relation to the autofs file system root. For example, given the indirect multi-mount map entry: idm1 \ /mm1 <server>:/<path1> /mm2 <server>:/<path2> For a request to mount idm1, IS_ROOT((idm1)->d_parent) will be always be true and the dentry may be negative. But directories idm1/mm1 and idm1/mm2 will always be created as part of the mount request for idm1. So any mount request within idm1 itself must have a positive dentry otherwise the map key is invalid. In version 4 these multi-mount entries are all mounted and umounted as a single request and in version 5 the directories idm1/mm1 and idm1/mm2 are created and an autofs fs mounted on them to act as a mount trigger so the above is also true. This also holds true for the autofs version 4 pseudo direct mount feature. When this feature is used without the "--ghost" option automount(8) will create internal submounts as we go down the map key paths which are essentially normal indirect mounts for which the above holds. If the "--ghost" option is given the directories for map keys are created at daemon startup so valid map entries correspond to postive dentries in the autofs fs. autofs version 5 direct mount maps are similar except that the IS_ROOT check is not needed. This has been addressed in a previous patch tittled "autofs4 - detect invalid direct mount requests". For example, given the direct multi-mount map entry: /test/dm1 \ /mm1 <server>:/<path1> /mm2 <server>:/<path2> An autofs fs is mounted on /test/dm1 as a trigger mount and when a mount is triggered for /test/dm1, the multi-mount offset directories /test/dm1/mm1 and /test/dm1/mm2 are created and an autofs fs is mounted on them to act as mount triggers. So valid direct mount requests must always have a positive dentry if they correspond to a valid map entry. Signed-off-by: Ian Kent <raven@themaw.net> Acked-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:23 +00:00
if (!dentry->d_inode) {
/*
* A wait for a negative dentry is invalid for certain
* cases. A direct or offset mount "always" has its mount
* point directory created and so the request dentry must
* be positive or the map key doesn't exist. The situation
* is very similar for indirect mounts except only dentrys
* in the root of the autofs file system may be negative.
*/
if (autofs_type_trigger(sbi->type))
autofs4: indirect dentry must almost always be positive We have been seeing mount requests comming to the automount daemon for keys of the form "<map key>/<non key directory>" which are lookups for invalid map keys. But we can check for this in the kernel module and return a fail immediately, without having to send a request to the daemon. It is possible to recognise these requests are invalid based on whether the request dentry is negative and its relation to the autofs file system root. For example, given the indirect multi-mount map entry: idm1 \ /mm1 <server>:/<path1> /mm2 <server>:/<path2> For a request to mount idm1, IS_ROOT((idm1)->d_parent) will be always be true and the dentry may be negative. But directories idm1/mm1 and idm1/mm2 will always be created as part of the mount request for idm1. So any mount request within idm1 itself must have a positive dentry otherwise the map key is invalid. In version 4 these multi-mount entries are all mounted and umounted as a single request and in version 5 the directories idm1/mm1 and idm1/mm2 are created and an autofs fs mounted on them to act as a mount trigger so the above is also true. This also holds true for the autofs version 4 pseudo direct mount feature. When this feature is used without the "--ghost" option automount(8) will create internal submounts as we go down the map key paths which are essentially normal indirect mounts for which the above holds. If the "--ghost" option is given the directories for map keys are created at daemon startup so valid map entries correspond to postive dentries in the autofs fs. autofs version 5 direct mount maps are similar except that the IS_ROOT check is not needed. This has been addressed in a previous patch tittled "autofs4 - detect invalid direct mount requests". For example, given the direct multi-mount map entry: /test/dm1 \ /mm1 <server>:/<path1> /mm2 <server>:/<path2> An autofs fs is mounted on /test/dm1 as a trigger mount and when a mount is triggered for /test/dm1, the multi-mount offset directories /test/dm1/mm1 and /test/dm1/mm2 are created and an autofs fs is mounted on them to act as mount triggers. So valid direct mount requests must always have a positive dentry if they correspond to a valid map entry. Signed-off-by: Ian Kent <raven@themaw.net> Acked-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:23 +00:00
return -ENOENT;
else if (!IS_ROOT(dentry->d_parent))
return -ENOENT;
}
name = kmalloc(NAME_MAX + 1, GFP_KERNEL);
if (!name)
return -ENOMEM;
/* If this is a direct mount request create a dummy name */
if (IS_ROOT(dentry) && autofs_type_trigger(sbi->type))
qstr.len = sprintf(name, "%p", dentry);
else {
qstr.len = autofs4_getpath(sbi, dentry, &name);
if (!qstr.len) {
kfree(name);
return -ENOENT;
}
}
qstr.name = name;
qstr.hash = full_name_hash(name, qstr.len);
if (mutex_lock_interruptible(&sbi->wq_mutex)) {
kfree(qstr.name);
return -EINTR;
}
autofs4: fix pending mount race Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:19 +00:00
ret = validate_request(&wq, sbi, &qstr, dentry, notify);
if (ret <= 0) {
if (ret != -EINTR)
mutex_unlock(&sbi->wq_mutex);
autofs4: fix pending mount race Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 04:30:19 +00:00
kfree(qstr.name);
return ret;
}
if (!wq) {
/* Create a new wait queue */
wq = kmalloc(sizeof(struct autofs_wait_queue),GFP_KERNEL);
if (!wq) {
kfree(qstr.name);
mutex_unlock(&sbi->wq_mutex);
return -ENOMEM;
}
wq->wait_queue_token = autofs4_next_wait_queue;
if (++autofs4_next_wait_queue == 0)
autofs4_next_wait_queue = 1;
wq->next = sbi->queues;
sbi->queues = wq;
init_waitqueue_head(&wq->queue);
memcpy(&wq->name, &qstr, sizeof(struct qstr));
wq->dev = autofs4_get_dev(sbi);
wq->ino = autofs4_get_ino(sbi);
wq->uid = current_uid();
wq->gid = current_gid();
wq->pid = current->pid;
wq->tgid = current->tgid;
wq->status = -EINTR; /* Status return if interrupted */
wq->wait_ctr = 2;
mutex_unlock(&sbi->wq_mutex);
if (sbi->version < 5) {
if (notify == NFY_MOUNT)
type = autofs_ptype_missing;
else
type = autofs_ptype_expire_multi;
} else {
if (notify == NFY_MOUNT)
type = autofs_type_trigger(sbi->type) ?
autofs_ptype_missing_direct :
autofs_ptype_missing_indirect;
else
type = autofs_type_trigger(sbi->type) ?
autofs_ptype_expire_direct :
autofs_ptype_expire_indirect;
}
DPRINTK("new wait id = 0x%08lx, name = %.*s, nfy=%d\n",
(unsigned long) wq->wait_queue_token, wq->name.len,
wq->name.name, notify);
/* autofs4_notify_daemon() may block */
autofs4_notify_daemon(sbi, wq, type);
} else {
wq->wait_ctr++;
mutex_unlock(&sbi->wq_mutex);
kfree(qstr.name);
DPRINTK("existing wait id = 0x%08lx, name = %.*s, nfy=%d",
(unsigned long) wq->wait_queue_token, wq->name.len,
wq->name.name, notify);
}
/*
* wq->name.name is NULL iff the lock is already released
* or the mount has been made catatonic.
*/
if (wq->name.name) {
/* Block all but "shutdown" signals while waiting */
sigset_t oldset;
unsigned long irqflags;
spin_lock_irqsave(&current->sighand->siglock, irqflags);
oldset = current->blocked;
siginitsetinv(&current->blocked, SHUTDOWN_SIGS & ~oldset.sig[0]);
recalc_sigpending();
spin_unlock_irqrestore(&current->sighand->siglock, irqflags);
wait_event_interruptible(wq->queue, wq->name.name == NULL);
spin_lock_irqsave(&current->sighand->siglock, irqflags);
current->blocked = oldset;
recalc_sigpending();
spin_unlock_irqrestore(&current->sighand->siglock, irqflags);
} else {
DPRINTK("skipped sleeping");
}
status = wq->status;
/*
* For direct and offset mounts we need to track the requester's
* uid and gid in the dentry info struct. This is so it can be
* supplied, on request, by the misc device ioctl interface.
* This is needed during daemon resatart when reconnecting
* to existing, active, autofs mounts. The uid and gid (and
* related string values) may be used for macro substitution
* in autofs mount maps.
*/
if (!status) {
struct autofs_info *ino;
struct dentry *de = NULL;
/* direct mount or browsable map */
ino = autofs4_dentry_ino(dentry);
if (!ino) {
/* If not lookup actual dentry used */
de = d_lookup(dentry->d_parent, &dentry->d_name);
if (de)
ino = autofs4_dentry_ino(de);
}
/* Set mount requester */
if (ino) {
spin_lock(&sbi->fs_lock);
ino->uid = wq->uid;
ino->gid = wq->gid;
spin_unlock(&sbi->fs_lock);
}
if (de)
dput(de);
}
/* Are we the last process to need status? */
mutex_lock(&sbi->wq_mutex);
if (!--wq->wait_ctr)
kfree(wq);
mutex_unlock(&sbi->wq_mutex);
return status;
}
int autofs4_wait_release(struct autofs_sb_info *sbi, autofs_wqt_t wait_queue_token, int status)
{
struct autofs_wait_queue *wq, **wql;
mutex_lock(&sbi->wq_mutex);
for (wql = &sbi->queues; (wq = *wql) != NULL; wql = &wq->next) {
if (wq->wait_queue_token == wait_queue_token)
break;
}
if (!wq) {
mutex_unlock(&sbi->wq_mutex);
return -EINVAL;
}
*wql = wq->next; /* Unlink from chain */
kfree(wq->name.name);
wq->name.name = NULL; /* Do not wait on this queue */
wq->status = status;
wake_up_interruptible(&wq->queue);
if (!--wq->wait_ctr)
kfree(wq);
mutex_unlock(&sbi->wq_mutex);
return 0;
}