linux/kernel/audit_tree.c

991 lines
23 KiB
C
Raw Normal View History

#include "audit.h"
#include <linux/fsnotify_backend.h>
#include <linux/namei.h>
#include <linux/mount.h>
#include <linux/kthread.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>
struct audit_tree;
struct audit_chunk;
struct audit_tree {
atomic_t count;
int goner;
struct audit_chunk *root;
struct list_head chunks;
struct list_head rules;
struct list_head list;
struct list_head same_root;
struct rcu_head head;
char pathname[];
};
struct audit_chunk {
struct list_head hash;
struct fsnotify_mark mark;
struct list_head trees; /* with root here */
int dead;
int count;
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
atomic_long_t refs;
struct rcu_head head;
struct node {
struct list_head list;
struct audit_tree *owner;
unsigned index; /* index; upper bit indicates 'will prune' */
} owners[];
};
static LIST_HEAD(tree_list);
static LIST_HEAD(prune_list);
static struct task_struct *prune_thread;
/*
* One struct chunk is attached to each inode of interest.
* We replace struct chunk on tagging/untagging.
* Rules have pointer to struct audit_tree.
* Rules have struct list_head rlist forming a list of rules over
* the same tree.
* References to struct chunk are collected at audit_inode{,_child}()
* time and used in AUDIT_TREE rule matching.
* These references are dropped at the same time we are calling
* audit_free_names(), etc.
*
* Cyclic lists galore:
* tree.chunks anchors chunk.owners[].list hash_lock
* tree.rules anchors rule.rlist audit_filter_mutex
* chunk.trees anchors tree.same_root hash_lock
* chunk.hash is a hash with middle bits of watch.inode as
* a hash function. RCU, hash_lock
*
* tree is refcounted; one reference for "some rules on rules_list refer to
* it", one for each chunk with pointer to it.
*
* chunk is refcounted by embedded fsnotify_mark + .refs (non-zero refcount
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
* of watch contributes 1 to .refs).
*
* node.index allows to get from node.list to containing chunk.
* MSB of that sucker is stolen to mark taggings that we might have to
* revert - several operations have very unpleasant cleanup logics and
* that makes a difference. Some.
*/
static struct fsnotify_group *audit_tree_group;
static struct audit_tree *alloc_tree(const char *s)
{
struct audit_tree *tree;
tree = kmalloc(sizeof(struct audit_tree) + strlen(s) + 1, GFP_KERNEL);
if (tree) {
atomic_set(&tree->count, 1);
tree->goner = 0;
INIT_LIST_HEAD(&tree->chunks);
INIT_LIST_HEAD(&tree->rules);
INIT_LIST_HEAD(&tree->list);
INIT_LIST_HEAD(&tree->same_root);
tree->root = NULL;
strcpy(tree->pathname, s);
}
return tree;
}
static inline void get_tree(struct audit_tree *tree)
{
atomic_inc(&tree->count);
}
static inline void put_tree(struct audit_tree *tree)
{
if (atomic_dec_and_test(&tree->count))
kfree_rcu(tree, head);
}
/* to avoid bringing the entire thing in audit.h */
const char *audit_tree_path(struct audit_tree *tree)
{
return tree->pathname;
}
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
static void free_chunk(struct audit_chunk *chunk)
{
int i;
for (i = 0; i < chunk->count; i++) {
if (chunk->owners[i].owner)
put_tree(chunk->owners[i].owner);
}
kfree(chunk);
}
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
void audit_put_chunk(struct audit_chunk *chunk)
{
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
if (atomic_long_dec_and_test(&chunk->refs))
free_chunk(chunk);
}
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
static void __put_chunk(struct rcu_head *rcu)
{
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
struct audit_chunk *chunk = container_of(rcu, struct audit_chunk, head);
audit_put_chunk(chunk);
}
static void audit_tree_destroy_watch(struct fsnotify_mark *entry)
{
struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
call_rcu(&chunk->head, __put_chunk);
}
static struct audit_chunk *alloc_chunk(int count)
{
struct audit_chunk *chunk;
size_t size;
int i;
size = offsetof(struct audit_chunk, owners) + count * sizeof(struct node);
chunk = kzalloc(size, GFP_KERNEL);
if (!chunk)
return NULL;
INIT_LIST_HEAD(&chunk->hash);
INIT_LIST_HEAD(&chunk->trees);
chunk->count = count;
atomic_long_set(&chunk->refs, 1);
for (i = 0; i < count; i++) {
INIT_LIST_HEAD(&chunk->owners[i].list);
chunk->owners[i].index = i;
}
fsnotify_init_mark(&chunk->mark, audit_tree_destroy_watch);
chunk->mark.mask = FS_IN_IGNORED;
return chunk;
}
enum {HASH_SIZE = 128};
static struct list_head chunk_hash_heads[HASH_SIZE];
static __cacheline_aligned_in_smp DEFINE_SPINLOCK(hash_lock);
static inline struct list_head *chunk_hash(const struct inode *inode)
{
unsigned long n = (unsigned long)inode / L1_CACHE_BYTES;
return chunk_hash_heads + n % HASH_SIZE;
}
/* hash_lock & entry->lock is held by caller */
static void insert_hash(struct audit_chunk *chunk)
{
struct fsnotify_mark *entry = &chunk->mark;
struct list_head *list;
if (!entry->inode)
return;
list = chunk_hash(entry->inode);
list_add_rcu(&chunk->hash, list);
}
/* called under rcu_read_lock */
struct audit_chunk *audit_tree_lookup(const struct inode *inode)
{
struct list_head *list = chunk_hash(inode);
struct audit_chunk *p;
list_for_each_entry_rcu(p, list, hash) {
/* mark.inode may have gone NULL, but who cares? */
if (p->mark.inode == inode) {
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
atomic_long_inc(&p->refs);
return p;
}
}
return NULL;
}
int audit_tree_match(struct audit_chunk *chunk, struct audit_tree *tree)
{
int n;
for (n = 0; n < chunk->count; n++)
if (chunk->owners[n].owner == tree)
return 1;
return 0;
}
/* tagging and untagging inodes with trees */
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
static struct audit_chunk *find_chunk(struct node *p)
{
int index = p->index & ~(1U<<31);
p -= index;
return container_of(p, struct audit_chunk, owners[0]);
}
static void untag_chunk(struct node *p)
{
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
struct audit_chunk *chunk = find_chunk(p);
struct fsnotify_mark *entry = &chunk->mark;
struct audit_chunk *new = NULL;
struct audit_tree *owner;
int size = chunk->count - 1;
int i, j;
fsnotify_get_mark(entry);
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
spin_unlock(&hash_lock);
if (size)
new = alloc_chunk(size);
spin_lock(&entry->lock);
if (chunk->dead || !entry->inode) {
spin_unlock(&entry->lock);
if (new)
free_chunk(new);
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
goto out;
}
owner = p->owner;
if (!size) {
chunk->dead = 1;
spin_lock(&hash_lock);
list_del_init(&chunk->trees);
if (owner->root == chunk)
owner->root = NULL;
list_del_init(&p->list);
list_del_rcu(&chunk->hash);
spin_unlock(&hash_lock);
spin_unlock(&entry->lock);
fsnotify_destroy_mark(entry, audit_tree_group);
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
goto out;
}
if (!new)
goto Fallback;
fsnotify_duplicate_mark(&new->mark, entry);
if (fsnotify_add_mark(&new->mark, new->mark.group, new->mark.inode, NULL, 1)) {
fsnotify_put_mark(&new->mark);
goto Fallback;
}
chunk->dead = 1;
spin_lock(&hash_lock);
list_replace_init(&chunk->trees, &new->trees);
if (owner->root == chunk) {
list_del_init(&owner->same_root);
owner->root = NULL;
}
for (i = j = 0; j <= size; i++, j++) {
struct audit_tree *s;
if (&chunk->owners[j] == p) {
list_del_init(&p->list);
i--;
continue;
}
s = chunk->owners[j].owner;
new->owners[i].owner = s;
new->owners[i].index = chunk->owners[j].index - j + i;
if (!s) /* result of earlier fallback */
continue;
get_tree(s);
list_replace_init(&chunk->owners[j].list, &new->owners[i].list);
}
list_replace_rcu(&chunk->hash, &new->hash);
list_for_each_entry(owner, &new->trees, same_root)
owner->root = new;
spin_unlock(&hash_lock);
spin_unlock(&entry->lock);
fsnotify_destroy_mark(entry, audit_tree_group);
fsnotify_put_mark(&new->mark); /* drop initial reference */
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
goto out;
Fallback:
// do the best we can
spin_lock(&hash_lock);
if (owner->root == chunk) {
list_del_init(&owner->same_root);
owner->root = NULL;
}
list_del_init(&p->list);
p->owner = NULL;
put_tree(owner);
spin_unlock(&hash_lock);
spin_unlock(&entry->lock);
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
out:
fsnotify_put_mark(entry);
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
spin_lock(&hash_lock);
}
static int create_chunk(struct inode *inode, struct audit_tree *tree)
{
struct fsnotify_mark *entry;
struct audit_chunk *chunk = alloc_chunk(1);
if (!chunk)
return -ENOMEM;
entry = &chunk->mark;
if (fsnotify_add_mark(entry, audit_tree_group, inode, NULL, 0)) {
fsnotify_put_mark(entry);
return -ENOSPC;
}
spin_lock(&entry->lock);
spin_lock(&hash_lock);
if (tree->goner) {
spin_unlock(&hash_lock);
chunk->dead = 1;
spin_unlock(&entry->lock);
fsnotify_destroy_mark(entry, audit_tree_group);
fsnotify_put_mark(entry);
return 0;
}
chunk->owners[0].index = (1U << 31);
chunk->owners[0].owner = tree;
get_tree(tree);
list_add(&chunk->owners[0].list, &tree->chunks);
if (!tree->root) {
tree->root = chunk;
list_add(&tree->same_root, &chunk->trees);
}
insert_hash(chunk);
spin_unlock(&hash_lock);
spin_unlock(&entry->lock);
fsnotify_put_mark(entry); /* drop initial reference */
return 0;
}
/* the first tagged inode becomes root of tree */
static int tag_chunk(struct inode *inode, struct audit_tree *tree)
{
struct fsnotify_mark *old_entry, *chunk_entry;
struct audit_tree *owner;
struct audit_chunk *chunk, *old;
struct node *p;
int n;
old_entry = fsnotify_find_inode_mark(audit_tree_group, inode);
if (!old_entry)
return create_chunk(inode, tree);
old = container_of(old_entry, struct audit_chunk, mark);
/* are we already there? */
spin_lock(&hash_lock);
for (n = 0; n < old->count; n++) {
if (old->owners[n].owner == tree) {
spin_unlock(&hash_lock);
fsnotify_put_mark(old_entry);
return 0;
}
}
spin_unlock(&hash_lock);
chunk = alloc_chunk(old->count + 1);
if (!chunk) {
fsnotify_put_mark(old_entry);
return -ENOMEM;
}
chunk_entry = &chunk->mark;
spin_lock(&old_entry->lock);
if (!old_entry->inode) {
/* old_entry is being shot, lets just lie */
spin_unlock(&old_entry->lock);
fsnotify_put_mark(old_entry);
free_chunk(chunk);
return -ENOENT;
}
fsnotify_duplicate_mark(chunk_entry, old_entry);
if (fsnotify_add_mark(chunk_entry, chunk_entry->group, chunk_entry->inode, NULL, 1)) {
spin_unlock(&old_entry->lock);
fsnotify_put_mark(chunk_entry);
fsnotify_put_mark(old_entry);
return -ENOSPC;
}
/* even though we hold old_entry->lock, this is safe since chunk_entry->lock could NEVER have been grabbed before */
spin_lock(&chunk_entry->lock);
spin_lock(&hash_lock);
/* we now hold old_entry->lock, chunk_entry->lock, and hash_lock */
if (tree->goner) {
spin_unlock(&hash_lock);
chunk->dead = 1;
spin_unlock(&chunk_entry->lock);
spin_unlock(&old_entry->lock);
fsnotify_destroy_mark(chunk_entry, audit_tree_group);
fsnotify_put_mark(chunk_entry);
fsnotify_put_mark(old_entry);
return 0;
}
list_replace_init(&old->trees, &chunk->trees);
for (n = 0, p = chunk->owners; n < old->count; n++, p++) {
struct audit_tree *s = old->owners[n].owner;
p->owner = s;
p->index = old->owners[n].index;
if (!s) /* result of fallback in untag */
continue;
get_tree(s);
list_replace_init(&old->owners[n].list, &p->list);
}
p->index = (chunk->count - 1) | (1U<<31);
p->owner = tree;
get_tree(tree);
list_add(&p->list, &tree->chunks);
list_replace_rcu(&old->hash, &chunk->hash);
list_for_each_entry(owner, &chunk->trees, same_root)
owner->root = chunk;
old->dead = 1;
if (!tree->root) {
tree->root = chunk;
list_add(&tree->same_root, &chunk->trees);
}
spin_unlock(&hash_lock);
spin_unlock(&chunk_entry->lock);
spin_unlock(&old_entry->lock);
fsnotify_destroy_mark(old_entry, audit_tree_group);
fsnotify_put_mark(chunk_entry); /* drop initial reference */
fsnotify_put_mark(old_entry); /* pair to fsnotify_find mark_entry */
return 0;
}
static void audit_tree_log_remove_rule(struct audit_krule *rule)
{
struct audit_buffer *ab;
ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
if (unlikely(!ab))
return;
audit_log_format(ab, "op=");
audit_log_string(ab, "remove_rule");
audit_log_format(ab, " dir=");
audit_log_untrustedstring(ab, rule->tree->pathname);
audit_log_key(ab, rule->filterkey);
audit_log_format(ab, " list=%d res=1", rule->listnr);
audit_log_end(ab);
}
static void kill_rules(struct audit_tree *tree)
{
struct audit_krule *rule, *next;
struct audit_entry *entry;
list_for_each_entry_safe(rule, next, &tree->rules, rlist) {
entry = container_of(rule, struct audit_entry, rule);
list_del_init(&rule->rlist);
if (rule->tree) {
/* not a half-baked one */
audit_tree_log_remove_rule(rule);
if (entry->rule.exe)
audit_remove_mark(entry->rule.exe);
rule->tree = NULL;
list_del_rcu(&entry->list);
list_del(&entry->rule.list);
call_rcu(&entry->rcu, audit_free_rule_rcu);
}
}
}
/*
* finish killing struct audit_tree
*/
static void prune_one(struct audit_tree *victim)
{
spin_lock(&hash_lock);
while (!list_empty(&victim->chunks)) {
struct node *p;
p = list_entry(victim->chunks.next, struct node, list);
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
untag_chunk(p);
}
spin_unlock(&hash_lock);
put_tree(victim);
}
/* trim the uncommitted chunks from tree */
static void trim_marked(struct audit_tree *tree)
{
struct list_head *p, *q;
spin_lock(&hash_lock);
if (tree->goner) {
spin_unlock(&hash_lock);
return;
}
/* reorder */
for (p = tree->chunks.next; p != &tree->chunks; p = q) {
struct node *node = list_entry(p, struct node, list);
q = p->next;
if (node->index & (1U<<31)) {
list_del_init(p);
list_add(p, &tree->chunks);
}
}
while (!list_empty(&tree->chunks)) {
struct node *node;
node = list_entry(tree->chunks.next, struct node, list);
/* have we run out of marked? */
if (!(node->index & (1U<<31)))
break;
Fix inotify watch removal/umount races Inotify watch removals suck violently. To kick the watch out we need (in this order) inode->inotify_mutex and ih->mutex. That's fine if we have a hold on inode; however, for all other cases we need to make damn sure we don't race with umount. We can *NOT* just grab a reference to a watch - inotify_unmount_inodes() will happily sail past it and we'll end with reference to inode potentially outliving its superblock. Ideally we just want to grab an active reference to superblock if we can; that will make sure we won't go into inotify_umount_inodes() until we are done. Cleanup is just deactivate_super(). However, that leaves a messy case - what if we *are* racing with umount() and active references to superblock can't be acquired anymore? We can bump ->s_count, grab ->s_umount, which will almost certainly wait until the superblock is shut down and the watch in question is pining for fjords. That's fine, but there is a problem - we might have hit the window between ->s_active getting to 0 / ->s_count - below S_BIAS (i.e. the moment when superblock is past the point of no return and is heading for shutdown) and the moment when deactivate_super() acquires ->s_umount. We could just do drop_super() yield() and retry, but that's rather antisocial and this stuff is luser-triggerable. OTOH, having grabbed ->s_umount and having found that we'd got there first (i.e. that ->s_root is non-NULL) we know that we won't race with inotify_umount_inodes(). So we could grab a reference to watch and do the rest as above, just with drop_super() instead of deactivate_super(), right? Wrong. We had to drop ih->mutex before we could grab ->s_umount. So the watch could've been gone already. That still can be dealt with - we need to save watch->wd, do idr_find() and compare its result with our pointer. If they match, we either have the damn thing still alive or we'd lost not one but two races at once, the watch had been killed and a new one got created with the same ->wd at the same address. That couldn't have happened in inotify_destroy(), but inotify_rm_wd() could run into that. Still, "new one got created" is not a problem - we have every right to kill it or leave it alone, whatever's more convenient. So we can use idr_find(...) == watch && watch->inode->i_sb == sb as "grab it and kill it" check. If it's been our original watch, we are fine, if it's a newcomer - nevermind, just pretend that we'd won the race and kill the fscker anyway; we are safe since we know that its superblock won't be going away. And yes, this is far beyond mere "not very pretty"; so's the entire concept of inotify to start with. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Greg KH <greg@kroah.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-15 01:15:43 +00:00
untag_chunk(node);
}
if (!tree->root && !tree->goner) {
tree->goner = 1;
spin_unlock(&hash_lock);
mutex_lock(&audit_filter_mutex);
kill_rules(tree);
list_del_init(&tree->list);
mutex_unlock(&audit_filter_mutex);
prune_one(tree);
} else {
spin_unlock(&hash_lock);
}
}
static void audit_schedule_prune(void);
/* called with audit_filter_mutex */
int audit_remove_tree_rule(struct audit_krule *rule)
{
struct audit_tree *tree;
tree = rule->tree;
if (tree) {
spin_lock(&hash_lock);
list_del_init(&rule->rlist);
if (list_empty(&tree->rules) && !tree->goner) {
tree->root = NULL;
list_del_init(&tree->same_root);
tree->goner = 1;
list_move(&tree->list, &prune_list);
rule->tree = NULL;
spin_unlock(&hash_lock);
audit_schedule_prune();
return 1;
}
rule->tree = NULL;
spin_unlock(&hash_lock);
return 1;
}
return 0;
}
static int compare_root(struct vfsmount *mnt, void *arg)
{
return d_backing_inode(mnt->mnt_root) == arg;
}
void audit_trim_trees(void)
{
struct list_head cursor;
mutex_lock(&audit_filter_mutex);
list_add(&cursor, &tree_list);
while (cursor.next != &tree_list) {
struct audit_tree *tree;
struct path path;
struct vfsmount *root_mnt;
struct node *node;
int err;
tree = container_of(cursor.next, struct audit_tree, list);
get_tree(tree);
list_del(&cursor);
list_add(&cursor, &tree->list);
mutex_unlock(&audit_filter_mutex);
err = kern_path(tree->pathname, 0, &path);
if (err)
goto skip_it;
root_mnt = collect_mounts(&path);
path_put(&path);
if (IS_ERR(root_mnt))
goto skip_it;
spin_lock(&hash_lock);
list_for_each_entry(node, &tree->chunks, list) {
struct audit_chunk *chunk = find_chunk(node);
/* this could be NULL if the watch is dying else where... */
struct inode *inode = chunk->mark.inode;
node->index |= 1U<<31;
if (iterate_mounts(compare_root, inode, root_mnt))
node->index &= ~(1U<<31);
}
spin_unlock(&hash_lock);
trim_marked(tree);
drop_collected_mounts(root_mnt);
skip_it:
put_tree(tree);
mutex_lock(&audit_filter_mutex);
}
list_del(&cursor);
mutex_unlock(&audit_filter_mutex);
}
int audit_make_tree(struct audit_krule *rule, char *pathname, u32 op)
{
if (pathname[0] != '/' ||
rule->listnr != AUDIT_FILTER_EXIT ||
op != Audit_equal ||
rule->inode_f || rule->watch || rule->tree)
return -EINVAL;
rule->tree = alloc_tree(pathname);
if (!rule->tree)
return -ENOMEM;
return 0;
}
void audit_put_tree(struct audit_tree *tree)
{
put_tree(tree);
}
static int tag_mount(struct vfsmount *mnt, void *arg)
{
return tag_chunk(d_backing_inode(mnt->mnt_root), arg);
}
/*
* That gets run when evict_chunk() ends up needing to kill audit_tree.
* Runs from a separate thread.
*/
static int prune_tree_thread(void *unused)
{
for (;;) {
set_current_state(TASK_INTERRUPTIBLE);
if (list_empty(&prune_list))
schedule();
__set_current_state(TASK_RUNNING);
mutex_lock(&audit_cmd_mutex);
mutex_lock(&audit_filter_mutex);
while (!list_empty(&prune_list)) {
struct audit_tree *victim;
victim = list_entry(prune_list.next,
struct audit_tree, list);
list_del_init(&victim->list);
mutex_unlock(&audit_filter_mutex);
prune_one(victim);
mutex_lock(&audit_filter_mutex);
}
mutex_unlock(&audit_filter_mutex);
mutex_unlock(&audit_cmd_mutex);
}
return 0;
}
static int audit_launch_prune(void)
{
if (prune_thread)
return 0;
prune_thread = kthread_create(prune_tree_thread, NULL,
"audit_prune_tree");
if (IS_ERR(prune_thread)) {
pr_err("cannot start thread audit_prune_tree");
prune_thread = NULL;
return -ENOMEM;
} else {
wake_up_process(prune_thread);
return 0;
}
}
/* called with audit_filter_mutex */
int audit_add_tree_rule(struct audit_krule *rule)
{
struct audit_tree *seed = rule->tree, *tree;
struct path path;
struct vfsmount *mnt;
int err;
kernel/audit_tree.c:audit_add_tree_rule(): protect `rule' from kill_rules() audit_add_tree_rule() must set 'rule->tree = NULL;' firstly, to protect the rule itself freed in kill_rules(). The reason is when it is killed, the 'rule' itself may have already released, we should not access it. one example: we add a rule to an inode, just at the same time the other task is deleting this inode. The work flow for adding a rule: audit_receive() -> (need audit_cmd_mutex lock) audit_receive_skb() -> audit_receive_msg() -> audit_receive_filter() -> audit_add_rule() -> audit_add_tree_rule() -> (need audit_filter_mutex lock) ... unlock audit_filter_mutex get_tree() ... iterate_mounts() -> (iterate all related inodes) tag_mount() -> tag_trunk() -> create_trunk() -> (assume it is 1st rule) fsnotify_add_mark() -> fsnotify_add_inode_mark() -> (add mark to inode->i_fsnotify_marks) ... get_tree(); (each inode will get one) ... lock audit_filter_mutex The work flow for deleting an inode: __destroy_inode() -> fsnotify_inode_delete() -> __fsnotify_inode_delete() -> fsnotify_clear_marks_by_inode() -> (get mark from inode->i_fsnotify_marks) fsnotify_destroy_mark() -> fsnotify_destroy_mark_locked() -> audit_tree_freeing_mark() -> evict_chunk() -> ... tree->goner = 1 ... kill_rules() -> (assume current->audit_context == NULL) call_rcu() -> (rule->tree != NULL) audit_free_rule_rcu() -> audit_free_rule() ... audit_schedule_prune() -> (assume current->audit_context == NULL) kthread_run() -> (need audit_cmd_mutex and audit_filter_mutex lock) prune_one() -> (delete it from prue_list) put_tree(); (match the original get_tree above) Signed-off-by: Chen Gang <gang.chen@asianux.com> Cc: Eric Paris <eparis@redhat.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-06-12 21:05:07 +00:00
rule->tree = NULL;
list_for_each_entry(tree, &tree_list, list) {
if (!strcmp(seed->pathname, tree->pathname)) {
put_tree(seed);
rule->tree = tree;
list_add(&rule->rlist, &tree->rules);
return 0;
}
}
tree = seed;
list_add(&tree->list, &tree_list);
list_add(&rule->rlist, &tree->rules);
/* do not set rule->tree yet */
mutex_unlock(&audit_filter_mutex);
if (unlikely(!prune_thread)) {
err = audit_launch_prune();
if (err)
goto Err;
}
err = kern_path(tree->pathname, 0, &path);
if (err)
goto Err;
mnt = collect_mounts(&path);
path_put(&path);
if (IS_ERR(mnt)) {
err = PTR_ERR(mnt);
goto Err;
}
get_tree(tree);
err = iterate_mounts(tag_mount, tree, mnt);
drop_collected_mounts(mnt);
if (!err) {
struct node *node;
spin_lock(&hash_lock);
list_for_each_entry(node, &tree->chunks, list)
node->index &= ~(1U<<31);
spin_unlock(&hash_lock);
} else {
trim_marked(tree);
goto Err;
}
mutex_lock(&audit_filter_mutex);
if (list_empty(&rule->rlist)) {
put_tree(tree);
return -ENOENT;
}
rule->tree = tree;
put_tree(tree);
return 0;
Err:
mutex_lock(&audit_filter_mutex);
list_del_init(&tree->list);
list_del_init(&tree->rules);
put_tree(tree);
return err;
}
int audit_tag_tree(char *old, char *new)
{
struct list_head cursor, barrier;
int failed = 0;
struct path path1, path2;
struct vfsmount *tagged;
int err;
err = kern_path(new, 0, &path2);
if (err)
return err;
tagged = collect_mounts(&path2);
path_put(&path2);
if (IS_ERR(tagged))
return PTR_ERR(tagged);
err = kern_path(old, 0, &path1);
if (err) {
drop_collected_mounts(tagged);
return err;
}
mutex_lock(&audit_filter_mutex);
list_add(&barrier, &tree_list);
list_add(&cursor, &barrier);
while (cursor.next != &tree_list) {
struct audit_tree *tree;
int good_one = 0;
tree = container_of(cursor.next, struct audit_tree, list);
get_tree(tree);
list_del(&cursor);
list_add(&cursor, &tree->list);
mutex_unlock(&audit_filter_mutex);
err = kern_path(tree->pathname, 0, &path2);
if (!err) {
good_one = path_is_under(&path1, &path2);
path_put(&path2);
}
if (!good_one) {
put_tree(tree);
mutex_lock(&audit_filter_mutex);
continue;
}
failed = iterate_mounts(tag_mount, tree, tagged);
if (failed) {
put_tree(tree);
mutex_lock(&audit_filter_mutex);
break;
}
mutex_lock(&audit_filter_mutex);
spin_lock(&hash_lock);
if (!tree->goner) {
list_del(&tree->list);
list_add(&tree->list, &tree_list);
}
spin_unlock(&hash_lock);
put_tree(tree);
}
while (barrier.prev != &tree_list) {
struct audit_tree *tree;
tree = container_of(barrier.prev, struct audit_tree, list);
get_tree(tree);
list_del(&tree->list);
list_add(&tree->list, &barrier);
mutex_unlock(&audit_filter_mutex);
if (!failed) {
struct node *node;
spin_lock(&hash_lock);
list_for_each_entry(node, &tree->chunks, list)
node->index &= ~(1U<<31);
spin_unlock(&hash_lock);
} else {
trim_marked(tree);
}
put_tree(tree);
mutex_lock(&audit_filter_mutex);
}
list_del(&barrier);
list_del(&cursor);
mutex_unlock(&audit_filter_mutex);
path_put(&path1);
drop_collected_mounts(tagged);
return failed;
}
static void audit_schedule_prune(void)
{
wake_up_process(prune_thread);
}
/*
* ... and that one is done if evict_chunk() decides to delay until the end
* of syscall. Runs synchronously.
*/
void audit_kill_trees(struct list_head *list)
{
mutex_lock(&audit_cmd_mutex);
mutex_lock(&audit_filter_mutex);
while (!list_empty(list)) {
struct audit_tree *victim;
victim = list_entry(list->next, struct audit_tree, list);
kill_rules(victim);
list_del_init(&victim->list);
mutex_unlock(&audit_filter_mutex);
prune_one(victim);
mutex_lock(&audit_filter_mutex);
}
mutex_unlock(&audit_filter_mutex);
mutex_unlock(&audit_cmd_mutex);
}
/*
* Here comes the stuff asynchronous to auditctl operations
*/
static void evict_chunk(struct audit_chunk *chunk)
{
struct audit_tree *owner;
struct list_head *postponed = audit_killed_trees();
int need_prune = 0;
int n;
if (chunk->dead)
return;
chunk->dead = 1;
mutex_lock(&audit_filter_mutex);
spin_lock(&hash_lock);
while (!list_empty(&chunk->trees)) {
owner = list_entry(chunk->trees.next,
struct audit_tree, same_root);
owner->goner = 1;
owner->root = NULL;
list_del_init(&owner->same_root);
spin_unlock(&hash_lock);
if (!postponed) {
kill_rules(owner);
list_move(&owner->list, &prune_list);
need_prune = 1;
} else {
list_move(&owner->list, postponed);
}
spin_lock(&hash_lock);
}
list_del_rcu(&chunk->hash);
for (n = 0; n < chunk->count; n++)
list_del_init(&chunk->owners[n].list);
spin_unlock(&hash_lock);
mutex_unlock(&audit_filter_mutex);
if (need_prune)
audit_schedule_prune();
}
static int audit_tree_handle_event(struct fsnotify_group *group,
fsnotify: do not share events between notification groups Currently fsnotify framework creates one event structure for each notification event and links this event into all interested notification groups. This is done so that we save memory when several notification groups are interested in the event. However the need for event structure shared between inotify & fanotify bloats the event structure so the result is often higher memory consumption. Another problem is that fsnotify framework keeps path references with outstanding events so that fanotify can return open file descriptors with its events. This has the undesirable effect that filesystem cannot be unmounted while there are outstanding events - a regression for inotify compared to a situation before it was converted to fsnotify framework. For fanotify this problem is hard to avoid and users of fanotify should kind of expect this behavior when they ask for file descriptors from notified files. This patch changes fsnotify and its users to create separate event structure for each group. This allows for much simpler code (~400 lines removed by this patch) and also smaller event structures. For example on 64-bit system original struct fsnotify_event consumes 120 bytes, plus additional space for file name, additional 24 bytes for second and each subsequent group linking the event, and additional 32 bytes for each inotify group for private data. After the conversion inotify event consumes 48 bytes plus space for file name which is considerably less memory unless file names are long and there are several groups interested in the events (both of which are uncommon). Fanotify event fits in 56 bytes after the conversion (fanotify doesn't care about file names so its events don't have to have it allocated). A win unless there are four or more fanotify groups interested in the event. The conversion also solves the problem with unmount when only inotify is used as we don't have to grab path references for inotify events. [hughd@google.com: fanotify: fix corruption preventing startup] Signed-off-by: Jan Kara <jack@suse.cz> Reviewed-by: Christoph Hellwig <hch@lst.de> Cc: Eric Paris <eparis@parisplace.org> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-21 23:48:14 +00:00
struct inode *to_tell,
struct fsnotify_mark *inode_mark,
fsnotify: do not share events between notification groups Currently fsnotify framework creates one event structure for each notification event and links this event into all interested notification groups. This is done so that we save memory when several notification groups are interested in the event. However the need for event structure shared between inotify & fanotify bloats the event structure so the result is often higher memory consumption. Another problem is that fsnotify framework keeps path references with outstanding events so that fanotify can return open file descriptors with its events. This has the undesirable effect that filesystem cannot be unmounted while there are outstanding events - a regression for inotify compared to a situation before it was converted to fsnotify framework. For fanotify this problem is hard to avoid and users of fanotify should kind of expect this behavior when they ask for file descriptors from notified files. This patch changes fsnotify and its users to create separate event structure for each group. This allows for much simpler code (~400 lines removed by this patch) and also smaller event structures. For example on 64-bit system original struct fsnotify_event consumes 120 bytes, plus additional space for file name, additional 24 bytes for second and each subsequent group linking the event, and additional 32 bytes for each inotify group for private data. After the conversion inotify event consumes 48 bytes plus space for file name which is considerably less memory unless file names are long and there are several groups interested in the events (both of which are uncommon). Fanotify event fits in 56 bytes after the conversion (fanotify doesn't care about file names so its events don't have to have it allocated). A win unless there are four or more fanotify groups interested in the event. The conversion also solves the problem with unmount when only inotify is used as we don't have to grab path references for inotify events. [hughd@google.com: fanotify: fix corruption preventing startup] Signed-off-by: Jan Kara <jack@suse.cz> Reviewed-by: Christoph Hellwig <hch@lst.de> Cc: Eric Paris <eparis@parisplace.org> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-21 23:48:14 +00:00
struct fsnotify_mark *vfsmount_mark,
u32 mask, void *data, int data_type,
const unsigned char *file_name, u32 cookie)
{
return 0;
}
static void audit_tree_freeing_mark(struct fsnotify_mark *entry, struct fsnotify_group *group)
{
struct audit_chunk *chunk = container_of(entry, struct audit_chunk, mark);
evict_chunk(chunk);
/*
* We are guaranteed to have at least one reference to the mark from
* either the inode or the caller of fsnotify_destroy_mark().
*/
BUG_ON(atomic_read(&entry->refcnt) < 1);
}
static const struct fsnotify_ops audit_tree_ops = {
.handle_event = audit_tree_handle_event,
.freeing_mark = audit_tree_freeing_mark,
};
static int __init audit_tree_init(void)
{
int i;
audit_tree_group = fsnotify_alloc_group(&audit_tree_ops);
if (IS_ERR(audit_tree_group))
audit_panic("cannot initialize fsnotify group for rectree watches");
for (i = 0; i < HASH_SIZE; i++)
INIT_LIST_HEAD(&chunk_hash_heads[i]);
return 0;
}
__initcall(audit_tree_init);