Fanotify code has its own lock (access_lock) to protect a list of events
waiting for a response from userspace.
However this is somewhat awkward as the same list_head in the event is
protected by notification_lock if it is part of the notification queue
and by access_lock if it is part of the fanotify private queue which
makes it difficult for any reliable checks in the generic code. So make
fanotify use the same lock - notification_lock - for protecting its
private event list.
Link: http://lkml.kernel.org/r/1473797711-14111-6-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Cc: Miklos Szeredi <mszeredi@redhat.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>
fsnotify_flush_notify() and fanotify_release() destroy notification
event while holding notification_mutex.
The destruction of fanotify event includes a path_put() call which may
end up calling into a filesystem to delete an inode if we happen to be
the last holders of dentry reference which happens to be the last holder
of inode reference.
That in turn may violate lock ordering for some filesystems since
notification_mutex is also acquired e. g. during write when generating
fanotify event.
Also this is the only thing that forces notification_mutex to be a
sleeping lock. So drop notification_mutex before destroying a
notification event.
Link: http://lkml.kernel.org/r/1473797711-14111-4-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Miklos Szeredi <mszeredi@redhat.com>
Cc: Lino Sanfilippo <LinoSanfilippo@gmx.de>
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>
fanotify_get_response() calls fsnotify_remove_event() when it finds that
group is being released from fanotify_release() (bypass_perm is set).
However the event it removes need not be only in the group's notification
queue but it can have already moved to access_list (userspace read the
event before closing the fanotify instance fd) which is protected by a
different lock. Thus when fsnotify_remove_event() races with
fanotify_release() operating on access_list, the list can get corrupted.
Fix the problem by moving all the logic removing permission events from
the lists to one place - fanotify_release().
Fixes: 5838d4442b ("fanotify: fix double free of pending permission events")
Link: http://lkml.kernel.org/r/1473797711-14111-3-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Reported-by: Miklos Szeredi <mszeredi@redhat.com>
Tested-by: Miklos Szeredi <mszeredi@redhat.com>
Reviewed-by: Miklos Szeredi <mszeredi@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Inotify instance is destroyed when all references to it are dropped.
That not only means that the corresponding file descriptor needs to be
closed but also that all corresponding instance marks are freed (as each
mark holds a reference to the inotify instance). However marks are
freed only after SRCU period ends which can take some time and thus if
user rapidly creates and frees inotify instances, number of existing
inotify instances can exceed max_user_instances limit although from user
point of view there is always at most one existing instance. Thus
inotify_init() returns EMFILE error which is hard to justify from user
point of view. This problem is exposed by LTP inotify06 testcase on
some machines.
We fix the problem by making sure all group marks are properly freed
while destroying inotify instance. We wait for SRCU period to end in
that path anyway since we have to make sure there is no event being
added to the instance while we are tearing down the instance. So it
takes only some plumbing to allow for marks to be destroyed in that path
as well and not from a dedicated work item.
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Jan Kara <jack@suse.cz>
Reported-by: Xiaoguang Wang <wangxg.fnst@cn.fujitsu.com>
Tested-by: Xiaoguang Wang <wangxg.fnst@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We don't require a dedicated thread for fsnotify cleanup. Switch it
over to a workqueue job instead that runs on the system_unbound_wq.
In the interest of not thrashing the queued job too often when there are
a lot of marks being removed, we delay the reaper job slightly when
queueing it, to allow several to gather on the list.
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Tested-by: Eryu Guan <guaneryu@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Eric Paris <eparis@parisplace.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This reverts commit c510eff6be ("fsnotify: destroy marks with
call_srcu instead of dedicated thread").
Eryu reported that he was seeing some OOM kills kick in when running a
testcase that adds and removes inotify marks on a file in a tight loop.
The above commit changed the code to use call_srcu to clean up the
marks. While that does (in principle) work, the srcu callback job is
limited to cleaning up entries in small batches and only once per jiffy.
It's easily possible to overwhelm that machinery with too many call_srcu
callbacks, and Eryu's reproduer did just that.
There's also another potential problem with using call_srcu here. While
you can obviously sleep while holding the srcu_read_lock, the callbacks
run under local_bh_disable, so you can't sleep there.
It's possible when putting the last reference to the fsnotify_mark that
we'll end up putting a chain of references including the fsnotify_group,
uid, and associated keys. While I don't see any obvious ways that that
could occurs, it's probably still best to avoid using call_srcu here
after all.
This patch reverts the above patch. A later patch will take a different
approach to eliminated the dedicated thread here.
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Reported-by: Eryu Guan <guaneryu@gmail.com>
Tested-by: Eryu Guan <guaneryu@gmail.com>
Cc: Jan Kara <jack@suse.com>
Cc: Eric Paris <eparis@parisplace.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
At the time that this code was originally written, call_srcu didn't
exist, so this thread was required to ensure that we waited for that
SRCU grace period to settle before finally freeing the object.
It does exist now however and we can much more efficiently use call_srcu
to handle this. That also allows us to potentially use srcu_barrier to
ensure that they are all of the callbacks have run before proceeding.
In order to conserve space, we union the rcu_head with the g_list.
This will be necessary for nfsd which will allocate marks from a
dedicated slabcache. We have to be able to ensure that all of the
objects are destroyed before destroying the cache. That's fairly
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Cc: Eric Paris <eparis@parisplace.org>
Reviewed-by: Jan Kara <jack@suse.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The comment here says that it is checking for invalid bits. But, the mask
is *actually* checking to ensure that _any_ valid bit is set, which is
quite different.
Without this check, an unexpected bit could get set on an inotify object.
Since these bits are also interpreted by the fsnotify/dnotify code, there
is the potential for an object to be mishandled inside the kernel. For
instance, can we be sure that setting the dnotify flag FS_DN_RENAME on an
inotify watch is harmless?
Add the actual check which was intended. Retain the existing inotify bits
are being added to the watch. Plus, this is existing behavior which would
be nice to preserve.
I did a quick sniff test that inotify functions and that my
'inotify-tools' package passes 'make check'.
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Cc: John McCutchan <john@johnmccutchan.com>
Cc: Robert Love <rlove@rlove.org>
Cc: Eric Paris <eparis@parisplace.org>
Cc: Josh Boyer <jwboyer@fedoraproject.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There was a report that my patch:
inotify: actually check for invalid bits in sys_inotify_add_watch()
broke CRIU.
The reason is that CRIU looks up raw flags in /proc/$pid/fdinfo/* to
figure out how to rebuild inotify watches and then passes those flags
directly back in to the inotify API. One of those flags
(FS_EVENT_ON_CHILD) is set in mark->mask, but is not part of the inotify
API. It is used inside the kernel to _implement_ inotify but it is not
and has never been part of the API.
My patch above ensured that we only allow bits which are part of the API
(IN_ALL_EVENTS). This broke CRIU.
FS_EVENT_ON_CHILD is really internal to the kernel. It is set _anyway_ on
all inotify marks. So, CRIU was really just trying to set a bit that was
already set.
This patch hides that bit from fdinfo. CRIU will not see the bit, not try
to set it, and should work as before. We should not have been exposing
this bit in the first place, so this is a good patch independent of the
CRIU problem.
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reported-by: Andrey Wagin <avagin@gmail.com>
Acked-by: Andrey Vagin <avagin@openvz.org>
Acked-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Eric Paris <eparis@redhat.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: John McCutchan <john@johnmccutchan.com>
Cc: Robert Love <rlove@rlove.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull vfs updates from Al Viro:
"In this one:
- d_move fixes (Eric Biederman)
- UFS fixes (me; locking is mostly sane now, a bunch of bugs in error
handling ought to be fixed)
- switch of sb_writers to percpu rwsem (Oleg Nesterov)
- superblock scalability (Josef Bacik and Dave Chinner)
- swapon(2) race fix (Hugh Dickins)"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (65 commits)
vfs: Test for and handle paths that are unreachable from their mnt_root
dcache: Reduce the scope of i_lock in d_splice_alias
dcache: Handle escaped paths in prepend_path
mm: fix potential data race in SyS_swapon
inode: don't softlockup when evicting inodes
inode: rename i_wb_list to i_io_list
sync: serialise per-superblock sync operations
inode: convert inode_sb_list_lock to per-sb
inode: add hlist_fake to avoid the inode hash lock in evict
writeback: plug writeback at a high level
change sb_writers to use percpu_rw_semaphore
shift percpu_counter_destroy() into destroy_super_work()
percpu-rwsem: kill CONFIG_PERCPU_RWSEM
percpu-rwsem: introduce percpu_rwsem_release() and percpu_rwsem_acquire()
percpu-rwsem: introduce percpu_down_read_trylock()
document rwsem_release() in sb_wait_write()
fix the broken lockdep logic in __sb_start_write()
introduce __sb_writers_{acquired,release}() helpers
ufs_inode_get{frag,block}(): get rid of 'phys' argument
ufs_getfrag_block(): tidy up a bit
...
fsnotify_destroy_mark_locked() is subtle to use because it temporarily
releases group->mark_mutex. To avoid future problems with this
function, split it into two.
fsnotify_detach_mark() is the part that needs group->mark_mutex and
fsnotify_free_mark() is the part that must be called outside of
group->mark_mutex. This way it's much clearer what's going on and we
also avoid some pointless acquisitions of group->mark_mutex.
Signed-off-by: Jan Kara <jack@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Free list is used when all marks on given inode / mount should be
destroyed when inode / mount is going away. However we can free all of
the marks without using a special list with some care.
Signed-off-by: Jan Kara <jack@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A check in inotify_fdinfo() checking whether mark is valid was always
true due to a bug. Luckily we can never get to invalidated marks since
we hold mark_mutex and invalidated marks get removed from the group list
when they are invalidated under that mutex.
Anyway fix the check to make code more future proof.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I have a _tiny_ microbenchmark that sits in a loop and writes single
bytes to a file. Writing one byte to a tmpfs file is around 2x slower
than reading one byte from a file, which is a _bit_ more than I expecte.
This is a dumb benchmark, but I think it's hard to deny that write() is
a hot path and we should avoid unnecessary overhead there.
I did a 'perf record' of 30-second samples of read and write. The top
item in a diffprofile is srcu_read_lock() from fsnotify(). There are
active inotify fd's from systemd, but nothing is actually listening to
the file or its part of the filesystem.
I *think* we can avoid taking the srcu_read_lock() for the common case
where there are no actual marks on the file. This means that there will
both be nothing to notify for *and* implies that there is no need for
clearing the ignore mask.
This patch gave a 13.1% speedup in writes/second on my test, which is an
improvement from the 10.8% that I saw with the last version.
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Eric Paris <eparis@redhat.com>
Cc: John McCutchan <john@johnmccutchan.com>
Cc: Robert Love <rlove@rlove.org>
Cc: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The process of reducing contention on per-superblock inode lists
starts with moving the locking to match the per-superblock inode
list. This takes the global lock out of the picture and reduces the
contention problems to within a single filesystem. This doesn't get
rid of contention as the locks still have global CPU scope, but it
does isolate operations on different superblocks form each other.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Dave Chinner <dchinner@redhat.com>
fsnotify_clear_marks_by_group_flags() can race with
fsnotify_destroy_marks() so that when fsnotify_destroy_mark_locked()
drops mark_mutex, a mark from the list iterated by
fsnotify_clear_marks_by_group_flags() can be freed and thus the next
entry pointer we have cached may become stale and we dereference free
memory.
Fix the problem by first moving marks to free to a special private list
and then always free the first entry in the special list. This method
is safe even when entries from the list can disappear once we drop the
lock.
Signed-off-by: Jan Kara <jack@suse.com>
Reported-by: Ashish Sangwan <a.sangwan@samsung.com>
Reviewed-by: Ashish Sangwan <a.sangwan@samsung.com>
Cc: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This reverts commit a2673b6e04.
Kinglong Mee reports a memory leak with that patch, and Jan Kara confirms:
"Thanks for report! You are right that my patch introduces a race
between fsnotify kthread and fsnotify_destroy_group() which can result
in leaking inotify event on group destruction.
I haven't yet decided whether the right fix is not to queue events for
dying notification group (as that is pointless anyway) or whether we
should just fix the original problem differently... Whenever I look
at fsnotify code mark handling I get lost in the maze of locks, lists,
and subtle differences between how different notification systems
handle notification marks :( I'll think about it over night"
and after thinking about it, Jan says:
"OK, I have looked into the code some more and I found another
relatively simple way of fixing the original oops. It will be IMHO
better than trying to fixup this issue which has more potential for
breakage. I'll ask Linus to revert the fsnotify fix he already merged
and send a new fix"
Reported-by: Kinglong Mee <kinglongmee@gmail.com>
Requested-by: Jan Kara <jack@suse.cz>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fsnotify_clear_marks_by_group_flags() can race with
fsnotify_destroy_marks() so when fsnotify_destroy_mark_locked() drops
mark_mutex, a mark from the list iterated by
fsnotify_clear_marks_by_group_flags() can be freed and we dereference free
memory in the loop there.
Fix the problem by keeping mark_mutex held in
fsnotify_destroy_mark_locked(). The reason why we drop that mutex is that
we need to call a ->freeing_mark() callback which may acquire mark_mutex
again. To avoid this and similar lock inversion issues, we move the call
to ->freeing_mark() callback to the kthread destroying the mark.
Signed-off-by: Jan Kara <jack@suse.cz>
Reported-by: Ashish Sangwan <a.sangwan@samsung.com>
Suggested-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The INOTIFY_USER option is bool, and hence this code is either
present or absent. It will never be modular, so using
module_init as an alias for __initcall is rather misleading.
Fix this up now, so that we can relocate module_init from
init.h into module.h in the future. If we don't do this, we'd
have to add module.h to obviously non-modular code, and that
would be a worse thing.
Note that direct use of __initcall is discouraged, vs. one
of the priority categorized subgroups. As __initcall gets
mapped onto device_initcall, our use of fs_initcall (which
makes sense for fs code) will thus change this registration
from level 6-device to level 5-fs (i.e. slightly earlier).
However no observable impact of that small difference has
been observed during testing, or is expected.
Cc: John McCutchan <john@johnmccutchan.com>
Cc: Robert Love <rlove@rlove.org>
Cc: Eric Paris <eparis@parisplace.org>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
With FAN_ONDIR set, the user can end up getting events, which it hasn't
marked. This was revealed with fanotify04 testcase failure on
Linux-4.0-rc1, and is a regression from 3.19, revealed with 66ba93c0d7
("fanotify: don't set FAN_ONDIR implicitly on a marks ignored mask").
# /opt/ltp/testcases/bin/fanotify04
[ ... ]
fanotify04 7 TPASS : event generated properly for type 100000
fanotify04 8 TFAIL : fanotify04.c:147: got unexpected event 30
fanotify04 9 TPASS : No event as expected
The testcase sets the adds the following marks : FAN_OPEN | FAN_ONDIR for
a fanotify on a dir. Then does an open(), followed by close() of the
directory and expects to see an event FAN_OPEN(0x20). However, the
fanotify returns (FAN_OPEN|FAN_CLOSE_NOWRITE(0x10)). This happens due to
the flaw in the check for event_mask in fanotify_should_send_event() which
does:
if (event_mask & marks_mask & ~marks_ignored_mask)
return true;
where, event_mask == (FAN_ONDIR | FAN_CLOSE_NOWRITE),
marks_mask == (FAN_ONDIR | FAN_OPEN),
marks_ignored_mask == 0
Fix this by masking the outgoing events to the user, as we already take
care of FAN_ONDIR and FAN_EVENT_ON_CHILD.
Signed-off-by: Suzuki K. Poulose <suzuki.poulose@arm.com>
Tested-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Eric Paris <eparis@redhat.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Jan Kara
Jeff Layton
Geliang Tang
Dave Hansen
Linus Torvalds
Jan Kara
Dave Chinner
Paul Gortmaker
Suzuki K. Poulose