Instead of removing mark from object list from fsnotify_detach_mark(),
remove the mark when last reference to the mark is dropped. This will
allow fanotify to wait for userspace response to event without having to
hold onto fsnotify_mark_srcu.

To avoid pinning inodes by elevated refcount (and thus e.g. delaying
file deletion) while someone holds mark reference, we detach connector
from the object also from fsnotify_destroy_marks() and not only after
removing last mark from the list as it was now.
Reviewed-by: NMiklos Szeredi <mszeredi@redhat.com>
Reviewed-by: NAmir Goldstein <amir73il@gmail.com>
Signed-off-by: NJan Kara <jack@suse.cz>

Currently we free fsnotify_mark_connector structure only when inode /
vfsmount is getting freed. This can however impose noticeable memory
overhead when marks get attached to inodes only temporarily. So free the
connector structure once the last mark is detached from the object.
Since notification infrastructure can be working with the connector
under the protection of fsnotify_mark_srcu, we have to be careful and
free the fsnotify_mark_connector only after SRCU period passes.
Reviewed-by: NMiklos Szeredi <mszeredi@redhat.com>
Reviewed-by: NAmir Goldstein <amir73il@gmail.com>
Signed-off-by: NJan Kara <jack@suse.cz>

So far list of marks attached to an object (inode / vfsmount) was
protected by i_lock or mnt_root->d_lock. This dictates that the list
must be empty before the object can be destroyed although the list is
now anchored in the fsnotify_mark_connector structure. Protect the list
by a spinlock in the fsnotify_mark_connector structure to decouple
lifetime of a list of marks from a lifetime of the object. This also
simplifies the code quite a bit since we don't have to differentiate
between inode and vfsmount lists in quite a few places anymore.
Reviewed-by: NMiklos Szeredi <mszeredi@redhat.com>
Reviewed-by: NAmir Goldstein <amir73il@gmail.com>
Signed-off-by: NJan Kara <jack@suse.cz>

Move locking of locks protecting a list of marks into
fsnotify_recalc_mask(). This reduces code churn in the following patch
which changes the lock protecting the list of marks.
Reviewed-by: NMiklos Szeredi <mszeredi@redhat.com>
Reviewed-by: NAmir Goldstein <amir73il@gmail.com>
Signed-off-by: NJan Kara <jack@suse.cz>

Currently inode reference is held by fsnotify marks. Change the rules so
that inode reference is held by fsnotify_mark_connector structure
whenever the list is non-empty. This simplifies the code and is more
logical.
Reviewed-by: NMiklos Szeredi <mszeredi@redhat.com>
Reviewed-by: NAmir Goldstein <amir73il@gmail.com>
Signed-off-by: NJan Kara <jack@suse.cz>

Move pointer to inode / vfsmount from mark itself to the
fsnotify_mark_connector structure. This is another step on the path
towards decoupling inode / vfsmount lifetime from notification mark
lifetime.
Reviewed-by: NMiklos Szeredi <mszeredi@redhat.com>
Reviewed-by: NAmir Goldstein <amir73il@gmail.com>
Signed-off-by: NJan Kara <jack@suse.cz>

Currently notification marks are attached to object (inode or vfsmnt) by
a hlist_head in the object. The list is also protected by a spinlock in
the object. So while there is any mark attached to the list of marks,
the object must be pinned in memory (and thus e.g. last iput() deleting
inode cannot happen). Also for list iteration in fsnotify() to work, we
must hold fsnotify_mark_srcu lock so that mark itself and
mark->obj_list.next cannot get freed. Thus we are required to wait for
response to fanotify events from userspace process with
fsnotify_mark_srcu lock held. That causes issues when userspace process
is buggy and does not reply to some event - basically the whole
notification subsystem gets eventually stuck.

So to be able to drop fsnotify_mark_srcu lock while waiting for
response, we have to pin the mark in memory and make sure it stays in
the object list (as removing the mark waiting for response could lead to
lost notification events for groups later in the list). However we don't
want inode reclaim to block on such mark as that would lead to system
just locking up elsewhere.

This commit is the first in the series that paves way towards solving
these conflicting lifetime needs. Instead of anchoring the list of marks
directly in the object, we anchor it in a dedicated structure
(fsnotify_mark_connector) and just point to that structure from the
object. The following commits will also add spinlock protecting the list
and object pointer to the structure.
Reviewed-by: NMiklos Szeredi <mszeredi@redhat.com>
Reviewed-by: NAmir Goldstein <amir73il@gmail.com>
Signed-off-by: NJan Kara <jack@suse.cz>

This patchset converts inotify to using the newly introduced
per-userns sysctl infrastructure.

Currently the inotify instances/watches are being accounted in the
user_struct structure. This means that in setups where multiple
users in unprivileged containers map to the same underlying
real user (i.e. pointing to the same user_struct) the inotify limits
are going to be shared as well, allowing one user(or application) to exhaust
all others limits.

Fix this by switching the inotify sysctls to using the
per-namespace/per-user limits. This will allow the server admin to
set sensible global limits, which can further be tuned inside every
individual user namespace. Additionally, in order to preserve the
sysctl ABI make the existing inotify instances/watches sysctls
modify the values of the initial user namespace.
Signed-off-by: NNikolay Borisov <n.borisov.lkml@gmail.com>
Acked-by: NJan Kara <jack@suse.cz>
Acked-by: NSerge Hallyn <serge@hallyn.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

There are only two calls sites of fsnotify_duplicate_mark(). Those are
in kernel/audit_tree.c and both are bogus. Vfsmount pointer is unused
for audit tree, inode pointer and group gets set in
fsnotify_add_mark_locked() later anyway, mask and free_mark are already
set in alloc_chunk(). In fact, calling fsnotify_duplicate_mark() is
actively harmful because following fsnotify_add_mark_locked() will leak
group reference by overwriting the group pointer. So just remove the two
calls to fsnotify_duplicate_mark() and the function.
Signed-off-by: NJan Kara <jack@suse.cz>
[PM: line wrapping to fit in 80 chars]
Signed-off-by: NPaul Moore <paul@paul-moore.com>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

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.czSigned-off-by: NJan Kara <jack@suse.cz>
Reviewed-by: NLino 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: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

notification_mutex is used to protect the list of pending events.  As such
there's no reason to use a sleeping lock for it.  Convert it to a
spinlock.

[jack@suse.cz: fixed version]
  Link: http://lkml.kernel.org/r/1474031567-1831-1-git-send-email-jack@suse.cz
Link: http://lkml.kernel.org/r/1473797711-14111-5-git-send-email-jack@suse.czSigned-off-by: NJan Kara <jack@suse.cz>
Reviewed-by: NLino Sanfilippo <LinoSanfilippo@gmx.de>
Tested-by: NGuenter Roeck <linux@roeck-us.net>
Cc: Miklos Szeredi <mszeredi@redhat.com>
Cc: Eric Paris <eparis@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus 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: 5838d444 ("fanotify: fix double free of pending permission events")
Link: http://lkml.kernel.org/r/1473797711-14111-3-git-send-email-jack@suse.czSigned-off-by: NJan Kara <jack@suse.cz>
Reported-by: NMiklos Szeredi <mszeredi@redhat.com>
Tested-by: NMiklos Szeredi <mszeredi@redhat.com>
Reviewed-by: NMiklos Szeredi <mszeredi@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Implement a function that can be called when a group is being shutdown
to stop queueing new events to the group.  Fanotify will use this.

Fixes: 5838d444 ("fanotify: fix double free of pending permission events")
Link: http://lkml.kernel.org/r/1473797711-14111-2-git-send-email-jack@suse.czSigned-off-by: NJan Kara <jack@suse.cz>
Reviewed-by: NMiklos Szeredi <mszeredi@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

fsnotify_d_move()/__fsnotify_d_instantiate()/__fsnotify_update_dcache_flags()
are identical to each other, regardless of the config.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Background: spufs used to mangle the order in which it had been building
dentry trees.  It was broken in a lot of ways, but most of them required
the right timing to trigger until an fsnotify change had added one more
- the one that was always triggered.

Unfortunately, insteading of fixing their long-standing bug the spufs
folks had chosen to paper over the fsnotify trigger.  Eventually said
bug had been spotted and killed off, but the pointless check in
fsnotify has remained, complete with the implication that one *could*
do that kind of crap.

Again, a parent of any dentry should always be positive.  Any code
can rely upon that and anything violating that assert is a bug,
*not* something to be accomodated.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

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: NJan Kara <jack@suse.cz>
Reported-by: NXiaoguang Wang <wangxg.fnst@cn.fujitsu.com>
Tested-by: NXiaoguang Wang <wangxg.fnst@cn.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

First of all, don't bother calling it if inode is NULL -
that makes inode argument unused.  Moreover, do it *before*
dropping ->d_lock, not right after that (and don't bother
grabbing ->d_lock in it, of course).
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

This reverts commit c510eff6 ("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: NJeff Layton <jeff.layton@primarydata.com>
Reported-by: NEryu Guan <guaneryu@gmail.com>
Tested-by: NEryu Guan <guaneryu@gmail.com>
Cc: Jan Kara <jack@suse.com>
Cc: Eric Paris <eparis@parisplace.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus 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: NJeff Layton <jeff.layton@primarydata.com>
Cc: Eric Paris <eparis@parisplace.org>
Reviewed-by: NJan Kara <jack@suse.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

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: NJan Kara <jack@suse.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus 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: NJan Kara <jack@suse.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NJan Kara <jack@suse.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus 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: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NJosef Bacik <jbacik@fb.com>
Reviewed-by: NJan Kara <jack@suse.cz>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Tested-by: NDave Chinner <dchinner@redhat.com>

should_send_event is no longer part of struct fsnotify_ops, so remove it.
Signed-off-by: NNikolay Borisov <kernel@kyup.com>
Reviewed-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

destroy_list is used to track marks which still need waiting for srcu
period end before they can be freed.  However by the time mark is added to
destroy_list it isn't in group's list of marks anymore and thus we can
reuse fsnotify_mark->g_list for queueing into destroy_list.  This saves
two pointers for each fsnotify_mark.
Signed-off-by: NJan Kara <jack@suse.cz>
Cc: Eric Paris <eparis@redhat.com>
Cc: Heinrich Schuchardt <xypron.glpk@gmx.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

There's a lot of common code in inode and mount marks handling.  Factor it
out to a common helper function.
Signed-off-by: NJan Kara <jack@suse.cz>
Cc: Eric Paris <eparis@redhat.com>
Cc: Heinrich Schuchardt <xypron.glpk@gmx.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Commit 85816794 ("fanotify: Fix use after free for permission
events") introduced a double free issue for permission events which are
pending in group's notification queue while group is being destroyed.
These events are freed from fanotify_handle_event() but they are not
removed from groups notification queue and thus they get freed again
from fsnotify_flush_notify().

Fix the problem by removing permission events from notification queue
before freeing them if we skip processing access response.  Also expand
comments in fanotify_release() to explain group shutdown in detail.

Fixes: 85816794Signed-off-by: NJan Kara <jack@suse.cz>
Reported-by: NDouglas Leeder <douglas.leeder@sophos.com>
Tested-by: NDouglas Leeder <douglas.leeder@sophos.com>
Reported-by: NHeinrich Schuchard <xypron.glpk@gmx.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Rename fsnotify_add_notify_event() to fsnotify_add_event() since the
"notify" part is duplicit.  Rename fsnotify_remove_notify_event() and
fsnotify_peek_notify_event() to fsnotify_remove_first_event() and
fsnotify_peek_first_event() respectively since "notify" part is duplicit
and they really look at the first event in the queue.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: NJan Kara <jack@suse.cz>
Cc: Eric Paris <eparis@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

access_mutex is used only to guard operations on access_list.  There's
no need for sleeping within this lock so just make a spinlock out of it.
Signed-off-by: NJan Kara <jack@suse.cz>
Cc: Eric Paris <eparis@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Commit 7053aee2 "fsnotify: do not share events between notification
groups" used overflow event statically allocated in a group with the
size of the generic notification event. This causes problems because
some code looks at type specific parts of event structure and gets
confused by a random data it sees there and causes crashes.

Fix the problem by allocating overflow event with type corresponding to
the group type so code cannot get confused.
Signed-off-by: NJan Kara <jack@suse.cz>

My rework of handling of notification events (namely commit 7053aee2
"fsnotify: do not share events between notification groups") broke
sending of cookies with inotify events. We didn't propagate the value
passed to fsnotify() properly and passed 4 uninitialized bytes to
userspace instead (so it is also an information leak). Sadly I didn't
notice this during my testing because inotify cookies aren't used very
much and LTP inotify tests ignore them.

Fix the problem by passing the cookie value properly.

Fixes: 7053aee2Reported-by: NVegard Nossum <vegard.nossum@oracle.com>
Signed-off-by: NJan Kara <jack@suse.cz>

The event returned from fsnotify_add_notify_event() cannot ever be used
safely as the event may be freed by the time the function returns (after
dropping notification_mutex). So change the prototype to just return
whether the event was added or merged into some existing event.
Reported-and-tested-by: NJiri Kosina <jkosina@suse.cz>
Reported-and-tested-by: NDave Jones <davej@fedoraproject.org>
Signed-off-by: NJan Kara <jack@suse.cz>

After removing event structure creation from the generic layer there is
no reason for separate .should_send_event and .handle_event callbacks.
So just remove the first one.
Signed-off-by: NJan Kara <jack@suse.cz>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Cc: Eric Paris <eparis@parisplace.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

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: NJan Kara <jack@suse.cz>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Cc: Eric Paris <eparis@parisplace.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NHugh Dickins <hughd@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NJeff Layton <jlayton@redhat.com>
Cc: John McCutchan <john@johnmccutchan.com>
Cc: Robert Love <rlove@rlove.org>
Cc: Eric Paris <eparis@parisplace.org>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

inotify is supposed to support async signal notification when information
is available on the inotify fd.  This patch moves that support to generic
fsnotify functions so it can be used by all notification mechanisms.
Signed-off-by: NEric Paris <eparis@redhat.com>

On Mon, Aug 01, 2011 at 04:38:22PM -0400, Eric Paris wrote:
>
> I finally built and tested a v3.0 kernel with these patches (I know I'm
> SOOOOOO far behind).  Not what I hoped for:
>
> > [  150.937798] VFS: Busy inodes after unmount of tmpfs. Self-destruct in 5 seconds.  Have a nice day...
> > [  150.945290] BUG: unable to handle kernel NULL pointer dereference at 0000000000000070
> > [  150.946012] IP: [<ffffffff810ffd58>] shmem_free_inode+0x18/0x50
> > [  150.946012] PGD 2bf9e067 PUD 2bf9f067 PMD 0
> > [  150.946012] Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
> > [  150.946012] CPU 0
> > [  150.946012] Modules linked in: nfs lockd fscache auth_rpcgss nfs_acl sunrpc ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_filter ip6_tables ext4 jbd2 crc16 joydev ata_piix i2c_piix4 pcspkr uinput ipv6 autofs4 usbhid [last unloaded: scsi_wait_scan]
> > [  150.946012]
> > [  150.946012] Pid: 2764, comm: syscall_thrash Not tainted 3.0.0+ #1 Red Hat KVM
> > [  150.946012] RIP: 0010:[<ffffffff810ffd58>]  [<ffffffff810ffd58>] shmem_free_inode+0x18/0x50
> > [  150.946012] RSP: 0018:ffff88002c2e5df8  EFLAGS: 00010282
> > [  150.946012] RAX: 000000004e370d9f RBX: 0000000000000000 RCX: ffff88003a029438
> > [  150.946012] RDX: 0000000033630a5f RSI: 0000000000000000 RDI: ffff88003491c240
> > [  150.946012] RBP: ffff88002c2e5e08 R08: 0000000000000000 R09: 0000000000000000
> > [  150.946012] R10: 0000000000000000 R11: 0000000000000000 R12: ffff88003a029428
> > [  150.946012] R13: ffff88003a029428 R14: ffff88003a029428 R15: ffff88003499a610
> > [  150.946012] FS:  00007f5a05420700(0000) GS:ffff88003f600000(0000) knlGS:0000000000000000
> > [  150.946012] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
> > [  150.946012] CR2: 0000000000000070 CR3: 000000002a662000 CR4: 00000000000006f0
> > [  150.946012] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > [  150.946012] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> > [  150.946012] Process syscall_thrash (pid: 2764, threadinfo ffff88002c2e4000, task ffff88002bfbc760)
> > [  150.946012] Stack:
> > [  150.946012]  ffff88003a029438 ffff88003a029428 ffff88002c2e5e38 ffffffff81102f76
> > [  150.946012]  ffff88003a029438 ffff88003a029598 ffffffff8160f9c0 ffff88002c221250
> > [  150.946012]  ffff88002c2e5e68 ffffffff8115e9be ffff88002c2e5e68 ffff88003a029438
> > [  150.946012] Call Trace:
> > [  150.946012]  [<ffffffff81102f76>] shmem_evict_inode+0x76/0x130
> > [  150.946012]  [<ffffffff8115e9be>] evict+0x7e/0x170
> > [  150.946012]  [<ffffffff8115ee40>] iput_final+0xd0/0x190
> > [  150.946012]  [<ffffffff8115ef33>] iput+0x33/0x40
> > [  150.946012]  [<ffffffff81180205>] fsnotify_destroy_mark_locked+0x145/0x160
> > [  150.946012]  [<ffffffff81180316>] fsnotify_destroy_mark+0x36/0x50
> > [  150.946012]  [<ffffffff81181937>] sys_inotify_rm_watch+0x77/0xd0
> > [  150.946012]  [<ffffffff815aca52>] system_call_fastpath+0x16/0x1b
> > [  150.946012] Code: 67 4a 00 b8 e4 ff ff ff eb aa 66 0f 1f 84 00 00 00 00 00 55 48 89 e5 48 83 ec 10 48 89 1c 24 4c 89 64 24 08 48 8b 9f 40 05 00 00
> > [  150.946012]  83 7b 70 00 74 1c 4c 8d a3 80 00 00 00 4c 89 e7 e8 d2 5d 4a
> > [  150.946012] RIP  [<ffffffff810ffd58>] shmem_free_inode+0x18/0x50
> > [  150.946012]  RSP <ffff88002c2e5df8>
> > [  150.946012] CR2: 0000000000000070
>
> Looks at aweful lot like the problem from:
> http://www.spinics.net/lists/linux-fsdevel/msg46101.html
>

I tried to reproduce this bug with your test program, but without success.
However, if I understand correctly, this occurs since we dont hold any locks when
we call iput() in mark_destroy(), right?
With the patches you tested, iput() is also not called within any lock, since the
groups mark_mutex is released temporarily before iput() is called.  This is, since
the original codes behaviour is similar.
However since we now have a mutex as the biggest lock, we can do what you
suggested (http://www.spinics.net/lists/linux-fsdevel/msg46107.html) and
call iput() with the mutex held to avoid the race.
The patch below implements this. It uses nested locking to avoid deadlock in case
we do the final iput() on an inode which still holds marks and thus would take
the mutex again when calling fsnotify_inode_delete() in destroy_inode().
Signed-off-by: NLino Sanfilippo <LinoSanfilippo@gmx.de>
Signed-off-by: NEric Paris <eparis@redhat.com>