kill-the-bkl/reiserfs: fix reiserfs lock to cpu_add_remove_lock dependency
authorFrederic Weisbecker <fweisbec@gmail.com>
Mon, 5 Oct 2009 14:31:37 +0000 (16:31 +0200)
committerFrederic Weisbecker <fweisbec@gmail.com>
Mon, 5 Oct 2009 14:31:37 +0000 (16:31 +0200)
commit48f6ba5e691948caba2e7bc362153fb28e4f1e09
tree318e755ec8c3664b1276a5a87f455b8dc37c3862
parent193be0ee17dd7ea309ddab1093da17e5924d7f36
kill-the-bkl/reiserfs: fix reiserfs lock to cpu_add_remove_lock dependency

While creating the reiserfs workqueue during the journal
initialization, we are holding the reiserfs lock, but
create_workqueue() also holds the cpu_add_remove_lock, creating
then the following dependency:

- reiserfs lock -> cpu_add_remove_lock

But we also have the following existing dependencies:

- mm->mmap_sem -> reiserfs lock
- cpu_add_remove_lock -> cpu_hotplug.lock -> slub_lock -> sysfs_mutex

The merged dependency chain then becomes:

- mm->mmap_sem -> reiserfs lock -> cpu_add_remove_lock ->
cpu_hotplug.lock -> slub_lock -> sysfs_mutex

But when we fill a dir entry in sysfs_readir(), we are holding the
sysfs_mutex and we also might fault while copying the directory entry
to the user, leading to the following dependency:

- sysfs_mutex -> mm->mmap_sem

The end result is then a lock inversion between sysfs_mutex and
mm->mmap_sem, as reported in the following lockdep warning:

[ INFO: possible circular locking dependency detected ]
2.6.31-07095-g25a3912 #4
-------------------------------------------------------
udevadm/790 is trying to acquire lock:
 (&mm->mmap_sem){++++++}, at: [<c1098942>] might_fault+0x72/0xc0

but task is already holding lock:
 (sysfs_mutex){+.+.+.}, at: [<c110813c>] sysfs_readdir+0x7c/0x260

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #5 (sysfs_mutex){+.+.+.}:
      [...]

-> #4 (slub_lock){+++++.}:
      [...]

-> #3 (cpu_hotplug.lock){+.+.+.}:
      [...]

-> #2 (cpu_add_remove_lock){+.+.+.}:
      [...]

-> #1 (&REISERFS_SB(s)->lock){+.+.+.}:
      [...]

-> #0 (&mm->mmap_sem){++++++}:
      [...]

This can be fixed by relaxing the reiserfs lock while creating the
workqueue.
This is fine to relax the lock here, we just keep it around to pass
through reiserfs lock checks and for paranoid reasons.

Reported-by: Alexander Beregalov <a.beregalov@gmail.com>
Tested-by: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Cc: Laurent Riffard <laurent.riffard@free.fr>
fs/reiserfs/journal.c