mm: memcontrol: fix lockless reclaim hierarchy iterator
authorJohannes Weiner <hannes@cmpxchg.org>
Wed, 12 Jun 2013 21:05:09 +0000 (14:05 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 12 Jun 2013 23:29:46 +0000 (16:29 -0700)
commit89dc991f0f5272c307c746fdd57d0bff382b1ba2
treef0d71da8f791864e38819b9e576b7fd5ddb217bc
parent7b57976da48e60b66fdbb9e97f5711b5382a49d7
mm: memcontrol: fix lockless reclaim hierarchy iterator

The lockless reclaim hierarchy iterator currently has a misplaced
barrier that can lead to use-after-free crashes.

The reclaim hierarchy iterator consist of a sequence count and a
position pointer that are read and written locklessly, with memory
barriers enforcing ordering.

The write side sets the position pointer first, then updates the
sequence count to "publish" the new position.  Likewise, the read side
must read the sequence count first, then the position.  If the sequence
count is up to date, it's guaranteed that the position is up to date as
well:

  writer:                         reader:
  iter->position = position       if iter->sequence == expected:
  smp_wmb()                           smp_rmb()
  iter->sequence = sequence           position = iter->position

However, the read side barrier is currently misplaced, which can lead to
dereferencing stale position pointers that no longer point to valid
memory.  Fix this.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reported-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Tejun Heo <tj@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Glauber Costa <glommer@parallels.com>
Cc: <stable@kernel.org> [3.10+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c