memcg: avoid deadlock caused by race between oom and cpuset_attach
authorDaisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Thu, 8 Jan 2009 02:08:29 +0000 (18:08 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 8 Jan 2009 16:31:09 +0000 (08:31 -0800)
commit7f4d454dee2e0bdd21bafd413d1c53e443a26540
treeabf54c2bd7c91fe09685e42b3a92d84679403058
parenta5e924f5f8abf97944e625d74967cc9452cfbce8
memcg: avoid deadlock caused by race between oom and cpuset_attach

mpol_rebind_mm(), which can be called from cpuset_attach(), does
down_write(mm->mmap_sem).  This means down_write(mm->mmap_sem) can be
called under cgroup_mutex.

OTOH, page fault path does down_read(mm->mmap_sem) and calls
mem_cgroup_try_charge_xxx(), which may eventually calls
mem_cgroup_out_of_memory().  And mem_cgroup_out_of_memory() calls
cgroup_lock().  This means cgroup_lock() can be called under
down_read(mm->mmap_sem).

If those two paths race, deadlock can happen.

This patch avoid this deadlock by:
  - remove cgroup_lock() from mem_cgroup_out_of_memory().
  - define new mutex (memcg_tasklist) and serialize mem_cgroup_move_task()
    (->attach handler of memory cgroup) and mem_cgroup_out_of_memory.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c
mm/oom_kill.c