memcg: fix hierarchical oom locking
Johannes Weiner [Thu, 25 Aug 2011 22:59:16 +0000 (15:59 -0700)]
Commit 79dfdaccd1d5 ("memcg: make oom_lock 0 and 1 based rather than
counter") tried to oom lock the hierarchy and roll back upon
encountering an already locked memcg.

The code is confused when it comes to detecting a locked memcg, though,
so it would fail and rollback after locking one memcg and encountering
an unlocked second one.

The result is that oom-locking hierarchies fails unconditionally and
that every oom killer invocation simply goes to sleep on the oom
waitqueue forever.  The tasks practically hang forever without anyone
intervening, possibly holding locks that trip up unrelated tasks, too.

Signed-off-by: Johannes Weiner <jweiner@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcontrol.c

index 0e40f02..ebd1e86 100644 (file)
@@ -1841,29 +1841,23 @@ static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
  */
 static bool mem_cgroup_oom_lock(struct mem_cgroup *mem)
 {
-       int lock_count = -1;
        struct mem_cgroup *iter, *failed = NULL;
        bool cond = true;
 
        for_each_mem_cgroup_tree_cond(iter, mem, cond) {
-               bool locked = iter->oom_lock;
-
-               iter->oom_lock = true;
-               if (lock_count == -1)
-                       lock_count = iter->oom_lock;
-               else if (lock_count != locked) {
+               if (iter->oom_lock) {
                        /*
                         * this subtree of our hierarchy is already locked
                         * so we cannot give a lock.
                         */
-                       lock_count = 0;
                        failed = iter;
                        cond = false;
-               }
+               } else
+                       iter->oom_lock = true;
        }
 
        if (!failed)
-               goto done;
+               return true;
 
        /*
         * OK, we failed to lock the whole subtree so we have to clean up
@@ -1877,8 +1871,7 @@ static bool mem_cgroup_oom_lock(struct mem_cgroup *mem)
                }
                iter->oom_lock = false;
        }
-done:
-       return lock_count;
+       return false;
 }
 
 /*