cgroup_freezer: fix freezing groups with stopped tasks
authorMichal Hocko <mhocko@suse.cz>
Tue, 22 Nov 2011 15:44:47 +0000 (07:44 -0800)
committerTejun Heo <tj@kernel.org>
Thu, 24 Nov 2011 19:58:22 +0000 (11:58 -0800)
2d3cbf8b (cgroup_freezer: update_freezer_state() does incorrect state
transitions) removed is_task_frozen_enough and replaced it with a simple
frozen call. This, however, breaks freezing for a group with stopped tasks
because those cannot be frozen and so the group remains in CGROUP_FREEZING
state (update_if_frozen doesn't count stopped tasks) and never reaches
CGROUP_FROZEN.

Let's add is_task_frozen_enough back and use it at the original locations
(update_if_frozen and try_to_freeze_cgroup). Semantically we consider
stopped tasks as frozen enough so we should consider both cases when
testing frozen tasks.

Testcase:
mkdir /dev/freezer
mount -t cgroup -o freezer none /dev/freezer
mkdir /dev/freezer/foo
sleep 1h &
pid=$!
kill -STOP $pid
echo $pid > /dev/freezer/foo/tasks
echo FROZEN > /dev/freezer/foo/freezer.state
while true
do
cat /dev/freezer/foo/freezer.state
[ "`cat /dev/freezer/foo/freezer.state`" = "FROZEN" ] && break
sleep 1
done
echo OK

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Tomasz Buchert <tomasz.buchert@inria.fr>
Cc: Paul Menage <paul@paulmenage.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: stable@kernel.org
Signed-off-by: Tejun Heo <htejun@gmail.com>
kernel/cgroup_freezer.c

index 5e828a2ca8e64641749da837fe093cefcfc958dc..213c0351dad8fc8315d6d79a5bf3706a59c823bd 100644 (file)
@@ -153,6 +153,13 @@ static void freezer_destroy(struct cgroup_subsys *ss,
        kfree(cgroup_freezer(cgroup));
 }
 
+/* task is frozen or will freeze immediately when next it gets woken */
+static bool is_task_frozen_enough(struct task_struct *task)
+{
+       return frozen(task) ||
+               (task_is_stopped_or_traced(task) && freezing(task));
+}
+
 /*
  * The call to cgroup_lock() in the freezer.state write method prevents
  * a write to that file racing against an attach, and hence the
@@ -231,7 +238,7 @@ static void update_if_frozen(struct cgroup *cgroup,
        cgroup_iter_start(cgroup, &it);
        while ((task = cgroup_iter_next(cgroup, &it))) {
                ntotal++;
-               if (frozen(task))
+               if (is_task_frozen_enough(task))
                        nfrozen++;
        }
 
@@ -284,7 +291,7 @@ static int try_to_freeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
        while ((task = cgroup_iter_next(cgroup, &it))) {
                if (!freeze_task(task, true))
                        continue;
-               if (frozen(task))
+               if (is_task_frozen_enough(task))
                        continue;
                if (!freezing(task) && !freezer_should_skip(task))
                        num_cant_freeze_now++;