cpuset: use trialcs->mems_allowed as a temp variable
authorAlban Crequy <alban.crequy@gmail.com>
Thu, 6 Aug 2015 14:21:05 +0000 (16:21 +0200)
committerTejun Heo <tj@kernel.org>
Mon, 10 Aug 2015 15:18:41 +0000 (11:18 -0400)
The comment says it's using trialcs->mems_allowed as a temp variable but
it didn't match the code. Change the code to match the comment.

This fixes an issue when writing in cpuset.mems when a sub-directory
exists: we need to write several times for the information to persist:

| root@alban:/sys/fs/cgroup/cpuset# mkdir footest9
| root@alban:/sys/fs/cgroup/cpuset# cd footest9
| root@alban:/sys/fs/cgroup/cpuset/footest9# mkdir aa
| root@alban:/sys/fs/cgroup/cpuset/footest9# cat cpuset.mems
|
| root@alban:/sys/fs/cgroup/cpuset/footest9# echo 0 > cpuset.mems
| root@alban:/sys/fs/cgroup/cpuset/footest9# cat cpuset.mems
|
| root@alban:/sys/fs/cgroup/cpuset/footest9# echo 0 > cpuset.mems
| root@alban:/sys/fs/cgroup/cpuset/footest9# cat cpuset.mems
| 0
| root@alban:/sys/fs/cgroup/cpuset/footest9# cat aa/cpuset.mems
|
| root@alban:/sys/fs/cgroup/cpuset/footest9# echo 0 > aa/cpuset.mems
| root@alban:/sys/fs/cgroup/cpuset/footest9# cat aa/cpuset.mems
| 0
| root@alban:/sys/fs/cgroup/cpuset/footest9#

This should help to fix the following issue in Docker:
https://github.com/opencontainers/runc/issues/133
In some conditions, a Docker container needs to be started twice in
order to work.

Signed-off-by: Alban Crequy <alban@endocode.com>
Tested-by: Iago López Galeiras <iago@endocode.com>
Cc: <stable@vger.kernel.org> # 3.17+
Acked-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/cpuset.c

index ee14e3a..f0acff0 100644 (file)
@@ -1223,7 +1223,7 @@ static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
        spin_unlock_irq(&callback_lock);
 
        /* use trialcs->mems_allowed as a temp variable */
-       update_nodemasks_hier(cs, &cs->mems_allowed);
+       update_nodemasks_hier(cs, &trialcs->mems_allowed);
 done:
        return retval;
 }