From: Vladimir Davydov <vdavydov@xxxxxxxxxxxxx> Subject: kernfs: do not account ino_ida allocations to memcg root->ino_ida is used for kernfs inode number allocations. Since IDA has a layered structure, different IDs can reside on the same layer, which is currently accounted to some memory cgroup. The problem is that each kmem cache of a memory cgroup has its own directory on sysfs (under /sys/fs/kernel/<cache-name>/cgroup). If the inode number of such a directory or any file in it gets allocated from a layer accounted to the cgroup which the cache is created for, the cgroup will get pinned for good, because one has to free all kmem allocations accounted to a cgroup in order to release it and destroy all its kmem caches. That said we must not account layers of ino_ida to any memory cgroup. Since per net init operations may create new sysfs entries directly (e.g. lo device) or indirectly (nf_conntrack creates a new kmem cache per each namespace, which, in turn, creates new sysfs entries), an easy way to reproduce this issue is by creating network namespace(s) from inside a kmem-active memory cgroup. Signed-off-by: Vladimir Davydov <vdavydov@xxxxxxxxxxxxx> Acked-by: Tejun Heo <tj@xxxxxxxxxx> Cc: Johannes Weiner <hannes@xxxxxxxxxxx> Cc: Michal Hocko <mhocko@xxxxxxx> Cc: Christoph Lameter <cl@xxxxxxxxx> Cc: Pekka Enberg <penberg@xxxxxxxxxx> Cc: David Rientjes <rientjes@xxxxxxxxxx> Cc: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx> Cc: Greg Thelen <gthelen@xxxxxxxxxx> Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> Cc: <stable@xxxxxxxxxxxxxxx> [4.0.x] Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- fs/kernfs/dir.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff -puN fs/kernfs/dir.c~kernfs-do-not-account-ino_ida-allocations-to-memcg fs/kernfs/dir.c --- a/fs/kernfs/dir.c~kernfs-do-not-account-ino_ida-allocations-to-memcg +++ a/fs/kernfs/dir.c @@ -518,7 +518,14 @@ static struct kernfs_node *__kernfs_new_ if (!kn) goto err_out1; - ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL); + /* + * If the ino of the sysfs entry created for a kmem cache gets + * allocated from an ida layer, which is accounted to the memcg that + * owns the cache, the memcg will get pinned forever. So do not account + * ino ida allocations. + */ + ret = ida_simple_get(&root->ino_ida, 1, 0, + GFP_KERNEL | __GFP_NOACCOUNT); if (ret < 0) goto err_out2; kn->ino = ret; _ -- To unsubscribe from this list: send the line "unsubscribe stable" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html