2014-02-15 04:03:13

by Zefan Li

Subject: [PATCH v2] cgroup: fix top cgroup refcnt leak

If we mount the same cgroupfs in serveral mount points, and then
umount all of them, kill_sb() will be called only once.

Therefore it's wrong to increment top_cgroup's refcnt when we find
an existing cgroup_root.

Try:
# mount -t cgroup -o cpuacct xxx /cgroup
# mount -t cgroup -o cpuacct xxx /cgroup2
# cat /proc/cgroups | grep cpuacct
cpuacct 2 1 1
# umount /cgroup
# umount /cgroup2
# cat /proc/cgroups | grep cpuacct
cpuacct 2 1 1

You'll see cgroupfs will never be freed.

v2: change to take the refcnt and drop it after kernfs_mount().

Signed-off-by: Li Zefan <[email protected]>
---
kernel/cgroup.c | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 37d94a2..eaffc08 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -1477,6 +1477,7 @@ static struct dentry *cgroup_mount(struct file_system_type *fs_type,
struct cgroup_sb_opts opts;
struct dentry *dentry;
int ret;
+ bool new_root = false;

/*
* The first time anyone tries to mount a cgroup, enable the list
@@ -1536,6 +1537,10 @@ retry:
* destruction to complete so that the subsystems are free.
* We can use wait_queue for the wait but this path is
* super cold. Let's just sleep for a bit and retry.
+ *
+ * Take a reference so root won't be freed after we drop
+ * cgroup mutexes, and drop it after we've done the real
+ * mount through kernfs.
*/
if (!atomic_inc_not_zero(&root->top_cgroup.refcnt)) {
mutex_unlock(&cgroup_mutex);
@@ -1551,6 +1556,7 @@ retry:
}

/* no such thing, create a new one */
+ new_root = true;
root = cgroup_root_from_opts(&opts);
if (IS_ERR(root)) {
ret = PTR_ERR(root);
@@ -1572,7 +1578,7 @@ out_unlock:
return ERR_PTR(ret);

dentry = kernfs_mount(fs_type, flags, root->kf_root);
- if (IS_ERR(dentry))
+ if (IS_ERR(dentry) || !new_root)
cgroup_put(&root->top_cgroup);
return dentry;
}
--
1.8.0.2

2014-02-15 09:31:50

by Zefan Li

Subject: Re: [PATCH v2] cgroup: fix top cgroup refcnt leak

On 2014/2/15 12:03, Li Zefan wrote:
> If we mount the same cgroupfs in serveral mount points, and then
> umount all of them, kill_sb() will be called only once.
>
> Therefore it's wrong to increment top_cgroup's refcnt when we find
> an existing cgroup_root.
>
> Try:
> # mount -t cgroup -o cpuacct xxx /cgroup
> # mount -t cgroup -o cpuacct xxx /cgroup2
> # cat /proc/cgroups | grep cpuacct
> cpuacct 2 1 1
> # umount /cgroup
> # umount /cgroup2
> # cat /proc/cgroups | grep cpuacct
> cpuacct 2 1 1
>
> You'll see cgroupfs will never be freed.
>
> v2: change to take the refcnt and drop it after kernfs_mount().
>
> Signed-off-by: Li Zefan <[email protected]>

Please hold off applying this patch. It's still buggy.