[PATCH] CGroups: Fix a race between rmdir and remount

[PATCH] CGroups: Fix a race between rmdir and remount

[Paul Menage]

Fix a race between rmdir and remount

When a cgroup is removed, it's unlinked from its parent's children
list, but not actually freed until the last dentry on it is released
(at which point cgrp->root->number_of_cgroups is decremented).

Currently rebind_subsystems checks for the top cgroup's child list
being empty in order to rebind subsystems into or out of a hierarchy -
this can result in the set of subsystems bound to a hierarchy being
different than the set of subsystems with state in the
removed-but-not-freed cgroup.

The simplest fix for this is to forbid remounts that change the set of
subsystems on a hierarchy that has removed-but-not-freed cgroups.

This bug can be reproduced via:

mkdir /mnt/cg
mount -t cgroup -o ns,freezer cgroup /mnt/cg
mkdir /mnt/cg/foo
sleep 1h < /mnt/cg/foo &
rmdir /mnt/cg/foo
mount -t cgroup -o remount,ns,devices,freezer cgroup /mnt/cg
kill $!

Signed-off-by: Paul Menage <menage@google.com>

---
 kernel/cgroup.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Index: hierarchy_lock-mmotm-2008-12-09/kernel/cgroup.c
===================================================================
--- hierarchy_lock-mmotm-2008-12-09.orig/kernel/cgroup.c
+++ hierarchy_lock-mmotm-2008-12-09/kernel/cgroup.c
@@ -702,7 +702,7 @@ static int rebind_subsystems(struct cgro
 	 * any child cgroups exist. This is theoretically supportable
 	 * but involves complex error handling, so it's being left until
 	 * later */
-	if (!list_empty(&cgrp->children))
+	if (root->number_of_cgroups > 1)
 		return -EBUSY;

 	/* Process each subsystem */

Re: [PATCH] CGroups: Fix a race between rmdir and remount

[Li Zefan]

Paul Menage wrote:
> Fix a race between rmdir and remount
> 
> When a cgroup is removed, it's unlinked from its parent's children
> list, but not actually freed until the last dentry on it is released
> (at which point cgrp->root->number_of_cgroups is decremented).
> 
> Currently rebind_subsystems checks for the top cgroup's child list
> being empty in order to rebind subsystems into or out of a hierarchy -
> this can result in the set of subsystems bound to a hierarchy being
> different than the set of subsystems with state in the
> removed-but-not-freed cgroup.
> 
> The simplest fix for this is to forbid remounts that change the set of
> subsystems on a hierarchy that has removed-but-not-freed cgroups.
> 
> This bug can be reproduced via:
> 

Seems this bug is revealed by my patch:

cgroups-remove-some-redundant-null-checks.patch
(http://marc.info/?l=linux-mm-commits&m=122730918427045&w=2)

@@ -611,10 +611,8 @@ static void cgroup_diput(struct dentry *
...
-		for_each_subsys(cgrp->root, ss) {
-			if (cgrp->subsys[ss->subsys_id])
-				ss->destroy(ss, cgrp);
-		}
+		for_each_subsys(cgrp->root, ss)
+			ss->destroy(ss, cgrp);

But this patch is not guilty. :)

The original code leaked memory silently due to this race, if the remount
removes some subsystems from the hierarchy.

> mkdir /mnt/cg
> mount -t cgroup -o ns,freezer cgroup /mnt/cg
> mkdir /mnt/cg/foo
> sleep 1h < /mnt/cg/foo &
> rmdir /mnt/cg/foo
> mount -t cgroup -o remount,ns,devices,freezer cgroup /mnt/cg
> kill $!
> 
> Signed-off-by: Paul Menage <menage@google.com>
> 

Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>

> ---
>  kernel/cgroup.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> Index: hierarchy_lock-mmotm-2008-12-09/kernel/cgroup.c
> ===================================================================
> --- hierarchy_lock-mmotm-2008-12-09.orig/kernel/cgroup.c
> +++ hierarchy_lock-mmotm-2008-12-09/kernel/cgroup.c
> @@ -702,7 +702,7 @@ static int rebind_subsystems(struct cgro
>  	 * any child cgroups exist. This is theoretically supportable
>  	 * but involves complex error handling, so it's being left until
>  	 * later */
> -	if (!list_empty(&cgrp->children))
> +	if (root->number_of_cgroups > 1)
>  		return -EBUSY;
> 
>  	/* Process each subsystem */