[PATCH 1/2] cgroup: fix exit() vsrmdir() race

[PATCH 1/2] cgroup: fix exit() vsrmdir() race

[Li Zefan]

In cgroup_exit() put_css_set_taskexit() is called without any lock,
which might lead to accessing a freed cgroup:

thread1                           thread2
---------------------------------------------
exit()
  cgroup_exit()
    put_css_set_taskexit()
      atomic_dec(cgrp->count);
                                   rmdir();
      /* not safe !! */
      check_for_release(cgrp);

rcu_read_lock() can be used to make sure the cgroup is alive.

Signed-off-by: Li Zefan <lizefan@huawei.com>
---
 kernel/cgroup.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 368cff1..09c5869 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -422,12 +422,20 @@ static void __put_css_set(struct css_set *cg, int taskexit)
 		struct cgroup *cgrp = link->cgrp;
 		list_del(&link->cg_link_list);
 		list_del(&link->cgrp_link_list);
+
+		/*
+		 * We may not be holding cgroup_mutex, and if cgrp->count is
+		 * dropped to 0 the cgroup can be destroyed at any time, hence
+		 * rcu_read_lock is used to keep it alive.
+		 */
+		rcu_read_lock();
 		if (atomic_dec_and_test(&cgrp->count) &&
 		    notify_on_release(cgrp)) {
 			if (taskexit)
 				set_bit(CGRP_RELEASABLE, &cgrp->flags);
 			check_for_release(cgrp);
 		}
+		rcu_read_unlock();
 
 		kfree(link);
 	}
-- 
1.8.0.2

[PATCH 2/2] cgroup: remove a NULL check in cgroup_exit()

[Li Zefan]

init_task.cgroups is initialized at boot phase, and whenver a ask
is forked, it's cgroups pointer is inherited from its parent, and
it's never set to NULL afterwards.

Signed-off-by: Li Zefan <lizefan@huawei.com>
---
 kernel/cgroup.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 09c5869..5d4c92e 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -5011,8 +5011,7 @@ void cgroup_exit(struct task_struct *tsk, int run_callbacks)
 	}
 	task_unlock(tsk);
 
-	if (cg)
-		put_css_set_taskexit(cg);
+	put_css_set_taskexit(cg);
 }
 
 /**
-- 
1.8.0.2

Re: [PATCH 1/2] cgroup: fix exit() vsrmdir() race

[Tejun Heo]

On Thu, Jan 24, 2013 at 02:43:28PM +0800, Li Zefan wrote:
> In cgroup_exit() put_css_set_taskexit() is called without any lock,
> which might lead to accessing a freed cgroup:
> 
> thread1                           thread2
> ---------------------------------------------
> exit()
>   cgroup_exit()
>     put_css_set_taskexit()
>       atomic_dec(cgrp->count);
>                                    rmdir();
>       /* not safe !! */
>       check_for_release(cgrp);
> 
> rcu_read_lock() can be used to make sure the cgroup is alive.
> 
> Signed-off-by: Li Zefan <lizefan@huawei.com>

Applying to cgroup/for-3.8-fixes w/ stable cc'd.

* Can you please add WARN_ON_ONCE(!rcu_read_lock_held()) in
  check_for_release()?

* cgroup_release_agent() seems broken too.  It's accessing cgrp after
  removing it from release_list.  Can you please fix it too?

Thanks.

-- 
tejun

Re: [PATCH 2/2] cgroup: remove a NULL check in cgroup_exit()

[Tejun Heo]

On Thu, Jan 24, 2013 at 02:43:51PM +0800, Li Zefan wrote:
> init_task.cgroups is initialized at boot phase, and whenver a ask
> is forked, it's cgroups pointer is inherited from its parent, and
> it's never set to NULL afterwards.
> 
> Signed-off-by: Li Zefan <lizefan@huawei.com>

Applied to cgroup/for-3.9.

Thanks.

-- 
tejun

Re: [PATCH 1/2] cgroup: fix exit() vsrmdir() race

[Tejun Heo]

On Thu, Jan 24, 2013 at 10:19:23AM -0800, Tejun Heo wrote:
> * cgroup_release_agent() seems broken too.  It's accessing cgrp after
>   removing it from release_list.  Can you please fix it too?

Oh, we should be okay as we're holding cgroup_mutex there.  Please
forget about the above.

Thanks.

-- 
tejun