cgroup: fix potential deadlock in pre_destroy

As Balbir pointed out, memcg's pre_destroy handler has potential deadlock.

It has following lock sequence.

	cgroup_mutex (cgroup_rmdir)
	    -> pre_destroy -> mem_cgroup_pre_destroy-> force_empty
		-> cpu_hotplug.lock. (lru_add_drain_all->
				      schedule_work->
                                      get_online_cpus)

But, cpuset has following.
	cpu_hotplug.lock (call notifier)
		-> cgroup_mutex. (within notifier)

Then, this lock sequence should be fixed.

Considering how pre_destroy works, it's not necessary to holding
cgroup_mutex() while calling it.

As a side effect, we don't have to wait at this mutex while memcg's
force_empty works.(it can be long when there are tons of pages.)

Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Acked-by: Balbir Singh <[email protected]>
Cc: Li Zefan <[email protected]>
Cc: Paul Menage <[email protected]>
Cc: Daisuke Nishimura <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kernel/cgroup.c

@@ -2472,18 +2472,22 @@ static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
 		mutex_unlock(&cgroup_mutex);
 		return -EBUSY;
 	}
-
-	parent = cgrp->parent;
-	root = cgrp->root;
-	sb = root->sb;
+	mutex_unlock(&cgroup_mutex);
 
 	/*
 	 * Call pre_destroy handlers of subsys. Notify subsystems
 	 * that rmdir() request comes.
 	 */
 	cgroup_call_pre_destroy(cgrp);
 
-	if (cgroup_has_css_refs(cgrp)) {
+	mutex_lock(&cgroup_mutex);
+	parent = cgrp->parent;
+	root = cgrp->root;
+	sb = root->sb;
+
+	if (atomic_read(&cgrp->count)
+	    || !list_empty(&cgrp->children)
+	    || cgroup_has_css_refs(cgrp)) {
 		mutex_unlock(&cgroup_mutex);
 		return -EBUSY;
 	}