cgroup: break kernfs active_ref protection in cgroup directory operations

From e1b2dc176f2d5be7952c47a4e4e8f3b06a90db1c Mon Sep 17 00:00:00 2001
From: Tejun Heo <[email protected]>
Date: Thu, 20 Mar 2014 11:10:15 -0400

cgroup_tree_mutex should nest above the kernfs active_ref protection; however, cgroup_create() and cgroup_rename() were grabbing cgroup_tree_mutex while under kernfs active_ref protection. This has actualy possibility to lead to deadlocks in case these operations race against cgroup_rmdir() which invokes kernfs_remove() on directory kernfs_node while holding cgroup_tree_mutex.

Neither cgroup_create() or cgroup_rename() requires active_ref protection. The former already has enough synchronization through cgroup_lock_live_group() and the latter doesn't care, so this can be fixed by updating both functions to break all active_ref protections before grabbing cgroup_tree_mutex.

While this patch fixes the immediate issue, it probably needs further work in the long term - kernfs directories should enable lockdep annotations and maybe the better way to handle this is marking directory nodes as not needing active_ref protection rather than breaking it in each operation.

---
Hey,

Another last minute fix that I just committed. I'm thinking maybe the right thing to do for cgroup directories is implementing a flag similar KERNFS_NO_ACTIVE_REF, which achieves similar thing but doesn't require each operation to break protection explicitly. Anyways, this should be enough for now.

Thanks!

 kernel/cgroup.c | 27 ++++++++++++++++++++++++++-
 1 file changed, 26 insertions(+), 1 deletion(-)

index 98a8045..58c67b3 100644
@@ -2324,6 +2324,14 @@ static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
 	if (cgroup_sane_behavior(cgrp))
 		return -EPERM;
 
+	/*
+	 * We're gonna grab cgroup_tree_mutex which nests outside kernfs
+	 * active_ref.  kernfs_rename() doesn't require active_ref
+	 * protection.  Break them before grabbing cgroup_tree_mutex.
+	 */
+	kernfs_break_active_protection(new_parent);
+	kernfs_break_active_protection(kn);
+
 	mutex_lock(&cgroup_tree_mutex);
 	mutex_lock(&cgroup_mutex);
 

@@ -2331,6 +2339,9 @@ static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
 
 	mutex_unlock(&cgroup_mutex);
 	mutex_unlock(&cgroup_tree_mutex);
+
+	kernfs_unbreak_active_protection(kn);
+	kernfs_unbreak_active_protection(new_parent);
 	return ret;
 }
 

@@ -3778,8 +3789,22 @@ static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
 			umode_t mode)
 {
 	struct cgroup *parent = parent_kn->priv;
+	int ret;
 
-	return cgroup_create(parent, name, mode);
+	/*
+	 * cgroup_create() grabs cgroup_tree_mutex which nests outside
+	 * kernfs active_ref and cgroup_create() already synchronizes
+	 * properly against removal through cgroup_lock_live_group().
+	 * Break it before calling cgroup_create().
+	 */
+	cgroup_get(parent);
+	kernfs_break_active_protection(parent_kn);
+
+	ret = cgroup_create(parent, name, mode);
+
+	kernfs_unbreak_active_protection(parent_kn);
+	cgroup_put(parent);
+	return ret;
 }
 
 /*

1.8.5.3