[PATCH -next v2 3/3] blk-cgroup: synchronize pd_free_fn() from blkg_free_workfn() and blkcg_deactivate_policy()

[Yu Kuai <yukuai1@xxxxxxxxxxxxxxx>]

From: Yu Kuai <yukuai3@xxxxxxxxxx>

Currently parent pd can be freed before child pd:

t1: remove cgroup C1
blkcg_destroy_blkgs
 blkg_destroy
  list_del_init(&blkg->q_node)
  // remove blkg from queue list
  percpu_ref_kill(&blkg->refcnt)
   blkg_release
    call_rcu

t2: from t1
__blkg_release
 blkg_free
  schedule_work
			t4: deactivate policy
			blkcg_deactivate_policy
			 pd_free_fn
			 // parent of C1 is freed first
t3: from t2
 blkg_free_workfn
  pd_free_fn

If policy(for example, ioc_timer_fn() from iocost) access parent pd from
child pd after pd_offline_fn(), then UAF can be triggered.

Fix the problem by delaying 'list_del_init(&blkg->q_node)' from
blkg_destroy() to blkg_free_workfn(), and use a new disk level mutex to
protect blkg_free_workfn() and blkcg_deactivate_policy).

Signed-off-by: Yu Kuai <yukuai3@xxxxxxxxxx>
---
 block/blk-cgroup.c     | 33 +++++++++++++++++++++++++++------
 include/linux/blkdev.h |  1 +
 2 files changed, 28 insertions(+), 6 deletions(-)

diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index 75f3c4460715..4098e8030e01 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -118,16 +118,26 @@ static void blkg_free_workfn(struct work_struct *work)
 {
 	struct blkcg_gq *blkg = container_of(work, struct blkcg_gq,
 					     free_work);
+	struct request_queue *q = blkg->q;
 	int i;
 
+	if (q)
+		mutex_lock(&q->blkcg_mutex);
+
 	for (i = 0; i < BLKCG_MAX_POLS; i++)
 		if (blkg->pd[i])
 			blkcg_policy[i]->pd_free_fn(blkg->pd[i]);
 
 	if (blkg->parent)
 		blkg_put(blkg->parent);
-	if (blkg->q)
-		blk_put_queue(blkg->q);
+
+	if (q) {
+		if (!list_empty(&blkg->q_node))
+			list_del_init(&blkg->q_node);
+		mutex_unlock(&q->blkcg_mutex);
+		blk_put_queue(q);
+	}
+
 	free_percpu(blkg->iostat_cpu);
 	percpu_ref_exit(&blkg->refcnt);
 	kfree(blkg);
@@ -462,9 +472,14 @@ static void blkg_destroy(struct blkcg_gq *blkg)
 	lockdep_assert_held(&blkg->q->queue_lock);
 	lockdep_assert_held(&blkcg->lock);
 
-	/* Something wrong if we are trying to remove same group twice */
-	WARN_ON_ONCE(list_empty(&blkg->q_node));
-	WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
+	/*
+	 * blkg is removed from queue list in blkg_free_workfn(), hence this
+	 * function can be called from blkcg_destroy_blkgs() first, and then
+	 * before blkg_free_workfn(), this function can be called again in
+	 * blkg_destroy_all().
+	 */
+	if (hlist_unhashed(&blkg->blkcg_node))
+		return;
 
 	for (i = 0; i < BLKCG_MAX_POLS; i++) {
 		struct blkcg_policy *pol = blkcg_policy[i];
@@ -478,8 +493,11 @@ static void blkg_destroy(struct blkcg_gq *blkg)
 
 	blkg->online = false;
 
+	/*
+	 * Delay deleting list blkg->q_node to blkg_free_workfn() to synchronize
+	 * pd_free_fn() from blkg_free_workfn() and blkcg_deactivate_policy().
+	 */
 	radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
-	list_del_init(&blkg->q_node);
 	hlist_del_init_rcu(&blkg->blkcg_node);
 
 	/*
@@ -1280,6 +1298,7 @@ int blkcg_init_disk(struct gendisk *disk)
 	int ret;
 
 	INIT_LIST_HEAD(&q->blkg_list);
+	mutex_init(&q->blkcg_mutex);
 
 	new_blkg = blkg_alloc(&blkcg_root, disk, GFP_KERNEL);
 	if (!new_blkg)
@@ -1520,6 +1539,7 @@ void blkcg_deactivate_policy(struct request_queue *q,
 	if (queue_is_mq(q))
 		blk_mq_freeze_queue(q);
 
+	mutex_lock(&q->blkcg_mutex);
 	spin_lock_irq(&q->queue_lock);
 
 	__clear_bit(pol->plid, q->blkcg_pols);
@@ -1538,6 +1558,7 @@ void blkcg_deactivate_policy(struct request_queue *q,
 	}
 
 	spin_unlock_irq(&q->queue_lock);
+	mutex_unlock(&q->blkcg_mutex);
 
 	if (queue_is_mq(q))
 		blk_mq_unfreeze_queue(q);
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index b87ed829ab94..53ae0a7fe377 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -485,6 +485,7 @@ struct request_queue {
 	DECLARE_BITMAP		(blkcg_pols, BLKCG_MAX_POLS);
 	struct blkcg_gq		*root_blkg;
 	struct list_head	blkg_list;
+	struct mutex		blkcg_mutex;
 #endif
 
 	struct queue_limits	limits;
-- 
2.31.1