blk-wbt: fix possible deadlock to nest pcpu_alloc_mutex under q_usage_counter

If wbt is disabled by default and user configures wbt by sysfs, queue
will be frozen first and then pcpu_alloc_mutex will be held in
blk_stat_alloc_callback().

Fix this problem by allocating memory first before queue frozen.

Signed-off-by: Yu Kuai <yukuai@fnnas.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

1 files changed, 63 insertions, 45 deletions

diff --git a/block/blk-wbt.c b/block/blk-wbt.c
index 0a37d97bda75..665760274e60 100644
--- a/block/blk-wbt.c
+++ b/block/blk-wbt.c
@@ -93,7 +93,7 @@ struct rq_wb {
 	struct rq_depth rq_depth;
 };
 
-static int wbt_init(struct gendisk *disk);
+static int wbt_init(struct gendisk *disk, struct rq_wb *rwb);
 
 static inline struct rq_wb *RQWB(struct rq_qos *rqos)
 {
@@ -698,6 +698,41 @@ static void wbt_requeue(struct rq_qos *rqos, struct request *rq)
 	}
 }
 
+static int wbt_data_dir(const struct request *rq)
+{
+	const enum req_op op = req_op(rq);
+
+	if (op == REQ_OP_READ)
+		return READ;
+	else if (op_is_write(op))
+		return WRITE;
+
+	/* don't account */
+	return -1;
+}
+
+static struct rq_wb *wbt_alloc(void)
+{
+	struct rq_wb *rwb = kzalloc(sizeof(*rwb), GFP_KERNEL);
+
+	if (!rwb)
+		return NULL;
+
+	rwb->cb = blk_stat_alloc_callback(wb_timer_fn, wbt_data_dir, 2, rwb);
+	if (!rwb->cb) {
+		kfree(rwb);
+		return NULL;
+	}
+
+	return rwb;
+}
+
+static void wbt_free(struct rq_wb *rwb)
+{
+	blk_stat_free_callback(rwb->cb);
+	kfree(rwb);
+}
+
 /*
  * Enable wbt if defaults are configured that way
  */
@@ -739,8 +774,17 @@ EXPORT_SYMBOL_GPL(wbt_enable_default);
 
 void wbt_init_enable_default(struct gendisk *disk)
 {
-	if (__wbt_enable_default(disk))
-		WARN_ON_ONCE(wbt_init(disk));
+	struct rq_wb *rwb;
+
+	if (!__wbt_enable_default(disk))
+		return;
+
+	rwb = wbt_alloc();
+	if (WARN_ON_ONCE(!rwb))
+		return;
+
+	if (WARN_ON_ONCE(wbt_init(disk, rwb)))
+		wbt_free(rwb);
 }
 
 static u64 wbt_default_latency_nsec(struct request_queue *q)
@@ -754,19 +798,6 @@ static u64 wbt_default_latency_nsec(struct request_queue *q)
 	return 2000000ULL;
 }
 
-static int wbt_data_dir(const struct request *rq)
-{
-	const enum req_op op = req_op(rq);
-
-	if (op == REQ_OP_READ)
-		return READ;
-	else if (op_is_write(op))
-		return WRITE;
-
-	/* don't account */
-	return -1;
-}
-
 static void wbt_queue_depth_changed(struct rq_qos *rqos)
 {
 	RQWB(rqos)->rq_depth.queue_depth = blk_queue_depth(rqos->disk->queue);
@@ -778,8 +809,7 @@ static void wbt_exit(struct rq_qos *rqos)
 	struct rq_wb *rwb = RQWB(rqos);
 
 	blk_stat_remove_callback(rqos->disk->queue, rwb->cb);
-	blk_stat_free_callback(rwb->cb);
-	kfree(rwb);
+	wbt_free(rwb);
 }
 
 /*
@@ -903,22 +933,11 @@ static const struct rq_qos_ops wbt_rqos_ops = {
 #endif
 };
 
-static int wbt_init(struct gendisk *disk)
+static int wbt_init(struct gendisk *disk, struct rq_wb *rwb)
 {
 	struct request_queue *q = disk->queue;
-	struct rq_wb *rwb;
-	int i;
 	int ret;
-
-	rwb = kzalloc(sizeof(*rwb), GFP_KERNEL);
-	if (!rwb)
-		return -ENOMEM;
-
-	rwb->cb = blk_stat_alloc_callback(wb_timer_fn, wbt_data_dir, 2, rwb);
-	if (!rwb->cb) {
-		kfree(rwb);
-		return -ENOMEM;
-	}
+	int i;
 
 	for (i = 0; i < WBT_NUM_RWQ; i++)
 		rq_wait_init(&rwb->rq_wait[i]);
@@ -938,38 +957,38 @@ static int wbt_init(struct gendisk *disk)
 	ret = rq_qos_add(&rwb->rqos, disk, RQ_QOS_WBT, &wbt_rqos_ops);
 	mutex_unlock(&q->rq_qos_mutex);
 	if (ret)
-		goto err_free;
+		return ret;
 
 	blk_stat_add_callback(q, rwb->cb);
-
 	return 0;
-
-err_free:
-	blk_stat_free_callback(rwb->cb);
-	kfree(rwb);
-	return ret;
-
 }
 
 int wbt_set_lat(struct gendisk *disk, s64 val)
 {
 	struct request_queue *q = disk->queue;
+	struct rq_qos *rqos = wbt_rq_qos(q);
+	struct rq_wb *rwb = NULL;
 	unsigned int memflags;
-	struct rq_qos *rqos;
 	int ret = 0;
 
+	if (!rqos) {
+		rwb = wbt_alloc();
+		if (!rwb)
+			return -ENOMEM;
+	}
+
 	/*
 	 * Ensure that the queue is idled, in case the latency update
 	 * ends up either enabling or disabling wbt completely. We can't
 	 * have IO inflight if that happens.
 	 */
 	memflags = blk_mq_freeze_queue(q);
-
-	rqos = wbt_rq_qos(q);
 	if (!rqos) {
-		ret = wbt_init(disk);
-		if (ret)
+		ret = wbt_init(disk, rwb);
+		if (ret) {
+			wbt_free(rwb);
 			goto out;
+		}
 	}
 
 	if (val == -1)
@@ -989,6 +1008,5 @@ int wbt_set_lat(struct gendisk *disk, s64 val)
 	blk_mq_unquiesce_queue(q);
 out:
 	blk_mq_unfreeze_queue(q, memflags);
-
 	return ret;
 }