blk-mq: use static_rqs instead of rqs to iterate tags

hulk inclusion
category: bugfix
bugzilla: 34280
CVE: NA

---------------------------

tags->rqs[] will not been cleaned when free driver tag to avoid
an extra store on a shared area in the per io path. But there
is a window between get driver tag and write tags->rqs[], so we
may see stale rq in tags->rqs[] which may have been freed, as
the following case:

blk_mq_get_request         blk_mq_queue_tag_busy_iter
  -> blk_mq_get_tag
                             -> bt_for_each
                               -> bt_iter
                                 -> rq = tags->rqs[]
                                 -> rq->q
  -> blk_mq_rq_ctx_init
    -> data->hctx->tags->rqs[rq->tag] = rq;

In additiion, tags->rqs[] only contains the requests that get
driver tag. It is not accurate for io-scheduler case when account
busy tags in part_in_flight.

To fix both of them, the blk_mq_queue_tag_busy_iter is changed
in this patch to use tags->static_rqs[] instead of tags->rqs[].
We have to identify whether there is a io scheduler attached to
decide to use hctx->tags or hctx->sched_tags. And we will try to
get a non-zero q_usage_counter before that, then could avoid race
with update nr_hw_queues, switch io-scheduler and even queue cleanup.

Add 'inflight' parameter to determine to iterate in-flight
requests or just busy tags and add a new helper interface
blk_mq_queue_tag_inflight_iter to iterate all of the in-flight
tags and export this interface for drivers.
Signed-off-by: Nyu kuai <yukuai3@huawei.com>
Reviewed-by: NHou Tao <houtao1@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

block/blk-mq-tag.c

@@ -216,37 +216,51 @@ struct bt_iter_data {
 	busy_iter_fn *fn;
 	void *data;
 	bool reserved;
+	bool inflight;
 };
 
 static bool bt_iter(struct sbitmap *bitmap, unsigned int bitnr, void *data)
 {
 	struct bt_iter_data *iter_data = data;
 	struct blk_mq_hw_ctx *hctx = iter_data->hctx;
-	struct blk_mq_tags *tags = hctx->tags;
 	bool reserved = iter_data->reserved;
+	struct blk_mq_tags *tags;
 	struct request *rq;
 
+	tags = hctx->sched_tags ? hctx->sched_tags : hctx->tags;
+
 	if (!reserved)
 		bitnr += tags->nr_reserved_tags;
-	rq = tags->rqs[bitnr];
 
 	/*
-	 * We can hit rq == NULL here, because the tagging functions
-	 * test and set the bit before assining ->rqs[].
+	 * Because tags->rqs[] will not been cleaned when free driver tag
+	 * and there is a window between get driver tag and write tags->rqs[],
+	 * so we may see stale rq in tags->rqs[] which may have been freed.
+	 * Using static_rqs[] is safer.
+	 */
+	rq = tags->static_rqs[bitnr];
+
+	/*
+	 * There is a small window between get tag and blk_mq_rq_ctx_init,
+	 * so rq->q and rq->mq_hctx maybe different.
 	 */
-	if (rq && rq->q == hctx->queue)
+	if (rq && rq->q == hctx->queue &&
+	    (!iter_data->inflight ||
+	     blk_mq_rq_state(rq) == MQ_RQ_IN_FLIGHT))
 		iter_data->fn(hctx, rq, iter_data->data, reserved);
 	return true;
 }
 
-static void bt_for_each(struct blk_mq_hw_ctx *hctx, struct sbitmap_queue *bt,
-			busy_iter_fn *fn, void *data, bool reserved)
+static void bt_for_each(struct blk_mq_hw_ctx *hctx,
+		struct sbitmap_queue *bt, busy_iter_fn *fn,
+		void *data, bool reserved, bool inflight)
 {
 	struct bt_iter_data iter_data = {
 		.hctx = hctx,
 		.fn = fn,
 		.data = data,
 		.reserved = reserved,
+		.inflight = inflight,
 	};
 
 	sbitmap_for_each_set(&bt->sb, bt_iter, &iter_data);
@@ -314,22 +328,23 @@ void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
 }
 EXPORT_SYMBOL(blk_mq_tagset_busy_iter);
 
-void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn,
-		void *priv)
+static void __blk_mq_queue_tag_busy_iter(struct request_queue *q,
+		busy_iter_fn *fn, void *priv, bool inflight)
 {
 	struct blk_mq_hw_ctx *hctx;
 	int i;
 
 	/*
-	 * __blk_mq_update_nr_hw_queues will update the nr_hw_queues and
-	 * queue_hw_ctx after freeze the queue, so we use q_usage_counter
-	 * to avoid race with it.
+	 * Get a reference of the queue unless it has been zero. We use this
+	 * to avoid the race with the code that would modify the hctxs after
+	 * freeze and drain the queue, including updating nr_hw_queues, io
+	 * scheduler switching and queue clean up.
 	 */
 	if (!percpu_ref_tryget(&q->q_usage_counter))
 		return;
 
 	queue_for_each_hw_ctx(q, hctx, i) {
-		struct blk_mq_tags *tags = hctx->tags;
+		struct blk_mq_tags *tags;
 
 		/*
 		 * If not software queues are currently mapped to this
@@ -338,13 +353,45 @@ void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn,
 		if (!blk_mq_hw_queue_mapped(hctx))
 			continue;
 
+		tags = hctx->sched_tags ? hctx->sched_tags : hctx->tags;
+
 		if (tags->nr_reserved_tags)
-			bt_for_each(hctx, &tags->breserved_tags, fn, priv, true);
-		bt_for_each(hctx, &tags->bitmap_tags, fn, priv, false);
+			bt_for_each(hctx, &tags->breserved_tags,
+				    fn, priv, true, inflight);
+		bt_for_each(hctx, &tags->bitmap_tags,
+			    fn, priv, false, inflight);
+		/*
+		 * flush_rq represents the rq with REQ_PREFLUSH and REQ_FUA
+		 * (if FUA is not supported by device) to be issued to
+		 * device. So we need to consider it when iterate inflight
+		 * rqs, but needn't to count it when iterate busy tags.
+		 */
+		if (inflight &&
+		    blk_mq_rq_state(hctx->fq->flush_rq) == MQ_RQ_IN_FLIGHT)
+			fn(hctx, hctx->fq->flush_rq, priv, false);
 	}
 	blk_queue_exit(q);
 }
 
+/*
+ * Iterate all the busy tags including pending and in-flight ones.
+ */
+void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn,
+				void *priv)
+{
+	__blk_mq_queue_tag_busy_iter(q, fn, priv, false);
+}
+
+/*
+ * Iterate all the inflight tags.
+ */
+void blk_mq_queue_tag_inflight_iter(struct request_queue *q,
+				    busy_iter_fn *fn, void *priv)
+{
+	__blk_mq_queue_tag_busy_iter(q, fn, priv, true);
+}
+EXPORT_SYMBOL(blk_mq_queue_tag_inflight_iter);
+
 static int bt_alloc(struct sbitmap_queue *bt, unsigned int depth,
 		    bool round_robin, int node)
 {

block/blk-mq.c

@@ -112,7 +112,7 @@ void blk_mq_in_flight(struct request_queue *q, struct hd_struct *part,
 	struct mq_inflight mi = { .part = part, .inflight = inflight, };
 
 	inflight[0] = inflight[1] = 0;
-	blk_mq_queue_tag_busy_iter(q, blk_mq_check_inflight, &mi);
+	blk_mq_queue_tag_inflight_iter(q, blk_mq_check_inflight, &mi);
 }
 
 static void blk_mq_check_inflight_rw(struct blk_mq_hw_ctx *hctx,
@@ -131,7 +131,7 @@ void blk_mq_in_flight_rw(struct request_queue *q, struct hd_struct *part,
 	struct mq_inflight mi = { .part = part, .inflight = inflight, };
 
 	inflight[0] = inflight[1] = 0;
-	blk_mq_queue_tag_busy_iter(q, blk_mq_check_inflight_rw, &mi);
+	blk_mq_queue_tag_inflight_iter(q, blk_mq_check_inflight_rw, &mi);
 }
 
 void blk_freeze_queue_start(struct request_queue *q)
@@ -875,7 +875,7 @@ static void blk_mq_timeout_work(struct work_struct *work)
 	if (!percpu_ref_tryget(&q->q_usage_counter))
 		return;
 
-	blk_mq_queue_tag_busy_iter(q, blk_mq_check_expired, &next);
+	blk_mq_queue_tag_inflight_iter(q, blk_mq_check_expired, &next);
 
 	if (next != 0) {
 		mod_timer(&q->timeout, next);

include/linux/blk-mq.h

@@ -320,7 +320,8 @@ void blk_freeze_queue_start(struct request_queue *q);
 void blk_mq_freeze_queue_wait(struct request_queue *q);
 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
 				     unsigned long timeout);
-
+void blk_mq_queue_tag_inflight_iter(struct request_queue *q, busy_iter_fn *fn,
+		void *priv);
 int blk_mq_map_queues(struct blk_mq_tag_set *set);
 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);