5 files changed, 35 insertions, 18 deletions

diff --git a/block/blk-flush.c b/block/blk-flush.c
index 20badd7b9d1b6..9c423e53324a2 100644
--- a/block/blk-flush.c
+++ b/block/blk-flush.c
@@ -73,6 +73,7 @@
 
 #include "blk.h"
 #include "blk-mq.h"
+#include "blk-mq-tag.h"
 
 /* FLUSH/FUA sequences */
 enum {
@@ -226,7 +227,12 @@ static void flush_end_io(struct request *flush_rq, int error)
 	struct blk_flush_queue *fq = blk_get_flush_queue(q, flush_rq->mq_ctx);
 
 	if (q->mq_ops) {
+		struct blk_mq_hw_ctx *hctx;
+
+		/* release the tag's ownership to the req cloned from */
 		spin_lock_irqsave(&fq->mq_flush_lock, flags);
+		hctx = q->mq_ops->map_queue(q, flush_rq->mq_ctx->cpu);
+		blk_mq_tag_set_rq(hctx, flush_rq->tag, fq->orig_rq);
 		flush_rq->tag = -1;
 	}
 
@@ -308,11 +314,18 @@ static bool blk_kick_flush(struct request_queue *q, struct blk_flush_queue *fq)
 
 	/*
 	 * Borrow tag from the first request since they can't
-	 * be in flight at the same time.
+	 * be in flight at the same time. And acquire the tag's
+	 * ownership for flush req.
 	 */
 	if (q->mq_ops) {
+		struct blk_mq_hw_ctx *hctx;
+
 		flush_rq->mq_ctx = first_rq->mq_ctx;
 		flush_rq->tag = first_rq->tag;
+		fq->orig_rq = first_rq;
+
+		hctx = q->mq_ops->map_queue(q, first_rq->mq_ctx->cpu);
+		blk_mq_tag_set_rq(hctx, first_rq->tag, flush_rq);
 	}
 
 	flush_rq->cmd_type = REQ_TYPE_FS;
diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
index 9b6e28830b823..9115c6d59948a 100644
--- a/block/blk-mq-tag.c
+++ b/block/blk-mq-tag.c
@@ -429,7 +429,7 @@ static void bt_for_each(struct blk_mq_hw_ctx *hctx,
 		for (bit = find_first_bit(&bm->word, bm->depth);
 		     bit < bm->depth;
 		     bit = find_next_bit(&bm->word, bm->depth, bit + 1)) {
-		     	rq = blk_mq_tag_to_rq(hctx->tags, off + bit);
+			rq = hctx->tags->rqs[off + bit];
 			if (rq->q == hctx->queue)
 				fn(hctx, rq, data, reserved);
 		}
@@ -453,7 +453,7 @@ static void bt_tags_for_each(struct blk_mq_tags *tags,
 		for (bit = find_first_bit(&bm->word, bm->depth);
 		     bit < bm->depth;
 		     bit = find_next_bit(&bm->word, bm->depth, bit + 1)) {
-			rq = blk_mq_tag_to_rq(tags, off + bit);
+			rq = tags->rqs[off + bit];
 			fn(rq, data, reserved);
 		}
 
diff --git a/block/blk-mq-tag.h b/block/blk-mq-tag.h
index 75893a34237d2..9eb2cf4f01cb8 100644
--- a/block/blk-mq-tag.h
+++ b/block/blk-mq-tag.h
@@ -89,4 +89,16 @@ static inline void blk_mq_tag_idle(struct blk_mq_hw_ctx *hctx)
 	__blk_mq_tag_idle(hctx);
 }
 
+/*
+ * This helper should only be used for flush request to share tag
+ * with the request cloned from, and both the two requests can't be
+ * in flight at the same time. The caller has to make sure the tag
+ * can't be freed.
+ */
+static inline void blk_mq_tag_set_rq(struct blk_mq_hw_ctx *hctx,
+		unsigned int tag, struct request *rq)
+{
+	hctx->tags->rqs[tag] = rq;
+}
+
 #endif
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 81edbd95bda87..f2d67b4047a04 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -559,23 +559,9 @@ void blk_mq_abort_requeue_list(struct request_queue *q)
 }
 EXPORT_SYMBOL(blk_mq_abort_requeue_list);
 
-static inline bool is_flush_request(struct request *rq,
-		struct blk_flush_queue *fq, unsigned int tag)
-{
-	return ((rq->cmd_flags & REQ_FLUSH_SEQ) &&
-			fq->flush_rq->tag == tag);
-}
-
 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag)
 {
-	struct request *rq = tags->rqs[tag];
-	/* mq_ctx of flush rq is always cloned from the corresponding req */
-	struct blk_flush_queue *fq = blk_get_flush_queue(rq->q, rq->mq_ctx);
-
-	if (!is_flush_request(rq, fq, tag))
-		return rq;
-
-	return fq->flush_rq;
+	return tags->rqs[tag];
 }
 EXPORT_SYMBOL(blk_mq_tag_to_rq);
 
diff --git a/block/blk.h b/block/blk.h
index 026d9594142bd..838188b35a83f 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -22,6 +22,12 @@ struct blk_flush_queue {
 	struct list_head	flush_queue[2];
 	struct list_head	flush_data_in_flight;
 	struct request		*flush_rq;
+
+	/*
+	 * flush_rq shares tag with this rq, both can't be active
+	 * at the same time
+	 */
+	struct request		*orig_rq;
 	spinlock_t		mq_flush_lock;
 };