diff --git a/block/blk-throttle.c b/block/blk-throttle.c
index 3836af6d45a20612b44325a5d808e89c14876ff1..d3527af8803ab1501389cf05b6d4d364c139ee10 100644
--- a/block/blk-throttle.c
+++ b/block/blk-throttle.c
@@ -1088,6 +1088,10 @@ static void throtl_bio_end_io(struct bio *bio)
 	struct throtl_grp *tg;
 
 	rcu_read_lock();
+	/* see comments in throtl_bio_stats_start() */
+	if (bio_flagged(bio, BIO_THROTL_STATED))
+		goto out;
+
 	tg = (struct throtl_grp *)bio->bi_tg_private;
 	if (!tg)
 		goto out;
@@ -1096,6 +1100,7 @@ static void throtl_bio_end_io(struct bio *bio)
 				       bio_io_start_time_ns(bio),
 				       bio_op(bio));
 	blkg_put(tg_to_blkg(tg));
+	bio_clear_flag(bio, BIO_THROTL_STATED);
 out:
 	rcu_read_unlock();
 }
@@ -1104,11 +1109,19 @@ static inline void throtl_bio_stats_start(struct bio *bio, struct throtl_grp *tg
 {
 	int op = bio_op(bio);
 
-	if (op == REQ_OP_READ || op == REQ_OP_WRITE) {
+	/*
+	 * It may happen that end_io will be called twice like dm-thin,
+	 * which will save origin end_io first, and call its overwrite
+	 * end_io and then the saved end_io. We use bio flag
+	 * BIO_THROTL_STATED to do only once statistics.
+	 */
+	if ((op == REQ_OP_READ || op == REQ_OP_WRITE) &&
+	    !bio_flagged(bio, BIO_THROTL_STATED)) {
+		blkg_get(tg_to_blkg(tg));
+		bio_set_flag(bio, BIO_THROTL_STATED);
 		bio->bi_tg_end_io = throtl_bio_end_io;
 		bio->bi_tg_private = tg;
 		bio_set_start_time_ns(bio);
-		blkg_get(tg_to_blkg(tg));
 	}
 }
 
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index dedee389966c5adbb80a0ead92da01371e4bcf38..4badbaae2a9cdbb8e3728c207d1f22af127e4940 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -238,6 +238,7 @@ struct bio {
 #define BIO_TRACE_COMPLETION 10	/* bio_endio() should trace the final completion
 				 * of this bio. */
 #define BIO_QUEUE_ENTERED 11	/* can use blk_queue_enter_live() */
+#define BIO_THROTL_STATED 12	/* bio already stated */
 
 /* See BVEC_POOL_OFFSET below before adding new flags */