diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index fceddbdfdd3d..dac6653d4cce 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -351,6 +351,7 @@ static inline unsigned long btrfs_chunk_item_size(int num_stripes)
 #define BTRFS_FS_STATE_ERROR		0
 #define BTRFS_FS_STATE_REMOUNTING	1
 #define BTRFS_FS_STATE_TRANS_ABORTED	2
+#define BTRFS_FS_STATE_DEV_REPLACING	3
 
 /* Super block flags */
 /* Errors detected */
@@ -1674,6 +1675,9 @@ struct btrfs_fs_info {
 
 	atomic_t mutually_exclusive_operation_running;
 
+	struct percpu_counter bio_counter;
+	wait_queue_head_t replace_wait;
+
 	struct semaphore uuid_tree_rescan_sem;
 	unsigned int update_uuid_tree_gen:1;
 };
@@ -4008,6 +4012,11 @@ int btrfs_scrub_cancel_dev(struct btrfs_fs_info *info,
 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
 			 struct btrfs_scrub_progress *progress);
 
+/* dev-replace.c */
+void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info);
+void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info);
+void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info);
+
 /* reada.c */
 struct reada_control {
 	struct btrfs_root	*root;		/* tree to prefetch */
diff --git a/fs/btrfs/dev-replace.c b/fs/btrfs/dev-replace.c
index b20d59e5e5dd..ec1c3f3a775d 100644
--- a/fs/btrfs/dev-replace.c
+++ b/fs/btrfs/dev-replace.c
@@ -431,6 +431,35 @@ leave_no_lock:
 	return ret;
 }
 
+/*
+ * blocked until all flighting bios are finished.
+ */
+static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
+{
+	s64 writers;
+	DEFINE_WAIT(wait);
+
+	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
+	do {
+		prepare_to_wait(&fs_info->replace_wait, &wait,
+				TASK_UNINTERRUPTIBLE);
+		writers = percpu_counter_sum(&fs_info->bio_counter);
+		if (writers)
+			schedule();
+		finish_wait(&fs_info->replace_wait, &wait);
+	} while (writers);
+}
+
+/*
+ * we have removed target device, it is safe to allow new bios request.
+ */
+static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
+{
+	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
+	if (waitqueue_active(&fs_info->replace_wait))
+		wake_up(&fs_info->replace_wait);
+}
+
 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 				       int scrub_ret)
 {
@@ -458,12 +487,6 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 	src_device = dev_replace->srcdev;
 	btrfs_dev_replace_unlock(dev_replace);
 
-	/* replace old device with new one in mapping tree */
-	if (!scrub_ret)
-		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
-								src_device,
-								tgt_device);
-
 	/*
 	 * flush all outstanding I/O and inode extent mappings before the
 	 * copy operation is declared as being finished
@@ -495,7 +518,12 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 	dev_replace->time_stopped = get_seconds();
 	dev_replace->item_needs_writeback = 1;
 
-	if (scrub_ret) {
+	/* replace old device with new one in mapping tree */
+	if (!scrub_ret) {
+		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
+								src_device,
+								tgt_device);
+	} else {
 		printk_in_rcu(KERN_ERR
 			      "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
 			      src_device->missing ? "<missing disk>" :
@@ -534,8 +562,12 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 		fs_info->fs_devices->latest_bdev = tgt_device->bdev;
 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 
+	btrfs_rm_dev_replace_blocked(fs_info);
+
 	btrfs_rm_dev_replace_srcdev(fs_info, src_device);
 
+	btrfs_rm_dev_replace_unblocked(fs_info);
+
 	/*
 	 * this is again a consistent state where no dev_replace procedure
 	 * is running, the target device is part of the filesystem, the
@@ -865,3 +897,31 @@ void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
 		mutex_unlock(&dev_replace->lock_management_lock);
 	}
 }
+
+void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
+{
+	percpu_counter_inc(&fs_info->bio_counter);
+}
+
+void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info)
+{
+	percpu_counter_dec(&fs_info->bio_counter);
+
+	if (waitqueue_active(&fs_info->replace_wait))
+		wake_up(&fs_info->replace_wait);
+}
+
+void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
+{
+	DEFINE_WAIT(wait);
+again:
+	percpu_counter_inc(&fs_info->bio_counter);
+	if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) {
+		btrfs_bio_counter_dec(fs_info);
+		wait_event(fs_info->replace_wait,
+			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
+				     &fs_info->fs_state));
+		goto again;
+	}
+
+}
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index fcf367581073..0cafacb07b43 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -2136,10 +2136,16 @@ int open_ctree(struct super_block *sb,
 		goto fail_dirty_metadata_bytes;
 	}
 
+	ret = percpu_counter_init(&fs_info->bio_counter, 0);
+	if (ret) {
+		err = ret;
+		goto fail_delalloc_bytes;
+	}
+
 	fs_info->btree_inode = new_inode(sb);
 	if (!fs_info->btree_inode) {
 		err = -ENOMEM;
-		goto fail_delalloc_bytes;
+		goto fail_bio_counter;
 	}
 
 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
@@ -2214,6 +2220,7 @@ int open_ctree(struct super_block *sb,
 	atomic_set(&fs_info->scrub_pause_req, 0);
 	atomic_set(&fs_info->scrubs_paused, 0);
 	atomic_set(&fs_info->scrub_cancel_req, 0);
+	init_waitqueue_head(&fs_info->replace_wait);
 	init_waitqueue_head(&fs_info->scrub_pause_wait);
 	fs_info->scrub_workers_refcnt = 0;
 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
@@ -2966,6 +2973,8 @@ fail_iput:
 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
 
 	iput(fs_info->btree_inode);
+fail_bio_counter:
+	percpu_counter_destroy(&fs_info->bio_counter);
 fail_delalloc_bytes:
 	percpu_counter_destroy(&fs_info->delalloc_bytes);
 fail_dirty_metadata_bytes:
@@ -3613,6 +3622,7 @@ int close_ctree(struct btrfs_root *root)
 
 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
 	percpu_counter_destroy(&fs_info->delalloc_bytes);
+	percpu_counter_destroy(&fs_info->bio_counter);
 	bdi_destroy(&fs_info->bdi);
 	cleanup_srcu_struct(&fs_info->subvol_srcu);
 
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index b68afe32419f..07629e99809a 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -5263,6 +5263,7 @@ int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
 static void btrfs_end_bio(struct bio *bio, int err)
 {
 	struct btrfs_bio *bbio = bio->bi_private;
+	struct btrfs_device *dev = bbio->stripes[0].dev;
 	int is_orig_bio = 0;
 
 	if (err) {
@@ -5270,7 +5271,6 @@ static void btrfs_end_bio(struct bio *bio, int err)
 		if (err == -EIO || err == -EREMOTEIO) {
 			unsigned int stripe_index =
 				btrfs_io_bio(bio)->stripe_index;
-			struct btrfs_device *dev;
 
 			BUG_ON(stripe_index >= bbio->num_stripes);
 			dev = bbio->stripes[stripe_index].dev;
@@ -5292,6 +5292,8 @@ static void btrfs_end_bio(struct bio *bio, int err)
 	if (bio == bbio->orig_bio)
 		is_orig_bio = 1;
 
+	btrfs_bio_counter_dec(bbio->fs_info);
+
 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
 		if (!is_orig_bio) {
 			bio_put(bio);
@@ -5440,6 +5442,9 @@ static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
 	}
 #endif
 	bio->bi_bdev = dev->bdev;
+
+	btrfs_bio_counter_inc_noblocked(root->fs_info);
+
 	if (async)
 		btrfs_schedule_bio(root, dev, rw, bio);
 	else
@@ -5508,28 +5513,38 @@ int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
 	length = bio->bi_size;
 	map_length = length;
 
+	btrfs_bio_counter_inc_blocked(root->fs_info);
 	ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
 			      mirror_num, &raid_map);
-	if (ret) /* -ENOMEM */
+	if (ret) {
+		btrfs_bio_counter_dec(root->fs_info);
 		return ret;
+	}
 
 	total_devs = bbio->num_stripes;
 	bbio->orig_bio = first_bio;
 	bbio->private = first_bio->bi_private;
 	bbio->end_io = first_bio->bi_end_io;
+	bbio->fs_info = root->fs_info;
 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
 
 	if (raid_map) {
 		/* In this case, map_length has been set to the length of
 		   a single stripe; not the whole write */
 		if (rw & WRITE) {
-			return raid56_parity_write(root, bio, bbio,
-						   raid_map, map_length);
+			ret = raid56_parity_write(root, bio, bbio,
+						  raid_map, map_length);
 		} else {
-			return raid56_parity_recover(root, bio, bbio,
-						     raid_map, map_length,
-						     mirror_num);
+			ret = raid56_parity_recover(root, bio, bbio,
+						    raid_map, map_length,
+						    mirror_num);
 		}
+		/*
+		 * FIXME, replace dosen't support raid56 yet, please fix
+		 * it in the future.
+		 */
+		btrfs_bio_counter_dec(root->fs_info);
+		return ret;
 	}
 
 	if (map_length < length) {
@@ -5571,6 +5586,7 @@ int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
 				  async_submit);
 		dev_nr++;
 	}
+	btrfs_bio_counter_dec(root->fs_info);
 	return 0;
 }
 
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index 8b3cd142b373..80754f9dd3df 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -192,6 +192,7 @@ typedef void (btrfs_bio_end_io_t) (struct btrfs_bio *bio, int err);
 
 struct btrfs_bio {
 	atomic_t stripes_pending;
+	struct btrfs_fs_info *fs_info;
 	bio_end_io_t *end_io;
 	struct bio *orig_bio;
 	void *private;