diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 559170464d7c50f43d6804d59dfd9f73e57a707f..e96634a725c3e441a311e36b40b43532031bf5c5 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -718,12 +718,13 @@ int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr,
 	return count;
 }
 
-void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr,
+int btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr,
 			      const u64 range_start, const u64 range_len)
 {
 	struct btrfs_root *root;
 	struct list_head splice;
 	int done;
+	int total_done = 0;
 
 	INIT_LIST_HEAD(&splice);
 
@@ -742,6 +743,7 @@ void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr,
 		done = btrfs_wait_ordered_extents(root, nr,
 						  range_start, range_len);
 		btrfs_put_fs_root(root);
+		total_done += done;
 
 		spin_lock(&fs_info->ordered_root_lock);
 		if (nr != -1) {
@@ -752,6 +754,8 @@ void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr,
 	list_splice_tail(&splice, &fs_info->ordered_roots);
 	spin_unlock(&fs_info->ordered_root_lock);
 	mutex_unlock(&fs_info->ordered_operations_mutex);
+
+	return total_done;
 }
 
 /*
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
index 2049c9be85ee304f41b11753db609fd802de7fd6..451507776ff59f87fdc286db823f4d08ff71899a 100644
--- a/fs/btrfs/ordered-data.h
+++ b/fs/btrfs/ordered-data.h
@@ -199,7 +199,7 @@ int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
 			   u32 *sum, int len);
 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr,
 			       const u64 range_start, const u64 range_len);
-void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr,
+int btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr,
 			      const u64 range_start, const u64 range_len);
 void btrfs_get_logged_extents(struct inode *inode,
 			      struct list_head *logged_list,
diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c
index 46d847f66e4b870dcd54dc18d023f8ca321f40fb..1611572d47bd0ff32fdd0e6bed4afa069264f5b6 100644
--- a/fs/btrfs/scrub.c
+++ b/fs/btrfs/scrub.c
@@ -3582,6 +3582,46 @@ int scrub_enumerate_chunks(struct scrub_ctx *sctx,
 		 */
 		scrub_pause_on(fs_info);
 		ret = btrfs_inc_block_group_ro(root, cache);
+		if (!ret && is_dev_replace) {
+			/*
+			 * If we are doing a device replace wait for any tasks
+			 * that started dellaloc right before we set the block
+			 * group to RO mode, as they might have just allocated
+			 * an extent from it or decided they could do a nocow
+			 * write. And if any such tasks did that, wait for their
+			 * ordered extents to complete and then commit the
+			 * current transaction, so that we can later see the new
+			 * extent items in the extent tree - the ordered extents
+			 * create delayed data references (for cow writes) when
+			 * they complete, which will be run and insert the
+			 * corresponding extent items into the extent tree when
+			 * we commit the transaction they used when running
+			 * inode.c:btrfs_finish_ordered_io(). We later use
+			 * the commit root of the extent tree to find extents
+			 * to copy from the srcdev into the tgtdev, and we don't
+			 * want to miss any new extents.
+			 */
+			btrfs_wait_block_group_reservations(cache);
+			btrfs_wait_nocow_writers(cache);
+			ret = btrfs_wait_ordered_roots(fs_info, -1,
+						       cache->key.objectid,
+						       cache->key.offset);
+			if (ret > 0) {
+				struct btrfs_trans_handle *trans;
+
+				trans = btrfs_join_transaction(root);
+				if (IS_ERR(trans))
+					ret = PTR_ERR(trans);
+				else
+					ret = btrfs_commit_transaction(trans,
+								       root);
+				if (ret) {
+					scrub_pause_off(fs_info);
+					btrfs_put_block_group(cache);
+					break;
+				}
+			}
+		}
 		scrub_pause_off(fs_info);
 
 		if (ret == 0) {