diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index fc6401a..6016500 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -120,30 +120,26 @@ static int btrfs_dirty_inode(struct inode *inode);
 
 static void __endio_write_update_ordered(struct inode *inode,
 					 u64 offset, u64 bytes,
-					 bool uptodate, bool cleanup);
-
-static inline void btrfs_endio_direct_write_update_ordered(struct inode *inode,
-							   u64 offset,
-							   u64 bytes,
-							   bool uptodate)
-{
-	return __endio_write_update_ordered(inode, offset, bytes, uptodate, false);
-}
+					 bool uptodate);
 
 /*
- * Cleanup all submitted ordered extents in specified range to handle error
- * in cow_file_range() and run_delalloc_nocow().
- * Compression handles error and ordered extent submission by itself,
- * so no need to call this function.
+ * Cleanup all submitted ordered extents in specified range to handle errors
+ * from the fill_dellaloc() callback.
  *
- * NOTE: caller must ensure extent_clear_unlock_delalloc() in error handler
- * doesn't cover any range of submitted ordered extent.
- * Or we will double free metadata for submitted ordered extent.
+ * NOTE: caller must ensure that when an error happens, it can not call
+ * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
+ * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
+ * to be released, which we want to happen only when finishing the ordered
+ * extent (btrfs_finish_ordered_io()). Also note that the caller of the
+ * fill_dealloc() callback already does proper cleanup for the first page of
+ * the range, that is, it invokes the callback writepage_end_io_hook() for the
+ * range of the first page.
  */
 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
 						 u64 offset, u64 bytes)
 {
-	return __endio_write_update_ordered(inode, offset, bytes, false, true);
+	return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
+					    bytes - PAGE_SIZE, false);
 }
 
 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
@@ -958,7 +954,6 @@ static noinline int cow_file_range(struct inode *inode,
 	u64 disk_num_bytes;
 	u64 cur_alloc_size;
 	u64 blocksize = fs_info->sectorsize;
-	u64 orig_start = start;
 	struct btrfs_key ins;
 	struct extent_map *em;
 	int ret = 0;
@@ -1100,7 +1095,6 @@ static noinline int cow_file_range(struct inode *inode,
 				     EXTENT_DELALLOC | EXTENT_DEFRAG,
 				     PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 				     PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
-	btrfs_cleanup_ordered_extents(inode, orig_start, end - orig_start + 1);
 	goto out;
 }
 
@@ -1526,8 +1520,6 @@ static noinline int run_delalloc_nocow(struct inode *inode,
 					     PAGE_CLEAR_DIRTY |
 					     PAGE_SET_WRITEBACK |
 					     PAGE_END_WRITEBACK);
-	if (ret)
-		btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
 	btrfs_free_path(path);
 	return ret;
 }
@@ -1577,6 +1569,8 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,
 		ret = cow_file_range_async(inode, locked_page, start, end,
 					   page_started, nr_written);
 	}
+	if (ret)
+		btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
 	return ret;
 }
 
@@ -8176,7 +8170,7 @@ static void btrfs_endio_direct_read(struct bio *bio)
 
 static void __endio_write_update_ordered(struct inode *inode,
 					 u64 offset, u64 bytes,
-					 bool uptodate, bool cleanup)
+					 bool uptodate)
 {
 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 	struct btrfs_ordered_extent *ordered = NULL;
@@ -8194,16 +8188,6 @@ static void __endio_write_update_ordered(struct inode *inode,
 		func = btrfs_endio_write_helper;
 	}
 
-	/*
-	 * In cleanup case, the first page of the range will be handled
-	 * by end_extent_writepage() when called from __extent_writepage()
-	 *
-	 * So we must skip first page, or we will underflow ordered->bytes_left
-	 */
-	if (cleanup) {
-		ordered_offset += PAGE_SIZE;
-		ordered_bytes -= PAGE_SIZE;
-	}
 again:
 	ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
 						   &ordered_offset,
@@ -8231,10 +8215,10 @@ static void btrfs_endio_direct_write(struct bio *bio)
 	struct btrfs_dio_private *dip = bio->bi_private;
 	struct bio *dio_bio = dip->dio_bio;
 
-	btrfs_endio_direct_write_update_ordered(dip->inode,
-						dip->logical_offset,
-						dip->bytes,
-						!bio->bi_error);
+	__endio_write_update_ordered(dip->inode,
+				     dip->logical_offset,
+				     dip->bytes,
+				     !bio->bi_error);
 
 	kfree(dip);
 
@@ -8595,10 +8579,10 @@ static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
 		io_bio = NULL;
 	} else {
 		if (write)
-			btrfs_endio_direct_write_update_ordered(inode,
+			__endio_write_update_ordered(inode,
 						file_offset,
 						dio_bio->bi_iter.bi_size,
-						0);
+						false);
 		else
 			unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
 			      file_offset + dio_bio->bi_iter.bi_size - 1);
@@ -8733,11 +8717,11 @@ static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 			 */
 			if (dio_data.unsubmitted_oe_range_start <
 			    dio_data.unsubmitted_oe_range_end)
-				btrfs_endio_direct_write_update_ordered(inode,
+				__endio_write_update_ordered(inode,
 					dio_data.unsubmitted_oe_range_start,
 					dio_data.unsubmitted_oe_range_end -
 					dio_data.unsubmitted_oe_range_start,
-					0);
+					false);
 		} else if (ret >= 0 && (size_t)ret < count)
 			btrfs_delalloc_release_space(inode, offset,
 						     count - (size_t)ret);