This is based on Josef's "Btrfs: turbo charge fsync".
We should cleanup those extents after we've finished logging inode,
otherwise we may do redundant work on them.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
                        err = ret;
                        goto out_unlock;
                }
+       } else {
+               struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
+               struct extent_map *em, *n;
+
+               list_for_each_entry_safe(em, n, &tree->modified_extents, list)
+                       list_del_init(&em->list);
        }
 
        if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {