Btrfs: run delayed directory updates during log replay
authorChris Mason <chris.mason@fusionio.com>
Mon, 2 Jul 2012 19:29:53 +0000 (15:29 -0400)
committerChris Mason <chris.mason@fusionio.com>
Mon, 2 Jul 2012 19:39:19 +0000 (15:39 -0400)
While we are resolving directory modifications in the
tree log, we are triggering delayed metadata updates to
the filesystem btrees.

This commit forces the delayed updates to run so the
replay code can find any modifications done.  It stops
us from crashing because the directory deleltion replay
expects items to be removed immediately from the tree.

Signed-off-by: Chris Mason <chris.mason@fusionio.com>
cc: stable@kernel.org

fs/btrfs/tree-log.c

index 2017d0f..8abeae4 100644 (file)
@@ -690,6 +690,8 @@ static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
        kfree(name);
 
        iput(inode);
+
+       btrfs_run_delayed_items(trans, root);
        return ret;
 }
 
@@ -895,6 +897,7 @@ again:
                                ret = btrfs_unlink_inode(trans, root, dir,
                                                         inode, victim_name,
                                                         victim_name_len);
+                               btrfs_run_delayed_items(trans, root);
                        }
                        kfree(victim_name);
                        ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
@@ -1475,6 +1478,9 @@ again:
                        ret = btrfs_unlink_inode(trans, root, dir, inode,
                                                 name, name_len);
                        BUG_ON(ret);
+
+                       btrfs_run_delayed_items(trans, root);
+
                        kfree(name);
                        iput(inode);