Btrfs: be smarter about dropping things from the tree log
authorJosef Bacik <jbacik@fusionio.com>
Fri, 28 Sep 2012 15:56:28 +0000 (11:56 -0400)
committerChris Mason <chris.mason@fusionio.com>
Tue, 9 Oct 2012 13:15:41 +0000 (09:15 -0400)
When we truncate existing items in the tree log we've been searching for
each individual item and removing them.  This is unnecessary churn and
searching, just keep track of the slot we are on and how many items we need
to delete and delete them all at once.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
fs/btrfs/tree-log.c

index 31b46a9e94cccfd5a8017c32d859431779fe5504..179fda964601a3f8a026113e96e5733d7285c5f5 100644 (file)
@@ -2901,6 +2901,7 @@ static int drop_objectid_items(struct btrfs_trans_handle *trans,
        int ret;
        struct btrfs_key key;
        struct btrfs_key found_key;
+       int start_slot;
 
        key.objectid = objectid;
        key.type = max_key_type;
@@ -2922,8 +2923,18 @@ static int drop_objectid_items(struct btrfs_trans_handle *trans,
                if (found_key.objectid != objectid)
                        break;
 
-               ret = btrfs_del_item(trans, log, path);
-               if (ret)
+               found_key.offset = 0;
+               found_key.type = 0;
+               ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
+                                      &start_slot);
+
+               ret = btrfs_del_items(trans, log, path, start_slot,
+                                     path->slots[0] - start_slot + 1);
+               /*
+                * If start slot isn't 0 then we don't need to re-search, we've
+                * found the last guy with the objectid in this tree.
+                */
+               if (ret || start_slot != 0)
                        break;
                btrfs_release_path(path);
        }