Btrfs: fix panic in balance due to EIO

During build_backref_tree(), if we fail to read a btree node,
we can eventually run into BUG_ON(cache->nr_nodes) that we put
in backref_cache_cleanup(), meaning we have at least one
memory leak.

This frees the backref_node that we's allocated at the very
beginning of build_backref_tree().
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>

fs/btrfs/relocation.c

@@ -1171,8 +1171,12 @@ struct backref_node *build_backref_tree(struct reloc_control *rc,
 			lower = list_entry(useless.next,
 					   struct backref_node, list);
 			list_del_init(&lower->list);
+			if (lower == node)
+				node = NULL;
 			free_backref_node(cache, lower);
 		}
+
+		free_backref_node(cache, node);
 		return ERR_PTR(err);
 	}
 	ASSERT(!node || !node->detached);