Btrfs: fix memory leak of block group cache

While processing delayed refs, we may update block group's statistics
and attach it to cur_trans->dirty_bgs, and later writing dirty block
groups will process the list, which happens during
btrfs_commit_transaction().

For whatever reason, the transaction is aborted and dirty_bgs
is not processed in cleanup_transaction(), we end up with memory leak
of these dirty block group cache.

Since btrfs_start_dirty_block_groups() doesn't make it go to the commit
critical section, this also adds the cleanup work inside it.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 665da8f..616c455 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -3694,6 +3694,8 @@
 			goto again;
 		}
 		spin_unlock(&cur_trans->dirty_bgs_lock);
+	} else if (ret < 0) {
+		btrfs_cleanup_dirty_bgs(cur_trans, root);
 	}
 
 	btrfs_free_path(path);