From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-qk0-f179.google.com ([209.85.220.179]:37075 "EHLO mail-qk0-f179.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750957AbdIKVMk (ORCPT ); Mon, 11 Sep 2017 17:12:40 -0400 Received: by mail-qk0-f179.google.com with SMTP id b82so21392900qkc.4 for ; Mon, 11 Sep 2017 14:12:40 -0700 (PDT) From: josef@toxicpanda.com To: kernel-team@fb.com, linux-btrfs@vger.kernel.org Subject: [RFC][0/10] Cleanup and enhance delayed refs Date: Mon, 11 Sep 2017 17:12:26 -0400 Message-Id: <1505164356-13474-1-git-send-email-jbacik@fb.com> Sender: linux-btrfs-owner@vger.kernel.org List-ID: I'm leaving for Linux Plumbers tomorrow so I wanted to get this out for people to look at and see if there were any general comments. These patches have only been compiled, not tested in any way, so I wouldn't recommend trying them. They do a few things 1) Cleanup empty delayed ref head handling. We clean this up in a variety of different ways in different places, so I've unified it all in one helper and reduced the complexity of __btrfs_run_delayed_refs. 2) Removed the btrfs_delayed_ref_node from the btrfs_delayed_ref_head. This used to be this way because we had heads and nodes all on the same list. Since this is no longer the case it's just wasted space and weird complexity, so kill it. 3) Track delayed ref updates in an rb_tree on the head ref instead of in a list. Our merging stuff really sucks if we end up with lots of different changes to the same block. You can end up blowing loads of cpu time during the transaction commit, which isn't helpful. With the tree we can not loop as much, and make our insert time merging much better. Here is the diffstat fs/btrfs/backref.c | 9 +- fs/btrfs/delayed-ref.c | 294 ++++++++++++++++----------------- fs/btrfs/delayed-ref.h | 54 +++---- fs/btrfs/disk-io.c | 22 +-- fs/btrfs/extent-tree.c | 376 ++++++++++++++++++++----------------------- include/trace/events/btrfs.h | 15 +- 6 files changed, 363 insertions(+), 407 deletions(-) Let me know if you have any questions or feedback. Thanks, Josef