From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from rcsinet15.oracle.com ([148.87.113.117]:33465 "EHLO rcsinet15.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S934833Ab2JXOSr (ORCPT ); Wed, 24 Oct 2012 10:18:47 -0400 Message-ID: <5087F8C8.40605@oracle.com> Date: Wed, 24 Oct 2012 22:18:48 +0800 From: Liu Bo MIME-Version: 1.0 To: Jan Schmidt CC: chris.mason@fusionio.com, linux-btrfs@vger.kernel.org Subject: Re: [PATCH] Btrfs: drop locks before calling read_tree_block from get_old_root References: <5087BB1D.4020501@oracle.com> <1351075516-16344-1-git-send-email-list.btrfs@jan-o-sch.net> In-Reply-To: <1351075516-16344-1-git-send-email-list.btrfs@jan-o-sch.net> Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-btrfs-owner@vger.kernel.org List-ID: On 10/24/2012 06:45 PM, Jan Schmidt wrote: > get_old_root needs its lock on the root node only to clone that buffer. > However, when we call read_tree_block, we know already that we're not > going to clone the root node. Thus we can safely unlock before, to avoid > issues where read_tree_block makes potentially sleeping allocations. > I tested this for 20 times, it's no more blocked. So you can add Reviewed-by: Liu Bo Tested-by: Liu Bo thanks, liubo > Signed-off-by: Jan Schmidt > --- > Thanks for pointing that out, liubo! > > I added this additional patch on top of my patch staple at: > > git://git.jan-o-sch.net/btrfs-unstable for-chris > > I also made a second branch, because this last commit actually fixes > that's introduced by one of the patches in the patch set. So I folded > it into the respective commit: > > git://git.jan-o-sch.net/btrfs-unstable for-chris-fixed > > I made sure that the resulting code is the same. > > Chris: please pull either one for rc3. > > -Jan > > --- > fs/btrfs/ctree.c | 9 +++++++-- > 1 files changed, 7 insertions(+), 2 deletions(-) > > diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c > index c608b3c..eba44b0 100644 > --- a/fs/btrfs/ctree.c > +++ b/fs/btrfs/ctree.c > @@ -1261,6 +1261,8 @@ get_old_root(struct btrfs_root *root, u64 time_seq) > > tm = tree_mod_log_search(root->fs_info, logical, time_seq); > if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { > + btrfs_tree_read_unlock(root->node); > + free_extent_buffer(root->node); > blocksize = btrfs_level_size(root, old_root->level); > eb = read_tree_block(root, logical, blocksize, 0); > if (!eb) { > @@ -1271,12 +1273,15 @@ get_old_root(struct btrfs_root *root, u64 time_seq) > eb = btrfs_clone_extent_buffer(eb); > } > } else if (old_root) { > + btrfs_tree_read_unlock(root->node); > + free_extent_buffer(root->node); > eb = alloc_dummy_extent_buffer(logical, root->nodesize); > } else { > eb = btrfs_clone_extent_buffer(root->node); > + btrfs_tree_read_unlock(root->node); > + free_extent_buffer(root->node); > } > - btrfs_tree_read_unlock(root->node); > - free_extent_buffer(root->node); > + > if (!eb) > return NULL; > extent_buffer_get(eb); >