From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from userp1040.oracle.com ([156.151.31.81]:39897 "EHLO userp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753136AbcFPApw (ORCPT ); Wed, 15 Jun 2016 20:45:52 -0400 Date: Wed, 15 Jun 2016 17:47:20 -0700 From: Liu Bo To: Chris Mason , Josef Bacik , linux-btrfs@vger.kernel.org Cc: David Sterba Subject: Re: [PATCH] Btrfs: track transid for delayed ref flushing Message-ID: <20160616004720.GD8071@localhost.localdomain> Reply-To: bo.li.liu@oracle.com References: <1460410660-17430-1-git-send-email-jbacik@fb.com> <20160427135938.pczv7f4ulrvwehqp@floor.thefacebook.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <20160427135938.pczv7f4ulrvwehqp@floor.thefacebook.com> Sender: linux-btrfs-owner@vger.kernel.org List-ID: On Wed, Apr 27, 2016 at 09:59:38AM -0400, Chris Mason wrote: > On Mon, Apr 11, 2016 at 05:37:40PM -0400, Josef Bacik wrote: > > Using the offwakecputime bpf script I noticed most of our time was spent waiting > > on the delayed ref throttling. This is what is supposed to happen, but > > sometimes the transaction can commit and then we're waiting for throttling that > > doesn't matter anymore. So change this stuff to be a little smarter by tracking > > the transid we were in when we initiated the throttling. If the transaction we > > get is different then we can just bail out. This resulted in a 50% speedup in > > my fs_mark test, and reduced the amount of time spent throttling by 60 seconds > > over the entire run (which is about 30 minutes). Thanks, > > > > Signed-off-by: Josef Bacik > > --- > > fs/btrfs/ctree.h | 2 +- > > fs/btrfs/extent-tree.c | 15 ++++++++++++--- > > fs/btrfs/inode.c | 1 + > > fs/btrfs/transaction.c | 3 ++- > > 4 files changed, 16 insertions(+), 5 deletions(-) > > > > diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h > > index 55a24c5..4222936 100644 > > --- a/fs/btrfs/ctree.h > > +++ b/fs/btrfs/ctree.h > > @@ -3505,7 +3505,7 @@ void btrfs_put_block_group(struct btrfs_block_group_cache *cache); > > int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, > > struct btrfs_root *root, unsigned long count); > > int btrfs_async_run_delayed_refs(struct btrfs_root *root, > > - unsigned long count, int wait); > > + unsigned long count, u64 transid, int wait); > > int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len); > > int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, > > struct btrfs_root *root, u64 bytenr, > > diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c > > index 4b5a517..f23f426 100644 > > --- a/fs/btrfs/extent-tree.c > > +++ b/fs/btrfs/extent-tree.c > > @@ -2839,6 +2839,7 @@ int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, > > > > struct async_delayed_refs { > > struct btrfs_root *root; > > + u64 transid; > > int count; > > int error; > > int sync; > > @@ -2854,9 +2855,16 @@ static void delayed_ref_async_start(struct btrfs_work *work) > > > > async = container_of(work, struct async_delayed_refs, work); > > > > - trans = btrfs_join_transaction(async->root); > > + trans = btrfs_attach_transaction(async->root); > > if (IS_ERR(trans)) { > > - async->error = PTR_ERR(trans); > > + if (PTR_ERR(trans) != -ENOENT) > > + async->error = PTR_ERR(trans); > > + goto done; > > + } > > This ends up deadlocking because btrfs_attach_transaction waits in ways > that join does not. The differences between these two are really > subtle, and we manage to make this mistake every year or so. > > Subject: [PATCH] btrfs: fix deadlock in delayed_ref_async_start > > "Btrfs: track transid for delayed ref flushing" was deadlocking on > btrfs_attach_transaction because its not safe to call from the async > delayed ref start code. This commit brings back btrfs_join_transaction > instead and checks for a blocked commit. Are we going to take this patch? Thanks, -liubo > > Signed-off-by: Josef Bacik > Signed-off-by: Chris Mason > --- > fs/btrfs/extent-tree.c | 20 +++++++++++--------- > 1 file changed, 11 insertions(+), 9 deletions(-) > > diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c > index 6ce5b6c..44da4ac 100644 > --- a/fs/btrfs/extent-tree.c > +++ b/fs/btrfs/extent-tree.c > @@ -2845,16 +2845,13 @@ static void delayed_ref_async_start(struct btrfs_work *work) > > async = container_of(work, struct async_delayed_refs, work); > > - trans = btrfs_attach_transaction(async->root); > - if (IS_ERR(trans)) { > - if (PTR_ERR(trans) != -ENOENT) > - async->error = PTR_ERR(trans); > + /* if the commit is already started, we don't need to wait here */ > + if (btrfs_transaction_blocked(async->root->fs_info)) > goto done; > - } > > - /* Don't bother flushing if we got into a different transaction */ > - if (trans->transid != async->transid) { > - btrfs_end_transaction(trans, async->root); > + trans = btrfs_join_transaction(async->root); > + if (IS_ERR(trans)) { > + async->error = PTR_ERR(trans); > goto done; > } > > @@ -2863,10 +2860,15 @@ static void delayed_ref_async_start(struct btrfs_work *work) > * wait on delayed refs > */ > trans->sync = true; > + > + /* Don't bother flushing if we got into a different transaction */ > + if (trans->transid > async->transid) > + goto end; > + > ret = btrfs_run_delayed_refs(trans, async->root, async->count); > if (ret) > async->error = ret; > - > +end: > ret = btrfs_end_transaction(trans, async->root); > if (ret && !async->error) > async->error = ret; > -- > 2.8.0.rc2 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html