From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from cn.fujitsu.com ([59.151.112.132]:22715 "EHLO heian.cn.fujitsu.com" rhost-flags-OK-FAIL-OK-FAIL) by vger.kernel.org with ESMTP id S1751182AbbETG1Y convert rfc822-to-8bit (ORCPT ); Wed, 20 May 2015 02:27:24 -0400 Message-ID: <555C2943.4070800@cn.fujitsu.com> Date: Wed, 20 May 2015 14:27:15 +0800 From: Qu Wenruo MIME-Version: 1.0 To: Filipe Manana , , Chris Mason CC: Subject: Re: [PATCH] Btrfs: avoid syncing log in the fast fsync path when not necessary References: <1427850043-27603-1-git-send-email-fdmanana@suse.com> In-Reply-To: <1427850043-27603-1-git-send-email-fdmanana@suse.com> Content-Type: text/plain; charset="utf-8"; format=flowed Sender: linux-btrfs-owner@vger.kernel.org List-ID: To Chris: Would you please consider including this patch to 4.1 merge window? As it fixes a performance regression in 4.0. Thanks, Qu -------- Original Message -------- Subject: [PATCH] Btrfs: avoid syncing log in the fast fsync path when not necessary From: Filipe Manana To: Date: 2015年04月01日 09:00 > Commit 3a8b36f37806 ("Btrfs: fix data loss in the fast fsync path") added > a performance regression for that causes an unnecessary sync of the log > trees (fs/subvol and root log trees) when 2 consecutive fsyncs are done > against a file, without no writes or any metadata updates to the inode in > between them and if a transaction is committed before the second fsync is > called. > > Huang Ying reported this to lkml (https://lkml.org/lkml/2015/3/18/99) > after a test sysbench test that measured a -62% decrease of file io > requests per second for that tests' workload. > > The test is: > > echo performance > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor > echo performance > /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor > echo performance > /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor > echo performance > /sys/devices/system/cpu/cpu3/cpufreq/scaling_governor > mkfs -t btrfs /dev/sda2 > mount -t btrfs /dev/sda2 /fs/sda2 > cd /fs/sda2 > for ((i = 0; i < 1024; i++)); do fallocate -l 67108864 testfile.$i; done > sysbench --test=fileio --max-requests=0 --num-threads=4 --max-time=600 \ > --file-test-mode=rndwr --file-total-size=68719476736 --file-io-mode=sync \ > --file-num=1024 run > > A test on kvm guest, running a debug kernel gave me the following results: > > Without 3a8b36f378060d: 16.01 reqs/sec > With 3a8b36f378060d: 3.39 reqs/sec > With 3a8b36f378060d and this patch: 16.04 reqs/sec > > Reported-by: Huang Ying > Tested-by: Huang, Ying > Signed-off-by: Filipe Manana > --- > fs/btrfs/file.c | 9 ++++++--- > fs/btrfs/ordered-data.c | 14 ++++++++++++++ > fs/btrfs/ordered-data.h | 3 +++ > 3 files changed, 23 insertions(+), 3 deletions(-) > > diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c > index 309dd57..379275c 100644 > --- a/fs/btrfs/file.c > +++ b/fs/btrfs/file.c > @@ -1878,6 +1878,7 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) > struct btrfs_log_ctx ctx; > int ret = 0; > bool full_sync = 0; > + const u64 len = end - start + 1; > > trace_btrfs_sync_file(file, datasync); > > @@ -1906,7 +1907,7 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) > * all extents are persisted and the respective file extent > * items are in the fs/subvol btree. > */ > - ret = btrfs_wait_ordered_range(inode, start, end - start + 1); > + ret = btrfs_wait_ordered_range(inode, start, len); > } else { > /* > * Start any new ordered operations before starting to log the > @@ -1978,8 +1979,10 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) > */ > smp_mb(); > if (btrfs_inode_in_log(inode, root->fs_info->generation) || > - (full_sync && BTRFS_I(inode)->last_trans <= > - root->fs_info->last_trans_committed)) { > + (BTRFS_I(inode)->last_trans <= > + root->fs_info->last_trans_committed && > + (full_sync || > + !btrfs_have_ordered_extents_in_range(inode, start, len)))) { > /* > * We'v had everything committed since the last time we were > * modified so clear this flag in case it was set for whatever > diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c > index 157cc54..72b6f0d 100644 > --- a/fs/btrfs/ordered-data.c > +++ b/fs/btrfs/ordered-data.c > @@ -838,6 +838,20 @@ out: > return entry; > } > > +bool btrfs_have_ordered_extents_in_range(struct inode *inode, > + u64 file_offset, > + u64 len) > +{ > + struct btrfs_ordered_extent *oe; > + > + oe = btrfs_lookup_ordered_range(inode, file_offset, len); > + if (oe) { > + btrfs_put_ordered_extent(oe); > + return true; > + } > + return false; > +} > + > /* > * lookup and return any extent before 'file_offset'. NULL is returned > * if none is found > diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h > index e96cd4c..9ba7209 100644 > --- a/fs/btrfs/ordered-data.h > +++ b/fs/btrfs/ordered-data.h > @@ -191,6 +191,9 @@ btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset); > struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode, > u64 file_offset, > u64 len); > +bool btrfs_have_ordered_extents_in_range(struct inode *inode, > + u64 file_offset, > + u64 len); > int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, > struct btrfs_ordered_extent *ordered); > int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, >