From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from out30-98.freemail.mail.aliyun.com (out30-98.freemail.mail.aliyun.com [115.124.30.98]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 742A02DC787 for ; Fri, 27 Mar 2026 03:12:19 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=115.124.30.98 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1774581142; cv=none; b=FzuZQFkFJUl+YqRZehlZh2VIJcNpzx9zycPq/2RFjcMnrgnP/X0Ly1Nl8hEvPHte0x/Q/+UOBbb8wqmjMD22MOV5Be+woEkg8im4Q8yYdoR23Ux8qIy+OQUmVsxMczvwOqihBttBnf9xT0Ip3KPNTv3TEThYV2v3aIdt1DyosJg= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1774581142; c=relaxed/simple; bh=fY1H1mhVbiv3v+6X5YM0di8lhmYkTEc7JIWCXTOyQBw=; h=Message-ID:Date:MIME-Version:Subject:To:Cc:References:From: In-Reply-To:Content-Type; b=MbbnZQQlUsArcdOAoPOuk2SkdHGoTsnVW2nidQBcFB9PwKLS0Ekb8BZX5zibSRFMNPfHbTuHOSdVxK7u352/ByIZa6R0/+OZMbxKmRUFiMVsIvjsqJjnKo4Re0T95l0YWTjAaSTn/J8QtapJsneZ2SUpbb9LJH4SxogBHIpMGns= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.alibaba.com; spf=pass smtp.mailfrom=linux.alibaba.com; dkim=pass (1024-bit key) header.d=linux.alibaba.com header.i=@linux.alibaba.com header.b=CtvSHx6w; arc=none smtp.client-ip=115.124.30.98 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.alibaba.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=linux.alibaba.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (1024-bit key) header.d=linux.alibaba.com header.i=@linux.alibaba.com header.b="CtvSHx6w" DKIM-Signature:v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.alibaba.com; s=default; t=1774581136; h=Message-ID:Date:MIME-Version:Subject:To:From:Content-Type; bh=Z3ER4uZfQFN8UsxUbmx5OlL1opWvl7soPiJH7TZQ/J8=; b=CtvSHx6w4cDyq9a63ljA/gzTNhDK8HoxdC6T07mamov56JBRqOD2EFFNV5MMNP+OCS8UvJLuJBvcqtQPXWAv2jU+wcef4+Hd+2upAkxqT7lB7b1grnQX0YdXQjKmfTrXANeUy6pYt8BG0pytNd+SZBSnI78CuEAkkZSZ42Ur0xs= X-Alimail-AntiSpam:AC=PASS;BC=-1|-1;BR=01201311R141e4;CH=green;DM=||false|;DS=||;FP=0|-1|-1|-1|0|-1|-1|-1;HT=maildocker-contentspam033032089153;MF=joseph.qi@linux.alibaba.com;NM=1;PH=DS;RN=5;SR=0;TI=SMTPD_---0X.mlyR0_1774581135; Received: from 30.221.145.61(mailfrom:joseph.qi@linux.alibaba.com fp:SMTPD_---0X.mlyR0_1774581135 cluster:ay36) by smtp.aliyun-inc.com; Fri, 27 Mar 2026 11:12:16 +0800 Message-ID: Date: Fri, 27 Mar 2026 11:12:15 +0800 Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Subject: Re: [PATCH v4 1/1] ocfs2: split transactions in dio completion to avoid credit exhaustion To: Heming Zhao Cc: ocfs2-devel@lists.linux.dev, linux-kernel@vger.kernel.org, glass.su@suse.com, Jan Kara References: <20260326142640.20077-1-heming.zhao@suse.com> <20260326142640.20077-2-heming.zhao@suse.com> From: Joseph Qi In-Reply-To: Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit On 3/27/26 11:02 AM, Heming Zhao wrote: > On Fri, Mar 27, 2026 at 09:42:55AM +0800, Joseph Qi wrote: >> Hi, >> >> On 3/26/26 10:26 PM, Heming Zhao wrote: >>> During ocfs2 dio operations, JBD2 may report warnings via following call trace: >>> ocfs2_dio_end_io_write >>> ocfs2_mark_extent_written >>> ocfs2_change_extent_flag >>> ocfs2_split_extent >>> ocfs2_try_to_merge_extent >>> ocfs2_extend_rotate_transaction >>> ocfs2_extend_trans >>> jbd2__journal_restart >>> start_this_handle >>> output: JBD2: kworker/6:2 wants too many credits credits:5450 rsv_credits:0 max:5449 >>> >>> To prevent exceeding the credits limit, modify ocfs2_dio_end_io_write() >>> to handle extent in a batch of transaction. >>> >>> Additionally, relocate ocfs2_del_inode_from_orphan(). The orphan inode should >>> only be removed from the orphan list after the extent tree update is complete. >>> this ensures that if a crash occurs in the middle of extent tree updates, we >>> won't leave stale blocks beyond EOF. >>> >>> Finally, thanks to Jans and Joseph for providing the bug fix prototype and >>> suggestions. >>> >>> Suggested-by: Jan Kara >>> Suggested-by: Joseph Qi >>> Reviewed-by: Jan Kara >>> Signed-off-by: Heming Zhao >>> --- >>> fs/ocfs2/aops.c | 72 ++++++++++++++++++++++++++++++------------------- >>> 1 file changed, 44 insertions(+), 28 deletions(-) >>> >>> diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c >>> index 09146b43d1f0..60f1b607022f 100644 >>> --- a/fs/ocfs2/aops.c >>> +++ b/fs/ocfs2/aops.c >>> @@ -37,6 +37,8 @@ >>> #include "namei.h" >>> ... ... <--- snip ---> >>> + if (++batch == OCFS2_DIO_MARK_EXTENT_BATCH) { >>> + ocfs2_commit_trans(osb, handle); >>> + handle = NULL; >>> + batch = 0; >>> + } >>> } >>> >>> if (end > i_size_read(inode)) { >> >> I still don't think it is a good idea to update inode size in case error. >> The original logic behaves inconsistent, if ocfs2_start_trans() and >> ocfs2_journal_access_di() fails, it won't update inode size, but if >> ocfs2_assure_trans_credits() and ocfs2_mark_extent_written(), it will do. >> So let's make it behave consistent by both checking 'ret' here. >> >> Other looks fine. >> >> Joseph > > After sending the v4 patch, I did some tests and identified the key of the > performance degradation in the dynamic credit adjustment patch. > I found that by using half of the jbd2_max value, the time consumption returned > to the same level as the "direct loop" style. > > key code change: > (base on patch[1]: 0001-ocfs2-dynamic-extending-jbd2-credits-during-dio-end-.patch) > int jbd2_max = (journal->j_max_transaction_buffers - > journal->j_transaction_overhead_buffers) >> 2; > > my test env: > - 10G ocfs2 volume > - DIO 2G file, write with 64MB block (--bs=64M). > > observations: > - ocfs2_assure_trans_credits calls "commit & start trans" times on every 64MB block: > - using jbd2_max (5449): ~9 times. > - using half value (2724): ~20 times. > - v4 patch ("batch" style), call "commit & start trans" on every 64MB block: ~62 times > > time consumption: > > jbd2_max:5449 case > real 2m9.346s > user 0m0.277s > sys 0m22.748s > > half jbd2_max:2724 case > real 1m49.400s > user 0m0.343s > sys 0m22.734s > > v4 patch: > real 1m49.650s > user 0m0.337s > sys 0m22.780s > > > the reason (I guess): > The original patch only performed a "commit and start new trans" when jbd2 > credits were nearly exhausted. I suspect contention between the commit flush > operation and the jbd2 operations in the dio end path, leading to high latency. > By using half the jbd2_max value, trigger commits more frequently, which appears > to reduce jbd2 contention and racing. > > Finally, I suggest we use the dynamic credit extension patch for v5. This > version only adds logic to ocfs2_assure_trans_credits() and does not affect > i_size changes during error cases. > Ummm... I think v4 looks simpler and direct. For i_size update logic, I think the existing logic looks wried. Take a look at ext4, it also doesn't do this if converting unwritten extents fails. Joseph