linux-kernel - Re: [PATCH] block: fix disordered IO in the case recursive split

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <c145302a-175e-da38-2d28-f92dd285b819@huaweicloud.com>
Date: Thu, 21 Aug 2025 17:42:08 +0800
From: Yu Kuai <yukuai1@...weicloud.com>
To: Hannes Reinecke <hare@...e.de>, Christoph Hellwig <hch@...radead.org>,
 Yu Kuai <yukuai1@...weicloud.com>
Cc: axboe@...nel.dk, neil@...wn.name, akpm@...ux-foundation.org,
 linux-block@...r.kernel.org, linux-raid@...r.kernel.org,
 linux-kernel@...r.kernel.org, colyli@...nel.org, xni@...hat.com,
 yi.zhang@...wei.com, yangerkun@...wei.com, johnny.chenyi@...wei.com,
 "yukuai (C)" <yukuai3@...wei.com>
Subject: Re: [PATCH] block: fix disordered IO in the case recursive split

Hi,

在 2025/08/21 17:33, Hannes Reinecke 写道:
> On 8/21/25 11:02, Christoph Hellwig wrote:
>> On Thu, Aug 21, 2025 at 04:56:33PM +0800, Yu Kuai wrote:
>>> Can you give some examples as how to chain the right way?
>>
>> fs/xfs/xfs_bio_io.c: xfs_rw_bdev
>> fs/xfs/xfs_buf.c: xfs_buf_submit_bio
>> fs/xfs/xfs_log.c: xlog_write_iclog
>>
>>> BTW, for all
>>> the io split case, should this order be fixed? I feel we should, this
>>> disorder can happen on any stack case, where top max_sector is greater
>>> than stacked disk.
>>
>> Yes, I've been trying get Bart to fix this for a while instead of
>> putting in a workaround very similar to the one proposed here,
>> but so far nothing happened.
>>
>>
> This feels like a really stupid fix, but wouldn't that help?
> 
> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
> index 023649fe2476..2b342bb59612 100644
> --- a/drivers/md/raid5.c
> +++ b/drivers/md/raid5.c
> @@ -5478,7 +5478,6 @@ static struct bio *chunk_aligned_read(struct mddev 
> *mddev, struct bio *raid_bio)
>                  split = bio_split(raid_bio, sectors, GFP_NOIO, 
> &conf->bio_split);
>                  bio_chain(split, raid_bio);
>                  submit_bio_noacct(raid_bio);
> -               raid_bio = split;
>          }
> 

I do not understand how can this help, do you miss that submit split
instead?

And with this change, this can help, however, I think we'll still submit
the last lba bio first, like bio_last -> bio0 -> bio1 ... where the
following is sequential.

BTW, this is not just a raid5 problem, this is also possible for
raid0/10 and all other recursive split case.

Thanks,
Kuai

> 
> Cheers,
> 
> Hannes