linux-kernel - Re: bio linked list corruption.

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]

Date:   Wed, 26 Oct 2016 19:38:08 -0400
From:   Chris Mason <clm@...com>
To:     Jens Axboe <axboe@...com>
CC:     Linus Torvalds <torvalds@...ux-foundation.org>,
        Dave Jones <davej@...emonkey.org.uk>,
        Andy Lutomirski <luto@...capital.net>,
        "Andy Lutomirski" <luto@...nel.org>,
        Al Viro <viro@...iv.linux.org.uk>, Josef Bacik <jbacik@...com>,
        David Sterba <dsterba@...e.com>,
        linux-btrfs <linux-btrfs@...r.kernel.org>,
        Linux Kernel <linux-kernel@...r.kernel.org>,
        Dave Chinner <david@...morbit.com>
Subject: Re: bio linked list corruption.

On Wed, Oct 26, 2016 at 05:20:01PM -0600, Jens Axboe wrote:
>On 10/26/2016 05:08 PM, Linus Torvalds wrote:
>>On Wed, Oct 26, 2016 at 4:03 PM, Jens Axboe <axboe@...com> wrote:
>>>
>>>Actually, I think I see what might trigger it. You are on nvme, iirc,
>>>and that has a deep queue.
>>
>>Yes. I have long since moved on from slow disks, so all my systems are
>>not just flash, but m.2 nvme ssd's.
>>
>>So at least that could explain why Dave sees it at bootup but I don't.
>
>Yep, you'd never sleep during normal boot. The original patch had a
>problem, though... This one should be better. Too many 'data's, we'll
>still need to assign ctx/hctx, we should just use the current ones, not
>the original ones.
>
>diff --git a/block/blk-mq.c b/block/blk-mq.c
>index ddc2eed64771..e56fec187ba6 100644
>--- a/block/blk-mq.c
>+++ b/block/blk-mq.c
>@@ -1217,9 +1224,9 @@ static struct request *blk_mq_map_request(struct 
>request_queue *q,
> 	blk_mq_set_alloc_data(&alloc_data, q, 0, ctx, hctx);
> 	rq = __blk_mq_alloc_request(&alloc_data, op, op_flags);
>
>-	hctx->queued++;
>-	data->hctx = hctx;
>-	data->ctx = ctx;
>+	data->hctx = alloc_data.hctx;
>+	data->ctx = alloc_data.ctx;
>+	data->hctx->queued++;
> 	return rq;
> }

This made it through an entire dbench 2048 run on btrfs.  My script has 
it running in a loop, but this is farther than I've gotten before.  
Looking great so far.

-chris