linux-kernel - Re: [PATCH] bio: decrease bi_iter.bi

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:	Thu, 29 May 2014 00:59:19 +0800
From:	Ming Lei <tom.leiming@...il.com>
To:	Dongsu Park <dongsu.park@...fitbricks.com>
Cc:	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Jens Axboe <axboe@...nel.dk>, Jet Chen <jet.chen@...el.com>,
	Maurizio Lombardi <mlombard@...hat.com>
Subject: Re: [PATCH] bio: decrease bi_iter.bi_size by len in the fail path

On Wed, May 28, 2014 at 11:42 PM, Ming Lei <tom.leiming@...il.com> wrote:
> Hi Dongsu,
>
> On Wed, May 28, 2014 at 11:09 PM, Dongsu Park
> <dongsu.park@...fitbricks.com> wrote:
>> From: Dongsu Park <dongsu.park@...fitbricks.com>
>>
>> Commit 3979ef4dcf3d1de55a560a3a4016c30a835df44d ("bio-modify-
>> __bio_add_page-to-accept-pages-that-dont-start-a-new-segment-v3")
>> introduced a regression as reported by Jet Chen.
>> That results in a kernel BUG at drivers/block/virtio_blk.c:166.
>>
>> To fix that, bi_iter.bi_size must be decreased by len, before
>> recounting the number of physical segments.
>>
>> Tested on with kernel 3.15.0-rc7-next-20140527 on qemu guest,
>> by running xfstests/ext4/271.
>>
>> Cc: Jens Axboe <axboe@...nel.dk>
>> Cc: Jet Chen <jet.chen@...el.com>
>> Cc: Maurizio Lombardi <mlombard@...hat.com>
>> Signed-off-by: Dongsu Park <dongsu.park@...fitbricks.com>
>> ---
>>  block/bio.c | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/block/bio.c b/block/bio.c
>> index 0443694ccbb4..67d7cba1e5fd 100644
>> --- a/block/bio.c
>> +++ b/block/bio.c
>> @@ -810,6 +810,7 @@ static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
>>         bvec->bv_len = 0;
>>         bvec->bv_offset = 0;
>>         bio->bi_vcnt--;
>> +       bio->bi_iter.bi_size -= len;
>
> Would you mind explaining why bi_iter.bi_size need to be
> decreased by 'len'? In the failure path, it wasn't added by
> 'len', was it?

Actually, the correct thing may be like what did in the
attached patch, as Maurizio discussed with me[1].

Very interestingly, I have reproduced the problem one time
with ext4/271 ext4/301 ext4/305, but won't with the attached
patch after running it for 3 rounds.

[tom@...alhost xfstests]$ sudo ./check ext4/271 ext4/301 ext4/305
FSTYP         -- ext4
PLATFORM      -- Linux/x86_64 localhost 3.15.0-rc7-next-20140527+
MKFS_OPTIONS  -- /dev/vdc
MOUNT_OPTIONS -- -o acl,user_xattr /dev/vdc /mnt/scratch

ext4/271 1s ... 1s
ext4/301 31s ... 32s
ext4/305 181s ... 180s
Ran: ext4/271 ext4/301 ext4/305
Passed all 3 tests

Jet, could you test the attached patch?

[1], https://lkml.org/lkml/2014/5/27/327


Thanks,
-- 
Ming Lei

Download attachment "lll" of type "application/octet-stream" (825 bytes)