Re: [PATCH] bio: decrease bi_iter.bi_size by len in the fail path

[Jet Chen <jet.chen@intel.com>]

[-- Attachment #1: Type: text/plain, Size: 1996 bytes --]

On 05/29/2014 01:44 AM, Ming Lei wrote:
> On Thu, May 29, 2014 at 1:21 AM, Maurizio Lombardi <mlombard@redhat.com> wrote:
>> Hi Ming,
>>
>> On Thu, May 29, 2014 at 12:59:19AM +0800, Ming Lei wrote:
>>>
>>> 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@localhost 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
>>
>> There is a little mistake in your patch, you removed bio->bi_iter.bi_size += len;
>> after the "done" label,
>> but be careful that at line 747 there is a "goto done"... bi_size should be incremented
>> before jumping there.
>
> Good catch, thanks Maurizio.
>
> Jet, please test the attached patch in this mail and ignore previous
> one.
>
> The story behind the patch should be like below:
>
> - one page is added in __bio_add_page() 'successfully',
> and bio->bi_phys_segments is equal to queue_max_segments(q),
> but it should have been rejected since the last vector isn't covered
>
> - next time, __bio_add_page() is called to add one page, but this
> time blk_recount_segments() can figure out the actual physical
> segments and find it is more than max segments, so failure is
> triggered, but the bio->bi_phys_segments is updated with
> max segments plus one
>
> - the oops is triggered and reported by Jet, :-)
>
>
> Thanks,
>
This patch works, thanks.

Tested-by: Jet Chen <jet.chen@intel.com>


[-- Attachment #2: fix_compute_segments.patch --]
[-- Type: text/x-patch, Size: 997 bytes --]

diff --git a/block/bio.c b/block/bio.c
index 0443694..f9bae56 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -744,6 +744,7 @@ static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
 				}
 			}
 
+			bio->bi_iter.bi_size += len;
 			goto done;
 		}
 	}
@@ -761,6 +762,7 @@ static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
 	bvec->bv_offset = offset;
 	bio->bi_vcnt++;
 	bio->bi_phys_segments++;
+	bio->bi_iter.bi_size += len;
 
 	/*
 	 * Perform a recount if the number of segments is greater
@@ -802,7 +804,6 @@ static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
 		bio->bi_flags &= ~(1 << BIO_SEG_VALID);
 
  done:
-	bio->bi_iter.bi_size += len;
 	return len;
 
  failed:
@@ -810,6 +811,7 @@ 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;
 	blk_recount_segments(q, bio);
 	return 0;
 }