Re: [PATCHSET v2 0/2] io_uring: handle short reads internally

[Jens Axboe <axboe@kernel.dk>]

On 8/19/20 1:31 AM, Stefan Metzmacher wrote:
> Hi Anoop,
> 
>> @metze,
>>
>> Looks like issue is fixed with patched version of v5.7.15. For the
>> record I am noting down the steps followed:
>>
>> Following is my configuration:
>> OS: Fedora 32
>> Kernel: 5.7.15 [5.7.15-200]
>>         5.7.15+patches from 5.7-stable branch [5.7.15-202]
>> Samba: 4.12.6 [4.12.6-0]
>>        4.12.6+git reverts for vfs_io_uring [4.12.6-1]
>> liburing: 0.7
>> Client: Win 10, build 1909
>> Issue: SHA256 mismatch on files copied from io_uring enabled Samba
>>        share using Windows explorer
>>
>> * Issue not seen with unpatched 5.7.15 and samba-4.12.6
>>
>> * prepared samba-4.12.6 build[1] with reverts(see attachment) for
>> vfs_io_uring(as per https://bugzilla.samba.org/show_bug.cgi?id=14361)
>> and reproduced issue with unpatched 5.7.15.
>>
>> * prepared kernel-5.7.15 build[2] with extra commits(see attachment) as
>> per 5.7-stable branch from git://git.kernel.dk/linux-block. SHA256
>> mismatch is no longer seen.
>>
>> [1] https://koji.fedoraproject.org/koji/taskinfo?taskID=49539069
>> [2] https://koji.fedoraproject.org/koji/taskinfo?taskID=49585679
> 
> Great, thanks!

Indeed, thanks a lot for retesting! It helps validate that the backport
is sane in general, both from a "now handles short reads and retries them
without issue" and the "didn't mess anything else up majorly" side
of things :-)

-- 
Jens Axboe