Data Corruption bug with Samba's vfs_iouring and Linux 5.6.7/5.7rc3

[Jens Axboe]

On 5/7/20 12:31 PM, Jeremy Allison wrote:
> On Thu, May 07, 2020 at 10:50:40AM -0600, Jens Axboe wrote:
>> On 5/7/20 10:48 AM, Jeremy Allison wrote:
>>> On Thu, May 07, 2020 at 10:43:17AM -0600, Jens Axboe wrote:
>>>>
>>>> Just like for regular system calls, applications must be able to deal
>>>> with short IO.
>>>
>>> Thanks, that's a helpful definitive reply. Of course, the SMB3
>>> protocol is designed to deal with short IO replies as well, and
>>> the Samba and linux kernel clients are well-enough written that
>>> they do so. MacOS and Windows however..
>>
>> I'm honestly surprised that such broken clients exists! Even being
>> a somewhat old timer cynic...
>>
>>> Unfortunately they're the most popular clients on the planet,
>>> so we'll probably have to fix Samba to never return short IOs.
>>
>> That does sound like the best way forward, short IOs is possible
>> with regular system calls as well, but will definitely be a lot
>> more frequent with io_uring depending on the access patterns,
>> page cache, number of threads, and so on.
> 
> OK, I just want to be *REALLY CLEAR* what you're telling me
> (I've already written the pread/pwrite wrappers for Samba
> that deal with short IO but want to ensure I understand
> fully before making any changes to Samba).
> 
> You're saying that on a bog-standard ext4 disk file:
> 
> ret = pread(fd, buf, count, offset);
> 
> can return *less* than count bytes if there's no IO
> error and the file size is greater than offset+count
> and no one else is in the middle of a truncate etc. ?
> 
> And:
> 
> ret = pwrite(fd, buf, count, offset);
> 
> can return less* than count bytes if there's no IO
> error and there's ample space on disk ?
> 
> I have to say I've *never* seen that happen, and
> Samba is widely enough used that IO corruption from
> short reads/writes from MacOSX and Windows clients
> would have been widely reported by now.
> 
> Look at how quickly someone spotted disk corruption
> because of the change in userspace-visible behavior
> of the io_uring interface. We only shipped that code
> 03 March 2020 and someone *already* found it.

I _think_ that will only happen on regular files if you use RWF_NOWAIT
or similar, for regular blocking it should not happen. So I don't think
you're at risk there, though I do think that anyone should write
applications with short IOs in mind or they will run into surprises down
the line. Should have been more clear!

-- 
Jens Axboe