Re: possible deadlock in __io_queue_deferred

[Jens Axboe]

On 8/11/20 8:44 AM, Stefano Garzarella wrote:
> On Tue, Aug 11, 2020 at 08:21:12AM -0600, Jens Axboe wrote:
>> On 8/11/20 8:00 AM, Stefano Garzarella wrote:
>>> On Mon, Aug 10, 2020 at 09:55:17AM -0600, Jens Axboe wrote:
>>>> On 8/10/20 9:36 AM, syzbot wrote:
>>>>> Hello,
>>>>>
>>>>> syzbot found the following issue on:
>>>>>
>>>>> HEAD commit:    449dc8c9 Merge tag 'for-v5.9' of git://git.kernel.org/pub/..
>>>>> git tree:       upstream
>>>>> console output: https://syzkaller.appspot.com/x/log.txt?x=14d41e02900000
>>>>> kernel config:  https://syzkaller.appspot.com/x/.config?x=9d25235bf0162fbc
>>>>> dashboard link: https://syzkaller.appspot.com/bug?extid=996f91b6ec3812c48042
>>>>> compiler:       clang version 10.0.0 (https://github.com/llvm/llvm-project/ c2443155a0fb245c8f17f2c1c72b6ea391e86e81)
>>>>> syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=133c9006900000
>>>>> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=1191cb1a900000
>>>>>
>>>>> IMPORTANT: if you fix the issue, please add the following tag to the commit:
>>>>> Reported-by: syzbot+996f91b6ec3812c48042@xxxxxxxxxxxxxxxxxxxxxxxxx
>>>>
>>>> Thanks, the below should fix this one.
>>>
>>> Yeah, it seems right to me, since only __io_queue_deferred() (invoked by
>>> io_commit_cqring()) can be called with 'completion_lock' held.
>>
>> Right
>>
>>> Just out of curiosity, while exploring the code I noticed that we call
>>> io_commit_cqring() always with the 'completion_lock' held, except in the
>>> io_poll_* functions.
>>>
>>> That's because then there can't be any concurrency?
>>
>> Do you mean the iopoll functions? Because we're definitely holding it
>> for the io_poll_* functions.
> 
> Right, the only one seems io_iopoll_complete().
> 
> So, IIUC, in this case we are actively polling the level below,
> so there shouldn't be any asynchronous events, is it right?

Right, that's serialized by itself.

-- 
Jens Axboe