On 3/19/20 4:41 AM, Pavel Begunkov wrote:
> On 3/19/2020 6:51 AM, Jens Axboe wrote:
>> A previous patch ensured that we always prepped requests that are
>> forced async, but it did so too late in the process. This can result
>> in 'sqe' already being NULL by the time we get to it:
>
> Isn't it fixed by f1d96a8fcbbbb ("io_uring: NULL-deref for
> IOSQE_{ASYNC,DRAIN}")? BTW, the same can happen with draining in
> io_req_defer() -> io_req_defer_prep().
>
> Can't look through your patches/RFC properly, but I will try do that
> this weekends.
Ah I think that it is, I was running the 5.7 branch and that one
doesn't have that patch. Disregard this one, I think we're fine.
--
Jens Axboe
