On 2/17/25 7:12 AM, Pavel Begunkov wrote:
> On 2/17/25 13:58, Jens Axboe wrote:
>> On 2/17/25 6:37 AM, Pavel Begunkov wrote:
>>> At the moment we can't sanely handle queuing an async request from a
>>> multishot context, so disable them. It shouldn't matter as pollable
>>> files / socekts don't normally do async.
>>
>> Having something pollable that can return -EIOCBQUEUED is odd, but
>> that's just a side comment.
>>
>>
>>> diff --git a/io_uring/rw.c b/io_uring/rw.c
>>> index 96b42c331267..4bda46c5eb20 100644
>>> --- a/io_uring/rw.c
>>> +++ b/io_uring/rw.c
>>> @@ -878,7 +878,15 @@ static int __io_read(struct io_kiocb *req, unsigned int issue_flags)
>>> if (unlikely(ret))
>>> return ret;
>>> - ret = io_iter_do_read(rw, &io->iter);
>>> + if (unlikely(req->opcode == IORING_OP_READ_MULTISHOT)) {
>>> + void *cb_copy = rw->kiocb.ki_complete;
>>> +
>>> + rw->kiocb.ki_complete = NULL;
>>> + ret = io_iter_do_read(rw, &io->iter);
>>> + rw->kiocb.ki_complete = cb_copy;
>>> + } else {
>>> + ret = io_iter_do_read(rw, &io->iter);
>>> + }
>>
>> This looks a bit odd. Why can't io_read_mshot() just clear
>> ->ki_complete?
>
> Forgot about that one, as for restoring it back, io_uring compares
> or calls ->ki_complete in a couple of places, this way the patch
> is more contained. It can definitely be refactored on top.
I'd be tempted to do that for the fix too, the patch as-is is a
bit of an eye sore... Hmm.
--
Jens Axboe
