Re: [PATCH] io_uring: avoid iowq again trap

[Pavel Begunkov <asml.silence@gmail.com>]

On 5/15/22 08:31, Hao Xu wrote:
> On 5/13/22 18:24, Pavel Begunkov wrote:
>> If an opcode handler semi-reliably returns -EAGAIN, io_wq_submit_work()
> 
> Hi Pavel,
> When would it return -EAGAIN in non-IOPOLL mode?

I didn't see it in the wild but stumbled upon while preparing some
future patches. I hope it's not a real issue, but it's better to not
leave a way for some driver/etc. to abuse it.


>> might continue busily hammer the same handler over and over again, which
>> is not ideal. The -EAGAIN handling in question was put there only for
>> IOPOLL, so restrict it to IOPOLL mode only.
>>
>> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
>> ---
>>   fs/io_uring.c | 2 ++
>>   1 file changed, 2 insertions(+)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index e01f595f5b7d..3af1905efc78 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -7319,6 +7319,8 @@ static void io_wq_submit_work(struct io_wq_work *work)
>>            * wait for request slots on the block side.
>>            */
>>           if (!needs_poll) {
>> +            if (!(req->ctx->flags & IORING_SETUP_IOPOLL))
>> +                break;
>>               cond_resched();
>>               continue;
>>           }
> 

-- 
Pavel Begunkov