Re: [PATCH for-next] io_uring: add async data clear/free helpers

[Jens Axboe <axboe@kernel.dk>]

On 8/26/25 10:10 AM, Caleb Sander Mateos wrote:
> On Tue, Aug 26, 2025 at 9:04 AM Jens Axboe <axboe@kernel.dk> wrote:
>>
>> On 8/26/25 9:27 AM, Caleb Sander Mateos wrote:
>>>> diff --git a/io_uring/io_uring.h b/io_uring/io_uring.h
>>>> index 2e4f7223a767..86613b8224bd 100644
>>>> --- a/io_uring/io_uring.h
>>>> +++ b/io_uring/io_uring.h
>>>> @@ -281,6 +281,19 @@ static inline bool req_has_async_data(struct io_kiocb *req)
>>>>         return req->flags & REQ_F_ASYNC_DATA;
>>>>  }
>>>>
>>>> +static inline void io_req_async_data_clear(struct io_kiocb *req,
>>>> +                                          io_req_flags_t extra_flags)
>>>> +{
>>>> +       req->flags &= ~(REQ_F_ASYNC_DATA|extra_flags);
>>>> +       req->async_data = NULL;
>>>> +}
>>>> +
>>>> +static inline void io_req_async_data_free(struct io_kiocb *req)
>>>> +{
>>>> +       kfree(req->async_data);
>>>> +       io_req_async_data_clear(req, 0);
>>>> +}
>>>
>>> Would it make sense to also add a helper for assigning async_data that
>>> would also make sure to set REQ_F_ASYNC_DATA?
>>
>> I did consider that, but it's only futex that'd use it and just in those
>> two spots. So decided against it. At least for now.
> 
> Makes sense. I think it could also be used in
> io_uring_alloc_async_data(), but it's true there are fewer sites
> assigning to async_data than clearing it.

Indeed, most of the setting happens through the alloc async data
helper already. And yes it could get unified with that, but then
we'd probably want to extend that a bit more to return -ENOMEM
too for the cached case, and futex has one of each. IOW, I just
stopped at that point :-)

-- 
Jens Axboe