Re: [PATCH v2] io_uring: add support for multishot timeouts

[David Wei <davidhwei@meta.com>]

On 14/04/2023 07:10, Pavel Begunkov wrote:
> > 
> On 4/12/23 23:27, David Wei wrote:
>> A multishot timeout submission will repeatedly generate completions with
>> the IORING_CQE_F_MORE cflag set. Depending on the value of the `off' field
>> in the submission, these timeouts can either repeat indefinitely until
>> cancelled (`off' = 0) or for a fixed number of times (`off' > 0).
>>
>> Only noseq timeouts (i.e. not dependent on the number of I/O
>> completions) are supported.
>>
>> An indefinite timer will be cancelled with EOVERFLOW if the CQ ever
>> overflows.
> 
> Seems mostly fine, two comments below
> 
> 
>> Signed-off-by: David Wei <davidhwei@meta.com>
>> ---
>>   include/uapi/linux/io_uring.h |  1 +
>>   io_uring/timeout.c            | 59 +++++++++++++++++++++++++++++++++--
>>   2 files changed, 57 insertions(+), 3 deletions(-)
>>
>> diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
>> index f8d14d1c58d3..0716cb17e436 100644
>> --- a/include/uapi/linux/io_uring.h
>> +++ b/include/uapi/linux/io_uring.h
>> @@ -250,6 +250,7 @@ enum io_uring_op {
>>   #define IORING_TIMEOUT_REALTIME        (1U << 3)
>>   #define IORING_LINK_TIMEOUT_UPDATE    (1U << 4)
>>   #define IORING_TIMEOUT_ETIME_SUCCESS    (1U << 5)
>> +#define IORING_TIMEOUT_MULTISHOT    (1U << 6)
>>   #define IORING_TIMEOUT_CLOCK_MASK    (IORING_TIMEOUT_BOOTTIME | IORING_TIMEOUT_REALTIME)
>>   #define IORING_TIMEOUT_UPDATE_MASK    (IORING_TIMEOUT_UPDATE | IORING_LINK_TIMEOUT_UPDATE)
>>   /*
>> diff --git a/io_uring/timeout.c b/io_uring/timeout.c
>> index 5c6c6f720809..61b8488565ce 100644
> ...
>> +static void io_timeout_complete(struct io_kiocb *req, struct io_tw_state *ts)
>> +{
>> +    struct io_timeout *timeout = io_kiocb_to_cmd(req, struct io_timeout);
>> +    struct io_timeout_data *data = req->async_data;
>> +    struct io_ring_ctx *ctx = req->ctx;
>> +
>> +    if (!io_timeout_finish(timeout, data)) {
>> +        bool filled;
>> +        filled = io_aux_cqe(ctx, false, req->cqe.user_data, -ETIME,
>> +                      IORING_CQE_F_MORE, false);
>> +        if (filled) {
>> +            /* re-arm timer */
>> +            spin_lock_irq(&ctx->timeout_lock);
>> +            list_add(&timeout->list, ctx->timeout_list.prev);
>> +            data->timer.function = io_timeout_fn;
>> +            hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
>> +            spin_unlock_irq(&ctx->timeout_lock);
>> +            return;
>> +        }
>> +        io_req_set_res(req, -EOVERFLOW, 0);
> 
> Let's not change the return value. It's considered a normal completion
> and we don't change the code for them. And there is IORING_CQE_F_MORE
> for userspace to figure out that it has been terminated.

Makes sense. I'll keep the return value as-is.

> 
> 
>> +    }
>> +
>> +    io_req_task_complete(req, ts);
>> +}
>> +
>>   static bool io_kill_timeout(struct io_kiocb *req, int status)
>>       __must_hold(&req->ctx->timeout_lock)
>>   {
>> @@ -212,7 +253,7 @@ static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
>>           req_set_fail(req);
>>         io_req_set_res(req, -ETIME, 0);
>> -    req->io_task_work.func = io_req_task_complete;
>> +    req->io_task_work.func = io_timeout_complete;
>>       io_req_task_work_add(req);
>>       return HRTIMER_NORESTART;
>>   }
>> @@ -470,16 +511,28 @@ static int __io_timeout_prep(struct io_kiocb *req,
>>           return -EINVAL;
>>       flags = READ_ONCE(sqe->timeout_flags);
>>       if (flags & ~(IORING_TIMEOUT_ABS | IORING_TIMEOUT_CLOCK_MASK |
>> -              IORING_TIMEOUT_ETIME_SUCCESS))
>> +              IORING_TIMEOUT_ETIME_SUCCESS |
>> +              IORING_TIMEOUT_MULTISHOT)) {
>>           return -EINVAL;
>> +    }
> 
> Please, don't add braces, they're not needed here.

Thanks, will remove.

> 
>>       /* more than one clock specified is invalid, obviously */
>>       if (hweight32(flags & IORING_TIMEOUT_CLOCK_MASK) > 1)
>>           return -EINVAL;
>> +    /* multishot requests only make sense with rel values */
>> +    if (!(~flags & (IORING_TIMEOUT_MULTISHOT | IORING_TIMEOUT_ABS)))
>> +        return -EINVAL;
>>         INIT_LIST_HEAD(&timeout->list);
>>       timeout->off = off;
>>       if (unlikely(off && !req->ctx->off_timeout_used))
>>           req->ctx->off_timeout_used = true;
>> +    /*
>> +     * for multishot reqs w/ fixed nr of repeats, target_seq tracks the
>> +     * remaining nr
>> +     */
>> +    timeout->repeats = 0;
>> +    if ((flags & IORING_TIMEOUT_MULTISHOT) && off > 0)
>> +        timeout->repeats = off;
>>         if (WARN_ON_ONCE(req_has_async_data(req)))
>>           return -EFAULT;
>