[matthewtolman]

What Changed

Added wrapper for io_uring_prep_link_timeout which allows adding timeouts to I/O operations (e.g. read).

I/O Uring will dispatch two events for a link timeout, one for the original I/O operation (e.g. read) with -ETIMEDOUT when there's a timeout, and another for the timeout itself with code -ETIME on a timeout.

If there isn't a timeout, then two events are still emitted, one for the original I/O operation (e.g. read) with the normal return value, and another for the timeout itself with code -ECANCELLED.

To prevent confusion around doing two co_await or managing two events for a timeout event, I opted to set the timeout to have an empty callback. Since it's using an empty callback and not a coroutine, there is no need to keep the return object "in scope", so I don't return anything. That said, I'm open to feedback on changing the API.

I also added a demo on how to use it. The demo adapts the echo server demo and adds a timeout.

Usage:

// Timeout duration
__kernel_timespec timeout = { .tv_sec = 1, .tv_nsec = 0 };

// Queue an operation (but don't submit yet)
// Operation MUST be linked to follow up operation
auto task = service.read(fd, buff, buffLen, -1, IOSQE_IO_LINK);

// Link a timeout to our queued I/O operation
// Since this has an internal callback, we don't save it or see the event on the CQE
service.link_timeout(&timeout);

// Now we can dispatch the task
auto res = co_await task;

// check for errors
if (res < 0) {
  // See if we got a timeout
  if (res == -ETIMEDOUT) {
    fmt::println("Timeout detected!");
  }
  co_return;
}

// process bytes