On 3/23/23 23:10, 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.
It's ok, I'm not sure there is anyone using sequences
For the second case, the `target_seq' field in `struct io_timeout' is
re-purposed to track the remaining number of timeouts.
We have space in struct io_timeout, let's just add another
field there.
[...]
static bool io_kill_timeout(struct io_kiocb *req, int status)
__must_hold(&req->ctx->timeout_lock)
{
@@ -202,6 +215,13 @@ static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
struct io_ring_ctx *ctx = req->ctx;
unsigned long flags;
+ if (!io_timeout_finish(timeout, data)) {
+ io_aux_cqe(req->ctx, false, req->cqe.user_data, -ETIME,
+ IORING_CQE_F_MORE, true);
We can't post a cqe from here, it should be a task context,
e.g. using tw
+ hrtimer_forward_now(&data->timer, timespec64_to_ktime(data->ts));
+ return HRTIMER_RESTART;
+ }
+
[...]
--
Pavel Begunkov