Re: [PATCH v2] io_uring: add support for linked SQE timeouts

[Jackie Liu <liuyun01@kylinos.cn>]

> 2019年11月7日 10:05，Jens Axboe <axboe@kernel.dk> 写道：
> 
> While we have support for generic timeouts, we don't have a way to tie
> a timeout to a specific SQE. The generic timeouts simply trigger wakeups
> on the CQ ring.
> 
> This adds support for IORING_OP_LINK_TIMEOUT. This command is only valid
> as a link to a previous command. The timeout specific can be either
> relative or absolute, following the same rules as IORING_OP_TIMEOUT. If
> the timeout triggers before the dependent command completes, it will
> attempt to cancel that command. Likewise, if the dependent command
> completes before the timeout triggers, it will cancel the timeout.
> 
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> 
> ---
> 
> Changes since v1:
> - Move required locking outside of io_req_link_next(), it's much
>  cleaner this way and avoids sparse complaining.
> - Avoid 32-bit complaint on casting to pointer of different size
> - Fix IORING_TIMEOUT_ABS, used sqe->flags instead of timeout_flags
> - Rebase on top of current tree
> 
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index c39d1c50a3be..e29ecc1b0218 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -329,6 +329,7 @@ struct io_kiocb {
> #define REQ_F_IO_DRAIN		16	/* drain existing IO first */
> #define REQ_F_IO_DRAINED	32	/* drain done */
> #define REQ_F_LINK		64	/* linked sqes */
> +#define REQ_F_LINK_TIMEOUT	128	/* has linked timeout */
> #define REQ_F_FAIL_LINK		256	/* fail rest of links */
> #define REQ_F_SHADOW_DRAIN	512	/* link-drain shadow req */
> #define REQ_F_TIMEOUT		1024	/* timeout request */
> @@ -371,6 +372,7 @@ static void io_wq_submit_work(struct io_wq_work **workptr);
> static void io_cqring_fill_event(struct io_ring_ctx *ctx, u64 ki_user_data,
> 				 long res);
> static void __io_free_req(struct io_kiocb *req);
> +static void io_put_req(struct io_kiocb *req, struct io_kiocb **nxtptr);
> 
> static struct kmem_cache *req_cachep;
> 
> @@ -712,9 +714,28 @@ static void __io_free_req(struct io_kiocb *req)
> 	kmem_cache_free(req_cachep, req);
> }
> 
> +static bool io_link_cancel_timeout(struct io_ring_ctx *ctx,
> +				   struct io_kiocb *req)
> +{
> +	int ret;
> +
> +	ret = hrtimer_try_to_cancel(&req->timeout.timer);
> +	if (ret != -1) {
> +		io_cqring_fill_event(ctx, req->user_data, -ECANCELED);
> +		io_commit_cqring(ctx);
> +		req->flags &= ~REQ_F_LINK;
> +		__io_free_req(req);
> +		return true;
> +	}
> +
> +	return false;
> +}
> +
> static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr)
> {
> +	struct io_ring_ctx *ctx = req->ctx;
> 	struct io_kiocb *nxt;
> +	bool wake_ev = false;
> 
> 	/*
> 	 * The list should never be empty when we are called here. But could
> @@ -722,7 +743,7 @@ static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr)
> 	 * safe side.
> 	 */
> 	nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb, list);
> -	if (nxt) {
> +	while (nxt) {
> 		list_del(&nxt->list);
> 		if (!list_empty(&req->link_list)) {
> 			INIT_LIST_HEAD(&nxt->link_list);
> @@ -734,11 +755,23 @@ static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr)
> 		 * If we're in async work, we can continue processing the chain
> 		 * in this context instead of having to queue up new async work.
> 		 */
> -		if (nxtptr && current_work())
> +		if (req->flags & REQ_F_LINK_TIMEOUT) {
> +			wake_ev = io_link_cancel_timeout(ctx, nxt);
> +
> +			/* we dropped this link, get next */
> +			nxt = list_first_entry_or_null(&req->link_list,
> +							struct io_kiocb, list);
> +		} else if (nxtptr && current_work()) {
> 			*nxtptr = nxt;
> -		else
> +			nxt = NULL;

Use 'break' is more better? no need to set variables and compare.

> +		} else {
> 			io_queue_async_work(req->ctx, nxt);
> +			nxt = NULL;
> +		}
> 	}

--
BR, Jackie Liu