[PATCH] io_uring: move iopoll reissue into regular IO path

[PATCH] io_uring: move iopoll reissue into regular IO path

[Pavel Begunkov]

230d50d448acb ("io_uring: move reissue into regular IO path")
made non-IOPOLL I/O to not retry from ki_complete handler. Follow it
steps and do the same for IOPOLL. Same problems, same implementation,
same -EAGAIN assumptions.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---

Needs more testing and with NVMe. It's on top of io_uring-5.15 for now,
we'll be rebased on whatever is needed after we're happy with everything

 fs/io_uring.c | 23 +++++------------------
 1 file changed, 5 insertions(+), 18 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index f6fae01c039c..7fb5f2acd274 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -736,7 +736,6 @@ enum {
 	REQ_F_BUFFER_SELECTED_BIT,
 	REQ_F_COMPLETE_INLINE_BIT,
 	REQ_F_REISSUE_BIT,
-	REQ_F_DONT_REISSUE_BIT,
 	REQ_F_CREDS_BIT,
 	REQ_F_REFCOUNT_BIT,
 	REQ_F_ARM_LTIMEOUT_BIT,
@@ -783,8 +782,6 @@ enum {
 	REQ_F_COMPLETE_INLINE	= BIT(REQ_F_COMPLETE_INLINE_BIT),
 	/* caller should reissue async */
 	REQ_F_REISSUE		= BIT(REQ_F_REISSUE_BIT),
-	/* don't attempt request reissue, see io_rw_reissue() */
-	REQ_F_DONT_REISSUE	= BIT(REQ_F_DONT_REISSUE_BIT),
 	/* supports async reads */
 	REQ_F_NOWAIT_READ	= BIT(REQ_F_NOWAIT_READ_BIT),
 	/* supports async writes */
@@ -2428,14 +2425,6 @@ static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
 	while (!list_empty(done)) {
 		req = list_first_entry(done, struct io_kiocb, inflight_entry);
 		list_del(&req->inflight_entry);
-
-		if (READ_ONCE(req->result) == -EAGAIN &&
-		    !(req->flags & REQ_F_DONT_REISSUE)) {
-			req->iopoll_completed = 0;
-			io_req_task_queue_reissue(req);
-			continue;
-		}
-
 		__io_fill_cqe(ctx, req->user_data, req->result,
 			      io_put_rw_kbuf(req));
 		(*nr_events)++;
@@ -2699,10 +2688,9 @@ static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
 	if (kiocb->ki_flags & IOCB_WRITE)
 		kiocb_end_write(req);
 	if (unlikely(res != req->result)) {
-		if (!(res == -EAGAIN && io_rw_should_reissue(req) &&
-		    io_resubmit_prep(req))) {
-			req_set_fail(req);
-			req->flags |= REQ_F_DONT_REISSUE;
+		if (res == -EAGAIN && io_rw_should_reissue(req)) {
+			req->flags |= REQ_F_REISSUE;
+			return;
 		}
 	}
 
@@ -2916,7 +2904,6 @@ static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
 {
 	struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
 	struct io_async_rw *io = req->async_data;
-	bool check_reissue = kiocb->ki_complete == io_complete_rw;
 
 	/* add previously done IO, if any */
 	if (io && io->bytes_done > 0) {
@@ -2928,12 +2915,12 @@ static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
 
 	if (req->flags & REQ_F_CUR_POS)
 		req->file->f_pos = kiocb->ki_pos;
-	if (ret >= 0 && check_reissue)
+	if (ret >= 0 && (kiocb->ki_complete == io_complete_rw))
 		__io_complete_rw(req, ret, 0, issue_flags);
 	else
 		io_rw_done(kiocb, ret);
 
-	if (check_reissue && (req->flags & REQ_F_REISSUE)) {
+	if (req->flags & REQ_F_REISSUE) {
 		req->flags &= ~REQ_F_REISSUE;
 		if (io_resubmit_prep(req)) {
 			io_req_task_queue_reissue(req);
-- 
2.33.0

Re: [PATCH] io_uring: move iopoll reissue into regular IO path

[Jens Axboe]

On 9/14/21 4:34 PM, Pavel Begunkov wrote:
> 230d50d448acb ("io_uring: move reissue into regular IO path")
> made non-IOPOLL I/O to not retry from ki_complete handler. Follow it
> steps and do the same for IOPOLL. Same problems, same implementation,
> same -EAGAIN assumptions.

This looks good to me. But I don't think it's against io_uring-5.15?
Trivial reject, but looks like -next to me.


-- 
Jens Axboe

Re: [PATCH] io_uring: move iopoll reissue into regular IO path

[Pavel Begunkov]

On 9/15/21 2:14 AM, Jens Axboe wrote:
> On 9/14/21 4:34 PM, Pavel Begunkov wrote:
>> 230d50d448acb ("io_uring: move reissue into regular IO path")
>> made non-IOPOLL I/O to not retry from ki_complete handler. Follow it
>> steps and do the same for IOPOLL. Same problems, same implementation,
>> same -EAGAIN assumptions.
> 
> This looks good to me. But I don't think it's against io_uring-5.15?
> Trivial reject, but looks like -next to me.

Was checking out to io_uring-5.15, apparently something gone wrong.
Not only that, but also a chunk left un-amended...

I'll resend against io_uring-5.15

-- 
Pavel Begunkov