[PATCHSET 0/2] Accept tweaks

[PATCHSET 0/2] Accept tweaks

[Jens Axboe]

Hi,

Add support for a forced nonblock accept attempt, returning -EAGAIN
if nothing is there to accept rather than arm poll. This is similar to
similarly provided "-EAGAIN is final" for other opcodes.

Add support for doing polling first on accept. Again similar to support
we have for eg recv/recvmsg on relying on poll before initiating the
first accept request, avoiding expensive setup only to tear it down and
wait for a retry.

 include/uapi/linux/io_uring.h |  2 ++
 io_uring/net.c                | 22 ++++++++++++++++------
 2 files changed, 18 insertions(+), 6 deletions(-)

-- 
Jens Axboe

[PATCH 1/2] io_uring/net: add IORING_ACCEPT_DONTWAIT flag

[Jens Axboe]

This allows the caller to perform a non-blocking attempt, similarly to
how recvmsg has MSG_DONTWAIT. If set, and we get -EAGAIN on a connection
attempt, propagate the result to userspace rather than arm poll and
wait for a retry.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/uapi/linux/io_uring.h |  1 +
 io_uring/net.c                | 15 +++++++++------
 2 files changed, 10 insertions(+), 6 deletions(-)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index f093cb2300d9..4a645d15516f 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -379,6 +379,7 @@ enum io_uring_op {
  * accept flags stored in sqe->ioprio
  */
 #define IORING_ACCEPT_MULTISHOT	(1U << 0)
+#define IORING_ACCEPT_DONTWAIT	(1U << 1)
 
 /*
  * IORING_OP_MSG_RING command types, stored in sqe->addr
diff --git a/io_uring/net.c b/io_uring/net.c
index b0bf8471ecb7..7861bc8fe8b1 100644
--- a/io_uring/net.c
+++ b/io_uring/net.c
@@ -28,6 +28,7 @@ struct io_accept {
 	struct sockaddr __user		*addr;
 	int __user			*addr_len;
 	int				flags;
+	int				iou_flags;
 	u32				file_slot;
 	unsigned long			nofile;
 };
@@ -1489,7 +1490,6 @@ void io_sendrecv_fail(struct io_kiocb *req)
 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
-	unsigned flags;
 
 	if (sqe->len || sqe->buf_index)
 		return -EINVAL;
@@ -1498,15 +1498,15 @@ int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
 	accept->flags = READ_ONCE(sqe->accept_flags);
 	accept->nofile = rlimit(RLIMIT_NOFILE);
-	flags = READ_ONCE(sqe->ioprio);
-	if (flags & ~IORING_ACCEPT_MULTISHOT)
+	accept->iou_flags = READ_ONCE(sqe->ioprio);
+	if (accept->iou_flags & ~(IORING_ACCEPT_MULTISHOT | IORING_ACCEPT_DONTWAIT))
 		return -EINVAL;
 
 	accept->file_slot = READ_ONCE(sqe->file_index);
 	if (accept->file_slot) {
 		if (accept->flags & SOCK_CLOEXEC)
 			return -EINVAL;
-		if (flags & IORING_ACCEPT_MULTISHOT &&
+		if (accept->iou_flags & IORING_ACCEPT_MULTISHOT &&
 		    accept->file_slot != IORING_FILE_INDEX_ALLOC)
 			return -EINVAL;
 	}
@@ -1514,8 +1514,10 @@ int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 		return -EINVAL;
 	if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
 		accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
-	if (flags & IORING_ACCEPT_MULTISHOT)
+	if (accept->iou_flags & IORING_ACCEPT_MULTISHOT)
 		req->flags |= REQ_F_APOLL_MULTISHOT;
+	if (accept->iou_flags & IORING_ACCEPT_DONTWAIT)
+		req->flags |= REQ_F_NOWAIT;
 	return 0;
 }
 
@@ -1540,7 +1542,8 @@ int io_accept(struct io_kiocb *req, unsigned int issue_flags)
 		if (!fixed)
 			put_unused_fd(fd);
 		ret = PTR_ERR(file);
-		if (ret == -EAGAIN && force_nonblock) {
+		if (ret == -EAGAIN && force_nonblock &&
+		    !(accept->iou_flags & IORING_ACCEPT_DONTWAIT)) {
 			/*
 			 * if it's multishot and polled, we don't need to
 			 * return EAGAIN to arm the poll infra since it
-- 
2.43.0

[PATCH 2/2] io_uring/net: add IORING_ACCEPT_POLL_FIRST flag

[Jens Axboe]

Similarly to how polling first is supported for receive, it makes sense
to provide the same for accept. An accept operation does a lot of
expensive setup, like allocating an fd, a socket/inode, etc. If no
connection request is already pending, this is wasted and will just be
cleaned up and freed, only to retry via the usual poll trigger.

Add IORING_ACCEPT_POLL_FIRST, which tells accept to only initiate the
accept request if poll says we have something to accept.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/uapi/linux/io_uring.h | 1 +
 io_uring/net.c                | 9 ++++++++-
 2 files changed, 9 insertions(+), 1 deletion(-)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 4a645d15516f..6dbac55f8686 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -380,6 +380,7 @@ enum io_uring_op {
  */
 #define IORING_ACCEPT_MULTISHOT	(1U << 0)
 #define IORING_ACCEPT_DONTWAIT	(1U << 1)
+#define IORING_ACCEPT_POLL_FIRST	(1U << 2)
 
 /*
  * IORING_OP_MSG_RING command types, stored in sqe->addr
diff --git a/io_uring/net.c b/io_uring/net.c
index 7861bc8fe8b1..070dea9a4eda 100644
--- a/io_uring/net.c
+++ b/io_uring/net.c
@@ -1487,6 +1487,9 @@ void io_sendrecv_fail(struct io_kiocb *req)
 		req->cqe.flags |= IORING_CQE_F_MORE;
 }
 
+#define ACCEPT_FLAGS	(IORING_ACCEPT_MULTISHOT | IORING_ACCEPT_DONTWAIT | \
+			 IORING_ACCEPT_POLL_FIRST)
+
 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
@@ -1499,7 +1502,7 @@ int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	accept->flags = READ_ONCE(sqe->accept_flags);
 	accept->nofile = rlimit(RLIMIT_NOFILE);
 	accept->iou_flags = READ_ONCE(sqe->ioprio);
-	if (accept->iou_flags & ~(IORING_ACCEPT_MULTISHOT | IORING_ACCEPT_DONTWAIT))
+	if (accept->iou_flags & ~ACCEPT_FLAGS)
 		return -EINVAL;
 
 	accept->file_slot = READ_ONCE(sqe->file_index);
@@ -1530,6 +1533,10 @@ int io_accept(struct io_kiocb *req, unsigned int issue_flags)
 	struct file *file;
 	int ret, fd;
 
+	if (!(req->flags & REQ_F_POLLED) &&
+	    accept->iou_flags & IORING_ACCEPT_POLL_FIRST)
+		return -EAGAIN;
+
 retry:
 	if (!fixed) {
 		fd = __get_unused_fd_flags(accept->flags, accept->nofile);
-- 
2.43.0