[PATCHSET 0/2] Support for provided buffers for send

[PATCHSET 0/2] Support for provided buffers for send

[Jens Axboe]

Hi,

We never supported provided buffers for sends, because it didn't seem
to make a lot of sense. But it actually does make a lot of sense! If
an app is receiving data, doing something with it, and then sending
either the same or another buffer out based on that, then if we use
provided buffers for sends we can guarantee that the sends are
serialized. This is because provided buffer rings are FIFO ordered,
as it's a ring buffer, and hence it doesn't really matter if you
have more than one send inflight.

This provides a nice efficiency win, but more importantly, it reduces
the complexity in the application as it no longer needs to track a
potential backlog of sends. The app just sets up a send based buffer
ring, exactly like it does for incoming data. And that's it, no more
dealing with serialized sends.

In some testing with proxy [1], in basic shuffling of packets I see
a 36% improvement with this over manually dealing with sends. That's
a pretty big win on top of making the app simpler.

This also opens the door for multishot send requests, which is an
interesting future venue to pursue!

You can also find the patches here:

https://git.kernel.dk/cgit/linux/log/?h=io_uring-send-queue

[1] https://git.kernel.dk/cgit/liburing/tree/examples/proxy.c

-- 
Jens Axboe

[PATCH 1/2] io_uring/net: add provided buffer support for IORING_OP_SEND

[Jens Axboe]

It's pretty trivial to wire up provided buffer support for the send
side, just like we do on the receive side. This enables setting up
a buffer ring that an application can use to push pending sends to,
and then have a send pick a buffer from that ring.

One of the challenges with async IO and networking sends is that you
can get into reordering conditions if you have more than one inflight
at the same time. Consider the following scenario where everything is
fine:

1) App queues sendA for socket1
2) App queues sendB for socket1
3) App does io_uring_submit()
4) sendA is issued, completes successfully, posts CQE
5) sendB is issued, completes successfully, posts CQE

All is fine. Requests are always issued in-order, and both complete
inline as most sends do.

However, if we're flooding socket1 with sends, the following could
also result from the same sequence:

1) App queues sendA for socket1
2) App queues sendB for socket1
3) App does io_uring_submit()
4) sendA is issued, socket1 is full, poll is armed for retry
5) Space frees up in socket1, this triggers sendA retry via task_work
6) sendB is issued, completes successfully, posts CQE
7) sendA is retried, completes successfully, posts CQE

Now we've sent sendB before sendA, which can make things unhappy. If
both sendA and sendB had been using provided buffers, then it would look
as follows instead:

1) App queues dataA for sendA, queues sendA for socket1
2) App queues dataB for sendB queues sendB for socket1
3) App does io_uring_submit()
4) sendA is issued, socket1 is full, poll is armed for retry
5) Space frees up in socket1, this triggers sendA retry via task_work
6) sendB is issued, picks first buffer (dataA), completes successfully,
   posts CQE (which says "I sent dataA")
7) sendA is retried, picks first buffer (dataB), completes successfully,
   posts CQE (which says "I sent dataB")

Now we've sent the data in order, and everybody is happy.

It's worth noting that this also opens the door for supporting multishot
sends, as provided buffers would be a prerequisite for that. Those can
trigger either when new buffers are added to the outgoing ring, or (if
stalled due to lack of space) when space frees up in the socket.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/uapi/linux/io_uring.h |  1 +
 io_uring/io_uring.c           |  3 ++-
 io_uring/net.c                | 19 ++++++++++++++++---
 io_uring/opdef.c              |  1 +
 4 files changed, 20 insertions(+), 4 deletions(-)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 7bd10201a02b..74c3afac9c63 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -522,6 +522,7 @@ struct io_uring_params {
 #define IORING_FEAT_CQE_SKIP		(1U << 11)
 #define IORING_FEAT_LINKED_FILE		(1U << 12)
 #define IORING_FEAT_REG_REG_RING	(1U << 13)
+#define IORING_FEAT_SEND_BUFS		(1U << 14)
 
 /*
  * io_uring_register(2) opcodes and arguments
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index cf2f514b7cc0..f6332fc56bed 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -3962,7 +3962,8 @@ static __cold int io_uring_create(unsigned entries, struct io_uring_params *p,
 			IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
 			IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS |
 			IORING_FEAT_RSRC_TAGS | IORING_FEAT_CQE_SKIP |
-			IORING_FEAT_LINKED_FILE | IORING_FEAT_REG_REG_RING;
+			IORING_FEAT_LINKED_FILE | IORING_FEAT_REG_REG_RING |
+			IORING_FEAT_SEND_BUFS;
 
 	if (copy_to_user(params, p, sizeof(*p))) {
 		ret = -EFAULT;
diff --git a/io_uring/net.c b/io_uring/net.c
index 161622029147..6c22091a0c77 100644
--- a/io_uring/net.c
+++ b/io_uring/net.c
@@ -365,8 +365,10 @@ int io_send(struct io_kiocb *req, unsigned int issue_flags)
 {
 	struct sockaddr_storage __address;
 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
-	struct msghdr msg;
+	size_t len = sr->len;
+	unsigned int cflags;
 	struct socket *sock;
+	struct msghdr msg;
 	unsigned flags;
 	int min_ret = 0;
 	int ret;
@@ -399,7 +401,17 @@ int io_send(struct io_kiocb *req, unsigned int issue_flags)
 	if (unlikely(!sock))
 		return -ENOTSOCK;
 
-	ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len, &msg.msg_iter);
+	if (io_do_buffer_select(req)) {
+		void __user *buf;
+
+		buf = io_buffer_select(req, &len, issue_flags);
+		if (!buf)
+			return -ENOBUFS;
+		sr->buf = buf;
+		sr->len = len;
+	}
+
+	ret = import_ubuf(ITER_SOURCE, sr->buf, len, &msg.msg_iter);
 	if (unlikely(ret))
 		return ret;
 
@@ -431,7 +443,8 @@ int io_send(struct io_kiocb *req, unsigned int issue_flags)
 		ret += sr->done_io;
 	else if (sr->done_io)
 		ret = sr->done_io;
-	io_req_set_res(req, ret, 0);
+	cflags = io_put_kbuf(req, issue_flags);
+	io_req_set_res(req, ret, cflags);
 	return IOU_OK;
 }
 
diff --git a/io_uring/opdef.c b/io_uring/opdef.c
index 9c080aadc5a6..88fbe5cfd379 100644
--- a/io_uring/opdef.c
+++ b/io_uring/opdef.c
@@ -273,6 +273,7 @@ const struct io_issue_def io_issue_defs[] = {
 		.audit_skip		= 1,
 		.ioprio			= 1,
 		.manual_alloc		= 1,
+		.buffer_select		= 1,
 #if defined(CONFIG_NET)
 		.prep			= io_sendmsg_prep,
 		.issue			= io_send,
-- 
2.43.0

[PATCH 2/2] io_uring/net: add provided buffer support for IORING_OP_SENDMSG

[Jens Axboe]

Adds provided buffer support for sendmsg as well, see the previous commit
that added it to IORING_OP_SEND for a longer explanation of why this
makes sense.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/net.c   | 15 ++++++++++++++-
 io_uring/opdef.c |  1 +
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/io_uring/net.c b/io_uring/net.c
index 6c22091a0c77..fcc0ce5fecd6 100644
--- a/io_uring/net.c
+++ b/io_uring/net.c
@@ -304,6 +304,7 @@ int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
 	struct io_async_msghdr iomsg, *kmsg;
 	struct socket *sock;
+	unsigned int cflags;
 	unsigned flags;
 	int min_ret = 0;
 	int ret;
@@ -326,6 +327,17 @@ int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
 		return io_setup_async_msg(req, kmsg, issue_flags);
 
+	if (io_do_buffer_select(req)) {
+		void __user *buf;
+		size_t len = 0;
+
+		buf = io_buffer_select(req, &len, issue_flags);
+		if (!buf)
+			return -ENOBUFS;
+
+		iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_SOURCE, buf, len);
+	}
+
 	flags = sr->msg_flags;
 	if (issue_flags & IO_URING_F_NONBLOCK)
 		flags |= MSG_DONTWAIT;
@@ -357,7 +369,8 @@ int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
 		ret += sr->done_io;
 	else if (sr->done_io)
 		ret = sr->done_io;
-	io_req_set_res(req, ret, 0);
+	cflags = io_put_kbuf(req, issue_flags);
+	io_req_set_res(req, ret, cflags);
 	return IOU_OK;
 }
 
diff --git a/io_uring/opdef.c b/io_uring/opdef.c
index 88fbe5cfd379..1f6b09e61ef8 100644
--- a/io_uring/opdef.c
+++ b/io_uring/opdef.c
@@ -139,6 +139,7 @@ const struct io_issue_def io_issue_defs[] = {
 		.pollout		= 1,
 		.ioprio			= 1,
 		.manual_alloc		= 1,
+		.buffer_select		= 1,
 #if defined(CONFIG_NET)
 		.prep			= io_sendmsg_prep,
 		.issue			= io_sendmsg,
-- 
2.43.0