[PATCH v2 for-next 0/3] io_uring: multishot recvmsg

[PATCH v2 for-next 0/3] io_uring: multishot recvmsg

[Dylan Yudaken]

This series adds multishot support to recvmsg in io_uring.

The idea is that you submit a single multishot recvmsg and then receive
completions as and when data arrives. For recvmsg each completion also has
control data, and this is necessarily included in the same buffer as the
payload.

In order to do this a new structure is used: io_uring_recvmsg_out. This
specifies the length written of the name, control and payload. As well as
including the flags.
The layout of the buffer is <header><name><control><payload> where the
lengths are those specified in the original msghdr used to issue the recvmsg.

I suspect this API will be the most contentious part of this series and would
appreciate any comments on it.

For completeness I considered having the original struct msghdr as the header,
but size wise it is much bigger (72 bytes including an iovec vs 16 bytes here).
Testing also showed a 1% slowdown in terms of QPS.

Using a mini network tester [1] shows 14% QPS improvment using this API, however
this is likely to go down to ~8% with the latest allocation cache added by Jens.

[1]: https://github.com/DylanZA/netbench/tree/main

Patches 1,2 change the copy_msghdr code to take a user_msghdr as input
Patch 3 is the multishot feature

v2:
 * Rebase without netbuf recycling provided by io_uring
 * Fix payload field output with MSG_TRUNC set to match recvmsg(2)

Dylan Yudaken (3):
  net: copy from user before calling __copy_msghdr
  net: copy from user before calling __get_compat_msghdr
  io_uring: support multishot in recvmsg

 include/linux/socket.h        |   7 +-
 include/net/compat.h          |   5 +-
 include/uapi/linux/io_uring.h |   7 ++
 io_uring/net.c                | 212 ++++++++++++++++++++++++++++------
 io_uring/net.h                |   6 +
 net/compat.c                  |  39 +++----
 net/socket.c                  |  37 +++---
 7 files changed, 228 insertions(+), 85 deletions(-)


base-commit: 20898aeac6b82d8eb6247754494584b2f6cafd53
-- 
2.30.2

[PATCH v2 for-next 1/3] net: copy from user before calling __copy_msghdr

[Dylan Yudaken]

this is in preparation for multishot receive from io_uring, where it needs
to have access to the original struct user_msghdr.

functionally this should be a no-op.

Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Dylan Yudaken <dylany@fb.com>
---
 include/linux/socket.h |  7 +++----
 io_uring/net.c         | 17 +++++++++--------
 net/socket.c           | 37 ++++++++++++++++---------------------
 3 files changed, 28 insertions(+), 33 deletions(-)

diff --git a/include/linux/socket.h b/include/linux/socket.h
index 17311ad9f9af..be24f1c8568a 100644
--- a/include/linux/socket.h
+++ b/include/linux/socket.h
@@ -416,10 +416,9 @@ extern int recvmsg_copy_msghdr(struct msghdr *msg,
 			       struct user_msghdr __user *umsg, unsigned flags,
 			       struct sockaddr __user **uaddr,
 			       struct iovec **iov);
-extern int __copy_msghdr_from_user(struct msghdr *kmsg,
-				   struct user_msghdr __user *umsg,
-				   struct sockaddr __user **save_addr,
-				   struct iovec __user **uiov, size_t *nsegs);
+extern int __copy_msghdr(struct msghdr *kmsg,
+			 struct user_msghdr *umsg,
+			 struct sockaddr __user **save_addr);
 
 /* helpers which do the actual work for syscalls */
 extern int __sys_recvfrom(int fd, void __user *ubuf, size_t size,
diff --git a/io_uring/net.c b/io_uring/net.c
index dc9190eafbe7..da7667ed3610 100644
--- a/io_uring/net.c
+++ b/io_uring/net.c
@@ -329,31 +329,32 @@ static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
 				 struct io_async_msghdr *iomsg)
 {
 	struct io_sr_msg *sr = io_kiocb_to_cmd(req);
-	struct iovec __user *uiov;
-	size_t iov_len;
+	struct user_msghdr msg;
 	int ret;
 
-	ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
-					&iomsg->uaddr, &uiov, &iov_len);
+	if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
+		return -EFAULT;
+
+	ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
 	if (ret)
 		return ret;
 
 	if (req->flags & REQ_F_BUFFER_SELECT) {
-		if (iov_len == 0) {
+		if (msg.msg_iovlen == 0) {
 			sr->len = iomsg->fast_iov[0].iov_len = 0;
 			iomsg->fast_iov[0].iov_base = NULL;
 			iomsg->free_iov = NULL;
-		} else if (iov_len > 1) {
+		} else if (msg.msg_iovlen > 1) {
 			return -EINVAL;
 		} else {
-			if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov)))
+			if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
 				return -EFAULT;
 			sr->len = iomsg->fast_iov[0].iov_len;
 			iomsg->free_iov = NULL;
 		}
 	} else {
 		iomsg->free_iov = iomsg->fast_iov;
-		ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
+		ret = __import_iovec(READ, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
 				     &iomsg->free_iov, &iomsg->msg.msg_iter,
 				     false);
 		if (ret > 0)
diff --git a/net/socket.c b/net/socket.c
index 96300cdc0625..843545c21ec2 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -2358,25 +2358,20 @@ struct used_address {
 	unsigned int name_len;
 };
 
-int __copy_msghdr_from_user(struct msghdr *kmsg,
-			    struct user_msghdr __user *umsg,
-			    struct sockaddr __user **save_addr,
-			    struct iovec __user **uiov, size_t *nsegs)
+int __copy_msghdr(struct msghdr *kmsg,
+		  struct user_msghdr *msg,
+		  struct sockaddr __user **save_addr)
 {
-	struct user_msghdr msg;
 	ssize_t err;
 
-	if (copy_from_user(&msg, umsg, sizeof(*umsg)))
-		return -EFAULT;
-
 	kmsg->msg_control_is_user = true;
 	kmsg->msg_get_inq = 0;
-	kmsg->msg_control_user = msg.msg_control;
-	kmsg->msg_controllen = msg.msg_controllen;
-	kmsg->msg_flags = msg.msg_flags;
+	kmsg->msg_control_user = msg->msg_control;
+	kmsg->msg_controllen = msg->msg_controllen;
+	kmsg->msg_flags = msg->msg_flags;
 
-	kmsg->msg_namelen = msg.msg_namelen;
-	if (!msg.msg_name)
+	kmsg->msg_namelen = msg->msg_namelen;
+	if (!msg->msg_name)
 		kmsg->msg_namelen = 0;
 
 	if (kmsg->msg_namelen < 0)
@@ -2386,11 +2381,11 @@ int __copy_msghdr_from_user(struct msghdr *kmsg,
 		kmsg->msg_namelen = sizeof(struct sockaddr_storage);
 
 	if (save_addr)
-		*save_addr = msg.msg_name;
+		*save_addr = msg->msg_name;
 
-	if (msg.msg_name && kmsg->msg_namelen) {
+	if (msg->msg_name && kmsg->msg_namelen) {
 		if (!save_addr) {
-			err = move_addr_to_kernel(msg.msg_name,
+			err = move_addr_to_kernel(msg->msg_name,
 						  kmsg->msg_namelen,
 						  kmsg->msg_name);
 			if (err < 0)
@@ -2401,12 +2396,10 @@ int __copy_msghdr_from_user(struct msghdr *kmsg,
 		kmsg->msg_namelen = 0;
 	}
 
-	if (msg.msg_iovlen > UIO_MAXIOV)
+	if (msg->msg_iovlen > UIO_MAXIOV)
 		return -EMSGSIZE;
 
 	kmsg->msg_iocb = NULL;
-	*uiov = msg.msg_iov;
-	*nsegs = msg.msg_iovlen;
 	return 0;
 }
 
@@ -2418,8 +2411,10 @@ static int copy_msghdr_from_user(struct msghdr *kmsg,
 	struct user_msghdr msg;
 	ssize_t err;
 
-	err = __copy_msghdr_from_user(kmsg, umsg, save_addr, &msg.msg_iov,
-					&msg.msg_iovlen);
+	if (copy_from_user(&msg, umsg, sizeof(*umsg)))
+		return -EFAULT;
+
+	err = __copy_msghdr(kmsg, &msg, save_addr);
 	if (err)
 		return err;
 
-- 
2.30.2

[PATCH v2 for-next 2/3] net: copy from user before calling __get_compat_msghdr

[Dylan Yudaken]

this is in preparation for multishot receive from io_uring, where it needs
to have access to the original struct user_msghdr.

functionally this should be a no-op.

Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Dylan Yudaken <dylany@fb.com>
---
 include/net/compat.h |  5 ++---
 io_uring/net.c       | 17 +++++++++--------
 net/compat.c         | 39 +++++++++++++++++----------------------
 3 files changed, 28 insertions(+), 33 deletions(-)

diff --git a/include/net/compat.h b/include/net/compat.h
index 595fee069b82..84c163f40f38 100644
--- a/include/net/compat.h
+++ b/include/net/compat.h
@@ -46,9 +46,8 @@ struct compat_rtentry {
 	unsigned short  rt_irtt;        /* Initial RTT                  */
 };
 
-int __get_compat_msghdr(struct msghdr *kmsg, struct compat_msghdr __user *umsg,
-			struct sockaddr __user **save_addr, compat_uptr_t *ptr,
-			compat_size_t *len);
+int __get_compat_msghdr(struct msghdr *kmsg, struct compat_msghdr *msg,
+			struct sockaddr __user **save_addr);
 int get_compat_msghdr(struct msghdr *, struct compat_msghdr __user *,
 		      struct sockaddr __user **, struct iovec **);
 int put_cmsg_compat(struct msghdr*, int, int, int, void *);
diff --git a/io_uring/net.c b/io_uring/net.c
index da7667ed3610..5bc3440a8290 100644
--- a/io_uring/net.c
+++ b/io_uring/net.c
@@ -369,24 +369,25 @@ static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
 					struct io_async_msghdr *iomsg)
 {
 	struct io_sr_msg *sr = io_kiocb_to_cmd(req);
+	struct compat_msghdr msg;
 	struct compat_iovec __user *uiov;
-	compat_uptr_t ptr;
-	compat_size_t len;
 	int ret;
 
-	ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr,
-				  &ptr, &len);
+	if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
+		return -EFAULT;
+
+	ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr);
 	if (ret)
 		return ret;
 
-	uiov = compat_ptr(ptr);
+	uiov = compat_ptr(msg.msg_iov);
 	if (req->flags & REQ_F_BUFFER_SELECT) {
 		compat_ssize_t clen;
 
-		if (len == 0) {
+		if (msg.msg_iovlen == 0) {
 			sr->len = 0;
 			iomsg->free_iov = NULL;
-		} else if (len > 1) {
+		} else if (msg.msg_iovlen > 1) {
 			return -EINVAL;
 		} else {
 			if (!access_ok(uiov, sizeof(*uiov)))
@@ -400,7 +401,7 @@ static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
 		}
 	} else {
 		iomsg->free_iov = iomsg->fast_iov;
-		ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
+		ret = __import_iovec(READ, (struct iovec __user *)uiov, msg.msg_iovlen,
 				   UIO_FASTIOV, &iomsg->free_iov,
 				   &iomsg->msg.msg_iter, true);
 		if (ret < 0)
diff --git a/net/compat.c b/net/compat.c
index 210fc3b4d0d8..513aa9a3fc64 100644
--- a/net/compat.c
+++ b/net/compat.c
@@ -34,20 +34,15 @@
 #include <net/compat.h>
 
 int __get_compat_msghdr(struct msghdr *kmsg,
-			struct compat_msghdr __user *umsg,
-			struct sockaddr __user **save_addr,
-			compat_uptr_t *ptr, compat_size_t *len)
+			struct compat_msghdr *msg,
+			struct sockaddr __user **save_addr)
 {
-	struct compat_msghdr msg;
 	ssize_t err;
 
-	if (copy_from_user(&msg, umsg, sizeof(*umsg)))
-		return -EFAULT;
-
-	kmsg->msg_flags = msg.msg_flags;
-	kmsg->msg_namelen = msg.msg_namelen;
+	kmsg->msg_flags = msg->msg_flags;
+	kmsg->msg_namelen = msg->msg_namelen;
 
-	if (!msg.msg_name)
+	if (!msg->msg_name)
 		kmsg->msg_namelen = 0;
 
 	if (kmsg->msg_namelen < 0)
@@ -57,15 +52,15 @@ int __get_compat_msghdr(struct msghdr *kmsg,
 		kmsg->msg_namelen = sizeof(struct sockaddr_storage);
 
 	kmsg->msg_control_is_user = true;
-	kmsg->msg_control_user = compat_ptr(msg.msg_control);
-	kmsg->msg_controllen = msg.msg_controllen;
+	kmsg->msg_control_user = compat_ptr(msg->msg_control);
+	kmsg->msg_controllen = msg->msg_controllen;
 
 	if (save_addr)
-		*save_addr = compat_ptr(msg.msg_name);
+		*save_addr = compat_ptr(msg->msg_name);
 
-	if (msg.msg_name && kmsg->msg_namelen) {
+	if (msg->msg_name && kmsg->msg_namelen) {
 		if (!save_addr) {
-			err = move_addr_to_kernel(compat_ptr(msg.msg_name),
+			err = move_addr_to_kernel(compat_ptr(msg->msg_name),
 						  kmsg->msg_namelen,
 						  kmsg->msg_name);
 			if (err < 0)
@@ -76,12 +71,10 @@ int __get_compat_msghdr(struct msghdr *kmsg,
 		kmsg->msg_namelen = 0;
 	}
 
-	if (msg.msg_iovlen > UIO_MAXIOV)
+	if (msg->msg_iovlen > UIO_MAXIOV)
 		return -EMSGSIZE;
 
 	kmsg->msg_iocb = NULL;
-	*ptr = msg.msg_iov;
-	*len = msg.msg_iovlen;
 	return 0;
 }
 
@@ -90,15 +83,17 @@ int get_compat_msghdr(struct msghdr *kmsg,
 		      struct sockaddr __user **save_addr,
 		      struct iovec **iov)
 {
-	compat_uptr_t ptr;
-	compat_size_t len;
+	struct compat_msghdr msg;
 	ssize_t err;
 
-	err = __get_compat_msghdr(kmsg, umsg, save_addr, &ptr, &len);
+	if (copy_from_user(&msg, umsg, sizeof(*umsg)))
+		return -EFAULT;
+
+	err = __get_compat_msghdr(kmsg, umsg, save_addr);
 	if (err)
 		return err;
 
-	err = import_iovec(save_addr ? READ : WRITE, compat_ptr(ptr), len,
+	err = import_iovec(save_addr ? READ : WRITE, compat_ptr(msg.msg_iov), msg.msg_iovlen,
 			   UIO_FASTIOV, iov, &kmsg->msg_iter);
 	return err < 0 ? err : 0;
 }
-- 
2.30.2

[PATCH v2 for-next 3/3] io_uring: support multishot in recvmsg

[Dylan Yudaken]

Similar to multishot recv, this will require provided buffers to be
used. However recvmsg is much more complex than recv as it has multiple
outputs. Specifically flags, name, and control messages.

Support this by introducing a new struct io_uring_recvmsg_out with 4
fields. namelen, controllen and flags match the similar out fields in
msghdr from standard recvmsg(2), payloadlen is the length of the payload
following the header.
This struct is placed at the start of the returned buffer. Based on what
the user specifies in struct msghdr, the next bytes of the buffer will be
name (the next msg_namelen bytes), and then control (the next
msg_controllen bytes). The payload will come at the end. The return value
in the CQE is the total used size of the provided buffer.

Signed-off-by: Dylan Yudaken <dylany@fb.com>
---
 include/uapi/linux/io_uring.h |   7 ++
 io_uring/net.c                | 178 ++++++++++++++++++++++++++++++----
 io_uring/net.h                |   6 ++
 3 files changed, 172 insertions(+), 19 deletions(-)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 499679134961..4c9b11e2e991 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -613,4 +613,11 @@ struct io_uring_file_index_range {
 	__u64	resv;
 };
 
+struct io_uring_recvmsg_out {
+	__u32 namelen;
+	__u32 controllen;
+	__u32 payloadlen;
+	__u32 flags;
+};
+
 #endif
diff --git a/io_uring/net.c b/io_uring/net.c
index 5bc3440a8290..56f734acced6 100644
--- a/io_uring/net.c
+++ b/io_uring/net.c
@@ -352,6 +352,11 @@ static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
 			sr->len = iomsg->fast_iov[0].iov_len;
 			iomsg->free_iov = NULL;
 		}
+
+		if (req->flags & REQ_F_APOLL_MULTISHOT) {
+			iomsg->namelen = msg.msg_namelen;
+			iomsg->controllen = msg.msg_controllen;
+		}
 	} else {
 		iomsg->free_iov = iomsg->fast_iov;
 		ret = __import_iovec(READ, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
@@ -399,6 +404,11 @@ static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
 			sr->len = clen;
 			iomsg->free_iov = NULL;
 		}
+
+		if (req->flags & REQ_F_APOLL_MULTISHOT) {
+			iomsg->namelen = msg.msg_namelen;
+			iomsg->controllen = msg.msg_controllen;
+		}
 	} else {
 		iomsg->free_iov = iomsg->fast_iov;
 		ret = __import_iovec(READ, (struct iovec __user *)uiov, msg.msg_iovlen,
@@ -455,8 +465,6 @@ int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	if (sr->msg_flags & MSG_ERRQUEUE)
 		req->flags |= REQ_F_CLEAR_POLLIN;
 	if (sr->flags & IORING_RECV_MULTISHOT) {
-		if (req->opcode == IORING_OP_RECVMSG)
-			return -EINVAL;
 		if (!(req->flags & REQ_F_BUFFER_SELECT))
 			return -EINVAL;
 		if (sr->msg_flags & MSG_WAITALL)
@@ -483,12 +491,15 @@ static inline void io_recv_prep_retry(struct io_kiocb *req)
 }
 
 /*
- * Finishes io_recv
+ * Finishes io_recv and io_recvmsg.
  *
  * Returns true if it is actually finished, or false if it should run
  * again (for multishot).
  */
-static inline bool io_recv_finish(struct io_kiocb *req, int *ret, unsigned int cflags)
+static inline bool io_recv_finish(struct io_kiocb *req,
+				  int *ret,
+				  unsigned int cflags,
+				  bool multishot_finished)
 {
 	if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
 		io_req_set_res(req, *ret, cflags);
@@ -496,7 +507,7 @@ static inline bool io_recv_finish(struct io_kiocb *req, int *ret, unsigned int c
 		return true;
 	}
 
-	if (*ret > 0) {
+	if (!multishot_finished) {
 		if (io_post_aux_cqe(req->ctx, req->cqe.user_data, *ret,
 				    cflags | IORING_CQE_F_MORE, false)) {
 			io_recv_prep_retry(req);
@@ -518,6 +529,104 @@ static inline bool io_recv_finish(struct io_kiocb *req, int *ret, unsigned int c
 	return true;
 }
 
+static int io_recvmsg_prep_multishot(
+	struct io_async_msghdr *kmsg,
+	struct io_sr_msg *sr,
+	void __user **buf,
+	size_t *len)
+{
+	unsigned long used = 0;
+
+	if (*len < sizeof(struct io_uring_recvmsg_out))
+		return -EFAULT;
+	used += sizeof(struct io_uring_recvmsg_out);
+
+	if (kmsg->namelen) {
+		if (kmsg->namelen + used > *len)
+			return -EFAULT;
+		used += kmsg->namelen;
+	}
+	if (kmsg->controllen) {
+		if (kmsg->controllen + used > *len)
+			return -EFAULT;
+		kmsg->msg.msg_control_user = (void *)((unsigned long)*buf + used);
+		kmsg->msg.msg_controllen = kmsg->controllen;
+		used += kmsg->controllen;
+	}
+	if (used >= UINT_MAX)
+		return -EOVERFLOW;
+
+	sr->buf = *buf; /* stash for later copy */
+	*buf = (void *)((unsigned long)*buf + used);
+	*len -= used;
+	kmsg->payloadlen = *len;
+	return 0;
+}
+
+struct io_recvmsg_multishot_hdr {
+	struct io_uring_recvmsg_out msg;
+	struct sockaddr_storage addr;
+} __packed;
+
+static int io_recvmsg_multishot(
+	struct socket *sock,
+	struct io_sr_msg *io,
+	struct io_async_msghdr *kmsg,
+	unsigned int flags,
+	bool *finished)
+{
+	int err;
+	int copy_len;
+	struct io_recvmsg_multishot_hdr hdr;
+
+	if (kmsg->namelen)
+		kmsg->msg.msg_name = &hdr.addr;
+	kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
+	kmsg->msg.msg_namelen = 0;
+
+	if (sock->file->f_flags & O_NONBLOCK)
+		flags |= MSG_DONTWAIT;
+
+	err = sock_recvmsg(sock, &kmsg->msg, flags);
+	*finished = err <= 0;
+	if (err < 0)
+		return err;
+
+	hdr.msg = (struct io_uring_recvmsg_out) {
+		.controllen = kmsg->controllen - kmsg->msg.msg_controllen,
+		.flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
+	};
+
+	hdr.msg.payloadlen = err;
+	if (err > kmsg->payloadlen)
+		err = kmsg->payloadlen;
+
+	copy_len = sizeof(struct io_uring_recvmsg_out);
+	if (kmsg->msg.msg_namelen > kmsg->namelen)
+		copy_len += kmsg->namelen;
+	else
+		copy_len += kmsg->msg.msg_namelen;
+
+	/*
+	 *      "fromlen shall refer to the value before truncation.."
+	 *                      1003.1g
+	 */
+	hdr.msg.namelen = kmsg->msg.msg_namelen;
+
+	/* ensure that there is no gap between hdr and sockaddr_storage */
+	BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
+		     sizeof(struct io_uring_recvmsg_out));
+	if (copy_to_user(io->buf, &hdr, copy_len)) {
+		*finished = true;
+		return -EFAULT;
+	}
+
+	return sizeof(struct io_uring_recvmsg_out) +
+		kmsg->namelen +
+		kmsg->controllen +
+		err;
+}
+
 int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
 {
 	struct io_sr_msg *sr = io_kiocb_to_cmd(req);
@@ -527,6 +636,7 @@ int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
 	unsigned flags;
 	int ret, min_ret = 0;
 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
+	bool multishot_finished = true;
 
 	sock = sock_from_file(req->file);
 	if (unlikely(!sock))
@@ -545,16 +655,29 @@ int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
 		return io_setup_async_msg(req, kmsg, issue_flags);
 
+retry_multishot:
 	if (io_do_buffer_select(req)) {
 		void __user *buf;
+		size_t len = sr->len;
 
-		buf = io_buffer_select(req, &sr->len, issue_flags);
+		buf = io_buffer_select(req, &len, issue_flags);
 		if (!buf)
 			return -ENOBUFS;
+
+		if (req->flags & REQ_F_APOLL_MULTISHOT) {
+			ret = io_recvmsg_prep_multishot(kmsg, sr,
+							&buf, &len);
+
+			if (ret) {
+				io_kbuf_recycle(req, issue_flags);
+				return ret;
+			}
+		}
+
 		kmsg->fast_iov[0].iov_base = buf;
-		kmsg->fast_iov[0].iov_len = sr->len;
+		kmsg->fast_iov[0].iov_len = len;
 		iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, 1,
-				sr->len);
+				len);
 	}
 
 	flags = sr->msg_flags;
@@ -564,10 +687,22 @@ int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
 		min_ret = iov_iter_count(&kmsg->msg.msg_iter);
 
 	kmsg->msg.msg_get_inq = 1;
-	ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg, kmsg->uaddr, flags);
+	if (req->flags & REQ_F_APOLL_MULTISHOT)
+		ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
+					   &multishot_finished);
+	else
+		ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg, kmsg->uaddr, flags);
+
 	if (ret < min_ret) {
-		if (ret == -EAGAIN && force_nonblock)
-			return io_setup_async_msg(req, kmsg, issue_flags);
+		if (ret == -EAGAIN && force_nonblock) {
+			ret = io_setup_async_msg(req, kmsg, issue_flags);
+			if (ret == -EAGAIN && (req->flags & IO_APOLL_MULTI_POLLED) ==
+					       IO_APOLL_MULTI_POLLED) {
+				io_kbuf_recycle(req, issue_flags);
+				return IOU_ISSUE_SKIP_COMPLETE;
+			}
+			return ret;
+		}
 		if (ret == -ERESTARTSYS)
 			ret = -EINTR;
 		if (ret > 0 && io_net_retry(sock, flags)) {
@@ -580,11 +715,6 @@ int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
 		req_set_fail(req);
 	}
 
-	/* fast path, check for non-NULL to avoid function call */
-	if (kmsg->free_iov)
-		kfree(kmsg->free_iov);
-	io_netmsg_recycle(req, issue_flags);
-	req->flags &= ~REQ_F_NEED_CLEANUP;
 	if (ret > 0)
 		ret += sr->done_io;
 	else if (sr->done_io)
@@ -596,8 +726,18 @@ int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
 	if (kmsg->msg.msg_inq)
 		cflags |= IORING_CQE_F_SOCK_NONEMPTY;
 
-	io_req_set_res(req, ret, cflags);
-	return IOU_OK;
+	if (!io_recv_finish(req, &ret, cflags, multishot_finished))
+		goto retry_multishot;
+
+	if (multishot_finished) {
+		io_netmsg_recycle(req, issue_flags);
+		/* fast path, check for non-NULL to avoid function call */
+		if (kmsg->free_iov)
+			kfree(kmsg->free_iov);
+		req->flags &= ~REQ_F_NEED_CLEANUP;
+	}
+
+	return ret;
 }
 
 int io_recv(struct io_kiocb *req, unsigned int issue_flags)
@@ -684,7 +824,7 @@ int io_recv(struct io_kiocb *req, unsigned int issue_flags)
 	if (msg.msg_inq)
 		cflags |= IORING_CQE_F_SOCK_NONEMPTY;
 
-	if (!io_recv_finish(req, &ret, cflags))
+	if (!io_recv_finish(req, &ret, cflags, ret <= 0))
 		goto retry_multishot;
 
 	return ret;
diff --git a/io_uring/net.h b/io_uring/net.h
index 178a6d8b76e0..db20ce9d6546 100644
--- a/io_uring/net.h
+++ b/io_uring/net.h
@@ -9,6 +9,12 @@
 struct io_async_msghdr {
 	union {
 		struct iovec		fast_iov[UIO_FASTIOV];
+		struct {
+			struct iovec	fast_iov_one;
+			__kernel_size_t	controllen;
+			int		namelen;
+			__kernel_size_t	payloadlen;
+		};
 		struct io_cache_entry	cache;
 	};
 	/* points to an allocated iov, if NULL we use fast_iov instead */
-- 
2.30.2

Re: [PATCH v2 for-next 3/3] io_uring: support multishot in recvmsg

[Jens Axboe]

On 7/13/22 2:23 AM, Dylan Yudaken wrote:
> diff --git a/io_uring/net.c b/io_uring/net.c
> index 5bc3440a8290..56f734acced6 100644
> --- a/io_uring/net.c
> +++ b/io_uring/net.c
> @@ -483,12 +491,15 @@ static inline void io_recv_prep_retry(struct io_kiocb *req)
>  }
>  
>  /*
> - * Finishes io_recv
> + * Finishes io_recv and io_recvmsg.
>   *
>   * Returns true if it is actually finished, or false if it should run
>   * again (for multishot).
>   */
> -static inline bool io_recv_finish(struct io_kiocb *req, int *ret, unsigned int cflags)
> +static inline bool io_recv_finish(struct io_kiocb *req,
> +				  int *ret,
> +				  unsigned int cflags,
> +				  bool multishot_finished)
>  {
>  	if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {

Minor nit, but this should look like:

static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
				  unsigned int cflags, bool mshot_finished)
> @@ -518,6 +529,104 @@ static inline bool io_recv_finish(struct io_kiocb *req, int *ret, unsigned int c
>  	return true;
>  }
>  
> +static int io_recvmsg_prep_multishot(
> +	struct io_async_msghdr *kmsg,
> +	struct io_sr_msg *sr,
> +	void __user **buf,
> +	size_t *len)
> +{

Ditto on the function formating.

> +	unsigned long used = 0;
> +
> +	if (*len < sizeof(struct io_uring_recvmsg_out))
> +		return -EFAULT;
> +	used += sizeof(struct io_uring_recvmsg_out);
> +
> +	if (kmsg->namelen) {
> +		if (kmsg->namelen + used > *len)
> +			return -EFAULT;
> +		used += kmsg->namelen;
> +	}
> +	if (kmsg->controllen) {
> +		if (kmsg->controllen + used > *len)
> +			return -EFAULT;
> +		kmsg->msg.msg_control_user = (void *)((unsigned long)*buf + used);
> +		kmsg->msg.msg_controllen = kmsg->controllen;
> +		used += kmsg->controllen;
> +	}
> +	if (used >= UINT_MAX)
> +		return -EOVERFLOW;
> +
> +	sr->buf = *buf; /* stash for later copy */
> +	*buf = (void *)((unsigned long)*buf + used);
> +	*len -= used;
> +	kmsg->payloadlen = *len;
> +	return 0;
> +}

Not sure if it's just me, but the *buf and casting is really hard to
read here. Can we make that any clearer? Maybe cast to an unsigned long
* at the top of change the buf argument to be that?

> +struct io_recvmsg_multishot_hdr {
> +	struct io_uring_recvmsg_out msg;
> +	struct sockaddr_storage addr;
> +} __packed;

This __packed shouldn't be necessary, and I'm always a bit wary of
adding it on kernel structures as if it's really needed, then we're most
likely doing something wrong (and things will run slower, notably on
some archs). Looks like you have a BUILD_BUG_ON() for this too, so we'd
catch any potential issues here upfront.

> +static int io_recvmsg_multishot(
> +	struct socket *sock,
> +	struct io_sr_msg *io,
> +	struct io_async_msghdr *kmsg,
> +	unsigned int flags,
> +	bool *finished)
> +{
> +	int err;
> +	int copy_len;
> +	struct io_recvmsg_multishot_hdr hdr;
> +
> +	if (kmsg->namelen)
> +		kmsg->msg.msg_name = &hdr.addr;
> +	kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
> +	kmsg->msg.msg_namelen = 0;
> +
> +	if (sock->file->f_flags & O_NONBLOCK)
> +		flags |= MSG_DONTWAIT;
> +
> +	err = sock_recvmsg(sock, &kmsg->msg, flags);
> +	*finished = err <= 0;
> +	if (err < 0)
> +		return err;
> +
> +	hdr.msg = (struct io_uring_recvmsg_out) {
> +		.controllen = kmsg->controllen - kmsg->msg.msg_controllen,
> +		.flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
> +	};
> +
> +	hdr.msg.payloadlen = err;
> +	if (err > kmsg->payloadlen)
> +		err = kmsg->payloadlen;
> +
> +	copy_len = sizeof(struct io_uring_recvmsg_out);
> +	if (kmsg->msg.msg_namelen > kmsg->namelen)
> +		copy_len += kmsg->namelen;
> +	else
> +		copy_len += kmsg->msg.msg_namelen;
> +
> +	/*
> +	 *      "fromlen shall refer to the value before truncation.."
> +	 *                      1003.1g
> +	 */
> +	hdr.msg.namelen = kmsg->msg.msg_namelen;
> +
> +	/* ensure that there is no gap between hdr and sockaddr_storage */
> +	BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
> +		     sizeof(struct io_uring_recvmsg_out));
> +	if (copy_to_user(io->buf, &hdr, copy_len)) {
> +		*finished = true;
> +		return -EFAULT;
> +	}
> +
> +	return sizeof(struct io_uring_recvmsg_out) +
> +		kmsg->namelen +
> +		kmsg->controllen +
> +		err;
> +}

	return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
			kmsg->controllen + err;

would be closer to the kernel style.

In general I'm not a big fan of the bool pointer 'finished'. But I also
don't have a good suggestion on how to make it cleaner, so... Would be
nice if we could just have an error return (< 0), and then return >= 0
in two variants for MSHOT_OK and MSHOT_TERMINATE or something.

> @@ -527,6 +636,7 @@ int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
>  	unsigned flags;
>  	int ret, min_ret = 0;
>  	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
> +	bool multishot_finished = true;
>  
>  	sock = sock_from_file(req->file);
>  	if (unlikely(!sock))
> @@ -545,16 +655,29 @@ int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
>  	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
>  		return io_setup_async_msg(req, kmsg, issue_flags);
>  
> +retry_multishot:
>  	if (io_do_buffer_select(req)) {
>  		void __user *buf;
> +		size_t len = sr->len;
>  
> -		buf = io_buffer_select(req, &sr->len, issue_flags);
> +		buf = io_buffer_select(req, &len, issue_flags);
>  		if (!buf)
>  			return -ENOBUFS;
> +
> +		if (req->flags & REQ_F_APOLL_MULTISHOT) {
> +			ret = io_recvmsg_prep_multishot(kmsg, sr,
> +							&buf, &len);
> +
			ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);

Apart from these nits, looks pretty good to me.

-- 
Jens Axboe

Re: [PATCH v2 for-next 3/3] io_uring: support multishot in recvmsg

[Dylan Yudaken]

On Wed, 2022-07-13 at 06:48 -0600, Jens Axboe wrote:
> On 7/13/22 2:23 AM, Dylan Yudaken wrote:
> > diff --git a/io_uring/net.c b/io_uring/net.c
> > index 5bc3440a8290..56f734acced6 100644
> > --- a/io_uring/net.c
> > +++ b/io_uring/net.c
> > @@ -483,12 +491,15 @@ static inline void io_recv_prep_retry(struct
> > io_kiocb *req)
> >  }
> >  
> >  /*
> > - * Finishes io_recv
> > + * Finishes io_recv and io_recvmsg.
> >   *
> >   * Returns true if it is actually finished, or false if it should
> > run
> >   * again (for multishot).
> >   */
> > -static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
> > unsigned int cflags)
> > +static inline bool io_recv_finish(struct io_kiocb *req,
> > +                                 int *ret,
> > +                                 unsigned int cflags,
> > +                                 bool multishot_finished)
> >  {
> >         if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
> 
> Minor nit, but this should look like:
> 
> static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
>                                   unsigned int cflags, bool
> mshot_finished)
> > @@ -518,6 +529,104 @@ static inline bool io_recv_finish(struct
> > io_kiocb *req, int *ret, unsigned int c
> >         return true;
> >  }
> >  
> > +static int io_recvmsg_prep_multishot(
> > +       struct io_async_msghdr *kmsg,
> > +       struct io_sr_msg *sr,
> > +       void __user **buf,
> > +       size_t *len)
> > +{
> 
> Ditto on the function formating.
> 

will update both in a v3

> > +       unsigned long used = 0;
> > +
> > +       if (*len < sizeof(struct io_uring_recvmsg_out))
> > +               return -EFAULT;
> > +       used += sizeof(struct io_uring_recvmsg_out);
> > +
> > +       if (kmsg->namelen) {
> > +               if (kmsg->namelen + used > *len)
> > +                       return -EFAULT;
> > +               used += kmsg->namelen;
> > +       }
> > +       if (kmsg->controllen) {
> > +               if (kmsg->controllen + used > *len)
> > +                       return -EFAULT;
> > +               kmsg->msg.msg_control_user = (void *)((unsigned
> > long)*buf + used);
> > +               kmsg->msg.msg_controllen = kmsg->controllen;
> > +               used += kmsg->controllen;
> > +       }
> > +       if (used >= UINT_MAX)
> > +               return -EOVERFLOW;
> > +
> > +       sr->buf = *buf; /* stash for later copy */
> > +       *buf = (void *)((unsigned long)*buf + used);
> > +       *len -= used;
> > +       kmsg->payloadlen = *len;
> > +       return 0;
> > +}
> 
> Not sure if it's just me, but the *buf and casting is really hard to
> read here. Can we make that any clearer? Maybe cast to an unsigned
> long
> * at the top of change the buf argument to be that?

I think that should be doable. Not an (unsigned long *) though as it is
all for incrementing the pointer address, but probably an (unsigned
char *).

> 
> > +struct io_recvmsg_multishot_hdr {
> > +       struct io_uring_recvmsg_out msg;
> > +       struct sockaddr_storage addr;
> > +} __packed;
> 
> This __packed shouldn't be necessary, and I'm always a bit wary of
> adding it on kernel structures as if it's really needed, then we're
> most
> likely doing something wrong (and things will run slower, notably on
> some archs). Looks like you have a BUILD_BUG_ON() for this too, so
> we'd
> catch any potential issues here upfront.

Makes sense.

> 
> > +static int io_recvmsg_multishot(
> > +       struct socket *sock,
> > +       struct io_sr_msg *io,
> > +       struct io_async_msghdr *kmsg,
> > +       unsigned int flags,
> > +       bool *finished)
> > +{
> > +       int err;
> > +       int copy_len;
> > +       struct io_recvmsg_multishot_hdr hdr;
> > +
> > +       if (kmsg->namelen)
> > +               kmsg->msg.msg_name = &hdr.addr;
> > +       kmsg->msg.msg_flags = flags &
> > (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
> > +       kmsg->msg.msg_namelen = 0;
> > +
> > +       if (sock->file->f_flags & O_NONBLOCK)
> > +               flags |= MSG_DONTWAIT;
> > +
> > +       err = sock_recvmsg(sock, &kmsg->msg, flags);
> > +       *finished = err <= 0;
> > +       if (err < 0)
> > +               return err;
> > +
> > +       hdr.msg = (struct io_uring_recvmsg_out) {
> > +               .controllen = kmsg->controllen - kmsg-
> > >msg.msg_controllen,
> > +               .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
> > +       };
> > +
> > +       hdr.msg.payloadlen = err;
> > +       if (err > kmsg->payloadlen)
> > +               err = kmsg->payloadlen;
> > +
> > +       copy_len = sizeof(struct io_uring_recvmsg_out);
> > +       if (kmsg->msg.msg_namelen > kmsg->namelen)
> > +               copy_len += kmsg->namelen;
> > +       else
> > +               copy_len += kmsg->msg.msg_namelen;
> > +
> > +       /*
> > +        *      "fromlen shall refer to the value before
> > truncation.."
> > +        *                      1003.1g
> > +        */
> > +       hdr.msg.namelen = kmsg->msg.msg_namelen;
> > +
> > +       /* ensure that there is no gap between hdr and
> > sockaddr_storage */
> > +       BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr,
> > addr) !=
> > +                    sizeof(struct io_uring_recvmsg_out));
> > +       if (copy_to_user(io->buf, &hdr, copy_len)) {
> > +               *finished = true;
> > +               return -EFAULT;
> > +       }
> > +
> > +       return sizeof(struct io_uring_recvmsg_out) +
> > +               kmsg->namelen +
> > +               kmsg->controllen +
> > +               err;
> > +}
> 
>         return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
>                         kmsg->controllen + err;
> 
> would be closer to the kernel style.
> 
> In general I'm not a big fan of the bool pointer 'finished'. But I
> also
> don't have a good suggestion on how to make it cleaner, so... Would
> be
> nice if we could just have an error return (< 0), and then return >=
> 0
> in two variants for MSHOT_OK and MSHOT_TERMINATE or something.

I will try for v3, but the issue is that we have 2 things to return:
how many bytes are copied to the output buffer, and if multishot is
finished, and these have to be kept separately as there is no way to
derive one from the other. 

> 
> > @@ -527,6 +636,7 @@ int io_recvmsg(struct io_kiocb *req, unsigned
> > int issue_flags)
> >         unsigned flags;
> >         int ret, min_ret = 0;
> >         bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
> > +       bool multishot_finished = true;
> >  
> >         sock = sock_from_file(req->file);
> >         if (unlikely(!sock))
> > @@ -545,16 +655,29 @@ int io_recvmsg(struct io_kiocb *req, unsigned
> > int issue_flags)
> >             (sr->flags & IORING_RECVSEND_POLL_FIRST))
> >                 return io_setup_async_msg(req, kmsg, issue_flags);
> >  
> > +retry_multishot:
> >         if (io_do_buffer_select(req)) {
> >                 void __user *buf;
> > +               size_t len = sr->len;
> >  
> > -               buf = io_buffer_select(req, &sr->len, issue_flags);
> > +               buf = io_buffer_select(req, &len, issue_flags);
> >                 if (!buf)
> >                         return -ENOBUFS;
> > +
> > +               if (req->flags & REQ_F_APOLL_MULTISHOT) {
> > +                       ret = io_recvmsg_prep_multishot(kmsg, sr,
> > +                                                       &buf,
> > &len);
> > +
>                         ret = io_recvmsg_prep_multishot(kmsg, sr,
> &buf, &len);
> 
> Apart from these nits, looks pretty good to me.
> 

RE: [PATCH v2 for-next 3/3] io_uring: support multishot in recvmsg

[David Laight]

From: Dylan Yudaken
> Sent: 13 July 2022 09:23
> 
> Similar to multishot recv, this will require provided buffers to be
> used. However recvmsg is much more complex than recv as it has multiple
> outputs. Specifically flags, name, and control messages.
...

Why is this any different from adding several 'recvmsg' requests
into the request ring?

IIUC the requests are all processed sequentially by a single thread.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

Re: [PATCH v2 for-next 3/3] io_uring: support multishot in recvmsg

[Dylan Yudaken]

On Thu, 2022-07-14 at 07:46 +0000, David Laight wrote:
> From: Dylan Yudaken
> > Sent: 13 July 2022 09:23
> > 
> > Similar to multishot recv, this will require provided buffers to be
> > used. However recvmsg is much more complex than recv as it has
> > multiple
> > outputs. Specifically flags, name, and control messages.
> ...
> 
> Why is this any different from adding several 'recvmsg' requests
> into the request ring?
> 
> IIUC the requests are all processed sequentially by a single thread.
> 

In practice it is mostly the same for UDP (apart from userspace having
to predict how many recvmsg requests it should add).  That approach
would also have a slight ordering race, which probably is not a problem
for protocols that don't guarantee ordering, but might be annoying.

For stream like TCP it would not be possible due to the ordering issues
(completions of separate SQEs have no guaranteed ordering). You would
have to link the SQE's to preserve ordering, but then you would still
have to wait for a batch to complete before submitting a new batch.

Apart from those practical reasons, performance wise it is better for a
few reasons
 * userspace doesn't have to submit new requests
 * the io_uring knows it's multishot, so doesn't teardown the poll each
time
 * io_uring can allocate the request and associated async data only
once and reuse it

Regards,
Dylan

>         David
> 
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes,
> MK1 1PT, UK
> Registration No: 1397386 (Wales)
>