[RFC 0/3] io_uring: add support for IORING_OP_MSGSND/IORING_OP_MSGRCV

[RFC 0/3] io_uring: add support for IORING_OP_MSGSND/IORING_OP_MSGRCV

[Usama Arif]

The first patch just refractors the do_msg function so that its different
parts can be used in io_uring.

The last 2 patches add the support for the async msgsnd/msgrcv.
For msgsnd the pointer to the loaded msgbuf is saved in 
io_setup_async_msg_msg for the case when the call is punted to async
context.
For msgrcv, the pointer to user-defined buffer is already stored in the
prep function.

The correspoding prep helpers and a simple testcase for these in liburing
are present at https://github.com/uarif1/liburing/commit/17f6685ca398265b54a07ce4246ee344b5390aad.

These patches are currently missing compat support.
If the patches and approach look acceptable to start review,
I can send the next series with compat support included and also add the
man pages in the above liburing patch.

Thanks!

Usama Arif (3):
  ipc/msg: split do_msgsnd into functions
  io_uring: add support for IORING_OP_MSGSND
  io_uring: add support for IORING_OP_MSGRCV

 fs/io_uring.c                 | 152 ++++++++++++++++++++++++++++++++++
 include/linux/msg.h           |  11 +++
 include/uapi/linux/io_uring.h |   2 +
 ipc/msg.c                     |  52 +++++++++---
 4 files changed, 205 insertions(+), 12 deletions(-)

-- 
2.25.1

[RFC 1/3] ipc/msg: split do_msgsnd into functions

[Usama Arif]

do_msgsnd is split into check_and_load_msgsnd and __do_msgsnd.
This does not introduce any change in the functions' operation.
Its only needed for async msgsnd in io_uring which will be added
in a later patch.

Functions used for msgsnd and msgrcv are also declared in the
header file for use in io_uring patches later.

Signed-off-by: Usama Arif <usama.arif@bytedance.com>
---
 include/linux/msg.h | 11 ++++++++++
 ipc/msg.c           | 52 ++++++++++++++++++++++++++++++++++-----------
 2 files changed, 51 insertions(+), 12 deletions(-)

diff --git a/include/linux/msg.h b/include/linux/msg.h
index 9a972a296b95..36e3116fed86 100644
--- a/include/linux/msg.h
+++ b/include/linux/msg.h
@@ -15,4 +15,15 @@ struct msg_msg {
 	/* the actual message follows immediately */
 };
 
+long check_and_load_msgsnd(int msqid, long mtype, void __user *mtext,
+			   struct msg_msg **msg, size_t msgsz);
+
+void free_msg(struct msg_msg *msg);
+
+long __do_msgsnd(int msqid, long mtype, struct msg_msg **msg,
+		 size_t msgsz, int msgflg);
+
+long ksys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
+		 long msgtyp, int msgflg);
+
 #endif /* _LINUX_MSG_H */
diff --git a/ipc/msg.c b/ipc/msg.c
index a0d05775af2c..0682204a684e 100644
--- a/ipc/msg.c
+++ b/ipc/msg.c
@@ -839,14 +839,11 @@ static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg,
 	return 0;
 }
 
-static long do_msgsnd(int msqid, long mtype, void __user *mtext,
-		size_t msgsz, int msgflg)
+long check_and_load_msgsnd(int msqid, long mtype, void __user *mtext,
+			   struct msg_msg **msg, size_t msgsz)
 {
-	struct msg_queue *msq;
-	struct msg_msg *msg;
-	int err;
 	struct ipc_namespace *ns;
-	DEFINE_WAKE_Q(wake_q);
+	struct msg_msg *tmp;
 
 	ns = current->nsproxy->ipc_ns;
 
@@ -855,12 +852,45 @@ static long do_msgsnd(int msqid, long mtype, void __user *mtext,
 	if (mtype < 1)
 		return -EINVAL;
 
-	msg = load_msg(mtext, msgsz);
+	tmp = load_msg(mtext, msgsz);
 	if (IS_ERR(msg))
 		return PTR_ERR(msg);
 
-	msg->m_type = mtype;
-	msg->m_ts = msgsz;
+	tmp->m_type = mtype;
+	tmp->m_ts = msgsz;
+
+	*msg = tmp;
+	return 0;
+}
+
+static long do_msgsnd(int msqid, long mtype, void __user *mtext,
+		      size_t msgsz, int msgflg)
+{
+	struct msg_msg *msg;
+	int err;
+
+	err = check_and_load_msgsnd(msqid, mtype, mtext, &msg, msgsz);
+	if (err)
+		return err;
+
+	err = __do_msgsnd(msqid, mtype, &msg, msgsz, msgflg);
+	if (msg != NULL)
+		free_msg(msg);
+
+	return err;
+}
+
+long __do_msgsnd(int msqid, long mtype, struct msg_msg **_msg,
+		 size_t msgsz, int msgflg)
+{
+	struct msg_queue *msq;
+	struct msg_msg *msg;
+	int err;
+	struct ipc_namespace *ns;
+	DEFINE_WAKE_Q(wake_q);
+
+	msg = *_msg;
+	ns = current->nsproxy->ipc_ns;
 
 	rcu_read_lock();
 	msq = msq_obtain_object_check(ns, msqid);
@@ -940,15 +970,13 @@ static long do_msgsnd(int msqid, long mtype, void __user *mtext,
 	}
 
 	err = 0;
-	msg = NULL;
+	*_msg = NULL;
 
 out_unlock0:
 	ipc_unlock_object(&msq->q_perm);
 	wake_up_q(&wake_q);
 out_unlock1:
 	rcu_read_unlock();
-	if (msg != NULL)
-		free_msg(msg);
 	return err;
 }
 
-- 
2.25.1

[RFC 2/3] io_uring: add support for IORING_OP_MSGSND

[Usama Arif]

This adds support for async msgsnd through io_uring.

The message is stored in msg pointer in io_msgsnd and is saved
in io_setup_async_msgq if we need to punt to async context.

Signed-off-by: Usama Arif <usama.arif@bytedance.com>
---
 fs/io_uring.c                 | 107 ++++++++++++++++++++++++++++++++++
 include/uapi/linux/io_uring.h |   1 +
 2 files changed, 108 insertions(+)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 3aab4182fd89..5949fcadb380 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -81,6 +81,7 @@
 #include <linux/audit.h>
 #include <linux/security.h>
 #include <linux/xattr.h>
+#include <linux/msg.h>
 
 #define CREATE_TRACE_POINTS
 #include <trace/events/io_uring.h>
@@ -669,6 +670,15 @@ struct io_sr_msg {
 	unsigned int			flags;
 };
 
+struct io_msg_sr {
+	struct file			*file;
+	int				msq_id;
+	struct msgbuf __user		*msg_p;
+	size_t				msg_sz;
+	long				msg_type;
+	int				msg_flags;
+};
+
 struct io_open {
 	struct file			*file;
 	int				dfd;
@@ -803,6 +813,10 @@ struct io_async_msghdr {
 	struct sockaddr_storage		addr;
 };
 
+struct io_async_msg_msg {
+	struct msg_msg			*msg;
+};
+
 struct io_rw_state {
 	struct iov_iter			iter;
 	struct iov_iter_state		iter_state;
@@ -996,6 +1010,7 @@ struct io_kiocb {
 		struct io_socket	sock;
 		struct io_nop		nop;
 		struct io_uring_cmd	uring_cmd;
+		struct io_msg_sr	msg_sr;
 	};
 
 	u8				opcode;
@@ -1199,6 +1214,9 @@ static const struct io_op_def io_op_defs[] = {
 		.needs_async_setup	= 1,
 		.async_size		= sizeof(struct io_async_msghdr),
 	},
+	[IORING_OP_MSGSND] = {
+		.async_size		= sizeof(struct io_async_msg_msg),
+	},
 	[IORING_OP_TIMEOUT] = {
 		.audit_skip		= 1,
 		.async_size		= sizeof(struct io_timeout_data),
@@ -1404,6 +1422,8 @@ const char *io_uring_get_opcode(u8 opcode)
 		return "SENDMSG";
 	case IORING_OP_RECVMSG:
 		return "RECVMSG";
+	case IORING_OP_MSGSND:
+		return "MSGSND";
 	case IORING_OP_TIMEOUT:
 		return "TIMEOUT";
 	case IORING_OP_TIMEOUT_REMOVE:
@@ -6180,6 +6200,81 @@ static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	return 0;
 }
 
+static int io_setup_async_msg_msg(struct io_kiocb *req, struct msg_msg *msg)
+{
+	struct io_async_msg_msg *async_msg_msg = req->async_data;
+
+	if (async_msg_msg)
+		return -EAGAIN;
+	if (io_alloc_async_data(req))
+		return -ENOMEM;
+	async_msg_msg = req->async_data;
+
+	req->flags |= REQ_F_NEED_CLEANUP;
+	async_msg_msg->msg = msg;
+
+	return -EAGAIN;
+}
+
+static int io_msgsnd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_msg_sr *msg_sr = &req->msg_sr;
+	struct msgbuf __user	*msg_p;
+	long mtype;
+
+	if (unlikely(sqe->addr2 || sqe->file_index))
+		return -EINVAL;
+
+	msg_sr->msq_id = READ_ONCE(sqe->fd);
+	msg_p = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	msg_sr->msg_p = msg_p;
+	if (get_user(mtype, &msg_p->mtype))
+		return -EFAULT;
+	msg_sr->msg_type = mtype;
+	msg_sr->msg_sz = READ_ONCE(sqe->len);
+	msg_sr->msg_flags = READ_ONCE(sqe->msg_flags);
+	if (msg_sr->msg_flags & IPC_NOWAIT)
+		req->flags |= REQ_F_NOWAIT;
+
+	return 0;
+}
+
+static int io_msgsnd(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_async_msg_msg *async_msg_msg;
+	struct io_msg_sr *msg_sr = &req->msg_sr;
+	int ret;
+	int flags;
+	struct msg_msg *msg;
+	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
+
+	if (req_has_async_data(req)) {
+		async_msg_msg = req->async_data;
+		msg = async_msg_msg->msg;
+	} else {
+		ret = check_and_load_msgsnd(msg_sr->msq_id, msg_sr->msg_type,
+					    msg_sr->msg_p->mtext,
+					    &msg, msg_sr->msg_sz);
+		if (ret)
+			return ret;
+	}
+
+	if (force_nonblock)
+		flags = msg_sr->msg_flags | IPC_NOWAIT;
+
+	ret = __do_msgsnd(msg_sr->msq_id, msg_sr->msg_type, &msg,
+					msg_sr->msg_sz, flags);
+
+	if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
+		return io_setup_async_msg_msg(req, msg);
+
+	if (msg != NULL)
+		free_msg(msg);
+	req->flags &= ~REQ_F_NEED_CLEANUP;
+
+	io_req_complete(req, ret);
+	return ret;
+}
 static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
 {
 	struct io_async_msghdr iomsg, *kmsg;
@@ -8192,6 +8287,8 @@ static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 		return io_socket_prep(req, sqe);
 	case IORING_OP_URING_CMD:
 		return io_uring_cmd_prep(req, sqe);
+	case IORING_OP_MSGSND:
+		return io_msgsnd_prep(req, sqe);
 	}
 
 	printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
@@ -8316,6 +8413,13 @@ static void io_clean_op(struct io_kiocb *req)
 			kfree(io->free_iov);
 			break;
 			}
+		case IORING_OP_MSGSND: {
+			struct io_async_msg_msg *io = req->async_data;
+
+			if (io->msg != NULL)
+				free_msg(io->msg);
+			break;
+			}
 		case IORING_OP_OPENAT:
 		case IORING_OP_OPENAT2:
 			if (req->open.filename)
@@ -8529,6 +8633,9 @@ static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
 	case IORING_OP_URING_CMD:
 		ret = io_uring_cmd(req, issue_flags);
 		break;
+	case IORING_OP_MSGSND:
+		ret = io_msgsnd(req, issue_flags);
+		break;
 	default:
 		ret = -EINVAL;
 		break;
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 776e0278f9dd..fa29bd96207d 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -190,6 +190,7 @@ enum io_uring_op {
 	IORING_OP_GETXATTR,
 	IORING_OP_SOCKET,
 	IORING_OP_URING_CMD,
+	IORING_OP_MSGSND,
 
 	/* this goes last, obviously */
 	IORING_OP_LAST,
-- 
2.25.1

[RFC 3/3] io_uring: add support for IORING_OP_MSGRCV

[Usama Arif]

This adds support for async msgrcv through io_uring.

All the information needed after punt to async context
is already stored in the io_msgrcv_prep call.

Signed-off-by: Usama Arif <usama.arif@bytedance.com>
---
 fs/io_uring.c                 | 45 +++++++++++++++++++++++++++++++++++
 include/uapi/linux/io_uring.h |  1 +
 2 files changed, 46 insertions(+)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 5949fcadb380..124914d8ee50 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1217,6 +1217,8 @@ static const struct io_op_def io_op_defs[] = {
 	[IORING_OP_MSGSND] = {
 		.async_size		= sizeof(struct io_async_msg_msg),
 	},
+	[IORING_OP_MSGRCV] = {
+	},
 	[IORING_OP_TIMEOUT] = {
 		.audit_skip		= 1,
 		.async_size		= sizeof(struct io_timeout_data),
@@ -1424,6 +1426,8 @@ const char *io_uring_get_opcode(u8 opcode)
 		return "RECVMSG";
 	case IORING_OP_MSGSND:
 		return "MSGSND";
+	case IORING_OP_MSGRCV:
+		return "MSGRCV";
 	case IORING_OP_TIMEOUT:
 		return "TIMEOUT";
 	case IORING_OP_TIMEOUT_REMOVE:
@@ -6275,6 +6279,42 @@ static int io_msgsnd(struct io_kiocb *req, unsigned int issue_flags)
 	io_req_complete(req, ret);
 	return ret;
 }
+
+static int io_msgrcv_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_msg_sr *msg_sr = &req->msg_sr;
+
+	if (unlikely(sqe->file_index))
+		return -EINVAL;
+
+	msg_sr->msq_id = READ_ONCE(sqe->fd);
+	msg_sr->msg_p = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	msg_sr->msg_type = READ_ONCE(sqe->off);
+	msg_sr->msg_sz = READ_ONCE(sqe->len);
+	msg_sr->msg_flags = READ_ONCE(sqe->msg_flags);
+	return 0;
+}
+
+static int io_msgrcv(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_msg_sr *msg_sr = &req->msg_sr;
+	int ret;
+	int flags;
+	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
+
+	if (force_nonblock)
+		flags = msg_sr->msg_flags | IPC_NOWAIT;
+
+	ret = ksys_msgrcv(msg_sr->msq_id, msg_sr->msg_p, msg_sr->msg_sz,
+			  msg_sr->msg_type, flags);
+
+	if (ret == -ENOMSG)
+		return -EAGAIN;
+
+	io_req_complete(req, ret);
+	return 0;
+}
+
 static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
 {
 	struct io_async_msghdr iomsg, *kmsg;
@@ -8289,6 +8329,8 @@ static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 		return io_uring_cmd_prep(req, sqe);
 	case IORING_OP_MSGSND:
 		return io_msgsnd_prep(req, sqe);
+	case IORING_OP_MSGRCV:
+		return io_msgrcv_prep(req, sqe);
 	}
 
 	printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
@@ -8636,6 +8678,9 @@ static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
 	case IORING_OP_MSGSND:
 		ret = io_msgsnd(req, issue_flags);
 		break;
+	case IORING_OP_MSGRCV:
+		ret = io_msgrcv(req, issue_flags);
+		break;
 	default:
 		ret = -EINVAL;
 		break;
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index fa29bd96207d..b5dcaac30d9d 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -191,6 +191,7 @@ enum io_uring_op {
 	IORING_OP_SOCKET,
 	IORING_OP_URING_CMD,
 	IORING_OP_MSGSND,
+	IORING_OP_MSGRCV,
 
 	/* this goes last, obviously */
 	IORING_OP_LAST,
-- 
2.25.1