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

From: Usama Arif
Date: Mon Jun 13 2022 - 16:37:21 EST


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@xxxxxxxxxxxxx>
---
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