1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
3 #include <linux/errno.h>
4 #include <linux/file.h>
5 #include <linux/slab.h>
6 #include <linux/nospec.h>
7 #include <linux/io_uring.h>
9 #include <uapi/linux/io_uring.h>
13 #include "filetable.h"
26 static int io_msg_ring_data(struct io_kiocb *req)
28 struct io_ring_ctx *target_ctx = req->file->private_data;
29 struct io_msg *msg = io_kiocb_to_cmd(req, struct io_msg);
31 if (msg->src_fd || msg->dst_fd || msg->flags)
34 if (io_post_aux_cqe(target_ctx, msg->user_data, msg->len, 0, true))
40 static void io_double_unlock_ctx(struct io_ring_ctx *ctx,
41 struct io_ring_ctx *octx,
42 unsigned int issue_flags)
44 if (issue_flags & IO_URING_F_UNLOCKED)
45 mutex_unlock(&ctx->uring_lock);
46 mutex_unlock(&octx->uring_lock);
49 static int io_double_lock_ctx(struct io_ring_ctx *ctx,
50 struct io_ring_ctx *octx,
51 unsigned int issue_flags)
54 * To ensure proper ordering between the two ctxs, we can only
55 * attempt a trylock on the target. If that fails and we already have
56 * the source ctx lock, punt to io-wq.
58 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
59 if (!mutex_trylock(&octx->uring_lock))
64 /* Always grab smallest value ctx first. We know ctx != octx. */
66 mutex_lock(&ctx->uring_lock);
67 mutex_lock(&octx->uring_lock);
69 mutex_lock(&octx->uring_lock);
70 mutex_lock(&ctx->uring_lock);
76 static int io_msg_send_fd(struct io_kiocb *req, unsigned int issue_flags)
78 struct io_ring_ctx *target_ctx = req->file->private_data;
79 struct io_msg *msg = io_kiocb_to_cmd(req, struct io_msg);
80 struct io_ring_ctx *ctx = req->ctx;
81 unsigned long file_ptr;
82 struct file *src_file;
85 if (target_ctx == ctx)
88 ret = io_double_lock_ctx(ctx, target_ctx, issue_flags);
93 if (unlikely(msg->src_fd >= ctx->nr_user_files))
96 msg->src_fd = array_index_nospec(msg->src_fd, ctx->nr_user_files);
97 file_ptr = io_fixed_file_slot(&ctx->file_table, msg->src_fd)->file_ptr;
98 src_file = (struct file *) (file_ptr & FFS_MASK);
101 ret = __io_fixed_fd_install(target_ctx, src_file, msg->dst_fd);
107 if (msg->flags & IORING_MSG_RING_CQE_SKIP)
111 * If this fails, the target still received the file descriptor but
112 * wasn't notified of the fact. This means that if this request
113 * completes with -EOVERFLOW, then the sender must ensure that a
114 * later IORING_OP_MSG_RING delivers the message.
116 if (!io_post_aux_cqe(target_ctx, msg->user_data, msg->len, 0, true))
119 io_double_unlock_ctx(ctx, target_ctx, issue_flags);
123 int io_msg_ring_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
125 struct io_msg *msg = io_kiocb_to_cmd(req, struct io_msg);
127 if (unlikely(sqe->buf_index || sqe->personality))
130 msg->user_data = READ_ONCE(sqe->off);
131 msg->len = READ_ONCE(sqe->len);
132 msg->cmd = READ_ONCE(sqe->addr);
133 msg->src_fd = READ_ONCE(sqe->addr3);
134 msg->dst_fd = READ_ONCE(sqe->file_index);
135 msg->flags = READ_ONCE(sqe->msg_ring_flags);
136 if (msg->flags & ~IORING_MSG_RING_CQE_SKIP)
142 int io_msg_ring(struct io_kiocb *req, unsigned int issue_flags)
144 struct io_msg *msg = io_kiocb_to_cmd(req, struct io_msg);
148 if (!io_is_uring_fops(req->file))
152 case IORING_MSG_DATA:
153 ret = io_msg_ring_data(req);
155 case IORING_MSG_SEND_FD:
156 ret = io_msg_send_fd(req, issue_flags);
166 io_req_set_res(req, ret, 0);
167 /* put file to avoid an attempt to IOPOLL the req */
168 io_put_file(req->file);