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>
7 #include <linux/compat.h>
8 #include <net/compat.h>
9 #include <linux/io_uring.h>
11 #include <uapi/linux/io_uring.h>
15 #include "alloc_cache.h"
20 #if defined(CONFIG_NET)
28 struct sockaddr __user *addr;
47 struct sockaddr __user *addr;
50 bool seen_econnaborted;
56 struct compat_msghdr __user *umsg_compat;
57 struct user_msghdr __user *umsg;
63 unsigned nr_multishot_loops;
65 /* initialised and used only by !msg send variants */
69 void __user *msg_control;
70 /* used only for send zerocopy */
71 struct io_kiocb *notif;
75 * Number of times we'll try and do receives if there's more data. If we
76 * exceed this limit, then add us to the back of the queue and retry from
77 * there. This helps fairness between flooding clients.
79 #define MULTISHOT_MAX_RETRY 32
81 int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
83 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
85 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
86 sqe->buf_index || sqe->splice_fd_in))
89 shutdown->how = READ_ONCE(sqe->len);
90 req->flags |= REQ_F_FORCE_ASYNC;
94 int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
96 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
100 WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
102 sock = sock_from_file(req->file);
106 ret = __sys_shutdown_sock(sock, shutdown->how);
107 io_req_set_res(req, ret, 0);
111 static bool io_net_retry(struct socket *sock, int flags)
113 if (!(flags & MSG_WAITALL))
115 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
118 static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
120 struct io_async_msghdr *hdr = req->async_data;
122 if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
125 /* Let normal cleanup path reap it if we fail adding to the cache */
126 if (io_alloc_cache_put(&req->ctx->netmsg_cache, &hdr->cache)) {
127 req->async_data = NULL;
128 req->flags &= ~REQ_F_ASYNC_DATA;
132 static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
133 unsigned int issue_flags)
135 struct io_ring_ctx *ctx = req->ctx;
136 struct io_cache_entry *entry;
137 struct io_async_msghdr *hdr;
139 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
140 entry = io_alloc_cache_get(&ctx->netmsg_cache);
142 hdr = container_of(entry, struct io_async_msghdr, cache);
143 hdr->free_iov = NULL;
144 req->flags |= REQ_F_ASYNC_DATA;
145 req->async_data = hdr;
150 if (!io_alloc_async_data(req)) {
151 hdr = req->async_data;
152 hdr->free_iov = NULL;
158 static inline struct io_async_msghdr *io_msg_alloc_async_prep(struct io_kiocb *req)
160 /* ->prep_async is always called from the submission context */
161 return io_msg_alloc_async(req, 0);
164 static int io_setup_async_msg(struct io_kiocb *req,
165 struct io_async_msghdr *kmsg,
166 unsigned int issue_flags)
168 struct io_async_msghdr *async_msg;
170 if (req_has_async_data(req))
172 async_msg = io_msg_alloc_async(req, issue_flags);
174 kfree(kmsg->free_iov);
177 req->flags |= REQ_F_NEED_CLEANUP;
178 memcpy(async_msg, kmsg, sizeof(*kmsg));
179 if (async_msg->msg.msg_name)
180 async_msg->msg.msg_name = &async_msg->addr;
182 if ((req->flags & REQ_F_BUFFER_SELECT) && !async_msg->msg.msg_iter.nr_segs)
185 /* if were using fast_iov, set it to the new one */
186 if (iter_is_iovec(&kmsg->msg.msg_iter) && !kmsg->free_iov) {
187 size_t fast_idx = iter_iov(&kmsg->msg.msg_iter) - kmsg->fast_iov;
188 async_msg->msg.msg_iter.__iov = &async_msg->fast_iov[fast_idx];
195 static int io_compat_msg_copy_hdr(struct io_kiocb *req,
196 struct io_async_msghdr *iomsg,
197 struct compat_msghdr *msg, int ddir)
199 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
200 struct compat_iovec __user *uiov;
203 if (copy_from_user(msg, sr->umsg_compat, sizeof(*msg)))
206 uiov = compat_ptr(msg->msg_iov);
207 if (req->flags & REQ_F_BUFFER_SELECT) {
210 iomsg->free_iov = NULL;
211 if (msg->msg_iovlen == 0) {
213 } else if (msg->msg_iovlen > 1) {
216 if (!access_ok(uiov, sizeof(*uiov)))
218 if (__get_user(clen, &uiov->iov_len))
228 iomsg->free_iov = iomsg->fast_iov;
229 ret = __import_iovec(ddir, (struct iovec __user *)uiov, msg->msg_iovlen,
230 UIO_FASTIOV, &iomsg->free_iov,
231 &iomsg->msg.msg_iter, true);
232 if (unlikely(ret < 0))
239 static int io_msg_copy_hdr(struct io_kiocb *req, struct io_async_msghdr *iomsg,
240 struct user_msghdr *msg, int ddir)
242 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
245 if (!user_access_begin(sr->umsg, sizeof(*sr->umsg)))
249 unsafe_get_user(msg->msg_name, &sr->umsg->msg_name, ua_end);
250 unsafe_get_user(msg->msg_namelen, &sr->umsg->msg_namelen, ua_end);
251 unsafe_get_user(msg->msg_iov, &sr->umsg->msg_iov, ua_end);
252 unsafe_get_user(msg->msg_iovlen, &sr->umsg->msg_iovlen, ua_end);
253 unsafe_get_user(msg->msg_control, &sr->umsg->msg_control, ua_end);
254 unsafe_get_user(msg->msg_controllen, &sr->umsg->msg_controllen, ua_end);
257 if (req->flags & REQ_F_BUFFER_SELECT) {
258 if (msg->msg_iovlen == 0) {
259 sr->len = iomsg->fast_iov[0].iov_len = 0;
260 iomsg->fast_iov[0].iov_base = NULL;
261 iomsg->free_iov = NULL;
262 } else if (msg->msg_iovlen > 1) {
266 /* we only need the length for provided buffers */
267 if (!access_ok(&msg->msg_iov[0].iov_len, sizeof(__kernel_size_t)))
269 unsafe_get_user(iomsg->fast_iov[0].iov_len,
270 &msg->msg_iov[0].iov_len, ua_end);
271 sr->len = iomsg->fast_iov[0].iov_len;
272 iomsg->free_iov = NULL;
281 iomsg->free_iov = iomsg->fast_iov;
282 ret = __import_iovec(ddir, msg->msg_iov, msg->msg_iovlen, UIO_FASTIOV,
283 &iomsg->free_iov, &iomsg->msg.msg_iter, false);
284 if (unlikely(ret < 0))
290 static int io_sendmsg_copy_hdr(struct io_kiocb *req,
291 struct io_async_msghdr *iomsg)
293 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
294 struct user_msghdr msg;
297 iomsg->msg.msg_name = &iomsg->addr;
298 iomsg->msg.msg_iter.nr_segs = 0;
301 if (unlikely(req->ctx->compat)) {
302 struct compat_msghdr cmsg;
304 ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ITER_SOURCE);
308 return __get_compat_msghdr(&iomsg->msg, &cmsg, NULL);
312 ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_SOURCE);
316 ret = __copy_msghdr(&iomsg->msg, &msg, NULL);
318 /* save msg_control as sys_sendmsg() overwrites it */
319 sr->msg_control = iomsg->msg.msg_control_user;
323 int io_send_prep_async(struct io_kiocb *req)
325 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
326 struct io_async_msghdr *io;
329 if (!zc->addr || req_has_async_data(req))
331 io = io_msg_alloc_async_prep(req);
334 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
338 static int io_setup_async_addr(struct io_kiocb *req,
339 struct sockaddr_storage *addr_storage,
340 unsigned int issue_flags)
342 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
343 struct io_async_msghdr *io;
345 if (!sr->addr || req_has_async_data(req))
347 io = io_msg_alloc_async(req, issue_flags);
350 memcpy(&io->addr, addr_storage, sizeof(io->addr));
354 int io_sendmsg_prep_async(struct io_kiocb *req)
358 if (!io_msg_alloc_async_prep(req))
360 ret = io_sendmsg_copy_hdr(req, req->async_data);
362 req->flags |= REQ_F_NEED_CLEANUP;
366 void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
368 struct io_async_msghdr *io = req->async_data;
373 int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
375 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
379 if (req->opcode == IORING_OP_SEND) {
380 if (READ_ONCE(sqe->__pad3[0]))
382 sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
383 sr->addr_len = READ_ONCE(sqe->addr_len);
384 } else if (sqe->addr2 || sqe->file_index) {
388 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
389 sr->len = READ_ONCE(sqe->len);
390 sr->flags = READ_ONCE(sqe->ioprio);
391 if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
393 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
394 if (sr->msg_flags & MSG_DONTWAIT)
395 req->flags |= REQ_F_NOWAIT;
398 if (req->ctx->compat)
399 sr->msg_flags |= MSG_CMSG_COMPAT;
404 static void io_req_msg_cleanup(struct io_kiocb *req,
405 struct io_async_msghdr *kmsg,
406 unsigned int issue_flags)
408 req->flags &= ~REQ_F_NEED_CLEANUP;
409 /* fast path, check for non-NULL to avoid function call */
411 kfree(kmsg->free_iov);
412 io_netmsg_recycle(req, issue_flags);
415 int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
417 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
418 struct io_async_msghdr iomsg, *kmsg;
424 sock = sock_from_file(req->file);
428 if (req_has_async_data(req)) {
429 kmsg = req->async_data;
430 kmsg->msg.msg_control_user = sr->msg_control;
432 ret = io_sendmsg_copy_hdr(req, &iomsg);
438 if (!(req->flags & REQ_F_POLLED) &&
439 (sr->flags & IORING_RECVSEND_POLL_FIRST))
440 return io_setup_async_msg(req, kmsg, issue_flags);
442 flags = sr->msg_flags;
443 if (issue_flags & IO_URING_F_NONBLOCK)
444 flags |= MSG_DONTWAIT;
445 if (flags & MSG_WAITALL)
446 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
448 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
451 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
452 return io_setup_async_msg(req, kmsg, issue_flags);
453 if (ret > 0 && io_net_retry(sock, flags)) {
454 kmsg->msg.msg_controllen = 0;
455 kmsg->msg.msg_control = NULL;
457 req->flags |= REQ_F_BL_NO_RECYCLE;
458 return io_setup_async_msg(req, kmsg, issue_flags);
460 if (ret == -ERESTARTSYS)
464 io_req_msg_cleanup(req, kmsg, issue_flags);
467 else if (sr->done_io)
469 io_req_set_res(req, ret, 0);
473 int io_send(struct io_kiocb *req, unsigned int issue_flags)
475 struct sockaddr_storage __address;
476 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
484 msg.msg_control = NULL;
485 msg.msg_controllen = 0;
490 if (req_has_async_data(req)) {
491 struct io_async_msghdr *io = req->async_data;
493 msg.msg_name = &io->addr;
495 ret = move_addr_to_kernel(sr->addr, sr->addr_len, &__address);
496 if (unlikely(ret < 0))
498 msg.msg_name = (struct sockaddr *)&__address;
500 msg.msg_namelen = sr->addr_len;
503 if (!(req->flags & REQ_F_POLLED) &&
504 (sr->flags & IORING_RECVSEND_POLL_FIRST))
505 return io_setup_async_addr(req, &__address, issue_flags);
507 sock = sock_from_file(req->file);
511 ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len, &msg.msg_iter);
515 flags = sr->msg_flags;
516 if (issue_flags & IO_URING_F_NONBLOCK)
517 flags |= MSG_DONTWAIT;
518 if (flags & MSG_WAITALL)
519 min_ret = iov_iter_count(&msg.msg_iter);
521 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
522 msg.msg_flags = flags;
523 ret = sock_sendmsg(sock, &msg);
525 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
526 return io_setup_async_addr(req, &__address, issue_flags);
528 if (ret > 0 && io_net_retry(sock, flags)) {
532 req->flags |= REQ_F_BL_NO_RECYCLE;
533 return io_setup_async_addr(req, &__address, issue_flags);
535 if (ret == -ERESTARTSYS)
541 else if (sr->done_io)
543 io_req_set_res(req, ret, 0);
547 static int io_recvmsg_mshot_prep(struct io_kiocb *req,
548 struct io_async_msghdr *iomsg,
549 int namelen, size_t controllen)
551 if ((req->flags & (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) ==
552 (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) {
555 if (unlikely(namelen < 0))
557 if (check_add_overflow(sizeof(struct io_uring_recvmsg_out),
560 if (check_add_overflow(hdr, controllen, &hdr))
563 iomsg->namelen = namelen;
564 iomsg->controllen = controllen;
571 static int io_recvmsg_copy_hdr(struct io_kiocb *req,
572 struct io_async_msghdr *iomsg)
574 struct user_msghdr msg;
577 iomsg->msg.msg_name = &iomsg->addr;
578 iomsg->msg.msg_iter.nr_segs = 0;
581 if (unlikely(req->ctx->compat)) {
582 struct compat_msghdr cmsg;
584 ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ITER_DEST);
588 ret = __get_compat_msghdr(&iomsg->msg, &cmsg, &iomsg->uaddr);
592 return io_recvmsg_mshot_prep(req, iomsg, cmsg.msg_namelen,
593 cmsg.msg_controllen);
597 ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_DEST);
601 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
605 return io_recvmsg_mshot_prep(req, iomsg, msg.msg_namelen,
609 int io_recvmsg_prep_async(struct io_kiocb *req)
611 struct io_async_msghdr *iomsg;
614 if (!io_msg_alloc_async_prep(req))
616 iomsg = req->async_data;
617 ret = io_recvmsg_copy_hdr(req, iomsg);
619 req->flags |= REQ_F_NEED_CLEANUP;
623 #define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
625 int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
627 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
631 if (unlikely(sqe->file_index || sqe->addr2))
634 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
635 sr->len = READ_ONCE(sqe->len);
636 sr->flags = READ_ONCE(sqe->ioprio);
637 if (sr->flags & ~(RECVMSG_FLAGS))
639 sr->msg_flags = READ_ONCE(sqe->msg_flags);
640 if (sr->msg_flags & MSG_DONTWAIT)
641 req->flags |= REQ_F_NOWAIT;
642 if (sr->msg_flags & MSG_ERRQUEUE)
643 req->flags |= REQ_F_CLEAR_POLLIN;
644 if (sr->flags & IORING_RECV_MULTISHOT) {
645 if (!(req->flags & REQ_F_BUFFER_SELECT))
647 if (sr->msg_flags & MSG_WAITALL)
649 if (req->opcode == IORING_OP_RECV && sr->len)
651 req->flags |= REQ_F_APOLL_MULTISHOT;
653 * Store the buffer group for this multishot receive separately,
654 * as if we end up doing an io-wq based issue that selects a
655 * buffer, it has to be committed immediately and that will
656 * clear ->buf_list. This means we lose the link to the buffer
657 * list, and the eventual buffer put on completion then cannot
660 sr->buf_group = req->buf_index;
664 if (req->ctx->compat)
665 sr->msg_flags |= MSG_CMSG_COMPAT;
667 sr->nr_multishot_loops = 0;
671 static inline void io_recv_prep_retry(struct io_kiocb *req)
673 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
675 req->flags &= ~REQ_F_BL_EMPTY;
677 sr->len = 0; /* get from the provided buffer */
678 req->buf_index = sr->buf_group;
682 * Finishes io_recv and io_recvmsg.
684 * Returns true if it is actually finished, or false if it should run
685 * again (for multishot).
687 static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
688 struct msghdr *msg, bool mshot_finished,
689 unsigned issue_flags)
693 cflags = io_put_kbuf(req, issue_flags);
694 if (msg->msg_inq > 0)
695 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
698 * Fill CQE for this receive and see if we should keep trying to
699 * receive from this socket.
701 if ((req->flags & REQ_F_APOLL_MULTISHOT) && !mshot_finished &&
702 io_fill_cqe_req_aux(req, issue_flags & IO_URING_F_COMPLETE_DEFER,
703 *ret, cflags | IORING_CQE_F_MORE)) {
704 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
705 int mshot_retry_ret = IOU_ISSUE_SKIP_COMPLETE;
707 io_recv_prep_retry(req);
708 /* Known not-empty or unknown state, retry */
709 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || msg->msg_inq < 0) {
710 if (sr->nr_multishot_loops++ < MULTISHOT_MAX_RETRY)
712 /* mshot retries exceeded, force a requeue */
713 sr->nr_multishot_loops = 0;
714 mshot_retry_ret = IOU_REQUEUE;
716 if (issue_flags & IO_URING_F_MULTISHOT)
717 *ret = mshot_retry_ret;
723 /* Finish the request / stop multishot. */
724 io_req_set_res(req, *ret, cflags);
726 if (issue_flags & IO_URING_F_MULTISHOT)
727 *ret = IOU_STOP_MULTISHOT;
733 static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
734 struct io_sr_msg *sr, void __user **buf,
737 unsigned long ubuf = (unsigned long) *buf;
740 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
745 if (kmsg->controllen) {
746 unsigned long control = ubuf + hdr - kmsg->controllen;
748 kmsg->msg.msg_control_user = (void __user *) control;
749 kmsg->msg.msg_controllen = kmsg->controllen;
752 sr->buf = *buf; /* stash for later copy */
753 *buf = (void __user *) (ubuf + hdr);
754 kmsg->payloadlen = *len = *len - hdr;
758 struct io_recvmsg_multishot_hdr {
759 struct io_uring_recvmsg_out msg;
760 struct sockaddr_storage addr;
763 static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
764 struct io_async_msghdr *kmsg,
765 unsigned int flags, bool *finished)
769 struct io_recvmsg_multishot_hdr hdr;
772 kmsg->msg.msg_name = &hdr.addr;
773 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
774 kmsg->msg.msg_namelen = 0;
776 if (sock->file->f_flags & O_NONBLOCK)
777 flags |= MSG_DONTWAIT;
779 err = sock_recvmsg(sock, &kmsg->msg, flags);
780 *finished = err <= 0;
784 hdr.msg = (struct io_uring_recvmsg_out) {
785 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
786 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
789 hdr.msg.payloadlen = err;
790 if (err > kmsg->payloadlen)
791 err = kmsg->payloadlen;
793 copy_len = sizeof(struct io_uring_recvmsg_out);
794 if (kmsg->msg.msg_namelen > kmsg->namelen)
795 copy_len += kmsg->namelen;
797 copy_len += kmsg->msg.msg_namelen;
800 * "fromlen shall refer to the value before truncation.."
803 hdr.msg.namelen = kmsg->msg.msg_namelen;
805 /* ensure that there is no gap between hdr and sockaddr_storage */
806 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
807 sizeof(struct io_uring_recvmsg_out));
808 if (copy_to_user(io->buf, &hdr, copy_len)) {
813 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
814 kmsg->controllen + err;
817 int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
819 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
820 struct io_async_msghdr iomsg, *kmsg;
823 int ret, min_ret = 0;
824 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
825 bool mshot_finished = true;
827 sock = sock_from_file(req->file);
831 if (req_has_async_data(req)) {
832 kmsg = req->async_data;
834 ret = io_recvmsg_copy_hdr(req, &iomsg);
840 if (!(req->flags & REQ_F_POLLED) &&
841 (sr->flags & IORING_RECVSEND_POLL_FIRST))
842 return io_setup_async_msg(req, kmsg, issue_flags);
844 flags = sr->msg_flags;
846 flags |= MSG_DONTWAIT;
849 if (io_do_buffer_select(req)) {
851 size_t len = sr->len;
853 buf = io_buffer_select(req, &len, issue_flags);
857 if (req->flags & REQ_F_APOLL_MULTISHOT) {
858 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
860 io_kbuf_recycle(req, issue_flags);
865 iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_DEST, buf, len);
868 kmsg->msg.msg_get_inq = 1;
869 kmsg->msg.msg_inq = -1;
870 if (req->flags & REQ_F_APOLL_MULTISHOT) {
871 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
874 /* disable partial retry for recvmsg with cmsg attached */
875 if (flags & MSG_WAITALL && !kmsg->msg.msg_controllen)
876 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
878 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
883 if (ret == -EAGAIN && force_nonblock) {
884 ret = io_setup_async_msg(req, kmsg, issue_flags);
885 if (ret == -EAGAIN && (issue_flags & IO_URING_F_MULTISHOT)) {
886 io_kbuf_recycle(req, issue_flags);
887 return IOU_ISSUE_SKIP_COMPLETE;
891 if (ret > 0 && io_net_retry(sock, flags)) {
893 req->flags |= REQ_F_BL_NO_RECYCLE;
894 return io_setup_async_msg(req, kmsg, issue_flags);
896 if (ret == -ERESTARTSYS)
899 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
905 else if (sr->done_io)
908 io_kbuf_recycle(req, issue_flags);
910 if (!io_recv_finish(req, &ret, &kmsg->msg, mshot_finished, issue_flags))
911 goto retry_multishot;
914 io_req_msg_cleanup(req, kmsg, issue_flags);
915 else if (ret == -EAGAIN)
916 return io_setup_async_msg(req, kmsg, issue_flags);
921 int io_recv(struct io_kiocb *req, unsigned int issue_flags)
923 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
927 int ret, min_ret = 0;
928 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
929 size_t len = sr->len;
931 if (!(req->flags & REQ_F_POLLED) &&
932 (sr->flags & IORING_RECVSEND_POLL_FIRST))
935 sock = sock_from_file(req->file);
941 msg.msg_control = NULL;
943 msg.msg_controllen = 0;
947 flags = sr->msg_flags;
949 flags |= MSG_DONTWAIT;
952 if (io_do_buffer_select(req)) {
955 buf = io_buffer_select(req, &len, issue_flags);
962 ret = import_ubuf(ITER_DEST, sr->buf, len, &msg.msg_iter);
969 if (flags & MSG_WAITALL)
970 min_ret = iov_iter_count(&msg.msg_iter);
972 ret = sock_recvmsg(sock, &msg, flags);
974 if (ret == -EAGAIN && force_nonblock) {
975 if (issue_flags & IO_URING_F_MULTISHOT) {
976 io_kbuf_recycle(req, issue_flags);
977 return IOU_ISSUE_SKIP_COMPLETE;
982 if (ret > 0 && io_net_retry(sock, flags)) {
986 req->flags |= REQ_F_BL_NO_RECYCLE;
989 if (ret == -ERESTARTSYS)
992 } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
999 else if (sr->done_io)
1002 io_kbuf_recycle(req, issue_flags);
1004 if (!io_recv_finish(req, &ret, &msg, ret <= 0, issue_flags))
1005 goto retry_multishot;
1010 void io_send_zc_cleanup(struct io_kiocb *req)
1012 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1013 struct io_async_msghdr *io;
1015 if (req_has_async_data(req)) {
1016 io = req->async_data;
1017 /* might be ->fast_iov if *msg_copy_hdr failed */
1018 if (io->free_iov != io->fast_iov)
1019 kfree(io->free_iov);
1022 io_notif_flush(zc->notif);
1027 #define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF)
1028 #define IO_ZC_FLAGS_VALID (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE)
1030 int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1032 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1033 struct io_ring_ctx *ctx = req->ctx;
1034 struct io_kiocb *notif;
1038 if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
1040 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
1041 if (req->flags & REQ_F_CQE_SKIP)
1044 notif = zc->notif = io_alloc_notif(ctx);
1047 notif->cqe.user_data = req->cqe.user_data;
1049 notif->cqe.flags = IORING_CQE_F_NOTIF;
1050 req->flags |= REQ_F_NEED_CLEANUP;
1052 zc->flags = READ_ONCE(sqe->ioprio);
1053 if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) {
1054 if (zc->flags & ~IO_ZC_FLAGS_VALID)
1056 if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) {
1057 io_notif_set_extended(notif);
1058 io_notif_to_data(notif)->zc_report = true;
1062 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1063 unsigned idx = READ_ONCE(sqe->buf_index);
1065 if (unlikely(idx >= ctx->nr_user_bufs))
1067 idx = array_index_nospec(idx, ctx->nr_user_bufs);
1068 req->imu = READ_ONCE(ctx->user_bufs[idx]);
1069 io_req_set_rsrc_node(notif, ctx, 0);
1072 if (req->opcode == IORING_OP_SEND_ZC) {
1073 if (READ_ONCE(sqe->__pad3[0]))
1075 zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1076 zc->addr_len = READ_ONCE(sqe->addr_len);
1078 if (unlikely(sqe->addr2 || sqe->file_index))
1080 if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
1084 zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
1085 zc->len = READ_ONCE(sqe->len);
1086 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
1087 if (zc->msg_flags & MSG_DONTWAIT)
1088 req->flags |= REQ_F_NOWAIT;
1090 #ifdef CONFIG_COMPAT
1091 if (req->ctx->compat)
1092 zc->msg_flags |= MSG_CMSG_COMPAT;
1097 static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
1098 struct iov_iter *from, size_t length)
1100 skb_zcopy_downgrade_managed(skb);
1101 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1104 static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1105 struct iov_iter *from, size_t length)
1107 struct skb_shared_info *shinfo = skb_shinfo(skb);
1108 int frag = shinfo->nr_frags;
1110 struct bvec_iter bi;
1112 unsigned long truesize = 0;
1115 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
1116 else if (unlikely(!skb_zcopy_managed(skb)))
1117 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1119 bi.bi_size = min(from->count, length);
1120 bi.bi_bvec_done = from->iov_offset;
1123 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1124 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1127 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1128 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1129 v.bv_offset, v.bv_len);
1130 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1135 shinfo->nr_frags = frag;
1136 from->bvec += bi.bi_idx;
1137 from->nr_segs -= bi.bi_idx;
1138 from->count -= copied;
1139 from->iov_offset = bi.bi_bvec_done;
1141 skb->data_len += copied;
1143 skb->truesize += truesize;
1145 if (sk && sk->sk_type == SOCK_STREAM) {
1146 sk_wmem_queued_add(sk, truesize);
1147 if (!skb_zcopy_pure(skb))
1148 sk_mem_charge(sk, truesize);
1150 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1155 int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
1157 struct sockaddr_storage __address;
1158 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1160 struct socket *sock;
1162 int ret, min_ret = 0;
1164 sock = sock_from_file(req->file);
1165 if (unlikely(!sock))
1167 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1170 msg.msg_name = NULL;
1171 msg.msg_control = NULL;
1172 msg.msg_controllen = 0;
1173 msg.msg_namelen = 0;
1176 if (req_has_async_data(req)) {
1177 struct io_async_msghdr *io = req->async_data;
1179 msg.msg_name = &io->addr;
1181 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1182 if (unlikely(ret < 0))
1184 msg.msg_name = (struct sockaddr *)&__address;
1186 msg.msg_namelen = zc->addr_len;
1189 if (!(req->flags & REQ_F_POLLED) &&
1190 (zc->flags & IORING_RECVSEND_POLL_FIRST))
1191 return io_setup_async_addr(req, &__address, issue_flags);
1193 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1194 ret = io_import_fixed(ITER_SOURCE, &msg.msg_iter, req->imu,
1195 (u64)(uintptr_t)zc->buf, zc->len);
1198 msg.sg_from_iter = io_sg_from_iter;
1200 io_notif_set_extended(zc->notif);
1201 ret = import_ubuf(ITER_SOURCE, zc->buf, zc->len, &msg.msg_iter);
1204 ret = io_notif_account_mem(zc->notif, zc->len);
1207 msg.sg_from_iter = io_sg_from_iter_iovec;
1210 msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1211 if (issue_flags & IO_URING_F_NONBLOCK)
1212 msg_flags |= MSG_DONTWAIT;
1213 if (msg_flags & MSG_WAITALL)
1214 min_ret = iov_iter_count(&msg.msg_iter);
1215 msg_flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
1217 msg.msg_flags = msg_flags;
1218 msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
1219 ret = sock_sendmsg(sock, &msg);
1221 if (unlikely(ret < min_ret)) {
1222 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1223 return io_setup_async_addr(req, &__address, issue_flags);
1225 if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1229 req->flags |= REQ_F_BL_NO_RECYCLE;
1230 return io_setup_async_addr(req, &__address, issue_flags);
1232 if (ret == -ERESTARTSYS)
1239 else if (zc->done_io)
1243 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1244 * flushing notif to io_send_zc_cleanup()
1246 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1247 io_notif_flush(zc->notif);
1248 req->flags &= ~REQ_F_NEED_CLEANUP;
1250 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1254 int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1256 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1257 struct io_async_msghdr iomsg, *kmsg;
1258 struct socket *sock;
1260 int ret, min_ret = 0;
1262 io_notif_set_extended(sr->notif);
1264 sock = sock_from_file(req->file);
1265 if (unlikely(!sock))
1267 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1270 if (req_has_async_data(req)) {
1271 kmsg = req->async_data;
1273 ret = io_sendmsg_copy_hdr(req, &iomsg);
1279 if (!(req->flags & REQ_F_POLLED) &&
1280 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1281 return io_setup_async_msg(req, kmsg, issue_flags);
1283 flags = sr->msg_flags | MSG_ZEROCOPY;
1284 if (issue_flags & IO_URING_F_NONBLOCK)
1285 flags |= MSG_DONTWAIT;
1286 if (flags & MSG_WAITALL)
1287 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1289 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1290 kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1291 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1293 if (unlikely(ret < min_ret)) {
1294 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1295 return io_setup_async_msg(req, kmsg, issue_flags);
1297 if (ret > 0 && io_net_retry(sock, flags)) {
1299 req->flags |= REQ_F_BL_NO_RECYCLE;
1300 return io_setup_async_msg(req, kmsg, issue_flags);
1302 if (ret == -ERESTARTSYS)
1306 /* fast path, check for non-NULL to avoid function call */
1307 if (kmsg->free_iov) {
1308 kfree(kmsg->free_iov);
1309 kmsg->free_iov = NULL;
1312 io_netmsg_recycle(req, issue_flags);
1315 else if (sr->done_io)
1319 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1320 * flushing notif to io_send_zc_cleanup()
1322 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1323 io_notif_flush(sr->notif);
1324 req->flags &= ~REQ_F_NEED_CLEANUP;
1326 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1330 void io_sendrecv_fail(struct io_kiocb *req)
1332 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1335 req->cqe.res = sr->done_io;
1337 if ((req->flags & REQ_F_NEED_CLEANUP) &&
1338 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1339 req->cqe.flags |= IORING_CQE_F_MORE;
1342 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1344 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1347 if (sqe->len || sqe->buf_index)
1350 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1351 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1352 accept->flags = READ_ONCE(sqe->accept_flags);
1353 accept->nofile = rlimit(RLIMIT_NOFILE);
1354 flags = READ_ONCE(sqe->ioprio);
1355 if (flags & ~IORING_ACCEPT_MULTISHOT)
1358 accept->file_slot = READ_ONCE(sqe->file_index);
1359 if (accept->file_slot) {
1360 if (accept->flags & SOCK_CLOEXEC)
1362 if (flags & IORING_ACCEPT_MULTISHOT &&
1363 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1366 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1368 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1369 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1370 if (flags & IORING_ACCEPT_MULTISHOT)
1371 req->flags |= REQ_F_APOLL_MULTISHOT;
1375 int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1377 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1378 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1379 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1380 bool fixed = !!accept->file_slot;
1386 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1387 if (unlikely(fd < 0))
1390 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1395 ret = PTR_ERR(file);
1396 if (ret == -EAGAIN && force_nonblock) {
1398 * if it's multishot and polled, we don't need to
1399 * return EAGAIN to arm the poll infra since it
1400 * has already been done
1402 if (issue_flags & IO_URING_F_MULTISHOT)
1403 return IOU_ISSUE_SKIP_COMPLETE;
1406 if (ret == -ERESTARTSYS)
1409 } else if (!fixed) {
1410 fd_install(fd, file);
1413 ret = io_fixed_fd_install(req, issue_flags, file,
1417 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1418 io_req_set_res(req, ret, 0);
1424 if (io_fill_cqe_req_aux(req, issue_flags & IO_URING_F_COMPLETE_DEFER,
1425 ret, IORING_CQE_F_MORE))
1428 io_req_set_res(req, ret, 0);
1429 return IOU_STOP_MULTISHOT;
1432 int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1434 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1436 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1439 sock->domain = READ_ONCE(sqe->fd);
1440 sock->type = READ_ONCE(sqe->off);
1441 sock->protocol = READ_ONCE(sqe->len);
1442 sock->file_slot = READ_ONCE(sqe->file_index);
1443 sock->nofile = rlimit(RLIMIT_NOFILE);
1445 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1446 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1448 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1453 int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1455 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1456 bool fixed = !!sock->file_slot;
1461 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1462 if (unlikely(fd < 0))
1465 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1469 ret = PTR_ERR(file);
1470 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1472 if (ret == -ERESTARTSYS)
1475 } else if (!fixed) {
1476 fd_install(fd, file);
1479 ret = io_fixed_fd_install(req, issue_flags, file,
1482 io_req_set_res(req, ret, 0);
1486 int io_connect_prep_async(struct io_kiocb *req)
1488 struct io_async_connect *io = req->async_data;
1489 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1491 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1494 int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1496 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1498 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1501 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1502 conn->addr_len = READ_ONCE(sqe->addr2);
1503 conn->in_progress = conn->seen_econnaborted = false;
1507 int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1509 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1510 struct io_async_connect __io, *io;
1511 unsigned file_flags;
1513 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1515 if (req_has_async_data(req)) {
1516 io = req->async_data;
1518 ret = move_addr_to_kernel(connect->addr,
1526 file_flags = force_nonblock ? O_NONBLOCK : 0;
1528 ret = __sys_connect_file(req->file, &io->address,
1529 connect->addr_len, file_flags);
1530 if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED)
1531 && force_nonblock) {
1532 if (ret == -EINPROGRESS) {
1533 connect->in_progress = true;
1534 } else if (ret == -ECONNABORTED) {
1535 if (connect->seen_econnaborted)
1537 connect->seen_econnaborted = true;
1539 if (req_has_async_data(req))
1541 if (io_alloc_async_data(req)) {
1545 memcpy(req->async_data, &__io, sizeof(__io));
1548 if (connect->in_progress) {
1550 * At least bluetooth will return -EBADFD on a re-connect
1551 * attempt, and it's (supposedly) also valid to get -EISCONN
1552 * which means the previous result is good. For both of these,
1553 * grab the sock_error() and use that for the completion.
1555 if (ret == -EBADFD || ret == -EISCONN)
1556 ret = sock_error(sock_from_file(req->file)->sk);
1558 if (ret == -ERESTARTSYS)
1563 io_req_set_res(req, ret, 0);
1567 void io_netmsg_cache_free(struct io_cache_entry *entry)
1569 kfree(container_of(entry, struct io_async_msghdr, cache));