Loading...
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/net.h>
7#include <linux/compat.h>
8#include <net/compat.h>
9#include <linux/io_uring.h>
10
11#include <uapi/linux/io_uring.h>
12
13#include "io_uring.h"
14#include "kbuf.h"
15#include "alloc_cache.h"
16#include "net.h"
17#include "notif.h"
18#include "rsrc.h"
19
20#if defined(CONFIG_NET)
21struct io_shutdown {
22 struct file *file;
23 int how;
24};
25
26struct io_accept {
27 struct file *file;
28 struct sockaddr __user *addr;
29 int __user *addr_len;
30 int flags;
31 u32 file_slot;
32 unsigned long nofile;
33};
34
35struct io_socket {
36 struct file *file;
37 int domain;
38 int type;
39 int protocol;
40 int flags;
41 u32 file_slot;
42 unsigned long nofile;
43};
44
45struct io_connect {
46 struct file *file;
47 struct sockaddr __user *addr;
48 int addr_len;
49 bool in_progress;
50 bool seen_econnaborted;
51};
52
53struct io_sr_msg {
54 struct file *file;
55 union {
56 struct compat_msghdr __user *umsg_compat;
57 struct user_msghdr __user *umsg;
58 void __user *buf;
59 };
60 unsigned len;
61 unsigned done_io;
62 unsigned msg_flags;
63 unsigned nr_multishot_loops;
64 u16 flags;
65 /* initialised and used only by !msg send variants */
66 u16 addr_len;
67 u16 buf_group;
68 void __user *addr;
69 void __user *msg_control;
70 /* used only for send zerocopy */
71 struct io_kiocb *notif;
72};
73
74/*
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.
78 */
79#define MULTISHOT_MAX_RETRY 32
80
81static inline bool io_check_multishot(struct io_kiocb *req,
82 unsigned int issue_flags)
83{
84 /*
85 * When ->locked_cq is set we only allow to post CQEs from the original
86 * task context. Usual request completions will be handled in other
87 * generic paths but multipoll may decide to post extra cqes.
88 */
89 return !(issue_flags & IO_URING_F_IOWQ) ||
90 !(issue_flags & IO_URING_F_MULTISHOT) ||
91 !req->ctx->task_complete;
92}
93
94int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
95{
96 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
97
98 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
99 sqe->buf_index || sqe->splice_fd_in))
100 return -EINVAL;
101
102 shutdown->how = READ_ONCE(sqe->len);
103 req->flags |= REQ_F_FORCE_ASYNC;
104 return 0;
105}
106
107int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
108{
109 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
110 struct socket *sock;
111 int ret;
112
113 WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
114
115 sock = sock_from_file(req->file);
116 if (unlikely(!sock))
117 return -ENOTSOCK;
118
119 ret = __sys_shutdown_sock(sock, shutdown->how);
120 io_req_set_res(req, ret, 0);
121 return IOU_OK;
122}
123
124static bool io_net_retry(struct socket *sock, int flags)
125{
126 if (!(flags & MSG_WAITALL))
127 return false;
128 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
129}
130
131static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
132{
133 struct io_async_msghdr *hdr = req->async_data;
134
135 if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
136 return;
137
138 /* Let normal cleanup path reap it if we fail adding to the cache */
139 if (io_alloc_cache_put(&req->ctx->netmsg_cache, &hdr->cache)) {
140 req->async_data = NULL;
141 req->flags &= ~REQ_F_ASYNC_DATA;
142 }
143}
144
145static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
146 unsigned int issue_flags)
147{
148 struct io_ring_ctx *ctx = req->ctx;
149 struct io_cache_entry *entry;
150 struct io_async_msghdr *hdr;
151
152 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
153 entry = io_alloc_cache_get(&ctx->netmsg_cache);
154 if (entry) {
155 hdr = container_of(entry, struct io_async_msghdr, cache);
156 hdr->free_iov = NULL;
157 req->flags |= REQ_F_ASYNC_DATA;
158 req->async_data = hdr;
159 return hdr;
160 }
161 }
162
163 if (!io_alloc_async_data(req)) {
164 hdr = req->async_data;
165 hdr->free_iov = NULL;
166 return hdr;
167 }
168 return NULL;
169}
170
171static inline struct io_async_msghdr *io_msg_alloc_async_prep(struct io_kiocb *req)
172{
173 /* ->prep_async is always called from the submission context */
174 return io_msg_alloc_async(req, 0);
175}
176
177static int io_setup_async_msg(struct io_kiocb *req,
178 struct io_async_msghdr *kmsg,
179 unsigned int issue_flags)
180{
181 struct io_async_msghdr *async_msg;
182
183 if (req_has_async_data(req))
184 return -EAGAIN;
185 async_msg = io_msg_alloc_async(req, issue_flags);
186 if (!async_msg) {
187 kfree(kmsg->free_iov);
188 return -ENOMEM;
189 }
190 req->flags |= REQ_F_NEED_CLEANUP;
191 memcpy(async_msg, kmsg, sizeof(*kmsg));
192 if (async_msg->msg.msg_name)
193 async_msg->msg.msg_name = &async_msg->addr;
194
195 if ((req->flags & REQ_F_BUFFER_SELECT) && !async_msg->msg.msg_iter.nr_segs)
196 return -EAGAIN;
197
198 /* if were using fast_iov, set it to the new one */
199 if (iter_is_iovec(&kmsg->msg.msg_iter) && !kmsg->free_iov) {
200 size_t fast_idx = iter_iov(&kmsg->msg.msg_iter) - kmsg->fast_iov;
201 async_msg->msg.msg_iter.__iov = &async_msg->fast_iov[fast_idx];
202 }
203
204 return -EAGAIN;
205}
206
207static int io_sendmsg_copy_hdr(struct io_kiocb *req,
208 struct io_async_msghdr *iomsg)
209{
210 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
211 int ret;
212
213 iomsg->msg.msg_name = &iomsg->addr;
214 iomsg->free_iov = iomsg->fast_iov;
215 ret = sendmsg_copy_msghdr(&iomsg->msg, sr->umsg, sr->msg_flags,
216 &iomsg->free_iov);
217 /* save msg_control as sys_sendmsg() overwrites it */
218 sr->msg_control = iomsg->msg.msg_control_user;
219 return ret;
220}
221
222int io_send_prep_async(struct io_kiocb *req)
223{
224 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
225 struct io_async_msghdr *io;
226 int ret;
227
228 if (!zc->addr || req_has_async_data(req))
229 return 0;
230 io = io_msg_alloc_async_prep(req);
231 if (!io)
232 return -ENOMEM;
233 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
234 return ret;
235}
236
237static int io_setup_async_addr(struct io_kiocb *req,
238 struct sockaddr_storage *addr_storage,
239 unsigned int issue_flags)
240{
241 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
242 struct io_async_msghdr *io;
243
244 if (!sr->addr || req_has_async_data(req))
245 return -EAGAIN;
246 io = io_msg_alloc_async(req, issue_flags);
247 if (!io)
248 return -ENOMEM;
249 memcpy(&io->addr, addr_storage, sizeof(io->addr));
250 return -EAGAIN;
251}
252
253int io_sendmsg_prep_async(struct io_kiocb *req)
254{
255 int ret;
256
257 if (!io_msg_alloc_async_prep(req))
258 return -ENOMEM;
259 ret = io_sendmsg_copy_hdr(req, req->async_data);
260 if (!ret)
261 req->flags |= REQ_F_NEED_CLEANUP;
262 return ret;
263}
264
265void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
266{
267 struct io_async_msghdr *io = req->async_data;
268
269 kfree(io->free_iov);
270}
271
272int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
273{
274 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
275
276 if (req->opcode == IORING_OP_SEND) {
277 if (READ_ONCE(sqe->__pad3[0]))
278 return -EINVAL;
279 sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
280 sr->addr_len = READ_ONCE(sqe->addr_len);
281 } else if (sqe->addr2 || sqe->file_index) {
282 return -EINVAL;
283 }
284
285 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
286 sr->len = READ_ONCE(sqe->len);
287 sr->flags = READ_ONCE(sqe->ioprio);
288 if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
289 return -EINVAL;
290 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
291 if (sr->msg_flags & MSG_DONTWAIT)
292 req->flags |= REQ_F_NOWAIT;
293
294#ifdef CONFIG_COMPAT
295 if (req->ctx->compat)
296 sr->msg_flags |= MSG_CMSG_COMPAT;
297#endif
298 sr->done_io = 0;
299 return 0;
300}
301
302int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
303{
304 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
305 struct io_async_msghdr iomsg, *kmsg;
306 struct socket *sock;
307 unsigned flags;
308 int min_ret = 0;
309 int ret;
310
311 sock = sock_from_file(req->file);
312 if (unlikely(!sock))
313 return -ENOTSOCK;
314
315 if (req_has_async_data(req)) {
316 kmsg = req->async_data;
317 kmsg->msg.msg_control_user = sr->msg_control;
318 } else {
319 ret = io_sendmsg_copy_hdr(req, &iomsg);
320 if (ret)
321 return ret;
322 kmsg = &iomsg;
323 }
324
325 if (!(req->flags & REQ_F_POLLED) &&
326 (sr->flags & IORING_RECVSEND_POLL_FIRST))
327 return io_setup_async_msg(req, kmsg, issue_flags);
328
329 flags = sr->msg_flags;
330 if (issue_flags & IO_URING_F_NONBLOCK)
331 flags |= MSG_DONTWAIT;
332 if (flags & MSG_WAITALL)
333 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
334
335 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
336
337 if (ret < min_ret) {
338 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
339 return io_setup_async_msg(req, kmsg, issue_flags);
340 if (ret > 0 && io_net_retry(sock, flags)) {
341 kmsg->msg.msg_controllen = 0;
342 kmsg->msg.msg_control = NULL;
343 sr->done_io += ret;
344 req->flags |= REQ_F_PARTIAL_IO;
345 return io_setup_async_msg(req, kmsg, issue_flags);
346 }
347 if (ret == -ERESTARTSYS)
348 ret = -EINTR;
349 req_set_fail(req);
350 }
351 /* fast path, check for non-NULL to avoid function call */
352 if (kmsg->free_iov)
353 kfree(kmsg->free_iov);
354 req->flags &= ~REQ_F_NEED_CLEANUP;
355 io_netmsg_recycle(req, issue_flags);
356 if (ret >= 0)
357 ret += sr->done_io;
358 else if (sr->done_io)
359 ret = sr->done_io;
360 io_req_set_res(req, ret, 0);
361 return IOU_OK;
362}
363
364int io_send(struct io_kiocb *req, unsigned int issue_flags)
365{
366 struct sockaddr_storage __address;
367 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
368 struct msghdr msg;
369 struct socket *sock;
370 unsigned flags;
371 int min_ret = 0;
372 int ret;
373
374 msg.msg_name = NULL;
375 msg.msg_control = NULL;
376 msg.msg_controllen = 0;
377 msg.msg_namelen = 0;
378 msg.msg_ubuf = NULL;
379
380 if (sr->addr) {
381 if (req_has_async_data(req)) {
382 struct io_async_msghdr *io = req->async_data;
383
384 msg.msg_name = &io->addr;
385 } else {
386 ret = move_addr_to_kernel(sr->addr, sr->addr_len, &__address);
387 if (unlikely(ret < 0))
388 return ret;
389 msg.msg_name = (struct sockaddr *)&__address;
390 }
391 msg.msg_namelen = sr->addr_len;
392 }
393
394 if (!(req->flags & REQ_F_POLLED) &&
395 (sr->flags & IORING_RECVSEND_POLL_FIRST))
396 return io_setup_async_addr(req, &__address, issue_flags);
397
398 sock = sock_from_file(req->file);
399 if (unlikely(!sock))
400 return -ENOTSOCK;
401
402 ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len, &msg.msg_iter);
403 if (unlikely(ret))
404 return ret;
405
406 flags = sr->msg_flags;
407 if (issue_flags & IO_URING_F_NONBLOCK)
408 flags |= MSG_DONTWAIT;
409 if (flags & MSG_WAITALL)
410 min_ret = iov_iter_count(&msg.msg_iter);
411
412 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
413 msg.msg_flags = flags;
414 ret = sock_sendmsg(sock, &msg);
415 if (ret < min_ret) {
416 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
417 return io_setup_async_addr(req, &__address, issue_flags);
418
419 if (ret > 0 && io_net_retry(sock, flags)) {
420 sr->len -= ret;
421 sr->buf += ret;
422 sr->done_io += ret;
423 req->flags |= REQ_F_PARTIAL_IO;
424 return io_setup_async_addr(req, &__address, issue_flags);
425 }
426 if (ret == -ERESTARTSYS)
427 ret = -EINTR;
428 req_set_fail(req);
429 }
430 if (ret >= 0)
431 ret += sr->done_io;
432 else if (sr->done_io)
433 ret = sr->done_io;
434 io_req_set_res(req, ret, 0);
435 return IOU_OK;
436}
437
438static bool io_recvmsg_multishot_overflow(struct io_async_msghdr *iomsg)
439{
440 int hdr;
441
442 if (iomsg->namelen < 0)
443 return true;
444 if (check_add_overflow((int)sizeof(struct io_uring_recvmsg_out),
445 iomsg->namelen, &hdr))
446 return true;
447 if (check_add_overflow(hdr, (int)iomsg->controllen, &hdr))
448 return true;
449
450 return false;
451}
452
453static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
454 struct io_async_msghdr *iomsg)
455{
456 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
457 struct user_msghdr msg;
458 int ret;
459
460 if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
461 return -EFAULT;
462
463 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
464 if (ret)
465 return ret;
466
467 if (req->flags & REQ_F_BUFFER_SELECT) {
468 if (msg.msg_iovlen == 0) {
469 sr->len = iomsg->fast_iov[0].iov_len = 0;
470 iomsg->fast_iov[0].iov_base = NULL;
471 iomsg->free_iov = NULL;
472 } else if (msg.msg_iovlen > 1) {
473 return -EINVAL;
474 } else {
475 if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
476 return -EFAULT;
477 sr->len = iomsg->fast_iov[0].iov_len;
478 iomsg->free_iov = NULL;
479 }
480
481 if (req->flags & REQ_F_APOLL_MULTISHOT) {
482 iomsg->namelen = msg.msg_namelen;
483 iomsg->controllen = msg.msg_controllen;
484 if (io_recvmsg_multishot_overflow(iomsg))
485 return -EOVERFLOW;
486 }
487 } else {
488 iomsg->free_iov = iomsg->fast_iov;
489 ret = __import_iovec(ITER_DEST, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
490 &iomsg->free_iov, &iomsg->msg.msg_iter,
491 false);
492 if (ret > 0)
493 ret = 0;
494 }
495
496 return ret;
497}
498
499#ifdef CONFIG_COMPAT
500static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
501 struct io_async_msghdr *iomsg)
502{
503 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
504 struct compat_msghdr msg;
505 struct compat_iovec __user *uiov;
506 int ret;
507
508 if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
509 return -EFAULT;
510
511 ret = __get_compat_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
512 if (ret)
513 return ret;
514
515 uiov = compat_ptr(msg.msg_iov);
516 if (req->flags & REQ_F_BUFFER_SELECT) {
517 compat_ssize_t clen;
518
519 iomsg->free_iov = NULL;
520 if (msg.msg_iovlen == 0) {
521 sr->len = 0;
522 } else if (msg.msg_iovlen > 1) {
523 return -EINVAL;
524 } else {
525 if (!access_ok(uiov, sizeof(*uiov)))
526 return -EFAULT;
527 if (__get_user(clen, &uiov->iov_len))
528 return -EFAULT;
529 if (clen < 0)
530 return -EINVAL;
531 sr->len = clen;
532 }
533
534 if (req->flags & REQ_F_APOLL_MULTISHOT) {
535 iomsg->namelen = msg.msg_namelen;
536 iomsg->controllen = msg.msg_controllen;
537 if (io_recvmsg_multishot_overflow(iomsg))
538 return -EOVERFLOW;
539 }
540 } else {
541 iomsg->free_iov = iomsg->fast_iov;
542 ret = __import_iovec(ITER_DEST, (struct iovec __user *)uiov, msg.msg_iovlen,
543 UIO_FASTIOV, &iomsg->free_iov,
544 &iomsg->msg.msg_iter, true);
545 if (ret < 0)
546 return ret;
547 }
548
549 return 0;
550}
551#endif
552
553static int io_recvmsg_copy_hdr(struct io_kiocb *req,
554 struct io_async_msghdr *iomsg)
555{
556 iomsg->msg.msg_name = &iomsg->addr;
557 iomsg->msg.msg_iter.nr_segs = 0;
558
559#ifdef CONFIG_COMPAT
560 if (req->ctx->compat)
561 return __io_compat_recvmsg_copy_hdr(req, iomsg);
562#endif
563
564 return __io_recvmsg_copy_hdr(req, iomsg);
565}
566
567int io_recvmsg_prep_async(struct io_kiocb *req)
568{
569 int ret;
570
571 if (!io_msg_alloc_async_prep(req))
572 return -ENOMEM;
573 ret = io_recvmsg_copy_hdr(req, req->async_data);
574 if (!ret)
575 req->flags |= REQ_F_NEED_CLEANUP;
576 return ret;
577}
578
579#define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
580
581int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
582{
583 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
584
585 if (unlikely(sqe->file_index || sqe->addr2))
586 return -EINVAL;
587
588 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
589 sr->len = READ_ONCE(sqe->len);
590 sr->flags = READ_ONCE(sqe->ioprio);
591 if (sr->flags & ~(RECVMSG_FLAGS))
592 return -EINVAL;
593 sr->msg_flags = READ_ONCE(sqe->msg_flags);
594 if (sr->msg_flags & MSG_DONTWAIT)
595 req->flags |= REQ_F_NOWAIT;
596 if (sr->msg_flags & MSG_ERRQUEUE)
597 req->flags |= REQ_F_CLEAR_POLLIN;
598 if (sr->flags & IORING_RECV_MULTISHOT) {
599 if (!(req->flags & REQ_F_BUFFER_SELECT))
600 return -EINVAL;
601 if (sr->msg_flags & MSG_WAITALL)
602 return -EINVAL;
603 if (req->opcode == IORING_OP_RECV && sr->len)
604 return -EINVAL;
605 req->flags |= REQ_F_APOLL_MULTISHOT;
606 /*
607 * Store the buffer group for this multishot receive separately,
608 * as if we end up doing an io-wq based issue that selects a
609 * buffer, it has to be committed immediately and that will
610 * clear ->buf_list. This means we lose the link to the buffer
611 * list, and the eventual buffer put on completion then cannot
612 * restore it.
613 */
614 sr->buf_group = req->buf_index;
615 }
616
617#ifdef CONFIG_COMPAT
618 if (req->ctx->compat)
619 sr->msg_flags |= MSG_CMSG_COMPAT;
620#endif
621 sr->done_io = 0;
622 sr->nr_multishot_loops = 0;
623 return 0;
624}
625
626static inline void io_recv_prep_retry(struct io_kiocb *req)
627{
628 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
629
630 sr->done_io = 0;
631 sr->len = 0; /* get from the provided buffer */
632 req->buf_index = sr->buf_group;
633}
634
635/*
636 * Finishes io_recv and io_recvmsg.
637 *
638 * Returns true if it is actually finished, or false if it should run
639 * again (for multishot).
640 */
641static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
642 struct msghdr *msg, bool mshot_finished,
643 unsigned issue_flags)
644{
645 unsigned int cflags;
646
647 cflags = io_put_kbuf(req, issue_flags);
648 if (msg->msg_inq && msg->msg_inq != -1)
649 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
650
651 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
652 io_req_set_res(req, *ret, cflags);
653 *ret = IOU_OK;
654 return true;
655 }
656
657 if (mshot_finished)
658 goto finish;
659
660 /*
661 * Fill CQE for this receive and see if we should keep trying to
662 * receive from this socket.
663 */
664 if (io_fill_cqe_req_aux(req, issue_flags & IO_URING_F_COMPLETE_DEFER,
665 *ret, cflags | IORING_CQE_F_MORE)) {
666 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
667 int mshot_retry_ret = IOU_ISSUE_SKIP_COMPLETE;
668
669 io_recv_prep_retry(req);
670 /* Known not-empty or unknown state, retry */
671 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || msg->msg_inq == -1) {
672 if (sr->nr_multishot_loops++ < MULTISHOT_MAX_RETRY)
673 return false;
674 /* mshot retries exceeded, force a requeue */
675 sr->nr_multishot_loops = 0;
676 mshot_retry_ret = IOU_REQUEUE;
677 }
678 if (issue_flags & IO_URING_F_MULTISHOT)
679 *ret = mshot_retry_ret;
680 else
681 *ret = -EAGAIN;
682 return true;
683 }
684 /* Otherwise stop multishot but use the current result. */
685finish:
686 io_req_set_res(req, *ret, cflags);
687
688 if (issue_flags & IO_URING_F_MULTISHOT)
689 *ret = IOU_STOP_MULTISHOT;
690 else
691 *ret = IOU_OK;
692 return true;
693}
694
695static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
696 struct io_sr_msg *sr, void __user **buf,
697 size_t *len)
698{
699 unsigned long ubuf = (unsigned long) *buf;
700 unsigned long hdr;
701
702 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
703 kmsg->controllen;
704 if (*len < hdr)
705 return -EFAULT;
706
707 if (kmsg->controllen) {
708 unsigned long control = ubuf + hdr - kmsg->controllen;
709
710 kmsg->msg.msg_control_user = (void __user *) control;
711 kmsg->msg.msg_controllen = kmsg->controllen;
712 }
713
714 sr->buf = *buf; /* stash for later copy */
715 *buf = (void __user *) (ubuf + hdr);
716 kmsg->payloadlen = *len = *len - hdr;
717 return 0;
718}
719
720struct io_recvmsg_multishot_hdr {
721 struct io_uring_recvmsg_out msg;
722 struct sockaddr_storage addr;
723};
724
725static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
726 struct io_async_msghdr *kmsg,
727 unsigned int flags, bool *finished)
728{
729 int err;
730 int copy_len;
731 struct io_recvmsg_multishot_hdr hdr;
732
733 if (kmsg->namelen)
734 kmsg->msg.msg_name = &hdr.addr;
735 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
736 kmsg->msg.msg_namelen = 0;
737
738 if (sock->file->f_flags & O_NONBLOCK)
739 flags |= MSG_DONTWAIT;
740
741 err = sock_recvmsg(sock, &kmsg->msg, flags);
742 *finished = err <= 0;
743 if (err < 0)
744 return err;
745
746 hdr.msg = (struct io_uring_recvmsg_out) {
747 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
748 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
749 };
750
751 hdr.msg.payloadlen = err;
752 if (err > kmsg->payloadlen)
753 err = kmsg->payloadlen;
754
755 copy_len = sizeof(struct io_uring_recvmsg_out);
756 if (kmsg->msg.msg_namelen > kmsg->namelen)
757 copy_len += kmsg->namelen;
758 else
759 copy_len += kmsg->msg.msg_namelen;
760
761 /*
762 * "fromlen shall refer to the value before truncation.."
763 * 1003.1g
764 */
765 hdr.msg.namelen = kmsg->msg.msg_namelen;
766
767 /* ensure that there is no gap between hdr and sockaddr_storage */
768 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
769 sizeof(struct io_uring_recvmsg_out));
770 if (copy_to_user(io->buf, &hdr, copy_len)) {
771 *finished = true;
772 return -EFAULT;
773 }
774
775 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
776 kmsg->controllen + err;
777}
778
779int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
780{
781 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
782 struct io_async_msghdr iomsg, *kmsg;
783 struct socket *sock;
784 unsigned flags;
785 int ret, min_ret = 0;
786 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
787 bool mshot_finished = true;
788
789 sock = sock_from_file(req->file);
790 if (unlikely(!sock))
791 return -ENOTSOCK;
792
793 if (req_has_async_data(req)) {
794 kmsg = req->async_data;
795 } else {
796 ret = io_recvmsg_copy_hdr(req, &iomsg);
797 if (ret)
798 return ret;
799 kmsg = &iomsg;
800 }
801
802 if (!(req->flags & REQ_F_POLLED) &&
803 (sr->flags & IORING_RECVSEND_POLL_FIRST))
804 return io_setup_async_msg(req, kmsg, issue_flags);
805
806 if (!io_check_multishot(req, issue_flags))
807 return io_setup_async_msg(req, kmsg, issue_flags);
808
809retry_multishot:
810 if (io_do_buffer_select(req)) {
811 void __user *buf;
812 size_t len = sr->len;
813
814 buf = io_buffer_select(req, &len, issue_flags);
815 if (!buf)
816 return -ENOBUFS;
817
818 if (req->flags & REQ_F_APOLL_MULTISHOT) {
819 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
820 if (ret) {
821 io_kbuf_recycle(req, issue_flags);
822 return ret;
823 }
824 }
825
826 iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_DEST, buf, len);
827 }
828
829 flags = sr->msg_flags;
830 if (force_nonblock)
831 flags |= MSG_DONTWAIT;
832
833 kmsg->msg.msg_get_inq = 1;
834 kmsg->msg.msg_inq = -1;
835 if (req->flags & REQ_F_APOLL_MULTISHOT) {
836 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
837 &mshot_finished);
838 } else {
839 /* disable partial retry for recvmsg with cmsg attached */
840 if (flags & MSG_WAITALL && !kmsg->msg.msg_controllen)
841 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
842
843 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
844 kmsg->uaddr, flags);
845 }
846
847 if (ret < min_ret) {
848 if (ret == -EAGAIN && force_nonblock) {
849 ret = io_setup_async_msg(req, kmsg, issue_flags);
850 if (ret == -EAGAIN && (issue_flags & IO_URING_F_MULTISHOT)) {
851 io_kbuf_recycle(req, issue_flags);
852 return IOU_ISSUE_SKIP_COMPLETE;
853 }
854 return ret;
855 }
856 if (ret > 0 && io_net_retry(sock, flags)) {
857 sr->done_io += ret;
858 req->flags |= REQ_F_PARTIAL_IO;
859 return io_setup_async_msg(req, kmsg, issue_flags);
860 }
861 if (ret == -ERESTARTSYS)
862 ret = -EINTR;
863 req_set_fail(req);
864 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
865 req_set_fail(req);
866 }
867
868 if (ret > 0)
869 ret += sr->done_io;
870 else if (sr->done_io)
871 ret = sr->done_io;
872 else
873 io_kbuf_recycle(req, issue_flags);
874
875 if (!io_recv_finish(req, &ret, &kmsg->msg, mshot_finished, issue_flags))
876 goto retry_multishot;
877
878 if (mshot_finished) {
879 /* fast path, check for non-NULL to avoid function call */
880 if (kmsg->free_iov)
881 kfree(kmsg->free_iov);
882 io_netmsg_recycle(req, issue_flags);
883 req->flags &= ~REQ_F_NEED_CLEANUP;
884 }
885
886 return ret;
887}
888
889int io_recv(struct io_kiocb *req, unsigned int issue_flags)
890{
891 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
892 struct msghdr msg;
893 struct socket *sock;
894 unsigned flags;
895 int ret, min_ret = 0;
896 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
897 size_t len = sr->len;
898
899 if (!(req->flags & REQ_F_POLLED) &&
900 (sr->flags & IORING_RECVSEND_POLL_FIRST))
901 return -EAGAIN;
902
903 if (!io_check_multishot(req, issue_flags))
904 return -EAGAIN;
905
906 sock = sock_from_file(req->file);
907 if (unlikely(!sock))
908 return -ENOTSOCK;
909
910 msg.msg_name = NULL;
911 msg.msg_namelen = 0;
912 msg.msg_control = NULL;
913 msg.msg_get_inq = 1;
914 msg.msg_controllen = 0;
915 msg.msg_iocb = NULL;
916 msg.msg_ubuf = NULL;
917
918retry_multishot:
919 if (io_do_buffer_select(req)) {
920 void __user *buf;
921
922 buf = io_buffer_select(req, &len, issue_flags);
923 if (!buf)
924 return -ENOBUFS;
925 sr->buf = buf;
926 sr->len = len;
927 }
928
929 ret = import_ubuf(ITER_DEST, sr->buf, len, &msg.msg_iter);
930 if (unlikely(ret))
931 goto out_free;
932
933 msg.msg_inq = -1;
934 msg.msg_flags = 0;
935
936 flags = sr->msg_flags;
937 if (force_nonblock)
938 flags |= MSG_DONTWAIT;
939 if (flags & MSG_WAITALL)
940 min_ret = iov_iter_count(&msg.msg_iter);
941
942 ret = sock_recvmsg(sock, &msg, flags);
943 if (ret < min_ret) {
944 if (ret == -EAGAIN && force_nonblock) {
945 if (issue_flags & IO_URING_F_MULTISHOT) {
946 io_kbuf_recycle(req, issue_flags);
947 return IOU_ISSUE_SKIP_COMPLETE;
948 }
949
950 return -EAGAIN;
951 }
952 if (ret > 0 && io_net_retry(sock, flags)) {
953 sr->len -= ret;
954 sr->buf += ret;
955 sr->done_io += ret;
956 req->flags |= REQ_F_PARTIAL_IO;
957 return -EAGAIN;
958 }
959 if (ret == -ERESTARTSYS)
960 ret = -EINTR;
961 req_set_fail(req);
962 } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
963out_free:
964 req_set_fail(req);
965 }
966
967 if (ret > 0)
968 ret += sr->done_io;
969 else if (sr->done_io)
970 ret = sr->done_io;
971 else
972 io_kbuf_recycle(req, issue_flags);
973
974 if (!io_recv_finish(req, &ret, &msg, ret <= 0, issue_flags))
975 goto retry_multishot;
976
977 return ret;
978}
979
980void io_send_zc_cleanup(struct io_kiocb *req)
981{
982 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
983 struct io_async_msghdr *io;
984
985 if (req_has_async_data(req)) {
986 io = req->async_data;
987 /* might be ->fast_iov if *msg_copy_hdr failed */
988 if (io->free_iov != io->fast_iov)
989 kfree(io->free_iov);
990 }
991 if (zc->notif) {
992 io_notif_flush(zc->notif);
993 zc->notif = NULL;
994 }
995}
996
997#define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF)
998#define IO_ZC_FLAGS_VALID (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE)
999
1000int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1001{
1002 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1003 struct io_ring_ctx *ctx = req->ctx;
1004 struct io_kiocb *notif;
1005
1006 if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
1007 return -EINVAL;
1008 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
1009 if (req->flags & REQ_F_CQE_SKIP)
1010 return -EINVAL;
1011
1012 notif = zc->notif = io_alloc_notif(ctx);
1013 if (!notif)
1014 return -ENOMEM;
1015 notif->cqe.user_data = req->cqe.user_data;
1016 notif->cqe.res = 0;
1017 notif->cqe.flags = IORING_CQE_F_NOTIF;
1018 req->flags |= REQ_F_NEED_CLEANUP;
1019
1020 zc->flags = READ_ONCE(sqe->ioprio);
1021 if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) {
1022 if (zc->flags & ~IO_ZC_FLAGS_VALID)
1023 return -EINVAL;
1024 if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) {
1025 io_notif_set_extended(notif);
1026 io_notif_to_data(notif)->zc_report = true;
1027 }
1028 }
1029
1030 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1031 unsigned idx = READ_ONCE(sqe->buf_index);
1032
1033 if (unlikely(idx >= ctx->nr_user_bufs))
1034 return -EFAULT;
1035 idx = array_index_nospec(idx, ctx->nr_user_bufs);
1036 req->imu = READ_ONCE(ctx->user_bufs[idx]);
1037 io_req_set_rsrc_node(notif, ctx, 0);
1038 }
1039
1040 if (req->opcode == IORING_OP_SEND_ZC) {
1041 if (READ_ONCE(sqe->__pad3[0]))
1042 return -EINVAL;
1043 zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1044 zc->addr_len = READ_ONCE(sqe->addr_len);
1045 } else {
1046 if (unlikely(sqe->addr2 || sqe->file_index))
1047 return -EINVAL;
1048 if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
1049 return -EINVAL;
1050 }
1051
1052 zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
1053 zc->len = READ_ONCE(sqe->len);
1054 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
1055 if (zc->msg_flags & MSG_DONTWAIT)
1056 req->flags |= REQ_F_NOWAIT;
1057
1058 zc->done_io = 0;
1059
1060#ifdef CONFIG_COMPAT
1061 if (req->ctx->compat)
1062 zc->msg_flags |= MSG_CMSG_COMPAT;
1063#endif
1064 return 0;
1065}
1066
1067static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
1068 struct iov_iter *from, size_t length)
1069{
1070 skb_zcopy_downgrade_managed(skb);
1071 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1072}
1073
1074static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1075 struct iov_iter *from, size_t length)
1076{
1077 struct skb_shared_info *shinfo = skb_shinfo(skb);
1078 int frag = shinfo->nr_frags;
1079 int ret = 0;
1080 struct bvec_iter bi;
1081 ssize_t copied = 0;
1082 unsigned long truesize = 0;
1083
1084 if (!frag)
1085 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
1086 else if (unlikely(!skb_zcopy_managed(skb)))
1087 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1088
1089 bi.bi_size = min(from->count, length);
1090 bi.bi_bvec_done = from->iov_offset;
1091 bi.bi_idx = 0;
1092
1093 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1094 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1095
1096 copied += v.bv_len;
1097 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1098 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1099 v.bv_offset, v.bv_len);
1100 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1101 }
1102 if (bi.bi_size)
1103 ret = -EMSGSIZE;
1104
1105 shinfo->nr_frags = frag;
1106 from->bvec += bi.bi_idx;
1107 from->nr_segs -= bi.bi_idx;
1108 from->count -= copied;
1109 from->iov_offset = bi.bi_bvec_done;
1110
1111 skb->data_len += copied;
1112 skb->len += copied;
1113 skb->truesize += truesize;
1114
1115 if (sk && sk->sk_type == SOCK_STREAM) {
1116 sk_wmem_queued_add(sk, truesize);
1117 if (!skb_zcopy_pure(skb))
1118 sk_mem_charge(sk, truesize);
1119 } else {
1120 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1121 }
1122 return ret;
1123}
1124
1125int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
1126{
1127 struct sockaddr_storage __address;
1128 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1129 struct msghdr msg;
1130 struct socket *sock;
1131 unsigned msg_flags;
1132 int ret, min_ret = 0;
1133
1134 sock = sock_from_file(req->file);
1135 if (unlikely(!sock))
1136 return -ENOTSOCK;
1137 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1138 return -EOPNOTSUPP;
1139
1140 msg.msg_name = NULL;
1141 msg.msg_control = NULL;
1142 msg.msg_controllen = 0;
1143 msg.msg_namelen = 0;
1144
1145 if (zc->addr) {
1146 if (req_has_async_data(req)) {
1147 struct io_async_msghdr *io = req->async_data;
1148
1149 msg.msg_name = &io->addr;
1150 } else {
1151 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1152 if (unlikely(ret < 0))
1153 return ret;
1154 msg.msg_name = (struct sockaddr *)&__address;
1155 }
1156 msg.msg_namelen = zc->addr_len;
1157 }
1158
1159 if (!(req->flags & REQ_F_POLLED) &&
1160 (zc->flags & IORING_RECVSEND_POLL_FIRST))
1161 return io_setup_async_addr(req, &__address, issue_flags);
1162
1163 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1164 ret = io_import_fixed(ITER_SOURCE, &msg.msg_iter, req->imu,
1165 (u64)(uintptr_t)zc->buf, zc->len);
1166 if (unlikely(ret))
1167 return ret;
1168 msg.sg_from_iter = io_sg_from_iter;
1169 } else {
1170 io_notif_set_extended(zc->notif);
1171 ret = import_ubuf(ITER_SOURCE, zc->buf, zc->len, &msg.msg_iter);
1172 if (unlikely(ret))
1173 return ret;
1174 ret = io_notif_account_mem(zc->notif, zc->len);
1175 if (unlikely(ret))
1176 return ret;
1177 msg.sg_from_iter = io_sg_from_iter_iovec;
1178 }
1179
1180 msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1181 if (issue_flags & IO_URING_F_NONBLOCK)
1182 msg_flags |= MSG_DONTWAIT;
1183 if (msg_flags & MSG_WAITALL)
1184 min_ret = iov_iter_count(&msg.msg_iter);
1185 msg_flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
1186
1187 msg.msg_flags = msg_flags;
1188 msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
1189 ret = sock_sendmsg(sock, &msg);
1190
1191 if (unlikely(ret < min_ret)) {
1192 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1193 return io_setup_async_addr(req, &__address, issue_flags);
1194
1195 if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1196 zc->len -= ret;
1197 zc->buf += ret;
1198 zc->done_io += ret;
1199 req->flags |= REQ_F_PARTIAL_IO;
1200 return io_setup_async_addr(req, &__address, issue_flags);
1201 }
1202 if (ret == -ERESTARTSYS)
1203 ret = -EINTR;
1204 req_set_fail(req);
1205 }
1206
1207 if (ret >= 0)
1208 ret += zc->done_io;
1209 else if (zc->done_io)
1210 ret = zc->done_io;
1211
1212 /*
1213 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1214 * flushing notif to io_send_zc_cleanup()
1215 */
1216 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1217 io_notif_flush(zc->notif);
1218 req->flags &= ~REQ_F_NEED_CLEANUP;
1219 }
1220 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1221 return IOU_OK;
1222}
1223
1224int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1225{
1226 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1227 struct io_async_msghdr iomsg, *kmsg;
1228 struct socket *sock;
1229 unsigned flags;
1230 int ret, min_ret = 0;
1231
1232 io_notif_set_extended(sr->notif);
1233
1234 sock = sock_from_file(req->file);
1235 if (unlikely(!sock))
1236 return -ENOTSOCK;
1237 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1238 return -EOPNOTSUPP;
1239
1240 if (req_has_async_data(req)) {
1241 kmsg = req->async_data;
1242 } else {
1243 ret = io_sendmsg_copy_hdr(req, &iomsg);
1244 if (ret)
1245 return ret;
1246 kmsg = &iomsg;
1247 }
1248
1249 if (!(req->flags & REQ_F_POLLED) &&
1250 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1251 return io_setup_async_msg(req, kmsg, issue_flags);
1252
1253 flags = sr->msg_flags | MSG_ZEROCOPY;
1254 if (issue_flags & IO_URING_F_NONBLOCK)
1255 flags |= MSG_DONTWAIT;
1256 if (flags & MSG_WAITALL)
1257 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1258
1259 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1260 kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1261 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1262
1263 if (unlikely(ret < min_ret)) {
1264 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1265 return io_setup_async_msg(req, kmsg, issue_flags);
1266
1267 if (ret > 0 && io_net_retry(sock, flags)) {
1268 sr->done_io += ret;
1269 req->flags |= REQ_F_PARTIAL_IO;
1270 return io_setup_async_msg(req, kmsg, issue_flags);
1271 }
1272 if (ret == -ERESTARTSYS)
1273 ret = -EINTR;
1274 req_set_fail(req);
1275 }
1276 /* fast path, check for non-NULL to avoid function call */
1277 if (kmsg->free_iov) {
1278 kfree(kmsg->free_iov);
1279 kmsg->free_iov = NULL;
1280 }
1281
1282 io_netmsg_recycle(req, issue_flags);
1283 if (ret >= 0)
1284 ret += sr->done_io;
1285 else if (sr->done_io)
1286 ret = sr->done_io;
1287
1288 /*
1289 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1290 * flushing notif to io_send_zc_cleanup()
1291 */
1292 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1293 io_notif_flush(sr->notif);
1294 req->flags &= ~REQ_F_NEED_CLEANUP;
1295 }
1296 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1297 return IOU_OK;
1298}
1299
1300void io_sendrecv_fail(struct io_kiocb *req)
1301{
1302 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1303
1304 if (req->flags & REQ_F_PARTIAL_IO)
1305 req->cqe.res = sr->done_io;
1306
1307 if ((req->flags & REQ_F_NEED_CLEANUP) &&
1308 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1309 req->cqe.flags |= IORING_CQE_F_MORE;
1310}
1311
1312int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1313{
1314 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1315 unsigned flags;
1316
1317 if (sqe->len || sqe->buf_index)
1318 return -EINVAL;
1319
1320 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1321 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1322 accept->flags = READ_ONCE(sqe->accept_flags);
1323 accept->nofile = rlimit(RLIMIT_NOFILE);
1324 flags = READ_ONCE(sqe->ioprio);
1325 if (flags & ~IORING_ACCEPT_MULTISHOT)
1326 return -EINVAL;
1327
1328 accept->file_slot = READ_ONCE(sqe->file_index);
1329 if (accept->file_slot) {
1330 if (accept->flags & SOCK_CLOEXEC)
1331 return -EINVAL;
1332 if (flags & IORING_ACCEPT_MULTISHOT &&
1333 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1334 return -EINVAL;
1335 }
1336 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1337 return -EINVAL;
1338 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1339 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1340 if (flags & IORING_ACCEPT_MULTISHOT)
1341 req->flags |= REQ_F_APOLL_MULTISHOT;
1342 return 0;
1343}
1344
1345int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1346{
1347 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1348 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1349 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1350 bool fixed = !!accept->file_slot;
1351 struct file *file;
1352 int ret, fd;
1353
1354 if (!io_check_multishot(req, issue_flags))
1355 return -EAGAIN;
1356retry:
1357 if (!fixed) {
1358 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1359 if (unlikely(fd < 0))
1360 return fd;
1361 }
1362 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1363 accept->flags);
1364 if (IS_ERR(file)) {
1365 if (!fixed)
1366 put_unused_fd(fd);
1367 ret = PTR_ERR(file);
1368 if (ret == -EAGAIN && force_nonblock) {
1369 /*
1370 * if it's multishot and polled, we don't need to
1371 * return EAGAIN to arm the poll infra since it
1372 * has already been done
1373 */
1374 if (issue_flags & IO_URING_F_MULTISHOT)
1375 return IOU_ISSUE_SKIP_COMPLETE;
1376 return ret;
1377 }
1378 if (ret == -ERESTARTSYS)
1379 ret = -EINTR;
1380 req_set_fail(req);
1381 } else if (!fixed) {
1382 fd_install(fd, file);
1383 ret = fd;
1384 } else {
1385 ret = io_fixed_fd_install(req, issue_flags, file,
1386 accept->file_slot);
1387 }
1388
1389 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1390 io_req_set_res(req, ret, 0);
1391 return IOU_OK;
1392 }
1393
1394 if (ret < 0)
1395 return ret;
1396 if (io_fill_cqe_req_aux(req, issue_flags & IO_URING_F_COMPLETE_DEFER,
1397 ret, IORING_CQE_F_MORE))
1398 goto retry;
1399
1400 io_req_set_res(req, ret, 0);
1401 return IOU_STOP_MULTISHOT;
1402}
1403
1404int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1405{
1406 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1407
1408 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1409 return -EINVAL;
1410
1411 sock->domain = READ_ONCE(sqe->fd);
1412 sock->type = READ_ONCE(sqe->off);
1413 sock->protocol = READ_ONCE(sqe->len);
1414 sock->file_slot = READ_ONCE(sqe->file_index);
1415 sock->nofile = rlimit(RLIMIT_NOFILE);
1416
1417 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1418 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1419 return -EINVAL;
1420 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1421 return -EINVAL;
1422 return 0;
1423}
1424
1425int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1426{
1427 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1428 bool fixed = !!sock->file_slot;
1429 struct file *file;
1430 int ret, fd;
1431
1432 if (!fixed) {
1433 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1434 if (unlikely(fd < 0))
1435 return fd;
1436 }
1437 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1438 if (IS_ERR(file)) {
1439 if (!fixed)
1440 put_unused_fd(fd);
1441 ret = PTR_ERR(file);
1442 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1443 return -EAGAIN;
1444 if (ret == -ERESTARTSYS)
1445 ret = -EINTR;
1446 req_set_fail(req);
1447 } else if (!fixed) {
1448 fd_install(fd, file);
1449 ret = fd;
1450 } else {
1451 ret = io_fixed_fd_install(req, issue_flags, file,
1452 sock->file_slot);
1453 }
1454 io_req_set_res(req, ret, 0);
1455 return IOU_OK;
1456}
1457
1458int io_connect_prep_async(struct io_kiocb *req)
1459{
1460 struct io_async_connect *io = req->async_data;
1461 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1462
1463 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1464}
1465
1466int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1467{
1468 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1469
1470 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1471 return -EINVAL;
1472
1473 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1474 conn->addr_len = READ_ONCE(sqe->addr2);
1475 conn->in_progress = conn->seen_econnaborted = false;
1476 return 0;
1477}
1478
1479int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1480{
1481 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1482 struct io_async_connect __io, *io;
1483 unsigned file_flags;
1484 int ret;
1485 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1486
1487 if (req_has_async_data(req)) {
1488 io = req->async_data;
1489 } else {
1490 ret = move_addr_to_kernel(connect->addr,
1491 connect->addr_len,
1492 &__io.address);
1493 if (ret)
1494 goto out;
1495 io = &__io;
1496 }
1497
1498 file_flags = force_nonblock ? O_NONBLOCK : 0;
1499
1500 ret = __sys_connect_file(req->file, &io->address,
1501 connect->addr_len, file_flags);
1502 if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED)
1503 && force_nonblock) {
1504 if (ret == -EINPROGRESS) {
1505 connect->in_progress = true;
1506 } else if (ret == -ECONNABORTED) {
1507 if (connect->seen_econnaborted)
1508 goto out;
1509 connect->seen_econnaborted = true;
1510 }
1511 if (req_has_async_data(req))
1512 return -EAGAIN;
1513 if (io_alloc_async_data(req)) {
1514 ret = -ENOMEM;
1515 goto out;
1516 }
1517 memcpy(req->async_data, &__io, sizeof(__io));
1518 return -EAGAIN;
1519 }
1520 if (connect->in_progress) {
1521 /*
1522 * At least bluetooth will return -EBADFD on a re-connect
1523 * attempt, and it's (supposedly) also valid to get -EISCONN
1524 * which means the previous result is good. For both of these,
1525 * grab the sock_error() and use that for the completion.
1526 */
1527 if (ret == -EBADFD || ret == -EISCONN)
1528 ret = sock_error(sock_from_file(req->file)->sk);
1529 }
1530 if (ret == -ERESTARTSYS)
1531 ret = -EINTR;
1532out:
1533 if (ret < 0)
1534 req_set_fail(req);
1535 io_req_set_res(req, ret, 0);
1536 return IOU_OK;
1537}
1538
1539void io_netmsg_cache_free(struct io_cache_entry *entry)
1540{
1541 kfree(container_of(entry, struct io_async_msghdr, cache));
1542}
1543#endif
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/net.h>
7#include <linux/compat.h>
8#include <net/compat.h>
9#include <linux/io_uring.h>
10
11#include <uapi/linux/io_uring.h>
12
13#include "io_uring.h"
14#include "kbuf.h"
15#include "alloc_cache.h"
16#include "net.h"
17#include "notif.h"
18#include "rsrc.h"
19
20#if defined(CONFIG_NET)
21struct io_shutdown {
22 struct file *file;
23 int how;
24};
25
26struct io_accept {
27 struct file *file;
28 struct sockaddr __user *addr;
29 int __user *addr_len;
30 int flags;
31 u32 file_slot;
32 unsigned long nofile;
33};
34
35struct io_socket {
36 struct file *file;
37 int domain;
38 int type;
39 int protocol;
40 int flags;
41 u32 file_slot;
42 unsigned long nofile;
43};
44
45struct io_connect {
46 struct file *file;
47 struct sockaddr __user *addr;
48 int addr_len;
49 bool in_progress;
50 bool seen_econnaborted;
51};
52
53struct io_sr_msg {
54 struct file *file;
55 union {
56 struct compat_msghdr __user *umsg_compat;
57 struct user_msghdr __user *umsg;
58 void __user *buf;
59 };
60 unsigned len;
61 unsigned done_io;
62 unsigned msg_flags;
63 unsigned nr_multishot_loops;
64 u16 flags;
65 /* initialised and used only by !msg send variants */
66 u16 addr_len;
67 u16 buf_group;
68 void __user *addr;
69 void __user *msg_control;
70 /* used only for send zerocopy */
71 struct io_kiocb *notif;
72};
73
74/*
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.
78 */
79#define MULTISHOT_MAX_RETRY 32
80
81int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
82{
83 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
84
85 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
86 sqe->buf_index || sqe->splice_fd_in))
87 return -EINVAL;
88
89 shutdown->how = READ_ONCE(sqe->len);
90 req->flags |= REQ_F_FORCE_ASYNC;
91 return 0;
92}
93
94int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
95{
96 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
97 struct socket *sock;
98 int ret;
99
100 WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
101
102 sock = sock_from_file(req->file);
103 if (unlikely(!sock))
104 return -ENOTSOCK;
105
106 ret = __sys_shutdown_sock(sock, shutdown->how);
107 io_req_set_res(req, ret, 0);
108 return IOU_OK;
109}
110
111static bool io_net_retry(struct socket *sock, int flags)
112{
113 if (!(flags & MSG_WAITALL))
114 return false;
115 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
116}
117
118static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
119{
120 struct io_async_msghdr *hdr = req->async_data;
121
122 if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
123 return;
124
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;
129 }
130}
131
132static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
133 unsigned int issue_flags)
134{
135 struct io_ring_ctx *ctx = req->ctx;
136 struct io_cache_entry *entry;
137 struct io_async_msghdr *hdr;
138
139 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
140 entry = io_alloc_cache_get(&ctx->netmsg_cache);
141 if (entry) {
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;
146 return hdr;
147 }
148 }
149
150 if (!io_alloc_async_data(req)) {
151 hdr = req->async_data;
152 hdr->free_iov = NULL;
153 return hdr;
154 }
155 return NULL;
156}
157
158static inline struct io_async_msghdr *io_msg_alloc_async_prep(struct io_kiocb *req)
159{
160 /* ->prep_async is always called from the submission context */
161 return io_msg_alloc_async(req, 0);
162}
163
164static int io_setup_async_msg(struct io_kiocb *req,
165 struct io_async_msghdr *kmsg,
166 unsigned int issue_flags)
167{
168 struct io_async_msghdr *async_msg;
169
170 if (req_has_async_data(req))
171 return -EAGAIN;
172 async_msg = io_msg_alloc_async(req, issue_flags);
173 if (!async_msg) {
174 kfree(kmsg->free_iov);
175 return -ENOMEM;
176 }
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;
181
182 if ((req->flags & REQ_F_BUFFER_SELECT) && !async_msg->msg.msg_iter.nr_segs)
183 return -EAGAIN;
184
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];
189 }
190
191 return -EAGAIN;
192}
193
194#ifdef CONFIG_COMPAT
195static int io_compat_msg_copy_hdr(struct io_kiocb *req,
196 struct io_async_msghdr *iomsg,
197 struct compat_msghdr *msg, int ddir)
198{
199 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
200 struct compat_iovec __user *uiov;
201 int ret;
202
203 if (copy_from_user(msg, sr->umsg_compat, sizeof(*msg)))
204 return -EFAULT;
205
206 uiov = compat_ptr(msg->msg_iov);
207 if (req->flags & REQ_F_BUFFER_SELECT) {
208 compat_ssize_t clen;
209
210 iomsg->free_iov = NULL;
211 if (msg->msg_iovlen == 0) {
212 sr->len = 0;
213 } else if (msg->msg_iovlen > 1) {
214 return -EINVAL;
215 } else {
216 if (!access_ok(uiov, sizeof(*uiov)))
217 return -EFAULT;
218 if (__get_user(clen, &uiov->iov_len))
219 return -EFAULT;
220 if (clen < 0)
221 return -EINVAL;
222 sr->len = clen;
223 }
224
225 return 0;
226 }
227
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))
233 return ret;
234
235 return 0;
236}
237#endif
238
239static int io_msg_copy_hdr(struct io_kiocb *req, struct io_async_msghdr *iomsg,
240 struct user_msghdr *msg, int ddir)
241{
242 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
243 int ret;
244
245 if (!user_access_begin(sr->umsg, sizeof(*sr->umsg)))
246 return -EFAULT;
247
248 ret = -EFAULT;
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);
255 msg->msg_flags = 0;
256
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) {
263 ret = -EINVAL;
264 goto ua_end;
265 } else {
266 /* we only need the length for provided buffers */
267 if (!access_ok(&msg->msg_iov[0].iov_len, sizeof(__kernel_size_t)))
268 goto ua_end;
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;
273 }
274 ret = 0;
275ua_end:
276 user_access_end();
277 return ret;
278 }
279
280 user_access_end();
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))
285 return ret;
286
287 return 0;
288}
289
290static int io_sendmsg_copy_hdr(struct io_kiocb *req,
291 struct io_async_msghdr *iomsg)
292{
293 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
294 struct user_msghdr msg;
295 int ret;
296
297 iomsg->msg.msg_name = &iomsg->addr;
298 iomsg->msg.msg_iter.nr_segs = 0;
299
300#ifdef CONFIG_COMPAT
301 if (unlikely(req->ctx->compat)) {
302 struct compat_msghdr cmsg;
303
304 ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ITER_SOURCE);
305 if (unlikely(ret))
306 return ret;
307
308 return __get_compat_msghdr(&iomsg->msg, &cmsg, NULL);
309 }
310#endif
311
312 ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_SOURCE);
313 if (unlikely(ret))
314 return ret;
315
316 ret = __copy_msghdr(&iomsg->msg, &msg, NULL);
317
318 /* save msg_control as sys_sendmsg() overwrites it */
319 sr->msg_control = iomsg->msg.msg_control_user;
320 return ret;
321}
322
323int io_send_prep_async(struct io_kiocb *req)
324{
325 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
326 struct io_async_msghdr *io;
327 int ret;
328
329 if (req_has_async_data(req))
330 return 0;
331 zc->done_io = 0;
332 if (!zc->addr)
333 return 0;
334 io = io_msg_alloc_async_prep(req);
335 if (!io)
336 return -ENOMEM;
337 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
338 return ret;
339}
340
341static int io_setup_async_addr(struct io_kiocb *req,
342 struct sockaddr_storage *addr_storage,
343 unsigned int issue_flags)
344{
345 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
346 struct io_async_msghdr *io;
347
348 if (!sr->addr || req_has_async_data(req))
349 return -EAGAIN;
350 io = io_msg_alloc_async(req, issue_flags);
351 if (!io)
352 return -ENOMEM;
353 memcpy(&io->addr, addr_storage, sizeof(io->addr));
354 return -EAGAIN;
355}
356
357int io_sendmsg_prep_async(struct io_kiocb *req)
358{
359 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
360 int ret;
361
362 sr->done_io = 0;
363 if (!io_msg_alloc_async_prep(req))
364 return -ENOMEM;
365 ret = io_sendmsg_copy_hdr(req, req->async_data);
366 if (!ret)
367 req->flags |= REQ_F_NEED_CLEANUP;
368 return ret;
369}
370
371void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
372{
373 struct io_async_msghdr *io = req->async_data;
374
375 kfree(io->free_iov);
376}
377
378int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
379{
380 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
381
382 sr->done_io = 0;
383
384 if (req->opcode == IORING_OP_SEND) {
385 if (READ_ONCE(sqe->__pad3[0]))
386 return -EINVAL;
387 sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
388 sr->addr_len = READ_ONCE(sqe->addr_len);
389 } else if (sqe->addr2 || sqe->file_index) {
390 return -EINVAL;
391 }
392
393 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
394 sr->len = READ_ONCE(sqe->len);
395 sr->flags = READ_ONCE(sqe->ioprio);
396 if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
397 return -EINVAL;
398 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
399 if (sr->msg_flags & MSG_DONTWAIT)
400 req->flags |= REQ_F_NOWAIT;
401
402#ifdef CONFIG_COMPAT
403 if (req->ctx->compat)
404 sr->msg_flags |= MSG_CMSG_COMPAT;
405#endif
406 return 0;
407}
408
409static void io_req_msg_cleanup(struct io_kiocb *req,
410 struct io_async_msghdr *kmsg,
411 unsigned int issue_flags)
412{
413 req->flags &= ~REQ_F_NEED_CLEANUP;
414 /* fast path, check for non-NULL to avoid function call */
415 if (kmsg->free_iov)
416 kfree(kmsg->free_iov);
417 io_netmsg_recycle(req, issue_flags);
418}
419
420int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
421{
422 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
423 struct io_async_msghdr iomsg, *kmsg;
424 struct socket *sock;
425 unsigned flags;
426 int min_ret = 0;
427 int ret;
428
429 sock = sock_from_file(req->file);
430 if (unlikely(!sock))
431 return -ENOTSOCK;
432
433 if (req_has_async_data(req)) {
434 kmsg = req->async_data;
435 kmsg->msg.msg_control_user = sr->msg_control;
436 } else {
437 ret = io_sendmsg_copy_hdr(req, &iomsg);
438 if (ret)
439 return ret;
440 kmsg = &iomsg;
441 }
442
443 if (!(req->flags & REQ_F_POLLED) &&
444 (sr->flags & IORING_RECVSEND_POLL_FIRST))
445 return io_setup_async_msg(req, kmsg, issue_flags);
446
447 flags = sr->msg_flags;
448 if (issue_flags & IO_URING_F_NONBLOCK)
449 flags |= MSG_DONTWAIT;
450 if (flags & MSG_WAITALL)
451 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
452
453 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
454
455 if (ret < min_ret) {
456 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
457 return io_setup_async_msg(req, kmsg, issue_flags);
458 if (ret > 0 && io_net_retry(sock, flags)) {
459 kmsg->msg.msg_controllen = 0;
460 kmsg->msg.msg_control = NULL;
461 sr->done_io += ret;
462 req->flags |= REQ_F_BL_NO_RECYCLE;
463 return io_setup_async_msg(req, kmsg, issue_flags);
464 }
465 if (ret == -ERESTARTSYS)
466 ret = -EINTR;
467 req_set_fail(req);
468 }
469 io_req_msg_cleanup(req, kmsg, issue_flags);
470 if (ret >= 0)
471 ret += sr->done_io;
472 else if (sr->done_io)
473 ret = sr->done_io;
474 io_req_set_res(req, ret, 0);
475 return IOU_OK;
476}
477
478int io_send(struct io_kiocb *req, unsigned int issue_flags)
479{
480 struct sockaddr_storage __address;
481 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
482 struct msghdr msg;
483 struct socket *sock;
484 unsigned flags;
485 int min_ret = 0;
486 int ret;
487
488 msg.msg_name = NULL;
489 msg.msg_control = NULL;
490 msg.msg_controllen = 0;
491 msg.msg_namelen = 0;
492 msg.msg_ubuf = NULL;
493
494 if (sr->addr) {
495 if (req_has_async_data(req)) {
496 struct io_async_msghdr *io = req->async_data;
497
498 msg.msg_name = &io->addr;
499 } else {
500 ret = move_addr_to_kernel(sr->addr, sr->addr_len, &__address);
501 if (unlikely(ret < 0))
502 return ret;
503 msg.msg_name = (struct sockaddr *)&__address;
504 }
505 msg.msg_namelen = sr->addr_len;
506 }
507
508 if (!(req->flags & REQ_F_POLLED) &&
509 (sr->flags & IORING_RECVSEND_POLL_FIRST))
510 return io_setup_async_addr(req, &__address, issue_flags);
511
512 sock = sock_from_file(req->file);
513 if (unlikely(!sock))
514 return -ENOTSOCK;
515
516 ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len, &msg.msg_iter);
517 if (unlikely(ret))
518 return ret;
519
520 flags = sr->msg_flags;
521 if (issue_flags & IO_URING_F_NONBLOCK)
522 flags |= MSG_DONTWAIT;
523 if (flags & MSG_WAITALL)
524 min_ret = iov_iter_count(&msg.msg_iter);
525
526 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
527 msg.msg_flags = flags;
528 ret = sock_sendmsg(sock, &msg);
529 if (ret < min_ret) {
530 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
531 return io_setup_async_addr(req, &__address, issue_flags);
532
533 if (ret > 0 && io_net_retry(sock, flags)) {
534 sr->len -= ret;
535 sr->buf += ret;
536 sr->done_io += ret;
537 req->flags |= REQ_F_BL_NO_RECYCLE;
538 return io_setup_async_addr(req, &__address, issue_flags);
539 }
540 if (ret == -ERESTARTSYS)
541 ret = -EINTR;
542 req_set_fail(req);
543 }
544 if (ret >= 0)
545 ret += sr->done_io;
546 else if (sr->done_io)
547 ret = sr->done_io;
548 io_req_set_res(req, ret, 0);
549 return IOU_OK;
550}
551
552static int io_recvmsg_mshot_prep(struct io_kiocb *req,
553 struct io_async_msghdr *iomsg,
554 int namelen, size_t controllen)
555{
556 if ((req->flags & (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) ==
557 (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) {
558 int hdr;
559
560 if (unlikely(namelen < 0))
561 return -EOVERFLOW;
562 if (check_add_overflow(sizeof(struct io_uring_recvmsg_out),
563 namelen, &hdr))
564 return -EOVERFLOW;
565 if (check_add_overflow(hdr, controllen, &hdr))
566 return -EOVERFLOW;
567
568 iomsg->namelen = namelen;
569 iomsg->controllen = controllen;
570 return 0;
571 }
572
573 return 0;
574}
575
576static int io_recvmsg_copy_hdr(struct io_kiocb *req,
577 struct io_async_msghdr *iomsg)
578{
579 struct user_msghdr msg;
580 int ret;
581
582 iomsg->msg.msg_name = &iomsg->addr;
583 iomsg->msg.msg_iter.nr_segs = 0;
584
585#ifdef CONFIG_COMPAT
586 if (unlikely(req->ctx->compat)) {
587 struct compat_msghdr cmsg;
588
589 ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ITER_DEST);
590 if (unlikely(ret))
591 return ret;
592
593 ret = __get_compat_msghdr(&iomsg->msg, &cmsg, &iomsg->uaddr);
594 if (unlikely(ret))
595 return ret;
596
597 return io_recvmsg_mshot_prep(req, iomsg, cmsg.msg_namelen,
598 cmsg.msg_controllen);
599 }
600#endif
601
602 ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_DEST);
603 if (unlikely(ret))
604 return ret;
605
606 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
607 if (unlikely(ret))
608 return ret;
609
610 return io_recvmsg_mshot_prep(req, iomsg, msg.msg_namelen,
611 msg.msg_controllen);
612}
613
614int io_recvmsg_prep_async(struct io_kiocb *req)
615{
616 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
617 struct io_async_msghdr *iomsg;
618 int ret;
619
620 sr->done_io = 0;
621 if (!io_msg_alloc_async_prep(req))
622 return -ENOMEM;
623 iomsg = req->async_data;
624 ret = io_recvmsg_copy_hdr(req, iomsg);
625 if (!ret)
626 req->flags |= REQ_F_NEED_CLEANUP;
627 return ret;
628}
629
630#define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
631
632int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
633{
634 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
635
636 sr->done_io = 0;
637
638 if (unlikely(sqe->file_index || sqe->addr2))
639 return -EINVAL;
640
641 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
642 sr->len = READ_ONCE(sqe->len);
643 sr->flags = READ_ONCE(sqe->ioprio);
644 if (sr->flags & ~(RECVMSG_FLAGS))
645 return -EINVAL;
646 sr->msg_flags = READ_ONCE(sqe->msg_flags);
647 if (sr->msg_flags & MSG_DONTWAIT)
648 req->flags |= REQ_F_NOWAIT;
649 if (sr->msg_flags & MSG_ERRQUEUE)
650 req->flags |= REQ_F_CLEAR_POLLIN;
651 if (sr->flags & IORING_RECV_MULTISHOT) {
652 if (!(req->flags & REQ_F_BUFFER_SELECT))
653 return -EINVAL;
654 if (sr->msg_flags & MSG_WAITALL)
655 return -EINVAL;
656 if (req->opcode == IORING_OP_RECV && sr->len)
657 return -EINVAL;
658 req->flags |= REQ_F_APOLL_MULTISHOT;
659 /*
660 * Store the buffer group for this multishot receive separately,
661 * as if we end up doing an io-wq based issue that selects a
662 * buffer, it has to be committed immediately and that will
663 * clear ->buf_list. This means we lose the link to the buffer
664 * list, and the eventual buffer put on completion then cannot
665 * restore it.
666 */
667 sr->buf_group = req->buf_index;
668 }
669
670#ifdef CONFIG_COMPAT
671 if (req->ctx->compat)
672 sr->msg_flags |= MSG_CMSG_COMPAT;
673#endif
674 sr->nr_multishot_loops = 0;
675 return 0;
676}
677
678static inline void io_recv_prep_retry(struct io_kiocb *req)
679{
680 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
681
682 req->flags &= ~REQ_F_BL_EMPTY;
683 sr->done_io = 0;
684 sr->len = 0; /* get from the provided buffer */
685 req->buf_index = sr->buf_group;
686}
687
688/*
689 * Finishes io_recv and io_recvmsg.
690 *
691 * Returns true if it is actually finished, or false if it should run
692 * again (for multishot).
693 */
694static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
695 struct msghdr *msg, bool mshot_finished,
696 unsigned issue_flags)
697{
698 unsigned int cflags;
699
700 cflags = io_put_kbuf(req, issue_flags);
701 if (msg->msg_inq > 0)
702 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
703
704 /*
705 * Fill CQE for this receive and see if we should keep trying to
706 * receive from this socket.
707 */
708 if ((req->flags & REQ_F_APOLL_MULTISHOT) && !mshot_finished &&
709 io_fill_cqe_req_aux(req, issue_flags & IO_URING_F_COMPLETE_DEFER,
710 *ret, cflags | IORING_CQE_F_MORE)) {
711 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
712 int mshot_retry_ret = IOU_ISSUE_SKIP_COMPLETE;
713
714 io_recv_prep_retry(req);
715 /* Known not-empty or unknown state, retry */
716 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || msg->msg_inq < 0) {
717 if (sr->nr_multishot_loops++ < MULTISHOT_MAX_RETRY)
718 return false;
719 /* mshot retries exceeded, force a requeue */
720 sr->nr_multishot_loops = 0;
721 mshot_retry_ret = IOU_REQUEUE;
722 }
723 if (issue_flags & IO_URING_F_MULTISHOT)
724 *ret = mshot_retry_ret;
725 else
726 *ret = -EAGAIN;
727 return true;
728 }
729
730 /* Finish the request / stop multishot. */
731 io_req_set_res(req, *ret, cflags);
732
733 if (issue_flags & IO_URING_F_MULTISHOT)
734 *ret = IOU_STOP_MULTISHOT;
735 else
736 *ret = IOU_OK;
737 return true;
738}
739
740static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
741 struct io_sr_msg *sr, void __user **buf,
742 size_t *len)
743{
744 unsigned long ubuf = (unsigned long) *buf;
745 unsigned long hdr;
746
747 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
748 kmsg->controllen;
749 if (*len < hdr)
750 return -EFAULT;
751
752 if (kmsg->controllen) {
753 unsigned long control = ubuf + hdr - kmsg->controllen;
754
755 kmsg->msg.msg_control_user = (void __user *) control;
756 kmsg->msg.msg_controllen = kmsg->controllen;
757 }
758
759 sr->buf = *buf; /* stash for later copy */
760 *buf = (void __user *) (ubuf + hdr);
761 kmsg->payloadlen = *len = *len - hdr;
762 return 0;
763}
764
765struct io_recvmsg_multishot_hdr {
766 struct io_uring_recvmsg_out msg;
767 struct sockaddr_storage addr;
768};
769
770static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
771 struct io_async_msghdr *kmsg,
772 unsigned int flags, bool *finished)
773{
774 int err;
775 int copy_len;
776 struct io_recvmsg_multishot_hdr hdr;
777
778 if (kmsg->namelen)
779 kmsg->msg.msg_name = &hdr.addr;
780 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
781 kmsg->msg.msg_namelen = 0;
782
783 if (sock->file->f_flags & O_NONBLOCK)
784 flags |= MSG_DONTWAIT;
785
786 err = sock_recvmsg(sock, &kmsg->msg, flags);
787 *finished = err <= 0;
788 if (err < 0)
789 return err;
790
791 hdr.msg = (struct io_uring_recvmsg_out) {
792 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
793 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
794 };
795
796 hdr.msg.payloadlen = err;
797 if (err > kmsg->payloadlen)
798 err = kmsg->payloadlen;
799
800 copy_len = sizeof(struct io_uring_recvmsg_out);
801 if (kmsg->msg.msg_namelen > kmsg->namelen)
802 copy_len += kmsg->namelen;
803 else
804 copy_len += kmsg->msg.msg_namelen;
805
806 /*
807 * "fromlen shall refer to the value before truncation.."
808 * 1003.1g
809 */
810 hdr.msg.namelen = kmsg->msg.msg_namelen;
811
812 /* ensure that there is no gap between hdr and sockaddr_storage */
813 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
814 sizeof(struct io_uring_recvmsg_out));
815 if (copy_to_user(io->buf, &hdr, copy_len)) {
816 *finished = true;
817 return -EFAULT;
818 }
819
820 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
821 kmsg->controllen + err;
822}
823
824int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
825{
826 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
827 struct io_async_msghdr iomsg, *kmsg;
828 struct socket *sock;
829 unsigned flags;
830 int ret, min_ret = 0;
831 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
832 bool mshot_finished = true;
833
834 sock = sock_from_file(req->file);
835 if (unlikely(!sock))
836 return -ENOTSOCK;
837
838 if (req_has_async_data(req)) {
839 kmsg = req->async_data;
840 } else {
841 ret = io_recvmsg_copy_hdr(req, &iomsg);
842 if (ret)
843 return ret;
844 kmsg = &iomsg;
845 }
846
847 if (!(req->flags & REQ_F_POLLED) &&
848 (sr->flags & IORING_RECVSEND_POLL_FIRST))
849 return io_setup_async_msg(req, kmsg, issue_flags);
850
851 flags = sr->msg_flags;
852 if (force_nonblock)
853 flags |= MSG_DONTWAIT;
854
855retry_multishot:
856 if (io_do_buffer_select(req)) {
857 void __user *buf;
858 size_t len = sr->len;
859
860 buf = io_buffer_select(req, &len, issue_flags);
861 if (!buf)
862 return -ENOBUFS;
863
864 if (req->flags & REQ_F_APOLL_MULTISHOT) {
865 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
866 if (ret) {
867 io_kbuf_recycle(req, issue_flags);
868 return ret;
869 }
870 }
871
872 iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_DEST, buf, len);
873 }
874
875 kmsg->msg.msg_get_inq = 1;
876 kmsg->msg.msg_inq = -1;
877 if (req->flags & REQ_F_APOLL_MULTISHOT) {
878 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
879 &mshot_finished);
880 } else {
881 /* disable partial retry for recvmsg with cmsg attached */
882 if (flags & MSG_WAITALL && !kmsg->msg.msg_controllen)
883 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
884
885 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
886 kmsg->uaddr, flags);
887 }
888
889 if (ret < min_ret) {
890 if (ret == -EAGAIN && force_nonblock) {
891 ret = io_setup_async_msg(req, kmsg, issue_flags);
892 if (ret == -EAGAIN && (issue_flags & IO_URING_F_MULTISHOT)) {
893 io_kbuf_recycle(req, issue_flags);
894 return IOU_ISSUE_SKIP_COMPLETE;
895 }
896 return ret;
897 }
898 if (ret > 0 && io_net_retry(sock, flags)) {
899 sr->done_io += ret;
900 req->flags |= REQ_F_BL_NO_RECYCLE;
901 return io_setup_async_msg(req, kmsg, issue_flags);
902 }
903 if (ret == -ERESTARTSYS)
904 ret = -EINTR;
905 req_set_fail(req);
906 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
907 req_set_fail(req);
908 }
909
910 if (ret > 0)
911 ret += sr->done_io;
912 else if (sr->done_io)
913 ret = sr->done_io;
914 else
915 io_kbuf_recycle(req, issue_flags);
916
917 if (!io_recv_finish(req, &ret, &kmsg->msg, mshot_finished, issue_flags))
918 goto retry_multishot;
919
920 if (mshot_finished)
921 io_req_msg_cleanup(req, kmsg, issue_flags);
922 else if (ret == -EAGAIN)
923 return io_setup_async_msg(req, kmsg, issue_flags);
924
925 return ret;
926}
927
928int io_recv(struct io_kiocb *req, unsigned int issue_flags)
929{
930 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
931 struct msghdr msg;
932 struct socket *sock;
933 unsigned flags;
934 int ret, min_ret = 0;
935 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
936 size_t len = sr->len;
937
938 if (!(req->flags & REQ_F_POLLED) &&
939 (sr->flags & IORING_RECVSEND_POLL_FIRST))
940 return -EAGAIN;
941
942 sock = sock_from_file(req->file);
943 if (unlikely(!sock))
944 return -ENOTSOCK;
945
946 msg.msg_name = NULL;
947 msg.msg_namelen = 0;
948 msg.msg_control = NULL;
949 msg.msg_get_inq = 1;
950 msg.msg_controllen = 0;
951 msg.msg_iocb = NULL;
952 msg.msg_ubuf = NULL;
953
954 flags = sr->msg_flags;
955 if (force_nonblock)
956 flags |= MSG_DONTWAIT;
957
958retry_multishot:
959 if (io_do_buffer_select(req)) {
960 void __user *buf;
961
962 buf = io_buffer_select(req, &len, issue_flags);
963 if (!buf)
964 return -ENOBUFS;
965 sr->buf = buf;
966 sr->len = len;
967 }
968
969 ret = import_ubuf(ITER_DEST, sr->buf, len, &msg.msg_iter);
970 if (unlikely(ret))
971 goto out_free;
972
973 msg.msg_inq = -1;
974 msg.msg_flags = 0;
975
976 if (flags & MSG_WAITALL)
977 min_ret = iov_iter_count(&msg.msg_iter);
978
979 ret = sock_recvmsg(sock, &msg, flags);
980 if (ret < min_ret) {
981 if (ret == -EAGAIN && force_nonblock) {
982 if (issue_flags & IO_URING_F_MULTISHOT) {
983 io_kbuf_recycle(req, issue_flags);
984 return IOU_ISSUE_SKIP_COMPLETE;
985 }
986
987 return -EAGAIN;
988 }
989 if (ret > 0 && io_net_retry(sock, flags)) {
990 sr->len -= ret;
991 sr->buf += ret;
992 sr->done_io += ret;
993 req->flags |= REQ_F_BL_NO_RECYCLE;
994 return -EAGAIN;
995 }
996 if (ret == -ERESTARTSYS)
997 ret = -EINTR;
998 req_set_fail(req);
999 } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
1000out_free:
1001 req_set_fail(req);
1002 }
1003
1004 if (ret > 0)
1005 ret += sr->done_io;
1006 else if (sr->done_io)
1007 ret = sr->done_io;
1008 else
1009 io_kbuf_recycle(req, issue_flags);
1010
1011 if (!io_recv_finish(req, &ret, &msg, ret <= 0, issue_flags))
1012 goto retry_multishot;
1013
1014 return ret;
1015}
1016
1017void io_send_zc_cleanup(struct io_kiocb *req)
1018{
1019 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1020 struct io_async_msghdr *io;
1021
1022 if (req_has_async_data(req)) {
1023 io = req->async_data;
1024 /* might be ->fast_iov if *msg_copy_hdr failed */
1025 if (io->free_iov != io->fast_iov)
1026 kfree(io->free_iov);
1027 }
1028 if (zc->notif) {
1029 io_notif_flush(zc->notif);
1030 zc->notif = NULL;
1031 }
1032}
1033
1034#define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF)
1035#define IO_ZC_FLAGS_VALID (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE)
1036
1037int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1038{
1039 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1040 struct io_ring_ctx *ctx = req->ctx;
1041 struct io_kiocb *notif;
1042
1043 zc->done_io = 0;
1044 req->flags |= REQ_F_POLL_NO_LAZY;
1045
1046 if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
1047 return -EINVAL;
1048 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
1049 if (req->flags & REQ_F_CQE_SKIP)
1050 return -EINVAL;
1051
1052 notif = zc->notif = io_alloc_notif(ctx);
1053 if (!notif)
1054 return -ENOMEM;
1055 notif->cqe.user_data = req->cqe.user_data;
1056 notif->cqe.res = 0;
1057 notif->cqe.flags = IORING_CQE_F_NOTIF;
1058 req->flags |= REQ_F_NEED_CLEANUP;
1059
1060 zc->flags = READ_ONCE(sqe->ioprio);
1061 if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) {
1062 if (zc->flags & ~IO_ZC_FLAGS_VALID)
1063 return -EINVAL;
1064 if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) {
1065 io_notif_set_extended(notif);
1066 io_notif_to_data(notif)->zc_report = true;
1067 }
1068 }
1069
1070 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1071 unsigned idx = READ_ONCE(sqe->buf_index);
1072
1073 if (unlikely(idx >= ctx->nr_user_bufs))
1074 return -EFAULT;
1075 idx = array_index_nospec(idx, ctx->nr_user_bufs);
1076 req->imu = READ_ONCE(ctx->user_bufs[idx]);
1077 io_req_set_rsrc_node(notif, ctx, 0);
1078 }
1079
1080 if (req->opcode == IORING_OP_SEND_ZC) {
1081 if (READ_ONCE(sqe->__pad3[0]))
1082 return -EINVAL;
1083 zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1084 zc->addr_len = READ_ONCE(sqe->addr_len);
1085 } else {
1086 if (unlikely(sqe->addr2 || sqe->file_index))
1087 return -EINVAL;
1088 if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
1089 return -EINVAL;
1090 }
1091
1092 zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
1093 zc->len = READ_ONCE(sqe->len);
1094 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
1095 if (zc->msg_flags & MSG_DONTWAIT)
1096 req->flags |= REQ_F_NOWAIT;
1097
1098#ifdef CONFIG_COMPAT
1099 if (req->ctx->compat)
1100 zc->msg_flags |= MSG_CMSG_COMPAT;
1101#endif
1102 return 0;
1103}
1104
1105static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
1106 struct iov_iter *from, size_t length)
1107{
1108 skb_zcopy_downgrade_managed(skb);
1109 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1110}
1111
1112static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1113 struct iov_iter *from, size_t length)
1114{
1115 struct skb_shared_info *shinfo = skb_shinfo(skb);
1116 int frag = shinfo->nr_frags;
1117 int ret = 0;
1118 struct bvec_iter bi;
1119 ssize_t copied = 0;
1120 unsigned long truesize = 0;
1121
1122 if (!frag)
1123 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
1124 else if (unlikely(!skb_zcopy_managed(skb)))
1125 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1126
1127 bi.bi_size = min(from->count, length);
1128 bi.bi_bvec_done = from->iov_offset;
1129 bi.bi_idx = 0;
1130
1131 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1132 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1133
1134 copied += v.bv_len;
1135 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1136 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1137 v.bv_offset, v.bv_len);
1138 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1139 }
1140 if (bi.bi_size)
1141 ret = -EMSGSIZE;
1142
1143 shinfo->nr_frags = frag;
1144 from->bvec += bi.bi_idx;
1145 from->nr_segs -= bi.bi_idx;
1146 from->count -= copied;
1147 from->iov_offset = bi.bi_bvec_done;
1148
1149 skb->data_len += copied;
1150 skb->len += copied;
1151 skb->truesize += truesize;
1152
1153 if (sk && sk->sk_type == SOCK_STREAM) {
1154 sk_wmem_queued_add(sk, truesize);
1155 if (!skb_zcopy_pure(skb))
1156 sk_mem_charge(sk, truesize);
1157 } else {
1158 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1159 }
1160 return ret;
1161}
1162
1163int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
1164{
1165 struct sockaddr_storage __address;
1166 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1167 struct msghdr msg;
1168 struct socket *sock;
1169 unsigned msg_flags;
1170 int ret, min_ret = 0;
1171
1172 sock = sock_from_file(req->file);
1173 if (unlikely(!sock))
1174 return -ENOTSOCK;
1175 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1176 return -EOPNOTSUPP;
1177
1178 msg.msg_name = NULL;
1179 msg.msg_control = NULL;
1180 msg.msg_controllen = 0;
1181 msg.msg_namelen = 0;
1182
1183 if (zc->addr) {
1184 if (req_has_async_data(req)) {
1185 struct io_async_msghdr *io = req->async_data;
1186
1187 msg.msg_name = &io->addr;
1188 } else {
1189 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1190 if (unlikely(ret < 0))
1191 return ret;
1192 msg.msg_name = (struct sockaddr *)&__address;
1193 }
1194 msg.msg_namelen = zc->addr_len;
1195 }
1196
1197 if (!(req->flags & REQ_F_POLLED) &&
1198 (zc->flags & IORING_RECVSEND_POLL_FIRST))
1199 return io_setup_async_addr(req, &__address, issue_flags);
1200
1201 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1202 ret = io_import_fixed(ITER_SOURCE, &msg.msg_iter, req->imu,
1203 (u64)(uintptr_t)zc->buf, zc->len);
1204 if (unlikely(ret))
1205 return ret;
1206 msg.sg_from_iter = io_sg_from_iter;
1207 } else {
1208 io_notif_set_extended(zc->notif);
1209 ret = import_ubuf(ITER_SOURCE, zc->buf, zc->len, &msg.msg_iter);
1210 if (unlikely(ret))
1211 return ret;
1212 ret = io_notif_account_mem(zc->notif, zc->len);
1213 if (unlikely(ret))
1214 return ret;
1215 msg.sg_from_iter = io_sg_from_iter_iovec;
1216 }
1217
1218 msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1219 if (issue_flags & IO_URING_F_NONBLOCK)
1220 msg_flags |= MSG_DONTWAIT;
1221 if (msg_flags & MSG_WAITALL)
1222 min_ret = iov_iter_count(&msg.msg_iter);
1223 msg_flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
1224
1225 msg.msg_flags = msg_flags;
1226 msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
1227 ret = sock_sendmsg(sock, &msg);
1228
1229 if (unlikely(ret < min_ret)) {
1230 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1231 return io_setup_async_addr(req, &__address, issue_flags);
1232
1233 if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1234 zc->len -= ret;
1235 zc->buf += ret;
1236 zc->done_io += ret;
1237 req->flags |= REQ_F_BL_NO_RECYCLE;
1238 return io_setup_async_addr(req, &__address, issue_flags);
1239 }
1240 if (ret == -ERESTARTSYS)
1241 ret = -EINTR;
1242 req_set_fail(req);
1243 }
1244
1245 if (ret >= 0)
1246 ret += zc->done_io;
1247 else if (zc->done_io)
1248 ret = zc->done_io;
1249
1250 /*
1251 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1252 * flushing notif to io_send_zc_cleanup()
1253 */
1254 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1255 io_notif_flush(zc->notif);
1256 req->flags &= ~REQ_F_NEED_CLEANUP;
1257 }
1258 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1259 return IOU_OK;
1260}
1261
1262int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1263{
1264 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1265 struct io_async_msghdr iomsg, *kmsg;
1266 struct socket *sock;
1267 unsigned flags;
1268 int ret, min_ret = 0;
1269
1270 io_notif_set_extended(sr->notif);
1271
1272 sock = sock_from_file(req->file);
1273 if (unlikely(!sock))
1274 return -ENOTSOCK;
1275 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1276 return -EOPNOTSUPP;
1277
1278 if (req_has_async_data(req)) {
1279 kmsg = req->async_data;
1280 kmsg->msg.msg_control_user = sr->msg_control;
1281 } else {
1282 ret = io_sendmsg_copy_hdr(req, &iomsg);
1283 if (ret)
1284 return ret;
1285 kmsg = &iomsg;
1286 }
1287
1288 if (!(req->flags & REQ_F_POLLED) &&
1289 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1290 return io_setup_async_msg(req, kmsg, issue_flags);
1291
1292 flags = sr->msg_flags | MSG_ZEROCOPY;
1293 if (issue_flags & IO_URING_F_NONBLOCK)
1294 flags |= MSG_DONTWAIT;
1295 if (flags & MSG_WAITALL)
1296 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1297
1298 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1299 kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1300 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1301
1302 if (unlikely(ret < min_ret)) {
1303 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1304 return io_setup_async_msg(req, kmsg, issue_flags);
1305
1306 if (ret > 0 && io_net_retry(sock, flags)) {
1307 sr->done_io += ret;
1308 req->flags |= REQ_F_BL_NO_RECYCLE;
1309 return io_setup_async_msg(req, kmsg, issue_flags);
1310 }
1311 if (ret == -ERESTARTSYS)
1312 ret = -EINTR;
1313 req_set_fail(req);
1314 }
1315 /* fast path, check for non-NULL to avoid function call */
1316 if (kmsg->free_iov) {
1317 kfree(kmsg->free_iov);
1318 kmsg->free_iov = NULL;
1319 }
1320
1321 io_netmsg_recycle(req, issue_flags);
1322 if (ret >= 0)
1323 ret += sr->done_io;
1324 else if (sr->done_io)
1325 ret = sr->done_io;
1326
1327 /*
1328 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1329 * flushing notif to io_send_zc_cleanup()
1330 */
1331 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1332 io_notif_flush(sr->notif);
1333 req->flags &= ~REQ_F_NEED_CLEANUP;
1334 }
1335 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1336 return IOU_OK;
1337}
1338
1339void io_sendrecv_fail(struct io_kiocb *req)
1340{
1341 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1342
1343 if (sr->done_io)
1344 req->cqe.res = sr->done_io;
1345
1346 if ((req->flags & REQ_F_NEED_CLEANUP) &&
1347 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1348 req->cqe.flags |= IORING_CQE_F_MORE;
1349}
1350
1351int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1352{
1353 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1354 unsigned flags;
1355
1356 if (sqe->len || sqe->buf_index)
1357 return -EINVAL;
1358
1359 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1360 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1361 accept->flags = READ_ONCE(sqe->accept_flags);
1362 accept->nofile = rlimit(RLIMIT_NOFILE);
1363 flags = READ_ONCE(sqe->ioprio);
1364 if (flags & ~IORING_ACCEPT_MULTISHOT)
1365 return -EINVAL;
1366
1367 accept->file_slot = READ_ONCE(sqe->file_index);
1368 if (accept->file_slot) {
1369 if (accept->flags & SOCK_CLOEXEC)
1370 return -EINVAL;
1371 if (flags & IORING_ACCEPT_MULTISHOT &&
1372 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1373 return -EINVAL;
1374 }
1375 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1376 return -EINVAL;
1377 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1378 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1379 if (flags & IORING_ACCEPT_MULTISHOT)
1380 req->flags |= REQ_F_APOLL_MULTISHOT;
1381 return 0;
1382}
1383
1384int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1385{
1386 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1387 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1388 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1389 bool fixed = !!accept->file_slot;
1390 struct file *file;
1391 int ret, fd;
1392
1393retry:
1394 if (!fixed) {
1395 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1396 if (unlikely(fd < 0))
1397 return fd;
1398 }
1399 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1400 accept->flags);
1401 if (IS_ERR(file)) {
1402 if (!fixed)
1403 put_unused_fd(fd);
1404 ret = PTR_ERR(file);
1405 if (ret == -EAGAIN && force_nonblock) {
1406 /*
1407 * if it's multishot and polled, we don't need to
1408 * return EAGAIN to arm the poll infra since it
1409 * has already been done
1410 */
1411 if (issue_flags & IO_URING_F_MULTISHOT)
1412 return IOU_ISSUE_SKIP_COMPLETE;
1413 return ret;
1414 }
1415 if (ret == -ERESTARTSYS)
1416 ret = -EINTR;
1417 req_set_fail(req);
1418 } else if (!fixed) {
1419 fd_install(fd, file);
1420 ret = fd;
1421 } else {
1422 ret = io_fixed_fd_install(req, issue_flags, file,
1423 accept->file_slot);
1424 }
1425
1426 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1427 io_req_set_res(req, ret, 0);
1428 return IOU_OK;
1429 }
1430
1431 if (ret < 0)
1432 return ret;
1433 if (io_fill_cqe_req_aux(req, issue_flags & IO_URING_F_COMPLETE_DEFER,
1434 ret, IORING_CQE_F_MORE))
1435 goto retry;
1436
1437 io_req_set_res(req, ret, 0);
1438 return IOU_STOP_MULTISHOT;
1439}
1440
1441int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1442{
1443 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1444
1445 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1446 return -EINVAL;
1447
1448 sock->domain = READ_ONCE(sqe->fd);
1449 sock->type = READ_ONCE(sqe->off);
1450 sock->protocol = READ_ONCE(sqe->len);
1451 sock->file_slot = READ_ONCE(sqe->file_index);
1452 sock->nofile = rlimit(RLIMIT_NOFILE);
1453
1454 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1455 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1456 return -EINVAL;
1457 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1458 return -EINVAL;
1459 return 0;
1460}
1461
1462int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1463{
1464 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1465 bool fixed = !!sock->file_slot;
1466 struct file *file;
1467 int ret, fd;
1468
1469 if (!fixed) {
1470 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1471 if (unlikely(fd < 0))
1472 return fd;
1473 }
1474 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1475 if (IS_ERR(file)) {
1476 if (!fixed)
1477 put_unused_fd(fd);
1478 ret = PTR_ERR(file);
1479 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1480 return -EAGAIN;
1481 if (ret == -ERESTARTSYS)
1482 ret = -EINTR;
1483 req_set_fail(req);
1484 } else if (!fixed) {
1485 fd_install(fd, file);
1486 ret = fd;
1487 } else {
1488 ret = io_fixed_fd_install(req, issue_flags, file,
1489 sock->file_slot);
1490 }
1491 io_req_set_res(req, ret, 0);
1492 return IOU_OK;
1493}
1494
1495int io_connect_prep_async(struct io_kiocb *req)
1496{
1497 struct io_async_connect *io = req->async_data;
1498 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1499
1500 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1501}
1502
1503int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1504{
1505 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1506
1507 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1508 return -EINVAL;
1509
1510 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1511 conn->addr_len = READ_ONCE(sqe->addr2);
1512 conn->in_progress = conn->seen_econnaborted = false;
1513 return 0;
1514}
1515
1516int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1517{
1518 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1519 struct io_async_connect __io, *io;
1520 unsigned file_flags;
1521 int ret;
1522 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1523
1524 if (req_has_async_data(req)) {
1525 io = req->async_data;
1526 } else {
1527 ret = move_addr_to_kernel(connect->addr,
1528 connect->addr_len,
1529 &__io.address);
1530 if (ret)
1531 goto out;
1532 io = &__io;
1533 }
1534
1535 file_flags = force_nonblock ? O_NONBLOCK : 0;
1536
1537 ret = __sys_connect_file(req->file, &io->address,
1538 connect->addr_len, file_flags);
1539 if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED)
1540 && force_nonblock) {
1541 if (ret == -EINPROGRESS) {
1542 connect->in_progress = true;
1543 } else if (ret == -ECONNABORTED) {
1544 if (connect->seen_econnaborted)
1545 goto out;
1546 connect->seen_econnaborted = true;
1547 }
1548 if (req_has_async_data(req))
1549 return -EAGAIN;
1550 if (io_alloc_async_data(req)) {
1551 ret = -ENOMEM;
1552 goto out;
1553 }
1554 memcpy(req->async_data, &__io, sizeof(__io));
1555 return -EAGAIN;
1556 }
1557 if (connect->in_progress) {
1558 /*
1559 * At least bluetooth will return -EBADFD on a re-connect
1560 * attempt, and it's (supposedly) also valid to get -EISCONN
1561 * which means the previous result is good. For both of these,
1562 * grab the sock_error() and use that for the completion.
1563 */
1564 if (ret == -EBADFD || ret == -EISCONN)
1565 ret = sock_error(sock_from_file(req->file)->sk);
1566 }
1567 if (ret == -ERESTARTSYS)
1568 ret = -EINTR;
1569out:
1570 if (ret < 0)
1571 req_set_fail(req);
1572 io_req_set_res(req, ret, 0);
1573 return IOU_OK;
1574}
1575
1576void io_netmsg_cache_free(struct io_cache_entry *entry)
1577{
1578 kfree(container_of(entry, struct io_async_msghdr, cache));
1579}
1580#endif