]> git.ipfire.org Git - thirdparty/linux.git/blob - io_uring/net.c
projects / thirdparty / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'devicetree-fixes-for-6.8-1' of git://git.kernel.org/pub/scm/linux/kernel...
[thirdparty/linux.git] / io_uring / net.c
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)
21 struct io_shutdown {
22 struct file *file;
23 int how;
24 };
25
26 struct 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
35 struct 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
45 struct 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
53 struct 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
81 static 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
94 int 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
107 int 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
124 static 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
131 static 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
145 static 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
171 static 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
177 static 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
207 static 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
222 int 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
237 static 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
253 int 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
265 void 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
272 int 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
302 int 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
364 int 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
438 static 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
453 static 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
500 static 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
553 static 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
567 int 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
581 int 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
626 static 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 */
641 static 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. */
685 finish:
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
695 static 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
720 struct io_recvmsg_multishot_hdr {
721 struct io_uring_recvmsg_out msg;
722 struct sockaddr_storage addr;
723 };
724
725 static 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
779 int 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
809 retry_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
889 int 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
918 retry_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))) {
963 out_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
980 void 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
1000 int 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
1067 static 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
1074 static 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
1125 int 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
1224 int 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
1300 void 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
1312 int 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
1345 int 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;
1356 retry:
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 ret = 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 return -ECANCELED;
1401 }
1402
1403 int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1404 {
1405 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1406
1407 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1408 return -EINVAL;
1409
1410 sock->domain = READ_ONCE(sqe->fd);
1411 sock->type = READ_ONCE(sqe->off);
1412 sock->protocol = READ_ONCE(sqe->len);
1413 sock->file_slot = READ_ONCE(sqe->file_index);
1414 sock->nofile = rlimit(RLIMIT_NOFILE);
1415
1416 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1417 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1418 return -EINVAL;
1419 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1420 return -EINVAL;
1421 return 0;
1422 }
1423
1424 int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1425 {
1426 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1427 bool fixed = !!sock->file_slot;
1428 struct file *file;
1429 int ret, fd;
1430
1431 if (!fixed) {
1432 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1433 if (unlikely(fd < 0))
1434 return fd;
1435 }
1436 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1437 if (IS_ERR(file)) {
1438 if (!fixed)
1439 put_unused_fd(fd);
1440 ret = PTR_ERR(file);
1441 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1442 return -EAGAIN;
1443 if (ret == -ERESTARTSYS)
1444 ret = -EINTR;
1445 req_set_fail(req);
1446 } else if (!fixed) {
1447 fd_install(fd, file);
1448 ret = fd;
1449 } else {
1450 ret = io_fixed_fd_install(req, issue_flags, file,
1451 sock->file_slot);
1452 }
1453 io_req_set_res(req, ret, 0);
1454 return IOU_OK;
1455 }
1456
1457 int io_connect_prep_async(struct io_kiocb *req)
1458 {
1459 struct io_async_connect *io = req->async_data;
1460 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1461
1462 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1463 }
1464
1465 int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1466 {
1467 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1468
1469 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1470 return -EINVAL;
1471
1472 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1473 conn->addr_len = READ_ONCE(sqe->addr2);
1474 conn->in_progress = conn->seen_econnaborted = false;
1475 return 0;
1476 }
1477
1478 int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1479 {
1480 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1481 struct io_async_connect __io, *io;
1482 unsigned file_flags;
1483 int ret;
1484 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1485
1486 if (req_has_async_data(req)) {
1487 io = req->async_data;
1488 } else {
1489 ret = move_addr_to_kernel(connect->addr,
1490 connect->addr_len,
1491 &__io.address);
1492 if (ret)
1493 goto out;
1494 io = &__io;
1495 }
1496
1497 file_flags = force_nonblock ? O_NONBLOCK : 0;
1498
1499 ret = __sys_connect_file(req->file, &io->address,
1500 connect->addr_len, file_flags);
1501 if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED)
1502 && force_nonblock) {
1503 if (ret == -EINPROGRESS) {
1504 connect->in_progress = true;
1505 } else if (ret == -ECONNABORTED) {
1506 if (connect->seen_econnaborted)
1507 goto out;
1508 connect->seen_econnaborted = true;
1509 }
1510 if (req_has_async_data(req))
1511 return -EAGAIN;
1512 if (io_alloc_async_data(req)) {
1513 ret = -ENOMEM;
1514 goto out;
1515 }
1516 memcpy(req->async_data, &__io, sizeof(__io));
1517 return -EAGAIN;
1518 }
1519 if (connect->in_progress) {
1520 /*
1521 * At least bluetooth will return -EBADFD on a re-connect
1522 * attempt, and it's (supposedly) also valid to get -EISCONN
1523 * which means the previous result is good. For both of these,
1524 * grab the sock_error() and use that for the completion.
1525 */
1526 if (ret == -EBADFD || ret == -EISCONN)
1527 ret = sock_error(sock_from_file(req->file)->sk);
1528 }
1529 if (ret == -ERESTARTSYS)
1530 ret = -EINTR;
1531 out:
1532 if (ret < 0)
1533 req_set_fail(req);
1534 io_req_set_res(req, ret, 0);
1535 return IOU_OK;
1536 }
1537
1538 void io_netmsg_cache_free(struct io_cache_entry *entry)
1539 {
1540 kfree(container_of(entry, struct io_async_msghdr, cache));
1541 }
1542 #endif
A mirror of Linus' kernel repository