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