projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
io_uring/net: fix zc send link failing
[linux-2.6-block.git] / io_uring / net.c
CommitLineData
f9ead18c
JA		 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
f9ead18c 13#include "io_uring.h"
3b77495a 14#include "kbuf.h"
43e0bbbd 15#include "alloc_cache.h"
f9ead18c 16#include "net.h"
06a5464b 17#include "notif.h"
10c7d33e 18#include "rsrc.h"
f9ead18c
JA		 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};
50
51struct io_sr_msg {
52 struct file *file;
53 union {
54 struct compat_msghdr __user *umsg_compat;
55 struct user_msghdr __user *umsg;
56 void __user *buf;
57 };
293402e5
PB		 58 unsigned msg_flags;
59 unsigned flags;
f9ead18c
JA		 60 size_t len;
61 size_t done_io;
f9ead18c
JA		 62};
63
06a5464b
PB		 64struct io_sendzc {
65 struct file *file;
66 void __user *buf;
67 size_t len;
68 u16 slot_idx;
69 unsigned msg_flags;
70 unsigned flags;
092aeedb
PB		 71 unsigned addr_len;
72 void __user *addr;
4a933e62 73 size_t done_io;
06a5464b
PB		 74};
75
f9ead18c
JA		 76#define IO_APOLL_MULTI_POLLED (REQ_F_APOLL_MULTISHOT | REQ_F_POLLED)
77
78int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
79{
f2ccb5ae 80 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
f9ead18c
JA		 81
82 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
83 sqe->buf_index || sqe->splice_fd_in))
84 return -EINVAL;
85
86 shutdown->how = READ_ONCE(sqe->len);
87 return 0;
88}
89
90int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
91{
f2ccb5ae 92 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
f9ead18c
JA		 93 struct socket *sock;
94 int ret;
95
96 if (issue_flags & IO_URING_F_NONBLOCK)
97 return -EAGAIN;
98
99 sock = sock_from_file(req->file);
100 if (unlikely(!sock))
101 return -ENOTSOCK;
102
103 ret = __sys_shutdown_sock(sock, shutdown->how);
104 io_req_set_res(req, ret, 0);
105 return IOU_OK;
106}
107
108static bool io_net_retry(struct socket *sock, int flags)
109{
110 if (!(flags & MSG_WAITALL))
111 return false;
112 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
113}
114
43e0bbbd
JA		 115static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
116{
117 struct io_async_msghdr *hdr = req->async_data;
118
06360426 119 if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
43e0bbbd
JA		 120 return;
121
122 /* Let normal cleanup path reap it if we fail adding to the cache */
123 if (io_alloc_cache_put(&req->ctx->netmsg_cache, &hdr->cache)) {
124 req->async_data = NULL;
125 req->flags &= ~REQ_F_ASYNC_DATA;
126 }
127}
128
129static struct io_async_msghdr *io_recvmsg_alloc_async(struct io_kiocb *req,
130 unsigned int issue_flags)
131{
132 struct io_ring_ctx *ctx = req->ctx;
133 struct io_cache_entry *entry;
134
135 if (!(issue_flags & IO_URING_F_UNLOCKED) &&
136 (entry = io_alloc_cache_get(&ctx->netmsg_cache)) != NULL) {
137 struct io_async_msghdr *hdr;
138
139 hdr = container_of(entry, struct io_async_msghdr, cache);
140 req->flags |= REQ_F_ASYNC_DATA;
141 req->async_data = hdr;
142 return hdr;
143 }
144
145 if (!io_alloc_async_data(req))
146 return req->async_data;
147
148 return NULL;
149}
150
f9ead18c 151static int io_setup_async_msg(struct io_kiocb *req,
43e0bbbd
JA		 152 struct io_async_msghdr *kmsg,
153 unsigned int issue_flags)
f9ead18c 154{
3f743e9b 155 struct io_async_msghdr *async_msg;
f9ead18c 156
3f743e9b 157 if (req_has_async_data(req))
f9ead18c 158 return -EAGAIN;
43e0bbbd
JA		 159 async_msg = io_recvmsg_alloc_async(req, issue_flags);
160 if (!async_msg) {
f9ead18c
JA		 161 kfree(kmsg->free_iov);
162 return -ENOMEM;
163 }
f9ead18c
JA		 164 req->flags |= REQ_F_NEED_CLEANUP;
165 memcpy(async_msg, kmsg, sizeof(*kmsg));
166 async_msg->msg.msg_name = &async_msg->addr;
167 /* if were using fast_iov, set it to the new one */
168 if (!async_msg->free_iov)
169 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
170
171 return -EAGAIN;
172}
173
174static int io_sendmsg_copy_hdr(struct io_kiocb *req,
175 struct io_async_msghdr *iomsg)
176{
f2ccb5ae 177 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 178
179 iomsg->msg.msg_name = &iomsg->addr;
180 iomsg->free_iov = iomsg->fast_iov;
181 return sendmsg_copy_msghdr(&iomsg->msg, sr->umsg, sr->msg_flags,
182 &iomsg->free_iov);
183}
184
185int io_sendmsg_prep_async(struct io_kiocb *req)
186{
187 int ret;
188
189 ret = io_sendmsg_copy_hdr(req, req->async_data);
190 if (!ret)
191 req->flags |= REQ_F_NEED_CLEANUP;
192 return ret;
193}
194
195void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
196{
197 struct io_async_msghdr *io = req->async_data;
198
199 kfree(io->free_iov);
200}
201
202int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
203{
f2ccb5ae 204 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 205
206 if (unlikely(sqe->file_index || sqe->addr2))
207 return -EINVAL;
208
209 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
210 sr->len = READ_ONCE(sqe->len);
211 sr->flags = READ_ONCE(sqe->ioprio);
212 if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
213 return -EINVAL;
214 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
215 if (sr->msg_flags & MSG_DONTWAIT)
216 req->flags |= REQ_F_NOWAIT;
217
218#ifdef CONFIG_COMPAT
219 if (req->ctx->compat)
220 sr->msg_flags |= MSG_CMSG_COMPAT;
221#endif
222 sr->done_io = 0;
223 return 0;
224}
225
226int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
227{
f2ccb5ae 228 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 229 struct io_async_msghdr iomsg, *kmsg;
230 struct socket *sock;
231 unsigned flags;
232 int min_ret = 0;
233 int ret;
234
235 sock = sock_from_file(req->file);
236 if (unlikely(!sock))
237 return -ENOTSOCK;
238
239 if (req_has_async_data(req)) {
240 kmsg = req->async_data;
241 } else {
242 ret = io_sendmsg_copy_hdr(req, &iomsg);
243 if (ret)
244 return ret;
245 kmsg = &iomsg;
246 }
247
248 if (!(req->flags & REQ_F_POLLED) &&
249 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 250 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 251
252 flags = sr->msg_flags;
253 if (issue_flags & IO_URING_F_NONBLOCK)
254 flags |= MSG_DONTWAIT;
255 if (flags & MSG_WAITALL)
256 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
257
258 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
259
260 if (ret < min_ret) {
261 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
43e0bbbd 262 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 263 if (ret == -ERESTARTSYS)
264 ret = -EINTR;
265 if (ret > 0 && io_net_retry(sock, flags)) {
266 sr->done_io += ret;
267 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 268 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 269 }
270 req_set_fail(req);
271 }
272 /* fast path, check for non-NULL to avoid function call */
273 if (kmsg->free_iov)
274 kfree(kmsg->free_iov);
275 req->flags &= ~REQ_F_NEED_CLEANUP;
43e0bbbd 276 io_netmsg_recycle(req, issue_flags);
f9ead18c
JA		 277 if (ret >= 0)
278 ret += sr->done_io;
279 else if (sr->done_io)
280 ret = sr->done_io;
281 io_req_set_res(req, ret, 0);
282 return IOU_OK;
283}
284
285int io_send(struct io_kiocb *req, unsigned int issue_flags)
286{
f2ccb5ae 287 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 288 struct msghdr msg;
289 struct iovec iov;
290 struct socket *sock;
291 unsigned flags;
292 int min_ret = 0;
293 int ret;
294
295 if (!(req->flags & REQ_F_POLLED) &&
296 (sr->flags & IORING_RECVSEND_POLL_FIRST))
297 return -EAGAIN;
298
299 sock = sock_from_file(req->file);
300 if (unlikely(!sock))
301 return -ENOTSOCK;
302
303 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
304 if (unlikely(ret))
305 return ret;
306
307 msg.msg_name = NULL;
308 msg.msg_control = NULL;
309 msg.msg_controllen = 0;
310 msg.msg_namelen = 0;
e02b6651 311 msg.msg_ubuf = NULL;
f9ead18c
JA		 312
313 flags = sr->msg_flags;
314 if (issue_flags & IO_URING_F_NONBLOCK)
315 flags |= MSG_DONTWAIT;
316 if (flags & MSG_WAITALL)
317 min_ret = iov_iter_count(&msg.msg_iter);
318
319 msg.msg_flags = flags;
320 ret = sock_sendmsg(sock, &msg);
321 if (ret < min_ret) {
322 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
323 return -EAGAIN;
324 if (ret == -ERESTARTSYS)
325 ret = -EINTR;
326 if (ret > 0 && io_net_retry(sock, flags)) {
327 sr->len -= ret;
328 sr->buf += ret;
329 sr->done_io += ret;
330 req->flags |= REQ_F_PARTIAL_IO;
331 return -EAGAIN;
332 }
333 req_set_fail(req);
334 }
335 if (ret >= 0)
336 ret += sr->done_io;
337 else if (sr->done_io)
338 ret = sr->done_io;
339 io_req_set_res(req, ret, 0);
340 return IOU_OK;
341}
342
9bb66906
DY		 343static bool io_recvmsg_multishot_overflow(struct io_async_msghdr *iomsg)
344{
9b0fc3c0 345 int hdr;
9bb66906 346
9b0fc3c0 347 if (iomsg->namelen < 0)
9bb66906 348 return true;
9b0fc3c0
DY		 349 if (check_add_overflow((int)sizeof(struct io_uring_recvmsg_out),
350 iomsg->namelen, &hdr))
9bb66906 351 return true;
9b0fc3c0 352 if (check_add_overflow(hdr, (int)iomsg->controllen, &hdr))
9bb66906
DY		 353 return true;
354
355 return false;
356}
357
f9ead18c
JA		 358static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
359 struct io_async_msghdr *iomsg)
360{
f2ccb5ae 361 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
7fa875b8 362 struct user_msghdr msg;
f9ead18c
JA		 363 int ret;
364
7fa875b8
DY		 365 if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
366 return -EFAULT;
367
368 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 369 if (ret)
370 return ret;
371
372 if (req->flags & REQ_F_BUFFER_SELECT) {
7fa875b8 373 if (msg.msg_iovlen == 0) {
5702196e
DY		 374 sr->len = iomsg->fast_iov[0].iov_len = 0;
375 iomsg->fast_iov[0].iov_base = NULL;
376 iomsg->free_iov = NULL;
7fa875b8 377 } else if (msg.msg_iovlen > 1) {
f9ead18c 378 return -EINVAL;
5702196e 379 } else {
7fa875b8 380 if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
5702196e
DY		 381 return -EFAULT;
382 sr->len = iomsg->fast_iov[0].iov_len;
383 iomsg->free_iov = NULL;
384 }
9bb66906
DY		 385
386 if (req->flags & REQ_F_APOLL_MULTISHOT) {
387 iomsg->namelen = msg.msg_namelen;
388 iomsg->controllen = msg.msg_controllen;
389 if (io_recvmsg_multishot_overflow(iomsg))
390 return -EOVERFLOW;
391 }
f9ead18c
JA		 392 } else {
393 iomsg->free_iov = iomsg->fast_iov;
7fa875b8 394 ret = __import_iovec(READ, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
f9ead18c
JA		 395 &iomsg->free_iov, &iomsg->msg.msg_iter,
396 false);
397 if (ret > 0)
398 ret = 0;
399 }
400
401 return ret;
402}
403
404#ifdef CONFIG_COMPAT
405static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
406 struct io_async_msghdr *iomsg)
407{
f2ccb5ae 408 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
72c531f8 409 struct compat_msghdr msg;
f9ead18c 410 struct compat_iovec __user *uiov;
f9ead18c
JA		 411 int ret;
412
72c531f8
DY		 413 if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
414 return -EFAULT;
415
4f6a94d3 416 ret = __get_compat_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 417 if (ret)
418 return ret;
419
72c531f8 420 uiov = compat_ptr(msg.msg_iov);
f9ead18c
JA		 421 if (req->flags & REQ_F_BUFFER_SELECT) {
422 compat_ssize_t clen;
423
72c531f8 424 if (msg.msg_iovlen == 0) {
6d2f75a0
DY		 425 sr->len = 0;
426 iomsg->free_iov = NULL;
72c531f8 427 } else if (msg.msg_iovlen > 1) {
f9ead18c 428 return -EINVAL;
6d2f75a0
DY		 429 } else {
430 if (!access_ok(uiov, sizeof(*uiov)))
431 return -EFAULT;
432 if (__get_user(clen, &uiov->iov_len))
433 return -EFAULT;
434 if (clen < 0)
435 return -EINVAL;
436 sr->len = clen;
437 iomsg->free_iov = NULL;
438 }
9bb66906
DY		 439
440 if (req->flags & REQ_F_APOLL_MULTISHOT) {
441 iomsg->namelen = msg.msg_namelen;
442 iomsg->controllen = msg.msg_controllen;
443 if (io_recvmsg_multishot_overflow(iomsg))
444 return -EOVERFLOW;
445 }
f9ead18c
JA		 446 } else {
447 iomsg->free_iov = iomsg->fast_iov;
72c531f8 448 ret = __import_iovec(READ, (struct iovec __user *)uiov, msg.msg_iovlen,
f9ead18c
JA		 449 UIO_FASTIOV, &iomsg->free_iov,
450 &iomsg->msg.msg_iter, true);
451 if (ret < 0)
452 return ret;
453 }
454
455 return 0;
456}
457#endif
458
459static int io_recvmsg_copy_hdr(struct io_kiocb *req,
460 struct io_async_msghdr *iomsg)
461{
462 iomsg->msg.msg_name = &iomsg->addr;
463
464#ifdef CONFIG_COMPAT
465 if (req->ctx->compat)
466 return __io_compat_recvmsg_copy_hdr(req, iomsg);
467#endif
468
469 return __io_recvmsg_copy_hdr(req, iomsg);
470}
471
472int io_recvmsg_prep_async(struct io_kiocb *req)
473{
474 int ret;
475
476 ret = io_recvmsg_copy_hdr(req, req->async_data);
477 if (!ret)
478 req->flags |= REQ_F_NEED_CLEANUP;
479 return ret;
480}
481
b3fdea6e
DY		 482#define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
483
f9ead18c
JA		 484int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
485{
f2ccb5ae 486 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 487
488 if (unlikely(sqe->file_index || sqe->addr2))
489 return -EINVAL;
490
491 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
492 sr->len = READ_ONCE(sqe->len);
493 sr->flags = READ_ONCE(sqe->ioprio);
b3fdea6e 494 if (sr->flags & ~(RECVMSG_FLAGS))
f9ead18c
JA		 495 return -EINVAL;
496 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
497 if (sr->msg_flags & MSG_DONTWAIT)
498 req->flags |= REQ_F_NOWAIT;
499 if (sr->msg_flags & MSG_ERRQUEUE)
500 req->flags |= REQ_F_CLEAR_POLLIN;
b3fdea6e
DY		 501 if (sr->flags & IORING_RECV_MULTISHOT) {
502 if (!(req->flags & REQ_F_BUFFER_SELECT))
503 return -EINVAL;
504 if (sr->msg_flags & MSG_WAITALL)
505 return -EINVAL;
506 if (req->opcode == IORING_OP_RECV && sr->len)
507 return -EINVAL;
508 req->flags |= REQ_F_APOLL_MULTISHOT;
509 }
f9ead18c
JA		 510
511#ifdef CONFIG_COMPAT
512 if (req->ctx->compat)
513 sr->msg_flags |= MSG_CMSG_COMPAT;
514#endif
515 sr->done_io = 0;
516 return 0;
517}
518
b3fdea6e
DY		 519static inline void io_recv_prep_retry(struct io_kiocb *req)
520{
f2ccb5ae 521 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
b3fdea6e
DY		 522
523 sr->done_io = 0;
524 sr->len = 0; /* get from the provided buffer */
525}
526
527/*
9bb66906 528 * Finishes io_recv and io_recvmsg.
b3fdea6e
DY		 529 *
530 * Returns true if it is actually finished, or false if it should run
531 * again (for multishot).
532 */
9bb66906
DY		 533static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
534 unsigned int cflags, bool mshot_finished)
b3fdea6e
DY		 535{
536 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
537 io_req_set_res(req, *ret, cflags);
538 *ret = IOU_OK;
539 return true;
540 }
541
9bb66906 542 if (!mshot_finished) {
b3fdea6e
DY		 543 if (io_post_aux_cqe(req->ctx, req->cqe.user_data, *ret,
544 cflags | IORING_CQE_F_MORE, false)) {
545 io_recv_prep_retry(req);
546 return false;
547 }
548 /*
549 * Otherwise stop multishot but use the current result.
550 * Probably will end up going into overflow, but this means
551 * we cannot trust the ordering anymore
552 */
553 }
554
555 io_req_set_res(req, *ret, cflags);
556
557 if (req->flags & REQ_F_POLLED)
558 *ret = IOU_STOP_MULTISHOT;
e2df2ccb
DY		 559 else
560 *ret = IOU_OK;
b3fdea6e
DY		 561 return true;
562}
563
9bb66906
DY		 564static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
565 struct io_sr_msg *sr, void __user **buf,
566 size_t *len)
567{
568 unsigned long ubuf = (unsigned long) *buf;
569 unsigned long hdr;
570
571 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
572 kmsg->controllen;
573 if (*len < hdr)
574 return -EFAULT;
575
576 if (kmsg->controllen) {
577 unsigned long control = ubuf + hdr - kmsg->controllen;
578
d1f6222c 579 kmsg->msg.msg_control_user = (void __user *) control;
9bb66906
DY		 580 kmsg->msg.msg_controllen = kmsg->controllen;
581 }
582
583 sr->buf = *buf; /* stash for later copy */
d1f6222c 584 *buf = (void __user *) (ubuf + hdr);
9bb66906
DY		 585 kmsg->payloadlen = *len = *len - hdr;
586 return 0;
587}
588
589struct io_recvmsg_multishot_hdr {
590 struct io_uring_recvmsg_out msg;
591 struct sockaddr_storage addr;
592};
593
594static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
595 struct io_async_msghdr *kmsg,
596 unsigned int flags, bool *finished)
597{
598 int err;
599 int copy_len;
600 struct io_recvmsg_multishot_hdr hdr;
601
602 if (kmsg->namelen)
603 kmsg->msg.msg_name = &hdr.addr;
604 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
605 kmsg->msg.msg_namelen = 0;
606
607 if (sock->file->f_flags & O_NONBLOCK)
608 flags |= MSG_DONTWAIT;
609
610 err = sock_recvmsg(sock, &kmsg->msg, flags);
611 *finished = err <= 0;
612 if (err < 0)
613 return err;
614
615 hdr.msg = (struct io_uring_recvmsg_out) {
616 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
617 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
618 };
619
620 hdr.msg.payloadlen = err;
621 if (err > kmsg->payloadlen)
622 err = kmsg->payloadlen;
623
624 copy_len = sizeof(struct io_uring_recvmsg_out);
625 if (kmsg->msg.msg_namelen > kmsg->namelen)
626 copy_len += kmsg->namelen;
627 else
628 copy_len += kmsg->msg.msg_namelen;
629
630 /*
631 * "fromlen shall refer to the value before truncation.."
632 * 1003.1g
633 */
634 hdr.msg.namelen = kmsg->msg.msg_namelen;
635
636 /* ensure that there is no gap between hdr and sockaddr_storage */
637 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
638 sizeof(struct io_uring_recvmsg_out));
639 if (copy_to_user(io->buf, &hdr, copy_len)) {
640 *finished = true;
641 return -EFAULT;
642 }
643
644 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
645 kmsg->controllen + err;
646}
647
f9ead18c
JA		 648int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
649{
f2ccb5ae 650 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 651 struct io_async_msghdr iomsg, *kmsg;
652 struct socket *sock;
653 unsigned int cflags;
654 unsigned flags;
655 int ret, min_ret = 0;
656 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
9bb66906 657 bool mshot_finished = true;
f9ead18c
JA		 658
659 sock = sock_from_file(req->file);
660 if (unlikely(!sock))
661 return -ENOTSOCK;
662
663 if (req_has_async_data(req)) {
664 kmsg = req->async_data;
665 } else {
666 ret = io_recvmsg_copy_hdr(req, &iomsg);
667 if (ret)
668 return ret;
669 kmsg = &iomsg;
670 }
671
672 if (!(req->flags & REQ_F_POLLED) &&
673 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 674 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 675
9bb66906 676retry_multishot:
f9ead18c
JA		 677 if (io_do_buffer_select(req)) {
678 void __user *buf;
9bb66906 679 size_t len = sr->len;
f9ead18c 680
9bb66906 681 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 682 if (!buf)
683 return -ENOBUFS;
9bb66906
DY		 684
685 if (req->flags & REQ_F_APOLL_MULTISHOT) {
686 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
687 if (ret) {
688 io_kbuf_recycle(req, issue_flags);
689 return ret;
690 }
691 }
692
f9ead18c 693 kmsg->fast_iov[0].iov_base = buf;
9bb66906 694 kmsg->fast_iov[0].iov_len = len;
f9ead18c 695 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, 1,
9bb66906 696 len);
f9ead18c
JA		 697 }
698
699 flags = sr->msg_flags;
700 if (force_nonblock)
701 flags |= MSG_DONTWAIT;
702 if (flags & MSG_WAITALL)
703 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
704
705 kmsg->msg.msg_get_inq = 1;
9bb66906
DY		 706 if (req->flags & REQ_F_APOLL_MULTISHOT)
707 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
708 &mshot_finished);
709 else
710 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
711 kmsg->uaddr, flags);
712
f9ead18c 713 if (ret < min_ret) {
9bb66906
DY		 714 if (ret == -EAGAIN && force_nonblock) {
715 ret = io_setup_async_msg(req, kmsg, issue_flags);
716 if (ret == -EAGAIN && (req->flags & IO_APOLL_MULTI_POLLED) ==
717 IO_APOLL_MULTI_POLLED) {
718 io_kbuf_recycle(req, issue_flags);
719 return IOU_ISSUE_SKIP_COMPLETE;
720 }
721 return ret;
722 }
f9ead18c
JA		 723 if (ret == -ERESTARTSYS)
724 ret = -EINTR;
725 if (ret > 0 && io_net_retry(sock, flags)) {
726 sr->done_io += ret;
727 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 728 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 729 }
730 req_set_fail(req);
731 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
732 req_set_fail(req);
733 }
734
d4e097da 735 if (ret > 0)
f9ead18c
JA		 736 ret += sr->done_io;
737 else if (sr->done_io)
738 ret = sr->done_io;
d4e097da
DY		 739 else
740 io_kbuf_recycle(req, issue_flags);
741
f9ead18c
JA		 742 cflags = io_put_kbuf(req, issue_flags);
743 if (kmsg->msg.msg_inq)
744 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 745
9bb66906
DY		 746 if (!io_recv_finish(req, &ret, cflags, mshot_finished))
747 goto retry_multishot;
748
749 if (mshot_finished) {
750 io_netmsg_recycle(req, issue_flags);
751 /* fast path, check for non-NULL to avoid function call */
752 if (kmsg->free_iov)
753 kfree(kmsg->free_iov);
754 req->flags &= ~REQ_F_NEED_CLEANUP;
755 }
756
757 return ret;
f9ead18c
JA		 758}
759
760int io_recv(struct io_kiocb *req, unsigned int issue_flags)
761{
f2ccb5ae 762 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 763 struct msghdr msg;
764 struct socket *sock;
765 struct iovec iov;
766 unsigned int cflags;
767 unsigned flags;
768 int ret, min_ret = 0;
769 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
b3fdea6e 770 size_t len = sr->len;
f9ead18c
JA		 771
772 if (!(req->flags & REQ_F_POLLED) &&
773 (sr->flags & IORING_RECVSEND_POLL_FIRST))
774 return -EAGAIN;
775
776 sock = sock_from_file(req->file);
777 if (unlikely(!sock))
778 return -ENOTSOCK;
779
b3fdea6e 780retry_multishot:
f9ead18c
JA		 781 if (io_do_buffer_select(req)) {
782 void __user *buf;
783
b3fdea6e 784 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 785 if (!buf)
786 return -ENOBUFS;
787 sr->buf = buf;
788 }
789
b3fdea6e 790 ret = import_single_range(READ, sr->buf, len, &iov, &msg.msg_iter);
f9ead18c
JA		 791 if (unlikely(ret))
792 goto out_free;
793
794 msg.msg_name = NULL;
795 msg.msg_namelen = 0;
796 msg.msg_control = NULL;
797 msg.msg_get_inq = 1;
798 msg.msg_flags = 0;
799 msg.msg_controllen = 0;
800 msg.msg_iocb = NULL;
e02b6651 801 msg.msg_ubuf = NULL;
f9ead18c
JA		 802
803 flags = sr->msg_flags;
804 if (force_nonblock)
805 flags |= MSG_DONTWAIT;
806 if (flags & MSG_WAITALL)
807 min_ret = iov_iter_count(&msg.msg_iter);
808
809 ret = sock_recvmsg(sock, &msg, flags);
810 if (ret < min_ret) {
b3fdea6e
DY		 811 if (ret == -EAGAIN && force_nonblock) {
812 if ((req->flags & IO_APOLL_MULTI_POLLED) == IO_APOLL_MULTI_POLLED) {
813 io_kbuf_recycle(req, issue_flags);
814 return IOU_ISSUE_SKIP_COMPLETE;
815 }
816
f9ead18c 817 return -EAGAIN;
b3fdea6e 818 }
f9ead18c
JA		 819 if (ret == -ERESTARTSYS)
820 ret = -EINTR;
821 if (ret > 0 && io_net_retry(sock, flags)) {
822 sr->len -= ret;
823 sr->buf += ret;
824 sr->done_io += ret;
825 req->flags |= REQ_F_PARTIAL_IO;
826 return -EAGAIN;
827 }
828 req_set_fail(req);
829 } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
830out_free:
831 req_set_fail(req);
832 }
833
d4e097da 834 if (ret > 0)
f9ead18c
JA		 835 ret += sr->done_io;
836 else if (sr->done_io)
837 ret = sr->done_io;
d4e097da
DY		 838 else
839 io_kbuf_recycle(req, issue_flags);
840
f9ead18c
JA		 841 cflags = io_put_kbuf(req, issue_flags);
842 if (msg.msg_inq)
843 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 844
9bb66906 845 if (!io_recv_finish(req, &ret, cflags, ret <= 0))
b3fdea6e
DY		 846 goto retry_multishot;
847
848 return ret;
f9ead18c
JA		 849}
850
06a5464b
PB		 851int io_sendzc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
852{
f2ccb5ae 853 struct io_sendzc *zc = io_kiocb_to_cmd(req, struct io_sendzc);
10c7d33e 854 struct io_ring_ctx *ctx = req->ctx;
06a5464b 855
092aeedb 856 if (READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3))
06a5464b
PB		 857 return -EINVAL;
858
859 zc->flags = READ_ONCE(sqe->ioprio);
63809137
PB		 860 if (zc->flags & ~(IORING_RECVSEND_POLL_FIRST |
861 IORING_RECVSEND_FIXED_BUF | IORING_RECVSEND_NOTIF_FLUSH))
06a5464b 862 return -EINVAL;
10c7d33e
PB		 863 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
864 unsigned idx = READ_ONCE(sqe->buf_index);
865
866 if (unlikely(idx >= ctx->nr_user_bufs))
867 return -EFAULT;
868 idx = array_index_nospec(idx, ctx->nr_user_bufs);
869 req->imu = READ_ONCE(ctx->user_bufs[idx]);
870 io_req_set_rsrc_node(req, ctx, 0);
871 }
06a5464b
PB		 872
873 zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
874 zc->len = READ_ONCE(sqe->len);
875 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
876 zc->slot_idx = READ_ONCE(sqe->notification_idx);
877 if (zc->msg_flags & MSG_DONTWAIT)
878 req->flags |= REQ_F_NOWAIT;
092aeedb
PB		 879
880 zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
881 zc->addr_len = READ_ONCE(sqe->addr_len);
4a933e62 882 zc->done_io = 0;
092aeedb 883
06a5464b
PB		 884#ifdef CONFIG_COMPAT
885 if (req->ctx->compat)
886 zc->msg_flags |= MSG_CMSG_COMPAT;
887#endif
888 return 0;
889}
890
3ff1a0d3
PB		 891static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
892 struct iov_iter *from, size_t length)
893{
894 struct skb_shared_info *shinfo = skb_shinfo(skb);
895 int frag = shinfo->nr_frags;
896 int ret = 0;
897 struct bvec_iter bi;
898 ssize_t copied = 0;
899 unsigned long truesize = 0;
900
901 if (!shinfo->nr_frags)
902 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
903
904 if (!skb_zcopy_managed(skb) || !iov_iter_is_bvec(from)) {
905 skb_zcopy_downgrade_managed(skb);
906 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
907 }
908
909 bi.bi_size = min(from->count, length);
910 bi.bi_bvec_done = from->iov_offset;
911 bi.bi_idx = 0;
912
913 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
914 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
915
916 copied += v.bv_len;
917 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
918 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
919 v.bv_offset, v.bv_len);
920 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
921 }
922 if (bi.bi_size)
923 ret = -EMSGSIZE;
924
925 shinfo->nr_frags = frag;
926 from->bvec += bi.bi_idx;
927 from->nr_segs -= bi.bi_idx;
928 from->count = bi.bi_size;
929 from->iov_offset = bi.bi_bvec_done;
930
931 skb->data_len += copied;
932 skb->len += copied;
933 skb->truesize += truesize;
934
935 if (sk && sk->sk_type == SOCK_STREAM) {
936 sk_wmem_queued_add(sk, truesize);
937 if (!skb_zcopy_pure(skb))
938 sk_mem_charge(sk, truesize);
939 } else {
940 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
941 }
942 return ret;
943}
944
06a5464b
PB		 945int io_sendzc(struct io_kiocb *req, unsigned int issue_flags)
946{
092aeedb 947 struct sockaddr_storage address;
06a5464b 948 struct io_ring_ctx *ctx = req->ctx;
f2ccb5ae 949 struct io_sendzc *zc = io_kiocb_to_cmd(req, struct io_sendzc);
06a5464b 950 struct io_notif_slot *notif_slot;
14b146b6 951 struct io_kiocb *notif;
06a5464b
PB		 952 struct msghdr msg;
953 struct iovec iov;
954 struct socket *sock;
955 unsigned msg_flags;
956 int ret, min_ret = 0;
957
958 if (!(req->flags & REQ_F_POLLED) &&
959 (zc->flags & IORING_RECVSEND_POLL_FIRST))
960 return -EAGAIN;
961
962 if (issue_flags & IO_URING_F_UNLOCKED)
963 return -EAGAIN;
964 sock = sock_from_file(req->file);
965 if (unlikely(!sock))
966 return -ENOTSOCK;
967
968 notif_slot = io_get_notif_slot(ctx, zc->slot_idx);
969 if (!notif_slot)
970 return -EINVAL;
971 notif = io_get_notif(ctx, notif_slot);
972 if (!notif)
973 return -ENOMEM;
974
975 msg.msg_name = NULL;
976 msg.msg_control = NULL;
977 msg.msg_controllen = 0;
978 msg.msg_namelen = 0;
979
86dc8f23
PB		 980 if (zc->addr) {
981 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &address);
982 if (unlikely(ret < 0))
983 return ret;
984 msg.msg_name = (struct sockaddr *)&address;
985 msg.msg_namelen = zc->addr_len;
986 }
987
10c7d33e
PB		 988 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
989 ret = io_import_fixed(WRITE, &msg.msg_iter, req->imu,
990 (u64)(uintptr_t)zc->buf, zc->len);
991 if (unlikely(ret))
992 return ret;
993 } else {
994 ret = import_single_range(WRITE, zc->buf, zc->len, &iov,
995 &msg.msg_iter);
996 if (unlikely(ret))
997 return ret;
6a9ce66f 998 ret = io_notif_account_mem(notif, zc->len);
2e32ba56
PB		 999 if (unlikely(ret))
1000 return ret;
10c7d33e 1001 }
06a5464b
PB		 1002
1003 msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1004 if (issue_flags & IO_URING_F_NONBLOCK)
1005 msg_flags |= MSG_DONTWAIT;
1006 if (msg_flags & MSG_WAITALL)
1007 min_ret = iov_iter_count(&msg.msg_iter);
1008
1009 msg.msg_flags = msg_flags;
14b146b6 1010 msg.msg_ubuf = &io_notif_to_data(notif)->uarg;
3ff1a0d3 1011 msg.sg_from_iter = io_sg_from_iter;
06a5464b
PB		 1012 ret = sock_sendmsg(sock, &msg);
1013
1014 if (unlikely(ret < min_ret)) {
1015 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1016 return -EAGAIN;
4a933e62
PB		 1017 if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1018 zc->len -= ret;
1019 zc->buf += ret;
1020 zc->done_io += ret;
1021 req->flags |= REQ_F_PARTIAL_IO;
1022 return -EAGAIN;
1023 }
1024 if (ret == -ERESTARTSYS)
1025 ret = -EINTR;
5a848b7c 1026 req_set_fail(req);
4a933e62
PB		 1027 } else if (zc->flags & IORING_RECVSEND_NOTIF_FLUSH) {
1028 io_notif_slot_flush_submit(notif_slot, 0);
06a5464b
PB		 1029 }
1030
4a933e62
PB		 1031 if (ret >= 0)
1032 ret += zc->done_io;
1033 else if (zc->done_io)
1034 ret = zc->done_io;
06a5464b
PB		 1035 io_req_set_res(req, ret, 0);
1036 return IOU_OK;
1037}
1038
f9ead18c
JA		 1039int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1040{
f2ccb5ae 1041 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1042 unsigned flags;
1043
1044 if (sqe->len || sqe->buf_index)
1045 return -EINVAL;
1046
1047 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1048 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1049 accept->flags = READ_ONCE(sqe->accept_flags);
1050 accept->nofile = rlimit(RLIMIT_NOFILE);
1051 flags = READ_ONCE(sqe->ioprio);
1052 if (flags & ~IORING_ACCEPT_MULTISHOT)
1053 return -EINVAL;
1054
1055 accept->file_slot = READ_ONCE(sqe->file_index);
1056 if (accept->file_slot) {
1057 if (accept->flags & SOCK_CLOEXEC)
1058 return -EINVAL;
1059 if (flags & IORING_ACCEPT_MULTISHOT &&
1060 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1061 return -EINVAL;
1062 }
1063 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1064 return -EINVAL;
1065 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1066 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1067 if (flags & IORING_ACCEPT_MULTISHOT)
1068 req->flags |= REQ_F_APOLL_MULTISHOT;
1069 return 0;
1070}
1071
1072int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1073{
1074 struct io_ring_ctx *ctx = req->ctx;
f2ccb5ae 1075 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1076 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1077 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1078 bool fixed = !!accept->file_slot;
1079 struct file *file;
1080 int ret, fd;
1081
1082retry:
1083 if (!fixed) {
1084 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1085 if (unlikely(fd < 0))
1086 return fd;
1087 }
1088 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1089 accept->flags);
1090 if (IS_ERR(file)) {
1091 if (!fixed)
1092 put_unused_fd(fd);
1093 ret = PTR_ERR(file);
1094 if (ret == -EAGAIN && force_nonblock) {
1095 /*
1096 * if it's multishot and polled, we don't need to
1097 * return EAGAIN to arm the poll infra since it
1098 * has already been done
1099 */
1100 if ((req->flags & IO_APOLL_MULTI_POLLED) ==
1101 IO_APOLL_MULTI_POLLED)
1102 ret = IOU_ISSUE_SKIP_COMPLETE;
1103 return ret;
1104 }
1105 if (ret == -ERESTARTSYS)
1106 ret = -EINTR;
1107 req_set_fail(req);
1108 } else if (!fixed) {
1109 fd_install(fd, file);
1110 ret = fd;
1111 } else {
1112 ret = io_fixed_fd_install(req, issue_flags, file,
1113 accept->file_slot);
1114 }
1115
1116 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1117 io_req_set_res(req, ret, 0);
1118 return IOU_OK;
1119 }
f9ead18c 1120
cbd25748
DY		 1121 if (ret >= 0 &&
1122 io_post_aux_cqe(ctx, req->cqe.user_data, ret, IORING_CQE_F_MORE, false))
d245bca6 1123 goto retry;
cbd25748
DY		 1124
1125 io_req_set_res(req, ret, 0);
1126 if (req->flags & REQ_F_POLLED)
1127 return IOU_STOP_MULTISHOT;
1128 return IOU_OK;
f9ead18c
JA		 1129}
1130
1131int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1132{
f2ccb5ae 1133 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1134
1135 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1136 return -EINVAL;
1137
1138 sock->domain = READ_ONCE(sqe->fd);
1139 sock->type = READ_ONCE(sqe->off);
1140 sock->protocol = READ_ONCE(sqe->len);
1141 sock->file_slot = READ_ONCE(sqe->file_index);
1142 sock->nofile = rlimit(RLIMIT_NOFILE);
1143
1144 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1145 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1146 return -EINVAL;
1147 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1148 return -EINVAL;
1149 return 0;
1150}
1151
1152int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1153{
f2ccb5ae 1154 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1155 bool fixed = !!sock->file_slot;
1156 struct file *file;
1157 int ret, fd;
1158
1159 if (!fixed) {
1160 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1161 if (unlikely(fd < 0))
1162 return fd;
1163 }
1164 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1165 if (IS_ERR(file)) {
1166 if (!fixed)
1167 put_unused_fd(fd);
1168 ret = PTR_ERR(file);
1169 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1170 return -EAGAIN;
1171 if (ret == -ERESTARTSYS)
1172 ret = -EINTR;
1173 req_set_fail(req);
1174 } else if (!fixed) {
1175 fd_install(fd, file);
1176 ret = fd;
1177 } else {
1178 ret = io_fixed_fd_install(req, issue_flags, file,
1179 sock->file_slot);
1180 }
1181 io_req_set_res(req, ret, 0);
1182 return IOU_OK;
1183}
1184
1185int io_connect_prep_async(struct io_kiocb *req)
1186{
1187 struct io_async_connect *io = req->async_data;
f2ccb5ae 1188 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1189
1190 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1191}
1192
1193int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1194{
f2ccb5ae 1195 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1196
1197 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1198 return -EINVAL;
1199
1200 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1201 conn->addr_len = READ_ONCE(sqe->addr2);
1202 return 0;
1203}
1204
1205int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1206{
f2ccb5ae 1207 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1208 struct io_async_connect __io, *io;
1209 unsigned file_flags;
1210 int ret;
1211 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1212
1213 if (req_has_async_data(req)) {
1214 io = req->async_data;
1215 } else {
1216 ret = move_addr_to_kernel(connect->addr,
1217 connect->addr_len,
1218 &__io.address);
1219 if (ret)
1220 goto out;
1221 io = &__io;
1222 }
1223
1224 file_flags = force_nonblock ? O_NONBLOCK : 0;
1225
1226 ret = __sys_connect_file(req->file, &io->address,
1227 connect->addr_len, file_flags);
1228 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
1229 if (req_has_async_data(req))
1230 return -EAGAIN;
1231 if (io_alloc_async_data(req)) {
1232 ret = -ENOMEM;
1233 goto out;
1234 }
1235 memcpy(req->async_data, &__io, sizeof(__io));
1236 return -EAGAIN;
1237 }
1238 if (ret == -ERESTARTSYS)
1239 ret = -EINTR;
1240out:
1241 if (ret < 0)
1242 req_set_fail(req);
1243 io_req_set_res(req, ret, 0);
1244 return IOU_OK;
1245}
43e0bbbd
JA		 1246
1247void io_netmsg_cache_free(struct io_cache_entry *entry)
1248{
1249 kfree(container_of(entry, struct io_async_msghdr, cache));
1250}
f9ead18c 1251#endif
Linux 6.x block layer and io_uring tree(s)