projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
io_uring/net: refactor io_sr_msg types
[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 };
0b048557
PB		 58 unsigned len;
59 unsigned done_io;
293402e5 60 unsigned msg_flags;
0b048557 61 u16 flags;
f9ead18c
JA		 62};
63
06a5464b
PB		 64struct io_sendzc {
65 struct file *file;
66 void __user *buf;
0b048557
PB		 67 unsigned len;
68 unsigned done_io;
06a5464b 69 unsigned msg_flags;
0b048557
PB		 70 u16 flags;
71 u16 addr_len;
092aeedb 72 void __user *addr;
b48c312b 73 struct io_kiocb *notif;
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
858c293e
PB		 129static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
130 unsigned int issue_flags)
43e0bbbd
JA		 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
858c293e
PB		 151static inline struct io_async_msghdr *io_msg_alloc_async_prep(struct io_kiocb *req)
152{
153 /* ->prep_async is always called from the submission context */
154 return io_msg_alloc_async(req, 0);
155}
156
f9ead18c 157static int io_setup_async_msg(struct io_kiocb *req,
43e0bbbd
JA		 158 struct io_async_msghdr *kmsg,
159 unsigned int issue_flags)
f9ead18c 160{
3f743e9b 161 struct io_async_msghdr *async_msg;
f9ead18c 162
3f743e9b 163 if (req_has_async_data(req))
f9ead18c 164 return -EAGAIN;
858c293e 165 async_msg = io_msg_alloc_async(req, issue_flags);
43e0bbbd 166 if (!async_msg) {
f9ead18c
JA		 167 kfree(kmsg->free_iov);
168 return -ENOMEM;
169 }
f9ead18c
JA		 170 req->flags |= REQ_F_NEED_CLEANUP;
171 memcpy(async_msg, kmsg, sizeof(*kmsg));
172 async_msg->msg.msg_name = &async_msg->addr;
173 /* if were using fast_iov, set it to the new one */
174 if (!async_msg->free_iov)
175 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
176
177 return -EAGAIN;
178}
179
180static int io_sendmsg_copy_hdr(struct io_kiocb *req,
181 struct io_async_msghdr *iomsg)
182{
f2ccb5ae 183 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 184
185 iomsg->msg.msg_name = &iomsg->addr;
186 iomsg->free_iov = iomsg->fast_iov;
187 return sendmsg_copy_msghdr(&iomsg->msg, sr->umsg, sr->msg_flags,
188 &iomsg->free_iov);
189}
190
581711c4
PB		 191int io_sendzc_prep_async(struct io_kiocb *req)
192{
193 struct io_sendzc *zc = io_kiocb_to_cmd(req, struct io_sendzc);
194 struct io_async_msghdr *io;
195 int ret;
196
197 if (!zc->addr || req_has_async_data(req))
198 return 0;
6bf8ad25
PB		 199 io = io_msg_alloc_async_prep(req);
200 if (!io)
581711c4 201 return -ENOMEM;
581711c4
PB		 202 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
203 return ret;
204}
205
206static int io_setup_async_addr(struct io_kiocb *req,
207 struct sockaddr_storage *addr,
208 unsigned int issue_flags)
209{
210 struct io_async_msghdr *io;
211
212 if (!addr || req_has_async_data(req))
213 return -EAGAIN;
6bf8ad25
PB		 214 io = io_msg_alloc_async(req, issue_flags);
215 if (!io)
581711c4 216 return -ENOMEM;
581711c4
PB		 217 memcpy(&io->addr, addr, sizeof(io->addr));
218 return -EAGAIN;
219}
220
f9ead18c
JA		 221int io_sendmsg_prep_async(struct io_kiocb *req)
222{
223 int ret;
224
858c293e
PB		 225 if (!io_msg_alloc_async_prep(req))
226 return -ENOMEM;
f9ead18c
JA		 227 ret = io_sendmsg_copy_hdr(req, req->async_data);
228 if (!ret)
229 req->flags |= REQ_F_NEED_CLEANUP;
230 return ret;
231}
232
233void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
234{
235 struct io_async_msghdr *io = req->async_data;
236
237 kfree(io->free_iov);
238}
239
240int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
241{
f2ccb5ae 242 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 243
244 if (unlikely(sqe->file_index || sqe->addr2))
245 return -EINVAL;
246
247 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
248 sr->len = READ_ONCE(sqe->len);
249 sr->flags = READ_ONCE(sqe->ioprio);
250 if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
251 return -EINVAL;
252 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
253 if (sr->msg_flags & MSG_DONTWAIT)
254 req->flags |= REQ_F_NOWAIT;
255
256#ifdef CONFIG_COMPAT
257 if (req->ctx->compat)
258 sr->msg_flags |= MSG_CMSG_COMPAT;
259#endif
260 sr->done_io = 0;
261 return 0;
262}
263
264int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
265{
f2ccb5ae 266 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 267 struct io_async_msghdr iomsg, *kmsg;
268 struct socket *sock;
269 unsigned flags;
270 int min_ret = 0;
271 int ret;
272
273 sock = sock_from_file(req->file);
274 if (unlikely(!sock))
275 return -ENOTSOCK;
276
277 if (req_has_async_data(req)) {
278 kmsg = req->async_data;
279 } else {
280 ret = io_sendmsg_copy_hdr(req, &iomsg);
281 if (ret)
282 return ret;
283 kmsg = &iomsg;
284 }
285
286 if (!(req->flags & REQ_F_POLLED) &&
287 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 288 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 289
290 flags = sr->msg_flags;
291 if (issue_flags & IO_URING_F_NONBLOCK)
292 flags |= MSG_DONTWAIT;
293 if (flags & MSG_WAITALL)
294 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
295
296 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
297
298 if (ret < min_ret) {
299 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
43e0bbbd 300 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 301 if (ret > 0 && io_net_retry(sock, flags)) {
302 sr->done_io += ret;
303 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 304 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 305 }
95eafc74
PB		 306 if (ret == -ERESTARTSYS)
307 ret = -EINTR;
f9ead18c
JA		 308 req_set_fail(req);
309 }
310 /* fast path, check for non-NULL to avoid function call */
311 if (kmsg->free_iov)
312 kfree(kmsg->free_iov);
313 req->flags &= ~REQ_F_NEED_CLEANUP;
43e0bbbd 314 io_netmsg_recycle(req, issue_flags);
f9ead18c
JA		 315 if (ret >= 0)
316 ret += sr->done_io;
317 else if (sr->done_io)
318 ret = sr->done_io;
319 io_req_set_res(req, ret, 0);
320 return IOU_OK;
321}
322
323int io_send(struct io_kiocb *req, unsigned int issue_flags)
324{
f2ccb5ae 325 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 326 struct msghdr msg;
327 struct iovec iov;
328 struct socket *sock;
329 unsigned flags;
330 int min_ret = 0;
331 int ret;
332
333 if (!(req->flags & REQ_F_POLLED) &&
334 (sr->flags & IORING_RECVSEND_POLL_FIRST))
335 return -EAGAIN;
336
337 sock = sock_from_file(req->file);
338 if (unlikely(!sock))
339 return -ENOTSOCK;
340
341 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
342 if (unlikely(ret))
343 return ret;
344
345 msg.msg_name = NULL;
346 msg.msg_control = NULL;
347 msg.msg_controllen = 0;
348 msg.msg_namelen = 0;
e02b6651 349 msg.msg_ubuf = NULL;
f9ead18c
JA		 350
351 flags = sr->msg_flags;
352 if (issue_flags & IO_URING_F_NONBLOCK)
353 flags |= MSG_DONTWAIT;
354 if (flags & MSG_WAITALL)
355 min_ret = iov_iter_count(&msg.msg_iter);
356
357 msg.msg_flags = flags;
358 ret = sock_sendmsg(sock, &msg);
359 if (ret < min_ret) {
360 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
361 return -EAGAIN;
f9ead18c
JA		 362 if (ret > 0 && io_net_retry(sock, flags)) {
363 sr->len -= ret;
364 sr->buf += ret;
365 sr->done_io += ret;
366 req->flags |= REQ_F_PARTIAL_IO;
367 return -EAGAIN;
368 }
95eafc74
PB		 369 if (ret == -ERESTARTSYS)
370 ret = -EINTR;
f9ead18c
JA		 371 req_set_fail(req);
372 }
373 if (ret >= 0)
374 ret += sr->done_io;
375 else if (sr->done_io)
376 ret = sr->done_io;
377 io_req_set_res(req, ret, 0);
378 return IOU_OK;
379}
380
9bb66906
DY		 381static bool io_recvmsg_multishot_overflow(struct io_async_msghdr *iomsg)
382{
9b0fc3c0 383 int hdr;
9bb66906 384
9b0fc3c0 385 if (iomsg->namelen < 0)
9bb66906 386 return true;
9b0fc3c0
DY		 387 if (check_add_overflow((int)sizeof(struct io_uring_recvmsg_out),
388 iomsg->namelen, &hdr))
9bb66906 389 return true;
9b0fc3c0 390 if (check_add_overflow(hdr, (int)iomsg->controllen, &hdr))
9bb66906
DY		 391 return true;
392
393 return false;
394}
395
f9ead18c
JA		 396static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
397 struct io_async_msghdr *iomsg)
398{
f2ccb5ae 399 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
7fa875b8 400 struct user_msghdr msg;
f9ead18c
JA		 401 int ret;
402
7fa875b8
DY		 403 if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
404 return -EFAULT;
405
406 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 407 if (ret)
408 return ret;
409
410 if (req->flags & REQ_F_BUFFER_SELECT) {
7fa875b8 411 if (msg.msg_iovlen == 0) {
5702196e
DY		 412 sr->len = iomsg->fast_iov[0].iov_len = 0;
413 iomsg->fast_iov[0].iov_base = NULL;
414 iomsg->free_iov = NULL;
7fa875b8 415 } else if (msg.msg_iovlen > 1) {
f9ead18c 416 return -EINVAL;
5702196e 417 } else {
7fa875b8 418 if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
5702196e
DY		 419 return -EFAULT;
420 sr->len = iomsg->fast_iov[0].iov_len;
421 iomsg->free_iov = NULL;
422 }
9bb66906
DY		 423
424 if (req->flags & REQ_F_APOLL_MULTISHOT) {
425 iomsg->namelen = msg.msg_namelen;
426 iomsg->controllen = msg.msg_controllen;
427 if (io_recvmsg_multishot_overflow(iomsg))
428 return -EOVERFLOW;
429 }
f9ead18c
JA		 430 } else {
431 iomsg->free_iov = iomsg->fast_iov;
7fa875b8 432 ret = __import_iovec(READ, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
f9ead18c
JA		 433 &iomsg->free_iov, &iomsg->msg.msg_iter,
434 false);
435 if (ret > 0)
436 ret = 0;
437 }
438
439 return ret;
440}
441
442#ifdef CONFIG_COMPAT
443static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
444 struct io_async_msghdr *iomsg)
445{
f2ccb5ae 446 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
72c531f8 447 struct compat_msghdr msg;
f9ead18c 448 struct compat_iovec __user *uiov;
f9ead18c
JA		 449 int ret;
450
72c531f8
DY		 451 if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
452 return -EFAULT;
453
4f6a94d3 454 ret = __get_compat_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 455 if (ret)
456 return ret;
457
72c531f8 458 uiov = compat_ptr(msg.msg_iov);
f9ead18c
JA		 459 if (req->flags & REQ_F_BUFFER_SELECT) {
460 compat_ssize_t clen;
461
72c531f8 462 if (msg.msg_iovlen == 0) {
6d2f75a0
DY		 463 sr->len = 0;
464 iomsg->free_iov = NULL;
72c531f8 465 } else if (msg.msg_iovlen > 1) {
f9ead18c 466 return -EINVAL;
6d2f75a0
DY		 467 } else {
468 if (!access_ok(uiov, sizeof(*uiov)))
469 return -EFAULT;
470 if (__get_user(clen, &uiov->iov_len))
471 return -EFAULT;
472 if (clen < 0)
473 return -EINVAL;
474 sr->len = clen;
475 iomsg->free_iov = NULL;
476 }
9bb66906
DY		 477
478 if (req->flags & REQ_F_APOLL_MULTISHOT) {
479 iomsg->namelen = msg.msg_namelen;
480 iomsg->controllen = msg.msg_controllen;
481 if (io_recvmsg_multishot_overflow(iomsg))
482 return -EOVERFLOW;
483 }
f9ead18c
JA		 484 } else {
485 iomsg->free_iov = iomsg->fast_iov;
72c531f8 486 ret = __import_iovec(READ, (struct iovec __user *)uiov, msg.msg_iovlen,
f9ead18c
JA		 487 UIO_FASTIOV, &iomsg->free_iov,
488 &iomsg->msg.msg_iter, true);
489 if (ret < 0)
490 return ret;
491 }
492
493 return 0;
494}
495#endif
496
497static int io_recvmsg_copy_hdr(struct io_kiocb *req,
498 struct io_async_msghdr *iomsg)
499{
500 iomsg->msg.msg_name = &iomsg->addr;
501
502#ifdef CONFIG_COMPAT
503 if (req->ctx->compat)
504 return __io_compat_recvmsg_copy_hdr(req, iomsg);
505#endif
506
507 return __io_recvmsg_copy_hdr(req, iomsg);
508}
509
510int io_recvmsg_prep_async(struct io_kiocb *req)
511{
512 int ret;
513
858c293e
PB		 514 if (!io_msg_alloc_async_prep(req))
515 return -ENOMEM;
f9ead18c
JA		 516 ret = io_recvmsg_copy_hdr(req, req->async_data);
517 if (!ret)
518 req->flags |= REQ_F_NEED_CLEANUP;
519 return ret;
520}
521
b3fdea6e
DY		 522#define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
523
f9ead18c
JA		 524int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
525{
f2ccb5ae 526 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 527
528 if (unlikely(sqe->file_index || sqe->addr2))
529 return -EINVAL;
530
531 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
532 sr->len = READ_ONCE(sqe->len);
533 sr->flags = READ_ONCE(sqe->ioprio);
b3fdea6e 534 if (sr->flags & ~(RECVMSG_FLAGS))
f9ead18c
JA		 535 return -EINVAL;
536 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
537 if (sr->msg_flags & MSG_DONTWAIT)
538 req->flags |= REQ_F_NOWAIT;
539 if (sr->msg_flags & MSG_ERRQUEUE)
540 req->flags |= REQ_F_CLEAR_POLLIN;
b3fdea6e
DY		 541 if (sr->flags & IORING_RECV_MULTISHOT) {
542 if (!(req->flags & REQ_F_BUFFER_SELECT))
543 return -EINVAL;
544 if (sr->msg_flags & MSG_WAITALL)
545 return -EINVAL;
546 if (req->opcode == IORING_OP_RECV && sr->len)
547 return -EINVAL;
548 req->flags |= REQ_F_APOLL_MULTISHOT;
549 }
f9ead18c
JA		 550
551#ifdef CONFIG_COMPAT
552 if (req->ctx->compat)
553 sr->msg_flags |= MSG_CMSG_COMPAT;
554#endif
555 sr->done_io = 0;
556 return 0;
557}
558
b3fdea6e
DY		 559static inline void io_recv_prep_retry(struct io_kiocb *req)
560{
f2ccb5ae 561 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
b3fdea6e
DY		 562
563 sr->done_io = 0;
564 sr->len = 0; /* get from the provided buffer */
565}
566
567/*
9bb66906 568 * Finishes io_recv and io_recvmsg.
b3fdea6e
DY		 569 *
570 * Returns true if it is actually finished, or false if it should run
571 * again (for multishot).
572 */
9bb66906
DY		 573static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
574 unsigned int cflags, bool mshot_finished)
b3fdea6e
DY		 575{
576 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
577 io_req_set_res(req, *ret, cflags);
578 *ret = IOU_OK;
579 return true;
580 }
581
9bb66906 582 if (!mshot_finished) {
b3fdea6e
DY		 583 if (io_post_aux_cqe(req->ctx, req->cqe.user_data, *ret,
584 cflags | IORING_CQE_F_MORE, false)) {
585 io_recv_prep_retry(req);
586 return false;
587 }
588 /*
589 * Otherwise stop multishot but use the current result.
590 * Probably will end up going into overflow, but this means
591 * we cannot trust the ordering anymore
592 */
593 }
594
595 io_req_set_res(req, *ret, cflags);
596
597 if (req->flags & REQ_F_POLLED)
598 *ret = IOU_STOP_MULTISHOT;
e2df2ccb
DY		 599 else
600 *ret = IOU_OK;
b3fdea6e
DY		 601 return true;
602}
603
9bb66906
DY		 604static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
605 struct io_sr_msg *sr, void __user **buf,
606 size_t *len)
607{
608 unsigned long ubuf = (unsigned long) *buf;
609 unsigned long hdr;
610
611 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
612 kmsg->controllen;
613 if (*len < hdr)
614 return -EFAULT;
615
616 if (kmsg->controllen) {
617 unsigned long control = ubuf + hdr - kmsg->controllen;
618
d1f6222c 619 kmsg->msg.msg_control_user = (void __user *) control;
9bb66906
DY		 620 kmsg->msg.msg_controllen = kmsg->controllen;
621 }
622
623 sr->buf = *buf; /* stash for later copy */
d1f6222c 624 *buf = (void __user *) (ubuf + hdr);
9bb66906
DY		 625 kmsg->payloadlen = *len = *len - hdr;
626 return 0;
627}
628
629struct io_recvmsg_multishot_hdr {
630 struct io_uring_recvmsg_out msg;
631 struct sockaddr_storage addr;
632};
633
634static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
635 struct io_async_msghdr *kmsg,
636 unsigned int flags, bool *finished)
637{
638 int err;
639 int copy_len;
640 struct io_recvmsg_multishot_hdr hdr;
641
642 if (kmsg->namelen)
643 kmsg->msg.msg_name = &hdr.addr;
644 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
645 kmsg->msg.msg_namelen = 0;
646
647 if (sock->file->f_flags & O_NONBLOCK)
648 flags |= MSG_DONTWAIT;
649
650 err = sock_recvmsg(sock, &kmsg->msg, flags);
651 *finished = err <= 0;
652 if (err < 0)
653 return err;
654
655 hdr.msg = (struct io_uring_recvmsg_out) {
656 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
657 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
658 };
659
660 hdr.msg.payloadlen = err;
661 if (err > kmsg->payloadlen)
662 err = kmsg->payloadlen;
663
664 copy_len = sizeof(struct io_uring_recvmsg_out);
665 if (kmsg->msg.msg_namelen > kmsg->namelen)
666 copy_len += kmsg->namelen;
667 else
668 copy_len += kmsg->msg.msg_namelen;
669
670 /*
671 * "fromlen shall refer to the value before truncation.."
672 * 1003.1g
673 */
674 hdr.msg.namelen = kmsg->msg.msg_namelen;
675
676 /* ensure that there is no gap between hdr and sockaddr_storage */
677 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
678 sizeof(struct io_uring_recvmsg_out));
679 if (copy_to_user(io->buf, &hdr, copy_len)) {
680 *finished = true;
681 return -EFAULT;
682 }
683
684 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
685 kmsg->controllen + err;
686}
687
f9ead18c
JA		 688int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
689{
f2ccb5ae 690 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 691 struct io_async_msghdr iomsg, *kmsg;
692 struct socket *sock;
693 unsigned int cflags;
694 unsigned flags;
695 int ret, min_ret = 0;
696 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
9bb66906 697 bool mshot_finished = true;
f9ead18c
JA		 698
699 sock = sock_from_file(req->file);
700 if (unlikely(!sock))
701 return -ENOTSOCK;
702
703 if (req_has_async_data(req)) {
704 kmsg = req->async_data;
705 } else {
706 ret = io_recvmsg_copy_hdr(req, &iomsg);
707 if (ret)
708 return ret;
709 kmsg = &iomsg;
710 }
711
712 if (!(req->flags & REQ_F_POLLED) &&
713 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 714 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 715
9bb66906 716retry_multishot:
f9ead18c
JA		 717 if (io_do_buffer_select(req)) {
718 void __user *buf;
9bb66906 719 size_t len = sr->len;
f9ead18c 720
9bb66906 721 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 722 if (!buf)
723 return -ENOBUFS;
9bb66906
DY		 724
725 if (req->flags & REQ_F_APOLL_MULTISHOT) {
726 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
727 if (ret) {
728 io_kbuf_recycle(req, issue_flags);
729 return ret;
730 }
731 }
732
f9ead18c 733 kmsg->fast_iov[0].iov_base = buf;
9bb66906 734 kmsg->fast_iov[0].iov_len = len;
f9ead18c 735 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, 1,
9bb66906 736 len);
f9ead18c
JA		 737 }
738
739 flags = sr->msg_flags;
740 if (force_nonblock)
741 flags |= MSG_DONTWAIT;
742 if (flags & MSG_WAITALL)
743 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
744
745 kmsg->msg.msg_get_inq = 1;
9bb66906
DY		 746 if (req->flags & REQ_F_APOLL_MULTISHOT)
747 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
748 &mshot_finished);
749 else
750 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
751 kmsg->uaddr, flags);
752
f9ead18c 753 if (ret < min_ret) {
9bb66906
DY		 754 if (ret == -EAGAIN && force_nonblock) {
755 ret = io_setup_async_msg(req, kmsg, issue_flags);
756 if (ret == -EAGAIN && (req->flags & IO_APOLL_MULTI_POLLED) ==
757 IO_APOLL_MULTI_POLLED) {
758 io_kbuf_recycle(req, issue_flags);
759 return IOU_ISSUE_SKIP_COMPLETE;
760 }
761 return ret;
762 }
f9ead18c
JA		 763 if (ret > 0 && io_net_retry(sock, flags)) {
764 sr->done_io += ret;
765 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 766 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 767 }
95eafc74
PB		 768 if (ret == -ERESTARTSYS)
769 ret = -EINTR;
f9ead18c
JA		 770 req_set_fail(req);
771 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
772 req_set_fail(req);
773 }
774
d4e097da 775 if (ret > 0)
f9ead18c
JA		 776 ret += sr->done_io;
777 else if (sr->done_io)
778 ret = sr->done_io;
d4e097da
DY		 779 else
780 io_kbuf_recycle(req, issue_flags);
781
f9ead18c
JA		 782 cflags = io_put_kbuf(req, issue_flags);
783 if (kmsg->msg.msg_inq)
784 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 785
9bb66906
DY		 786 if (!io_recv_finish(req, &ret, cflags, mshot_finished))
787 goto retry_multishot;
788
789 if (mshot_finished) {
790 io_netmsg_recycle(req, issue_flags);
791 /* fast path, check for non-NULL to avoid function call */
792 if (kmsg->free_iov)
793 kfree(kmsg->free_iov);
794 req->flags &= ~REQ_F_NEED_CLEANUP;
795 }
796
797 return ret;
f9ead18c
JA		 798}
799
800int io_recv(struct io_kiocb *req, unsigned int issue_flags)
801{
f2ccb5ae 802 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 803 struct msghdr msg;
804 struct socket *sock;
805 struct iovec iov;
806 unsigned int cflags;
807 unsigned flags;
808 int ret, min_ret = 0;
809 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
b3fdea6e 810 size_t len = sr->len;
f9ead18c
JA		 811
812 if (!(req->flags & REQ_F_POLLED) &&
813 (sr->flags & IORING_RECVSEND_POLL_FIRST))
814 return -EAGAIN;
815
816 sock = sock_from_file(req->file);
817 if (unlikely(!sock))
818 return -ENOTSOCK;
819
b3fdea6e 820retry_multishot:
f9ead18c
JA		 821 if (io_do_buffer_select(req)) {
822 void __user *buf;
823
b3fdea6e 824 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 825 if (!buf)
826 return -ENOBUFS;
827 sr->buf = buf;
828 }
829
b3fdea6e 830 ret = import_single_range(READ, sr->buf, len, &iov, &msg.msg_iter);
f9ead18c
JA		 831 if (unlikely(ret))
832 goto out_free;
833
834 msg.msg_name = NULL;
835 msg.msg_namelen = 0;
836 msg.msg_control = NULL;
837 msg.msg_get_inq = 1;
838 msg.msg_flags = 0;
839 msg.msg_controllen = 0;
840 msg.msg_iocb = NULL;
e02b6651 841 msg.msg_ubuf = NULL;
f9ead18c
JA		 842
843 flags = sr->msg_flags;
844 if (force_nonblock)
845 flags |= MSG_DONTWAIT;
846 if (flags & MSG_WAITALL)
847 min_ret = iov_iter_count(&msg.msg_iter);
848
849 ret = sock_recvmsg(sock, &msg, flags);
850 if (ret < min_ret) {
b3fdea6e
DY		 851 if (ret == -EAGAIN && force_nonblock) {
852 if ((req->flags & IO_APOLL_MULTI_POLLED) == IO_APOLL_MULTI_POLLED) {
853 io_kbuf_recycle(req, issue_flags);
854 return IOU_ISSUE_SKIP_COMPLETE;
855 }
856
f9ead18c 857 return -EAGAIN;
b3fdea6e 858 }
f9ead18c
JA		 859 if (ret > 0 && io_net_retry(sock, flags)) {
860 sr->len -= ret;
861 sr->buf += ret;
862 sr->done_io += ret;
863 req->flags |= REQ_F_PARTIAL_IO;
864 return -EAGAIN;
865 }
95eafc74
PB		 866 if (ret == -ERESTARTSYS)
867 ret = -EINTR;
f9ead18c
JA		 868 req_set_fail(req);
869 } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
870out_free:
871 req_set_fail(req);
872 }
873
d4e097da 874 if (ret > 0)
f9ead18c
JA		 875 ret += sr->done_io;
876 else if (sr->done_io)
877 ret = sr->done_io;
d4e097da
DY		 878 else
879 io_kbuf_recycle(req, issue_flags);
880
f9ead18c
JA		 881 cflags = io_put_kbuf(req, issue_flags);
882 if (msg.msg_inq)
883 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 884
9bb66906 885 if (!io_recv_finish(req, &ret, cflags, ret <= 0))
b3fdea6e
DY		 886 goto retry_multishot;
887
888 return ret;
f9ead18c
JA		 889}
890
b48c312b
PB		 891void io_sendzc_cleanup(struct io_kiocb *req)
892{
893 struct io_sendzc *zc = io_kiocb_to_cmd(req, struct io_sendzc);
894
895 zc->notif->flags |= REQ_F_CQE_SKIP;
896 io_notif_flush(zc->notif);
897 zc->notif = NULL;
898}
899
06a5464b
PB		 900int io_sendzc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
901{
f2ccb5ae 902 struct io_sendzc *zc = io_kiocb_to_cmd(req, struct io_sendzc);
10c7d33e 903 struct io_ring_ctx *ctx = req->ctx;
b48c312b 904 struct io_kiocb *notif;
06a5464b 905
b48c312b
PB		 906 if (READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3) ||
907 READ_ONCE(sqe->__pad3[0]))
908 return -EINVAL;
909 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
910 if (req->flags & REQ_F_CQE_SKIP)
06a5464b
PB		 911 return -EINVAL;
912
913 zc->flags = READ_ONCE(sqe->ioprio);
63809137 914 if (zc->flags & ~(IORING_RECVSEND_POLL_FIRST |
57f33224 915 IORING_RECVSEND_FIXED_BUF))
06a5464b 916 return -EINVAL;
e3366e02
PB		 917 notif = zc->notif = io_alloc_notif(ctx);
918 if (!notif)
919 return -ENOMEM;
920 notif->cqe.user_data = req->cqe.user_data;
921 notif->cqe.res = 0;
922 notif->cqe.flags = IORING_CQE_F_NOTIF;
923 req->flags |= REQ_F_NEED_CLEANUP;
10c7d33e
PB		 924 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
925 unsigned idx = READ_ONCE(sqe->buf_index);
926
927 if (unlikely(idx >= ctx->nr_user_bufs))
928 return -EFAULT;
929 idx = array_index_nospec(idx, ctx->nr_user_bufs);
930 req->imu = READ_ONCE(ctx->user_bufs[idx]);
e3366e02 931 io_req_set_rsrc_node(notif, ctx, 0);
10c7d33e 932 }
06a5464b
PB		 933
934 zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
935 zc->len = READ_ONCE(sqe->len);
936 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
06a5464b
PB		 937 if (zc->msg_flags & MSG_DONTWAIT)
938 req->flags |= REQ_F_NOWAIT;
092aeedb
PB		 939
940 zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
941 zc->addr_len = READ_ONCE(sqe->addr_len);
4a933e62 942 zc->done_io = 0;
092aeedb 943
06a5464b
PB		 944#ifdef CONFIG_COMPAT
945 if (req->ctx->compat)
946 zc->msg_flags |= MSG_CMSG_COMPAT;
947#endif
948 return 0;
949}
950
cd9021e8
PB		 951static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
952 struct iov_iter *from, size_t length)
953{
954 skb_zcopy_downgrade_managed(skb);
955 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
956}
957
3ff1a0d3
PB		 958static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
959 struct iov_iter *from, size_t length)
960{
961 struct skb_shared_info *shinfo = skb_shinfo(skb);
962 int frag = shinfo->nr_frags;
963 int ret = 0;
964 struct bvec_iter bi;
965 ssize_t copied = 0;
966 unsigned long truesize = 0;
967
cd9021e8 968 if (!frag)
3ff1a0d3 969 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
cd9021e8 970 else if (unlikely(!skb_zcopy_managed(skb)))
3ff1a0d3 971 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
3ff1a0d3
PB		 972
973 bi.bi_size = min(from->count, length);
974 bi.bi_bvec_done = from->iov_offset;
975 bi.bi_idx = 0;
976
977 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
978 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
979
980 copied += v.bv_len;
981 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
982 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
983 v.bv_offset, v.bv_len);
984 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
985 }
986 if (bi.bi_size)
987 ret = -EMSGSIZE;
988
989 shinfo->nr_frags = frag;
990 from->bvec += bi.bi_idx;
991 from->nr_segs -= bi.bi_idx;
dfb58b17 992 from->count -= copied;
3ff1a0d3
PB		 993 from->iov_offset = bi.bi_bvec_done;
994
995 skb->data_len += copied;
996 skb->len += copied;
997 skb->truesize += truesize;
998
999 if (sk && sk->sk_type == SOCK_STREAM) {
1000 sk_wmem_queued_add(sk, truesize);
1001 if (!skb_zcopy_pure(skb))
1002 sk_mem_charge(sk, truesize);
1003 } else {
1004 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1005 }
1006 return ret;
1007}
1008
06a5464b
PB		 1009int io_sendzc(struct io_kiocb *req, unsigned int issue_flags)
1010{
581711c4 1011 struct sockaddr_storage __address, *addr = NULL;
f2ccb5ae 1012 struct io_sendzc *zc = io_kiocb_to_cmd(req, struct io_sendzc);
06a5464b
PB		 1013 struct msghdr msg;
1014 struct iovec iov;
1015 struct socket *sock;
b48c312b 1016 unsigned msg_flags, cflags;
06a5464b
PB		 1017 int ret, min_ret = 0;
1018
06a5464b
PB		 1019 sock = sock_from_file(req->file);
1020 if (unlikely(!sock))
1021 return -ENOTSOCK;
1022
06a5464b
PB		 1023 msg.msg_name = NULL;
1024 msg.msg_control = NULL;
1025 msg.msg_controllen = 0;
1026 msg.msg_namelen = 0;
1027
86dc8f23 1028 if (zc->addr) {
581711c4
PB		 1029 if (req_has_async_data(req)) {
1030 struct io_async_msghdr *io = req->async_data;
1031
1032 msg.msg_name = addr = &io->addr;
1033 } else {
1034 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1035 if (unlikely(ret < 0))
1036 return ret;
1037 msg.msg_name = (struct sockaddr *)&__address;
1038 addr = &__address;
1039 }
86dc8f23
PB		 1040 msg.msg_namelen = zc->addr_len;
1041 }
1042
3c840053
PB		 1043 if (!(req->flags & REQ_F_POLLED) &&
1044 (zc->flags & IORING_RECVSEND_POLL_FIRST))
1045 return io_setup_async_addr(req, addr, issue_flags);
1046
10c7d33e
PB		 1047 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1048 ret = io_import_fixed(WRITE, &msg.msg_iter, req->imu,
1049 (u64)(uintptr_t)zc->buf, zc->len);
1050 if (unlikely(ret))
986e263d 1051 return ret;
cd9021e8 1052 msg.sg_from_iter = io_sg_from_iter;
10c7d33e
PB		 1053 } else {
1054 ret = import_single_range(WRITE, zc->buf, zc->len, &iov,
1055 &msg.msg_iter);
1056 if (unlikely(ret))
1057 return ret;
b48c312b 1058 ret = io_notif_account_mem(zc->notif, zc->len);
2e32ba56
PB		 1059 if (unlikely(ret))
1060 return ret;
cd9021e8 1061 msg.sg_from_iter = io_sg_from_iter_iovec;
10c7d33e 1062 }
06a5464b
PB		 1063
1064 msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1065 if (issue_flags & IO_URING_F_NONBLOCK)
1066 msg_flags |= MSG_DONTWAIT;
1067 if (msg_flags & MSG_WAITALL)
1068 min_ret = iov_iter_count(&msg.msg_iter);
1069
1070 msg.msg_flags = msg_flags;
b48c312b 1071 msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
06a5464b
PB		 1072 ret = sock_sendmsg(sock, &msg);
1073
1074 if (unlikely(ret < min_ret)) {
1075 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
581711c4
PB		 1076 return io_setup_async_addr(req, addr, issue_flags);
1077
4a933e62
PB		 1078 if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1079 zc->len -= ret;
1080 zc->buf += ret;
1081 zc->done_io += ret;
1082 req->flags |= REQ_F_PARTIAL_IO;
581711c4 1083 return io_setup_async_addr(req, addr, issue_flags);
4a933e62 1084 }
b48c312b
PB		 1085 if (ret < 0 && !zc->done_io)
1086 zc->notif->flags |= REQ_F_CQE_SKIP;
4a933e62
PB		 1087 if (ret == -ERESTARTSYS)
1088 ret = -EINTR;
5a848b7c 1089 req_set_fail(req);
06a5464b
PB		 1090 }
1091
4a933e62
PB		 1092 if (ret >= 0)
1093 ret += zc->done_io;
1094 else if (zc->done_io)
1095 ret = zc->done_io;
b48c312b
PB		 1096
1097 io_notif_flush(zc->notif);
1098 req->flags &= ~REQ_F_NEED_CLEANUP;
1099 cflags = ret >= 0 ? IORING_CQE_F_MORE : 0;
1100 io_req_set_res(req, ret, cflags);
06a5464b
PB		 1101 return IOU_OK;
1102}
1103
f9ead18c
JA		 1104int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1105{
f2ccb5ae 1106 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1107 unsigned flags;
1108
1109 if (sqe->len || sqe->buf_index)
1110 return -EINVAL;
1111
1112 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1113 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1114 accept->flags = READ_ONCE(sqe->accept_flags);
1115 accept->nofile = rlimit(RLIMIT_NOFILE);
1116 flags = READ_ONCE(sqe->ioprio);
1117 if (flags & ~IORING_ACCEPT_MULTISHOT)
1118 return -EINVAL;
1119
1120 accept->file_slot = READ_ONCE(sqe->file_index);
1121 if (accept->file_slot) {
1122 if (accept->flags & SOCK_CLOEXEC)
1123 return -EINVAL;
1124 if (flags & IORING_ACCEPT_MULTISHOT &&
1125 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1126 return -EINVAL;
1127 }
1128 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1129 return -EINVAL;
1130 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1131 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1132 if (flags & IORING_ACCEPT_MULTISHOT)
1133 req->flags |= REQ_F_APOLL_MULTISHOT;
1134 return 0;
1135}
1136
1137int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1138{
1139 struct io_ring_ctx *ctx = req->ctx;
f2ccb5ae 1140 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1141 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1142 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1143 bool fixed = !!accept->file_slot;
1144 struct file *file;
1145 int ret, fd;
1146
1147retry:
1148 if (!fixed) {
1149 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1150 if (unlikely(fd < 0))
1151 return fd;
1152 }
1153 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1154 accept->flags);
1155 if (IS_ERR(file)) {
1156 if (!fixed)
1157 put_unused_fd(fd);
1158 ret = PTR_ERR(file);
1159 if (ret == -EAGAIN && force_nonblock) {
1160 /*
1161 * if it's multishot and polled, we don't need to
1162 * return EAGAIN to arm the poll infra since it
1163 * has already been done
1164 */
1165 if ((req->flags & IO_APOLL_MULTI_POLLED) ==
1166 IO_APOLL_MULTI_POLLED)
1167 ret = IOU_ISSUE_SKIP_COMPLETE;
1168 return ret;
1169 }
1170 if (ret == -ERESTARTSYS)
1171 ret = -EINTR;
1172 req_set_fail(req);
1173 } else if (!fixed) {
1174 fd_install(fd, file);
1175 ret = fd;
1176 } else {
1177 ret = io_fixed_fd_install(req, issue_flags, file,
1178 accept->file_slot);
1179 }
1180
1181 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1182 io_req_set_res(req, ret, 0);
1183 return IOU_OK;
1184 }
f9ead18c 1185
cbd25748
DY		 1186 if (ret >= 0 &&
1187 io_post_aux_cqe(ctx, req->cqe.user_data, ret, IORING_CQE_F_MORE, false))
d245bca6 1188 goto retry;
cbd25748
DY		 1189
1190 io_req_set_res(req, ret, 0);
1191 if (req->flags & REQ_F_POLLED)
1192 return IOU_STOP_MULTISHOT;
1193 return IOU_OK;
f9ead18c
JA		 1194}
1195
1196int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1197{
f2ccb5ae 1198 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1199
1200 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1201 return -EINVAL;
1202
1203 sock->domain = READ_ONCE(sqe->fd);
1204 sock->type = READ_ONCE(sqe->off);
1205 sock->protocol = READ_ONCE(sqe->len);
1206 sock->file_slot = READ_ONCE(sqe->file_index);
1207 sock->nofile = rlimit(RLIMIT_NOFILE);
1208
1209 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1210 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1211 return -EINVAL;
1212 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1213 return -EINVAL;
1214 return 0;
1215}
1216
1217int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1218{
f2ccb5ae 1219 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1220 bool fixed = !!sock->file_slot;
1221 struct file *file;
1222 int ret, fd;
1223
1224 if (!fixed) {
1225 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1226 if (unlikely(fd < 0))
1227 return fd;
1228 }
1229 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1230 if (IS_ERR(file)) {
1231 if (!fixed)
1232 put_unused_fd(fd);
1233 ret = PTR_ERR(file);
1234 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1235 return -EAGAIN;
1236 if (ret == -ERESTARTSYS)
1237 ret = -EINTR;
1238 req_set_fail(req);
1239 } else if (!fixed) {
1240 fd_install(fd, file);
1241 ret = fd;
1242 } else {
1243 ret = io_fixed_fd_install(req, issue_flags, file,
1244 sock->file_slot);
1245 }
1246 io_req_set_res(req, ret, 0);
1247 return IOU_OK;
1248}
1249
1250int io_connect_prep_async(struct io_kiocb *req)
1251{
1252 struct io_async_connect *io = req->async_data;
f2ccb5ae 1253 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1254
1255 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1256}
1257
1258int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1259{
f2ccb5ae 1260 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1261
1262 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1263 return -EINVAL;
1264
1265 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1266 conn->addr_len = READ_ONCE(sqe->addr2);
1267 return 0;
1268}
1269
1270int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1271{
f2ccb5ae 1272 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1273 struct io_async_connect __io, *io;
1274 unsigned file_flags;
1275 int ret;
1276 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1277
1278 if (req_has_async_data(req)) {
1279 io = req->async_data;
1280 } else {
1281 ret = move_addr_to_kernel(connect->addr,
1282 connect->addr_len,
1283 &__io.address);
1284 if (ret)
1285 goto out;
1286 io = &__io;
1287 }
1288
1289 file_flags = force_nonblock ? O_NONBLOCK : 0;
1290
1291 ret = __sys_connect_file(req->file, &io->address,
1292 connect->addr_len, file_flags);
1293 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
1294 if (req_has_async_data(req))
1295 return -EAGAIN;
1296 if (io_alloc_async_data(req)) {
1297 ret = -ENOMEM;
1298 goto out;
1299 }
1300 memcpy(req->async_data, &__io, sizeof(__io));
1301 return -EAGAIN;
1302 }
1303 if (ret == -ERESTARTSYS)
1304 ret = -EINTR;
1305out:
1306 if (ret < 0)
1307 req_set_fail(req);
1308 io_req_set_res(req, ret, 0);
1309 return IOU_OK;
1310}
43e0bbbd
JA		 1311
1312void io_netmsg_cache_free(struct io_cache_entry *entry)
1313{
1314 kfree(container_of(entry, struct io_async_msghdr, cache));
1315}
f9ead18c 1316#endif
Linux 6.x block layer and io_uring tree(s)