projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
io_uring: don't gate task_work run on TIF_NOTIFY_SIGNAL
[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;
516e82f0 62 /* initialised and used only by !msg send variants */
0b048557 63 u16 addr_len;
092aeedb 64 void __user *addr;
516e82f0 65 /* used only for send zerocopy */
b48c312b 66 struct io_kiocb *notif;
06a5464b
PB		 67};
68
f9ead18c
JA		 69#define IO_APOLL_MULTI_POLLED (REQ_F_APOLL_MULTISHOT | REQ_F_POLLED)
70
71int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
72{
f2ccb5ae 73 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
f9ead18c
JA		 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
83int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
84{
f2ccb5ae 85 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
f9ead18c
JA		 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
101static 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
43e0bbbd
JA		 108static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
109{
110 struct io_async_msghdr *hdr = req->async_data;
111
06360426 112 if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
43e0bbbd
JA		 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
858c293e
PB		 122static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
123 unsigned int issue_flags)
43e0bbbd
JA		 124{
125 struct io_ring_ctx *ctx = req->ctx;
126 struct io_cache_entry *entry;
4c17a496 127 struct io_async_msghdr *hdr;
43e0bbbd
JA		 128
129 if (!(issue_flags & IO_URING_F_UNLOCKED) &&
130 (entry = io_alloc_cache_get(&ctx->netmsg_cache)) != NULL) {
43e0bbbd 131 hdr = container_of(entry, struct io_async_msghdr, cache);
4c17a496 132 hdr->free_iov = NULL;
43e0bbbd
JA		 133 req->flags |= REQ_F_ASYNC_DATA;
134 req->async_data = hdr;
135 return hdr;
136 }
137
4c17a496
PB		 138 if (!io_alloc_async_data(req)) {
139 hdr = req->async_data;
140 hdr->free_iov = NULL;
141 return hdr;
142 }
43e0bbbd
JA		 143 return NULL;
144}
145
858c293e
PB		 146static 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
f9ead18c 152static int io_setup_async_msg(struct io_kiocb *req,
43e0bbbd
JA		 153 struct io_async_msghdr *kmsg,
154 unsigned int issue_flags)
f9ead18c 155{
3f743e9b 156 struct io_async_msghdr *async_msg;
f9ead18c 157
3f743e9b 158 if (req_has_async_data(req))
f9ead18c 159 return -EAGAIN;
858c293e 160 async_msg = io_msg_alloc_async(req, issue_flags);
43e0bbbd 161 if (!async_msg) {
f9ead18c
JA		 162 kfree(kmsg->free_iov);
163 return -ENOMEM;
164 }
f9ead18c
JA		 165 req->flags |= REQ_F_NEED_CLEANUP;
166 memcpy(async_msg, kmsg, sizeof(*kmsg));
167 async_msg->msg.msg_name = &async_msg->addr;
168 /* if were using fast_iov, set it to the new one */
3e4cb6eb
SM		 169 if (!kmsg->free_iov) {
170 size_t fast_idx = kmsg->msg.msg_iter.iov - kmsg->fast_iov;
171 async_msg->msg.msg_iter.iov = &async_msg->fast_iov[fast_idx];
172 }
f9ead18c
JA		 173
174 return -EAGAIN;
175}
176
177static int io_sendmsg_copy_hdr(struct io_kiocb *req,
178 struct io_async_msghdr *iomsg)
179{
f2ccb5ae 180 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 181
182 iomsg->msg.msg_name = &iomsg->addr;
183 iomsg->free_iov = iomsg->fast_iov;
184 return sendmsg_copy_msghdr(&iomsg->msg, sr->umsg, sr->msg_flags,
185 &iomsg->free_iov);
186}
187
516e82f0 188int io_send_prep_async(struct io_kiocb *req)
581711c4 189{
ac9e5784 190 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
581711c4
PB		 191 struct io_async_msghdr *io;
192 int ret;
193
194 if (!zc->addr || req_has_async_data(req))
195 return 0;
6bf8ad25
PB		 196 io = io_msg_alloc_async_prep(req);
197 if (!io)
581711c4 198 return -ENOMEM;
581711c4
PB		 199 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
200 return ret;
201}
202
203static int io_setup_async_addr(struct io_kiocb *req,
6ae61b7a 204 struct sockaddr_storage *addr_storage,
581711c4
PB		 205 unsigned int issue_flags)
206{
6ae61b7a 207 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
581711c4
PB		 208 struct io_async_msghdr *io;
209
6ae61b7a 210 if (!sr->addr || req_has_async_data(req))
581711c4 211 return -EAGAIN;
6bf8ad25
PB		 212 io = io_msg_alloc_async(req, issue_flags);
213 if (!io)
581711c4 214 return -ENOMEM;
6ae61b7a 215 memcpy(&io->addr, addr_storage, sizeof(io->addr));
581711c4
PB		 216 return -EAGAIN;
217}
218
f9ead18c
JA		 219int io_sendmsg_prep_async(struct io_kiocb *req)
220{
221 int ret;
222
858c293e
PB		 223 if (!io_msg_alloc_async_prep(req))
224 return -ENOMEM;
f9ead18c
JA		 225 ret = io_sendmsg_copy_hdr(req, req->async_data);
226 if (!ret)
227 req->flags |= REQ_F_NEED_CLEANUP;
228 return ret;
229}
230
231void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
232{
233 struct io_async_msghdr *io = req->async_data;
234
235 kfree(io->free_iov);
236}
237
238int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
239{
f2ccb5ae 240 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c 241
516e82f0
PB		 242 if (req->opcode == IORING_OP_SEND) {
243 if (READ_ONCE(sqe->__pad3[0]))
244 return -EINVAL;
245 sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
246 sr->addr_len = READ_ONCE(sqe->addr_len);
247 } else if (sqe->addr2 || sqe->file_index) {
f9ead18c 248 return -EINVAL;
516e82f0 249 }
f9ead18c
JA		 250
251 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
252 sr->len = READ_ONCE(sqe->len);
253 sr->flags = READ_ONCE(sqe->ioprio);
254 if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
255 return -EINVAL;
256 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
257 if (sr->msg_flags & MSG_DONTWAIT)
258 req->flags |= REQ_F_NOWAIT;
259
260#ifdef CONFIG_COMPAT
261 if (req->ctx->compat)
262 sr->msg_flags |= MSG_CMSG_COMPAT;
263#endif
264 sr->done_io = 0;
265 return 0;
266}
267
268int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
269{
f2ccb5ae 270 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 271 struct io_async_msghdr iomsg, *kmsg;
272 struct socket *sock;
273 unsigned flags;
274 int min_ret = 0;
275 int ret;
276
277 sock = sock_from_file(req->file);
278 if (unlikely(!sock))
279 return -ENOTSOCK;
280
281 if (req_has_async_data(req)) {
282 kmsg = req->async_data;
283 } else {
284 ret = io_sendmsg_copy_hdr(req, &iomsg);
285 if (ret)
286 return ret;
287 kmsg = &iomsg;
288 }
289
290 if (!(req->flags & REQ_F_POLLED) &&
291 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 292 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 293
294 flags = sr->msg_flags;
295 if (issue_flags & IO_URING_F_NONBLOCK)
296 flags |= MSG_DONTWAIT;
297 if (flags & MSG_WAITALL)
298 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
299
300 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
301
302 if (ret < min_ret) {
303 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
43e0bbbd 304 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 305 if (ret > 0 && io_net_retry(sock, flags)) {
306 sr->done_io += ret;
307 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 308 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 309 }
95eafc74
PB		 310 if (ret == -ERESTARTSYS)
311 ret = -EINTR;
f9ead18c
JA		 312 req_set_fail(req);
313 }
314 /* fast path, check for non-NULL to avoid function call */
315 if (kmsg->free_iov)
316 kfree(kmsg->free_iov);
317 req->flags &= ~REQ_F_NEED_CLEANUP;
43e0bbbd 318 io_netmsg_recycle(req, issue_flags);
f9ead18c
JA		 319 if (ret >= 0)
320 ret += sr->done_io;
321 else if (sr->done_io)
322 ret = sr->done_io;
323 io_req_set_res(req, ret, 0);
324 return IOU_OK;
325}
326
327int io_send(struct io_kiocb *req, unsigned int issue_flags)
328{
516e82f0 329 struct sockaddr_storage __address;
f2ccb5ae 330 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 331 struct msghdr msg;
332 struct iovec iov;
333 struct socket *sock;
334 unsigned flags;
335 int min_ret = 0;
336 int ret;
337
04360d3e
PB		 338 msg.msg_name = NULL;
339 msg.msg_control = NULL;
340 msg.msg_controllen = 0;
341 msg.msg_namelen = 0;
342 msg.msg_ubuf = NULL;
343
516e82f0
PB		 344 if (sr->addr) {
345 if (req_has_async_data(req)) {
346 struct io_async_msghdr *io = req->async_data;
347
348 msg.msg_name = &io->addr;
349 } else {
350 ret = move_addr_to_kernel(sr->addr, sr->addr_len, &__address);
351 if (unlikely(ret < 0))
352 return ret;
353 msg.msg_name = (struct sockaddr *)&__address;
354 }
355 msg.msg_namelen = sr->addr_len;
356 }
357
f9ead18c
JA		 358 if (!(req->flags & REQ_F_POLLED) &&
359 (sr->flags & IORING_RECVSEND_POLL_FIRST))
516e82f0 360 return io_setup_async_addr(req, &__address, issue_flags);
f9ead18c
JA		 361
362 sock = sock_from_file(req->file);
363 if (unlikely(!sock))
364 return -ENOTSOCK;
365
366 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
367 if (unlikely(ret))
368 return ret;
369
f9ead18c
JA		 370 flags = sr->msg_flags;
371 if (issue_flags & IO_URING_F_NONBLOCK)
372 flags |= MSG_DONTWAIT;
373 if (flags & MSG_WAITALL)
374 min_ret = iov_iter_count(&msg.msg_iter);
375
376 msg.msg_flags = flags;
377 ret = sock_sendmsg(sock, &msg);
378 if (ret < min_ret) {
379 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
516e82f0
PB		 380 return io_setup_async_addr(req, &__address, issue_flags);
381
f9ead18c
JA		 382 if (ret > 0 && io_net_retry(sock, flags)) {
383 sr->len -= ret;
384 sr->buf += ret;
385 sr->done_io += ret;
386 req->flags |= REQ_F_PARTIAL_IO;
516e82f0 387 return io_setup_async_addr(req, &__address, issue_flags);
f9ead18c 388 }
95eafc74
PB		 389 if (ret == -ERESTARTSYS)
390 ret = -EINTR;
f9ead18c
JA		 391 req_set_fail(req);
392 }
393 if (ret >= 0)
394 ret += sr->done_io;
395 else if (sr->done_io)
396 ret = sr->done_io;
397 io_req_set_res(req, ret, 0);
398 return IOU_OK;
399}
400
9bb66906
DY		 401static bool io_recvmsg_multishot_overflow(struct io_async_msghdr *iomsg)
402{
9b0fc3c0 403 int hdr;
9bb66906 404
9b0fc3c0 405 if (iomsg->namelen < 0)
9bb66906 406 return true;
9b0fc3c0
DY		 407 if (check_add_overflow((int)sizeof(struct io_uring_recvmsg_out),
408 iomsg->namelen, &hdr))
9bb66906 409 return true;
9b0fc3c0 410 if (check_add_overflow(hdr, (int)iomsg->controllen, &hdr))
9bb66906
DY		 411 return true;
412
413 return false;
414}
415
f9ead18c
JA		 416static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
417 struct io_async_msghdr *iomsg)
418{
f2ccb5ae 419 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
7fa875b8 420 struct user_msghdr msg;
f9ead18c
JA		 421 int ret;
422
7fa875b8
DY		 423 if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
424 return -EFAULT;
425
426 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 427 if (ret)
428 return ret;
429
430 if (req->flags & REQ_F_BUFFER_SELECT) {
7fa875b8 431 if (msg.msg_iovlen == 0) {
5702196e
DY		 432 sr->len = iomsg->fast_iov[0].iov_len = 0;
433 iomsg->fast_iov[0].iov_base = NULL;
434 iomsg->free_iov = NULL;
7fa875b8 435 } else if (msg.msg_iovlen > 1) {
f9ead18c 436 return -EINVAL;
5702196e 437 } else {
7fa875b8 438 if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
5702196e
DY		 439 return -EFAULT;
440 sr->len = iomsg->fast_iov[0].iov_len;
441 iomsg->free_iov = NULL;
442 }
9bb66906
DY		 443
444 if (req->flags & REQ_F_APOLL_MULTISHOT) {
445 iomsg->namelen = msg.msg_namelen;
446 iomsg->controllen = msg.msg_controllen;
447 if (io_recvmsg_multishot_overflow(iomsg))
448 return -EOVERFLOW;
449 }
f9ead18c
JA		 450 } else {
451 iomsg->free_iov = iomsg->fast_iov;
7fa875b8 452 ret = __import_iovec(READ, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
f9ead18c
JA		 453 &iomsg->free_iov, &iomsg->msg.msg_iter,
454 false);
455 if (ret > 0)
456 ret = 0;
457 }
458
459 return ret;
460}
461
462#ifdef CONFIG_COMPAT
463static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
464 struct io_async_msghdr *iomsg)
465{
f2ccb5ae 466 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
72c531f8 467 struct compat_msghdr msg;
f9ead18c 468 struct compat_iovec __user *uiov;
f9ead18c
JA		 469 int ret;
470
72c531f8
DY		 471 if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
472 return -EFAULT;
473
4f6a94d3 474 ret = __get_compat_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 475 if (ret)
476 return ret;
477
72c531f8 478 uiov = compat_ptr(msg.msg_iov);
f9ead18c
JA		 479 if (req->flags & REQ_F_BUFFER_SELECT) {
480 compat_ssize_t clen;
481
72c531f8 482 if (msg.msg_iovlen == 0) {
6d2f75a0 483 sr->len = 0;
72c531f8 484 } else if (msg.msg_iovlen > 1) {
f9ead18c 485 return -EINVAL;
6d2f75a0
DY		 486 } else {
487 if (!access_ok(uiov, sizeof(*uiov)))
488 return -EFAULT;
489 if (__get_user(clen, &uiov->iov_len))
490 return -EFAULT;
491 if (clen < 0)
492 return -EINVAL;
493 sr->len = clen;
6d2f75a0 494 }
9bb66906
DY		 495
496 if (req->flags & REQ_F_APOLL_MULTISHOT) {
497 iomsg->namelen = msg.msg_namelen;
498 iomsg->controllen = msg.msg_controllen;
499 if (io_recvmsg_multishot_overflow(iomsg))
500 return -EOVERFLOW;
501 }
f9ead18c
JA		 502 } else {
503 iomsg->free_iov = iomsg->fast_iov;
72c531f8 504 ret = __import_iovec(READ, (struct iovec __user *)uiov, msg.msg_iovlen,
f9ead18c
JA		 505 UIO_FASTIOV, &iomsg->free_iov,
506 &iomsg->msg.msg_iter, true);
507 if (ret < 0)
508 return ret;
509 }
510
511 return 0;
512}
513#endif
514
515static int io_recvmsg_copy_hdr(struct io_kiocb *req,
516 struct io_async_msghdr *iomsg)
517{
518 iomsg->msg.msg_name = &iomsg->addr;
519
520#ifdef CONFIG_COMPAT
521 if (req->ctx->compat)
522 return __io_compat_recvmsg_copy_hdr(req, iomsg);
523#endif
524
525 return __io_recvmsg_copy_hdr(req, iomsg);
526}
527
528int io_recvmsg_prep_async(struct io_kiocb *req)
529{
530 int ret;
531
858c293e
PB		 532 if (!io_msg_alloc_async_prep(req))
533 return -ENOMEM;
f9ead18c
JA		 534 ret = io_recvmsg_copy_hdr(req, req->async_data);
535 if (!ret)
536 req->flags |= REQ_F_NEED_CLEANUP;
537 return ret;
538}
539
b3fdea6e
DY		 540#define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
541
f9ead18c
JA		 542int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
543{
f2ccb5ae 544 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 545
546 if (unlikely(sqe->file_index || sqe->addr2))
547 return -EINVAL;
548
549 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
550 sr->len = READ_ONCE(sqe->len);
551 sr->flags = READ_ONCE(sqe->ioprio);
b3fdea6e 552 if (sr->flags & ~(RECVMSG_FLAGS))
f9ead18c
JA		 553 return -EINVAL;
554 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
555 if (sr->msg_flags & MSG_DONTWAIT)
556 req->flags |= REQ_F_NOWAIT;
557 if (sr->msg_flags & MSG_ERRQUEUE)
558 req->flags |= REQ_F_CLEAR_POLLIN;
b3fdea6e
DY		 559 if (sr->flags & IORING_RECV_MULTISHOT) {
560 if (!(req->flags & REQ_F_BUFFER_SELECT))
561 return -EINVAL;
562 if (sr->msg_flags & MSG_WAITALL)
563 return -EINVAL;
564 if (req->opcode == IORING_OP_RECV && sr->len)
565 return -EINVAL;
566 req->flags |= REQ_F_APOLL_MULTISHOT;
567 }
f9ead18c
JA		 568
569#ifdef CONFIG_COMPAT
570 if (req->ctx->compat)
571 sr->msg_flags |= MSG_CMSG_COMPAT;
572#endif
573 sr->done_io = 0;
574 return 0;
575}
576
b3fdea6e
DY		 577static inline void io_recv_prep_retry(struct io_kiocb *req)
578{
f2ccb5ae 579 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
b3fdea6e
DY		 580
581 sr->done_io = 0;
582 sr->len = 0; /* get from the provided buffer */
583}
584
585/*
9bb66906 586 * Finishes io_recv and io_recvmsg.
b3fdea6e
DY		 587 *
588 * Returns true if it is actually finished, or false if it should run
589 * again (for multishot).
590 */
9bb66906
DY		 591static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
592 unsigned int cflags, bool mshot_finished)
b3fdea6e
DY		 593{
594 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
595 io_req_set_res(req, *ret, cflags);
596 *ret = IOU_OK;
597 return true;
598 }
599
9bb66906 600 if (!mshot_finished) {
b3fdea6e
DY		 601 if (io_post_aux_cqe(req->ctx, req->cqe.user_data, *ret,
602 cflags | IORING_CQE_F_MORE, false)) {
603 io_recv_prep_retry(req);
604 return false;
605 }
606 /*
607 * Otherwise stop multishot but use the current result.
608 * Probably will end up going into overflow, but this means
609 * we cannot trust the ordering anymore
610 */
611 }
612
613 io_req_set_res(req, *ret, cflags);
614
615 if (req->flags & REQ_F_POLLED)
616 *ret = IOU_STOP_MULTISHOT;
e2df2ccb
DY		 617 else
618 *ret = IOU_OK;
b3fdea6e
DY		 619 return true;
620}
621
9bb66906
DY		 622static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
623 struct io_sr_msg *sr, void __user **buf,
624 size_t *len)
625{
626 unsigned long ubuf = (unsigned long) *buf;
627 unsigned long hdr;
628
629 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
630 kmsg->controllen;
631 if (*len < hdr)
632 return -EFAULT;
633
634 if (kmsg->controllen) {
635 unsigned long control = ubuf + hdr - kmsg->controllen;
636
d1f6222c 637 kmsg->msg.msg_control_user = (void __user *) control;
9bb66906
DY		 638 kmsg->msg.msg_controllen = kmsg->controllen;
639 }
640
641 sr->buf = *buf; /* stash for later copy */
d1f6222c 642 *buf = (void __user *) (ubuf + hdr);
9bb66906
DY		 643 kmsg->payloadlen = *len = *len - hdr;
644 return 0;
645}
646
647struct io_recvmsg_multishot_hdr {
648 struct io_uring_recvmsg_out msg;
649 struct sockaddr_storage addr;
650};
651
652static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
653 struct io_async_msghdr *kmsg,
654 unsigned int flags, bool *finished)
655{
656 int err;
657 int copy_len;
658 struct io_recvmsg_multishot_hdr hdr;
659
660 if (kmsg->namelen)
661 kmsg->msg.msg_name = &hdr.addr;
662 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
663 kmsg->msg.msg_namelen = 0;
664
665 if (sock->file->f_flags & O_NONBLOCK)
666 flags |= MSG_DONTWAIT;
667
668 err = sock_recvmsg(sock, &kmsg->msg, flags);
669 *finished = err <= 0;
670 if (err < 0)
671 return err;
672
673 hdr.msg = (struct io_uring_recvmsg_out) {
674 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
675 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
676 };
677
678 hdr.msg.payloadlen = err;
679 if (err > kmsg->payloadlen)
680 err = kmsg->payloadlen;
681
682 copy_len = sizeof(struct io_uring_recvmsg_out);
683 if (kmsg->msg.msg_namelen > kmsg->namelen)
684 copy_len += kmsg->namelen;
685 else
686 copy_len += kmsg->msg.msg_namelen;
687
688 /*
689 * "fromlen shall refer to the value before truncation.."
690 * 1003.1g
691 */
692 hdr.msg.namelen = kmsg->msg.msg_namelen;
693
694 /* ensure that there is no gap between hdr and sockaddr_storage */
695 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
696 sizeof(struct io_uring_recvmsg_out));
697 if (copy_to_user(io->buf, &hdr, copy_len)) {
698 *finished = true;
699 return -EFAULT;
700 }
701
702 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
703 kmsg->controllen + err;
704}
705
f9ead18c
JA		 706int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
707{
f2ccb5ae 708 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 709 struct io_async_msghdr iomsg, *kmsg;
710 struct socket *sock;
711 unsigned int cflags;
712 unsigned flags;
713 int ret, min_ret = 0;
714 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
9bb66906 715 bool mshot_finished = true;
f9ead18c
JA		 716
717 sock = sock_from_file(req->file);
718 if (unlikely(!sock))
719 return -ENOTSOCK;
720
721 if (req_has_async_data(req)) {
722 kmsg = req->async_data;
723 } else {
724 ret = io_recvmsg_copy_hdr(req, &iomsg);
725 if (ret)
726 return ret;
727 kmsg = &iomsg;
728 }
729
730 if (!(req->flags & REQ_F_POLLED) &&
731 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 732 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 733
9bb66906 734retry_multishot:
f9ead18c
JA		 735 if (io_do_buffer_select(req)) {
736 void __user *buf;
9bb66906 737 size_t len = sr->len;
f9ead18c 738
9bb66906 739 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 740 if (!buf)
741 return -ENOBUFS;
9bb66906
DY		 742
743 if (req->flags & REQ_F_APOLL_MULTISHOT) {
744 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
745 if (ret) {
746 io_kbuf_recycle(req, issue_flags);
747 return ret;
748 }
749 }
750
f9ead18c 751 kmsg->fast_iov[0].iov_base = buf;
9bb66906 752 kmsg->fast_iov[0].iov_len = len;
f9ead18c 753 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, 1,
9bb66906 754 len);
f9ead18c
JA		 755 }
756
757 flags = sr->msg_flags;
758 if (force_nonblock)
759 flags |= MSG_DONTWAIT;
760 if (flags & MSG_WAITALL)
761 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
762
763 kmsg->msg.msg_get_inq = 1;
9bb66906
DY		 764 if (req->flags & REQ_F_APOLL_MULTISHOT)
765 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
766 &mshot_finished);
767 else
768 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
769 kmsg->uaddr, flags);
770
f9ead18c 771 if (ret < min_ret) {
9bb66906
DY		 772 if (ret == -EAGAIN && force_nonblock) {
773 ret = io_setup_async_msg(req, kmsg, issue_flags);
774 if (ret == -EAGAIN && (req->flags & IO_APOLL_MULTI_POLLED) ==
775 IO_APOLL_MULTI_POLLED) {
776 io_kbuf_recycle(req, issue_flags);
777 return IOU_ISSUE_SKIP_COMPLETE;
778 }
779 return ret;
780 }
f9ead18c
JA		 781 if (ret > 0 && io_net_retry(sock, flags)) {
782 sr->done_io += ret;
783 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 784 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 785 }
95eafc74
PB		 786 if (ret == -ERESTARTSYS)
787 ret = -EINTR;
f9ead18c
JA		 788 req_set_fail(req);
789 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
790 req_set_fail(req);
791 }
792
d4e097da 793 if (ret > 0)
f9ead18c
JA		 794 ret += sr->done_io;
795 else if (sr->done_io)
796 ret = sr->done_io;
d4e097da
DY		 797 else
798 io_kbuf_recycle(req, issue_flags);
799
f9ead18c
JA		 800 cflags = io_put_kbuf(req, issue_flags);
801 if (kmsg->msg.msg_inq)
802 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 803
9bb66906
DY		 804 if (!io_recv_finish(req, &ret, cflags, mshot_finished))
805 goto retry_multishot;
806
807 if (mshot_finished) {
808 io_netmsg_recycle(req, issue_flags);
809 /* fast path, check for non-NULL to avoid function call */
810 if (kmsg->free_iov)
811 kfree(kmsg->free_iov);
812 req->flags &= ~REQ_F_NEED_CLEANUP;
813 }
814
815 return ret;
f9ead18c
JA		 816}
817
818int io_recv(struct io_kiocb *req, unsigned int issue_flags)
819{
f2ccb5ae 820 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 821 struct msghdr msg;
822 struct socket *sock;
823 struct iovec iov;
824 unsigned int cflags;
825 unsigned flags;
826 int ret, min_ret = 0;
827 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
b3fdea6e 828 size_t len = sr->len;
f9ead18c
JA		 829
830 if (!(req->flags & REQ_F_POLLED) &&
831 (sr->flags & IORING_RECVSEND_POLL_FIRST))
832 return -EAGAIN;
833
834 sock = sock_from_file(req->file);
835 if (unlikely(!sock))
836 return -ENOTSOCK;
837
b3fdea6e 838retry_multishot:
f9ead18c
JA		 839 if (io_do_buffer_select(req)) {
840 void __user *buf;
841
b3fdea6e 842 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 843 if (!buf)
844 return -ENOBUFS;
845 sr->buf = buf;
846 }
847
b3fdea6e 848 ret = import_single_range(READ, sr->buf, len, &iov, &msg.msg_iter);
f9ead18c
JA		 849 if (unlikely(ret))
850 goto out_free;
851
852 msg.msg_name = NULL;
853 msg.msg_namelen = 0;
854 msg.msg_control = NULL;
855 msg.msg_get_inq = 1;
856 msg.msg_flags = 0;
857 msg.msg_controllen = 0;
858 msg.msg_iocb = NULL;
e02b6651 859 msg.msg_ubuf = NULL;
f9ead18c
JA		 860
861 flags = sr->msg_flags;
862 if (force_nonblock)
863 flags |= MSG_DONTWAIT;
864 if (flags & MSG_WAITALL)
865 min_ret = iov_iter_count(&msg.msg_iter);
866
867 ret = sock_recvmsg(sock, &msg, flags);
868 if (ret < min_ret) {
b3fdea6e
DY		 869 if (ret == -EAGAIN && force_nonblock) {
870 if ((req->flags & IO_APOLL_MULTI_POLLED) == IO_APOLL_MULTI_POLLED) {
871 io_kbuf_recycle(req, issue_flags);
872 return IOU_ISSUE_SKIP_COMPLETE;
873 }
874
f9ead18c 875 return -EAGAIN;
b3fdea6e 876 }
f9ead18c
JA		 877 if (ret > 0 && io_net_retry(sock, flags)) {
878 sr->len -= ret;
879 sr->buf += ret;
880 sr->done_io += ret;
881 req->flags |= REQ_F_PARTIAL_IO;
882 return -EAGAIN;
883 }
95eafc74
PB		 884 if (ret == -ERESTARTSYS)
885 ret = -EINTR;
f9ead18c
JA		 886 req_set_fail(req);
887 } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
888out_free:
889 req_set_fail(req);
890 }
891
d4e097da 892 if (ret > 0)
f9ead18c
JA		 893 ret += sr->done_io;
894 else if (sr->done_io)
895 ret = sr->done_io;
d4e097da
DY		 896 else
897 io_kbuf_recycle(req, issue_flags);
898
f9ead18c
JA		 899 cflags = io_put_kbuf(req, issue_flags);
900 if (msg.msg_inq)
901 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 902
9bb66906 903 if (!io_recv_finish(req, &ret, cflags, ret <= 0))
b3fdea6e
DY		 904 goto retry_multishot;
905
906 return ret;
f9ead18c
JA		 907}
908
b0e9b551 909void io_send_zc_cleanup(struct io_kiocb *req)
b48c312b 910{
ac9e5784 911 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
493108d9 912 struct io_async_msghdr *io;
b48c312b 913
493108d9
PB		 914 if (req_has_async_data(req)) {
915 io = req->async_data;
4c17a496
PB		 916 /* might be ->fast_iov if *msg_copy_hdr failed */
917 if (io->free_iov != io->fast_iov)
918 kfree(io->free_iov);
493108d9 919 }
a75155fa 920 if (zc->notif) {
a75155fa
PB		 921 io_notif_flush(zc->notif);
922 zc->notif = NULL;
923 }
b48c312b
PB		 924}
925
b0e9b551 926int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
06a5464b 927{
ac9e5784 928 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
10c7d33e 929 struct io_ring_ctx *ctx = req->ctx;
b48c312b 930 struct io_kiocb *notif;
06a5464b 931
493108d9 932 if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
b48c312b
PB		 933 return -EINVAL;
934 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
935 if (req->flags & REQ_F_CQE_SKIP)
06a5464b
PB		 936 return -EINVAL;
937
938 zc->flags = READ_ONCE(sqe->ioprio);
63809137 939 if (zc->flags & ~(IORING_RECVSEND_POLL_FIRST |
57f33224 940 IORING_RECVSEND_FIXED_BUF))
06a5464b 941 return -EINVAL;
e3366e02
PB		 942 notif = zc->notif = io_alloc_notif(ctx);
943 if (!notif)
944 return -ENOMEM;
945 notif->cqe.user_data = req->cqe.user_data;
946 notif->cqe.res = 0;
947 notif->cqe.flags = IORING_CQE_F_NOTIF;
948 req->flags |= REQ_F_NEED_CLEANUP;
10c7d33e
PB		 949 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
950 unsigned idx = READ_ONCE(sqe->buf_index);
951
952 if (unlikely(idx >= ctx->nr_user_bufs))
953 return -EFAULT;
954 idx = array_index_nospec(idx, ctx->nr_user_bufs);
955 req->imu = READ_ONCE(ctx->user_bufs[idx]);
e3366e02 956 io_req_set_rsrc_node(notif, ctx, 0);
10c7d33e 957 }
06a5464b 958
493108d9
PB		 959 if (req->opcode == IORING_OP_SEND_ZC) {
960 if (READ_ONCE(sqe->__pad3[0]))
961 return -EINVAL;
962 zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
963 zc->addr_len = READ_ONCE(sqe->addr_len);
964 } else {
965 if (unlikely(sqe->addr2 || sqe->file_index))
966 return -EINVAL;
967 if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
968 return -EINVAL;
969 }
970
06a5464b
PB		 971 zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
972 zc->len = READ_ONCE(sqe->len);
973 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
06a5464b
PB		 974 if (zc->msg_flags & MSG_DONTWAIT)
975 req->flags |= REQ_F_NOWAIT;
092aeedb 976
4a933e62 977 zc->done_io = 0;
092aeedb 978
06a5464b
PB		 979#ifdef CONFIG_COMPAT
980 if (req->ctx->compat)
981 zc->msg_flags |= MSG_CMSG_COMPAT;
982#endif
983 return 0;
984}
985
cd9021e8
PB		 986static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
987 struct iov_iter *from, size_t length)
988{
989 skb_zcopy_downgrade_managed(skb);
990 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
991}
992
3ff1a0d3
PB		 993static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
994 struct iov_iter *from, size_t length)
995{
996 struct skb_shared_info *shinfo = skb_shinfo(skb);
997 int frag = shinfo->nr_frags;
998 int ret = 0;
999 struct bvec_iter bi;
1000 ssize_t copied = 0;
1001 unsigned long truesize = 0;
1002
cd9021e8 1003 if (!frag)
3ff1a0d3 1004 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
cd9021e8 1005 else if (unlikely(!skb_zcopy_managed(skb)))
3ff1a0d3 1006 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
3ff1a0d3
PB		 1007
1008 bi.bi_size = min(from->count, length);
1009 bi.bi_bvec_done = from->iov_offset;
1010 bi.bi_idx = 0;
1011
1012 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1013 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1014
1015 copied += v.bv_len;
1016 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1017 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1018 v.bv_offset, v.bv_len);
1019 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1020 }
1021 if (bi.bi_size)
1022 ret = -EMSGSIZE;
1023
1024 shinfo->nr_frags = frag;
1025 from->bvec += bi.bi_idx;
1026 from->nr_segs -= bi.bi_idx;
dfb58b17 1027 from->count -= copied;
3ff1a0d3
PB		 1028 from->iov_offset = bi.bi_bvec_done;
1029
1030 skb->data_len += copied;
1031 skb->len += copied;
1032 skb->truesize += truesize;
1033
1034 if (sk && sk->sk_type == SOCK_STREAM) {
1035 sk_wmem_queued_add(sk, truesize);
1036 if (!skb_zcopy_pure(skb))
1037 sk_mem_charge(sk, truesize);
1038 } else {
1039 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1040 }
1041 return ret;
1042}
1043
b0e9b551 1044int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
06a5464b 1045{
6ae61b7a 1046 struct sockaddr_storage __address;
ac9e5784 1047 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
06a5464b
PB		 1048 struct msghdr msg;
1049 struct iovec iov;
1050 struct socket *sock;
6ae91ac9 1051 unsigned msg_flags;
06a5464b
PB		 1052 int ret, min_ret = 0;
1053
06a5464b
PB		 1054 sock = sock_from_file(req->file);
1055 if (unlikely(!sock))
1056 return -ENOTSOCK;
1057
06a5464b
PB		 1058 msg.msg_name = NULL;
1059 msg.msg_control = NULL;
1060 msg.msg_controllen = 0;
1061 msg.msg_namelen = 0;
1062
86dc8f23 1063 if (zc->addr) {
581711c4
PB		 1064 if (req_has_async_data(req)) {
1065 struct io_async_msghdr *io = req->async_data;
1066
6ae61b7a 1067 msg.msg_name = &io->addr;
581711c4
PB		 1068 } else {
1069 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1070 if (unlikely(ret < 0))
1071 return ret;
1072 msg.msg_name = (struct sockaddr *)&__address;
581711c4 1073 }
86dc8f23
PB		 1074 msg.msg_namelen = zc->addr_len;
1075 }
1076
3c840053
PB		 1077 if (!(req->flags & REQ_F_POLLED) &&
1078 (zc->flags & IORING_RECVSEND_POLL_FIRST))
6ae61b7a 1079 return io_setup_async_addr(req, &__address, issue_flags);
3c840053 1080
10c7d33e
PB		 1081 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1082 ret = io_import_fixed(WRITE, &msg.msg_iter, req->imu,
1083 (u64)(uintptr_t)zc->buf, zc->len);
1084 if (unlikely(ret))
986e263d 1085 return ret;
cd9021e8 1086 msg.sg_from_iter = io_sg_from_iter;
10c7d33e
PB		 1087 } else {
1088 ret = import_single_range(WRITE, zc->buf, zc->len, &iov,
1089 &msg.msg_iter);
1090 if (unlikely(ret))
1091 return ret;
b48c312b 1092 ret = io_notif_account_mem(zc->notif, zc->len);
2e32ba56
PB		 1093 if (unlikely(ret))
1094 return ret;
cd9021e8 1095 msg.sg_from_iter = io_sg_from_iter_iovec;
10c7d33e 1096 }
06a5464b
PB		 1097
1098 msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1099 if (issue_flags & IO_URING_F_NONBLOCK)
1100 msg_flags |= MSG_DONTWAIT;
1101 if (msg_flags & MSG_WAITALL)
1102 min_ret = iov_iter_count(&msg.msg_iter);
1103
1104 msg.msg_flags = msg_flags;
b48c312b 1105 msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
06a5464b
PB		 1106 ret = sock_sendmsg(sock, &msg);
1107
1108 if (unlikely(ret < min_ret)) {
1109 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
6ae61b7a 1110 return io_setup_async_addr(req, &__address, issue_flags);
581711c4 1111
4a933e62
PB		 1112 if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1113 zc->len -= ret;
1114 zc->buf += ret;
1115 zc->done_io += ret;
1116 req->flags |= REQ_F_PARTIAL_IO;
6ae61b7a 1117 return io_setup_async_addr(req, &__address, issue_flags);
4a933e62
PB		 1118 }
1119 if (ret == -ERESTARTSYS)
1120 ret = -EINTR;
5a848b7c 1121 req_set_fail(req);
06a5464b
PB		 1122 }
1123
4a933e62
PB		 1124 if (ret >= 0)
1125 ret += zc->done_io;
1126 else if (zc->done_io)
1127 ret = zc->done_io;
b48c312b
PB		 1128
1129 io_notif_flush(zc->notif);
1130 req->flags &= ~REQ_F_NEED_CLEANUP;
6ae91ac9 1131 io_req_set_res(req, ret, IORING_CQE_F_MORE);
06a5464b
PB		 1132 return IOU_OK;
1133}
1134
493108d9
PB		 1135int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1136{
1137 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1138 struct io_async_msghdr iomsg, *kmsg;
1139 struct socket *sock;
6ae91ac9 1140 unsigned flags;
493108d9
PB		 1141 int ret, min_ret = 0;
1142
1143 sock = sock_from_file(req->file);
1144 if (unlikely(!sock))
1145 return -ENOTSOCK;
1146
1147 if (req_has_async_data(req)) {
1148 kmsg = req->async_data;
1149 } else {
1150 ret = io_sendmsg_copy_hdr(req, &iomsg);
1151 if (ret)
1152 return ret;
1153 kmsg = &iomsg;
1154 }
1155
1156 if (!(req->flags & REQ_F_POLLED) &&
1157 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1158 return io_setup_async_msg(req, kmsg, issue_flags);
1159
1160 flags = sr->msg_flags | MSG_ZEROCOPY;
1161 if (issue_flags & IO_URING_F_NONBLOCK)
1162 flags |= MSG_DONTWAIT;
1163 if (flags & MSG_WAITALL)
1164 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1165
1166 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1167 kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1168 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1169
1170 if (unlikely(ret < min_ret)) {
1171 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1172 return io_setup_async_msg(req, kmsg, issue_flags);
1173
1174 if (ret > 0 && io_net_retry(sock, flags)) {
1175 sr->done_io += ret;
1176 req->flags |= REQ_F_PARTIAL_IO;
1177 return io_setup_async_msg(req, kmsg, issue_flags);
1178 }
493108d9
PB		 1179 if (ret == -ERESTARTSYS)
1180 ret = -EINTR;
1181 req_set_fail(req);
1182 }
1183 /* fast path, check for non-NULL to avoid function call */
1184 if (kmsg->free_iov)
1185 kfree(kmsg->free_iov);
1186
1187 io_netmsg_recycle(req, issue_flags);
1188 if (ret >= 0)
1189 ret += sr->done_io;
1190 else if (sr->done_io)
1191 ret = sr->done_io;
1192
1193 io_notif_flush(sr->notif);
1194 req->flags &= ~REQ_F_NEED_CLEANUP;
6ae91ac9 1195 io_req_set_res(req, ret, IORING_CQE_F_MORE);
493108d9
PB		 1196 return IOU_OK;
1197}
1198
7e6b638e
PB		 1199void io_sendrecv_fail(struct io_kiocb *req)
1200{
1201 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
7e6b638e
PB		 1202
1203 if (req->flags & REQ_F_PARTIAL_IO)
6ae91ac9
PB		 1204 req->cqe.res = sr->done_io;
1205
c4c0009e 1206 if ((req->flags & REQ_F_NEED_CLEANUP) &&
6ae91ac9
PB		 1207 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1208 req->cqe.flags |= IORING_CQE_F_MORE;
5693bcce
PB		 1209}
1210
f9ead18c
JA		 1211int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1212{
f2ccb5ae 1213 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1214 unsigned flags;
1215
1216 if (sqe->len || sqe->buf_index)
1217 return -EINVAL;
1218
1219 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1220 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1221 accept->flags = READ_ONCE(sqe->accept_flags);
1222 accept->nofile = rlimit(RLIMIT_NOFILE);
1223 flags = READ_ONCE(sqe->ioprio);
1224 if (flags & ~IORING_ACCEPT_MULTISHOT)
1225 return -EINVAL;
1226
1227 accept->file_slot = READ_ONCE(sqe->file_index);
1228 if (accept->file_slot) {
1229 if (accept->flags & SOCK_CLOEXEC)
1230 return -EINVAL;
1231 if (flags & IORING_ACCEPT_MULTISHOT &&
1232 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1233 return -EINVAL;
1234 }
1235 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1236 return -EINVAL;
1237 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1238 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1239 if (flags & IORING_ACCEPT_MULTISHOT)
1240 req->flags |= REQ_F_APOLL_MULTISHOT;
1241 return 0;
1242}
1243
1244int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1245{
1246 struct io_ring_ctx *ctx = req->ctx;
f2ccb5ae 1247 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1248 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1249 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1250 bool fixed = !!accept->file_slot;
1251 struct file *file;
1252 int ret, fd;
1253
1254retry:
1255 if (!fixed) {
1256 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1257 if (unlikely(fd < 0))
1258 return fd;
1259 }
1260 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1261 accept->flags);
1262 if (IS_ERR(file)) {
1263 if (!fixed)
1264 put_unused_fd(fd);
1265 ret = PTR_ERR(file);
1266 if (ret == -EAGAIN && force_nonblock) {
1267 /*
1268 * if it's multishot and polled, we don't need to
1269 * return EAGAIN to arm the poll infra since it
1270 * has already been done
1271 */
1272 if ((req->flags & IO_APOLL_MULTI_POLLED) ==
1273 IO_APOLL_MULTI_POLLED)
1274 ret = IOU_ISSUE_SKIP_COMPLETE;
1275 return ret;
1276 }
1277 if (ret == -ERESTARTSYS)
1278 ret = -EINTR;
1279 req_set_fail(req);
1280 } else if (!fixed) {
1281 fd_install(fd, file);
1282 ret = fd;
1283 } else {
1284 ret = io_fixed_fd_install(req, issue_flags, file,
1285 accept->file_slot);
1286 }
1287
1288 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1289 io_req_set_res(req, ret, 0);
1290 return IOU_OK;
1291 }
f9ead18c 1292
cbd25748
DY		 1293 if (ret >= 0 &&
1294 io_post_aux_cqe(ctx, req->cqe.user_data, ret, IORING_CQE_F_MORE, false))
d245bca6 1295 goto retry;
cbd25748
DY		 1296
1297 io_req_set_res(req, ret, 0);
1298 if (req->flags & REQ_F_POLLED)
1299 return IOU_STOP_MULTISHOT;
1300 return IOU_OK;
f9ead18c
JA		 1301}
1302
1303int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1304{
f2ccb5ae 1305 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1306
1307 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1308 return -EINVAL;
1309
1310 sock->domain = READ_ONCE(sqe->fd);
1311 sock->type = READ_ONCE(sqe->off);
1312 sock->protocol = READ_ONCE(sqe->len);
1313 sock->file_slot = READ_ONCE(sqe->file_index);
1314 sock->nofile = rlimit(RLIMIT_NOFILE);
1315
1316 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1317 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1318 return -EINVAL;
1319 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1320 return -EINVAL;
1321 return 0;
1322}
1323
1324int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1325{
f2ccb5ae 1326 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1327 bool fixed = !!sock->file_slot;
1328 struct file *file;
1329 int ret, fd;
1330
1331 if (!fixed) {
1332 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1333 if (unlikely(fd < 0))
1334 return fd;
1335 }
1336 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1337 if (IS_ERR(file)) {
1338 if (!fixed)
1339 put_unused_fd(fd);
1340 ret = PTR_ERR(file);
1341 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1342 return -EAGAIN;
1343 if (ret == -ERESTARTSYS)
1344 ret = -EINTR;
1345 req_set_fail(req);
1346 } else if (!fixed) {
1347 fd_install(fd, file);
1348 ret = fd;
1349 } else {
1350 ret = io_fixed_fd_install(req, issue_flags, file,
1351 sock->file_slot);
1352 }
1353 io_req_set_res(req, ret, 0);
1354 return IOU_OK;
1355}
1356
1357int io_connect_prep_async(struct io_kiocb *req)
1358{
1359 struct io_async_connect *io = req->async_data;
f2ccb5ae 1360 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1361
1362 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1363}
1364
1365int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1366{
f2ccb5ae 1367 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1368
1369 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1370 return -EINVAL;
1371
1372 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1373 conn->addr_len = READ_ONCE(sqe->addr2);
1374 return 0;
1375}
1376
1377int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1378{
f2ccb5ae 1379 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1380 struct io_async_connect __io, *io;
1381 unsigned file_flags;
1382 int ret;
1383 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1384
1385 if (req_has_async_data(req)) {
1386 io = req->async_data;
1387 } else {
1388 ret = move_addr_to_kernel(connect->addr,
1389 connect->addr_len,
1390 &__io.address);
1391 if (ret)
1392 goto out;
1393 io = &__io;
1394 }
1395
1396 file_flags = force_nonblock ? O_NONBLOCK : 0;
1397
1398 ret = __sys_connect_file(req->file, &io->address,
1399 connect->addr_len, file_flags);
1400 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
1401 if (req_has_async_data(req))
1402 return -EAGAIN;
1403 if (io_alloc_async_data(req)) {
1404 ret = -ENOMEM;
1405 goto out;
1406 }
1407 memcpy(req->async_data, &__io, sizeof(__io));
1408 return -EAGAIN;
1409 }
1410 if (ret == -ERESTARTSYS)
1411 ret = -EINTR;
1412out:
1413 if (ret < 0)
1414 req_set_fail(req);
1415 io_req_set_res(req, ret, 0);
1416 return IOU_OK;
1417}
43e0bbbd
JA		 1418
1419void io_netmsg_cache_free(struct io_cache_entry *entry)
1420{
1421 kfree(container_of(entry, struct io_async_msghdr, cache));
1422}
f9ead18c 1423#endif
Linux 6.x block layer and io_uring tree(s)