projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
io_uring: include task_work run after scheduling in wait for events
[linux-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;
3fb1bd68 49 bool in_progress;
f9ead18c
JA		 50};
51
52struct io_sr_msg {
53 struct file *file;
54 union {
55 struct compat_msghdr __user *umsg_compat;
56 struct user_msghdr __user *umsg;
57 void __user *buf;
58 };
0b048557
PB		 59 unsigned len;
60 unsigned done_io;
293402e5 61 unsigned msg_flags;
0b048557 62 u16 flags;
516e82f0 63 /* initialised and used only by !msg send variants */
0b048557 64 u16 addr_len;
092aeedb 65 void __user *addr;
516e82f0 66 /* used only for send zerocopy */
b48c312b 67 struct io_kiocb *notif;
06a5464b
PB		 68};
69
17add5ce
PB		 70static inline bool io_check_multishot(struct io_kiocb *req,
71 unsigned int issue_flags)
72{
73 /*
74 * When ->locked_cq is set we only allow to post CQEs from the original
75 * task context. Usual request completions will be handled in other
76 * generic paths but multipoll may decide to post extra cqes.
77 */
78 return !(issue_flags & IO_URING_F_IOWQ) ||
79 !(issue_flags & IO_URING_F_MULTISHOT) ||
80 !req->ctx->task_complete;
81}
82
f9ead18c
JA		 83int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
84{
f2ccb5ae 85 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
f9ead18c
JA		 86
87 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
88 sqe->buf_index || sqe->splice_fd_in))
89 return -EINVAL;
90
91 shutdown->how = READ_ONCE(sqe->len);
92 return 0;
93}
94
95int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
96{
f2ccb5ae 97 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
f9ead18c
JA		 98 struct socket *sock;
99 int ret;
100
101 if (issue_flags & IO_URING_F_NONBLOCK)
102 return -EAGAIN;
103
104 sock = sock_from_file(req->file);
105 if (unlikely(!sock))
106 return -ENOTSOCK;
107
108 ret = __sys_shutdown_sock(sock, shutdown->how);
109 io_req_set_res(req, ret, 0);
110 return IOU_OK;
111}
112
113static bool io_net_retry(struct socket *sock, int flags)
114{
115 if (!(flags & MSG_WAITALL))
116 return false;
117 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
118}
119
43e0bbbd
JA		 120static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
121{
122 struct io_async_msghdr *hdr = req->async_data;
123
06360426 124 if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
43e0bbbd
JA		 125 return;
126
127 /* Let normal cleanup path reap it if we fail adding to the cache */
128 if (io_alloc_cache_put(&req->ctx->netmsg_cache, &hdr->cache)) {
129 req->async_data = NULL;
130 req->flags &= ~REQ_F_ASYNC_DATA;
131 }
132}
133
858c293e
PB		 134static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
135 unsigned int issue_flags)
43e0bbbd
JA		 136{
137 struct io_ring_ctx *ctx = req->ctx;
138 struct io_cache_entry *entry;
4c17a496 139 struct io_async_msghdr *hdr;
43e0bbbd 140
df730ec2
XL		 141 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
142 entry = io_alloc_cache_get(&ctx->netmsg_cache);
143 if (entry) {
144 hdr = container_of(entry, struct io_async_msghdr, cache);
145 hdr->free_iov = NULL;
146 req->flags |= REQ_F_ASYNC_DATA;
147 req->async_data = hdr;
148 return hdr;
149 }
43e0bbbd
JA		 150 }
151
4c17a496
PB		 152 if (!io_alloc_async_data(req)) {
153 hdr = req->async_data;
154 hdr->free_iov = NULL;
155 return hdr;
156 }
43e0bbbd
JA		 157 return NULL;
158}
159
858c293e
PB		 160static inline struct io_async_msghdr *io_msg_alloc_async_prep(struct io_kiocb *req)
161{
162 /* ->prep_async is always called from the submission context */
163 return io_msg_alloc_async(req, 0);
164}
165
f9ead18c 166static int io_setup_async_msg(struct io_kiocb *req,
43e0bbbd
JA		 167 struct io_async_msghdr *kmsg,
168 unsigned int issue_flags)
f9ead18c 169{
3f743e9b 170 struct io_async_msghdr *async_msg;
f9ead18c 171
3f743e9b 172 if (req_has_async_data(req))
f9ead18c 173 return -EAGAIN;
858c293e 174 async_msg = io_msg_alloc_async(req, issue_flags);
43e0bbbd 175 if (!async_msg) {
f9ead18c
JA		 176 kfree(kmsg->free_iov);
177 return -ENOMEM;
178 }
f9ead18c
JA		 179 req->flags |= REQ_F_NEED_CLEANUP;
180 memcpy(async_msg, kmsg, sizeof(*kmsg));
6f10ae8a
PB		 181 if (async_msg->msg.msg_name)
182 async_msg->msg.msg_name = &async_msg->addr;
f9ead18c 183 /* if were using fast_iov, set it to the new one */
3e4cb6eb
SM		 184 if (!kmsg->free_iov) {
185 size_t fast_idx = kmsg->msg.msg_iter.iov - kmsg->fast_iov;
186 async_msg->msg.msg_iter.iov = &async_msg->fast_iov[fast_idx];
187 }
f9ead18c
JA		 188
189 return -EAGAIN;
190}
191
192static int io_sendmsg_copy_hdr(struct io_kiocb *req,
193 struct io_async_msghdr *iomsg)
194{
f2ccb5ae 195 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 196
197 iomsg->msg.msg_name = &iomsg->addr;
198 iomsg->free_iov = iomsg->fast_iov;
199 return sendmsg_copy_msghdr(&iomsg->msg, sr->umsg, sr->msg_flags,
200 &iomsg->free_iov);
201}
202
516e82f0 203int io_send_prep_async(struct io_kiocb *req)
581711c4 204{
ac9e5784 205 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
581711c4
PB		 206 struct io_async_msghdr *io;
207 int ret;
208
209 if (!zc->addr || req_has_async_data(req))
210 return 0;
6bf8ad25
PB		 211 io = io_msg_alloc_async_prep(req);
212 if (!io)
581711c4 213 return -ENOMEM;
581711c4
PB		 214 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
215 return ret;
216}
217
218static int io_setup_async_addr(struct io_kiocb *req,
6ae61b7a 219 struct sockaddr_storage *addr_storage,
581711c4
PB		 220 unsigned int issue_flags)
221{
6ae61b7a 222 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
581711c4
PB		 223 struct io_async_msghdr *io;
224
6ae61b7a 225 if (!sr->addr || req_has_async_data(req))
581711c4 226 return -EAGAIN;
6bf8ad25
PB		 227 io = io_msg_alloc_async(req, issue_flags);
228 if (!io)
581711c4 229 return -ENOMEM;
6ae61b7a 230 memcpy(&io->addr, addr_storage, sizeof(io->addr));
581711c4
PB		 231 return -EAGAIN;
232}
233
f9ead18c
JA		 234int io_sendmsg_prep_async(struct io_kiocb *req)
235{
236 int ret;
237
858c293e
PB		 238 if (!io_msg_alloc_async_prep(req))
239 return -ENOMEM;
f9ead18c
JA		 240 ret = io_sendmsg_copy_hdr(req, req->async_data);
241 if (!ret)
242 req->flags |= REQ_F_NEED_CLEANUP;
243 return ret;
244}
245
246void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
247{
248 struct io_async_msghdr *io = req->async_data;
249
250 kfree(io->free_iov);
251}
252
253int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
254{
f2ccb5ae 255 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c 256
516e82f0
PB		 257 if (req->opcode == IORING_OP_SEND) {
258 if (READ_ONCE(sqe->__pad3[0]))
259 return -EINVAL;
260 sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
261 sr->addr_len = READ_ONCE(sqe->addr_len);
262 } else if (sqe->addr2 || sqe->file_index) {
f9ead18c 263 return -EINVAL;
516e82f0 264 }
f9ead18c
JA		 265
266 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
267 sr->len = READ_ONCE(sqe->len);
268 sr->flags = READ_ONCE(sqe->ioprio);
269 if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
270 return -EINVAL;
271 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
272 if (sr->msg_flags & MSG_DONTWAIT)
273 req->flags |= REQ_F_NOWAIT;
274
275#ifdef CONFIG_COMPAT
276 if (req->ctx->compat)
277 sr->msg_flags |= MSG_CMSG_COMPAT;
278#endif
279 sr->done_io = 0;
280 return 0;
281}
282
283int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
284{
f2ccb5ae 285 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 286 struct io_async_msghdr iomsg, *kmsg;
287 struct socket *sock;
288 unsigned flags;
289 int min_ret = 0;
290 int ret;
291
292 sock = sock_from_file(req->file);
293 if (unlikely(!sock))
294 return -ENOTSOCK;
295
296 if (req_has_async_data(req)) {
297 kmsg = req->async_data;
298 } else {
299 ret = io_sendmsg_copy_hdr(req, &iomsg);
300 if (ret)
301 return ret;
302 kmsg = &iomsg;
303 }
304
305 if (!(req->flags & REQ_F_POLLED) &&
306 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 307 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 308
309 flags = sr->msg_flags;
310 if (issue_flags & IO_URING_F_NONBLOCK)
311 flags |= MSG_DONTWAIT;
312 if (flags & MSG_WAITALL)
313 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
314
315 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
316
317 if (ret < min_ret) {
318 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
43e0bbbd 319 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 320 if (ret > 0 && io_net_retry(sock, flags)) {
321 sr->done_io += ret;
322 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 323 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 324 }
95eafc74
PB		 325 if (ret == -ERESTARTSYS)
326 ret = -EINTR;
f9ead18c
JA		 327 req_set_fail(req);
328 }
329 /* fast path, check for non-NULL to avoid function call */
330 if (kmsg->free_iov)
331 kfree(kmsg->free_iov);
332 req->flags &= ~REQ_F_NEED_CLEANUP;
43e0bbbd 333 io_netmsg_recycle(req, issue_flags);
f9ead18c
JA		 334 if (ret >= 0)
335 ret += sr->done_io;
336 else if (sr->done_io)
337 ret = sr->done_io;
338 io_req_set_res(req, ret, 0);
339 return IOU_OK;
340}
341
342int io_send(struct io_kiocb *req, unsigned int issue_flags)
343{
516e82f0 344 struct sockaddr_storage __address;
f2ccb5ae 345 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 346 struct msghdr msg;
347 struct iovec iov;
348 struct socket *sock;
349 unsigned flags;
350 int min_ret = 0;
351 int ret;
352
04360d3e
PB		 353 msg.msg_name = NULL;
354 msg.msg_control = NULL;
355 msg.msg_controllen = 0;
356 msg.msg_namelen = 0;
357 msg.msg_ubuf = NULL;
358
516e82f0
PB		 359 if (sr->addr) {
360 if (req_has_async_data(req)) {
361 struct io_async_msghdr *io = req->async_data;
362
363 msg.msg_name = &io->addr;
364 } else {
365 ret = move_addr_to_kernel(sr->addr, sr->addr_len, &__address);
366 if (unlikely(ret < 0))
367 return ret;
368 msg.msg_name = (struct sockaddr *)&__address;
369 }
370 msg.msg_namelen = sr->addr_len;
371 }
372
f9ead18c
JA		 373 if (!(req->flags & REQ_F_POLLED) &&
374 (sr->flags & IORING_RECVSEND_POLL_FIRST))
516e82f0 375 return io_setup_async_addr(req, &__address, issue_flags);
f9ead18c
JA		 376
377 sock = sock_from_file(req->file);
378 if (unlikely(!sock))
379 return -ENOTSOCK;
380
de4eda9d 381 ret = import_single_range(ITER_SOURCE, sr->buf, sr->len, &iov, &msg.msg_iter);
f9ead18c
JA		 382 if (unlikely(ret))
383 return ret;
384
f9ead18c
JA		 385 flags = sr->msg_flags;
386 if (issue_flags & IO_URING_F_NONBLOCK)
387 flags |= MSG_DONTWAIT;
388 if (flags & MSG_WAITALL)
389 min_ret = iov_iter_count(&msg.msg_iter);
390
391 msg.msg_flags = flags;
392 ret = sock_sendmsg(sock, &msg);
393 if (ret < min_ret) {
394 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
516e82f0
PB		 395 return io_setup_async_addr(req, &__address, issue_flags);
396
f9ead18c
JA		 397 if (ret > 0 && io_net_retry(sock, flags)) {
398 sr->len -= ret;
399 sr->buf += ret;
400 sr->done_io += ret;
401 req->flags |= REQ_F_PARTIAL_IO;
516e82f0 402 return io_setup_async_addr(req, &__address, issue_flags);
f9ead18c 403 }
95eafc74
PB		 404 if (ret == -ERESTARTSYS)
405 ret = -EINTR;
f9ead18c
JA		 406 req_set_fail(req);
407 }
408 if (ret >= 0)
409 ret += sr->done_io;
410 else if (sr->done_io)
411 ret = sr->done_io;
412 io_req_set_res(req, ret, 0);
413 return IOU_OK;
414}
415
9bb66906
DY		 416static bool io_recvmsg_multishot_overflow(struct io_async_msghdr *iomsg)
417{
9b0fc3c0 418 int hdr;
9bb66906 419
9b0fc3c0 420 if (iomsg->namelen < 0)
9bb66906 421 return true;
9b0fc3c0
DY		 422 if (check_add_overflow((int)sizeof(struct io_uring_recvmsg_out),
423 iomsg->namelen, &hdr))
9bb66906 424 return true;
9b0fc3c0 425 if (check_add_overflow(hdr, (int)iomsg->controllen, &hdr))
9bb66906
DY		 426 return true;
427
428 return false;
429}
430
f9ead18c
JA		 431static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
432 struct io_async_msghdr *iomsg)
433{
f2ccb5ae 434 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
7fa875b8 435 struct user_msghdr msg;
f9ead18c
JA		 436 int ret;
437
7fa875b8
DY		 438 if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
439 return -EFAULT;
440
441 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 442 if (ret)
443 return ret;
444
445 if (req->flags & REQ_F_BUFFER_SELECT) {
7fa875b8 446 if (msg.msg_iovlen == 0) {
5702196e
DY		 447 sr->len = iomsg->fast_iov[0].iov_len = 0;
448 iomsg->fast_iov[0].iov_base = NULL;
449 iomsg->free_iov = NULL;
7fa875b8 450 } else if (msg.msg_iovlen > 1) {
f9ead18c 451 return -EINVAL;
5702196e 452 } else {
7fa875b8 453 if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
5702196e
DY		 454 return -EFAULT;
455 sr->len = iomsg->fast_iov[0].iov_len;
456 iomsg->free_iov = NULL;
457 }
9bb66906
DY		 458
459 if (req->flags & REQ_F_APOLL_MULTISHOT) {
460 iomsg->namelen = msg.msg_namelen;
461 iomsg->controllen = msg.msg_controllen;
462 if (io_recvmsg_multishot_overflow(iomsg))
463 return -EOVERFLOW;
464 }
f9ead18c
JA		 465 } else {
466 iomsg->free_iov = iomsg->fast_iov;
de4eda9d 467 ret = __import_iovec(ITER_DEST, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
f9ead18c
JA		 468 &iomsg->free_iov, &iomsg->msg.msg_iter,
469 false);
470 if (ret > 0)
471 ret = 0;
472 }
473
474 return ret;
475}
476
477#ifdef CONFIG_COMPAT
478static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
479 struct io_async_msghdr *iomsg)
480{
f2ccb5ae 481 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
72c531f8 482 struct compat_msghdr msg;
f9ead18c 483 struct compat_iovec __user *uiov;
f9ead18c
JA		 484 int ret;
485
72c531f8
DY		 486 if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
487 return -EFAULT;
488
4f6a94d3 489 ret = __get_compat_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 490 if (ret)
491 return ret;
492
72c531f8 493 uiov = compat_ptr(msg.msg_iov);
f9ead18c
JA		 494 if (req->flags & REQ_F_BUFFER_SELECT) {
495 compat_ssize_t clen;
496
72c531f8 497 if (msg.msg_iovlen == 0) {
6d2f75a0 498 sr->len = 0;
72c531f8 499 } else if (msg.msg_iovlen > 1) {
f9ead18c 500 return -EINVAL;
6d2f75a0
DY		 501 } else {
502 if (!access_ok(uiov, sizeof(*uiov)))
503 return -EFAULT;
504 if (__get_user(clen, &uiov->iov_len))
505 return -EFAULT;
506 if (clen < 0)
507 return -EINVAL;
508 sr->len = clen;
6d2f75a0 509 }
9bb66906
DY		 510
511 if (req->flags & REQ_F_APOLL_MULTISHOT) {
512 iomsg->namelen = msg.msg_namelen;
513 iomsg->controllen = msg.msg_controllen;
514 if (io_recvmsg_multishot_overflow(iomsg))
515 return -EOVERFLOW;
516 }
f9ead18c
JA		 517 } else {
518 iomsg->free_iov = iomsg->fast_iov;
de4eda9d 519 ret = __import_iovec(ITER_DEST, (struct iovec __user *)uiov, msg.msg_iovlen,
f9ead18c
JA		 520 UIO_FASTIOV, &iomsg->free_iov,
521 &iomsg->msg.msg_iter, true);
522 if (ret < 0)
523 return ret;
524 }
525
526 return 0;
527}
528#endif
529
530static int io_recvmsg_copy_hdr(struct io_kiocb *req,
531 struct io_async_msghdr *iomsg)
532{
533 iomsg->msg.msg_name = &iomsg->addr;
534
535#ifdef CONFIG_COMPAT
536 if (req->ctx->compat)
537 return __io_compat_recvmsg_copy_hdr(req, iomsg);
538#endif
539
540 return __io_recvmsg_copy_hdr(req, iomsg);
541}
542
543int io_recvmsg_prep_async(struct io_kiocb *req)
544{
545 int ret;
546
858c293e
PB		 547 if (!io_msg_alloc_async_prep(req))
548 return -ENOMEM;
f9ead18c
JA		 549 ret = io_recvmsg_copy_hdr(req, req->async_data);
550 if (!ret)
551 req->flags |= REQ_F_NEED_CLEANUP;
552 return ret;
553}
554
b3fdea6e
DY		 555#define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
556
f9ead18c
JA		 557int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
558{
f2ccb5ae 559 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 560
561 if (unlikely(sqe->file_index || sqe->addr2))
562 return -EINVAL;
563
564 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
565 sr->len = READ_ONCE(sqe->len);
566 sr->flags = READ_ONCE(sqe->ioprio);
b3fdea6e 567 if (sr->flags & ~(RECVMSG_FLAGS))
f9ead18c
JA		 568 return -EINVAL;
569 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
570 if (sr->msg_flags & MSG_DONTWAIT)
571 req->flags |= REQ_F_NOWAIT;
572 if (sr->msg_flags & MSG_ERRQUEUE)
573 req->flags |= REQ_F_CLEAR_POLLIN;
b3fdea6e
DY		 574 if (sr->flags & IORING_RECV_MULTISHOT) {
575 if (!(req->flags & REQ_F_BUFFER_SELECT))
576 return -EINVAL;
577 if (sr->msg_flags & MSG_WAITALL)
578 return -EINVAL;
579 if (req->opcode == IORING_OP_RECV && sr->len)
580 return -EINVAL;
581 req->flags |= REQ_F_APOLL_MULTISHOT;
582 }
f9ead18c
JA		 583
584#ifdef CONFIG_COMPAT
585 if (req->ctx->compat)
586 sr->msg_flags |= MSG_CMSG_COMPAT;
587#endif
588 sr->done_io = 0;
589 return 0;
590}
591
b3fdea6e
DY		 592static inline void io_recv_prep_retry(struct io_kiocb *req)
593{
f2ccb5ae 594 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
b3fdea6e
DY		 595
596 sr->done_io = 0;
597 sr->len = 0; /* get from the provided buffer */
598}
599
600/*
9bb66906 601 * Finishes io_recv and io_recvmsg.
b3fdea6e
DY		 602 *
603 * Returns true if it is actually finished, or false if it should run
604 * again (for multishot).
605 */
9bb66906 606static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
100d6b17
PB		 607 unsigned int cflags, bool mshot_finished,
608 unsigned issue_flags)
b3fdea6e
DY		 609{
610 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
611 io_req_set_res(req, *ret, cflags);
612 *ret = IOU_OK;
613 return true;
614 }
615
9bb66906 616 if (!mshot_finished) {
9b8c5475
DY		 617 if (io_aux_cqe(req->ctx, issue_flags & IO_URING_F_COMPLETE_DEFER,
618 req->cqe.user_data, *ret, cflags | IORING_CQE_F_MORE, true)) {
b3fdea6e
DY		 619 io_recv_prep_retry(req);
620 return false;
621 }
e2ad599d 622 /* Otherwise stop multishot but use the current result. */
b3fdea6e
DY		 623 }
624
625 io_req_set_res(req, *ret, cflags);
626
100d6b17 627 if (issue_flags & IO_URING_F_MULTISHOT)
b3fdea6e 628 *ret = IOU_STOP_MULTISHOT;
e2df2ccb
DY		 629 else
630 *ret = IOU_OK;
b3fdea6e
DY		 631 return true;
632}
633
9bb66906
DY		 634static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
635 struct io_sr_msg *sr, void __user **buf,
636 size_t *len)
637{
638 unsigned long ubuf = (unsigned long) *buf;
639 unsigned long hdr;
640
641 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
642 kmsg->controllen;
643 if (*len < hdr)
644 return -EFAULT;
645
646 if (kmsg->controllen) {
647 unsigned long control = ubuf + hdr - kmsg->controllen;
648
d1f6222c 649 kmsg->msg.msg_control_user = (void __user *) control;
9bb66906
DY		 650 kmsg->msg.msg_controllen = kmsg->controllen;
651 }
652
653 sr->buf = *buf; /* stash for later copy */
d1f6222c 654 *buf = (void __user *) (ubuf + hdr);
9bb66906
DY		 655 kmsg->payloadlen = *len = *len - hdr;
656 return 0;
657}
658
659struct io_recvmsg_multishot_hdr {
660 struct io_uring_recvmsg_out msg;
661 struct sockaddr_storage addr;
662};
663
664static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
665 struct io_async_msghdr *kmsg,
666 unsigned int flags, bool *finished)
667{
668 int err;
669 int copy_len;
670 struct io_recvmsg_multishot_hdr hdr;
671
672 if (kmsg->namelen)
673 kmsg->msg.msg_name = &hdr.addr;
674 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
675 kmsg->msg.msg_namelen = 0;
676
677 if (sock->file->f_flags & O_NONBLOCK)
678 flags |= MSG_DONTWAIT;
679
680 err = sock_recvmsg(sock, &kmsg->msg, flags);
681 *finished = err <= 0;
682 if (err < 0)
683 return err;
684
685 hdr.msg = (struct io_uring_recvmsg_out) {
686 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
687 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
688 };
689
690 hdr.msg.payloadlen = err;
691 if (err > kmsg->payloadlen)
692 err = kmsg->payloadlen;
693
694 copy_len = sizeof(struct io_uring_recvmsg_out);
695 if (kmsg->msg.msg_namelen > kmsg->namelen)
696 copy_len += kmsg->namelen;
697 else
698 copy_len += kmsg->msg.msg_namelen;
699
700 /*
701 * "fromlen shall refer to the value before truncation.."
702 * 1003.1g
703 */
704 hdr.msg.namelen = kmsg->msg.msg_namelen;
705
706 /* ensure that there is no gap between hdr and sockaddr_storage */
707 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
708 sizeof(struct io_uring_recvmsg_out));
709 if (copy_to_user(io->buf, &hdr, copy_len)) {
710 *finished = true;
711 return -EFAULT;
712 }
713
714 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
715 kmsg->controllen + err;
716}
717
f9ead18c
JA		 718int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
719{
f2ccb5ae 720 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 721 struct io_async_msghdr iomsg, *kmsg;
722 struct socket *sock;
723 unsigned int cflags;
724 unsigned flags;
725 int ret, min_ret = 0;
726 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
9bb66906 727 bool mshot_finished = true;
f9ead18c
JA		 728
729 sock = sock_from_file(req->file);
730 if (unlikely(!sock))
731 return -ENOTSOCK;
732
733 if (req_has_async_data(req)) {
734 kmsg = req->async_data;
735 } else {
736 ret = io_recvmsg_copy_hdr(req, &iomsg);
737 if (ret)
738 return ret;
739 kmsg = &iomsg;
740 }
741
742 if (!(req->flags & REQ_F_POLLED) &&
743 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 744 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 745
17add5ce
PB		 746 if (!io_check_multishot(req, issue_flags))
747 return io_setup_async_msg(req, kmsg, issue_flags);
748
9bb66906 749retry_multishot:
f9ead18c
JA		 750 if (io_do_buffer_select(req)) {
751 void __user *buf;
9bb66906 752 size_t len = sr->len;
f9ead18c 753
9bb66906 754 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 755 if (!buf)
756 return -ENOBUFS;
9bb66906
DY		 757
758 if (req->flags & REQ_F_APOLL_MULTISHOT) {
759 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
760 if (ret) {
761 io_kbuf_recycle(req, issue_flags);
762 return ret;
763 }
764 }
765
f9ead18c 766 kmsg->fast_iov[0].iov_base = buf;
9bb66906 767 kmsg->fast_iov[0].iov_len = len;
de4eda9d 768 iov_iter_init(&kmsg->msg.msg_iter, ITER_DEST, kmsg->fast_iov, 1,
9bb66906 769 len);
f9ead18c
JA		 770 }
771
772 flags = sr->msg_flags;
773 if (force_nonblock)
774 flags |= MSG_DONTWAIT;
775 if (flags & MSG_WAITALL)
776 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
777
778 kmsg->msg.msg_get_inq = 1;
9bb66906
DY		 779 if (req->flags & REQ_F_APOLL_MULTISHOT)
780 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
781 &mshot_finished);
782 else
783 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
784 kmsg->uaddr, flags);
785
f9ead18c 786 if (ret < min_ret) {
9bb66906
DY		 787 if (ret == -EAGAIN && force_nonblock) {
788 ret = io_setup_async_msg(req, kmsg, issue_flags);
100d6b17 789 if (ret == -EAGAIN && (issue_flags & IO_URING_F_MULTISHOT)) {
9bb66906
DY		 790 io_kbuf_recycle(req, issue_flags);
791 return IOU_ISSUE_SKIP_COMPLETE;
792 }
793 return ret;
794 }
f9ead18c
JA		 795 if (ret > 0 && io_net_retry(sock, flags)) {
796 sr->done_io += ret;
797 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 798 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 799 }
95eafc74
PB		 800 if (ret == -ERESTARTSYS)
801 ret = -EINTR;
f9ead18c
JA		 802 req_set_fail(req);
803 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
804 req_set_fail(req);
805 }
806
d4e097da 807 if (ret > 0)
f9ead18c
JA		 808 ret += sr->done_io;
809 else if (sr->done_io)
810 ret = sr->done_io;
d4e097da
DY		 811 else
812 io_kbuf_recycle(req, issue_flags);
813
f9ead18c
JA		 814 cflags = io_put_kbuf(req, issue_flags);
815 if (kmsg->msg.msg_inq)
816 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 817
100d6b17 818 if (!io_recv_finish(req, &ret, cflags, mshot_finished, issue_flags))
9bb66906
DY		 819 goto retry_multishot;
820
821 if (mshot_finished) {
822 io_netmsg_recycle(req, issue_flags);
823 /* fast path, check for non-NULL to avoid function call */
824 if (kmsg->free_iov)
825 kfree(kmsg->free_iov);
826 req->flags &= ~REQ_F_NEED_CLEANUP;
827 }
828
829 return ret;
f9ead18c
JA		 830}
831
832int io_recv(struct io_kiocb *req, unsigned int issue_flags)
833{
f2ccb5ae 834 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 835 struct msghdr msg;
836 struct socket *sock;
837 struct iovec iov;
838 unsigned int cflags;
839 unsigned flags;
840 int ret, min_ret = 0;
841 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
b3fdea6e 842 size_t len = sr->len;
f9ead18c
JA		 843
844 if (!(req->flags & REQ_F_POLLED) &&
845 (sr->flags & IORING_RECVSEND_POLL_FIRST))
846 return -EAGAIN;
847
17add5ce
PB		 848 if (!io_check_multishot(req, issue_flags))
849 return -EAGAIN;
850
f9ead18c
JA		 851 sock = sock_from_file(req->file);
852 if (unlikely(!sock))
853 return -ENOTSOCK;
854
b3fdea6e 855retry_multishot:
f9ead18c
JA		 856 if (io_do_buffer_select(req)) {
857 void __user *buf;
858
b3fdea6e 859 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 860 if (!buf)
861 return -ENOBUFS;
862 sr->buf = buf;
863 }
864
de4eda9d 865 ret = import_single_range(ITER_DEST, sr->buf, len, &iov, &msg.msg_iter);
f9ead18c
JA		 866 if (unlikely(ret))
867 goto out_free;
868
869 msg.msg_name = NULL;
870 msg.msg_namelen = 0;
871 msg.msg_control = NULL;
872 msg.msg_get_inq = 1;
873 msg.msg_flags = 0;
874 msg.msg_controllen = 0;
875 msg.msg_iocb = NULL;
e02b6651 876 msg.msg_ubuf = NULL;
f9ead18c
JA		 877
878 flags = sr->msg_flags;
879 if (force_nonblock)
880 flags |= MSG_DONTWAIT;
881 if (flags & MSG_WAITALL)
882 min_ret = iov_iter_count(&msg.msg_iter);
883
884 ret = sock_recvmsg(sock, &msg, flags);
885 if (ret < min_ret) {
b3fdea6e 886 if (ret == -EAGAIN && force_nonblock) {
100d6b17 887 if (issue_flags & IO_URING_F_MULTISHOT) {
b3fdea6e
DY		 888 io_kbuf_recycle(req, issue_flags);
889 return IOU_ISSUE_SKIP_COMPLETE;
890 }
891
f9ead18c 892 return -EAGAIN;
b3fdea6e 893 }
f9ead18c
JA		 894 if (ret > 0 && io_net_retry(sock, flags)) {
895 sr->len -= ret;
896 sr->buf += ret;
897 sr->done_io += ret;
898 req->flags |= REQ_F_PARTIAL_IO;
899 return -EAGAIN;
900 }
95eafc74
PB		 901 if (ret == -ERESTARTSYS)
902 ret = -EINTR;
f9ead18c
JA		 903 req_set_fail(req);
904 } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
905out_free:
906 req_set_fail(req);
907 }
908
d4e097da 909 if (ret > 0)
f9ead18c
JA		 910 ret += sr->done_io;
911 else if (sr->done_io)
912 ret = sr->done_io;
d4e097da
DY		 913 else
914 io_kbuf_recycle(req, issue_flags);
915
f9ead18c
JA		 916 cflags = io_put_kbuf(req, issue_flags);
917 if (msg.msg_inq)
918 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 919
100d6b17 920 if (!io_recv_finish(req, &ret, cflags, ret <= 0, issue_flags))
b3fdea6e
DY		 921 goto retry_multishot;
922
923 return ret;
f9ead18c
JA		 924}
925
b0e9b551 926void io_send_zc_cleanup(struct io_kiocb *req)
b48c312b 927{
ac9e5784 928 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
493108d9 929 struct io_async_msghdr *io;
b48c312b 930
493108d9
PB		 931 if (req_has_async_data(req)) {
932 io = req->async_data;
4c17a496
PB		 933 /* might be ->fast_iov if *msg_copy_hdr failed */
934 if (io->free_iov != io->fast_iov)
935 kfree(io->free_iov);
493108d9 936 }
a75155fa 937 if (zc->notif) {
a75155fa
PB		 938 io_notif_flush(zc->notif);
939 zc->notif = NULL;
940 }
b48c312b
PB		 941}
942
40725d1b
PB		 943#define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF)
944#define IO_ZC_FLAGS_VALID (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE)
945
b0e9b551 946int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
06a5464b 947{
ac9e5784 948 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
10c7d33e 949 struct io_ring_ctx *ctx = req->ctx;
b48c312b 950 struct io_kiocb *notif;
06a5464b 951
493108d9 952 if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
b48c312b
PB		 953 return -EINVAL;
954 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
955 if (req->flags & REQ_F_CQE_SKIP)
06a5464b
PB		 956 return -EINVAL;
957
e3366e02
PB		 958 notif = zc->notif = io_alloc_notif(ctx);
959 if (!notif)
960 return -ENOMEM;
961 notif->cqe.user_data = req->cqe.user_data;
962 notif->cqe.res = 0;
963 notif->cqe.flags = IORING_CQE_F_NOTIF;
964 req->flags |= REQ_F_NEED_CLEANUP;
40725d1b
PB		 965
966 zc->flags = READ_ONCE(sqe->ioprio);
967 if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) {
968 if (zc->flags & ~IO_ZC_FLAGS_VALID)
969 return -EINVAL;
970 if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) {
971 io_notif_set_extended(notif);
972 io_notif_to_data(notif)->zc_report = true;
973 }
974 }
975
10c7d33e
PB		 976 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
977 unsigned idx = READ_ONCE(sqe->buf_index);
978
979 if (unlikely(idx >= ctx->nr_user_bufs))
980 return -EFAULT;
981 idx = array_index_nospec(idx, ctx->nr_user_bufs);
982 req->imu = READ_ONCE(ctx->user_bufs[idx]);
e3366e02 983 io_req_set_rsrc_node(notif, ctx, 0);
10c7d33e 984 }
06a5464b 985
493108d9
PB		 986 if (req->opcode == IORING_OP_SEND_ZC) {
987 if (READ_ONCE(sqe->__pad3[0]))
988 return -EINVAL;
989 zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
990 zc->addr_len = READ_ONCE(sqe->addr_len);
991 } else {
992 if (unlikely(sqe->addr2 || sqe->file_index))
993 return -EINVAL;
994 if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
995 return -EINVAL;
996 }
997
06a5464b
PB		 998 zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
999 zc->len = READ_ONCE(sqe->len);
1000 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
06a5464b
PB		 1001 if (zc->msg_flags & MSG_DONTWAIT)
1002 req->flags |= REQ_F_NOWAIT;
092aeedb 1003
4a933e62 1004 zc->done_io = 0;
092aeedb 1005
06a5464b
PB		 1006#ifdef CONFIG_COMPAT
1007 if (req->ctx->compat)
1008 zc->msg_flags |= MSG_CMSG_COMPAT;
1009#endif
1010 return 0;
1011}
1012
cd9021e8
PB		 1013static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
1014 struct iov_iter *from, size_t length)
1015{
1016 skb_zcopy_downgrade_managed(skb);
1017 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1018}
1019
3ff1a0d3
PB		 1020static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1021 struct iov_iter *from, size_t length)
1022{
1023 struct skb_shared_info *shinfo = skb_shinfo(skb);
1024 int frag = shinfo->nr_frags;
1025 int ret = 0;
1026 struct bvec_iter bi;
1027 ssize_t copied = 0;
1028 unsigned long truesize = 0;
1029
cd9021e8 1030 if (!frag)
3ff1a0d3 1031 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
cd9021e8 1032 else if (unlikely(!skb_zcopy_managed(skb)))
3ff1a0d3 1033 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
3ff1a0d3
PB		 1034
1035 bi.bi_size = min(from->count, length);
1036 bi.bi_bvec_done = from->iov_offset;
1037 bi.bi_idx = 0;
1038
1039 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1040 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1041
1042 copied += v.bv_len;
1043 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1044 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1045 v.bv_offset, v.bv_len);
1046 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1047 }
1048 if (bi.bi_size)
1049 ret = -EMSGSIZE;
1050
1051 shinfo->nr_frags = frag;
1052 from->bvec += bi.bi_idx;
1053 from->nr_segs -= bi.bi_idx;
dfb58b17 1054 from->count -= copied;
3ff1a0d3
PB		 1055 from->iov_offset = bi.bi_bvec_done;
1056
1057 skb->data_len += copied;
1058 skb->len += copied;
1059 skb->truesize += truesize;
1060
1061 if (sk && sk->sk_type == SOCK_STREAM) {
1062 sk_wmem_queued_add(sk, truesize);
1063 if (!skb_zcopy_pure(skb))
1064 sk_mem_charge(sk, truesize);
1065 } else {
1066 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1067 }
1068 return ret;
1069}
1070
b0e9b551 1071int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
06a5464b 1072{
6ae61b7a 1073 struct sockaddr_storage __address;
ac9e5784 1074 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
06a5464b
PB		 1075 struct msghdr msg;
1076 struct iovec iov;
1077 struct socket *sock;
6ae91ac9 1078 unsigned msg_flags;
06a5464b
PB		 1079 int ret, min_ret = 0;
1080
06a5464b
PB		 1081 sock = sock_from_file(req->file);
1082 if (unlikely(!sock))
1083 return -ENOTSOCK;
edf81438
PB		 1084 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1085 return -EOPNOTSUPP;
06a5464b 1086
06a5464b
PB		 1087 msg.msg_name = NULL;
1088 msg.msg_control = NULL;
1089 msg.msg_controllen = 0;
1090 msg.msg_namelen = 0;
1091
86dc8f23 1092 if (zc->addr) {
581711c4
PB		 1093 if (req_has_async_data(req)) {
1094 struct io_async_msghdr *io = req->async_data;
1095
6ae61b7a 1096 msg.msg_name = &io->addr;
581711c4
PB		 1097 } else {
1098 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1099 if (unlikely(ret < 0))
1100 return ret;
1101 msg.msg_name = (struct sockaddr *)&__address;
581711c4 1102 }
86dc8f23
PB		 1103 msg.msg_namelen = zc->addr_len;
1104 }
1105
3c840053
PB		 1106 if (!(req->flags & REQ_F_POLLED) &&
1107 (zc->flags & IORING_RECVSEND_POLL_FIRST))
6ae61b7a 1108 return io_setup_async_addr(req, &__address, issue_flags);
3c840053 1109
10c7d33e 1110 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
de4eda9d 1111 ret = io_import_fixed(ITER_SOURCE, &msg.msg_iter, req->imu,
10c7d33e
PB		 1112 (u64)(uintptr_t)zc->buf, zc->len);
1113 if (unlikely(ret))
986e263d 1114 return ret;
cd9021e8 1115 msg.sg_from_iter = io_sg_from_iter;
10c7d33e 1116 } else {
42385b02 1117 io_notif_set_extended(zc->notif);
de4eda9d 1118 ret = import_single_range(ITER_SOURCE, zc->buf, zc->len, &iov,
10c7d33e
PB		 1119 &msg.msg_iter);
1120 if (unlikely(ret))
1121 return ret;
b48c312b 1122 ret = io_notif_account_mem(zc->notif, zc->len);
2e32ba56
PB		 1123 if (unlikely(ret))
1124 return ret;
cd9021e8 1125 msg.sg_from_iter = io_sg_from_iter_iovec;
10c7d33e 1126 }
06a5464b
PB		 1127
1128 msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1129 if (issue_flags & IO_URING_F_NONBLOCK)
1130 msg_flags |= MSG_DONTWAIT;
1131 if (msg_flags & MSG_WAITALL)
1132 min_ret = iov_iter_count(&msg.msg_iter);
1133
1134 msg.msg_flags = msg_flags;
b48c312b 1135 msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
06a5464b
PB		 1136 ret = sock_sendmsg(sock, &msg);
1137
1138 if (unlikely(ret < min_ret)) {
1139 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
6ae61b7a 1140 return io_setup_async_addr(req, &__address, issue_flags);
581711c4 1141
4a933e62
PB		 1142 if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1143 zc->len -= ret;
1144 zc->buf += ret;
1145 zc->done_io += ret;
1146 req->flags |= REQ_F_PARTIAL_IO;
6ae61b7a 1147 return io_setup_async_addr(req, &__address, issue_flags);
4a933e62
PB		 1148 }
1149 if (ret == -ERESTARTSYS)
1150 ret = -EINTR;
5a848b7c 1151 req_set_fail(req);
06a5464b
PB		 1152 }
1153
4a933e62
PB		 1154 if (ret >= 0)
1155 ret += zc->done_io;
1156 else if (zc->done_io)
1157 ret = zc->done_io;
b48c312b 1158
108893dd
PB		 1159 /*
1160 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1161 * flushing notif to io_send_zc_cleanup()
1162 */
1163 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1164 io_notif_flush(zc->notif);
1165 req->flags &= ~REQ_F_NEED_CLEANUP;
1166 }
6ae91ac9 1167 io_req_set_res(req, ret, IORING_CQE_F_MORE);
06a5464b
PB		 1168 return IOU_OK;
1169}
1170
493108d9
PB		 1171int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1172{
1173 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1174 struct io_async_msghdr iomsg, *kmsg;
1175 struct socket *sock;
6ae91ac9 1176 unsigned flags;
493108d9
PB		 1177 int ret, min_ret = 0;
1178
42385b02
PB		 1179 io_notif_set_extended(sr->notif);
1180
493108d9
PB		 1181 sock = sock_from_file(req->file);
1182 if (unlikely(!sock))
1183 return -ENOTSOCK;
cc767e7c
PB		 1184 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1185 return -EOPNOTSUPP;
493108d9
PB		 1186
1187 if (req_has_async_data(req)) {
1188 kmsg = req->async_data;
1189 } else {
1190 ret = io_sendmsg_copy_hdr(req, &iomsg);
1191 if (ret)
1192 return ret;
1193 kmsg = &iomsg;
1194 }
1195
1196 if (!(req->flags & REQ_F_POLLED) &&
1197 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1198 return io_setup_async_msg(req, kmsg, issue_flags);
1199
1200 flags = sr->msg_flags | MSG_ZEROCOPY;
1201 if (issue_flags & IO_URING_F_NONBLOCK)
1202 flags |= MSG_DONTWAIT;
1203 if (flags & MSG_WAITALL)
1204 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1205
1206 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1207 kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1208 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1209
1210 if (unlikely(ret < min_ret)) {
1211 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1212 return io_setup_async_msg(req, kmsg, issue_flags);
1213
1214 if (ret > 0 && io_net_retry(sock, flags)) {
1215 sr->done_io += ret;
1216 req->flags |= REQ_F_PARTIAL_IO;
1217 return io_setup_async_msg(req, kmsg, issue_flags);
1218 }
493108d9
PB		 1219 if (ret == -ERESTARTSYS)
1220 ret = -EINTR;
1221 req_set_fail(req);
1222 }
1223 /* fast path, check for non-NULL to avoid function call */
108893dd 1224 if (kmsg->free_iov) {
493108d9 1225 kfree(kmsg->free_iov);
108893dd
PB		 1226 kmsg->free_iov = NULL;
1227 }
493108d9
PB		 1228
1229 io_netmsg_recycle(req, issue_flags);
1230 if (ret >= 0)
1231 ret += sr->done_io;
1232 else if (sr->done_io)
1233 ret = sr->done_io;
1234
108893dd
PB		 1235 /*
1236 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1237 * flushing notif to io_send_zc_cleanup()
1238 */
1239 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1240 io_notif_flush(sr->notif);
1241 req->flags &= ~REQ_F_NEED_CLEANUP;
1242 }
6ae91ac9 1243 io_req_set_res(req, ret, IORING_CQE_F_MORE);
493108d9
PB		 1244 return IOU_OK;
1245}
1246
7e6b638e
PB		 1247void io_sendrecv_fail(struct io_kiocb *req)
1248{
1249 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
7e6b638e
PB		 1250
1251 if (req->flags & REQ_F_PARTIAL_IO)
6ae91ac9
PB		 1252 req->cqe.res = sr->done_io;
1253
c4c0009e 1254 if ((req->flags & REQ_F_NEED_CLEANUP) &&
6ae91ac9
PB		 1255 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1256 req->cqe.flags |= IORING_CQE_F_MORE;
5693bcce
PB		 1257}
1258
f9ead18c
JA		 1259int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1260{
f2ccb5ae 1261 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1262 unsigned flags;
1263
1264 if (sqe->len || sqe->buf_index)
1265 return -EINVAL;
1266
1267 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1268 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1269 accept->flags = READ_ONCE(sqe->accept_flags);
1270 accept->nofile = rlimit(RLIMIT_NOFILE);
1271 flags = READ_ONCE(sqe->ioprio);
1272 if (flags & ~IORING_ACCEPT_MULTISHOT)
1273 return -EINVAL;
1274
1275 accept->file_slot = READ_ONCE(sqe->file_index);
1276 if (accept->file_slot) {
1277 if (accept->flags & SOCK_CLOEXEC)
1278 return -EINVAL;
1279 if (flags & IORING_ACCEPT_MULTISHOT &&
1280 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1281 return -EINVAL;
1282 }
1283 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1284 return -EINVAL;
1285 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1286 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1287 if (flags & IORING_ACCEPT_MULTISHOT)
1288 req->flags |= REQ_F_APOLL_MULTISHOT;
1289 return 0;
1290}
1291
1292int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1293{
1294 struct io_ring_ctx *ctx = req->ctx;
f2ccb5ae 1295 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1296 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1297 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1298 bool fixed = !!accept->file_slot;
1299 struct file *file;
1300 int ret, fd;
1301
17add5ce
PB		 1302 if (!io_check_multishot(req, issue_flags))
1303 return -EAGAIN;
f9ead18c
JA		 1304retry:
1305 if (!fixed) {
1306 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1307 if (unlikely(fd < 0))
1308 return fd;
1309 }
1310 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1311 accept->flags);
1312 if (IS_ERR(file)) {
1313 if (!fixed)
1314 put_unused_fd(fd);
1315 ret = PTR_ERR(file);
1316 if (ret == -EAGAIN && force_nonblock) {
1317 /*
1318 * if it's multishot and polled, we don't need to
1319 * return EAGAIN to arm the poll infra since it
1320 * has already been done
1321 */
91482864 1322 if (issue_flags & IO_URING_F_MULTISHOT)
f9ead18c
JA		 1323 ret = IOU_ISSUE_SKIP_COMPLETE;
1324 return ret;
1325 }
1326 if (ret == -ERESTARTSYS)
1327 ret = -EINTR;
1328 req_set_fail(req);
1329 } else if (!fixed) {
1330 fd_install(fd, file);
1331 ret = fd;
1332 } else {
1333 ret = io_fixed_fd_install(req, issue_flags, file,
1334 accept->file_slot);
1335 }
1336
1337 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1338 io_req_set_res(req, ret, 0);
1339 return IOU_OK;
1340 }
f9ead18c 1341
515e2696
DY		 1342 if (ret < 0)
1343 return ret;
9b8c5475
DY		 1344 if (io_aux_cqe(ctx, issue_flags & IO_URING_F_COMPLETE_DEFER,
1345 req->cqe.user_data, ret, IORING_CQE_F_MORE, true))
d245bca6 1346 goto retry;
cbd25748 1347
515e2696 1348 return -ECANCELED;
f9ead18c
JA		 1349}
1350
1351int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1352{
f2ccb5ae 1353 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1354
1355 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1356 return -EINVAL;
1357
1358 sock->domain = READ_ONCE(sqe->fd);
1359 sock->type = READ_ONCE(sqe->off);
1360 sock->protocol = READ_ONCE(sqe->len);
1361 sock->file_slot = READ_ONCE(sqe->file_index);
1362 sock->nofile = rlimit(RLIMIT_NOFILE);
1363
1364 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1365 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1366 return -EINVAL;
1367 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1368 return -EINVAL;
1369 return 0;
1370}
1371
1372int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1373{
f2ccb5ae 1374 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1375 bool fixed = !!sock->file_slot;
1376 struct file *file;
1377 int ret, fd;
1378
1379 if (!fixed) {
1380 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1381 if (unlikely(fd < 0))
1382 return fd;
1383 }
1384 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1385 if (IS_ERR(file)) {
1386 if (!fixed)
1387 put_unused_fd(fd);
1388 ret = PTR_ERR(file);
1389 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1390 return -EAGAIN;
1391 if (ret == -ERESTARTSYS)
1392 ret = -EINTR;
1393 req_set_fail(req);
1394 } else if (!fixed) {
1395 fd_install(fd, file);
1396 ret = fd;
1397 } else {
1398 ret = io_fixed_fd_install(req, issue_flags, file,
1399 sock->file_slot);
1400 }
1401 io_req_set_res(req, ret, 0);
1402 return IOU_OK;
1403}
1404
1405int io_connect_prep_async(struct io_kiocb *req)
1406{
1407 struct io_async_connect *io = req->async_data;
f2ccb5ae 1408 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1409
1410 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1411}
1412
1413int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1414{
f2ccb5ae 1415 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1416
1417 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1418 return -EINVAL;
1419
1420 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1421 conn->addr_len = READ_ONCE(sqe->addr2);
3fb1bd68 1422 conn->in_progress = false;
f9ead18c
JA		 1423 return 0;
1424}
1425
1426int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1427{
f2ccb5ae 1428 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1429 struct io_async_connect __io, *io;
1430 unsigned file_flags;
1431 int ret;
1432 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1433
3fb1bd68
JA		 1434 if (connect->in_progress) {
1435 struct socket *socket;
1436
1437 ret = -ENOTSOCK;
1438 socket = sock_from_file(req->file);
1439 if (socket)
1440 ret = sock_error(socket->sk);
1441 goto out;
1442 }
1443
f9ead18c
JA		 1444 if (req_has_async_data(req)) {
1445 io = req->async_data;
1446 } else {
1447 ret = move_addr_to_kernel(connect->addr,
1448 connect->addr_len,
1449 &__io.address);
1450 if (ret)
1451 goto out;
1452 io = &__io;
1453 }
1454
1455 file_flags = force_nonblock ? O_NONBLOCK : 0;
1456
1457 ret = __sys_connect_file(req->file, &io->address,
1458 connect->addr_len, file_flags);
1459 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
3fb1bd68
JA		 1460 if (ret == -EINPROGRESS) {
1461 connect->in_progress = true;
1462 } else {
1463 if (req_has_async_data(req))
1464 return -EAGAIN;
1465 if (io_alloc_async_data(req)) {
1466 ret = -ENOMEM;
1467 goto out;
1468 }
1469 memcpy(req->async_data, &__io, sizeof(__io));
f9ead18c 1470 }
f9ead18c
JA		 1471 return -EAGAIN;
1472 }
1473 if (ret == -ERESTARTSYS)
1474 ret = -EINTR;
1475out:
1476 if (ret < 0)
1477 req_set_fail(req);
1478 io_req_set_res(req, ret, 0);
1479 return IOU_OK;
1480}
43e0bbbd
JA		 1481
1482void io_netmsg_cache_free(struct io_cache_entry *entry)
1483{
1484 kfree(container_of(entry, struct io_async_msghdr, cache));
1485}
f9ead18c 1486#endif
Linux 6.x block layer and io_uring tree(s)