projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
io_uring/net: fail zc send when unsupported by socket
[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
f9ead18c
JA		 70#define IO_APOLL_MULTI_POLLED (REQ_F_APOLL_MULTISHOT | REQ_F_POLLED)
71
72int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
73{
f2ccb5ae 74 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
f9ead18c
JA		 75
76 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
77 sqe->buf_index || sqe->splice_fd_in))
78 return -EINVAL;
79
80 shutdown->how = READ_ONCE(sqe->len);
81 return 0;
82}
83
84int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
85{
f2ccb5ae 86 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
f9ead18c
JA		 87 struct socket *sock;
88 int ret;
89
90 if (issue_flags & IO_URING_F_NONBLOCK)
91 return -EAGAIN;
92
93 sock = sock_from_file(req->file);
94 if (unlikely(!sock))
95 return -ENOTSOCK;
96
97 ret = __sys_shutdown_sock(sock, shutdown->how);
98 io_req_set_res(req, ret, 0);
99 return IOU_OK;
100}
101
102static bool io_net_retry(struct socket *sock, int flags)
103{
104 if (!(flags & MSG_WAITALL))
105 return false;
106 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
107}
108
43e0bbbd
JA		 109static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
110{
111 struct io_async_msghdr *hdr = req->async_data;
112
06360426 113 if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
43e0bbbd
JA		 114 return;
115
116 /* Let normal cleanup path reap it if we fail adding to the cache */
117 if (io_alloc_cache_put(&req->ctx->netmsg_cache, &hdr->cache)) {
118 req->async_data = NULL;
119 req->flags &= ~REQ_F_ASYNC_DATA;
120 }
121}
122
858c293e
PB		 123static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
124 unsigned int issue_flags)
43e0bbbd
JA		 125{
126 struct io_ring_ctx *ctx = req->ctx;
127 struct io_cache_entry *entry;
4c17a496 128 struct io_async_msghdr *hdr;
43e0bbbd
JA		 129
130 if (!(issue_flags & IO_URING_F_UNLOCKED) &&
131 (entry = io_alloc_cache_get(&ctx->netmsg_cache)) != NULL) {
43e0bbbd 132 hdr = container_of(entry, struct io_async_msghdr, cache);
4c17a496 133 hdr->free_iov = NULL;
43e0bbbd
JA		 134 req->flags |= REQ_F_ASYNC_DATA;
135 req->async_data = hdr;
136 return hdr;
137 }
138
4c17a496
PB		 139 if (!io_alloc_async_data(req)) {
140 hdr = req->async_data;
141 hdr->free_iov = NULL;
142 return hdr;
143 }
43e0bbbd
JA		 144 return NULL;
145}
146
858c293e
PB		 147static inline struct io_async_msghdr *io_msg_alloc_async_prep(struct io_kiocb *req)
148{
149 /* ->prep_async is always called from the submission context */
150 return io_msg_alloc_async(req, 0);
151}
152
f9ead18c 153static int io_setup_async_msg(struct io_kiocb *req,
43e0bbbd
JA		 154 struct io_async_msghdr *kmsg,
155 unsigned int issue_flags)
f9ead18c 156{
3f743e9b 157 struct io_async_msghdr *async_msg;
f9ead18c 158
3f743e9b 159 if (req_has_async_data(req))
f9ead18c 160 return -EAGAIN;
858c293e 161 async_msg = io_msg_alloc_async(req, issue_flags);
43e0bbbd 162 if (!async_msg) {
f9ead18c
JA		 163 kfree(kmsg->free_iov);
164 return -ENOMEM;
165 }
f9ead18c
JA		 166 req->flags |= REQ_F_NEED_CLEANUP;
167 memcpy(async_msg, kmsg, sizeof(*kmsg));
6f10ae8a
PB		 168 if (async_msg->msg.msg_name)
169 async_msg->msg.msg_name = &async_msg->addr;
f9ead18c 170 /* if were using fast_iov, set it to the new one */
3e4cb6eb
SM		 171 if (!kmsg->free_iov) {
172 size_t fast_idx = kmsg->msg.msg_iter.iov - kmsg->fast_iov;
173 async_msg->msg.msg_iter.iov = &async_msg->fast_iov[fast_idx];
174 }
f9ead18c
JA		 175
176 return -EAGAIN;
177}
178
179static int io_sendmsg_copy_hdr(struct io_kiocb *req,
180 struct io_async_msghdr *iomsg)
181{
f2ccb5ae 182 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 183
184 iomsg->msg.msg_name = &iomsg->addr;
185 iomsg->free_iov = iomsg->fast_iov;
186 return sendmsg_copy_msghdr(&iomsg->msg, sr->umsg, sr->msg_flags,
187 &iomsg->free_iov);
188}
189
516e82f0 190int io_send_prep_async(struct io_kiocb *req)
581711c4 191{
ac9e5784 192 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
581711c4
PB		 193 struct io_async_msghdr *io;
194 int ret;
195
196 if (!zc->addr || req_has_async_data(req))
197 return 0;
6bf8ad25
PB		 198 io = io_msg_alloc_async_prep(req);
199 if (!io)
581711c4 200 return -ENOMEM;
581711c4
PB		 201 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
202 return ret;
203}
204
205static int io_setup_async_addr(struct io_kiocb *req,
6ae61b7a 206 struct sockaddr_storage *addr_storage,
581711c4
PB		 207 unsigned int issue_flags)
208{
6ae61b7a 209 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
581711c4
PB		 210 struct io_async_msghdr *io;
211
6ae61b7a 212 if (!sr->addr || req_has_async_data(req))
581711c4 213 return -EAGAIN;
6bf8ad25
PB		 214 io = io_msg_alloc_async(req, issue_flags);
215 if (!io)
581711c4 216 return -ENOMEM;
6ae61b7a 217 memcpy(&io->addr, addr_storage, sizeof(io->addr));
581711c4
PB		 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 243
516e82f0
PB		 244 if (req->opcode == IORING_OP_SEND) {
245 if (READ_ONCE(sqe->__pad3[0]))
246 return -EINVAL;
247 sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
248 sr->addr_len = READ_ONCE(sqe->addr_len);
249 } else if (sqe->addr2 || sqe->file_index) {
f9ead18c 250 return -EINVAL;
516e82f0 251 }
f9ead18c
JA		 252
253 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
254 sr->len = READ_ONCE(sqe->len);
255 sr->flags = READ_ONCE(sqe->ioprio);
256 if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
257 return -EINVAL;
258 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
259 if (sr->msg_flags & MSG_DONTWAIT)
260 req->flags |= REQ_F_NOWAIT;
261
262#ifdef CONFIG_COMPAT
263 if (req->ctx->compat)
264 sr->msg_flags |= MSG_CMSG_COMPAT;
265#endif
266 sr->done_io = 0;
267 return 0;
268}
269
270int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
271{
f2ccb5ae 272 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 273 struct io_async_msghdr iomsg, *kmsg;
274 struct socket *sock;
275 unsigned flags;
276 int min_ret = 0;
277 int ret;
278
279 sock = sock_from_file(req->file);
280 if (unlikely(!sock))
281 return -ENOTSOCK;
282
283 if (req_has_async_data(req)) {
284 kmsg = req->async_data;
285 } else {
286 ret = io_sendmsg_copy_hdr(req, &iomsg);
287 if (ret)
288 return ret;
289 kmsg = &iomsg;
290 }
291
292 if (!(req->flags & REQ_F_POLLED) &&
293 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 294 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 295
296 flags = sr->msg_flags;
297 if (issue_flags & IO_URING_F_NONBLOCK)
298 flags |= MSG_DONTWAIT;
299 if (flags & MSG_WAITALL)
300 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
301
302 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
303
304 if (ret < min_ret) {
305 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
43e0bbbd 306 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 307 if (ret > 0 && io_net_retry(sock, flags)) {
308 sr->done_io += ret;
309 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 310 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 311 }
95eafc74
PB		 312 if (ret == -ERESTARTSYS)
313 ret = -EINTR;
f9ead18c
JA		 314 req_set_fail(req);
315 }
316 /* fast path, check for non-NULL to avoid function call */
317 if (kmsg->free_iov)
318 kfree(kmsg->free_iov);
319 req->flags &= ~REQ_F_NEED_CLEANUP;
43e0bbbd 320 io_netmsg_recycle(req, issue_flags);
f9ead18c
JA		 321 if (ret >= 0)
322 ret += sr->done_io;
323 else if (sr->done_io)
324 ret = sr->done_io;
325 io_req_set_res(req, ret, 0);
326 return IOU_OK;
327}
328
329int io_send(struct io_kiocb *req, unsigned int issue_flags)
330{
516e82f0 331 struct sockaddr_storage __address;
f2ccb5ae 332 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 333 struct msghdr msg;
334 struct iovec iov;
335 struct socket *sock;
336 unsigned flags;
337 int min_ret = 0;
338 int ret;
339
04360d3e
PB		 340 msg.msg_name = NULL;
341 msg.msg_control = NULL;
342 msg.msg_controllen = 0;
343 msg.msg_namelen = 0;
344 msg.msg_ubuf = NULL;
345
516e82f0
PB		 346 if (sr->addr) {
347 if (req_has_async_data(req)) {
348 struct io_async_msghdr *io = req->async_data;
349
350 msg.msg_name = &io->addr;
351 } else {
352 ret = move_addr_to_kernel(sr->addr, sr->addr_len, &__address);
353 if (unlikely(ret < 0))
354 return ret;
355 msg.msg_name = (struct sockaddr *)&__address;
356 }
357 msg.msg_namelen = sr->addr_len;
358 }
359
f9ead18c
JA		 360 if (!(req->flags & REQ_F_POLLED) &&
361 (sr->flags & IORING_RECVSEND_POLL_FIRST))
516e82f0 362 return io_setup_async_addr(req, &__address, issue_flags);
f9ead18c
JA		 363
364 sock = sock_from_file(req->file);
365 if (unlikely(!sock))
366 return -ENOTSOCK;
367
368 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
369 if (unlikely(ret))
370 return ret;
371
f9ead18c
JA		 372 flags = sr->msg_flags;
373 if (issue_flags & IO_URING_F_NONBLOCK)
374 flags |= MSG_DONTWAIT;
375 if (flags & MSG_WAITALL)
376 min_ret = iov_iter_count(&msg.msg_iter);
377
378 msg.msg_flags = flags;
379 ret = sock_sendmsg(sock, &msg);
380 if (ret < min_ret) {
381 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
516e82f0
PB		 382 return io_setup_async_addr(req, &__address, issue_flags);
383
f9ead18c
JA		 384 if (ret > 0 && io_net_retry(sock, flags)) {
385 sr->len -= ret;
386 sr->buf += ret;
387 sr->done_io += ret;
388 req->flags |= REQ_F_PARTIAL_IO;
516e82f0 389 return io_setup_async_addr(req, &__address, issue_flags);
f9ead18c 390 }
95eafc74
PB		 391 if (ret == -ERESTARTSYS)
392 ret = -EINTR;
f9ead18c
JA		 393 req_set_fail(req);
394 }
395 if (ret >= 0)
396 ret += sr->done_io;
397 else if (sr->done_io)
398 ret = sr->done_io;
399 io_req_set_res(req, ret, 0);
400 return IOU_OK;
401}
402
9bb66906
DY		 403static bool io_recvmsg_multishot_overflow(struct io_async_msghdr *iomsg)
404{
9b0fc3c0 405 int hdr;
9bb66906 406
9b0fc3c0 407 if (iomsg->namelen < 0)
9bb66906 408 return true;
9b0fc3c0
DY		 409 if (check_add_overflow((int)sizeof(struct io_uring_recvmsg_out),
410 iomsg->namelen, &hdr))
9bb66906 411 return true;
9b0fc3c0 412 if (check_add_overflow(hdr, (int)iomsg->controllen, &hdr))
9bb66906
DY		 413 return true;
414
415 return false;
416}
417
f9ead18c
JA		 418static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
419 struct io_async_msghdr *iomsg)
420{
f2ccb5ae 421 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
7fa875b8 422 struct user_msghdr msg;
f9ead18c
JA		 423 int ret;
424
7fa875b8
DY		 425 if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
426 return -EFAULT;
427
428 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 429 if (ret)
430 return ret;
431
432 if (req->flags & REQ_F_BUFFER_SELECT) {
7fa875b8 433 if (msg.msg_iovlen == 0) {
5702196e
DY		 434 sr->len = iomsg->fast_iov[0].iov_len = 0;
435 iomsg->fast_iov[0].iov_base = NULL;
436 iomsg->free_iov = NULL;
7fa875b8 437 } else if (msg.msg_iovlen > 1) {
f9ead18c 438 return -EINVAL;
5702196e 439 } else {
7fa875b8 440 if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
5702196e
DY		 441 return -EFAULT;
442 sr->len = iomsg->fast_iov[0].iov_len;
443 iomsg->free_iov = NULL;
444 }
9bb66906
DY		 445
446 if (req->flags & REQ_F_APOLL_MULTISHOT) {
447 iomsg->namelen = msg.msg_namelen;
448 iomsg->controllen = msg.msg_controllen;
449 if (io_recvmsg_multishot_overflow(iomsg))
450 return -EOVERFLOW;
451 }
f9ead18c
JA		 452 } else {
453 iomsg->free_iov = iomsg->fast_iov;
7fa875b8 454 ret = __import_iovec(READ, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
f9ead18c
JA		 455 &iomsg->free_iov, &iomsg->msg.msg_iter,
456 false);
457 if (ret > 0)
458 ret = 0;
459 }
460
461 return ret;
462}
463
464#ifdef CONFIG_COMPAT
465static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
466 struct io_async_msghdr *iomsg)
467{
f2ccb5ae 468 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
72c531f8 469 struct compat_msghdr msg;
f9ead18c 470 struct compat_iovec __user *uiov;
f9ead18c
JA		 471 int ret;
472
72c531f8
DY		 473 if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
474 return -EFAULT;
475
4f6a94d3 476 ret = __get_compat_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 477 if (ret)
478 return ret;
479
72c531f8 480 uiov = compat_ptr(msg.msg_iov);
f9ead18c
JA		 481 if (req->flags & REQ_F_BUFFER_SELECT) {
482 compat_ssize_t clen;
483
72c531f8 484 if (msg.msg_iovlen == 0) {
6d2f75a0 485 sr->len = 0;
72c531f8 486 } else if (msg.msg_iovlen > 1) {
f9ead18c 487 return -EINVAL;
6d2f75a0
DY		 488 } else {
489 if (!access_ok(uiov, sizeof(*uiov)))
490 return -EFAULT;
491 if (__get_user(clen, &uiov->iov_len))
492 return -EFAULT;
493 if (clen < 0)
494 return -EINVAL;
495 sr->len = clen;
6d2f75a0 496 }
9bb66906
DY		 497
498 if (req->flags & REQ_F_APOLL_MULTISHOT) {
499 iomsg->namelen = msg.msg_namelen;
500 iomsg->controllen = msg.msg_controllen;
501 if (io_recvmsg_multishot_overflow(iomsg))
502 return -EOVERFLOW;
503 }
f9ead18c
JA		 504 } else {
505 iomsg->free_iov = iomsg->fast_iov;
72c531f8 506 ret = __import_iovec(READ, (struct iovec __user *)uiov, msg.msg_iovlen,
f9ead18c
JA		 507 UIO_FASTIOV, &iomsg->free_iov,
508 &iomsg->msg.msg_iter, true);
509 if (ret < 0)
510 return ret;
511 }
512
513 return 0;
514}
515#endif
516
517static int io_recvmsg_copy_hdr(struct io_kiocb *req,
518 struct io_async_msghdr *iomsg)
519{
520 iomsg->msg.msg_name = &iomsg->addr;
521
522#ifdef CONFIG_COMPAT
523 if (req->ctx->compat)
524 return __io_compat_recvmsg_copy_hdr(req, iomsg);
525#endif
526
527 return __io_recvmsg_copy_hdr(req, iomsg);
528}
529
530int io_recvmsg_prep_async(struct io_kiocb *req)
531{
532 int ret;
533
858c293e
PB		 534 if (!io_msg_alloc_async_prep(req))
535 return -ENOMEM;
f9ead18c
JA		 536 ret = io_recvmsg_copy_hdr(req, req->async_data);
537 if (!ret)
538 req->flags |= REQ_F_NEED_CLEANUP;
539 return ret;
540}
541
b3fdea6e
DY		 542#define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
543
f9ead18c
JA		 544int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
545{
f2ccb5ae 546 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 547
548 if (unlikely(sqe->file_index || sqe->addr2))
549 return -EINVAL;
550
551 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
552 sr->len = READ_ONCE(sqe->len);
553 sr->flags = READ_ONCE(sqe->ioprio);
b3fdea6e 554 if (sr->flags & ~(RECVMSG_FLAGS))
f9ead18c
JA		 555 return -EINVAL;
556 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
557 if (sr->msg_flags & MSG_DONTWAIT)
558 req->flags |= REQ_F_NOWAIT;
559 if (sr->msg_flags & MSG_ERRQUEUE)
560 req->flags |= REQ_F_CLEAR_POLLIN;
b3fdea6e
DY		 561 if (sr->flags & IORING_RECV_MULTISHOT) {
562 if (!(req->flags & REQ_F_BUFFER_SELECT))
563 return -EINVAL;
564 if (sr->msg_flags & MSG_WAITALL)
565 return -EINVAL;
566 if (req->opcode == IORING_OP_RECV && sr->len)
567 return -EINVAL;
568 req->flags |= REQ_F_APOLL_MULTISHOT;
569 }
f9ead18c
JA		 570
571#ifdef CONFIG_COMPAT
572 if (req->ctx->compat)
573 sr->msg_flags |= MSG_CMSG_COMPAT;
574#endif
575 sr->done_io = 0;
576 return 0;
577}
578
b3fdea6e
DY		 579static inline void io_recv_prep_retry(struct io_kiocb *req)
580{
f2ccb5ae 581 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
b3fdea6e
DY		 582
583 sr->done_io = 0;
584 sr->len = 0; /* get from the provided buffer */
585}
586
587/*
9bb66906 588 * Finishes io_recv and io_recvmsg.
b3fdea6e
DY		 589 *
590 * Returns true if it is actually finished, or false if it should run
591 * again (for multishot).
592 */
9bb66906
DY		 593static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
594 unsigned int cflags, bool mshot_finished)
b3fdea6e
DY		 595{
596 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
597 io_req_set_res(req, *ret, cflags);
598 *ret = IOU_OK;
599 return true;
600 }
601
9bb66906 602 if (!mshot_finished) {
b3fdea6e
DY		 603 if (io_post_aux_cqe(req->ctx, req->cqe.user_data, *ret,
604 cflags | IORING_CQE_F_MORE, false)) {
605 io_recv_prep_retry(req);
606 return false;
607 }
608 /*
609 * Otherwise stop multishot but use the current result.
610 * Probably will end up going into overflow, but this means
611 * we cannot trust the ordering anymore
612 */
613 }
614
615 io_req_set_res(req, *ret, cflags);
616
617 if (req->flags & REQ_F_POLLED)
618 *ret = IOU_STOP_MULTISHOT;
e2df2ccb
DY		 619 else
620 *ret = IOU_OK;
b3fdea6e
DY		 621 return true;
622}
623
9bb66906
DY		 624static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
625 struct io_sr_msg *sr, void __user **buf,
626 size_t *len)
627{
628 unsigned long ubuf = (unsigned long) *buf;
629 unsigned long hdr;
630
631 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
632 kmsg->controllen;
633 if (*len < hdr)
634 return -EFAULT;
635
636 if (kmsg->controllen) {
637 unsigned long control = ubuf + hdr - kmsg->controllen;
638
d1f6222c 639 kmsg->msg.msg_control_user = (void __user *) control;
9bb66906
DY		 640 kmsg->msg.msg_controllen = kmsg->controllen;
641 }
642
643 sr->buf = *buf; /* stash for later copy */
d1f6222c 644 *buf = (void __user *) (ubuf + hdr);
9bb66906
DY		 645 kmsg->payloadlen = *len = *len - hdr;
646 return 0;
647}
648
649struct io_recvmsg_multishot_hdr {
650 struct io_uring_recvmsg_out msg;
651 struct sockaddr_storage addr;
652};
653
654static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
655 struct io_async_msghdr *kmsg,
656 unsigned int flags, bool *finished)
657{
658 int err;
659 int copy_len;
660 struct io_recvmsg_multishot_hdr hdr;
661
662 if (kmsg->namelen)
663 kmsg->msg.msg_name = &hdr.addr;
664 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
665 kmsg->msg.msg_namelen = 0;
666
667 if (sock->file->f_flags & O_NONBLOCK)
668 flags |= MSG_DONTWAIT;
669
670 err = sock_recvmsg(sock, &kmsg->msg, flags);
671 *finished = err <= 0;
672 if (err < 0)
673 return err;
674
675 hdr.msg = (struct io_uring_recvmsg_out) {
676 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
677 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
678 };
679
680 hdr.msg.payloadlen = err;
681 if (err > kmsg->payloadlen)
682 err = kmsg->payloadlen;
683
684 copy_len = sizeof(struct io_uring_recvmsg_out);
685 if (kmsg->msg.msg_namelen > kmsg->namelen)
686 copy_len += kmsg->namelen;
687 else
688 copy_len += kmsg->msg.msg_namelen;
689
690 /*
691 * "fromlen shall refer to the value before truncation.."
692 * 1003.1g
693 */
694 hdr.msg.namelen = kmsg->msg.msg_namelen;
695
696 /* ensure that there is no gap between hdr and sockaddr_storage */
697 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
698 sizeof(struct io_uring_recvmsg_out));
699 if (copy_to_user(io->buf, &hdr, copy_len)) {
700 *finished = true;
701 return -EFAULT;
702 }
703
704 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
705 kmsg->controllen + err;
706}
707
f9ead18c
JA		 708int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
709{
f2ccb5ae 710 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 711 struct io_async_msghdr iomsg, *kmsg;
712 struct socket *sock;
713 unsigned int cflags;
714 unsigned flags;
715 int ret, min_ret = 0;
716 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
9bb66906 717 bool mshot_finished = true;
f9ead18c
JA		 718
719 sock = sock_from_file(req->file);
720 if (unlikely(!sock))
721 return -ENOTSOCK;
722
723 if (req_has_async_data(req)) {
724 kmsg = req->async_data;
725 } else {
726 ret = io_recvmsg_copy_hdr(req, &iomsg);
727 if (ret)
728 return ret;
729 kmsg = &iomsg;
730 }
731
732 if (!(req->flags & REQ_F_POLLED) &&
733 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 734 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 735
9bb66906 736retry_multishot:
f9ead18c
JA		 737 if (io_do_buffer_select(req)) {
738 void __user *buf;
9bb66906 739 size_t len = sr->len;
f9ead18c 740
9bb66906 741 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 742 if (!buf)
743 return -ENOBUFS;
9bb66906
DY		 744
745 if (req->flags & REQ_F_APOLL_MULTISHOT) {
746 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
747 if (ret) {
748 io_kbuf_recycle(req, issue_flags);
749 return ret;
750 }
751 }
752
f9ead18c 753 kmsg->fast_iov[0].iov_base = buf;
9bb66906 754 kmsg->fast_iov[0].iov_len = len;
f9ead18c 755 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, 1,
9bb66906 756 len);
f9ead18c
JA		 757 }
758
759 flags = sr->msg_flags;
760 if (force_nonblock)
761 flags |= MSG_DONTWAIT;
762 if (flags & MSG_WAITALL)
763 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
764
765 kmsg->msg.msg_get_inq = 1;
9bb66906
DY		 766 if (req->flags & REQ_F_APOLL_MULTISHOT)
767 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
768 &mshot_finished);
769 else
770 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
771 kmsg->uaddr, flags);
772
f9ead18c 773 if (ret < min_ret) {
9bb66906
DY		 774 if (ret == -EAGAIN && force_nonblock) {
775 ret = io_setup_async_msg(req, kmsg, issue_flags);
776 if (ret == -EAGAIN && (req->flags & IO_APOLL_MULTI_POLLED) ==
777 IO_APOLL_MULTI_POLLED) {
778 io_kbuf_recycle(req, issue_flags);
779 return IOU_ISSUE_SKIP_COMPLETE;
780 }
781 return ret;
782 }
f9ead18c
JA		 783 if (ret > 0 && io_net_retry(sock, flags)) {
784 sr->done_io += ret;
785 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 786 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 787 }
95eafc74
PB		 788 if (ret == -ERESTARTSYS)
789 ret = -EINTR;
f9ead18c
JA		 790 req_set_fail(req);
791 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
792 req_set_fail(req);
793 }
794
d4e097da 795 if (ret > 0)
f9ead18c
JA		 796 ret += sr->done_io;
797 else if (sr->done_io)
798 ret = sr->done_io;
d4e097da
DY		 799 else
800 io_kbuf_recycle(req, issue_flags);
801
f9ead18c
JA		 802 cflags = io_put_kbuf(req, issue_flags);
803 if (kmsg->msg.msg_inq)
804 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 805
9bb66906
DY		 806 if (!io_recv_finish(req, &ret, cflags, mshot_finished))
807 goto retry_multishot;
808
809 if (mshot_finished) {
810 io_netmsg_recycle(req, issue_flags);
811 /* fast path, check for non-NULL to avoid function call */
812 if (kmsg->free_iov)
813 kfree(kmsg->free_iov);
814 req->flags &= ~REQ_F_NEED_CLEANUP;
815 }
816
817 return ret;
f9ead18c
JA		 818}
819
820int io_recv(struct io_kiocb *req, unsigned int issue_flags)
821{
f2ccb5ae 822 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
f9ead18c
JA		 823 struct msghdr msg;
824 struct socket *sock;
825 struct iovec iov;
826 unsigned int cflags;
827 unsigned flags;
828 int ret, min_ret = 0;
829 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
b3fdea6e 830 size_t len = sr->len;
f9ead18c
JA		 831
832 if (!(req->flags & REQ_F_POLLED) &&
833 (sr->flags & IORING_RECVSEND_POLL_FIRST))
834 return -EAGAIN;
835
836 sock = sock_from_file(req->file);
837 if (unlikely(!sock))
838 return -ENOTSOCK;
839
b3fdea6e 840retry_multishot:
f9ead18c
JA		 841 if (io_do_buffer_select(req)) {
842 void __user *buf;
843
b3fdea6e 844 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 845 if (!buf)
846 return -ENOBUFS;
847 sr->buf = buf;
848 }
849
b3fdea6e 850 ret = import_single_range(READ, sr->buf, len, &iov, &msg.msg_iter);
f9ead18c
JA		 851 if (unlikely(ret))
852 goto out_free;
853
854 msg.msg_name = NULL;
855 msg.msg_namelen = 0;
856 msg.msg_control = NULL;
857 msg.msg_get_inq = 1;
858 msg.msg_flags = 0;
859 msg.msg_controllen = 0;
860 msg.msg_iocb = NULL;
e02b6651 861 msg.msg_ubuf = NULL;
f9ead18c
JA		 862
863 flags = sr->msg_flags;
864 if (force_nonblock)
865 flags |= MSG_DONTWAIT;
866 if (flags & MSG_WAITALL)
867 min_ret = iov_iter_count(&msg.msg_iter);
868
869 ret = sock_recvmsg(sock, &msg, flags);
870 if (ret < min_ret) {
b3fdea6e
DY		 871 if (ret == -EAGAIN && force_nonblock) {
872 if ((req->flags & IO_APOLL_MULTI_POLLED) == IO_APOLL_MULTI_POLLED) {
873 io_kbuf_recycle(req, issue_flags);
874 return IOU_ISSUE_SKIP_COMPLETE;
875 }
876
f9ead18c 877 return -EAGAIN;
b3fdea6e 878 }
f9ead18c
JA		 879 if (ret > 0 && io_net_retry(sock, flags)) {
880 sr->len -= ret;
881 sr->buf += ret;
882 sr->done_io += ret;
883 req->flags |= REQ_F_PARTIAL_IO;
884 return -EAGAIN;
885 }
95eafc74
PB		 886 if (ret == -ERESTARTSYS)
887 ret = -EINTR;
f9ead18c
JA		 888 req_set_fail(req);
889 } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
890out_free:
891 req_set_fail(req);
892 }
893
d4e097da 894 if (ret > 0)
f9ead18c
JA		 895 ret += sr->done_io;
896 else if (sr->done_io)
897 ret = sr->done_io;
d4e097da
DY		 898 else
899 io_kbuf_recycle(req, issue_flags);
900
f9ead18c
JA		 901 cflags = io_put_kbuf(req, issue_flags);
902 if (msg.msg_inq)
903 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 904
9bb66906 905 if (!io_recv_finish(req, &ret, cflags, ret <= 0))
b3fdea6e
DY		 906 goto retry_multishot;
907
908 return ret;
f9ead18c
JA		 909}
910
b0e9b551 911void io_send_zc_cleanup(struct io_kiocb *req)
b48c312b 912{
ac9e5784 913 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
493108d9 914 struct io_async_msghdr *io;
b48c312b 915
493108d9
PB		 916 if (req_has_async_data(req)) {
917 io = req->async_data;
4c17a496
PB		 918 /* might be ->fast_iov if *msg_copy_hdr failed */
919 if (io->free_iov != io->fast_iov)
920 kfree(io->free_iov);
493108d9 921 }
a75155fa 922 if (zc->notif) {
a75155fa
PB		 923 io_notif_flush(zc->notif);
924 zc->notif = NULL;
925 }
b48c312b
PB		 926}
927
b0e9b551 928int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
06a5464b 929{
ac9e5784 930 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
10c7d33e 931 struct io_ring_ctx *ctx = req->ctx;
b48c312b 932 struct io_kiocb *notif;
06a5464b 933
493108d9 934 if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
b48c312b
PB		 935 return -EINVAL;
936 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
937 if (req->flags & REQ_F_CQE_SKIP)
06a5464b
PB		 938 return -EINVAL;
939
940 zc->flags = READ_ONCE(sqe->ioprio);
63809137 941 if (zc->flags & ~(IORING_RECVSEND_POLL_FIRST |
57f33224 942 IORING_RECVSEND_FIXED_BUF))
06a5464b 943 return -EINVAL;
e3366e02
PB		 944 notif = zc->notif = io_alloc_notif(ctx);
945 if (!notif)
946 return -ENOMEM;
947 notif->cqe.user_data = req->cqe.user_data;
948 notif->cqe.res = 0;
949 notif->cqe.flags = IORING_CQE_F_NOTIF;
950 req->flags |= REQ_F_NEED_CLEANUP;
10c7d33e
PB		 951 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
952 unsigned idx = READ_ONCE(sqe->buf_index);
953
954 if (unlikely(idx >= ctx->nr_user_bufs))
955 return -EFAULT;
956 idx = array_index_nospec(idx, ctx->nr_user_bufs);
957 req->imu = READ_ONCE(ctx->user_bufs[idx]);
e3366e02 958 io_req_set_rsrc_node(notif, ctx, 0);
10c7d33e 959 }
06a5464b 960
493108d9
PB		 961 if (req->opcode == IORING_OP_SEND_ZC) {
962 if (READ_ONCE(sqe->__pad3[0]))
963 return -EINVAL;
964 zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
965 zc->addr_len = READ_ONCE(sqe->addr_len);
966 } else {
967 if (unlikely(sqe->addr2 || sqe->file_index))
968 return -EINVAL;
969 if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
970 return -EINVAL;
971 }
972
06a5464b
PB		 973 zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
974 zc->len = READ_ONCE(sqe->len);
975 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
06a5464b
PB		 976 if (zc->msg_flags & MSG_DONTWAIT)
977 req->flags |= REQ_F_NOWAIT;
092aeedb 978
4a933e62 979 zc->done_io = 0;
092aeedb 980
06a5464b
PB		 981#ifdef CONFIG_COMPAT
982 if (req->ctx->compat)
983 zc->msg_flags |= MSG_CMSG_COMPAT;
984#endif
985 return 0;
986}
987
cd9021e8
PB		 988static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
989 struct iov_iter *from, size_t length)
990{
991 skb_zcopy_downgrade_managed(skb);
992 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
993}
994
3ff1a0d3
PB		 995static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
996 struct iov_iter *from, size_t length)
997{
998 struct skb_shared_info *shinfo = skb_shinfo(skb);
999 int frag = shinfo->nr_frags;
1000 int ret = 0;
1001 struct bvec_iter bi;
1002 ssize_t copied = 0;
1003 unsigned long truesize = 0;
1004
cd9021e8 1005 if (!frag)
3ff1a0d3 1006 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
cd9021e8 1007 else if (unlikely(!skb_zcopy_managed(skb)))
3ff1a0d3 1008 return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
3ff1a0d3
PB		 1009
1010 bi.bi_size = min(from->count, length);
1011 bi.bi_bvec_done = from->iov_offset;
1012 bi.bi_idx = 0;
1013
1014 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1015 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1016
1017 copied += v.bv_len;
1018 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1019 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1020 v.bv_offset, v.bv_len);
1021 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1022 }
1023 if (bi.bi_size)
1024 ret = -EMSGSIZE;
1025
1026 shinfo->nr_frags = frag;
1027 from->bvec += bi.bi_idx;
1028 from->nr_segs -= bi.bi_idx;
dfb58b17 1029 from->count -= copied;
3ff1a0d3
PB		 1030 from->iov_offset = bi.bi_bvec_done;
1031
1032 skb->data_len += copied;
1033 skb->len += copied;
1034 skb->truesize += truesize;
1035
1036 if (sk && sk->sk_type == SOCK_STREAM) {
1037 sk_wmem_queued_add(sk, truesize);
1038 if (!skb_zcopy_pure(skb))
1039 sk_mem_charge(sk, truesize);
1040 } else {
1041 refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1042 }
1043 return ret;
1044}
1045
b0e9b551 1046int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
06a5464b 1047{
6ae61b7a 1048 struct sockaddr_storage __address;
ac9e5784 1049 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
06a5464b
PB		 1050 struct msghdr msg;
1051 struct iovec iov;
1052 struct socket *sock;
6ae91ac9 1053 unsigned msg_flags;
06a5464b
PB		 1054 int ret, min_ret = 0;
1055
06a5464b
PB		 1056 sock = sock_from_file(req->file);
1057 if (unlikely(!sock))
1058 return -ENOTSOCK;
edf81438
PB		 1059 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1060 return -EOPNOTSUPP;
06a5464b 1061
06a5464b
PB		 1062 msg.msg_name = NULL;
1063 msg.msg_control = NULL;
1064 msg.msg_controllen = 0;
1065 msg.msg_namelen = 0;
1066
86dc8f23 1067 if (zc->addr) {
581711c4
PB		 1068 if (req_has_async_data(req)) {
1069 struct io_async_msghdr *io = req->async_data;
1070
6ae61b7a 1071 msg.msg_name = &io->addr;
581711c4
PB		 1072 } else {
1073 ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1074 if (unlikely(ret < 0))
1075 return ret;
1076 msg.msg_name = (struct sockaddr *)&__address;
581711c4 1077 }
86dc8f23
PB		 1078 msg.msg_namelen = zc->addr_len;
1079 }
1080
3c840053
PB		 1081 if (!(req->flags & REQ_F_POLLED) &&
1082 (zc->flags & IORING_RECVSEND_POLL_FIRST))
6ae61b7a 1083 return io_setup_async_addr(req, &__address, issue_flags);
3c840053 1084
10c7d33e
PB		 1085 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1086 ret = io_import_fixed(WRITE, &msg.msg_iter, req->imu,
1087 (u64)(uintptr_t)zc->buf, zc->len);
1088 if (unlikely(ret))
986e263d 1089 return ret;
cd9021e8 1090 msg.sg_from_iter = io_sg_from_iter;
10c7d33e
PB		 1091 } else {
1092 ret = import_single_range(WRITE, zc->buf, zc->len, &iov,
1093 &msg.msg_iter);
1094 if (unlikely(ret))
1095 return ret;
b48c312b 1096 ret = io_notif_account_mem(zc->notif, zc->len);
2e32ba56
PB		 1097 if (unlikely(ret))
1098 return ret;
cd9021e8 1099 msg.sg_from_iter = io_sg_from_iter_iovec;
10c7d33e 1100 }
06a5464b
PB		 1101
1102 msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1103 if (issue_flags & IO_URING_F_NONBLOCK)
1104 msg_flags |= MSG_DONTWAIT;
1105 if (msg_flags & MSG_WAITALL)
1106 min_ret = iov_iter_count(&msg.msg_iter);
1107
1108 msg.msg_flags = msg_flags;
b48c312b 1109 msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
06a5464b
PB		 1110 ret = sock_sendmsg(sock, &msg);
1111
1112 if (unlikely(ret < min_ret)) {
1113 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
6ae61b7a 1114 return io_setup_async_addr(req, &__address, issue_flags);
581711c4 1115
4a933e62
PB		 1116 if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1117 zc->len -= ret;
1118 zc->buf += ret;
1119 zc->done_io += ret;
1120 req->flags |= REQ_F_PARTIAL_IO;
6ae61b7a 1121 return io_setup_async_addr(req, &__address, issue_flags);
4a933e62
PB		 1122 }
1123 if (ret == -ERESTARTSYS)
1124 ret = -EINTR;
5a848b7c 1125 req_set_fail(req);
06a5464b
PB		 1126 }
1127
4a933e62
PB		 1128 if (ret >= 0)
1129 ret += zc->done_io;
1130 else if (zc->done_io)
1131 ret = zc->done_io;
b48c312b 1132
108893dd
PB		 1133 /*
1134 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1135 * flushing notif to io_send_zc_cleanup()
1136 */
1137 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1138 io_notif_flush(zc->notif);
1139 req->flags &= ~REQ_F_NEED_CLEANUP;
1140 }
6ae91ac9 1141 io_req_set_res(req, ret, IORING_CQE_F_MORE);
06a5464b
PB		 1142 return IOU_OK;
1143}
1144
493108d9
PB		 1145int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1146{
1147 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1148 struct io_async_msghdr iomsg, *kmsg;
1149 struct socket *sock;
6ae91ac9 1150 unsigned flags;
493108d9
PB		 1151 int ret, min_ret = 0;
1152
1153 sock = sock_from_file(req->file);
1154 if (unlikely(!sock))
1155 return -ENOTSOCK;
1156
1157 if (req_has_async_data(req)) {
1158 kmsg = req->async_data;
1159 } else {
1160 ret = io_sendmsg_copy_hdr(req, &iomsg);
1161 if (ret)
1162 return ret;
1163 kmsg = &iomsg;
1164 }
1165
1166 if (!(req->flags & REQ_F_POLLED) &&
1167 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1168 return io_setup_async_msg(req, kmsg, issue_flags);
1169
1170 flags = sr->msg_flags | MSG_ZEROCOPY;
1171 if (issue_flags & IO_URING_F_NONBLOCK)
1172 flags |= MSG_DONTWAIT;
1173 if (flags & MSG_WAITALL)
1174 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1175
1176 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1177 kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1178 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1179
1180 if (unlikely(ret < min_ret)) {
1181 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1182 return io_setup_async_msg(req, kmsg, issue_flags);
1183
1184 if (ret > 0 && io_net_retry(sock, flags)) {
1185 sr->done_io += ret;
1186 req->flags |= REQ_F_PARTIAL_IO;
1187 return io_setup_async_msg(req, kmsg, issue_flags);
1188 }
493108d9
PB		 1189 if (ret == -ERESTARTSYS)
1190 ret = -EINTR;
1191 req_set_fail(req);
1192 }
1193 /* fast path, check for non-NULL to avoid function call */
108893dd 1194 if (kmsg->free_iov) {
493108d9 1195 kfree(kmsg->free_iov);
108893dd
PB		 1196 kmsg->free_iov = NULL;
1197 }
493108d9
PB		 1198
1199 io_netmsg_recycle(req, issue_flags);
1200 if (ret >= 0)
1201 ret += sr->done_io;
1202 else if (sr->done_io)
1203 ret = sr->done_io;
1204
108893dd
PB		 1205 /*
1206 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1207 * flushing notif to io_send_zc_cleanup()
1208 */
1209 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1210 io_notif_flush(sr->notif);
1211 req->flags &= ~REQ_F_NEED_CLEANUP;
1212 }
6ae91ac9 1213 io_req_set_res(req, ret, IORING_CQE_F_MORE);
493108d9
PB		 1214 return IOU_OK;
1215}
1216
7e6b638e
PB		 1217void io_sendrecv_fail(struct io_kiocb *req)
1218{
1219 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
7e6b638e
PB		 1220
1221 if (req->flags & REQ_F_PARTIAL_IO)
6ae91ac9
PB		 1222 req->cqe.res = sr->done_io;
1223
c4c0009e 1224 if ((req->flags & REQ_F_NEED_CLEANUP) &&
6ae91ac9
PB		 1225 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1226 req->cqe.flags |= IORING_CQE_F_MORE;
5693bcce
PB		 1227}
1228
f9ead18c
JA		 1229int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1230{
f2ccb5ae 1231 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1232 unsigned flags;
1233
1234 if (sqe->len || sqe->buf_index)
1235 return -EINVAL;
1236
1237 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1238 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1239 accept->flags = READ_ONCE(sqe->accept_flags);
1240 accept->nofile = rlimit(RLIMIT_NOFILE);
1241 flags = READ_ONCE(sqe->ioprio);
1242 if (flags & ~IORING_ACCEPT_MULTISHOT)
1243 return -EINVAL;
1244
1245 accept->file_slot = READ_ONCE(sqe->file_index);
1246 if (accept->file_slot) {
1247 if (accept->flags & SOCK_CLOEXEC)
1248 return -EINVAL;
1249 if (flags & IORING_ACCEPT_MULTISHOT &&
1250 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1251 return -EINVAL;
1252 }
1253 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1254 return -EINVAL;
1255 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1256 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1257 if (flags & IORING_ACCEPT_MULTISHOT)
1258 req->flags |= REQ_F_APOLL_MULTISHOT;
1259 return 0;
1260}
1261
1262int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1263{
1264 struct io_ring_ctx *ctx = req->ctx;
f2ccb5ae 1265 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
f9ead18c
JA		 1266 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1267 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1268 bool fixed = !!accept->file_slot;
1269 struct file *file;
1270 int ret, fd;
1271
1272retry:
1273 if (!fixed) {
1274 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1275 if (unlikely(fd < 0))
1276 return fd;
1277 }
1278 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1279 accept->flags);
1280 if (IS_ERR(file)) {
1281 if (!fixed)
1282 put_unused_fd(fd);
1283 ret = PTR_ERR(file);
1284 if (ret == -EAGAIN && force_nonblock) {
1285 /*
1286 * if it's multishot and polled, we don't need to
1287 * return EAGAIN to arm the poll infra since it
1288 * has already been done
1289 */
1290 if ((req->flags & IO_APOLL_MULTI_POLLED) ==
1291 IO_APOLL_MULTI_POLLED)
1292 ret = IOU_ISSUE_SKIP_COMPLETE;
1293 return ret;
1294 }
1295 if (ret == -ERESTARTSYS)
1296 ret = -EINTR;
1297 req_set_fail(req);
1298 } else if (!fixed) {
1299 fd_install(fd, file);
1300 ret = fd;
1301 } else {
1302 ret = io_fixed_fd_install(req, issue_flags, file,
1303 accept->file_slot);
1304 }
1305
1306 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1307 io_req_set_res(req, ret, 0);
1308 return IOU_OK;
1309 }
f9ead18c 1310
cbd25748
DY		 1311 if (ret >= 0 &&
1312 io_post_aux_cqe(ctx, req->cqe.user_data, ret, IORING_CQE_F_MORE, false))
d245bca6 1313 goto retry;
cbd25748
DY		 1314
1315 io_req_set_res(req, ret, 0);
1316 if (req->flags & REQ_F_POLLED)
1317 return IOU_STOP_MULTISHOT;
1318 return IOU_OK;
f9ead18c
JA		 1319}
1320
1321int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1322{
f2ccb5ae 1323 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1324
1325 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1326 return -EINVAL;
1327
1328 sock->domain = READ_ONCE(sqe->fd);
1329 sock->type = READ_ONCE(sqe->off);
1330 sock->protocol = READ_ONCE(sqe->len);
1331 sock->file_slot = READ_ONCE(sqe->file_index);
1332 sock->nofile = rlimit(RLIMIT_NOFILE);
1333
1334 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1335 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1336 return -EINVAL;
1337 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1338 return -EINVAL;
1339 return 0;
1340}
1341
1342int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1343{
f2ccb5ae 1344 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
f9ead18c
JA		 1345 bool fixed = !!sock->file_slot;
1346 struct file *file;
1347 int ret, fd;
1348
1349 if (!fixed) {
1350 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1351 if (unlikely(fd < 0))
1352 return fd;
1353 }
1354 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1355 if (IS_ERR(file)) {
1356 if (!fixed)
1357 put_unused_fd(fd);
1358 ret = PTR_ERR(file);
1359 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1360 return -EAGAIN;
1361 if (ret == -ERESTARTSYS)
1362 ret = -EINTR;
1363 req_set_fail(req);
1364 } else if (!fixed) {
1365 fd_install(fd, file);
1366 ret = fd;
1367 } else {
1368 ret = io_fixed_fd_install(req, issue_flags, file,
1369 sock->file_slot);
1370 }
1371 io_req_set_res(req, ret, 0);
1372 return IOU_OK;
1373}
1374
1375int io_connect_prep_async(struct io_kiocb *req)
1376{
1377 struct io_async_connect *io = req->async_data;
f2ccb5ae 1378 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1379
1380 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1381}
1382
1383int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1384{
f2ccb5ae 1385 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1386
1387 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1388 return -EINVAL;
1389
1390 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1391 conn->addr_len = READ_ONCE(sqe->addr2);
3fb1bd68 1392 conn->in_progress = false;
f9ead18c
JA		 1393 return 0;
1394}
1395
1396int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1397{
f2ccb5ae 1398 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
f9ead18c
JA		 1399 struct io_async_connect __io, *io;
1400 unsigned file_flags;
1401 int ret;
1402 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1403
3fb1bd68
JA		 1404 if (connect->in_progress) {
1405 struct socket *socket;
1406
1407 ret = -ENOTSOCK;
1408 socket = sock_from_file(req->file);
1409 if (socket)
1410 ret = sock_error(socket->sk);
1411 goto out;
1412 }
1413
f9ead18c
JA		 1414 if (req_has_async_data(req)) {
1415 io = req->async_data;
1416 } else {
1417 ret = move_addr_to_kernel(connect->addr,
1418 connect->addr_len,
1419 &__io.address);
1420 if (ret)
1421 goto out;
1422 io = &__io;
1423 }
1424
1425 file_flags = force_nonblock ? O_NONBLOCK : 0;
1426
1427 ret = __sys_connect_file(req->file, &io->address,
1428 connect->addr_len, file_flags);
1429 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
3fb1bd68
JA		 1430 if (ret == -EINPROGRESS) {
1431 connect->in_progress = true;
1432 } else {
1433 if (req_has_async_data(req))
1434 return -EAGAIN;
1435 if (io_alloc_async_data(req)) {
1436 ret = -ENOMEM;
1437 goto out;
1438 }
1439 memcpy(req->async_data, &__io, sizeof(__io));
f9ead18c 1440 }
f9ead18c
JA		 1441 return -EAGAIN;
1442 }
1443 if (ret == -ERESTARTSYS)
1444 ret = -EINTR;
1445out:
1446 if (ret < 0)
1447 req_set_fail(req);
1448 io_req_set_res(req, ret, 0);
1449 return IOU_OK;
1450}
43e0bbbd
JA		 1451
1452void io_netmsg_cache_free(struct io_cache_entry *entry)
1453{
1454 kfree(container_of(entry, struct io_async_msghdr, cache));
1455}
f9ead18c 1456#endif
Linux 6.x block layer and io_uring tree(s)