projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
io_uring: Don't require reinitable percpu_ref
[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
JA		 16#include "net.h"
17
18#if defined(CONFIG_NET)
19struct io_shutdown {
20 struct file *file;
21 int how;
22};
23
24struct io_accept {
25 struct file *file;
26 struct sockaddr __user *addr;
27 int __user *addr_len;
28 int flags;
29 u32 file_slot;
30 unsigned long nofile;
31};
32
33struct io_socket {
34 struct file *file;
35 int domain;
36 int type;
37 int protocol;
38 int flags;
39 u32 file_slot;
40 unsigned long nofile;
41};
42
43struct io_connect {
44 struct file *file;
45 struct sockaddr __user *addr;
46 int addr_len;
47};
48
49struct io_sr_msg {
50 struct file *file;
51 union {
52 struct compat_msghdr __user *umsg_compat;
53 struct user_msghdr __user *umsg;
54 void __user *buf;
55 };
56 int msg_flags;
57 size_t len;
58 size_t done_io;
59 unsigned int flags;
60};
61
62#define IO_APOLL_MULTI_POLLED (REQ_F_APOLL_MULTISHOT | REQ_F_POLLED)
63
64int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
65{
66 struct io_shutdown *shutdown = io_kiocb_to_cmd(req);
67
68 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
69 sqe->buf_index || sqe->splice_fd_in))
70 return -EINVAL;
71
72 shutdown->how = READ_ONCE(sqe->len);
73 return 0;
74}
75
76int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
77{
78 struct io_shutdown *shutdown = io_kiocb_to_cmd(req);
79 struct socket *sock;
80 int ret;
81
82 if (issue_flags & IO_URING_F_NONBLOCK)
83 return -EAGAIN;
84
85 sock = sock_from_file(req->file);
86 if (unlikely(!sock))
87 return -ENOTSOCK;
88
89 ret = __sys_shutdown_sock(sock, shutdown->how);
90 io_req_set_res(req, ret, 0);
91 return IOU_OK;
92}
93
94static bool io_net_retry(struct socket *sock, int flags)
95{
96 if (!(flags & MSG_WAITALL))
97 return false;
98 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
99}
100
43e0bbbd
JA		 101static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
102{
103 struct io_async_msghdr *hdr = req->async_data;
104
105 if (!hdr || issue_flags & IO_URING_F_UNLOCKED)
106 return;
107
108 /* Let normal cleanup path reap it if we fail adding to the cache */
109 if (io_alloc_cache_put(&req->ctx->netmsg_cache, &hdr->cache)) {
110 req->async_data = NULL;
111 req->flags &= ~REQ_F_ASYNC_DATA;
112 }
113}
114
115static struct io_async_msghdr *io_recvmsg_alloc_async(struct io_kiocb *req,
116 unsigned int issue_flags)
117{
118 struct io_ring_ctx *ctx = req->ctx;
119 struct io_cache_entry *entry;
120
121 if (!(issue_flags & IO_URING_F_UNLOCKED) &&
122 (entry = io_alloc_cache_get(&ctx->netmsg_cache)) != NULL) {
123 struct io_async_msghdr *hdr;
124
125 hdr = container_of(entry, struct io_async_msghdr, cache);
126 req->flags |= REQ_F_ASYNC_DATA;
127 req->async_data = hdr;
128 return hdr;
129 }
130
131 if (!io_alloc_async_data(req))
132 return req->async_data;
133
134 return NULL;
135}
136
f9ead18c 137static int io_setup_async_msg(struct io_kiocb *req,
43e0bbbd
JA		 138 struct io_async_msghdr *kmsg,
139 unsigned int issue_flags)
f9ead18c
JA		 140{
141 struct io_async_msghdr *async_msg = req->async_data;
142
143 if (async_msg)
144 return -EAGAIN;
43e0bbbd
JA		 145 async_msg = io_recvmsg_alloc_async(req, issue_flags);
146 if (!async_msg) {
f9ead18c
JA		 147 kfree(kmsg->free_iov);
148 return -ENOMEM;
149 }
f9ead18c
JA		 150 req->flags |= REQ_F_NEED_CLEANUP;
151 memcpy(async_msg, kmsg, sizeof(*kmsg));
152 async_msg->msg.msg_name = &async_msg->addr;
153 /* if were using fast_iov, set it to the new one */
154 if (!async_msg->free_iov)
155 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
156
157 return -EAGAIN;
158}
159
160static int io_sendmsg_copy_hdr(struct io_kiocb *req,
161 struct io_async_msghdr *iomsg)
162{
163 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
164
165 iomsg->msg.msg_name = &iomsg->addr;
166 iomsg->free_iov = iomsg->fast_iov;
167 return sendmsg_copy_msghdr(&iomsg->msg, sr->umsg, sr->msg_flags,
168 &iomsg->free_iov);
169}
170
171int io_sendmsg_prep_async(struct io_kiocb *req)
172{
173 int ret;
174
175 ret = io_sendmsg_copy_hdr(req, req->async_data);
176 if (!ret)
177 req->flags |= REQ_F_NEED_CLEANUP;
178 return ret;
179}
180
181void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
182{
183 struct io_async_msghdr *io = req->async_data;
184
185 kfree(io->free_iov);
186}
187
188int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
189{
190 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
191
192 if (unlikely(sqe->file_index || sqe->addr2))
193 return -EINVAL;
194
195 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
196 sr->len = READ_ONCE(sqe->len);
197 sr->flags = READ_ONCE(sqe->ioprio);
198 if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
199 return -EINVAL;
200 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
201 if (sr->msg_flags & MSG_DONTWAIT)
202 req->flags |= REQ_F_NOWAIT;
203
204#ifdef CONFIG_COMPAT
205 if (req->ctx->compat)
206 sr->msg_flags |= MSG_CMSG_COMPAT;
207#endif
208 sr->done_io = 0;
209 return 0;
210}
211
212int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
213{
214 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
215 struct io_async_msghdr iomsg, *kmsg;
216 struct socket *sock;
217 unsigned flags;
218 int min_ret = 0;
219 int ret;
220
221 sock = sock_from_file(req->file);
222 if (unlikely(!sock))
223 return -ENOTSOCK;
224
225 if (req_has_async_data(req)) {
226 kmsg = req->async_data;
227 } else {
228 ret = io_sendmsg_copy_hdr(req, &iomsg);
229 if (ret)
230 return ret;
231 kmsg = &iomsg;
232 }
233
234 if (!(req->flags & REQ_F_POLLED) &&
235 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 236 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 237
238 flags = sr->msg_flags;
239 if (issue_flags & IO_URING_F_NONBLOCK)
240 flags |= MSG_DONTWAIT;
241 if (flags & MSG_WAITALL)
242 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
243
244 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
245
246 if (ret < min_ret) {
247 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
43e0bbbd 248 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 249 if (ret == -ERESTARTSYS)
250 ret = -EINTR;
251 if (ret > 0 && io_net_retry(sock, flags)) {
252 sr->done_io += ret;
253 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 254 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 255 }
256 req_set_fail(req);
257 }
258 /* fast path, check for non-NULL to avoid function call */
259 if (kmsg->free_iov)
260 kfree(kmsg->free_iov);
261 req->flags &= ~REQ_F_NEED_CLEANUP;
43e0bbbd 262 io_netmsg_recycle(req, issue_flags);
f9ead18c
JA		 263 if (ret >= 0)
264 ret += sr->done_io;
265 else if (sr->done_io)
266 ret = sr->done_io;
267 io_req_set_res(req, ret, 0);
268 return IOU_OK;
269}
270
271int io_send(struct io_kiocb *req, unsigned int issue_flags)
272{
273 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
274 struct msghdr msg;
275 struct iovec iov;
276 struct socket *sock;
277 unsigned flags;
278 int min_ret = 0;
279 int ret;
280
281 if (!(req->flags & REQ_F_POLLED) &&
282 (sr->flags & IORING_RECVSEND_POLL_FIRST))
283 return -EAGAIN;
284
285 sock = sock_from_file(req->file);
286 if (unlikely(!sock))
287 return -ENOTSOCK;
288
289 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
290 if (unlikely(ret))
291 return ret;
292
293 msg.msg_name = NULL;
294 msg.msg_control = NULL;
295 msg.msg_controllen = 0;
296 msg.msg_namelen = 0;
297
298 flags = sr->msg_flags;
299 if (issue_flags & IO_URING_F_NONBLOCK)
300 flags |= MSG_DONTWAIT;
301 if (flags & MSG_WAITALL)
302 min_ret = iov_iter_count(&msg.msg_iter);
303
304 msg.msg_flags = flags;
305 ret = sock_sendmsg(sock, &msg);
306 if (ret < min_ret) {
307 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
308 return -EAGAIN;
309 if (ret == -ERESTARTSYS)
310 ret = -EINTR;
311 if (ret > 0 && io_net_retry(sock, flags)) {
312 sr->len -= ret;
313 sr->buf += ret;
314 sr->done_io += ret;
315 req->flags |= REQ_F_PARTIAL_IO;
316 return -EAGAIN;
317 }
318 req_set_fail(req);
319 }
320 if (ret >= 0)
321 ret += sr->done_io;
322 else if (sr->done_io)
323 ret = sr->done_io;
324 io_req_set_res(req, ret, 0);
325 return IOU_OK;
326}
327
9bb66906
DY		 328static bool io_recvmsg_multishot_overflow(struct io_async_msghdr *iomsg)
329{
9b0fc3c0 330 int hdr;
9bb66906 331
9b0fc3c0 332 if (iomsg->namelen < 0)
9bb66906 333 return true;
9b0fc3c0
DY		 334 if (check_add_overflow((int)sizeof(struct io_uring_recvmsg_out),
335 iomsg->namelen, &hdr))
9bb66906 336 return true;
9b0fc3c0 337 if (check_add_overflow(hdr, (int)iomsg->controllen, &hdr))
9bb66906
DY		 338 return true;
339
340 return false;
341}
342
f9ead18c
JA		 343static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
344 struct io_async_msghdr *iomsg)
345{
346 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
7fa875b8 347 struct user_msghdr msg;
f9ead18c
JA		 348 int ret;
349
7fa875b8
DY		 350 if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
351 return -EFAULT;
352
353 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
f9ead18c
JA		 354 if (ret)
355 return ret;
356
357 if (req->flags & REQ_F_BUFFER_SELECT) {
7fa875b8 358 if (msg.msg_iovlen == 0) {
5702196e
DY		 359 sr->len = iomsg->fast_iov[0].iov_len = 0;
360 iomsg->fast_iov[0].iov_base = NULL;
361 iomsg->free_iov = NULL;
7fa875b8 362 } else if (msg.msg_iovlen > 1) {
f9ead18c 363 return -EINVAL;
5702196e 364 } else {
7fa875b8 365 if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
5702196e
DY		 366 return -EFAULT;
367 sr->len = iomsg->fast_iov[0].iov_len;
368 iomsg->free_iov = NULL;
369 }
9bb66906
DY		 370
371 if (req->flags & REQ_F_APOLL_MULTISHOT) {
372 iomsg->namelen = msg.msg_namelen;
373 iomsg->controllen = msg.msg_controllen;
374 if (io_recvmsg_multishot_overflow(iomsg))
375 return -EOVERFLOW;
376 }
f9ead18c
JA		 377 } else {
378 iomsg->free_iov = iomsg->fast_iov;
7fa875b8 379 ret = __import_iovec(READ, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
f9ead18c
JA		 380 &iomsg->free_iov, &iomsg->msg.msg_iter,
381 false);
382 if (ret > 0)
383 ret = 0;
384 }
385
386 return ret;
387}
388
389#ifdef CONFIG_COMPAT
390static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
391 struct io_async_msghdr *iomsg)
392{
393 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
72c531f8 394 struct compat_msghdr msg;
f9ead18c 395 struct compat_iovec __user *uiov;
f9ead18c
JA		 396 int ret;
397
72c531f8
DY		 398 if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
399 return -EFAULT;
400
401 ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr);
f9ead18c
JA		 402 if (ret)
403 return ret;
404
72c531f8 405 uiov = compat_ptr(msg.msg_iov);
f9ead18c
JA		 406 if (req->flags & REQ_F_BUFFER_SELECT) {
407 compat_ssize_t clen;
408
72c531f8 409 if (msg.msg_iovlen == 0) {
6d2f75a0
DY		 410 sr->len = 0;
411 iomsg->free_iov = NULL;
72c531f8 412 } else if (msg.msg_iovlen > 1) {
f9ead18c 413 return -EINVAL;
6d2f75a0
DY		 414 } else {
415 if (!access_ok(uiov, sizeof(*uiov)))
416 return -EFAULT;
417 if (__get_user(clen, &uiov->iov_len))
418 return -EFAULT;
419 if (clen < 0)
420 return -EINVAL;
421 sr->len = clen;
422 iomsg->free_iov = NULL;
423 }
9bb66906
DY		 424
425 if (req->flags & REQ_F_APOLL_MULTISHOT) {
426 iomsg->namelen = msg.msg_namelen;
427 iomsg->controllen = msg.msg_controllen;
428 if (io_recvmsg_multishot_overflow(iomsg))
429 return -EOVERFLOW;
430 }
f9ead18c
JA		 431 } else {
432 iomsg->free_iov = iomsg->fast_iov;
72c531f8 433 ret = __import_iovec(READ, (struct iovec __user *)uiov, msg.msg_iovlen,
f9ead18c
JA		 434 UIO_FASTIOV, &iomsg->free_iov,
435 &iomsg->msg.msg_iter, true);
436 if (ret < 0)
437 return ret;
438 }
439
440 return 0;
441}
442#endif
443
444static int io_recvmsg_copy_hdr(struct io_kiocb *req,
445 struct io_async_msghdr *iomsg)
446{
447 iomsg->msg.msg_name = &iomsg->addr;
448
449#ifdef CONFIG_COMPAT
450 if (req->ctx->compat)
451 return __io_compat_recvmsg_copy_hdr(req, iomsg);
452#endif
453
454 return __io_recvmsg_copy_hdr(req, iomsg);
455}
456
457int io_recvmsg_prep_async(struct io_kiocb *req)
458{
459 int ret;
460
461 ret = io_recvmsg_copy_hdr(req, req->async_data);
462 if (!ret)
463 req->flags |= REQ_F_NEED_CLEANUP;
464 return ret;
465}
466
b3fdea6e
DY		 467#define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
468
f9ead18c
JA		 469int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
470{
471 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
472
473 if (unlikely(sqe->file_index || sqe->addr2))
474 return -EINVAL;
475
476 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
477 sr->len = READ_ONCE(sqe->len);
478 sr->flags = READ_ONCE(sqe->ioprio);
b3fdea6e 479 if (sr->flags & ~(RECVMSG_FLAGS))
f9ead18c
JA		 480 return -EINVAL;
481 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
482 if (sr->msg_flags & MSG_DONTWAIT)
483 req->flags |= REQ_F_NOWAIT;
484 if (sr->msg_flags & MSG_ERRQUEUE)
485 req->flags |= REQ_F_CLEAR_POLLIN;
b3fdea6e
DY		 486 if (sr->flags & IORING_RECV_MULTISHOT) {
487 if (!(req->flags & REQ_F_BUFFER_SELECT))
488 return -EINVAL;
489 if (sr->msg_flags & MSG_WAITALL)
490 return -EINVAL;
491 if (req->opcode == IORING_OP_RECV && sr->len)
492 return -EINVAL;
493 req->flags |= REQ_F_APOLL_MULTISHOT;
494 }
f9ead18c
JA		 495
496#ifdef CONFIG_COMPAT
497 if (req->ctx->compat)
498 sr->msg_flags |= MSG_CMSG_COMPAT;
499#endif
500 sr->done_io = 0;
501 return 0;
502}
503
b3fdea6e
DY		 504static inline void io_recv_prep_retry(struct io_kiocb *req)
505{
506 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
507
508 sr->done_io = 0;
509 sr->len = 0; /* get from the provided buffer */
510}
511
512/*
9bb66906 513 * Finishes io_recv and io_recvmsg.
b3fdea6e
DY		 514 *
515 * Returns true if it is actually finished, or false if it should run
516 * again (for multishot).
517 */
9bb66906
DY		 518static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
519 unsigned int cflags, bool mshot_finished)
b3fdea6e
DY		 520{
521 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
522 io_req_set_res(req, *ret, cflags);
523 *ret = IOU_OK;
524 return true;
525 }
526
9bb66906 527 if (!mshot_finished) {
b3fdea6e
DY		 528 if (io_post_aux_cqe(req->ctx, req->cqe.user_data, *ret,
529 cflags | IORING_CQE_F_MORE, false)) {
530 io_recv_prep_retry(req);
531 return false;
532 }
533 /*
534 * Otherwise stop multishot but use the current result.
535 * Probably will end up going into overflow, but this means
536 * we cannot trust the ordering anymore
537 */
538 }
539
540 io_req_set_res(req, *ret, cflags);
541
542 if (req->flags & REQ_F_POLLED)
543 *ret = IOU_STOP_MULTISHOT;
e2df2ccb
DY		 544 else
545 *ret = IOU_OK;
b3fdea6e
DY		 546 return true;
547}
548
9bb66906
DY		 549static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
550 struct io_sr_msg *sr, void __user **buf,
551 size_t *len)
552{
553 unsigned long ubuf = (unsigned long) *buf;
554 unsigned long hdr;
555
556 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
557 kmsg->controllen;
558 if (*len < hdr)
559 return -EFAULT;
560
561 if (kmsg->controllen) {
562 unsigned long control = ubuf + hdr - kmsg->controllen;
563
564 kmsg->msg.msg_control_user = (void *) control;
565 kmsg->msg.msg_controllen = kmsg->controllen;
566 }
567
568 sr->buf = *buf; /* stash for later copy */
569 *buf = (void *) (ubuf + hdr);
570 kmsg->payloadlen = *len = *len - hdr;
571 return 0;
572}
573
574struct io_recvmsg_multishot_hdr {
575 struct io_uring_recvmsg_out msg;
576 struct sockaddr_storage addr;
577};
578
579static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
580 struct io_async_msghdr *kmsg,
581 unsigned int flags, bool *finished)
582{
583 int err;
584 int copy_len;
585 struct io_recvmsg_multishot_hdr hdr;
586
587 if (kmsg->namelen)
588 kmsg->msg.msg_name = &hdr.addr;
589 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
590 kmsg->msg.msg_namelen = 0;
591
592 if (sock->file->f_flags & O_NONBLOCK)
593 flags |= MSG_DONTWAIT;
594
595 err = sock_recvmsg(sock, &kmsg->msg, flags);
596 *finished = err <= 0;
597 if (err < 0)
598 return err;
599
600 hdr.msg = (struct io_uring_recvmsg_out) {
601 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
602 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
603 };
604
605 hdr.msg.payloadlen = err;
606 if (err > kmsg->payloadlen)
607 err = kmsg->payloadlen;
608
609 copy_len = sizeof(struct io_uring_recvmsg_out);
610 if (kmsg->msg.msg_namelen > kmsg->namelen)
611 copy_len += kmsg->namelen;
612 else
613 copy_len += kmsg->msg.msg_namelen;
614
615 /*
616 * "fromlen shall refer to the value before truncation.."
617 * 1003.1g
618 */
619 hdr.msg.namelen = kmsg->msg.msg_namelen;
620
621 /* ensure that there is no gap between hdr and sockaddr_storage */
622 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
623 sizeof(struct io_uring_recvmsg_out));
624 if (copy_to_user(io->buf, &hdr, copy_len)) {
625 *finished = true;
626 return -EFAULT;
627 }
628
629 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
630 kmsg->controllen + err;
631}
632
f9ead18c
JA		 633int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
634{
635 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
636 struct io_async_msghdr iomsg, *kmsg;
637 struct socket *sock;
638 unsigned int cflags;
639 unsigned flags;
640 int ret, min_ret = 0;
641 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
9bb66906 642 bool mshot_finished = true;
f9ead18c
JA		 643
644 sock = sock_from_file(req->file);
645 if (unlikely(!sock))
646 return -ENOTSOCK;
647
648 if (req_has_async_data(req)) {
649 kmsg = req->async_data;
650 } else {
651 ret = io_recvmsg_copy_hdr(req, &iomsg);
652 if (ret)
653 return ret;
654 kmsg = &iomsg;
655 }
656
657 if (!(req->flags & REQ_F_POLLED) &&
658 (sr->flags & IORING_RECVSEND_POLL_FIRST))
43e0bbbd 659 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c 660
9bb66906 661retry_multishot:
f9ead18c
JA		 662 if (io_do_buffer_select(req)) {
663 void __user *buf;
9bb66906 664 size_t len = sr->len;
f9ead18c 665
9bb66906 666 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 667 if (!buf)
668 return -ENOBUFS;
9bb66906
DY		 669
670 if (req->flags & REQ_F_APOLL_MULTISHOT) {
671 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
672 if (ret) {
673 io_kbuf_recycle(req, issue_flags);
674 return ret;
675 }
676 }
677
f9ead18c 678 kmsg->fast_iov[0].iov_base = buf;
9bb66906 679 kmsg->fast_iov[0].iov_len = len;
f9ead18c 680 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, 1,
9bb66906 681 len);
f9ead18c
JA		 682 }
683
684 flags = sr->msg_flags;
685 if (force_nonblock)
686 flags |= MSG_DONTWAIT;
687 if (flags & MSG_WAITALL)
688 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
689
690 kmsg->msg.msg_get_inq = 1;
9bb66906
DY		 691 if (req->flags & REQ_F_APOLL_MULTISHOT)
692 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
693 &mshot_finished);
694 else
695 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
696 kmsg->uaddr, flags);
697
f9ead18c 698 if (ret < min_ret) {
9bb66906
DY		 699 if (ret == -EAGAIN && force_nonblock) {
700 ret = io_setup_async_msg(req, kmsg, issue_flags);
701 if (ret == -EAGAIN && (req->flags & IO_APOLL_MULTI_POLLED) ==
702 IO_APOLL_MULTI_POLLED) {
703 io_kbuf_recycle(req, issue_flags);
704 return IOU_ISSUE_SKIP_COMPLETE;
705 }
706 return ret;
707 }
f9ead18c
JA		 708 if (ret == -ERESTARTSYS)
709 ret = -EINTR;
710 if (ret > 0 && io_net_retry(sock, flags)) {
711 sr->done_io += ret;
712 req->flags |= REQ_F_PARTIAL_IO;
43e0bbbd 713 return io_setup_async_msg(req, kmsg, issue_flags);
f9ead18c
JA		 714 }
715 req_set_fail(req);
716 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
717 req_set_fail(req);
718 }
719
d4e097da 720 if (ret > 0)
f9ead18c
JA		 721 ret += sr->done_io;
722 else if (sr->done_io)
723 ret = sr->done_io;
d4e097da
DY		 724 else
725 io_kbuf_recycle(req, issue_flags);
726
f9ead18c
JA		 727 cflags = io_put_kbuf(req, issue_flags);
728 if (kmsg->msg.msg_inq)
729 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 730
9bb66906
DY		 731 if (!io_recv_finish(req, &ret, cflags, mshot_finished))
732 goto retry_multishot;
733
734 if (mshot_finished) {
735 io_netmsg_recycle(req, issue_flags);
736 /* fast path, check for non-NULL to avoid function call */
737 if (kmsg->free_iov)
738 kfree(kmsg->free_iov);
739 req->flags &= ~REQ_F_NEED_CLEANUP;
740 }
741
742 return ret;
f9ead18c
JA		 743}
744
745int io_recv(struct io_kiocb *req, unsigned int issue_flags)
746{
747 struct io_sr_msg *sr = io_kiocb_to_cmd(req);
748 struct msghdr msg;
749 struct socket *sock;
750 struct iovec iov;
751 unsigned int cflags;
752 unsigned flags;
753 int ret, min_ret = 0;
754 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
b3fdea6e 755 size_t len = sr->len;
f9ead18c
JA		 756
757 if (!(req->flags & REQ_F_POLLED) &&
758 (sr->flags & IORING_RECVSEND_POLL_FIRST))
759 return -EAGAIN;
760
761 sock = sock_from_file(req->file);
762 if (unlikely(!sock))
763 return -ENOTSOCK;
764
b3fdea6e 765retry_multishot:
f9ead18c
JA		 766 if (io_do_buffer_select(req)) {
767 void __user *buf;
768
b3fdea6e 769 buf = io_buffer_select(req, &len, issue_flags);
f9ead18c
JA		 770 if (!buf)
771 return -ENOBUFS;
772 sr->buf = buf;
773 }
774
b3fdea6e 775 ret = import_single_range(READ, sr->buf, len, &iov, &msg.msg_iter);
f9ead18c
JA		 776 if (unlikely(ret))
777 goto out_free;
778
779 msg.msg_name = NULL;
780 msg.msg_namelen = 0;
781 msg.msg_control = NULL;
782 msg.msg_get_inq = 1;
783 msg.msg_flags = 0;
784 msg.msg_controllen = 0;
785 msg.msg_iocb = NULL;
786
787 flags = sr->msg_flags;
788 if (force_nonblock)
789 flags |= MSG_DONTWAIT;
790 if (flags & MSG_WAITALL)
791 min_ret = iov_iter_count(&msg.msg_iter);
792
793 ret = sock_recvmsg(sock, &msg, flags);
794 if (ret < min_ret) {
b3fdea6e
DY		 795 if (ret == -EAGAIN && force_nonblock) {
796 if ((req->flags & IO_APOLL_MULTI_POLLED) == IO_APOLL_MULTI_POLLED) {
797 io_kbuf_recycle(req, issue_flags);
798 return IOU_ISSUE_SKIP_COMPLETE;
799 }
800
f9ead18c 801 return -EAGAIN;
b3fdea6e 802 }
f9ead18c
JA		 803 if (ret == -ERESTARTSYS)
804 ret = -EINTR;
805 if (ret > 0 && io_net_retry(sock, flags)) {
806 sr->len -= ret;
807 sr->buf += ret;
808 sr->done_io += ret;
809 req->flags |= REQ_F_PARTIAL_IO;
810 return -EAGAIN;
811 }
812 req_set_fail(req);
813 } else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
814out_free:
815 req_set_fail(req);
816 }
817
d4e097da 818 if (ret > 0)
f9ead18c
JA		 819 ret += sr->done_io;
820 else if (sr->done_io)
821 ret = sr->done_io;
d4e097da
DY		 822 else
823 io_kbuf_recycle(req, issue_flags);
824
f9ead18c
JA		 825 cflags = io_put_kbuf(req, issue_flags);
826 if (msg.msg_inq)
827 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
b3fdea6e 828
9bb66906 829 if (!io_recv_finish(req, &ret, cflags, ret <= 0))
b3fdea6e
DY		 830 goto retry_multishot;
831
832 return ret;
f9ead18c
JA		 833}
834
835int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
836{
837 struct io_accept *accept = io_kiocb_to_cmd(req);
838 unsigned flags;
839
840 if (sqe->len || sqe->buf_index)
841 return -EINVAL;
842
843 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
844 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
845 accept->flags = READ_ONCE(sqe->accept_flags);
846 accept->nofile = rlimit(RLIMIT_NOFILE);
847 flags = READ_ONCE(sqe->ioprio);
848 if (flags & ~IORING_ACCEPT_MULTISHOT)
849 return -EINVAL;
850
851 accept->file_slot = READ_ONCE(sqe->file_index);
852 if (accept->file_slot) {
853 if (accept->flags & SOCK_CLOEXEC)
854 return -EINVAL;
855 if (flags & IORING_ACCEPT_MULTISHOT &&
856 accept->file_slot != IORING_FILE_INDEX_ALLOC)
857 return -EINVAL;
858 }
859 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
860 return -EINVAL;
861 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
862 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
863 if (flags & IORING_ACCEPT_MULTISHOT)
864 req->flags |= REQ_F_APOLL_MULTISHOT;
865 return 0;
866}
867
868int io_accept(struct io_kiocb *req, unsigned int issue_flags)
869{
870 struct io_ring_ctx *ctx = req->ctx;
871 struct io_accept *accept = io_kiocb_to_cmd(req);
872 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
873 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
874 bool fixed = !!accept->file_slot;
875 struct file *file;
876 int ret, fd;
877
878retry:
879 if (!fixed) {
880 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
881 if (unlikely(fd < 0))
882 return fd;
883 }
884 file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
885 accept->flags);
886 if (IS_ERR(file)) {
887 if (!fixed)
888 put_unused_fd(fd);
889 ret = PTR_ERR(file);
890 if (ret == -EAGAIN && force_nonblock) {
891 /*
892 * if it's multishot and polled, we don't need to
893 * return EAGAIN to arm the poll infra since it
894 * has already been done
895 */
896 if ((req->flags & IO_APOLL_MULTI_POLLED) ==
897 IO_APOLL_MULTI_POLLED)
898 ret = IOU_ISSUE_SKIP_COMPLETE;
899 return ret;
900 }
901 if (ret == -ERESTARTSYS)
902 ret = -EINTR;
903 req_set_fail(req);
904 } else if (!fixed) {
905 fd_install(fd, file);
906 ret = fd;
907 } else {
908 ret = io_fixed_fd_install(req, issue_flags, file,
909 accept->file_slot);
910 }
911
912 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
913 io_req_set_res(req, ret, 0);
914 return IOU_OK;
915 }
f9ead18c 916
cbd25748
DY		 917 if (ret >= 0 &&
918 io_post_aux_cqe(ctx, req->cqe.user_data, ret, IORING_CQE_F_MORE, false))
d245bca6 919 goto retry;
cbd25748
DY		 920
921 io_req_set_res(req, ret, 0);
922 if (req->flags & REQ_F_POLLED)
923 return IOU_STOP_MULTISHOT;
924 return IOU_OK;
f9ead18c
JA		 925}
926
927int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
928{
929 struct io_socket *sock = io_kiocb_to_cmd(req);
930
931 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
932 return -EINVAL;
933
934 sock->domain = READ_ONCE(sqe->fd);
935 sock->type = READ_ONCE(sqe->off);
936 sock->protocol = READ_ONCE(sqe->len);
937 sock->file_slot = READ_ONCE(sqe->file_index);
938 sock->nofile = rlimit(RLIMIT_NOFILE);
939
940 sock->flags = sock->type & ~SOCK_TYPE_MASK;
941 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
942 return -EINVAL;
943 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
944 return -EINVAL;
945 return 0;
946}
947
948int io_socket(struct io_kiocb *req, unsigned int issue_flags)
949{
950 struct io_socket *sock = io_kiocb_to_cmd(req);
951 bool fixed = !!sock->file_slot;
952 struct file *file;
953 int ret, fd;
954
955 if (!fixed) {
956 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
957 if (unlikely(fd < 0))
958 return fd;
959 }
960 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
961 if (IS_ERR(file)) {
962 if (!fixed)
963 put_unused_fd(fd);
964 ret = PTR_ERR(file);
965 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
966 return -EAGAIN;
967 if (ret == -ERESTARTSYS)
968 ret = -EINTR;
969 req_set_fail(req);
970 } else if (!fixed) {
971 fd_install(fd, file);
972 ret = fd;
973 } else {
974 ret = io_fixed_fd_install(req, issue_flags, file,
975 sock->file_slot);
976 }
977 io_req_set_res(req, ret, 0);
978 return IOU_OK;
979}
980
981int io_connect_prep_async(struct io_kiocb *req)
982{
983 struct io_async_connect *io = req->async_data;
984 struct io_connect *conn = io_kiocb_to_cmd(req);
985
986 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
987}
988
989int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
990{
991 struct io_connect *conn = io_kiocb_to_cmd(req);
992
993 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
994 return -EINVAL;
995
996 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
997 conn->addr_len = READ_ONCE(sqe->addr2);
998 return 0;
999}
1000
1001int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1002{
1003 struct io_connect *connect = io_kiocb_to_cmd(req);
1004 struct io_async_connect __io, *io;
1005 unsigned file_flags;
1006 int ret;
1007 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1008
1009 if (req_has_async_data(req)) {
1010 io = req->async_data;
1011 } else {
1012 ret = move_addr_to_kernel(connect->addr,
1013 connect->addr_len,
1014 &__io.address);
1015 if (ret)
1016 goto out;
1017 io = &__io;
1018 }
1019
1020 file_flags = force_nonblock ? O_NONBLOCK : 0;
1021
1022 ret = __sys_connect_file(req->file, &io->address,
1023 connect->addr_len, file_flags);
1024 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
1025 if (req_has_async_data(req))
1026 return -EAGAIN;
1027 if (io_alloc_async_data(req)) {
1028 ret = -ENOMEM;
1029 goto out;
1030 }
1031 memcpy(req->async_data, &__io, sizeof(__io));
1032 return -EAGAIN;
1033 }
1034 if (ret == -ERESTARTSYS)
1035 ret = -EINTR;
1036out:
1037 if (ret < 0)
1038 req_set_fail(req);
1039 io_req_set_res(req, ret, 0);
1040 return IOU_OK;
1041}
43e0bbbd
JA		 1042
1043void io_netmsg_cache_free(struct io_cache_entry *entry)
1044{
1045 kfree(container_of(entry, struct io_async_msghdr, cache));
1046}
f9ead18c 1047#endif
Linux 6.x block layer and io_uring tree(s)