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
13 #include "io_uring.h"
14 #include "kbuf.h"
15 #include "alloc_cache.h"
16 #include "net.h"
17 #include "notif.h"
18 #include "rsrc.h"
19
20 #if defined(CONFIG_NET)
21 struct io_shutdown {
22 struct file *file;
23 int how;
24 };
25
26 struct io_accept {
27 struct file *file;
28 struct sockaddr __user *addr;
29 int __user *addr_len;
30 int flags;
31 int iou_flags;
32 u32 file_slot;
33 unsigned long nofile;
34 };
35
36 struct io_socket {
37 struct file *file;
38 int domain;
39 int type;
40 int protocol;
41 int flags;
42 u32 file_slot;
43 unsigned long nofile;
44 };
45
46 struct io_connect {
47 struct file *file;
48 struct sockaddr __user *addr;
49 int addr_len;
50 bool in_progress;
51 bool seen_econnaborted;
52 };
53
54 struct io_bind {
55 struct file *file;
56 int addr_len;
57 };
58
59 struct io_listen {
60 struct file *file;
61 int backlog;
62 };
63
64 struct io_sr_msg {
65 struct file *file;
66 union {
67 struct compat_msghdr __user *umsg_compat;
68 struct user_msghdr __user *umsg;
69 void __user *buf;
70 };
71 int len;
72 unsigned done_io;
73 unsigned msg_flags;
74 unsigned nr_multishot_loops;
75 u16 flags;
76 /* initialised and used only by !msg send variants */
77 u16 addr_len;
78 u16 buf_group;
79 unsigned short retry_flags;
80 void __user *addr;
81 void __user *msg_control;
82 /* used only for send zerocopy */
83 struct io_kiocb *notif;
84 };
85
86 enum sr_retry_flags {
87 IO_SR_MSG_RETRY = 1,
88 IO_SR_MSG_PARTIAL_MAP = 2,
89 };
90
91 /*
92 * Number of times we'll try and do receives if there's more data. If we
93 * exceed this limit, then add us to the back of the queue and retry from
94 * there. This helps fairness between flooding clients.
95 */
96 #define MULTISHOT_MAX_RETRY 32
97
io_shutdown_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)98 int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
99 {
100 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
101
102 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
103 sqe->buf_index || sqe->splice_fd_in))
104 return -EINVAL;
105
106 shutdown->how = READ_ONCE(sqe->len);
107 req->flags |= REQ_F_FORCE_ASYNC;
108 return 0;
109 }
110
io_shutdown(struct io_kiocb * req,unsigned int issue_flags)111 int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
112 {
113 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
114 struct socket *sock;
115 int ret;
116
117 WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
118
119 sock = sock_from_file(req->file);
120 if (unlikely(!sock))
121 return -ENOTSOCK;
122
123 ret = __sys_shutdown_sock(sock, shutdown->how);
124 io_req_set_res(req, ret, 0);
125 return IOU_OK;
126 }
127
io_net_retry(struct socket * sock,int flags)128 static bool io_net_retry(struct socket *sock, int flags)
129 {
130 if (!(flags & MSG_WAITALL))
131 return false;
132 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
133 }
134
io_netmsg_iovec_free(struct io_async_msghdr * kmsg)135 static void io_netmsg_iovec_free(struct io_async_msghdr *kmsg)
136 {
137 if (kmsg->free_iov) {
138 kfree(kmsg->free_iov);
139 kmsg->free_iov_nr = 0;
140 kmsg->free_iov = NULL;
141 }
142 }
143
io_netmsg_recycle(struct io_kiocb * req,unsigned int issue_flags)144 static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
145 {
146 struct io_async_msghdr *hdr = req->async_data;
147 struct iovec *iov;
148
149 /* can't recycle, ensure we free the iovec if we have one */
150 if (unlikely(issue_flags & IO_URING_F_UNLOCKED)) {
151 io_netmsg_iovec_free(hdr);
152 return;
153 }
154
155 /* Let normal cleanup path reap it if we fail adding to the cache */
156 iov = hdr->free_iov;
157 if (io_alloc_cache_put(&req->ctx->netmsg_cache, hdr)) {
158 if (iov)
159 kasan_mempool_poison_object(iov);
160 req->async_data = NULL;
161 req->flags &= ~(REQ_F_ASYNC_DATA|REQ_F_NEED_CLEANUP);
162 }
163 }
164
io_msg_alloc_async(struct io_kiocb * req)165 static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req)
166 {
167 struct io_ring_ctx *ctx = req->ctx;
168 struct io_async_msghdr *hdr;
169
170 hdr = io_alloc_cache_get(&ctx->netmsg_cache);
171 if (hdr) {
172 if (hdr->free_iov) {
173 kasan_mempool_unpoison_object(hdr->free_iov,
174 hdr->free_iov_nr * sizeof(struct iovec));
175 req->flags |= REQ_F_NEED_CLEANUP;
176 }
177 req->flags |= REQ_F_ASYNC_DATA;
178 req->async_data = hdr;
179 return hdr;
180 }
181
182 if (!io_alloc_async_data(req)) {
183 hdr = req->async_data;
184 hdr->free_iov_nr = 0;
185 hdr->free_iov = NULL;
186 return hdr;
187 }
188 return NULL;
189 }
190
191 /* assign new iovec to kmsg, if we need to */
io_net_vec_assign(struct io_kiocb * req,struct io_async_msghdr * kmsg,struct iovec * iov)192 static int io_net_vec_assign(struct io_kiocb *req, struct io_async_msghdr *kmsg,
193 struct iovec *iov)
194 {
195 if (iov) {
196 req->flags |= REQ_F_NEED_CLEANUP;
197 kmsg->free_iov_nr = kmsg->msg.msg_iter.nr_segs;
198 if (kmsg->free_iov)
199 kfree(kmsg->free_iov);
200 kmsg->free_iov = iov;
201 }
202 return 0;
203 }
204
io_mshot_prep_retry(struct io_kiocb * req,struct io_async_msghdr * kmsg)205 static inline void io_mshot_prep_retry(struct io_kiocb *req,
206 struct io_async_msghdr *kmsg)
207 {
208 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
209
210 req->flags &= ~REQ_F_BL_EMPTY;
211 sr->done_io = 0;
212 sr->retry_flags = 0;
213 sr->len = 0; /* get from the provided buffer */
214 req->buf_index = sr->buf_group;
215 }
216
217 #ifdef CONFIG_COMPAT
io_compat_msg_copy_hdr(struct io_kiocb * req,struct io_async_msghdr * iomsg,struct compat_msghdr * msg,int ddir)218 static int io_compat_msg_copy_hdr(struct io_kiocb *req,
219 struct io_async_msghdr *iomsg,
220 struct compat_msghdr *msg, int ddir)
221 {
222 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
223 struct compat_iovec __user *uiov;
224 struct iovec *iov;
225 int ret, nr_segs;
226
227 if (iomsg->free_iov) {
228 nr_segs = iomsg->free_iov_nr;
229 iov = iomsg->free_iov;
230 } else {
231 iov = &iomsg->fast_iov;
232 nr_segs = 1;
233 }
234
235 if (copy_from_user(msg, sr->umsg_compat, sizeof(*msg)))
236 return -EFAULT;
237
238 uiov = compat_ptr(msg->msg_iov);
239 if (req->flags & REQ_F_BUFFER_SELECT) {
240 compat_ssize_t clen;
241
242 if (msg->msg_iovlen == 0) {
243 sr->len = iov->iov_len = 0;
244 iov->iov_base = NULL;
245 } else if (msg->msg_iovlen > 1) {
246 return -EINVAL;
247 } else {
248 if (!access_ok(uiov, sizeof(*uiov)))
249 return -EFAULT;
250 if (__get_user(clen, &uiov->iov_len))
251 return -EFAULT;
252 if (clen < 0)
253 return -EINVAL;
254 sr->len = clen;
255 }
256
257 return 0;
258 }
259
260 ret = __import_iovec(ddir, (struct iovec __user *)uiov, msg->msg_iovlen,
261 nr_segs, &iov, &iomsg->msg.msg_iter, true);
262 if (unlikely(ret < 0))
263 return ret;
264
265 return io_net_vec_assign(req, iomsg, iov);
266 }
267 #endif
268
io_msg_copy_hdr(struct io_kiocb * req,struct io_async_msghdr * iomsg,struct user_msghdr * msg,int ddir)269 static int io_msg_copy_hdr(struct io_kiocb *req, struct io_async_msghdr *iomsg,
270 struct user_msghdr *msg, int ddir)
271 {
272 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
273 struct iovec *iov;
274 int ret, nr_segs;
275
276 if (iomsg->free_iov) {
277 nr_segs = iomsg->free_iov_nr;
278 iov = iomsg->free_iov;
279 } else {
280 iov = &iomsg->fast_iov;
281 nr_segs = 1;
282 }
283
284 if (!user_access_begin(sr->umsg, sizeof(*sr->umsg)))
285 return -EFAULT;
286
287 ret = -EFAULT;
288 unsafe_get_user(msg->msg_name, &sr->umsg->msg_name, ua_end);
289 unsafe_get_user(msg->msg_namelen, &sr->umsg->msg_namelen, ua_end);
290 unsafe_get_user(msg->msg_iov, &sr->umsg->msg_iov, ua_end);
291 unsafe_get_user(msg->msg_iovlen, &sr->umsg->msg_iovlen, ua_end);
292 unsafe_get_user(msg->msg_control, &sr->umsg->msg_control, ua_end);
293 unsafe_get_user(msg->msg_controllen, &sr->umsg->msg_controllen, ua_end);
294 msg->msg_flags = 0;
295
296 if (req->flags & REQ_F_BUFFER_SELECT) {
297 if (msg->msg_iovlen == 0) {
298 sr->len = iov->iov_len = 0;
299 iov->iov_base = NULL;
300 } else if (msg->msg_iovlen > 1) {
301 ret = -EINVAL;
302 goto ua_end;
303 } else {
304 /* we only need the length for provided buffers */
305 if (!access_ok(&msg->msg_iov[0].iov_len, sizeof(__kernel_size_t)))
306 goto ua_end;
307 unsafe_get_user(iov->iov_len, &msg->msg_iov[0].iov_len,
308 ua_end);
309 sr->len = iov->iov_len;
310 }
311 ret = 0;
312 ua_end:
313 user_access_end();
314 return ret;
315 }
316
317 user_access_end();
318 ret = __import_iovec(ddir, msg->msg_iov, msg->msg_iovlen, nr_segs,
319 &iov, &iomsg->msg.msg_iter, false);
320 if (unlikely(ret < 0))
321 return ret;
322
323 return io_net_vec_assign(req, iomsg, iov);
324 }
325
io_sendmsg_copy_hdr(struct io_kiocb * req,struct io_async_msghdr * iomsg)326 static int io_sendmsg_copy_hdr(struct io_kiocb *req,
327 struct io_async_msghdr *iomsg)
328 {
329 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
330 struct user_msghdr msg;
331 int ret;
332
333 iomsg->msg.msg_name = &iomsg->addr;
334 iomsg->msg.msg_iter.nr_segs = 0;
335
336 #ifdef CONFIG_COMPAT
337 if (unlikely(req->ctx->compat)) {
338 struct compat_msghdr cmsg;
339
340 ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ITER_SOURCE);
341 if (unlikely(ret))
342 return ret;
343
344 ret = __get_compat_msghdr(&iomsg->msg, &cmsg, NULL);
345 sr->msg_control = iomsg->msg.msg_control_user;
346 return ret;
347 }
348 #endif
349
350 ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_SOURCE);
351 if (unlikely(ret))
352 return ret;
353
354 ret = __copy_msghdr(&iomsg->msg, &msg, NULL);
355
356 /* save msg_control as sys_sendmsg() overwrites it */
357 sr->msg_control = iomsg->msg.msg_control_user;
358 return ret;
359 }
360
io_sendmsg_recvmsg_cleanup(struct io_kiocb * req)361 void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
362 {
363 struct io_async_msghdr *io = req->async_data;
364
365 io_netmsg_iovec_free(io);
366 }
367
io_send_setup(struct io_kiocb * req)368 static int io_send_setup(struct io_kiocb *req)
369 {
370 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
371 struct io_async_msghdr *kmsg = req->async_data;
372 int ret;
373
374 kmsg->msg.msg_name = NULL;
375 kmsg->msg.msg_namelen = 0;
376 kmsg->msg.msg_control = NULL;
377 kmsg->msg.msg_controllen = 0;
378 kmsg->msg.msg_ubuf = NULL;
379
380 if (sr->addr) {
381 ret = move_addr_to_kernel(sr->addr, sr->addr_len, &kmsg->addr);
382 if (unlikely(ret < 0))
383 return ret;
384 kmsg->msg.msg_name = &kmsg->addr;
385 kmsg->msg.msg_namelen = sr->addr_len;
386 }
387 if (!io_do_buffer_select(req)) {
388 ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len,
389 &kmsg->msg.msg_iter);
390 if (unlikely(ret < 0))
391 return ret;
392 }
393 return 0;
394 }
395
io_sendmsg_prep_setup(struct io_kiocb * req,int is_msg)396 static int io_sendmsg_prep_setup(struct io_kiocb *req, int is_msg)
397 {
398 struct io_async_msghdr *kmsg;
399 int ret;
400
401 kmsg = io_msg_alloc_async(req);
402 if (unlikely(!kmsg))
403 return -ENOMEM;
404 if (!is_msg)
405 return io_send_setup(req);
406 ret = io_sendmsg_copy_hdr(req, kmsg);
407 if (!ret)
408 req->flags |= REQ_F_NEED_CLEANUP;
409 return ret;
410 }
411
412 #define SENDMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_BUNDLE)
413
io_sendmsg_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)414 int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
415 {
416 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
417
418 sr->done_io = 0;
419 sr->retry_flags = 0;
420
421 if (req->opcode == IORING_OP_SEND) {
422 if (READ_ONCE(sqe->__pad3[0]))
423 return -EINVAL;
424 sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
425 sr->addr_len = READ_ONCE(sqe->addr_len);
426 } else if (sqe->addr2 || sqe->file_index) {
427 return -EINVAL;
428 }
429
430 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
431 sr->len = READ_ONCE(sqe->len);
432 sr->flags = READ_ONCE(sqe->ioprio);
433 if (sr->flags & ~SENDMSG_FLAGS)
434 return -EINVAL;
435 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
436 if (sr->msg_flags & MSG_DONTWAIT)
437 req->flags |= REQ_F_NOWAIT;
438 if (sr->flags & IORING_RECVSEND_BUNDLE) {
439 if (req->opcode == IORING_OP_SENDMSG)
440 return -EINVAL;
441 if (!(req->flags & REQ_F_BUFFER_SELECT))
442 return -EINVAL;
443 sr->msg_flags |= MSG_WAITALL;
444 sr->buf_group = req->buf_index;
445 req->buf_list = NULL;
446 req->flags |= REQ_F_MULTISHOT;
447 }
448
449 #ifdef CONFIG_COMPAT
450 if (req->ctx->compat)
451 sr->msg_flags |= MSG_CMSG_COMPAT;
452 #endif
453 return io_sendmsg_prep_setup(req, req->opcode == IORING_OP_SENDMSG);
454 }
455
io_req_msg_cleanup(struct io_kiocb * req,unsigned int issue_flags)456 static void io_req_msg_cleanup(struct io_kiocb *req,
457 unsigned int issue_flags)
458 {
459 io_netmsg_recycle(req, issue_flags);
460 }
461
462 /*
463 * For bundle completions, we need to figure out how many segments we consumed.
464 * A bundle could be using a single ITER_UBUF if that's all we mapped, or it
465 * could be using an ITER_IOVEC. If the latter, then if we consumed all of
466 * the segments, then it's a trivial questiont o answer. If we have residual
467 * data in the iter, then loop the segments to figure out how much we
468 * transferred.
469 */
io_bundle_nbufs(struct io_async_msghdr * kmsg,int ret)470 static int io_bundle_nbufs(struct io_async_msghdr *kmsg, int ret)
471 {
472 struct iovec *iov;
473 int nbufs;
474
475 /* no data is always zero segments, and a ubuf is always 1 segment */
476 if (ret <= 0)
477 return 0;
478 if (iter_is_ubuf(&kmsg->msg.msg_iter))
479 return 1;
480
481 iov = kmsg->free_iov;
482 if (!iov)
483 iov = &kmsg->fast_iov;
484
485 /* if all data was transferred, it's basic pointer math */
486 if (!iov_iter_count(&kmsg->msg.msg_iter))
487 return iter_iov(&kmsg->msg.msg_iter) - iov;
488
489 /* short transfer, count segments */
490 nbufs = 0;
491 do {
492 int this_len = min_t(int, iov[nbufs].iov_len, ret);
493
494 nbufs++;
495 ret -= this_len;
496 } while (ret);
497
498 return nbufs;
499 }
500
io_net_kbuf_recyle(struct io_kiocb * req,struct io_async_msghdr * kmsg,int len)501 static int io_net_kbuf_recyle(struct io_kiocb *req,
502 struct io_async_msghdr *kmsg, int len)
503 {
504 req->flags |= REQ_F_BL_NO_RECYCLE;
505 if (req->flags & REQ_F_BUFFERS_COMMIT)
506 io_kbuf_commit(req, req->buf_list, len, io_bundle_nbufs(kmsg, len));
507 return -EAGAIN;
508 }
509
io_send_finish(struct io_kiocb * req,int * ret,struct io_async_msghdr * kmsg,unsigned issue_flags)510 static inline bool io_send_finish(struct io_kiocb *req, int *ret,
511 struct io_async_msghdr *kmsg,
512 unsigned issue_flags)
513 {
514 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
515 bool bundle_finished = *ret <= 0;
516 unsigned int cflags;
517
518 if (!(sr->flags & IORING_RECVSEND_BUNDLE)) {
519 cflags = io_put_kbuf(req, *ret, issue_flags);
520 goto finish;
521 }
522
523 cflags = io_put_kbufs(req, *ret, io_bundle_nbufs(kmsg, *ret), issue_flags);
524
525 if (bundle_finished || req->flags & REQ_F_BL_EMPTY)
526 goto finish;
527
528 /*
529 * Fill CQE for this receive and see if we should keep trying to
530 * receive from this socket.
531 */
532 if (io_req_post_cqe(req, *ret, cflags | IORING_CQE_F_MORE)) {
533 io_mshot_prep_retry(req, kmsg);
534 return false;
535 }
536
537 /* Otherwise stop bundle and use the current result. */
538 finish:
539 io_req_set_res(req, *ret, cflags);
540 *ret = IOU_OK;
541 return true;
542 }
543
io_sendmsg(struct io_kiocb * req,unsigned int issue_flags)544 int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
545 {
546 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
547 struct io_async_msghdr *kmsg = req->async_data;
548 struct socket *sock;
549 unsigned flags;
550 int min_ret = 0;
551 int ret;
552
553 sock = sock_from_file(req->file);
554 if (unlikely(!sock))
555 return -ENOTSOCK;
556
557 if (!(req->flags & REQ_F_POLLED) &&
558 (sr->flags & IORING_RECVSEND_POLL_FIRST))
559 return -EAGAIN;
560
561 flags = sr->msg_flags;
562 if (issue_flags & IO_URING_F_NONBLOCK)
563 flags |= MSG_DONTWAIT;
564 if (flags & MSG_WAITALL)
565 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
566
567 kmsg->msg.msg_control_user = sr->msg_control;
568
569 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
570
571 if (ret < min_ret) {
572 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
573 return -EAGAIN;
574 if (ret > 0 && io_net_retry(sock, flags)) {
575 kmsg->msg.msg_controllen = 0;
576 kmsg->msg.msg_control = NULL;
577 sr->done_io += ret;
578 return io_net_kbuf_recyle(req, kmsg, ret);
579 }
580 if (ret == -ERESTARTSYS)
581 ret = -EINTR;
582 req_set_fail(req);
583 }
584 io_req_msg_cleanup(req, issue_flags);
585 if (ret >= 0)
586 ret += sr->done_io;
587 else if (sr->done_io)
588 ret = sr->done_io;
589 io_req_set_res(req, ret, 0);
590 return IOU_OK;
591 }
592
io_send(struct io_kiocb * req,unsigned int issue_flags)593 int io_send(struct io_kiocb *req, unsigned int issue_flags)
594 {
595 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
596 struct io_async_msghdr *kmsg = req->async_data;
597 struct socket *sock;
598 unsigned flags;
599 int min_ret = 0;
600 int ret;
601
602 sock = sock_from_file(req->file);
603 if (unlikely(!sock))
604 return -ENOTSOCK;
605
606 if (!(req->flags & REQ_F_POLLED) &&
607 (sr->flags & IORING_RECVSEND_POLL_FIRST))
608 return -EAGAIN;
609
610 flags = sr->msg_flags;
611 if (issue_flags & IO_URING_F_NONBLOCK)
612 flags |= MSG_DONTWAIT;
613
614 retry_bundle:
615 if (io_do_buffer_select(req)) {
616 struct buf_sel_arg arg = {
617 .iovs = &kmsg->fast_iov,
618 .max_len = min_not_zero(sr->len, INT_MAX),
619 .nr_iovs = 1,
620 };
621
622 if (kmsg->free_iov) {
623 arg.nr_iovs = kmsg->free_iov_nr;
624 arg.iovs = kmsg->free_iov;
625 arg.mode = KBUF_MODE_FREE;
626 }
627
628 if (!(sr->flags & IORING_RECVSEND_BUNDLE))
629 arg.nr_iovs = 1;
630 else
631 arg.mode |= KBUF_MODE_EXPAND;
632
633 ret = io_buffers_select(req, &arg, issue_flags);
634 if (unlikely(ret < 0))
635 return ret;
636
637 if (arg.iovs != &kmsg->fast_iov && arg.iovs != kmsg->free_iov) {
638 kmsg->free_iov_nr = ret;
639 kmsg->free_iov = arg.iovs;
640 req->flags |= REQ_F_NEED_CLEANUP;
641 }
642 sr->len = arg.out_len;
643
644 if (ret == 1) {
645 sr->buf = arg.iovs[0].iov_base;
646 ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len,
647 &kmsg->msg.msg_iter);
648 if (unlikely(ret))
649 return ret;
650 } else {
651 iov_iter_init(&kmsg->msg.msg_iter, ITER_SOURCE,
652 arg.iovs, ret, arg.out_len);
653 }
654 }
655
656 /*
657 * If MSG_WAITALL is set, or this is a bundle send, then we need
658 * the full amount. If just bundle is set, if we do a short send
659 * then we complete the bundle sequence rather than continue on.
660 */
661 if (flags & MSG_WAITALL || sr->flags & IORING_RECVSEND_BUNDLE)
662 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
663
664 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
665 kmsg->msg.msg_flags = flags;
666 ret = sock_sendmsg(sock, &kmsg->msg);
667 if (ret < min_ret) {
668 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
669 return -EAGAIN;
670
671 if (ret > 0 && io_net_retry(sock, flags)) {
672 sr->len -= ret;
673 sr->buf += ret;
674 sr->done_io += ret;
675 return io_net_kbuf_recyle(req, kmsg, ret);
676 }
677 if (ret == -ERESTARTSYS)
678 ret = -EINTR;
679 req_set_fail(req);
680 }
681 if (ret >= 0)
682 ret += sr->done_io;
683 else if (sr->done_io)
684 ret = sr->done_io;
685
686 if (!io_send_finish(req, &ret, kmsg, issue_flags))
687 goto retry_bundle;
688
689 io_req_msg_cleanup(req, issue_flags);
690 return ret;
691 }
692
io_recvmsg_mshot_prep(struct io_kiocb * req,struct io_async_msghdr * iomsg,int namelen,size_t controllen)693 static int io_recvmsg_mshot_prep(struct io_kiocb *req,
694 struct io_async_msghdr *iomsg,
695 int namelen, size_t controllen)
696 {
697 if ((req->flags & (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) ==
698 (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) {
699 int hdr;
700
701 if (unlikely(namelen < 0))
702 return -EOVERFLOW;
703 if (check_add_overflow(sizeof(struct io_uring_recvmsg_out),
704 namelen, &hdr))
705 return -EOVERFLOW;
706 if (check_add_overflow(hdr, controllen, &hdr))
707 return -EOVERFLOW;
708
709 iomsg->namelen = namelen;
710 iomsg->controllen = controllen;
711 return 0;
712 }
713
714 return 0;
715 }
716
io_recvmsg_copy_hdr(struct io_kiocb * req,struct io_async_msghdr * iomsg)717 static int io_recvmsg_copy_hdr(struct io_kiocb *req,
718 struct io_async_msghdr *iomsg)
719 {
720 struct user_msghdr msg;
721 int ret;
722
723 iomsg->msg.msg_name = &iomsg->addr;
724 iomsg->msg.msg_iter.nr_segs = 0;
725
726 #ifdef CONFIG_COMPAT
727 if (unlikely(req->ctx->compat)) {
728 struct compat_msghdr cmsg;
729
730 ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ITER_DEST);
731 if (unlikely(ret))
732 return ret;
733
734 ret = __get_compat_msghdr(&iomsg->msg, &cmsg, &iomsg->uaddr);
735 if (unlikely(ret))
736 return ret;
737
738 return io_recvmsg_mshot_prep(req, iomsg, cmsg.msg_namelen,
739 cmsg.msg_controllen);
740 }
741 #endif
742
743 ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_DEST);
744 if (unlikely(ret))
745 return ret;
746
747 ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
748 if (unlikely(ret))
749 return ret;
750
751 return io_recvmsg_mshot_prep(req, iomsg, msg.msg_namelen,
752 msg.msg_controllen);
753 }
754
io_recvmsg_prep_setup(struct io_kiocb * req)755 static int io_recvmsg_prep_setup(struct io_kiocb *req)
756 {
757 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
758 struct io_async_msghdr *kmsg;
759 int ret;
760
761 kmsg = io_msg_alloc_async(req);
762 if (unlikely(!kmsg))
763 return -ENOMEM;
764
765 if (req->opcode == IORING_OP_RECV) {
766 kmsg->msg.msg_name = NULL;
767 kmsg->msg.msg_namelen = 0;
768 kmsg->msg.msg_inq = 0;
769 kmsg->msg.msg_control = NULL;
770 kmsg->msg.msg_get_inq = 1;
771 kmsg->msg.msg_controllen = 0;
772 kmsg->msg.msg_iocb = NULL;
773 kmsg->msg.msg_ubuf = NULL;
774
775 if (!io_do_buffer_select(req)) {
776 ret = import_ubuf(ITER_DEST, sr->buf, sr->len,
777 &kmsg->msg.msg_iter);
778 if (unlikely(ret))
779 return ret;
780 }
781 return 0;
782 }
783
784 ret = io_recvmsg_copy_hdr(req, kmsg);
785 if (!ret)
786 req->flags |= REQ_F_NEED_CLEANUP;
787 return ret;
788 }
789
790 #define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT | \
791 IORING_RECVSEND_BUNDLE)
792
io_recvmsg_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)793 int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
794 {
795 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
796
797 sr->done_io = 0;
798 sr->retry_flags = 0;
799
800 if (unlikely(sqe->file_index || sqe->addr2))
801 return -EINVAL;
802
803 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
804 sr->len = READ_ONCE(sqe->len);
805 sr->flags = READ_ONCE(sqe->ioprio);
806 if (sr->flags & ~RECVMSG_FLAGS)
807 return -EINVAL;
808 sr->msg_flags = READ_ONCE(sqe->msg_flags);
809 if (sr->msg_flags & MSG_DONTWAIT)
810 req->flags |= REQ_F_NOWAIT;
811 if (sr->msg_flags & MSG_ERRQUEUE)
812 req->flags |= REQ_F_CLEAR_POLLIN;
813 if (req->flags & REQ_F_BUFFER_SELECT) {
814 /*
815 * Store the buffer group for this multishot receive separately,
816 * as if we end up doing an io-wq based issue that selects a
817 * buffer, it has to be committed immediately and that will
818 * clear ->buf_list. This means we lose the link to the buffer
819 * list, and the eventual buffer put on completion then cannot
820 * restore it.
821 */
822 sr->buf_group = req->buf_index;
823 req->buf_list = NULL;
824 }
825 if (sr->flags & IORING_RECV_MULTISHOT) {
826 if (!(req->flags & REQ_F_BUFFER_SELECT))
827 return -EINVAL;
828 if (sr->msg_flags & MSG_WAITALL)
829 return -EINVAL;
830 if (req->opcode == IORING_OP_RECV && sr->len)
831 return -EINVAL;
832 req->flags |= REQ_F_APOLL_MULTISHOT;
833 }
834 if (sr->flags & IORING_RECVSEND_BUNDLE) {
835 if (req->opcode == IORING_OP_RECVMSG)
836 return -EINVAL;
837 }
838
839 #ifdef CONFIG_COMPAT
840 if (req->ctx->compat)
841 sr->msg_flags |= MSG_CMSG_COMPAT;
842 #endif
843 sr->nr_multishot_loops = 0;
844 return io_recvmsg_prep_setup(req);
845 }
846
847 /* bits to clear in old and inherit in new cflags on bundle retry */
848 #define CQE_F_MASK (IORING_CQE_F_SOCK_NONEMPTY|IORING_CQE_F_MORE)
849
850 /*
851 * Finishes io_recv and io_recvmsg.
852 *
853 * Returns true if it is actually finished, or false if it should run
854 * again (for multishot).
855 */
io_recv_finish(struct io_kiocb * req,int * ret,struct io_async_msghdr * kmsg,bool mshot_finished,unsigned issue_flags)856 static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
857 struct io_async_msghdr *kmsg,
858 bool mshot_finished, unsigned issue_flags)
859 {
860 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
861 unsigned int cflags = 0;
862
863 if (kmsg->msg.msg_inq > 0)
864 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
865
866 if (sr->flags & IORING_RECVSEND_BUNDLE) {
867 size_t this_ret = *ret - sr->done_io;
868
869 cflags |= io_put_kbufs(req, this_ret, io_bundle_nbufs(kmsg, this_ret),
870 issue_flags);
871 if (sr->retry_flags & IO_SR_MSG_RETRY)
872 cflags = req->cqe.flags | (cflags & CQE_F_MASK);
873 /* bundle with no more immediate buffers, we're done */
874 if (req->flags & REQ_F_BL_EMPTY)
875 goto finish;
876 /*
877 * If more is available AND it was a full transfer, retry and
878 * append to this one
879 */
880 if (!sr->retry_flags && kmsg->msg.msg_inq > 1 && this_ret > 0 &&
881 !iov_iter_count(&kmsg->msg.msg_iter)) {
882 req->cqe.flags = cflags & ~CQE_F_MASK;
883 sr->len = kmsg->msg.msg_inq;
884 sr->done_io += this_ret;
885 sr->retry_flags |= IO_SR_MSG_RETRY;
886 return false;
887 }
888 } else {
889 cflags |= io_put_kbuf(req, *ret, issue_flags);
890 }
891
892 /*
893 * Fill CQE for this receive and see if we should keep trying to
894 * receive from this socket.
895 */
896 if ((req->flags & REQ_F_APOLL_MULTISHOT) && !mshot_finished &&
897 io_req_post_cqe(req, *ret, cflags | IORING_CQE_F_MORE)) {
898 int mshot_retry_ret = IOU_ISSUE_SKIP_COMPLETE;
899
900 io_mshot_prep_retry(req, kmsg);
901 /* Known not-empty or unknown state, retry */
902 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || kmsg->msg.msg_inq < 0) {
903 if (sr->nr_multishot_loops++ < MULTISHOT_MAX_RETRY)
904 return false;
905 /* mshot retries exceeded, force a requeue */
906 sr->nr_multishot_loops = 0;
907 mshot_retry_ret = IOU_REQUEUE;
908 }
909 if (issue_flags & IO_URING_F_MULTISHOT)
910 *ret = mshot_retry_ret;
911 else
912 *ret = -EAGAIN;
913 return true;
914 }
915
916 /* Finish the request / stop multishot. */
917 finish:
918 io_req_set_res(req, *ret, cflags);
919
920 if (issue_flags & IO_URING_F_MULTISHOT)
921 *ret = IOU_STOP_MULTISHOT;
922 else
923 *ret = IOU_OK;
924 io_req_msg_cleanup(req, issue_flags);
925 return true;
926 }
927
io_recvmsg_prep_multishot(struct io_async_msghdr * kmsg,struct io_sr_msg * sr,void __user ** buf,size_t * len)928 static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
929 struct io_sr_msg *sr, void __user **buf,
930 size_t *len)
931 {
932 unsigned long ubuf = (unsigned long) *buf;
933 unsigned long hdr;
934
935 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
936 kmsg->controllen;
937 if (*len < hdr)
938 return -EFAULT;
939
940 if (kmsg->controllen) {
941 unsigned long control = ubuf + hdr - kmsg->controllen;
942
943 kmsg->msg.msg_control_user = (void __user *) control;
944 kmsg->msg.msg_controllen = kmsg->controllen;
945 }
946
947 sr->buf = *buf; /* stash for later copy */
948 *buf = (void __user *) (ubuf + hdr);
949 kmsg->payloadlen = *len = *len - hdr;
950 return 0;
951 }
952
953 struct io_recvmsg_multishot_hdr {
954 struct io_uring_recvmsg_out msg;
955 struct sockaddr_storage addr;
956 };
957
io_recvmsg_multishot(struct socket * sock,struct io_sr_msg * io,struct io_async_msghdr * kmsg,unsigned int flags,bool * finished)958 static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
959 struct io_async_msghdr *kmsg,
960 unsigned int flags, bool *finished)
961 {
962 int err;
963 int copy_len;
964 struct io_recvmsg_multishot_hdr hdr;
965
966 if (kmsg->namelen)
967 kmsg->msg.msg_name = &hdr.addr;
968 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
969 kmsg->msg.msg_namelen = 0;
970
971 if (sock->file->f_flags & O_NONBLOCK)
972 flags |= MSG_DONTWAIT;
973
974 err = sock_recvmsg(sock, &kmsg->msg, flags);
975 *finished = err <= 0;
976 if (err < 0)
977 return err;
978
979 hdr.msg = (struct io_uring_recvmsg_out) {
980 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
981 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
982 };
983
984 hdr.msg.payloadlen = err;
985 if (err > kmsg->payloadlen)
986 err = kmsg->payloadlen;
987
988 copy_len = sizeof(struct io_uring_recvmsg_out);
989 if (kmsg->msg.msg_namelen > kmsg->namelen)
990 copy_len += kmsg->namelen;
991 else
992 copy_len += kmsg->msg.msg_namelen;
993
994 /*
995 * "fromlen shall refer to the value before truncation.."
996 * 1003.1g
997 */
998 hdr.msg.namelen = kmsg->msg.msg_namelen;
999
1000 /* ensure that there is no gap between hdr and sockaddr_storage */
1001 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
1002 sizeof(struct io_uring_recvmsg_out));
1003 if (copy_to_user(io->buf, &hdr, copy_len)) {
1004 *finished = true;
1005 return -EFAULT;
1006 }
1007
1008 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
1009 kmsg->controllen + err;
1010 }
1011
io_recvmsg(struct io_kiocb * req,unsigned int issue_flags)1012 int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
1013 {
1014 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1015 struct io_async_msghdr *kmsg = req->async_data;
1016 struct socket *sock;
1017 unsigned flags;
1018 int ret, min_ret = 0;
1019 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1020 bool mshot_finished = true;
1021
1022 sock = sock_from_file(req->file);
1023 if (unlikely(!sock))
1024 return -ENOTSOCK;
1025
1026 if (!(req->flags & REQ_F_POLLED) &&
1027 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1028 return -EAGAIN;
1029
1030 flags = sr->msg_flags;
1031 if (force_nonblock)
1032 flags |= MSG_DONTWAIT;
1033
1034 retry_multishot:
1035 if (io_do_buffer_select(req)) {
1036 void __user *buf;
1037 size_t len = sr->len;
1038
1039 buf = io_buffer_select(req, &len, issue_flags);
1040 if (!buf)
1041 return -ENOBUFS;
1042
1043 if (req->flags & REQ_F_APOLL_MULTISHOT) {
1044 ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
1045 if (ret) {
1046 io_kbuf_recycle(req, issue_flags);
1047 return ret;
1048 }
1049 }
1050
1051 iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_DEST, buf, len);
1052 }
1053
1054 kmsg->msg.msg_get_inq = 1;
1055 kmsg->msg.msg_inq = -1;
1056 if (req->flags & REQ_F_APOLL_MULTISHOT) {
1057 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
1058 &mshot_finished);
1059 } else {
1060 /* disable partial retry for recvmsg with cmsg attached */
1061 if (flags & MSG_WAITALL && !kmsg->msg.msg_controllen)
1062 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1063
1064 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
1065 kmsg->uaddr, flags);
1066 }
1067
1068 if (ret < min_ret) {
1069 if (ret == -EAGAIN && force_nonblock) {
1070 if (issue_flags & IO_URING_F_MULTISHOT) {
1071 io_kbuf_recycle(req, issue_flags);
1072 return IOU_ISSUE_SKIP_COMPLETE;
1073 }
1074 return -EAGAIN;
1075 }
1076 if (ret > 0 && io_net_retry(sock, flags)) {
1077 sr->done_io += ret;
1078 return io_net_kbuf_recyle(req, kmsg, ret);
1079 }
1080 if (ret == -ERESTARTSYS)
1081 ret = -EINTR;
1082 req_set_fail(req);
1083 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
1084 req_set_fail(req);
1085 }
1086
1087 if (ret > 0)
1088 ret += sr->done_io;
1089 else if (sr->done_io)
1090 ret = sr->done_io;
1091 else
1092 io_kbuf_recycle(req, issue_flags);
1093
1094 if (!io_recv_finish(req, &ret, kmsg, mshot_finished, issue_flags))
1095 goto retry_multishot;
1096
1097 return ret;
1098 }
1099
io_recv_buf_select(struct io_kiocb * req,struct io_async_msghdr * kmsg,size_t * len,unsigned int issue_flags)1100 static int io_recv_buf_select(struct io_kiocb *req, struct io_async_msghdr *kmsg,
1101 size_t *len, unsigned int issue_flags)
1102 {
1103 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1104 int ret;
1105
1106 /*
1107 * If the ring isn't locked, then don't use the peek interface
1108 * to grab multiple buffers as we will lock/unlock between
1109 * this selection and posting the buffers.
1110 */
1111 if (!(issue_flags & IO_URING_F_UNLOCKED) &&
1112 sr->flags & IORING_RECVSEND_BUNDLE) {
1113 struct buf_sel_arg arg = {
1114 .iovs = &kmsg->fast_iov,
1115 .nr_iovs = 1,
1116 .mode = KBUF_MODE_EXPAND,
1117 };
1118
1119 if (kmsg->free_iov) {
1120 arg.nr_iovs = kmsg->free_iov_nr;
1121 arg.iovs = kmsg->free_iov;
1122 arg.mode |= KBUF_MODE_FREE;
1123 }
1124
1125 if (kmsg->msg.msg_inq > 1)
1126 arg.max_len = min_not_zero(sr->len, kmsg->msg.msg_inq);
1127
1128 ret = io_buffers_peek(req, &arg);
1129 if (unlikely(ret < 0))
1130 return ret;
1131
1132 if (arg.iovs != &kmsg->fast_iov && arg.iovs != kmsg->free_iov) {
1133 kmsg->free_iov_nr = ret;
1134 kmsg->free_iov = arg.iovs;
1135 req->flags |= REQ_F_NEED_CLEANUP;
1136 }
1137 if (arg.partial_map)
1138 sr->retry_flags |= IO_SR_MSG_PARTIAL_MAP;
1139
1140 /* special case 1 vec, can be a fast path */
1141 if (ret == 1) {
1142 sr->buf = arg.iovs[0].iov_base;
1143 sr->len = arg.iovs[0].iov_len;
1144 goto map_ubuf;
1145 }
1146 iov_iter_init(&kmsg->msg.msg_iter, ITER_DEST, arg.iovs, ret,
1147 arg.out_len);
1148 } else {
1149 void __user *buf;
1150
1151 *len = sr->len;
1152 buf = io_buffer_select(req, len, issue_flags);
1153 if (!buf)
1154 return -ENOBUFS;
1155 sr->buf = buf;
1156 sr->len = *len;
1157 map_ubuf:
1158 ret = import_ubuf(ITER_DEST, sr->buf, sr->len,
1159 &kmsg->msg.msg_iter);
1160 if (unlikely(ret))
1161 return ret;
1162 }
1163
1164 return 0;
1165 }
1166
io_recv(struct io_kiocb * req,unsigned int issue_flags)1167 int io_recv(struct io_kiocb *req, unsigned int issue_flags)
1168 {
1169 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1170 struct io_async_msghdr *kmsg = req->async_data;
1171 struct socket *sock;
1172 unsigned flags;
1173 int ret, min_ret = 0;
1174 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1175 size_t len = sr->len;
1176 bool mshot_finished;
1177
1178 if (!(req->flags & REQ_F_POLLED) &&
1179 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1180 return -EAGAIN;
1181
1182 sock = sock_from_file(req->file);
1183 if (unlikely(!sock))
1184 return -ENOTSOCK;
1185
1186 flags = sr->msg_flags;
1187 if (force_nonblock)
1188 flags |= MSG_DONTWAIT;
1189
1190 retry_multishot:
1191 if (io_do_buffer_select(req)) {
1192 ret = io_recv_buf_select(req, kmsg, &len, issue_flags);
1193 if (unlikely(ret)) {
1194 kmsg->msg.msg_inq = -1;
1195 goto out_free;
1196 }
1197 sr->buf = NULL;
1198 }
1199
1200 kmsg->msg.msg_flags = 0;
1201 kmsg->msg.msg_inq = -1;
1202
1203 if (flags & MSG_WAITALL)
1204 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1205
1206 ret = sock_recvmsg(sock, &kmsg->msg, flags);
1207 if (ret < min_ret) {
1208 if (ret == -EAGAIN && force_nonblock) {
1209 if (issue_flags & IO_URING_F_MULTISHOT) {
1210 io_kbuf_recycle(req, issue_flags);
1211 return IOU_ISSUE_SKIP_COMPLETE;
1212 }
1213
1214 return -EAGAIN;
1215 }
1216 if (ret > 0 && io_net_retry(sock, flags)) {
1217 sr->len -= ret;
1218 sr->buf += ret;
1219 sr->done_io += ret;
1220 return io_net_kbuf_recyle(req, kmsg, ret);
1221 }
1222 if (ret == -ERESTARTSYS)
1223 ret = -EINTR;
1224 req_set_fail(req);
1225 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
1226 out_free:
1227 req_set_fail(req);
1228 }
1229
1230 mshot_finished = ret <= 0;
1231 if (ret > 0)
1232 ret += sr->done_io;
1233 else if (sr->done_io)
1234 ret = sr->done_io;
1235 else
1236 io_kbuf_recycle(req, issue_flags);
1237
1238 if (!io_recv_finish(req, &ret, kmsg, mshot_finished, issue_flags))
1239 goto retry_multishot;
1240
1241 return ret;
1242 }
1243
io_send_zc_cleanup(struct io_kiocb * req)1244 void io_send_zc_cleanup(struct io_kiocb *req)
1245 {
1246 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1247 struct io_async_msghdr *io = req->async_data;
1248
1249 if (req_has_async_data(req))
1250 io_netmsg_iovec_free(io);
1251 if (zc->notif) {
1252 io_notif_flush(zc->notif);
1253 zc->notif = NULL;
1254 }
1255 }
1256
1257 #define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF)
1258 #define IO_ZC_FLAGS_VALID (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE)
1259
io_send_zc_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1260 int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1261 {
1262 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1263 struct io_ring_ctx *ctx = req->ctx;
1264 struct io_kiocb *notif;
1265
1266 zc->done_io = 0;
1267 zc->retry_flags = 0;
1268 req->flags |= REQ_F_POLL_NO_LAZY;
1269
1270 if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
1271 return -EINVAL;
1272 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
1273 if (req->flags & REQ_F_CQE_SKIP)
1274 return -EINVAL;
1275
1276 notif = zc->notif = io_alloc_notif(ctx);
1277 if (!notif)
1278 return -ENOMEM;
1279 notif->cqe.user_data = req->cqe.user_data;
1280 notif->cqe.res = 0;
1281 notif->cqe.flags = IORING_CQE_F_NOTIF;
1282 req->flags |= REQ_F_NEED_CLEANUP;
1283
1284 zc->flags = READ_ONCE(sqe->ioprio);
1285 if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) {
1286 if (zc->flags & ~IO_ZC_FLAGS_VALID)
1287 return -EINVAL;
1288 if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) {
1289 struct io_notif_data *nd = io_notif_to_data(notif);
1290
1291 nd->zc_report = true;
1292 nd->zc_used = false;
1293 nd->zc_copied = false;
1294 }
1295 }
1296
1297 if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1298 unsigned idx = READ_ONCE(sqe->buf_index);
1299
1300 if (unlikely(idx >= ctx->nr_user_bufs))
1301 return -EFAULT;
1302 idx = array_index_nospec(idx, ctx->nr_user_bufs);
1303 req->imu = READ_ONCE(ctx->user_bufs[idx]);
1304 io_req_set_rsrc_node(notif, ctx, 0);
1305 }
1306
1307 if (req->opcode == IORING_OP_SEND_ZC) {
1308 if (READ_ONCE(sqe->__pad3[0]))
1309 return -EINVAL;
1310 zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1311 zc->addr_len = READ_ONCE(sqe->addr_len);
1312 } else {
1313 if (unlikely(sqe->addr2 || sqe->file_index))
1314 return -EINVAL;
1315 if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
1316 return -EINVAL;
1317 }
1318
1319 zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
1320 zc->len = READ_ONCE(sqe->len);
1321 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL | MSG_ZEROCOPY;
1322 if (zc->msg_flags & MSG_DONTWAIT)
1323 req->flags |= REQ_F_NOWAIT;
1324
1325 #ifdef CONFIG_COMPAT
1326 if (req->ctx->compat)
1327 zc->msg_flags |= MSG_CMSG_COMPAT;
1328 #endif
1329 return io_sendmsg_prep_setup(req, req->opcode == IORING_OP_SENDMSG_ZC);
1330 }
1331
io_sg_from_iter_iovec(struct sk_buff * skb,struct iov_iter * from,size_t length)1332 static int io_sg_from_iter_iovec(struct sk_buff *skb,
1333 struct iov_iter *from, size_t length)
1334 {
1335 skb_zcopy_downgrade_managed(skb);
1336 return zerocopy_fill_skb_from_iter(skb, from, length);
1337 }
1338
io_sg_from_iter(struct sk_buff * skb,struct iov_iter * from,size_t length)1339 static int io_sg_from_iter(struct sk_buff *skb,
1340 struct iov_iter *from, size_t length)
1341 {
1342 struct skb_shared_info *shinfo = skb_shinfo(skb);
1343 int frag = shinfo->nr_frags;
1344 int ret = 0;
1345 struct bvec_iter bi;
1346 ssize_t copied = 0;
1347 unsigned long truesize = 0;
1348
1349 if (!frag)
1350 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
1351 else if (unlikely(!skb_zcopy_managed(skb)))
1352 return zerocopy_fill_skb_from_iter(skb, from, length);
1353
1354 bi.bi_size = min(from->count, length);
1355 bi.bi_bvec_done = from->iov_offset;
1356 bi.bi_idx = 0;
1357
1358 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1359 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1360
1361 copied += v.bv_len;
1362 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1363 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1364 v.bv_offset, v.bv_len);
1365 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1366 }
1367 if (bi.bi_size)
1368 ret = -EMSGSIZE;
1369
1370 shinfo->nr_frags = frag;
1371 from->bvec += bi.bi_idx;
1372 from->nr_segs -= bi.bi_idx;
1373 from->count -= copied;
1374 from->iov_offset = bi.bi_bvec_done;
1375
1376 skb->data_len += copied;
1377 skb->len += copied;
1378 skb->truesize += truesize;
1379 return ret;
1380 }
1381
io_send_zc_import(struct io_kiocb * req,struct io_async_msghdr * kmsg)1382 static int io_send_zc_import(struct io_kiocb *req, struct io_async_msghdr *kmsg)
1383 {
1384 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1385 int ret;
1386
1387 if (sr->flags & IORING_RECVSEND_FIXED_BUF) {
1388 ret = io_import_fixed(ITER_SOURCE, &kmsg->msg.msg_iter, req->imu,
1389 (u64)(uintptr_t)sr->buf, sr->len);
1390 if (unlikely(ret))
1391 return ret;
1392 kmsg->msg.sg_from_iter = io_sg_from_iter;
1393 } else {
1394 ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len, &kmsg->msg.msg_iter);
1395 if (unlikely(ret))
1396 return ret;
1397 ret = io_notif_account_mem(sr->notif, sr->len);
1398 if (unlikely(ret))
1399 return ret;
1400 kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1401 }
1402
1403 return ret;
1404 }
1405
io_send_zc(struct io_kiocb * req,unsigned int issue_flags)1406 int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
1407 {
1408 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1409 struct io_async_msghdr *kmsg = req->async_data;
1410 struct socket *sock;
1411 unsigned msg_flags;
1412 int ret, min_ret = 0;
1413
1414 sock = sock_from_file(req->file);
1415 if (unlikely(!sock))
1416 return -ENOTSOCK;
1417 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1418 return -EOPNOTSUPP;
1419
1420 if (!(req->flags & REQ_F_POLLED) &&
1421 (zc->flags & IORING_RECVSEND_POLL_FIRST))
1422 return -EAGAIN;
1423
1424 if (!zc->done_io) {
1425 ret = io_send_zc_import(req, kmsg);
1426 if (unlikely(ret))
1427 return ret;
1428 }
1429
1430 msg_flags = zc->msg_flags;
1431 if (issue_flags & IO_URING_F_NONBLOCK)
1432 msg_flags |= MSG_DONTWAIT;
1433 if (msg_flags & MSG_WAITALL)
1434 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1435 msg_flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
1436
1437 kmsg->msg.msg_flags = msg_flags;
1438 kmsg->msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
1439 ret = sock_sendmsg(sock, &kmsg->msg);
1440
1441 if (unlikely(ret < min_ret)) {
1442 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1443 return -EAGAIN;
1444
1445 if (ret > 0 && io_net_retry(sock, kmsg->msg.msg_flags)) {
1446 zc->len -= ret;
1447 zc->buf += ret;
1448 zc->done_io += ret;
1449 return io_net_kbuf_recyle(req, kmsg, ret);
1450 }
1451 if (ret == -ERESTARTSYS)
1452 ret = -EINTR;
1453 req_set_fail(req);
1454 }
1455
1456 if (ret >= 0)
1457 ret += zc->done_io;
1458 else if (zc->done_io)
1459 ret = zc->done_io;
1460
1461 /*
1462 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1463 * flushing notif to io_send_zc_cleanup()
1464 */
1465 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1466 io_notif_flush(zc->notif);
1467 zc->notif = NULL;
1468 io_req_msg_cleanup(req, 0);
1469 }
1470 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1471 return IOU_OK;
1472 }
1473
io_sendmsg_zc(struct io_kiocb * req,unsigned int issue_flags)1474 int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1475 {
1476 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1477 struct io_async_msghdr *kmsg = req->async_data;
1478 struct socket *sock;
1479 unsigned flags;
1480 int ret, min_ret = 0;
1481
1482 sock = sock_from_file(req->file);
1483 if (unlikely(!sock))
1484 return -ENOTSOCK;
1485 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1486 return -EOPNOTSUPP;
1487
1488 if (!(req->flags & REQ_F_POLLED) &&
1489 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1490 return -EAGAIN;
1491
1492 flags = sr->msg_flags;
1493 if (issue_flags & IO_URING_F_NONBLOCK)
1494 flags |= MSG_DONTWAIT;
1495 if (flags & MSG_WAITALL)
1496 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1497
1498 kmsg->msg.msg_control_user = sr->msg_control;
1499 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1500 kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1501 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1502
1503 if (unlikely(ret < min_ret)) {
1504 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1505 return -EAGAIN;
1506
1507 if (ret > 0 && io_net_retry(sock, flags)) {
1508 sr->done_io += ret;
1509 return io_net_kbuf_recyle(req, kmsg, ret);
1510 }
1511 if (ret == -ERESTARTSYS)
1512 ret = -EINTR;
1513 req_set_fail(req);
1514 }
1515
1516 if (ret >= 0)
1517 ret += sr->done_io;
1518 else if (sr->done_io)
1519 ret = sr->done_io;
1520
1521 /*
1522 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1523 * flushing notif to io_send_zc_cleanup()
1524 */
1525 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1526 io_notif_flush(sr->notif);
1527 sr->notif = NULL;
1528 io_req_msg_cleanup(req, 0);
1529 }
1530 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1531 return IOU_OK;
1532 }
1533
io_sendrecv_fail(struct io_kiocb * req)1534 void io_sendrecv_fail(struct io_kiocb *req)
1535 {
1536 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1537
1538 if (sr->done_io)
1539 req->cqe.res = sr->done_io;
1540
1541 if ((req->flags & REQ_F_NEED_CLEANUP) &&
1542 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1543 req->cqe.flags |= IORING_CQE_F_MORE;
1544 }
1545
1546 #define ACCEPT_FLAGS (IORING_ACCEPT_MULTISHOT | IORING_ACCEPT_DONTWAIT | \
1547 IORING_ACCEPT_POLL_FIRST)
1548
io_accept_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1549 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1550 {
1551 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1552
1553 if (sqe->len || sqe->buf_index)
1554 return -EINVAL;
1555
1556 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1557 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1558 accept->flags = READ_ONCE(sqe->accept_flags);
1559 accept->nofile = rlimit(RLIMIT_NOFILE);
1560 accept->iou_flags = READ_ONCE(sqe->ioprio);
1561 if (accept->iou_flags & ~ACCEPT_FLAGS)
1562 return -EINVAL;
1563
1564 accept->file_slot = READ_ONCE(sqe->file_index);
1565 if (accept->file_slot) {
1566 if (accept->flags & SOCK_CLOEXEC)
1567 return -EINVAL;
1568 if (accept->iou_flags & IORING_ACCEPT_MULTISHOT &&
1569 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1570 return -EINVAL;
1571 }
1572 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1573 return -EINVAL;
1574 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1575 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1576 if (accept->iou_flags & IORING_ACCEPT_MULTISHOT)
1577 req->flags |= REQ_F_APOLL_MULTISHOT;
1578 if (accept->iou_flags & IORING_ACCEPT_DONTWAIT)
1579 req->flags |= REQ_F_NOWAIT;
1580 return 0;
1581 }
1582
io_accept(struct io_kiocb * req,unsigned int issue_flags)1583 int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1584 {
1585 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1586 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1587 bool fixed = !!accept->file_slot;
1588 struct proto_accept_arg arg = {
1589 .flags = force_nonblock ? O_NONBLOCK : 0,
1590 };
1591 struct file *file;
1592 unsigned cflags;
1593 int ret, fd;
1594
1595 if (!(req->flags & REQ_F_POLLED) &&
1596 accept->iou_flags & IORING_ACCEPT_POLL_FIRST)
1597 return -EAGAIN;
1598
1599 retry:
1600 if (!fixed) {
1601 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1602 if (unlikely(fd < 0))
1603 return fd;
1604 }
1605 arg.err = 0;
1606 arg.is_empty = -1;
1607 file = do_accept(req->file, &arg, accept->addr, accept->addr_len,
1608 accept->flags);
1609 if (IS_ERR(file)) {
1610 if (!fixed)
1611 put_unused_fd(fd);
1612 ret = PTR_ERR(file);
1613 if (ret == -EAGAIN && force_nonblock &&
1614 !(accept->iou_flags & IORING_ACCEPT_DONTWAIT)) {
1615 /*
1616 * if it's multishot and polled, we don't need to
1617 * return EAGAIN to arm the poll infra since it
1618 * has already been done
1619 */
1620 if (issue_flags & IO_URING_F_MULTISHOT)
1621 return IOU_ISSUE_SKIP_COMPLETE;
1622 return ret;
1623 }
1624 if (ret == -ERESTARTSYS)
1625 ret = -EINTR;
1626 req_set_fail(req);
1627 } else if (!fixed) {
1628 fd_install(fd, file);
1629 ret = fd;
1630 } else {
1631 ret = io_fixed_fd_install(req, issue_flags, file,
1632 accept->file_slot);
1633 }
1634
1635 cflags = 0;
1636 if (!arg.is_empty)
1637 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
1638
1639 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1640 io_req_set_res(req, ret, cflags);
1641 return IOU_OK;
1642 }
1643
1644 if (ret < 0)
1645 return ret;
1646 if (io_req_post_cqe(req, ret, cflags | IORING_CQE_F_MORE)) {
1647 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || arg.is_empty == -1)
1648 goto retry;
1649 if (issue_flags & IO_URING_F_MULTISHOT)
1650 return IOU_ISSUE_SKIP_COMPLETE;
1651 return -EAGAIN;
1652 }
1653
1654 io_req_set_res(req, ret, cflags);
1655 if (!(issue_flags & IO_URING_F_MULTISHOT))
1656 return IOU_OK;
1657 return IOU_STOP_MULTISHOT;
1658 }
1659
io_socket_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1660 int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1661 {
1662 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1663
1664 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1665 return -EINVAL;
1666
1667 sock->domain = READ_ONCE(sqe->fd);
1668 sock->type = READ_ONCE(sqe->off);
1669 sock->protocol = READ_ONCE(sqe->len);
1670 sock->file_slot = READ_ONCE(sqe->file_index);
1671 sock->nofile = rlimit(RLIMIT_NOFILE);
1672
1673 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1674 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1675 return -EINVAL;
1676 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1677 return -EINVAL;
1678 return 0;
1679 }
1680
io_socket(struct io_kiocb * req,unsigned int issue_flags)1681 int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1682 {
1683 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1684 bool fixed = !!sock->file_slot;
1685 struct file *file;
1686 int ret, fd;
1687
1688 if (!fixed) {
1689 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1690 if (unlikely(fd < 0))
1691 return fd;
1692 }
1693 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1694 if (IS_ERR(file)) {
1695 if (!fixed)
1696 put_unused_fd(fd);
1697 ret = PTR_ERR(file);
1698 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1699 return -EAGAIN;
1700 if (ret == -ERESTARTSYS)
1701 ret = -EINTR;
1702 req_set_fail(req);
1703 } else if (!fixed) {
1704 fd_install(fd, file);
1705 ret = fd;
1706 } else {
1707 ret = io_fixed_fd_install(req, issue_flags, file,
1708 sock->file_slot);
1709 }
1710 io_req_set_res(req, ret, 0);
1711 return IOU_OK;
1712 }
1713
io_connect_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1714 int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1715 {
1716 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1717 struct io_async_msghdr *io;
1718
1719 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1720 return -EINVAL;
1721
1722 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1723 conn->addr_len = READ_ONCE(sqe->addr2);
1724 conn->in_progress = conn->seen_econnaborted = false;
1725
1726 io = io_msg_alloc_async(req);
1727 if (unlikely(!io))
1728 return -ENOMEM;
1729
1730 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->addr);
1731 }
1732
io_connect(struct io_kiocb * req,unsigned int issue_flags)1733 int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1734 {
1735 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1736 struct io_async_msghdr *io = req->async_data;
1737 unsigned file_flags;
1738 int ret;
1739 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1740
1741 if (connect->in_progress) {
1742 struct poll_table_struct pt = { ._key = EPOLLERR };
1743
1744 if (vfs_poll(req->file, &pt) & EPOLLERR)
1745 goto get_sock_err;
1746 }
1747
1748 file_flags = force_nonblock ? O_NONBLOCK : 0;
1749
1750 ret = __sys_connect_file(req->file, &io->addr, connect->addr_len,
1751 file_flags);
1752 if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED)
1753 && force_nonblock) {
1754 if (ret == -EINPROGRESS) {
1755 connect->in_progress = true;
1756 } else if (ret == -ECONNABORTED) {
1757 if (connect->seen_econnaborted)
1758 goto out;
1759 connect->seen_econnaborted = true;
1760 }
1761 return -EAGAIN;
1762 }
1763 if (connect->in_progress) {
1764 /*
1765 * At least bluetooth will return -EBADFD on a re-connect
1766 * attempt, and it's (supposedly) also valid to get -EISCONN
1767 * which means the previous result is good. For both of these,
1768 * grab the sock_error() and use that for the completion.
1769 */
1770 if (ret == -EBADFD || ret == -EISCONN) {
1771 get_sock_err:
1772 ret = sock_error(sock_from_file(req->file)->sk);
1773 }
1774 }
1775 if (ret == -ERESTARTSYS)
1776 ret = -EINTR;
1777 out:
1778 if (ret < 0)
1779 req_set_fail(req);
1780 io_req_msg_cleanup(req, issue_flags);
1781 io_req_set_res(req, ret, 0);
1782 return IOU_OK;
1783 }
1784
io_bind_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1785 int io_bind_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1786 {
1787 struct io_bind *bind = io_kiocb_to_cmd(req, struct io_bind);
1788 struct sockaddr __user *uaddr;
1789 struct io_async_msghdr *io;
1790
1791 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1792 return -EINVAL;
1793
1794 uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1795 bind->addr_len = READ_ONCE(sqe->addr2);
1796
1797 io = io_msg_alloc_async(req);
1798 if (unlikely(!io))
1799 return -ENOMEM;
1800 return move_addr_to_kernel(uaddr, bind->addr_len, &io->addr);
1801 }
1802
io_bind(struct io_kiocb * req,unsigned int issue_flags)1803 int io_bind(struct io_kiocb *req, unsigned int issue_flags)
1804 {
1805 struct io_bind *bind = io_kiocb_to_cmd(req, struct io_bind);
1806 struct io_async_msghdr *io = req->async_data;
1807 struct socket *sock;
1808 int ret;
1809
1810 sock = sock_from_file(req->file);
1811 if (unlikely(!sock))
1812 return -ENOTSOCK;
1813
1814 ret = __sys_bind_socket(sock, &io->addr, bind->addr_len);
1815 if (ret < 0)
1816 req_set_fail(req);
1817 io_req_set_res(req, ret, 0);
1818 return 0;
1819 }
1820
io_listen_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1821 int io_listen_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1822 {
1823 struct io_listen *listen = io_kiocb_to_cmd(req, struct io_listen);
1824
1825 if (sqe->addr || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in || sqe->addr2)
1826 return -EINVAL;
1827
1828 listen->backlog = READ_ONCE(sqe->len);
1829 return 0;
1830 }
1831
io_listen(struct io_kiocb * req,unsigned int issue_flags)1832 int io_listen(struct io_kiocb *req, unsigned int issue_flags)
1833 {
1834 struct io_listen *listen = io_kiocb_to_cmd(req, struct io_listen);
1835 struct socket *sock;
1836 int ret;
1837
1838 sock = sock_from_file(req->file);
1839 if (unlikely(!sock))
1840 return -ENOTSOCK;
1841
1842 ret = __sys_listen_socket(sock, listen->backlog);
1843 if (ret < 0)
1844 req_set_fail(req);
1845 io_req_set_res(req, ret, 0);
1846 return 0;
1847 }
1848
io_netmsg_cache_free(const void * entry)1849 void io_netmsg_cache_free(const void *entry)
1850 {
1851 struct io_async_msghdr *kmsg = (struct io_async_msghdr *) entry;
1852
1853 if (kmsg->free_iov) {
1854 kasan_mempool_unpoison_object(kmsg->free_iov,
1855 kmsg->free_iov_nr * sizeof(struct iovec));
1856 io_netmsg_iovec_free(kmsg);
1857 }
1858 kfree(kmsg);
1859 }
1860 #endif
1861