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 "filetable.h"
14 #include "io_uring.h"
15 #include "kbuf.h"
16 #include "alloc_cache.h"
17 #include "net.h"
18 #include "notif.h"
19 #include "rsrc.h"
20 #include "zcrx.h"
21
22 struct io_shutdown {
23 struct file *file;
24 int how;
25 };
26
27 struct io_accept {
28 struct file *file;
29 struct sockaddr __user *addr;
30 int __user *addr_len;
31 int flags;
32 int iou_flags;
33 u32 file_slot;
34 unsigned long nofile;
35 };
36
37 struct io_socket {
38 struct file *file;
39 int domain;
40 int type;
41 int protocol;
42 int flags;
43 u32 file_slot;
44 unsigned long nofile;
45 };
46
47 struct io_connect {
48 struct file *file;
49 struct sockaddr __user *addr;
50 int addr_len;
51 bool in_progress;
52 bool seen_econnaborted;
53 };
54
55 struct io_bind {
56 struct file *file;
57 int addr_len;
58 };
59
60 struct io_listen {
61 struct file *file;
62 int backlog;
63 };
64
65 struct io_sr_msg {
66 struct file *file;
67 union {
68 struct compat_msghdr __user *umsg_compat;
69 struct user_msghdr __user *umsg;
70 void __user *buf;
71 };
72 int len;
73 unsigned done_io;
74 unsigned msg_flags;
75 unsigned nr_multishot_loops;
76 u16 flags;
77 /* initialised and used only by !msg send variants */
78 u16 buf_group;
79 /* per-invocation mshot limit */
80 unsigned mshot_len;
81 /* overall mshot byte limit */
82 unsigned mshot_total_len;
83 void __user *msg_control;
84 /* used only for send zerocopy */
85 struct io_kiocb *notif;
86 };
87
88 /*
89 * The UAPI flags are the lower 8 bits, as that's all sqe->ioprio will hold
90 * anyway. Use the upper 8 bits for internal uses.
91 */
92 enum sr_retry_flags {
93 IORING_RECV_RETRY = (1U << 15),
94 IORING_RECV_PARTIAL_MAP = (1U << 14),
95 IORING_RECV_MSHOT_CAP = (1U << 13),
96 IORING_RECV_MSHOT_LIM = (1U << 12),
97 IORING_RECV_MSHOT_DONE = (1U << 11),
98
99 IORING_RECV_RETRY_CLEAR = IORING_RECV_RETRY | IORING_RECV_PARTIAL_MAP,
100 IORING_RECV_NO_RETRY = IORING_RECV_RETRY | IORING_RECV_PARTIAL_MAP |
101 IORING_RECV_MSHOT_CAP | IORING_RECV_MSHOT_DONE,
102 };
103
104 /*
105 * Number of times we'll try and do receives if there's more data. If we
106 * exceed this limit, then add us to the back of the queue and retry from
107 * there. This helps fairness between flooding clients.
108 */
109 #define MULTISHOT_MAX_RETRY 32
110
111 struct io_recvzc {
112 struct file *file;
113 u16 flags;
114 u32 len;
115 struct io_zcrx_ifq *ifq;
116 };
117
118 static int io_sg_from_iter_iovec(struct sk_buff *skb,
119 struct iov_iter *from, size_t length);
120 static int io_sg_from_iter(struct sk_buff *skb,
121 struct iov_iter *from, size_t length);
122
io_shutdown_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)123 int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
124 {
125 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
126
127 if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
128 sqe->buf_index || sqe->splice_fd_in))
129 return -EINVAL;
130
131 shutdown->how = READ_ONCE(sqe->len);
132 req->flags |= REQ_F_FORCE_ASYNC;
133 return 0;
134 }
135
io_shutdown(struct io_kiocb * req,unsigned int issue_flags)136 int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
137 {
138 struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
139 struct socket *sock;
140 int ret;
141
142 WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
143
144 sock = sock_from_file(req->file);
145 if (unlikely(!sock))
146 return -ENOTSOCK;
147
148 ret = __sys_shutdown_sock(sock, shutdown->how);
149 io_req_set_res(req, ret, 0);
150 return IOU_COMPLETE;
151 }
152
io_net_retry(struct socket * sock,int flags)153 static bool io_net_retry(struct socket *sock, int flags)
154 {
155 if (!(flags & MSG_WAITALL))
156 return false;
157 return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
158 }
159
io_netmsg_iovec_free(struct io_async_msghdr * kmsg)160 static void io_netmsg_iovec_free(struct io_async_msghdr *kmsg)
161 {
162 if (kmsg->vec.iovec)
163 io_vec_free(&kmsg->vec);
164 }
165
io_netmsg_recycle(struct io_kiocb * req,unsigned int issue_flags)166 static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
167 {
168 struct io_async_msghdr *hdr = req->async_data;
169
170 /* can't recycle, ensure we free the iovec if we have one */
171 if (unlikely(issue_flags & IO_URING_F_UNLOCKED)) {
172 io_netmsg_iovec_free(hdr);
173 return;
174 }
175
176 /* Let normal cleanup path reap it if we fail adding to the cache */
177 io_alloc_cache_vec_kasan(&hdr->vec);
178 if (hdr->vec.nr > IO_VEC_CACHE_SOFT_CAP)
179 io_vec_free(&hdr->vec);
180
181 if (io_alloc_cache_put(&req->ctx->netmsg_cache, hdr))
182 io_req_async_data_clear(req, REQ_F_NEED_CLEANUP);
183 }
184
io_msg_alloc_async(struct io_kiocb * req)185 static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req)
186 {
187 struct io_ring_ctx *ctx = req->ctx;
188 struct io_async_msghdr *hdr;
189
190 hdr = io_uring_alloc_async_data(&ctx->netmsg_cache, req);
191 if (!hdr)
192 return NULL;
193
194 /* If the async data was cached, we might have an iov cached inside. */
195 if (hdr->vec.iovec)
196 req->flags |= REQ_F_NEED_CLEANUP;
197 return hdr;
198 }
199
io_mshot_prep_retry(struct io_kiocb * req,struct io_async_msghdr * kmsg)200 static inline void io_mshot_prep_retry(struct io_kiocb *req,
201 struct io_async_msghdr *kmsg)
202 {
203 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
204
205 req->flags &= ~REQ_F_BL_EMPTY;
206 sr->done_io = 0;
207 sr->flags &= ~IORING_RECV_RETRY_CLEAR;
208 sr->len = sr->mshot_len;
209 }
210
io_net_import_vec(struct io_kiocb * req,struct io_async_msghdr * iomsg,const struct iovec __user * uiov,unsigned uvec_seg,int ddir)211 static int io_net_import_vec(struct io_kiocb *req, struct io_async_msghdr *iomsg,
212 const struct iovec __user *uiov, unsigned uvec_seg,
213 int ddir)
214 {
215 struct iovec *iov;
216 int ret, nr_segs;
217
218 if (iomsg->vec.iovec) {
219 nr_segs = iomsg->vec.nr;
220 iov = iomsg->vec.iovec;
221 } else {
222 nr_segs = 1;
223 iov = &iomsg->fast_iov;
224 }
225
226 ret = __import_iovec(ddir, uiov, uvec_seg, nr_segs, &iov,
227 &iomsg->msg.msg_iter, io_is_compat(req->ctx));
228 if (unlikely(ret < 0))
229 return ret;
230
231 if (iov) {
232 req->flags |= REQ_F_NEED_CLEANUP;
233 io_vec_reset_iovec(&iomsg->vec, iov, iomsg->msg.msg_iter.nr_segs);
234 }
235 return 0;
236 }
237
io_compat_msg_copy_hdr(struct io_kiocb * req,struct io_async_msghdr * iomsg,struct compat_msghdr * msg,int ddir,struct sockaddr __user ** save_addr)238 static int io_compat_msg_copy_hdr(struct io_kiocb *req,
239 struct io_async_msghdr *iomsg,
240 struct compat_msghdr *msg, int ddir,
241 struct sockaddr __user **save_addr)
242 {
243 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
244 struct compat_iovec __user *uiov;
245 int ret;
246
247 if (copy_from_user(msg, sr->umsg_compat, sizeof(*msg)))
248 return -EFAULT;
249
250 ret = __get_compat_msghdr(&iomsg->msg, msg, save_addr);
251 if (ret)
252 return ret;
253
254 uiov = compat_ptr(msg->msg_iov);
255 if (req->flags & REQ_F_BUFFER_SELECT) {
256 if (msg->msg_iovlen == 0) {
257 sr->len = 0;
258 } else if (msg->msg_iovlen > 1) {
259 return -EINVAL;
260 } else {
261 struct compat_iovec tmp_iov;
262
263 if (copy_from_user(&tmp_iov, uiov, sizeof(tmp_iov)))
264 return -EFAULT;
265 sr->len = tmp_iov.iov_len;
266 }
267 }
268 return 0;
269 }
270
io_copy_msghdr_from_user(struct user_msghdr * msg,struct user_msghdr __user * umsg)271 static int io_copy_msghdr_from_user(struct user_msghdr *msg,
272 struct user_msghdr __user *umsg)
273 {
274 if (!user_access_begin(umsg, sizeof(*umsg)))
275 return -EFAULT;
276 unsafe_get_user(msg->msg_name, &umsg->msg_name, ua_end);
277 unsafe_get_user(msg->msg_namelen, &umsg->msg_namelen, ua_end);
278 unsafe_get_user(msg->msg_iov, &umsg->msg_iov, ua_end);
279 unsafe_get_user(msg->msg_iovlen, &umsg->msg_iovlen, ua_end);
280 unsafe_get_user(msg->msg_control, &umsg->msg_control, ua_end);
281 unsafe_get_user(msg->msg_controllen, &umsg->msg_controllen, ua_end);
282 user_access_end();
283 return 0;
284 ua_end:
285 user_access_end();
286 return -EFAULT;
287 }
288
io_msg_copy_hdr(struct io_kiocb * req,struct io_async_msghdr * iomsg,struct user_msghdr * msg,int ddir,struct sockaddr __user ** save_addr)289 static int io_msg_copy_hdr(struct io_kiocb *req, struct io_async_msghdr *iomsg,
290 struct user_msghdr *msg, int ddir,
291 struct sockaddr __user **save_addr)
292 {
293 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
294 struct user_msghdr __user *umsg = sr->umsg;
295 int ret;
296
297 iomsg->msg.msg_name = &iomsg->addr;
298 iomsg->msg.msg_iter.nr_segs = 0;
299
300 if (io_is_compat(req->ctx)) {
301 struct compat_msghdr cmsg;
302
303 ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ddir, save_addr);
304 if (ret)
305 return ret;
306
307 memset(msg, 0, sizeof(*msg));
308 msg->msg_namelen = cmsg.msg_namelen;
309 msg->msg_controllen = cmsg.msg_controllen;
310 msg->msg_iov = compat_ptr(cmsg.msg_iov);
311 msg->msg_iovlen = cmsg.msg_iovlen;
312 return 0;
313 }
314
315 ret = io_copy_msghdr_from_user(msg, umsg);
316 if (unlikely(ret))
317 return ret;
318
319 msg->msg_flags = 0;
320
321 ret = __copy_msghdr(&iomsg->msg, msg, save_addr);
322 if (ret)
323 return ret;
324
325 if (req->flags & REQ_F_BUFFER_SELECT) {
326 if (msg->msg_iovlen == 0) {
327 sr->len = 0;
328 } else if (msg->msg_iovlen > 1) {
329 return -EINVAL;
330 } else {
331 struct iovec __user *uiov = msg->msg_iov;
332 struct iovec tmp_iov;
333
334 if (copy_from_user(&tmp_iov, uiov, sizeof(tmp_iov)))
335 return -EFAULT;
336 sr->len = tmp_iov.iov_len;
337 }
338 }
339 return 0;
340 }
341
io_sendmsg_recvmsg_cleanup(struct io_kiocb * req)342 void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
343 {
344 struct io_async_msghdr *io = req->async_data;
345
346 io_netmsg_iovec_free(io);
347 }
348
io_send_setup(struct io_kiocb * req,const struct io_uring_sqe * sqe)349 static int io_send_setup(struct io_kiocb *req, const struct io_uring_sqe *sqe)
350 {
351 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
352 struct io_async_msghdr *kmsg = req->async_data;
353 void __user *addr;
354 u16 addr_len;
355 int ret;
356
357 sr->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
358
359 if (READ_ONCE(sqe->__pad3[0]))
360 return -EINVAL;
361
362 kmsg->msg.msg_name = NULL;
363 kmsg->msg.msg_namelen = 0;
364 kmsg->msg.msg_control = NULL;
365 kmsg->msg.msg_controllen = 0;
366 kmsg->msg.msg_ubuf = NULL;
367
368 addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
369 addr_len = READ_ONCE(sqe->addr_len);
370 if (addr) {
371 ret = move_addr_to_kernel(addr, addr_len, &kmsg->addr);
372 if (unlikely(ret < 0))
373 return ret;
374 kmsg->msg.msg_name = &kmsg->addr;
375 kmsg->msg.msg_namelen = addr_len;
376 }
377 if (sr->flags & IORING_RECVSEND_FIXED_BUF) {
378 if (!(sr->flags & IORING_SEND_VECTORIZED)) {
379 req->flags |= REQ_F_IMPORT_BUFFER;
380 return 0;
381 }
382
383 kmsg->msg.msg_iter.nr_segs = sr->len;
384 return io_prep_reg_iovec(req, &kmsg->vec, sr->buf, sr->len);
385 }
386 if (req->flags & REQ_F_BUFFER_SELECT)
387 return 0;
388
389 if (sr->flags & IORING_SEND_VECTORIZED)
390 return io_net_import_vec(req, kmsg, sr->buf, sr->len, ITER_SOURCE);
391
392 return import_ubuf(ITER_SOURCE, sr->buf, sr->len, &kmsg->msg.msg_iter);
393 }
394
io_sendmsg_setup(struct io_kiocb * req,const struct io_uring_sqe * sqe)395 static int io_sendmsg_setup(struct io_kiocb *req, const struct io_uring_sqe *sqe)
396 {
397 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
398 struct io_async_msghdr *kmsg = req->async_data;
399 struct user_msghdr msg;
400 int ret;
401
402 sr->flags |= IORING_SEND_VECTORIZED;
403 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
404 ret = io_msg_copy_hdr(req, kmsg, &msg, ITER_SOURCE, NULL);
405 if (unlikely(ret))
406 return ret;
407 /* save msg_control as sys_sendmsg() overwrites it */
408 sr->msg_control = kmsg->msg.msg_control_user;
409
410 if (sr->flags & IORING_RECVSEND_FIXED_BUF) {
411 kmsg->msg.msg_iter.nr_segs = msg.msg_iovlen;
412 return io_prep_reg_iovec(req, &kmsg->vec, msg.msg_iov,
413 msg.msg_iovlen);
414 }
415 if (req->flags & REQ_F_BUFFER_SELECT)
416 return 0;
417 return io_net_import_vec(req, kmsg, msg.msg_iov, msg.msg_iovlen, ITER_SOURCE);
418 }
419
420 #define SENDMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_BUNDLE | IORING_SEND_VECTORIZED)
421
io_sendmsg_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)422 int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
423 {
424 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
425
426 sr->done_io = 0;
427 sr->len = READ_ONCE(sqe->len);
428 if (unlikely(sr->len < 0))
429 return -EINVAL;
430 sr->flags = READ_ONCE(sqe->ioprio);
431 if (sr->flags & ~SENDMSG_FLAGS)
432 return -EINVAL;
433 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
434 if (sr->msg_flags & MSG_DONTWAIT)
435 req->flags |= REQ_F_NOWAIT;
436 if (req->flags & REQ_F_BUFFER_SELECT)
437 sr->buf_group = req->buf_index;
438 if (sr->flags & IORING_RECVSEND_BUNDLE) {
439 if (req->opcode == IORING_OP_SENDMSG)
440 return -EINVAL;
441 sr->msg_flags |= MSG_WAITALL;
442 req->flags |= REQ_F_MULTISHOT;
443 }
444
445 if (io_is_compat(req->ctx))
446 sr->msg_flags |= MSG_CMSG_COMPAT;
447
448 if (unlikely(!io_msg_alloc_async(req)))
449 return -ENOMEM;
450 if (req->opcode != IORING_OP_SENDMSG)
451 return io_send_setup(req, sqe);
452 if (unlikely(sqe->addr2 || sqe->file_index))
453 return -EINVAL;
454 return io_sendmsg_setup(req, sqe);
455 }
456
io_req_msg_cleanup(struct io_kiocb * req,unsigned int issue_flags)457 static void io_req_msg_cleanup(struct io_kiocb *req,
458 unsigned int issue_flags)
459 {
460 io_netmsg_recycle(req, issue_flags);
461 }
462
463 /*
464 * For bundle completions, we need to figure out how many segments we consumed.
465 * A bundle could be using a single ITER_UBUF if that's all we mapped, or it
466 * could be using an ITER_IOVEC. If the latter, then if we consumed all of
467 * the segments, then it's a trivial questiont o answer. If we have residual
468 * data in the iter, then loop the segments to figure out how much we
469 * transferred.
470 */
io_bundle_nbufs(struct io_async_msghdr * kmsg,int ret)471 static int io_bundle_nbufs(struct io_async_msghdr *kmsg, int ret)
472 {
473 struct iovec *iov;
474 int nbufs;
475
476 /* no data is always zero segments, and a ubuf is always 1 segment */
477 if (ret <= 0)
478 return 0;
479 if (iter_is_ubuf(&kmsg->msg.msg_iter))
480 return 1;
481
482 iov = kmsg->vec.iovec;
483 if (!iov)
484 iov = &kmsg->fast_iov;
485
486 /* if all data was transferred, it's basic pointer math */
487 if (!iov_iter_count(&kmsg->msg.msg_iter))
488 return iter_iov(&kmsg->msg.msg_iter) - iov;
489
490 /* short transfer, count segments */
491 nbufs = 0;
492 do {
493 int this_len = min_t(int, iov[nbufs].iov_len, ret);
494
495 nbufs++;
496 ret -= this_len;
497 } while (ret);
498
499 return nbufs;
500 }
501
io_net_kbuf_recyle(struct io_kiocb * req,struct io_buffer_list * bl,struct io_async_msghdr * kmsg,int len)502 static int io_net_kbuf_recyle(struct io_kiocb *req, struct io_buffer_list *bl,
503 struct io_async_msghdr *kmsg, int len)
504 {
505 req->flags |= REQ_F_BL_NO_RECYCLE;
506 if (req->flags & REQ_F_BUFFERS_COMMIT)
507 io_kbuf_commit(req, bl, len, io_bundle_nbufs(kmsg, len));
508 return IOU_RETRY;
509 }
510
io_send_finish(struct io_kiocb * req,struct io_async_msghdr * kmsg,struct io_br_sel * sel)511 static inline bool io_send_finish(struct io_kiocb *req,
512 struct io_async_msghdr *kmsg,
513 struct io_br_sel *sel)
514 {
515 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
516 bool bundle_finished = sel->val <= 0;
517 unsigned int cflags;
518
519 if (!(sr->flags & IORING_RECVSEND_BUNDLE)) {
520 cflags = io_put_kbuf(req, sel->val, sel->buf_list);
521 goto finish;
522 }
523
524 cflags = io_put_kbufs(req, sel->val, sel->buf_list, io_bundle_nbufs(kmsg, sel->val));
525
526 /*
527 * Don't start new bundles if the buffer list is empty, or if the
528 * current operation needed to go through polling to complete.
529 */
530 if (bundle_finished || req->flags & (REQ_F_BL_EMPTY | REQ_F_POLLED))
531 goto finish;
532
533 /*
534 * Fill CQE for this receive and see if we should keep trying to
535 * receive from this socket.
536 */
537 if (io_req_post_cqe(req, sel->val, cflags | IORING_CQE_F_MORE)) {
538 io_mshot_prep_retry(req, kmsg);
539 return false;
540 }
541
542 /* Otherwise stop bundle and use the current result. */
543 finish:
544 io_req_set_res(req, sel->val, cflags);
545 sel->val = IOU_COMPLETE;
546 return true;
547 }
548
io_sendmsg(struct io_kiocb * req,unsigned int issue_flags)549 int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
550 {
551 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
552 struct io_async_msghdr *kmsg = req->async_data;
553 struct socket *sock;
554 unsigned flags;
555 int min_ret = 0;
556 int ret;
557
558 sock = sock_from_file(req->file);
559 if (unlikely(!sock))
560 return -ENOTSOCK;
561
562 if (!(req->flags & REQ_F_POLLED) &&
563 (sr->flags & IORING_RECVSEND_POLL_FIRST))
564 return -EAGAIN;
565
566 flags = sr->msg_flags;
567 if (issue_flags & IO_URING_F_NONBLOCK)
568 flags |= MSG_DONTWAIT;
569 if (flags & MSG_WAITALL)
570 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
571
572 kmsg->msg.msg_control_user = sr->msg_control;
573
574 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
575
576 if (ret < min_ret) {
577 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
578 return -EAGAIN;
579 if (ret > 0 && io_net_retry(sock, flags)) {
580 kmsg->msg.msg_controllen = 0;
581 kmsg->msg.msg_control = NULL;
582 sr->done_io += ret;
583 return -EAGAIN;
584 }
585 if (ret == -ERESTARTSYS)
586 ret = -EINTR;
587 req_set_fail(req);
588 }
589 io_req_msg_cleanup(req, issue_flags);
590 if (ret >= 0)
591 ret += sr->done_io;
592 else if (sr->done_io)
593 ret = sr->done_io;
594 io_req_set_res(req, ret, 0);
595 return IOU_COMPLETE;
596 }
597
io_send_select_buffer(struct io_kiocb * req,unsigned int issue_flags,struct io_br_sel * sel,struct io_async_msghdr * kmsg)598 static int io_send_select_buffer(struct io_kiocb *req, unsigned int issue_flags,
599 struct io_br_sel *sel, struct io_async_msghdr *kmsg)
600 {
601 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
602 struct buf_sel_arg arg = {
603 .iovs = &kmsg->fast_iov,
604 .max_len = min_not_zero(sr->len, INT_MAX),
605 .nr_iovs = 1,
606 .buf_group = sr->buf_group,
607 };
608 int ret;
609
610 if (kmsg->vec.iovec) {
611 arg.nr_iovs = kmsg->vec.nr;
612 arg.iovs = kmsg->vec.iovec;
613 arg.mode = KBUF_MODE_FREE;
614 }
615
616 if (!(sr->flags & IORING_RECVSEND_BUNDLE))
617 arg.nr_iovs = 1;
618 else
619 arg.mode |= KBUF_MODE_EXPAND;
620
621 ret = io_buffers_select(req, &arg, sel, issue_flags);
622 if (unlikely(ret < 0))
623 return ret;
624
625 if (arg.iovs != &kmsg->fast_iov && arg.iovs != kmsg->vec.iovec) {
626 kmsg->vec.nr = ret;
627 kmsg->vec.iovec = arg.iovs;
628 req->flags |= REQ_F_NEED_CLEANUP;
629 }
630 sr->len = arg.out_len;
631
632 if (ret == 1) {
633 sr->buf = arg.iovs[0].iov_base;
634 ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len,
635 &kmsg->msg.msg_iter);
636 if (unlikely(ret))
637 return ret;
638 } else {
639 iov_iter_init(&kmsg->msg.msg_iter, ITER_SOURCE,
640 arg.iovs, ret, arg.out_len);
641 }
642
643 return 0;
644 }
645
io_send(struct io_kiocb * req,unsigned int issue_flags)646 int io_send(struct io_kiocb *req, unsigned int issue_flags)
647 {
648 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
649 struct io_async_msghdr *kmsg = req->async_data;
650 struct io_br_sel sel = { };
651 struct socket *sock;
652 unsigned flags;
653 int min_ret = 0;
654 int ret;
655
656 sock = sock_from_file(req->file);
657 if (unlikely(!sock))
658 return -ENOTSOCK;
659
660 if (!(req->flags & REQ_F_POLLED) &&
661 (sr->flags & IORING_RECVSEND_POLL_FIRST))
662 return -EAGAIN;
663
664 flags = sr->msg_flags;
665 if (issue_flags & IO_URING_F_NONBLOCK)
666 flags |= MSG_DONTWAIT;
667
668 retry_bundle:
669 sel.buf_list = NULL;
670 if (io_do_buffer_select(req)) {
671 ret = io_send_select_buffer(req, issue_flags, &sel, kmsg);
672 if (ret)
673 return ret;
674 }
675
676 /*
677 * If MSG_WAITALL is set, or this is a bundle send, then we need
678 * the full amount. If just bundle is set, if we do a short send
679 * then we complete the bundle sequence rather than continue on.
680 */
681 if (flags & MSG_WAITALL || sr->flags & IORING_RECVSEND_BUNDLE)
682 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
683
684 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
685 kmsg->msg.msg_flags = flags;
686 ret = sock_sendmsg(sock, &kmsg->msg);
687 if (ret < min_ret) {
688 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
689 return -EAGAIN;
690
691 if (ret > 0 && io_net_retry(sock, flags)) {
692 sr->len -= ret;
693 sr->buf += ret;
694 sr->done_io += ret;
695 return io_net_kbuf_recyle(req, sel.buf_list, kmsg, ret);
696 }
697 if (ret == -ERESTARTSYS)
698 ret = -EINTR;
699 req_set_fail(req);
700 }
701 if (ret >= 0)
702 ret += sr->done_io;
703 else if (sr->done_io)
704 ret = sr->done_io;
705
706 sel.val = ret;
707 if (!io_send_finish(req, kmsg, &sel))
708 goto retry_bundle;
709
710 io_req_msg_cleanup(req, issue_flags);
711 return sel.val;
712 }
713
io_recvmsg_mshot_prep(struct io_kiocb * req,struct io_async_msghdr * iomsg,int namelen,size_t controllen)714 static int io_recvmsg_mshot_prep(struct io_kiocb *req,
715 struct io_async_msghdr *iomsg,
716 int namelen, size_t controllen)
717 {
718 if ((req->flags & (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) ==
719 (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) {
720 int hdr;
721
722 if (unlikely(namelen < 0))
723 return -EOVERFLOW;
724 if (check_add_overflow(sizeof(struct io_uring_recvmsg_out),
725 namelen, &hdr))
726 return -EOVERFLOW;
727 if (check_add_overflow(hdr, controllen, &hdr))
728 return -EOVERFLOW;
729
730 iomsg->namelen = namelen;
731 iomsg->controllen = controllen;
732 return 0;
733 }
734
735 return 0;
736 }
737
io_recvmsg_copy_hdr(struct io_kiocb * req,struct io_async_msghdr * iomsg)738 static int io_recvmsg_copy_hdr(struct io_kiocb *req,
739 struct io_async_msghdr *iomsg)
740 {
741 struct user_msghdr msg;
742 int ret;
743
744 ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_DEST, &iomsg->uaddr);
745 if (unlikely(ret))
746 return ret;
747
748 if (!(req->flags & REQ_F_BUFFER_SELECT)) {
749 ret = io_net_import_vec(req, iomsg, msg.msg_iov, msg.msg_iovlen,
750 ITER_DEST);
751 if (unlikely(ret))
752 return ret;
753 }
754 return io_recvmsg_mshot_prep(req, iomsg, msg.msg_namelen,
755 msg.msg_controllen);
756 }
757
io_recvmsg_prep_setup(struct io_kiocb * req)758 static int io_recvmsg_prep_setup(struct io_kiocb *req)
759 {
760 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
761 struct io_async_msghdr *kmsg;
762
763 kmsg = io_msg_alloc_async(req);
764 if (unlikely(!kmsg))
765 return -ENOMEM;
766
767 if (req->opcode == IORING_OP_RECV) {
768 kmsg->msg.msg_name = NULL;
769 kmsg->msg.msg_namelen = 0;
770 kmsg->msg.msg_inq = 0;
771 kmsg->msg.msg_control = NULL;
772 kmsg->msg.msg_get_inq = 1;
773 kmsg->msg.msg_controllen = 0;
774 kmsg->msg.msg_iocb = NULL;
775 kmsg->msg.msg_ubuf = NULL;
776
777 if (req->flags & REQ_F_BUFFER_SELECT)
778 return 0;
779 return import_ubuf(ITER_DEST, sr->buf, sr->len,
780 &kmsg->msg.msg_iter);
781 }
782
783 return io_recvmsg_copy_hdr(req, kmsg);
784 }
785
786 #define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT | \
787 IORING_RECVSEND_BUNDLE)
788
io_recvmsg_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)789 int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
790 {
791 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
792
793 sr->done_io = 0;
794
795 if (unlikely(sqe->addr2))
796 return -EINVAL;
797
798 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
799 sr->len = READ_ONCE(sqe->len);
800 if (unlikely(sr->len < 0))
801 return -EINVAL;
802 sr->flags = READ_ONCE(sqe->ioprio);
803 if (sr->flags & ~RECVMSG_FLAGS)
804 return -EINVAL;
805 sr->msg_flags = READ_ONCE(sqe->msg_flags);
806 if (sr->msg_flags & MSG_DONTWAIT)
807 req->flags |= REQ_F_NOWAIT;
808 if (sr->msg_flags & MSG_ERRQUEUE)
809 req->flags |= REQ_F_CLEAR_POLLIN;
810 if (req->flags & REQ_F_BUFFER_SELECT)
811 sr->buf_group = req->buf_index;
812 sr->mshot_total_len = sr->mshot_len = 0;
813 if (sr->flags & IORING_RECV_MULTISHOT) {
814 if (!(req->flags & REQ_F_BUFFER_SELECT))
815 return -EINVAL;
816 if (sr->msg_flags & MSG_WAITALL)
817 return -EINVAL;
818 if (req->opcode == IORING_OP_RECV) {
819 sr->mshot_len = sr->len;
820 sr->mshot_total_len = READ_ONCE(sqe->optlen);
821 if (sr->mshot_total_len)
822 sr->flags |= IORING_RECV_MSHOT_LIM;
823 } else if (sqe->optlen) {
824 return -EINVAL;
825 }
826 req->flags |= REQ_F_APOLL_MULTISHOT;
827 } else if (sqe->optlen) {
828 return -EINVAL;
829 }
830
831 if (sr->flags & IORING_RECVSEND_BUNDLE) {
832 if (req->opcode == IORING_OP_RECVMSG)
833 return -EINVAL;
834 }
835
836 if (io_is_compat(req->ctx))
837 sr->msg_flags |= MSG_CMSG_COMPAT;
838
839 sr->nr_multishot_loops = 0;
840 return io_recvmsg_prep_setup(req);
841 }
842
843 /* bits to clear in old and inherit in new cflags on bundle retry */
844 #define CQE_F_MASK (IORING_CQE_F_SOCK_NONEMPTY|IORING_CQE_F_MORE)
845
846 /*
847 * Finishes io_recv and io_recvmsg.
848 *
849 * Returns true if it is actually finished, or false if it should run
850 * again (for multishot).
851 */
io_recv_finish(struct io_kiocb * req,struct io_async_msghdr * kmsg,struct io_br_sel * sel,bool mshot_finished,unsigned issue_flags)852 static inline bool io_recv_finish(struct io_kiocb *req,
853 struct io_async_msghdr *kmsg,
854 struct io_br_sel *sel, bool mshot_finished,
855 unsigned issue_flags)
856 {
857 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
858 unsigned int cflags = 0;
859
860 if (kmsg->msg.msg_inq > 0)
861 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
862
863 if (sel->val > 0 && sr->flags & IORING_RECV_MSHOT_LIM) {
864 /*
865 * If sr->len hits zero, the limit has been reached. Mark
866 * mshot as finished, and flag MSHOT_DONE as well to prevent
867 * a potential bundle from being retried.
868 */
869 sr->mshot_total_len -= min_t(int, sel->val, sr->mshot_total_len);
870 if (!sr->mshot_total_len) {
871 sr->flags |= IORING_RECV_MSHOT_DONE;
872 mshot_finished = true;
873 }
874 }
875
876 if (sr->flags & IORING_RECVSEND_BUNDLE) {
877 size_t this_ret = sel->val - sr->done_io;
878
879 cflags |= io_put_kbufs(req, this_ret, sel->buf_list, io_bundle_nbufs(kmsg, this_ret));
880 if (sr->flags & IORING_RECV_RETRY)
881 cflags = req->cqe.flags | (cflags & CQE_F_MASK);
882 if (sr->mshot_len && sel->val >= sr->mshot_len)
883 sr->flags |= IORING_RECV_MSHOT_CAP;
884 /* bundle with no more immediate buffers, we're done */
885 if (req->flags & REQ_F_BL_EMPTY)
886 goto finish;
887 /*
888 * If more is available AND it was a full transfer, retry and
889 * append to this one
890 */
891 if (!(sr->flags & IORING_RECV_NO_RETRY) &&
892 kmsg->msg.msg_inq > 1 && this_ret > 0 &&
893 !iov_iter_count(&kmsg->msg.msg_iter)) {
894 req->cqe.flags = cflags & ~CQE_F_MASK;
895 sr->len = kmsg->msg.msg_inq;
896 sr->done_io += this_ret;
897 sr->flags |= IORING_RECV_RETRY;
898 return false;
899 }
900 } else {
901 cflags |= io_put_kbuf(req, sel->val, sel->buf_list);
902 }
903
904 /*
905 * Fill CQE for this receive and see if we should keep trying to
906 * receive from this socket.
907 */
908 if ((req->flags & REQ_F_APOLL_MULTISHOT) && !mshot_finished &&
909 io_req_post_cqe(req, sel->val, cflags | IORING_CQE_F_MORE)) {
910 sel->val = IOU_RETRY;
911 io_mshot_prep_retry(req, kmsg);
912 /* Known not-empty or unknown state, retry */
913 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || kmsg->msg.msg_inq < 0) {
914 if (sr->nr_multishot_loops++ < MULTISHOT_MAX_RETRY &&
915 !(sr->flags & IORING_RECV_MSHOT_CAP)) {
916 return false;
917 }
918 /* mshot retries exceeded, force a requeue */
919 sr->nr_multishot_loops = 0;
920 sr->flags &= ~IORING_RECV_MSHOT_CAP;
921 if (issue_flags & IO_URING_F_MULTISHOT)
922 sel->val = IOU_REQUEUE;
923 }
924 return true;
925 }
926
927 /* Finish the request / stop multishot. */
928 finish:
929 io_req_set_res(req, sel->val, cflags);
930 sel->val = IOU_COMPLETE;
931 io_req_msg_cleanup(req, issue_flags);
932 return true;
933 }
934
io_recvmsg_prep_multishot(struct io_async_msghdr * kmsg,struct io_sr_msg * sr,void __user ** buf,size_t * len)935 static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
936 struct io_sr_msg *sr, void __user **buf,
937 size_t *len)
938 {
939 unsigned long ubuf = (unsigned long) *buf;
940 unsigned long hdr;
941
942 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
943 kmsg->controllen;
944 if (*len < hdr)
945 return -EFAULT;
946
947 if (kmsg->controllen) {
948 unsigned long control = ubuf + hdr - kmsg->controllen;
949
950 kmsg->msg.msg_control_user = (void __user *) control;
951 kmsg->msg.msg_controllen = kmsg->controllen;
952 }
953
954 sr->buf = *buf; /* stash for later copy */
955 *buf = (void __user *) (ubuf + hdr);
956 kmsg->payloadlen = *len = *len - hdr;
957 return 0;
958 }
959
960 struct io_recvmsg_multishot_hdr {
961 struct io_uring_recvmsg_out msg;
962 struct sockaddr_storage addr;
963 };
964
io_recvmsg_multishot(struct socket * sock,struct io_sr_msg * io,struct io_async_msghdr * kmsg,unsigned int flags,bool * finished)965 static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
966 struct io_async_msghdr *kmsg,
967 unsigned int flags, bool *finished)
968 {
969 int err;
970 int copy_len;
971 struct io_recvmsg_multishot_hdr hdr;
972
973 if (kmsg->namelen)
974 kmsg->msg.msg_name = &hdr.addr;
975 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
976 kmsg->msg.msg_namelen = 0;
977
978 if (sock->file->f_flags & O_NONBLOCK)
979 flags |= MSG_DONTWAIT;
980
981 err = sock_recvmsg(sock, &kmsg->msg, flags);
982 *finished = err <= 0;
983 if (err < 0)
984 return err;
985
986 hdr.msg = (struct io_uring_recvmsg_out) {
987 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
988 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
989 };
990
991 hdr.msg.payloadlen = err;
992 if (err > kmsg->payloadlen)
993 err = kmsg->payloadlen;
994
995 copy_len = sizeof(struct io_uring_recvmsg_out);
996 if (kmsg->msg.msg_namelen > kmsg->namelen)
997 copy_len += kmsg->namelen;
998 else
999 copy_len += kmsg->msg.msg_namelen;
1000
1001 /*
1002 * "fromlen shall refer to the value before truncation.."
1003 * 1003.1g
1004 */
1005 hdr.msg.namelen = kmsg->msg.msg_namelen;
1006
1007 /* ensure that there is no gap between hdr and sockaddr_storage */
1008 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
1009 sizeof(struct io_uring_recvmsg_out));
1010 if (copy_to_user(io->buf, &hdr, copy_len)) {
1011 *finished = true;
1012 return -EFAULT;
1013 }
1014
1015 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
1016 kmsg->controllen + err;
1017 }
1018
io_recvmsg(struct io_kiocb * req,unsigned int issue_flags)1019 int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
1020 {
1021 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1022 struct io_async_msghdr *kmsg = req->async_data;
1023 struct io_br_sel sel = { };
1024 struct socket *sock;
1025 unsigned flags;
1026 int ret, min_ret = 0;
1027 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1028 bool mshot_finished = true;
1029
1030 sock = sock_from_file(req->file);
1031 if (unlikely(!sock))
1032 return -ENOTSOCK;
1033
1034 if (!(req->flags & REQ_F_POLLED) &&
1035 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1036 return -EAGAIN;
1037
1038 flags = sr->msg_flags;
1039 if (force_nonblock)
1040 flags |= MSG_DONTWAIT;
1041
1042 retry_multishot:
1043 sel.buf_list = NULL;
1044 if (io_do_buffer_select(req)) {
1045 size_t len = sr->len;
1046
1047 sel = io_buffer_select(req, &len, sr->buf_group, issue_flags);
1048 if (!sel.addr)
1049 return -ENOBUFS;
1050
1051 if (req->flags & REQ_F_APOLL_MULTISHOT) {
1052 ret = io_recvmsg_prep_multishot(kmsg, sr, &sel.addr, &len);
1053 if (ret) {
1054 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1055 return ret;
1056 }
1057 }
1058
1059 iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_DEST, sel.addr, len);
1060 }
1061
1062 kmsg->msg.msg_get_inq = 1;
1063 kmsg->msg.msg_inq = -1;
1064 if (req->flags & REQ_F_APOLL_MULTISHOT) {
1065 ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
1066 &mshot_finished);
1067 } else {
1068 /* disable partial retry for recvmsg with cmsg attached */
1069 if (flags & MSG_WAITALL && !kmsg->msg.msg_controllen)
1070 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1071
1072 ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
1073 kmsg->uaddr, flags);
1074 }
1075
1076 if (ret < min_ret) {
1077 if (ret == -EAGAIN && force_nonblock) {
1078 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1079 return IOU_RETRY;
1080 }
1081 if (ret > 0 && io_net_retry(sock, flags)) {
1082 sr->done_io += ret;
1083 return io_net_kbuf_recyle(req, sel.buf_list, kmsg, ret);
1084 }
1085 if (ret == -ERESTARTSYS)
1086 ret = -EINTR;
1087 req_set_fail(req);
1088 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
1089 req_set_fail(req);
1090 }
1091
1092 if (ret > 0)
1093 ret += sr->done_io;
1094 else if (sr->done_io)
1095 ret = sr->done_io;
1096 else
1097 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1098
1099 sel.val = ret;
1100 if (!io_recv_finish(req, kmsg, &sel, mshot_finished, issue_flags))
1101 goto retry_multishot;
1102
1103 return sel.val;
1104 }
1105
io_recv_buf_select(struct io_kiocb * req,struct io_async_msghdr * kmsg,struct io_br_sel * sel,unsigned int issue_flags)1106 static int io_recv_buf_select(struct io_kiocb *req, struct io_async_msghdr *kmsg,
1107 struct io_br_sel *sel, unsigned int issue_flags)
1108 {
1109 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1110 int ret;
1111
1112 /*
1113 * If the ring isn't locked, then don't use the peek interface
1114 * to grab multiple buffers as we will lock/unlock between
1115 * this selection and posting the buffers.
1116 */
1117 if (!(issue_flags & IO_URING_F_UNLOCKED) &&
1118 sr->flags & IORING_RECVSEND_BUNDLE) {
1119 struct buf_sel_arg arg = {
1120 .iovs = &kmsg->fast_iov,
1121 .nr_iovs = 1,
1122 .mode = KBUF_MODE_EXPAND,
1123 .buf_group = sr->buf_group,
1124 };
1125
1126 if (kmsg->vec.iovec) {
1127 arg.nr_iovs = kmsg->vec.nr;
1128 arg.iovs = kmsg->vec.iovec;
1129 arg.mode |= KBUF_MODE_FREE;
1130 }
1131
1132 if (sel->val)
1133 arg.max_len = sel->val;
1134 else if (kmsg->msg.msg_inq > 1)
1135 arg.max_len = min_not_zero(sel->val, (ssize_t) kmsg->msg.msg_inq);
1136
1137 /* if mshot limited, ensure we don't go over */
1138 if (sr->flags & IORING_RECV_MSHOT_LIM)
1139 arg.max_len = min_not_zero(arg.max_len, sr->mshot_total_len);
1140 ret = io_buffers_peek(req, &arg, sel);
1141 if (unlikely(ret < 0))
1142 return ret;
1143
1144 if (arg.iovs != &kmsg->fast_iov && arg.iovs != kmsg->vec.iovec) {
1145 kmsg->vec.nr = ret;
1146 kmsg->vec.iovec = arg.iovs;
1147 req->flags |= REQ_F_NEED_CLEANUP;
1148 }
1149 if (arg.partial_map)
1150 sr->flags |= IORING_RECV_PARTIAL_MAP;
1151
1152 /* special case 1 vec, can be a fast path */
1153 if (ret == 1) {
1154 sr->buf = arg.iovs[0].iov_base;
1155 sr->len = arg.iovs[0].iov_len;
1156 goto map_ubuf;
1157 }
1158 iov_iter_init(&kmsg->msg.msg_iter, ITER_DEST, arg.iovs, ret,
1159 arg.out_len);
1160 } else {
1161 size_t len = sel->val;
1162
1163 *sel = io_buffer_select(req, &len, sr->buf_group, issue_flags);
1164 if (!sel->addr)
1165 return -ENOBUFS;
1166 sr->buf = sel->addr;
1167 sr->len = len;
1168 map_ubuf:
1169 ret = import_ubuf(ITER_DEST, sr->buf, sr->len,
1170 &kmsg->msg.msg_iter);
1171 if (unlikely(ret))
1172 return ret;
1173 }
1174
1175 return 0;
1176 }
1177
io_recv(struct io_kiocb * req,unsigned int issue_flags)1178 int io_recv(struct io_kiocb *req, unsigned int issue_flags)
1179 {
1180 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1181 struct io_async_msghdr *kmsg = req->async_data;
1182 struct io_br_sel sel;
1183 struct socket *sock;
1184 unsigned flags;
1185 int ret, min_ret = 0;
1186 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1187 bool mshot_finished;
1188
1189 if (!(req->flags & REQ_F_POLLED) &&
1190 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1191 return -EAGAIN;
1192
1193 sock = sock_from_file(req->file);
1194 if (unlikely(!sock))
1195 return -ENOTSOCK;
1196
1197 flags = sr->msg_flags;
1198 if (force_nonblock)
1199 flags |= MSG_DONTWAIT;
1200
1201 retry_multishot:
1202 sel.buf_list = NULL;
1203 if (io_do_buffer_select(req)) {
1204 sel.val = sr->len;
1205 ret = io_recv_buf_select(req, kmsg, &sel, issue_flags);
1206 if (unlikely(ret < 0)) {
1207 kmsg->msg.msg_inq = -1;
1208 goto out_free;
1209 }
1210 sr->buf = NULL;
1211 }
1212
1213 kmsg->msg.msg_flags = 0;
1214 kmsg->msg.msg_inq = -1;
1215
1216 if (flags & MSG_WAITALL)
1217 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1218
1219 ret = sock_recvmsg(sock, &kmsg->msg, flags);
1220 if (ret < min_ret) {
1221 if (ret == -EAGAIN && force_nonblock) {
1222 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1223 return IOU_RETRY;
1224 }
1225 if (ret > 0 && io_net_retry(sock, flags)) {
1226 sr->len -= ret;
1227 sr->buf += ret;
1228 sr->done_io += ret;
1229 return io_net_kbuf_recyle(req, sel.buf_list, kmsg, ret);
1230 }
1231 if (ret == -ERESTARTSYS)
1232 ret = -EINTR;
1233 req_set_fail(req);
1234 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
1235 out_free:
1236 req_set_fail(req);
1237 }
1238
1239 mshot_finished = ret <= 0;
1240 if (ret > 0)
1241 ret += sr->done_io;
1242 else if (sr->done_io)
1243 ret = sr->done_io;
1244 else
1245 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1246
1247 sel.val = ret;
1248 if (!io_recv_finish(req, kmsg, &sel, mshot_finished, issue_flags))
1249 goto retry_multishot;
1250
1251 return sel.val;
1252 }
1253
io_recvzc_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1254 int io_recvzc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1255 {
1256 struct io_recvzc *zc = io_kiocb_to_cmd(req, struct io_recvzc);
1257 unsigned ifq_idx;
1258
1259 if (unlikely(sqe->addr2 || sqe->addr || sqe->addr3))
1260 return -EINVAL;
1261
1262 ifq_idx = READ_ONCE(sqe->zcrx_ifq_idx);
1263 zc->ifq = xa_load(&req->ctx->zcrx_ctxs, ifq_idx);
1264 if (!zc->ifq)
1265 return -EINVAL;
1266
1267 zc->len = READ_ONCE(sqe->len);
1268 zc->flags = READ_ONCE(sqe->ioprio);
1269 if (READ_ONCE(sqe->msg_flags))
1270 return -EINVAL;
1271 if (zc->flags & ~(IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT))
1272 return -EINVAL;
1273 /* multishot required */
1274 if (!(zc->flags & IORING_RECV_MULTISHOT))
1275 return -EINVAL;
1276 /* All data completions are posted as aux CQEs. */
1277 req->flags |= REQ_F_APOLL_MULTISHOT;
1278
1279 return 0;
1280 }
1281
io_recvzc(struct io_kiocb * req,unsigned int issue_flags)1282 int io_recvzc(struct io_kiocb *req, unsigned int issue_flags)
1283 {
1284 struct io_recvzc *zc = io_kiocb_to_cmd(req, struct io_recvzc);
1285 struct socket *sock;
1286 unsigned int len;
1287 int ret;
1288
1289 if (!(req->flags & REQ_F_POLLED) &&
1290 (zc->flags & IORING_RECVSEND_POLL_FIRST))
1291 return -EAGAIN;
1292
1293 sock = sock_from_file(req->file);
1294 if (unlikely(!sock))
1295 return -ENOTSOCK;
1296
1297 len = zc->len;
1298 ret = io_zcrx_recv(req, zc->ifq, sock, 0, issue_flags, &zc->len);
1299 if (len && zc->len == 0) {
1300 io_req_set_res(req, 0, 0);
1301
1302 return IOU_COMPLETE;
1303 }
1304 if (unlikely(ret <= 0) && ret != -EAGAIN) {
1305 if (ret == -ERESTARTSYS)
1306 ret = -EINTR;
1307 if (ret == IOU_REQUEUE)
1308 return IOU_REQUEUE;
1309
1310 req_set_fail(req);
1311 io_req_set_res(req, ret, 0);
1312 return IOU_COMPLETE;
1313 }
1314 return IOU_RETRY;
1315 }
1316
io_send_zc_cleanup(struct io_kiocb * req)1317 void io_send_zc_cleanup(struct io_kiocb *req)
1318 {
1319 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1320 struct io_async_msghdr *io = req->async_data;
1321
1322 if (req_has_async_data(req))
1323 io_netmsg_iovec_free(io);
1324 if (zc->notif) {
1325 io_notif_flush(zc->notif);
1326 zc->notif = NULL;
1327 }
1328 }
1329
1330 #define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF)
1331 #define IO_ZC_FLAGS_VALID (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE | \
1332 IORING_SEND_VECTORIZED)
1333
io_send_zc_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1334 int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1335 {
1336 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1337 struct io_ring_ctx *ctx = req->ctx;
1338 struct io_async_msghdr *iomsg;
1339 struct io_kiocb *notif;
1340 u64 user_data;
1341 int ret;
1342
1343 zc->done_io = 0;
1344
1345 if (unlikely(READ_ONCE(sqe->__pad2[0])))
1346 return -EINVAL;
1347 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
1348 if (req->flags & REQ_F_CQE_SKIP)
1349 return -EINVAL;
1350
1351 notif = zc->notif = io_alloc_notif(ctx);
1352 if (!notif)
1353 return -ENOMEM;
1354 user_data = READ_ONCE(sqe->addr3);
1355 if (!user_data)
1356 user_data = req->cqe.user_data;
1357
1358 notif->cqe.user_data = user_data;
1359 notif->cqe.res = 0;
1360 notif->cqe.flags = IORING_CQE_F_NOTIF;
1361 req->flags |= REQ_F_NEED_CLEANUP | REQ_F_POLL_NO_LAZY;
1362
1363 zc->flags = READ_ONCE(sqe->ioprio);
1364 if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) {
1365 if (zc->flags & ~IO_ZC_FLAGS_VALID)
1366 return -EINVAL;
1367 if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) {
1368 struct io_notif_data *nd = io_notif_to_data(notif);
1369
1370 nd->zc_report = true;
1371 nd->zc_used = false;
1372 nd->zc_copied = false;
1373 }
1374 }
1375
1376 zc->len = READ_ONCE(sqe->len);
1377 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL | MSG_ZEROCOPY;
1378 req->buf_index = READ_ONCE(sqe->buf_index);
1379 if (zc->msg_flags & MSG_DONTWAIT)
1380 req->flags |= REQ_F_NOWAIT;
1381
1382 if (io_is_compat(ctx))
1383 zc->msg_flags |= MSG_CMSG_COMPAT;
1384
1385 iomsg = io_msg_alloc_async(req);
1386 if (unlikely(!iomsg))
1387 return -ENOMEM;
1388
1389 if (req->opcode == IORING_OP_SEND_ZC) {
1390 ret = io_send_setup(req, sqe);
1391 } else {
1392 if (unlikely(sqe->addr2 || sqe->file_index))
1393 return -EINVAL;
1394 ret = io_sendmsg_setup(req, sqe);
1395 }
1396 if (unlikely(ret))
1397 return ret;
1398
1399 if (!(zc->flags & IORING_RECVSEND_FIXED_BUF)) {
1400 iomsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1401 return io_notif_account_mem(zc->notif, iomsg->msg.msg_iter.count);
1402 }
1403 iomsg->msg.sg_from_iter = io_sg_from_iter;
1404 return 0;
1405 }
1406
io_sg_from_iter_iovec(struct sk_buff * skb,struct iov_iter * from,size_t length)1407 static int io_sg_from_iter_iovec(struct sk_buff *skb,
1408 struct iov_iter *from, size_t length)
1409 {
1410 skb_zcopy_downgrade_managed(skb);
1411 return zerocopy_fill_skb_from_iter(skb, from, length);
1412 }
1413
io_sg_from_iter(struct sk_buff * skb,struct iov_iter * from,size_t length)1414 static int io_sg_from_iter(struct sk_buff *skb,
1415 struct iov_iter *from, size_t length)
1416 {
1417 struct skb_shared_info *shinfo = skb_shinfo(skb);
1418 int frag = shinfo->nr_frags;
1419 int ret = 0;
1420 struct bvec_iter bi;
1421 ssize_t copied = 0;
1422 unsigned long truesize = 0;
1423
1424 if (!frag)
1425 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
1426 else if (unlikely(!skb_zcopy_managed(skb)))
1427 return zerocopy_fill_skb_from_iter(skb, from, length);
1428
1429 bi.bi_size = min(from->count, length);
1430 bi.bi_bvec_done = from->iov_offset;
1431 bi.bi_idx = 0;
1432
1433 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1434 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1435
1436 copied += v.bv_len;
1437 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1438 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1439 v.bv_offset, v.bv_len);
1440 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1441 }
1442 if (bi.bi_size)
1443 ret = -EMSGSIZE;
1444
1445 shinfo->nr_frags = frag;
1446 from->bvec += bi.bi_idx;
1447 from->nr_segs -= bi.bi_idx;
1448 from->count -= copied;
1449 from->iov_offset = bi.bi_bvec_done;
1450
1451 skb->data_len += copied;
1452 skb->len += copied;
1453 skb->truesize += truesize;
1454 return ret;
1455 }
1456
io_send_zc_import(struct io_kiocb * req,struct io_async_msghdr * kmsg,unsigned int issue_flags)1457 static int io_send_zc_import(struct io_kiocb *req,
1458 struct io_async_msghdr *kmsg,
1459 unsigned int issue_flags)
1460 {
1461 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1462 struct io_kiocb *notif = sr->notif;
1463 int ret;
1464
1465 WARN_ON_ONCE(!(sr->flags & IORING_RECVSEND_FIXED_BUF));
1466
1467 notif->buf_index = req->buf_index;
1468
1469 if (!(sr->flags & IORING_SEND_VECTORIZED)) {
1470 ret = io_import_reg_buf(notif, &kmsg->msg.msg_iter,
1471 (u64)(uintptr_t)sr->buf, sr->len,
1472 ITER_SOURCE, issue_flags);
1473 } else {
1474 unsigned uvec_segs = kmsg->msg.msg_iter.nr_segs;
1475
1476 ret = io_import_reg_vec(ITER_SOURCE, &kmsg->msg.msg_iter,
1477 notif, &kmsg->vec, uvec_segs,
1478 issue_flags);
1479 }
1480
1481 if (unlikely(ret))
1482 return ret;
1483 req->flags &= ~REQ_F_IMPORT_BUFFER;
1484 return 0;
1485 }
1486
io_sendmsg_zc(struct io_kiocb * req,unsigned int issue_flags)1487 int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1488 {
1489 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1490 struct io_async_msghdr *kmsg = req->async_data;
1491 struct socket *sock;
1492 unsigned msg_flags;
1493 int ret, min_ret = 0;
1494
1495 sock = sock_from_file(req->file);
1496 if (unlikely(!sock))
1497 return -ENOTSOCK;
1498 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1499 return -EOPNOTSUPP;
1500 if (!(req->flags & REQ_F_POLLED) &&
1501 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1502 return -EAGAIN;
1503
1504 if (req->flags & REQ_F_IMPORT_BUFFER) {
1505 ret = io_send_zc_import(req, kmsg, issue_flags);
1506 if (unlikely(ret))
1507 return ret;
1508 }
1509
1510 msg_flags = sr->msg_flags;
1511 if (issue_flags & IO_URING_F_NONBLOCK)
1512 msg_flags |= MSG_DONTWAIT;
1513 if (msg_flags & MSG_WAITALL)
1514 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1515
1516 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1517
1518 if (req->opcode == IORING_OP_SEND_ZC) {
1519 msg_flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
1520 kmsg->msg.msg_flags = msg_flags;
1521 ret = sock_sendmsg(sock, &kmsg->msg);
1522 } else {
1523 kmsg->msg.msg_control_user = sr->msg_control;
1524 ret = __sys_sendmsg_sock(sock, &kmsg->msg, msg_flags);
1525 }
1526
1527 if (unlikely(ret < min_ret)) {
1528 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1529 return -EAGAIN;
1530
1531 if (ret > 0 && io_net_retry(sock, sr->msg_flags)) {
1532 sr->done_io += ret;
1533 return -EAGAIN;
1534 }
1535 if (ret == -ERESTARTSYS)
1536 ret = -EINTR;
1537 req_set_fail(req);
1538 }
1539
1540 if (ret >= 0)
1541 ret += sr->done_io;
1542 else if (sr->done_io)
1543 ret = sr->done_io;
1544
1545 /*
1546 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1547 * flushing notif to io_send_zc_cleanup()
1548 */
1549 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1550 io_notif_flush(sr->notif);
1551 sr->notif = NULL;
1552 io_req_msg_cleanup(req, 0);
1553 }
1554 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1555 return IOU_COMPLETE;
1556 }
1557
io_sendrecv_fail(struct io_kiocb * req)1558 void io_sendrecv_fail(struct io_kiocb *req)
1559 {
1560 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1561
1562 if (sr->done_io)
1563 req->cqe.res = sr->done_io;
1564
1565 if ((req->flags & REQ_F_NEED_CLEANUP) &&
1566 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1567 req->cqe.flags |= IORING_CQE_F_MORE;
1568 }
1569
1570 #define ACCEPT_FLAGS (IORING_ACCEPT_MULTISHOT | IORING_ACCEPT_DONTWAIT | \
1571 IORING_ACCEPT_POLL_FIRST)
1572
io_accept_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1573 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1574 {
1575 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1576
1577 if (sqe->len || sqe->buf_index)
1578 return -EINVAL;
1579
1580 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1581 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1582 accept->flags = READ_ONCE(sqe->accept_flags);
1583 accept->nofile = rlimit(RLIMIT_NOFILE);
1584 accept->iou_flags = READ_ONCE(sqe->ioprio);
1585 if (accept->iou_flags & ~ACCEPT_FLAGS)
1586 return -EINVAL;
1587
1588 accept->file_slot = READ_ONCE(sqe->file_index);
1589 if (accept->file_slot) {
1590 if (accept->flags & SOCK_CLOEXEC)
1591 return -EINVAL;
1592 if (accept->iou_flags & IORING_ACCEPT_MULTISHOT &&
1593 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1594 return -EINVAL;
1595 }
1596 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1597 return -EINVAL;
1598 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1599 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1600 if (accept->iou_flags & IORING_ACCEPT_MULTISHOT)
1601 req->flags |= REQ_F_APOLL_MULTISHOT;
1602 if (accept->iou_flags & IORING_ACCEPT_DONTWAIT)
1603 req->flags |= REQ_F_NOWAIT;
1604 return 0;
1605 }
1606
io_accept(struct io_kiocb * req,unsigned int issue_flags)1607 int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1608 {
1609 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1610 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1611 bool fixed = !!accept->file_slot;
1612 struct proto_accept_arg arg = {
1613 .flags = force_nonblock ? O_NONBLOCK : 0,
1614 };
1615 struct file *file;
1616 unsigned cflags;
1617 int ret, fd;
1618
1619 if (!(req->flags & REQ_F_POLLED) &&
1620 accept->iou_flags & IORING_ACCEPT_POLL_FIRST)
1621 return -EAGAIN;
1622
1623 retry:
1624 if (!fixed) {
1625 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1626 if (unlikely(fd < 0))
1627 return fd;
1628 }
1629 arg.err = 0;
1630 arg.is_empty = -1;
1631 file = do_accept(req->file, &arg, accept->addr, accept->addr_len,
1632 accept->flags);
1633 if (IS_ERR(file)) {
1634 if (!fixed)
1635 put_unused_fd(fd);
1636 ret = PTR_ERR(file);
1637 if (ret == -EAGAIN && force_nonblock &&
1638 !(accept->iou_flags & IORING_ACCEPT_DONTWAIT))
1639 return IOU_RETRY;
1640
1641 if (ret == -ERESTARTSYS)
1642 ret = -EINTR;
1643 } else if (!fixed) {
1644 fd_install(fd, file);
1645 ret = fd;
1646 } else {
1647 ret = io_fixed_fd_install(req, issue_flags, file,
1648 accept->file_slot);
1649 }
1650
1651 cflags = 0;
1652 if (!arg.is_empty)
1653 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
1654
1655 if (ret >= 0 && (req->flags & REQ_F_APOLL_MULTISHOT) &&
1656 io_req_post_cqe(req, ret, cflags | IORING_CQE_F_MORE)) {
1657 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || arg.is_empty == -1)
1658 goto retry;
1659 return IOU_RETRY;
1660 }
1661
1662 io_req_set_res(req, ret, cflags);
1663 if (ret < 0)
1664 req_set_fail(req);
1665 return IOU_COMPLETE;
1666 }
1667
io_socket_bpf_populate(struct io_uring_bpf_ctx * bctx,struct io_kiocb * req)1668 void io_socket_bpf_populate(struct io_uring_bpf_ctx *bctx, struct io_kiocb *req)
1669 {
1670 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1671
1672 bctx->socket.family = sock->domain;
1673 bctx->socket.type = sock->type;
1674 bctx->socket.protocol = sock->protocol;
1675 }
1676
io_socket_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1677 int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1678 {
1679 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1680
1681 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1682 return -EINVAL;
1683
1684 sock->domain = READ_ONCE(sqe->fd);
1685 sock->type = READ_ONCE(sqe->off);
1686 sock->protocol = READ_ONCE(sqe->len);
1687 sock->file_slot = READ_ONCE(sqe->file_index);
1688 sock->nofile = rlimit(RLIMIT_NOFILE);
1689
1690 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1691 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1692 return -EINVAL;
1693 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1694 return -EINVAL;
1695 return 0;
1696 }
1697
io_socket(struct io_kiocb * req,unsigned int issue_flags)1698 int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1699 {
1700 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1701 bool fixed = !!sock->file_slot;
1702 struct file *file;
1703 int ret, fd;
1704
1705 if (!fixed) {
1706 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1707 if (unlikely(fd < 0))
1708 return fd;
1709 }
1710 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1711 if (IS_ERR(file)) {
1712 if (!fixed)
1713 put_unused_fd(fd);
1714 ret = PTR_ERR(file);
1715 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1716 return -EAGAIN;
1717 if (ret == -ERESTARTSYS)
1718 ret = -EINTR;
1719 req_set_fail(req);
1720 } else if (!fixed) {
1721 fd_install(fd, file);
1722 ret = fd;
1723 } else {
1724 ret = io_fixed_fd_install(req, issue_flags, file,
1725 sock->file_slot);
1726 }
1727 io_req_set_res(req, ret, 0);
1728 return IOU_COMPLETE;
1729 }
1730
io_connect_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1731 int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1732 {
1733 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1734 struct io_async_msghdr *io;
1735
1736 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1737 return -EINVAL;
1738
1739 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1740 conn->addr_len = READ_ONCE(sqe->addr2);
1741 conn->in_progress = conn->seen_econnaborted = false;
1742
1743 io = io_msg_alloc_async(req);
1744 if (unlikely(!io))
1745 return -ENOMEM;
1746
1747 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->addr);
1748 }
1749
io_connect(struct io_kiocb * req,unsigned int issue_flags)1750 int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1751 {
1752 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1753 struct io_async_msghdr *io = req->async_data;
1754 unsigned file_flags;
1755 int ret;
1756 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1757
1758 if (connect->in_progress) {
1759 struct poll_table_struct pt = { ._key = EPOLLERR };
1760
1761 if (vfs_poll(req->file, &pt) & EPOLLERR)
1762 goto get_sock_err;
1763 }
1764
1765 file_flags = force_nonblock ? O_NONBLOCK : 0;
1766
1767 ret = __sys_connect_file(req->file, &io->addr, connect->addr_len,
1768 file_flags);
1769 if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED)
1770 && force_nonblock) {
1771 if (ret == -EINPROGRESS) {
1772 connect->in_progress = true;
1773 } else if (ret == -ECONNABORTED) {
1774 if (connect->seen_econnaborted)
1775 goto out;
1776 connect->seen_econnaborted = true;
1777 }
1778 return -EAGAIN;
1779 }
1780 if (connect->in_progress) {
1781 /*
1782 * At least bluetooth will return -EBADFD on a re-connect
1783 * attempt, and it's (supposedly) also valid to get -EISCONN
1784 * which means the previous result is good. For both of these,
1785 * grab the sock_error() and use that for the completion.
1786 */
1787 if (ret == -EBADFD || ret == -EISCONN) {
1788 get_sock_err:
1789 ret = sock_error(sock_from_file(req->file)->sk);
1790 }
1791 }
1792 if (ret == -ERESTARTSYS)
1793 ret = -EINTR;
1794 out:
1795 if (ret < 0)
1796 req_set_fail(req);
1797 io_req_msg_cleanup(req, issue_flags);
1798 io_req_set_res(req, ret, 0);
1799 return IOU_COMPLETE;
1800 }
1801
io_bind_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1802 int io_bind_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1803 {
1804 struct io_bind *bind = io_kiocb_to_cmd(req, struct io_bind);
1805 struct sockaddr __user *uaddr;
1806 struct io_async_msghdr *io;
1807
1808 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1809 return -EINVAL;
1810
1811 uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1812 bind->addr_len = READ_ONCE(sqe->addr2);
1813
1814 io = io_msg_alloc_async(req);
1815 if (unlikely(!io))
1816 return -ENOMEM;
1817 return move_addr_to_kernel(uaddr, bind->addr_len, &io->addr);
1818 }
1819
io_bind(struct io_kiocb * req,unsigned int issue_flags)1820 int io_bind(struct io_kiocb *req, unsigned int issue_flags)
1821 {
1822 struct io_bind *bind = io_kiocb_to_cmd(req, struct io_bind);
1823 struct io_async_msghdr *io = req->async_data;
1824 struct socket *sock;
1825 int ret;
1826
1827 sock = sock_from_file(req->file);
1828 if (unlikely(!sock))
1829 return -ENOTSOCK;
1830
1831 ret = __sys_bind_socket(sock, &io->addr, bind->addr_len);
1832 if (ret < 0)
1833 req_set_fail(req);
1834 io_req_set_res(req, ret, 0);
1835 return 0;
1836 }
1837
io_listen_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1838 int io_listen_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1839 {
1840 struct io_listen *listen = io_kiocb_to_cmd(req, struct io_listen);
1841
1842 if (sqe->addr || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in || sqe->addr2)
1843 return -EINVAL;
1844
1845 listen->backlog = READ_ONCE(sqe->len);
1846 return 0;
1847 }
1848
io_listen(struct io_kiocb * req,unsigned int issue_flags)1849 int io_listen(struct io_kiocb *req, unsigned int issue_flags)
1850 {
1851 struct io_listen *listen = io_kiocb_to_cmd(req, struct io_listen);
1852 struct socket *sock;
1853 int ret;
1854
1855 sock = sock_from_file(req->file);
1856 if (unlikely(!sock))
1857 return -ENOTSOCK;
1858
1859 ret = __sys_listen_socket(sock, listen->backlog);
1860 if (ret < 0)
1861 req_set_fail(req);
1862 io_req_set_res(req, ret, 0);
1863 return 0;
1864 }
1865
io_netmsg_cache_free(const void * entry)1866 void io_netmsg_cache_free(const void *entry)
1867 {
1868 struct io_async_msghdr *kmsg = (struct io_async_msghdr *) entry;
1869
1870 io_vec_free(&kmsg->vec);
1871 kfree(kmsg);
1872 }
1873