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 return -EINVAL;
380 req->flags |= REQ_F_IMPORT_BUFFER;
381 return 0;
382 }
383 if (req->flags & REQ_F_BUFFER_SELECT)
384 return 0;
385
386 if (sr->flags & IORING_SEND_VECTORIZED)
387 return io_net_import_vec(req, kmsg, sr->buf, sr->len, ITER_SOURCE);
388
389 return import_ubuf(ITER_SOURCE, sr->buf, sr->len, &kmsg->msg.msg_iter);
390 }
391
io_sendmsg_setup(struct io_kiocb * req,const struct io_uring_sqe * sqe)392 static int io_sendmsg_setup(struct io_kiocb *req, const struct io_uring_sqe *sqe)
393 {
394 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
395 struct io_async_msghdr *kmsg = req->async_data;
396 struct user_msghdr msg;
397 int ret;
398
399 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
400 ret = io_msg_copy_hdr(req, kmsg, &msg, ITER_SOURCE, NULL);
401 if (unlikely(ret))
402 return ret;
403 /* save msg_control as sys_sendmsg() overwrites it */
404 sr->msg_control = kmsg->msg.msg_control_user;
405
406 if (sr->flags & IORING_RECVSEND_FIXED_BUF) {
407 kmsg->msg.msg_iter.nr_segs = msg.msg_iovlen;
408 return io_prep_reg_iovec(req, &kmsg->vec, msg.msg_iov,
409 msg.msg_iovlen);
410 }
411 if (req->flags & REQ_F_BUFFER_SELECT)
412 return 0;
413 return io_net_import_vec(req, kmsg, msg.msg_iov, msg.msg_iovlen, ITER_SOURCE);
414 }
415
416 #define SENDMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_BUNDLE | IORING_SEND_VECTORIZED)
417
io_sendmsg_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)418 int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
419 {
420 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
421
422 sr->done_io = 0;
423 sr->len = READ_ONCE(sqe->len);
424 sr->flags = READ_ONCE(sqe->ioprio);
425 if (sr->flags & ~SENDMSG_FLAGS)
426 return -EINVAL;
427 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
428 if (sr->msg_flags & MSG_DONTWAIT)
429 req->flags |= REQ_F_NOWAIT;
430 if (req->flags & REQ_F_BUFFER_SELECT)
431 sr->buf_group = req->buf_index;
432 if (sr->flags & IORING_RECVSEND_BUNDLE) {
433 if (req->opcode == IORING_OP_SENDMSG)
434 return -EINVAL;
435 sr->msg_flags |= MSG_WAITALL;
436 req->flags |= REQ_F_MULTISHOT;
437 }
438
439 if (io_is_compat(req->ctx))
440 sr->msg_flags |= MSG_CMSG_COMPAT;
441
442 if (unlikely(!io_msg_alloc_async(req)))
443 return -ENOMEM;
444 if (req->opcode != IORING_OP_SENDMSG)
445 return io_send_setup(req, sqe);
446 if (unlikely(sqe->addr2 || sqe->file_index))
447 return -EINVAL;
448 return io_sendmsg_setup(req, sqe);
449 }
450
io_req_msg_cleanup(struct io_kiocb * req,unsigned int issue_flags)451 static void io_req_msg_cleanup(struct io_kiocb *req,
452 unsigned int issue_flags)
453 {
454 io_netmsg_recycle(req, issue_flags);
455 }
456
457 /*
458 * For bundle completions, we need to figure out how many segments we consumed.
459 * A bundle could be using a single ITER_UBUF if that's all we mapped, or it
460 * could be using an ITER_IOVEC. If the latter, then if we consumed all of
461 * the segments, then it's a trivial questiont o answer. If we have residual
462 * data in the iter, then loop the segments to figure out how much we
463 * transferred.
464 */
io_bundle_nbufs(struct io_async_msghdr * kmsg,int ret)465 static int io_bundle_nbufs(struct io_async_msghdr *kmsg, int ret)
466 {
467 struct iovec *iov;
468 int nbufs;
469
470 /* no data is always zero segments, and a ubuf is always 1 segment */
471 if (ret <= 0)
472 return 0;
473 if (iter_is_ubuf(&kmsg->msg.msg_iter))
474 return 1;
475
476 iov = kmsg->vec.iovec;
477 if (!iov)
478 iov = &kmsg->fast_iov;
479
480 /* if all data was transferred, it's basic pointer math */
481 if (!iov_iter_count(&kmsg->msg.msg_iter))
482 return iter_iov(&kmsg->msg.msg_iter) - iov;
483
484 /* short transfer, count segments */
485 nbufs = 0;
486 do {
487 int this_len = min_t(int, iov[nbufs].iov_len, ret);
488
489 nbufs++;
490 ret -= this_len;
491 } while (ret);
492
493 return nbufs;
494 }
495
io_net_kbuf_recyle(struct io_kiocb * req,struct io_buffer_list * bl,struct io_async_msghdr * kmsg,int len)496 static int io_net_kbuf_recyle(struct io_kiocb *req, struct io_buffer_list *bl,
497 struct io_async_msghdr *kmsg, int len)
498 {
499 req->flags |= REQ_F_BL_NO_RECYCLE;
500 if (req->flags & REQ_F_BUFFERS_COMMIT)
501 io_kbuf_commit(req, bl, len, io_bundle_nbufs(kmsg, len));
502 return IOU_RETRY;
503 }
504
io_send_finish(struct io_kiocb * req,struct io_async_msghdr * kmsg,struct io_br_sel * sel)505 static inline bool io_send_finish(struct io_kiocb *req,
506 struct io_async_msghdr *kmsg,
507 struct io_br_sel *sel)
508 {
509 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
510 bool bundle_finished = sel->val <= 0;
511 unsigned int cflags;
512
513 if (!(sr->flags & IORING_RECVSEND_BUNDLE)) {
514 cflags = io_put_kbuf(req, sel->val, sel->buf_list);
515 goto finish;
516 }
517
518 cflags = io_put_kbufs(req, sel->val, sel->buf_list, io_bundle_nbufs(kmsg, sel->val));
519
520 /*
521 * Don't start new bundles if the buffer list is empty, or if the
522 * current operation needed to go through polling to complete.
523 */
524 if (bundle_finished || req->flags & (REQ_F_BL_EMPTY | REQ_F_POLLED))
525 goto finish;
526
527 /*
528 * Fill CQE for this receive and see if we should keep trying to
529 * receive from this socket.
530 */
531 if (io_req_post_cqe(req, sel->val, cflags | IORING_CQE_F_MORE)) {
532 io_mshot_prep_retry(req, kmsg);
533 return false;
534 }
535
536 /* Otherwise stop bundle and use the current result. */
537 finish:
538 io_req_set_res(req, sel->val, cflags);
539 sel->val = IOU_COMPLETE;
540 return true;
541 }
542
io_sendmsg(struct io_kiocb * req,unsigned int issue_flags)543 int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
544 {
545 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
546 struct io_async_msghdr *kmsg = req->async_data;
547 struct socket *sock;
548 unsigned flags;
549 int min_ret = 0;
550 int ret;
551
552 sock = sock_from_file(req->file);
553 if (unlikely(!sock))
554 return -ENOTSOCK;
555
556 if (!(req->flags & REQ_F_POLLED) &&
557 (sr->flags & IORING_RECVSEND_POLL_FIRST))
558 return -EAGAIN;
559
560 flags = sr->msg_flags;
561 if (issue_flags & IO_URING_F_NONBLOCK)
562 flags |= MSG_DONTWAIT;
563 if (flags & MSG_WAITALL)
564 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
565
566 kmsg->msg.msg_control_user = sr->msg_control;
567
568 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
569
570 if (ret < min_ret) {
571 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
572 return -EAGAIN;
573 if (ret > 0 && io_net_retry(sock, flags)) {
574 kmsg->msg.msg_controllen = 0;
575 kmsg->msg.msg_control = NULL;
576 sr->done_io += ret;
577 return -EAGAIN;
578 }
579 if (ret == -ERESTARTSYS)
580 ret = -EINTR;
581 req_set_fail(req);
582 }
583 io_req_msg_cleanup(req, issue_flags);
584 if (ret >= 0)
585 ret += sr->done_io;
586 else if (sr->done_io)
587 ret = sr->done_io;
588 io_req_set_res(req, ret, 0);
589 return IOU_COMPLETE;
590 }
591
io_send_select_buffer(struct io_kiocb * req,unsigned int issue_flags,struct io_br_sel * sel,struct io_async_msghdr * kmsg)592 static int io_send_select_buffer(struct io_kiocb *req, unsigned int issue_flags,
593 struct io_br_sel *sel, struct io_async_msghdr *kmsg)
594 {
595 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
596 struct buf_sel_arg arg = {
597 .iovs = &kmsg->fast_iov,
598 .max_len = min_not_zero(sr->len, INT_MAX),
599 .nr_iovs = 1,
600 .buf_group = sr->buf_group,
601 };
602 int ret;
603
604 if (kmsg->vec.iovec) {
605 arg.nr_iovs = kmsg->vec.nr;
606 arg.iovs = kmsg->vec.iovec;
607 arg.mode = KBUF_MODE_FREE;
608 }
609
610 if (!(sr->flags & IORING_RECVSEND_BUNDLE))
611 arg.nr_iovs = 1;
612 else
613 arg.mode |= KBUF_MODE_EXPAND;
614
615 ret = io_buffers_select(req, &arg, sel, issue_flags);
616 if (unlikely(ret < 0))
617 return ret;
618
619 if (arg.iovs != &kmsg->fast_iov && arg.iovs != kmsg->vec.iovec) {
620 kmsg->vec.nr = ret;
621 kmsg->vec.iovec = arg.iovs;
622 req->flags |= REQ_F_NEED_CLEANUP;
623 }
624 sr->len = arg.out_len;
625
626 if (ret == 1) {
627 sr->buf = arg.iovs[0].iov_base;
628 ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len,
629 &kmsg->msg.msg_iter);
630 if (unlikely(ret))
631 return ret;
632 } else {
633 iov_iter_init(&kmsg->msg.msg_iter, ITER_SOURCE,
634 arg.iovs, ret, arg.out_len);
635 }
636
637 return 0;
638 }
639
io_send(struct io_kiocb * req,unsigned int issue_flags)640 int io_send(struct io_kiocb *req, unsigned int issue_flags)
641 {
642 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
643 struct io_async_msghdr *kmsg = req->async_data;
644 struct io_br_sel sel = { };
645 struct socket *sock;
646 unsigned flags;
647 int min_ret = 0;
648 int ret;
649
650 sock = sock_from_file(req->file);
651 if (unlikely(!sock))
652 return -ENOTSOCK;
653
654 if (!(req->flags & REQ_F_POLLED) &&
655 (sr->flags & IORING_RECVSEND_POLL_FIRST))
656 return -EAGAIN;
657
658 flags = sr->msg_flags;
659 if (issue_flags & IO_URING_F_NONBLOCK)
660 flags |= MSG_DONTWAIT;
661
662 retry_bundle:
663 sel.buf_list = NULL;
664 if (io_do_buffer_select(req)) {
665 ret = io_send_select_buffer(req, issue_flags, &sel, kmsg);
666 if (ret)
667 return ret;
668 }
669
670 /*
671 * If MSG_WAITALL is set, or this is a bundle send, then we need
672 * the full amount. If just bundle is set, if we do a short send
673 * then we complete the bundle sequence rather than continue on.
674 */
675 if (flags & MSG_WAITALL || sr->flags & IORING_RECVSEND_BUNDLE)
676 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
677
678 flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
679 kmsg->msg.msg_flags = flags;
680 ret = sock_sendmsg(sock, &kmsg->msg);
681 if (ret < min_ret) {
682 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
683 return -EAGAIN;
684
685 if (ret > 0 && io_net_retry(sock, flags)) {
686 sr->len -= ret;
687 sr->buf += ret;
688 sr->done_io += ret;
689 return io_net_kbuf_recyle(req, sel.buf_list, kmsg, ret);
690 }
691 if (ret == -ERESTARTSYS)
692 ret = -EINTR;
693 req_set_fail(req);
694 }
695 if (ret >= 0)
696 ret += sr->done_io;
697 else if (sr->done_io)
698 ret = sr->done_io;
699
700 sel.val = ret;
701 if (!io_send_finish(req, kmsg, &sel))
702 goto retry_bundle;
703
704 io_req_msg_cleanup(req, issue_flags);
705 return sel.val;
706 }
707
io_recvmsg_mshot_prep(struct io_kiocb * req,struct io_async_msghdr * iomsg,int namelen,size_t controllen)708 static int io_recvmsg_mshot_prep(struct io_kiocb *req,
709 struct io_async_msghdr *iomsg,
710 int namelen, size_t controllen)
711 {
712 if ((req->flags & (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) ==
713 (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) {
714 int hdr;
715
716 if (unlikely(namelen < 0))
717 return -EOVERFLOW;
718 if (check_add_overflow(sizeof(struct io_uring_recvmsg_out),
719 namelen, &hdr))
720 return -EOVERFLOW;
721 if (check_add_overflow(hdr, controllen, &hdr))
722 return -EOVERFLOW;
723
724 iomsg->namelen = namelen;
725 iomsg->controllen = controllen;
726 return 0;
727 }
728
729 return 0;
730 }
731
io_recvmsg_copy_hdr(struct io_kiocb * req,struct io_async_msghdr * iomsg)732 static int io_recvmsg_copy_hdr(struct io_kiocb *req,
733 struct io_async_msghdr *iomsg)
734 {
735 struct user_msghdr msg;
736 int ret;
737
738 ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_DEST, &iomsg->uaddr);
739 if (unlikely(ret))
740 return ret;
741
742 if (!(req->flags & REQ_F_BUFFER_SELECT)) {
743 ret = io_net_import_vec(req, iomsg, msg.msg_iov, msg.msg_iovlen,
744 ITER_DEST);
745 if (unlikely(ret))
746 return ret;
747 }
748 return io_recvmsg_mshot_prep(req, iomsg, msg.msg_namelen,
749 msg.msg_controllen);
750 }
751
io_recvmsg_prep_setup(struct io_kiocb * req)752 static int io_recvmsg_prep_setup(struct io_kiocb *req)
753 {
754 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
755 struct io_async_msghdr *kmsg;
756
757 kmsg = io_msg_alloc_async(req);
758 if (unlikely(!kmsg))
759 return -ENOMEM;
760
761 if (req->opcode == IORING_OP_RECV) {
762 kmsg->msg.msg_name = NULL;
763 kmsg->msg.msg_namelen = 0;
764 kmsg->msg.msg_inq = 0;
765 kmsg->msg.msg_control = NULL;
766 kmsg->msg.msg_get_inq = 1;
767 kmsg->msg.msg_controllen = 0;
768 kmsg->msg.msg_iocb = NULL;
769 kmsg->msg.msg_ubuf = NULL;
770
771 if (req->flags & REQ_F_BUFFER_SELECT)
772 return 0;
773 return import_ubuf(ITER_DEST, sr->buf, sr->len,
774 &kmsg->msg.msg_iter);
775 }
776
777 return io_recvmsg_copy_hdr(req, kmsg);
778 }
779
780 #define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT | \
781 IORING_RECVSEND_BUNDLE)
782
io_recvmsg_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)783 int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
784 {
785 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
786
787 sr->done_io = 0;
788
789 if (unlikely(sqe->addr2))
790 return -EINVAL;
791
792 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
793 sr->len = READ_ONCE(sqe->len);
794 sr->flags = READ_ONCE(sqe->ioprio);
795 if (sr->flags & ~RECVMSG_FLAGS)
796 return -EINVAL;
797 sr->msg_flags = READ_ONCE(sqe->msg_flags);
798 if (sr->msg_flags & MSG_DONTWAIT)
799 req->flags |= REQ_F_NOWAIT;
800 if (sr->msg_flags & MSG_ERRQUEUE)
801 req->flags |= REQ_F_CLEAR_POLLIN;
802 if (req->flags & REQ_F_BUFFER_SELECT)
803 sr->buf_group = req->buf_index;
804 sr->mshot_total_len = sr->mshot_len = 0;
805 if (sr->flags & IORING_RECV_MULTISHOT) {
806 if (!(req->flags & REQ_F_BUFFER_SELECT))
807 return -EINVAL;
808 if (sr->msg_flags & MSG_WAITALL)
809 return -EINVAL;
810 if (req->opcode == IORING_OP_RECV) {
811 sr->mshot_len = sr->len;
812 sr->mshot_total_len = READ_ONCE(sqe->optlen);
813 if (sr->mshot_total_len)
814 sr->flags |= IORING_RECV_MSHOT_LIM;
815 } else if (sqe->optlen) {
816 return -EINVAL;
817 }
818 req->flags |= REQ_F_APOLL_MULTISHOT;
819 } else if (sqe->optlen) {
820 return -EINVAL;
821 }
822
823 if (sr->flags & IORING_RECVSEND_BUNDLE) {
824 if (req->opcode == IORING_OP_RECVMSG)
825 return -EINVAL;
826 }
827
828 if (io_is_compat(req->ctx))
829 sr->msg_flags |= MSG_CMSG_COMPAT;
830
831 sr->nr_multishot_loops = 0;
832 return io_recvmsg_prep_setup(req);
833 }
834
835 /* bits to clear in old and inherit in new cflags on bundle retry */
836 #define CQE_F_MASK (IORING_CQE_F_SOCK_NONEMPTY|IORING_CQE_F_MORE)
837
838 /*
839 * Finishes io_recv and io_recvmsg.
840 *
841 * Returns true if it is actually finished, or false if it should run
842 * again (for multishot).
843 */
io_recv_finish(struct io_kiocb * req,struct io_async_msghdr * kmsg,struct io_br_sel * sel,bool mshot_finished,unsigned issue_flags)844 static inline bool io_recv_finish(struct io_kiocb *req,
845 struct io_async_msghdr *kmsg,
846 struct io_br_sel *sel, bool mshot_finished,
847 unsigned issue_flags)
848 {
849 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
850 unsigned int cflags = 0;
851
852 if (kmsg->msg.msg_inq > 0)
853 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
854
855 if (sel->val > 0 && sr->flags & IORING_RECV_MSHOT_LIM) {
856 /*
857 * If sr->len hits zero, the limit has been reached. Mark
858 * mshot as finished, and flag MSHOT_DONE as well to prevent
859 * a potential bundle from being retried.
860 */
861 sr->mshot_total_len -= min_t(int, sel->val, sr->mshot_total_len);
862 if (!sr->mshot_total_len) {
863 sr->flags |= IORING_RECV_MSHOT_DONE;
864 mshot_finished = true;
865 }
866 }
867
868 if (sr->flags & IORING_RECVSEND_BUNDLE) {
869 size_t this_ret = sel->val - sr->done_io;
870
871 cflags |= io_put_kbufs(req, this_ret, sel->buf_list, io_bundle_nbufs(kmsg, this_ret));
872 if (sr->flags & IORING_RECV_RETRY)
873 cflags = req->cqe.flags | (cflags & CQE_F_MASK);
874 if (sr->mshot_len && sel->val >= sr->mshot_len)
875 sr->flags |= IORING_RECV_MSHOT_CAP;
876 /* bundle with no more immediate buffers, we're done */
877 if (req->flags & REQ_F_BL_EMPTY)
878 goto finish;
879 /*
880 * If more is available AND it was a full transfer, retry and
881 * append to this one
882 */
883 if (!(sr->flags & IORING_RECV_NO_RETRY) &&
884 kmsg->msg.msg_inq > 1 && this_ret > 0 &&
885 !iov_iter_count(&kmsg->msg.msg_iter)) {
886 req->cqe.flags = cflags & ~CQE_F_MASK;
887 sr->len = kmsg->msg.msg_inq;
888 sr->done_io += this_ret;
889 sr->flags |= IORING_RECV_RETRY;
890 return false;
891 }
892 } else {
893 cflags |= io_put_kbuf(req, sel->val, sel->buf_list);
894 }
895
896 /*
897 * Fill CQE for this receive and see if we should keep trying to
898 * receive from this socket.
899 */
900 if ((req->flags & REQ_F_APOLL_MULTISHOT) && !mshot_finished &&
901 io_req_post_cqe(req, sel->val, cflags | IORING_CQE_F_MORE)) {
902 sel->val = IOU_RETRY;
903 io_mshot_prep_retry(req, kmsg);
904 /* Known not-empty or unknown state, retry */
905 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || kmsg->msg.msg_inq < 0) {
906 if (sr->nr_multishot_loops++ < MULTISHOT_MAX_RETRY &&
907 !(sr->flags & IORING_RECV_MSHOT_CAP)) {
908 return false;
909 }
910 /* mshot retries exceeded, force a requeue */
911 sr->nr_multishot_loops = 0;
912 sr->flags &= ~IORING_RECV_MSHOT_CAP;
913 if (issue_flags & IO_URING_F_MULTISHOT)
914 sel->val = IOU_REQUEUE;
915 }
916 return true;
917 }
918
919 /* Finish the request / stop multishot. */
920 finish:
921 io_req_set_res(req, sel->val, cflags);
922 sel->val = IOU_COMPLETE;
923 io_req_msg_cleanup(req, issue_flags);
924 return true;
925 }
926
io_recvmsg_prep_multishot(struct io_async_msghdr * kmsg,struct io_sr_msg * sr,void __user ** buf,size_t * len)927 static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
928 struct io_sr_msg *sr, void __user **buf,
929 size_t *len)
930 {
931 unsigned long ubuf = (unsigned long) *buf;
932 unsigned long hdr;
933
934 hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
935 kmsg->controllen;
936 if (*len < hdr)
937 return -EFAULT;
938
939 if (kmsg->controllen) {
940 unsigned long control = ubuf + hdr - kmsg->controllen;
941
942 kmsg->msg.msg_control_user = (void __user *) control;
943 kmsg->msg.msg_controllen = kmsg->controllen;
944 }
945
946 sr->buf = *buf; /* stash for later copy */
947 *buf = (void __user *) (ubuf + hdr);
948 kmsg->payloadlen = *len = *len - hdr;
949 return 0;
950 }
951
952 struct io_recvmsg_multishot_hdr {
953 struct io_uring_recvmsg_out msg;
954 struct sockaddr_storage addr;
955 };
956
io_recvmsg_multishot(struct socket * sock,struct io_sr_msg * io,struct io_async_msghdr * kmsg,unsigned int flags,bool * finished)957 static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
958 struct io_async_msghdr *kmsg,
959 unsigned int flags, bool *finished)
960 {
961 int err;
962 int copy_len;
963 struct io_recvmsg_multishot_hdr hdr;
964
965 if (kmsg->namelen)
966 kmsg->msg.msg_name = &hdr.addr;
967 kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
968 kmsg->msg.msg_namelen = 0;
969
970 if (sock->file->f_flags & O_NONBLOCK)
971 flags |= MSG_DONTWAIT;
972
973 err = sock_recvmsg(sock, &kmsg->msg, flags);
974 *finished = err <= 0;
975 if (err < 0)
976 return err;
977
978 hdr.msg = (struct io_uring_recvmsg_out) {
979 .controllen = kmsg->controllen - kmsg->msg.msg_controllen,
980 .flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
981 };
982
983 hdr.msg.payloadlen = err;
984 if (err > kmsg->payloadlen)
985 err = kmsg->payloadlen;
986
987 copy_len = sizeof(struct io_uring_recvmsg_out);
988 if (kmsg->msg.msg_namelen > kmsg->namelen)
989 copy_len += kmsg->namelen;
990 else
991 copy_len += kmsg->msg.msg_namelen;
992
993 /*
994 * "fromlen shall refer to the value before truncation.."
995 * 1003.1g
996 */
997 hdr.msg.namelen = kmsg->msg.msg_namelen;
998
999 /* ensure that there is no gap between hdr and sockaddr_storage */
1000 BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
1001 sizeof(struct io_uring_recvmsg_out));
1002 if (copy_to_user(io->buf, &hdr, copy_len)) {
1003 *finished = true;
1004 return -EFAULT;
1005 }
1006
1007 return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
1008 kmsg->controllen + err;
1009 }
1010
io_recvmsg(struct io_kiocb * req,unsigned int issue_flags)1011 int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
1012 {
1013 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1014 struct io_async_msghdr *kmsg = req->async_data;
1015 struct io_br_sel sel = { };
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 sel.buf_list = NULL;
1036 if (io_do_buffer_select(req)) {
1037 size_t len = sr->len;
1038
1039 sel = io_buffer_select(req, &len, sr->buf_group, issue_flags);
1040 if (!sel.addr)
1041 return -ENOBUFS;
1042
1043 if (req->flags & REQ_F_APOLL_MULTISHOT) {
1044 ret = io_recvmsg_prep_multishot(kmsg, sr, &sel.addr, &len);
1045 if (ret) {
1046 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1047 return ret;
1048 }
1049 }
1050
1051 iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_DEST, sel.addr, 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 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1071 return IOU_RETRY;
1072 }
1073 if (ret > 0 && io_net_retry(sock, flags)) {
1074 sr->done_io += ret;
1075 return io_net_kbuf_recyle(req, sel.buf_list, kmsg, ret);
1076 }
1077 if (ret == -ERESTARTSYS)
1078 ret = -EINTR;
1079 req_set_fail(req);
1080 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
1081 req_set_fail(req);
1082 }
1083
1084 if (ret > 0)
1085 ret += sr->done_io;
1086 else if (sr->done_io)
1087 ret = sr->done_io;
1088 else
1089 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1090
1091 sel.val = ret;
1092 if (!io_recv_finish(req, kmsg, &sel, mshot_finished, issue_flags))
1093 goto retry_multishot;
1094
1095 return sel.val;
1096 }
1097
io_recv_buf_select(struct io_kiocb * req,struct io_async_msghdr * kmsg,struct io_br_sel * sel,unsigned int issue_flags)1098 static int io_recv_buf_select(struct io_kiocb *req, struct io_async_msghdr *kmsg,
1099 struct io_br_sel *sel, unsigned int issue_flags)
1100 {
1101 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1102 int ret;
1103
1104 /*
1105 * If the ring isn't locked, then don't use the peek interface
1106 * to grab multiple buffers as we will lock/unlock between
1107 * this selection and posting the buffers.
1108 */
1109 if (!(issue_flags & IO_URING_F_UNLOCKED) &&
1110 sr->flags & IORING_RECVSEND_BUNDLE) {
1111 struct buf_sel_arg arg = {
1112 .iovs = &kmsg->fast_iov,
1113 .nr_iovs = 1,
1114 .mode = KBUF_MODE_EXPAND,
1115 .buf_group = sr->buf_group,
1116 };
1117
1118 if (kmsg->vec.iovec) {
1119 arg.nr_iovs = kmsg->vec.nr;
1120 arg.iovs = kmsg->vec.iovec;
1121 arg.mode |= KBUF_MODE_FREE;
1122 }
1123
1124 if (sel->val)
1125 arg.max_len = sel->val;
1126 else if (kmsg->msg.msg_inq > 1)
1127 arg.max_len = min_not_zero(sel->val, (ssize_t) kmsg->msg.msg_inq);
1128
1129 /* if mshot limited, ensure we don't go over */
1130 if (sr->flags & IORING_RECV_MSHOT_LIM)
1131 arg.max_len = min_not_zero(arg.max_len, sr->mshot_total_len);
1132 ret = io_buffers_peek(req, &arg, sel);
1133 if (unlikely(ret < 0))
1134 return ret;
1135
1136 if (arg.iovs != &kmsg->fast_iov && arg.iovs != kmsg->vec.iovec) {
1137 kmsg->vec.nr = ret;
1138 kmsg->vec.iovec = arg.iovs;
1139 req->flags |= REQ_F_NEED_CLEANUP;
1140 }
1141 if (arg.partial_map)
1142 sr->flags |= IORING_RECV_PARTIAL_MAP;
1143
1144 /* special case 1 vec, can be a fast path */
1145 if (ret == 1) {
1146 sr->buf = arg.iovs[0].iov_base;
1147 sr->len = arg.iovs[0].iov_len;
1148 goto map_ubuf;
1149 }
1150 iov_iter_init(&kmsg->msg.msg_iter, ITER_DEST, arg.iovs, ret,
1151 arg.out_len);
1152 } else {
1153 size_t len = sel->val;
1154
1155 *sel = io_buffer_select(req, &len, sr->buf_group, issue_flags);
1156 if (!sel->addr)
1157 return -ENOBUFS;
1158 sr->buf = sel->addr;
1159 sr->len = len;
1160 map_ubuf:
1161 ret = import_ubuf(ITER_DEST, sr->buf, sr->len,
1162 &kmsg->msg.msg_iter);
1163 if (unlikely(ret))
1164 return ret;
1165 }
1166
1167 return 0;
1168 }
1169
io_recv(struct io_kiocb * req,unsigned int issue_flags)1170 int io_recv(struct io_kiocb *req, unsigned int issue_flags)
1171 {
1172 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1173 struct io_async_msghdr *kmsg = req->async_data;
1174 struct io_br_sel sel;
1175 struct socket *sock;
1176 unsigned flags;
1177 int ret, min_ret = 0;
1178 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1179 bool mshot_finished;
1180
1181 if (!(req->flags & REQ_F_POLLED) &&
1182 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1183 return -EAGAIN;
1184
1185 sock = sock_from_file(req->file);
1186 if (unlikely(!sock))
1187 return -ENOTSOCK;
1188
1189 flags = sr->msg_flags;
1190 if (force_nonblock)
1191 flags |= MSG_DONTWAIT;
1192
1193 retry_multishot:
1194 sel.buf_list = NULL;
1195 if (io_do_buffer_select(req)) {
1196 sel.val = sr->len;
1197 ret = io_recv_buf_select(req, kmsg, &sel, issue_flags);
1198 if (unlikely(ret < 0)) {
1199 kmsg->msg.msg_inq = -1;
1200 goto out_free;
1201 }
1202 sr->buf = NULL;
1203 }
1204
1205 kmsg->msg.msg_flags = 0;
1206 kmsg->msg.msg_inq = -1;
1207
1208 if (flags & MSG_WAITALL)
1209 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1210
1211 ret = sock_recvmsg(sock, &kmsg->msg, flags);
1212 if (ret < min_ret) {
1213 if (ret == -EAGAIN && force_nonblock) {
1214 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1215 return IOU_RETRY;
1216 }
1217 if (ret > 0 && io_net_retry(sock, flags)) {
1218 sr->len -= ret;
1219 sr->buf += ret;
1220 sr->done_io += ret;
1221 return io_net_kbuf_recyle(req, sel.buf_list, kmsg, ret);
1222 }
1223 if (ret == -ERESTARTSYS)
1224 ret = -EINTR;
1225 req_set_fail(req);
1226 } else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
1227 out_free:
1228 req_set_fail(req);
1229 }
1230
1231 mshot_finished = ret <= 0;
1232 if (ret > 0)
1233 ret += sr->done_io;
1234 else if (sr->done_io)
1235 ret = sr->done_io;
1236 else
1237 io_kbuf_recycle(req, sel.buf_list, issue_flags);
1238
1239 sel.val = ret;
1240 if (!io_recv_finish(req, kmsg, &sel, mshot_finished, issue_flags))
1241 goto retry_multishot;
1242
1243 return sel.val;
1244 }
1245
io_recvzc_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1246 int io_recvzc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1247 {
1248 struct io_recvzc *zc = io_kiocb_to_cmd(req, struct io_recvzc);
1249 unsigned ifq_idx;
1250
1251 if (unlikely(sqe->addr2 || sqe->addr || sqe->addr3))
1252 return -EINVAL;
1253
1254 ifq_idx = READ_ONCE(sqe->zcrx_ifq_idx);
1255 zc->ifq = xa_load(&req->ctx->zcrx_ctxs, ifq_idx);
1256 if (!zc->ifq)
1257 return -EINVAL;
1258
1259 zc->len = READ_ONCE(sqe->len);
1260 zc->flags = READ_ONCE(sqe->ioprio);
1261 if (READ_ONCE(sqe->msg_flags))
1262 return -EINVAL;
1263 if (zc->flags & ~(IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT))
1264 return -EINVAL;
1265 /* multishot required */
1266 if (!(zc->flags & IORING_RECV_MULTISHOT))
1267 return -EINVAL;
1268 /* All data completions are posted as aux CQEs. */
1269 req->flags |= REQ_F_APOLL_MULTISHOT;
1270
1271 return 0;
1272 }
1273
io_recvzc(struct io_kiocb * req,unsigned int issue_flags)1274 int io_recvzc(struct io_kiocb *req, unsigned int issue_flags)
1275 {
1276 struct io_recvzc *zc = io_kiocb_to_cmd(req, struct io_recvzc);
1277 struct socket *sock;
1278 unsigned int len;
1279 int ret;
1280
1281 if (!(req->flags & REQ_F_POLLED) &&
1282 (zc->flags & IORING_RECVSEND_POLL_FIRST))
1283 return -EAGAIN;
1284
1285 sock = sock_from_file(req->file);
1286 if (unlikely(!sock))
1287 return -ENOTSOCK;
1288
1289 len = zc->len;
1290 ret = io_zcrx_recv(req, zc->ifq, sock, 0, issue_flags, &zc->len);
1291 if (len && zc->len == 0) {
1292 io_req_set_res(req, 0, 0);
1293
1294 return IOU_COMPLETE;
1295 }
1296 if (unlikely(ret <= 0) && ret != -EAGAIN) {
1297 if (ret == -ERESTARTSYS)
1298 ret = -EINTR;
1299 if (ret == IOU_REQUEUE)
1300 return IOU_REQUEUE;
1301
1302 req_set_fail(req);
1303 io_req_set_res(req, ret, 0);
1304 return IOU_COMPLETE;
1305 }
1306 return IOU_RETRY;
1307 }
1308
io_send_zc_cleanup(struct io_kiocb * req)1309 void io_send_zc_cleanup(struct io_kiocb *req)
1310 {
1311 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1312 struct io_async_msghdr *io = req->async_data;
1313
1314 if (req_has_async_data(req))
1315 io_netmsg_iovec_free(io);
1316 if (zc->notif) {
1317 io_notif_flush(zc->notif);
1318 zc->notif = NULL;
1319 }
1320 }
1321
1322 #define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF)
1323 #define IO_ZC_FLAGS_VALID (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE | \
1324 IORING_SEND_VECTORIZED)
1325
io_send_zc_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1326 int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1327 {
1328 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1329 struct io_ring_ctx *ctx = req->ctx;
1330 struct io_async_msghdr *iomsg;
1331 struct io_kiocb *notif;
1332 int ret;
1333
1334 zc->done_io = 0;
1335
1336 if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
1337 return -EINVAL;
1338 /* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
1339 if (req->flags & REQ_F_CQE_SKIP)
1340 return -EINVAL;
1341
1342 notif = zc->notif = io_alloc_notif(ctx);
1343 if (!notif)
1344 return -ENOMEM;
1345 notif->cqe.user_data = req->cqe.user_data;
1346 notif->cqe.res = 0;
1347 notif->cqe.flags = IORING_CQE_F_NOTIF;
1348 req->flags |= REQ_F_NEED_CLEANUP | REQ_F_POLL_NO_LAZY;
1349
1350 zc->flags = READ_ONCE(sqe->ioprio);
1351 if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) {
1352 if (zc->flags & ~IO_ZC_FLAGS_VALID)
1353 return -EINVAL;
1354 if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) {
1355 struct io_notif_data *nd = io_notif_to_data(notif);
1356
1357 nd->zc_report = true;
1358 nd->zc_used = false;
1359 nd->zc_copied = false;
1360 }
1361 }
1362
1363 zc->len = READ_ONCE(sqe->len);
1364 zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL | MSG_ZEROCOPY;
1365 req->buf_index = READ_ONCE(sqe->buf_index);
1366 if (zc->msg_flags & MSG_DONTWAIT)
1367 req->flags |= REQ_F_NOWAIT;
1368
1369 if (io_is_compat(req->ctx))
1370 zc->msg_flags |= MSG_CMSG_COMPAT;
1371
1372 iomsg = io_msg_alloc_async(req);
1373 if (unlikely(!iomsg))
1374 return -ENOMEM;
1375
1376 if (req->opcode == IORING_OP_SEND_ZC) {
1377 ret = io_send_setup(req, sqe);
1378 } else {
1379 if (unlikely(sqe->addr2 || sqe->file_index))
1380 return -EINVAL;
1381 ret = io_sendmsg_setup(req, sqe);
1382 }
1383 if (unlikely(ret))
1384 return ret;
1385
1386 if (!(zc->flags & IORING_RECVSEND_FIXED_BUF)) {
1387 iomsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1388 return io_notif_account_mem(zc->notif, iomsg->msg.msg_iter.count);
1389 }
1390 iomsg->msg.sg_from_iter = io_sg_from_iter;
1391 return 0;
1392 }
1393
io_sg_from_iter_iovec(struct sk_buff * skb,struct iov_iter * from,size_t length)1394 static int io_sg_from_iter_iovec(struct sk_buff *skb,
1395 struct iov_iter *from, size_t length)
1396 {
1397 skb_zcopy_downgrade_managed(skb);
1398 return zerocopy_fill_skb_from_iter(skb, from, length);
1399 }
1400
io_sg_from_iter(struct sk_buff * skb,struct iov_iter * from,size_t length)1401 static int io_sg_from_iter(struct sk_buff *skb,
1402 struct iov_iter *from, size_t length)
1403 {
1404 struct skb_shared_info *shinfo = skb_shinfo(skb);
1405 int frag = shinfo->nr_frags;
1406 int ret = 0;
1407 struct bvec_iter bi;
1408 ssize_t copied = 0;
1409 unsigned long truesize = 0;
1410
1411 if (!frag)
1412 shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
1413 else if (unlikely(!skb_zcopy_managed(skb)))
1414 return zerocopy_fill_skb_from_iter(skb, from, length);
1415
1416 bi.bi_size = min(from->count, length);
1417 bi.bi_bvec_done = from->iov_offset;
1418 bi.bi_idx = 0;
1419
1420 while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1421 struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1422
1423 copied += v.bv_len;
1424 truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1425 __skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1426 v.bv_offset, v.bv_len);
1427 bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1428 }
1429 if (bi.bi_size)
1430 ret = -EMSGSIZE;
1431
1432 shinfo->nr_frags = frag;
1433 from->bvec += bi.bi_idx;
1434 from->nr_segs -= bi.bi_idx;
1435 from->count -= copied;
1436 from->iov_offset = bi.bi_bvec_done;
1437
1438 skb->data_len += copied;
1439 skb->len += copied;
1440 skb->truesize += truesize;
1441 return ret;
1442 }
1443
io_send_zc_import(struct io_kiocb * req,unsigned int issue_flags)1444 static int io_send_zc_import(struct io_kiocb *req, unsigned int issue_flags)
1445 {
1446 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1447 struct io_async_msghdr *kmsg = req->async_data;
1448
1449 WARN_ON_ONCE(!(sr->flags & IORING_RECVSEND_FIXED_BUF));
1450
1451 sr->notif->buf_index = req->buf_index;
1452 return io_import_reg_buf(sr->notif, &kmsg->msg.msg_iter,
1453 (u64)(uintptr_t)sr->buf, sr->len,
1454 ITER_SOURCE, issue_flags);
1455 }
1456
io_send_zc(struct io_kiocb * req,unsigned int issue_flags)1457 int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
1458 {
1459 struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1460 struct io_async_msghdr *kmsg = req->async_data;
1461 struct socket *sock;
1462 unsigned msg_flags;
1463 int ret, min_ret = 0;
1464
1465 sock = sock_from_file(req->file);
1466 if (unlikely(!sock))
1467 return -ENOTSOCK;
1468 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1469 return -EOPNOTSUPP;
1470
1471 if (!(req->flags & REQ_F_POLLED) &&
1472 (zc->flags & IORING_RECVSEND_POLL_FIRST))
1473 return -EAGAIN;
1474
1475 if (req->flags & REQ_F_IMPORT_BUFFER) {
1476 req->flags &= ~REQ_F_IMPORT_BUFFER;
1477 ret = io_send_zc_import(req, issue_flags);
1478 if (unlikely(ret))
1479 return ret;
1480 }
1481
1482 msg_flags = zc->msg_flags;
1483 if (issue_flags & IO_URING_F_NONBLOCK)
1484 msg_flags |= MSG_DONTWAIT;
1485 if (msg_flags & MSG_WAITALL)
1486 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1487 msg_flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
1488
1489 kmsg->msg.msg_flags = msg_flags;
1490 kmsg->msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
1491 ret = sock_sendmsg(sock, &kmsg->msg);
1492
1493 if (unlikely(ret < min_ret)) {
1494 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1495 return -EAGAIN;
1496
1497 if (ret > 0 && io_net_retry(sock, kmsg->msg.msg_flags)) {
1498 zc->done_io += ret;
1499 return -EAGAIN;
1500 }
1501 if (ret == -ERESTARTSYS)
1502 ret = -EINTR;
1503 req_set_fail(req);
1504 }
1505
1506 if (ret >= 0)
1507 ret += zc->done_io;
1508 else if (zc->done_io)
1509 ret = zc->done_io;
1510
1511 /*
1512 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1513 * flushing notif to io_send_zc_cleanup()
1514 */
1515 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1516 io_notif_flush(zc->notif);
1517 zc->notif = NULL;
1518 io_req_msg_cleanup(req, 0);
1519 }
1520 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1521 return IOU_COMPLETE;
1522 }
1523
io_sendmsg_zc(struct io_kiocb * req,unsigned int issue_flags)1524 int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1525 {
1526 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1527 struct io_async_msghdr *kmsg = req->async_data;
1528 struct socket *sock;
1529 unsigned flags;
1530 int ret, min_ret = 0;
1531
1532 if (req->flags & REQ_F_IMPORT_BUFFER) {
1533 unsigned uvec_segs = kmsg->msg.msg_iter.nr_segs;
1534 int ret;
1535
1536 sr->notif->buf_index = req->buf_index;
1537 ret = io_import_reg_vec(ITER_SOURCE, &kmsg->msg.msg_iter,
1538 sr->notif, &kmsg->vec, uvec_segs,
1539 issue_flags);
1540 if (unlikely(ret))
1541 return ret;
1542 req->flags &= ~REQ_F_IMPORT_BUFFER;
1543 }
1544
1545 sock = sock_from_file(req->file);
1546 if (unlikely(!sock))
1547 return -ENOTSOCK;
1548 if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1549 return -EOPNOTSUPP;
1550
1551 if (!(req->flags & REQ_F_POLLED) &&
1552 (sr->flags & IORING_RECVSEND_POLL_FIRST))
1553 return -EAGAIN;
1554
1555 flags = sr->msg_flags;
1556 if (issue_flags & IO_URING_F_NONBLOCK)
1557 flags |= MSG_DONTWAIT;
1558 if (flags & MSG_WAITALL)
1559 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1560
1561 kmsg->msg.msg_control_user = sr->msg_control;
1562 kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1563 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1564
1565 if (unlikely(ret < min_ret)) {
1566 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1567 return -EAGAIN;
1568
1569 if (ret > 0 && io_net_retry(sock, flags)) {
1570 sr->done_io += ret;
1571 return -EAGAIN;
1572 }
1573 if (ret == -ERESTARTSYS)
1574 ret = -EINTR;
1575 req_set_fail(req);
1576 }
1577
1578 if (ret >= 0)
1579 ret += sr->done_io;
1580 else if (sr->done_io)
1581 ret = sr->done_io;
1582
1583 /*
1584 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1585 * flushing notif to io_send_zc_cleanup()
1586 */
1587 if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1588 io_notif_flush(sr->notif);
1589 sr->notif = NULL;
1590 io_req_msg_cleanup(req, 0);
1591 }
1592 io_req_set_res(req, ret, IORING_CQE_F_MORE);
1593 return IOU_COMPLETE;
1594 }
1595
io_sendrecv_fail(struct io_kiocb * req)1596 void io_sendrecv_fail(struct io_kiocb *req)
1597 {
1598 struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1599
1600 if (sr->done_io)
1601 req->cqe.res = sr->done_io;
1602
1603 if ((req->flags & REQ_F_NEED_CLEANUP) &&
1604 (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1605 req->cqe.flags |= IORING_CQE_F_MORE;
1606 }
1607
1608 #define ACCEPT_FLAGS (IORING_ACCEPT_MULTISHOT | IORING_ACCEPT_DONTWAIT | \
1609 IORING_ACCEPT_POLL_FIRST)
1610
io_accept_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1611 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1612 {
1613 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1614
1615 if (sqe->len || sqe->buf_index)
1616 return -EINVAL;
1617
1618 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1619 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1620 accept->flags = READ_ONCE(sqe->accept_flags);
1621 accept->nofile = rlimit(RLIMIT_NOFILE);
1622 accept->iou_flags = READ_ONCE(sqe->ioprio);
1623 if (accept->iou_flags & ~ACCEPT_FLAGS)
1624 return -EINVAL;
1625
1626 accept->file_slot = READ_ONCE(sqe->file_index);
1627 if (accept->file_slot) {
1628 if (accept->flags & SOCK_CLOEXEC)
1629 return -EINVAL;
1630 if (accept->iou_flags & IORING_ACCEPT_MULTISHOT &&
1631 accept->file_slot != IORING_FILE_INDEX_ALLOC)
1632 return -EINVAL;
1633 }
1634 if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1635 return -EINVAL;
1636 if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1637 accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1638 if (accept->iou_flags & IORING_ACCEPT_MULTISHOT)
1639 req->flags |= REQ_F_APOLL_MULTISHOT;
1640 if (accept->iou_flags & IORING_ACCEPT_DONTWAIT)
1641 req->flags |= REQ_F_NOWAIT;
1642 return 0;
1643 }
1644
io_accept(struct io_kiocb * req,unsigned int issue_flags)1645 int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1646 {
1647 struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1648 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1649 bool fixed = !!accept->file_slot;
1650 struct proto_accept_arg arg = {
1651 .flags = force_nonblock ? O_NONBLOCK : 0,
1652 };
1653 struct file *file;
1654 unsigned cflags;
1655 int ret, fd;
1656
1657 if (!(req->flags & REQ_F_POLLED) &&
1658 accept->iou_flags & IORING_ACCEPT_POLL_FIRST)
1659 return -EAGAIN;
1660
1661 retry:
1662 if (!fixed) {
1663 fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1664 if (unlikely(fd < 0))
1665 return fd;
1666 }
1667 arg.err = 0;
1668 arg.is_empty = -1;
1669 file = do_accept(req->file, &arg, accept->addr, accept->addr_len,
1670 accept->flags);
1671 if (IS_ERR(file)) {
1672 if (!fixed)
1673 put_unused_fd(fd);
1674 ret = PTR_ERR(file);
1675 if (ret == -EAGAIN && force_nonblock &&
1676 !(accept->iou_flags & IORING_ACCEPT_DONTWAIT))
1677 return IOU_RETRY;
1678
1679 if (ret == -ERESTARTSYS)
1680 ret = -EINTR;
1681 } else if (!fixed) {
1682 fd_install(fd, file);
1683 ret = fd;
1684 } else {
1685 ret = io_fixed_fd_install(req, issue_flags, file,
1686 accept->file_slot);
1687 }
1688
1689 cflags = 0;
1690 if (!arg.is_empty)
1691 cflags |= IORING_CQE_F_SOCK_NONEMPTY;
1692
1693 if (ret >= 0 && (req->flags & REQ_F_APOLL_MULTISHOT) &&
1694 io_req_post_cqe(req, ret, cflags | IORING_CQE_F_MORE)) {
1695 if (cflags & IORING_CQE_F_SOCK_NONEMPTY || arg.is_empty == -1)
1696 goto retry;
1697 return IOU_RETRY;
1698 }
1699
1700 io_req_set_res(req, ret, cflags);
1701 if (ret < 0)
1702 req_set_fail(req);
1703 return IOU_COMPLETE;
1704 }
1705
io_socket_bpf_populate(struct io_uring_bpf_ctx * bctx,struct io_kiocb * req)1706 void io_socket_bpf_populate(struct io_uring_bpf_ctx *bctx, struct io_kiocb *req)
1707 {
1708 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1709
1710 bctx->socket.family = sock->domain;
1711 bctx->socket.type = sock->type;
1712 bctx->socket.protocol = sock->protocol;
1713 }
1714
io_socket_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1715 int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1716 {
1717 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1718
1719 if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1720 return -EINVAL;
1721
1722 sock->domain = READ_ONCE(sqe->fd);
1723 sock->type = READ_ONCE(sqe->off);
1724 sock->protocol = READ_ONCE(sqe->len);
1725 sock->file_slot = READ_ONCE(sqe->file_index);
1726 sock->nofile = rlimit(RLIMIT_NOFILE);
1727
1728 sock->flags = sock->type & ~SOCK_TYPE_MASK;
1729 if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1730 return -EINVAL;
1731 if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1732 return -EINVAL;
1733 return 0;
1734 }
1735
io_socket(struct io_kiocb * req,unsigned int issue_flags)1736 int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1737 {
1738 struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1739 bool fixed = !!sock->file_slot;
1740 struct file *file;
1741 int ret, fd;
1742
1743 if (!fixed) {
1744 fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1745 if (unlikely(fd < 0))
1746 return fd;
1747 }
1748 file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1749 if (IS_ERR(file)) {
1750 if (!fixed)
1751 put_unused_fd(fd);
1752 ret = PTR_ERR(file);
1753 if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1754 return -EAGAIN;
1755 if (ret == -ERESTARTSYS)
1756 ret = -EINTR;
1757 req_set_fail(req);
1758 } else if (!fixed) {
1759 fd_install(fd, file);
1760 ret = fd;
1761 } else {
1762 ret = io_fixed_fd_install(req, issue_flags, file,
1763 sock->file_slot);
1764 }
1765 io_req_set_res(req, ret, 0);
1766 return IOU_COMPLETE;
1767 }
1768
io_connect_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1769 int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1770 {
1771 struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1772 struct io_async_msghdr *io;
1773
1774 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1775 return -EINVAL;
1776
1777 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1778 conn->addr_len = READ_ONCE(sqe->addr2);
1779 conn->in_progress = conn->seen_econnaborted = false;
1780
1781 io = io_msg_alloc_async(req);
1782 if (unlikely(!io))
1783 return -ENOMEM;
1784
1785 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->addr);
1786 }
1787
io_connect(struct io_kiocb * req,unsigned int issue_flags)1788 int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1789 {
1790 struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1791 struct io_async_msghdr *io = req->async_data;
1792 unsigned file_flags;
1793 int ret;
1794 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1795
1796 if (connect->in_progress) {
1797 struct poll_table_struct pt = { ._key = EPOLLERR };
1798
1799 if (vfs_poll(req->file, &pt) & EPOLLERR)
1800 goto get_sock_err;
1801 }
1802
1803 file_flags = force_nonblock ? O_NONBLOCK : 0;
1804
1805 ret = __sys_connect_file(req->file, &io->addr, connect->addr_len,
1806 file_flags);
1807 if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED)
1808 && force_nonblock) {
1809 if (ret == -EINPROGRESS) {
1810 connect->in_progress = true;
1811 } else if (ret == -ECONNABORTED) {
1812 if (connect->seen_econnaborted)
1813 goto out;
1814 connect->seen_econnaborted = true;
1815 }
1816 return -EAGAIN;
1817 }
1818 if (connect->in_progress) {
1819 /*
1820 * At least bluetooth will return -EBADFD on a re-connect
1821 * attempt, and it's (supposedly) also valid to get -EISCONN
1822 * which means the previous result is good. For both of these,
1823 * grab the sock_error() and use that for the completion.
1824 */
1825 if (ret == -EBADFD || ret == -EISCONN) {
1826 get_sock_err:
1827 ret = sock_error(sock_from_file(req->file)->sk);
1828 }
1829 }
1830 if (ret == -ERESTARTSYS)
1831 ret = -EINTR;
1832 out:
1833 if (ret < 0)
1834 req_set_fail(req);
1835 io_req_msg_cleanup(req, issue_flags);
1836 io_req_set_res(req, ret, 0);
1837 return IOU_COMPLETE;
1838 }
1839
io_bind_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1840 int io_bind_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1841 {
1842 struct io_bind *bind = io_kiocb_to_cmd(req, struct io_bind);
1843 struct sockaddr __user *uaddr;
1844 struct io_async_msghdr *io;
1845
1846 if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1847 return -EINVAL;
1848
1849 uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1850 bind->addr_len = READ_ONCE(sqe->addr2);
1851
1852 io = io_msg_alloc_async(req);
1853 if (unlikely(!io))
1854 return -ENOMEM;
1855 return move_addr_to_kernel(uaddr, bind->addr_len, &io->addr);
1856 }
1857
io_bind(struct io_kiocb * req,unsigned int issue_flags)1858 int io_bind(struct io_kiocb *req, unsigned int issue_flags)
1859 {
1860 struct io_bind *bind = io_kiocb_to_cmd(req, struct io_bind);
1861 struct io_async_msghdr *io = req->async_data;
1862 struct socket *sock;
1863 int ret;
1864
1865 sock = sock_from_file(req->file);
1866 if (unlikely(!sock))
1867 return -ENOTSOCK;
1868
1869 ret = __sys_bind_socket(sock, &io->addr, bind->addr_len);
1870 if (ret < 0)
1871 req_set_fail(req);
1872 io_req_set_res(req, ret, 0);
1873 return 0;
1874 }
1875
io_listen_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)1876 int io_listen_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1877 {
1878 struct io_listen *listen = io_kiocb_to_cmd(req, struct io_listen);
1879
1880 if (sqe->addr || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in || sqe->addr2)
1881 return -EINVAL;
1882
1883 listen->backlog = READ_ONCE(sqe->len);
1884 return 0;
1885 }
1886
io_listen(struct io_kiocb * req,unsigned int issue_flags)1887 int io_listen(struct io_kiocb *req, unsigned int issue_flags)
1888 {
1889 struct io_listen *listen = io_kiocb_to_cmd(req, struct io_listen);
1890 struct socket *sock;
1891 int ret;
1892
1893 sock = sock_from_file(req->file);
1894 if (unlikely(!sock))
1895 return -ENOTSOCK;
1896
1897 ret = __sys_listen_socket(sock, listen->backlog);
1898 if (ret < 0)
1899 req_set_fail(req);
1900 io_req_set_res(req, ret, 0);
1901 return 0;
1902 }
1903
io_netmsg_cache_free(const void * entry)1904 void io_netmsg_cache_free(const void *entry)
1905 {
1906 struct io_async_msghdr *kmsg = (struct io_async_msghdr *) entry;
1907
1908 io_vec_free(&kmsg->vec);
1909 kfree(kmsg);
1910 }
1911