xref: /linux/io_uring/net.c (revision e1c4c5436b4ad579762fbe78bfabc8aef59bd5b1)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
3 #include <linux/errno.h>
4 #include <linux/file.h>
5 #include <linux/slab.h>
6 #include <linux/net.h>
7 #include <linux/compat.h>
8 #include <net/compat.h>
9 #include <linux/io_uring.h>
10 
11 #include <uapi/linux/io_uring.h>
12 
13 #include "io_uring.h"
14 #include "kbuf.h"
15 #include "alloc_cache.h"
16 #include "net.h"
17 #include "notif.h"
18 #include "rsrc.h"
19 
20 #if defined(CONFIG_NET)
21 struct io_shutdown {
22 	struct file			*file;
23 	int				how;
24 };
25 
26 struct io_accept {
27 	struct file			*file;
28 	struct sockaddr __user		*addr;
29 	int __user			*addr_len;
30 	int				flags;
31 	u32				file_slot;
32 	unsigned long			nofile;
33 };
34 
35 struct io_socket {
36 	struct file			*file;
37 	int				domain;
38 	int				type;
39 	int				protocol;
40 	int				flags;
41 	u32				file_slot;
42 	unsigned long			nofile;
43 };
44 
45 struct io_connect {
46 	struct file			*file;
47 	struct sockaddr __user		*addr;
48 	int				addr_len;
49 	bool				in_progress;
50 };
51 
52 struct io_sr_msg {
53 	struct file			*file;
54 	union {
55 		struct compat_msghdr __user	*umsg_compat;
56 		struct user_msghdr __user	*umsg;
57 		void __user			*buf;
58 	};
59 	unsigned			len;
60 	unsigned			done_io;
61 	unsigned			msg_flags;
62 	u16				flags;
63 	/* initialised and used only by !msg send variants */
64 	u16				addr_len;
65 	void __user			*addr;
66 	/* used only for send zerocopy */
67 	struct io_kiocb 		*notif;
68 };
69 
70 #define IO_APOLL_MULTI_POLLED (REQ_F_APOLL_MULTISHOT | REQ_F_POLLED)
71 
72 int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
73 {
74 	struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
75 
76 	if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
77 		     sqe->buf_index || sqe->splice_fd_in))
78 		return -EINVAL;
79 
80 	shutdown->how = READ_ONCE(sqe->len);
81 	return 0;
82 }
83 
84 int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
85 {
86 	struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
87 	struct socket *sock;
88 	int ret;
89 
90 	if (issue_flags & IO_URING_F_NONBLOCK)
91 		return -EAGAIN;
92 
93 	sock = sock_from_file(req->file);
94 	if (unlikely(!sock))
95 		return -ENOTSOCK;
96 
97 	ret = __sys_shutdown_sock(sock, shutdown->how);
98 	io_req_set_res(req, ret, 0);
99 	return IOU_OK;
100 }
101 
102 static bool io_net_retry(struct socket *sock, int flags)
103 {
104 	if (!(flags & MSG_WAITALL))
105 		return false;
106 	return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
107 }
108 
109 static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
110 {
111 	struct io_async_msghdr *hdr = req->async_data;
112 
113 	if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
114 		return;
115 
116 	/* Let normal cleanup path reap it if we fail adding to the cache */
117 	if (io_alloc_cache_put(&req->ctx->netmsg_cache, &hdr->cache)) {
118 		req->async_data = NULL;
119 		req->flags &= ~REQ_F_ASYNC_DATA;
120 	}
121 }
122 
123 static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
124 						  unsigned int issue_flags)
125 {
126 	struct io_ring_ctx *ctx = req->ctx;
127 	struct io_cache_entry *entry;
128 	struct io_async_msghdr *hdr;
129 
130 	if (!(issue_flags & IO_URING_F_UNLOCKED) &&
131 	    (entry = io_alloc_cache_get(&ctx->netmsg_cache)) != NULL) {
132 		hdr = container_of(entry, struct io_async_msghdr, cache);
133 		hdr->free_iov = NULL;
134 		req->flags |= REQ_F_ASYNC_DATA;
135 		req->async_data = hdr;
136 		return hdr;
137 	}
138 
139 	if (!io_alloc_async_data(req)) {
140 		hdr = req->async_data;
141 		hdr->free_iov = NULL;
142 		return hdr;
143 	}
144 	return NULL;
145 }
146 
147 static inline struct io_async_msghdr *io_msg_alloc_async_prep(struct io_kiocb *req)
148 {
149 	/* ->prep_async is always called from the submission context */
150 	return io_msg_alloc_async(req, 0);
151 }
152 
153 static int io_setup_async_msg(struct io_kiocb *req,
154 			      struct io_async_msghdr *kmsg,
155 			      unsigned int issue_flags)
156 {
157 	struct io_async_msghdr *async_msg;
158 
159 	if (req_has_async_data(req))
160 		return -EAGAIN;
161 	async_msg = io_msg_alloc_async(req, issue_flags);
162 	if (!async_msg) {
163 		kfree(kmsg->free_iov);
164 		return -ENOMEM;
165 	}
166 	req->flags |= REQ_F_NEED_CLEANUP;
167 	memcpy(async_msg, kmsg, sizeof(*kmsg));
168 	if (async_msg->msg.msg_name)
169 		async_msg->msg.msg_name = &async_msg->addr;
170 	/* if were using fast_iov, set it to the new one */
171 	if (!kmsg->free_iov) {
172 		size_t fast_idx = kmsg->msg.msg_iter.iov - kmsg->fast_iov;
173 		async_msg->msg.msg_iter.iov = &async_msg->fast_iov[fast_idx];
174 	}
175 
176 	return -EAGAIN;
177 }
178 
179 static int io_sendmsg_copy_hdr(struct io_kiocb *req,
180 			       struct io_async_msghdr *iomsg)
181 {
182 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
183 
184 	iomsg->msg.msg_name = &iomsg->addr;
185 	iomsg->free_iov = iomsg->fast_iov;
186 	return sendmsg_copy_msghdr(&iomsg->msg, sr->umsg, sr->msg_flags,
187 					&iomsg->free_iov);
188 }
189 
190 int io_send_prep_async(struct io_kiocb *req)
191 {
192 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
193 	struct io_async_msghdr *io;
194 	int ret;
195 
196 	if (!zc->addr || req_has_async_data(req))
197 		return 0;
198 	io = io_msg_alloc_async_prep(req);
199 	if (!io)
200 		return -ENOMEM;
201 	ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
202 	return ret;
203 }
204 
205 static int io_setup_async_addr(struct io_kiocb *req,
206 			      struct sockaddr_storage *addr_storage,
207 			      unsigned int issue_flags)
208 {
209 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
210 	struct io_async_msghdr *io;
211 
212 	if (!sr->addr || req_has_async_data(req))
213 		return -EAGAIN;
214 	io = io_msg_alloc_async(req, issue_flags);
215 	if (!io)
216 		return -ENOMEM;
217 	memcpy(&io->addr, addr_storage, sizeof(io->addr));
218 	return -EAGAIN;
219 }
220 
221 int io_sendmsg_prep_async(struct io_kiocb *req)
222 {
223 	int ret;
224 
225 	if (!io_msg_alloc_async_prep(req))
226 		return -ENOMEM;
227 	ret = io_sendmsg_copy_hdr(req, req->async_data);
228 	if (!ret)
229 		req->flags |= REQ_F_NEED_CLEANUP;
230 	return ret;
231 }
232 
233 void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
234 {
235 	struct io_async_msghdr *io = req->async_data;
236 
237 	kfree(io->free_iov);
238 }
239 
240 int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
241 {
242 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
243 
244 	if (req->opcode == IORING_OP_SEND) {
245 		if (READ_ONCE(sqe->__pad3[0]))
246 			return -EINVAL;
247 		sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
248 		sr->addr_len = READ_ONCE(sqe->addr_len);
249 	} else if (sqe->addr2 || sqe->file_index) {
250 		return -EINVAL;
251 	}
252 
253 	sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
254 	sr->len = READ_ONCE(sqe->len);
255 	sr->flags = READ_ONCE(sqe->ioprio);
256 	if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
257 		return -EINVAL;
258 	sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
259 	if (sr->msg_flags & MSG_DONTWAIT)
260 		req->flags |= REQ_F_NOWAIT;
261 
262 #ifdef CONFIG_COMPAT
263 	if (req->ctx->compat)
264 		sr->msg_flags |= MSG_CMSG_COMPAT;
265 #endif
266 	sr->done_io = 0;
267 	return 0;
268 }
269 
270 int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
271 {
272 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
273 	struct io_async_msghdr iomsg, *kmsg;
274 	struct socket *sock;
275 	unsigned flags;
276 	int min_ret = 0;
277 	int ret;
278 
279 	sock = sock_from_file(req->file);
280 	if (unlikely(!sock))
281 		return -ENOTSOCK;
282 
283 	if (req_has_async_data(req)) {
284 		kmsg = req->async_data;
285 	} else {
286 		ret = io_sendmsg_copy_hdr(req, &iomsg);
287 		if (ret)
288 			return ret;
289 		kmsg = &iomsg;
290 	}
291 
292 	if (!(req->flags & REQ_F_POLLED) &&
293 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
294 		return io_setup_async_msg(req, kmsg, issue_flags);
295 
296 	flags = sr->msg_flags;
297 	if (issue_flags & IO_URING_F_NONBLOCK)
298 		flags |= MSG_DONTWAIT;
299 	if (flags & MSG_WAITALL)
300 		min_ret = iov_iter_count(&kmsg->msg.msg_iter);
301 
302 	ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
303 
304 	if (ret < min_ret) {
305 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
306 			return io_setup_async_msg(req, kmsg, issue_flags);
307 		if (ret > 0 && io_net_retry(sock, flags)) {
308 			sr->done_io += ret;
309 			req->flags |= REQ_F_PARTIAL_IO;
310 			return io_setup_async_msg(req, kmsg, issue_flags);
311 		}
312 		if (ret == -ERESTARTSYS)
313 			ret = -EINTR;
314 		req_set_fail(req);
315 	}
316 	/* fast path, check for non-NULL to avoid function call */
317 	if (kmsg->free_iov)
318 		kfree(kmsg->free_iov);
319 	req->flags &= ~REQ_F_NEED_CLEANUP;
320 	io_netmsg_recycle(req, issue_flags);
321 	if (ret >= 0)
322 		ret += sr->done_io;
323 	else if (sr->done_io)
324 		ret = sr->done_io;
325 	io_req_set_res(req, ret, 0);
326 	return IOU_OK;
327 }
328 
329 int io_send(struct io_kiocb *req, unsigned int issue_flags)
330 {
331 	struct sockaddr_storage __address;
332 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
333 	struct msghdr msg;
334 	struct iovec iov;
335 	struct socket *sock;
336 	unsigned flags;
337 	int min_ret = 0;
338 	int ret;
339 
340 	msg.msg_name = NULL;
341 	msg.msg_control = NULL;
342 	msg.msg_controllen = 0;
343 	msg.msg_namelen = 0;
344 	msg.msg_ubuf = NULL;
345 
346 	if (sr->addr) {
347 		if (req_has_async_data(req)) {
348 			struct io_async_msghdr *io = req->async_data;
349 
350 			msg.msg_name = &io->addr;
351 		} else {
352 			ret = move_addr_to_kernel(sr->addr, sr->addr_len, &__address);
353 			if (unlikely(ret < 0))
354 				return ret;
355 			msg.msg_name = (struct sockaddr *)&__address;
356 		}
357 		msg.msg_namelen = sr->addr_len;
358 	}
359 
360 	if (!(req->flags & REQ_F_POLLED) &&
361 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
362 		return io_setup_async_addr(req, &__address, issue_flags);
363 
364 	sock = sock_from_file(req->file);
365 	if (unlikely(!sock))
366 		return -ENOTSOCK;
367 
368 	ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
369 	if (unlikely(ret))
370 		return ret;
371 
372 	flags = sr->msg_flags;
373 	if (issue_flags & IO_URING_F_NONBLOCK)
374 		flags |= MSG_DONTWAIT;
375 	if (flags & MSG_WAITALL)
376 		min_ret = iov_iter_count(&msg.msg_iter);
377 
378 	msg.msg_flags = flags;
379 	ret = sock_sendmsg(sock, &msg);
380 	if (ret < min_ret) {
381 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
382 			return io_setup_async_addr(req, &__address, issue_flags);
383 
384 		if (ret > 0 && io_net_retry(sock, flags)) {
385 			sr->len -= ret;
386 			sr->buf += ret;
387 			sr->done_io += ret;
388 			req->flags |= REQ_F_PARTIAL_IO;
389 			return io_setup_async_addr(req, &__address, issue_flags);
390 		}
391 		if (ret == -ERESTARTSYS)
392 			ret = -EINTR;
393 		req_set_fail(req);
394 	}
395 	if (ret >= 0)
396 		ret += sr->done_io;
397 	else if (sr->done_io)
398 		ret = sr->done_io;
399 	io_req_set_res(req, ret, 0);
400 	return IOU_OK;
401 }
402 
403 static bool io_recvmsg_multishot_overflow(struct io_async_msghdr *iomsg)
404 {
405 	int hdr;
406 
407 	if (iomsg->namelen < 0)
408 		return true;
409 	if (check_add_overflow((int)sizeof(struct io_uring_recvmsg_out),
410 			       iomsg->namelen, &hdr))
411 		return true;
412 	if (check_add_overflow(hdr, (int)iomsg->controllen, &hdr))
413 		return true;
414 
415 	return false;
416 }
417 
418 static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
419 				 struct io_async_msghdr *iomsg)
420 {
421 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
422 	struct user_msghdr msg;
423 	int ret;
424 
425 	if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
426 		return -EFAULT;
427 
428 	ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
429 	if (ret)
430 		return ret;
431 
432 	if (req->flags & REQ_F_BUFFER_SELECT) {
433 		if (msg.msg_iovlen == 0) {
434 			sr->len = iomsg->fast_iov[0].iov_len = 0;
435 			iomsg->fast_iov[0].iov_base = NULL;
436 			iomsg->free_iov = NULL;
437 		} else if (msg.msg_iovlen > 1) {
438 			return -EINVAL;
439 		} else {
440 			if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
441 				return -EFAULT;
442 			sr->len = iomsg->fast_iov[0].iov_len;
443 			iomsg->free_iov = NULL;
444 		}
445 
446 		if (req->flags & REQ_F_APOLL_MULTISHOT) {
447 			iomsg->namelen = msg.msg_namelen;
448 			iomsg->controllen = msg.msg_controllen;
449 			if (io_recvmsg_multishot_overflow(iomsg))
450 				return -EOVERFLOW;
451 		}
452 	} else {
453 		iomsg->free_iov = iomsg->fast_iov;
454 		ret = __import_iovec(READ, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
455 				     &iomsg->free_iov, &iomsg->msg.msg_iter,
456 				     false);
457 		if (ret > 0)
458 			ret = 0;
459 	}
460 
461 	return ret;
462 }
463 
464 #ifdef CONFIG_COMPAT
465 static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
466 					struct io_async_msghdr *iomsg)
467 {
468 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
469 	struct compat_msghdr msg;
470 	struct compat_iovec __user *uiov;
471 	int ret;
472 
473 	if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
474 		return -EFAULT;
475 
476 	ret = __get_compat_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
477 	if (ret)
478 		return ret;
479 
480 	uiov = compat_ptr(msg.msg_iov);
481 	if (req->flags & REQ_F_BUFFER_SELECT) {
482 		compat_ssize_t clen;
483 
484 		if (msg.msg_iovlen == 0) {
485 			sr->len = 0;
486 		} else if (msg.msg_iovlen > 1) {
487 			return -EINVAL;
488 		} else {
489 			if (!access_ok(uiov, sizeof(*uiov)))
490 				return -EFAULT;
491 			if (__get_user(clen, &uiov->iov_len))
492 				return -EFAULT;
493 			if (clen < 0)
494 				return -EINVAL;
495 			sr->len = clen;
496 		}
497 
498 		if (req->flags & REQ_F_APOLL_MULTISHOT) {
499 			iomsg->namelen = msg.msg_namelen;
500 			iomsg->controllen = msg.msg_controllen;
501 			if (io_recvmsg_multishot_overflow(iomsg))
502 				return -EOVERFLOW;
503 		}
504 	} else {
505 		iomsg->free_iov = iomsg->fast_iov;
506 		ret = __import_iovec(READ, (struct iovec __user *)uiov, msg.msg_iovlen,
507 				   UIO_FASTIOV, &iomsg->free_iov,
508 				   &iomsg->msg.msg_iter, true);
509 		if (ret < 0)
510 			return ret;
511 	}
512 
513 	return 0;
514 }
515 #endif
516 
517 static int io_recvmsg_copy_hdr(struct io_kiocb *req,
518 			       struct io_async_msghdr *iomsg)
519 {
520 	iomsg->msg.msg_name = &iomsg->addr;
521 
522 #ifdef CONFIG_COMPAT
523 	if (req->ctx->compat)
524 		return __io_compat_recvmsg_copy_hdr(req, iomsg);
525 #endif
526 
527 	return __io_recvmsg_copy_hdr(req, iomsg);
528 }
529 
530 int io_recvmsg_prep_async(struct io_kiocb *req)
531 {
532 	int ret;
533 
534 	if (!io_msg_alloc_async_prep(req))
535 		return -ENOMEM;
536 	ret = io_recvmsg_copy_hdr(req, req->async_data);
537 	if (!ret)
538 		req->flags |= REQ_F_NEED_CLEANUP;
539 	return ret;
540 }
541 
542 #define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
543 
544 int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
545 {
546 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
547 
548 	if (unlikely(sqe->file_index || sqe->addr2))
549 		return -EINVAL;
550 
551 	sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
552 	sr->len = READ_ONCE(sqe->len);
553 	sr->flags = READ_ONCE(sqe->ioprio);
554 	if (sr->flags & ~(RECVMSG_FLAGS))
555 		return -EINVAL;
556 	sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
557 	if (sr->msg_flags & MSG_DONTWAIT)
558 		req->flags |= REQ_F_NOWAIT;
559 	if (sr->msg_flags & MSG_ERRQUEUE)
560 		req->flags |= REQ_F_CLEAR_POLLIN;
561 	if (sr->flags & IORING_RECV_MULTISHOT) {
562 		if (!(req->flags & REQ_F_BUFFER_SELECT))
563 			return -EINVAL;
564 		if (sr->msg_flags & MSG_WAITALL)
565 			return -EINVAL;
566 		if (req->opcode == IORING_OP_RECV && sr->len)
567 			return -EINVAL;
568 		req->flags |= REQ_F_APOLL_MULTISHOT;
569 	}
570 
571 #ifdef CONFIG_COMPAT
572 	if (req->ctx->compat)
573 		sr->msg_flags |= MSG_CMSG_COMPAT;
574 #endif
575 	sr->done_io = 0;
576 	return 0;
577 }
578 
579 static inline void io_recv_prep_retry(struct io_kiocb *req)
580 {
581 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
582 
583 	sr->done_io = 0;
584 	sr->len = 0; /* get from the provided buffer */
585 }
586 
587 /*
588  * Finishes io_recv and io_recvmsg.
589  *
590  * Returns true if it is actually finished, or false if it should run
591  * again (for multishot).
592  */
593 static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
594 				  unsigned int cflags, bool mshot_finished)
595 {
596 	if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
597 		io_req_set_res(req, *ret, cflags);
598 		*ret = IOU_OK;
599 		return true;
600 	}
601 
602 	if (!mshot_finished) {
603 		if (io_post_aux_cqe(req->ctx, req->cqe.user_data, *ret,
604 				    cflags | IORING_CQE_F_MORE, false)) {
605 			io_recv_prep_retry(req);
606 			return false;
607 		}
608 		/*
609 		 * Otherwise stop multishot but use the current result.
610 		 * Probably will end up going into overflow, but this means
611 		 * we cannot trust the ordering anymore
612 		 */
613 	}
614 
615 	io_req_set_res(req, *ret, cflags);
616 
617 	if (req->flags & REQ_F_POLLED)
618 		*ret = IOU_STOP_MULTISHOT;
619 	else
620 		*ret = IOU_OK;
621 	return true;
622 }
623 
624 static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
625 				     struct io_sr_msg *sr, void __user **buf,
626 				     size_t *len)
627 {
628 	unsigned long ubuf = (unsigned long) *buf;
629 	unsigned long hdr;
630 
631 	hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
632 		kmsg->controllen;
633 	if (*len < hdr)
634 		return -EFAULT;
635 
636 	if (kmsg->controllen) {
637 		unsigned long control = ubuf + hdr - kmsg->controllen;
638 
639 		kmsg->msg.msg_control_user = (void __user *) control;
640 		kmsg->msg.msg_controllen = kmsg->controllen;
641 	}
642 
643 	sr->buf = *buf; /* stash for later copy */
644 	*buf = (void __user *) (ubuf + hdr);
645 	kmsg->payloadlen = *len = *len - hdr;
646 	return 0;
647 }
648 
649 struct io_recvmsg_multishot_hdr {
650 	struct io_uring_recvmsg_out msg;
651 	struct sockaddr_storage addr;
652 };
653 
654 static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
655 				struct io_async_msghdr *kmsg,
656 				unsigned int flags, bool *finished)
657 {
658 	int err;
659 	int copy_len;
660 	struct io_recvmsg_multishot_hdr hdr;
661 
662 	if (kmsg->namelen)
663 		kmsg->msg.msg_name = &hdr.addr;
664 	kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
665 	kmsg->msg.msg_namelen = 0;
666 
667 	if (sock->file->f_flags & O_NONBLOCK)
668 		flags |= MSG_DONTWAIT;
669 
670 	err = sock_recvmsg(sock, &kmsg->msg, flags);
671 	*finished = err <= 0;
672 	if (err < 0)
673 		return err;
674 
675 	hdr.msg = (struct io_uring_recvmsg_out) {
676 		.controllen = kmsg->controllen - kmsg->msg.msg_controllen,
677 		.flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
678 	};
679 
680 	hdr.msg.payloadlen = err;
681 	if (err > kmsg->payloadlen)
682 		err = kmsg->payloadlen;
683 
684 	copy_len = sizeof(struct io_uring_recvmsg_out);
685 	if (kmsg->msg.msg_namelen > kmsg->namelen)
686 		copy_len += kmsg->namelen;
687 	else
688 		copy_len += kmsg->msg.msg_namelen;
689 
690 	/*
691 	 *      "fromlen shall refer to the value before truncation.."
692 	 *                      1003.1g
693 	 */
694 	hdr.msg.namelen = kmsg->msg.msg_namelen;
695 
696 	/* ensure that there is no gap between hdr and sockaddr_storage */
697 	BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
698 		     sizeof(struct io_uring_recvmsg_out));
699 	if (copy_to_user(io->buf, &hdr, copy_len)) {
700 		*finished = true;
701 		return -EFAULT;
702 	}
703 
704 	return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
705 			kmsg->controllen + err;
706 }
707 
708 int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
709 {
710 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
711 	struct io_async_msghdr iomsg, *kmsg;
712 	struct socket *sock;
713 	unsigned int cflags;
714 	unsigned flags;
715 	int ret, min_ret = 0;
716 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
717 	bool mshot_finished = true;
718 
719 	sock = sock_from_file(req->file);
720 	if (unlikely(!sock))
721 		return -ENOTSOCK;
722 
723 	if (req_has_async_data(req)) {
724 		kmsg = req->async_data;
725 	} else {
726 		ret = io_recvmsg_copy_hdr(req, &iomsg);
727 		if (ret)
728 			return ret;
729 		kmsg = &iomsg;
730 	}
731 
732 	if (!(req->flags & REQ_F_POLLED) &&
733 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
734 		return io_setup_async_msg(req, kmsg, issue_flags);
735 
736 retry_multishot:
737 	if (io_do_buffer_select(req)) {
738 		void __user *buf;
739 		size_t len = sr->len;
740 
741 		buf = io_buffer_select(req, &len, issue_flags);
742 		if (!buf)
743 			return -ENOBUFS;
744 
745 		if (req->flags & REQ_F_APOLL_MULTISHOT) {
746 			ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
747 			if (ret) {
748 				io_kbuf_recycle(req, issue_flags);
749 				return ret;
750 			}
751 		}
752 
753 		kmsg->fast_iov[0].iov_base = buf;
754 		kmsg->fast_iov[0].iov_len = len;
755 		iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov, 1,
756 				len);
757 	}
758 
759 	flags = sr->msg_flags;
760 	if (force_nonblock)
761 		flags |= MSG_DONTWAIT;
762 	if (flags & MSG_WAITALL)
763 		min_ret = iov_iter_count(&kmsg->msg.msg_iter);
764 
765 	kmsg->msg.msg_get_inq = 1;
766 	if (req->flags & REQ_F_APOLL_MULTISHOT)
767 		ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
768 					   &mshot_finished);
769 	else
770 		ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
771 					 kmsg->uaddr, flags);
772 
773 	if (ret < min_ret) {
774 		if (ret == -EAGAIN && force_nonblock) {
775 			ret = io_setup_async_msg(req, kmsg, issue_flags);
776 			if (ret == -EAGAIN && (req->flags & IO_APOLL_MULTI_POLLED) ==
777 					       IO_APOLL_MULTI_POLLED) {
778 				io_kbuf_recycle(req, issue_flags);
779 				return IOU_ISSUE_SKIP_COMPLETE;
780 			}
781 			return ret;
782 		}
783 		if (ret > 0 && io_net_retry(sock, flags)) {
784 			sr->done_io += ret;
785 			req->flags |= REQ_F_PARTIAL_IO;
786 			return io_setup_async_msg(req, kmsg, issue_flags);
787 		}
788 		if (ret == -ERESTARTSYS)
789 			ret = -EINTR;
790 		req_set_fail(req);
791 	} else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
792 		req_set_fail(req);
793 	}
794 
795 	if (ret > 0)
796 		ret += sr->done_io;
797 	else if (sr->done_io)
798 		ret = sr->done_io;
799 	else
800 		io_kbuf_recycle(req, issue_flags);
801 
802 	cflags = io_put_kbuf(req, issue_flags);
803 	if (kmsg->msg.msg_inq)
804 		cflags |= IORING_CQE_F_SOCK_NONEMPTY;
805 
806 	if (!io_recv_finish(req, &ret, cflags, mshot_finished))
807 		goto retry_multishot;
808 
809 	if (mshot_finished) {
810 		io_netmsg_recycle(req, issue_flags);
811 		/* fast path, check for non-NULL to avoid function call */
812 		if (kmsg->free_iov)
813 			kfree(kmsg->free_iov);
814 		req->flags &= ~REQ_F_NEED_CLEANUP;
815 	}
816 
817 	return ret;
818 }
819 
820 int io_recv(struct io_kiocb *req, unsigned int issue_flags)
821 {
822 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
823 	struct msghdr msg;
824 	struct socket *sock;
825 	struct iovec iov;
826 	unsigned int cflags;
827 	unsigned flags;
828 	int ret, min_ret = 0;
829 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
830 	size_t len = sr->len;
831 
832 	if (!(req->flags & REQ_F_POLLED) &&
833 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
834 		return -EAGAIN;
835 
836 	sock = sock_from_file(req->file);
837 	if (unlikely(!sock))
838 		return -ENOTSOCK;
839 
840 retry_multishot:
841 	if (io_do_buffer_select(req)) {
842 		void __user *buf;
843 
844 		buf = io_buffer_select(req, &len, issue_flags);
845 		if (!buf)
846 			return -ENOBUFS;
847 		sr->buf = buf;
848 	}
849 
850 	ret = import_single_range(READ, sr->buf, len, &iov, &msg.msg_iter);
851 	if (unlikely(ret))
852 		goto out_free;
853 
854 	msg.msg_name = NULL;
855 	msg.msg_namelen = 0;
856 	msg.msg_control = NULL;
857 	msg.msg_get_inq = 1;
858 	msg.msg_flags = 0;
859 	msg.msg_controllen = 0;
860 	msg.msg_iocb = NULL;
861 	msg.msg_ubuf = NULL;
862 
863 	flags = sr->msg_flags;
864 	if (force_nonblock)
865 		flags |= MSG_DONTWAIT;
866 	if (flags & MSG_WAITALL)
867 		min_ret = iov_iter_count(&msg.msg_iter);
868 
869 	ret = sock_recvmsg(sock, &msg, flags);
870 	if (ret < min_ret) {
871 		if (ret == -EAGAIN && force_nonblock) {
872 			if ((req->flags & IO_APOLL_MULTI_POLLED) == IO_APOLL_MULTI_POLLED) {
873 				io_kbuf_recycle(req, issue_flags);
874 				return IOU_ISSUE_SKIP_COMPLETE;
875 			}
876 
877 			return -EAGAIN;
878 		}
879 		if (ret > 0 && io_net_retry(sock, flags)) {
880 			sr->len -= ret;
881 			sr->buf += ret;
882 			sr->done_io += ret;
883 			req->flags |= REQ_F_PARTIAL_IO;
884 			return -EAGAIN;
885 		}
886 		if (ret == -ERESTARTSYS)
887 			ret = -EINTR;
888 		req_set_fail(req);
889 	} else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
890 out_free:
891 		req_set_fail(req);
892 	}
893 
894 	if (ret > 0)
895 		ret += sr->done_io;
896 	else if (sr->done_io)
897 		ret = sr->done_io;
898 	else
899 		io_kbuf_recycle(req, issue_flags);
900 
901 	cflags = io_put_kbuf(req, issue_flags);
902 	if (msg.msg_inq)
903 		cflags |= IORING_CQE_F_SOCK_NONEMPTY;
904 
905 	if (!io_recv_finish(req, &ret, cflags, ret <= 0))
906 		goto retry_multishot;
907 
908 	return ret;
909 }
910 
911 void io_send_zc_cleanup(struct io_kiocb *req)
912 {
913 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
914 	struct io_async_msghdr *io;
915 
916 	if (req_has_async_data(req)) {
917 		io = req->async_data;
918 		/* might be ->fast_iov if *msg_copy_hdr failed */
919 		if (io->free_iov != io->fast_iov)
920 			kfree(io->free_iov);
921 	}
922 	if (zc->notif) {
923 		io_notif_flush(zc->notif);
924 		zc->notif = NULL;
925 	}
926 }
927 
928 int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
929 {
930 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
931 	struct io_ring_ctx *ctx = req->ctx;
932 	struct io_kiocb *notif;
933 
934 	if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
935 		return -EINVAL;
936 	/* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
937 	if (req->flags & REQ_F_CQE_SKIP)
938 		return -EINVAL;
939 
940 	zc->flags = READ_ONCE(sqe->ioprio);
941 	if (zc->flags & ~(IORING_RECVSEND_POLL_FIRST |
942 			  IORING_RECVSEND_FIXED_BUF))
943 		return -EINVAL;
944 	notif = zc->notif = io_alloc_notif(ctx);
945 	if (!notif)
946 		return -ENOMEM;
947 	notif->cqe.user_data = req->cqe.user_data;
948 	notif->cqe.res = 0;
949 	notif->cqe.flags = IORING_CQE_F_NOTIF;
950 	req->flags |= REQ_F_NEED_CLEANUP;
951 	if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
952 		unsigned idx = READ_ONCE(sqe->buf_index);
953 
954 		if (unlikely(idx >= ctx->nr_user_bufs))
955 			return -EFAULT;
956 		idx = array_index_nospec(idx, ctx->nr_user_bufs);
957 		req->imu = READ_ONCE(ctx->user_bufs[idx]);
958 		io_req_set_rsrc_node(notif, ctx, 0);
959 	}
960 
961 	if (req->opcode == IORING_OP_SEND_ZC) {
962 		if (READ_ONCE(sqe->__pad3[0]))
963 			return -EINVAL;
964 		zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
965 		zc->addr_len = READ_ONCE(sqe->addr_len);
966 	} else {
967 		if (unlikely(sqe->addr2 || sqe->file_index))
968 			return -EINVAL;
969 		if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
970 			return -EINVAL;
971 	}
972 
973 	zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
974 	zc->len = READ_ONCE(sqe->len);
975 	zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
976 	if (zc->msg_flags & MSG_DONTWAIT)
977 		req->flags |= REQ_F_NOWAIT;
978 
979 	zc->done_io = 0;
980 
981 #ifdef CONFIG_COMPAT
982 	if (req->ctx->compat)
983 		zc->msg_flags |= MSG_CMSG_COMPAT;
984 #endif
985 	return 0;
986 }
987 
988 static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
989 				 struct iov_iter *from, size_t length)
990 {
991 	skb_zcopy_downgrade_managed(skb);
992 	return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
993 }
994 
995 static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
996 			   struct iov_iter *from, size_t length)
997 {
998 	struct skb_shared_info *shinfo = skb_shinfo(skb);
999 	int frag = shinfo->nr_frags;
1000 	int ret = 0;
1001 	struct bvec_iter bi;
1002 	ssize_t copied = 0;
1003 	unsigned long truesize = 0;
1004 
1005 	if (!frag)
1006 		shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
1007 	else if (unlikely(!skb_zcopy_managed(skb)))
1008 		return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1009 
1010 	bi.bi_size = min(from->count, length);
1011 	bi.bi_bvec_done = from->iov_offset;
1012 	bi.bi_idx = 0;
1013 
1014 	while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1015 		struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1016 
1017 		copied += v.bv_len;
1018 		truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1019 		__skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1020 					   v.bv_offset, v.bv_len);
1021 		bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1022 	}
1023 	if (bi.bi_size)
1024 		ret = -EMSGSIZE;
1025 
1026 	shinfo->nr_frags = frag;
1027 	from->bvec += bi.bi_idx;
1028 	from->nr_segs -= bi.bi_idx;
1029 	from->count -= copied;
1030 	from->iov_offset = bi.bi_bvec_done;
1031 
1032 	skb->data_len += copied;
1033 	skb->len += copied;
1034 	skb->truesize += truesize;
1035 
1036 	if (sk && sk->sk_type == SOCK_STREAM) {
1037 		sk_wmem_queued_add(sk, truesize);
1038 		if (!skb_zcopy_pure(skb))
1039 			sk_mem_charge(sk, truesize);
1040 	} else {
1041 		refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1042 	}
1043 	return ret;
1044 }
1045 
1046 int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
1047 {
1048 	struct sockaddr_storage __address;
1049 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1050 	struct msghdr msg;
1051 	struct iovec iov;
1052 	struct socket *sock;
1053 	unsigned msg_flags;
1054 	int ret, min_ret = 0;
1055 
1056 	sock = sock_from_file(req->file);
1057 	if (unlikely(!sock))
1058 		return -ENOTSOCK;
1059 
1060 	msg.msg_name = NULL;
1061 	msg.msg_control = NULL;
1062 	msg.msg_controllen = 0;
1063 	msg.msg_namelen = 0;
1064 
1065 	if (zc->addr) {
1066 		if (req_has_async_data(req)) {
1067 			struct io_async_msghdr *io = req->async_data;
1068 
1069 			msg.msg_name = &io->addr;
1070 		} else {
1071 			ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1072 			if (unlikely(ret < 0))
1073 				return ret;
1074 			msg.msg_name = (struct sockaddr *)&__address;
1075 		}
1076 		msg.msg_namelen = zc->addr_len;
1077 	}
1078 
1079 	if (!(req->flags & REQ_F_POLLED) &&
1080 	    (zc->flags & IORING_RECVSEND_POLL_FIRST))
1081 		return io_setup_async_addr(req, &__address, issue_flags);
1082 
1083 	if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1084 		ret = io_import_fixed(WRITE, &msg.msg_iter, req->imu,
1085 					(u64)(uintptr_t)zc->buf, zc->len);
1086 		if (unlikely(ret))
1087 			return ret;
1088 		msg.sg_from_iter = io_sg_from_iter;
1089 	} else {
1090 		ret = import_single_range(WRITE, zc->buf, zc->len, &iov,
1091 					  &msg.msg_iter);
1092 		if (unlikely(ret))
1093 			return ret;
1094 		ret = io_notif_account_mem(zc->notif, zc->len);
1095 		if (unlikely(ret))
1096 			return ret;
1097 		msg.sg_from_iter = io_sg_from_iter_iovec;
1098 	}
1099 
1100 	msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1101 	if (issue_flags & IO_URING_F_NONBLOCK)
1102 		msg_flags |= MSG_DONTWAIT;
1103 	if (msg_flags & MSG_WAITALL)
1104 		min_ret = iov_iter_count(&msg.msg_iter);
1105 
1106 	msg.msg_flags = msg_flags;
1107 	msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
1108 	ret = sock_sendmsg(sock, &msg);
1109 
1110 	if (unlikely(ret < min_ret)) {
1111 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1112 			return io_setup_async_addr(req, &__address, issue_flags);
1113 
1114 		if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1115 			zc->len -= ret;
1116 			zc->buf += ret;
1117 			zc->done_io += ret;
1118 			req->flags |= REQ_F_PARTIAL_IO;
1119 			return io_setup_async_addr(req, &__address, issue_flags);
1120 		}
1121 		if (ret == -ERESTARTSYS)
1122 			ret = -EINTR;
1123 		req_set_fail(req);
1124 	}
1125 
1126 	if (ret >= 0)
1127 		ret += zc->done_io;
1128 	else if (zc->done_io)
1129 		ret = zc->done_io;
1130 
1131 	/*
1132 	 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1133 	 * flushing notif to io_send_zc_cleanup()
1134 	 */
1135 	if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1136 		io_notif_flush(zc->notif);
1137 		req->flags &= ~REQ_F_NEED_CLEANUP;
1138 	}
1139 	io_req_set_res(req, ret, IORING_CQE_F_MORE);
1140 	return IOU_OK;
1141 }
1142 
1143 int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1144 {
1145 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1146 	struct io_async_msghdr iomsg, *kmsg;
1147 	struct socket *sock;
1148 	unsigned flags;
1149 	int ret, min_ret = 0;
1150 
1151 	sock = sock_from_file(req->file);
1152 	if (unlikely(!sock))
1153 		return -ENOTSOCK;
1154 
1155 	if (req_has_async_data(req)) {
1156 		kmsg = req->async_data;
1157 	} else {
1158 		ret = io_sendmsg_copy_hdr(req, &iomsg);
1159 		if (ret)
1160 			return ret;
1161 		kmsg = &iomsg;
1162 	}
1163 
1164 	if (!(req->flags & REQ_F_POLLED) &&
1165 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
1166 		return io_setup_async_msg(req, kmsg, issue_flags);
1167 
1168 	flags = sr->msg_flags | MSG_ZEROCOPY;
1169 	if (issue_flags & IO_URING_F_NONBLOCK)
1170 		flags |= MSG_DONTWAIT;
1171 	if (flags & MSG_WAITALL)
1172 		min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1173 
1174 	kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1175 	kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1176 	ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1177 
1178 	if (unlikely(ret < min_ret)) {
1179 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1180 			return io_setup_async_msg(req, kmsg, issue_flags);
1181 
1182 		if (ret > 0 && io_net_retry(sock, flags)) {
1183 			sr->done_io += ret;
1184 			req->flags |= REQ_F_PARTIAL_IO;
1185 			return io_setup_async_msg(req, kmsg, issue_flags);
1186 		}
1187 		if (ret == -ERESTARTSYS)
1188 			ret = -EINTR;
1189 		req_set_fail(req);
1190 	}
1191 	/* fast path, check for non-NULL to avoid function call */
1192 	if (kmsg->free_iov) {
1193 		kfree(kmsg->free_iov);
1194 		kmsg->free_iov = NULL;
1195 	}
1196 
1197 	io_netmsg_recycle(req, issue_flags);
1198 	if (ret >= 0)
1199 		ret += sr->done_io;
1200 	else if (sr->done_io)
1201 		ret = sr->done_io;
1202 
1203 	/*
1204 	 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1205 	 * flushing notif to io_send_zc_cleanup()
1206 	 */
1207 	if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1208 		io_notif_flush(sr->notif);
1209 		req->flags &= ~REQ_F_NEED_CLEANUP;
1210 	}
1211 	io_req_set_res(req, ret, IORING_CQE_F_MORE);
1212 	return IOU_OK;
1213 }
1214 
1215 void io_sendrecv_fail(struct io_kiocb *req)
1216 {
1217 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1218 
1219 	if (req->flags & REQ_F_PARTIAL_IO)
1220 		req->cqe.res = sr->done_io;
1221 
1222 	if ((req->flags & REQ_F_NEED_CLEANUP) &&
1223 	    (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1224 		req->cqe.flags |= IORING_CQE_F_MORE;
1225 }
1226 
1227 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1228 {
1229 	struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1230 	unsigned flags;
1231 
1232 	if (sqe->len || sqe->buf_index)
1233 		return -EINVAL;
1234 
1235 	accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1236 	accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1237 	accept->flags = READ_ONCE(sqe->accept_flags);
1238 	accept->nofile = rlimit(RLIMIT_NOFILE);
1239 	flags = READ_ONCE(sqe->ioprio);
1240 	if (flags & ~IORING_ACCEPT_MULTISHOT)
1241 		return -EINVAL;
1242 
1243 	accept->file_slot = READ_ONCE(sqe->file_index);
1244 	if (accept->file_slot) {
1245 		if (accept->flags & SOCK_CLOEXEC)
1246 			return -EINVAL;
1247 		if (flags & IORING_ACCEPT_MULTISHOT &&
1248 		    accept->file_slot != IORING_FILE_INDEX_ALLOC)
1249 			return -EINVAL;
1250 	}
1251 	if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1252 		return -EINVAL;
1253 	if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1254 		accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1255 	if (flags & IORING_ACCEPT_MULTISHOT)
1256 		req->flags |= REQ_F_APOLL_MULTISHOT;
1257 	return 0;
1258 }
1259 
1260 int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1261 {
1262 	struct io_ring_ctx *ctx = req->ctx;
1263 	struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1264 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1265 	unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1266 	bool fixed = !!accept->file_slot;
1267 	struct file *file;
1268 	int ret, fd;
1269 
1270 retry:
1271 	if (!fixed) {
1272 		fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1273 		if (unlikely(fd < 0))
1274 			return fd;
1275 	}
1276 	file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1277 			 accept->flags);
1278 	if (IS_ERR(file)) {
1279 		if (!fixed)
1280 			put_unused_fd(fd);
1281 		ret = PTR_ERR(file);
1282 		if (ret == -EAGAIN && force_nonblock) {
1283 			/*
1284 			 * if it's multishot and polled, we don't need to
1285 			 * return EAGAIN to arm the poll infra since it
1286 			 * has already been done
1287 			 */
1288 			if ((req->flags & IO_APOLL_MULTI_POLLED) ==
1289 			    IO_APOLL_MULTI_POLLED)
1290 				ret = IOU_ISSUE_SKIP_COMPLETE;
1291 			return ret;
1292 		}
1293 		if (ret == -ERESTARTSYS)
1294 			ret = -EINTR;
1295 		req_set_fail(req);
1296 	} else if (!fixed) {
1297 		fd_install(fd, file);
1298 		ret = fd;
1299 	} else {
1300 		ret = io_fixed_fd_install(req, issue_flags, file,
1301 						accept->file_slot);
1302 	}
1303 
1304 	if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1305 		io_req_set_res(req, ret, 0);
1306 		return IOU_OK;
1307 	}
1308 
1309 	if (ret >= 0 &&
1310 	    io_post_aux_cqe(ctx, req->cqe.user_data, ret, IORING_CQE_F_MORE, false))
1311 		goto retry;
1312 
1313 	io_req_set_res(req, ret, 0);
1314 	if (req->flags & REQ_F_POLLED)
1315 		return IOU_STOP_MULTISHOT;
1316 	return IOU_OK;
1317 }
1318 
1319 int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1320 {
1321 	struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1322 
1323 	if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1324 		return -EINVAL;
1325 
1326 	sock->domain = READ_ONCE(sqe->fd);
1327 	sock->type = READ_ONCE(sqe->off);
1328 	sock->protocol = READ_ONCE(sqe->len);
1329 	sock->file_slot = READ_ONCE(sqe->file_index);
1330 	sock->nofile = rlimit(RLIMIT_NOFILE);
1331 
1332 	sock->flags = sock->type & ~SOCK_TYPE_MASK;
1333 	if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1334 		return -EINVAL;
1335 	if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1336 		return -EINVAL;
1337 	return 0;
1338 }
1339 
1340 int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1341 {
1342 	struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1343 	bool fixed = !!sock->file_slot;
1344 	struct file *file;
1345 	int ret, fd;
1346 
1347 	if (!fixed) {
1348 		fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1349 		if (unlikely(fd < 0))
1350 			return fd;
1351 	}
1352 	file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1353 	if (IS_ERR(file)) {
1354 		if (!fixed)
1355 			put_unused_fd(fd);
1356 		ret = PTR_ERR(file);
1357 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1358 			return -EAGAIN;
1359 		if (ret == -ERESTARTSYS)
1360 			ret = -EINTR;
1361 		req_set_fail(req);
1362 	} else if (!fixed) {
1363 		fd_install(fd, file);
1364 		ret = fd;
1365 	} else {
1366 		ret = io_fixed_fd_install(req, issue_flags, file,
1367 					    sock->file_slot);
1368 	}
1369 	io_req_set_res(req, ret, 0);
1370 	return IOU_OK;
1371 }
1372 
1373 int io_connect_prep_async(struct io_kiocb *req)
1374 {
1375 	struct io_async_connect *io = req->async_data;
1376 	struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1377 
1378 	return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1379 }
1380 
1381 int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1382 {
1383 	struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1384 
1385 	if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1386 		return -EINVAL;
1387 
1388 	conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1389 	conn->addr_len =  READ_ONCE(sqe->addr2);
1390 	conn->in_progress = false;
1391 	return 0;
1392 }
1393 
1394 int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1395 {
1396 	struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1397 	struct io_async_connect __io, *io;
1398 	unsigned file_flags;
1399 	int ret;
1400 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1401 
1402 	if (connect->in_progress) {
1403 		struct socket *socket;
1404 
1405 		ret = -ENOTSOCK;
1406 		socket = sock_from_file(req->file);
1407 		if (socket)
1408 			ret = sock_error(socket->sk);
1409 		goto out;
1410 	}
1411 
1412 	if (req_has_async_data(req)) {
1413 		io = req->async_data;
1414 	} else {
1415 		ret = move_addr_to_kernel(connect->addr,
1416 						connect->addr_len,
1417 						&__io.address);
1418 		if (ret)
1419 			goto out;
1420 		io = &__io;
1421 	}
1422 
1423 	file_flags = force_nonblock ? O_NONBLOCK : 0;
1424 
1425 	ret = __sys_connect_file(req->file, &io->address,
1426 					connect->addr_len, file_flags);
1427 	if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
1428 		if (ret == -EINPROGRESS) {
1429 			connect->in_progress = true;
1430 		} else {
1431 			if (req_has_async_data(req))
1432 				return -EAGAIN;
1433 			if (io_alloc_async_data(req)) {
1434 				ret = -ENOMEM;
1435 				goto out;
1436 			}
1437 			memcpy(req->async_data, &__io, sizeof(__io));
1438 		}
1439 		return -EAGAIN;
1440 	}
1441 	if (ret == -ERESTARTSYS)
1442 		ret = -EINTR;
1443 out:
1444 	if (ret < 0)
1445 		req_set_fail(req);
1446 	io_req_set_res(req, ret, 0);
1447 	return IOU_OK;
1448 }
1449 
1450 void io_netmsg_cache_free(struct io_cache_entry *entry)
1451 {
1452 	kfree(container_of(entry, struct io_async_msghdr, cache));
1453 }
1454 #endif
1455