xref: /linux/io_uring/net.c (revision 40d269c000bda9fcd276a0412a9cebd3f6e344c5)
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 	bool				seen_econnaborted;
51 };
52 
53 struct io_sr_msg {
54 	struct file			*file;
55 	union {
56 		struct compat_msghdr __user	*umsg_compat;
57 		struct user_msghdr __user	*umsg;
58 		void __user			*buf;
59 	};
60 	unsigned			len;
61 	unsigned			done_io;
62 	unsigned			msg_flags;
63 	unsigned			nr_multishot_loops;
64 	u16				flags;
65 	/* initialised and used only by !msg send variants */
66 	u16				addr_len;
67 	u16				buf_group;
68 	void __user			*addr;
69 	void __user			*msg_control;
70 	/* used only for send zerocopy */
71 	struct io_kiocb 		*notif;
72 };
73 
74 /*
75  * Number of times we'll try and do receives if there's more data. If we
76  * exceed this limit, then add us to the back of the queue and retry from
77  * there. This helps fairness between flooding clients.
78  */
79 #define MULTISHOT_MAX_RETRY	32
80 
81 int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
82 {
83 	struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
84 
85 	if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
86 		     sqe->buf_index || sqe->splice_fd_in))
87 		return -EINVAL;
88 
89 	shutdown->how = READ_ONCE(sqe->len);
90 	req->flags |= REQ_F_FORCE_ASYNC;
91 	return 0;
92 }
93 
94 int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
95 {
96 	struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
97 	struct socket *sock;
98 	int ret;
99 
100 	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
101 
102 	sock = sock_from_file(req->file);
103 	if (unlikely(!sock))
104 		return -ENOTSOCK;
105 
106 	ret = __sys_shutdown_sock(sock, shutdown->how);
107 	io_req_set_res(req, ret, 0);
108 	return IOU_OK;
109 }
110 
111 static bool io_net_retry(struct socket *sock, int flags)
112 {
113 	if (!(flags & MSG_WAITALL))
114 		return false;
115 	return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
116 }
117 
118 static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
119 {
120 	struct io_async_msghdr *hdr = req->async_data;
121 
122 	if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
123 		return;
124 
125 	/* Let normal cleanup path reap it if we fail adding to the cache */
126 	if (io_alloc_cache_put(&req->ctx->netmsg_cache, &hdr->cache)) {
127 		req->async_data = NULL;
128 		req->flags &= ~REQ_F_ASYNC_DATA;
129 	}
130 }
131 
132 static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
133 						  unsigned int issue_flags)
134 {
135 	struct io_ring_ctx *ctx = req->ctx;
136 	struct io_cache_entry *entry;
137 	struct io_async_msghdr *hdr;
138 
139 	if (!(issue_flags & IO_URING_F_UNLOCKED)) {
140 		entry = io_alloc_cache_get(&ctx->netmsg_cache);
141 		if (entry) {
142 			hdr = container_of(entry, struct io_async_msghdr, cache);
143 			hdr->free_iov = NULL;
144 			req->flags |= REQ_F_ASYNC_DATA;
145 			req->async_data = hdr;
146 			return hdr;
147 		}
148 	}
149 
150 	if (!io_alloc_async_data(req)) {
151 		hdr = req->async_data;
152 		hdr->free_iov = NULL;
153 		return hdr;
154 	}
155 	return NULL;
156 }
157 
158 static inline struct io_async_msghdr *io_msg_alloc_async_prep(struct io_kiocb *req)
159 {
160 	/* ->prep_async is always called from the submission context */
161 	return io_msg_alloc_async(req, 0);
162 }
163 
164 static int io_setup_async_msg(struct io_kiocb *req,
165 			      struct io_async_msghdr *kmsg,
166 			      unsigned int issue_flags)
167 {
168 	struct io_async_msghdr *async_msg;
169 
170 	if (req_has_async_data(req))
171 		return -EAGAIN;
172 	async_msg = io_msg_alloc_async(req, issue_flags);
173 	if (!async_msg) {
174 		kfree(kmsg->free_iov);
175 		return -ENOMEM;
176 	}
177 	req->flags |= REQ_F_NEED_CLEANUP;
178 	memcpy(async_msg, kmsg, sizeof(*kmsg));
179 	if (async_msg->msg.msg_name)
180 		async_msg->msg.msg_name = &async_msg->addr;
181 
182 	if ((req->flags & REQ_F_BUFFER_SELECT) && !async_msg->msg.msg_iter.nr_segs)
183 		return -EAGAIN;
184 
185 	/* if were using fast_iov, set it to the new one */
186 	if (iter_is_iovec(&kmsg->msg.msg_iter) && !kmsg->free_iov) {
187 		size_t fast_idx = iter_iov(&kmsg->msg.msg_iter) - kmsg->fast_iov;
188 		async_msg->msg.msg_iter.__iov = &async_msg->fast_iov[fast_idx];
189 	}
190 
191 	return -EAGAIN;
192 }
193 
194 #ifdef CONFIG_COMPAT
195 static int io_compat_msg_copy_hdr(struct io_kiocb *req,
196 				  struct io_async_msghdr *iomsg,
197 				  struct compat_msghdr *msg, int ddir)
198 {
199 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
200 	struct compat_iovec __user *uiov;
201 	int ret;
202 
203 	if (copy_from_user(msg, sr->umsg_compat, sizeof(*msg)))
204 		return -EFAULT;
205 
206 	uiov = compat_ptr(msg->msg_iov);
207 	if (req->flags & REQ_F_BUFFER_SELECT) {
208 		compat_ssize_t clen;
209 
210 		iomsg->free_iov = NULL;
211 		if (msg->msg_iovlen == 0) {
212 			sr->len = 0;
213 		} else if (msg->msg_iovlen > 1) {
214 			return -EINVAL;
215 		} else {
216 			if (!access_ok(uiov, sizeof(*uiov)))
217 				return -EFAULT;
218 			if (__get_user(clen, &uiov->iov_len))
219 				return -EFAULT;
220 			if (clen < 0)
221 				return -EINVAL;
222 			sr->len = clen;
223 		}
224 
225 		return 0;
226 	}
227 
228 	iomsg->free_iov = iomsg->fast_iov;
229 	ret = __import_iovec(ddir, (struct iovec __user *)uiov, msg->msg_iovlen,
230 				UIO_FASTIOV, &iomsg->free_iov,
231 				&iomsg->msg.msg_iter, true);
232 	if (unlikely(ret < 0))
233 		return ret;
234 
235 	return 0;
236 }
237 #endif
238 
239 static int io_msg_copy_hdr(struct io_kiocb *req, struct io_async_msghdr *iomsg,
240 			   struct user_msghdr *msg, int ddir)
241 {
242 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
243 	int ret;
244 
245 	if (!user_access_begin(sr->umsg, sizeof(*sr->umsg)))
246 		return -EFAULT;
247 
248 	ret = -EFAULT;
249 	unsafe_get_user(msg->msg_name, &sr->umsg->msg_name, ua_end);
250 	unsafe_get_user(msg->msg_namelen, &sr->umsg->msg_namelen, ua_end);
251 	unsafe_get_user(msg->msg_iov, &sr->umsg->msg_iov, ua_end);
252 	unsafe_get_user(msg->msg_iovlen, &sr->umsg->msg_iovlen, ua_end);
253 	unsafe_get_user(msg->msg_control, &sr->umsg->msg_control, ua_end);
254 	unsafe_get_user(msg->msg_controllen, &sr->umsg->msg_controllen, ua_end);
255 	msg->msg_flags = 0;
256 
257 	if (req->flags & REQ_F_BUFFER_SELECT) {
258 		if (msg->msg_iovlen == 0) {
259 			sr->len = iomsg->fast_iov[0].iov_len = 0;
260 			iomsg->fast_iov[0].iov_base = NULL;
261 			iomsg->free_iov = NULL;
262 		} else if (msg->msg_iovlen > 1) {
263 			ret = -EINVAL;
264 			goto ua_end;
265 		} else {
266 			/* we only need the length for provided buffers */
267 			if (!access_ok(&msg->msg_iov[0].iov_len, sizeof(__kernel_size_t)))
268 				goto ua_end;
269 			unsafe_get_user(iomsg->fast_iov[0].iov_len,
270 					&msg->msg_iov[0].iov_len, ua_end);
271 			sr->len = iomsg->fast_iov[0].iov_len;
272 			iomsg->free_iov = NULL;
273 		}
274 		ret = 0;
275 ua_end:
276 		user_access_end();
277 		return ret;
278 	}
279 
280 	user_access_end();
281 	iomsg->free_iov = iomsg->fast_iov;
282 	ret = __import_iovec(ddir, msg->msg_iov, msg->msg_iovlen, UIO_FASTIOV,
283 				&iomsg->free_iov, &iomsg->msg.msg_iter, false);
284 	if (unlikely(ret < 0))
285 		return ret;
286 
287 	return 0;
288 }
289 
290 static int io_sendmsg_copy_hdr(struct io_kiocb *req,
291 			       struct io_async_msghdr *iomsg)
292 {
293 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
294 	struct user_msghdr msg;
295 	int ret;
296 
297 	iomsg->msg.msg_name = &iomsg->addr;
298 	iomsg->msg.msg_iter.nr_segs = 0;
299 
300 #ifdef CONFIG_COMPAT
301 	if (unlikely(req->ctx->compat)) {
302 		struct compat_msghdr cmsg;
303 
304 		ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ITER_SOURCE);
305 		if (unlikely(ret))
306 			return ret;
307 
308 		return __get_compat_msghdr(&iomsg->msg, &cmsg, NULL);
309 	}
310 #endif
311 
312 	ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_SOURCE);
313 	if (unlikely(ret))
314 		return ret;
315 
316 	ret = __copy_msghdr(&iomsg->msg, &msg, NULL);
317 
318 	/* save msg_control as sys_sendmsg() overwrites it */
319 	sr->msg_control = iomsg->msg.msg_control_user;
320 	return ret;
321 }
322 
323 int io_send_prep_async(struct io_kiocb *req)
324 {
325 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
326 	struct io_async_msghdr *io;
327 	int ret;
328 
329 	if (req_has_async_data(req))
330 		return 0;
331 	zc->done_io = 0;
332 	if (!zc->addr)
333 		return 0;
334 	io = io_msg_alloc_async_prep(req);
335 	if (!io)
336 		return -ENOMEM;
337 	ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
338 	return ret;
339 }
340 
341 static int io_setup_async_addr(struct io_kiocb *req,
342 			      struct sockaddr_storage *addr_storage,
343 			      unsigned int issue_flags)
344 {
345 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
346 	struct io_async_msghdr *io;
347 
348 	if (!sr->addr || req_has_async_data(req))
349 		return -EAGAIN;
350 	io = io_msg_alloc_async(req, issue_flags);
351 	if (!io)
352 		return -ENOMEM;
353 	memcpy(&io->addr, addr_storage, sizeof(io->addr));
354 	return -EAGAIN;
355 }
356 
357 int io_sendmsg_prep_async(struct io_kiocb *req)
358 {
359 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
360 	int ret;
361 
362 	sr->done_io = 0;
363 	if (!io_msg_alloc_async_prep(req))
364 		return -ENOMEM;
365 	ret = io_sendmsg_copy_hdr(req, req->async_data);
366 	if (!ret)
367 		req->flags |= REQ_F_NEED_CLEANUP;
368 	return ret;
369 }
370 
371 void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
372 {
373 	struct io_async_msghdr *io = req->async_data;
374 
375 	kfree(io->free_iov);
376 }
377 
378 int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
379 {
380 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
381 
382 	sr->done_io = 0;
383 
384 	if (req->opcode == IORING_OP_SEND) {
385 		if (READ_ONCE(sqe->__pad3[0]))
386 			return -EINVAL;
387 		sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
388 		sr->addr_len = READ_ONCE(sqe->addr_len);
389 	} else if (sqe->addr2 || sqe->file_index) {
390 		return -EINVAL;
391 	}
392 
393 	sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
394 	sr->len = READ_ONCE(sqe->len);
395 	sr->flags = READ_ONCE(sqe->ioprio);
396 	if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
397 		return -EINVAL;
398 	sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
399 	if (sr->msg_flags & MSG_DONTWAIT)
400 		req->flags |= REQ_F_NOWAIT;
401 
402 #ifdef CONFIG_COMPAT
403 	if (req->ctx->compat)
404 		sr->msg_flags |= MSG_CMSG_COMPAT;
405 #endif
406 	return 0;
407 }
408 
409 static void io_req_msg_cleanup(struct io_kiocb *req,
410 			       struct io_async_msghdr *kmsg,
411 			       unsigned int issue_flags)
412 {
413 	req->flags &= ~REQ_F_NEED_CLEANUP;
414 	/* fast path, check for non-NULL to avoid function call */
415 	if (kmsg->free_iov)
416 		kfree(kmsg->free_iov);
417 	io_netmsg_recycle(req, issue_flags);
418 }
419 
420 int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
421 {
422 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
423 	struct io_async_msghdr iomsg, *kmsg;
424 	struct socket *sock;
425 	unsigned flags;
426 	int min_ret = 0;
427 	int ret;
428 
429 	sock = sock_from_file(req->file);
430 	if (unlikely(!sock))
431 		return -ENOTSOCK;
432 
433 	if (req_has_async_data(req)) {
434 		kmsg = req->async_data;
435 		kmsg->msg.msg_control_user = sr->msg_control;
436 	} else {
437 		ret = io_sendmsg_copy_hdr(req, &iomsg);
438 		if (ret)
439 			return ret;
440 		kmsg = &iomsg;
441 	}
442 
443 	if (!(req->flags & REQ_F_POLLED) &&
444 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
445 		return io_setup_async_msg(req, kmsg, issue_flags);
446 
447 	flags = sr->msg_flags;
448 	if (issue_flags & IO_URING_F_NONBLOCK)
449 		flags |= MSG_DONTWAIT;
450 	if (flags & MSG_WAITALL)
451 		min_ret = iov_iter_count(&kmsg->msg.msg_iter);
452 
453 	ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
454 
455 	if (ret < min_ret) {
456 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
457 			return io_setup_async_msg(req, kmsg, issue_flags);
458 		if (ret > 0 && io_net_retry(sock, flags)) {
459 			kmsg->msg.msg_controllen = 0;
460 			kmsg->msg.msg_control = NULL;
461 			sr->done_io += ret;
462 			req->flags |= REQ_F_BL_NO_RECYCLE;
463 			return io_setup_async_msg(req, kmsg, issue_flags);
464 		}
465 		if (ret == -ERESTARTSYS)
466 			ret = -EINTR;
467 		req_set_fail(req);
468 	}
469 	io_req_msg_cleanup(req, kmsg, issue_flags);
470 	if (ret >= 0)
471 		ret += sr->done_io;
472 	else if (sr->done_io)
473 		ret = sr->done_io;
474 	io_req_set_res(req, ret, 0);
475 	return IOU_OK;
476 }
477 
478 int io_send(struct io_kiocb *req, unsigned int issue_flags)
479 {
480 	struct sockaddr_storage __address;
481 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
482 	struct msghdr msg;
483 	struct socket *sock;
484 	unsigned flags;
485 	int min_ret = 0;
486 	int ret;
487 
488 	msg.msg_name = NULL;
489 	msg.msg_control = NULL;
490 	msg.msg_controllen = 0;
491 	msg.msg_namelen = 0;
492 	msg.msg_ubuf = NULL;
493 
494 	if (sr->addr) {
495 		if (req_has_async_data(req)) {
496 			struct io_async_msghdr *io = req->async_data;
497 
498 			msg.msg_name = &io->addr;
499 		} else {
500 			ret = move_addr_to_kernel(sr->addr, sr->addr_len, &__address);
501 			if (unlikely(ret < 0))
502 				return ret;
503 			msg.msg_name = (struct sockaddr *)&__address;
504 		}
505 		msg.msg_namelen = sr->addr_len;
506 	}
507 
508 	if (!(req->flags & REQ_F_POLLED) &&
509 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
510 		return io_setup_async_addr(req, &__address, issue_flags);
511 
512 	sock = sock_from_file(req->file);
513 	if (unlikely(!sock))
514 		return -ENOTSOCK;
515 
516 	ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len, &msg.msg_iter);
517 	if (unlikely(ret))
518 		return ret;
519 
520 	flags = sr->msg_flags;
521 	if (issue_flags & IO_URING_F_NONBLOCK)
522 		flags |= MSG_DONTWAIT;
523 	if (flags & MSG_WAITALL)
524 		min_ret = iov_iter_count(&msg.msg_iter);
525 
526 	flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
527 	msg.msg_flags = flags;
528 	ret = sock_sendmsg(sock, &msg);
529 	if (ret < min_ret) {
530 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
531 			return io_setup_async_addr(req, &__address, issue_flags);
532 
533 		if (ret > 0 && io_net_retry(sock, flags)) {
534 			sr->len -= ret;
535 			sr->buf += ret;
536 			sr->done_io += ret;
537 			req->flags |= REQ_F_BL_NO_RECYCLE;
538 			return io_setup_async_addr(req, &__address, issue_flags);
539 		}
540 		if (ret == -ERESTARTSYS)
541 			ret = -EINTR;
542 		req_set_fail(req);
543 	}
544 	if (ret >= 0)
545 		ret += sr->done_io;
546 	else if (sr->done_io)
547 		ret = sr->done_io;
548 	io_req_set_res(req, ret, 0);
549 	return IOU_OK;
550 }
551 
552 static int io_recvmsg_mshot_prep(struct io_kiocb *req,
553 				 struct io_async_msghdr *iomsg,
554 				 int namelen, size_t controllen)
555 {
556 	if ((req->flags & (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) ==
557 			  (REQ_F_APOLL_MULTISHOT|REQ_F_BUFFER_SELECT)) {
558 		int hdr;
559 
560 		if (unlikely(namelen < 0))
561 			return -EOVERFLOW;
562 		if (check_add_overflow(sizeof(struct io_uring_recvmsg_out),
563 					namelen, &hdr))
564 			return -EOVERFLOW;
565 		if (check_add_overflow(hdr, controllen, &hdr))
566 			return -EOVERFLOW;
567 
568 		iomsg->namelen = namelen;
569 		iomsg->controllen = controllen;
570 		return 0;
571 	}
572 
573 	return 0;
574 }
575 
576 static int io_recvmsg_copy_hdr(struct io_kiocb *req,
577 			       struct io_async_msghdr *iomsg)
578 {
579 	struct user_msghdr msg;
580 	int ret;
581 
582 	iomsg->msg.msg_name = &iomsg->addr;
583 	iomsg->msg.msg_iter.nr_segs = 0;
584 
585 #ifdef CONFIG_COMPAT
586 	if (unlikely(req->ctx->compat)) {
587 		struct compat_msghdr cmsg;
588 
589 		ret = io_compat_msg_copy_hdr(req, iomsg, &cmsg, ITER_DEST);
590 		if (unlikely(ret))
591 			return ret;
592 
593 		ret = __get_compat_msghdr(&iomsg->msg, &cmsg, &iomsg->uaddr);
594 		if (unlikely(ret))
595 			return ret;
596 
597 		return io_recvmsg_mshot_prep(req, iomsg, cmsg.msg_namelen,
598 						cmsg.msg_controllen);
599 	}
600 #endif
601 
602 	ret = io_msg_copy_hdr(req, iomsg, &msg, ITER_DEST);
603 	if (unlikely(ret))
604 		return ret;
605 
606 	ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
607 	if (unlikely(ret))
608 		return ret;
609 
610 	return io_recvmsg_mshot_prep(req, iomsg, msg.msg_namelen,
611 					msg.msg_controllen);
612 }
613 
614 int io_recvmsg_prep_async(struct io_kiocb *req)
615 {
616 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
617 	struct io_async_msghdr *iomsg;
618 	int ret;
619 
620 	sr->done_io = 0;
621 	if (!io_msg_alloc_async_prep(req))
622 		return -ENOMEM;
623 	iomsg = req->async_data;
624 	ret = io_recvmsg_copy_hdr(req, iomsg);
625 	if (!ret)
626 		req->flags |= REQ_F_NEED_CLEANUP;
627 	return ret;
628 }
629 
630 #define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
631 
632 int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
633 {
634 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
635 
636 	sr->done_io = 0;
637 
638 	if (unlikely(sqe->file_index || sqe->addr2))
639 		return -EINVAL;
640 
641 	sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
642 	sr->len = READ_ONCE(sqe->len);
643 	sr->flags = READ_ONCE(sqe->ioprio);
644 	if (sr->flags & ~(RECVMSG_FLAGS))
645 		return -EINVAL;
646 	sr->msg_flags = READ_ONCE(sqe->msg_flags);
647 	if (sr->msg_flags & MSG_DONTWAIT)
648 		req->flags |= REQ_F_NOWAIT;
649 	if (sr->msg_flags & MSG_ERRQUEUE)
650 		req->flags |= REQ_F_CLEAR_POLLIN;
651 	if (sr->flags & IORING_RECV_MULTISHOT) {
652 		if (!(req->flags & REQ_F_BUFFER_SELECT))
653 			return -EINVAL;
654 		if (sr->msg_flags & MSG_WAITALL)
655 			return -EINVAL;
656 		if (req->opcode == IORING_OP_RECV && sr->len)
657 			return -EINVAL;
658 		req->flags |= REQ_F_APOLL_MULTISHOT;
659 		/*
660 		 * Store the buffer group for this multishot receive separately,
661 		 * as if we end up doing an io-wq based issue that selects a
662 		 * buffer, it has to be committed immediately and that will
663 		 * clear ->buf_list. This means we lose the link to the buffer
664 		 * list, and the eventual buffer put on completion then cannot
665 		 * restore it.
666 		 */
667 		sr->buf_group = req->buf_index;
668 	}
669 
670 #ifdef CONFIG_COMPAT
671 	if (req->ctx->compat)
672 		sr->msg_flags |= MSG_CMSG_COMPAT;
673 #endif
674 	sr->nr_multishot_loops = 0;
675 	return 0;
676 }
677 
678 static inline void io_recv_prep_retry(struct io_kiocb *req)
679 {
680 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
681 
682 	req->flags &= ~REQ_F_BL_EMPTY;
683 	sr->done_io = 0;
684 	sr->len = 0; /* get from the provided buffer */
685 	req->buf_index = sr->buf_group;
686 }
687 
688 /*
689  * Finishes io_recv and io_recvmsg.
690  *
691  * Returns true if it is actually finished, or false if it should run
692  * again (for multishot).
693  */
694 static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
695 				  struct msghdr *msg, bool mshot_finished,
696 				  unsigned issue_flags)
697 {
698 	unsigned int cflags;
699 
700 	cflags = io_put_kbuf(req, issue_flags);
701 	if (msg->msg_inq > 0)
702 		cflags |= IORING_CQE_F_SOCK_NONEMPTY;
703 
704 	/*
705 	 * Fill CQE for this receive and see if we should keep trying to
706 	 * receive from this socket.
707 	 */
708 	if ((req->flags & REQ_F_APOLL_MULTISHOT) && !mshot_finished &&
709 	    io_fill_cqe_req_aux(req, issue_flags & IO_URING_F_COMPLETE_DEFER,
710 				*ret, cflags | IORING_CQE_F_MORE)) {
711 		struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
712 		int mshot_retry_ret = IOU_ISSUE_SKIP_COMPLETE;
713 
714 		io_recv_prep_retry(req);
715 		/* Known not-empty or unknown state, retry */
716 		if (cflags & IORING_CQE_F_SOCK_NONEMPTY || msg->msg_inq < 0) {
717 			if (sr->nr_multishot_loops++ < MULTISHOT_MAX_RETRY)
718 				return false;
719 			/* mshot retries exceeded, force a requeue */
720 			sr->nr_multishot_loops = 0;
721 			mshot_retry_ret = IOU_REQUEUE;
722 		}
723 		if (issue_flags & IO_URING_F_MULTISHOT)
724 			*ret = mshot_retry_ret;
725 		else
726 			*ret = -EAGAIN;
727 		return true;
728 	}
729 
730 	/* Finish the request / stop multishot. */
731 	io_req_set_res(req, *ret, cflags);
732 
733 	if (issue_flags & IO_URING_F_MULTISHOT)
734 		*ret = IOU_STOP_MULTISHOT;
735 	else
736 		*ret = IOU_OK;
737 	return true;
738 }
739 
740 static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
741 				     struct io_sr_msg *sr, void __user **buf,
742 				     size_t *len)
743 {
744 	unsigned long ubuf = (unsigned long) *buf;
745 	unsigned long hdr;
746 
747 	hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
748 		kmsg->controllen;
749 	if (*len < hdr)
750 		return -EFAULT;
751 
752 	if (kmsg->controllen) {
753 		unsigned long control = ubuf + hdr - kmsg->controllen;
754 
755 		kmsg->msg.msg_control_user = (void __user *) control;
756 		kmsg->msg.msg_controllen = kmsg->controllen;
757 	}
758 
759 	sr->buf = *buf; /* stash for later copy */
760 	*buf = (void __user *) (ubuf + hdr);
761 	kmsg->payloadlen = *len = *len - hdr;
762 	return 0;
763 }
764 
765 struct io_recvmsg_multishot_hdr {
766 	struct io_uring_recvmsg_out msg;
767 	struct sockaddr_storage addr;
768 };
769 
770 static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
771 				struct io_async_msghdr *kmsg,
772 				unsigned int flags, bool *finished)
773 {
774 	int err;
775 	int copy_len;
776 	struct io_recvmsg_multishot_hdr hdr;
777 
778 	if (kmsg->namelen)
779 		kmsg->msg.msg_name = &hdr.addr;
780 	kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
781 	kmsg->msg.msg_namelen = 0;
782 
783 	if (sock->file->f_flags & O_NONBLOCK)
784 		flags |= MSG_DONTWAIT;
785 
786 	err = sock_recvmsg(sock, &kmsg->msg, flags);
787 	*finished = err <= 0;
788 	if (err < 0)
789 		return err;
790 
791 	hdr.msg = (struct io_uring_recvmsg_out) {
792 		.controllen = kmsg->controllen - kmsg->msg.msg_controllen,
793 		.flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
794 	};
795 
796 	hdr.msg.payloadlen = err;
797 	if (err > kmsg->payloadlen)
798 		err = kmsg->payloadlen;
799 
800 	copy_len = sizeof(struct io_uring_recvmsg_out);
801 	if (kmsg->msg.msg_namelen > kmsg->namelen)
802 		copy_len += kmsg->namelen;
803 	else
804 		copy_len += kmsg->msg.msg_namelen;
805 
806 	/*
807 	 *      "fromlen shall refer to the value before truncation.."
808 	 *                      1003.1g
809 	 */
810 	hdr.msg.namelen = kmsg->msg.msg_namelen;
811 
812 	/* ensure that there is no gap between hdr and sockaddr_storage */
813 	BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
814 		     sizeof(struct io_uring_recvmsg_out));
815 	if (copy_to_user(io->buf, &hdr, copy_len)) {
816 		*finished = true;
817 		return -EFAULT;
818 	}
819 
820 	return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
821 			kmsg->controllen + err;
822 }
823 
824 int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
825 {
826 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
827 	struct io_async_msghdr iomsg, *kmsg;
828 	struct socket *sock;
829 	unsigned flags;
830 	int ret, min_ret = 0;
831 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
832 	bool mshot_finished = true;
833 
834 	sock = sock_from_file(req->file);
835 	if (unlikely(!sock))
836 		return -ENOTSOCK;
837 
838 	if (req_has_async_data(req)) {
839 		kmsg = req->async_data;
840 	} else {
841 		ret = io_recvmsg_copy_hdr(req, &iomsg);
842 		if (ret)
843 			return ret;
844 		kmsg = &iomsg;
845 	}
846 
847 	if (!(req->flags & REQ_F_POLLED) &&
848 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
849 		return io_setup_async_msg(req, kmsg, issue_flags);
850 
851 	flags = sr->msg_flags;
852 	if (force_nonblock)
853 		flags |= MSG_DONTWAIT;
854 
855 retry_multishot:
856 	if (io_do_buffer_select(req)) {
857 		void __user *buf;
858 		size_t len = sr->len;
859 
860 		buf = io_buffer_select(req, &len, issue_flags);
861 		if (!buf)
862 			return -ENOBUFS;
863 
864 		if (req->flags & REQ_F_APOLL_MULTISHOT) {
865 			ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
866 			if (ret) {
867 				io_kbuf_recycle(req, issue_flags);
868 				return ret;
869 			}
870 		}
871 
872 		iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_DEST, buf, len);
873 	}
874 
875 	kmsg->msg.msg_get_inq = 1;
876 	kmsg->msg.msg_inq = -1;
877 	if (req->flags & REQ_F_APOLL_MULTISHOT) {
878 		ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
879 					   &mshot_finished);
880 	} else {
881 		/* disable partial retry for recvmsg with cmsg attached */
882 		if (flags & MSG_WAITALL && !kmsg->msg.msg_controllen)
883 			min_ret = iov_iter_count(&kmsg->msg.msg_iter);
884 
885 		ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
886 					 kmsg->uaddr, flags);
887 	}
888 
889 	if (ret < min_ret) {
890 		if (ret == -EAGAIN && force_nonblock) {
891 			ret = io_setup_async_msg(req, kmsg, issue_flags);
892 			if (ret == -EAGAIN && (issue_flags & IO_URING_F_MULTISHOT)) {
893 				io_kbuf_recycle(req, issue_flags);
894 				return IOU_ISSUE_SKIP_COMPLETE;
895 			}
896 			return ret;
897 		}
898 		if (ret > 0 && io_net_retry(sock, flags)) {
899 			sr->done_io += ret;
900 			req->flags |= REQ_F_BL_NO_RECYCLE;
901 			return io_setup_async_msg(req, kmsg, issue_flags);
902 		}
903 		if (ret == -ERESTARTSYS)
904 			ret = -EINTR;
905 		req_set_fail(req);
906 	} else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
907 		req_set_fail(req);
908 	}
909 
910 	if (ret > 0)
911 		ret += sr->done_io;
912 	else if (sr->done_io)
913 		ret = sr->done_io;
914 	else
915 		io_kbuf_recycle(req, issue_flags);
916 
917 	if (!io_recv_finish(req, &ret, &kmsg->msg, mshot_finished, issue_flags))
918 		goto retry_multishot;
919 
920 	if (mshot_finished)
921 		io_req_msg_cleanup(req, kmsg, issue_flags);
922 	else if (ret == -EAGAIN)
923 		return io_setup_async_msg(req, kmsg, issue_flags);
924 
925 	return ret;
926 }
927 
928 int io_recv(struct io_kiocb *req, unsigned int issue_flags)
929 {
930 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
931 	struct msghdr msg;
932 	struct socket *sock;
933 	unsigned flags;
934 	int ret, min_ret = 0;
935 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
936 	size_t len = sr->len;
937 
938 	if (!(req->flags & REQ_F_POLLED) &&
939 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
940 		return -EAGAIN;
941 
942 	sock = sock_from_file(req->file);
943 	if (unlikely(!sock))
944 		return -ENOTSOCK;
945 
946 	msg.msg_name = NULL;
947 	msg.msg_namelen = 0;
948 	msg.msg_control = NULL;
949 	msg.msg_get_inq = 1;
950 	msg.msg_controllen = 0;
951 	msg.msg_iocb = NULL;
952 	msg.msg_ubuf = NULL;
953 
954 	flags = sr->msg_flags;
955 	if (force_nonblock)
956 		flags |= MSG_DONTWAIT;
957 
958 retry_multishot:
959 	if (io_do_buffer_select(req)) {
960 		void __user *buf;
961 
962 		buf = io_buffer_select(req, &len, issue_flags);
963 		if (!buf)
964 			return -ENOBUFS;
965 		sr->buf = buf;
966 		sr->len = len;
967 	}
968 
969 	ret = import_ubuf(ITER_DEST, sr->buf, len, &msg.msg_iter);
970 	if (unlikely(ret))
971 		goto out_free;
972 
973 	msg.msg_inq = -1;
974 	msg.msg_flags = 0;
975 
976 	if (flags & MSG_WAITALL)
977 		min_ret = iov_iter_count(&msg.msg_iter);
978 
979 	ret = sock_recvmsg(sock, &msg, flags);
980 	if (ret < min_ret) {
981 		if (ret == -EAGAIN && force_nonblock) {
982 			if (issue_flags & IO_URING_F_MULTISHOT) {
983 				io_kbuf_recycle(req, issue_flags);
984 				return IOU_ISSUE_SKIP_COMPLETE;
985 			}
986 
987 			return -EAGAIN;
988 		}
989 		if (ret > 0 && io_net_retry(sock, flags)) {
990 			sr->len -= ret;
991 			sr->buf += ret;
992 			sr->done_io += ret;
993 			req->flags |= REQ_F_BL_NO_RECYCLE;
994 			return -EAGAIN;
995 		}
996 		if (ret == -ERESTARTSYS)
997 			ret = -EINTR;
998 		req_set_fail(req);
999 	} else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
1000 out_free:
1001 		req_set_fail(req);
1002 	}
1003 
1004 	if (ret > 0)
1005 		ret += sr->done_io;
1006 	else if (sr->done_io)
1007 		ret = sr->done_io;
1008 	else
1009 		io_kbuf_recycle(req, issue_flags);
1010 
1011 	if (!io_recv_finish(req, &ret, &msg, ret <= 0, issue_flags))
1012 		goto retry_multishot;
1013 
1014 	return ret;
1015 }
1016 
1017 void io_send_zc_cleanup(struct io_kiocb *req)
1018 {
1019 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1020 	struct io_async_msghdr *io;
1021 
1022 	if (req_has_async_data(req)) {
1023 		io = req->async_data;
1024 		/* might be ->fast_iov if *msg_copy_hdr failed */
1025 		if (io->free_iov != io->fast_iov)
1026 			kfree(io->free_iov);
1027 	}
1028 	if (zc->notif) {
1029 		io_notif_flush(zc->notif);
1030 		zc->notif = NULL;
1031 	}
1032 }
1033 
1034 #define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF)
1035 #define IO_ZC_FLAGS_VALID  (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE)
1036 
1037 int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1038 {
1039 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1040 	struct io_ring_ctx *ctx = req->ctx;
1041 	struct io_kiocb *notif;
1042 
1043 	zc->done_io = 0;
1044 
1045 	if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
1046 		return -EINVAL;
1047 	/* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
1048 	if (req->flags & REQ_F_CQE_SKIP)
1049 		return -EINVAL;
1050 
1051 	notif = zc->notif = io_alloc_notif(ctx);
1052 	if (!notif)
1053 		return -ENOMEM;
1054 	notif->cqe.user_data = req->cqe.user_data;
1055 	notif->cqe.res = 0;
1056 	notif->cqe.flags = IORING_CQE_F_NOTIF;
1057 	req->flags |= REQ_F_NEED_CLEANUP;
1058 
1059 	zc->flags = READ_ONCE(sqe->ioprio);
1060 	if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) {
1061 		if (zc->flags & ~IO_ZC_FLAGS_VALID)
1062 			return -EINVAL;
1063 		if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) {
1064 			io_notif_set_extended(notif);
1065 			io_notif_to_data(notif)->zc_report = true;
1066 		}
1067 	}
1068 
1069 	if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1070 		unsigned idx = READ_ONCE(sqe->buf_index);
1071 
1072 		if (unlikely(idx >= ctx->nr_user_bufs))
1073 			return -EFAULT;
1074 		idx = array_index_nospec(idx, ctx->nr_user_bufs);
1075 		req->imu = READ_ONCE(ctx->user_bufs[idx]);
1076 		io_req_set_rsrc_node(notif, ctx, 0);
1077 	}
1078 
1079 	if (req->opcode == IORING_OP_SEND_ZC) {
1080 		if (READ_ONCE(sqe->__pad3[0]))
1081 			return -EINVAL;
1082 		zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1083 		zc->addr_len = READ_ONCE(sqe->addr_len);
1084 	} else {
1085 		if (unlikely(sqe->addr2 || sqe->file_index))
1086 			return -EINVAL;
1087 		if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
1088 			return -EINVAL;
1089 	}
1090 
1091 	zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
1092 	zc->len = READ_ONCE(sqe->len);
1093 	zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
1094 	if (zc->msg_flags & MSG_DONTWAIT)
1095 		req->flags |= REQ_F_NOWAIT;
1096 
1097 #ifdef CONFIG_COMPAT
1098 	if (req->ctx->compat)
1099 		zc->msg_flags |= MSG_CMSG_COMPAT;
1100 #endif
1101 	return 0;
1102 }
1103 
1104 static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
1105 				 struct iov_iter *from, size_t length)
1106 {
1107 	skb_zcopy_downgrade_managed(skb);
1108 	return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1109 }
1110 
1111 static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1112 			   struct iov_iter *from, size_t length)
1113 {
1114 	struct skb_shared_info *shinfo = skb_shinfo(skb);
1115 	int frag = shinfo->nr_frags;
1116 	int ret = 0;
1117 	struct bvec_iter bi;
1118 	ssize_t copied = 0;
1119 	unsigned long truesize = 0;
1120 
1121 	if (!frag)
1122 		shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
1123 	else if (unlikely(!skb_zcopy_managed(skb)))
1124 		return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1125 
1126 	bi.bi_size = min(from->count, length);
1127 	bi.bi_bvec_done = from->iov_offset;
1128 	bi.bi_idx = 0;
1129 
1130 	while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1131 		struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1132 
1133 		copied += v.bv_len;
1134 		truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1135 		__skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1136 					   v.bv_offset, v.bv_len);
1137 		bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1138 	}
1139 	if (bi.bi_size)
1140 		ret = -EMSGSIZE;
1141 
1142 	shinfo->nr_frags = frag;
1143 	from->bvec += bi.bi_idx;
1144 	from->nr_segs -= bi.bi_idx;
1145 	from->count -= copied;
1146 	from->iov_offset = bi.bi_bvec_done;
1147 
1148 	skb->data_len += copied;
1149 	skb->len += copied;
1150 	skb->truesize += truesize;
1151 
1152 	if (sk && sk->sk_type == SOCK_STREAM) {
1153 		sk_wmem_queued_add(sk, truesize);
1154 		if (!skb_zcopy_pure(skb))
1155 			sk_mem_charge(sk, truesize);
1156 	} else {
1157 		refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1158 	}
1159 	return ret;
1160 }
1161 
1162 int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
1163 {
1164 	struct sockaddr_storage __address;
1165 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1166 	struct msghdr msg;
1167 	struct socket *sock;
1168 	unsigned msg_flags;
1169 	int ret, min_ret = 0;
1170 
1171 	sock = sock_from_file(req->file);
1172 	if (unlikely(!sock))
1173 		return -ENOTSOCK;
1174 	if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1175 		return -EOPNOTSUPP;
1176 
1177 	msg.msg_name = NULL;
1178 	msg.msg_control = NULL;
1179 	msg.msg_controllen = 0;
1180 	msg.msg_namelen = 0;
1181 
1182 	if (zc->addr) {
1183 		if (req_has_async_data(req)) {
1184 			struct io_async_msghdr *io = req->async_data;
1185 
1186 			msg.msg_name = &io->addr;
1187 		} else {
1188 			ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1189 			if (unlikely(ret < 0))
1190 				return ret;
1191 			msg.msg_name = (struct sockaddr *)&__address;
1192 		}
1193 		msg.msg_namelen = zc->addr_len;
1194 	}
1195 
1196 	if (!(req->flags & REQ_F_POLLED) &&
1197 	    (zc->flags & IORING_RECVSEND_POLL_FIRST))
1198 		return io_setup_async_addr(req, &__address, issue_flags);
1199 
1200 	if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1201 		ret = io_import_fixed(ITER_SOURCE, &msg.msg_iter, req->imu,
1202 					(u64)(uintptr_t)zc->buf, zc->len);
1203 		if (unlikely(ret))
1204 			return ret;
1205 		msg.sg_from_iter = io_sg_from_iter;
1206 	} else {
1207 		io_notif_set_extended(zc->notif);
1208 		ret = import_ubuf(ITER_SOURCE, zc->buf, zc->len, &msg.msg_iter);
1209 		if (unlikely(ret))
1210 			return ret;
1211 		ret = io_notif_account_mem(zc->notif, zc->len);
1212 		if (unlikely(ret))
1213 			return ret;
1214 		msg.sg_from_iter = io_sg_from_iter_iovec;
1215 	}
1216 
1217 	msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1218 	if (issue_flags & IO_URING_F_NONBLOCK)
1219 		msg_flags |= MSG_DONTWAIT;
1220 	if (msg_flags & MSG_WAITALL)
1221 		min_ret = iov_iter_count(&msg.msg_iter);
1222 	msg_flags &= ~MSG_INTERNAL_SENDMSG_FLAGS;
1223 
1224 	msg.msg_flags = msg_flags;
1225 	msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
1226 	ret = sock_sendmsg(sock, &msg);
1227 
1228 	if (unlikely(ret < min_ret)) {
1229 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1230 			return io_setup_async_addr(req, &__address, issue_flags);
1231 
1232 		if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1233 			zc->len -= ret;
1234 			zc->buf += ret;
1235 			zc->done_io += ret;
1236 			req->flags |= REQ_F_BL_NO_RECYCLE;
1237 			return io_setup_async_addr(req, &__address, issue_flags);
1238 		}
1239 		if (ret == -ERESTARTSYS)
1240 			ret = -EINTR;
1241 		req_set_fail(req);
1242 	}
1243 
1244 	if (ret >= 0)
1245 		ret += zc->done_io;
1246 	else if (zc->done_io)
1247 		ret = zc->done_io;
1248 
1249 	/*
1250 	 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1251 	 * flushing notif to io_send_zc_cleanup()
1252 	 */
1253 	if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1254 		io_notif_flush(zc->notif);
1255 		req->flags &= ~REQ_F_NEED_CLEANUP;
1256 	}
1257 	io_req_set_res(req, ret, IORING_CQE_F_MORE);
1258 	return IOU_OK;
1259 }
1260 
1261 int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1262 {
1263 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1264 	struct io_async_msghdr iomsg, *kmsg;
1265 	struct socket *sock;
1266 	unsigned flags;
1267 	int ret, min_ret = 0;
1268 
1269 	io_notif_set_extended(sr->notif);
1270 
1271 	sock = sock_from_file(req->file);
1272 	if (unlikely(!sock))
1273 		return -ENOTSOCK;
1274 	if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1275 		return -EOPNOTSUPP;
1276 
1277 	if (req_has_async_data(req)) {
1278 		kmsg = req->async_data;
1279 	} else {
1280 		ret = io_sendmsg_copy_hdr(req, &iomsg);
1281 		if (ret)
1282 			return ret;
1283 		kmsg = &iomsg;
1284 	}
1285 
1286 	if (!(req->flags & REQ_F_POLLED) &&
1287 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
1288 		return io_setup_async_msg(req, kmsg, issue_flags);
1289 
1290 	flags = sr->msg_flags | MSG_ZEROCOPY;
1291 	if (issue_flags & IO_URING_F_NONBLOCK)
1292 		flags |= MSG_DONTWAIT;
1293 	if (flags & MSG_WAITALL)
1294 		min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1295 
1296 	kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1297 	kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1298 	ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1299 
1300 	if (unlikely(ret < min_ret)) {
1301 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1302 			return io_setup_async_msg(req, kmsg, issue_flags);
1303 
1304 		if (ret > 0 && io_net_retry(sock, flags)) {
1305 			sr->done_io += ret;
1306 			req->flags |= REQ_F_BL_NO_RECYCLE;
1307 			return io_setup_async_msg(req, kmsg, issue_flags);
1308 		}
1309 		if (ret == -ERESTARTSYS)
1310 			ret = -EINTR;
1311 		req_set_fail(req);
1312 	}
1313 	/* fast path, check for non-NULL to avoid function call */
1314 	if (kmsg->free_iov) {
1315 		kfree(kmsg->free_iov);
1316 		kmsg->free_iov = NULL;
1317 	}
1318 
1319 	io_netmsg_recycle(req, issue_flags);
1320 	if (ret >= 0)
1321 		ret += sr->done_io;
1322 	else if (sr->done_io)
1323 		ret = sr->done_io;
1324 
1325 	/*
1326 	 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1327 	 * flushing notif to io_send_zc_cleanup()
1328 	 */
1329 	if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1330 		io_notif_flush(sr->notif);
1331 		req->flags &= ~REQ_F_NEED_CLEANUP;
1332 	}
1333 	io_req_set_res(req, ret, IORING_CQE_F_MORE);
1334 	return IOU_OK;
1335 }
1336 
1337 void io_sendrecv_fail(struct io_kiocb *req)
1338 {
1339 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1340 
1341 	if (sr->done_io)
1342 		req->cqe.res = sr->done_io;
1343 
1344 	if ((req->flags & REQ_F_NEED_CLEANUP) &&
1345 	    (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1346 		req->cqe.flags |= IORING_CQE_F_MORE;
1347 }
1348 
1349 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1350 {
1351 	struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1352 	unsigned flags;
1353 
1354 	if (sqe->len || sqe->buf_index)
1355 		return -EINVAL;
1356 
1357 	accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1358 	accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1359 	accept->flags = READ_ONCE(sqe->accept_flags);
1360 	accept->nofile = rlimit(RLIMIT_NOFILE);
1361 	flags = READ_ONCE(sqe->ioprio);
1362 	if (flags & ~IORING_ACCEPT_MULTISHOT)
1363 		return -EINVAL;
1364 
1365 	accept->file_slot = READ_ONCE(sqe->file_index);
1366 	if (accept->file_slot) {
1367 		if (accept->flags & SOCK_CLOEXEC)
1368 			return -EINVAL;
1369 		if (flags & IORING_ACCEPT_MULTISHOT &&
1370 		    accept->file_slot != IORING_FILE_INDEX_ALLOC)
1371 			return -EINVAL;
1372 	}
1373 	if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1374 		return -EINVAL;
1375 	if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1376 		accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1377 	if (flags & IORING_ACCEPT_MULTISHOT)
1378 		req->flags |= REQ_F_APOLL_MULTISHOT;
1379 	return 0;
1380 }
1381 
1382 int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1383 {
1384 	struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1385 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1386 	unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1387 	bool fixed = !!accept->file_slot;
1388 	struct file *file;
1389 	int ret, fd;
1390 
1391 retry:
1392 	if (!fixed) {
1393 		fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1394 		if (unlikely(fd < 0))
1395 			return fd;
1396 	}
1397 	file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1398 			 accept->flags);
1399 	if (IS_ERR(file)) {
1400 		if (!fixed)
1401 			put_unused_fd(fd);
1402 		ret = PTR_ERR(file);
1403 		if (ret == -EAGAIN && force_nonblock) {
1404 			/*
1405 			 * if it's multishot and polled, we don't need to
1406 			 * return EAGAIN to arm the poll infra since it
1407 			 * has already been done
1408 			 */
1409 			if (issue_flags & IO_URING_F_MULTISHOT)
1410 				return IOU_ISSUE_SKIP_COMPLETE;
1411 			return ret;
1412 		}
1413 		if (ret == -ERESTARTSYS)
1414 			ret = -EINTR;
1415 		req_set_fail(req);
1416 	} else if (!fixed) {
1417 		fd_install(fd, file);
1418 		ret = fd;
1419 	} else {
1420 		ret = io_fixed_fd_install(req, issue_flags, file,
1421 						accept->file_slot);
1422 	}
1423 
1424 	if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1425 		io_req_set_res(req, ret, 0);
1426 		return IOU_OK;
1427 	}
1428 
1429 	if (ret < 0)
1430 		return ret;
1431 	if (io_fill_cqe_req_aux(req, issue_flags & IO_URING_F_COMPLETE_DEFER,
1432 				ret, IORING_CQE_F_MORE))
1433 		goto retry;
1434 
1435 	io_req_set_res(req, ret, 0);
1436 	return IOU_STOP_MULTISHOT;
1437 }
1438 
1439 int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1440 {
1441 	struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1442 
1443 	if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1444 		return -EINVAL;
1445 
1446 	sock->domain = READ_ONCE(sqe->fd);
1447 	sock->type = READ_ONCE(sqe->off);
1448 	sock->protocol = READ_ONCE(sqe->len);
1449 	sock->file_slot = READ_ONCE(sqe->file_index);
1450 	sock->nofile = rlimit(RLIMIT_NOFILE);
1451 
1452 	sock->flags = sock->type & ~SOCK_TYPE_MASK;
1453 	if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1454 		return -EINVAL;
1455 	if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1456 		return -EINVAL;
1457 	return 0;
1458 }
1459 
1460 int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1461 {
1462 	struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1463 	bool fixed = !!sock->file_slot;
1464 	struct file *file;
1465 	int ret, fd;
1466 
1467 	if (!fixed) {
1468 		fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1469 		if (unlikely(fd < 0))
1470 			return fd;
1471 	}
1472 	file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1473 	if (IS_ERR(file)) {
1474 		if (!fixed)
1475 			put_unused_fd(fd);
1476 		ret = PTR_ERR(file);
1477 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1478 			return -EAGAIN;
1479 		if (ret == -ERESTARTSYS)
1480 			ret = -EINTR;
1481 		req_set_fail(req);
1482 	} else if (!fixed) {
1483 		fd_install(fd, file);
1484 		ret = fd;
1485 	} else {
1486 		ret = io_fixed_fd_install(req, issue_flags, file,
1487 					    sock->file_slot);
1488 	}
1489 	io_req_set_res(req, ret, 0);
1490 	return IOU_OK;
1491 }
1492 
1493 int io_connect_prep_async(struct io_kiocb *req)
1494 {
1495 	struct io_async_connect *io = req->async_data;
1496 	struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1497 
1498 	return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1499 }
1500 
1501 int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1502 {
1503 	struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1504 
1505 	if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1506 		return -EINVAL;
1507 
1508 	conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1509 	conn->addr_len =  READ_ONCE(sqe->addr2);
1510 	conn->in_progress = conn->seen_econnaborted = false;
1511 	return 0;
1512 }
1513 
1514 int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1515 {
1516 	struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1517 	struct io_async_connect __io, *io;
1518 	unsigned file_flags;
1519 	int ret;
1520 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1521 
1522 	if (req_has_async_data(req)) {
1523 		io = req->async_data;
1524 	} else {
1525 		ret = move_addr_to_kernel(connect->addr,
1526 						connect->addr_len,
1527 						&__io.address);
1528 		if (ret)
1529 			goto out;
1530 		io = &__io;
1531 	}
1532 
1533 	file_flags = force_nonblock ? O_NONBLOCK : 0;
1534 
1535 	ret = __sys_connect_file(req->file, &io->address,
1536 					connect->addr_len, file_flags);
1537 	if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED)
1538 	    && force_nonblock) {
1539 		if (ret == -EINPROGRESS) {
1540 			connect->in_progress = true;
1541 		} else if (ret == -ECONNABORTED) {
1542 			if (connect->seen_econnaborted)
1543 				goto out;
1544 			connect->seen_econnaborted = true;
1545 		}
1546 		if (req_has_async_data(req))
1547 			return -EAGAIN;
1548 		if (io_alloc_async_data(req)) {
1549 			ret = -ENOMEM;
1550 			goto out;
1551 		}
1552 		memcpy(req->async_data, &__io, sizeof(__io));
1553 		return -EAGAIN;
1554 	}
1555 	if (connect->in_progress) {
1556 		/*
1557 		 * At least bluetooth will return -EBADFD on a re-connect
1558 		 * attempt, and it's (supposedly) also valid to get -EISCONN
1559 		 * which means the previous result is good. For both of these,
1560 		 * grab the sock_error() and use that for the completion.
1561 		 */
1562 		if (ret == -EBADFD || ret == -EISCONN)
1563 			ret = sock_error(sock_from_file(req->file)->sk);
1564 	}
1565 	if (ret == -ERESTARTSYS)
1566 		ret = -EINTR;
1567 out:
1568 	if (ret < 0)
1569 		req_set_fail(req);
1570 	io_req_set_res(req, ret, 0);
1571 	return IOU_OK;
1572 }
1573 
1574 void io_netmsg_cache_free(struct io_cache_entry *entry)
1575 {
1576 	kfree(container_of(entry, struct io_async_msghdr, cache));
1577 }
1578 #endif
1579