xref: /openbmc/linux/io_uring/net.c (revision 9f7cc57f)
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 	u16				flags;
64 	/* initialised and used only by !msg send variants */
65 	u16				addr_len;
66 	u16				buf_group;
67 	void __user			*addr;
68 	/* used only for send zerocopy */
69 	struct io_kiocb 		*notif;
70 };
71 
72 static inline bool io_check_multishot(struct io_kiocb *req,
73 				      unsigned int issue_flags)
74 {
75 	/*
76 	 * When ->locked_cq is set we only allow to post CQEs from the original
77 	 * task context. Usual request completions will be handled in other
78 	 * generic paths but multipoll may decide to post extra cqes.
79 	 */
80 	return !(issue_flags & IO_URING_F_IOWQ) ||
81 		!(issue_flags & IO_URING_F_MULTISHOT) ||
82 		!req->ctx->task_complete;
83 }
84 
85 int io_shutdown_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
86 {
87 	struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
88 
89 	if (unlikely(sqe->off || sqe->addr || sqe->rw_flags ||
90 		     sqe->buf_index || sqe->splice_fd_in))
91 		return -EINVAL;
92 
93 	shutdown->how = READ_ONCE(sqe->len);
94 	req->flags |= REQ_F_FORCE_ASYNC;
95 	return 0;
96 }
97 
98 int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
99 {
100 	struct io_shutdown *shutdown = io_kiocb_to_cmd(req, struct io_shutdown);
101 	struct socket *sock;
102 	int ret;
103 
104 	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
105 
106 	sock = sock_from_file(req->file);
107 	if (unlikely(!sock))
108 		return -ENOTSOCK;
109 
110 	ret = __sys_shutdown_sock(sock, shutdown->how);
111 	io_req_set_res(req, ret, 0);
112 	return IOU_OK;
113 }
114 
115 static bool io_net_retry(struct socket *sock, int flags)
116 {
117 	if (!(flags & MSG_WAITALL))
118 		return false;
119 	return sock->type == SOCK_STREAM || sock->type == SOCK_SEQPACKET;
120 }
121 
122 static void io_netmsg_recycle(struct io_kiocb *req, unsigned int issue_flags)
123 {
124 	struct io_async_msghdr *hdr = req->async_data;
125 
126 	if (!req_has_async_data(req) || issue_flags & IO_URING_F_UNLOCKED)
127 		return;
128 
129 	/* Let normal cleanup path reap it if we fail adding to the cache */
130 	if (io_alloc_cache_put(&req->ctx->netmsg_cache, &hdr->cache)) {
131 		req->async_data = NULL;
132 		req->flags &= ~REQ_F_ASYNC_DATA;
133 	}
134 }
135 
136 static struct io_async_msghdr *io_msg_alloc_async(struct io_kiocb *req,
137 						  unsigned int issue_flags)
138 {
139 	struct io_ring_ctx *ctx = req->ctx;
140 	struct io_cache_entry *entry;
141 	struct io_async_msghdr *hdr;
142 
143 	if (!(issue_flags & IO_URING_F_UNLOCKED)) {
144 		entry = io_alloc_cache_get(&ctx->netmsg_cache);
145 		if (entry) {
146 			hdr = container_of(entry, struct io_async_msghdr, cache);
147 			hdr->free_iov = NULL;
148 			req->flags |= REQ_F_ASYNC_DATA;
149 			req->async_data = hdr;
150 			return hdr;
151 		}
152 	}
153 
154 	if (!io_alloc_async_data(req)) {
155 		hdr = req->async_data;
156 		hdr->free_iov = NULL;
157 		return hdr;
158 	}
159 	return NULL;
160 }
161 
162 static inline struct io_async_msghdr *io_msg_alloc_async_prep(struct io_kiocb *req)
163 {
164 	/* ->prep_async is always called from the submission context */
165 	return io_msg_alloc_async(req, 0);
166 }
167 
168 static int io_setup_async_msg(struct io_kiocb *req,
169 			      struct io_async_msghdr *kmsg,
170 			      unsigned int issue_flags)
171 {
172 	struct io_async_msghdr *async_msg;
173 
174 	if (req_has_async_data(req))
175 		return -EAGAIN;
176 	async_msg = io_msg_alloc_async(req, issue_flags);
177 	if (!async_msg) {
178 		kfree(kmsg->free_iov);
179 		return -ENOMEM;
180 	}
181 	req->flags |= REQ_F_NEED_CLEANUP;
182 	memcpy(async_msg, kmsg, sizeof(*kmsg));
183 	if (async_msg->msg.msg_name)
184 		async_msg->msg.msg_name = &async_msg->addr;
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 = kmsg->msg.msg_iter.iov - kmsg->fast_iov;
188 		async_msg->msg.msg_iter.iov = &async_msg->fast_iov[fast_idx];
189 	}
190 
191 	return -EAGAIN;
192 }
193 
194 static int io_sendmsg_copy_hdr(struct io_kiocb *req,
195 			       struct io_async_msghdr *iomsg)
196 {
197 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
198 
199 	iomsg->msg.msg_name = &iomsg->addr;
200 	iomsg->free_iov = iomsg->fast_iov;
201 	return sendmsg_copy_msghdr(&iomsg->msg, sr->umsg, sr->msg_flags,
202 					&iomsg->free_iov);
203 }
204 
205 int io_send_prep_async(struct io_kiocb *req)
206 {
207 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
208 	struct io_async_msghdr *io;
209 	int ret;
210 
211 	if (!zc->addr || req_has_async_data(req))
212 		return 0;
213 	io = io_msg_alloc_async_prep(req);
214 	if (!io)
215 		return -ENOMEM;
216 	ret = move_addr_to_kernel(zc->addr, zc->addr_len, &io->addr);
217 	return ret;
218 }
219 
220 static int io_setup_async_addr(struct io_kiocb *req,
221 			      struct sockaddr_storage *addr_storage,
222 			      unsigned int issue_flags)
223 {
224 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
225 	struct io_async_msghdr *io;
226 
227 	if (!sr->addr || req_has_async_data(req))
228 		return -EAGAIN;
229 	io = io_msg_alloc_async(req, issue_flags);
230 	if (!io)
231 		return -ENOMEM;
232 	memcpy(&io->addr, addr_storage, sizeof(io->addr));
233 	return -EAGAIN;
234 }
235 
236 int io_sendmsg_prep_async(struct io_kiocb *req)
237 {
238 	int ret;
239 
240 	if (!io_msg_alloc_async_prep(req))
241 		return -ENOMEM;
242 	ret = io_sendmsg_copy_hdr(req, req->async_data);
243 	if (!ret)
244 		req->flags |= REQ_F_NEED_CLEANUP;
245 	return ret;
246 }
247 
248 void io_sendmsg_recvmsg_cleanup(struct io_kiocb *req)
249 {
250 	struct io_async_msghdr *io = req->async_data;
251 
252 	kfree(io->free_iov);
253 }
254 
255 int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
256 {
257 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
258 
259 	if (req->opcode == IORING_OP_SEND) {
260 		if (READ_ONCE(sqe->__pad3[0]))
261 			return -EINVAL;
262 		sr->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
263 		sr->addr_len = READ_ONCE(sqe->addr_len);
264 	} else if (sqe->addr2 || sqe->file_index) {
265 		return -EINVAL;
266 	}
267 
268 	sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
269 	sr->len = READ_ONCE(sqe->len);
270 	sr->flags = READ_ONCE(sqe->ioprio);
271 	if (sr->flags & ~IORING_RECVSEND_POLL_FIRST)
272 		return -EINVAL;
273 	sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
274 	if (sr->msg_flags & MSG_DONTWAIT)
275 		req->flags |= REQ_F_NOWAIT;
276 
277 #ifdef CONFIG_COMPAT
278 	if (req->ctx->compat)
279 		sr->msg_flags |= MSG_CMSG_COMPAT;
280 #endif
281 	sr->done_io = 0;
282 	return 0;
283 }
284 
285 int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
286 {
287 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
288 	struct io_async_msghdr iomsg, *kmsg;
289 	struct socket *sock;
290 	unsigned flags;
291 	int min_ret = 0;
292 	int ret;
293 
294 	sock = sock_from_file(req->file);
295 	if (unlikely(!sock))
296 		return -ENOTSOCK;
297 
298 	if (req_has_async_data(req)) {
299 		kmsg = req->async_data;
300 	} else {
301 		ret = io_sendmsg_copy_hdr(req, &iomsg);
302 		if (ret)
303 			return ret;
304 		kmsg = &iomsg;
305 	}
306 
307 	if (!(req->flags & REQ_F_POLLED) &&
308 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
309 		return io_setup_async_msg(req, kmsg, issue_flags);
310 
311 	flags = sr->msg_flags;
312 	if (issue_flags & IO_URING_F_NONBLOCK)
313 		flags |= MSG_DONTWAIT;
314 	if (flags & MSG_WAITALL)
315 		min_ret = iov_iter_count(&kmsg->msg.msg_iter);
316 
317 	ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
318 
319 	if (ret < min_ret) {
320 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
321 			return io_setup_async_msg(req, kmsg, issue_flags);
322 		if (ret > 0 && io_net_retry(sock, flags)) {
323 			sr->done_io += ret;
324 			req->flags |= REQ_F_PARTIAL_IO;
325 			return io_setup_async_msg(req, kmsg, issue_flags);
326 		}
327 		if (ret == -ERESTARTSYS)
328 			ret = -EINTR;
329 		req_set_fail(req);
330 	}
331 	/* fast path, check for non-NULL to avoid function call */
332 	if (kmsg->free_iov)
333 		kfree(kmsg->free_iov);
334 	req->flags &= ~REQ_F_NEED_CLEANUP;
335 	io_netmsg_recycle(req, issue_flags);
336 	if (ret >= 0)
337 		ret += sr->done_io;
338 	else if (sr->done_io)
339 		ret = sr->done_io;
340 	io_req_set_res(req, ret, 0);
341 	return IOU_OK;
342 }
343 
344 int io_send(struct io_kiocb *req, unsigned int issue_flags)
345 {
346 	struct sockaddr_storage __address;
347 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
348 	struct msghdr msg;
349 	struct socket *sock;
350 	unsigned flags;
351 	int min_ret = 0;
352 	int ret;
353 
354 	msg.msg_name = NULL;
355 	msg.msg_control = NULL;
356 	msg.msg_controllen = 0;
357 	msg.msg_namelen = 0;
358 	msg.msg_ubuf = NULL;
359 
360 	if (sr->addr) {
361 		if (req_has_async_data(req)) {
362 			struct io_async_msghdr *io = req->async_data;
363 
364 			msg.msg_name = &io->addr;
365 		} else {
366 			ret = move_addr_to_kernel(sr->addr, sr->addr_len, &__address);
367 			if (unlikely(ret < 0))
368 				return ret;
369 			msg.msg_name = (struct sockaddr *)&__address;
370 		}
371 		msg.msg_namelen = sr->addr_len;
372 	}
373 
374 	if (!(req->flags & REQ_F_POLLED) &&
375 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
376 		return io_setup_async_addr(req, &__address, issue_flags);
377 
378 	sock = sock_from_file(req->file);
379 	if (unlikely(!sock))
380 		return -ENOTSOCK;
381 
382 	ret = import_ubuf(ITER_SOURCE, sr->buf, sr->len, &msg.msg_iter);
383 	if (unlikely(ret))
384 		return ret;
385 
386 	flags = sr->msg_flags;
387 	if (issue_flags & IO_URING_F_NONBLOCK)
388 		flags |= MSG_DONTWAIT;
389 	if (flags & MSG_WAITALL)
390 		min_ret = iov_iter_count(&msg.msg_iter);
391 
392 	msg.msg_flags = flags;
393 	ret = sock_sendmsg(sock, &msg);
394 	if (ret < min_ret) {
395 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
396 			return io_setup_async_addr(req, &__address, issue_flags);
397 
398 		if (ret > 0 && io_net_retry(sock, flags)) {
399 			sr->len -= ret;
400 			sr->buf += ret;
401 			sr->done_io += ret;
402 			req->flags |= REQ_F_PARTIAL_IO;
403 			return io_setup_async_addr(req, &__address, issue_flags);
404 		}
405 		if (ret == -ERESTARTSYS)
406 			ret = -EINTR;
407 		req_set_fail(req);
408 	}
409 	if (ret >= 0)
410 		ret += sr->done_io;
411 	else if (sr->done_io)
412 		ret = sr->done_io;
413 	io_req_set_res(req, ret, 0);
414 	return IOU_OK;
415 }
416 
417 static bool io_recvmsg_multishot_overflow(struct io_async_msghdr *iomsg)
418 {
419 	int hdr;
420 
421 	if (iomsg->namelen < 0)
422 		return true;
423 	if (check_add_overflow((int)sizeof(struct io_uring_recvmsg_out),
424 			       iomsg->namelen, &hdr))
425 		return true;
426 	if (check_add_overflow(hdr, (int)iomsg->controllen, &hdr))
427 		return true;
428 
429 	return false;
430 }
431 
432 static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
433 				 struct io_async_msghdr *iomsg)
434 {
435 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
436 	struct user_msghdr msg;
437 	int ret;
438 
439 	if (copy_from_user(&msg, sr->umsg, sizeof(*sr->umsg)))
440 		return -EFAULT;
441 
442 	ret = __copy_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
443 	if (ret)
444 		return ret;
445 
446 	if (req->flags & REQ_F_BUFFER_SELECT) {
447 		if (msg.msg_iovlen == 0) {
448 			sr->len = iomsg->fast_iov[0].iov_len = 0;
449 			iomsg->fast_iov[0].iov_base = NULL;
450 			iomsg->free_iov = NULL;
451 		} else if (msg.msg_iovlen > 1) {
452 			return -EINVAL;
453 		} else {
454 			if (copy_from_user(iomsg->fast_iov, msg.msg_iov, sizeof(*msg.msg_iov)))
455 				return -EFAULT;
456 			sr->len = iomsg->fast_iov[0].iov_len;
457 			iomsg->free_iov = NULL;
458 		}
459 
460 		if (req->flags & REQ_F_APOLL_MULTISHOT) {
461 			iomsg->namelen = msg.msg_namelen;
462 			iomsg->controllen = msg.msg_controllen;
463 			if (io_recvmsg_multishot_overflow(iomsg))
464 				return -EOVERFLOW;
465 		}
466 	} else {
467 		iomsg->free_iov = iomsg->fast_iov;
468 		ret = __import_iovec(ITER_DEST, msg.msg_iov, msg.msg_iovlen, UIO_FASTIOV,
469 				     &iomsg->free_iov, &iomsg->msg.msg_iter,
470 				     false);
471 		if (ret > 0)
472 			ret = 0;
473 	}
474 
475 	return ret;
476 }
477 
478 #ifdef CONFIG_COMPAT
479 static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
480 					struct io_async_msghdr *iomsg)
481 {
482 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
483 	struct compat_msghdr msg;
484 	struct compat_iovec __user *uiov;
485 	int ret;
486 
487 	if (copy_from_user(&msg, sr->umsg_compat, sizeof(msg)))
488 		return -EFAULT;
489 
490 	ret = __get_compat_msghdr(&iomsg->msg, &msg, &iomsg->uaddr);
491 	if (ret)
492 		return ret;
493 
494 	uiov = compat_ptr(msg.msg_iov);
495 	if (req->flags & REQ_F_BUFFER_SELECT) {
496 		compat_ssize_t clen;
497 
498 		iomsg->free_iov = NULL;
499 		if (msg.msg_iovlen == 0) {
500 			sr->len = 0;
501 		} else if (msg.msg_iovlen > 1) {
502 			return -EINVAL;
503 		} else {
504 			if (!access_ok(uiov, sizeof(*uiov)))
505 				return -EFAULT;
506 			if (__get_user(clen, &uiov->iov_len))
507 				return -EFAULT;
508 			if (clen < 0)
509 				return -EINVAL;
510 			sr->len = clen;
511 		}
512 
513 		if (req->flags & REQ_F_APOLL_MULTISHOT) {
514 			iomsg->namelen = msg.msg_namelen;
515 			iomsg->controllen = msg.msg_controllen;
516 			if (io_recvmsg_multishot_overflow(iomsg))
517 				return -EOVERFLOW;
518 		}
519 	} else {
520 		iomsg->free_iov = iomsg->fast_iov;
521 		ret = __import_iovec(ITER_DEST, (struct iovec __user *)uiov, msg.msg_iovlen,
522 				   UIO_FASTIOV, &iomsg->free_iov,
523 				   &iomsg->msg.msg_iter, true);
524 		if (ret < 0)
525 			return ret;
526 	}
527 
528 	return 0;
529 }
530 #endif
531 
532 static int io_recvmsg_copy_hdr(struct io_kiocb *req,
533 			       struct io_async_msghdr *iomsg)
534 {
535 	iomsg->msg.msg_name = &iomsg->addr;
536 
537 #ifdef CONFIG_COMPAT
538 	if (req->ctx->compat)
539 		return __io_compat_recvmsg_copy_hdr(req, iomsg);
540 #endif
541 
542 	return __io_recvmsg_copy_hdr(req, iomsg);
543 }
544 
545 int io_recvmsg_prep_async(struct io_kiocb *req)
546 {
547 	int ret;
548 
549 	if (!io_msg_alloc_async_prep(req))
550 		return -ENOMEM;
551 	ret = io_recvmsg_copy_hdr(req, req->async_data);
552 	if (!ret)
553 		req->flags |= REQ_F_NEED_CLEANUP;
554 	return ret;
555 }
556 
557 #define RECVMSG_FLAGS (IORING_RECVSEND_POLL_FIRST | IORING_RECV_MULTISHOT)
558 
559 int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
560 {
561 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
562 
563 	if (unlikely(sqe->file_index || sqe->addr2))
564 		return -EINVAL;
565 
566 	sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
567 	sr->len = READ_ONCE(sqe->len);
568 	sr->flags = READ_ONCE(sqe->ioprio);
569 	if (sr->flags & ~(RECVMSG_FLAGS))
570 		return -EINVAL;
571 	sr->msg_flags = READ_ONCE(sqe->msg_flags);
572 	if (sr->msg_flags & MSG_DONTWAIT)
573 		req->flags |= REQ_F_NOWAIT;
574 	if (sr->msg_flags & MSG_ERRQUEUE)
575 		req->flags |= REQ_F_CLEAR_POLLIN;
576 	if (sr->flags & IORING_RECV_MULTISHOT) {
577 		if (!(req->flags & REQ_F_BUFFER_SELECT))
578 			return -EINVAL;
579 		if (sr->msg_flags & MSG_WAITALL)
580 			return -EINVAL;
581 		if (req->opcode == IORING_OP_RECV && sr->len)
582 			return -EINVAL;
583 		req->flags |= REQ_F_APOLL_MULTISHOT;
584 		/*
585 		 * Store the buffer group for this multishot receive separately,
586 		 * as if we end up doing an io-wq based issue that selects a
587 		 * buffer, it has to be committed immediately and that will
588 		 * clear ->buf_list. This means we lose the link to the buffer
589 		 * list, and the eventual buffer put on completion then cannot
590 		 * restore it.
591 		 */
592 		sr->buf_group = req->buf_index;
593 	}
594 
595 #ifdef CONFIG_COMPAT
596 	if (req->ctx->compat)
597 		sr->msg_flags |= MSG_CMSG_COMPAT;
598 #endif
599 	sr->done_io = 0;
600 	return 0;
601 }
602 
603 static inline void io_recv_prep_retry(struct io_kiocb *req)
604 {
605 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
606 
607 	sr->done_io = 0;
608 	sr->len = 0; /* get from the provided buffer */
609 	req->buf_index = sr->buf_group;
610 }
611 
612 /*
613  * Finishes io_recv and io_recvmsg.
614  *
615  * Returns true if it is actually finished, or false if it should run
616  * again (for multishot).
617  */
618 static inline bool io_recv_finish(struct io_kiocb *req, int *ret,
619 				  unsigned int cflags, bool mshot_finished,
620 				  unsigned issue_flags)
621 {
622 	if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
623 		io_req_set_res(req, *ret, cflags);
624 		*ret = IOU_OK;
625 		return true;
626 	}
627 
628 	if (!mshot_finished) {
629 		if (io_aux_cqe(req->ctx, issue_flags & IO_URING_F_COMPLETE_DEFER,
630 			       req->cqe.user_data, *ret, cflags | IORING_CQE_F_MORE, true)) {
631 			io_recv_prep_retry(req);
632 			return false;
633 		}
634 		/* Otherwise stop multishot but use the current result. */
635 	}
636 
637 	io_req_set_res(req, *ret, cflags);
638 
639 	if (issue_flags & IO_URING_F_MULTISHOT)
640 		*ret = IOU_STOP_MULTISHOT;
641 	else
642 		*ret = IOU_OK;
643 	return true;
644 }
645 
646 static int io_recvmsg_prep_multishot(struct io_async_msghdr *kmsg,
647 				     struct io_sr_msg *sr, void __user **buf,
648 				     size_t *len)
649 {
650 	unsigned long ubuf = (unsigned long) *buf;
651 	unsigned long hdr;
652 
653 	hdr = sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
654 		kmsg->controllen;
655 	if (*len < hdr)
656 		return -EFAULT;
657 
658 	if (kmsg->controllen) {
659 		unsigned long control = ubuf + hdr - kmsg->controllen;
660 
661 		kmsg->msg.msg_control_user = (void __user *) control;
662 		kmsg->msg.msg_controllen = kmsg->controllen;
663 	}
664 
665 	sr->buf = *buf; /* stash for later copy */
666 	*buf = (void __user *) (ubuf + hdr);
667 	kmsg->payloadlen = *len = *len - hdr;
668 	return 0;
669 }
670 
671 struct io_recvmsg_multishot_hdr {
672 	struct io_uring_recvmsg_out msg;
673 	struct sockaddr_storage addr;
674 };
675 
676 static int io_recvmsg_multishot(struct socket *sock, struct io_sr_msg *io,
677 				struct io_async_msghdr *kmsg,
678 				unsigned int flags, bool *finished)
679 {
680 	int err;
681 	int copy_len;
682 	struct io_recvmsg_multishot_hdr hdr;
683 
684 	if (kmsg->namelen)
685 		kmsg->msg.msg_name = &hdr.addr;
686 	kmsg->msg.msg_flags = flags & (MSG_CMSG_CLOEXEC|MSG_CMSG_COMPAT);
687 	kmsg->msg.msg_namelen = 0;
688 
689 	if (sock->file->f_flags & O_NONBLOCK)
690 		flags |= MSG_DONTWAIT;
691 
692 	err = sock_recvmsg(sock, &kmsg->msg, flags);
693 	*finished = err <= 0;
694 	if (err < 0)
695 		return err;
696 
697 	hdr.msg = (struct io_uring_recvmsg_out) {
698 		.controllen = kmsg->controllen - kmsg->msg.msg_controllen,
699 		.flags = kmsg->msg.msg_flags & ~MSG_CMSG_COMPAT
700 	};
701 
702 	hdr.msg.payloadlen = err;
703 	if (err > kmsg->payloadlen)
704 		err = kmsg->payloadlen;
705 
706 	copy_len = sizeof(struct io_uring_recvmsg_out);
707 	if (kmsg->msg.msg_namelen > kmsg->namelen)
708 		copy_len += kmsg->namelen;
709 	else
710 		copy_len += kmsg->msg.msg_namelen;
711 
712 	/*
713 	 *      "fromlen shall refer to the value before truncation.."
714 	 *                      1003.1g
715 	 */
716 	hdr.msg.namelen = kmsg->msg.msg_namelen;
717 
718 	/* ensure that there is no gap between hdr and sockaddr_storage */
719 	BUILD_BUG_ON(offsetof(struct io_recvmsg_multishot_hdr, addr) !=
720 		     sizeof(struct io_uring_recvmsg_out));
721 	if (copy_to_user(io->buf, &hdr, copy_len)) {
722 		*finished = true;
723 		return -EFAULT;
724 	}
725 
726 	return sizeof(struct io_uring_recvmsg_out) + kmsg->namelen +
727 			kmsg->controllen + err;
728 }
729 
730 int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
731 {
732 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
733 	struct io_async_msghdr iomsg, *kmsg;
734 	struct socket *sock;
735 	unsigned int cflags;
736 	unsigned flags;
737 	int ret, min_ret = 0;
738 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
739 	bool mshot_finished = true;
740 
741 	sock = sock_from_file(req->file);
742 	if (unlikely(!sock))
743 		return -ENOTSOCK;
744 
745 	if (req_has_async_data(req)) {
746 		kmsg = req->async_data;
747 	} else {
748 		ret = io_recvmsg_copy_hdr(req, &iomsg);
749 		if (ret)
750 			return ret;
751 		kmsg = &iomsg;
752 	}
753 
754 	if (!(req->flags & REQ_F_POLLED) &&
755 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
756 		return io_setup_async_msg(req, kmsg, issue_flags);
757 
758 	if (!io_check_multishot(req, issue_flags))
759 		return io_setup_async_msg(req, kmsg, issue_flags);
760 
761 retry_multishot:
762 	if (io_do_buffer_select(req)) {
763 		void __user *buf;
764 		size_t len = sr->len;
765 
766 		buf = io_buffer_select(req, &len, issue_flags);
767 		if (!buf)
768 			return -ENOBUFS;
769 
770 		if (req->flags & REQ_F_APOLL_MULTISHOT) {
771 			ret = io_recvmsg_prep_multishot(kmsg, sr, &buf, &len);
772 			if (ret) {
773 				io_kbuf_recycle(req, issue_flags);
774 				return ret;
775 			}
776 		}
777 
778 		iov_iter_ubuf(&kmsg->msg.msg_iter, ITER_DEST, buf, len);
779 	}
780 
781 	flags = sr->msg_flags;
782 	if (force_nonblock)
783 		flags |= MSG_DONTWAIT;
784 	if (flags & MSG_WAITALL)
785 		min_ret = iov_iter_count(&kmsg->msg.msg_iter);
786 
787 	kmsg->msg.msg_get_inq = 1;
788 	if (req->flags & REQ_F_APOLL_MULTISHOT)
789 		ret = io_recvmsg_multishot(sock, sr, kmsg, flags,
790 					   &mshot_finished);
791 	else
792 		ret = __sys_recvmsg_sock(sock, &kmsg->msg, sr->umsg,
793 					 kmsg->uaddr, flags);
794 
795 	if (ret < min_ret) {
796 		if (ret == -EAGAIN && force_nonblock) {
797 			ret = io_setup_async_msg(req, kmsg, issue_flags);
798 			if (ret == -EAGAIN && (issue_flags & IO_URING_F_MULTISHOT)) {
799 				io_kbuf_recycle(req, issue_flags);
800 				return IOU_ISSUE_SKIP_COMPLETE;
801 			}
802 			return ret;
803 		}
804 		if (ret > 0 && io_net_retry(sock, flags)) {
805 			sr->done_io += ret;
806 			req->flags |= REQ_F_PARTIAL_IO;
807 			return io_setup_async_msg(req, kmsg, issue_flags);
808 		}
809 		if (ret == -ERESTARTSYS)
810 			ret = -EINTR;
811 		req_set_fail(req);
812 	} else if ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
813 		req_set_fail(req);
814 	}
815 
816 	if (ret > 0)
817 		ret += sr->done_io;
818 	else if (sr->done_io)
819 		ret = sr->done_io;
820 	else
821 		io_kbuf_recycle(req, issue_flags);
822 
823 	cflags = io_put_kbuf(req, issue_flags);
824 	if (kmsg->msg.msg_inq)
825 		cflags |= IORING_CQE_F_SOCK_NONEMPTY;
826 
827 	if (!io_recv_finish(req, &ret, cflags, mshot_finished, issue_flags))
828 		goto retry_multishot;
829 
830 	if (mshot_finished) {
831 		/* fast path, check for non-NULL to avoid function call */
832 		if (kmsg->free_iov)
833 			kfree(kmsg->free_iov);
834 		io_netmsg_recycle(req, issue_flags);
835 		req->flags &= ~REQ_F_NEED_CLEANUP;
836 	}
837 
838 	return ret;
839 }
840 
841 int io_recv(struct io_kiocb *req, unsigned int issue_flags)
842 {
843 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
844 	struct msghdr msg;
845 	struct socket *sock;
846 	unsigned int cflags;
847 	unsigned flags;
848 	int ret, min_ret = 0;
849 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
850 	size_t len = sr->len;
851 
852 	if (!(req->flags & REQ_F_POLLED) &&
853 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
854 		return -EAGAIN;
855 
856 	if (!io_check_multishot(req, issue_flags))
857 		return -EAGAIN;
858 
859 	sock = sock_from_file(req->file);
860 	if (unlikely(!sock))
861 		return -ENOTSOCK;
862 
863 retry_multishot:
864 	if (io_do_buffer_select(req)) {
865 		void __user *buf;
866 
867 		buf = io_buffer_select(req, &len, issue_flags);
868 		if (!buf)
869 			return -ENOBUFS;
870 		sr->buf = buf;
871 	}
872 
873 	ret = import_ubuf(ITER_DEST, sr->buf, len, &msg.msg_iter);
874 	if (unlikely(ret))
875 		goto out_free;
876 
877 	msg.msg_name = NULL;
878 	msg.msg_namelen = 0;
879 	msg.msg_control = NULL;
880 	msg.msg_get_inq = 1;
881 	msg.msg_flags = 0;
882 	msg.msg_controllen = 0;
883 	msg.msg_iocb = NULL;
884 	msg.msg_ubuf = NULL;
885 
886 	flags = sr->msg_flags;
887 	if (force_nonblock)
888 		flags |= MSG_DONTWAIT;
889 	if (flags & MSG_WAITALL)
890 		min_ret = iov_iter_count(&msg.msg_iter);
891 
892 	ret = sock_recvmsg(sock, &msg, flags);
893 	if (ret < min_ret) {
894 		if (ret == -EAGAIN && force_nonblock) {
895 			if (issue_flags & IO_URING_F_MULTISHOT) {
896 				io_kbuf_recycle(req, issue_flags);
897 				return IOU_ISSUE_SKIP_COMPLETE;
898 			}
899 
900 			return -EAGAIN;
901 		}
902 		if (ret > 0 && io_net_retry(sock, flags)) {
903 			sr->len -= ret;
904 			sr->buf += ret;
905 			sr->done_io += ret;
906 			req->flags |= REQ_F_PARTIAL_IO;
907 			return -EAGAIN;
908 		}
909 		if (ret == -ERESTARTSYS)
910 			ret = -EINTR;
911 		req_set_fail(req);
912 	} else if ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))) {
913 out_free:
914 		req_set_fail(req);
915 	}
916 
917 	if (ret > 0)
918 		ret += sr->done_io;
919 	else if (sr->done_io)
920 		ret = sr->done_io;
921 	else
922 		io_kbuf_recycle(req, issue_flags);
923 
924 	cflags = io_put_kbuf(req, issue_flags);
925 	if (msg.msg_inq)
926 		cflags |= IORING_CQE_F_SOCK_NONEMPTY;
927 
928 	if (!io_recv_finish(req, &ret, cflags, ret <= 0, issue_flags))
929 		goto retry_multishot;
930 
931 	return ret;
932 }
933 
934 void io_send_zc_cleanup(struct io_kiocb *req)
935 {
936 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
937 	struct io_async_msghdr *io;
938 
939 	if (req_has_async_data(req)) {
940 		io = req->async_data;
941 		/* might be ->fast_iov if *msg_copy_hdr failed */
942 		if (io->free_iov != io->fast_iov)
943 			kfree(io->free_iov);
944 	}
945 	if (zc->notif) {
946 		io_notif_flush(zc->notif);
947 		zc->notif = NULL;
948 	}
949 }
950 
951 #define IO_ZC_FLAGS_COMMON (IORING_RECVSEND_POLL_FIRST | IORING_RECVSEND_FIXED_BUF)
952 #define IO_ZC_FLAGS_VALID  (IO_ZC_FLAGS_COMMON | IORING_SEND_ZC_REPORT_USAGE)
953 
954 int io_send_zc_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
955 {
956 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
957 	struct io_ring_ctx *ctx = req->ctx;
958 	struct io_kiocb *notif;
959 
960 	if (unlikely(READ_ONCE(sqe->__pad2[0]) || READ_ONCE(sqe->addr3)))
961 		return -EINVAL;
962 	/* we don't support IOSQE_CQE_SKIP_SUCCESS just yet */
963 	if (req->flags & REQ_F_CQE_SKIP)
964 		return -EINVAL;
965 
966 	notif = zc->notif = io_alloc_notif(ctx);
967 	if (!notif)
968 		return -ENOMEM;
969 	notif->cqe.user_data = req->cqe.user_data;
970 	notif->cqe.res = 0;
971 	notif->cqe.flags = IORING_CQE_F_NOTIF;
972 	req->flags |= REQ_F_NEED_CLEANUP;
973 
974 	zc->flags = READ_ONCE(sqe->ioprio);
975 	if (unlikely(zc->flags & ~IO_ZC_FLAGS_COMMON)) {
976 		if (zc->flags & ~IO_ZC_FLAGS_VALID)
977 			return -EINVAL;
978 		if (zc->flags & IORING_SEND_ZC_REPORT_USAGE) {
979 			io_notif_set_extended(notif);
980 			io_notif_to_data(notif)->zc_report = true;
981 		}
982 	}
983 
984 	if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
985 		unsigned idx = READ_ONCE(sqe->buf_index);
986 
987 		if (unlikely(idx >= ctx->nr_user_bufs))
988 			return -EFAULT;
989 		idx = array_index_nospec(idx, ctx->nr_user_bufs);
990 		req->imu = READ_ONCE(ctx->user_bufs[idx]);
991 		io_req_set_rsrc_node(notif, ctx, 0);
992 	}
993 
994 	if (req->opcode == IORING_OP_SEND_ZC) {
995 		if (READ_ONCE(sqe->__pad3[0]))
996 			return -EINVAL;
997 		zc->addr = u64_to_user_ptr(READ_ONCE(sqe->addr2));
998 		zc->addr_len = READ_ONCE(sqe->addr_len);
999 	} else {
1000 		if (unlikely(sqe->addr2 || sqe->file_index))
1001 			return -EINVAL;
1002 		if (unlikely(zc->flags & IORING_RECVSEND_FIXED_BUF))
1003 			return -EINVAL;
1004 	}
1005 
1006 	zc->buf = u64_to_user_ptr(READ_ONCE(sqe->addr));
1007 	zc->len = READ_ONCE(sqe->len);
1008 	zc->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
1009 	if (zc->msg_flags & MSG_DONTWAIT)
1010 		req->flags |= REQ_F_NOWAIT;
1011 
1012 	zc->done_io = 0;
1013 
1014 #ifdef CONFIG_COMPAT
1015 	if (req->ctx->compat)
1016 		zc->msg_flags |= MSG_CMSG_COMPAT;
1017 #endif
1018 	return 0;
1019 }
1020 
1021 static int io_sg_from_iter_iovec(struct sock *sk, struct sk_buff *skb,
1022 				 struct iov_iter *from, size_t length)
1023 {
1024 	skb_zcopy_downgrade_managed(skb);
1025 	return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1026 }
1027 
1028 static int io_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1029 			   struct iov_iter *from, size_t length)
1030 {
1031 	struct skb_shared_info *shinfo = skb_shinfo(skb);
1032 	int frag = shinfo->nr_frags;
1033 	int ret = 0;
1034 	struct bvec_iter bi;
1035 	ssize_t copied = 0;
1036 	unsigned long truesize = 0;
1037 
1038 	if (!frag)
1039 		shinfo->flags |= SKBFL_MANAGED_FRAG_REFS;
1040 	else if (unlikely(!skb_zcopy_managed(skb)))
1041 		return __zerocopy_sg_from_iter(NULL, sk, skb, from, length);
1042 
1043 	bi.bi_size = min(from->count, length);
1044 	bi.bi_bvec_done = from->iov_offset;
1045 	bi.bi_idx = 0;
1046 
1047 	while (bi.bi_size && frag < MAX_SKB_FRAGS) {
1048 		struct bio_vec v = mp_bvec_iter_bvec(from->bvec, bi);
1049 
1050 		copied += v.bv_len;
1051 		truesize += PAGE_ALIGN(v.bv_len + v.bv_offset);
1052 		__skb_fill_page_desc_noacc(shinfo, frag++, v.bv_page,
1053 					   v.bv_offset, v.bv_len);
1054 		bvec_iter_advance_single(from->bvec, &bi, v.bv_len);
1055 	}
1056 	if (bi.bi_size)
1057 		ret = -EMSGSIZE;
1058 
1059 	shinfo->nr_frags = frag;
1060 	from->bvec += bi.bi_idx;
1061 	from->nr_segs -= bi.bi_idx;
1062 	from->count -= copied;
1063 	from->iov_offset = bi.bi_bvec_done;
1064 
1065 	skb->data_len += copied;
1066 	skb->len += copied;
1067 	skb->truesize += truesize;
1068 
1069 	if (sk && sk->sk_type == SOCK_STREAM) {
1070 		sk_wmem_queued_add(sk, truesize);
1071 		if (!skb_zcopy_pure(skb))
1072 			sk_mem_charge(sk, truesize);
1073 	} else {
1074 		refcount_add(truesize, &skb->sk->sk_wmem_alloc);
1075 	}
1076 	return ret;
1077 }
1078 
1079 int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
1080 {
1081 	struct sockaddr_storage __address;
1082 	struct io_sr_msg *zc = io_kiocb_to_cmd(req, struct io_sr_msg);
1083 	struct msghdr msg;
1084 	struct socket *sock;
1085 	unsigned msg_flags;
1086 	int ret, min_ret = 0;
1087 
1088 	sock = sock_from_file(req->file);
1089 	if (unlikely(!sock))
1090 		return -ENOTSOCK;
1091 	if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1092 		return -EOPNOTSUPP;
1093 
1094 	msg.msg_name = NULL;
1095 	msg.msg_control = NULL;
1096 	msg.msg_controllen = 0;
1097 	msg.msg_namelen = 0;
1098 
1099 	if (zc->addr) {
1100 		if (req_has_async_data(req)) {
1101 			struct io_async_msghdr *io = req->async_data;
1102 
1103 			msg.msg_name = &io->addr;
1104 		} else {
1105 			ret = move_addr_to_kernel(zc->addr, zc->addr_len, &__address);
1106 			if (unlikely(ret < 0))
1107 				return ret;
1108 			msg.msg_name = (struct sockaddr *)&__address;
1109 		}
1110 		msg.msg_namelen = zc->addr_len;
1111 	}
1112 
1113 	if (!(req->flags & REQ_F_POLLED) &&
1114 	    (zc->flags & IORING_RECVSEND_POLL_FIRST))
1115 		return io_setup_async_addr(req, &__address, issue_flags);
1116 
1117 	if (zc->flags & IORING_RECVSEND_FIXED_BUF) {
1118 		ret = io_import_fixed(ITER_SOURCE, &msg.msg_iter, req->imu,
1119 					(u64)(uintptr_t)zc->buf, zc->len);
1120 		if (unlikely(ret))
1121 			return ret;
1122 		msg.sg_from_iter = io_sg_from_iter;
1123 	} else {
1124 		io_notif_set_extended(zc->notif);
1125 		ret = import_ubuf(ITER_SOURCE, zc->buf, zc->len, &msg.msg_iter);
1126 		if (unlikely(ret))
1127 			return ret;
1128 		ret = io_notif_account_mem(zc->notif, zc->len);
1129 		if (unlikely(ret))
1130 			return ret;
1131 		msg.sg_from_iter = io_sg_from_iter_iovec;
1132 	}
1133 
1134 	msg_flags = zc->msg_flags | MSG_ZEROCOPY;
1135 	if (issue_flags & IO_URING_F_NONBLOCK)
1136 		msg_flags |= MSG_DONTWAIT;
1137 	if (msg_flags & MSG_WAITALL)
1138 		min_ret = iov_iter_count(&msg.msg_iter);
1139 
1140 	msg.msg_flags = msg_flags;
1141 	msg.msg_ubuf = &io_notif_to_data(zc->notif)->uarg;
1142 	ret = sock_sendmsg(sock, &msg);
1143 
1144 	if (unlikely(ret < min_ret)) {
1145 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1146 			return io_setup_async_addr(req, &__address, issue_flags);
1147 
1148 		if (ret > 0 && io_net_retry(sock, msg.msg_flags)) {
1149 			zc->len -= ret;
1150 			zc->buf += ret;
1151 			zc->done_io += ret;
1152 			req->flags |= REQ_F_PARTIAL_IO;
1153 			return io_setup_async_addr(req, &__address, issue_flags);
1154 		}
1155 		if (ret == -ERESTARTSYS)
1156 			ret = -EINTR;
1157 		req_set_fail(req);
1158 	}
1159 
1160 	if (ret >= 0)
1161 		ret += zc->done_io;
1162 	else if (zc->done_io)
1163 		ret = zc->done_io;
1164 
1165 	/*
1166 	 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1167 	 * flushing notif to io_send_zc_cleanup()
1168 	 */
1169 	if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1170 		io_notif_flush(zc->notif);
1171 		req->flags &= ~REQ_F_NEED_CLEANUP;
1172 	}
1173 	io_req_set_res(req, ret, IORING_CQE_F_MORE);
1174 	return IOU_OK;
1175 }
1176 
1177 int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
1178 {
1179 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1180 	struct io_async_msghdr iomsg, *kmsg;
1181 	struct socket *sock;
1182 	unsigned flags;
1183 	int ret, min_ret = 0;
1184 
1185 	io_notif_set_extended(sr->notif);
1186 
1187 	sock = sock_from_file(req->file);
1188 	if (unlikely(!sock))
1189 		return -ENOTSOCK;
1190 	if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
1191 		return -EOPNOTSUPP;
1192 
1193 	if (req_has_async_data(req)) {
1194 		kmsg = req->async_data;
1195 	} else {
1196 		ret = io_sendmsg_copy_hdr(req, &iomsg);
1197 		if (ret)
1198 			return ret;
1199 		kmsg = &iomsg;
1200 	}
1201 
1202 	if (!(req->flags & REQ_F_POLLED) &&
1203 	    (sr->flags & IORING_RECVSEND_POLL_FIRST))
1204 		return io_setup_async_msg(req, kmsg, issue_flags);
1205 
1206 	flags = sr->msg_flags | MSG_ZEROCOPY;
1207 	if (issue_flags & IO_URING_F_NONBLOCK)
1208 		flags |= MSG_DONTWAIT;
1209 	if (flags & MSG_WAITALL)
1210 		min_ret = iov_iter_count(&kmsg->msg.msg_iter);
1211 
1212 	kmsg->msg.msg_ubuf = &io_notif_to_data(sr->notif)->uarg;
1213 	kmsg->msg.sg_from_iter = io_sg_from_iter_iovec;
1214 	ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
1215 
1216 	if (unlikely(ret < min_ret)) {
1217 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1218 			return io_setup_async_msg(req, kmsg, issue_flags);
1219 
1220 		if (ret > 0 && io_net_retry(sock, flags)) {
1221 			sr->done_io += ret;
1222 			req->flags |= REQ_F_PARTIAL_IO;
1223 			return io_setup_async_msg(req, kmsg, issue_flags);
1224 		}
1225 		if (ret == -ERESTARTSYS)
1226 			ret = -EINTR;
1227 		req_set_fail(req);
1228 	}
1229 	/* fast path, check for non-NULL to avoid function call */
1230 	if (kmsg->free_iov) {
1231 		kfree(kmsg->free_iov);
1232 		kmsg->free_iov = NULL;
1233 	}
1234 
1235 	io_netmsg_recycle(req, issue_flags);
1236 	if (ret >= 0)
1237 		ret += sr->done_io;
1238 	else if (sr->done_io)
1239 		ret = sr->done_io;
1240 
1241 	/*
1242 	 * If we're in io-wq we can't rely on tw ordering guarantees, defer
1243 	 * flushing notif to io_send_zc_cleanup()
1244 	 */
1245 	if (!(issue_flags & IO_URING_F_UNLOCKED)) {
1246 		io_notif_flush(sr->notif);
1247 		req->flags &= ~REQ_F_NEED_CLEANUP;
1248 	}
1249 	io_req_set_res(req, ret, IORING_CQE_F_MORE);
1250 	return IOU_OK;
1251 }
1252 
1253 void io_sendrecv_fail(struct io_kiocb *req)
1254 {
1255 	struct io_sr_msg *sr = io_kiocb_to_cmd(req, struct io_sr_msg);
1256 
1257 	if (req->flags & REQ_F_PARTIAL_IO)
1258 		req->cqe.res = sr->done_io;
1259 
1260 	if ((req->flags & REQ_F_NEED_CLEANUP) &&
1261 	    (req->opcode == IORING_OP_SEND_ZC || req->opcode == IORING_OP_SENDMSG_ZC))
1262 		req->cqe.flags |= IORING_CQE_F_MORE;
1263 }
1264 
1265 int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1266 {
1267 	struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1268 	unsigned flags;
1269 
1270 	if (sqe->len || sqe->buf_index)
1271 		return -EINVAL;
1272 
1273 	accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1274 	accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
1275 	accept->flags = READ_ONCE(sqe->accept_flags);
1276 	accept->nofile = rlimit(RLIMIT_NOFILE);
1277 	flags = READ_ONCE(sqe->ioprio);
1278 	if (flags & ~IORING_ACCEPT_MULTISHOT)
1279 		return -EINVAL;
1280 
1281 	accept->file_slot = READ_ONCE(sqe->file_index);
1282 	if (accept->file_slot) {
1283 		if (accept->flags & SOCK_CLOEXEC)
1284 			return -EINVAL;
1285 		if (flags & IORING_ACCEPT_MULTISHOT &&
1286 		    accept->file_slot != IORING_FILE_INDEX_ALLOC)
1287 			return -EINVAL;
1288 	}
1289 	if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1290 		return -EINVAL;
1291 	if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
1292 		accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
1293 	if (flags & IORING_ACCEPT_MULTISHOT)
1294 		req->flags |= REQ_F_APOLL_MULTISHOT;
1295 	return 0;
1296 }
1297 
1298 int io_accept(struct io_kiocb *req, unsigned int issue_flags)
1299 {
1300 	struct io_ring_ctx *ctx = req->ctx;
1301 	struct io_accept *accept = io_kiocb_to_cmd(req, struct io_accept);
1302 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1303 	unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
1304 	bool fixed = !!accept->file_slot;
1305 	struct file *file;
1306 	int ret, fd;
1307 
1308 	if (!io_check_multishot(req, issue_flags))
1309 		return -EAGAIN;
1310 retry:
1311 	if (!fixed) {
1312 		fd = __get_unused_fd_flags(accept->flags, accept->nofile);
1313 		if (unlikely(fd < 0))
1314 			return fd;
1315 	}
1316 	file = do_accept(req->file, file_flags, accept->addr, accept->addr_len,
1317 			 accept->flags);
1318 	if (IS_ERR(file)) {
1319 		if (!fixed)
1320 			put_unused_fd(fd);
1321 		ret = PTR_ERR(file);
1322 		if (ret == -EAGAIN && force_nonblock) {
1323 			/*
1324 			 * if it's multishot and polled, we don't need to
1325 			 * return EAGAIN to arm the poll infra since it
1326 			 * has already been done
1327 			 */
1328 			if (issue_flags & IO_URING_F_MULTISHOT)
1329 				ret = IOU_ISSUE_SKIP_COMPLETE;
1330 			return ret;
1331 		}
1332 		if (ret == -ERESTARTSYS)
1333 			ret = -EINTR;
1334 		req_set_fail(req);
1335 	} else if (!fixed) {
1336 		fd_install(fd, file);
1337 		ret = fd;
1338 	} else {
1339 		ret = io_fixed_fd_install(req, issue_flags, file,
1340 						accept->file_slot);
1341 	}
1342 
1343 	if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
1344 		io_req_set_res(req, ret, 0);
1345 		return IOU_OK;
1346 	}
1347 
1348 	if (ret < 0)
1349 		return ret;
1350 	if (io_aux_cqe(ctx, issue_flags & IO_URING_F_COMPLETE_DEFER,
1351 		       req->cqe.user_data, ret, IORING_CQE_F_MORE, true))
1352 		goto retry;
1353 
1354 	return -ECANCELED;
1355 }
1356 
1357 int io_socket_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1358 {
1359 	struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1360 
1361 	if (sqe->addr || sqe->rw_flags || sqe->buf_index)
1362 		return -EINVAL;
1363 
1364 	sock->domain = READ_ONCE(sqe->fd);
1365 	sock->type = READ_ONCE(sqe->off);
1366 	sock->protocol = READ_ONCE(sqe->len);
1367 	sock->file_slot = READ_ONCE(sqe->file_index);
1368 	sock->nofile = rlimit(RLIMIT_NOFILE);
1369 
1370 	sock->flags = sock->type & ~SOCK_TYPE_MASK;
1371 	if (sock->file_slot && (sock->flags & SOCK_CLOEXEC))
1372 		return -EINVAL;
1373 	if (sock->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
1374 		return -EINVAL;
1375 	return 0;
1376 }
1377 
1378 int io_socket(struct io_kiocb *req, unsigned int issue_flags)
1379 {
1380 	struct io_socket *sock = io_kiocb_to_cmd(req, struct io_socket);
1381 	bool fixed = !!sock->file_slot;
1382 	struct file *file;
1383 	int ret, fd;
1384 
1385 	if (!fixed) {
1386 		fd = __get_unused_fd_flags(sock->flags, sock->nofile);
1387 		if (unlikely(fd < 0))
1388 			return fd;
1389 	}
1390 	file = __sys_socket_file(sock->domain, sock->type, sock->protocol);
1391 	if (IS_ERR(file)) {
1392 		if (!fixed)
1393 			put_unused_fd(fd);
1394 		ret = PTR_ERR(file);
1395 		if (ret == -EAGAIN && (issue_flags & IO_URING_F_NONBLOCK))
1396 			return -EAGAIN;
1397 		if (ret == -ERESTARTSYS)
1398 			ret = -EINTR;
1399 		req_set_fail(req);
1400 	} else if (!fixed) {
1401 		fd_install(fd, file);
1402 		ret = fd;
1403 	} else {
1404 		ret = io_fixed_fd_install(req, issue_flags, file,
1405 					    sock->file_slot);
1406 	}
1407 	io_req_set_res(req, ret, 0);
1408 	return IOU_OK;
1409 }
1410 
1411 int io_connect_prep_async(struct io_kiocb *req)
1412 {
1413 	struct io_async_connect *io = req->async_data;
1414 	struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1415 
1416 	return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
1417 }
1418 
1419 int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
1420 {
1421 	struct io_connect *conn = io_kiocb_to_cmd(req, struct io_connect);
1422 
1423 	if (sqe->len || sqe->buf_index || sqe->rw_flags || sqe->splice_fd_in)
1424 		return -EINVAL;
1425 
1426 	conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
1427 	conn->addr_len =  READ_ONCE(sqe->addr2);
1428 	conn->in_progress = conn->seen_econnaborted = false;
1429 	return 0;
1430 }
1431 
1432 int io_connect(struct io_kiocb *req, unsigned int issue_flags)
1433 {
1434 	struct io_connect *connect = io_kiocb_to_cmd(req, struct io_connect);
1435 	struct io_async_connect __io, *io;
1436 	unsigned file_flags;
1437 	int ret;
1438 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
1439 
1440 	if (connect->in_progress) {
1441 		struct socket *socket;
1442 
1443 		ret = -ENOTSOCK;
1444 		socket = sock_from_file(req->file);
1445 		if (socket)
1446 			ret = sock_error(socket->sk);
1447 		goto out;
1448 	}
1449 
1450 	if (req_has_async_data(req)) {
1451 		io = req->async_data;
1452 	} else {
1453 		ret = move_addr_to_kernel(connect->addr,
1454 						connect->addr_len,
1455 						&__io.address);
1456 		if (ret)
1457 			goto out;
1458 		io = &__io;
1459 	}
1460 
1461 	file_flags = force_nonblock ? O_NONBLOCK : 0;
1462 
1463 	ret = __sys_connect_file(req->file, &io->address,
1464 					connect->addr_len, file_flags);
1465 	if ((ret == -EAGAIN || ret == -EINPROGRESS || ret == -ECONNABORTED)
1466 	    && force_nonblock) {
1467 		if (ret == -EINPROGRESS) {
1468 			connect->in_progress = true;
1469 			return -EAGAIN;
1470 		}
1471 		if (ret == -ECONNABORTED) {
1472 			if (connect->seen_econnaborted)
1473 				goto out;
1474 			connect->seen_econnaborted = true;
1475 		}
1476 		if (req_has_async_data(req))
1477 			return -EAGAIN;
1478 		if (io_alloc_async_data(req)) {
1479 			ret = -ENOMEM;
1480 			goto out;
1481 		}
1482 		memcpy(req->async_data, &__io, sizeof(__io));
1483 		return -EAGAIN;
1484 	}
1485 	if (ret == -ERESTARTSYS)
1486 		ret = -EINTR;
1487 out:
1488 	if (ret < 0)
1489 		req_set_fail(req);
1490 	io_req_set_res(req, ret, 0);
1491 	return IOU_OK;
1492 }
1493 
1494 void io_netmsg_cache_free(struct io_cache_entry *entry)
1495 {
1496 	kfree(container_of(entry, struct io_async_msghdr, cache));
1497 }
1498 #endif
1499