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