• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3  *
4  * Copyright (c) 2021, Red Hat.
5  */
6 
7 #define pr_fmt(fmt) "MPTCP: " fmt
8 
9 #include <linux/kernel.h>
10 #include <linux/module.h>
11 #include <net/sock.h>
12 #include <net/protocol.h>
13 #include <net/tcp.h>
14 #include <net/mptcp.h>
15 #include "protocol.h"
16 
17 #define MIN_INFO_OPTLEN_SIZE		16
18 #define MIN_FULL_INFO_OPTLEN_SIZE	40
19 
__mptcp_tcp_fallback(struct mptcp_sock * msk)20 static struct sock *__mptcp_tcp_fallback(struct mptcp_sock *msk)
21 {
22 	msk_owned_by_me(msk);
23 
24 	if (likely(!__mptcp_check_fallback(msk)))
25 		return NULL;
26 
27 	return msk->first;
28 }
29 
sockopt_seq_reset(const struct sock * sk)30 static u32 sockopt_seq_reset(const struct sock *sk)
31 {
32 	sock_owned_by_me(sk);
33 
34 	/* Highbits contain state.  Allows to distinguish sockopt_seq
35 	 * of listener and established:
36 	 * s0 = new_listener()
37 	 * sockopt(s0) - seq is 1
38 	 * s1 = accept(s0) - s1 inherits seq 1 if listener sk (s0)
39 	 * sockopt(s0) - seq increments to 2 on s0
40 	 * sockopt(s1) // seq increments to 2 on s1 (different option)
41 	 * new ssk completes join, inherits options from s0 // seq 2
42 	 * Needs sync from mptcp join logic, but ssk->seq == msk->seq
43 	 *
44 	 * Set High order bits to sk_state so ssk->seq == msk->seq test
45 	 * will fail.
46 	 */
47 
48 	return (u32)sk->sk_state << 24u;
49 }
50 
sockopt_seq_inc(struct mptcp_sock * msk)51 static void sockopt_seq_inc(struct mptcp_sock *msk)
52 {
53 	u32 seq = (msk->setsockopt_seq + 1) & 0x00ffffff;
54 
55 	msk->setsockopt_seq = sockopt_seq_reset((struct sock *)msk) + seq;
56 }
57 
mptcp_get_int_option(struct mptcp_sock * msk,sockptr_t optval,unsigned int optlen,int * val)58 static int mptcp_get_int_option(struct mptcp_sock *msk, sockptr_t optval,
59 				unsigned int optlen, int *val)
60 {
61 	if (optlen < sizeof(int))
62 		return -EINVAL;
63 
64 	if (copy_from_sockptr(val, optval, sizeof(*val)))
65 		return -EFAULT;
66 
67 	return 0;
68 }
69 
mptcp_sol_socket_sync_intval(struct mptcp_sock * msk,int optname,int val)70 static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, int val)
71 {
72 	struct mptcp_subflow_context *subflow;
73 	struct sock *sk = (struct sock *)msk;
74 
75 	lock_sock(sk);
76 	sockopt_seq_inc(msk);
77 
78 	mptcp_for_each_subflow(msk, subflow) {
79 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
80 		bool slow = lock_sock_fast(ssk);
81 
82 		switch (optname) {
83 		case SO_DEBUG:
84 			sock_valbool_flag(ssk, SOCK_DBG, !!val);
85 			break;
86 		case SO_KEEPALIVE:
87 			if (ssk->sk_prot->keepalive)
88 				ssk->sk_prot->keepalive(ssk, !!val);
89 			sock_valbool_flag(ssk, SOCK_KEEPOPEN, !!val);
90 			break;
91 		case SO_PRIORITY:
92 			ssk->sk_priority = val;
93 			break;
94 		case SO_SNDBUF:
95 		case SO_SNDBUFFORCE:
96 			ssk->sk_userlocks |= SOCK_SNDBUF_LOCK;
97 			WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
98 			mptcp_subflow_ctx(ssk)->cached_sndbuf = sk->sk_sndbuf;
99 			break;
100 		case SO_RCVBUF:
101 		case SO_RCVBUFFORCE:
102 			ssk->sk_userlocks |= SOCK_RCVBUF_LOCK;
103 			WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
104 			break;
105 		case SO_MARK:
106 			if (READ_ONCE(ssk->sk_mark) != sk->sk_mark) {
107 				WRITE_ONCE(ssk->sk_mark, sk->sk_mark);
108 				sk_dst_reset(ssk);
109 			}
110 			break;
111 		case SO_INCOMING_CPU:
112 			WRITE_ONCE(ssk->sk_incoming_cpu, val);
113 			break;
114 		}
115 
116 		subflow->setsockopt_seq = msk->setsockopt_seq;
117 		unlock_sock_fast(ssk, slow);
118 	}
119 
120 	release_sock(sk);
121 }
122 
mptcp_sol_socket_intval(struct mptcp_sock * msk,int optname,int val)123 static int mptcp_sol_socket_intval(struct mptcp_sock *msk, int optname, int val)
124 {
125 	sockptr_t optval = KERNEL_SOCKPTR(&val);
126 	struct sock *sk = (struct sock *)msk;
127 	int ret;
128 
129 	ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
130 			      optval, sizeof(val));
131 	if (ret)
132 		return ret;
133 
134 	mptcp_sol_socket_sync_intval(msk, optname, val);
135 	return 0;
136 }
137 
mptcp_so_incoming_cpu(struct mptcp_sock * msk,int val)138 static void mptcp_so_incoming_cpu(struct mptcp_sock *msk, int val)
139 {
140 	struct sock *sk = (struct sock *)msk;
141 
142 	WRITE_ONCE(sk->sk_incoming_cpu, val);
143 
144 	mptcp_sol_socket_sync_intval(msk, SO_INCOMING_CPU, val);
145 }
146 
mptcp_setsockopt_sol_socket_tstamp(struct mptcp_sock * msk,int optname,int val)147 static int mptcp_setsockopt_sol_socket_tstamp(struct mptcp_sock *msk, int optname, int val)
148 {
149 	sockptr_t optval = KERNEL_SOCKPTR(&val);
150 	struct mptcp_subflow_context *subflow;
151 	struct sock *sk = (struct sock *)msk;
152 	int ret;
153 
154 	ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
155 			      optval, sizeof(val));
156 	if (ret)
157 		return ret;
158 
159 	lock_sock(sk);
160 	mptcp_for_each_subflow(msk, subflow) {
161 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
162 		bool slow = lock_sock_fast(ssk);
163 
164 		sock_set_timestamp(sk, optname, !!val);
165 		unlock_sock_fast(ssk, slow);
166 	}
167 
168 	release_sock(sk);
169 	return 0;
170 }
171 
mptcp_setsockopt_sol_socket_int(struct mptcp_sock * msk,int optname,sockptr_t optval,unsigned int optlen)172 static int mptcp_setsockopt_sol_socket_int(struct mptcp_sock *msk, int optname,
173 					   sockptr_t optval,
174 					   unsigned int optlen)
175 {
176 	int val, ret;
177 
178 	ret = mptcp_get_int_option(msk, optval, optlen, &val);
179 	if (ret)
180 		return ret;
181 
182 	switch (optname) {
183 	case SO_KEEPALIVE:
184 	case SO_DEBUG:
185 	case SO_MARK:
186 	case SO_PRIORITY:
187 	case SO_SNDBUF:
188 	case SO_SNDBUFFORCE:
189 	case SO_RCVBUF:
190 	case SO_RCVBUFFORCE:
191 		return mptcp_sol_socket_intval(msk, optname, val);
192 	case SO_INCOMING_CPU:
193 		mptcp_so_incoming_cpu(msk, val);
194 		return 0;
195 	case SO_TIMESTAMP_OLD:
196 	case SO_TIMESTAMP_NEW:
197 	case SO_TIMESTAMPNS_OLD:
198 	case SO_TIMESTAMPNS_NEW:
199 		return mptcp_setsockopt_sol_socket_tstamp(msk, optname, val);
200 	}
201 
202 	return -ENOPROTOOPT;
203 }
204 
mptcp_setsockopt_sol_socket_timestamping(struct mptcp_sock * msk,int optname,sockptr_t optval,unsigned int optlen)205 static int mptcp_setsockopt_sol_socket_timestamping(struct mptcp_sock *msk,
206 						    int optname,
207 						    sockptr_t optval,
208 						    unsigned int optlen)
209 {
210 	struct mptcp_subflow_context *subflow;
211 	struct sock *sk = (struct sock *)msk;
212 	struct so_timestamping timestamping;
213 	int ret;
214 
215 	if (optlen == sizeof(timestamping)) {
216 		if (copy_from_sockptr(&timestamping, optval,
217 				      sizeof(timestamping)))
218 			return -EFAULT;
219 	} else if (optlen == sizeof(int)) {
220 		memset(&timestamping, 0, sizeof(timestamping));
221 
222 		if (copy_from_sockptr(&timestamping.flags, optval, sizeof(int)))
223 			return -EFAULT;
224 	} else {
225 		return -EINVAL;
226 	}
227 
228 	ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
229 			      KERNEL_SOCKPTR(&timestamping),
230 			      sizeof(timestamping));
231 	if (ret)
232 		return ret;
233 
234 	lock_sock(sk);
235 
236 	mptcp_for_each_subflow(msk, subflow) {
237 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
238 		bool slow = lock_sock_fast(ssk);
239 
240 		sock_set_timestamping(sk, optname, timestamping);
241 		unlock_sock_fast(ssk, slow);
242 	}
243 
244 	release_sock(sk);
245 
246 	return 0;
247 }
248 
mptcp_setsockopt_sol_socket_linger(struct mptcp_sock * msk,sockptr_t optval,unsigned int optlen)249 static int mptcp_setsockopt_sol_socket_linger(struct mptcp_sock *msk, sockptr_t optval,
250 					      unsigned int optlen)
251 {
252 	struct mptcp_subflow_context *subflow;
253 	struct sock *sk = (struct sock *)msk;
254 	struct linger ling;
255 	sockptr_t kopt;
256 	int ret;
257 
258 	if (optlen < sizeof(ling))
259 		return -EINVAL;
260 
261 	if (copy_from_sockptr(&ling, optval, sizeof(ling)))
262 		return -EFAULT;
263 
264 	kopt = KERNEL_SOCKPTR(&ling);
265 	ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, SO_LINGER, kopt, sizeof(ling));
266 	if (ret)
267 		return ret;
268 
269 	lock_sock(sk);
270 	sockopt_seq_inc(msk);
271 	mptcp_for_each_subflow(msk, subflow) {
272 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
273 		bool slow = lock_sock_fast(ssk);
274 
275 		if (!ling.l_onoff) {
276 			sock_reset_flag(ssk, SOCK_LINGER);
277 		} else {
278 			ssk->sk_lingertime = sk->sk_lingertime;
279 			sock_set_flag(ssk, SOCK_LINGER);
280 		}
281 
282 		subflow->setsockopt_seq = msk->setsockopt_seq;
283 		unlock_sock_fast(ssk, slow);
284 	}
285 
286 	release_sock(sk);
287 	return 0;
288 }
289 
mptcp_setsockopt_sol_socket(struct mptcp_sock * msk,int optname,sockptr_t optval,unsigned int optlen)290 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
291 				       sockptr_t optval, unsigned int optlen)
292 {
293 	struct sock *sk = (struct sock *)msk;
294 	struct sock *ssk;
295 	int ret;
296 
297 	switch (optname) {
298 	case SO_REUSEPORT:
299 	case SO_REUSEADDR:
300 	case SO_BINDTODEVICE:
301 	case SO_BINDTOIFINDEX:
302 		lock_sock(sk);
303 		ssk = __mptcp_nmpc_sk(msk);
304 		if (IS_ERR(ssk)) {
305 			release_sock(sk);
306 			return PTR_ERR(ssk);
307 		}
308 
309 		ret = sk_setsockopt(ssk, SOL_SOCKET, optname, optval, optlen);
310 		if (ret == 0) {
311 			if (optname == SO_REUSEPORT)
312 				sk->sk_reuseport = ssk->sk_reuseport;
313 			else if (optname == SO_REUSEADDR)
314 				sk->sk_reuse = ssk->sk_reuse;
315 			else if (optname == SO_BINDTODEVICE)
316 				sk->sk_bound_dev_if = ssk->sk_bound_dev_if;
317 			else if (optname == SO_BINDTOIFINDEX)
318 				sk->sk_bound_dev_if = ssk->sk_bound_dev_if;
319 		}
320 		release_sock(sk);
321 		return ret;
322 	case SO_KEEPALIVE:
323 	case SO_PRIORITY:
324 	case SO_SNDBUF:
325 	case SO_SNDBUFFORCE:
326 	case SO_RCVBUF:
327 	case SO_RCVBUFFORCE:
328 	case SO_MARK:
329 	case SO_INCOMING_CPU:
330 	case SO_DEBUG:
331 	case SO_TIMESTAMP_OLD:
332 	case SO_TIMESTAMP_NEW:
333 	case SO_TIMESTAMPNS_OLD:
334 	case SO_TIMESTAMPNS_NEW:
335 		return mptcp_setsockopt_sol_socket_int(msk, optname, optval,
336 						       optlen);
337 	case SO_TIMESTAMPING_OLD:
338 	case SO_TIMESTAMPING_NEW:
339 		return mptcp_setsockopt_sol_socket_timestamping(msk, optname,
340 								optval, optlen);
341 	case SO_LINGER:
342 		return mptcp_setsockopt_sol_socket_linger(msk, optval, optlen);
343 	case SO_RCVLOWAT:
344 	case SO_RCVTIMEO_OLD:
345 	case SO_RCVTIMEO_NEW:
346 	case SO_SNDTIMEO_OLD:
347 	case SO_SNDTIMEO_NEW:
348 	case SO_BUSY_POLL:
349 	case SO_PREFER_BUSY_POLL:
350 	case SO_BUSY_POLL_BUDGET:
351 		/* No need to copy: only relevant for msk */
352 		return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
353 	case SO_NO_CHECK:
354 	case SO_DONTROUTE:
355 	case SO_BROADCAST:
356 	case SO_BSDCOMPAT:
357 	case SO_PASSCRED:
358 	case SO_PASSPIDFD:
359 	case SO_PASSSEC:
360 	case SO_RXQ_OVFL:
361 	case SO_WIFI_STATUS:
362 	case SO_NOFCS:
363 	case SO_SELECT_ERR_QUEUE:
364 		return 0;
365 	}
366 
367 	/* SO_OOBINLINE is not supported, let's avoid the related mess
368 	 * SO_ATTACH_FILTER, SO_ATTACH_BPF, SO_ATTACH_REUSEPORT_CBPF,
369 	 * SO_DETACH_REUSEPORT_BPF, SO_DETACH_FILTER, SO_LOCK_FILTER,
370 	 * we must be careful with subflows
371 	 *
372 	 * SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks
373 	 * explicitly the sk_protocol field
374 	 *
375 	 * SO_PEEK_OFF is unsupported, as it is for plain TCP
376 	 * SO_MAX_PACING_RATE is unsupported, we must be careful with subflows
377 	 * SO_CNX_ADVICE is currently unsupported, could possibly be relevant,
378 	 * but likely needs careful design
379 	 *
380 	 * SO_ZEROCOPY is currently unsupported, TODO in sndmsg
381 	 * SO_TXTIME is currently unsupported
382 	 */
383 
384 	return -EOPNOTSUPP;
385 }
386 
mptcp_setsockopt_v6(struct mptcp_sock * msk,int optname,sockptr_t optval,unsigned int optlen)387 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
388 			       sockptr_t optval, unsigned int optlen)
389 {
390 	struct sock *sk = (struct sock *)msk;
391 	int ret = -EOPNOTSUPP;
392 	struct sock *ssk;
393 
394 	switch (optname) {
395 	case IPV6_V6ONLY:
396 	case IPV6_TRANSPARENT:
397 	case IPV6_FREEBIND:
398 		lock_sock(sk);
399 		ssk = __mptcp_nmpc_sk(msk);
400 		if (IS_ERR(ssk)) {
401 			release_sock(sk);
402 			return PTR_ERR(ssk);
403 		}
404 
405 		ret = tcp_setsockopt(ssk, SOL_IPV6, optname, optval, optlen);
406 		if (ret != 0) {
407 			release_sock(sk);
408 			return ret;
409 		}
410 
411 		sockopt_seq_inc(msk);
412 
413 		switch (optname) {
414 		case IPV6_V6ONLY:
415 			sk->sk_ipv6only = ssk->sk_ipv6only;
416 			break;
417 		case IPV6_TRANSPARENT:
418 			inet_assign_bit(TRANSPARENT, sk,
419 					inet_test_bit(TRANSPARENT, ssk));
420 			break;
421 		case IPV6_FREEBIND:
422 			inet_assign_bit(FREEBIND, sk,
423 					inet_test_bit(FREEBIND, ssk));
424 			break;
425 		}
426 
427 		release_sock(sk);
428 		break;
429 	}
430 
431 	return ret;
432 }
433 
mptcp_supported_sockopt(int level,int optname)434 static bool mptcp_supported_sockopt(int level, int optname)
435 {
436 	if (level == SOL_IP) {
437 		switch (optname) {
438 		/* should work fine */
439 		case IP_FREEBIND:
440 		case IP_TRANSPARENT:
441 
442 		/* the following are control cmsg related */
443 		case IP_PKTINFO:
444 		case IP_RECVTTL:
445 		case IP_RECVTOS:
446 		case IP_RECVOPTS:
447 		case IP_RETOPTS:
448 		case IP_PASSSEC:
449 		case IP_RECVORIGDSTADDR:
450 		case IP_CHECKSUM:
451 		case IP_RECVFRAGSIZE:
452 
453 		/* common stuff that need some love */
454 		case IP_TOS:
455 		case IP_TTL:
456 		case IP_BIND_ADDRESS_NO_PORT:
457 		case IP_MTU_DISCOVER:
458 		case IP_RECVERR:
459 
460 		/* possibly less common may deserve some love */
461 		case IP_MINTTL:
462 
463 		/* the following is apparently a no-op for plain TCP */
464 		case IP_RECVERR_RFC4884:
465 			return true;
466 		}
467 
468 		/* IP_OPTIONS is not supported, needs subflow care */
469 		/* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */
470 		/* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF,
471 		 * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP,
472 		 * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE,
473 		 * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP,
474 		 * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE,
475 		 * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal
476 		 * with mcast stuff
477 		 */
478 		/* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */
479 		return false;
480 	}
481 	if (level == SOL_IPV6) {
482 		switch (optname) {
483 		case IPV6_V6ONLY:
484 
485 		/* the following are control cmsg related */
486 		case IPV6_RECVPKTINFO:
487 		case IPV6_2292PKTINFO:
488 		case IPV6_RECVHOPLIMIT:
489 		case IPV6_2292HOPLIMIT:
490 		case IPV6_RECVRTHDR:
491 		case IPV6_2292RTHDR:
492 		case IPV6_RECVHOPOPTS:
493 		case IPV6_2292HOPOPTS:
494 		case IPV6_RECVDSTOPTS:
495 		case IPV6_2292DSTOPTS:
496 		case IPV6_RECVTCLASS:
497 		case IPV6_FLOWINFO:
498 		case IPV6_RECVPATHMTU:
499 		case IPV6_RECVORIGDSTADDR:
500 		case IPV6_RECVFRAGSIZE:
501 
502 		/* the following ones need some love but are quite common */
503 		case IPV6_TCLASS:
504 		case IPV6_TRANSPARENT:
505 		case IPV6_FREEBIND:
506 		case IPV6_PKTINFO:
507 		case IPV6_2292PKTOPTIONS:
508 		case IPV6_UNICAST_HOPS:
509 		case IPV6_MTU_DISCOVER:
510 		case IPV6_MTU:
511 		case IPV6_RECVERR:
512 		case IPV6_FLOWINFO_SEND:
513 		case IPV6_FLOWLABEL_MGR:
514 		case IPV6_MINHOPCOUNT:
515 		case IPV6_DONTFRAG:
516 		case IPV6_AUTOFLOWLABEL:
517 
518 		/* the following one is a no-op for plain TCP */
519 		case IPV6_RECVERR_RFC4884:
520 			return true;
521 		}
522 
523 		/* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are
524 		 * not supported
525 		 */
526 		/* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF,
527 		 * IPV6_MULTICAST_IF, IPV6_ADDRFORM,
528 		 * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST,
529 		 * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP,
530 		 * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP,
531 		 * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER
532 		 * are not supported better not deal with mcast
533 		 */
534 		/* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */
535 
536 		/* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */
537 		/* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */
538 		return false;
539 	}
540 	if (level == SOL_TCP) {
541 		switch (optname) {
542 		/* the following are no-op or should work just fine */
543 		case TCP_THIN_DUPACK:
544 		case TCP_DEFER_ACCEPT:
545 
546 		/* the following need some love */
547 		case TCP_MAXSEG:
548 		case TCP_NODELAY:
549 		case TCP_THIN_LINEAR_TIMEOUTS:
550 		case TCP_CONGESTION:
551 		case TCP_CORK:
552 		case TCP_KEEPIDLE:
553 		case TCP_KEEPINTVL:
554 		case TCP_KEEPCNT:
555 		case TCP_SYNCNT:
556 		case TCP_SAVE_SYN:
557 		case TCP_LINGER2:
558 		case TCP_WINDOW_CLAMP:
559 		case TCP_QUICKACK:
560 		case TCP_USER_TIMEOUT:
561 		case TCP_TIMESTAMP:
562 		case TCP_NOTSENT_LOWAT:
563 		case TCP_TX_DELAY:
564 		case TCP_INQ:
565 		case TCP_FASTOPEN:
566 		case TCP_FASTOPEN_CONNECT:
567 		case TCP_FASTOPEN_KEY:
568 		case TCP_FASTOPEN_NO_COOKIE:
569 			return true;
570 		}
571 
572 		/* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */
573 
574 		/* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS,
575 		 * TCP_REPAIR_WINDOW are not supported, better avoid this mess
576 		 */
577 	}
578 	return false;
579 }
580 
mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock * msk,sockptr_t optval,unsigned int optlen)581 static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval,
582 					       unsigned int optlen)
583 {
584 	struct mptcp_subflow_context *subflow;
585 	struct sock *sk = (struct sock *)msk;
586 	char name[TCP_CA_NAME_MAX];
587 	bool cap_net_admin;
588 	int ret;
589 
590 	if (optlen < 1)
591 		return -EINVAL;
592 
593 	ret = strncpy_from_sockptr(name, optval,
594 				   min_t(long, TCP_CA_NAME_MAX - 1, optlen));
595 	if (ret < 0)
596 		return -EFAULT;
597 
598 	name[ret] = 0;
599 
600 	cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN);
601 
602 	ret = 0;
603 	lock_sock(sk);
604 	sockopt_seq_inc(msk);
605 	mptcp_for_each_subflow(msk, subflow) {
606 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
607 		int err;
608 
609 		lock_sock(ssk);
610 		err = tcp_set_congestion_control(ssk, name, true, cap_net_admin);
611 		if (err < 0 && ret == 0)
612 			ret = err;
613 		subflow->setsockopt_seq = msk->setsockopt_seq;
614 		release_sock(ssk);
615 	}
616 
617 	if (ret == 0)
618 		strcpy(msk->ca_name, name);
619 
620 	release_sock(sk);
621 	return ret;
622 }
623 
__mptcp_setsockopt_set_val(struct mptcp_sock * msk,int max,int (* set_val)(struct sock *,int),int * msk_val,int val)624 static int __mptcp_setsockopt_set_val(struct mptcp_sock *msk, int max,
625 				      int (*set_val)(struct sock *, int),
626 				      int *msk_val, int val)
627 {
628 	struct mptcp_subflow_context *subflow;
629 	int err = 0;
630 
631 	mptcp_for_each_subflow(msk, subflow) {
632 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
633 		int ret;
634 
635 		lock_sock(ssk);
636 		ret = set_val(ssk, val);
637 		err = err ? : ret;
638 		release_sock(ssk);
639 	}
640 
641 	if (!err) {
642 		*msk_val = val;
643 		sockopt_seq_inc(msk);
644 	}
645 
646 	return err;
647 }
648 
__mptcp_setsockopt_sol_tcp_cork(struct mptcp_sock * msk,int val)649 static int __mptcp_setsockopt_sol_tcp_cork(struct mptcp_sock *msk, int val)
650 {
651 	struct mptcp_subflow_context *subflow;
652 	struct sock *sk = (struct sock *)msk;
653 
654 	sockopt_seq_inc(msk);
655 	msk->cork = !!val;
656 	mptcp_for_each_subflow(msk, subflow) {
657 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
658 
659 		lock_sock(ssk);
660 		__tcp_sock_set_cork(ssk, !!val);
661 		release_sock(ssk);
662 	}
663 	if (!val)
664 		mptcp_check_and_set_pending(sk);
665 
666 	return 0;
667 }
668 
__mptcp_setsockopt_sol_tcp_nodelay(struct mptcp_sock * msk,int val)669 static int __mptcp_setsockopt_sol_tcp_nodelay(struct mptcp_sock *msk, int val)
670 {
671 	struct mptcp_subflow_context *subflow;
672 	struct sock *sk = (struct sock *)msk;
673 
674 	sockopt_seq_inc(msk);
675 	msk->nodelay = !!val;
676 	mptcp_for_each_subflow(msk, subflow) {
677 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
678 
679 		lock_sock(ssk);
680 		__tcp_sock_set_nodelay(ssk, !!val);
681 		release_sock(ssk);
682 	}
683 	if (val)
684 		mptcp_check_and_set_pending(sk);
685 	return 0;
686 }
687 
mptcp_setsockopt_sol_ip_set_transparent(struct mptcp_sock * msk,int optname,sockptr_t optval,unsigned int optlen)688 static int mptcp_setsockopt_sol_ip_set_transparent(struct mptcp_sock *msk, int optname,
689 						   sockptr_t optval, unsigned int optlen)
690 {
691 	struct sock *sk = (struct sock *)msk;
692 	struct sock *ssk;
693 	int err;
694 
695 	err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen);
696 	if (err != 0)
697 		return err;
698 
699 	lock_sock(sk);
700 
701 	ssk = __mptcp_nmpc_sk(msk);
702 	if (IS_ERR(ssk)) {
703 		release_sock(sk);
704 		return PTR_ERR(ssk);
705 	}
706 
707 	switch (optname) {
708 	case IP_FREEBIND:
709 		inet_assign_bit(FREEBIND, ssk, inet_test_bit(FREEBIND, sk));
710 		break;
711 	case IP_TRANSPARENT:
712 		inet_assign_bit(TRANSPARENT, ssk,
713 				inet_test_bit(TRANSPARENT, sk));
714 		break;
715 	default:
716 		release_sock(sk);
717 		WARN_ON_ONCE(1);
718 		return -EOPNOTSUPP;
719 	}
720 
721 	sockopt_seq_inc(msk);
722 	release_sock(sk);
723 	return 0;
724 }
725 
mptcp_setsockopt_v4_set_tos(struct mptcp_sock * msk,int optname,sockptr_t optval,unsigned int optlen)726 static int mptcp_setsockopt_v4_set_tos(struct mptcp_sock *msk, int optname,
727 				       sockptr_t optval, unsigned int optlen)
728 {
729 	struct mptcp_subflow_context *subflow;
730 	struct sock *sk = (struct sock *)msk;
731 	int err, val;
732 
733 	err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen);
734 
735 	if (err != 0)
736 		return err;
737 
738 	lock_sock(sk);
739 	sockopt_seq_inc(msk);
740 	val = inet_sk(sk)->tos;
741 	mptcp_for_each_subflow(msk, subflow) {
742 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
743 		bool slow;
744 
745 		slow = lock_sock_fast(ssk);
746 		__ip_sock_set_tos(ssk, val);
747 		unlock_sock_fast(ssk, slow);
748 	}
749 	release_sock(sk);
750 
751 	return 0;
752 }
753 
mptcp_setsockopt_v4(struct mptcp_sock * msk,int optname,sockptr_t optval,unsigned int optlen)754 static int mptcp_setsockopt_v4(struct mptcp_sock *msk, int optname,
755 			       sockptr_t optval, unsigned int optlen)
756 {
757 	switch (optname) {
758 	case IP_FREEBIND:
759 	case IP_TRANSPARENT:
760 		return mptcp_setsockopt_sol_ip_set_transparent(msk, optname, optval, optlen);
761 	case IP_TOS:
762 		return mptcp_setsockopt_v4_set_tos(msk, optname, optval, optlen);
763 	}
764 
765 	return -EOPNOTSUPP;
766 }
767 
mptcp_setsockopt_first_sf_only(struct mptcp_sock * msk,int level,int optname,sockptr_t optval,unsigned int optlen)768 static int mptcp_setsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
769 					  sockptr_t optval, unsigned int optlen)
770 {
771 	struct sock *sk = (struct sock *)msk;
772 	struct sock *ssk;
773 	int ret;
774 
775 	/* Limit to first subflow, before the connection establishment */
776 	lock_sock(sk);
777 	ssk = __mptcp_nmpc_sk(msk);
778 	if (IS_ERR(ssk)) {
779 		ret = PTR_ERR(ssk);
780 		goto unlock;
781 	}
782 
783 	ret = tcp_setsockopt(ssk, level, optname, optval, optlen);
784 
785 unlock:
786 	release_sock(sk);
787 	return ret;
788 }
789 
mptcp_setsockopt_sol_tcp(struct mptcp_sock * msk,int optname,sockptr_t optval,unsigned int optlen)790 static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
791 				    sockptr_t optval, unsigned int optlen)
792 {
793 	struct sock *sk = (void *)msk;
794 	int ret, val;
795 
796 	switch (optname) {
797 	case TCP_ULP:
798 		return -EOPNOTSUPP;
799 	case TCP_CONGESTION:
800 		return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen);
801 	case TCP_DEFER_ACCEPT:
802 		/* See tcp.c: TCP_DEFER_ACCEPT does not fail */
803 		mptcp_setsockopt_first_sf_only(msk, SOL_TCP, optname, optval, optlen);
804 		return 0;
805 	case TCP_FASTOPEN:
806 	case TCP_FASTOPEN_CONNECT:
807 	case TCP_FASTOPEN_KEY:
808 	case TCP_FASTOPEN_NO_COOKIE:
809 		return mptcp_setsockopt_first_sf_only(msk, SOL_TCP, optname,
810 						      optval, optlen);
811 	}
812 
813 	ret = mptcp_get_int_option(msk, optval, optlen, &val);
814 	if (ret)
815 		return ret;
816 
817 	lock_sock(sk);
818 	switch (optname) {
819 	case TCP_INQ:
820 		if (val < 0 || val > 1)
821 			ret = -EINVAL;
822 		else
823 			msk->recvmsg_inq = !!val;
824 		break;
825 	case TCP_CORK:
826 		ret = __mptcp_setsockopt_sol_tcp_cork(msk, val);
827 		break;
828 	case TCP_NODELAY:
829 		ret = __mptcp_setsockopt_sol_tcp_nodelay(msk, val);
830 		break;
831 	case TCP_KEEPIDLE:
832 		ret = __mptcp_setsockopt_set_val(msk, MAX_TCP_KEEPIDLE,
833 						 &tcp_sock_set_keepidle_locked,
834 						 &msk->keepalive_idle, val);
835 		break;
836 	case TCP_KEEPINTVL:
837 		ret = __mptcp_setsockopt_set_val(msk, MAX_TCP_KEEPINTVL,
838 						 &tcp_sock_set_keepintvl,
839 						 &msk->keepalive_intvl, val);
840 		break;
841 	case TCP_KEEPCNT:
842 		ret = __mptcp_setsockopt_set_val(msk, MAX_TCP_KEEPCNT,
843 						 &tcp_sock_set_keepcnt,
844 						 &msk->keepalive_cnt,
845 						 val);
846 		break;
847 	default:
848 		ret = -ENOPROTOOPT;
849 	}
850 
851 	release_sock(sk);
852 	return ret;
853 }
854 
mptcp_setsockopt(struct sock * sk,int level,int optname,sockptr_t optval,unsigned int optlen)855 int mptcp_setsockopt(struct sock *sk, int level, int optname,
856 		     sockptr_t optval, unsigned int optlen)
857 {
858 	struct mptcp_sock *msk = mptcp_sk(sk);
859 	struct sock *ssk;
860 
861 	pr_debug("msk=%p", msk);
862 
863 	if (level == SOL_SOCKET)
864 		return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
865 
866 	if (!mptcp_supported_sockopt(level, optname))
867 		return -ENOPROTOOPT;
868 
869 	/* @@ the meaning of setsockopt() when the socket is connected and
870 	 * there are multiple subflows is not yet defined. It is up to the
871 	 * MPTCP-level socket to configure the subflows until the subflow
872 	 * is in TCP fallback, when TCP socket options are passed through
873 	 * to the one remaining subflow.
874 	 */
875 	lock_sock(sk);
876 	ssk = __mptcp_tcp_fallback(msk);
877 	release_sock(sk);
878 	if (ssk)
879 		return tcp_setsockopt(ssk, level, optname, optval, optlen);
880 
881 	if (level == SOL_IP)
882 		return mptcp_setsockopt_v4(msk, optname, optval, optlen);
883 
884 	if (level == SOL_IPV6)
885 		return mptcp_setsockopt_v6(msk, optname, optval, optlen);
886 
887 	if (level == SOL_TCP)
888 		return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen);
889 
890 	return -EOPNOTSUPP;
891 }
892 
mptcp_getsockopt_first_sf_only(struct mptcp_sock * msk,int level,int optname,char __user * optval,int __user * optlen)893 static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
894 					  char __user *optval, int __user *optlen)
895 {
896 	struct sock *sk = (struct sock *)msk;
897 	struct sock *ssk;
898 	int ret;
899 
900 	lock_sock(sk);
901 	ssk = msk->first;
902 	if (ssk) {
903 		ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
904 		goto out;
905 	}
906 
907 	ssk = __mptcp_nmpc_sk(msk);
908 	if (IS_ERR(ssk)) {
909 		ret = PTR_ERR(ssk);
910 		goto out;
911 	}
912 
913 	ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
914 
915 out:
916 	release_sock(sk);
917 	return ret;
918 }
919 
mptcp_diag_fill_info(struct mptcp_sock * msk,struct mptcp_info * info)920 void mptcp_diag_fill_info(struct mptcp_sock *msk, struct mptcp_info *info)
921 {
922 	struct sock *sk = (struct sock *)msk;
923 	u32 flags = 0;
924 	bool slow;
925 
926 	memset(info, 0, sizeof(*info));
927 
928 	info->mptcpi_subflows = READ_ONCE(msk->pm.subflows);
929 	info->mptcpi_add_addr_signal = READ_ONCE(msk->pm.add_addr_signaled);
930 	info->mptcpi_add_addr_accepted = READ_ONCE(msk->pm.add_addr_accepted);
931 	info->mptcpi_local_addr_used = READ_ONCE(msk->pm.local_addr_used);
932 
933 	if (inet_sk_state_load(sk) == TCP_LISTEN)
934 		return;
935 
936 	/* The following limits only make sense for the in-kernel PM */
937 	if (mptcp_pm_is_kernel(msk)) {
938 		info->mptcpi_subflows_max =
939 			mptcp_pm_get_subflows_max(msk);
940 		info->mptcpi_add_addr_signal_max =
941 			mptcp_pm_get_add_addr_signal_max(msk);
942 		info->mptcpi_add_addr_accepted_max =
943 			mptcp_pm_get_add_addr_accept_max(msk);
944 		info->mptcpi_local_addr_max =
945 			mptcp_pm_get_local_addr_max(msk);
946 	}
947 
948 	if (test_bit(MPTCP_FALLBACK_DONE, &msk->flags))
949 		flags |= MPTCP_INFO_FLAG_FALLBACK;
950 	if (READ_ONCE(msk->can_ack))
951 		flags |= MPTCP_INFO_FLAG_REMOTE_KEY_RECEIVED;
952 	info->mptcpi_flags = flags;
953 	mptcp_data_lock(sk);
954 	info->mptcpi_snd_una = msk->snd_una;
955 	info->mptcpi_rcv_nxt = msk->ack_seq;
956 	info->mptcpi_bytes_acked = msk->bytes_acked;
957 	mptcp_data_unlock(sk);
958 
959 	slow = lock_sock_fast(sk);
960 	info->mptcpi_csum_enabled = msk->csum_enabled;
961 	info->mptcpi_token = msk->token;
962 	info->mptcpi_write_seq = msk->write_seq;
963 	info->mptcpi_retransmits = inet_csk(sk)->icsk_retransmits;
964 	info->mptcpi_bytes_sent = msk->bytes_sent;
965 	info->mptcpi_bytes_received = msk->bytes_received;
966 	info->mptcpi_bytes_retrans = msk->bytes_retrans;
967 	unlock_sock_fast(sk, slow);
968 }
969 EXPORT_SYMBOL_GPL(mptcp_diag_fill_info);
970 
mptcp_getsockopt_info(struct mptcp_sock * msk,char __user * optval,int __user * optlen)971 static int mptcp_getsockopt_info(struct mptcp_sock *msk, char __user *optval, int __user *optlen)
972 {
973 	struct mptcp_info m_info;
974 	int len;
975 
976 	if (get_user(len, optlen))
977 		return -EFAULT;
978 
979 	len = min_t(unsigned int, len, sizeof(struct mptcp_info));
980 
981 	mptcp_diag_fill_info(msk, &m_info);
982 
983 	if (put_user(len, optlen))
984 		return -EFAULT;
985 
986 	if (copy_to_user(optval, &m_info, len))
987 		return -EFAULT;
988 
989 	return 0;
990 }
991 
mptcp_put_subflow_data(struct mptcp_subflow_data * sfd,char __user * optval,u32 copied,int __user * optlen)992 static int mptcp_put_subflow_data(struct mptcp_subflow_data *sfd,
993 				  char __user *optval,
994 				  u32 copied,
995 				  int __user *optlen)
996 {
997 	u32 copylen = min_t(u32, sfd->size_subflow_data, sizeof(*sfd));
998 
999 	if (copied)
1000 		copied += sfd->size_subflow_data;
1001 	else
1002 		copied = copylen;
1003 
1004 	if (put_user(copied, optlen))
1005 		return -EFAULT;
1006 
1007 	if (copy_to_user(optval, sfd, copylen))
1008 		return -EFAULT;
1009 
1010 	return 0;
1011 }
1012 
mptcp_get_subflow_data(struct mptcp_subflow_data * sfd,char __user * optval,int __user * optlen)1013 static int mptcp_get_subflow_data(struct mptcp_subflow_data *sfd,
1014 				  char __user *optval,
1015 				  int __user *optlen)
1016 {
1017 	int len, copylen;
1018 
1019 	if (get_user(len, optlen))
1020 		return -EFAULT;
1021 
1022 	/* if mptcp_subflow_data size is changed, need to adjust
1023 	 * this function to deal with programs using old version.
1024 	 */
1025 	BUILD_BUG_ON(sizeof(*sfd) != MIN_INFO_OPTLEN_SIZE);
1026 
1027 	if (len < MIN_INFO_OPTLEN_SIZE)
1028 		return -EINVAL;
1029 
1030 	memset(sfd, 0, sizeof(*sfd));
1031 
1032 	copylen = min_t(unsigned int, len, sizeof(*sfd));
1033 	if (copy_from_user(sfd, optval, copylen))
1034 		return -EFAULT;
1035 
1036 	/* size_subflow_data is u32, but len is signed */
1037 	if (sfd->size_subflow_data > INT_MAX ||
1038 	    sfd->size_user > INT_MAX)
1039 		return -EINVAL;
1040 
1041 	if (sfd->size_subflow_data < MIN_INFO_OPTLEN_SIZE ||
1042 	    sfd->size_subflow_data > len)
1043 		return -EINVAL;
1044 
1045 	if (sfd->num_subflows || sfd->size_kernel)
1046 		return -EINVAL;
1047 
1048 	return len - sfd->size_subflow_data;
1049 }
1050 
mptcp_getsockopt_tcpinfo(struct mptcp_sock * msk,char __user * optval,int __user * optlen)1051 static int mptcp_getsockopt_tcpinfo(struct mptcp_sock *msk, char __user *optval,
1052 				    int __user *optlen)
1053 {
1054 	struct mptcp_subflow_context *subflow;
1055 	struct sock *sk = (struct sock *)msk;
1056 	unsigned int sfcount = 0, copied = 0;
1057 	struct mptcp_subflow_data sfd;
1058 	char __user *infoptr;
1059 	int len;
1060 
1061 	len = mptcp_get_subflow_data(&sfd, optval, optlen);
1062 	if (len < 0)
1063 		return len;
1064 
1065 	sfd.size_kernel = sizeof(struct tcp_info);
1066 	sfd.size_user = min_t(unsigned int, sfd.size_user,
1067 			      sizeof(struct tcp_info));
1068 
1069 	infoptr = optval + sfd.size_subflow_data;
1070 
1071 	lock_sock(sk);
1072 
1073 	mptcp_for_each_subflow(msk, subflow) {
1074 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1075 
1076 		++sfcount;
1077 
1078 		if (len && len >= sfd.size_user) {
1079 			struct tcp_info info;
1080 
1081 			tcp_get_info(ssk, &info);
1082 
1083 			if (copy_to_user(infoptr, &info, sfd.size_user)) {
1084 				release_sock(sk);
1085 				return -EFAULT;
1086 			}
1087 
1088 			infoptr += sfd.size_user;
1089 			copied += sfd.size_user;
1090 			len -= sfd.size_user;
1091 		}
1092 	}
1093 
1094 	release_sock(sk);
1095 
1096 	sfd.num_subflows = sfcount;
1097 
1098 	if (mptcp_put_subflow_data(&sfd, optval, copied, optlen))
1099 		return -EFAULT;
1100 
1101 	return 0;
1102 }
1103 
mptcp_get_sub_addrs(const struct sock * sk,struct mptcp_subflow_addrs * a)1104 static void mptcp_get_sub_addrs(const struct sock *sk, struct mptcp_subflow_addrs *a)
1105 {
1106 	const struct inet_sock *inet = inet_sk(sk);
1107 
1108 	memset(a, 0, sizeof(*a));
1109 
1110 	if (sk->sk_family == AF_INET) {
1111 		a->sin_local.sin_family = AF_INET;
1112 		a->sin_local.sin_port = inet->inet_sport;
1113 		a->sin_local.sin_addr.s_addr = inet->inet_rcv_saddr;
1114 
1115 		if (!a->sin_local.sin_addr.s_addr)
1116 			a->sin_local.sin_addr.s_addr = inet->inet_saddr;
1117 
1118 		a->sin_remote.sin_family = AF_INET;
1119 		a->sin_remote.sin_port = inet->inet_dport;
1120 		a->sin_remote.sin_addr.s_addr = inet->inet_daddr;
1121 #if IS_ENABLED(CONFIG_IPV6)
1122 	} else if (sk->sk_family == AF_INET6) {
1123 		const struct ipv6_pinfo *np = inet6_sk(sk);
1124 
1125 		if (WARN_ON_ONCE(!np))
1126 			return;
1127 
1128 		a->sin6_local.sin6_family = AF_INET6;
1129 		a->sin6_local.sin6_port = inet->inet_sport;
1130 
1131 		if (ipv6_addr_any(&sk->sk_v6_rcv_saddr))
1132 			a->sin6_local.sin6_addr = np->saddr;
1133 		else
1134 			a->sin6_local.sin6_addr = sk->sk_v6_rcv_saddr;
1135 
1136 		a->sin6_remote.sin6_family = AF_INET6;
1137 		a->sin6_remote.sin6_port = inet->inet_dport;
1138 		a->sin6_remote.sin6_addr = sk->sk_v6_daddr;
1139 #endif
1140 	}
1141 }
1142 
mptcp_getsockopt_subflow_addrs(struct mptcp_sock * msk,char __user * optval,int __user * optlen)1143 static int mptcp_getsockopt_subflow_addrs(struct mptcp_sock *msk, char __user *optval,
1144 					  int __user *optlen)
1145 {
1146 	struct mptcp_subflow_context *subflow;
1147 	struct sock *sk = (struct sock *)msk;
1148 	unsigned int sfcount = 0, copied = 0;
1149 	struct mptcp_subflow_data sfd;
1150 	char __user *addrptr;
1151 	int len;
1152 
1153 	len = mptcp_get_subflow_data(&sfd, optval, optlen);
1154 	if (len < 0)
1155 		return len;
1156 
1157 	sfd.size_kernel = sizeof(struct mptcp_subflow_addrs);
1158 	sfd.size_user = min_t(unsigned int, sfd.size_user,
1159 			      sizeof(struct mptcp_subflow_addrs));
1160 
1161 	addrptr = optval + sfd.size_subflow_data;
1162 
1163 	lock_sock(sk);
1164 
1165 	mptcp_for_each_subflow(msk, subflow) {
1166 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1167 
1168 		++sfcount;
1169 
1170 		if (len && len >= sfd.size_user) {
1171 			struct mptcp_subflow_addrs a;
1172 
1173 			mptcp_get_sub_addrs(ssk, &a);
1174 
1175 			if (copy_to_user(addrptr, &a, sfd.size_user)) {
1176 				release_sock(sk);
1177 				return -EFAULT;
1178 			}
1179 
1180 			addrptr += sfd.size_user;
1181 			copied += sfd.size_user;
1182 			len -= sfd.size_user;
1183 		}
1184 	}
1185 
1186 	release_sock(sk);
1187 
1188 	sfd.num_subflows = sfcount;
1189 
1190 	if (mptcp_put_subflow_data(&sfd, optval, copied, optlen))
1191 		return -EFAULT;
1192 
1193 	return 0;
1194 }
1195 
mptcp_get_full_info(struct mptcp_full_info * mfi,char __user * optval,int __user * optlen)1196 static int mptcp_get_full_info(struct mptcp_full_info *mfi,
1197 			       char __user *optval,
1198 			       int __user *optlen)
1199 {
1200 	int len;
1201 
1202 	BUILD_BUG_ON(offsetof(struct mptcp_full_info, mptcp_info) !=
1203 		     MIN_FULL_INFO_OPTLEN_SIZE);
1204 
1205 	if (get_user(len, optlen))
1206 		return -EFAULT;
1207 
1208 	if (len < MIN_FULL_INFO_OPTLEN_SIZE)
1209 		return -EINVAL;
1210 
1211 	memset(mfi, 0, sizeof(*mfi));
1212 	if (copy_from_user(mfi, optval, MIN_FULL_INFO_OPTLEN_SIZE))
1213 		return -EFAULT;
1214 
1215 	if (mfi->size_tcpinfo_kernel ||
1216 	    mfi->size_sfinfo_kernel ||
1217 	    mfi->num_subflows)
1218 		return -EINVAL;
1219 
1220 	if (mfi->size_sfinfo_user > INT_MAX ||
1221 	    mfi->size_tcpinfo_user > INT_MAX)
1222 		return -EINVAL;
1223 
1224 	return len - MIN_FULL_INFO_OPTLEN_SIZE;
1225 }
1226 
mptcp_put_full_info(struct mptcp_full_info * mfi,char __user * optval,u32 copylen,int __user * optlen)1227 static int mptcp_put_full_info(struct mptcp_full_info *mfi,
1228 			       char __user *optval,
1229 			       u32 copylen,
1230 			       int __user *optlen)
1231 {
1232 	copylen += MIN_FULL_INFO_OPTLEN_SIZE;
1233 	if (put_user(copylen, optlen))
1234 		return -EFAULT;
1235 
1236 	if (copy_to_user(optval, mfi, copylen))
1237 		return -EFAULT;
1238 	return 0;
1239 }
1240 
mptcp_getsockopt_full_info(struct mptcp_sock * msk,char __user * optval,int __user * optlen)1241 static int mptcp_getsockopt_full_info(struct mptcp_sock *msk, char __user *optval,
1242 				      int __user *optlen)
1243 {
1244 	unsigned int sfcount = 0, copylen = 0;
1245 	struct mptcp_subflow_context *subflow;
1246 	struct sock *sk = (struct sock *)msk;
1247 	void __user *tcpinfoptr, *sfinfoptr;
1248 	struct mptcp_full_info mfi;
1249 	int len;
1250 
1251 	len = mptcp_get_full_info(&mfi, optval, optlen);
1252 	if (len < 0)
1253 		return len;
1254 
1255 	/* don't bother filling the mptcp info if there is not enough
1256 	 * user-space-provided storage
1257 	 */
1258 	if (len > 0) {
1259 		mptcp_diag_fill_info(msk, &mfi.mptcp_info);
1260 		copylen += min_t(unsigned int, len, sizeof(struct mptcp_info));
1261 	}
1262 
1263 	mfi.size_tcpinfo_kernel = sizeof(struct tcp_info);
1264 	mfi.size_tcpinfo_user = min_t(unsigned int, mfi.size_tcpinfo_user,
1265 				      sizeof(struct tcp_info));
1266 	sfinfoptr = u64_to_user_ptr(mfi.subflow_info);
1267 	mfi.size_sfinfo_kernel = sizeof(struct mptcp_subflow_info);
1268 	mfi.size_sfinfo_user = min_t(unsigned int, mfi.size_sfinfo_user,
1269 				     sizeof(struct mptcp_subflow_info));
1270 	tcpinfoptr = u64_to_user_ptr(mfi.tcp_info);
1271 
1272 	lock_sock(sk);
1273 	mptcp_for_each_subflow(msk, subflow) {
1274 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1275 		struct mptcp_subflow_info sfinfo;
1276 		struct tcp_info tcp_info;
1277 
1278 		if (sfcount++ >= mfi.size_arrays_user)
1279 			continue;
1280 
1281 		/* fetch addr/tcp_info only if the user space buffers
1282 		 * are wide enough
1283 		 */
1284 		memset(&sfinfo, 0, sizeof(sfinfo));
1285 		sfinfo.id = subflow->subflow_id;
1286 		if (mfi.size_sfinfo_user >
1287 		    offsetof(struct mptcp_subflow_info, addrs))
1288 			mptcp_get_sub_addrs(ssk, &sfinfo.addrs);
1289 		if (copy_to_user(sfinfoptr, &sfinfo, mfi.size_sfinfo_user))
1290 			goto fail_release;
1291 
1292 		if (mfi.size_tcpinfo_user) {
1293 			tcp_get_info(ssk, &tcp_info);
1294 			if (copy_to_user(tcpinfoptr, &tcp_info,
1295 					 mfi.size_tcpinfo_user))
1296 				goto fail_release;
1297 		}
1298 
1299 		tcpinfoptr += mfi.size_tcpinfo_user;
1300 		sfinfoptr += mfi.size_sfinfo_user;
1301 	}
1302 	release_sock(sk);
1303 
1304 	mfi.num_subflows = sfcount;
1305 	if (mptcp_put_full_info(&mfi, optval, copylen, optlen))
1306 		return -EFAULT;
1307 
1308 	return 0;
1309 
1310 fail_release:
1311 	release_sock(sk);
1312 	return -EFAULT;
1313 }
1314 
mptcp_put_int_option(struct mptcp_sock * msk,char __user * optval,int __user * optlen,int val)1315 static int mptcp_put_int_option(struct mptcp_sock *msk, char __user *optval,
1316 				int __user *optlen, int val)
1317 {
1318 	int len;
1319 
1320 	if (get_user(len, optlen))
1321 		return -EFAULT;
1322 	if (len < 0)
1323 		return -EINVAL;
1324 
1325 	if (len < sizeof(int) && len > 0 && val >= 0 && val <= 255) {
1326 		unsigned char ucval = (unsigned char)val;
1327 
1328 		len = 1;
1329 		if (put_user(len, optlen))
1330 			return -EFAULT;
1331 		if (copy_to_user(optval, &ucval, 1))
1332 			return -EFAULT;
1333 	} else {
1334 		len = min_t(unsigned int, len, sizeof(int));
1335 		if (put_user(len, optlen))
1336 			return -EFAULT;
1337 		if (copy_to_user(optval, &val, len))
1338 			return -EFAULT;
1339 	}
1340 
1341 	return 0;
1342 }
1343 
mptcp_getsockopt_sol_tcp(struct mptcp_sock * msk,int optname,char __user * optval,int __user * optlen)1344 static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
1345 				    char __user *optval, int __user *optlen)
1346 {
1347 	struct sock *sk = (void *)msk;
1348 
1349 	switch (optname) {
1350 	case TCP_ULP:
1351 	case TCP_CONGESTION:
1352 	case TCP_INFO:
1353 	case TCP_CC_INFO:
1354 	case TCP_DEFER_ACCEPT:
1355 	case TCP_FASTOPEN:
1356 	case TCP_FASTOPEN_CONNECT:
1357 	case TCP_FASTOPEN_KEY:
1358 	case TCP_FASTOPEN_NO_COOKIE:
1359 		return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname,
1360 						      optval, optlen);
1361 	case TCP_INQ:
1362 		return mptcp_put_int_option(msk, optval, optlen, msk->recvmsg_inq);
1363 	case TCP_CORK:
1364 		return mptcp_put_int_option(msk, optval, optlen, msk->cork);
1365 	case TCP_NODELAY:
1366 		return mptcp_put_int_option(msk, optval, optlen, msk->nodelay);
1367 	case TCP_KEEPIDLE:
1368 		return mptcp_put_int_option(msk, optval, optlen,
1369 					    msk->keepalive_idle ? :
1370 					    READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_keepalive_time) / HZ);
1371 	case TCP_KEEPINTVL:
1372 		return mptcp_put_int_option(msk, optval, optlen,
1373 					    msk->keepalive_intvl ? :
1374 					    READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_keepalive_intvl) / HZ);
1375 	case TCP_KEEPCNT:
1376 		return mptcp_put_int_option(msk, optval, optlen,
1377 					    msk->keepalive_cnt ? :
1378 					    READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_keepalive_probes));
1379 	}
1380 	return -EOPNOTSUPP;
1381 }
1382 
mptcp_getsockopt_v4(struct mptcp_sock * msk,int optname,char __user * optval,int __user * optlen)1383 static int mptcp_getsockopt_v4(struct mptcp_sock *msk, int optname,
1384 			       char __user *optval, int __user *optlen)
1385 {
1386 	struct sock *sk = (void *)msk;
1387 
1388 	switch (optname) {
1389 	case IP_TOS:
1390 		return mptcp_put_int_option(msk, optval, optlen, inet_sk(sk)->tos);
1391 	}
1392 
1393 	return -EOPNOTSUPP;
1394 }
1395 
mptcp_getsockopt_sol_mptcp(struct mptcp_sock * msk,int optname,char __user * optval,int __user * optlen)1396 static int mptcp_getsockopt_sol_mptcp(struct mptcp_sock *msk, int optname,
1397 				      char __user *optval, int __user *optlen)
1398 {
1399 	switch (optname) {
1400 	case MPTCP_INFO:
1401 		return mptcp_getsockopt_info(msk, optval, optlen);
1402 	case MPTCP_FULL_INFO:
1403 		return mptcp_getsockopt_full_info(msk, optval, optlen);
1404 	case MPTCP_TCPINFO:
1405 		return mptcp_getsockopt_tcpinfo(msk, optval, optlen);
1406 	case MPTCP_SUBFLOW_ADDRS:
1407 		return mptcp_getsockopt_subflow_addrs(msk, optval, optlen);
1408 	}
1409 
1410 	return -EOPNOTSUPP;
1411 }
1412 
mptcp_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * option)1413 int mptcp_getsockopt(struct sock *sk, int level, int optname,
1414 		     char __user *optval, int __user *option)
1415 {
1416 	struct mptcp_sock *msk = mptcp_sk(sk);
1417 	struct sock *ssk;
1418 
1419 	pr_debug("msk=%p", msk);
1420 
1421 	/* @@ the meaning of setsockopt() when the socket is connected and
1422 	 * there are multiple subflows is not yet defined. It is up to the
1423 	 * MPTCP-level socket to configure the subflows until the subflow
1424 	 * is in TCP fallback, when socket options are passed through
1425 	 * to the one remaining subflow.
1426 	 */
1427 	lock_sock(sk);
1428 	ssk = __mptcp_tcp_fallback(msk);
1429 	release_sock(sk);
1430 	if (ssk)
1431 		return tcp_getsockopt(ssk, level, optname, optval, option);
1432 
1433 	if (level == SOL_IP)
1434 		return mptcp_getsockopt_v4(msk, optname, optval, option);
1435 	if (level == SOL_TCP)
1436 		return mptcp_getsockopt_sol_tcp(msk, optname, optval, option);
1437 	if (level == SOL_MPTCP)
1438 		return mptcp_getsockopt_sol_mptcp(msk, optname, optval, option);
1439 	return -EOPNOTSUPP;
1440 }
1441 
sync_socket_options(struct mptcp_sock * msk,struct sock * ssk)1442 static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
1443 {
1444 	static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK;
1445 	struct sock *sk = (struct sock *)msk;
1446 
1447 	if (ssk->sk_prot->keepalive) {
1448 		if (sock_flag(sk, SOCK_KEEPOPEN))
1449 			ssk->sk_prot->keepalive(ssk, 1);
1450 		else
1451 			ssk->sk_prot->keepalive(ssk, 0);
1452 	}
1453 
1454 	ssk->sk_priority = sk->sk_priority;
1455 	ssk->sk_bound_dev_if = sk->sk_bound_dev_if;
1456 	ssk->sk_incoming_cpu = sk->sk_incoming_cpu;
1457 	ssk->sk_ipv6only = sk->sk_ipv6only;
1458 	__ip_sock_set_tos(ssk, inet_sk(sk)->tos);
1459 
1460 	if (sk->sk_userlocks & tx_rx_locks) {
1461 		ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks;
1462 		if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) {
1463 			WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
1464 			mptcp_subflow_ctx(ssk)->cached_sndbuf = sk->sk_sndbuf;
1465 		}
1466 		if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1467 			WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
1468 	}
1469 
1470 	if (sock_flag(sk, SOCK_LINGER)) {
1471 		ssk->sk_lingertime = sk->sk_lingertime;
1472 		sock_set_flag(ssk, SOCK_LINGER);
1473 	} else {
1474 		sock_reset_flag(ssk, SOCK_LINGER);
1475 	}
1476 
1477 	if (sk->sk_mark != ssk->sk_mark) {
1478 		ssk->sk_mark = sk->sk_mark;
1479 		sk_dst_reset(ssk);
1480 	}
1481 
1482 	sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG));
1483 
1484 	if (inet_csk(sk)->icsk_ca_ops != inet_csk(ssk)->icsk_ca_ops)
1485 		tcp_set_congestion_control(ssk, msk->ca_name, false, true);
1486 	__tcp_sock_set_cork(ssk, !!msk->cork);
1487 	__tcp_sock_set_nodelay(ssk, !!msk->nodelay);
1488 	tcp_sock_set_keepidle_locked(ssk, msk->keepalive_idle);
1489 	tcp_sock_set_keepintvl(ssk, msk->keepalive_intvl);
1490 	tcp_sock_set_keepcnt(ssk, msk->keepalive_cnt);
1491 
1492 	inet_assign_bit(TRANSPARENT, ssk, inet_test_bit(TRANSPARENT, sk));
1493 	inet_assign_bit(FREEBIND, ssk, inet_test_bit(FREEBIND, sk));
1494 }
1495 
__mptcp_sockopt_sync(struct mptcp_sock * msk,struct sock * ssk)1496 static void __mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
1497 {
1498 	bool slow = lock_sock_fast(ssk);
1499 
1500 	sync_socket_options(msk, ssk);
1501 
1502 	unlock_sock_fast(ssk, slow);
1503 }
1504 
mptcp_sockopt_sync(struct mptcp_sock * msk,struct sock * ssk)1505 void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
1506 {
1507 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1508 
1509 	msk_owned_by_me(msk);
1510 
1511 	if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
1512 		__mptcp_sockopt_sync(msk, ssk);
1513 
1514 		subflow->setsockopt_seq = msk->setsockopt_seq;
1515 	}
1516 }
1517 
mptcp_sockopt_sync_locked(struct mptcp_sock * msk,struct sock * ssk)1518 void mptcp_sockopt_sync_locked(struct mptcp_sock *msk, struct sock *ssk)
1519 {
1520 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1521 
1522 	msk_owned_by_me(msk);
1523 
1524 	ssk->sk_rcvlowat = 0;
1525 
1526 	if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
1527 		sync_socket_options(msk, ssk);
1528 
1529 		subflow->setsockopt_seq = msk->setsockopt_seq;
1530 	}
1531 }
1532 
1533 /* unfortunately this is different enough from the tcp version so
1534  * that we can't factor it out
1535  */
mptcp_set_rcvlowat(struct sock * sk,int val)1536 int mptcp_set_rcvlowat(struct sock *sk, int val)
1537 {
1538 	struct mptcp_subflow_context *subflow;
1539 	int space, cap;
1540 
1541 	/* bpf can land here with a wrong sk type */
1542 	if (sk->sk_protocol == IPPROTO_TCP)
1543 		return -EINVAL;
1544 
1545 	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1546 		cap = sk->sk_rcvbuf >> 1;
1547 	else
1548 		cap = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]) >> 1;
1549 	val = min(val, cap);
1550 	WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1551 
1552 	/* Check if we need to signal EPOLLIN right now */
1553 	if (mptcp_epollin_ready(sk))
1554 		sk->sk_data_ready(sk);
1555 
1556 	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1557 		return 0;
1558 
1559 	space = __tcp_space_from_win(mptcp_sk(sk)->scaling_ratio, val);
1560 	if (space <= sk->sk_rcvbuf)
1561 		return 0;
1562 
1563 	/* propagate the rcvbuf changes to all the subflows */
1564 	WRITE_ONCE(sk->sk_rcvbuf, space);
1565 	mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
1566 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1567 		bool slow;
1568 
1569 		slow = lock_sock_fast(ssk);
1570 		WRITE_ONCE(ssk->sk_rcvbuf, space);
1571 		WRITE_ONCE(tcp_sk(ssk)->window_clamp, val);
1572 		unlock_sock_fast(ssk, slow);
1573 	}
1574 	return 0;
1575 }
1576