xref: /openbmc/linux/net/mptcp/sockopt.c (revision 1d7a0395)
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 static struct sock *__mptcp_tcp_fallback(struct mptcp_sock *msk)
18 {
19 	sock_owned_by_me((const struct sock *)msk);
20 
21 	if (likely(!__mptcp_check_fallback(msk)))
22 		return NULL;
23 
24 	return msk->first;
25 }
26 
27 static u32 sockopt_seq_reset(const struct sock *sk)
28 {
29 	sock_owned_by_me(sk);
30 
31 	/* Highbits contain state.  Allows to distinguish sockopt_seq
32 	 * of listener and established:
33 	 * s0 = new_listener()
34 	 * sockopt(s0) - seq is 1
35 	 * s1 = accept(s0) - s1 inherits seq 1 if listener sk (s0)
36 	 * sockopt(s0) - seq increments to 2 on s0
37 	 * sockopt(s1) // seq increments to 2 on s1 (different option)
38 	 * new ssk completes join, inherits options from s0 // seq 2
39 	 * Needs sync from mptcp join logic, but ssk->seq == msk->seq
40 	 *
41 	 * Set High order bits to sk_state so ssk->seq == msk->seq test
42 	 * will fail.
43 	 */
44 
45 	return (u32)sk->sk_state << 24u;
46 }
47 
48 static void sockopt_seq_inc(struct mptcp_sock *msk)
49 {
50 	u32 seq = (msk->setsockopt_seq + 1) & 0x00ffffff;
51 
52 	msk->setsockopt_seq = sockopt_seq_reset((struct sock *)msk) + seq;
53 }
54 
55 static int mptcp_get_int_option(struct mptcp_sock *msk, sockptr_t optval,
56 				unsigned int optlen, int *val)
57 {
58 	if (optlen < sizeof(int))
59 		return -EINVAL;
60 
61 	if (copy_from_sockptr(val, optval, sizeof(*val)))
62 		return -EFAULT;
63 
64 	return 0;
65 }
66 
67 static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, int val)
68 {
69 	struct mptcp_subflow_context *subflow;
70 	struct sock *sk = (struct sock *)msk;
71 
72 	lock_sock(sk);
73 	sockopt_seq_inc(msk);
74 
75 	mptcp_for_each_subflow(msk, subflow) {
76 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
77 		bool slow = lock_sock_fast(ssk);
78 
79 		switch (optname) {
80 		case SO_DEBUG:
81 			sock_valbool_flag(ssk, SOCK_DBG, !!val);
82 			break;
83 		case SO_KEEPALIVE:
84 			if (ssk->sk_prot->keepalive)
85 				ssk->sk_prot->keepalive(ssk, !!val);
86 			sock_valbool_flag(ssk, SOCK_KEEPOPEN, !!val);
87 			break;
88 		case SO_PRIORITY:
89 			ssk->sk_priority = val;
90 			break;
91 		case SO_SNDBUF:
92 		case SO_SNDBUFFORCE:
93 			ssk->sk_userlocks |= SOCK_SNDBUF_LOCK;
94 			WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
95 			break;
96 		case SO_RCVBUF:
97 		case SO_RCVBUFFORCE:
98 			ssk->sk_userlocks |= SOCK_RCVBUF_LOCK;
99 			WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
100 			break;
101 		case SO_MARK:
102 			if (READ_ONCE(ssk->sk_mark) != sk->sk_mark) {
103 				ssk->sk_mark = sk->sk_mark;
104 				sk_dst_reset(ssk);
105 			}
106 			break;
107 		case SO_INCOMING_CPU:
108 			WRITE_ONCE(ssk->sk_incoming_cpu, val);
109 			break;
110 		}
111 
112 		subflow->setsockopt_seq = msk->setsockopt_seq;
113 		unlock_sock_fast(ssk, slow);
114 	}
115 
116 	release_sock(sk);
117 }
118 
119 static int mptcp_sol_socket_intval(struct mptcp_sock *msk, int optname, int val)
120 {
121 	sockptr_t optval = KERNEL_SOCKPTR(&val);
122 	struct sock *sk = (struct sock *)msk;
123 	int ret;
124 
125 	ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
126 			      optval, sizeof(val));
127 	if (ret)
128 		return ret;
129 
130 	mptcp_sol_socket_sync_intval(msk, optname, val);
131 	return 0;
132 }
133 
134 static void mptcp_so_incoming_cpu(struct mptcp_sock *msk, int val)
135 {
136 	struct sock *sk = (struct sock *)msk;
137 
138 	WRITE_ONCE(sk->sk_incoming_cpu, val);
139 
140 	mptcp_sol_socket_sync_intval(msk, SO_INCOMING_CPU, val);
141 }
142 
143 static int mptcp_setsockopt_sol_socket_tstamp(struct mptcp_sock *msk, int optname, int val)
144 {
145 	sockptr_t optval = KERNEL_SOCKPTR(&val);
146 	struct mptcp_subflow_context *subflow;
147 	struct sock *sk = (struct sock *)msk;
148 	int ret;
149 
150 	ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
151 			      optval, sizeof(val));
152 	if (ret)
153 		return ret;
154 
155 	lock_sock(sk);
156 	mptcp_for_each_subflow(msk, subflow) {
157 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
158 		bool slow = lock_sock_fast(ssk);
159 
160 		switch (optname) {
161 		case SO_TIMESTAMP_OLD:
162 		case SO_TIMESTAMP_NEW:
163 		case SO_TIMESTAMPNS_OLD:
164 		case SO_TIMESTAMPNS_NEW:
165 			sock_set_timestamp(sk, optname, !!val);
166 			break;
167 		case SO_TIMESTAMPING_NEW:
168 		case SO_TIMESTAMPING_OLD:
169 			sock_set_timestamping(sk, optname, val);
170 			break;
171 		}
172 
173 		unlock_sock_fast(ssk, slow);
174 	}
175 
176 	release_sock(sk);
177 	return 0;
178 }
179 
180 static int mptcp_setsockopt_sol_socket_int(struct mptcp_sock *msk, int optname,
181 					   sockptr_t optval, unsigned int optlen)
182 {
183 	int val, ret;
184 
185 	ret = mptcp_get_int_option(msk, optval, optlen, &val);
186 	if (ret)
187 		return ret;
188 
189 	switch (optname) {
190 	case SO_KEEPALIVE:
191 		mptcp_sol_socket_sync_intval(msk, optname, val);
192 		return 0;
193 	case SO_DEBUG:
194 	case SO_MARK:
195 	case SO_PRIORITY:
196 	case SO_SNDBUF:
197 	case SO_SNDBUFFORCE:
198 	case SO_RCVBUF:
199 	case SO_RCVBUFFORCE:
200 		return mptcp_sol_socket_intval(msk, optname, val);
201 	case SO_INCOMING_CPU:
202 		mptcp_so_incoming_cpu(msk, val);
203 		return 0;
204 	case SO_TIMESTAMP_OLD:
205 	case SO_TIMESTAMP_NEW:
206 	case SO_TIMESTAMPNS_OLD:
207 	case SO_TIMESTAMPNS_NEW:
208 	case SO_TIMESTAMPING_OLD:
209 	case SO_TIMESTAMPING_NEW:
210 		return mptcp_setsockopt_sol_socket_tstamp(msk, optname, val);
211 	}
212 
213 	return -ENOPROTOOPT;
214 }
215 
216 static int mptcp_setsockopt_sol_socket_linger(struct mptcp_sock *msk, sockptr_t optval,
217 					      unsigned int optlen)
218 {
219 	struct mptcp_subflow_context *subflow;
220 	struct sock *sk = (struct sock *)msk;
221 	struct linger ling;
222 	sockptr_t kopt;
223 	int ret;
224 
225 	if (optlen < sizeof(ling))
226 		return -EINVAL;
227 
228 	if (copy_from_sockptr(&ling, optval, sizeof(ling)))
229 		return -EFAULT;
230 
231 	kopt = KERNEL_SOCKPTR(&ling);
232 	ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, SO_LINGER, kopt, sizeof(ling));
233 	if (ret)
234 		return ret;
235 
236 	lock_sock(sk);
237 	sockopt_seq_inc(msk);
238 	mptcp_for_each_subflow(msk, subflow) {
239 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
240 		bool slow = lock_sock_fast(ssk);
241 
242 		if (!ling.l_onoff) {
243 			sock_reset_flag(ssk, SOCK_LINGER);
244 		} else {
245 			ssk->sk_lingertime = sk->sk_lingertime;
246 			sock_set_flag(ssk, SOCK_LINGER);
247 		}
248 
249 		subflow->setsockopt_seq = msk->setsockopt_seq;
250 		unlock_sock_fast(ssk, slow);
251 	}
252 
253 	release_sock(sk);
254 	return 0;
255 }
256 
257 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
258 				       sockptr_t optval, unsigned int optlen)
259 {
260 	struct sock *sk = (struct sock *)msk;
261 	struct socket *ssock;
262 	int ret;
263 
264 	switch (optname) {
265 	case SO_REUSEPORT:
266 	case SO_REUSEADDR:
267 	case SO_BINDTODEVICE:
268 	case SO_BINDTOIFINDEX:
269 		lock_sock(sk);
270 		ssock = __mptcp_nmpc_socket(msk);
271 		if (!ssock) {
272 			release_sock(sk);
273 			return -EINVAL;
274 		}
275 
276 		ret = sock_setsockopt(ssock, SOL_SOCKET, optname, optval, optlen);
277 		if (ret == 0) {
278 			if (optname == SO_REUSEPORT)
279 				sk->sk_reuseport = ssock->sk->sk_reuseport;
280 			else if (optname == SO_REUSEADDR)
281 				sk->sk_reuse = ssock->sk->sk_reuse;
282 			else if (optname == SO_BINDTODEVICE)
283 				sk->sk_bound_dev_if = ssock->sk->sk_bound_dev_if;
284 			else if (optname == SO_BINDTOIFINDEX)
285 				sk->sk_bound_dev_if = ssock->sk->sk_bound_dev_if;
286 		}
287 		release_sock(sk);
288 		return ret;
289 	case SO_KEEPALIVE:
290 	case SO_PRIORITY:
291 	case SO_SNDBUF:
292 	case SO_SNDBUFFORCE:
293 	case SO_RCVBUF:
294 	case SO_RCVBUFFORCE:
295 	case SO_MARK:
296 	case SO_INCOMING_CPU:
297 	case SO_DEBUG:
298 	case SO_TIMESTAMP_OLD:
299 	case SO_TIMESTAMP_NEW:
300 	case SO_TIMESTAMPNS_OLD:
301 	case SO_TIMESTAMPNS_NEW:
302 	case SO_TIMESTAMPING_OLD:
303 	case SO_TIMESTAMPING_NEW:
304 		return mptcp_setsockopt_sol_socket_int(msk, optname, optval, optlen);
305 	case SO_LINGER:
306 		return mptcp_setsockopt_sol_socket_linger(msk, optval, optlen);
307 	case SO_RCVLOWAT:
308 	case SO_RCVTIMEO_OLD:
309 	case SO_RCVTIMEO_NEW:
310 	case SO_BUSY_POLL:
311 	case SO_PREFER_BUSY_POLL:
312 	case SO_BUSY_POLL_BUDGET:
313 		/* No need to copy: only relevant for msk */
314 		return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
315 	case SO_NO_CHECK:
316 	case SO_DONTROUTE:
317 	case SO_BROADCAST:
318 	case SO_BSDCOMPAT:
319 	case SO_PASSCRED:
320 	case SO_PASSSEC:
321 	case SO_RXQ_OVFL:
322 	case SO_WIFI_STATUS:
323 	case SO_NOFCS:
324 	case SO_SELECT_ERR_QUEUE:
325 		return 0;
326 	}
327 
328 	/* SO_OOBINLINE is not supported, let's avoid the related mess
329 	 * SO_ATTACH_FILTER, SO_ATTACH_BPF, SO_ATTACH_REUSEPORT_CBPF,
330 	 * SO_DETACH_REUSEPORT_BPF, SO_DETACH_FILTER, SO_LOCK_FILTER,
331 	 * we must be careful with subflows
332 	 *
333 	 * SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks
334 	 * explicitly the sk_protocol field
335 	 *
336 	 * SO_PEEK_OFF is unsupported, as it is for plain TCP
337 	 * SO_MAX_PACING_RATE is unsupported, we must be careful with subflows
338 	 * SO_CNX_ADVICE is currently unsupported, could possibly be relevant,
339 	 * but likely needs careful design
340 	 *
341 	 * SO_ZEROCOPY is currently unsupported, TODO in sndmsg
342 	 * SO_TXTIME is currently unsupported
343 	 */
344 
345 	return -EOPNOTSUPP;
346 }
347 
348 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
349 			       sockptr_t optval, unsigned int optlen)
350 {
351 	struct sock *sk = (struct sock *)msk;
352 	int ret = -EOPNOTSUPP;
353 	struct socket *ssock;
354 
355 	switch (optname) {
356 	case IPV6_V6ONLY:
357 		lock_sock(sk);
358 		ssock = __mptcp_nmpc_socket(msk);
359 		if (!ssock) {
360 			release_sock(sk);
361 			return -EINVAL;
362 		}
363 
364 		ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen);
365 		if (ret == 0)
366 			sk->sk_ipv6only = ssock->sk->sk_ipv6only;
367 
368 		release_sock(sk);
369 		break;
370 	}
371 
372 	return ret;
373 }
374 
375 static bool mptcp_supported_sockopt(int level, int optname)
376 {
377 	if (level == SOL_IP) {
378 		switch (optname) {
379 		/* should work fine */
380 		case IP_FREEBIND:
381 		case IP_TRANSPARENT:
382 
383 		/* the following are control cmsg related */
384 		case IP_PKTINFO:
385 		case IP_RECVTTL:
386 		case IP_RECVTOS:
387 		case IP_RECVOPTS:
388 		case IP_RETOPTS:
389 		case IP_PASSSEC:
390 		case IP_RECVORIGDSTADDR:
391 		case IP_CHECKSUM:
392 		case IP_RECVFRAGSIZE:
393 
394 		/* common stuff that need some love */
395 		case IP_TOS:
396 		case IP_TTL:
397 		case IP_BIND_ADDRESS_NO_PORT:
398 		case IP_MTU_DISCOVER:
399 		case IP_RECVERR:
400 
401 		/* possibly less common may deserve some love */
402 		case IP_MINTTL:
403 
404 		/* the following is apparently a no-op for plain TCP */
405 		case IP_RECVERR_RFC4884:
406 			return true;
407 		}
408 
409 		/* IP_OPTIONS is not supported, needs subflow care */
410 		/* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */
411 		/* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF,
412 		 * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP,
413 		 * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE,
414 		 * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP,
415 		 * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE,
416 		 * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal
417 		 * with mcast stuff
418 		 */
419 		/* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */
420 		return false;
421 	}
422 	if (level == SOL_IPV6) {
423 		switch (optname) {
424 		case IPV6_V6ONLY:
425 
426 		/* the following are control cmsg related */
427 		case IPV6_RECVPKTINFO:
428 		case IPV6_2292PKTINFO:
429 		case IPV6_RECVHOPLIMIT:
430 		case IPV6_2292HOPLIMIT:
431 		case IPV6_RECVRTHDR:
432 		case IPV6_2292RTHDR:
433 		case IPV6_RECVHOPOPTS:
434 		case IPV6_2292HOPOPTS:
435 		case IPV6_RECVDSTOPTS:
436 		case IPV6_2292DSTOPTS:
437 		case IPV6_RECVTCLASS:
438 		case IPV6_FLOWINFO:
439 		case IPV6_RECVPATHMTU:
440 		case IPV6_RECVORIGDSTADDR:
441 		case IPV6_RECVFRAGSIZE:
442 
443 		/* the following ones need some love but are quite common */
444 		case IPV6_TCLASS:
445 		case IPV6_TRANSPARENT:
446 		case IPV6_FREEBIND:
447 		case IPV6_PKTINFO:
448 		case IPV6_2292PKTOPTIONS:
449 		case IPV6_UNICAST_HOPS:
450 		case IPV6_MTU_DISCOVER:
451 		case IPV6_MTU:
452 		case IPV6_RECVERR:
453 		case IPV6_FLOWINFO_SEND:
454 		case IPV6_FLOWLABEL_MGR:
455 		case IPV6_MINHOPCOUNT:
456 		case IPV6_DONTFRAG:
457 		case IPV6_AUTOFLOWLABEL:
458 
459 		/* the following one is a no-op for plain TCP */
460 		case IPV6_RECVERR_RFC4884:
461 			return true;
462 		}
463 
464 		/* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are
465 		 * not supported
466 		 */
467 		/* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF,
468 		 * IPV6_MULTICAST_IF, IPV6_ADDRFORM,
469 		 * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST,
470 		 * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP,
471 		 * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP,
472 		 * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER
473 		 * are not supported better not deal with mcast
474 		 */
475 		/* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */
476 
477 		/* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */
478 		/* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */
479 		return false;
480 	}
481 	if (level == SOL_TCP) {
482 		switch (optname) {
483 		/* the following are no-op or should work just fine */
484 		case TCP_THIN_DUPACK:
485 		case TCP_DEFER_ACCEPT:
486 
487 		/* the following need some love */
488 		case TCP_MAXSEG:
489 		case TCP_NODELAY:
490 		case TCP_THIN_LINEAR_TIMEOUTS:
491 		case TCP_CONGESTION:
492 		case TCP_ULP:
493 		case TCP_CORK:
494 		case TCP_KEEPIDLE:
495 		case TCP_KEEPINTVL:
496 		case TCP_KEEPCNT:
497 		case TCP_SYNCNT:
498 		case TCP_SAVE_SYN:
499 		case TCP_LINGER2:
500 		case TCP_WINDOW_CLAMP:
501 		case TCP_QUICKACK:
502 		case TCP_USER_TIMEOUT:
503 		case TCP_TIMESTAMP:
504 		case TCP_NOTSENT_LOWAT:
505 		case TCP_TX_DELAY:
506 			return true;
507 		}
508 
509 		/* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */
510 
511 		/* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS,
512 		 * TCP_REPAIR_WINDOW are not supported, better avoid this mess
513 		 */
514 		/* TCP_FASTOPEN_KEY, TCP_FASTOPEN TCP_FASTOPEN_CONNECT, TCP_FASTOPEN_NO_COOKIE,
515 		 * are not supported fastopen is currently unsupported
516 		 */
517 		/* TCP_INQ is currently unsupported, needs some recvmsg work */
518 	}
519 	return false;
520 }
521 
522 static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval,
523 					       unsigned int optlen)
524 {
525 	struct mptcp_subflow_context *subflow;
526 	struct sock *sk = (struct sock *)msk;
527 	char name[TCP_CA_NAME_MAX];
528 	bool cap_net_admin;
529 	int ret;
530 
531 	if (optlen < 1)
532 		return -EINVAL;
533 
534 	ret = strncpy_from_sockptr(name, optval,
535 				   min_t(long, TCP_CA_NAME_MAX - 1, optlen));
536 	if (ret < 0)
537 		return -EFAULT;
538 
539 	name[ret] = 0;
540 
541 	cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN);
542 
543 	ret = 0;
544 	lock_sock(sk);
545 	sockopt_seq_inc(msk);
546 	mptcp_for_each_subflow(msk, subflow) {
547 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
548 		int err;
549 
550 		lock_sock(ssk);
551 		err = tcp_set_congestion_control(ssk, name, true, cap_net_admin);
552 		if (err < 0 && ret == 0)
553 			ret = err;
554 		subflow->setsockopt_seq = msk->setsockopt_seq;
555 		release_sock(ssk);
556 	}
557 
558 	if (ret == 0)
559 		strcpy(msk->ca_name, name);
560 
561 	release_sock(sk);
562 	return ret;
563 }
564 
565 static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
566 				    sockptr_t optval, unsigned int optlen)
567 {
568 	switch (optname) {
569 	case TCP_ULP:
570 		return -EOPNOTSUPP;
571 	case TCP_CONGESTION:
572 		return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen);
573 	}
574 
575 	return -EOPNOTSUPP;
576 }
577 
578 int mptcp_setsockopt(struct sock *sk, int level, int optname,
579 		     sockptr_t optval, unsigned int optlen)
580 {
581 	struct mptcp_sock *msk = mptcp_sk(sk);
582 	struct sock *ssk;
583 
584 	pr_debug("msk=%p", msk);
585 
586 	if (level == SOL_SOCKET)
587 		return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
588 
589 	if (!mptcp_supported_sockopt(level, optname))
590 		return -ENOPROTOOPT;
591 
592 	/* @@ the meaning of setsockopt() when the socket is connected and
593 	 * there are multiple subflows is not yet defined. It is up to the
594 	 * MPTCP-level socket to configure the subflows until the subflow
595 	 * is in TCP fallback, when TCP socket options are passed through
596 	 * to the one remaining subflow.
597 	 */
598 	lock_sock(sk);
599 	ssk = __mptcp_tcp_fallback(msk);
600 	release_sock(sk);
601 	if (ssk)
602 		return tcp_setsockopt(ssk, level, optname, optval, optlen);
603 
604 	if (level == SOL_IPV6)
605 		return mptcp_setsockopt_v6(msk, optname, optval, optlen);
606 
607 	if (level == SOL_TCP)
608 		return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen);
609 
610 	return -EOPNOTSUPP;
611 }
612 
613 static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
614 					  char __user *optval, int __user *optlen)
615 {
616 	struct sock *sk = (struct sock *)msk;
617 	struct socket *ssock;
618 	int ret = -EINVAL;
619 	struct sock *ssk;
620 
621 	lock_sock(sk);
622 	ssk = msk->first;
623 	if (ssk) {
624 		ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
625 		goto out;
626 	}
627 
628 	ssock = __mptcp_nmpc_socket(msk);
629 	if (!ssock)
630 		goto out;
631 
632 	ret = tcp_getsockopt(ssock->sk, level, optname, optval, optlen);
633 
634 out:
635 	release_sock(sk);
636 	return ret;
637 }
638 
639 static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
640 				    char __user *optval, int __user *optlen)
641 {
642 	switch (optname) {
643 	case TCP_ULP:
644 	case TCP_CONGESTION:
645 	case TCP_INFO:
646 	case TCP_CC_INFO:
647 		return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname,
648 						      optval, optlen);
649 	}
650 	return -EOPNOTSUPP;
651 }
652 
653 int mptcp_getsockopt(struct sock *sk, int level, int optname,
654 		     char __user *optval, int __user *option)
655 {
656 	struct mptcp_sock *msk = mptcp_sk(sk);
657 	struct sock *ssk;
658 
659 	pr_debug("msk=%p", msk);
660 
661 	/* @@ the meaning of setsockopt() when the socket is connected and
662 	 * there are multiple subflows is not yet defined. It is up to the
663 	 * MPTCP-level socket to configure the subflows until the subflow
664 	 * is in TCP fallback, when socket options are passed through
665 	 * to the one remaining subflow.
666 	 */
667 	lock_sock(sk);
668 	ssk = __mptcp_tcp_fallback(msk);
669 	release_sock(sk);
670 	if (ssk)
671 		return tcp_getsockopt(ssk, level, optname, optval, option);
672 
673 	if (level == SOL_TCP)
674 		return mptcp_getsockopt_sol_tcp(msk, optname, optval, option);
675 	return -EOPNOTSUPP;
676 }
677 
678 static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
679 {
680 	static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK;
681 	struct sock *sk = (struct sock *)msk;
682 
683 	if (ssk->sk_prot->keepalive) {
684 		if (sock_flag(sk, SOCK_KEEPOPEN))
685 			ssk->sk_prot->keepalive(ssk, 1);
686 		else
687 			ssk->sk_prot->keepalive(ssk, 0);
688 	}
689 
690 	ssk->sk_priority = sk->sk_priority;
691 	ssk->sk_bound_dev_if = sk->sk_bound_dev_if;
692 	ssk->sk_incoming_cpu = sk->sk_incoming_cpu;
693 
694 	if (sk->sk_userlocks & tx_rx_locks) {
695 		ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks;
696 		if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
697 			WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
698 		if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
699 			WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
700 	}
701 
702 	if (sock_flag(sk, SOCK_LINGER)) {
703 		ssk->sk_lingertime = sk->sk_lingertime;
704 		sock_set_flag(ssk, SOCK_LINGER);
705 	} else {
706 		sock_reset_flag(ssk, SOCK_LINGER);
707 	}
708 
709 	if (sk->sk_mark != ssk->sk_mark) {
710 		ssk->sk_mark = sk->sk_mark;
711 		sk_dst_reset(ssk);
712 	}
713 
714 	sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG));
715 
716 	if (inet_csk(sk)->icsk_ca_ops != inet_csk(ssk)->icsk_ca_ops)
717 		tcp_set_congestion_control(ssk, msk->ca_name, false, true);
718 }
719 
720 static void __mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
721 {
722 	bool slow = lock_sock_fast(ssk);
723 
724 	sync_socket_options(msk, ssk);
725 
726 	unlock_sock_fast(ssk, slow);
727 }
728 
729 void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
730 {
731 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
732 
733 	msk_owned_by_me(msk);
734 
735 	if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
736 		__mptcp_sockopt_sync(msk, ssk);
737 
738 		subflow->setsockopt_seq = msk->setsockopt_seq;
739 	}
740 }
741 
742 void mptcp_sockopt_sync_all(struct mptcp_sock *msk)
743 {
744 	struct mptcp_subflow_context *subflow;
745 	struct sock *sk = (struct sock *)msk;
746 	u32 seq;
747 
748 	seq = sockopt_seq_reset(sk);
749 
750 	mptcp_for_each_subflow(msk, subflow) {
751 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
752 		u32 sseq = READ_ONCE(subflow->setsockopt_seq);
753 
754 		if (sseq != msk->setsockopt_seq) {
755 			__mptcp_sockopt_sync(msk, ssk);
756 			WRITE_ONCE(subflow->setsockopt_seq, seq);
757 		} else if (sseq != seq) {
758 			WRITE_ONCE(subflow->setsockopt_seq, seq);
759 		}
760 
761 		cond_resched();
762 	}
763 
764 	msk->setsockopt_seq = seq;
765 }
766