projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mptcp: sockopt: propagate timestamp request to subflows
[linux-block.git] / net / mptcp / sockopt.c
CommitLineData
0abdde82
PA		 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
17static 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
df00b087
FW		 27static 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
1b3e7ede
FW		 48static 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
55static 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
67static 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) {
a03c99b2
FW		 80 case SO_DEBUG:
81 sock_valbool_flag(ssk, SOCK_DBG, !!val);
82 break;
1b3e7ede
FW		 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;
5d0a6bc8
FW		 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;
36704413
FW		 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;
6f0d7198
FW		 107 case SO_INCOMING_CPU:
108 WRITE_ONCE(ssk->sk_incoming_cpu, val);
109 break;
1b3e7ede
FW		 110 }
111
112 subflow->setsockopt_seq = msk->setsockopt_seq;
113 unlock_sock_fast(ssk, slow);
114 }
115
116 release_sock(sk);
117}
118
119static 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
6f0d7198
FW		 134static 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
9061f24b
FW		 143static 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
1b3e7ede
FW		 180static 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;
a03c99b2 193 case SO_DEBUG:
36704413 194 case SO_MARK:
1b3e7ede 195 case SO_PRIORITY:
5d0a6bc8
FW		 196 case SO_SNDBUF:
197 case SO_SNDBUFFORCE:
198 case SO_RCVBUF:
199 case SO_RCVBUFFORCE:
1b3e7ede 200 return mptcp_sol_socket_intval(msk, optname, val);
6f0d7198
FW		 201 case SO_INCOMING_CPU:
202 mptcp_so_incoming_cpu(msk, val);
203 return 0;
9061f24b
FW		 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);
1b3e7ede
FW		 211 }
212
213 return -ENOPROTOOPT;
214}
215
268b1238
FW		 216static 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
0abdde82
PA		 257static 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:
5d0a6bc8
FW		 267 case SO_BINDTODEVICE:
268 case SO_BINDTOIFINDEX:
0abdde82
PA		 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;
5d0a6bc8
FW		 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;
0abdde82
PA		 286 }
287 release_sock(sk);
288 return ret;
1b3e7ede
FW		 289 case SO_KEEPALIVE:
290 case SO_PRIORITY:
5d0a6bc8
FW		 291 case SO_SNDBUF:
292 case SO_SNDBUFFORCE:
293 case SO_RCVBUF:
294 case SO_RCVBUFFORCE:
36704413 295 case SO_MARK:
6f0d7198 296 case SO_INCOMING_CPU:
a03c99b2 297 case SO_DEBUG:
9061f24b
FW		 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:
1b3e7ede 304 return mptcp_setsockopt_sol_socket_int(msk, optname, optval, optlen);
268b1238
FW		 305 case SO_LINGER:
306 return mptcp_setsockopt_sol_socket_linger(msk, optval, optlen);
a03c99b2
FW		 307 case SO_NO_CHECK:
308 case SO_DONTROUTE:
309 case SO_BROADCAST:
310 case SO_BSDCOMPAT:
311 case SO_PASSCRED:
312 case SO_PASSSEC:
313 case SO_RXQ_OVFL:
314 case SO_WIFI_STATUS:
315 case SO_NOFCS:
316 case SO_SELECT_ERR_QUEUE:
317 return 0;
0abdde82
PA		 318 }
319
320 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
321}
322
323static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
324 sockptr_t optval, unsigned int optlen)
325{
326 struct sock *sk = (struct sock *)msk;
327 int ret = -EOPNOTSUPP;
328 struct socket *ssock;
329
330 switch (optname) {
331 case IPV6_V6ONLY:
332 lock_sock(sk);
333 ssock = __mptcp_nmpc_socket(msk);
334 if (!ssock) {
335 release_sock(sk);
336 return -EINVAL;
337 }
338
339 ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen);
340 if (ret == 0)
341 sk->sk_ipv6only = ssock->sk->sk_ipv6only;
342
343 release_sock(sk);
344 break;
345 }
346
347 return ret;
348}
349
d9e4c129
PA		 350static bool mptcp_supported_sockopt(int level, int optname)
351{
352 if (level == SOL_SOCKET) {
353 switch (optname) {
354 case SO_DEBUG:
355 case SO_REUSEPORT:
356 case SO_REUSEADDR:
357
358 /* the following ones need a better implementation,
359 * but are quite common we want to preserve them
360 */
361 case SO_BINDTODEVICE:
362 case SO_SNDBUF:
363 case SO_SNDBUFFORCE:
364 case SO_RCVBUF:
365 case SO_RCVBUFFORCE:
366 case SO_KEEPALIVE:
367 case SO_PRIORITY:
368 case SO_LINGER:
369 case SO_TIMESTAMP_OLD:
370 case SO_TIMESTAMP_NEW:
371 case SO_TIMESTAMPNS_OLD:
372 case SO_TIMESTAMPNS_NEW:
373 case SO_TIMESTAMPING_OLD:
374 case SO_TIMESTAMPING_NEW:
375 case SO_RCVLOWAT:
376 case SO_RCVTIMEO_OLD:
377 case SO_RCVTIMEO_NEW:
378 case SO_SNDTIMEO_OLD:
379 case SO_SNDTIMEO_NEW:
380 case SO_MARK:
381 case SO_INCOMING_CPU:
382 case SO_BINDTOIFINDEX:
383 case SO_BUSY_POLL:
384 case SO_PREFER_BUSY_POLL:
385 case SO_BUSY_POLL_BUDGET:
386
387 /* next ones are no-op for plain TCP */
388 case SO_NO_CHECK:
389 case SO_DONTROUTE:
390 case SO_BROADCAST:
391 case SO_BSDCOMPAT:
392 case SO_PASSCRED:
393 case SO_PASSSEC:
394 case SO_RXQ_OVFL:
395 case SO_WIFI_STATUS:
396 case SO_NOFCS:
397 case SO_SELECT_ERR_QUEUE:
398 return true;
399 }
400
401 /* SO_OOBINLINE is not supported, let's avoid the related mess */
402 /* SO_ATTACH_FILTER, SO_ATTACH_BPF, SO_ATTACH_REUSEPORT_CBPF,
403 * SO_DETACH_REUSEPORT_BPF, SO_DETACH_FILTER, SO_LOCK_FILTER,
404 * we must be careful with subflows
405 */
406 /* SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks
407 * explicitly the sk_protocol field
408 */
409 /* SO_PEEK_OFF is unsupported, as it is for plain TCP */
410 /* SO_MAX_PACING_RATE is unsupported, we must be careful with subflows */
411 /* SO_CNX_ADVICE is currently unsupported, could possibly be relevant,
412 * but likely needs careful design
413 */
414 /* SO_ZEROCOPY is currently unsupported, TODO in sndmsg */
415 /* SO_TXTIME is currently unsupported */
416 return false;
417 }
418 if (level == SOL_IP) {
419 switch (optname) {
420 /* should work fine */
421 case IP_FREEBIND:
422 case IP_TRANSPARENT:
423
424 /* the following are control cmsg related */
425 case IP_PKTINFO:
426 case IP_RECVTTL:
427 case IP_RECVTOS:
428 case IP_RECVOPTS:
429 case IP_RETOPTS:
430 case IP_PASSSEC:
431 case IP_RECVORIGDSTADDR:
432 case IP_CHECKSUM:
433 case IP_RECVFRAGSIZE:
434
435 /* common stuff that need some love */
436 case IP_TOS:
437 case IP_TTL:
438 case IP_BIND_ADDRESS_NO_PORT:
439 case IP_MTU_DISCOVER:
440 case IP_RECVERR:
441
442 /* possibly less common may deserve some love */
443 case IP_MINTTL:
444
445 /* the following is apparently a no-op for plain TCP */
446 case IP_RECVERR_RFC4884:
447 return true;
448 }
449
450 /* IP_OPTIONS is not supported, needs subflow care */
451 /* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */
452 /* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF,
453 * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP,
454 * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE,
455 * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP,
456 * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE,
457 * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal
458 * with mcast stuff
459 */
460 /* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */
461 return false;
462 }
463 if (level == SOL_IPV6) {
464 switch (optname) {
465 case IPV6_V6ONLY:
466
467 /* the following are control cmsg related */
468 case IPV6_RECVPKTINFO:
469 case IPV6_2292PKTINFO:
470 case IPV6_RECVHOPLIMIT:
471 case IPV6_2292HOPLIMIT:
472 case IPV6_RECVRTHDR:
473 case IPV6_2292RTHDR:
474 case IPV6_RECVHOPOPTS:
475 case IPV6_2292HOPOPTS:
476 case IPV6_RECVDSTOPTS:
477 case IPV6_2292DSTOPTS:
478 case IPV6_RECVTCLASS:
479 case IPV6_FLOWINFO:
480 case IPV6_RECVPATHMTU:
481 case IPV6_RECVORIGDSTADDR:
482 case IPV6_RECVFRAGSIZE:
483
484 /* the following ones need some love but are quite common */
485 case IPV6_TCLASS:
486 case IPV6_TRANSPARENT:
487 case IPV6_FREEBIND:
488 case IPV6_PKTINFO:
489 case IPV6_2292PKTOPTIONS:
490 case IPV6_UNICAST_HOPS:
491 case IPV6_MTU_DISCOVER:
492 case IPV6_MTU:
493 case IPV6_RECVERR:
494 case IPV6_FLOWINFO_SEND:
495 case IPV6_FLOWLABEL_MGR:
496 case IPV6_MINHOPCOUNT:
497 case IPV6_DONTFRAG:
498 case IPV6_AUTOFLOWLABEL:
499
500 /* the following one is a no-op for plain TCP */
501 case IPV6_RECVERR_RFC4884:
502 return true;
503 }
504
505 /* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are
506 * not supported
507 */
508 /* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF,
509 * IPV6_MULTICAST_IF, IPV6_ADDRFORM,
510 * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST,
511 * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP,
512 * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP,
513 * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER
514 * are not supported better not deal with mcast
515 */
516 /* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */
517
518 /* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */
519 /* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */
520 return false;
521 }
522 if (level == SOL_TCP) {
523 switch (optname) {
524 /* the following are no-op or should work just fine */
525 case TCP_THIN_DUPACK:
526 case TCP_DEFER_ACCEPT:
527
528 /* the following need some love */
529 case TCP_MAXSEG:
530 case TCP_NODELAY:
531 case TCP_THIN_LINEAR_TIMEOUTS:
532 case TCP_CONGESTION:
533 case TCP_ULP:
534 case TCP_CORK:
535 case TCP_KEEPIDLE:
536 case TCP_KEEPINTVL:
537 case TCP_KEEPCNT:
538 case TCP_SYNCNT:
539 case TCP_SAVE_SYN:
540 case TCP_LINGER2:
541 case TCP_WINDOW_CLAMP:
542 case TCP_QUICKACK:
543 case TCP_USER_TIMEOUT:
544 case TCP_TIMESTAMP:
545 case TCP_NOTSENT_LOWAT:
546 case TCP_TX_DELAY:
547 return true;
548 }
549
550 /* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */
551
552 /* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS,
553 * TCP_REPAIR_WINDOW are not supported, better avoid this mess
554 */
555 /* TCP_FASTOPEN_KEY, TCP_FASTOPEN TCP_FASTOPEN_CONNECT, TCP_FASTOPEN_NO_COOKIE,
556 * are not supported fastopen is currently unsupported
557 */
558 /* TCP_INQ is currently unsupported, needs some recvmsg work */
559 }
560 return false;
561}
562
aa1fbd94
FW		 563static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval,
564 unsigned int optlen)
565{
566 struct mptcp_subflow_context *subflow;
567 struct sock *sk = (struct sock *)msk;
568 char name[TCP_CA_NAME_MAX];
569 bool cap_net_admin;
570 int ret;
571
572 if (optlen < 1)
573 return -EINVAL;
574
575 ret = strncpy_from_sockptr(name, optval,
576 min_t(long, TCP_CA_NAME_MAX - 1, optlen));
577 if (ret < 0)
578 return -EFAULT;
579
580 name[ret] = 0;
581
582 cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN);
583
584 ret = 0;
585 lock_sock(sk);
586 sockopt_seq_inc(msk);
587 mptcp_for_each_subflow(msk, subflow) {
588 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
589 int err;
590
591 lock_sock(ssk);
592 err = tcp_set_congestion_control(ssk, name, true, cap_net_admin);
593 if (err < 0 && ret == 0)
594 ret = err;
595 subflow->setsockopt_seq = msk->setsockopt_seq;
596 release_sock(ssk);
597 }
598
599 if (ret == 0)
20b5759f 600 strcpy(msk->ca_name, name);
aa1fbd94
FW		 601
602 release_sock(sk);
603 return ret;
604}
605
606static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
607 sockptr_t optval, unsigned int optlen)
608{
609 switch (optname) {
610 case TCP_ULP:
611 return -EOPNOTSUPP;
612 case TCP_CONGESTION:
613 return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen);
614 }
615
616 return -EOPNOTSUPP;
617}
618
0abdde82
PA		 619int mptcp_setsockopt(struct sock *sk, int level, int optname,
620 sockptr_t optval, unsigned int optlen)
621{
622 struct mptcp_sock *msk = mptcp_sk(sk);
623 struct sock *ssk;
624
625 pr_debug("msk=%p", msk);
626
d9e4c129
PA		 627 if (!mptcp_supported_sockopt(level, optname))
628 return -ENOPROTOOPT;
629
0abdde82
PA		 630 if (level == SOL_SOCKET)
631 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
632
633 /* @@ the meaning of setsockopt() when the socket is connected and
634 * there are multiple subflows is not yet defined. It is up to the
635 * MPTCP-level socket to configure the subflows until the subflow
636 * is in TCP fallback, when TCP socket options are passed through
637 * to the one remaining subflow.
638 */
639 lock_sock(sk);
640 ssk = __mptcp_tcp_fallback(msk);
641 release_sock(sk);
642 if (ssk)
643 return tcp_setsockopt(ssk, level, optname, optval, optlen);
644
645 if (level == SOL_IPV6)
646 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
647
aa1fbd94
FW		 648 if (level == SOL_TCP)
649 return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen);
650
651 return -EOPNOTSUPP;
652}
653
654static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
655 char __user *optval, int __user *optlen)
656{
657 struct sock *sk = (struct sock *)msk;
658 struct socket *ssock;
659 int ret = -EINVAL;
660 struct sock *ssk;
661
662 lock_sock(sk);
663 ssk = msk->first;
664 if (ssk) {
665 ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
666 goto out;
667 }
668
669 ssock = __mptcp_nmpc_socket(msk);
670 if (!ssock)
671 goto out;
672
673 ret = tcp_getsockopt(ssock->sk, level, optname, optval, optlen);
674
675out:
676 release_sock(sk);
677 return ret;
678}
679
680static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
681 char __user *optval, int __user *optlen)
682{
683 switch (optname) {
684 case TCP_ULP:
685 case TCP_CONGESTION:
686 case TCP_INFO:
687 case TCP_CC_INFO:
688 return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname,
689 optval, optlen);
690 }
0abdde82
PA		 691 return -EOPNOTSUPP;
692}
693
694int mptcp_getsockopt(struct sock *sk, int level, int optname,
695 char __user *optval, int __user *option)
696{
697 struct mptcp_sock *msk = mptcp_sk(sk);
698 struct sock *ssk;
699
700 pr_debug("msk=%p", msk);
701
702 /* @@ the meaning of setsockopt() when the socket is connected and
703 * there are multiple subflows is not yet defined. It is up to the
704 * MPTCP-level socket to configure the subflows until the subflow
705 * is in TCP fallback, when socket options are passed through
706 * to the one remaining subflow.
707 */
708 lock_sock(sk);
709 ssk = __mptcp_tcp_fallback(msk);
710 release_sock(sk);
711 if (ssk)
712 return tcp_getsockopt(ssk, level, optname, optval, option);
713
aa1fbd94
FW		 714 if (level == SOL_TCP)
715 return mptcp_getsockopt_sol_tcp(msk, optname, optval, option);
0abdde82
PA		 716 return -EOPNOTSUPP;
717}
718
1b3e7ede
FW		 719static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
720{
5d0a6bc8 721 static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK;
1b3e7ede
FW		 722 struct sock *sk = (struct sock *)msk;
723
724 if (ssk->sk_prot->keepalive) {
725 if (sock_flag(sk, SOCK_KEEPOPEN))
726 ssk->sk_prot->keepalive(ssk, 1);
727 else
728 ssk->sk_prot->keepalive(ssk, 0);
729 }
730
731 ssk->sk_priority = sk->sk_priority;
5d0a6bc8
FW		 732 ssk->sk_bound_dev_if = sk->sk_bound_dev_if;
733 ssk->sk_incoming_cpu = sk->sk_incoming_cpu;
734
735 if (sk->sk_userlocks & tx_rx_locks) {
736 ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks;
737 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
738 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
739 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
740 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
741 }
742
743 if (sock_flag(sk, SOCK_LINGER)) {
744 ssk->sk_lingertime = sk->sk_lingertime;
745 sock_set_flag(ssk, SOCK_LINGER);
746 } else {
747 sock_reset_flag(ssk, SOCK_LINGER);
748 }
749
750 if (sk->sk_mark != ssk->sk_mark) {
751 ssk->sk_mark = sk->sk_mark;
752 sk_dst_reset(ssk);
753 }
754
755 sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG));
756
757 if (inet_csk(sk)->icsk_ca_ops != inet_csk(ssk)->icsk_ca_ops)
20b5759f 758 tcp_set_congestion_control(ssk, msk->ca_name, false, true);
1b3e7ede
FW		 759}
760
df00b087
FW		 761static void __mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
762{
1b3e7ede
FW		 763 bool slow = lock_sock_fast(ssk);
764
765 sync_socket_options(msk, ssk);
766
767 unlock_sock_fast(ssk, slow);
df00b087
FW		 768}
769
78962489
FW		 770void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
771{
df00b087
FW		 772 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
773
78962489 774 msk_owned_by_me(msk);
df00b087
FW		 775
776 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
777 __mptcp_sockopt_sync(msk, ssk);
778
779 subflow->setsockopt_seq = msk->setsockopt_seq;
780 }
78962489
FW		 781}
782
783void mptcp_sockopt_sync_all(struct mptcp_sock *msk)
784{
785 struct mptcp_subflow_context *subflow;
df00b087
FW		 786 struct sock *sk = (struct sock *)msk;
787 u32 seq;
78962489 788
df00b087 789 seq = sockopt_seq_reset(sk);
78962489
FW		 790
791 mptcp_for_each_subflow(msk, subflow) {
792 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
df00b087 793 u32 sseq = READ_ONCE(subflow->setsockopt_seq);
78962489 794
df00b087
FW		 795 if (sseq != msk->setsockopt_seq) {
796 __mptcp_sockopt_sync(msk, ssk);
797 WRITE_ONCE(subflow->setsockopt_seq, seq);
798 } else if (sseq != seq) {
799 WRITE_ONCE(subflow->setsockopt_seq, seq);
800 }
78962489
FW		 801
802 cond_resched();
803 }
df00b087
FW		 804
805 msk->setsockopt_seq = seq;
78962489 806}
Linux 6.x block layer and io_uring tree(s)