1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (c) 2021, Red Hat.
7 #define pr_fmt(fmt) "MPTCP: " fmt
9 #include <linux/kernel.h>
10 #include <linux/module.h>
12 #include <net/protocol.h>
14 #include <net/mptcp.h>
17 static struct sock *__mptcp_tcp_fallback(struct mptcp_sock *msk)
19 sock_owned_by_me((const struct sock *)msk);
21 if (likely(!__mptcp_check_fallback(msk)))
27 static u32 sockopt_seq_reset(const struct sock *sk)
31 /* Highbits contain state. Allows to distinguish sockopt_seq
32 * of listener and established:
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
41 * Set High order bits to sk_state so ssk->seq == msk->seq test
45 return (u32)sk->sk_state << 24u;
48 static void sockopt_seq_inc(struct mptcp_sock *msk)
50 u32 seq = (msk->setsockopt_seq + 1) & 0x00ffffff;
52 msk->setsockopt_seq = sockopt_seq_reset((struct sock *)msk) + seq;
55 static int mptcp_get_int_option(struct mptcp_sock *msk, sockptr_t optval,
56 unsigned int optlen, int *val)
58 if (optlen < sizeof(int))
61 if (copy_from_sockptr(val, optval, sizeof(*val)))
67 static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, int val)
69 struct mptcp_subflow_context *subflow;
70 struct sock *sk = (struct sock *)msk;
75 mptcp_for_each_subflow(msk, subflow) {
76 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
77 bool slow = lock_sock_fast(ssk);
81 sock_valbool_flag(ssk, SOCK_DBG, !!val);
84 if (ssk->sk_prot->keepalive)
85 ssk->sk_prot->keepalive(ssk, !!val);
86 sock_valbool_flag(ssk, SOCK_KEEPOPEN, !!val);
89 ssk->sk_priority = val;
93 ssk->sk_userlocks |= SOCK_SNDBUF_LOCK;
94 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
98 ssk->sk_userlocks |= SOCK_RCVBUF_LOCK;
99 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
102 if (READ_ONCE(ssk->sk_mark) != sk->sk_mark) {
103 ssk->sk_mark = sk->sk_mark;
107 case SO_INCOMING_CPU:
108 WRITE_ONCE(ssk->sk_incoming_cpu, val);
112 subflow->setsockopt_seq = msk->setsockopt_seq;
113 unlock_sock_fast(ssk, slow);
119 static int mptcp_sol_socket_intval(struct mptcp_sock *msk, int optname, int val)
121 sockptr_t optval = KERNEL_SOCKPTR(&val);
122 struct sock *sk = (struct sock *)msk;
125 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
126 optval, sizeof(val));
130 mptcp_sol_socket_sync_intval(msk, optname, val);
134 static void mptcp_so_incoming_cpu(struct mptcp_sock *msk, int val)
136 struct sock *sk = (struct sock *)msk;
138 WRITE_ONCE(sk->sk_incoming_cpu, val);
140 mptcp_sol_socket_sync_intval(msk, SO_INCOMING_CPU, val);
143 static int mptcp_setsockopt_sol_socket_int(struct mptcp_sock *msk, int optname,
144 sockptr_t optval, unsigned int optlen)
148 ret = mptcp_get_int_option(msk, optval, optlen, &val);
154 mptcp_sol_socket_sync_intval(msk, optname, val);
163 return mptcp_sol_socket_intval(msk, optname, val);
164 case SO_INCOMING_CPU:
165 mptcp_so_incoming_cpu(msk, val);
172 static int mptcp_setsockopt_sol_socket_linger(struct mptcp_sock *msk, sockptr_t optval,
175 struct mptcp_subflow_context *subflow;
176 struct sock *sk = (struct sock *)msk;
181 if (optlen < sizeof(ling))
184 if (copy_from_sockptr(&ling, optval, sizeof(ling)))
187 kopt = KERNEL_SOCKPTR(&ling);
188 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, SO_LINGER, kopt, sizeof(ling));
193 sockopt_seq_inc(msk);
194 mptcp_for_each_subflow(msk, subflow) {
195 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
196 bool slow = lock_sock_fast(ssk);
199 sock_reset_flag(ssk, SOCK_LINGER);
201 ssk->sk_lingertime = sk->sk_lingertime;
202 sock_set_flag(ssk, SOCK_LINGER);
205 subflow->setsockopt_seq = msk->setsockopt_seq;
206 unlock_sock_fast(ssk, slow);
213 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
214 sockptr_t optval, unsigned int optlen)
216 struct sock *sk = (struct sock *)msk;
217 struct socket *ssock;
223 case SO_BINDTODEVICE:
224 case SO_BINDTOIFINDEX:
226 ssock = __mptcp_nmpc_socket(msk);
232 ret = sock_setsockopt(ssock, SOL_SOCKET, optname, optval, optlen);
234 if (optname == SO_REUSEPORT)
235 sk->sk_reuseport = ssock->sk->sk_reuseport;
236 else if (optname == SO_REUSEADDR)
237 sk->sk_reuse = ssock->sk->sk_reuse;
238 else if (optname == SO_BINDTODEVICE)
239 sk->sk_bound_dev_if = ssock->sk->sk_bound_dev_if;
240 else if (optname == SO_BINDTOIFINDEX)
241 sk->sk_bound_dev_if = ssock->sk->sk_bound_dev_if;
252 case SO_INCOMING_CPU:
254 return mptcp_setsockopt_sol_socket_int(msk, optname, optval, optlen);
256 return mptcp_setsockopt_sol_socket_linger(msk, optval, optlen);
266 case SO_SELECT_ERR_QUEUE:
270 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
273 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
274 sockptr_t optval, unsigned int optlen)
276 struct sock *sk = (struct sock *)msk;
277 int ret = -EOPNOTSUPP;
278 struct socket *ssock;
283 ssock = __mptcp_nmpc_socket(msk);
289 ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen);
291 sk->sk_ipv6only = ssock->sk->sk_ipv6only;
300 static bool mptcp_supported_sockopt(int level, int optname)
302 if (level == SOL_SOCKET) {
308 /* the following ones need a better implementation,
309 * but are quite common we want to preserve them
311 case SO_BINDTODEVICE:
319 case SO_TIMESTAMP_OLD:
320 case SO_TIMESTAMP_NEW:
321 case SO_TIMESTAMPNS_OLD:
322 case SO_TIMESTAMPNS_NEW:
323 case SO_TIMESTAMPING_OLD:
324 case SO_TIMESTAMPING_NEW:
326 case SO_RCVTIMEO_OLD:
327 case SO_RCVTIMEO_NEW:
328 case SO_SNDTIMEO_OLD:
329 case SO_SNDTIMEO_NEW:
331 case SO_INCOMING_CPU:
332 case SO_BINDTOIFINDEX:
334 case SO_PREFER_BUSY_POLL:
335 case SO_BUSY_POLL_BUDGET:
337 /* next ones are no-op for plain TCP */
347 case SO_SELECT_ERR_QUEUE:
351 /* SO_OOBINLINE is not supported, let's avoid the related mess */
352 /* SO_ATTACH_FILTER, SO_ATTACH_BPF, SO_ATTACH_REUSEPORT_CBPF,
353 * SO_DETACH_REUSEPORT_BPF, SO_DETACH_FILTER, SO_LOCK_FILTER,
354 * we must be careful with subflows
356 /* SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks
357 * explicitly the sk_protocol field
359 /* SO_PEEK_OFF is unsupported, as it is for plain TCP */
360 /* SO_MAX_PACING_RATE is unsupported, we must be careful with subflows */
361 /* SO_CNX_ADVICE is currently unsupported, could possibly be relevant,
362 * but likely needs careful design
364 /* SO_ZEROCOPY is currently unsupported, TODO in sndmsg */
365 /* SO_TXTIME is currently unsupported */
368 if (level == SOL_IP) {
370 /* should work fine */
374 /* the following are control cmsg related */
381 case IP_RECVORIGDSTADDR:
383 case IP_RECVFRAGSIZE:
385 /* common stuff that need some love */
388 case IP_BIND_ADDRESS_NO_PORT:
389 case IP_MTU_DISCOVER:
392 /* possibly less common may deserve some love */
395 /* the following is apparently a no-op for plain TCP */
396 case IP_RECVERR_RFC4884:
400 /* IP_OPTIONS is not supported, needs subflow care */
401 /* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */
402 /* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF,
403 * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP,
404 * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE,
405 * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP,
406 * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE,
407 * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal
410 /* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */
413 if (level == SOL_IPV6) {
417 /* the following are control cmsg related */
418 case IPV6_RECVPKTINFO:
419 case IPV6_2292PKTINFO:
420 case IPV6_RECVHOPLIMIT:
421 case IPV6_2292HOPLIMIT:
424 case IPV6_RECVHOPOPTS:
425 case IPV6_2292HOPOPTS:
426 case IPV6_RECVDSTOPTS:
427 case IPV6_2292DSTOPTS:
428 case IPV6_RECVTCLASS:
430 case IPV6_RECVPATHMTU:
431 case IPV6_RECVORIGDSTADDR:
432 case IPV6_RECVFRAGSIZE:
434 /* the following ones need some love but are quite common */
436 case IPV6_TRANSPARENT:
439 case IPV6_2292PKTOPTIONS:
440 case IPV6_UNICAST_HOPS:
441 case IPV6_MTU_DISCOVER:
444 case IPV6_FLOWINFO_SEND:
445 case IPV6_FLOWLABEL_MGR:
446 case IPV6_MINHOPCOUNT:
448 case IPV6_AUTOFLOWLABEL:
450 /* the following one is a no-op for plain TCP */
451 case IPV6_RECVERR_RFC4884:
455 /* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are
458 /* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF,
459 * IPV6_MULTICAST_IF, IPV6_ADDRFORM,
460 * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST,
461 * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP,
462 * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP,
463 * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER
464 * are not supported better not deal with mcast
466 /* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */
468 /* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */
469 /* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */
472 if (level == SOL_TCP) {
474 /* the following are no-op or should work just fine */
475 case TCP_THIN_DUPACK:
476 case TCP_DEFER_ACCEPT:
478 /* the following need some love */
481 case TCP_THIN_LINEAR_TIMEOUTS:
491 case TCP_WINDOW_CLAMP:
493 case TCP_USER_TIMEOUT:
495 case TCP_NOTSENT_LOWAT:
500 /* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */
502 /* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS,
503 * TCP_REPAIR_WINDOW are not supported, better avoid this mess
505 /* TCP_FASTOPEN_KEY, TCP_FASTOPEN TCP_FASTOPEN_CONNECT, TCP_FASTOPEN_NO_COOKIE,
506 * are not supported fastopen is currently unsupported
508 /* TCP_INQ is currently unsupported, needs some recvmsg work */
513 static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval,
516 struct mptcp_subflow_context *subflow;
517 struct sock *sk = (struct sock *)msk;
518 char name[TCP_CA_NAME_MAX];
525 ret = strncpy_from_sockptr(name, optval,
526 min_t(long, TCP_CA_NAME_MAX - 1, optlen));
532 cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN);
536 sockopt_seq_inc(msk);
537 mptcp_for_each_subflow(msk, subflow) {
538 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
542 err = tcp_set_congestion_control(ssk, name, true, cap_net_admin);
543 if (err < 0 && ret == 0)
545 subflow->setsockopt_seq = msk->setsockopt_seq;
550 tcp_set_congestion_control(sk, name, false, cap_net_admin);
556 static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
557 sockptr_t optval, unsigned int optlen)
563 return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen);
569 int mptcp_setsockopt(struct sock *sk, int level, int optname,
570 sockptr_t optval, unsigned int optlen)
572 struct mptcp_sock *msk = mptcp_sk(sk);
575 pr_debug("msk=%p", msk);
577 if (!mptcp_supported_sockopt(level, optname))
580 if (level == SOL_SOCKET)
581 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
583 /* @@ the meaning of setsockopt() when the socket is connected and
584 * there are multiple subflows is not yet defined. It is up to the
585 * MPTCP-level socket to configure the subflows until the subflow
586 * is in TCP fallback, when TCP socket options are passed through
587 * to the one remaining subflow.
590 ssk = __mptcp_tcp_fallback(msk);
593 return tcp_setsockopt(ssk, level, optname, optval, optlen);
595 if (level == SOL_IPV6)
596 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
598 if (level == SOL_TCP)
599 return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen);
604 static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
605 char __user *optval, int __user *optlen)
607 struct sock *sk = (struct sock *)msk;
608 struct socket *ssock;
615 ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
619 ssock = __mptcp_nmpc_socket(msk);
623 ret = tcp_getsockopt(ssock->sk, level, optname, optval, optlen);
630 static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
631 char __user *optval, int __user *optlen)
638 return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname,
644 int mptcp_getsockopt(struct sock *sk, int level, int optname,
645 char __user *optval, int __user *option)
647 struct mptcp_sock *msk = mptcp_sk(sk);
650 pr_debug("msk=%p", msk);
652 /* @@ the meaning of setsockopt() when the socket is connected and
653 * there are multiple subflows is not yet defined. It is up to the
654 * MPTCP-level socket to configure the subflows until the subflow
655 * is in TCP fallback, when socket options are passed through
656 * to the one remaining subflow.
659 ssk = __mptcp_tcp_fallback(msk);
662 return tcp_getsockopt(ssk, level, optname, optval, option);
664 if (level == SOL_TCP)
665 return mptcp_getsockopt_sol_tcp(msk, optname, optval, option);
669 static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
671 static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK;
672 struct sock *sk = (struct sock *)msk;
674 if (ssk->sk_prot->keepalive) {
675 if (sock_flag(sk, SOCK_KEEPOPEN))
676 ssk->sk_prot->keepalive(ssk, 1);
678 ssk->sk_prot->keepalive(ssk, 0);
681 ssk->sk_priority = sk->sk_priority;
682 ssk->sk_bound_dev_if = sk->sk_bound_dev_if;
683 ssk->sk_incoming_cpu = sk->sk_incoming_cpu;
685 if (sk->sk_userlocks & tx_rx_locks) {
686 ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks;
687 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
688 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
689 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
690 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
693 if (sock_flag(sk, SOCK_LINGER)) {
694 ssk->sk_lingertime = sk->sk_lingertime;
695 sock_set_flag(ssk, SOCK_LINGER);
697 sock_reset_flag(ssk, SOCK_LINGER);
700 if (sk->sk_mark != ssk->sk_mark) {
701 ssk->sk_mark = sk->sk_mark;
705 sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG));
707 if (inet_csk(sk)->icsk_ca_ops != inet_csk(ssk)->icsk_ca_ops)
708 tcp_set_congestion_control(ssk, inet_csk(sk)->icsk_ca_ops->name, false, true);
711 static void __mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
713 bool slow = lock_sock_fast(ssk);
715 sync_socket_options(msk, ssk);
717 unlock_sock_fast(ssk, slow);
720 void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
722 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
724 msk_owned_by_me(msk);
726 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
727 __mptcp_sockopt_sync(msk, ssk);
729 subflow->setsockopt_seq = msk->setsockopt_seq;
733 void mptcp_sockopt_sync_all(struct mptcp_sock *msk)
735 struct mptcp_subflow_context *subflow;
736 struct sock *sk = (struct sock *)msk;
739 seq = sockopt_seq_reset(sk);
741 mptcp_for_each_subflow(msk, subflow) {
742 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
743 u32 sseq = READ_ONCE(subflow->setsockopt_seq);
745 if (sseq != msk->setsockopt_seq) {
746 __mptcp_sockopt_sync(msk, ssk);
747 WRITE_ONCE(subflow->setsockopt_seq, seq);
748 } else if (sseq != seq) {
749 WRITE_ONCE(subflow->setsockopt_seq, seq);
755 msk->setsockopt_seq = seq;