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_tstamp(struct mptcp_sock *msk, int optname, int val)
145 sockptr_t optval = KERNEL_SOCKPTR(&val);
146 struct mptcp_subflow_context *subflow;
147 struct sock *sk = (struct sock *)msk;
150 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
151 optval, sizeof(val));
156 mptcp_for_each_subflow(msk, subflow) {
157 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
158 bool slow = lock_sock_fast(ssk);
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);
167 case SO_TIMESTAMPING_NEW:
168 case SO_TIMESTAMPING_OLD:
169 sock_set_timestamping(sk, optname, val);
173 unlock_sock_fast(ssk, slow);
180 static int mptcp_setsockopt_sol_socket_int(struct mptcp_sock *msk, int optname,
181 sockptr_t optval, unsigned int optlen)
185 ret = mptcp_get_int_option(msk, optval, optlen, &val);
191 mptcp_sol_socket_sync_intval(msk, optname, val);
200 return mptcp_sol_socket_intval(msk, optname, val);
201 case SO_INCOMING_CPU:
202 mptcp_so_incoming_cpu(msk, val);
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);
216 static int mptcp_setsockopt_sol_socket_linger(struct mptcp_sock *msk, sockptr_t optval,
219 struct mptcp_subflow_context *subflow;
220 struct sock *sk = (struct sock *)msk;
225 if (optlen < sizeof(ling))
228 if (copy_from_sockptr(&ling, optval, sizeof(ling)))
231 kopt = KERNEL_SOCKPTR(&ling);
232 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, SO_LINGER, kopt, sizeof(ling));
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);
243 sock_reset_flag(ssk, SOCK_LINGER);
245 ssk->sk_lingertime = sk->sk_lingertime;
246 sock_set_flag(ssk, SOCK_LINGER);
249 subflow->setsockopt_seq = msk->setsockopt_seq;
250 unlock_sock_fast(ssk, slow);
257 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
258 sockptr_t optval, unsigned int optlen)
260 struct sock *sk = (struct sock *)msk;
261 struct socket *ssock;
267 case SO_BINDTODEVICE:
268 case SO_BINDTOIFINDEX:
270 ssock = __mptcp_nmpc_socket(msk);
276 ret = sock_setsockopt(ssock, SOL_SOCKET, optname, optval, optlen);
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;
296 case SO_INCOMING_CPU:
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);
306 return mptcp_setsockopt_sol_socket_linger(msk, optval, optlen);
308 case SO_RCVTIMEO_OLD:
309 case SO_RCVTIMEO_NEW:
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);
324 case SO_SELECT_ERR_QUEUE:
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
333 * SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks
334 * explicitly the sk_protocol field
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
341 * SO_ZEROCOPY is currently unsupported, TODO in sndmsg
342 * SO_TXTIME is currently unsupported
348 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
349 sockptr_t optval, unsigned int optlen)
351 struct sock *sk = (struct sock *)msk;
352 int ret = -EOPNOTSUPP;
353 struct socket *ssock;
358 ssock = __mptcp_nmpc_socket(msk);
364 ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen);
366 sk->sk_ipv6only = ssock->sk->sk_ipv6only;
375 static bool mptcp_supported_sockopt(int level, int optname)
377 if (level == SOL_IP) {
379 /* should work fine */
383 /* the following are control cmsg related */
390 case IP_RECVORIGDSTADDR:
392 case IP_RECVFRAGSIZE:
394 /* common stuff that need some love */
397 case IP_BIND_ADDRESS_NO_PORT:
398 case IP_MTU_DISCOVER:
401 /* possibly less common may deserve some love */
404 /* the following is apparently a no-op for plain TCP */
405 case IP_RECVERR_RFC4884:
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
419 /* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */
422 if (level == SOL_IPV6) {
426 /* the following are control cmsg related */
427 case IPV6_RECVPKTINFO:
428 case IPV6_2292PKTINFO:
429 case IPV6_RECVHOPLIMIT:
430 case IPV6_2292HOPLIMIT:
433 case IPV6_RECVHOPOPTS:
434 case IPV6_2292HOPOPTS:
435 case IPV6_RECVDSTOPTS:
436 case IPV6_2292DSTOPTS:
437 case IPV6_RECVTCLASS:
439 case IPV6_RECVPATHMTU:
440 case IPV6_RECVORIGDSTADDR:
441 case IPV6_RECVFRAGSIZE:
443 /* the following ones need some love but are quite common */
445 case IPV6_TRANSPARENT:
448 case IPV6_2292PKTOPTIONS:
449 case IPV6_UNICAST_HOPS:
450 case IPV6_MTU_DISCOVER:
453 case IPV6_FLOWINFO_SEND:
454 case IPV6_FLOWLABEL_MGR:
455 case IPV6_MINHOPCOUNT:
457 case IPV6_AUTOFLOWLABEL:
459 /* the following one is a no-op for plain TCP */
460 case IPV6_RECVERR_RFC4884:
464 /* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are
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
475 /* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */
477 /* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */
478 /* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */
481 if (level == SOL_TCP) {
483 /* the following are no-op or should work just fine */
484 case TCP_THIN_DUPACK:
485 case TCP_DEFER_ACCEPT:
487 /* the following need some love */
490 case TCP_THIN_LINEAR_TIMEOUTS:
500 case TCP_WINDOW_CLAMP:
502 case TCP_USER_TIMEOUT:
504 case TCP_NOTSENT_LOWAT:
509 /* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */
511 /* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS,
512 * TCP_REPAIR_WINDOW are not supported, better avoid this mess
514 /* TCP_FASTOPEN_KEY, TCP_FASTOPEN TCP_FASTOPEN_CONNECT, TCP_FASTOPEN_NO_COOKIE,
515 * are not supported fastopen is currently unsupported
517 /* TCP_INQ is currently unsupported, needs some recvmsg work */
522 static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval,
525 struct mptcp_subflow_context *subflow;
526 struct sock *sk = (struct sock *)msk;
527 char name[TCP_CA_NAME_MAX];
534 ret = strncpy_from_sockptr(name, optval,
535 min_t(long, TCP_CA_NAME_MAX - 1, optlen));
541 cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN);
545 sockopt_seq_inc(msk);
546 mptcp_for_each_subflow(msk, subflow) {
547 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
551 err = tcp_set_congestion_control(ssk, name, true, cap_net_admin);
552 if (err < 0 && ret == 0)
554 subflow->setsockopt_seq = msk->setsockopt_seq;
559 strcpy(msk->ca_name, name);
565 static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
566 sockptr_t optval, unsigned int optlen)
572 return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen);
578 int mptcp_setsockopt(struct sock *sk, int level, int optname,
579 sockptr_t optval, unsigned int optlen)
581 struct mptcp_sock *msk = mptcp_sk(sk);
584 pr_debug("msk=%p", msk);
586 if (level == SOL_SOCKET)
587 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
589 if (!mptcp_supported_sockopt(level, optname))
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.
599 ssk = __mptcp_tcp_fallback(msk);
602 return tcp_setsockopt(ssk, level, optname, optval, optlen);
604 if (level == SOL_IPV6)
605 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
607 if (level == SOL_TCP)
608 return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen);
613 static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
614 char __user *optval, int __user *optlen)
616 struct sock *sk = (struct sock *)msk;
617 struct socket *ssock;
624 ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
628 ssock = __mptcp_nmpc_socket(msk);
632 ret = tcp_getsockopt(ssock->sk, level, optname, optval, optlen);
639 static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
640 char __user *optval, int __user *optlen)
647 return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname,
653 int mptcp_getsockopt(struct sock *sk, int level, int optname,
654 char __user *optval, int __user *option)
656 struct mptcp_sock *msk = mptcp_sk(sk);
659 pr_debug("msk=%p", msk);
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.
668 ssk = __mptcp_tcp_fallback(msk);
671 return tcp_getsockopt(ssk, level, optname, optval, option);
673 if (level == SOL_TCP)
674 return mptcp_getsockopt_sol_tcp(msk, optname, optval, option);
678 static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
680 static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK;
681 struct sock *sk = (struct sock *)msk;
683 if (ssk->sk_prot->keepalive) {
684 if (sock_flag(sk, SOCK_KEEPOPEN))
685 ssk->sk_prot->keepalive(ssk, 1);
687 ssk->sk_prot->keepalive(ssk, 0);
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;
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);
702 if (sock_flag(sk, SOCK_LINGER)) {
703 ssk->sk_lingertime = sk->sk_lingertime;
704 sock_set_flag(ssk, SOCK_LINGER);
706 sock_reset_flag(ssk, SOCK_LINGER);
709 if (sk->sk_mark != ssk->sk_mark) {
710 ssk->sk_mark = sk->sk_mark;
714 sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG));
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);
720 static void __mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
722 bool slow = lock_sock_fast(ssk);
724 sync_socket_options(msk, ssk);
726 unlock_sock_fast(ssk, slow);
729 void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
731 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
733 msk_owned_by_me(msk);
735 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
736 __mptcp_sockopt_sync(msk, ssk);
738 subflow->setsockopt_seq = msk->setsockopt_seq;
742 void mptcp_sockopt_sync_all(struct mptcp_sock *msk)
744 struct mptcp_subflow_context *subflow;
745 struct sock *sk = (struct sock *)msk;
748 seq = sockopt_seq_reset(sk);
750 mptcp_for_each_subflow(msk, subflow) {
751 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
752 u32 sseq = READ_ONCE(subflow->setsockopt_seq);
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);
764 msk->setsockopt_seq = seq;