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);
160 sock_set_timestamp(sk, optname, !!val);
161 unlock_sock_fast(ssk, slow);
168 static int mptcp_setsockopt_sol_socket_int(struct mptcp_sock *msk, int optname,
174 ret = mptcp_get_int_option(msk, optval, optlen, &val);
180 mptcp_sol_socket_sync_intval(msk, optname, val);
189 return mptcp_sol_socket_intval(msk, optname, val);
190 case SO_INCOMING_CPU:
191 mptcp_so_incoming_cpu(msk, val);
193 case SO_TIMESTAMP_OLD:
194 case SO_TIMESTAMP_NEW:
195 case SO_TIMESTAMPNS_OLD:
196 case SO_TIMESTAMPNS_NEW:
197 return mptcp_setsockopt_sol_socket_tstamp(msk, optname, val);
203 static int mptcp_setsockopt_sol_socket_timestamping(struct mptcp_sock *msk,
208 struct mptcp_subflow_context *subflow;
209 struct sock *sk = (struct sock *)msk;
212 ret = mptcp_get_int_option(msk, optval, optlen, &val);
216 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
217 KERNEL_SOCKPTR(&val), sizeof(val));
223 mptcp_for_each_subflow(msk, subflow) {
224 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
225 bool slow = lock_sock_fast(ssk);
227 sock_set_timestamping(sk, optname, val);
228 unlock_sock_fast(ssk, slow);
236 static int mptcp_setsockopt_sol_socket_linger(struct mptcp_sock *msk, sockptr_t optval,
239 struct mptcp_subflow_context *subflow;
240 struct sock *sk = (struct sock *)msk;
245 if (optlen < sizeof(ling))
248 if (copy_from_sockptr(&ling, optval, sizeof(ling)))
251 kopt = KERNEL_SOCKPTR(&ling);
252 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, SO_LINGER, kopt, sizeof(ling));
257 sockopt_seq_inc(msk);
258 mptcp_for_each_subflow(msk, subflow) {
259 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
260 bool slow = lock_sock_fast(ssk);
263 sock_reset_flag(ssk, SOCK_LINGER);
265 ssk->sk_lingertime = sk->sk_lingertime;
266 sock_set_flag(ssk, SOCK_LINGER);
269 subflow->setsockopt_seq = msk->setsockopt_seq;
270 unlock_sock_fast(ssk, slow);
277 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
278 sockptr_t optval, unsigned int optlen)
280 struct sock *sk = (struct sock *)msk;
281 struct socket *ssock;
287 case SO_BINDTODEVICE:
288 case SO_BINDTOIFINDEX:
290 ssock = __mptcp_nmpc_socket(msk);
296 ret = sock_setsockopt(ssock, SOL_SOCKET, optname, optval, optlen);
298 if (optname == SO_REUSEPORT)
299 sk->sk_reuseport = ssock->sk->sk_reuseport;
300 else if (optname == SO_REUSEADDR)
301 sk->sk_reuse = ssock->sk->sk_reuse;
302 else if (optname == SO_BINDTODEVICE)
303 sk->sk_bound_dev_if = ssock->sk->sk_bound_dev_if;
304 else if (optname == SO_BINDTOIFINDEX)
305 sk->sk_bound_dev_if = ssock->sk->sk_bound_dev_if;
316 case SO_INCOMING_CPU:
318 case SO_TIMESTAMP_OLD:
319 case SO_TIMESTAMP_NEW:
320 case SO_TIMESTAMPNS_OLD:
321 case SO_TIMESTAMPNS_NEW:
322 return mptcp_setsockopt_sol_socket_int(msk, optname, optval,
324 case SO_TIMESTAMPING_OLD:
325 case SO_TIMESTAMPING_NEW:
326 return mptcp_setsockopt_sol_socket_timestamping(msk, optname,
329 return mptcp_setsockopt_sol_socket_linger(msk, optval, optlen);
331 case SO_RCVTIMEO_OLD:
332 case SO_RCVTIMEO_NEW:
334 case SO_PREFER_BUSY_POLL:
335 case SO_BUSY_POLL_BUDGET:
336 /* No need to copy: only relevant for msk */
337 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
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
371 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
372 sockptr_t optval, unsigned int optlen)
374 struct sock *sk = (struct sock *)msk;
375 int ret = -EOPNOTSUPP;
376 struct socket *ssock;
381 ssock = __mptcp_nmpc_socket(msk);
387 ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen);
389 sk->sk_ipv6only = ssock->sk->sk_ipv6only;
398 static bool mptcp_supported_sockopt(int level, int optname)
400 if (level == SOL_IP) {
402 /* should work fine */
406 /* the following are control cmsg related */
413 case IP_RECVORIGDSTADDR:
415 case IP_RECVFRAGSIZE:
417 /* common stuff that need some love */
420 case IP_BIND_ADDRESS_NO_PORT:
421 case IP_MTU_DISCOVER:
424 /* possibly less common may deserve some love */
427 /* the following is apparently a no-op for plain TCP */
428 case IP_RECVERR_RFC4884:
432 /* IP_OPTIONS is not supported, needs subflow care */
433 /* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */
434 /* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF,
435 * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP,
436 * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE,
437 * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP,
438 * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE,
439 * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal
442 /* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */
445 if (level == SOL_IPV6) {
449 /* the following are control cmsg related */
450 case IPV6_RECVPKTINFO:
451 case IPV6_2292PKTINFO:
452 case IPV6_RECVHOPLIMIT:
453 case IPV6_2292HOPLIMIT:
456 case IPV6_RECVHOPOPTS:
457 case IPV6_2292HOPOPTS:
458 case IPV6_RECVDSTOPTS:
459 case IPV6_2292DSTOPTS:
460 case IPV6_RECVTCLASS:
462 case IPV6_RECVPATHMTU:
463 case IPV6_RECVORIGDSTADDR:
464 case IPV6_RECVFRAGSIZE:
466 /* the following ones need some love but are quite common */
468 case IPV6_TRANSPARENT:
471 case IPV6_2292PKTOPTIONS:
472 case IPV6_UNICAST_HOPS:
473 case IPV6_MTU_DISCOVER:
476 case IPV6_FLOWINFO_SEND:
477 case IPV6_FLOWLABEL_MGR:
478 case IPV6_MINHOPCOUNT:
480 case IPV6_AUTOFLOWLABEL:
482 /* the following one is a no-op for plain TCP */
483 case IPV6_RECVERR_RFC4884:
487 /* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are
490 /* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF,
491 * IPV6_MULTICAST_IF, IPV6_ADDRFORM,
492 * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST,
493 * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP,
494 * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP,
495 * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER
496 * are not supported better not deal with mcast
498 /* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */
500 /* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */
501 /* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */
504 if (level == SOL_TCP) {
506 /* the following are no-op or should work just fine */
507 case TCP_THIN_DUPACK:
508 case TCP_DEFER_ACCEPT:
510 /* the following need some love */
513 case TCP_THIN_LINEAR_TIMEOUTS:
523 case TCP_WINDOW_CLAMP:
525 case TCP_USER_TIMEOUT:
527 case TCP_NOTSENT_LOWAT:
532 /* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */
534 /* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS,
535 * TCP_REPAIR_WINDOW are not supported, better avoid this mess
537 /* TCP_FASTOPEN_KEY, TCP_FASTOPEN TCP_FASTOPEN_CONNECT, TCP_FASTOPEN_NO_COOKIE,
538 * are not supported fastopen is currently unsupported
540 /* TCP_INQ is currently unsupported, needs some recvmsg work */
545 static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval,
548 struct mptcp_subflow_context *subflow;
549 struct sock *sk = (struct sock *)msk;
550 char name[TCP_CA_NAME_MAX];
557 ret = strncpy_from_sockptr(name, optval,
558 min_t(long, TCP_CA_NAME_MAX - 1, optlen));
564 cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN);
568 sockopt_seq_inc(msk);
569 mptcp_for_each_subflow(msk, subflow) {
570 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
574 err = tcp_set_congestion_control(ssk, name, true, cap_net_admin);
575 if (err < 0 && ret == 0)
577 subflow->setsockopt_seq = msk->setsockopt_seq;
582 strcpy(msk->ca_name, name);
588 static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
589 sockptr_t optval, unsigned int optlen)
595 return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen);
601 int mptcp_setsockopt(struct sock *sk, int level, int optname,
602 sockptr_t optval, unsigned int optlen)
604 struct mptcp_sock *msk = mptcp_sk(sk);
607 pr_debug("msk=%p", msk);
609 if (level == SOL_SOCKET)
610 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
612 if (!mptcp_supported_sockopt(level, optname))
615 /* @@ the meaning of setsockopt() when the socket is connected and
616 * there are multiple subflows is not yet defined. It is up to the
617 * MPTCP-level socket to configure the subflows until the subflow
618 * is in TCP fallback, when TCP socket options are passed through
619 * to the one remaining subflow.
622 ssk = __mptcp_tcp_fallback(msk);
625 return tcp_setsockopt(ssk, level, optname, optval, optlen);
627 if (level == SOL_IPV6)
628 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
630 if (level == SOL_TCP)
631 return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen);
636 static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
637 char __user *optval, int __user *optlen)
639 struct sock *sk = (struct sock *)msk;
640 struct socket *ssock;
647 ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
651 ssock = __mptcp_nmpc_socket(msk);
655 ret = tcp_getsockopt(ssock->sk, level, optname, optval, optlen);
662 static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
663 char __user *optval, int __user *optlen)
670 return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname,
676 int mptcp_getsockopt(struct sock *sk, int level, int optname,
677 char __user *optval, int __user *option)
679 struct mptcp_sock *msk = mptcp_sk(sk);
682 pr_debug("msk=%p", msk);
684 /* @@ the meaning of setsockopt() when the socket is connected and
685 * there are multiple subflows is not yet defined. It is up to the
686 * MPTCP-level socket to configure the subflows until the subflow
687 * is in TCP fallback, when socket options are passed through
688 * to the one remaining subflow.
691 ssk = __mptcp_tcp_fallback(msk);
694 return tcp_getsockopt(ssk, level, optname, optval, option);
696 if (level == SOL_TCP)
697 return mptcp_getsockopt_sol_tcp(msk, optname, optval, option);
701 static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
703 static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK;
704 struct sock *sk = (struct sock *)msk;
706 if (ssk->sk_prot->keepalive) {
707 if (sock_flag(sk, SOCK_KEEPOPEN))
708 ssk->sk_prot->keepalive(ssk, 1);
710 ssk->sk_prot->keepalive(ssk, 0);
713 ssk->sk_priority = sk->sk_priority;
714 ssk->sk_bound_dev_if = sk->sk_bound_dev_if;
715 ssk->sk_incoming_cpu = sk->sk_incoming_cpu;
717 if (sk->sk_userlocks & tx_rx_locks) {
718 ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks;
719 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
720 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
721 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
722 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
725 if (sock_flag(sk, SOCK_LINGER)) {
726 ssk->sk_lingertime = sk->sk_lingertime;
727 sock_set_flag(ssk, SOCK_LINGER);
729 sock_reset_flag(ssk, SOCK_LINGER);
732 if (sk->sk_mark != ssk->sk_mark) {
733 ssk->sk_mark = sk->sk_mark;
737 sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG));
739 if (inet_csk(sk)->icsk_ca_ops != inet_csk(ssk)->icsk_ca_ops)
740 tcp_set_congestion_control(ssk, msk->ca_name, false, true);
743 static void __mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
745 bool slow = lock_sock_fast(ssk);
747 sync_socket_options(msk, ssk);
749 unlock_sock_fast(ssk, slow);
752 void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
754 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
756 msk_owned_by_me(msk);
758 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
759 __mptcp_sockopt_sync(msk, ssk);
761 subflow->setsockopt_seq = msk->setsockopt_seq;
765 void mptcp_sockopt_sync_all(struct mptcp_sock *msk)
767 struct mptcp_subflow_context *subflow;
768 struct sock *sk = (struct sock *)msk;
771 seq = sockopt_seq_reset(sk);
773 mptcp_for_each_subflow(msk, subflow) {
774 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
775 u32 sseq = READ_ONCE(subflow->setsockopt_seq);
777 if (sseq != msk->setsockopt_seq) {
778 __mptcp_sockopt_sync(msk, ssk);
779 WRITE_ONCE(subflow->setsockopt_seq, seq);
780 } else if (sseq != seq) {
781 WRITE_ONCE(subflow->setsockopt_seq, seq);
787 msk->setsockopt_seq = seq;