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 #define MIN_INFO_OPTLEN_SIZE 16
18 #define MIN_FULL_INFO_OPTLEN_SIZE 40
20 static struct sock *__mptcp_tcp_fallback(struct mptcp_sock *msk)
24 if (likely(!__mptcp_check_fallback(msk)))
30 static u32 sockopt_seq_reset(const struct sock *sk)
34 /* Highbits contain state. Allows to distinguish sockopt_seq
35 * of listener and established:
37 * sockopt(s0) - seq is 1
38 * s1 = accept(s0) - s1 inherits seq 1 if listener sk (s0)
39 * sockopt(s0) - seq increments to 2 on s0
40 * sockopt(s1) // seq increments to 2 on s1 (different option)
41 * new ssk completes join, inherits options from s0 // seq 2
42 * Needs sync from mptcp join logic, but ssk->seq == msk->seq
44 * Set High order bits to sk_state so ssk->seq == msk->seq test
48 return (u32)sk->sk_state << 24u;
51 static void sockopt_seq_inc(struct mptcp_sock *msk)
53 u32 seq = (msk->setsockopt_seq + 1) & 0x00ffffff;
55 msk->setsockopt_seq = sockopt_seq_reset((struct sock *)msk) + seq;
58 static int mptcp_get_int_option(struct mptcp_sock *msk, sockptr_t optval,
59 unsigned int optlen, int *val)
61 if (optlen < sizeof(int))
64 if (copy_from_sockptr(val, optval, sizeof(*val)))
70 static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, int val)
72 struct mptcp_subflow_context *subflow;
73 struct sock *sk = (struct sock *)msk;
78 mptcp_for_each_subflow(msk, subflow) {
79 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
80 bool slow = lock_sock_fast(ssk);
84 sock_valbool_flag(ssk, SOCK_DBG, !!val);
87 if (ssk->sk_prot->keepalive)
88 ssk->sk_prot->keepalive(ssk, !!val);
89 sock_valbool_flag(ssk, SOCK_KEEPOPEN, !!val);
92 WRITE_ONCE(ssk->sk_priority, val);
96 ssk->sk_userlocks |= SOCK_SNDBUF_LOCK;
97 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
98 mptcp_subflow_ctx(ssk)->cached_sndbuf = sk->sk_sndbuf;
102 ssk->sk_userlocks |= SOCK_RCVBUF_LOCK;
103 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
106 if (READ_ONCE(ssk->sk_mark) != sk->sk_mark) {
107 WRITE_ONCE(ssk->sk_mark, sk->sk_mark);
111 case SO_INCOMING_CPU:
112 WRITE_ONCE(ssk->sk_incoming_cpu, val);
116 subflow->setsockopt_seq = msk->setsockopt_seq;
117 unlock_sock_fast(ssk, slow);
123 static int mptcp_sol_socket_intval(struct mptcp_sock *msk, int optname, int val)
125 sockptr_t optval = KERNEL_SOCKPTR(&val);
126 struct sock *sk = (struct sock *)msk;
129 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
130 optval, sizeof(val));
134 mptcp_sol_socket_sync_intval(msk, optname, val);
138 static void mptcp_so_incoming_cpu(struct mptcp_sock *msk, int val)
140 struct sock *sk = (struct sock *)msk;
142 WRITE_ONCE(sk->sk_incoming_cpu, val);
144 mptcp_sol_socket_sync_intval(msk, SO_INCOMING_CPU, val);
147 static int mptcp_setsockopt_sol_socket_tstamp(struct mptcp_sock *msk, int optname, int val)
149 sockptr_t optval = KERNEL_SOCKPTR(&val);
150 struct mptcp_subflow_context *subflow;
151 struct sock *sk = (struct sock *)msk;
154 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
155 optval, sizeof(val));
160 mptcp_for_each_subflow(msk, subflow) {
161 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
162 bool slow = lock_sock_fast(ssk);
164 sock_set_timestamp(sk, optname, !!val);
165 unlock_sock_fast(ssk, slow);
172 static int mptcp_setsockopt_sol_socket_int(struct mptcp_sock *msk, int optname,
178 ret = mptcp_get_int_option(msk, optval, optlen, &val);
184 mptcp_sol_socket_sync_intval(msk, optname, val);
193 return mptcp_sol_socket_intval(msk, optname, val);
194 case SO_INCOMING_CPU:
195 mptcp_so_incoming_cpu(msk, val);
197 case SO_TIMESTAMP_OLD:
198 case SO_TIMESTAMP_NEW:
199 case SO_TIMESTAMPNS_OLD:
200 case SO_TIMESTAMPNS_NEW:
201 return mptcp_setsockopt_sol_socket_tstamp(msk, optname, val);
207 static int mptcp_setsockopt_sol_socket_timestamping(struct mptcp_sock *msk,
212 struct mptcp_subflow_context *subflow;
213 struct sock *sk = (struct sock *)msk;
214 struct so_timestamping timestamping;
217 if (optlen == sizeof(timestamping)) {
218 if (copy_from_sockptr(×tamping, optval,
219 sizeof(timestamping)))
221 } else if (optlen == sizeof(int)) {
222 memset(×tamping, 0, sizeof(timestamping));
224 if (copy_from_sockptr(×tamping.flags, optval, sizeof(int)))
230 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
231 KERNEL_SOCKPTR(×tamping),
232 sizeof(timestamping));
238 mptcp_for_each_subflow(msk, subflow) {
239 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
240 bool slow = lock_sock_fast(ssk);
242 sock_set_timestamping(sk, optname, timestamping);
243 unlock_sock_fast(ssk, slow);
251 static int mptcp_setsockopt_sol_socket_linger(struct mptcp_sock *msk, sockptr_t optval,
254 struct mptcp_subflow_context *subflow;
255 struct sock *sk = (struct sock *)msk;
260 if (optlen < sizeof(ling))
263 if (copy_from_sockptr(&ling, optval, sizeof(ling)))
266 kopt = KERNEL_SOCKPTR(&ling);
267 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, SO_LINGER, kopt, sizeof(ling));
272 sockopt_seq_inc(msk);
273 mptcp_for_each_subflow(msk, subflow) {
274 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
275 bool slow = lock_sock_fast(ssk);
278 sock_reset_flag(ssk, SOCK_LINGER);
280 ssk->sk_lingertime = sk->sk_lingertime;
281 sock_set_flag(ssk, SOCK_LINGER);
284 subflow->setsockopt_seq = msk->setsockopt_seq;
285 unlock_sock_fast(ssk, slow);
292 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
293 sockptr_t optval, unsigned int optlen)
295 struct sock *sk = (struct sock *)msk;
302 case SO_BINDTODEVICE:
303 case SO_BINDTOIFINDEX:
305 ssk = __mptcp_nmpc_sk(msk);
311 ret = sk_setsockopt(ssk, SOL_SOCKET, optname, optval, optlen);
313 if (optname == SO_REUSEPORT)
314 sk->sk_reuseport = ssk->sk_reuseport;
315 else if (optname == SO_REUSEADDR)
316 sk->sk_reuse = ssk->sk_reuse;
317 else if (optname == SO_BINDTODEVICE)
318 sk->sk_bound_dev_if = ssk->sk_bound_dev_if;
319 else if (optname == SO_BINDTOIFINDEX)
320 sk->sk_bound_dev_if = ssk->sk_bound_dev_if;
331 case SO_INCOMING_CPU:
333 case SO_TIMESTAMP_OLD:
334 case SO_TIMESTAMP_NEW:
335 case SO_TIMESTAMPNS_OLD:
336 case SO_TIMESTAMPNS_NEW:
337 return mptcp_setsockopt_sol_socket_int(msk, optname, optval,
339 case SO_TIMESTAMPING_OLD:
340 case SO_TIMESTAMPING_NEW:
341 return mptcp_setsockopt_sol_socket_timestamping(msk, optname,
344 return mptcp_setsockopt_sol_socket_linger(msk, optval, optlen);
346 case SO_RCVTIMEO_OLD:
347 case SO_RCVTIMEO_NEW:
348 case SO_SNDTIMEO_OLD:
349 case SO_SNDTIMEO_NEW:
351 case SO_PREFER_BUSY_POLL:
352 case SO_BUSY_POLL_BUDGET:
353 /* No need to copy: only relevant for msk */
354 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
365 case SO_SELECT_ERR_QUEUE:
369 /* SO_OOBINLINE is not supported, let's avoid the related mess
370 * SO_ATTACH_FILTER, SO_ATTACH_BPF, SO_ATTACH_REUSEPORT_CBPF,
371 * SO_DETACH_REUSEPORT_BPF, SO_DETACH_FILTER, SO_LOCK_FILTER,
372 * we must be careful with subflows
374 * SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks
375 * explicitly the sk_protocol field
377 * SO_PEEK_OFF is unsupported, as it is for plain TCP
378 * SO_MAX_PACING_RATE is unsupported, we must be careful with subflows
379 * SO_CNX_ADVICE is currently unsupported, could possibly be relevant,
380 * but likely needs careful design
382 * SO_ZEROCOPY is currently unsupported, TODO in sndmsg
383 * SO_TXTIME is currently unsupported
389 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
390 sockptr_t optval, unsigned int optlen)
392 struct sock *sk = (struct sock *)msk;
393 int ret = -EOPNOTSUPP;
398 case IPV6_TRANSPARENT:
401 ssk = __mptcp_nmpc_sk(msk);
407 ret = tcp_setsockopt(ssk, SOL_IPV6, optname, optval, optlen);
413 sockopt_seq_inc(msk);
417 sk->sk_ipv6only = ssk->sk_ipv6only;
419 case IPV6_TRANSPARENT:
420 inet_assign_bit(TRANSPARENT, sk,
421 inet_test_bit(TRANSPARENT, ssk));
424 inet_assign_bit(FREEBIND, sk,
425 inet_test_bit(FREEBIND, ssk));
436 static bool mptcp_supported_sockopt(int level, int optname)
438 if (level == SOL_IP) {
440 /* should work fine */
443 case IP_BIND_ADDRESS_NO_PORT:
444 case IP_LOCAL_PORT_RANGE:
446 /* the following are control cmsg related */
453 case IP_RECVORIGDSTADDR:
455 case IP_RECVFRAGSIZE:
457 /* common stuff that need some love */
460 case IP_MTU_DISCOVER:
463 /* possibly less common may deserve some love */
466 /* the following is apparently a no-op for plain TCP */
467 case IP_RECVERR_RFC4884:
471 /* IP_OPTIONS is not supported, needs subflow care */
472 /* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */
473 /* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF,
474 * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP,
475 * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE,
476 * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP,
477 * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE,
478 * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal
481 /* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */
484 if (level == SOL_IPV6) {
488 /* the following are control cmsg related */
489 case IPV6_RECVPKTINFO:
490 case IPV6_2292PKTINFO:
491 case IPV6_RECVHOPLIMIT:
492 case IPV6_2292HOPLIMIT:
495 case IPV6_RECVHOPOPTS:
496 case IPV6_2292HOPOPTS:
497 case IPV6_RECVDSTOPTS:
498 case IPV6_2292DSTOPTS:
499 case IPV6_RECVTCLASS:
501 case IPV6_RECVPATHMTU:
502 case IPV6_RECVORIGDSTADDR:
503 case IPV6_RECVFRAGSIZE:
505 /* the following ones need some love but are quite common */
507 case IPV6_TRANSPARENT:
510 case IPV6_2292PKTOPTIONS:
511 case IPV6_UNICAST_HOPS:
512 case IPV6_MTU_DISCOVER:
515 case IPV6_FLOWINFO_SEND:
516 case IPV6_FLOWLABEL_MGR:
517 case IPV6_MINHOPCOUNT:
519 case IPV6_AUTOFLOWLABEL:
521 /* the following one is a no-op for plain TCP */
522 case IPV6_RECVERR_RFC4884:
526 /* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are
529 /* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF,
530 * IPV6_MULTICAST_IF, IPV6_ADDRFORM,
531 * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST,
532 * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP,
533 * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP,
534 * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER
535 * are not supported better not deal with mcast
537 /* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */
539 /* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */
540 /* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */
543 if (level == SOL_TCP) {
545 /* the following are no-op or should work just fine */
546 case TCP_THIN_DUPACK:
547 case TCP_DEFER_ACCEPT:
549 /* the following need some love */
552 case TCP_THIN_LINEAR_TIMEOUTS:
561 case TCP_WINDOW_CLAMP:
563 case TCP_USER_TIMEOUT:
565 case TCP_NOTSENT_LOWAT:
569 case TCP_FASTOPEN_CONNECT:
570 case TCP_FASTOPEN_KEY:
571 case TCP_FASTOPEN_NO_COOKIE:
575 /* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */
577 /* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS,
578 * TCP_REPAIR_WINDOW are not supported, better avoid this mess
584 static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval,
587 struct mptcp_subflow_context *subflow;
588 struct sock *sk = (struct sock *)msk;
589 char name[TCP_CA_NAME_MAX];
596 ret = strncpy_from_sockptr(name, optval,
597 min_t(long, TCP_CA_NAME_MAX - 1, optlen));
603 cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN);
607 sockopt_seq_inc(msk);
608 mptcp_for_each_subflow(msk, subflow) {
609 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
613 err = tcp_set_congestion_control(ssk, name, true, cap_net_admin);
614 if (err < 0 && ret == 0)
616 subflow->setsockopt_seq = msk->setsockopt_seq;
621 strcpy(msk->ca_name, name);
627 static int __mptcp_setsockopt_sol_tcp_cork(struct mptcp_sock *msk, int val)
629 struct mptcp_subflow_context *subflow;
630 struct sock *sk = (struct sock *)msk;
632 sockopt_seq_inc(msk);
634 mptcp_for_each_subflow(msk, subflow) {
635 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
638 __tcp_sock_set_cork(ssk, !!val);
642 mptcp_check_and_set_pending(sk);
647 static int __mptcp_setsockopt_sol_tcp_nodelay(struct mptcp_sock *msk, int val)
649 struct mptcp_subflow_context *subflow;
650 struct sock *sk = (struct sock *)msk;
652 sockopt_seq_inc(msk);
653 msk->nodelay = !!val;
654 mptcp_for_each_subflow(msk, subflow) {
655 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
658 __tcp_sock_set_nodelay(ssk, !!val);
662 mptcp_check_and_set_pending(sk);
666 static int mptcp_setsockopt_sol_ip_set(struct mptcp_sock *msk, int optname,
667 sockptr_t optval, unsigned int optlen)
669 struct sock *sk = (struct sock *)msk;
673 err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen);
679 ssk = __mptcp_nmpc_sk(msk);
687 inet_assign_bit(FREEBIND, ssk, inet_test_bit(FREEBIND, sk));
690 inet_assign_bit(TRANSPARENT, ssk,
691 inet_test_bit(TRANSPARENT, sk));
693 case IP_BIND_ADDRESS_NO_PORT:
694 inet_assign_bit(BIND_ADDRESS_NO_PORT, ssk,
695 inet_test_bit(BIND_ADDRESS_NO_PORT, sk));
697 case IP_LOCAL_PORT_RANGE:
698 WRITE_ONCE(inet_sk(ssk)->local_port_range,
699 READ_ONCE(inet_sk(sk)->local_port_range));
707 sockopt_seq_inc(msk);
712 static int mptcp_setsockopt_v4_set_tos(struct mptcp_sock *msk, int optname,
713 sockptr_t optval, unsigned int optlen)
715 struct mptcp_subflow_context *subflow;
716 struct sock *sk = (struct sock *)msk;
719 err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen);
725 sockopt_seq_inc(msk);
726 val = READ_ONCE(inet_sk(sk)->tos);
727 mptcp_for_each_subflow(msk, subflow) {
728 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
731 slow = lock_sock_fast(ssk);
732 __ip_sock_set_tos(ssk, val);
733 unlock_sock_fast(ssk, slow);
740 static int mptcp_setsockopt_v4(struct mptcp_sock *msk, int optname,
741 sockptr_t optval, unsigned int optlen)
746 case IP_BIND_ADDRESS_NO_PORT:
747 case IP_LOCAL_PORT_RANGE:
748 return mptcp_setsockopt_sol_ip_set(msk, optname, optval, optlen);
750 return mptcp_setsockopt_v4_set_tos(msk, optname, optval, optlen);
756 static int mptcp_setsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
757 sockptr_t optval, unsigned int optlen)
759 struct sock *sk = (struct sock *)msk;
763 /* Limit to first subflow, before the connection establishment */
765 ssk = __mptcp_nmpc_sk(msk);
771 ret = tcp_setsockopt(ssk, level, optname, optval, optlen);
778 static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
779 sockptr_t optval, unsigned int optlen)
781 struct sock *sk = (void *)msk;
788 return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen);
789 case TCP_DEFER_ACCEPT:
790 /* See tcp.c: TCP_DEFER_ACCEPT does not fail */
791 mptcp_setsockopt_first_sf_only(msk, SOL_TCP, optname, optval, optlen);
794 case TCP_FASTOPEN_CONNECT:
795 case TCP_FASTOPEN_KEY:
796 case TCP_FASTOPEN_NO_COOKIE:
797 return mptcp_setsockopt_first_sf_only(msk, SOL_TCP, optname,
801 ret = mptcp_get_int_option(msk, optval, optlen, &val);
808 if (val < 0 || val > 1)
811 msk->recvmsg_inq = !!val;
813 case TCP_NOTSENT_LOWAT:
814 WRITE_ONCE(msk->notsent_lowat, val);
815 mptcp_write_space(sk);
818 ret = __mptcp_setsockopt_sol_tcp_cork(msk, val);
821 ret = __mptcp_setsockopt_sol_tcp_nodelay(msk, val);
831 int mptcp_setsockopt(struct sock *sk, int level, int optname,
832 sockptr_t optval, unsigned int optlen)
834 struct mptcp_sock *msk = mptcp_sk(sk);
837 pr_debug("msk=%p", msk);
839 if (level == SOL_SOCKET)
840 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
842 if (!mptcp_supported_sockopt(level, optname))
845 /* @@ the meaning of setsockopt() when the socket is connected and
846 * there are multiple subflows is not yet defined. It is up to the
847 * MPTCP-level socket to configure the subflows until the subflow
848 * is in TCP fallback, when TCP socket options are passed through
849 * to the one remaining subflow.
852 ssk = __mptcp_tcp_fallback(msk);
855 return tcp_setsockopt(ssk, level, optname, optval, optlen);
858 return mptcp_setsockopt_v4(msk, optname, optval, optlen);
860 if (level == SOL_IPV6)
861 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
863 if (level == SOL_TCP)
864 return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen);
869 static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
870 char __user *optval, int __user *optlen)
872 struct sock *sk = (struct sock *)msk;
879 ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
883 ssk = __mptcp_nmpc_sk(msk);
889 ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
896 void mptcp_diag_fill_info(struct mptcp_sock *msk, struct mptcp_info *info)
898 struct sock *sk = (struct sock *)msk;
902 memset(info, 0, sizeof(*info));
904 info->mptcpi_subflows = READ_ONCE(msk->pm.subflows);
905 info->mptcpi_add_addr_signal = READ_ONCE(msk->pm.add_addr_signaled);
906 info->mptcpi_add_addr_accepted = READ_ONCE(msk->pm.add_addr_accepted);
907 info->mptcpi_local_addr_used = READ_ONCE(msk->pm.local_addr_used);
909 if (inet_sk_state_load(sk) == TCP_LISTEN)
912 /* The following limits only make sense for the in-kernel PM */
913 if (mptcp_pm_is_kernel(msk)) {
914 info->mptcpi_subflows_max =
915 mptcp_pm_get_subflows_max(msk);
916 info->mptcpi_add_addr_signal_max =
917 mptcp_pm_get_add_addr_signal_max(msk);
918 info->mptcpi_add_addr_accepted_max =
919 mptcp_pm_get_add_addr_accept_max(msk);
920 info->mptcpi_local_addr_max =
921 mptcp_pm_get_local_addr_max(msk);
924 if (__mptcp_check_fallback(msk))
925 flags |= MPTCP_INFO_FLAG_FALLBACK;
926 if (READ_ONCE(msk->can_ack))
927 flags |= MPTCP_INFO_FLAG_REMOTE_KEY_RECEIVED;
928 info->mptcpi_flags = flags;
930 info->mptcpi_snd_una = msk->snd_una;
931 info->mptcpi_rcv_nxt = msk->ack_seq;
932 info->mptcpi_bytes_acked = msk->bytes_acked;
933 mptcp_data_unlock(sk);
935 slow = lock_sock_fast(sk);
936 info->mptcpi_csum_enabled = READ_ONCE(msk->csum_enabled);
937 info->mptcpi_token = msk->token;
938 info->mptcpi_write_seq = msk->write_seq;
939 info->mptcpi_retransmits = inet_csk(sk)->icsk_retransmits;
940 info->mptcpi_bytes_sent = msk->bytes_sent;
941 info->mptcpi_bytes_received = msk->bytes_received;
942 info->mptcpi_bytes_retrans = msk->bytes_retrans;
943 info->mptcpi_subflows_total = info->mptcpi_subflows +
944 __mptcp_has_initial_subflow(msk);
945 unlock_sock_fast(sk, slow);
947 EXPORT_SYMBOL_GPL(mptcp_diag_fill_info);
949 static int mptcp_getsockopt_info(struct mptcp_sock *msk, char __user *optval, int __user *optlen)
951 struct mptcp_info m_info;
954 if (get_user(len, optlen))
957 len = min_t(unsigned int, len, sizeof(struct mptcp_info));
959 mptcp_diag_fill_info(msk, &m_info);
961 if (put_user(len, optlen))
964 if (copy_to_user(optval, &m_info, len))
970 static int mptcp_put_subflow_data(struct mptcp_subflow_data *sfd,
975 u32 copylen = min_t(u32, sfd->size_subflow_data, sizeof(*sfd));
978 copied += sfd->size_subflow_data;
982 if (put_user(copied, optlen))
985 if (copy_to_user(optval, sfd, copylen))
991 static int mptcp_get_subflow_data(struct mptcp_subflow_data *sfd,
997 if (get_user(len, optlen))
1000 /* if mptcp_subflow_data size is changed, need to adjust
1001 * this function to deal with programs using old version.
1003 BUILD_BUG_ON(sizeof(*sfd) != MIN_INFO_OPTLEN_SIZE);
1005 if (len < MIN_INFO_OPTLEN_SIZE)
1008 memset(sfd, 0, sizeof(*sfd));
1010 copylen = min_t(unsigned int, len, sizeof(*sfd));
1011 if (copy_from_user(sfd, optval, copylen))
1014 /* size_subflow_data is u32, but len is signed */
1015 if (sfd->size_subflow_data > INT_MAX ||
1016 sfd->size_user > INT_MAX)
1019 if (sfd->size_subflow_data < MIN_INFO_OPTLEN_SIZE ||
1020 sfd->size_subflow_data > len)
1023 if (sfd->num_subflows || sfd->size_kernel)
1026 return len - sfd->size_subflow_data;
1029 static int mptcp_getsockopt_tcpinfo(struct mptcp_sock *msk, char __user *optval,
1032 struct mptcp_subflow_context *subflow;
1033 struct sock *sk = (struct sock *)msk;
1034 unsigned int sfcount = 0, copied = 0;
1035 struct mptcp_subflow_data sfd;
1036 char __user *infoptr;
1039 len = mptcp_get_subflow_data(&sfd, optval, optlen);
1043 sfd.size_kernel = sizeof(struct tcp_info);
1044 sfd.size_user = min_t(unsigned int, sfd.size_user,
1045 sizeof(struct tcp_info));
1047 infoptr = optval + sfd.size_subflow_data;
1051 mptcp_for_each_subflow(msk, subflow) {
1052 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1056 if (len && len >= sfd.size_user) {
1057 struct tcp_info info;
1059 tcp_get_info(ssk, &info);
1061 if (copy_to_user(infoptr, &info, sfd.size_user)) {
1066 infoptr += sfd.size_user;
1067 copied += sfd.size_user;
1068 len -= sfd.size_user;
1074 sfd.num_subflows = sfcount;
1076 if (mptcp_put_subflow_data(&sfd, optval, copied, optlen))
1082 static void mptcp_get_sub_addrs(const struct sock *sk, struct mptcp_subflow_addrs *a)
1084 const struct inet_sock *inet = inet_sk(sk);
1086 memset(a, 0, sizeof(*a));
1088 if (sk->sk_family == AF_INET) {
1089 a->sin_local.sin_family = AF_INET;
1090 a->sin_local.sin_port = inet->inet_sport;
1091 a->sin_local.sin_addr.s_addr = inet->inet_rcv_saddr;
1093 if (!a->sin_local.sin_addr.s_addr)
1094 a->sin_local.sin_addr.s_addr = inet->inet_saddr;
1096 a->sin_remote.sin_family = AF_INET;
1097 a->sin_remote.sin_port = inet->inet_dport;
1098 a->sin_remote.sin_addr.s_addr = inet->inet_daddr;
1099 #if IS_ENABLED(CONFIG_IPV6)
1100 } else if (sk->sk_family == AF_INET6) {
1101 const struct ipv6_pinfo *np = inet6_sk(sk);
1103 if (WARN_ON_ONCE(!np))
1106 a->sin6_local.sin6_family = AF_INET6;
1107 a->sin6_local.sin6_port = inet->inet_sport;
1109 if (ipv6_addr_any(&sk->sk_v6_rcv_saddr))
1110 a->sin6_local.sin6_addr = np->saddr;
1112 a->sin6_local.sin6_addr = sk->sk_v6_rcv_saddr;
1114 a->sin6_remote.sin6_family = AF_INET6;
1115 a->sin6_remote.sin6_port = inet->inet_dport;
1116 a->sin6_remote.sin6_addr = sk->sk_v6_daddr;
1121 static int mptcp_getsockopt_subflow_addrs(struct mptcp_sock *msk, char __user *optval,
1124 struct mptcp_subflow_context *subflow;
1125 struct sock *sk = (struct sock *)msk;
1126 unsigned int sfcount = 0, copied = 0;
1127 struct mptcp_subflow_data sfd;
1128 char __user *addrptr;
1131 len = mptcp_get_subflow_data(&sfd, optval, optlen);
1135 sfd.size_kernel = sizeof(struct mptcp_subflow_addrs);
1136 sfd.size_user = min_t(unsigned int, sfd.size_user,
1137 sizeof(struct mptcp_subflow_addrs));
1139 addrptr = optval + sfd.size_subflow_data;
1143 mptcp_for_each_subflow(msk, subflow) {
1144 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1148 if (len && len >= sfd.size_user) {
1149 struct mptcp_subflow_addrs a;
1151 mptcp_get_sub_addrs(ssk, &a);
1153 if (copy_to_user(addrptr, &a, sfd.size_user)) {
1158 addrptr += sfd.size_user;
1159 copied += sfd.size_user;
1160 len -= sfd.size_user;
1166 sfd.num_subflows = sfcount;
1168 if (mptcp_put_subflow_data(&sfd, optval, copied, optlen))
1174 static int mptcp_get_full_info(struct mptcp_full_info *mfi,
1175 char __user *optval,
1180 BUILD_BUG_ON(offsetof(struct mptcp_full_info, mptcp_info) !=
1181 MIN_FULL_INFO_OPTLEN_SIZE);
1183 if (get_user(len, optlen))
1186 if (len < MIN_FULL_INFO_OPTLEN_SIZE)
1189 memset(mfi, 0, sizeof(*mfi));
1190 if (copy_from_user(mfi, optval, MIN_FULL_INFO_OPTLEN_SIZE))
1193 if (mfi->size_tcpinfo_kernel ||
1194 mfi->size_sfinfo_kernel ||
1198 if (mfi->size_sfinfo_user > INT_MAX ||
1199 mfi->size_tcpinfo_user > INT_MAX)
1202 return len - MIN_FULL_INFO_OPTLEN_SIZE;
1205 static int mptcp_put_full_info(struct mptcp_full_info *mfi,
1206 char __user *optval,
1210 copylen += MIN_FULL_INFO_OPTLEN_SIZE;
1211 if (put_user(copylen, optlen))
1214 if (copy_to_user(optval, mfi, copylen))
1219 static int mptcp_getsockopt_full_info(struct mptcp_sock *msk, char __user *optval,
1222 unsigned int sfcount = 0, copylen = 0;
1223 struct mptcp_subflow_context *subflow;
1224 struct sock *sk = (struct sock *)msk;
1225 void __user *tcpinfoptr, *sfinfoptr;
1226 struct mptcp_full_info mfi;
1229 len = mptcp_get_full_info(&mfi, optval, optlen);
1233 /* don't bother filling the mptcp info if there is not enough
1234 * user-space-provided storage
1237 mptcp_diag_fill_info(msk, &mfi.mptcp_info);
1238 copylen += min_t(unsigned int, len, sizeof(struct mptcp_info));
1241 mfi.size_tcpinfo_kernel = sizeof(struct tcp_info);
1242 mfi.size_tcpinfo_user = min_t(unsigned int, mfi.size_tcpinfo_user,
1243 sizeof(struct tcp_info));
1244 sfinfoptr = u64_to_user_ptr(mfi.subflow_info);
1245 mfi.size_sfinfo_kernel = sizeof(struct mptcp_subflow_info);
1246 mfi.size_sfinfo_user = min_t(unsigned int, mfi.size_sfinfo_user,
1247 sizeof(struct mptcp_subflow_info));
1248 tcpinfoptr = u64_to_user_ptr(mfi.tcp_info);
1251 mptcp_for_each_subflow(msk, subflow) {
1252 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1253 struct mptcp_subflow_info sfinfo;
1254 struct tcp_info tcp_info;
1256 if (sfcount++ >= mfi.size_arrays_user)
1259 /* fetch addr/tcp_info only if the user space buffers
1262 memset(&sfinfo, 0, sizeof(sfinfo));
1263 sfinfo.id = subflow->subflow_id;
1264 if (mfi.size_sfinfo_user >
1265 offsetof(struct mptcp_subflow_info, addrs))
1266 mptcp_get_sub_addrs(ssk, &sfinfo.addrs);
1267 if (copy_to_user(sfinfoptr, &sfinfo, mfi.size_sfinfo_user))
1270 if (mfi.size_tcpinfo_user) {
1271 tcp_get_info(ssk, &tcp_info);
1272 if (copy_to_user(tcpinfoptr, &tcp_info,
1273 mfi.size_tcpinfo_user))
1277 tcpinfoptr += mfi.size_tcpinfo_user;
1278 sfinfoptr += mfi.size_sfinfo_user;
1282 mfi.num_subflows = sfcount;
1283 if (mptcp_put_full_info(&mfi, optval, copylen, optlen))
1293 static int mptcp_put_int_option(struct mptcp_sock *msk, char __user *optval,
1294 int __user *optlen, int val)
1298 if (get_user(len, optlen))
1303 if (len < sizeof(int) && len > 0 && val >= 0 && val <= 255) {
1304 unsigned char ucval = (unsigned char)val;
1307 if (put_user(len, optlen))
1309 if (copy_to_user(optval, &ucval, 1))
1312 len = min_t(unsigned int, len, sizeof(int));
1313 if (put_user(len, optlen))
1315 if (copy_to_user(optval, &val, len))
1322 static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
1323 char __user *optval, int __user *optlen)
1327 case TCP_CONGESTION:
1330 case TCP_DEFER_ACCEPT:
1332 case TCP_FASTOPEN_CONNECT:
1333 case TCP_FASTOPEN_KEY:
1334 case TCP_FASTOPEN_NO_COOKIE:
1335 return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname,
1338 return mptcp_put_int_option(msk, optval, optlen, msk->recvmsg_inq);
1340 return mptcp_put_int_option(msk, optval, optlen, msk->cork);
1342 return mptcp_put_int_option(msk, optval, optlen, msk->nodelay);
1343 case TCP_NOTSENT_LOWAT:
1344 return mptcp_put_int_option(msk, optval, optlen, msk->notsent_lowat);
1349 static int mptcp_getsockopt_v4(struct mptcp_sock *msk, int optname,
1350 char __user *optval, int __user *optlen)
1352 struct sock *sk = (void *)msk;
1356 return mptcp_put_int_option(msk, optval, optlen, READ_ONCE(inet_sk(sk)->tos));
1357 case IP_BIND_ADDRESS_NO_PORT:
1358 return mptcp_put_int_option(msk, optval, optlen,
1359 inet_test_bit(BIND_ADDRESS_NO_PORT, sk));
1360 case IP_LOCAL_PORT_RANGE:
1361 return mptcp_put_int_option(msk, optval, optlen,
1362 READ_ONCE(inet_sk(sk)->local_port_range));
1368 static int mptcp_getsockopt_sol_mptcp(struct mptcp_sock *msk, int optname,
1369 char __user *optval, int __user *optlen)
1373 return mptcp_getsockopt_info(msk, optval, optlen);
1374 case MPTCP_FULL_INFO:
1375 return mptcp_getsockopt_full_info(msk, optval, optlen);
1377 return mptcp_getsockopt_tcpinfo(msk, optval, optlen);
1378 case MPTCP_SUBFLOW_ADDRS:
1379 return mptcp_getsockopt_subflow_addrs(msk, optval, optlen);
1385 int mptcp_getsockopt(struct sock *sk, int level, int optname,
1386 char __user *optval, int __user *option)
1388 struct mptcp_sock *msk = mptcp_sk(sk);
1391 pr_debug("msk=%p", msk);
1393 /* @@ the meaning of setsockopt() when the socket is connected and
1394 * there are multiple subflows is not yet defined. It is up to the
1395 * MPTCP-level socket to configure the subflows until the subflow
1396 * is in TCP fallback, when socket options are passed through
1397 * to the one remaining subflow.
1400 ssk = __mptcp_tcp_fallback(msk);
1403 return tcp_getsockopt(ssk, level, optname, optval, option);
1405 if (level == SOL_IP)
1406 return mptcp_getsockopt_v4(msk, optname, optval, option);
1407 if (level == SOL_TCP)
1408 return mptcp_getsockopt_sol_tcp(msk, optname, optval, option);
1409 if (level == SOL_MPTCP)
1410 return mptcp_getsockopt_sol_mptcp(msk, optname, optval, option);
1414 static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
1416 static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK;
1417 struct sock *sk = (struct sock *)msk;
1419 if (ssk->sk_prot->keepalive) {
1420 if (sock_flag(sk, SOCK_KEEPOPEN))
1421 ssk->sk_prot->keepalive(ssk, 1);
1423 ssk->sk_prot->keepalive(ssk, 0);
1426 ssk->sk_priority = sk->sk_priority;
1427 ssk->sk_bound_dev_if = sk->sk_bound_dev_if;
1428 ssk->sk_incoming_cpu = sk->sk_incoming_cpu;
1429 ssk->sk_ipv6only = sk->sk_ipv6only;
1430 __ip_sock_set_tos(ssk, inet_sk(sk)->tos);
1432 if (sk->sk_userlocks & tx_rx_locks) {
1433 ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks;
1434 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) {
1435 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
1436 mptcp_subflow_ctx(ssk)->cached_sndbuf = sk->sk_sndbuf;
1438 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1439 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
1442 if (sock_flag(sk, SOCK_LINGER)) {
1443 ssk->sk_lingertime = sk->sk_lingertime;
1444 sock_set_flag(ssk, SOCK_LINGER);
1446 sock_reset_flag(ssk, SOCK_LINGER);
1449 if (sk->sk_mark != ssk->sk_mark) {
1450 ssk->sk_mark = sk->sk_mark;
1454 sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG));
1456 if (inet_csk(sk)->icsk_ca_ops != inet_csk(ssk)->icsk_ca_ops)
1457 tcp_set_congestion_control(ssk, msk->ca_name, false, true);
1458 __tcp_sock_set_cork(ssk, !!msk->cork);
1459 __tcp_sock_set_nodelay(ssk, !!msk->nodelay);
1461 inet_assign_bit(TRANSPARENT, ssk, inet_test_bit(TRANSPARENT, sk));
1462 inet_assign_bit(FREEBIND, ssk, inet_test_bit(FREEBIND, sk));
1463 inet_assign_bit(BIND_ADDRESS_NO_PORT, ssk, inet_test_bit(BIND_ADDRESS_NO_PORT, sk));
1464 WRITE_ONCE(inet_sk(ssk)->local_port_range, READ_ONCE(inet_sk(sk)->local_port_range));
1467 void mptcp_sockopt_sync_locked(struct mptcp_sock *msk, struct sock *ssk)
1469 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1471 msk_owned_by_me(msk);
1473 ssk->sk_rcvlowat = 0;
1475 /* subflows must ignore any latency-related settings: will not affect
1476 * the user-space - only the msk is relevant - but will foul the
1479 tcp_sk(ssk)->notsent_lowat = UINT_MAX;
1481 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
1482 sync_socket_options(msk, ssk);
1484 subflow->setsockopt_seq = msk->setsockopt_seq;
1488 /* unfortunately this is different enough from the tcp version so
1489 * that we can't factor it out
1491 int mptcp_set_rcvlowat(struct sock *sk, int val)
1493 struct mptcp_subflow_context *subflow;
1496 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1497 cap = sk->sk_rcvbuf >> 1;
1499 cap = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]) >> 1;
1500 val = min(val, cap);
1501 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1503 /* Check if we need to signal EPOLLIN right now */
1504 if (mptcp_epollin_ready(sk))
1505 sk->sk_data_ready(sk);
1507 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1510 space = __tcp_space_from_win(mptcp_sk(sk)->scaling_ratio, val);
1511 if (space <= sk->sk_rcvbuf)
1514 /* propagate the rcvbuf changes to all the subflows */
1515 WRITE_ONCE(sk->sk_rcvbuf, space);
1516 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
1517 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1520 slow = lock_sock_fast(ssk);
1521 WRITE_ONCE(ssk->sk_rcvbuf, space);
1522 tcp_sk(ssk)->window_clamp = val;
1523 unlock_sock_fast(ssk, slow);