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 ssk->sk_priority = val;
96 ssk->sk_userlocks |= SOCK_SNDBUF_LOCK;
97 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
101 ssk->sk_userlocks |= SOCK_RCVBUF_LOCK;
102 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
105 if (READ_ONCE(ssk->sk_mark) != sk->sk_mark) {
106 WRITE_ONCE(ssk->sk_mark, sk->sk_mark);
110 case SO_INCOMING_CPU:
111 WRITE_ONCE(ssk->sk_incoming_cpu, val);
115 subflow->setsockopt_seq = msk->setsockopt_seq;
116 unlock_sock_fast(ssk, slow);
122 static int mptcp_sol_socket_intval(struct mptcp_sock *msk, int optname, int val)
124 sockptr_t optval = KERNEL_SOCKPTR(&val);
125 struct sock *sk = (struct sock *)msk;
128 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
129 optval, sizeof(val));
133 mptcp_sol_socket_sync_intval(msk, optname, val);
137 static void mptcp_so_incoming_cpu(struct mptcp_sock *msk, int val)
139 struct sock *sk = (struct sock *)msk;
141 WRITE_ONCE(sk->sk_incoming_cpu, val);
143 mptcp_sol_socket_sync_intval(msk, SO_INCOMING_CPU, val);
146 static int mptcp_setsockopt_sol_socket_tstamp(struct mptcp_sock *msk, int optname, int val)
148 sockptr_t optval = KERNEL_SOCKPTR(&val);
149 struct mptcp_subflow_context *subflow;
150 struct sock *sk = (struct sock *)msk;
153 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
154 optval, sizeof(val));
159 mptcp_for_each_subflow(msk, subflow) {
160 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
161 bool slow = lock_sock_fast(ssk);
163 sock_set_timestamp(sk, optname, !!val);
164 unlock_sock_fast(ssk, slow);
171 static int mptcp_setsockopt_sol_socket_int(struct mptcp_sock *msk, int optname,
177 ret = mptcp_get_int_option(msk, optval, optlen, &val);
183 mptcp_sol_socket_sync_intval(msk, optname, val);
192 return mptcp_sol_socket_intval(msk, optname, val);
193 case SO_INCOMING_CPU:
194 mptcp_so_incoming_cpu(msk, val);
196 case SO_TIMESTAMP_OLD:
197 case SO_TIMESTAMP_NEW:
198 case SO_TIMESTAMPNS_OLD:
199 case SO_TIMESTAMPNS_NEW:
200 return mptcp_setsockopt_sol_socket_tstamp(msk, optname, val);
206 static int mptcp_setsockopt_sol_socket_timestamping(struct mptcp_sock *msk,
211 struct mptcp_subflow_context *subflow;
212 struct sock *sk = (struct sock *)msk;
213 struct so_timestamping timestamping;
216 if (optlen == sizeof(timestamping)) {
217 if (copy_from_sockptr(×tamping, optval,
218 sizeof(timestamping)))
220 } else if (optlen == sizeof(int)) {
221 memset(×tamping, 0, sizeof(timestamping));
223 if (copy_from_sockptr(×tamping.flags, optval, sizeof(int)))
229 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname,
230 KERNEL_SOCKPTR(×tamping),
231 sizeof(timestamping));
237 mptcp_for_each_subflow(msk, subflow) {
238 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
239 bool slow = lock_sock_fast(ssk);
241 sock_set_timestamping(sk, optname, timestamping);
242 unlock_sock_fast(ssk, slow);
250 static int mptcp_setsockopt_sol_socket_linger(struct mptcp_sock *msk, sockptr_t optval,
253 struct mptcp_subflow_context *subflow;
254 struct sock *sk = (struct sock *)msk;
259 if (optlen < sizeof(ling))
262 if (copy_from_sockptr(&ling, optval, sizeof(ling)))
265 kopt = KERNEL_SOCKPTR(&ling);
266 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, SO_LINGER, kopt, sizeof(ling));
271 sockopt_seq_inc(msk);
272 mptcp_for_each_subflow(msk, subflow) {
273 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
274 bool slow = lock_sock_fast(ssk);
277 sock_reset_flag(ssk, SOCK_LINGER);
279 ssk->sk_lingertime = sk->sk_lingertime;
280 sock_set_flag(ssk, SOCK_LINGER);
283 subflow->setsockopt_seq = msk->setsockopt_seq;
284 unlock_sock_fast(ssk, slow);
291 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
292 sockptr_t optval, unsigned int optlen)
294 struct sock *sk = (struct sock *)msk;
301 case SO_BINDTODEVICE:
302 case SO_BINDTOIFINDEX:
304 ssk = __mptcp_nmpc_sk(msk);
310 ret = sk_setsockopt(ssk, SOL_SOCKET, optname, optval, optlen);
312 if (optname == SO_REUSEPORT)
313 sk->sk_reuseport = ssk->sk_reuseport;
314 else if (optname == SO_REUSEADDR)
315 sk->sk_reuse = ssk->sk_reuse;
316 else if (optname == SO_BINDTODEVICE)
317 sk->sk_bound_dev_if = ssk->sk_bound_dev_if;
318 else if (optname == SO_BINDTOIFINDEX)
319 sk->sk_bound_dev_if = ssk->sk_bound_dev_if;
330 case SO_INCOMING_CPU:
332 case SO_TIMESTAMP_OLD:
333 case SO_TIMESTAMP_NEW:
334 case SO_TIMESTAMPNS_OLD:
335 case SO_TIMESTAMPNS_NEW:
336 return mptcp_setsockopt_sol_socket_int(msk, optname, optval,
338 case SO_TIMESTAMPING_OLD:
339 case SO_TIMESTAMPING_NEW:
340 return mptcp_setsockopt_sol_socket_timestamping(msk, optname,
343 return mptcp_setsockopt_sol_socket_linger(msk, optval, optlen);
345 case SO_RCVTIMEO_OLD:
346 case SO_RCVTIMEO_NEW:
347 case SO_SNDTIMEO_OLD:
348 case SO_SNDTIMEO_NEW:
350 case SO_PREFER_BUSY_POLL:
351 case SO_BUSY_POLL_BUDGET:
352 /* No need to copy: only relevant for msk */
353 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
364 case SO_SELECT_ERR_QUEUE:
368 /* SO_OOBINLINE is not supported, let's avoid the related mess
369 * SO_ATTACH_FILTER, SO_ATTACH_BPF, SO_ATTACH_REUSEPORT_CBPF,
370 * SO_DETACH_REUSEPORT_BPF, SO_DETACH_FILTER, SO_LOCK_FILTER,
371 * we must be careful with subflows
373 * SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks
374 * explicitly the sk_protocol field
376 * SO_PEEK_OFF is unsupported, as it is for plain TCP
377 * SO_MAX_PACING_RATE is unsupported, we must be careful with subflows
378 * SO_CNX_ADVICE is currently unsupported, could possibly be relevant,
379 * but likely needs careful design
381 * SO_ZEROCOPY is currently unsupported, TODO in sndmsg
382 * SO_TXTIME is currently unsupported
388 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
389 sockptr_t optval, unsigned int optlen)
391 struct sock *sk = (struct sock *)msk;
392 int ret = -EOPNOTSUPP;
397 case IPV6_TRANSPARENT:
400 ssk = __mptcp_nmpc_sk(msk);
406 ret = tcp_setsockopt(ssk, SOL_IPV6, optname, optval, optlen);
412 sockopt_seq_inc(msk);
416 sk->sk_ipv6only = ssk->sk_ipv6only;
418 case IPV6_TRANSPARENT:
419 inet_assign_bit(TRANSPARENT, sk,
420 inet_test_bit(TRANSPARENT, ssk));
423 inet_assign_bit(FREEBIND, sk,
424 inet_test_bit(FREEBIND, ssk));
435 static bool mptcp_supported_sockopt(int level, int optname)
437 if (level == SOL_IP) {
439 /* should work fine */
443 /* the following are control cmsg related */
450 case IP_RECVORIGDSTADDR:
452 case IP_RECVFRAGSIZE:
454 /* common stuff that need some love */
457 case IP_BIND_ADDRESS_NO_PORT:
458 case IP_MTU_DISCOVER:
461 /* possibly less common may deserve some love */
464 /* the following is apparently a no-op for plain TCP */
465 case IP_RECVERR_RFC4884:
469 /* IP_OPTIONS is not supported, needs subflow care */
470 /* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */
471 /* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF,
472 * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP,
473 * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE,
474 * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP,
475 * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE,
476 * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal
479 /* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */
482 if (level == SOL_IPV6) {
486 /* the following are control cmsg related */
487 case IPV6_RECVPKTINFO:
488 case IPV6_2292PKTINFO:
489 case IPV6_RECVHOPLIMIT:
490 case IPV6_2292HOPLIMIT:
493 case IPV6_RECVHOPOPTS:
494 case IPV6_2292HOPOPTS:
495 case IPV6_RECVDSTOPTS:
496 case IPV6_2292DSTOPTS:
497 case IPV6_RECVTCLASS:
499 case IPV6_RECVPATHMTU:
500 case IPV6_RECVORIGDSTADDR:
501 case IPV6_RECVFRAGSIZE:
503 /* the following ones need some love but are quite common */
505 case IPV6_TRANSPARENT:
508 case IPV6_2292PKTOPTIONS:
509 case IPV6_UNICAST_HOPS:
510 case IPV6_MTU_DISCOVER:
513 case IPV6_FLOWINFO_SEND:
514 case IPV6_FLOWLABEL_MGR:
515 case IPV6_MINHOPCOUNT:
517 case IPV6_AUTOFLOWLABEL:
519 /* the following one is a no-op for plain TCP */
520 case IPV6_RECVERR_RFC4884:
524 /* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are
527 /* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF,
528 * IPV6_MULTICAST_IF, IPV6_ADDRFORM,
529 * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST,
530 * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP,
531 * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP,
532 * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER
533 * are not supported better not deal with mcast
535 /* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */
537 /* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */
538 /* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */
541 if (level == SOL_TCP) {
543 /* the following are no-op or should work just fine */
544 case TCP_THIN_DUPACK:
545 case TCP_DEFER_ACCEPT:
547 /* the following need some love */
550 case TCP_THIN_LINEAR_TIMEOUTS:
559 case TCP_WINDOW_CLAMP:
561 case TCP_USER_TIMEOUT:
563 case TCP_NOTSENT_LOWAT:
567 case TCP_FASTOPEN_CONNECT:
568 case TCP_FASTOPEN_KEY:
569 case TCP_FASTOPEN_NO_COOKIE:
573 /* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */
575 /* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS,
576 * TCP_REPAIR_WINDOW are not supported, better avoid this mess
582 static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval,
585 struct mptcp_subflow_context *subflow;
586 struct sock *sk = (struct sock *)msk;
587 char name[TCP_CA_NAME_MAX];
594 ret = strncpy_from_sockptr(name, optval,
595 min_t(long, TCP_CA_NAME_MAX - 1, optlen));
601 cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN);
605 sockopt_seq_inc(msk);
606 mptcp_for_each_subflow(msk, subflow) {
607 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
611 err = tcp_set_congestion_control(ssk, name, true, cap_net_admin);
612 if (err < 0 && ret == 0)
614 subflow->setsockopt_seq = msk->setsockopt_seq;
619 strcpy(msk->ca_name, name);
625 static int mptcp_setsockopt_sol_tcp_cork(struct mptcp_sock *msk, sockptr_t optval,
628 struct mptcp_subflow_context *subflow;
629 struct sock *sk = (struct sock *)msk;
632 if (optlen < sizeof(int))
635 if (copy_from_sockptr(&val, optval, sizeof(val)))
639 sockopt_seq_inc(msk);
641 mptcp_for_each_subflow(msk, subflow) {
642 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
645 __tcp_sock_set_cork(ssk, !!val);
649 mptcp_check_and_set_pending(sk);
655 static int mptcp_setsockopt_sol_tcp_nodelay(struct mptcp_sock *msk, sockptr_t optval,
658 struct mptcp_subflow_context *subflow;
659 struct sock *sk = (struct sock *)msk;
662 if (optlen < sizeof(int))
665 if (copy_from_sockptr(&val, optval, sizeof(val)))
669 sockopt_seq_inc(msk);
670 msk->nodelay = !!val;
671 mptcp_for_each_subflow(msk, subflow) {
672 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
675 __tcp_sock_set_nodelay(ssk, !!val);
679 mptcp_check_and_set_pending(sk);
685 static int mptcp_setsockopt_sol_ip_set_transparent(struct mptcp_sock *msk, int optname,
686 sockptr_t optval, unsigned int optlen)
688 struct sock *sk = (struct sock *)msk;
692 err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen);
698 ssk = __mptcp_nmpc_sk(msk);
706 inet_assign_bit(FREEBIND, ssk, inet_test_bit(FREEBIND, sk));
709 inet_assign_bit(TRANSPARENT, ssk,
710 inet_test_bit(TRANSPARENT, sk));
718 sockopt_seq_inc(msk);
723 static int mptcp_setsockopt_v4_set_tos(struct mptcp_sock *msk, int optname,
724 sockptr_t optval, unsigned int optlen)
726 struct mptcp_subflow_context *subflow;
727 struct sock *sk = (struct sock *)msk;
730 err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen);
736 sockopt_seq_inc(msk);
737 val = inet_sk(sk)->tos;
738 mptcp_for_each_subflow(msk, subflow) {
739 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
741 __ip_sock_set_tos(ssk, val);
748 static int mptcp_setsockopt_v4(struct mptcp_sock *msk, int optname,
749 sockptr_t optval, unsigned int optlen)
754 return mptcp_setsockopt_sol_ip_set_transparent(msk, optname, optval, optlen);
756 return mptcp_setsockopt_v4_set_tos(msk, optname, optval, optlen);
762 static int mptcp_setsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
763 sockptr_t optval, unsigned int optlen)
765 struct sock *sk = (struct sock *)msk;
769 /* Limit to first subflow, before the connection establishment */
771 ssk = __mptcp_nmpc_sk(msk);
777 ret = tcp_setsockopt(ssk, level, optname, optval, optlen);
784 static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
785 sockptr_t optval, unsigned int optlen)
787 struct sock *sk = (void *)msk;
792 ret = mptcp_get_int_option(msk, optval, optlen, &val);
795 if (val < 0 || val > 1)
799 msk->recvmsg_inq = !!val;
805 return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen);
807 return mptcp_setsockopt_sol_tcp_cork(msk, optval, optlen);
809 return mptcp_setsockopt_sol_tcp_nodelay(msk, optval, optlen);
810 case TCP_DEFER_ACCEPT:
811 /* See tcp.c: TCP_DEFER_ACCEPT does not fail */
812 mptcp_setsockopt_first_sf_only(msk, SOL_TCP, optname, optval, optlen);
815 case TCP_FASTOPEN_CONNECT:
816 case TCP_FASTOPEN_KEY:
817 case TCP_FASTOPEN_NO_COOKIE:
818 return mptcp_setsockopt_first_sf_only(msk, SOL_TCP, optname,
825 int mptcp_setsockopt(struct sock *sk, int level, int optname,
826 sockptr_t optval, unsigned int optlen)
828 struct mptcp_sock *msk = mptcp_sk(sk);
831 pr_debug("msk=%p", msk);
833 if (level == SOL_SOCKET)
834 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
836 if (!mptcp_supported_sockopt(level, optname))
839 /* @@ the meaning of setsockopt() when the socket is connected and
840 * there are multiple subflows is not yet defined. It is up to the
841 * MPTCP-level socket to configure the subflows until the subflow
842 * is in TCP fallback, when TCP socket options are passed through
843 * to the one remaining subflow.
846 ssk = __mptcp_tcp_fallback(msk);
849 return tcp_setsockopt(ssk, level, optname, optval, optlen);
852 return mptcp_setsockopt_v4(msk, optname, optval, optlen);
854 if (level == SOL_IPV6)
855 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
857 if (level == SOL_TCP)
858 return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen);
863 static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname,
864 char __user *optval, int __user *optlen)
866 struct sock *sk = (struct sock *)msk;
873 ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
877 ssk = __mptcp_nmpc_sk(msk);
883 ret = tcp_getsockopt(ssk, level, optname, optval, optlen);
890 void mptcp_diag_fill_info(struct mptcp_sock *msk, struct mptcp_info *info)
892 struct sock *sk = (struct sock *)msk;
896 memset(info, 0, sizeof(*info));
898 info->mptcpi_subflows = READ_ONCE(msk->pm.subflows);
899 info->mptcpi_add_addr_signal = READ_ONCE(msk->pm.add_addr_signaled);
900 info->mptcpi_add_addr_accepted = READ_ONCE(msk->pm.add_addr_accepted);
901 info->mptcpi_local_addr_used = READ_ONCE(msk->pm.local_addr_used);
903 if (inet_sk_state_load(sk) == TCP_LISTEN)
906 /* The following limits only make sense for the in-kernel PM */
907 if (mptcp_pm_is_kernel(msk)) {
908 info->mptcpi_subflows_max =
909 mptcp_pm_get_subflows_max(msk);
910 info->mptcpi_add_addr_signal_max =
911 mptcp_pm_get_add_addr_signal_max(msk);
912 info->mptcpi_add_addr_accepted_max =
913 mptcp_pm_get_add_addr_accept_max(msk);
914 info->mptcpi_local_addr_max =
915 mptcp_pm_get_local_addr_max(msk);
918 if (test_bit(MPTCP_FALLBACK_DONE, &msk->flags))
919 flags |= MPTCP_INFO_FLAG_FALLBACK;
920 if (READ_ONCE(msk->can_ack))
921 flags |= MPTCP_INFO_FLAG_REMOTE_KEY_RECEIVED;
922 info->mptcpi_flags = flags;
924 info->mptcpi_snd_una = msk->snd_una;
925 info->mptcpi_rcv_nxt = msk->ack_seq;
926 info->mptcpi_bytes_acked = msk->bytes_acked;
927 mptcp_data_unlock(sk);
929 slow = lock_sock_fast(sk);
930 info->mptcpi_csum_enabled = msk->csum_enabled;
931 info->mptcpi_token = msk->token;
932 info->mptcpi_write_seq = msk->write_seq;
933 info->mptcpi_retransmits = inet_csk(sk)->icsk_retransmits;
934 info->mptcpi_bytes_sent = msk->bytes_sent;
935 info->mptcpi_bytes_received = msk->bytes_received;
936 info->mptcpi_bytes_retrans = msk->bytes_retrans;
937 unlock_sock_fast(sk, slow);
939 EXPORT_SYMBOL_GPL(mptcp_diag_fill_info);
941 static int mptcp_getsockopt_info(struct mptcp_sock *msk, char __user *optval, int __user *optlen)
943 struct mptcp_info m_info;
946 if (get_user(len, optlen))
949 len = min_t(unsigned int, len, sizeof(struct mptcp_info));
951 mptcp_diag_fill_info(msk, &m_info);
953 if (put_user(len, optlen))
956 if (copy_to_user(optval, &m_info, len))
962 static int mptcp_put_subflow_data(struct mptcp_subflow_data *sfd,
967 u32 copylen = min_t(u32, sfd->size_subflow_data, sizeof(*sfd));
970 copied += sfd->size_subflow_data;
974 if (put_user(copied, optlen))
977 if (copy_to_user(optval, sfd, copylen))
983 static int mptcp_get_subflow_data(struct mptcp_subflow_data *sfd,
989 if (get_user(len, optlen))
992 /* if mptcp_subflow_data size is changed, need to adjust
993 * this function to deal with programs using old version.
995 BUILD_BUG_ON(sizeof(*sfd) != MIN_INFO_OPTLEN_SIZE);
997 if (len < MIN_INFO_OPTLEN_SIZE)
1000 memset(sfd, 0, sizeof(*sfd));
1002 copylen = min_t(unsigned int, len, sizeof(*sfd));
1003 if (copy_from_user(sfd, optval, copylen))
1006 /* size_subflow_data is u32, but len is signed */
1007 if (sfd->size_subflow_data > INT_MAX ||
1008 sfd->size_user > INT_MAX)
1011 if (sfd->size_subflow_data < MIN_INFO_OPTLEN_SIZE ||
1012 sfd->size_subflow_data > len)
1015 if (sfd->num_subflows || sfd->size_kernel)
1018 return len - sfd->size_subflow_data;
1021 static int mptcp_getsockopt_tcpinfo(struct mptcp_sock *msk, char __user *optval,
1024 struct mptcp_subflow_context *subflow;
1025 struct sock *sk = (struct sock *)msk;
1026 unsigned int sfcount = 0, copied = 0;
1027 struct mptcp_subflow_data sfd;
1028 char __user *infoptr;
1031 len = mptcp_get_subflow_data(&sfd, optval, optlen);
1035 sfd.size_kernel = sizeof(struct tcp_info);
1036 sfd.size_user = min_t(unsigned int, sfd.size_user,
1037 sizeof(struct tcp_info));
1039 infoptr = optval + sfd.size_subflow_data;
1043 mptcp_for_each_subflow(msk, subflow) {
1044 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1048 if (len && len >= sfd.size_user) {
1049 struct tcp_info info;
1051 tcp_get_info(ssk, &info);
1053 if (copy_to_user(infoptr, &info, sfd.size_user)) {
1058 infoptr += sfd.size_user;
1059 copied += sfd.size_user;
1060 len -= sfd.size_user;
1066 sfd.num_subflows = sfcount;
1068 if (mptcp_put_subflow_data(&sfd, optval, copied, optlen))
1074 static void mptcp_get_sub_addrs(const struct sock *sk, struct mptcp_subflow_addrs *a)
1076 const struct inet_sock *inet = inet_sk(sk);
1078 memset(a, 0, sizeof(*a));
1080 if (sk->sk_family == AF_INET) {
1081 a->sin_local.sin_family = AF_INET;
1082 a->sin_local.sin_port = inet->inet_sport;
1083 a->sin_local.sin_addr.s_addr = inet->inet_rcv_saddr;
1085 if (!a->sin_local.sin_addr.s_addr)
1086 a->sin_local.sin_addr.s_addr = inet->inet_saddr;
1088 a->sin_remote.sin_family = AF_INET;
1089 a->sin_remote.sin_port = inet->inet_dport;
1090 a->sin_remote.sin_addr.s_addr = inet->inet_daddr;
1091 #if IS_ENABLED(CONFIG_IPV6)
1092 } else if (sk->sk_family == AF_INET6) {
1093 const struct ipv6_pinfo *np = inet6_sk(sk);
1095 if (WARN_ON_ONCE(!np))
1098 a->sin6_local.sin6_family = AF_INET6;
1099 a->sin6_local.sin6_port = inet->inet_sport;
1101 if (ipv6_addr_any(&sk->sk_v6_rcv_saddr))
1102 a->sin6_local.sin6_addr = np->saddr;
1104 a->sin6_local.sin6_addr = sk->sk_v6_rcv_saddr;
1106 a->sin6_remote.sin6_family = AF_INET6;
1107 a->sin6_remote.sin6_port = inet->inet_dport;
1108 a->sin6_remote.sin6_addr = sk->sk_v6_daddr;
1113 static int mptcp_getsockopt_subflow_addrs(struct mptcp_sock *msk, char __user *optval,
1116 struct mptcp_subflow_context *subflow;
1117 struct sock *sk = (struct sock *)msk;
1118 unsigned int sfcount = 0, copied = 0;
1119 struct mptcp_subflow_data sfd;
1120 char __user *addrptr;
1123 len = mptcp_get_subflow_data(&sfd, optval, optlen);
1127 sfd.size_kernel = sizeof(struct mptcp_subflow_addrs);
1128 sfd.size_user = min_t(unsigned int, sfd.size_user,
1129 sizeof(struct mptcp_subflow_addrs));
1131 addrptr = optval + sfd.size_subflow_data;
1135 mptcp_for_each_subflow(msk, subflow) {
1136 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1140 if (len && len >= sfd.size_user) {
1141 struct mptcp_subflow_addrs a;
1143 mptcp_get_sub_addrs(ssk, &a);
1145 if (copy_to_user(addrptr, &a, sfd.size_user)) {
1150 addrptr += sfd.size_user;
1151 copied += sfd.size_user;
1152 len -= sfd.size_user;
1158 sfd.num_subflows = sfcount;
1160 if (mptcp_put_subflow_data(&sfd, optval, copied, optlen))
1166 static int mptcp_get_full_info(struct mptcp_full_info *mfi,
1167 char __user *optval,
1172 BUILD_BUG_ON(offsetof(struct mptcp_full_info, mptcp_info) !=
1173 MIN_FULL_INFO_OPTLEN_SIZE);
1175 if (get_user(len, optlen))
1178 if (len < MIN_FULL_INFO_OPTLEN_SIZE)
1181 memset(mfi, 0, sizeof(*mfi));
1182 if (copy_from_user(mfi, optval, MIN_FULL_INFO_OPTLEN_SIZE))
1185 if (mfi->size_tcpinfo_kernel ||
1186 mfi->size_sfinfo_kernel ||
1190 if (mfi->size_sfinfo_user > INT_MAX ||
1191 mfi->size_tcpinfo_user > INT_MAX)
1194 return len - MIN_FULL_INFO_OPTLEN_SIZE;
1197 static int mptcp_put_full_info(struct mptcp_full_info *mfi,
1198 char __user *optval,
1202 copylen += MIN_FULL_INFO_OPTLEN_SIZE;
1203 if (put_user(copylen, optlen))
1206 if (copy_to_user(optval, mfi, copylen))
1211 static int mptcp_getsockopt_full_info(struct mptcp_sock *msk, char __user *optval,
1214 unsigned int sfcount = 0, copylen = 0;
1215 struct mptcp_subflow_context *subflow;
1216 struct sock *sk = (struct sock *)msk;
1217 void __user *tcpinfoptr, *sfinfoptr;
1218 struct mptcp_full_info mfi;
1221 len = mptcp_get_full_info(&mfi, optval, optlen);
1225 /* don't bother filling the mptcp info if there is not enough
1226 * user-space-provided storage
1229 mptcp_diag_fill_info(msk, &mfi.mptcp_info);
1230 copylen += min_t(unsigned int, len, sizeof(struct mptcp_info));
1233 mfi.size_tcpinfo_kernel = sizeof(struct tcp_info);
1234 mfi.size_tcpinfo_user = min_t(unsigned int, mfi.size_tcpinfo_user,
1235 sizeof(struct tcp_info));
1236 sfinfoptr = u64_to_user_ptr(mfi.subflow_info);
1237 mfi.size_sfinfo_kernel = sizeof(struct mptcp_subflow_info);
1238 mfi.size_sfinfo_user = min_t(unsigned int, mfi.size_sfinfo_user,
1239 sizeof(struct mptcp_subflow_info));
1240 tcpinfoptr = u64_to_user_ptr(mfi.tcp_info);
1243 mptcp_for_each_subflow(msk, subflow) {
1244 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1245 struct mptcp_subflow_info sfinfo;
1246 struct tcp_info tcp_info;
1248 if (sfcount++ >= mfi.size_arrays_user)
1251 /* fetch addr/tcp_info only if the user space buffers
1254 memset(&sfinfo, 0, sizeof(sfinfo));
1255 sfinfo.id = subflow->subflow_id;
1256 if (mfi.size_sfinfo_user >
1257 offsetof(struct mptcp_subflow_info, addrs))
1258 mptcp_get_sub_addrs(ssk, &sfinfo.addrs);
1259 if (copy_to_user(sfinfoptr, &sfinfo, mfi.size_sfinfo_user))
1262 if (mfi.size_tcpinfo_user) {
1263 tcp_get_info(ssk, &tcp_info);
1264 if (copy_to_user(tcpinfoptr, &tcp_info,
1265 mfi.size_tcpinfo_user))
1269 tcpinfoptr += mfi.size_tcpinfo_user;
1270 sfinfoptr += mfi.size_sfinfo_user;
1274 mfi.num_subflows = sfcount;
1275 if (mptcp_put_full_info(&mfi, optval, copylen, optlen))
1285 static int mptcp_put_int_option(struct mptcp_sock *msk, char __user *optval,
1286 int __user *optlen, int val)
1290 if (get_user(len, optlen))
1295 if (len < sizeof(int) && len > 0 && val >= 0 && val <= 255) {
1296 unsigned char ucval = (unsigned char)val;
1299 if (put_user(len, optlen))
1301 if (copy_to_user(optval, &ucval, 1))
1304 len = min_t(unsigned int, len, sizeof(int));
1305 if (put_user(len, optlen))
1307 if (copy_to_user(optval, &val, len))
1314 static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
1315 char __user *optval, int __user *optlen)
1319 case TCP_CONGESTION:
1322 case TCP_DEFER_ACCEPT:
1324 case TCP_FASTOPEN_CONNECT:
1325 case TCP_FASTOPEN_KEY:
1326 case TCP_FASTOPEN_NO_COOKIE:
1327 return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname,
1330 return mptcp_put_int_option(msk, optval, optlen, msk->recvmsg_inq);
1332 return mptcp_put_int_option(msk, optval, optlen, msk->cork);
1334 return mptcp_put_int_option(msk, optval, optlen, msk->nodelay);
1339 static int mptcp_getsockopt_v4(struct mptcp_sock *msk, int optname,
1340 char __user *optval, int __user *optlen)
1342 struct sock *sk = (void *)msk;
1346 return mptcp_put_int_option(msk, optval, optlen, inet_sk(sk)->tos);
1352 static int mptcp_getsockopt_sol_mptcp(struct mptcp_sock *msk, int optname,
1353 char __user *optval, int __user *optlen)
1357 return mptcp_getsockopt_info(msk, optval, optlen);
1358 case MPTCP_FULL_INFO:
1359 return mptcp_getsockopt_full_info(msk, optval, optlen);
1361 return mptcp_getsockopt_tcpinfo(msk, optval, optlen);
1362 case MPTCP_SUBFLOW_ADDRS:
1363 return mptcp_getsockopt_subflow_addrs(msk, optval, optlen);
1369 int mptcp_getsockopt(struct sock *sk, int level, int optname,
1370 char __user *optval, int __user *option)
1372 struct mptcp_sock *msk = mptcp_sk(sk);
1375 pr_debug("msk=%p", msk);
1377 /* @@ the meaning of setsockopt() when the socket is connected and
1378 * there are multiple subflows is not yet defined. It is up to the
1379 * MPTCP-level socket to configure the subflows until the subflow
1380 * is in TCP fallback, when socket options are passed through
1381 * to the one remaining subflow.
1384 ssk = __mptcp_tcp_fallback(msk);
1387 return tcp_getsockopt(ssk, level, optname, optval, option);
1389 if (level == SOL_IP)
1390 return mptcp_getsockopt_v4(msk, optname, optval, option);
1391 if (level == SOL_TCP)
1392 return mptcp_getsockopt_sol_tcp(msk, optname, optval, option);
1393 if (level == SOL_MPTCP)
1394 return mptcp_getsockopt_sol_mptcp(msk, optname, optval, option);
1398 static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
1400 static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK;
1401 struct sock *sk = (struct sock *)msk;
1403 if (ssk->sk_prot->keepalive) {
1404 if (sock_flag(sk, SOCK_KEEPOPEN))
1405 ssk->sk_prot->keepalive(ssk, 1);
1407 ssk->sk_prot->keepalive(ssk, 0);
1410 ssk->sk_priority = sk->sk_priority;
1411 ssk->sk_bound_dev_if = sk->sk_bound_dev_if;
1412 ssk->sk_incoming_cpu = sk->sk_incoming_cpu;
1413 ssk->sk_ipv6only = sk->sk_ipv6only;
1414 __ip_sock_set_tos(ssk, inet_sk(sk)->tos);
1416 if (sk->sk_userlocks & tx_rx_locks) {
1417 ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks;
1418 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
1419 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf);
1420 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
1421 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
1424 if (sock_flag(sk, SOCK_LINGER)) {
1425 ssk->sk_lingertime = sk->sk_lingertime;
1426 sock_set_flag(ssk, SOCK_LINGER);
1428 sock_reset_flag(ssk, SOCK_LINGER);
1431 if (sk->sk_mark != ssk->sk_mark) {
1432 ssk->sk_mark = sk->sk_mark;
1436 sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG));
1438 if (inet_csk(sk)->icsk_ca_ops != inet_csk(ssk)->icsk_ca_ops)
1439 tcp_set_congestion_control(ssk, msk->ca_name, false, true);
1440 __tcp_sock_set_cork(ssk, !!msk->cork);
1441 __tcp_sock_set_nodelay(ssk, !!msk->nodelay);
1443 inet_assign_bit(TRANSPARENT, ssk, inet_test_bit(TRANSPARENT, sk));
1444 inet_assign_bit(FREEBIND, ssk, inet_test_bit(FREEBIND, sk));
1447 static void __mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
1449 bool slow = lock_sock_fast(ssk);
1451 sync_socket_options(msk, ssk);
1453 unlock_sock_fast(ssk, slow);
1456 void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk)
1458 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1460 msk_owned_by_me(msk);
1462 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
1463 __mptcp_sockopt_sync(msk, ssk);
1465 subflow->setsockopt_seq = msk->setsockopt_seq;
1469 void mptcp_sockopt_sync_locked(struct mptcp_sock *msk, struct sock *ssk)
1471 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1473 msk_owned_by_me(msk);
1475 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) {
1476 sync_socket_options(msk, ssk);
1478 subflow->setsockopt_seq = msk->setsockopt_seq;