1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Linux INET6 implementation
7 * Pedro Roque <roque@di.fc.ul.pt>
10 * linux/net/ipv4/tcp.c
11 * linux/net/ipv4/tcp_input.c
12 * linux/net/ipv4/tcp_output.c
15 * Hideaki YOSHIFUJI : sin6_scope_id support
16 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
17 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
18 * a single port at the same time.
19 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
22 #include <linux/bottom_half.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/jiffies.h>
31 #include <linux/in6.h>
32 #include <linux/netdevice.h>
33 #include <linux/init.h>
34 #include <linux/jhash.h>
35 #include <linux/ipsec.h>
36 #include <linux/times.h>
37 #include <linux/slab.h>
38 #include <linux/uaccess.h>
39 #include <linux/ipv6.h>
40 #include <linux/icmpv6.h>
41 #include <linux/random.h>
42 #include <linux/indirect_call_wrapper.h>
45 #include <net/ndisc.h>
46 #include <net/inet6_hashtables.h>
47 #include <net/inet6_connection_sock.h>
49 #include <net/transp_v6.h>
50 #include <net/addrconf.h>
51 #include <net/ip6_route.h>
52 #include <net/ip6_checksum.h>
53 #include <net/inet_ecn.h>
54 #include <net/protocol.h>
57 #include <net/dsfield.h>
58 #include <net/timewait_sock.h>
59 #include <net/inet_common.h>
60 #include <net/secure_seq.h>
61 #include <net/busy_poll.h>
63 #include <linux/proc_fs.h>
64 #include <linux/seq_file.h>
66 #include <crypto/hash.h>
67 #include <linux/scatterlist.h>
69 #include <trace/events/tcp.h>
71 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
72 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
73 struct request_sock *req);
75 INDIRECT_CALLABLE_SCOPE int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
77 static const struct inet_connection_sock_af_ops ipv6_mapped;
78 const struct inet_connection_sock_af_ops ipv6_specific;
79 #ifdef CONFIG_TCP_MD5SIG
80 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
81 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
83 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
84 const struct in6_addr *addr,
91 /* Helper returning the inet6 address from a given tcp socket.
92 * It can be used in TCP stack instead of inet6_sk(sk).
93 * This avoids a dereference and allow compiler optimizations.
94 * It is a specialized version of inet6_sk_generic().
96 static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk)
98 unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo);
100 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
103 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
105 struct dst_entry *dst = skb_dst(skb);
107 if (dst && dst_hold_safe(dst)) {
108 const struct rt6_info *rt = (const struct rt6_info *)dst;
110 rcu_assign_pointer(sk->sk_rx_dst, dst);
111 sk->sk_rx_dst_ifindex = skb->skb_iif;
112 sk->sk_rx_dst_cookie = rt6_get_cookie(rt);
116 static u32 tcp_v6_init_seq(const struct sk_buff *skb)
118 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
119 ipv6_hdr(skb)->saddr.s6_addr32,
121 tcp_hdr(skb)->source);
124 static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
126 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
127 ipv6_hdr(skb)->saddr.s6_addr32);
130 static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
133 /* This check is replicated from tcp_v6_connect() and intended to
134 * prevent BPF program called below from accessing bytes that are out
135 * of the bound specified by user in addr_len.
137 if (addr_len < SIN6_LEN_RFC2133)
140 sock_owned_by_me(sk);
142 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
145 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
148 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
149 struct inet_sock *inet = inet_sk(sk);
150 struct inet_connection_sock *icsk = inet_csk(sk);
151 struct inet_timewait_death_row *tcp_death_row;
152 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
153 struct tcp_sock *tp = tcp_sk(sk);
154 struct in6_addr *saddr = NULL, *final_p, final;
155 struct ipv6_txoptions *opt;
157 struct dst_entry *dst;
161 if (addr_len < SIN6_LEN_RFC2133)
164 if (usin->sin6_family != AF_INET6)
165 return -EAFNOSUPPORT;
167 memset(&fl6, 0, sizeof(fl6));
170 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
171 IP6_ECN_flow_init(fl6.flowlabel);
172 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
173 struct ip6_flowlabel *flowlabel;
174 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
175 if (IS_ERR(flowlabel))
177 fl6_sock_release(flowlabel);
182 * connect() to INADDR_ANY means loopback (BSD'ism).
185 if (ipv6_addr_any(&usin->sin6_addr)) {
186 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
187 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
190 usin->sin6_addr = in6addr_loopback;
193 addr_type = ipv6_addr_type(&usin->sin6_addr);
195 if (addr_type & IPV6_ADDR_MULTICAST)
198 if (addr_type&IPV6_ADDR_LINKLOCAL) {
199 if (addr_len >= sizeof(struct sockaddr_in6) &&
200 usin->sin6_scope_id) {
201 /* If interface is set while binding, indices
204 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
207 sk->sk_bound_dev_if = usin->sin6_scope_id;
210 /* Connect to link-local address requires an interface */
211 if (!sk->sk_bound_dev_if)
215 if (tp->rx_opt.ts_recent_stamp &&
216 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
217 tp->rx_opt.ts_recent = 0;
218 tp->rx_opt.ts_recent_stamp = 0;
219 WRITE_ONCE(tp->write_seq, 0);
222 sk->sk_v6_daddr = usin->sin6_addr;
223 np->flow_label = fl6.flowlabel;
229 if (addr_type & IPV6_ADDR_MAPPED) {
230 u32 exthdrlen = icsk->icsk_ext_hdr_len;
231 struct sockaddr_in sin;
233 if (__ipv6_only_sock(sk))
236 sin.sin_family = AF_INET;
237 sin.sin_port = usin->sin6_port;
238 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
240 icsk->icsk_af_ops = &ipv6_mapped;
242 mptcpv6_handle_mapped(sk, true);
243 sk->sk_backlog_rcv = tcp_v4_do_rcv;
244 #ifdef CONFIG_TCP_MD5SIG
245 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
248 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
251 icsk->icsk_ext_hdr_len = exthdrlen;
252 icsk->icsk_af_ops = &ipv6_specific;
254 mptcpv6_handle_mapped(sk, false);
255 sk->sk_backlog_rcv = tcp_v6_do_rcv;
256 #ifdef CONFIG_TCP_MD5SIG
257 tp->af_specific = &tcp_sock_ipv6_specific;
261 np->saddr = sk->sk_v6_rcv_saddr;
266 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
267 saddr = &sk->sk_v6_rcv_saddr;
269 fl6.flowi6_proto = IPPROTO_TCP;
270 fl6.daddr = sk->sk_v6_daddr;
271 fl6.saddr = saddr ? *saddr : np->saddr;
272 fl6.flowi6_oif = sk->sk_bound_dev_if;
273 fl6.flowi6_mark = sk->sk_mark;
274 fl6.fl6_dport = usin->sin6_port;
275 fl6.fl6_sport = inet->inet_sport;
276 fl6.flowi6_uid = sk->sk_uid;
278 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
279 final_p = fl6_update_dst(&fl6, opt, &final);
281 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
283 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
291 sk->sk_v6_rcv_saddr = *saddr;
294 /* set the source address */
296 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
298 sk->sk_gso_type = SKB_GSO_TCPV6;
299 ip6_dst_store(sk, dst, NULL, NULL);
301 icsk->icsk_ext_hdr_len = 0;
303 icsk->icsk_ext_hdr_len = opt->opt_flen +
306 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
308 inet->inet_dport = usin->sin6_port;
310 tcp_set_state(sk, TCP_SYN_SENT);
311 tcp_death_row = sock_net(sk)->ipv4.tcp_death_row;
312 err = inet6_hash_connect(tcp_death_row, sk);
318 if (likely(!tp->repair)) {
320 WRITE_ONCE(tp->write_seq,
321 secure_tcpv6_seq(np->saddr.s6_addr32,
322 sk->sk_v6_daddr.s6_addr32,
325 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
327 sk->sk_v6_daddr.s6_addr32);
330 if (tcp_fastopen_defer_connect(sk, &err))
335 err = tcp_connect(sk);
342 tcp_set_state(sk, TCP_CLOSE);
344 inet->inet_dport = 0;
345 sk->sk_route_caps = 0;
349 static void tcp_v6_mtu_reduced(struct sock *sk)
351 struct dst_entry *dst;
354 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
357 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
359 /* Drop requests trying to increase our current mss.
360 * Check done in __ip6_rt_update_pmtu() is too late.
362 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
365 dst = inet6_csk_update_pmtu(sk, mtu);
369 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
370 tcp_sync_mss(sk, dst_mtu(dst));
371 tcp_simple_retransmit(sk);
375 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
376 u8 type, u8 code, int offset, __be32 info)
378 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
379 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
380 struct net *net = dev_net(skb->dev);
381 struct request_sock *fastopen;
382 struct ipv6_pinfo *np;
389 sk = __inet6_lookup_established(net, &tcp_hashinfo,
390 &hdr->daddr, th->dest,
391 &hdr->saddr, ntohs(th->source),
392 skb->dev->ifindex, inet6_sdif(skb));
395 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
400 if (sk->sk_state == TCP_TIME_WAIT) {
401 inet_twsk_put(inet_twsk(sk));
404 seq = ntohl(th->seq);
405 fatal = icmpv6_err_convert(type, code, &err);
406 if (sk->sk_state == TCP_NEW_SYN_RECV) {
407 tcp_req_err(sk, seq, fatal);
412 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
413 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
415 if (sk->sk_state == TCP_CLOSE)
418 if (static_branch_unlikely(&ip6_min_hopcount)) {
419 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
420 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
421 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
427 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
428 fastopen = rcu_dereference(tp->fastopen_rsk);
429 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
430 if (sk->sk_state != TCP_LISTEN &&
431 !between(seq, snd_una, tp->snd_nxt)) {
432 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
436 np = tcp_inet6_sk(sk);
438 if (type == NDISC_REDIRECT) {
439 if (!sock_owned_by_user(sk)) {
440 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
443 dst->ops->redirect(dst, sk, skb);
448 if (type == ICMPV6_PKT_TOOBIG) {
449 u32 mtu = ntohl(info);
451 /* We are not interested in TCP_LISTEN and open_requests
452 * (SYN-ACKs send out by Linux are always <576bytes so
453 * they should go through unfragmented).
455 if (sk->sk_state == TCP_LISTEN)
458 if (!ip6_sk_accept_pmtu(sk))
461 if (mtu < IPV6_MIN_MTU)
464 WRITE_ONCE(tp->mtu_info, mtu);
466 if (!sock_owned_by_user(sk))
467 tcp_v6_mtu_reduced(sk);
468 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
475 /* Might be for an request_sock */
476 switch (sk->sk_state) {
479 /* Only in fast or simultaneous open. If a fast open socket is
480 * already accepted it is treated as a connected one below.
482 if (fastopen && !fastopen->sk)
485 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
487 if (!sock_owned_by_user(sk)) {
489 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
493 sk->sk_err_soft = err;
498 /* check if this ICMP message allows revert of backoff.
501 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
502 code == ICMPV6_NOROUTE)
503 tcp_ld_RTO_revert(sk, seq);
506 if (!sock_owned_by_user(sk) && np->recverr) {
510 sk->sk_err_soft = err;
519 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
521 struct request_sock *req,
522 struct tcp_fastopen_cookie *foc,
523 enum tcp_synack_type synack_type,
524 struct sk_buff *syn_skb)
526 struct inet_request_sock *ireq = inet_rsk(req);
527 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
528 struct ipv6_txoptions *opt;
529 struct flowi6 *fl6 = &fl->u.ip6;
534 /* First, grab a route. */
535 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
536 IPPROTO_TCP)) == NULL)
539 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
542 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
543 &ireq->ir_v6_rmt_addr);
545 fl6->daddr = ireq->ir_v6_rmt_addr;
546 if (np->repflow && ireq->pktopts)
547 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
549 tclass = sock_net(sk)->ipv4.sysctl_tcp_reflect_tos ?
550 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
551 (np->tclass & INET_ECN_MASK) :
554 if (!INET_ECN_is_capable(tclass) &&
555 tcp_bpf_ca_needs_ecn((struct sock *)req))
556 tclass |= INET_ECN_ECT_0;
559 opt = ireq->ipv6_opt;
561 opt = rcu_dereference(np->opt);
562 err = ip6_xmit(sk, skb, fl6, skb->mark ? : sk->sk_mark, opt,
563 tclass, sk->sk_priority);
565 err = net_xmit_eval(err);
573 static void tcp_v6_reqsk_destructor(struct request_sock *req)
575 kfree(inet_rsk(req)->ipv6_opt);
576 consume_skb(inet_rsk(req)->pktopts);
579 #ifdef CONFIG_TCP_MD5SIG
580 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
581 const struct in6_addr *addr,
584 return tcp_md5_do_lookup(sk, l3index,
585 (union tcp_md5_addr *)addr, AF_INET6);
588 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
589 const struct sock *addr_sk)
593 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
594 addr_sk->sk_bound_dev_if);
595 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
599 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
600 sockptr_t optval, int optlen)
602 struct tcp_md5sig cmd;
603 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
608 if (optlen < sizeof(cmd))
611 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
614 if (sin6->sin6_family != AF_INET6)
617 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
619 if (optname == TCP_MD5SIG_EXT &&
620 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
621 prefixlen = cmd.tcpm_prefixlen;
622 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
626 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
629 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
630 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
631 struct net_device *dev;
634 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
635 if (dev && netif_is_l3_master(dev))
636 l3index = dev->ifindex;
639 /* ok to reference set/not set outside of rcu;
640 * right now device MUST be an L3 master
642 if (!dev || !l3index)
646 if (!cmd.tcpm_keylen) {
647 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
648 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
651 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
652 AF_INET6, prefixlen, l3index, flags);
655 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
658 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
659 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
660 AF_INET, prefixlen, l3index, flags,
661 cmd.tcpm_key, cmd.tcpm_keylen,
664 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
665 AF_INET6, prefixlen, l3index, flags,
666 cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
669 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
670 const struct in6_addr *daddr,
671 const struct in6_addr *saddr,
672 const struct tcphdr *th, int nbytes)
674 struct tcp6_pseudohdr *bp;
675 struct scatterlist sg;
679 /* 1. TCP pseudo-header (RFC2460) */
682 bp->protocol = cpu_to_be32(IPPROTO_TCP);
683 bp->len = cpu_to_be32(nbytes);
685 _th = (struct tcphdr *)(bp + 1);
686 memcpy(_th, th, sizeof(*th));
689 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
690 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
691 sizeof(*bp) + sizeof(*th));
692 return crypto_ahash_update(hp->md5_req);
695 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
696 const struct in6_addr *daddr, struct in6_addr *saddr,
697 const struct tcphdr *th)
699 struct tcp_md5sig_pool *hp;
700 struct ahash_request *req;
702 hp = tcp_get_md5sig_pool();
704 goto clear_hash_noput;
707 if (crypto_ahash_init(req))
709 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
711 if (tcp_md5_hash_key(hp, key))
713 ahash_request_set_crypt(req, NULL, md5_hash, 0);
714 if (crypto_ahash_final(req))
717 tcp_put_md5sig_pool();
721 tcp_put_md5sig_pool();
723 memset(md5_hash, 0, 16);
727 static int tcp_v6_md5_hash_skb(char *md5_hash,
728 const struct tcp_md5sig_key *key,
729 const struct sock *sk,
730 const struct sk_buff *skb)
732 const struct in6_addr *saddr, *daddr;
733 struct tcp_md5sig_pool *hp;
734 struct ahash_request *req;
735 const struct tcphdr *th = tcp_hdr(skb);
737 if (sk) { /* valid for establish/request sockets */
738 saddr = &sk->sk_v6_rcv_saddr;
739 daddr = &sk->sk_v6_daddr;
741 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
742 saddr = &ip6h->saddr;
743 daddr = &ip6h->daddr;
746 hp = tcp_get_md5sig_pool();
748 goto clear_hash_noput;
751 if (crypto_ahash_init(req))
754 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
756 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
758 if (tcp_md5_hash_key(hp, key))
760 ahash_request_set_crypt(req, NULL, md5_hash, 0);
761 if (crypto_ahash_final(req))
764 tcp_put_md5sig_pool();
768 tcp_put_md5sig_pool();
770 memset(md5_hash, 0, 16);
776 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
777 const struct sk_buff *skb,
780 #ifdef CONFIG_TCP_MD5SIG
781 const __u8 *hash_location = NULL;
782 struct tcp_md5sig_key *hash_expected;
783 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
784 const struct tcphdr *th = tcp_hdr(skb);
785 int genhash, l3index;
788 /* sdif set, means packet ingressed via a device
789 * in an L3 domain and dif is set to the l3mdev
791 l3index = sdif ? dif : 0;
793 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr, l3index);
794 hash_location = tcp_parse_md5sig_option(th);
796 /* We've parsed the options - do we have a hash? */
797 if (!hash_expected && !hash_location)
800 if (hash_expected && !hash_location) {
801 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
805 if (!hash_expected && hash_location) {
806 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
810 /* check the signature */
811 genhash = tcp_v6_md5_hash_skb(newhash,
815 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
816 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
817 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u L3 index %d\n",
818 genhash ? "failed" : "mismatch",
819 &ip6h->saddr, ntohs(th->source),
820 &ip6h->daddr, ntohs(th->dest), l3index);
827 static void tcp_v6_init_req(struct request_sock *req,
828 const struct sock *sk_listener,
831 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
832 struct inet_request_sock *ireq = inet_rsk(req);
833 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
835 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
836 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
838 /* So that link locals have meaning */
839 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
840 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
841 ireq->ir_iif = tcp_v6_iif(skb);
843 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
844 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
845 np->rxopt.bits.rxinfo ||
846 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
847 np->rxopt.bits.rxohlim || np->repflow)) {
848 refcount_inc(&skb->users);
853 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
856 struct request_sock *req)
858 tcp_v6_init_req(req, sk, skb);
860 if (security_inet_conn_request(sk, skb, req))
863 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
866 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
868 .obj_size = sizeof(struct tcp6_request_sock),
869 .rtx_syn_ack = tcp_rtx_synack,
870 .send_ack = tcp_v6_reqsk_send_ack,
871 .destructor = tcp_v6_reqsk_destructor,
872 .send_reset = tcp_v6_send_reset,
873 .syn_ack_timeout = tcp_syn_ack_timeout,
876 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
877 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
878 sizeof(struct ipv6hdr),
879 #ifdef CONFIG_TCP_MD5SIG
880 .req_md5_lookup = tcp_v6_md5_lookup,
881 .calc_md5_hash = tcp_v6_md5_hash_skb,
883 #ifdef CONFIG_SYN_COOKIES
884 .cookie_init_seq = cookie_v6_init_sequence,
886 .route_req = tcp_v6_route_req,
887 .init_seq = tcp_v6_init_seq,
888 .init_ts_off = tcp_v6_init_ts_off,
889 .send_synack = tcp_v6_send_synack,
892 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
893 u32 ack, u32 win, u32 tsval, u32 tsecr,
894 int oif, struct tcp_md5sig_key *key, int rst,
895 u8 tclass, __be32 label, u32 priority)
897 const struct tcphdr *th = tcp_hdr(skb);
899 struct sk_buff *buff;
901 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
902 struct sock *ctl_sk = net->ipv6.tcp_sk;
903 unsigned int tot_len = sizeof(struct tcphdr);
904 __be32 mrst = 0, *topt;
905 struct dst_entry *dst;
909 tot_len += TCPOLEN_TSTAMP_ALIGNED;
910 #ifdef CONFIG_TCP_MD5SIG
912 tot_len += TCPOLEN_MD5SIG_ALIGNED;
917 mrst = mptcp_reset_option(skb);
920 tot_len += sizeof(__be32);
924 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
929 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
931 t1 = skb_push(buff, tot_len);
932 skb_reset_transport_header(buff);
934 /* Swap the send and the receive. */
935 memset(t1, 0, sizeof(*t1));
936 t1->dest = th->source;
937 t1->source = th->dest;
938 t1->doff = tot_len / 4;
939 t1->seq = htonl(seq);
940 t1->ack_seq = htonl(ack);
941 t1->ack = !rst || !th->ack;
943 t1->window = htons(win);
945 topt = (__be32 *)(t1 + 1);
948 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
949 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
950 *topt++ = htonl(tsval);
951 *topt++ = htonl(tsecr);
957 #ifdef CONFIG_TCP_MD5SIG
959 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
960 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
961 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
962 &ipv6_hdr(skb)->saddr,
963 &ipv6_hdr(skb)->daddr, t1);
967 memset(&fl6, 0, sizeof(fl6));
968 fl6.daddr = ipv6_hdr(skb)->saddr;
969 fl6.saddr = ipv6_hdr(skb)->daddr;
970 fl6.flowlabel = label;
972 buff->ip_summed = CHECKSUM_PARTIAL;
974 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
976 fl6.flowi6_proto = IPPROTO_TCP;
977 if (rt6_need_strict(&fl6.daddr) && !oif)
978 fl6.flowi6_oif = tcp_v6_iif(skb);
980 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
983 fl6.flowi6_oif = oif;
987 if (sk->sk_state == TCP_TIME_WAIT) {
988 mark = inet_twsk(sk)->tw_mark;
989 /* autoflowlabel relies on buff->hash */
990 skb_set_hash(buff, inet_twsk(sk)->tw_txhash,
995 buff->tstamp = tcp_transmit_time(sk);
997 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
998 fl6.fl6_dport = t1->dest;
999 fl6.fl6_sport = t1->source;
1000 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
1001 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
1003 /* Pass a socket to ip6_dst_lookup either it is for RST
1004 * Underlying function will use this to retrieve the network
1007 dst = ip6_dst_lookup_flow(sock_net(ctl_sk), ctl_sk, &fl6, NULL);
1009 skb_dst_set(buff, dst);
1010 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
1011 tclass & ~INET_ECN_MASK, priority);
1012 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1014 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
1021 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
1023 const struct tcphdr *th = tcp_hdr(skb);
1024 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
1025 u32 seq = 0, ack_seq = 0;
1026 struct tcp_md5sig_key *key = NULL;
1027 #ifdef CONFIG_TCP_MD5SIG
1028 const __u8 *hash_location = NULL;
1029 unsigned char newhash[16];
1031 struct sock *sk1 = NULL;
1041 /* If sk not NULL, it means we did a successful lookup and incoming
1042 * route had to be correct. prequeue might have dropped our dst.
1044 if (!sk && !ipv6_unicast_destination(skb))
1047 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1048 #ifdef CONFIG_TCP_MD5SIG
1050 hash_location = tcp_parse_md5sig_option(th);
1051 if (sk && sk_fullsock(sk)) {
1054 /* sdif set, means packet ingressed via a device
1055 * in an L3 domain and inet_iif is set to it.
1057 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1058 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1059 } else if (hash_location) {
1060 int dif = tcp_v6_iif_l3_slave(skb);
1061 int sdif = tcp_v6_sdif(skb);
1065 * active side is lost. Try to find listening socket through
1066 * source port, and then find md5 key through listening socket.
1067 * we are not loose security here:
1068 * Incoming packet is checked with md5 hash with finding key,
1069 * no RST generated if md5 hash doesn't match.
1071 sk1 = inet6_lookup_listener(net,
1072 &tcp_hashinfo, NULL, 0,
1074 th->source, &ipv6h->daddr,
1075 ntohs(th->source), dif, sdif);
1079 /* sdif set, means packet ingressed via a device
1080 * in an L3 domain and dif is set to it.
1082 l3index = tcp_v6_sdif(skb) ? dif : 0;
1084 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1088 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1089 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1095 seq = ntohl(th->ack_seq);
1097 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1101 oif = sk->sk_bound_dev_if;
1102 if (sk_fullsock(sk)) {
1103 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1105 trace_tcp_send_reset(sk, skb);
1107 label = ip6_flowlabel(ipv6h);
1108 priority = sk->sk_priority;
1110 if (sk->sk_state == TCP_TIME_WAIT) {
1111 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1112 priority = inet_twsk(sk)->tw_priority;
1115 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1116 label = ip6_flowlabel(ipv6h);
1119 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1,
1120 ipv6_get_dsfield(ipv6h), label, priority);
1122 #ifdef CONFIG_TCP_MD5SIG
1128 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1129 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1130 struct tcp_md5sig_key *key, u8 tclass,
1131 __be32 label, u32 priority)
1133 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1134 tclass, label, priority);
1137 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1139 struct inet_timewait_sock *tw = inet_twsk(sk);
1140 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1142 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1143 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1144 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1145 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1146 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority);
1151 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1152 struct request_sock *req)
1156 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1158 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1159 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1162 * The window field (SEG.WND) of every outgoing segment, with the
1163 * exception of <SYN> segments, MUST be right-shifted by
1164 * Rcv.Wind.Shift bits:
1166 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1167 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1168 tcp_rsk(req)->rcv_nxt,
1169 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1170 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1171 req->ts_recent, sk->sk_bound_dev_if,
1172 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1173 ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority);
1177 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1179 #ifdef CONFIG_SYN_COOKIES
1180 const struct tcphdr *th = tcp_hdr(skb);
1183 sk = cookie_v6_check(sk, skb);
1188 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1189 struct tcphdr *th, u32 *cookie)
1192 #ifdef CONFIG_SYN_COOKIES
1193 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1194 &tcp_request_sock_ipv6_ops, sk, th);
1196 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1197 tcp_synq_overflow(sk);
1203 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1205 if (skb->protocol == htons(ETH_P_IP))
1206 return tcp_v4_conn_request(sk, skb);
1208 if (!ipv6_unicast_destination(skb))
1211 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1212 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1216 return tcp_conn_request(&tcp6_request_sock_ops,
1217 &tcp_request_sock_ipv6_ops, sk, skb);
1221 return 0; /* don't send reset */
1224 static void tcp_v6_restore_cb(struct sk_buff *skb)
1226 /* We need to move header back to the beginning if xfrm6_policy_check()
1227 * and tcp_v6_fill_cb() are going to be called again.
1228 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1230 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1231 sizeof(struct inet6_skb_parm));
1234 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1235 struct request_sock *req,
1236 struct dst_entry *dst,
1237 struct request_sock *req_unhash,
1240 struct inet_request_sock *ireq;
1241 struct ipv6_pinfo *newnp;
1242 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1243 struct ipv6_txoptions *opt;
1244 struct inet_sock *newinet;
1245 bool found_dup_sk = false;
1246 struct tcp_sock *newtp;
1248 #ifdef CONFIG_TCP_MD5SIG
1249 struct tcp_md5sig_key *key;
1254 if (skb->protocol == htons(ETH_P_IP)) {
1259 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1260 req_unhash, own_req);
1265 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1267 newnp = tcp_inet6_sk(newsk);
1268 newtp = tcp_sk(newsk);
1270 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1272 newnp->saddr = newsk->sk_v6_rcv_saddr;
1274 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1275 if (sk_is_mptcp(newsk))
1276 mptcpv6_handle_mapped(newsk, true);
1277 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1278 #ifdef CONFIG_TCP_MD5SIG
1279 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1282 newnp->ipv6_mc_list = NULL;
1283 newnp->ipv6_ac_list = NULL;
1284 newnp->ipv6_fl_list = NULL;
1285 newnp->pktoptions = NULL;
1287 newnp->mcast_oif = inet_iif(skb);
1288 newnp->mcast_hops = ip_hdr(skb)->ttl;
1289 newnp->rcv_flowinfo = 0;
1291 newnp->flow_label = 0;
1294 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1295 * here, tcp_create_openreq_child now does this for us, see the comment in
1296 * that function for the gory details. -acme
1299 /* It is tricky place. Until this moment IPv4 tcp
1300 worked with IPv6 icsk.icsk_af_ops.
1303 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1308 ireq = inet_rsk(req);
1310 if (sk_acceptq_is_full(sk))
1314 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1319 newsk = tcp_create_openreq_child(sk, req, skb);
1324 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1325 * count here, tcp_create_openreq_child now does this for us, see the
1326 * comment in that function for the gory details. -acme
1329 newsk->sk_gso_type = SKB_GSO_TCPV6;
1330 ip6_dst_store(newsk, dst, NULL, NULL);
1331 inet6_sk_rx_dst_set(newsk, skb);
1333 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1335 newtp = tcp_sk(newsk);
1336 newinet = inet_sk(newsk);
1337 newnp = tcp_inet6_sk(newsk);
1339 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1341 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1342 newnp->saddr = ireq->ir_v6_loc_addr;
1343 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1344 newsk->sk_bound_dev_if = ireq->ir_iif;
1346 /* Now IPv6 options...
1348 First: no IPv4 options.
1350 newinet->inet_opt = NULL;
1351 newnp->ipv6_mc_list = NULL;
1352 newnp->ipv6_ac_list = NULL;
1353 newnp->ipv6_fl_list = NULL;
1356 newnp->rxopt.all = np->rxopt.all;
1358 newnp->pktoptions = NULL;
1360 newnp->mcast_oif = tcp_v6_iif(skb);
1361 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1362 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1364 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1366 /* Set ToS of the new socket based upon the value of incoming SYN.
1367 * ECT bits are set later in tcp_init_transfer().
1369 if (sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)
1370 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1372 /* Clone native IPv6 options from listening socket (if any)
1374 Yes, keeping reference count would be much more clever,
1375 but we make one more one thing there: reattach optmem
1378 opt = ireq->ipv6_opt;
1380 opt = rcu_dereference(np->opt);
1382 opt = ipv6_dup_options(newsk, opt);
1383 RCU_INIT_POINTER(newnp->opt, opt);
1385 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1387 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1390 tcp_ca_openreq_child(newsk, dst);
1392 tcp_sync_mss(newsk, dst_mtu(dst));
1393 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1395 tcp_initialize_rcv_mss(newsk);
1397 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1398 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1400 #ifdef CONFIG_TCP_MD5SIG
1401 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1403 /* Copy over the MD5 key from the original socket */
1404 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1406 /* We're using one, so create a matching key
1407 * on the newsk structure. If we fail to get
1408 * memory, then we end up not copying the key
1411 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1412 AF_INET6, 128, l3index, key->flags, key->key, key->keylen,
1413 sk_gfp_mask(sk, GFP_ATOMIC));
1417 if (__inet_inherit_port(sk, newsk) < 0) {
1418 inet_csk_prepare_forced_close(newsk);
1422 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1425 tcp_move_syn(newtp, req);
1427 /* Clone pktoptions received with SYN, if we own the req */
1428 if (ireq->pktopts) {
1429 newnp->pktoptions = skb_clone(ireq->pktopts,
1430 sk_gfp_mask(sk, GFP_ATOMIC));
1431 consume_skb(ireq->pktopts);
1432 ireq->pktopts = NULL;
1433 if (newnp->pktoptions) {
1434 tcp_v6_restore_cb(newnp->pktoptions);
1435 skb_set_owner_r(newnp->pktoptions, newsk);
1439 if (!req_unhash && found_dup_sk) {
1440 /* This code path should only be executed in the
1441 * syncookie case only
1443 bh_unlock_sock(newsk);
1452 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1460 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1462 /* The socket must have it's spinlock held when we get
1463 * here, unless it is a TCP_LISTEN socket.
1465 * We have a potential double-lock case here, so even when
1466 * doing backlog processing we use the BH locking scheme.
1467 * This is because we cannot sleep with the original spinlock
1470 INDIRECT_CALLABLE_SCOPE
1471 int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1473 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1474 struct sk_buff *opt_skb = NULL;
1475 struct tcp_sock *tp;
1477 /* Imagine: socket is IPv6. IPv4 packet arrives,
1478 goes to IPv4 receive handler and backlogged.
1479 From backlog it always goes here. Kerboom...
1480 Fortunately, tcp_rcv_established and rcv_established
1481 handle them correctly, but it is not case with
1482 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1485 if (skb->protocol == htons(ETH_P_IP))
1486 return tcp_v4_do_rcv(sk, skb);
1489 * socket locking is here for SMP purposes as backlog rcv
1490 * is currently called with bh processing disabled.
1493 /* Do Stevens' IPV6_PKTOPTIONS.
1495 Yes, guys, it is the only place in our code, where we
1496 may make it not affecting IPv4.
1497 The rest of code is protocol independent,
1498 and I do not like idea to uglify IPv4.
1500 Actually, all the idea behind IPV6_PKTOPTIONS
1501 looks not very well thought. For now we latch
1502 options, received in the last packet, enqueued
1503 by tcp. Feel free to propose better solution.
1507 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1509 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1510 struct dst_entry *dst;
1512 dst = rcu_dereference_protected(sk->sk_rx_dst,
1513 lockdep_sock_is_held(sk));
1515 sock_rps_save_rxhash(sk, skb);
1516 sk_mark_napi_id(sk, skb);
1518 if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1519 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1520 dst, sk->sk_rx_dst_cookie) == NULL) {
1521 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1526 tcp_rcv_established(sk, skb);
1528 goto ipv6_pktoptions;
1532 if (tcp_checksum_complete(skb))
1535 if (sk->sk_state == TCP_LISTEN) {
1536 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1542 if (tcp_child_process(sk, nsk, skb))
1545 __kfree_skb(opt_skb);
1549 sock_rps_save_rxhash(sk, skb);
1551 if (tcp_rcv_state_process(sk, skb))
1554 goto ipv6_pktoptions;
1558 tcp_v6_send_reset(sk, skb);
1561 __kfree_skb(opt_skb);
1565 trace_tcp_bad_csum(skb);
1566 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1567 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1572 /* Do you ask, what is it?
1574 1. skb was enqueued by tcp.
1575 2. skb is added to tail of read queue, rather than out of order.
1576 3. socket is not in passive state.
1577 4. Finally, it really contains options, which user wants to receive.
1580 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1581 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1582 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1583 np->mcast_oif = tcp_v6_iif(opt_skb);
1584 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1585 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1586 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1587 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1589 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1590 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1591 skb_set_owner_r(opt_skb, sk);
1592 tcp_v6_restore_cb(opt_skb);
1593 opt_skb = xchg(&np->pktoptions, opt_skb);
1595 __kfree_skb(opt_skb);
1596 opt_skb = xchg(&np->pktoptions, NULL);
1600 consume_skb(opt_skb);
1604 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1605 const struct tcphdr *th)
1607 /* This is tricky: we move IP6CB at its correct location into
1608 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1609 * _decode_session6() uses IP6CB().
1610 * barrier() makes sure compiler won't play aliasing games.
1612 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1613 sizeof(struct inet6_skb_parm));
1616 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1617 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1618 skb->len - th->doff*4);
1619 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1620 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1621 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1622 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1623 TCP_SKB_CB(skb)->sacked = 0;
1624 TCP_SKB_CB(skb)->has_rxtstamp =
1625 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1628 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1630 int sdif = inet6_sdif(skb);
1631 int dif = inet6_iif(skb);
1632 const struct tcphdr *th;
1633 const struct ipv6hdr *hdr;
1637 struct net *net = dev_net(skb->dev);
1639 if (skb->pkt_type != PACKET_HOST)
1643 * Count it even if it's bad.
1645 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1647 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1650 th = (const struct tcphdr *)skb->data;
1652 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1654 if (!pskb_may_pull(skb, th->doff*4))
1657 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1660 th = (const struct tcphdr *)skb->data;
1661 hdr = ipv6_hdr(skb);
1664 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1665 th->source, th->dest, inet6_iif(skb), sdif,
1671 if (sk->sk_state == TCP_TIME_WAIT)
1674 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1675 struct request_sock *req = inet_reqsk(sk);
1676 bool req_stolen = false;
1679 sk = req->rsk_listener;
1680 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif)) {
1681 sk_drops_add(sk, skb);
1685 if (tcp_checksum_complete(skb)) {
1689 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1690 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1692 inet_csk_reqsk_queue_drop_and_put(sk, req);
1696 /* reuseport_migrate_sock() has already held one sk_refcnt
1704 if (!tcp_filter(sk, skb)) {
1705 th = (const struct tcphdr *)skb->data;
1706 hdr = ipv6_hdr(skb);
1707 tcp_v6_fill_cb(skb, hdr, th);
1708 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1713 /* Another cpu got exclusive access to req
1714 * and created a full blown socket.
1715 * Try to feed this packet to this socket
1716 * instead of discarding it.
1718 tcp_v6_restore_cb(skb);
1722 goto discard_and_relse;
1726 tcp_v6_restore_cb(skb);
1727 } else if (tcp_child_process(sk, nsk, skb)) {
1728 tcp_v6_send_reset(nsk, skb);
1729 goto discard_and_relse;
1736 if (static_branch_unlikely(&ip6_min_hopcount)) {
1737 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
1738 if (hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
1739 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1740 goto discard_and_relse;
1744 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1745 goto discard_and_relse;
1747 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif))
1748 goto discard_and_relse;
1750 if (tcp_filter(sk, skb))
1751 goto discard_and_relse;
1752 th = (const struct tcphdr *)skb->data;
1753 hdr = ipv6_hdr(skb);
1754 tcp_v6_fill_cb(skb, hdr, th);
1758 if (sk->sk_state == TCP_LISTEN) {
1759 ret = tcp_v6_do_rcv(sk, skb);
1760 goto put_and_return;
1763 sk_incoming_cpu_update(sk);
1765 sk_defer_free_flush(sk);
1766 bh_lock_sock_nested(sk);
1767 tcp_segs_in(tcp_sk(sk), skb);
1769 if (!sock_owned_by_user(sk)) {
1770 ret = tcp_v6_do_rcv(sk, skb);
1772 if (tcp_add_backlog(sk, skb))
1773 goto discard_and_relse;
1779 return ret ? -1 : 0;
1782 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1785 tcp_v6_fill_cb(skb, hdr, th);
1787 if (tcp_checksum_complete(skb)) {
1789 trace_tcp_bad_csum(skb);
1790 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1792 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1794 tcp_v6_send_reset(NULL, skb);
1802 sk_drops_add(sk, skb);
1808 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1809 inet_twsk_put(inet_twsk(sk));
1813 tcp_v6_fill_cb(skb, hdr, th);
1815 if (tcp_checksum_complete(skb)) {
1816 inet_twsk_put(inet_twsk(sk));
1820 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1825 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1826 skb, __tcp_hdrlen(th),
1827 &ipv6_hdr(skb)->saddr, th->source,
1828 &ipv6_hdr(skb)->daddr,
1830 tcp_v6_iif_l3_slave(skb),
1833 struct inet_timewait_sock *tw = inet_twsk(sk);
1834 inet_twsk_deschedule_put(tw);
1836 tcp_v6_restore_cb(skb);
1844 tcp_v6_timewait_ack(sk, skb);
1847 tcp_v6_send_reset(sk, skb);
1848 inet_twsk_deschedule_put(inet_twsk(sk));
1850 case TCP_TW_SUCCESS:
1856 INDIRECT_CALLABLE_SCOPE void tcp_v6_early_demux(struct sk_buff *skb)
1858 const struct ipv6hdr *hdr;
1859 const struct tcphdr *th;
1862 if (skb->pkt_type != PACKET_HOST)
1865 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1868 hdr = ipv6_hdr(skb);
1871 if (th->doff < sizeof(struct tcphdr) / 4)
1874 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1875 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1876 &hdr->saddr, th->source,
1877 &hdr->daddr, ntohs(th->dest),
1878 inet6_iif(skb), inet6_sdif(skb));
1881 skb->destructor = sock_edemux;
1882 if (sk_fullsock(sk)) {
1883 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1886 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1888 sk->sk_rx_dst_ifindex == skb->skb_iif)
1889 skb_dst_set_noref(skb, dst);
1894 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1895 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1896 .twsk_unique = tcp_twsk_unique,
1897 .twsk_destructor = tcp_twsk_destructor,
1900 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1902 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr);
1905 const struct inet_connection_sock_af_ops ipv6_specific = {
1906 .queue_xmit = inet6_csk_xmit,
1907 .send_check = tcp_v6_send_check,
1908 .rebuild_header = inet6_sk_rebuild_header,
1909 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1910 .conn_request = tcp_v6_conn_request,
1911 .syn_recv_sock = tcp_v6_syn_recv_sock,
1912 .net_header_len = sizeof(struct ipv6hdr),
1913 .net_frag_header_len = sizeof(struct frag_hdr),
1914 .setsockopt = ipv6_setsockopt,
1915 .getsockopt = ipv6_getsockopt,
1916 .addr2sockaddr = inet6_csk_addr2sockaddr,
1917 .sockaddr_len = sizeof(struct sockaddr_in6),
1918 .mtu_reduced = tcp_v6_mtu_reduced,
1921 #ifdef CONFIG_TCP_MD5SIG
1922 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1923 .md5_lookup = tcp_v6_md5_lookup,
1924 .calc_md5_hash = tcp_v6_md5_hash_skb,
1925 .md5_parse = tcp_v6_parse_md5_keys,
1930 * TCP over IPv4 via INET6 API
1932 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1933 .queue_xmit = ip_queue_xmit,
1934 .send_check = tcp_v4_send_check,
1935 .rebuild_header = inet_sk_rebuild_header,
1936 .sk_rx_dst_set = inet_sk_rx_dst_set,
1937 .conn_request = tcp_v6_conn_request,
1938 .syn_recv_sock = tcp_v6_syn_recv_sock,
1939 .net_header_len = sizeof(struct iphdr),
1940 .setsockopt = ipv6_setsockopt,
1941 .getsockopt = ipv6_getsockopt,
1942 .addr2sockaddr = inet6_csk_addr2sockaddr,
1943 .sockaddr_len = sizeof(struct sockaddr_in6),
1944 .mtu_reduced = tcp_v4_mtu_reduced,
1947 #ifdef CONFIG_TCP_MD5SIG
1948 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1949 .md5_lookup = tcp_v4_md5_lookup,
1950 .calc_md5_hash = tcp_v4_md5_hash_skb,
1951 .md5_parse = tcp_v6_parse_md5_keys,
1955 /* NOTE: A lot of things set to zero explicitly by call to
1956 * sk_alloc() so need not be done here.
1958 static int tcp_v6_init_sock(struct sock *sk)
1960 struct inet_connection_sock *icsk = inet_csk(sk);
1964 icsk->icsk_af_ops = &ipv6_specific;
1966 #ifdef CONFIG_TCP_MD5SIG
1967 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1973 static void tcp_v6_destroy_sock(struct sock *sk)
1975 tcp_v4_destroy_sock(sk);
1976 inet6_destroy_sock(sk);
1979 #ifdef CONFIG_PROC_FS
1980 /* Proc filesystem TCPv6 sock list dumping. */
1981 static void get_openreq6(struct seq_file *seq,
1982 const struct request_sock *req, int i)
1984 long ttd = req->rsk_timer.expires - jiffies;
1985 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1986 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1992 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1993 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1995 src->s6_addr32[0], src->s6_addr32[1],
1996 src->s6_addr32[2], src->s6_addr32[3],
1997 inet_rsk(req)->ir_num,
1998 dest->s6_addr32[0], dest->s6_addr32[1],
1999 dest->s6_addr32[2], dest->s6_addr32[3],
2000 ntohs(inet_rsk(req)->ir_rmt_port),
2002 0, 0, /* could print option size, but that is af dependent. */
2003 1, /* timers active (only the expire timer) */
2004 jiffies_to_clock_t(ttd),
2006 from_kuid_munged(seq_user_ns(seq),
2007 sock_i_uid(req->rsk_listener)),
2008 0, /* non standard timer */
2009 0, /* open_requests have no inode */
2013 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2015 const struct in6_addr *dest, *src;
2018 unsigned long timer_expires;
2019 const struct inet_sock *inet = inet_sk(sp);
2020 const struct tcp_sock *tp = tcp_sk(sp);
2021 const struct inet_connection_sock *icsk = inet_csk(sp);
2022 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2026 dest = &sp->sk_v6_daddr;
2027 src = &sp->sk_v6_rcv_saddr;
2028 destp = ntohs(inet->inet_dport);
2029 srcp = ntohs(inet->inet_sport);
2031 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2032 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2033 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2035 timer_expires = icsk->icsk_timeout;
2036 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2038 timer_expires = icsk->icsk_timeout;
2039 } else if (timer_pending(&sp->sk_timer)) {
2041 timer_expires = sp->sk_timer.expires;
2044 timer_expires = jiffies;
2047 state = inet_sk_state_load(sp);
2048 if (state == TCP_LISTEN)
2049 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2051 /* Because we don't lock the socket,
2052 * we might find a transient negative value.
2054 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2055 READ_ONCE(tp->copied_seq), 0);
2058 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2059 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2061 src->s6_addr32[0], src->s6_addr32[1],
2062 src->s6_addr32[2], src->s6_addr32[3], srcp,
2063 dest->s6_addr32[0], dest->s6_addr32[1],
2064 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2066 READ_ONCE(tp->write_seq) - tp->snd_una,
2069 jiffies_delta_to_clock_t(timer_expires - jiffies),
2070 icsk->icsk_retransmits,
2071 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2072 icsk->icsk_probes_out,
2074 refcount_read(&sp->sk_refcnt), sp,
2075 jiffies_to_clock_t(icsk->icsk_rto),
2076 jiffies_to_clock_t(icsk->icsk_ack.ato),
2077 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2079 state == TCP_LISTEN ?
2080 fastopenq->max_qlen :
2081 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2085 static void get_timewait6_sock(struct seq_file *seq,
2086 struct inet_timewait_sock *tw, int i)
2088 long delta = tw->tw_timer.expires - jiffies;
2089 const struct in6_addr *dest, *src;
2092 dest = &tw->tw_v6_daddr;
2093 src = &tw->tw_v6_rcv_saddr;
2094 destp = ntohs(tw->tw_dport);
2095 srcp = ntohs(tw->tw_sport);
2098 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2099 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2101 src->s6_addr32[0], src->s6_addr32[1],
2102 src->s6_addr32[2], src->s6_addr32[3], srcp,
2103 dest->s6_addr32[0], dest->s6_addr32[1],
2104 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2105 tw->tw_substate, 0, 0,
2106 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2107 refcount_read(&tw->tw_refcnt), tw);
2110 static int tcp6_seq_show(struct seq_file *seq, void *v)
2112 struct tcp_iter_state *st;
2113 struct sock *sk = v;
2115 if (v == SEQ_START_TOKEN) {
2120 "st tx_queue rx_queue tr tm->when retrnsmt"
2121 " uid timeout inode\n");
2126 if (sk->sk_state == TCP_TIME_WAIT)
2127 get_timewait6_sock(seq, v, st->num);
2128 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2129 get_openreq6(seq, v, st->num);
2131 get_tcp6_sock(seq, v, st->num);
2136 static const struct seq_operations tcp6_seq_ops = {
2137 .show = tcp6_seq_show,
2138 .start = tcp_seq_start,
2139 .next = tcp_seq_next,
2140 .stop = tcp_seq_stop,
2143 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2147 int __net_init tcp6_proc_init(struct net *net)
2149 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2150 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2155 void tcp6_proc_exit(struct net *net)
2157 remove_proc_entry("tcp6", net->proc_net);
2161 struct proto tcpv6_prot = {
2163 .owner = THIS_MODULE,
2165 .pre_connect = tcp_v6_pre_connect,
2166 .connect = tcp_v6_connect,
2167 .disconnect = tcp_disconnect,
2168 .accept = inet_csk_accept,
2170 .init = tcp_v6_init_sock,
2171 .destroy = tcp_v6_destroy_sock,
2172 .shutdown = tcp_shutdown,
2173 .setsockopt = tcp_setsockopt,
2174 .getsockopt = tcp_getsockopt,
2175 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
2176 .keepalive = tcp_set_keepalive,
2177 .recvmsg = tcp_recvmsg,
2178 .sendmsg = tcp_sendmsg,
2179 .sendpage = tcp_sendpage,
2180 .backlog_rcv = tcp_v6_do_rcv,
2181 .release_cb = tcp_release_cb,
2183 .unhash = inet_unhash,
2184 .get_port = inet_csk_get_port,
2185 .put_port = inet_put_port,
2186 #ifdef CONFIG_BPF_SYSCALL
2187 .psock_update_sk_prot = tcp_bpf_update_proto,
2189 .enter_memory_pressure = tcp_enter_memory_pressure,
2190 .leave_memory_pressure = tcp_leave_memory_pressure,
2191 .stream_memory_free = tcp_stream_memory_free,
2192 .sockets_allocated = &tcp_sockets_allocated,
2193 .memory_allocated = &tcp_memory_allocated,
2194 .memory_pressure = &tcp_memory_pressure,
2195 .orphan_count = &tcp_orphan_count,
2196 .sysctl_mem = sysctl_tcp_mem,
2197 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2198 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2199 .max_header = MAX_TCP_HEADER,
2200 .obj_size = sizeof(struct tcp6_sock),
2201 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2202 .twsk_prot = &tcp6_timewait_sock_ops,
2203 .rsk_prot = &tcp6_request_sock_ops,
2204 .h.hashinfo = &tcp_hashinfo,
2205 .no_autobind = true,
2206 .diag_destroy = tcp_abort,
2208 EXPORT_SYMBOL_GPL(tcpv6_prot);
2210 /* thinking of making this const? Don't.
2211 * early_demux can change based on sysctl.
2213 static struct inet6_protocol tcpv6_protocol = {
2214 .early_demux = tcp_v6_early_demux,
2215 .early_demux_handler = tcp_v6_early_demux,
2216 .handler = tcp_v6_rcv,
2217 .err_handler = tcp_v6_err,
2218 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2221 static struct inet_protosw tcpv6_protosw = {
2222 .type = SOCK_STREAM,
2223 .protocol = IPPROTO_TCP,
2224 .prot = &tcpv6_prot,
2225 .ops = &inet6_stream_ops,
2226 .flags = INET_PROTOSW_PERMANENT |
2230 static int __net_init tcpv6_net_init(struct net *net)
2232 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2233 SOCK_RAW, IPPROTO_TCP, net);
2236 static void __net_exit tcpv6_net_exit(struct net *net)
2238 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2241 static struct pernet_operations tcpv6_net_ops = {
2242 .init = tcpv6_net_init,
2243 .exit = tcpv6_net_exit,
2246 int __init tcpv6_init(void)
2250 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2254 /* register inet6 protocol */
2255 ret = inet6_register_protosw(&tcpv6_protosw);
2257 goto out_tcpv6_protocol;
2259 ret = register_pernet_subsys(&tcpv6_net_ops);
2261 goto out_tcpv6_protosw;
2263 ret = mptcpv6_init();
2265 goto out_tcpv6_pernet_subsys;
2270 out_tcpv6_pernet_subsys:
2271 unregister_pernet_subsys(&tcpv6_net_ops);
2273 inet6_unregister_protosw(&tcpv6_protosw);
2275 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2279 void tcpv6_exit(void)
2281 unregister_pernet_subsys(&tcpv6_net_ops);
2282 inet6_unregister_protosw(&tcpv6_protosw);
2283 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);