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_connection_sock *icsk = inet_csk(sk);
150 struct in6_addr *saddr = NULL, *final_p, final;
151 struct inet_timewait_death_row *tcp_death_row;
152 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
153 struct inet_sock *inet = inet_sk(sk);
154 struct tcp_sock *tp = tcp_sk(sk);
155 struct net *net = sock_net(sk);
156 struct ipv6_txoptions *opt;
157 struct dst_entry *dst;
162 if (addr_len < SIN6_LEN_RFC2133)
165 if (usin->sin6_family != AF_INET6)
166 return -EAFNOSUPPORT;
168 memset(&fl6, 0, sizeof(fl6));
171 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
172 IP6_ECN_flow_init(fl6.flowlabel);
173 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
174 struct ip6_flowlabel *flowlabel;
175 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
176 if (IS_ERR(flowlabel))
178 fl6_sock_release(flowlabel);
183 * connect() to INADDR_ANY means loopback (BSD'ism).
186 if (ipv6_addr_any(&usin->sin6_addr)) {
187 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
188 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
191 usin->sin6_addr = in6addr_loopback;
194 addr_type = ipv6_addr_type(&usin->sin6_addr);
196 if (addr_type & IPV6_ADDR_MULTICAST)
199 if (addr_type&IPV6_ADDR_LINKLOCAL) {
200 if (addr_len >= sizeof(struct sockaddr_in6) &&
201 usin->sin6_scope_id) {
202 /* If interface is set while binding, indices
205 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
208 sk->sk_bound_dev_if = usin->sin6_scope_id;
211 /* Connect to link-local address requires an interface */
212 if (!sk->sk_bound_dev_if)
216 if (tp->rx_opt.ts_recent_stamp &&
217 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
218 tp->rx_opt.ts_recent = 0;
219 tp->rx_opt.ts_recent_stamp = 0;
220 WRITE_ONCE(tp->write_seq, 0);
223 sk->sk_v6_daddr = usin->sin6_addr;
224 np->flow_label = fl6.flowlabel;
230 if (addr_type & IPV6_ADDR_MAPPED) {
231 u32 exthdrlen = icsk->icsk_ext_hdr_len;
232 struct sockaddr_in sin;
234 if (ipv6_only_sock(sk))
237 sin.sin_family = AF_INET;
238 sin.sin_port = usin->sin6_port;
239 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
241 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
242 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_mapped);
244 mptcpv6_handle_mapped(sk, true);
245 sk->sk_backlog_rcv = tcp_v4_do_rcv;
246 #ifdef CONFIG_TCP_MD5SIG
247 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
250 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
253 icsk->icsk_ext_hdr_len = exthdrlen;
254 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
255 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_specific);
257 mptcpv6_handle_mapped(sk, false);
258 sk->sk_backlog_rcv = tcp_v6_do_rcv;
259 #ifdef CONFIG_TCP_MD5SIG
260 tp->af_specific = &tcp_sock_ipv6_specific;
264 np->saddr = sk->sk_v6_rcv_saddr;
269 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
270 saddr = &sk->sk_v6_rcv_saddr;
272 fl6.flowi6_proto = IPPROTO_TCP;
273 fl6.daddr = sk->sk_v6_daddr;
274 fl6.saddr = saddr ? *saddr : np->saddr;
275 fl6.flowlabel = ip6_make_flowinfo(np->tclass, np->flow_label);
276 fl6.flowi6_oif = sk->sk_bound_dev_if;
277 fl6.flowi6_mark = sk->sk_mark;
278 fl6.fl6_dport = usin->sin6_port;
279 fl6.fl6_sport = inet->inet_sport;
280 fl6.flowi6_uid = sk->sk_uid;
282 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
283 final_p = fl6_update_dst(&fl6, opt, &final);
285 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
287 dst = ip6_dst_lookup_flow(net, sk, &fl6, final_p);
293 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
298 err = inet_bhash2_update_saddr(sk, saddr, AF_INET6);
303 /* set the source address */
305 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
307 sk->sk_gso_type = SKB_GSO_TCPV6;
308 ip6_dst_store(sk, dst, NULL, NULL);
310 icsk->icsk_ext_hdr_len = 0;
312 icsk->icsk_ext_hdr_len = opt->opt_flen +
315 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
317 inet->inet_dport = usin->sin6_port;
319 tcp_set_state(sk, TCP_SYN_SENT);
320 err = inet6_hash_connect(tcp_death_row, sk);
326 if (likely(!tp->repair)) {
328 WRITE_ONCE(tp->write_seq,
329 secure_tcpv6_seq(np->saddr.s6_addr32,
330 sk->sk_v6_daddr.s6_addr32,
333 tp->tsoffset = secure_tcpv6_ts_off(net, np->saddr.s6_addr32,
334 sk->sk_v6_daddr.s6_addr32);
337 if (tcp_fastopen_defer_connect(sk, &err))
342 err = tcp_connect(sk);
349 tcp_set_state(sk, TCP_CLOSE);
350 inet_bhash2_reset_saddr(sk);
352 inet->inet_dport = 0;
353 sk->sk_route_caps = 0;
357 static void tcp_v6_mtu_reduced(struct sock *sk)
359 struct dst_entry *dst;
362 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
365 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
367 /* Drop requests trying to increase our current mss.
368 * Check done in __ip6_rt_update_pmtu() is too late.
370 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
373 dst = inet6_csk_update_pmtu(sk, mtu);
377 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
378 tcp_sync_mss(sk, dst_mtu(dst));
379 tcp_simple_retransmit(sk);
383 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
384 u8 type, u8 code, int offset, __be32 info)
386 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
387 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
388 struct net *net = dev_net(skb->dev);
389 struct request_sock *fastopen;
390 struct ipv6_pinfo *np;
397 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
398 &hdr->daddr, th->dest,
399 &hdr->saddr, ntohs(th->source),
400 skb->dev->ifindex, inet6_sdif(skb));
403 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
408 if (sk->sk_state == TCP_TIME_WAIT) {
409 inet_twsk_put(inet_twsk(sk));
412 seq = ntohl(th->seq);
413 fatal = icmpv6_err_convert(type, code, &err);
414 if (sk->sk_state == TCP_NEW_SYN_RECV) {
415 tcp_req_err(sk, seq, fatal);
420 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
421 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
423 if (sk->sk_state == TCP_CLOSE)
426 if (static_branch_unlikely(&ip6_min_hopcount)) {
427 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
428 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
429 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
435 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
436 fastopen = rcu_dereference(tp->fastopen_rsk);
437 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
438 if (sk->sk_state != TCP_LISTEN &&
439 !between(seq, snd_una, tp->snd_nxt)) {
440 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
444 np = tcp_inet6_sk(sk);
446 if (type == NDISC_REDIRECT) {
447 if (!sock_owned_by_user(sk)) {
448 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
451 dst->ops->redirect(dst, sk, skb);
456 if (type == ICMPV6_PKT_TOOBIG) {
457 u32 mtu = ntohl(info);
459 /* We are not interested in TCP_LISTEN and open_requests
460 * (SYN-ACKs send out by Linux are always <576bytes so
461 * they should go through unfragmented).
463 if (sk->sk_state == TCP_LISTEN)
466 if (!ip6_sk_accept_pmtu(sk))
469 if (mtu < IPV6_MIN_MTU)
472 WRITE_ONCE(tp->mtu_info, mtu);
474 if (!sock_owned_by_user(sk))
475 tcp_v6_mtu_reduced(sk);
476 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
483 /* Might be for an request_sock */
484 switch (sk->sk_state) {
487 /* Only in fast or simultaneous open. If a fast open socket is
488 * already accepted it is treated as a connected one below.
490 if (fastopen && !fastopen->sk)
493 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
495 if (!sock_owned_by_user(sk)) {
497 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
501 sk->sk_err_soft = err;
506 /* check if this ICMP message allows revert of backoff.
509 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
510 code == ICMPV6_NOROUTE)
511 tcp_ld_RTO_revert(sk, seq);
514 if (!sock_owned_by_user(sk) && np->recverr) {
518 sk->sk_err_soft = err;
527 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
529 struct request_sock *req,
530 struct tcp_fastopen_cookie *foc,
531 enum tcp_synack_type synack_type,
532 struct sk_buff *syn_skb)
534 struct inet_request_sock *ireq = inet_rsk(req);
535 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
536 struct ipv6_txoptions *opt;
537 struct flowi6 *fl6 = &fl->u.ip6;
542 /* First, grab a route. */
543 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
544 IPPROTO_TCP)) == NULL)
547 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
550 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
551 &ireq->ir_v6_rmt_addr);
553 fl6->daddr = ireq->ir_v6_rmt_addr;
554 if (np->repflow && ireq->pktopts)
555 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
557 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
558 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
559 (np->tclass & INET_ECN_MASK) :
562 if (!INET_ECN_is_capable(tclass) &&
563 tcp_bpf_ca_needs_ecn((struct sock *)req))
564 tclass |= INET_ECN_ECT_0;
567 opt = ireq->ipv6_opt;
569 opt = rcu_dereference(np->opt);
570 err = ip6_xmit(sk, skb, fl6, skb->mark ? : sk->sk_mark, opt,
571 tclass, sk->sk_priority);
573 err = net_xmit_eval(err);
581 static void tcp_v6_reqsk_destructor(struct request_sock *req)
583 kfree(inet_rsk(req)->ipv6_opt);
584 consume_skb(inet_rsk(req)->pktopts);
587 #ifdef CONFIG_TCP_MD5SIG
588 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
589 const struct in6_addr *addr,
592 return tcp_md5_do_lookup(sk, l3index,
593 (union tcp_md5_addr *)addr, AF_INET6);
596 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
597 const struct sock *addr_sk)
601 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
602 addr_sk->sk_bound_dev_if);
603 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
607 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
608 sockptr_t optval, int optlen)
610 struct tcp_md5sig cmd;
611 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
616 if (optlen < sizeof(cmd))
619 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
622 if (sin6->sin6_family != AF_INET6)
625 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
627 if (optname == TCP_MD5SIG_EXT &&
628 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
629 prefixlen = cmd.tcpm_prefixlen;
630 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
634 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
637 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
638 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
639 struct net_device *dev;
642 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
643 if (dev && netif_is_l3_master(dev))
644 l3index = dev->ifindex;
647 /* ok to reference set/not set outside of rcu;
648 * right now device MUST be an L3 master
650 if (!dev || !l3index)
654 if (!cmd.tcpm_keylen) {
655 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
656 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
659 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
660 AF_INET6, prefixlen, l3index, flags);
663 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
666 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
667 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
668 AF_INET, prefixlen, l3index, flags,
669 cmd.tcpm_key, cmd.tcpm_keylen,
672 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
673 AF_INET6, prefixlen, l3index, flags,
674 cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
677 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
678 const struct in6_addr *daddr,
679 const struct in6_addr *saddr,
680 const struct tcphdr *th, int nbytes)
682 struct tcp6_pseudohdr *bp;
683 struct scatterlist sg;
687 /* 1. TCP pseudo-header (RFC2460) */
690 bp->protocol = cpu_to_be32(IPPROTO_TCP);
691 bp->len = cpu_to_be32(nbytes);
693 _th = (struct tcphdr *)(bp + 1);
694 memcpy(_th, th, sizeof(*th));
697 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
698 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
699 sizeof(*bp) + sizeof(*th));
700 return crypto_ahash_update(hp->md5_req);
703 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
704 const struct in6_addr *daddr, struct in6_addr *saddr,
705 const struct tcphdr *th)
707 struct tcp_md5sig_pool *hp;
708 struct ahash_request *req;
710 hp = tcp_get_md5sig_pool();
712 goto clear_hash_noput;
715 if (crypto_ahash_init(req))
717 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
719 if (tcp_md5_hash_key(hp, key))
721 ahash_request_set_crypt(req, NULL, md5_hash, 0);
722 if (crypto_ahash_final(req))
725 tcp_put_md5sig_pool();
729 tcp_put_md5sig_pool();
731 memset(md5_hash, 0, 16);
735 static int tcp_v6_md5_hash_skb(char *md5_hash,
736 const struct tcp_md5sig_key *key,
737 const struct sock *sk,
738 const struct sk_buff *skb)
740 const struct in6_addr *saddr, *daddr;
741 struct tcp_md5sig_pool *hp;
742 struct ahash_request *req;
743 const struct tcphdr *th = tcp_hdr(skb);
745 if (sk) { /* valid for establish/request sockets */
746 saddr = &sk->sk_v6_rcv_saddr;
747 daddr = &sk->sk_v6_daddr;
749 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
750 saddr = &ip6h->saddr;
751 daddr = &ip6h->daddr;
754 hp = tcp_get_md5sig_pool();
756 goto clear_hash_noput;
759 if (crypto_ahash_init(req))
762 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
764 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
766 if (tcp_md5_hash_key(hp, key))
768 ahash_request_set_crypt(req, NULL, md5_hash, 0);
769 if (crypto_ahash_final(req))
772 tcp_put_md5sig_pool();
776 tcp_put_md5sig_pool();
778 memset(md5_hash, 0, 16);
784 static void tcp_v6_init_req(struct request_sock *req,
785 const struct sock *sk_listener,
788 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
789 struct inet_request_sock *ireq = inet_rsk(req);
790 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
792 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
793 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
795 /* So that link locals have meaning */
796 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
797 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
798 ireq->ir_iif = tcp_v6_iif(skb);
800 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
801 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
802 np->rxopt.bits.rxinfo ||
803 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
804 np->rxopt.bits.rxohlim || np->repflow)) {
805 refcount_inc(&skb->users);
810 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
813 struct request_sock *req)
815 tcp_v6_init_req(req, sk, skb);
817 if (security_inet_conn_request(sk, skb, req))
820 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
823 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
825 .obj_size = sizeof(struct tcp6_request_sock),
826 .rtx_syn_ack = tcp_rtx_synack,
827 .send_ack = tcp_v6_reqsk_send_ack,
828 .destructor = tcp_v6_reqsk_destructor,
829 .send_reset = tcp_v6_send_reset,
830 .syn_ack_timeout = tcp_syn_ack_timeout,
833 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
834 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
835 sizeof(struct ipv6hdr),
836 #ifdef CONFIG_TCP_MD5SIG
837 .req_md5_lookup = tcp_v6_md5_lookup,
838 .calc_md5_hash = tcp_v6_md5_hash_skb,
840 #ifdef CONFIG_SYN_COOKIES
841 .cookie_init_seq = cookie_v6_init_sequence,
843 .route_req = tcp_v6_route_req,
844 .init_seq = tcp_v6_init_seq,
845 .init_ts_off = tcp_v6_init_ts_off,
846 .send_synack = tcp_v6_send_synack,
849 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
850 u32 ack, u32 win, u32 tsval, u32 tsecr,
851 int oif, struct tcp_md5sig_key *key, int rst,
852 u8 tclass, __be32 label, u32 priority, u32 txhash)
854 const struct tcphdr *th = tcp_hdr(skb);
856 struct sk_buff *buff;
858 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
859 struct sock *ctl_sk = net->ipv6.tcp_sk;
860 unsigned int tot_len = sizeof(struct tcphdr);
861 __be32 mrst = 0, *topt;
862 struct dst_entry *dst;
866 tot_len += TCPOLEN_TSTAMP_ALIGNED;
867 #ifdef CONFIG_TCP_MD5SIG
869 tot_len += TCPOLEN_MD5SIG_ALIGNED;
874 mrst = mptcp_reset_option(skb);
877 tot_len += sizeof(__be32);
881 buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
885 skb_reserve(buff, MAX_TCP_HEADER);
887 t1 = skb_push(buff, tot_len);
888 skb_reset_transport_header(buff);
890 /* Swap the send and the receive. */
891 memset(t1, 0, sizeof(*t1));
892 t1->dest = th->source;
893 t1->source = th->dest;
894 t1->doff = tot_len / 4;
895 t1->seq = htonl(seq);
896 t1->ack_seq = htonl(ack);
897 t1->ack = !rst || !th->ack;
899 t1->window = htons(win);
901 topt = (__be32 *)(t1 + 1);
904 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
905 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
906 *topt++ = htonl(tsval);
907 *topt++ = htonl(tsecr);
913 #ifdef CONFIG_TCP_MD5SIG
915 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
916 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
917 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
918 &ipv6_hdr(skb)->saddr,
919 &ipv6_hdr(skb)->daddr, t1);
923 memset(&fl6, 0, sizeof(fl6));
924 fl6.daddr = ipv6_hdr(skb)->saddr;
925 fl6.saddr = ipv6_hdr(skb)->daddr;
926 fl6.flowlabel = label;
928 buff->ip_summed = CHECKSUM_PARTIAL;
930 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
932 fl6.flowi6_proto = IPPROTO_TCP;
933 if (rt6_need_strict(&fl6.daddr) && !oif)
934 fl6.flowi6_oif = tcp_v6_iif(skb);
936 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
939 fl6.flowi6_oif = oif;
943 if (sk->sk_state == TCP_TIME_WAIT)
944 mark = inet_twsk(sk)->tw_mark;
947 skb_set_delivery_time(buff, tcp_transmit_time(sk), true);
950 /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */
951 skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4);
953 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
954 fl6.fl6_dport = t1->dest;
955 fl6.fl6_sport = t1->source;
956 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
957 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
959 /* Pass a socket to ip6_dst_lookup either it is for RST
960 * Underlying function will use this to retrieve the network
963 if (sk && sk->sk_state != TCP_TIME_WAIT)
964 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/
966 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL);
968 skb_dst_set(buff, dst);
969 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
970 tclass & ~INET_ECN_MASK, priority);
971 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
973 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
980 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
982 const struct tcphdr *th = tcp_hdr(skb);
983 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
984 u32 seq = 0, ack_seq = 0;
985 struct tcp_md5sig_key *key = NULL;
986 #ifdef CONFIG_TCP_MD5SIG
987 const __u8 *hash_location = NULL;
988 unsigned char newhash[16];
990 struct sock *sk1 = NULL;
1001 /* If sk not NULL, it means we did a successful lookup and incoming
1002 * route had to be correct. prequeue might have dropped our dst.
1004 if (!sk && !ipv6_unicast_destination(skb))
1007 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1008 #ifdef CONFIG_TCP_MD5SIG
1010 hash_location = tcp_parse_md5sig_option(th);
1011 if (sk && sk_fullsock(sk)) {
1014 /* sdif set, means packet ingressed via a device
1015 * in an L3 domain and inet_iif is set to it.
1017 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1018 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1019 } else if (hash_location) {
1020 int dif = tcp_v6_iif_l3_slave(skb);
1021 int sdif = tcp_v6_sdif(skb);
1025 * active side is lost. Try to find listening socket through
1026 * source port, and then find md5 key through listening socket.
1027 * we are not loose security here:
1028 * Incoming packet is checked with md5 hash with finding key,
1029 * no RST generated if md5 hash doesn't match.
1031 sk1 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1032 NULL, 0, &ipv6h->saddr, th->source,
1033 &ipv6h->daddr, ntohs(th->source),
1038 /* sdif set, means packet ingressed via a device
1039 * in an L3 domain and dif is set to it.
1041 l3index = tcp_v6_sdif(skb) ? dif : 0;
1043 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1047 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1048 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1054 seq = ntohl(th->ack_seq);
1056 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1060 oif = sk->sk_bound_dev_if;
1061 if (sk_fullsock(sk)) {
1062 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1064 trace_tcp_send_reset(sk, skb);
1066 label = ip6_flowlabel(ipv6h);
1067 priority = sk->sk_priority;
1068 txhash = sk->sk_hash;
1070 if (sk->sk_state == TCP_TIME_WAIT) {
1071 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1072 priority = inet_twsk(sk)->tw_priority;
1073 txhash = inet_twsk(sk)->tw_txhash;
1076 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1077 label = ip6_flowlabel(ipv6h);
1080 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1,
1081 ipv6_get_dsfield(ipv6h), label, priority, txhash);
1083 #ifdef CONFIG_TCP_MD5SIG
1089 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1090 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1091 struct tcp_md5sig_key *key, u8 tclass,
1092 __be32 label, u32 priority, u32 txhash)
1094 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1095 tclass, label, priority, txhash);
1098 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1100 struct inet_timewait_sock *tw = inet_twsk(sk);
1101 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1103 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1104 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1105 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1106 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1107 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority,
1113 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1114 struct request_sock *req)
1118 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1120 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1121 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1124 * The window field (SEG.WND) of every outgoing segment, with the
1125 * exception of <SYN> segments, MUST be right-shifted by
1126 * Rcv.Wind.Shift bits:
1128 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1129 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1130 tcp_rsk(req)->rcv_nxt,
1131 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1132 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1133 req->ts_recent, sk->sk_bound_dev_if,
1134 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1135 ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority,
1136 tcp_rsk(req)->txhash);
1140 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1142 #ifdef CONFIG_SYN_COOKIES
1143 const struct tcphdr *th = tcp_hdr(skb);
1146 sk = cookie_v6_check(sk, skb);
1151 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1152 struct tcphdr *th, u32 *cookie)
1155 #ifdef CONFIG_SYN_COOKIES
1156 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1157 &tcp_request_sock_ipv6_ops, sk, th);
1159 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1160 tcp_synq_overflow(sk);
1166 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1168 if (skb->protocol == htons(ETH_P_IP))
1169 return tcp_v4_conn_request(sk, skb);
1171 if (!ipv6_unicast_destination(skb))
1174 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1175 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1179 return tcp_conn_request(&tcp6_request_sock_ops,
1180 &tcp_request_sock_ipv6_ops, sk, skb);
1184 return 0; /* don't send reset */
1187 static void tcp_v6_restore_cb(struct sk_buff *skb)
1189 /* We need to move header back to the beginning if xfrm6_policy_check()
1190 * and tcp_v6_fill_cb() are going to be called again.
1191 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1193 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1194 sizeof(struct inet6_skb_parm));
1197 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1198 struct request_sock *req,
1199 struct dst_entry *dst,
1200 struct request_sock *req_unhash,
1203 struct inet_request_sock *ireq;
1204 struct ipv6_pinfo *newnp;
1205 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1206 struct ipv6_txoptions *opt;
1207 struct inet_sock *newinet;
1208 bool found_dup_sk = false;
1209 struct tcp_sock *newtp;
1211 #ifdef CONFIG_TCP_MD5SIG
1212 struct tcp_md5sig_key *key;
1217 if (skb->protocol == htons(ETH_P_IP)) {
1222 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1223 req_unhash, own_req);
1228 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1230 newnp = tcp_inet6_sk(newsk);
1231 newtp = tcp_sk(newsk);
1233 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1235 newnp->saddr = newsk->sk_v6_rcv_saddr;
1237 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1238 if (sk_is_mptcp(newsk))
1239 mptcpv6_handle_mapped(newsk, true);
1240 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1241 #ifdef CONFIG_TCP_MD5SIG
1242 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1245 newnp->ipv6_mc_list = NULL;
1246 newnp->ipv6_ac_list = NULL;
1247 newnp->ipv6_fl_list = NULL;
1248 newnp->pktoptions = NULL;
1250 newnp->mcast_oif = inet_iif(skb);
1251 newnp->mcast_hops = ip_hdr(skb)->ttl;
1252 newnp->rcv_flowinfo = 0;
1254 newnp->flow_label = 0;
1257 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1258 * here, tcp_create_openreq_child now does this for us, see the comment in
1259 * that function for the gory details. -acme
1262 /* It is tricky place. Until this moment IPv4 tcp
1263 worked with IPv6 icsk.icsk_af_ops.
1266 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1271 ireq = inet_rsk(req);
1273 if (sk_acceptq_is_full(sk))
1277 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1282 newsk = tcp_create_openreq_child(sk, req, skb);
1287 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1288 * count here, tcp_create_openreq_child now does this for us, see the
1289 * comment in that function for the gory details. -acme
1292 newsk->sk_gso_type = SKB_GSO_TCPV6;
1293 ip6_dst_store(newsk, dst, NULL, NULL);
1294 inet6_sk_rx_dst_set(newsk, skb);
1296 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1298 newtp = tcp_sk(newsk);
1299 newinet = inet_sk(newsk);
1300 newnp = tcp_inet6_sk(newsk);
1302 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1304 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1305 newnp->saddr = ireq->ir_v6_loc_addr;
1306 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1307 newsk->sk_bound_dev_if = ireq->ir_iif;
1309 /* Now IPv6 options...
1311 First: no IPv4 options.
1313 newinet->inet_opt = NULL;
1314 newnp->ipv6_mc_list = NULL;
1315 newnp->ipv6_ac_list = NULL;
1316 newnp->ipv6_fl_list = NULL;
1319 newnp->rxopt.all = np->rxopt.all;
1321 newnp->pktoptions = NULL;
1323 newnp->mcast_oif = tcp_v6_iif(skb);
1324 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1325 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1327 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1329 /* Set ToS of the new socket based upon the value of incoming SYN.
1330 * ECT bits are set later in tcp_init_transfer().
1332 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1333 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1335 /* Clone native IPv6 options from listening socket (if any)
1337 Yes, keeping reference count would be much more clever,
1338 but we make one more one thing there: reattach optmem
1341 opt = ireq->ipv6_opt;
1343 opt = rcu_dereference(np->opt);
1345 opt = ipv6_dup_options(newsk, opt);
1346 RCU_INIT_POINTER(newnp->opt, opt);
1348 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1350 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1353 tcp_ca_openreq_child(newsk, dst);
1355 tcp_sync_mss(newsk, dst_mtu(dst));
1356 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1358 tcp_initialize_rcv_mss(newsk);
1360 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1361 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1363 #ifdef CONFIG_TCP_MD5SIG
1364 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1366 /* Copy over the MD5 key from the original socket */
1367 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1369 /* We're using one, so create a matching key
1370 * on the newsk structure. If we fail to get
1371 * memory, then we end up not copying the key
1374 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1375 AF_INET6, 128, l3index, key->flags, key->key, key->keylen,
1376 sk_gfp_mask(sk, GFP_ATOMIC));
1380 if (__inet_inherit_port(sk, newsk) < 0) {
1381 inet_csk_prepare_forced_close(newsk);
1385 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1388 tcp_move_syn(newtp, req);
1390 /* Clone pktoptions received with SYN, if we own the req */
1391 if (ireq->pktopts) {
1392 newnp->pktoptions = skb_clone_and_charge_r(ireq->pktopts, newsk);
1393 consume_skb(ireq->pktopts);
1394 ireq->pktopts = NULL;
1395 if (newnp->pktoptions)
1396 tcp_v6_restore_cb(newnp->pktoptions);
1399 if (!req_unhash && found_dup_sk) {
1400 /* This code path should only be executed in the
1401 * syncookie case only
1403 bh_unlock_sock(newsk);
1412 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1420 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1422 /* The socket must have it's spinlock held when we get
1423 * here, unless it is a TCP_LISTEN socket.
1425 * We have a potential double-lock case here, so even when
1426 * doing backlog processing we use the BH locking scheme.
1427 * This is because we cannot sleep with the original spinlock
1430 INDIRECT_CALLABLE_SCOPE
1431 int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1433 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1434 struct sk_buff *opt_skb = NULL;
1435 enum skb_drop_reason reason;
1436 struct tcp_sock *tp;
1438 /* Imagine: socket is IPv6. IPv4 packet arrives,
1439 goes to IPv4 receive handler and backlogged.
1440 From backlog it always goes here. Kerboom...
1441 Fortunately, tcp_rcv_established and rcv_established
1442 handle them correctly, but it is not case with
1443 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1446 if (skb->protocol == htons(ETH_P_IP))
1447 return tcp_v4_do_rcv(sk, skb);
1450 * socket locking is here for SMP purposes as backlog rcv
1451 * is currently called with bh processing disabled.
1454 /* Do Stevens' IPV6_PKTOPTIONS.
1456 Yes, guys, it is the only place in our code, where we
1457 may make it not affecting IPv4.
1458 The rest of code is protocol independent,
1459 and I do not like idea to uglify IPv4.
1461 Actually, all the idea behind IPV6_PKTOPTIONS
1462 looks not very well thought. For now we latch
1463 options, received in the last packet, enqueued
1464 by tcp. Feel free to propose better solution.
1468 opt_skb = skb_clone_and_charge_r(skb, sk);
1470 reason = SKB_DROP_REASON_NOT_SPECIFIED;
1471 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1472 struct dst_entry *dst;
1474 dst = rcu_dereference_protected(sk->sk_rx_dst,
1475 lockdep_sock_is_held(sk));
1477 sock_rps_save_rxhash(sk, skb);
1478 sk_mark_napi_id(sk, skb);
1480 if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1481 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1482 dst, sk->sk_rx_dst_cookie) == NULL) {
1483 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1488 tcp_rcv_established(sk, skb);
1490 goto ipv6_pktoptions;
1494 if (tcp_checksum_complete(skb))
1497 if (sk->sk_state == TCP_LISTEN) {
1498 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1504 if (tcp_child_process(sk, nsk, skb))
1507 __kfree_skb(opt_skb);
1511 sock_rps_save_rxhash(sk, skb);
1513 if (tcp_rcv_state_process(sk, skb))
1516 goto ipv6_pktoptions;
1520 tcp_v6_send_reset(sk, skb);
1523 __kfree_skb(opt_skb);
1524 kfree_skb_reason(skb, reason);
1527 reason = SKB_DROP_REASON_TCP_CSUM;
1528 trace_tcp_bad_csum(skb);
1529 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1530 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1535 /* Do you ask, what is it?
1537 1. skb was enqueued by tcp.
1538 2. skb is added to tail of read queue, rather than out of order.
1539 3. socket is not in passive state.
1540 4. Finally, it really contains options, which user wants to receive.
1543 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1544 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1545 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1546 np->mcast_oif = tcp_v6_iif(opt_skb);
1547 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1548 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1549 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1550 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1552 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1553 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1554 tcp_v6_restore_cb(opt_skb);
1555 opt_skb = xchg(&np->pktoptions, opt_skb);
1557 __kfree_skb(opt_skb);
1558 opt_skb = xchg(&np->pktoptions, NULL);
1562 consume_skb(opt_skb);
1566 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1567 const struct tcphdr *th)
1569 /* This is tricky: we move IP6CB at its correct location into
1570 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1571 * _decode_session6() uses IP6CB().
1572 * barrier() makes sure compiler won't play aliasing games.
1574 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1575 sizeof(struct inet6_skb_parm));
1578 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1579 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1580 skb->len - th->doff*4);
1581 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1582 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1583 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1584 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1585 TCP_SKB_CB(skb)->sacked = 0;
1586 TCP_SKB_CB(skb)->has_rxtstamp =
1587 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1590 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1592 enum skb_drop_reason drop_reason;
1593 int sdif = inet6_sdif(skb);
1594 int dif = inet6_iif(skb);
1595 const struct tcphdr *th;
1596 const struct ipv6hdr *hdr;
1600 struct net *net = dev_net(skb->dev);
1602 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1603 if (skb->pkt_type != PACKET_HOST)
1607 * Count it even if it's bad.
1609 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1611 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1614 th = (const struct tcphdr *)skb->data;
1616 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
1617 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1620 if (!pskb_may_pull(skb, th->doff*4))
1623 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1626 th = (const struct tcphdr *)skb->data;
1627 hdr = ipv6_hdr(skb);
1630 sk = __inet6_lookup_skb(net->ipv4.tcp_death_row.hashinfo, skb, __tcp_hdrlen(th),
1631 th->source, th->dest, inet6_iif(skb), sdif,
1637 if (sk->sk_state == TCP_TIME_WAIT)
1640 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1641 struct request_sock *req = inet_reqsk(sk);
1642 bool req_stolen = false;
1645 sk = req->rsk_listener;
1646 drop_reason = tcp_inbound_md5_hash(sk, skb,
1647 &hdr->saddr, &hdr->daddr,
1648 AF_INET6, dif, sdif);
1650 sk_drops_add(sk, skb);
1654 if (tcp_checksum_complete(skb)) {
1658 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1659 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1661 inet_csk_reqsk_queue_drop_and_put(sk, req);
1665 /* reuseport_migrate_sock() has already held one sk_refcnt
1673 if (!tcp_filter(sk, skb)) {
1674 th = (const struct tcphdr *)skb->data;
1675 hdr = ipv6_hdr(skb);
1676 tcp_v6_fill_cb(skb, hdr, th);
1677 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1679 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1684 /* Another cpu got exclusive access to req
1685 * and created a full blown socket.
1686 * Try to feed this packet to this socket
1687 * instead of discarding it.
1689 tcp_v6_restore_cb(skb);
1693 goto discard_and_relse;
1697 tcp_v6_restore_cb(skb);
1698 } else if (tcp_child_process(sk, nsk, skb)) {
1699 tcp_v6_send_reset(nsk, skb);
1700 goto discard_and_relse;
1707 if (static_branch_unlikely(&ip6_min_hopcount)) {
1708 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
1709 if (hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
1710 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1711 goto discard_and_relse;
1715 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
1716 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1717 goto discard_and_relse;
1720 drop_reason = tcp_inbound_md5_hash(sk, skb, &hdr->saddr, &hdr->daddr,
1721 AF_INET6, dif, sdif);
1723 goto discard_and_relse;
1725 if (tcp_filter(sk, skb)) {
1726 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1727 goto discard_and_relse;
1729 th = (const struct tcphdr *)skb->data;
1730 hdr = ipv6_hdr(skb);
1731 tcp_v6_fill_cb(skb, hdr, th);
1735 if (sk->sk_state == TCP_LISTEN) {
1736 ret = tcp_v6_do_rcv(sk, skb);
1737 goto put_and_return;
1740 sk_incoming_cpu_update(sk);
1742 bh_lock_sock_nested(sk);
1743 tcp_segs_in(tcp_sk(sk), skb);
1745 if (!sock_owned_by_user(sk)) {
1746 ret = tcp_v6_do_rcv(sk, skb);
1748 if (tcp_add_backlog(sk, skb, &drop_reason))
1749 goto discard_and_relse;
1755 return ret ? -1 : 0;
1758 drop_reason = SKB_DROP_REASON_NO_SOCKET;
1759 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1762 tcp_v6_fill_cb(skb, hdr, th);
1764 if (tcp_checksum_complete(skb)) {
1766 drop_reason = SKB_DROP_REASON_TCP_CSUM;
1767 trace_tcp_bad_csum(skb);
1768 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1770 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1772 tcp_v6_send_reset(NULL, skb);
1776 SKB_DR_OR(drop_reason, NOT_SPECIFIED);
1777 kfree_skb_reason(skb, drop_reason);
1781 sk_drops_add(sk, skb);
1787 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1788 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1789 inet_twsk_put(inet_twsk(sk));
1793 tcp_v6_fill_cb(skb, hdr, th);
1795 if (tcp_checksum_complete(skb)) {
1796 inet_twsk_put(inet_twsk(sk));
1800 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1805 sk2 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1806 skb, __tcp_hdrlen(th),
1807 &ipv6_hdr(skb)->saddr, th->source,
1808 &ipv6_hdr(skb)->daddr,
1810 tcp_v6_iif_l3_slave(skb),
1813 struct inet_timewait_sock *tw = inet_twsk(sk);
1814 inet_twsk_deschedule_put(tw);
1816 tcp_v6_restore_cb(skb);
1824 tcp_v6_timewait_ack(sk, skb);
1827 tcp_v6_send_reset(sk, skb);
1828 inet_twsk_deschedule_put(inet_twsk(sk));
1830 case TCP_TW_SUCCESS:
1836 void tcp_v6_early_demux(struct sk_buff *skb)
1838 struct net *net = dev_net(skb->dev);
1839 const struct ipv6hdr *hdr;
1840 const struct tcphdr *th;
1843 if (skb->pkt_type != PACKET_HOST)
1846 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1849 hdr = ipv6_hdr(skb);
1852 if (th->doff < sizeof(struct tcphdr) / 4)
1855 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1856 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
1857 &hdr->saddr, th->source,
1858 &hdr->daddr, ntohs(th->dest),
1859 inet6_iif(skb), inet6_sdif(skb));
1862 skb->destructor = sock_edemux;
1863 if (sk_fullsock(sk)) {
1864 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1867 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1869 sk->sk_rx_dst_ifindex == skb->skb_iif)
1870 skb_dst_set_noref(skb, dst);
1875 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1876 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1877 .twsk_unique = tcp_twsk_unique,
1878 .twsk_destructor = tcp_twsk_destructor,
1881 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1883 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr);
1886 const struct inet_connection_sock_af_ops ipv6_specific = {
1887 .queue_xmit = inet6_csk_xmit,
1888 .send_check = tcp_v6_send_check,
1889 .rebuild_header = inet6_sk_rebuild_header,
1890 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1891 .conn_request = tcp_v6_conn_request,
1892 .syn_recv_sock = tcp_v6_syn_recv_sock,
1893 .net_header_len = sizeof(struct ipv6hdr),
1894 .net_frag_header_len = sizeof(struct frag_hdr),
1895 .setsockopt = ipv6_setsockopt,
1896 .getsockopt = ipv6_getsockopt,
1897 .addr2sockaddr = inet6_csk_addr2sockaddr,
1898 .sockaddr_len = sizeof(struct sockaddr_in6),
1899 .mtu_reduced = tcp_v6_mtu_reduced,
1902 #ifdef CONFIG_TCP_MD5SIG
1903 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1904 .md5_lookup = tcp_v6_md5_lookup,
1905 .calc_md5_hash = tcp_v6_md5_hash_skb,
1906 .md5_parse = tcp_v6_parse_md5_keys,
1911 * TCP over IPv4 via INET6 API
1913 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1914 .queue_xmit = ip_queue_xmit,
1915 .send_check = tcp_v4_send_check,
1916 .rebuild_header = inet_sk_rebuild_header,
1917 .sk_rx_dst_set = inet_sk_rx_dst_set,
1918 .conn_request = tcp_v6_conn_request,
1919 .syn_recv_sock = tcp_v6_syn_recv_sock,
1920 .net_header_len = sizeof(struct iphdr),
1921 .setsockopt = ipv6_setsockopt,
1922 .getsockopt = ipv6_getsockopt,
1923 .addr2sockaddr = inet6_csk_addr2sockaddr,
1924 .sockaddr_len = sizeof(struct sockaddr_in6),
1925 .mtu_reduced = tcp_v4_mtu_reduced,
1928 #ifdef CONFIG_TCP_MD5SIG
1929 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1930 .md5_lookup = tcp_v4_md5_lookup,
1931 .calc_md5_hash = tcp_v4_md5_hash_skb,
1932 .md5_parse = tcp_v6_parse_md5_keys,
1936 /* NOTE: A lot of things set to zero explicitly by call to
1937 * sk_alloc() so need not be done here.
1939 static int tcp_v6_init_sock(struct sock *sk)
1941 struct inet_connection_sock *icsk = inet_csk(sk);
1945 icsk->icsk_af_ops = &ipv6_specific;
1947 #ifdef CONFIG_TCP_MD5SIG
1948 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1954 static void tcp_v6_destroy_sock(struct sock *sk)
1956 tcp_v4_destroy_sock(sk);
1957 inet6_destroy_sock(sk);
1960 #ifdef CONFIG_PROC_FS
1961 /* Proc filesystem TCPv6 sock list dumping. */
1962 static void get_openreq6(struct seq_file *seq,
1963 const struct request_sock *req, int i)
1965 long ttd = req->rsk_timer.expires - jiffies;
1966 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1967 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1973 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1974 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1976 src->s6_addr32[0], src->s6_addr32[1],
1977 src->s6_addr32[2], src->s6_addr32[3],
1978 inet_rsk(req)->ir_num,
1979 dest->s6_addr32[0], dest->s6_addr32[1],
1980 dest->s6_addr32[2], dest->s6_addr32[3],
1981 ntohs(inet_rsk(req)->ir_rmt_port),
1983 0, 0, /* could print option size, but that is af dependent. */
1984 1, /* timers active (only the expire timer) */
1985 jiffies_to_clock_t(ttd),
1987 from_kuid_munged(seq_user_ns(seq),
1988 sock_i_uid(req->rsk_listener)),
1989 0, /* non standard timer */
1990 0, /* open_requests have no inode */
1994 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1996 const struct in6_addr *dest, *src;
1999 unsigned long timer_expires;
2000 const struct inet_sock *inet = inet_sk(sp);
2001 const struct tcp_sock *tp = tcp_sk(sp);
2002 const struct inet_connection_sock *icsk = inet_csk(sp);
2003 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2007 dest = &sp->sk_v6_daddr;
2008 src = &sp->sk_v6_rcv_saddr;
2009 destp = ntohs(inet->inet_dport);
2010 srcp = ntohs(inet->inet_sport);
2012 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2013 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2014 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2016 timer_expires = icsk->icsk_timeout;
2017 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2019 timer_expires = icsk->icsk_timeout;
2020 } else if (timer_pending(&sp->sk_timer)) {
2022 timer_expires = sp->sk_timer.expires;
2025 timer_expires = jiffies;
2028 state = inet_sk_state_load(sp);
2029 if (state == TCP_LISTEN)
2030 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2032 /* Because we don't lock the socket,
2033 * we might find a transient negative value.
2035 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2036 READ_ONCE(tp->copied_seq), 0);
2039 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2040 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2042 src->s6_addr32[0], src->s6_addr32[1],
2043 src->s6_addr32[2], src->s6_addr32[3], srcp,
2044 dest->s6_addr32[0], dest->s6_addr32[1],
2045 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2047 READ_ONCE(tp->write_seq) - tp->snd_una,
2050 jiffies_delta_to_clock_t(timer_expires - jiffies),
2051 icsk->icsk_retransmits,
2052 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2053 icsk->icsk_probes_out,
2055 refcount_read(&sp->sk_refcnt), sp,
2056 jiffies_to_clock_t(icsk->icsk_rto),
2057 jiffies_to_clock_t(icsk->icsk_ack.ato),
2058 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2060 state == TCP_LISTEN ?
2061 fastopenq->max_qlen :
2062 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2066 static void get_timewait6_sock(struct seq_file *seq,
2067 struct inet_timewait_sock *tw, int i)
2069 long delta = tw->tw_timer.expires - jiffies;
2070 const struct in6_addr *dest, *src;
2073 dest = &tw->tw_v6_daddr;
2074 src = &tw->tw_v6_rcv_saddr;
2075 destp = ntohs(tw->tw_dport);
2076 srcp = ntohs(tw->tw_sport);
2079 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2080 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2082 src->s6_addr32[0], src->s6_addr32[1],
2083 src->s6_addr32[2], src->s6_addr32[3], srcp,
2084 dest->s6_addr32[0], dest->s6_addr32[1],
2085 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2086 tw->tw_substate, 0, 0,
2087 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2088 refcount_read(&tw->tw_refcnt), tw);
2091 static int tcp6_seq_show(struct seq_file *seq, void *v)
2093 struct tcp_iter_state *st;
2094 struct sock *sk = v;
2096 if (v == SEQ_START_TOKEN) {
2101 "st tx_queue rx_queue tr tm->when retrnsmt"
2102 " uid timeout inode\n");
2107 if (sk->sk_state == TCP_TIME_WAIT)
2108 get_timewait6_sock(seq, v, st->num);
2109 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2110 get_openreq6(seq, v, st->num);
2112 get_tcp6_sock(seq, v, st->num);
2117 static const struct seq_operations tcp6_seq_ops = {
2118 .show = tcp6_seq_show,
2119 .start = tcp_seq_start,
2120 .next = tcp_seq_next,
2121 .stop = tcp_seq_stop,
2124 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2128 int __net_init tcp6_proc_init(struct net *net)
2130 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2131 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2136 void tcp6_proc_exit(struct net *net)
2138 remove_proc_entry("tcp6", net->proc_net);
2142 struct proto tcpv6_prot = {
2144 .owner = THIS_MODULE,
2146 .pre_connect = tcp_v6_pre_connect,
2147 .connect = tcp_v6_connect,
2148 .disconnect = tcp_disconnect,
2149 .accept = inet_csk_accept,
2151 .init = tcp_v6_init_sock,
2152 .destroy = tcp_v6_destroy_sock,
2153 .shutdown = tcp_shutdown,
2154 .setsockopt = tcp_setsockopt,
2155 .getsockopt = tcp_getsockopt,
2156 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
2157 .keepalive = tcp_set_keepalive,
2158 .recvmsg = tcp_recvmsg,
2159 .sendmsg = tcp_sendmsg,
2160 .sendpage = tcp_sendpage,
2161 .backlog_rcv = tcp_v6_do_rcv,
2162 .release_cb = tcp_release_cb,
2164 .unhash = inet_unhash,
2165 .get_port = inet_csk_get_port,
2166 .put_port = inet_put_port,
2167 #ifdef CONFIG_BPF_SYSCALL
2168 .psock_update_sk_prot = tcp_bpf_update_proto,
2170 .enter_memory_pressure = tcp_enter_memory_pressure,
2171 .leave_memory_pressure = tcp_leave_memory_pressure,
2172 .stream_memory_free = tcp_stream_memory_free,
2173 .sockets_allocated = &tcp_sockets_allocated,
2175 .memory_allocated = &tcp_memory_allocated,
2176 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
2178 .memory_pressure = &tcp_memory_pressure,
2179 .orphan_count = &tcp_orphan_count,
2180 .sysctl_mem = sysctl_tcp_mem,
2181 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2182 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2183 .max_header = MAX_TCP_HEADER,
2184 .obj_size = sizeof(struct tcp6_sock),
2185 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2186 .twsk_prot = &tcp6_timewait_sock_ops,
2187 .rsk_prot = &tcp6_request_sock_ops,
2189 .no_autobind = true,
2190 .diag_destroy = tcp_abort,
2192 EXPORT_SYMBOL_GPL(tcpv6_prot);
2194 static const struct inet6_protocol tcpv6_protocol = {
2195 .handler = tcp_v6_rcv,
2196 .err_handler = tcp_v6_err,
2197 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2200 static struct inet_protosw tcpv6_protosw = {
2201 .type = SOCK_STREAM,
2202 .protocol = IPPROTO_TCP,
2203 .prot = &tcpv6_prot,
2204 .ops = &inet6_stream_ops,
2205 .flags = INET_PROTOSW_PERMANENT |
2209 static int __net_init tcpv6_net_init(struct net *net)
2211 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2212 SOCK_RAW, IPPROTO_TCP, net);
2215 static void __net_exit tcpv6_net_exit(struct net *net)
2217 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2220 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2222 tcp_twsk_purge(net_exit_list, AF_INET6);
2225 static struct pernet_operations tcpv6_net_ops = {
2226 .init = tcpv6_net_init,
2227 .exit = tcpv6_net_exit,
2228 .exit_batch = tcpv6_net_exit_batch,
2231 int __init tcpv6_init(void)
2235 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2239 /* register inet6 protocol */
2240 ret = inet6_register_protosw(&tcpv6_protosw);
2242 goto out_tcpv6_protocol;
2244 ret = register_pernet_subsys(&tcpv6_net_ops);
2246 goto out_tcpv6_protosw;
2248 ret = mptcpv6_init();
2250 goto out_tcpv6_pernet_subsys;
2255 out_tcpv6_pernet_subsys:
2256 unregister_pernet_subsys(&tcpv6_net_ops);
2258 inet6_unregister_protosw(&tcpv6_protosw);
2260 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2264 void tcpv6_exit(void)
2266 unregister_pernet_subsys(&tcpv6_net_ops);
2267 inet6_unregister_protosw(&tcpv6_protosw);
2268 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);