1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Linux INET6 implementation
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);
290 struct inet_bind_hashbucket *prev_addr_hashbucket = NULL;
291 struct in6_addr prev_v6_rcv_saddr;
293 if (icsk->icsk_bind2_hash) {
294 prev_addr_hashbucket = inet_bhashfn_portaddr(&tcp_hashinfo,
297 prev_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
300 sk->sk_v6_rcv_saddr = *saddr;
302 if (prev_addr_hashbucket) {
303 err = inet_bhash2_update_saddr(prev_addr_hashbucket, sk);
305 sk->sk_v6_rcv_saddr = prev_v6_rcv_saddr;
311 /* set the source address */
313 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
315 sk->sk_gso_type = SKB_GSO_TCPV6;
316 ip6_dst_store(sk, dst, NULL, NULL);
318 icsk->icsk_ext_hdr_len = 0;
320 icsk->icsk_ext_hdr_len = opt->opt_flen +
323 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
325 inet->inet_dport = usin->sin6_port;
327 tcp_set_state(sk, TCP_SYN_SENT);
328 tcp_death_row = sock_net(sk)->ipv4.tcp_death_row;
329 err = inet6_hash_connect(tcp_death_row, sk);
335 if (likely(!tp->repair)) {
337 WRITE_ONCE(tp->write_seq,
338 secure_tcpv6_seq(np->saddr.s6_addr32,
339 sk->sk_v6_daddr.s6_addr32,
342 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
344 sk->sk_v6_daddr.s6_addr32);
347 if (tcp_fastopen_defer_connect(sk, &err))
352 err = tcp_connect(sk);
359 tcp_set_state(sk, TCP_CLOSE);
361 inet->inet_dport = 0;
362 sk->sk_route_caps = 0;
366 static void tcp_v6_mtu_reduced(struct sock *sk)
368 struct dst_entry *dst;
371 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
374 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
376 /* Drop requests trying to increase our current mss.
377 * Check done in __ip6_rt_update_pmtu() is too late.
379 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
382 dst = inet6_csk_update_pmtu(sk, mtu);
386 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
387 tcp_sync_mss(sk, dst_mtu(dst));
388 tcp_simple_retransmit(sk);
392 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
393 u8 type, u8 code, int offset, __be32 info)
395 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
396 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
397 struct net *net = dev_net(skb->dev);
398 struct request_sock *fastopen;
399 struct ipv6_pinfo *np;
406 sk = __inet6_lookup_established(net, &tcp_hashinfo,
407 &hdr->daddr, th->dest,
408 &hdr->saddr, ntohs(th->source),
409 skb->dev->ifindex, inet6_sdif(skb));
412 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
417 if (sk->sk_state == TCP_TIME_WAIT) {
418 inet_twsk_put(inet_twsk(sk));
421 seq = ntohl(th->seq);
422 fatal = icmpv6_err_convert(type, code, &err);
423 if (sk->sk_state == TCP_NEW_SYN_RECV) {
424 tcp_req_err(sk, seq, fatal);
429 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
430 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
432 if (sk->sk_state == TCP_CLOSE)
435 if (static_branch_unlikely(&ip6_min_hopcount)) {
436 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
437 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
438 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
444 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
445 fastopen = rcu_dereference(tp->fastopen_rsk);
446 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
447 if (sk->sk_state != TCP_LISTEN &&
448 !between(seq, snd_una, tp->snd_nxt)) {
449 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
453 np = tcp_inet6_sk(sk);
455 if (type == NDISC_REDIRECT) {
456 if (!sock_owned_by_user(sk)) {
457 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
460 dst->ops->redirect(dst, sk, skb);
465 if (type == ICMPV6_PKT_TOOBIG) {
466 u32 mtu = ntohl(info);
468 /* We are not interested in TCP_LISTEN and open_requests
469 * (SYN-ACKs send out by Linux are always <576bytes so
470 * they should go through unfragmented).
472 if (sk->sk_state == TCP_LISTEN)
475 if (!ip6_sk_accept_pmtu(sk))
478 if (mtu < IPV6_MIN_MTU)
481 WRITE_ONCE(tp->mtu_info, mtu);
483 if (!sock_owned_by_user(sk))
484 tcp_v6_mtu_reduced(sk);
485 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
492 /* Might be for an request_sock */
493 switch (sk->sk_state) {
496 /* Only in fast or simultaneous open. If a fast open socket is
497 * already accepted it is treated as a connected one below.
499 if (fastopen && !fastopen->sk)
502 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
504 if (!sock_owned_by_user(sk)) {
506 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
510 sk->sk_err_soft = err;
515 /* check if this ICMP message allows revert of backoff.
518 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
519 code == ICMPV6_NOROUTE)
520 tcp_ld_RTO_revert(sk, seq);
523 if (!sock_owned_by_user(sk) && np->recverr) {
527 sk->sk_err_soft = err;
536 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
538 struct request_sock *req,
539 struct tcp_fastopen_cookie *foc,
540 enum tcp_synack_type synack_type,
541 struct sk_buff *syn_skb)
543 struct inet_request_sock *ireq = inet_rsk(req);
544 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
545 struct ipv6_txoptions *opt;
546 struct flowi6 *fl6 = &fl->u.ip6;
551 /* First, grab a route. */
552 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
553 IPPROTO_TCP)) == NULL)
556 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
559 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
560 &ireq->ir_v6_rmt_addr);
562 fl6->daddr = ireq->ir_v6_rmt_addr;
563 if (np->repflow && ireq->pktopts)
564 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
566 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
567 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
568 (np->tclass & INET_ECN_MASK) :
571 if (!INET_ECN_is_capable(tclass) &&
572 tcp_bpf_ca_needs_ecn((struct sock *)req))
573 tclass |= INET_ECN_ECT_0;
576 opt = ireq->ipv6_opt;
578 opt = rcu_dereference(np->opt);
579 err = ip6_xmit(sk, skb, fl6, skb->mark ? : sk->sk_mark, opt,
580 tclass, sk->sk_priority);
582 err = net_xmit_eval(err);
590 static void tcp_v6_reqsk_destructor(struct request_sock *req)
592 kfree(inet_rsk(req)->ipv6_opt);
593 consume_skb(inet_rsk(req)->pktopts);
596 #ifdef CONFIG_TCP_MD5SIG
597 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
598 const struct in6_addr *addr,
601 return tcp_md5_do_lookup(sk, l3index,
602 (union tcp_md5_addr *)addr, AF_INET6);
605 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
606 const struct sock *addr_sk)
610 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
611 addr_sk->sk_bound_dev_if);
612 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
616 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
617 sockptr_t optval, int optlen)
619 struct tcp_md5sig cmd;
620 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
625 if (optlen < sizeof(cmd))
628 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
631 if (sin6->sin6_family != AF_INET6)
634 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
636 if (optname == TCP_MD5SIG_EXT &&
637 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
638 prefixlen = cmd.tcpm_prefixlen;
639 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
643 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
646 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
647 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
648 struct net_device *dev;
651 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
652 if (dev && netif_is_l3_master(dev))
653 l3index = dev->ifindex;
656 /* ok to reference set/not set outside of rcu;
657 * right now device MUST be an L3 master
659 if (!dev || !l3index)
663 if (!cmd.tcpm_keylen) {
664 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
665 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
668 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
669 AF_INET6, prefixlen, l3index, flags);
672 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
675 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
676 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
677 AF_INET, prefixlen, l3index, flags,
678 cmd.tcpm_key, cmd.tcpm_keylen,
681 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
682 AF_INET6, prefixlen, l3index, flags,
683 cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
686 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
687 const struct in6_addr *daddr,
688 const struct in6_addr *saddr,
689 const struct tcphdr *th, int nbytes)
691 struct tcp6_pseudohdr *bp;
692 struct scatterlist sg;
696 /* 1. TCP pseudo-header (RFC2460) */
699 bp->protocol = cpu_to_be32(IPPROTO_TCP);
700 bp->len = cpu_to_be32(nbytes);
702 _th = (struct tcphdr *)(bp + 1);
703 memcpy(_th, th, sizeof(*th));
706 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
707 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
708 sizeof(*bp) + sizeof(*th));
709 return crypto_ahash_update(hp->md5_req);
712 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
713 const struct in6_addr *daddr, struct in6_addr *saddr,
714 const struct tcphdr *th)
716 struct tcp_md5sig_pool *hp;
717 struct ahash_request *req;
719 hp = tcp_get_md5sig_pool();
721 goto clear_hash_noput;
724 if (crypto_ahash_init(req))
726 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
728 if (tcp_md5_hash_key(hp, key))
730 ahash_request_set_crypt(req, NULL, md5_hash, 0);
731 if (crypto_ahash_final(req))
734 tcp_put_md5sig_pool();
738 tcp_put_md5sig_pool();
740 memset(md5_hash, 0, 16);
744 static int tcp_v6_md5_hash_skb(char *md5_hash,
745 const struct tcp_md5sig_key *key,
746 const struct sock *sk,
747 const struct sk_buff *skb)
749 const struct in6_addr *saddr, *daddr;
750 struct tcp_md5sig_pool *hp;
751 struct ahash_request *req;
752 const struct tcphdr *th = tcp_hdr(skb);
754 if (sk) { /* valid for establish/request sockets */
755 saddr = &sk->sk_v6_rcv_saddr;
756 daddr = &sk->sk_v6_daddr;
758 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
759 saddr = &ip6h->saddr;
760 daddr = &ip6h->daddr;
763 hp = tcp_get_md5sig_pool();
765 goto clear_hash_noput;
768 if (crypto_ahash_init(req))
771 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
773 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
775 if (tcp_md5_hash_key(hp, key))
777 ahash_request_set_crypt(req, NULL, md5_hash, 0);
778 if (crypto_ahash_final(req))
781 tcp_put_md5sig_pool();
785 tcp_put_md5sig_pool();
787 memset(md5_hash, 0, 16);
793 static void tcp_v6_init_req(struct request_sock *req,
794 const struct sock *sk_listener,
797 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
798 struct inet_request_sock *ireq = inet_rsk(req);
799 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
801 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
802 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
804 /* So that link locals have meaning */
805 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
806 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
807 ireq->ir_iif = tcp_v6_iif(skb);
809 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
810 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
811 np->rxopt.bits.rxinfo ||
812 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
813 np->rxopt.bits.rxohlim || np->repflow)) {
814 refcount_inc(&skb->users);
819 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
822 struct request_sock *req)
824 tcp_v6_init_req(req, sk, skb);
826 if (security_inet_conn_request(sk, skb, req))
829 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
832 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
834 .obj_size = sizeof(struct tcp6_request_sock),
835 .rtx_syn_ack = tcp_rtx_synack,
836 .send_ack = tcp_v6_reqsk_send_ack,
837 .destructor = tcp_v6_reqsk_destructor,
838 .send_reset = tcp_v6_send_reset,
839 .syn_ack_timeout = tcp_syn_ack_timeout,
842 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
843 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
844 sizeof(struct ipv6hdr),
845 #ifdef CONFIG_TCP_MD5SIG
846 .req_md5_lookup = tcp_v6_md5_lookup,
847 .calc_md5_hash = tcp_v6_md5_hash_skb,
849 #ifdef CONFIG_SYN_COOKIES
850 .cookie_init_seq = cookie_v6_init_sequence,
852 .route_req = tcp_v6_route_req,
853 .init_seq = tcp_v6_init_seq,
854 .init_ts_off = tcp_v6_init_ts_off,
855 .send_synack = tcp_v6_send_synack,
858 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
859 u32 ack, u32 win, u32 tsval, u32 tsecr,
860 int oif, struct tcp_md5sig_key *key, int rst,
861 u8 tclass, __be32 label, u32 priority, u32 txhash)
863 const struct tcphdr *th = tcp_hdr(skb);
865 struct sk_buff *buff;
867 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
868 struct sock *ctl_sk = net->ipv6.tcp_sk;
869 unsigned int tot_len = sizeof(struct tcphdr);
870 __be32 mrst = 0, *topt;
871 struct dst_entry *dst;
875 tot_len += TCPOLEN_TSTAMP_ALIGNED;
876 #ifdef CONFIG_TCP_MD5SIG
878 tot_len += TCPOLEN_MD5SIG_ALIGNED;
883 mrst = mptcp_reset_option(skb);
886 tot_len += sizeof(__be32);
890 buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
894 skb_reserve(buff, MAX_TCP_HEADER);
896 t1 = skb_push(buff, tot_len);
897 skb_reset_transport_header(buff);
899 /* Swap the send and the receive. */
900 memset(t1, 0, sizeof(*t1));
901 t1->dest = th->source;
902 t1->source = th->dest;
903 t1->doff = tot_len / 4;
904 t1->seq = htonl(seq);
905 t1->ack_seq = htonl(ack);
906 t1->ack = !rst || !th->ack;
908 t1->window = htons(win);
910 topt = (__be32 *)(t1 + 1);
913 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
914 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
915 *topt++ = htonl(tsval);
916 *topt++ = htonl(tsecr);
922 #ifdef CONFIG_TCP_MD5SIG
924 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
925 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
926 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
927 &ipv6_hdr(skb)->saddr,
928 &ipv6_hdr(skb)->daddr, t1);
932 memset(&fl6, 0, sizeof(fl6));
933 fl6.daddr = ipv6_hdr(skb)->saddr;
934 fl6.saddr = ipv6_hdr(skb)->daddr;
935 fl6.flowlabel = label;
937 buff->ip_summed = CHECKSUM_PARTIAL;
939 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
941 fl6.flowi6_proto = IPPROTO_TCP;
942 if (rt6_need_strict(&fl6.daddr) && !oif)
943 fl6.flowi6_oif = tcp_v6_iif(skb);
945 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
948 fl6.flowi6_oif = oif;
952 if (sk->sk_state == TCP_TIME_WAIT)
953 mark = inet_twsk(sk)->tw_mark;
956 skb_set_delivery_time(buff, tcp_transmit_time(sk), true);
959 /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */
960 skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4);
962 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
963 fl6.fl6_dport = t1->dest;
964 fl6.fl6_sport = t1->source;
965 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
966 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
968 /* Pass a socket to ip6_dst_lookup either it is for RST
969 * Underlying function will use this to retrieve the network
972 if (sk && sk->sk_state != TCP_TIME_WAIT)
973 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/
975 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL);
977 skb_dst_set(buff, dst);
978 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
979 tclass & ~INET_ECN_MASK, priority);
980 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
982 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
989 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
991 const struct tcphdr *th = tcp_hdr(skb);
992 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
993 u32 seq = 0, ack_seq = 0;
994 struct tcp_md5sig_key *key = NULL;
995 #ifdef CONFIG_TCP_MD5SIG
996 const __u8 *hash_location = NULL;
997 unsigned char newhash[16];
999 struct sock *sk1 = NULL;
1009 /* If sk not NULL, it means we did a successful lookup and incoming
1010 * route had to be correct. prequeue might have dropped our dst.
1012 if (!sk && !ipv6_unicast_destination(skb))
1015 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1016 #ifdef CONFIG_TCP_MD5SIG
1018 hash_location = tcp_parse_md5sig_option(th);
1019 if (sk && sk_fullsock(sk)) {
1022 /* sdif set, means packet ingressed via a device
1023 * in an L3 domain and inet_iif is set to it.
1025 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1026 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1027 } else if (hash_location) {
1028 int dif = tcp_v6_iif_l3_slave(skb);
1029 int sdif = tcp_v6_sdif(skb);
1033 * active side is lost. Try to find listening socket through
1034 * source port, and then find md5 key through listening socket.
1035 * we are not loose security here:
1036 * Incoming packet is checked with md5 hash with finding key,
1037 * no RST generated if md5 hash doesn't match.
1039 sk1 = inet6_lookup_listener(net,
1040 &tcp_hashinfo, NULL, 0,
1042 th->source, &ipv6h->daddr,
1043 ntohs(th->source), dif, sdif);
1047 /* sdif set, means packet ingressed via a device
1048 * in an L3 domain and dif is set to it.
1050 l3index = tcp_v6_sdif(skb) ? dif : 0;
1052 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1056 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1057 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1063 seq = ntohl(th->ack_seq);
1065 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1069 oif = sk->sk_bound_dev_if;
1070 if (sk_fullsock(sk)) {
1071 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1073 trace_tcp_send_reset(sk, skb);
1075 label = ip6_flowlabel(ipv6h);
1076 priority = sk->sk_priority;
1078 if (sk->sk_state == TCP_TIME_WAIT) {
1079 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1080 priority = inet_twsk(sk)->tw_priority;
1083 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1084 label = ip6_flowlabel(ipv6h);
1087 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1,
1088 ipv6_get_dsfield(ipv6h), label, priority, 0);
1090 #ifdef CONFIG_TCP_MD5SIG
1096 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1097 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1098 struct tcp_md5sig_key *key, u8 tclass,
1099 __be32 label, u32 priority, u32 txhash)
1101 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1102 tclass, label, priority, txhash);
1105 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1107 struct inet_timewait_sock *tw = inet_twsk(sk);
1108 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1110 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1111 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1112 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1113 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1114 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority,
1120 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1121 struct request_sock *req)
1125 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1127 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1128 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1131 * The window field (SEG.WND) of every outgoing segment, with the
1132 * exception of <SYN> segments, MUST be right-shifted by
1133 * Rcv.Wind.Shift bits:
1135 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1136 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1137 tcp_rsk(req)->rcv_nxt,
1138 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1139 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1140 req->ts_recent, sk->sk_bound_dev_if,
1141 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1142 ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority,
1143 tcp_rsk(req)->txhash);
1147 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1149 #ifdef CONFIG_SYN_COOKIES
1150 const struct tcphdr *th = tcp_hdr(skb);
1153 sk = cookie_v6_check(sk, skb);
1158 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1159 struct tcphdr *th, u32 *cookie)
1162 #ifdef CONFIG_SYN_COOKIES
1163 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1164 &tcp_request_sock_ipv6_ops, sk, th);
1166 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1167 tcp_synq_overflow(sk);
1173 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1175 if (skb->protocol == htons(ETH_P_IP))
1176 return tcp_v4_conn_request(sk, skb);
1178 if (!ipv6_unicast_destination(skb))
1181 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1182 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1186 return tcp_conn_request(&tcp6_request_sock_ops,
1187 &tcp_request_sock_ipv6_ops, sk, skb);
1191 return 0; /* don't send reset */
1194 static void tcp_v6_restore_cb(struct sk_buff *skb)
1196 /* We need to move header back to the beginning if xfrm6_policy_check()
1197 * and tcp_v6_fill_cb() are going to be called again.
1198 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1200 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1201 sizeof(struct inet6_skb_parm));
1204 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1205 struct request_sock *req,
1206 struct dst_entry *dst,
1207 struct request_sock *req_unhash,
1210 struct inet_request_sock *ireq;
1211 struct ipv6_pinfo *newnp;
1212 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1213 struct ipv6_txoptions *opt;
1214 struct inet_sock *newinet;
1215 bool found_dup_sk = false;
1216 struct tcp_sock *newtp;
1218 #ifdef CONFIG_TCP_MD5SIG
1219 struct tcp_md5sig_key *key;
1224 if (skb->protocol == htons(ETH_P_IP)) {
1229 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1230 req_unhash, own_req);
1235 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1237 newnp = tcp_inet6_sk(newsk);
1238 newtp = tcp_sk(newsk);
1240 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1242 newnp->saddr = newsk->sk_v6_rcv_saddr;
1244 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1245 if (sk_is_mptcp(newsk))
1246 mptcpv6_handle_mapped(newsk, true);
1247 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1248 #ifdef CONFIG_TCP_MD5SIG
1249 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1252 newnp->ipv6_mc_list = NULL;
1253 newnp->ipv6_ac_list = NULL;
1254 newnp->ipv6_fl_list = NULL;
1255 newnp->pktoptions = NULL;
1257 newnp->mcast_oif = inet_iif(skb);
1258 newnp->mcast_hops = ip_hdr(skb)->ttl;
1259 newnp->rcv_flowinfo = 0;
1261 newnp->flow_label = 0;
1264 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1265 * here, tcp_create_openreq_child now does this for us, see the comment in
1266 * that function for the gory details. -acme
1269 /* It is tricky place. Until this moment IPv4 tcp
1270 worked with IPv6 icsk.icsk_af_ops.
1273 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1278 ireq = inet_rsk(req);
1280 if (sk_acceptq_is_full(sk))
1284 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1289 newsk = tcp_create_openreq_child(sk, req, skb);
1294 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1295 * count here, tcp_create_openreq_child now does this for us, see the
1296 * comment in that function for the gory details. -acme
1299 newsk->sk_gso_type = SKB_GSO_TCPV6;
1300 ip6_dst_store(newsk, dst, NULL, NULL);
1301 inet6_sk_rx_dst_set(newsk, skb);
1303 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1305 newtp = tcp_sk(newsk);
1306 newinet = inet_sk(newsk);
1307 newnp = tcp_inet6_sk(newsk);
1309 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1311 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1312 newnp->saddr = ireq->ir_v6_loc_addr;
1313 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1314 newsk->sk_bound_dev_if = ireq->ir_iif;
1316 /* Now IPv6 options...
1318 First: no IPv4 options.
1320 newinet->inet_opt = NULL;
1321 newnp->ipv6_mc_list = NULL;
1322 newnp->ipv6_ac_list = NULL;
1323 newnp->ipv6_fl_list = NULL;
1326 newnp->rxopt.all = np->rxopt.all;
1328 newnp->pktoptions = NULL;
1330 newnp->mcast_oif = tcp_v6_iif(skb);
1331 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1332 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1334 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1336 /* Set ToS of the new socket based upon the value of incoming SYN.
1337 * ECT bits are set later in tcp_init_transfer().
1339 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1340 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1342 /* Clone native IPv6 options from listening socket (if any)
1344 Yes, keeping reference count would be much more clever,
1345 but we make one more one thing there: reattach optmem
1348 opt = ireq->ipv6_opt;
1350 opt = rcu_dereference(np->opt);
1352 opt = ipv6_dup_options(newsk, opt);
1353 RCU_INIT_POINTER(newnp->opt, opt);
1355 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1357 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1360 tcp_ca_openreq_child(newsk, dst);
1362 tcp_sync_mss(newsk, dst_mtu(dst));
1363 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1365 tcp_initialize_rcv_mss(newsk);
1367 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1368 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1370 #ifdef CONFIG_TCP_MD5SIG
1371 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1373 /* Copy over the MD5 key from the original socket */
1374 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1376 /* We're using one, so create a matching key
1377 * on the newsk structure. If we fail to get
1378 * memory, then we end up not copying the key
1381 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1382 AF_INET6, 128, l3index, key->flags, key->key, key->keylen,
1383 sk_gfp_mask(sk, GFP_ATOMIC));
1387 if (__inet_inherit_port(sk, newsk) < 0) {
1388 inet_csk_prepare_forced_close(newsk);
1392 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1395 tcp_move_syn(newtp, req);
1397 /* Clone pktoptions received with SYN, if we own the req */
1398 if (ireq->pktopts) {
1399 newnp->pktoptions = skb_clone(ireq->pktopts,
1400 sk_gfp_mask(sk, GFP_ATOMIC));
1401 consume_skb(ireq->pktopts);
1402 ireq->pktopts = NULL;
1403 if (newnp->pktoptions) {
1404 tcp_v6_restore_cb(newnp->pktoptions);
1405 skb_set_owner_r(newnp->pktoptions, newsk);
1409 if (!req_unhash && found_dup_sk) {
1410 /* This code path should only be executed in the
1411 * syncookie case only
1413 bh_unlock_sock(newsk);
1422 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1430 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1432 /* The socket must have it's spinlock held when we get
1433 * here, unless it is a TCP_LISTEN socket.
1435 * We have a potential double-lock case here, so even when
1436 * doing backlog processing we use the BH locking scheme.
1437 * This is because we cannot sleep with the original spinlock
1440 INDIRECT_CALLABLE_SCOPE
1441 int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1443 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1444 struct sk_buff *opt_skb = NULL;
1445 enum skb_drop_reason reason;
1446 struct tcp_sock *tp;
1448 /* Imagine: socket is IPv6. IPv4 packet arrives,
1449 goes to IPv4 receive handler and backlogged.
1450 From backlog it always goes here. Kerboom...
1451 Fortunately, tcp_rcv_established and rcv_established
1452 handle them correctly, but it is not case with
1453 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1456 if (skb->protocol == htons(ETH_P_IP))
1457 return tcp_v4_do_rcv(sk, skb);
1460 * socket locking is here for SMP purposes as backlog rcv
1461 * is currently called with bh processing disabled.
1464 /* Do Stevens' IPV6_PKTOPTIONS.
1466 Yes, guys, it is the only place in our code, where we
1467 may make it not affecting IPv4.
1468 The rest of code is protocol independent,
1469 and I do not like idea to uglify IPv4.
1471 Actually, all the idea behind IPV6_PKTOPTIONS
1472 looks not very well thought. For now we latch
1473 options, received in the last packet, enqueued
1474 by tcp. Feel free to propose better solution.
1478 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1480 reason = SKB_DROP_REASON_NOT_SPECIFIED;
1481 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1482 struct dst_entry *dst;
1484 dst = rcu_dereference_protected(sk->sk_rx_dst,
1485 lockdep_sock_is_held(sk));
1487 sock_rps_save_rxhash(sk, skb);
1488 sk_mark_napi_id(sk, skb);
1490 if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1491 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1492 dst, sk->sk_rx_dst_cookie) == NULL) {
1493 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1498 tcp_rcv_established(sk, skb);
1500 goto ipv6_pktoptions;
1504 if (tcp_checksum_complete(skb))
1507 if (sk->sk_state == TCP_LISTEN) {
1508 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1514 if (tcp_child_process(sk, nsk, skb))
1517 __kfree_skb(opt_skb);
1521 sock_rps_save_rxhash(sk, skb);
1523 if (tcp_rcv_state_process(sk, skb))
1526 goto ipv6_pktoptions;
1530 tcp_v6_send_reset(sk, skb);
1533 __kfree_skb(opt_skb);
1534 kfree_skb_reason(skb, reason);
1537 reason = SKB_DROP_REASON_TCP_CSUM;
1538 trace_tcp_bad_csum(skb);
1539 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1540 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1545 /* Do you ask, what is it?
1547 1. skb was enqueued by tcp.
1548 2. skb is added to tail of read queue, rather than out of order.
1549 3. socket is not in passive state.
1550 4. Finally, it really contains options, which user wants to receive.
1553 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1554 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1555 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1556 np->mcast_oif = tcp_v6_iif(opt_skb);
1557 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1558 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1559 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1560 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1562 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1563 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1564 skb_set_owner_r(opt_skb, sk);
1565 tcp_v6_restore_cb(opt_skb);
1566 opt_skb = xchg(&np->pktoptions, opt_skb);
1568 __kfree_skb(opt_skb);
1569 opt_skb = xchg(&np->pktoptions, NULL);
1573 consume_skb(opt_skb);
1577 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1578 const struct tcphdr *th)
1580 /* This is tricky: we move IP6CB at its correct location into
1581 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1582 * _decode_session6() uses IP6CB().
1583 * barrier() makes sure compiler won't play aliasing games.
1585 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1586 sizeof(struct inet6_skb_parm));
1589 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1590 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1591 skb->len - th->doff*4);
1592 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1593 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1594 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1595 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1596 TCP_SKB_CB(skb)->sacked = 0;
1597 TCP_SKB_CB(skb)->has_rxtstamp =
1598 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1601 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1603 enum skb_drop_reason drop_reason;
1604 int sdif = inet6_sdif(skb);
1605 int dif = inet6_iif(skb);
1606 const struct tcphdr *th;
1607 const struct ipv6hdr *hdr;
1611 struct net *net = dev_net(skb->dev);
1613 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1614 if (skb->pkt_type != PACKET_HOST)
1618 * Count it even if it's bad.
1620 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1622 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1625 th = (const struct tcphdr *)skb->data;
1627 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
1628 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1631 if (!pskb_may_pull(skb, th->doff*4))
1634 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1637 th = (const struct tcphdr *)skb->data;
1638 hdr = ipv6_hdr(skb);
1641 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1642 th->source, th->dest, inet6_iif(skb), sdif,
1648 if (sk->sk_state == TCP_TIME_WAIT)
1651 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1652 struct request_sock *req = inet_reqsk(sk);
1653 bool req_stolen = false;
1656 sk = req->rsk_listener;
1657 drop_reason = tcp_inbound_md5_hash(sk, skb,
1658 &hdr->saddr, &hdr->daddr,
1659 AF_INET6, dif, sdif);
1661 sk_drops_add(sk, skb);
1665 if (tcp_checksum_complete(skb)) {
1669 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1670 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1672 inet_csk_reqsk_queue_drop_and_put(sk, req);
1676 /* reuseport_migrate_sock() has already held one sk_refcnt
1684 if (!tcp_filter(sk, skb)) {
1685 th = (const struct tcphdr *)skb->data;
1686 hdr = ipv6_hdr(skb);
1687 tcp_v6_fill_cb(skb, hdr, th);
1688 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1690 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1695 /* Another cpu got exclusive access to req
1696 * and created a full blown socket.
1697 * Try to feed this packet to this socket
1698 * instead of discarding it.
1700 tcp_v6_restore_cb(skb);
1704 goto discard_and_relse;
1708 tcp_v6_restore_cb(skb);
1709 } else if (tcp_child_process(sk, nsk, skb)) {
1710 tcp_v6_send_reset(nsk, skb);
1711 goto discard_and_relse;
1718 if (static_branch_unlikely(&ip6_min_hopcount)) {
1719 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
1720 if (hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
1721 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1722 goto discard_and_relse;
1726 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
1727 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1728 goto discard_and_relse;
1731 drop_reason = tcp_inbound_md5_hash(sk, skb, &hdr->saddr, &hdr->daddr,
1732 AF_INET6, dif, sdif);
1734 goto discard_and_relse;
1736 if (tcp_filter(sk, skb)) {
1737 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1738 goto discard_and_relse;
1740 th = (const struct tcphdr *)skb->data;
1741 hdr = ipv6_hdr(skb);
1742 tcp_v6_fill_cb(skb, hdr, th);
1746 if (sk->sk_state == TCP_LISTEN) {
1747 ret = tcp_v6_do_rcv(sk, skb);
1748 goto put_and_return;
1751 sk_incoming_cpu_update(sk);
1753 bh_lock_sock_nested(sk);
1754 tcp_segs_in(tcp_sk(sk), skb);
1756 if (!sock_owned_by_user(sk)) {
1757 ret = tcp_v6_do_rcv(sk, skb);
1759 if (tcp_add_backlog(sk, skb, &drop_reason))
1760 goto discard_and_relse;
1766 return ret ? -1 : 0;
1769 drop_reason = SKB_DROP_REASON_NO_SOCKET;
1770 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1773 tcp_v6_fill_cb(skb, hdr, th);
1775 if (tcp_checksum_complete(skb)) {
1777 drop_reason = SKB_DROP_REASON_TCP_CSUM;
1778 trace_tcp_bad_csum(skb);
1779 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1781 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1783 tcp_v6_send_reset(NULL, skb);
1787 SKB_DR_OR(drop_reason, NOT_SPECIFIED);
1788 kfree_skb_reason(skb, drop_reason);
1792 sk_drops_add(sk, skb);
1798 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1799 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1800 inet_twsk_put(inet_twsk(sk));
1804 tcp_v6_fill_cb(skb, hdr, th);
1806 if (tcp_checksum_complete(skb)) {
1807 inet_twsk_put(inet_twsk(sk));
1811 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1816 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1817 skb, __tcp_hdrlen(th),
1818 &ipv6_hdr(skb)->saddr, th->source,
1819 &ipv6_hdr(skb)->daddr,
1821 tcp_v6_iif_l3_slave(skb),
1824 struct inet_timewait_sock *tw = inet_twsk(sk);
1825 inet_twsk_deschedule_put(tw);
1827 tcp_v6_restore_cb(skb);
1835 tcp_v6_timewait_ack(sk, skb);
1838 tcp_v6_send_reset(sk, skb);
1839 inet_twsk_deschedule_put(inet_twsk(sk));
1841 case TCP_TW_SUCCESS:
1847 void tcp_v6_early_demux(struct sk_buff *skb)
1849 const struct ipv6hdr *hdr;
1850 const struct tcphdr *th;
1853 if (skb->pkt_type != PACKET_HOST)
1856 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1859 hdr = ipv6_hdr(skb);
1862 if (th->doff < sizeof(struct tcphdr) / 4)
1865 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1866 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1867 &hdr->saddr, th->source,
1868 &hdr->daddr, ntohs(th->dest),
1869 inet6_iif(skb), inet6_sdif(skb));
1872 skb->destructor = sock_edemux;
1873 if (sk_fullsock(sk)) {
1874 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1877 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1879 sk->sk_rx_dst_ifindex == skb->skb_iif)
1880 skb_dst_set_noref(skb, dst);
1885 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1886 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1887 .twsk_unique = tcp_twsk_unique,
1888 .twsk_destructor = tcp_twsk_destructor,
1891 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1893 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr);
1896 const struct inet_connection_sock_af_ops ipv6_specific = {
1897 .queue_xmit = inet6_csk_xmit,
1898 .send_check = tcp_v6_send_check,
1899 .rebuild_header = inet6_sk_rebuild_header,
1900 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1901 .conn_request = tcp_v6_conn_request,
1902 .syn_recv_sock = tcp_v6_syn_recv_sock,
1903 .net_header_len = sizeof(struct ipv6hdr),
1904 .net_frag_header_len = sizeof(struct frag_hdr),
1905 .setsockopt = ipv6_setsockopt,
1906 .getsockopt = ipv6_getsockopt,
1907 .addr2sockaddr = inet6_csk_addr2sockaddr,
1908 .sockaddr_len = sizeof(struct sockaddr_in6),
1909 .mtu_reduced = tcp_v6_mtu_reduced,
1912 #ifdef CONFIG_TCP_MD5SIG
1913 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1914 .md5_lookup = tcp_v6_md5_lookup,
1915 .calc_md5_hash = tcp_v6_md5_hash_skb,
1916 .md5_parse = tcp_v6_parse_md5_keys,
1921 * TCP over IPv4 via INET6 API
1923 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1924 .queue_xmit = ip_queue_xmit,
1925 .send_check = tcp_v4_send_check,
1926 .rebuild_header = inet_sk_rebuild_header,
1927 .sk_rx_dst_set = inet_sk_rx_dst_set,
1928 .conn_request = tcp_v6_conn_request,
1929 .syn_recv_sock = tcp_v6_syn_recv_sock,
1930 .net_header_len = sizeof(struct iphdr),
1931 .setsockopt = ipv6_setsockopt,
1932 .getsockopt = ipv6_getsockopt,
1933 .addr2sockaddr = inet6_csk_addr2sockaddr,
1934 .sockaddr_len = sizeof(struct sockaddr_in6),
1935 .mtu_reduced = tcp_v4_mtu_reduced,
1938 #ifdef CONFIG_TCP_MD5SIG
1939 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1940 .md5_lookup = tcp_v4_md5_lookup,
1941 .calc_md5_hash = tcp_v4_md5_hash_skb,
1942 .md5_parse = tcp_v6_parse_md5_keys,
1946 /* NOTE: A lot of things set to zero explicitly by call to
1947 * sk_alloc() so need not be done here.
1949 static int tcp_v6_init_sock(struct sock *sk)
1951 struct inet_connection_sock *icsk = inet_csk(sk);
1955 icsk->icsk_af_ops = &ipv6_specific;
1957 #ifdef CONFIG_TCP_MD5SIG
1958 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1964 static void tcp_v6_destroy_sock(struct sock *sk)
1966 tcp_v4_destroy_sock(sk);
1967 inet6_destroy_sock(sk);
1970 #ifdef CONFIG_PROC_FS
1971 /* Proc filesystem TCPv6 sock list dumping. */
1972 static void get_openreq6(struct seq_file *seq,
1973 const struct request_sock *req, int i)
1975 long ttd = req->rsk_timer.expires - jiffies;
1976 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1977 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1983 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1984 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1986 src->s6_addr32[0], src->s6_addr32[1],
1987 src->s6_addr32[2], src->s6_addr32[3],
1988 inet_rsk(req)->ir_num,
1989 dest->s6_addr32[0], dest->s6_addr32[1],
1990 dest->s6_addr32[2], dest->s6_addr32[3],
1991 ntohs(inet_rsk(req)->ir_rmt_port),
1993 0, 0, /* could print option size, but that is af dependent. */
1994 1, /* timers active (only the expire timer) */
1995 jiffies_to_clock_t(ttd),
1997 from_kuid_munged(seq_user_ns(seq),
1998 sock_i_uid(req->rsk_listener)),
1999 0, /* non standard timer */
2000 0, /* open_requests have no inode */
2004 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2006 const struct in6_addr *dest, *src;
2009 unsigned long timer_expires;
2010 const struct inet_sock *inet = inet_sk(sp);
2011 const struct tcp_sock *tp = tcp_sk(sp);
2012 const struct inet_connection_sock *icsk = inet_csk(sp);
2013 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2017 dest = &sp->sk_v6_daddr;
2018 src = &sp->sk_v6_rcv_saddr;
2019 destp = ntohs(inet->inet_dport);
2020 srcp = ntohs(inet->inet_sport);
2022 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2023 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2024 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2026 timer_expires = icsk->icsk_timeout;
2027 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2029 timer_expires = icsk->icsk_timeout;
2030 } else if (timer_pending(&sp->sk_timer)) {
2032 timer_expires = sp->sk_timer.expires;
2035 timer_expires = jiffies;
2038 state = inet_sk_state_load(sp);
2039 if (state == TCP_LISTEN)
2040 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2042 /* Because we don't lock the socket,
2043 * we might find a transient negative value.
2045 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2046 READ_ONCE(tp->copied_seq), 0);
2049 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2050 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2052 src->s6_addr32[0], src->s6_addr32[1],
2053 src->s6_addr32[2], src->s6_addr32[3], srcp,
2054 dest->s6_addr32[0], dest->s6_addr32[1],
2055 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2057 READ_ONCE(tp->write_seq) - tp->snd_una,
2060 jiffies_delta_to_clock_t(timer_expires - jiffies),
2061 icsk->icsk_retransmits,
2062 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2063 icsk->icsk_probes_out,
2065 refcount_read(&sp->sk_refcnt), sp,
2066 jiffies_to_clock_t(icsk->icsk_rto),
2067 jiffies_to_clock_t(icsk->icsk_ack.ato),
2068 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2070 state == TCP_LISTEN ?
2071 fastopenq->max_qlen :
2072 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2076 static void get_timewait6_sock(struct seq_file *seq,
2077 struct inet_timewait_sock *tw, int i)
2079 long delta = tw->tw_timer.expires - jiffies;
2080 const struct in6_addr *dest, *src;
2083 dest = &tw->tw_v6_daddr;
2084 src = &tw->tw_v6_rcv_saddr;
2085 destp = ntohs(tw->tw_dport);
2086 srcp = ntohs(tw->tw_sport);
2089 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2090 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2092 src->s6_addr32[0], src->s6_addr32[1],
2093 src->s6_addr32[2], src->s6_addr32[3], srcp,
2094 dest->s6_addr32[0], dest->s6_addr32[1],
2095 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2096 tw->tw_substate, 0, 0,
2097 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2098 refcount_read(&tw->tw_refcnt), tw);
2101 static int tcp6_seq_show(struct seq_file *seq, void *v)
2103 struct tcp_iter_state *st;
2104 struct sock *sk = v;
2106 if (v == SEQ_START_TOKEN) {
2111 "st tx_queue rx_queue tr tm->when retrnsmt"
2112 " uid timeout inode\n");
2117 if (sk->sk_state == TCP_TIME_WAIT)
2118 get_timewait6_sock(seq, v, st->num);
2119 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2120 get_openreq6(seq, v, st->num);
2122 get_tcp6_sock(seq, v, st->num);
2127 static const struct seq_operations tcp6_seq_ops = {
2128 .show = tcp6_seq_show,
2129 .start = tcp_seq_start,
2130 .next = tcp_seq_next,
2131 .stop = tcp_seq_stop,
2134 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2138 int __net_init tcp6_proc_init(struct net *net)
2140 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2141 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2146 void tcp6_proc_exit(struct net *net)
2148 remove_proc_entry("tcp6", net->proc_net);
2152 struct proto tcpv6_prot = {
2154 .owner = THIS_MODULE,
2156 .pre_connect = tcp_v6_pre_connect,
2157 .connect = tcp_v6_connect,
2158 .disconnect = tcp_disconnect,
2159 .accept = inet_csk_accept,
2161 .init = tcp_v6_init_sock,
2162 .destroy = tcp_v6_destroy_sock,
2163 .shutdown = tcp_shutdown,
2164 .setsockopt = tcp_setsockopt,
2165 .getsockopt = tcp_getsockopt,
2166 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
2167 .keepalive = tcp_set_keepalive,
2168 .recvmsg = tcp_recvmsg,
2169 .sendmsg = tcp_sendmsg,
2170 .sendpage = tcp_sendpage,
2171 .backlog_rcv = tcp_v6_do_rcv,
2172 .release_cb = tcp_release_cb,
2174 .unhash = inet_unhash,
2175 .get_port = inet_csk_get_port,
2176 .put_port = inet_put_port,
2177 #ifdef CONFIG_BPF_SYSCALL
2178 .psock_update_sk_prot = tcp_bpf_update_proto,
2180 .enter_memory_pressure = tcp_enter_memory_pressure,
2181 .leave_memory_pressure = tcp_leave_memory_pressure,
2182 .stream_memory_free = tcp_stream_memory_free,
2183 .sockets_allocated = &tcp_sockets_allocated,
2185 .memory_allocated = &tcp_memory_allocated,
2186 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
2188 .memory_pressure = &tcp_memory_pressure,
2189 .orphan_count = &tcp_orphan_count,
2190 .sysctl_mem = sysctl_tcp_mem,
2191 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2192 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2193 .max_header = MAX_TCP_HEADER,
2194 .obj_size = sizeof(struct tcp6_sock),
2195 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2196 .twsk_prot = &tcp6_timewait_sock_ops,
2197 .rsk_prot = &tcp6_request_sock_ops,
2198 .h.hashinfo = &tcp_hashinfo,
2199 .no_autobind = true,
2200 .diag_destroy = tcp_abort,
2202 EXPORT_SYMBOL_GPL(tcpv6_prot);
2204 static const struct inet6_protocol tcpv6_protocol = {
2205 .handler = tcp_v6_rcv,
2206 .err_handler = tcp_v6_err,
2207 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2210 static struct inet_protosw tcpv6_protosw = {
2211 .type = SOCK_STREAM,
2212 .protocol = IPPROTO_TCP,
2213 .prot = &tcpv6_prot,
2214 .ops = &inet6_stream_ops,
2215 .flags = INET_PROTOSW_PERMANENT |
2219 static int __net_init tcpv6_net_init(struct net *net)
2221 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2222 SOCK_RAW, IPPROTO_TCP, net);
2225 static void __net_exit tcpv6_net_exit(struct net *net)
2227 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2230 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2232 inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2235 static struct pernet_operations tcpv6_net_ops = {
2236 .init = tcpv6_net_init,
2237 .exit = tcpv6_net_exit,
2238 .exit_batch = tcpv6_net_exit_batch,
2241 int __init tcpv6_init(void)
2245 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2249 /* register inet6 protocol */
2250 ret = inet6_register_protosw(&tcpv6_protosw);
2252 goto out_tcpv6_protocol;
2254 ret = register_pernet_subsys(&tcpv6_net_ops);
2256 goto out_tcpv6_protosw;
2258 ret = mptcpv6_init();
2260 goto out_tcpv6_pernet_subsys;
2265 out_tcpv6_pernet_subsys:
2266 unregister_pernet_subsys(&tcpv6_net_ops);
2268 inet6_unregister_protosw(&tcpv6_protosw);
2270 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2274 void tcpv6_exit(void)
2276 unregister_pernet_subsys(&tcpv6_net_ops);
2277 inet6_unregister_protosw(&tcpv6_protosw);
2278 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);