3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
14 * Hideaki YOSHIFUJI : sin6_scope_id support
15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
17 * a single port at the same time.
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
26 #include <linux/bottom_half.h>
27 #include <linux/module.h>
28 #include <linux/errno.h>
29 #include <linux/types.h>
30 #include <linux/socket.h>
31 #include <linux/sockios.h>
32 #include <linux/net.h>
33 #include <linux/jiffies.h>
35 #include <linux/in6.h>
36 #include <linux/netdevice.h>
37 #include <linux/init.h>
38 #include <linux/jhash.h>
39 #include <linux/ipsec.h>
40 #include <linux/times.h>
41 #include <linux/slab.h>
42 #include <linux/uaccess.h>
43 #include <linux/ipv6.h>
44 #include <linux/icmpv6.h>
45 #include <linux/random.h>
48 #include <net/ndisc.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/inet6_connection_sock.h>
52 #include <net/transp_v6.h>
53 #include <net/addrconf.h>
54 #include <net/ip6_route.h>
55 #include <net/ip6_checksum.h>
56 #include <net/inet_ecn.h>
57 #include <net/protocol.h>
60 #include <net/dsfield.h>
61 #include <net/timewait_sock.h>
62 #include <net/inet_common.h>
63 #include <net/secure_seq.h>
64 #include <net/tcp_memcontrol.h>
65 #include <net/busy_poll.h>
67 #include <linux/proc_fs.h>
68 #include <linux/seq_file.h>
70 #include <linux/crypto.h>
71 #include <linux/scatterlist.h>
73 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
74 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
75 struct request_sock *req);
77 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
79 static const struct inet_connection_sock_af_ops ipv6_mapped;
80 static const struct inet_connection_sock_af_ops ipv6_specific;
81 #ifdef CONFIG_TCP_MD5SIG
82 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
83 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
85 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
86 const struct in6_addr *addr)
92 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
94 struct dst_entry *dst = skb_dst(skb);
96 if (dst && dst_hold_safe(dst)) {
97 const struct rt6_info *rt = (const struct rt6_info *)dst;
100 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
101 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
105 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
107 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
108 ipv6_hdr(skb)->saddr.s6_addr32,
110 tcp_hdr(skb)->source);
113 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
116 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
117 struct inet_sock *inet = inet_sk(sk);
118 struct inet_connection_sock *icsk = inet_csk(sk);
119 struct ipv6_pinfo *np = inet6_sk(sk);
120 struct tcp_sock *tp = tcp_sk(sk);
121 struct in6_addr *saddr = NULL, *final_p, final;
122 struct ipv6_txoptions *opt;
124 struct dst_entry *dst;
128 if (addr_len < SIN6_LEN_RFC2133)
131 if (usin->sin6_family != AF_INET6)
132 return -EAFNOSUPPORT;
134 memset(&fl6, 0, sizeof(fl6));
137 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
138 IP6_ECN_flow_init(fl6.flowlabel);
139 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
140 struct ip6_flowlabel *flowlabel;
141 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
144 fl6_sock_release(flowlabel);
149 * connect() to INADDR_ANY means loopback (BSD'ism).
152 if (ipv6_addr_any(&usin->sin6_addr))
153 usin->sin6_addr.s6_addr[15] = 0x1;
155 addr_type = ipv6_addr_type(&usin->sin6_addr);
157 if (addr_type & IPV6_ADDR_MULTICAST)
160 if (addr_type&IPV6_ADDR_LINKLOCAL) {
161 if (addr_len >= sizeof(struct sockaddr_in6) &&
162 usin->sin6_scope_id) {
163 /* If interface is set while binding, indices
166 if (sk->sk_bound_dev_if &&
167 sk->sk_bound_dev_if != usin->sin6_scope_id)
170 sk->sk_bound_dev_if = usin->sin6_scope_id;
173 /* Connect to link-local address requires an interface */
174 if (!sk->sk_bound_dev_if)
178 if (tp->rx_opt.ts_recent_stamp &&
179 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
180 tp->rx_opt.ts_recent = 0;
181 tp->rx_opt.ts_recent_stamp = 0;
185 sk->sk_v6_daddr = usin->sin6_addr;
186 np->flow_label = fl6.flowlabel;
192 if (addr_type == IPV6_ADDR_MAPPED) {
193 u32 exthdrlen = icsk->icsk_ext_hdr_len;
194 struct sockaddr_in sin;
196 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
198 if (__ipv6_only_sock(sk))
201 sin.sin_family = AF_INET;
202 sin.sin_port = usin->sin6_port;
203 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
205 icsk->icsk_af_ops = &ipv6_mapped;
206 sk->sk_backlog_rcv = tcp_v4_do_rcv;
207 #ifdef CONFIG_TCP_MD5SIG
208 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
211 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
214 icsk->icsk_ext_hdr_len = exthdrlen;
215 icsk->icsk_af_ops = &ipv6_specific;
216 sk->sk_backlog_rcv = tcp_v6_do_rcv;
217 #ifdef CONFIG_TCP_MD5SIG
218 tp->af_specific = &tcp_sock_ipv6_specific;
222 np->saddr = sk->sk_v6_rcv_saddr;
227 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
228 saddr = &sk->sk_v6_rcv_saddr;
230 fl6.flowi6_proto = IPPROTO_TCP;
231 fl6.daddr = sk->sk_v6_daddr;
232 fl6.saddr = saddr ? *saddr : np->saddr;
233 fl6.flowi6_oif = sk->sk_bound_dev_if;
234 fl6.flowi6_mark = sk->sk_mark;
235 fl6.fl6_dport = usin->sin6_port;
236 fl6.fl6_sport = inet->inet_sport;
238 opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
239 final_p = fl6_update_dst(&fl6, opt, &final);
241 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
243 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
251 sk->sk_v6_rcv_saddr = *saddr;
254 /* set the source address */
256 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
258 sk->sk_gso_type = SKB_GSO_TCPV6;
259 ip6_dst_store(sk, dst, NULL, NULL);
261 if (tcp_death_row.sysctl_tw_recycle &&
262 !tp->rx_opt.ts_recent_stamp &&
263 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
264 tcp_fetch_timewait_stamp(sk, dst);
266 icsk->icsk_ext_hdr_len = 0;
268 icsk->icsk_ext_hdr_len = opt->opt_flen +
271 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
273 inet->inet_dport = usin->sin6_port;
275 tcp_set_state(sk, TCP_SYN_SENT);
276 err = inet6_hash_connect(&tcp_death_row, sk);
282 if (!tp->write_seq && likely(!tp->repair))
283 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
284 sk->sk_v6_daddr.s6_addr32,
288 err = tcp_connect(sk);
295 tcp_set_state(sk, TCP_CLOSE);
298 inet->inet_dport = 0;
299 sk->sk_route_caps = 0;
303 static void tcp_v6_mtu_reduced(struct sock *sk)
305 struct dst_entry *dst;
307 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
310 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
314 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
315 tcp_sync_mss(sk, dst_mtu(dst));
316 tcp_simple_retransmit(sk);
320 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
321 u8 type, u8 code, int offset, __be32 info)
323 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
324 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
325 struct net *net = dev_net(skb->dev);
326 struct request_sock *fastopen;
327 struct ipv6_pinfo *np;
334 sk = __inet6_lookup_established(net, &tcp_hashinfo,
335 &hdr->daddr, th->dest,
336 &hdr->saddr, ntohs(th->source),
340 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
345 if (sk->sk_state == TCP_TIME_WAIT) {
346 inet_twsk_put(inet_twsk(sk));
349 seq = ntohl(th->seq);
350 fatal = icmpv6_err_convert(type, code, &err);
351 if (sk->sk_state == TCP_NEW_SYN_RECV)
352 return tcp_req_err(sk, seq, fatal);
355 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
356 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
358 if (sk->sk_state == TCP_CLOSE)
361 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
362 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
367 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
368 fastopen = tp->fastopen_rsk;
369 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
370 if (sk->sk_state != TCP_LISTEN &&
371 !between(seq, snd_una, tp->snd_nxt)) {
372 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
378 if (type == NDISC_REDIRECT) {
379 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
382 dst->ops->redirect(dst, sk, skb);
386 if (type == ICMPV6_PKT_TOOBIG) {
387 /* We are not interested in TCP_LISTEN and open_requests
388 * (SYN-ACKs send out by Linux are always <576bytes so
389 * they should go through unfragmented).
391 if (sk->sk_state == TCP_LISTEN)
394 if (!ip6_sk_accept_pmtu(sk))
397 tp->mtu_info = ntohl(info);
398 if (!sock_owned_by_user(sk))
399 tcp_v6_mtu_reduced(sk);
400 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
407 /* Might be for an request_sock */
408 switch (sk->sk_state) {
411 /* Only in fast or simultaneous open. If a fast open socket is
412 * is already accepted it is treated as a connected one below.
414 if (fastopen && !fastopen->sk)
417 if (!sock_owned_by_user(sk)) {
419 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
423 sk->sk_err_soft = err;
427 if (!sock_owned_by_user(sk) && np->recverr) {
429 sk->sk_error_report(sk);
431 sk->sk_err_soft = err;
439 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
441 struct request_sock *req,
442 struct tcp_fastopen_cookie *foc,
445 struct inet_request_sock *ireq = inet_rsk(req);
446 struct ipv6_pinfo *np = inet6_sk(sk);
447 struct flowi6 *fl6 = &fl->u.ip6;
451 /* First, grab a route. */
452 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
453 IPPROTO_TCP)) == NULL)
456 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
459 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
460 &ireq->ir_v6_rmt_addr);
462 fl6->daddr = ireq->ir_v6_rmt_addr;
463 if (np->repflow && ireq->pktopts)
464 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
467 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
470 err = net_xmit_eval(err);
478 static void tcp_v6_reqsk_destructor(struct request_sock *req)
480 kfree_skb(inet_rsk(req)->pktopts);
483 #ifdef CONFIG_TCP_MD5SIG
484 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
485 const struct in6_addr *addr)
487 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
490 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
491 const struct sock *addr_sk)
493 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
496 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
499 struct tcp_md5sig cmd;
500 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
502 if (optlen < sizeof(cmd))
505 if (copy_from_user(&cmd, optval, sizeof(cmd)))
508 if (sin6->sin6_family != AF_INET6)
511 if (!cmd.tcpm_keylen) {
512 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
513 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
515 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
519 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
522 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
523 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
524 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
526 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
527 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
530 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
531 const struct in6_addr *daddr,
532 const struct in6_addr *saddr, int nbytes)
534 struct tcp6_pseudohdr *bp;
535 struct scatterlist sg;
537 bp = &hp->md5_blk.ip6;
538 /* 1. TCP pseudo-header (RFC2460) */
541 bp->protocol = cpu_to_be32(IPPROTO_TCP);
542 bp->len = cpu_to_be32(nbytes);
544 sg_init_one(&sg, bp, sizeof(*bp));
545 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
548 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
549 const struct in6_addr *daddr, struct in6_addr *saddr,
550 const struct tcphdr *th)
552 struct tcp_md5sig_pool *hp;
553 struct hash_desc *desc;
555 hp = tcp_get_md5sig_pool();
557 goto clear_hash_noput;
558 desc = &hp->md5_desc;
560 if (crypto_hash_init(desc))
562 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
564 if (tcp_md5_hash_header(hp, th))
566 if (tcp_md5_hash_key(hp, key))
568 if (crypto_hash_final(desc, md5_hash))
571 tcp_put_md5sig_pool();
575 tcp_put_md5sig_pool();
577 memset(md5_hash, 0, 16);
581 static int tcp_v6_md5_hash_skb(char *md5_hash,
582 const struct tcp_md5sig_key *key,
583 const struct sock *sk,
584 const struct sk_buff *skb)
586 const struct in6_addr *saddr, *daddr;
587 struct tcp_md5sig_pool *hp;
588 struct hash_desc *desc;
589 const struct tcphdr *th = tcp_hdr(skb);
591 if (sk) { /* valid for establish/request sockets */
592 saddr = &sk->sk_v6_rcv_saddr;
593 daddr = &sk->sk_v6_daddr;
595 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
596 saddr = &ip6h->saddr;
597 daddr = &ip6h->daddr;
600 hp = tcp_get_md5sig_pool();
602 goto clear_hash_noput;
603 desc = &hp->md5_desc;
605 if (crypto_hash_init(desc))
608 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
610 if (tcp_md5_hash_header(hp, th))
612 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
614 if (tcp_md5_hash_key(hp, key))
616 if (crypto_hash_final(desc, md5_hash))
619 tcp_put_md5sig_pool();
623 tcp_put_md5sig_pool();
625 memset(md5_hash, 0, 16);
631 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
632 const struct sk_buff *skb)
634 #ifdef CONFIG_TCP_MD5SIG
635 const __u8 *hash_location = NULL;
636 struct tcp_md5sig_key *hash_expected;
637 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
638 const struct tcphdr *th = tcp_hdr(skb);
642 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
643 hash_location = tcp_parse_md5sig_option(th);
645 /* We've parsed the options - do we have a hash? */
646 if (!hash_expected && !hash_location)
649 if (hash_expected && !hash_location) {
650 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
654 if (!hash_expected && hash_location) {
655 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
659 /* check the signature */
660 genhash = tcp_v6_md5_hash_skb(newhash,
664 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
665 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
666 genhash ? "failed" : "mismatch",
667 &ip6h->saddr, ntohs(th->source),
668 &ip6h->daddr, ntohs(th->dest));
675 static void tcp_v6_init_req(struct request_sock *req,
676 const struct sock *sk_listener,
679 struct inet_request_sock *ireq = inet_rsk(req);
680 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
682 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
683 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
685 /* So that link locals have meaning */
686 if (!sk_listener->sk_bound_dev_if &&
687 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
688 ireq->ir_iif = tcp_v6_iif(skb);
690 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
691 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
692 np->rxopt.bits.rxinfo ||
693 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
694 np->rxopt.bits.rxohlim || np->repflow)) {
695 atomic_inc(&skb->users);
700 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
702 const struct request_sock *req,
707 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
710 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
712 .obj_size = sizeof(struct tcp6_request_sock),
713 .rtx_syn_ack = tcp_rtx_synack,
714 .send_ack = tcp_v6_reqsk_send_ack,
715 .destructor = tcp_v6_reqsk_destructor,
716 .send_reset = tcp_v6_send_reset,
717 .syn_ack_timeout = tcp_syn_ack_timeout,
720 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
721 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
722 sizeof(struct ipv6hdr),
723 #ifdef CONFIG_TCP_MD5SIG
724 .req_md5_lookup = tcp_v6_md5_lookup,
725 .calc_md5_hash = tcp_v6_md5_hash_skb,
727 .init_req = tcp_v6_init_req,
728 #ifdef CONFIG_SYN_COOKIES
729 .cookie_init_seq = cookie_v6_init_sequence,
731 .route_req = tcp_v6_route_req,
732 .init_seq = tcp_v6_init_sequence,
733 .send_synack = tcp_v6_send_synack,
736 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
737 u32 ack, u32 win, u32 tsval, u32 tsecr,
738 int oif, struct tcp_md5sig_key *key, int rst,
739 u8 tclass, u32 label)
741 const struct tcphdr *th = tcp_hdr(skb);
743 struct sk_buff *buff;
745 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
746 struct sock *ctl_sk = net->ipv6.tcp_sk;
747 unsigned int tot_len = sizeof(struct tcphdr);
748 struct dst_entry *dst;
752 tot_len += TCPOLEN_TSTAMP_ALIGNED;
753 #ifdef CONFIG_TCP_MD5SIG
755 tot_len += TCPOLEN_MD5SIG_ALIGNED;
758 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
763 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
765 t1 = (struct tcphdr *) skb_push(buff, tot_len);
766 skb_reset_transport_header(buff);
768 /* Swap the send and the receive. */
769 memset(t1, 0, sizeof(*t1));
770 t1->dest = th->source;
771 t1->source = th->dest;
772 t1->doff = tot_len / 4;
773 t1->seq = htonl(seq);
774 t1->ack_seq = htonl(ack);
775 t1->ack = !rst || !th->ack;
777 t1->window = htons(win);
779 topt = (__be32 *)(t1 + 1);
782 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
783 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
784 *topt++ = htonl(tsval);
785 *topt++ = htonl(tsecr);
788 #ifdef CONFIG_TCP_MD5SIG
790 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
791 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
792 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
793 &ipv6_hdr(skb)->saddr,
794 &ipv6_hdr(skb)->daddr, t1);
798 memset(&fl6, 0, sizeof(fl6));
799 fl6.daddr = ipv6_hdr(skb)->saddr;
800 fl6.saddr = ipv6_hdr(skb)->daddr;
801 fl6.flowlabel = label;
803 buff->ip_summed = CHECKSUM_PARTIAL;
806 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
808 fl6.flowi6_proto = IPPROTO_TCP;
809 if (rt6_need_strict(&fl6.daddr) && !oif)
810 fl6.flowi6_oif = tcp_v6_iif(skb);
812 fl6.flowi6_oif = oif;
813 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
814 fl6.fl6_dport = t1->dest;
815 fl6.fl6_sport = t1->source;
816 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
818 /* Pass a socket to ip6_dst_lookup either it is for RST
819 * Underlying function will use this to retrieve the network
822 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
824 skb_dst_set(buff, dst);
825 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
826 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
828 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
835 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
837 const struct tcphdr *th = tcp_hdr(skb);
838 u32 seq = 0, ack_seq = 0;
839 struct tcp_md5sig_key *key = NULL;
840 #ifdef CONFIG_TCP_MD5SIG
841 const __u8 *hash_location = NULL;
842 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
843 unsigned char newhash[16];
845 struct sock *sk1 = NULL;
852 /* If sk not NULL, it means we did a successful lookup and incoming
853 * route had to be correct. prequeue might have dropped our dst.
855 if (!sk && !ipv6_unicast_destination(skb))
858 #ifdef CONFIG_TCP_MD5SIG
859 hash_location = tcp_parse_md5sig_option(th);
860 if (!sk && hash_location) {
862 * active side is lost. Try to find listening socket through
863 * source port, and then find md5 key through listening socket.
864 * we are not loose security here:
865 * Incoming packet is checked with md5 hash with finding key,
866 * no RST generated if md5 hash doesn't match.
868 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
869 &tcp_hashinfo, &ipv6h->saddr,
870 th->source, &ipv6h->daddr,
871 ntohs(th->source), tcp_v6_iif(skb));
876 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
880 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
881 if (genhash || memcmp(hash_location, newhash, 16) != 0)
884 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
889 seq = ntohl(th->ack_seq);
891 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
894 oif = sk ? sk->sk_bound_dev_if : 0;
895 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
897 #ifdef CONFIG_TCP_MD5SIG
906 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
907 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
908 struct tcp_md5sig_key *key, u8 tclass,
911 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
915 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
917 struct inet_timewait_sock *tw = inet_twsk(sk);
918 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
920 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
921 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
922 tcp_time_stamp + tcptw->tw_ts_offset,
923 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
924 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
929 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
930 struct request_sock *req)
932 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
933 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
935 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
936 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
937 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
938 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
939 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
944 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
946 #ifdef CONFIG_SYN_COOKIES
947 const struct tcphdr *th = tcp_hdr(skb);
950 sk = cookie_v6_check(sk, skb);
955 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
957 if (skb->protocol == htons(ETH_P_IP))
958 return tcp_v4_conn_request(sk, skb);
960 if (!ipv6_unicast_destination(skb))
963 return tcp_conn_request(&tcp6_request_sock_ops,
964 &tcp_request_sock_ipv6_ops, sk, skb);
967 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
968 return 0; /* don't send reset */
971 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
972 struct request_sock *req,
973 struct dst_entry *dst,
974 struct request_sock *req_unhash,
977 struct inet_request_sock *ireq;
978 struct ipv6_pinfo *newnp;
979 const struct ipv6_pinfo *np = inet6_sk(sk);
980 struct ipv6_txoptions *opt;
981 struct tcp6_sock *newtcp6sk;
982 struct inet_sock *newinet;
983 struct tcp_sock *newtp;
985 #ifdef CONFIG_TCP_MD5SIG
986 struct tcp_md5sig_key *key;
990 if (skb->protocol == htons(ETH_P_IP)) {
995 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
996 req_unhash, own_req);
1001 newtcp6sk = (struct tcp6_sock *)newsk;
1002 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1004 newinet = inet_sk(newsk);
1005 newnp = inet6_sk(newsk);
1006 newtp = tcp_sk(newsk);
1008 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1010 newnp->saddr = newsk->sk_v6_rcv_saddr;
1012 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1013 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1014 #ifdef CONFIG_TCP_MD5SIG
1015 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1018 newnp->ipv6_ac_list = NULL;
1019 newnp->ipv6_fl_list = NULL;
1020 newnp->pktoptions = NULL;
1022 newnp->mcast_oif = tcp_v6_iif(skb);
1023 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1024 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1026 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1029 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1030 * here, tcp_create_openreq_child now does this for us, see the comment in
1031 * that function for the gory details. -acme
1034 /* It is tricky place. Until this moment IPv4 tcp
1035 worked with IPv6 icsk.icsk_af_ops.
1038 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1043 ireq = inet_rsk(req);
1045 if (sk_acceptq_is_full(sk))
1049 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1054 newsk = tcp_create_openreq_child(sk, req, skb);
1059 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1060 * count here, tcp_create_openreq_child now does this for us, see the
1061 * comment in that function for the gory details. -acme
1064 newsk->sk_gso_type = SKB_GSO_TCPV6;
1065 ip6_dst_store(newsk, dst, NULL, NULL);
1066 inet6_sk_rx_dst_set(newsk, skb);
1068 newtcp6sk = (struct tcp6_sock *)newsk;
1069 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1071 newtp = tcp_sk(newsk);
1072 newinet = inet_sk(newsk);
1073 newnp = inet6_sk(newsk);
1075 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1077 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1078 newnp->saddr = ireq->ir_v6_loc_addr;
1079 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1080 newsk->sk_bound_dev_if = ireq->ir_iif;
1082 /* Now IPv6 options...
1084 First: no IPv4 options.
1086 newinet->inet_opt = NULL;
1087 newnp->ipv6_ac_list = NULL;
1088 newnp->ipv6_fl_list = NULL;
1091 newnp->rxopt.all = np->rxopt.all;
1093 newnp->pktoptions = NULL;
1095 newnp->mcast_oif = tcp_v6_iif(skb);
1096 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1097 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1099 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1101 /* Clone native IPv6 options from listening socket (if any)
1103 Yes, keeping reference count would be much more clever,
1104 but we make one more one thing there: reattach optmem
1107 opt = rcu_dereference(np->opt);
1109 opt = ipv6_dup_options(newsk, opt);
1110 RCU_INIT_POINTER(newnp->opt, opt);
1112 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1114 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1117 tcp_ca_openreq_child(newsk, dst);
1119 tcp_sync_mss(newsk, dst_mtu(dst));
1120 newtp->advmss = dst_metric_advmss(dst);
1121 if (tcp_sk(sk)->rx_opt.user_mss &&
1122 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1123 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1125 tcp_initialize_rcv_mss(newsk);
1127 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1128 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1130 #ifdef CONFIG_TCP_MD5SIG
1131 /* Copy over the MD5 key from the original socket */
1132 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1134 /* We're using one, so create a matching key
1135 * on the newsk structure. If we fail to get
1136 * memory, then we end up not copying the key
1139 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1140 AF_INET6, key->key, key->keylen,
1141 sk_gfp_atomic(sk, GFP_ATOMIC));
1145 if (__inet_inherit_port(sk, newsk) < 0) {
1146 inet_csk_prepare_forced_close(newsk);
1150 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1152 tcp_move_syn(newtp, req);
1154 /* Clone pktoptions received with SYN, if we own the req */
1155 if (ireq->pktopts) {
1156 newnp->pktoptions = skb_clone(ireq->pktopts,
1157 sk_gfp_atomic(sk, GFP_ATOMIC));
1158 consume_skb(ireq->pktopts);
1159 ireq->pktopts = NULL;
1160 if (newnp->pktoptions)
1161 skb_set_owner_r(newnp->pktoptions, newsk);
1168 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1172 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1176 /* The socket must have it's spinlock held when we get
1177 * here, unless it is a TCP_LISTEN socket.
1179 * We have a potential double-lock case here, so even when
1180 * doing backlog processing we use the BH locking scheme.
1181 * This is because we cannot sleep with the original spinlock
1184 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1186 struct ipv6_pinfo *np = inet6_sk(sk);
1187 struct tcp_sock *tp;
1188 struct sk_buff *opt_skb = NULL;
1190 /* Imagine: socket is IPv6. IPv4 packet arrives,
1191 goes to IPv4 receive handler and backlogged.
1192 From backlog it always goes here. Kerboom...
1193 Fortunately, tcp_rcv_established and rcv_established
1194 handle them correctly, but it is not case with
1195 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1198 if (skb->protocol == htons(ETH_P_IP))
1199 return tcp_v4_do_rcv(sk, skb);
1201 if (sk_filter(sk, skb))
1205 * socket locking is here for SMP purposes as backlog rcv
1206 * is currently called with bh processing disabled.
1209 /* Do Stevens' IPV6_PKTOPTIONS.
1211 Yes, guys, it is the only place in our code, where we
1212 may make it not affecting IPv4.
1213 The rest of code is protocol independent,
1214 and I do not like idea to uglify IPv4.
1216 Actually, all the idea behind IPV6_PKTOPTIONS
1217 looks not very well thought. For now we latch
1218 options, received in the last packet, enqueued
1219 by tcp. Feel free to propose better solution.
1223 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1225 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1226 struct dst_entry *dst = sk->sk_rx_dst;
1228 sock_rps_save_rxhash(sk, skb);
1229 sk_mark_napi_id(sk, skb);
1231 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1232 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1234 sk->sk_rx_dst = NULL;
1238 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1240 goto ipv6_pktoptions;
1244 if (tcp_checksum_complete(skb))
1247 if (sk->sk_state == TCP_LISTEN) {
1248 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1254 sock_rps_save_rxhash(nsk, skb);
1255 sk_mark_napi_id(nsk, skb);
1256 if (tcp_child_process(sk, nsk, skb))
1259 __kfree_skb(opt_skb);
1263 sock_rps_save_rxhash(sk, skb);
1265 if (tcp_rcv_state_process(sk, skb))
1268 goto ipv6_pktoptions;
1272 tcp_v6_send_reset(sk, skb);
1275 __kfree_skb(opt_skb);
1279 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1280 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1285 /* Do you ask, what is it?
1287 1. skb was enqueued by tcp.
1288 2. skb is added to tail of read queue, rather than out of order.
1289 3. socket is not in passive state.
1290 4. Finally, it really contains options, which user wants to receive.
1293 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1294 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1295 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1296 np->mcast_oif = tcp_v6_iif(opt_skb);
1297 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1298 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1299 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1300 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1302 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1303 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1304 skb_set_owner_r(opt_skb, sk);
1305 opt_skb = xchg(&np->pktoptions, opt_skb);
1307 __kfree_skb(opt_skb);
1308 opt_skb = xchg(&np->pktoptions, NULL);
1316 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1317 const struct tcphdr *th)
1319 /* This is tricky: we move IP6CB at its correct location into
1320 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1321 * _decode_session6() uses IP6CB().
1322 * barrier() makes sure compiler won't play aliasing games.
1324 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1325 sizeof(struct inet6_skb_parm));
1328 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1329 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1330 skb->len - th->doff*4);
1331 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1332 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1333 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1334 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1335 TCP_SKB_CB(skb)->sacked = 0;
1338 static void tcp_v6_restore_cb(struct sk_buff *skb)
1340 /* We need to move header back to the beginning if xfrm6_policy_check()
1341 * and tcp_v6_fill_cb() are going to be called again.
1343 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1344 sizeof(struct inet6_skb_parm));
1347 static int tcp_v6_rcv(struct sk_buff *skb)
1349 const struct tcphdr *th;
1350 const struct ipv6hdr *hdr;
1353 struct net *net = dev_net(skb->dev);
1355 if (skb->pkt_type != PACKET_HOST)
1359 * Count it even if it's bad.
1361 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1363 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1368 if (th->doff < sizeof(struct tcphdr)/4)
1370 if (!pskb_may_pull(skb, th->doff*4))
1373 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1377 hdr = ipv6_hdr(skb);
1380 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest,
1386 if (sk->sk_state == TCP_TIME_WAIT)
1389 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1390 struct request_sock *req = inet_reqsk(sk);
1393 sk = req->rsk_listener;
1394 tcp_v6_fill_cb(skb, hdr, th);
1395 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1399 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1400 inet_csk_reqsk_queue_drop_and_put(sk, req);
1404 nsk = tcp_check_req(sk, skb, req, false);
1407 goto discard_and_relse;
1411 tcp_v6_restore_cb(skb);
1412 } else if (tcp_child_process(sk, nsk, skb)) {
1413 tcp_v6_send_reset(nsk, skb);
1414 goto discard_and_relse;
1420 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1421 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1422 goto discard_and_relse;
1425 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1426 goto discard_and_relse;
1428 tcp_v6_fill_cb(skb, hdr, th);
1430 if (tcp_v6_inbound_md5_hash(sk, skb))
1431 goto discard_and_relse;
1433 if (sk_filter(sk, skb))
1434 goto discard_and_relse;
1438 if (sk->sk_state == TCP_LISTEN) {
1439 ret = tcp_v6_do_rcv(sk, skb);
1440 goto put_and_return;
1443 sk_incoming_cpu_update(sk);
1445 bh_lock_sock_nested(sk);
1446 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1448 if (!sock_owned_by_user(sk)) {
1449 if (!tcp_prequeue(sk, skb))
1450 ret = tcp_v6_do_rcv(sk, skb);
1451 } else if (unlikely(sk_add_backlog(sk, skb,
1452 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1454 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1455 goto discard_and_relse;
1461 return ret ? -1 : 0;
1464 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1467 tcp_v6_fill_cb(skb, hdr, th);
1469 if (tcp_checksum_complete(skb)) {
1471 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1473 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1475 tcp_v6_send_reset(NULL, skb);
1487 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1488 inet_twsk_put(inet_twsk(sk));
1492 tcp_v6_fill_cb(skb, hdr, th);
1494 if (tcp_checksum_complete(skb)) {
1495 inet_twsk_put(inet_twsk(sk));
1499 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1504 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1505 &ipv6_hdr(skb)->saddr, th->source,
1506 &ipv6_hdr(skb)->daddr,
1507 ntohs(th->dest), tcp_v6_iif(skb));
1509 struct inet_timewait_sock *tw = inet_twsk(sk);
1510 inet_twsk_deschedule_put(tw);
1512 tcp_v6_restore_cb(skb);
1515 /* Fall through to ACK */
1518 tcp_v6_timewait_ack(sk, skb);
1521 tcp_v6_restore_cb(skb);
1523 case TCP_TW_SUCCESS:
1529 static void tcp_v6_early_demux(struct sk_buff *skb)
1531 const struct ipv6hdr *hdr;
1532 const struct tcphdr *th;
1535 if (skb->pkt_type != PACKET_HOST)
1538 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1541 hdr = ipv6_hdr(skb);
1544 if (th->doff < sizeof(struct tcphdr) / 4)
1547 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1548 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1549 &hdr->saddr, th->source,
1550 &hdr->daddr, ntohs(th->dest),
1554 skb->destructor = sock_edemux;
1555 if (sk_fullsock(sk)) {
1556 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1559 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1561 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1562 skb_dst_set_noref(skb, dst);
1567 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1568 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1569 .twsk_unique = tcp_twsk_unique,
1570 .twsk_destructor = tcp_twsk_destructor,
1573 static const struct inet_connection_sock_af_ops ipv6_specific = {
1574 .queue_xmit = inet6_csk_xmit,
1575 .send_check = tcp_v6_send_check,
1576 .rebuild_header = inet6_sk_rebuild_header,
1577 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1578 .conn_request = tcp_v6_conn_request,
1579 .syn_recv_sock = tcp_v6_syn_recv_sock,
1580 .net_header_len = sizeof(struct ipv6hdr),
1581 .net_frag_header_len = sizeof(struct frag_hdr),
1582 .setsockopt = ipv6_setsockopt,
1583 .getsockopt = ipv6_getsockopt,
1584 .addr2sockaddr = inet6_csk_addr2sockaddr,
1585 .sockaddr_len = sizeof(struct sockaddr_in6),
1586 .bind_conflict = inet6_csk_bind_conflict,
1587 #ifdef CONFIG_COMPAT
1588 .compat_setsockopt = compat_ipv6_setsockopt,
1589 .compat_getsockopt = compat_ipv6_getsockopt,
1591 .mtu_reduced = tcp_v6_mtu_reduced,
1594 #ifdef CONFIG_TCP_MD5SIG
1595 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1596 .md5_lookup = tcp_v6_md5_lookup,
1597 .calc_md5_hash = tcp_v6_md5_hash_skb,
1598 .md5_parse = tcp_v6_parse_md5_keys,
1603 * TCP over IPv4 via INET6 API
1605 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1606 .queue_xmit = ip_queue_xmit,
1607 .send_check = tcp_v4_send_check,
1608 .rebuild_header = inet_sk_rebuild_header,
1609 .sk_rx_dst_set = inet_sk_rx_dst_set,
1610 .conn_request = tcp_v6_conn_request,
1611 .syn_recv_sock = tcp_v6_syn_recv_sock,
1612 .net_header_len = sizeof(struct iphdr),
1613 .setsockopt = ipv6_setsockopt,
1614 .getsockopt = ipv6_getsockopt,
1615 .addr2sockaddr = inet6_csk_addr2sockaddr,
1616 .sockaddr_len = sizeof(struct sockaddr_in6),
1617 .bind_conflict = inet6_csk_bind_conflict,
1618 #ifdef CONFIG_COMPAT
1619 .compat_setsockopt = compat_ipv6_setsockopt,
1620 .compat_getsockopt = compat_ipv6_getsockopt,
1622 .mtu_reduced = tcp_v4_mtu_reduced,
1625 #ifdef CONFIG_TCP_MD5SIG
1626 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1627 .md5_lookup = tcp_v4_md5_lookup,
1628 .calc_md5_hash = tcp_v4_md5_hash_skb,
1629 .md5_parse = tcp_v6_parse_md5_keys,
1633 /* NOTE: A lot of things set to zero explicitly by call to
1634 * sk_alloc() so need not be done here.
1636 static int tcp_v6_init_sock(struct sock *sk)
1638 struct inet_connection_sock *icsk = inet_csk(sk);
1642 icsk->icsk_af_ops = &ipv6_specific;
1644 #ifdef CONFIG_TCP_MD5SIG
1645 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1651 static void tcp_v6_destroy_sock(struct sock *sk)
1653 tcp_v4_destroy_sock(sk);
1654 inet6_destroy_sock(sk);
1657 #ifdef CONFIG_PROC_FS
1658 /* Proc filesystem TCPv6 sock list dumping. */
1659 static void get_openreq6(struct seq_file *seq,
1660 const struct request_sock *req, int i)
1662 long ttd = req->rsk_timer.expires - jiffies;
1663 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1664 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1670 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1671 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1673 src->s6_addr32[0], src->s6_addr32[1],
1674 src->s6_addr32[2], src->s6_addr32[3],
1675 inet_rsk(req)->ir_num,
1676 dest->s6_addr32[0], dest->s6_addr32[1],
1677 dest->s6_addr32[2], dest->s6_addr32[3],
1678 ntohs(inet_rsk(req)->ir_rmt_port),
1680 0, 0, /* could print option size, but that is af dependent. */
1681 1, /* timers active (only the expire timer) */
1682 jiffies_to_clock_t(ttd),
1684 from_kuid_munged(seq_user_ns(seq),
1685 sock_i_uid(req->rsk_listener)),
1686 0, /* non standard timer */
1687 0, /* open_requests have no inode */
1691 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1693 const struct in6_addr *dest, *src;
1696 unsigned long timer_expires;
1697 const struct inet_sock *inet = inet_sk(sp);
1698 const struct tcp_sock *tp = tcp_sk(sp);
1699 const struct inet_connection_sock *icsk = inet_csk(sp);
1700 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1704 dest = &sp->sk_v6_daddr;
1705 src = &sp->sk_v6_rcv_saddr;
1706 destp = ntohs(inet->inet_dport);
1707 srcp = ntohs(inet->inet_sport);
1709 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1711 timer_expires = icsk->icsk_timeout;
1712 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1714 timer_expires = icsk->icsk_timeout;
1715 } else if (timer_pending(&sp->sk_timer)) {
1717 timer_expires = sp->sk_timer.expires;
1720 timer_expires = jiffies;
1723 state = sk_state_load(sp);
1724 if (state == TCP_LISTEN)
1725 rx_queue = sp->sk_ack_backlog;
1727 /* Because we don't lock the socket,
1728 * we might find a transient negative value.
1730 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1733 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1734 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1736 src->s6_addr32[0], src->s6_addr32[1],
1737 src->s6_addr32[2], src->s6_addr32[3], srcp,
1738 dest->s6_addr32[0], dest->s6_addr32[1],
1739 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1741 tp->write_seq - tp->snd_una,
1744 jiffies_delta_to_clock_t(timer_expires - jiffies),
1745 icsk->icsk_retransmits,
1746 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1747 icsk->icsk_probes_out,
1749 atomic_read(&sp->sk_refcnt), sp,
1750 jiffies_to_clock_t(icsk->icsk_rto),
1751 jiffies_to_clock_t(icsk->icsk_ack.ato),
1752 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1754 state == TCP_LISTEN ?
1755 fastopenq->max_qlen :
1756 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1760 static void get_timewait6_sock(struct seq_file *seq,
1761 struct inet_timewait_sock *tw, int i)
1763 long delta = tw->tw_timer.expires - jiffies;
1764 const struct in6_addr *dest, *src;
1767 dest = &tw->tw_v6_daddr;
1768 src = &tw->tw_v6_rcv_saddr;
1769 destp = ntohs(tw->tw_dport);
1770 srcp = ntohs(tw->tw_sport);
1773 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1774 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1776 src->s6_addr32[0], src->s6_addr32[1],
1777 src->s6_addr32[2], src->s6_addr32[3], srcp,
1778 dest->s6_addr32[0], dest->s6_addr32[1],
1779 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1780 tw->tw_substate, 0, 0,
1781 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1782 atomic_read(&tw->tw_refcnt), tw);
1785 static int tcp6_seq_show(struct seq_file *seq, void *v)
1787 struct tcp_iter_state *st;
1788 struct sock *sk = v;
1790 if (v == SEQ_START_TOKEN) {
1795 "st tx_queue rx_queue tr tm->when retrnsmt"
1796 " uid timeout inode\n");
1801 if (sk->sk_state == TCP_TIME_WAIT)
1802 get_timewait6_sock(seq, v, st->num);
1803 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1804 get_openreq6(seq, v, st->num);
1806 get_tcp6_sock(seq, v, st->num);
1811 static const struct file_operations tcp6_afinfo_seq_fops = {
1812 .owner = THIS_MODULE,
1813 .open = tcp_seq_open,
1815 .llseek = seq_lseek,
1816 .release = seq_release_net
1819 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1822 .seq_fops = &tcp6_afinfo_seq_fops,
1824 .show = tcp6_seq_show,
1828 int __net_init tcp6_proc_init(struct net *net)
1830 return tcp_proc_register(net, &tcp6_seq_afinfo);
1833 void tcp6_proc_exit(struct net *net)
1835 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1839 static void tcp_v6_clear_sk(struct sock *sk, int size)
1841 struct inet_sock *inet = inet_sk(sk);
1843 /* we do not want to clear pinet6 field, because of RCU lookups */
1844 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1846 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1847 memset(&inet->pinet6 + 1, 0, size);
1850 struct proto tcpv6_prot = {
1852 .owner = THIS_MODULE,
1854 .connect = tcp_v6_connect,
1855 .disconnect = tcp_disconnect,
1856 .accept = inet_csk_accept,
1858 .init = tcp_v6_init_sock,
1859 .destroy = tcp_v6_destroy_sock,
1860 .shutdown = tcp_shutdown,
1861 .setsockopt = tcp_setsockopt,
1862 .getsockopt = tcp_getsockopt,
1863 .recvmsg = tcp_recvmsg,
1864 .sendmsg = tcp_sendmsg,
1865 .sendpage = tcp_sendpage,
1866 .backlog_rcv = tcp_v6_do_rcv,
1867 .release_cb = tcp_release_cb,
1869 .unhash = inet_unhash,
1870 .get_port = inet_csk_get_port,
1871 .enter_memory_pressure = tcp_enter_memory_pressure,
1872 .stream_memory_free = tcp_stream_memory_free,
1873 .sockets_allocated = &tcp_sockets_allocated,
1874 .memory_allocated = &tcp_memory_allocated,
1875 .memory_pressure = &tcp_memory_pressure,
1876 .orphan_count = &tcp_orphan_count,
1877 .sysctl_mem = sysctl_tcp_mem,
1878 .sysctl_wmem = sysctl_tcp_wmem,
1879 .sysctl_rmem = sysctl_tcp_rmem,
1880 .max_header = MAX_TCP_HEADER,
1881 .obj_size = sizeof(struct tcp6_sock),
1882 .slab_flags = SLAB_DESTROY_BY_RCU,
1883 .twsk_prot = &tcp6_timewait_sock_ops,
1884 .rsk_prot = &tcp6_request_sock_ops,
1885 .h.hashinfo = &tcp_hashinfo,
1886 .no_autobind = true,
1887 #ifdef CONFIG_COMPAT
1888 .compat_setsockopt = compat_tcp_setsockopt,
1889 .compat_getsockopt = compat_tcp_getsockopt,
1891 #ifdef CONFIG_MEMCG_KMEM
1892 .proto_cgroup = tcp_proto_cgroup,
1894 .clear_sk = tcp_v6_clear_sk,
1897 static const struct inet6_protocol tcpv6_protocol = {
1898 .early_demux = tcp_v6_early_demux,
1899 .handler = tcp_v6_rcv,
1900 .err_handler = tcp_v6_err,
1901 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1904 static struct inet_protosw tcpv6_protosw = {
1905 .type = SOCK_STREAM,
1906 .protocol = IPPROTO_TCP,
1907 .prot = &tcpv6_prot,
1908 .ops = &inet6_stream_ops,
1909 .flags = INET_PROTOSW_PERMANENT |
1913 static int __net_init tcpv6_net_init(struct net *net)
1915 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1916 SOCK_RAW, IPPROTO_TCP, net);
1919 static void __net_exit tcpv6_net_exit(struct net *net)
1921 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1924 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1926 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1929 static struct pernet_operations tcpv6_net_ops = {
1930 .init = tcpv6_net_init,
1931 .exit = tcpv6_net_exit,
1932 .exit_batch = tcpv6_net_exit_batch,
1935 int __init tcpv6_init(void)
1939 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1943 /* register inet6 protocol */
1944 ret = inet6_register_protosw(&tcpv6_protosw);
1946 goto out_tcpv6_protocol;
1948 ret = register_pernet_subsys(&tcpv6_net_ops);
1950 goto out_tcpv6_protosw;
1955 inet6_unregister_protosw(&tcpv6_protosw);
1957 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1961 void tcpv6_exit(void)
1963 unregister_pernet_subsys(&tcpv6_net_ops);
1964 inet6_unregister_protosw(&tcpv6_protosw);
1965 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);