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;
237 fl6.flowi6_uid = sk->sk_uid;
239 opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
240 final_p = fl6_update_dst(&fl6, opt, &final);
242 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
244 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
252 sk->sk_v6_rcv_saddr = *saddr;
255 /* set the source address */
257 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
259 sk->sk_gso_type = SKB_GSO_TCPV6;
260 ip6_dst_store(sk, dst, NULL, NULL);
262 if (tcp_death_row.sysctl_tw_recycle &&
263 !tp->rx_opt.ts_recent_stamp &&
264 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
265 tcp_fetch_timewait_stamp(sk, dst);
267 icsk->icsk_ext_hdr_len = 0;
269 icsk->icsk_ext_hdr_len = opt->opt_flen +
272 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
274 inet->inet_dport = usin->sin6_port;
276 tcp_set_state(sk, TCP_SYN_SENT);
277 err = inet6_hash_connect(&tcp_death_row, sk);
283 if (!tp->write_seq && likely(!tp->repair))
284 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
285 sk->sk_v6_daddr.s6_addr32,
289 err = tcp_connect(sk);
296 tcp_set_state(sk, TCP_CLOSE);
299 inet->inet_dport = 0;
300 sk->sk_route_caps = 0;
304 static void tcp_v6_mtu_reduced(struct sock *sk)
306 struct dst_entry *dst;
308 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
311 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
315 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
316 tcp_sync_mss(sk, dst_mtu(dst));
317 tcp_simple_retransmit(sk);
321 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
322 u8 type, u8 code, int offset, __be32 info)
324 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
325 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
326 struct net *net = dev_net(skb->dev);
327 struct request_sock *fastopen;
328 struct ipv6_pinfo *np;
335 sk = __inet6_lookup_established(net, &tcp_hashinfo,
336 &hdr->daddr, th->dest,
337 &hdr->saddr, ntohs(th->source),
341 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
346 if (sk->sk_state == TCP_TIME_WAIT) {
347 inet_twsk_put(inet_twsk(sk));
350 seq = ntohl(th->seq);
351 fatal = icmpv6_err_convert(type, code, &err);
352 if (sk->sk_state == TCP_NEW_SYN_RECV)
353 return tcp_req_err(sk, seq, fatal);
356 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
357 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
359 if (sk->sk_state == TCP_CLOSE)
362 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
363 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
368 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
369 fastopen = tp->fastopen_rsk;
370 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
371 if (sk->sk_state != TCP_LISTEN &&
372 !between(seq, snd_una, tp->snd_nxt)) {
373 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
379 if (type == NDISC_REDIRECT) {
380 if (!sock_owned_by_user(sk)) {
381 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
384 dst->ops->redirect(dst, sk, skb);
389 if (type == ICMPV6_PKT_TOOBIG) {
390 /* We are not interested in TCP_LISTEN and open_requests
391 * (SYN-ACKs send out by Linux are always <576bytes so
392 * they should go through unfragmented).
394 if (sk->sk_state == TCP_LISTEN)
397 if (!ip6_sk_accept_pmtu(sk))
400 tp->mtu_info = ntohl(info);
401 if (!sock_owned_by_user(sk))
402 tcp_v6_mtu_reduced(sk);
403 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
410 /* Might be for an request_sock */
411 switch (sk->sk_state) {
414 /* Only in fast or simultaneous open. If a fast open socket is
415 * is already accepted it is treated as a connected one below.
417 if (fastopen && !fastopen->sk)
420 if (!sock_owned_by_user(sk)) {
422 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
426 sk->sk_err_soft = err;
430 if (!sock_owned_by_user(sk) && np->recverr) {
432 sk->sk_error_report(sk);
434 sk->sk_err_soft = err;
442 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
444 struct request_sock *req,
445 struct tcp_fastopen_cookie *foc,
448 struct inet_request_sock *ireq = inet_rsk(req);
449 struct ipv6_pinfo *np = inet6_sk(sk);
450 struct flowi6 *fl6 = &fl->u.ip6;
454 /* First, grab a route. */
455 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
456 IPPROTO_TCP)) == NULL)
459 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
462 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
463 &ireq->ir_v6_rmt_addr);
465 fl6->daddr = ireq->ir_v6_rmt_addr;
466 if (np->repflow && ireq->pktopts)
467 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
470 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
473 err = net_xmit_eval(err);
481 static void tcp_v6_reqsk_destructor(struct request_sock *req)
483 kfree_skb(inet_rsk(req)->pktopts);
486 #ifdef CONFIG_TCP_MD5SIG
487 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
488 const struct in6_addr *addr)
490 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
493 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
494 const struct sock *addr_sk)
496 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
499 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
502 struct tcp_md5sig cmd;
503 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
505 if (optlen < sizeof(cmd))
508 if (copy_from_user(&cmd, optval, sizeof(cmd)))
511 if (sin6->sin6_family != AF_INET6)
514 if (!cmd.tcpm_keylen) {
515 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
516 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
518 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
522 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
525 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
526 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
527 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
529 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
530 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
533 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
534 const struct in6_addr *daddr,
535 const struct in6_addr *saddr, int nbytes)
537 struct tcp6_pseudohdr *bp;
538 struct scatterlist sg;
540 bp = &hp->md5_blk.ip6;
541 /* 1. TCP pseudo-header (RFC2460) */
544 bp->protocol = cpu_to_be32(IPPROTO_TCP);
545 bp->len = cpu_to_be32(nbytes);
547 sg_init_one(&sg, bp, sizeof(*bp));
548 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
551 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
552 const struct in6_addr *daddr, struct in6_addr *saddr,
553 const struct tcphdr *th)
555 struct tcp_md5sig_pool *hp;
556 struct hash_desc *desc;
558 hp = tcp_get_md5sig_pool();
560 goto clear_hash_noput;
561 desc = &hp->md5_desc;
563 if (crypto_hash_init(desc))
565 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
567 if (tcp_md5_hash_header(hp, th))
569 if (tcp_md5_hash_key(hp, key))
571 if (crypto_hash_final(desc, md5_hash))
574 tcp_put_md5sig_pool();
578 tcp_put_md5sig_pool();
580 memset(md5_hash, 0, 16);
584 static int tcp_v6_md5_hash_skb(char *md5_hash,
585 const struct tcp_md5sig_key *key,
586 const struct sock *sk,
587 const struct sk_buff *skb)
589 const struct in6_addr *saddr, *daddr;
590 struct tcp_md5sig_pool *hp;
591 struct hash_desc *desc;
592 const struct tcphdr *th = tcp_hdr(skb);
594 if (sk) { /* valid for establish/request sockets */
595 saddr = &sk->sk_v6_rcv_saddr;
596 daddr = &sk->sk_v6_daddr;
598 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
599 saddr = &ip6h->saddr;
600 daddr = &ip6h->daddr;
603 hp = tcp_get_md5sig_pool();
605 goto clear_hash_noput;
606 desc = &hp->md5_desc;
608 if (crypto_hash_init(desc))
611 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
613 if (tcp_md5_hash_header(hp, th))
615 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
617 if (tcp_md5_hash_key(hp, key))
619 if (crypto_hash_final(desc, md5_hash))
622 tcp_put_md5sig_pool();
626 tcp_put_md5sig_pool();
628 memset(md5_hash, 0, 16);
634 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
635 const struct sk_buff *skb)
637 #ifdef CONFIG_TCP_MD5SIG
638 const __u8 *hash_location = NULL;
639 struct tcp_md5sig_key *hash_expected;
640 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
641 const struct tcphdr *th = tcp_hdr(skb);
645 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
646 hash_location = tcp_parse_md5sig_option(th);
648 /* We've parsed the options - do we have a hash? */
649 if (!hash_expected && !hash_location)
652 if (hash_expected && !hash_location) {
653 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
657 if (!hash_expected && hash_location) {
658 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
662 /* check the signature */
663 genhash = tcp_v6_md5_hash_skb(newhash,
667 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
668 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
669 genhash ? "failed" : "mismatch",
670 &ip6h->saddr, ntohs(th->source),
671 &ip6h->daddr, ntohs(th->dest));
678 static void tcp_v6_init_req(struct request_sock *req,
679 const struct sock *sk_listener,
682 struct inet_request_sock *ireq = inet_rsk(req);
683 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
685 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
686 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
688 /* So that link locals have meaning */
689 if (!sk_listener->sk_bound_dev_if &&
690 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
691 ireq->ir_iif = tcp_v6_iif(skb);
693 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
694 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
695 np->rxopt.bits.rxinfo ||
696 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
697 np->rxopt.bits.rxohlim || np->repflow)) {
698 atomic_inc(&skb->users);
703 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
705 const struct request_sock *req,
710 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
713 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
715 .obj_size = sizeof(struct tcp6_request_sock),
716 .rtx_syn_ack = tcp_rtx_synack,
717 .send_ack = tcp_v6_reqsk_send_ack,
718 .destructor = tcp_v6_reqsk_destructor,
719 .send_reset = tcp_v6_send_reset,
720 .syn_ack_timeout = tcp_syn_ack_timeout,
723 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
724 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
725 sizeof(struct ipv6hdr),
726 #ifdef CONFIG_TCP_MD5SIG
727 .req_md5_lookup = tcp_v6_md5_lookup,
728 .calc_md5_hash = tcp_v6_md5_hash_skb,
730 .init_req = tcp_v6_init_req,
731 #ifdef CONFIG_SYN_COOKIES
732 .cookie_init_seq = cookie_v6_init_sequence,
734 .route_req = tcp_v6_route_req,
735 .init_seq = tcp_v6_init_sequence,
736 .send_synack = tcp_v6_send_synack,
739 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
740 u32 ack, u32 win, u32 tsval, u32 tsecr,
741 int oif, struct tcp_md5sig_key *key, int rst,
742 u8 tclass, u32 label)
744 const struct tcphdr *th = tcp_hdr(skb);
746 struct sk_buff *buff;
748 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
749 struct sock *ctl_sk = net->ipv6.tcp_sk;
750 unsigned int tot_len = sizeof(struct tcphdr);
751 struct dst_entry *dst;
755 tot_len += TCPOLEN_TSTAMP_ALIGNED;
756 #ifdef CONFIG_TCP_MD5SIG
758 tot_len += TCPOLEN_MD5SIG_ALIGNED;
761 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
766 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
768 t1 = (struct tcphdr *) skb_push(buff, tot_len);
769 skb_reset_transport_header(buff);
771 /* Swap the send and the receive. */
772 memset(t1, 0, sizeof(*t1));
773 t1->dest = th->source;
774 t1->source = th->dest;
775 t1->doff = tot_len / 4;
776 t1->seq = htonl(seq);
777 t1->ack_seq = htonl(ack);
778 t1->ack = !rst || !th->ack;
780 t1->window = htons(win);
782 topt = (__be32 *)(t1 + 1);
785 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
786 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
787 *topt++ = htonl(tsval);
788 *topt++ = htonl(tsecr);
791 #ifdef CONFIG_TCP_MD5SIG
793 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
794 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
795 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
796 &ipv6_hdr(skb)->saddr,
797 &ipv6_hdr(skb)->daddr, t1);
801 memset(&fl6, 0, sizeof(fl6));
802 fl6.daddr = ipv6_hdr(skb)->saddr;
803 fl6.saddr = ipv6_hdr(skb)->daddr;
804 fl6.flowlabel = label;
806 buff->ip_summed = CHECKSUM_PARTIAL;
809 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
811 fl6.flowi6_proto = IPPROTO_TCP;
812 if (rt6_need_strict(&fl6.daddr) && !oif)
813 fl6.flowi6_oif = tcp_v6_iif(skb);
815 fl6.flowi6_oif = oif;
816 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
817 fl6.fl6_dport = t1->dest;
818 fl6.fl6_sport = t1->source;
819 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
820 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
822 /* Pass a socket to ip6_dst_lookup either it is for RST
823 * Underlying function will use this to retrieve the network
826 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
828 skb_dst_set(buff, dst);
829 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
830 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
832 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
839 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
841 const struct tcphdr *th = tcp_hdr(skb);
842 u32 seq = 0, ack_seq = 0;
843 struct tcp_md5sig_key *key = NULL;
844 #ifdef CONFIG_TCP_MD5SIG
845 const __u8 *hash_location = NULL;
846 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
847 unsigned char newhash[16];
849 struct sock *sk1 = NULL;
856 /* If sk not NULL, it means we did a successful lookup and incoming
857 * route had to be correct. prequeue might have dropped our dst.
859 if (!sk && !ipv6_unicast_destination(skb))
862 #ifdef CONFIG_TCP_MD5SIG
863 hash_location = tcp_parse_md5sig_option(th);
864 if (!sk && hash_location) {
866 * active side is lost. Try to find listening socket through
867 * source port, and then find md5 key through listening socket.
868 * we are not loose security here:
869 * Incoming packet is checked with md5 hash with finding key,
870 * no RST generated if md5 hash doesn't match.
872 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
873 &tcp_hashinfo, &ipv6h->saddr,
874 th->source, &ipv6h->daddr,
875 ntohs(th->source), tcp_v6_iif(skb));
880 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
884 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
885 if (genhash || memcmp(hash_location, newhash, 16) != 0)
888 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
893 seq = ntohl(th->ack_seq);
895 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
898 oif = sk ? sk->sk_bound_dev_if : 0;
899 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
901 #ifdef CONFIG_TCP_MD5SIG
910 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
911 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
912 struct tcp_md5sig_key *key, u8 tclass,
915 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
919 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
921 struct inet_timewait_sock *tw = inet_twsk(sk);
922 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
924 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
925 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
926 tcp_time_stamp + tcptw->tw_ts_offset,
927 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
928 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
933 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
934 struct request_sock *req)
936 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
937 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
940 * The window field (SEG.WND) of every outgoing segment, with the
941 * exception of <SYN> segments, MUST be right-shifted by
942 * Rcv.Wind.Shift bits:
944 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
945 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
946 tcp_rsk(req)->rcv_nxt,
947 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
948 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
949 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
954 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
956 #ifdef CONFIG_SYN_COOKIES
957 const struct tcphdr *th = tcp_hdr(skb);
960 sk = cookie_v6_check(sk, skb);
965 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
967 if (skb->protocol == htons(ETH_P_IP))
968 return tcp_v4_conn_request(sk, skb);
970 if (!ipv6_unicast_destination(skb))
973 return tcp_conn_request(&tcp6_request_sock_ops,
974 &tcp_request_sock_ipv6_ops, sk, skb);
977 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
978 return 0; /* don't send reset */
981 static void tcp_v6_restore_cb(struct sk_buff *skb)
983 /* We need to move header back to the beginning if xfrm6_policy_check()
984 * and tcp_v6_fill_cb() are going to be called again.
985 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
987 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
988 sizeof(struct inet6_skb_parm));
991 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
992 struct request_sock *req,
993 struct dst_entry *dst,
994 struct request_sock *req_unhash,
997 struct inet_request_sock *ireq;
998 struct ipv6_pinfo *newnp;
999 const struct ipv6_pinfo *np = inet6_sk(sk);
1000 struct ipv6_txoptions *opt;
1001 struct tcp6_sock *newtcp6sk;
1002 struct inet_sock *newinet;
1003 struct tcp_sock *newtp;
1005 #ifdef CONFIG_TCP_MD5SIG
1006 struct tcp_md5sig_key *key;
1010 if (skb->protocol == htons(ETH_P_IP)) {
1015 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1016 req_unhash, own_req);
1021 newtcp6sk = (struct tcp6_sock *)newsk;
1022 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1024 newinet = inet_sk(newsk);
1025 newnp = inet6_sk(newsk);
1026 newtp = tcp_sk(newsk);
1028 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1030 newnp->saddr = newsk->sk_v6_rcv_saddr;
1032 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1033 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1034 #ifdef CONFIG_TCP_MD5SIG
1035 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1038 newnp->ipv6_ac_list = NULL;
1039 newnp->ipv6_fl_list = NULL;
1040 newnp->pktoptions = NULL;
1042 newnp->mcast_oif = tcp_v6_iif(skb);
1043 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1044 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1046 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1049 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1050 * here, tcp_create_openreq_child now does this for us, see the comment in
1051 * that function for the gory details. -acme
1054 /* It is tricky place. Until this moment IPv4 tcp
1055 worked with IPv6 icsk.icsk_af_ops.
1058 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1063 ireq = inet_rsk(req);
1065 if (sk_acceptq_is_full(sk))
1069 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1074 newsk = tcp_create_openreq_child(sk, req, skb);
1079 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1080 * count here, tcp_create_openreq_child now does this for us, see the
1081 * comment in that function for the gory details. -acme
1084 newsk->sk_gso_type = SKB_GSO_TCPV6;
1085 ip6_dst_store(newsk, dst, NULL, NULL);
1086 inet6_sk_rx_dst_set(newsk, skb);
1088 newtcp6sk = (struct tcp6_sock *)newsk;
1089 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1091 newtp = tcp_sk(newsk);
1092 newinet = inet_sk(newsk);
1093 newnp = inet6_sk(newsk);
1095 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1097 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1098 newnp->saddr = ireq->ir_v6_loc_addr;
1099 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1100 newsk->sk_bound_dev_if = ireq->ir_iif;
1102 /* Now IPv6 options...
1104 First: no IPv4 options.
1106 newinet->inet_opt = NULL;
1107 newnp->ipv6_ac_list = NULL;
1108 newnp->ipv6_fl_list = NULL;
1111 newnp->rxopt.all = np->rxopt.all;
1113 newnp->pktoptions = NULL;
1115 newnp->mcast_oif = tcp_v6_iif(skb);
1116 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1117 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1119 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1121 /* Clone native IPv6 options from listening socket (if any)
1123 Yes, keeping reference count would be much more clever,
1124 but we make one more one thing there: reattach optmem
1127 opt = rcu_dereference(np->opt);
1129 opt = ipv6_dup_options(newsk, opt);
1130 RCU_INIT_POINTER(newnp->opt, opt);
1132 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1134 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1137 tcp_ca_openreq_child(newsk, dst);
1139 tcp_sync_mss(newsk, dst_mtu(dst));
1140 newtp->advmss = dst_metric_advmss(dst);
1141 if (tcp_sk(sk)->rx_opt.user_mss &&
1142 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1143 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1145 tcp_initialize_rcv_mss(newsk);
1147 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1148 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1150 #ifdef CONFIG_TCP_MD5SIG
1151 /* Copy over the MD5 key from the original socket */
1152 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1154 /* We're using one, so create a matching key
1155 * on the newsk structure. If we fail to get
1156 * memory, then we end up not copying the key
1159 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1160 AF_INET6, key->key, key->keylen,
1161 sk_gfp_atomic(sk, GFP_ATOMIC));
1165 if (__inet_inherit_port(sk, newsk) < 0) {
1166 inet_csk_prepare_forced_close(newsk);
1170 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1172 tcp_move_syn(newtp, req);
1174 /* Clone pktoptions received with SYN, if we own the req */
1175 if (ireq->pktopts) {
1176 newnp->pktoptions = skb_clone(ireq->pktopts,
1177 sk_gfp_atomic(sk, GFP_ATOMIC));
1178 consume_skb(ireq->pktopts);
1179 ireq->pktopts = NULL;
1180 if (newnp->pktoptions) {
1181 tcp_v6_restore_cb(newnp->pktoptions);
1182 skb_set_owner_r(newnp->pktoptions, newsk);
1190 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1194 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1198 /* The socket must have it's spinlock held when we get
1199 * here, unless it is a TCP_LISTEN socket.
1201 * We have a potential double-lock case here, so even when
1202 * doing backlog processing we use the BH locking scheme.
1203 * This is because we cannot sleep with the original spinlock
1206 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1208 struct ipv6_pinfo *np = inet6_sk(sk);
1209 struct tcp_sock *tp;
1210 struct sk_buff *opt_skb = NULL;
1212 /* Imagine: socket is IPv6. IPv4 packet arrives,
1213 goes to IPv4 receive handler and backlogged.
1214 From backlog it always goes here. Kerboom...
1215 Fortunately, tcp_rcv_established and rcv_established
1216 handle them correctly, but it is not case with
1217 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1220 if (skb->protocol == htons(ETH_P_IP))
1221 return tcp_v4_do_rcv(sk, skb);
1223 if (tcp_filter(sk, skb))
1227 * socket locking is here for SMP purposes as backlog rcv
1228 * is currently called with bh processing disabled.
1231 /* Do Stevens' IPV6_PKTOPTIONS.
1233 Yes, guys, it is the only place in our code, where we
1234 may make it not affecting IPv4.
1235 The rest of code is protocol independent,
1236 and I do not like idea to uglify IPv4.
1238 Actually, all the idea behind IPV6_PKTOPTIONS
1239 looks not very well thought. For now we latch
1240 options, received in the last packet, enqueued
1241 by tcp. Feel free to propose better solution.
1245 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1247 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1248 struct dst_entry *dst = sk->sk_rx_dst;
1250 sock_rps_save_rxhash(sk, skb);
1251 sk_mark_napi_id(sk, skb);
1253 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1254 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1256 sk->sk_rx_dst = NULL;
1260 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1262 goto ipv6_pktoptions;
1266 if (tcp_checksum_complete(skb))
1269 if (sk->sk_state == TCP_LISTEN) {
1270 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1276 sock_rps_save_rxhash(nsk, skb);
1277 sk_mark_napi_id(nsk, skb);
1278 if (tcp_child_process(sk, nsk, skb))
1281 __kfree_skb(opt_skb);
1285 sock_rps_save_rxhash(sk, skb);
1287 if (tcp_rcv_state_process(sk, skb))
1290 goto ipv6_pktoptions;
1294 tcp_v6_send_reset(sk, skb);
1297 __kfree_skb(opt_skb);
1301 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1302 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1307 /* Do you ask, what is it?
1309 1. skb was enqueued by tcp.
1310 2. skb is added to tail of read queue, rather than out of order.
1311 3. socket is not in passive state.
1312 4. Finally, it really contains options, which user wants to receive.
1315 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1316 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1317 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1318 np->mcast_oif = tcp_v6_iif(opt_skb);
1319 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1320 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1321 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1322 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1324 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1325 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1326 skb_set_owner_r(opt_skb, sk);
1327 tcp_v6_restore_cb(opt_skb);
1328 opt_skb = xchg(&np->pktoptions, opt_skb);
1330 __kfree_skb(opt_skb);
1331 opt_skb = xchg(&np->pktoptions, NULL);
1339 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1340 const struct tcphdr *th)
1342 /* This is tricky: we move IP6CB at its correct location into
1343 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1344 * _decode_session6() uses IP6CB().
1345 * barrier() makes sure compiler won't play aliasing games.
1347 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1348 sizeof(struct inet6_skb_parm));
1351 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1352 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1353 skb->len - th->doff*4);
1354 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1355 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1356 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1357 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1358 TCP_SKB_CB(skb)->sacked = 0;
1361 static int tcp_v6_rcv(struct sk_buff *skb)
1363 const struct tcphdr *th;
1364 const struct ipv6hdr *hdr;
1367 struct net *net = dev_net(skb->dev);
1369 if (skb->pkt_type != PACKET_HOST)
1373 * Count it even if it's bad.
1375 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1377 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1382 if (th->doff < sizeof(struct tcphdr)/4)
1384 if (!pskb_may_pull(skb, th->doff*4))
1387 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1391 hdr = ipv6_hdr(skb);
1394 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest,
1400 if (sk->sk_state == TCP_TIME_WAIT)
1403 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1404 struct request_sock *req = inet_reqsk(sk);
1407 sk = req->rsk_listener;
1408 tcp_v6_fill_cb(skb, hdr, th);
1409 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1413 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1414 inet_csk_reqsk_queue_drop_and_put(sk, req);
1418 nsk = tcp_check_req(sk, skb, req, false);
1421 goto discard_and_relse;
1425 tcp_v6_restore_cb(skb);
1426 } else if (tcp_child_process(sk, nsk, skb)) {
1427 tcp_v6_send_reset(nsk, skb);
1428 goto discard_and_relse;
1434 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1435 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1436 goto discard_and_relse;
1439 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1440 goto discard_and_relse;
1442 tcp_v6_fill_cb(skb, hdr, th);
1444 if (tcp_v6_inbound_md5_hash(sk, skb))
1445 goto discard_and_relse;
1447 if (tcp_filter(sk, skb))
1448 goto discard_and_relse;
1449 th = (const struct tcphdr *)skb->data;
1450 hdr = ipv6_hdr(skb);
1454 if (sk->sk_state == TCP_LISTEN) {
1455 ret = tcp_v6_do_rcv(sk, skb);
1456 goto put_and_return;
1459 sk_incoming_cpu_update(sk);
1461 bh_lock_sock_nested(sk);
1462 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1464 if (!sock_owned_by_user(sk)) {
1465 if (!tcp_prequeue(sk, skb))
1466 ret = tcp_v6_do_rcv(sk, skb);
1467 } else if (unlikely(sk_add_backlog(sk, skb,
1468 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1470 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1471 goto discard_and_relse;
1477 return ret ? -1 : 0;
1480 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1483 tcp_v6_fill_cb(skb, hdr, th);
1485 if (tcp_checksum_complete(skb)) {
1487 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1489 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1491 tcp_v6_send_reset(NULL, skb);
1503 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1504 inet_twsk_put(inet_twsk(sk));
1508 tcp_v6_fill_cb(skb, hdr, th);
1510 if (tcp_checksum_complete(skb)) {
1511 inet_twsk_put(inet_twsk(sk));
1515 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1520 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1521 &ipv6_hdr(skb)->saddr, th->source,
1522 &ipv6_hdr(skb)->daddr,
1523 ntohs(th->dest), tcp_v6_iif(skb));
1525 struct inet_timewait_sock *tw = inet_twsk(sk);
1526 inet_twsk_deschedule_put(tw);
1528 tcp_v6_restore_cb(skb);
1531 /* Fall through to ACK */
1534 tcp_v6_timewait_ack(sk, skb);
1537 tcp_v6_restore_cb(skb);
1539 case TCP_TW_SUCCESS:
1545 static void tcp_v6_early_demux(struct sk_buff *skb)
1547 const struct ipv6hdr *hdr;
1548 const struct tcphdr *th;
1551 if (skb->pkt_type != PACKET_HOST)
1554 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1557 hdr = ipv6_hdr(skb);
1560 if (th->doff < sizeof(struct tcphdr) / 4)
1563 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1564 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1565 &hdr->saddr, th->source,
1566 &hdr->daddr, ntohs(th->dest),
1570 skb->destructor = sock_edemux;
1571 if (sk_fullsock(sk)) {
1572 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1575 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1577 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1578 skb_dst_set_noref(skb, dst);
1583 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1584 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1585 .twsk_unique = tcp_twsk_unique,
1586 .twsk_destructor = tcp_twsk_destructor,
1589 static const struct inet_connection_sock_af_ops ipv6_specific = {
1590 .queue_xmit = inet6_csk_xmit,
1591 .send_check = tcp_v6_send_check,
1592 .rebuild_header = inet6_sk_rebuild_header,
1593 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1594 .conn_request = tcp_v6_conn_request,
1595 .syn_recv_sock = tcp_v6_syn_recv_sock,
1596 .net_header_len = sizeof(struct ipv6hdr),
1597 .net_frag_header_len = sizeof(struct frag_hdr),
1598 .setsockopt = ipv6_setsockopt,
1599 .getsockopt = ipv6_getsockopt,
1600 .addr2sockaddr = inet6_csk_addr2sockaddr,
1601 .sockaddr_len = sizeof(struct sockaddr_in6),
1602 .bind_conflict = inet6_csk_bind_conflict,
1603 #ifdef CONFIG_COMPAT
1604 .compat_setsockopt = compat_ipv6_setsockopt,
1605 .compat_getsockopt = compat_ipv6_getsockopt,
1607 .mtu_reduced = tcp_v6_mtu_reduced,
1610 #ifdef CONFIG_TCP_MD5SIG
1611 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1612 .md5_lookup = tcp_v6_md5_lookup,
1613 .calc_md5_hash = tcp_v6_md5_hash_skb,
1614 .md5_parse = tcp_v6_parse_md5_keys,
1619 * TCP over IPv4 via INET6 API
1621 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1622 .queue_xmit = ip_queue_xmit,
1623 .send_check = tcp_v4_send_check,
1624 .rebuild_header = inet_sk_rebuild_header,
1625 .sk_rx_dst_set = inet_sk_rx_dst_set,
1626 .conn_request = tcp_v6_conn_request,
1627 .syn_recv_sock = tcp_v6_syn_recv_sock,
1628 .net_header_len = sizeof(struct iphdr),
1629 .setsockopt = ipv6_setsockopt,
1630 .getsockopt = ipv6_getsockopt,
1631 .addr2sockaddr = inet6_csk_addr2sockaddr,
1632 .sockaddr_len = sizeof(struct sockaddr_in6),
1633 .bind_conflict = inet6_csk_bind_conflict,
1634 #ifdef CONFIG_COMPAT
1635 .compat_setsockopt = compat_ipv6_setsockopt,
1636 .compat_getsockopt = compat_ipv6_getsockopt,
1638 .mtu_reduced = tcp_v4_mtu_reduced,
1641 #ifdef CONFIG_TCP_MD5SIG
1642 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1643 .md5_lookup = tcp_v4_md5_lookup,
1644 .calc_md5_hash = tcp_v4_md5_hash_skb,
1645 .md5_parse = tcp_v6_parse_md5_keys,
1649 /* NOTE: A lot of things set to zero explicitly by call to
1650 * sk_alloc() so need not be done here.
1652 static int tcp_v6_init_sock(struct sock *sk)
1654 struct inet_connection_sock *icsk = inet_csk(sk);
1658 icsk->icsk_af_ops = &ipv6_specific;
1660 #ifdef CONFIG_TCP_MD5SIG
1661 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1667 static void tcp_v6_destroy_sock(struct sock *sk)
1669 tcp_v4_destroy_sock(sk);
1670 inet6_destroy_sock(sk);
1673 #ifdef CONFIG_PROC_FS
1674 /* Proc filesystem TCPv6 sock list dumping. */
1675 static void get_openreq6(struct seq_file *seq,
1676 const struct request_sock *req, int i)
1678 long ttd = req->rsk_timer.expires - jiffies;
1679 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1680 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1686 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1687 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1689 src->s6_addr32[0], src->s6_addr32[1],
1690 src->s6_addr32[2], src->s6_addr32[3],
1691 inet_rsk(req)->ir_num,
1692 dest->s6_addr32[0], dest->s6_addr32[1],
1693 dest->s6_addr32[2], dest->s6_addr32[3],
1694 ntohs(inet_rsk(req)->ir_rmt_port),
1696 0, 0, /* could print option size, but that is af dependent. */
1697 1, /* timers active (only the expire timer) */
1698 jiffies_to_clock_t(ttd),
1700 from_kuid_munged(seq_user_ns(seq),
1701 sock_i_uid(req->rsk_listener)),
1702 0, /* non standard timer */
1703 0, /* open_requests have no inode */
1707 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1709 const struct in6_addr *dest, *src;
1712 unsigned long timer_expires;
1713 const struct inet_sock *inet = inet_sk(sp);
1714 const struct tcp_sock *tp = tcp_sk(sp);
1715 const struct inet_connection_sock *icsk = inet_csk(sp);
1716 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1720 dest = &sp->sk_v6_daddr;
1721 src = &sp->sk_v6_rcv_saddr;
1722 destp = ntohs(inet->inet_dport);
1723 srcp = ntohs(inet->inet_sport);
1725 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1726 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
1727 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1729 timer_expires = icsk->icsk_timeout;
1730 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1732 timer_expires = icsk->icsk_timeout;
1733 } else if (timer_pending(&sp->sk_timer)) {
1735 timer_expires = sp->sk_timer.expires;
1738 timer_expires = jiffies;
1741 state = sk_state_load(sp);
1742 if (state == TCP_LISTEN)
1743 rx_queue = sp->sk_ack_backlog;
1745 /* Because we don't lock the socket,
1746 * we might find a transient negative value.
1748 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1751 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1752 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1754 src->s6_addr32[0], src->s6_addr32[1],
1755 src->s6_addr32[2], src->s6_addr32[3], srcp,
1756 dest->s6_addr32[0], dest->s6_addr32[1],
1757 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1759 tp->write_seq - tp->snd_una,
1762 jiffies_delta_to_clock_t(timer_expires - jiffies),
1763 icsk->icsk_retransmits,
1764 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1765 icsk->icsk_probes_out,
1767 atomic_read(&sp->sk_refcnt), sp,
1768 jiffies_to_clock_t(icsk->icsk_rto),
1769 jiffies_to_clock_t(icsk->icsk_ack.ato),
1770 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1772 state == TCP_LISTEN ?
1773 fastopenq->max_qlen :
1774 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1778 static void get_timewait6_sock(struct seq_file *seq,
1779 struct inet_timewait_sock *tw, int i)
1781 long delta = tw->tw_timer.expires - jiffies;
1782 const struct in6_addr *dest, *src;
1785 dest = &tw->tw_v6_daddr;
1786 src = &tw->tw_v6_rcv_saddr;
1787 destp = ntohs(tw->tw_dport);
1788 srcp = ntohs(tw->tw_sport);
1791 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1792 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1794 src->s6_addr32[0], src->s6_addr32[1],
1795 src->s6_addr32[2], src->s6_addr32[3], srcp,
1796 dest->s6_addr32[0], dest->s6_addr32[1],
1797 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1798 tw->tw_substate, 0, 0,
1799 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1800 atomic_read(&tw->tw_refcnt), tw);
1803 static int tcp6_seq_show(struct seq_file *seq, void *v)
1805 struct tcp_iter_state *st;
1806 struct sock *sk = v;
1808 if (v == SEQ_START_TOKEN) {
1813 "st tx_queue rx_queue tr tm->when retrnsmt"
1814 " uid timeout inode\n");
1819 if (sk->sk_state == TCP_TIME_WAIT)
1820 get_timewait6_sock(seq, v, st->num);
1821 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1822 get_openreq6(seq, v, st->num);
1824 get_tcp6_sock(seq, v, st->num);
1829 static const struct file_operations tcp6_afinfo_seq_fops = {
1830 .owner = THIS_MODULE,
1831 .open = tcp_seq_open,
1833 .llseek = seq_lseek,
1834 .release = seq_release_net
1837 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1840 .seq_fops = &tcp6_afinfo_seq_fops,
1842 .show = tcp6_seq_show,
1846 int __net_init tcp6_proc_init(struct net *net)
1848 return tcp_proc_register(net, &tcp6_seq_afinfo);
1851 void tcp6_proc_exit(struct net *net)
1853 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1857 static void tcp_v6_clear_sk(struct sock *sk, int size)
1859 struct inet_sock *inet = inet_sk(sk);
1861 /* we do not want to clear pinet6 field, because of RCU lookups */
1862 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1864 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1865 memset(&inet->pinet6 + 1, 0, size);
1868 struct proto tcpv6_prot = {
1870 .owner = THIS_MODULE,
1872 .connect = tcp_v6_connect,
1873 .disconnect = tcp_disconnect,
1874 .accept = inet_csk_accept,
1876 .init = tcp_v6_init_sock,
1877 .destroy = tcp_v6_destroy_sock,
1878 .shutdown = tcp_shutdown,
1879 .setsockopt = tcp_setsockopt,
1880 .getsockopt = tcp_getsockopt,
1881 .recvmsg = tcp_recvmsg,
1882 .sendmsg = tcp_sendmsg,
1883 .sendpage = tcp_sendpage,
1884 .backlog_rcv = tcp_v6_do_rcv,
1885 .release_cb = tcp_release_cb,
1887 .unhash = inet_unhash,
1888 .get_port = inet_csk_get_port,
1889 .enter_memory_pressure = tcp_enter_memory_pressure,
1890 .stream_memory_free = tcp_stream_memory_free,
1891 .sockets_allocated = &tcp_sockets_allocated,
1892 .memory_allocated = &tcp_memory_allocated,
1893 .memory_pressure = &tcp_memory_pressure,
1894 .orphan_count = &tcp_orphan_count,
1895 .sysctl_mem = sysctl_tcp_mem,
1896 .sysctl_wmem = sysctl_tcp_wmem,
1897 .sysctl_rmem = sysctl_tcp_rmem,
1898 .max_header = MAX_TCP_HEADER,
1899 .obj_size = sizeof(struct tcp6_sock),
1900 .slab_flags = SLAB_DESTROY_BY_RCU,
1901 .twsk_prot = &tcp6_timewait_sock_ops,
1902 .rsk_prot = &tcp6_request_sock_ops,
1903 .h.hashinfo = &tcp_hashinfo,
1904 .no_autobind = true,
1905 #ifdef CONFIG_COMPAT
1906 .compat_setsockopt = compat_tcp_setsockopt,
1907 .compat_getsockopt = compat_tcp_getsockopt,
1909 #ifdef CONFIG_MEMCG_KMEM
1910 .proto_cgroup = tcp_proto_cgroup,
1912 .clear_sk = tcp_v6_clear_sk,
1913 .diag_destroy = tcp_abort,
1916 static const struct inet6_protocol tcpv6_protocol = {
1917 .early_demux = tcp_v6_early_demux,
1918 .handler = tcp_v6_rcv,
1919 .err_handler = tcp_v6_err,
1920 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1923 static struct inet_protosw tcpv6_protosw = {
1924 .type = SOCK_STREAM,
1925 .protocol = IPPROTO_TCP,
1926 .prot = &tcpv6_prot,
1927 .ops = &inet6_stream_ops,
1928 .flags = INET_PROTOSW_PERMANENT |
1932 static int __net_init tcpv6_net_init(struct net *net)
1934 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1935 SOCK_RAW, IPPROTO_TCP, net);
1938 static void __net_exit tcpv6_net_exit(struct net *net)
1940 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1943 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1945 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1948 static struct pernet_operations tcpv6_net_ops = {
1949 .init = tcpv6_net_init,
1950 .exit = tcpv6_net_exit,
1951 .exit_batch = tcpv6_net_exit_batch,
1954 int __init tcpv6_init(void)
1958 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1962 /* register inet6 protocol */
1963 ret = inet6_register_protosw(&tcpv6_protosw);
1965 goto out_tcpv6_protocol;
1967 ret = register_pernet_subsys(&tcpv6_net_ops);
1969 goto out_tcpv6_protosw;
1974 inet6_unregister_protosw(&tcpv6_protosw);
1976 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1980 void tcpv6_exit(void)
1982 unregister_pernet_subsys(&tcpv6_net_ops);
1983 inet6_unregister_protosw(&tcpv6_protosw);
1984 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);