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.
936 * The window field (SEG.WND) of every outgoing segment, with the
937 * exception of <SYN> segments, MUST be right-shifted by
938 * Rcv.Wind.Shift bits:
940 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
941 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
942 tcp_rsk(req)->rcv_nxt,
943 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
944 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
945 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
950 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
952 #ifdef CONFIG_SYN_COOKIES
953 const struct tcphdr *th = tcp_hdr(skb);
956 sk = cookie_v6_check(sk, skb);
961 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
963 if (skb->protocol == htons(ETH_P_IP))
964 return tcp_v4_conn_request(sk, skb);
966 if (!ipv6_unicast_destination(skb))
969 return tcp_conn_request(&tcp6_request_sock_ops,
970 &tcp_request_sock_ipv6_ops, sk, skb);
973 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
974 return 0; /* don't send reset */
977 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
978 struct request_sock *req,
979 struct dst_entry *dst,
980 struct request_sock *req_unhash,
983 struct inet_request_sock *ireq;
984 struct ipv6_pinfo *newnp;
985 const struct ipv6_pinfo *np = inet6_sk(sk);
986 struct ipv6_txoptions *opt;
987 struct tcp6_sock *newtcp6sk;
988 struct inet_sock *newinet;
989 struct tcp_sock *newtp;
991 #ifdef CONFIG_TCP_MD5SIG
992 struct tcp_md5sig_key *key;
996 if (skb->protocol == htons(ETH_P_IP)) {
1001 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1002 req_unhash, own_req);
1007 newtcp6sk = (struct tcp6_sock *)newsk;
1008 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1010 newinet = inet_sk(newsk);
1011 newnp = inet6_sk(newsk);
1012 newtp = tcp_sk(newsk);
1014 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1016 newnp->saddr = newsk->sk_v6_rcv_saddr;
1018 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1019 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1020 #ifdef CONFIG_TCP_MD5SIG
1021 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1024 newnp->ipv6_ac_list = NULL;
1025 newnp->ipv6_fl_list = NULL;
1026 newnp->pktoptions = NULL;
1028 newnp->mcast_oif = tcp_v6_iif(skb);
1029 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1030 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1032 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1035 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1036 * here, tcp_create_openreq_child now does this for us, see the comment in
1037 * that function for the gory details. -acme
1040 /* It is tricky place. Until this moment IPv4 tcp
1041 worked with IPv6 icsk.icsk_af_ops.
1044 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1049 ireq = inet_rsk(req);
1051 if (sk_acceptq_is_full(sk))
1055 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1060 newsk = tcp_create_openreq_child(sk, req, skb);
1065 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1066 * count here, tcp_create_openreq_child now does this for us, see the
1067 * comment in that function for the gory details. -acme
1070 newsk->sk_gso_type = SKB_GSO_TCPV6;
1071 ip6_dst_store(newsk, dst, NULL, NULL);
1072 inet6_sk_rx_dst_set(newsk, skb);
1074 newtcp6sk = (struct tcp6_sock *)newsk;
1075 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1077 newtp = tcp_sk(newsk);
1078 newinet = inet_sk(newsk);
1079 newnp = inet6_sk(newsk);
1081 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1083 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1084 newnp->saddr = ireq->ir_v6_loc_addr;
1085 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1086 newsk->sk_bound_dev_if = ireq->ir_iif;
1088 /* Now IPv6 options...
1090 First: no IPv4 options.
1092 newinet->inet_opt = NULL;
1093 newnp->ipv6_ac_list = NULL;
1094 newnp->ipv6_fl_list = NULL;
1097 newnp->rxopt.all = np->rxopt.all;
1099 newnp->pktoptions = NULL;
1101 newnp->mcast_oif = tcp_v6_iif(skb);
1102 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1103 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1105 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1107 /* Clone native IPv6 options from listening socket (if any)
1109 Yes, keeping reference count would be much more clever,
1110 but we make one more one thing there: reattach optmem
1113 opt = rcu_dereference(np->opt);
1115 opt = ipv6_dup_options(newsk, opt);
1116 RCU_INIT_POINTER(newnp->opt, opt);
1118 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1120 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1123 tcp_ca_openreq_child(newsk, dst);
1125 tcp_sync_mss(newsk, dst_mtu(dst));
1126 newtp->advmss = dst_metric_advmss(dst);
1127 if (tcp_sk(sk)->rx_opt.user_mss &&
1128 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1129 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1131 tcp_initialize_rcv_mss(newsk);
1133 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1134 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1136 #ifdef CONFIG_TCP_MD5SIG
1137 /* Copy over the MD5 key from the original socket */
1138 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1140 /* We're using one, so create a matching key
1141 * on the newsk structure. If we fail to get
1142 * memory, then we end up not copying the key
1145 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1146 AF_INET6, key->key, key->keylen,
1147 sk_gfp_atomic(sk, GFP_ATOMIC));
1151 if (__inet_inherit_port(sk, newsk) < 0) {
1152 inet_csk_prepare_forced_close(newsk);
1156 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1158 tcp_move_syn(newtp, req);
1160 /* Clone pktoptions received with SYN, if we own the req */
1161 if (ireq->pktopts) {
1162 newnp->pktoptions = skb_clone(ireq->pktopts,
1163 sk_gfp_atomic(sk, GFP_ATOMIC));
1164 consume_skb(ireq->pktopts);
1165 ireq->pktopts = NULL;
1166 if (newnp->pktoptions)
1167 skb_set_owner_r(newnp->pktoptions, newsk);
1174 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1178 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1182 /* The socket must have it's spinlock held when we get
1183 * here, unless it is a TCP_LISTEN socket.
1185 * We have a potential double-lock case here, so even when
1186 * doing backlog processing we use the BH locking scheme.
1187 * This is because we cannot sleep with the original spinlock
1190 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1192 struct ipv6_pinfo *np = inet6_sk(sk);
1193 struct tcp_sock *tp;
1194 struct sk_buff *opt_skb = NULL;
1196 /* Imagine: socket is IPv6. IPv4 packet arrives,
1197 goes to IPv4 receive handler and backlogged.
1198 From backlog it always goes here. Kerboom...
1199 Fortunately, tcp_rcv_established and rcv_established
1200 handle them correctly, but it is not case with
1201 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1204 if (skb->protocol == htons(ETH_P_IP))
1205 return tcp_v4_do_rcv(sk, skb);
1207 if (sk_filter(sk, skb))
1211 * socket locking is here for SMP purposes as backlog rcv
1212 * is currently called with bh processing disabled.
1215 /* Do Stevens' IPV6_PKTOPTIONS.
1217 Yes, guys, it is the only place in our code, where we
1218 may make it not affecting IPv4.
1219 The rest of code is protocol independent,
1220 and I do not like idea to uglify IPv4.
1222 Actually, all the idea behind IPV6_PKTOPTIONS
1223 looks not very well thought. For now we latch
1224 options, received in the last packet, enqueued
1225 by tcp. Feel free to propose better solution.
1229 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1231 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1232 struct dst_entry *dst = sk->sk_rx_dst;
1234 sock_rps_save_rxhash(sk, skb);
1235 sk_mark_napi_id(sk, skb);
1237 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1238 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1240 sk->sk_rx_dst = NULL;
1244 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1246 goto ipv6_pktoptions;
1250 if (tcp_checksum_complete(skb))
1253 if (sk->sk_state == TCP_LISTEN) {
1254 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1260 sock_rps_save_rxhash(nsk, skb);
1261 sk_mark_napi_id(nsk, skb);
1262 if (tcp_child_process(sk, nsk, skb))
1265 __kfree_skb(opt_skb);
1269 sock_rps_save_rxhash(sk, skb);
1271 if (tcp_rcv_state_process(sk, skb))
1274 goto ipv6_pktoptions;
1278 tcp_v6_send_reset(sk, skb);
1281 __kfree_skb(opt_skb);
1285 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1286 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1291 /* Do you ask, what is it?
1293 1. skb was enqueued by tcp.
1294 2. skb is added to tail of read queue, rather than out of order.
1295 3. socket is not in passive state.
1296 4. Finally, it really contains options, which user wants to receive.
1299 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1300 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1301 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1302 np->mcast_oif = tcp_v6_iif(opt_skb);
1303 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1304 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1305 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1306 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1308 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1309 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1310 skb_set_owner_r(opt_skb, sk);
1311 opt_skb = xchg(&np->pktoptions, opt_skb);
1313 __kfree_skb(opt_skb);
1314 opt_skb = xchg(&np->pktoptions, NULL);
1322 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1323 const struct tcphdr *th)
1325 /* This is tricky: we move IP6CB at its correct location into
1326 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1327 * _decode_session6() uses IP6CB().
1328 * barrier() makes sure compiler won't play aliasing games.
1330 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1331 sizeof(struct inet6_skb_parm));
1334 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1335 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1336 skb->len - th->doff*4);
1337 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1338 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1339 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1340 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1341 TCP_SKB_CB(skb)->sacked = 0;
1344 static void tcp_v6_restore_cb(struct sk_buff *skb)
1346 /* We need to move header back to the beginning if xfrm6_policy_check()
1347 * and tcp_v6_fill_cb() are going to be called again.
1349 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1350 sizeof(struct inet6_skb_parm));
1353 static int tcp_v6_rcv(struct sk_buff *skb)
1355 const struct tcphdr *th;
1356 const struct ipv6hdr *hdr;
1359 struct net *net = dev_net(skb->dev);
1361 if (skb->pkt_type != PACKET_HOST)
1365 * Count it even if it's bad.
1367 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1369 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1374 if (th->doff < sizeof(struct tcphdr)/4)
1376 if (!pskb_may_pull(skb, th->doff*4))
1379 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1383 hdr = ipv6_hdr(skb);
1386 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest,
1392 if (sk->sk_state == TCP_TIME_WAIT)
1395 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1396 struct request_sock *req = inet_reqsk(sk);
1399 sk = req->rsk_listener;
1400 tcp_v6_fill_cb(skb, hdr, th);
1401 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1405 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1406 inet_csk_reqsk_queue_drop_and_put(sk, req);
1410 nsk = tcp_check_req(sk, skb, req, false);
1413 goto discard_and_relse;
1417 tcp_v6_restore_cb(skb);
1418 } else if (tcp_child_process(sk, nsk, skb)) {
1419 tcp_v6_send_reset(nsk, skb);
1420 goto discard_and_relse;
1426 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1427 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1428 goto discard_and_relse;
1431 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1432 goto discard_and_relse;
1434 tcp_v6_fill_cb(skb, hdr, th);
1436 if (tcp_v6_inbound_md5_hash(sk, skb))
1437 goto discard_and_relse;
1439 if (sk_filter(sk, skb))
1440 goto discard_and_relse;
1444 if (sk->sk_state == TCP_LISTEN) {
1445 ret = tcp_v6_do_rcv(sk, skb);
1446 goto put_and_return;
1449 sk_incoming_cpu_update(sk);
1451 bh_lock_sock_nested(sk);
1452 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1454 if (!sock_owned_by_user(sk)) {
1455 if (!tcp_prequeue(sk, skb))
1456 ret = tcp_v6_do_rcv(sk, skb);
1457 } else if (unlikely(sk_add_backlog(sk, skb,
1458 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1460 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1461 goto discard_and_relse;
1467 return ret ? -1 : 0;
1470 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1473 tcp_v6_fill_cb(skb, hdr, th);
1475 if (tcp_checksum_complete(skb)) {
1477 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1479 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1481 tcp_v6_send_reset(NULL, skb);
1493 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1494 inet_twsk_put(inet_twsk(sk));
1498 tcp_v6_fill_cb(skb, hdr, th);
1500 if (tcp_checksum_complete(skb)) {
1501 inet_twsk_put(inet_twsk(sk));
1505 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1510 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1511 &ipv6_hdr(skb)->saddr, th->source,
1512 &ipv6_hdr(skb)->daddr,
1513 ntohs(th->dest), tcp_v6_iif(skb));
1515 struct inet_timewait_sock *tw = inet_twsk(sk);
1516 inet_twsk_deschedule_put(tw);
1518 tcp_v6_restore_cb(skb);
1521 /* Fall through to ACK */
1524 tcp_v6_timewait_ack(sk, skb);
1527 tcp_v6_restore_cb(skb);
1529 case TCP_TW_SUCCESS:
1535 static void tcp_v6_early_demux(struct sk_buff *skb)
1537 const struct ipv6hdr *hdr;
1538 const struct tcphdr *th;
1541 if (skb->pkt_type != PACKET_HOST)
1544 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1547 hdr = ipv6_hdr(skb);
1550 if (th->doff < sizeof(struct tcphdr) / 4)
1553 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1554 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1555 &hdr->saddr, th->source,
1556 &hdr->daddr, ntohs(th->dest),
1560 skb->destructor = sock_edemux;
1561 if (sk_fullsock(sk)) {
1562 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1565 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1567 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1568 skb_dst_set_noref(skb, dst);
1573 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1574 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1575 .twsk_unique = tcp_twsk_unique,
1576 .twsk_destructor = tcp_twsk_destructor,
1579 static const struct inet_connection_sock_af_ops ipv6_specific = {
1580 .queue_xmit = inet6_csk_xmit,
1581 .send_check = tcp_v6_send_check,
1582 .rebuild_header = inet6_sk_rebuild_header,
1583 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1584 .conn_request = tcp_v6_conn_request,
1585 .syn_recv_sock = tcp_v6_syn_recv_sock,
1586 .net_header_len = sizeof(struct ipv6hdr),
1587 .net_frag_header_len = sizeof(struct frag_hdr),
1588 .setsockopt = ipv6_setsockopt,
1589 .getsockopt = ipv6_getsockopt,
1590 .addr2sockaddr = inet6_csk_addr2sockaddr,
1591 .sockaddr_len = sizeof(struct sockaddr_in6),
1592 .bind_conflict = inet6_csk_bind_conflict,
1593 #ifdef CONFIG_COMPAT
1594 .compat_setsockopt = compat_ipv6_setsockopt,
1595 .compat_getsockopt = compat_ipv6_getsockopt,
1597 .mtu_reduced = tcp_v6_mtu_reduced,
1600 #ifdef CONFIG_TCP_MD5SIG
1601 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1602 .md5_lookup = tcp_v6_md5_lookup,
1603 .calc_md5_hash = tcp_v6_md5_hash_skb,
1604 .md5_parse = tcp_v6_parse_md5_keys,
1609 * TCP over IPv4 via INET6 API
1611 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1612 .queue_xmit = ip_queue_xmit,
1613 .send_check = tcp_v4_send_check,
1614 .rebuild_header = inet_sk_rebuild_header,
1615 .sk_rx_dst_set = inet_sk_rx_dst_set,
1616 .conn_request = tcp_v6_conn_request,
1617 .syn_recv_sock = tcp_v6_syn_recv_sock,
1618 .net_header_len = sizeof(struct iphdr),
1619 .setsockopt = ipv6_setsockopt,
1620 .getsockopt = ipv6_getsockopt,
1621 .addr2sockaddr = inet6_csk_addr2sockaddr,
1622 .sockaddr_len = sizeof(struct sockaddr_in6),
1623 .bind_conflict = inet6_csk_bind_conflict,
1624 #ifdef CONFIG_COMPAT
1625 .compat_setsockopt = compat_ipv6_setsockopt,
1626 .compat_getsockopt = compat_ipv6_getsockopt,
1628 .mtu_reduced = tcp_v4_mtu_reduced,
1631 #ifdef CONFIG_TCP_MD5SIG
1632 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1633 .md5_lookup = tcp_v4_md5_lookup,
1634 .calc_md5_hash = tcp_v4_md5_hash_skb,
1635 .md5_parse = tcp_v6_parse_md5_keys,
1639 /* NOTE: A lot of things set to zero explicitly by call to
1640 * sk_alloc() so need not be done here.
1642 static int tcp_v6_init_sock(struct sock *sk)
1644 struct inet_connection_sock *icsk = inet_csk(sk);
1648 icsk->icsk_af_ops = &ipv6_specific;
1650 #ifdef CONFIG_TCP_MD5SIG
1651 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1657 static void tcp_v6_destroy_sock(struct sock *sk)
1659 tcp_v4_destroy_sock(sk);
1660 inet6_destroy_sock(sk);
1663 #ifdef CONFIG_PROC_FS
1664 /* Proc filesystem TCPv6 sock list dumping. */
1665 static void get_openreq6(struct seq_file *seq,
1666 const struct request_sock *req, int i)
1668 long ttd = req->rsk_timer.expires - jiffies;
1669 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1670 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1676 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1677 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1679 src->s6_addr32[0], src->s6_addr32[1],
1680 src->s6_addr32[2], src->s6_addr32[3],
1681 inet_rsk(req)->ir_num,
1682 dest->s6_addr32[0], dest->s6_addr32[1],
1683 dest->s6_addr32[2], dest->s6_addr32[3],
1684 ntohs(inet_rsk(req)->ir_rmt_port),
1686 0, 0, /* could print option size, but that is af dependent. */
1687 1, /* timers active (only the expire timer) */
1688 jiffies_to_clock_t(ttd),
1690 from_kuid_munged(seq_user_ns(seq),
1691 sock_i_uid(req->rsk_listener)),
1692 0, /* non standard timer */
1693 0, /* open_requests have no inode */
1697 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1699 const struct in6_addr *dest, *src;
1702 unsigned long timer_expires;
1703 const struct inet_sock *inet = inet_sk(sp);
1704 const struct tcp_sock *tp = tcp_sk(sp);
1705 const struct inet_connection_sock *icsk = inet_csk(sp);
1706 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1710 dest = &sp->sk_v6_daddr;
1711 src = &sp->sk_v6_rcv_saddr;
1712 destp = ntohs(inet->inet_dport);
1713 srcp = ntohs(inet->inet_sport);
1715 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1716 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
1717 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1719 timer_expires = icsk->icsk_timeout;
1720 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1722 timer_expires = icsk->icsk_timeout;
1723 } else if (timer_pending(&sp->sk_timer)) {
1725 timer_expires = sp->sk_timer.expires;
1728 timer_expires = jiffies;
1731 state = sk_state_load(sp);
1732 if (state == TCP_LISTEN)
1733 rx_queue = sp->sk_ack_backlog;
1735 /* Because we don't lock the socket,
1736 * we might find a transient negative value.
1738 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1741 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1742 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1744 src->s6_addr32[0], src->s6_addr32[1],
1745 src->s6_addr32[2], src->s6_addr32[3], srcp,
1746 dest->s6_addr32[0], dest->s6_addr32[1],
1747 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1749 tp->write_seq - tp->snd_una,
1752 jiffies_delta_to_clock_t(timer_expires - jiffies),
1753 icsk->icsk_retransmits,
1754 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1755 icsk->icsk_probes_out,
1757 atomic_read(&sp->sk_refcnt), sp,
1758 jiffies_to_clock_t(icsk->icsk_rto),
1759 jiffies_to_clock_t(icsk->icsk_ack.ato),
1760 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1762 state == TCP_LISTEN ?
1763 fastopenq->max_qlen :
1764 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1768 static void get_timewait6_sock(struct seq_file *seq,
1769 struct inet_timewait_sock *tw, int i)
1771 long delta = tw->tw_timer.expires - jiffies;
1772 const struct in6_addr *dest, *src;
1775 dest = &tw->tw_v6_daddr;
1776 src = &tw->tw_v6_rcv_saddr;
1777 destp = ntohs(tw->tw_dport);
1778 srcp = ntohs(tw->tw_sport);
1781 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1782 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1784 src->s6_addr32[0], src->s6_addr32[1],
1785 src->s6_addr32[2], src->s6_addr32[3], srcp,
1786 dest->s6_addr32[0], dest->s6_addr32[1],
1787 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1788 tw->tw_substate, 0, 0,
1789 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1790 atomic_read(&tw->tw_refcnt), tw);
1793 static int tcp6_seq_show(struct seq_file *seq, void *v)
1795 struct tcp_iter_state *st;
1796 struct sock *sk = v;
1798 if (v == SEQ_START_TOKEN) {
1803 "st tx_queue rx_queue tr tm->when retrnsmt"
1804 " uid timeout inode\n");
1809 if (sk->sk_state == TCP_TIME_WAIT)
1810 get_timewait6_sock(seq, v, st->num);
1811 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1812 get_openreq6(seq, v, st->num);
1814 get_tcp6_sock(seq, v, st->num);
1819 static const struct file_operations tcp6_afinfo_seq_fops = {
1820 .owner = THIS_MODULE,
1821 .open = tcp_seq_open,
1823 .llseek = seq_lseek,
1824 .release = seq_release_net
1827 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1830 .seq_fops = &tcp6_afinfo_seq_fops,
1832 .show = tcp6_seq_show,
1836 int __net_init tcp6_proc_init(struct net *net)
1838 return tcp_proc_register(net, &tcp6_seq_afinfo);
1841 void tcp6_proc_exit(struct net *net)
1843 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1847 static void tcp_v6_clear_sk(struct sock *sk, int size)
1849 struct inet_sock *inet = inet_sk(sk);
1851 /* we do not want to clear pinet6 field, because of RCU lookups */
1852 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1854 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1855 memset(&inet->pinet6 + 1, 0, size);
1858 struct proto tcpv6_prot = {
1860 .owner = THIS_MODULE,
1862 .connect = tcp_v6_connect,
1863 .disconnect = tcp_disconnect,
1864 .accept = inet_csk_accept,
1866 .init = tcp_v6_init_sock,
1867 .destroy = tcp_v6_destroy_sock,
1868 .shutdown = tcp_shutdown,
1869 .setsockopt = tcp_setsockopt,
1870 .getsockopt = tcp_getsockopt,
1871 .recvmsg = tcp_recvmsg,
1872 .sendmsg = tcp_sendmsg,
1873 .sendpage = tcp_sendpage,
1874 .backlog_rcv = tcp_v6_do_rcv,
1875 .release_cb = tcp_release_cb,
1877 .unhash = inet_unhash,
1878 .get_port = inet_csk_get_port,
1879 .enter_memory_pressure = tcp_enter_memory_pressure,
1880 .stream_memory_free = tcp_stream_memory_free,
1881 .sockets_allocated = &tcp_sockets_allocated,
1882 .memory_allocated = &tcp_memory_allocated,
1883 .memory_pressure = &tcp_memory_pressure,
1884 .orphan_count = &tcp_orphan_count,
1885 .sysctl_mem = sysctl_tcp_mem,
1886 .sysctl_wmem = sysctl_tcp_wmem,
1887 .sysctl_rmem = sysctl_tcp_rmem,
1888 .max_header = MAX_TCP_HEADER,
1889 .obj_size = sizeof(struct tcp6_sock),
1890 .slab_flags = SLAB_DESTROY_BY_RCU,
1891 .twsk_prot = &tcp6_timewait_sock_ops,
1892 .rsk_prot = &tcp6_request_sock_ops,
1893 .h.hashinfo = &tcp_hashinfo,
1894 .no_autobind = true,
1895 #ifdef CONFIG_COMPAT
1896 .compat_setsockopt = compat_tcp_setsockopt,
1897 .compat_getsockopt = compat_tcp_getsockopt,
1899 #ifdef CONFIG_MEMCG_KMEM
1900 .proto_cgroup = tcp_proto_cgroup,
1902 .clear_sk = tcp_v6_clear_sk,
1905 static const struct inet6_protocol tcpv6_protocol = {
1906 .early_demux = tcp_v6_early_demux,
1907 .handler = tcp_v6_rcv,
1908 .err_handler = tcp_v6_err,
1909 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1912 static struct inet_protosw tcpv6_protosw = {
1913 .type = SOCK_STREAM,
1914 .protocol = IPPROTO_TCP,
1915 .prot = &tcpv6_prot,
1916 .ops = &inet6_stream_ops,
1917 .flags = INET_PROTOSW_PERMANENT |
1921 static int __net_init tcpv6_net_init(struct net *net)
1923 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1924 SOCK_RAW, IPPROTO_TCP, net);
1927 static void __net_exit tcpv6_net_exit(struct net *net)
1929 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1932 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1934 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1937 static struct pernet_operations tcpv6_net_ops = {
1938 .init = tcpv6_net_init,
1939 .exit = tcpv6_net_exit,
1940 .exit_batch = tcpv6_net_exit_batch,
1943 int __init tcpv6_init(void)
1947 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1951 /* register inet6 protocol */
1952 ret = inet6_register_protosw(&tcpv6_protosw);
1954 goto out_tcpv6_protocol;
1956 ret = register_pernet_subsys(&tcpv6_net_ops);
1958 goto out_tcpv6_protosw;
1963 inet6_unregister_protosw(&tcpv6_protosw);
1965 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1969 void tcpv6_exit(void)
1971 unregister_pernet_subsys(&tcpv6_net_ops);
1972 inet6_unregister_protosw(&tcpv6_protosw);
1973 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);