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/netdma.h>
63 #include <net/inet_common.h>
64 #include <net/secure_seq.h>
65 #include <net/tcp_memcontrol.h>
66 #include <net/busy_poll.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
71 #include <linux/crypto.h>
72 #include <linux/scatterlist.h>
74 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
75 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
76 struct request_sock *req);
78 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
80 static const struct inet_connection_sock_af_ops ipv6_mapped;
81 static const struct inet_connection_sock_af_ops ipv6_specific;
82 #ifdef CONFIG_TCP_MD5SIG
83 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
84 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
86 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
87 const struct in6_addr *addr)
93 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
95 struct dst_entry *dst = skb_dst(skb);
96 const struct rt6_info *rt = (const struct rt6_info *)dst;
100 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
102 inet6_sk(sk)->rx_dst_cookie = rt->rt6i_node->fn_sernum;
105 static void tcp_v6_hash(struct sock *sk)
107 if (sk->sk_state != TCP_CLOSE) {
108 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
113 __inet6_hash(sk, NULL);
118 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
120 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
121 ipv6_hdr(skb)->saddr.s6_addr32,
123 tcp_hdr(skb)->source);
126 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
129 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
130 struct inet_sock *inet = inet_sk(sk);
131 struct inet_connection_sock *icsk = inet_csk(sk);
132 struct ipv6_pinfo *np = inet6_sk(sk);
133 struct tcp_sock *tp = tcp_sk(sk);
134 struct in6_addr *saddr = NULL, *final_p, final;
137 struct dst_entry *dst;
141 if (addr_len < SIN6_LEN_RFC2133)
144 if (usin->sin6_family != AF_INET6)
145 return -EAFNOSUPPORT;
147 memset(&fl6, 0, sizeof(fl6));
150 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
151 IP6_ECN_flow_init(fl6.flowlabel);
152 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
153 struct ip6_flowlabel *flowlabel;
154 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
155 if (flowlabel == NULL)
157 fl6_sock_release(flowlabel);
162 * connect() to INADDR_ANY means loopback (BSD'ism).
165 if (ipv6_addr_any(&usin->sin6_addr))
166 usin->sin6_addr.s6_addr[15] = 0x1;
168 addr_type = ipv6_addr_type(&usin->sin6_addr);
170 if (addr_type & IPV6_ADDR_MULTICAST)
173 if (addr_type&IPV6_ADDR_LINKLOCAL) {
174 if (addr_len >= sizeof(struct sockaddr_in6) &&
175 usin->sin6_scope_id) {
176 /* If interface is set while binding, indices
179 if (sk->sk_bound_dev_if &&
180 sk->sk_bound_dev_if != usin->sin6_scope_id)
183 sk->sk_bound_dev_if = usin->sin6_scope_id;
186 /* Connect to link-local address requires an interface */
187 if (!sk->sk_bound_dev_if)
191 if (tp->rx_opt.ts_recent_stamp &&
192 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
198 sk->sk_v6_daddr = usin->sin6_addr;
199 np->flow_label = fl6.flowlabel;
207 if (addr_type == IPV6_ADDR_MAPPED) {
208 u32 exthdrlen = icsk->icsk_ext_hdr_len;
209 struct sockaddr_in sin;
211 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
213 if (__ipv6_only_sock(sk))
216 sin.sin_family = AF_INET;
217 sin.sin_port = usin->sin6_port;
218 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
220 icsk->icsk_af_ops = &ipv6_mapped;
221 sk->sk_backlog_rcv = tcp_v4_do_rcv;
222 #ifdef CONFIG_TCP_MD5SIG
223 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
226 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
229 icsk->icsk_ext_hdr_len = exthdrlen;
230 icsk->icsk_af_ops = &ipv6_specific;
231 sk->sk_backlog_rcv = tcp_v6_do_rcv;
232 #ifdef CONFIG_TCP_MD5SIG
233 tp->af_specific = &tcp_sock_ipv6_specific;
237 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
238 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
239 &sk->sk_v6_rcv_saddr);
245 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
246 saddr = &sk->sk_v6_rcv_saddr;
248 fl6.flowi6_proto = IPPROTO_TCP;
249 fl6.daddr = sk->sk_v6_daddr;
250 fl6.saddr = saddr ? *saddr : np->saddr;
251 fl6.flowi6_oif = sk->sk_bound_dev_if;
252 fl6.flowi6_mark = sk->sk_mark;
253 fl6.fl6_dport = usin->sin6_port;
254 fl6.fl6_sport = inet->inet_sport;
256 final_p = fl6_update_dst(&fl6, np->opt, &final);
258 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
260 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
268 sk->sk_v6_rcv_saddr = *saddr;
271 /* set the source address */
273 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
275 sk->sk_gso_type = SKB_GSO_TCPV6;
276 __ip6_dst_store(sk, dst, NULL, NULL);
278 rt = (struct rt6_info *) dst;
279 if (tcp_death_row.sysctl_tw_recycle &&
280 !tp->rx_opt.ts_recent_stamp &&
281 ipv6_addr_equal(&rt->rt6i_dst.addr, &sk->sk_v6_daddr))
282 tcp_fetch_timewait_stamp(sk, dst);
284 icsk->icsk_ext_hdr_len = 0;
286 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
289 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
291 inet->inet_dport = usin->sin6_port;
293 tcp_set_state(sk, TCP_SYN_SENT);
294 err = inet6_hash_connect(&tcp_death_row, sk);
298 if (!tp->write_seq && likely(!tp->repair))
299 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
300 sk->sk_v6_daddr.s6_addr32,
304 err = tcp_connect(sk);
311 tcp_set_state(sk, TCP_CLOSE);
314 inet->inet_dport = 0;
315 sk->sk_route_caps = 0;
319 static void tcp_v6_mtu_reduced(struct sock *sk)
321 struct dst_entry *dst;
323 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
326 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
330 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
331 tcp_sync_mss(sk, dst_mtu(dst));
332 tcp_simple_retransmit(sk);
336 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
337 u8 type, u8 code, int offset, __be32 info)
339 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
340 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
341 struct ipv6_pinfo *np;
345 struct request_sock *fastopen;
347 struct net *net = dev_net(skb->dev);
349 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
350 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
353 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
358 if (sk->sk_state == TCP_TIME_WAIT) {
359 inet_twsk_put(inet_twsk(sk));
364 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
365 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
367 if (sk->sk_state == TCP_CLOSE)
370 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
371 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
376 seq = ntohl(th->seq);
377 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
378 fastopen = tp->fastopen_rsk;
379 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
380 if (sk->sk_state != TCP_LISTEN &&
381 !between(seq, snd_una, tp->snd_nxt)) {
382 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
388 if (type == NDISC_REDIRECT) {
389 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
392 dst->ops->redirect(dst, sk, skb);
396 if (type == ICMPV6_PKT_TOOBIG) {
397 /* We are not interested in TCP_LISTEN and open_requests
398 * (SYN-ACKs send out by Linux are always <576bytes so
399 * they should go through unfragmented).
401 if (sk->sk_state == TCP_LISTEN)
404 if (!ip6_sk_accept_pmtu(sk))
407 tp->mtu_info = ntohl(info);
408 if (!sock_owned_by_user(sk))
409 tcp_v6_mtu_reduced(sk);
410 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
416 icmpv6_err_convert(type, code, &err);
418 /* Might be for an request_sock */
419 switch (sk->sk_state) {
420 struct request_sock *req, **prev;
422 if (sock_owned_by_user(sk))
425 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
426 &hdr->saddr, inet6_iif(skb));
430 /* ICMPs are not backlogged, hence we cannot get
431 * an established socket here.
433 WARN_ON(req->sk != NULL);
435 if (seq != tcp_rsk(req)->snt_isn) {
436 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
440 inet_csk_reqsk_queue_drop(sk, req, prev);
441 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
446 /* Only in fast or simultaneous open. If a fast open socket is
447 * is already accepted it is treated as a connected one below.
449 if (fastopen && fastopen->sk == NULL)
452 if (!sock_owned_by_user(sk)) {
454 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
458 sk->sk_err_soft = err;
462 if (!sock_owned_by_user(sk) && np->recverr) {
464 sk->sk_error_report(sk);
466 sk->sk_err_soft = err;
474 static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
476 struct request_sock *req,
478 struct tcp_fastopen_cookie *foc)
480 struct inet_request_sock *ireq = inet_rsk(req);
481 struct ipv6_pinfo *np = inet6_sk(sk);
482 struct flowi6 *fl6 = &fl->u.ip6;
486 /* First, grab a route. */
487 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
490 skb = tcp_make_synack(sk, dst, req, foc);
493 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
494 &ireq->ir_v6_rmt_addr);
496 fl6->daddr = ireq->ir_v6_rmt_addr;
497 if (np->repflow && (ireq->pktopts != NULL))
498 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
500 skb_set_queue_mapping(skb, queue_mapping);
501 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
502 err = net_xmit_eval(err);
510 static void tcp_v6_reqsk_destructor(struct request_sock *req)
512 kfree_skb(inet_rsk(req)->pktopts);
515 #ifdef CONFIG_TCP_MD5SIG
516 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
517 const struct in6_addr *addr)
519 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
522 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
523 struct sock *addr_sk)
525 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
528 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
529 struct request_sock *req)
531 return tcp_v6_md5_do_lookup(sk, &inet_rsk(req)->ir_v6_rmt_addr);
534 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
537 struct tcp_md5sig cmd;
538 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
540 if (optlen < sizeof(cmd))
543 if (copy_from_user(&cmd, optval, sizeof(cmd)))
546 if (sin6->sin6_family != AF_INET6)
549 if (!cmd.tcpm_keylen) {
550 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
551 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
553 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
557 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
560 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
561 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
562 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
564 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
565 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
568 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
569 const struct in6_addr *daddr,
570 const struct in6_addr *saddr, int nbytes)
572 struct tcp6_pseudohdr *bp;
573 struct scatterlist sg;
575 bp = &hp->md5_blk.ip6;
576 /* 1. TCP pseudo-header (RFC2460) */
579 bp->protocol = cpu_to_be32(IPPROTO_TCP);
580 bp->len = cpu_to_be32(nbytes);
582 sg_init_one(&sg, bp, sizeof(*bp));
583 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
586 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
587 const struct in6_addr *daddr, struct in6_addr *saddr,
588 const struct tcphdr *th)
590 struct tcp_md5sig_pool *hp;
591 struct hash_desc *desc;
593 hp = tcp_get_md5sig_pool();
595 goto clear_hash_noput;
596 desc = &hp->md5_desc;
598 if (crypto_hash_init(desc))
600 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
602 if (tcp_md5_hash_header(hp, th))
604 if (tcp_md5_hash_key(hp, key))
606 if (crypto_hash_final(desc, md5_hash))
609 tcp_put_md5sig_pool();
613 tcp_put_md5sig_pool();
615 memset(md5_hash, 0, 16);
619 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
620 const struct sock *sk,
621 const struct request_sock *req,
622 const struct sk_buff *skb)
624 const struct in6_addr *saddr, *daddr;
625 struct tcp_md5sig_pool *hp;
626 struct hash_desc *desc;
627 const struct tcphdr *th = tcp_hdr(skb);
630 saddr = &inet6_sk(sk)->saddr;
631 daddr = &sk->sk_v6_daddr;
633 saddr = &inet_rsk(req)->ir_v6_loc_addr;
634 daddr = &inet_rsk(req)->ir_v6_rmt_addr;
636 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
637 saddr = &ip6h->saddr;
638 daddr = &ip6h->daddr;
641 hp = tcp_get_md5sig_pool();
643 goto clear_hash_noput;
644 desc = &hp->md5_desc;
646 if (crypto_hash_init(desc))
649 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
651 if (tcp_md5_hash_header(hp, th))
653 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
655 if (tcp_md5_hash_key(hp, key))
657 if (crypto_hash_final(desc, md5_hash))
660 tcp_put_md5sig_pool();
664 tcp_put_md5sig_pool();
666 memset(md5_hash, 0, 16);
670 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
672 const __u8 *hash_location = NULL;
673 struct tcp_md5sig_key *hash_expected;
674 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
675 const struct tcphdr *th = tcp_hdr(skb);
679 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
680 hash_location = tcp_parse_md5sig_option(th);
682 /* We've parsed the options - do we have a hash? */
683 if (!hash_expected && !hash_location)
686 if (hash_expected && !hash_location) {
687 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
691 if (!hash_expected && hash_location) {
692 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
696 /* check the signature */
697 genhash = tcp_v6_md5_hash_skb(newhash,
701 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
702 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
703 genhash ? "failed" : "mismatch",
704 &ip6h->saddr, ntohs(th->source),
705 &ip6h->daddr, ntohs(th->dest));
712 static void tcp_v6_init_req(struct request_sock *req, struct sock *sk,
715 struct inet_request_sock *ireq = inet_rsk(req);
716 struct ipv6_pinfo *np = inet6_sk(sk);
718 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
719 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
721 ireq->ir_iif = sk->sk_bound_dev_if;
723 /* So that link locals have meaning */
724 if (!sk->sk_bound_dev_if &&
725 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
726 ireq->ir_iif = inet6_iif(skb);
728 if (!TCP_SKB_CB(skb)->when &&
729 (ipv6_opt_accepted(sk, skb) || np->rxopt.bits.rxinfo ||
730 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
731 np->rxopt.bits.rxohlim || np->repflow)) {
732 atomic_inc(&skb->users);
737 static struct dst_entry *tcp_v6_route_req(struct sock *sk, struct flowi *fl,
738 const struct request_sock *req,
743 return inet6_csk_route_req(sk, &fl->u.ip6, req);
746 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
748 .obj_size = sizeof(struct tcp6_request_sock),
749 .rtx_syn_ack = tcp_rtx_synack,
750 .send_ack = tcp_v6_reqsk_send_ack,
751 .destructor = tcp_v6_reqsk_destructor,
752 .send_reset = tcp_v6_send_reset,
753 .syn_ack_timeout = tcp_syn_ack_timeout,
756 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
757 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
758 sizeof(struct ipv6hdr),
759 #ifdef CONFIG_TCP_MD5SIG
760 .md5_lookup = tcp_v6_reqsk_md5_lookup,
761 .calc_md5_hash = tcp_v6_md5_hash_skb,
763 .init_req = tcp_v6_init_req,
764 #ifdef CONFIG_SYN_COOKIES
765 .cookie_init_seq = cookie_v6_init_sequence,
767 .route_req = tcp_v6_route_req,
768 .init_seq = tcp_v6_init_sequence,
769 .send_synack = tcp_v6_send_synack,
770 .queue_hash_add = inet6_csk_reqsk_queue_hash_add,
773 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
774 u32 tsval, u32 tsecr, int oif,
775 struct tcp_md5sig_key *key, int rst, u8 tclass,
778 const struct tcphdr *th = tcp_hdr(skb);
780 struct sk_buff *buff;
782 struct net *net = dev_net(skb_dst(skb)->dev);
783 struct sock *ctl_sk = net->ipv6.tcp_sk;
784 unsigned int tot_len = sizeof(struct tcphdr);
785 struct dst_entry *dst;
789 tot_len += TCPOLEN_TSTAMP_ALIGNED;
790 #ifdef CONFIG_TCP_MD5SIG
792 tot_len += TCPOLEN_MD5SIG_ALIGNED;
795 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
800 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
802 t1 = (struct tcphdr *) skb_push(buff, tot_len);
803 skb_reset_transport_header(buff);
805 /* Swap the send and the receive. */
806 memset(t1, 0, sizeof(*t1));
807 t1->dest = th->source;
808 t1->source = th->dest;
809 t1->doff = tot_len / 4;
810 t1->seq = htonl(seq);
811 t1->ack_seq = htonl(ack);
812 t1->ack = !rst || !th->ack;
814 t1->window = htons(win);
816 topt = (__be32 *)(t1 + 1);
819 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
820 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
821 *topt++ = htonl(tsval);
822 *topt++ = htonl(tsecr);
825 #ifdef CONFIG_TCP_MD5SIG
827 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
828 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
829 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
830 &ipv6_hdr(skb)->saddr,
831 &ipv6_hdr(skb)->daddr, t1);
835 memset(&fl6, 0, sizeof(fl6));
836 fl6.daddr = ipv6_hdr(skb)->saddr;
837 fl6.saddr = ipv6_hdr(skb)->daddr;
838 fl6.flowlabel = label;
840 buff->ip_summed = CHECKSUM_PARTIAL;
843 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
845 fl6.flowi6_proto = IPPROTO_TCP;
846 if (rt6_need_strict(&fl6.daddr) && !oif)
847 fl6.flowi6_oif = inet6_iif(skb);
849 fl6.flowi6_oif = oif;
850 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
851 fl6.fl6_dport = t1->dest;
852 fl6.fl6_sport = t1->source;
853 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
855 /* Pass a socket to ip6_dst_lookup either it is for RST
856 * Underlying function will use this to retrieve the network
859 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
861 skb_dst_set(buff, dst);
862 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
863 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
865 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
872 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
874 const struct tcphdr *th = tcp_hdr(skb);
875 u32 seq = 0, ack_seq = 0;
876 struct tcp_md5sig_key *key = NULL;
877 #ifdef CONFIG_TCP_MD5SIG
878 const __u8 *hash_location = NULL;
879 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
880 unsigned char newhash[16];
882 struct sock *sk1 = NULL;
889 if (!ipv6_unicast_destination(skb))
892 #ifdef CONFIG_TCP_MD5SIG
893 hash_location = tcp_parse_md5sig_option(th);
894 if (!sk && hash_location) {
896 * active side is lost. Try to find listening socket through
897 * source port, and then find md5 key through listening socket.
898 * we are not loose security here:
899 * Incoming packet is checked with md5 hash with finding key,
900 * no RST generated if md5 hash doesn't match.
902 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
903 &tcp_hashinfo, &ipv6h->saddr,
904 th->source, &ipv6h->daddr,
905 ntohs(th->source), inet6_iif(skb));
910 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
914 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
915 if (genhash || memcmp(hash_location, newhash, 16) != 0)
918 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
923 seq = ntohl(th->ack_seq);
925 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
928 oif = sk ? sk->sk_bound_dev_if : 0;
929 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
931 #ifdef CONFIG_TCP_MD5SIG
940 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
941 u32 win, u32 tsval, u32 tsecr, int oif,
942 struct tcp_md5sig_key *key, u8 tclass,
945 tcp_v6_send_response(skb, seq, ack, win, tsval, tsecr, oif, key, 0, tclass,
949 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
951 struct inet_timewait_sock *tw = inet_twsk(sk);
952 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
954 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
955 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
956 tcp_time_stamp + tcptw->tw_ts_offset,
957 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
958 tw->tw_tclass, (tw->tw_flowlabel << 12));
963 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
964 struct request_sock *req)
966 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
967 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
969 tcp_v6_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
970 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
971 tcp_rsk(req)->rcv_nxt,
972 req->rcv_wnd, tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
973 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
978 static struct sock *tcp_v6_hnd_req(struct sock *sk, struct sk_buff *skb)
980 struct request_sock *req, **prev;
981 const struct tcphdr *th = tcp_hdr(skb);
984 /* Find possible connection requests. */
985 req = inet6_csk_search_req(sk, &prev, th->source,
986 &ipv6_hdr(skb)->saddr,
987 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
989 return tcp_check_req(sk, skb, req, prev, false);
991 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
992 &ipv6_hdr(skb)->saddr, th->source,
993 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
996 if (nsk->sk_state != TCP_TIME_WAIT) {
1000 inet_twsk_put(inet_twsk(nsk));
1004 #ifdef CONFIG_SYN_COOKIES
1006 sk = cookie_v6_check(sk, skb);
1011 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1013 if (skb->protocol == htons(ETH_P_IP))
1014 return tcp_v4_conn_request(sk, skb);
1016 if (!ipv6_unicast_destination(skb))
1019 return tcp_conn_request(&tcp6_request_sock_ops,
1020 &tcp_request_sock_ipv6_ops, sk, skb);
1023 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1024 return 0; /* don't send reset */
1027 static struct sock *tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1028 struct request_sock *req,
1029 struct dst_entry *dst)
1031 struct inet_request_sock *ireq;
1032 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1033 struct tcp6_sock *newtcp6sk;
1034 struct inet_sock *newinet;
1035 struct tcp_sock *newtp;
1037 #ifdef CONFIG_TCP_MD5SIG
1038 struct tcp_md5sig_key *key;
1042 if (skb->protocol == htons(ETH_P_IP)) {
1047 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1052 newtcp6sk = (struct tcp6_sock *)newsk;
1053 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1055 newinet = inet_sk(newsk);
1056 newnp = inet6_sk(newsk);
1057 newtp = tcp_sk(newsk);
1059 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1061 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newsk->sk_v6_daddr);
1063 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1065 newsk->sk_v6_rcv_saddr = newnp->saddr;
1067 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1068 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1069 #ifdef CONFIG_TCP_MD5SIG
1070 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1073 newnp->ipv6_ac_list = NULL;
1074 newnp->ipv6_fl_list = NULL;
1075 newnp->pktoptions = NULL;
1077 newnp->mcast_oif = inet6_iif(skb);
1078 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1079 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1081 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1084 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1085 * here, tcp_create_openreq_child now does this for us, see the comment in
1086 * that function for the gory details. -acme
1089 /* It is tricky place. Until this moment IPv4 tcp
1090 worked with IPv6 icsk.icsk_af_ops.
1093 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1098 ireq = inet_rsk(req);
1100 if (sk_acceptq_is_full(sk))
1104 dst = inet6_csk_route_req(sk, &fl6, req);
1109 newsk = tcp_create_openreq_child(sk, req, skb);
1114 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1115 * count here, tcp_create_openreq_child now does this for us, see the
1116 * comment in that function for the gory details. -acme
1119 newsk->sk_gso_type = SKB_GSO_TCPV6;
1120 __ip6_dst_store(newsk, dst, NULL, NULL);
1121 inet6_sk_rx_dst_set(newsk, skb);
1123 newtcp6sk = (struct tcp6_sock *)newsk;
1124 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1126 newtp = tcp_sk(newsk);
1127 newinet = inet_sk(newsk);
1128 newnp = inet6_sk(newsk);
1130 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1132 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1133 newnp->saddr = ireq->ir_v6_loc_addr;
1134 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1135 newsk->sk_bound_dev_if = ireq->ir_iif;
1137 ip6_set_txhash(newsk);
1139 /* Now IPv6 options...
1141 First: no IPv4 options.
1143 newinet->inet_opt = NULL;
1144 newnp->ipv6_ac_list = NULL;
1145 newnp->ipv6_fl_list = NULL;
1148 newnp->rxopt.all = np->rxopt.all;
1150 /* Clone pktoptions received with SYN */
1151 newnp->pktoptions = NULL;
1152 if (ireq->pktopts != NULL) {
1153 newnp->pktoptions = skb_clone(ireq->pktopts,
1154 sk_gfp_atomic(sk, GFP_ATOMIC));
1155 consume_skb(ireq->pktopts);
1156 ireq->pktopts = NULL;
1157 if (newnp->pktoptions)
1158 skb_set_owner_r(newnp->pktoptions, newsk);
1161 newnp->mcast_oif = inet6_iif(skb);
1162 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1163 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1165 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1167 /* Clone native IPv6 options from listening socket (if any)
1169 Yes, keeping reference count would be much more clever,
1170 but we make one more one thing there: reattach optmem
1174 newnp->opt = ipv6_dup_options(newsk, np->opt);
1176 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1178 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1179 newnp->opt->opt_flen);
1181 tcp_sync_mss(newsk, dst_mtu(dst));
1182 newtp->advmss = dst_metric_advmss(dst);
1183 if (tcp_sk(sk)->rx_opt.user_mss &&
1184 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1185 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1187 tcp_initialize_rcv_mss(newsk);
1189 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1190 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1192 #ifdef CONFIG_TCP_MD5SIG
1193 /* Copy over the MD5 key from the original socket */
1194 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1196 /* We're using one, so create a matching key
1197 * on the newsk structure. If we fail to get
1198 * memory, then we end up not copying the key
1201 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1202 AF_INET6, key->key, key->keylen,
1203 sk_gfp_atomic(sk, GFP_ATOMIC));
1207 if (__inet_inherit_port(sk, newsk) < 0) {
1208 inet_csk_prepare_forced_close(newsk);
1212 __inet6_hash(newsk, NULL);
1217 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1221 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1225 /* The socket must have it's spinlock held when we get
1228 * We have a potential double-lock case here, so even when
1229 * doing backlog processing we use the BH locking scheme.
1230 * This is because we cannot sleep with the original spinlock
1233 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1235 struct ipv6_pinfo *np = inet6_sk(sk);
1236 struct tcp_sock *tp;
1237 struct sk_buff *opt_skb = NULL;
1239 /* Imagine: socket is IPv6. IPv4 packet arrives,
1240 goes to IPv4 receive handler and backlogged.
1241 From backlog it always goes here. Kerboom...
1242 Fortunately, tcp_rcv_established and rcv_established
1243 handle them correctly, but it is not case with
1244 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1247 if (skb->protocol == htons(ETH_P_IP))
1248 return tcp_v4_do_rcv(sk, skb);
1250 #ifdef CONFIG_TCP_MD5SIG
1251 if (tcp_v6_inbound_md5_hash(sk, skb))
1255 if (sk_filter(sk, skb))
1259 * socket locking is here for SMP purposes as backlog rcv
1260 * is currently called with bh processing disabled.
1263 /* Do Stevens' IPV6_PKTOPTIONS.
1265 Yes, guys, it is the only place in our code, where we
1266 may make it not affecting IPv4.
1267 The rest of code is protocol independent,
1268 and I do not like idea to uglify IPv4.
1270 Actually, all the idea behind IPV6_PKTOPTIONS
1271 looks not very well thought. For now we latch
1272 options, received in the last packet, enqueued
1273 by tcp. Feel free to propose better solution.
1277 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1279 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1280 struct dst_entry *dst = sk->sk_rx_dst;
1282 sock_rps_save_rxhash(sk, skb);
1284 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1285 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1287 sk->sk_rx_dst = NULL;
1291 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1293 goto ipv6_pktoptions;
1297 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1300 if (sk->sk_state == TCP_LISTEN) {
1301 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1306 * Queue it on the new socket if the new socket is active,
1307 * otherwise we just shortcircuit this and continue with
1311 sock_rps_save_rxhash(nsk, skb);
1312 if (tcp_child_process(sk, nsk, skb))
1315 __kfree_skb(opt_skb);
1319 sock_rps_save_rxhash(sk, skb);
1321 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1324 goto ipv6_pktoptions;
1328 tcp_v6_send_reset(sk, skb);
1331 __kfree_skb(opt_skb);
1335 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1336 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1341 /* Do you ask, what is it?
1343 1. skb was enqueued by tcp.
1344 2. skb is added to tail of read queue, rather than out of order.
1345 3. socket is not in passive state.
1346 4. Finally, it really contains options, which user wants to receive.
1349 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1350 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1351 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1352 np->mcast_oif = inet6_iif(opt_skb);
1353 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1354 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1355 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1356 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1358 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1359 if (ipv6_opt_accepted(sk, opt_skb)) {
1360 skb_set_owner_r(opt_skb, sk);
1361 opt_skb = xchg(&np->pktoptions, opt_skb);
1363 __kfree_skb(opt_skb);
1364 opt_skb = xchg(&np->pktoptions, NULL);
1372 static int tcp_v6_rcv(struct sk_buff *skb)
1374 const struct tcphdr *th;
1375 const struct ipv6hdr *hdr;
1378 struct net *net = dev_net(skb->dev);
1380 if (skb->pkt_type != PACKET_HOST)
1384 * Count it even if it's bad.
1386 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1388 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1393 if (th->doff < sizeof(struct tcphdr)/4)
1395 if (!pskb_may_pull(skb, th->doff*4))
1398 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1402 hdr = ipv6_hdr(skb);
1403 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1404 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1405 skb->len - th->doff*4);
1406 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1407 TCP_SKB_CB(skb)->when = 0;
1408 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1409 TCP_SKB_CB(skb)->sacked = 0;
1411 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1416 if (sk->sk_state == TCP_TIME_WAIT)
1419 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1420 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1421 goto discard_and_relse;
1424 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1425 goto discard_and_relse;
1427 if (sk_filter(sk, skb))
1428 goto discard_and_relse;
1430 sk_mark_napi_id(sk, skb);
1433 bh_lock_sock_nested(sk);
1435 if (!sock_owned_by_user(sk)) {
1436 #ifdef CONFIG_NET_DMA
1437 struct tcp_sock *tp = tcp_sk(sk);
1438 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1439 tp->ucopy.dma_chan = net_dma_find_channel();
1440 if (tp->ucopy.dma_chan)
1441 ret = tcp_v6_do_rcv(sk, skb);
1445 if (!tcp_prequeue(sk, skb))
1446 ret = tcp_v6_do_rcv(sk, skb);
1448 } else if (unlikely(sk_add_backlog(sk, skb,
1449 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1451 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1452 goto discard_and_relse;
1457 return ret ? -1 : 0;
1460 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1463 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1465 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1467 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1469 tcp_v6_send_reset(NULL, skb);
1481 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1482 inet_twsk_put(inet_twsk(sk));
1486 if (skb->len < (th->doff<<2)) {
1487 inet_twsk_put(inet_twsk(sk));
1490 if (tcp_checksum_complete(skb)) {
1491 inet_twsk_put(inet_twsk(sk));
1495 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1500 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1501 &ipv6_hdr(skb)->saddr, th->source,
1502 &ipv6_hdr(skb)->daddr,
1503 ntohs(th->dest), inet6_iif(skb));
1505 struct inet_timewait_sock *tw = inet_twsk(sk);
1506 inet_twsk_deschedule(tw, &tcp_death_row);
1511 /* Fall through to ACK */
1514 tcp_v6_timewait_ack(sk, skb);
1518 case TCP_TW_SUCCESS:
1524 static void tcp_v6_early_demux(struct sk_buff *skb)
1526 const struct ipv6hdr *hdr;
1527 const struct tcphdr *th;
1530 if (skb->pkt_type != PACKET_HOST)
1533 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1536 hdr = ipv6_hdr(skb);
1539 if (th->doff < sizeof(struct tcphdr) / 4)
1542 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1543 &hdr->saddr, th->source,
1544 &hdr->daddr, ntohs(th->dest),
1548 skb->destructor = sock_edemux;
1549 if (sk->sk_state != TCP_TIME_WAIT) {
1550 struct dst_entry *dst = sk->sk_rx_dst;
1553 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1555 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1556 skb_dst_set_noref(skb, dst);
1561 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1562 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1563 .twsk_unique = tcp_twsk_unique,
1564 .twsk_destructor = tcp_twsk_destructor,
1567 static const struct inet_connection_sock_af_ops ipv6_specific = {
1568 .queue_xmit = inet6_csk_xmit,
1569 .send_check = tcp_v6_send_check,
1570 .rebuild_header = inet6_sk_rebuild_header,
1571 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1572 .conn_request = tcp_v6_conn_request,
1573 .syn_recv_sock = tcp_v6_syn_recv_sock,
1574 .net_header_len = sizeof(struct ipv6hdr),
1575 .net_frag_header_len = sizeof(struct frag_hdr),
1576 .setsockopt = ipv6_setsockopt,
1577 .getsockopt = ipv6_getsockopt,
1578 .addr2sockaddr = inet6_csk_addr2sockaddr,
1579 .sockaddr_len = sizeof(struct sockaddr_in6),
1580 .bind_conflict = inet6_csk_bind_conflict,
1581 #ifdef CONFIG_COMPAT
1582 .compat_setsockopt = compat_ipv6_setsockopt,
1583 .compat_getsockopt = compat_ipv6_getsockopt,
1587 #ifdef CONFIG_TCP_MD5SIG
1588 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1589 .md5_lookup = tcp_v6_md5_lookup,
1590 .calc_md5_hash = tcp_v6_md5_hash_skb,
1591 .md5_parse = tcp_v6_parse_md5_keys,
1596 * TCP over IPv4 via INET6 API
1598 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1599 .queue_xmit = ip_queue_xmit,
1600 .send_check = tcp_v4_send_check,
1601 .rebuild_header = inet_sk_rebuild_header,
1602 .sk_rx_dst_set = inet_sk_rx_dst_set,
1603 .conn_request = tcp_v6_conn_request,
1604 .syn_recv_sock = tcp_v6_syn_recv_sock,
1605 .net_header_len = sizeof(struct iphdr),
1606 .setsockopt = ipv6_setsockopt,
1607 .getsockopt = ipv6_getsockopt,
1608 .addr2sockaddr = inet6_csk_addr2sockaddr,
1609 .sockaddr_len = sizeof(struct sockaddr_in6),
1610 .bind_conflict = inet6_csk_bind_conflict,
1611 #ifdef CONFIG_COMPAT
1612 .compat_setsockopt = compat_ipv6_setsockopt,
1613 .compat_getsockopt = compat_ipv6_getsockopt,
1617 #ifdef CONFIG_TCP_MD5SIG
1618 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1619 .md5_lookup = tcp_v4_md5_lookup,
1620 .calc_md5_hash = tcp_v4_md5_hash_skb,
1621 .md5_parse = tcp_v6_parse_md5_keys,
1625 /* NOTE: A lot of things set to zero explicitly by call to
1626 * sk_alloc() so need not be done here.
1628 static int tcp_v6_init_sock(struct sock *sk)
1630 struct inet_connection_sock *icsk = inet_csk(sk);
1634 icsk->icsk_af_ops = &ipv6_specific;
1636 #ifdef CONFIG_TCP_MD5SIG
1637 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1643 static void tcp_v6_destroy_sock(struct sock *sk)
1645 tcp_v4_destroy_sock(sk);
1646 inet6_destroy_sock(sk);
1649 #ifdef CONFIG_PROC_FS
1650 /* Proc filesystem TCPv6 sock list dumping. */
1651 static void get_openreq6(struct seq_file *seq,
1652 const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1654 int ttd = req->expires - jiffies;
1655 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1656 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1662 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1663 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1665 src->s6_addr32[0], src->s6_addr32[1],
1666 src->s6_addr32[2], src->s6_addr32[3],
1667 inet_rsk(req)->ir_num,
1668 dest->s6_addr32[0], dest->s6_addr32[1],
1669 dest->s6_addr32[2], dest->s6_addr32[3],
1670 ntohs(inet_rsk(req)->ir_rmt_port),
1672 0, 0, /* could print option size, but that is af dependent. */
1673 1, /* timers active (only the expire timer) */
1674 jiffies_to_clock_t(ttd),
1676 from_kuid_munged(seq_user_ns(seq), uid),
1677 0, /* non standard timer */
1678 0, /* open_requests have no inode */
1682 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1684 const struct in6_addr *dest, *src;
1687 unsigned long timer_expires;
1688 const struct inet_sock *inet = inet_sk(sp);
1689 const struct tcp_sock *tp = tcp_sk(sp);
1690 const struct inet_connection_sock *icsk = inet_csk(sp);
1691 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
1693 dest = &sp->sk_v6_daddr;
1694 src = &sp->sk_v6_rcv_saddr;
1695 destp = ntohs(inet->inet_dport);
1696 srcp = ntohs(inet->inet_sport);
1698 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1700 timer_expires = icsk->icsk_timeout;
1701 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1703 timer_expires = icsk->icsk_timeout;
1704 } else if (timer_pending(&sp->sk_timer)) {
1706 timer_expires = sp->sk_timer.expires;
1709 timer_expires = jiffies;
1713 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1714 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1716 src->s6_addr32[0], src->s6_addr32[1],
1717 src->s6_addr32[2], src->s6_addr32[3], srcp,
1718 dest->s6_addr32[0], dest->s6_addr32[1],
1719 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1721 tp->write_seq-tp->snd_una,
1722 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1724 jiffies_delta_to_clock_t(timer_expires - jiffies),
1725 icsk->icsk_retransmits,
1726 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1727 icsk->icsk_probes_out,
1729 atomic_read(&sp->sk_refcnt), sp,
1730 jiffies_to_clock_t(icsk->icsk_rto),
1731 jiffies_to_clock_t(icsk->icsk_ack.ato),
1732 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1734 sp->sk_state == TCP_LISTEN ?
1735 (fastopenq ? fastopenq->max_qlen : 0) :
1736 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1740 static void get_timewait6_sock(struct seq_file *seq,
1741 struct inet_timewait_sock *tw, int i)
1743 const struct in6_addr *dest, *src;
1745 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
1747 dest = &tw->tw_v6_daddr;
1748 src = &tw->tw_v6_rcv_saddr;
1749 destp = ntohs(tw->tw_dport);
1750 srcp = ntohs(tw->tw_sport);
1753 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1754 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1756 src->s6_addr32[0], src->s6_addr32[1],
1757 src->s6_addr32[2], src->s6_addr32[3], srcp,
1758 dest->s6_addr32[0], dest->s6_addr32[1],
1759 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1760 tw->tw_substate, 0, 0,
1761 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1762 atomic_read(&tw->tw_refcnt), tw);
1765 static int tcp6_seq_show(struct seq_file *seq, void *v)
1767 struct tcp_iter_state *st;
1768 struct sock *sk = v;
1770 if (v == SEQ_START_TOKEN) {
1775 "st tx_queue rx_queue tr tm->when retrnsmt"
1776 " uid timeout inode\n");
1781 switch (st->state) {
1782 case TCP_SEQ_STATE_LISTENING:
1783 case TCP_SEQ_STATE_ESTABLISHED:
1784 if (sk->sk_state == TCP_TIME_WAIT)
1785 get_timewait6_sock(seq, v, st->num);
1787 get_tcp6_sock(seq, v, st->num);
1789 case TCP_SEQ_STATE_OPENREQ:
1790 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1797 static const struct file_operations tcp6_afinfo_seq_fops = {
1798 .owner = THIS_MODULE,
1799 .open = tcp_seq_open,
1801 .llseek = seq_lseek,
1802 .release = seq_release_net
1805 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1808 .seq_fops = &tcp6_afinfo_seq_fops,
1810 .show = tcp6_seq_show,
1814 int __net_init tcp6_proc_init(struct net *net)
1816 return tcp_proc_register(net, &tcp6_seq_afinfo);
1819 void tcp6_proc_exit(struct net *net)
1821 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1825 static void tcp_v6_clear_sk(struct sock *sk, int size)
1827 struct inet_sock *inet = inet_sk(sk);
1829 /* we do not want to clear pinet6 field, because of RCU lookups */
1830 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1832 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1833 memset(&inet->pinet6 + 1, 0, size);
1836 struct proto tcpv6_prot = {
1838 .owner = THIS_MODULE,
1840 .connect = tcp_v6_connect,
1841 .disconnect = tcp_disconnect,
1842 .accept = inet_csk_accept,
1844 .init = tcp_v6_init_sock,
1845 .destroy = tcp_v6_destroy_sock,
1846 .shutdown = tcp_shutdown,
1847 .setsockopt = tcp_setsockopt,
1848 .getsockopt = tcp_getsockopt,
1849 .recvmsg = tcp_recvmsg,
1850 .sendmsg = tcp_sendmsg,
1851 .sendpage = tcp_sendpage,
1852 .backlog_rcv = tcp_v6_do_rcv,
1853 .release_cb = tcp_release_cb,
1854 .mtu_reduced = tcp_v6_mtu_reduced,
1855 .hash = tcp_v6_hash,
1856 .unhash = inet_unhash,
1857 .get_port = inet_csk_get_port,
1858 .enter_memory_pressure = tcp_enter_memory_pressure,
1859 .stream_memory_free = tcp_stream_memory_free,
1860 .sockets_allocated = &tcp_sockets_allocated,
1861 .memory_allocated = &tcp_memory_allocated,
1862 .memory_pressure = &tcp_memory_pressure,
1863 .orphan_count = &tcp_orphan_count,
1864 .sysctl_mem = sysctl_tcp_mem,
1865 .sysctl_wmem = sysctl_tcp_wmem,
1866 .sysctl_rmem = sysctl_tcp_rmem,
1867 .max_header = MAX_TCP_HEADER,
1868 .obj_size = sizeof(struct tcp6_sock),
1869 .slab_flags = SLAB_DESTROY_BY_RCU,
1870 .twsk_prot = &tcp6_timewait_sock_ops,
1871 .rsk_prot = &tcp6_request_sock_ops,
1872 .h.hashinfo = &tcp_hashinfo,
1873 .no_autobind = true,
1874 #ifdef CONFIG_COMPAT
1875 .compat_setsockopt = compat_tcp_setsockopt,
1876 .compat_getsockopt = compat_tcp_getsockopt,
1878 #ifdef CONFIG_MEMCG_KMEM
1879 .proto_cgroup = tcp_proto_cgroup,
1881 .clear_sk = tcp_v6_clear_sk,
1884 static const struct inet6_protocol tcpv6_protocol = {
1885 .early_demux = tcp_v6_early_demux,
1886 .handler = tcp_v6_rcv,
1887 .err_handler = tcp_v6_err,
1888 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1891 static struct inet_protosw tcpv6_protosw = {
1892 .type = SOCK_STREAM,
1893 .protocol = IPPROTO_TCP,
1894 .prot = &tcpv6_prot,
1895 .ops = &inet6_stream_ops,
1896 .flags = INET_PROTOSW_PERMANENT |
1900 static int __net_init tcpv6_net_init(struct net *net)
1902 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1903 SOCK_RAW, IPPROTO_TCP, net);
1906 static void __net_exit tcpv6_net_exit(struct net *net)
1908 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1911 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1913 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1916 static struct pernet_operations tcpv6_net_ops = {
1917 .init = tcpv6_net_init,
1918 .exit = tcpv6_net_exit,
1919 .exit_batch = tcpv6_net_exit_batch,
1922 int __init tcpv6_init(void)
1926 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1930 /* register inet6 protocol */
1931 ret = inet6_register_protosw(&tcpv6_protosw);
1933 goto out_tcpv6_protocol;
1935 ret = register_pernet_subsys(&tcpv6_net_ops);
1937 goto out_tcpv6_protosw;
1942 inet6_unregister_protosw(&tcpv6_protosw);
1944 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1948 void tcpv6_exit(void)
1950 unregister_pernet_subsys(&tcpv6_net_ops);
1951 inet6_unregister_protosw(&tcpv6_protosw);
1952 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);