3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: tcp_ipv6.c,v 1.144 2002/02/01 22:01:04 davem Exp $
11 * linux/net/ipv4/tcp.c
12 * linux/net/ipv4/tcp_input.c
13 * linux/net/ipv4/tcp_output.c
16 * Hideaki YOSHIFUJI : sin6_scope_id support
17 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
18 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
19 * a single port at the same time.
20 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
22 * This program is free software; you can redistribute it and/or
23 * modify it under the terms of the GNU General Public License
24 * as published by the Free Software Foundation; either version
25 * 2 of the License, or (at your option) any later version.
28 #include <linux/module.h>
29 #include <linux/config.h>
30 #include <linux/errno.h>
31 #include <linux/types.h>
32 #include <linux/socket.h>
33 #include <linux/sockios.h>
34 #include <linux/net.h>
35 #include <linux/jiffies.h>
37 #include <linux/in6.h>
38 #include <linux/netdevice.h>
39 #include <linux/init.h>
40 #include <linux/jhash.h>
41 #include <linux/ipsec.h>
42 #include <linux/times.h>
44 #include <linux/ipv6.h>
45 #include <linux/icmpv6.h>
46 #include <linux/random.h>
49 #include <net/ndisc.h>
51 #include <net/transp_v6.h>
52 #include <net/addrconf.h>
53 #include <net/ip6_route.h>
54 #include <net/ip6_checksum.h>
55 #include <net/inet_ecn.h>
56 #include <net/protocol.h>
59 #include <asm/uaccess.h>
61 #include <linux/proc_fs.h>
62 #include <linux/seq_file.h>
64 static void tcp_v6_send_reset(struct sk_buff *skb);
65 static void tcp_v6_or_send_ack(struct sk_buff *skb, struct open_request *req);
66 static void tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len,
69 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
70 static int tcp_v6_xmit(struct sk_buff *skb, int ipfragok);
72 static struct tcp_func ipv6_mapped;
73 static struct tcp_func ipv6_specific;
75 /* I have no idea if this is a good hash for v6 or not. -DaveM */
76 static __inline__ int tcp_v6_hashfn(struct in6_addr *laddr, u16 lport,
77 struct in6_addr *faddr, u16 fport)
79 int hashent = (lport ^ fport);
81 hashent ^= (laddr->s6_addr32[3] ^ faddr->s6_addr32[3]);
82 hashent ^= hashent>>16;
83 hashent ^= hashent>>8;
84 return (hashent & (tcp_ehash_size - 1));
87 static __inline__ int tcp_v6_sk_hashfn(struct sock *sk)
89 struct inet_opt *inet = inet_sk(sk);
90 struct ipv6_pinfo *np = inet6_sk(sk);
91 struct in6_addr *laddr = &np->rcv_saddr;
92 struct in6_addr *faddr = &np->daddr;
93 __u16 lport = inet->num;
94 __u16 fport = inet->dport;
95 return tcp_v6_hashfn(laddr, lport, faddr, fport);
98 static inline int tcp_v6_bind_conflict(struct sock *sk,
99 struct tcp_bind_bucket *tb)
102 struct hlist_node *node;
104 /* We must walk the whole port owner list in this case. -DaveM */
105 sk_for_each_bound(sk2, node, &tb->owners) {
107 (!sk->sk_bound_dev_if ||
108 !sk2->sk_bound_dev_if ||
109 sk->sk_bound_dev_if == sk2->sk_bound_dev_if) &&
110 (!sk->sk_reuse || !sk2->sk_reuse ||
111 sk2->sk_state == TCP_LISTEN) &&
112 ipv6_rcv_saddr_equal(sk, sk2))
119 /* Grrr, addr_type already calculated by caller, but I don't want
120 * to add some silly "cookie" argument to this method just for that.
121 * But it doesn't matter, the recalculation is in the rarest path
122 * this function ever takes.
124 static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
126 struct tcp_bind_hashbucket *head;
127 struct tcp_bind_bucket *tb;
128 struct hlist_node *node;
133 int low = sysctl_local_port_range[0];
134 int high = sysctl_local_port_range[1];
135 int remaining = (high - low) + 1;
138 spin_lock(&tcp_portalloc_lock);
139 rover = tcp_port_rover;
141 if ((rover < low) || (rover > high))
143 head = &tcp_bhash[tcp_bhashfn(rover)];
144 spin_lock(&head->lock);
145 tb_for_each(tb, node, &head->chain)
146 if (tb->port == rover)
150 spin_unlock(&head->lock);
151 } while (--remaining > 0);
152 tcp_port_rover = rover;
153 spin_unlock(&tcp_portalloc_lock);
155 /* Exhausted local port range during search? */
160 /* OK, here is the one we will use. */
163 head = &tcp_bhash[tcp_bhashfn(snum)];
164 spin_lock(&head->lock);
165 tb_for_each(tb, node, &head->chain)
166 if (tb->port == snum)
172 if (tb && !hlist_empty(&tb->owners)) {
173 if (tb->fastreuse > 0 && sk->sk_reuse &&
174 sk->sk_state != TCP_LISTEN) {
178 if (tcp_v6_bind_conflict(sk, tb))
184 if (!tb && (tb = tcp_bucket_create(head, snum)) == NULL)
186 if (hlist_empty(&tb->owners)) {
187 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
191 } else if (tb->fastreuse &&
192 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
196 if (!tcp_sk(sk)->bind_hash)
197 tcp_bind_hash(sk, tb, snum);
198 BUG_TRAP(tcp_sk(sk)->bind_hash == tb);
202 spin_unlock(&head->lock);
208 static __inline__ void __tcp_v6_hash(struct sock *sk)
210 struct hlist_head *list;
213 BUG_TRAP(sk_unhashed(sk));
215 if (sk->sk_state == TCP_LISTEN) {
216 list = &tcp_listening_hash[tcp_sk_listen_hashfn(sk)];
217 lock = &tcp_lhash_lock;
220 sk->sk_hashent = tcp_v6_sk_hashfn(sk);
221 list = &tcp_ehash[sk->sk_hashent].chain;
222 lock = &tcp_ehash[sk->sk_hashent].lock;
226 __sk_add_node(sk, list);
227 sock_prot_inc_use(sk->sk_prot);
232 static void tcp_v6_hash(struct sock *sk)
234 if (sk->sk_state != TCP_CLOSE) {
235 struct tcp_opt *tp = tcp_sk(sk);
237 if (tp->af_specific == &ipv6_mapped) {
247 static struct sock *tcp_v6_lookup_listener(struct in6_addr *daddr, unsigned short hnum, int dif)
250 struct hlist_node *node;
251 struct sock *result = NULL;
255 read_lock(&tcp_lhash_lock);
256 sk_for_each(sk, node, &tcp_listening_hash[tcp_lhashfn(hnum)]) {
257 if (inet_sk(sk)->num == hnum && sk->sk_family == PF_INET6) {
258 struct ipv6_pinfo *np = inet6_sk(sk);
261 if (!ipv6_addr_any(&np->rcv_saddr)) {
262 if (ipv6_addr_cmp(&np->rcv_saddr, daddr))
266 if (sk->sk_bound_dev_if) {
267 if (sk->sk_bound_dev_if != dif)
275 if (score > hiscore) {
283 read_unlock(&tcp_lhash_lock);
287 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
288 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
290 * The sockhash lock must be held as a reader here.
293 static inline struct sock *__tcp_v6_lookup_established(struct in6_addr *saddr, u16 sport,
294 struct in6_addr *daddr, u16 hnum,
297 struct tcp_ehash_bucket *head;
299 struct hlist_node *node;
300 __u32 ports = TCP_COMBINED_PORTS(sport, hnum);
303 /* Optimize here for direct hit, only listening connections can
304 * have wildcards anyways.
306 hash = tcp_v6_hashfn(daddr, hnum, saddr, sport);
307 head = &tcp_ehash[hash];
308 read_lock(&head->lock);
309 sk_for_each(sk, node, &head->chain) {
310 /* For IPV6 do the cheaper port and family tests first. */
311 if(TCP_IPV6_MATCH(sk, saddr, daddr, ports, dif))
312 goto hit; /* You sunk my battleship! */
314 /* Must check for a TIME_WAIT'er before going to listener hash. */
315 sk_for_each(sk, node, &(head + tcp_ehash_size)->chain) {
316 /* FIXME: acme: check this... */
317 struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk;
319 if(*((__u32 *)&(tw->tw_dport)) == ports &&
320 sk->sk_family == PF_INET6) {
321 if(!ipv6_addr_cmp(&tw->tw_v6_daddr, saddr) &&
322 !ipv6_addr_cmp(&tw->tw_v6_rcv_saddr, daddr) &&
323 (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif))
327 read_unlock(&head->lock);
332 read_unlock(&head->lock);
337 static inline struct sock *__tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
338 struct in6_addr *daddr, u16 hnum,
343 sk = __tcp_v6_lookup_established(saddr, sport, daddr, hnum, dif);
348 return tcp_v6_lookup_listener(daddr, hnum, dif);
351 inline struct sock *tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
352 struct in6_addr *daddr, u16 dport,
358 sk = __tcp_v6_lookup(saddr, sport, daddr, ntohs(dport), dif);
366 * Open request hash tables.
369 static u32 tcp_v6_synq_hash(struct in6_addr *raddr, u16 rport, u32 rnd)
373 a = raddr->s6_addr32[0];
374 b = raddr->s6_addr32[1];
375 c = raddr->s6_addr32[2];
377 a += JHASH_GOLDEN_RATIO;
378 b += JHASH_GOLDEN_RATIO;
380 __jhash_mix(a, b, c);
382 a += raddr->s6_addr32[3];
384 __jhash_mix(a, b, c);
386 return c & (TCP_SYNQ_HSIZE - 1);
389 static struct open_request *tcp_v6_search_req(struct tcp_opt *tp,
390 struct open_request ***prevp,
392 struct in6_addr *raddr,
393 struct in6_addr *laddr,
396 struct tcp_listen_opt *lopt = tp->listen_opt;
397 struct open_request *req, **prev;
399 for (prev = &lopt->syn_table[tcp_v6_synq_hash(raddr, rport, lopt->hash_rnd)];
400 (req = *prev) != NULL;
401 prev = &req->dl_next) {
402 if (req->rmt_port == rport &&
403 req->class->family == AF_INET6 &&
404 !ipv6_addr_cmp(&req->af.v6_req.rmt_addr, raddr) &&
405 !ipv6_addr_cmp(&req->af.v6_req.loc_addr, laddr) &&
406 (!req->af.v6_req.iif || req->af.v6_req.iif == iif)) {
407 BUG_TRAP(req->sk == NULL);
416 static __inline__ u16 tcp_v6_check(struct tcphdr *th, int len,
417 struct in6_addr *saddr,
418 struct in6_addr *daddr,
421 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
424 static __u32 tcp_v6_init_sequence(struct sock *sk, struct sk_buff *skb)
426 if (skb->protocol == htons(ETH_P_IPV6)) {
427 return secure_tcpv6_sequence_number(skb->nh.ipv6h->daddr.s6_addr32,
428 skb->nh.ipv6h->saddr.s6_addr32,
432 return secure_tcp_sequence_number(skb->nh.iph->daddr,
439 static int tcp_v6_check_established(struct sock *sk)
441 struct inet_opt *inet = inet_sk(sk);
442 struct ipv6_pinfo *np = inet6_sk(sk);
443 struct in6_addr *daddr = &np->rcv_saddr;
444 struct in6_addr *saddr = &np->daddr;
445 int dif = sk->sk_bound_dev_if;
446 u32 ports = TCP_COMBINED_PORTS(inet->dport, inet->num);
447 int hash = tcp_v6_hashfn(daddr, inet->num, saddr, inet->dport);
448 struct tcp_ehash_bucket *head = &tcp_ehash[hash];
450 struct hlist_node *node;
451 struct tcp_tw_bucket *tw;
453 write_lock_bh(&head->lock);
455 /* Check TIME-WAIT sockets first. */
456 sk_for_each(sk2, node, &(head + tcp_ehash_size)->chain) {
457 tw = (struct tcp_tw_bucket*)sk2;
459 if(*((__u32 *)&(tw->tw_dport)) == ports &&
460 sk2->sk_family == PF_INET6 &&
461 !ipv6_addr_cmp(&tw->tw_v6_daddr, saddr) &&
462 !ipv6_addr_cmp(&tw->tw_v6_rcv_saddr, daddr) &&
463 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) {
464 struct tcp_opt *tp = tcp_sk(sk);
466 if (tw->tw_ts_recent_stamp) {
467 /* See comment in tcp_ipv4.c */
468 tp->write_seq = tw->tw_snd_nxt + 65535 + 2;
471 tp->ts_recent = tw->tw_ts_recent;
472 tp->ts_recent_stamp = tw->tw_ts_recent_stamp;
481 /* And established part... */
482 sk_for_each(sk2, node, &head->chain) {
483 if(TCP_IPV6_MATCH(sk2, saddr, daddr, ports, dif))
488 BUG_TRAP(sk_unhashed(sk));
489 __sk_add_node(sk, &head->chain);
490 sk->sk_hashent = hash;
491 sock_prot_inc_use(sk->sk_prot);
492 write_unlock_bh(&head->lock);
495 /* Silly. Should hash-dance instead... */
497 tcp_tw_deschedule(tw);
498 NET_INC_STATS_BH(TimeWaitRecycled);
506 write_unlock_bh(&head->lock);
507 return -EADDRNOTAVAIL;
510 static int tcp_v6_hash_connect(struct sock *sk)
512 struct tcp_bind_hashbucket *head;
513 struct tcp_bind_bucket *tb;
516 if (inet_sk(sk)->num == 0) {
517 int err = tcp_v6_get_port(sk, inet_sk(sk)->num);
520 inet_sk(sk)->sport = htons(inet_sk(sk)->num);
523 head = &tcp_bhash[tcp_bhashfn(inet_sk(sk)->num)];
526 spin_lock_bh(&head->lock);
528 if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
530 spin_unlock_bh(&head->lock);
533 spin_unlock_bh(&head->lock);
534 return tcp_v6_check_established(sk);
538 static __inline__ int tcp_v6_iif(struct sk_buff *skb)
540 return IP6CB(skb)->iif;
543 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
546 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
547 struct inet_opt *inet = inet_sk(sk);
548 struct ipv6_pinfo *np = inet6_sk(sk);
549 struct tcp_opt *tp = tcp_sk(sk);
550 struct in6_addr *saddr = NULL;
552 struct dst_entry *dst;
556 if (addr_len < SIN6_LEN_RFC2133)
559 if (usin->sin6_family != AF_INET6)
560 return(-EAFNOSUPPORT);
562 memset(&fl, 0, sizeof(fl));
565 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
566 IP6_ECN_flow_init(fl.fl6_flowlabel);
567 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
568 struct ip6_flowlabel *flowlabel;
569 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
570 if (flowlabel == NULL)
572 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
573 fl6_sock_release(flowlabel);
578 * connect() to INADDR_ANY means loopback (BSD'ism).
581 if(ipv6_addr_any(&usin->sin6_addr))
582 usin->sin6_addr.s6_addr[15] = 0x1;
584 addr_type = ipv6_addr_type(&usin->sin6_addr);
586 if(addr_type & IPV6_ADDR_MULTICAST)
589 if (addr_type&IPV6_ADDR_LINKLOCAL) {
590 if (addr_len >= sizeof(struct sockaddr_in6) &&
591 usin->sin6_scope_id) {
592 /* If interface is set while binding, indices
595 if (sk->sk_bound_dev_if &&
596 sk->sk_bound_dev_if != usin->sin6_scope_id)
599 sk->sk_bound_dev_if = usin->sin6_scope_id;
602 /* Connect to link-local address requires an interface */
603 if (!sk->sk_bound_dev_if)
607 if (tp->ts_recent_stamp &&
608 ipv6_addr_cmp(&np->daddr, &usin->sin6_addr)) {
610 tp->ts_recent_stamp = 0;
614 ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
615 np->flow_label = fl.fl6_flowlabel;
621 if (addr_type == IPV6_ADDR_MAPPED) {
622 u32 exthdrlen = tp->ext_header_len;
623 struct sockaddr_in sin;
625 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
627 if (__ipv6_only_sock(sk))
630 sin.sin_family = AF_INET;
631 sin.sin_port = usin->sin6_port;
632 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
634 tp->af_specific = &ipv6_mapped;
635 sk->sk_backlog_rcv = tcp_v4_do_rcv;
637 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
640 tp->ext_header_len = exthdrlen;
641 tp->af_specific = &ipv6_specific;
642 sk->sk_backlog_rcv = tcp_v6_do_rcv;
645 ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
647 ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
654 if (!ipv6_addr_any(&np->rcv_saddr))
655 saddr = &np->rcv_saddr;
657 fl.proto = IPPROTO_TCP;
658 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
659 ipv6_addr_copy(&fl.fl6_src,
660 (saddr ? saddr : &np->saddr));
661 fl.oif = sk->sk_bound_dev_if;
662 fl.fl_ip_dport = usin->sin6_port;
663 fl.fl_ip_sport = inet->sport;
665 if (np->opt && np->opt->srcrt) {
666 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
667 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
670 err = ip6_dst_lookup(sk, &dst, &fl);
677 ipv6_addr_copy(&np->rcv_saddr, saddr);
680 /* set the source address */
681 ipv6_addr_copy(&np->saddr, saddr);
682 inet->rcv_saddr = LOOPBACK4_IPV6;
684 ip6_dst_store(sk, dst, NULL);
685 sk->sk_route_caps = dst->dev->features &
686 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
688 tp->ext_header_len = 0;
690 tp->ext_header_len = np->opt->opt_flen + np->opt->opt_nflen;
691 tp->ext2_header_len = dst->header_len;
693 tp->mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
695 inet->dport = usin->sin6_port;
697 tcp_set_state(sk, TCP_SYN_SENT);
698 err = tcp_v6_hash_connect(sk);
703 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
708 err = tcp_connect(sk);
715 tcp_set_state(sk, TCP_CLOSE);
719 sk->sk_route_caps = 0;
723 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
724 int type, int code, int offset, __u32 info)
726 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
727 struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
728 struct ipv6_pinfo *np;
734 sk = tcp_v6_lookup(&hdr->daddr, th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
737 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), Icmp6InErrors);
741 if (sk->sk_state == TCP_TIME_WAIT) {
742 tcp_tw_put((struct tcp_tw_bucket*)sk);
747 if (sock_owned_by_user(sk))
748 NET_INC_STATS_BH(LockDroppedIcmps);
750 if (sk->sk_state == TCP_CLOSE)
754 seq = ntohl(th->seq);
755 if (sk->sk_state != TCP_LISTEN &&
756 !between(seq, tp->snd_una, tp->snd_nxt)) {
757 NET_INC_STATS_BH(OutOfWindowIcmps);
763 if (type == ICMPV6_PKT_TOOBIG) {
764 struct dst_entry *dst = NULL;
766 if (sock_owned_by_user(sk))
768 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
771 /* icmp should have updated the destination cache entry */
772 dst = __sk_dst_check(sk, np->dst_cookie);
775 struct inet_opt *inet = inet_sk(sk);
778 /* BUGGG_FUTURE: Again, it is not clear how
779 to handle rthdr case. Ignore this complexity
782 memset(&fl, 0, sizeof(fl));
783 fl.proto = IPPROTO_TCP;
784 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
785 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
786 fl.oif = sk->sk_bound_dev_if;
787 fl.fl_ip_dport = inet->dport;
788 fl.fl_ip_sport = inet->sport;
790 if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
791 sk->sk_err_soft = -err;
797 if (tp->pmtu_cookie > dst_pmtu(dst)) {
798 tcp_sync_mss(sk, dst_pmtu(dst));
799 tcp_simple_retransmit(sk);
800 } /* else let the usual retransmit timer handle it */
805 icmpv6_err_convert(type, code, &err);
807 /* Might be for an open_request */
808 switch (sk->sk_state) {
809 struct open_request *req, **prev;
811 if (sock_owned_by_user(sk))
814 req = tcp_v6_search_req(tp, &prev, th->dest, &hdr->daddr,
815 &hdr->saddr, tcp_v6_iif(skb));
819 /* ICMPs are not backlogged, hence we cannot get
820 * an established socket here.
822 BUG_TRAP(req->sk == NULL);
824 if (seq != req->snt_isn) {
825 NET_INC_STATS_BH(OutOfWindowIcmps);
829 tcp_synq_drop(sk, req, prev);
833 case TCP_SYN_RECV: /* Cannot happen.
834 It can, it SYNs are crossed. --ANK */
835 if (!sock_owned_by_user(sk)) {
836 TCP_INC_STATS_BH(TcpAttemptFails);
838 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
842 sk->sk_err_soft = err;
846 if (!sock_owned_by_user(sk) && np->recverr) {
848 sk->sk_error_report(sk);
850 sk->sk_err_soft = err;
858 static int tcp_v6_send_synack(struct sock *sk, struct open_request *req,
859 struct dst_entry *dst)
861 struct ipv6_pinfo *np = inet6_sk(sk);
862 struct sk_buff * skb;
863 struct ipv6_txoptions *opt = NULL;
867 memset(&fl, 0, sizeof(fl));
868 fl.proto = IPPROTO_TCP;
869 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
870 ipv6_addr_copy(&fl.fl6_src, &req->af.v6_req.loc_addr);
871 fl.fl6_flowlabel = 0;
872 fl.oif = req->af.v6_req.iif;
873 fl.fl_ip_dport = req->rmt_port;
874 fl.fl_ip_sport = inet_sk(sk)->sport;
879 np->rxopt.bits.srcrt == 2 &&
880 req->af.v6_req.pktopts) {
881 struct sk_buff *pktopts = req->af.v6_req.pktopts;
882 struct inet6_skb_parm *rxopt = IP6CB(pktopts);
884 opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(pktopts->nh.raw + rxopt->srcrt));
887 if (opt && opt->srcrt) {
888 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
889 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
892 err = ip6_dst_lookup(sk, &dst, &fl);
897 skb = tcp_make_synack(sk, dst, req);
899 struct tcphdr *th = skb->h.th;
901 th->check = tcp_v6_check(th, skb->len,
902 &req->af.v6_req.loc_addr, &req->af.v6_req.rmt_addr,
903 csum_partial((char *)th, skb->len, skb->csum));
905 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
906 err = ip6_xmit(sk, skb, &fl, opt, 0);
907 if (err == NET_XMIT_CN)
913 if (opt && opt != np->opt)
914 sock_kfree_s(sk, opt, opt->tot_len);
918 static void tcp_v6_or_free(struct open_request *req)
920 if (req->af.v6_req.pktopts)
921 kfree_skb(req->af.v6_req.pktopts);
924 static struct or_calltable or_ipv6 = {
926 .rtx_syn_ack = tcp_v6_send_synack,
927 .send_ack = tcp_v6_or_send_ack,
928 .destructor = tcp_v6_or_free,
929 .send_reset = tcp_v6_send_reset
932 static int ipv6_opt_accepted(struct sock *sk, struct sk_buff *skb)
934 struct ipv6_pinfo *np = inet6_sk(sk);
935 struct inet6_skb_parm *opt = IP6CB(skb);
938 if ((opt->hop && np->rxopt.bits.hopopts) ||
939 ((IPV6_FLOWINFO_MASK&*(u32*)skb->nh.raw) &&
940 np->rxopt.bits.rxflow) ||
941 (opt->srcrt && np->rxopt.bits.srcrt) ||
942 ((opt->dst1 || opt->dst0) && np->rxopt.bits.dstopts))
949 static void tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len,
952 struct ipv6_pinfo *np = inet6_sk(sk);
954 if (skb->ip_summed == CHECKSUM_HW) {
955 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 0);
956 skb->csum = offsetof(struct tcphdr, check);
958 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
959 csum_partial((char *)th, th->doff<<2,
965 static void tcp_v6_send_reset(struct sk_buff *skb)
967 struct tcphdr *th = skb->h.th, *t1;
968 struct sk_buff *buff;
974 if (!ipv6_unicast_destination(skb))
978 * We need to grab some memory, and put together an RST,
979 * and then put it into the queue to be sent.
982 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr), GFP_ATOMIC);
986 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr));
988 t1 = (struct tcphdr *) skb_push(buff,sizeof(struct tcphdr));
990 /* Swap the send and the receive. */
991 memset(t1, 0, sizeof(*t1));
992 t1->dest = th->source;
993 t1->source = th->dest;
994 t1->doff = sizeof(*t1)/4;
998 t1->seq = th->ack_seq;
1001 t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
1002 + skb->len - (th->doff<<2));
1005 buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1007 memset(&fl, 0, sizeof(fl));
1008 ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
1009 ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
1011 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1012 sizeof(*t1), IPPROTO_TCP,
1015 fl.proto = IPPROTO_TCP;
1016 fl.oif = tcp_v6_iif(skb);
1017 fl.fl_ip_dport = t1->dest;
1018 fl.fl_ip_sport = t1->source;
1020 /* sk = NULL, but it is safe for now. RST socket required. */
1021 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1022 ip6_xmit(NULL, buff, &fl, NULL, 0);
1023 TCP_INC_STATS_BH(TcpOutSegs);
1024 TCP_INC_STATS_BH(TcpOutRsts);
1031 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
1033 struct tcphdr *th = skb->h.th, *t1;
1034 struct sk_buff *buff;
1036 int tot_len = sizeof(struct tcphdr);
1038 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr), GFP_ATOMIC);
1042 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr));
1047 t1 = (struct tcphdr *) skb_push(buff,tot_len);
1049 /* Swap the send and the receive. */
1050 memset(t1, 0, sizeof(*t1));
1051 t1->dest = th->source;
1052 t1->source = th->dest;
1053 t1->doff = tot_len/4;
1054 t1->seq = htonl(seq);
1055 t1->ack_seq = htonl(ack);
1057 t1->window = htons(win);
1060 u32 *ptr = (u32*)(t1 + 1);
1061 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1062 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1063 *ptr++ = htonl(tcp_time_stamp);
1067 buff->csum = csum_partial((char *)t1, tot_len, 0);
1069 memset(&fl, 0, sizeof(fl));
1070 ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
1071 ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
1073 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1074 tot_len, IPPROTO_TCP,
1077 fl.proto = IPPROTO_TCP;
1078 fl.oif = tcp_v6_iif(skb);
1079 fl.fl_ip_dport = t1->dest;
1080 fl.fl_ip_sport = t1->source;
1082 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1083 ip6_xmit(NULL, buff, &fl, NULL, 0);
1084 TCP_INC_STATS_BH(TcpOutSegs);
1091 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1093 struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk;
1095 tcp_v6_send_ack(skb, tw->tw_snd_nxt, tw->tw_rcv_nxt,
1096 tw->tw_rcv_wnd >> tw->tw_rcv_wscale, tw->tw_ts_recent);
1101 static void tcp_v6_or_send_ack(struct sk_buff *skb, struct open_request *req)
1103 tcp_v6_send_ack(skb, req->snt_isn+1, req->rcv_isn+1, req->rcv_wnd, req->ts_recent);
1107 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1109 struct open_request *req, **prev;
1110 struct tcphdr *th = skb->h.th;
1111 struct tcp_opt *tp = tcp_sk(sk);
1114 /* Find possible connection requests. */
1115 req = tcp_v6_search_req(tp, &prev, th->source, &skb->nh.ipv6h->saddr,
1116 &skb->nh.ipv6h->daddr, tcp_v6_iif(skb));
1118 return tcp_check_req(sk, skb, req, prev);
1120 nsk = __tcp_v6_lookup_established(&skb->nh.ipv6h->saddr,
1122 &skb->nh.ipv6h->daddr,
1127 if (nsk->sk_state != TCP_TIME_WAIT) {
1131 tcp_tw_put((struct tcp_tw_bucket*)nsk);
1135 #if 0 /*def CONFIG_SYN_COOKIES*/
1136 if (!th->rst && !th->syn && th->ack)
1137 sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
1142 static void tcp_v6_synq_add(struct sock *sk, struct open_request *req)
1144 struct tcp_opt *tp = tcp_sk(sk);
1145 struct tcp_listen_opt *lopt = tp->listen_opt;
1146 u32 h = tcp_v6_synq_hash(&req->af.v6_req.rmt_addr, req->rmt_port, lopt->hash_rnd);
1149 req->expires = jiffies + TCP_TIMEOUT_INIT;
1151 req->dl_next = lopt->syn_table[h];
1153 write_lock(&tp->syn_wait_lock);
1154 lopt->syn_table[h] = req;
1155 write_unlock(&tp->syn_wait_lock);
1161 /* FIXME: this is substantially similar to the ipv4 code.
1162 * Can some kind of merge be done? -- erics
1164 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1166 struct ipv6_pinfo *np = inet6_sk(sk);
1167 struct tcp_opt tmptp, *tp = tcp_sk(sk);
1168 struct open_request *req = NULL;
1169 __u32 isn = TCP_SKB_CB(skb)->when;
1171 if (skb->protocol == htons(ETH_P_IP))
1172 return tcp_v4_conn_request(sk, skb);
1174 if (!ipv6_unicast_destination(skb))
1178 * There are no SYN attacks on IPv6, yet...
1180 if (tcp_synq_is_full(sk) && !isn) {
1181 if (net_ratelimit())
1182 printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
1186 if (sk_acceptq_is_full(sk) && tcp_synq_young(sk) > 1)
1189 req = tcp_openreq_alloc();
1193 tcp_clear_options(&tmptp);
1194 tmptp.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1195 tmptp.user_mss = tp->user_mss;
1197 tcp_parse_options(skb, &tmptp, 0);
1199 tmptp.tstamp_ok = tmptp.saw_tstamp;
1200 tcp_openreq_init(req, &tmptp, skb);
1202 req->class = &or_ipv6;
1203 ipv6_addr_copy(&req->af.v6_req.rmt_addr, &skb->nh.ipv6h->saddr);
1204 ipv6_addr_copy(&req->af.v6_req.loc_addr, &skb->nh.ipv6h->daddr);
1205 TCP_ECN_create_request(req, skb->h.th);
1206 req->af.v6_req.pktopts = NULL;
1207 if (ipv6_opt_accepted(sk, skb) ||
1208 np->rxopt.bits.rxinfo ||
1209 np->rxopt.bits.rxhlim) {
1210 atomic_inc(&skb->users);
1211 req->af.v6_req.pktopts = skb;
1213 req->af.v6_req.iif = sk->sk_bound_dev_if;
1215 /* So that link locals have meaning */
1216 if (!sk->sk_bound_dev_if &&
1217 ipv6_addr_type(&req->af.v6_req.rmt_addr) & IPV6_ADDR_LINKLOCAL)
1218 req->af.v6_req.iif = tcp_v6_iif(skb);
1221 isn = tcp_v6_init_sequence(sk,skb);
1225 if (tcp_v6_send_synack(sk, req, NULL))
1228 tcp_v6_synq_add(sk, req);
1234 tcp_openreq_free(req);
1236 TCP_INC_STATS_BH(TcpAttemptFails);
1237 return 0; /* don't send reset */
1240 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1241 struct open_request *req,
1242 struct dst_entry *dst)
1244 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1245 struct tcp6_sock *newtcp6sk;
1246 struct inet_opt *newinet;
1247 struct tcp_opt *newtp;
1249 struct ipv6_txoptions *opt;
1251 if (skb->protocol == htons(ETH_P_IP)) {
1256 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1261 newtcp6sk = (struct tcp6_sock *)newsk;
1262 newtcp6sk->pinet6 = &newtcp6sk->inet6;
1264 newinet = inet_sk(newsk);
1265 newnp = inet6_sk(newsk);
1266 newtp = tcp_sk(newsk);
1268 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1270 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1273 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1276 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1278 newtp->af_specific = &ipv6_mapped;
1279 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1280 newnp->pktoptions = NULL;
1282 newnp->mcast_oif = tcp_v6_iif(skb);
1283 newnp->mcast_hops = skb->nh.ipv6h->hop_limit;
1285 /* Charge newly allocated IPv6 socket. Though it is mapped,
1288 #ifdef INET_REFCNT_DEBUG
1289 atomic_inc(&inet6_sock_nr);
1292 /* It is tricky place. Until this moment IPv4 tcp
1293 worked with IPv6 af_tcp.af_specific.
1296 tcp_sync_mss(newsk, newtp->pmtu_cookie);
1303 if (sk_acceptq_is_full(sk))
1306 if (np->rxopt.bits.srcrt == 2 &&
1307 opt == NULL && req->af.v6_req.pktopts) {
1308 struct inet6_skb_parm *rxopt = IP6CB(req->af.v6_req.pktopts);
1310 opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(req->af.v6_req.pktopts->nh.raw+rxopt->srcrt));
1316 memset(&fl, 0, sizeof(fl));
1317 fl.proto = IPPROTO_TCP;
1318 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
1319 if (opt && opt->srcrt) {
1320 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1321 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1323 ipv6_addr_copy(&fl.fl6_src, &req->af.v6_req.loc_addr);
1324 fl.oif = sk->sk_bound_dev_if;
1325 fl.fl_ip_dport = req->rmt_port;
1326 fl.fl_ip_sport = inet_sk(sk)->sport;
1328 if (ip6_dst_lookup(sk, &dst, &fl))
1332 newsk = tcp_create_openreq_child(sk, req, skb);
1336 /* Charge newly allocated IPv6 socket */
1337 #ifdef INET_REFCNT_DEBUG
1338 atomic_inc(&inet6_sock_nr);
1341 ip6_dst_store(newsk, dst, NULL);
1342 newsk->sk_route_caps = dst->dev->features &
1343 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1345 newtcp6sk = (struct tcp6_sock *)newsk;
1346 newtcp6sk->pinet6 = &newtcp6sk->inet6;
1348 newtp = tcp_sk(newsk);
1349 newinet = inet_sk(newsk);
1350 newnp = inet6_sk(newsk);
1352 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1354 ipv6_addr_copy(&newnp->daddr, &req->af.v6_req.rmt_addr);
1355 ipv6_addr_copy(&newnp->saddr, &req->af.v6_req.loc_addr);
1356 ipv6_addr_copy(&newnp->rcv_saddr, &req->af.v6_req.loc_addr);
1357 newsk->sk_bound_dev_if = req->af.v6_req.iif;
1359 /* Now IPv6 options...
1361 First: no IPv4 options.
1363 newinet->opt = NULL;
1366 newnp->rxopt.all = np->rxopt.all;
1368 /* Clone pktoptions received with SYN */
1369 newnp->pktoptions = NULL;
1370 if (req->af.v6_req.pktopts) {
1371 newnp->pktoptions = skb_clone(req->af.v6_req.pktopts,
1373 kfree_skb(req->af.v6_req.pktopts);
1374 req->af.v6_req.pktopts = NULL;
1375 if (newnp->pktoptions)
1376 skb_set_owner_r(newnp->pktoptions, newsk);
1379 newnp->mcast_oif = tcp_v6_iif(skb);
1380 newnp->mcast_hops = skb->nh.ipv6h->hop_limit;
1382 /* Clone native IPv6 options from listening socket (if any)
1384 Yes, keeping reference count would be much more clever,
1385 but we make one more one thing there: reattach optmem
1389 newnp->opt = ipv6_dup_options(newsk, opt);
1391 sock_kfree_s(sk, opt, opt->tot_len);
1394 newtp->ext_header_len = 0;
1396 newtp->ext_header_len = newnp->opt->opt_nflen +
1397 newnp->opt->opt_flen;
1398 newtp->ext2_header_len = dst->header_len;
1400 tcp_sync_mss(newsk, dst_pmtu(dst));
1401 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1402 tcp_initialize_rcv_mss(newsk);
1404 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1406 __tcp_v6_hash(newsk);
1407 tcp_inherit_port(sk, newsk);
1412 NET_INC_STATS_BH(ListenOverflows);
1414 NET_INC_STATS_BH(ListenDrops);
1415 if (opt && opt != np->opt)
1416 sock_kfree_s(sk, opt, opt->tot_len);
1421 static int tcp_v6_checksum_init(struct sk_buff *skb)
1423 if (skb->ip_summed == CHECKSUM_HW) {
1424 skb->ip_summed = CHECKSUM_UNNECESSARY;
1425 if (!tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1426 &skb->nh.ipv6h->daddr,skb->csum))
1428 LIMIT_NETDEBUG(printk(KERN_DEBUG "hw tcp v6 csum failed\n"));
1430 if (skb->len <= 76) {
1431 if (tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1432 &skb->nh.ipv6h->daddr,skb_checksum(skb, 0, skb->len, 0)))
1434 skb->ip_summed = CHECKSUM_UNNECESSARY;
1436 skb->csum = ~tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1437 &skb->nh.ipv6h->daddr,0);
1442 /* The socket must have it's spinlock held when we get
1445 * We have a potential double-lock case here, so even when
1446 * doing backlog processing we use the BH locking scheme.
1447 * This is because we cannot sleep with the original spinlock
1450 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1452 struct ipv6_pinfo *np = inet6_sk(sk);
1454 struct sk_buff *opt_skb = NULL;
1456 /* Imagine: socket is IPv6. IPv4 packet arrives,
1457 goes to IPv4 receive handler and backlogged.
1458 From backlog it always goes here. Kerboom...
1459 Fortunately, tcp_rcv_established and rcv_established
1460 handle them correctly, but it is not case with
1461 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1464 if (skb->protocol == htons(ETH_P_IP))
1465 return tcp_v4_do_rcv(sk, skb);
1467 if (sk_filter(sk, skb, 0))
1471 * socket locking is here for SMP purposes as backlog rcv
1472 * is currently called with bh processing disabled.
1475 /* Do Stevens' IPV6_PKTOPTIONS.
1477 Yes, guys, it is the only place in our code, where we
1478 may make it not affecting IPv4.
1479 The rest of code is protocol independent,
1480 and I do not like idea to uglify IPv4.
1482 Actually, all the idea behind IPV6_PKTOPTIONS
1483 looks not very well thought. For now we latch
1484 options, received in the last packet, enqueued
1485 by tcp. Feel free to propose better solution.
1489 opt_skb = skb_clone(skb, GFP_ATOMIC);
1491 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1492 TCP_CHECK_TIMER(sk);
1493 if (tcp_rcv_established(sk, skb, skb->h.th, skb->len))
1495 TCP_CHECK_TIMER(sk);
1497 goto ipv6_pktoptions;
1501 if (skb->len < (skb->h.th->doff<<2) || tcp_checksum_complete(skb))
1504 if (sk->sk_state == TCP_LISTEN) {
1505 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1510 * Queue it on the new socket if the new socket is active,
1511 * otherwise we just shortcircuit this and continue with
1515 if (tcp_child_process(sk, nsk, skb))
1518 __kfree_skb(opt_skb);
1523 TCP_CHECK_TIMER(sk);
1524 if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len))
1526 TCP_CHECK_TIMER(sk);
1528 goto ipv6_pktoptions;
1532 tcp_v6_send_reset(skb);
1535 __kfree_skb(opt_skb);
1539 TCP_INC_STATS_BH(TcpInErrs);
1544 /* Do you ask, what is it?
1546 1. skb was enqueued by tcp.
1547 2. skb is added to tail of read queue, rather than out of order.
1548 3. socket is not in passive state.
1549 4. Finally, it really contains options, which user wants to receive.
1552 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1553 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1554 if (np->rxopt.bits.rxinfo)
1555 np->mcast_oif = tcp_v6_iif(opt_skb);
1556 if (np->rxopt.bits.rxhlim)
1557 np->mcast_hops = opt_skb->nh.ipv6h->hop_limit;
1558 if (ipv6_opt_accepted(sk, opt_skb)) {
1559 skb_set_owner_r(opt_skb, sk);
1560 opt_skb = xchg(&np->pktoptions, opt_skb);
1562 __kfree_skb(opt_skb);
1563 opt_skb = xchg(&np->pktoptions, NULL);
1572 static int tcp_v6_rcv(struct sk_buff **pskb, unsigned int *nhoffp)
1574 struct sk_buff *skb = *pskb;
1579 if (skb->pkt_type != PACKET_HOST)
1583 * Count it even if it's bad.
1585 TCP_INC_STATS_BH(TcpInSegs);
1587 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1592 if (th->doff < sizeof(struct tcphdr)/4)
1594 if (!pskb_may_pull(skb, th->doff*4))
1597 if ((skb->ip_summed != CHECKSUM_UNNECESSARY &&
1598 tcp_v6_checksum_init(skb) < 0))
1602 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1603 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1604 skb->len - th->doff*4);
1605 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1606 TCP_SKB_CB(skb)->when = 0;
1607 TCP_SKB_CB(skb)->flags = ip6_get_dsfield(skb->nh.ipv6h);
1608 TCP_SKB_CB(skb)->sacked = 0;
1610 sk = __tcp_v6_lookup(&skb->nh.ipv6h->saddr, th->source,
1611 &skb->nh.ipv6h->daddr, ntohs(th->dest), tcp_v6_iif(skb));
1617 if (sk->sk_state == TCP_TIME_WAIT)
1620 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1621 goto discard_and_relse;
1623 if (sk_filter(sk, skb, 0))
1624 goto discard_and_relse;
1630 if (!sock_owned_by_user(sk)) {
1631 if (!tcp_prequeue(sk, skb))
1632 ret = tcp_v6_do_rcv(sk, skb);
1634 sk_add_backlog(sk, skb);
1638 return ret ? -1 : 0;
1641 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1644 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1646 TCP_INC_STATS_BH(TcpInErrs);
1648 tcp_v6_send_reset(skb);
1665 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1666 tcp_tw_put((struct tcp_tw_bucket *) sk);
1670 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1671 TCP_INC_STATS_BH(TcpInErrs);
1672 tcp_tw_put((struct tcp_tw_bucket *) sk);
1676 switch(tcp_timewait_state_process((struct tcp_tw_bucket *)sk,
1677 skb, th, skb->len)) {
1682 sk2 = tcp_v6_lookup_listener(&skb->nh.ipv6h->daddr, ntohs(th->dest), tcp_v6_iif(skb));
1684 tcp_tw_deschedule((struct tcp_tw_bucket *)sk);
1685 tcp_tw_put((struct tcp_tw_bucket *)sk);
1689 /* Fall through to ACK */
1692 tcp_v6_timewait_ack(sk, skb);
1696 case TCP_TW_SUCCESS:;
1701 static int tcp_v6_rebuild_header(struct sock *sk)
1704 struct dst_entry *dst;
1705 struct ipv6_pinfo *np = inet6_sk(sk);
1707 dst = __sk_dst_check(sk, np->dst_cookie);
1710 struct inet_opt *inet = inet_sk(sk);
1713 memset(&fl, 0, sizeof(fl));
1714 fl.proto = IPPROTO_TCP;
1715 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1716 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1717 fl.fl6_flowlabel = np->flow_label;
1718 fl.oif = sk->sk_bound_dev_if;
1719 fl.fl_ip_dport = inet->dport;
1720 fl.fl_ip_sport = inet->sport;
1722 if (np->opt && np->opt->srcrt) {
1723 struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1724 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1727 err = ip6_dst_lookup(sk, &dst, &fl);
1730 sk->sk_route_caps = 0;
1734 ip6_dst_store(sk, dst, NULL);
1735 sk->sk_route_caps = dst->dev->features &
1736 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1737 tcp_sk(sk)->ext2_header_len = dst->header_len;
1743 static int tcp_v6_xmit(struct sk_buff *skb, int ipfragok)
1745 struct sock *sk = skb->sk;
1746 struct inet_opt *inet = inet_sk(sk);
1747 struct ipv6_pinfo *np = inet6_sk(sk);
1749 struct dst_entry *dst;
1751 memset(&fl, 0, sizeof(fl));
1752 fl.proto = IPPROTO_TCP;
1753 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1754 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1755 fl.fl6_flowlabel = np->flow_label;
1756 IP6_ECN_flow_xmit(sk, fl.fl6_flowlabel);
1757 fl.oif = sk->sk_bound_dev_if;
1758 fl.fl_ip_sport = inet->sport;
1759 fl.fl_ip_dport = inet->dport;
1761 if (np->opt && np->opt->srcrt) {
1762 struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1763 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1766 dst = __sk_dst_check(sk, np->dst_cookie);
1769 int err = ip6_dst_lookup(sk, &dst, &fl);
1772 sk->sk_err_soft = -err;
1776 ip6_dst_store(sk, dst, NULL);
1777 sk->sk_route_caps = dst->dev->features &
1778 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1779 tcp_sk(sk)->ext2_header_len = dst->header_len;
1782 skb->dst = dst_clone(dst);
1784 /* Restore final destination back after routing done */
1785 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1787 return ip6_xmit(sk, skb, &fl, np->opt, 0);
1790 static void v6_addr2sockaddr(struct sock *sk, struct sockaddr * uaddr)
1792 struct ipv6_pinfo *np = inet6_sk(sk);
1793 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) uaddr;
1795 sin6->sin6_family = AF_INET6;
1796 ipv6_addr_copy(&sin6->sin6_addr, &np->daddr);
1797 sin6->sin6_port = inet_sk(sk)->dport;
1798 /* We do not store received flowlabel for TCP */
1799 sin6->sin6_flowinfo = 0;
1800 sin6->sin6_scope_id = 0;
1801 if (sk->sk_bound_dev_if &&
1802 ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
1803 sin6->sin6_scope_id = sk->sk_bound_dev_if;
1806 static int tcp_v6_remember_stamp(struct sock *sk)
1808 /* Alas, not yet... */
1812 static struct tcp_func ipv6_specific = {
1813 .queue_xmit = tcp_v6_xmit,
1814 .send_check = tcp_v6_send_check,
1815 .rebuild_header = tcp_v6_rebuild_header,
1816 .conn_request = tcp_v6_conn_request,
1817 .syn_recv_sock = tcp_v6_syn_recv_sock,
1818 .remember_stamp = tcp_v6_remember_stamp,
1819 .net_header_len = sizeof(struct ipv6hdr),
1821 .setsockopt = ipv6_setsockopt,
1822 .getsockopt = ipv6_getsockopt,
1823 .addr2sockaddr = v6_addr2sockaddr,
1824 .sockaddr_len = sizeof(struct sockaddr_in6)
1828 * TCP over IPv4 via INET6 API
1831 static struct tcp_func ipv6_mapped = {
1832 .queue_xmit = ip_queue_xmit,
1833 .send_check = tcp_v4_send_check,
1834 .rebuild_header = tcp_v4_rebuild_header,
1835 .conn_request = tcp_v6_conn_request,
1836 .syn_recv_sock = tcp_v6_syn_recv_sock,
1837 .remember_stamp = tcp_v4_remember_stamp,
1838 .net_header_len = sizeof(struct iphdr),
1840 .setsockopt = ipv6_setsockopt,
1841 .getsockopt = ipv6_getsockopt,
1842 .addr2sockaddr = v6_addr2sockaddr,
1843 .sockaddr_len = sizeof(struct sockaddr_in6)
1848 /* NOTE: A lot of things set to zero explicitly by call to
1849 * sk_alloc() so need not be done here.
1851 static int tcp_v6_init_sock(struct sock *sk)
1853 struct tcp_opt *tp = tcp_sk(sk);
1855 skb_queue_head_init(&tp->out_of_order_queue);
1856 tcp_init_xmit_timers(sk);
1857 tcp_prequeue_init(tp);
1859 tp->rto = TCP_TIMEOUT_INIT;
1860 tp->mdev = TCP_TIMEOUT_INIT;
1862 /* So many TCP implementations out there (incorrectly) count the
1863 * initial SYN frame in their delayed-ACK and congestion control
1864 * algorithms that we must have the following bandaid to talk
1865 * efficiently to them. -DaveM
1869 /* See draft-stevens-tcpca-spec-01 for discussion of the
1870 * initialization of these values.
1872 tp->snd_ssthresh = 0x7fffffff;
1873 tp->snd_cwnd_clamp = ~0;
1874 tp->mss_cache = 536;
1876 tp->reordering = sysctl_tcp_reordering;
1878 sk->sk_state = TCP_CLOSE;
1880 tp->af_specific = &ipv6_specific;
1882 sk->sk_write_space = sk_stream_write_space;
1883 sk->sk_use_write_queue = 1;
1885 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1886 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1888 atomic_inc(&tcp_sockets_allocated);
1893 static int tcp_v6_destroy_sock(struct sock *sk)
1895 extern int tcp_v4_destroy_sock(struct sock *sk);
1897 tcp_v4_destroy_sock(sk);
1898 return inet6_destroy_sock(sk);
1901 /* Proc filesystem TCPv6 sock list dumping. */
1902 static void get_openreq6(struct seq_file *seq,
1903 struct sock *sk, struct open_request *req, int i, int uid)
1905 struct in6_addr *dest, *src;
1906 int ttd = req->expires - jiffies;
1911 src = &req->af.v6_req.loc_addr;
1912 dest = &req->af.v6_req.rmt_addr;
1914 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1915 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1917 src->s6_addr32[0], src->s6_addr32[1],
1918 src->s6_addr32[2], src->s6_addr32[3],
1919 ntohs(inet_sk(sk)->sport),
1920 dest->s6_addr32[0], dest->s6_addr32[1],
1921 dest->s6_addr32[2], dest->s6_addr32[3],
1922 ntohs(req->rmt_port),
1924 0,0, /* could print option size, but that is af dependent. */
1925 1, /* timers active (only the expire timer) */
1926 jiffies_to_clock_t(ttd),
1929 0, /* non standard timer */
1930 0, /* open_requests have no inode */
1934 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1936 struct in6_addr *dest, *src;
1939 unsigned long timer_expires;
1940 struct inet_opt *inet = inet_sk(sp);
1941 struct tcp_opt *tp = tcp_sk(sp);
1942 struct ipv6_pinfo *np = inet6_sk(sp);
1945 src = &np->rcv_saddr;
1946 destp = ntohs(inet->dport);
1947 srcp = ntohs(inet->sport);
1948 if (tp->pending == TCP_TIME_RETRANS) {
1950 timer_expires = tp->timeout;
1951 } else if (tp->pending == TCP_TIME_PROBE0) {
1953 timer_expires = tp->timeout;
1954 } else if (timer_pending(&sp->sk_timer)) {
1956 timer_expires = sp->sk_timer.expires;
1959 timer_expires = jiffies;
1963 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1964 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
1966 src->s6_addr32[0], src->s6_addr32[1],
1967 src->s6_addr32[2], src->s6_addr32[3], srcp,
1968 dest->s6_addr32[0], dest->s6_addr32[1],
1969 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1971 tp->write_seq-tp->snd_una, tp->rcv_nxt-tp->copied_seq,
1973 jiffies_to_clock_t(timer_expires - jiffies),
1978 atomic_read(&sp->sk_refcnt), sp,
1979 tp->rto, tp->ack.ato, (tp->ack.quick<<1)|tp->ack.pingpong,
1980 tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
1984 static void get_timewait6_sock(struct seq_file *seq,
1985 struct tcp_tw_bucket *tw, int i)
1987 struct in6_addr *dest, *src;
1989 int ttd = tw->tw_ttd - jiffies;
1994 dest = &tw->tw_v6_daddr;
1995 src = &tw->tw_v6_rcv_saddr;
1996 destp = ntohs(tw->tw_dport);
1997 srcp = ntohs(tw->tw_sport);
2000 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2001 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2003 src->s6_addr32[0], src->s6_addr32[1],
2004 src->s6_addr32[2], src->s6_addr32[3], srcp,
2005 dest->s6_addr32[0], dest->s6_addr32[1],
2006 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2007 tw->tw_substate, 0, 0,
2008 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2009 atomic_read(&tw->tw_refcnt), tw);
2012 #ifdef CONFIG_PROC_FS
2013 static int tcp6_seq_show(struct seq_file *seq, void *v)
2015 struct tcp_iter_state *st;
2017 if (v == SEQ_START_TOKEN) {
2022 "st tx_queue rx_queue tr tm->when retrnsmt"
2023 " uid timeout inode\n");
2028 switch (st->state) {
2029 case TCP_SEQ_STATE_LISTENING:
2030 case TCP_SEQ_STATE_ESTABLISHED:
2031 get_tcp6_sock(seq, v, st->num);
2033 case TCP_SEQ_STATE_OPENREQ:
2034 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2036 case TCP_SEQ_STATE_TIME_WAIT:
2037 get_timewait6_sock(seq, v, st->num);
2044 static struct file_operations tcp6_seq_fops;
2045 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2046 .owner = THIS_MODULE,
2049 .seq_show = tcp6_seq_show,
2050 .seq_fops = &tcp6_seq_fops,
2053 int __init tcp6_proc_init(void)
2055 return tcp_proc_register(&tcp6_seq_afinfo);
2058 void tcp6_proc_exit(void)
2060 tcp_proc_unregister(&tcp6_seq_afinfo);
2064 struct proto tcpv6_prot = {
2067 .connect = tcp_v6_connect,
2068 .disconnect = tcp_disconnect,
2069 .accept = tcp_accept,
2071 .init = tcp_v6_init_sock,
2072 .destroy = tcp_v6_destroy_sock,
2073 .shutdown = tcp_shutdown,
2074 .setsockopt = tcp_setsockopt,
2075 .getsockopt = tcp_getsockopt,
2076 .sendmsg = tcp_sendmsg,
2077 .recvmsg = tcp_recvmsg,
2078 .backlog_rcv = tcp_v6_do_rcv,
2079 .hash = tcp_v6_hash,
2080 .unhash = tcp_unhash,
2081 .get_port = tcp_v6_get_port,
2082 .enter_memory_pressure = tcp_enter_memory_pressure,
2083 .sockets_allocated = &tcp_sockets_allocated,
2084 .memory_allocated = &tcp_memory_allocated,
2085 .memory_pressure = &tcp_memory_pressure,
2086 .sysctl_mem = sysctl_tcp_mem,
2087 .sysctl_wmem = sysctl_tcp_wmem,
2088 .sysctl_rmem = sysctl_tcp_rmem,
2089 .max_header = MAX_TCP_HEADER,
2092 static struct inet6_protocol tcpv6_protocol = {
2093 .handler = tcp_v6_rcv,
2094 .err_handler = tcp_v6_err,
2095 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2098 extern struct proto_ops inet6_stream_ops;
2100 static struct inet_protosw tcpv6_protosw = {
2101 .type = SOCK_STREAM,
2102 .protocol = IPPROTO_TCP,
2103 .prot = &tcpv6_prot,
2104 .ops = &inet6_stream_ops,
2107 .flags = INET_PROTOSW_PERMANENT,
2110 void __init tcpv6_init(void)
2112 /* register inet6 protocol */
2113 if (inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP) < 0)
2114 printk(KERN_ERR "tcpv6_init: Could not register protocol\n");
2115 inet6_register_protosw(&tcpv6_protosw);