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>
58 #include <net/addrconf.h>
60 #include <net/dsfield.h>
62 #include <asm/uaccess.h>
64 #include <linux/proc_fs.h>
65 #include <linux/seq_file.h>
67 static void tcp_v6_send_reset(struct sk_buff *skb);
68 static void tcp_v6_or_send_ack(struct sk_buff *skb, struct open_request *req);
69 static void tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len,
72 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
73 static int tcp_v6_xmit(struct sk_buff *skb, int ipfragok);
75 static struct tcp_func ipv6_mapped;
76 static struct tcp_func ipv6_specific;
78 /* I have no idea if this is a good hash for v6 or not. -DaveM */
79 static __inline__ int tcp_v6_hashfn(struct in6_addr *laddr, u16 lport,
80 struct in6_addr *faddr, u16 fport)
82 int hashent = (lport ^ fport);
84 hashent ^= (laddr->s6_addr32[3] ^ faddr->s6_addr32[3]);
85 hashent ^= hashent>>16;
86 hashent ^= hashent>>8;
87 return (hashent & (tcp_ehash_size - 1));
90 static __inline__ int tcp_v6_sk_hashfn(struct sock *sk)
92 struct inet_opt *inet = inet_sk(sk);
93 struct ipv6_pinfo *np = inet6_sk(sk);
94 struct in6_addr *laddr = &np->rcv_saddr;
95 struct in6_addr *faddr = &np->daddr;
96 __u16 lport = inet->num;
97 __u16 fport = inet->dport;
98 return tcp_v6_hashfn(laddr, lport, faddr, fport);
101 static inline int tcp_v6_bind_conflict(struct sock *sk,
102 struct tcp_bind_bucket *tb)
105 struct hlist_node *node;
107 /* We must walk the whole port owner list in this case. -DaveM */
108 sk_for_each_bound(sk2, node, &tb->owners) {
110 (!sk->sk_bound_dev_if ||
111 !sk2->sk_bound_dev_if ||
112 sk->sk_bound_dev_if == sk2->sk_bound_dev_if) &&
113 (!sk->sk_reuse || !sk2->sk_reuse ||
114 sk2->sk_state == TCP_LISTEN) &&
115 ipv6_rcv_saddr_equal(sk, sk2))
122 /* Grrr, addr_type already calculated by caller, but I don't want
123 * to add some silly "cookie" argument to this method just for that.
124 * But it doesn't matter, the recalculation is in the rarest path
125 * this function ever takes.
127 static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
129 struct tcp_bind_hashbucket *head;
130 struct tcp_bind_bucket *tb;
131 struct hlist_node *node;
136 int low = sysctl_local_port_range[0];
137 int high = sysctl_local_port_range[1];
138 int remaining = (high - low) + 1;
141 spin_lock(&tcp_portalloc_lock);
142 rover = tcp_port_rover;
144 if ((rover < low) || (rover > high))
146 head = &tcp_bhash[tcp_bhashfn(rover)];
147 spin_lock(&head->lock);
148 tb_for_each(tb, node, &head->chain)
149 if (tb->port == rover)
153 spin_unlock(&head->lock);
154 } while (--remaining > 0);
155 tcp_port_rover = rover;
156 spin_unlock(&tcp_portalloc_lock);
158 /* Exhausted local port range during search? */
163 /* OK, here is the one we will use. */
166 head = &tcp_bhash[tcp_bhashfn(snum)];
167 spin_lock(&head->lock);
168 tb_for_each(tb, node, &head->chain)
169 if (tb->port == snum)
175 if (tb && !hlist_empty(&tb->owners)) {
176 if (tb->fastreuse > 0 && sk->sk_reuse &&
177 sk->sk_state != TCP_LISTEN) {
181 if (tcp_v6_bind_conflict(sk, tb))
187 if (!tb && (tb = tcp_bucket_create(head, snum)) == NULL)
189 if (hlist_empty(&tb->owners)) {
190 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
194 } else if (tb->fastreuse &&
195 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
199 if (!tcp_sk(sk)->bind_hash)
200 tcp_bind_hash(sk, tb, snum);
201 BUG_TRAP(tcp_sk(sk)->bind_hash == tb);
205 spin_unlock(&head->lock);
211 static __inline__ void __tcp_v6_hash(struct sock *sk)
213 struct hlist_head *list;
216 BUG_TRAP(sk_unhashed(sk));
218 if (sk->sk_state == TCP_LISTEN) {
219 list = &tcp_listening_hash[tcp_sk_listen_hashfn(sk)];
220 lock = &tcp_lhash_lock;
223 sk->sk_hashent = tcp_v6_sk_hashfn(sk);
224 list = &tcp_ehash[sk->sk_hashent].chain;
225 lock = &tcp_ehash[sk->sk_hashent].lock;
229 __sk_add_node(sk, list);
230 sock_prot_inc_use(sk->sk_prot);
235 static void tcp_v6_hash(struct sock *sk)
237 if (sk->sk_state != TCP_CLOSE) {
238 struct tcp_opt *tp = tcp_sk(sk);
240 if (tp->af_specific == &ipv6_mapped) {
250 static struct sock *tcp_v6_lookup_listener(struct in6_addr *daddr, unsigned short hnum, int dif)
253 struct hlist_node *node;
254 struct sock *result = NULL;
258 read_lock(&tcp_lhash_lock);
259 sk_for_each(sk, node, &tcp_listening_hash[tcp_lhashfn(hnum)]) {
260 if (inet_sk(sk)->num == hnum && sk->sk_family == PF_INET6) {
261 struct ipv6_pinfo *np = inet6_sk(sk);
264 if (!ipv6_addr_any(&np->rcv_saddr)) {
265 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
269 if (sk->sk_bound_dev_if) {
270 if (sk->sk_bound_dev_if != dif)
278 if (score > hiscore) {
286 read_unlock(&tcp_lhash_lock);
290 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
291 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
293 * The sockhash lock must be held as a reader here.
296 static inline struct sock *__tcp_v6_lookup_established(struct in6_addr *saddr, u16 sport,
297 struct in6_addr *daddr, u16 hnum,
300 struct tcp_ehash_bucket *head;
302 struct hlist_node *node;
303 __u32 ports = TCP_COMBINED_PORTS(sport, hnum);
306 /* Optimize here for direct hit, only listening connections can
307 * have wildcards anyways.
309 hash = tcp_v6_hashfn(daddr, hnum, saddr, sport);
310 head = &tcp_ehash[hash];
311 read_lock(&head->lock);
312 sk_for_each(sk, node, &head->chain) {
313 /* For IPV6 do the cheaper port and family tests first. */
314 if(TCP_IPV6_MATCH(sk, saddr, daddr, ports, dif))
315 goto hit; /* You sunk my battleship! */
317 /* Must check for a TIME_WAIT'er before going to listener hash. */
318 sk_for_each(sk, node, &(head + tcp_ehash_size)->chain) {
319 /* FIXME: acme: check this... */
320 struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk;
322 if(*((__u32 *)&(tw->tw_dport)) == ports &&
323 sk->sk_family == PF_INET6) {
324 if(ipv6_addr_equal(&tw->tw_v6_daddr, saddr) &&
325 ipv6_addr_equal(&tw->tw_v6_rcv_saddr, daddr) &&
326 (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif))
330 read_unlock(&head->lock);
335 read_unlock(&head->lock);
340 static inline struct sock *__tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
341 struct in6_addr *daddr, u16 hnum,
346 sk = __tcp_v6_lookup_established(saddr, sport, daddr, hnum, dif);
351 return tcp_v6_lookup_listener(daddr, hnum, dif);
354 inline struct sock *tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
355 struct in6_addr *daddr, u16 dport,
361 sk = __tcp_v6_lookup(saddr, sport, daddr, ntohs(dport), dif);
367 EXPORT_SYMBOL_GPL(tcp_v6_lookup);
371 * Open request hash tables.
374 static u32 tcp_v6_synq_hash(struct in6_addr *raddr, u16 rport, u32 rnd)
378 a = raddr->s6_addr32[0];
379 b = raddr->s6_addr32[1];
380 c = raddr->s6_addr32[2];
382 a += JHASH_GOLDEN_RATIO;
383 b += JHASH_GOLDEN_RATIO;
385 __jhash_mix(a, b, c);
387 a += raddr->s6_addr32[3];
389 __jhash_mix(a, b, c);
391 return c & (TCP_SYNQ_HSIZE - 1);
394 static struct open_request *tcp_v6_search_req(struct tcp_opt *tp,
395 struct open_request ***prevp,
397 struct in6_addr *raddr,
398 struct in6_addr *laddr,
401 struct tcp_listen_opt *lopt = tp->listen_opt;
402 struct open_request *req, **prev;
404 for (prev = &lopt->syn_table[tcp_v6_synq_hash(raddr, rport, lopt->hash_rnd)];
405 (req = *prev) != NULL;
406 prev = &req->dl_next) {
407 if (req->rmt_port == rport &&
408 req->class->family == AF_INET6 &&
409 ipv6_addr_equal(&req->af.v6_req.rmt_addr, raddr) &&
410 ipv6_addr_equal(&req->af.v6_req.loc_addr, laddr) &&
411 (!req->af.v6_req.iif || req->af.v6_req.iif == iif)) {
412 BUG_TRAP(req->sk == NULL);
421 static __inline__ u16 tcp_v6_check(struct tcphdr *th, int len,
422 struct in6_addr *saddr,
423 struct in6_addr *daddr,
426 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
429 static __u32 tcp_v6_init_sequence(struct sock *sk, struct sk_buff *skb)
431 if (skb->protocol == htons(ETH_P_IPV6)) {
432 return secure_tcpv6_sequence_number(skb->nh.ipv6h->daddr.s6_addr32,
433 skb->nh.ipv6h->saddr.s6_addr32,
437 return secure_tcp_sequence_number(skb->nh.iph->daddr,
444 static int tcp_v6_check_established(struct sock *sk)
446 struct inet_opt *inet = inet_sk(sk);
447 struct ipv6_pinfo *np = inet6_sk(sk);
448 struct in6_addr *daddr = &np->rcv_saddr;
449 struct in6_addr *saddr = &np->daddr;
450 int dif = sk->sk_bound_dev_if;
451 u32 ports = TCP_COMBINED_PORTS(inet->dport, inet->num);
452 int hash = tcp_v6_hashfn(daddr, inet->num, saddr, inet->dport);
453 struct tcp_ehash_bucket *head = &tcp_ehash[hash];
455 struct hlist_node *node;
456 struct tcp_tw_bucket *tw;
458 write_lock_bh(&head->lock);
460 /* Check TIME-WAIT sockets first. */
461 sk_for_each(sk2, node, &(head + tcp_ehash_size)->chain) {
462 tw = (struct tcp_tw_bucket*)sk2;
464 if(*((__u32 *)&(tw->tw_dport)) == ports &&
465 sk2->sk_family == PF_INET6 &&
466 ipv6_addr_equal(&tw->tw_v6_daddr, saddr) &&
467 ipv6_addr_equal(&tw->tw_v6_rcv_saddr, daddr) &&
468 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) {
469 struct tcp_opt *tp = tcp_sk(sk);
471 if (tw->tw_ts_recent_stamp) {
472 /* See comment in tcp_ipv4.c */
473 tp->write_seq = tw->tw_snd_nxt + 65535 + 2;
476 tp->ts_recent = tw->tw_ts_recent;
477 tp->ts_recent_stamp = tw->tw_ts_recent_stamp;
486 /* And established part... */
487 sk_for_each(sk2, node, &head->chain) {
488 if(TCP_IPV6_MATCH(sk2, saddr, daddr, ports, dif))
493 BUG_TRAP(sk_unhashed(sk));
494 __sk_add_node(sk, &head->chain);
495 sk->sk_hashent = hash;
496 sock_prot_inc_use(sk->sk_prot);
497 write_unlock_bh(&head->lock);
500 /* Silly. Should hash-dance instead... */
502 tcp_tw_deschedule(tw);
503 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
511 write_unlock_bh(&head->lock);
512 return -EADDRNOTAVAIL;
515 static int tcp_v6_hash_connect(struct sock *sk)
517 struct tcp_bind_hashbucket *head;
518 struct tcp_bind_bucket *tb;
521 if (inet_sk(sk)->num == 0) {
522 int err = tcp_v6_get_port(sk, inet_sk(sk)->num);
525 inet_sk(sk)->sport = htons(inet_sk(sk)->num);
528 head = &tcp_bhash[tcp_bhashfn(inet_sk(sk)->num)];
531 spin_lock_bh(&head->lock);
533 if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
535 spin_unlock_bh(&head->lock);
538 spin_unlock_bh(&head->lock);
539 return tcp_v6_check_established(sk);
543 static __inline__ int tcp_v6_iif(struct sk_buff *skb)
545 return IP6CB(skb)->iif;
548 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
551 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
552 struct inet_opt *inet = inet_sk(sk);
553 struct ipv6_pinfo *np = inet6_sk(sk);
554 struct tcp_opt *tp = tcp_sk(sk);
555 struct in6_addr *saddr = NULL, *final_p = NULL, final;
557 struct dst_entry *dst;
561 if (addr_len < SIN6_LEN_RFC2133)
564 if (usin->sin6_family != AF_INET6)
565 return(-EAFNOSUPPORT);
567 memset(&fl, 0, sizeof(fl));
570 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
571 IP6_ECN_flow_init(fl.fl6_flowlabel);
572 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
573 struct ip6_flowlabel *flowlabel;
574 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
575 if (flowlabel == NULL)
577 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
578 fl6_sock_release(flowlabel);
583 * connect() to INADDR_ANY means loopback (BSD'ism).
586 if(ipv6_addr_any(&usin->sin6_addr))
587 usin->sin6_addr.s6_addr[15] = 0x1;
589 addr_type = ipv6_addr_type(&usin->sin6_addr);
591 if(addr_type & IPV6_ADDR_MULTICAST)
594 if (addr_type&IPV6_ADDR_LINKLOCAL) {
595 if (addr_len >= sizeof(struct sockaddr_in6) &&
596 usin->sin6_scope_id) {
597 /* If interface is set while binding, indices
600 if (sk->sk_bound_dev_if &&
601 sk->sk_bound_dev_if != usin->sin6_scope_id)
604 sk->sk_bound_dev_if = usin->sin6_scope_id;
607 /* Connect to link-local address requires an interface */
608 if (!sk->sk_bound_dev_if)
612 if (tp->ts_recent_stamp &&
613 !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
615 tp->ts_recent_stamp = 0;
619 ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
620 np->flow_label = fl.fl6_flowlabel;
626 if (addr_type == IPV6_ADDR_MAPPED) {
627 u32 exthdrlen = tp->ext_header_len;
628 struct sockaddr_in sin;
630 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
632 if (__ipv6_only_sock(sk))
635 sin.sin_family = AF_INET;
636 sin.sin_port = usin->sin6_port;
637 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
639 tp->af_specific = &ipv6_mapped;
640 sk->sk_backlog_rcv = tcp_v4_do_rcv;
642 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
645 tp->ext_header_len = exthdrlen;
646 tp->af_specific = &ipv6_specific;
647 sk->sk_backlog_rcv = tcp_v6_do_rcv;
650 ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
652 ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
659 if (!ipv6_addr_any(&np->rcv_saddr))
660 saddr = &np->rcv_saddr;
662 fl.proto = IPPROTO_TCP;
663 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
664 ipv6_addr_copy(&fl.fl6_src,
665 (saddr ? saddr : &np->saddr));
666 fl.oif = sk->sk_bound_dev_if;
667 fl.fl_ip_dport = usin->sin6_port;
668 fl.fl_ip_sport = inet->sport;
670 if (np->opt && np->opt->srcrt) {
671 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
672 ipv6_addr_copy(&final, &fl.fl6_dst);
673 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
677 err = ip6_dst_lookup(sk, &dst, &fl);
681 ipv6_addr_copy(&fl.fl6_dst, final_p);
683 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
690 ipv6_addr_copy(&np->rcv_saddr, saddr);
693 /* set the source address */
694 ipv6_addr_copy(&np->saddr, saddr);
695 inet->rcv_saddr = LOOPBACK4_IPV6;
697 ip6_dst_store(sk, dst, NULL);
698 sk->sk_route_caps = dst->dev->features &
699 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
701 tp->ext_header_len = 0;
703 tp->ext_header_len = np->opt->opt_flen + np->opt->opt_nflen;
704 tp->ext2_header_len = dst->header_len;
706 tp->mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
708 inet->dport = usin->sin6_port;
710 tcp_set_state(sk, TCP_SYN_SENT);
711 err = tcp_v6_hash_connect(sk);
716 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
721 err = tcp_connect(sk);
728 tcp_set_state(sk, TCP_CLOSE);
732 sk->sk_route_caps = 0;
736 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
737 int type, int code, int offset, __u32 info)
739 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
740 struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
741 struct ipv6_pinfo *np;
747 sk = tcp_v6_lookup(&hdr->daddr, th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
750 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
754 if (sk->sk_state == TCP_TIME_WAIT) {
755 tcp_tw_put((struct tcp_tw_bucket*)sk);
760 if (sock_owned_by_user(sk))
761 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
763 if (sk->sk_state == TCP_CLOSE)
767 seq = ntohl(th->seq);
768 if (sk->sk_state != TCP_LISTEN &&
769 !between(seq, tp->snd_una, tp->snd_nxt)) {
770 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
776 if (type == ICMPV6_PKT_TOOBIG) {
777 struct dst_entry *dst = NULL;
779 if (sock_owned_by_user(sk))
781 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
784 /* icmp should have updated the destination cache entry */
785 dst = __sk_dst_check(sk, np->dst_cookie);
788 struct inet_opt *inet = inet_sk(sk);
791 /* BUGGG_FUTURE: Again, it is not clear how
792 to handle rthdr case. Ignore this complexity
795 memset(&fl, 0, sizeof(fl));
796 fl.proto = IPPROTO_TCP;
797 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
798 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
799 fl.oif = sk->sk_bound_dev_if;
800 fl.fl_ip_dport = inet->dport;
801 fl.fl_ip_sport = inet->sport;
803 if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
804 sk->sk_err_soft = -err;
808 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
809 sk->sk_err_soft = -err;
816 if (tp->pmtu_cookie > dst_pmtu(dst)) {
817 tcp_sync_mss(sk, dst_pmtu(dst));
818 tcp_simple_retransmit(sk);
819 } /* else let the usual retransmit timer handle it */
824 icmpv6_err_convert(type, code, &err);
826 /* Might be for an open_request */
827 switch (sk->sk_state) {
828 struct open_request *req, **prev;
830 if (sock_owned_by_user(sk))
833 req = tcp_v6_search_req(tp, &prev, th->dest, &hdr->daddr,
834 &hdr->saddr, tcp_v6_iif(skb));
838 /* ICMPs are not backlogged, hence we cannot get
839 * an established socket here.
841 BUG_TRAP(req->sk == NULL);
843 if (seq != req->snt_isn) {
844 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
848 tcp_synq_drop(sk, req, prev);
852 case TCP_SYN_RECV: /* Cannot happen.
853 It can, it SYNs are crossed. --ANK */
854 if (!sock_owned_by_user(sk)) {
855 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
857 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
861 sk->sk_err_soft = err;
865 if (!sock_owned_by_user(sk) && np->recverr) {
867 sk->sk_error_report(sk);
869 sk->sk_err_soft = err;
877 static int tcp_v6_send_synack(struct sock *sk, struct open_request *req,
878 struct dst_entry *dst)
880 struct ipv6_pinfo *np = inet6_sk(sk);
881 struct sk_buff * skb;
882 struct ipv6_txoptions *opt = NULL;
883 struct in6_addr * final_p = NULL, final;
887 memset(&fl, 0, sizeof(fl));
888 fl.proto = IPPROTO_TCP;
889 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
890 ipv6_addr_copy(&fl.fl6_src, &req->af.v6_req.loc_addr);
891 fl.fl6_flowlabel = 0;
892 fl.oif = req->af.v6_req.iif;
893 fl.fl_ip_dport = req->rmt_port;
894 fl.fl_ip_sport = inet_sk(sk)->sport;
899 np->rxopt.bits.srcrt == 2 &&
900 req->af.v6_req.pktopts) {
901 struct sk_buff *pktopts = req->af.v6_req.pktopts;
902 struct inet6_skb_parm *rxopt = IP6CB(pktopts);
904 opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(pktopts->nh.raw + rxopt->srcrt));
907 if (opt && opt->srcrt) {
908 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
909 ipv6_addr_copy(&final, &fl.fl6_dst);
910 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
914 err = ip6_dst_lookup(sk, &dst, &fl);
918 ipv6_addr_copy(&fl.fl6_dst, final_p);
919 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
923 skb = tcp_make_synack(sk, dst, req);
925 struct tcphdr *th = skb->h.th;
927 th->check = tcp_v6_check(th, skb->len,
928 &req->af.v6_req.loc_addr, &req->af.v6_req.rmt_addr,
929 csum_partial((char *)th, skb->len, skb->csum));
931 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
932 err = ip6_xmit(sk, skb, &fl, opt, 0);
933 if (err == NET_XMIT_CN)
939 if (opt && opt != np->opt)
940 sock_kfree_s(sk, opt, opt->tot_len);
944 static void tcp_v6_or_free(struct open_request *req)
946 if (req->af.v6_req.pktopts)
947 kfree_skb(req->af.v6_req.pktopts);
950 static struct or_calltable or_ipv6 = {
952 .rtx_syn_ack = tcp_v6_send_synack,
953 .send_ack = tcp_v6_or_send_ack,
954 .destructor = tcp_v6_or_free,
955 .send_reset = tcp_v6_send_reset
958 static int ipv6_opt_accepted(struct sock *sk, struct sk_buff *skb)
960 struct ipv6_pinfo *np = inet6_sk(sk);
961 struct inet6_skb_parm *opt = IP6CB(skb);
964 if ((opt->hop && np->rxopt.bits.hopopts) ||
965 ((IPV6_FLOWINFO_MASK&*(u32*)skb->nh.raw) &&
966 np->rxopt.bits.rxflow) ||
967 (opt->srcrt && np->rxopt.bits.srcrt) ||
968 ((opt->dst1 || opt->dst0) && np->rxopt.bits.dstopts))
975 static void tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len,
978 struct ipv6_pinfo *np = inet6_sk(sk);
980 if (skb->ip_summed == CHECKSUM_HW) {
981 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 0);
982 skb->csum = offsetof(struct tcphdr, check);
984 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
985 csum_partial((char *)th, th->doff<<2,
991 static void tcp_v6_send_reset(struct sk_buff *skb)
993 struct tcphdr *th = skb->h.th, *t1;
994 struct sk_buff *buff;
1000 if (!ipv6_unicast_destination(skb))
1004 * We need to grab some memory, and put together an RST,
1005 * and then put it into the queue to be sent.
1008 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + sizeof(struct tcphdr),
1013 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + sizeof(struct tcphdr));
1015 t1 = (struct tcphdr *) skb_push(buff,sizeof(struct tcphdr));
1017 /* Swap the send and the receive. */
1018 memset(t1, 0, sizeof(*t1));
1019 t1->dest = th->source;
1020 t1->source = th->dest;
1021 t1->doff = sizeof(*t1)/4;
1025 t1->seq = th->ack_seq;
1028 t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
1029 + skb->len - (th->doff<<2));
1032 buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1034 memset(&fl, 0, sizeof(fl));
1035 ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
1036 ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
1038 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1039 sizeof(*t1), IPPROTO_TCP,
1042 fl.proto = IPPROTO_TCP;
1043 fl.oif = tcp_v6_iif(skb);
1044 fl.fl_ip_dport = t1->dest;
1045 fl.fl_ip_sport = t1->source;
1047 /* sk = NULL, but it is safe for now. RST socket required. */
1048 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1050 if ((xfrm_lookup(&buff->dst, &fl, NULL, 0)) < 0) {
1051 dst_release(buff->dst);
1055 ip6_xmit(NULL, buff, &fl, NULL, 0);
1056 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1057 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
1064 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
1066 struct tcphdr *th = skb->h.th, *t1;
1067 struct sk_buff *buff;
1069 int tot_len = sizeof(struct tcphdr);
1074 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1079 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1081 t1 = (struct tcphdr *) skb_push(buff,tot_len);
1083 /* Swap the send and the receive. */
1084 memset(t1, 0, sizeof(*t1));
1085 t1->dest = th->source;
1086 t1->source = th->dest;
1087 t1->doff = tot_len/4;
1088 t1->seq = htonl(seq);
1089 t1->ack_seq = htonl(ack);
1091 t1->window = htons(win);
1094 u32 *ptr = (u32*)(t1 + 1);
1095 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1096 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1097 *ptr++ = htonl(tcp_time_stamp);
1101 buff->csum = csum_partial((char *)t1, tot_len, 0);
1103 memset(&fl, 0, sizeof(fl));
1104 ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
1105 ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
1107 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1108 tot_len, IPPROTO_TCP,
1111 fl.proto = IPPROTO_TCP;
1112 fl.oif = tcp_v6_iif(skb);
1113 fl.fl_ip_dport = t1->dest;
1114 fl.fl_ip_sport = t1->source;
1116 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1117 if ((xfrm_lookup(&buff->dst, &fl, NULL, 0)) < 0) {
1118 dst_release(buff->dst);
1121 ip6_xmit(NULL, buff, &fl, NULL, 0);
1122 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1129 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1131 struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk;
1133 tcp_v6_send_ack(skb, tw->tw_snd_nxt, tw->tw_rcv_nxt,
1134 tw->tw_rcv_wnd >> tw->tw_rcv_wscale, tw->tw_ts_recent);
1139 static void tcp_v6_or_send_ack(struct sk_buff *skb, struct open_request *req)
1141 tcp_v6_send_ack(skb, req->snt_isn+1, req->rcv_isn+1, req->rcv_wnd, req->ts_recent);
1145 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1147 struct open_request *req, **prev;
1148 struct tcphdr *th = skb->h.th;
1149 struct tcp_opt *tp = tcp_sk(sk);
1152 /* Find possible connection requests. */
1153 req = tcp_v6_search_req(tp, &prev, th->source, &skb->nh.ipv6h->saddr,
1154 &skb->nh.ipv6h->daddr, tcp_v6_iif(skb));
1156 return tcp_check_req(sk, skb, req, prev);
1158 nsk = __tcp_v6_lookup_established(&skb->nh.ipv6h->saddr,
1160 &skb->nh.ipv6h->daddr,
1165 if (nsk->sk_state != TCP_TIME_WAIT) {
1169 tcp_tw_put((struct tcp_tw_bucket*)nsk);
1173 #if 0 /*def CONFIG_SYN_COOKIES*/
1174 if (!th->rst && !th->syn && th->ack)
1175 sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
1180 static void tcp_v6_synq_add(struct sock *sk, struct open_request *req)
1182 struct tcp_opt *tp = tcp_sk(sk);
1183 struct tcp_listen_opt *lopt = tp->listen_opt;
1184 u32 h = tcp_v6_synq_hash(&req->af.v6_req.rmt_addr, req->rmt_port, lopt->hash_rnd);
1187 req->expires = jiffies + TCP_TIMEOUT_INIT;
1189 req->dl_next = lopt->syn_table[h];
1191 write_lock(&tp->syn_wait_lock);
1192 lopt->syn_table[h] = req;
1193 write_unlock(&tp->syn_wait_lock);
1199 /* FIXME: this is substantially similar to the ipv4 code.
1200 * Can some kind of merge be done? -- erics
1202 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1204 struct ipv6_pinfo *np = inet6_sk(sk);
1205 struct tcp_opt tmptp, *tp = tcp_sk(sk);
1206 struct open_request *req = NULL;
1207 __u32 isn = TCP_SKB_CB(skb)->when;
1209 if (skb->protocol == htons(ETH_P_IP))
1210 return tcp_v4_conn_request(sk, skb);
1212 if (!ipv6_unicast_destination(skb))
1216 * There are no SYN attacks on IPv6, yet...
1218 if (tcp_synq_is_full(sk) && !isn) {
1219 if (net_ratelimit())
1220 printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
1224 if (sk_acceptq_is_full(sk) && tcp_synq_young(sk) > 1)
1227 req = tcp_openreq_alloc();
1231 tcp_clear_options(&tmptp);
1232 tmptp.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1233 tmptp.user_mss = tp->user_mss;
1235 tcp_parse_options(skb, &tmptp, 0);
1237 tmptp.tstamp_ok = tmptp.saw_tstamp;
1238 tcp_openreq_init(req, &tmptp, skb);
1240 req->class = &or_ipv6;
1241 ipv6_addr_copy(&req->af.v6_req.rmt_addr, &skb->nh.ipv6h->saddr);
1242 ipv6_addr_copy(&req->af.v6_req.loc_addr, &skb->nh.ipv6h->daddr);
1243 TCP_ECN_create_request(req, skb->h.th);
1244 req->af.v6_req.pktopts = NULL;
1245 if (ipv6_opt_accepted(sk, skb) ||
1246 np->rxopt.bits.rxinfo ||
1247 np->rxopt.bits.rxhlim) {
1248 atomic_inc(&skb->users);
1249 req->af.v6_req.pktopts = skb;
1251 req->af.v6_req.iif = sk->sk_bound_dev_if;
1253 /* So that link locals have meaning */
1254 if (!sk->sk_bound_dev_if &&
1255 ipv6_addr_type(&req->af.v6_req.rmt_addr) & IPV6_ADDR_LINKLOCAL)
1256 req->af.v6_req.iif = tcp_v6_iif(skb);
1259 isn = tcp_v6_init_sequence(sk,skb);
1263 if (tcp_v6_send_synack(sk, req, NULL))
1266 tcp_v6_synq_add(sk, req);
1272 tcp_openreq_free(req);
1274 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
1275 return 0; /* don't send reset */
1278 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1279 struct open_request *req,
1280 struct dst_entry *dst)
1282 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1283 struct tcp6_sock *newtcp6sk;
1284 struct inet_opt *newinet;
1285 struct tcp_opt *newtp;
1287 struct ipv6_txoptions *opt;
1289 if (skb->protocol == htons(ETH_P_IP)) {
1294 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1299 newtcp6sk = (struct tcp6_sock *)newsk;
1300 newtcp6sk->pinet6 = &newtcp6sk->inet6;
1302 newinet = inet_sk(newsk);
1303 newnp = inet6_sk(newsk);
1304 newtp = tcp_sk(newsk);
1306 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1308 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1311 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1314 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1316 newtp->af_specific = &ipv6_mapped;
1317 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1318 newnp->pktoptions = NULL;
1320 newnp->mcast_oif = tcp_v6_iif(skb);
1321 newnp->mcast_hops = skb->nh.ipv6h->hop_limit;
1323 /* Charge newly allocated IPv6 socket. Though it is mapped,
1326 #ifdef INET_REFCNT_DEBUG
1327 atomic_inc(&inet6_sock_nr);
1330 /* It is tricky place. Until this moment IPv4 tcp
1331 worked with IPv6 af_tcp.af_specific.
1334 tcp_sync_mss(newsk, newtp->pmtu_cookie);
1341 if (sk_acceptq_is_full(sk))
1344 if (np->rxopt.bits.srcrt == 2 &&
1345 opt == NULL && req->af.v6_req.pktopts) {
1346 struct inet6_skb_parm *rxopt = IP6CB(req->af.v6_req.pktopts);
1348 opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(req->af.v6_req.pktopts->nh.raw+rxopt->srcrt));
1352 struct in6_addr *final_p = NULL, final;
1355 memset(&fl, 0, sizeof(fl));
1356 fl.proto = IPPROTO_TCP;
1357 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
1358 if (opt && opt->srcrt) {
1359 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1360 ipv6_addr_copy(&final, &fl.fl6_dst);
1361 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1364 ipv6_addr_copy(&fl.fl6_src, &req->af.v6_req.loc_addr);
1365 fl.oif = sk->sk_bound_dev_if;
1366 fl.fl_ip_dport = req->rmt_port;
1367 fl.fl_ip_sport = inet_sk(sk)->sport;
1369 if (ip6_dst_lookup(sk, &dst, &fl))
1373 ipv6_addr_copy(&fl.fl6_dst, final_p);
1375 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1379 newsk = tcp_create_openreq_child(sk, req, skb);
1383 /* Charge newly allocated IPv6 socket */
1384 #ifdef INET_REFCNT_DEBUG
1385 atomic_inc(&inet6_sock_nr);
1388 ip6_dst_store(newsk, dst, NULL);
1389 newsk->sk_route_caps = dst->dev->features &
1390 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1392 newtcp6sk = (struct tcp6_sock *)newsk;
1393 newtcp6sk->pinet6 = &newtcp6sk->inet6;
1395 newtp = tcp_sk(newsk);
1396 newinet = inet_sk(newsk);
1397 newnp = inet6_sk(newsk);
1399 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1401 ipv6_addr_copy(&newnp->daddr, &req->af.v6_req.rmt_addr);
1402 ipv6_addr_copy(&newnp->saddr, &req->af.v6_req.loc_addr);
1403 ipv6_addr_copy(&newnp->rcv_saddr, &req->af.v6_req.loc_addr);
1404 newsk->sk_bound_dev_if = req->af.v6_req.iif;
1406 /* Now IPv6 options...
1408 First: no IPv4 options.
1410 newinet->opt = NULL;
1413 newnp->rxopt.all = np->rxopt.all;
1415 /* Clone pktoptions received with SYN */
1416 newnp->pktoptions = NULL;
1417 if (req->af.v6_req.pktopts) {
1418 newnp->pktoptions = skb_clone(req->af.v6_req.pktopts,
1420 kfree_skb(req->af.v6_req.pktopts);
1421 req->af.v6_req.pktopts = NULL;
1422 if (newnp->pktoptions)
1423 skb_set_owner_r(newnp->pktoptions, newsk);
1426 newnp->mcast_oif = tcp_v6_iif(skb);
1427 newnp->mcast_hops = skb->nh.ipv6h->hop_limit;
1429 /* Clone native IPv6 options from listening socket (if any)
1431 Yes, keeping reference count would be much more clever,
1432 but we make one more one thing there: reattach optmem
1436 newnp->opt = ipv6_dup_options(newsk, opt);
1438 sock_kfree_s(sk, opt, opt->tot_len);
1441 newtp->ext_header_len = 0;
1443 newtp->ext_header_len = newnp->opt->opt_nflen +
1444 newnp->opt->opt_flen;
1445 newtp->ext2_header_len = dst->header_len;
1447 tcp_sync_mss(newsk, dst_pmtu(dst));
1448 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1449 tcp_initialize_rcv_mss(newsk);
1451 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1453 __tcp_v6_hash(newsk);
1454 tcp_inherit_port(sk, newsk);
1459 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1461 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1462 if (opt && opt != np->opt)
1463 sock_kfree_s(sk, opt, opt->tot_len);
1468 static int tcp_v6_checksum_init(struct sk_buff *skb)
1470 if (skb->ip_summed == CHECKSUM_HW) {
1471 skb->ip_summed = CHECKSUM_UNNECESSARY;
1472 if (!tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1473 &skb->nh.ipv6h->daddr,skb->csum))
1475 LIMIT_NETDEBUG(printk(KERN_DEBUG "hw tcp v6 csum failed\n"));
1477 if (skb->len <= 76) {
1478 if (tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1479 &skb->nh.ipv6h->daddr,skb_checksum(skb, 0, skb->len, 0)))
1481 skb->ip_summed = CHECKSUM_UNNECESSARY;
1483 skb->csum = ~tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1484 &skb->nh.ipv6h->daddr,0);
1489 /* The socket must have it's spinlock held when we get
1492 * We have a potential double-lock case here, so even when
1493 * doing backlog processing we use the BH locking scheme.
1494 * This is because we cannot sleep with the original spinlock
1497 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1499 struct ipv6_pinfo *np = inet6_sk(sk);
1501 struct sk_buff *opt_skb = NULL;
1503 /* Imagine: socket is IPv6. IPv4 packet arrives,
1504 goes to IPv4 receive handler and backlogged.
1505 From backlog it always goes here. Kerboom...
1506 Fortunately, tcp_rcv_established and rcv_established
1507 handle them correctly, but it is not case with
1508 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1511 if (skb->protocol == htons(ETH_P_IP))
1512 return tcp_v4_do_rcv(sk, skb);
1514 if (sk_filter(sk, skb, 0))
1518 * socket locking is here for SMP purposes as backlog rcv
1519 * is currently called with bh processing disabled.
1522 /* Do Stevens' IPV6_PKTOPTIONS.
1524 Yes, guys, it is the only place in our code, where we
1525 may make it not affecting IPv4.
1526 The rest of code is protocol independent,
1527 and I do not like idea to uglify IPv4.
1529 Actually, all the idea behind IPV6_PKTOPTIONS
1530 looks not very well thought. For now we latch
1531 options, received in the last packet, enqueued
1532 by tcp. Feel free to propose better solution.
1536 opt_skb = skb_clone(skb, GFP_ATOMIC);
1538 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1539 TCP_CHECK_TIMER(sk);
1540 if (tcp_rcv_established(sk, skb, skb->h.th, skb->len))
1542 TCP_CHECK_TIMER(sk);
1544 goto ipv6_pktoptions;
1548 if (skb->len < (skb->h.th->doff<<2) || tcp_checksum_complete(skb))
1551 if (sk->sk_state == TCP_LISTEN) {
1552 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1557 * Queue it on the new socket if the new socket is active,
1558 * otherwise we just shortcircuit this and continue with
1562 if (tcp_child_process(sk, nsk, skb))
1565 __kfree_skb(opt_skb);
1570 TCP_CHECK_TIMER(sk);
1571 if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len))
1573 TCP_CHECK_TIMER(sk);
1575 goto ipv6_pktoptions;
1579 tcp_v6_send_reset(skb);
1582 __kfree_skb(opt_skb);
1586 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1591 /* Do you ask, what is it?
1593 1. skb was enqueued by tcp.
1594 2. skb is added to tail of read queue, rather than out of order.
1595 3. socket is not in passive state.
1596 4. Finally, it really contains options, which user wants to receive.
1599 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1600 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1601 if (np->rxopt.bits.rxinfo)
1602 np->mcast_oif = tcp_v6_iif(opt_skb);
1603 if (np->rxopt.bits.rxhlim)
1604 np->mcast_hops = opt_skb->nh.ipv6h->hop_limit;
1605 if (ipv6_opt_accepted(sk, opt_skb)) {
1606 skb_set_owner_r(opt_skb, sk);
1607 opt_skb = xchg(&np->pktoptions, opt_skb);
1609 __kfree_skb(opt_skb);
1610 opt_skb = xchg(&np->pktoptions, NULL);
1619 static int tcp_v6_rcv(struct sk_buff **pskb, unsigned int *nhoffp)
1621 struct sk_buff *skb = *pskb;
1626 if (skb->pkt_type != PACKET_HOST)
1630 * Count it even if it's bad.
1632 TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1634 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1639 if (th->doff < sizeof(struct tcphdr)/4)
1641 if (!pskb_may_pull(skb, th->doff*4))
1644 if ((skb->ip_summed != CHECKSUM_UNNECESSARY &&
1645 tcp_v6_checksum_init(skb) < 0))
1649 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1650 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1651 skb->len - th->doff*4);
1652 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1653 TCP_SKB_CB(skb)->when = 0;
1654 TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(skb->nh.ipv6h);
1655 TCP_SKB_CB(skb)->sacked = 0;
1657 sk = __tcp_v6_lookup(&skb->nh.ipv6h->saddr, th->source,
1658 &skb->nh.ipv6h->daddr, ntohs(th->dest), tcp_v6_iif(skb));
1664 if (sk->sk_state == TCP_TIME_WAIT)
1667 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1668 goto discard_and_relse;
1670 if (sk_filter(sk, skb, 0))
1671 goto discard_and_relse;
1677 if (!sock_owned_by_user(sk)) {
1678 if (!tcp_prequeue(sk, skb))
1679 ret = tcp_v6_do_rcv(sk, skb);
1681 sk_add_backlog(sk, skb);
1685 return ret ? -1 : 0;
1688 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1691 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1693 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1695 tcp_v6_send_reset(skb);
1712 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1713 tcp_tw_put((struct tcp_tw_bucket *) sk);
1717 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1718 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1719 tcp_tw_put((struct tcp_tw_bucket *) sk);
1723 switch(tcp_timewait_state_process((struct tcp_tw_bucket *)sk,
1724 skb, th, skb->len)) {
1729 sk2 = tcp_v6_lookup_listener(&skb->nh.ipv6h->daddr, ntohs(th->dest), tcp_v6_iif(skb));
1731 tcp_tw_deschedule((struct tcp_tw_bucket *)sk);
1732 tcp_tw_put((struct tcp_tw_bucket *)sk);
1736 /* Fall through to ACK */
1739 tcp_v6_timewait_ack(sk, skb);
1743 case TCP_TW_SUCCESS:;
1748 static int tcp_v6_rebuild_header(struct sock *sk)
1751 struct dst_entry *dst;
1752 struct ipv6_pinfo *np = inet6_sk(sk);
1754 dst = __sk_dst_check(sk, np->dst_cookie);
1757 struct inet_opt *inet = inet_sk(sk);
1758 struct in6_addr *final_p = NULL, final;
1761 memset(&fl, 0, sizeof(fl));
1762 fl.proto = IPPROTO_TCP;
1763 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1764 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1765 fl.fl6_flowlabel = np->flow_label;
1766 fl.oif = sk->sk_bound_dev_if;
1767 fl.fl_ip_dport = inet->dport;
1768 fl.fl_ip_sport = inet->sport;
1770 if (np->opt && np->opt->srcrt) {
1771 struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1772 ipv6_addr_copy(&final, &fl.fl6_dst);
1773 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1777 err = ip6_dst_lookup(sk, &dst, &fl);
1779 sk->sk_route_caps = 0;
1783 ipv6_addr_copy(&fl.fl6_dst, final_p);
1785 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
1786 sk->sk_err_soft = -err;
1791 ip6_dst_store(sk, dst, NULL);
1792 sk->sk_route_caps = dst->dev->features &
1793 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1794 tcp_sk(sk)->ext2_header_len = dst->header_len;
1800 static int tcp_v6_xmit(struct sk_buff *skb, int ipfragok)
1802 struct sock *sk = skb->sk;
1803 struct inet_opt *inet = inet_sk(sk);
1804 struct ipv6_pinfo *np = inet6_sk(sk);
1806 struct dst_entry *dst;
1807 struct in6_addr *final_p = NULL, final;
1809 memset(&fl, 0, sizeof(fl));
1810 fl.proto = IPPROTO_TCP;
1811 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1812 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1813 fl.fl6_flowlabel = np->flow_label;
1814 IP6_ECN_flow_xmit(sk, fl.fl6_flowlabel);
1815 fl.oif = sk->sk_bound_dev_if;
1816 fl.fl_ip_sport = inet->sport;
1817 fl.fl_ip_dport = inet->dport;
1819 if (np->opt && np->opt->srcrt) {
1820 struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1821 ipv6_addr_copy(&final, &fl.fl6_dst);
1822 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1826 dst = __sk_dst_check(sk, np->dst_cookie);
1829 int err = ip6_dst_lookup(sk, &dst, &fl);
1832 sk->sk_err_soft = -err;
1837 ipv6_addr_copy(&fl.fl6_dst, final_p);
1839 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
1840 sk->sk_route_caps = 0;
1845 ip6_dst_store(sk, dst, NULL);
1846 sk->sk_route_caps = dst->dev->features &
1847 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1848 tcp_sk(sk)->ext2_header_len = dst->header_len;
1851 skb->dst = dst_clone(dst);
1853 /* Restore final destination back after routing done */
1854 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1856 return ip6_xmit(sk, skb, &fl, np->opt, 0);
1859 static void v6_addr2sockaddr(struct sock *sk, struct sockaddr * uaddr)
1861 struct ipv6_pinfo *np = inet6_sk(sk);
1862 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) uaddr;
1864 sin6->sin6_family = AF_INET6;
1865 ipv6_addr_copy(&sin6->sin6_addr, &np->daddr);
1866 sin6->sin6_port = inet_sk(sk)->dport;
1867 /* We do not store received flowlabel for TCP */
1868 sin6->sin6_flowinfo = 0;
1869 sin6->sin6_scope_id = 0;
1870 if (sk->sk_bound_dev_if &&
1871 ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
1872 sin6->sin6_scope_id = sk->sk_bound_dev_if;
1875 static int tcp_v6_remember_stamp(struct sock *sk)
1877 /* Alas, not yet... */
1881 static struct tcp_func ipv6_specific = {
1882 .queue_xmit = tcp_v6_xmit,
1883 .send_check = tcp_v6_send_check,
1884 .rebuild_header = tcp_v6_rebuild_header,
1885 .conn_request = tcp_v6_conn_request,
1886 .syn_recv_sock = tcp_v6_syn_recv_sock,
1887 .remember_stamp = tcp_v6_remember_stamp,
1888 .net_header_len = sizeof(struct ipv6hdr),
1890 .setsockopt = ipv6_setsockopt,
1891 .getsockopt = ipv6_getsockopt,
1892 .addr2sockaddr = v6_addr2sockaddr,
1893 .sockaddr_len = sizeof(struct sockaddr_in6)
1897 * TCP over IPv4 via INET6 API
1900 static struct tcp_func ipv6_mapped = {
1901 .queue_xmit = ip_queue_xmit,
1902 .send_check = tcp_v4_send_check,
1903 .rebuild_header = tcp_v4_rebuild_header,
1904 .conn_request = tcp_v6_conn_request,
1905 .syn_recv_sock = tcp_v6_syn_recv_sock,
1906 .remember_stamp = tcp_v4_remember_stamp,
1907 .net_header_len = sizeof(struct iphdr),
1909 .setsockopt = ipv6_setsockopt,
1910 .getsockopt = ipv6_getsockopt,
1911 .addr2sockaddr = v6_addr2sockaddr,
1912 .sockaddr_len = sizeof(struct sockaddr_in6)
1917 /* NOTE: A lot of things set to zero explicitly by call to
1918 * sk_alloc() so need not be done here.
1920 static int tcp_v6_init_sock(struct sock *sk)
1922 struct tcp_opt *tp = tcp_sk(sk);
1924 skb_queue_head_init(&tp->out_of_order_queue);
1925 tcp_init_xmit_timers(sk);
1926 tcp_prequeue_init(tp);
1928 tp->rto = TCP_TIMEOUT_INIT;
1929 tp->mdev = TCP_TIMEOUT_INIT;
1931 /* So many TCP implementations out there (incorrectly) count the
1932 * initial SYN frame in their delayed-ACK and congestion control
1933 * algorithms that we must have the following bandaid to talk
1934 * efficiently to them. -DaveM
1938 /* See draft-stevens-tcpca-spec-01 for discussion of the
1939 * initialization of these values.
1941 tp->snd_ssthresh = 0x7fffffff;
1942 tp->snd_cwnd_clamp = ~0;
1943 tp->mss_cache_std = tp->mss_cache = 536;
1945 tp->reordering = sysctl_tcp_reordering;
1947 sk->sk_state = TCP_CLOSE;
1949 tp->af_specific = &ipv6_specific;
1951 sk->sk_write_space = sk_stream_write_space;
1952 sk->sk_use_write_queue = 1;
1954 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1955 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1957 atomic_inc(&tcp_sockets_allocated);
1962 static int tcp_v6_destroy_sock(struct sock *sk)
1964 extern int tcp_v4_destroy_sock(struct sock *sk);
1966 tcp_v4_destroy_sock(sk);
1967 return inet6_destroy_sock(sk);
1970 /* Proc filesystem TCPv6 sock list dumping. */
1971 static void get_openreq6(struct seq_file *seq,
1972 struct sock *sk, struct open_request *req, int i, int uid)
1974 struct in6_addr *dest, *src;
1975 int ttd = req->expires - jiffies;
1980 src = &req->af.v6_req.loc_addr;
1981 dest = &req->af.v6_req.rmt_addr;
1983 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1984 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1986 src->s6_addr32[0], src->s6_addr32[1],
1987 src->s6_addr32[2], src->s6_addr32[3],
1988 ntohs(inet_sk(sk)->sport),
1989 dest->s6_addr32[0], dest->s6_addr32[1],
1990 dest->s6_addr32[2], dest->s6_addr32[3],
1991 ntohs(req->rmt_port),
1993 0,0, /* could print option size, but that is af dependent. */
1994 1, /* timers active (only the expire timer) */
1995 jiffies_to_clock_t(ttd),
1998 0, /* non standard timer */
1999 0, /* open_requests have no inode */
2003 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2005 struct in6_addr *dest, *src;
2008 unsigned long timer_expires;
2009 struct inet_opt *inet = inet_sk(sp);
2010 struct tcp_opt *tp = tcp_sk(sp);
2011 struct ipv6_pinfo *np = inet6_sk(sp);
2014 src = &np->rcv_saddr;
2015 destp = ntohs(inet->dport);
2016 srcp = ntohs(inet->sport);
2017 if (tp->pending == TCP_TIME_RETRANS) {
2019 timer_expires = tp->timeout;
2020 } else if (tp->pending == TCP_TIME_PROBE0) {
2022 timer_expires = tp->timeout;
2023 } else if (timer_pending(&sp->sk_timer)) {
2025 timer_expires = sp->sk_timer.expires;
2028 timer_expires = jiffies;
2032 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2033 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
2035 src->s6_addr32[0], src->s6_addr32[1],
2036 src->s6_addr32[2], src->s6_addr32[3], srcp,
2037 dest->s6_addr32[0], dest->s6_addr32[1],
2038 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2040 tp->write_seq-tp->snd_una, tp->rcv_nxt-tp->copied_seq,
2042 jiffies_to_clock_t(timer_expires - jiffies),
2047 atomic_read(&sp->sk_refcnt), sp,
2048 tp->rto, tp->ack.ato, (tp->ack.quick<<1)|tp->ack.pingpong,
2049 tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2053 static void get_timewait6_sock(struct seq_file *seq,
2054 struct tcp_tw_bucket *tw, int i)
2056 struct in6_addr *dest, *src;
2058 int ttd = tw->tw_ttd - jiffies;
2063 dest = &tw->tw_v6_daddr;
2064 src = &tw->tw_v6_rcv_saddr;
2065 destp = ntohs(tw->tw_dport);
2066 srcp = ntohs(tw->tw_sport);
2069 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2070 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2072 src->s6_addr32[0], src->s6_addr32[1],
2073 src->s6_addr32[2], src->s6_addr32[3], srcp,
2074 dest->s6_addr32[0], dest->s6_addr32[1],
2075 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2076 tw->tw_substate, 0, 0,
2077 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2078 atomic_read(&tw->tw_refcnt), tw);
2081 #ifdef CONFIG_PROC_FS
2082 static int tcp6_seq_show(struct seq_file *seq, void *v)
2084 struct tcp_iter_state *st;
2086 if (v == SEQ_START_TOKEN) {
2091 "st tx_queue rx_queue tr tm->when retrnsmt"
2092 " uid timeout inode\n");
2097 switch (st->state) {
2098 case TCP_SEQ_STATE_LISTENING:
2099 case TCP_SEQ_STATE_ESTABLISHED:
2100 get_tcp6_sock(seq, v, st->num);
2102 case TCP_SEQ_STATE_OPENREQ:
2103 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2105 case TCP_SEQ_STATE_TIME_WAIT:
2106 get_timewait6_sock(seq, v, st->num);
2113 static struct file_operations tcp6_seq_fops;
2114 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2115 .owner = THIS_MODULE,
2118 .seq_show = tcp6_seq_show,
2119 .seq_fops = &tcp6_seq_fops,
2122 int __init tcp6_proc_init(void)
2124 return tcp_proc_register(&tcp6_seq_afinfo);
2127 void tcp6_proc_exit(void)
2129 tcp_proc_unregister(&tcp6_seq_afinfo);
2133 struct proto tcpv6_prot = {
2135 .owner = THIS_MODULE,
2137 .connect = tcp_v6_connect,
2138 .disconnect = tcp_disconnect,
2139 .accept = tcp_accept,
2141 .init = tcp_v6_init_sock,
2142 .destroy = tcp_v6_destroy_sock,
2143 .shutdown = tcp_shutdown,
2144 .setsockopt = tcp_setsockopt,
2145 .getsockopt = tcp_getsockopt,
2146 .sendmsg = tcp_sendmsg,
2147 .recvmsg = tcp_recvmsg,
2148 .backlog_rcv = tcp_v6_do_rcv,
2149 .hash = tcp_v6_hash,
2150 .unhash = tcp_unhash,
2151 .get_port = tcp_v6_get_port,
2152 .enter_memory_pressure = tcp_enter_memory_pressure,
2153 .sockets_allocated = &tcp_sockets_allocated,
2154 .memory_allocated = &tcp_memory_allocated,
2155 .memory_pressure = &tcp_memory_pressure,
2156 .sysctl_mem = sysctl_tcp_mem,
2157 .sysctl_wmem = sysctl_tcp_wmem,
2158 .sysctl_rmem = sysctl_tcp_rmem,
2159 .max_header = MAX_TCP_HEADER,
2160 .slab_obj_size = sizeof(struct tcp6_sock),
2163 static struct inet6_protocol tcpv6_protocol = {
2164 .handler = tcp_v6_rcv,
2165 .err_handler = tcp_v6_err,
2166 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2169 extern struct proto_ops inet6_stream_ops;
2171 static struct inet_protosw tcpv6_protosw = {
2172 .type = SOCK_STREAM,
2173 .protocol = IPPROTO_TCP,
2174 .prot = &tcpv6_prot,
2175 .ops = &inet6_stream_ops,
2178 .flags = INET_PROTOSW_PERMANENT,
2181 void __init tcpv6_init(void)
2183 /* register inet6 protocol */
2184 if (inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP) < 0)
2185 printk(KERN_ERR "tcpv6_init: Could not register protocol\n");
2186 inet6_register_protosw(&tcpv6_protosw);