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/errno.h>
30 #include <linux/types.h>
31 #include <linux/socket.h>
32 #include <linux/sockios.h>
33 #include <linux/net.h>
34 #include <linux/jiffies.h>
36 #include <linux/in6.h>
37 #include <linux/netdevice.h>
38 #include <linux/init.h>
39 #include <linux/jhash.h>
40 #include <linux/ipsec.h>
41 #include <linux/times.h>
43 #include <linux/ipv6.h>
44 #include <linux/icmpv6.h>
45 #include <linux/random.h>
48 #include <net/ndisc.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/inet6_connection_sock.h>
52 #include <net/transp_v6.h>
53 #include <net/addrconf.h>
54 #include <net/ip6_route.h>
55 #include <net/ip6_checksum.h>
56 #include <net/inet_ecn.h>
57 #include <net/protocol.h>
60 #include <net/dsfield.h>
61 #include <net/timewait_sock.h>
63 #include <asm/uaccess.h>
65 #include <linux/proc_fs.h>
66 #include <linux/seq_file.h>
68 #include <linux/crypto.h>
69 #include <linux/scatterlist.h>
71 /* Socket used for sending RSTs and ACKs */
72 static struct socket *tcp6_socket;
74 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
75 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req);
76 static void tcp_v6_send_check(struct sock *sk, int len,
79 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
81 static struct inet_connection_sock_af_ops ipv6_mapped;
82 static struct inet_connection_sock_af_ops ipv6_specific;
83 #ifdef CONFIG_TCP_MD5SIG
84 static struct tcp_sock_af_ops tcp_sock_ipv6_specific;
85 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
88 static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
90 return inet_csk_get_port(&tcp_hashinfo, sk, snum,
91 inet6_csk_bind_conflict);
94 static void tcp_v6_hash(struct sock *sk)
96 if (sk->sk_state != TCP_CLOSE) {
97 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
102 __inet6_hash(&tcp_hashinfo, sk);
107 static __inline__ __sum16 tcp_v6_check(struct tcphdr *th, int len,
108 struct in6_addr *saddr,
109 struct in6_addr *daddr,
112 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
115 static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
117 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
118 ipv6_hdr(skb)->saddr.s6_addr32,
120 tcp_hdr(skb)->source);
123 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
126 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
127 struct inet_sock *inet = inet_sk(sk);
128 struct inet_connection_sock *icsk = inet_csk(sk);
129 struct ipv6_pinfo *np = inet6_sk(sk);
130 struct tcp_sock *tp = tcp_sk(sk);
131 struct in6_addr *saddr = NULL, *final_p = NULL, final;
133 struct dst_entry *dst;
137 if (addr_len < SIN6_LEN_RFC2133)
140 if (usin->sin6_family != AF_INET6)
141 return(-EAFNOSUPPORT);
143 memset(&fl, 0, sizeof(fl));
146 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
147 IP6_ECN_flow_init(fl.fl6_flowlabel);
148 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
149 struct ip6_flowlabel *flowlabel;
150 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
151 if (flowlabel == NULL)
153 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
154 fl6_sock_release(flowlabel);
159 * connect() to INADDR_ANY means loopback (BSD'ism).
162 if(ipv6_addr_any(&usin->sin6_addr))
163 usin->sin6_addr.s6_addr[15] = 0x1;
165 addr_type = ipv6_addr_type(&usin->sin6_addr);
167 if(addr_type & IPV6_ADDR_MULTICAST)
170 if (addr_type&IPV6_ADDR_LINKLOCAL) {
171 if (addr_len >= sizeof(struct sockaddr_in6) &&
172 usin->sin6_scope_id) {
173 /* If interface is set while binding, indices
176 if (sk->sk_bound_dev_if &&
177 sk->sk_bound_dev_if != usin->sin6_scope_id)
180 sk->sk_bound_dev_if = usin->sin6_scope_id;
183 /* Connect to link-local address requires an interface */
184 if (!sk->sk_bound_dev_if)
188 if (tp->rx_opt.ts_recent_stamp &&
189 !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
190 tp->rx_opt.ts_recent = 0;
191 tp->rx_opt.ts_recent_stamp = 0;
195 ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
196 np->flow_label = fl.fl6_flowlabel;
202 if (addr_type == IPV6_ADDR_MAPPED) {
203 u32 exthdrlen = icsk->icsk_ext_hdr_len;
204 struct sockaddr_in sin;
206 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
208 if (__ipv6_only_sock(sk))
211 sin.sin_family = AF_INET;
212 sin.sin_port = usin->sin6_port;
213 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
215 icsk->icsk_af_ops = &ipv6_mapped;
216 sk->sk_backlog_rcv = tcp_v4_do_rcv;
217 #ifdef CONFIG_TCP_MD5SIG
218 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
221 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
224 icsk->icsk_ext_hdr_len = exthdrlen;
225 icsk->icsk_af_ops = &ipv6_specific;
226 sk->sk_backlog_rcv = tcp_v6_do_rcv;
227 #ifdef CONFIG_TCP_MD5SIG
228 tp->af_specific = &tcp_sock_ipv6_specific;
232 ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
234 ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
241 if (!ipv6_addr_any(&np->rcv_saddr))
242 saddr = &np->rcv_saddr;
244 fl.proto = IPPROTO_TCP;
245 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
246 ipv6_addr_copy(&fl.fl6_src,
247 (saddr ? saddr : &np->saddr));
248 fl.oif = sk->sk_bound_dev_if;
249 fl.fl_ip_dport = usin->sin6_port;
250 fl.fl_ip_sport = inet->sport;
252 if (np->opt && np->opt->srcrt) {
253 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
254 ipv6_addr_copy(&final, &fl.fl6_dst);
255 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
259 security_sk_classify_flow(sk, &fl);
261 err = ip6_dst_lookup(sk, &dst, &fl);
265 ipv6_addr_copy(&fl.fl6_dst, final_p);
267 if ((err = __xfrm_lookup(&dst, &fl, sk, 1)) < 0) {
269 err = ip6_dst_blackhole(sk, &dst, &fl);
276 ipv6_addr_copy(&np->rcv_saddr, saddr);
279 /* set the source address */
280 ipv6_addr_copy(&np->saddr, saddr);
281 inet->rcv_saddr = LOOPBACK4_IPV6;
283 sk->sk_gso_type = SKB_GSO_TCPV6;
284 __ip6_dst_store(sk, dst, NULL, NULL);
286 icsk->icsk_ext_hdr_len = 0;
288 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
291 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
293 inet->dport = usin->sin6_port;
295 tcp_set_state(sk, TCP_SYN_SENT);
296 err = inet6_hash_connect(&tcp_death_row, sk);
301 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
306 err = tcp_connect(sk);
313 tcp_set_state(sk, TCP_CLOSE);
317 sk->sk_route_caps = 0;
321 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
322 int type, int code, int offset, __be32 info)
324 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
325 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
326 struct ipv6_pinfo *np;
332 sk = inet6_lookup(&tcp_hashinfo, &hdr->daddr, th->dest, &hdr->saddr,
333 th->source, skb->dev->ifindex);
336 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
340 if (sk->sk_state == TCP_TIME_WAIT) {
341 inet_twsk_put(inet_twsk(sk));
346 if (sock_owned_by_user(sk))
347 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
349 if (sk->sk_state == TCP_CLOSE)
353 seq = ntohl(th->seq);
354 if (sk->sk_state != TCP_LISTEN &&
355 !between(seq, tp->snd_una, tp->snd_nxt)) {
356 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
362 if (type == ICMPV6_PKT_TOOBIG) {
363 struct dst_entry *dst = NULL;
365 if (sock_owned_by_user(sk))
367 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
370 /* icmp should have updated the destination cache entry */
371 dst = __sk_dst_check(sk, np->dst_cookie);
374 struct inet_sock *inet = inet_sk(sk);
377 /* BUGGG_FUTURE: Again, it is not clear how
378 to handle rthdr case. Ignore this complexity
381 memset(&fl, 0, sizeof(fl));
382 fl.proto = IPPROTO_TCP;
383 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
384 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
385 fl.oif = sk->sk_bound_dev_if;
386 fl.fl_ip_dport = inet->dport;
387 fl.fl_ip_sport = inet->sport;
388 security_skb_classify_flow(skb, &fl);
390 if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
391 sk->sk_err_soft = -err;
395 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
396 sk->sk_err_soft = -err;
403 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
404 tcp_sync_mss(sk, dst_mtu(dst));
405 tcp_simple_retransmit(sk);
406 } /* else let the usual retransmit timer handle it */
411 icmpv6_err_convert(type, code, &err);
413 /* Might be for an request_sock */
414 switch (sk->sk_state) {
415 struct request_sock *req, **prev;
417 if (sock_owned_by_user(sk))
420 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
421 &hdr->saddr, inet6_iif(skb));
425 /* ICMPs are not backlogged, hence we cannot get
426 * an established socket here.
428 BUG_TRAP(req->sk == NULL);
430 if (seq != tcp_rsk(req)->snt_isn) {
431 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
435 inet_csk_reqsk_queue_drop(sk, req, prev);
439 case TCP_SYN_RECV: /* Cannot happen.
440 It can, it SYNs are crossed. --ANK */
441 if (!sock_owned_by_user(sk)) {
443 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
447 sk->sk_err_soft = err;
451 if (!sock_owned_by_user(sk) && np->recverr) {
453 sk->sk_error_report(sk);
455 sk->sk_err_soft = err;
463 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
464 struct dst_entry *dst)
466 struct inet6_request_sock *treq = inet6_rsk(req);
467 struct ipv6_pinfo *np = inet6_sk(sk);
468 struct sk_buff * skb;
469 struct ipv6_txoptions *opt = NULL;
470 struct in6_addr * final_p = NULL, final;
474 memset(&fl, 0, sizeof(fl));
475 fl.proto = IPPROTO_TCP;
476 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
477 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
478 fl.fl6_flowlabel = 0;
480 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
481 fl.fl_ip_sport = inet_sk(sk)->sport;
482 security_req_classify_flow(req, &fl);
486 if (opt && opt->srcrt) {
487 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
488 ipv6_addr_copy(&final, &fl.fl6_dst);
489 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
493 err = ip6_dst_lookup(sk, &dst, &fl);
497 ipv6_addr_copy(&fl.fl6_dst, final_p);
498 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
502 skb = tcp_make_synack(sk, dst, req);
504 struct tcphdr *th = tcp_hdr(skb);
506 th->check = tcp_v6_check(th, skb->len,
507 &treq->loc_addr, &treq->rmt_addr,
508 csum_partial((char *)th, skb->len, skb->csum));
510 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
511 err = ip6_xmit(sk, skb, &fl, opt, 0);
512 err = net_xmit_eval(err);
516 if (opt && opt != np->opt)
517 sock_kfree_s(sk, opt, opt->tot_len);
522 static void tcp_v6_reqsk_destructor(struct request_sock *req)
524 if (inet6_rsk(req)->pktopts)
525 kfree_skb(inet6_rsk(req)->pktopts);
528 #ifdef CONFIG_TCP_MD5SIG
529 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
530 struct in6_addr *addr)
532 struct tcp_sock *tp = tcp_sk(sk);
537 if (!tp->md5sig_info || !tp->md5sig_info->entries6)
540 for (i = 0; i < tp->md5sig_info->entries6; i++) {
541 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, addr) == 0)
542 return &tp->md5sig_info->keys6[i].base;
547 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
548 struct sock *addr_sk)
550 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
553 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
554 struct request_sock *req)
556 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
559 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
560 char *newkey, u8 newkeylen)
562 /* Add key to the list */
563 struct tcp6_md5sig_key *key;
564 struct tcp_sock *tp = tcp_sk(sk);
565 struct tcp6_md5sig_key *keys;
567 key = (struct tcp6_md5sig_key*) tcp_v6_md5_do_lookup(sk, peer);
569 /* modify existing entry - just update that one */
570 kfree(key->base.key);
571 key->base.key = newkey;
572 key->base.keylen = newkeylen;
574 /* reallocate new list if current one is full. */
575 if (!tp->md5sig_info) {
576 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
577 if (!tp->md5sig_info) {
581 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
583 tcp_alloc_md5sig_pool();
584 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
585 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
586 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
589 tcp_free_md5sig_pool();
594 if (tp->md5sig_info->entries6)
595 memmove(keys, tp->md5sig_info->keys6,
596 (sizeof (tp->md5sig_info->keys6[0]) *
597 tp->md5sig_info->entries6));
599 kfree(tp->md5sig_info->keys6);
600 tp->md5sig_info->keys6 = keys;
601 tp->md5sig_info->alloced6++;
604 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
606 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
607 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
609 tp->md5sig_info->entries6++;
614 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
615 u8 *newkey, __u8 newkeylen)
617 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
621 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
623 struct tcp_sock *tp = tcp_sk(sk);
626 for (i = 0; i < tp->md5sig_info->entries6; i++) {
627 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, peer) == 0) {
629 kfree(tp->md5sig_info->keys6[i].base.key);
630 tp->md5sig_info->entries6--;
632 if (tp->md5sig_info->entries6 == 0) {
633 kfree(tp->md5sig_info->keys6);
634 tp->md5sig_info->keys6 = NULL;
635 tp->md5sig_info->alloced6 = 0;
637 tcp_free_md5sig_pool();
641 /* shrink the database */
642 if (tp->md5sig_info->entries6 != i)
643 memmove(&tp->md5sig_info->keys6[i],
644 &tp->md5sig_info->keys6[i+1],
645 (tp->md5sig_info->entries6 - i)
646 * sizeof (tp->md5sig_info->keys6[0]));
653 static void tcp_v6_clear_md5_list (struct sock *sk)
655 struct tcp_sock *tp = tcp_sk(sk);
658 if (tp->md5sig_info->entries6) {
659 for (i = 0; i < tp->md5sig_info->entries6; i++)
660 kfree(tp->md5sig_info->keys6[i].base.key);
661 tp->md5sig_info->entries6 = 0;
662 tcp_free_md5sig_pool();
665 kfree(tp->md5sig_info->keys6);
666 tp->md5sig_info->keys6 = NULL;
667 tp->md5sig_info->alloced6 = 0;
669 if (tp->md5sig_info->entries4) {
670 for (i = 0; i < tp->md5sig_info->entries4; i++)
671 kfree(tp->md5sig_info->keys4[i].base.key);
672 tp->md5sig_info->entries4 = 0;
673 tcp_free_md5sig_pool();
676 kfree(tp->md5sig_info->keys4);
677 tp->md5sig_info->keys4 = NULL;
678 tp->md5sig_info->alloced4 = 0;
681 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
684 struct tcp_md5sig cmd;
685 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
688 if (optlen < sizeof(cmd))
691 if (copy_from_user(&cmd, optval, sizeof(cmd)))
694 if (sin6->sin6_family != AF_INET6)
697 if (!cmd.tcpm_keylen) {
698 if (!tcp_sk(sk)->md5sig_info)
700 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_MAPPED)
701 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
702 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
705 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
708 if (!tcp_sk(sk)->md5sig_info) {
709 struct tcp_sock *tp = tcp_sk(sk);
710 struct tcp_md5sig_info *p;
712 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
717 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
720 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
723 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_MAPPED) {
724 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
725 newkey, cmd.tcpm_keylen);
727 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
730 static int tcp_v6_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
731 struct in6_addr *saddr,
732 struct in6_addr *daddr,
733 struct tcphdr *th, int protocol,
736 struct scatterlist sg[4];
740 struct tcp_md5sig_pool *hp;
741 struct tcp6_pseudohdr *bp;
742 struct hash_desc *desc;
744 unsigned int nbytes = 0;
746 hp = tcp_get_md5sig_pool();
748 printk(KERN_WARNING "%s(): hash pool not found...\n", __FUNCTION__);
749 goto clear_hash_noput;
751 bp = &hp->md5_blk.ip6;
752 desc = &hp->md5_desc;
754 /* 1. TCP pseudo-header (RFC2460) */
755 ipv6_addr_copy(&bp->saddr, saddr);
756 ipv6_addr_copy(&bp->daddr, daddr);
757 bp->len = htonl(tcplen);
758 bp->protocol = htonl(protocol);
760 sg_set_buf(&sg[block++], bp, sizeof(*bp));
761 nbytes += sizeof(*bp);
763 /* 2. TCP header, excluding options */
766 sg_set_buf(&sg[block++], th, sizeof(*th));
767 nbytes += sizeof(*th);
769 /* 3. TCP segment data (if any) */
770 data_len = tcplen - (th->doff << 2);
772 u8 *data = (u8 *)th + (th->doff << 2);
773 sg_set_buf(&sg[block++], data, data_len);
778 sg_set_buf(&sg[block++], key->key, key->keylen);
779 nbytes += key->keylen;
781 /* Now store the hash into the packet */
782 err = crypto_hash_init(desc);
784 printk(KERN_WARNING "%s(): hash_init failed\n", __FUNCTION__);
787 err = crypto_hash_update(desc, sg, nbytes);
789 printk(KERN_WARNING "%s(): hash_update failed\n", __FUNCTION__);
792 err = crypto_hash_final(desc, md5_hash);
794 printk(KERN_WARNING "%s(): hash_final failed\n", __FUNCTION__);
798 /* Reset header, and free up the crypto */
799 tcp_put_md5sig_pool();
804 tcp_put_md5sig_pool();
806 memset(md5_hash, 0, 16);
810 static int tcp_v6_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
812 struct dst_entry *dst,
813 struct request_sock *req,
814 struct tcphdr *th, int protocol,
817 struct in6_addr *saddr, *daddr;
820 saddr = &inet6_sk(sk)->saddr;
821 daddr = &inet6_sk(sk)->daddr;
823 saddr = &inet6_rsk(req)->loc_addr;
824 daddr = &inet6_rsk(req)->rmt_addr;
826 return tcp_v6_do_calc_md5_hash(md5_hash, key,
828 th, protocol, tcplen);
831 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
833 __u8 *hash_location = NULL;
834 struct tcp_md5sig_key *hash_expected;
835 struct ipv6hdr *ip6h = ipv6_hdr(skb);
836 struct tcphdr *th = tcp_hdr(skb);
837 int length = (th->doff << 2) - sizeof (*th);
842 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
844 /* If the TCP option is too short, we can short cut */
845 if (length < TCPOLEN_MD5SIG)
846 return hash_expected ? 1 : 0;
862 if (opsize < 2 || opsize > length)
864 if (opcode == TCPOPT_MD5SIG) {
874 /* do we have a hash as expected? */
875 if (!hash_expected) {
878 if (net_ratelimit()) {
879 printk(KERN_INFO "MD5 Hash NOT expected but found "
880 "(" NIP6_FMT ", %u)->"
881 "(" NIP6_FMT ", %u)\n",
882 NIP6(ip6h->saddr), ntohs(th->source),
883 NIP6(ip6h->daddr), ntohs(th->dest));
888 if (!hash_location) {
889 if (net_ratelimit()) {
890 printk(KERN_INFO "MD5 Hash expected but NOT found "
891 "(" NIP6_FMT ", %u)->"
892 "(" NIP6_FMT ", %u)\n",
893 NIP6(ip6h->saddr), ntohs(th->source),
894 NIP6(ip6h->daddr), ntohs(th->dest));
899 /* check the signature */
900 genhash = tcp_v6_do_calc_md5_hash(newhash,
902 &ip6h->saddr, &ip6h->daddr,
905 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
906 if (net_ratelimit()) {
907 printk(KERN_INFO "MD5 Hash %s for "
908 "(" NIP6_FMT ", %u)->"
909 "(" NIP6_FMT ", %u)\n",
910 genhash ? "failed" : "mismatch",
911 NIP6(ip6h->saddr), ntohs(th->source),
912 NIP6(ip6h->daddr), ntohs(th->dest));
920 static struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
922 .obj_size = sizeof(struct tcp6_request_sock),
923 .rtx_syn_ack = tcp_v6_send_synack,
924 .send_ack = tcp_v6_reqsk_send_ack,
925 .destructor = tcp_v6_reqsk_destructor,
926 .send_reset = tcp_v6_send_reset
929 #ifdef CONFIG_TCP_MD5SIG
930 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
931 .md5_lookup = tcp_v6_reqsk_md5_lookup,
935 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
936 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
937 .twsk_unique = tcp_twsk_unique,
938 .twsk_destructor= tcp_twsk_destructor,
941 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
943 struct ipv6_pinfo *np = inet6_sk(sk);
944 struct tcphdr *th = tcp_hdr(skb);
946 if (skb->ip_summed == CHECKSUM_PARTIAL) {
947 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 0);
948 skb->csum_start = skb_transport_header(skb) - skb->head;
949 skb->csum_offset = offsetof(struct tcphdr, check);
951 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
952 csum_partial((char *)th, th->doff<<2,
957 static int tcp_v6_gso_send_check(struct sk_buff *skb)
959 struct ipv6hdr *ipv6h;
962 if (!pskb_may_pull(skb, sizeof(*th)))
965 ipv6h = ipv6_hdr(skb);
969 th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
971 skb->csum_start = skb_transport_header(skb) - skb->head;
972 skb->csum_offset = offsetof(struct tcphdr, check);
973 skb->ip_summed = CHECKSUM_PARTIAL;
977 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
979 struct tcphdr *th = tcp_hdr(skb), *t1;
980 struct sk_buff *buff;
982 int tot_len = sizeof(*th);
983 #ifdef CONFIG_TCP_MD5SIG
984 struct tcp_md5sig_key *key;
990 if (!ipv6_unicast_destination(skb))
993 #ifdef CONFIG_TCP_MD5SIG
995 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1000 tot_len += TCPOLEN_MD5SIG_ALIGNED;
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) + tot_len,
1013 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1015 t1 = (struct tcphdr *) skb_push(buff, tot_len);
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 = tot_len / 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 #ifdef CONFIG_TCP_MD5SIG
1034 __be32 *opt = (__be32*)(t1 + 1);
1035 opt[0] = htonl((TCPOPT_NOP << 24) |
1036 (TCPOPT_NOP << 16) |
1037 (TCPOPT_MD5SIG << 8) |
1039 tcp_v6_do_calc_md5_hash((__u8 *)&opt[1], key,
1040 &ipv6_hdr(skb)->daddr,
1041 &ipv6_hdr(skb)->saddr,
1042 t1, IPPROTO_TCP, tot_len);
1046 buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1048 memset(&fl, 0, sizeof(fl));
1049 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1050 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1052 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1053 sizeof(*t1), IPPROTO_TCP,
1056 fl.proto = IPPROTO_TCP;
1057 fl.oif = inet6_iif(skb);
1058 fl.fl_ip_dport = t1->dest;
1059 fl.fl_ip_sport = t1->source;
1060 security_skb_classify_flow(skb, &fl);
1062 /* sk = NULL, but it is safe for now. RST socket required. */
1063 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1065 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1066 ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1067 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1068 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
1076 static void tcp_v6_send_ack(struct tcp_timewait_sock *tw,
1077 struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
1079 struct tcphdr *th = tcp_hdr(skb), *t1;
1080 struct sk_buff *buff;
1082 int tot_len = sizeof(struct tcphdr);
1084 #ifdef CONFIG_TCP_MD5SIG
1085 struct tcp_md5sig_key *key;
1086 struct tcp_md5sig_key tw_key;
1089 #ifdef CONFIG_TCP_MD5SIG
1090 if (!tw && skb->sk) {
1091 key = tcp_v6_md5_do_lookup(skb->sk, &ipv6_hdr(skb)->daddr);
1092 } else if (tw && tw->tw_md5_keylen) {
1093 tw_key.key = tw->tw_md5_key;
1094 tw_key.keylen = tw->tw_md5_keylen;
1102 tot_len += TCPOLEN_TSTAMP_ALIGNED;
1103 #ifdef CONFIG_TCP_MD5SIG
1105 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1108 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1113 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1115 t1 = (struct tcphdr *) skb_push(buff,tot_len);
1117 /* Swap the send and the receive. */
1118 memset(t1, 0, sizeof(*t1));
1119 t1->dest = th->source;
1120 t1->source = th->dest;
1121 t1->doff = tot_len/4;
1122 t1->seq = htonl(seq);
1123 t1->ack_seq = htonl(ack);
1125 t1->window = htons(win);
1127 topt = (__be32 *)(t1 + 1);
1130 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1131 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1132 *topt++ = htonl(tcp_time_stamp);
1136 #ifdef CONFIG_TCP_MD5SIG
1138 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1139 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1140 tcp_v6_do_calc_md5_hash((__u8 *)topt, key,
1141 &ipv6_hdr(skb)->daddr,
1142 &ipv6_hdr(skb)->saddr,
1143 t1, IPPROTO_TCP, tot_len);
1147 buff->csum = csum_partial((char *)t1, tot_len, 0);
1149 memset(&fl, 0, sizeof(fl));
1150 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1151 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1153 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1154 tot_len, IPPROTO_TCP,
1157 fl.proto = IPPROTO_TCP;
1158 fl.oif = inet6_iif(skb);
1159 fl.fl_ip_dport = t1->dest;
1160 fl.fl_ip_sport = t1->source;
1161 security_skb_classify_flow(skb, &fl);
1163 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1164 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1165 ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1166 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1174 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1176 struct inet_timewait_sock *tw = inet_twsk(sk);
1177 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1179 tcp_v6_send_ack(tcptw, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1180 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1181 tcptw->tw_ts_recent);
1186 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
1188 tcp_v6_send_ack(NULL, skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent);
1192 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1194 struct request_sock *req, **prev;
1195 const struct tcphdr *th = tcp_hdr(skb);
1198 /* Find possible connection requests. */
1199 req = inet6_csk_search_req(sk, &prev, th->source,
1200 &ipv6_hdr(skb)->saddr,
1201 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1203 return tcp_check_req(sk, skb, req, prev);
1205 nsk = __inet6_lookup_established(&tcp_hashinfo, &ipv6_hdr(skb)->saddr,
1206 th->source, &ipv6_hdr(skb)->daddr,
1207 ntohs(th->dest), inet6_iif(skb));
1210 if (nsk->sk_state != TCP_TIME_WAIT) {
1214 inet_twsk_put(inet_twsk(nsk));
1218 #if 0 /*def CONFIG_SYN_COOKIES*/
1219 if (!th->rst && !th->syn && th->ack)
1220 sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
1225 /* FIXME: this is substantially similar to the ipv4 code.
1226 * Can some kind of merge be done? -- erics
1228 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1230 struct inet6_request_sock *treq;
1231 struct ipv6_pinfo *np = inet6_sk(sk);
1232 struct tcp_options_received tmp_opt;
1233 struct tcp_sock *tp = tcp_sk(sk);
1234 struct request_sock *req = NULL;
1235 __u32 isn = TCP_SKB_CB(skb)->when;
1237 if (skb->protocol == htons(ETH_P_IP))
1238 return tcp_v4_conn_request(sk, skb);
1240 if (!ipv6_unicast_destination(skb))
1244 * There are no SYN attacks on IPv6, yet...
1246 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1247 if (net_ratelimit())
1248 printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
1252 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1255 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1259 #ifdef CONFIG_TCP_MD5SIG
1260 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1263 tcp_clear_options(&tmp_opt);
1264 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1265 tmp_opt.user_mss = tp->rx_opt.user_mss;
1267 tcp_parse_options(skb, &tmp_opt, 0);
1269 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1270 tcp_openreq_init(req, &tmp_opt, skb);
1272 treq = inet6_rsk(req);
1273 ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1274 ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1275 TCP_ECN_create_request(req, tcp_hdr(skb));
1276 treq->pktopts = NULL;
1277 if (ipv6_opt_accepted(sk, skb) ||
1278 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1279 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1280 atomic_inc(&skb->users);
1281 treq->pktopts = skb;
1283 treq->iif = sk->sk_bound_dev_if;
1285 /* So that link locals have meaning */
1286 if (!sk->sk_bound_dev_if &&
1287 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1288 treq->iif = inet6_iif(skb);
1291 isn = tcp_v6_init_sequence(skb);
1293 tcp_rsk(req)->snt_isn = isn;
1295 security_inet_conn_request(sk, skb, req);
1297 if (tcp_v6_send_synack(sk, req, NULL))
1300 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1307 return 0; /* don't send reset */
1310 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1311 struct request_sock *req,
1312 struct dst_entry *dst)
1314 struct inet6_request_sock *treq = inet6_rsk(req);
1315 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1316 struct tcp6_sock *newtcp6sk;
1317 struct inet_sock *newinet;
1318 struct tcp_sock *newtp;
1320 struct ipv6_txoptions *opt;
1321 #ifdef CONFIG_TCP_MD5SIG
1322 struct tcp_md5sig_key *key;
1325 if (skb->protocol == htons(ETH_P_IP)) {
1330 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1335 newtcp6sk = (struct tcp6_sock *)newsk;
1336 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1338 newinet = inet_sk(newsk);
1339 newnp = inet6_sk(newsk);
1340 newtp = tcp_sk(newsk);
1342 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1344 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1347 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1350 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1352 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1353 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1354 #ifdef CONFIG_TCP_MD5SIG
1355 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1358 newnp->pktoptions = NULL;
1360 newnp->mcast_oif = inet6_iif(skb);
1361 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1364 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1365 * here, tcp_create_openreq_child now does this for us, see the comment in
1366 * that function for the gory details. -acme
1369 /* It is tricky place. Until this moment IPv4 tcp
1370 worked with IPv6 icsk.icsk_af_ops.
1373 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1380 if (sk_acceptq_is_full(sk))
1384 struct in6_addr *final_p = NULL, final;
1387 memset(&fl, 0, sizeof(fl));
1388 fl.proto = IPPROTO_TCP;
1389 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1390 if (opt && opt->srcrt) {
1391 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1392 ipv6_addr_copy(&final, &fl.fl6_dst);
1393 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1396 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1397 fl.oif = sk->sk_bound_dev_if;
1398 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1399 fl.fl_ip_sport = inet_sk(sk)->sport;
1400 security_req_classify_flow(req, &fl);
1402 if (ip6_dst_lookup(sk, &dst, &fl))
1406 ipv6_addr_copy(&fl.fl6_dst, final_p);
1408 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1412 newsk = tcp_create_openreq_child(sk, req, skb);
1417 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1418 * count here, tcp_create_openreq_child now does this for us, see the
1419 * comment in that function for the gory details. -acme
1422 newsk->sk_gso_type = SKB_GSO_TCPV6;
1423 __ip6_dst_store(newsk, dst, NULL, NULL);
1425 newtcp6sk = (struct tcp6_sock *)newsk;
1426 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1428 newtp = tcp_sk(newsk);
1429 newinet = inet_sk(newsk);
1430 newnp = inet6_sk(newsk);
1432 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1434 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1435 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1436 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1437 newsk->sk_bound_dev_if = treq->iif;
1439 /* Now IPv6 options...
1441 First: no IPv4 options.
1443 newinet->opt = NULL;
1444 newnp->ipv6_fl_list = NULL;
1447 newnp->rxopt.all = np->rxopt.all;
1449 /* Clone pktoptions received with SYN */
1450 newnp->pktoptions = NULL;
1451 if (treq->pktopts != NULL) {
1452 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1453 kfree_skb(treq->pktopts);
1454 treq->pktopts = NULL;
1455 if (newnp->pktoptions)
1456 skb_set_owner_r(newnp->pktoptions, newsk);
1459 newnp->mcast_oif = inet6_iif(skb);
1460 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1462 /* Clone native IPv6 options from listening socket (if any)
1464 Yes, keeping reference count would be much more clever,
1465 but we make one more one thing there: reattach optmem
1469 newnp->opt = ipv6_dup_options(newsk, opt);
1471 sock_kfree_s(sk, opt, opt->tot_len);
1474 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1476 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1477 newnp->opt->opt_flen);
1479 tcp_mtup_init(newsk);
1480 tcp_sync_mss(newsk, dst_mtu(dst));
1481 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1482 tcp_initialize_rcv_mss(newsk);
1484 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1486 #ifdef CONFIG_TCP_MD5SIG
1487 /* Copy over the MD5 key from the original socket */
1488 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1489 /* We're using one, so create a matching key
1490 * on the newsk structure. If we fail to get
1491 * memory, then we end up not copying the key
1494 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1496 tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1497 newkey, key->keylen);
1501 __inet6_hash(&tcp_hashinfo, newsk);
1502 inet_inherit_port(&tcp_hashinfo, sk, newsk);
1507 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1509 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1510 if (opt && opt != np->opt)
1511 sock_kfree_s(sk, opt, opt->tot_len);
1516 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1518 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1519 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1520 &ipv6_hdr(skb)->daddr, skb->csum)) {
1521 skb->ip_summed = CHECKSUM_UNNECESSARY;
1526 skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1527 &ipv6_hdr(skb)->saddr,
1528 &ipv6_hdr(skb)->daddr, 0));
1530 if (skb->len <= 76) {
1531 return __skb_checksum_complete(skb);
1536 /* The socket must have it's spinlock held when we get
1539 * We have a potential double-lock case here, so even when
1540 * doing backlog processing we use the BH locking scheme.
1541 * This is because we cannot sleep with the original spinlock
1544 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1546 struct ipv6_pinfo *np = inet6_sk(sk);
1547 struct tcp_sock *tp;
1548 struct sk_buff *opt_skb = NULL;
1550 /* Imagine: socket is IPv6. IPv4 packet arrives,
1551 goes to IPv4 receive handler and backlogged.
1552 From backlog it always goes here. Kerboom...
1553 Fortunately, tcp_rcv_established and rcv_established
1554 handle them correctly, but it is not case with
1555 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1558 if (skb->protocol == htons(ETH_P_IP))
1559 return tcp_v4_do_rcv(sk, skb);
1561 #ifdef CONFIG_TCP_MD5SIG
1562 if (tcp_v6_inbound_md5_hash (sk, skb))
1566 if (sk_filter(sk, skb))
1570 * socket locking is here for SMP purposes as backlog rcv
1571 * is currently called with bh processing disabled.
1574 /* Do Stevens' IPV6_PKTOPTIONS.
1576 Yes, guys, it is the only place in our code, where we
1577 may make it not affecting IPv4.
1578 The rest of code is protocol independent,
1579 and I do not like idea to uglify IPv4.
1581 Actually, all the idea behind IPV6_PKTOPTIONS
1582 looks not very well thought. For now we latch
1583 options, received in the last packet, enqueued
1584 by tcp. Feel free to propose better solution.
1588 opt_skb = skb_clone(skb, GFP_ATOMIC);
1590 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1591 TCP_CHECK_TIMER(sk);
1592 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1594 TCP_CHECK_TIMER(sk);
1596 goto ipv6_pktoptions;
1600 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1603 if (sk->sk_state == TCP_LISTEN) {
1604 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1609 * Queue it on the new socket if the new socket is active,
1610 * otherwise we just shortcircuit this and continue with
1614 if (tcp_child_process(sk, nsk, skb))
1617 __kfree_skb(opt_skb);
1622 TCP_CHECK_TIMER(sk);
1623 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1625 TCP_CHECK_TIMER(sk);
1627 goto ipv6_pktoptions;
1631 tcp_v6_send_reset(sk, skb);
1634 __kfree_skb(opt_skb);
1638 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1643 /* Do you ask, what is it?
1645 1. skb was enqueued by tcp.
1646 2. skb is added to tail of read queue, rather than out of order.
1647 3. socket is not in passive state.
1648 4. Finally, it really contains options, which user wants to receive.
1651 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1652 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1653 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1654 np->mcast_oif = inet6_iif(opt_skb);
1655 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1656 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1657 if (ipv6_opt_accepted(sk, opt_skb)) {
1658 skb_set_owner_r(opt_skb, sk);
1659 opt_skb = xchg(&np->pktoptions, opt_skb);
1661 __kfree_skb(opt_skb);
1662 opt_skb = xchg(&np->pktoptions, NULL);
1671 static int tcp_v6_rcv(struct sk_buff **pskb)
1673 struct sk_buff *skb = *pskb;
1678 if (skb->pkt_type != PACKET_HOST)
1682 * Count it even if it's bad.
1684 TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1686 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1691 if (th->doff < sizeof(struct tcphdr)/4)
1693 if (!pskb_may_pull(skb, th->doff*4))
1696 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1700 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1701 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1702 skb->len - th->doff*4);
1703 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1704 TCP_SKB_CB(skb)->when = 0;
1705 TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1706 TCP_SKB_CB(skb)->sacked = 0;
1708 sk = __inet6_lookup(&tcp_hashinfo, &ipv6_hdr(skb)->saddr, th->source,
1709 &ipv6_hdr(skb)->daddr, ntohs(th->dest),
1716 if (sk->sk_state == TCP_TIME_WAIT)
1719 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1720 goto discard_and_relse;
1722 if (sk_filter(sk, skb))
1723 goto discard_and_relse;
1727 bh_lock_sock_nested(sk);
1729 if (!sock_owned_by_user(sk)) {
1730 #ifdef CONFIG_NET_DMA
1731 struct tcp_sock *tp = tcp_sk(sk);
1732 if (tp->ucopy.dma_chan)
1733 ret = tcp_v6_do_rcv(sk, skb);
1737 if (!tcp_prequeue(sk, skb))
1738 ret = tcp_v6_do_rcv(sk, skb);
1741 sk_add_backlog(sk, skb);
1745 return ret ? -1 : 0;
1748 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1751 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1753 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1755 tcp_v6_send_reset(NULL, skb);
1772 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1773 inet_twsk_put(inet_twsk(sk));
1777 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1778 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1779 inet_twsk_put(inet_twsk(sk));
1783 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1788 sk2 = inet6_lookup_listener(&tcp_hashinfo,
1789 &ipv6_hdr(skb)->daddr,
1790 ntohs(th->dest), inet6_iif(skb));
1792 struct inet_timewait_sock *tw = inet_twsk(sk);
1793 inet_twsk_deschedule(tw, &tcp_death_row);
1798 /* Fall through to ACK */
1801 tcp_v6_timewait_ack(sk, skb);
1805 case TCP_TW_SUCCESS:;
1810 static int tcp_v6_remember_stamp(struct sock *sk)
1812 /* Alas, not yet... */
1816 static struct inet_connection_sock_af_ops ipv6_specific = {
1817 .queue_xmit = inet6_csk_xmit,
1818 .send_check = tcp_v6_send_check,
1819 .rebuild_header = inet6_sk_rebuild_header,
1820 .conn_request = tcp_v6_conn_request,
1821 .syn_recv_sock = tcp_v6_syn_recv_sock,
1822 .remember_stamp = tcp_v6_remember_stamp,
1823 .net_header_len = sizeof(struct ipv6hdr),
1824 .setsockopt = ipv6_setsockopt,
1825 .getsockopt = ipv6_getsockopt,
1826 .addr2sockaddr = inet6_csk_addr2sockaddr,
1827 .sockaddr_len = sizeof(struct sockaddr_in6),
1828 #ifdef CONFIG_COMPAT
1829 .compat_setsockopt = compat_ipv6_setsockopt,
1830 .compat_getsockopt = compat_ipv6_getsockopt,
1834 #ifdef CONFIG_TCP_MD5SIG
1835 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1836 .md5_lookup = tcp_v6_md5_lookup,
1837 .calc_md5_hash = tcp_v6_calc_md5_hash,
1838 .md5_add = tcp_v6_md5_add_func,
1839 .md5_parse = tcp_v6_parse_md5_keys,
1844 * TCP over IPv4 via INET6 API
1847 static struct inet_connection_sock_af_ops ipv6_mapped = {
1848 .queue_xmit = ip_queue_xmit,
1849 .send_check = tcp_v4_send_check,
1850 .rebuild_header = inet_sk_rebuild_header,
1851 .conn_request = tcp_v6_conn_request,
1852 .syn_recv_sock = tcp_v6_syn_recv_sock,
1853 .remember_stamp = tcp_v4_remember_stamp,
1854 .net_header_len = sizeof(struct iphdr),
1855 .setsockopt = ipv6_setsockopt,
1856 .getsockopt = ipv6_getsockopt,
1857 .addr2sockaddr = inet6_csk_addr2sockaddr,
1858 .sockaddr_len = sizeof(struct sockaddr_in6),
1859 #ifdef CONFIG_COMPAT
1860 .compat_setsockopt = compat_ipv6_setsockopt,
1861 .compat_getsockopt = compat_ipv6_getsockopt,
1865 #ifdef CONFIG_TCP_MD5SIG
1866 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1867 .md5_lookup = tcp_v4_md5_lookup,
1868 .calc_md5_hash = tcp_v4_calc_md5_hash,
1869 .md5_add = tcp_v6_md5_add_func,
1870 .md5_parse = tcp_v6_parse_md5_keys,
1874 /* NOTE: A lot of things set to zero explicitly by call to
1875 * sk_alloc() so need not be done here.
1877 static int tcp_v6_init_sock(struct sock *sk)
1879 struct inet_connection_sock *icsk = inet_csk(sk);
1880 struct tcp_sock *tp = tcp_sk(sk);
1882 skb_queue_head_init(&tp->out_of_order_queue);
1883 tcp_init_xmit_timers(sk);
1884 tcp_prequeue_init(tp);
1886 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1887 tp->mdev = TCP_TIMEOUT_INIT;
1889 /* So many TCP implementations out there (incorrectly) count the
1890 * initial SYN frame in their delayed-ACK and congestion control
1891 * algorithms that we must have the following bandaid to talk
1892 * efficiently to them. -DaveM
1896 /* See draft-stevens-tcpca-spec-01 for discussion of the
1897 * initialization of these values.
1899 tp->snd_ssthresh = 0x7fffffff;
1900 tp->snd_cwnd_clamp = ~0;
1901 tp->mss_cache = 536;
1903 tp->reordering = sysctl_tcp_reordering;
1905 sk->sk_state = TCP_CLOSE;
1907 icsk->icsk_af_ops = &ipv6_specific;
1908 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1909 icsk->icsk_sync_mss = tcp_sync_mss;
1910 sk->sk_write_space = sk_stream_write_space;
1911 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1913 #ifdef CONFIG_TCP_MD5SIG
1914 tp->af_specific = &tcp_sock_ipv6_specific;
1917 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1918 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1920 atomic_inc(&tcp_sockets_allocated);
1925 static int tcp_v6_destroy_sock(struct sock *sk)
1927 #ifdef CONFIG_TCP_MD5SIG
1928 /* Clean up the MD5 key list */
1929 if (tcp_sk(sk)->md5sig_info)
1930 tcp_v6_clear_md5_list(sk);
1932 tcp_v4_destroy_sock(sk);
1933 return inet6_destroy_sock(sk);
1936 #ifdef CONFIG_PROC_FS
1937 /* Proc filesystem TCPv6 sock list dumping. */
1938 static void get_openreq6(struct seq_file *seq,
1939 struct sock *sk, struct request_sock *req, int i, int uid)
1941 int ttd = req->expires - jiffies;
1942 struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1943 struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1949 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1950 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1952 src->s6_addr32[0], src->s6_addr32[1],
1953 src->s6_addr32[2], src->s6_addr32[3],
1954 ntohs(inet_sk(sk)->sport),
1955 dest->s6_addr32[0], dest->s6_addr32[1],
1956 dest->s6_addr32[2], dest->s6_addr32[3],
1957 ntohs(inet_rsk(req)->rmt_port),
1959 0,0, /* could print option size, but that is af dependent. */
1960 1, /* timers active (only the expire timer) */
1961 jiffies_to_clock_t(ttd),
1964 0, /* non standard timer */
1965 0, /* open_requests have no inode */
1969 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1971 struct in6_addr *dest, *src;
1974 unsigned long timer_expires;
1975 struct inet_sock *inet = inet_sk(sp);
1976 struct tcp_sock *tp = tcp_sk(sp);
1977 const struct inet_connection_sock *icsk = inet_csk(sp);
1978 struct ipv6_pinfo *np = inet6_sk(sp);
1981 src = &np->rcv_saddr;
1982 destp = ntohs(inet->dport);
1983 srcp = ntohs(inet->sport);
1985 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1987 timer_expires = icsk->icsk_timeout;
1988 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1990 timer_expires = icsk->icsk_timeout;
1991 } else if (timer_pending(&sp->sk_timer)) {
1993 timer_expires = sp->sk_timer.expires;
1996 timer_expires = jiffies;
2000 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2001 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\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,
2008 tp->write_seq-tp->snd_una,
2009 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2011 jiffies_to_clock_t(timer_expires - jiffies),
2012 icsk->icsk_retransmits,
2014 icsk->icsk_probes_out,
2016 atomic_read(&sp->sk_refcnt), sp,
2019 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2020 tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2024 static void get_timewait6_sock(struct seq_file *seq,
2025 struct inet_timewait_sock *tw, int i)
2027 struct in6_addr *dest, *src;
2029 struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2030 int ttd = tw->tw_ttd - jiffies;
2035 dest = &tw6->tw_v6_daddr;
2036 src = &tw6->tw_v6_rcv_saddr;
2037 destp = ntohs(tw->tw_dport);
2038 srcp = ntohs(tw->tw_sport);
2041 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2042 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2044 src->s6_addr32[0], src->s6_addr32[1],
2045 src->s6_addr32[2], src->s6_addr32[3], srcp,
2046 dest->s6_addr32[0], dest->s6_addr32[1],
2047 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2048 tw->tw_substate, 0, 0,
2049 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2050 atomic_read(&tw->tw_refcnt), tw);
2053 static int tcp6_seq_show(struct seq_file *seq, void *v)
2055 struct tcp_iter_state *st;
2057 if (v == SEQ_START_TOKEN) {
2062 "st tx_queue rx_queue tr tm->when retrnsmt"
2063 " uid timeout inode\n");
2068 switch (st->state) {
2069 case TCP_SEQ_STATE_LISTENING:
2070 case TCP_SEQ_STATE_ESTABLISHED:
2071 get_tcp6_sock(seq, v, st->num);
2073 case TCP_SEQ_STATE_OPENREQ:
2074 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2076 case TCP_SEQ_STATE_TIME_WAIT:
2077 get_timewait6_sock(seq, v, st->num);
2084 static struct file_operations tcp6_seq_fops;
2085 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2086 .owner = THIS_MODULE,
2089 .seq_show = tcp6_seq_show,
2090 .seq_fops = &tcp6_seq_fops,
2093 int __init tcp6_proc_init(void)
2095 return tcp_proc_register(&tcp6_seq_afinfo);
2098 void tcp6_proc_exit(void)
2100 tcp_proc_unregister(&tcp6_seq_afinfo);
2104 struct proto tcpv6_prot = {
2106 .owner = THIS_MODULE,
2108 .connect = tcp_v6_connect,
2109 .disconnect = tcp_disconnect,
2110 .accept = inet_csk_accept,
2112 .init = tcp_v6_init_sock,
2113 .destroy = tcp_v6_destroy_sock,
2114 .shutdown = tcp_shutdown,
2115 .setsockopt = tcp_setsockopt,
2116 .getsockopt = tcp_getsockopt,
2117 .recvmsg = tcp_recvmsg,
2118 .backlog_rcv = tcp_v6_do_rcv,
2119 .hash = tcp_v6_hash,
2120 .unhash = tcp_unhash,
2121 .get_port = tcp_v6_get_port,
2122 .enter_memory_pressure = tcp_enter_memory_pressure,
2123 .sockets_allocated = &tcp_sockets_allocated,
2124 .memory_allocated = &tcp_memory_allocated,
2125 .memory_pressure = &tcp_memory_pressure,
2126 .orphan_count = &tcp_orphan_count,
2127 .sysctl_mem = sysctl_tcp_mem,
2128 .sysctl_wmem = sysctl_tcp_wmem,
2129 .sysctl_rmem = sysctl_tcp_rmem,
2130 .max_header = MAX_TCP_HEADER,
2131 .obj_size = sizeof(struct tcp6_sock),
2132 .twsk_prot = &tcp6_timewait_sock_ops,
2133 .rsk_prot = &tcp6_request_sock_ops,
2134 #ifdef CONFIG_COMPAT
2135 .compat_setsockopt = compat_tcp_setsockopt,
2136 .compat_getsockopt = compat_tcp_getsockopt,
2140 static struct inet6_protocol tcpv6_protocol = {
2141 .handler = tcp_v6_rcv,
2142 .err_handler = tcp_v6_err,
2143 .gso_send_check = tcp_v6_gso_send_check,
2144 .gso_segment = tcp_tso_segment,
2145 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2148 static struct inet_protosw tcpv6_protosw = {
2149 .type = SOCK_STREAM,
2150 .protocol = IPPROTO_TCP,
2151 .prot = &tcpv6_prot,
2152 .ops = &inet6_stream_ops,
2155 .flags = INET_PROTOSW_PERMANENT |
2159 void __init tcpv6_init(void)
2161 /* register inet6 protocol */
2162 if (inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP) < 0)
2163 printk(KERN_ERR "tcpv6_init: Could not register protocol\n");
2164 inet6_register_protosw(&tcpv6_protosw);
2166 if (inet_csk_ctl_sock_create(&tcp6_socket, PF_INET6, SOCK_RAW,
2168 panic("Failed to create the TCPv6 control socket.\n");