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>
59 #include <net/addrconf.h>
61 #include <net/dsfield.h>
62 #include <net/timewait_sock.h>
64 #include <asm/uaccess.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
69 #include <linux/crypto.h>
70 #include <linux/scatterlist.h>
72 /* Socket used for sending RSTs and ACKs */
73 static struct socket *tcp6_socket;
75 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
76 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req);
77 static void tcp_v6_send_check(struct sock *sk, int len,
80 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
82 static struct inet_connection_sock_af_ops ipv6_mapped;
83 static struct inet_connection_sock_af_ops ipv6_specific;
84 #ifdef CONFIG_TCP_MD5SIG
85 static struct tcp_sock_af_ops tcp_sock_ipv6_specific;
86 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
89 static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
91 return inet_csk_get_port(&tcp_hashinfo, sk, snum,
92 inet6_csk_bind_conflict);
95 static void tcp_v6_hash(struct sock *sk)
97 if (sk->sk_state != TCP_CLOSE) {
98 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
103 __inet6_hash(&tcp_hashinfo, sk);
108 static __inline__ __sum16 tcp_v6_check(struct tcphdr *th, int len,
109 struct in6_addr *saddr,
110 struct in6_addr *daddr,
113 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
116 static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
118 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
119 ipv6_hdr(skb)->saddr.s6_addr32,
121 tcp_hdr(skb)->source);
124 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
127 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
128 struct inet_sock *inet = inet_sk(sk);
129 struct inet_connection_sock *icsk = inet_csk(sk);
130 struct ipv6_pinfo *np = inet6_sk(sk);
131 struct tcp_sock *tp = tcp_sk(sk);
132 struct in6_addr *saddr = NULL, *final_p = NULL, final;
134 struct dst_entry *dst;
138 if (addr_len < SIN6_LEN_RFC2133)
141 if (usin->sin6_family != AF_INET6)
142 return(-EAFNOSUPPORT);
144 memset(&fl, 0, sizeof(fl));
147 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
148 IP6_ECN_flow_init(fl.fl6_flowlabel);
149 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
150 struct ip6_flowlabel *flowlabel;
151 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
152 if (flowlabel == NULL)
154 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
155 fl6_sock_release(flowlabel);
160 * connect() to INADDR_ANY means loopback (BSD'ism).
163 if(ipv6_addr_any(&usin->sin6_addr))
164 usin->sin6_addr.s6_addr[15] = 0x1;
166 addr_type = ipv6_addr_type(&usin->sin6_addr);
168 if(addr_type & IPV6_ADDR_MULTICAST)
171 if (addr_type&IPV6_ADDR_LINKLOCAL) {
172 if (addr_len >= sizeof(struct sockaddr_in6) &&
173 usin->sin6_scope_id) {
174 /* If interface is set while binding, indices
177 if (sk->sk_bound_dev_if &&
178 sk->sk_bound_dev_if != usin->sin6_scope_id)
181 sk->sk_bound_dev_if = usin->sin6_scope_id;
184 /* Connect to link-local address requires an interface */
185 if (!sk->sk_bound_dev_if)
189 if (tp->rx_opt.ts_recent_stamp &&
190 !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
191 tp->rx_opt.ts_recent = 0;
192 tp->rx_opt.ts_recent_stamp = 0;
196 ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
197 np->flow_label = fl.fl6_flowlabel;
203 if (addr_type == IPV6_ADDR_MAPPED) {
204 u32 exthdrlen = icsk->icsk_ext_hdr_len;
205 struct sockaddr_in sin;
207 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
209 if (__ipv6_only_sock(sk))
212 sin.sin_family = AF_INET;
213 sin.sin_port = usin->sin6_port;
214 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
216 icsk->icsk_af_ops = &ipv6_mapped;
217 sk->sk_backlog_rcv = tcp_v4_do_rcv;
218 #ifdef CONFIG_TCP_MD5SIG
219 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
222 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
225 icsk->icsk_ext_hdr_len = exthdrlen;
226 icsk->icsk_af_ops = &ipv6_specific;
227 sk->sk_backlog_rcv = tcp_v6_do_rcv;
228 #ifdef CONFIG_TCP_MD5SIG
229 tp->af_specific = &tcp_sock_ipv6_specific;
233 ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
235 ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
242 if (!ipv6_addr_any(&np->rcv_saddr))
243 saddr = &np->rcv_saddr;
245 fl.proto = IPPROTO_TCP;
246 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
247 ipv6_addr_copy(&fl.fl6_src,
248 (saddr ? saddr : &np->saddr));
249 fl.oif = sk->sk_bound_dev_if;
250 fl.fl_ip_dport = usin->sin6_port;
251 fl.fl_ip_sport = inet->sport;
253 if (np->opt && np->opt->srcrt) {
254 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
255 ipv6_addr_copy(&final, &fl.fl6_dst);
256 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
260 security_sk_classify_flow(sk, &fl);
262 err = ip6_dst_lookup(sk, &dst, &fl);
266 ipv6_addr_copy(&fl.fl6_dst, final_p);
268 if ((err = xfrm_lookup(&dst, &fl, sk, 1)) < 0)
273 ipv6_addr_copy(&np->rcv_saddr, saddr);
276 /* set the source address */
277 ipv6_addr_copy(&np->saddr, saddr);
278 inet->rcv_saddr = LOOPBACK4_IPV6;
280 sk->sk_gso_type = SKB_GSO_TCPV6;
281 __ip6_dst_store(sk, dst, NULL, NULL);
283 icsk->icsk_ext_hdr_len = 0;
285 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
288 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
290 inet->dport = usin->sin6_port;
292 tcp_set_state(sk, TCP_SYN_SENT);
293 err = inet6_hash_connect(&tcp_death_row, sk);
298 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
303 err = tcp_connect(sk);
310 tcp_set_state(sk, TCP_CLOSE);
314 sk->sk_route_caps = 0;
318 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
319 int type, int code, int offset, __be32 info)
321 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
322 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
323 struct ipv6_pinfo *np;
329 sk = inet6_lookup(&tcp_hashinfo, &hdr->daddr, th->dest, &hdr->saddr,
330 th->source, skb->dev->ifindex);
333 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
337 if (sk->sk_state == TCP_TIME_WAIT) {
338 inet_twsk_put(inet_twsk(sk));
343 if (sock_owned_by_user(sk))
344 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
346 if (sk->sk_state == TCP_CLOSE)
350 seq = ntohl(th->seq);
351 if (sk->sk_state != TCP_LISTEN &&
352 !between(seq, tp->snd_una, tp->snd_nxt)) {
353 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
359 if (type == ICMPV6_PKT_TOOBIG) {
360 struct dst_entry *dst = NULL;
362 if (sock_owned_by_user(sk))
364 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
367 /* icmp should have updated the destination cache entry */
368 dst = __sk_dst_check(sk, np->dst_cookie);
371 struct inet_sock *inet = inet_sk(sk);
374 /* BUGGG_FUTURE: Again, it is not clear how
375 to handle rthdr case. Ignore this complexity
378 memset(&fl, 0, sizeof(fl));
379 fl.proto = IPPROTO_TCP;
380 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
381 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
382 fl.oif = sk->sk_bound_dev_if;
383 fl.fl_ip_dport = inet->dport;
384 fl.fl_ip_sport = inet->sport;
385 security_skb_classify_flow(skb, &fl);
387 if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
388 sk->sk_err_soft = -err;
392 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
393 sk->sk_err_soft = -err;
400 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
401 tcp_sync_mss(sk, dst_mtu(dst));
402 tcp_simple_retransmit(sk);
403 } /* else let the usual retransmit timer handle it */
408 icmpv6_err_convert(type, code, &err);
410 /* Might be for an request_sock */
411 switch (sk->sk_state) {
412 struct request_sock *req, **prev;
414 if (sock_owned_by_user(sk))
417 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
418 &hdr->saddr, inet6_iif(skb));
422 /* ICMPs are not backlogged, hence we cannot get
423 * an established socket here.
425 BUG_TRAP(req->sk == NULL);
427 if (seq != tcp_rsk(req)->snt_isn) {
428 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
432 inet_csk_reqsk_queue_drop(sk, req, prev);
436 case TCP_SYN_RECV: /* Cannot happen.
437 It can, it SYNs are crossed. --ANK */
438 if (!sock_owned_by_user(sk)) {
440 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
444 sk->sk_err_soft = err;
448 if (!sock_owned_by_user(sk) && np->recverr) {
450 sk->sk_error_report(sk);
452 sk->sk_err_soft = err;
460 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
461 struct dst_entry *dst)
463 struct inet6_request_sock *treq = inet6_rsk(req);
464 struct ipv6_pinfo *np = inet6_sk(sk);
465 struct sk_buff * skb;
466 struct ipv6_txoptions *opt = NULL;
467 struct in6_addr * final_p = NULL, final;
471 memset(&fl, 0, sizeof(fl));
472 fl.proto = IPPROTO_TCP;
473 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
474 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
475 fl.fl6_flowlabel = 0;
477 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
478 fl.fl_ip_sport = inet_sk(sk)->sport;
479 security_req_classify_flow(req, &fl);
484 np->rxopt.bits.osrcrt == 2 &&
486 struct sk_buff *pktopts = treq->pktopts;
487 struct inet6_skb_parm *rxopt = IP6CB(pktopts);
489 opt = ipv6_invert_rthdr(sk,
490 (struct ipv6_rt_hdr *)(skb_network_header(pktopts) +
494 if (opt && opt->srcrt) {
495 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
496 ipv6_addr_copy(&final, &fl.fl6_dst);
497 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
501 err = ip6_dst_lookup(sk, &dst, &fl);
505 ipv6_addr_copy(&fl.fl6_dst, final_p);
506 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
510 skb = tcp_make_synack(sk, dst, req);
512 struct tcphdr *th = tcp_hdr(skb);
514 th->check = tcp_v6_check(th, skb->len,
515 &treq->loc_addr, &treq->rmt_addr,
516 csum_partial((char *)th, skb->len, skb->csum));
518 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
519 err = ip6_xmit(sk, skb, &fl, opt, 0);
520 err = net_xmit_eval(err);
524 if (opt && opt != np->opt)
525 sock_kfree_s(sk, opt, opt->tot_len);
530 static void tcp_v6_reqsk_destructor(struct request_sock *req)
532 if (inet6_rsk(req)->pktopts)
533 kfree_skb(inet6_rsk(req)->pktopts);
536 #ifdef CONFIG_TCP_MD5SIG
537 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
538 struct in6_addr *addr)
540 struct tcp_sock *tp = tcp_sk(sk);
545 if (!tp->md5sig_info || !tp->md5sig_info->entries6)
548 for (i = 0; i < tp->md5sig_info->entries6; i++) {
549 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, addr) == 0)
550 return (struct tcp_md5sig_key *)&tp->md5sig_info->keys6[i];
555 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
556 struct sock *addr_sk)
558 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
561 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
562 struct request_sock *req)
564 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
567 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
568 char *newkey, u8 newkeylen)
570 /* Add key to the list */
571 struct tcp6_md5sig_key *key;
572 struct tcp_sock *tp = tcp_sk(sk);
573 struct tcp6_md5sig_key *keys;
575 key = (struct tcp6_md5sig_key*) tcp_v6_md5_do_lookup(sk, peer);
577 /* modify existing entry - just update that one */
580 key->keylen = newkeylen;
582 /* reallocate new list if current one is full. */
583 if (!tp->md5sig_info) {
584 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
585 if (!tp->md5sig_info) {
590 tcp_alloc_md5sig_pool();
591 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
592 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
593 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
596 tcp_free_md5sig_pool();
601 if (tp->md5sig_info->entries6)
602 memmove(keys, tp->md5sig_info->keys6,
603 (sizeof (tp->md5sig_info->keys6[0]) *
604 tp->md5sig_info->entries6));
606 kfree(tp->md5sig_info->keys6);
607 tp->md5sig_info->keys6 = keys;
608 tp->md5sig_info->alloced6++;
611 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
613 tp->md5sig_info->keys6[tp->md5sig_info->entries6].key = newkey;
614 tp->md5sig_info->keys6[tp->md5sig_info->entries6].keylen = newkeylen;
616 tp->md5sig_info->entries6++;
621 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
622 u8 *newkey, __u8 newkeylen)
624 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
628 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
630 struct tcp_sock *tp = tcp_sk(sk);
633 for (i = 0; i < tp->md5sig_info->entries6; i++) {
634 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, peer) == 0) {
636 kfree(tp->md5sig_info->keys6[i].key);
637 tp->md5sig_info->entries6--;
639 if (tp->md5sig_info->entries6 == 0) {
640 kfree(tp->md5sig_info->keys6);
641 tp->md5sig_info->keys6 = NULL;
643 tcp_free_md5sig_pool();
647 /* shrink the database */
648 if (tp->md5sig_info->entries6 != i)
649 memmove(&tp->md5sig_info->keys6[i],
650 &tp->md5sig_info->keys6[i+1],
651 (tp->md5sig_info->entries6 - i)
652 * sizeof (tp->md5sig_info->keys6[0]));
659 static void tcp_v6_clear_md5_list (struct sock *sk)
661 struct tcp_sock *tp = tcp_sk(sk);
664 if (tp->md5sig_info->entries6) {
665 for (i = 0; i < tp->md5sig_info->entries6; i++)
666 kfree(tp->md5sig_info->keys6[i].key);
667 tp->md5sig_info->entries6 = 0;
668 tcp_free_md5sig_pool();
671 kfree(tp->md5sig_info->keys6);
672 tp->md5sig_info->keys6 = NULL;
673 tp->md5sig_info->alloced6 = 0;
675 if (tp->md5sig_info->entries4) {
676 for (i = 0; i < tp->md5sig_info->entries4; i++)
677 kfree(tp->md5sig_info->keys4[i].key);
678 tp->md5sig_info->entries4 = 0;
679 tcp_free_md5sig_pool();
682 kfree(tp->md5sig_info->keys4);
683 tp->md5sig_info->keys4 = NULL;
684 tp->md5sig_info->alloced4 = 0;
687 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
690 struct tcp_md5sig cmd;
691 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
694 if (optlen < sizeof(cmd))
697 if (copy_from_user(&cmd, optval, sizeof(cmd)))
700 if (sin6->sin6_family != AF_INET6)
703 if (!cmd.tcpm_keylen) {
704 if (!tcp_sk(sk)->md5sig_info)
706 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_MAPPED)
707 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
708 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
711 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
714 if (!tcp_sk(sk)->md5sig_info) {
715 struct tcp_sock *tp = tcp_sk(sk);
716 struct tcp_md5sig_info *p;
718 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
725 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
728 if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_MAPPED) {
729 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
730 newkey, cmd.tcpm_keylen);
732 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
735 static int tcp_v6_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
736 struct in6_addr *saddr,
737 struct in6_addr *daddr,
738 struct tcphdr *th, int protocol,
741 struct scatterlist sg[4];
745 struct tcp_md5sig_pool *hp;
746 struct tcp6_pseudohdr *bp;
747 struct hash_desc *desc;
749 unsigned int nbytes = 0;
751 hp = tcp_get_md5sig_pool();
753 printk(KERN_WARNING "%s(): hash pool not found...\n", __FUNCTION__);
754 goto clear_hash_noput;
756 bp = &hp->md5_blk.ip6;
757 desc = &hp->md5_desc;
759 /* 1. TCP pseudo-header (RFC2460) */
760 ipv6_addr_copy(&bp->saddr, saddr);
761 ipv6_addr_copy(&bp->daddr, daddr);
762 bp->len = htonl(tcplen);
763 bp->protocol = htonl(protocol);
765 sg_set_buf(&sg[block++], bp, sizeof(*bp));
766 nbytes += sizeof(*bp);
768 /* 2. TCP header, excluding options */
771 sg_set_buf(&sg[block++], th, sizeof(*th));
772 nbytes += sizeof(*th);
774 /* 3. TCP segment data (if any) */
775 data_len = tcplen - (th->doff << 2);
777 u8 *data = (u8 *)th + (th->doff << 2);
778 sg_set_buf(&sg[block++], data, data_len);
783 sg_set_buf(&sg[block++], key->key, key->keylen);
784 nbytes += key->keylen;
786 /* Now store the hash into the packet */
787 err = crypto_hash_init(desc);
789 printk(KERN_WARNING "%s(): hash_init failed\n", __FUNCTION__);
792 err = crypto_hash_update(desc, sg, nbytes);
794 printk(KERN_WARNING "%s(): hash_update failed\n", __FUNCTION__);
797 err = crypto_hash_final(desc, md5_hash);
799 printk(KERN_WARNING "%s(): hash_final failed\n", __FUNCTION__);
803 /* Reset header, and free up the crypto */
804 tcp_put_md5sig_pool();
809 tcp_put_md5sig_pool();
811 memset(md5_hash, 0, 16);
815 static int tcp_v6_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
817 struct dst_entry *dst,
818 struct request_sock *req,
819 struct tcphdr *th, int protocol,
822 struct in6_addr *saddr, *daddr;
825 saddr = &inet6_sk(sk)->saddr;
826 daddr = &inet6_sk(sk)->daddr;
828 saddr = &inet6_rsk(req)->loc_addr;
829 daddr = &inet6_rsk(req)->rmt_addr;
831 return tcp_v6_do_calc_md5_hash(md5_hash, key,
833 th, protocol, tcplen);
836 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
838 __u8 *hash_location = NULL;
839 struct tcp_md5sig_key *hash_expected;
840 struct ipv6hdr *ip6h = ipv6_hdr(skb);
841 struct tcphdr *th = tcp_hdr(skb);
842 int length = (th->doff << 2) - sizeof (*th);
847 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
849 /* If the TCP option is too short, we can short cut */
850 if (length < TCPOLEN_MD5SIG)
851 return hash_expected ? 1 : 0;
867 if (opsize < 2 || opsize > length)
869 if (opcode == TCPOPT_MD5SIG) {
879 /* do we have a hash as expected? */
880 if (!hash_expected) {
883 if (net_ratelimit()) {
884 printk(KERN_INFO "MD5 Hash NOT expected but found "
885 "(" NIP6_FMT ", %u)->"
886 "(" NIP6_FMT ", %u)\n",
887 NIP6(ip6h->saddr), ntohs(th->source),
888 NIP6(ip6h->daddr), ntohs(th->dest));
893 if (!hash_location) {
894 if (net_ratelimit()) {
895 printk(KERN_INFO "MD5 Hash expected but NOT found "
896 "(" NIP6_FMT ", %u)->"
897 "(" NIP6_FMT ", %u)\n",
898 NIP6(ip6h->saddr), ntohs(th->source),
899 NIP6(ip6h->daddr), ntohs(th->dest));
904 /* check the signature */
905 genhash = tcp_v6_do_calc_md5_hash(newhash,
907 &ip6h->saddr, &ip6h->daddr,
910 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
911 if (net_ratelimit()) {
912 printk(KERN_INFO "MD5 Hash %s for "
913 "(" NIP6_FMT ", %u)->"
914 "(" NIP6_FMT ", %u)\n",
915 genhash ? "failed" : "mismatch",
916 NIP6(ip6h->saddr), ntohs(th->source),
917 NIP6(ip6h->daddr), ntohs(th->dest));
925 static struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
927 .obj_size = sizeof(struct tcp6_request_sock),
928 .rtx_syn_ack = tcp_v6_send_synack,
929 .send_ack = tcp_v6_reqsk_send_ack,
930 .destructor = tcp_v6_reqsk_destructor,
931 .send_reset = tcp_v6_send_reset
934 #ifdef CONFIG_TCP_MD5SIG
935 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
936 .md5_lookup = tcp_v6_reqsk_md5_lookup,
940 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
941 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
942 .twsk_unique = tcp_twsk_unique,
943 .twsk_destructor= tcp_twsk_destructor,
946 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
948 struct ipv6_pinfo *np = inet6_sk(sk);
949 struct tcphdr *th = tcp_hdr(skb);
951 if (skb->ip_summed == CHECKSUM_PARTIAL) {
952 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 0);
953 skb->csum_start = skb_transport_header(skb) - skb->head;
954 skb->csum_offset = offsetof(struct tcphdr, check);
956 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
957 csum_partial((char *)th, th->doff<<2,
962 static int tcp_v6_gso_send_check(struct sk_buff *skb)
964 struct ipv6hdr *ipv6h;
967 if (!pskb_may_pull(skb, sizeof(*th)))
970 ipv6h = ipv6_hdr(skb);
974 th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
976 skb->csum_start = skb_transport_header(skb) - skb->head;
977 skb->csum_offset = offsetof(struct tcphdr, check);
978 skb->ip_summed = CHECKSUM_PARTIAL;
982 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
984 struct tcphdr *th = tcp_hdr(skb), *t1;
985 struct sk_buff *buff;
987 int tot_len = sizeof(*th);
988 #ifdef CONFIG_TCP_MD5SIG
989 struct tcp_md5sig_key *key;
995 if (!ipv6_unicast_destination(skb))
998 #ifdef CONFIG_TCP_MD5SIG
1000 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1005 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1009 * We need to grab some memory, and put together an RST,
1010 * and then put it into the queue to be sent.
1013 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1018 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1020 t1 = (struct tcphdr *) skb_push(buff, tot_len);
1022 /* Swap the send and the receive. */
1023 memset(t1, 0, sizeof(*t1));
1024 t1->dest = th->source;
1025 t1->source = th->dest;
1026 t1->doff = tot_len / 4;
1030 t1->seq = th->ack_seq;
1033 t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
1034 + skb->len - (th->doff<<2));
1037 #ifdef CONFIG_TCP_MD5SIG
1039 __be32 *opt = (__be32*)(t1 + 1);
1040 opt[0] = htonl((TCPOPT_NOP << 24) |
1041 (TCPOPT_NOP << 16) |
1042 (TCPOPT_MD5SIG << 8) |
1044 tcp_v6_do_calc_md5_hash((__u8 *)&opt[1], key,
1045 &ipv6_hdr(skb)->daddr,
1046 &ipv6_hdr(skb)->saddr,
1047 t1, IPPROTO_TCP, tot_len);
1051 buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1053 memset(&fl, 0, sizeof(fl));
1054 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1055 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1057 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1058 sizeof(*t1), IPPROTO_TCP,
1061 fl.proto = IPPROTO_TCP;
1062 fl.oif = inet6_iif(skb);
1063 fl.fl_ip_dport = t1->dest;
1064 fl.fl_ip_sport = t1->source;
1065 security_skb_classify_flow(skb, &fl);
1067 /* sk = NULL, but it is safe for now. RST socket required. */
1068 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1070 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1071 ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1072 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1073 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
1081 static void tcp_v6_send_ack(struct tcp_timewait_sock *tw,
1082 struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
1084 struct tcphdr *th = tcp_hdr(skb), *t1;
1085 struct sk_buff *buff;
1087 int tot_len = sizeof(struct tcphdr);
1089 #ifdef CONFIG_TCP_MD5SIG
1090 struct tcp_md5sig_key *key;
1091 struct tcp_md5sig_key tw_key;
1094 #ifdef CONFIG_TCP_MD5SIG
1095 if (!tw && skb->sk) {
1096 key = tcp_v6_md5_do_lookup(skb->sk, &ipv6_hdr(skb)->daddr);
1097 } else if (tw && tw->tw_md5_keylen) {
1098 tw_key.key = tw->tw_md5_key;
1099 tw_key.keylen = tw->tw_md5_keylen;
1107 tot_len += TCPOLEN_TSTAMP_ALIGNED;
1108 #ifdef CONFIG_TCP_MD5SIG
1110 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1113 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1118 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1120 t1 = (struct tcphdr *) skb_push(buff,tot_len);
1122 /* Swap the send and the receive. */
1123 memset(t1, 0, sizeof(*t1));
1124 t1->dest = th->source;
1125 t1->source = th->dest;
1126 t1->doff = tot_len/4;
1127 t1->seq = htonl(seq);
1128 t1->ack_seq = htonl(ack);
1130 t1->window = htons(win);
1132 topt = (__be32 *)(t1 + 1);
1135 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1136 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1137 *topt++ = htonl(tcp_time_stamp);
1141 #ifdef CONFIG_TCP_MD5SIG
1143 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1144 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1145 tcp_v6_do_calc_md5_hash((__u8 *)topt, key,
1146 &ipv6_hdr(skb)->daddr,
1147 &ipv6_hdr(skb)->saddr,
1148 t1, IPPROTO_TCP, tot_len);
1152 buff->csum = csum_partial((char *)t1, tot_len, 0);
1154 memset(&fl, 0, sizeof(fl));
1155 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1156 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1158 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1159 tot_len, IPPROTO_TCP,
1162 fl.proto = IPPROTO_TCP;
1163 fl.oif = inet6_iif(skb);
1164 fl.fl_ip_dport = t1->dest;
1165 fl.fl_ip_sport = t1->source;
1166 security_skb_classify_flow(skb, &fl);
1168 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1169 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1170 ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1171 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1179 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1181 struct inet_timewait_sock *tw = inet_twsk(sk);
1182 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1184 tcp_v6_send_ack(tcptw, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1185 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1186 tcptw->tw_ts_recent);
1191 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
1193 tcp_v6_send_ack(NULL, skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent);
1197 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1199 struct request_sock *req, **prev;
1200 const struct tcphdr *th = tcp_hdr(skb);
1203 /* Find possible connection requests. */
1204 req = inet6_csk_search_req(sk, &prev, th->source,
1205 &ipv6_hdr(skb)->saddr,
1206 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1208 return tcp_check_req(sk, skb, req, prev);
1210 nsk = __inet6_lookup_established(&tcp_hashinfo, &ipv6_hdr(skb)->saddr,
1211 th->source, &ipv6_hdr(skb)->daddr,
1212 ntohs(th->dest), inet6_iif(skb));
1215 if (nsk->sk_state != TCP_TIME_WAIT) {
1219 inet_twsk_put(inet_twsk(nsk));
1223 #if 0 /*def CONFIG_SYN_COOKIES*/
1224 if (!th->rst && !th->syn && th->ack)
1225 sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
1230 /* FIXME: this is substantially similar to the ipv4 code.
1231 * Can some kind of merge be done? -- erics
1233 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1235 struct inet6_request_sock *treq;
1236 struct ipv6_pinfo *np = inet6_sk(sk);
1237 struct tcp_options_received tmp_opt;
1238 struct tcp_sock *tp = tcp_sk(sk);
1239 struct request_sock *req = NULL;
1240 __u32 isn = TCP_SKB_CB(skb)->when;
1242 if (skb->protocol == htons(ETH_P_IP))
1243 return tcp_v4_conn_request(sk, skb);
1245 if (!ipv6_unicast_destination(skb))
1249 * There are no SYN attacks on IPv6, yet...
1251 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1252 if (net_ratelimit())
1253 printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
1257 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1260 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1264 #ifdef CONFIG_TCP_MD5SIG
1265 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1268 tcp_clear_options(&tmp_opt);
1269 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1270 tmp_opt.user_mss = tp->rx_opt.user_mss;
1272 tcp_parse_options(skb, &tmp_opt, 0);
1274 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1275 tcp_openreq_init(req, &tmp_opt, skb);
1277 treq = inet6_rsk(req);
1278 ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1279 ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1280 TCP_ECN_create_request(req, tcp_hdr(skb));
1281 treq->pktopts = NULL;
1282 if (ipv6_opt_accepted(sk, skb) ||
1283 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1284 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1285 atomic_inc(&skb->users);
1286 treq->pktopts = skb;
1288 treq->iif = sk->sk_bound_dev_if;
1290 /* So that link locals have meaning */
1291 if (!sk->sk_bound_dev_if &&
1292 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1293 treq->iif = inet6_iif(skb);
1296 isn = tcp_v6_init_sequence(skb);
1298 tcp_rsk(req)->snt_isn = isn;
1300 security_inet_conn_request(sk, skb, req);
1302 if (tcp_v6_send_synack(sk, req, NULL))
1305 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1312 return 0; /* don't send reset */
1315 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1316 struct request_sock *req,
1317 struct dst_entry *dst)
1319 struct inet6_request_sock *treq = inet6_rsk(req);
1320 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1321 struct tcp6_sock *newtcp6sk;
1322 struct inet_sock *newinet;
1323 struct tcp_sock *newtp;
1325 struct ipv6_txoptions *opt;
1326 #ifdef CONFIG_TCP_MD5SIG
1327 struct tcp_md5sig_key *key;
1330 if (skb->protocol == htons(ETH_P_IP)) {
1335 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1340 newtcp6sk = (struct tcp6_sock *)newsk;
1341 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1343 newinet = inet_sk(newsk);
1344 newnp = inet6_sk(newsk);
1345 newtp = tcp_sk(newsk);
1347 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1349 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1352 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1355 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1357 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1358 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1359 #ifdef CONFIG_TCP_MD5SIG
1360 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1363 newnp->pktoptions = NULL;
1365 newnp->mcast_oif = inet6_iif(skb);
1366 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1369 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1370 * here, tcp_create_openreq_child now does this for us, see the comment in
1371 * that function for the gory details. -acme
1374 /* It is tricky place. Until this moment IPv4 tcp
1375 worked with IPv6 icsk.icsk_af_ops.
1378 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1385 if (sk_acceptq_is_full(sk))
1388 if (np->rxopt.bits.osrcrt == 2 &&
1389 opt == NULL && treq->pktopts) {
1390 struct inet6_skb_parm *rxopt = IP6CB(treq->pktopts);
1392 opt = ipv6_invert_rthdr(sk,
1393 (struct ipv6_rt_hdr *)(skb_network_header(treq->pktopts) +
1398 struct in6_addr *final_p = NULL, final;
1401 memset(&fl, 0, sizeof(fl));
1402 fl.proto = IPPROTO_TCP;
1403 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1404 if (opt && opt->srcrt) {
1405 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1406 ipv6_addr_copy(&final, &fl.fl6_dst);
1407 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1410 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1411 fl.oif = sk->sk_bound_dev_if;
1412 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1413 fl.fl_ip_sport = inet_sk(sk)->sport;
1414 security_req_classify_flow(req, &fl);
1416 if (ip6_dst_lookup(sk, &dst, &fl))
1420 ipv6_addr_copy(&fl.fl6_dst, final_p);
1422 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1426 newsk = tcp_create_openreq_child(sk, req, skb);
1431 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1432 * count here, tcp_create_openreq_child now does this for us, see the
1433 * comment in that function for the gory details. -acme
1436 newsk->sk_gso_type = SKB_GSO_TCPV6;
1437 __ip6_dst_store(newsk, dst, NULL, NULL);
1439 newtcp6sk = (struct tcp6_sock *)newsk;
1440 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1442 newtp = tcp_sk(newsk);
1443 newinet = inet_sk(newsk);
1444 newnp = inet6_sk(newsk);
1446 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1448 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1449 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1450 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1451 newsk->sk_bound_dev_if = treq->iif;
1453 /* Now IPv6 options...
1455 First: no IPv4 options.
1457 newinet->opt = NULL;
1458 newnp->ipv6_fl_list = NULL;
1461 newnp->rxopt.all = np->rxopt.all;
1463 /* Clone pktoptions received with SYN */
1464 newnp->pktoptions = NULL;
1465 if (treq->pktopts != NULL) {
1466 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1467 kfree_skb(treq->pktopts);
1468 treq->pktopts = NULL;
1469 if (newnp->pktoptions)
1470 skb_set_owner_r(newnp->pktoptions, newsk);
1473 newnp->mcast_oif = inet6_iif(skb);
1474 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1476 /* Clone native IPv6 options from listening socket (if any)
1478 Yes, keeping reference count would be much more clever,
1479 but we make one more one thing there: reattach optmem
1483 newnp->opt = ipv6_dup_options(newsk, opt);
1485 sock_kfree_s(sk, opt, opt->tot_len);
1488 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1490 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1491 newnp->opt->opt_flen);
1493 tcp_mtup_init(newsk);
1494 tcp_sync_mss(newsk, dst_mtu(dst));
1495 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1496 tcp_initialize_rcv_mss(newsk);
1498 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1500 #ifdef CONFIG_TCP_MD5SIG
1501 /* Copy over the MD5 key from the original socket */
1502 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1503 /* We're using one, so create a matching key
1504 * on the newsk structure. If we fail to get
1505 * memory, then we end up not copying the key
1508 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1510 tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1511 newkey, key->keylen);
1515 __inet6_hash(&tcp_hashinfo, newsk);
1516 inet_inherit_port(&tcp_hashinfo, sk, newsk);
1521 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1523 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1524 if (opt && opt != np->opt)
1525 sock_kfree_s(sk, opt, opt->tot_len);
1530 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1532 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1533 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1534 &ipv6_hdr(skb)->daddr, skb->csum)) {
1535 skb->ip_summed = CHECKSUM_UNNECESSARY;
1540 skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1541 &ipv6_hdr(skb)->saddr,
1542 &ipv6_hdr(skb)->daddr, 0));
1544 if (skb->len <= 76) {
1545 return __skb_checksum_complete(skb);
1550 /* The socket must have it's spinlock held when we get
1553 * We have a potential double-lock case here, so even when
1554 * doing backlog processing we use the BH locking scheme.
1555 * This is because we cannot sleep with the original spinlock
1558 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1560 struct ipv6_pinfo *np = inet6_sk(sk);
1561 struct tcp_sock *tp;
1562 struct sk_buff *opt_skb = NULL;
1564 /* Imagine: socket is IPv6. IPv4 packet arrives,
1565 goes to IPv4 receive handler and backlogged.
1566 From backlog it always goes here. Kerboom...
1567 Fortunately, tcp_rcv_established and rcv_established
1568 handle them correctly, but it is not case with
1569 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1572 if (skb->protocol == htons(ETH_P_IP))
1573 return tcp_v4_do_rcv(sk, skb);
1575 #ifdef CONFIG_TCP_MD5SIG
1576 if (tcp_v6_inbound_md5_hash (sk, skb))
1580 if (sk_filter(sk, skb))
1584 * socket locking is here for SMP purposes as backlog rcv
1585 * is currently called with bh processing disabled.
1588 /* Do Stevens' IPV6_PKTOPTIONS.
1590 Yes, guys, it is the only place in our code, where we
1591 may make it not affecting IPv4.
1592 The rest of code is protocol independent,
1593 and I do not like idea to uglify IPv4.
1595 Actually, all the idea behind IPV6_PKTOPTIONS
1596 looks not very well thought. For now we latch
1597 options, received in the last packet, enqueued
1598 by tcp. Feel free to propose better solution.
1602 opt_skb = skb_clone(skb, GFP_ATOMIC);
1604 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1605 TCP_CHECK_TIMER(sk);
1606 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1608 TCP_CHECK_TIMER(sk);
1610 goto ipv6_pktoptions;
1614 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1617 if (sk->sk_state == TCP_LISTEN) {
1618 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1623 * Queue it on the new socket if the new socket is active,
1624 * otherwise we just shortcircuit this and continue with
1628 if (tcp_child_process(sk, nsk, skb))
1631 __kfree_skb(opt_skb);
1636 TCP_CHECK_TIMER(sk);
1637 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1639 TCP_CHECK_TIMER(sk);
1641 goto ipv6_pktoptions;
1645 tcp_v6_send_reset(sk, skb);
1648 __kfree_skb(opt_skb);
1652 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1657 /* Do you ask, what is it?
1659 1. skb was enqueued by tcp.
1660 2. skb is added to tail of read queue, rather than out of order.
1661 3. socket is not in passive state.
1662 4. Finally, it really contains options, which user wants to receive.
1665 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1666 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1667 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1668 np->mcast_oif = inet6_iif(opt_skb);
1669 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1670 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1671 if (ipv6_opt_accepted(sk, opt_skb)) {
1672 skb_set_owner_r(opt_skb, sk);
1673 opt_skb = xchg(&np->pktoptions, opt_skb);
1675 __kfree_skb(opt_skb);
1676 opt_skb = xchg(&np->pktoptions, NULL);
1685 static int tcp_v6_rcv(struct sk_buff **pskb)
1687 struct sk_buff *skb = *pskb;
1692 if (skb->pkt_type != PACKET_HOST)
1696 * Count it even if it's bad.
1698 TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1700 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1705 if (th->doff < sizeof(struct tcphdr)/4)
1707 if (!pskb_may_pull(skb, th->doff*4))
1710 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1714 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1715 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1716 skb->len - th->doff*4);
1717 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1718 TCP_SKB_CB(skb)->when = 0;
1719 TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1720 TCP_SKB_CB(skb)->sacked = 0;
1722 sk = __inet6_lookup(&tcp_hashinfo, &ipv6_hdr(skb)->saddr, th->source,
1723 &ipv6_hdr(skb)->daddr, ntohs(th->dest),
1730 if (sk->sk_state == TCP_TIME_WAIT)
1733 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1734 goto discard_and_relse;
1736 if (sk_filter(sk, skb))
1737 goto discard_and_relse;
1741 bh_lock_sock_nested(sk);
1743 if (!sock_owned_by_user(sk)) {
1744 #ifdef CONFIG_NET_DMA
1745 struct tcp_sock *tp = tcp_sk(sk);
1746 if (tp->ucopy.dma_chan)
1747 ret = tcp_v6_do_rcv(sk, skb);
1751 if (!tcp_prequeue(sk, skb))
1752 ret = tcp_v6_do_rcv(sk, skb);
1755 sk_add_backlog(sk, skb);
1759 return ret ? -1 : 0;
1762 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1765 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1767 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1769 tcp_v6_send_reset(NULL, skb);
1786 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1787 inet_twsk_put(inet_twsk(sk));
1791 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1792 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1793 inet_twsk_put(inet_twsk(sk));
1797 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1802 sk2 = inet6_lookup_listener(&tcp_hashinfo,
1803 &ipv6_hdr(skb)->daddr,
1804 ntohs(th->dest), inet6_iif(skb));
1806 struct inet_timewait_sock *tw = inet_twsk(sk);
1807 inet_twsk_deschedule(tw, &tcp_death_row);
1812 /* Fall through to ACK */
1815 tcp_v6_timewait_ack(sk, skb);
1819 case TCP_TW_SUCCESS:;
1824 static int tcp_v6_remember_stamp(struct sock *sk)
1826 /* Alas, not yet... */
1830 static struct inet_connection_sock_af_ops ipv6_specific = {
1831 .queue_xmit = inet6_csk_xmit,
1832 .send_check = tcp_v6_send_check,
1833 .rebuild_header = inet6_sk_rebuild_header,
1834 .conn_request = tcp_v6_conn_request,
1835 .syn_recv_sock = tcp_v6_syn_recv_sock,
1836 .remember_stamp = tcp_v6_remember_stamp,
1837 .net_header_len = sizeof(struct ipv6hdr),
1838 .setsockopt = ipv6_setsockopt,
1839 .getsockopt = ipv6_getsockopt,
1840 .addr2sockaddr = inet6_csk_addr2sockaddr,
1841 .sockaddr_len = sizeof(struct sockaddr_in6),
1842 #ifdef CONFIG_COMPAT
1843 .compat_setsockopt = compat_ipv6_setsockopt,
1844 .compat_getsockopt = compat_ipv6_getsockopt,
1848 #ifdef CONFIG_TCP_MD5SIG
1849 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1850 .md5_lookup = tcp_v6_md5_lookup,
1851 .calc_md5_hash = tcp_v6_calc_md5_hash,
1852 .md5_add = tcp_v6_md5_add_func,
1853 .md5_parse = tcp_v6_parse_md5_keys,
1858 * TCP over IPv4 via INET6 API
1861 static struct inet_connection_sock_af_ops ipv6_mapped = {
1862 .queue_xmit = ip_queue_xmit,
1863 .send_check = tcp_v4_send_check,
1864 .rebuild_header = inet_sk_rebuild_header,
1865 .conn_request = tcp_v6_conn_request,
1866 .syn_recv_sock = tcp_v6_syn_recv_sock,
1867 .remember_stamp = tcp_v4_remember_stamp,
1868 .net_header_len = sizeof(struct iphdr),
1869 .setsockopt = ipv6_setsockopt,
1870 .getsockopt = ipv6_getsockopt,
1871 .addr2sockaddr = inet6_csk_addr2sockaddr,
1872 .sockaddr_len = sizeof(struct sockaddr_in6),
1873 #ifdef CONFIG_COMPAT
1874 .compat_setsockopt = compat_ipv6_setsockopt,
1875 .compat_getsockopt = compat_ipv6_getsockopt,
1879 #ifdef CONFIG_TCP_MD5SIG
1880 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1881 .md5_lookup = tcp_v4_md5_lookup,
1882 .calc_md5_hash = tcp_v4_calc_md5_hash,
1883 .md5_add = tcp_v6_md5_add_func,
1884 .md5_parse = tcp_v6_parse_md5_keys,
1888 /* NOTE: A lot of things set to zero explicitly by call to
1889 * sk_alloc() so need not be done here.
1891 static int tcp_v6_init_sock(struct sock *sk)
1893 struct inet_connection_sock *icsk = inet_csk(sk);
1894 struct tcp_sock *tp = tcp_sk(sk);
1896 skb_queue_head_init(&tp->out_of_order_queue);
1897 tcp_init_xmit_timers(sk);
1898 tcp_prequeue_init(tp);
1900 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1901 tp->mdev = TCP_TIMEOUT_INIT;
1903 /* So many TCP implementations out there (incorrectly) count the
1904 * initial SYN frame in their delayed-ACK and congestion control
1905 * algorithms that we must have the following bandaid to talk
1906 * efficiently to them. -DaveM
1910 /* See draft-stevens-tcpca-spec-01 for discussion of the
1911 * initialization of these values.
1913 tp->snd_ssthresh = 0x7fffffff;
1914 tp->snd_cwnd_clamp = ~0;
1915 tp->mss_cache = 536;
1917 tp->reordering = sysctl_tcp_reordering;
1919 sk->sk_state = TCP_CLOSE;
1921 icsk->icsk_af_ops = &ipv6_specific;
1922 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1923 icsk->icsk_sync_mss = tcp_sync_mss;
1924 sk->sk_write_space = sk_stream_write_space;
1925 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1927 #ifdef CONFIG_TCP_MD5SIG
1928 tp->af_specific = &tcp_sock_ipv6_specific;
1931 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1932 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1934 atomic_inc(&tcp_sockets_allocated);
1939 static int tcp_v6_destroy_sock(struct sock *sk)
1941 #ifdef CONFIG_TCP_MD5SIG
1942 /* Clean up the MD5 key list */
1943 if (tcp_sk(sk)->md5sig_info)
1944 tcp_v6_clear_md5_list(sk);
1946 tcp_v4_destroy_sock(sk);
1947 return inet6_destroy_sock(sk);
1950 #ifdef CONFIG_PROC_FS
1951 /* Proc filesystem TCPv6 sock list dumping. */
1952 static void get_openreq6(struct seq_file *seq,
1953 struct sock *sk, struct request_sock *req, int i, int uid)
1955 int ttd = req->expires - jiffies;
1956 struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1957 struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1963 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1964 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1966 src->s6_addr32[0], src->s6_addr32[1],
1967 src->s6_addr32[2], src->s6_addr32[3],
1968 ntohs(inet_sk(sk)->sport),
1969 dest->s6_addr32[0], dest->s6_addr32[1],
1970 dest->s6_addr32[2], dest->s6_addr32[3],
1971 ntohs(inet_rsk(req)->rmt_port),
1973 0,0, /* could print option size, but that is af dependent. */
1974 1, /* timers active (only the expire timer) */
1975 jiffies_to_clock_t(ttd),
1978 0, /* non standard timer */
1979 0, /* open_requests have no inode */
1983 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1985 struct in6_addr *dest, *src;
1988 unsigned long timer_expires;
1989 struct inet_sock *inet = inet_sk(sp);
1990 struct tcp_sock *tp = tcp_sk(sp);
1991 const struct inet_connection_sock *icsk = inet_csk(sp);
1992 struct ipv6_pinfo *np = inet6_sk(sp);
1995 src = &np->rcv_saddr;
1996 destp = ntohs(inet->dport);
1997 srcp = ntohs(inet->sport);
1999 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2001 timer_expires = icsk->icsk_timeout;
2002 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2004 timer_expires = icsk->icsk_timeout;
2005 } else if (timer_pending(&sp->sk_timer)) {
2007 timer_expires = sp->sk_timer.expires;
2010 timer_expires = jiffies;
2014 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2015 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
2017 src->s6_addr32[0], src->s6_addr32[1],
2018 src->s6_addr32[2], src->s6_addr32[3], srcp,
2019 dest->s6_addr32[0], dest->s6_addr32[1],
2020 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2022 tp->write_seq-tp->snd_una,
2023 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2025 jiffies_to_clock_t(timer_expires - jiffies),
2026 icsk->icsk_retransmits,
2028 icsk->icsk_probes_out,
2030 atomic_read(&sp->sk_refcnt), sp,
2033 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2034 tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2038 static void get_timewait6_sock(struct seq_file *seq,
2039 struct inet_timewait_sock *tw, int i)
2041 struct in6_addr *dest, *src;
2043 struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2044 int ttd = tw->tw_ttd - jiffies;
2049 dest = &tw6->tw_v6_daddr;
2050 src = &tw6->tw_v6_rcv_saddr;
2051 destp = ntohs(tw->tw_dport);
2052 srcp = ntohs(tw->tw_sport);
2055 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2056 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2058 src->s6_addr32[0], src->s6_addr32[1],
2059 src->s6_addr32[2], src->s6_addr32[3], srcp,
2060 dest->s6_addr32[0], dest->s6_addr32[1],
2061 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2062 tw->tw_substate, 0, 0,
2063 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2064 atomic_read(&tw->tw_refcnt), tw);
2067 static int tcp6_seq_show(struct seq_file *seq, void *v)
2069 struct tcp_iter_state *st;
2071 if (v == SEQ_START_TOKEN) {
2076 "st tx_queue rx_queue tr tm->when retrnsmt"
2077 " uid timeout inode\n");
2082 switch (st->state) {
2083 case TCP_SEQ_STATE_LISTENING:
2084 case TCP_SEQ_STATE_ESTABLISHED:
2085 get_tcp6_sock(seq, v, st->num);
2087 case TCP_SEQ_STATE_OPENREQ:
2088 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2090 case TCP_SEQ_STATE_TIME_WAIT:
2091 get_timewait6_sock(seq, v, st->num);
2098 static struct file_operations tcp6_seq_fops;
2099 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2100 .owner = THIS_MODULE,
2103 .seq_show = tcp6_seq_show,
2104 .seq_fops = &tcp6_seq_fops,
2107 int __init tcp6_proc_init(void)
2109 return tcp_proc_register(&tcp6_seq_afinfo);
2112 void tcp6_proc_exit(void)
2114 tcp_proc_unregister(&tcp6_seq_afinfo);
2118 struct proto tcpv6_prot = {
2120 .owner = THIS_MODULE,
2122 .connect = tcp_v6_connect,
2123 .disconnect = tcp_disconnect,
2124 .accept = inet_csk_accept,
2126 .init = tcp_v6_init_sock,
2127 .destroy = tcp_v6_destroy_sock,
2128 .shutdown = tcp_shutdown,
2129 .setsockopt = tcp_setsockopt,
2130 .getsockopt = tcp_getsockopt,
2131 .sendmsg = tcp_sendmsg,
2132 .recvmsg = tcp_recvmsg,
2133 .backlog_rcv = tcp_v6_do_rcv,
2134 .hash = tcp_v6_hash,
2135 .unhash = tcp_unhash,
2136 .get_port = tcp_v6_get_port,
2137 .enter_memory_pressure = tcp_enter_memory_pressure,
2138 .sockets_allocated = &tcp_sockets_allocated,
2139 .memory_allocated = &tcp_memory_allocated,
2140 .memory_pressure = &tcp_memory_pressure,
2141 .orphan_count = &tcp_orphan_count,
2142 .sysctl_mem = sysctl_tcp_mem,
2143 .sysctl_wmem = sysctl_tcp_wmem,
2144 .sysctl_rmem = sysctl_tcp_rmem,
2145 .max_header = MAX_TCP_HEADER,
2146 .obj_size = sizeof(struct tcp6_sock),
2147 .twsk_prot = &tcp6_timewait_sock_ops,
2148 .rsk_prot = &tcp6_request_sock_ops,
2149 #ifdef CONFIG_COMPAT
2150 .compat_setsockopt = compat_tcp_setsockopt,
2151 .compat_getsockopt = compat_tcp_getsockopt,
2155 static struct inet6_protocol tcpv6_protocol = {
2156 .handler = tcp_v6_rcv,
2157 .err_handler = tcp_v6_err,
2158 .gso_send_check = tcp_v6_gso_send_check,
2159 .gso_segment = tcp_tso_segment,
2160 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2163 static struct inet_protosw tcpv6_protosw = {
2164 .type = SOCK_STREAM,
2165 .protocol = IPPROTO_TCP,
2166 .prot = &tcpv6_prot,
2167 .ops = &inet6_stream_ops,
2170 .flags = INET_PROTOSW_PERMANENT |
2174 void __init tcpv6_init(void)
2176 /* register inet6 protocol */
2177 if (inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP) < 0)
2178 printk(KERN_ERR "tcpv6_init: Could not register protocol\n");
2179 inet6_register_protosw(&tcpv6_protosw);
2181 if (inet_csk_ctl_sock_create(&tcp6_socket, PF_INET6, SOCK_RAW,
2183 panic("Failed to create the TCPv6 control socket.\n");