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>
62 #include <net/netdma.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 void tcp_v6_hash(struct sock *sk)
91 if (sk->sk_state != TCP_CLOSE) {
92 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
102 static __inline__ __sum16 tcp_v6_check(struct tcphdr *th, int len,
103 struct in6_addr *saddr,
104 struct in6_addr *daddr,
107 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
110 static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
112 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
113 ipv6_hdr(skb)->saddr.s6_addr32,
115 tcp_hdr(skb)->source);
118 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
121 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
122 struct inet_sock *inet = inet_sk(sk);
123 struct inet_connection_sock *icsk = inet_csk(sk);
124 struct ipv6_pinfo *np = inet6_sk(sk);
125 struct tcp_sock *tp = tcp_sk(sk);
126 struct in6_addr *saddr = NULL, *final_p = NULL, final;
128 struct dst_entry *dst;
132 if (addr_len < SIN6_LEN_RFC2133)
135 if (usin->sin6_family != AF_INET6)
136 return(-EAFNOSUPPORT);
138 memset(&fl, 0, sizeof(fl));
141 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
142 IP6_ECN_flow_init(fl.fl6_flowlabel);
143 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
144 struct ip6_flowlabel *flowlabel;
145 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
146 if (flowlabel == NULL)
148 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
149 fl6_sock_release(flowlabel);
154 * connect() to INADDR_ANY means loopback (BSD'ism).
157 if(ipv6_addr_any(&usin->sin6_addr))
158 usin->sin6_addr.s6_addr[15] = 0x1;
160 addr_type = ipv6_addr_type(&usin->sin6_addr);
162 if(addr_type & IPV6_ADDR_MULTICAST)
165 if (addr_type&IPV6_ADDR_LINKLOCAL) {
166 if (addr_len >= sizeof(struct sockaddr_in6) &&
167 usin->sin6_scope_id) {
168 /* If interface is set while binding, indices
171 if (sk->sk_bound_dev_if &&
172 sk->sk_bound_dev_if != usin->sin6_scope_id)
175 sk->sk_bound_dev_if = usin->sin6_scope_id;
178 /* Connect to link-local address requires an interface */
179 if (!sk->sk_bound_dev_if)
183 if (tp->rx_opt.ts_recent_stamp &&
184 !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
185 tp->rx_opt.ts_recent = 0;
186 tp->rx_opt.ts_recent_stamp = 0;
190 ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
191 np->flow_label = fl.fl6_flowlabel;
197 if (addr_type == IPV6_ADDR_MAPPED) {
198 u32 exthdrlen = icsk->icsk_ext_hdr_len;
199 struct sockaddr_in sin;
201 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
203 if (__ipv6_only_sock(sk))
206 sin.sin_family = AF_INET;
207 sin.sin_port = usin->sin6_port;
208 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
210 icsk->icsk_af_ops = &ipv6_mapped;
211 sk->sk_backlog_rcv = tcp_v4_do_rcv;
212 #ifdef CONFIG_TCP_MD5SIG
213 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
216 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
219 icsk->icsk_ext_hdr_len = exthdrlen;
220 icsk->icsk_af_ops = &ipv6_specific;
221 sk->sk_backlog_rcv = tcp_v6_do_rcv;
222 #ifdef CONFIG_TCP_MD5SIG
223 tp->af_specific = &tcp_sock_ipv6_specific;
227 ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
229 ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
236 if (!ipv6_addr_any(&np->rcv_saddr))
237 saddr = &np->rcv_saddr;
239 fl.proto = IPPROTO_TCP;
240 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
241 ipv6_addr_copy(&fl.fl6_src,
242 (saddr ? saddr : &np->saddr));
243 fl.oif = sk->sk_bound_dev_if;
244 fl.fl_ip_dport = usin->sin6_port;
245 fl.fl_ip_sport = inet->sport;
247 if (np->opt && np->opt->srcrt) {
248 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
249 ipv6_addr_copy(&final, &fl.fl6_dst);
250 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
254 security_sk_classify_flow(sk, &fl);
256 err = ip6_dst_lookup(sk, &dst, &fl);
260 ipv6_addr_copy(&fl.fl6_dst, final_p);
262 if ((err = __xfrm_lookup(&dst, &fl, sk, XFRM_LOOKUP_WAIT)) < 0) {
264 err = ip6_dst_blackhole(sk, &dst, &fl);
271 ipv6_addr_copy(&np->rcv_saddr, saddr);
274 /* set the source address */
275 ipv6_addr_copy(&np->saddr, saddr);
276 inet->rcv_saddr = LOOPBACK4_IPV6;
278 sk->sk_gso_type = SKB_GSO_TCPV6;
279 __ip6_dst_store(sk, dst, NULL, NULL);
281 icsk->icsk_ext_hdr_len = 0;
283 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
286 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
288 inet->dport = usin->sin6_port;
290 tcp_set_state(sk, TCP_SYN_SENT);
291 err = inet6_hash_connect(&tcp_death_row, sk);
296 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
301 err = tcp_connect(sk);
308 tcp_set_state(sk, TCP_CLOSE);
312 sk->sk_route_caps = 0;
316 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
317 int type, int code, int offset, __be32 info)
319 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
320 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
321 struct ipv6_pinfo *np;
327 sk = inet6_lookup(skb->dev->nd_net, &tcp_hashinfo, &hdr->daddr,
328 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
331 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
335 if (sk->sk_state == TCP_TIME_WAIT) {
336 inet_twsk_put(inet_twsk(sk));
341 if (sock_owned_by_user(sk))
342 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
344 if (sk->sk_state == TCP_CLOSE)
348 seq = ntohl(th->seq);
349 if (sk->sk_state != TCP_LISTEN &&
350 !between(seq, tp->snd_una, tp->snd_nxt)) {
351 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
357 if (type == ICMPV6_PKT_TOOBIG) {
358 struct dst_entry *dst = NULL;
360 if (sock_owned_by_user(sk))
362 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
365 /* icmp should have updated the destination cache entry */
366 dst = __sk_dst_check(sk, np->dst_cookie);
369 struct inet_sock *inet = inet_sk(sk);
372 /* BUGGG_FUTURE: Again, it is not clear how
373 to handle rthdr case. Ignore this complexity
376 memset(&fl, 0, sizeof(fl));
377 fl.proto = IPPROTO_TCP;
378 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
379 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
380 fl.oif = sk->sk_bound_dev_if;
381 fl.fl_ip_dport = inet->dport;
382 fl.fl_ip_sport = inet->sport;
383 security_skb_classify_flow(skb, &fl);
385 if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
386 sk->sk_err_soft = -err;
390 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
391 sk->sk_err_soft = -err;
398 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
399 tcp_sync_mss(sk, dst_mtu(dst));
400 tcp_simple_retransmit(sk);
401 } /* else let the usual retransmit timer handle it */
406 icmpv6_err_convert(type, code, &err);
408 /* Might be for an request_sock */
409 switch (sk->sk_state) {
410 struct request_sock *req, **prev;
412 if (sock_owned_by_user(sk))
415 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
416 &hdr->saddr, inet6_iif(skb));
420 /* ICMPs are not backlogged, hence we cannot get
421 * an established socket here.
423 BUG_TRAP(req->sk == NULL);
425 if (seq != tcp_rsk(req)->snt_isn) {
426 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
430 inet_csk_reqsk_queue_drop(sk, req, prev);
434 case TCP_SYN_RECV: /* Cannot happen.
435 It can, it SYNs are crossed. --ANK */
436 if (!sock_owned_by_user(sk)) {
438 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
442 sk->sk_err_soft = err;
446 if (!sock_owned_by_user(sk) && np->recverr) {
448 sk->sk_error_report(sk);
450 sk->sk_err_soft = err;
458 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req)
460 struct inet6_request_sock *treq = inet6_rsk(req);
461 struct ipv6_pinfo *np = inet6_sk(sk);
462 struct sk_buff * skb;
463 struct ipv6_txoptions *opt = NULL;
464 struct in6_addr * final_p = NULL, final;
466 struct dst_entry *dst;
469 memset(&fl, 0, sizeof(fl));
470 fl.proto = IPPROTO_TCP;
471 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
472 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
473 fl.fl6_flowlabel = 0;
475 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
476 fl.fl_ip_sport = inet_sk(sk)->sport;
477 security_req_classify_flow(req, &fl);
480 if (opt && opt->srcrt) {
481 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
482 ipv6_addr_copy(&final, &fl.fl6_dst);
483 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
487 err = ip6_dst_lookup(sk, &dst, &fl);
491 ipv6_addr_copy(&fl.fl6_dst, final_p);
492 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
495 skb = tcp_make_synack(sk, dst, req);
497 struct tcphdr *th = tcp_hdr(skb);
499 th->check = tcp_v6_check(th, skb->len,
500 &treq->loc_addr, &treq->rmt_addr,
501 csum_partial((char *)th, skb->len, skb->csum));
503 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
504 err = ip6_xmit(sk, skb, &fl, opt, 0);
505 err = net_xmit_eval(err);
509 if (opt && opt != np->opt)
510 sock_kfree_s(sk, opt, opt->tot_len);
515 static inline void syn_flood_warning(struct sk_buff *skb)
517 #ifdef CONFIG_SYN_COOKIES
518 if (sysctl_tcp_syncookies)
520 "TCPv6: Possible SYN flooding on port %d. "
521 "Sending cookies.\n", ntohs(tcp_hdr(skb)->dest));
525 "TCPv6: Possible SYN flooding on port %d. "
526 "Dropping request.\n", ntohs(tcp_hdr(skb)->dest));
529 static void tcp_v6_reqsk_destructor(struct request_sock *req)
531 if (inet6_rsk(req)->pktopts)
532 kfree_skb(inet6_rsk(req)->pktopts);
535 #ifdef CONFIG_TCP_MD5SIG
536 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
537 struct in6_addr *addr)
539 struct tcp_sock *tp = tcp_sk(sk);
544 if (!tp->md5sig_info || !tp->md5sig_info->entries6)
547 for (i = 0; i < tp->md5sig_info->entries6; i++) {
548 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, addr) == 0)
549 return &tp->md5sig_info->keys6[i].base;
554 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
555 struct sock *addr_sk)
557 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
560 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
561 struct request_sock *req)
563 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
566 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
567 char *newkey, u8 newkeylen)
569 /* Add key to the list */
570 struct tcp_md5sig_key *key;
571 struct tcp_sock *tp = tcp_sk(sk);
572 struct tcp6_md5sig_key *keys;
574 key = tcp_v6_md5_do_lookup(sk, peer);
576 /* modify existing entry - just update that one */
579 key->keylen = newkeylen;
581 /* reallocate new list if current one is full. */
582 if (!tp->md5sig_info) {
583 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
584 if (!tp->md5sig_info) {
588 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
590 if (tcp_alloc_md5sig_pool() == NULL) {
594 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
595 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
596 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
599 tcp_free_md5sig_pool();
604 if (tp->md5sig_info->entries6)
605 memmove(keys, tp->md5sig_info->keys6,
606 (sizeof (tp->md5sig_info->keys6[0]) *
607 tp->md5sig_info->entries6));
609 kfree(tp->md5sig_info->keys6);
610 tp->md5sig_info->keys6 = keys;
611 tp->md5sig_info->alloced6++;
614 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
616 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
617 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
619 tp->md5sig_info->entries6++;
624 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
625 u8 *newkey, __u8 newkeylen)
627 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
631 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
633 struct tcp_sock *tp = tcp_sk(sk);
636 for (i = 0; i < tp->md5sig_info->entries6; i++) {
637 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, peer) == 0) {
639 kfree(tp->md5sig_info->keys6[i].base.key);
640 tp->md5sig_info->entries6--;
642 if (tp->md5sig_info->entries6 == 0) {
643 kfree(tp->md5sig_info->keys6);
644 tp->md5sig_info->keys6 = NULL;
645 tp->md5sig_info->alloced6 = 0;
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]));
654 tcp_free_md5sig_pool();
661 static void tcp_v6_clear_md5_list (struct sock *sk)
663 struct tcp_sock *tp = tcp_sk(sk);
666 if (tp->md5sig_info->entries6) {
667 for (i = 0; i < tp->md5sig_info->entries6; i++)
668 kfree(tp->md5sig_info->keys6[i].base.key);
669 tp->md5sig_info->entries6 = 0;
670 tcp_free_md5sig_pool();
673 kfree(tp->md5sig_info->keys6);
674 tp->md5sig_info->keys6 = NULL;
675 tp->md5sig_info->alloced6 = 0;
677 if (tp->md5sig_info->entries4) {
678 for (i = 0; i < tp->md5sig_info->entries4; i++)
679 kfree(tp->md5sig_info->keys4[i].base.key);
680 tp->md5sig_info->entries4 = 0;
681 tcp_free_md5sig_pool();
684 kfree(tp->md5sig_info->keys4);
685 tp->md5sig_info->keys4 = NULL;
686 tp->md5sig_info->alloced4 = 0;
689 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
692 struct tcp_md5sig cmd;
693 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
696 if (optlen < sizeof(cmd))
699 if (copy_from_user(&cmd, optval, sizeof(cmd)))
702 if (sin6->sin6_family != AF_INET6)
705 if (!cmd.tcpm_keylen) {
706 if (!tcp_sk(sk)->md5sig_info)
708 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
709 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
710 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
713 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
716 if (!tcp_sk(sk)->md5sig_info) {
717 struct tcp_sock *tp = tcp_sk(sk);
718 struct tcp_md5sig_info *p;
720 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
725 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
728 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
731 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
732 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
733 newkey, cmd.tcpm_keylen);
735 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
738 static int tcp_v6_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
739 struct in6_addr *saddr,
740 struct in6_addr *daddr,
741 struct tcphdr *th, int protocol,
744 struct scatterlist sg[4];
748 struct tcp_md5sig_pool *hp;
749 struct tcp6_pseudohdr *bp;
750 struct hash_desc *desc;
752 unsigned int nbytes = 0;
754 hp = tcp_get_md5sig_pool();
756 printk(KERN_WARNING "%s(): hash pool not found...\n", __func__);
757 goto clear_hash_noput;
759 bp = &hp->md5_blk.ip6;
760 desc = &hp->md5_desc;
762 /* 1. TCP pseudo-header (RFC2460) */
763 ipv6_addr_copy(&bp->saddr, saddr);
764 ipv6_addr_copy(&bp->daddr, daddr);
765 bp->len = htonl(tcplen);
766 bp->protocol = htonl(protocol);
768 sg_init_table(sg, 4);
770 sg_set_buf(&sg[block++], bp, sizeof(*bp));
771 nbytes += sizeof(*bp);
773 /* 2. TCP header, excluding options */
776 sg_set_buf(&sg[block++], th, sizeof(*th));
777 nbytes += sizeof(*th);
779 /* 3. TCP segment data (if any) */
780 data_len = tcplen - (th->doff << 2);
782 u8 *data = (u8 *)th + (th->doff << 2);
783 sg_set_buf(&sg[block++], data, data_len);
788 sg_set_buf(&sg[block++], key->key, key->keylen);
789 nbytes += key->keylen;
791 sg_mark_end(&sg[block - 1]);
793 /* Now store the hash into the packet */
794 err = crypto_hash_init(desc);
796 printk(KERN_WARNING "%s(): hash_init failed\n", __func__);
799 err = crypto_hash_update(desc, sg, nbytes);
801 printk(KERN_WARNING "%s(): hash_update failed\n", __func__);
804 err = crypto_hash_final(desc, md5_hash);
806 printk(KERN_WARNING "%s(): hash_final failed\n", __func__);
810 /* Reset header, and free up the crypto */
811 tcp_put_md5sig_pool();
816 tcp_put_md5sig_pool();
818 memset(md5_hash, 0, 16);
822 static int tcp_v6_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
824 struct dst_entry *dst,
825 struct request_sock *req,
826 struct tcphdr *th, int protocol,
829 struct in6_addr *saddr, *daddr;
832 saddr = &inet6_sk(sk)->saddr;
833 daddr = &inet6_sk(sk)->daddr;
835 saddr = &inet6_rsk(req)->loc_addr;
836 daddr = &inet6_rsk(req)->rmt_addr;
838 return tcp_v6_do_calc_md5_hash(md5_hash, key,
840 th, protocol, tcplen);
843 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
845 __u8 *hash_location = NULL;
846 struct tcp_md5sig_key *hash_expected;
847 struct ipv6hdr *ip6h = ipv6_hdr(skb);
848 struct tcphdr *th = tcp_hdr(skb);
849 int length = (th->doff << 2) - sizeof (*th);
854 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
856 /* If the TCP option is too short, we can short cut */
857 if (length < TCPOLEN_MD5SIG)
858 return hash_expected ? 1 : 0;
874 if (opsize < 2 || opsize > length)
876 if (opcode == TCPOPT_MD5SIG) {
886 /* do we have a hash as expected? */
887 if (!hash_expected) {
890 if (net_ratelimit()) {
891 printk(KERN_INFO "MD5 Hash NOT expected but found "
892 "(" NIP6_FMT ", %u)->"
893 "(" NIP6_FMT ", %u)\n",
894 NIP6(ip6h->saddr), ntohs(th->source),
895 NIP6(ip6h->daddr), ntohs(th->dest));
900 if (!hash_location) {
901 if (net_ratelimit()) {
902 printk(KERN_INFO "MD5 Hash expected but NOT found "
903 "(" NIP6_FMT ", %u)->"
904 "(" NIP6_FMT ", %u)\n",
905 NIP6(ip6h->saddr), ntohs(th->source),
906 NIP6(ip6h->daddr), ntohs(th->dest));
911 /* check the signature */
912 genhash = tcp_v6_do_calc_md5_hash(newhash,
914 &ip6h->saddr, &ip6h->daddr,
917 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
918 if (net_ratelimit()) {
919 printk(KERN_INFO "MD5 Hash %s for "
920 "(" NIP6_FMT ", %u)->"
921 "(" NIP6_FMT ", %u)\n",
922 genhash ? "failed" : "mismatch",
923 NIP6(ip6h->saddr), ntohs(th->source),
924 NIP6(ip6h->daddr), ntohs(th->dest));
932 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
934 .obj_size = sizeof(struct tcp6_request_sock),
935 .rtx_syn_ack = tcp_v6_send_synack,
936 .send_ack = tcp_v6_reqsk_send_ack,
937 .destructor = tcp_v6_reqsk_destructor,
938 .send_reset = tcp_v6_send_reset
941 #ifdef CONFIG_TCP_MD5SIG
942 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
943 .md5_lookup = tcp_v6_reqsk_md5_lookup,
947 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
948 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
949 .twsk_unique = tcp_twsk_unique,
950 .twsk_destructor= tcp_twsk_destructor,
953 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
955 struct ipv6_pinfo *np = inet6_sk(sk);
956 struct tcphdr *th = tcp_hdr(skb);
958 if (skb->ip_summed == CHECKSUM_PARTIAL) {
959 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 0);
960 skb->csum_start = skb_transport_header(skb) - skb->head;
961 skb->csum_offset = offsetof(struct tcphdr, check);
963 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
964 csum_partial((char *)th, th->doff<<2,
969 static int tcp_v6_gso_send_check(struct sk_buff *skb)
971 struct ipv6hdr *ipv6h;
974 if (!pskb_may_pull(skb, sizeof(*th)))
977 ipv6h = ipv6_hdr(skb);
981 th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
983 skb->csum_start = skb_transport_header(skb) - skb->head;
984 skb->csum_offset = offsetof(struct tcphdr, check);
985 skb->ip_summed = CHECKSUM_PARTIAL;
989 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
991 struct tcphdr *th = tcp_hdr(skb), *t1;
992 struct sk_buff *buff;
994 unsigned int tot_len = sizeof(*th);
995 #ifdef CONFIG_TCP_MD5SIG
996 struct tcp_md5sig_key *key;
1002 if (!ipv6_unicast_destination(skb))
1005 #ifdef CONFIG_TCP_MD5SIG
1007 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1012 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1016 * We need to grab some memory, and put together an RST,
1017 * and then put it into the queue to be sent.
1020 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1025 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1027 t1 = (struct tcphdr *) skb_push(buff, tot_len);
1029 /* Swap the send and the receive. */
1030 memset(t1, 0, sizeof(*t1));
1031 t1->dest = th->source;
1032 t1->source = th->dest;
1033 t1->doff = tot_len / 4;
1037 t1->seq = th->ack_seq;
1040 t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
1041 + skb->len - (th->doff<<2));
1044 #ifdef CONFIG_TCP_MD5SIG
1046 __be32 *opt = (__be32*)(t1 + 1);
1047 opt[0] = htonl((TCPOPT_NOP << 24) |
1048 (TCPOPT_NOP << 16) |
1049 (TCPOPT_MD5SIG << 8) |
1051 tcp_v6_do_calc_md5_hash((__u8 *)&opt[1], key,
1052 &ipv6_hdr(skb)->daddr,
1053 &ipv6_hdr(skb)->saddr,
1054 t1, IPPROTO_TCP, tot_len);
1058 buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1060 memset(&fl, 0, sizeof(fl));
1061 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1062 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1064 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1065 sizeof(*t1), IPPROTO_TCP,
1068 fl.proto = IPPROTO_TCP;
1069 fl.oif = inet6_iif(skb);
1070 fl.fl_ip_dport = t1->dest;
1071 fl.fl_ip_sport = t1->source;
1072 security_skb_classify_flow(skb, &fl);
1074 /* Pass a socket to ip6_dst_lookup either it is for RST
1075 * Underlying function will use this to retrieve the network
1078 if (!ip6_dst_lookup(tcp6_socket->sk, &buff->dst, &fl)) {
1080 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1081 ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1082 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1083 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
1091 static void tcp_v6_send_ack(struct tcp_timewait_sock *tw,
1092 struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
1094 struct tcphdr *th = tcp_hdr(skb), *t1;
1095 struct sk_buff *buff;
1097 unsigned int tot_len = sizeof(struct tcphdr);
1099 #ifdef CONFIG_TCP_MD5SIG
1100 struct tcp_md5sig_key *key;
1101 struct tcp_md5sig_key tw_key;
1104 #ifdef CONFIG_TCP_MD5SIG
1105 if (!tw && skb->sk) {
1106 key = tcp_v6_md5_do_lookup(skb->sk, &ipv6_hdr(skb)->daddr);
1107 } else if (tw && tw->tw_md5_keylen) {
1108 tw_key.key = tw->tw_md5_key;
1109 tw_key.keylen = tw->tw_md5_keylen;
1117 tot_len += TCPOLEN_TSTAMP_ALIGNED;
1118 #ifdef CONFIG_TCP_MD5SIG
1120 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1123 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1128 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1130 t1 = (struct tcphdr *) skb_push(buff,tot_len);
1132 /* Swap the send and the receive. */
1133 memset(t1, 0, sizeof(*t1));
1134 t1->dest = th->source;
1135 t1->source = th->dest;
1136 t1->doff = tot_len/4;
1137 t1->seq = htonl(seq);
1138 t1->ack_seq = htonl(ack);
1140 t1->window = htons(win);
1142 topt = (__be32 *)(t1 + 1);
1145 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1146 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1147 *topt++ = htonl(tcp_time_stamp);
1151 #ifdef CONFIG_TCP_MD5SIG
1153 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1154 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1155 tcp_v6_do_calc_md5_hash((__u8 *)topt, key,
1156 &ipv6_hdr(skb)->daddr,
1157 &ipv6_hdr(skb)->saddr,
1158 t1, IPPROTO_TCP, tot_len);
1162 buff->csum = csum_partial((char *)t1, tot_len, 0);
1164 memset(&fl, 0, sizeof(fl));
1165 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1166 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1168 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1169 tot_len, IPPROTO_TCP,
1172 fl.proto = IPPROTO_TCP;
1173 fl.oif = inet6_iif(skb);
1174 fl.fl_ip_dport = t1->dest;
1175 fl.fl_ip_sport = t1->source;
1176 security_skb_classify_flow(skb, &fl);
1178 if (!ip6_dst_lookup(tcp6_socket->sk, &buff->dst, &fl)) {
1179 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1180 ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1181 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1189 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1191 struct inet_timewait_sock *tw = inet_twsk(sk);
1192 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1194 tcp_v6_send_ack(tcptw, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1195 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1196 tcptw->tw_ts_recent);
1201 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
1203 tcp_v6_send_ack(NULL, skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent);
1207 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1209 struct request_sock *req, **prev;
1210 const struct tcphdr *th = tcp_hdr(skb);
1213 /* Find possible connection requests. */
1214 req = inet6_csk_search_req(sk, &prev, th->source,
1215 &ipv6_hdr(skb)->saddr,
1216 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1218 return tcp_check_req(sk, skb, req, prev);
1220 nsk = __inet6_lookup_established(sk->sk_net, &tcp_hashinfo,
1221 &ipv6_hdr(skb)->saddr, th->source,
1222 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1225 if (nsk->sk_state != TCP_TIME_WAIT) {
1229 inet_twsk_put(inet_twsk(nsk));
1233 #ifdef CONFIG_SYN_COOKIES
1234 if (!th->rst && !th->syn && th->ack)
1235 sk = cookie_v6_check(sk, skb);
1240 /* FIXME: this is substantially similar to the ipv4 code.
1241 * Can some kind of merge be done? -- erics
1243 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1245 struct inet6_request_sock *treq;
1246 struct ipv6_pinfo *np = inet6_sk(sk);
1247 struct tcp_options_received tmp_opt;
1248 struct tcp_sock *tp = tcp_sk(sk);
1249 struct request_sock *req = NULL;
1250 __u32 isn = TCP_SKB_CB(skb)->when;
1251 #ifdef CONFIG_SYN_COOKIES
1252 int want_cookie = 0;
1254 #define want_cookie 0
1257 if (skb->protocol == htons(ETH_P_IP))
1258 return tcp_v4_conn_request(sk, skb);
1260 if (!ipv6_unicast_destination(skb))
1263 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1264 if (net_ratelimit())
1265 syn_flood_warning(skb);
1266 #ifdef CONFIG_SYN_COOKIES
1267 if (sysctl_tcp_syncookies)
1274 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1277 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1281 #ifdef CONFIG_TCP_MD5SIG
1282 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1285 tcp_clear_options(&tmp_opt);
1286 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1287 tmp_opt.user_mss = tp->rx_opt.user_mss;
1289 tcp_parse_options(skb, &tmp_opt, 0);
1292 tcp_clear_options(&tmp_opt);
1293 tmp_opt.saw_tstamp = 0;
1296 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1297 tcp_openreq_init(req, &tmp_opt, skb);
1299 treq = inet6_rsk(req);
1300 ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1301 ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1302 treq->pktopts = NULL;
1304 TCP_ECN_create_request(req, tcp_hdr(skb));
1307 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1309 if (ipv6_opt_accepted(sk, skb) ||
1310 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1311 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1312 atomic_inc(&skb->users);
1313 treq->pktopts = skb;
1315 treq->iif = sk->sk_bound_dev_if;
1317 /* So that link locals have meaning */
1318 if (!sk->sk_bound_dev_if &&
1319 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1320 treq->iif = inet6_iif(skb);
1322 isn = tcp_v6_init_sequence(skb);
1325 tcp_rsk(req)->snt_isn = isn;
1327 security_inet_conn_request(sk, skb, req);
1329 if (tcp_v6_send_synack(sk, req))
1333 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1341 return 0; /* don't send reset */
1344 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1345 struct request_sock *req,
1346 struct dst_entry *dst)
1348 struct inet6_request_sock *treq = inet6_rsk(req);
1349 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1350 struct tcp6_sock *newtcp6sk;
1351 struct inet_sock *newinet;
1352 struct tcp_sock *newtp;
1354 struct ipv6_txoptions *opt;
1355 #ifdef CONFIG_TCP_MD5SIG
1356 struct tcp_md5sig_key *key;
1359 if (skb->protocol == htons(ETH_P_IP)) {
1364 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1369 newtcp6sk = (struct tcp6_sock *)newsk;
1370 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1372 newinet = inet_sk(newsk);
1373 newnp = inet6_sk(newsk);
1374 newtp = tcp_sk(newsk);
1376 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1378 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1381 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1384 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1386 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1387 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1388 #ifdef CONFIG_TCP_MD5SIG
1389 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1392 newnp->pktoptions = NULL;
1394 newnp->mcast_oif = inet6_iif(skb);
1395 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1398 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1399 * here, tcp_create_openreq_child now does this for us, see the comment in
1400 * that function for the gory details. -acme
1403 /* It is tricky place. Until this moment IPv4 tcp
1404 worked with IPv6 icsk.icsk_af_ops.
1407 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1414 if (sk_acceptq_is_full(sk))
1418 struct in6_addr *final_p = NULL, final;
1421 memset(&fl, 0, sizeof(fl));
1422 fl.proto = IPPROTO_TCP;
1423 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1424 if (opt && opt->srcrt) {
1425 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1426 ipv6_addr_copy(&final, &fl.fl6_dst);
1427 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1430 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1431 fl.oif = sk->sk_bound_dev_if;
1432 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1433 fl.fl_ip_sport = inet_sk(sk)->sport;
1434 security_req_classify_flow(req, &fl);
1436 if (ip6_dst_lookup(sk, &dst, &fl))
1440 ipv6_addr_copy(&fl.fl6_dst, final_p);
1442 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1446 newsk = tcp_create_openreq_child(sk, req, skb);
1451 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1452 * count here, tcp_create_openreq_child now does this for us, see the
1453 * comment in that function for the gory details. -acme
1456 newsk->sk_gso_type = SKB_GSO_TCPV6;
1457 __ip6_dst_store(newsk, dst, NULL, NULL);
1459 newtcp6sk = (struct tcp6_sock *)newsk;
1460 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1462 newtp = tcp_sk(newsk);
1463 newinet = inet_sk(newsk);
1464 newnp = inet6_sk(newsk);
1466 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1468 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1469 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1470 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1471 newsk->sk_bound_dev_if = treq->iif;
1473 /* Now IPv6 options...
1475 First: no IPv4 options.
1477 newinet->opt = NULL;
1478 newnp->ipv6_fl_list = NULL;
1481 newnp->rxopt.all = np->rxopt.all;
1483 /* Clone pktoptions received with SYN */
1484 newnp->pktoptions = NULL;
1485 if (treq->pktopts != NULL) {
1486 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1487 kfree_skb(treq->pktopts);
1488 treq->pktopts = NULL;
1489 if (newnp->pktoptions)
1490 skb_set_owner_r(newnp->pktoptions, newsk);
1493 newnp->mcast_oif = inet6_iif(skb);
1494 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1496 /* Clone native IPv6 options from listening socket (if any)
1498 Yes, keeping reference count would be much more clever,
1499 but we make one more one thing there: reattach optmem
1503 newnp->opt = ipv6_dup_options(newsk, opt);
1505 sock_kfree_s(sk, opt, opt->tot_len);
1508 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1510 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1511 newnp->opt->opt_flen);
1513 tcp_mtup_init(newsk);
1514 tcp_sync_mss(newsk, dst_mtu(dst));
1515 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1516 tcp_initialize_rcv_mss(newsk);
1518 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1520 #ifdef CONFIG_TCP_MD5SIG
1521 /* Copy over the MD5 key from the original socket */
1522 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1523 /* We're using one, so create a matching key
1524 * on the newsk structure. If we fail to get
1525 * memory, then we end up not copying the key
1528 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1530 tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1531 newkey, key->keylen);
1535 __inet6_hash(newsk);
1536 inet_inherit_port(sk, newsk);
1541 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1543 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1544 if (opt && opt != np->opt)
1545 sock_kfree_s(sk, opt, opt->tot_len);
1550 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1552 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1553 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1554 &ipv6_hdr(skb)->daddr, skb->csum)) {
1555 skb->ip_summed = CHECKSUM_UNNECESSARY;
1560 skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1561 &ipv6_hdr(skb)->saddr,
1562 &ipv6_hdr(skb)->daddr, 0));
1564 if (skb->len <= 76) {
1565 return __skb_checksum_complete(skb);
1570 /* The socket must have it's spinlock held when we get
1573 * We have a potential double-lock case here, so even when
1574 * doing backlog processing we use the BH locking scheme.
1575 * This is because we cannot sleep with the original spinlock
1578 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1580 struct ipv6_pinfo *np = inet6_sk(sk);
1581 struct tcp_sock *tp;
1582 struct sk_buff *opt_skb = NULL;
1584 /* Imagine: socket is IPv6. IPv4 packet arrives,
1585 goes to IPv4 receive handler and backlogged.
1586 From backlog it always goes here. Kerboom...
1587 Fortunately, tcp_rcv_established and rcv_established
1588 handle them correctly, but it is not case with
1589 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1592 if (skb->protocol == htons(ETH_P_IP))
1593 return tcp_v4_do_rcv(sk, skb);
1595 #ifdef CONFIG_TCP_MD5SIG
1596 if (tcp_v6_inbound_md5_hash (sk, skb))
1600 if (sk_filter(sk, skb))
1604 * socket locking is here for SMP purposes as backlog rcv
1605 * is currently called with bh processing disabled.
1608 /* Do Stevens' IPV6_PKTOPTIONS.
1610 Yes, guys, it is the only place in our code, where we
1611 may make it not affecting IPv4.
1612 The rest of code is protocol independent,
1613 and I do not like idea to uglify IPv4.
1615 Actually, all the idea behind IPV6_PKTOPTIONS
1616 looks not very well thought. For now we latch
1617 options, received in the last packet, enqueued
1618 by tcp. Feel free to propose better solution.
1622 opt_skb = skb_clone(skb, GFP_ATOMIC);
1624 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1625 TCP_CHECK_TIMER(sk);
1626 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1628 TCP_CHECK_TIMER(sk);
1630 goto ipv6_pktoptions;
1634 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1637 if (sk->sk_state == TCP_LISTEN) {
1638 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1643 * Queue it on the new socket if the new socket is active,
1644 * otherwise we just shortcircuit this and continue with
1648 if (tcp_child_process(sk, nsk, skb))
1651 __kfree_skb(opt_skb);
1656 TCP_CHECK_TIMER(sk);
1657 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1659 TCP_CHECK_TIMER(sk);
1661 goto ipv6_pktoptions;
1665 tcp_v6_send_reset(sk, skb);
1668 __kfree_skb(opt_skb);
1672 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1677 /* Do you ask, what is it?
1679 1. skb was enqueued by tcp.
1680 2. skb is added to tail of read queue, rather than out of order.
1681 3. socket is not in passive state.
1682 4. Finally, it really contains options, which user wants to receive.
1685 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1686 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1687 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1688 np->mcast_oif = inet6_iif(opt_skb);
1689 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1690 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1691 if (ipv6_opt_accepted(sk, opt_skb)) {
1692 skb_set_owner_r(opt_skb, sk);
1693 opt_skb = xchg(&np->pktoptions, opt_skb);
1695 __kfree_skb(opt_skb);
1696 opt_skb = xchg(&np->pktoptions, NULL);
1705 static int tcp_v6_rcv(struct sk_buff *skb)
1711 if (skb->pkt_type != PACKET_HOST)
1715 * Count it even if it's bad.
1717 TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1719 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1724 if (th->doff < sizeof(struct tcphdr)/4)
1726 if (!pskb_may_pull(skb, th->doff*4))
1729 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1733 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1734 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1735 skb->len - th->doff*4);
1736 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1737 TCP_SKB_CB(skb)->when = 0;
1738 TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1739 TCP_SKB_CB(skb)->sacked = 0;
1741 sk = __inet6_lookup(skb->dev->nd_net, &tcp_hashinfo,
1742 &ipv6_hdr(skb)->saddr, th->source,
1743 &ipv6_hdr(skb)->daddr, ntohs(th->dest),
1750 if (sk->sk_state == TCP_TIME_WAIT)
1753 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1754 goto discard_and_relse;
1756 if (sk_filter(sk, skb))
1757 goto discard_and_relse;
1761 bh_lock_sock_nested(sk);
1763 if (!sock_owned_by_user(sk)) {
1764 #ifdef CONFIG_NET_DMA
1765 struct tcp_sock *tp = tcp_sk(sk);
1766 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1767 tp->ucopy.dma_chan = get_softnet_dma();
1768 if (tp->ucopy.dma_chan)
1769 ret = tcp_v6_do_rcv(sk, skb);
1773 if (!tcp_prequeue(sk, skb))
1774 ret = tcp_v6_do_rcv(sk, skb);
1777 sk_add_backlog(sk, skb);
1781 return ret ? -1 : 0;
1784 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1787 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1789 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1791 tcp_v6_send_reset(NULL, skb);
1808 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1809 inet_twsk_put(inet_twsk(sk));
1813 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1814 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1815 inet_twsk_put(inet_twsk(sk));
1819 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1824 sk2 = inet6_lookup_listener(skb->dev->nd_net, &tcp_hashinfo,
1825 &ipv6_hdr(skb)->daddr,
1826 ntohs(th->dest), inet6_iif(skb));
1828 struct inet_timewait_sock *tw = inet_twsk(sk);
1829 inet_twsk_deschedule(tw, &tcp_death_row);
1834 /* Fall through to ACK */
1837 tcp_v6_timewait_ack(sk, skb);
1841 case TCP_TW_SUCCESS:;
1846 static int tcp_v6_remember_stamp(struct sock *sk)
1848 /* Alas, not yet... */
1852 static struct inet_connection_sock_af_ops ipv6_specific = {
1853 .queue_xmit = inet6_csk_xmit,
1854 .send_check = tcp_v6_send_check,
1855 .rebuild_header = inet6_sk_rebuild_header,
1856 .conn_request = tcp_v6_conn_request,
1857 .syn_recv_sock = tcp_v6_syn_recv_sock,
1858 .remember_stamp = tcp_v6_remember_stamp,
1859 .net_header_len = sizeof(struct ipv6hdr),
1860 .setsockopt = ipv6_setsockopt,
1861 .getsockopt = ipv6_getsockopt,
1862 .addr2sockaddr = inet6_csk_addr2sockaddr,
1863 .sockaddr_len = sizeof(struct sockaddr_in6),
1864 .bind_conflict = inet6_csk_bind_conflict,
1865 #ifdef CONFIG_COMPAT
1866 .compat_setsockopt = compat_ipv6_setsockopt,
1867 .compat_getsockopt = compat_ipv6_getsockopt,
1871 #ifdef CONFIG_TCP_MD5SIG
1872 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1873 .md5_lookup = tcp_v6_md5_lookup,
1874 .calc_md5_hash = tcp_v6_calc_md5_hash,
1875 .md5_add = tcp_v6_md5_add_func,
1876 .md5_parse = tcp_v6_parse_md5_keys,
1881 * TCP over IPv4 via INET6 API
1884 static struct inet_connection_sock_af_ops ipv6_mapped = {
1885 .queue_xmit = ip_queue_xmit,
1886 .send_check = tcp_v4_send_check,
1887 .rebuild_header = inet_sk_rebuild_header,
1888 .conn_request = tcp_v6_conn_request,
1889 .syn_recv_sock = tcp_v6_syn_recv_sock,
1890 .remember_stamp = tcp_v4_remember_stamp,
1891 .net_header_len = sizeof(struct iphdr),
1892 .setsockopt = ipv6_setsockopt,
1893 .getsockopt = ipv6_getsockopt,
1894 .addr2sockaddr = inet6_csk_addr2sockaddr,
1895 .sockaddr_len = sizeof(struct sockaddr_in6),
1896 .bind_conflict = inet6_csk_bind_conflict,
1897 #ifdef CONFIG_COMPAT
1898 .compat_setsockopt = compat_ipv6_setsockopt,
1899 .compat_getsockopt = compat_ipv6_getsockopt,
1903 #ifdef CONFIG_TCP_MD5SIG
1904 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1905 .md5_lookup = tcp_v4_md5_lookup,
1906 .calc_md5_hash = tcp_v4_calc_md5_hash,
1907 .md5_add = tcp_v6_md5_add_func,
1908 .md5_parse = tcp_v6_parse_md5_keys,
1912 /* NOTE: A lot of things set to zero explicitly by call to
1913 * sk_alloc() so need not be done here.
1915 static int tcp_v6_init_sock(struct sock *sk)
1917 struct inet_connection_sock *icsk = inet_csk(sk);
1918 struct tcp_sock *tp = tcp_sk(sk);
1920 skb_queue_head_init(&tp->out_of_order_queue);
1921 tcp_init_xmit_timers(sk);
1922 tcp_prequeue_init(tp);
1924 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1925 tp->mdev = TCP_TIMEOUT_INIT;
1927 /* So many TCP implementations out there (incorrectly) count the
1928 * initial SYN frame in their delayed-ACK and congestion control
1929 * algorithms that we must have the following bandaid to talk
1930 * efficiently to them. -DaveM
1934 /* See draft-stevens-tcpca-spec-01 for discussion of the
1935 * initialization of these values.
1937 tp->snd_ssthresh = 0x7fffffff;
1938 tp->snd_cwnd_clamp = ~0;
1939 tp->mss_cache = 536;
1941 tp->reordering = sysctl_tcp_reordering;
1943 sk->sk_state = TCP_CLOSE;
1945 icsk->icsk_af_ops = &ipv6_specific;
1946 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1947 icsk->icsk_sync_mss = tcp_sync_mss;
1948 sk->sk_write_space = sk_stream_write_space;
1949 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1951 #ifdef CONFIG_TCP_MD5SIG
1952 tp->af_specific = &tcp_sock_ipv6_specific;
1955 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1956 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1958 atomic_inc(&tcp_sockets_allocated);
1963 static int tcp_v6_destroy_sock(struct sock *sk)
1965 #ifdef CONFIG_TCP_MD5SIG
1966 /* Clean up the MD5 key list */
1967 if (tcp_sk(sk)->md5sig_info)
1968 tcp_v6_clear_md5_list(sk);
1970 tcp_v4_destroy_sock(sk);
1971 return inet6_destroy_sock(sk);
1974 #ifdef CONFIG_PROC_FS
1975 /* Proc filesystem TCPv6 sock list dumping. */
1976 static void get_openreq6(struct seq_file *seq,
1977 struct sock *sk, struct request_sock *req, int i, int uid)
1979 int ttd = req->expires - jiffies;
1980 struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1981 struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1987 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1988 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1990 src->s6_addr32[0], src->s6_addr32[1],
1991 src->s6_addr32[2], src->s6_addr32[3],
1992 ntohs(inet_sk(sk)->sport),
1993 dest->s6_addr32[0], dest->s6_addr32[1],
1994 dest->s6_addr32[2], dest->s6_addr32[3],
1995 ntohs(inet_rsk(req)->rmt_port),
1997 0,0, /* could print option size, but that is af dependent. */
1998 1, /* timers active (only the expire timer) */
1999 jiffies_to_clock_t(ttd),
2002 0, /* non standard timer */
2003 0, /* open_requests have no inode */
2007 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2009 struct in6_addr *dest, *src;
2012 unsigned long timer_expires;
2013 struct inet_sock *inet = inet_sk(sp);
2014 struct tcp_sock *tp = tcp_sk(sp);
2015 const struct inet_connection_sock *icsk = inet_csk(sp);
2016 struct ipv6_pinfo *np = inet6_sk(sp);
2019 src = &np->rcv_saddr;
2020 destp = ntohs(inet->dport);
2021 srcp = ntohs(inet->sport);
2023 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2025 timer_expires = icsk->icsk_timeout;
2026 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2028 timer_expires = icsk->icsk_timeout;
2029 } else if (timer_pending(&sp->sk_timer)) {
2031 timer_expires = sp->sk_timer.expires;
2034 timer_expires = jiffies;
2038 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2039 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
2041 src->s6_addr32[0], src->s6_addr32[1],
2042 src->s6_addr32[2], src->s6_addr32[3], srcp,
2043 dest->s6_addr32[0], dest->s6_addr32[1],
2044 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2046 tp->write_seq-tp->snd_una,
2047 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2049 jiffies_to_clock_t(timer_expires - jiffies),
2050 icsk->icsk_retransmits,
2052 icsk->icsk_probes_out,
2054 atomic_read(&sp->sk_refcnt), sp,
2057 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2058 tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2062 static void get_timewait6_sock(struct seq_file *seq,
2063 struct inet_timewait_sock *tw, int i)
2065 struct in6_addr *dest, *src;
2067 struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2068 int ttd = tw->tw_ttd - jiffies;
2073 dest = &tw6->tw_v6_daddr;
2074 src = &tw6->tw_v6_rcv_saddr;
2075 destp = ntohs(tw->tw_dport);
2076 srcp = ntohs(tw->tw_sport);
2079 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2080 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2082 src->s6_addr32[0], src->s6_addr32[1],
2083 src->s6_addr32[2], src->s6_addr32[3], srcp,
2084 dest->s6_addr32[0], dest->s6_addr32[1],
2085 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2086 tw->tw_substate, 0, 0,
2087 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2088 atomic_read(&tw->tw_refcnt), tw);
2091 static int tcp6_seq_show(struct seq_file *seq, void *v)
2093 struct tcp_iter_state *st;
2095 if (v == SEQ_START_TOKEN) {
2100 "st tx_queue rx_queue tr tm->when retrnsmt"
2101 " uid timeout inode\n");
2106 switch (st->state) {
2107 case TCP_SEQ_STATE_LISTENING:
2108 case TCP_SEQ_STATE_ESTABLISHED:
2109 get_tcp6_sock(seq, v, st->num);
2111 case TCP_SEQ_STATE_OPENREQ:
2112 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2114 case TCP_SEQ_STATE_TIME_WAIT:
2115 get_timewait6_sock(seq, v, st->num);
2122 static struct file_operations tcp6_seq_fops;
2123 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2124 .owner = THIS_MODULE,
2127 .seq_show = tcp6_seq_show,
2128 .seq_fops = &tcp6_seq_fops,
2131 int __init tcp6_proc_init(void)
2133 return tcp_proc_register(&tcp6_seq_afinfo);
2136 void tcp6_proc_exit(void)
2138 tcp_proc_unregister(&tcp6_seq_afinfo);
2142 DEFINE_PROTO_INUSE(tcpv6)
2144 struct proto tcpv6_prot = {
2146 .owner = THIS_MODULE,
2148 .connect = tcp_v6_connect,
2149 .disconnect = tcp_disconnect,
2150 .accept = inet_csk_accept,
2152 .init = tcp_v6_init_sock,
2153 .destroy = tcp_v6_destroy_sock,
2154 .shutdown = tcp_shutdown,
2155 .setsockopt = tcp_setsockopt,
2156 .getsockopt = tcp_getsockopt,
2157 .recvmsg = tcp_recvmsg,
2158 .backlog_rcv = tcp_v6_do_rcv,
2159 .hash = tcp_v6_hash,
2160 .unhash = inet_unhash,
2161 .get_port = inet_csk_get_port,
2162 .enter_memory_pressure = tcp_enter_memory_pressure,
2163 .sockets_allocated = &tcp_sockets_allocated,
2164 .memory_allocated = &tcp_memory_allocated,
2165 .memory_pressure = &tcp_memory_pressure,
2166 .orphan_count = &tcp_orphan_count,
2167 .sysctl_mem = sysctl_tcp_mem,
2168 .sysctl_wmem = sysctl_tcp_wmem,
2169 .sysctl_rmem = sysctl_tcp_rmem,
2170 .max_header = MAX_TCP_HEADER,
2171 .obj_size = sizeof(struct tcp6_sock),
2172 .twsk_prot = &tcp6_timewait_sock_ops,
2173 .rsk_prot = &tcp6_request_sock_ops,
2174 .hashinfo = &tcp_hashinfo,
2175 #ifdef CONFIG_COMPAT
2176 .compat_setsockopt = compat_tcp_setsockopt,
2177 .compat_getsockopt = compat_tcp_getsockopt,
2179 REF_PROTO_INUSE(tcpv6)
2182 static struct inet6_protocol tcpv6_protocol = {
2183 .handler = tcp_v6_rcv,
2184 .err_handler = tcp_v6_err,
2185 .gso_send_check = tcp_v6_gso_send_check,
2186 .gso_segment = tcp_tso_segment,
2187 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2190 static struct inet_protosw tcpv6_protosw = {
2191 .type = SOCK_STREAM,
2192 .protocol = IPPROTO_TCP,
2193 .prot = &tcpv6_prot,
2194 .ops = &inet6_stream_ops,
2197 .flags = INET_PROTOSW_PERMANENT |
2201 int __init tcpv6_init(void)
2205 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2209 /* register inet6 protocol */
2210 ret = inet6_register_protosw(&tcpv6_protosw);
2212 goto out_tcpv6_protocol;
2214 ret = inet_csk_ctl_sock_create(&tcp6_socket, PF_INET6,
2215 SOCK_RAW, IPPROTO_TCP);
2217 goto out_tcpv6_protosw;
2222 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2224 inet6_unregister_protosw(&tcpv6_protosw);
2228 void tcpv6_exit(void)
2230 sock_release(tcp6_socket);
2231 inet6_unregister_protosw(&tcpv6_protosw);
2232 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);