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 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) {
270 err = ip6_dst_blackhole(sk, &dst, &fl);
277 ipv6_addr_copy(&np->rcv_saddr, saddr);
280 /* set the source address */
281 ipv6_addr_copy(&np->saddr, saddr);
282 inet->rcv_saddr = LOOPBACK4_IPV6;
284 sk->sk_gso_type = SKB_GSO_TCPV6;
285 __ip6_dst_store(sk, dst, NULL, NULL);
287 icsk->icsk_ext_hdr_len = 0;
289 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
292 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
294 inet->dport = usin->sin6_port;
296 tcp_set_state(sk, TCP_SYN_SENT);
297 err = inet6_hash_connect(&tcp_death_row, sk);
302 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
307 err = tcp_connect(sk);
314 tcp_set_state(sk, TCP_CLOSE);
318 sk->sk_route_caps = 0;
322 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
323 int type, int code, int offset, __be32 info)
325 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
326 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
327 struct ipv6_pinfo *np;
333 sk = inet6_lookup(&tcp_hashinfo, &hdr->daddr, th->dest, &hdr->saddr,
334 th->source, skb->dev->ifindex);
337 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
341 if (sk->sk_state == TCP_TIME_WAIT) {
342 inet_twsk_put(inet_twsk(sk));
347 if (sock_owned_by_user(sk))
348 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
350 if (sk->sk_state == TCP_CLOSE)
354 seq = ntohl(th->seq);
355 if (sk->sk_state != TCP_LISTEN &&
356 !between(seq, tp->snd_una, tp->snd_nxt)) {
357 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
363 if (type == ICMPV6_PKT_TOOBIG) {
364 struct dst_entry *dst = NULL;
366 if (sock_owned_by_user(sk))
368 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
371 /* icmp should have updated the destination cache entry */
372 dst = __sk_dst_check(sk, np->dst_cookie);
375 struct inet_sock *inet = inet_sk(sk);
378 /* BUGGG_FUTURE: Again, it is not clear how
379 to handle rthdr case. Ignore this complexity
382 memset(&fl, 0, sizeof(fl));
383 fl.proto = IPPROTO_TCP;
384 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
385 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
386 fl.oif = sk->sk_bound_dev_if;
387 fl.fl_ip_dport = inet->dport;
388 fl.fl_ip_sport = inet->sport;
389 security_skb_classify_flow(skb, &fl);
391 if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
392 sk->sk_err_soft = -err;
396 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
397 sk->sk_err_soft = -err;
404 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
405 tcp_sync_mss(sk, dst_mtu(dst));
406 tcp_simple_retransmit(sk);
407 } /* else let the usual retransmit timer handle it */
412 icmpv6_err_convert(type, code, &err);
414 /* Might be for an request_sock */
415 switch (sk->sk_state) {
416 struct request_sock *req, **prev;
418 if (sock_owned_by_user(sk))
421 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
422 &hdr->saddr, inet6_iif(skb));
426 /* ICMPs are not backlogged, hence we cannot get
427 * an established socket here.
429 BUG_TRAP(req->sk == NULL);
431 if (seq != tcp_rsk(req)->snt_isn) {
432 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
436 inet_csk_reqsk_queue_drop(sk, req, prev);
440 case TCP_SYN_RECV: /* Cannot happen.
441 It can, it SYNs are crossed. --ANK */
442 if (!sock_owned_by_user(sk)) {
444 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
448 sk->sk_err_soft = err;
452 if (!sock_owned_by_user(sk) && np->recverr) {
454 sk->sk_error_report(sk);
456 sk->sk_err_soft = err;
464 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
465 struct dst_entry *dst)
467 struct inet6_request_sock *treq = inet6_rsk(req);
468 struct ipv6_pinfo *np = inet6_sk(sk);
469 struct sk_buff * skb;
470 struct ipv6_txoptions *opt = NULL;
471 struct in6_addr * final_p = NULL, final;
475 memset(&fl, 0, sizeof(fl));
476 fl.proto = IPPROTO_TCP;
477 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
478 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
479 fl.fl6_flowlabel = 0;
481 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
482 fl.fl_ip_sport = inet_sk(sk)->sport;
483 security_req_classify_flow(req, &fl);
487 if (opt && opt->srcrt) {
488 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
489 ipv6_addr_copy(&final, &fl.fl6_dst);
490 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
494 err = ip6_dst_lookup(sk, &dst, &fl);
498 ipv6_addr_copy(&fl.fl6_dst, final_p);
499 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
503 skb = tcp_make_synack(sk, dst, req);
505 struct tcphdr *th = tcp_hdr(skb);
507 th->check = tcp_v6_check(th, skb->len,
508 &treq->loc_addr, &treq->rmt_addr,
509 csum_partial((char *)th, skb->len, skb->csum));
511 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
512 err = ip6_xmit(sk, skb, &fl, opt, 0);
513 err = net_xmit_eval(err);
517 if (opt && opt != np->opt)
518 sock_kfree_s(sk, opt, opt->tot_len);
523 static void tcp_v6_reqsk_destructor(struct request_sock *req)
525 if (inet6_rsk(req)->pktopts)
526 kfree_skb(inet6_rsk(req)->pktopts);
529 #ifdef CONFIG_TCP_MD5SIG
530 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
531 struct in6_addr *addr)
533 struct tcp_sock *tp = tcp_sk(sk);
538 if (!tp->md5sig_info || !tp->md5sig_info->entries6)
541 for (i = 0; i < tp->md5sig_info->entries6; i++) {
542 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, addr) == 0)
543 return &tp->md5sig_info->keys6[i].base;
548 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
549 struct sock *addr_sk)
551 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
554 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
555 struct request_sock *req)
557 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
560 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
561 char *newkey, u8 newkeylen)
563 /* Add key to the list */
564 struct tcp_md5sig_key *key;
565 struct tcp_sock *tp = tcp_sk(sk);
566 struct tcp6_md5sig_key *keys;
568 key = tcp_v6_md5_do_lookup(sk, peer);
570 /* modify existing entry - just update that one */
573 key->keylen = newkeylen;
575 /* reallocate new list if current one is full. */
576 if (!tp->md5sig_info) {
577 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
578 if (!tp->md5sig_info) {
582 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
584 tcp_alloc_md5sig_pool();
585 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
586 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
587 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
590 tcp_free_md5sig_pool();
595 if (tp->md5sig_info->entries6)
596 memmove(keys, tp->md5sig_info->keys6,
597 (sizeof (tp->md5sig_info->keys6[0]) *
598 tp->md5sig_info->entries6));
600 kfree(tp->md5sig_info->keys6);
601 tp->md5sig_info->keys6 = keys;
602 tp->md5sig_info->alloced6++;
605 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
607 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
608 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
610 tp->md5sig_info->entries6++;
615 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
616 u8 *newkey, __u8 newkeylen)
618 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
622 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
624 struct tcp_sock *tp = tcp_sk(sk);
627 for (i = 0; i < tp->md5sig_info->entries6; i++) {
628 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, peer) == 0) {
630 kfree(tp->md5sig_info->keys6[i].base.key);
631 tp->md5sig_info->entries6--;
633 if (tp->md5sig_info->entries6 == 0) {
634 kfree(tp->md5sig_info->keys6);
635 tp->md5sig_info->keys6 = NULL;
636 tp->md5sig_info->alloced6 = 0;
638 tcp_free_md5sig_pool();
642 /* shrink the database */
643 if (tp->md5sig_info->entries6 != i)
644 memmove(&tp->md5sig_info->keys6[i],
645 &tp->md5sig_info->keys6[i+1],
646 (tp->md5sig_info->entries6 - i)
647 * sizeof (tp->md5sig_info->keys6[0]));
654 static void tcp_v6_clear_md5_list (struct sock *sk)
656 struct tcp_sock *tp = tcp_sk(sk);
659 if (tp->md5sig_info->entries6) {
660 for (i = 0; i < tp->md5sig_info->entries6; i++)
661 kfree(tp->md5sig_info->keys6[i].base.key);
662 tp->md5sig_info->entries6 = 0;
663 tcp_free_md5sig_pool();
666 kfree(tp->md5sig_info->keys6);
667 tp->md5sig_info->keys6 = NULL;
668 tp->md5sig_info->alloced6 = 0;
670 if (tp->md5sig_info->entries4) {
671 for (i = 0; i < tp->md5sig_info->entries4; i++)
672 kfree(tp->md5sig_info->keys4[i].base.key);
673 tp->md5sig_info->entries4 = 0;
674 tcp_free_md5sig_pool();
677 kfree(tp->md5sig_info->keys4);
678 tp->md5sig_info->keys4 = NULL;
679 tp->md5sig_info->alloced4 = 0;
682 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
685 struct tcp_md5sig cmd;
686 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
689 if (optlen < sizeof(cmd))
692 if (copy_from_user(&cmd, optval, sizeof(cmd)))
695 if (sin6->sin6_family != AF_INET6)
698 if (!cmd.tcpm_keylen) {
699 if (!tcp_sk(sk)->md5sig_info)
701 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
702 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
703 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
706 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
709 if (!tcp_sk(sk)->md5sig_info) {
710 struct tcp_sock *tp = tcp_sk(sk);
711 struct tcp_md5sig_info *p;
713 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
718 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
721 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
724 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
725 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
726 newkey, cmd.tcpm_keylen);
728 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
731 static int tcp_v6_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
732 struct in6_addr *saddr,
733 struct in6_addr *daddr,
734 struct tcphdr *th, int protocol,
737 struct scatterlist sg[4];
741 struct tcp_md5sig_pool *hp;
742 struct tcp6_pseudohdr *bp;
743 struct hash_desc *desc;
745 unsigned int nbytes = 0;
747 hp = tcp_get_md5sig_pool();
749 printk(KERN_WARNING "%s(): hash pool not found...\n", __FUNCTION__);
750 goto clear_hash_noput;
752 bp = &hp->md5_blk.ip6;
753 desc = &hp->md5_desc;
755 /* 1. TCP pseudo-header (RFC2460) */
756 ipv6_addr_copy(&bp->saddr, saddr);
757 ipv6_addr_copy(&bp->daddr, daddr);
758 bp->len = htonl(tcplen);
759 bp->protocol = htonl(protocol);
761 sg_init_table(sg, 4);
763 sg_set_buf(&sg[block++], bp, sizeof(*bp));
764 nbytes += sizeof(*bp);
766 /* 2. TCP header, excluding options */
769 sg_set_buf(&sg[block++], th, sizeof(*th));
770 nbytes += sizeof(*th);
772 /* 3. TCP segment data (if any) */
773 data_len = tcplen - (th->doff << 2);
775 u8 *data = (u8 *)th + (th->doff << 2);
776 sg_set_buf(&sg[block++], data, data_len);
781 sg_set_buf(&sg[block++], key->key, key->keylen);
782 nbytes += key->keylen;
784 sg_mark_end(&sg[block - 1]);
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))
1389 struct in6_addr *final_p = NULL, final;
1392 memset(&fl, 0, sizeof(fl));
1393 fl.proto = IPPROTO_TCP;
1394 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1395 if (opt && opt->srcrt) {
1396 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1397 ipv6_addr_copy(&final, &fl.fl6_dst);
1398 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1401 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1402 fl.oif = sk->sk_bound_dev_if;
1403 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1404 fl.fl_ip_sport = inet_sk(sk)->sport;
1405 security_req_classify_flow(req, &fl);
1407 if (ip6_dst_lookup(sk, &dst, &fl))
1411 ipv6_addr_copy(&fl.fl6_dst, final_p);
1413 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1417 newsk = tcp_create_openreq_child(sk, req, skb);
1422 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1423 * count here, tcp_create_openreq_child now does this for us, see the
1424 * comment in that function for the gory details. -acme
1427 newsk->sk_gso_type = SKB_GSO_TCPV6;
1428 __ip6_dst_store(newsk, dst, NULL, NULL);
1430 newtcp6sk = (struct tcp6_sock *)newsk;
1431 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1433 newtp = tcp_sk(newsk);
1434 newinet = inet_sk(newsk);
1435 newnp = inet6_sk(newsk);
1437 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1439 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1440 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1441 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1442 newsk->sk_bound_dev_if = treq->iif;
1444 /* Now IPv6 options...
1446 First: no IPv4 options.
1448 newinet->opt = NULL;
1449 newnp->ipv6_fl_list = NULL;
1452 newnp->rxopt.all = np->rxopt.all;
1454 /* Clone pktoptions received with SYN */
1455 newnp->pktoptions = NULL;
1456 if (treq->pktopts != NULL) {
1457 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1458 kfree_skb(treq->pktopts);
1459 treq->pktopts = NULL;
1460 if (newnp->pktoptions)
1461 skb_set_owner_r(newnp->pktoptions, newsk);
1464 newnp->mcast_oif = inet6_iif(skb);
1465 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1467 /* Clone native IPv6 options from listening socket (if any)
1469 Yes, keeping reference count would be much more clever,
1470 but we make one more one thing there: reattach optmem
1474 newnp->opt = ipv6_dup_options(newsk, opt);
1476 sock_kfree_s(sk, opt, opt->tot_len);
1479 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1481 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1482 newnp->opt->opt_flen);
1484 tcp_mtup_init(newsk);
1485 tcp_sync_mss(newsk, dst_mtu(dst));
1486 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1487 tcp_initialize_rcv_mss(newsk);
1489 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1491 #ifdef CONFIG_TCP_MD5SIG
1492 /* Copy over the MD5 key from the original socket */
1493 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1494 /* We're using one, so create a matching key
1495 * on the newsk structure. If we fail to get
1496 * memory, then we end up not copying the key
1499 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1501 tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1502 newkey, key->keylen);
1506 __inet6_hash(&tcp_hashinfo, newsk);
1507 inet_inherit_port(&tcp_hashinfo, sk, newsk);
1512 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1514 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1515 if (opt && opt != np->opt)
1516 sock_kfree_s(sk, opt, opt->tot_len);
1521 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1523 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1524 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1525 &ipv6_hdr(skb)->daddr, skb->csum)) {
1526 skb->ip_summed = CHECKSUM_UNNECESSARY;
1531 skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1532 &ipv6_hdr(skb)->saddr,
1533 &ipv6_hdr(skb)->daddr, 0));
1535 if (skb->len <= 76) {
1536 return __skb_checksum_complete(skb);
1541 /* The socket must have it's spinlock held when we get
1544 * We have a potential double-lock case here, so even when
1545 * doing backlog processing we use the BH locking scheme.
1546 * This is because we cannot sleep with the original spinlock
1549 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1551 struct ipv6_pinfo *np = inet6_sk(sk);
1552 struct tcp_sock *tp;
1553 struct sk_buff *opt_skb = NULL;
1555 /* Imagine: socket is IPv6. IPv4 packet arrives,
1556 goes to IPv4 receive handler and backlogged.
1557 From backlog it always goes here. Kerboom...
1558 Fortunately, tcp_rcv_established and rcv_established
1559 handle them correctly, but it is not case with
1560 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1563 if (skb->protocol == htons(ETH_P_IP))
1564 return tcp_v4_do_rcv(sk, skb);
1566 #ifdef CONFIG_TCP_MD5SIG
1567 if (tcp_v6_inbound_md5_hash (sk, skb))
1571 if (sk_filter(sk, skb))
1575 * socket locking is here for SMP purposes as backlog rcv
1576 * is currently called with bh processing disabled.
1579 /* Do Stevens' IPV6_PKTOPTIONS.
1581 Yes, guys, it is the only place in our code, where we
1582 may make it not affecting IPv4.
1583 The rest of code is protocol independent,
1584 and I do not like idea to uglify IPv4.
1586 Actually, all the idea behind IPV6_PKTOPTIONS
1587 looks not very well thought. For now we latch
1588 options, received in the last packet, enqueued
1589 by tcp. Feel free to propose better solution.
1593 opt_skb = skb_clone(skb, GFP_ATOMIC);
1595 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1596 TCP_CHECK_TIMER(sk);
1597 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1599 TCP_CHECK_TIMER(sk);
1601 goto ipv6_pktoptions;
1605 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1608 if (sk->sk_state == TCP_LISTEN) {
1609 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1614 * Queue it on the new socket if the new socket is active,
1615 * otherwise we just shortcircuit this and continue with
1619 if (tcp_child_process(sk, nsk, skb))
1622 __kfree_skb(opt_skb);
1627 TCP_CHECK_TIMER(sk);
1628 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1630 TCP_CHECK_TIMER(sk);
1632 goto ipv6_pktoptions;
1636 tcp_v6_send_reset(sk, skb);
1639 __kfree_skb(opt_skb);
1643 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1648 /* Do you ask, what is it?
1650 1. skb was enqueued by tcp.
1651 2. skb is added to tail of read queue, rather than out of order.
1652 3. socket is not in passive state.
1653 4. Finally, it really contains options, which user wants to receive.
1656 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1657 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1658 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1659 np->mcast_oif = inet6_iif(opt_skb);
1660 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1661 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1662 if (ipv6_opt_accepted(sk, opt_skb)) {
1663 skb_set_owner_r(opt_skb, sk);
1664 opt_skb = xchg(&np->pktoptions, opt_skb);
1666 __kfree_skb(opt_skb);
1667 opt_skb = xchg(&np->pktoptions, NULL);
1676 static int tcp_v6_rcv(struct sk_buff *skb)
1682 if (skb->pkt_type != PACKET_HOST)
1686 * Count it even if it's bad.
1688 TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1690 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1695 if (th->doff < sizeof(struct tcphdr)/4)
1697 if (!pskb_may_pull(skb, th->doff*4))
1700 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1704 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1705 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1706 skb->len - th->doff*4);
1707 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1708 TCP_SKB_CB(skb)->when = 0;
1709 TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1710 TCP_SKB_CB(skb)->sacked = 0;
1712 sk = __inet6_lookup(&tcp_hashinfo, &ipv6_hdr(skb)->saddr, th->source,
1713 &ipv6_hdr(skb)->daddr, ntohs(th->dest),
1720 if (sk->sk_state == TCP_TIME_WAIT)
1723 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1724 goto discard_and_relse;
1726 if (sk_filter(sk, skb))
1727 goto discard_and_relse;
1731 bh_lock_sock_nested(sk);
1733 if (!sock_owned_by_user(sk)) {
1734 #ifdef CONFIG_NET_DMA
1735 struct tcp_sock *tp = tcp_sk(sk);
1736 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1737 tp->ucopy.dma_chan = get_softnet_dma();
1738 if (tp->ucopy.dma_chan)
1739 ret = tcp_v6_do_rcv(sk, skb);
1743 if (!tcp_prequeue(sk, skb))
1744 ret = tcp_v6_do_rcv(sk, skb);
1747 sk_add_backlog(sk, skb);
1751 return ret ? -1 : 0;
1754 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1757 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1759 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1761 tcp_v6_send_reset(NULL, skb);
1778 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1779 inet_twsk_put(inet_twsk(sk));
1783 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1784 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1785 inet_twsk_put(inet_twsk(sk));
1789 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1794 sk2 = inet6_lookup_listener(&tcp_hashinfo,
1795 &ipv6_hdr(skb)->daddr,
1796 ntohs(th->dest), inet6_iif(skb));
1798 struct inet_timewait_sock *tw = inet_twsk(sk);
1799 inet_twsk_deschedule(tw, &tcp_death_row);
1804 /* Fall through to ACK */
1807 tcp_v6_timewait_ack(sk, skb);
1811 case TCP_TW_SUCCESS:;
1816 static int tcp_v6_remember_stamp(struct sock *sk)
1818 /* Alas, not yet... */
1822 static struct inet_connection_sock_af_ops ipv6_specific = {
1823 .queue_xmit = inet6_csk_xmit,
1824 .send_check = tcp_v6_send_check,
1825 .rebuild_header = inet6_sk_rebuild_header,
1826 .conn_request = tcp_v6_conn_request,
1827 .syn_recv_sock = tcp_v6_syn_recv_sock,
1828 .remember_stamp = tcp_v6_remember_stamp,
1829 .net_header_len = sizeof(struct ipv6hdr),
1830 .setsockopt = ipv6_setsockopt,
1831 .getsockopt = ipv6_getsockopt,
1832 .addr2sockaddr = inet6_csk_addr2sockaddr,
1833 .sockaddr_len = sizeof(struct sockaddr_in6),
1834 #ifdef CONFIG_COMPAT
1835 .compat_setsockopt = compat_ipv6_setsockopt,
1836 .compat_getsockopt = compat_ipv6_getsockopt,
1840 #ifdef CONFIG_TCP_MD5SIG
1841 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1842 .md5_lookup = tcp_v6_md5_lookup,
1843 .calc_md5_hash = tcp_v6_calc_md5_hash,
1844 .md5_add = tcp_v6_md5_add_func,
1845 .md5_parse = tcp_v6_parse_md5_keys,
1850 * TCP over IPv4 via INET6 API
1853 static struct inet_connection_sock_af_ops ipv6_mapped = {
1854 .queue_xmit = ip_queue_xmit,
1855 .send_check = tcp_v4_send_check,
1856 .rebuild_header = inet_sk_rebuild_header,
1857 .conn_request = tcp_v6_conn_request,
1858 .syn_recv_sock = tcp_v6_syn_recv_sock,
1859 .remember_stamp = tcp_v4_remember_stamp,
1860 .net_header_len = sizeof(struct iphdr),
1861 .setsockopt = ipv6_setsockopt,
1862 .getsockopt = ipv6_getsockopt,
1863 .addr2sockaddr = inet6_csk_addr2sockaddr,
1864 .sockaddr_len = sizeof(struct sockaddr_in6),
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_mapped_specific = {
1873 .md5_lookup = tcp_v4_md5_lookup,
1874 .calc_md5_hash = tcp_v4_calc_md5_hash,
1875 .md5_add = tcp_v6_md5_add_func,
1876 .md5_parse = tcp_v6_parse_md5_keys,
1880 /* NOTE: A lot of things set to zero explicitly by call to
1881 * sk_alloc() so need not be done here.
1883 static int tcp_v6_init_sock(struct sock *sk)
1885 struct inet_connection_sock *icsk = inet_csk(sk);
1886 struct tcp_sock *tp = tcp_sk(sk);
1888 skb_queue_head_init(&tp->out_of_order_queue);
1889 tcp_init_xmit_timers(sk);
1890 tcp_prequeue_init(tp);
1892 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1893 tp->mdev = TCP_TIMEOUT_INIT;
1895 /* So many TCP implementations out there (incorrectly) count the
1896 * initial SYN frame in their delayed-ACK and congestion control
1897 * algorithms that we must have the following bandaid to talk
1898 * efficiently to them. -DaveM
1902 /* See draft-stevens-tcpca-spec-01 for discussion of the
1903 * initialization of these values.
1905 tp->snd_ssthresh = 0x7fffffff;
1906 tp->snd_cwnd_clamp = ~0;
1907 tp->mss_cache = 536;
1909 tp->reordering = sysctl_tcp_reordering;
1911 sk->sk_state = TCP_CLOSE;
1913 icsk->icsk_af_ops = &ipv6_specific;
1914 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1915 icsk->icsk_sync_mss = tcp_sync_mss;
1916 sk->sk_write_space = sk_stream_write_space;
1917 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1919 #ifdef CONFIG_TCP_MD5SIG
1920 tp->af_specific = &tcp_sock_ipv6_specific;
1923 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1924 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1926 atomic_inc(&tcp_sockets_allocated);
1931 static int tcp_v6_destroy_sock(struct sock *sk)
1933 #ifdef CONFIG_TCP_MD5SIG
1934 /* Clean up the MD5 key list */
1935 if (tcp_sk(sk)->md5sig_info)
1936 tcp_v6_clear_md5_list(sk);
1938 tcp_v4_destroy_sock(sk);
1939 return inet6_destroy_sock(sk);
1942 #ifdef CONFIG_PROC_FS
1943 /* Proc filesystem TCPv6 sock list dumping. */
1944 static void get_openreq6(struct seq_file *seq,
1945 struct sock *sk, struct request_sock *req, int i, int uid)
1947 int ttd = req->expires - jiffies;
1948 struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1949 struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1955 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1956 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1958 src->s6_addr32[0], src->s6_addr32[1],
1959 src->s6_addr32[2], src->s6_addr32[3],
1960 ntohs(inet_sk(sk)->sport),
1961 dest->s6_addr32[0], dest->s6_addr32[1],
1962 dest->s6_addr32[2], dest->s6_addr32[3],
1963 ntohs(inet_rsk(req)->rmt_port),
1965 0,0, /* could print option size, but that is af dependent. */
1966 1, /* timers active (only the expire timer) */
1967 jiffies_to_clock_t(ttd),
1970 0, /* non standard timer */
1971 0, /* open_requests have no inode */
1975 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1977 struct in6_addr *dest, *src;
1980 unsigned long timer_expires;
1981 struct inet_sock *inet = inet_sk(sp);
1982 struct tcp_sock *tp = tcp_sk(sp);
1983 const struct inet_connection_sock *icsk = inet_csk(sp);
1984 struct ipv6_pinfo *np = inet6_sk(sp);
1987 src = &np->rcv_saddr;
1988 destp = ntohs(inet->dport);
1989 srcp = ntohs(inet->sport);
1991 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1993 timer_expires = icsk->icsk_timeout;
1994 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1996 timer_expires = icsk->icsk_timeout;
1997 } else if (timer_pending(&sp->sk_timer)) {
1999 timer_expires = sp->sk_timer.expires;
2002 timer_expires = jiffies;
2006 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2007 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
2009 src->s6_addr32[0], src->s6_addr32[1],
2010 src->s6_addr32[2], src->s6_addr32[3], srcp,
2011 dest->s6_addr32[0], dest->s6_addr32[1],
2012 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2014 tp->write_seq-tp->snd_una,
2015 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2017 jiffies_to_clock_t(timer_expires - jiffies),
2018 icsk->icsk_retransmits,
2020 icsk->icsk_probes_out,
2022 atomic_read(&sp->sk_refcnt), sp,
2025 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2026 tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2030 static void get_timewait6_sock(struct seq_file *seq,
2031 struct inet_timewait_sock *tw, int i)
2033 struct in6_addr *dest, *src;
2035 struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2036 int ttd = tw->tw_ttd - jiffies;
2041 dest = &tw6->tw_v6_daddr;
2042 src = &tw6->tw_v6_rcv_saddr;
2043 destp = ntohs(tw->tw_dport);
2044 srcp = ntohs(tw->tw_sport);
2047 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2048 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2050 src->s6_addr32[0], src->s6_addr32[1],
2051 src->s6_addr32[2], src->s6_addr32[3], srcp,
2052 dest->s6_addr32[0], dest->s6_addr32[1],
2053 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2054 tw->tw_substate, 0, 0,
2055 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2056 atomic_read(&tw->tw_refcnt), tw);
2059 static int tcp6_seq_show(struct seq_file *seq, void *v)
2061 struct tcp_iter_state *st;
2063 if (v == SEQ_START_TOKEN) {
2068 "st tx_queue rx_queue tr tm->when retrnsmt"
2069 " uid timeout inode\n");
2074 switch (st->state) {
2075 case TCP_SEQ_STATE_LISTENING:
2076 case TCP_SEQ_STATE_ESTABLISHED:
2077 get_tcp6_sock(seq, v, st->num);
2079 case TCP_SEQ_STATE_OPENREQ:
2080 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2082 case TCP_SEQ_STATE_TIME_WAIT:
2083 get_timewait6_sock(seq, v, st->num);
2090 static struct file_operations tcp6_seq_fops;
2091 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2092 .owner = THIS_MODULE,
2095 .seq_show = tcp6_seq_show,
2096 .seq_fops = &tcp6_seq_fops,
2099 int __init tcp6_proc_init(void)
2101 return tcp_proc_register(&tcp6_seq_afinfo);
2104 void tcp6_proc_exit(void)
2106 tcp_proc_unregister(&tcp6_seq_afinfo);
2110 DEFINE_PROTO_INUSE(tcpv6)
2112 struct proto tcpv6_prot = {
2114 .owner = THIS_MODULE,
2116 .connect = tcp_v6_connect,
2117 .disconnect = tcp_disconnect,
2118 .accept = inet_csk_accept,
2120 .init = tcp_v6_init_sock,
2121 .destroy = tcp_v6_destroy_sock,
2122 .shutdown = tcp_shutdown,
2123 .setsockopt = tcp_setsockopt,
2124 .getsockopt = tcp_getsockopt,
2125 .recvmsg = tcp_recvmsg,
2126 .backlog_rcv = tcp_v6_do_rcv,
2127 .hash = tcp_v6_hash,
2128 .unhash = tcp_unhash,
2129 .get_port = tcp_v6_get_port,
2130 .enter_memory_pressure = tcp_enter_memory_pressure,
2131 .sockets_allocated = &tcp_sockets_allocated,
2132 .memory_allocated = &tcp_memory_allocated,
2133 .memory_pressure = &tcp_memory_pressure,
2134 .orphan_count = &tcp_orphan_count,
2135 .sysctl_mem = sysctl_tcp_mem,
2136 .sysctl_wmem = sysctl_tcp_wmem,
2137 .sysctl_rmem = sysctl_tcp_rmem,
2138 .max_header = MAX_TCP_HEADER,
2139 .obj_size = sizeof(struct tcp6_sock),
2140 .twsk_prot = &tcp6_timewait_sock_ops,
2141 .rsk_prot = &tcp6_request_sock_ops,
2142 #ifdef CONFIG_COMPAT
2143 .compat_setsockopt = compat_tcp_setsockopt,
2144 .compat_getsockopt = compat_tcp_getsockopt,
2146 REF_PROTO_INUSE(tcpv6)
2149 static struct inet6_protocol tcpv6_protocol = {
2150 .handler = tcp_v6_rcv,
2151 .err_handler = tcp_v6_err,
2152 .gso_send_check = tcp_v6_gso_send_check,
2153 .gso_segment = tcp_tso_segment,
2154 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2157 static struct inet_protosw tcpv6_protosw = {
2158 .type = SOCK_STREAM,
2159 .protocol = IPPROTO_TCP,
2160 .prot = &tcpv6_prot,
2161 .ops = &inet6_stream_ops,
2164 .flags = INET_PROTOSW_PERMANENT |
2168 void __init tcpv6_init(void)
2170 /* register inet6 protocol */
2171 if (inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP) < 0)
2172 printk(KERN_ERR "tcpv6_init: Could not register protocol\n");
2173 inet6_register_protosw(&tcpv6_protosw);
2175 if (inet_csk_ctl_sock_create(&tcp6_socket, PF_INET6, SOCK_RAW,
2177 panic("Failed to create the TCPv6 control socket.\n");