3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
14 * Hideaki YOSHIFUJI : sin6_scope_id support
15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
17 * a single port at the same time.
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
26 #include <linux/module.h>
27 #include <linux/errno.h>
28 #include <linux/types.h>
29 #include <linux/socket.h>
30 #include <linux/sockios.h>
31 #include <linux/net.h>
32 #include <linux/jiffies.h>
34 #include <linux/in6.h>
35 #include <linux/netdevice.h>
36 #include <linux/init.h>
37 #include <linux/jhash.h>
38 #include <linux/ipsec.h>
39 #include <linux/times.h>
41 #include <linux/ipv6.h>
42 #include <linux/icmpv6.h>
43 #include <linux/random.h>
46 #include <net/ndisc.h>
47 #include <net/inet6_hashtables.h>
48 #include <net/inet6_connection_sock.h>
50 #include <net/transp_v6.h>
51 #include <net/addrconf.h>
52 #include <net/ip6_route.h>
53 #include <net/ip6_checksum.h>
54 #include <net/inet_ecn.h>
55 #include <net/protocol.h>
58 #include <net/dsfield.h>
59 #include <net/timewait_sock.h>
60 #include <net/netdma.h>
61 #include <net/inet_common.h>
63 #include <asm/uaccess.h>
65 #include <linux/proc_fs.h>
66 #include <linux/seq_file.h>
68 #include <linux/crypto.h>
69 #include <linux/scatterlist.h>
71 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
72 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req);
74 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
76 static struct inet_connection_sock_af_ops ipv6_mapped;
77 static struct inet_connection_sock_af_ops ipv6_specific;
78 #ifdef CONFIG_TCP_MD5SIG
79 static struct tcp_sock_af_ops tcp_sock_ipv6_specific;
80 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
82 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
83 struct in6_addr *addr)
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;
326 struct net *net = dev_net(skb->dev);
328 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
329 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
332 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
336 if (sk->sk_state == TCP_TIME_WAIT) {
337 inet_twsk_put(inet_twsk(sk));
342 if (sock_owned_by_user(sk))
343 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
345 if (sk->sk_state == TCP_CLOSE)
349 seq = ntohl(th->seq);
350 if (sk->sk_state != TCP_LISTEN &&
351 !between(seq, tp->snd_una, tp->snd_nxt)) {
352 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
358 if (type == ICMPV6_PKT_TOOBIG) {
359 struct dst_entry *dst = NULL;
361 if (sock_owned_by_user(sk))
363 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
366 /* icmp should have updated the destination cache entry */
367 dst = __sk_dst_check(sk, np->dst_cookie);
370 struct inet_sock *inet = inet_sk(sk);
373 /* BUGGG_FUTURE: Again, it is not clear how
374 to handle rthdr case. Ignore this complexity
377 memset(&fl, 0, sizeof(fl));
378 fl.proto = IPPROTO_TCP;
379 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
380 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
381 fl.oif = sk->sk_bound_dev_if;
382 fl.fl_ip_dport = inet->dport;
383 fl.fl_ip_sport = inet->sport;
384 security_skb_classify_flow(skb, &fl);
386 if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
387 sk->sk_err_soft = -err;
391 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
392 sk->sk_err_soft = -err;
399 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
400 tcp_sync_mss(sk, dst_mtu(dst));
401 tcp_simple_retransmit(sk);
402 } /* else let the usual retransmit timer handle it */
407 icmpv6_err_convert(type, code, &err);
409 /* Might be for an request_sock */
410 switch (sk->sk_state) {
411 struct request_sock *req, **prev;
413 if (sock_owned_by_user(sk))
416 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
417 &hdr->saddr, inet6_iif(skb));
421 /* ICMPs are not backlogged, hence we cannot get
422 * an established socket here.
424 BUG_TRAP(req->sk == NULL);
426 if (seq != tcp_rsk(req)->snt_isn) {
427 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
431 inet_csk_reqsk_queue_drop(sk, req, prev);
435 case TCP_SYN_RECV: /* Cannot happen.
436 It can, it SYNs are crossed. --ANK */
437 if (!sock_owned_by_user(sk)) {
439 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
443 sk->sk_err_soft = err;
447 if (!sock_owned_by_user(sk) && np->recverr) {
449 sk->sk_error_report(sk);
451 sk->sk_err_soft = err;
459 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req)
461 struct inet6_request_sock *treq = inet6_rsk(req);
462 struct ipv6_pinfo *np = inet6_sk(sk);
463 struct sk_buff * skb;
464 struct ipv6_txoptions *opt = NULL;
465 struct in6_addr * final_p = NULL, final;
467 struct dst_entry *dst;
470 memset(&fl, 0, sizeof(fl));
471 fl.proto = IPPROTO_TCP;
472 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
473 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
474 fl.fl6_flowlabel = 0;
476 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
477 fl.fl_ip_sport = inet_sk(sk)->sport;
478 security_req_classify_flow(req, &fl);
481 if (opt && opt->srcrt) {
482 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
483 ipv6_addr_copy(&final, &fl.fl6_dst);
484 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
488 err = ip6_dst_lookup(sk, &dst, &fl);
492 ipv6_addr_copy(&fl.fl6_dst, final_p);
493 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
496 skb = tcp_make_synack(sk, dst, req);
498 struct tcphdr *th = tcp_hdr(skb);
500 th->check = tcp_v6_check(th, skb->len,
501 &treq->loc_addr, &treq->rmt_addr,
502 csum_partial((char *)th, skb->len, skb->csum));
504 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
505 err = ip6_xmit(sk, skb, &fl, opt, 0);
506 err = net_xmit_eval(err);
510 if (opt && opt != np->opt)
511 sock_kfree_s(sk, opt, opt->tot_len);
516 static inline void syn_flood_warning(struct sk_buff *skb)
518 #ifdef CONFIG_SYN_COOKIES
519 if (sysctl_tcp_syncookies)
521 "TCPv6: Possible SYN flooding on port %d. "
522 "Sending cookies.\n", ntohs(tcp_hdr(skb)->dest));
526 "TCPv6: Possible SYN flooding on port %d. "
527 "Dropping request.\n", ntohs(tcp_hdr(skb)->dest));
530 static void tcp_v6_reqsk_destructor(struct request_sock *req)
532 if (inet6_rsk(req)->pktopts)
533 kfree_skb(inet6_rsk(req)->pktopts);
536 #ifdef CONFIG_TCP_MD5SIG
537 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
538 struct in6_addr *addr)
540 struct tcp_sock *tp = tcp_sk(sk);
545 if (!tp->md5sig_info || !tp->md5sig_info->entries6)
548 for (i = 0; i < tp->md5sig_info->entries6; i++) {
549 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, addr))
550 return &tp->md5sig_info->keys6[i].base;
555 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
556 struct sock *addr_sk)
558 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
561 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
562 struct request_sock *req)
564 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
567 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
568 char *newkey, u8 newkeylen)
570 /* Add key to the list */
571 struct tcp_md5sig_key *key;
572 struct tcp_sock *tp = tcp_sk(sk);
573 struct tcp6_md5sig_key *keys;
575 key = tcp_v6_md5_do_lookup(sk, peer);
577 /* modify existing entry - just update that one */
580 key->keylen = newkeylen;
582 /* reallocate new list if current one is full. */
583 if (!tp->md5sig_info) {
584 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
585 if (!tp->md5sig_info) {
589 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
591 if (tcp_alloc_md5sig_pool() == NULL) {
595 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
596 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
597 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
600 tcp_free_md5sig_pool();
605 if (tp->md5sig_info->entries6)
606 memmove(keys, tp->md5sig_info->keys6,
607 (sizeof (tp->md5sig_info->keys6[0]) *
608 tp->md5sig_info->entries6));
610 kfree(tp->md5sig_info->keys6);
611 tp->md5sig_info->keys6 = keys;
612 tp->md5sig_info->alloced6++;
615 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
617 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
618 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
620 tp->md5sig_info->entries6++;
625 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
626 u8 *newkey, __u8 newkeylen)
628 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
632 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
634 struct tcp_sock *tp = tcp_sk(sk);
637 for (i = 0; i < tp->md5sig_info->entries6; i++) {
638 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) {
640 kfree(tp->md5sig_info->keys6[i].base.key);
641 tp->md5sig_info->entries6--;
643 if (tp->md5sig_info->entries6 == 0) {
644 kfree(tp->md5sig_info->keys6);
645 tp->md5sig_info->keys6 = NULL;
646 tp->md5sig_info->alloced6 = 0;
648 /* shrink the database */
649 if (tp->md5sig_info->entries6 != i)
650 memmove(&tp->md5sig_info->keys6[i],
651 &tp->md5sig_info->keys6[i+1],
652 (tp->md5sig_info->entries6 - i)
653 * sizeof (tp->md5sig_info->keys6[0]));
655 tcp_free_md5sig_pool();
662 static void tcp_v6_clear_md5_list (struct sock *sk)
664 struct tcp_sock *tp = tcp_sk(sk);
667 if (tp->md5sig_info->entries6) {
668 for (i = 0; i < tp->md5sig_info->entries6; i++)
669 kfree(tp->md5sig_info->keys6[i].base.key);
670 tp->md5sig_info->entries6 = 0;
671 tcp_free_md5sig_pool();
674 kfree(tp->md5sig_info->keys6);
675 tp->md5sig_info->keys6 = NULL;
676 tp->md5sig_info->alloced6 = 0;
678 if (tp->md5sig_info->entries4) {
679 for (i = 0; i < tp->md5sig_info->entries4; i++)
680 kfree(tp->md5sig_info->keys4[i].base.key);
681 tp->md5sig_info->entries4 = 0;
682 tcp_free_md5sig_pool();
685 kfree(tp->md5sig_info->keys4);
686 tp->md5sig_info->keys4 = NULL;
687 tp->md5sig_info->alloced4 = 0;
690 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
693 struct tcp_md5sig cmd;
694 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
697 if (optlen < sizeof(cmd))
700 if (copy_from_user(&cmd, optval, sizeof(cmd)))
703 if (sin6->sin6_family != AF_INET6)
706 if (!cmd.tcpm_keylen) {
707 if (!tcp_sk(sk)->md5sig_info)
709 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
710 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
711 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
714 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
717 if (!tcp_sk(sk)->md5sig_info) {
718 struct tcp_sock *tp = tcp_sk(sk);
719 struct tcp_md5sig_info *p;
721 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
726 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
729 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
732 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
733 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
734 newkey, cmd.tcpm_keylen);
736 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
739 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
740 struct in6_addr *daddr,
741 struct in6_addr *saddr, int nbytes)
743 struct tcp6_pseudohdr *bp;
744 struct scatterlist sg;
746 bp = &hp->md5_blk.ip6;
747 /* 1. TCP pseudo-header (RFC2460) */
748 ipv6_addr_copy(&bp->saddr, saddr);
749 ipv6_addr_copy(&bp->daddr, daddr);
750 bp->protocol = cpu_to_be32(IPPROTO_TCP);
751 bp->len = cpu_to_be16(nbytes);
753 sg_init_one(&sg, bp, sizeof(*bp));
754 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
757 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
758 struct in6_addr *daddr, struct in6_addr *saddr,
761 struct tcp_md5sig_pool *hp;
762 struct hash_desc *desc;
764 hp = tcp_get_md5sig_pool();
766 goto clear_hash_noput;
767 desc = &hp->md5_desc;
769 if (crypto_hash_init(desc))
771 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
773 if (tcp_md5_hash_header(hp, th))
775 if (tcp_md5_hash_key(hp, key))
777 if (crypto_hash_final(desc, md5_hash))
780 tcp_put_md5sig_pool();
784 tcp_put_md5sig_pool();
786 memset(md5_hash, 0, 16);
790 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
791 struct sock *sk, struct request_sock *req,
794 struct in6_addr *saddr, *daddr;
795 struct tcp_md5sig_pool *hp;
796 struct hash_desc *desc;
797 struct tcphdr *th = tcp_hdr(skb);
800 saddr = &inet6_sk(sk)->saddr;
801 daddr = &inet6_sk(sk)->daddr;
803 saddr = &inet6_rsk(req)->loc_addr;
804 daddr = &inet6_rsk(req)->rmt_addr;
806 struct ipv6hdr *ip6h = ipv6_hdr(skb);
807 saddr = &ip6h->saddr;
808 daddr = &ip6h->daddr;
811 hp = tcp_get_md5sig_pool();
813 goto clear_hash_noput;
814 desc = &hp->md5_desc;
816 if (crypto_hash_init(desc))
819 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
821 if (tcp_md5_hash_header(hp, th))
823 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
825 if (tcp_md5_hash_key(hp, key))
827 if (crypto_hash_final(desc, md5_hash))
830 tcp_put_md5sig_pool();
834 tcp_put_md5sig_pool();
836 memset(md5_hash, 0, 16);
840 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
842 __u8 *hash_location = NULL;
843 struct tcp_md5sig_key *hash_expected;
844 struct ipv6hdr *ip6h = ipv6_hdr(skb);
845 struct tcphdr *th = tcp_hdr(skb);
849 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
850 hash_location = tcp_parse_md5sig_option(th);
852 /* do we have a hash as expected? */
853 if (!hash_expected) {
856 if (net_ratelimit()) {
857 printk(KERN_INFO "MD5 Hash NOT expected but found "
858 "(" NIP6_FMT ", %u)->"
859 "(" NIP6_FMT ", %u)\n",
860 NIP6(ip6h->saddr), ntohs(th->source),
861 NIP6(ip6h->daddr), ntohs(th->dest));
866 if (!hash_location) {
867 if (net_ratelimit()) {
868 printk(KERN_INFO "MD5 Hash expected but NOT found "
869 "(" NIP6_FMT ", %u)->"
870 "(" NIP6_FMT ", %u)\n",
871 NIP6(ip6h->saddr), ntohs(th->source),
872 NIP6(ip6h->daddr), ntohs(th->dest));
877 /* check the signature */
878 genhash = tcp_v6_md5_hash_skb(newhash,
882 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
883 if (net_ratelimit()) {
884 printk(KERN_INFO "MD5 Hash %s for "
885 "(" NIP6_FMT ", %u)->"
886 "(" NIP6_FMT ", %u)\n",
887 genhash ? "failed" : "mismatch",
888 NIP6(ip6h->saddr), ntohs(th->source),
889 NIP6(ip6h->daddr), ntohs(th->dest));
897 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
899 .obj_size = sizeof(struct tcp6_request_sock),
900 .rtx_syn_ack = tcp_v6_send_synack,
901 .send_ack = tcp_v6_reqsk_send_ack,
902 .destructor = tcp_v6_reqsk_destructor,
903 .send_reset = tcp_v6_send_reset
906 #ifdef CONFIG_TCP_MD5SIG
907 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
908 .md5_lookup = tcp_v6_reqsk_md5_lookup,
912 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
913 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
914 .twsk_unique = tcp_twsk_unique,
915 .twsk_destructor= tcp_twsk_destructor,
918 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
920 struct ipv6_pinfo *np = inet6_sk(sk);
921 struct tcphdr *th = tcp_hdr(skb);
923 if (skb->ip_summed == CHECKSUM_PARTIAL) {
924 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 0);
925 skb->csum_start = skb_transport_header(skb) - skb->head;
926 skb->csum_offset = offsetof(struct tcphdr, check);
928 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
929 csum_partial((char *)th, th->doff<<2,
934 static int tcp_v6_gso_send_check(struct sk_buff *skb)
936 struct ipv6hdr *ipv6h;
939 if (!pskb_may_pull(skb, sizeof(*th)))
942 ipv6h = ipv6_hdr(skb);
946 th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
948 skb->csum_start = skb_transport_header(skb) - skb->head;
949 skb->csum_offset = offsetof(struct tcphdr, check);
950 skb->ip_summed = CHECKSUM_PARTIAL;
954 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
956 struct tcphdr *th = tcp_hdr(skb), *t1;
957 struct sk_buff *buff;
959 struct net *net = dev_net(skb->dst->dev);
960 struct sock *ctl_sk = net->ipv6.tcp_sk;
961 unsigned int tot_len = sizeof(*th);
962 #ifdef CONFIG_TCP_MD5SIG
963 struct tcp_md5sig_key *key;
969 if (!ipv6_unicast_destination(skb))
972 #ifdef CONFIG_TCP_MD5SIG
974 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
979 tot_len += TCPOLEN_MD5SIG_ALIGNED;
983 * We need to grab some memory, and put together an RST,
984 * and then put it into the queue to be sent.
987 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
992 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
994 t1 = (struct tcphdr *) skb_push(buff, tot_len);
996 /* Swap the send and the receive. */
997 memset(t1, 0, sizeof(*t1));
998 t1->dest = th->source;
999 t1->source = th->dest;
1000 t1->doff = tot_len / 4;
1004 t1->seq = th->ack_seq;
1007 t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
1008 + skb->len - (th->doff<<2));
1011 #ifdef CONFIG_TCP_MD5SIG
1013 __be32 *opt = (__be32*)(t1 + 1);
1014 opt[0] = htonl((TCPOPT_NOP << 24) |
1015 (TCPOPT_NOP << 16) |
1016 (TCPOPT_MD5SIG << 8) |
1018 tcp_v6_md5_hash_hdr((__u8 *)&opt[1], key,
1019 &ipv6_hdr(skb)->daddr,
1020 &ipv6_hdr(skb)->saddr, t1);
1024 buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1026 memset(&fl, 0, sizeof(fl));
1027 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1028 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1030 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1031 sizeof(*t1), IPPROTO_TCP,
1034 fl.proto = IPPROTO_TCP;
1035 fl.oif = inet6_iif(skb);
1036 fl.fl_ip_dport = t1->dest;
1037 fl.fl_ip_sport = t1->source;
1038 security_skb_classify_flow(skb, &fl);
1040 /* Pass a socket to ip6_dst_lookup either it is for RST
1041 * Underlying function will use this to retrieve the network
1044 if (!ip6_dst_lookup(ctl_sk, &buff->dst, &fl)) {
1046 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1047 ip6_xmit(ctl_sk, buff, &fl, NULL, 0);
1048 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1049 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1057 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1058 struct tcp_md5sig_key *key)
1060 struct tcphdr *th = tcp_hdr(skb), *t1;
1061 struct sk_buff *buff;
1063 struct net *net = dev_net(skb->dev);
1064 struct sock *ctl_sk = net->ipv6.tcp_sk;
1065 unsigned int tot_len = sizeof(struct tcphdr);
1069 tot_len += TCPOLEN_TSTAMP_ALIGNED;
1070 #ifdef CONFIG_TCP_MD5SIG
1072 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1075 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1080 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1082 t1 = (struct tcphdr *) skb_push(buff,tot_len);
1084 /* Swap the send and the receive. */
1085 memset(t1, 0, sizeof(*t1));
1086 t1->dest = th->source;
1087 t1->source = th->dest;
1088 t1->doff = tot_len/4;
1089 t1->seq = htonl(seq);
1090 t1->ack_seq = htonl(ack);
1092 t1->window = htons(win);
1094 topt = (__be32 *)(t1 + 1);
1097 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1098 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1099 *topt++ = htonl(tcp_time_stamp);
1103 #ifdef CONFIG_TCP_MD5SIG
1105 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1106 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1107 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
1108 &ipv6_hdr(skb)->daddr,
1109 &ipv6_hdr(skb)->saddr, t1);
1113 buff->csum = csum_partial((char *)t1, tot_len, 0);
1115 memset(&fl, 0, sizeof(fl));
1116 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1117 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1119 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1120 tot_len, IPPROTO_TCP,
1123 fl.proto = IPPROTO_TCP;
1124 fl.oif = inet6_iif(skb);
1125 fl.fl_ip_dport = t1->dest;
1126 fl.fl_ip_sport = t1->source;
1127 security_skb_classify_flow(skb, &fl);
1129 if (!ip6_dst_lookup(ctl_sk, &buff->dst, &fl)) {
1130 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1131 ip6_xmit(ctl_sk, buff, &fl, NULL, 0);
1132 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1140 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1142 struct inet_timewait_sock *tw = inet_twsk(sk);
1143 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1145 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1146 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1147 tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw));
1152 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
1154 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1155 tcp_v6_md5_do_lookup(skb->sk, &ipv6_hdr(skb)->daddr));
1159 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1161 struct request_sock *req, **prev;
1162 const struct tcphdr *th = tcp_hdr(skb);
1165 /* Find possible connection requests. */
1166 req = inet6_csk_search_req(sk, &prev, th->source,
1167 &ipv6_hdr(skb)->saddr,
1168 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1170 return tcp_check_req(sk, skb, req, prev);
1172 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1173 &ipv6_hdr(skb)->saddr, th->source,
1174 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1177 if (nsk->sk_state != TCP_TIME_WAIT) {
1181 inet_twsk_put(inet_twsk(nsk));
1185 #ifdef CONFIG_SYN_COOKIES
1186 if (!th->rst && !th->syn && th->ack)
1187 sk = cookie_v6_check(sk, skb);
1192 /* FIXME: this is substantially similar to the ipv4 code.
1193 * Can some kind of merge be done? -- erics
1195 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1197 struct inet6_request_sock *treq;
1198 struct ipv6_pinfo *np = inet6_sk(sk);
1199 struct tcp_options_received tmp_opt;
1200 struct tcp_sock *tp = tcp_sk(sk);
1201 struct request_sock *req = NULL;
1202 __u32 isn = TCP_SKB_CB(skb)->when;
1203 #ifdef CONFIG_SYN_COOKIES
1204 int want_cookie = 0;
1206 #define want_cookie 0
1209 if (skb->protocol == htons(ETH_P_IP))
1210 return tcp_v4_conn_request(sk, skb);
1212 if (!ipv6_unicast_destination(skb))
1215 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1216 if (net_ratelimit())
1217 syn_flood_warning(skb);
1218 #ifdef CONFIG_SYN_COOKIES
1219 if (sysctl_tcp_syncookies)
1226 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1229 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1233 #ifdef CONFIG_TCP_MD5SIG
1234 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1237 tcp_clear_options(&tmp_opt);
1238 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1239 tmp_opt.user_mss = tp->rx_opt.user_mss;
1241 tcp_parse_options(skb, &tmp_opt, 0);
1243 if (want_cookie && !tmp_opt.saw_tstamp)
1244 tcp_clear_options(&tmp_opt);
1246 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1247 tcp_openreq_init(req, &tmp_opt, skb);
1249 treq = inet6_rsk(req);
1250 ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1251 ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1253 TCP_ECN_create_request(req, tcp_hdr(skb));
1256 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1257 req->cookie_ts = tmp_opt.tstamp_ok;
1259 if (ipv6_opt_accepted(sk, skb) ||
1260 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1261 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1262 atomic_inc(&skb->users);
1263 treq->pktopts = skb;
1265 treq->iif = sk->sk_bound_dev_if;
1267 /* So that link locals have meaning */
1268 if (!sk->sk_bound_dev_if &&
1269 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1270 treq->iif = inet6_iif(skb);
1272 isn = tcp_v6_init_sequence(skb);
1275 tcp_rsk(req)->snt_isn = isn;
1277 security_inet_conn_request(sk, skb, req);
1279 if (tcp_v6_send_synack(sk, req))
1283 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1291 return 0; /* don't send reset */
1294 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1295 struct request_sock *req,
1296 struct dst_entry *dst)
1298 struct inet6_request_sock *treq = inet6_rsk(req);
1299 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1300 struct tcp6_sock *newtcp6sk;
1301 struct inet_sock *newinet;
1302 struct tcp_sock *newtp;
1304 struct ipv6_txoptions *opt;
1305 #ifdef CONFIG_TCP_MD5SIG
1306 struct tcp_md5sig_key *key;
1309 if (skb->protocol == htons(ETH_P_IP)) {
1314 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1319 newtcp6sk = (struct tcp6_sock *)newsk;
1320 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1322 newinet = inet_sk(newsk);
1323 newnp = inet6_sk(newsk);
1324 newtp = tcp_sk(newsk);
1326 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1328 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1331 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1334 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1336 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1337 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1338 #ifdef CONFIG_TCP_MD5SIG
1339 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1342 newnp->pktoptions = NULL;
1344 newnp->mcast_oif = inet6_iif(skb);
1345 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1348 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1349 * here, tcp_create_openreq_child now does this for us, see the comment in
1350 * that function for the gory details. -acme
1353 /* It is tricky place. Until this moment IPv4 tcp
1354 worked with IPv6 icsk.icsk_af_ops.
1357 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1364 if (sk_acceptq_is_full(sk))
1368 struct in6_addr *final_p = NULL, final;
1371 memset(&fl, 0, sizeof(fl));
1372 fl.proto = IPPROTO_TCP;
1373 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1374 if (opt && opt->srcrt) {
1375 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1376 ipv6_addr_copy(&final, &fl.fl6_dst);
1377 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1380 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1381 fl.oif = sk->sk_bound_dev_if;
1382 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1383 fl.fl_ip_sport = inet_sk(sk)->sport;
1384 security_req_classify_flow(req, &fl);
1386 if (ip6_dst_lookup(sk, &dst, &fl))
1390 ipv6_addr_copy(&fl.fl6_dst, final_p);
1392 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1396 newsk = tcp_create_openreq_child(sk, req, skb);
1401 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1402 * count here, tcp_create_openreq_child now does this for us, see the
1403 * comment in that function for the gory details. -acme
1406 newsk->sk_gso_type = SKB_GSO_TCPV6;
1407 __ip6_dst_store(newsk, dst, NULL, NULL);
1409 newtcp6sk = (struct tcp6_sock *)newsk;
1410 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1412 newtp = tcp_sk(newsk);
1413 newinet = inet_sk(newsk);
1414 newnp = inet6_sk(newsk);
1416 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1418 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1419 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1420 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1421 newsk->sk_bound_dev_if = treq->iif;
1423 /* Now IPv6 options...
1425 First: no IPv4 options.
1427 newinet->opt = NULL;
1428 newnp->ipv6_fl_list = NULL;
1431 newnp->rxopt.all = np->rxopt.all;
1433 /* Clone pktoptions received with SYN */
1434 newnp->pktoptions = NULL;
1435 if (treq->pktopts != NULL) {
1436 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1437 kfree_skb(treq->pktopts);
1438 treq->pktopts = NULL;
1439 if (newnp->pktoptions)
1440 skb_set_owner_r(newnp->pktoptions, newsk);
1443 newnp->mcast_oif = inet6_iif(skb);
1444 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1446 /* Clone native IPv6 options from listening socket (if any)
1448 Yes, keeping reference count would be much more clever,
1449 but we make one more one thing there: reattach optmem
1453 newnp->opt = ipv6_dup_options(newsk, opt);
1455 sock_kfree_s(sk, opt, opt->tot_len);
1458 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1460 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1461 newnp->opt->opt_flen);
1463 tcp_mtup_init(newsk);
1464 tcp_sync_mss(newsk, dst_mtu(dst));
1465 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1466 tcp_initialize_rcv_mss(newsk);
1468 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1470 #ifdef CONFIG_TCP_MD5SIG
1471 /* Copy over the MD5 key from the original socket */
1472 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1473 /* We're using one, so create a matching key
1474 * on the newsk structure. If we fail to get
1475 * memory, then we end up not copying the key
1478 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1480 tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1481 newkey, key->keylen);
1485 __inet6_hash(newsk);
1486 __inet_inherit_port(sk, newsk);
1491 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1493 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1494 if (opt && opt != np->opt)
1495 sock_kfree_s(sk, opt, opt->tot_len);
1500 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1502 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1503 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1504 &ipv6_hdr(skb)->daddr, skb->csum)) {
1505 skb->ip_summed = CHECKSUM_UNNECESSARY;
1510 skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1511 &ipv6_hdr(skb)->saddr,
1512 &ipv6_hdr(skb)->daddr, 0));
1514 if (skb->len <= 76) {
1515 return __skb_checksum_complete(skb);
1520 /* The socket must have it's spinlock held when we get
1523 * We have a potential double-lock case here, so even when
1524 * doing backlog processing we use the BH locking scheme.
1525 * This is because we cannot sleep with the original spinlock
1528 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1530 struct ipv6_pinfo *np = inet6_sk(sk);
1531 struct tcp_sock *tp;
1532 struct sk_buff *opt_skb = NULL;
1534 /* Imagine: socket is IPv6. IPv4 packet arrives,
1535 goes to IPv4 receive handler and backlogged.
1536 From backlog it always goes here. Kerboom...
1537 Fortunately, tcp_rcv_established and rcv_established
1538 handle them correctly, but it is not case with
1539 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1542 if (skb->protocol == htons(ETH_P_IP))
1543 return tcp_v4_do_rcv(sk, skb);
1545 #ifdef CONFIG_TCP_MD5SIG
1546 if (tcp_v6_inbound_md5_hash (sk, skb))
1550 if (sk_filter(sk, skb))
1554 * socket locking is here for SMP purposes as backlog rcv
1555 * is currently called with bh processing disabled.
1558 /* Do Stevens' IPV6_PKTOPTIONS.
1560 Yes, guys, it is the only place in our code, where we
1561 may make it not affecting IPv4.
1562 The rest of code is protocol independent,
1563 and I do not like idea to uglify IPv4.
1565 Actually, all the idea behind IPV6_PKTOPTIONS
1566 looks not very well thought. For now we latch
1567 options, received in the last packet, enqueued
1568 by tcp. Feel free to propose better solution.
1572 opt_skb = skb_clone(skb, GFP_ATOMIC);
1574 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1575 TCP_CHECK_TIMER(sk);
1576 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1578 TCP_CHECK_TIMER(sk);
1580 goto ipv6_pktoptions;
1584 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1587 if (sk->sk_state == TCP_LISTEN) {
1588 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1593 * Queue it on the new socket if the new socket is active,
1594 * otherwise we just shortcircuit this and continue with
1598 if (tcp_child_process(sk, nsk, skb))
1601 __kfree_skb(opt_skb);
1606 TCP_CHECK_TIMER(sk);
1607 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1609 TCP_CHECK_TIMER(sk);
1611 goto ipv6_pktoptions;
1615 tcp_v6_send_reset(sk, skb);
1618 __kfree_skb(opt_skb);
1622 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1627 /* Do you ask, what is it?
1629 1. skb was enqueued by tcp.
1630 2. skb is added to tail of read queue, rather than out of order.
1631 3. socket is not in passive state.
1632 4. Finally, it really contains options, which user wants to receive.
1635 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1636 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1637 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1638 np->mcast_oif = inet6_iif(opt_skb);
1639 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1640 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1641 if (ipv6_opt_accepted(sk, opt_skb)) {
1642 skb_set_owner_r(opt_skb, sk);
1643 opt_skb = xchg(&np->pktoptions, opt_skb);
1645 __kfree_skb(opt_skb);
1646 opt_skb = xchg(&np->pktoptions, NULL);
1655 static int tcp_v6_rcv(struct sk_buff *skb)
1660 struct net *net = dev_net(skb->dev);
1662 if (skb->pkt_type != PACKET_HOST)
1666 * Count it even if it's bad.
1668 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1670 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1675 if (th->doff < sizeof(struct tcphdr)/4)
1677 if (!pskb_may_pull(skb, th->doff*4))
1680 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1684 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1685 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1686 skb->len - th->doff*4);
1687 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1688 TCP_SKB_CB(skb)->when = 0;
1689 TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1690 TCP_SKB_CB(skb)->sacked = 0;
1692 sk = __inet6_lookup(net, &tcp_hashinfo,
1693 &ipv6_hdr(skb)->saddr, th->source,
1694 &ipv6_hdr(skb)->daddr, ntohs(th->dest),
1701 if (sk->sk_state == TCP_TIME_WAIT)
1704 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1705 goto discard_and_relse;
1707 if (sk_filter(sk, skb))
1708 goto discard_and_relse;
1712 bh_lock_sock_nested(sk);
1714 if (!sock_owned_by_user(sk)) {
1715 #ifdef CONFIG_NET_DMA
1716 struct tcp_sock *tp = tcp_sk(sk);
1717 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1718 tp->ucopy.dma_chan = get_softnet_dma();
1719 if (tp->ucopy.dma_chan)
1720 ret = tcp_v6_do_rcv(sk, skb);
1724 if (!tcp_prequeue(sk, skb))
1725 ret = tcp_v6_do_rcv(sk, skb);
1728 sk_add_backlog(sk, skb);
1732 return ret ? -1 : 0;
1735 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1738 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1740 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1742 tcp_v6_send_reset(NULL, skb);
1759 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1760 inet_twsk_put(inet_twsk(sk));
1764 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1765 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1766 inet_twsk_put(inet_twsk(sk));
1770 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1775 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1776 &ipv6_hdr(skb)->daddr,
1777 ntohs(th->dest), inet6_iif(skb));
1779 struct inet_timewait_sock *tw = inet_twsk(sk);
1780 inet_twsk_deschedule(tw, &tcp_death_row);
1785 /* Fall through to ACK */
1788 tcp_v6_timewait_ack(sk, skb);
1792 case TCP_TW_SUCCESS:;
1797 static int tcp_v6_remember_stamp(struct sock *sk)
1799 /* Alas, not yet... */
1803 static struct inet_connection_sock_af_ops ipv6_specific = {
1804 .queue_xmit = inet6_csk_xmit,
1805 .send_check = tcp_v6_send_check,
1806 .rebuild_header = inet6_sk_rebuild_header,
1807 .conn_request = tcp_v6_conn_request,
1808 .syn_recv_sock = tcp_v6_syn_recv_sock,
1809 .remember_stamp = tcp_v6_remember_stamp,
1810 .net_header_len = sizeof(struct ipv6hdr),
1811 .setsockopt = ipv6_setsockopt,
1812 .getsockopt = ipv6_getsockopt,
1813 .addr2sockaddr = inet6_csk_addr2sockaddr,
1814 .sockaddr_len = sizeof(struct sockaddr_in6),
1815 .bind_conflict = inet6_csk_bind_conflict,
1816 #ifdef CONFIG_COMPAT
1817 .compat_setsockopt = compat_ipv6_setsockopt,
1818 .compat_getsockopt = compat_ipv6_getsockopt,
1822 #ifdef CONFIG_TCP_MD5SIG
1823 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1824 .md5_lookup = tcp_v6_md5_lookup,
1825 .calc_md5_hash = tcp_v6_md5_hash_skb,
1826 .md5_add = tcp_v6_md5_add_func,
1827 .md5_parse = tcp_v6_parse_md5_keys,
1832 * TCP over IPv4 via INET6 API
1835 static struct inet_connection_sock_af_ops ipv6_mapped = {
1836 .queue_xmit = ip_queue_xmit,
1837 .send_check = tcp_v4_send_check,
1838 .rebuild_header = inet_sk_rebuild_header,
1839 .conn_request = tcp_v6_conn_request,
1840 .syn_recv_sock = tcp_v6_syn_recv_sock,
1841 .remember_stamp = tcp_v4_remember_stamp,
1842 .net_header_len = sizeof(struct iphdr),
1843 .setsockopt = ipv6_setsockopt,
1844 .getsockopt = ipv6_getsockopt,
1845 .addr2sockaddr = inet6_csk_addr2sockaddr,
1846 .sockaddr_len = sizeof(struct sockaddr_in6),
1847 .bind_conflict = inet6_csk_bind_conflict,
1848 #ifdef CONFIG_COMPAT
1849 .compat_setsockopt = compat_ipv6_setsockopt,
1850 .compat_getsockopt = compat_ipv6_getsockopt,
1854 #ifdef CONFIG_TCP_MD5SIG
1855 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1856 .md5_lookup = tcp_v4_md5_lookup,
1857 .calc_md5_hash = tcp_v4_md5_hash_skb,
1858 .md5_add = tcp_v6_md5_add_func,
1859 .md5_parse = tcp_v6_parse_md5_keys,
1863 /* NOTE: A lot of things set to zero explicitly by call to
1864 * sk_alloc() so need not be done here.
1866 static int tcp_v6_init_sock(struct sock *sk)
1868 struct inet_connection_sock *icsk = inet_csk(sk);
1869 struct tcp_sock *tp = tcp_sk(sk);
1871 skb_queue_head_init(&tp->out_of_order_queue);
1872 tcp_init_xmit_timers(sk);
1873 tcp_prequeue_init(tp);
1875 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1876 tp->mdev = TCP_TIMEOUT_INIT;
1878 /* So many TCP implementations out there (incorrectly) count the
1879 * initial SYN frame in their delayed-ACK and congestion control
1880 * algorithms that we must have the following bandaid to talk
1881 * efficiently to them. -DaveM
1885 /* See draft-stevens-tcpca-spec-01 for discussion of the
1886 * initialization of these values.
1888 tp->snd_ssthresh = 0x7fffffff;
1889 tp->snd_cwnd_clamp = ~0;
1890 tp->mss_cache = 536;
1892 tp->reordering = sysctl_tcp_reordering;
1894 sk->sk_state = TCP_CLOSE;
1896 icsk->icsk_af_ops = &ipv6_specific;
1897 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1898 icsk->icsk_sync_mss = tcp_sync_mss;
1899 sk->sk_write_space = sk_stream_write_space;
1900 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1902 #ifdef CONFIG_TCP_MD5SIG
1903 tp->af_specific = &tcp_sock_ipv6_specific;
1906 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1907 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1909 atomic_inc(&tcp_sockets_allocated);
1914 static void tcp_v6_destroy_sock(struct sock *sk)
1916 #ifdef CONFIG_TCP_MD5SIG
1917 /* Clean up the MD5 key list */
1918 if (tcp_sk(sk)->md5sig_info)
1919 tcp_v6_clear_md5_list(sk);
1921 tcp_v4_destroy_sock(sk);
1922 inet6_destroy_sock(sk);
1925 #ifdef CONFIG_PROC_FS
1926 /* Proc filesystem TCPv6 sock list dumping. */
1927 static void get_openreq6(struct seq_file *seq,
1928 struct sock *sk, struct request_sock *req, int i, int uid)
1930 int ttd = req->expires - jiffies;
1931 struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1932 struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1938 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1939 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1941 src->s6_addr32[0], src->s6_addr32[1],
1942 src->s6_addr32[2], src->s6_addr32[3],
1943 ntohs(inet_sk(sk)->sport),
1944 dest->s6_addr32[0], dest->s6_addr32[1],
1945 dest->s6_addr32[2], dest->s6_addr32[3],
1946 ntohs(inet_rsk(req)->rmt_port),
1948 0,0, /* could print option size, but that is af dependent. */
1949 1, /* timers active (only the expire timer) */
1950 jiffies_to_clock_t(ttd),
1953 0, /* non standard timer */
1954 0, /* open_requests have no inode */
1958 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1960 struct in6_addr *dest, *src;
1963 unsigned long timer_expires;
1964 struct inet_sock *inet = inet_sk(sp);
1965 struct tcp_sock *tp = tcp_sk(sp);
1966 const struct inet_connection_sock *icsk = inet_csk(sp);
1967 struct ipv6_pinfo *np = inet6_sk(sp);
1970 src = &np->rcv_saddr;
1971 destp = ntohs(inet->dport);
1972 srcp = ntohs(inet->sport);
1974 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1976 timer_expires = icsk->icsk_timeout;
1977 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1979 timer_expires = icsk->icsk_timeout;
1980 } else if (timer_pending(&sp->sk_timer)) {
1982 timer_expires = sp->sk_timer.expires;
1985 timer_expires = jiffies;
1989 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1990 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %lu %lu %u %u %d\n",
1992 src->s6_addr32[0], src->s6_addr32[1],
1993 src->s6_addr32[2], src->s6_addr32[3], srcp,
1994 dest->s6_addr32[0], dest->s6_addr32[1],
1995 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1997 tp->write_seq-tp->snd_una,
1998 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2000 jiffies_to_clock_t(timer_expires - jiffies),
2001 icsk->icsk_retransmits,
2003 icsk->icsk_probes_out,
2005 atomic_read(&sp->sk_refcnt), sp,
2006 jiffies_to_clock_t(icsk->icsk_rto),
2007 jiffies_to_clock_t(icsk->icsk_ack.ato),
2008 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2009 tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2013 static void get_timewait6_sock(struct seq_file *seq,
2014 struct inet_timewait_sock *tw, int i)
2016 struct in6_addr *dest, *src;
2018 struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2019 int ttd = tw->tw_ttd - jiffies;
2024 dest = &tw6->tw_v6_daddr;
2025 src = &tw6->tw_v6_rcv_saddr;
2026 destp = ntohs(tw->tw_dport);
2027 srcp = ntohs(tw->tw_sport);
2030 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2031 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2033 src->s6_addr32[0], src->s6_addr32[1],
2034 src->s6_addr32[2], src->s6_addr32[3], srcp,
2035 dest->s6_addr32[0], dest->s6_addr32[1],
2036 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2037 tw->tw_substate, 0, 0,
2038 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2039 atomic_read(&tw->tw_refcnt), tw);
2042 static int tcp6_seq_show(struct seq_file *seq, void *v)
2044 struct tcp_iter_state *st;
2046 if (v == SEQ_START_TOKEN) {
2051 "st tx_queue rx_queue tr tm->when retrnsmt"
2052 " uid timeout inode\n");
2057 switch (st->state) {
2058 case TCP_SEQ_STATE_LISTENING:
2059 case TCP_SEQ_STATE_ESTABLISHED:
2060 get_tcp6_sock(seq, v, st->num);
2062 case TCP_SEQ_STATE_OPENREQ:
2063 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2065 case TCP_SEQ_STATE_TIME_WAIT:
2066 get_timewait6_sock(seq, v, st->num);
2073 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2077 .owner = THIS_MODULE,
2080 .show = tcp6_seq_show,
2084 int tcp6_proc_init(struct net *net)
2086 return tcp_proc_register(net, &tcp6_seq_afinfo);
2089 void tcp6_proc_exit(struct net *net)
2091 tcp_proc_unregister(net, &tcp6_seq_afinfo);
2095 struct proto tcpv6_prot = {
2097 .owner = THIS_MODULE,
2099 .connect = tcp_v6_connect,
2100 .disconnect = tcp_disconnect,
2101 .accept = inet_csk_accept,
2103 .init = tcp_v6_init_sock,
2104 .destroy = tcp_v6_destroy_sock,
2105 .shutdown = tcp_shutdown,
2106 .setsockopt = tcp_setsockopt,
2107 .getsockopt = tcp_getsockopt,
2108 .recvmsg = tcp_recvmsg,
2109 .backlog_rcv = tcp_v6_do_rcv,
2110 .hash = tcp_v6_hash,
2111 .unhash = inet_unhash,
2112 .get_port = inet_csk_get_port,
2113 .enter_memory_pressure = tcp_enter_memory_pressure,
2114 .sockets_allocated = &tcp_sockets_allocated,
2115 .memory_allocated = &tcp_memory_allocated,
2116 .memory_pressure = &tcp_memory_pressure,
2117 .orphan_count = &tcp_orphan_count,
2118 .sysctl_mem = sysctl_tcp_mem,
2119 .sysctl_wmem = sysctl_tcp_wmem,
2120 .sysctl_rmem = sysctl_tcp_rmem,
2121 .max_header = MAX_TCP_HEADER,
2122 .obj_size = sizeof(struct tcp6_sock),
2123 .twsk_prot = &tcp6_timewait_sock_ops,
2124 .rsk_prot = &tcp6_request_sock_ops,
2125 .h.hashinfo = &tcp_hashinfo,
2126 #ifdef CONFIG_COMPAT
2127 .compat_setsockopt = compat_tcp_setsockopt,
2128 .compat_getsockopt = compat_tcp_getsockopt,
2132 static struct inet6_protocol tcpv6_protocol = {
2133 .handler = tcp_v6_rcv,
2134 .err_handler = tcp_v6_err,
2135 .gso_send_check = tcp_v6_gso_send_check,
2136 .gso_segment = tcp_tso_segment,
2137 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2140 static struct inet_protosw tcpv6_protosw = {
2141 .type = SOCK_STREAM,
2142 .protocol = IPPROTO_TCP,
2143 .prot = &tcpv6_prot,
2144 .ops = &inet6_stream_ops,
2147 .flags = INET_PROTOSW_PERMANENT |
2151 static int tcpv6_net_init(struct net *net)
2153 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2154 SOCK_RAW, IPPROTO_TCP, net);
2157 static void tcpv6_net_exit(struct net *net)
2159 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2162 static struct pernet_operations tcpv6_net_ops = {
2163 .init = tcpv6_net_init,
2164 .exit = tcpv6_net_exit,
2167 int __init tcpv6_init(void)
2171 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2175 /* register inet6 protocol */
2176 ret = inet6_register_protosw(&tcpv6_protosw);
2178 goto out_tcpv6_protocol;
2180 ret = register_pernet_subsys(&tcpv6_net_ops);
2182 goto out_tcpv6_protosw;
2187 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2189 inet6_unregister_protosw(&tcpv6_protosw);
2193 void tcpv6_exit(void)
2195 unregister_pernet_subsys(&tcpv6_net_ops);
2196 inet6_unregister_protosw(&tcpv6_protosw);
2197 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);