3 * Linux INET6 implementation
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: tcp_ipv6.c,v 1.144 2002/02/01 22:01:04 davem Exp $
11 * linux/net/ipv4/tcp.c
12 * linux/net/ipv4/tcp_input.c
13 * linux/net/ipv4/tcp_output.c
16 * Hideaki YOSHIFUJI : sin6_scope_id support
17 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
18 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
19 * a single port at the same time.
20 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
22 * This program is free software; you can redistribute it and/or
23 * modify it under the terms of the GNU General Public License
24 * as published by the Free Software Foundation; either version
25 * 2 of the License, or (at your option) any later version.
28 #include <linux/module.h>
29 #include <linux/errno.h>
30 #include <linux/types.h>
31 #include <linux/socket.h>
32 #include <linux/sockios.h>
33 #include <linux/net.h>
34 #include <linux/jiffies.h>
36 #include <linux/in6.h>
37 #include <linux/netdevice.h>
38 #include <linux/init.h>
39 #include <linux/jhash.h>
40 #include <linux/ipsec.h>
41 #include <linux/times.h>
43 #include <linux/ipv6.h>
44 #include <linux/icmpv6.h>
45 #include <linux/random.h>
48 #include <net/ndisc.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/inet6_connection_sock.h>
52 #include <net/transp_v6.h>
53 #include <net/addrconf.h>
54 #include <net/ip6_route.h>
55 #include <net/ip6_checksum.h>
56 #include <net/inet_ecn.h>
57 #include <net/protocol.h>
60 #include <net/dsfield.h>
61 #include <net/timewait_sock.h>
62 #include <net/netdma.h>
64 #include <asm/uaccess.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
69 #include <linux/crypto.h>
70 #include <linux/scatterlist.h>
72 /* Socket used for sending RSTs and ACKs */
73 static struct socket *tcp6_socket;
75 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
76 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req);
77 static void tcp_v6_send_check(struct sock *sk, int len,
80 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
82 static struct inet_connection_sock_af_ops ipv6_mapped;
83 static struct inet_connection_sock_af_ops ipv6_specific;
84 #ifdef CONFIG_TCP_MD5SIG
85 static struct tcp_sock_af_ops tcp_sock_ipv6_specific;
86 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
89 static void tcp_v6_hash(struct sock *sk)
91 if (sk->sk_state != TCP_CLOSE) {
92 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
102 static __inline__ __sum16 tcp_v6_check(struct tcphdr *th, int len,
103 struct in6_addr *saddr,
104 struct in6_addr *daddr,
107 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
110 static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
112 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
113 ipv6_hdr(skb)->saddr.s6_addr32,
115 tcp_hdr(skb)->source);
118 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
121 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
122 struct inet_sock *inet = inet_sk(sk);
123 struct inet_connection_sock *icsk = inet_csk(sk);
124 struct ipv6_pinfo *np = inet6_sk(sk);
125 struct tcp_sock *tp = tcp_sk(sk);
126 struct in6_addr *saddr = NULL, *final_p = NULL, final;
128 struct dst_entry *dst;
132 if (addr_len < SIN6_LEN_RFC2133)
135 if (usin->sin6_family != AF_INET6)
136 return(-EAFNOSUPPORT);
138 memset(&fl, 0, sizeof(fl));
141 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
142 IP6_ECN_flow_init(fl.fl6_flowlabel);
143 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
144 struct ip6_flowlabel *flowlabel;
145 flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
146 if (flowlabel == NULL)
148 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
149 fl6_sock_release(flowlabel);
154 * connect() to INADDR_ANY means loopback (BSD'ism).
157 if(ipv6_addr_any(&usin->sin6_addr))
158 usin->sin6_addr.s6_addr[15] = 0x1;
160 addr_type = ipv6_addr_type(&usin->sin6_addr);
162 if(addr_type & IPV6_ADDR_MULTICAST)
165 if (addr_type&IPV6_ADDR_LINKLOCAL) {
166 if (addr_len >= sizeof(struct sockaddr_in6) &&
167 usin->sin6_scope_id) {
168 /* If interface is set while binding, indices
171 if (sk->sk_bound_dev_if &&
172 sk->sk_bound_dev_if != usin->sin6_scope_id)
175 sk->sk_bound_dev_if = usin->sin6_scope_id;
178 /* Connect to link-local address requires an interface */
179 if (!sk->sk_bound_dev_if)
183 if (tp->rx_opt.ts_recent_stamp &&
184 !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
185 tp->rx_opt.ts_recent = 0;
186 tp->rx_opt.ts_recent_stamp = 0;
190 ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
191 np->flow_label = fl.fl6_flowlabel;
197 if (addr_type == IPV6_ADDR_MAPPED) {
198 u32 exthdrlen = icsk->icsk_ext_hdr_len;
199 struct sockaddr_in sin;
201 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
203 if (__ipv6_only_sock(sk))
206 sin.sin_family = AF_INET;
207 sin.sin_port = usin->sin6_port;
208 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
210 icsk->icsk_af_ops = &ipv6_mapped;
211 sk->sk_backlog_rcv = tcp_v4_do_rcv;
212 #ifdef CONFIG_TCP_MD5SIG
213 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
216 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
219 icsk->icsk_ext_hdr_len = exthdrlen;
220 icsk->icsk_af_ops = &ipv6_specific;
221 sk->sk_backlog_rcv = tcp_v6_do_rcv;
222 #ifdef CONFIG_TCP_MD5SIG
223 tp->af_specific = &tcp_sock_ipv6_specific;
227 ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
229 ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
236 if (!ipv6_addr_any(&np->rcv_saddr))
237 saddr = &np->rcv_saddr;
239 fl.proto = IPPROTO_TCP;
240 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
241 ipv6_addr_copy(&fl.fl6_src,
242 (saddr ? saddr : &np->saddr));
243 fl.oif = sk->sk_bound_dev_if;
244 fl.fl_ip_dport = usin->sin6_port;
245 fl.fl_ip_sport = inet->sport;
247 if (np->opt && np->opt->srcrt) {
248 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
249 ipv6_addr_copy(&final, &fl.fl6_dst);
250 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
254 security_sk_classify_flow(sk, &fl);
256 err = ip6_dst_lookup(sk, &dst, &fl);
260 ipv6_addr_copy(&fl.fl6_dst, final_p);
262 if ((err = __xfrm_lookup(&dst, &fl, sk, XFRM_LOOKUP_WAIT)) < 0) {
264 err = ip6_dst_blackhole(sk, &dst, &fl);
271 ipv6_addr_copy(&np->rcv_saddr, saddr);
274 /* set the source address */
275 ipv6_addr_copy(&np->saddr, saddr);
276 inet->rcv_saddr = LOOPBACK4_IPV6;
278 sk->sk_gso_type = SKB_GSO_TCPV6;
279 __ip6_dst_store(sk, dst, NULL, NULL);
281 icsk->icsk_ext_hdr_len = 0;
283 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
286 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
288 inet->dport = usin->sin6_port;
290 tcp_set_state(sk, TCP_SYN_SENT);
291 err = inet6_hash_connect(&tcp_death_row, sk);
296 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
301 err = tcp_connect(sk);
308 tcp_set_state(sk, TCP_CLOSE);
312 sk->sk_route_caps = 0;
316 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
317 int type, int code, int offset, __be32 info)
319 struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
320 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
321 struct ipv6_pinfo *np;
327 sk = inet6_lookup(skb->dev->nd_net, &tcp_hashinfo, &hdr->daddr,
328 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
331 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
335 if (sk->sk_state == TCP_TIME_WAIT) {
336 inet_twsk_put(inet_twsk(sk));
341 if (sock_owned_by_user(sk))
342 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
344 if (sk->sk_state == TCP_CLOSE)
348 seq = ntohl(th->seq);
349 if (sk->sk_state != TCP_LISTEN &&
350 !between(seq, tp->snd_una, tp->snd_nxt)) {
351 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
357 if (type == ICMPV6_PKT_TOOBIG) {
358 struct dst_entry *dst = NULL;
360 if (sock_owned_by_user(sk))
362 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
365 /* icmp should have updated the destination cache entry */
366 dst = __sk_dst_check(sk, np->dst_cookie);
369 struct inet_sock *inet = inet_sk(sk);
372 /* BUGGG_FUTURE: Again, it is not clear how
373 to handle rthdr case. Ignore this complexity
376 memset(&fl, 0, sizeof(fl));
377 fl.proto = IPPROTO_TCP;
378 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
379 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
380 fl.oif = sk->sk_bound_dev_if;
381 fl.fl_ip_dport = inet->dport;
382 fl.fl_ip_sport = inet->sport;
383 security_skb_classify_flow(skb, &fl);
385 if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
386 sk->sk_err_soft = -err;
390 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
391 sk->sk_err_soft = -err;
398 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
399 tcp_sync_mss(sk, dst_mtu(dst));
400 tcp_simple_retransmit(sk);
401 } /* else let the usual retransmit timer handle it */
406 icmpv6_err_convert(type, code, &err);
408 /* Might be for an request_sock */
409 switch (sk->sk_state) {
410 struct request_sock *req, **prev;
412 if (sock_owned_by_user(sk))
415 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
416 &hdr->saddr, inet6_iif(skb));
420 /* ICMPs are not backlogged, hence we cannot get
421 * an established socket here.
423 BUG_TRAP(req->sk == NULL);
425 if (seq != tcp_rsk(req)->snt_isn) {
426 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
430 inet_csk_reqsk_queue_drop(sk, req, prev);
434 case TCP_SYN_RECV: /* Cannot happen.
435 It can, it SYNs are crossed. --ANK */
436 if (!sock_owned_by_user(sk)) {
438 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
442 sk->sk_err_soft = err;
446 if (!sock_owned_by_user(sk) && np->recverr) {
448 sk->sk_error_report(sk);
450 sk->sk_err_soft = err;
458 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
459 struct dst_entry *dst)
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;
469 memset(&fl, 0, sizeof(fl));
470 fl.proto = IPPROTO_TCP;
471 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
472 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
473 fl.fl6_flowlabel = 0;
475 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
476 fl.fl_ip_sport = inet_sk(sk)->sport;
477 security_req_classify_flow(req, &fl);
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)
497 skb = tcp_make_synack(sk, dst, req);
499 struct tcphdr *th = tcp_hdr(skb);
501 th->check = tcp_v6_check(th, skb->len,
502 &treq->loc_addr, &treq->rmt_addr,
503 csum_partial((char *)th, skb->len, skb->csum));
505 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
506 err = ip6_xmit(sk, skb, &fl, opt, 0);
507 err = net_xmit_eval(err);
511 if (opt && opt != np->opt)
512 sock_kfree_s(sk, opt, opt->tot_len);
517 static void tcp_v6_reqsk_destructor(struct request_sock *req)
519 if (inet6_rsk(req)->pktopts)
520 kfree_skb(inet6_rsk(req)->pktopts);
523 #ifdef CONFIG_TCP_MD5SIG
524 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
525 struct in6_addr *addr)
527 struct tcp_sock *tp = tcp_sk(sk);
532 if (!tp->md5sig_info || !tp->md5sig_info->entries6)
535 for (i = 0; i < tp->md5sig_info->entries6; i++) {
536 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, addr) == 0)
537 return &tp->md5sig_info->keys6[i].base;
542 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
543 struct sock *addr_sk)
545 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
548 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
549 struct request_sock *req)
551 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
554 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
555 char *newkey, u8 newkeylen)
557 /* Add key to the list */
558 struct tcp_md5sig_key *key;
559 struct tcp_sock *tp = tcp_sk(sk);
560 struct tcp6_md5sig_key *keys;
562 key = tcp_v6_md5_do_lookup(sk, peer);
564 /* modify existing entry - just update that one */
567 key->keylen = newkeylen;
569 /* reallocate new list if current one is full. */
570 if (!tp->md5sig_info) {
571 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
572 if (!tp->md5sig_info) {
576 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
578 if (tcp_alloc_md5sig_pool() == NULL) {
582 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
583 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
584 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
587 tcp_free_md5sig_pool();
592 if (tp->md5sig_info->entries6)
593 memmove(keys, tp->md5sig_info->keys6,
594 (sizeof (tp->md5sig_info->keys6[0]) *
595 tp->md5sig_info->entries6));
597 kfree(tp->md5sig_info->keys6);
598 tp->md5sig_info->keys6 = keys;
599 tp->md5sig_info->alloced6++;
602 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
604 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
605 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
607 tp->md5sig_info->entries6++;
612 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
613 u8 *newkey, __u8 newkeylen)
615 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
619 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
621 struct tcp_sock *tp = tcp_sk(sk);
624 for (i = 0; i < tp->md5sig_info->entries6; i++) {
625 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, peer) == 0) {
627 kfree(tp->md5sig_info->keys6[i].base.key);
628 tp->md5sig_info->entries6--;
630 if (tp->md5sig_info->entries6 == 0) {
631 kfree(tp->md5sig_info->keys6);
632 tp->md5sig_info->keys6 = NULL;
633 tp->md5sig_info->alloced6 = 0;
635 /* shrink the database */
636 if (tp->md5sig_info->entries6 != i)
637 memmove(&tp->md5sig_info->keys6[i],
638 &tp->md5sig_info->keys6[i+1],
639 (tp->md5sig_info->entries6 - i)
640 * sizeof (tp->md5sig_info->keys6[0]));
642 tcp_free_md5sig_pool();
649 static void tcp_v6_clear_md5_list (struct sock *sk)
651 struct tcp_sock *tp = tcp_sk(sk);
654 if (tp->md5sig_info->entries6) {
655 for (i = 0; i < tp->md5sig_info->entries6; i++)
656 kfree(tp->md5sig_info->keys6[i].base.key);
657 tp->md5sig_info->entries6 = 0;
658 tcp_free_md5sig_pool();
661 kfree(tp->md5sig_info->keys6);
662 tp->md5sig_info->keys6 = NULL;
663 tp->md5sig_info->alloced6 = 0;
665 if (tp->md5sig_info->entries4) {
666 for (i = 0; i < tp->md5sig_info->entries4; i++)
667 kfree(tp->md5sig_info->keys4[i].base.key);
668 tp->md5sig_info->entries4 = 0;
669 tcp_free_md5sig_pool();
672 kfree(tp->md5sig_info->keys4);
673 tp->md5sig_info->keys4 = NULL;
674 tp->md5sig_info->alloced4 = 0;
677 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
680 struct tcp_md5sig cmd;
681 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
684 if (optlen < sizeof(cmd))
687 if (copy_from_user(&cmd, optval, sizeof(cmd)))
690 if (sin6->sin6_family != AF_INET6)
693 if (!cmd.tcpm_keylen) {
694 if (!tcp_sk(sk)->md5sig_info)
696 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
697 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
698 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
701 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
704 if (!tcp_sk(sk)->md5sig_info) {
705 struct tcp_sock *tp = tcp_sk(sk);
706 struct tcp_md5sig_info *p;
708 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
713 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
716 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
719 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
720 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
721 newkey, cmd.tcpm_keylen);
723 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
726 static int tcp_v6_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
727 struct in6_addr *saddr,
728 struct in6_addr *daddr,
729 struct tcphdr *th, int protocol,
732 struct scatterlist sg[4];
736 struct tcp_md5sig_pool *hp;
737 struct tcp6_pseudohdr *bp;
738 struct hash_desc *desc;
740 unsigned int nbytes = 0;
742 hp = tcp_get_md5sig_pool();
744 printk(KERN_WARNING "%s(): hash pool not found...\n", __FUNCTION__);
745 goto clear_hash_noput;
747 bp = &hp->md5_blk.ip6;
748 desc = &hp->md5_desc;
750 /* 1. TCP pseudo-header (RFC2460) */
751 ipv6_addr_copy(&bp->saddr, saddr);
752 ipv6_addr_copy(&bp->daddr, daddr);
753 bp->len = htonl(tcplen);
754 bp->protocol = htonl(protocol);
756 sg_init_table(sg, 4);
758 sg_set_buf(&sg[block++], bp, sizeof(*bp));
759 nbytes += sizeof(*bp);
761 /* 2. TCP header, excluding options */
764 sg_set_buf(&sg[block++], th, sizeof(*th));
765 nbytes += sizeof(*th);
767 /* 3. TCP segment data (if any) */
768 data_len = tcplen - (th->doff << 2);
770 u8 *data = (u8 *)th + (th->doff << 2);
771 sg_set_buf(&sg[block++], data, data_len);
776 sg_set_buf(&sg[block++], key->key, key->keylen);
777 nbytes += key->keylen;
779 sg_mark_end(&sg[block - 1]);
781 /* Now store the hash into the packet */
782 err = crypto_hash_init(desc);
784 printk(KERN_WARNING "%s(): hash_init failed\n", __FUNCTION__);
787 err = crypto_hash_update(desc, sg, nbytes);
789 printk(KERN_WARNING "%s(): hash_update failed\n", __FUNCTION__);
792 err = crypto_hash_final(desc, md5_hash);
794 printk(KERN_WARNING "%s(): hash_final failed\n", __FUNCTION__);
798 /* Reset header, and free up the crypto */
799 tcp_put_md5sig_pool();
804 tcp_put_md5sig_pool();
806 memset(md5_hash, 0, 16);
810 static int tcp_v6_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
812 struct dst_entry *dst,
813 struct request_sock *req,
814 struct tcphdr *th, int protocol,
817 struct in6_addr *saddr, *daddr;
820 saddr = &inet6_sk(sk)->saddr;
821 daddr = &inet6_sk(sk)->daddr;
823 saddr = &inet6_rsk(req)->loc_addr;
824 daddr = &inet6_rsk(req)->rmt_addr;
826 return tcp_v6_do_calc_md5_hash(md5_hash, key,
828 th, protocol, tcplen);
831 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
833 __u8 *hash_location = NULL;
834 struct tcp_md5sig_key *hash_expected;
835 struct ipv6hdr *ip6h = ipv6_hdr(skb);
836 struct tcphdr *th = tcp_hdr(skb);
837 int length = (th->doff << 2) - sizeof (*th);
842 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
844 /* If the TCP option is too short, we can short cut */
845 if (length < TCPOLEN_MD5SIG)
846 return hash_expected ? 1 : 0;
862 if (opsize < 2 || opsize > length)
864 if (opcode == TCPOPT_MD5SIG) {
874 /* do we have a hash as expected? */
875 if (!hash_expected) {
878 if (net_ratelimit()) {
879 printk(KERN_INFO "MD5 Hash NOT expected but found "
880 "(" NIP6_FMT ", %u)->"
881 "(" NIP6_FMT ", %u)\n",
882 NIP6(ip6h->saddr), ntohs(th->source),
883 NIP6(ip6h->daddr), ntohs(th->dest));
888 if (!hash_location) {
889 if (net_ratelimit()) {
890 printk(KERN_INFO "MD5 Hash expected but NOT found "
891 "(" NIP6_FMT ", %u)->"
892 "(" NIP6_FMT ", %u)\n",
893 NIP6(ip6h->saddr), ntohs(th->source),
894 NIP6(ip6h->daddr), ntohs(th->dest));
899 /* check the signature */
900 genhash = tcp_v6_do_calc_md5_hash(newhash,
902 &ip6h->saddr, &ip6h->daddr,
905 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
906 if (net_ratelimit()) {
907 printk(KERN_INFO "MD5 Hash %s for "
908 "(" NIP6_FMT ", %u)->"
909 "(" NIP6_FMT ", %u)\n",
910 genhash ? "failed" : "mismatch",
911 NIP6(ip6h->saddr), ntohs(th->source),
912 NIP6(ip6h->daddr), ntohs(th->dest));
920 static struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
922 .obj_size = sizeof(struct tcp6_request_sock),
923 .rtx_syn_ack = tcp_v6_send_synack,
924 .send_ack = tcp_v6_reqsk_send_ack,
925 .destructor = tcp_v6_reqsk_destructor,
926 .send_reset = tcp_v6_send_reset
929 #ifdef CONFIG_TCP_MD5SIG
930 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
931 .md5_lookup = tcp_v6_reqsk_md5_lookup,
935 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
936 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
937 .twsk_unique = tcp_twsk_unique,
938 .twsk_destructor= tcp_twsk_destructor,
941 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
943 struct ipv6_pinfo *np = inet6_sk(sk);
944 struct tcphdr *th = tcp_hdr(skb);
946 if (skb->ip_summed == CHECKSUM_PARTIAL) {
947 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 0);
948 skb->csum_start = skb_transport_header(skb) - skb->head;
949 skb->csum_offset = offsetof(struct tcphdr, check);
951 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
952 csum_partial((char *)th, th->doff<<2,
957 static int tcp_v6_gso_send_check(struct sk_buff *skb)
959 struct ipv6hdr *ipv6h;
962 if (!pskb_may_pull(skb, sizeof(*th)))
965 ipv6h = ipv6_hdr(skb);
969 th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
971 skb->csum_start = skb_transport_header(skb) - skb->head;
972 skb->csum_offset = offsetof(struct tcphdr, check);
973 skb->ip_summed = CHECKSUM_PARTIAL;
977 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
979 struct tcphdr *th = tcp_hdr(skb), *t1;
980 struct sk_buff *buff;
982 unsigned int tot_len = sizeof(*th);
983 #ifdef CONFIG_TCP_MD5SIG
984 struct tcp_md5sig_key *key;
990 if (!ipv6_unicast_destination(skb))
993 #ifdef CONFIG_TCP_MD5SIG
995 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1000 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1004 * We need to grab some memory, and put together an RST,
1005 * and then put it into the queue to be sent.
1008 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1013 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1015 t1 = (struct tcphdr *) skb_push(buff, tot_len);
1017 /* Swap the send and the receive. */
1018 memset(t1, 0, sizeof(*t1));
1019 t1->dest = th->source;
1020 t1->source = th->dest;
1021 t1->doff = tot_len / 4;
1025 t1->seq = th->ack_seq;
1028 t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
1029 + skb->len - (th->doff<<2));
1032 #ifdef CONFIG_TCP_MD5SIG
1034 __be32 *opt = (__be32*)(t1 + 1);
1035 opt[0] = htonl((TCPOPT_NOP << 24) |
1036 (TCPOPT_NOP << 16) |
1037 (TCPOPT_MD5SIG << 8) |
1039 tcp_v6_do_calc_md5_hash((__u8 *)&opt[1], key,
1040 &ipv6_hdr(skb)->daddr,
1041 &ipv6_hdr(skb)->saddr,
1042 t1, IPPROTO_TCP, tot_len);
1046 buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1048 memset(&fl, 0, sizeof(fl));
1049 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1050 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1052 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1053 sizeof(*t1), IPPROTO_TCP,
1056 fl.proto = IPPROTO_TCP;
1057 fl.oif = inet6_iif(skb);
1058 fl.fl_ip_dport = t1->dest;
1059 fl.fl_ip_sport = t1->source;
1060 security_skb_classify_flow(skb, &fl);
1062 /* sk = NULL, but it is safe for now. RST socket required. */
1063 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1065 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1066 ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1067 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1068 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
1076 static void tcp_v6_send_ack(struct tcp_timewait_sock *tw,
1077 struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
1079 struct tcphdr *th = tcp_hdr(skb), *t1;
1080 struct sk_buff *buff;
1082 unsigned int tot_len = sizeof(struct tcphdr);
1084 #ifdef CONFIG_TCP_MD5SIG
1085 struct tcp_md5sig_key *key;
1086 struct tcp_md5sig_key tw_key;
1089 #ifdef CONFIG_TCP_MD5SIG
1090 if (!tw && skb->sk) {
1091 key = tcp_v6_md5_do_lookup(skb->sk, &ipv6_hdr(skb)->daddr);
1092 } else if (tw && tw->tw_md5_keylen) {
1093 tw_key.key = tw->tw_md5_key;
1094 tw_key.keylen = tw->tw_md5_keylen;
1102 tot_len += TCPOLEN_TSTAMP_ALIGNED;
1103 #ifdef CONFIG_TCP_MD5SIG
1105 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1108 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1113 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1115 t1 = (struct tcphdr *) skb_push(buff,tot_len);
1117 /* Swap the send and the receive. */
1118 memset(t1, 0, sizeof(*t1));
1119 t1->dest = th->source;
1120 t1->source = th->dest;
1121 t1->doff = tot_len/4;
1122 t1->seq = htonl(seq);
1123 t1->ack_seq = htonl(ack);
1125 t1->window = htons(win);
1127 topt = (__be32 *)(t1 + 1);
1130 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1131 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1132 *topt++ = htonl(tcp_time_stamp);
1136 #ifdef CONFIG_TCP_MD5SIG
1138 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1139 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1140 tcp_v6_do_calc_md5_hash((__u8 *)topt, key,
1141 &ipv6_hdr(skb)->daddr,
1142 &ipv6_hdr(skb)->saddr,
1143 t1, IPPROTO_TCP, tot_len);
1147 buff->csum = csum_partial((char *)t1, tot_len, 0);
1149 memset(&fl, 0, sizeof(fl));
1150 ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1151 ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1153 t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1154 tot_len, IPPROTO_TCP,
1157 fl.proto = IPPROTO_TCP;
1158 fl.oif = inet6_iif(skb);
1159 fl.fl_ip_dport = t1->dest;
1160 fl.fl_ip_sport = t1->source;
1161 security_skb_classify_flow(skb, &fl);
1163 if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1164 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1165 ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1166 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1174 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1176 struct inet_timewait_sock *tw = inet_twsk(sk);
1177 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1179 tcp_v6_send_ack(tcptw, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1180 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1181 tcptw->tw_ts_recent);
1186 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
1188 tcp_v6_send_ack(NULL, skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent);
1192 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1194 struct request_sock *req, **prev;
1195 const struct tcphdr *th = tcp_hdr(skb);
1198 /* Find possible connection requests. */
1199 req = inet6_csk_search_req(sk, &prev, th->source,
1200 &ipv6_hdr(skb)->saddr,
1201 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1203 return tcp_check_req(sk, skb, req, prev);
1205 nsk = __inet6_lookup_established(sk->sk_net, &tcp_hashinfo,
1206 &ipv6_hdr(skb)->saddr, th->source,
1207 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1210 if (nsk->sk_state != TCP_TIME_WAIT) {
1214 inet_twsk_put(inet_twsk(nsk));
1218 #if 0 /*def CONFIG_SYN_COOKIES*/
1219 if (!th->rst && !th->syn && th->ack)
1220 sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
1225 /* FIXME: this is substantially similar to the ipv4 code.
1226 * Can some kind of merge be done? -- erics
1228 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1230 struct inet6_request_sock *treq;
1231 struct ipv6_pinfo *np = inet6_sk(sk);
1232 struct tcp_options_received tmp_opt;
1233 struct tcp_sock *tp = tcp_sk(sk);
1234 struct request_sock *req = NULL;
1235 __u32 isn = TCP_SKB_CB(skb)->when;
1237 if (skb->protocol == htons(ETH_P_IP))
1238 return tcp_v4_conn_request(sk, skb);
1240 if (!ipv6_unicast_destination(skb))
1244 * There are no SYN attacks on IPv6, yet...
1246 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1247 if (net_ratelimit())
1248 printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
1252 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1255 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1259 #ifdef CONFIG_TCP_MD5SIG
1260 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1263 tcp_clear_options(&tmp_opt);
1264 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1265 tmp_opt.user_mss = tp->rx_opt.user_mss;
1267 tcp_parse_options(skb, &tmp_opt, 0);
1269 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1270 tcp_openreq_init(req, &tmp_opt, skb);
1272 treq = inet6_rsk(req);
1273 ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1274 ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1275 TCP_ECN_create_request(req, tcp_hdr(skb));
1276 treq->pktopts = NULL;
1277 if (ipv6_opt_accepted(sk, skb) ||
1278 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1279 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1280 atomic_inc(&skb->users);
1281 treq->pktopts = skb;
1283 treq->iif = sk->sk_bound_dev_if;
1285 /* So that link locals have meaning */
1286 if (!sk->sk_bound_dev_if &&
1287 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1288 treq->iif = inet6_iif(skb);
1291 isn = tcp_v6_init_sequence(skb);
1293 tcp_rsk(req)->snt_isn = isn;
1295 security_inet_conn_request(sk, skb, req);
1297 if (tcp_v6_send_synack(sk, req, NULL))
1300 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1307 return 0; /* don't send reset */
1310 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1311 struct request_sock *req,
1312 struct dst_entry *dst)
1314 struct inet6_request_sock *treq = inet6_rsk(req);
1315 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1316 struct tcp6_sock *newtcp6sk;
1317 struct inet_sock *newinet;
1318 struct tcp_sock *newtp;
1320 struct ipv6_txoptions *opt;
1321 #ifdef CONFIG_TCP_MD5SIG
1322 struct tcp_md5sig_key *key;
1325 if (skb->protocol == htons(ETH_P_IP)) {
1330 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1335 newtcp6sk = (struct tcp6_sock *)newsk;
1336 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1338 newinet = inet_sk(newsk);
1339 newnp = inet6_sk(newsk);
1340 newtp = tcp_sk(newsk);
1342 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1344 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1347 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1350 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1352 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1353 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1354 #ifdef CONFIG_TCP_MD5SIG
1355 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1358 newnp->pktoptions = NULL;
1360 newnp->mcast_oif = inet6_iif(skb);
1361 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1364 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1365 * here, tcp_create_openreq_child now does this for us, see the comment in
1366 * that function for the gory details. -acme
1369 /* It is tricky place. Until this moment IPv4 tcp
1370 worked with IPv6 icsk.icsk_af_ops.
1373 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1380 if (sk_acceptq_is_full(sk))
1384 struct in6_addr *final_p = NULL, final;
1387 memset(&fl, 0, sizeof(fl));
1388 fl.proto = IPPROTO_TCP;
1389 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1390 if (opt && opt->srcrt) {
1391 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1392 ipv6_addr_copy(&final, &fl.fl6_dst);
1393 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1396 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1397 fl.oif = sk->sk_bound_dev_if;
1398 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1399 fl.fl_ip_sport = inet_sk(sk)->sport;
1400 security_req_classify_flow(req, &fl);
1402 if (ip6_dst_lookup(sk, &dst, &fl))
1406 ipv6_addr_copy(&fl.fl6_dst, final_p);
1408 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1412 newsk = tcp_create_openreq_child(sk, req, skb);
1417 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1418 * count here, tcp_create_openreq_child now does this for us, see the
1419 * comment in that function for the gory details. -acme
1422 newsk->sk_gso_type = SKB_GSO_TCPV6;
1423 __ip6_dst_store(newsk, dst, NULL, NULL);
1425 newtcp6sk = (struct tcp6_sock *)newsk;
1426 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1428 newtp = tcp_sk(newsk);
1429 newinet = inet_sk(newsk);
1430 newnp = inet6_sk(newsk);
1432 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1434 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1435 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1436 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1437 newsk->sk_bound_dev_if = treq->iif;
1439 /* Now IPv6 options...
1441 First: no IPv4 options.
1443 newinet->opt = NULL;
1444 newnp->ipv6_fl_list = NULL;
1447 newnp->rxopt.all = np->rxopt.all;
1449 /* Clone pktoptions received with SYN */
1450 newnp->pktoptions = NULL;
1451 if (treq->pktopts != NULL) {
1452 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1453 kfree_skb(treq->pktopts);
1454 treq->pktopts = NULL;
1455 if (newnp->pktoptions)
1456 skb_set_owner_r(newnp->pktoptions, newsk);
1459 newnp->mcast_oif = inet6_iif(skb);
1460 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1462 /* Clone native IPv6 options from listening socket (if any)
1464 Yes, keeping reference count would be much more clever,
1465 but we make one more one thing there: reattach optmem
1469 newnp->opt = ipv6_dup_options(newsk, opt);
1471 sock_kfree_s(sk, opt, opt->tot_len);
1474 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1476 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1477 newnp->opt->opt_flen);
1479 tcp_mtup_init(newsk);
1480 tcp_sync_mss(newsk, dst_mtu(dst));
1481 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1482 tcp_initialize_rcv_mss(newsk);
1484 newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1486 #ifdef CONFIG_TCP_MD5SIG
1487 /* Copy over the MD5 key from the original socket */
1488 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1489 /* We're using one, so create a matching key
1490 * on the newsk structure. If we fail to get
1491 * memory, then we end up not copying the key
1494 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1496 tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1497 newkey, key->keylen);
1501 __inet6_hash(newsk);
1502 inet_inherit_port(sk, newsk);
1507 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1509 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1510 if (opt && opt != np->opt)
1511 sock_kfree_s(sk, opt, opt->tot_len);
1516 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1518 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1519 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1520 &ipv6_hdr(skb)->daddr, skb->csum)) {
1521 skb->ip_summed = CHECKSUM_UNNECESSARY;
1526 skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1527 &ipv6_hdr(skb)->saddr,
1528 &ipv6_hdr(skb)->daddr, 0));
1530 if (skb->len <= 76) {
1531 return __skb_checksum_complete(skb);
1536 /* The socket must have it's spinlock held when we get
1539 * We have a potential double-lock case here, so even when
1540 * doing backlog processing we use the BH locking scheme.
1541 * This is because we cannot sleep with the original spinlock
1544 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1546 struct ipv6_pinfo *np = inet6_sk(sk);
1547 struct tcp_sock *tp;
1548 struct sk_buff *opt_skb = NULL;
1550 /* Imagine: socket is IPv6. IPv4 packet arrives,
1551 goes to IPv4 receive handler and backlogged.
1552 From backlog it always goes here. Kerboom...
1553 Fortunately, tcp_rcv_established and rcv_established
1554 handle them correctly, but it is not case with
1555 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1558 if (skb->protocol == htons(ETH_P_IP))
1559 return tcp_v4_do_rcv(sk, skb);
1561 #ifdef CONFIG_TCP_MD5SIG
1562 if (tcp_v6_inbound_md5_hash (sk, skb))
1566 if (sk_filter(sk, skb))
1570 * socket locking is here for SMP purposes as backlog rcv
1571 * is currently called with bh processing disabled.
1574 /* Do Stevens' IPV6_PKTOPTIONS.
1576 Yes, guys, it is the only place in our code, where we
1577 may make it not affecting IPv4.
1578 The rest of code is protocol independent,
1579 and I do not like idea to uglify IPv4.
1581 Actually, all the idea behind IPV6_PKTOPTIONS
1582 looks not very well thought. For now we latch
1583 options, received in the last packet, enqueued
1584 by tcp. Feel free to propose better solution.
1588 opt_skb = skb_clone(skb, GFP_ATOMIC);
1590 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1591 TCP_CHECK_TIMER(sk);
1592 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1594 TCP_CHECK_TIMER(sk);
1596 goto ipv6_pktoptions;
1600 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1603 if (sk->sk_state == TCP_LISTEN) {
1604 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1609 * Queue it on the new socket if the new socket is active,
1610 * otherwise we just shortcircuit this and continue with
1614 if (tcp_child_process(sk, nsk, skb))
1617 __kfree_skb(opt_skb);
1622 TCP_CHECK_TIMER(sk);
1623 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1625 TCP_CHECK_TIMER(sk);
1627 goto ipv6_pktoptions;
1631 tcp_v6_send_reset(sk, skb);
1634 __kfree_skb(opt_skb);
1638 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1643 /* Do you ask, what is it?
1645 1. skb was enqueued by tcp.
1646 2. skb is added to tail of read queue, rather than out of order.
1647 3. socket is not in passive state.
1648 4. Finally, it really contains options, which user wants to receive.
1651 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1652 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1653 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1654 np->mcast_oif = inet6_iif(opt_skb);
1655 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1656 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1657 if (ipv6_opt_accepted(sk, opt_skb)) {
1658 skb_set_owner_r(opt_skb, sk);
1659 opt_skb = xchg(&np->pktoptions, opt_skb);
1661 __kfree_skb(opt_skb);
1662 opt_skb = xchg(&np->pktoptions, NULL);
1671 static int tcp_v6_rcv(struct sk_buff *skb)
1677 if (skb->pkt_type != PACKET_HOST)
1681 * Count it even if it's bad.
1683 TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1685 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1690 if (th->doff < sizeof(struct tcphdr)/4)
1692 if (!pskb_may_pull(skb, th->doff*4))
1695 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1699 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1700 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1701 skb->len - th->doff*4);
1702 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1703 TCP_SKB_CB(skb)->when = 0;
1704 TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1705 TCP_SKB_CB(skb)->sacked = 0;
1707 sk = __inet6_lookup(skb->dev->nd_net, &tcp_hashinfo,
1708 &ipv6_hdr(skb)->saddr, th->source,
1709 &ipv6_hdr(skb)->daddr, ntohs(th->dest),
1716 if (sk->sk_state == TCP_TIME_WAIT)
1719 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1720 goto discard_and_relse;
1722 if (sk_filter(sk, skb))
1723 goto discard_and_relse;
1727 bh_lock_sock_nested(sk);
1729 if (!sock_owned_by_user(sk)) {
1730 #ifdef CONFIG_NET_DMA
1731 struct tcp_sock *tp = tcp_sk(sk);
1732 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1733 tp->ucopy.dma_chan = get_softnet_dma();
1734 if (tp->ucopy.dma_chan)
1735 ret = tcp_v6_do_rcv(sk, skb);
1739 if (!tcp_prequeue(sk, skb))
1740 ret = tcp_v6_do_rcv(sk, skb);
1743 sk_add_backlog(sk, skb);
1747 return ret ? -1 : 0;
1750 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1753 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1755 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1757 tcp_v6_send_reset(NULL, skb);
1774 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1775 inet_twsk_put(inet_twsk(sk));
1779 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1780 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1781 inet_twsk_put(inet_twsk(sk));
1785 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1790 sk2 = inet6_lookup_listener(skb->dev->nd_net, &tcp_hashinfo,
1791 &ipv6_hdr(skb)->daddr,
1792 ntohs(th->dest), inet6_iif(skb));
1794 struct inet_timewait_sock *tw = inet_twsk(sk);
1795 inet_twsk_deschedule(tw, &tcp_death_row);
1800 /* Fall through to ACK */
1803 tcp_v6_timewait_ack(sk, skb);
1807 case TCP_TW_SUCCESS:;
1812 static int tcp_v6_remember_stamp(struct sock *sk)
1814 /* Alas, not yet... */
1818 static struct inet_connection_sock_af_ops ipv6_specific = {
1819 .queue_xmit = inet6_csk_xmit,
1820 .send_check = tcp_v6_send_check,
1821 .rebuild_header = inet6_sk_rebuild_header,
1822 .conn_request = tcp_v6_conn_request,
1823 .syn_recv_sock = tcp_v6_syn_recv_sock,
1824 .remember_stamp = tcp_v6_remember_stamp,
1825 .net_header_len = sizeof(struct ipv6hdr),
1826 .setsockopt = ipv6_setsockopt,
1827 .getsockopt = ipv6_getsockopt,
1828 .addr2sockaddr = inet6_csk_addr2sockaddr,
1829 .sockaddr_len = sizeof(struct sockaddr_in6),
1830 .bind_conflict = inet6_csk_bind_conflict,
1831 #ifdef CONFIG_COMPAT
1832 .compat_setsockopt = compat_ipv6_setsockopt,
1833 .compat_getsockopt = compat_ipv6_getsockopt,
1837 #ifdef CONFIG_TCP_MD5SIG
1838 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1839 .md5_lookup = tcp_v6_md5_lookup,
1840 .calc_md5_hash = tcp_v6_calc_md5_hash,
1841 .md5_add = tcp_v6_md5_add_func,
1842 .md5_parse = tcp_v6_parse_md5_keys,
1847 * TCP over IPv4 via INET6 API
1850 static struct inet_connection_sock_af_ops ipv6_mapped = {
1851 .queue_xmit = ip_queue_xmit,
1852 .send_check = tcp_v4_send_check,
1853 .rebuild_header = inet_sk_rebuild_header,
1854 .conn_request = tcp_v6_conn_request,
1855 .syn_recv_sock = tcp_v6_syn_recv_sock,
1856 .remember_stamp = tcp_v4_remember_stamp,
1857 .net_header_len = sizeof(struct iphdr),
1858 .setsockopt = ipv6_setsockopt,
1859 .getsockopt = ipv6_getsockopt,
1860 .addr2sockaddr = inet6_csk_addr2sockaddr,
1861 .sockaddr_len = sizeof(struct sockaddr_in6),
1862 .bind_conflict = inet6_csk_bind_conflict,
1863 #ifdef CONFIG_COMPAT
1864 .compat_setsockopt = compat_ipv6_setsockopt,
1865 .compat_getsockopt = compat_ipv6_getsockopt,
1869 #ifdef CONFIG_TCP_MD5SIG
1870 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1871 .md5_lookup = tcp_v4_md5_lookup,
1872 .calc_md5_hash = tcp_v4_calc_md5_hash,
1873 .md5_add = tcp_v6_md5_add_func,
1874 .md5_parse = tcp_v6_parse_md5_keys,
1878 /* NOTE: A lot of things set to zero explicitly by call to
1879 * sk_alloc() so need not be done here.
1881 static int tcp_v6_init_sock(struct sock *sk)
1883 struct inet_connection_sock *icsk = inet_csk(sk);
1884 struct tcp_sock *tp = tcp_sk(sk);
1886 skb_queue_head_init(&tp->out_of_order_queue);
1887 tcp_init_xmit_timers(sk);
1888 tcp_prequeue_init(tp);
1890 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1891 tp->mdev = TCP_TIMEOUT_INIT;
1893 /* So many TCP implementations out there (incorrectly) count the
1894 * initial SYN frame in their delayed-ACK and congestion control
1895 * algorithms that we must have the following bandaid to talk
1896 * efficiently to them. -DaveM
1900 /* See draft-stevens-tcpca-spec-01 for discussion of the
1901 * initialization of these values.
1903 tp->snd_ssthresh = 0x7fffffff;
1904 tp->snd_cwnd_clamp = ~0;
1905 tp->mss_cache = 536;
1907 tp->reordering = sysctl_tcp_reordering;
1909 sk->sk_state = TCP_CLOSE;
1911 icsk->icsk_af_ops = &ipv6_specific;
1912 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1913 icsk->icsk_sync_mss = tcp_sync_mss;
1914 sk->sk_write_space = sk_stream_write_space;
1915 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1917 #ifdef CONFIG_TCP_MD5SIG
1918 tp->af_specific = &tcp_sock_ipv6_specific;
1921 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1922 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1924 atomic_inc(&tcp_sockets_allocated);
1929 static int tcp_v6_destroy_sock(struct sock *sk)
1931 #ifdef CONFIG_TCP_MD5SIG
1932 /* Clean up the MD5 key list */
1933 if (tcp_sk(sk)->md5sig_info)
1934 tcp_v6_clear_md5_list(sk);
1936 tcp_v4_destroy_sock(sk);
1937 return inet6_destroy_sock(sk);
1940 #ifdef CONFIG_PROC_FS
1941 /* Proc filesystem TCPv6 sock list dumping. */
1942 static void get_openreq6(struct seq_file *seq,
1943 struct sock *sk, struct request_sock *req, int i, int uid)
1945 int ttd = req->expires - jiffies;
1946 struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1947 struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1953 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1954 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1956 src->s6_addr32[0], src->s6_addr32[1],
1957 src->s6_addr32[2], src->s6_addr32[3],
1958 ntohs(inet_sk(sk)->sport),
1959 dest->s6_addr32[0], dest->s6_addr32[1],
1960 dest->s6_addr32[2], dest->s6_addr32[3],
1961 ntohs(inet_rsk(req)->rmt_port),
1963 0,0, /* could print option size, but that is af dependent. */
1964 1, /* timers active (only the expire timer) */
1965 jiffies_to_clock_t(ttd),
1968 0, /* non standard timer */
1969 0, /* open_requests have no inode */
1973 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1975 struct in6_addr *dest, *src;
1978 unsigned long timer_expires;
1979 struct inet_sock *inet = inet_sk(sp);
1980 struct tcp_sock *tp = tcp_sk(sp);
1981 const struct inet_connection_sock *icsk = inet_csk(sp);
1982 struct ipv6_pinfo *np = inet6_sk(sp);
1985 src = &np->rcv_saddr;
1986 destp = ntohs(inet->dport);
1987 srcp = ntohs(inet->sport);
1989 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1991 timer_expires = icsk->icsk_timeout;
1992 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1994 timer_expires = icsk->icsk_timeout;
1995 } else if (timer_pending(&sp->sk_timer)) {
1997 timer_expires = sp->sk_timer.expires;
2000 timer_expires = jiffies;
2004 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2005 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
2007 src->s6_addr32[0], src->s6_addr32[1],
2008 src->s6_addr32[2], src->s6_addr32[3], srcp,
2009 dest->s6_addr32[0], dest->s6_addr32[1],
2010 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2012 tp->write_seq-tp->snd_una,
2013 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2015 jiffies_to_clock_t(timer_expires - jiffies),
2016 icsk->icsk_retransmits,
2018 icsk->icsk_probes_out,
2020 atomic_read(&sp->sk_refcnt), sp,
2023 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2024 tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2028 static void get_timewait6_sock(struct seq_file *seq,
2029 struct inet_timewait_sock *tw, int i)
2031 struct in6_addr *dest, *src;
2033 struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2034 int ttd = tw->tw_ttd - jiffies;
2039 dest = &tw6->tw_v6_daddr;
2040 src = &tw6->tw_v6_rcv_saddr;
2041 destp = ntohs(tw->tw_dport);
2042 srcp = ntohs(tw->tw_sport);
2045 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2046 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2048 src->s6_addr32[0], src->s6_addr32[1],
2049 src->s6_addr32[2], src->s6_addr32[3], srcp,
2050 dest->s6_addr32[0], dest->s6_addr32[1],
2051 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2052 tw->tw_substate, 0, 0,
2053 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2054 atomic_read(&tw->tw_refcnt), tw);
2057 static int tcp6_seq_show(struct seq_file *seq, void *v)
2059 struct tcp_iter_state *st;
2061 if (v == SEQ_START_TOKEN) {
2066 "st tx_queue rx_queue tr tm->when retrnsmt"
2067 " uid timeout inode\n");
2072 switch (st->state) {
2073 case TCP_SEQ_STATE_LISTENING:
2074 case TCP_SEQ_STATE_ESTABLISHED:
2075 get_tcp6_sock(seq, v, st->num);
2077 case TCP_SEQ_STATE_OPENREQ:
2078 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2080 case TCP_SEQ_STATE_TIME_WAIT:
2081 get_timewait6_sock(seq, v, st->num);
2088 static struct file_operations tcp6_seq_fops;
2089 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2090 .owner = THIS_MODULE,
2093 .seq_show = tcp6_seq_show,
2094 .seq_fops = &tcp6_seq_fops,
2097 int __init tcp6_proc_init(void)
2099 return tcp_proc_register(&tcp6_seq_afinfo);
2102 void tcp6_proc_exit(void)
2104 tcp_proc_unregister(&tcp6_seq_afinfo);
2108 DEFINE_PROTO_INUSE(tcpv6)
2110 struct proto tcpv6_prot = {
2112 .owner = THIS_MODULE,
2114 .connect = tcp_v6_connect,
2115 .disconnect = tcp_disconnect,
2116 .accept = inet_csk_accept,
2118 .init = tcp_v6_init_sock,
2119 .destroy = tcp_v6_destroy_sock,
2120 .shutdown = tcp_shutdown,
2121 .setsockopt = tcp_setsockopt,
2122 .getsockopt = tcp_getsockopt,
2123 .recvmsg = tcp_recvmsg,
2124 .backlog_rcv = tcp_v6_do_rcv,
2125 .hash = tcp_v6_hash,
2126 .unhash = inet_unhash,
2127 .get_port = inet_csk_get_port,
2128 .enter_memory_pressure = tcp_enter_memory_pressure,
2129 .sockets_allocated = &tcp_sockets_allocated,
2130 .memory_allocated = &tcp_memory_allocated,
2131 .memory_pressure = &tcp_memory_pressure,
2132 .orphan_count = &tcp_orphan_count,
2133 .sysctl_mem = sysctl_tcp_mem,
2134 .sysctl_wmem = sysctl_tcp_wmem,
2135 .sysctl_rmem = sysctl_tcp_rmem,
2136 .max_header = MAX_TCP_HEADER,
2137 .obj_size = sizeof(struct tcp6_sock),
2138 .twsk_prot = &tcp6_timewait_sock_ops,
2139 .rsk_prot = &tcp6_request_sock_ops,
2140 .hashinfo = &tcp_hashinfo,
2141 #ifdef CONFIG_COMPAT
2142 .compat_setsockopt = compat_tcp_setsockopt,
2143 .compat_getsockopt = compat_tcp_getsockopt,
2145 REF_PROTO_INUSE(tcpv6)
2148 static struct inet6_protocol tcpv6_protocol = {
2149 .handler = tcp_v6_rcv,
2150 .err_handler = tcp_v6_err,
2151 .gso_send_check = tcp_v6_gso_send_check,
2152 .gso_segment = tcp_tso_segment,
2153 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2156 static struct inet_protosw tcpv6_protosw = {
2157 .type = SOCK_STREAM,
2158 .protocol = IPPROTO_TCP,
2159 .prot = &tcpv6_prot,
2160 .ops = &inet6_stream_ops,
2163 .flags = INET_PROTOSW_PERMANENT |
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 = inet_csk_ctl_sock_create(&tcp6_socket, PF_INET6,
2181 SOCK_RAW, IPPROTO_TCP);
2183 goto out_tcpv6_protosw;
2188 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2190 inet6_unregister_protosw(&tcpv6_protosw);
2194 void tcpv6_exit(void)
2196 sock_release(tcp6_socket);
2197 inet6_unregister_protosw(&tcpv6_protosw);
2198 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);