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/bottom_half.h>
27 #include <linux/module.h>
28 #include <linux/errno.h>
29 #include <linux/types.h>
30 #include <linux/socket.h>
31 #include <linux/sockios.h>
32 #include <linux/net.h>
33 #include <linux/jiffies.h>
35 #include <linux/in6.h>
36 #include <linux/netdevice.h>
37 #include <linux/init.h>
38 #include <linux/jhash.h>
39 #include <linux/ipsec.h>
40 #include <linux/times.h>
41 #include <linux/slab.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>
63 #include <net/inet_common.h>
64 #include <net/secure_seq.h>
65 #include <net/tcp_memcontrol.h>
67 #include <asm/uaccess.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
72 #include <linux/crypto.h>
73 #include <linux/scatterlist.h>
75 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
76 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
77 struct request_sock *req);
79 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
81 static const struct inet_connection_sock_af_ops ipv6_mapped;
82 static const struct inet_connection_sock_af_ops ipv6_specific;
83 #ifdef CONFIG_TCP_MD5SIG
84 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
85 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
87 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
88 const struct in6_addr *addr)
94 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
96 struct dst_entry *dst = skb_dst(skb);
97 const struct rt6_info *rt = (const struct rt6_info *)dst;
101 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
103 inet6_sk(sk)->rx_dst_cookie = rt->rt6i_node->fn_sernum;
106 static void tcp_v6_hash(struct sock *sk)
108 if (sk->sk_state != TCP_CLOSE) {
109 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
114 __inet6_hash(sk, NULL);
119 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
121 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
122 ipv6_hdr(skb)->saddr.s6_addr32,
124 tcp_hdr(skb)->source);
127 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
130 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
131 struct inet_sock *inet = inet_sk(sk);
132 struct inet_connection_sock *icsk = inet_csk(sk);
133 struct ipv6_pinfo *np = inet6_sk(sk);
134 struct tcp_sock *tp = tcp_sk(sk);
135 struct in6_addr *saddr = NULL, *final_p, final;
138 struct dst_entry *dst;
142 if (addr_len < SIN6_LEN_RFC2133)
145 if (usin->sin6_family != AF_INET6)
146 return -EAFNOSUPPORT;
148 memset(&fl6, 0, sizeof(fl6));
151 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
152 IP6_ECN_flow_init(fl6.flowlabel);
153 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
154 struct ip6_flowlabel *flowlabel;
155 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
156 if (flowlabel == NULL)
158 usin->sin6_addr = flowlabel->dst;
159 fl6_sock_release(flowlabel);
164 * connect() to INADDR_ANY means loopback (BSD'ism).
167 if(ipv6_addr_any(&usin->sin6_addr))
168 usin->sin6_addr.s6_addr[15] = 0x1;
170 addr_type = ipv6_addr_type(&usin->sin6_addr);
172 if(addr_type & IPV6_ADDR_MULTICAST)
175 if (addr_type&IPV6_ADDR_LINKLOCAL) {
176 if (addr_len >= sizeof(struct sockaddr_in6) &&
177 usin->sin6_scope_id) {
178 /* If interface is set while binding, indices
181 if (sk->sk_bound_dev_if &&
182 sk->sk_bound_dev_if != usin->sin6_scope_id)
185 sk->sk_bound_dev_if = usin->sin6_scope_id;
188 /* Connect to link-local address requires an interface */
189 if (!sk->sk_bound_dev_if)
193 if (tp->rx_opt.ts_recent_stamp &&
194 !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
195 tp->rx_opt.ts_recent = 0;
196 tp->rx_opt.ts_recent_stamp = 0;
200 np->daddr = usin->sin6_addr;
201 np->flow_label = fl6.flowlabel;
207 if (addr_type == IPV6_ADDR_MAPPED) {
208 u32 exthdrlen = icsk->icsk_ext_hdr_len;
209 struct sockaddr_in sin;
211 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
213 if (__ipv6_only_sock(sk))
216 sin.sin_family = AF_INET;
217 sin.sin_port = usin->sin6_port;
218 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
220 icsk->icsk_af_ops = &ipv6_mapped;
221 sk->sk_backlog_rcv = tcp_v4_do_rcv;
222 #ifdef CONFIG_TCP_MD5SIG
223 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
226 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
229 icsk->icsk_ext_hdr_len = exthdrlen;
230 icsk->icsk_af_ops = &ipv6_specific;
231 sk->sk_backlog_rcv = tcp_v6_do_rcv;
232 #ifdef CONFIG_TCP_MD5SIG
233 tp->af_specific = &tcp_sock_ipv6_specific;
237 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
238 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
245 if (!ipv6_addr_any(&np->rcv_saddr))
246 saddr = &np->rcv_saddr;
248 fl6.flowi6_proto = IPPROTO_TCP;
249 fl6.daddr = np->daddr;
250 fl6.saddr = saddr ? *saddr : np->saddr;
251 fl6.flowi6_oif = sk->sk_bound_dev_if;
252 fl6.flowi6_mark = sk->sk_mark;
253 fl6.fl6_dport = usin->sin6_port;
254 fl6.fl6_sport = inet->inet_sport;
256 final_p = fl6_update_dst(&fl6, np->opt, &final);
258 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
260 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, true);
268 np->rcv_saddr = *saddr;
271 /* set the source address */
273 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
275 sk->sk_gso_type = SKB_GSO_TCPV6;
276 __ip6_dst_store(sk, dst, NULL, NULL);
278 rt = (struct rt6_info *) dst;
279 if (tcp_death_row.sysctl_tw_recycle &&
280 !tp->rx_opt.ts_recent_stamp &&
281 ipv6_addr_equal(&rt->rt6i_dst.addr, &np->daddr))
282 tcp_fetch_timewait_stamp(sk, dst);
284 icsk->icsk_ext_hdr_len = 0;
286 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
289 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
291 inet->inet_dport = usin->sin6_port;
293 tcp_set_state(sk, TCP_SYN_SENT);
294 err = inet6_hash_connect(&tcp_death_row, sk);
298 if (!tp->write_seq && likely(!tp->repair))
299 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
304 err = tcp_connect(sk);
311 tcp_set_state(sk, TCP_CLOSE);
314 inet->inet_dport = 0;
315 sk->sk_route_caps = 0;
319 static void tcp_v6_mtu_reduced(struct sock *sk)
321 struct dst_entry *dst;
323 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
326 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
330 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
331 tcp_sync_mss(sk, dst_mtu(dst));
332 tcp_simple_retransmit(sk);
336 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
337 u8 type, u8 code, int offset, __be32 info)
339 const struct ipv6hdr *hdr = (const struct ipv6hdr*)skb->data;
340 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
341 struct ipv6_pinfo *np;
346 struct net *net = dev_net(skb->dev);
348 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
349 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
352 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
357 if (sk->sk_state == TCP_TIME_WAIT) {
358 inet_twsk_put(inet_twsk(sk));
363 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
364 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
366 if (sk->sk_state == TCP_CLOSE)
369 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
370 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
375 seq = ntohl(th->seq);
376 if (sk->sk_state != TCP_LISTEN &&
377 !between(seq, tp->snd_una, tp->snd_nxt)) {
378 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
384 if (type == NDISC_REDIRECT) {
385 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
388 dst->ops->redirect(dst, sk, skb);
392 if (type == ICMPV6_PKT_TOOBIG) {
393 /* We are not interested in TCP_LISTEN and open_requests
394 * (SYN-ACKs send out by Linux are always <576bytes so
395 * they should go through unfragmented).
397 if (sk->sk_state == TCP_LISTEN)
400 tp->mtu_info = ntohl(info);
401 if (!sock_owned_by_user(sk))
402 tcp_v6_mtu_reduced(sk);
403 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
409 icmpv6_err_convert(type, code, &err);
411 /* Might be for an request_sock */
412 switch (sk->sk_state) {
413 struct request_sock *req, **prev;
415 if (sock_owned_by_user(sk))
418 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
419 &hdr->saddr, inet6_iif(skb));
423 /* ICMPs are not backlogged, hence we cannot get
424 * an established socket here.
426 WARN_ON(req->sk != NULL);
428 if (seq != tcp_rsk(req)->snt_isn) {
429 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
433 inet_csk_reqsk_queue_drop(sk, req, prev);
434 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
438 case TCP_SYN_RECV: /* Cannot happen.
439 It can, it SYNs are crossed. --ANK */
440 if (!sock_owned_by_user(sk)) {
442 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
446 sk->sk_err_soft = err;
450 if (!sock_owned_by_user(sk) && np->recverr) {
452 sk->sk_error_report(sk);
454 sk->sk_err_soft = err;
462 static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
464 struct request_sock *req,
465 struct request_values *rvp,
468 struct inet6_request_sock *treq = inet6_rsk(req);
469 struct ipv6_pinfo *np = inet6_sk(sk);
470 struct sk_buff * skb;
473 /* First, grab a route. */
474 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
477 skb = tcp_make_synack(sk, dst, req, rvp, NULL);
480 __tcp_v6_send_check(skb, &treq->loc_addr, &treq->rmt_addr);
482 fl6->daddr = treq->rmt_addr;
483 skb_set_queue_mapping(skb, queue_mapping);
484 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
485 err = net_xmit_eval(err);
492 static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req,
493 struct request_values *rvp)
498 res = tcp_v6_send_synack(sk, NULL, &fl6, req, rvp, 0);
500 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
504 static void tcp_v6_reqsk_destructor(struct request_sock *req)
506 kfree_skb(inet6_rsk(req)->pktopts);
509 #ifdef CONFIG_TCP_MD5SIG
510 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
511 const struct in6_addr *addr)
513 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
516 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
517 struct sock *addr_sk)
519 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
522 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
523 struct request_sock *req)
525 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
528 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
531 struct tcp_md5sig cmd;
532 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
534 if (optlen < sizeof(cmd))
537 if (copy_from_user(&cmd, optval, sizeof(cmd)))
540 if (sin6->sin6_family != AF_INET6)
543 if (!cmd.tcpm_keylen) {
544 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
545 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
547 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
551 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
554 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
555 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
556 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
558 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
559 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
562 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
563 const struct in6_addr *daddr,
564 const struct in6_addr *saddr, int nbytes)
566 struct tcp6_pseudohdr *bp;
567 struct scatterlist sg;
569 bp = &hp->md5_blk.ip6;
570 /* 1. TCP pseudo-header (RFC2460) */
573 bp->protocol = cpu_to_be32(IPPROTO_TCP);
574 bp->len = cpu_to_be32(nbytes);
576 sg_init_one(&sg, bp, sizeof(*bp));
577 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
580 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
581 const struct in6_addr *daddr, struct in6_addr *saddr,
582 const struct tcphdr *th)
584 struct tcp_md5sig_pool *hp;
585 struct hash_desc *desc;
587 hp = tcp_get_md5sig_pool();
589 goto clear_hash_noput;
590 desc = &hp->md5_desc;
592 if (crypto_hash_init(desc))
594 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
596 if (tcp_md5_hash_header(hp, th))
598 if (tcp_md5_hash_key(hp, key))
600 if (crypto_hash_final(desc, md5_hash))
603 tcp_put_md5sig_pool();
607 tcp_put_md5sig_pool();
609 memset(md5_hash, 0, 16);
613 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
614 const struct sock *sk,
615 const struct request_sock *req,
616 const struct sk_buff *skb)
618 const struct in6_addr *saddr, *daddr;
619 struct tcp_md5sig_pool *hp;
620 struct hash_desc *desc;
621 const struct tcphdr *th = tcp_hdr(skb);
624 saddr = &inet6_sk(sk)->saddr;
625 daddr = &inet6_sk(sk)->daddr;
627 saddr = &inet6_rsk(req)->loc_addr;
628 daddr = &inet6_rsk(req)->rmt_addr;
630 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
631 saddr = &ip6h->saddr;
632 daddr = &ip6h->daddr;
635 hp = tcp_get_md5sig_pool();
637 goto clear_hash_noput;
638 desc = &hp->md5_desc;
640 if (crypto_hash_init(desc))
643 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
645 if (tcp_md5_hash_header(hp, th))
647 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
649 if (tcp_md5_hash_key(hp, key))
651 if (crypto_hash_final(desc, md5_hash))
654 tcp_put_md5sig_pool();
658 tcp_put_md5sig_pool();
660 memset(md5_hash, 0, 16);
664 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
666 const __u8 *hash_location = NULL;
667 struct tcp_md5sig_key *hash_expected;
668 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
669 const struct tcphdr *th = tcp_hdr(skb);
673 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
674 hash_location = tcp_parse_md5sig_option(th);
676 /* We've parsed the options - do we have a hash? */
677 if (!hash_expected && !hash_location)
680 if (hash_expected && !hash_location) {
681 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
685 if (!hash_expected && hash_location) {
686 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
690 /* check the signature */
691 genhash = tcp_v6_md5_hash_skb(newhash,
695 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
696 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
697 genhash ? "failed" : "mismatch",
698 &ip6h->saddr, ntohs(th->source),
699 &ip6h->daddr, ntohs(th->dest));
706 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
708 .obj_size = sizeof(struct tcp6_request_sock),
709 .rtx_syn_ack = tcp_v6_rtx_synack,
710 .send_ack = tcp_v6_reqsk_send_ack,
711 .destructor = tcp_v6_reqsk_destructor,
712 .send_reset = tcp_v6_send_reset,
713 .syn_ack_timeout = tcp_syn_ack_timeout,
716 #ifdef CONFIG_TCP_MD5SIG
717 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
718 .md5_lookup = tcp_v6_reqsk_md5_lookup,
719 .calc_md5_hash = tcp_v6_md5_hash_skb,
723 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
724 u32 tsval, u32 tsecr,
725 struct tcp_md5sig_key *key, int rst, u8 tclass)
727 const struct tcphdr *th = tcp_hdr(skb);
729 struct sk_buff *buff;
731 struct net *net = dev_net(skb_dst(skb)->dev);
732 struct sock *ctl_sk = net->ipv6.tcp_sk;
733 unsigned int tot_len = sizeof(struct tcphdr);
734 struct dst_entry *dst;
738 tot_len += TCPOLEN_TSTAMP_ALIGNED;
739 #ifdef CONFIG_TCP_MD5SIG
741 tot_len += TCPOLEN_MD5SIG_ALIGNED;
744 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
749 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
751 t1 = (struct tcphdr *) skb_push(buff, tot_len);
752 skb_reset_transport_header(buff);
754 /* Swap the send and the receive. */
755 memset(t1, 0, sizeof(*t1));
756 t1->dest = th->source;
757 t1->source = th->dest;
758 t1->doff = tot_len / 4;
759 t1->seq = htonl(seq);
760 t1->ack_seq = htonl(ack);
761 t1->ack = !rst || !th->ack;
763 t1->window = htons(win);
765 topt = (__be32 *)(t1 + 1);
768 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
769 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
770 *topt++ = htonl(tsval);
771 *topt++ = htonl(tsecr);
774 #ifdef CONFIG_TCP_MD5SIG
776 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
777 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
778 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
779 &ipv6_hdr(skb)->saddr,
780 &ipv6_hdr(skb)->daddr, t1);
784 memset(&fl6, 0, sizeof(fl6));
785 fl6.daddr = ipv6_hdr(skb)->saddr;
786 fl6.saddr = ipv6_hdr(skb)->daddr;
788 buff->ip_summed = CHECKSUM_PARTIAL;
791 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
793 fl6.flowi6_proto = IPPROTO_TCP;
794 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
795 fl6.flowi6_oif = inet6_iif(skb);
796 fl6.fl6_dport = t1->dest;
797 fl6.fl6_sport = t1->source;
798 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
800 /* Pass a socket to ip6_dst_lookup either it is for RST
801 * Underlying function will use this to retrieve the network
804 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false);
806 skb_dst_set(buff, dst);
807 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
808 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
810 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
817 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
819 const struct tcphdr *th = tcp_hdr(skb);
820 u32 seq = 0, ack_seq = 0;
821 struct tcp_md5sig_key *key = NULL;
822 #ifdef CONFIG_TCP_MD5SIG
823 const __u8 *hash_location = NULL;
824 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
825 unsigned char newhash[16];
827 struct sock *sk1 = NULL;
833 if (!ipv6_unicast_destination(skb))
836 #ifdef CONFIG_TCP_MD5SIG
837 hash_location = tcp_parse_md5sig_option(th);
838 if (!sk && hash_location) {
840 * active side is lost. Try to find listening socket through
841 * source port, and then find md5 key through listening socket.
842 * we are not loose security here:
843 * Incoming packet is checked with md5 hash with finding key,
844 * no RST generated if md5 hash doesn't match.
846 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
847 &tcp_hashinfo, &ipv6h->saddr,
848 th->source, &ipv6h->daddr,
849 ntohs(th->source), inet6_iif(skb));
854 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
858 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
859 if (genhash || memcmp(hash_location, newhash, 16) != 0)
862 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
867 seq = ntohl(th->ack_seq);
869 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
872 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, 0, key, 1, 0);
874 #ifdef CONFIG_TCP_MD5SIG
883 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
884 u32 win, u32 tsval, u32 tsecr,
885 struct tcp_md5sig_key *key, u8 tclass)
887 tcp_v6_send_response(skb, seq, ack, win, tsval, tsecr, key, 0, tclass);
890 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
892 struct inet_timewait_sock *tw = inet_twsk(sk);
893 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
895 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
896 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
897 tcp_time_stamp + tcptw->tw_ts_offset,
898 tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw),
904 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
905 struct request_sock *req)
907 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1,
908 req->rcv_wnd, tcp_time_stamp, req->ts_recent,
909 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr), 0);
913 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
915 struct request_sock *req, **prev;
916 const struct tcphdr *th = tcp_hdr(skb);
919 /* Find possible connection requests. */
920 req = inet6_csk_search_req(sk, &prev, th->source,
921 &ipv6_hdr(skb)->saddr,
922 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
924 return tcp_check_req(sk, skb, req, prev, false);
926 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
927 &ipv6_hdr(skb)->saddr, th->source,
928 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
931 if (nsk->sk_state != TCP_TIME_WAIT) {
935 inet_twsk_put(inet_twsk(nsk));
939 #ifdef CONFIG_SYN_COOKIES
941 sk = cookie_v6_check(sk, skb);
946 /* FIXME: this is substantially similar to the ipv4 code.
947 * Can some kind of merge be done? -- erics
949 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
951 struct tcp_extend_values tmp_ext;
952 struct tcp_options_received tmp_opt;
953 const u8 *hash_location;
954 struct request_sock *req;
955 struct inet6_request_sock *treq;
956 struct ipv6_pinfo *np = inet6_sk(sk);
957 struct tcp_sock *tp = tcp_sk(sk);
958 __u32 isn = TCP_SKB_CB(skb)->when;
959 struct dst_entry *dst = NULL;
961 bool want_cookie = false;
963 if (skb->protocol == htons(ETH_P_IP))
964 return tcp_v4_conn_request(sk, skb);
966 if (!ipv6_unicast_destination(skb))
969 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
970 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
975 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
976 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
980 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
984 #ifdef CONFIG_TCP_MD5SIG
985 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
988 tcp_clear_options(&tmp_opt);
989 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
990 tmp_opt.user_mss = tp->rx_opt.user_mss;
991 tcp_parse_options(skb, &tmp_opt, &hash_location, 0, NULL);
993 if (tmp_opt.cookie_plus > 0 &&
994 tmp_opt.saw_tstamp &&
995 !tp->rx_opt.cookie_out_never &&
996 (sysctl_tcp_cookie_size > 0 ||
997 (tp->cookie_values != NULL &&
998 tp->cookie_values->cookie_desired > 0))) {
1001 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1002 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1004 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1007 /* Secret recipe starts with IP addresses */
1008 d = (__force u32 *)&ipv6_hdr(skb)->daddr.s6_addr32[0];
1013 d = (__force u32 *)&ipv6_hdr(skb)->saddr.s6_addr32[0];
1019 /* plus variable length Initiator Cookie */
1022 *c++ ^= *hash_location++;
1024 want_cookie = false; /* not our kind of cookie */
1025 tmp_ext.cookie_out_never = 0; /* false */
1026 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1027 } else if (!tp->rx_opt.cookie_in_always) {
1028 /* redundant indications, but ensure initialization. */
1029 tmp_ext.cookie_out_never = 1; /* true */
1030 tmp_ext.cookie_plus = 0;
1034 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1036 if (want_cookie && !tmp_opt.saw_tstamp)
1037 tcp_clear_options(&tmp_opt);
1039 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1040 tcp_openreq_init(req, &tmp_opt, skb);
1042 treq = inet6_rsk(req);
1043 treq->rmt_addr = ipv6_hdr(skb)->saddr;
1044 treq->loc_addr = ipv6_hdr(skb)->daddr;
1045 if (!want_cookie || tmp_opt.tstamp_ok)
1046 TCP_ECN_create_request(req, skb, sock_net(sk));
1048 treq->iif = sk->sk_bound_dev_if;
1050 /* So that link locals have meaning */
1051 if (!sk->sk_bound_dev_if &&
1052 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1053 treq->iif = inet6_iif(skb);
1056 if (ipv6_opt_accepted(sk, skb) ||
1057 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1058 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1059 atomic_inc(&skb->users);
1060 treq->pktopts = skb;
1064 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1065 req->cookie_ts = tmp_opt.tstamp_ok;
1069 /* VJ's idea. We save last timestamp seen
1070 * from the destination in peer table, when entering
1071 * state TIME-WAIT, and check against it before
1072 * accepting new connection request.
1074 * If "isn" is not zero, this request hit alive
1075 * timewait bucket, so that all the necessary checks
1076 * are made in the function processing timewait state.
1078 if (tmp_opt.saw_tstamp &&
1079 tcp_death_row.sysctl_tw_recycle &&
1080 (dst = inet6_csk_route_req(sk, &fl6, req)) != NULL) {
1081 if (!tcp_peer_is_proven(req, dst, true)) {
1082 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1083 goto drop_and_release;
1086 /* Kill the following clause, if you dislike this way. */
1087 else if (!sysctl_tcp_syncookies &&
1088 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1089 (sysctl_max_syn_backlog >> 2)) &&
1090 !tcp_peer_is_proven(req, dst, false)) {
1091 /* Without syncookies last quarter of
1092 * backlog is filled with destinations,
1093 * proven to be alive.
1094 * It means that we continue to communicate
1095 * to destinations, already remembered
1096 * to the moment of synflood.
1098 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1099 &treq->rmt_addr, ntohs(tcp_hdr(skb)->source));
1100 goto drop_and_release;
1103 isn = tcp_v6_init_sequence(skb);
1106 tcp_rsk(req)->snt_isn = isn;
1108 if (security_inet_conn_request(sk, skb, req))
1109 goto drop_and_release;
1111 if (tcp_v6_send_synack(sk, dst, &fl6, req,
1112 (struct request_values *)&tmp_ext,
1113 skb_get_queue_mapping(skb)) ||
1117 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1118 tcp_rsk(req)->listener = NULL;
1119 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1127 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1128 return 0; /* don't send reset */
1131 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1132 struct request_sock *req,
1133 struct dst_entry *dst)
1135 struct inet6_request_sock *treq;
1136 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1137 struct tcp6_sock *newtcp6sk;
1138 struct inet_sock *newinet;
1139 struct tcp_sock *newtp;
1141 #ifdef CONFIG_TCP_MD5SIG
1142 struct tcp_md5sig_key *key;
1146 if (skb->protocol == htons(ETH_P_IP)) {
1151 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1156 newtcp6sk = (struct tcp6_sock *)newsk;
1157 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1159 newinet = inet_sk(newsk);
1160 newnp = inet6_sk(newsk);
1161 newtp = tcp_sk(newsk);
1163 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1165 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr);
1167 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1169 newnp->rcv_saddr = newnp->saddr;
1171 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1172 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1173 #ifdef CONFIG_TCP_MD5SIG
1174 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1177 newnp->ipv6_ac_list = NULL;
1178 newnp->ipv6_fl_list = NULL;
1179 newnp->pktoptions = NULL;
1181 newnp->mcast_oif = inet6_iif(skb);
1182 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1183 newnp->rcv_tclass = ipv6_get_dsfield(ipv6_hdr(skb));
1186 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1187 * here, tcp_create_openreq_child now does this for us, see the comment in
1188 * that function for the gory details. -acme
1191 /* It is tricky place. Until this moment IPv4 tcp
1192 worked with IPv6 icsk.icsk_af_ops.
1195 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1200 treq = inet6_rsk(req);
1202 if (sk_acceptq_is_full(sk))
1206 dst = inet6_csk_route_req(sk, &fl6, req);
1211 newsk = tcp_create_openreq_child(sk, req, skb);
1216 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1217 * count here, tcp_create_openreq_child now does this for us, see the
1218 * comment in that function for the gory details. -acme
1221 newsk->sk_gso_type = SKB_GSO_TCPV6;
1222 __ip6_dst_store(newsk, dst, NULL, NULL);
1223 inet6_sk_rx_dst_set(newsk, skb);
1225 newtcp6sk = (struct tcp6_sock *)newsk;
1226 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1228 newtp = tcp_sk(newsk);
1229 newinet = inet_sk(newsk);
1230 newnp = inet6_sk(newsk);
1232 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1234 newnp->daddr = treq->rmt_addr;
1235 newnp->saddr = treq->loc_addr;
1236 newnp->rcv_saddr = treq->loc_addr;
1237 newsk->sk_bound_dev_if = treq->iif;
1239 /* Now IPv6 options...
1241 First: no IPv4 options.
1243 newinet->inet_opt = NULL;
1244 newnp->ipv6_ac_list = NULL;
1245 newnp->ipv6_fl_list = NULL;
1248 newnp->rxopt.all = np->rxopt.all;
1250 /* Clone pktoptions received with SYN */
1251 newnp->pktoptions = NULL;
1252 if (treq->pktopts != NULL) {
1253 newnp->pktoptions = skb_clone(treq->pktopts,
1254 sk_gfp_atomic(sk, GFP_ATOMIC));
1255 consume_skb(treq->pktopts);
1256 treq->pktopts = NULL;
1257 if (newnp->pktoptions)
1258 skb_set_owner_r(newnp->pktoptions, newsk);
1261 newnp->mcast_oif = inet6_iif(skb);
1262 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1263 newnp->rcv_tclass = ipv6_get_dsfield(ipv6_hdr(skb));
1265 /* Clone native IPv6 options from listening socket (if any)
1267 Yes, keeping reference count would be much more clever,
1268 but we make one more one thing there: reattach optmem
1272 newnp->opt = ipv6_dup_options(newsk, np->opt);
1274 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1276 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1277 newnp->opt->opt_flen);
1279 tcp_mtup_init(newsk);
1280 tcp_sync_mss(newsk, dst_mtu(dst));
1281 newtp->advmss = dst_metric_advmss(dst);
1282 if (tcp_sk(sk)->rx_opt.user_mss &&
1283 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1284 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1286 tcp_initialize_rcv_mss(newsk);
1287 tcp_synack_rtt_meas(newsk, req);
1288 newtp->total_retrans = req->num_retrans;
1290 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1291 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1293 #ifdef CONFIG_TCP_MD5SIG
1294 /* Copy over the MD5 key from the original socket */
1295 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1296 /* We're using one, so create a matching key
1297 * on the newsk structure. If we fail to get
1298 * memory, then we end up not copying the key
1301 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newnp->daddr,
1302 AF_INET6, key->key, key->keylen,
1303 sk_gfp_atomic(sk, GFP_ATOMIC));
1307 if (__inet_inherit_port(sk, newsk) < 0) {
1308 inet_csk_prepare_forced_close(newsk);
1312 __inet6_hash(newsk, NULL);
1317 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1321 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1325 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1327 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1328 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1329 &ipv6_hdr(skb)->daddr, skb->csum)) {
1330 skb->ip_summed = CHECKSUM_UNNECESSARY;
1335 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1336 &ipv6_hdr(skb)->saddr,
1337 &ipv6_hdr(skb)->daddr, 0));
1339 if (skb->len <= 76) {
1340 return __skb_checksum_complete(skb);
1345 /* The socket must have it's spinlock held when we get
1348 * We have a potential double-lock case here, so even when
1349 * doing backlog processing we use the BH locking scheme.
1350 * This is because we cannot sleep with the original spinlock
1353 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1355 struct ipv6_pinfo *np = inet6_sk(sk);
1356 struct tcp_sock *tp;
1357 struct sk_buff *opt_skb = NULL;
1359 /* Imagine: socket is IPv6. IPv4 packet arrives,
1360 goes to IPv4 receive handler and backlogged.
1361 From backlog it always goes here. Kerboom...
1362 Fortunately, tcp_rcv_established and rcv_established
1363 handle them correctly, but it is not case with
1364 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1367 if (skb->protocol == htons(ETH_P_IP))
1368 return tcp_v4_do_rcv(sk, skb);
1370 #ifdef CONFIG_TCP_MD5SIG
1371 if (tcp_v6_inbound_md5_hash (sk, skb))
1375 if (sk_filter(sk, skb))
1379 * socket locking is here for SMP purposes as backlog rcv
1380 * is currently called with bh processing disabled.
1383 /* Do Stevens' IPV6_PKTOPTIONS.
1385 Yes, guys, it is the only place in our code, where we
1386 may make it not affecting IPv4.
1387 The rest of code is protocol independent,
1388 and I do not like idea to uglify IPv4.
1390 Actually, all the idea behind IPV6_PKTOPTIONS
1391 looks not very well thought. For now we latch
1392 options, received in the last packet, enqueued
1393 by tcp. Feel free to propose better solution.
1397 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1399 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1400 struct dst_entry *dst = sk->sk_rx_dst;
1402 sock_rps_save_rxhash(sk, skb);
1404 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1405 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1407 sk->sk_rx_dst = NULL;
1411 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1414 goto ipv6_pktoptions;
1418 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1421 if (sk->sk_state == TCP_LISTEN) {
1422 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1427 * Queue it on the new socket if the new socket is active,
1428 * otherwise we just shortcircuit this and continue with
1432 sock_rps_save_rxhash(nsk, skb);
1433 if (tcp_child_process(sk, nsk, skb))
1436 __kfree_skb(opt_skb);
1440 sock_rps_save_rxhash(sk, skb);
1442 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1445 goto ipv6_pktoptions;
1449 tcp_v6_send_reset(sk, skb);
1452 __kfree_skb(opt_skb);
1456 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1461 /* Do you ask, what is it?
1463 1. skb was enqueued by tcp.
1464 2. skb is added to tail of read queue, rather than out of order.
1465 3. socket is not in passive state.
1466 4. Finally, it really contains options, which user wants to receive.
1469 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1470 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1471 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1472 np->mcast_oif = inet6_iif(opt_skb);
1473 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1474 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1475 if (np->rxopt.bits.rxtclass)
1476 np->rcv_tclass = ipv6_get_dsfield(ipv6_hdr(skb));
1477 if (ipv6_opt_accepted(sk, opt_skb)) {
1478 skb_set_owner_r(opt_skb, sk);
1479 opt_skb = xchg(&np->pktoptions, opt_skb);
1481 __kfree_skb(opt_skb);
1482 opt_skb = xchg(&np->pktoptions, NULL);
1490 static int tcp_v6_rcv(struct sk_buff *skb)
1492 const struct tcphdr *th;
1493 const struct ipv6hdr *hdr;
1496 struct net *net = dev_net(skb->dev);
1498 if (skb->pkt_type != PACKET_HOST)
1502 * Count it even if it's bad.
1504 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1506 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1511 if (th->doff < sizeof(struct tcphdr)/4)
1513 if (!pskb_may_pull(skb, th->doff*4))
1516 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1520 hdr = ipv6_hdr(skb);
1521 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1522 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1523 skb->len - th->doff*4);
1524 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1525 TCP_SKB_CB(skb)->when = 0;
1526 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1527 TCP_SKB_CB(skb)->sacked = 0;
1529 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1534 if (sk->sk_state == TCP_TIME_WAIT)
1537 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1538 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1539 goto discard_and_relse;
1542 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1543 goto discard_and_relse;
1545 if (sk_filter(sk, skb))
1546 goto discard_and_relse;
1550 bh_lock_sock_nested(sk);
1552 if (!sock_owned_by_user(sk)) {
1553 #ifdef CONFIG_NET_DMA
1554 struct tcp_sock *tp = tcp_sk(sk);
1555 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1556 tp->ucopy.dma_chan = net_dma_find_channel();
1557 if (tp->ucopy.dma_chan)
1558 ret = tcp_v6_do_rcv(sk, skb);
1562 if (!tcp_prequeue(sk, skb))
1563 ret = tcp_v6_do_rcv(sk, skb);
1565 } else if (unlikely(sk_add_backlog(sk, skb,
1566 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1568 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1569 goto discard_and_relse;
1574 return ret ? -1 : 0;
1577 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1580 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1582 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1584 tcp_v6_send_reset(NULL, skb);
1601 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1602 inet_twsk_put(inet_twsk(sk));
1606 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1607 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1608 inet_twsk_put(inet_twsk(sk));
1612 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1617 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1618 &ipv6_hdr(skb)->saddr, th->source,
1619 &ipv6_hdr(skb)->daddr,
1620 ntohs(th->dest), inet6_iif(skb));
1622 struct inet_timewait_sock *tw = inet_twsk(sk);
1623 inet_twsk_deschedule(tw, &tcp_death_row);
1628 /* Fall through to ACK */
1631 tcp_v6_timewait_ack(sk, skb);
1635 case TCP_TW_SUCCESS:;
1640 static void tcp_v6_early_demux(struct sk_buff *skb)
1642 const struct ipv6hdr *hdr;
1643 const struct tcphdr *th;
1646 if (skb->pkt_type != PACKET_HOST)
1649 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1652 hdr = ipv6_hdr(skb);
1655 if (th->doff < sizeof(struct tcphdr) / 4)
1658 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1659 &hdr->saddr, th->source,
1660 &hdr->daddr, ntohs(th->dest),
1664 skb->destructor = sock_edemux;
1665 if (sk->sk_state != TCP_TIME_WAIT) {
1666 struct dst_entry *dst = sk->sk_rx_dst;
1669 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1671 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1672 skb_dst_set_noref(skb, dst);
1677 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1678 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1679 .twsk_unique = tcp_twsk_unique,
1680 .twsk_destructor= tcp_twsk_destructor,
1683 static const struct inet_connection_sock_af_ops ipv6_specific = {
1684 .queue_xmit = inet6_csk_xmit,
1685 .send_check = tcp_v6_send_check,
1686 .rebuild_header = inet6_sk_rebuild_header,
1687 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1688 .conn_request = tcp_v6_conn_request,
1689 .syn_recv_sock = tcp_v6_syn_recv_sock,
1690 .net_header_len = sizeof(struct ipv6hdr),
1691 .net_frag_header_len = sizeof(struct frag_hdr),
1692 .setsockopt = ipv6_setsockopt,
1693 .getsockopt = ipv6_getsockopt,
1694 .addr2sockaddr = inet6_csk_addr2sockaddr,
1695 .sockaddr_len = sizeof(struct sockaddr_in6),
1696 .bind_conflict = inet6_csk_bind_conflict,
1697 #ifdef CONFIG_COMPAT
1698 .compat_setsockopt = compat_ipv6_setsockopt,
1699 .compat_getsockopt = compat_ipv6_getsockopt,
1703 #ifdef CONFIG_TCP_MD5SIG
1704 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1705 .md5_lookup = tcp_v6_md5_lookup,
1706 .calc_md5_hash = tcp_v6_md5_hash_skb,
1707 .md5_parse = tcp_v6_parse_md5_keys,
1712 * TCP over IPv4 via INET6 API
1715 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1716 .queue_xmit = ip_queue_xmit,
1717 .send_check = tcp_v4_send_check,
1718 .rebuild_header = inet_sk_rebuild_header,
1719 .sk_rx_dst_set = inet_sk_rx_dst_set,
1720 .conn_request = tcp_v6_conn_request,
1721 .syn_recv_sock = tcp_v6_syn_recv_sock,
1722 .net_header_len = sizeof(struct iphdr),
1723 .setsockopt = ipv6_setsockopt,
1724 .getsockopt = ipv6_getsockopt,
1725 .addr2sockaddr = inet6_csk_addr2sockaddr,
1726 .sockaddr_len = sizeof(struct sockaddr_in6),
1727 .bind_conflict = inet6_csk_bind_conflict,
1728 #ifdef CONFIG_COMPAT
1729 .compat_setsockopt = compat_ipv6_setsockopt,
1730 .compat_getsockopt = compat_ipv6_getsockopt,
1734 #ifdef CONFIG_TCP_MD5SIG
1735 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1736 .md5_lookup = tcp_v4_md5_lookup,
1737 .calc_md5_hash = tcp_v4_md5_hash_skb,
1738 .md5_parse = tcp_v6_parse_md5_keys,
1742 /* NOTE: A lot of things set to zero explicitly by call to
1743 * sk_alloc() so need not be done here.
1745 static int tcp_v6_init_sock(struct sock *sk)
1747 struct inet_connection_sock *icsk = inet_csk(sk);
1751 icsk->icsk_af_ops = &ipv6_specific;
1753 #ifdef CONFIG_TCP_MD5SIG
1754 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1760 static void tcp_v6_destroy_sock(struct sock *sk)
1762 tcp_v4_destroy_sock(sk);
1763 inet6_destroy_sock(sk);
1766 #ifdef CONFIG_PROC_FS
1767 /* Proc filesystem TCPv6 sock list dumping. */
1768 static void get_openreq6(struct seq_file *seq,
1769 const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1771 int ttd = req->expires - jiffies;
1772 const struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1773 const struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1779 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1780 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1782 src->s6_addr32[0], src->s6_addr32[1],
1783 src->s6_addr32[2], src->s6_addr32[3],
1784 ntohs(inet_rsk(req)->loc_port),
1785 dest->s6_addr32[0], dest->s6_addr32[1],
1786 dest->s6_addr32[2], dest->s6_addr32[3],
1787 ntohs(inet_rsk(req)->rmt_port),
1789 0,0, /* could print option size, but that is af dependent. */
1790 1, /* timers active (only the expire timer) */
1791 jiffies_to_clock_t(ttd),
1793 from_kuid_munged(seq_user_ns(seq), uid),
1794 0, /* non standard timer */
1795 0, /* open_requests have no inode */
1799 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1801 const struct in6_addr *dest, *src;
1804 unsigned long timer_expires;
1805 const struct inet_sock *inet = inet_sk(sp);
1806 const struct tcp_sock *tp = tcp_sk(sp);
1807 const struct inet_connection_sock *icsk = inet_csk(sp);
1808 const struct ipv6_pinfo *np = inet6_sk(sp);
1811 src = &np->rcv_saddr;
1812 destp = ntohs(inet->inet_dport);
1813 srcp = ntohs(inet->inet_sport);
1815 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1817 timer_expires = icsk->icsk_timeout;
1818 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1820 timer_expires = icsk->icsk_timeout;
1821 } else if (timer_pending(&sp->sk_timer)) {
1823 timer_expires = sp->sk_timer.expires;
1826 timer_expires = jiffies;
1830 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1831 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %lu %lu %u %u %d\n",
1833 src->s6_addr32[0], src->s6_addr32[1],
1834 src->s6_addr32[2], src->s6_addr32[3], srcp,
1835 dest->s6_addr32[0], dest->s6_addr32[1],
1836 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1838 tp->write_seq-tp->snd_una,
1839 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1841 jiffies_delta_to_clock_t(timer_expires - jiffies),
1842 icsk->icsk_retransmits,
1843 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1844 icsk->icsk_probes_out,
1846 atomic_read(&sp->sk_refcnt), sp,
1847 jiffies_to_clock_t(icsk->icsk_rto),
1848 jiffies_to_clock_t(icsk->icsk_ack.ato),
1849 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1851 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
1855 static void get_timewait6_sock(struct seq_file *seq,
1856 struct inet_timewait_sock *tw, int i)
1858 const struct in6_addr *dest, *src;
1860 const struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
1861 long delta = tw->tw_ttd - jiffies;
1863 dest = &tw6->tw_v6_daddr;
1864 src = &tw6->tw_v6_rcv_saddr;
1865 destp = ntohs(tw->tw_dport);
1866 srcp = ntohs(tw->tw_sport);
1869 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1870 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1872 src->s6_addr32[0], src->s6_addr32[1],
1873 src->s6_addr32[2], src->s6_addr32[3], srcp,
1874 dest->s6_addr32[0], dest->s6_addr32[1],
1875 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1876 tw->tw_substate, 0, 0,
1877 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1878 atomic_read(&tw->tw_refcnt), tw);
1881 static int tcp6_seq_show(struct seq_file *seq, void *v)
1883 struct tcp_iter_state *st;
1885 if (v == SEQ_START_TOKEN) {
1890 "st tx_queue rx_queue tr tm->when retrnsmt"
1891 " uid timeout inode\n");
1896 switch (st->state) {
1897 case TCP_SEQ_STATE_LISTENING:
1898 case TCP_SEQ_STATE_ESTABLISHED:
1899 get_tcp6_sock(seq, v, st->num);
1901 case TCP_SEQ_STATE_OPENREQ:
1902 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1904 case TCP_SEQ_STATE_TIME_WAIT:
1905 get_timewait6_sock(seq, v, st->num);
1912 static const struct file_operations tcp6_afinfo_seq_fops = {
1913 .owner = THIS_MODULE,
1914 .open = tcp_seq_open,
1916 .llseek = seq_lseek,
1917 .release = seq_release_net
1920 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1923 .seq_fops = &tcp6_afinfo_seq_fops,
1925 .show = tcp6_seq_show,
1929 int __net_init tcp6_proc_init(struct net *net)
1931 return tcp_proc_register(net, &tcp6_seq_afinfo);
1934 void tcp6_proc_exit(struct net *net)
1936 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1940 struct proto tcpv6_prot = {
1942 .owner = THIS_MODULE,
1944 .connect = tcp_v6_connect,
1945 .disconnect = tcp_disconnect,
1946 .accept = inet_csk_accept,
1948 .init = tcp_v6_init_sock,
1949 .destroy = tcp_v6_destroy_sock,
1950 .shutdown = tcp_shutdown,
1951 .setsockopt = tcp_setsockopt,
1952 .getsockopt = tcp_getsockopt,
1953 .recvmsg = tcp_recvmsg,
1954 .sendmsg = tcp_sendmsg,
1955 .sendpage = tcp_sendpage,
1956 .backlog_rcv = tcp_v6_do_rcv,
1957 .release_cb = tcp_release_cb,
1958 .mtu_reduced = tcp_v6_mtu_reduced,
1959 .hash = tcp_v6_hash,
1960 .unhash = inet_unhash,
1961 .get_port = inet_csk_get_port,
1962 .enter_memory_pressure = tcp_enter_memory_pressure,
1963 .sockets_allocated = &tcp_sockets_allocated,
1964 .memory_allocated = &tcp_memory_allocated,
1965 .memory_pressure = &tcp_memory_pressure,
1966 .orphan_count = &tcp_orphan_count,
1967 .sysctl_wmem = sysctl_tcp_wmem,
1968 .sysctl_rmem = sysctl_tcp_rmem,
1969 .max_header = MAX_TCP_HEADER,
1970 .obj_size = sizeof(struct tcp6_sock),
1971 .slab_flags = SLAB_DESTROY_BY_RCU,
1972 .twsk_prot = &tcp6_timewait_sock_ops,
1973 .rsk_prot = &tcp6_request_sock_ops,
1974 .h.hashinfo = &tcp_hashinfo,
1975 .no_autobind = true,
1976 #ifdef CONFIG_COMPAT
1977 .compat_setsockopt = compat_tcp_setsockopt,
1978 .compat_getsockopt = compat_tcp_getsockopt,
1980 #ifdef CONFIG_MEMCG_KMEM
1981 .proto_cgroup = tcp_proto_cgroup,
1985 static const struct inet6_protocol tcpv6_protocol = {
1986 .early_demux = tcp_v6_early_demux,
1987 .handler = tcp_v6_rcv,
1988 .err_handler = tcp_v6_err,
1989 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1992 static struct inet_protosw tcpv6_protosw = {
1993 .type = SOCK_STREAM,
1994 .protocol = IPPROTO_TCP,
1995 .prot = &tcpv6_prot,
1996 .ops = &inet6_stream_ops,
1998 .flags = INET_PROTOSW_PERMANENT |
2002 static int __net_init tcpv6_net_init(struct net *net)
2004 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2005 SOCK_RAW, IPPROTO_TCP, net);
2008 static void __net_exit tcpv6_net_exit(struct net *net)
2010 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2013 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2015 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2018 static struct pernet_operations tcpv6_net_ops = {
2019 .init = tcpv6_net_init,
2020 .exit = tcpv6_net_exit,
2021 .exit_batch = tcpv6_net_exit_batch,
2024 int __init tcpv6_init(void)
2028 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2032 /* register inet6 protocol */
2033 ret = inet6_register_protosw(&tcpv6_protosw);
2035 goto out_tcpv6_protocol;
2037 ret = register_pernet_subsys(&tcpv6_net_ops);
2039 goto out_tcpv6_protosw;
2044 inet6_unregister_protosw(&tcpv6_protosw);
2046 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2050 void tcpv6_exit(void)
2052 unregister_pernet_subsys(&tcpv6_net_ops);
2053 inet6_unregister_protosw(&tcpv6_protosw);
2054 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);