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);
391 if (type == ICMPV6_PKT_TOOBIG) {
392 tp->mtu_info = ntohl(info);
393 if (!sock_owned_by_user(sk))
394 tcp_v6_mtu_reduced(sk);
395 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
401 icmpv6_err_convert(type, code, &err);
403 /* Might be for an request_sock */
404 switch (sk->sk_state) {
405 struct request_sock *req, **prev;
407 if (sock_owned_by_user(sk))
410 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
411 &hdr->saddr, inet6_iif(skb));
415 /* ICMPs are not backlogged, hence we cannot get
416 * an established socket here.
418 WARN_ON(req->sk != NULL);
420 if (seq != tcp_rsk(req)->snt_isn) {
421 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
425 inet_csk_reqsk_queue_drop(sk, req, prev);
426 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
430 case TCP_SYN_RECV: /* Cannot happen.
431 It can, it SYNs are crossed. --ANK */
432 if (!sock_owned_by_user(sk)) {
434 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
438 sk->sk_err_soft = err;
442 if (!sock_owned_by_user(sk) && np->recverr) {
444 sk->sk_error_report(sk);
446 sk->sk_err_soft = err;
454 static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
456 struct request_sock *req,
457 struct request_values *rvp,
460 struct inet6_request_sock *treq = inet6_rsk(req);
461 struct ipv6_pinfo *np = inet6_sk(sk);
462 struct sk_buff * skb;
465 /* First, grab a route. */
466 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
469 skb = tcp_make_synack(sk, dst, req, rvp, NULL);
472 __tcp_v6_send_check(skb, &treq->loc_addr, &treq->rmt_addr);
474 fl6->daddr = treq->rmt_addr;
475 skb_set_queue_mapping(skb, queue_mapping);
476 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
477 err = net_xmit_eval(err);
484 static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req,
485 struct request_values *rvp)
490 res = tcp_v6_send_synack(sk, NULL, &fl6, req, rvp, 0);
492 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
496 static void tcp_v6_reqsk_destructor(struct request_sock *req)
498 kfree_skb(inet6_rsk(req)->pktopts);
501 #ifdef CONFIG_TCP_MD5SIG
502 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
503 const struct in6_addr *addr)
505 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
508 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
509 struct sock *addr_sk)
511 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
514 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
515 struct request_sock *req)
517 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
520 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
523 struct tcp_md5sig cmd;
524 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
526 if (optlen < sizeof(cmd))
529 if (copy_from_user(&cmd, optval, sizeof(cmd)))
532 if (sin6->sin6_family != AF_INET6)
535 if (!cmd.tcpm_keylen) {
536 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
537 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
539 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
543 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
546 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
547 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
548 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
550 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
551 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
554 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
555 const struct in6_addr *daddr,
556 const struct in6_addr *saddr, int nbytes)
558 struct tcp6_pseudohdr *bp;
559 struct scatterlist sg;
561 bp = &hp->md5_blk.ip6;
562 /* 1. TCP pseudo-header (RFC2460) */
565 bp->protocol = cpu_to_be32(IPPROTO_TCP);
566 bp->len = cpu_to_be32(nbytes);
568 sg_init_one(&sg, bp, sizeof(*bp));
569 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
572 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
573 const struct in6_addr *daddr, struct in6_addr *saddr,
574 const struct tcphdr *th)
576 struct tcp_md5sig_pool *hp;
577 struct hash_desc *desc;
579 hp = tcp_get_md5sig_pool();
581 goto clear_hash_noput;
582 desc = &hp->md5_desc;
584 if (crypto_hash_init(desc))
586 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
588 if (tcp_md5_hash_header(hp, th))
590 if (tcp_md5_hash_key(hp, key))
592 if (crypto_hash_final(desc, md5_hash))
595 tcp_put_md5sig_pool();
599 tcp_put_md5sig_pool();
601 memset(md5_hash, 0, 16);
605 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
606 const struct sock *sk,
607 const struct request_sock *req,
608 const struct sk_buff *skb)
610 const struct in6_addr *saddr, *daddr;
611 struct tcp_md5sig_pool *hp;
612 struct hash_desc *desc;
613 const struct tcphdr *th = tcp_hdr(skb);
616 saddr = &inet6_sk(sk)->saddr;
617 daddr = &inet6_sk(sk)->daddr;
619 saddr = &inet6_rsk(req)->loc_addr;
620 daddr = &inet6_rsk(req)->rmt_addr;
622 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
623 saddr = &ip6h->saddr;
624 daddr = &ip6h->daddr;
627 hp = tcp_get_md5sig_pool();
629 goto clear_hash_noput;
630 desc = &hp->md5_desc;
632 if (crypto_hash_init(desc))
635 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
637 if (tcp_md5_hash_header(hp, th))
639 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
641 if (tcp_md5_hash_key(hp, key))
643 if (crypto_hash_final(desc, md5_hash))
646 tcp_put_md5sig_pool();
650 tcp_put_md5sig_pool();
652 memset(md5_hash, 0, 16);
656 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
658 const __u8 *hash_location = NULL;
659 struct tcp_md5sig_key *hash_expected;
660 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
661 const struct tcphdr *th = tcp_hdr(skb);
665 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
666 hash_location = tcp_parse_md5sig_option(th);
668 /* We've parsed the options - do we have a hash? */
669 if (!hash_expected && !hash_location)
672 if (hash_expected && !hash_location) {
673 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
677 if (!hash_expected && hash_location) {
678 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
682 /* check the signature */
683 genhash = tcp_v6_md5_hash_skb(newhash,
687 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
688 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
689 genhash ? "failed" : "mismatch",
690 &ip6h->saddr, ntohs(th->source),
691 &ip6h->daddr, ntohs(th->dest));
698 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
700 .obj_size = sizeof(struct tcp6_request_sock),
701 .rtx_syn_ack = tcp_v6_rtx_synack,
702 .send_ack = tcp_v6_reqsk_send_ack,
703 .destructor = tcp_v6_reqsk_destructor,
704 .send_reset = tcp_v6_send_reset,
705 .syn_ack_timeout = tcp_syn_ack_timeout,
708 #ifdef CONFIG_TCP_MD5SIG
709 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
710 .md5_lookup = tcp_v6_reqsk_md5_lookup,
711 .calc_md5_hash = tcp_v6_md5_hash_skb,
715 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
716 u32 ts, struct tcp_md5sig_key *key, int rst, u8 tclass)
718 const struct tcphdr *th = tcp_hdr(skb);
720 struct sk_buff *buff;
722 struct net *net = dev_net(skb_dst(skb)->dev);
723 struct sock *ctl_sk = net->ipv6.tcp_sk;
724 unsigned int tot_len = sizeof(struct tcphdr);
725 struct dst_entry *dst;
729 tot_len += TCPOLEN_TSTAMP_ALIGNED;
730 #ifdef CONFIG_TCP_MD5SIG
732 tot_len += TCPOLEN_MD5SIG_ALIGNED;
735 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
740 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
742 t1 = (struct tcphdr *) skb_push(buff, tot_len);
743 skb_reset_transport_header(buff);
745 /* Swap the send and the receive. */
746 memset(t1, 0, sizeof(*t1));
747 t1->dest = th->source;
748 t1->source = th->dest;
749 t1->doff = tot_len / 4;
750 t1->seq = htonl(seq);
751 t1->ack_seq = htonl(ack);
752 t1->ack = !rst || !th->ack;
754 t1->window = htons(win);
756 topt = (__be32 *)(t1 + 1);
759 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
760 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
761 *topt++ = htonl(tcp_time_stamp);
765 #ifdef CONFIG_TCP_MD5SIG
767 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
768 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
769 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
770 &ipv6_hdr(skb)->saddr,
771 &ipv6_hdr(skb)->daddr, t1);
775 memset(&fl6, 0, sizeof(fl6));
776 fl6.daddr = ipv6_hdr(skb)->saddr;
777 fl6.saddr = ipv6_hdr(skb)->daddr;
779 buff->ip_summed = CHECKSUM_PARTIAL;
782 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
784 fl6.flowi6_proto = IPPROTO_TCP;
785 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
786 fl6.flowi6_oif = inet6_iif(skb);
787 fl6.fl6_dport = t1->dest;
788 fl6.fl6_sport = t1->source;
789 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
791 /* Pass a socket to ip6_dst_lookup either it is for RST
792 * Underlying function will use this to retrieve the network
795 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false);
797 skb_dst_set(buff, dst);
798 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
799 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
801 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
808 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
810 const struct tcphdr *th = tcp_hdr(skb);
811 u32 seq = 0, ack_seq = 0;
812 struct tcp_md5sig_key *key = NULL;
813 #ifdef CONFIG_TCP_MD5SIG
814 const __u8 *hash_location = NULL;
815 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
816 unsigned char newhash[16];
818 struct sock *sk1 = NULL;
824 if (!ipv6_unicast_destination(skb))
827 #ifdef CONFIG_TCP_MD5SIG
828 hash_location = tcp_parse_md5sig_option(th);
829 if (!sk && hash_location) {
831 * active side is lost. Try to find listening socket through
832 * source port, and then find md5 key through listening socket.
833 * we are not loose security here:
834 * Incoming packet is checked with md5 hash with finding key,
835 * no RST generated if md5 hash doesn't match.
837 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
838 &tcp_hashinfo, &ipv6h->saddr,
839 th->source, &ipv6h->daddr,
840 ntohs(th->source), inet6_iif(skb));
845 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
849 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
850 if (genhash || memcmp(hash_location, newhash, 16) != 0)
853 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
858 seq = ntohl(th->ack_seq);
860 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
863 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1, 0);
865 #ifdef CONFIG_TCP_MD5SIG
874 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
875 struct tcp_md5sig_key *key, u8 tclass)
877 tcp_v6_send_response(skb, seq, ack, win, ts, key, 0, tclass);
880 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
882 struct inet_timewait_sock *tw = inet_twsk(sk);
883 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
885 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
886 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
887 tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw),
893 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
894 struct request_sock *req)
896 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
897 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr), 0);
901 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
903 struct request_sock *req, **prev;
904 const struct tcphdr *th = tcp_hdr(skb);
907 /* Find possible connection requests. */
908 req = inet6_csk_search_req(sk, &prev, th->source,
909 &ipv6_hdr(skb)->saddr,
910 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
912 return tcp_check_req(sk, skb, req, prev, false);
914 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
915 &ipv6_hdr(skb)->saddr, th->source,
916 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
919 if (nsk->sk_state != TCP_TIME_WAIT) {
923 inet_twsk_put(inet_twsk(nsk));
927 #ifdef CONFIG_SYN_COOKIES
929 sk = cookie_v6_check(sk, skb);
934 /* FIXME: this is substantially similar to the ipv4 code.
935 * Can some kind of merge be done? -- erics
937 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
939 struct tcp_extend_values tmp_ext;
940 struct tcp_options_received tmp_opt;
941 const u8 *hash_location;
942 struct request_sock *req;
943 struct inet6_request_sock *treq;
944 struct ipv6_pinfo *np = inet6_sk(sk);
945 struct tcp_sock *tp = tcp_sk(sk);
946 __u32 isn = TCP_SKB_CB(skb)->when;
947 struct dst_entry *dst = NULL;
949 bool want_cookie = false;
951 if (skb->protocol == htons(ETH_P_IP))
952 return tcp_v4_conn_request(sk, skb);
954 if (!ipv6_unicast_destination(skb))
957 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
958 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
963 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
964 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
968 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
972 #ifdef CONFIG_TCP_MD5SIG
973 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
976 tcp_clear_options(&tmp_opt);
977 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
978 tmp_opt.user_mss = tp->rx_opt.user_mss;
979 tcp_parse_options(skb, &tmp_opt, &hash_location, 0, NULL);
981 if (tmp_opt.cookie_plus > 0 &&
982 tmp_opt.saw_tstamp &&
983 !tp->rx_opt.cookie_out_never &&
984 (sysctl_tcp_cookie_size > 0 ||
985 (tp->cookie_values != NULL &&
986 tp->cookie_values->cookie_desired > 0))) {
989 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
990 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
992 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
995 /* Secret recipe starts with IP addresses */
996 d = (__force u32 *)&ipv6_hdr(skb)->daddr.s6_addr32[0];
1001 d = (__force u32 *)&ipv6_hdr(skb)->saddr.s6_addr32[0];
1007 /* plus variable length Initiator Cookie */
1010 *c++ ^= *hash_location++;
1012 want_cookie = false; /* not our kind of cookie */
1013 tmp_ext.cookie_out_never = 0; /* false */
1014 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1015 } else if (!tp->rx_opt.cookie_in_always) {
1016 /* redundant indications, but ensure initialization. */
1017 tmp_ext.cookie_out_never = 1; /* true */
1018 tmp_ext.cookie_plus = 0;
1022 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1024 if (want_cookie && !tmp_opt.saw_tstamp)
1025 tcp_clear_options(&tmp_opt);
1027 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1028 tcp_openreq_init(req, &tmp_opt, skb);
1030 treq = inet6_rsk(req);
1031 treq->rmt_addr = ipv6_hdr(skb)->saddr;
1032 treq->loc_addr = ipv6_hdr(skb)->daddr;
1033 if (!want_cookie || tmp_opt.tstamp_ok)
1034 TCP_ECN_create_request(req, skb, sock_net(sk));
1036 treq->iif = sk->sk_bound_dev_if;
1038 /* So that link locals have meaning */
1039 if (!sk->sk_bound_dev_if &&
1040 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1041 treq->iif = inet6_iif(skb);
1044 if (ipv6_opt_accepted(sk, skb) ||
1045 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1046 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1047 atomic_inc(&skb->users);
1048 treq->pktopts = skb;
1052 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1053 req->cookie_ts = tmp_opt.tstamp_ok;
1057 /* VJ's idea. We save last timestamp seen
1058 * from the destination in peer table, when entering
1059 * state TIME-WAIT, and check against it before
1060 * accepting new connection request.
1062 * If "isn" is not zero, this request hit alive
1063 * timewait bucket, so that all the necessary checks
1064 * are made in the function processing timewait state.
1066 if (tmp_opt.saw_tstamp &&
1067 tcp_death_row.sysctl_tw_recycle &&
1068 (dst = inet6_csk_route_req(sk, &fl6, req)) != NULL) {
1069 if (!tcp_peer_is_proven(req, dst, true)) {
1070 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1071 goto drop_and_release;
1074 /* Kill the following clause, if you dislike this way. */
1075 else if (!sysctl_tcp_syncookies &&
1076 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1077 (sysctl_max_syn_backlog >> 2)) &&
1078 !tcp_peer_is_proven(req, dst, false)) {
1079 /* Without syncookies last quarter of
1080 * backlog is filled with destinations,
1081 * proven to be alive.
1082 * It means that we continue to communicate
1083 * to destinations, already remembered
1084 * to the moment of synflood.
1086 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1087 &treq->rmt_addr, ntohs(tcp_hdr(skb)->source));
1088 goto drop_and_release;
1091 isn = tcp_v6_init_sequence(skb);
1094 tcp_rsk(req)->snt_isn = isn;
1096 if (security_inet_conn_request(sk, skb, req))
1097 goto drop_and_release;
1099 if (tcp_v6_send_synack(sk, dst, &fl6, req,
1100 (struct request_values *)&tmp_ext,
1101 skb_get_queue_mapping(skb)) ||
1105 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1106 tcp_rsk(req)->listener = NULL;
1107 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1115 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1116 return 0; /* don't send reset */
1119 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1120 struct request_sock *req,
1121 struct dst_entry *dst)
1123 struct inet6_request_sock *treq;
1124 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1125 struct tcp6_sock *newtcp6sk;
1126 struct inet_sock *newinet;
1127 struct tcp_sock *newtp;
1129 #ifdef CONFIG_TCP_MD5SIG
1130 struct tcp_md5sig_key *key;
1134 if (skb->protocol == htons(ETH_P_IP)) {
1139 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1144 newtcp6sk = (struct tcp6_sock *)newsk;
1145 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1147 newinet = inet_sk(newsk);
1148 newnp = inet6_sk(newsk);
1149 newtp = tcp_sk(newsk);
1151 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1153 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr);
1155 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1157 newnp->rcv_saddr = newnp->saddr;
1159 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1160 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1161 #ifdef CONFIG_TCP_MD5SIG
1162 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1165 newnp->ipv6_ac_list = NULL;
1166 newnp->ipv6_fl_list = NULL;
1167 newnp->pktoptions = NULL;
1169 newnp->mcast_oif = inet6_iif(skb);
1170 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1171 newnp->rcv_tclass = ipv6_get_dsfield(ipv6_hdr(skb));
1174 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1175 * here, tcp_create_openreq_child now does this for us, see the comment in
1176 * that function for the gory details. -acme
1179 /* It is tricky place. Until this moment IPv4 tcp
1180 worked with IPv6 icsk.icsk_af_ops.
1183 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1188 treq = inet6_rsk(req);
1190 if (sk_acceptq_is_full(sk))
1194 dst = inet6_csk_route_req(sk, &fl6, req);
1199 newsk = tcp_create_openreq_child(sk, req, skb);
1204 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1205 * count here, tcp_create_openreq_child now does this for us, see the
1206 * comment in that function for the gory details. -acme
1209 newsk->sk_gso_type = SKB_GSO_TCPV6;
1210 __ip6_dst_store(newsk, dst, NULL, NULL);
1211 inet6_sk_rx_dst_set(newsk, skb);
1213 newtcp6sk = (struct tcp6_sock *)newsk;
1214 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1216 newtp = tcp_sk(newsk);
1217 newinet = inet_sk(newsk);
1218 newnp = inet6_sk(newsk);
1220 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1222 newnp->daddr = treq->rmt_addr;
1223 newnp->saddr = treq->loc_addr;
1224 newnp->rcv_saddr = treq->loc_addr;
1225 newsk->sk_bound_dev_if = treq->iif;
1227 /* Now IPv6 options...
1229 First: no IPv4 options.
1231 newinet->inet_opt = NULL;
1232 newnp->ipv6_ac_list = NULL;
1233 newnp->ipv6_fl_list = NULL;
1236 newnp->rxopt.all = np->rxopt.all;
1238 /* Clone pktoptions received with SYN */
1239 newnp->pktoptions = NULL;
1240 if (treq->pktopts != NULL) {
1241 newnp->pktoptions = skb_clone(treq->pktopts,
1242 sk_gfp_atomic(sk, GFP_ATOMIC));
1243 consume_skb(treq->pktopts);
1244 treq->pktopts = NULL;
1245 if (newnp->pktoptions)
1246 skb_set_owner_r(newnp->pktoptions, newsk);
1249 newnp->mcast_oif = inet6_iif(skb);
1250 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1251 newnp->rcv_tclass = ipv6_get_dsfield(ipv6_hdr(skb));
1253 /* Clone native IPv6 options from listening socket (if any)
1255 Yes, keeping reference count would be much more clever,
1256 but we make one more one thing there: reattach optmem
1260 newnp->opt = ipv6_dup_options(newsk, np->opt);
1262 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1264 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1265 newnp->opt->opt_flen);
1267 tcp_mtup_init(newsk);
1268 tcp_sync_mss(newsk, dst_mtu(dst));
1269 newtp->advmss = dst_metric_advmss(dst);
1270 if (tcp_sk(sk)->rx_opt.user_mss &&
1271 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1272 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1274 tcp_initialize_rcv_mss(newsk);
1275 tcp_synack_rtt_meas(newsk, req);
1276 newtp->total_retrans = req->num_retrans;
1278 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1279 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1281 #ifdef CONFIG_TCP_MD5SIG
1282 /* Copy over the MD5 key from the original socket */
1283 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1284 /* We're using one, so create a matching key
1285 * on the newsk structure. If we fail to get
1286 * memory, then we end up not copying the key
1289 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newnp->daddr,
1290 AF_INET6, key->key, key->keylen,
1291 sk_gfp_atomic(sk, GFP_ATOMIC));
1295 if (__inet_inherit_port(sk, newsk) < 0) {
1296 inet_csk_prepare_forced_close(newsk);
1300 __inet6_hash(newsk, NULL);
1305 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1309 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1313 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1315 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1316 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1317 &ipv6_hdr(skb)->daddr, skb->csum)) {
1318 skb->ip_summed = CHECKSUM_UNNECESSARY;
1323 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1324 &ipv6_hdr(skb)->saddr,
1325 &ipv6_hdr(skb)->daddr, 0));
1327 if (skb->len <= 76) {
1328 return __skb_checksum_complete(skb);
1333 /* The socket must have it's spinlock held when we get
1336 * We have a potential double-lock case here, so even when
1337 * doing backlog processing we use the BH locking scheme.
1338 * This is because we cannot sleep with the original spinlock
1341 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1343 struct ipv6_pinfo *np = inet6_sk(sk);
1344 struct tcp_sock *tp;
1345 struct sk_buff *opt_skb = NULL;
1347 /* Imagine: socket is IPv6. IPv4 packet arrives,
1348 goes to IPv4 receive handler and backlogged.
1349 From backlog it always goes here. Kerboom...
1350 Fortunately, tcp_rcv_established and rcv_established
1351 handle them correctly, but it is not case with
1352 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1355 if (skb->protocol == htons(ETH_P_IP))
1356 return tcp_v4_do_rcv(sk, skb);
1358 #ifdef CONFIG_TCP_MD5SIG
1359 if (tcp_v6_inbound_md5_hash (sk, skb))
1363 if (sk_filter(sk, skb))
1367 * socket locking is here for SMP purposes as backlog rcv
1368 * is currently called with bh processing disabled.
1371 /* Do Stevens' IPV6_PKTOPTIONS.
1373 Yes, guys, it is the only place in our code, where we
1374 may make it not affecting IPv4.
1375 The rest of code is protocol independent,
1376 and I do not like idea to uglify IPv4.
1378 Actually, all the idea behind IPV6_PKTOPTIONS
1379 looks not very well thought. For now we latch
1380 options, received in the last packet, enqueued
1381 by tcp. Feel free to propose better solution.
1385 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1387 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1388 struct dst_entry *dst = sk->sk_rx_dst;
1390 sock_rps_save_rxhash(sk, skb);
1392 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1393 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1395 sk->sk_rx_dst = NULL;
1399 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1402 goto ipv6_pktoptions;
1406 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1409 if (sk->sk_state == TCP_LISTEN) {
1410 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1415 * Queue it on the new socket if the new socket is active,
1416 * otherwise we just shortcircuit this and continue with
1420 sock_rps_save_rxhash(nsk, skb);
1421 if (tcp_child_process(sk, nsk, skb))
1424 __kfree_skb(opt_skb);
1428 sock_rps_save_rxhash(sk, skb);
1430 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1433 goto ipv6_pktoptions;
1437 tcp_v6_send_reset(sk, skb);
1440 __kfree_skb(opt_skb);
1444 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1449 /* Do you ask, what is it?
1451 1. skb was enqueued by tcp.
1452 2. skb is added to tail of read queue, rather than out of order.
1453 3. socket is not in passive state.
1454 4. Finally, it really contains options, which user wants to receive.
1457 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1458 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1459 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1460 np->mcast_oif = inet6_iif(opt_skb);
1461 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1462 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1463 if (np->rxopt.bits.rxtclass)
1464 np->rcv_tclass = ipv6_get_dsfield(ipv6_hdr(skb));
1465 if (ipv6_opt_accepted(sk, opt_skb)) {
1466 skb_set_owner_r(opt_skb, sk);
1467 opt_skb = xchg(&np->pktoptions, opt_skb);
1469 __kfree_skb(opt_skb);
1470 opt_skb = xchg(&np->pktoptions, NULL);
1478 static int tcp_v6_rcv(struct sk_buff *skb)
1480 const struct tcphdr *th;
1481 const struct ipv6hdr *hdr;
1484 struct net *net = dev_net(skb->dev);
1486 if (skb->pkt_type != PACKET_HOST)
1490 * Count it even if it's bad.
1492 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1494 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1499 if (th->doff < sizeof(struct tcphdr)/4)
1501 if (!pskb_may_pull(skb, th->doff*4))
1504 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1508 hdr = ipv6_hdr(skb);
1509 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1510 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1511 skb->len - th->doff*4);
1512 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1513 TCP_SKB_CB(skb)->when = 0;
1514 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1515 TCP_SKB_CB(skb)->sacked = 0;
1517 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1522 if (sk->sk_state == TCP_TIME_WAIT)
1525 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1526 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1527 goto discard_and_relse;
1530 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1531 goto discard_and_relse;
1533 if (sk_filter(sk, skb))
1534 goto discard_and_relse;
1538 bh_lock_sock_nested(sk);
1540 if (!sock_owned_by_user(sk)) {
1541 #ifdef CONFIG_NET_DMA
1542 struct tcp_sock *tp = tcp_sk(sk);
1543 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1544 tp->ucopy.dma_chan = net_dma_find_channel();
1545 if (tp->ucopy.dma_chan)
1546 ret = tcp_v6_do_rcv(sk, skb);
1550 if (!tcp_prequeue(sk, skb))
1551 ret = tcp_v6_do_rcv(sk, skb);
1553 } else if (unlikely(sk_add_backlog(sk, skb,
1554 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1556 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1557 goto discard_and_relse;
1562 return ret ? -1 : 0;
1565 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1568 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1570 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1572 tcp_v6_send_reset(NULL, skb);
1589 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1590 inet_twsk_put(inet_twsk(sk));
1594 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1595 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1596 inet_twsk_put(inet_twsk(sk));
1600 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1605 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1606 &ipv6_hdr(skb)->saddr, th->source,
1607 &ipv6_hdr(skb)->daddr,
1608 ntohs(th->dest), inet6_iif(skb));
1610 struct inet_timewait_sock *tw = inet_twsk(sk);
1611 inet_twsk_deschedule(tw, &tcp_death_row);
1616 /* Fall through to ACK */
1619 tcp_v6_timewait_ack(sk, skb);
1623 case TCP_TW_SUCCESS:;
1628 static void tcp_v6_early_demux(struct sk_buff *skb)
1630 const struct ipv6hdr *hdr;
1631 const struct tcphdr *th;
1634 if (skb->pkt_type != PACKET_HOST)
1637 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1640 hdr = ipv6_hdr(skb);
1643 if (th->doff < sizeof(struct tcphdr) / 4)
1646 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1647 &hdr->saddr, th->source,
1648 &hdr->daddr, ntohs(th->dest),
1652 skb->destructor = sock_edemux;
1653 if (sk->sk_state != TCP_TIME_WAIT) {
1654 struct dst_entry *dst = sk->sk_rx_dst;
1657 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1659 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1660 skb_dst_set_noref(skb, dst);
1665 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1666 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1667 .twsk_unique = tcp_twsk_unique,
1668 .twsk_destructor= tcp_twsk_destructor,
1671 static const struct inet_connection_sock_af_ops ipv6_specific = {
1672 .queue_xmit = inet6_csk_xmit,
1673 .send_check = tcp_v6_send_check,
1674 .rebuild_header = inet6_sk_rebuild_header,
1675 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1676 .conn_request = tcp_v6_conn_request,
1677 .syn_recv_sock = tcp_v6_syn_recv_sock,
1678 .net_header_len = sizeof(struct ipv6hdr),
1679 .net_frag_header_len = sizeof(struct frag_hdr),
1680 .setsockopt = ipv6_setsockopt,
1681 .getsockopt = ipv6_getsockopt,
1682 .addr2sockaddr = inet6_csk_addr2sockaddr,
1683 .sockaddr_len = sizeof(struct sockaddr_in6),
1684 .bind_conflict = inet6_csk_bind_conflict,
1685 #ifdef CONFIG_COMPAT
1686 .compat_setsockopt = compat_ipv6_setsockopt,
1687 .compat_getsockopt = compat_ipv6_getsockopt,
1691 #ifdef CONFIG_TCP_MD5SIG
1692 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1693 .md5_lookup = tcp_v6_md5_lookup,
1694 .calc_md5_hash = tcp_v6_md5_hash_skb,
1695 .md5_parse = tcp_v6_parse_md5_keys,
1700 * TCP over IPv4 via INET6 API
1703 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1704 .queue_xmit = ip_queue_xmit,
1705 .send_check = tcp_v4_send_check,
1706 .rebuild_header = inet_sk_rebuild_header,
1707 .sk_rx_dst_set = inet_sk_rx_dst_set,
1708 .conn_request = tcp_v6_conn_request,
1709 .syn_recv_sock = tcp_v6_syn_recv_sock,
1710 .net_header_len = sizeof(struct iphdr),
1711 .setsockopt = ipv6_setsockopt,
1712 .getsockopt = ipv6_getsockopt,
1713 .addr2sockaddr = inet6_csk_addr2sockaddr,
1714 .sockaddr_len = sizeof(struct sockaddr_in6),
1715 .bind_conflict = inet6_csk_bind_conflict,
1716 #ifdef CONFIG_COMPAT
1717 .compat_setsockopt = compat_ipv6_setsockopt,
1718 .compat_getsockopt = compat_ipv6_getsockopt,
1722 #ifdef CONFIG_TCP_MD5SIG
1723 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1724 .md5_lookup = tcp_v4_md5_lookup,
1725 .calc_md5_hash = tcp_v4_md5_hash_skb,
1726 .md5_parse = tcp_v6_parse_md5_keys,
1730 /* NOTE: A lot of things set to zero explicitly by call to
1731 * sk_alloc() so need not be done here.
1733 static int tcp_v6_init_sock(struct sock *sk)
1735 struct inet_connection_sock *icsk = inet_csk(sk);
1739 icsk->icsk_af_ops = &ipv6_specific;
1741 #ifdef CONFIG_TCP_MD5SIG
1742 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1748 static void tcp_v6_destroy_sock(struct sock *sk)
1750 tcp_v4_destroy_sock(sk);
1751 inet6_destroy_sock(sk);
1754 #ifdef CONFIG_PROC_FS
1755 /* Proc filesystem TCPv6 sock list dumping. */
1756 static void get_openreq6(struct seq_file *seq,
1757 const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1759 int ttd = req->expires - jiffies;
1760 const struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1761 const struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1767 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1768 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1770 src->s6_addr32[0], src->s6_addr32[1],
1771 src->s6_addr32[2], src->s6_addr32[3],
1772 ntohs(inet_rsk(req)->loc_port),
1773 dest->s6_addr32[0], dest->s6_addr32[1],
1774 dest->s6_addr32[2], dest->s6_addr32[3],
1775 ntohs(inet_rsk(req)->rmt_port),
1777 0,0, /* could print option size, but that is af dependent. */
1778 1, /* timers active (only the expire timer) */
1779 jiffies_to_clock_t(ttd),
1781 from_kuid_munged(seq_user_ns(seq), uid),
1782 0, /* non standard timer */
1783 0, /* open_requests have no inode */
1787 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1789 const struct in6_addr *dest, *src;
1792 unsigned long timer_expires;
1793 const struct inet_sock *inet = inet_sk(sp);
1794 const struct tcp_sock *tp = tcp_sk(sp);
1795 const struct inet_connection_sock *icsk = inet_csk(sp);
1796 const struct ipv6_pinfo *np = inet6_sk(sp);
1799 src = &np->rcv_saddr;
1800 destp = ntohs(inet->inet_dport);
1801 srcp = ntohs(inet->inet_sport);
1803 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1805 timer_expires = icsk->icsk_timeout;
1806 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1808 timer_expires = icsk->icsk_timeout;
1809 } else if (timer_pending(&sp->sk_timer)) {
1811 timer_expires = sp->sk_timer.expires;
1814 timer_expires = jiffies;
1818 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1819 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %lu %lu %u %u %d\n",
1821 src->s6_addr32[0], src->s6_addr32[1],
1822 src->s6_addr32[2], src->s6_addr32[3], srcp,
1823 dest->s6_addr32[0], dest->s6_addr32[1],
1824 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1826 tp->write_seq-tp->snd_una,
1827 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1829 jiffies_delta_to_clock_t(timer_expires - jiffies),
1830 icsk->icsk_retransmits,
1831 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1832 icsk->icsk_probes_out,
1834 atomic_read(&sp->sk_refcnt), sp,
1835 jiffies_to_clock_t(icsk->icsk_rto),
1836 jiffies_to_clock_t(icsk->icsk_ack.ato),
1837 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1839 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
1843 static void get_timewait6_sock(struct seq_file *seq,
1844 struct inet_timewait_sock *tw, int i)
1846 const struct in6_addr *dest, *src;
1848 const struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
1849 long delta = tw->tw_ttd - jiffies;
1851 dest = &tw6->tw_v6_daddr;
1852 src = &tw6->tw_v6_rcv_saddr;
1853 destp = ntohs(tw->tw_dport);
1854 srcp = ntohs(tw->tw_sport);
1857 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1858 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1860 src->s6_addr32[0], src->s6_addr32[1],
1861 src->s6_addr32[2], src->s6_addr32[3], srcp,
1862 dest->s6_addr32[0], dest->s6_addr32[1],
1863 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1864 tw->tw_substate, 0, 0,
1865 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1866 atomic_read(&tw->tw_refcnt), tw);
1869 static int tcp6_seq_show(struct seq_file *seq, void *v)
1871 struct tcp_iter_state *st;
1873 if (v == SEQ_START_TOKEN) {
1878 "st tx_queue rx_queue tr tm->when retrnsmt"
1879 " uid timeout inode\n");
1884 switch (st->state) {
1885 case TCP_SEQ_STATE_LISTENING:
1886 case TCP_SEQ_STATE_ESTABLISHED:
1887 get_tcp6_sock(seq, v, st->num);
1889 case TCP_SEQ_STATE_OPENREQ:
1890 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1892 case TCP_SEQ_STATE_TIME_WAIT:
1893 get_timewait6_sock(seq, v, st->num);
1900 static const struct file_operations tcp6_afinfo_seq_fops = {
1901 .owner = THIS_MODULE,
1902 .open = tcp_seq_open,
1904 .llseek = seq_lseek,
1905 .release = seq_release_net
1908 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1911 .seq_fops = &tcp6_afinfo_seq_fops,
1913 .show = tcp6_seq_show,
1917 int __net_init tcp6_proc_init(struct net *net)
1919 return tcp_proc_register(net, &tcp6_seq_afinfo);
1922 void tcp6_proc_exit(struct net *net)
1924 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1928 struct proto tcpv6_prot = {
1930 .owner = THIS_MODULE,
1932 .connect = tcp_v6_connect,
1933 .disconnect = tcp_disconnect,
1934 .accept = inet_csk_accept,
1936 .init = tcp_v6_init_sock,
1937 .destroy = tcp_v6_destroy_sock,
1938 .shutdown = tcp_shutdown,
1939 .setsockopt = tcp_setsockopt,
1940 .getsockopt = tcp_getsockopt,
1941 .recvmsg = tcp_recvmsg,
1942 .sendmsg = tcp_sendmsg,
1943 .sendpage = tcp_sendpage,
1944 .backlog_rcv = tcp_v6_do_rcv,
1945 .release_cb = tcp_release_cb,
1946 .mtu_reduced = tcp_v6_mtu_reduced,
1947 .hash = tcp_v6_hash,
1948 .unhash = inet_unhash,
1949 .get_port = inet_csk_get_port,
1950 .enter_memory_pressure = tcp_enter_memory_pressure,
1951 .sockets_allocated = &tcp_sockets_allocated,
1952 .memory_allocated = &tcp_memory_allocated,
1953 .memory_pressure = &tcp_memory_pressure,
1954 .orphan_count = &tcp_orphan_count,
1955 .sysctl_wmem = sysctl_tcp_wmem,
1956 .sysctl_rmem = sysctl_tcp_rmem,
1957 .max_header = MAX_TCP_HEADER,
1958 .obj_size = sizeof(struct tcp6_sock),
1959 .slab_flags = SLAB_DESTROY_BY_RCU,
1960 .twsk_prot = &tcp6_timewait_sock_ops,
1961 .rsk_prot = &tcp6_request_sock_ops,
1962 .h.hashinfo = &tcp_hashinfo,
1963 .no_autobind = true,
1964 #ifdef CONFIG_COMPAT
1965 .compat_setsockopt = compat_tcp_setsockopt,
1966 .compat_getsockopt = compat_tcp_getsockopt,
1968 #ifdef CONFIG_MEMCG_KMEM
1969 .proto_cgroup = tcp_proto_cgroup,
1973 static const struct inet6_protocol tcpv6_protocol = {
1974 .early_demux = tcp_v6_early_demux,
1975 .handler = tcp_v6_rcv,
1976 .err_handler = tcp_v6_err,
1977 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1980 static struct inet_protosw tcpv6_protosw = {
1981 .type = SOCK_STREAM,
1982 .protocol = IPPROTO_TCP,
1983 .prot = &tcpv6_prot,
1984 .ops = &inet6_stream_ops,
1986 .flags = INET_PROTOSW_PERMANENT |
1990 static int __net_init tcpv6_net_init(struct net *net)
1992 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1993 SOCK_RAW, IPPROTO_TCP, net);
1996 static void __net_exit tcpv6_net_exit(struct net *net)
1998 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2001 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2003 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2006 static struct pernet_operations tcpv6_net_ops = {
2007 .init = tcpv6_net_init,
2008 .exit = tcpv6_net_exit,
2009 .exit_batch = tcpv6_net_exit_batch,
2012 int __init tcpv6_init(void)
2016 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2020 /* register inet6 protocol */
2021 ret = inet6_register_protosw(&tcpv6_protosw);
2023 goto out_tcpv6_protocol;
2025 ret = register_pernet_subsys(&tcpv6_net_ops);
2027 goto out_tcpv6_protosw;
2032 inet6_unregister_protosw(&tcpv6_protosw);
2034 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2038 void tcpv6_exit(void)
2040 unregister_pernet_subsys(&tcpv6_net_ops);
2041 inet6_unregister_protosw(&tcpv6_protosw);
2042 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);