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>
42 #include <linux/uaccess.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/inet_common.h>
63 #include <net/secure_seq.h>
64 #include <net/busy_poll.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
69 #include <linux/crypto.h>
70 #include <linux/scatterlist.h>
72 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
73 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
74 struct request_sock *req);
76 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
78 static const struct inet_connection_sock_af_ops ipv6_mapped;
79 static const struct inet_connection_sock_af_ops ipv6_specific;
80 #ifdef CONFIG_TCP_MD5SIG
81 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
82 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
84 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
85 const struct in6_addr *addr)
91 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
93 struct dst_entry *dst = skb_dst(skb);
95 if (dst && dst_hold_safe(dst)) {
96 const struct rt6_info *rt = (const struct rt6_info *)dst;
99 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
100 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
104 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
106 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
107 ipv6_hdr(skb)->saddr.s6_addr32,
109 tcp_hdr(skb)->source);
112 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
115 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
116 struct inet_sock *inet = inet_sk(sk);
117 struct inet_connection_sock *icsk = inet_csk(sk);
118 struct ipv6_pinfo *np = inet6_sk(sk);
119 struct tcp_sock *tp = tcp_sk(sk);
120 struct in6_addr *saddr = NULL, *final_p, final;
121 struct ipv6_txoptions *opt;
123 struct dst_entry *dst;
127 if (addr_len < SIN6_LEN_RFC2133)
130 if (usin->sin6_family != AF_INET6)
131 return -EAFNOSUPPORT;
133 memset(&fl6, 0, sizeof(fl6));
136 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
137 IP6_ECN_flow_init(fl6.flowlabel);
138 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
139 struct ip6_flowlabel *flowlabel;
140 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
143 fl6_sock_release(flowlabel);
148 * connect() to INADDR_ANY means loopback (BSD'ism).
151 if (ipv6_addr_any(&usin->sin6_addr))
152 usin->sin6_addr.s6_addr[15] = 0x1;
154 addr_type = ipv6_addr_type(&usin->sin6_addr);
156 if (addr_type & IPV6_ADDR_MULTICAST)
159 if (addr_type&IPV6_ADDR_LINKLOCAL) {
160 if (addr_len >= sizeof(struct sockaddr_in6) &&
161 usin->sin6_scope_id) {
162 /* If interface is set while binding, indices
165 if (sk->sk_bound_dev_if &&
166 sk->sk_bound_dev_if != usin->sin6_scope_id)
169 sk->sk_bound_dev_if = usin->sin6_scope_id;
172 /* Connect to link-local address requires an interface */
173 if (!sk->sk_bound_dev_if)
177 if (tp->rx_opt.ts_recent_stamp &&
178 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
179 tp->rx_opt.ts_recent = 0;
180 tp->rx_opt.ts_recent_stamp = 0;
184 sk->sk_v6_daddr = usin->sin6_addr;
185 np->flow_label = fl6.flowlabel;
191 if (addr_type == IPV6_ADDR_MAPPED) {
192 u32 exthdrlen = icsk->icsk_ext_hdr_len;
193 struct sockaddr_in sin;
195 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
197 if (__ipv6_only_sock(sk))
200 sin.sin_family = AF_INET;
201 sin.sin_port = usin->sin6_port;
202 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
204 icsk->icsk_af_ops = &ipv6_mapped;
205 sk->sk_backlog_rcv = tcp_v4_do_rcv;
206 #ifdef CONFIG_TCP_MD5SIG
207 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
210 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
213 icsk->icsk_ext_hdr_len = exthdrlen;
214 icsk->icsk_af_ops = &ipv6_specific;
215 sk->sk_backlog_rcv = tcp_v6_do_rcv;
216 #ifdef CONFIG_TCP_MD5SIG
217 tp->af_specific = &tcp_sock_ipv6_specific;
221 np->saddr = sk->sk_v6_rcv_saddr;
226 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
227 saddr = &sk->sk_v6_rcv_saddr;
229 fl6.flowi6_proto = IPPROTO_TCP;
230 fl6.daddr = sk->sk_v6_daddr;
231 fl6.saddr = saddr ? *saddr : np->saddr;
232 fl6.flowi6_oif = sk->sk_bound_dev_if;
233 fl6.flowi6_mark = sk->sk_mark;
234 fl6.fl6_dport = usin->sin6_port;
235 fl6.fl6_sport = inet->inet_sport;
237 opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
238 final_p = fl6_update_dst(&fl6, opt, &final);
240 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
242 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
250 sk->sk_v6_rcv_saddr = *saddr;
253 /* set the source address */
255 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
257 sk->sk_gso_type = SKB_GSO_TCPV6;
258 ip6_dst_store(sk, dst, NULL, NULL);
260 if (tcp_death_row.sysctl_tw_recycle &&
261 !tp->rx_opt.ts_recent_stamp &&
262 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
263 tcp_fetch_timewait_stamp(sk, dst);
265 icsk->icsk_ext_hdr_len = 0;
267 icsk->icsk_ext_hdr_len = opt->opt_flen +
270 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
272 inet->inet_dport = usin->sin6_port;
274 tcp_set_state(sk, TCP_SYN_SENT);
275 err = inet6_hash_connect(&tcp_death_row, sk);
281 if (!tp->write_seq && likely(!tp->repair))
282 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
283 sk->sk_v6_daddr.s6_addr32,
287 err = tcp_connect(sk);
294 tcp_set_state(sk, TCP_CLOSE);
297 inet->inet_dport = 0;
298 sk->sk_route_caps = 0;
302 static void tcp_v6_mtu_reduced(struct sock *sk)
304 struct dst_entry *dst;
306 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
309 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
313 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
314 tcp_sync_mss(sk, dst_mtu(dst));
315 tcp_simple_retransmit(sk);
319 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
320 u8 type, u8 code, int offset, __be32 info)
322 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
323 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
324 struct net *net = dev_net(skb->dev);
325 struct request_sock *fastopen;
326 struct ipv6_pinfo *np;
332 sk = __inet6_lookup_established(net, &tcp_hashinfo,
333 &hdr->daddr, th->dest,
334 &hdr->saddr, ntohs(th->source),
338 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
343 if (sk->sk_state == TCP_TIME_WAIT) {
344 inet_twsk_put(inet_twsk(sk));
347 seq = ntohl(th->seq);
348 if (sk->sk_state == TCP_NEW_SYN_RECV)
349 return tcp_req_err(sk, seq);
352 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
353 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
355 if (sk->sk_state == TCP_CLOSE)
358 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
359 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
364 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
365 fastopen = tp->fastopen_rsk;
366 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
367 if (sk->sk_state != TCP_LISTEN &&
368 !between(seq, snd_una, tp->snd_nxt)) {
369 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
375 if (type == NDISC_REDIRECT) {
376 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
379 dst->ops->redirect(dst, sk, skb);
383 if (type == ICMPV6_PKT_TOOBIG) {
384 /* We are not interested in TCP_LISTEN and open_requests
385 * (SYN-ACKs send out by Linux are always <576bytes so
386 * they should go through unfragmented).
388 if (sk->sk_state == TCP_LISTEN)
391 if (!ip6_sk_accept_pmtu(sk))
394 tp->mtu_info = ntohl(info);
395 if (!sock_owned_by_user(sk))
396 tcp_v6_mtu_reduced(sk);
397 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
403 icmpv6_err_convert(type, code, &err);
405 /* Might be for an request_sock */
406 switch (sk->sk_state) {
409 /* Only in fast or simultaneous open. If a fast open socket is
410 * is already accepted it is treated as a connected one below.
412 if (fastopen && !fastopen->sk)
415 if (!sock_owned_by_user(sk)) {
417 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
421 sk->sk_err_soft = err;
425 if (!sock_owned_by_user(sk) && np->recverr) {
427 sk->sk_error_report(sk);
429 sk->sk_err_soft = err;
437 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
439 struct request_sock *req,
440 struct tcp_fastopen_cookie *foc,
443 struct inet_request_sock *ireq = inet_rsk(req);
444 struct ipv6_pinfo *np = inet6_sk(sk);
445 struct flowi6 *fl6 = &fl->u.ip6;
449 /* First, grab a route. */
450 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
451 IPPROTO_TCP)) == NULL)
454 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
457 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
458 &ireq->ir_v6_rmt_addr);
460 fl6->daddr = ireq->ir_v6_rmt_addr;
461 if (np->repflow && ireq->pktopts)
462 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
465 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
468 err = net_xmit_eval(err);
476 static void tcp_v6_reqsk_destructor(struct request_sock *req)
478 kfree_skb(inet_rsk(req)->pktopts);
481 #ifdef CONFIG_TCP_MD5SIG
482 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
483 const struct in6_addr *addr)
485 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
488 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
489 const struct sock *addr_sk)
491 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
494 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
497 struct tcp_md5sig cmd;
498 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
500 if (optlen < sizeof(cmd))
503 if (copy_from_user(&cmd, optval, sizeof(cmd)))
506 if (sin6->sin6_family != AF_INET6)
509 if (!cmd.tcpm_keylen) {
510 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
511 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
513 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
517 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
520 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
521 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
522 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
524 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
525 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
528 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
529 const struct in6_addr *daddr,
530 const struct in6_addr *saddr, int nbytes)
532 struct tcp6_pseudohdr *bp;
533 struct scatterlist sg;
535 bp = &hp->md5_blk.ip6;
536 /* 1. TCP pseudo-header (RFC2460) */
539 bp->protocol = cpu_to_be32(IPPROTO_TCP);
540 bp->len = cpu_to_be32(nbytes);
542 sg_init_one(&sg, bp, sizeof(*bp));
543 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
546 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
547 const struct in6_addr *daddr, struct in6_addr *saddr,
548 const struct tcphdr *th)
550 struct tcp_md5sig_pool *hp;
551 struct hash_desc *desc;
553 hp = tcp_get_md5sig_pool();
555 goto clear_hash_noput;
556 desc = &hp->md5_desc;
558 if (crypto_hash_init(desc))
560 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
562 if (tcp_md5_hash_header(hp, th))
564 if (tcp_md5_hash_key(hp, key))
566 if (crypto_hash_final(desc, md5_hash))
569 tcp_put_md5sig_pool();
573 tcp_put_md5sig_pool();
575 memset(md5_hash, 0, 16);
579 static int tcp_v6_md5_hash_skb(char *md5_hash,
580 const struct tcp_md5sig_key *key,
581 const struct sock *sk,
582 const struct sk_buff *skb)
584 const struct in6_addr *saddr, *daddr;
585 struct tcp_md5sig_pool *hp;
586 struct hash_desc *desc;
587 const struct tcphdr *th = tcp_hdr(skb);
589 if (sk) { /* valid for establish/request sockets */
590 saddr = &sk->sk_v6_rcv_saddr;
591 daddr = &sk->sk_v6_daddr;
593 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
594 saddr = &ip6h->saddr;
595 daddr = &ip6h->daddr;
598 hp = tcp_get_md5sig_pool();
600 goto clear_hash_noput;
601 desc = &hp->md5_desc;
603 if (crypto_hash_init(desc))
606 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
608 if (tcp_md5_hash_header(hp, th))
610 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
612 if (tcp_md5_hash_key(hp, key))
614 if (crypto_hash_final(desc, md5_hash))
617 tcp_put_md5sig_pool();
621 tcp_put_md5sig_pool();
623 memset(md5_hash, 0, 16);
629 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
630 const struct sk_buff *skb)
632 #ifdef CONFIG_TCP_MD5SIG
633 const __u8 *hash_location = NULL;
634 struct tcp_md5sig_key *hash_expected;
635 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
636 const struct tcphdr *th = tcp_hdr(skb);
640 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
641 hash_location = tcp_parse_md5sig_option(th);
643 /* We've parsed the options - do we have a hash? */
644 if (!hash_expected && !hash_location)
647 if (hash_expected && !hash_location) {
648 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
652 if (!hash_expected && hash_location) {
653 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
657 /* check the signature */
658 genhash = tcp_v6_md5_hash_skb(newhash,
662 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
663 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
664 genhash ? "failed" : "mismatch",
665 &ip6h->saddr, ntohs(th->source),
666 &ip6h->daddr, ntohs(th->dest));
673 static void tcp_v6_init_req(struct request_sock *req,
674 const struct sock *sk_listener,
677 struct inet_request_sock *ireq = inet_rsk(req);
678 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
680 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
681 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
683 /* So that link locals have meaning */
684 if (!sk_listener->sk_bound_dev_if &&
685 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
686 ireq->ir_iif = tcp_v6_iif(skb);
688 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
689 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
690 np->rxopt.bits.rxinfo ||
691 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
692 np->rxopt.bits.rxohlim || np->repflow)) {
693 atomic_inc(&skb->users);
698 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
700 const struct request_sock *req,
705 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
708 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
710 .obj_size = sizeof(struct tcp6_request_sock),
711 .rtx_syn_ack = tcp_rtx_synack,
712 .send_ack = tcp_v6_reqsk_send_ack,
713 .destructor = tcp_v6_reqsk_destructor,
714 .send_reset = tcp_v6_send_reset,
715 .syn_ack_timeout = tcp_syn_ack_timeout,
718 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
719 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
720 sizeof(struct ipv6hdr),
721 #ifdef CONFIG_TCP_MD5SIG
722 .req_md5_lookup = tcp_v6_md5_lookup,
723 .calc_md5_hash = tcp_v6_md5_hash_skb,
725 .init_req = tcp_v6_init_req,
726 #ifdef CONFIG_SYN_COOKIES
727 .cookie_init_seq = cookie_v6_init_sequence,
729 .route_req = tcp_v6_route_req,
730 .init_seq = tcp_v6_init_sequence,
731 .send_synack = tcp_v6_send_synack,
734 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
735 u32 ack, u32 win, u32 tsval, u32 tsecr,
736 int oif, struct tcp_md5sig_key *key, int rst,
737 u8 tclass, u32 label)
739 const struct tcphdr *th = tcp_hdr(skb);
741 struct sk_buff *buff;
743 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
744 struct sock *ctl_sk = net->ipv6.tcp_sk;
745 unsigned int tot_len = sizeof(struct tcphdr);
746 struct dst_entry *dst;
750 tot_len += TCPOLEN_TSTAMP_ALIGNED;
751 #ifdef CONFIG_TCP_MD5SIG
753 tot_len += TCPOLEN_MD5SIG_ALIGNED;
756 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
761 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
763 t1 = (struct tcphdr *) skb_push(buff, tot_len);
764 skb_reset_transport_header(buff);
766 /* Swap the send and the receive. */
767 memset(t1, 0, sizeof(*t1));
768 t1->dest = th->source;
769 t1->source = th->dest;
770 t1->doff = tot_len / 4;
771 t1->seq = htonl(seq);
772 t1->ack_seq = htonl(ack);
773 t1->ack = !rst || !th->ack;
775 t1->window = htons(win);
777 topt = (__be32 *)(t1 + 1);
780 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
781 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
782 *topt++ = htonl(tsval);
783 *topt++ = htonl(tsecr);
786 #ifdef CONFIG_TCP_MD5SIG
788 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
789 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
790 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
791 &ipv6_hdr(skb)->saddr,
792 &ipv6_hdr(skb)->daddr, t1);
796 memset(&fl6, 0, sizeof(fl6));
797 fl6.daddr = ipv6_hdr(skb)->saddr;
798 fl6.saddr = ipv6_hdr(skb)->daddr;
799 fl6.flowlabel = label;
801 buff->ip_summed = CHECKSUM_PARTIAL;
804 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
806 fl6.flowi6_proto = IPPROTO_TCP;
807 if (rt6_need_strict(&fl6.daddr) && !oif)
808 fl6.flowi6_oif = tcp_v6_iif(skb);
810 fl6.flowi6_oif = oif;
811 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
812 fl6.fl6_dport = t1->dest;
813 fl6.fl6_sport = t1->source;
814 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
816 /* Pass a socket to ip6_dst_lookup either it is for RST
817 * Underlying function will use this to retrieve the network
820 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
822 skb_dst_set(buff, dst);
823 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
824 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
826 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
833 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
835 const struct tcphdr *th = tcp_hdr(skb);
836 u32 seq = 0, ack_seq = 0;
837 struct tcp_md5sig_key *key = NULL;
838 #ifdef CONFIG_TCP_MD5SIG
839 const __u8 *hash_location = NULL;
840 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
841 unsigned char newhash[16];
843 struct sock *sk1 = NULL;
850 /* If sk not NULL, it means we did a successful lookup and incoming
851 * route had to be correct. prequeue might have dropped our dst.
853 if (!sk && !ipv6_unicast_destination(skb))
856 #ifdef CONFIG_TCP_MD5SIG
857 hash_location = tcp_parse_md5sig_option(th);
858 if (sk && sk_fullsock(sk)) {
859 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
860 } else if (hash_location) {
862 * active side is lost. Try to find listening socket through
863 * source port, and then find md5 key through listening socket.
864 * we are not loose security here:
865 * Incoming packet is checked with md5 hash with finding key,
866 * no RST generated if md5 hash doesn't match.
868 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
869 &tcp_hashinfo, NULL, 0,
871 th->source, &ipv6h->daddr,
872 ntohs(th->source), tcp_v6_iif(skb));
877 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
881 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
882 if (genhash || memcmp(hash_location, newhash, 16) != 0)
888 seq = ntohl(th->ack_seq);
890 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
893 oif = sk ? sk->sk_bound_dev_if : 0;
894 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
896 #ifdef CONFIG_TCP_MD5SIG
905 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
906 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
907 struct tcp_md5sig_key *key, u8 tclass,
910 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
914 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
916 struct inet_timewait_sock *tw = inet_twsk(sk);
917 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
919 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
920 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
921 tcp_time_stamp + tcptw->tw_ts_offset,
922 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
923 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
928 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
929 struct request_sock *req)
931 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
932 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
934 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
935 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
936 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
937 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
938 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
943 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
945 #ifdef CONFIG_SYN_COOKIES
946 const struct tcphdr *th = tcp_hdr(skb);
949 sk = cookie_v6_check(sk, skb);
954 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
956 if (skb->protocol == htons(ETH_P_IP))
957 return tcp_v4_conn_request(sk, skb);
959 if (!ipv6_unicast_destination(skb))
962 return tcp_conn_request(&tcp6_request_sock_ops,
963 &tcp_request_sock_ipv6_ops, sk, skb);
966 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
967 return 0; /* don't send reset */
970 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
971 struct request_sock *req,
972 struct dst_entry *dst,
973 struct request_sock *req_unhash,
976 struct inet_request_sock *ireq;
977 struct ipv6_pinfo *newnp;
978 const struct ipv6_pinfo *np = inet6_sk(sk);
979 struct ipv6_txoptions *opt;
980 struct tcp6_sock *newtcp6sk;
981 struct inet_sock *newinet;
982 struct tcp_sock *newtp;
984 #ifdef CONFIG_TCP_MD5SIG
985 struct tcp_md5sig_key *key;
989 if (skb->protocol == htons(ETH_P_IP)) {
994 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
995 req_unhash, own_req);
1000 newtcp6sk = (struct tcp6_sock *)newsk;
1001 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1003 newinet = inet_sk(newsk);
1004 newnp = inet6_sk(newsk);
1005 newtp = tcp_sk(newsk);
1007 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1009 newnp->saddr = newsk->sk_v6_rcv_saddr;
1011 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1012 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1013 #ifdef CONFIG_TCP_MD5SIG
1014 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1017 newnp->ipv6_ac_list = NULL;
1018 newnp->ipv6_fl_list = NULL;
1019 newnp->pktoptions = NULL;
1021 newnp->mcast_oif = tcp_v6_iif(skb);
1022 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1023 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1025 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1028 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1029 * here, tcp_create_openreq_child now does this for us, see the comment in
1030 * that function for the gory details. -acme
1033 /* It is tricky place. Until this moment IPv4 tcp
1034 worked with IPv6 icsk.icsk_af_ops.
1037 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1042 ireq = inet_rsk(req);
1044 if (sk_acceptq_is_full(sk))
1048 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1053 newsk = tcp_create_openreq_child(sk, req, skb);
1058 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1059 * count here, tcp_create_openreq_child now does this for us, see the
1060 * comment in that function for the gory details. -acme
1063 newsk->sk_gso_type = SKB_GSO_TCPV6;
1064 ip6_dst_store(newsk, dst, NULL, NULL);
1065 inet6_sk_rx_dst_set(newsk, skb);
1067 newtcp6sk = (struct tcp6_sock *)newsk;
1068 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1070 newtp = tcp_sk(newsk);
1071 newinet = inet_sk(newsk);
1072 newnp = inet6_sk(newsk);
1074 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1076 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1077 newnp->saddr = ireq->ir_v6_loc_addr;
1078 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1079 newsk->sk_bound_dev_if = ireq->ir_iif;
1081 /* Now IPv6 options...
1083 First: no IPv4 options.
1085 newinet->inet_opt = NULL;
1086 newnp->ipv6_ac_list = NULL;
1087 newnp->ipv6_fl_list = NULL;
1090 newnp->rxopt.all = np->rxopt.all;
1092 newnp->pktoptions = NULL;
1094 newnp->mcast_oif = tcp_v6_iif(skb);
1095 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1096 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1098 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1100 /* Clone native IPv6 options from listening socket (if any)
1102 Yes, keeping reference count would be much more clever,
1103 but we make one more one thing there: reattach optmem
1106 opt = rcu_dereference(np->opt);
1108 opt = ipv6_dup_options(newsk, opt);
1109 RCU_INIT_POINTER(newnp->opt, opt);
1111 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1113 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1116 tcp_ca_openreq_child(newsk, dst);
1118 tcp_sync_mss(newsk, dst_mtu(dst));
1119 newtp->advmss = dst_metric_advmss(dst);
1120 if (tcp_sk(sk)->rx_opt.user_mss &&
1121 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1122 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1124 tcp_initialize_rcv_mss(newsk);
1126 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1127 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1129 #ifdef CONFIG_TCP_MD5SIG
1130 /* Copy over the MD5 key from the original socket */
1131 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1133 /* We're using one, so create a matching key
1134 * on the newsk structure. If we fail to get
1135 * memory, then we end up not copying the key
1138 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1139 AF_INET6, key->key, key->keylen,
1140 sk_gfp_mask(sk, GFP_ATOMIC));
1144 if (__inet_inherit_port(sk, newsk) < 0) {
1145 inet_csk_prepare_forced_close(newsk);
1149 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1151 tcp_move_syn(newtp, req);
1153 /* Clone pktoptions received with SYN, if we own the req */
1154 if (ireq->pktopts) {
1155 newnp->pktoptions = skb_clone(ireq->pktopts,
1156 sk_gfp_mask(sk, GFP_ATOMIC));
1157 consume_skb(ireq->pktopts);
1158 ireq->pktopts = NULL;
1159 if (newnp->pktoptions)
1160 skb_set_owner_r(newnp->pktoptions, newsk);
1167 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1171 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1175 /* The socket must have it's spinlock held when we get
1176 * here, unless it is a TCP_LISTEN socket.
1178 * We have a potential double-lock case here, so even when
1179 * doing backlog processing we use the BH locking scheme.
1180 * This is because we cannot sleep with the original spinlock
1183 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1185 struct ipv6_pinfo *np = inet6_sk(sk);
1186 struct tcp_sock *tp;
1187 struct sk_buff *opt_skb = NULL;
1189 /* Imagine: socket is IPv6. IPv4 packet arrives,
1190 goes to IPv4 receive handler and backlogged.
1191 From backlog it always goes here. Kerboom...
1192 Fortunately, tcp_rcv_established and rcv_established
1193 handle them correctly, but it is not case with
1194 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1197 if (skb->protocol == htons(ETH_P_IP))
1198 return tcp_v4_do_rcv(sk, skb);
1200 if (sk_filter(sk, skb))
1204 * socket locking is here for SMP purposes as backlog rcv
1205 * is currently called with bh processing disabled.
1208 /* Do Stevens' IPV6_PKTOPTIONS.
1210 Yes, guys, it is the only place in our code, where we
1211 may make it not affecting IPv4.
1212 The rest of code is protocol independent,
1213 and I do not like idea to uglify IPv4.
1215 Actually, all the idea behind IPV6_PKTOPTIONS
1216 looks not very well thought. For now we latch
1217 options, received in the last packet, enqueued
1218 by tcp. Feel free to propose better solution.
1222 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1224 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1225 struct dst_entry *dst = sk->sk_rx_dst;
1227 sock_rps_save_rxhash(sk, skb);
1228 sk_mark_napi_id(sk, skb);
1230 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1231 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1233 sk->sk_rx_dst = NULL;
1237 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1239 goto ipv6_pktoptions;
1243 if (tcp_checksum_complete(skb))
1246 if (sk->sk_state == TCP_LISTEN) {
1247 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1253 sock_rps_save_rxhash(nsk, skb);
1254 sk_mark_napi_id(nsk, skb);
1255 if (tcp_child_process(sk, nsk, skb))
1258 __kfree_skb(opt_skb);
1262 sock_rps_save_rxhash(sk, skb);
1264 if (tcp_rcv_state_process(sk, skb))
1267 goto ipv6_pktoptions;
1271 tcp_v6_send_reset(sk, skb);
1274 __kfree_skb(opt_skb);
1278 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1279 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1284 /* Do you ask, what is it?
1286 1. skb was enqueued by tcp.
1287 2. skb is added to tail of read queue, rather than out of order.
1288 3. socket is not in passive state.
1289 4. Finally, it really contains options, which user wants to receive.
1292 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1293 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1294 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1295 np->mcast_oif = tcp_v6_iif(opt_skb);
1296 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1297 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1298 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1299 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1301 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1302 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1303 skb_set_owner_r(opt_skb, sk);
1304 opt_skb = xchg(&np->pktoptions, opt_skb);
1306 __kfree_skb(opt_skb);
1307 opt_skb = xchg(&np->pktoptions, NULL);
1315 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1316 const struct tcphdr *th)
1318 /* This is tricky: we move IP6CB at its correct location into
1319 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1320 * _decode_session6() uses IP6CB().
1321 * barrier() makes sure compiler won't play aliasing games.
1323 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1324 sizeof(struct inet6_skb_parm));
1327 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1328 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1329 skb->len - th->doff*4);
1330 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1331 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1332 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1333 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1334 TCP_SKB_CB(skb)->sacked = 0;
1337 static void tcp_v6_restore_cb(struct sk_buff *skb)
1339 /* We need to move header back to the beginning if xfrm6_policy_check()
1340 * and tcp_v6_fill_cb() are going to be called again.
1342 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1343 sizeof(struct inet6_skb_parm));
1346 static int tcp_v6_rcv(struct sk_buff *skb)
1348 const struct tcphdr *th;
1349 const struct ipv6hdr *hdr;
1352 struct net *net = dev_net(skb->dev);
1354 if (skb->pkt_type != PACKET_HOST)
1358 * Count it even if it's bad.
1360 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1362 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1367 if (th->doff < sizeof(struct tcphdr)/4)
1369 if (!pskb_may_pull(skb, th->doff*4))
1372 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1376 hdr = ipv6_hdr(skb);
1379 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1380 th->source, th->dest, inet6_iif(skb));
1385 if (sk->sk_state == TCP_TIME_WAIT)
1388 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1389 struct request_sock *req = inet_reqsk(sk);
1390 struct sock *nsk = NULL;
1392 sk = req->rsk_listener;
1393 tcp_v6_fill_cb(skb, hdr, th);
1394 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1398 if (likely(sk->sk_state == TCP_LISTEN)) {
1399 nsk = tcp_check_req(sk, skb, req, false);
1401 inet_csk_reqsk_queue_drop_and_put(sk, req);
1411 tcp_v6_restore_cb(skb);
1412 } else if (tcp_child_process(sk, nsk, skb)) {
1413 tcp_v6_send_reset(nsk, skb);
1419 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1420 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1421 goto discard_and_relse;
1424 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1425 goto discard_and_relse;
1427 tcp_v6_fill_cb(skb, hdr, th);
1429 if (tcp_v6_inbound_md5_hash(sk, skb))
1430 goto discard_and_relse;
1432 if (sk_filter(sk, skb))
1433 goto discard_and_relse;
1437 if (sk->sk_state == TCP_LISTEN) {
1438 ret = tcp_v6_do_rcv(sk, skb);
1439 goto put_and_return;
1442 sk_incoming_cpu_update(sk);
1444 bh_lock_sock_nested(sk);
1445 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1447 if (!sock_owned_by_user(sk)) {
1448 if (!tcp_prequeue(sk, skb))
1449 ret = tcp_v6_do_rcv(sk, skb);
1450 } else if (unlikely(sk_add_backlog(sk, skb,
1451 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1453 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1454 goto discard_and_relse;
1460 return ret ? -1 : 0;
1463 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1466 tcp_v6_fill_cb(skb, hdr, th);
1468 if (tcp_checksum_complete(skb)) {
1470 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1472 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1474 tcp_v6_send_reset(NULL, skb);
1486 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1487 inet_twsk_put(inet_twsk(sk));
1491 tcp_v6_fill_cb(skb, hdr, th);
1493 if (tcp_checksum_complete(skb)) {
1494 inet_twsk_put(inet_twsk(sk));
1498 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1503 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1504 skb, __tcp_hdrlen(th),
1505 &ipv6_hdr(skb)->saddr, th->source,
1506 &ipv6_hdr(skb)->daddr,
1507 ntohs(th->dest), tcp_v6_iif(skb));
1509 struct inet_timewait_sock *tw = inet_twsk(sk);
1510 inet_twsk_deschedule_put(tw);
1512 tcp_v6_restore_cb(skb);
1515 /* Fall through to ACK */
1518 tcp_v6_timewait_ack(sk, skb);
1521 tcp_v6_restore_cb(skb);
1522 tcp_v6_send_reset(sk, skb);
1523 inet_twsk_deschedule_put(inet_twsk(sk));
1525 case TCP_TW_SUCCESS:
1531 static void tcp_v6_early_demux(struct sk_buff *skb)
1533 const struct ipv6hdr *hdr;
1534 const struct tcphdr *th;
1537 if (skb->pkt_type != PACKET_HOST)
1540 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1543 hdr = ipv6_hdr(skb);
1546 if (th->doff < sizeof(struct tcphdr) / 4)
1549 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1550 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1551 &hdr->saddr, th->source,
1552 &hdr->daddr, ntohs(th->dest),
1556 skb->destructor = sock_edemux;
1557 if (sk_fullsock(sk)) {
1558 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1561 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1563 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1564 skb_dst_set_noref(skb, dst);
1569 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1570 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1571 .twsk_unique = tcp_twsk_unique,
1572 .twsk_destructor = tcp_twsk_destructor,
1575 static const struct inet_connection_sock_af_ops ipv6_specific = {
1576 .queue_xmit = inet6_csk_xmit,
1577 .send_check = tcp_v6_send_check,
1578 .rebuild_header = inet6_sk_rebuild_header,
1579 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1580 .conn_request = tcp_v6_conn_request,
1581 .syn_recv_sock = tcp_v6_syn_recv_sock,
1582 .net_header_len = sizeof(struct ipv6hdr),
1583 .net_frag_header_len = sizeof(struct frag_hdr),
1584 .setsockopt = ipv6_setsockopt,
1585 .getsockopt = ipv6_getsockopt,
1586 .addr2sockaddr = inet6_csk_addr2sockaddr,
1587 .sockaddr_len = sizeof(struct sockaddr_in6),
1588 .bind_conflict = inet6_csk_bind_conflict,
1589 #ifdef CONFIG_COMPAT
1590 .compat_setsockopt = compat_ipv6_setsockopt,
1591 .compat_getsockopt = compat_ipv6_getsockopt,
1593 .mtu_reduced = tcp_v6_mtu_reduced,
1596 #ifdef CONFIG_TCP_MD5SIG
1597 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1598 .md5_lookup = tcp_v6_md5_lookup,
1599 .calc_md5_hash = tcp_v6_md5_hash_skb,
1600 .md5_parse = tcp_v6_parse_md5_keys,
1605 * TCP over IPv4 via INET6 API
1607 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1608 .queue_xmit = ip_queue_xmit,
1609 .send_check = tcp_v4_send_check,
1610 .rebuild_header = inet_sk_rebuild_header,
1611 .sk_rx_dst_set = inet_sk_rx_dst_set,
1612 .conn_request = tcp_v6_conn_request,
1613 .syn_recv_sock = tcp_v6_syn_recv_sock,
1614 .net_header_len = sizeof(struct iphdr),
1615 .setsockopt = ipv6_setsockopt,
1616 .getsockopt = ipv6_getsockopt,
1617 .addr2sockaddr = inet6_csk_addr2sockaddr,
1618 .sockaddr_len = sizeof(struct sockaddr_in6),
1619 .bind_conflict = inet6_csk_bind_conflict,
1620 #ifdef CONFIG_COMPAT
1621 .compat_setsockopt = compat_ipv6_setsockopt,
1622 .compat_getsockopt = compat_ipv6_getsockopt,
1624 .mtu_reduced = tcp_v4_mtu_reduced,
1627 #ifdef CONFIG_TCP_MD5SIG
1628 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1629 .md5_lookup = tcp_v4_md5_lookup,
1630 .calc_md5_hash = tcp_v4_md5_hash_skb,
1631 .md5_parse = tcp_v6_parse_md5_keys,
1635 /* NOTE: A lot of things set to zero explicitly by call to
1636 * sk_alloc() so need not be done here.
1638 static int tcp_v6_init_sock(struct sock *sk)
1640 struct inet_connection_sock *icsk = inet_csk(sk);
1644 icsk->icsk_af_ops = &ipv6_specific;
1646 #ifdef CONFIG_TCP_MD5SIG
1647 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1653 static void tcp_v6_destroy_sock(struct sock *sk)
1655 tcp_v4_destroy_sock(sk);
1656 inet6_destroy_sock(sk);
1659 #ifdef CONFIG_PROC_FS
1660 /* Proc filesystem TCPv6 sock list dumping. */
1661 static void get_openreq6(struct seq_file *seq,
1662 const struct request_sock *req, int i)
1664 long ttd = req->rsk_timer.expires - jiffies;
1665 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1666 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1672 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1673 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1675 src->s6_addr32[0], src->s6_addr32[1],
1676 src->s6_addr32[2], src->s6_addr32[3],
1677 inet_rsk(req)->ir_num,
1678 dest->s6_addr32[0], dest->s6_addr32[1],
1679 dest->s6_addr32[2], dest->s6_addr32[3],
1680 ntohs(inet_rsk(req)->ir_rmt_port),
1682 0, 0, /* could print option size, but that is af dependent. */
1683 1, /* timers active (only the expire timer) */
1684 jiffies_to_clock_t(ttd),
1686 from_kuid_munged(seq_user_ns(seq),
1687 sock_i_uid(req->rsk_listener)),
1688 0, /* non standard timer */
1689 0, /* open_requests have no inode */
1693 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1695 const struct in6_addr *dest, *src;
1698 unsigned long timer_expires;
1699 const struct inet_sock *inet = inet_sk(sp);
1700 const struct tcp_sock *tp = tcp_sk(sp);
1701 const struct inet_connection_sock *icsk = inet_csk(sp);
1702 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1706 dest = &sp->sk_v6_daddr;
1707 src = &sp->sk_v6_rcv_saddr;
1708 destp = ntohs(inet->inet_dport);
1709 srcp = ntohs(inet->inet_sport);
1711 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1713 timer_expires = icsk->icsk_timeout;
1714 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1716 timer_expires = icsk->icsk_timeout;
1717 } else if (timer_pending(&sp->sk_timer)) {
1719 timer_expires = sp->sk_timer.expires;
1722 timer_expires = jiffies;
1725 state = sk_state_load(sp);
1726 if (state == TCP_LISTEN)
1727 rx_queue = sp->sk_ack_backlog;
1729 /* Because we don't lock the socket,
1730 * we might find a transient negative value.
1732 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1735 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1736 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1738 src->s6_addr32[0], src->s6_addr32[1],
1739 src->s6_addr32[2], src->s6_addr32[3], srcp,
1740 dest->s6_addr32[0], dest->s6_addr32[1],
1741 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1743 tp->write_seq - tp->snd_una,
1746 jiffies_delta_to_clock_t(timer_expires - jiffies),
1747 icsk->icsk_retransmits,
1748 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1749 icsk->icsk_probes_out,
1751 atomic_read(&sp->sk_refcnt), sp,
1752 jiffies_to_clock_t(icsk->icsk_rto),
1753 jiffies_to_clock_t(icsk->icsk_ack.ato),
1754 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1756 state == TCP_LISTEN ?
1757 fastopenq->max_qlen :
1758 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1762 static void get_timewait6_sock(struct seq_file *seq,
1763 struct inet_timewait_sock *tw, int i)
1765 long delta = tw->tw_timer.expires - jiffies;
1766 const struct in6_addr *dest, *src;
1769 dest = &tw->tw_v6_daddr;
1770 src = &tw->tw_v6_rcv_saddr;
1771 destp = ntohs(tw->tw_dport);
1772 srcp = ntohs(tw->tw_sport);
1775 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1776 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1778 src->s6_addr32[0], src->s6_addr32[1],
1779 src->s6_addr32[2], src->s6_addr32[3], srcp,
1780 dest->s6_addr32[0], dest->s6_addr32[1],
1781 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1782 tw->tw_substate, 0, 0,
1783 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1784 atomic_read(&tw->tw_refcnt), tw);
1787 static int tcp6_seq_show(struct seq_file *seq, void *v)
1789 struct tcp_iter_state *st;
1790 struct sock *sk = v;
1792 if (v == SEQ_START_TOKEN) {
1797 "st tx_queue rx_queue tr tm->when retrnsmt"
1798 " uid timeout inode\n");
1803 if (sk->sk_state == TCP_TIME_WAIT)
1804 get_timewait6_sock(seq, v, st->num);
1805 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1806 get_openreq6(seq, v, st->num);
1808 get_tcp6_sock(seq, v, st->num);
1813 static const struct file_operations tcp6_afinfo_seq_fops = {
1814 .owner = THIS_MODULE,
1815 .open = tcp_seq_open,
1817 .llseek = seq_lseek,
1818 .release = seq_release_net
1821 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1824 .seq_fops = &tcp6_afinfo_seq_fops,
1826 .show = tcp6_seq_show,
1830 int __net_init tcp6_proc_init(struct net *net)
1832 return tcp_proc_register(net, &tcp6_seq_afinfo);
1835 void tcp6_proc_exit(struct net *net)
1837 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1841 static void tcp_v6_clear_sk(struct sock *sk, int size)
1843 struct inet_sock *inet = inet_sk(sk);
1845 /* we do not want to clear pinet6 field, because of RCU lookups */
1846 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1848 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1849 memset(&inet->pinet6 + 1, 0, size);
1852 struct proto tcpv6_prot = {
1854 .owner = THIS_MODULE,
1856 .connect = tcp_v6_connect,
1857 .disconnect = tcp_disconnect,
1858 .accept = inet_csk_accept,
1860 .init = tcp_v6_init_sock,
1861 .destroy = tcp_v6_destroy_sock,
1862 .shutdown = tcp_shutdown,
1863 .setsockopt = tcp_setsockopt,
1864 .getsockopt = tcp_getsockopt,
1865 .recvmsg = tcp_recvmsg,
1866 .sendmsg = tcp_sendmsg,
1867 .sendpage = tcp_sendpage,
1868 .backlog_rcv = tcp_v6_do_rcv,
1869 .release_cb = tcp_release_cb,
1871 .unhash = inet_unhash,
1872 .get_port = inet_csk_get_port,
1873 .enter_memory_pressure = tcp_enter_memory_pressure,
1874 .stream_memory_free = tcp_stream_memory_free,
1875 .sockets_allocated = &tcp_sockets_allocated,
1876 .memory_allocated = &tcp_memory_allocated,
1877 .memory_pressure = &tcp_memory_pressure,
1878 .orphan_count = &tcp_orphan_count,
1879 .sysctl_mem = sysctl_tcp_mem,
1880 .sysctl_wmem = sysctl_tcp_wmem,
1881 .sysctl_rmem = sysctl_tcp_rmem,
1882 .max_header = MAX_TCP_HEADER,
1883 .obj_size = sizeof(struct tcp6_sock),
1884 .slab_flags = SLAB_DESTROY_BY_RCU,
1885 .twsk_prot = &tcp6_timewait_sock_ops,
1886 .rsk_prot = &tcp6_request_sock_ops,
1887 .h.hashinfo = &tcp_hashinfo,
1888 .no_autobind = true,
1889 #ifdef CONFIG_COMPAT
1890 .compat_setsockopt = compat_tcp_setsockopt,
1891 .compat_getsockopt = compat_tcp_getsockopt,
1893 .clear_sk = tcp_v6_clear_sk,
1894 .diag_destroy = tcp_abort,
1897 static const struct inet6_protocol tcpv6_protocol = {
1898 .early_demux = tcp_v6_early_demux,
1899 .handler = tcp_v6_rcv,
1900 .err_handler = tcp_v6_err,
1901 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1904 static struct inet_protosw tcpv6_protosw = {
1905 .type = SOCK_STREAM,
1906 .protocol = IPPROTO_TCP,
1907 .prot = &tcpv6_prot,
1908 .ops = &inet6_stream_ops,
1909 .flags = INET_PROTOSW_PERMANENT |
1913 static int __net_init tcpv6_net_init(struct net *net)
1915 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1916 SOCK_RAW, IPPROTO_TCP, net);
1919 static void __net_exit tcpv6_net_exit(struct net *net)
1921 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1924 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1926 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1929 static struct pernet_operations tcpv6_net_ops = {
1930 .init = tcpv6_net_init,
1931 .exit = tcpv6_net_exit,
1932 .exit_batch = tcpv6_net_exit_batch,
1935 int __init tcpv6_init(void)
1939 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1943 /* register inet6 protocol */
1944 ret = inet6_register_protosw(&tcpv6_protosw);
1946 goto out_tcpv6_protocol;
1948 ret = register_pernet_subsys(&tcpv6_net_ops);
1950 goto out_tcpv6_protosw;
1955 inet6_unregister_protosw(&tcpv6_protosw);
1957 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1961 void tcpv6_exit(void)
1963 unregister_pernet_subsys(&tcpv6_net_ops);
1964 inet6_unregister_protosw(&tcpv6_protosw);
1965 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);