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 <crypto/hash.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, lockdep_sock_is_held(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;
333 sk = __inet6_lookup_established(net, &tcp_hashinfo,
334 &hdr->daddr, th->dest,
335 &hdr->saddr, ntohs(th->source),
339 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
344 if (sk->sk_state == TCP_TIME_WAIT) {
345 inet_twsk_put(inet_twsk(sk));
348 seq = ntohl(th->seq);
349 fatal = icmpv6_err_convert(type, code, &err);
350 if (sk->sk_state == TCP_NEW_SYN_RECV)
351 return tcp_req_err(sk, seq, fatal);
354 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
355 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
357 if (sk->sk_state == TCP_CLOSE)
360 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
361 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
366 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
367 fastopen = tp->fastopen_rsk;
368 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
369 if (sk->sk_state != TCP_LISTEN &&
370 !between(seq, snd_una, tp->snd_nxt)) {
371 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
377 if (type == NDISC_REDIRECT) {
378 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
381 dst->ops->redirect(dst, sk, skb);
385 if (type == ICMPV6_PKT_TOOBIG) {
386 /* We are not interested in TCP_LISTEN and open_requests
387 * (SYN-ACKs send out by Linux are always <576bytes so
388 * they should go through unfragmented).
390 if (sk->sk_state == TCP_LISTEN)
393 if (!ip6_sk_accept_pmtu(sk))
396 tp->mtu_info = ntohl(info);
397 if (!sock_owned_by_user(sk))
398 tcp_v6_mtu_reduced(sk);
399 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
406 /* Might be for an request_sock */
407 switch (sk->sk_state) {
410 /* Only in fast or simultaneous open. If a fast open socket is
411 * is already accepted it is treated as a connected one below.
413 if (fastopen && !fastopen->sk)
416 if (!sock_owned_by_user(sk)) {
418 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
422 sk->sk_err_soft = err;
426 if (!sock_owned_by_user(sk) && np->recverr) {
428 sk->sk_error_report(sk);
430 sk->sk_err_soft = err;
438 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
440 struct request_sock *req,
441 struct tcp_fastopen_cookie *foc,
442 enum tcp_synack_type synack_type)
444 struct inet_request_sock *ireq = inet_rsk(req);
445 struct ipv6_pinfo *np = inet6_sk(sk);
446 struct ipv6_txoptions *opt;
447 struct flowi6 *fl6 = &fl->u.ip6;
451 /* First, grab a route. */
452 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
453 IPPROTO_TCP)) == NULL)
456 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
459 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
460 &ireq->ir_v6_rmt_addr);
462 fl6->daddr = ireq->ir_v6_rmt_addr;
463 if (np->repflow && ireq->pktopts)
464 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
467 opt = ireq->ipv6_opt;
469 opt = rcu_dereference(np->opt);
470 err = ip6_xmit(sk, skb, fl6, opt, np->tclass);
472 err = net_xmit_eval(err);
480 static void tcp_v6_reqsk_destructor(struct request_sock *req)
482 kfree(inet_rsk(req)->ipv6_opt);
483 kfree_skb(inet_rsk(req)->pktopts);
486 #ifdef CONFIG_TCP_MD5SIG
487 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
488 const struct in6_addr *addr)
490 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
493 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
494 const struct sock *addr_sk)
496 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
499 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
502 struct tcp_md5sig cmd;
503 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
505 if (optlen < sizeof(cmd))
508 if (copy_from_user(&cmd, optval, sizeof(cmd)))
511 if (sin6->sin6_family != AF_INET6)
514 if (!cmd.tcpm_keylen) {
515 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
516 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
518 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
522 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
525 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
526 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
527 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
529 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
530 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
533 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
534 const struct in6_addr *daddr,
535 const struct in6_addr *saddr,
536 const struct tcphdr *th, int nbytes)
538 struct tcp6_pseudohdr *bp;
539 struct scatterlist sg;
543 /* 1. TCP pseudo-header (RFC2460) */
546 bp->protocol = cpu_to_be32(IPPROTO_TCP);
547 bp->len = cpu_to_be32(nbytes);
549 _th = (struct tcphdr *)(bp + 1);
550 memcpy(_th, th, sizeof(*th));
553 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
554 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
555 sizeof(*bp) + sizeof(*th));
556 return crypto_ahash_update(hp->md5_req);
559 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
560 const struct in6_addr *daddr, struct in6_addr *saddr,
561 const struct tcphdr *th)
563 struct tcp_md5sig_pool *hp;
564 struct ahash_request *req;
566 hp = tcp_get_md5sig_pool();
568 goto clear_hash_noput;
571 if (crypto_ahash_init(req))
573 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
575 if (tcp_md5_hash_key(hp, key))
577 ahash_request_set_crypt(req, NULL, md5_hash, 0);
578 if (crypto_ahash_final(req))
581 tcp_put_md5sig_pool();
585 tcp_put_md5sig_pool();
587 memset(md5_hash, 0, 16);
591 static int tcp_v6_md5_hash_skb(char *md5_hash,
592 const struct tcp_md5sig_key *key,
593 const struct sock *sk,
594 const struct sk_buff *skb)
596 const struct in6_addr *saddr, *daddr;
597 struct tcp_md5sig_pool *hp;
598 struct ahash_request *req;
599 const struct tcphdr *th = tcp_hdr(skb);
601 if (sk) { /* valid for establish/request sockets */
602 saddr = &sk->sk_v6_rcv_saddr;
603 daddr = &sk->sk_v6_daddr;
605 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
606 saddr = &ip6h->saddr;
607 daddr = &ip6h->daddr;
610 hp = tcp_get_md5sig_pool();
612 goto clear_hash_noput;
615 if (crypto_ahash_init(req))
618 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
620 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
622 if (tcp_md5_hash_key(hp, key))
624 ahash_request_set_crypt(req, NULL, md5_hash, 0);
625 if (crypto_ahash_final(req))
628 tcp_put_md5sig_pool();
632 tcp_put_md5sig_pool();
634 memset(md5_hash, 0, 16);
640 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
641 const struct sk_buff *skb)
643 #ifdef CONFIG_TCP_MD5SIG
644 const __u8 *hash_location = NULL;
645 struct tcp_md5sig_key *hash_expected;
646 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
647 const struct tcphdr *th = tcp_hdr(skb);
651 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
652 hash_location = tcp_parse_md5sig_option(th);
654 /* We've parsed the options - do we have a hash? */
655 if (!hash_expected && !hash_location)
658 if (hash_expected && !hash_location) {
659 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
663 if (!hash_expected && hash_location) {
664 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
668 /* check the signature */
669 genhash = tcp_v6_md5_hash_skb(newhash,
673 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
674 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
675 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
676 genhash ? "failed" : "mismatch",
677 &ip6h->saddr, ntohs(th->source),
678 &ip6h->daddr, ntohs(th->dest));
685 static void tcp_v6_init_req(struct request_sock *req,
686 const struct sock *sk_listener,
689 struct inet_request_sock *ireq = inet_rsk(req);
690 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
692 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
693 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
695 /* So that link locals have meaning */
696 if (!sk_listener->sk_bound_dev_if &&
697 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
698 ireq->ir_iif = tcp_v6_iif(skb);
700 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
701 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
702 np->rxopt.bits.rxinfo ||
703 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
704 np->rxopt.bits.rxohlim || np->repflow)) {
705 atomic_inc(&skb->users);
710 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
712 const struct request_sock *req,
717 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
720 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
722 .obj_size = sizeof(struct tcp6_request_sock),
723 .rtx_syn_ack = tcp_rtx_synack,
724 .send_ack = tcp_v6_reqsk_send_ack,
725 .destructor = tcp_v6_reqsk_destructor,
726 .send_reset = tcp_v6_send_reset,
727 .syn_ack_timeout = tcp_syn_ack_timeout,
730 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
731 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
732 sizeof(struct ipv6hdr),
733 #ifdef CONFIG_TCP_MD5SIG
734 .req_md5_lookup = tcp_v6_md5_lookup,
735 .calc_md5_hash = tcp_v6_md5_hash_skb,
737 .init_req = tcp_v6_init_req,
738 #ifdef CONFIG_SYN_COOKIES
739 .cookie_init_seq = cookie_v6_init_sequence,
741 .route_req = tcp_v6_route_req,
742 .init_seq = tcp_v6_init_sequence,
743 .send_synack = tcp_v6_send_synack,
746 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
747 u32 ack, u32 win, u32 tsval, u32 tsecr,
748 int oif, struct tcp_md5sig_key *key, int rst,
749 u8 tclass, __be32 label)
751 const struct tcphdr *th = tcp_hdr(skb);
753 struct sk_buff *buff;
755 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
756 struct sock *ctl_sk = net->ipv6.tcp_sk;
757 unsigned int tot_len = sizeof(struct tcphdr);
758 struct dst_entry *dst;
762 tot_len += TCPOLEN_TSTAMP_ALIGNED;
763 #ifdef CONFIG_TCP_MD5SIG
765 tot_len += TCPOLEN_MD5SIG_ALIGNED;
768 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
773 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
775 t1 = (struct tcphdr *) skb_push(buff, tot_len);
776 skb_reset_transport_header(buff);
778 /* Swap the send and the receive. */
779 memset(t1, 0, sizeof(*t1));
780 t1->dest = th->source;
781 t1->source = th->dest;
782 t1->doff = tot_len / 4;
783 t1->seq = htonl(seq);
784 t1->ack_seq = htonl(ack);
785 t1->ack = !rst || !th->ack;
787 t1->window = htons(win);
789 topt = (__be32 *)(t1 + 1);
792 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
793 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
794 *topt++ = htonl(tsval);
795 *topt++ = htonl(tsecr);
798 #ifdef CONFIG_TCP_MD5SIG
800 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
801 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
802 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
803 &ipv6_hdr(skb)->saddr,
804 &ipv6_hdr(skb)->daddr, t1);
808 memset(&fl6, 0, sizeof(fl6));
809 fl6.daddr = ipv6_hdr(skb)->saddr;
810 fl6.saddr = ipv6_hdr(skb)->daddr;
811 fl6.flowlabel = label;
813 buff->ip_summed = CHECKSUM_PARTIAL;
816 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
818 fl6.flowi6_proto = IPPROTO_TCP;
819 if (rt6_need_strict(&fl6.daddr) && !oif)
820 fl6.flowi6_oif = tcp_v6_iif(skb);
822 fl6.flowi6_oif = oif ? : skb->skb_iif;
824 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
825 fl6.fl6_dport = t1->dest;
826 fl6.fl6_sport = t1->source;
827 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
829 /* Pass a socket to ip6_dst_lookup either it is for RST
830 * Underlying function will use this to retrieve the network
833 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
835 skb_dst_set(buff, dst);
836 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
837 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
839 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
846 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
848 const struct tcphdr *th = tcp_hdr(skb);
849 u32 seq = 0, ack_seq = 0;
850 struct tcp_md5sig_key *key = NULL;
851 #ifdef CONFIG_TCP_MD5SIG
852 const __u8 *hash_location = NULL;
853 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
854 unsigned char newhash[16];
856 struct sock *sk1 = NULL;
863 /* If sk not NULL, it means we did a successful lookup and incoming
864 * route had to be correct. prequeue might have dropped our dst.
866 if (!sk && !ipv6_unicast_destination(skb))
869 #ifdef CONFIG_TCP_MD5SIG
871 hash_location = tcp_parse_md5sig_option(th);
872 if (sk && sk_fullsock(sk)) {
873 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
874 } else if (hash_location) {
876 * active side is lost. Try to find listening socket through
877 * source port, and then find md5 key through listening socket.
878 * we are not loose security here:
879 * Incoming packet is checked with md5 hash with finding key,
880 * no RST generated if md5 hash doesn't match.
882 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
883 &tcp_hashinfo, NULL, 0,
885 th->source, &ipv6h->daddr,
886 ntohs(th->source), tcp_v6_iif(skb));
890 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
894 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
895 if (genhash || memcmp(hash_location, newhash, 16) != 0)
901 seq = ntohl(th->ack_seq);
903 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
906 oif = sk ? sk->sk_bound_dev_if : 0;
907 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
909 #ifdef CONFIG_TCP_MD5SIG
915 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
916 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
917 struct tcp_md5sig_key *key, u8 tclass,
920 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
924 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
926 struct inet_timewait_sock *tw = inet_twsk(sk);
927 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
929 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
930 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
931 tcp_time_stamp + tcptw->tw_ts_offset,
932 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
933 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
938 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
939 struct request_sock *req)
941 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
942 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
945 * The window field (SEG.WND) of every outgoing segment, with the
946 * exception of <SYN> segments, MUST be right-shifted by
947 * Rcv.Wind.Shift bits:
949 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
950 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
951 tcp_rsk(req)->rcv_nxt,
952 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
953 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
954 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
959 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
961 #ifdef CONFIG_SYN_COOKIES
962 const struct tcphdr *th = tcp_hdr(skb);
965 sk = cookie_v6_check(sk, skb);
970 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
972 if (skb->protocol == htons(ETH_P_IP))
973 return tcp_v4_conn_request(sk, skb);
975 if (!ipv6_unicast_destination(skb))
978 return tcp_conn_request(&tcp6_request_sock_ops,
979 &tcp_request_sock_ipv6_ops, sk, skb);
983 return 0; /* don't send reset */
986 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
987 struct request_sock *req,
988 struct dst_entry *dst,
989 struct request_sock *req_unhash,
992 struct inet_request_sock *ireq;
993 struct ipv6_pinfo *newnp;
994 const struct ipv6_pinfo *np = inet6_sk(sk);
995 struct ipv6_txoptions *opt;
996 struct tcp6_sock *newtcp6sk;
997 struct inet_sock *newinet;
998 struct tcp_sock *newtp;
1000 #ifdef CONFIG_TCP_MD5SIG
1001 struct tcp_md5sig_key *key;
1005 if (skb->protocol == htons(ETH_P_IP)) {
1010 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1011 req_unhash, own_req);
1016 newtcp6sk = (struct tcp6_sock *)newsk;
1017 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1019 newinet = inet_sk(newsk);
1020 newnp = inet6_sk(newsk);
1021 newtp = tcp_sk(newsk);
1023 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1025 newnp->saddr = newsk->sk_v6_rcv_saddr;
1027 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1028 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1029 #ifdef CONFIG_TCP_MD5SIG
1030 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1033 newnp->ipv6_ac_list = NULL;
1034 newnp->ipv6_fl_list = NULL;
1035 newnp->pktoptions = NULL;
1037 newnp->mcast_oif = tcp_v6_iif(skb);
1038 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1039 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1041 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1044 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1045 * here, tcp_create_openreq_child now does this for us, see the comment in
1046 * that function for the gory details. -acme
1049 /* It is tricky place. Until this moment IPv4 tcp
1050 worked with IPv6 icsk.icsk_af_ops.
1053 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1058 ireq = inet_rsk(req);
1060 if (sk_acceptq_is_full(sk))
1064 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1069 newsk = tcp_create_openreq_child(sk, req, skb);
1074 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1075 * count here, tcp_create_openreq_child now does this for us, see the
1076 * comment in that function for the gory details. -acme
1079 newsk->sk_gso_type = SKB_GSO_TCPV6;
1080 ip6_dst_store(newsk, dst, NULL, NULL);
1081 inet6_sk_rx_dst_set(newsk, skb);
1083 newtcp6sk = (struct tcp6_sock *)newsk;
1084 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1086 newtp = tcp_sk(newsk);
1087 newinet = inet_sk(newsk);
1088 newnp = inet6_sk(newsk);
1090 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1092 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1093 newnp->saddr = ireq->ir_v6_loc_addr;
1094 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1095 newsk->sk_bound_dev_if = ireq->ir_iif;
1097 /* Now IPv6 options...
1099 First: no IPv4 options.
1101 newinet->inet_opt = NULL;
1102 newnp->ipv6_ac_list = NULL;
1103 newnp->ipv6_fl_list = NULL;
1106 newnp->rxopt.all = np->rxopt.all;
1108 newnp->pktoptions = NULL;
1110 newnp->mcast_oif = tcp_v6_iif(skb);
1111 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1112 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1114 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1116 /* Clone native IPv6 options from listening socket (if any)
1118 Yes, keeping reference count would be much more clever,
1119 but we make one more one thing there: reattach optmem
1122 opt = ireq->ipv6_opt;
1124 opt = rcu_dereference(np->opt);
1126 opt = ipv6_dup_options(newsk, opt);
1127 RCU_INIT_POINTER(newnp->opt, opt);
1129 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1131 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1134 tcp_ca_openreq_child(newsk, dst);
1136 tcp_sync_mss(newsk, dst_mtu(dst));
1137 newtp->advmss = dst_metric_advmss(dst);
1138 if (tcp_sk(sk)->rx_opt.user_mss &&
1139 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1140 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1142 tcp_initialize_rcv_mss(newsk);
1144 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1145 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1147 #ifdef CONFIG_TCP_MD5SIG
1148 /* Copy over the MD5 key from the original socket */
1149 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1151 /* We're using one, so create a matching key
1152 * on the newsk structure. If we fail to get
1153 * memory, then we end up not copying the key
1156 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1157 AF_INET6, key->key, key->keylen,
1158 sk_gfp_mask(sk, GFP_ATOMIC));
1162 if (__inet_inherit_port(sk, newsk) < 0) {
1163 inet_csk_prepare_forced_close(newsk);
1167 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1169 tcp_move_syn(newtp, req);
1171 /* Clone pktoptions received with SYN, if we own the req */
1172 if (ireq->pktopts) {
1173 newnp->pktoptions = skb_clone(ireq->pktopts,
1174 sk_gfp_mask(sk, GFP_ATOMIC));
1175 consume_skb(ireq->pktopts);
1176 ireq->pktopts = NULL;
1177 if (newnp->pktoptions)
1178 skb_set_owner_r(newnp->pktoptions, newsk);
1185 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1193 /* The socket must have it's spinlock held when we get
1194 * here, unless it is a TCP_LISTEN socket.
1196 * We have a potential double-lock case here, so even when
1197 * doing backlog processing we use the BH locking scheme.
1198 * This is because we cannot sleep with the original spinlock
1201 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1203 struct ipv6_pinfo *np = inet6_sk(sk);
1204 struct tcp_sock *tp;
1205 struct sk_buff *opt_skb = NULL;
1207 /* Imagine: socket is IPv6. IPv4 packet arrives,
1208 goes to IPv4 receive handler and backlogged.
1209 From backlog it always goes here. Kerboom...
1210 Fortunately, tcp_rcv_established and rcv_established
1211 handle them correctly, but it is not case with
1212 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1215 if (skb->protocol == htons(ETH_P_IP))
1216 return tcp_v4_do_rcv(sk, skb);
1218 if (sk_filter(sk, skb))
1222 * socket locking is here for SMP purposes as backlog rcv
1223 * is currently called with bh processing disabled.
1226 /* Do Stevens' IPV6_PKTOPTIONS.
1228 Yes, guys, it is the only place in our code, where we
1229 may make it not affecting IPv4.
1230 The rest of code is protocol independent,
1231 and I do not like idea to uglify IPv4.
1233 Actually, all the idea behind IPV6_PKTOPTIONS
1234 looks not very well thought. For now we latch
1235 options, received in the last packet, enqueued
1236 by tcp. Feel free to propose better solution.
1240 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1242 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1243 struct dst_entry *dst = sk->sk_rx_dst;
1245 sock_rps_save_rxhash(sk, skb);
1246 sk_mark_napi_id(sk, skb);
1248 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1249 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1251 sk->sk_rx_dst = NULL;
1255 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1257 goto ipv6_pktoptions;
1261 if (tcp_checksum_complete(skb))
1264 if (sk->sk_state == TCP_LISTEN) {
1265 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1271 sock_rps_save_rxhash(nsk, skb);
1272 sk_mark_napi_id(nsk, skb);
1273 if (tcp_child_process(sk, nsk, skb))
1276 __kfree_skb(opt_skb);
1280 sock_rps_save_rxhash(sk, skb);
1282 if (tcp_rcv_state_process(sk, skb))
1285 goto ipv6_pktoptions;
1289 tcp_v6_send_reset(sk, skb);
1292 __kfree_skb(opt_skb);
1296 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1297 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1302 /* Do you ask, what is it?
1304 1. skb was enqueued by tcp.
1305 2. skb is added to tail of read queue, rather than out of order.
1306 3. socket is not in passive state.
1307 4. Finally, it really contains options, which user wants to receive.
1310 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1311 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1312 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1313 np->mcast_oif = tcp_v6_iif(opt_skb);
1314 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1315 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1316 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1317 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1319 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1320 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1321 skb_set_owner_r(opt_skb, sk);
1322 opt_skb = xchg(&np->pktoptions, opt_skb);
1324 __kfree_skb(opt_skb);
1325 opt_skb = xchg(&np->pktoptions, NULL);
1333 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1334 const struct tcphdr *th)
1336 /* This is tricky: we move IP6CB at its correct location into
1337 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1338 * _decode_session6() uses IP6CB().
1339 * barrier() makes sure compiler won't play aliasing games.
1341 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1342 sizeof(struct inet6_skb_parm));
1345 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1346 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1347 skb->len - th->doff*4);
1348 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1349 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1350 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1351 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1352 TCP_SKB_CB(skb)->sacked = 0;
1355 static void tcp_v6_restore_cb(struct sk_buff *skb)
1357 /* We need to move header back to the beginning if xfrm6_policy_check()
1358 * and tcp_v6_fill_cb() are going to be called again.
1360 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1361 sizeof(struct inet6_skb_parm));
1364 static int tcp_v6_rcv(struct sk_buff *skb)
1366 const struct tcphdr *th;
1367 const struct ipv6hdr *hdr;
1371 struct net *net = dev_net(skb->dev);
1373 if (skb->pkt_type != PACKET_HOST)
1377 * Count it even if it's bad.
1379 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1381 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1384 th = (const struct tcphdr *)skb->data;
1386 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1388 if (!pskb_may_pull(skb, th->doff*4))
1391 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1394 th = (const struct tcphdr *)skb->data;
1395 hdr = ipv6_hdr(skb);
1398 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1399 th->source, th->dest, inet6_iif(skb),
1405 if (sk->sk_state == TCP_TIME_WAIT)
1408 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1409 struct request_sock *req = inet_reqsk(sk);
1412 sk = req->rsk_listener;
1413 tcp_v6_fill_cb(skb, hdr, th);
1414 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1415 sk_drops_add(sk, skb);
1419 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1420 inet_csk_reqsk_queue_drop_and_put(sk, req);
1425 nsk = tcp_check_req(sk, skb, req, false);
1428 goto discard_and_relse;
1432 tcp_v6_restore_cb(skb);
1433 } else if (tcp_child_process(sk, nsk, skb)) {
1434 tcp_v6_send_reset(nsk, skb);
1435 goto discard_and_relse;
1441 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1442 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1443 goto discard_and_relse;
1446 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1447 goto discard_and_relse;
1449 tcp_v6_fill_cb(skb, hdr, th);
1451 if (tcp_v6_inbound_md5_hash(sk, skb))
1452 goto discard_and_relse;
1454 if (sk_filter(sk, skb))
1455 goto discard_and_relse;
1459 if (sk->sk_state == TCP_LISTEN) {
1460 ret = tcp_v6_do_rcv(sk, skb);
1461 goto put_and_return;
1464 sk_incoming_cpu_update(sk);
1466 bh_lock_sock_nested(sk);
1467 tcp_segs_in(tcp_sk(sk), skb);
1469 if (!sock_owned_by_user(sk)) {
1470 if (!tcp_prequeue(sk, skb))
1471 ret = tcp_v6_do_rcv(sk, skb);
1472 } else if (tcp_add_backlog(sk, skb)) {
1473 goto discard_and_relse;
1480 return ret ? -1 : 0;
1483 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1486 tcp_v6_fill_cb(skb, hdr, th);
1488 if (tcp_checksum_complete(skb)) {
1490 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1492 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1494 tcp_v6_send_reset(NULL, skb);
1502 sk_drops_add(sk, skb);
1508 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1509 inet_twsk_put(inet_twsk(sk));
1513 tcp_v6_fill_cb(skb, hdr, th);
1515 if (tcp_checksum_complete(skb)) {
1516 inet_twsk_put(inet_twsk(sk));
1520 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1525 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1526 skb, __tcp_hdrlen(th),
1527 &ipv6_hdr(skb)->saddr, th->source,
1528 &ipv6_hdr(skb)->daddr,
1529 ntohs(th->dest), tcp_v6_iif(skb));
1531 struct inet_timewait_sock *tw = inet_twsk(sk);
1532 inet_twsk_deschedule_put(tw);
1534 tcp_v6_restore_cb(skb);
1538 /* Fall through to ACK */
1541 tcp_v6_timewait_ack(sk, skb);
1544 tcp_v6_restore_cb(skb);
1545 tcp_v6_send_reset(sk, skb);
1546 inet_twsk_deschedule_put(inet_twsk(sk));
1548 case TCP_TW_SUCCESS:
1554 static void tcp_v6_early_demux(struct sk_buff *skb)
1556 const struct ipv6hdr *hdr;
1557 const struct tcphdr *th;
1560 if (skb->pkt_type != PACKET_HOST)
1563 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1566 hdr = ipv6_hdr(skb);
1569 if (th->doff < sizeof(struct tcphdr) / 4)
1572 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1573 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1574 &hdr->saddr, th->source,
1575 &hdr->daddr, ntohs(th->dest),
1579 skb->destructor = sock_edemux;
1580 if (sk_fullsock(sk)) {
1581 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1584 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1586 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1587 skb_dst_set_noref(skb, dst);
1592 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1593 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1594 .twsk_unique = tcp_twsk_unique,
1595 .twsk_destructor = tcp_twsk_destructor,
1598 static const struct inet_connection_sock_af_ops ipv6_specific = {
1599 .queue_xmit = inet6_csk_xmit,
1600 .send_check = tcp_v6_send_check,
1601 .rebuild_header = inet6_sk_rebuild_header,
1602 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1603 .conn_request = tcp_v6_conn_request,
1604 .syn_recv_sock = tcp_v6_syn_recv_sock,
1605 .net_header_len = sizeof(struct ipv6hdr),
1606 .net_frag_header_len = sizeof(struct frag_hdr),
1607 .setsockopt = ipv6_setsockopt,
1608 .getsockopt = ipv6_getsockopt,
1609 .addr2sockaddr = inet6_csk_addr2sockaddr,
1610 .sockaddr_len = sizeof(struct sockaddr_in6),
1611 .bind_conflict = inet6_csk_bind_conflict,
1612 #ifdef CONFIG_COMPAT
1613 .compat_setsockopt = compat_ipv6_setsockopt,
1614 .compat_getsockopt = compat_ipv6_getsockopt,
1616 .mtu_reduced = tcp_v6_mtu_reduced,
1619 #ifdef CONFIG_TCP_MD5SIG
1620 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1621 .md5_lookup = tcp_v6_md5_lookup,
1622 .calc_md5_hash = tcp_v6_md5_hash_skb,
1623 .md5_parse = tcp_v6_parse_md5_keys,
1628 * TCP over IPv4 via INET6 API
1630 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1631 .queue_xmit = ip_queue_xmit,
1632 .send_check = tcp_v4_send_check,
1633 .rebuild_header = inet_sk_rebuild_header,
1634 .sk_rx_dst_set = inet_sk_rx_dst_set,
1635 .conn_request = tcp_v6_conn_request,
1636 .syn_recv_sock = tcp_v6_syn_recv_sock,
1637 .net_header_len = sizeof(struct iphdr),
1638 .setsockopt = ipv6_setsockopt,
1639 .getsockopt = ipv6_getsockopt,
1640 .addr2sockaddr = inet6_csk_addr2sockaddr,
1641 .sockaddr_len = sizeof(struct sockaddr_in6),
1642 .bind_conflict = inet6_csk_bind_conflict,
1643 #ifdef CONFIG_COMPAT
1644 .compat_setsockopt = compat_ipv6_setsockopt,
1645 .compat_getsockopt = compat_ipv6_getsockopt,
1647 .mtu_reduced = tcp_v4_mtu_reduced,
1650 #ifdef CONFIG_TCP_MD5SIG
1651 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1652 .md5_lookup = tcp_v4_md5_lookup,
1653 .calc_md5_hash = tcp_v4_md5_hash_skb,
1654 .md5_parse = tcp_v6_parse_md5_keys,
1658 /* NOTE: A lot of things set to zero explicitly by call to
1659 * sk_alloc() so need not be done here.
1661 static int tcp_v6_init_sock(struct sock *sk)
1663 struct inet_connection_sock *icsk = inet_csk(sk);
1667 icsk->icsk_af_ops = &ipv6_specific;
1669 #ifdef CONFIG_TCP_MD5SIG
1670 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1676 static void tcp_v6_destroy_sock(struct sock *sk)
1678 tcp_v4_destroy_sock(sk);
1679 inet6_destroy_sock(sk);
1682 #ifdef CONFIG_PROC_FS
1683 /* Proc filesystem TCPv6 sock list dumping. */
1684 static void get_openreq6(struct seq_file *seq,
1685 const struct request_sock *req, int i)
1687 long ttd = req->rsk_timer.expires - jiffies;
1688 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1689 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1695 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1696 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1698 src->s6_addr32[0], src->s6_addr32[1],
1699 src->s6_addr32[2], src->s6_addr32[3],
1700 inet_rsk(req)->ir_num,
1701 dest->s6_addr32[0], dest->s6_addr32[1],
1702 dest->s6_addr32[2], dest->s6_addr32[3],
1703 ntohs(inet_rsk(req)->ir_rmt_port),
1705 0, 0, /* could print option size, but that is af dependent. */
1706 1, /* timers active (only the expire timer) */
1707 jiffies_to_clock_t(ttd),
1709 from_kuid_munged(seq_user_ns(seq),
1710 sock_i_uid(req->rsk_listener)),
1711 0, /* non standard timer */
1712 0, /* open_requests have no inode */
1716 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1718 const struct in6_addr *dest, *src;
1721 unsigned long timer_expires;
1722 const struct inet_sock *inet = inet_sk(sp);
1723 const struct tcp_sock *tp = tcp_sk(sp);
1724 const struct inet_connection_sock *icsk = inet_csk(sp);
1725 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1729 dest = &sp->sk_v6_daddr;
1730 src = &sp->sk_v6_rcv_saddr;
1731 destp = ntohs(inet->inet_dport);
1732 srcp = ntohs(inet->inet_sport);
1734 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1735 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
1736 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1738 timer_expires = icsk->icsk_timeout;
1739 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1741 timer_expires = icsk->icsk_timeout;
1742 } else if (timer_pending(&sp->sk_timer)) {
1744 timer_expires = sp->sk_timer.expires;
1747 timer_expires = jiffies;
1750 state = sk_state_load(sp);
1751 if (state == TCP_LISTEN)
1752 rx_queue = sp->sk_ack_backlog;
1754 /* Because we don't lock the socket,
1755 * we might find a transient negative value.
1757 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1760 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1761 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1763 src->s6_addr32[0], src->s6_addr32[1],
1764 src->s6_addr32[2], src->s6_addr32[3], srcp,
1765 dest->s6_addr32[0], dest->s6_addr32[1],
1766 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1768 tp->write_seq - tp->snd_una,
1771 jiffies_delta_to_clock_t(timer_expires - jiffies),
1772 icsk->icsk_retransmits,
1773 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1774 icsk->icsk_probes_out,
1776 atomic_read(&sp->sk_refcnt), sp,
1777 jiffies_to_clock_t(icsk->icsk_rto),
1778 jiffies_to_clock_t(icsk->icsk_ack.ato),
1779 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1781 state == TCP_LISTEN ?
1782 fastopenq->max_qlen :
1783 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1787 static void get_timewait6_sock(struct seq_file *seq,
1788 struct inet_timewait_sock *tw, int i)
1790 long delta = tw->tw_timer.expires - jiffies;
1791 const struct in6_addr *dest, *src;
1794 dest = &tw->tw_v6_daddr;
1795 src = &tw->tw_v6_rcv_saddr;
1796 destp = ntohs(tw->tw_dport);
1797 srcp = ntohs(tw->tw_sport);
1800 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1801 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1803 src->s6_addr32[0], src->s6_addr32[1],
1804 src->s6_addr32[2], src->s6_addr32[3], srcp,
1805 dest->s6_addr32[0], dest->s6_addr32[1],
1806 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1807 tw->tw_substate, 0, 0,
1808 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1809 atomic_read(&tw->tw_refcnt), tw);
1812 static int tcp6_seq_show(struct seq_file *seq, void *v)
1814 struct tcp_iter_state *st;
1815 struct sock *sk = v;
1817 if (v == SEQ_START_TOKEN) {
1822 "st tx_queue rx_queue tr tm->when retrnsmt"
1823 " uid timeout inode\n");
1828 if (sk->sk_state == TCP_TIME_WAIT)
1829 get_timewait6_sock(seq, v, st->num);
1830 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1831 get_openreq6(seq, v, st->num);
1833 get_tcp6_sock(seq, v, st->num);
1838 static const struct file_operations tcp6_afinfo_seq_fops = {
1839 .owner = THIS_MODULE,
1840 .open = tcp_seq_open,
1842 .llseek = seq_lseek,
1843 .release = seq_release_net
1846 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1849 .seq_fops = &tcp6_afinfo_seq_fops,
1851 .show = tcp6_seq_show,
1855 int __net_init tcp6_proc_init(struct net *net)
1857 return tcp_proc_register(net, &tcp6_seq_afinfo);
1860 void tcp6_proc_exit(struct net *net)
1862 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1866 struct proto tcpv6_prot = {
1868 .owner = THIS_MODULE,
1870 .connect = tcp_v6_connect,
1871 .disconnect = tcp_disconnect,
1872 .accept = inet_csk_accept,
1874 .init = tcp_v6_init_sock,
1875 .destroy = tcp_v6_destroy_sock,
1876 .shutdown = tcp_shutdown,
1877 .setsockopt = tcp_setsockopt,
1878 .getsockopt = tcp_getsockopt,
1879 .recvmsg = tcp_recvmsg,
1880 .sendmsg = tcp_sendmsg,
1881 .sendpage = tcp_sendpage,
1882 .backlog_rcv = tcp_v6_do_rcv,
1883 .release_cb = tcp_release_cb,
1885 .unhash = inet_unhash,
1886 .get_port = inet_csk_get_port,
1887 .enter_memory_pressure = tcp_enter_memory_pressure,
1888 .stream_memory_free = tcp_stream_memory_free,
1889 .sockets_allocated = &tcp_sockets_allocated,
1890 .memory_allocated = &tcp_memory_allocated,
1891 .memory_pressure = &tcp_memory_pressure,
1892 .orphan_count = &tcp_orphan_count,
1893 .sysctl_mem = sysctl_tcp_mem,
1894 .sysctl_wmem = sysctl_tcp_wmem,
1895 .sysctl_rmem = sysctl_tcp_rmem,
1896 .max_header = MAX_TCP_HEADER,
1897 .obj_size = sizeof(struct tcp6_sock),
1898 .slab_flags = SLAB_DESTROY_BY_RCU,
1899 .twsk_prot = &tcp6_timewait_sock_ops,
1900 .rsk_prot = &tcp6_request_sock_ops,
1901 .h.hashinfo = &tcp_hashinfo,
1902 .no_autobind = true,
1903 #ifdef CONFIG_COMPAT
1904 .compat_setsockopt = compat_tcp_setsockopt,
1905 .compat_getsockopt = compat_tcp_getsockopt,
1907 .diag_destroy = tcp_abort,
1910 static const struct inet6_protocol tcpv6_protocol = {
1911 .early_demux = tcp_v6_early_demux,
1912 .handler = tcp_v6_rcv,
1913 .err_handler = tcp_v6_err,
1914 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1917 static struct inet_protosw tcpv6_protosw = {
1918 .type = SOCK_STREAM,
1919 .protocol = IPPROTO_TCP,
1920 .prot = &tcpv6_prot,
1921 .ops = &inet6_stream_ops,
1922 .flags = INET_PROTOSW_PERMANENT |
1926 static int __net_init tcpv6_net_init(struct net *net)
1928 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1929 SOCK_RAW, IPPROTO_TCP, net);
1932 static void __net_exit tcpv6_net_exit(struct net *net)
1934 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1937 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1939 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1942 static struct pernet_operations tcpv6_net_ops = {
1943 .init = tcpv6_net_init,
1944 .exit = tcpv6_net_exit,
1945 .exit_batch = tcpv6_net_exit_batch,
1948 int __init tcpv6_init(void)
1952 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1956 /* register inet6 protocol */
1957 ret = inet6_register_protosw(&tcpv6_protosw);
1959 goto out_tcpv6_protocol;
1961 ret = register_pernet_subsys(&tcpv6_net_ops);
1963 goto out_tcpv6_protosw;
1968 inet6_unregister_protosw(&tcpv6_protosw);
1970 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1974 void tcpv6_exit(void)
1976 unregister_pernet_subsys(&tcpv6_net_ops);
1977 inet6_unregister_protosw(&tcpv6_protosw);
1978 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);