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/tcp_memcontrol.h>
65 #include <net/busy_poll.h>
67 #include <linux/proc_fs.h>
68 #include <linux/seq_file.h>
70 #include <linux/crypto.h>
71 #include <linux/scatterlist.h>
73 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
74 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
75 struct request_sock *req);
77 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
79 static const struct inet_connection_sock_af_ops ipv6_mapped;
80 static const struct inet_connection_sock_af_ops ipv6_specific;
81 #ifdef CONFIG_TCP_MD5SIG
82 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
83 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
85 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
86 const struct in6_addr *addr)
92 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
94 struct dst_entry *dst = skb_dst(skb);
96 if (dst && dst_hold_safe(dst)) {
97 const struct rt6_info *rt = (const struct rt6_info *)dst;
100 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
101 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
105 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
107 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
108 ipv6_hdr(skb)->saddr.s6_addr32,
110 tcp_hdr(skb)->source);
113 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
116 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
117 struct inet_sock *inet = inet_sk(sk);
118 struct inet_connection_sock *icsk = inet_csk(sk);
119 struct ipv6_pinfo *np = inet6_sk(sk);
120 struct tcp_sock *tp = tcp_sk(sk);
121 struct in6_addr *saddr = NULL, *final_p, final;
122 struct ipv6_txoptions *opt;
124 struct dst_entry *dst;
128 if (addr_len < SIN6_LEN_RFC2133)
131 if (usin->sin6_family != AF_INET6)
132 return -EAFNOSUPPORT;
134 memset(&fl6, 0, sizeof(fl6));
137 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
138 IP6_ECN_flow_init(fl6.flowlabel);
139 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
140 struct ip6_flowlabel *flowlabel;
141 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
144 fl6_sock_release(flowlabel);
149 * connect() to INADDR_ANY means loopback (BSD'ism).
152 if (ipv6_addr_any(&usin->sin6_addr))
153 usin->sin6_addr.s6_addr[15] = 0x1;
155 addr_type = ipv6_addr_type(&usin->sin6_addr);
157 if (addr_type & IPV6_ADDR_MULTICAST)
160 if (addr_type&IPV6_ADDR_LINKLOCAL) {
161 if (addr_len >= sizeof(struct sockaddr_in6) &&
162 usin->sin6_scope_id) {
163 /* If interface is set while binding, indices
166 if (sk->sk_bound_dev_if &&
167 sk->sk_bound_dev_if != usin->sin6_scope_id)
170 sk->sk_bound_dev_if = usin->sin6_scope_id;
173 /* Connect to link-local address requires an interface */
174 if (!sk->sk_bound_dev_if)
178 if (tp->rx_opt.ts_recent_stamp &&
179 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
180 tp->rx_opt.ts_recent = 0;
181 tp->rx_opt.ts_recent_stamp = 0;
185 sk->sk_v6_daddr = usin->sin6_addr;
186 np->flow_label = fl6.flowlabel;
192 if (addr_type == IPV6_ADDR_MAPPED) {
193 u32 exthdrlen = icsk->icsk_ext_hdr_len;
194 struct sockaddr_in sin;
196 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
198 if (__ipv6_only_sock(sk))
201 sin.sin_family = AF_INET;
202 sin.sin_port = usin->sin6_port;
203 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
205 icsk->icsk_af_ops = &ipv6_mapped;
206 sk->sk_backlog_rcv = tcp_v4_do_rcv;
207 #ifdef CONFIG_TCP_MD5SIG
208 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
211 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
214 icsk->icsk_ext_hdr_len = exthdrlen;
215 icsk->icsk_af_ops = &ipv6_specific;
216 sk->sk_backlog_rcv = tcp_v6_do_rcv;
217 #ifdef CONFIG_TCP_MD5SIG
218 tp->af_specific = &tcp_sock_ipv6_specific;
222 np->saddr = sk->sk_v6_rcv_saddr;
227 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
228 saddr = &sk->sk_v6_rcv_saddr;
230 fl6.flowi6_proto = IPPROTO_TCP;
231 fl6.daddr = sk->sk_v6_daddr;
232 fl6.saddr = saddr ? *saddr : np->saddr;
233 fl6.flowi6_oif = sk->sk_bound_dev_if;
234 fl6.flowi6_mark = sk->sk_mark;
235 fl6.fl6_dport = usin->sin6_port;
236 fl6.fl6_sport = inet->inet_sport;
237 fl6.flowi6_uid = sock_i_uid(sk);
239 opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
240 final_p = fl6_update_dst(&fl6, opt, &final);
242 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
244 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
252 sk->sk_v6_rcv_saddr = *saddr;
255 /* set the source address */
257 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
259 sk->sk_gso_type = SKB_GSO_TCPV6;
260 ip6_dst_store(sk, dst, NULL, NULL);
262 if (tcp_death_row.sysctl_tw_recycle &&
263 !tp->rx_opt.ts_recent_stamp &&
264 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
265 tcp_fetch_timewait_stamp(sk, dst);
267 icsk->icsk_ext_hdr_len = 0;
269 icsk->icsk_ext_hdr_len = opt->opt_flen +
272 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
274 inet->inet_dport = usin->sin6_port;
276 tcp_set_state(sk, TCP_SYN_SENT);
277 err = inet6_hash_connect(&tcp_death_row, sk);
283 if (!tp->write_seq && likely(!tp->repair))
284 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
285 sk->sk_v6_daddr.s6_addr32,
289 err = tcp_connect(sk);
296 tcp_set_state(sk, TCP_CLOSE);
299 inet->inet_dport = 0;
300 sk->sk_route_caps = 0;
304 static void tcp_v6_mtu_reduced(struct sock *sk)
306 struct dst_entry *dst;
308 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
311 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
315 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
316 tcp_sync_mss(sk, dst_mtu(dst));
317 tcp_simple_retransmit(sk);
321 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
322 u8 type, u8 code, int offset, __be32 info)
324 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
325 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
326 struct net *net = dev_net(skb->dev);
327 struct request_sock *fastopen;
328 struct ipv6_pinfo *np;
335 sk = __inet6_lookup_established(net, &tcp_hashinfo,
336 &hdr->daddr, th->dest,
337 &hdr->saddr, ntohs(th->source),
341 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
346 if (sk->sk_state == TCP_TIME_WAIT) {
347 inet_twsk_put(inet_twsk(sk));
350 seq = ntohl(th->seq);
351 fatal = icmpv6_err_convert(type, code, &err);
352 if (sk->sk_state == TCP_NEW_SYN_RECV)
353 return tcp_req_err(sk, seq, fatal);
356 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
357 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
359 if (sk->sk_state == TCP_CLOSE)
362 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
363 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
368 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
369 fastopen = tp->fastopen_rsk;
370 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
371 if (sk->sk_state != TCP_LISTEN &&
372 !between(seq, snd_una, tp->snd_nxt)) {
373 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
379 if (type == NDISC_REDIRECT) {
380 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
383 dst->ops->redirect(dst, sk, skb);
387 if (type == ICMPV6_PKT_TOOBIG) {
388 /* We are not interested in TCP_LISTEN and open_requests
389 * (SYN-ACKs send out by Linux are always <576bytes so
390 * they should go through unfragmented).
392 if (sk->sk_state == TCP_LISTEN)
395 if (!ip6_sk_accept_pmtu(sk))
398 tp->mtu_info = ntohl(info);
399 if (!sock_owned_by_user(sk))
400 tcp_v6_mtu_reduced(sk);
401 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
408 /* Might be for an request_sock */
409 switch (sk->sk_state) {
412 /* Only in fast or simultaneous open. If a fast open socket is
413 * is already accepted it is treated as a connected one below.
415 if (fastopen && !fastopen->sk)
418 if (!sock_owned_by_user(sk)) {
420 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
424 sk->sk_err_soft = err;
428 if (!sock_owned_by_user(sk) && np->recverr) {
430 sk->sk_error_report(sk);
432 sk->sk_err_soft = err;
440 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
442 struct request_sock *req,
443 struct tcp_fastopen_cookie *foc,
446 struct inet_request_sock *ireq = inet_rsk(req);
447 struct ipv6_pinfo *np = inet6_sk(sk);
448 struct flowi6 *fl6 = &fl->u.ip6;
452 /* First, grab a route. */
453 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
454 IPPROTO_TCP)) == NULL)
457 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
460 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
461 &ireq->ir_v6_rmt_addr);
463 fl6->daddr = ireq->ir_v6_rmt_addr;
464 if (np->repflow && ireq->pktopts)
465 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
468 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
471 err = net_xmit_eval(err);
479 static void tcp_v6_reqsk_destructor(struct request_sock *req)
481 kfree_skb(inet_rsk(req)->pktopts);
484 #ifdef CONFIG_TCP_MD5SIG
485 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
486 const struct in6_addr *addr)
488 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
491 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
492 const struct sock *addr_sk)
494 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
497 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
500 struct tcp_md5sig cmd;
501 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
503 if (optlen < sizeof(cmd))
506 if (copy_from_user(&cmd, optval, sizeof(cmd)))
509 if (sin6->sin6_family != AF_INET6)
512 if (!cmd.tcpm_keylen) {
513 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
514 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
516 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
520 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
523 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
524 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
525 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
527 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
528 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
531 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
532 const struct in6_addr *daddr,
533 const struct in6_addr *saddr, int nbytes)
535 struct tcp6_pseudohdr *bp;
536 struct scatterlist sg;
538 bp = &hp->md5_blk.ip6;
539 /* 1. TCP pseudo-header (RFC2460) */
542 bp->protocol = cpu_to_be32(IPPROTO_TCP);
543 bp->len = cpu_to_be32(nbytes);
545 sg_init_one(&sg, bp, sizeof(*bp));
546 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
549 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
550 const struct in6_addr *daddr, struct in6_addr *saddr,
551 const struct tcphdr *th)
553 struct tcp_md5sig_pool *hp;
554 struct hash_desc *desc;
556 hp = tcp_get_md5sig_pool();
558 goto clear_hash_noput;
559 desc = &hp->md5_desc;
561 if (crypto_hash_init(desc))
563 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
565 if (tcp_md5_hash_header(hp, th))
567 if (tcp_md5_hash_key(hp, key))
569 if (crypto_hash_final(desc, md5_hash))
572 tcp_put_md5sig_pool();
576 tcp_put_md5sig_pool();
578 memset(md5_hash, 0, 16);
582 static int tcp_v6_md5_hash_skb(char *md5_hash,
583 const struct tcp_md5sig_key *key,
584 const struct sock *sk,
585 const struct sk_buff *skb)
587 const struct in6_addr *saddr, *daddr;
588 struct tcp_md5sig_pool *hp;
589 struct hash_desc *desc;
590 const struct tcphdr *th = tcp_hdr(skb);
592 if (sk) { /* valid for establish/request sockets */
593 saddr = &sk->sk_v6_rcv_saddr;
594 daddr = &sk->sk_v6_daddr;
596 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
597 saddr = &ip6h->saddr;
598 daddr = &ip6h->daddr;
601 hp = tcp_get_md5sig_pool();
603 goto clear_hash_noput;
604 desc = &hp->md5_desc;
606 if (crypto_hash_init(desc))
609 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
611 if (tcp_md5_hash_header(hp, th))
613 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
615 if (tcp_md5_hash_key(hp, key))
617 if (crypto_hash_final(desc, md5_hash))
620 tcp_put_md5sig_pool();
624 tcp_put_md5sig_pool();
626 memset(md5_hash, 0, 16);
632 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
633 const struct sk_buff *skb)
635 #ifdef CONFIG_TCP_MD5SIG
636 const __u8 *hash_location = NULL;
637 struct tcp_md5sig_key *hash_expected;
638 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
639 const struct tcphdr *th = tcp_hdr(skb);
643 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
644 hash_location = tcp_parse_md5sig_option(th);
646 /* We've parsed the options - do we have a hash? */
647 if (!hash_expected && !hash_location)
650 if (hash_expected && !hash_location) {
651 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
655 if (!hash_expected && hash_location) {
656 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
660 /* check the signature */
661 genhash = tcp_v6_md5_hash_skb(newhash,
665 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
666 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
667 genhash ? "failed" : "mismatch",
668 &ip6h->saddr, ntohs(th->source),
669 &ip6h->daddr, ntohs(th->dest));
676 static void tcp_v6_init_req(struct request_sock *req,
677 const struct sock *sk_listener,
680 struct inet_request_sock *ireq = inet_rsk(req);
681 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
683 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
684 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
686 /* So that link locals have meaning */
687 if (!sk_listener->sk_bound_dev_if &&
688 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
689 ireq->ir_iif = tcp_v6_iif(skb);
691 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
692 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
693 np->rxopt.bits.rxinfo ||
694 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
695 np->rxopt.bits.rxohlim || np->repflow)) {
696 atomic_inc(&skb->users);
701 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
703 const struct request_sock *req,
708 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
711 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
713 .obj_size = sizeof(struct tcp6_request_sock),
714 .rtx_syn_ack = tcp_rtx_synack,
715 .send_ack = tcp_v6_reqsk_send_ack,
716 .destructor = tcp_v6_reqsk_destructor,
717 .send_reset = tcp_v6_send_reset,
718 .syn_ack_timeout = tcp_syn_ack_timeout,
721 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
722 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
723 sizeof(struct ipv6hdr),
724 #ifdef CONFIG_TCP_MD5SIG
725 .req_md5_lookup = tcp_v6_md5_lookup,
726 .calc_md5_hash = tcp_v6_md5_hash_skb,
728 .init_req = tcp_v6_init_req,
729 #ifdef CONFIG_SYN_COOKIES
730 .cookie_init_seq = cookie_v6_init_sequence,
732 .route_req = tcp_v6_route_req,
733 .init_seq = tcp_v6_init_sequence,
734 .send_synack = tcp_v6_send_synack,
737 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
738 u32 ack, u32 win, u32 tsval, u32 tsecr,
739 int oif, struct tcp_md5sig_key *key, int rst,
740 u8 tclass, u32 label)
742 const struct tcphdr *th = tcp_hdr(skb);
744 struct sk_buff *buff;
746 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
747 struct sock *ctl_sk = net->ipv6.tcp_sk;
748 unsigned int tot_len = sizeof(struct tcphdr);
749 struct dst_entry *dst;
753 tot_len += TCPOLEN_TSTAMP_ALIGNED;
754 #ifdef CONFIG_TCP_MD5SIG
756 tot_len += TCPOLEN_MD5SIG_ALIGNED;
759 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
764 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
766 t1 = (struct tcphdr *) skb_push(buff, tot_len);
767 skb_reset_transport_header(buff);
769 /* Swap the send and the receive. */
770 memset(t1, 0, sizeof(*t1));
771 t1->dest = th->source;
772 t1->source = th->dest;
773 t1->doff = tot_len / 4;
774 t1->seq = htonl(seq);
775 t1->ack_seq = htonl(ack);
776 t1->ack = !rst || !th->ack;
778 t1->window = htons(win);
780 topt = (__be32 *)(t1 + 1);
783 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
784 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
785 *topt++ = htonl(tsval);
786 *topt++ = htonl(tsecr);
789 #ifdef CONFIG_TCP_MD5SIG
791 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
792 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
793 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
794 &ipv6_hdr(skb)->saddr,
795 &ipv6_hdr(skb)->daddr, t1);
799 memset(&fl6, 0, sizeof(fl6));
800 fl6.daddr = ipv6_hdr(skb)->saddr;
801 fl6.saddr = ipv6_hdr(skb)->daddr;
802 fl6.flowlabel = label;
804 buff->ip_summed = CHECKSUM_PARTIAL;
807 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
809 fl6.flowi6_proto = IPPROTO_TCP;
810 if (rt6_need_strict(&fl6.daddr) && !oif)
811 fl6.flowi6_oif = tcp_v6_iif(skb);
813 fl6.flowi6_oif = oif;
814 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
815 fl6.fl6_dport = t1->dest;
816 fl6.fl6_sport = t1->source;
817 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
819 /* Pass a socket to ip6_dst_lookup either it is for RST
820 * Underlying function will use this to retrieve the network
823 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
825 skb_dst_set(buff, dst);
826 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
827 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
829 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
836 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
838 const struct tcphdr *th = tcp_hdr(skb);
839 u32 seq = 0, ack_seq = 0;
840 struct tcp_md5sig_key *key = NULL;
841 #ifdef CONFIG_TCP_MD5SIG
842 const __u8 *hash_location = NULL;
843 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
844 unsigned char newhash[16];
846 struct sock *sk1 = NULL;
853 /* If sk not NULL, it means we did a successful lookup and incoming
854 * route had to be correct. prequeue might have dropped our dst.
856 if (!sk && !ipv6_unicast_destination(skb))
859 #ifdef CONFIG_TCP_MD5SIG
860 hash_location = tcp_parse_md5sig_option(th);
861 if (!sk && hash_location) {
863 * active side is lost. Try to find listening socket through
864 * source port, and then find md5 key through listening socket.
865 * we are not loose security here:
866 * Incoming packet is checked with md5 hash with finding key,
867 * no RST generated if md5 hash doesn't match.
869 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
870 &tcp_hashinfo, &ipv6h->saddr,
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)
885 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
890 seq = ntohl(th->ack_seq);
892 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
895 oif = sk ? sk->sk_bound_dev_if : 0;
896 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
898 #ifdef CONFIG_TCP_MD5SIG
907 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
908 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
909 struct tcp_md5sig_key *key, u8 tclass,
912 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
916 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
918 struct inet_timewait_sock *tw = inet_twsk(sk);
919 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
921 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
922 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
923 tcp_time_stamp + tcptw->tw_ts_offset,
924 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
925 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
930 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
931 struct request_sock *req)
933 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
934 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
937 * The window field (SEG.WND) of every outgoing segment, with the
938 * exception of <SYN> segments, MUST be right-shifted by
939 * Rcv.Wind.Shift bits:
941 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
942 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
943 tcp_rsk(req)->rcv_nxt,
944 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
945 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
946 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
951 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
953 #ifdef CONFIG_SYN_COOKIES
954 const struct tcphdr *th = tcp_hdr(skb);
957 sk = cookie_v6_check(sk, skb);
962 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
964 if (skb->protocol == htons(ETH_P_IP))
965 return tcp_v4_conn_request(sk, skb);
967 if (!ipv6_unicast_destination(skb))
970 return tcp_conn_request(&tcp6_request_sock_ops,
971 &tcp_request_sock_ipv6_ops, sk, skb);
974 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
975 return 0; /* don't send reset */
978 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
979 struct request_sock *req,
980 struct dst_entry *dst,
981 struct request_sock *req_unhash,
984 struct inet_request_sock *ireq;
985 struct ipv6_pinfo *newnp;
986 const struct ipv6_pinfo *np = inet6_sk(sk);
987 struct ipv6_txoptions *opt;
988 struct tcp6_sock *newtcp6sk;
989 struct inet_sock *newinet;
990 struct tcp_sock *newtp;
992 #ifdef CONFIG_TCP_MD5SIG
993 struct tcp_md5sig_key *key;
997 if (skb->protocol == htons(ETH_P_IP)) {
1002 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1003 req_unhash, own_req);
1008 newtcp6sk = (struct tcp6_sock *)newsk;
1009 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1011 newinet = inet_sk(newsk);
1012 newnp = inet6_sk(newsk);
1013 newtp = tcp_sk(newsk);
1015 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1017 newnp->saddr = newsk->sk_v6_rcv_saddr;
1019 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1020 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1021 #ifdef CONFIG_TCP_MD5SIG
1022 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1025 newnp->ipv6_ac_list = NULL;
1026 newnp->ipv6_fl_list = NULL;
1027 newnp->pktoptions = NULL;
1029 newnp->mcast_oif = tcp_v6_iif(skb);
1030 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1031 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1033 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1036 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1037 * here, tcp_create_openreq_child now does this for us, see the comment in
1038 * that function for the gory details. -acme
1041 /* It is tricky place. Until this moment IPv4 tcp
1042 worked with IPv6 icsk.icsk_af_ops.
1045 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1050 ireq = inet_rsk(req);
1052 if (sk_acceptq_is_full(sk))
1056 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1061 newsk = tcp_create_openreq_child(sk, req, skb);
1066 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1067 * count here, tcp_create_openreq_child now does this for us, see the
1068 * comment in that function for the gory details. -acme
1071 newsk->sk_gso_type = SKB_GSO_TCPV6;
1072 ip6_dst_store(newsk, dst, NULL, NULL);
1073 inet6_sk_rx_dst_set(newsk, skb);
1075 newtcp6sk = (struct tcp6_sock *)newsk;
1076 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1078 newtp = tcp_sk(newsk);
1079 newinet = inet_sk(newsk);
1080 newnp = inet6_sk(newsk);
1082 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1084 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1085 newnp->saddr = ireq->ir_v6_loc_addr;
1086 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1087 newsk->sk_bound_dev_if = ireq->ir_iif;
1089 /* Now IPv6 options...
1091 First: no IPv4 options.
1093 newinet->inet_opt = NULL;
1094 newnp->ipv6_ac_list = NULL;
1095 newnp->ipv6_fl_list = NULL;
1098 newnp->rxopt.all = np->rxopt.all;
1100 newnp->pktoptions = NULL;
1102 newnp->mcast_oif = tcp_v6_iif(skb);
1103 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1104 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1106 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1108 /* Clone native IPv6 options from listening socket (if any)
1110 Yes, keeping reference count would be much more clever,
1111 but we make one more one thing there: reattach optmem
1114 opt = rcu_dereference(np->opt);
1116 opt = ipv6_dup_options(newsk, opt);
1117 RCU_INIT_POINTER(newnp->opt, opt);
1119 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1121 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1124 tcp_ca_openreq_child(newsk, dst);
1126 tcp_sync_mss(newsk, dst_mtu(dst));
1127 newtp->advmss = dst_metric_advmss(dst);
1128 if (tcp_sk(sk)->rx_opt.user_mss &&
1129 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1130 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1132 tcp_initialize_rcv_mss(newsk);
1134 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1135 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1137 #ifdef CONFIG_TCP_MD5SIG
1138 /* Copy over the MD5 key from the original socket */
1139 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1141 /* We're using one, so create a matching key
1142 * on the newsk structure. If we fail to get
1143 * memory, then we end up not copying the key
1146 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1147 AF_INET6, key->key, key->keylen,
1148 sk_gfp_atomic(sk, GFP_ATOMIC));
1152 if (__inet_inherit_port(sk, newsk) < 0) {
1153 inet_csk_prepare_forced_close(newsk);
1157 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1159 tcp_move_syn(newtp, req);
1161 /* Clone pktoptions received with SYN, if we own the req */
1162 if (ireq->pktopts) {
1163 newnp->pktoptions = skb_clone(ireq->pktopts,
1164 sk_gfp_atomic(sk, GFP_ATOMIC));
1165 consume_skb(ireq->pktopts);
1166 ireq->pktopts = NULL;
1167 if (newnp->pktoptions)
1168 skb_set_owner_r(newnp->pktoptions, newsk);
1175 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1179 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1183 static void tcp_v6_restore_cb(struct sk_buff *skb)
1185 /* We need to move header back to the beginning if xfrm6_policy_check()
1186 * and tcp_v6_fill_cb() are going to be called again.
1187 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1189 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1190 sizeof(struct inet6_skb_parm));
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_atomic(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_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1297 TCP_INC_STATS_BH(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 tcp_v6_restore_cb(opt_skb);
1323 opt_skb = xchg(&np->pktoptions, opt_skb);
1325 __kfree_skb(opt_skb);
1326 opt_skb = xchg(&np->pktoptions, NULL);
1334 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1335 const struct tcphdr *th)
1337 /* This is tricky: we move IP6CB at its correct location into
1338 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1339 * _decode_session6() uses IP6CB().
1340 * barrier() makes sure compiler won't play aliasing games.
1342 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1343 sizeof(struct inet6_skb_parm));
1346 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1347 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1348 skb->len - th->doff*4);
1349 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1350 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1351 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1352 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1353 TCP_SKB_CB(skb)->sacked = 0;
1356 static int tcp_v6_rcv(struct sk_buff *skb)
1358 const struct tcphdr *th;
1359 const struct ipv6hdr *hdr;
1362 struct net *net = dev_net(skb->dev);
1364 if (skb->pkt_type != PACKET_HOST)
1368 * Count it even if it's bad.
1370 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1372 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1377 if (th->doff < sizeof(struct tcphdr)/4)
1379 if (!pskb_may_pull(skb, th->doff*4))
1382 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1386 hdr = ipv6_hdr(skb);
1389 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest,
1395 if (sk->sk_state == TCP_TIME_WAIT)
1398 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1399 struct request_sock *req = inet_reqsk(sk);
1402 sk = req->rsk_listener;
1403 tcp_v6_fill_cb(skb, hdr, th);
1404 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1408 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1409 inet_csk_reqsk_queue_drop_and_put(sk, req);
1413 nsk = tcp_check_req(sk, skb, req, false);
1416 goto discard_and_relse;
1420 tcp_v6_restore_cb(skb);
1421 } else if (tcp_child_process(sk, nsk, skb)) {
1422 tcp_v6_send_reset(nsk, skb);
1423 goto discard_and_relse;
1429 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1430 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1431 goto discard_and_relse;
1434 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1435 goto discard_and_relse;
1437 tcp_v6_fill_cb(skb, hdr, th);
1439 if (tcp_v6_inbound_md5_hash(sk, skb))
1440 goto discard_and_relse;
1442 if (sk_filter(sk, skb))
1443 goto discard_and_relse;
1447 if (sk->sk_state == TCP_LISTEN) {
1448 ret = tcp_v6_do_rcv(sk, skb);
1449 goto put_and_return;
1452 sk_incoming_cpu_update(sk);
1454 bh_lock_sock_nested(sk);
1455 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1457 if (!sock_owned_by_user(sk)) {
1458 if (!tcp_prequeue(sk, skb))
1459 ret = tcp_v6_do_rcv(sk, skb);
1460 } else if (unlikely(sk_add_backlog(sk, skb,
1461 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1463 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1464 goto discard_and_relse;
1470 return ret ? -1 : 0;
1473 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1476 tcp_v6_fill_cb(skb, hdr, th);
1478 if (tcp_checksum_complete(skb)) {
1480 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1482 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1484 tcp_v6_send_reset(NULL, skb);
1496 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1497 inet_twsk_put(inet_twsk(sk));
1501 tcp_v6_fill_cb(skb, hdr, th);
1503 if (tcp_checksum_complete(skb)) {
1504 inet_twsk_put(inet_twsk(sk));
1508 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1513 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1514 &ipv6_hdr(skb)->saddr, th->source,
1515 &ipv6_hdr(skb)->daddr,
1516 ntohs(th->dest), tcp_v6_iif(skb));
1518 struct inet_timewait_sock *tw = inet_twsk(sk);
1519 inet_twsk_deschedule_put(tw);
1521 tcp_v6_restore_cb(skb);
1524 /* Fall through to ACK */
1527 tcp_v6_timewait_ack(sk, skb);
1530 tcp_v6_restore_cb(skb);
1532 case TCP_TW_SUCCESS:
1538 static void tcp_v6_early_demux(struct sk_buff *skb)
1540 const struct ipv6hdr *hdr;
1541 const struct tcphdr *th;
1544 if (skb->pkt_type != PACKET_HOST)
1547 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1550 hdr = ipv6_hdr(skb);
1553 if (th->doff < sizeof(struct tcphdr) / 4)
1556 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1557 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1558 &hdr->saddr, th->source,
1559 &hdr->daddr, ntohs(th->dest),
1563 skb->destructor = sock_edemux;
1564 if (sk_fullsock(sk)) {
1565 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1568 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1570 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1571 skb_dst_set_noref(skb, dst);
1576 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1577 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1578 .twsk_unique = tcp_twsk_unique,
1579 .twsk_destructor = tcp_twsk_destructor,
1582 static const struct inet_connection_sock_af_ops ipv6_specific = {
1583 .queue_xmit = inet6_csk_xmit,
1584 .send_check = tcp_v6_send_check,
1585 .rebuild_header = inet6_sk_rebuild_header,
1586 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1587 .conn_request = tcp_v6_conn_request,
1588 .syn_recv_sock = tcp_v6_syn_recv_sock,
1589 .net_header_len = sizeof(struct ipv6hdr),
1590 .net_frag_header_len = sizeof(struct frag_hdr),
1591 .setsockopt = ipv6_setsockopt,
1592 .getsockopt = ipv6_getsockopt,
1593 .addr2sockaddr = inet6_csk_addr2sockaddr,
1594 .sockaddr_len = sizeof(struct sockaddr_in6),
1595 .bind_conflict = inet6_csk_bind_conflict,
1596 #ifdef CONFIG_COMPAT
1597 .compat_setsockopt = compat_ipv6_setsockopt,
1598 .compat_getsockopt = compat_ipv6_getsockopt,
1600 .mtu_reduced = tcp_v6_mtu_reduced,
1603 #ifdef CONFIG_TCP_MD5SIG
1604 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1605 .md5_lookup = tcp_v6_md5_lookup,
1606 .calc_md5_hash = tcp_v6_md5_hash_skb,
1607 .md5_parse = tcp_v6_parse_md5_keys,
1612 * TCP over IPv4 via INET6 API
1614 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1615 .queue_xmit = ip_queue_xmit,
1616 .send_check = tcp_v4_send_check,
1617 .rebuild_header = inet_sk_rebuild_header,
1618 .sk_rx_dst_set = inet_sk_rx_dst_set,
1619 .conn_request = tcp_v6_conn_request,
1620 .syn_recv_sock = tcp_v6_syn_recv_sock,
1621 .net_header_len = sizeof(struct iphdr),
1622 .setsockopt = ipv6_setsockopt,
1623 .getsockopt = ipv6_getsockopt,
1624 .addr2sockaddr = inet6_csk_addr2sockaddr,
1625 .sockaddr_len = sizeof(struct sockaddr_in6),
1626 .bind_conflict = inet6_csk_bind_conflict,
1627 #ifdef CONFIG_COMPAT
1628 .compat_setsockopt = compat_ipv6_setsockopt,
1629 .compat_getsockopt = compat_ipv6_getsockopt,
1631 .mtu_reduced = tcp_v4_mtu_reduced,
1634 #ifdef CONFIG_TCP_MD5SIG
1635 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1636 .md5_lookup = tcp_v4_md5_lookup,
1637 .calc_md5_hash = tcp_v4_md5_hash_skb,
1638 .md5_parse = tcp_v6_parse_md5_keys,
1642 /* NOTE: A lot of things set to zero explicitly by call to
1643 * sk_alloc() so need not be done here.
1645 static int tcp_v6_init_sock(struct sock *sk)
1647 struct inet_connection_sock *icsk = inet_csk(sk);
1651 icsk->icsk_af_ops = &ipv6_specific;
1653 #ifdef CONFIG_TCP_MD5SIG
1654 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1660 static void tcp_v6_destroy_sock(struct sock *sk)
1662 tcp_v4_destroy_sock(sk);
1663 inet6_destroy_sock(sk);
1666 #ifdef CONFIG_PROC_FS
1667 /* Proc filesystem TCPv6 sock list dumping. */
1668 static void get_openreq6(struct seq_file *seq,
1669 const struct request_sock *req, int i)
1671 long ttd = req->rsk_timer.expires - jiffies;
1672 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1673 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1679 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1680 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1682 src->s6_addr32[0], src->s6_addr32[1],
1683 src->s6_addr32[2], src->s6_addr32[3],
1684 inet_rsk(req)->ir_num,
1685 dest->s6_addr32[0], dest->s6_addr32[1],
1686 dest->s6_addr32[2], dest->s6_addr32[3],
1687 ntohs(inet_rsk(req)->ir_rmt_port),
1689 0, 0, /* could print option size, but that is af dependent. */
1690 1, /* timers active (only the expire timer) */
1691 jiffies_to_clock_t(ttd),
1693 from_kuid_munged(seq_user_ns(seq),
1694 sock_i_uid(req->rsk_listener)),
1695 0, /* non standard timer */
1696 0, /* open_requests have no inode */
1700 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1702 const struct in6_addr *dest, *src;
1705 unsigned long timer_expires;
1706 const struct inet_sock *inet = inet_sk(sp);
1707 const struct tcp_sock *tp = tcp_sk(sp);
1708 const struct inet_connection_sock *icsk = inet_csk(sp);
1709 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1713 dest = &sp->sk_v6_daddr;
1714 src = &sp->sk_v6_rcv_saddr;
1715 destp = ntohs(inet->inet_dport);
1716 srcp = ntohs(inet->inet_sport);
1718 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1719 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
1720 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1722 timer_expires = icsk->icsk_timeout;
1723 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1725 timer_expires = icsk->icsk_timeout;
1726 } else if (timer_pending(&sp->sk_timer)) {
1728 timer_expires = sp->sk_timer.expires;
1731 timer_expires = jiffies;
1734 state = sk_state_load(sp);
1735 if (state == TCP_LISTEN)
1736 rx_queue = sp->sk_ack_backlog;
1738 /* Because we don't lock the socket,
1739 * we might find a transient negative value.
1741 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1744 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1745 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1747 src->s6_addr32[0], src->s6_addr32[1],
1748 src->s6_addr32[2], src->s6_addr32[3], srcp,
1749 dest->s6_addr32[0], dest->s6_addr32[1],
1750 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1752 tp->write_seq - tp->snd_una,
1755 jiffies_delta_to_clock_t(timer_expires - jiffies),
1756 icsk->icsk_retransmits,
1757 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1758 icsk->icsk_probes_out,
1760 atomic_read(&sp->sk_refcnt), sp,
1761 jiffies_to_clock_t(icsk->icsk_rto),
1762 jiffies_to_clock_t(icsk->icsk_ack.ato),
1763 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1765 state == TCP_LISTEN ?
1766 fastopenq->max_qlen :
1767 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1771 static void get_timewait6_sock(struct seq_file *seq,
1772 struct inet_timewait_sock *tw, int i)
1774 long delta = tw->tw_timer.expires - jiffies;
1775 const struct in6_addr *dest, *src;
1778 dest = &tw->tw_v6_daddr;
1779 src = &tw->tw_v6_rcv_saddr;
1780 destp = ntohs(tw->tw_dport);
1781 srcp = ntohs(tw->tw_sport);
1784 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1785 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1787 src->s6_addr32[0], src->s6_addr32[1],
1788 src->s6_addr32[2], src->s6_addr32[3], srcp,
1789 dest->s6_addr32[0], dest->s6_addr32[1],
1790 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1791 tw->tw_substate, 0, 0,
1792 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1793 atomic_read(&tw->tw_refcnt), tw);
1796 static int tcp6_seq_show(struct seq_file *seq, void *v)
1798 struct tcp_iter_state *st;
1799 struct sock *sk = v;
1801 if (v == SEQ_START_TOKEN) {
1806 "st tx_queue rx_queue tr tm->when retrnsmt"
1807 " uid timeout inode\n");
1812 if (sk->sk_state == TCP_TIME_WAIT)
1813 get_timewait6_sock(seq, v, st->num);
1814 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1815 get_openreq6(seq, v, st->num);
1817 get_tcp6_sock(seq, v, st->num);
1822 static const struct file_operations tcp6_afinfo_seq_fops = {
1823 .owner = THIS_MODULE,
1824 .open = tcp_seq_open,
1826 .llseek = seq_lseek,
1827 .release = seq_release_net
1830 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1833 .seq_fops = &tcp6_afinfo_seq_fops,
1835 .show = tcp6_seq_show,
1839 int __net_init tcp6_proc_init(struct net *net)
1841 return tcp_proc_register(net, &tcp6_seq_afinfo);
1844 void tcp6_proc_exit(struct net *net)
1846 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1850 static void tcp_v6_clear_sk(struct sock *sk, int size)
1852 struct inet_sock *inet = inet_sk(sk);
1854 /* we do not want to clear pinet6 field, because of RCU lookups */
1855 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1857 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1858 memset(&inet->pinet6 + 1, 0, size);
1861 struct proto tcpv6_prot = {
1863 .owner = THIS_MODULE,
1865 .connect = tcp_v6_connect,
1866 .disconnect = tcp_disconnect,
1867 .accept = inet_csk_accept,
1869 .init = tcp_v6_init_sock,
1870 .destroy = tcp_v6_destroy_sock,
1871 .shutdown = tcp_shutdown,
1872 .setsockopt = tcp_setsockopt,
1873 .getsockopt = tcp_getsockopt,
1874 .recvmsg = tcp_recvmsg,
1875 .sendmsg = tcp_sendmsg,
1876 .sendpage = tcp_sendpage,
1877 .backlog_rcv = tcp_v6_do_rcv,
1878 .release_cb = tcp_release_cb,
1880 .unhash = inet_unhash,
1881 .get_port = inet_csk_get_port,
1882 .enter_memory_pressure = tcp_enter_memory_pressure,
1883 .stream_memory_free = tcp_stream_memory_free,
1884 .sockets_allocated = &tcp_sockets_allocated,
1885 .memory_allocated = &tcp_memory_allocated,
1886 .memory_pressure = &tcp_memory_pressure,
1887 .orphan_count = &tcp_orphan_count,
1888 .sysctl_mem = sysctl_tcp_mem,
1889 .sysctl_wmem = sysctl_tcp_wmem,
1890 .sysctl_rmem = sysctl_tcp_rmem,
1891 .max_header = MAX_TCP_HEADER,
1892 .obj_size = sizeof(struct tcp6_sock),
1893 .slab_flags = SLAB_DESTROY_BY_RCU,
1894 .twsk_prot = &tcp6_timewait_sock_ops,
1895 .rsk_prot = &tcp6_request_sock_ops,
1896 .h.hashinfo = &tcp_hashinfo,
1897 .no_autobind = true,
1898 #ifdef CONFIG_COMPAT
1899 .compat_setsockopt = compat_tcp_setsockopt,
1900 .compat_getsockopt = compat_tcp_getsockopt,
1902 #ifdef CONFIG_MEMCG_KMEM
1903 .proto_cgroup = tcp_proto_cgroup,
1905 .clear_sk = tcp_v6_clear_sk,
1906 .diag_destroy = tcp_abort,
1909 static const struct inet6_protocol tcpv6_protocol = {
1910 .early_demux = tcp_v6_early_demux,
1911 .handler = tcp_v6_rcv,
1912 .err_handler = tcp_v6_err,
1913 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1916 static struct inet_protosw tcpv6_protosw = {
1917 .type = SOCK_STREAM,
1918 .protocol = IPPROTO_TCP,
1919 .prot = &tcpv6_prot,
1920 .ops = &inet6_stream_ops,
1921 .flags = INET_PROTOSW_PERMANENT |
1925 static int __net_init tcpv6_net_init(struct net *net)
1927 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1928 SOCK_RAW, IPPROTO_TCP, net);
1931 static void __net_exit tcpv6_net_exit(struct net *net)
1933 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1936 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1938 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1941 static struct pernet_operations tcpv6_net_ops = {
1942 .init = tcpv6_net_init,
1943 .exit = tcpv6_net_exit,
1944 .exit_batch = tcpv6_net_exit_batch,
1947 int __init tcpv6_init(void)
1951 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1955 /* register inet6 protocol */
1956 ret = inet6_register_protosw(&tcpv6_protosw);
1958 goto out_tcpv6_protocol;
1960 ret = register_pernet_subsys(&tcpv6_net_ops);
1962 goto out_tcpv6_protosw;
1967 inet6_unregister_protosw(&tcpv6_protosw);
1969 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1973 void tcpv6_exit(void)
1975 unregister_pernet_subsys(&tcpv6_net_ops);
1976 inet6_unregister_protosw(&tcpv6_protosw);
1977 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);