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;
238 opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
239 final_p = fl6_update_dst(&fl6, opt, &final);
241 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
243 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
251 sk->sk_v6_rcv_saddr = *saddr;
254 /* set the source address */
256 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
258 sk->sk_gso_type = SKB_GSO_TCPV6;
259 ip6_dst_store(sk, dst, NULL, NULL);
261 if (tcp_death_row.sysctl_tw_recycle &&
262 !tp->rx_opt.ts_recent_stamp &&
263 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
264 tcp_fetch_timewait_stamp(sk, dst);
266 icsk->icsk_ext_hdr_len = 0;
268 icsk->icsk_ext_hdr_len = opt->opt_flen +
271 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
273 inet->inet_dport = usin->sin6_port;
275 tcp_set_state(sk, TCP_SYN_SENT);
276 err = inet6_hash_connect(&tcp_death_row, sk);
282 if (!tp->write_seq && likely(!tp->repair))
283 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
284 sk->sk_v6_daddr.s6_addr32,
288 err = tcp_connect(sk);
295 tcp_set_state(sk, TCP_CLOSE);
298 inet->inet_dport = 0;
299 sk->sk_route_caps = 0;
303 static void tcp_v6_mtu_reduced(struct sock *sk)
305 struct dst_entry *dst;
307 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
310 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
314 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
315 tcp_sync_mss(sk, dst_mtu(dst));
316 tcp_simple_retransmit(sk);
320 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
321 u8 type, u8 code, int offset, __be32 info)
323 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
324 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
325 struct net *net = dev_net(skb->dev);
326 struct request_sock *fastopen;
327 struct ipv6_pinfo *np;
334 sk = __inet6_lookup_established(net, &tcp_hashinfo,
335 &hdr->daddr, th->dest,
336 &hdr->saddr, ntohs(th->source),
340 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
345 if (sk->sk_state == TCP_TIME_WAIT) {
346 inet_twsk_put(inet_twsk(sk));
349 seq = ntohl(th->seq);
350 fatal = icmpv6_err_convert(type, code, &err);
351 if (sk->sk_state == TCP_NEW_SYN_RECV)
352 return tcp_req_err(sk, seq, fatal);
355 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
356 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
358 if (sk->sk_state == TCP_CLOSE)
361 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
362 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
367 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
368 fastopen = tp->fastopen_rsk;
369 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
370 if (sk->sk_state != TCP_LISTEN &&
371 !between(seq, snd_una, tp->snd_nxt)) {
372 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
378 if (type == NDISC_REDIRECT) {
379 if (!sock_owned_by_user(sk)) {
380 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
383 dst->ops->redirect(dst, sk, skb);
388 if (type == ICMPV6_PKT_TOOBIG) {
389 /* We are not interested in TCP_LISTEN and open_requests
390 * (SYN-ACKs send out by Linux are always <576bytes so
391 * they should go through unfragmented).
393 if (sk->sk_state == TCP_LISTEN)
396 if (!ip6_sk_accept_pmtu(sk))
399 tp->mtu_info = ntohl(info);
400 if (!sock_owned_by_user(sk))
401 tcp_v6_mtu_reduced(sk);
402 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
409 /* Might be for an request_sock */
410 switch (sk->sk_state) {
413 /* Only in fast or simultaneous open. If a fast open socket is
414 * is already accepted it is treated as a connected one below.
416 if (fastopen && !fastopen->sk)
419 if (!sock_owned_by_user(sk)) {
421 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
425 sk->sk_err_soft = err;
429 if (!sock_owned_by_user(sk) && np->recverr) {
431 sk->sk_error_report(sk);
433 sk->sk_err_soft = err;
441 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
443 struct request_sock *req,
444 struct tcp_fastopen_cookie *foc,
447 struct inet_request_sock *ireq = inet_rsk(req);
448 struct ipv6_pinfo *np = inet6_sk(sk);
449 struct flowi6 *fl6 = &fl->u.ip6;
453 /* First, grab a route. */
454 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
455 IPPROTO_TCP)) == NULL)
458 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
461 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
462 &ireq->ir_v6_rmt_addr);
464 fl6->daddr = ireq->ir_v6_rmt_addr;
465 if (np->repflow && ireq->pktopts)
466 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
469 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
472 err = net_xmit_eval(err);
480 static void tcp_v6_reqsk_destructor(struct request_sock *req)
482 kfree_skb(inet_rsk(req)->pktopts);
485 #ifdef CONFIG_TCP_MD5SIG
486 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
487 const struct in6_addr *addr)
489 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
492 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
493 const struct sock *addr_sk)
495 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
498 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
501 struct tcp_md5sig cmd;
502 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
504 if (optlen < sizeof(cmd))
507 if (copy_from_user(&cmd, optval, sizeof(cmd)))
510 if (sin6->sin6_family != AF_INET6)
513 if (!cmd.tcpm_keylen) {
514 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
515 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
517 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
521 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
524 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
525 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
526 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
528 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
529 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
532 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
533 const struct in6_addr *daddr,
534 const struct in6_addr *saddr, int nbytes)
536 struct tcp6_pseudohdr *bp;
537 struct scatterlist sg;
539 bp = &hp->md5_blk.ip6;
540 /* 1. TCP pseudo-header (RFC2460) */
543 bp->protocol = cpu_to_be32(IPPROTO_TCP);
544 bp->len = cpu_to_be32(nbytes);
546 sg_init_one(&sg, bp, sizeof(*bp));
547 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
550 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
551 const struct in6_addr *daddr, struct in6_addr *saddr,
552 const struct tcphdr *th)
554 struct tcp_md5sig_pool *hp;
555 struct hash_desc *desc;
557 hp = tcp_get_md5sig_pool();
559 goto clear_hash_noput;
560 desc = &hp->md5_desc;
562 if (crypto_hash_init(desc))
564 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
566 if (tcp_md5_hash_header(hp, th))
568 if (tcp_md5_hash_key(hp, key))
570 if (crypto_hash_final(desc, md5_hash))
573 tcp_put_md5sig_pool();
577 tcp_put_md5sig_pool();
579 memset(md5_hash, 0, 16);
583 static int tcp_v6_md5_hash_skb(char *md5_hash,
584 const struct tcp_md5sig_key *key,
585 const struct sock *sk,
586 const struct sk_buff *skb)
588 const struct in6_addr *saddr, *daddr;
589 struct tcp_md5sig_pool *hp;
590 struct hash_desc *desc;
591 const struct tcphdr *th = tcp_hdr(skb);
593 if (sk) { /* valid for establish/request sockets */
594 saddr = &sk->sk_v6_rcv_saddr;
595 daddr = &sk->sk_v6_daddr;
597 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
598 saddr = &ip6h->saddr;
599 daddr = &ip6h->daddr;
602 hp = tcp_get_md5sig_pool();
604 goto clear_hash_noput;
605 desc = &hp->md5_desc;
607 if (crypto_hash_init(desc))
610 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
612 if (tcp_md5_hash_header(hp, th))
614 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
616 if (tcp_md5_hash_key(hp, key))
618 if (crypto_hash_final(desc, md5_hash))
621 tcp_put_md5sig_pool();
625 tcp_put_md5sig_pool();
627 memset(md5_hash, 0, 16);
633 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
634 const struct sk_buff *skb)
636 #ifdef CONFIG_TCP_MD5SIG
637 const __u8 *hash_location = NULL;
638 struct tcp_md5sig_key *hash_expected;
639 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
640 const struct tcphdr *th = tcp_hdr(skb);
644 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
645 hash_location = tcp_parse_md5sig_option(th);
647 /* We've parsed the options - do we have a hash? */
648 if (!hash_expected && !hash_location)
651 if (hash_expected && !hash_location) {
652 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
656 if (!hash_expected && hash_location) {
657 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
661 /* check the signature */
662 genhash = tcp_v6_md5_hash_skb(newhash,
666 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
667 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
668 genhash ? "failed" : "mismatch",
669 &ip6h->saddr, ntohs(th->source),
670 &ip6h->daddr, ntohs(th->dest));
677 static void tcp_v6_init_req(struct request_sock *req,
678 const struct sock *sk_listener,
681 struct inet_request_sock *ireq = inet_rsk(req);
682 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
684 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
685 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
687 /* So that link locals have meaning */
688 if (!sk_listener->sk_bound_dev_if &&
689 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
690 ireq->ir_iif = tcp_v6_iif(skb);
692 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
693 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
694 np->rxopt.bits.rxinfo ||
695 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
696 np->rxopt.bits.rxohlim || np->repflow)) {
697 atomic_inc(&skb->users);
702 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
704 const struct request_sock *req,
709 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
712 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
714 .obj_size = sizeof(struct tcp6_request_sock),
715 .rtx_syn_ack = tcp_rtx_synack,
716 .send_ack = tcp_v6_reqsk_send_ack,
717 .destructor = tcp_v6_reqsk_destructor,
718 .send_reset = tcp_v6_send_reset,
719 .syn_ack_timeout = tcp_syn_ack_timeout,
722 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
723 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
724 sizeof(struct ipv6hdr),
725 #ifdef CONFIG_TCP_MD5SIG
726 .req_md5_lookup = tcp_v6_md5_lookup,
727 .calc_md5_hash = tcp_v6_md5_hash_skb,
729 .init_req = tcp_v6_init_req,
730 #ifdef CONFIG_SYN_COOKIES
731 .cookie_init_seq = cookie_v6_init_sequence,
733 .route_req = tcp_v6_route_req,
734 .init_seq = tcp_v6_init_sequence,
735 .send_synack = tcp_v6_send_synack,
738 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
739 u32 ack, u32 win, u32 tsval, u32 tsecr,
740 int oif, struct tcp_md5sig_key *key, int rst,
741 u8 tclass, u32 label)
743 const struct tcphdr *th = tcp_hdr(skb);
745 struct sk_buff *buff;
747 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
748 struct sock *ctl_sk = net->ipv6.tcp_sk;
749 unsigned int tot_len = sizeof(struct tcphdr);
750 struct dst_entry *dst;
754 tot_len += TCPOLEN_TSTAMP_ALIGNED;
755 #ifdef CONFIG_TCP_MD5SIG
757 tot_len += TCPOLEN_MD5SIG_ALIGNED;
760 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
765 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
767 t1 = (struct tcphdr *) skb_push(buff, tot_len);
768 skb_reset_transport_header(buff);
770 /* Swap the send and the receive. */
771 memset(t1, 0, sizeof(*t1));
772 t1->dest = th->source;
773 t1->source = th->dest;
774 t1->doff = tot_len / 4;
775 t1->seq = htonl(seq);
776 t1->ack_seq = htonl(ack);
777 t1->ack = !rst || !th->ack;
779 t1->window = htons(win);
781 topt = (__be32 *)(t1 + 1);
784 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
785 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
786 *topt++ = htonl(tsval);
787 *topt++ = htonl(tsecr);
790 #ifdef CONFIG_TCP_MD5SIG
792 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
793 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
794 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
795 &ipv6_hdr(skb)->saddr,
796 &ipv6_hdr(skb)->daddr, t1);
800 memset(&fl6, 0, sizeof(fl6));
801 fl6.daddr = ipv6_hdr(skb)->saddr;
802 fl6.saddr = ipv6_hdr(skb)->daddr;
803 fl6.flowlabel = label;
805 buff->ip_summed = CHECKSUM_PARTIAL;
808 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
810 fl6.flowi6_proto = IPPROTO_TCP;
811 if (rt6_need_strict(&fl6.daddr) && !oif)
812 fl6.flowi6_oif = tcp_v6_iif(skb);
814 fl6.flowi6_oif = oif;
815 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
816 fl6.fl6_dport = t1->dest;
817 fl6.fl6_sport = t1->source;
818 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
820 /* Pass a socket to ip6_dst_lookup either it is for RST
821 * Underlying function will use this to retrieve the network
824 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
826 skb_dst_set(buff, dst);
827 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
828 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
830 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
837 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
839 const struct tcphdr *th = tcp_hdr(skb);
840 u32 seq = 0, ack_seq = 0;
841 struct tcp_md5sig_key *key = NULL;
842 #ifdef CONFIG_TCP_MD5SIG
843 const __u8 *hash_location = NULL;
844 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
845 unsigned char newhash[16];
847 struct sock *sk1 = NULL;
854 /* If sk not NULL, it means we did a successful lookup and incoming
855 * route had to be correct. prequeue might have dropped our dst.
857 if (!sk && !ipv6_unicast_destination(skb))
860 #ifdef CONFIG_TCP_MD5SIG
861 hash_location = tcp_parse_md5sig_option(th);
862 if (!sk && hash_location) {
864 * active side is lost. Try to find listening socket through
865 * source port, and then find md5 key through listening socket.
866 * we are not loose security here:
867 * Incoming packet is checked with md5 hash with finding key,
868 * no RST generated if md5 hash doesn't match.
870 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
871 &tcp_hashinfo, &ipv6h->saddr,
872 th->source, &ipv6h->daddr,
873 ntohs(th->source), tcp_v6_iif(skb));
878 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
882 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
883 if (genhash || memcmp(hash_location, newhash, 16) != 0)
886 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
891 seq = ntohl(th->ack_seq);
893 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
896 oif = sk ? sk->sk_bound_dev_if : 0;
897 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
899 #ifdef CONFIG_TCP_MD5SIG
908 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
909 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
910 struct tcp_md5sig_key *key, u8 tclass,
913 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
917 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
919 struct inet_timewait_sock *tw = inet_twsk(sk);
920 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
922 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
923 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
924 tcp_time_stamp + tcptw->tw_ts_offset,
925 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
926 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
931 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
932 struct request_sock *req)
934 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
935 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
938 * The window field (SEG.WND) of every outgoing segment, with the
939 * exception of <SYN> segments, MUST be right-shifted by
940 * Rcv.Wind.Shift bits:
942 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
943 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
944 tcp_rsk(req)->rcv_nxt,
945 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
946 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
947 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
952 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
954 #ifdef CONFIG_SYN_COOKIES
955 const struct tcphdr *th = tcp_hdr(skb);
958 sk = cookie_v6_check(sk, skb);
963 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
965 if (skb->protocol == htons(ETH_P_IP))
966 return tcp_v4_conn_request(sk, skb);
968 if (!ipv6_unicast_destination(skb))
971 return tcp_conn_request(&tcp6_request_sock_ops,
972 &tcp_request_sock_ipv6_ops, sk, skb);
975 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
976 return 0; /* don't send reset */
979 static void tcp_v6_restore_cb(struct sk_buff *skb)
981 /* We need to move header back to the beginning if xfrm6_policy_check()
982 * and tcp_v6_fill_cb() are going to be called again.
983 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
985 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
986 sizeof(struct inet6_skb_parm));
989 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
990 struct request_sock *req,
991 struct dst_entry *dst,
992 struct request_sock *req_unhash,
995 struct inet_request_sock *ireq;
996 struct ipv6_pinfo *newnp;
997 const struct ipv6_pinfo *np = inet6_sk(sk);
998 struct ipv6_txoptions *opt;
999 struct tcp6_sock *newtcp6sk;
1000 struct inet_sock *newinet;
1001 struct tcp_sock *newtp;
1003 #ifdef CONFIG_TCP_MD5SIG
1004 struct tcp_md5sig_key *key;
1008 if (skb->protocol == htons(ETH_P_IP)) {
1013 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1014 req_unhash, own_req);
1019 newtcp6sk = (struct tcp6_sock *)newsk;
1020 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1022 newinet = inet_sk(newsk);
1023 newnp = inet6_sk(newsk);
1024 newtp = tcp_sk(newsk);
1026 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1028 newnp->saddr = newsk->sk_v6_rcv_saddr;
1030 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1031 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1032 #ifdef CONFIG_TCP_MD5SIG
1033 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1036 newnp->ipv6_ac_list = NULL;
1037 newnp->ipv6_fl_list = NULL;
1038 newnp->pktoptions = NULL;
1040 newnp->mcast_oif = tcp_v6_iif(skb);
1041 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1042 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1044 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1047 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1048 * here, tcp_create_openreq_child now does this for us, see the comment in
1049 * that function for the gory details. -acme
1052 /* It is tricky place. Until this moment IPv4 tcp
1053 worked with IPv6 icsk.icsk_af_ops.
1056 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1061 ireq = inet_rsk(req);
1063 if (sk_acceptq_is_full(sk))
1067 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1072 newsk = tcp_create_openreq_child(sk, req, skb);
1077 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1078 * count here, tcp_create_openreq_child now does this for us, see the
1079 * comment in that function for the gory details. -acme
1082 newsk->sk_gso_type = SKB_GSO_TCPV6;
1083 ip6_dst_store(newsk, dst, NULL, NULL);
1084 inet6_sk_rx_dst_set(newsk, skb);
1086 newtcp6sk = (struct tcp6_sock *)newsk;
1087 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1089 newtp = tcp_sk(newsk);
1090 newinet = inet_sk(newsk);
1091 newnp = inet6_sk(newsk);
1093 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1095 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1096 newnp->saddr = ireq->ir_v6_loc_addr;
1097 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1098 newsk->sk_bound_dev_if = ireq->ir_iif;
1100 /* Now IPv6 options...
1102 First: no IPv4 options.
1104 newinet->inet_opt = NULL;
1105 newnp->ipv6_ac_list = NULL;
1106 newnp->ipv6_fl_list = NULL;
1109 newnp->rxopt.all = np->rxopt.all;
1111 newnp->pktoptions = NULL;
1113 newnp->mcast_oif = tcp_v6_iif(skb);
1114 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1115 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1117 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1119 /* Clone native IPv6 options from listening socket (if any)
1121 Yes, keeping reference count would be much more clever,
1122 but we make one more one thing there: reattach optmem
1125 opt = rcu_dereference(np->opt);
1127 opt = ipv6_dup_options(newsk, opt);
1128 RCU_INIT_POINTER(newnp->opt, opt);
1130 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1132 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1135 tcp_ca_openreq_child(newsk, dst);
1137 tcp_sync_mss(newsk, dst_mtu(dst));
1138 newtp->advmss = dst_metric_advmss(dst);
1139 if (tcp_sk(sk)->rx_opt.user_mss &&
1140 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1141 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1143 tcp_initialize_rcv_mss(newsk);
1145 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1146 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1148 #ifdef CONFIG_TCP_MD5SIG
1149 /* Copy over the MD5 key from the original socket */
1150 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1152 /* We're using one, so create a matching key
1153 * on the newsk structure. If we fail to get
1154 * memory, then we end up not copying the key
1157 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1158 AF_INET6, key->key, key->keylen,
1159 sk_gfp_atomic(sk, GFP_ATOMIC));
1163 if (__inet_inherit_port(sk, newsk) < 0) {
1164 inet_csk_prepare_forced_close(newsk);
1168 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1170 tcp_move_syn(newtp, req);
1172 /* Clone pktoptions received with SYN, if we own the req */
1173 if (ireq->pktopts) {
1174 newnp->pktoptions = skb_clone(ireq->pktopts,
1175 sk_gfp_atomic(sk, GFP_ATOMIC));
1176 consume_skb(ireq->pktopts);
1177 ireq->pktopts = NULL;
1178 if (newnp->pktoptions) {
1179 tcp_v6_restore_cb(newnp->pktoptions);
1180 skb_set_owner_r(newnp->pktoptions, newsk);
1188 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1192 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1196 /* The socket must have it's spinlock held when we get
1197 * here, unless it is a TCP_LISTEN socket.
1199 * We have a potential double-lock case here, so even when
1200 * doing backlog processing we use the BH locking scheme.
1201 * This is because we cannot sleep with the original spinlock
1204 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1206 struct ipv6_pinfo *np = inet6_sk(sk);
1207 struct tcp_sock *tp;
1208 struct sk_buff *opt_skb = NULL;
1210 /* Imagine: socket is IPv6. IPv4 packet arrives,
1211 goes to IPv4 receive handler and backlogged.
1212 From backlog it always goes here. Kerboom...
1213 Fortunately, tcp_rcv_established and rcv_established
1214 handle them correctly, but it is not case with
1215 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1218 if (skb->protocol == htons(ETH_P_IP))
1219 return tcp_v4_do_rcv(sk, skb);
1221 if (tcp_filter(sk, skb))
1225 * socket locking is here for SMP purposes as backlog rcv
1226 * is currently called with bh processing disabled.
1229 /* Do Stevens' IPV6_PKTOPTIONS.
1231 Yes, guys, it is the only place in our code, where we
1232 may make it not affecting IPv4.
1233 The rest of code is protocol independent,
1234 and I do not like idea to uglify IPv4.
1236 Actually, all the idea behind IPV6_PKTOPTIONS
1237 looks not very well thought. For now we latch
1238 options, received in the last packet, enqueued
1239 by tcp. Feel free to propose better solution.
1243 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1245 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1246 struct dst_entry *dst = sk->sk_rx_dst;
1248 sock_rps_save_rxhash(sk, skb);
1249 sk_mark_napi_id(sk, skb);
1251 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1252 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1254 sk->sk_rx_dst = NULL;
1258 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1260 goto ipv6_pktoptions;
1264 if (tcp_checksum_complete(skb))
1267 if (sk->sk_state == TCP_LISTEN) {
1268 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1274 sock_rps_save_rxhash(nsk, skb);
1275 sk_mark_napi_id(nsk, skb);
1276 if (tcp_child_process(sk, nsk, skb))
1279 __kfree_skb(opt_skb);
1283 sock_rps_save_rxhash(sk, skb);
1285 if (tcp_rcv_state_process(sk, skb))
1288 goto ipv6_pktoptions;
1292 tcp_v6_send_reset(sk, skb);
1295 __kfree_skb(opt_skb);
1299 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1300 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1305 /* Do you ask, what is it?
1307 1. skb was enqueued by tcp.
1308 2. skb is added to tail of read queue, rather than out of order.
1309 3. socket is not in passive state.
1310 4. Finally, it really contains options, which user wants to receive.
1313 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1314 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1315 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1316 np->mcast_oif = tcp_v6_iif(opt_skb);
1317 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1318 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1319 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1320 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1322 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1323 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1324 skb_set_owner_r(opt_skb, sk);
1325 tcp_v6_restore_cb(opt_skb);
1326 opt_skb = xchg(&np->pktoptions, opt_skb);
1328 __kfree_skb(opt_skb);
1329 opt_skb = xchg(&np->pktoptions, NULL);
1337 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1338 const struct tcphdr *th)
1340 /* This is tricky: we move IP6CB at its correct location into
1341 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1342 * _decode_session6() uses IP6CB().
1343 * barrier() makes sure compiler won't play aliasing games.
1345 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1346 sizeof(struct inet6_skb_parm));
1349 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1350 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1351 skb->len - th->doff*4);
1352 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1353 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1354 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1355 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1356 TCP_SKB_CB(skb)->sacked = 0;
1359 static int tcp_v6_rcv(struct sk_buff *skb)
1361 const struct tcphdr *th;
1362 const struct ipv6hdr *hdr;
1365 struct net *net = dev_net(skb->dev);
1367 if (skb->pkt_type != PACKET_HOST)
1371 * Count it even if it's bad.
1373 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1375 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1380 if (th->doff < sizeof(struct tcphdr)/4)
1382 if (!pskb_may_pull(skb, th->doff*4))
1385 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1389 hdr = ipv6_hdr(skb);
1392 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest,
1398 if (sk->sk_state == TCP_TIME_WAIT)
1401 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1402 struct request_sock *req = inet_reqsk(sk);
1405 sk = req->rsk_listener;
1406 tcp_v6_fill_cb(skb, hdr, th);
1407 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1411 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1412 inet_csk_reqsk_queue_drop_and_put(sk, req);
1416 nsk = tcp_check_req(sk, skb, req, false);
1419 goto discard_and_relse;
1423 tcp_v6_restore_cb(skb);
1424 } else if (tcp_child_process(sk, nsk, skb)) {
1425 tcp_v6_send_reset(nsk, skb);
1426 goto discard_and_relse;
1432 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1433 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1434 goto discard_and_relse;
1437 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1438 goto discard_and_relse;
1440 tcp_v6_fill_cb(skb, hdr, th);
1442 if (tcp_v6_inbound_md5_hash(sk, skb))
1443 goto discard_and_relse;
1445 if (tcp_filter(sk, skb))
1446 goto discard_and_relse;
1447 th = (const struct tcphdr *)skb->data;
1448 hdr = ipv6_hdr(skb);
1452 if (sk->sk_state == TCP_LISTEN) {
1453 ret = tcp_v6_do_rcv(sk, skb);
1454 goto put_and_return;
1457 sk_incoming_cpu_update(sk);
1459 bh_lock_sock_nested(sk);
1460 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1462 if (!sock_owned_by_user(sk)) {
1463 if (!tcp_prequeue(sk, skb))
1464 ret = tcp_v6_do_rcv(sk, skb);
1465 } else if (unlikely(sk_add_backlog(sk, skb,
1466 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1468 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1469 goto discard_and_relse;
1475 return ret ? -1 : 0;
1478 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1481 tcp_v6_fill_cb(skb, hdr, th);
1483 if (tcp_checksum_complete(skb)) {
1485 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1487 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1489 tcp_v6_send_reset(NULL, skb);
1501 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1502 inet_twsk_put(inet_twsk(sk));
1506 tcp_v6_fill_cb(skb, hdr, th);
1508 if (tcp_checksum_complete(skb)) {
1509 inet_twsk_put(inet_twsk(sk));
1513 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1518 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1519 &ipv6_hdr(skb)->saddr, th->source,
1520 &ipv6_hdr(skb)->daddr,
1521 ntohs(th->dest), tcp_v6_iif(skb));
1523 struct inet_timewait_sock *tw = inet_twsk(sk);
1524 inet_twsk_deschedule_put(tw);
1526 tcp_v6_restore_cb(skb);
1529 /* Fall through to ACK */
1532 tcp_v6_timewait_ack(sk, skb);
1535 tcp_v6_restore_cb(skb);
1537 case TCP_TW_SUCCESS:
1543 static void tcp_v6_early_demux(struct sk_buff *skb)
1545 const struct ipv6hdr *hdr;
1546 const struct tcphdr *th;
1549 if (skb->pkt_type != PACKET_HOST)
1552 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1555 hdr = ipv6_hdr(skb);
1558 if (th->doff < sizeof(struct tcphdr) / 4)
1561 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1562 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1563 &hdr->saddr, th->source,
1564 &hdr->daddr, ntohs(th->dest),
1568 skb->destructor = sock_edemux;
1569 if (sk_fullsock(sk)) {
1570 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1573 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1575 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1576 skb_dst_set_noref(skb, dst);
1581 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1582 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1583 .twsk_unique = tcp_twsk_unique,
1584 .twsk_destructor = tcp_twsk_destructor,
1587 static const struct inet_connection_sock_af_ops ipv6_specific = {
1588 .queue_xmit = inet6_csk_xmit,
1589 .send_check = tcp_v6_send_check,
1590 .rebuild_header = inet6_sk_rebuild_header,
1591 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1592 .conn_request = tcp_v6_conn_request,
1593 .syn_recv_sock = tcp_v6_syn_recv_sock,
1594 .net_header_len = sizeof(struct ipv6hdr),
1595 .net_frag_header_len = sizeof(struct frag_hdr),
1596 .setsockopt = ipv6_setsockopt,
1597 .getsockopt = ipv6_getsockopt,
1598 .addr2sockaddr = inet6_csk_addr2sockaddr,
1599 .sockaddr_len = sizeof(struct sockaddr_in6),
1600 .bind_conflict = inet6_csk_bind_conflict,
1601 #ifdef CONFIG_COMPAT
1602 .compat_setsockopt = compat_ipv6_setsockopt,
1603 .compat_getsockopt = compat_ipv6_getsockopt,
1605 .mtu_reduced = tcp_v6_mtu_reduced,
1608 #ifdef CONFIG_TCP_MD5SIG
1609 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1610 .md5_lookup = tcp_v6_md5_lookup,
1611 .calc_md5_hash = tcp_v6_md5_hash_skb,
1612 .md5_parse = tcp_v6_parse_md5_keys,
1617 * TCP over IPv4 via INET6 API
1619 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1620 .queue_xmit = ip_queue_xmit,
1621 .send_check = tcp_v4_send_check,
1622 .rebuild_header = inet_sk_rebuild_header,
1623 .sk_rx_dst_set = inet_sk_rx_dst_set,
1624 .conn_request = tcp_v6_conn_request,
1625 .syn_recv_sock = tcp_v6_syn_recv_sock,
1626 .net_header_len = sizeof(struct iphdr),
1627 .setsockopt = ipv6_setsockopt,
1628 .getsockopt = ipv6_getsockopt,
1629 .addr2sockaddr = inet6_csk_addr2sockaddr,
1630 .sockaddr_len = sizeof(struct sockaddr_in6),
1631 .bind_conflict = inet6_csk_bind_conflict,
1632 #ifdef CONFIG_COMPAT
1633 .compat_setsockopt = compat_ipv6_setsockopt,
1634 .compat_getsockopt = compat_ipv6_getsockopt,
1636 .mtu_reduced = tcp_v4_mtu_reduced,
1639 #ifdef CONFIG_TCP_MD5SIG
1640 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1641 .md5_lookup = tcp_v4_md5_lookup,
1642 .calc_md5_hash = tcp_v4_md5_hash_skb,
1643 .md5_parse = tcp_v6_parse_md5_keys,
1647 /* NOTE: A lot of things set to zero explicitly by call to
1648 * sk_alloc() so need not be done here.
1650 static int tcp_v6_init_sock(struct sock *sk)
1652 struct inet_connection_sock *icsk = inet_csk(sk);
1656 icsk->icsk_af_ops = &ipv6_specific;
1658 #ifdef CONFIG_TCP_MD5SIG
1659 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1665 static void tcp_v6_destroy_sock(struct sock *sk)
1667 tcp_v4_destroy_sock(sk);
1668 inet6_destroy_sock(sk);
1671 #ifdef CONFIG_PROC_FS
1672 /* Proc filesystem TCPv6 sock list dumping. */
1673 static void get_openreq6(struct seq_file *seq,
1674 const struct request_sock *req, int i)
1676 long ttd = req->rsk_timer.expires - jiffies;
1677 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1678 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1684 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1685 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1687 src->s6_addr32[0], src->s6_addr32[1],
1688 src->s6_addr32[2], src->s6_addr32[3],
1689 inet_rsk(req)->ir_num,
1690 dest->s6_addr32[0], dest->s6_addr32[1],
1691 dest->s6_addr32[2], dest->s6_addr32[3],
1692 ntohs(inet_rsk(req)->ir_rmt_port),
1694 0, 0, /* could print option size, but that is af dependent. */
1695 1, /* timers active (only the expire timer) */
1696 jiffies_to_clock_t(ttd),
1698 from_kuid_munged(seq_user_ns(seq),
1699 sock_i_uid(req->rsk_listener)),
1700 0, /* non standard timer */
1701 0, /* open_requests have no inode */
1705 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1707 const struct in6_addr *dest, *src;
1710 unsigned long timer_expires;
1711 const struct inet_sock *inet = inet_sk(sp);
1712 const struct tcp_sock *tp = tcp_sk(sp);
1713 const struct inet_connection_sock *icsk = inet_csk(sp);
1714 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1718 dest = &sp->sk_v6_daddr;
1719 src = &sp->sk_v6_rcv_saddr;
1720 destp = ntohs(inet->inet_dport);
1721 srcp = ntohs(inet->inet_sport);
1723 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1724 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
1725 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1727 timer_expires = icsk->icsk_timeout;
1728 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1730 timer_expires = icsk->icsk_timeout;
1731 } else if (timer_pending(&sp->sk_timer)) {
1733 timer_expires = sp->sk_timer.expires;
1736 timer_expires = jiffies;
1739 state = sk_state_load(sp);
1740 if (state == TCP_LISTEN)
1741 rx_queue = sp->sk_ack_backlog;
1743 /* Because we don't lock the socket,
1744 * we might find a transient negative value.
1746 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1749 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1750 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1752 src->s6_addr32[0], src->s6_addr32[1],
1753 src->s6_addr32[2], src->s6_addr32[3], srcp,
1754 dest->s6_addr32[0], dest->s6_addr32[1],
1755 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1757 tp->write_seq - tp->snd_una,
1760 jiffies_delta_to_clock_t(timer_expires - jiffies),
1761 icsk->icsk_retransmits,
1762 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1763 icsk->icsk_probes_out,
1765 atomic_read(&sp->sk_refcnt), sp,
1766 jiffies_to_clock_t(icsk->icsk_rto),
1767 jiffies_to_clock_t(icsk->icsk_ack.ato),
1768 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1770 state == TCP_LISTEN ?
1771 fastopenq->max_qlen :
1772 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1776 static void get_timewait6_sock(struct seq_file *seq,
1777 struct inet_timewait_sock *tw, int i)
1779 long delta = tw->tw_timer.expires - jiffies;
1780 const struct in6_addr *dest, *src;
1783 dest = &tw->tw_v6_daddr;
1784 src = &tw->tw_v6_rcv_saddr;
1785 destp = ntohs(tw->tw_dport);
1786 srcp = ntohs(tw->tw_sport);
1789 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1790 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1792 src->s6_addr32[0], src->s6_addr32[1],
1793 src->s6_addr32[2], src->s6_addr32[3], srcp,
1794 dest->s6_addr32[0], dest->s6_addr32[1],
1795 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1796 tw->tw_substate, 0, 0,
1797 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1798 atomic_read(&tw->tw_refcnt), tw);
1801 static int tcp6_seq_show(struct seq_file *seq, void *v)
1803 struct tcp_iter_state *st;
1804 struct sock *sk = v;
1806 if (v == SEQ_START_TOKEN) {
1811 "st tx_queue rx_queue tr tm->when retrnsmt"
1812 " uid timeout inode\n");
1817 if (sk->sk_state == TCP_TIME_WAIT)
1818 get_timewait6_sock(seq, v, st->num);
1819 else if (sk->sk_state == TCP_NEW_SYN_RECV)
1820 get_openreq6(seq, v, st->num);
1822 get_tcp6_sock(seq, v, st->num);
1827 static const struct file_operations tcp6_afinfo_seq_fops = {
1828 .owner = THIS_MODULE,
1829 .open = tcp_seq_open,
1831 .llseek = seq_lseek,
1832 .release = seq_release_net
1835 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1838 .seq_fops = &tcp6_afinfo_seq_fops,
1840 .show = tcp6_seq_show,
1844 int __net_init tcp6_proc_init(struct net *net)
1846 return tcp_proc_register(net, &tcp6_seq_afinfo);
1849 void tcp6_proc_exit(struct net *net)
1851 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1855 static void tcp_v6_clear_sk(struct sock *sk, int size)
1857 struct inet_sock *inet = inet_sk(sk);
1859 /* we do not want to clear pinet6 field, because of RCU lookups */
1860 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1862 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1863 memset(&inet->pinet6 + 1, 0, size);
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 #ifdef CONFIG_MEMCG_KMEM
1908 .proto_cgroup = tcp_proto_cgroup,
1910 .clear_sk = tcp_v6_clear_sk,
1913 static const struct inet6_protocol tcpv6_protocol = {
1914 .early_demux = tcp_v6_early_demux,
1915 .handler = tcp_v6_rcv,
1916 .err_handler = tcp_v6_err,
1917 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1920 static struct inet_protosw tcpv6_protosw = {
1921 .type = SOCK_STREAM,
1922 .protocol = IPPROTO_TCP,
1923 .prot = &tcpv6_prot,
1924 .ops = &inet6_stream_ops,
1925 .flags = INET_PROTOSW_PERMANENT |
1929 static int __net_init tcpv6_net_init(struct net *net)
1931 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1932 SOCK_RAW, IPPROTO_TCP, net);
1935 static void __net_exit tcpv6_net_exit(struct net *net)
1937 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1940 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1942 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1945 static struct pernet_operations tcpv6_net_ops = {
1946 .init = tcpv6_net_init,
1947 .exit = tcpv6_net_exit,
1948 .exit_batch = tcpv6_net_exit_batch,
1951 int __init tcpv6_init(void)
1955 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1959 /* register inet6 protocol */
1960 ret = inet6_register_protosw(&tcpv6_protosw);
1962 goto out_tcpv6_protocol;
1964 ret = register_pernet_subsys(&tcpv6_net_ops);
1966 goto out_tcpv6_protosw;
1971 inet6_unregister_protosw(&tcpv6_protosw);
1973 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1977 void tcpv6_exit(void)
1979 unregister_pernet_subsys(&tcpv6_net_ops);
1980 inet6_unregister_protosw(&tcpv6_protosw);
1981 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);