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(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);
97 const struct rt6_info *rt = (const struct rt6_info *)dst;
101 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
102 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
106 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
108 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
109 ipv6_hdr(skb)->saddr.s6_addr32,
111 tcp_hdr(skb)->source);
114 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
117 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
118 struct inet_sock *inet = inet_sk(sk);
119 struct inet_connection_sock *icsk = inet_csk(sk);
120 struct ipv6_pinfo *np = inet6_sk(sk);
121 struct tcp_sock *tp = tcp_sk(sk);
122 struct in6_addr *saddr = NULL, *final_p, final;
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 final_p = fl6_update_dst(&fl6, np->opt, &final);
240 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
242 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
250 sk->sk_v6_rcv_saddr = *saddr;
253 /* set the source address */
255 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
257 sk->sk_gso_type = SKB_GSO_TCPV6;
258 __ip6_dst_store(sk, dst, NULL, NULL);
260 if (tcp_death_row.sysctl_tw_recycle &&
261 !tp->rx_opt.ts_recent_stamp &&
262 ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
263 tcp_fetch_timewait_stamp(sk, dst);
265 icsk->icsk_ext_hdr_len = 0;
267 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
270 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
272 inet->inet_dport = usin->sin6_port;
274 tcp_set_state(sk, TCP_SYN_SENT);
275 err = inet6_hash_connect(&tcp_death_row, sk);
281 if (!tp->write_seq && likely(!tp->repair))
282 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
283 sk->sk_v6_daddr.s6_addr32,
287 err = tcp_connect(sk);
294 tcp_set_state(sk, TCP_CLOSE);
297 inet->inet_dport = 0;
298 sk->sk_route_caps = 0;
302 static void tcp_v6_mtu_reduced(struct sock *sk)
304 struct dst_entry *dst;
306 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
309 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
313 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
314 tcp_sync_mss(sk, dst_mtu(dst));
315 tcp_simple_retransmit(sk);
319 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
320 u8 type, u8 code, int offset, __be32 info)
322 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
323 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
324 struct net *net = dev_net(skb->dev);
325 struct request_sock *fastopen;
326 struct ipv6_pinfo *np;
332 sk = __inet6_lookup_established(net, &tcp_hashinfo,
333 &hdr->daddr, th->dest,
334 &hdr->saddr, ntohs(th->source),
338 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
343 if (sk->sk_state == TCP_TIME_WAIT) {
344 inet_twsk_put(inet_twsk(sk));
347 seq = ntohl(th->seq);
348 if (sk->sk_state == TCP_NEW_SYN_RECV)
349 return tcp_req_err(sk, seq);
352 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
353 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
355 if (sk->sk_state == TCP_CLOSE)
358 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
359 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
364 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
365 fastopen = tp->fastopen_rsk;
366 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
367 if (sk->sk_state != TCP_LISTEN &&
368 !between(seq, snd_una, tp->snd_nxt)) {
369 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
375 if (type == NDISC_REDIRECT) {
376 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
379 dst->ops->redirect(dst, sk, skb);
383 if (type == ICMPV6_PKT_TOOBIG) {
384 /* We are not interested in TCP_LISTEN and open_requests
385 * (SYN-ACKs send out by Linux are always <576bytes so
386 * they should go through unfragmented).
388 if (sk->sk_state == TCP_LISTEN)
391 if (!ip6_sk_accept_pmtu(sk))
394 tp->mtu_info = ntohl(info);
395 if (!sock_owned_by_user(sk))
396 tcp_v6_mtu_reduced(sk);
397 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
403 icmpv6_err_convert(type, code, &err);
405 /* Might be for an request_sock */
406 switch (sk->sk_state) {
409 /* Only in fast or simultaneous open. If a fast open socket is
410 * is already accepted it is treated as a connected one below.
412 if (fastopen && !fastopen->sk)
415 if (!sock_owned_by_user(sk)) {
417 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
421 sk->sk_err_soft = err;
425 if (!sock_owned_by_user(sk) && np->recverr) {
427 sk->sk_error_report(sk);
429 sk->sk_err_soft = err;
437 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
439 struct request_sock *req,
441 struct tcp_fastopen_cookie *foc)
443 struct inet_request_sock *ireq = inet_rsk(req);
444 struct ipv6_pinfo *np = inet6_sk(sk);
445 struct flowi6 *fl6 = &fl->u.ip6;
449 /* First, grab a route. */
450 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
453 skb = tcp_make_synack(sk, dst, req, foc);
456 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
457 &ireq->ir_v6_rmt_addr);
459 fl6->daddr = ireq->ir_v6_rmt_addr;
460 if (np->repflow && ireq->pktopts)
461 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
463 skb_set_queue_mapping(skb, queue_mapping);
464 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
465 err = net_xmit_eval(err);
473 static void tcp_v6_reqsk_destructor(struct request_sock *req)
475 kfree_skb(inet_rsk(req)->pktopts);
478 #ifdef CONFIG_TCP_MD5SIG
479 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
480 const struct in6_addr *addr)
482 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
485 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
486 const struct sock *addr_sk)
488 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
491 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
494 struct tcp_md5sig cmd;
495 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
497 if (optlen < sizeof(cmd))
500 if (copy_from_user(&cmd, optval, sizeof(cmd)))
503 if (sin6->sin6_family != AF_INET6)
506 if (!cmd.tcpm_keylen) {
507 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
508 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
510 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
514 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
517 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
518 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
519 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
521 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
522 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
525 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
526 const struct in6_addr *daddr,
527 const struct in6_addr *saddr, int nbytes)
529 struct tcp6_pseudohdr *bp;
530 struct scatterlist sg;
532 bp = &hp->md5_blk.ip6;
533 /* 1. TCP pseudo-header (RFC2460) */
536 bp->protocol = cpu_to_be32(IPPROTO_TCP);
537 bp->len = cpu_to_be32(nbytes);
539 sg_init_one(&sg, bp, sizeof(*bp));
540 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
543 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
544 const struct in6_addr *daddr, struct in6_addr *saddr,
545 const struct tcphdr *th)
547 struct tcp_md5sig_pool *hp;
548 struct hash_desc *desc;
550 hp = tcp_get_md5sig_pool();
552 goto clear_hash_noput;
553 desc = &hp->md5_desc;
555 if (crypto_hash_init(desc))
557 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
559 if (tcp_md5_hash_header(hp, th))
561 if (tcp_md5_hash_key(hp, key))
563 if (crypto_hash_final(desc, md5_hash))
566 tcp_put_md5sig_pool();
570 tcp_put_md5sig_pool();
572 memset(md5_hash, 0, 16);
576 static int tcp_v6_md5_hash_skb(char *md5_hash,
577 const struct tcp_md5sig_key *key,
578 const struct sock *sk,
579 const struct sk_buff *skb)
581 const struct in6_addr *saddr, *daddr;
582 struct tcp_md5sig_pool *hp;
583 struct hash_desc *desc;
584 const struct tcphdr *th = tcp_hdr(skb);
586 if (sk) { /* valid for establish/request sockets */
587 saddr = &sk->sk_v6_rcv_saddr;
588 daddr = &sk->sk_v6_daddr;
590 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
591 saddr = &ip6h->saddr;
592 daddr = &ip6h->daddr;
595 hp = tcp_get_md5sig_pool();
597 goto clear_hash_noput;
598 desc = &hp->md5_desc;
600 if (crypto_hash_init(desc))
603 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
605 if (tcp_md5_hash_header(hp, th))
607 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
609 if (tcp_md5_hash_key(hp, key))
611 if (crypto_hash_final(desc, md5_hash))
614 tcp_put_md5sig_pool();
618 tcp_put_md5sig_pool();
620 memset(md5_hash, 0, 16);
624 static bool tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
626 const __u8 *hash_location = NULL;
627 struct tcp_md5sig_key *hash_expected;
628 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
629 const struct tcphdr *th = tcp_hdr(skb);
633 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
634 hash_location = tcp_parse_md5sig_option(th);
636 /* We've parsed the options - do we have a hash? */
637 if (!hash_expected && !hash_location)
640 if (hash_expected && !hash_location) {
641 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
645 if (!hash_expected && hash_location) {
646 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
650 /* check the signature */
651 genhash = tcp_v6_md5_hash_skb(newhash,
655 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
656 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
657 genhash ? "failed" : "mismatch",
658 &ip6h->saddr, ntohs(th->source),
659 &ip6h->daddr, ntohs(th->dest));
666 static void tcp_v6_init_req(struct request_sock *req,
667 const struct sock *sk_listener,
670 struct inet_request_sock *ireq = inet_rsk(req);
671 const struct ipv6_pinfo *np = inet6_sk(sk_listener);
673 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
674 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
676 /* So that link locals have meaning */
677 if (!sk_listener->sk_bound_dev_if &&
678 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
679 ireq->ir_iif = tcp_v6_iif(skb);
681 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
682 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
683 np->rxopt.bits.rxinfo ||
684 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
685 np->rxopt.bits.rxohlim || np->repflow)) {
686 atomic_inc(&skb->users);
691 static struct dst_entry *tcp_v6_route_req(struct sock *sk, struct flowi *fl,
692 const struct request_sock *req,
697 return inet6_csk_route_req(sk, &fl->u.ip6, req);
700 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
702 .obj_size = sizeof(struct tcp6_request_sock),
703 .rtx_syn_ack = tcp_rtx_synack,
704 .send_ack = tcp_v6_reqsk_send_ack,
705 .destructor = tcp_v6_reqsk_destructor,
706 .send_reset = tcp_v6_send_reset,
707 .syn_ack_timeout = tcp_syn_ack_timeout,
710 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
711 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
712 sizeof(struct ipv6hdr),
713 #ifdef CONFIG_TCP_MD5SIG
714 .req_md5_lookup = tcp_v6_md5_lookup,
715 .calc_md5_hash = tcp_v6_md5_hash_skb,
717 .init_req = tcp_v6_init_req,
718 #ifdef CONFIG_SYN_COOKIES
719 .cookie_init_seq = cookie_v6_init_sequence,
721 .route_req = tcp_v6_route_req,
722 .init_seq = tcp_v6_init_sequence,
723 .send_synack = tcp_v6_send_synack,
724 .queue_hash_add = inet6_csk_reqsk_queue_hash_add,
727 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
728 u32 ack, u32 win, u32 tsval, u32 tsecr,
729 int oif, struct tcp_md5sig_key *key, int rst,
730 u8 tclass, u32 label)
732 const struct tcphdr *th = tcp_hdr(skb);
734 struct sk_buff *buff;
736 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
737 struct sock *ctl_sk = net->ipv6.tcp_sk;
738 unsigned int tot_len = sizeof(struct tcphdr);
739 struct dst_entry *dst;
743 tot_len += TCPOLEN_TSTAMP_ALIGNED;
744 #ifdef CONFIG_TCP_MD5SIG
746 tot_len += TCPOLEN_MD5SIG_ALIGNED;
749 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
754 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
756 t1 = (struct tcphdr *) skb_push(buff, tot_len);
757 skb_reset_transport_header(buff);
759 /* Swap the send and the receive. */
760 memset(t1, 0, sizeof(*t1));
761 t1->dest = th->source;
762 t1->source = th->dest;
763 t1->doff = tot_len / 4;
764 t1->seq = htonl(seq);
765 t1->ack_seq = htonl(ack);
766 t1->ack = !rst || !th->ack;
768 t1->window = htons(win);
770 topt = (__be32 *)(t1 + 1);
773 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
774 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
775 *topt++ = htonl(tsval);
776 *topt++ = htonl(tsecr);
779 #ifdef CONFIG_TCP_MD5SIG
781 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
782 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
783 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
784 &ipv6_hdr(skb)->saddr,
785 &ipv6_hdr(skb)->daddr, t1);
789 memset(&fl6, 0, sizeof(fl6));
790 fl6.daddr = ipv6_hdr(skb)->saddr;
791 fl6.saddr = ipv6_hdr(skb)->daddr;
792 fl6.flowlabel = label;
794 buff->ip_summed = CHECKSUM_PARTIAL;
797 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
799 fl6.flowi6_proto = IPPROTO_TCP;
800 if (rt6_need_strict(&fl6.daddr) && !oif)
801 fl6.flowi6_oif = tcp_v6_iif(skb);
803 fl6.flowi6_oif = oif;
804 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
805 fl6.fl6_dport = t1->dest;
806 fl6.fl6_sport = t1->source;
807 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
809 /* Pass a socket to ip6_dst_lookup either it is for RST
810 * Underlying function will use this to retrieve the network
813 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
815 skb_dst_set(buff, dst);
816 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
817 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
819 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
826 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
828 const struct tcphdr *th = tcp_hdr(skb);
829 u32 seq = 0, ack_seq = 0;
830 struct tcp_md5sig_key *key = NULL;
831 #ifdef CONFIG_TCP_MD5SIG
832 const __u8 *hash_location = NULL;
833 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
834 unsigned char newhash[16];
836 struct sock *sk1 = NULL;
843 /* If sk not NULL, it means we did a successful lookup and incoming
844 * route had to be correct. prequeue might have dropped our dst.
846 if (!sk && !ipv6_unicast_destination(skb))
849 #ifdef CONFIG_TCP_MD5SIG
850 hash_location = tcp_parse_md5sig_option(th);
851 if (!sk && hash_location) {
853 * active side is lost. Try to find listening socket through
854 * source port, and then find md5 key through listening socket.
855 * we are not loose security here:
856 * Incoming packet is checked with md5 hash with finding key,
857 * no RST generated if md5 hash doesn't match.
859 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
860 &tcp_hashinfo, &ipv6h->saddr,
861 th->source, &ipv6h->daddr,
862 ntohs(th->source), tcp_v6_iif(skb));
867 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
871 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
872 if (genhash || memcmp(hash_location, newhash, 16) != 0)
875 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
880 seq = ntohl(th->ack_seq);
882 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
885 oif = sk ? sk->sk_bound_dev_if : 0;
886 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
888 #ifdef CONFIG_TCP_MD5SIG
897 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
898 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
899 struct tcp_md5sig_key *key, u8 tclass,
902 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
906 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
908 struct inet_timewait_sock *tw = inet_twsk(sk);
909 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
911 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
912 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
913 tcp_time_stamp + tcptw->tw_ts_offset,
914 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
915 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
920 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
921 struct request_sock *req)
923 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
924 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
926 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
927 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
928 tcp_rsk(req)->rcv_nxt, req->rcv_wnd,
929 tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
930 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
935 static struct sock *tcp_v6_hnd_req(struct sock *sk, struct sk_buff *skb)
937 const struct tcphdr *th = tcp_hdr(skb);
938 struct request_sock *req;
941 /* Find possible connection requests. */
942 req = inet6_csk_search_req(sk, th->source,
943 &ipv6_hdr(skb)->saddr,
944 &ipv6_hdr(skb)->daddr, tcp_v6_iif(skb));
946 nsk = tcp_check_req(sk, skb, req, false);
947 if (!nsk || nsk == sk)
951 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
952 &ipv6_hdr(skb)->saddr, th->source,
953 &ipv6_hdr(skb)->daddr, ntohs(th->dest),
957 if (nsk->sk_state != TCP_TIME_WAIT) {
961 inet_twsk_put(inet_twsk(nsk));
965 #ifdef CONFIG_SYN_COOKIES
967 sk = cookie_v6_check(sk, skb);
972 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
974 if (skb->protocol == htons(ETH_P_IP))
975 return tcp_v4_conn_request(sk, skb);
977 if (!ipv6_unicast_destination(skb))
980 return tcp_conn_request(&tcp6_request_sock_ops,
981 &tcp_request_sock_ipv6_ops, sk, skb);
984 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
985 return 0; /* don't send reset */
988 static struct sock *tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
989 struct request_sock *req,
990 struct dst_entry *dst)
992 struct inet_request_sock *ireq;
993 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
994 struct tcp6_sock *newtcp6sk;
995 struct inet_sock *newinet;
996 struct tcp_sock *newtp;
998 #ifdef CONFIG_TCP_MD5SIG
999 struct tcp_md5sig_key *key;
1003 if (skb->protocol == htons(ETH_P_IP)) {
1008 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1013 newtcp6sk = (struct tcp6_sock *)newsk;
1014 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1016 newinet = inet_sk(newsk);
1017 newnp = inet6_sk(newsk);
1018 newtp = tcp_sk(newsk);
1020 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1022 newnp->saddr = newsk->sk_v6_rcv_saddr;
1024 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1025 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1026 #ifdef CONFIG_TCP_MD5SIG
1027 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1030 newnp->ipv6_ac_list = NULL;
1031 newnp->ipv6_fl_list = NULL;
1032 newnp->pktoptions = NULL;
1034 newnp->mcast_oif = tcp_v6_iif(skb);
1035 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1036 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1038 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1041 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1042 * here, tcp_create_openreq_child now does this for us, see the comment in
1043 * that function for the gory details. -acme
1046 /* It is tricky place. Until this moment IPv4 tcp
1047 worked with IPv6 icsk.icsk_af_ops.
1050 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1055 ireq = inet_rsk(req);
1057 if (sk_acceptq_is_full(sk))
1061 dst = inet6_csk_route_req(sk, &fl6, req);
1066 newsk = tcp_create_openreq_child(sk, req, skb);
1071 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1072 * count here, tcp_create_openreq_child now does this for us, see the
1073 * comment in that function for the gory details. -acme
1076 newsk->sk_gso_type = SKB_GSO_TCPV6;
1077 __ip6_dst_store(newsk, dst, NULL, NULL);
1078 inet6_sk_rx_dst_set(newsk, skb);
1080 newtcp6sk = (struct tcp6_sock *)newsk;
1081 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1083 newtp = tcp_sk(newsk);
1084 newinet = inet_sk(newsk);
1085 newnp = inet6_sk(newsk);
1087 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1089 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1090 newnp->saddr = ireq->ir_v6_loc_addr;
1091 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1092 newsk->sk_bound_dev_if = ireq->ir_iif;
1094 /* Now IPv6 options...
1096 First: no IPv4 options.
1098 newinet->inet_opt = NULL;
1099 newnp->ipv6_ac_list = NULL;
1100 newnp->ipv6_fl_list = NULL;
1103 newnp->rxopt.all = np->rxopt.all;
1105 /* Clone pktoptions received with SYN */
1106 newnp->pktoptions = NULL;
1107 if (ireq->pktopts) {
1108 newnp->pktoptions = skb_clone(ireq->pktopts,
1109 sk_gfp_atomic(sk, GFP_ATOMIC));
1110 consume_skb(ireq->pktopts);
1111 ireq->pktopts = NULL;
1112 if (newnp->pktoptions)
1113 skb_set_owner_r(newnp->pktoptions, newsk);
1116 newnp->mcast_oif = tcp_v6_iif(skb);
1117 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1118 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1120 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1122 /* Clone native IPv6 options from listening socket (if any)
1124 Yes, keeping reference count would be much more clever,
1125 but we make one more one thing there: reattach optmem
1129 newnp->opt = ipv6_dup_options(newsk, np->opt);
1131 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1133 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1134 newnp->opt->opt_flen);
1136 tcp_ca_openreq_child(newsk, dst);
1138 tcp_sync_mss(newsk, dst_mtu(dst));
1139 newtp->advmss = dst_metric_advmss(dst);
1140 if (tcp_sk(sk)->rx_opt.user_mss &&
1141 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1142 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1144 tcp_initialize_rcv_mss(newsk);
1146 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1147 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1149 #ifdef CONFIG_TCP_MD5SIG
1150 /* Copy over the MD5 key from the original socket */
1151 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1153 /* We're using one, so create a matching key
1154 * on the newsk structure. If we fail to get
1155 * memory, then we end up not copying the key
1158 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1159 AF_INET6, key->key, key->keylen,
1160 sk_gfp_atomic(sk, GFP_ATOMIC));
1164 if (__inet_inherit_port(sk, newsk) < 0) {
1165 inet_csk_prepare_forced_close(newsk);
1169 __inet_hash(newsk, NULL);
1174 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1178 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1182 /* The socket must have it's spinlock held when we get
1185 * We have a potential double-lock case here, so even when
1186 * doing backlog processing we use the BH locking scheme.
1187 * This is because we cannot sleep with the original spinlock
1190 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1192 struct ipv6_pinfo *np = inet6_sk(sk);
1193 struct tcp_sock *tp;
1194 struct sk_buff *opt_skb = NULL;
1196 /* Imagine: socket is IPv6. IPv4 packet arrives,
1197 goes to IPv4 receive handler and backlogged.
1198 From backlog it always goes here. Kerboom...
1199 Fortunately, tcp_rcv_established and rcv_established
1200 handle them correctly, but it is not case with
1201 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1204 if (skb->protocol == htons(ETH_P_IP))
1205 return tcp_v4_do_rcv(sk, skb);
1207 if (sk_filter(sk, skb))
1211 * socket locking is here for SMP purposes as backlog rcv
1212 * is currently called with bh processing disabled.
1215 /* Do Stevens' IPV6_PKTOPTIONS.
1217 Yes, guys, it is the only place in our code, where we
1218 may make it not affecting IPv4.
1219 The rest of code is protocol independent,
1220 and I do not like idea to uglify IPv4.
1222 Actually, all the idea behind IPV6_PKTOPTIONS
1223 looks not very well thought. For now we latch
1224 options, received in the last packet, enqueued
1225 by tcp. Feel free to propose better solution.
1229 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1231 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1232 struct dst_entry *dst = sk->sk_rx_dst;
1234 sock_rps_save_rxhash(sk, skb);
1235 sk_mark_napi_id(sk, skb);
1237 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1238 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1240 sk->sk_rx_dst = NULL;
1244 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1246 goto ipv6_pktoptions;
1250 if (tcp_checksum_complete(skb))
1253 if (sk->sk_state == TCP_LISTEN) {
1254 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1259 * Queue it on the new socket if the new socket is active,
1260 * otherwise we just shortcircuit this and continue with
1264 sock_rps_save_rxhash(nsk, skb);
1265 sk_mark_napi_id(sk, skb);
1266 if (tcp_child_process(sk, nsk, skb))
1269 __kfree_skb(opt_skb);
1273 sock_rps_save_rxhash(sk, skb);
1275 if (tcp_rcv_state_process(sk, skb))
1278 goto ipv6_pktoptions;
1282 tcp_v6_send_reset(sk, skb);
1285 __kfree_skb(opt_skb);
1289 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1290 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1295 /* Do you ask, what is it?
1297 1. skb was enqueued by tcp.
1298 2. skb is added to tail of read queue, rather than out of order.
1299 3. socket is not in passive state.
1300 4. Finally, it really contains options, which user wants to receive.
1303 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1304 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1305 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1306 np->mcast_oif = tcp_v6_iif(opt_skb);
1307 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1308 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1309 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1310 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1312 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1313 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1314 skb_set_owner_r(opt_skb, sk);
1315 opt_skb = xchg(&np->pktoptions, opt_skb);
1317 __kfree_skb(opt_skb);
1318 opt_skb = xchg(&np->pktoptions, NULL);
1326 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1327 const struct tcphdr *th)
1329 /* This is tricky: we move IP6CB at its correct location into
1330 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1331 * _decode_session6() uses IP6CB().
1332 * barrier() makes sure compiler won't play aliasing games.
1334 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1335 sizeof(struct inet6_skb_parm));
1338 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1339 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1340 skb->len - th->doff*4);
1341 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1342 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1343 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1344 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1345 TCP_SKB_CB(skb)->sacked = 0;
1348 static void tcp_v6_restore_cb(struct sk_buff *skb)
1350 /* We need to move header back to the beginning if xfrm6_policy_check()
1351 * and tcp_v6_fill_cb() are going to be called again.
1353 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1354 sizeof(struct inet6_skb_parm));
1357 static int tcp_v6_rcv(struct sk_buff *skb)
1359 const struct tcphdr *th;
1360 const struct ipv6hdr *hdr;
1363 struct net *net = dev_net(skb->dev);
1365 if (skb->pkt_type != PACKET_HOST)
1369 * Count it even if it's bad.
1371 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1373 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1378 if (th->doff < sizeof(struct tcphdr)/4)
1380 if (!pskb_may_pull(skb, th->doff*4))
1383 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1387 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 (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1399 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1400 goto discard_and_relse;
1403 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1404 goto discard_and_relse;
1406 tcp_v6_fill_cb(skb, hdr, th);
1408 #ifdef CONFIG_TCP_MD5SIG
1409 if (tcp_v6_inbound_md5_hash(sk, skb))
1410 goto discard_and_relse;
1413 if (sk_filter(sk, skb))
1414 goto discard_and_relse;
1416 sk_incoming_cpu_update(sk);
1419 bh_lock_sock_nested(sk);
1420 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1422 if (!sock_owned_by_user(sk)) {
1423 if (!tcp_prequeue(sk, skb))
1424 ret = tcp_v6_do_rcv(sk, skb);
1425 } else if (unlikely(sk_add_backlog(sk, skb,
1426 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1428 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1429 goto discard_and_relse;
1434 return ret ? -1 : 0;
1437 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1440 tcp_v6_fill_cb(skb, hdr, th);
1442 if (tcp_checksum_complete(skb)) {
1444 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1446 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1448 tcp_v6_send_reset(NULL, skb);
1460 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1461 inet_twsk_put(inet_twsk(sk));
1465 tcp_v6_fill_cb(skb, hdr, th);
1467 if (tcp_checksum_complete(skb)) {
1468 inet_twsk_put(inet_twsk(sk));
1472 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1477 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1478 &ipv6_hdr(skb)->saddr, th->source,
1479 &ipv6_hdr(skb)->daddr,
1480 ntohs(th->dest), tcp_v6_iif(skb));
1482 struct inet_timewait_sock *tw = inet_twsk(sk);
1483 inet_twsk_deschedule_put(tw);
1485 tcp_v6_restore_cb(skb);
1488 /* Fall through to ACK */
1491 tcp_v6_timewait_ack(sk, skb);
1494 tcp_v6_restore_cb(skb);
1496 case TCP_TW_SUCCESS:
1502 static void tcp_v6_early_demux(struct sk_buff *skb)
1504 const struct ipv6hdr *hdr;
1505 const struct tcphdr *th;
1508 if (skb->pkt_type != PACKET_HOST)
1511 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1514 hdr = ipv6_hdr(skb);
1517 if (th->doff < sizeof(struct tcphdr) / 4)
1520 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1521 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1522 &hdr->saddr, th->source,
1523 &hdr->daddr, ntohs(th->dest),
1527 skb->destructor = sock_edemux;
1528 if (sk_fullsock(sk)) {
1529 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1532 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1534 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1535 skb_dst_set_noref(skb, dst);
1540 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1541 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1542 .twsk_unique = tcp_twsk_unique,
1543 .twsk_destructor = tcp_twsk_destructor,
1546 static const struct inet_connection_sock_af_ops ipv6_specific = {
1547 .queue_xmit = inet6_csk_xmit,
1548 .send_check = tcp_v6_send_check,
1549 .rebuild_header = inet6_sk_rebuild_header,
1550 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1551 .conn_request = tcp_v6_conn_request,
1552 .syn_recv_sock = tcp_v6_syn_recv_sock,
1553 .net_header_len = sizeof(struct ipv6hdr),
1554 .net_frag_header_len = sizeof(struct frag_hdr),
1555 .setsockopt = ipv6_setsockopt,
1556 .getsockopt = ipv6_getsockopt,
1557 .addr2sockaddr = inet6_csk_addr2sockaddr,
1558 .sockaddr_len = sizeof(struct sockaddr_in6),
1559 .bind_conflict = inet6_csk_bind_conflict,
1560 #ifdef CONFIG_COMPAT
1561 .compat_setsockopt = compat_ipv6_setsockopt,
1562 .compat_getsockopt = compat_ipv6_getsockopt,
1564 .mtu_reduced = tcp_v6_mtu_reduced,
1567 #ifdef CONFIG_TCP_MD5SIG
1568 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1569 .md5_lookup = tcp_v6_md5_lookup,
1570 .calc_md5_hash = tcp_v6_md5_hash_skb,
1571 .md5_parse = tcp_v6_parse_md5_keys,
1576 * TCP over IPv4 via INET6 API
1578 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1579 .queue_xmit = ip_queue_xmit,
1580 .send_check = tcp_v4_send_check,
1581 .rebuild_header = inet_sk_rebuild_header,
1582 .sk_rx_dst_set = inet_sk_rx_dst_set,
1583 .conn_request = tcp_v6_conn_request,
1584 .syn_recv_sock = tcp_v6_syn_recv_sock,
1585 .net_header_len = sizeof(struct iphdr),
1586 .setsockopt = ipv6_setsockopt,
1587 .getsockopt = ipv6_getsockopt,
1588 .addr2sockaddr = inet6_csk_addr2sockaddr,
1589 .sockaddr_len = sizeof(struct sockaddr_in6),
1590 .bind_conflict = inet6_csk_bind_conflict,
1591 #ifdef CONFIG_COMPAT
1592 .compat_setsockopt = compat_ipv6_setsockopt,
1593 .compat_getsockopt = compat_ipv6_getsockopt,
1595 .mtu_reduced = tcp_v4_mtu_reduced,
1598 #ifdef CONFIG_TCP_MD5SIG
1599 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1600 .md5_lookup = tcp_v4_md5_lookup,
1601 .calc_md5_hash = tcp_v4_md5_hash_skb,
1602 .md5_parse = tcp_v6_parse_md5_keys,
1606 /* NOTE: A lot of things set to zero explicitly by call to
1607 * sk_alloc() so need not be done here.
1609 static int tcp_v6_init_sock(struct sock *sk)
1611 struct inet_connection_sock *icsk = inet_csk(sk);
1615 icsk->icsk_af_ops = &ipv6_specific;
1617 #ifdef CONFIG_TCP_MD5SIG
1618 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1624 static void tcp_v6_destroy_sock(struct sock *sk)
1626 tcp_v4_destroy_sock(sk);
1627 inet6_destroy_sock(sk);
1630 #ifdef CONFIG_PROC_FS
1631 /* Proc filesystem TCPv6 sock list dumping. */
1632 static void get_openreq6(struct seq_file *seq,
1633 struct request_sock *req, int i, kuid_t uid)
1635 long ttd = req->rsk_timer.expires - jiffies;
1636 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1637 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1643 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1644 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1646 src->s6_addr32[0], src->s6_addr32[1],
1647 src->s6_addr32[2], src->s6_addr32[3],
1648 inet_rsk(req)->ir_num,
1649 dest->s6_addr32[0], dest->s6_addr32[1],
1650 dest->s6_addr32[2], dest->s6_addr32[3],
1651 ntohs(inet_rsk(req)->ir_rmt_port),
1653 0, 0, /* could print option size, but that is af dependent. */
1654 1, /* timers active (only the expire timer) */
1655 jiffies_to_clock_t(ttd),
1657 from_kuid_munged(seq_user_ns(seq), uid),
1658 0, /* non standard timer */
1659 0, /* open_requests have no inode */
1663 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1665 const struct in6_addr *dest, *src;
1668 unsigned long timer_expires;
1669 const struct inet_sock *inet = inet_sk(sp);
1670 const struct tcp_sock *tp = tcp_sk(sp);
1671 const struct inet_connection_sock *icsk = inet_csk(sp);
1672 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
1674 dest = &sp->sk_v6_daddr;
1675 src = &sp->sk_v6_rcv_saddr;
1676 destp = ntohs(inet->inet_dport);
1677 srcp = ntohs(inet->inet_sport);
1679 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1681 timer_expires = icsk->icsk_timeout;
1682 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1684 timer_expires = icsk->icsk_timeout;
1685 } else if (timer_pending(&sp->sk_timer)) {
1687 timer_expires = sp->sk_timer.expires;
1690 timer_expires = jiffies;
1694 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1695 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1697 src->s6_addr32[0], src->s6_addr32[1],
1698 src->s6_addr32[2], src->s6_addr32[3], srcp,
1699 dest->s6_addr32[0], dest->s6_addr32[1],
1700 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1702 tp->write_seq-tp->snd_una,
1703 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1705 jiffies_delta_to_clock_t(timer_expires - jiffies),
1706 icsk->icsk_retransmits,
1707 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1708 icsk->icsk_probes_out,
1710 atomic_read(&sp->sk_refcnt), sp,
1711 jiffies_to_clock_t(icsk->icsk_rto),
1712 jiffies_to_clock_t(icsk->icsk_ack.ato),
1713 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1715 sp->sk_state == TCP_LISTEN ?
1716 (fastopenq ? fastopenq->max_qlen : 0) :
1717 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1721 static void get_timewait6_sock(struct seq_file *seq,
1722 struct inet_timewait_sock *tw, int i)
1724 long delta = tw->tw_timer.expires - jiffies;
1725 const struct in6_addr *dest, *src;
1728 dest = &tw->tw_v6_daddr;
1729 src = &tw->tw_v6_rcv_saddr;
1730 destp = ntohs(tw->tw_dport);
1731 srcp = ntohs(tw->tw_sport);
1734 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1735 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1737 src->s6_addr32[0], src->s6_addr32[1],
1738 src->s6_addr32[2], src->s6_addr32[3], srcp,
1739 dest->s6_addr32[0], dest->s6_addr32[1],
1740 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1741 tw->tw_substate, 0, 0,
1742 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1743 atomic_read(&tw->tw_refcnt), tw);
1746 static int tcp6_seq_show(struct seq_file *seq, void *v)
1748 struct tcp_iter_state *st;
1749 struct sock *sk = v;
1751 if (v == SEQ_START_TOKEN) {
1756 "st tx_queue rx_queue tr tm->when retrnsmt"
1757 " uid timeout inode\n");
1762 switch (st->state) {
1763 case TCP_SEQ_STATE_LISTENING:
1764 case TCP_SEQ_STATE_ESTABLISHED:
1765 if (sk->sk_state == TCP_TIME_WAIT)
1766 get_timewait6_sock(seq, v, st->num);
1768 get_tcp6_sock(seq, v, st->num);
1770 case TCP_SEQ_STATE_OPENREQ:
1771 get_openreq6(seq, v, st->num, st->uid);
1778 static const struct file_operations tcp6_afinfo_seq_fops = {
1779 .owner = THIS_MODULE,
1780 .open = tcp_seq_open,
1782 .llseek = seq_lseek,
1783 .release = seq_release_net
1786 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1789 .seq_fops = &tcp6_afinfo_seq_fops,
1791 .show = tcp6_seq_show,
1795 int __net_init tcp6_proc_init(struct net *net)
1797 return tcp_proc_register(net, &tcp6_seq_afinfo);
1800 void tcp6_proc_exit(struct net *net)
1802 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1806 static void tcp_v6_clear_sk(struct sock *sk, int size)
1808 struct inet_sock *inet = inet_sk(sk);
1810 /* we do not want to clear pinet6 field, because of RCU lookups */
1811 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1813 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1814 memset(&inet->pinet6 + 1, 0, size);
1817 struct proto tcpv6_prot = {
1819 .owner = THIS_MODULE,
1821 .connect = tcp_v6_connect,
1822 .disconnect = tcp_disconnect,
1823 .accept = inet_csk_accept,
1825 .init = tcp_v6_init_sock,
1826 .destroy = tcp_v6_destroy_sock,
1827 .shutdown = tcp_shutdown,
1828 .setsockopt = tcp_setsockopt,
1829 .getsockopt = tcp_getsockopt,
1830 .recvmsg = tcp_recvmsg,
1831 .sendmsg = tcp_sendmsg,
1832 .sendpage = tcp_sendpage,
1833 .backlog_rcv = tcp_v6_do_rcv,
1834 .release_cb = tcp_release_cb,
1836 .unhash = inet_unhash,
1837 .get_port = inet_csk_get_port,
1838 .enter_memory_pressure = tcp_enter_memory_pressure,
1839 .stream_memory_free = tcp_stream_memory_free,
1840 .sockets_allocated = &tcp_sockets_allocated,
1841 .memory_allocated = &tcp_memory_allocated,
1842 .memory_pressure = &tcp_memory_pressure,
1843 .orphan_count = &tcp_orphan_count,
1844 .sysctl_mem = sysctl_tcp_mem,
1845 .sysctl_wmem = sysctl_tcp_wmem,
1846 .sysctl_rmem = sysctl_tcp_rmem,
1847 .max_header = MAX_TCP_HEADER,
1848 .obj_size = sizeof(struct tcp6_sock),
1849 .slab_flags = SLAB_DESTROY_BY_RCU,
1850 .twsk_prot = &tcp6_timewait_sock_ops,
1851 .rsk_prot = &tcp6_request_sock_ops,
1852 .h.hashinfo = &tcp_hashinfo,
1853 .no_autobind = true,
1854 #ifdef CONFIG_COMPAT
1855 .compat_setsockopt = compat_tcp_setsockopt,
1856 .compat_getsockopt = compat_tcp_getsockopt,
1858 #ifdef CONFIG_MEMCG_KMEM
1859 .proto_cgroup = tcp_proto_cgroup,
1861 .clear_sk = tcp_v6_clear_sk,
1864 static const struct inet6_protocol tcpv6_protocol = {
1865 .early_demux = tcp_v6_early_demux,
1866 .handler = tcp_v6_rcv,
1867 .err_handler = tcp_v6_err,
1868 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1871 static struct inet_protosw tcpv6_protosw = {
1872 .type = SOCK_STREAM,
1873 .protocol = IPPROTO_TCP,
1874 .prot = &tcpv6_prot,
1875 .ops = &inet6_stream_ops,
1876 .flags = INET_PROTOSW_PERMANENT |
1880 static int __net_init tcpv6_net_init(struct net *net)
1882 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1883 SOCK_RAW, IPPROTO_TCP, net);
1886 static void __net_exit tcpv6_net_exit(struct net *net)
1888 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1891 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1893 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1896 static struct pernet_operations tcpv6_net_ops = {
1897 .init = tcpv6_net_init,
1898 .exit = tcpv6_net_exit,
1899 .exit_batch = tcpv6_net_exit_batch,
1902 int __init tcpv6_init(void)
1906 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1910 /* register inet6 protocol */
1911 ret = inet6_register_protosw(&tcpv6_protosw);
1913 goto out_tcpv6_protocol;
1915 ret = register_pernet_subsys(&tcpv6_net_ops);
1917 goto out_tcpv6_protosw;
1922 inet6_unregister_protosw(&tcpv6_protosw);
1924 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1928 void tcpv6_exit(void)
1930 unregister_pernet_subsys(&tcpv6_net_ops);
1931 inet6_unregister_protosw(&tcpv6_protosw);
1932 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);