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/netdma.h>
63 #include <net/inet_common.h>
64 #include <net/secure_seq.h>
65 #include <net/tcp_memcontrol.h>
66 #include <net/busy_poll.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
71 #include <linux/crypto.h>
72 #include <linux/scatterlist.h>
74 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
75 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
76 struct request_sock *req);
78 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
80 static const struct inet_connection_sock_af_ops ipv6_mapped;
81 static const struct inet_connection_sock_af_ops ipv6_specific;
82 #ifdef CONFIG_TCP_MD5SIG
83 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
84 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
86 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
87 const struct in6_addr *addr)
93 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
95 struct dst_entry *dst = skb_dst(skb);
96 const struct rt6_info *rt = (const struct rt6_info *)dst;
100 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
102 inet6_sk(sk)->rx_dst_cookie = rt->rt6i_node->fn_sernum;
105 static void tcp_v6_hash(struct sock *sk)
107 if (sk->sk_state != TCP_CLOSE) {
108 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
113 __inet6_hash(sk, NULL);
118 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
120 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
121 ipv6_hdr(skb)->saddr.s6_addr32,
123 tcp_hdr(skb)->source);
126 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
129 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
130 struct inet_sock *inet = inet_sk(sk);
131 struct inet_connection_sock *icsk = inet_csk(sk);
132 struct ipv6_pinfo *np = inet6_sk(sk);
133 struct tcp_sock *tp = tcp_sk(sk);
134 struct in6_addr *saddr = NULL, *final_p, final;
137 struct dst_entry *dst;
141 if (addr_len < SIN6_LEN_RFC2133)
144 if (usin->sin6_family != AF_INET6)
145 return -EAFNOSUPPORT;
147 memset(&fl6, 0, sizeof(fl6));
150 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
151 IP6_ECN_flow_init(fl6.flowlabel);
152 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
153 struct ip6_flowlabel *flowlabel;
154 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
155 if (flowlabel == NULL)
157 fl6_sock_release(flowlabel);
162 * connect() to INADDR_ANY means loopback (BSD'ism).
165 if (ipv6_addr_any(&usin->sin6_addr))
166 usin->sin6_addr.s6_addr[15] = 0x1;
168 addr_type = ipv6_addr_type(&usin->sin6_addr);
170 if (addr_type & IPV6_ADDR_MULTICAST)
173 if (addr_type&IPV6_ADDR_LINKLOCAL) {
174 if (addr_len >= sizeof(struct sockaddr_in6) &&
175 usin->sin6_scope_id) {
176 /* If interface is set while binding, indices
179 if (sk->sk_bound_dev_if &&
180 sk->sk_bound_dev_if != usin->sin6_scope_id)
183 sk->sk_bound_dev_if = usin->sin6_scope_id;
186 /* Connect to link-local address requires an interface */
187 if (!sk->sk_bound_dev_if)
191 if (tp->rx_opt.ts_recent_stamp &&
192 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
198 sk->sk_v6_daddr = usin->sin6_addr;
199 np->flow_label = fl6.flowlabel;
205 if (addr_type == IPV6_ADDR_MAPPED) {
206 u32 exthdrlen = icsk->icsk_ext_hdr_len;
207 struct sockaddr_in sin;
209 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
211 if (__ipv6_only_sock(sk))
214 sin.sin_family = AF_INET;
215 sin.sin_port = usin->sin6_port;
216 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
218 icsk->icsk_af_ops = &ipv6_mapped;
219 sk->sk_backlog_rcv = tcp_v4_do_rcv;
220 #ifdef CONFIG_TCP_MD5SIG
221 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
224 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
227 icsk->icsk_ext_hdr_len = exthdrlen;
228 icsk->icsk_af_ops = &ipv6_specific;
229 sk->sk_backlog_rcv = tcp_v6_do_rcv;
230 #ifdef CONFIG_TCP_MD5SIG
231 tp->af_specific = &tcp_sock_ipv6_specific;
235 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
236 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
237 &sk->sk_v6_rcv_saddr);
243 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
244 saddr = &sk->sk_v6_rcv_saddr;
246 fl6.flowi6_proto = IPPROTO_TCP;
247 fl6.daddr = sk->sk_v6_daddr;
248 fl6.saddr = saddr ? *saddr : np->saddr;
249 fl6.flowi6_oif = sk->sk_bound_dev_if;
250 fl6.flowi6_mark = sk->sk_mark;
251 fl6.fl6_dport = usin->sin6_port;
252 fl6.fl6_sport = inet->inet_sport;
254 final_p = fl6_update_dst(&fl6, np->opt, &final);
256 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
258 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
266 sk->sk_v6_rcv_saddr = *saddr;
269 /* set the source address */
271 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
273 sk->sk_gso_type = SKB_GSO_TCPV6;
274 __ip6_dst_store(sk, dst, NULL, NULL);
276 rt = (struct rt6_info *) dst;
277 if (tcp_death_row.sysctl_tw_recycle &&
278 !tp->rx_opt.ts_recent_stamp &&
279 ipv6_addr_equal(&rt->rt6i_dst.addr, &sk->sk_v6_daddr))
280 tcp_fetch_timewait_stamp(sk, dst);
282 icsk->icsk_ext_hdr_len = 0;
284 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
287 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
289 inet->inet_dport = usin->sin6_port;
291 tcp_set_state(sk, TCP_SYN_SENT);
292 err = inet6_hash_connect(&tcp_death_row, sk);
296 if (!tp->write_seq && likely(!tp->repair))
297 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
298 sk->sk_v6_daddr.s6_addr32,
302 err = tcp_connect(sk);
309 tcp_set_state(sk, TCP_CLOSE);
312 inet->inet_dport = 0;
313 sk->sk_route_caps = 0;
317 static void tcp_v6_mtu_reduced(struct sock *sk)
319 struct dst_entry *dst;
321 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
324 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
328 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
329 tcp_sync_mss(sk, dst_mtu(dst));
330 tcp_simple_retransmit(sk);
334 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
335 u8 type, u8 code, int offset, __be32 info)
337 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
338 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
339 struct ipv6_pinfo *np;
344 struct net *net = dev_net(skb->dev);
346 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
347 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
350 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
355 if (sk->sk_state == TCP_TIME_WAIT) {
356 inet_twsk_put(inet_twsk(sk));
361 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
362 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
364 if (sk->sk_state == TCP_CLOSE)
367 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
368 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
373 seq = ntohl(th->seq);
374 if (sk->sk_state != TCP_LISTEN &&
375 !between(seq, tp->snd_una, tp->snd_nxt)) {
376 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
382 if (type == NDISC_REDIRECT) {
383 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
386 dst->ops->redirect(dst, sk, skb);
390 if (type == ICMPV6_PKT_TOOBIG) {
391 /* We are not interested in TCP_LISTEN and open_requests
392 * (SYN-ACKs send out by Linux are always <576bytes so
393 * they should go through unfragmented).
395 if (sk->sk_state == TCP_LISTEN)
398 if (!ip6_sk_accept_pmtu(sk))
401 tp->mtu_info = ntohl(info);
402 if (!sock_owned_by_user(sk))
403 tcp_v6_mtu_reduced(sk);
404 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
410 icmpv6_err_convert(type, code, &err);
412 /* Might be for an request_sock */
413 switch (sk->sk_state) {
414 struct request_sock *req, **prev;
416 if (sock_owned_by_user(sk))
419 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
420 &hdr->saddr, inet6_iif(skb));
424 /* ICMPs are not backlogged, hence we cannot get
425 * an established socket here.
427 WARN_ON(req->sk != NULL);
429 if (seq != tcp_rsk(req)->snt_isn) {
430 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
434 inet_csk_reqsk_queue_drop(sk, req, prev);
435 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
439 case TCP_SYN_RECV: /* Cannot happen.
440 It can, it SYNs are crossed. --ANK */
441 if (!sock_owned_by_user(sk)) {
443 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
447 sk->sk_err_soft = err;
451 if (!sock_owned_by_user(sk) && np->recverr) {
453 sk->sk_error_report(sk);
455 sk->sk_err_soft = err;
463 static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
465 struct request_sock *req,
468 struct inet_request_sock *ireq = inet_rsk(req);
469 struct ipv6_pinfo *np = inet6_sk(sk);
473 /* First, grab a route. */
474 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
477 skb = tcp_make_synack(sk, dst, req, NULL);
480 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
481 &ireq->ir_v6_rmt_addr);
483 fl6->daddr = ireq->ir_v6_rmt_addr;
484 if (np->repflow && (ireq->pktopts != NULL))
485 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
487 skb_set_queue_mapping(skb, queue_mapping);
488 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
489 err = net_xmit_eval(err);
496 static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req)
501 res = tcp_v6_send_synack(sk, NULL, &fl6, req, 0);
503 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
504 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
509 static void tcp_v6_reqsk_destructor(struct request_sock *req)
511 kfree_skb(inet_rsk(req)->pktopts);
514 #ifdef CONFIG_TCP_MD5SIG
515 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
516 const struct in6_addr *addr)
518 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
521 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
522 struct sock *addr_sk)
524 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
527 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
528 struct request_sock *req)
530 return tcp_v6_md5_do_lookup(sk, &inet_rsk(req)->ir_v6_rmt_addr);
533 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
536 struct tcp_md5sig cmd;
537 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
539 if (optlen < sizeof(cmd))
542 if (copy_from_user(&cmd, optval, sizeof(cmd)))
545 if (sin6->sin6_family != AF_INET6)
548 if (!cmd.tcpm_keylen) {
549 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
550 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
552 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
556 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
559 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
560 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
561 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
563 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
564 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
567 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
568 const struct in6_addr *daddr,
569 const struct in6_addr *saddr, int nbytes)
571 struct tcp6_pseudohdr *bp;
572 struct scatterlist sg;
574 bp = &hp->md5_blk.ip6;
575 /* 1. TCP pseudo-header (RFC2460) */
578 bp->protocol = cpu_to_be32(IPPROTO_TCP);
579 bp->len = cpu_to_be32(nbytes);
581 sg_init_one(&sg, bp, sizeof(*bp));
582 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
585 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
586 const struct in6_addr *daddr, struct in6_addr *saddr,
587 const struct tcphdr *th)
589 struct tcp_md5sig_pool *hp;
590 struct hash_desc *desc;
592 hp = tcp_get_md5sig_pool();
594 goto clear_hash_noput;
595 desc = &hp->md5_desc;
597 if (crypto_hash_init(desc))
599 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
601 if (tcp_md5_hash_header(hp, th))
603 if (tcp_md5_hash_key(hp, key))
605 if (crypto_hash_final(desc, md5_hash))
608 tcp_put_md5sig_pool();
612 tcp_put_md5sig_pool();
614 memset(md5_hash, 0, 16);
618 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
619 const struct sock *sk,
620 const struct request_sock *req,
621 const struct sk_buff *skb)
623 const struct in6_addr *saddr, *daddr;
624 struct tcp_md5sig_pool *hp;
625 struct hash_desc *desc;
626 const struct tcphdr *th = tcp_hdr(skb);
629 saddr = &inet6_sk(sk)->saddr;
630 daddr = &sk->sk_v6_daddr;
632 saddr = &inet_rsk(req)->ir_v6_loc_addr;
633 daddr = &inet_rsk(req)->ir_v6_rmt_addr;
635 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
636 saddr = &ip6h->saddr;
637 daddr = &ip6h->daddr;
640 hp = tcp_get_md5sig_pool();
642 goto clear_hash_noput;
643 desc = &hp->md5_desc;
645 if (crypto_hash_init(desc))
648 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
650 if (tcp_md5_hash_header(hp, th))
652 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
654 if (tcp_md5_hash_key(hp, key))
656 if (crypto_hash_final(desc, md5_hash))
659 tcp_put_md5sig_pool();
663 tcp_put_md5sig_pool();
665 memset(md5_hash, 0, 16);
669 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
671 const __u8 *hash_location = NULL;
672 struct tcp_md5sig_key *hash_expected;
673 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
674 const struct tcphdr *th = tcp_hdr(skb);
678 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
679 hash_location = tcp_parse_md5sig_option(th);
681 /* We've parsed the options - do we have a hash? */
682 if (!hash_expected && !hash_location)
685 if (hash_expected && !hash_location) {
686 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
690 if (!hash_expected && hash_location) {
691 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
695 /* check the signature */
696 genhash = tcp_v6_md5_hash_skb(newhash,
700 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
701 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
702 genhash ? "failed" : "mismatch",
703 &ip6h->saddr, ntohs(th->source),
704 &ip6h->daddr, ntohs(th->dest));
711 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
713 .obj_size = sizeof(struct tcp6_request_sock),
714 .rtx_syn_ack = tcp_v6_rtx_synack,
715 .send_ack = tcp_v6_reqsk_send_ack,
716 .destructor = tcp_v6_reqsk_destructor,
717 .send_reset = tcp_v6_send_reset,
718 .syn_ack_timeout = tcp_syn_ack_timeout,
721 #ifdef CONFIG_TCP_MD5SIG
722 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
723 .md5_lookup = tcp_v6_reqsk_md5_lookup,
724 .calc_md5_hash = tcp_v6_md5_hash_skb,
728 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
729 u32 tsval, u32 tsecr, int oif,
730 struct tcp_md5sig_key *key, int rst, u8 tclass,
733 const struct tcphdr *th = tcp_hdr(skb);
735 struct sk_buff *buff;
737 struct net *net = dev_net(skb_dst(skb)->dev);
738 struct sock *ctl_sk = net->ipv6.tcp_sk;
739 unsigned int tot_len = sizeof(struct tcphdr);
740 struct dst_entry *dst;
744 tot_len += TCPOLEN_TSTAMP_ALIGNED;
745 #ifdef CONFIG_TCP_MD5SIG
747 tot_len += TCPOLEN_MD5SIG_ALIGNED;
750 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
755 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
757 t1 = (struct tcphdr *) skb_push(buff, tot_len);
758 skb_reset_transport_header(buff);
760 /* Swap the send and the receive. */
761 memset(t1, 0, sizeof(*t1));
762 t1->dest = th->source;
763 t1->source = th->dest;
764 t1->doff = tot_len / 4;
765 t1->seq = htonl(seq);
766 t1->ack_seq = htonl(ack);
767 t1->ack = !rst || !th->ack;
769 t1->window = htons(win);
771 topt = (__be32 *)(t1 + 1);
774 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
775 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
776 *topt++ = htonl(tsval);
777 *topt++ = htonl(tsecr);
780 #ifdef CONFIG_TCP_MD5SIG
782 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
783 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
784 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
785 &ipv6_hdr(skb)->saddr,
786 &ipv6_hdr(skb)->daddr, t1);
790 memset(&fl6, 0, sizeof(fl6));
791 fl6.daddr = ipv6_hdr(skb)->saddr;
792 fl6.saddr = ipv6_hdr(skb)->daddr;
793 fl6.flowlabel = label;
795 buff->ip_summed = CHECKSUM_PARTIAL;
798 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
800 fl6.flowi6_proto = IPPROTO_TCP;
801 if (rt6_need_strict(&fl6.daddr) && !oif)
802 fl6.flowi6_oif = inet6_iif(skb);
804 fl6.flowi6_oif = oif;
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(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 (!ipv6_unicast_destination(skb))
846 #ifdef CONFIG_TCP_MD5SIG
847 hash_location = tcp_parse_md5sig_option(th);
848 if (!sk && hash_location) {
850 * active side is lost. Try to find listening socket through
851 * source port, and then find md5 key through listening socket.
852 * we are not loose security here:
853 * Incoming packet is checked with md5 hash with finding key,
854 * no RST generated if md5 hash doesn't match.
856 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
857 &tcp_hashinfo, &ipv6h->saddr,
858 th->source, &ipv6h->daddr,
859 ntohs(th->source), inet6_iif(skb));
864 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
868 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
869 if (genhash || memcmp(hash_location, newhash, 16) != 0)
872 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
877 seq = ntohl(th->ack_seq);
879 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
882 oif = sk ? sk->sk_bound_dev_if : 0;
883 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
885 #ifdef CONFIG_TCP_MD5SIG
894 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
895 u32 win, u32 tsval, u32 tsecr, int oif,
896 struct tcp_md5sig_key *key, u8 tclass,
899 tcp_v6_send_response(skb, seq, ack, win, tsval, tsecr, oif, key, 0, tclass,
903 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
905 struct inet_timewait_sock *tw = inet_twsk(sk);
906 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
908 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
909 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
910 tcp_time_stamp + tcptw->tw_ts_offset,
911 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
912 tw->tw_tclass, (tw->tw_flowlabel << 12));
917 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
918 struct request_sock *req)
920 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1,
921 req->rcv_wnd, tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
922 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
927 static struct sock *tcp_v6_hnd_req(struct sock *sk, struct sk_buff *skb)
929 struct request_sock *req, **prev;
930 const struct tcphdr *th = tcp_hdr(skb);
933 /* Find possible connection requests. */
934 req = inet6_csk_search_req(sk, &prev, th->source,
935 &ipv6_hdr(skb)->saddr,
936 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
938 return tcp_check_req(sk, skb, req, prev, false);
940 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
941 &ipv6_hdr(skb)->saddr, th->source,
942 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
945 if (nsk->sk_state != TCP_TIME_WAIT) {
949 inet_twsk_put(inet_twsk(nsk));
953 #ifdef CONFIG_SYN_COOKIES
955 sk = cookie_v6_check(sk, skb);
960 /* FIXME: this is substantially similar to the ipv4 code.
961 * Can some kind of merge be done? -- erics
963 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
965 struct tcp_options_received tmp_opt;
966 struct request_sock *req;
967 struct inet_request_sock *ireq;
968 struct ipv6_pinfo *np = inet6_sk(sk);
969 struct tcp_sock *tp = tcp_sk(sk);
970 __u32 isn = TCP_SKB_CB(skb)->when;
971 struct dst_entry *dst = NULL;
973 bool want_cookie = false;
975 if (skb->protocol == htons(ETH_P_IP))
976 return tcp_v4_conn_request(sk, skb);
978 if (!ipv6_unicast_destination(skb))
981 if ((sysctl_tcp_syncookies == 2 ||
982 inet_csk_reqsk_queue_is_full(sk)) && !isn) {
983 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
988 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
989 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
993 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
997 #ifdef CONFIG_TCP_MD5SIG
998 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1001 tcp_clear_options(&tmp_opt);
1002 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1003 tmp_opt.user_mss = tp->rx_opt.user_mss;
1004 tcp_parse_options(skb, &tmp_opt, 0, NULL);
1006 if (want_cookie && !tmp_opt.saw_tstamp)
1007 tcp_clear_options(&tmp_opt);
1009 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1010 tcp_openreq_init(req, &tmp_opt, skb);
1012 ireq = inet_rsk(req);
1013 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
1014 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
1015 if (!want_cookie || tmp_opt.tstamp_ok)
1016 TCP_ECN_create_request(req, skb, sock_net(sk));
1018 ireq->ir_iif = sk->sk_bound_dev_if;
1020 /* So that link locals have meaning */
1021 if (!sk->sk_bound_dev_if &&
1022 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
1023 ireq->ir_iif = inet6_iif(skb);
1026 if (ipv6_opt_accepted(sk, skb) ||
1027 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1028 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim ||
1030 atomic_inc(&skb->users);
1031 ireq->pktopts = skb;
1035 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1036 req->cookie_ts = tmp_opt.tstamp_ok;
1040 /* VJ's idea. We save last timestamp seen
1041 * from the destination in peer table, when entering
1042 * state TIME-WAIT, and check against it before
1043 * accepting new connection request.
1045 * If "isn" is not zero, this request hit alive
1046 * timewait bucket, so that all the necessary checks
1047 * are made in the function processing timewait state.
1049 if (tmp_opt.saw_tstamp &&
1050 tcp_death_row.sysctl_tw_recycle &&
1051 (dst = inet6_csk_route_req(sk, &fl6, req)) != NULL) {
1052 if (!tcp_peer_is_proven(req, dst, true)) {
1053 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1054 goto drop_and_release;
1057 /* Kill the following clause, if you dislike this way. */
1058 else if (!sysctl_tcp_syncookies &&
1059 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1060 (sysctl_max_syn_backlog >> 2)) &&
1061 !tcp_peer_is_proven(req, dst, false)) {
1062 /* Without syncookies last quarter of
1063 * backlog is filled with destinations,
1064 * proven to be alive.
1065 * It means that we continue to communicate
1066 * to destinations, already remembered
1067 * to the moment of synflood.
1069 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1070 &ireq->ir_v6_rmt_addr, ntohs(tcp_hdr(skb)->source));
1071 goto drop_and_release;
1074 isn = tcp_v6_init_sequence(skb);
1077 tcp_rsk(req)->snt_isn = isn;
1079 if (security_inet_conn_request(sk, skb, req))
1080 goto drop_and_release;
1082 if (tcp_v6_send_synack(sk, dst, &fl6, req,
1083 skb_get_queue_mapping(skb)) ||
1087 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1088 tcp_rsk(req)->listener = NULL;
1089 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1097 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1098 return 0; /* don't send reset */
1101 static struct sock *tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1102 struct request_sock *req,
1103 struct dst_entry *dst)
1105 struct inet_request_sock *ireq;
1106 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1107 struct tcp6_sock *newtcp6sk;
1108 struct inet_sock *newinet;
1109 struct tcp_sock *newtp;
1111 #ifdef CONFIG_TCP_MD5SIG
1112 struct tcp_md5sig_key *key;
1116 if (skb->protocol == htons(ETH_P_IP)) {
1121 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1126 newtcp6sk = (struct tcp6_sock *)newsk;
1127 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1129 newinet = inet_sk(newsk);
1130 newnp = inet6_sk(newsk);
1131 newtp = tcp_sk(newsk);
1133 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1135 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newsk->sk_v6_daddr);
1137 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1139 newsk->sk_v6_rcv_saddr = newnp->saddr;
1141 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1142 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1143 #ifdef CONFIG_TCP_MD5SIG
1144 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1147 newnp->ipv6_ac_list = NULL;
1148 newnp->ipv6_fl_list = NULL;
1149 newnp->pktoptions = NULL;
1151 newnp->mcast_oif = inet6_iif(skb);
1152 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1153 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1155 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1158 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1159 * here, tcp_create_openreq_child now does this for us, see the comment in
1160 * that function for the gory details. -acme
1163 /* It is tricky place. Until this moment IPv4 tcp
1164 worked with IPv6 icsk.icsk_af_ops.
1167 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1172 ireq = inet_rsk(req);
1174 if (sk_acceptq_is_full(sk))
1178 dst = inet6_csk_route_req(sk, &fl6, req);
1183 newsk = tcp_create_openreq_child(sk, req, skb);
1188 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1189 * count here, tcp_create_openreq_child now does this for us, see the
1190 * comment in that function for the gory details. -acme
1193 newsk->sk_gso_type = SKB_GSO_TCPV6;
1194 __ip6_dst_store(newsk, dst, NULL, NULL);
1195 inet6_sk_rx_dst_set(newsk, skb);
1197 newtcp6sk = (struct tcp6_sock *)newsk;
1198 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1200 newtp = tcp_sk(newsk);
1201 newinet = inet_sk(newsk);
1202 newnp = inet6_sk(newsk);
1204 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1206 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1207 newnp->saddr = ireq->ir_v6_loc_addr;
1208 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1209 newsk->sk_bound_dev_if = ireq->ir_iif;
1211 /* Now IPv6 options...
1213 First: no IPv4 options.
1215 newinet->inet_opt = NULL;
1216 newnp->ipv6_ac_list = NULL;
1217 newnp->ipv6_fl_list = NULL;
1220 newnp->rxopt.all = np->rxopt.all;
1222 /* Clone pktoptions received with SYN */
1223 newnp->pktoptions = NULL;
1224 if (ireq->pktopts != NULL) {
1225 newnp->pktoptions = skb_clone(ireq->pktopts,
1226 sk_gfp_atomic(sk, GFP_ATOMIC));
1227 consume_skb(ireq->pktopts);
1228 ireq->pktopts = NULL;
1229 if (newnp->pktoptions)
1230 skb_set_owner_r(newnp->pktoptions, newsk);
1233 newnp->mcast_oif = inet6_iif(skb);
1234 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1235 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1237 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1239 /* Clone native IPv6 options from listening socket (if any)
1241 Yes, keeping reference count would be much more clever,
1242 but we make one more one thing there: reattach optmem
1246 newnp->opt = ipv6_dup_options(newsk, np->opt);
1248 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1250 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1251 newnp->opt->opt_flen);
1253 tcp_sync_mss(newsk, dst_mtu(dst));
1254 newtp->advmss = dst_metric_advmss(dst);
1255 if (tcp_sk(sk)->rx_opt.user_mss &&
1256 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1257 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1259 tcp_initialize_rcv_mss(newsk);
1261 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1262 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1264 #ifdef CONFIG_TCP_MD5SIG
1265 /* Copy over the MD5 key from the original socket */
1266 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1268 /* We're using one, so create a matching key
1269 * on the newsk structure. If we fail to get
1270 * memory, then we end up not copying the key
1273 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1274 AF_INET6, key->key, key->keylen,
1275 sk_gfp_atomic(sk, GFP_ATOMIC));
1279 if (__inet_inherit_port(sk, newsk) < 0) {
1280 inet_csk_prepare_forced_close(newsk);
1284 __inet6_hash(newsk, NULL);
1289 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1293 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1297 /* The socket must have it's spinlock held when we get
1300 * We have a potential double-lock case here, so even when
1301 * doing backlog processing we use the BH locking scheme.
1302 * This is because we cannot sleep with the original spinlock
1305 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1307 struct ipv6_pinfo *np = inet6_sk(sk);
1308 struct tcp_sock *tp;
1309 struct sk_buff *opt_skb = NULL;
1311 /* Imagine: socket is IPv6. IPv4 packet arrives,
1312 goes to IPv4 receive handler and backlogged.
1313 From backlog it always goes here. Kerboom...
1314 Fortunately, tcp_rcv_established and rcv_established
1315 handle them correctly, but it is not case with
1316 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1319 if (skb->protocol == htons(ETH_P_IP))
1320 return tcp_v4_do_rcv(sk, skb);
1322 #ifdef CONFIG_TCP_MD5SIG
1323 if (tcp_v6_inbound_md5_hash(sk, skb))
1327 if (sk_filter(sk, skb))
1331 * socket locking is here for SMP purposes as backlog rcv
1332 * is currently called with bh processing disabled.
1335 /* Do Stevens' IPV6_PKTOPTIONS.
1337 Yes, guys, it is the only place in our code, where we
1338 may make it not affecting IPv4.
1339 The rest of code is protocol independent,
1340 and I do not like idea to uglify IPv4.
1342 Actually, all the idea behind IPV6_PKTOPTIONS
1343 looks not very well thought. For now we latch
1344 options, received in the last packet, enqueued
1345 by tcp. Feel free to propose better solution.
1349 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1351 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1352 struct dst_entry *dst = sk->sk_rx_dst;
1354 sock_rps_save_rxhash(sk, skb);
1356 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1357 dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1359 sk->sk_rx_dst = NULL;
1363 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1365 goto ipv6_pktoptions;
1369 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1372 if (sk->sk_state == TCP_LISTEN) {
1373 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1378 * Queue it on the new socket if the new socket is active,
1379 * otherwise we just shortcircuit this and continue with
1383 sock_rps_save_rxhash(nsk, skb);
1384 if (tcp_child_process(sk, nsk, skb))
1387 __kfree_skb(opt_skb);
1391 sock_rps_save_rxhash(sk, skb);
1393 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1396 goto ipv6_pktoptions;
1400 tcp_v6_send_reset(sk, skb);
1403 __kfree_skb(opt_skb);
1407 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1408 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1413 /* Do you ask, what is it?
1415 1. skb was enqueued by tcp.
1416 2. skb is added to tail of read queue, rather than out of order.
1417 3. socket is not in passive state.
1418 4. Finally, it really contains options, which user wants to receive.
1421 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1422 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1423 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1424 np->mcast_oif = inet6_iif(opt_skb);
1425 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1426 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1427 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1428 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1430 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1431 if (ipv6_opt_accepted(sk, opt_skb)) {
1432 skb_set_owner_r(opt_skb, sk);
1433 opt_skb = xchg(&np->pktoptions, opt_skb);
1435 __kfree_skb(opt_skb);
1436 opt_skb = xchg(&np->pktoptions, NULL);
1444 static int tcp_v6_rcv(struct sk_buff *skb)
1446 const struct tcphdr *th;
1447 const struct ipv6hdr *hdr;
1450 struct net *net = dev_net(skb->dev);
1452 if (skb->pkt_type != PACKET_HOST)
1456 * Count it even if it's bad.
1458 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1460 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1465 if (th->doff < sizeof(struct tcphdr)/4)
1467 if (!pskb_may_pull(skb, th->doff*4))
1470 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1474 hdr = ipv6_hdr(skb);
1475 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1476 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1477 skb->len - th->doff*4);
1478 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1479 TCP_SKB_CB(skb)->when = 0;
1480 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1481 TCP_SKB_CB(skb)->sacked = 0;
1483 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1488 if (sk->sk_state == TCP_TIME_WAIT)
1491 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1492 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1493 goto discard_and_relse;
1496 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1497 goto discard_and_relse;
1499 if (sk_filter(sk, skb))
1500 goto discard_and_relse;
1502 sk_mark_napi_id(sk, skb);
1505 bh_lock_sock_nested(sk);
1507 if (!sock_owned_by_user(sk)) {
1508 #ifdef CONFIG_NET_DMA
1509 struct tcp_sock *tp = tcp_sk(sk);
1510 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1511 tp->ucopy.dma_chan = net_dma_find_channel();
1512 if (tp->ucopy.dma_chan)
1513 ret = tcp_v6_do_rcv(sk, skb);
1517 if (!tcp_prequeue(sk, skb))
1518 ret = tcp_v6_do_rcv(sk, skb);
1520 } else if (unlikely(sk_add_backlog(sk, skb,
1521 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1523 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1524 goto discard_and_relse;
1529 return ret ? -1 : 0;
1532 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1535 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1537 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1539 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1541 tcp_v6_send_reset(NULL, skb);
1553 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1554 inet_twsk_put(inet_twsk(sk));
1558 if (skb->len < (th->doff<<2)) {
1559 inet_twsk_put(inet_twsk(sk));
1562 if (tcp_checksum_complete(skb)) {
1563 inet_twsk_put(inet_twsk(sk));
1567 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1572 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1573 &ipv6_hdr(skb)->saddr, th->source,
1574 &ipv6_hdr(skb)->daddr,
1575 ntohs(th->dest), inet6_iif(skb));
1577 struct inet_timewait_sock *tw = inet_twsk(sk);
1578 inet_twsk_deschedule(tw, &tcp_death_row);
1583 /* Fall through to ACK */
1586 tcp_v6_timewait_ack(sk, skb);
1590 case TCP_TW_SUCCESS:
1596 static void tcp_v6_early_demux(struct sk_buff *skb)
1598 const struct ipv6hdr *hdr;
1599 const struct tcphdr *th;
1602 if (skb->pkt_type != PACKET_HOST)
1605 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1608 hdr = ipv6_hdr(skb);
1611 if (th->doff < sizeof(struct tcphdr) / 4)
1614 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1615 &hdr->saddr, th->source,
1616 &hdr->daddr, ntohs(th->dest),
1620 skb->destructor = sock_edemux;
1621 if (sk->sk_state != TCP_TIME_WAIT) {
1622 struct dst_entry *dst = sk->sk_rx_dst;
1625 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1627 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1628 skb_dst_set_noref(skb, dst);
1633 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1634 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1635 .twsk_unique = tcp_twsk_unique,
1636 .twsk_destructor = tcp_twsk_destructor,
1639 static const struct inet_connection_sock_af_ops ipv6_specific = {
1640 .queue_xmit = inet6_csk_xmit,
1641 .send_check = tcp_v6_send_check,
1642 .rebuild_header = inet6_sk_rebuild_header,
1643 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1644 .conn_request = tcp_v6_conn_request,
1645 .syn_recv_sock = tcp_v6_syn_recv_sock,
1646 .net_header_len = sizeof(struct ipv6hdr),
1647 .net_frag_header_len = sizeof(struct frag_hdr),
1648 .setsockopt = ipv6_setsockopt,
1649 .getsockopt = ipv6_getsockopt,
1650 .addr2sockaddr = inet6_csk_addr2sockaddr,
1651 .sockaddr_len = sizeof(struct sockaddr_in6),
1652 .bind_conflict = inet6_csk_bind_conflict,
1653 #ifdef CONFIG_COMPAT
1654 .compat_setsockopt = compat_ipv6_setsockopt,
1655 .compat_getsockopt = compat_ipv6_getsockopt,
1659 #ifdef CONFIG_TCP_MD5SIG
1660 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1661 .md5_lookup = tcp_v6_md5_lookup,
1662 .calc_md5_hash = tcp_v6_md5_hash_skb,
1663 .md5_parse = tcp_v6_parse_md5_keys,
1668 * TCP over IPv4 via INET6 API
1670 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1671 .queue_xmit = ip_queue_xmit,
1672 .send_check = tcp_v4_send_check,
1673 .rebuild_header = inet_sk_rebuild_header,
1674 .sk_rx_dst_set = inet_sk_rx_dst_set,
1675 .conn_request = tcp_v6_conn_request,
1676 .syn_recv_sock = tcp_v6_syn_recv_sock,
1677 .net_header_len = sizeof(struct iphdr),
1678 .setsockopt = ipv6_setsockopt,
1679 .getsockopt = ipv6_getsockopt,
1680 .addr2sockaddr = inet6_csk_addr2sockaddr,
1681 .sockaddr_len = sizeof(struct sockaddr_in6),
1682 .bind_conflict = inet6_csk_bind_conflict,
1683 #ifdef CONFIG_COMPAT
1684 .compat_setsockopt = compat_ipv6_setsockopt,
1685 .compat_getsockopt = compat_ipv6_getsockopt,
1689 #ifdef CONFIG_TCP_MD5SIG
1690 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1691 .md5_lookup = tcp_v4_md5_lookup,
1692 .calc_md5_hash = tcp_v4_md5_hash_skb,
1693 .md5_parse = tcp_v6_parse_md5_keys,
1697 /* NOTE: A lot of things set to zero explicitly by call to
1698 * sk_alloc() so need not be done here.
1700 static int tcp_v6_init_sock(struct sock *sk)
1702 struct inet_connection_sock *icsk = inet_csk(sk);
1706 icsk->icsk_af_ops = &ipv6_specific;
1708 #ifdef CONFIG_TCP_MD5SIG
1709 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1715 static void tcp_v6_destroy_sock(struct sock *sk)
1717 tcp_v4_destroy_sock(sk);
1718 inet6_destroy_sock(sk);
1721 #ifdef CONFIG_PROC_FS
1722 /* Proc filesystem TCPv6 sock list dumping. */
1723 static void get_openreq6(struct seq_file *seq,
1724 const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1726 int ttd = req->expires - jiffies;
1727 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1728 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1734 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1735 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1737 src->s6_addr32[0], src->s6_addr32[1],
1738 src->s6_addr32[2], src->s6_addr32[3],
1739 inet_rsk(req)->ir_num,
1740 dest->s6_addr32[0], dest->s6_addr32[1],
1741 dest->s6_addr32[2], dest->s6_addr32[3],
1742 ntohs(inet_rsk(req)->ir_rmt_port),
1744 0, 0, /* could print option size, but that is af dependent. */
1745 1, /* timers active (only the expire timer) */
1746 jiffies_to_clock_t(ttd),
1748 from_kuid_munged(seq_user_ns(seq), uid),
1749 0, /* non standard timer */
1750 0, /* open_requests have no inode */
1754 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1756 const struct in6_addr *dest, *src;
1759 unsigned long timer_expires;
1760 const struct inet_sock *inet = inet_sk(sp);
1761 const struct tcp_sock *tp = tcp_sk(sp);
1762 const struct inet_connection_sock *icsk = inet_csk(sp);
1764 dest = &sp->sk_v6_daddr;
1765 src = &sp->sk_v6_rcv_saddr;
1766 destp = ntohs(inet->inet_dport);
1767 srcp = ntohs(inet->inet_sport);
1769 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1771 timer_expires = icsk->icsk_timeout;
1772 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1774 timer_expires = icsk->icsk_timeout;
1775 } else if (timer_pending(&sp->sk_timer)) {
1777 timer_expires = sp->sk_timer.expires;
1780 timer_expires = jiffies;
1784 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1785 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1787 src->s6_addr32[0], src->s6_addr32[1],
1788 src->s6_addr32[2], src->s6_addr32[3], srcp,
1789 dest->s6_addr32[0], dest->s6_addr32[1],
1790 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1792 tp->write_seq-tp->snd_una,
1793 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1795 jiffies_delta_to_clock_t(timer_expires - jiffies),
1796 icsk->icsk_retransmits,
1797 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1798 icsk->icsk_probes_out,
1800 atomic_read(&sp->sk_refcnt), sp,
1801 jiffies_to_clock_t(icsk->icsk_rto),
1802 jiffies_to_clock_t(icsk->icsk_ack.ato),
1803 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1805 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
1809 static void get_timewait6_sock(struct seq_file *seq,
1810 struct inet_timewait_sock *tw, int i)
1812 const struct in6_addr *dest, *src;
1814 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
1816 dest = &tw->tw_v6_daddr;
1817 src = &tw->tw_v6_rcv_saddr;
1818 destp = ntohs(tw->tw_dport);
1819 srcp = ntohs(tw->tw_sport);
1822 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1823 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1825 src->s6_addr32[0], src->s6_addr32[1],
1826 src->s6_addr32[2], src->s6_addr32[3], srcp,
1827 dest->s6_addr32[0], dest->s6_addr32[1],
1828 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1829 tw->tw_substate, 0, 0,
1830 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1831 atomic_read(&tw->tw_refcnt), tw);
1834 static int tcp6_seq_show(struct seq_file *seq, void *v)
1836 struct tcp_iter_state *st;
1837 struct sock *sk = v;
1839 if (v == SEQ_START_TOKEN) {
1844 "st tx_queue rx_queue tr tm->when retrnsmt"
1845 " uid timeout inode\n");
1850 switch (st->state) {
1851 case TCP_SEQ_STATE_LISTENING:
1852 case TCP_SEQ_STATE_ESTABLISHED:
1853 if (sk->sk_state == TCP_TIME_WAIT)
1854 get_timewait6_sock(seq, v, st->num);
1856 get_tcp6_sock(seq, v, st->num);
1858 case TCP_SEQ_STATE_OPENREQ:
1859 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1866 static const struct file_operations tcp6_afinfo_seq_fops = {
1867 .owner = THIS_MODULE,
1868 .open = tcp_seq_open,
1870 .llseek = seq_lseek,
1871 .release = seq_release_net
1874 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1877 .seq_fops = &tcp6_afinfo_seq_fops,
1879 .show = tcp6_seq_show,
1883 int __net_init tcp6_proc_init(struct net *net)
1885 return tcp_proc_register(net, &tcp6_seq_afinfo);
1888 void tcp6_proc_exit(struct net *net)
1890 tcp_proc_unregister(net, &tcp6_seq_afinfo);
1894 static void tcp_v6_clear_sk(struct sock *sk, int size)
1896 struct inet_sock *inet = inet_sk(sk);
1898 /* we do not want to clear pinet6 field, because of RCU lookups */
1899 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1901 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1902 memset(&inet->pinet6 + 1, 0, size);
1905 struct proto tcpv6_prot = {
1907 .owner = THIS_MODULE,
1909 .connect = tcp_v6_connect,
1910 .disconnect = tcp_disconnect,
1911 .accept = inet_csk_accept,
1913 .init = tcp_v6_init_sock,
1914 .destroy = tcp_v6_destroy_sock,
1915 .shutdown = tcp_shutdown,
1916 .setsockopt = tcp_setsockopt,
1917 .getsockopt = tcp_getsockopt,
1918 .recvmsg = tcp_recvmsg,
1919 .sendmsg = tcp_sendmsg,
1920 .sendpage = tcp_sendpage,
1921 .backlog_rcv = tcp_v6_do_rcv,
1922 .release_cb = tcp_release_cb,
1923 .mtu_reduced = tcp_v6_mtu_reduced,
1924 .hash = tcp_v6_hash,
1925 .unhash = inet_unhash,
1926 .get_port = inet_csk_get_port,
1927 .enter_memory_pressure = tcp_enter_memory_pressure,
1928 .stream_memory_free = tcp_stream_memory_free,
1929 .sockets_allocated = &tcp_sockets_allocated,
1930 .memory_allocated = &tcp_memory_allocated,
1931 .memory_pressure = &tcp_memory_pressure,
1932 .orphan_count = &tcp_orphan_count,
1933 .sysctl_mem = sysctl_tcp_mem,
1934 .sysctl_wmem = sysctl_tcp_wmem,
1935 .sysctl_rmem = sysctl_tcp_rmem,
1936 .max_header = MAX_TCP_HEADER,
1937 .obj_size = sizeof(struct tcp6_sock),
1938 .slab_flags = SLAB_DESTROY_BY_RCU,
1939 .twsk_prot = &tcp6_timewait_sock_ops,
1940 .rsk_prot = &tcp6_request_sock_ops,
1941 .h.hashinfo = &tcp_hashinfo,
1942 .no_autobind = true,
1943 #ifdef CONFIG_COMPAT
1944 .compat_setsockopt = compat_tcp_setsockopt,
1945 .compat_getsockopt = compat_tcp_getsockopt,
1947 #ifdef CONFIG_MEMCG_KMEM
1948 .proto_cgroup = tcp_proto_cgroup,
1950 .clear_sk = tcp_v6_clear_sk,
1953 static const struct inet6_protocol tcpv6_protocol = {
1954 .early_demux = tcp_v6_early_demux,
1955 .handler = tcp_v6_rcv,
1956 .err_handler = tcp_v6_err,
1957 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1960 static struct inet_protosw tcpv6_protosw = {
1961 .type = SOCK_STREAM,
1962 .protocol = IPPROTO_TCP,
1963 .prot = &tcpv6_prot,
1964 .ops = &inet6_stream_ops,
1966 .flags = INET_PROTOSW_PERMANENT |
1970 static int __net_init tcpv6_net_init(struct net *net)
1972 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1973 SOCK_RAW, IPPROTO_TCP, net);
1976 static void __net_exit tcpv6_net_exit(struct net *net)
1978 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1981 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1983 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1986 static struct pernet_operations tcpv6_net_ops = {
1987 .init = tcpv6_net_init,
1988 .exit = tcpv6_net_exit,
1989 .exit_batch = tcpv6_net_exit_batch,
1992 int __init tcpv6_init(void)
1996 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2000 /* register inet6 protocol */
2001 ret = inet6_register_protosw(&tcpv6_protosw);
2003 goto out_tcpv6_protocol;
2005 ret = register_pernet_subsys(&tcpv6_net_ops);
2007 goto out_tcpv6_protosw;
2012 inet6_unregister_protosw(&tcpv6_protosw);
2014 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2018 void tcpv6_exit(void)
2020 unregister_pernet_subsys(&tcpv6_net_ops);
2021 inet6_unregister_protosw(&tcpv6_protosw);
2022 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);