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>
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>
66 #include <asm/uaccess.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);
79 static void __tcp_v6_send_check(struct sk_buff *skb,
80 const struct in6_addr *saddr,
81 const struct in6_addr *daddr);
83 static const struct inet_connection_sock_af_ops ipv6_mapped;
84 static const struct inet_connection_sock_af_ops ipv6_specific;
85 #ifdef CONFIG_TCP_MD5SIG
86 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
87 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
89 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
90 const struct in6_addr *addr)
96 static void tcp_v6_hash(struct sock *sk)
98 if (sk->sk_state != TCP_CLOSE) {
99 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
104 __inet6_hash(sk, NULL);
109 static __inline__ __sum16 tcp_v6_check(int len,
110 const struct in6_addr *saddr,
111 const struct in6_addr *daddr,
114 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
117 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
119 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
120 ipv6_hdr(skb)->saddr.s6_addr32,
122 tcp_hdr(skb)->source);
125 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
128 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
129 struct inet_sock *inet = inet_sk(sk);
130 struct inet_connection_sock *icsk = inet_csk(sk);
131 struct ipv6_pinfo *np = inet6_sk(sk);
132 struct tcp_sock *tp = tcp_sk(sk);
133 struct in6_addr *saddr = NULL, *final_p, final;
136 struct dst_entry *dst;
140 if (addr_len < SIN6_LEN_RFC2133)
143 if (usin->sin6_family != AF_INET6)
144 return -EAFNOSUPPORT;
146 memset(&fl6, 0, sizeof(fl6));
149 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
150 IP6_ECN_flow_init(fl6.flowlabel);
151 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
152 struct ip6_flowlabel *flowlabel;
153 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
154 if (flowlabel == NULL)
156 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
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(&np->daddr, &usin->sin6_addr)) {
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
198 ipv6_addr_copy(&np->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,
243 if (!ipv6_addr_any(&np->rcv_saddr))
244 saddr = &np->rcv_saddr;
246 fl6.flowi6_proto = IPPROTO_TCP;
247 ipv6_addr_copy(&fl6.daddr, &np->daddr);
248 ipv6_addr_copy(&fl6.saddr,
249 (saddr ? saddr : &np->saddr));
250 fl6.flowi6_oif = sk->sk_bound_dev_if;
251 fl6.flowi6_mark = sk->sk_mark;
252 fl6.fl6_dport = usin->sin6_port;
253 fl6.fl6_sport = inet->inet_sport;
255 final_p = fl6_update_dst(&fl6, np->opt, &final);
257 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
259 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, true);
267 ipv6_addr_copy(&np->rcv_saddr, saddr);
270 /* set the source address */
271 ipv6_addr_copy(&np->saddr, saddr);
272 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
274 sk->sk_gso_type = SKB_GSO_TCPV6;
275 __ip6_dst_store(sk, dst, NULL, NULL);
277 rt = (struct rt6_info *) dst;
278 if (tcp_death_row.sysctl_tw_recycle &&
279 !tp->rx_opt.ts_recent_stamp &&
280 ipv6_addr_equal(&rt->rt6i_dst.addr, &np->daddr)) {
281 struct inet_peer *peer = rt6_get_peer(rt);
283 * VJ's idea. We save last timestamp seen from
284 * the destination in peer table, when entering state
285 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
286 * when trying new connection.
289 inet_peer_refcheck(peer);
290 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
291 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
292 tp->rx_opt.ts_recent = peer->tcp_ts;
297 icsk->icsk_ext_hdr_len = 0;
299 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
302 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
304 inet->inet_dport = usin->sin6_port;
306 tcp_set_state(sk, TCP_SYN_SENT);
307 err = inet6_hash_connect(&tcp_death_row, sk);
312 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
317 err = tcp_connect(sk);
324 tcp_set_state(sk, TCP_CLOSE);
327 inet->inet_dport = 0;
328 sk->sk_route_caps = 0;
332 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
333 u8 type, u8 code, int offset, __be32 info)
335 const struct ipv6hdr *hdr = (const struct ipv6hdr*)skb->data;
336 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
337 struct ipv6_pinfo *np;
342 struct net *net = dev_net(skb->dev);
344 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
345 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
348 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
353 if (sk->sk_state == TCP_TIME_WAIT) {
354 inet_twsk_put(inet_twsk(sk));
359 if (sock_owned_by_user(sk))
360 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
362 if (sk->sk_state == TCP_CLOSE)
365 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
366 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
371 seq = ntohl(th->seq);
372 if (sk->sk_state != TCP_LISTEN &&
373 !between(seq, tp->snd_una, tp->snd_nxt)) {
374 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
380 if (type == ICMPV6_PKT_TOOBIG) {
381 struct dst_entry *dst;
383 if (sock_owned_by_user(sk))
385 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
388 /* icmp should have updated the destination cache entry */
389 dst = __sk_dst_check(sk, np->dst_cookie);
392 struct inet_sock *inet = inet_sk(sk);
395 /* BUGGG_FUTURE: Again, it is not clear how
396 to handle rthdr case. Ignore this complexity
399 memset(&fl6, 0, sizeof(fl6));
400 fl6.flowi6_proto = IPPROTO_TCP;
401 ipv6_addr_copy(&fl6.daddr, &np->daddr);
402 ipv6_addr_copy(&fl6.saddr, &np->saddr);
403 fl6.flowi6_oif = sk->sk_bound_dev_if;
404 fl6.flowi6_mark = sk->sk_mark;
405 fl6.fl6_dport = inet->inet_dport;
406 fl6.fl6_sport = inet->inet_sport;
407 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
409 dst = ip6_dst_lookup_flow(sk, &fl6, NULL, false);
411 sk->sk_err_soft = -PTR_ERR(dst);
418 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
419 tcp_sync_mss(sk, dst_mtu(dst));
420 tcp_simple_retransmit(sk);
421 } /* else let the usual retransmit timer handle it */
426 icmpv6_err_convert(type, code, &err);
428 /* Might be for an request_sock */
429 switch (sk->sk_state) {
430 struct request_sock *req, **prev;
432 if (sock_owned_by_user(sk))
435 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
436 &hdr->saddr, inet6_iif(skb));
440 /* ICMPs are not backlogged, hence we cannot get
441 * an established socket here.
443 WARN_ON(req->sk != NULL);
445 if (seq != tcp_rsk(req)->snt_isn) {
446 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
450 inet_csk_reqsk_queue_drop(sk, req, prev);
454 case TCP_SYN_RECV: /* Cannot happen.
455 It can, it SYNs are crossed. --ANK */
456 if (!sock_owned_by_user(sk)) {
458 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
462 sk->sk_err_soft = err;
466 if (!sock_owned_by_user(sk) && np->recverr) {
468 sk->sk_error_report(sk);
470 sk->sk_err_soft = err;
478 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
479 struct request_values *rvp)
481 struct inet6_request_sock *treq = inet6_rsk(req);
482 struct ipv6_pinfo *np = inet6_sk(sk);
483 struct sk_buff * skb;
484 struct ipv6_txoptions *opt = NULL;
485 struct in6_addr * final_p, final;
487 struct dst_entry *dst;
490 memset(&fl6, 0, sizeof(fl6));
491 fl6.flowi6_proto = IPPROTO_TCP;
492 ipv6_addr_copy(&fl6.daddr, &treq->rmt_addr);
493 ipv6_addr_copy(&fl6.saddr, &treq->loc_addr);
495 fl6.flowi6_oif = treq->iif;
496 fl6.flowi6_mark = sk->sk_mark;
497 fl6.fl6_dport = inet_rsk(req)->rmt_port;
498 fl6.fl6_sport = inet_rsk(req)->loc_port;
499 security_req_classify_flow(req, flowi6_to_flowi(&fl6));
502 final_p = fl6_update_dst(&fl6, opt, &final);
504 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false);
510 skb = tcp_make_synack(sk, dst, req, rvp);
513 __tcp_v6_send_check(skb, &treq->loc_addr, &treq->rmt_addr);
515 ipv6_addr_copy(&fl6.daddr, &treq->rmt_addr);
516 err = ip6_xmit(sk, skb, &fl6, opt, np->tclass);
517 err = net_xmit_eval(err);
521 if (opt && opt != np->opt)
522 sock_kfree_s(sk, opt, opt->tot_len);
527 static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req,
528 struct request_values *rvp)
530 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
531 return tcp_v6_send_synack(sk, req, rvp);
534 static void tcp_v6_reqsk_destructor(struct request_sock *req)
536 kfree_skb(inet6_rsk(req)->pktopts);
539 #ifdef CONFIG_TCP_MD5SIG
540 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
541 const struct in6_addr *addr)
543 struct tcp_sock *tp = tcp_sk(sk);
548 if (!tp->md5sig_info || !tp->md5sig_info->entries6)
551 for (i = 0; i < tp->md5sig_info->entries6; i++) {
552 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, addr))
553 return &tp->md5sig_info->keys6[i].base;
558 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
559 struct sock *addr_sk)
561 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
564 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
565 struct request_sock *req)
567 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
570 static int tcp_v6_md5_do_add(struct sock *sk, const struct in6_addr *peer,
571 char *newkey, u8 newkeylen)
573 /* Add key to the list */
574 struct tcp_md5sig_key *key;
575 struct tcp_sock *tp = tcp_sk(sk);
576 struct tcp6_md5sig_key *keys;
578 key = tcp_v6_md5_do_lookup(sk, peer);
580 /* modify existing entry - just update that one */
583 key->keylen = newkeylen;
585 /* reallocate new list if current one is full. */
586 if (!tp->md5sig_info) {
587 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
588 if (!tp->md5sig_info) {
592 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
594 if (tp->md5sig_info->entries6 == 0 &&
595 tcp_alloc_md5sig_pool(sk) == NULL) {
599 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
600 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
601 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
605 if (tp->md5sig_info->entries6 == 0)
606 tcp_free_md5sig_pool();
610 if (tp->md5sig_info->entries6)
611 memmove(keys, tp->md5sig_info->keys6,
612 (sizeof (tp->md5sig_info->keys6[0]) *
613 tp->md5sig_info->entries6));
615 kfree(tp->md5sig_info->keys6);
616 tp->md5sig_info->keys6 = keys;
617 tp->md5sig_info->alloced6++;
620 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
622 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
623 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
625 tp->md5sig_info->entries6++;
630 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
631 u8 *newkey, __u8 newkeylen)
633 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
637 static int tcp_v6_md5_do_del(struct sock *sk, const struct in6_addr *peer)
639 struct tcp_sock *tp = tcp_sk(sk);
642 for (i = 0; i < tp->md5sig_info->entries6; i++) {
643 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) {
645 kfree(tp->md5sig_info->keys6[i].base.key);
646 tp->md5sig_info->entries6--;
648 if (tp->md5sig_info->entries6 == 0) {
649 kfree(tp->md5sig_info->keys6);
650 tp->md5sig_info->keys6 = NULL;
651 tp->md5sig_info->alloced6 = 0;
652 tcp_free_md5sig_pool();
654 /* shrink the database */
655 if (tp->md5sig_info->entries6 != i)
656 memmove(&tp->md5sig_info->keys6[i],
657 &tp->md5sig_info->keys6[i+1],
658 (tp->md5sig_info->entries6 - i)
659 * sizeof (tp->md5sig_info->keys6[0]));
667 static void tcp_v6_clear_md5_list (struct sock *sk)
669 struct tcp_sock *tp = tcp_sk(sk);
672 if (tp->md5sig_info->entries6) {
673 for (i = 0; i < tp->md5sig_info->entries6; i++)
674 kfree(tp->md5sig_info->keys6[i].base.key);
675 tp->md5sig_info->entries6 = 0;
676 tcp_free_md5sig_pool();
679 kfree(tp->md5sig_info->keys6);
680 tp->md5sig_info->keys6 = NULL;
681 tp->md5sig_info->alloced6 = 0;
683 if (tp->md5sig_info->entries4) {
684 for (i = 0; i < tp->md5sig_info->entries4; i++)
685 kfree(tp->md5sig_info->keys4[i].base.key);
686 tp->md5sig_info->entries4 = 0;
687 tcp_free_md5sig_pool();
690 kfree(tp->md5sig_info->keys4);
691 tp->md5sig_info->keys4 = NULL;
692 tp->md5sig_info->alloced4 = 0;
695 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
698 struct tcp_md5sig cmd;
699 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
702 if (optlen < sizeof(cmd))
705 if (copy_from_user(&cmd, optval, sizeof(cmd)))
708 if (sin6->sin6_family != AF_INET6)
711 if (!cmd.tcpm_keylen) {
712 if (!tcp_sk(sk)->md5sig_info)
714 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
715 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
716 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
719 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
722 if (!tcp_sk(sk)->md5sig_info) {
723 struct tcp_sock *tp = tcp_sk(sk);
724 struct tcp_md5sig_info *p;
726 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
731 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
734 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
737 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
738 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
739 newkey, cmd.tcpm_keylen);
741 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
744 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
745 const struct in6_addr *daddr,
746 const struct in6_addr *saddr, int nbytes)
748 struct tcp6_pseudohdr *bp;
749 struct scatterlist sg;
751 bp = &hp->md5_blk.ip6;
752 /* 1. TCP pseudo-header (RFC2460) */
753 ipv6_addr_copy(&bp->saddr, saddr);
754 ipv6_addr_copy(&bp->daddr, daddr);
755 bp->protocol = cpu_to_be32(IPPROTO_TCP);
756 bp->len = cpu_to_be32(nbytes);
758 sg_init_one(&sg, bp, sizeof(*bp));
759 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
762 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
763 const struct in6_addr *daddr, struct in6_addr *saddr,
764 const struct tcphdr *th)
766 struct tcp_md5sig_pool *hp;
767 struct hash_desc *desc;
769 hp = tcp_get_md5sig_pool();
771 goto clear_hash_noput;
772 desc = &hp->md5_desc;
774 if (crypto_hash_init(desc))
776 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
778 if (tcp_md5_hash_header(hp, th))
780 if (tcp_md5_hash_key(hp, key))
782 if (crypto_hash_final(desc, md5_hash))
785 tcp_put_md5sig_pool();
789 tcp_put_md5sig_pool();
791 memset(md5_hash, 0, 16);
795 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
796 const struct sock *sk,
797 const struct request_sock *req,
798 const struct sk_buff *skb)
800 const struct in6_addr *saddr, *daddr;
801 struct tcp_md5sig_pool *hp;
802 struct hash_desc *desc;
803 const struct tcphdr *th = tcp_hdr(skb);
806 saddr = &inet6_sk(sk)->saddr;
807 daddr = &inet6_sk(sk)->daddr;
809 saddr = &inet6_rsk(req)->loc_addr;
810 daddr = &inet6_rsk(req)->rmt_addr;
812 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
813 saddr = &ip6h->saddr;
814 daddr = &ip6h->daddr;
817 hp = tcp_get_md5sig_pool();
819 goto clear_hash_noput;
820 desc = &hp->md5_desc;
822 if (crypto_hash_init(desc))
825 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
827 if (tcp_md5_hash_header(hp, th))
829 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
831 if (tcp_md5_hash_key(hp, key))
833 if (crypto_hash_final(desc, md5_hash))
836 tcp_put_md5sig_pool();
840 tcp_put_md5sig_pool();
842 memset(md5_hash, 0, 16);
846 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
848 const __u8 *hash_location = NULL;
849 struct tcp_md5sig_key *hash_expected;
850 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
851 const struct tcphdr *th = tcp_hdr(skb);
855 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
856 hash_location = tcp_parse_md5sig_option(th);
858 /* We've parsed the options - do we have a hash? */
859 if (!hash_expected && !hash_location)
862 if (hash_expected && !hash_location) {
863 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
867 if (!hash_expected && hash_location) {
868 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
872 /* check the signature */
873 genhash = tcp_v6_md5_hash_skb(newhash,
877 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
878 if (net_ratelimit()) {
879 printk(KERN_INFO "MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
880 genhash ? "failed" : "mismatch",
881 &ip6h->saddr, ntohs(th->source),
882 &ip6h->daddr, ntohs(th->dest));
890 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
892 .obj_size = sizeof(struct tcp6_request_sock),
893 .rtx_syn_ack = tcp_v6_rtx_synack,
894 .send_ack = tcp_v6_reqsk_send_ack,
895 .destructor = tcp_v6_reqsk_destructor,
896 .send_reset = tcp_v6_send_reset,
897 .syn_ack_timeout = tcp_syn_ack_timeout,
900 #ifdef CONFIG_TCP_MD5SIG
901 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
902 .md5_lookup = tcp_v6_reqsk_md5_lookup,
903 .calc_md5_hash = tcp_v6_md5_hash_skb,
907 static void __tcp_v6_send_check(struct sk_buff *skb,
908 const struct in6_addr *saddr, const struct in6_addr *daddr)
910 struct tcphdr *th = tcp_hdr(skb);
912 if (skb->ip_summed == CHECKSUM_PARTIAL) {
913 th->check = ~tcp_v6_check(skb->len, saddr, daddr, 0);
914 skb->csum_start = skb_transport_header(skb) - skb->head;
915 skb->csum_offset = offsetof(struct tcphdr, check);
917 th->check = tcp_v6_check(skb->len, saddr, daddr,
918 csum_partial(th, th->doff << 2,
923 static void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
925 struct ipv6_pinfo *np = inet6_sk(sk);
927 __tcp_v6_send_check(skb, &np->saddr, &np->daddr);
930 static int tcp_v6_gso_send_check(struct sk_buff *skb)
932 const struct ipv6hdr *ipv6h;
935 if (!pskb_may_pull(skb, sizeof(*th)))
938 ipv6h = ipv6_hdr(skb);
942 skb->ip_summed = CHECKSUM_PARTIAL;
943 __tcp_v6_send_check(skb, &ipv6h->saddr, &ipv6h->daddr);
947 static struct sk_buff **tcp6_gro_receive(struct sk_buff **head,
950 const struct ipv6hdr *iph = skb_gro_network_header(skb);
952 switch (skb->ip_summed) {
953 case CHECKSUM_COMPLETE:
954 if (!tcp_v6_check(skb_gro_len(skb), &iph->saddr, &iph->daddr,
956 skb->ip_summed = CHECKSUM_UNNECESSARY;
962 NAPI_GRO_CB(skb)->flush = 1;
966 return tcp_gro_receive(head, skb);
969 static int tcp6_gro_complete(struct sk_buff *skb)
971 const struct ipv6hdr *iph = ipv6_hdr(skb);
972 struct tcphdr *th = tcp_hdr(skb);
974 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
975 &iph->saddr, &iph->daddr, 0);
976 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
978 return tcp_gro_complete(skb);
981 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
982 u32 ts, struct tcp_md5sig_key *key, int rst, u8 tclass)
984 const struct tcphdr *th = tcp_hdr(skb);
986 struct sk_buff *buff;
988 struct net *net = dev_net(skb_dst(skb)->dev);
989 struct sock *ctl_sk = net->ipv6.tcp_sk;
990 unsigned int tot_len = sizeof(struct tcphdr);
991 struct dst_entry *dst;
995 tot_len += TCPOLEN_TSTAMP_ALIGNED;
996 #ifdef CONFIG_TCP_MD5SIG
998 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1001 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1006 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1008 t1 = (struct tcphdr *) skb_push(buff, tot_len);
1009 skb_reset_transport_header(buff);
1011 /* Swap the send and the receive. */
1012 memset(t1, 0, sizeof(*t1));
1013 t1->dest = th->source;
1014 t1->source = th->dest;
1015 t1->doff = tot_len / 4;
1016 t1->seq = htonl(seq);
1017 t1->ack_seq = htonl(ack);
1018 t1->ack = !rst || !th->ack;
1020 t1->window = htons(win);
1022 topt = (__be32 *)(t1 + 1);
1025 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1026 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1027 *topt++ = htonl(tcp_time_stamp);
1028 *topt++ = htonl(ts);
1031 #ifdef CONFIG_TCP_MD5SIG
1033 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1034 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1035 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
1036 &ipv6_hdr(skb)->saddr,
1037 &ipv6_hdr(skb)->daddr, t1);
1041 memset(&fl6, 0, sizeof(fl6));
1042 ipv6_addr_copy(&fl6.daddr, &ipv6_hdr(skb)->saddr);
1043 ipv6_addr_copy(&fl6.saddr, &ipv6_hdr(skb)->daddr);
1045 buff->ip_summed = CHECKSUM_PARTIAL;
1048 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
1050 fl6.flowi6_proto = IPPROTO_TCP;
1051 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
1052 fl6.flowi6_oif = inet6_iif(skb);
1053 fl6.fl6_dport = t1->dest;
1054 fl6.fl6_sport = t1->source;
1055 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
1057 /* Pass a socket to ip6_dst_lookup either it is for RST
1058 * Underlying function will use this to retrieve the network
1061 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false);
1063 skb_dst_set(buff, dst);
1064 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
1065 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1067 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1074 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
1076 const struct tcphdr *th = tcp_hdr(skb);
1077 u32 seq = 0, ack_seq = 0;
1078 struct tcp_md5sig_key *key = NULL;
1083 if (!ipv6_unicast_destination(skb))
1086 #ifdef CONFIG_TCP_MD5SIG
1088 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr);
1092 seq = ntohl(th->ack_seq);
1094 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1097 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1, 0);
1100 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1101 struct tcp_md5sig_key *key, u8 tclass)
1103 tcp_v6_send_response(skb, seq, ack, win, ts, key, 0, tclass);
1106 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1108 struct inet_timewait_sock *tw = inet_twsk(sk);
1109 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1111 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1112 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1113 tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw),
1119 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1120 struct request_sock *req)
1122 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1123 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr), 0);
1127 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1129 struct request_sock *req, **prev;
1130 const struct tcphdr *th = tcp_hdr(skb);
1133 /* Find possible connection requests. */
1134 req = inet6_csk_search_req(sk, &prev, th->source,
1135 &ipv6_hdr(skb)->saddr,
1136 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1138 return tcp_check_req(sk, skb, req, prev);
1140 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1141 &ipv6_hdr(skb)->saddr, th->source,
1142 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1145 if (nsk->sk_state != TCP_TIME_WAIT) {
1149 inet_twsk_put(inet_twsk(nsk));
1153 #ifdef CONFIG_SYN_COOKIES
1155 sk = cookie_v6_check(sk, skb);
1160 /* FIXME: this is substantially similar to the ipv4 code.
1161 * Can some kind of merge be done? -- erics
1163 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1165 struct tcp_extend_values tmp_ext;
1166 struct tcp_options_received tmp_opt;
1167 const u8 *hash_location;
1168 struct request_sock *req;
1169 struct inet6_request_sock *treq;
1170 struct ipv6_pinfo *np = inet6_sk(sk);
1171 struct tcp_sock *tp = tcp_sk(sk);
1172 __u32 isn = TCP_SKB_CB(skb)->when;
1173 struct dst_entry *dst = NULL;
1174 int want_cookie = 0;
1176 if (skb->protocol == htons(ETH_P_IP))
1177 return tcp_v4_conn_request(sk, skb);
1179 if (!ipv6_unicast_destination(skb))
1182 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1183 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
1188 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1191 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1195 #ifdef CONFIG_TCP_MD5SIG
1196 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1199 tcp_clear_options(&tmp_opt);
1200 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1201 tmp_opt.user_mss = tp->rx_opt.user_mss;
1202 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1204 if (tmp_opt.cookie_plus > 0 &&
1205 tmp_opt.saw_tstamp &&
1206 !tp->rx_opt.cookie_out_never &&
1207 (sysctl_tcp_cookie_size > 0 ||
1208 (tp->cookie_values != NULL &&
1209 tp->cookie_values->cookie_desired > 0))) {
1212 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1213 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1215 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1218 /* Secret recipe starts with IP addresses */
1219 d = (__force u32 *)&ipv6_hdr(skb)->daddr.s6_addr32[0];
1224 d = (__force u32 *)&ipv6_hdr(skb)->saddr.s6_addr32[0];
1230 /* plus variable length Initiator Cookie */
1233 *c++ ^= *hash_location++;
1235 want_cookie = 0; /* not our kind of cookie */
1236 tmp_ext.cookie_out_never = 0; /* false */
1237 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1238 } else if (!tp->rx_opt.cookie_in_always) {
1239 /* redundant indications, but ensure initialization. */
1240 tmp_ext.cookie_out_never = 1; /* true */
1241 tmp_ext.cookie_plus = 0;
1245 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1247 if (want_cookie && !tmp_opt.saw_tstamp)
1248 tcp_clear_options(&tmp_opt);
1250 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1251 tcp_openreq_init(req, &tmp_opt, skb);
1253 treq = inet6_rsk(req);
1254 ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1255 ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1256 if (!want_cookie || tmp_opt.tstamp_ok)
1257 TCP_ECN_create_request(req, tcp_hdr(skb));
1259 treq->iif = sk->sk_bound_dev_if;
1261 /* So that link locals have meaning */
1262 if (!sk->sk_bound_dev_if &&
1263 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1264 treq->iif = inet6_iif(skb);
1267 struct inet_peer *peer = NULL;
1269 if (ipv6_opt_accepted(sk, skb) ||
1270 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1271 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1272 atomic_inc(&skb->users);
1273 treq->pktopts = skb;
1277 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1278 req->cookie_ts = tmp_opt.tstamp_ok;
1282 /* VJ's idea. We save last timestamp seen
1283 * from the destination in peer table, when entering
1284 * state TIME-WAIT, and check against it before
1285 * accepting new connection request.
1287 * If "isn" is not zero, this request hit alive
1288 * timewait bucket, so that all the necessary checks
1289 * are made in the function processing timewait state.
1291 if (tmp_opt.saw_tstamp &&
1292 tcp_death_row.sysctl_tw_recycle &&
1293 (dst = inet6_csk_route_req(sk, req)) != NULL &&
1294 (peer = rt6_get_peer((struct rt6_info *)dst)) != NULL &&
1295 ipv6_addr_equal((struct in6_addr *)peer->daddr.addr.a6,
1297 inet_peer_refcheck(peer);
1298 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1299 (s32)(peer->tcp_ts - req->ts_recent) >
1301 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1302 goto drop_and_release;
1305 /* Kill the following clause, if you dislike this way. */
1306 else if (!sysctl_tcp_syncookies &&
1307 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1308 (sysctl_max_syn_backlog >> 2)) &&
1309 (!peer || !peer->tcp_ts_stamp) &&
1310 (!dst || !dst_metric(dst, RTAX_RTT))) {
1311 /* Without syncookies last quarter of
1312 * backlog is filled with destinations,
1313 * proven to be alive.
1314 * It means that we continue to communicate
1315 * to destinations, already remembered
1316 * to the moment of synflood.
1318 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1319 &treq->rmt_addr, ntohs(tcp_hdr(skb)->source));
1320 goto drop_and_release;
1323 isn = tcp_v6_init_sequence(skb);
1326 tcp_rsk(req)->snt_isn = isn;
1327 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1329 security_inet_conn_request(sk, skb, req);
1331 if (tcp_v6_send_synack(sk, req,
1332 (struct request_values *)&tmp_ext) ||
1336 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1344 return 0; /* don't send reset */
1347 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1348 struct request_sock *req,
1349 struct dst_entry *dst)
1351 struct inet6_request_sock *treq;
1352 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1353 struct tcp6_sock *newtcp6sk;
1354 struct inet_sock *newinet;
1355 struct tcp_sock *newtp;
1357 struct ipv6_txoptions *opt;
1358 #ifdef CONFIG_TCP_MD5SIG
1359 struct tcp_md5sig_key *key;
1362 if (skb->protocol == htons(ETH_P_IP)) {
1367 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1372 newtcp6sk = (struct tcp6_sock *)newsk;
1373 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1375 newinet = inet_sk(newsk);
1376 newnp = inet6_sk(newsk);
1377 newtp = tcp_sk(newsk);
1379 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1381 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr);
1383 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1385 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1387 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1388 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1389 #ifdef CONFIG_TCP_MD5SIG
1390 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1393 newnp->ipv6_ac_list = NULL;
1394 newnp->ipv6_fl_list = NULL;
1395 newnp->pktoptions = NULL;
1397 newnp->mcast_oif = inet6_iif(skb);
1398 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1401 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1402 * here, tcp_create_openreq_child now does this for us, see the comment in
1403 * that function for the gory details. -acme
1406 /* It is tricky place. Until this moment IPv4 tcp
1407 worked with IPv6 icsk.icsk_af_ops.
1410 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1415 treq = inet6_rsk(req);
1418 if (sk_acceptq_is_full(sk))
1422 dst = inet6_csk_route_req(sk, req);
1427 newsk = tcp_create_openreq_child(sk, req, skb);
1432 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1433 * count here, tcp_create_openreq_child now does this for us, see the
1434 * comment in that function for the gory details. -acme
1437 newsk->sk_gso_type = SKB_GSO_TCPV6;
1438 __ip6_dst_store(newsk, dst, NULL, NULL);
1440 newtcp6sk = (struct tcp6_sock *)newsk;
1441 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1443 newtp = tcp_sk(newsk);
1444 newinet = inet_sk(newsk);
1445 newnp = inet6_sk(newsk);
1447 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1449 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1450 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1451 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1452 newsk->sk_bound_dev_if = treq->iif;
1454 /* Now IPv6 options...
1456 First: no IPv4 options.
1458 newinet->inet_opt = NULL;
1459 newnp->ipv6_ac_list = NULL;
1460 newnp->ipv6_fl_list = NULL;
1463 newnp->rxopt.all = np->rxopt.all;
1465 /* Clone pktoptions received with SYN */
1466 newnp->pktoptions = NULL;
1467 if (treq->pktopts != NULL) {
1468 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1469 kfree_skb(treq->pktopts);
1470 treq->pktopts = NULL;
1471 if (newnp->pktoptions)
1472 skb_set_owner_r(newnp->pktoptions, newsk);
1475 newnp->mcast_oif = inet6_iif(skb);
1476 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1478 /* Clone native IPv6 options from listening socket (if any)
1480 Yes, keeping reference count would be much more clever,
1481 but we make one more one thing there: reattach optmem
1485 newnp->opt = ipv6_dup_options(newsk, opt);
1487 sock_kfree_s(sk, opt, opt->tot_len);
1490 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1492 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1493 newnp->opt->opt_flen);
1495 tcp_mtup_init(newsk);
1496 tcp_sync_mss(newsk, dst_mtu(dst));
1497 newtp->advmss = dst_metric_advmss(dst);
1498 if (tcp_sk(sk)->rx_opt.user_mss &&
1499 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1500 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1502 tcp_initialize_rcv_mss(newsk);
1503 if (tcp_rsk(req)->snt_synack)
1504 tcp_valid_rtt_meas(newsk,
1505 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1506 newtp->total_retrans = req->retrans;
1508 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1509 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1511 #ifdef CONFIG_TCP_MD5SIG
1512 /* Copy over the MD5 key from the original socket */
1513 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1514 /* We're using one, so create a matching key
1515 * on the newsk structure. If we fail to get
1516 * memory, then we end up not copying the key
1519 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1521 tcp_v6_md5_do_add(newsk, &newnp->daddr,
1522 newkey, key->keylen);
1526 if (__inet_inherit_port(sk, newsk) < 0) {
1527 inet_csk_prepare_forced_close(newsk);
1531 __inet6_hash(newsk, NULL);
1536 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1538 if (opt && opt != np->opt)
1539 sock_kfree_s(sk, opt, opt->tot_len);
1542 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1546 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1548 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1549 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1550 &ipv6_hdr(skb)->daddr, skb->csum)) {
1551 skb->ip_summed = CHECKSUM_UNNECESSARY;
1556 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1557 &ipv6_hdr(skb)->saddr,
1558 &ipv6_hdr(skb)->daddr, 0));
1560 if (skb->len <= 76) {
1561 return __skb_checksum_complete(skb);
1566 /* The socket must have it's spinlock held when we get
1569 * We have a potential double-lock case here, so even when
1570 * doing backlog processing we use the BH locking scheme.
1571 * This is because we cannot sleep with the original spinlock
1574 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1576 struct ipv6_pinfo *np = inet6_sk(sk);
1577 struct tcp_sock *tp;
1578 struct sk_buff *opt_skb = NULL;
1580 /* Imagine: socket is IPv6. IPv4 packet arrives,
1581 goes to IPv4 receive handler and backlogged.
1582 From backlog it always goes here. Kerboom...
1583 Fortunately, tcp_rcv_established and rcv_established
1584 handle them correctly, but it is not case with
1585 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1588 if (skb->protocol == htons(ETH_P_IP))
1589 return tcp_v4_do_rcv(sk, skb);
1591 #ifdef CONFIG_TCP_MD5SIG
1592 if (tcp_v6_inbound_md5_hash (sk, skb))
1596 if (sk_filter(sk, skb))
1600 * socket locking is here for SMP purposes as backlog rcv
1601 * is currently called with bh processing disabled.
1604 /* Do Stevens' IPV6_PKTOPTIONS.
1606 Yes, guys, it is the only place in our code, where we
1607 may make it not affecting IPv4.
1608 The rest of code is protocol independent,
1609 and I do not like idea to uglify IPv4.
1611 Actually, all the idea behind IPV6_PKTOPTIONS
1612 looks not very well thought. For now we latch
1613 options, received in the last packet, enqueued
1614 by tcp. Feel free to propose better solution.
1618 opt_skb = skb_clone(skb, GFP_ATOMIC);
1620 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1621 sock_rps_save_rxhash(sk, skb);
1622 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1625 goto ipv6_pktoptions;
1629 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1632 if (sk->sk_state == TCP_LISTEN) {
1633 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1638 * Queue it on the new socket if the new socket is active,
1639 * otherwise we just shortcircuit this and continue with
1643 sock_rps_save_rxhash(nsk, skb);
1644 if (tcp_child_process(sk, nsk, skb))
1647 __kfree_skb(opt_skb);
1651 sock_rps_save_rxhash(sk, skb);
1653 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1656 goto ipv6_pktoptions;
1660 tcp_v6_send_reset(sk, skb);
1663 __kfree_skb(opt_skb);
1667 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1672 /* Do you ask, what is it?
1674 1. skb was enqueued by tcp.
1675 2. skb is added to tail of read queue, rather than out of order.
1676 3. socket is not in passive state.
1677 4. Finally, it really contains options, which user wants to receive.
1680 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1681 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1682 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1683 np->mcast_oif = inet6_iif(opt_skb);
1684 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1685 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1686 if (ipv6_opt_accepted(sk, opt_skb)) {
1687 skb_set_owner_r(opt_skb, sk);
1688 opt_skb = xchg(&np->pktoptions, opt_skb);
1690 __kfree_skb(opt_skb);
1691 opt_skb = xchg(&np->pktoptions, NULL);
1699 static int tcp_v6_rcv(struct sk_buff *skb)
1701 const struct tcphdr *th;
1702 const struct ipv6hdr *hdr;
1705 struct net *net = dev_net(skb->dev);
1707 if (skb->pkt_type != PACKET_HOST)
1711 * Count it even if it's bad.
1713 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1715 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1720 if (th->doff < sizeof(struct tcphdr)/4)
1722 if (!pskb_may_pull(skb, th->doff*4))
1725 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1729 hdr = ipv6_hdr(skb);
1730 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1731 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1732 skb->len - th->doff*4);
1733 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1734 TCP_SKB_CB(skb)->when = 0;
1735 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1736 TCP_SKB_CB(skb)->sacked = 0;
1738 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1743 if (sk->sk_state == TCP_TIME_WAIT)
1746 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1747 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1748 goto discard_and_relse;
1751 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1752 goto discard_and_relse;
1754 if (sk_filter(sk, skb))
1755 goto discard_and_relse;
1759 bh_lock_sock_nested(sk);
1761 if (!sock_owned_by_user(sk)) {
1762 #ifdef CONFIG_NET_DMA
1763 struct tcp_sock *tp = tcp_sk(sk);
1764 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1765 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1766 if (tp->ucopy.dma_chan)
1767 ret = tcp_v6_do_rcv(sk, skb);
1771 if (!tcp_prequeue(sk, skb))
1772 ret = tcp_v6_do_rcv(sk, skb);
1774 } else if (unlikely(sk_add_backlog(sk, skb))) {
1776 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1777 goto discard_and_relse;
1782 return ret ? -1 : 0;
1785 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1788 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1790 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1792 tcp_v6_send_reset(NULL, skb);
1809 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1810 inet_twsk_put(inet_twsk(sk));
1814 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1815 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1816 inet_twsk_put(inet_twsk(sk));
1820 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1825 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1826 &ipv6_hdr(skb)->daddr,
1827 ntohs(th->dest), inet6_iif(skb));
1829 struct inet_timewait_sock *tw = inet_twsk(sk);
1830 inet_twsk_deschedule(tw, &tcp_death_row);
1835 /* Fall through to ACK */
1838 tcp_v6_timewait_ack(sk, skb);
1842 case TCP_TW_SUCCESS:;
1847 static struct inet_peer *tcp_v6_get_peer(struct sock *sk, bool *release_it)
1849 struct rt6_info *rt = (struct rt6_info *) __sk_dst_get(sk);
1850 struct ipv6_pinfo *np = inet6_sk(sk);
1851 struct inet_peer *peer;
1854 !ipv6_addr_equal(&np->daddr, &rt->rt6i_dst.addr)) {
1855 peer = inet_getpeer_v6(&np->daddr, 1);
1859 rt6_bind_peer(rt, 1);
1860 peer = rt->rt6i_peer;
1861 *release_it = false;
1867 static void *tcp_v6_tw_get_peer(struct sock *sk)
1869 const struct inet6_timewait_sock *tw6 = inet6_twsk(sk);
1870 const struct inet_timewait_sock *tw = inet_twsk(sk);
1872 if (tw->tw_family == AF_INET)
1873 return tcp_v4_tw_get_peer(sk);
1875 return inet_getpeer_v6(&tw6->tw_v6_daddr, 1);
1878 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1879 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1880 .twsk_unique = tcp_twsk_unique,
1881 .twsk_destructor= tcp_twsk_destructor,
1882 .twsk_getpeer = tcp_v6_tw_get_peer,
1885 static const struct inet_connection_sock_af_ops ipv6_specific = {
1886 .queue_xmit = inet6_csk_xmit,
1887 .send_check = tcp_v6_send_check,
1888 .rebuild_header = inet6_sk_rebuild_header,
1889 .conn_request = tcp_v6_conn_request,
1890 .syn_recv_sock = tcp_v6_syn_recv_sock,
1891 .get_peer = tcp_v6_get_peer,
1892 .net_header_len = sizeof(struct ipv6hdr),
1893 .setsockopt = ipv6_setsockopt,
1894 .getsockopt = ipv6_getsockopt,
1895 .addr2sockaddr = inet6_csk_addr2sockaddr,
1896 .sockaddr_len = sizeof(struct sockaddr_in6),
1897 .bind_conflict = inet6_csk_bind_conflict,
1898 #ifdef CONFIG_COMPAT
1899 .compat_setsockopt = compat_ipv6_setsockopt,
1900 .compat_getsockopt = compat_ipv6_getsockopt,
1904 #ifdef CONFIG_TCP_MD5SIG
1905 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1906 .md5_lookup = tcp_v6_md5_lookup,
1907 .calc_md5_hash = tcp_v6_md5_hash_skb,
1908 .md5_add = tcp_v6_md5_add_func,
1909 .md5_parse = tcp_v6_parse_md5_keys,
1914 * TCP over IPv4 via INET6 API
1917 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1918 .queue_xmit = ip_queue_xmit,
1919 .send_check = tcp_v4_send_check,
1920 .rebuild_header = inet_sk_rebuild_header,
1921 .conn_request = tcp_v6_conn_request,
1922 .syn_recv_sock = tcp_v6_syn_recv_sock,
1923 .get_peer = tcp_v4_get_peer,
1924 .net_header_len = sizeof(struct iphdr),
1925 .setsockopt = ipv6_setsockopt,
1926 .getsockopt = ipv6_getsockopt,
1927 .addr2sockaddr = inet6_csk_addr2sockaddr,
1928 .sockaddr_len = sizeof(struct sockaddr_in6),
1929 .bind_conflict = inet6_csk_bind_conflict,
1930 #ifdef CONFIG_COMPAT
1931 .compat_setsockopt = compat_ipv6_setsockopt,
1932 .compat_getsockopt = compat_ipv6_getsockopt,
1936 #ifdef CONFIG_TCP_MD5SIG
1937 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1938 .md5_lookup = tcp_v4_md5_lookup,
1939 .calc_md5_hash = tcp_v4_md5_hash_skb,
1940 .md5_add = tcp_v6_md5_add_func,
1941 .md5_parse = tcp_v6_parse_md5_keys,
1945 /* NOTE: A lot of things set to zero explicitly by call to
1946 * sk_alloc() so need not be done here.
1948 static int tcp_v6_init_sock(struct sock *sk)
1950 struct inet_connection_sock *icsk = inet_csk(sk);
1951 struct tcp_sock *tp = tcp_sk(sk);
1953 skb_queue_head_init(&tp->out_of_order_queue);
1954 tcp_init_xmit_timers(sk);
1955 tcp_prequeue_init(tp);
1957 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1958 tp->mdev = TCP_TIMEOUT_INIT;
1960 /* So many TCP implementations out there (incorrectly) count the
1961 * initial SYN frame in their delayed-ACK and congestion control
1962 * algorithms that we must have the following bandaid to talk
1963 * efficiently to them. -DaveM
1967 /* See draft-stevens-tcpca-spec-01 for discussion of the
1968 * initialization of these values.
1970 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1971 tp->snd_cwnd_clamp = ~0;
1972 tp->mss_cache = TCP_MSS_DEFAULT;
1974 tp->reordering = sysctl_tcp_reordering;
1976 sk->sk_state = TCP_CLOSE;
1978 icsk->icsk_af_ops = &ipv6_specific;
1979 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1980 icsk->icsk_sync_mss = tcp_sync_mss;
1981 sk->sk_write_space = sk_stream_write_space;
1982 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1984 #ifdef CONFIG_TCP_MD5SIG
1985 tp->af_specific = &tcp_sock_ipv6_specific;
1988 /* TCP Cookie Transactions */
1989 if (sysctl_tcp_cookie_size > 0) {
1990 /* Default, cookies without s_data_payload. */
1992 kzalloc(sizeof(*tp->cookie_values),
1994 if (tp->cookie_values != NULL)
1995 kref_init(&tp->cookie_values->kref);
1997 /* Presumed zeroed, in order of appearance:
1998 * cookie_in_always, cookie_out_never,
1999 * s_data_constant, s_data_in, s_data_out
2001 sk->sk_sndbuf = sysctl_tcp_wmem[1];
2002 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
2005 percpu_counter_inc(&tcp_sockets_allocated);
2011 static void tcp_v6_destroy_sock(struct sock *sk)
2013 #ifdef CONFIG_TCP_MD5SIG
2014 /* Clean up the MD5 key list */
2015 if (tcp_sk(sk)->md5sig_info)
2016 tcp_v6_clear_md5_list(sk);
2018 tcp_v4_destroy_sock(sk);
2019 inet6_destroy_sock(sk);
2022 #ifdef CONFIG_PROC_FS
2023 /* Proc filesystem TCPv6 sock list dumping. */
2024 static void get_openreq6(struct seq_file *seq,
2025 const struct sock *sk, struct request_sock *req, int i, int uid)
2027 int ttd = req->expires - jiffies;
2028 const struct in6_addr *src = &inet6_rsk(req)->loc_addr;
2029 const struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
2035 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2036 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2038 src->s6_addr32[0], src->s6_addr32[1],
2039 src->s6_addr32[2], src->s6_addr32[3],
2040 ntohs(inet_rsk(req)->loc_port),
2041 dest->s6_addr32[0], dest->s6_addr32[1],
2042 dest->s6_addr32[2], dest->s6_addr32[3],
2043 ntohs(inet_rsk(req)->rmt_port),
2045 0,0, /* could print option size, but that is af dependent. */
2046 1, /* timers active (only the expire timer) */
2047 jiffies_to_clock_t(ttd),
2050 0, /* non standard timer */
2051 0, /* open_requests have no inode */
2055 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2057 const struct in6_addr *dest, *src;
2060 unsigned long timer_expires;
2061 const struct inet_sock *inet = inet_sk(sp);
2062 const struct tcp_sock *tp = tcp_sk(sp);
2063 const struct inet_connection_sock *icsk = inet_csk(sp);
2064 const struct ipv6_pinfo *np = inet6_sk(sp);
2067 src = &np->rcv_saddr;
2068 destp = ntohs(inet->inet_dport);
2069 srcp = ntohs(inet->inet_sport);
2071 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2073 timer_expires = icsk->icsk_timeout;
2074 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2076 timer_expires = icsk->icsk_timeout;
2077 } else if (timer_pending(&sp->sk_timer)) {
2079 timer_expires = sp->sk_timer.expires;
2082 timer_expires = jiffies;
2086 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2087 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %lu %lu %u %u %d\n",
2089 src->s6_addr32[0], src->s6_addr32[1],
2090 src->s6_addr32[2], src->s6_addr32[3], srcp,
2091 dest->s6_addr32[0], dest->s6_addr32[1],
2092 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2094 tp->write_seq-tp->snd_una,
2095 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2097 jiffies_to_clock_t(timer_expires - jiffies),
2098 icsk->icsk_retransmits,
2100 icsk->icsk_probes_out,
2102 atomic_read(&sp->sk_refcnt), sp,
2103 jiffies_to_clock_t(icsk->icsk_rto),
2104 jiffies_to_clock_t(icsk->icsk_ack.ato),
2105 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2107 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
2111 static void get_timewait6_sock(struct seq_file *seq,
2112 struct inet_timewait_sock *tw, int i)
2114 const struct in6_addr *dest, *src;
2116 const struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2117 int ttd = tw->tw_ttd - jiffies;
2122 dest = &tw6->tw_v6_daddr;
2123 src = &tw6->tw_v6_rcv_saddr;
2124 destp = ntohs(tw->tw_dport);
2125 srcp = ntohs(tw->tw_sport);
2128 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2129 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2131 src->s6_addr32[0], src->s6_addr32[1],
2132 src->s6_addr32[2], src->s6_addr32[3], srcp,
2133 dest->s6_addr32[0], dest->s6_addr32[1],
2134 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2135 tw->tw_substate, 0, 0,
2136 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2137 atomic_read(&tw->tw_refcnt), tw);
2140 static int tcp6_seq_show(struct seq_file *seq, void *v)
2142 struct tcp_iter_state *st;
2144 if (v == SEQ_START_TOKEN) {
2149 "st tx_queue rx_queue tr tm->when retrnsmt"
2150 " uid timeout inode\n");
2155 switch (st->state) {
2156 case TCP_SEQ_STATE_LISTENING:
2157 case TCP_SEQ_STATE_ESTABLISHED:
2158 get_tcp6_sock(seq, v, st->num);
2160 case TCP_SEQ_STATE_OPENREQ:
2161 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2163 case TCP_SEQ_STATE_TIME_WAIT:
2164 get_timewait6_sock(seq, v, st->num);
2171 static const struct file_operations tcp6_afinfo_seq_fops = {
2172 .owner = THIS_MODULE,
2173 .open = tcp_seq_open,
2175 .llseek = seq_lseek,
2176 .release = seq_release_net
2179 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2182 .seq_fops = &tcp6_afinfo_seq_fops,
2184 .show = tcp6_seq_show,
2188 int __net_init tcp6_proc_init(struct net *net)
2190 return tcp_proc_register(net, &tcp6_seq_afinfo);
2193 void tcp6_proc_exit(struct net *net)
2195 tcp_proc_unregister(net, &tcp6_seq_afinfo);
2199 static void tcp_v6_clear_sk(struct sock *sk, int size)
2201 struct inet_sock *inet = inet_sk(sk);
2203 /* we do not want to clear pinet6 field, because of RCU lookups */
2204 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
2206 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
2207 memset(&inet->pinet6 + 1, 0, size);
2210 struct proto tcpv6_prot = {
2212 .owner = THIS_MODULE,
2214 .connect = tcp_v6_connect,
2215 .disconnect = tcp_disconnect,
2216 .accept = inet_csk_accept,
2218 .init = tcp_v6_init_sock,
2219 .destroy = tcp_v6_destroy_sock,
2220 .shutdown = tcp_shutdown,
2221 .setsockopt = tcp_setsockopt,
2222 .getsockopt = tcp_getsockopt,
2223 .recvmsg = tcp_recvmsg,
2224 .sendmsg = tcp_sendmsg,
2225 .sendpage = tcp_sendpage,
2226 .backlog_rcv = tcp_v6_do_rcv,
2227 .hash = tcp_v6_hash,
2228 .unhash = inet_unhash,
2229 .get_port = inet_csk_get_port,
2230 .enter_memory_pressure = tcp_enter_memory_pressure,
2231 .sockets_allocated = &tcp_sockets_allocated,
2232 .memory_allocated = &tcp_memory_allocated,
2233 .memory_pressure = &tcp_memory_pressure,
2234 .orphan_count = &tcp_orphan_count,
2235 .sysctl_mem = sysctl_tcp_mem,
2236 .sysctl_wmem = sysctl_tcp_wmem,
2237 .sysctl_rmem = sysctl_tcp_rmem,
2238 .max_header = MAX_TCP_HEADER,
2239 .obj_size = sizeof(struct tcp6_sock),
2240 .slab_flags = SLAB_DESTROY_BY_RCU,
2241 .twsk_prot = &tcp6_timewait_sock_ops,
2242 .rsk_prot = &tcp6_request_sock_ops,
2243 .h.hashinfo = &tcp_hashinfo,
2244 .no_autobind = true,
2245 #ifdef CONFIG_COMPAT
2246 .compat_setsockopt = compat_tcp_setsockopt,
2247 .compat_getsockopt = compat_tcp_getsockopt,
2249 .clear_sk = tcp_v6_clear_sk,
2252 static const struct inet6_protocol tcpv6_protocol = {
2253 .handler = tcp_v6_rcv,
2254 .err_handler = tcp_v6_err,
2255 .gso_send_check = tcp_v6_gso_send_check,
2256 .gso_segment = tcp_tso_segment,
2257 .gro_receive = tcp6_gro_receive,
2258 .gro_complete = tcp6_gro_complete,
2259 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2262 static struct inet_protosw tcpv6_protosw = {
2263 .type = SOCK_STREAM,
2264 .protocol = IPPROTO_TCP,
2265 .prot = &tcpv6_prot,
2266 .ops = &inet6_stream_ops,
2268 .flags = INET_PROTOSW_PERMANENT |
2272 static int __net_init tcpv6_net_init(struct net *net)
2274 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2275 SOCK_RAW, IPPROTO_TCP, net);
2278 static void __net_exit tcpv6_net_exit(struct net *net)
2280 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2283 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2285 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2288 static struct pernet_operations tcpv6_net_ops = {
2289 .init = tcpv6_net_init,
2290 .exit = tcpv6_net_exit,
2291 .exit_batch = tcpv6_net_exit_batch,
2294 int __init tcpv6_init(void)
2298 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2302 /* register inet6 protocol */
2303 ret = inet6_register_protosw(&tcpv6_protosw);
2305 goto out_tcpv6_protocol;
2307 ret = register_pernet_subsys(&tcpv6_net_ops);
2309 goto out_tcpv6_protosw;
2314 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2316 inet6_unregister_protosw(&tcpv6_protosw);
2320 void tcpv6_exit(void)
2322 unregister_pernet_subsys(&tcpv6_net_ops);
2323 inet6_unregister_protosw(&tcpv6_protosw);
2324 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);