b6b356b7912a57c298a542731c2c9daf24a77bbc
[linux-2.6.git] / net / ipv6 / tcp_ipv6.c
1 /*
2  *      TCP over IPv6
3  *      Linux INET6 implementation
4  *
5  *      Authors:
6  *      Pedro Roque             <roque@di.fc.ul.pt>
7  *
8  *      Based on:
9  *      linux/net/ipv4/tcp.c
10  *      linux/net/ipv4/tcp_input.c
11  *      linux/net/ipv4/tcp_output.c
12  *
13  *      Fixes:
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.
19  *
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.
24  */
25
26 #include <linux/module.h>
27 #include <linux/errno.h>
28 #include <linux/types.h>
29 #include <linux/socket.h>
30 #include <linux/sockios.h>
31 #include <linux/net.h>
32 #include <linux/jiffies.h>
33 #include <linux/in.h>
34 #include <linux/in6.h>
35 #include <linux/netdevice.h>
36 #include <linux/init.h>
37 #include <linux/jhash.h>
38 #include <linux/ipsec.h>
39 #include <linux/times.h>
40
41 #include <linux/ipv6.h>
42 #include <linux/icmpv6.h>
43 #include <linux/random.h>
44
45 #include <net/tcp.h>
46 #include <net/ndisc.h>
47 #include <net/inet6_hashtables.h>
48 #include <net/inet6_connection_sock.h>
49 #include <net/ipv6.h>
50 #include <net/transp_v6.h>
51 #include <net/addrconf.h>
52 #include <net/ip6_route.h>
53 #include <net/ip6_checksum.h>
54 #include <net/inet_ecn.h>
55 #include <net/protocol.h>
56 #include <net/xfrm.h>
57 #include <net/snmp.h>
58 #include <net/dsfield.h>
59 #include <net/timewait_sock.h>
60 #include <net/netdma.h>
61 #include <net/inet_common.h>
62
63 #include <asm/uaccess.h>
64
65 #include <linux/proc_fs.h>
66 #include <linux/seq_file.h>
67
68 #include <linux/crypto.h>
69 #include <linux/scatterlist.h>
70
71 static void     tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
72 static void     tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
73                                       struct request_sock *req);
74
75 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
76
77 static struct inet_connection_sock_af_ops ipv6_mapped;
78 static struct inet_connection_sock_af_ops ipv6_specific;
79 #ifdef CONFIG_TCP_MD5SIG
80 static struct tcp_sock_af_ops tcp_sock_ipv6_specific;
81 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
82 #else
83 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
84                                                    struct in6_addr *addr)
85 {
86         return NULL;
87 }
88 #endif
89
90 static void tcp_v6_hash(struct sock *sk)
91 {
92         if (sk->sk_state != TCP_CLOSE) {
93                 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
94                         tcp_prot.hash(sk);
95                         return;
96                 }
97                 local_bh_disable();
98                 __inet6_hash(sk);
99                 local_bh_enable();
100         }
101 }
102
103 static __inline__ __sum16 tcp_v6_check(struct tcphdr *th, int len,
104                                    struct in6_addr *saddr,
105                                    struct in6_addr *daddr,
106                                    __wsum base)
107 {
108         return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
109 }
110
111 static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
112 {
113         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
114                                             ipv6_hdr(skb)->saddr.s6_addr32,
115                                             tcp_hdr(skb)->dest,
116                                             tcp_hdr(skb)->source);
117 }
118
119 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
120                           int addr_len)
121 {
122         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
123         struct inet_sock *inet = inet_sk(sk);
124         struct inet_connection_sock *icsk = inet_csk(sk);
125         struct ipv6_pinfo *np = inet6_sk(sk);
126         struct tcp_sock *tp = tcp_sk(sk);
127         struct in6_addr *saddr = NULL, *final_p = NULL, final;
128         struct flowi fl;
129         struct dst_entry *dst;
130         int addr_type;
131         int err;
132
133         if (addr_len < SIN6_LEN_RFC2133)
134                 return -EINVAL;
135
136         if (usin->sin6_family != AF_INET6)
137                 return(-EAFNOSUPPORT);
138
139         memset(&fl, 0, sizeof(fl));
140
141         if (np->sndflow) {
142                 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
143                 IP6_ECN_flow_init(fl.fl6_flowlabel);
144                 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
145                         struct ip6_flowlabel *flowlabel;
146                         flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
147                         if (flowlabel == NULL)
148                                 return -EINVAL;
149                         ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
150                         fl6_sock_release(flowlabel);
151                 }
152         }
153
154         /*
155          *      connect() to INADDR_ANY means loopback (BSD'ism).
156          */
157
158         if(ipv6_addr_any(&usin->sin6_addr))
159                 usin->sin6_addr.s6_addr[15] = 0x1;
160
161         addr_type = ipv6_addr_type(&usin->sin6_addr);
162
163         if(addr_type & IPV6_ADDR_MULTICAST)
164                 return -ENETUNREACH;
165
166         if (addr_type&IPV6_ADDR_LINKLOCAL) {
167                 if (addr_len >= sizeof(struct sockaddr_in6) &&
168                     usin->sin6_scope_id) {
169                         /* If interface is set while binding, indices
170                          * must coincide.
171                          */
172                         if (sk->sk_bound_dev_if &&
173                             sk->sk_bound_dev_if != usin->sin6_scope_id)
174                                 return -EINVAL;
175
176                         sk->sk_bound_dev_if = usin->sin6_scope_id;
177                 }
178
179                 /* Connect to link-local address requires an interface */
180                 if (!sk->sk_bound_dev_if)
181                         return -EINVAL;
182         }
183
184         if (tp->rx_opt.ts_recent_stamp &&
185             !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
186                 tp->rx_opt.ts_recent = 0;
187                 tp->rx_opt.ts_recent_stamp = 0;
188                 tp->write_seq = 0;
189         }
190
191         ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
192         np->flow_label = fl.fl6_flowlabel;
193
194         /*
195          *      TCP over IPv4
196          */
197
198         if (addr_type == IPV6_ADDR_MAPPED) {
199                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
200                 struct sockaddr_in sin;
201
202                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
203
204                 if (__ipv6_only_sock(sk))
205                         return -ENETUNREACH;
206
207                 sin.sin_family = AF_INET;
208                 sin.sin_port = usin->sin6_port;
209                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
210
211                 icsk->icsk_af_ops = &ipv6_mapped;
212                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
213 #ifdef CONFIG_TCP_MD5SIG
214                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
215 #endif
216
217                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
218
219                 if (err) {
220                         icsk->icsk_ext_hdr_len = exthdrlen;
221                         icsk->icsk_af_ops = &ipv6_specific;
222                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
223 #ifdef CONFIG_TCP_MD5SIG
224                         tp->af_specific = &tcp_sock_ipv6_specific;
225 #endif
226                         goto failure;
227                 } else {
228                         ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
229                                       inet->saddr);
230                         ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
231                                       inet->rcv_saddr);
232                 }
233
234                 return err;
235         }
236
237         if (!ipv6_addr_any(&np->rcv_saddr))
238                 saddr = &np->rcv_saddr;
239
240         fl.proto = IPPROTO_TCP;
241         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
242         ipv6_addr_copy(&fl.fl6_src,
243                        (saddr ? saddr : &np->saddr));
244         fl.oif = sk->sk_bound_dev_if;
245         fl.fl_ip_dport = usin->sin6_port;
246         fl.fl_ip_sport = inet->sport;
247
248         if (np->opt && np->opt->srcrt) {
249                 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
250                 ipv6_addr_copy(&final, &fl.fl6_dst);
251                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
252                 final_p = &final;
253         }
254
255         security_sk_classify_flow(sk, &fl);
256
257         err = ip6_dst_lookup(sk, &dst, &fl);
258         if (err)
259                 goto failure;
260         if (final_p)
261                 ipv6_addr_copy(&fl.fl6_dst, final_p);
262
263         if ((err = __xfrm_lookup(&dst, &fl, sk, XFRM_LOOKUP_WAIT)) < 0) {
264                 if (err == -EREMOTE)
265                         err = ip6_dst_blackhole(sk, &dst, &fl);
266                 if (err < 0)
267                         goto failure;
268         }
269
270         if (saddr == NULL) {
271                 saddr = &fl.fl6_src;
272                 ipv6_addr_copy(&np->rcv_saddr, saddr);
273         }
274
275         /* set the source address */
276         ipv6_addr_copy(&np->saddr, saddr);
277         inet->rcv_saddr = LOOPBACK4_IPV6;
278
279         sk->sk_gso_type = SKB_GSO_TCPV6;
280         __ip6_dst_store(sk, dst, NULL, NULL);
281
282         icsk->icsk_ext_hdr_len = 0;
283         if (np->opt)
284                 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
285                                           np->opt->opt_nflen);
286
287         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
288
289         inet->dport = usin->sin6_port;
290
291         tcp_set_state(sk, TCP_SYN_SENT);
292         err = inet6_hash_connect(&tcp_death_row, sk);
293         if (err)
294                 goto late_failure;
295
296         if (!tp->write_seq)
297                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
298                                                              np->daddr.s6_addr32,
299                                                              inet->sport,
300                                                              inet->dport);
301
302         err = tcp_connect(sk);
303         if (err)
304                 goto late_failure;
305
306         return 0;
307
308 late_failure:
309         tcp_set_state(sk, TCP_CLOSE);
310         __sk_dst_reset(sk);
311 failure:
312         inet->dport = 0;
313         sk->sk_route_caps = 0;
314         return err;
315 }
316
317 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
318                 int type, int code, int offset, __be32 info)
319 {
320         struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
321         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
322         struct ipv6_pinfo *np;
323         struct sock *sk;
324         int err;
325         struct tcp_sock *tp;
326         __u32 seq;
327         struct net *net = dev_net(skb->dev);
328
329         sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
330                         th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
331
332         if (sk == NULL) {
333                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
334                                    ICMP6_MIB_INERRORS);
335                 return;
336         }
337
338         if (sk->sk_state == TCP_TIME_WAIT) {
339                 inet_twsk_put(inet_twsk(sk));
340                 return;
341         }
342
343         bh_lock_sock(sk);
344         if (sock_owned_by_user(sk))
345                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
346
347         if (sk->sk_state == TCP_CLOSE)
348                 goto out;
349
350         tp = tcp_sk(sk);
351         seq = ntohl(th->seq);
352         if (sk->sk_state != TCP_LISTEN &&
353             !between(seq, tp->snd_una, tp->snd_nxt)) {
354                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
355                 goto out;
356         }
357
358         np = inet6_sk(sk);
359
360         if (type == ICMPV6_PKT_TOOBIG) {
361                 struct dst_entry *dst = NULL;
362
363                 if (sock_owned_by_user(sk))
364                         goto out;
365                 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
366                         goto out;
367
368                 /* icmp should have updated the destination cache entry */
369                 dst = __sk_dst_check(sk, np->dst_cookie);
370
371                 if (dst == NULL) {
372                         struct inet_sock *inet = inet_sk(sk);
373                         struct flowi fl;
374
375                         /* BUGGG_FUTURE: Again, it is not clear how
376                            to handle rthdr case. Ignore this complexity
377                            for now.
378                          */
379                         memset(&fl, 0, sizeof(fl));
380                         fl.proto = IPPROTO_TCP;
381                         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
382                         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
383                         fl.oif = sk->sk_bound_dev_if;
384                         fl.fl_ip_dport = inet->dport;
385                         fl.fl_ip_sport = inet->sport;
386                         security_skb_classify_flow(skb, &fl);
387
388                         if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
389                                 sk->sk_err_soft = -err;
390                                 goto out;
391                         }
392
393                         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
394                                 sk->sk_err_soft = -err;
395                                 goto out;
396                         }
397
398                 } else
399                         dst_hold(dst);
400
401                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
402                         tcp_sync_mss(sk, dst_mtu(dst));
403                         tcp_simple_retransmit(sk);
404                 } /* else let the usual retransmit timer handle it */
405                 dst_release(dst);
406                 goto out;
407         }
408
409         icmpv6_err_convert(type, code, &err);
410
411         /* Might be for an request_sock */
412         switch (sk->sk_state) {
413                 struct request_sock *req, **prev;
414         case TCP_LISTEN:
415                 if (sock_owned_by_user(sk))
416                         goto out;
417
418                 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
419                                            &hdr->saddr, inet6_iif(skb));
420                 if (!req)
421                         goto out;
422
423                 /* ICMPs are not backlogged, hence we cannot get
424                  * an established socket here.
425                  */
426                 WARN_ON(req->sk != NULL);
427
428                 if (seq != tcp_rsk(req)->snt_isn) {
429                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
430                         goto out;
431                 }
432
433                 inet_csk_reqsk_queue_drop(sk, req, prev);
434                 goto out;
435
436         case TCP_SYN_SENT:
437         case TCP_SYN_RECV:  /* Cannot happen.
438                                It can, it SYNs are crossed. --ANK */
439                 if (!sock_owned_by_user(sk)) {
440                         sk->sk_err = err;
441                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
442
443                         tcp_done(sk);
444                 } else
445                         sk->sk_err_soft = err;
446                 goto out;
447         }
448
449         if (!sock_owned_by_user(sk) && np->recverr) {
450                 sk->sk_err = err;
451                 sk->sk_error_report(sk);
452         } else
453                 sk->sk_err_soft = err;
454
455 out:
456         bh_unlock_sock(sk);
457         sock_put(sk);
458 }
459
460
461 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req)
462 {
463         struct inet6_request_sock *treq = inet6_rsk(req);
464         struct ipv6_pinfo *np = inet6_sk(sk);
465         struct sk_buff * skb;
466         struct ipv6_txoptions *opt = NULL;
467         struct in6_addr * final_p = NULL, final;
468         struct flowi fl;
469         struct dst_entry *dst;
470         int err = -1;
471
472         memset(&fl, 0, sizeof(fl));
473         fl.proto = IPPROTO_TCP;
474         ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
475         ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
476         fl.fl6_flowlabel = 0;
477         fl.oif = treq->iif;
478         fl.fl_ip_dport = inet_rsk(req)->rmt_port;
479         fl.fl_ip_sport = inet_rsk(req)->loc_port;
480         security_req_classify_flow(req, &fl);
481
482         opt = np->opt;
483         if (opt && opt->srcrt) {
484                 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
485                 ipv6_addr_copy(&final, &fl.fl6_dst);
486                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
487                 final_p = &final;
488         }
489
490         err = ip6_dst_lookup(sk, &dst, &fl);
491         if (err)
492                 goto done;
493         if (final_p)
494                 ipv6_addr_copy(&fl.fl6_dst, final_p);
495         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
496                 goto done;
497
498         skb = tcp_make_synack(sk, dst, req);
499         if (skb) {
500                 struct tcphdr *th = tcp_hdr(skb);
501
502                 th->check = tcp_v6_check(th, skb->len,
503                                          &treq->loc_addr, &treq->rmt_addr,
504                                          csum_partial((char *)th, skb->len, skb->csum));
505
506                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
507                 err = ip6_xmit(sk, skb, &fl, opt, 0);
508                 err = net_xmit_eval(err);
509         }
510
511 done:
512         if (opt && opt != np->opt)
513                 sock_kfree_s(sk, opt, opt->tot_len);
514         dst_release(dst);
515         return err;
516 }
517
518 static inline void syn_flood_warning(struct sk_buff *skb)
519 {
520 #ifdef CONFIG_SYN_COOKIES
521         if (sysctl_tcp_syncookies)
522                 printk(KERN_INFO
523                        "TCPv6: Possible SYN flooding on port %d. "
524                        "Sending cookies.\n", ntohs(tcp_hdr(skb)->dest));
525         else
526 #endif
527                 printk(KERN_INFO
528                        "TCPv6: Possible SYN flooding on port %d. "
529                        "Dropping request.\n", ntohs(tcp_hdr(skb)->dest));
530 }
531
532 static void tcp_v6_reqsk_destructor(struct request_sock *req)
533 {
534         if (inet6_rsk(req)->pktopts)
535                 kfree_skb(inet6_rsk(req)->pktopts);
536 }
537
538 #ifdef CONFIG_TCP_MD5SIG
539 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
540                                                    struct in6_addr *addr)
541 {
542         struct tcp_sock *tp = tcp_sk(sk);
543         int i;
544
545         BUG_ON(tp == NULL);
546
547         if (!tp->md5sig_info || !tp->md5sig_info->entries6)
548                 return NULL;
549
550         for (i = 0; i < tp->md5sig_info->entries6; i++) {
551                 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, addr))
552                         return &tp->md5sig_info->keys6[i].base;
553         }
554         return NULL;
555 }
556
557 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
558                                                 struct sock *addr_sk)
559 {
560         return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
561 }
562
563 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
564                                                       struct request_sock *req)
565 {
566         return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
567 }
568
569 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
570                              char *newkey, u8 newkeylen)
571 {
572         /* Add key to the list */
573         struct tcp_md5sig_key *key;
574         struct tcp_sock *tp = tcp_sk(sk);
575         struct tcp6_md5sig_key *keys;
576
577         key = tcp_v6_md5_do_lookup(sk, peer);
578         if (key) {
579                 /* modify existing entry - just update that one */
580                 kfree(key->key);
581                 key->key = newkey;
582                 key->keylen = newkeylen;
583         } else {
584                 /* reallocate new list if current one is full. */
585                 if (!tp->md5sig_info) {
586                         tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
587                         if (!tp->md5sig_info) {
588                                 kfree(newkey);
589                                 return -ENOMEM;
590                         }
591                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
592                 }
593                 if (tcp_alloc_md5sig_pool() == NULL) {
594                         kfree(newkey);
595                         return -ENOMEM;
596                 }
597                 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
598                         keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
599                                        (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
600
601                         if (!keys) {
602                                 tcp_free_md5sig_pool();
603                                 kfree(newkey);
604                                 return -ENOMEM;
605                         }
606
607                         if (tp->md5sig_info->entries6)
608                                 memmove(keys, tp->md5sig_info->keys6,
609                                         (sizeof (tp->md5sig_info->keys6[0]) *
610                                          tp->md5sig_info->entries6));
611
612                         kfree(tp->md5sig_info->keys6);
613                         tp->md5sig_info->keys6 = keys;
614                         tp->md5sig_info->alloced6++;
615                 }
616
617                 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
618                                peer);
619                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
620                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
621
622                 tp->md5sig_info->entries6++;
623         }
624         return 0;
625 }
626
627 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
628                                u8 *newkey, __u8 newkeylen)
629 {
630         return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
631                                  newkey, newkeylen);
632 }
633
634 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
635 {
636         struct tcp_sock *tp = tcp_sk(sk);
637         int i;
638
639         for (i = 0; i < tp->md5sig_info->entries6; i++) {
640                 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) {
641                         /* Free the key */
642                         kfree(tp->md5sig_info->keys6[i].base.key);
643                         tp->md5sig_info->entries6--;
644
645                         if (tp->md5sig_info->entries6 == 0) {
646                                 kfree(tp->md5sig_info->keys6);
647                                 tp->md5sig_info->keys6 = NULL;
648                                 tp->md5sig_info->alloced6 = 0;
649                         } else {
650                                 /* shrink the database */
651                                 if (tp->md5sig_info->entries6 != i)
652                                         memmove(&tp->md5sig_info->keys6[i],
653                                                 &tp->md5sig_info->keys6[i+1],
654                                                 (tp->md5sig_info->entries6 - i)
655                                                 * sizeof (tp->md5sig_info->keys6[0]));
656                         }
657                         tcp_free_md5sig_pool();
658                         return 0;
659                 }
660         }
661         return -ENOENT;
662 }
663
664 static void tcp_v6_clear_md5_list (struct sock *sk)
665 {
666         struct tcp_sock *tp = tcp_sk(sk);
667         int i;
668
669         if (tp->md5sig_info->entries6) {
670                 for (i = 0; i < tp->md5sig_info->entries6; i++)
671                         kfree(tp->md5sig_info->keys6[i].base.key);
672                 tp->md5sig_info->entries6 = 0;
673                 tcp_free_md5sig_pool();
674         }
675
676         kfree(tp->md5sig_info->keys6);
677         tp->md5sig_info->keys6 = NULL;
678         tp->md5sig_info->alloced6 = 0;
679
680         if (tp->md5sig_info->entries4) {
681                 for (i = 0; i < tp->md5sig_info->entries4; i++)
682                         kfree(tp->md5sig_info->keys4[i].base.key);
683                 tp->md5sig_info->entries4 = 0;
684                 tcp_free_md5sig_pool();
685         }
686
687         kfree(tp->md5sig_info->keys4);
688         tp->md5sig_info->keys4 = NULL;
689         tp->md5sig_info->alloced4 = 0;
690 }
691
692 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
693                                   int optlen)
694 {
695         struct tcp_md5sig cmd;
696         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
697         u8 *newkey;
698
699         if (optlen < sizeof(cmd))
700                 return -EINVAL;
701
702         if (copy_from_user(&cmd, optval, sizeof(cmd)))
703                 return -EFAULT;
704
705         if (sin6->sin6_family != AF_INET6)
706                 return -EINVAL;
707
708         if (!cmd.tcpm_keylen) {
709                 if (!tcp_sk(sk)->md5sig_info)
710                         return -ENOENT;
711                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
712                         return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
713                 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
714         }
715
716         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
717                 return -EINVAL;
718
719         if (!tcp_sk(sk)->md5sig_info) {
720                 struct tcp_sock *tp = tcp_sk(sk);
721                 struct tcp_md5sig_info *p;
722
723                 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
724                 if (!p)
725                         return -ENOMEM;
726
727                 tp->md5sig_info = p;
728                 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
729         }
730
731         newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
732         if (!newkey)
733                 return -ENOMEM;
734         if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
735                 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
736                                          newkey, cmd.tcpm_keylen);
737         }
738         return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
739 }
740
741 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
742                                         struct in6_addr *daddr,
743                                         struct in6_addr *saddr, int nbytes)
744 {
745         struct tcp6_pseudohdr *bp;
746         struct scatterlist sg;
747
748         bp = &hp->md5_blk.ip6;
749         /* 1. TCP pseudo-header (RFC2460) */
750         ipv6_addr_copy(&bp->saddr, saddr);
751         ipv6_addr_copy(&bp->daddr, daddr);
752         bp->protocol = cpu_to_be32(IPPROTO_TCP);
753         bp->len = cpu_to_be32(nbytes);
754
755         sg_init_one(&sg, bp, sizeof(*bp));
756         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
757 }
758
759 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
760                                struct in6_addr *daddr, struct in6_addr *saddr,
761                                struct tcphdr *th)
762 {
763         struct tcp_md5sig_pool *hp;
764         struct hash_desc *desc;
765
766         hp = tcp_get_md5sig_pool();
767         if (!hp)
768                 goto clear_hash_noput;
769         desc = &hp->md5_desc;
770
771         if (crypto_hash_init(desc))
772                 goto clear_hash;
773         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
774                 goto clear_hash;
775         if (tcp_md5_hash_header(hp, th))
776                 goto clear_hash;
777         if (tcp_md5_hash_key(hp, key))
778                 goto clear_hash;
779         if (crypto_hash_final(desc, md5_hash))
780                 goto clear_hash;
781
782         tcp_put_md5sig_pool();
783         return 0;
784
785 clear_hash:
786         tcp_put_md5sig_pool();
787 clear_hash_noput:
788         memset(md5_hash, 0, 16);
789         return 1;
790 }
791
792 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
793                                struct sock *sk, struct request_sock *req,
794                                struct sk_buff *skb)
795 {
796         struct in6_addr *saddr, *daddr;
797         struct tcp_md5sig_pool *hp;
798         struct hash_desc *desc;
799         struct tcphdr *th = tcp_hdr(skb);
800
801         if (sk) {
802                 saddr = &inet6_sk(sk)->saddr;
803                 daddr = &inet6_sk(sk)->daddr;
804         } else if (req) {
805                 saddr = &inet6_rsk(req)->loc_addr;
806                 daddr = &inet6_rsk(req)->rmt_addr;
807         } else {
808                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
809                 saddr = &ip6h->saddr;
810                 daddr = &ip6h->daddr;
811         }
812
813         hp = tcp_get_md5sig_pool();
814         if (!hp)
815                 goto clear_hash_noput;
816         desc = &hp->md5_desc;
817
818         if (crypto_hash_init(desc))
819                 goto clear_hash;
820
821         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
822                 goto clear_hash;
823         if (tcp_md5_hash_header(hp, th))
824                 goto clear_hash;
825         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
826                 goto clear_hash;
827         if (tcp_md5_hash_key(hp, key))
828                 goto clear_hash;
829         if (crypto_hash_final(desc, md5_hash))
830                 goto clear_hash;
831
832         tcp_put_md5sig_pool();
833         return 0;
834
835 clear_hash:
836         tcp_put_md5sig_pool();
837 clear_hash_noput:
838         memset(md5_hash, 0, 16);
839         return 1;
840 }
841
842 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
843 {
844         __u8 *hash_location = NULL;
845         struct tcp_md5sig_key *hash_expected;
846         struct ipv6hdr *ip6h = ipv6_hdr(skb);
847         struct tcphdr *th = tcp_hdr(skb);
848         int genhash;
849         u8 newhash[16];
850
851         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
852         hash_location = tcp_parse_md5sig_option(th);
853
854         /* We've parsed the options - do we have a hash? */
855         if (!hash_expected && !hash_location)
856                 return 0;
857
858         if (hash_expected && !hash_location) {
859                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
860                 return 1;
861         }
862
863         if (!hash_expected && hash_location) {
864                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
865                 return 1;
866         }
867
868         /* check the signature */
869         genhash = tcp_v6_md5_hash_skb(newhash,
870                                       hash_expected,
871                                       NULL, NULL, skb);
872
873         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
874                 if (net_ratelimit()) {
875                         printk(KERN_INFO "MD5 Hash %s for "
876                                "(" NIP6_FMT ", %u)->"
877                                "(" NIP6_FMT ", %u)\n",
878                                genhash ? "failed" : "mismatch",
879                                NIP6(ip6h->saddr), ntohs(th->source),
880                                NIP6(ip6h->daddr), ntohs(th->dest));
881                 }
882                 return 1;
883         }
884         return 0;
885 }
886 #endif
887
888 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
889         .family         =       AF_INET6,
890         .obj_size       =       sizeof(struct tcp6_request_sock),
891         .rtx_syn_ack    =       tcp_v6_send_synack,
892         .send_ack       =       tcp_v6_reqsk_send_ack,
893         .destructor     =       tcp_v6_reqsk_destructor,
894         .send_reset     =       tcp_v6_send_reset
895 };
896
897 #ifdef CONFIG_TCP_MD5SIG
898 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
899         .md5_lookup     =       tcp_v6_reqsk_md5_lookup,
900 };
901 #endif
902
903 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
904         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
905         .twsk_unique    = tcp_twsk_unique,
906         .twsk_destructor= tcp_twsk_destructor,
907 };
908
909 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
910 {
911         struct ipv6_pinfo *np = inet6_sk(sk);
912         struct tcphdr *th = tcp_hdr(skb);
913
914         if (skb->ip_summed == CHECKSUM_PARTIAL) {
915                 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,  0);
916                 skb->csum_start = skb_transport_header(skb) - skb->head;
917                 skb->csum_offset = offsetof(struct tcphdr, check);
918         } else {
919                 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
920                                             csum_partial((char *)th, th->doff<<2,
921                                                          skb->csum));
922         }
923 }
924
925 static int tcp_v6_gso_send_check(struct sk_buff *skb)
926 {
927         struct ipv6hdr *ipv6h;
928         struct tcphdr *th;
929
930         if (!pskb_may_pull(skb, sizeof(*th)))
931                 return -EINVAL;
932
933         ipv6h = ipv6_hdr(skb);
934         th = tcp_hdr(skb);
935
936         th->check = 0;
937         th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
938                                      IPPROTO_TCP, 0);
939         skb->csum_start = skb_transport_header(skb) - skb->head;
940         skb->csum_offset = offsetof(struct tcphdr, check);
941         skb->ip_summed = CHECKSUM_PARTIAL;
942         return 0;
943 }
944
945 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
946                                  u32 ts, struct tcp_md5sig_key *key, int rst)
947 {
948         struct tcphdr *th = tcp_hdr(skb), *t1;
949         struct sk_buff *buff;
950         struct flowi fl;
951         struct net *net = dev_net(skb->dst->dev);
952         struct sock *ctl_sk = net->ipv6.tcp_sk;
953         unsigned int tot_len = sizeof(struct tcphdr);
954         __be32 *topt;
955
956         if (ts)
957                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
958 #ifdef CONFIG_TCP_MD5SIG
959         if (key)
960                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
961 #endif
962
963         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
964                          GFP_ATOMIC);
965         if (buff == NULL)
966                 return;
967
968         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
969
970         t1 = (struct tcphdr *) skb_push(buff, tot_len);
971
972         /* Swap the send and the receive. */
973         memset(t1, 0, sizeof(*t1));
974         t1->dest = th->source;
975         t1->source = th->dest;
976         t1->doff = tot_len / 4;
977         t1->seq = htonl(seq);
978         t1->ack_seq = htonl(ack);
979         t1->ack = !rst || !th->ack;
980         t1->rst = rst;
981         t1->window = htons(win);
982
983         topt = (__be32 *)(t1 + 1);
984
985         if (ts) {
986                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
987                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
988                 *topt++ = htonl(tcp_time_stamp);
989                 *topt++ = htonl(ts);
990         }
991
992 #ifdef CONFIG_TCP_MD5SIG
993         if (key) {
994                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
995                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
996                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
997                                     &ipv6_hdr(skb)->saddr,
998                                     &ipv6_hdr(skb)->daddr, t1);
999         }
1000 #endif
1001
1002         buff->csum = csum_partial((char *)t1, tot_len, 0);
1003
1004         memset(&fl, 0, sizeof(fl));
1005         ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1006         ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1007
1008         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1009                                     tot_len, IPPROTO_TCP,
1010                                     buff->csum);
1011
1012         fl.proto = IPPROTO_TCP;
1013         fl.oif = inet6_iif(skb);
1014         fl.fl_ip_dport = t1->dest;
1015         fl.fl_ip_sport = t1->source;
1016         security_skb_classify_flow(skb, &fl);
1017
1018         /* Pass a socket to ip6_dst_lookup either it is for RST
1019          * Underlying function will use this to retrieve the network
1020          * namespace
1021          */
1022         if (!ip6_dst_lookup(ctl_sk, &buff->dst, &fl)) {
1023                 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1024                         ip6_xmit(ctl_sk, buff, &fl, NULL, 0);
1025                         TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1026                         if (rst)
1027                                 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1028                         return;
1029                 }
1030         }
1031
1032         kfree_skb(buff);
1033 }
1034
1035 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
1036 {
1037         struct tcphdr *th = tcp_hdr(skb);
1038         u32 seq = 0, ack_seq = 0;
1039         struct tcp_md5sig_key *key = NULL;
1040
1041         if (th->rst)
1042                 return;
1043
1044         if (!ipv6_unicast_destination(skb))
1045                 return;
1046
1047 #ifdef CONFIG_TCP_MD5SIG
1048         if (sk)
1049                 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1050 #endif
1051
1052         if (th->ack)
1053                 seq = ntohl(th->ack_seq);
1054         else
1055                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1056                           (th->doff << 2);
1057
1058         tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1);
1059 }
1060
1061 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1062                             struct tcp_md5sig_key *key)
1063 {
1064         tcp_v6_send_response(skb, seq, ack, win, ts, key, 0);
1065 }
1066
1067 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1068 {
1069         struct inet_timewait_sock *tw = inet_twsk(sk);
1070         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1071
1072         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1073                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1074                         tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw));
1075
1076         inet_twsk_put(tw);
1077 }
1078
1079 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1080                                   struct request_sock *req)
1081 {
1082         tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1083                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr));
1084 }
1085
1086
1087 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1088 {
1089         struct request_sock *req, **prev;
1090         const struct tcphdr *th = tcp_hdr(skb);
1091         struct sock *nsk;
1092
1093         /* Find possible connection requests. */
1094         req = inet6_csk_search_req(sk, &prev, th->source,
1095                                    &ipv6_hdr(skb)->saddr,
1096                                    &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1097         if (req)
1098                 return tcp_check_req(sk, skb, req, prev);
1099
1100         nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1101                         &ipv6_hdr(skb)->saddr, th->source,
1102                         &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1103
1104         if (nsk) {
1105                 if (nsk->sk_state != TCP_TIME_WAIT) {
1106                         bh_lock_sock(nsk);
1107                         return nsk;
1108                 }
1109                 inet_twsk_put(inet_twsk(nsk));
1110                 return NULL;
1111         }
1112
1113 #ifdef CONFIG_SYN_COOKIES
1114         if (!th->rst && !th->syn && th->ack)
1115                 sk = cookie_v6_check(sk, skb);
1116 #endif
1117         return sk;
1118 }
1119
1120 /* FIXME: this is substantially similar to the ipv4 code.
1121  * Can some kind of merge be done? -- erics
1122  */
1123 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1124 {
1125         struct inet6_request_sock *treq;
1126         struct ipv6_pinfo *np = inet6_sk(sk);
1127         struct tcp_options_received tmp_opt;
1128         struct tcp_sock *tp = tcp_sk(sk);
1129         struct request_sock *req = NULL;
1130         __u32 isn = TCP_SKB_CB(skb)->when;
1131 #ifdef CONFIG_SYN_COOKIES
1132         int want_cookie = 0;
1133 #else
1134 #define want_cookie 0
1135 #endif
1136
1137         if (skb->protocol == htons(ETH_P_IP))
1138                 return tcp_v4_conn_request(sk, skb);
1139
1140         if (!ipv6_unicast_destination(skb))
1141                 goto drop;
1142
1143         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1144                 if (net_ratelimit())
1145                         syn_flood_warning(skb);
1146 #ifdef CONFIG_SYN_COOKIES
1147                 if (sysctl_tcp_syncookies)
1148                         want_cookie = 1;
1149                 else
1150 #endif
1151                 goto drop;
1152         }
1153
1154         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1155                 goto drop;
1156
1157         req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1158         if (req == NULL)
1159                 goto drop;
1160
1161 #ifdef CONFIG_TCP_MD5SIG
1162         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1163 #endif
1164
1165         tcp_clear_options(&tmp_opt);
1166         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1167         tmp_opt.user_mss = tp->rx_opt.user_mss;
1168
1169         tcp_parse_options(skb, &tmp_opt, 0);
1170
1171         if (want_cookie && !tmp_opt.saw_tstamp)
1172                 tcp_clear_options(&tmp_opt);
1173
1174         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1175         tcp_openreq_init(req, &tmp_opt, skb);
1176
1177         treq = inet6_rsk(req);
1178         ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1179         ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1180         if (!want_cookie)
1181                 TCP_ECN_create_request(req, tcp_hdr(skb));
1182
1183         if (want_cookie) {
1184                 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1185                 req->cookie_ts = tmp_opt.tstamp_ok;
1186         } else if (!isn) {
1187                 if (ipv6_opt_accepted(sk, skb) ||
1188                     np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1189                     np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1190                         atomic_inc(&skb->users);
1191                         treq->pktopts = skb;
1192                 }
1193                 treq->iif = sk->sk_bound_dev_if;
1194
1195                 /* So that link locals have meaning */
1196                 if (!sk->sk_bound_dev_if &&
1197                     ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1198                         treq->iif = inet6_iif(skb);
1199
1200                 isn = tcp_v6_init_sequence(skb);
1201         }
1202
1203         tcp_rsk(req)->snt_isn = isn;
1204
1205         security_inet_conn_request(sk, skb, req);
1206
1207         if (tcp_v6_send_synack(sk, req))
1208                 goto drop;
1209
1210         if (!want_cookie) {
1211                 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1212                 return 0;
1213         }
1214
1215 drop:
1216         if (req)
1217                 reqsk_free(req);
1218
1219         return 0; /* don't send reset */
1220 }
1221
1222 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1223                                           struct request_sock *req,
1224                                           struct dst_entry *dst)
1225 {
1226         struct inet6_request_sock *treq;
1227         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1228         struct tcp6_sock *newtcp6sk;
1229         struct inet_sock *newinet;
1230         struct tcp_sock *newtp;
1231         struct sock *newsk;
1232         struct ipv6_txoptions *opt;
1233 #ifdef CONFIG_TCP_MD5SIG
1234         struct tcp_md5sig_key *key;
1235 #endif
1236
1237         if (skb->protocol == htons(ETH_P_IP)) {
1238                 /*
1239                  *      v6 mapped
1240                  */
1241
1242                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1243
1244                 if (newsk == NULL)
1245                         return NULL;
1246
1247                 newtcp6sk = (struct tcp6_sock *)newsk;
1248                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1249
1250                 newinet = inet_sk(newsk);
1251                 newnp = inet6_sk(newsk);
1252                 newtp = tcp_sk(newsk);
1253
1254                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1255
1256                 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1257                               newinet->daddr);
1258
1259                 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1260                               newinet->saddr);
1261
1262                 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1263
1264                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1265                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1266 #ifdef CONFIG_TCP_MD5SIG
1267                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1268 #endif
1269
1270                 newnp->pktoptions  = NULL;
1271                 newnp->opt         = NULL;
1272                 newnp->mcast_oif   = inet6_iif(skb);
1273                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1274
1275                 /*
1276                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1277                  * here, tcp_create_openreq_child now does this for us, see the comment in
1278                  * that function for the gory details. -acme
1279                  */
1280
1281                 /* It is tricky place. Until this moment IPv4 tcp
1282                    worked with IPv6 icsk.icsk_af_ops.
1283                    Sync it now.
1284                  */
1285                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1286
1287                 return newsk;
1288         }
1289
1290         treq = inet6_rsk(req);
1291         opt = np->opt;
1292
1293         if (sk_acceptq_is_full(sk))
1294                 goto out_overflow;
1295
1296         if (dst == NULL) {
1297                 struct in6_addr *final_p = NULL, final;
1298                 struct flowi fl;
1299
1300                 memset(&fl, 0, sizeof(fl));
1301                 fl.proto = IPPROTO_TCP;
1302                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1303                 if (opt && opt->srcrt) {
1304                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1305                         ipv6_addr_copy(&final, &fl.fl6_dst);
1306                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1307                         final_p = &final;
1308                 }
1309                 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1310                 fl.oif = sk->sk_bound_dev_if;
1311                 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1312                 fl.fl_ip_sport = inet_rsk(req)->loc_port;
1313                 security_req_classify_flow(req, &fl);
1314
1315                 if (ip6_dst_lookup(sk, &dst, &fl))
1316                         goto out;
1317
1318                 if (final_p)
1319                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1320
1321                 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1322                         goto out;
1323         }
1324
1325         newsk = tcp_create_openreq_child(sk, req, skb);
1326         if (newsk == NULL)
1327                 goto out;
1328
1329         /*
1330          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1331          * count here, tcp_create_openreq_child now does this for us, see the
1332          * comment in that function for the gory details. -acme
1333          */
1334
1335         newsk->sk_gso_type = SKB_GSO_TCPV6;
1336         __ip6_dst_store(newsk, dst, NULL, NULL);
1337
1338         newtcp6sk = (struct tcp6_sock *)newsk;
1339         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1340
1341         newtp = tcp_sk(newsk);
1342         newinet = inet_sk(newsk);
1343         newnp = inet6_sk(newsk);
1344
1345         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1346
1347         ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1348         ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1349         ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1350         newsk->sk_bound_dev_if = treq->iif;
1351
1352         /* Now IPv6 options...
1353
1354            First: no IPv4 options.
1355          */
1356         newinet->opt = NULL;
1357         newnp->ipv6_fl_list = NULL;
1358
1359         /* Clone RX bits */
1360         newnp->rxopt.all = np->rxopt.all;
1361
1362         /* Clone pktoptions received with SYN */
1363         newnp->pktoptions = NULL;
1364         if (treq->pktopts != NULL) {
1365                 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1366                 kfree_skb(treq->pktopts);
1367                 treq->pktopts = NULL;
1368                 if (newnp->pktoptions)
1369                         skb_set_owner_r(newnp->pktoptions, newsk);
1370         }
1371         newnp->opt        = NULL;
1372         newnp->mcast_oif  = inet6_iif(skb);
1373         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1374
1375         /* Clone native IPv6 options from listening socket (if any)
1376
1377            Yes, keeping reference count would be much more clever,
1378            but we make one more one thing there: reattach optmem
1379            to newsk.
1380          */
1381         if (opt) {
1382                 newnp->opt = ipv6_dup_options(newsk, opt);
1383                 if (opt != np->opt)
1384                         sock_kfree_s(sk, opt, opt->tot_len);
1385         }
1386
1387         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1388         if (newnp->opt)
1389                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1390                                                      newnp->opt->opt_flen);
1391
1392         tcp_mtup_init(newsk);
1393         tcp_sync_mss(newsk, dst_mtu(dst));
1394         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1395         tcp_initialize_rcv_mss(newsk);
1396
1397         newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1398
1399 #ifdef CONFIG_TCP_MD5SIG
1400         /* Copy over the MD5 key from the original socket */
1401         if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1402                 /* We're using one, so create a matching key
1403                  * on the newsk structure. If we fail to get
1404                  * memory, then we end up not copying the key
1405                  * across. Shucks.
1406                  */
1407                 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1408                 if (newkey != NULL)
1409                         tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1410                                           newkey, key->keylen);
1411         }
1412 #endif
1413
1414         __inet6_hash(newsk);
1415         __inet_inherit_port(sk, newsk);
1416
1417         return newsk;
1418
1419 out_overflow:
1420         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1421 out:
1422         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1423         if (opt && opt != np->opt)
1424                 sock_kfree_s(sk, opt, opt->tot_len);
1425         dst_release(dst);
1426         return NULL;
1427 }
1428
1429 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1430 {
1431         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1432                 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1433                                   &ipv6_hdr(skb)->daddr, skb->csum)) {
1434                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1435                         return 0;
1436                 }
1437         }
1438
1439         skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1440                                               &ipv6_hdr(skb)->saddr,
1441                                               &ipv6_hdr(skb)->daddr, 0));
1442
1443         if (skb->len <= 76) {
1444                 return __skb_checksum_complete(skb);
1445         }
1446         return 0;
1447 }
1448
1449 /* The socket must have it's spinlock held when we get
1450  * here.
1451  *
1452  * We have a potential double-lock case here, so even when
1453  * doing backlog processing we use the BH locking scheme.
1454  * This is because we cannot sleep with the original spinlock
1455  * held.
1456  */
1457 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1458 {
1459         struct ipv6_pinfo *np = inet6_sk(sk);
1460         struct tcp_sock *tp;
1461         struct sk_buff *opt_skb = NULL;
1462
1463         /* Imagine: socket is IPv6. IPv4 packet arrives,
1464            goes to IPv4 receive handler and backlogged.
1465            From backlog it always goes here. Kerboom...
1466            Fortunately, tcp_rcv_established and rcv_established
1467            handle them correctly, but it is not case with
1468            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1469          */
1470
1471         if (skb->protocol == htons(ETH_P_IP))
1472                 return tcp_v4_do_rcv(sk, skb);
1473
1474 #ifdef CONFIG_TCP_MD5SIG
1475         if (tcp_v6_inbound_md5_hash (sk, skb))
1476                 goto discard;
1477 #endif
1478
1479         if (sk_filter(sk, skb))
1480                 goto discard;
1481
1482         /*
1483          *      socket locking is here for SMP purposes as backlog rcv
1484          *      is currently called with bh processing disabled.
1485          */
1486
1487         /* Do Stevens' IPV6_PKTOPTIONS.
1488
1489            Yes, guys, it is the only place in our code, where we
1490            may make it not affecting IPv4.
1491            The rest of code is protocol independent,
1492            and I do not like idea to uglify IPv4.
1493
1494            Actually, all the idea behind IPV6_PKTOPTIONS
1495            looks not very well thought. For now we latch
1496            options, received in the last packet, enqueued
1497            by tcp. Feel free to propose better solution.
1498                                                --ANK (980728)
1499          */
1500         if (np->rxopt.all)
1501                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1502
1503         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1504                 TCP_CHECK_TIMER(sk);
1505                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1506                         goto reset;
1507                 TCP_CHECK_TIMER(sk);
1508                 if (opt_skb)
1509                         goto ipv6_pktoptions;
1510                 return 0;
1511         }
1512
1513         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1514                 goto csum_err;
1515
1516         if (sk->sk_state == TCP_LISTEN) {
1517                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1518                 if (!nsk)
1519                         goto discard;
1520
1521                 /*
1522                  * Queue it on the new socket if the new socket is active,
1523                  * otherwise we just shortcircuit this and continue with
1524                  * the new socket..
1525                  */
1526                 if(nsk != sk) {
1527                         if (tcp_child_process(sk, nsk, skb))
1528                                 goto reset;
1529                         if (opt_skb)
1530                                 __kfree_skb(opt_skb);
1531                         return 0;
1532                 }
1533         }
1534
1535         TCP_CHECK_TIMER(sk);
1536         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1537                 goto reset;
1538         TCP_CHECK_TIMER(sk);
1539         if (opt_skb)
1540                 goto ipv6_pktoptions;
1541         return 0;
1542
1543 reset:
1544         tcp_v6_send_reset(sk, skb);
1545 discard:
1546         if (opt_skb)
1547                 __kfree_skb(opt_skb);
1548         kfree_skb(skb);
1549         return 0;
1550 csum_err:
1551         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1552         goto discard;
1553
1554
1555 ipv6_pktoptions:
1556         /* Do you ask, what is it?
1557
1558            1. skb was enqueued by tcp.
1559            2. skb is added to tail of read queue, rather than out of order.
1560            3. socket is not in passive state.
1561            4. Finally, it really contains options, which user wants to receive.
1562          */
1563         tp = tcp_sk(sk);
1564         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1565             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1566                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1567                         np->mcast_oif = inet6_iif(opt_skb);
1568                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1569                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1570                 if (ipv6_opt_accepted(sk, opt_skb)) {
1571                         skb_set_owner_r(opt_skb, sk);
1572                         opt_skb = xchg(&np->pktoptions, opt_skb);
1573                 } else {
1574                         __kfree_skb(opt_skb);
1575                         opt_skb = xchg(&np->pktoptions, NULL);
1576                 }
1577         }
1578
1579         if (opt_skb)
1580                 kfree_skb(opt_skb);
1581         return 0;
1582 }
1583
1584 static int tcp_v6_rcv(struct sk_buff *skb)
1585 {
1586         struct tcphdr *th;
1587         struct sock *sk;
1588         int ret;
1589         struct net *net = dev_net(skb->dev);
1590
1591         if (skb->pkt_type != PACKET_HOST)
1592                 goto discard_it;
1593
1594         /*
1595          *      Count it even if it's bad.
1596          */
1597         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1598
1599         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1600                 goto discard_it;
1601
1602         th = tcp_hdr(skb);
1603
1604         if (th->doff < sizeof(struct tcphdr)/4)
1605                 goto bad_packet;
1606         if (!pskb_may_pull(skb, th->doff*4))
1607                 goto discard_it;
1608
1609         if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1610                 goto bad_packet;
1611
1612         th = tcp_hdr(skb);
1613         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1614         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1615                                     skb->len - th->doff*4);
1616         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1617         TCP_SKB_CB(skb)->when = 0;
1618         TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1619         TCP_SKB_CB(skb)->sacked = 0;
1620
1621         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1622         if (!sk)
1623                 goto no_tcp_socket;
1624
1625 process:
1626         if (sk->sk_state == TCP_TIME_WAIT)
1627                 goto do_time_wait;
1628
1629         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1630                 goto discard_and_relse;
1631
1632         if (sk_filter(sk, skb))
1633                 goto discard_and_relse;
1634
1635         skb->dev = NULL;
1636
1637         bh_lock_sock_nested(sk);
1638         ret = 0;
1639         if (!sock_owned_by_user(sk)) {
1640 #ifdef CONFIG_NET_DMA
1641                 struct tcp_sock *tp = tcp_sk(sk);
1642                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1643                         tp->ucopy.dma_chan = get_softnet_dma();
1644                 if (tp->ucopy.dma_chan)
1645                         ret = tcp_v6_do_rcv(sk, skb);
1646                 else
1647 #endif
1648                 {
1649                         if (!tcp_prequeue(sk, skb))
1650                                 ret = tcp_v6_do_rcv(sk, skb);
1651                 }
1652         } else
1653                 sk_add_backlog(sk, skb);
1654         bh_unlock_sock(sk);
1655
1656         sock_put(sk);
1657         return ret ? -1 : 0;
1658
1659 no_tcp_socket:
1660         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1661                 goto discard_it;
1662
1663         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1664 bad_packet:
1665                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1666         } else {
1667                 tcp_v6_send_reset(NULL, skb);
1668         }
1669
1670 discard_it:
1671
1672         /*
1673          *      Discard frame
1674          */
1675
1676         kfree_skb(skb);
1677         return 0;
1678
1679 discard_and_relse:
1680         sock_put(sk);
1681         goto discard_it;
1682
1683 do_time_wait:
1684         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1685                 inet_twsk_put(inet_twsk(sk));
1686                 goto discard_it;
1687         }
1688
1689         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1690                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1691                 inet_twsk_put(inet_twsk(sk));
1692                 goto discard_it;
1693         }
1694
1695         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1696         case TCP_TW_SYN:
1697         {
1698                 struct sock *sk2;
1699
1700                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1701                                             &ipv6_hdr(skb)->daddr,
1702                                             ntohs(th->dest), inet6_iif(skb));
1703                 if (sk2 != NULL) {
1704                         struct inet_timewait_sock *tw = inet_twsk(sk);
1705                         inet_twsk_deschedule(tw, &tcp_death_row);
1706                         inet_twsk_put(tw);
1707                         sk = sk2;
1708                         goto process;
1709                 }
1710                 /* Fall through to ACK */
1711         }
1712         case TCP_TW_ACK:
1713                 tcp_v6_timewait_ack(sk, skb);
1714                 break;
1715         case TCP_TW_RST:
1716                 goto no_tcp_socket;
1717         case TCP_TW_SUCCESS:;
1718         }
1719         goto discard_it;
1720 }
1721
1722 static int tcp_v6_remember_stamp(struct sock *sk)
1723 {
1724         /* Alas, not yet... */
1725         return 0;
1726 }
1727
1728 static struct inet_connection_sock_af_ops ipv6_specific = {
1729         .queue_xmit        = inet6_csk_xmit,
1730         .send_check        = tcp_v6_send_check,
1731         .rebuild_header    = inet6_sk_rebuild_header,
1732         .conn_request      = tcp_v6_conn_request,
1733         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1734         .remember_stamp    = tcp_v6_remember_stamp,
1735         .net_header_len    = sizeof(struct ipv6hdr),
1736         .setsockopt        = ipv6_setsockopt,
1737         .getsockopt        = ipv6_getsockopt,
1738         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1739         .sockaddr_len      = sizeof(struct sockaddr_in6),
1740         .bind_conflict     = inet6_csk_bind_conflict,
1741 #ifdef CONFIG_COMPAT
1742         .compat_setsockopt = compat_ipv6_setsockopt,
1743         .compat_getsockopt = compat_ipv6_getsockopt,
1744 #endif
1745 };
1746
1747 #ifdef CONFIG_TCP_MD5SIG
1748 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1749         .md5_lookup     =       tcp_v6_md5_lookup,
1750         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1751         .md5_add        =       tcp_v6_md5_add_func,
1752         .md5_parse      =       tcp_v6_parse_md5_keys,
1753 };
1754 #endif
1755
1756 /*
1757  *      TCP over IPv4 via INET6 API
1758  */
1759
1760 static struct inet_connection_sock_af_ops ipv6_mapped = {
1761         .queue_xmit        = ip_queue_xmit,
1762         .send_check        = tcp_v4_send_check,
1763         .rebuild_header    = inet_sk_rebuild_header,
1764         .conn_request      = tcp_v6_conn_request,
1765         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1766         .remember_stamp    = tcp_v4_remember_stamp,
1767         .net_header_len    = sizeof(struct iphdr),
1768         .setsockopt        = ipv6_setsockopt,
1769         .getsockopt        = ipv6_getsockopt,
1770         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1771         .sockaddr_len      = sizeof(struct sockaddr_in6),
1772         .bind_conflict     = inet6_csk_bind_conflict,
1773 #ifdef CONFIG_COMPAT
1774         .compat_setsockopt = compat_ipv6_setsockopt,
1775         .compat_getsockopt = compat_ipv6_getsockopt,
1776 #endif
1777 };
1778
1779 #ifdef CONFIG_TCP_MD5SIG
1780 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1781         .md5_lookup     =       tcp_v4_md5_lookup,
1782         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1783         .md5_add        =       tcp_v6_md5_add_func,
1784         .md5_parse      =       tcp_v6_parse_md5_keys,
1785 };
1786 #endif
1787
1788 /* NOTE: A lot of things set to zero explicitly by call to
1789  *       sk_alloc() so need not be done here.
1790  */
1791 static int tcp_v6_init_sock(struct sock *sk)
1792 {
1793         struct inet_connection_sock *icsk = inet_csk(sk);
1794         struct tcp_sock *tp = tcp_sk(sk);
1795
1796         skb_queue_head_init(&tp->out_of_order_queue);
1797         tcp_init_xmit_timers(sk);
1798         tcp_prequeue_init(tp);
1799
1800         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1801         tp->mdev = TCP_TIMEOUT_INIT;
1802
1803         /* So many TCP implementations out there (incorrectly) count the
1804          * initial SYN frame in their delayed-ACK and congestion control
1805          * algorithms that we must have the following bandaid to talk
1806          * efficiently to them.  -DaveM
1807          */
1808         tp->snd_cwnd = 2;
1809
1810         /* See draft-stevens-tcpca-spec-01 for discussion of the
1811          * initialization of these values.
1812          */
1813         tp->snd_ssthresh = 0x7fffffff;
1814         tp->snd_cwnd_clamp = ~0;
1815         tp->mss_cache = 536;
1816
1817         tp->reordering = sysctl_tcp_reordering;
1818
1819         sk->sk_state = TCP_CLOSE;
1820
1821         icsk->icsk_af_ops = &ipv6_specific;
1822         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1823         icsk->icsk_sync_mss = tcp_sync_mss;
1824         sk->sk_write_space = sk_stream_write_space;
1825         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1826
1827 #ifdef CONFIG_TCP_MD5SIG
1828         tp->af_specific = &tcp_sock_ipv6_specific;
1829 #endif
1830
1831         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1832         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1833
1834         atomic_inc(&tcp_sockets_allocated);
1835
1836         return 0;
1837 }
1838
1839 static void tcp_v6_destroy_sock(struct sock *sk)
1840 {
1841 #ifdef CONFIG_TCP_MD5SIG
1842         /* Clean up the MD5 key list */
1843         if (tcp_sk(sk)->md5sig_info)
1844                 tcp_v6_clear_md5_list(sk);
1845 #endif
1846         tcp_v4_destroy_sock(sk);
1847         inet6_destroy_sock(sk);
1848 }
1849
1850 #ifdef CONFIG_PROC_FS
1851 /* Proc filesystem TCPv6 sock list dumping. */
1852 static void get_openreq6(struct seq_file *seq,
1853                          struct sock *sk, struct request_sock *req, int i, int uid)
1854 {
1855         int ttd = req->expires - jiffies;
1856         struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1857         struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1858
1859         if (ttd < 0)
1860                 ttd = 0;
1861
1862         seq_printf(seq,
1863                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1864                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1865                    i,
1866                    src->s6_addr32[0], src->s6_addr32[1],
1867                    src->s6_addr32[2], src->s6_addr32[3],
1868                    ntohs(inet_rsk(req)->loc_port),
1869                    dest->s6_addr32[0], dest->s6_addr32[1],
1870                    dest->s6_addr32[2], dest->s6_addr32[3],
1871                    ntohs(inet_rsk(req)->rmt_port),
1872                    TCP_SYN_RECV,
1873                    0,0, /* could print option size, but that is af dependent. */
1874                    1,   /* timers active (only the expire timer) */
1875                    jiffies_to_clock_t(ttd),
1876                    req->retrans,
1877                    uid,
1878                    0,  /* non standard timer */
1879                    0, /* open_requests have no inode */
1880                    0, req);
1881 }
1882
1883 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1884 {
1885         struct in6_addr *dest, *src;
1886         __u16 destp, srcp;
1887         int timer_active;
1888         unsigned long timer_expires;
1889         struct inet_sock *inet = inet_sk(sp);
1890         struct tcp_sock *tp = tcp_sk(sp);
1891         const struct inet_connection_sock *icsk = inet_csk(sp);
1892         struct ipv6_pinfo *np = inet6_sk(sp);
1893
1894         dest  = &np->daddr;
1895         src   = &np->rcv_saddr;
1896         destp = ntohs(inet->dport);
1897         srcp  = ntohs(inet->sport);
1898
1899         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1900                 timer_active    = 1;
1901                 timer_expires   = icsk->icsk_timeout;
1902         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1903                 timer_active    = 4;
1904                 timer_expires   = icsk->icsk_timeout;
1905         } else if (timer_pending(&sp->sk_timer)) {
1906                 timer_active    = 2;
1907                 timer_expires   = sp->sk_timer.expires;
1908         } else {
1909                 timer_active    = 0;
1910                 timer_expires = jiffies;
1911         }
1912
1913         seq_printf(seq,
1914                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1915                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %lu %lu %u %u %d\n",
1916                    i,
1917                    src->s6_addr32[0], src->s6_addr32[1],
1918                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1919                    dest->s6_addr32[0], dest->s6_addr32[1],
1920                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1921                    sp->sk_state,
1922                    tp->write_seq-tp->snd_una,
1923                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1924                    timer_active,
1925                    jiffies_to_clock_t(timer_expires - jiffies),
1926                    icsk->icsk_retransmits,
1927                    sock_i_uid(sp),
1928                    icsk->icsk_probes_out,
1929                    sock_i_ino(sp),
1930                    atomic_read(&sp->sk_refcnt), sp,
1931                    jiffies_to_clock_t(icsk->icsk_rto),
1932                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1933                    (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1934                    tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
1935                    );
1936 }
1937
1938 static void get_timewait6_sock(struct seq_file *seq,
1939                                struct inet_timewait_sock *tw, int i)
1940 {
1941         struct in6_addr *dest, *src;
1942         __u16 destp, srcp;
1943         struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
1944         int ttd = tw->tw_ttd - jiffies;
1945
1946         if (ttd < 0)
1947                 ttd = 0;
1948
1949         dest = &tw6->tw_v6_daddr;
1950         src  = &tw6->tw_v6_rcv_saddr;
1951         destp = ntohs(tw->tw_dport);
1952         srcp  = ntohs(tw->tw_sport);
1953
1954         seq_printf(seq,
1955                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1956                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1957                    i,
1958                    src->s6_addr32[0], src->s6_addr32[1],
1959                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1960                    dest->s6_addr32[0], dest->s6_addr32[1],
1961                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1962                    tw->tw_substate, 0, 0,
1963                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
1964                    atomic_read(&tw->tw_refcnt), tw);
1965 }
1966
1967 static int tcp6_seq_show(struct seq_file *seq, void *v)
1968 {
1969         struct tcp_iter_state *st;
1970
1971         if (v == SEQ_START_TOKEN) {
1972                 seq_puts(seq,
1973                          "  sl  "
1974                          "local_address                         "
1975                          "remote_address                        "
1976                          "st tx_queue rx_queue tr tm->when retrnsmt"
1977                          "   uid  timeout inode\n");
1978                 goto out;
1979         }
1980         st = seq->private;
1981
1982         switch (st->state) {
1983         case TCP_SEQ_STATE_LISTENING:
1984         case TCP_SEQ_STATE_ESTABLISHED:
1985                 get_tcp6_sock(seq, v, st->num);
1986                 break;
1987         case TCP_SEQ_STATE_OPENREQ:
1988                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1989                 break;
1990         case TCP_SEQ_STATE_TIME_WAIT:
1991                 get_timewait6_sock(seq, v, st->num);
1992                 break;
1993         }
1994 out:
1995         return 0;
1996 }
1997
1998 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1999         .name           = "tcp6",
2000         .family         = AF_INET6,
2001         .seq_fops       = {
2002                 .owner          = THIS_MODULE,
2003         },
2004         .seq_ops        = {
2005                 .show           = tcp6_seq_show,
2006         },
2007 };
2008
2009 int tcp6_proc_init(struct net *net)
2010 {
2011         return tcp_proc_register(net, &tcp6_seq_afinfo);
2012 }
2013
2014 void tcp6_proc_exit(struct net *net)
2015 {
2016         tcp_proc_unregister(net, &tcp6_seq_afinfo);
2017 }
2018 #endif
2019
2020 struct proto tcpv6_prot = {
2021         .name                   = "TCPv6",
2022         .owner                  = THIS_MODULE,
2023         .close                  = tcp_close,
2024         .connect                = tcp_v6_connect,
2025         .disconnect             = tcp_disconnect,
2026         .accept                 = inet_csk_accept,
2027         .ioctl                  = tcp_ioctl,
2028         .init                   = tcp_v6_init_sock,
2029         .destroy                = tcp_v6_destroy_sock,
2030         .shutdown               = tcp_shutdown,
2031         .setsockopt             = tcp_setsockopt,
2032         .getsockopt             = tcp_getsockopt,
2033         .recvmsg                = tcp_recvmsg,
2034         .backlog_rcv            = tcp_v6_do_rcv,
2035         .hash                   = tcp_v6_hash,
2036         .unhash                 = inet_unhash,
2037         .get_port               = inet_csk_get_port,
2038         .enter_memory_pressure  = tcp_enter_memory_pressure,
2039         .sockets_allocated      = &tcp_sockets_allocated,
2040         .memory_allocated       = &tcp_memory_allocated,
2041         .memory_pressure        = &tcp_memory_pressure,
2042         .orphan_count           = &tcp_orphan_count,
2043         .sysctl_mem             = sysctl_tcp_mem,
2044         .sysctl_wmem            = sysctl_tcp_wmem,
2045         .sysctl_rmem            = sysctl_tcp_rmem,
2046         .max_header             = MAX_TCP_HEADER,
2047         .obj_size               = sizeof(struct tcp6_sock),
2048         .twsk_prot              = &tcp6_timewait_sock_ops,
2049         .rsk_prot               = &tcp6_request_sock_ops,
2050         .h.hashinfo             = &tcp_hashinfo,
2051 #ifdef CONFIG_COMPAT
2052         .compat_setsockopt      = compat_tcp_setsockopt,
2053         .compat_getsockopt      = compat_tcp_getsockopt,
2054 #endif
2055 };
2056
2057 static struct inet6_protocol tcpv6_protocol = {
2058         .handler        =       tcp_v6_rcv,
2059         .err_handler    =       tcp_v6_err,
2060         .gso_send_check =       tcp_v6_gso_send_check,
2061         .gso_segment    =       tcp_tso_segment,
2062         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2063 };
2064
2065 static struct inet_protosw tcpv6_protosw = {
2066         .type           =       SOCK_STREAM,
2067         .protocol       =       IPPROTO_TCP,
2068         .prot           =       &tcpv6_prot,
2069         .ops            =       &inet6_stream_ops,
2070         .capability     =       -1,
2071         .no_check       =       0,
2072         .flags          =       INET_PROTOSW_PERMANENT |
2073                                 INET_PROTOSW_ICSK,
2074 };
2075
2076 static int tcpv6_net_init(struct net *net)
2077 {
2078         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2079                                     SOCK_RAW, IPPROTO_TCP, net);
2080 }
2081
2082 static void tcpv6_net_exit(struct net *net)
2083 {
2084         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2085         inet_twsk_purge(net, &tcp_hashinfo, &tcp_death_row, AF_INET6);
2086 }
2087
2088 static struct pernet_operations tcpv6_net_ops = {
2089         .init = tcpv6_net_init,
2090         .exit = tcpv6_net_exit,
2091 };
2092
2093 int __init tcpv6_init(void)
2094 {
2095         int ret;
2096
2097         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2098         if (ret)
2099                 goto out;
2100
2101         /* register inet6 protocol */
2102         ret = inet6_register_protosw(&tcpv6_protosw);
2103         if (ret)
2104                 goto out_tcpv6_protocol;
2105
2106         ret = register_pernet_subsys(&tcpv6_net_ops);
2107         if (ret)
2108                 goto out_tcpv6_protosw;
2109 out:
2110         return ret;
2111
2112 out_tcpv6_protocol:
2113         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2114 out_tcpv6_protosw:
2115         inet6_unregister_protosw(&tcpv6_protosw);
2116         goto out;
2117 }
2118
2119 void tcpv6_exit(void)
2120 {
2121         unregister_pernet_subsys(&tcpv6_net_ops);
2122         inet6_unregister_protosw(&tcpv6_protosw);
2123         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2124 }