Merge branch 'virtex-for-2.6.25' of git://git.secretlab.ca/git/linux-2.6-virtex into...
[linux-2.6] / 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  *      $Id: tcp_ipv6.c,v 1.144 2002/02/01 22:01:04 davem Exp $
9  *
10  *      Based on:
11  *      linux/net/ipv4/tcp.c
12  *      linux/net/ipv4/tcp_input.c
13  *      linux/net/ipv4/tcp_output.c
14  *
15  *      Fixes:
16  *      Hideaki YOSHIFUJI       :       sin6_scope_id support
17  *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
18  *      Alexey Kuznetsov                allow both IPv4 and IPv6 sockets to bind
19  *                                      a single port at the same time.
20  *      YOSHIFUJI Hideaki @USAGI:       convert /proc/net/tcp6 to seq_file.
21  *
22  *      This program is free software; you can redistribute it and/or
23  *      modify it under the terms of the GNU General Public License
24  *      as published by the Free Software Foundation; either version
25  *      2 of the License, or (at your option) any later version.
26  */
27
28 #include <linux/module.h>
29 #include <linux/errno.h>
30 #include <linux/types.h>
31 #include <linux/socket.h>
32 #include <linux/sockios.h>
33 #include <linux/net.h>
34 #include <linux/jiffies.h>
35 #include <linux/in.h>
36 #include <linux/in6.h>
37 #include <linux/netdevice.h>
38 #include <linux/init.h>
39 #include <linux/jhash.h>
40 #include <linux/ipsec.h>
41 #include <linux/times.h>
42
43 #include <linux/ipv6.h>
44 #include <linux/icmpv6.h>
45 #include <linux/random.h>
46
47 #include <net/tcp.h>
48 #include <net/ndisc.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/inet6_connection_sock.h>
51 #include <net/ipv6.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>
58 #include <net/xfrm.h>
59 #include <net/snmp.h>
60 #include <net/dsfield.h>
61 #include <net/timewait_sock.h>
62 #include <net/netdma.h>
63
64 #include <asm/uaccess.h>
65
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
68
69 #include <linux/crypto.h>
70 #include <linux/scatterlist.h>
71
72 /* Socket used for sending RSTs and ACKs */
73 static struct socket *tcp6_socket;
74
75 static void     tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
76 static void     tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req);
77 static void     tcp_v6_send_check(struct sock *sk, int len,
78                                   struct sk_buff *skb);
79
80 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
81
82 static struct inet_connection_sock_af_ops ipv6_mapped;
83 static struct inet_connection_sock_af_ops ipv6_specific;
84 #ifdef CONFIG_TCP_MD5SIG
85 static struct tcp_sock_af_ops tcp_sock_ipv6_specific;
86 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
87 #endif
88
89 static void tcp_v6_hash(struct sock *sk)
90 {
91         if (sk->sk_state != TCP_CLOSE) {
92                 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
93                         tcp_prot.hash(sk);
94                         return;
95                 }
96                 local_bh_disable();
97                 __inet6_hash(sk);
98                 local_bh_enable();
99         }
100 }
101
102 static __inline__ __sum16 tcp_v6_check(struct tcphdr *th, int len,
103                                    struct in6_addr *saddr,
104                                    struct in6_addr *daddr,
105                                    __wsum base)
106 {
107         return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
108 }
109
110 static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
111 {
112         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
113                                             ipv6_hdr(skb)->saddr.s6_addr32,
114                                             tcp_hdr(skb)->dest,
115                                             tcp_hdr(skb)->source);
116 }
117
118 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
119                           int addr_len)
120 {
121         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
122         struct inet_sock *inet = inet_sk(sk);
123         struct inet_connection_sock *icsk = inet_csk(sk);
124         struct ipv6_pinfo *np = inet6_sk(sk);
125         struct tcp_sock *tp = tcp_sk(sk);
126         struct in6_addr *saddr = NULL, *final_p = NULL, final;
127         struct flowi fl;
128         struct dst_entry *dst;
129         int addr_type;
130         int err;
131
132         if (addr_len < SIN6_LEN_RFC2133)
133                 return -EINVAL;
134
135         if (usin->sin6_family != AF_INET6)
136                 return(-EAFNOSUPPORT);
137
138         memset(&fl, 0, sizeof(fl));
139
140         if (np->sndflow) {
141                 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
142                 IP6_ECN_flow_init(fl.fl6_flowlabel);
143                 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
144                         struct ip6_flowlabel *flowlabel;
145                         flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
146                         if (flowlabel == NULL)
147                                 return -EINVAL;
148                         ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
149                         fl6_sock_release(flowlabel);
150                 }
151         }
152
153         /*
154          *      connect() to INADDR_ANY means loopback (BSD'ism).
155          */
156
157         if(ipv6_addr_any(&usin->sin6_addr))
158                 usin->sin6_addr.s6_addr[15] = 0x1;
159
160         addr_type = ipv6_addr_type(&usin->sin6_addr);
161
162         if(addr_type & IPV6_ADDR_MULTICAST)
163                 return -ENETUNREACH;
164
165         if (addr_type&IPV6_ADDR_LINKLOCAL) {
166                 if (addr_len >= sizeof(struct sockaddr_in6) &&
167                     usin->sin6_scope_id) {
168                         /* If interface is set while binding, indices
169                          * must coincide.
170                          */
171                         if (sk->sk_bound_dev_if &&
172                             sk->sk_bound_dev_if != usin->sin6_scope_id)
173                                 return -EINVAL;
174
175                         sk->sk_bound_dev_if = usin->sin6_scope_id;
176                 }
177
178                 /* Connect to link-local address requires an interface */
179                 if (!sk->sk_bound_dev_if)
180                         return -EINVAL;
181         }
182
183         if (tp->rx_opt.ts_recent_stamp &&
184             !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
185                 tp->rx_opt.ts_recent = 0;
186                 tp->rx_opt.ts_recent_stamp = 0;
187                 tp->write_seq = 0;
188         }
189
190         ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
191         np->flow_label = fl.fl6_flowlabel;
192
193         /*
194          *      TCP over IPv4
195          */
196
197         if (addr_type == IPV6_ADDR_MAPPED) {
198                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
199                 struct sockaddr_in sin;
200
201                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
202
203                 if (__ipv6_only_sock(sk))
204                         return -ENETUNREACH;
205
206                 sin.sin_family = AF_INET;
207                 sin.sin_port = usin->sin6_port;
208                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
209
210                 icsk->icsk_af_ops = &ipv6_mapped;
211                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
212 #ifdef CONFIG_TCP_MD5SIG
213                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
214 #endif
215
216                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
217
218                 if (err) {
219                         icsk->icsk_ext_hdr_len = exthdrlen;
220                         icsk->icsk_af_ops = &ipv6_specific;
221                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
222 #ifdef CONFIG_TCP_MD5SIG
223                         tp->af_specific = &tcp_sock_ipv6_specific;
224 #endif
225                         goto failure;
226                 } else {
227                         ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
228                                       inet->saddr);
229                         ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
230                                       inet->rcv_saddr);
231                 }
232
233                 return err;
234         }
235
236         if (!ipv6_addr_any(&np->rcv_saddr))
237                 saddr = &np->rcv_saddr;
238
239         fl.proto = IPPROTO_TCP;
240         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
241         ipv6_addr_copy(&fl.fl6_src,
242                        (saddr ? saddr : &np->saddr));
243         fl.oif = sk->sk_bound_dev_if;
244         fl.fl_ip_dport = usin->sin6_port;
245         fl.fl_ip_sport = inet->sport;
246
247         if (np->opt && np->opt->srcrt) {
248                 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
249                 ipv6_addr_copy(&final, &fl.fl6_dst);
250                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
251                 final_p = &final;
252         }
253
254         security_sk_classify_flow(sk, &fl);
255
256         err = ip6_dst_lookup(sk, &dst, &fl);
257         if (err)
258                 goto failure;
259         if (final_p)
260                 ipv6_addr_copy(&fl.fl6_dst, final_p);
261
262         if ((err = __xfrm_lookup(&dst, &fl, sk, XFRM_LOOKUP_WAIT)) < 0) {
263                 if (err == -EREMOTE)
264                         err = ip6_dst_blackhole(sk, &dst, &fl);
265                 if (err < 0)
266                         goto failure;
267         }
268
269         if (saddr == NULL) {
270                 saddr = &fl.fl6_src;
271                 ipv6_addr_copy(&np->rcv_saddr, saddr);
272         }
273
274         /* set the source address */
275         ipv6_addr_copy(&np->saddr, saddr);
276         inet->rcv_saddr = LOOPBACK4_IPV6;
277
278         sk->sk_gso_type = SKB_GSO_TCPV6;
279         __ip6_dst_store(sk, dst, NULL, NULL);
280
281         icsk->icsk_ext_hdr_len = 0;
282         if (np->opt)
283                 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
284                                           np->opt->opt_nflen);
285
286         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
287
288         inet->dport = usin->sin6_port;
289
290         tcp_set_state(sk, TCP_SYN_SENT);
291         err = inet6_hash_connect(&tcp_death_row, sk);
292         if (err)
293                 goto late_failure;
294
295         if (!tp->write_seq)
296                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
297                                                              np->daddr.s6_addr32,
298                                                              inet->sport,
299                                                              inet->dport);
300
301         err = tcp_connect(sk);
302         if (err)
303                 goto late_failure;
304
305         return 0;
306
307 late_failure:
308         tcp_set_state(sk, TCP_CLOSE);
309         __sk_dst_reset(sk);
310 failure:
311         inet->dport = 0;
312         sk->sk_route_caps = 0;
313         return err;
314 }
315
316 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
317                 int type, int code, int offset, __be32 info)
318 {
319         struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
320         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
321         struct ipv6_pinfo *np;
322         struct sock *sk;
323         int err;
324         struct tcp_sock *tp;
325         __u32 seq;
326
327         sk = inet6_lookup(skb->dev->nd_net, &tcp_hashinfo, &hdr->daddr,
328                         th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
329
330         if (sk == NULL) {
331                 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
332                 return;
333         }
334
335         if (sk->sk_state == TCP_TIME_WAIT) {
336                 inet_twsk_put(inet_twsk(sk));
337                 return;
338         }
339
340         bh_lock_sock(sk);
341         if (sock_owned_by_user(sk))
342                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
343
344         if (sk->sk_state == TCP_CLOSE)
345                 goto out;
346
347         tp = tcp_sk(sk);
348         seq = ntohl(th->seq);
349         if (sk->sk_state != TCP_LISTEN &&
350             !between(seq, tp->snd_una, tp->snd_nxt)) {
351                 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
352                 goto out;
353         }
354
355         np = inet6_sk(sk);
356
357         if (type == ICMPV6_PKT_TOOBIG) {
358                 struct dst_entry *dst = NULL;
359
360                 if (sock_owned_by_user(sk))
361                         goto out;
362                 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
363                         goto out;
364
365                 /* icmp should have updated the destination cache entry */
366                 dst = __sk_dst_check(sk, np->dst_cookie);
367
368                 if (dst == NULL) {
369                         struct inet_sock *inet = inet_sk(sk);
370                         struct flowi fl;
371
372                         /* BUGGG_FUTURE: Again, it is not clear how
373                            to handle rthdr case. Ignore this complexity
374                            for now.
375                          */
376                         memset(&fl, 0, sizeof(fl));
377                         fl.proto = IPPROTO_TCP;
378                         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
379                         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
380                         fl.oif = sk->sk_bound_dev_if;
381                         fl.fl_ip_dport = inet->dport;
382                         fl.fl_ip_sport = inet->sport;
383                         security_skb_classify_flow(skb, &fl);
384
385                         if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
386                                 sk->sk_err_soft = -err;
387                                 goto out;
388                         }
389
390                         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
391                                 sk->sk_err_soft = -err;
392                                 goto out;
393                         }
394
395                 } else
396                         dst_hold(dst);
397
398                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
399                         tcp_sync_mss(sk, dst_mtu(dst));
400                         tcp_simple_retransmit(sk);
401                 } /* else let the usual retransmit timer handle it */
402                 dst_release(dst);
403                 goto out;
404         }
405
406         icmpv6_err_convert(type, code, &err);
407
408         /* Might be for an request_sock */
409         switch (sk->sk_state) {
410                 struct request_sock *req, **prev;
411         case TCP_LISTEN:
412                 if (sock_owned_by_user(sk))
413                         goto out;
414
415                 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
416                                            &hdr->saddr, inet6_iif(skb));
417                 if (!req)
418                         goto out;
419
420                 /* ICMPs are not backlogged, hence we cannot get
421                  * an established socket here.
422                  */
423                 BUG_TRAP(req->sk == NULL);
424
425                 if (seq != tcp_rsk(req)->snt_isn) {
426                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
427                         goto out;
428                 }
429
430                 inet_csk_reqsk_queue_drop(sk, req, prev);
431                 goto out;
432
433         case TCP_SYN_SENT:
434         case TCP_SYN_RECV:  /* Cannot happen.
435                                It can, it SYNs are crossed. --ANK */
436                 if (!sock_owned_by_user(sk)) {
437                         sk->sk_err = err;
438                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
439
440                         tcp_done(sk);
441                 } else
442                         sk->sk_err_soft = err;
443                 goto out;
444         }
445
446         if (!sock_owned_by_user(sk) && np->recverr) {
447                 sk->sk_err = err;
448                 sk->sk_error_report(sk);
449         } else
450                 sk->sk_err_soft = err;
451
452 out:
453         bh_unlock_sock(sk);
454         sock_put(sk);
455 }
456
457
458 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
459                               struct dst_entry *dst)
460 {
461         struct inet6_request_sock *treq = inet6_rsk(req);
462         struct ipv6_pinfo *np = inet6_sk(sk);
463         struct sk_buff * skb;
464         struct ipv6_txoptions *opt = NULL;
465         struct in6_addr * final_p = NULL, final;
466         struct flowi fl;
467         int err = -1;
468
469         memset(&fl, 0, sizeof(fl));
470         fl.proto = IPPROTO_TCP;
471         ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
472         ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
473         fl.fl6_flowlabel = 0;
474         fl.oif = treq->iif;
475         fl.fl_ip_dport = inet_rsk(req)->rmt_port;
476         fl.fl_ip_sport = inet_sk(sk)->sport;
477         security_req_classify_flow(req, &fl);
478
479         if (dst == NULL) {
480                 opt = np->opt;
481                 if (opt && opt->srcrt) {
482                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
483                         ipv6_addr_copy(&final, &fl.fl6_dst);
484                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
485                         final_p = &final;
486                 }
487
488                 err = ip6_dst_lookup(sk, &dst, &fl);
489                 if (err)
490                         goto done;
491                 if (final_p)
492                         ipv6_addr_copy(&fl.fl6_dst, final_p);
493                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
494                         goto done;
495         }
496
497         skb = tcp_make_synack(sk, dst, req);
498         if (skb) {
499                 struct tcphdr *th = tcp_hdr(skb);
500
501                 th->check = tcp_v6_check(th, skb->len,
502                                          &treq->loc_addr, &treq->rmt_addr,
503                                          csum_partial((char *)th, skb->len, skb->csum));
504
505                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
506                 err = ip6_xmit(sk, skb, &fl, opt, 0);
507                 err = net_xmit_eval(err);
508         }
509
510 done:
511         if (opt && opt != np->opt)
512                 sock_kfree_s(sk, opt, opt->tot_len);
513         dst_release(dst);
514         return err;
515 }
516
517 static void tcp_v6_reqsk_destructor(struct request_sock *req)
518 {
519         if (inet6_rsk(req)->pktopts)
520                 kfree_skb(inet6_rsk(req)->pktopts);
521 }
522
523 #ifdef CONFIG_TCP_MD5SIG
524 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
525                                                    struct in6_addr *addr)
526 {
527         struct tcp_sock *tp = tcp_sk(sk);
528         int i;
529
530         BUG_ON(tp == NULL);
531
532         if (!tp->md5sig_info || !tp->md5sig_info->entries6)
533                 return NULL;
534
535         for (i = 0; i < tp->md5sig_info->entries6; i++) {
536                 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, addr) == 0)
537                         return &tp->md5sig_info->keys6[i].base;
538         }
539         return NULL;
540 }
541
542 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
543                                                 struct sock *addr_sk)
544 {
545         return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
546 }
547
548 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
549                                                       struct request_sock *req)
550 {
551         return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
552 }
553
554 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
555                              char *newkey, u8 newkeylen)
556 {
557         /* Add key to the list */
558         struct tcp_md5sig_key *key;
559         struct tcp_sock *tp = tcp_sk(sk);
560         struct tcp6_md5sig_key *keys;
561
562         key = tcp_v6_md5_do_lookup(sk, peer);
563         if (key) {
564                 /* modify existing entry - just update that one */
565                 kfree(key->key);
566                 key->key = newkey;
567                 key->keylen = newkeylen;
568         } else {
569                 /* reallocate new list if current one is full. */
570                 if (!tp->md5sig_info) {
571                         tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
572                         if (!tp->md5sig_info) {
573                                 kfree(newkey);
574                                 return -ENOMEM;
575                         }
576                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
577                 }
578                 if (tcp_alloc_md5sig_pool() == NULL) {
579                         kfree(newkey);
580                         return -ENOMEM;
581                 }
582                 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
583                         keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
584                                        (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
585
586                         if (!keys) {
587                                 tcp_free_md5sig_pool();
588                                 kfree(newkey);
589                                 return -ENOMEM;
590                         }
591
592                         if (tp->md5sig_info->entries6)
593                                 memmove(keys, tp->md5sig_info->keys6,
594                                         (sizeof (tp->md5sig_info->keys6[0]) *
595                                          tp->md5sig_info->entries6));
596
597                         kfree(tp->md5sig_info->keys6);
598                         tp->md5sig_info->keys6 = keys;
599                         tp->md5sig_info->alloced6++;
600                 }
601
602                 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
603                                peer);
604                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
605                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
606
607                 tp->md5sig_info->entries6++;
608         }
609         return 0;
610 }
611
612 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
613                                u8 *newkey, __u8 newkeylen)
614 {
615         return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
616                                  newkey, newkeylen);
617 }
618
619 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
620 {
621         struct tcp_sock *tp = tcp_sk(sk);
622         int i;
623
624         for (i = 0; i < tp->md5sig_info->entries6; i++) {
625                 if (ipv6_addr_cmp(&tp->md5sig_info->keys6[i].addr, peer) == 0) {
626                         /* Free the key */
627                         kfree(tp->md5sig_info->keys6[i].base.key);
628                         tp->md5sig_info->entries6--;
629
630                         if (tp->md5sig_info->entries6 == 0) {
631                                 kfree(tp->md5sig_info->keys6);
632                                 tp->md5sig_info->keys6 = NULL;
633                                 tp->md5sig_info->alloced6 = 0;
634                         } else {
635                                 /* shrink the database */
636                                 if (tp->md5sig_info->entries6 != i)
637                                         memmove(&tp->md5sig_info->keys6[i],
638                                                 &tp->md5sig_info->keys6[i+1],
639                                                 (tp->md5sig_info->entries6 - i)
640                                                 * sizeof (tp->md5sig_info->keys6[0]));
641                         }
642                         tcp_free_md5sig_pool();
643                         return 0;
644                 }
645         }
646         return -ENOENT;
647 }
648
649 static void tcp_v6_clear_md5_list (struct sock *sk)
650 {
651         struct tcp_sock *tp = tcp_sk(sk);
652         int i;
653
654         if (tp->md5sig_info->entries6) {
655                 for (i = 0; i < tp->md5sig_info->entries6; i++)
656                         kfree(tp->md5sig_info->keys6[i].base.key);
657                 tp->md5sig_info->entries6 = 0;
658                 tcp_free_md5sig_pool();
659         }
660
661         kfree(tp->md5sig_info->keys6);
662         tp->md5sig_info->keys6 = NULL;
663         tp->md5sig_info->alloced6 = 0;
664
665         if (tp->md5sig_info->entries4) {
666                 for (i = 0; i < tp->md5sig_info->entries4; i++)
667                         kfree(tp->md5sig_info->keys4[i].base.key);
668                 tp->md5sig_info->entries4 = 0;
669                 tcp_free_md5sig_pool();
670         }
671
672         kfree(tp->md5sig_info->keys4);
673         tp->md5sig_info->keys4 = NULL;
674         tp->md5sig_info->alloced4 = 0;
675 }
676
677 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
678                                   int optlen)
679 {
680         struct tcp_md5sig cmd;
681         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
682         u8 *newkey;
683
684         if (optlen < sizeof(cmd))
685                 return -EINVAL;
686
687         if (copy_from_user(&cmd, optval, sizeof(cmd)))
688                 return -EFAULT;
689
690         if (sin6->sin6_family != AF_INET6)
691                 return -EINVAL;
692
693         if (!cmd.tcpm_keylen) {
694                 if (!tcp_sk(sk)->md5sig_info)
695                         return -ENOENT;
696                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
697                         return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
698                 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
699         }
700
701         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
702                 return -EINVAL;
703
704         if (!tcp_sk(sk)->md5sig_info) {
705                 struct tcp_sock *tp = tcp_sk(sk);
706                 struct tcp_md5sig_info *p;
707
708                 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
709                 if (!p)
710                         return -ENOMEM;
711
712                 tp->md5sig_info = p;
713                 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
714         }
715
716         newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
717         if (!newkey)
718                 return -ENOMEM;
719         if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
720                 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
721                                          newkey, cmd.tcpm_keylen);
722         }
723         return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
724 }
725
726 static int tcp_v6_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
727                                    struct in6_addr *saddr,
728                                    struct in6_addr *daddr,
729                                    struct tcphdr *th, int protocol,
730                                    unsigned int tcplen)
731 {
732         struct scatterlist sg[4];
733         __u16 data_len;
734         int block = 0;
735         __sum16 cksum;
736         struct tcp_md5sig_pool *hp;
737         struct tcp6_pseudohdr *bp;
738         struct hash_desc *desc;
739         int err;
740         unsigned int nbytes = 0;
741
742         hp = tcp_get_md5sig_pool();
743         if (!hp) {
744                 printk(KERN_WARNING "%s(): hash pool not found...\n", __FUNCTION__);
745                 goto clear_hash_noput;
746         }
747         bp = &hp->md5_blk.ip6;
748         desc = &hp->md5_desc;
749
750         /* 1. TCP pseudo-header (RFC2460) */
751         ipv6_addr_copy(&bp->saddr, saddr);
752         ipv6_addr_copy(&bp->daddr, daddr);
753         bp->len = htonl(tcplen);
754         bp->protocol = htonl(protocol);
755
756         sg_init_table(sg, 4);
757
758         sg_set_buf(&sg[block++], bp, sizeof(*bp));
759         nbytes += sizeof(*bp);
760
761         /* 2. TCP header, excluding options */
762         cksum = th->check;
763         th->check = 0;
764         sg_set_buf(&sg[block++], th, sizeof(*th));
765         nbytes += sizeof(*th);
766
767         /* 3. TCP segment data (if any) */
768         data_len = tcplen - (th->doff << 2);
769         if (data_len > 0) {
770                 u8 *data = (u8 *)th + (th->doff << 2);
771                 sg_set_buf(&sg[block++], data, data_len);
772                 nbytes += data_len;
773         }
774
775         /* 4. shared key */
776         sg_set_buf(&sg[block++], key->key, key->keylen);
777         nbytes += key->keylen;
778
779         sg_mark_end(&sg[block - 1]);
780
781         /* Now store the hash into the packet */
782         err = crypto_hash_init(desc);
783         if (err) {
784                 printk(KERN_WARNING "%s(): hash_init failed\n", __FUNCTION__);
785                 goto clear_hash;
786         }
787         err = crypto_hash_update(desc, sg, nbytes);
788         if (err) {
789                 printk(KERN_WARNING "%s(): hash_update failed\n", __FUNCTION__);
790                 goto clear_hash;
791         }
792         err = crypto_hash_final(desc, md5_hash);
793         if (err) {
794                 printk(KERN_WARNING "%s(): hash_final failed\n", __FUNCTION__);
795                 goto clear_hash;
796         }
797
798         /* Reset header, and free up the crypto */
799         tcp_put_md5sig_pool();
800         th->check = cksum;
801 out:
802         return 0;
803 clear_hash:
804         tcp_put_md5sig_pool();
805 clear_hash_noput:
806         memset(md5_hash, 0, 16);
807         goto out;
808 }
809
810 static int tcp_v6_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
811                                 struct sock *sk,
812                                 struct dst_entry *dst,
813                                 struct request_sock *req,
814                                 struct tcphdr *th, int protocol,
815                                 unsigned int tcplen)
816 {
817         struct in6_addr *saddr, *daddr;
818
819         if (sk) {
820                 saddr = &inet6_sk(sk)->saddr;
821                 daddr = &inet6_sk(sk)->daddr;
822         } else {
823                 saddr = &inet6_rsk(req)->loc_addr;
824                 daddr = &inet6_rsk(req)->rmt_addr;
825         }
826         return tcp_v6_do_calc_md5_hash(md5_hash, key,
827                                        saddr, daddr,
828                                        th, protocol, tcplen);
829 }
830
831 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
832 {
833         __u8 *hash_location = NULL;
834         struct tcp_md5sig_key *hash_expected;
835         struct ipv6hdr *ip6h = ipv6_hdr(skb);
836         struct tcphdr *th = tcp_hdr(skb);
837         int length = (th->doff << 2) - sizeof (*th);
838         int genhash;
839         u8 *ptr;
840         u8 newhash[16];
841
842         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
843
844         /* If the TCP option is too short, we can short cut */
845         if (length < TCPOLEN_MD5SIG)
846                 return hash_expected ? 1 : 0;
847
848         /* parse options */
849         ptr = (u8*)(th + 1);
850         while (length > 0) {
851                 int opcode = *ptr++;
852                 int opsize;
853
854                 switch(opcode) {
855                 case TCPOPT_EOL:
856                         goto done_opts;
857                 case TCPOPT_NOP:
858                         length--;
859                         continue;
860                 default:
861                         opsize = *ptr++;
862                         if (opsize < 2 || opsize > length)
863                                 goto done_opts;
864                         if (opcode == TCPOPT_MD5SIG) {
865                                 hash_location = ptr;
866                                 goto done_opts;
867                         }
868                 }
869                 ptr += opsize - 2;
870                 length -= opsize;
871         }
872
873 done_opts:
874         /* do we have a hash as expected? */
875         if (!hash_expected) {
876                 if (!hash_location)
877                         return 0;
878                 if (net_ratelimit()) {
879                         printk(KERN_INFO "MD5 Hash NOT expected but found "
880                                "(" NIP6_FMT ", %u)->"
881                                "(" NIP6_FMT ", %u)\n",
882                                NIP6(ip6h->saddr), ntohs(th->source),
883                                NIP6(ip6h->daddr), ntohs(th->dest));
884                 }
885                 return 1;
886         }
887
888         if (!hash_location) {
889                 if (net_ratelimit()) {
890                         printk(KERN_INFO "MD5 Hash expected but NOT found "
891                                "(" NIP6_FMT ", %u)->"
892                                "(" NIP6_FMT ", %u)\n",
893                                NIP6(ip6h->saddr), ntohs(th->source),
894                                NIP6(ip6h->daddr), ntohs(th->dest));
895                 }
896                 return 1;
897         }
898
899         /* check the signature */
900         genhash = tcp_v6_do_calc_md5_hash(newhash,
901                                           hash_expected,
902                                           &ip6h->saddr, &ip6h->daddr,
903                                           th, sk->sk_protocol,
904                                           skb->len);
905         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
906                 if (net_ratelimit()) {
907                         printk(KERN_INFO "MD5 Hash %s for "
908                                "(" NIP6_FMT ", %u)->"
909                                "(" NIP6_FMT ", %u)\n",
910                                genhash ? "failed" : "mismatch",
911                                NIP6(ip6h->saddr), ntohs(th->source),
912                                NIP6(ip6h->daddr), ntohs(th->dest));
913                 }
914                 return 1;
915         }
916         return 0;
917 }
918 #endif
919
920 static struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
921         .family         =       AF_INET6,
922         .obj_size       =       sizeof(struct tcp6_request_sock),
923         .rtx_syn_ack    =       tcp_v6_send_synack,
924         .send_ack       =       tcp_v6_reqsk_send_ack,
925         .destructor     =       tcp_v6_reqsk_destructor,
926         .send_reset     =       tcp_v6_send_reset
927 };
928
929 #ifdef CONFIG_TCP_MD5SIG
930 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
931         .md5_lookup     =       tcp_v6_reqsk_md5_lookup,
932 };
933 #endif
934
935 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
936         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
937         .twsk_unique    = tcp_twsk_unique,
938         .twsk_destructor= tcp_twsk_destructor,
939 };
940
941 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
942 {
943         struct ipv6_pinfo *np = inet6_sk(sk);
944         struct tcphdr *th = tcp_hdr(skb);
945
946         if (skb->ip_summed == CHECKSUM_PARTIAL) {
947                 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,  0);
948                 skb->csum_start = skb_transport_header(skb) - skb->head;
949                 skb->csum_offset = offsetof(struct tcphdr, check);
950         } else {
951                 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
952                                             csum_partial((char *)th, th->doff<<2,
953                                                          skb->csum));
954         }
955 }
956
957 static int tcp_v6_gso_send_check(struct sk_buff *skb)
958 {
959         struct ipv6hdr *ipv6h;
960         struct tcphdr *th;
961
962         if (!pskb_may_pull(skb, sizeof(*th)))
963                 return -EINVAL;
964
965         ipv6h = ipv6_hdr(skb);
966         th = tcp_hdr(skb);
967
968         th->check = 0;
969         th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
970                                      IPPROTO_TCP, 0);
971         skb->csum_start = skb_transport_header(skb) - skb->head;
972         skb->csum_offset = offsetof(struct tcphdr, check);
973         skb->ip_summed = CHECKSUM_PARTIAL;
974         return 0;
975 }
976
977 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
978 {
979         struct tcphdr *th = tcp_hdr(skb), *t1;
980         struct sk_buff *buff;
981         struct flowi fl;
982         unsigned int tot_len = sizeof(*th);
983 #ifdef CONFIG_TCP_MD5SIG
984         struct tcp_md5sig_key *key;
985 #endif
986
987         if (th->rst)
988                 return;
989
990         if (!ipv6_unicast_destination(skb))
991                 return;
992
993 #ifdef CONFIG_TCP_MD5SIG
994         if (sk)
995                 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
996         else
997                 key = NULL;
998
999         if (key)
1000                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1001 #endif
1002
1003         /*
1004          * We need to grab some memory, and put together an RST,
1005          * and then put it into the queue to be sent.
1006          */
1007
1008         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1009                          GFP_ATOMIC);
1010         if (buff == NULL)
1011                 return;
1012
1013         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1014
1015         t1 = (struct tcphdr *) skb_push(buff, tot_len);
1016
1017         /* Swap the send and the receive. */
1018         memset(t1, 0, sizeof(*t1));
1019         t1->dest = th->source;
1020         t1->source = th->dest;
1021         t1->doff = tot_len / 4;
1022         t1->rst = 1;
1023
1024         if(th->ack) {
1025                 t1->seq = th->ack_seq;
1026         } else {
1027                 t1->ack = 1;
1028                 t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
1029                                     + skb->len - (th->doff<<2));
1030         }
1031
1032 #ifdef CONFIG_TCP_MD5SIG
1033         if (key) {
1034                 __be32 *opt = (__be32*)(t1 + 1);
1035                 opt[0] = htonl((TCPOPT_NOP << 24) |
1036                                (TCPOPT_NOP << 16) |
1037                                (TCPOPT_MD5SIG << 8) |
1038                                TCPOLEN_MD5SIG);
1039                 tcp_v6_do_calc_md5_hash((__u8 *)&opt[1], key,
1040                                         &ipv6_hdr(skb)->daddr,
1041                                         &ipv6_hdr(skb)->saddr,
1042                                         t1, IPPROTO_TCP, tot_len);
1043         }
1044 #endif
1045
1046         buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1047
1048         memset(&fl, 0, sizeof(fl));
1049         ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1050         ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1051
1052         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1053                                     sizeof(*t1), IPPROTO_TCP,
1054                                     buff->csum);
1055
1056         fl.proto = IPPROTO_TCP;
1057         fl.oif = inet6_iif(skb);
1058         fl.fl_ip_dport = t1->dest;
1059         fl.fl_ip_sport = t1->source;
1060         security_skb_classify_flow(skb, &fl);
1061
1062         /* sk = NULL, but it is safe for now. RST socket required. */
1063         if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1064
1065                 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1066                         ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1067                         TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1068                         TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
1069                         return;
1070                 }
1071         }
1072
1073         kfree_skb(buff);
1074 }
1075
1076 static void tcp_v6_send_ack(struct tcp_timewait_sock *tw,
1077                             struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
1078 {
1079         struct tcphdr *th = tcp_hdr(skb), *t1;
1080         struct sk_buff *buff;
1081         struct flowi fl;
1082         unsigned int tot_len = sizeof(struct tcphdr);
1083         __be32 *topt;
1084 #ifdef CONFIG_TCP_MD5SIG
1085         struct tcp_md5sig_key *key;
1086         struct tcp_md5sig_key tw_key;
1087 #endif
1088
1089 #ifdef CONFIG_TCP_MD5SIG
1090         if (!tw && skb->sk) {
1091                 key = tcp_v6_md5_do_lookup(skb->sk, &ipv6_hdr(skb)->daddr);
1092         } else if (tw && tw->tw_md5_keylen) {
1093                 tw_key.key = tw->tw_md5_key;
1094                 tw_key.keylen = tw->tw_md5_keylen;
1095                 key = &tw_key;
1096         } else {
1097                 key = NULL;
1098         }
1099 #endif
1100
1101         if (ts)
1102                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
1103 #ifdef CONFIG_TCP_MD5SIG
1104         if (key)
1105                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1106 #endif
1107
1108         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1109                          GFP_ATOMIC);
1110         if (buff == NULL)
1111                 return;
1112
1113         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1114
1115         t1 = (struct tcphdr *) skb_push(buff,tot_len);
1116
1117         /* Swap the send and the receive. */
1118         memset(t1, 0, sizeof(*t1));
1119         t1->dest = th->source;
1120         t1->source = th->dest;
1121         t1->doff = tot_len/4;
1122         t1->seq = htonl(seq);
1123         t1->ack_seq = htonl(ack);
1124         t1->ack = 1;
1125         t1->window = htons(win);
1126
1127         topt = (__be32 *)(t1 + 1);
1128
1129         if (ts) {
1130                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1131                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1132                 *topt++ = htonl(tcp_time_stamp);
1133                 *topt = htonl(ts);
1134         }
1135
1136 #ifdef CONFIG_TCP_MD5SIG
1137         if (key) {
1138                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1139                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1140                 tcp_v6_do_calc_md5_hash((__u8 *)topt, key,
1141                                         &ipv6_hdr(skb)->daddr,
1142                                         &ipv6_hdr(skb)->saddr,
1143                                         t1, IPPROTO_TCP, tot_len);
1144         }
1145 #endif
1146
1147         buff->csum = csum_partial((char *)t1, tot_len, 0);
1148
1149         memset(&fl, 0, sizeof(fl));
1150         ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1151         ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1152
1153         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1154                                     tot_len, IPPROTO_TCP,
1155                                     buff->csum);
1156
1157         fl.proto = IPPROTO_TCP;
1158         fl.oif = inet6_iif(skb);
1159         fl.fl_ip_dport = t1->dest;
1160         fl.fl_ip_sport = t1->source;
1161         security_skb_classify_flow(skb, &fl);
1162
1163         if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1164                 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1165                         ip6_xmit(tcp6_socket->sk, buff, &fl, NULL, 0);
1166                         TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1167                         return;
1168                 }
1169         }
1170
1171         kfree_skb(buff);
1172 }
1173
1174 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1175 {
1176         struct inet_timewait_sock *tw = inet_twsk(sk);
1177         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1178
1179         tcp_v6_send_ack(tcptw, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1180                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1181                         tcptw->tw_ts_recent);
1182
1183         inet_twsk_put(tw);
1184 }
1185
1186 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
1187 {
1188         tcp_v6_send_ack(NULL, skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent);
1189 }
1190
1191
1192 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1193 {
1194         struct request_sock *req, **prev;
1195         const struct tcphdr *th = tcp_hdr(skb);
1196         struct sock *nsk;
1197
1198         /* Find possible connection requests. */
1199         req = inet6_csk_search_req(sk, &prev, th->source,
1200                                    &ipv6_hdr(skb)->saddr,
1201                                    &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1202         if (req)
1203                 return tcp_check_req(sk, skb, req, prev);
1204
1205         nsk = __inet6_lookup_established(sk->sk_net, &tcp_hashinfo,
1206                         &ipv6_hdr(skb)->saddr, th->source,
1207                         &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1208
1209         if (nsk) {
1210                 if (nsk->sk_state != TCP_TIME_WAIT) {
1211                         bh_lock_sock(nsk);
1212                         return nsk;
1213                 }
1214                 inet_twsk_put(inet_twsk(nsk));
1215                 return NULL;
1216         }
1217
1218 #if 0 /*def CONFIG_SYN_COOKIES*/
1219         if (!th->rst && !th->syn && th->ack)
1220                 sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
1221 #endif
1222         return sk;
1223 }
1224
1225 /* FIXME: this is substantially similar to the ipv4 code.
1226  * Can some kind of merge be done? -- erics
1227  */
1228 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1229 {
1230         struct inet6_request_sock *treq;
1231         struct ipv6_pinfo *np = inet6_sk(sk);
1232         struct tcp_options_received tmp_opt;
1233         struct tcp_sock *tp = tcp_sk(sk);
1234         struct request_sock *req = NULL;
1235         __u32 isn = TCP_SKB_CB(skb)->when;
1236
1237         if (skb->protocol == htons(ETH_P_IP))
1238                 return tcp_v4_conn_request(sk, skb);
1239
1240         if (!ipv6_unicast_destination(skb))
1241                 goto drop;
1242
1243         /*
1244          *      There are no SYN attacks on IPv6, yet...
1245          */
1246         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1247                 if (net_ratelimit())
1248                         printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
1249                 goto drop;
1250         }
1251
1252         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1253                 goto drop;
1254
1255         req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1256         if (req == NULL)
1257                 goto drop;
1258
1259 #ifdef CONFIG_TCP_MD5SIG
1260         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1261 #endif
1262
1263         tcp_clear_options(&tmp_opt);
1264         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1265         tmp_opt.user_mss = tp->rx_opt.user_mss;
1266
1267         tcp_parse_options(skb, &tmp_opt, 0);
1268
1269         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1270         tcp_openreq_init(req, &tmp_opt, skb);
1271
1272         treq = inet6_rsk(req);
1273         ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1274         ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1275         TCP_ECN_create_request(req, tcp_hdr(skb));
1276         treq->pktopts = NULL;
1277         if (ipv6_opt_accepted(sk, skb) ||
1278             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1279             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1280                 atomic_inc(&skb->users);
1281                 treq->pktopts = skb;
1282         }
1283         treq->iif = sk->sk_bound_dev_if;
1284
1285         /* So that link locals have meaning */
1286         if (!sk->sk_bound_dev_if &&
1287             ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1288                 treq->iif = inet6_iif(skb);
1289
1290         if (isn == 0)
1291                 isn = tcp_v6_init_sequence(skb);
1292
1293         tcp_rsk(req)->snt_isn = isn;
1294
1295         security_inet_conn_request(sk, skb, req);
1296
1297         if (tcp_v6_send_synack(sk, req, NULL))
1298                 goto drop;
1299
1300         inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1301         return 0;
1302
1303 drop:
1304         if (req)
1305                 reqsk_free(req);
1306
1307         return 0; /* don't send reset */
1308 }
1309
1310 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1311                                           struct request_sock *req,
1312                                           struct dst_entry *dst)
1313 {
1314         struct inet6_request_sock *treq = inet6_rsk(req);
1315         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1316         struct tcp6_sock *newtcp6sk;
1317         struct inet_sock *newinet;
1318         struct tcp_sock *newtp;
1319         struct sock *newsk;
1320         struct ipv6_txoptions *opt;
1321 #ifdef CONFIG_TCP_MD5SIG
1322         struct tcp_md5sig_key *key;
1323 #endif
1324
1325         if (skb->protocol == htons(ETH_P_IP)) {
1326                 /*
1327                  *      v6 mapped
1328                  */
1329
1330                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1331
1332                 if (newsk == NULL)
1333                         return NULL;
1334
1335                 newtcp6sk = (struct tcp6_sock *)newsk;
1336                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1337
1338                 newinet = inet_sk(newsk);
1339                 newnp = inet6_sk(newsk);
1340                 newtp = tcp_sk(newsk);
1341
1342                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1343
1344                 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1345                               newinet->daddr);
1346
1347                 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1348                               newinet->saddr);
1349
1350                 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1351
1352                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1353                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1354 #ifdef CONFIG_TCP_MD5SIG
1355                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1356 #endif
1357
1358                 newnp->pktoptions  = NULL;
1359                 newnp->opt         = NULL;
1360                 newnp->mcast_oif   = inet6_iif(skb);
1361                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1362
1363                 /*
1364                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1365                  * here, tcp_create_openreq_child now does this for us, see the comment in
1366                  * that function for the gory details. -acme
1367                  */
1368
1369                 /* It is tricky place. Until this moment IPv4 tcp
1370                    worked with IPv6 icsk.icsk_af_ops.
1371                    Sync it now.
1372                  */
1373                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1374
1375                 return newsk;
1376         }
1377
1378         opt = np->opt;
1379
1380         if (sk_acceptq_is_full(sk))
1381                 goto out_overflow;
1382
1383         if (dst == NULL) {
1384                 struct in6_addr *final_p = NULL, final;
1385                 struct flowi fl;
1386
1387                 memset(&fl, 0, sizeof(fl));
1388                 fl.proto = IPPROTO_TCP;
1389                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1390                 if (opt && opt->srcrt) {
1391                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1392                         ipv6_addr_copy(&final, &fl.fl6_dst);
1393                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1394                         final_p = &final;
1395                 }
1396                 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1397                 fl.oif = sk->sk_bound_dev_if;
1398                 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1399                 fl.fl_ip_sport = inet_sk(sk)->sport;
1400                 security_req_classify_flow(req, &fl);
1401
1402                 if (ip6_dst_lookup(sk, &dst, &fl))
1403                         goto out;
1404
1405                 if (final_p)
1406                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1407
1408                 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1409                         goto out;
1410         }
1411
1412         newsk = tcp_create_openreq_child(sk, req, skb);
1413         if (newsk == NULL)
1414                 goto out;
1415
1416         /*
1417          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1418          * count here, tcp_create_openreq_child now does this for us, see the
1419          * comment in that function for the gory details. -acme
1420          */
1421
1422         newsk->sk_gso_type = SKB_GSO_TCPV6;
1423         __ip6_dst_store(newsk, dst, NULL, NULL);
1424
1425         newtcp6sk = (struct tcp6_sock *)newsk;
1426         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1427
1428         newtp = tcp_sk(newsk);
1429         newinet = inet_sk(newsk);
1430         newnp = inet6_sk(newsk);
1431
1432         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1433
1434         ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1435         ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1436         ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1437         newsk->sk_bound_dev_if = treq->iif;
1438
1439         /* Now IPv6 options...
1440
1441            First: no IPv4 options.
1442          */
1443         newinet->opt = NULL;
1444         newnp->ipv6_fl_list = NULL;
1445
1446         /* Clone RX bits */
1447         newnp->rxopt.all = np->rxopt.all;
1448
1449         /* Clone pktoptions received with SYN */
1450         newnp->pktoptions = NULL;
1451         if (treq->pktopts != NULL) {
1452                 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1453                 kfree_skb(treq->pktopts);
1454                 treq->pktopts = NULL;
1455                 if (newnp->pktoptions)
1456                         skb_set_owner_r(newnp->pktoptions, newsk);
1457         }
1458         newnp->opt        = NULL;
1459         newnp->mcast_oif  = inet6_iif(skb);
1460         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1461
1462         /* Clone native IPv6 options from listening socket (if any)
1463
1464            Yes, keeping reference count would be much more clever,
1465            but we make one more one thing there: reattach optmem
1466            to newsk.
1467          */
1468         if (opt) {
1469                 newnp->opt = ipv6_dup_options(newsk, opt);
1470                 if (opt != np->opt)
1471                         sock_kfree_s(sk, opt, opt->tot_len);
1472         }
1473
1474         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1475         if (newnp->opt)
1476                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1477                                                      newnp->opt->opt_flen);
1478
1479         tcp_mtup_init(newsk);
1480         tcp_sync_mss(newsk, dst_mtu(dst));
1481         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1482         tcp_initialize_rcv_mss(newsk);
1483
1484         newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1485
1486 #ifdef CONFIG_TCP_MD5SIG
1487         /* Copy over the MD5 key from the original socket */
1488         if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1489                 /* We're using one, so create a matching key
1490                  * on the newsk structure. If we fail to get
1491                  * memory, then we end up not copying the key
1492                  * across. Shucks.
1493                  */
1494                 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1495                 if (newkey != NULL)
1496                         tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1497                                           newkey, key->keylen);
1498         }
1499 #endif
1500
1501         __inet6_hash(newsk);
1502         inet_inherit_port(sk, newsk);
1503
1504         return newsk;
1505
1506 out_overflow:
1507         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1508 out:
1509         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1510         if (opt && opt != np->opt)
1511                 sock_kfree_s(sk, opt, opt->tot_len);
1512         dst_release(dst);
1513         return NULL;
1514 }
1515
1516 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1517 {
1518         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1519                 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1520                                   &ipv6_hdr(skb)->daddr, skb->csum)) {
1521                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1522                         return 0;
1523                 }
1524         }
1525
1526         skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1527                                               &ipv6_hdr(skb)->saddr,
1528                                               &ipv6_hdr(skb)->daddr, 0));
1529
1530         if (skb->len <= 76) {
1531                 return __skb_checksum_complete(skb);
1532         }
1533         return 0;
1534 }
1535
1536 /* The socket must have it's spinlock held when we get
1537  * here.
1538  *
1539  * We have a potential double-lock case here, so even when
1540  * doing backlog processing we use the BH locking scheme.
1541  * This is because we cannot sleep with the original spinlock
1542  * held.
1543  */
1544 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1545 {
1546         struct ipv6_pinfo *np = inet6_sk(sk);
1547         struct tcp_sock *tp;
1548         struct sk_buff *opt_skb = NULL;
1549
1550         /* Imagine: socket is IPv6. IPv4 packet arrives,
1551            goes to IPv4 receive handler and backlogged.
1552            From backlog it always goes here. Kerboom...
1553            Fortunately, tcp_rcv_established and rcv_established
1554            handle them correctly, but it is not case with
1555            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1556          */
1557
1558         if (skb->protocol == htons(ETH_P_IP))
1559                 return tcp_v4_do_rcv(sk, skb);
1560
1561 #ifdef CONFIG_TCP_MD5SIG
1562         if (tcp_v6_inbound_md5_hash (sk, skb))
1563                 goto discard;
1564 #endif
1565
1566         if (sk_filter(sk, skb))
1567                 goto discard;
1568
1569         /*
1570          *      socket locking is here for SMP purposes as backlog rcv
1571          *      is currently called with bh processing disabled.
1572          */
1573
1574         /* Do Stevens' IPV6_PKTOPTIONS.
1575
1576            Yes, guys, it is the only place in our code, where we
1577            may make it not affecting IPv4.
1578            The rest of code is protocol independent,
1579            and I do not like idea to uglify IPv4.
1580
1581            Actually, all the idea behind IPV6_PKTOPTIONS
1582            looks not very well thought. For now we latch
1583            options, received in the last packet, enqueued
1584            by tcp. Feel free to propose better solution.
1585                                                --ANK (980728)
1586          */
1587         if (np->rxopt.all)
1588                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1589
1590         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1591                 TCP_CHECK_TIMER(sk);
1592                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1593                         goto reset;
1594                 TCP_CHECK_TIMER(sk);
1595                 if (opt_skb)
1596                         goto ipv6_pktoptions;
1597                 return 0;
1598         }
1599
1600         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1601                 goto csum_err;
1602
1603         if (sk->sk_state == TCP_LISTEN) {
1604                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1605                 if (!nsk)
1606                         goto discard;
1607
1608                 /*
1609                  * Queue it on the new socket if the new socket is active,
1610                  * otherwise we just shortcircuit this and continue with
1611                  * the new socket..
1612                  */
1613                 if(nsk != sk) {
1614                         if (tcp_child_process(sk, nsk, skb))
1615                                 goto reset;
1616                         if (opt_skb)
1617                                 __kfree_skb(opt_skb);
1618                         return 0;
1619                 }
1620         }
1621
1622         TCP_CHECK_TIMER(sk);
1623         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1624                 goto reset;
1625         TCP_CHECK_TIMER(sk);
1626         if (opt_skb)
1627                 goto ipv6_pktoptions;
1628         return 0;
1629
1630 reset:
1631         tcp_v6_send_reset(sk, skb);
1632 discard:
1633         if (opt_skb)
1634                 __kfree_skb(opt_skb);
1635         kfree_skb(skb);
1636         return 0;
1637 csum_err:
1638         TCP_INC_STATS_BH(TCP_MIB_INERRS);
1639         goto discard;
1640
1641
1642 ipv6_pktoptions:
1643         /* Do you ask, what is it?
1644
1645            1. skb was enqueued by tcp.
1646            2. skb is added to tail of read queue, rather than out of order.
1647            3. socket is not in passive state.
1648            4. Finally, it really contains options, which user wants to receive.
1649          */
1650         tp = tcp_sk(sk);
1651         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1652             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1653                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1654                         np->mcast_oif = inet6_iif(opt_skb);
1655                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1656                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1657                 if (ipv6_opt_accepted(sk, opt_skb)) {
1658                         skb_set_owner_r(opt_skb, sk);
1659                         opt_skb = xchg(&np->pktoptions, opt_skb);
1660                 } else {
1661                         __kfree_skb(opt_skb);
1662                         opt_skb = xchg(&np->pktoptions, NULL);
1663                 }
1664         }
1665
1666         if (opt_skb)
1667                 kfree_skb(opt_skb);
1668         return 0;
1669 }
1670
1671 static int tcp_v6_rcv(struct sk_buff *skb)
1672 {
1673         struct tcphdr *th;
1674         struct sock *sk;
1675         int ret;
1676
1677         if (skb->pkt_type != PACKET_HOST)
1678                 goto discard_it;
1679
1680         /*
1681          *      Count it even if it's bad.
1682          */
1683         TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1684
1685         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1686                 goto discard_it;
1687
1688         th = tcp_hdr(skb);
1689
1690         if (th->doff < sizeof(struct tcphdr)/4)
1691                 goto bad_packet;
1692         if (!pskb_may_pull(skb, th->doff*4))
1693                 goto discard_it;
1694
1695         if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1696                 goto bad_packet;
1697
1698         th = tcp_hdr(skb);
1699         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1700         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1701                                     skb->len - th->doff*4);
1702         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1703         TCP_SKB_CB(skb)->when = 0;
1704         TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1705         TCP_SKB_CB(skb)->sacked = 0;
1706
1707         sk = __inet6_lookup(skb->dev->nd_net, &tcp_hashinfo,
1708                         &ipv6_hdr(skb)->saddr, th->source,
1709                         &ipv6_hdr(skb)->daddr, ntohs(th->dest),
1710                         inet6_iif(skb));
1711
1712         if (!sk)
1713                 goto no_tcp_socket;
1714
1715 process:
1716         if (sk->sk_state == TCP_TIME_WAIT)
1717                 goto do_time_wait;
1718
1719         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1720                 goto discard_and_relse;
1721
1722         if (sk_filter(sk, skb))
1723                 goto discard_and_relse;
1724
1725         skb->dev = NULL;
1726
1727         bh_lock_sock_nested(sk);
1728         ret = 0;
1729         if (!sock_owned_by_user(sk)) {
1730 #ifdef CONFIG_NET_DMA
1731                 struct tcp_sock *tp = tcp_sk(sk);
1732                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1733                         tp->ucopy.dma_chan = get_softnet_dma();
1734                 if (tp->ucopy.dma_chan)
1735                         ret = tcp_v6_do_rcv(sk, skb);
1736                 else
1737 #endif
1738                 {
1739                         if (!tcp_prequeue(sk, skb))
1740                                 ret = tcp_v6_do_rcv(sk, skb);
1741                 }
1742         } else
1743                 sk_add_backlog(sk, skb);
1744         bh_unlock_sock(sk);
1745
1746         sock_put(sk);
1747         return ret ? -1 : 0;
1748
1749 no_tcp_socket:
1750         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1751                 goto discard_it;
1752
1753         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1754 bad_packet:
1755                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1756         } else {
1757                 tcp_v6_send_reset(NULL, skb);
1758         }
1759
1760 discard_it:
1761
1762         /*
1763          *      Discard frame
1764          */
1765
1766         kfree_skb(skb);
1767         return 0;
1768
1769 discard_and_relse:
1770         sock_put(sk);
1771         goto discard_it;
1772
1773 do_time_wait:
1774         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1775                 inet_twsk_put(inet_twsk(sk));
1776                 goto discard_it;
1777         }
1778
1779         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1780                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1781                 inet_twsk_put(inet_twsk(sk));
1782                 goto discard_it;
1783         }
1784
1785         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1786         case TCP_TW_SYN:
1787         {
1788                 struct sock *sk2;
1789
1790                 sk2 = inet6_lookup_listener(skb->dev->nd_net, &tcp_hashinfo,
1791                                             &ipv6_hdr(skb)->daddr,
1792                                             ntohs(th->dest), inet6_iif(skb));
1793                 if (sk2 != NULL) {
1794                         struct inet_timewait_sock *tw = inet_twsk(sk);
1795                         inet_twsk_deschedule(tw, &tcp_death_row);
1796                         inet_twsk_put(tw);
1797                         sk = sk2;
1798                         goto process;
1799                 }
1800                 /* Fall through to ACK */
1801         }
1802         case TCP_TW_ACK:
1803                 tcp_v6_timewait_ack(sk, skb);
1804                 break;
1805         case TCP_TW_RST:
1806                 goto no_tcp_socket;
1807         case TCP_TW_SUCCESS:;
1808         }
1809         goto discard_it;
1810 }
1811
1812 static int tcp_v6_remember_stamp(struct sock *sk)
1813 {
1814         /* Alas, not yet... */
1815         return 0;
1816 }
1817
1818 static struct inet_connection_sock_af_ops ipv6_specific = {
1819         .queue_xmit        = inet6_csk_xmit,
1820         .send_check        = tcp_v6_send_check,
1821         .rebuild_header    = inet6_sk_rebuild_header,
1822         .conn_request      = tcp_v6_conn_request,
1823         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1824         .remember_stamp    = tcp_v6_remember_stamp,
1825         .net_header_len    = sizeof(struct ipv6hdr),
1826         .setsockopt        = ipv6_setsockopt,
1827         .getsockopt        = ipv6_getsockopt,
1828         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1829         .sockaddr_len      = sizeof(struct sockaddr_in6),
1830         .bind_conflict     = inet6_csk_bind_conflict,
1831 #ifdef CONFIG_COMPAT
1832         .compat_setsockopt = compat_ipv6_setsockopt,
1833         .compat_getsockopt = compat_ipv6_getsockopt,
1834 #endif
1835 };
1836
1837 #ifdef CONFIG_TCP_MD5SIG
1838 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1839         .md5_lookup     =       tcp_v6_md5_lookup,
1840         .calc_md5_hash  =       tcp_v6_calc_md5_hash,
1841         .md5_add        =       tcp_v6_md5_add_func,
1842         .md5_parse      =       tcp_v6_parse_md5_keys,
1843 };
1844 #endif
1845
1846 /*
1847  *      TCP over IPv4 via INET6 API
1848  */
1849
1850 static struct inet_connection_sock_af_ops ipv6_mapped = {
1851         .queue_xmit        = ip_queue_xmit,
1852         .send_check        = tcp_v4_send_check,
1853         .rebuild_header    = inet_sk_rebuild_header,
1854         .conn_request      = tcp_v6_conn_request,
1855         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1856         .remember_stamp    = tcp_v4_remember_stamp,
1857         .net_header_len    = sizeof(struct iphdr),
1858         .setsockopt        = ipv6_setsockopt,
1859         .getsockopt        = ipv6_getsockopt,
1860         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1861         .sockaddr_len      = sizeof(struct sockaddr_in6),
1862         .bind_conflict     = inet6_csk_bind_conflict,
1863 #ifdef CONFIG_COMPAT
1864         .compat_setsockopt = compat_ipv6_setsockopt,
1865         .compat_getsockopt = compat_ipv6_getsockopt,
1866 #endif
1867 };
1868
1869 #ifdef CONFIG_TCP_MD5SIG
1870 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1871         .md5_lookup     =       tcp_v4_md5_lookup,
1872         .calc_md5_hash  =       tcp_v4_calc_md5_hash,
1873         .md5_add        =       tcp_v6_md5_add_func,
1874         .md5_parse      =       tcp_v6_parse_md5_keys,
1875 };
1876 #endif
1877
1878 /* NOTE: A lot of things set to zero explicitly by call to
1879  *       sk_alloc() so need not be done here.
1880  */
1881 static int tcp_v6_init_sock(struct sock *sk)
1882 {
1883         struct inet_connection_sock *icsk = inet_csk(sk);
1884         struct tcp_sock *tp = tcp_sk(sk);
1885
1886         skb_queue_head_init(&tp->out_of_order_queue);
1887         tcp_init_xmit_timers(sk);
1888         tcp_prequeue_init(tp);
1889
1890         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1891         tp->mdev = TCP_TIMEOUT_INIT;
1892
1893         /* So many TCP implementations out there (incorrectly) count the
1894          * initial SYN frame in their delayed-ACK and congestion control
1895          * algorithms that we must have the following bandaid to talk
1896          * efficiently to them.  -DaveM
1897          */
1898         tp->snd_cwnd = 2;
1899
1900         /* See draft-stevens-tcpca-spec-01 for discussion of the
1901          * initialization of these values.
1902          */
1903         tp->snd_ssthresh = 0x7fffffff;
1904         tp->snd_cwnd_clamp = ~0;
1905         tp->mss_cache = 536;
1906
1907         tp->reordering = sysctl_tcp_reordering;
1908
1909         sk->sk_state = TCP_CLOSE;
1910
1911         icsk->icsk_af_ops = &ipv6_specific;
1912         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1913         icsk->icsk_sync_mss = tcp_sync_mss;
1914         sk->sk_write_space = sk_stream_write_space;
1915         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1916
1917 #ifdef CONFIG_TCP_MD5SIG
1918         tp->af_specific = &tcp_sock_ipv6_specific;
1919 #endif
1920
1921         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1922         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1923
1924         atomic_inc(&tcp_sockets_allocated);
1925
1926         return 0;
1927 }
1928
1929 static int tcp_v6_destroy_sock(struct sock *sk)
1930 {
1931 #ifdef CONFIG_TCP_MD5SIG
1932         /* Clean up the MD5 key list */
1933         if (tcp_sk(sk)->md5sig_info)
1934                 tcp_v6_clear_md5_list(sk);
1935 #endif
1936         tcp_v4_destroy_sock(sk);
1937         return inet6_destroy_sock(sk);
1938 }
1939
1940 #ifdef CONFIG_PROC_FS
1941 /* Proc filesystem TCPv6 sock list dumping. */
1942 static void get_openreq6(struct seq_file *seq,
1943                          struct sock *sk, struct request_sock *req, int i, int uid)
1944 {
1945         int ttd = req->expires - jiffies;
1946         struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1947         struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1948
1949         if (ttd < 0)
1950                 ttd = 0;
1951
1952         seq_printf(seq,
1953                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1954                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1955                    i,
1956                    src->s6_addr32[0], src->s6_addr32[1],
1957                    src->s6_addr32[2], src->s6_addr32[3],
1958                    ntohs(inet_sk(sk)->sport),
1959                    dest->s6_addr32[0], dest->s6_addr32[1],
1960                    dest->s6_addr32[2], dest->s6_addr32[3],
1961                    ntohs(inet_rsk(req)->rmt_port),
1962                    TCP_SYN_RECV,
1963                    0,0, /* could print option size, but that is af dependent. */
1964                    1,   /* timers active (only the expire timer) */
1965                    jiffies_to_clock_t(ttd),
1966                    req->retrans,
1967                    uid,
1968                    0,  /* non standard timer */
1969                    0, /* open_requests have no inode */
1970                    0, req);
1971 }
1972
1973 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1974 {
1975         struct in6_addr *dest, *src;
1976         __u16 destp, srcp;
1977         int timer_active;
1978         unsigned long timer_expires;
1979         struct inet_sock *inet = inet_sk(sp);
1980         struct tcp_sock *tp = tcp_sk(sp);
1981         const struct inet_connection_sock *icsk = inet_csk(sp);
1982         struct ipv6_pinfo *np = inet6_sk(sp);
1983
1984         dest  = &np->daddr;
1985         src   = &np->rcv_saddr;
1986         destp = ntohs(inet->dport);
1987         srcp  = ntohs(inet->sport);
1988
1989         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1990                 timer_active    = 1;
1991                 timer_expires   = icsk->icsk_timeout;
1992         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1993                 timer_active    = 4;
1994                 timer_expires   = icsk->icsk_timeout;
1995         } else if (timer_pending(&sp->sk_timer)) {
1996                 timer_active    = 2;
1997                 timer_expires   = sp->sk_timer.expires;
1998         } else {
1999                 timer_active    = 0;
2000                 timer_expires = jiffies;
2001         }
2002
2003         seq_printf(seq,
2004                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2005                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
2006                    i,
2007                    src->s6_addr32[0], src->s6_addr32[1],
2008                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2009                    dest->s6_addr32[0], dest->s6_addr32[1],
2010                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2011                    sp->sk_state,
2012                    tp->write_seq-tp->snd_una,
2013                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2014                    timer_active,
2015                    jiffies_to_clock_t(timer_expires - jiffies),
2016                    icsk->icsk_retransmits,
2017                    sock_i_uid(sp),
2018                    icsk->icsk_probes_out,
2019                    sock_i_ino(sp),
2020                    atomic_read(&sp->sk_refcnt), sp,
2021                    icsk->icsk_rto,
2022                    icsk->icsk_ack.ato,
2023                    (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2024                    tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2025                    );
2026 }
2027
2028 static void get_timewait6_sock(struct seq_file *seq,
2029                                struct inet_timewait_sock *tw, int i)
2030 {
2031         struct in6_addr *dest, *src;
2032         __u16 destp, srcp;
2033         struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2034         int ttd = tw->tw_ttd - jiffies;
2035
2036         if (ttd < 0)
2037                 ttd = 0;
2038
2039         dest = &tw6->tw_v6_daddr;
2040         src  = &tw6->tw_v6_rcv_saddr;
2041         destp = ntohs(tw->tw_dport);
2042         srcp  = ntohs(tw->tw_sport);
2043
2044         seq_printf(seq,
2045                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2046                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2047                    i,
2048                    src->s6_addr32[0], src->s6_addr32[1],
2049                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2050                    dest->s6_addr32[0], dest->s6_addr32[1],
2051                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2052                    tw->tw_substate, 0, 0,
2053                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2054                    atomic_read(&tw->tw_refcnt), tw);
2055 }
2056
2057 static int tcp6_seq_show(struct seq_file *seq, void *v)
2058 {
2059         struct tcp_iter_state *st;
2060
2061         if (v == SEQ_START_TOKEN) {
2062                 seq_puts(seq,
2063                          "  sl  "
2064                          "local_address                         "
2065                          "remote_address                        "
2066                          "st tx_queue rx_queue tr tm->when retrnsmt"
2067                          "   uid  timeout inode\n");
2068                 goto out;
2069         }
2070         st = seq->private;
2071
2072         switch (st->state) {
2073         case TCP_SEQ_STATE_LISTENING:
2074         case TCP_SEQ_STATE_ESTABLISHED:
2075                 get_tcp6_sock(seq, v, st->num);
2076                 break;
2077         case TCP_SEQ_STATE_OPENREQ:
2078                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2079                 break;
2080         case TCP_SEQ_STATE_TIME_WAIT:
2081                 get_timewait6_sock(seq, v, st->num);
2082                 break;
2083         }
2084 out:
2085         return 0;
2086 }
2087
2088 static struct file_operations tcp6_seq_fops;
2089 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2090         .owner          = THIS_MODULE,
2091         .name           = "tcp6",
2092         .family         = AF_INET6,
2093         .seq_show       = tcp6_seq_show,
2094         .seq_fops       = &tcp6_seq_fops,
2095 };
2096
2097 int __init tcp6_proc_init(void)
2098 {
2099         return tcp_proc_register(&tcp6_seq_afinfo);
2100 }
2101
2102 void tcp6_proc_exit(void)
2103 {
2104         tcp_proc_unregister(&tcp6_seq_afinfo);
2105 }
2106 #endif
2107
2108 DEFINE_PROTO_INUSE(tcpv6)
2109
2110 struct proto tcpv6_prot = {
2111         .name                   = "TCPv6",
2112         .owner                  = THIS_MODULE,
2113         .close                  = tcp_close,
2114         .connect                = tcp_v6_connect,
2115         .disconnect             = tcp_disconnect,
2116         .accept                 = inet_csk_accept,
2117         .ioctl                  = tcp_ioctl,
2118         .init                   = tcp_v6_init_sock,
2119         .destroy                = tcp_v6_destroy_sock,
2120         .shutdown               = tcp_shutdown,
2121         .setsockopt             = tcp_setsockopt,
2122         .getsockopt             = tcp_getsockopt,
2123         .recvmsg                = tcp_recvmsg,
2124         .backlog_rcv            = tcp_v6_do_rcv,
2125         .hash                   = tcp_v6_hash,
2126         .unhash                 = inet_unhash,
2127         .get_port               = inet_csk_get_port,
2128         .enter_memory_pressure  = tcp_enter_memory_pressure,
2129         .sockets_allocated      = &tcp_sockets_allocated,
2130         .memory_allocated       = &tcp_memory_allocated,
2131         .memory_pressure        = &tcp_memory_pressure,
2132         .orphan_count           = &tcp_orphan_count,
2133         .sysctl_mem             = sysctl_tcp_mem,
2134         .sysctl_wmem            = sysctl_tcp_wmem,
2135         .sysctl_rmem            = sysctl_tcp_rmem,
2136         .max_header             = MAX_TCP_HEADER,
2137         .obj_size               = sizeof(struct tcp6_sock),
2138         .twsk_prot              = &tcp6_timewait_sock_ops,
2139         .rsk_prot               = &tcp6_request_sock_ops,
2140         .hashinfo               = &tcp_hashinfo,
2141 #ifdef CONFIG_COMPAT
2142         .compat_setsockopt      = compat_tcp_setsockopt,
2143         .compat_getsockopt      = compat_tcp_getsockopt,
2144 #endif
2145         REF_PROTO_INUSE(tcpv6)
2146 };
2147
2148 static struct inet6_protocol tcpv6_protocol = {
2149         .handler        =       tcp_v6_rcv,
2150         .err_handler    =       tcp_v6_err,
2151         .gso_send_check =       tcp_v6_gso_send_check,
2152         .gso_segment    =       tcp_tso_segment,
2153         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2154 };
2155
2156 static struct inet_protosw tcpv6_protosw = {
2157         .type           =       SOCK_STREAM,
2158         .protocol       =       IPPROTO_TCP,
2159         .prot           =       &tcpv6_prot,
2160         .ops            =       &inet6_stream_ops,
2161         .capability     =       -1,
2162         .no_check       =       0,
2163         .flags          =       INET_PROTOSW_PERMANENT |
2164                                 INET_PROTOSW_ICSK,
2165 };
2166
2167 int __init tcpv6_init(void)
2168 {
2169         int ret;
2170
2171         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2172         if (ret)
2173                 goto out;
2174
2175         /* register inet6 protocol */
2176         ret = inet6_register_protosw(&tcpv6_protosw);
2177         if (ret)
2178                 goto out_tcpv6_protocol;
2179
2180         ret = inet_csk_ctl_sock_create(&tcp6_socket, PF_INET6,
2181                                        SOCK_RAW, IPPROTO_TCP);
2182         if (ret)
2183                 goto out_tcpv6_protosw;
2184 out:
2185         return ret;
2186
2187 out_tcpv6_protocol:
2188         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2189 out_tcpv6_protosw:
2190         inet6_unregister_protosw(&tcpv6_protosw);
2191         goto out;
2192 }
2193
2194 void tcpv6_exit(void)
2195 {
2196         sock_release(tcp6_socket);
2197         inet6_unregister_protosw(&tcpv6_protosw);
2198         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2199 }