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/config.h>
  30 #include <linux/errno.h>
  31 #include <linux/types.h>
  32 #include <linux/socket.h>
  33 #include <linux/sockios.h>
  34 #include <linux/net.h>
  35 #include <linux/jiffies.h>
  36 #include <linux/in.h>
  37 #include <linux/in6.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/init.h>
  40 #include <linux/jhash.h>
  41 #include <linux/ipsec.h>
  42 #include <linux/times.h>
  43
  44 #include <linux/ipv6.h>
  45 #include <linux/icmpv6.h>
  46 #include <linux/random.h>
  47
  48 #include <net/tcp.h>
  49 #include <net/ndisc.h>
  50 #include <net/ipv6.h>
  51 #include <net/transp_v6.h>
  52 #include <net/addrconf.h>
  53 #include <net/ip6_route.h>
  54 #include <net/ip6_checksum.h>
  55 #include <net/inet_ecn.h>
  56 #include <net/protocol.h>
  57 #include <net/xfrm.h>
  58 #include <net/addrconf.h>
  59 #include <net/snmp.h>
  60 #include <net/dsfield.h>
  61
  62 #include <asm/uaccess.h>
  63
  64 #include <linux/proc_fs.h>
  65 #include <linux/seq_file.h>
  66
  67 static void     tcp_v6_send_reset(struct sk_buff *skb);
  68 static void     tcp_v6_or_send_ack(struct sk_buff *skb, struct open_request *req);
  69 static void     tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len,
  70                                   struct sk_buff *skb);
  71
  72 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  73 static int      tcp_v6_xmit(struct sk_buff *skb, int ipfragok);
  74
  75 static struct tcp_func ipv6_mapped;
  76 static struct tcp_func ipv6_specific;
  77
  78 /* I have no idea if this is a good hash for v6 or not. -DaveM */
  79 static __inline__ int tcp_v6_hashfn(struct in6_addr *laddr, u16 lport,
  80                                     struct in6_addr *faddr, u16 fport)
  81 {
  82         int hashent = (lport ^ fport);
  83
  84         hashent ^= (laddr->s6_addr32[3] ^ faddr->s6_addr32[3]);
  85         hashent ^= hashent>>16;
  86         hashent ^= hashent>>8;
  87         return (hashent & (tcp_ehash_size - 1));
  88 }
  89
  90 static __inline__ int tcp_v6_sk_hashfn(struct sock *sk)
  91 {
  92         struct inet_sock *inet = inet_sk(sk);
  93         struct ipv6_pinfo *np = inet6_sk(sk);
  94         struct in6_addr *laddr = &np->rcv_saddr;
  95         struct in6_addr *faddr = &np->daddr;
  96         __u16 lport = inet->num;
  97         __u16 fport = inet->dport;
  98         return tcp_v6_hashfn(laddr, lport, faddr, fport);
  99 }
 100
 101 static inline int tcp_v6_bind_conflict(struct sock *sk,
 102                                        struct tcp_bind_bucket *tb)
 103 {
 104         struct sock *sk2;
 105         struct hlist_node *node;
 106
 107         /* We must walk the whole port owner list in this case. -DaveM */
 108         sk_for_each_bound(sk2, node, &tb->owners) {
 109                 if (sk != sk2 &&
 110                     (!sk->sk_bound_dev_if ||
 111                      !sk2->sk_bound_dev_if ||
 112                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if) &&
 113                     (!sk->sk_reuse || !sk2->sk_reuse ||
 114                      sk2->sk_state == TCP_LISTEN) &&
 115                      ipv6_rcv_saddr_equal(sk, sk2))
 116                         break;
 117         }
 118
 119         return node != NULL;
 120 }
 121
 122 /* Grrr, addr_type already calculated by caller, but I don't want
 123  * to add some silly "cookie" argument to this method just for that.
 124  * But it doesn't matter, the recalculation is in the rarest path
 125  * this function ever takes.
 126  */
 127 static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
 128 {
 129         struct tcp_bind_hashbucket *head;
 130         struct tcp_bind_bucket *tb;
 131         struct hlist_node *node;
 132         int ret;
 133
 134         local_bh_disable();
 135         if (snum == 0) {
 136                 int low = sysctl_local_port_range[0];
 137                 int high = sysctl_local_port_range[1];
 138                 int remaining = (high - low) + 1;
 139                 int rover;
 140
 141                 spin_lock(&tcp_portalloc_lock);
 142                 if (tcp_port_rover < low)
 143                         rover = low;
 144                 else
 145                         rover = tcp_port_rover;
 146                 do {    rover++;
 147                         if (rover > high)
 148                                 rover = low;
 149                         head = &tcp_bhash[tcp_bhashfn(rover)];
 150                         spin_lock(&head->lock);
 151                         tb_for_each(tb, node, &head->chain)
 152                                 if (tb->port == rover)
 153                                         goto next;
 154                         break;
 155                 next:
 156                         spin_unlock(&head->lock);
 157                 } while (--remaining > 0);
 158                 tcp_port_rover = rover;
 159                 spin_unlock(&tcp_portalloc_lock);
 160
 161                 /* Exhausted local port range during search? */
 162                 ret = 1;
 163                 if (remaining <= 0)
 164                         goto fail;
 165
 166                 /* OK, here is the one we will use. */
 167                 snum = rover;
 168         } else {
 169                 head = &tcp_bhash[tcp_bhashfn(snum)];
 170                 spin_lock(&head->lock);
 171                 tb_for_each(tb, node, &head->chain)
 172                         if (tb->port == snum)
 173                                 goto tb_found;
 174         }
 175         tb = NULL;
 176         goto tb_not_found;
 177 tb_found:
 178         if (tb && !hlist_empty(&tb->owners)) {
 179                 if (tb->fastreuse > 0 && sk->sk_reuse &&
 180                     sk->sk_state != TCP_LISTEN) {
 181                         goto success;
 182                 } else {
 183                         ret = 1;
 184                         if (tcp_v6_bind_conflict(sk, tb))
 185                                 goto fail_unlock;
 186                 }
 187         }
 188 tb_not_found:
 189         ret = 1;
 190         if (!tb && (tb = tcp_bucket_create(head, snum)) == NULL)
 191                 goto fail_unlock;
 192         if (hlist_empty(&tb->owners)) {
 193                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 194                         tb->fastreuse = 1;
 195                 else
 196                         tb->fastreuse = 0;
 197         } else if (tb->fastreuse &&
 198                    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 199                 tb->fastreuse = 0;
 200
 201 success:
 202         if (!tcp_sk(sk)->bind_hash)
 203                 tcp_bind_hash(sk, tb, snum);
 204         BUG_TRAP(tcp_sk(sk)->bind_hash == tb);
 205         ret = 0;
 206
 207 fail_unlock:
 208         spin_unlock(&head->lock);
 209 fail:
 210         local_bh_enable();
 211         return ret;
 212 }
 213
 214 static __inline__ void __tcp_v6_hash(struct sock *sk)
 215 {
 216         struct hlist_head *list;
 217         rwlock_t *lock;
 218
 219         BUG_TRAP(sk_unhashed(sk));
 220
 221         if (sk->sk_state == TCP_LISTEN) {
 222                 list = &tcp_listening_hash[tcp_sk_listen_hashfn(sk)];
 223                 lock = &tcp_lhash_lock;
 224                 tcp_listen_wlock();
 225         } else {
 226                 sk->sk_hashent = tcp_v6_sk_hashfn(sk);
 227                 list = &tcp_ehash[sk->sk_hashent].chain;
 228                 lock = &tcp_ehash[sk->sk_hashent].lock;
 229                 write_lock(lock);
 230         }
 231
 232         __sk_add_node(sk, list);
 233         sock_prot_inc_use(sk->sk_prot);
 234         write_unlock(lock);
 235 }
 236
 237
 238 static void tcp_v6_hash(struct sock *sk)
 239 {
 240         if (sk->sk_state != TCP_CLOSE) {
 241                 struct tcp_sock *tp = tcp_sk(sk);
 242
 243                 if (tp->af_specific == &ipv6_mapped) {
 244                         tcp_prot.hash(sk);
 245                         return;
 246                 }
 247                 local_bh_disable();
 248                 __tcp_v6_hash(sk);
 249                 local_bh_enable();
 250         }
 251 }
 252
 253 static struct sock *tcp_v6_lookup_listener(struct in6_addr *daddr, unsigned short hnum, int dif)
 254 {
 255         struct sock *sk;
 256         struct hlist_node *node;
 257         struct sock *result = NULL;
 258         int score, hiscore;
 259
 260         hiscore=0;
 261         read_lock(&tcp_lhash_lock);
 262         sk_for_each(sk, node, &tcp_listening_hash[tcp_lhashfn(hnum)]) {
 263                 if (inet_sk(sk)->num == hnum && sk->sk_family == PF_INET6) {
 264                         struct ipv6_pinfo *np = inet6_sk(sk);
 265
 266                         score = 1;
 267                         if (!ipv6_addr_any(&np->rcv_saddr)) {
 268                                 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
 269                                         continue;
 270                                 score++;
 271                         }
 272                         if (sk->sk_bound_dev_if) {
 273                                 if (sk->sk_bound_dev_if != dif)
 274                                         continue;
 275                                 score++;
 276                         }
 277                         if (score == 3) {
 278                                 result = sk;
 279                                 break;
 280                         }
 281                         if (score > hiscore) {
 282                                 hiscore = score;
 283                                 result = sk;
 284                         }
 285                 }
 286         }
 287         if (result)
 288                 sock_hold(result);
 289         read_unlock(&tcp_lhash_lock);
 290         return result;
 291 }
 292
 293 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
 294  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
 295  *
 296  * The sockhash lock must be held as a reader here.
 297  */
 298
 299 static inline struct sock *__tcp_v6_lookup_established(struct in6_addr *saddr, u16 sport,
 300                                                        struct in6_addr *daddr, u16 hnum,
 301                                                        int dif)
 302 {
 303         struct tcp_ehash_bucket *head;
 304         struct sock *sk;
 305         struct hlist_node *node;
 306         __u32 ports = TCP_COMBINED_PORTS(sport, hnum);
 307         int hash;
 308
 309         /* Optimize here for direct hit, only listening connections can
 310          * have wildcards anyways.
 311          */
 312         hash = tcp_v6_hashfn(daddr, hnum, saddr, sport);
 313         head = &tcp_ehash[hash];
 314         read_lock(&head->lock);
 315         sk_for_each(sk, node, &head->chain) {
 316                 /* For IPV6 do the cheaper port and family tests first. */
 317                 if(TCP_IPV6_MATCH(sk, saddr, daddr, ports, dif))
 318                         goto hit; /* You sunk my battleship! */
 319         }
 320         /* Must check for a TIME_WAIT'er before going to listener hash. */
 321         sk_for_each(sk, node, &(head + tcp_ehash_size)->chain) {
 322                 /* FIXME: acme: check this... */
 323                 struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk;
 324
 325                 if(*((__u32 *)&(tw->tw_dport))  == ports        &&
 326                    sk->sk_family                == PF_INET6) {
 327                         if(ipv6_addr_equal(&tw->tw_v6_daddr, saddr)     &&
 328                            ipv6_addr_equal(&tw->tw_v6_rcv_saddr, daddr) &&
 329                            (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif))
 330                                 goto hit;
 331                 }
 332         }
 333         read_unlock(&head->lock);
 334         return NULL;
 335
 336 hit:
 337         sock_hold(sk);
 338         read_unlock(&head->lock);
 339         return sk;
 340 }
 341
 342
 343 static inline struct sock *__tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
 344                                            struct in6_addr *daddr, u16 hnum,
 345                                            int dif)
 346 {
 347         struct sock *sk;
 348
 349         sk = __tcp_v6_lookup_established(saddr, sport, daddr, hnum, dif);
 350
 351         if (sk)
 352                 return sk;
 353
 354         return tcp_v6_lookup_listener(daddr, hnum, dif);
 355 }
 356
 357 inline struct sock *tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
 358                                   struct in6_addr *daddr, u16 dport,
 359                                   int dif)
 360 {
 361         struct sock *sk;
 362
 363         local_bh_disable();
 364         sk = __tcp_v6_lookup(saddr, sport, daddr, ntohs(dport), dif);
 365         local_bh_enable();
 366
 367         return sk;
 368 }
 369
 370 EXPORT_SYMBOL_GPL(tcp_v6_lookup);
 371
 372
 373 /*
 374  * Open request hash tables.
 375  */
 376
 377 static u32 tcp_v6_synq_hash(struct in6_addr *raddr, u16 rport, u32 rnd)
 378 {
 379         u32 a, b, c;
 380
 381         a = raddr->s6_addr32[0];
 382         b = raddr->s6_addr32[1];
 383         c = raddr->s6_addr32[2];
 384
 385         a += JHASH_GOLDEN_RATIO;
 386         b += JHASH_GOLDEN_RATIO;
 387         c += rnd;
 388         __jhash_mix(a, b, c);
 389
 390         a += raddr->s6_addr32[3];
 391         b += (u32) rport;
 392         __jhash_mix(a, b, c);
 393
 394         return c & (TCP_SYNQ_HSIZE - 1);
 395 }
 396
 397 static struct open_request *tcp_v6_search_req(struct tcp_sock *tp,
 398                                               struct open_request ***prevp,
 399                                               __u16 rport,
 400                                               struct in6_addr *raddr,
 401                                               struct in6_addr *laddr,
 402                                               int iif)
 403 {
 404         struct tcp_listen_opt *lopt = tp->listen_opt;
 405         struct open_request *req, **prev;
 406
 407         for (prev = &lopt->syn_table[tcp_v6_synq_hash(raddr, rport, lopt->hash_rnd)];
 408              (req = *prev) != NULL;
 409              prev = &req->dl_next) {
 410                 if (req->rmt_port == rport &&
 411                     req->class->family == AF_INET6 &&
 412                     ipv6_addr_equal(&req->af.v6_req.rmt_addr, raddr) &&
 413                     ipv6_addr_equal(&req->af.v6_req.loc_addr, laddr) &&
 414                     (!req->af.v6_req.iif || req->af.v6_req.iif == iif)) {
 415                         BUG_TRAP(req->sk == NULL);
 416                         *prevp = prev;
 417                         return req;
 418                 }
 419         }
 420
 421         return NULL;
 422 }
 423
 424 static __inline__ u16 tcp_v6_check(struct tcphdr *th, int len,
 425                                    struct in6_addr *saddr,
 426                                    struct in6_addr *daddr,
 427                                    unsigned long base)
 428 {
 429         return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
 430 }
 431
 432 static __u32 tcp_v6_init_sequence(struct sock *sk, struct sk_buff *skb)
 433 {
 434         if (skb->protocol == htons(ETH_P_IPV6)) {
 435                 return secure_tcpv6_sequence_number(skb->nh.ipv6h->daddr.s6_addr32,
 436                                                     skb->nh.ipv6h->saddr.s6_addr32,
 437                                                     skb->h.th->dest,
 438                                                     skb->h.th->source);
 439         } else {
 440                 return secure_tcp_sequence_number(skb->nh.iph->daddr,
 441                                                   skb->nh.iph->saddr,
 442                                                   skb->h.th->dest,
 443                                                   skb->h.th->source);
 444         }
 445 }
 446
 447 static int __tcp_v6_check_established(struct sock *sk, __u16 lport,
 448                                       struct tcp_tw_bucket **twp)
 449 {
 450         struct inet_sock *inet = inet_sk(sk);
 451         struct ipv6_pinfo *np = inet6_sk(sk);
 452         struct in6_addr *daddr = &np->rcv_saddr;
 453         struct in6_addr *saddr = &np->daddr;
 454         int dif = sk->sk_bound_dev_if;
 455         u32 ports = TCP_COMBINED_PORTS(inet->dport, lport);
 456         int hash = tcp_v6_hashfn(daddr, inet->num, saddr, inet->dport);
 457         struct tcp_ehash_bucket *head = &tcp_ehash[hash];
 458         struct sock *sk2;
 459         struct hlist_node *node;
 460         struct tcp_tw_bucket *tw;
 461
 462         write_lock(&head->lock);
 463
 464         /* Check TIME-WAIT sockets first. */
 465         sk_for_each(sk2, node, &(head + tcp_ehash_size)->chain) {
 466                 tw = (struct tcp_tw_bucket*)sk2;
 467
 468                 if(*((__u32 *)&(tw->tw_dport))  == ports        &&
 469                    sk2->sk_family               == PF_INET6     &&
 470                    ipv6_addr_equal(&tw->tw_v6_daddr, saddr)     &&
 471                    ipv6_addr_equal(&tw->tw_v6_rcv_saddr, daddr) &&
 472                    sk2->sk_bound_dev_if == sk->sk_bound_dev_if) {
 473                         struct tcp_sock *tp = tcp_sk(sk);
 474
 475                         if (tw->tw_ts_recent_stamp &&
 476                             (!twp || (sysctl_tcp_tw_reuse &&
 477                                       xtime.tv_sec -
 478                                       tw->tw_ts_recent_stamp > 1))) {
 479                                 /* See comment in tcp_ipv4.c */
 480                                 tp->write_seq = tw->tw_snd_nxt + 65535 + 2;
 481                                 if (!tp->write_seq)
 482                                         tp->write_seq = 1;
 483                                 tp->rx_opt.ts_recent = tw->tw_ts_recent;
 484                                 tp->rx_opt.ts_recent_stamp = tw->tw_ts_recent_stamp;
 485                                 sock_hold(sk2);
 486                                 goto unique;
 487                         } else
 488                                 goto not_unique;
 489                 }
 490         }
 491         tw = NULL;
 492
 493         /* And established part... */
 494         sk_for_each(sk2, node, &head->chain) {
 495                 if(TCP_IPV6_MATCH(sk2, saddr, daddr, ports, dif))
 496                         goto not_unique;
 497         }
 498
 499 unique:
 500         BUG_TRAP(sk_unhashed(sk));
 501         __sk_add_node(sk, &head->chain);
 502         sk->sk_hashent = hash;
 503         sock_prot_inc_use(sk->sk_prot);
 504         write_unlock(&head->lock);
 505
 506         if (twp) {
 507                 *twp = tw;
 508                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 509         } else if (tw) {
 510                 /* Silly. Should hash-dance instead... */
 511                 tcp_tw_deschedule(tw);
 512                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 513
 514                 tcp_tw_put(tw);
 515         }
 516         return 0;
 517
 518 not_unique:
 519         write_unlock(&head->lock);
 520         return -EADDRNOTAVAIL;
 521 }
 522
 523 static inline u32 tcpv6_port_offset(const struct sock *sk)
 524 {
 525         const struct inet_sock *inet = inet_sk(sk);
 526         const struct ipv6_pinfo *np = inet6_sk(sk);
 527
 528         return secure_tcpv6_port_ephemeral(np->rcv_saddr.s6_addr32,
 529                                            np->daddr.s6_addr32,
 530                                            inet->dport);
 531 }
 532
 533 static int tcp_v6_hash_connect(struct sock *sk)
 534 {
 535         unsigned short snum = inet_sk(sk)->num;
 536         struct tcp_bind_hashbucket *head;
 537         struct tcp_bind_bucket *tb;
 538         int ret;
 539
 540         if (!snum) {
 541                 int low = sysctl_local_port_range[0];
 542                 int high = sysctl_local_port_range[1];
 543                 int range = high - low;
 544                 int i;
 545                 int port;
 546                 static u32 hint;
 547                 u32 offset = hint + tcpv6_port_offset(sk);
 548                 struct hlist_node *node;
 549                 struct tcp_tw_bucket *tw = NULL;
 550
 551                 local_bh_disable();
 552                 for (i = 1; i <= range; i++) {
 553                         port = low + (i + offset) % range;
 554                         head = &tcp_bhash[tcp_bhashfn(port)];
 555                         spin_lock(&head->lock);
 556
 557                         /* Does not bother with rcv_saddr checks,
 558                          * because the established check is already
 559                          * unique enough.
 560                          */
 561                         tb_for_each(tb, node, &head->chain) {
 562                                 if (tb->port == port) {
 563                                         BUG_TRAP(!hlist_empty(&tb->owners));
 564                                         if (tb->fastreuse >= 0)
 565                                                 goto next_port;
 566                                         if (!__tcp_v6_check_established(sk,
 567                                                                         port,
 568                                                                         &tw))
 569                                                 goto ok;
 570                                         goto next_port;
 571                                 }
 572                         }
 573
 574                         tb = tcp_bucket_create(head, port);
 575                         if (!tb) {
 576                                 spin_unlock(&head->lock);
 577                                 break;
 578                         }
 579                         tb->fastreuse = -1;
 580                         goto ok;
 581
 582                 next_port:
 583                         spin_unlock(&head->lock);
 584                 }
 585                 local_bh_enable();
 586
 587                 return -EADDRNOTAVAIL;
 588
 589 ok:
 590                 hint += i;
 591
 592                 /* Head lock still held and bh's disabled */
 593                 tcp_bind_hash(sk, tb, port);
 594                 if (sk_unhashed(sk)) {
 595                         inet_sk(sk)->sport = htons(port);
 596                         __tcp_v6_hash(sk);
 597                 }
 598                 spin_unlock(&head->lock);
 599
 600                 if (tw) {
 601                         tcp_tw_deschedule(tw);
 602                         tcp_tw_put(tw);
 603                 }
 604
 605                 ret = 0;
 606                 goto out;
 607         }
 608
 609         head  = &tcp_bhash[tcp_bhashfn(snum)];
 610         tb  = tcp_sk(sk)->bind_hash;
 611         spin_lock_bh(&head->lock);
 612
 613         if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
 614                 __tcp_v6_hash(sk);
 615                 spin_unlock_bh(&head->lock);
 616                 return 0;
 617         } else {
 618                 spin_unlock(&head->lock);
 619                 /* No definite answer... Walk to established hash table */
 620                 ret = __tcp_v6_check_established(sk, snum, NULL);
 621 out:
 622                 local_bh_enable();
 623                 return ret;
 624         }
 625 }
 626
 627 static __inline__ int tcp_v6_iif(struct sk_buff *skb)
 628 {
 629         return IP6CB(skb)->iif;
 630 }
 631
 632 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 633                           int addr_len)
 634 {
 635         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 636         struct inet_sock *inet = inet_sk(sk);
 637         struct ipv6_pinfo *np = inet6_sk(sk);
 638         struct tcp_sock *tp = tcp_sk(sk);
 639         struct in6_addr *saddr = NULL, *final_p = NULL, final;
 640         struct flowi fl;
 641         struct dst_entry *dst;
 642         int addr_type;
 643         int err;
 644
 645         if (addr_len < SIN6_LEN_RFC2133)
 646                 return -EINVAL;
 647
 648         if (usin->sin6_family != AF_INET6)
 649                 return(-EAFNOSUPPORT);
 650
 651         memset(&fl, 0, sizeof(fl));
 652
 653         if (np->sndflow) {
 654                 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 655                 IP6_ECN_flow_init(fl.fl6_flowlabel);
 656                 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
 657                         struct ip6_flowlabel *flowlabel;
 658                         flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
 659                         if (flowlabel == NULL)
 660                                 return -EINVAL;
 661                         ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
 662                         fl6_sock_release(flowlabel);
 663                 }
 664         }
 665
 666         /*
 667          *      connect() to INADDR_ANY means loopback (BSD'ism).
 668          */
 669
 670         if(ipv6_addr_any(&usin->sin6_addr))
 671                 usin->sin6_addr.s6_addr[15] = 0x1;
 672
 673         addr_type = ipv6_addr_type(&usin->sin6_addr);
 674
 675         if(addr_type & IPV6_ADDR_MULTICAST)
 676                 return -ENETUNREACH;
 677
 678         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 679                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 680                     usin->sin6_scope_id) {
 681                         /* If interface is set while binding, indices
 682                          * must coincide.
 683                          */
 684                         if (sk->sk_bound_dev_if &&
 685                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 686                                 return -EINVAL;
 687
 688                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 689                 }
 690
 691                 /* Connect to link-local address requires an interface */
 692                 if (!sk->sk_bound_dev_if)
 693                         return -EINVAL;
 694         }
 695
 696         if (tp->rx_opt.ts_recent_stamp &&
 697             !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
 698                 tp->rx_opt.ts_recent = 0;
 699                 tp->rx_opt.ts_recent_stamp = 0;
 700                 tp->write_seq = 0;
 701         }
 702
 703         ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
 704         np->flow_label = fl.fl6_flowlabel;
 705
 706         /*
 707          *      TCP over IPv4
 708          */
 709
 710         if (addr_type == IPV6_ADDR_MAPPED) {
 711                 u32 exthdrlen = tp->ext_header_len;
 712                 struct sockaddr_in sin;
 713
 714                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 715
 716                 if (__ipv6_only_sock(sk))
 717                         return -ENETUNREACH;
 718
 719                 sin.sin_family = AF_INET;
 720                 sin.sin_port = usin->sin6_port;
 721                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 722
 723                 tp->af_specific = &ipv6_mapped;
 724                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 725
 726                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 727
 728                 if (err) {
 729                         tp->ext_header_len = exthdrlen;
 730                         tp->af_specific = &ipv6_specific;
 731                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 732                         goto failure;
 733                 } else {
 734                         ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
 735                                       inet->saddr);
 736                         ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
 737                                       inet->rcv_saddr);
 738                 }
 739
 740                 return err;
 741         }
 742
 743         if (!ipv6_addr_any(&np->rcv_saddr))
 744                 saddr = &np->rcv_saddr;
 745
 746         fl.proto = IPPROTO_TCP;
 747         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 748         ipv6_addr_copy(&fl.fl6_src,
 749                        (saddr ? saddr : &np->saddr));
 750         fl.oif = sk->sk_bound_dev_if;
 751         fl.fl_ip_dport = usin->sin6_port;
 752         fl.fl_ip_sport = inet->sport;
 753
 754         if (np->opt && np->opt->srcrt) {
 755                 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
 756                 ipv6_addr_copy(&final, &fl.fl6_dst);
 757                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 758                 final_p = &final;
 759         }
 760
 761         err = ip6_dst_lookup(sk, &dst, &fl);
 762         if (err)
 763                 goto failure;
 764         if (final_p)
 765                 ipv6_addr_copy(&fl.fl6_dst, final_p);
 766
 767         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
 768                 dst_release(dst);
 769                 goto failure;
 770         }
 771
 772         if (saddr == NULL) {
 773                 saddr = &fl.fl6_src;
 774                 ipv6_addr_copy(&np->rcv_saddr, saddr);
 775         }
 776
 777         /* set the source address */
 778         ipv6_addr_copy(&np->saddr, saddr);
 779         inet->rcv_saddr = LOOPBACK4_IPV6;
 780
 781         ip6_dst_store(sk, dst, NULL);
 782         sk->sk_route_caps = dst->dev->features &
 783                 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
 784
 785         tp->ext_header_len = 0;
 786         if (np->opt)
 787                 tp->ext_header_len = np->opt->opt_flen + np->opt->opt_nflen;
 788
 789         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 790
 791         inet->dport = usin->sin6_port;
 792
 793         tcp_set_state(sk, TCP_SYN_SENT);
 794         err = tcp_v6_hash_connect(sk);
 795         if (err)
 796                 goto late_failure;
 797
 798         if (!tp->write_seq)
 799                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 800                                                              np->daddr.s6_addr32,
 801                                                              inet->sport,
 802                                                              inet->dport);
 803
 804         err = tcp_connect(sk);
 805         if (err)
 806                 goto late_failure;
 807
 808         return 0;
 809
 810 late_failure:
 811         tcp_set_state(sk, TCP_CLOSE);
 812         __sk_dst_reset(sk);
 813 failure:
 814         inet->dport = 0;
 815         sk->sk_route_caps = 0;
 816         return err;
 817 }
 818
 819 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 820                 int type, int code, int offset, __u32 info)
 821 {
 822         struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
 823         struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 824         struct ipv6_pinfo *np;
 825         struct sock *sk;
 826         int err;
 827         struct tcp_sock *tp;
 828         __u32 seq;
 829
 830         sk = tcp_v6_lookup(&hdr->daddr, th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
 831
 832         if (sk == NULL) {
 833                 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
 834                 return;
 835         }
 836
 837         if (sk->sk_state == TCP_TIME_WAIT) {
 838                 tcp_tw_put((struct tcp_tw_bucket*)sk);
 839                 return;
 840         }
 841
 842         bh_lock_sock(sk);
 843         if (sock_owned_by_user(sk))
 844                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 845
 846         if (sk->sk_state == TCP_CLOSE)
 847                 goto out;
 848
 849         tp = tcp_sk(sk);
 850         seq = ntohl(th->seq);
 851         if (sk->sk_state != TCP_LISTEN &&
 852             !between(seq, tp->snd_una, tp->snd_nxt)) {
 853                 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 854                 goto out;
 855         }
 856
 857         np = inet6_sk(sk);
 858
 859         if (type == ICMPV6_PKT_TOOBIG) {
 860                 struct dst_entry *dst = NULL;
 861
 862                 if (sock_owned_by_user(sk))
 863                         goto out;
 864                 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 865                         goto out;
 866
 867                 /* icmp should have updated the destination cache entry */
 868                 dst = __sk_dst_check(sk, np->dst_cookie);
 869
 870                 if (dst == NULL) {
 871                         struct inet_sock *inet = inet_sk(sk);
 872                         struct flowi fl;
 873
 874                         /* BUGGG_FUTURE: Again, it is not clear how
 875                            to handle rthdr case. Ignore this complexity
 876                            for now.
 877                          */
 878                         memset(&fl, 0, sizeof(fl));
 879                         fl.proto = IPPROTO_TCP;
 880                         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 881                         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
 882                         fl.oif = sk->sk_bound_dev_if;
 883                         fl.fl_ip_dport = inet->dport;
 884                         fl.fl_ip_sport = inet->sport;
 885
 886                         if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
 887                                 sk->sk_err_soft = -err;
 888                                 goto out;
 889                         }
 890
 891                         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
 892                                 sk->sk_err_soft = -err;
 893                                 goto out;
 894                         }
 895
 896                 } else
 897                         dst_hold(dst);
 898
 899                 if (tp->pmtu_cookie > dst_mtu(dst)) {
 900                         tcp_sync_mss(sk, dst_mtu(dst));
 901                         tcp_simple_retransmit(sk);
 902                 } /* else let the usual retransmit timer handle it */
 903                 dst_release(dst);
 904                 goto out;
 905         }
 906
 907         icmpv6_err_convert(type, code, &err);
 908
 909         /* Might be for an open_request */
 910         switch (sk->sk_state) {
 911                 struct open_request *req, **prev;
 912         case TCP_LISTEN:
 913                 if (sock_owned_by_user(sk))
 914                         goto out;
 915
 916                 req = tcp_v6_search_req(tp, &prev, th->dest, &hdr->daddr,
 917                                         &hdr->saddr, tcp_v6_iif(skb));
 918                 if (!req)
 919                         goto out;
 920
 921                 /* ICMPs are not backlogged, hence we cannot get
 922                  * an established socket here.
 923                  */
 924                 BUG_TRAP(req->sk == NULL);
 925
 926                 if (seq != req->snt_isn) {
 927                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 928                         goto out;
 929                 }
 930
 931                 tcp_synq_drop(sk, req, prev);
 932                 goto out;
 933
 934         case TCP_SYN_SENT:
 935         case TCP_SYN_RECV:  /* Cannot happen.
 936                                It can, it SYNs are crossed. --ANK */
 937                 if (!sock_owned_by_user(sk)) {
 938                         TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
 939                         sk->sk_err = err;
 940                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 941
 942                         tcp_done(sk);
 943                 } else
 944                         sk->sk_err_soft = err;
 945                 goto out;
 946         }
 947
 948         if (!sock_owned_by_user(sk) && np->recverr) {
 949                 sk->sk_err = err;
 950                 sk->sk_error_report(sk);
 951         } else
 952                 sk->sk_err_soft = err;
 953
 954 out:
 955         bh_unlock_sock(sk);
 956         sock_put(sk);
 957 }
 958
 959
 960 static int tcp_v6_send_synack(struct sock *sk, struct open_request *req,
 961                               struct dst_entry *dst)
 962 {
 963         struct ipv6_pinfo *np = inet6_sk(sk);
 964         struct sk_buff * skb;
 965         struct ipv6_txoptions *opt = NULL;
 966         struct in6_addr * final_p = NULL, final;
 967         struct flowi fl;
 968         int err = -1;
 969
 970         memset(&fl, 0, sizeof(fl));
 971         fl.proto = IPPROTO_TCP;
 972         ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
 973         ipv6_addr_copy(&fl.fl6_src, &req->af.v6_req.loc_addr);
 974         fl.fl6_flowlabel = 0;
 975         fl.oif = req->af.v6_req.iif;
 976         fl.fl_ip_dport = req->rmt_port;
 977         fl.fl_ip_sport = inet_sk(sk)->sport;
 978
 979         if (dst == NULL) {
 980                 opt = np->opt;
 981                 if (opt == NULL &&
 982                     np->rxopt.bits.srcrt == 2 &&
 983                     req->af.v6_req.pktopts) {
 984                         struct sk_buff *pktopts = req->af.v6_req.pktopts;
 985                         struct inet6_skb_parm *rxopt = IP6CB(pktopts);
 986                         if (rxopt->srcrt)
 987                                 opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(pktopts->nh.raw + rxopt->srcrt));
 988                 }
 989
 990                 if (opt && opt->srcrt) {
 991                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
 992                         ipv6_addr_copy(&final, &fl.fl6_dst);
 993                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 994                         final_p = &final;
 995                 }
 996
 997                 err = ip6_dst_lookup(sk, &dst, &fl);
 998                 if (err)
 999                         goto done;
1000                 if (final_p)
1001                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1002                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1003                         goto done;
1004         }
1005
1006         skb = tcp_make_synack(sk, dst, req);
1007         if (skb) {
1008                 struct tcphdr *th = skb->h.th;
1009
1010                 th->check = tcp_v6_check(th, skb->len,
1011                                          &req->af.v6_req.loc_addr, &req->af.v6_req.rmt_addr,
1012                                          csum_partial((char *)th, skb->len, skb->csum));
1013
1014                 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
1015                 err = ip6_xmit(sk, skb, &fl, opt, 0);
1016                 if (err == NET_XMIT_CN)
1017                         err = 0;
1018         }
1019
1020 done:
1021         dst_release(dst);
1022         if (opt && opt != np->opt)
1023                 sock_kfree_s(sk, opt, opt->tot_len);
1024         return err;
1025 }
1026
1027 static void tcp_v6_or_free(struct open_request *req)
1028 {
1029         if (req->af.v6_req.pktopts)
1030                 kfree_skb(req->af.v6_req.pktopts);
1031 }
1032
1033 static struct or_calltable or_ipv6 = {
1034         .family         =       AF_INET6,
1035         .rtx_syn_ack    =       tcp_v6_send_synack,
1036         .send_ack       =       tcp_v6_or_send_ack,
1037         .destructor     =       tcp_v6_or_free,
1038         .send_reset     =       tcp_v6_send_reset
1039 };
1040
1041 static int ipv6_opt_accepted(struct sock *sk, struct sk_buff *skb)
1042 {
1043         struct ipv6_pinfo *np = inet6_sk(sk);
1044         struct inet6_skb_parm *opt = IP6CB(skb);
1045
1046         if (np->rxopt.all) {
1047                 if ((opt->hop && np->rxopt.bits.hopopts) ||
1048                     ((IPV6_FLOWINFO_MASK&*(u32*)skb->nh.raw) &&
1049                      np->rxopt.bits.rxflow) ||
1050                     (opt->srcrt && np->rxopt.bits.srcrt) ||
1051                     ((opt->dst1 || opt->dst0) && np->rxopt.bits.dstopts))
1052                         return 1;
1053         }
1054         return 0;
1055 }
1056
1057
1058 static void tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len,
1059                               struct sk_buff *skb)
1060 {
1061         struct ipv6_pinfo *np = inet6_sk(sk);
1062
1063         if (skb->ip_summed == CHECKSUM_HW) {
1064                 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,  0);
1065                 skb->csum = offsetof(struct tcphdr, check);
1066         } else {
1067                 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
1068                                             csum_partial((char *)th, th->doff<<2,
1069                                                          skb->csum));
1070         }
1071 }
1072
1073
1074 static void tcp_v6_send_reset(struct sk_buff *skb)
1075 {
1076         struct tcphdr *th = skb->h.th, *t1;
1077         struct sk_buff *buff;
1078         struct flowi fl;
1079
1080         if (th->rst)
1081                 return;
1082
1083         if (!ipv6_unicast_destination(skb))
1084                 return;
1085
1086         /*
1087          * We need to grab some memory, and put together an RST,
1088          * and then put it into the queue to be sent.
1089          */
1090
1091         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + sizeof(struct tcphdr),
1092                          GFP_ATOMIC);
1093         if (buff == NULL)
1094                 return;
1095
1096         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + sizeof(struct tcphdr));
1097
1098         t1 = (struct tcphdr *) skb_push(buff,sizeof(struct tcphdr));
1099
1100         /* Swap the send and the receive. */
1101         memset(t1, 0, sizeof(*t1));
1102         t1->dest = th->source;
1103         t1->source = th->dest;
1104         t1->doff = sizeof(*t1)/4;
1105         t1->rst = 1;
1106
1107         if(th->ack) {
1108                 t1->seq = th->ack_seq;
1109         } else {
1110                 t1->ack = 1;
1111                 t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
1112                                     + skb->len - (th->doff<<2));
1113         }
1114
1115         buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1116
1117         memset(&fl, 0, sizeof(fl));
1118         ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
1119         ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
1120
1121         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1122                                     sizeof(*t1), IPPROTO_TCP,
1123                                     buff->csum);
1124
1125         fl.proto = IPPROTO_TCP;
1126         fl.oif = tcp_v6_iif(skb);
1127         fl.fl_ip_dport = t1->dest;
1128         fl.fl_ip_sport = t1->source;
1129
1130         /* sk = NULL, but it is safe for now. RST socket required. */
1131         if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1132
1133                 if ((xfrm_lookup(&buff->dst, &fl, NULL, 0)) < 0) {
1134                         dst_release(buff->dst);
1135                         return;
1136                 }
1137
1138                 ip6_xmit(NULL, buff, &fl, NULL, 0);
1139                 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1140                 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
1141                 return;
1142         }
1143
1144         kfree_skb(buff);
1145 }
1146
1147 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
1148 {
1149         struct tcphdr *th = skb->h.th, *t1;
1150         struct sk_buff *buff;
1151         struct flowi fl;
1152         int tot_len = sizeof(struct tcphdr);
1153
1154         if (ts)
1155                 tot_len += 3*4;
1156
1157         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1158                          GFP_ATOMIC);
1159         if (buff == NULL)
1160                 return;
1161
1162         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1163
1164         t1 = (struct tcphdr *) skb_push(buff,tot_len);
1165
1166         /* Swap the send and the receive. */
1167         memset(t1, 0, sizeof(*t1));
1168         t1->dest = th->source;
1169         t1->source = th->dest;
1170         t1->doff = tot_len/4;
1171         t1->seq = htonl(seq);
1172         t1->ack_seq = htonl(ack);
1173         t1->ack = 1;
1174         t1->window = htons(win);
1175
1176         if (ts) {
1177                 u32 *ptr = (u32*)(t1 + 1);
1178                 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1179                                (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1180                 *ptr++ = htonl(tcp_time_stamp);
1181                 *ptr = htonl(ts);
1182         }
1183
1184         buff->csum = csum_partial((char *)t1, tot_len, 0);
1185
1186         memset(&fl, 0, sizeof(fl));
1187         ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
1188         ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
1189
1190         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1191                                     tot_len, IPPROTO_TCP,
1192                                     buff->csum);
1193
1194         fl.proto = IPPROTO_TCP;
1195         fl.oif = tcp_v6_iif(skb);
1196         fl.fl_ip_dport = t1->dest;
1197         fl.fl_ip_sport = t1->source;
1198
1199         if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1200                 if ((xfrm_lookup(&buff->dst, &fl, NULL, 0)) < 0) {
1201                         dst_release(buff->dst);
1202                         return;
1203                 }
1204                 ip6_xmit(NULL, buff, &fl, NULL, 0);
1205                 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1206                 return;
1207         }
1208
1209         kfree_skb(buff);
1210 }
1211
1212 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1213 {
1214         struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk;
1215
1216         tcp_v6_send_ack(skb, tw->tw_snd_nxt, tw->tw_rcv_nxt,
1217                         tw->tw_rcv_wnd >> tw->tw_rcv_wscale, tw->tw_ts_recent);
1218
1219         tcp_tw_put(tw);
1220 }
1221
1222 static void tcp_v6_or_send_ack(struct sk_buff *skb, struct open_request *req)
1223 {
1224         tcp_v6_send_ack(skb, req->snt_isn+1, req->rcv_isn+1, req->rcv_wnd, req->ts_recent);
1225 }
1226
1227
1228 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1229 {
1230         struct open_request *req, **prev;
1231         struct tcphdr *th = skb->h.th;
1232         struct tcp_sock *tp = tcp_sk(sk);
1233         struct sock *nsk;
1234
1235         /* Find possible connection requests. */
1236         req = tcp_v6_search_req(tp, &prev, th->source, &skb->nh.ipv6h->saddr,
1237                                 &skb->nh.ipv6h->daddr, tcp_v6_iif(skb));
1238         if (req)
1239                 return tcp_check_req(sk, skb, req, prev);
1240
1241         nsk = __tcp_v6_lookup_established(&skb->nh.ipv6h->saddr,
1242                                           th->source,
1243                                           &skb->nh.ipv6h->daddr,
1244                                           ntohs(th->dest),
1245                                           tcp_v6_iif(skb));
1246
1247         if (nsk) {
1248                 if (nsk->sk_state != TCP_TIME_WAIT) {
1249                         bh_lock_sock(nsk);
1250                         return nsk;
1251                 }
1252                 tcp_tw_put((struct tcp_tw_bucket*)nsk);
1253                 return NULL;
1254         }
1255
1256 #if 0 /*def CONFIG_SYN_COOKIES*/
1257         if (!th->rst && !th->syn && th->ack)
1258                 sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
1259 #endif
1260         return sk;
1261 }
1262
1263 static void tcp_v6_synq_add(struct sock *sk, struct open_request *req)
1264 {
1265         struct tcp_sock *tp = tcp_sk(sk);
1266         struct tcp_listen_opt *lopt = tp->listen_opt;
1267         u32 h = tcp_v6_synq_hash(&req->af.v6_req.rmt_addr, req->rmt_port, lopt->hash_rnd);
1268
1269         req->sk = NULL;
1270         req->expires = jiffies + TCP_TIMEOUT_INIT;
1271         req->retrans = 0;
1272         req->dl_next = lopt->syn_table[h];
1273
1274         write_lock(&tp->syn_wait_lock);
1275         lopt->syn_table[h] = req;
1276         write_unlock(&tp->syn_wait_lock);
1277
1278         tcp_synq_added(sk);
1279 }
1280
1281
1282 /* FIXME: this is substantially similar to the ipv4 code.
1283  * Can some kind of merge be done? -- erics
1284  */
1285 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1286 {
1287         struct ipv6_pinfo *np = inet6_sk(sk);
1288         struct tcp_options_received tmp_opt;
1289         struct tcp_sock *tp = tcp_sk(sk);
1290         struct open_request *req = NULL;
1291         __u32 isn = TCP_SKB_CB(skb)->when;
1292
1293         if (skb->protocol == htons(ETH_P_IP))
1294                 return tcp_v4_conn_request(sk, skb);
1295
1296         if (!ipv6_unicast_destination(skb))
1297                 goto drop;
1298
1299         /*
1300          *      There are no SYN attacks on IPv6, yet...
1301          */
1302         if (tcp_synq_is_full(sk) && !isn) {
1303                 if (net_ratelimit())
1304                         printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
1305                 goto drop;
1306         }
1307
1308         if (sk_acceptq_is_full(sk) && tcp_synq_young(sk) > 1)
1309                 goto drop;
1310
1311         req = tcp_openreq_alloc();
1312         if (req == NULL)
1313                 goto drop;
1314
1315         tcp_clear_options(&tmp_opt);
1316         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1317         tmp_opt.user_mss = tp->rx_opt.user_mss;
1318
1319         tcp_parse_options(skb, &tmp_opt, 0);
1320
1321         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1322         tcp_openreq_init(req, &tmp_opt, skb);
1323
1324         req->class = &or_ipv6;
1325         ipv6_addr_copy(&req->af.v6_req.rmt_addr, &skb->nh.ipv6h->saddr);
1326         ipv6_addr_copy(&req->af.v6_req.loc_addr, &skb->nh.ipv6h->daddr);
1327         TCP_ECN_create_request(req, skb->h.th);
1328         req->af.v6_req.pktopts = NULL;
1329         if (ipv6_opt_accepted(sk, skb) ||
1330             np->rxopt.bits.rxinfo ||
1331             np->rxopt.bits.rxhlim) {
1332                 atomic_inc(&skb->users);
1333                 req->af.v6_req.pktopts = skb;
1334         }
1335         req->af.v6_req.iif = sk->sk_bound_dev_if;
1336
1337         /* So that link locals have meaning */
1338         if (!sk->sk_bound_dev_if &&
1339             ipv6_addr_type(&req->af.v6_req.rmt_addr) & IPV6_ADDR_LINKLOCAL)
1340                 req->af.v6_req.iif = tcp_v6_iif(skb);
1341
1342         if (isn == 0)
1343                 isn = tcp_v6_init_sequence(sk,skb);
1344
1345         req->snt_isn = isn;
1346
1347         if (tcp_v6_send_synack(sk, req, NULL))
1348                 goto drop;
1349
1350         tcp_v6_synq_add(sk, req);
1351
1352         return 0;
1353
1354 drop:
1355         if (req)
1356                 tcp_openreq_free(req);
1357
1358         TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
1359         return 0; /* don't send reset */
1360 }
1361
1362 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1363                                           struct open_request *req,
1364                                           struct dst_entry *dst)
1365 {
1366         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1367         struct tcp6_sock *newtcp6sk;
1368         struct inet_sock *newinet;
1369         struct tcp_sock *newtp;
1370         struct sock *newsk;
1371         struct ipv6_txoptions *opt;
1372
1373         if (skb->protocol == htons(ETH_P_IP)) {
1374                 /*
1375                  *      v6 mapped
1376                  */
1377
1378                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1379
1380                 if (newsk == NULL)
1381                         return NULL;
1382
1383                 newtcp6sk = (struct tcp6_sock *)newsk;
1384                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1385
1386                 newinet = inet_sk(newsk);
1387                 newnp = inet6_sk(newsk);
1388                 newtp = tcp_sk(newsk);
1389
1390                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1391
1392                 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1393                               newinet->daddr);
1394
1395                 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1396                               newinet->saddr);
1397
1398                 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1399
1400                 newtp->af_specific = &ipv6_mapped;
1401                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1402                 newnp->pktoptions  = NULL;
1403                 newnp->opt         = NULL;
1404                 newnp->mcast_oif   = tcp_v6_iif(skb);
1405                 newnp->mcast_hops  = skb->nh.ipv6h->hop_limit;
1406
1407                 /* Charge newly allocated IPv6 socket. Though it is mapped,
1408                  * it is IPv6 yet.
1409                  */
1410 #ifdef INET_REFCNT_DEBUG
1411                 atomic_inc(&inet6_sock_nr);
1412 #endif
1413
1414                 /* It is tricky place. Until this moment IPv4 tcp
1415                    worked with IPv6 af_tcp.af_specific.
1416                    Sync it now.
1417                  */
1418                 tcp_sync_mss(newsk, newtp->pmtu_cookie);
1419
1420                 return newsk;
1421         }
1422
1423         opt = np->opt;
1424
1425         if (sk_acceptq_is_full(sk))
1426                 goto out_overflow;
1427
1428         if (np->rxopt.bits.srcrt == 2 &&
1429             opt == NULL && req->af.v6_req.pktopts) {
1430                 struct inet6_skb_parm *rxopt = IP6CB(req->af.v6_req.pktopts);
1431                 if (rxopt->srcrt)
1432                         opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(req->af.v6_req.pktopts->nh.raw+rxopt->srcrt));
1433         }
1434
1435         if (dst == NULL) {
1436                 struct in6_addr *final_p = NULL, final;
1437                 struct flowi fl;
1438
1439                 memset(&fl, 0, sizeof(fl));
1440                 fl.proto = IPPROTO_TCP;
1441                 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
1442                 if (opt && opt->srcrt) {
1443                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1444                         ipv6_addr_copy(&final, &fl.fl6_dst);
1445                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1446                         final_p = &final;
1447                 }
1448                 ipv6_addr_copy(&fl.fl6_src, &req->af.v6_req.loc_addr);
1449                 fl.oif = sk->sk_bound_dev_if;
1450                 fl.fl_ip_dport = req->rmt_port;
1451                 fl.fl_ip_sport = inet_sk(sk)->sport;
1452
1453                 if (ip6_dst_lookup(sk, &dst, &fl))
1454                         goto out;
1455
1456                 if (final_p)
1457                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1458
1459                 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1460                         goto out;
1461         }
1462
1463         newsk = tcp_create_openreq_child(sk, req, skb);
1464         if (newsk == NULL)
1465                 goto out;
1466
1467         /* Charge newly allocated IPv6 socket */
1468 #ifdef INET_REFCNT_DEBUG
1469         atomic_inc(&inet6_sock_nr);
1470 #endif
1471
1472         ip6_dst_store(newsk, dst, NULL);
1473         newsk->sk_route_caps = dst->dev->features &
1474                 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1475
1476         newtcp6sk = (struct tcp6_sock *)newsk;
1477         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1478
1479         newtp = tcp_sk(newsk);
1480         newinet = inet_sk(newsk);
1481         newnp = inet6_sk(newsk);
1482
1483         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1484
1485         ipv6_addr_copy(&newnp->daddr, &req->af.v6_req.rmt_addr);
1486         ipv6_addr_copy(&newnp->saddr, &req->af.v6_req.loc_addr);
1487         ipv6_addr_copy(&newnp->rcv_saddr, &req->af.v6_req.loc_addr);
1488         newsk->sk_bound_dev_if = req->af.v6_req.iif;
1489
1490         /* Now IPv6 options...
1491
1492            First: no IPv4 options.
1493          */
1494         newinet->opt = NULL;
1495
1496         /* Clone RX bits */
1497         newnp->rxopt.all = np->rxopt.all;
1498
1499         /* Clone pktoptions received with SYN */
1500         newnp->pktoptions = NULL;
1501         if (req->af.v6_req.pktopts) {
1502                 newnp->pktoptions = skb_clone(req->af.v6_req.pktopts,
1503                                               GFP_ATOMIC);
1504                 kfree_skb(req->af.v6_req.pktopts);
1505                 req->af.v6_req.pktopts = NULL;
1506                 if (newnp->pktoptions)
1507                         skb_set_owner_r(newnp->pktoptions, newsk);
1508         }
1509         newnp->opt        = NULL;
1510         newnp->mcast_oif  = tcp_v6_iif(skb);
1511         newnp->mcast_hops = skb->nh.ipv6h->hop_limit;
1512
1513         /* Clone native IPv6 options from listening socket (if any)
1514
1515            Yes, keeping reference count would be much more clever,
1516            but we make one more one thing there: reattach optmem
1517            to newsk.
1518          */
1519         if (opt) {
1520                 newnp->opt = ipv6_dup_options(newsk, opt);
1521                 if (opt != np->opt)
1522                         sock_kfree_s(sk, opt, opt->tot_len);
1523         }
1524
1525         newtp->ext_header_len = 0;
1526         if (newnp->opt)
1527                 newtp->ext_header_len = newnp->opt->opt_nflen +
1528                                         newnp->opt->opt_flen;
1529
1530         tcp_sync_mss(newsk, dst_mtu(dst));
1531         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1532         tcp_initialize_rcv_mss(newsk);
1533
1534         newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1535
1536         __tcp_v6_hash(newsk);
1537         tcp_inherit_port(sk, newsk);
1538
1539         return newsk;
1540
1541 out_overflow:
1542         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1543 out:
1544         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1545         if (opt && opt != np->opt)
1546                 sock_kfree_s(sk, opt, opt->tot_len);
1547         dst_release(dst);
1548         return NULL;
1549 }
1550
1551 static int tcp_v6_checksum_init(struct sk_buff *skb)
1552 {
1553         if (skb->ip_summed == CHECKSUM_HW) {
1554                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1555                 if (!tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1556                                   &skb->nh.ipv6h->daddr,skb->csum))
1557                         return 0;
1558                 LIMIT_NETDEBUG(printk(KERN_DEBUG "hw tcp v6 csum failed\n"));
1559         }
1560         if (skb->len <= 76) {
1561                 if (tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1562                                  &skb->nh.ipv6h->daddr,skb_checksum(skb, 0, skb->len, 0)))
1563                         return -1;
1564                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1565         } else {
1566                 skb->csum = ~tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1567                                           &skb->nh.ipv6h->daddr,0);
1568         }
1569         return 0;
1570 }
1571
1572 /* The socket must have it's spinlock held when we get
1573  * here.
1574  *
1575  * We have a potential double-lock case here, so even when
1576  * doing backlog processing we use the BH locking scheme.
1577  * This is because we cannot sleep with the original spinlock
1578  * held.
1579  */
1580 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1581 {
1582         struct ipv6_pinfo *np = inet6_sk(sk);
1583         struct tcp_sock *tp;
1584         struct sk_buff *opt_skb = NULL;
1585
1586         /* Imagine: socket is IPv6. IPv4 packet arrives,
1587            goes to IPv4 receive handler and backlogged.
1588            From backlog it always goes here. Kerboom...
1589            Fortunately, tcp_rcv_established and rcv_established
1590            handle them correctly, but it is not case with
1591            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1592          */
1593
1594         if (skb->protocol == htons(ETH_P_IP))
1595                 return tcp_v4_do_rcv(sk, skb);
1596
1597         if (sk_filter(sk, skb, 0))
1598                 goto discard;
1599
1600         /*
1601          *      socket locking is here for SMP purposes as backlog rcv
1602          *      is currently called with bh processing disabled.
1603          */
1604
1605         /* Do Stevens' IPV6_PKTOPTIONS.
1606
1607            Yes, guys, it is the only place in our code, where we
1608            may make it not affecting IPv4.
1609            The rest of code is protocol independent,
1610            and I do not like idea to uglify IPv4.
1611
1612            Actually, all the idea behind IPV6_PKTOPTIONS
1613            looks not very well thought. For now we latch
1614            options, received in the last packet, enqueued
1615            by tcp. Feel free to propose better solution.
1616                                                --ANK (980728)
1617          */
1618         if (np->rxopt.all)
1619                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1620
1621         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1622                 TCP_CHECK_TIMER(sk);
1623                 if (tcp_rcv_established(sk, skb, skb->h.th, skb->len))
1624                         goto reset;
1625                 TCP_CHECK_TIMER(sk);
1626                 if (opt_skb)
1627                         goto ipv6_pktoptions;
1628                 return 0;
1629         }
1630
1631         if (skb->len < (skb->h.th->doff<<2) || tcp_checksum_complete(skb))
1632                 goto csum_err;
1633
1634         if (sk->sk_state == TCP_LISTEN) {
1635                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1636                 if (!nsk)
1637                         goto discard;
1638
1639                 /*
1640                  * Queue it on the new socket if the new socket is active,
1641                  * otherwise we just shortcircuit this and continue with
1642                  * the new socket..
1643                  */
1644                 if(nsk != sk) {
1645                         if (tcp_child_process(sk, nsk, skb))
1646                                 goto reset;
1647                         if (opt_skb)
1648                                 __kfree_skb(opt_skb);
1649                         return 0;
1650                 }
1651         }
1652
1653         TCP_CHECK_TIMER(sk);
1654         if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len))
1655                 goto reset;
1656         TCP_CHECK_TIMER(sk);
1657         if (opt_skb)
1658                 goto ipv6_pktoptions;
1659         return 0;
1660
1661 reset:
1662         tcp_v6_send_reset(skb);
1663 discard:
1664         if (opt_skb)
1665                 __kfree_skb(opt_skb);
1666         kfree_skb(skb);
1667         return 0;
1668 csum_err:
1669         TCP_INC_STATS_BH(TCP_MIB_INERRS);
1670         goto discard;
1671
1672
1673 ipv6_pktoptions:
1674         /* Do you ask, what is it?
1675
1676            1. skb was enqueued by tcp.
1677            2. skb is added to tail of read queue, rather than out of order.
1678            3. socket is not in passive state.
1679            4. Finally, it really contains options, which user wants to receive.
1680          */
1681         tp = tcp_sk(sk);
1682         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1683             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1684                 if (np->rxopt.bits.rxinfo)
1685                         np->mcast_oif = tcp_v6_iif(opt_skb);
1686                 if (np->rxopt.bits.rxhlim)
1687                         np->mcast_hops = opt_skb->nh.ipv6h->hop_limit;
1688                 if (ipv6_opt_accepted(sk, opt_skb)) {
1689                         skb_set_owner_r(opt_skb, sk);
1690                         opt_skb = xchg(&np->pktoptions, opt_skb);
1691                 } else {
1692                         __kfree_skb(opt_skb);
1693                         opt_skb = xchg(&np->pktoptions, NULL);
1694                 }
1695         }
1696
1697         if (opt_skb)
1698                 kfree_skb(opt_skb);
1699         return 0;
1700 }
1701
1702 static int tcp_v6_rcv(struct sk_buff **pskb, unsigned int *nhoffp)
1703 {
1704         struct sk_buff *skb = *pskb;
1705         struct tcphdr *th;
1706         struct sock *sk;
1707         int ret;
1708
1709         if (skb->pkt_type != PACKET_HOST)
1710                 goto discard_it;
1711
1712         /*
1713          *      Count it even if it's bad.
1714          */
1715         TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1716
1717         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1718                 goto discard_it;
1719
1720         th = skb->h.th;
1721
1722         if (th->doff < sizeof(struct tcphdr)/4)
1723                 goto bad_packet;
1724         if (!pskb_may_pull(skb, th->doff*4))
1725                 goto discard_it;
1726
1727         if ((skb->ip_summed != CHECKSUM_UNNECESSARY &&
1728              tcp_v6_checksum_init(skb) < 0))
1729                 goto bad_packet;
1730
1731         th = skb->h.th;
1732         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1733         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1734                                     skb->len - th->doff*4);
1735         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1736         TCP_SKB_CB(skb)->when = 0;
1737         TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(skb->nh.ipv6h);
1738         TCP_SKB_CB(skb)->sacked = 0;
1739
1740         sk = __tcp_v6_lookup(&skb->nh.ipv6h->saddr, th->source,
1741                              &skb->nh.ipv6h->daddr, ntohs(th->dest), tcp_v6_iif(skb));
1742
1743         if (!sk)
1744                 goto no_tcp_socket;
1745
1746 process:
1747         if (sk->sk_state == TCP_TIME_WAIT)
1748                 goto do_time_wait;
1749
1750         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1751                 goto discard_and_relse;
1752
1753         if (sk_filter(sk, skb, 0))
1754                 goto discard_and_relse;
1755
1756         skb->dev = NULL;
1757
1758         bh_lock_sock(sk);
1759         ret = 0;
1760         if (!sock_owned_by_user(sk)) {
1761                 if (!tcp_prequeue(sk, skb))
1762                         ret = tcp_v6_do_rcv(sk, skb);
1763         } else
1764                 sk_add_backlog(sk, skb);
1765         bh_unlock_sock(sk);
1766
1767         sock_put(sk);
1768         return ret ? -1 : 0;
1769
1770 no_tcp_socket:
1771         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1772                 goto discard_it;
1773
1774         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1775 bad_packet:
1776                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1777         } else {
1778                 tcp_v6_send_reset(skb);
1779         }
1780
1781 discard_it:
1782
1783         /*
1784          *      Discard frame
1785          */
1786
1787         kfree_skb(skb);
1788         return 0;
1789
1790 discard_and_relse:
1791         sock_put(sk);
1792         goto discard_it;
1793
1794 do_time_wait:
1795         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1796                 tcp_tw_put((struct tcp_tw_bucket *) sk);
1797                 goto discard_it;
1798         }
1799
1800         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1801                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1802                 tcp_tw_put((struct tcp_tw_bucket *) sk);
1803                 goto discard_it;
1804         }
1805
1806         switch(tcp_timewait_state_process((struct tcp_tw_bucket *)sk,
1807                                           skb, th, skb->len)) {
1808         case TCP_TW_SYN:
1809         {
1810                 struct sock *sk2;
1811
1812                 sk2 = tcp_v6_lookup_listener(&skb->nh.ipv6h->daddr, ntohs(th->dest), tcp_v6_iif(skb));
1813                 if (sk2 != NULL) {
1814                         tcp_tw_deschedule((struct tcp_tw_bucket *)sk);
1815                         tcp_tw_put((struct tcp_tw_bucket *)sk);
1816                         sk = sk2;
1817                         goto process;
1818                 }
1819                 /* Fall through to ACK */
1820         }
1821         case TCP_TW_ACK:
1822                 tcp_v6_timewait_ack(sk, skb);
1823                 break;
1824         case TCP_TW_RST:
1825                 goto no_tcp_socket;
1826         case TCP_TW_SUCCESS:;
1827         }
1828         goto discard_it;
1829 }
1830
1831 static int tcp_v6_rebuild_header(struct sock *sk)
1832 {
1833         int err;
1834         struct dst_entry *dst;
1835         struct ipv6_pinfo *np = inet6_sk(sk);
1836
1837         dst = __sk_dst_check(sk, np->dst_cookie);
1838
1839         if (dst == NULL) {
1840                 struct inet_sock *inet = inet_sk(sk);
1841                 struct in6_addr *final_p = NULL, final;
1842                 struct flowi fl;
1843
1844                 memset(&fl, 0, sizeof(fl));
1845                 fl.proto = IPPROTO_TCP;
1846                 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1847                 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1848                 fl.fl6_flowlabel = np->flow_label;
1849                 fl.oif = sk->sk_bound_dev_if;
1850                 fl.fl_ip_dport = inet->dport;
1851                 fl.fl_ip_sport = inet->sport;
1852
1853                 if (np->opt && np->opt->srcrt) {
1854                         struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1855                         ipv6_addr_copy(&final, &fl.fl6_dst);
1856                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1857                         final_p = &final;
1858                 }
1859
1860                 err = ip6_dst_lookup(sk, &dst, &fl);
1861                 if (err) {
1862                         sk->sk_route_caps = 0;
1863                         return err;
1864                 }
1865                 if (final_p)
1866                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1867
1868                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
1869                         sk->sk_err_soft = -err;
1870                         dst_release(dst);
1871                         return err;
1872                 }
1873
1874                 ip6_dst_store(sk, dst, NULL);
1875                 sk->sk_route_caps = dst->dev->features &
1876                         ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1877         }
1878
1879         return 0;
1880 }
1881
1882 static int tcp_v6_xmit(struct sk_buff *skb, int ipfragok)
1883 {
1884         struct sock *sk = skb->sk;
1885         struct inet_sock *inet = inet_sk(sk);
1886         struct ipv6_pinfo *np = inet6_sk(sk);
1887         struct flowi fl;
1888         struct dst_entry *dst;
1889         struct in6_addr *final_p = NULL, final;
1890
1891         memset(&fl, 0, sizeof(fl));
1892         fl.proto = IPPROTO_TCP;
1893         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1894         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1895         fl.fl6_flowlabel = np->flow_label;
1896         IP6_ECN_flow_xmit(sk, fl.fl6_flowlabel);
1897         fl.oif = sk->sk_bound_dev_if;
1898         fl.fl_ip_sport = inet->sport;
1899         fl.fl_ip_dport = inet->dport;
1900
1901         if (np->opt && np->opt->srcrt) {
1902                 struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1903                 ipv6_addr_copy(&final, &fl.fl6_dst);
1904                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1905                 final_p = &final;
1906         }
1907
1908         dst = __sk_dst_check(sk, np->dst_cookie);
1909
1910         if (dst == NULL) {
1911                 int err = ip6_dst_lookup(sk, &dst, &fl);
1912
1913                 if (err) {
1914                         sk->sk_err_soft = -err;
1915                         return err;
1916                 }
1917
1918                 if (final_p)
1919                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1920
1921                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
1922                         sk->sk_route_caps = 0;
1923                         dst_release(dst);
1924                         return err;
1925                 }
1926
1927                 ip6_dst_store(sk, dst, NULL);
1928                 sk->sk_route_caps = dst->dev->features &
1929                         ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1930         }
1931
1932         skb->dst = dst_clone(dst);
1933
1934         /* Restore final destination back after routing done */
1935         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1936
1937         return ip6_xmit(sk, skb, &fl, np->opt, 0);
1938 }
1939
1940 static void v6_addr2sockaddr(struct sock *sk, struct sockaddr * uaddr)
1941 {
1942         struct ipv6_pinfo *np = inet6_sk(sk);
1943         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) uaddr;
1944
1945         sin6->sin6_family = AF_INET6;
1946         ipv6_addr_copy(&sin6->sin6_addr, &np->daddr);
1947         sin6->sin6_port = inet_sk(sk)->dport;
1948         /* We do not store received flowlabel for TCP */
1949         sin6->sin6_flowinfo = 0;
1950         sin6->sin6_scope_id = 0;
1951         if (sk->sk_bound_dev_if &&
1952             ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
1953                 sin6->sin6_scope_id = sk->sk_bound_dev_if;
1954 }
1955
1956 static int tcp_v6_remember_stamp(struct sock *sk)
1957 {
1958         /* Alas, not yet... */
1959         return 0;
1960 }
1961
1962 static struct tcp_func ipv6_specific = {
1963         .queue_xmit     =       tcp_v6_xmit,
1964         .send_check     =       tcp_v6_send_check,
1965         .rebuild_header =       tcp_v6_rebuild_header,
1966         .conn_request   =       tcp_v6_conn_request,
1967         .syn_recv_sock  =       tcp_v6_syn_recv_sock,
1968         .remember_stamp =       tcp_v6_remember_stamp,
1969         .net_header_len =       sizeof(struct ipv6hdr),
1970
1971         .setsockopt     =       ipv6_setsockopt,
1972         .getsockopt     =       ipv6_getsockopt,
1973         .addr2sockaddr  =       v6_addr2sockaddr,
1974         .sockaddr_len   =       sizeof(struct sockaddr_in6)
1975 };
1976
1977 /*
1978  *      TCP over IPv4 via INET6 API
1979  */
1980
1981 static struct tcp_func ipv6_mapped = {
1982         .queue_xmit     =       ip_queue_xmit,
1983         .send_check     =       tcp_v4_send_check,
1984         .rebuild_header =       tcp_v4_rebuild_header,
1985         .conn_request   =       tcp_v6_conn_request,
1986         .syn_recv_sock  =       tcp_v6_syn_recv_sock,
1987         .remember_stamp =       tcp_v4_remember_stamp,
1988         .net_header_len =       sizeof(struct iphdr),
1989
1990         .setsockopt     =       ipv6_setsockopt,
1991         .getsockopt     =       ipv6_getsockopt,
1992         .addr2sockaddr  =       v6_addr2sockaddr,
1993         .sockaddr_len   =       sizeof(struct sockaddr_in6)
1994 };
1995
1996
1997
1998 /* NOTE: A lot of things set to zero explicitly by call to
1999  *       sk_alloc() so need not be done here.
2000  */
2001 static int tcp_v6_init_sock(struct sock *sk)
2002 {
2003         struct tcp_sock *tp = tcp_sk(sk);
2004
2005         skb_queue_head_init(&tp->out_of_order_queue);
2006         tcp_init_xmit_timers(sk);
2007         tcp_prequeue_init(tp);
2008
2009         tp->rto  = TCP_TIMEOUT_INIT;
2010         tp->mdev = TCP_TIMEOUT_INIT;
2011
2012         /* So many TCP implementations out there (incorrectly) count the
2013          * initial SYN frame in their delayed-ACK and congestion control
2014          * algorithms that we must have the following bandaid to talk
2015          * efficiently to them.  -DaveM
2016          */
2017         tp->snd_cwnd = 2;
2018
2019         /* See draft-stevens-tcpca-spec-01 for discussion of the
2020          * initialization of these values.
2021          */
2022         tp->snd_ssthresh = 0x7fffffff;
2023         tp->snd_cwnd_clamp = ~0;
2024         tp->mss_cache_std = tp->mss_cache = 536;
2025
2026         tp->reordering = sysctl_tcp_reordering;
2027
2028         sk->sk_state = TCP_CLOSE;
2029
2030         tp->af_specific = &ipv6_specific;
2031
2032         sk->sk_write_space = sk_stream_write_space;
2033         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2034
2035         sk->sk_sndbuf = sysctl_tcp_wmem[1];
2036         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
2037
2038         atomic_inc(&tcp_sockets_allocated);
2039
2040         return 0;
2041 }
2042
2043 static int tcp_v6_destroy_sock(struct sock *sk)
2044 {
2045         extern int tcp_v4_destroy_sock(struct sock *sk);
2046
2047         tcp_v4_destroy_sock(sk);
2048         return inet6_destroy_sock(sk);
2049 }
2050
2051 /* Proc filesystem TCPv6 sock list dumping. */
2052 static void get_openreq6(struct seq_file *seq,
2053                          struct sock *sk, struct open_request *req, int i, int uid)
2054 {
2055         struct in6_addr *dest, *src;
2056         int ttd = req->expires - jiffies;
2057
2058         if (ttd < 0)
2059                 ttd = 0;
2060
2061         src = &req->af.v6_req.loc_addr;
2062         dest = &req->af.v6_req.rmt_addr;
2063         seq_printf(seq,
2064                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2065                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2066                    i,
2067                    src->s6_addr32[0], src->s6_addr32[1],
2068                    src->s6_addr32[2], src->s6_addr32[3],
2069                    ntohs(inet_sk(sk)->sport),
2070                    dest->s6_addr32[0], dest->s6_addr32[1],
2071                    dest->s6_addr32[2], dest->s6_addr32[3],
2072                    ntohs(req->rmt_port),
2073                    TCP_SYN_RECV,
2074                    0,0, /* could print option size, but that is af dependent. */
2075                    1,   /* timers active (only the expire timer) */
2076                    jiffies_to_clock_t(ttd),
2077                    req->retrans,
2078                    uid,
2079                    0,  /* non standard timer */
2080                    0, /* open_requests have no inode */
2081                    0, req);
2082 }
2083
2084 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2085 {
2086         struct in6_addr *dest, *src;
2087         __u16 destp, srcp;
2088         int timer_active;
2089         unsigned long timer_expires;
2090         struct inet_sock *inet = inet_sk(sp);
2091         struct tcp_sock *tp = tcp_sk(sp);
2092         struct ipv6_pinfo *np = inet6_sk(sp);
2093
2094         dest  = &np->daddr;
2095         src   = &np->rcv_saddr;
2096         destp = ntohs(inet->dport);
2097         srcp  = ntohs(inet->sport);
2098         if (tp->pending == TCP_TIME_RETRANS) {
2099                 timer_active    = 1;
2100                 timer_expires   = tp->timeout;
2101         } else if (tp->pending == TCP_TIME_PROBE0) {
2102                 timer_active    = 4;
2103                 timer_expires   = tp->timeout;
2104         } else if (timer_pending(&sp->sk_timer)) {
2105                 timer_active    = 2;
2106                 timer_expires   = sp->sk_timer.expires;
2107         } else {
2108                 timer_active    = 0;
2109                 timer_expires = jiffies;
2110         }
2111
2112         seq_printf(seq,
2113                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2114                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
2115                    i,
2116                    src->s6_addr32[0], src->s6_addr32[1],
2117                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2118                    dest->s6_addr32[0], dest->s6_addr32[1],
2119                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2120                    sp->sk_state,
2121                    tp->write_seq-tp->snd_una, tp->rcv_nxt-tp->copied_seq,
2122                    timer_active,
2123                    jiffies_to_clock_t(timer_expires - jiffies),
2124                    tp->retransmits,
2125                    sock_i_uid(sp),
2126                    tp->probes_out,
2127                    sock_i_ino(sp),
2128                    atomic_read(&sp->sk_refcnt), sp,
2129                    tp->rto, tp->ack.ato, (tp->ack.quick<<1)|tp->ack.pingpong,
2130                    tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2131                    );
2132 }
2133
2134 static void get_timewait6_sock(struct seq_file *seq,
2135                                struct tcp_tw_bucket *tw, int i)
2136 {
2137         struct in6_addr *dest, *src;
2138         __u16 destp, srcp;
2139         int ttd = tw->tw_ttd - jiffies;
2140
2141         if (ttd < 0)
2142                 ttd = 0;
2143
2144         dest  = &tw->tw_v6_daddr;
2145         src   = &tw->tw_v6_rcv_saddr;
2146         destp = ntohs(tw->tw_dport);
2147         srcp  = ntohs(tw->tw_sport);
2148
2149         seq_printf(seq,
2150                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2151                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2152                    i,
2153                    src->s6_addr32[0], src->s6_addr32[1],
2154                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2155                    dest->s6_addr32[0], dest->s6_addr32[1],
2156                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2157                    tw->tw_substate, 0, 0,
2158                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2159                    atomic_read(&tw->tw_refcnt), tw);
2160 }
2161
2162 #ifdef CONFIG_PROC_FS
2163 static int tcp6_seq_show(struct seq_file *seq, void *v)
2164 {
2165         struct tcp_iter_state *st;
2166
2167         if (v == SEQ_START_TOKEN) {
2168                 seq_puts(seq,
2169                          "  sl  "
2170                          "local_address                         "
2171                          "remote_address                        "
2172                          "st tx_queue rx_queue tr tm->when retrnsmt"
2173                          "   uid  timeout inode\n");
2174                 goto out;
2175         }
2176         st = seq->private;
2177
2178         switch (st->state) {
2179         case TCP_SEQ_STATE_LISTENING:
2180         case TCP_SEQ_STATE_ESTABLISHED:
2181                 get_tcp6_sock(seq, v, st->num);
2182                 break;
2183         case TCP_SEQ_STATE_OPENREQ:
2184                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2185                 break;
2186         case TCP_SEQ_STATE_TIME_WAIT:
2187                 get_timewait6_sock(seq, v, st->num);
2188                 break;
2189         }
2190 out:
2191         return 0;
2192 }
2193
2194 static struct file_operations tcp6_seq_fops;
2195 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2196         .owner          = THIS_MODULE,
2197         .name           = "tcp6",
2198         .family         = AF_INET6,
2199         .seq_show       = tcp6_seq_show,
2200         .seq_fops       = &tcp6_seq_fops,
2201 };
2202
2203 int __init tcp6_proc_init(void)
2204 {
2205         return tcp_proc_register(&tcp6_seq_afinfo);
2206 }
2207
2208 void tcp6_proc_exit(void)
2209 {
2210         tcp_proc_unregister(&tcp6_seq_afinfo);
2211 }
2212 #endif
2213
2214 struct proto tcpv6_prot = {
2215         .name                   = "TCPv6",
2216         .owner                  = THIS_MODULE,
2217         .close                  = tcp_close,
2218         .connect                = tcp_v6_connect,
2219         .disconnect             = tcp_disconnect,
2220         .accept                 = tcp_accept,
2221         .ioctl                  = tcp_ioctl,
2222         .init                   = tcp_v6_init_sock,
2223         .destroy                = tcp_v6_destroy_sock,
2224         .shutdown               = tcp_shutdown,
2225         .setsockopt             = tcp_setsockopt,
2226         .getsockopt             = tcp_getsockopt,
2227         .sendmsg                = tcp_sendmsg,
2228         .recvmsg                = tcp_recvmsg,
2229         .backlog_rcv            = tcp_v6_do_rcv,
2230         .hash                   = tcp_v6_hash,
2231         .unhash                 = tcp_unhash,
2232         .get_port               = tcp_v6_get_port,
2233         .enter_memory_pressure  = tcp_enter_memory_pressure,
2234         .sockets_allocated      = &tcp_sockets_allocated,
2235         .memory_allocated       = &tcp_memory_allocated,
2236         .memory_pressure        = &tcp_memory_pressure,
2237         .sysctl_mem             = sysctl_tcp_mem,
2238         .sysctl_wmem            = sysctl_tcp_wmem,
2239         .sysctl_rmem            = sysctl_tcp_rmem,
2240         .max_header             = MAX_TCP_HEADER,
2241         .obj_size               = sizeof(struct tcp6_sock),
2242 };
2243
2244 static struct inet6_protocol tcpv6_protocol = {
2245         .handler        =       tcp_v6_rcv,
2246         .err_handler    =       tcp_v6_err,
2247         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2248 };
2249
2250 extern struct proto_ops inet6_stream_ops;
2251
2252 static struct inet_protosw tcpv6_protosw = {
2253         .type           =       SOCK_STREAM,
2254         .protocol       =       IPPROTO_TCP,
2255         .prot           =       &tcpv6_prot,
2256         .ops            =       &inet6_stream_ops,
2257         .capability     =       -1,
2258         .no_check       =       0,
2259         .flags          =       INET_PROTOSW_PERMANENT,
2260 };
2261
2262 void __init tcpv6_init(void)
2263 {
2264         /* register inet6 protocol */
2265         if (inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP) < 0)
2266                 printk(KERN_ERR "tcpv6_init: Could not register protocol\n");
2267         inet6_register_protosw(&tcpv6_protosw);
2268 }