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                 rover = tcp_port_rover;
 143                 do {    rover++;
 144                         if ((rover < low) || (rover > high))
 145                                 rover = low;
 146                         head = &tcp_bhash[tcp_bhashfn(rover)];
 147                         spin_lock(&head->lock);
 148                         tb_for_each(tb, node, &head->chain)
 149                                 if (tb->port == rover)
 150                                         goto next;
 151                         break;
 152                 next:
 153                         spin_unlock(&head->lock);
 154                 } while (--remaining > 0);
 155                 tcp_port_rover = rover;
 156                 spin_unlock(&tcp_portalloc_lock);
 157
 158                 /* Exhausted local port range during search? */
 159                 ret = 1;
 160                 if (remaining <= 0)
 161                         goto fail;
 162
 163                 /* OK, here is the one we will use. */
 164                 snum = rover;
 165         } else {
 166                 head = &tcp_bhash[tcp_bhashfn(snum)];
 167                 spin_lock(&head->lock);
 168                 tb_for_each(tb, node, &head->chain)
 169                         if (tb->port == snum)
 170                                 goto tb_found;
 171         }
 172         tb = NULL;
 173         goto tb_not_found;
 174 tb_found:
 175         if (tb && !hlist_empty(&tb->owners)) {
 176                 if (tb->fastreuse > 0 && sk->sk_reuse &&
 177                     sk->sk_state != TCP_LISTEN) {
 178                         goto success;
 179                 } else {
 180                         ret = 1;
 181                         if (tcp_v6_bind_conflict(sk, tb))
 182                                 goto fail_unlock;
 183                 }
 184         }
 185 tb_not_found:
 186         ret = 1;
 187         if (!tb && (tb = tcp_bucket_create(head, snum)) == NULL)
 188                 goto fail_unlock;
 189         if (hlist_empty(&tb->owners)) {
 190                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 191                         tb->fastreuse = 1;
 192                 else
 193                         tb->fastreuse = 0;
 194         } else if (tb->fastreuse &&
 195                    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 196                 tb->fastreuse = 0;
 197
 198 success:
 199         if (!tcp_sk(sk)->bind_hash)
 200                 tcp_bind_hash(sk, tb, snum);
 201         BUG_TRAP(tcp_sk(sk)->bind_hash == tb);
 202         ret = 0;
 203
 204 fail_unlock:
 205         spin_unlock(&head->lock);
 206 fail:
 207         local_bh_enable();
 208         return ret;
 209 }
 210
 211 static __inline__ void __tcp_v6_hash(struct sock *sk)
 212 {
 213         struct hlist_head *list;
 214         rwlock_t *lock;
 215
 216         BUG_TRAP(sk_unhashed(sk));
 217
 218         if (sk->sk_state == TCP_LISTEN) {
 219                 list = &tcp_listening_hash[tcp_sk_listen_hashfn(sk)];
 220                 lock = &tcp_lhash_lock;
 221                 tcp_listen_wlock();
 222         } else {
 223                 sk->sk_hashent = tcp_v6_sk_hashfn(sk);
 224                 list = &tcp_ehash[sk->sk_hashent].chain;
 225                 lock = &tcp_ehash[sk->sk_hashent].lock;
 226                 write_lock(lock);
 227         }
 228
 229         __sk_add_node(sk, list);
 230         sock_prot_inc_use(sk->sk_prot);
 231         write_unlock(lock);
 232 }
 233
 234
 235 static void tcp_v6_hash(struct sock *sk)
 236 {
 237         if (sk->sk_state != TCP_CLOSE) {
 238                 struct tcp_sock *tp = tcp_sk(sk);
 239
 240                 if (tp->af_specific == &ipv6_mapped) {
 241                         tcp_prot.hash(sk);
 242                         return;
 243                 }
 244                 local_bh_disable();
 245                 __tcp_v6_hash(sk);
 246                 local_bh_enable();
 247         }
 248 }
 249
 250 static struct sock *tcp_v6_lookup_listener(struct in6_addr *daddr, unsigned short hnum, int dif)
 251 {
 252         struct sock *sk;
 253         struct hlist_node *node;
 254         struct sock *result = NULL;
 255         int score, hiscore;
 256
 257         hiscore=0;
 258         read_lock(&tcp_lhash_lock);
 259         sk_for_each(sk, node, &tcp_listening_hash[tcp_lhashfn(hnum)]) {
 260                 if (inet_sk(sk)->num == hnum && sk->sk_family == PF_INET6) {
 261                         struct ipv6_pinfo *np = inet6_sk(sk);
 262
 263                         score = 1;
 264                         if (!ipv6_addr_any(&np->rcv_saddr)) {
 265                                 if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
 266                                         continue;
 267                                 score++;
 268                         }
 269                         if (sk->sk_bound_dev_if) {
 270                                 if (sk->sk_bound_dev_if != dif)
 271                                         continue;
 272                                 score++;
 273                         }
 274                         if (score == 3) {
 275                                 result = sk;
 276                                 break;
 277                         }
 278                         if (score > hiscore) {
 279                                 hiscore = score;
 280                                 result = sk;
 281                         }
 282                 }
 283         }
 284         if (result)
 285                 sock_hold(result);
 286         read_unlock(&tcp_lhash_lock);
 287         return result;
 288 }
 289
 290 /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
 291  * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
 292  *
 293  * The sockhash lock must be held as a reader here.
 294  */
 295
 296 static inline struct sock *__tcp_v6_lookup_established(struct in6_addr *saddr, u16 sport,
 297                                                        struct in6_addr *daddr, u16 hnum,
 298                                                        int dif)
 299 {
 300         struct tcp_ehash_bucket *head;
 301         struct sock *sk;
 302         struct hlist_node *node;
 303         __u32 ports = TCP_COMBINED_PORTS(sport, hnum);
 304         int hash;
 305
 306         /* Optimize here for direct hit, only listening connections can
 307          * have wildcards anyways.
 308          */
 309         hash = tcp_v6_hashfn(daddr, hnum, saddr, sport);
 310         head = &tcp_ehash[hash];
 311         read_lock(&head->lock);
 312         sk_for_each(sk, node, &head->chain) {
 313                 /* For IPV6 do the cheaper port and family tests first. */
 314                 if(TCP_IPV6_MATCH(sk, saddr, daddr, ports, dif))
 315                         goto hit; /* You sunk my battleship! */
 316         }
 317         /* Must check for a TIME_WAIT'er before going to listener hash. */
 318         sk_for_each(sk, node, &(head + tcp_ehash_size)->chain) {
 319                 /* FIXME: acme: check this... */
 320                 struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk;
 321
 322                 if(*((__u32 *)&(tw->tw_dport))  == ports        &&
 323                    sk->sk_family                == PF_INET6) {
 324                         if(ipv6_addr_equal(&tw->tw_v6_daddr, saddr)     &&
 325                            ipv6_addr_equal(&tw->tw_v6_rcv_saddr, daddr) &&
 326                            (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif))
 327                                 goto hit;
 328                 }
 329         }
 330         read_unlock(&head->lock);
 331         return NULL;
 332
 333 hit:
 334         sock_hold(sk);
 335         read_unlock(&head->lock);
 336         return sk;
 337 }
 338
 339
 340 static inline struct sock *__tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
 341                                            struct in6_addr *daddr, u16 hnum,
 342                                            int dif)
 343 {
 344         struct sock *sk;
 345
 346         sk = __tcp_v6_lookup_established(saddr, sport, daddr, hnum, dif);
 347
 348         if (sk)
 349                 return sk;
 350
 351         return tcp_v6_lookup_listener(daddr, hnum, dif);
 352 }
 353
 354 inline struct sock *tcp_v6_lookup(struct in6_addr *saddr, u16 sport,
 355                                   struct in6_addr *daddr, u16 dport,
 356                                   int dif)
 357 {
 358         struct sock *sk;
 359
 360         local_bh_disable();
 361         sk = __tcp_v6_lookup(saddr, sport, daddr, ntohs(dport), dif);
 362         local_bh_enable();
 363
 364         return sk;
 365 }
 366
 367 EXPORT_SYMBOL_GPL(tcp_v6_lookup);
 368
 369
 370 /*
 371  * Open request hash tables.
 372  */
 373
 374 static u32 tcp_v6_synq_hash(struct in6_addr *raddr, u16 rport, u32 rnd)
 375 {
 376         u32 a, b, c;
 377
 378         a = raddr->s6_addr32[0];
 379         b = raddr->s6_addr32[1];
 380         c = raddr->s6_addr32[2];
 381
 382         a += JHASH_GOLDEN_RATIO;
 383         b += JHASH_GOLDEN_RATIO;
 384         c += rnd;
 385         __jhash_mix(a, b, c);
 386
 387         a += raddr->s6_addr32[3];
 388         b += (u32) rport;
 389         __jhash_mix(a, b, c);
 390
 391         return c & (TCP_SYNQ_HSIZE - 1);
 392 }
 393
 394 static struct open_request *tcp_v6_search_req(struct tcp_sock *tp,
 395                                               struct open_request ***prevp,
 396                                               __u16 rport,
 397                                               struct in6_addr *raddr,
 398                                               struct in6_addr *laddr,
 399                                               int iif)
 400 {
 401         struct tcp_listen_opt *lopt = tp->listen_opt;
 402         struct open_request *req, **prev;
 403
 404         for (prev = &lopt->syn_table[tcp_v6_synq_hash(raddr, rport, lopt->hash_rnd)];
 405              (req = *prev) != NULL;
 406              prev = &req->dl_next) {
 407                 if (req->rmt_port == rport &&
 408                     req->class->family == AF_INET6 &&
 409                     ipv6_addr_equal(&req->af.v6_req.rmt_addr, raddr) &&
 410                     ipv6_addr_equal(&req->af.v6_req.loc_addr, laddr) &&
 411                     (!req->af.v6_req.iif || req->af.v6_req.iif == iif)) {
 412                         BUG_TRAP(req->sk == NULL);
 413                         *prevp = prev;
 414                         return req;
 415                 }
 416         }
 417
 418         return NULL;
 419 }
 420
 421 static __inline__ u16 tcp_v6_check(struct tcphdr *th, int len,
 422                                    struct in6_addr *saddr,
 423                                    struct in6_addr *daddr,
 424                                    unsigned long base)
 425 {
 426         return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
 427 }
 428
 429 static __u32 tcp_v6_init_sequence(struct sock *sk, struct sk_buff *skb)
 430 {
 431         if (skb->protocol == htons(ETH_P_IPV6)) {
 432                 return secure_tcpv6_sequence_number(skb->nh.ipv6h->daddr.s6_addr32,
 433                                                     skb->nh.ipv6h->saddr.s6_addr32,
 434                                                     skb->h.th->dest,
 435                                                     skb->h.th->source);
 436         } else {
 437                 return secure_tcp_sequence_number(skb->nh.iph->daddr,
 438                                                   skb->nh.iph->saddr,
 439                                                   skb->h.th->dest,
 440                                                   skb->h.th->source);
 441         }
 442 }
 443
 444 static int tcp_v6_check_established(struct sock *sk)
 445 {
 446         struct inet_sock *inet = inet_sk(sk);
 447         struct ipv6_pinfo *np = inet6_sk(sk);
 448         struct in6_addr *daddr = &np->rcv_saddr;
 449         struct in6_addr *saddr = &np->daddr;
 450         int dif = sk->sk_bound_dev_if;
 451         u32 ports = TCP_COMBINED_PORTS(inet->dport, inet->num);
 452         int hash = tcp_v6_hashfn(daddr, inet->num, saddr, inet->dport);
 453         struct tcp_ehash_bucket *head = &tcp_ehash[hash];
 454         struct sock *sk2;
 455         struct hlist_node *node;
 456         struct tcp_tw_bucket *tw;
 457
 458         write_lock_bh(&head->lock);
 459
 460         /* Check TIME-WAIT sockets first. */
 461         sk_for_each(sk2, node, &(head + tcp_ehash_size)->chain) {
 462                 tw = (struct tcp_tw_bucket*)sk2;
 463
 464                 if(*((__u32 *)&(tw->tw_dport))  == ports        &&
 465                    sk2->sk_family               == PF_INET6     &&
 466                    ipv6_addr_equal(&tw->tw_v6_daddr, saddr)     &&
 467                    ipv6_addr_equal(&tw->tw_v6_rcv_saddr, daddr) &&
 468                    sk2->sk_bound_dev_if == sk->sk_bound_dev_if) {
 469                         struct tcp_sock *tp = tcp_sk(sk);
 470
 471                         if (tw->tw_ts_recent_stamp) {
 472                                 /* See comment in tcp_ipv4.c */
 473                                 tp->write_seq = tw->tw_snd_nxt + 65535 + 2;
 474                                 if (!tp->write_seq)
 475                                         tp->write_seq = 1;
 476                                 tp->rx_opt.ts_recent = tw->tw_ts_recent;
 477                                 tp->rx_opt.ts_recent_stamp = tw->tw_ts_recent_stamp;
 478                                 sock_hold(sk2);
 479                                 goto unique;
 480                         } else
 481                                 goto not_unique;
 482                 }
 483         }
 484         tw = NULL;
 485
 486         /* And established part... */
 487         sk_for_each(sk2, node, &head->chain) {
 488                 if(TCP_IPV6_MATCH(sk2, saddr, daddr, ports, dif))
 489                         goto not_unique;
 490         }
 491
 492 unique:
 493         BUG_TRAP(sk_unhashed(sk));
 494         __sk_add_node(sk, &head->chain);
 495         sk->sk_hashent = hash;
 496         sock_prot_inc_use(sk->sk_prot);
 497         write_unlock_bh(&head->lock);
 498
 499         if (tw) {
 500                 /* Silly. Should hash-dance instead... */
 501                 local_bh_disable();
 502                 tcp_tw_deschedule(tw);
 503                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 504                 local_bh_enable();
 505
 506                 tcp_tw_put(tw);
 507         }
 508         return 0;
 509
 510 not_unique:
 511         write_unlock_bh(&head->lock);
 512         return -EADDRNOTAVAIL;
 513 }
 514
 515 static int tcp_v6_hash_connect(struct sock *sk)
 516 {
 517         struct tcp_bind_hashbucket *head;
 518         struct tcp_bind_bucket *tb;
 519
 520         /* XXX */
 521         if (inet_sk(sk)->num == 0) {
 522                 int err = tcp_v6_get_port(sk, inet_sk(sk)->num);
 523                 if (err)
 524                         return err;
 525                 inet_sk(sk)->sport = htons(inet_sk(sk)->num);
 526         }
 527
 528         head = &tcp_bhash[tcp_bhashfn(inet_sk(sk)->num)];
 529         tb = tb_head(head);
 530
 531         spin_lock_bh(&head->lock);
 532
 533         if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
 534                 __tcp_v6_hash(sk);
 535                 spin_unlock_bh(&head->lock);
 536                 return 0;
 537         } else {
 538                 spin_unlock_bh(&head->lock);
 539                 return tcp_v6_check_established(sk);
 540         }
 541 }
 542
 543 static __inline__ int tcp_v6_iif(struct sk_buff *skb)
 544 {
 545         return IP6CB(skb)->iif;
 546 }
 547
 548 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 549                           int addr_len)
 550 {
 551         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 552         struct inet_sock *inet = inet_sk(sk);
 553         struct ipv6_pinfo *np = inet6_sk(sk);
 554         struct tcp_sock *tp = tcp_sk(sk);
 555         struct in6_addr *saddr = NULL, *final_p = NULL, final;
 556         struct flowi fl;
 557         struct dst_entry *dst;
 558         int addr_type;
 559         int err;
 560
 561         if (addr_len < SIN6_LEN_RFC2133)
 562                 return -EINVAL;
 563
 564         if (usin->sin6_family != AF_INET6)
 565                 return(-EAFNOSUPPORT);
 566
 567         memset(&fl, 0, sizeof(fl));
 568
 569         if (np->sndflow) {
 570                 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 571                 IP6_ECN_flow_init(fl.fl6_flowlabel);
 572                 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
 573                         struct ip6_flowlabel *flowlabel;
 574                         flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
 575                         if (flowlabel == NULL)
 576                                 return -EINVAL;
 577                         ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
 578                         fl6_sock_release(flowlabel);
 579                 }
 580         }
 581
 582         /*
 583          *      connect() to INADDR_ANY means loopback (BSD'ism).
 584          */
 585
 586         if(ipv6_addr_any(&usin->sin6_addr))
 587                 usin->sin6_addr.s6_addr[15] = 0x1;
 588
 589         addr_type = ipv6_addr_type(&usin->sin6_addr);
 590
 591         if(addr_type & IPV6_ADDR_MULTICAST)
 592                 return -ENETUNREACH;
 593
 594         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 595                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 596                     usin->sin6_scope_id) {
 597                         /* If interface is set while binding, indices
 598                          * must coincide.
 599                          */
 600                         if (sk->sk_bound_dev_if &&
 601                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 602                                 return -EINVAL;
 603
 604                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 605                 }
 606
 607                 /* Connect to link-local address requires an interface */
 608                 if (!sk->sk_bound_dev_if)
 609                         return -EINVAL;
 610         }
 611
 612         if (tp->rx_opt.ts_recent_stamp &&
 613             !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
 614                 tp->rx_opt.ts_recent = 0;
 615                 tp->rx_opt.ts_recent_stamp = 0;
 616                 tp->write_seq = 0;
 617         }
 618
 619         ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
 620         np->flow_label = fl.fl6_flowlabel;
 621
 622         /*
 623          *      TCP over IPv4
 624          */
 625
 626         if (addr_type == IPV6_ADDR_MAPPED) {
 627                 u32 exthdrlen = tp->ext_header_len;
 628                 struct sockaddr_in sin;
 629
 630                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 631
 632                 if (__ipv6_only_sock(sk))
 633                         return -ENETUNREACH;
 634
 635                 sin.sin_family = AF_INET;
 636                 sin.sin_port = usin->sin6_port;
 637                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 638
 639                 tp->af_specific = &ipv6_mapped;
 640                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 641
 642                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 643
 644                 if (err) {
 645                         tp->ext_header_len = exthdrlen;
 646                         tp->af_specific = &ipv6_specific;
 647                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 648                         goto failure;
 649                 } else {
 650                         ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
 651                                       inet->saddr);
 652                         ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
 653                                       inet->rcv_saddr);
 654                 }
 655
 656                 return err;
 657         }
 658
 659         if (!ipv6_addr_any(&np->rcv_saddr))
 660                 saddr = &np->rcv_saddr;
 661
 662         fl.proto = IPPROTO_TCP;
 663         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 664         ipv6_addr_copy(&fl.fl6_src,
 665                        (saddr ? saddr : &np->saddr));
 666         fl.oif = sk->sk_bound_dev_if;
 667         fl.fl_ip_dport = usin->sin6_port;
 668         fl.fl_ip_sport = inet->sport;
 669
 670         if (np->opt && np->opt->srcrt) {
 671                 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
 672                 ipv6_addr_copy(&final, &fl.fl6_dst);
 673                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 674                 final_p = &final;
 675         }
 676
 677         err = ip6_dst_lookup(sk, &dst, &fl);
 678         if (err)
 679                 goto failure;
 680         if (final_p)
 681                 ipv6_addr_copy(&fl.fl6_dst, final_p);
 682
 683         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
 684                 dst_release(dst);
 685                 goto failure;
 686         }
 687
 688         if (saddr == NULL) {
 689                 saddr = &fl.fl6_src;
 690                 ipv6_addr_copy(&np->rcv_saddr, saddr);
 691         }
 692
 693         /* set the source address */
 694         ipv6_addr_copy(&np->saddr, saddr);
 695         inet->rcv_saddr = LOOPBACK4_IPV6;
 696
 697         ip6_dst_store(sk, dst, NULL);
 698         sk->sk_route_caps = dst->dev->features &
 699                 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
 700
 701         tp->ext_header_len = 0;
 702         if (np->opt)
 703                 tp->ext_header_len = np->opt->opt_flen + np->opt->opt_nflen;
 704         tp->ext2_header_len = dst->header_len;
 705
 706         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 707
 708         inet->dport = usin->sin6_port;
 709
 710         tcp_set_state(sk, TCP_SYN_SENT);
 711         err = tcp_v6_hash_connect(sk);
 712         if (err)
 713                 goto late_failure;
 714
 715         if (!tp->write_seq)
 716                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 717                                                              np->daddr.s6_addr32,
 718                                                              inet->sport,
 719                                                              inet->dport);
 720
 721         err = tcp_connect(sk);
 722         if (err)
 723                 goto late_failure;
 724
 725         return 0;
 726
 727 late_failure:
 728         tcp_set_state(sk, TCP_CLOSE);
 729         __sk_dst_reset(sk);
 730 failure:
 731         inet->dport = 0;
 732         sk->sk_route_caps = 0;
 733         return err;
 734 }
 735
 736 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 737                 int type, int code, int offset, __u32 info)
 738 {
 739         struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
 740         struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 741         struct ipv6_pinfo *np;
 742         struct sock *sk;
 743         int err;
 744         struct tcp_sock *tp;
 745         __u32 seq;
 746
 747         sk = tcp_v6_lookup(&hdr->daddr, th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
 748
 749         if (sk == NULL) {
 750                 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
 751                 return;
 752         }
 753
 754         if (sk->sk_state == TCP_TIME_WAIT) {
 755                 tcp_tw_put((struct tcp_tw_bucket*)sk);
 756                 return;
 757         }
 758
 759         bh_lock_sock(sk);
 760         if (sock_owned_by_user(sk))
 761                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 762
 763         if (sk->sk_state == TCP_CLOSE)
 764                 goto out;
 765
 766         tp = tcp_sk(sk);
 767         seq = ntohl(th->seq);
 768         if (sk->sk_state != TCP_LISTEN &&
 769             !between(seq, tp->snd_una, tp->snd_nxt)) {
 770                 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 771                 goto out;
 772         }
 773
 774         np = inet6_sk(sk);
 775
 776         if (type == ICMPV6_PKT_TOOBIG) {
 777                 struct dst_entry *dst = NULL;
 778
 779                 if (sock_owned_by_user(sk))
 780                         goto out;
 781                 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 782                         goto out;
 783
 784                 /* icmp should have updated the destination cache entry */
 785                 dst = __sk_dst_check(sk, np->dst_cookie);
 786
 787                 if (dst == NULL) {
 788                         struct inet_sock *inet = inet_sk(sk);
 789                         struct flowi fl;
 790
 791                         /* BUGGG_FUTURE: Again, it is not clear how
 792                            to handle rthdr case. Ignore this complexity
 793                            for now.
 794                          */
 795                         memset(&fl, 0, sizeof(fl));
 796                         fl.proto = IPPROTO_TCP;
 797                         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 798                         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
 799                         fl.oif = sk->sk_bound_dev_if;
 800                         fl.fl_ip_dport = inet->dport;
 801                         fl.fl_ip_sport = inet->sport;
 802
 803                         if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
 804                                 sk->sk_err_soft = -err;
 805                                 goto out;
 806                         }
 807
 808                         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
 809                                 sk->sk_err_soft = -err;
 810                                 goto out;
 811                         }
 812
 813                 } else
 814                         dst_hold(dst);
 815
 816                 if (tp->pmtu_cookie > dst_pmtu(dst)) {
 817                         tcp_sync_mss(sk, dst_pmtu(dst));
 818                         tcp_simple_retransmit(sk);
 819                 } /* else let the usual retransmit timer handle it */
 820                 dst_release(dst);
 821                 goto out;
 822         }
 823
 824         icmpv6_err_convert(type, code, &err);
 825
 826         /* Might be for an open_request */
 827         switch (sk->sk_state) {
 828                 struct open_request *req, **prev;
 829         case TCP_LISTEN:
 830                 if (sock_owned_by_user(sk))
 831                         goto out;
 832
 833                 req = tcp_v6_search_req(tp, &prev, th->dest, &hdr->daddr,
 834                                         &hdr->saddr, tcp_v6_iif(skb));
 835                 if (!req)
 836                         goto out;
 837
 838                 /* ICMPs are not backlogged, hence we cannot get
 839                  * an established socket here.
 840                  */
 841                 BUG_TRAP(req->sk == NULL);
 842
 843                 if (seq != req->snt_isn) {
 844                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 845                         goto out;
 846                 }
 847
 848                 tcp_synq_drop(sk, req, prev);
 849                 goto out;
 850
 851         case TCP_SYN_SENT:
 852         case TCP_SYN_RECV:  /* Cannot happen.
 853                                It can, it SYNs are crossed. --ANK */
 854                 if (!sock_owned_by_user(sk)) {
 855                         TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
 856                         sk->sk_err = err;
 857                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 858
 859                         tcp_done(sk);
 860                 } else
 861                         sk->sk_err_soft = err;
 862                 goto out;
 863         }
 864
 865         if (!sock_owned_by_user(sk) && np->recverr) {
 866                 sk->sk_err = err;
 867                 sk->sk_error_report(sk);
 868         } else
 869                 sk->sk_err_soft = err;
 870
 871 out:
 872         bh_unlock_sock(sk);
 873         sock_put(sk);
 874 }
 875
 876
 877 static int tcp_v6_send_synack(struct sock *sk, struct open_request *req,
 878                               struct dst_entry *dst)
 879 {
 880         struct ipv6_pinfo *np = inet6_sk(sk);
 881         struct sk_buff * skb;
 882         struct ipv6_txoptions *opt = NULL;
 883         struct in6_addr * final_p = NULL, final;
 884         struct flowi fl;
 885         int err = -1;
 886
 887         memset(&fl, 0, sizeof(fl));
 888         fl.proto = IPPROTO_TCP;
 889         ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
 890         ipv6_addr_copy(&fl.fl6_src, &req->af.v6_req.loc_addr);
 891         fl.fl6_flowlabel = 0;
 892         fl.oif = req->af.v6_req.iif;
 893         fl.fl_ip_dport = req->rmt_port;
 894         fl.fl_ip_sport = inet_sk(sk)->sport;
 895
 896         if (dst == NULL) {
 897                 opt = np->opt;
 898                 if (opt == NULL &&
 899                     np->rxopt.bits.srcrt == 2 &&
 900                     req->af.v6_req.pktopts) {
 901                         struct sk_buff *pktopts = req->af.v6_req.pktopts;
 902                         struct inet6_skb_parm *rxopt = IP6CB(pktopts);
 903                         if (rxopt->srcrt)
 904                                 opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(pktopts->nh.raw + rxopt->srcrt));
 905                 }
 906
 907                 if (opt && opt->srcrt) {
 908                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
 909                         ipv6_addr_copy(&final, &fl.fl6_dst);
 910                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 911                         final_p = &final;
 912                 }
 913
 914                 err = ip6_dst_lookup(sk, &dst, &fl);
 915                 if (err)
 916                         goto done;
 917                 if (final_p)
 918                         ipv6_addr_copy(&fl.fl6_dst, final_p);
 919                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
 920                         goto done;
 921         }
 922
 923         skb = tcp_make_synack(sk, dst, req);
 924         if (skb) {
 925                 struct tcphdr *th = skb->h.th;
 926
 927                 th->check = tcp_v6_check(th, skb->len,
 928                                          &req->af.v6_req.loc_addr, &req->af.v6_req.rmt_addr,
 929                                          csum_partial((char *)th, skb->len, skb->csum));
 930
 931                 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
 932                 err = ip6_xmit(sk, skb, &fl, opt, 0);
 933                 if (err == NET_XMIT_CN)
 934                         err = 0;
 935         }
 936
 937 done:
 938         dst_release(dst);
 939         if (opt && opt != np->opt)
 940                 sock_kfree_s(sk, opt, opt->tot_len);
 941         return err;
 942 }
 943
 944 static void tcp_v6_or_free(struct open_request *req)
 945 {
 946         if (req->af.v6_req.pktopts)
 947                 kfree_skb(req->af.v6_req.pktopts);
 948 }
 949
 950 static struct or_calltable or_ipv6 = {
 951         .family         =       AF_INET6,
 952         .rtx_syn_ack    =       tcp_v6_send_synack,
 953         .send_ack       =       tcp_v6_or_send_ack,
 954         .destructor     =       tcp_v6_or_free,
 955         .send_reset     =       tcp_v6_send_reset
 956 };
 957
 958 static int ipv6_opt_accepted(struct sock *sk, struct sk_buff *skb)
 959 {
 960         struct ipv6_pinfo *np = inet6_sk(sk);
 961         struct inet6_skb_parm *opt = IP6CB(skb);
 962
 963         if (np->rxopt.all) {
 964                 if ((opt->hop && np->rxopt.bits.hopopts) ||
 965                     ((IPV6_FLOWINFO_MASK&*(u32*)skb->nh.raw) &&
 966                      np->rxopt.bits.rxflow) ||
 967                     (opt->srcrt && np->rxopt.bits.srcrt) ||
 968                     ((opt->dst1 || opt->dst0) && np->rxopt.bits.dstopts))
 969                         return 1;
 970         }
 971         return 0;
 972 }
 973
 974
 975 static void tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len,
 976                               struct sk_buff *skb)
 977 {
 978         struct ipv6_pinfo *np = inet6_sk(sk);
 979
 980         if (skb->ip_summed == CHECKSUM_HW) {
 981                 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,  0);
 982                 skb->csum = offsetof(struct tcphdr, check);
 983         } else {
 984                 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
 985                                             csum_partial((char *)th, th->doff<<2,
 986                                                          skb->csum));
 987         }
 988 }
 989
 990
 991 static void tcp_v6_send_reset(struct sk_buff *skb)
 992 {
 993         struct tcphdr *th = skb->h.th, *t1;
 994         struct sk_buff *buff;
 995         struct flowi fl;
 996
 997         if (th->rst)
 998                 return;
 999
1000         if (!ipv6_unicast_destination(skb))
1001                 return;
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) + sizeof(struct tcphdr),
1009                          GFP_ATOMIC);
1010         if (buff == NULL)
1011                 return;
1012
1013         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + sizeof(struct tcphdr));
1014
1015         t1 = (struct tcphdr *) skb_push(buff,sizeof(struct tcphdr));
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 = sizeof(*t1)/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         buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
1033
1034         memset(&fl, 0, sizeof(fl));
1035         ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
1036         ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
1037
1038         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1039                                     sizeof(*t1), IPPROTO_TCP,
1040                                     buff->csum);
1041
1042         fl.proto = IPPROTO_TCP;
1043         fl.oif = tcp_v6_iif(skb);
1044         fl.fl_ip_dport = t1->dest;
1045         fl.fl_ip_sport = t1->source;
1046
1047         /* sk = NULL, but it is safe for now. RST socket required. */
1048         if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1049
1050                 if ((xfrm_lookup(&buff->dst, &fl, NULL, 0)) < 0) {
1051                         dst_release(buff->dst);
1052                         return;
1053                 }
1054
1055                 ip6_xmit(NULL, buff, &fl, NULL, 0);
1056                 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1057                 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
1058                 return;
1059         }
1060
1061         kfree_skb(buff);
1062 }
1063
1064 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
1065 {
1066         struct tcphdr *th = skb->h.th, *t1;
1067         struct sk_buff *buff;
1068         struct flowi fl;
1069         int tot_len = sizeof(struct tcphdr);
1070
1071         if (ts)
1072                 tot_len += 3*4;
1073
1074         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1075                          GFP_ATOMIC);
1076         if (buff == NULL)
1077                 return;
1078
1079         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1080
1081         t1 = (struct tcphdr *) skb_push(buff,tot_len);
1082
1083         /* Swap the send and the receive. */
1084         memset(t1, 0, sizeof(*t1));
1085         t1->dest = th->source;
1086         t1->source = th->dest;
1087         t1->doff = tot_len/4;
1088         t1->seq = htonl(seq);
1089         t1->ack_seq = htonl(ack);
1090         t1->ack = 1;
1091         t1->window = htons(win);
1092
1093         if (ts) {
1094                 u32 *ptr = (u32*)(t1 + 1);
1095                 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1096                                (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1097                 *ptr++ = htonl(tcp_time_stamp);
1098                 *ptr = htonl(ts);
1099         }
1100
1101         buff->csum = csum_partial((char *)t1, tot_len, 0);
1102
1103         memset(&fl, 0, sizeof(fl));
1104         ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
1105         ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
1106
1107         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1108                                     tot_len, IPPROTO_TCP,
1109                                     buff->csum);
1110
1111         fl.proto = IPPROTO_TCP;
1112         fl.oif = tcp_v6_iif(skb);
1113         fl.fl_ip_dport = t1->dest;
1114         fl.fl_ip_sport = t1->source;
1115
1116         if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
1117                 if ((xfrm_lookup(&buff->dst, &fl, NULL, 0)) < 0) {
1118                         dst_release(buff->dst);
1119                         return;
1120                 }
1121                 ip6_xmit(NULL, buff, &fl, NULL, 0);
1122                 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
1123                 return;
1124         }
1125
1126         kfree_skb(buff);
1127 }
1128
1129 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1130 {
1131         struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk;
1132
1133         tcp_v6_send_ack(skb, tw->tw_snd_nxt, tw->tw_rcv_nxt,
1134                         tw->tw_rcv_wnd >> tw->tw_rcv_wscale, tw->tw_ts_recent);
1135
1136         tcp_tw_put(tw);
1137 }
1138
1139 static void tcp_v6_or_send_ack(struct sk_buff *skb, struct open_request *req)
1140 {
1141         tcp_v6_send_ack(skb, req->snt_isn+1, req->rcv_isn+1, req->rcv_wnd, req->ts_recent);
1142 }
1143
1144
1145 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1146 {
1147         struct open_request *req, **prev;
1148         struct tcphdr *th = skb->h.th;
1149         struct tcp_sock *tp = tcp_sk(sk);
1150         struct sock *nsk;
1151
1152         /* Find possible connection requests. */
1153         req = tcp_v6_search_req(tp, &prev, th->source, &skb->nh.ipv6h->saddr,
1154                                 &skb->nh.ipv6h->daddr, tcp_v6_iif(skb));
1155         if (req)
1156                 return tcp_check_req(sk, skb, req, prev);
1157
1158         nsk = __tcp_v6_lookup_established(&skb->nh.ipv6h->saddr,
1159                                           th->source,
1160                                           &skb->nh.ipv6h->daddr,
1161                                           ntohs(th->dest),
1162                                           tcp_v6_iif(skb));
1163
1164         if (nsk) {
1165                 if (nsk->sk_state != TCP_TIME_WAIT) {
1166                         bh_lock_sock(nsk);
1167                         return nsk;
1168                 }
1169                 tcp_tw_put((struct tcp_tw_bucket*)nsk);
1170                 return NULL;
1171         }
1172
1173 #if 0 /*def CONFIG_SYN_COOKIES*/
1174         if (!th->rst && !th->syn && th->ack)
1175                 sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
1176 #endif
1177         return sk;
1178 }
1179
1180 static void tcp_v6_synq_add(struct sock *sk, struct open_request *req)
1181 {
1182         struct tcp_sock *tp = tcp_sk(sk);
1183         struct tcp_listen_opt *lopt = tp->listen_opt;
1184         u32 h = tcp_v6_synq_hash(&req->af.v6_req.rmt_addr, req->rmt_port, lopt->hash_rnd);
1185
1186         req->sk = NULL;
1187         req->expires = jiffies + TCP_TIMEOUT_INIT;
1188         req->retrans = 0;
1189         req->dl_next = lopt->syn_table[h];
1190
1191         write_lock(&tp->syn_wait_lock);
1192         lopt->syn_table[h] = req;
1193         write_unlock(&tp->syn_wait_lock);
1194
1195         tcp_synq_added(sk);
1196 }
1197
1198
1199 /* FIXME: this is substantially similar to the ipv4 code.
1200  * Can some kind of merge be done? -- erics
1201  */
1202 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1203 {
1204         struct ipv6_pinfo *np = inet6_sk(sk);
1205         struct tcp_options_received tmp_opt;
1206         struct tcp_sock *tp = tcp_sk(sk);
1207         struct open_request *req = NULL;
1208         __u32 isn = TCP_SKB_CB(skb)->when;
1209
1210         if (skb->protocol == htons(ETH_P_IP))
1211                 return tcp_v4_conn_request(sk, skb);
1212
1213         if (!ipv6_unicast_destination(skb))
1214                 goto drop;
1215
1216         /*
1217          *      There are no SYN attacks on IPv6, yet...
1218          */
1219         if (tcp_synq_is_full(sk) && !isn) {
1220                 if (net_ratelimit())
1221                         printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
1222                 goto drop;
1223         }
1224
1225         if (sk_acceptq_is_full(sk) && tcp_synq_young(sk) > 1)
1226                 goto drop;
1227
1228         req = tcp_openreq_alloc();
1229         if (req == NULL)
1230                 goto drop;
1231
1232         tcp_clear_options(&tmp_opt);
1233         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1234         tmp_opt.user_mss = tp->rx_opt.user_mss;
1235
1236         tcp_parse_options(skb, &tmp_opt, 0);
1237
1238         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1239         tcp_openreq_init(req, &tmp_opt, skb);
1240
1241         req->class = &or_ipv6;
1242         ipv6_addr_copy(&req->af.v6_req.rmt_addr, &skb->nh.ipv6h->saddr);
1243         ipv6_addr_copy(&req->af.v6_req.loc_addr, &skb->nh.ipv6h->daddr);
1244         TCP_ECN_create_request(req, skb->h.th);
1245         req->af.v6_req.pktopts = NULL;
1246         if (ipv6_opt_accepted(sk, skb) ||
1247             np->rxopt.bits.rxinfo ||
1248             np->rxopt.bits.rxhlim) {
1249                 atomic_inc(&skb->users);
1250                 req->af.v6_req.pktopts = skb;
1251         }
1252         req->af.v6_req.iif = sk->sk_bound_dev_if;
1253
1254         /* So that link locals have meaning */
1255         if (!sk->sk_bound_dev_if &&
1256             ipv6_addr_type(&req->af.v6_req.rmt_addr) & IPV6_ADDR_LINKLOCAL)
1257                 req->af.v6_req.iif = tcp_v6_iif(skb);
1258
1259         if (isn == 0)
1260                 isn = tcp_v6_init_sequence(sk,skb);
1261
1262         req->snt_isn = isn;
1263
1264         if (tcp_v6_send_synack(sk, req, NULL))
1265                 goto drop;
1266
1267         tcp_v6_synq_add(sk, req);
1268
1269         return 0;
1270
1271 drop:
1272         if (req)
1273                 tcp_openreq_free(req);
1274
1275         TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
1276         return 0; /* don't send reset */
1277 }
1278
1279 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1280                                           struct open_request *req,
1281                                           struct dst_entry *dst)
1282 {
1283         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1284         struct tcp6_sock *newtcp6sk;
1285         struct inet_sock *newinet;
1286         struct tcp_sock *newtp;
1287         struct sock *newsk;
1288         struct ipv6_txoptions *opt;
1289
1290         if (skb->protocol == htons(ETH_P_IP)) {
1291                 /*
1292                  *      v6 mapped
1293                  */
1294
1295                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1296
1297                 if (newsk == NULL)
1298                         return NULL;
1299
1300                 newtcp6sk = (struct tcp6_sock *)newsk;
1301                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1302
1303                 newinet = inet_sk(newsk);
1304                 newnp = inet6_sk(newsk);
1305                 newtp = tcp_sk(newsk);
1306
1307                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1308
1309                 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1310                               newinet->daddr);
1311
1312                 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1313                               newinet->saddr);
1314
1315                 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1316
1317                 newtp->af_specific = &ipv6_mapped;
1318                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1319                 newnp->pktoptions  = NULL;
1320                 newnp->opt         = NULL;
1321                 newnp->mcast_oif   = tcp_v6_iif(skb);
1322                 newnp->mcast_hops  = skb->nh.ipv6h->hop_limit;
1323
1324                 /* Charge newly allocated IPv6 socket. Though it is mapped,
1325                  * it is IPv6 yet.
1326                  */
1327 #ifdef INET_REFCNT_DEBUG
1328                 atomic_inc(&inet6_sock_nr);
1329 #endif
1330
1331                 /* It is tricky place. Until this moment IPv4 tcp
1332                    worked with IPv6 af_tcp.af_specific.
1333                    Sync it now.
1334                  */
1335                 tcp_sync_mss(newsk, newtp->pmtu_cookie);
1336
1337                 return newsk;
1338         }
1339
1340         opt = np->opt;
1341
1342         if (sk_acceptq_is_full(sk))
1343                 goto out_overflow;
1344
1345         if (np->rxopt.bits.srcrt == 2 &&
1346             opt == NULL && req->af.v6_req.pktopts) {
1347                 struct inet6_skb_parm *rxopt = IP6CB(req->af.v6_req.pktopts);
1348                 if (rxopt->srcrt)
1349                         opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(req->af.v6_req.pktopts->nh.raw+rxopt->srcrt));
1350         }
1351
1352         if (dst == NULL) {
1353                 struct in6_addr *final_p = NULL, final;
1354                 struct flowi fl;
1355
1356                 memset(&fl, 0, sizeof(fl));
1357                 fl.proto = IPPROTO_TCP;
1358                 ipv6_addr_copy(&fl.fl6_dst, &req->af.v6_req.rmt_addr);
1359                 if (opt && opt->srcrt) {
1360                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1361                         ipv6_addr_copy(&final, &fl.fl6_dst);
1362                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1363                         final_p = &final;
1364                 }
1365                 ipv6_addr_copy(&fl.fl6_src, &req->af.v6_req.loc_addr);
1366                 fl.oif = sk->sk_bound_dev_if;
1367                 fl.fl_ip_dport = req->rmt_port;
1368                 fl.fl_ip_sport = inet_sk(sk)->sport;
1369
1370                 if (ip6_dst_lookup(sk, &dst, &fl))
1371                         goto out;
1372
1373                 if (final_p)
1374                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1375
1376                 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1377                         goto out;
1378         }
1379
1380         newsk = tcp_create_openreq_child(sk, req, skb);
1381         if (newsk == NULL)
1382                 goto out;
1383
1384         /* Charge newly allocated IPv6 socket */
1385 #ifdef INET_REFCNT_DEBUG
1386         atomic_inc(&inet6_sock_nr);
1387 #endif
1388
1389         ip6_dst_store(newsk, dst, NULL);
1390         newsk->sk_route_caps = dst->dev->features &
1391                 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1392
1393         newtcp6sk = (struct tcp6_sock *)newsk;
1394         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1395
1396         newtp = tcp_sk(newsk);
1397         newinet = inet_sk(newsk);
1398         newnp = inet6_sk(newsk);
1399
1400         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1401
1402         ipv6_addr_copy(&newnp->daddr, &req->af.v6_req.rmt_addr);
1403         ipv6_addr_copy(&newnp->saddr, &req->af.v6_req.loc_addr);
1404         ipv6_addr_copy(&newnp->rcv_saddr, &req->af.v6_req.loc_addr);
1405         newsk->sk_bound_dev_if = req->af.v6_req.iif;
1406
1407         /* Now IPv6 options...
1408
1409            First: no IPv4 options.
1410          */
1411         newinet->opt = NULL;
1412
1413         /* Clone RX bits */
1414         newnp->rxopt.all = np->rxopt.all;
1415
1416         /* Clone pktoptions received with SYN */
1417         newnp->pktoptions = NULL;
1418         if (req->af.v6_req.pktopts) {
1419                 newnp->pktoptions = skb_clone(req->af.v6_req.pktopts,
1420                                               GFP_ATOMIC);
1421                 kfree_skb(req->af.v6_req.pktopts);
1422                 req->af.v6_req.pktopts = NULL;
1423                 if (newnp->pktoptions)
1424                         skb_set_owner_r(newnp->pktoptions, newsk);
1425         }
1426         newnp->opt        = NULL;
1427         newnp->mcast_oif  = tcp_v6_iif(skb);
1428         newnp->mcast_hops = skb->nh.ipv6h->hop_limit;
1429
1430         /* Clone native IPv6 options from listening socket (if any)
1431
1432            Yes, keeping reference count would be much more clever,
1433            but we make one more one thing there: reattach optmem
1434            to newsk.
1435          */
1436         if (opt) {
1437                 newnp->opt = ipv6_dup_options(newsk, opt);
1438                 if (opt != np->opt)
1439                         sock_kfree_s(sk, opt, opt->tot_len);
1440         }
1441
1442         newtp->ext_header_len = 0;
1443         if (newnp->opt)
1444                 newtp->ext_header_len = newnp->opt->opt_nflen +
1445                                         newnp->opt->opt_flen;
1446         newtp->ext2_header_len = dst->header_len;
1447
1448         tcp_sync_mss(newsk, dst_pmtu(dst));
1449         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1450         tcp_initialize_rcv_mss(newsk);
1451
1452         newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1453
1454         __tcp_v6_hash(newsk);
1455         tcp_inherit_port(sk, newsk);
1456
1457         return newsk;
1458
1459 out_overflow:
1460         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1461 out:
1462         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1463         if (opt && opt != np->opt)
1464                 sock_kfree_s(sk, opt, opt->tot_len);
1465         dst_release(dst);
1466         return NULL;
1467 }
1468
1469 static int tcp_v6_checksum_init(struct sk_buff *skb)
1470 {
1471         if (skb->ip_summed == CHECKSUM_HW) {
1472                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1473                 if (!tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1474                                   &skb->nh.ipv6h->daddr,skb->csum))
1475                         return 0;
1476                 LIMIT_NETDEBUG(printk(KERN_DEBUG "hw tcp v6 csum failed\n"));
1477         }
1478         if (skb->len <= 76) {
1479                 if (tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1480                                  &skb->nh.ipv6h->daddr,skb_checksum(skb, 0, skb->len, 0)))
1481                         return -1;
1482                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1483         } else {
1484                 skb->csum = ~tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1485                                           &skb->nh.ipv6h->daddr,0);
1486         }
1487         return 0;
1488 }
1489
1490 /* The socket must have it's spinlock held when we get
1491  * here.
1492  *
1493  * We have a potential double-lock case here, so even when
1494  * doing backlog processing we use the BH locking scheme.
1495  * This is because we cannot sleep with the original spinlock
1496  * held.
1497  */
1498 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1499 {
1500         struct ipv6_pinfo *np = inet6_sk(sk);
1501         struct tcp_sock *tp;
1502         struct sk_buff *opt_skb = NULL;
1503
1504         /* Imagine: socket is IPv6. IPv4 packet arrives,
1505            goes to IPv4 receive handler and backlogged.
1506            From backlog it always goes here. Kerboom...
1507            Fortunately, tcp_rcv_established and rcv_established
1508            handle them correctly, but it is not case with
1509            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1510          */
1511
1512         if (skb->protocol == htons(ETH_P_IP))
1513                 return tcp_v4_do_rcv(sk, skb);
1514
1515         if (sk_filter(sk, skb, 0))
1516                 goto discard;
1517
1518         /*
1519          *      socket locking is here for SMP purposes as backlog rcv
1520          *      is currently called with bh processing disabled.
1521          */
1522
1523         /* Do Stevens' IPV6_PKTOPTIONS.
1524
1525            Yes, guys, it is the only place in our code, where we
1526            may make it not affecting IPv4.
1527            The rest of code is protocol independent,
1528            and I do not like idea to uglify IPv4.
1529
1530            Actually, all the idea behind IPV6_PKTOPTIONS
1531            looks not very well thought. For now we latch
1532            options, received in the last packet, enqueued
1533            by tcp. Feel free to propose better solution.
1534                                                --ANK (980728)
1535          */
1536         if (np->rxopt.all)
1537                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1538
1539         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1540                 TCP_CHECK_TIMER(sk);
1541                 if (tcp_rcv_established(sk, skb, skb->h.th, skb->len))
1542                         goto reset;
1543                 TCP_CHECK_TIMER(sk);
1544                 if (opt_skb)
1545                         goto ipv6_pktoptions;
1546                 return 0;
1547         }
1548
1549         if (skb->len < (skb->h.th->doff<<2) || tcp_checksum_complete(skb))
1550                 goto csum_err;
1551
1552         if (sk->sk_state == TCP_LISTEN) {
1553                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1554                 if (!nsk)
1555                         goto discard;
1556
1557                 /*
1558                  * Queue it on the new socket if the new socket is active,
1559                  * otherwise we just shortcircuit this and continue with
1560                  * the new socket..
1561                  */
1562                 if(nsk != sk) {
1563                         if (tcp_child_process(sk, nsk, skb))
1564                                 goto reset;
1565                         if (opt_skb)
1566                                 __kfree_skb(opt_skb);
1567                         return 0;
1568                 }
1569         }
1570
1571         TCP_CHECK_TIMER(sk);
1572         if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len))
1573                 goto reset;
1574         TCP_CHECK_TIMER(sk);
1575         if (opt_skb)
1576                 goto ipv6_pktoptions;
1577         return 0;
1578
1579 reset:
1580         tcp_v6_send_reset(skb);
1581 discard:
1582         if (opt_skb)
1583                 __kfree_skb(opt_skb);
1584         kfree_skb(skb);
1585         return 0;
1586 csum_err:
1587         TCP_INC_STATS_BH(TCP_MIB_INERRS);
1588         goto discard;
1589
1590
1591 ipv6_pktoptions:
1592         /* Do you ask, what is it?
1593
1594            1. skb was enqueued by tcp.
1595            2. skb is added to tail of read queue, rather than out of order.
1596            3. socket is not in passive state.
1597            4. Finally, it really contains options, which user wants to receive.
1598          */
1599         tp = tcp_sk(sk);
1600         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1601             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1602                 if (np->rxopt.bits.rxinfo)
1603                         np->mcast_oif = tcp_v6_iif(opt_skb);
1604                 if (np->rxopt.bits.rxhlim)
1605                         np->mcast_hops = opt_skb->nh.ipv6h->hop_limit;
1606                 if (ipv6_opt_accepted(sk, opt_skb)) {
1607                         skb_set_owner_r(opt_skb, sk);
1608                         opt_skb = xchg(&np->pktoptions, opt_skb);
1609                 } else {
1610                         __kfree_skb(opt_skb);
1611                         opt_skb = xchg(&np->pktoptions, NULL);
1612                 }
1613         }
1614
1615         if (opt_skb)
1616                 kfree_skb(opt_skb);
1617         return 0;
1618 }
1619
1620 static int tcp_v6_rcv(struct sk_buff **pskb, unsigned int *nhoffp)
1621 {
1622         struct sk_buff *skb = *pskb;
1623         struct tcphdr *th;
1624         struct sock *sk;
1625         int ret;
1626
1627         if (skb->pkt_type != PACKET_HOST)
1628                 goto discard_it;
1629
1630         /*
1631          *      Count it even if it's bad.
1632          */
1633         TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1634
1635         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1636                 goto discard_it;
1637
1638         th = skb->h.th;
1639
1640         if (th->doff < sizeof(struct tcphdr)/4)
1641                 goto bad_packet;
1642         if (!pskb_may_pull(skb, th->doff*4))
1643                 goto discard_it;
1644
1645         if ((skb->ip_summed != CHECKSUM_UNNECESSARY &&
1646              tcp_v6_checksum_init(skb) < 0))
1647                 goto bad_packet;
1648
1649         th = skb->h.th;
1650         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1651         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1652                                     skb->len - th->doff*4);
1653         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1654         TCP_SKB_CB(skb)->when = 0;
1655         TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(skb->nh.ipv6h);
1656         TCP_SKB_CB(skb)->sacked = 0;
1657
1658         sk = __tcp_v6_lookup(&skb->nh.ipv6h->saddr, th->source,
1659                              &skb->nh.ipv6h->daddr, ntohs(th->dest), tcp_v6_iif(skb));
1660
1661         if (!sk)
1662                 goto no_tcp_socket;
1663
1664 process:
1665         if (sk->sk_state == TCP_TIME_WAIT)
1666                 goto do_time_wait;
1667
1668         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1669                 goto discard_and_relse;
1670
1671         if (sk_filter(sk, skb, 0))
1672                 goto discard_and_relse;
1673
1674         skb->dev = NULL;
1675
1676         bh_lock_sock(sk);
1677         ret = 0;
1678         if (!sock_owned_by_user(sk)) {
1679                 if (!tcp_prequeue(sk, skb))
1680                         ret = tcp_v6_do_rcv(sk, skb);
1681         } else
1682                 sk_add_backlog(sk, skb);
1683         bh_unlock_sock(sk);
1684
1685         sock_put(sk);
1686         return ret ? -1 : 0;
1687
1688 no_tcp_socket:
1689         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1690                 goto discard_it;
1691
1692         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1693 bad_packet:
1694                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1695         } else {
1696                 tcp_v6_send_reset(skb);
1697         }
1698
1699 discard_it:
1700
1701         /*
1702          *      Discard frame
1703          */
1704
1705         kfree_skb(skb);
1706         return 0;
1707
1708 discard_and_relse:
1709         sock_put(sk);
1710         goto discard_it;
1711
1712 do_time_wait:
1713         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1714                 tcp_tw_put((struct tcp_tw_bucket *) sk);
1715                 goto discard_it;
1716         }
1717
1718         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1719                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1720                 tcp_tw_put((struct tcp_tw_bucket *) sk);
1721                 goto discard_it;
1722         }
1723
1724         switch(tcp_timewait_state_process((struct tcp_tw_bucket *)sk,
1725                                           skb, th, skb->len)) {
1726         case TCP_TW_SYN:
1727         {
1728                 struct sock *sk2;
1729
1730                 sk2 = tcp_v6_lookup_listener(&skb->nh.ipv6h->daddr, ntohs(th->dest), tcp_v6_iif(skb));
1731                 if (sk2 != NULL) {
1732                         tcp_tw_deschedule((struct tcp_tw_bucket *)sk);
1733                         tcp_tw_put((struct tcp_tw_bucket *)sk);
1734                         sk = sk2;
1735                         goto process;
1736                 }
1737                 /* Fall through to ACK */
1738         }
1739         case TCP_TW_ACK:
1740                 tcp_v6_timewait_ack(sk, skb);
1741                 break;
1742         case TCP_TW_RST:
1743                 goto no_tcp_socket;
1744         case TCP_TW_SUCCESS:;
1745         }
1746         goto discard_it;
1747 }
1748
1749 static int tcp_v6_rebuild_header(struct sock *sk)
1750 {
1751         int err;
1752         struct dst_entry *dst;
1753         struct ipv6_pinfo *np = inet6_sk(sk);
1754
1755         dst = __sk_dst_check(sk, np->dst_cookie);
1756
1757         if (dst == NULL) {
1758                 struct inet_sock *inet = inet_sk(sk);
1759                 struct in6_addr *final_p = NULL, final;
1760                 struct flowi fl;
1761
1762                 memset(&fl, 0, sizeof(fl));
1763                 fl.proto = IPPROTO_TCP;
1764                 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1765                 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1766                 fl.fl6_flowlabel = np->flow_label;
1767                 fl.oif = sk->sk_bound_dev_if;
1768                 fl.fl_ip_dport = inet->dport;
1769                 fl.fl_ip_sport = inet->sport;
1770
1771                 if (np->opt && np->opt->srcrt) {
1772                         struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1773                         ipv6_addr_copy(&final, &fl.fl6_dst);
1774                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1775                         final_p = &final;
1776                 }
1777
1778                 err = ip6_dst_lookup(sk, &dst, &fl);
1779                 if (err) {
1780                         sk->sk_route_caps = 0;
1781                         return err;
1782                 }
1783                 if (final_p)
1784                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1785
1786                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
1787                         sk->sk_err_soft = -err;
1788                         dst_release(dst);
1789                         return err;
1790                 }
1791
1792                 ip6_dst_store(sk, dst, NULL);
1793                 sk->sk_route_caps = dst->dev->features &
1794                         ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1795                 tcp_sk(sk)->ext2_header_len = dst->header_len;
1796         }
1797
1798         return 0;
1799 }
1800
1801 static int tcp_v6_xmit(struct sk_buff *skb, int ipfragok)
1802 {
1803         struct sock *sk = skb->sk;
1804         struct inet_sock *inet = inet_sk(sk);
1805         struct ipv6_pinfo *np = inet6_sk(sk);
1806         struct flowi fl;
1807         struct dst_entry *dst;
1808         struct in6_addr *final_p = NULL, final;
1809
1810         memset(&fl, 0, sizeof(fl));
1811         fl.proto = IPPROTO_TCP;
1812         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1813         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1814         fl.fl6_flowlabel = np->flow_label;
1815         IP6_ECN_flow_xmit(sk, fl.fl6_flowlabel);
1816         fl.oif = sk->sk_bound_dev_if;
1817         fl.fl_ip_sport = inet->sport;
1818         fl.fl_ip_dport = inet->dport;
1819
1820         if (np->opt && np->opt->srcrt) {
1821                 struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1822                 ipv6_addr_copy(&final, &fl.fl6_dst);
1823                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1824                 final_p = &final;
1825         }
1826
1827         dst = __sk_dst_check(sk, np->dst_cookie);
1828
1829         if (dst == NULL) {
1830                 int err = ip6_dst_lookup(sk, &dst, &fl);
1831
1832                 if (err) {
1833                         sk->sk_err_soft = -err;
1834                         return err;
1835                 }
1836
1837                 if (final_p)
1838                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1839
1840                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
1841                         sk->sk_route_caps = 0;
1842                         dst_release(dst);
1843                         return err;
1844                 }
1845
1846                 ip6_dst_store(sk, dst, NULL);
1847                 sk->sk_route_caps = dst->dev->features &
1848                         ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1849                 tcp_sk(sk)->ext2_header_len = dst->header_len;
1850         }
1851
1852         skb->dst = dst_clone(dst);
1853
1854         /* Restore final destination back after routing done */
1855         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1856
1857         return ip6_xmit(sk, skb, &fl, np->opt, 0);
1858 }
1859
1860 static void v6_addr2sockaddr(struct sock *sk, struct sockaddr * uaddr)
1861 {
1862         struct ipv6_pinfo *np = inet6_sk(sk);
1863         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) uaddr;
1864
1865         sin6->sin6_family = AF_INET6;
1866         ipv6_addr_copy(&sin6->sin6_addr, &np->daddr);
1867         sin6->sin6_port = inet_sk(sk)->dport;
1868         /* We do not store received flowlabel for TCP */
1869         sin6->sin6_flowinfo = 0;
1870         sin6->sin6_scope_id = 0;
1871         if (sk->sk_bound_dev_if &&
1872             ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
1873                 sin6->sin6_scope_id = sk->sk_bound_dev_if;
1874 }
1875
1876 static int tcp_v6_remember_stamp(struct sock *sk)
1877 {
1878         /* Alas, not yet... */
1879         return 0;
1880 }
1881
1882 static struct tcp_func ipv6_specific = {
1883         .queue_xmit     =       tcp_v6_xmit,
1884         .send_check     =       tcp_v6_send_check,
1885         .rebuild_header =       tcp_v6_rebuild_header,
1886         .conn_request   =       tcp_v6_conn_request,
1887         .syn_recv_sock  =       tcp_v6_syn_recv_sock,
1888         .remember_stamp =       tcp_v6_remember_stamp,
1889         .net_header_len =       sizeof(struct ipv6hdr),
1890
1891         .setsockopt     =       ipv6_setsockopt,
1892         .getsockopt     =       ipv6_getsockopt,
1893         .addr2sockaddr  =       v6_addr2sockaddr,
1894         .sockaddr_len   =       sizeof(struct sockaddr_in6)
1895 };
1896
1897 /*
1898  *      TCP over IPv4 via INET6 API
1899  */
1900
1901 static struct tcp_func ipv6_mapped = {
1902         .queue_xmit     =       ip_queue_xmit,
1903         .send_check     =       tcp_v4_send_check,
1904         .rebuild_header =       tcp_v4_rebuild_header,
1905         .conn_request   =       tcp_v6_conn_request,
1906         .syn_recv_sock  =       tcp_v6_syn_recv_sock,
1907         .remember_stamp =       tcp_v4_remember_stamp,
1908         .net_header_len =       sizeof(struct iphdr),
1909
1910         .setsockopt     =       ipv6_setsockopt,
1911         .getsockopt     =       ipv6_getsockopt,
1912         .addr2sockaddr  =       v6_addr2sockaddr,
1913         .sockaddr_len   =       sizeof(struct sockaddr_in6)
1914 };
1915
1916
1917
1918 /* NOTE: A lot of things set to zero explicitly by call to
1919  *       sk_alloc() so need not be done here.
1920  */
1921 static int tcp_v6_init_sock(struct sock *sk)
1922 {
1923         struct tcp_sock *tp = tcp_sk(sk);
1924
1925         skb_queue_head_init(&tp->out_of_order_queue);
1926         tcp_init_xmit_timers(sk);
1927         tcp_prequeue_init(tp);
1928
1929         tp->rto  = TCP_TIMEOUT_INIT;
1930         tp->mdev = TCP_TIMEOUT_INIT;
1931
1932         /* So many TCP implementations out there (incorrectly) count the
1933          * initial SYN frame in their delayed-ACK and congestion control
1934          * algorithms that we must have the following bandaid to talk
1935          * efficiently to them.  -DaveM
1936          */
1937         tp->snd_cwnd = 2;
1938
1939         /* See draft-stevens-tcpca-spec-01 for discussion of the
1940          * initialization of these values.
1941          */
1942         tp->snd_ssthresh = 0x7fffffff;
1943         tp->snd_cwnd_clamp = ~0;
1944         tp->mss_cache_std = tp->mss_cache = 536;
1945
1946         tp->reordering = sysctl_tcp_reordering;
1947
1948         sk->sk_state = TCP_CLOSE;
1949
1950         tp->af_specific = &ipv6_specific;
1951
1952         sk->sk_write_space = sk_stream_write_space;
1953         sk->sk_use_write_queue = 1;
1954
1955         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1956         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1957
1958         atomic_inc(&tcp_sockets_allocated);
1959
1960         return 0;
1961 }
1962
1963 static int tcp_v6_destroy_sock(struct sock *sk)
1964 {
1965         extern int tcp_v4_destroy_sock(struct sock *sk);
1966
1967         tcp_v4_destroy_sock(sk);
1968         return inet6_destroy_sock(sk);
1969 }
1970
1971 /* Proc filesystem TCPv6 sock list dumping. */
1972 static void get_openreq6(struct seq_file *seq,
1973                          struct sock *sk, struct open_request *req, int i, int uid)
1974 {
1975         struct in6_addr *dest, *src;
1976         int ttd = req->expires - jiffies;
1977
1978         if (ttd < 0)
1979                 ttd = 0;
1980
1981         src = &req->af.v6_req.loc_addr;
1982         dest = &req->af.v6_req.rmt_addr;
1983         seq_printf(seq,
1984                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1985                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1986                    i,
1987                    src->s6_addr32[0], src->s6_addr32[1],
1988                    src->s6_addr32[2], src->s6_addr32[3],
1989                    ntohs(inet_sk(sk)->sport),
1990                    dest->s6_addr32[0], dest->s6_addr32[1],
1991                    dest->s6_addr32[2], dest->s6_addr32[3],
1992                    ntohs(req->rmt_port),
1993                    TCP_SYN_RECV,
1994                    0,0, /* could print option size, but that is af dependent. */
1995                    1,   /* timers active (only the expire timer) */
1996                    jiffies_to_clock_t(ttd),
1997                    req->retrans,
1998                    uid,
1999                    0,  /* non standard timer */
2000                    0, /* open_requests have no inode */
2001                    0, req);
2002 }
2003
2004 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2005 {
2006         struct in6_addr *dest, *src;
2007         __u16 destp, srcp;
2008         int timer_active;
2009         unsigned long timer_expires;
2010         struct inet_sock *inet = inet_sk(sp);
2011         struct tcp_sock *tp = tcp_sk(sp);
2012         struct ipv6_pinfo *np = inet6_sk(sp);
2013
2014         dest  = &np->daddr;
2015         src   = &np->rcv_saddr;
2016         destp = ntohs(inet->dport);
2017         srcp  = ntohs(inet->sport);
2018         if (tp->pending == TCP_TIME_RETRANS) {
2019                 timer_active    = 1;
2020                 timer_expires   = tp->timeout;
2021         } else if (tp->pending == TCP_TIME_PROBE0) {
2022                 timer_active    = 4;
2023                 timer_expires   = tp->timeout;
2024         } else if (timer_pending(&sp->sk_timer)) {
2025                 timer_active    = 2;
2026                 timer_expires   = sp->sk_timer.expires;
2027         } else {
2028                 timer_active    = 0;
2029                 timer_expires = jiffies;
2030         }
2031
2032         seq_printf(seq,
2033                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2034                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
2035                    i,
2036                    src->s6_addr32[0], src->s6_addr32[1],
2037                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2038                    dest->s6_addr32[0], dest->s6_addr32[1],
2039                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2040                    sp->sk_state,
2041                    tp->write_seq-tp->snd_una, tp->rcv_nxt-tp->copied_seq,
2042                    timer_active,
2043                    jiffies_to_clock_t(timer_expires - jiffies),
2044                    tp->retransmits,
2045                    sock_i_uid(sp),
2046                    tp->probes_out,
2047                    sock_i_ino(sp),
2048                    atomic_read(&sp->sk_refcnt), sp,
2049                    tp->rto, tp->ack.ato, (tp->ack.quick<<1)|tp->ack.pingpong,
2050                    tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
2051                    );
2052 }
2053
2054 static void get_timewait6_sock(struct seq_file *seq,
2055                                struct tcp_tw_bucket *tw, int i)
2056 {
2057         struct in6_addr *dest, *src;
2058         __u16 destp, srcp;
2059         int ttd = tw->tw_ttd - jiffies;
2060
2061         if (ttd < 0)
2062                 ttd = 0;
2063
2064         dest  = &tw->tw_v6_daddr;
2065         src   = &tw->tw_v6_rcv_saddr;
2066         destp = ntohs(tw->tw_dport);
2067         srcp  = ntohs(tw->tw_sport);
2068
2069         seq_printf(seq,
2070                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2071                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2072                    i,
2073                    src->s6_addr32[0], src->s6_addr32[1],
2074                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2075                    dest->s6_addr32[0], dest->s6_addr32[1],
2076                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2077                    tw->tw_substate, 0, 0,
2078                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2079                    atomic_read(&tw->tw_refcnt), tw);
2080 }
2081
2082 #ifdef CONFIG_PROC_FS
2083 static int tcp6_seq_show(struct seq_file *seq, void *v)
2084 {
2085         struct tcp_iter_state *st;
2086
2087         if (v == SEQ_START_TOKEN) {
2088                 seq_puts(seq,
2089                          "  sl  "
2090                          "local_address                         "
2091                          "remote_address                        "
2092                          "st tx_queue rx_queue tr tm->when retrnsmt"
2093                          "   uid  timeout inode\n");
2094                 goto out;
2095         }
2096         st = seq->private;
2097
2098         switch (st->state) {
2099         case TCP_SEQ_STATE_LISTENING:
2100         case TCP_SEQ_STATE_ESTABLISHED:
2101                 get_tcp6_sock(seq, v, st->num);
2102                 break;
2103         case TCP_SEQ_STATE_OPENREQ:
2104                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2105                 break;
2106         case TCP_SEQ_STATE_TIME_WAIT:
2107                 get_timewait6_sock(seq, v, st->num);
2108                 break;
2109         }
2110 out:
2111         return 0;
2112 }
2113
2114 static struct file_operations tcp6_seq_fops;
2115 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2116         .owner          = THIS_MODULE,
2117         .name           = "tcp6",
2118         .family         = AF_INET6,
2119         .seq_show       = tcp6_seq_show,
2120         .seq_fops       = &tcp6_seq_fops,
2121 };
2122
2123 int __init tcp6_proc_init(void)
2124 {
2125         return tcp_proc_register(&tcp6_seq_afinfo);
2126 }
2127
2128 void tcp6_proc_exit(void)
2129 {
2130         tcp_proc_unregister(&tcp6_seq_afinfo);
2131 }
2132 #endif
2133
2134 struct proto tcpv6_prot = {
2135         .name                   = "TCPv6",
2136         .owner                  = THIS_MODULE,
2137         .close                  = tcp_close,
2138         .connect                = tcp_v6_connect,
2139         .disconnect             = tcp_disconnect,
2140         .accept                 = tcp_accept,
2141         .ioctl                  = tcp_ioctl,
2142         .init                   = tcp_v6_init_sock,
2143         .destroy                = tcp_v6_destroy_sock,
2144         .shutdown               = tcp_shutdown,
2145         .setsockopt             = tcp_setsockopt,
2146         .getsockopt             = tcp_getsockopt,
2147         .sendmsg                = tcp_sendmsg,
2148         .recvmsg                = tcp_recvmsg,
2149         .backlog_rcv            = tcp_v6_do_rcv,
2150         .hash                   = tcp_v6_hash,
2151         .unhash                 = tcp_unhash,
2152         .get_port               = tcp_v6_get_port,
2153         .enter_memory_pressure  = tcp_enter_memory_pressure,
2154         .sockets_allocated      = &tcp_sockets_allocated,
2155         .memory_allocated       = &tcp_memory_allocated,
2156         .memory_pressure        = &tcp_memory_pressure,
2157         .sysctl_mem             = sysctl_tcp_mem,
2158         .sysctl_wmem            = sysctl_tcp_wmem,
2159         .sysctl_rmem            = sysctl_tcp_rmem,
2160         .max_header             = MAX_TCP_HEADER,
2161         .slab_obj_size          = sizeof(struct tcp6_sock),
2162 };
2163
2164 static struct inet6_protocol tcpv6_protocol = {
2165         .handler        =       tcp_v6_rcv,
2166         .err_handler    =       tcp_v6_err,
2167         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2168 };
2169
2170 extern struct proto_ops inet6_stream_ops;
2171
2172 static struct inet_protosw tcpv6_protosw = {
2173         .type           =       SOCK_STREAM,
2174         .protocol       =       IPPROTO_TCP,
2175         .prot           =       &tcpv6_prot,
2176         .ops            =       &inet6_stream_ops,
2177         .capability     =       -1,
2178         .no_check       =       0,
2179         .flags          =       INET_PROTOSW_PERMANENT,
2180 };
2181
2182 void __init tcpv6_init(void)
2183 {
2184         /* register inet6 protocol */
2185         if (inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP) < 0)
2186                 printk(KERN_ERR "tcpv6_init: Could not register protocol\n");
2187         inet6_register_protosw(&tcpv6_protosw);
2188 }