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/inet6_hashtables.h>
  51 #include <net/ipv6.h>
  52 #include <net/transp_v6.h>
  53 #include <net/addrconf.h>
  54 #include <net/ip6_route.h>
  55 #include <net/ip6_checksum.h>
  56 #include <net/inet_ecn.h>
  57 #include <net/protocol.h>
  58 #include <net/xfrm.h>
  59 #include <net/addrconf.h>
  60 #include <net/snmp.h>
  61 #include <net/dsfield.h>
  62
  63 #include <asm/uaccess.h>
  64
  65 #include <linux/proc_fs.h>
  66 #include <linux/seq_file.h>
  67
  68 static void     tcp_v6_send_reset(struct sk_buff *skb);
  69 static void     tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req);
  70 static void     tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len,
  71                                   struct sk_buff *skb);
  72
  73 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  74 static int      tcp_v6_xmit(struct sk_buff *skb, int ipfragok);
  75
  76 static struct tcp_func ipv6_mapped;
  77 static struct tcp_func ipv6_specific;
  78
  79 int inet6_csk_bind_conflict(const struct sock *sk,
  80                             const struct inet_bind_bucket *tb)
  81 {
  82         const struct sock *sk2;
  83         const struct hlist_node *node;
  84
  85         /* We must walk the whole port owner list in this case. -DaveM */
  86         sk_for_each_bound(sk2, node, &tb->owners) {
  87                 if (sk != sk2 &&
  88                     (!sk->sk_bound_dev_if ||
  89                      !sk2->sk_bound_dev_if ||
  90                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if) &&
  91                     (!sk->sk_reuse || !sk2->sk_reuse ||
  92                      sk2->sk_state == TCP_LISTEN) &&
  93                      ipv6_rcv_saddr_equal(sk, sk2))
  94                         break;
  95         }
  96
  97         return node != NULL;
  98 }
  99
 100 static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
 101 {
 102         return inet_csk_get_port(&tcp_hashinfo, sk, snum,
 103                                  inet6_csk_bind_conflict);
 104 }
 105
 106 static void tcp_v6_hash(struct sock *sk)
 107 {
 108         if (sk->sk_state != TCP_CLOSE) {
 109                 struct tcp_sock *tp = tcp_sk(sk);
 110
 111                 if (tp->af_specific == &ipv6_mapped) {
 112                         tcp_prot.hash(sk);
 113                         return;
 114                 }
 115                 local_bh_disable();
 116                 __inet6_hash(&tcp_hashinfo, sk);
 117                 local_bh_enable();
 118         }
 119 }
 120
 121 /*
 122  * Open request hash tables.
 123  */
 124
 125 static u32 tcp_v6_synq_hash(const struct in6_addr *raddr, const u16 rport, const u32 rnd)
 126 {
 127         u32 a, b, c;
 128
 129         a = raddr->s6_addr32[0];
 130         b = raddr->s6_addr32[1];
 131         c = raddr->s6_addr32[2];
 132
 133         a += JHASH_GOLDEN_RATIO;
 134         b += JHASH_GOLDEN_RATIO;
 135         c += rnd;
 136         __jhash_mix(a, b, c);
 137
 138         a += raddr->s6_addr32[3];
 139         b += (u32) rport;
 140         __jhash_mix(a, b, c);
 141
 142         return c & (TCP_SYNQ_HSIZE - 1);
 143 }
 144
 145 static struct request_sock *tcp_v6_search_req(const struct sock *sk,
 146                                               struct request_sock ***prevp,
 147                                               __u16 rport,
 148                                               struct in6_addr *raddr,
 149                                               struct in6_addr *laddr,
 150                                               int iif)
 151 {
 152         const struct inet_connection_sock *icsk = inet_csk(sk);
 153         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 154         struct request_sock *req, **prev;
 155
 156         for (prev = &lopt->syn_table[tcp_v6_synq_hash(raddr, rport, lopt->hash_rnd)];
 157              (req = *prev) != NULL;
 158              prev = &req->dl_next) {
 159                 const struct tcp6_request_sock *treq = tcp6_rsk(req);
 160
 161                 if (inet_rsk(req)->rmt_port == rport &&
 162                     req->rsk_ops->family == AF_INET6 &&
 163                     ipv6_addr_equal(&treq->rmt_addr, raddr) &&
 164                     ipv6_addr_equal(&treq->loc_addr, laddr) &&
 165                     (!treq->iif || treq->iif == iif)) {
 166                         BUG_TRAP(req->sk == NULL);
 167                         *prevp = prev;
 168                         return req;
 169                 }
 170         }
 171
 172         return NULL;
 173 }
 174
 175 static __inline__ u16 tcp_v6_check(struct tcphdr *th, int len,
 176                                    struct in6_addr *saddr,
 177                                    struct in6_addr *daddr,
 178                                    unsigned long base)
 179 {
 180         return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
 181 }
 182
 183 static __u32 tcp_v6_init_sequence(struct sock *sk, struct sk_buff *skb)
 184 {
 185         if (skb->protocol == htons(ETH_P_IPV6)) {
 186                 return secure_tcpv6_sequence_number(skb->nh.ipv6h->daddr.s6_addr32,
 187                                                     skb->nh.ipv6h->saddr.s6_addr32,
 188                                                     skb->h.th->dest,
 189                                                     skb->h.th->source);
 190         } else {
 191                 return secure_tcp_sequence_number(skb->nh.iph->daddr,
 192                                                   skb->nh.iph->saddr,
 193                                                   skb->h.th->dest,
 194                                                   skb->h.th->source);
 195         }
 196 }
 197
 198 static int __tcp_v6_check_established(struct sock *sk, const __u16 lport,
 199                                       struct inet_timewait_sock **twp)
 200 {
 201         struct inet_sock *inet = inet_sk(sk);
 202         const struct ipv6_pinfo *np = inet6_sk(sk);
 203         const struct in6_addr *daddr = &np->rcv_saddr;
 204         const struct in6_addr *saddr = &np->daddr;
 205         const int dif = sk->sk_bound_dev_if;
 206         const u32 ports = INET_COMBINED_PORTS(inet->dport, lport);
 207         unsigned int hash = inet6_ehashfn(daddr, inet->num, saddr, inet->dport);
 208         struct inet_ehash_bucket *head = inet_ehash_bucket(&tcp_hashinfo, hash);
 209         struct sock *sk2;
 210         const struct hlist_node *node;
 211         struct inet_timewait_sock *tw;
 212
 213         prefetch(head->chain.first);
 214         write_lock(&head->lock);
 215
 216         /* Check TIME-WAIT sockets first. */
 217         sk_for_each(sk2, node, &(head + tcp_hashinfo.ehash_size)->chain) {
 218                 const struct tcp6_timewait_sock *tcp6tw = tcp6_twsk(sk2);
 219
 220                 tw = inet_twsk(sk2);
 221
 222                 if(*((__u32 *)&(tw->tw_dport))  == ports        &&
 223                    sk2->sk_family               == PF_INET6     &&
 224                    ipv6_addr_equal(&tcp6tw->tw_v6_daddr, saddr) &&
 225                    ipv6_addr_equal(&tcp6tw->tw_v6_rcv_saddr, daddr)     &&
 226                    sk2->sk_bound_dev_if == sk->sk_bound_dev_if) {
 227                         const struct tcp_timewait_sock *tcptw = tcp_twsk(sk2);
 228                         struct tcp_sock *tp = tcp_sk(sk);
 229
 230                         if (tcptw->tw_ts_recent_stamp &&
 231                             (!twp ||
 232                              (sysctl_tcp_tw_reuse &&
 233                               xtime.tv_sec - tcptw->tw_ts_recent_stamp > 1))) {
 234                                 /* See comment in tcp_ipv4.c */
 235                                 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
 236                                 if (!tp->write_seq)
 237                                         tp->write_seq = 1;
 238                                 tp->rx_opt.ts_recent       = tcptw->tw_ts_recent;
 239                                 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
 240                                 sock_hold(sk2);
 241                                 goto unique;
 242                         } else
 243                                 goto not_unique;
 244                 }
 245         }
 246         tw = NULL;
 247
 248         /* And established part... */
 249         sk_for_each(sk2, node, &head->chain) {
 250                 if (INET6_MATCH(sk2, hash, saddr, daddr, ports, dif))
 251                         goto not_unique;
 252         }
 253
 254 unique:
 255         BUG_TRAP(sk_unhashed(sk));
 256         __sk_add_node(sk, &head->chain);
 257         sk->sk_hash = hash;
 258         sock_prot_inc_use(sk->sk_prot);
 259         write_unlock(&head->lock);
 260
 261         if (twp) {
 262                 *twp = tw;
 263                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 264         } else if (tw) {
 265                 /* Silly. Should hash-dance instead... */
 266                 inet_twsk_deschedule(tw, &tcp_death_row);
 267                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 268
 269                 inet_twsk_put(tw);
 270         }
 271         return 0;
 272
 273 not_unique:
 274         write_unlock(&head->lock);
 275         return -EADDRNOTAVAIL;
 276 }
 277
 278 static inline u32 tcpv6_port_offset(const struct sock *sk)
 279 {
 280         const struct inet_sock *inet = inet_sk(sk);
 281         const struct ipv6_pinfo *np = inet6_sk(sk);
 282
 283         return secure_tcpv6_port_ephemeral(np->rcv_saddr.s6_addr32,
 284                                            np->daddr.s6_addr32,
 285                                            inet->dport);
 286 }
 287
 288 static int tcp_v6_hash_connect(struct sock *sk)
 289 {
 290         unsigned short snum = inet_sk(sk)->num;
 291         struct inet_bind_hashbucket *head;
 292         struct inet_bind_bucket *tb;
 293         int ret;
 294
 295         if (!snum) {
 296                 int low = sysctl_local_port_range[0];
 297                 int high = sysctl_local_port_range[1];
 298                 int range = high - low;
 299                 int i;
 300                 int port;
 301                 static u32 hint;
 302                 u32 offset = hint + tcpv6_port_offset(sk);
 303                 struct hlist_node *node;
 304                 struct inet_timewait_sock *tw = NULL;
 305
 306                 local_bh_disable();
 307                 for (i = 1; i <= range; i++) {
 308                         port = low + (i + offset) % range;
 309                         head = &tcp_hashinfo.bhash[inet_bhashfn(port, tcp_hashinfo.bhash_size)];
 310                         spin_lock(&head->lock);
 311
 312                         /* Does not bother with rcv_saddr checks,
 313                          * because the established check is already
 314                          * unique enough.
 315                          */
 316                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 317                                 if (tb->port == port) {
 318                                         BUG_TRAP(!hlist_empty(&tb->owners));
 319                                         if (tb->fastreuse >= 0)
 320                                                 goto next_port;
 321                                         if (!__tcp_v6_check_established(sk,
 322                                                                         port,
 323                                                                         &tw))
 324                                                 goto ok;
 325                                         goto next_port;
 326                                 }
 327                         }
 328
 329                         tb = inet_bind_bucket_create(tcp_hashinfo.bind_bucket_cachep, head, port);
 330                         if (!tb) {
 331                                 spin_unlock(&head->lock);
 332                                 break;
 333                         }
 334                         tb->fastreuse = -1;
 335                         goto ok;
 336
 337                 next_port:
 338                         spin_unlock(&head->lock);
 339                 }
 340                 local_bh_enable();
 341
 342                 return -EADDRNOTAVAIL;
 343
 344 ok:
 345                 hint += i;
 346
 347                 /* Head lock still held and bh's disabled */
 348                 inet_bind_hash(sk, tb, port);
 349                 if (sk_unhashed(sk)) {
 350                         inet_sk(sk)->sport = htons(port);
 351                         __inet6_hash(&tcp_hashinfo, sk);
 352                 }
 353                 spin_unlock(&head->lock);
 354
 355                 if (tw) {
 356                         inet_twsk_deschedule(tw, &tcp_death_row);
 357                         inet_twsk_put(tw);
 358                 }
 359
 360                 ret = 0;
 361                 goto out;
 362         }
 363
 364         head = &tcp_hashinfo.bhash[inet_bhashfn(snum, tcp_hashinfo.bhash_size)];
 365         tb   = inet_csk(sk)->icsk_bind_hash;
 366         spin_lock_bh(&head->lock);
 367
 368         if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
 369                 __inet6_hash(&tcp_hashinfo, sk);
 370                 spin_unlock_bh(&head->lock);
 371                 return 0;
 372         } else {
 373                 spin_unlock(&head->lock);
 374                 /* No definite answer... Walk to established hash table */
 375                 ret = __tcp_v6_check_established(sk, snum, NULL);
 376 out:
 377                 local_bh_enable();
 378                 return ret;
 379         }
 380 }
 381
 382 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 383                           int addr_len)
 384 {
 385         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 386         struct inet_sock *inet = inet_sk(sk);
 387         struct ipv6_pinfo *np = inet6_sk(sk);
 388         struct tcp_sock *tp = tcp_sk(sk);
 389         struct in6_addr *saddr = NULL, *final_p = NULL, final;
 390         struct flowi fl;
 391         struct dst_entry *dst;
 392         int addr_type;
 393         int err;
 394
 395         if (addr_len < SIN6_LEN_RFC2133)
 396                 return -EINVAL;
 397
 398         if (usin->sin6_family != AF_INET6)
 399                 return(-EAFNOSUPPORT);
 400
 401         memset(&fl, 0, sizeof(fl));
 402
 403         if (np->sndflow) {
 404                 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 405                 IP6_ECN_flow_init(fl.fl6_flowlabel);
 406                 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
 407                         struct ip6_flowlabel *flowlabel;
 408                         flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
 409                         if (flowlabel == NULL)
 410                                 return -EINVAL;
 411                         ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
 412                         fl6_sock_release(flowlabel);
 413                 }
 414         }
 415
 416         /*
 417          *      connect() to INADDR_ANY means loopback (BSD'ism).
 418          */
 419
 420         if(ipv6_addr_any(&usin->sin6_addr))
 421                 usin->sin6_addr.s6_addr[15] = 0x1;
 422
 423         addr_type = ipv6_addr_type(&usin->sin6_addr);
 424
 425         if(addr_type & IPV6_ADDR_MULTICAST)
 426                 return -ENETUNREACH;
 427
 428         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 429                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 430                     usin->sin6_scope_id) {
 431                         /* If interface is set while binding, indices
 432                          * must coincide.
 433                          */
 434                         if (sk->sk_bound_dev_if &&
 435                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 436                                 return -EINVAL;
 437
 438                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 439                 }
 440
 441                 /* Connect to link-local address requires an interface */
 442                 if (!sk->sk_bound_dev_if)
 443                         return -EINVAL;
 444         }
 445
 446         if (tp->rx_opt.ts_recent_stamp &&
 447             !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
 448                 tp->rx_opt.ts_recent = 0;
 449                 tp->rx_opt.ts_recent_stamp = 0;
 450                 tp->write_seq = 0;
 451         }
 452
 453         ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
 454         np->flow_label = fl.fl6_flowlabel;
 455
 456         /*
 457          *      TCP over IPv4
 458          */
 459
 460         if (addr_type == IPV6_ADDR_MAPPED) {
 461                 u32 exthdrlen = tp->ext_header_len;
 462                 struct sockaddr_in sin;
 463
 464                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 465
 466                 if (__ipv6_only_sock(sk))
 467                         return -ENETUNREACH;
 468
 469                 sin.sin_family = AF_INET;
 470                 sin.sin_port = usin->sin6_port;
 471                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 472
 473                 tp->af_specific = &ipv6_mapped;
 474                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 475
 476                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 477
 478                 if (err) {
 479                         tp->ext_header_len = exthdrlen;
 480                         tp->af_specific = &ipv6_specific;
 481                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 482                         goto failure;
 483                 } else {
 484                         ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
 485                                       inet->saddr);
 486                         ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
 487                                       inet->rcv_saddr);
 488                 }
 489
 490                 return err;
 491         }
 492
 493         if (!ipv6_addr_any(&np->rcv_saddr))
 494                 saddr = &np->rcv_saddr;
 495
 496         fl.proto = IPPROTO_TCP;
 497         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 498         ipv6_addr_copy(&fl.fl6_src,
 499                        (saddr ? saddr : &np->saddr));
 500         fl.oif = sk->sk_bound_dev_if;
 501         fl.fl_ip_dport = usin->sin6_port;
 502         fl.fl_ip_sport = inet->sport;
 503
 504         if (np->opt && np->opt->srcrt) {
 505                 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
 506                 ipv6_addr_copy(&final, &fl.fl6_dst);
 507                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 508                 final_p = &final;
 509         }
 510
 511         err = ip6_dst_lookup(sk, &dst, &fl);
 512         if (err)
 513                 goto failure;
 514         if (final_p)
 515                 ipv6_addr_copy(&fl.fl6_dst, final_p);
 516
 517         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
 518                 goto failure;
 519
 520         if (saddr == NULL) {
 521                 saddr = &fl.fl6_src;
 522                 ipv6_addr_copy(&np->rcv_saddr, saddr);
 523         }
 524
 525         /* set the source address */
 526         ipv6_addr_copy(&np->saddr, saddr);
 527         inet->rcv_saddr = LOOPBACK4_IPV6;
 528
 529         ip6_dst_store(sk, dst, NULL);
 530         sk->sk_route_caps = dst->dev->features &
 531                 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
 532
 533         tp->ext_header_len = 0;
 534         if (np->opt)
 535                 tp->ext_header_len = np->opt->opt_flen + np->opt->opt_nflen;
 536
 537         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 538
 539         inet->dport = usin->sin6_port;
 540
 541         tcp_set_state(sk, TCP_SYN_SENT);
 542         err = tcp_v6_hash_connect(sk);
 543         if (err)
 544                 goto late_failure;
 545
 546         if (!tp->write_seq)
 547                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 548                                                              np->daddr.s6_addr32,
 549                                                              inet->sport,
 550                                                              inet->dport);
 551
 552         err = tcp_connect(sk);
 553         if (err)
 554                 goto late_failure;
 555
 556         return 0;
 557
 558 late_failure:
 559         tcp_set_state(sk, TCP_CLOSE);
 560         __sk_dst_reset(sk);
 561 failure:
 562         inet->dport = 0;
 563         sk->sk_route_caps = 0;
 564         return err;
 565 }
 566
 567 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 568                 int type, int code, int offset, __u32 info)
 569 {
 570         struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
 571         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 572         struct ipv6_pinfo *np;
 573         struct sock *sk;
 574         int err;
 575         struct tcp_sock *tp;
 576         __u32 seq;
 577
 578         sk = inet6_lookup(&tcp_hashinfo, &hdr->daddr, th->dest, &hdr->saddr,
 579                           th->source, skb->dev->ifindex);
 580
 581         if (sk == NULL) {
 582                 ICMP6_INC_STATS_BH(__in6_dev_get(skb->dev), ICMP6_MIB_INERRORS);
 583                 return;
 584         }
 585
 586         if (sk->sk_state == TCP_TIME_WAIT) {
 587                 inet_twsk_put((struct inet_timewait_sock *)sk);
 588                 return;
 589         }
 590
 591         bh_lock_sock(sk);
 592         if (sock_owned_by_user(sk))
 593                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 594
 595         if (sk->sk_state == TCP_CLOSE)
 596                 goto out;
 597
 598         tp = tcp_sk(sk);
 599         seq = ntohl(th->seq);
 600         if (sk->sk_state != TCP_LISTEN &&
 601             !between(seq, tp->snd_una, tp->snd_nxt)) {
 602                 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 603                 goto out;
 604         }
 605
 606         np = inet6_sk(sk);
 607
 608         if (type == ICMPV6_PKT_TOOBIG) {
 609                 struct dst_entry *dst = NULL;
 610
 611                 if (sock_owned_by_user(sk))
 612                         goto out;
 613                 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 614                         goto out;
 615
 616                 /* icmp should have updated the destination cache entry */
 617                 dst = __sk_dst_check(sk, np->dst_cookie);
 618
 619                 if (dst == NULL) {
 620                         struct inet_sock *inet = inet_sk(sk);
 621                         struct flowi fl;
 622
 623                         /* BUGGG_FUTURE: Again, it is not clear how
 624                            to handle rthdr case. Ignore this complexity
 625                            for now.
 626                          */
 627                         memset(&fl, 0, sizeof(fl));
 628                         fl.proto = IPPROTO_TCP;
 629                         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 630                         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
 631                         fl.oif = sk->sk_bound_dev_if;
 632                         fl.fl_ip_dport = inet->dport;
 633                         fl.fl_ip_sport = inet->sport;
 634
 635                         if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
 636                                 sk->sk_err_soft = -err;
 637                                 goto out;
 638                         }
 639
 640                         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
 641                                 sk->sk_err_soft = -err;
 642                                 goto out;
 643                         }
 644
 645                 } else
 646                         dst_hold(dst);
 647
 648                 if (tp->pmtu_cookie > dst_mtu(dst)) {
 649                         tcp_sync_mss(sk, dst_mtu(dst));
 650                         tcp_simple_retransmit(sk);
 651                 } /* else let the usual retransmit timer handle it */
 652                 dst_release(dst);
 653                 goto out;
 654         }
 655
 656         icmpv6_err_convert(type, code, &err);
 657
 658         /* Might be for an request_sock */
 659         switch (sk->sk_state) {
 660                 struct request_sock *req, **prev;
 661         case TCP_LISTEN:
 662                 if (sock_owned_by_user(sk))
 663                         goto out;
 664
 665                 req = tcp_v6_search_req(sk, &prev, th->dest, &hdr->daddr,
 666                                         &hdr->saddr, inet6_iif(skb));
 667                 if (!req)
 668                         goto out;
 669
 670                 /* ICMPs are not backlogged, hence we cannot get
 671                  * an established socket here.
 672                  */
 673                 BUG_TRAP(req->sk == NULL);
 674
 675                 if (seq != tcp_rsk(req)->snt_isn) {
 676                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 677                         goto out;
 678                 }
 679
 680                 inet_csk_reqsk_queue_drop(sk, req, prev);
 681                 goto out;
 682
 683         case TCP_SYN_SENT:
 684         case TCP_SYN_RECV:  /* Cannot happen.
 685                                It can, it SYNs are crossed. --ANK */
 686                 if (!sock_owned_by_user(sk)) {
 687                         TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
 688                         sk->sk_err = err;
 689                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 690
 691                         tcp_done(sk);
 692                 } else
 693                         sk->sk_err_soft = err;
 694                 goto out;
 695         }
 696
 697         if (!sock_owned_by_user(sk) && np->recverr) {
 698                 sk->sk_err = err;
 699                 sk->sk_error_report(sk);
 700         } else
 701                 sk->sk_err_soft = err;
 702
 703 out:
 704         bh_unlock_sock(sk);
 705         sock_put(sk);
 706 }
 707
 708
 709 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
 710                               struct dst_entry *dst)
 711 {
 712         struct tcp6_request_sock *treq = tcp6_rsk(req);
 713         struct ipv6_pinfo *np = inet6_sk(sk);
 714         struct sk_buff * skb;
 715         struct ipv6_txoptions *opt = NULL;
 716         struct in6_addr * final_p = NULL, final;
 717         struct flowi fl;
 718         int err = -1;
 719
 720         memset(&fl, 0, sizeof(fl));
 721         fl.proto = IPPROTO_TCP;
 722         ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
 723         ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
 724         fl.fl6_flowlabel = 0;
 725         fl.oif = treq->iif;
 726         fl.fl_ip_dport = inet_rsk(req)->rmt_port;
 727         fl.fl_ip_sport = inet_sk(sk)->sport;
 728
 729         if (dst == NULL) {
 730                 opt = np->opt;
 731                 if (opt == NULL &&
 732                     np->rxopt.bits.osrcrt == 2 &&
 733                     treq->pktopts) {
 734                         struct sk_buff *pktopts = treq->pktopts;
 735                         struct inet6_skb_parm *rxopt = IP6CB(pktopts);
 736                         if (rxopt->srcrt)
 737                                 opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(pktopts->nh.raw + rxopt->srcrt));
 738                 }
 739
 740                 if (opt && opt->srcrt) {
 741                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
 742                         ipv6_addr_copy(&final, &fl.fl6_dst);
 743                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 744                         final_p = &final;
 745                 }
 746
 747                 err = ip6_dst_lookup(sk, &dst, &fl);
 748                 if (err)
 749                         goto done;
 750                 if (final_p)
 751                         ipv6_addr_copy(&fl.fl6_dst, final_p);
 752                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
 753                         goto done;
 754         }
 755
 756         skb = tcp_make_synack(sk, dst, req);
 757         if (skb) {
 758                 struct tcphdr *th = skb->h.th;
 759
 760                 th->check = tcp_v6_check(th, skb->len,
 761                                          &treq->loc_addr, &treq->rmt_addr,
 762                                          csum_partial((char *)th, skb->len, skb->csum));
 763
 764                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
 765                 err = ip6_xmit(sk, skb, &fl, opt, 0);
 766                 if (err == NET_XMIT_CN)
 767                         err = 0;
 768         }
 769
 770 done:
 771         if (opt && opt != np->opt)
 772                 sock_kfree_s(sk, opt, opt->tot_len);
 773         return err;
 774 }
 775
 776 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 777 {
 778         if (tcp6_rsk(req)->pktopts)
 779                 kfree_skb(tcp6_rsk(req)->pktopts);
 780 }
 781
 782 static struct request_sock_ops tcp6_request_sock_ops = {
 783         .family         =       AF_INET6,
 784         .obj_size       =       sizeof(struct tcp6_request_sock),
 785         .rtx_syn_ack    =       tcp_v6_send_synack,
 786         .send_ack       =       tcp_v6_reqsk_send_ack,
 787         .destructor     =       tcp_v6_reqsk_destructor,
 788         .send_reset     =       tcp_v6_send_reset
 789 };
 790
 791 static int ipv6_opt_accepted(struct sock *sk, struct sk_buff *skb)
 792 {
 793         struct ipv6_pinfo *np = inet6_sk(sk);
 794         struct inet6_skb_parm *opt = IP6CB(skb);
 795
 796         if (np->rxopt.all) {
 797                 if ((opt->hop && (np->rxopt.bits.hopopts || np->rxopt.bits.ohopopts)) ||
 798                     ((IPV6_FLOWINFO_MASK & *(u32*)skb->nh.raw) && np->rxopt.bits.rxflow) ||
 799                     (opt->srcrt && (np->rxopt.bits.srcrt || np->rxopt.bits.osrcrt)) ||
 800                     ((opt->dst1 || opt->dst0) && (np->rxopt.bits.dstopts || np->rxopt.bits.odstopts)))
 801                         return 1;
 802         }
 803         return 0;
 804 }
 805
 806
 807 static void tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len,
 808                               struct sk_buff *skb)
 809 {
 810         struct ipv6_pinfo *np = inet6_sk(sk);
 811
 812         if (skb->ip_summed == CHECKSUM_HW) {
 813                 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,  0);
 814                 skb->csum = offsetof(struct tcphdr, check);
 815         } else {
 816                 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
 817                                             csum_partial((char *)th, th->doff<<2,
 818                                                          skb->csum));
 819         }
 820 }
 821
 822
 823 static void tcp_v6_send_reset(struct sk_buff *skb)
 824 {
 825         struct tcphdr *th = skb->h.th, *t1;
 826         struct sk_buff *buff;
 827         struct flowi fl;
 828
 829         if (th->rst)
 830                 return;
 831
 832         if (!ipv6_unicast_destination(skb))
 833                 return;
 834
 835         /*
 836          * We need to grab some memory, and put together an RST,
 837          * and then put it into the queue to be sent.
 838          */
 839
 840         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + sizeof(struct tcphdr),
 841                          GFP_ATOMIC);
 842         if (buff == NULL)
 843                 return;
 844
 845         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + sizeof(struct tcphdr));
 846
 847         t1 = (struct tcphdr *) skb_push(buff,sizeof(struct tcphdr));
 848
 849         /* Swap the send and the receive. */
 850         memset(t1, 0, sizeof(*t1));
 851         t1->dest = th->source;
 852         t1->source = th->dest;
 853         t1->doff = sizeof(*t1)/4;
 854         t1->rst = 1;
 855
 856         if(th->ack) {
 857                 t1->seq = th->ack_seq;
 858         } else {
 859                 t1->ack = 1;
 860                 t1->ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin
 861                                     + skb->len - (th->doff<<2));
 862         }
 863
 864         buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);
 865
 866         memset(&fl, 0, sizeof(fl));
 867         ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
 868         ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
 869
 870         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
 871                                     sizeof(*t1), IPPROTO_TCP,
 872                                     buff->csum);
 873
 874         fl.proto = IPPROTO_TCP;
 875         fl.oif = inet6_iif(skb);
 876         fl.fl_ip_dport = t1->dest;
 877         fl.fl_ip_sport = t1->source;
 878
 879         /* sk = NULL, but it is safe for now. RST socket required. */
 880         if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
 881
 882                 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
 883                         ip6_xmit(NULL, buff, &fl, NULL, 0);
 884                         TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
 885                         TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
 886                         return;
 887                 }
 888         }
 889
 890         kfree_skb(buff);
 891 }
 892
 893 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts)
 894 {
 895         struct tcphdr *th = skb->h.th, *t1;
 896         struct sk_buff *buff;
 897         struct flowi fl;
 898         int tot_len = sizeof(struct tcphdr);
 899
 900         if (ts)
 901                 tot_len += 3*4;
 902
 903         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 904                          GFP_ATOMIC);
 905         if (buff == NULL)
 906                 return;
 907
 908         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 909
 910         t1 = (struct tcphdr *) skb_push(buff,tot_len);
 911
 912         /* Swap the send and the receive. */
 913         memset(t1, 0, sizeof(*t1));
 914         t1->dest = th->source;
 915         t1->source = th->dest;
 916         t1->doff = tot_len/4;
 917         t1->seq = htonl(seq);
 918         t1->ack_seq = htonl(ack);
 919         t1->ack = 1;
 920         t1->window = htons(win);
 921
 922         if (ts) {
 923                 u32 *ptr = (u32*)(t1 + 1);
 924                 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 925                                (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 926                 *ptr++ = htonl(tcp_time_stamp);
 927                 *ptr = htonl(ts);
 928         }
 929
 930         buff->csum = csum_partial((char *)t1, tot_len, 0);
 931
 932         memset(&fl, 0, sizeof(fl));
 933         ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
 934         ipv6_addr_copy(&fl.fl6_src, &skb->nh.ipv6h->daddr);
 935
 936         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
 937                                     tot_len, IPPROTO_TCP,
 938                                     buff->csum);
 939
 940         fl.proto = IPPROTO_TCP;
 941         fl.oif = inet6_iif(skb);
 942         fl.fl_ip_dport = t1->dest;
 943         fl.fl_ip_sport = t1->source;
 944
 945         if (!ip6_dst_lookup(NULL, &buff->dst, &fl)) {
 946                 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
 947                         ip6_xmit(NULL, buff, &fl, NULL, 0);
 948                         TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
 949                         return;
 950                 }
 951         }
 952
 953         kfree_skb(buff);
 954 }
 955
 956 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 957 {
 958         struct inet_timewait_sock *tw = inet_twsk(sk);
 959         const struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 960
 961         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 962                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 963                         tcptw->tw_ts_recent);
 964
 965         inet_twsk_put(tw);
 966 }
 967
 968 static void tcp_v6_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
 969 {
 970         tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent);
 971 }
 972
 973
 974 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
 975 {
 976         struct request_sock *req, **prev;
 977         const struct tcphdr *th = skb->h.th;
 978         struct sock *nsk;
 979
 980         /* Find possible connection requests. */
 981         req = tcp_v6_search_req(sk, &prev, th->source, &skb->nh.ipv6h->saddr,
 982                                 &skb->nh.ipv6h->daddr, inet6_iif(skb));
 983         if (req)
 984                 return tcp_check_req(sk, skb, req, prev);
 985
 986         nsk = __inet6_lookup_established(&tcp_hashinfo, &skb->nh.ipv6h->saddr,
 987                                          th->source, &skb->nh.ipv6h->daddr,
 988                                          ntohs(th->dest), inet6_iif(skb));
 989
 990         if (nsk) {
 991                 if (nsk->sk_state != TCP_TIME_WAIT) {
 992                         bh_lock_sock(nsk);
 993                         return nsk;
 994                 }
 995                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 996                 return NULL;
 997         }
 998
 999 #if 0 /*def CONFIG_SYN_COOKIES*/
1000         if (!th->rst && !th->syn && th->ack)
1001                 sk = cookie_v6_check(sk, skb, &(IPCB(skb)->opt));
1002 #endif
1003         return sk;
1004 }
1005
1006 static void tcp_v6_synq_add(struct sock *sk, struct request_sock *req)
1007 {
1008         struct inet_connection_sock *icsk = inet_csk(sk);
1009         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
1010         const u32 h = tcp_v6_synq_hash(&tcp6_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port, lopt->hash_rnd);
1011
1012         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, TCP_TIMEOUT_INIT);
1013         inet_csk_reqsk_queue_added(sk, TCP_TIMEOUT_INIT);
1014 }
1015
1016
1017 /* FIXME: this is substantially similar to the ipv4 code.
1018  * Can some kind of merge be done? -- erics
1019  */
1020 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1021 {
1022         struct tcp6_request_sock *treq;
1023         struct ipv6_pinfo *np = inet6_sk(sk);
1024         struct tcp_options_received tmp_opt;
1025         struct tcp_sock *tp = tcp_sk(sk);
1026         struct request_sock *req = NULL;
1027         __u32 isn = TCP_SKB_CB(skb)->when;
1028
1029         if (skb->protocol == htons(ETH_P_IP))
1030                 return tcp_v4_conn_request(sk, skb);
1031
1032         if (!ipv6_unicast_destination(skb))
1033                 goto drop;
1034
1035         /*
1036          *      There are no SYN attacks on IPv6, yet...
1037          */
1038         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1039                 if (net_ratelimit())
1040                         printk(KERN_INFO "TCPv6: dropping request, synflood is possible\n");
1041                 goto drop;
1042         }
1043
1044         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1045                 goto drop;
1046
1047         req = reqsk_alloc(&tcp6_request_sock_ops);
1048         if (req == NULL)
1049                 goto drop;
1050
1051         tcp_clear_options(&tmp_opt);
1052         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1053         tmp_opt.user_mss = tp->rx_opt.user_mss;
1054
1055         tcp_parse_options(skb, &tmp_opt, 0);
1056
1057         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1058         tcp_openreq_init(req, &tmp_opt, skb);
1059
1060         treq = tcp6_rsk(req);
1061         ipv6_addr_copy(&treq->rmt_addr, &skb->nh.ipv6h->saddr);
1062         ipv6_addr_copy(&treq->loc_addr, &skb->nh.ipv6h->daddr);
1063         TCP_ECN_create_request(req, skb->h.th);
1064         treq->pktopts = NULL;
1065         if (ipv6_opt_accepted(sk, skb) ||
1066             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1067             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1068                 atomic_inc(&skb->users);
1069                 treq->pktopts = skb;
1070         }
1071         treq->iif = sk->sk_bound_dev_if;
1072
1073         /* So that link locals have meaning */
1074         if (!sk->sk_bound_dev_if &&
1075             ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1076                 treq->iif = inet6_iif(skb);
1077
1078         if (isn == 0)
1079                 isn = tcp_v6_init_sequence(sk,skb);
1080
1081         tcp_rsk(req)->snt_isn = isn;
1082
1083         if (tcp_v6_send_synack(sk, req, NULL))
1084                 goto drop;
1085
1086         tcp_v6_synq_add(sk, req);
1087
1088         return 0;
1089
1090 drop:
1091         if (req)
1092                 reqsk_free(req);
1093
1094         TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
1095         return 0; /* don't send reset */
1096 }
1097
1098 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1099                                           struct request_sock *req,
1100                                           struct dst_entry *dst)
1101 {
1102         struct tcp6_request_sock *treq = tcp6_rsk(req);
1103         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1104         struct tcp6_sock *newtcp6sk;
1105         struct inet_sock *newinet;
1106         struct tcp_sock *newtp;
1107         struct sock *newsk;
1108         struct ipv6_txoptions *opt;
1109
1110         if (skb->protocol == htons(ETH_P_IP)) {
1111                 /*
1112                  *      v6 mapped
1113                  */
1114
1115                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1116
1117                 if (newsk == NULL)
1118                         return NULL;
1119
1120                 newtcp6sk = (struct tcp6_sock *)newsk;
1121                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1122
1123                 newinet = inet_sk(newsk);
1124                 newnp = inet6_sk(newsk);
1125                 newtp = tcp_sk(newsk);
1126
1127                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1128
1129                 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1130                               newinet->daddr);
1131
1132                 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1133                               newinet->saddr);
1134
1135                 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1136
1137                 newtp->af_specific = &ipv6_mapped;
1138                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1139                 newnp->pktoptions  = NULL;
1140                 newnp->opt         = NULL;
1141                 newnp->mcast_oif   = inet6_iif(skb);
1142                 newnp->mcast_hops  = skb->nh.ipv6h->hop_limit;
1143
1144                 /*
1145                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1146                  * here, tcp_create_openreq_child now does this for us, see the comment in
1147                  * that function for the gory details. -acme
1148                  */
1149
1150                 /* It is tricky place. Until this moment IPv4 tcp
1151                    worked with IPv6 af_tcp.af_specific.
1152                    Sync it now.
1153                  */
1154                 tcp_sync_mss(newsk, newtp->pmtu_cookie);
1155
1156                 return newsk;
1157         }
1158
1159         opt = np->opt;
1160
1161         if (sk_acceptq_is_full(sk))
1162                 goto out_overflow;
1163
1164         if (np->rxopt.bits.osrcrt == 2 &&
1165             opt == NULL && treq->pktopts) {
1166                 struct inet6_skb_parm *rxopt = IP6CB(treq->pktopts);
1167                 if (rxopt->srcrt)
1168                         opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr *)(treq->pktopts->nh.raw + rxopt->srcrt));
1169         }
1170
1171         if (dst == NULL) {
1172                 struct in6_addr *final_p = NULL, final;
1173                 struct flowi fl;
1174
1175                 memset(&fl, 0, sizeof(fl));
1176                 fl.proto = IPPROTO_TCP;
1177                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1178                 if (opt && opt->srcrt) {
1179                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1180                         ipv6_addr_copy(&final, &fl.fl6_dst);
1181                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1182                         final_p = &final;
1183                 }
1184                 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1185                 fl.oif = sk->sk_bound_dev_if;
1186                 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1187                 fl.fl_ip_sport = inet_sk(sk)->sport;
1188
1189                 if (ip6_dst_lookup(sk, &dst, &fl))
1190                         goto out;
1191
1192                 if (final_p)
1193                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1194
1195                 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1196                         goto out;
1197         }
1198
1199         newsk = tcp_create_openreq_child(sk, req, skb);
1200         if (newsk == NULL)
1201                 goto out;
1202
1203         /*
1204          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1205          * count here, tcp_create_openreq_child now does this for us, see the
1206          * comment in that function for the gory details. -acme
1207          */
1208
1209         ip6_dst_store(newsk, dst, NULL);
1210         newsk->sk_route_caps = dst->dev->features &
1211                 ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1212
1213         newtcp6sk = (struct tcp6_sock *)newsk;
1214         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1215
1216         newtp = tcp_sk(newsk);
1217         newinet = inet_sk(newsk);
1218         newnp = inet6_sk(newsk);
1219
1220         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1221
1222         ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1223         ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1224         ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1225         newsk->sk_bound_dev_if = treq->iif;
1226
1227         /* Now IPv6 options...
1228
1229            First: no IPv4 options.
1230          */
1231         newinet->opt = NULL;
1232
1233         /* Clone RX bits */
1234         newnp->rxopt.all = np->rxopt.all;
1235
1236         /* Clone pktoptions received with SYN */
1237         newnp->pktoptions = NULL;
1238         if (treq->pktopts != NULL) {
1239                 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1240                 kfree_skb(treq->pktopts);
1241                 treq->pktopts = NULL;
1242                 if (newnp->pktoptions)
1243                         skb_set_owner_r(newnp->pktoptions, newsk);
1244         }
1245         newnp->opt        = NULL;
1246         newnp->mcast_oif  = inet6_iif(skb);
1247         newnp->mcast_hops = skb->nh.ipv6h->hop_limit;
1248
1249         /* Clone native IPv6 options from listening socket (if any)
1250
1251            Yes, keeping reference count would be much more clever,
1252            but we make one more one thing there: reattach optmem
1253            to newsk.
1254          */
1255         if (opt) {
1256                 newnp->opt = ipv6_dup_options(newsk, opt);
1257                 if (opt != np->opt)
1258                         sock_kfree_s(sk, opt, opt->tot_len);
1259         }
1260
1261         newtp->ext_header_len = 0;
1262         if (newnp->opt)
1263                 newtp->ext_header_len = newnp->opt->opt_nflen +
1264                                         newnp->opt->opt_flen;
1265
1266         tcp_sync_mss(newsk, dst_mtu(dst));
1267         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1268         tcp_initialize_rcv_mss(newsk);
1269
1270         newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1271
1272         __inet6_hash(&tcp_hashinfo, newsk);
1273         inet_inherit_port(&tcp_hashinfo, sk, newsk);
1274
1275         return newsk;
1276
1277 out_overflow:
1278         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1279 out:
1280         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1281         if (opt && opt != np->opt)
1282                 sock_kfree_s(sk, opt, opt->tot_len);
1283         dst_release(dst);
1284         return NULL;
1285 }
1286
1287 static int tcp_v6_checksum_init(struct sk_buff *skb)
1288 {
1289         if (skb->ip_summed == CHECKSUM_HW) {
1290                 if (!tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1291                                   &skb->nh.ipv6h->daddr,skb->csum)) {
1292                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1293                         return 0;
1294                 }
1295         }
1296
1297         skb->csum = ~tcp_v6_check(skb->h.th,skb->len,&skb->nh.ipv6h->saddr,
1298                                   &skb->nh.ipv6h->daddr, 0);
1299
1300         if (skb->len <= 76) {
1301                 return __skb_checksum_complete(skb);
1302         }
1303         return 0;
1304 }
1305
1306 /* The socket must have it's spinlock held when we get
1307  * here.
1308  *
1309  * We have a potential double-lock case here, so even when
1310  * doing backlog processing we use the BH locking scheme.
1311  * This is because we cannot sleep with the original spinlock
1312  * held.
1313  */
1314 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1315 {
1316         struct ipv6_pinfo *np = inet6_sk(sk);
1317         struct tcp_sock *tp;
1318         struct sk_buff *opt_skb = NULL;
1319
1320         /* Imagine: socket is IPv6. IPv4 packet arrives,
1321            goes to IPv4 receive handler and backlogged.
1322            From backlog it always goes here. Kerboom...
1323            Fortunately, tcp_rcv_established and rcv_established
1324            handle them correctly, but it is not case with
1325            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1326          */
1327
1328         if (skb->protocol == htons(ETH_P_IP))
1329                 return tcp_v4_do_rcv(sk, skb);
1330
1331         if (sk_filter(sk, skb, 0))
1332                 goto discard;
1333
1334         /*
1335          *      socket locking is here for SMP purposes as backlog rcv
1336          *      is currently called with bh processing disabled.
1337          */
1338
1339         /* Do Stevens' IPV6_PKTOPTIONS.
1340
1341            Yes, guys, it is the only place in our code, where we
1342            may make it not affecting IPv4.
1343            The rest of code is protocol independent,
1344            and I do not like idea to uglify IPv4.
1345
1346            Actually, all the idea behind IPV6_PKTOPTIONS
1347            looks not very well thought. For now we latch
1348            options, received in the last packet, enqueued
1349            by tcp. Feel free to propose better solution.
1350                                                --ANK (980728)
1351          */
1352         if (np->rxopt.all)
1353                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1354
1355         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1356                 TCP_CHECK_TIMER(sk);
1357                 if (tcp_rcv_established(sk, skb, skb->h.th, skb->len))
1358                         goto reset;
1359                 TCP_CHECK_TIMER(sk);
1360                 if (opt_skb)
1361                         goto ipv6_pktoptions;
1362                 return 0;
1363         }
1364
1365         if (skb->len < (skb->h.th->doff<<2) || tcp_checksum_complete(skb))
1366                 goto csum_err;
1367
1368         if (sk->sk_state == TCP_LISTEN) {
1369                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1370                 if (!nsk)
1371                         goto discard;
1372
1373                 /*
1374                  * Queue it on the new socket if the new socket is active,
1375                  * otherwise we just shortcircuit this and continue with
1376                  * the new socket..
1377                  */
1378                 if(nsk != sk) {
1379                         if (tcp_child_process(sk, nsk, skb))
1380                                 goto reset;
1381                         if (opt_skb)
1382                                 __kfree_skb(opt_skb);
1383                         return 0;
1384                 }
1385         }
1386
1387         TCP_CHECK_TIMER(sk);
1388         if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len))
1389                 goto reset;
1390         TCP_CHECK_TIMER(sk);
1391         if (opt_skb)
1392                 goto ipv6_pktoptions;
1393         return 0;
1394
1395 reset:
1396         tcp_v6_send_reset(skb);
1397 discard:
1398         if (opt_skb)
1399                 __kfree_skb(opt_skb);
1400         kfree_skb(skb);
1401         return 0;
1402 csum_err:
1403         TCP_INC_STATS_BH(TCP_MIB_INERRS);
1404         goto discard;
1405
1406
1407 ipv6_pktoptions:
1408         /* Do you ask, what is it?
1409
1410            1. skb was enqueued by tcp.
1411            2. skb is added to tail of read queue, rather than out of order.
1412            3. socket is not in passive state.
1413            4. Finally, it really contains options, which user wants to receive.
1414          */
1415         tp = tcp_sk(sk);
1416         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1417             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1418                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1419                         np->mcast_oif = inet6_iif(opt_skb);
1420                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1421                         np->mcast_hops = opt_skb->nh.ipv6h->hop_limit;
1422                 if (ipv6_opt_accepted(sk, opt_skb)) {
1423                         skb_set_owner_r(opt_skb, sk);
1424                         opt_skb = xchg(&np->pktoptions, opt_skb);
1425                 } else {
1426                         __kfree_skb(opt_skb);
1427                         opt_skb = xchg(&np->pktoptions, NULL);
1428                 }
1429         }
1430
1431         if (opt_skb)
1432                 kfree_skb(opt_skb);
1433         return 0;
1434 }
1435
1436 static int tcp_v6_rcv(struct sk_buff **pskb, unsigned int *nhoffp)
1437 {
1438         struct sk_buff *skb = *pskb;
1439         struct tcphdr *th;
1440         struct sock *sk;
1441         int ret;
1442
1443         if (skb->pkt_type != PACKET_HOST)
1444                 goto discard_it;
1445
1446         /*
1447          *      Count it even if it's bad.
1448          */
1449         TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1450
1451         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1452                 goto discard_it;
1453
1454         th = skb->h.th;
1455
1456         if (th->doff < sizeof(struct tcphdr)/4)
1457                 goto bad_packet;
1458         if (!pskb_may_pull(skb, th->doff*4))
1459                 goto discard_it;
1460
1461         if ((skb->ip_summed != CHECKSUM_UNNECESSARY &&
1462              tcp_v6_checksum_init(skb)))
1463                 goto bad_packet;
1464
1465         th = skb->h.th;
1466         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1467         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1468                                     skb->len - th->doff*4);
1469         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1470         TCP_SKB_CB(skb)->when = 0;
1471         TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(skb->nh.ipv6h);
1472         TCP_SKB_CB(skb)->sacked = 0;
1473
1474         sk = __inet6_lookup(&tcp_hashinfo, &skb->nh.ipv6h->saddr, th->source,
1475                             &skb->nh.ipv6h->daddr, ntohs(th->dest),
1476                             inet6_iif(skb));
1477
1478         if (!sk)
1479                 goto no_tcp_socket;
1480
1481 process:
1482         if (sk->sk_state == TCP_TIME_WAIT)
1483                 goto do_time_wait;
1484
1485         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1486                 goto discard_and_relse;
1487
1488         if (sk_filter(sk, skb, 0))
1489                 goto discard_and_relse;
1490
1491         skb->dev = NULL;
1492
1493         bh_lock_sock(sk);
1494         ret = 0;
1495         if (!sock_owned_by_user(sk)) {
1496                 if (!tcp_prequeue(sk, skb))
1497                         ret = tcp_v6_do_rcv(sk, skb);
1498         } else
1499                 sk_add_backlog(sk, skb);
1500         bh_unlock_sock(sk);
1501
1502         sock_put(sk);
1503         return ret ? -1 : 0;
1504
1505 no_tcp_socket:
1506         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1507                 goto discard_it;
1508
1509         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1510 bad_packet:
1511                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1512         } else {
1513                 tcp_v6_send_reset(skb);
1514         }
1515
1516 discard_it:
1517
1518         /*
1519          *      Discard frame
1520          */
1521
1522         kfree_skb(skb);
1523         return 0;
1524
1525 discard_and_relse:
1526         sock_put(sk);
1527         goto discard_it;
1528
1529 do_time_wait:
1530         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1531                 inet_twsk_put((struct inet_timewait_sock *)sk);
1532                 goto discard_it;
1533         }
1534
1535         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1536                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1537                 inet_twsk_put((struct inet_timewait_sock *)sk);
1538                 goto discard_it;
1539         }
1540
1541         switch (tcp_timewait_state_process((struct inet_timewait_sock *)sk,
1542                                            skb, th)) {
1543         case TCP_TW_SYN:
1544         {
1545                 struct sock *sk2;
1546
1547                 sk2 = inet6_lookup_listener(&tcp_hashinfo,
1548                                             &skb->nh.ipv6h->daddr,
1549                                             ntohs(th->dest), inet6_iif(skb));
1550                 if (sk2 != NULL) {
1551                         struct inet_timewait_sock *tw = inet_twsk(sk);
1552                         inet_twsk_deschedule(tw, &tcp_death_row);
1553                         inet_twsk_put(tw);
1554                         sk = sk2;
1555                         goto process;
1556                 }
1557                 /* Fall through to ACK */
1558         }
1559         case TCP_TW_ACK:
1560                 tcp_v6_timewait_ack(sk, skb);
1561                 break;
1562         case TCP_TW_RST:
1563                 goto no_tcp_socket;
1564         case TCP_TW_SUCCESS:;
1565         }
1566         goto discard_it;
1567 }
1568
1569 static int tcp_v6_rebuild_header(struct sock *sk)
1570 {
1571         int err;
1572         struct dst_entry *dst;
1573         struct ipv6_pinfo *np = inet6_sk(sk);
1574
1575         dst = __sk_dst_check(sk, np->dst_cookie);
1576
1577         if (dst == NULL) {
1578                 struct inet_sock *inet = inet_sk(sk);
1579                 struct in6_addr *final_p = NULL, final;
1580                 struct flowi fl;
1581
1582                 memset(&fl, 0, sizeof(fl));
1583                 fl.proto = IPPROTO_TCP;
1584                 ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1585                 ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1586                 fl.fl6_flowlabel = np->flow_label;
1587                 fl.oif = sk->sk_bound_dev_if;
1588                 fl.fl_ip_dport = inet->dport;
1589                 fl.fl_ip_sport = inet->sport;
1590
1591                 if (np->opt && np->opt->srcrt) {
1592                         struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1593                         ipv6_addr_copy(&final, &fl.fl6_dst);
1594                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1595                         final_p = &final;
1596                 }
1597
1598                 err = ip6_dst_lookup(sk, &dst, &fl);
1599                 if (err) {
1600                         sk->sk_route_caps = 0;
1601                         return err;
1602                 }
1603                 if (final_p)
1604                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1605
1606                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
1607                         sk->sk_err_soft = -err;
1608                         return err;
1609                 }
1610
1611                 ip6_dst_store(sk, dst, NULL);
1612                 sk->sk_route_caps = dst->dev->features &
1613                         ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1614         }
1615
1616         return 0;
1617 }
1618
1619 static int tcp_v6_xmit(struct sk_buff *skb, int ipfragok)
1620 {
1621         struct sock *sk = skb->sk;
1622         struct inet_sock *inet = inet_sk(sk);
1623         struct ipv6_pinfo *np = inet6_sk(sk);
1624         struct flowi fl;
1625         struct dst_entry *dst;
1626         struct in6_addr *final_p = NULL, final;
1627
1628         memset(&fl, 0, sizeof(fl));
1629         fl.proto = IPPROTO_TCP;
1630         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1631         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
1632         fl.fl6_flowlabel = np->flow_label;
1633         IP6_ECN_flow_xmit(sk, fl.fl6_flowlabel);
1634         fl.oif = sk->sk_bound_dev_if;
1635         fl.fl_ip_sport = inet->sport;
1636         fl.fl_ip_dport = inet->dport;
1637
1638         if (np->opt && np->opt->srcrt) {
1639                 struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
1640                 ipv6_addr_copy(&final, &fl.fl6_dst);
1641                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1642                 final_p = &final;
1643         }
1644
1645         dst = __sk_dst_check(sk, np->dst_cookie);
1646
1647         if (dst == NULL) {
1648                 int err = ip6_dst_lookup(sk, &dst, &fl);
1649
1650                 if (err) {
1651                         sk->sk_err_soft = -err;
1652                         return err;
1653                 }
1654
1655                 if (final_p)
1656                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1657
1658                 if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
1659                         sk->sk_route_caps = 0;
1660                         return err;
1661                 }
1662
1663                 ip6_dst_store(sk, dst, NULL);
1664                 sk->sk_route_caps = dst->dev->features &
1665                         ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
1666         }
1667
1668         skb->dst = dst_clone(dst);
1669
1670         /* Restore final destination back after routing done */
1671         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
1672
1673         return ip6_xmit(sk, skb, &fl, np->opt, 0);
1674 }
1675
1676 static void v6_addr2sockaddr(struct sock *sk, struct sockaddr * uaddr)
1677 {
1678         struct ipv6_pinfo *np = inet6_sk(sk);
1679         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) uaddr;
1680
1681         sin6->sin6_family = AF_INET6;
1682         ipv6_addr_copy(&sin6->sin6_addr, &np->daddr);
1683         sin6->sin6_port = inet_sk(sk)->dport;
1684         /* We do not store received flowlabel for TCP */
1685         sin6->sin6_flowinfo = 0;
1686         sin6->sin6_scope_id = 0;
1687         if (sk->sk_bound_dev_if &&
1688             ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
1689                 sin6->sin6_scope_id = sk->sk_bound_dev_if;
1690 }
1691
1692 static int tcp_v6_remember_stamp(struct sock *sk)
1693 {
1694         /* Alas, not yet... */
1695         return 0;
1696 }
1697
1698 static struct tcp_func ipv6_specific = {
1699         .queue_xmit     =       tcp_v6_xmit,
1700         .send_check     =       tcp_v6_send_check,
1701         .rebuild_header =       tcp_v6_rebuild_header,
1702         .conn_request   =       tcp_v6_conn_request,
1703         .syn_recv_sock  =       tcp_v6_syn_recv_sock,
1704         .remember_stamp =       tcp_v6_remember_stamp,
1705         .net_header_len =       sizeof(struct ipv6hdr),
1706
1707         .setsockopt     =       ipv6_setsockopt,
1708         .getsockopt     =       ipv6_getsockopt,
1709         .addr2sockaddr  =       v6_addr2sockaddr,
1710         .sockaddr_len   =       sizeof(struct sockaddr_in6)
1711 };
1712
1713 /*
1714  *      TCP over IPv4 via INET6 API
1715  */
1716
1717 static struct tcp_func ipv6_mapped = {
1718         .queue_xmit     =       ip_queue_xmit,
1719         .send_check     =       tcp_v4_send_check,
1720         .rebuild_header =       inet_sk_rebuild_header,
1721         .conn_request   =       tcp_v6_conn_request,
1722         .syn_recv_sock  =       tcp_v6_syn_recv_sock,
1723         .remember_stamp =       tcp_v4_remember_stamp,
1724         .net_header_len =       sizeof(struct iphdr),
1725
1726         .setsockopt     =       ipv6_setsockopt,
1727         .getsockopt     =       ipv6_getsockopt,
1728         .addr2sockaddr  =       v6_addr2sockaddr,
1729         .sockaddr_len   =       sizeof(struct sockaddr_in6)
1730 };
1731
1732
1733
1734 /* NOTE: A lot of things set to zero explicitly by call to
1735  *       sk_alloc() so need not be done here.
1736  */
1737 static int tcp_v6_init_sock(struct sock *sk)
1738 {
1739         struct inet_connection_sock *icsk = inet_csk(sk);
1740         struct tcp_sock *tp = tcp_sk(sk);
1741
1742         skb_queue_head_init(&tp->out_of_order_queue);
1743         tcp_init_xmit_timers(sk);
1744         tcp_prequeue_init(tp);
1745
1746         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1747         tp->mdev = TCP_TIMEOUT_INIT;
1748
1749         /* So many TCP implementations out there (incorrectly) count the
1750          * initial SYN frame in their delayed-ACK and congestion control
1751          * algorithms that we must have the following bandaid to talk
1752          * efficiently to them.  -DaveM
1753          */
1754         tp->snd_cwnd = 2;
1755
1756         /* See draft-stevens-tcpca-spec-01 for discussion of the
1757          * initialization of these values.
1758          */
1759         tp->snd_ssthresh = 0x7fffffff;
1760         tp->snd_cwnd_clamp = ~0;
1761         tp->mss_cache = 536;
1762
1763         tp->reordering = sysctl_tcp_reordering;
1764
1765         sk->sk_state = TCP_CLOSE;
1766
1767         tp->af_specific = &ipv6_specific;
1768         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1769         sk->sk_write_space = sk_stream_write_space;
1770         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1771
1772         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1773         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1774
1775         atomic_inc(&tcp_sockets_allocated);
1776
1777         return 0;
1778 }
1779
1780 static int tcp_v6_destroy_sock(struct sock *sk)
1781 {
1782         tcp_v4_destroy_sock(sk);
1783         return inet6_destroy_sock(sk);
1784 }
1785
1786 /* Proc filesystem TCPv6 sock list dumping. */
1787 static void get_openreq6(struct seq_file *seq,
1788                          struct sock *sk, struct request_sock *req, int i, int uid)
1789 {
1790         struct in6_addr *dest, *src;
1791         int ttd = req->expires - jiffies;
1792
1793         if (ttd < 0)
1794                 ttd = 0;
1795
1796         src = &tcp6_rsk(req)->loc_addr;
1797         dest = &tcp6_rsk(req)->rmt_addr;
1798         seq_printf(seq,
1799                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1800                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1801                    i,
1802                    src->s6_addr32[0], src->s6_addr32[1],
1803                    src->s6_addr32[2], src->s6_addr32[3],
1804                    ntohs(inet_sk(sk)->sport),
1805                    dest->s6_addr32[0], dest->s6_addr32[1],
1806                    dest->s6_addr32[2], dest->s6_addr32[3],
1807                    ntohs(inet_rsk(req)->rmt_port),
1808                    TCP_SYN_RECV,
1809                    0,0, /* could print option size, but that is af dependent. */
1810                    1,   /* timers active (only the expire timer) */
1811                    jiffies_to_clock_t(ttd),
1812                    req->retrans,
1813                    uid,
1814                    0,  /* non standard timer */
1815                    0, /* open_requests have no inode */
1816                    0, req);
1817 }
1818
1819 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1820 {
1821         struct in6_addr *dest, *src;
1822         __u16 destp, srcp;
1823         int timer_active;
1824         unsigned long timer_expires;
1825         struct inet_sock *inet = inet_sk(sp);
1826         struct tcp_sock *tp = tcp_sk(sp);
1827         const struct inet_connection_sock *icsk = inet_csk(sp);
1828         struct ipv6_pinfo *np = inet6_sk(sp);
1829
1830         dest  = &np->daddr;
1831         src   = &np->rcv_saddr;
1832         destp = ntohs(inet->dport);
1833         srcp  = ntohs(inet->sport);
1834
1835         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1836                 timer_active    = 1;
1837                 timer_expires   = icsk->icsk_timeout;
1838         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1839                 timer_active    = 4;
1840                 timer_expires   = icsk->icsk_timeout;
1841         } else if (timer_pending(&sp->sk_timer)) {
1842                 timer_active    = 2;
1843                 timer_expires   = sp->sk_timer.expires;
1844         } else {
1845                 timer_active    = 0;
1846                 timer_expires = jiffies;
1847         }
1848
1849         seq_printf(seq,
1850                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1851                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %u %u %u %u %d\n",
1852                    i,
1853                    src->s6_addr32[0], src->s6_addr32[1],
1854                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1855                    dest->s6_addr32[0], dest->s6_addr32[1],
1856                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1857                    sp->sk_state,
1858                    tp->write_seq-tp->snd_una, tp->rcv_nxt-tp->copied_seq,
1859                    timer_active,
1860                    jiffies_to_clock_t(timer_expires - jiffies),
1861                    icsk->icsk_retransmits,
1862                    sock_i_uid(sp),
1863                    icsk->icsk_probes_out,
1864                    sock_i_ino(sp),
1865                    atomic_read(&sp->sk_refcnt), sp,
1866                    icsk->icsk_rto,
1867                    icsk->icsk_ack.ato,
1868                    (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1869                    tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
1870                    );
1871 }
1872
1873 static void get_timewait6_sock(struct seq_file *seq,
1874                                struct inet_timewait_sock *tw, int i)
1875 {
1876         struct in6_addr *dest, *src;
1877         __u16 destp, srcp;
1878         struct tcp6_timewait_sock *tcp6tw = tcp6_twsk((struct sock *)tw);
1879         int ttd = tw->tw_ttd - jiffies;
1880
1881         if (ttd < 0)
1882                 ttd = 0;
1883
1884         dest = &tcp6tw->tw_v6_daddr;
1885         src  = &tcp6tw->tw_v6_rcv_saddr;
1886         destp = ntohs(tw->tw_dport);
1887         srcp  = ntohs(tw->tw_sport);
1888
1889         seq_printf(seq,
1890                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1891                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1892                    i,
1893                    src->s6_addr32[0], src->s6_addr32[1],
1894                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1895                    dest->s6_addr32[0], dest->s6_addr32[1],
1896                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1897                    tw->tw_substate, 0, 0,
1898                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
1899                    atomic_read(&tw->tw_refcnt), tw);
1900 }
1901
1902 #ifdef CONFIG_PROC_FS
1903 static int tcp6_seq_show(struct seq_file *seq, void *v)
1904 {
1905         struct tcp_iter_state *st;
1906
1907         if (v == SEQ_START_TOKEN) {
1908                 seq_puts(seq,
1909                          "  sl  "
1910                          "local_address                         "
1911                          "remote_address                        "
1912                          "st tx_queue rx_queue tr tm->when retrnsmt"
1913                          "   uid  timeout inode\n");
1914                 goto out;
1915         }
1916         st = seq->private;
1917
1918         switch (st->state) {
1919         case TCP_SEQ_STATE_LISTENING:
1920         case TCP_SEQ_STATE_ESTABLISHED:
1921                 get_tcp6_sock(seq, v, st->num);
1922                 break;
1923         case TCP_SEQ_STATE_OPENREQ:
1924                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1925                 break;
1926         case TCP_SEQ_STATE_TIME_WAIT:
1927                 get_timewait6_sock(seq, v, st->num);
1928                 break;
1929         }
1930 out:
1931         return 0;
1932 }
1933
1934 static struct file_operations tcp6_seq_fops;
1935 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1936         .owner          = THIS_MODULE,
1937         .name           = "tcp6",
1938         .family         = AF_INET6,
1939         .seq_show       = tcp6_seq_show,
1940         .seq_fops       = &tcp6_seq_fops,
1941 };
1942
1943 int __init tcp6_proc_init(void)
1944 {
1945         return tcp_proc_register(&tcp6_seq_afinfo);
1946 }
1947
1948 void tcp6_proc_exit(void)
1949 {
1950         tcp_proc_unregister(&tcp6_seq_afinfo);
1951 }
1952 #endif
1953
1954 struct proto tcpv6_prot = {
1955         .name                   = "TCPv6",
1956         .owner                  = THIS_MODULE,
1957         .close                  = tcp_close,
1958         .connect                = tcp_v6_connect,
1959         .disconnect             = tcp_disconnect,
1960         .accept                 = inet_csk_accept,
1961         .ioctl                  = tcp_ioctl,
1962         .init                   = tcp_v6_init_sock,
1963         .destroy                = tcp_v6_destroy_sock,
1964         .shutdown               = tcp_shutdown,
1965         .setsockopt             = tcp_setsockopt,
1966         .getsockopt             = tcp_getsockopt,
1967         .sendmsg                = tcp_sendmsg,
1968         .recvmsg                = tcp_recvmsg,
1969         .backlog_rcv            = tcp_v6_do_rcv,
1970         .hash                   = tcp_v6_hash,
1971         .unhash                 = tcp_unhash,
1972         .get_port               = tcp_v6_get_port,
1973         .enter_memory_pressure  = tcp_enter_memory_pressure,
1974         .sockets_allocated      = &tcp_sockets_allocated,
1975         .memory_allocated       = &tcp_memory_allocated,
1976         .memory_pressure        = &tcp_memory_pressure,
1977         .orphan_count           = &tcp_orphan_count,
1978         .sysctl_mem             = sysctl_tcp_mem,
1979         .sysctl_wmem            = sysctl_tcp_wmem,
1980         .sysctl_rmem            = sysctl_tcp_rmem,
1981         .max_header             = MAX_TCP_HEADER,
1982         .obj_size               = sizeof(struct tcp6_sock),
1983         .twsk_obj_size          = sizeof(struct tcp6_timewait_sock),
1984         .rsk_prot               = &tcp6_request_sock_ops,
1985 };
1986
1987 static struct inet6_protocol tcpv6_protocol = {
1988         .handler        =       tcp_v6_rcv,
1989         .err_handler    =       tcp_v6_err,
1990         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1991 };
1992
1993 static struct inet_protosw tcpv6_protosw = {
1994         .type           =       SOCK_STREAM,
1995         .protocol       =       IPPROTO_TCP,
1996         .prot           =       &tcpv6_prot,
1997         .ops            =       &inet6_stream_ops,
1998         .capability     =       -1,
1999         .no_check       =       0,
2000         .flags          =       INET_PROTOSW_PERMANENT,
2001 };
2002
2003 void __init tcpv6_init(void)
2004 {
2005         /* register inet6 protocol */
2006         if (inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP) < 0)
2007                 printk(KERN_ERR "tcpv6_init: Could not register protocol\n");
2008         inet6_register_protosw(&tcpv6_protosw);
2009 }