git.oblomov.eu Git - linux-2.6/blob - net/ipv6/tcp_ipv6.c

   1 /*
   2  *      TCP over IPv6
   3  *      Linux INET6 implementation
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      Based on:
   9  *      linux/net/ipv4/tcp.c
  10  *      linux/net/ipv4/tcp_input.c
  11  *      linux/net/ipv4/tcp_output.c
  12  *
  13  *      Fixes:
  14  *      Hideaki YOSHIFUJI       :       sin6_scope_id support
  15  *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
  16  *      Alexey Kuznetsov                allow both IPv4 and IPv6 sockets to bind
  17  *                                      a single port at the same time.
  18  *      YOSHIFUJI Hideaki @USAGI:       convert /proc/net/tcp6 to seq_file.
  19  *
  20  *      This program is free software; you can redistribute it and/or
  21  *      modify it under the terms of the GNU General Public License
  22  *      as published by the Free Software Foundation; either version
  23  *      2 of the License, or (at your option) any later version.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/errno.h>
  28 #include <linux/types.h>
  29 #include <linux/socket.h>
  30 #include <linux/sockios.h>
  31 #include <linux/net.h>
  32 #include <linux/jiffies.h>
  33 #include <linux/in.h>
  34 #include <linux/in6.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/init.h>
  37 #include <linux/jhash.h>
  38 #include <linux/ipsec.h>
  39 #include <linux/times.h>
  40
  41 #include <linux/ipv6.h>
  42 #include <linux/icmpv6.h>
  43 #include <linux/random.h>
  44
  45 #include <net/tcp.h>
  46 #include <net/ndisc.h>
  47 #include <net/inet6_hashtables.h>
  48 #include <net/inet6_connection_sock.h>
  49 #include <net/ipv6.h>
  50 #include <net/transp_v6.h>
  51 #include <net/addrconf.h>
  52 #include <net/ip6_route.h>
  53 #include <net/ip6_checksum.h>
  54 #include <net/inet_ecn.h>
  55 #include <net/protocol.h>
  56 #include <net/xfrm.h>
  57 #include <net/snmp.h>
  58 #include <net/dsfield.h>
  59 #include <net/timewait_sock.h>
  60 #include <net/netdma.h>
  61 #include <net/inet_common.h>
  62
  63 #include <asm/uaccess.h>
  64
  65 #include <linux/proc_fs.h>
  66 #include <linux/seq_file.h>
  67
  68 #include <linux/crypto.h>
  69 #include <linux/scatterlist.h>
  70
  71 static void     tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
  72 static void     tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
  73                                       struct request_sock *req);
  74
  75 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  76
  77 static struct inet_connection_sock_af_ops ipv6_mapped;
  78 static struct inet_connection_sock_af_ops ipv6_specific;
  79 #ifdef CONFIG_TCP_MD5SIG
  80 static struct tcp_sock_af_ops tcp_sock_ipv6_specific;
  81 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
  82 #else
  83 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
  84                                                    struct in6_addr *addr)
  85 {
  86         return NULL;
  87 }
  88 #endif
  89
  90 static void tcp_v6_hash(struct sock *sk)
  91 {
  92         if (sk->sk_state != TCP_CLOSE) {
  93                 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
  94                         tcp_prot.hash(sk);
  95                         return;
  96                 }
  97                 local_bh_disable();
  98                 __inet6_hash(sk);
  99                 local_bh_enable();
 100         }
 101 }
 102
 103 static __inline__ __sum16 tcp_v6_check(struct tcphdr *th, int len,
 104                                    struct in6_addr *saddr,
 105                                    struct in6_addr *daddr,
 106                                    __wsum base)
 107 {
 108         return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
 109 }
 110
 111 static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
 112 {
 113         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
 114                                             ipv6_hdr(skb)->saddr.s6_addr32,
 115                                             tcp_hdr(skb)->dest,
 116                                             tcp_hdr(skb)->source);
 117 }
 118
 119 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 120                           int addr_len)
 121 {
 122         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 123         struct inet_sock *inet = inet_sk(sk);
 124         struct inet_connection_sock *icsk = inet_csk(sk);
 125         struct ipv6_pinfo *np = inet6_sk(sk);
 126         struct tcp_sock *tp = tcp_sk(sk);
 127         struct in6_addr *saddr = NULL, *final_p = NULL, final;
 128         struct flowi fl;
 129         struct dst_entry *dst;
 130         int addr_type;
 131         int err;
 132
 133         if (addr_len < SIN6_LEN_RFC2133)
 134                 return -EINVAL;
 135
 136         if (usin->sin6_family != AF_INET6)
 137                 return(-EAFNOSUPPORT);
 138
 139         memset(&fl, 0, sizeof(fl));
 140
 141         if (np->sndflow) {
 142                 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 143                 IP6_ECN_flow_init(fl.fl6_flowlabel);
 144                 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
 145                         struct ip6_flowlabel *flowlabel;
 146                         flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
 147                         if (flowlabel == NULL)
 148                                 return -EINVAL;
 149                         ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
 150                         fl6_sock_release(flowlabel);
 151                 }
 152         }
 153
 154         /*
 155          *      connect() to INADDR_ANY means loopback (BSD'ism).
 156          */
 157
 158         if(ipv6_addr_any(&usin->sin6_addr))
 159                 usin->sin6_addr.s6_addr[15] = 0x1;
 160
 161         addr_type = ipv6_addr_type(&usin->sin6_addr);
 162
 163         if(addr_type & IPV6_ADDR_MULTICAST)
 164                 return -ENETUNREACH;
 165
 166         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 167                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 168                     usin->sin6_scope_id) {
 169                         /* If interface is set while binding, indices
 170                          * must coincide.
 171                          */
 172                         if (sk->sk_bound_dev_if &&
 173                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 174                                 return -EINVAL;
 175
 176                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 177                 }
 178
 179                 /* Connect to link-local address requires an interface */
 180                 if (!sk->sk_bound_dev_if)
 181                         return -EINVAL;
 182         }
 183
 184         if (tp->rx_opt.ts_recent_stamp &&
 185             !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
 186                 tp->rx_opt.ts_recent = 0;
 187                 tp->rx_opt.ts_recent_stamp = 0;
 188                 tp->write_seq = 0;
 189         }
 190
 191         ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
 192         np->flow_label = fl.fl6_flowlabel;
 193
 194         /*
 195          *      TCP over IPv4
 196          */
 197
 198         if (addr_type == IPV6_ADDR_MAPPED) {
 199                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 200                 struct sockaddr_in sin;
 201
 202                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 203
 204                 if (__ipv6_only_sock(sk))
 205                         return -ENETUNREACH;
 206
 207                 sin.sin_family = AF_INET;
 208                 sin.sin_port = usin->sin6_port;
 209                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 210
 211                 icsk->icsk_af_ops = &ipv6_mapped;
 212                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 213 #ifdef CONFIG_TCP_MD5SIG
 214                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
 215 #endif
 216
 217                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 218
 219                 if (err) {
 220                         icsk->icsk_ext_hdr_len = exthdrlen;
 221                         icsk->icsk_af_ops = &ipv6_specific;
 222                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 223 #ifdef CONFIG_TCP_MD5SIG
 224                         tp->af_specific = &tcp_sock_ipv6_specific;
 225 #endif
 226                         goto failure;
 227                 } else {
 228                         ipv6_addr_set(&np->saddr, 0, 0, htonl(0x0000FFFF),
 229                                       inet->saddr);
 230                         ipv6_addr_set(&np->rcv_saddr, 0, 0, htonl(0x0000FFFF),
 231                                       inet->rcv_saddr);
 232                 }
 233
 234                 return err;
 235         }
 236
 237         if (!ipv6_addr_any(&np->rcv_saddr))
 238                 saddr = &np->rcv_saddr;
 239
 240         fl.proto = IPPROTO_TCP;
 241         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 242         ipv6_addr_copy(&fl.fl6_src,
 243                        (saddr ? saddr : &np->saddr));
 244         fl.oif = sk->sk_bound_dev_if;
 245         fl.fl_ip_dport = usin->sin6_port;
 246         fl.fl_ip_sport = inet->sport;
 247
 248         if (np->opt && np->opt->srcrt) {
 249                 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
 250                 ipv6_addr_copy(&final, &fl.fl6_dst);
 251                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 252                 final_p = &final;
 253         }
 254
 255         security_sk_classify_flow(sk, &fl);
 256
 257         err = ip6_dst_lookup(sk, &dst, &fl);
 258         if (err)
 259                 goto failure;
 260         if (final_p)
 261                 ipv6_addr_copy(&fl.fl6_dst, final_p);
 262
 263         if ((err = __xfrm_lookup(&dst, &fl, sk, XFRM_LOOKUP_WAIT)) < 0) {
 264                 if (err == -EREMOTE)
 265                         err = ip6_dst_blackhole(sk, &dst, &fl);
 266                 if (err < 0)
 267                         goto failure;
 268         }
 269
 270         if (saddr == NULL) {
 271                 saddr = &fl.fl6_src;
 272                 ipv6_addr_copy(&np->rcv_saddr, saddr);
 273         }
 274
 275         /* set the source address */
 276         ipv6_addr_copy(&np->saddr, saddr);
 277         inet->rcv_saddr = LOOPBACK4_IPV6;
 278
 279         sk->sk_gso_type = SKB_GSO_TCPV6;
 280         __ip6_dst_store(sk, dst, NULL, NULL);
 281
 282         icsk->icsk_ext_hdr_len = 0;
 283         if (np->opt)
 284                 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
 285                                           np->opt->opt_nflen);
 286
 287         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 288
 289         inet->dport = usin->sin6_port;
 290
 291         tcp_set_state(sk, TCP_SYN_SENT);
 292         err = inet6_hash_connect(&tcp_death_row, sk);
 293         if (err)
 294                 goto late_failure;
 295
 296         if (!tp->write_seq)
 297                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 298                                                              np->daddr.s6_addr32,
 299                                                              inet->sport,
 300                                                              inet->dport);
 301
 302         err = tcp_connect(sk);
 303         if (err)
 304                 goto late_failure;
 305
 306         return 0;
 307
 308 late_failure:
 309         tcp_set_state(sk, TCP_CLOSE);
 310         __sk_dst_reset(sk);
 311 failure:
 312         inet->dport = 0;
 313         sk->sk_route_caps = 0;
 314         return err;
 315 }
 316
 317 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 318                 int type, int code, int offset, __be32 info)
 319 {
 320         struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
 321         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 322         struct ipv6_pinfo *np;
 323         struct sock *sk;
 324         int err;
 325         struct tcp_sock *tp;
 326         __u32 seq;
 327         struct net *net = dev_net(skb->dev);
 328
 329         sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
 330                         th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
 331
 332         if (sk == NULL) {
 333                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 334                                    ICMP6_MIB_INERRORS);
 335                 return;
 336         }
 337
 338         if (sk->sk_state == TCP_TIME_WAIT) {
 339                 inet_twsk_put(inet_twsk(sk));
 340                 return;
 341         }
 342
 343         bh_lock_sock(sk);
 344         if (sock_owned_by_user(sk))
 345                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 346
 347         if (sk->sk_state == TCP_CLOSE)
 348                 goto out;
 349
 350         tp = tcp_sk(sk);
 351         seq = ntohl(th->seq);
 352         if (sk->sk_state != TCP_LISTEN &&
 353             !between(seq, tp->snd_una, tp->snd_nxt)) {
 354                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 355                 goto out;
 356         }
 357
 358         np = inet6_sk(sk);
 359
 360         if (type == ICMPV6_PKT_TOOBIG) {
 361                 struct dst_entry *dst = NULL;
 362
 363                 if (sock_owned_by_user(sk))
 364                         goto out;
 365                 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 366                         goto out;
 367
 368                 /* icmp should have updated the destination cache entry */
 369                 dst = __sk_dst_check(sk, np->dst_cookie);
 370
 371                 if (dst == NULL) {
 372                         struct inet_sock *inet = inet_sk(sk);
 373                         struct flowi fl;
 374
 375                         /* BUGGG_FUTURE: Again, it is not clear how
 376                            to handle rthdr case. Ignore this complexity
 377                            for now.
 378                          */
 379                         memset(&fl, 0, sizeof(fl));
 380                         fl.proto = IPPROTO_TCP;
 381                         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 382                         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
 383                         fl.oif = sk->sk_bound_dev_if;
 384                         fl.fl_ip_dport = inet->dport;
 385                         fl.fl_ip_sport = inet->sport;
 386                         security_skb_classify_flow(skb, &fl);
 387
 388                         if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
 389                                 sk->sk_err_soft = -err;
 390                                 goto out;
 391                         }
 392
 393                         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0) {
 394                                 sk->sk_err_soft = -err;
 395                                 goto out;
 396                         }
 397
 398                 } else
 399                         dst_hold(dst);
 400
 401                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
 402                         tcp_sync_mss(sk, dst_mtu(dst));
 403                         tcp_simple_retransmit(sk);
 404                 } /* else let the usual retransmit timer handle it */
 405                 dst_release(dst);
 406                 goto out;
 407         }
 408
 409         icmpv6_err_convert(type, code, &err);
 410
 411         /* Might be for an request_sock */
 412         switch (sk->sk_state) {
 413                 struct request_sock *req, **prev;
 414         case TCP_LISTEN:
 415                 if (sock_owned_by_user(sk))
 416                         goto out;
 417
 418                 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
 419                                            &hdr->saddr, inet6_iif(skb));
 420                 if (!req)
 421                         goto out;
 422
 423                 /* ICMPs are not backlogged, hence we cannot get
 424                  * an established socket here.
 425                  */
 426                 WARN_ON(req->sk != NULL);
 427
 428                 if (seq != tcp_rsk(req)->snt_isn) {
 429                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 430                         goto out;
 431                 }
 432
 433                 inet_csk_reqsk_queue_drop(sk, req, prev);
 434                 goto out;
 435
 436         case TCP_SYN_SENT:
 437         case TCP_SYN_RECV:  /* Cannot happen.
 438                                It can, it SYNs are crossed. --ANK */
 439                 if (!sock_owned_by_user(sk)) {
 440                         sk->sk_err = err;
 441                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 442
 443                         tcp_done(sk);
 444                 } else
 445                         sk->sk_err_soft = err;
 446                 goto out;
 447         }
 448
 449         if (!sock_owned_by_user(sk) && np->recverr) {
 450                 sk->sk_err = err;
 451                 sk->sk_error_report(sk);
 452         } else
 453                 sk->sk_err_soft = err;
 454
 455 out:
 456         bh_unlock_sock(sk);
 457         sock_put(sk);
 458 }
 459
 460
 461 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req)
 462 {
 463         struct inet6_request_sock *treq = inet6_rsk(req);
 464         struct ipv6_pinfo *np = inet6_sk(sk);
 465         struct sk_buff * skb;
 466         struct ipv6_txoptions *opt = NULL;
 467         struct in6_addr * final_p = NULL, final;
 468         struct flowi fl;
 469         struct dst_entry *dst;
 470         int err = -1;
 471
 472         memset(&fl, 0, sizeof(fl));
 473         fl.proto = IPPROTO_TCP;
 474         ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
 475         ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
 476         fl.fl6_flowlabel = 0;
 477         fl.oif = treq->iif;
 478         fl.fl_ip_dport = inet_rsk(req)->rmt_port;
 479         fl.fl_ip_sport = inet_rsk(req)->loc_port;
 480         security_req_classify_flow(req, &fl);
 481
 482         opt = np->opt;
 483         if (opt && opt->srcrt) {
 484                 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
 485                 ipv6_addr_copy(&final, &fl.fl6_dst);
 486                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 487                 final_p = &final;
 488         }
 489
 490         err = ip6_dst_lookup(sk, &dst, &fl);
 491         if (err)
 492                 goto done;
 493         if (final_p)
 494                 ipv6_addr_copy(&fl.fl6_dst, final_p);
 495         if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
 496                 goto done;
 497
 498         skb = tcp_make_synack(sk, dst, req);
 499         if (skb) {
 500                 struct tcphdr *th = tcp_hdr(skb);
 501
 502                 th->check = tcp_v6_check(th, skb->len,
 503                                          &treq->loc_addr, &treq->rmt_addr,
 504                                          csum_partial((char *)th, skb->len, skb->csum));
 505
 506                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
 507                 err = ip6_xmit(sk, skb, &fl, opt, 0);
 508                 err = net_xmit_eval(err);
 509         }
 510
 511 done:
 512         if (opt && opt != np->opt)
 513                 sock_kfree_s(sk, opt, opt->tot_len);
 514         dst_release(dst);
 515         return err;
 516 }
 517
 518 static inline void syn_flood_warning(struct sk_buff *skb)
 519 {
 520 #ifdef CONFIG_SYN_COOKIES
 521         if (sysctl_tcp_syncookies)
 522                 printk(KERN_INFO
 523                        "TCPv6: Possible SYN flooding on port %d. "
 524                        "Sending cookies.\n", ntohs(tcp_hdr(skb)->dest));
 525         else
 526 #endif
 527                 printk(KERN_INFO
 528                        "TCPv6: Possible SYN flooding on port %d. "
 529                        "Dropping request.\n", ntohs(tcp_hdr(skb)->dest));
 530 }
 531
 532 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 533 {
 534         if (inet6_rsk(req)->pktopts)
 535                 kfree_skb(inet6_rsk(req)->pktopts);
 536 }
 537
 538 #ifdef CONFIG_TCP_MD5SIG
 539 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
 540                                                    struct in6_addr *addr)
 541 {
 542         struct tcp_sock *tp = tcp_sk(sk);
 543         int i;
 544
 545         BUG_ON(tp == NULL);
 546
 547         if (!tp->md5sig_info || !tp->md5sig_info->entries6)
 548                 return NULL;
 549
 550         for (i = 0; i < tp->md5sig_info->entries6; i++) {
 551                 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, addr))
 552                         return &tp->md5sig_info->keys6[i].base;
 553         }
 554         return NULL;
 555 }
 556
 557 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
 558                                                 struct sock *addr_sk)
 559 {
 560         return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
 561 }
 562
 563 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
 564                                                       struct request_sock *req)
 565 {
 566         return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
 567 }
 568
 569 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
 570                              char *newkey, u8 newkeylen)
 571 {
 572         /* Add key to the list */
 573         struct tcp_md5sig_key *key;
 574         struct tcp_sock *tp = tcp_sk(sk);
 575         struct tcp6_md5sig_key *keys;
 576
 577         key = tcp_v6_md5_do_lookup(sk, peer);
 578         if (key) {
 579                 /* modify existing entry - just update that one */
 580                 kfree(key->key);
 581                 key->key = newkey;
 582                 key->keylen = newkeylen;
 583         } else {
 584                 /* reallocate new list if current one is full. */
 585                 if (!tp->md5sig_info) {
 586                         tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
 587                         if (!tp->md5sig_info) {
 588                                 kfree(newkey);
 589                                 return -ENOMEM;
 590                         }
 591                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
 592                 }
 593                 if (tcp_alloc_md5sig_pool() == NULL) {
 594                         kfree(newkey);
 595                         return -ENOMEM;
 596                 }
 597                 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
 598                         keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
 599                                        (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
 600
 601                         if (!keys) {
 602                                 tcp_free_md5sig_pool();
 603                                 kfree(newkey);
 604                                 return -ENOMEM;
 605                         }
 606
 607                         if (tp->md5sig_info->entries6)
 608                                 memmove(keys, tp->md5sig_info->keys6,
 609                                         (sizeof (tp->md5sig_info->keys6[0]) *
 610                                          tp->md5sig_info->entries6));
 611
 612                         kfree(tp->md5sig_info->keys6);
 613                         tp->md5sig_info->keys6 = keys;
 614                         tp->md5sig_info->alloced6++;
 615                 }
 616
 617                 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
 618                                peer);
 619                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
 620                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
 621
 622                 tp->md5sig_info->entries6++;
 623         }
 624         return 0;
 625 }
 626
 627 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
 628                                u8 *newkey, __u8 newkeylen)
 629 {
 630         return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
 631                                  newkey, newkeylen);
 632 }
 633
 634 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
 635 {
 636         struct tcp_sock *tp = tcp_sk(sk);
 637         int i;
 638
 639         for (i = 0; i < tp->md5sig_info->entries6; i++) {
 640                 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) {
 641                         /* Free the key */
 642                         kfree(tp->md5sig_info->keys6[i].base.key);
 643                         tp->md5sig_info->entries6--;
 644
 645                         if (tp->md5sig_info->entries6 == 0) {
 646                                 kfree(tp->md5sig_info->keys6);
 647                                 tp->md5sig_info->keys6 = NULL;
 648                                 tp->md5sig_info->alloced6 = 0;
 649                         } else {
 650                                 /* shrink the database */
 651                                 if (tp->md5sig_info->entries6 != i)
 652                                         memmove(&tp->md5sig_info->keys6[i],
 653                                                 &tp->md5sig_info->keys6[i+1],
 654                                                 (tp->md5sig_info->entries6 - i)
 655                                                 * sizeof (tp->md5sig_info->keys6[0]));
 656                         }
 657                         tcp_free_md5sig_pool();
 658                         return 0;
 659                 }
 660         }
 661         return -ENOENT;
 662 }
 663
 664 static void tcp_v6_clear_md5_list (struct sock *sk)
 665 {
 666         struct tcp_sock *tp = tcp_sk(sk);
 667         int i;
 668
 669         if (tp->md5sig_info->entries6) {
 670                 for (i = 0; i < tp->md5sig_info->entries6; i++)
 671                         kfree(tp->md5sig_info->keys6[i].base.key);
 672                 tp->md5sig_info->entries6 = 0;
 673                 tcp_free_md5sig_pool();
 674         }
 675
 676         kfree(tp->md5sig_info->keys6);
 677         tp->md5sig_info->keys6 = NULL;
 678         tp->md5sig_info->alloced6 = 0;
 679
 680         if (tp->md5sig_info->entries4) {
 681                 for (i = 0; i < tp->md5sig_info->entries4; i++)
 682                         kfree(tp->md5sig_info->keys4[i].base.key);
 683                 tp->md5sig_info->entries4 = 0;
 684                 tcp_free_md5sig_pool();
 685         }
 686
 687         kfree(tp->md5sig_info->keys4);
 688         tp->md5sig_info->keys4 = NULL;
 689         tp->md5sig_info->alloced4 = 0;
 690 }
 691
 692 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
 693                                   int optlen)
 694 {
 695         struct tcp_md5sig cmd;
 696         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 697         u8 *newkey;
 698
 699         if (optlen < sizeof(cmd))
 700                 return -EINVAL;
 701
 702         if (copy_from_user(&cmd, optval, sizeof(cmd)))
 703                 return -EFAULT;
 704
 705         if (sin6->sin6_family != AF_INET6)
 706                 return -EINVAL;
 707
 708         if (!cmd.tcpm_keylen) {
 709                 if (!tcp_sk(sk)->md5sig_info)
 710                         return -ENOENT;
 711                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 712                         return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
 713                 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
 714         }
 715
 716         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 717                 return -EINVAL;
 718
 719         if (!tcp_sk(sk)->md5sig_info) {
 720                 struct tcp_sock *tp = tcp_sk(sk);
 721                 struct tcp_md5sig_info *p;
 722
 723                 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
 724                 if (!p)
 725                         return -ENOMEM;
 726
 727                 tp->md5sig_info = p;
 728                 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
 729         }
 730
 731         newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 732         if (!newkey)
 733                 return -ENOMEM;
 734         if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
 735                 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
 736                                          newkey, cmd.tcpm_keylen);
 737         }
 738         return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
 739 }
 740
 741 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
 742                                         struct in6_addr *daddr,
 743                                         struct in6_addr *saddr, int nbytes)
 744 {
 745         struct tcp6_pseudohdr *bp;
 746         struct scatterlist sg;
 747
 748         bp = &hp->md5_blk.ip6;
 749         /* 1. TCP pseudo-header (RFC2460) */
 750         ipv6_addr_copy(&bp->saddr, saddr);
 751         ipv6_addr_copy(&bp->daddr, daddr);
 752         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 753         bp->len = cpu_to_be32(nbytes);
 754
 755         sg_init_one(&sg, bp, sizeof(*bp));
 756         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
 757 }
 758
 759 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
 760                                struct in6_addr *daddr, struct in6_addr *saddr,
 761                                struct tcphdr *th)
 762 {
 763         struct tcp_md5sig_pool *hp;
 764         struct hash_desc *desc;
 765
 766         hp = tcp_get_md5sig_pool();
 767         if (!hp)
 768                 goto clear_hash_noput;
 769         desc = &hp->md5_desc;
 770
 771         if (crypto_hash_init(desc))
 772                 goto clear_hash;
 773         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
 774                 goto clear_hash;
 775         if (tcp_md5_hash_header(hp, th))
 776                 goto clear_hash;
 777         if (tcp_md5_hash_key(hp, key))
 778                 goto clear_hash;
 779         if (crypto_hash_final(desc, md5_hash))
 780                 goto clear_hash;
 781
 782         tcp_put_md5sig_pool();
 783         return 0;
 784
 785 clear_hash:
 786         tcp_put_md5sig_pool();
 787 clear_hash_noput:
 788         memset(md5_hash, 0, 16);
 789         return 1;
 790 }
 791
 792 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
 793                                struct sock *sk, struct request_sock *req,
 794                                struct sk_buff *skb)
 795 {
 796         struct in6_addr *saddr, *daddr;
 797         struct tcp_md5sig_pool *hp;
 798         struct hash_desc *desc;
 799         struct tcphdr *th = tcp_hdr(skb);
 800
 801         if (sk) {
 802                 saddr = &inet6_sk(sk)->saddr;
 803                 daddr = &inet6_sk(sk)->daddr;
 804         } else if (req) {
 805                 saddr = &inet6_rsk(req)->loc_addr;
 806                 daddr = &inet6_rsk(req)->rmt_addr;
 807         } else {
 808                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 809                 saddr = &ip6h->saddr;
 810                 daddr = &ip6h->daddr;
 811         }
 812
 813         hp = tcp_get_md5sig_pool();
 814         if (!hp)
 815                 goto clear_hash_noput;
 816         desc = &hp->md5_desc;
 817
 818         if (crypto_hash_init(desc))
 819                 goto clear_hash;
 820
 821         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
 822                 goto clear_hash;
 823         if (tcp_md5_hash_header(hp, th))
 824                 goto clear_hash;
 825         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 826                 goto clear_hash;
 827         if (tcp_md5_hash_key(hp, key))
 828                 goto clear_hash;
 829         if (crypto_hash_final(desc, md5_hash))
 830                 goto clear_hash;
 831
 832         tcp_put_md5sig_pool();
 833         return 0;
 834
 835 clear_hash:
 836         tcp_put_md5sig_pool();
 837 clear_hash_noput:
 838         memset(md5_hash, 0, 16);
 839         return 1;
 840 }
 841
 842 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
 843 {
 844         __u8 *hash_location = NULL;
 845         struct tcp_md5sig_key *hash_expected;
 846         struct ipv6hdr *ip6h = ipv6_hdr(skb);
 847         struct tcphdr *th = tcp_hdr(skb);
 848         int genhash;
 849         u8 newhash[16];
 850
 851         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 852         hash_location = tcp_parse_md5sig_option(th);
 853
 854         /* We've parsed the options - do we have a hash? */
 855         if (!hash_expected && !hash_location)
 856                 return 0;
 857
 858         if (hash_expected && !hash_location) {
 859                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 860                 return 1;
 861         }
 862
 863         if (!hash_expected && hash_location) {
 864                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 865                 return 1;
 866         }
 867
 868         /* check the signature */
 869         genhash = tcp_v6_md5_hash_skb(newhash,
 870                                       hash_expected,
 871                                       NULL, NULL, skb);
 872
 873         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 874                 if (net_ratelimit()) {
 875                         printk(KERN_INFO "MD5 Hash %s for (%pI6, %u)->(%pI6, %u)\n",
 876                                genhash ? "failed" : "mismatch",
 877                                &ip6h->saddr, ntohs(th->source),
 878                                &ip6h->daddr, ntohs(th->dest));
 879                 }
 880                 return 1;
 881         }
 882         return 0;
 883 }
 884 #endif
 885
 886 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 887         .family         =       AF_INET6,
 888         .obj_size       =       sizeof(struct tcp6_request_sock),
 889         .rtx_syn_ack    =       tcp_v6_send_synack,
 890         .send_ack       =       tcp_v6_reqsk_send_ack,
 891         .destructor     =       tcp_v6_reqsk_destructor,
 892         .send_reset     =       tcp_v6_send_reset
 893 };
 894
 895 #ifdef CONFIG_TCP_MD5SIG
 896 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 897         .md5_lookup     =       tcp_v6_reqsk_md5_lookup,
 898 };
 899 #endif
 900
 901 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
 902         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
 903         .twsk_unique    = tcp_twsk_unique,
 904         .twsk_destructor= tcp_twsk_destructor,
 905 };
 906
 907 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
 908 {
 909         struct ipv6_pinfo *np = inet6_sk(sk);
 910         struct tcphdr *th = tcp_hdr(skb);
 911
 912         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 913                 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,  0);
 914                 skb->csum_start = skb_transport_header(skb) - skb->head;
 915                 skb->csum_offset = offsetof(struct tcphdr, check);
 916         } else {
 917                 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
 918                                             csum_partial((char *)th, th->doff<<2,
 919                                                          skb->csum));
 920         }
 921 }
 922
 923 static int tcp_v6_gso_send_check(struct sk_buff *skb)
 924 {
 925         struct ipv6hdr *ipv6h;
 926         struct tcphdr *th;
 927
 928         if (!pskb_may_pull(skb, sizeof(*th)))
 929                 return -EINVAL;
 930
 931         ipv6h = ipv6_hdr(skb);
 932         th = tcp_hdr(skb);
 933
 934         th->check = 0;
 935         th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
 936                                      IPPROTO_TCP, 0);
 937         skb->csum_start = skb_transport_header(skb) - skb->head;
 938         skb->csum_offset = offsetof(struct tcphdr, check);
 939         skb->ip_summed = CHECKSUM_PARTIAL;
 940         return 0;
 941 }
 942
 943 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
 944                                  u32 ts, struct tcp_md5sig_key *key, int rst)
 945 {
 946         struct tcphdr *th = tcp_hdr(skb), *t1;
 947         struct sk_buff *buff;
 948         struct flowi fl;
 949         struct net *net = dev_net(skb->dst->dev);
 950         struct sock *ctl_sk = net->ipv6.tcp_sk;
 951         unsigned int tot_len = sizeof(struct tcphdr);
 952         __be32 *topt;
 953
 954         if (ts)
 955                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 956 #ifdef CONFIG_TCP_MD5SIG
 957         if (key)
 958                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 959 #endif
 960
 961         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 962                          GFP_ATOMIC);
 963         if (buff == NULL)
 964                 return;
 965
 966         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 967
 968         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 969
 970         /* Swap the send and the receive. */
 971         memset(t1, 0, sizeof(*t1));
 972         t1->dest = th->source;
 973         t1->source = th->dest;
 974         t1->doff = tot_len / 4;
 975         t1->seq = htonl(seq);
 976         t1->ack_seq = htonl(ack);
 977         t1->ack = !rst || !th->ack;
 978         t1->rst = rst;
 979         t1->window = htons(win);
 980
 981         topt = (__be32 *)(t1 + 1);
 982
 983         if (ts) {
 984                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 985                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 986                 *topt++ = htonl(tcp_time_stamp);
 987                 *topt++ = htonl(ts);
 988         }
 989
 990 #ifdef CONFIG_TCP_MD5SIG
 991         if (key) {
 992                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 993                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 994                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 995                                     &ipv6_hdr(skb)->saddr,
 996                                     &ipv6_hdr(skb)->daddr, t1);
 997         }
 998 #endif
 999
1000         buff->csum = csum_partial((char *)t1, tot_len, 0);
1001
1002         memset(&fl, 0, sizeof(fl));
1003         ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1004         ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1005
1006         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1007                                     tot_len, IPPROTO_TCP,
1008                                     buff->csum);
1009
1010         fl.proto = IPPROTO_TCP;
1011         fl.oif = inet6_iif(skb);
1012         fl.fl_ip_dport = t1->dest;
1013         fl.fl_ip_sport = t1->source;
1014         security_skb_classify_flow(skb, &fl);
1015
1016         /* Pass a socket to ip6_dst_lookup either it is for RST
1017          * Underlying function will use this to retrieve the network
1018          * namespace
1019          */
1020         if (!ip6_dst_lookup(ctl_sk, &buff->dst, &fl)) {
1021                 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1022                         ip6_xmit(ctl_sk, buff, &fl, NULL, 0);
1023                         TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1024                         if (rst)
1025                                 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1026                         return;
1027                 }
1028         }
1029
1030         kfree_skb(buff);
1031 }
1032
1033 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
1034 {
1035         struct tcphdr *th = tcp_hdr(skb);
1036         u32 seq = 0, ack_seq = 0;
1037         struct tcp_md5sig_key *key = NULL;
1038
1039         if (th->rst)
1040                 return;
1041
1042         if (!ipv6_unicast_destination(skb))
1043                 return;
1044
1045 #ifdef CONFIG_TCP_MD5SIG
1046         if (sk)
1047                 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1048 #endif
1049
1050         if (th->ack)
1051                 seq = ntohl(th->ack_seq);
1052         else
1053                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1054                           (th->doff << 2);
1055
1056         tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1);
1057 }
1058
1059 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1060                             struct tcp_md5sig_key *key)
1061 {
1062         tcp_v6_send_response(skb, seq, ack, win, ts, key, 0);
1063 }
1064
1065 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1066 {
1067         struct inet_timewait_sock *tw = inet_twsk(sk);
1068         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1069
1070         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1071                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1072                         tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw));
1073
1074         inet_twsk_put(tw);
1075 }
1076
1077 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1078                                   struct request_sock *req)
1079 {
1080         tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1081                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr));
1082 }
1083
1084
1085 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1086 {
1087         struct request_sock *req, **prev;
1088         const struct tcphdr *th = tcp_hdr(skb);
1089         struct sock *nsk;
1090
1091         /* Find possible connection requests. */
1092         req = inet6_csk_search_req(sk, &prev, th->source,
1093                                    &ipv6_hdr(skb)->saddr,
1094                                    &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1095         if (req)
1096                 return tcp_check_req(sk, skb, req, prev);
1097
1098         nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1099                         &ipv6_hdr(skb)->saddr, th->source,
1100                         &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1101
1102         if (nsk) {
1103                 if (nsk->sk_state != TCP_TIME_WAIT) {
1104                         bh_lock_sock(nsk);
1105                         return nsk;
1106                 }
1107                 inet_twsk_put(inet_twsk(nsk));
1108                 return NULL;
1109         }
1110
1111 #ifdef CONFIG_SYN_COOKIES
1112         if (!th->rst && !th->syn && th->ack)
1113                 sk = cookie_v6_check(sk, skb);
1114 #endif
1115         return sk;
1116 }
1117
1118 /* FIXME: this is substantially similar to the ipv4 code.
1119  * Can some kind of merge be done? -- erics
1120  */
1121 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1122 {
1123         struct inet6_request_sock *treq;
1124         struct ipv6_pinfo *np = inet6_sk(sk);
1125         struct tcp_options_received tmp_opt;
1126         struct tcp_sock *tp = tcp_sk(sk);
1127         struct request_sock *req = NULL;
1128         __u32 isn = TCP_SKB_CB(skb)->when;
1129 #ifdef CONFIG_SYN_COOKIES
1130         int want_cookie = 0;
1131 #else
1132 #define want_cookie 0
1133 #endif
1134
1135         if (skb->protocol == htons(ETH_P_IP))
1136                 return tcp_v4_conn_request(sk, skb);
1137
1138         if (!ipv6_unicast_destination(skb))
1139                 goto drop;
1140
1141         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1142                 if (net_ratelimit())
1143                         syn_flood_warning(skb);
1144 #ifdef CONFIG_SYN_COOKIES
1145                 if (sysctl_tcp_syncookies)
1146                         want_cookie = 1;
1147                 else
1148 #endif
1149                 goto drop;
1150         }
1151
1152         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1153                 goto drop;
1154
1155         req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1156         if (req == NULL)
1157                 goto drop;
1158
1159 #ifdef CONFIG_TCP_MD5SIG
1160         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1161 #endif
1162
1163         tcp_clear_options(&tmp_opt);
1164         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1165         tmp_opt.user_mss = tp->rx_opt.user_mss;
1166
1167         tcp_parse_options(skb, &tmp_opt, 0);
1168
1169         if (want_cookie && !tmp_opt.saw_tstamp)
1170                 tcp_clear_options(&tmp_opt);
1171
1172         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1173         tcp_openreq_init(req, &tmp_opt, skb);
1174
1175         treq = inet6_rsk(req);
1176         ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1177         ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1178         if (!want_cookie)
1179                 TCP_ECN_create_request(req, tcp_hdr(skb));
1180
1181         if (want_cookie) {
1182                 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1183                 req->cookie_ts = tmp_opt.tstamp_ok;
1184         } else if (!isn) {
1185                 if (ipv6_opt_accepted(sk, skb) ||
1186                     np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1187                     np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1188                         atomic_inc(&skb->users);
1189                         treq->pktopts = skb;
1190                 }
1191                 treq->iif = sk->sk_bound_dev_if;
1192
1193                 /* So that link locals have meaning */
1194                 if (!sk->sk_bound_dev_if &&
1195                     ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1196                         treq->iif = inet6_iif(skb);
1197
1198                 isn = tcp_v6_init_sequence(skb);
1199         }
1200
1201         tcp_rsk(req)->snt_isn = isn;
1202
1203         security_inet_conn_request(sk, skb, req);
1204
1205         if (tcp_v6_send_synack(sk, req))
1206                 goto drop;
1207
1208         if (!want_cookie) {
1209                 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1210                 return 0;
1211         }
1212
1213 drop:
1214         if (req)
1215                 reqsk_free(req);
1216
1217         return 0; /* don't send reset */
1218 }
1219
1220 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1221                                           struct request_sock *req,
1222                                           struct dst_entry *dst)
1223 {
1224         struct inet6_request_sock *treq;
1225         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1226         struct tcp6_sock *newtcp6sk;
1227         struct inet_sock *newinet;
1228         struct tcp_sock *newtp;
1229         struct sock *newsk;
1230         struct ipv6_txoptions *opt;
1231 #ifdef CONFIG_TCP_MD5SIG
1232         struct tcp_md5sig_key *key;
1233 #endif
1234
1235         if (skb->protocol == htons(ETH_P_IP)) {
1236                 /*
1237                  *      v6 mapped
1238                  */
1239
1240                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1241
1242                 if (newsk == NULL)
1243                         return NULL;
1244
1245                 newtcp6sk = (struct tcp6_sock *)newsk;
1246                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1247
1248                 newinet = inet_sk(newsk);
1249                 newnp = inet6_sk(newsk);
1250                 newtp = tcp_sk(newsk);
1251
1252                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1253
1254                 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1255                               newinet->daddr);
1256
1257                 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1258                               newinet->saddr);
1259
1260                 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1261
1262                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1263                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1264 #ifdef CONFIG_TCP_MD5SIG
1265                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1266 #endif
1267
1268                 newnp->pktoptions  = NULL;
1269                 newnp->opt         = NULL;
1270                 newnp->mcast_oif   = inet6_iif(skb);
1271                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1272
1273                 /*
1274                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1275                  * here, tcp_create_openreq_child now does this for us, see the comment in
1276                  * that function for the gory details. -acme
1277                  */
1278
1279                 /* It is tricky place. Until this moment IPv4 tcp
1280                    worked with IPv6 icsk.icsk_af_ops.
1281                    Sync it now.
1282                  */
1283                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1284
1285                 return newsk;
1286         }
1287
1288         treq = inet6_rsk(req);
1289         opt = np->opt;
1290
1291         if (sk_acceptq_is_full(sk))
1292                 goto out_overflow;
1293
1294         if (dst == NULL) {
1295                 struct in6_addr *final_p = NULL, final;
1296                 struct flowi fl;
1297
1298                 memset(&fl, 0, sizeof(fl));
1299                 fl.proto = IPPROTO_TCP;
1300                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1301                 if (opt && opt->srcrt) {
1302                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1303                         ipv6_addr_copy(&final, &fl.fl6_dst);
1304                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1305                         final_p = &final;
1306                 }
1307                 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1308                 fl.oif = sk->sk_bound_dev_if;
1309                 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1310                 fl.fl_ip_sport = inet_rsk(req)->loc_port;
1311                 security_req_classify_flow(req, &fl);
1312
1313                 if (ip6_dst_lookup(sk, &dst, &fl))
1314                         goto out;
1315
1316                 if (final_p)
1317                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1318
1319                 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1320                         goto out;
1321         }
1322
1323         newsk = tcp_create_openreq_child(sk, req, skb);
1324         if (newsk == NULL)
1325                 goto out;
1326
1327         /*
1328          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1329          * count here, tcp_create_openreq_child now does this for us, see the
1330          * comment in that function for the gory details. -acme
1331          */
1332
1333         newsk->sk_gso_type = SKB_GSO_TCPV6;
1334         __ip6_dst_store(newsk, dst, NULL, NULL);
1335
1336         newtcp6sk = (struct tcp6_sock *)newsk;
1337         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1338
1339         newtp = tcp_sk(newsk);
1340         newinet = inet_sk(newsk);
1341         newnp = inet6_sk(newsk);
1342
1343         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1344
1345         ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1346         ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1347         ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1348         newsk->sk_bound_dev_if = treq->iif;
1349
1350         /* Now IPv6 options...
1351
1352            First: no IPv4 options.
1353          */
1354         newinet->opt = NULL;
1355         newnp->ipv6_fl_list = NULL;
1356
1357         /* Clone RX bits */
1358         newnp->rxopt.all = np->rxopt.all;
1359
1360         /* Clone pktoptions received with SYN */
1361         newnp->pktoptions = NULL;
1362         if (treq->pktopts != NULL) {
1363                 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1364                 kfree_skb(treq->pktopts);
1365                 treq->pktopts = NULL;
1366                 if (newnp->pktoptions)
1367                         skb_set_owner_r(newnp->pktoptions, newsk);
1368         }
1369         newnp->opt        = NULL;
1370         newnp->mcast_oif  = inet6_iif(skb);
1371         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1372
1373         /* Clone native IPv6 options from listening socket (if any)
1374
1375            Yes, keeping reference count would be much more clever,
1376            but we make one more one thing there: reattach optmem
1377            to newsk.
1378          */
1379         if (opt) {
1380                 newnp->opt = ipv6_dup_options(newsk, opt);
1381                 if (opt != np->opt)
1382                         sock_kfree_s(sk, opt, opt->tot_len);
1383         }
1384
1385         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1386         if (newnp->opt)
1387                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1388                                                      newnp->opt->opt_flen);
1389
1390         tcp_mtup_init(newsk);
1391         tcp_sync_mss(newsk, dst_mtu(dst));
1392         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1393         tcp_initialize_rcv_mss(newsk);
1394
1395         newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1396
1397 #ifdef CONFIG_TCP_MD5SIG
1398         /* Copy over the MD5 key from the original socket */
1399         if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1400                 /* We're using one, so create a matching key
1401                  * on the newsk structure. If we fail to get
1402                  * memory, then we end up not copying the key
1403                  * across. Shucks.
1404                  */
1405                 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1406                 if (newkey != NULL)
1407                         tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1408                                           newkey, key->keylen);
1409         }
1410 #endif
1411
1412         __inet6_hash(newsk);
1413         __inet_inherit_port(sk, newsk);
1414
1415         return newsk;
1416
1417 out_overflow:
1418         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1419 out:
1420         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1421         if (opt && opt != np->opt)
1422                 sock_kfree_s(sk, opt, opt->tot_len);
1423         dst_release(dst);
1424         return NULL;
1425 }
1426
1427 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1428 {
1429         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1430                 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1431                                   &ipv6_hdr(skb)->daddr, skb->csum)) {
1432                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1433                         return 0;
1434                 }
1435         }
1436
1437         skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1438                                               &ipv6_hdr(skb)->saddr,
1439                                               &ipv6_hdr(skb)->daddr, 0));
1440
1441         if (skb->len <= 76) {
1442                 return __skb_checksum_complete(skb);
1443         }
1444         return 0;
1445 }
1446
1447 /* The socket must have it's spinlock held when we get
1448  * here.
1449  *
1450  * We have a potential double-lock case here, so even when
1451  * doing backlog processing we use the BH locking scheme.
1452  * This is because we cannot sleep with the original spinlock
1453  * held.
1454  */
1455 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1456 {
1457         struct ipv6_pinfo *np = inet6_sk(sk);
1458         struct tcp_sock *tp;
1459         struct sk_buff *opt_skb = NULL;
1460
1461         /* Imagine: socket is IPv6. IPv4 packet arrives,
1462            goes to IPv4 receive handler and backlogged.
1463            From backlog it always goes here. Kerboom...
1464            Fortunately, tcp_rcv_established and rcv_established
1465            handle them correctly, but it is not case with
1466            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1467          */
1468
1469         if (skb->protocol == htons(ETH_P_IP))
1470                 return tcp_v4_do_rcv(sk, skb);
1471
1472 #ifdef CONFIG_TCP_MD5SIG
1473         if (tcp_v6_inbound_md5_hash (sk, skb))
1474                 goto discard;
1475 #endif
1476
1477         if (sk_filter(sk, skb))
1478                 goto discard;
1479
1480         /*
1481          *      socket locking is here for SMP purposes as backlog rcv
1482          *      is currently called with bh processing disabled.
1483          */
1484
1485         /* Do Stevens' IPV6_PKTOPTIONS.
1486
1487            Yes, guys, it is the only place in our code, where we
1488            may make it not affecting IPv4.
1489            The rest of code is protocol independent,
1490            and I do not like idea to uglify IPv4.
1491
1492            Actually, all the idea behind IPV6_PKTOPTIONS
1493            looks not very well thought. For now we latch
1494            options, received in the last packet, enqueued
1495            by tcp. Feel free to propose better solution.
1496                                                --ANK (980728)
1497          */
1498         if (np->rxopt.all)
1499                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1500
1501         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1502                 TCP_CHECK_TIMER(sk);
1503                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1504                         goto reset;
1505                 TCP_CHECK_TIMER(sk);
1506                 if (opt_skb)
1507                         goto ipv6_pktoptions;
1508                 return 0;
1509         }
1510
1511         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1512                 goto csum_err;
1513
1514         if (sk->sk_state == TCP_LISTEN) {
1515                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1516                 if (!nsk)
1517                         goto discard;
1518
1519                 /*
1520                  * Queue it on the new socket if the new socket is active,
1521                  * otherwise we just shortcircuit this and continue with
1522                  * the new socket..
1523                  */
1524                 if(nsk != sk) {
1525                         if (tcp_child_process(sk, nsk, skb))
1526                                 goto reset;
1527                         if (opt_skb)
1528                                 __kfree_skb(opt_skb);
1529                         return 0;
1530                 }
1531         }
1532
1533         TCP_CHECK_TIMER(sk);
1534         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1535                 goto reset;
1536         TCP_CHECK_TIMER(sk);
1537         if (opt_skb)
1538                 goto ipv6_pktoptions;
1539         return 0;
1540
1541 reset:
1542         tcp_v6_send_reset(sk, skb);
1543 discard:
1544         if (opt_skb)
1545                 __kfree_skb(opt_skb);
1546         kfree_skb(skb);
1547         return 0;
1548 csum_err:
1549         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1550         goto discard;
1551
1552
1553 ipv6_pktoptions:
1554         /* Do you ask, what is it?
1555
1556            1. skb was enqueued by tcp.
1557            2. skb is added to tail of read queue, rather than out of order.
1558            3. socket is not in passive state.
1559            4. Finally, it really contains options, which user wants to receive.
1560          */
1561         tp = tcp_sk(sk);
1562         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1563             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1564                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1565                         np->mcast_oif = inet6_iif(opt_skb);
1566                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1567                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1568                 if (ipv6_opt_accepted(sk, opt_skb)) {
1569                         skb_set_owner_r(opt_skb, sk);
1570                         opt_skb = xchg(&np->pktoptions, opt_skb);
1571                 } else {
1572                         __kfree_skb(opt_skb);
1573                         opt_skb = xchg(&np->pktoptions, NULL);
1574                 }
1575         }
1576
1577         if (opt_skb)
1578                 kfree_skb(opt_skb);
1579         return 0;
1580 }
1581
1582 static int tcp_v6_rcv(struct sk_buff *skb)
1583 {
1584         struct tcphdr *th;
1585         struct sock *sk;
1586         int ret;
1587         struct net *net = dev_net(skb->dev);
1588
1589         if (skb->pkt_type != PACKET_HOST)
1590                 goto discard_it;
1591
1592         /*
1593          *      Count it even if it's bad.
1594          */
1595         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1596
1597         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1598                 goto discard_it;
1599
1600         th = tcp_hdr(skb);
1601
1602         if (th->doff < sizeof(struct tcphdr)/4)
1603                 goto bad_packet;
1604         if (!pskb_may_pull(skb, th->doff*4))
1605                 goto discard_it;
1606
1607         if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1608                 goto bad_packet;
1609
1610         th = tcp_hdr(skb);
1611         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1612         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1613                                     skb->len - th->doff*4);
1614         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1615         TCP_SKB_CB(skb)->when = 0;
1616         TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1617         TCP_SKB_CB(skb)->sacked = 0;
1618
1619         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1620         if (!sk)
1621                 goto no_tcp_socket;
1622
1623 process:
1624         if (sk->sk_state == TCP_TIME_WAIT)
1625                 goto do_time_wait;
1626
1627         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1628                 goto discard_and_relse;
1629
1630         if (sk_filter(sk, skb))
1631                 goto discard_and_relse;
1632
1633         skb->dev = NULL;
1634
1635         bh_lock_sock_nested(sk);
1636         ret = 0;
1637         if (!sock_owned_by_user(sk)) {
1638 #ifdef CONFIG_NET_DMA
1639                 struct tcp_sock *tp = tcp_sk(sk);
1640                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1641                         tp->ucopy.dma_chan = get_softnet_dma();
1642                 if (tp->ucopy.dma_chan)
1643                         ret = tcp_v6_do_rcv(sk, skb);
1644                 else
1645 #endif
1646                 {
1647                         if (!tcp_prequeue(sk, skb))
1648                                 ret = tcp_v6_do_rcv(sk, skb);
1649                 }
1650         } else
1651                 sk_add_backlog(sk, skb);
1652         bh_unlock_sock(sk);
1653
1654         sock_put(sk);
1655         return ret ? -1 : 0;
1656
1657 no_tcp_socket:
1658         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1659                 goto discard_it;
1660
1661         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1662 bad_packet:
1663                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1664         } else {
1665                 tcp_v6_send_reset(NULL, skb);
1666         }
1667
1668 discard_it:
1669
1670         /*
1671          *      Discard frame
1672          */
1673
1674         kfree_skb(skb);
1675         return 0;
1676
1677 discard_and_relse:
1678         sock_put(sk);
1679         goto discard_it;
1680
1681 do_time_wait:
1682         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1683                 inet_twsk_put(inet_twsk(sk));
1684                 goto discard_it;
1685         }
1686
1687         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1688                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1689                 inet_twsk_put(inet_twsk(sk));
1690                 goto discard_it;
1691         }
1692
1693         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1694         case TCP_TW_SYN:
1695         {
1696                 struct sock *sk2;
1697
1698                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1699                                             &ipv6_hdr(skb)->daddr,
1700                                             ntohs(th->dest), inet6_iif(skb));
1701                 if (sk2 != NULL) {
1702                         struct inet_timewait_sock *tw = inet_twsk(sk);
1703                         inet_twsk_deschedule(tw, &tcp_death_row);
1704                         inet_twsk_put(tw);
1705                         sk = sk2;
1706                         goto process;
1707                 }
1708                 /* Fall through to ACK */
1709         }
1710         case TCP_TW_ACK:
1711                 tcp_v6_timewait_ack(sk, skb);
1712                 break;
1713         case TCP_TW_RST:
1714                 goto no_tcp_socket;
1715         case TCP_TW_SUCCESS:;
1716         }
1717         goto discard_it;
1718 }
1719
1720 static int tcp_v6_remember_stamp(struct sock *sk)
1721 {
1722         /* Alas, not yet... */
1723         return 0;
1724 }
1725
1726 static struct inet_connection_sock_af_ops ipv6_specific = {
1727         .queue_xmit        = inet6_csk_xmit,
1728         .send_check        = tcp_v6_send_check,
1729         .rebuild_header    = inet6_sk_rebuild_header,
1730         .conn_request      = tcp_v6_conn_request,
1731         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1732         .remember_stamp    = tcp_v6_remember_stamp,
1733         .net_header_len    = sizeof(struct ipv6hdr),
1734         .setsockopt        = ipv6_setsockopt,
1735         .getsockopt        = ipv6_getsockopt,
1736         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1737         .sockaddr_len      = sizeof(struct sockaddr_in6),
1738         .bind_conflict     = inet6_csk_bind_conflict,
1739 #ifdef CONFIG_COMPAT
1740         .compat_setsockopt = compat_ipv6_setsockopt,
1741         .compat_getsockopt = compat_ipv6_getsockopt,
1742 #endif
1743 };
1744
1745 #ifdef CONFIG_TCP_MD5SIG
1746 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1747         .md5_lookup     =       tcp_v6_md5_lookup,
1748         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1749         .md5_add        =       tcp_v6_md5_add_func,
1750         .md5_parse      =       tcp_v6_parse_md5_keys,
1751 };
1752 #endif
1753
1754 /*
1755  *      TCP over IPv4 via INET6 API
1756  */
1757
1758 static struct inet_connection_sock_af_ops ipv6_mapped = {
1759         .queue_xmit        = ip_queue_xmit,
1760         .send_check        = tcp_v4_send_check,
1761         .rebuild_header    = inet_sk_rebuild_header,
1762         .conn_request      = tcp_v6_conn_request,
1763         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1764         .remember_stamp    = tcp_v4_remember_stamp,
1765         .net_header_len    = sizeof(struct iphdr),
1766         .setsockopt        = ipv6_setsockopt,
1767         .getsockopt        = ipv6_getsockopt,
1768         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1769         .sockaddr_len      = sizeof(struct sockaddr_in6),
1770         .bind_conflict     = inet6_csk_bind_conflict,
1771 #ifdef CONFIG_COMPAT
1772         .compat_setsockopt = compat_ipv6_setsockopt,
1773         .compat_getsockopt = compat_ipv6_getsockopt,
1774 #endif
1775 };
1776
1777 #ifdef CONFIG_TCP_MD5SIG
1778 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1779         .md5_lookup     =       tcp_v4_md5_lookup,
1780         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1781         .md5_add        =       tcp_v6_md5_add_func,
1782         .md5_parse      =       tcp_v6_parse_md5_keys,
1783 };
1784 #endif
1785
1786 /* NOTE: A lot of things set to zero explicitly by call to
1787  *       sk_alloc() so need not be done here.
1788  */
1789 static int tcp_v6_init_sock(struct sock *sk)
1790 {
1791         struct inet_connection_sock *icsk = inet_csk(sk);
1792         struct tcp_sock *tp = tcp_sk(sk);
1793
1794         skb_queue_head_init(&tp->out_of_order_queue);
1795         tcp_init_xmit_timers(sk);
1796         tcp_prequeue_init(tp);
1797
1798         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1799         tp->mdev = TCP_TIMEOUT_INIT;
1800
1801         /* So many TCP implementations out there (incorrectly) count the
1802          * initial SYN frame in their delayed-ACK and congestion control
1803          * algorithms that we must have the following bandaid to talk
1804          * efficiently to them.  -DaveM
1805          */
1806         tp->snd_cwnd = 2;
1807
1808         /* See draft-stevens-tcpca-spec-01 for discussion of the
1809          * initialization of these values.
1810          */
1811         tp->snd_ssthresh = 0x7fffffff;
1812         tp->snd_cwnd_clamp = ~0;
1813         tp->mss_cache = 536;
1814
1815         tp->reordering = sysctl_tcp_reordering;
1816
1817         sk->sk_state = TCP_CLOSE;
1818
1819         icsk->icsk_af_ops = &ipv6_specific;
1820         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1821         icsk->icsk_sync_mss = tcp_sync_mss;
1822         sk->sk_write_space = sk_stream_write_space;
1823         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1824
1825 #ifdef CONFIG_TCP_MD5SIG
1826         tp->af_specific = &tcp_sock_ipv6_specific;
1827 #endif
1828
1829         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1830         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1831
1832         atomic_inc(&tcp_sockets_allocated);
1833
1834         return 0;
1835 }
1836
1837 static void tcp_v6_destroy_sock(struct sock *sk)
1838 {
1839 #ifdef CONFIG_TCP_MD5SIG
1840         /* Clean up the MD5 key list */
1841         if (tcp_sk(sk)->md5sig_info)
1842                 tcp_v6_clear_md5_list(sk);
1843 #endif
1844         tcp_v4_destroy_sock(sk);
1845         inet6_destroy_sock(sk);
1846 }
1847
1848 #ifdef CONFIG_PROC_FS
1849 /* Proc filesystem TCPv6 sock list dumping. */
1850 static void get_openreq6(struct seq_file *seq,
1851                          struct sock *sk, struct request_sock *req, int i, int uid)
1852 {
1853         int ttd = req->expires - jiffies;
1854         struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1855         struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1856
1857         if (ttd < 0)
1858                 ttd = 0;
1859
1860         seq_printf(seq,
1861                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1862                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1863                    i,
1864                    src->s6_addr32[0], src->s6_addr32[1],
1865                    src->s6_addr32[2], src->s6_addr32[3],
1866                    ntohs(inet_rsk(req)->loc_port),
1867                    dest->s6_addr32[0], dest->s6_addr32[1],
1868                    dest->s6_addr32[2], dest->s6_addr32[3],
1869                    ntohs(inet_rsk(req)->rmt_port),
1870                    TCP_SYN_RECV,
1871                    0,0, /* could print option size, but that is af dependent. */
1872                    1,   /* timers active (only the expire timer) */
1873                    jiffies_to_clock_t(ttd),
1874                    req->retrans,
1875                    uid,
1876                    0,  /* non standard timer */
1877                    0, /* open_requests have no inode */
1878                    0, req);
1879 }
1880
1881 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1882 {
1883         struct in6_addr *dest, *src;
1884         __u16 destp, srcp;
1885         int timer_active;
1886         unsigned long timer_expires;
1887         struct inet_sock *inet = inet_sk(sp);
1888         struct tcp_sock *tp = tcp_sk(sp);
1889         const struct inet_connection_sock *icsk = inet_csk(sp);
1890         struct ipv6_pinfo *np = inet6_sk(sp);
1891
1892         dest  = &np->daddr;
1893         src   = &np->rcv_saddr;
1894         destp = ntohs(inet->dport);
1895         srcp  = ntohs(inet->sport);
1896
1897         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1898                 timer_active    = 1;
1899                 timer_expires   = icsk->icsk_timeout;
1900         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1901                 timer_active    = 4;
1902                 timer_expires   = icsk->icsk_timeout;
1903         } else if (timer_pending(&sp->sk_timer)) {
1904                 timer_active    = 2;
1905                 timer_expires   = sp->sk_timer.expires;
1906         } else {
1907                 timer_active    = 0;
1908                 timer_expires = jiffies;
1909         }
1910
1911         seq_printf(seq,
1912                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1913                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %lu %lu %u %u %d\n",
1914                    i,
1915                    src->s6_addr32[0], src->s6_addr32[1],
1916                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1917                    dest->s6_addr32[0], dest->s6_addr32[1],
1918                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1919                    sp->sk_state,
1920                    tp->write_seq-tp->snd_una,
1921                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1922                    timer_active,
1923                    jiffies_to_clock_t(timer_expires - jiffies),
1924                    icsk->icsk_retransmits,
1925                    sock_i_uid(sp),
1926                    icsk->icsk_probes_out,
1927                    sock_i_ino(sp),
1928                    atomic_read(&sp->sk_refcnt), sp,
1929                    jiffies_to_clock_t(icsk->icsk_rto),
1930                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1931                    (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1932                    tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
1933                    );
1934 }
1935
1936 static void get_timewait6_sock(struct seq_file *seq,
1937                                struct inet_timewait_sock *tw, int i)
1938 {
1939         struct in6_addr *dest, *src;
1940         __u16 destp, srcp;
1941         struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
1942         int ttd = tw->tw_ttd - jiffies;
1943
1944         if (ttd < 0)
1945                 ttd = 0;
1946
1947         dest = &tw6->tw_v6_daddr;
1948         src  = &tw6->tw_v6_rcv_saddr;
1949         destp = ntohs(tw->tw_dport);
1950         srcp  = ntohs(tw->tw_sport);
1951
1952         seq_printf(seq,
1953                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1954                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1955                    i,
1956                    src->s6_addr32[0], src->s6_addr32[1],
1957                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1958                    dest->s6_addr32[0], dest->s6_addr32[1],
1959                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1960                    tw->tw_substate, 0, 0,
1961                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
1962                    atomic_read(&tw->tw_refcnt), tw);
1963 }
1964
1965 static int tcp6_seq_show(struct seq_file *seq, void *v)
1966 {
1967         struct tcp_iter_state *st;
1968
1969         if (v == SEQ_START_TOKEN) {
1970                 seq_puts(seq,
1971                          "  sl  "
1972                          "local_address                         "
1973                          "remote_address                        "
1974                          "st tx_queue rx_queue tr tm->when retrnsmt"
1975                          "   uid  timeout inode\n");
1976                 goto out;
1977         }
1978         st = seq->private;
1979
1980         switch (st->state) {
1981         case TCP_SEQ_STATE_LISTENING:
1982         case TCP_SEQ_STATE_ESTABLISHED:
1983                 get_tcp6_sock(seq, v, st->num);
1984                 break;
1985         case TCP_SEQ_STATE_OPENREQ:
1986                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1987                 break;
1988         case TCP_SEQ_STATE_TIME_WAIT:
1989                 get_timewait6_sock(seq, v, st->num);
1990                 break;
1991         }
1992 out:
1993         return 0;
1994 }
1995
1996 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1997         .name           = "tcp6",
1998         .family         = AF_INET6,
1999         .seq_fops       = {
2000                 .owner          = THIS_MODULE,
2001         },
2002         .seq_ops        = {
2003                 .show           = tcp6_seq_show,
2004         },
2005 };
2006
2007 int tcp6_proc_init(struct net *net)
2008 {
2009         return tcp_proc_register(net, &tcp6_seq_afinfo);
2010 }
2011
2012 void tcp6_proc_exit(struct net *net)
2013 {
2014         tcp_proc_unregister(net, &tcp6_seq_afinfo);
2015 }
2016 #endif
2017
2018 struct proto tcpv6_prot = {
2019         .name                   = "TCPv6",
2020         .owner                  = THIS_MODULE,
2021         .close                  = tcp_close,
2022         .connect                = tcp_v6_connect,
2023         .disconnect             = tcp_disconnect,
2024         .accept                 = inet_csk_accept,
2025         .ioctl                  = tcp_ioctl,
2026         .init                   = tcp_v6_init_sock,
2027         .destroy                = tcp_v6_destroy_sock,
2028         .shutdown               = tcp_shutdown,
2029         .setsockopt             = tcp_setsockopt,
2030         .getsockopt             = tcp_getsockopt,
2031         .recvmsg                = tcp_recvmsg,
2032         .backlog_rcv            = tcp_v6_do_rcv,
2033         .hash                   = tcp_v6_hash,
2034         .unhash                 = inet_unhash,
2035         .get_port               = inet_csk_get_port,
2036         .enter_memory_pressure  = tcp_enter_memory_pressure,
2037         .sockets_allocated      = &tcp_sockets_allocated,
2038         .memory_allocated       = &tcp_memory_allocated,
2039         .memory_pressure        = &tcp_memory_pressure,
2040         .orphan_count           = &tcp_orphan_count,
2041         .sysctl_mem             = sysctl_tcp_mem,
2042         .sysctl_wmem            = sysctl_tcp_wmem,
2043         .sysctl_rmem            = sysctl_tcp_rmem,
2044         .max_header             = MAX_TCP_HEADER,
2045         .obj_size               = sizeof(struct tcp6_sock),
2046         .twsk_prot              = &tcp6_timewait_sock_ops,
2047         .rsk_prot               = &tcp6_request_sock_ops,
2048         .h.hashinfo             = &tcp_hashinfo,
2049 #ifdef CONFIG_COMPAT
2050         .compat_setsockopt      = compat_tcp_setsockopt,
2051         .compat_getsockopt      = compat_tcp_getsockopt,
2052 #endif
2053 };
2054
2055 static struct inet6_protocol tcpv6_protocol = {
2056         .handler        =       tcp_v6_rcv,
2057         .err_handler    =       tcp_v6_err,
2058         .gso_send_check =       tcp_v6_gso_send_check,
2059         .gso_segment    =       tcp_tso_segment,
2060         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2061 };
2062
2063 static struct inet_protosw tcpv6_protosw = {
2064         .type           =       SOCK_STREAM,
2065         .protocol       =       IPPROTO_TCP,
2066         .prot           =       &tcpv6_prot,
2067         .ops            =       &inet6_stream_ops,
2068         .capability     =       -1,
2069         .no_check       =       0,
2070         .flags          =       INET_PROTOSW_PERMANENT |
2071                                 INET_PROTOSW_ICSK,
2072 };
2073
2074 static int tcpv6_net_init(struct net *net)
2075 {
2076         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2077                                     SOCK_RAW, IPPROTO_TCP, net);
2078 }
2079
2080 static void tcpv6_net_exit(struct net *net)
2081 {
2082         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2083         inet_twsk_purge(net, &tcp_hashinfo, &tcp_death_row, AF_INET6);
2084 }
2085
2086 static struct pernet_operations tcpv6_net_ops = {
2087         .init = tcpv6_net_init,
2088         .exit = tcpv6_net_exit,
2089 };
2090
2091 int __init tcpv6_init(void)
2092 {
2093         int ret;
2094
2095         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2096         if (ret)
2097                 goto out;
2098
2099         /* register inet6 protocol */
2100         ret = inet6_register_protosw(&tcpv6_protosw);
2101         if (ret)
2102                 goto out_tcpv6_protocol;
2103
2104         ret = register_pernet_subsys(&tcpv6_net_ops);
2105         if (ret)
2106                 goto out_tcpv6_protosw;
2107 out:
2108         return ret;
2109
2110 out_tcpv6_protocol:
2111         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2112 out_tcpv6_protosw:
2113         inet6_unregister_protosw(&tcpv6_protosw);
2114         goto out;
2115 }
2116
2117 void tcpv6_exit(void)
2118 {
2119         unregister_pernet_subsys(&tcpv6_net_ops);
2120         inet6_unregister_protosw(&tcpv6_protosw);
2121         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2122 }