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_sk(sk)->sport;
 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 "
 876                                "(" NIP6_FMT ", %u)->"
 877                                "(" NIP6_FMT ", %u)\n",
 878                                genhash ? "failed" : "mismatch",
 879                                NIP6(ip6h->saddr), ntohs(th->source),
 880                                NIP6(ip6h->daddr), ntohs(th->dest));
 881                 }
 882                 return 1;
 883         }
 884         return 0;
 885 }
 886 #endif
 887
 888 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 889         .family         =       AF_INET6,
 890         .obj_size       =       sizeof(struct tcp6_request_sock),
 891         .rtx_syn_ack    =       tcp_v6_send_synack,
 892         .send_ack       =       tcp_v6_reqsk_send_ack,
 893         .destructor     =       tcp_v6_reqsk_destructor,
 894         .send_reset     =       tcp_v6_send_reset
 895 };
 896
 897 #ifdef CONFIG_TCP_MD5SIG
 898 static struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 899         .md5_lookup     =       tcp_v6_reqsk_md5_lookup,
 900 };
 901 #endif
 902
 903 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
 904         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
 905         .twsk_unique    = tcp_twsk_unique,
 906         .twsk_destructor= tcp_twsk_destructor,
 907 };
 908
 909 static void tcp_v6_send_check(struct sock *sk, int len, struct sk_buff *skb)
 910 {
 911         struct ipv6_pinfo *np = inet6_sk(sk);
 912         struct tcphdr *th = tcp_hdr(skb);
 913
 914         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 915                 th->check = ~csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,  0);
 916                 skb->csum_start = skb_transport_header(skb) - skb->head;
 917                 skb->csum_offset = offsetof(struct tcphdr, check);
 918         } else {
 919                 th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP,
 920                                             csum_partial((char *)th, th->doff<<2,
 921                                                          skb->csum));
 922         }
 923 }
 924
 925 static int tcp_v6_gso_send_check(struct sk_buff *skb)
 926 {
 927         struct ipv6hdr *ipv6h;
 928         struct tcphdr *th;
 929
 930         if (!pskb_may_pull(skb, sizeof(*th)))
 931                 return -EINVAL;
 932
 933         ipv6h = ipv6_hdr(skb);
 934         th = tcp_hdr(skb);
 935
 936         th->check = 0;
 937         th->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len,
 938                                      IPPROTO_TCP, 0);
 939         skb->csum_start = skb_transport_header(skb) - skb->head;
 940         skb->csum_offset = offsetof(struct tcphdr, check);
 941         skb->ip_summed = CHECKSUM_PARTIAL;
 942         return 0;
 943 }
 944
 945 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
 946                                  u32 ts, struct tcp_md5sig_key *key, int rst)
 947 {
 948         struct tcphdr *th = tcp_hdr(skb), *t1;
 949         struct sk_buff *buff;
 950         struct flowi fl;
 951         struct net *net = dev_net(skb->dst->dev);
 952         struct sock *ctl_sk = net->ipv6.tcp_sk;
 953         unsigned int tot_len = sizeof(struct tcphdr);
 954         __be32 *topt;
 955
 956         if (ts)
 957                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 958 #ifdef CONFIG_TCP_MD5SIG
 959         if (key)
 960                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 961 #endif
 962
 963         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 964                          GFP_ATOMIC);
 965         if (buff == NULL)
 966                 return;
 967
 968         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 969
 970         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 971
 972         /* Swap the send and the receive. */
 973         memset(t1, 0, sizeof(*t1));
 974         t1->dest = th->source;
 975         t1->source = th->dest;
 976         t1->doff = tot_len / 4;
 977         t1->seq = htonl(seq);
 978         t1->ack_seq = htonl(ack);
 979         t1->ack = !rst || !th->ack;
 980         t1->rst = rst;
 981         t1->window = htons(win);
 982
 983         topt = (__be32 *)(t1 + 1);
 984
 985         if (ts) {
 986                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 987                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 988                 *topt++ = htonl(tcp_time_stamp);
 989                 *topt++ = htonl(ts);
 990         }
 991
 992 #ifdef CONFIG_TCP_MD5SIG
 993         if (key) {
 994                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 995                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 996                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 997                                     &ipv6_hdr(skb)->saddr,
 998                                     &ipv6_hdr(skb)->daddr, t1);
 999         }
1000 #endif
1001
1002         buff->csum = csum_partial((char *)t1, tot_len, 0);
1003
1004         memset(&fl, 0, sizeof(fl));
1005         ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1006         ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1007
1008         t1->check = csum_ipv6_magic(&fl.fl6_src, &fl.fl6_dst,
1009                                     tot_len, IPPROTO_TCP,
1010                                     buff->csum);
1011
1012         fl.proto = IPPROTO_TCP;
1013         fl.oif = inet6_iif(skb);
1014         fl.fl_ip_dport = t1->dest;
1015         fl.fl_ip_sport = t1->source;
1016         security_skb_classify_flow(skb, &fl);
1017
1018         /* Pass a socket to ip6_dst_lookup either it is for RST
1019          * Underlying function will use this to retrieve the network
1020          * namespace
1021          */
1022         if (!ip6_dst_lookup(ctl_sk, &buff->dst, &fl)) {
1023                 if (xfrm_lookup(&buff->dst, &fl, NULL, 0) >= 0) {
1024                         ip6_xmit(ctl_sk, buff, &fl, NULL, 0);
1025                         TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1026                         if (rst)
1027                                 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1028                         return;
1029                 }
1030         }
1031
1032         kfree_skb(buff);
1033 }
1034
1035 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
1036 {
1037         struct tcphdr *th = tcp_hdr(skb);
1038         u32 seq = 0, ack_seq = 0;
1039 #ifdef CONFIG_TCP_MD5SIG
1040         struct tcp_md5sig_key *key;
1041 #endif
1042
1043         if (th->rst)
1044                 return;
1045
1046         if (!ipv6_unicast_destination(skb))
1047                 return;
1048
1049 #ifdef CONFIG_TCP_MD5SIG
1050         if (sk)
1051                 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1052         else
1053                 key = NULL;
1054 #endif
1055
1056         if (th->ack)
1057                 seq = ntohl(th->ack_seq);
1058         else
1059                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1060                           (th->doff << 2);
1061
1062         tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1);
1063 }
1064
1065 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1066                             struct tcp_md5sig_key *key)
1067 {
1068         tcp_v6_send_response(skb, seq, ack, win, ts, key, 0);
1069 }
1070
1071 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1072 {
1073         struct inet_timewait_sock *tw = inet_twsk(sk);
1074         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1075
1076         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1077                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1078                         tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw));
1079
1080         inet_twsk_put(tw);
1081 }
1082
1083 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1084                                   struct request_sock *req)
1085 {
1086         tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1087                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr));
1088 }
1089
1090
1091 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1092 {
1093         struct request_sock *req, **prev;
1094         const struct tcphdr *th = tcp_hdr(skb);
1095         struct sock *nsk;
1096
1097         /* Find possible connection requests. */
1098         req = inet6_csk_search_req(sk, &prev, th->source,
1099                                    &ipv6_hdr(skb)->saddr,
1100                                    &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1101         if (req)
1102                 return tcp_check_req(sk, skb, req, prev);
1103
1104         nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1105                         &ipv6_hdr(skb)->saddr, th->source,
1106                         &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1107
1108         if (nsk) {
1109                 if (nsk->sk_state != TCP_TIME_WAIT) {
1110                         bh_lock_sock(nsk);
1111                         return nsk;
1112                 }
1113                 inet_twsk_put(inet_twsk(nsk));
1114                 return NULL;
1115         }
1116
1117 #ifdef CONFIG_SYN_COOKIES
1118         if (!th->rst && !th->syn && th->ack)
1119                 sk = cookie_v6_check(sk, skb);
1120 #endif
1121         return sk;
1122 }
1123
1124 /* FIXME: this is substantially similar to the ipv4 code.
1125  * Can some kind of merge be done? -- erics
1126  */
1127 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1128 {
1129         struct inet6_request_sock *treq;
1130         struct ipv6_pinfo *np = inet6_sk(sk);
1131         struct tcp_options_received tmp_opt;
1132         struct tcp_sock *tp = tcp_sk(sk);
1133         struct request_sock *req = NULL;
1134         __u32 isn = TCP_SKB_CB(skb)->when;
1135 #ifdef CONFIG_SYN_COOKIES
1136         int want_cookie = 0;
1137 #else
1138 #define want_cookie 0
1139 #endif
1140
1141         if (skb->protocol == htons(ETH_P_IP))
1142                 return tcp_v4_conn_request(sk, skb);
1143
1144         if (!ipv6_unicast_destination(skb))
1145                 goto drop;
1146
1147         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1148                 if (net_ratelimit())
1149                         syn_flood_warning(skb);
1150 #ifdef CONFIG_SYN_COOKIES
1151                 if (sysctl_tcp_syncookies)
1152                         want_cookie = 1;
1153                 else
1154 #endif
1155                 goto drop;
1156         }
1157
1158         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1159                 goto drop;
1160
1161         req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1162         if (req == NULL)
1163                 goto drop;
1164
1165 #ifdef CONFIG_TCP_MD5SIG
1166         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1167 #endif
1168
1169         tcp_clear_options(&tmp_opt);
1170         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1171         tmp_opt.user_mss = tp->rx_opt.user_mss;
1172
1173         tcp_parse_options(skb, &tmp_opt, 0);
1174
1175         if (want_cookie && !tmp_opt.saw_tstamp)
1176                 tcp_clear_options(&tmp_opt);
1177
1178         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1179         tcp_openreq_init(req, &tmp_opt, skb);
1180
1181         treq = inet6_rsk(req);
1182         ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1183         ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1184         if (!want_cookie)
1185                 TCP_ECN_create_request(req, tcp_hdr(skb));
1186
1187         if (want_cookie) {
1188                 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1189                 req->cookie_ts = tmp_opt.tstamp_ok;
1190         } else if (!isn) {
1191                 if (ipv6_opt_accepted(sk, skb) ||
1192                     np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1193                     np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1194                         atomic_inc(&skb->users);
1195                         treq->pktopts = skb;
1196                 }
1197                 treq->iif = sk->sk_bound_dev_if;
1198
1199                 /* So that link locals have meaning */
1200                 if (!sk->sk_bound_dev_if &&
1201                     ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1202                         treq->iif = inet6_iif(skb);
1203
1204                 isn = tcp_v6_init_sequence(skb);
1205         }
1206
1207         tcp_rsk(req)->snt_isn = isn;
1208
1209         security_inet_conn_request(sk, skb, req);
1210
1211         if (tcp_v6_send_synack(sk, req))
1212                 goto drop;
1213
1214         if (!want_cookie) {
1215                 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1216                 return 0;
1217         }
1218
1219 drop:
1220         if (req)
1221                 reqsk_free(req);
1222
1223         return 0; /* don't send reset */
1224 }
1225
1226 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1227                                           struct request_sock *req,
1228                                           struct dst_entry *dst)
1229 {
1230         struct inet6_request_sock *treq;
1231         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1232         struct tcp6_sock *newtcp6sk;
1233         struct inet_sock *newinet;
1234         struct tcp_sock *newtp;
1235         struct sock *newsk;
1236         struct ipv6_txoptions *opt;
1237 #ifdef CONFIG_TCP_MD5SIG
1238         struct tcp_md5sig_key *key;
1239 #endif
1240
1241         if (skb->protocol == htons(ETH_P_IP)) {
1242                 /*
1243                  *      v6 mapped
1244                  */
1245
1246                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1247
1248                 if (newsk == NULL)
1249                         return NULL;
1250
1251                 newtcp6sk = (struct tcp6_sock *)newsk;
1252                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1253
1254                 newinet = inet_sk(newsk);
1255                 newnp = inet6_sk(newsk);
1256                 newtp = tcp_sk(newsk);
1257
1258                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1259
1260                 ipv6_addr_set(&newnp->daddr, 0, 0, htonl(0x0000FFFF),
1261                               newinet->daddr);
1262
1263                 ipv6_addr_set(&newnp->saddr, 0, 0, htonl(0x0000FFFF),
1264                               newinet->saddr);
1265
1266                 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1267
1268                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1269                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1270 #ifdef CONFIG_TCP_MD5SIG
1271                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1272 #endif
1273
1274                 newnp->pktoptions  = NULL;
1275                 newnp->opt         = NULL;
1276                 newnp->mcast_oif   = inet6_iif(skb);
1277                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1278
1279                 /*
1280                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1281                  * here, tcp_create_openreq_child now does this for us, see the comment in
1282                  * that function for the gory details. -acme
1283                  */
1284
1285                 /* It is tricky place. Until this moment IPv4 tcp
1286                    worked with IPv6 icsk.icsk_af_ops.
1287                    Sync it now.
1288                  */
1289                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1290
1291                 return newsk;
1292         }
1293
1294         treq = inet6_rsk(req);
1295         opt = np->opt;
1296
1297         if (sk_acceptq_is_full(sk))
1298                 goto out_overflow;
1299
1300         if (dst == NULL) {
1301                 struct in6_addr *final_p = NULL, final;
1302                 struct flowi fl;
1303
1304                 memset(&fl, 0, sizeof(fl));
1305                 fl.proto = IPPROTO_TCP;
1306                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1307                 if (opt && opt->srcrt) {
1308                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1309                         ipv6_addr_copy(&final, &fl.fl6_dst);
1310                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1311                         final_p = &final;
1312                 }
1313                 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1314                 fl.oif = sk->sk_bound_dev_if;
1315                 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1316                 fl.fl_ip_sport = inet_sk(sk)->sport;
1317                 security_req_classify_flow(req, &fl);
1318
1319                 if (ip6_dst_lookup(sk, &dst, &fl))
1320                         goto out;
1321
1322                 if (final_p)
1323                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1324
1325                 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
1326                         goto out;
1327         }
1328
1329         newsk = tcp_create_openreq_child(sk, req, skb);
1330         if (newsk == NULL)
1331                 goto out;
1332
1333         /*
1334          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1335          * count here, tcp_create_openreq_child now does this for us, see the
1336          * comment in that function for the gory details. -acme
1337          */
1338
1339         newsk->sk_gso_type = SKB_GSO_TCPV6;
1340         __ip6_dst_store(newsk, dst, NULL, NULL);
1341
1342         newtcp6sk = (struct tcp6_sock *)newsk;
1343         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1344
1345         newtp = tcp_sk(newsk);
1346         newinet = inet_sk(newsk);
1347         newnp = inet6_sk(newsk);
1348
1349         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1350
1351         ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1352         ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1353         ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1354         newsk->sk_bound_dev_if = treq->iif;
1355
1356         /* Now IPv6 options...
1357
1358            First: no IPv4 options.
1359          */
1360         newinet->opt = NULL;
1361         newnp->ipv6_fl_list = NULL;
1362
1363         /* Clone RX bits */
1364         newnp->rxopt.all = np->rxopt.all;
1365
1366         /* Clone pktoptions received with SYN */
1367         newnp->pktoptions = NULL;
1368         if (treq->pktopts != NULL) {
1369                 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1370                 kfree_skb(treq->pktopts);
1371                 treq->pktopts = NULL;
1372                 if (newnp->pktoptions)
1373                         skb_set_owner_r(newnp->pktoptions, newsk);
1374         }
1375         newnp->opt        = NULL;
1376         newnp->mcast_oif  = inet6_iif(skb);
1377         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1378
1379         /* Clone native IPv6 options from listening socket (if any)
1380
1381            Yes, keeping reference count would be much more clever,
1382            but we make one more one thing there: reattach optmem
1383            to newsk.
1384          */
1385         if (opt) {
1386                 newnp->opt = ipv6_dup_options(newsk, opt);
1387                 if (opt != np->opt)
1388                         sock_kfree_s(sk, opt, opt->tot_len);
1389         }
1390
1391         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1392         if (newnp->opt)
1393                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1394                                                      newnp->opt->opt_flen);
1395
1396         tcp_mtup_init(newsk);
1397         tcp_sync_mss(newsk, dst_mtu(dst));
1398         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1399         tcp_initialize_rcv_mss(newsk);
1400
1401         newinet->daddr = newinet->saddr = newinet->rcv_saddr = LOOPBACK4_IPV6;
1402
1403 #ifdef CONFIG_TCP_MD5SIG
1404         /* Copy over the MD5 key from the original socket */
1405         if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1406                 /* We're using one, so create a matching key
1407                  * on the newsk structure. If we fail to get
1408                  * memory, then we end up not copying the key
1409                  * across. Shucks.
1410                  */
1411                 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1412                 if (newkey != NULL)
1413                         tcp_v6_md5_do_add(newsk, &inet6_sk(sk)->daddr,
1414                                           newkey, key->keylen);
1415         }
1416 #endif
1417
1418         __inet6_hash(newsk);
1419         __inet_inherit_port(sk, newsk);
1420
1421         return newsk;
1422
1423 out_overflow:
1424         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1425 out:
1426         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1427         if (opt && opt != np->opt)
1428                 sock_kfree_s(sk, opt, opt->tot_len);
1429         dst_release(dst);
1430         return NULL;
1431 }
1432
1433 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1434 {
1435         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1436                 if (!tcp_v6_check(tcp_hdr(skb), skb->len, &ipv6_hdr(skb)->saddr,
1437                                   &ipv6_hdr(skb)->daddr, skb->csum)) {
1438                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1439                         return 0;
1440                 }
1441         }
1442
1443         skb->csum = ~csum_unfold(tcp_v6_check(tcp_hdr(skb), skb->len,
1444                                               &ipv6_hdr(skb)->saddr,
1445                                               &ipv6_hdr(skb)->daddr, 0));
1446
1447         if (skb->len <= 76) {
1448                 return __skb_checksum_complete(skb);
1449         }
1450         return 0;
1451 }
1452
1453 /* The socket must have it's spinlock held when we get
1454  * here.
1455  *
1456  * We have a potential double-lock case here, so even when
1457  * doing backlog processing we use the BH locking scheme.
1458  * This is because we cannot sleep with the original spinlock
1459  * held.
1460  */
1461 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1462 {
1463         struct ipv6_pinfo *np = inet6_sk(sk);
1464         struct tcp_sock *tp;
1465         struct sk_buff *opt_skb = NULL;
1466
1467         /* Imagine: socket is IPv6. IPv4 packet arrives,
1468            goes to IPv4 receive handler and backlogged.
1469            From backlog it always goes here. Kerboom...
1470            Fortunately, tcp_rcv_established and rcv_established
1471            handle them correctly, but it is not case with
1472            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1473          */
1474
1475         if (skb->protocol == htons(ETH_P_IP))
1476                 return tcp_v4_do_rcv(sk, skb);
1477
1478 #ifdef CONFIG_TCP_MD5SIG
1479         if (tcp_v6_inbound_md5_hash (sk, skb))
1480                 goto discard;
1481 #endif
1482
1483         if (sk_filter(sk, skb))
1484                 goto discard;
1485
1486         /*
1487          *      socket locking is here for SMP purposes as backlog rcv
1488          *      is currently called with bh processing disabled.
1489          */
1490
1491         /* Do Stevens' IPV6_PKTOPTIONS.
1492
1493            Yes, guys, it is the only place in our code, where we
1494            may make it not affecting IPv4.
1495            The rest of code is protocol independent,
1496            and I do not like idea to uglify IPv4.
1497
1498            Actually, all the idea behind IPV6_PKTOPTIONS
1499            looks not very well thought. For now we latch
1500            options, received in the last packet, enqueued
1501            by tcp. Feel free to propose better solution.
1502                                                --ANK (980728)
1503          */
1504         if (np->rxopt.all)
1505                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1506
1507         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1508                 TCP_CHECK_TIMER(sk);
1509                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1510                         goto reset;
1511                 TCP_CHECK_TIMER(sk);
1512                 if (opt_skb)
1513                         goto ipv6_pktoptions;
1514                 return 0;
1515         }
1516
1517         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1518                 goto csum_err;
1519
1520         if (sk->sk_state == TCP_LISTEN) {
1521                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1522                 if (!nsk)
1523                         goto discard;
1524
1525                 /*
1526                  * Queue it on the new socket if the new socket is active,
1527                  * otherwise we just shortcircuit this and continue with
1528                  * the new socket..
1529                  */
1530                 if(nsk != sk) {
1531                         if (tcp_child_process(sk, nsk, skb))
1532                                 goto reset;
1533                         if (opt_skb)
1534                                 __kfree_skb(opt_skb);
1535                         return 0;
1536                 }
1537         }
1538
1539         TCP_CHECK_TIMER(sk);
1540         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1541                 goto reset;
1542         TCP_CHECK_TIMER(sk);
1543         if (opt_skb)
1544                 goto ipv6_pktoptions;
1545         return 0;
1546
1547 reset:
1548         tcp_v6_send_reset(sk, skb);
1549 discard:
1550         if (opt_skb)
1551                 __kfree_skb(opt_skb);
1552         kfree_skb(skb);
1553         return 0;
1554 csum_err:
1555         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1556         goto discard;
1557
1558
1559 ipv6_pktoptions:
1560         /* Do you ask, what is it?
1561
1562            1. skb was enqueued by tcp.
1563            2. skb is added to tail of read queue, rather than out of order.
1564            3. socket is not in passive state.
1565            4. Finally, it really contains options, which user wants to receive.
1566          */
1567         tp = tcp_sk(sk);
1568         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1569             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1570                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1571                         np->mcast_oif = inet6_iif(opt_skb);
1572                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1573                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1574                 if (ipv6_opt_accepted(sk, opt_skb)) {
1575                         skb_set_owner_r(opt_skb, sk);
1576                         opt_skb = xchg(&np->pktoptions, opt_skb);
1577                 } else {
1578                         __kfree_skb(opt_skb);
1579                         opt_skb = xchg(&np->pktoptions, NULL);
1580                 }
1581         }
1582
1583         if (opt_skb)
1584                 kfree_skb(opt_skb);
1585         return 0;
1586 }
1587
1588 static int tcp_v6_rcv(struct sk_buff *skb)
1589 {
1590         struct tcphdr *th;
1591         struct sock *sk;
1592         int ret;
1593         struct net *net = dev_net(skb->dev);
1594
1595         if (skb->pkt_type != PACKET_HOST)
1596                 goto discard_it;
1597
1598         /*
1599          *      Count it even if it's bad.
1600          */
1601         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1602
1603         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1604                 goto discard_it;
1605
1606         th = tcp_hdr(skb);
1607
1608         if (th->doff < sizeof(struct tcphdr)/4)
1609                 goto bad_packet;
1610         if (!pskb_may_pull(skb, th->doff*4))
1611                 goto discard_it;
1612
1613         if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1614                 goto bad_packet;
1615
1616         th = tcp_hdr(skb);
1617         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1618         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1619                                     skb->len - th->doff*4);
1620         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1621         TCP_SKB_CB(skb)->when = 0;
1622         TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1623         TCP_SKB_CB(skb)->sacked = 0;
1624
1625         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1626         if (!sk)
1627                 goto no_tcp_socket;
1628
1629 process:
1630         if (sk->sk_state == TCP_TIME_WAIT)
1631                 goto do_time_wait;
1632
1633         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1634                 goto discard_and_relse;
1635
1636         if (sk_filter(sk, skb))
1637                 goto discard_and_relse;
1638
1639         skb->dev = NULL;
1640
1641         bh_lock_sock_nested(sk);
1642         ret = 0;
1643         if (!sock_owned_by_user(sk)) {
1644 #ifdef CONFIG_NET_DMA
1645                 struct tcp_sock *tp = tcp_sk(sk);
1646                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1647                         tp->ucopy.dma_chan = get_softnet_dma();
1648                 if (tp->ucopy.dma_chan)
1649                         ret = tcp_v6_do_rcv(sk, skb);
1650                 else
1651 #endif
1652                 {
1653                         if (!tcp_prequeue(sk, skb))
1654                                 ret = tcp_v6_do_rcv(sk, skb);
1655                 }
1656         } else
1657                 sk_add_backlog(sk, skb);
1658         bh_unlock_sock(sk);
1659
1660         sock_put(sk);
1661         return ret ? -1 : 0;
1662
1663 no_tcp_socket:
1664         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1665                 goto discard_it;
1666
1667         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1668 bad_packet:
1669                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1670         } else {
1671                 tcp_v6_send_reset(NULL, skb);
1672         }
1673
1674 discard_it:
1675
1676         /*
1677          *      Discard frame
1678          */
1679
1680         kfree_skb(skb);
1681         return 0;
1682
1683 discard_and_relse:
1684         sock_put(sk);
1685         goto discard_it;
1686
1687 do_time_wait:
1688         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1689                 inet_twsk_put(inet_twsk(sk));
1690                 goto discard_it;
1691         }
1692
1693         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1694                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1695                 inet_twsk_put(inet_twsk(sk));
1696                 goto discard_it;
1697         }
1698
1699         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1700         case TCP_TW_SYN:
1701         {
1702                 struct sock *sk2;
1703
1704                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1705                                             &ipv6_hdr(skb)->daddr,
1706                                             ntohs(th->dest), inet6_iif(skb));
1707                 if (sk2 != NULL) {
1708                         struct inet_timewait_sock *tw = inet_twsk(sk);
1709                         inet_twsk_deschedule(tw, &tcp_death_row);
1710                         inet_twsk_put(tw);
1711                         sk = sk2;
1712                         goto process;
1713                 }
1714                 /* Fall through to ACK */
1715         }
1716         case TCP_TW_ACK:
1717                 tcp_v6_timewait_ack(sk, skb);
1718                 break;
1719         case TCP_TW_RST:
1720                 goto no_tcp_socket;
1721         case TCP_TW_SUCCESS:;
1722         }
1723         goto discard_it;
1724 }
1725
1726 static int tcp_v6_remember_stamp(struct sock *sk)
1727 {
1728         /* Alas, not yet... */
1729         return 0;
1730 }
1731
1732 static struct inet_connection_sock_af_ops ipv6_specific = {
1733         .queue_xmit        = inet6_csk_xmit,
1734         .send_check        = tcp_v6_send_check,
1735         .rebuild_header    = inet6_sk_rebuild_header,
1736         .conn_request      = tcp_v6_conn_request,
1737         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1738         .remember_stamp    = tcp_v6_remember_stamp,
1739         .net_header_len    = sizeof(struct ipv6hdr),
1740         .setsockopt        = ipv6_setsockopt,
1741         .getsockopt        = ipv6_getsockopt,
1742         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1743         .sockaddr_len      = sizeof(struct sockaddr_in6),
1744         .bind_conflict     = inet6_csk_bind_conflict,
1745 #ifdef CONFIG_COMPAT
1746         .compat_setsockopt = compat_ipv6_setsockopt,
1747         .compat_getsockopt = compat_ipv6_getsockopt,
1748 #endif
1749 };
1750
1751 #ifdef CONFIG_TCP_MD5SIG
1752 static struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1753         .md5_lookup     =       tcp_v6_md5_lookup,
1754         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1755         .md5_add        =       tcp_v6_md5_add_func,
1756         .md5_parse      =       tcp_v6_parse_md5_keys,
1757 };
1758 #endif
1759
1760 /*
1761  *      TCP over IPv4 via INET6 API
1762  */
1763
1764 static struct inet_connection_sock_af_ops ipv6_mapped = {
1765         .queue_xmit        = ip_queue_xmit,
1766         .send_check        = tcp_v4_send_check,
1767         .rebuild_header    = inet_sk_rebuild_header,
1768         .conn_request      = tcp_v6_conn_request,
1769         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1770         .remember_stamp    = tcp_v4_remember_stamp,
1771         .net_header_len    = sizeof(struct iphdr),
1772         .setsockopt        = ipv6_setsockopt,
1773         .getsockopt        = ipv6_getsockopt,
1774         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1775         .sockaddr_len      = sizeof(struct sockaddr_in6),
1776         .bind_conflict     = inet6_csk_bind_conflict,
1777 #ifdef CONFIG_COMPAT
1778         .compat_setsockopt = compat_ipv6_setsockopt,
1779         .compat_getsockopt = compat_ipv6_getsockopt,
1780 #endif
1781 };
1782
1783 #ifdef CONFIG_TCP_MD5SIG
1784 static struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1785         .md5_lookup     =       tcp_v4_md5_lookup,
1786         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1787         .md5_add        =       tcp_v6_md5_add_func,
1788         .md5_parse      =       tcp_v6_parse_md5_keys,
1789 };
1790 #endif
1791
1792 /* NOTE: A lot of things set to zero explicitly by call to
1793  *       sk_alloc() so need not be done here.
1794  */
1795 static int tcp_v6_init_sock(struct sock *sk)
1796 {
1797         struct inet_connection_sock *icsk = inet_csk(sk);
1798         struct tcp_sock *tp = tcp_sk(sk);
1799
1800         skb_queue_head_init(&tp->out_of_order_queue);
1801         tcp_init_xmit_timers(sk);
1802         tcp_prequeue_init(tp);
1803
1804         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1805         tp->mdev = TCP_TIMEOUT_INIT;
1806
1807         /* So many TCP implementations out there (incorrectly) count the
1808          * initial SYN frame in their delayed-ACK and congestion control
1809          * algorithms that we must have the following bandaid to talk
1810          * efficiently to them.  -DaveM
1811          */
1812         tp->snd_cwnd = 2;
1813
1814         /* See draft-stevens-tcpca-spec-01 for discussion of the
1815          * initialization of these values.
1816          */
1817         tp->snd_ssthresh = 0x7fffffff;
1818         tp->snd_cwnd_clamp = ~0;
1819         tp->mss_cache = 536;
1820
1821         tp->reordering = sysctl_tcp_reordering;
1822
1823         sk->sk_state = TCP_CLOSE;
1824
1825         icsk->icsk_af_ops = &ipv6_specific;
1826         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1827         icsk->icsk_sync_mss = tcp_sync_mss;
1828         sk->sk_write_space = sk_stream_write_space;
1829         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1830
1831 #ifdef CONFIG_TCP_MD5SIG
1832         tp->af_specific = &tcp_sock_ipv6_specific;
1833 #endif
1834
1835         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1836         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1837
1838         atomic_inc(&tcp_sockets_allocated);
1839
1840         return 0;
1841 }
1842
1843 static void tcp_v6_destroy_sock(struct sock *sk)
1844 {
1845 #ifdef CONFIG_TCP_MD5SIG
1846         /* Clean up the MD5 key list */
1847         if (tcp_sk(sk)->md5sig_info)
1848                 tcp_v6_clear_md5_list(sk);
1849 #endif
1850         tcp_v4_destroy_sock(sk);
1851         inet6_destroy_sock(sk);
1852 }
1853
1854 #ifdef CONFIG_PROC_FS
1855 /* Proc filesystem TCPv6 sock list dumping. */
1856 static void get_openreq6(struct seq_file *seq,
1857                          struct sock *sk, struct request_sock *req, int i, int uid)
1858 {
1859         int ttd = req->expires - jiffies;
1860         struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1861         struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1862
1863         if (ttd < 0)
1864                 ttd = 0;
1865
1866         seq_printf(seq,
1867                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1868                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1869                    i,
1870                    src->s6_addr32[0], src->s6_addr32[1],
1871                    src->s6_addr32[2], src->s6_addr32[3],
1872                    ntohs(inet_sk(sk)->sport),
1873                    dest->s6_addr32[0], dest->s6_addr32[1],
1874                    dest->s6_addr32[2], dest->s6_addr32[3],
1875                    ntohs(inet_rsk(req)->rmt_port),
1876                    TCP_SYN_RECV,
1877                    0,0, /* could print option size, but that is af dependent. */
1878                    1,   /* timers active (only the expire timer) */
1879                    jiffies_to_clock_t(ttd),
1880                    req->retrans,
1881                    uid,
1882                    0,  /* non standard timer */
1883                    0, /* open_requests have no inode */
1884                    0, req);
1885 }
1886
1887 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1888 {
1889         struct in6_addr *dest, *src;
1890         __u16 destp, srcp;
1891         int timer_active;
1892         unsigned long timer_expires;
1893         struct inet_sock *inet = inet_sk(sp);
1894         struct tcp_sock *tp = tcp_sk(sp);
1895         const struct inet_connection_sock *icsk = inet_csk(sp);
1896         struct ipv6_pinfo *np = inet6_sk(sp);
1897
1898         dest  = &np->daddr;
1899         src   = &np->rcv_saddr;
1900         destp = ntohs(inet->dport);
1901         srcp  = ntohs(inet->sport);
1902
1903         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1904                 timer_active    = 1;
1905                 timer_expires   = icsk->icsk_timeout;
1906         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1907                 timer_active    = 4;
1908                 timer_expires   = icsk->icsk_timeout;
1909         } else if (timer_pending(&sp->sk_timer)) {
1910                 timer_active    = 2;
1911                 timer_expires   = sp->sk_timer.expires;
1912         } else {
1913                 timer_active    = 0;
1914                 timer_expires = jiffies;
1915         }
1916
1917         seq_printf(seq,
1918                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1919                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %lu %lu %u %u %d\n",
1920                    i,
1921                    src->s6_addr32[0], src->s6_addr32[1],
1922                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1923                    dest->s6_addr32[0], dest->s6_addr32[1],
1924                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1925                    sp->sk_state,
1926                    tp->write_seq-tp->snd_una,
1927                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1928                    timer_active,
1929                    jiffies_to_clock_t(timer_expires - jiffies),
1930                    icsk->icsk_retransmits,
1931                    sock_i_uid(sp),
1932                    icsk->icsk_probes_out,
1933                    sock_i_ino(sp),
1934                    atomic_read(&sp->sk_refcnt), sp,
1935                    jiffies_to_clock_t(icsk->icsk_rto),
1936                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1937                    (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1938                    tp->snd_cwnd, tp->snd_ssthresh>=0xFFFF?-1:tp->snd_ssthresh
1939                    );
1940 }
1941
1942 static void get_timewait6_sock(struct seq_file *seq,
1943                                struct inet_timewait_sock *tw, int i)
1944 {
1945         struct in6_addr *dest, *src;
1946         __u16 destp, srcp;
1947         struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
1948         int ttd = tw->tw_ttd - jiffies;
1949
1950         if (ttd < 0)
1951                 ttd = 0;
1952
1953         dest = &tw6->tw_v6_daddr;
1954         src  = &tw6->tw_v6_rcv_saddr;
1955         destp = ntohs(tw->tw_dport);
1956         srcp  = ntohs(tw->tw_sport);
1957
1958         seq_printf(seq,
1959                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1960                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1961                    i,
1962                    src->s6_addr32[0], src->s6_addr32[1],
1963                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1964                    dest->s6_addr32[0], dest->s6_addr32[1],
1965                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1966                    tw->tw_substate, 0, 0,
1967                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
1968                    atomic_read(&tw->tw_refcnt), tw);
1969 }
1970
1971 static int tcp6_seq_show(struct seq_file *seq, void *v)
1972 {
1973         struct tcp_iter_state *st;
1974
1975         if (v == SEQ_START_TOKEN) {
1976                 seq_puts(seq,
1977                          "  sl  "
1978                          "local_address                         "
1979                          "remote_address                        "
1980                          "st tx_queue rx_queue tr tm->when retrnsmt"
1981                          "   uid  timeout inode\n");
1982                 goto out;
1983         }
1984         st = seq->private;
1985
1986         switch (st->state) {
1987         case TCP_SEQ_STATE_LISTENING:
1988         case TCP_SEQ_STATE_ESTABLISHED:
1989                 get_tcp6_sock(seq, v, st->num);
1990                 break;
1991         case TCP_SEQ_STATE_OPENREQ:
1992                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1993                 break;
1994         case TCP_SEQ_STATE_TIME_WAIT:
1995                 get_timewait6_sock(seq, v, st->num);
1996                 break;
1997         }
1998 out:
1999         return 0;
2000 }
2001
2002 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2003         .name           = "tcp6",
2004         .family         = AF_INET6,
2005         .seq_fops       = {
2006                 .owner          = THIS_MODULE,
2007         },
2008         .seq_ops        = {
2009                 .show           = tcp6_seq_show,
2010         },
2011 };
2012
2013 int tcp6_proc_init(struct net *net)
2014 {
2015         return tcp_proc_register(net, &tcp6_seq_afinfo);
2016 }
2017
2018 void tcp6_proc_exit(struct net *net)
2019 {
2020         tcp_proc_unregister(net, &tcp6_seq_afinfo);
2021 }
2022 #endif
2023
2024 struct proto tcpv6_prot = {
2025         .name                   = "TCPv6",
2026         .owner                  = THIS_MODULE,
2027         .close                  = tcp_close,
2028         .connect                = tcp_v6_connect,
2029         .disconnect             = tcp_disconnect,
2030         .accept                 = inet_csk_accept,
2031         .ioctl                  = tcp_ioctl,
2032         .init                   = tcp_v6_init_sock,
2033         .destroy                = tcp_v6_destroy_sock,
2034         .shutdown               = tcp_shutdown,
2035         .setsockopt             = tcp_setsockopt,
2036         .getsockopt             = tcp_getsockopt,
2037         .recvmsg                = tcp_recvmsg,
2038         .backlog_rcv            = tcp_v6_do_rcv,
2039         .hash                   = tcp_v6_hash,
2040         .unhash                 = inet_unhash,
2041         .get_port               = inet_csk_get_port,
2042         .enter_memory_pressure  = tcp_enter_memory_pressure,
2043         .sockets_allocated      = &tcp_sockets_allocated,
2044         .memory_allocated       = &tcp_memory_allocated,
2045         .memory_pressure        = &tcp_memory_pressure,
2046         .orphan_count           = &tcp_orphan_count,
2047         .sysctl_mem             = sysctl_tcp_mem,
2048         .sysctl_wmem            = sysctl_tcp_wmem,
2049         .sysctl_rmem            = sysctl_tcp_rmem,
2050         .max_header             = MAX_TCP_HEADER,
2051         .obj_size               = sizeof(struct tcp6_sock),
2052         .twsk_prot              = &tcp6_timewait_sock_ops,
2053         .rsk_prot               = &tcp6_request_sock_ops,
2054         .h.hashinfo             = &tcp_hashinfo,
2055 #ifdef CONFIG_COMPAT
2056         .compat_setsockopt      = compat_tcp_setsockopt,
2057         .compat_getsockopt      = compat_tcp_getsockopt,
2058 #endif
2059 };
2060
2061 static struct inet6_protocol tcpv6_protocol = {
2062         .handler        =       tcp_v6_rcv,
2063         .err_handler    =       tcp_v6_err,
2064         .gso_send_check =       tcp_v6_gso_send_check,
2065         .gso_segment    =       tcp_tso_segment,
2066         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2067 };
2068
2069 static struct inet_protosw tcpv6_protosw = {
2070         .type           =       SOCK_STREAM,
2071         .protocol       =       IPPROTO_TCP,
2072         .prot           =       &tcpv6_prot,
2073         .ops            =       &inet6_stream_ops,
2074         .capability     =       -1,
2075         .no_check       =       0,
2076         .flags          =       INET_PROTOSW_PERMANENT |
2077                                 INET_PROTOSW_ICSK,
2078 };
2079
2080 static int tcpv6_net_init(struct net *net)
2081 {
2082         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2083                                     SOCK_RAW, IPPROTO_TCP, net);
2084 }
2085
2086 static void tcpv6_net_exit(struct net *net)
2087 {
2088         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2089         inet_twsk_purge(net, &tcp_hashinfo, &tcp_death_row, AF_INET6);
2090 }
2091
2092 static struct pernet_operations tcpv6_net_ops = {
2093         .init = tcpv6_net_init,
2094         .exit = tcpv6_net_exit,
2095 };
2096
2097 int __init tcpv6_init(void)
2098 {
2099         int ret;
2100
2101         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2102         if (ret)
2103                 goto out;
2104
2105         /* register inet6 protocol */
2106         ret = inet6_register_protosw(&tcpv6_protosw);
2107         if (ret)
2108                 goto out_tcpv6_protocol;
2109
2110         ret = register_pernet_subsys(&tcpv6_net_ops);
2111         if (ret)
2112                 goto out_tcpv6_protosw;
2113 out:
2114         return ret;
2115
2116 out_tcpv6_protocol:
2117         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2118 out_tcpv6_protosw:
2119         inet6_unregister_protosw(&tcpv6_protosw);
2120         goto out;
2121 }
2122
2123 void tcpv6_exit(void)
2124 {
2125         unregister_pernet_subsys(&tcpv6_net_ops);
2126         inet6_unregister_protosw(&tcpv6_protosw);
2127         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2128 }