net/ipv6/tcp_ipv6.c

   1 /*
   2  *      TCP over IPv6
   3  *      Linux INET6 implementation
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <[email protected]>
   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/bottom_half.h>
  27 #include <linux/module.h>
  28 #include <linux/errno.h>
  29 #include <linux/types.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/jiffies.h>
  34 #include <linux/in.h>
  35 #include <linux/in6.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/init.h>
  38 #include <linux/jhash.h>
  39 #include <linux/ipsec.h>
  40 #include <linux/times.h>
  41 #include <linux/slab.h>
  42 #include <linux/uaccess.h>
  43 #include <linux/ipv6.h>
  44 #include <linux/icmpv6.h>
  45 #include <linux/random.h>
  46
  47 #include <net/tcp.h>
  48 #include <net/ndisc.h>
  49 #include <net/inet6_hashtables.h>
  50 #include <net/inet6_connection_sock.h>
  51 #include <net/ipv6.h>
  52 #include <net/transp_v6.h>
  53 #include <net/addrconf.h>
  54 #include <net/ip6_route.h>
  55 #include <net/ip6_checksum.h>
  56 #include <net/inet_ecn.h>
  57 #include <net/protocol.h>
  58 #include <net/xfrm.h>
  59 #include <net/snmp.h>
  60 #include <net/dsfield.h>
  61 #include <net/timewait_sock.h>
  62 #include <net/netdma.h>
  63 #include <net/inet_common.h>
  64 #include <net/secure_seq.h>
  65 #include <net/tcp_memcontrol.h>
  66 #include <net/busy_poll.h>
  67
  68 #include <linux/proc_fs.h>
  69 #include <linux/seq_file.h>
  70
  71 #include <linux/crypto.h>
  72 #include <linux/scatterlist.h>
  73
  74 static void     tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
  75 static void     tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
  76                                       struct request_sock *req);
  77
  78 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  79
  80 static const struct inet_connection_sock_af_ops ipv6_mapped;
  81 static const struct inet_connection_sock_af_ops ipv6_specific;
  82 #ifdef CONFIG_TCP_MD5SIG
  83 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
  84 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
  85 #else
  86 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
  87                                                    const struct in6_addr *addr)
  88 {
  89         return NULL;
  90 }
  91 #endif
  92
  93 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
  94 {
  95         struct dst_entry *dst = skb_dst(skb);
  96         const struct rt6_info *rt = (const struct rt6_info *)dst;
  97
  98         dst_hold(dst);
  99         sk->sk_rx_dst = dst;
 100         inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
 101         if (rt->rt6i_node)
 102                 inet6_sk(sk)->rx_dst_cookie = rt->rt6i_node->fn_sernum;
 103 }
 104
 105 static void tcp_v6_hash(struct sock *sk)
 106 {
 107         if (sk->sk_state != TCP_CLOSE) {
 108                 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
 109                         tcp_prot.hash(sk);
 110                         return;
 111                 }
 112                 local_bh_disable();
 113                 __inet6_hash(sk, NULL);
 114                 local_bh_enable();
 115         }
 116 }
 117
 118 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
 119 {
 120         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
 121                                             ipv6_hdr(skb)->saddr.s6_addr32,
 122                                             tcp_hdr(skb)->dest,
 123                                             tcp_hdr(skb)->source);
 124 }
 125
 126 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 127                           int addr_len)
 128 {
 129         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 130         struct inet_sock *inet = inet_sk(sk);
 131         struct inet_connection_sock *icsk = inet_csk(sk);
 132         struct ipv6_pinfo *np = inet6_sk(sk);
 133         struct tcp_sock *tp = tcp_sk(sk);
 134         struct in6_addr *saddr = NULL, *final_p, final;
 135         struct rt6_info *rt;
 136         struct flowi6 fl6;
 137         struct dst_entry *dst;
 138         int addr_type;
 139         int err;
 140
 141         if (addr_len < SIN6_LEN_RFC2133)
 142                 return -EINVAL;
 143
 144         if (usin->sin6_family != AF_INET6)
 145                 return -EAFNOSUPPORT;
 146
 147         memset(&fl6, 0, sizeof(fl6));
 148
 149         if (np->sndflow) {
 150                 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 151                 IP6_ECN_flow_init(fl6.flowlabel);
 152                 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
 153                         struct ip6_flowlabel *flowlabel;
 154                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 155                         if (flowlabel == NULL)
 156                                 return -EINVAL;
 157                         fl6_sock_release(flowlabel);
 158                 }
 159         }
 160
 161         /*
 162          *      connect() to INADDR_ANY means loopback (BSD'ism).
 163          */
 164
 165         if (ipv6_addr_any(&usin->sin6_addr))
 166                 usin->sin6_addr.s6_addr[15] = 0x1;
 167
 168         addr_type = ipv6_addr_type(&usin->sin6_addr);
 169
 170         if (addr_type & IPV6_ADDR_MULTICAST)
 171                 return -ENETUNREACH;
 172
 173         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 174                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 175                     usin->sin6_scope_id) {
 176                         /* If interface is set while binding, indices
 177                          * must coincide.
 178                          */
 179                         if (sk->sk_bound_dev_if &&
 180                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 181                                 return -EINVAL;
 182
 183                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 184                 }
 185
 186                 /* Connect to link-local address requires an interface */
 187                 if (!sk->sk_bound_dev_if)
 188                         return -EINVAL;
 189         }
 190
 191         if (tp->rx_opt.ts_recent_stamp &&
 192             !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
 193                 tp->rx_opt.ts_recent = 0;
 194                 tp->rx_opt.ts_recent_stamp = 0;
 195                 tp->write_seq = 0;
 196         }
 197
 198         sk->sk_v6_daddr = usin->sin6_addr;
 199         np->flow_label = fl6.flowlabel;
 200
 201         /*
 202          *      TCP over IPv4
 203          */
 204
 205         if (addr_type == IPV6_ADDR_MAPPED) {
 206                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 207                 struct sockaddr_in sin;
 208
 209                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 210
 211                 if (__ipv6_only_sock(sk))
 212                         return -ENETUNREACH;
 213
 214                 sin.sin_family = AF_INET;
 215                 sin.sin_port = usin->sin6_port;
 216                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 217
 218                 icsk->icsk_af_ops = &ipv6_mapped;
 219                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 220 #ifdef CONFIG_TCP_MD5SIG
 221                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
 222 #endif
 223
 224                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 225
 226                 if (err) {
 227                         icsk->icsk_ext_hdr_len = exthdrlen;
 228                         icsk->icsk_af_ops = &ipv6_specific;
 229                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 230 #ifdef CONFIG_TCP_MD5SIG
 231                         tp->af_specific = &tcp_sock_ipv6_specific;
 232 #endif
 233                         goto failure;
 234                 } else {
 235                         ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
 236                         ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
 237                                                &sk->sk_v6_rcv_saddr);
 238                 }
 239
 240                 return err;
 241         }
 242
 243         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 244                 saddr = &sk->sk_v6_rcv_saddr;
 245
 246         fl6.flowi6_proto = IPPROTO_TCP;
 247         fl6.daddr = sk->sk_v6_daddr;
 248         fl6.saddr = saddr ? *saddr : np->saddr;
 249         fl6.flowi6_oif = sk->sk_bound_dev_if;
 250         fl6.flowi6_mark = sk->sk_mark;
 251         fl6.fl6_dport = usin->sin6_port;
 252         fl6.fl6_sport = inet->inet_sport;
 253
 254         final_p = fl6_update_dst(&fl6, np->opt, &final);
 255
 256         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 257
 258         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 259         if (IS_ERR(dst)) {
 260                 err = PTR_ERR(dst);
 261                 goto failure;
 262         }
 263
 264         if (saddr == NULL) {
 265                 saddr = &fl6.saddr;
 266                 sk->sk_v6_rcv_saddr = *saddr;
 267         }
 268
 269         /* set the source address */
 270         np->saddr = *saddr;
 271         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 272
 273         sk->sk_gso_type = SKB_GSO_TCPV6;
 274         __ip6_dst_store(sk, dst, NULL, NULL);
 275
 276         rt = (struct rt6_info *) dst;
 277         if (tcp_death_row.sysctl_tw_recycle &&
 278             !tp->rx_opt.ts_recent_stamp &&
 279             ipv6_addr_equal(&rt->rt6i_dst.addr, &sk->sk_v6_daddr))
 280                 tcp_fetch_timewait_stamp(sk, dst);
 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->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 && likely(!tp->repair))
 297                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 298                                                              sk->sk_v6_daddr.s6_addr32,
 299                                                              inet->inet_sport,
 300                                                              inet->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->inet_dport = 0;
 313         sk->sk_route_caps = 0;
 314         return err;
 315 }
 316
 317 static void tcp_v6_mtu_reduced(struct sock *sk)
 318 {
 319         struct dst_entry *dst;
 320
 321         if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 322                 return;
 323
 324         dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
 325         if (!dst)
 326                 return;
 327
 328         if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
 329                 tcp_sync_mss(sk, dst_mtu(dst));
 330                 tcp_simple_retransmit(sk);
 331         }
 332 }
 333
 334 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 335                 u8 type, u8 code, int offset, __be32 info)
 336 {
 337         const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
 338         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 339         struct ipv6_pinfo *np;
 340         struct sock *sk;
 341         int err;
 342         struct tcp_sock *tp;
 343         struct request_sock *fastopen;
 344         __u32 seq, snd_una;
 345         struct net *net = dev_net(skb->dev);
 346
 347         sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
 348                         th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
 349
 350         if (sk == NULL) {
 351                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 352                                    ICMP6_MIB_INERRORS);
 353                 return;
 354         }
 355
 356         if (sk->sk_state == TCP_TIME_WAIT) {
 357                 inet_twsk_put(inet_twsk(sk));
 358                 return;
 359         }
 360
 361         bh_lock_sock(sk);
 362         if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
 363                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 364
 365         if (sk->sk_state == TCP_CLOSE)
 366                 goto out;
 367
 368         if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
 369                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
 370                 goto out;
 371         }
 372
 373         tp = tcp_sk(sk);
 374         seq = ntohl(th->seq);
 375         /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 376         fastopen = tp->fastopen_rsk;
 377         snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 378         if (sk->sk_state != TCP_LISTEN &&
 379             !between(seq, snd_una, tp->snd_nxt)) {
 380                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 381                 goto out;
 382         }
 383
 384         np = inet6_sk(sk);
 385
 386         if (type == NDISC_REDIRECT) {
 387                 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 388
 389                 if (dst)
 390                         dst->ops->redirect(dst, sk, skb);
 391                 goto out;
 392         }
 393
 394         if (type == ICMPV6_PKT_TOOBIG) {
 395                 /* We are not interested in TCP_LISTEN and open_requests
 396                  * (SYN-ACKs send out by Linux are always <576bytes so
 397                  * they should go through unfragmented).
 398                  */
 399                 if (sk->sk_state == TCP_LISTEN)
 400                         goto out;
 401
 402                 if (!ip6_sk_accept_pmtu(sk))
 403                         goto out;
 404
 405                 tp->mtu_info = ntohl(info);
 406                 if (!sock_owned_by_user(sk))
 407                         tcp_v6_mtu_reduced(sk);
 408                 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
 409                                            &tp->tsq_flags))
 410                         sock_hold(sk);
 411                 goto out;
 412         }
 413
 414         icmpv6_err_convert(type, code, &err);
 415
 416         /* Might be for an request_sock */
 417         switch (sk->sk_state) {
 418                 struct request_sock *req, **prev;
 419         case TCP_LISTEN:
 420                 if (sock_owned_by_user(sk))
 421                         goto out;
 422
 423                 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
 424                                            &hdr->saddr, inet6_iif(skb));
 425                 if (!req)
 426                         goto out;
 427
 428                 /* ICMPs are not backlogged, hence we cannot get
 429                  * an established socket here.
 430                  */
 431                 WARN_ON(req->sk != NULL);
 432
 433                 if (seq != tcp_rsk(req)->snt_isn) {
 434                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 435                         goto out;
 436                 }
 437
 438                 inet_csk_reqsk_queue_drop(sk, req, prev);
 439                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 440                 goto out;
 441
 442         case TCP_SYN_SENT:
 443         case TCP_SYN_RECV:
 444                 /* Only in fast or simultaneous open. If a fast open socket is
 445                  * is already accepted it is treated as a connected one below.
 446                  */
 447                 if (fastopen && fastopen->sk == NULL)
 448                         break;
 449
 450                 if (!sock_owned_by_user(sk)) {
 451                         sk->sk_err = err;
 452                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 453
 454                         tcp_done(sk);
 455                 } else
 456                         sk->sk_err_soft = err;
 457                 goto out;
 458         }
 459
 460         if (!sock_owned_by_user(sk) && np->recverr) {
 461                 sk->sk_err = err;
 462                 sk->sk_error_report(sk);
 463         } else
 464                 sk->sk_err_soft = err;
 465
 466 out:
 467         bh_unlock_sock(sk);
 468         sock_put(sk);
 469 }
 470
 471
 472 static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
 473                               struct flowi6 *fl6,
 474                               struct request_sock *req,
 475                               u16 queue_mapping,
 476                               struct tcp_fastopen_cookie *foc)
 477 {
 478         struct inet_request_sock *ireq = inet_rsk(req);
 479         struct ipv6_pinfo *np = inet6_sk(sk);
 480         struct sk_buff *skb;
 481         int err = -ENOMEM;
 482
 483         /* First, grab a route. */
 484         if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
 485                 goto done;
 486
 487         skb = tcp_make_synack(sk, dst, req, foc);
 488
 489         if (skb) {
 490                 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
 491                                     &ireq->ir_v6_rmt_addr);
 492
 493                 fl6->daddr = ireq->ir_v6_rmt_addr;
 494                 if (np->repflow && (ireq->pktopts != NULL))
 495                         fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
 496
 497                 skb_set_queue_mapping(skb, queue_mapping);
 498                 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
 499                 err = net_xmit_eval(err);
 500         }
 501
 502 done:
 503         return err;
 504 }
 505
 506 static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req)
 507 {
 508         struct flowi6 fl6;
 509         int res;
 510
 511         res = tcp_v6_send_synack(sk, NULL, &fl6, req, 0, NULL);
 512         if (!res) {
 513                 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
 514                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
 515         }
 516         return res;
 517 }
 518
 519 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 520 {
 521         kfree_skb(inet_rsk(req)->pktopts);
 522 }
 523
 524 #ifdef CONFIG_TCP_MD5SIG
 525 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
 526                                                    const struct in6_addr *addr)
 527 {
 528         return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
 529 }
 530
 531 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
 532                                                 struct sock *addr_sk)
 533 {
 534         return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
 535 }
 536
 537 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
 538                                                       struct request_sock *req)
 539 {
 540         return tcp_v6_md5_do_lookup(sk, &inet_rsk(req)->ir_v6_rmt_addr);
 541 }
 542
 543 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
 544                                  int optlen)
 545 {
 546         struct tcp_md5sig cmd;
 547         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 548
 549         if (optlen < sizeof(cmd))
 550                 return -EINVAL;
 551
 552         if (copy_from_user(&cmd, optval, sizeof(cmd)))
 553                 return -EFAULT;
 554
 555         if (sin6->sin6_family != AF_INET6)
 556                 return -EINVAL;
 557
 558         if (!cmd.tcpm_keylen) {
 559                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 560                         return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 561                                               AF_INET);
 562                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 563                                       AF_INET6);
 564         }
 565
 566         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 567                 return -EINVAL;
 568
 569         if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 570                 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 571                                       AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 572
 573         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 574                               AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 575 }
 576
 577 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
 578                                         const struct in6_addr *daddr,
 579                                         const struct in6_addr *saddr, int nbytes)
 580 {
 581         struct tcp6_pseudohdr *bp;
 582         struct scatterlist sg;
 583
 584         bp = &hp->md5_blk.ip6;
 585         /* 1. TCP pseudo-header (RFC2460) */
 586         bp->saddr = *saddr;
 587         bp->daddr = *daddr;
 588         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 589         bp->len = cpu_to_be32(nbytes);
 590
 591         sg_init_one(&sg, bp, sizeof(*bp));
 592         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
 593 }
 594
 595 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
 596                                const struct in6_addr *daddr, struct in6_addr *saddr,
 597                                const struct tcphdr *th)
 598 {
 599         struct tcp_md5sig_pool *hp;
 600         struct hash_desc *desc;
 601
 602         hp = tcp_get_md5sig_pool();
 603         if (!hp)
 604                 goto clear_hash_noput;
 605         desc = &hp->md5_desc;
 606
 607         if (crypto_hash_init(desc))
 608                 goto clear_hash;
 609         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
 610                 goto clear_hash;
 611         if (tcp_md5_hash_header(hp, th))
 612                 goto clear_hash;
 613         if (tcp_md5_hash_key(hp, key))
 614                 goto clear_hash;
 615         if (crypto_hash_final(desc, md5_hash))
 616                 goto clear_hash;
 617
 618         tcp_put_md5sig_pool();
 619         return 0;
 620
 621 clear_hash:
 622         tcp_put_md5sig_pool();
 623 clear_hash_noput:
 624         memset(md5_hash, 0, 16);
 625         return 1;
 626 }
 627
 628 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
 629                                const struct sock *sk,
 630                                const struct request_sock *req,
 631                                const struct sk_buff *skb)
 632 {
 633         const struct in6_addr *saddr, *daddr;
 634         struct tcp_md5sig_pool *hp;
 635         struct hash_desc *desc;
 636         const struct tcphdr *th = tcp_hdr(skb);
 637
 638         if (sk) {
 639                 saddr = &inet6_sk(sk)->saddr;
 640                 daddr = &sk->sk_v6_daddr;
 641         } else if (req) {
 642                 saddr = &inet_rsk(req)->ir_v6_loc_addr;
 643                 daddr = &inet_rsk(req)->ir_v6_rmt_addr;
 644         } else {
 645                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 646                 saddr = &ip6h->saddr;
 647                 daddr = &ip6h->daddr;
 648         }
 649
 650         hp = tcp_get_md5sig_pool();
 651         if (!hp)
 652                 goto clear_hash_noput;
 653         desc = &hp->md5_desc;
 654
 655         if (crypto_hash_init(desc))
 656                 goto clear_hash;
 657
 658         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
 659                 goto clear_hash;
 660         if (tcp_md5_hash_header(hp, th))
 661                 goto clear_hash;
 662         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 663                 goto clear_hash;
 664         if (tcp_md5_hash_key(hp, key))
 665                 goto clear_hash;
 666         if (crypto_hash_final(desc, md5_hash))
 667                 goto clear_hash;
 668
 669         tcp_put_md5sig_pool();
 670         return 0;
 671
 672 clear_hash:
 673         tcp_put_md5sig_pool();
 674 clear_hash_noput:
 675         memset(md5_hash, 0, 16);
 676         return 1;
 677 }
 678
 679 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
 680 {
 681         const __u8 *hash_location = NULL;
 682         struct tcp_md5sig_key *hash_expected;
 683         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 684         const struct tcphdr *th = tcp_hdr(skb);
 685         int genhash;
 686         u8 newhash[16];
 687
 688         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 689         hash_location = tcp_parse_md5sig_option(th);
 690
 691         /* We've parsed the options - do we have a hash? */
 692         if (!hash_expected && !hash_location)
 693                 return 0;
 694
 695         if (hash_expected && !hash_location) {
 696                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 697                 return 1;
 698         }
 699
 700         if (!hash_expected && hash_location) {
 701                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 702                 return 1;
 703         }
 704
 705         /* check the signature */
 706         genhash = tcp_v6_md5_hash_skb(newhash,
 707                                       hash_expected,
 708                                       NULL, NULL, skb);
 709
 710         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 711                 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
 712                                      genhash ? "failed" : "mismatch",
 713                                      &ip6h->saddr, ntohs(th->source),
 714                                      &ip6h->daddr, ntohs(th->dest));
 715                 return 1;
 716         }
 717         return 0;
 718 }
 719 #endif
 720
 721 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 722         .family         =       AF_INET6,
 723         .obj_size       =       sizeof(struct tcp6_request_sock),
 724         .rtx_syn_ack    =       tcp_v6_rtx_synack,
 725         .send_ack       =       tcp_v6_reqsk_send_ack,
 726         .destructor     =       tcp_v6_reqsk_destructor,
 727         .send_reset     =       tcp_v6_send_reset,
 728         .syn_ack_timeout =      tcp_syn_ack_timeout,
 729 };
 730
 731 #ifdef CONFIG_TCP_MD5SIG
 732 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 733         .md5_lookup     =       tcp_v6_reqsk_md5_lookup,
 734         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 735 };
 736 #endif
 737
 738 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
 739                                  u32 tsval, u32 tsecr, int oif,
 740                                  struct tcp_md5sig_key *key, int rst, u8 tclass,
 741                                  u32 label)
 742 {
 743         const struct tcphdr *th = tcp_hdr(skb);
 744         struct tcphdr *t1;
 745         struct sk_buff *buff;
 746         struct flowi6 fl6;
 747         struct net *net = dev_net(skb_dst(skb)->dev);
 748         struct sock *ctl_sk = net->ipv6.tcp_sk;
 749         unsigned int tot_len = sizeof(struct tcphdr);
 750         struct dst_entry *dst;
 751         __be32 *topt;
 752
 753         if (tsecr)
 754                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 755 #ifdef CONFIG_TCP_MD5SIG
 756         if (key)
 757                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 758 #endif
 759
 760         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 761                          GFP_ATOMIC);
 762         if (buff == NULL)
 763                 return;
 764
 765         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 766
 767         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 768         skb_reset_transport_header(buff);
 769
 770         /* Swap the send and the receive. */
 771         memset(t1, 0, sizeof(*t1));
 772         t1->dest = th->source;
 773         t1->source = th->dest;
 774         t1->doff = tot_len / 4;
 775         t1->seq = htonl(seq);
 776         t1->ack_seq = htonl(ack);
 777         t1->ack = !rst || !th->ack;
 778         t1->rst = rst;
 779         t1->window = htons(win);
 780
 781         topt = (__be32 *)(t1 + 1);
 782
 783         if (tsecr) {
 784                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 785                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 786                 *topt++ = htonl(tsval);
 787                 *topt++ = htonl(tsecr);
 788         }
 789
 790 #ifdef CONFIG_TCP_MD5SIG
 791         if (key) {
 792                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 793                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 794                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 795                                     &ipv6_hdr(skb)->saddr,
 796                                     &ipv6_hdr(skb)->daddr, t1);
 797         }
 798 #endif
 799
 800         memset(&fl6, 0, sizeof(fl6));
 801         fl6.daddr = ipv6_hdr(skb)->saddr;
 802         fl6.saddr = ipv6_hdr(skb)->daddr;
 803         fl6.flowlabel = label;
 804
 805         buff->ip_summed = CHECKSUM_PARTIAL;
 806         buff->csum = 0;
 807
 808         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
 809
 810         fl6.flowi6_proto = IPPROTO_TCP;
 811         if (rt6_need_strict(&fl6.daddr) && !oif)
 812                 fl6.flowi6_oif = inet6_iif(skb);
 813         else
 814                 fl6.flowi6_oif = oif;
 815         fl6.fl6_dport = t1->dest;
 816         fl6.fl6_sport = t1->source;
 817         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 818
 819         /* Pass a socket to ip6_dst_lookup either it is for RST
 820          * Underlying function will use this to retrieve the network
 821          * namespace
 822          */
 823         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 824         if (!IS_ERR(dst)) {
 825                 skb_dst_set(buff, dst);
 826                 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
 827                 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 828                 if (rst)
 829                         TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
 830                 return;
 831         }
 832
 833         kfree_skb(buff);
 834 }
 835
 836 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
 837 {
 838         const struct tcphdr *th = tcp_hdr(skb);
 839         u32 seq = 0, ack_seq = 0;
 840         struct tcp_md5sig_key *key = NULL;
 841 #ifdef CONFIG_TCP_MD5SIG
 842         const __u8 *hash_location = NULL;
 843         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 844         unsigned char newhash[16];
 845         int genhash;
 846         struct sock *sk1 = NULL;
 847 #endif
 848         int oif;
 849
 850         if (th->rst)
 851                 return;
 852
 853         if (!ipv6_unicast_destination(skb))
 854                 return;
 855
 856 #ifdef CONFIG_TCP_MD5SIG
 857         hash_location = tcp_parse_md5sig_option(th);
 858         if (!sk && hash_location) {
 859                 /*
 860                  * active side is lost. Try to find listening socket through
 861                  * source port, and then find md5 key through listening socket.
 862                  * we are not loose security here:
 863                  * Incoming packet is checked with md5 hash with finding key,
 864                  * no RST generated if md5 hash doesn't match.
 865                  */
 866                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
 867                                            &tcp_hashinfo, &ipv6h->saddr,
 868                                            th->source, &ipv6h->daddr,
 869                                            ntohs(th->source), inet6_iif(skb));
 870                 if (!sk1)
 871                         return;
 872
 873                 rcu_read_lock();
 874                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 875                 if (!key)
 876                         goto release_sk1;
 877
 878                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
 879                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
 880                         goto release_sk1;
 881         } else {
 882                 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
 883         }
 884 #endif
 885
 886         if (th->ack)
 887                 seq = ntohl(th->ack_seq);
 888         else
 889                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
 890                           (th->doff << 2);
 891
 892         oif = sk ? sk->sk_bound_dev_if : 0;
 893         tcp_v6_send_response(skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
 894
 895 #ifdef CONFIG_TCP_MD5SIG
 896 release_sk1:
 897         if (sk1) {
 898                 rcu_read_unlock();
 899                 sock_put(sk1);
 900         }
 901 #endif
 902 }
 903
 904 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
 905                             u32 win, u32 tsval, u32 tsecr, int oif,
 906                             struct tcp_md5sig_key *key, u8 tclass,
 907                             u32 label)
 908 {
 909         tcp_v6_send_response(skb, seq, ack, win, tsval, tsecr, oif, key, 0, tclass,
 910                              label);
 911 }
 912
 913 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 914 {
 915         struct inet_timewait_sock *tw = inet_twsk(sk);
 916         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 917
 918         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 919                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 920                         tcp_time_stamp + tcptw->tw_ts_offset,
 921                         tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
 922                         tw->tw_tclass, (tw->tw_flowlabel << 12));
 923
 924         inet_twsk_put(tw);
 925 }
 926
 927 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
 928                                   struct request_sock *req)
 929 {
 930         /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 931          * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 932          */
 933         tcp_v6_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
 934                         tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
 935                         tcp_rsk(req)->rcv_nxt,
 936                         req->rcv_wnd, tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
 937                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
 938                         0, 0);
 939 }
 940
 941
 942 static struct sock *tcp_v6_hnd_req(struct sock *sk, struct sk_buff *skb)
 943 {
 944         struct request_sock *req, **prev;
 945         const struct tcphdr *th = tcp_hdr(skb);
 946         struct sock *nsk;
 947
 948         /* Find possible connection requests. */
 949         req = inet6_csk_search_req(sk, &prev, th->source,
 950                                    &ipv6_hdr(skb)->saddr,
 951                                    &ipv6_hdr(skb)->daddr, inet6_iif(skb));
 952         if (req)
 953                 return tcp_check_req(sk, skb, req, prev, false);
 954
 955         nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
 956                         &ipv6_hdr(skb)->saddr, th->source,
 957                         &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
 958
 959         if (nsk) {
 960                 if (nsk->sk_state != TCP_TIME_WAIT) {
 961                         bh_lock_sock(nsk);
 962                         return nsk;
 963                 }
 964                 inet_twsk_put(inet_twsk(nsk));
 965                 return NULL;
 966         }
 967
 968 #ifdef CONFIG_SYN_COOKIES
 969         if (!th->syn)
 970                 sk = cookie_v6_check(sk, skb);
 971 #endif
 972         return sk;
 973 }
 974
 975 /* FIXME: this is substantially similar to the ipv4 code.
 976  * Can some kind of merge be done? -- erics
 977  */
 978 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 979 {
 980         struct tcp_options_received tmp_opt;
 981         struct request_sock *req;
 982         struct inet_request_sock *ireq;
 983         struct ipv6_pinfo *np = inet6_sk(sk);
 984         struct tcp_sock *tp = tcp_sk(sk);
 985         __u32 isn = TCP_SKB_CB(skb)->when;
 986         struct dst_entry *dst = NULL;
 987         struct tcp_fastopen_cookie foc = { .len = -1 };
 988         bool want_cookie = false, fastopen;
 989         struct flowi6 fl6;
 990         int err;
 991
 992         if (skb->protocol == htons(ETH_P_IP))
 993                 return tcp_v4_conn_request(sk, skb);
 994
 995         if (!ipv6_unicast_destination(skb))
 996                 goto drop;
 997
 998         if ((sysctl_tcp_syncookies == 2 ||
 999              inet_csk_reqsk_queue_is_full(sk)) && !isn) {
1000                 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
1001                 if (!want_cookie)
1002                         goto drop;
1003         }
1004
1005         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
1006                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1007                 goto drop;
1008         }
1009
1010         req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1011         if (req == NULL)
1012                 goto drop;
1013
1014 #ifdef CONFIG_TCP_MD5SIG
1015         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1016 #endif
1017
1018         tcp_clear_options(&tmp_opt);
1019         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1020         tmp_opt.user_mss = tp->rx_opt.user_mss;
1021         tcp_parse_options(skb, &tmp_opt, 0, want_cookie ? NULL : &foc);
1022
1023         if (want_cookie && !tmp_opt.saw_tstamp)
1024                 tcp_clear_options(&tmp_opt);
1025
1026         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1027         tcp_openreq_init(req, &tmp_opt, skb);
1028
1029         ireq = inet_rsk(req);
1030         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
1031         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
1032         if (!want_cookie || tmp_opt.tstamp_ok)
1033                 TCP_ECN_create_request(req, skb, sock_net(sk));
1034
1035         ireq->ir_iif = sk->sk_bound_dev_if;
1036
1037         /* So that link locals have meaning */
1038         if (!sk->sk_bound_dev_if &&
1039             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
1040                 ireq->ir_iif = inet6_iif(skb);
1041
1042         if (!isn) {
1043                 if (ipv6_opt_accepted(sk, skb) ||
1044                     np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1045                     np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim ||
1046                     np->repflow) {
1047                         atomic_inc(&skb->users);
1048                         ireq->pktopts = skb;
1049                 }
1050
1051                 if (want_cookie) {
1052                         isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1053                         req->cookie_ts = tmp_opt.tstamp_ok;
1054                         goto have_isn;
1055                 }
1056
1057                 /* VJ's idea. We save last timestamp seen
1058                  * from the destination in peer table, when entering
1059                  * state TIME-WAIT, and check against it before
1060                  * accepting new connection request.
1061                  *
1062                  * If "isn" is not zero, this request hit alive
1063                  * timewait bucket, so that all the necessary checks
1064                  * are made in the function processing timewait state.
1065                  */
1066                 if (tmp_opt.saw_tstamp &&
1067                     tcp_death_row.sysctl_tw_recycle &&
1068                     (dst = inet6_csk_route_req(sk, &fl6, req)) != NULL) {
1069                         if (!tcp_peer_is_proven(req, dst, true)) {
1070                                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1071                                 goto drop_and_release;
1072                         }
1073                 }
1074                 /* Kill the following clause, if you dislike this way. */
1075                 else if (!sysctl_tcp_syncookies &&
1076                          (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1077                           (sysctl_max_syn_backlog >> 2)) &&
1078                          !tcp_peer_is_proven(req, dst, false)) {
1079                         /* Without syncookies last quarter of
1080                          * backlog is filled with destinations,
1081                          * proven to be alive.
1082                          * It means that we continue to communicate
1083                          * to destinations, already remembered
1084                          * to the moment of synflood.
1085                          */
1086                         LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1087                                        &ireq->ir_v6_rmt_addr, ntohs(tcp_hdr(skb)->source));
1088                         goto drop_and_release;
1089                 }
1090
1091                 isn = tcp_v6_init_sequence(skb);
1092         }
1093 have_isn:
1094
1095         if (security_inet_conn_request(sk, skb, req))
1096                 goto drop_and_release;
1097
1098         if (!dst && (dst = inet6_csk_route_req(sk, &fl6, req)) == NULL)
1099                 goto drop_and_free;
1100
1101         tcp_rsk(req)->snt_isn = isn;
1102         tcp_rsk(req)->snt_synack = tcp_time_stamp;
1103         tcp_openreq_init_rwin(req, sk, dst);
1104         fastopen = !want_cookie &&
1105                    tcp_try_fastopen(sk, skb, req, &foc, dst);
1106         err = tcp_v6_send_synack(sk, dst, &fl6, req,
1107                                  skb_get_queue_mapping(skb), &foc);
1108         if (!fastopen) {
1109                 if (err || want_cookie)
1110                         goto drop_and_free;
1111
1112                 tcp_rsk(req)->listener = NULL;
1113                 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1114         }
1115         return 0;
1116
1117 drop_and_release:
1118         dst_release(dst);
1119 drop_and_free:
1120         reqsk_free(req);
1121 drop:
1122         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1123         return 0; /* don't send reset */
1124 }
1125
1126 static struct sock *tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1127                                          struct request_sock *req,
1128                                          struct dst_entry *dst)
1129 {
1130         struct inet_request_sock *ireq;
1131         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1132         struct tcp6_sock *newtcp6sk;
1133         struct inet_sock *newinet;
1134         struct tcp_sock *newtp;
1135         struct sock *newsk;
1136 #ifdef CONFIG_TCP_MD5SIG
1137         struct tcp_md5sig_key *key;
1138 #endif
1139         struct flowi6 fl6;
1140
1141         if (skb->protocol == htons(ETH_P_IP)) {
1142                 /*
1143                  *      v6 mapped
1144                  */
1145
1146                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1147
1148                 if (newsk == NULL)
1149                         return NULL;
1150
1151                 newtcp6sk = (struct tcp6_sock *)newsk;
1152                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1153
1154                 newinet = inet_sk(newsk);
1155                 newnp = inet6_sk(newsk);
1156                 newtp = tcp_sk(newsk);
1157
1158                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1159
1160                 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newsk->sk_v6_daddr);
1161
1162                 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1163
1164                 newsk->sk_v6_rcv_saddr = newnp->saddr;
1165
1166                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1167                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1168 #ifdef CONFIG_TCP_MD5SIG
1169                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1170 #endif
1171
1172                 newnp->ipv6_ac_list = NULL;
1173                 newnp->ipv6_fl_list = NULL;
1174                 newnp->pktoptions  = NULL;
1175                 newnp->opt         = NULL;
1176                 newnp->mcast_oif   = inet6_iif(skb);
1177                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1178                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1179                 if (np->repflow)
1180                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1181
1182                 /*
1183                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1184                  * here, tcp_create_openreq_child now does this for us, see the comment in
1185                  * that function for the gory details. -acme
1186                  */
1187
1188                 /* It is tricky place. Until this moment IPv4 tcp
1189                    worked with IPv6 icsk.icsk_af_ops.
1190                    Sync it now.
1191                  */
1192                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1193
1194                 return newsk;
1195         }
1196
1197         ireq = inet_rsk(req);
1198
1199         if (sk_acceptq_is_full(sk))
1200                 goto out_overflow;
1201
1202         if (!dst) {
1203                 dst = inet6_csk_route_req(sk, &fl6, req);
1204                 if (!dst)
1205                         goto out;
1206         }
1207
1208         newsk = tcp_create_openreq_child(sk, req, skb);
1209         if (newsk == NULL)
1210                 goto out_nonewsk;
1211
1212         /*
1213          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1214          * count here, tcp_create_openreq_child now does this for us, see the
1215          * comment in that function for the gory details. -acme
1216          */
1217
1218         newsk->sk_gso_type = SKB_GSO_TCPV6;
1219         __ip6_dst_store(newsk, dst, NULL, NULL);
1220         inet6_sk_rx_dst_set(newsk, skb);
1221
1222         newtcp6sk = (struct tcp6_sock *)newsk;
1223         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1224
1225         newtp = tcp_sk(newsk);
1226         newinet = inet_sk(newsk);
1227         newnp = inet6_sk(newsk);
1228
1229         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1230
1231         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1232         newnp->saddr = ireq->ir_v6_loc_addr;
1233         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1234         newsk->sk_bound_dev_if = ireq->ir_iif;
1235
1236         /* Now IPv6 options...
1237
1238            First: no IPv4 options.
1239          */
1240         newinet->inet_opt = NULL;
1241         newnp->ipv6_ac_list = NULL;
1242         newnp->ipv6_fl_list = NULL;
1243
1244         /* Clone RX bits */
1245         newnp->rxopt.all = np->rxopt.all;
1246
1247         /* Clone pktoptions received with SYN */
1248         newnp->pktoptions = NULL;
1249         if (ireq->pktopts != NULL) {
1250                 newnp->pktoptions = skb_clone(ireq->pktopts,
1251                                               sk_gfp_atomic(sk, GFP_ATOMIC));
1252                 consume_skb(ireq->pktopts);
1253                 ireq->pktopts = NULL;
1254                 if (newnp->pktoptions)
1255                         skb_set_owner_r(newnp->pktoptions, newsk);
1256         }
1257         newnp->opt        = NULL;
1258         newnp->mcast_oif  = inet6_iif(skb);
1259         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1260         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1261         if (np->repflow)
1262                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1263
1264         /* Clone native IPv6 options from listening socket (if any)
1265
1266            Yes, keeping reference count would be much more clever,
1267            but we make one more one thing there: reattach optmem
1268            to newsk.
1269          */
1270         if (np->opt)
1271                 newnp->opt = ipv6_dup_options(newsk, np->opt);
1272
1273         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1274         if (newnp->opt)
1275                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1276                                                      newnp->opt->opt_flen);
1277
1278         tcp_sync_mss(newsk, dst_mtu(dst));
1279         newtp->advmss = dst_metric_advmss(dst);
1280         if (tcp_sk(sk)->rx_opt.user_mss &&
1281             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1282                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1283
1284         tcp_initialize_rcv_mss(newsk);
1285
1286         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1287         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1288
1289 #ifdef CONFIG_TCP_MD5SIG
1290         /* Copy over the MD5 key from the original socket */
1291         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1292         if (key != NULL) {
1293                 /* We're using one, so create a matching key
1294                  * on the newsk structure. If we fail to get
1295                  * memory, then we end up not copying the key
1296                  * across. Shucks.
1297                  */
1298                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1299                                AF_INET6, key->key, key->keylen,
1300                                sk_gfp_atomic(sk, GFP_ATOMIC));
1301         }
1302 #endif
1303
1304         if (__inet_inherit_port(sk, newsk) < 0) {
1305                 inet_csk_prepare_forced_close(newsk);
1306                 tcp_done(newsk);
1307                 goto out;
1308         }
1309         __inet6_hash(newsk, NULL);
1310
1311         return newsk;
1312
1313 out_overflow:
1314         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1315 out_nonewsk:
1316         dst_release(dst);
1317 out:
1318         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1319         return NULL;
1320 }
1321
1322 /* The socket must have it's spinlock held when we get
1323  * here.
1324  *
1325  * We have a potential double-lock case here, so even when
1326  * doing backlog processing we use the BH locking scheme.
1327  * This is because we cannot sleep with the original spinlock
1328  * held.
1329  */
1330 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1331 {
1332         struct ipv6_pinfo *np = inet6_sk(sk);
1333         struct tcp_sock *tp;
1334         struct sk_buff *opt_skb = NULL;
1335
1336         /* Imagine: socket is IPv6. IPv4 packet arrives,
1337            goes to IPv4 receive handler and backlogged.
1338            From backlog it always goes here. Kerboom...
1339            Fortunately, tcp_rcv_established and rcv_established
1340            handle them correctly, but it is not case with
1341            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1342          */
1343
1344         if (skb->protocol == htons(ETH_P_IP))
1345                 return tcp_v4_do_rcv(sk, skb);
1346
1347 #ifdef CONFIG_TCP_MD5SIG
1348         if (tcp_v6_inbound_md5_hash(sk, skb))
1349                 goto discard;
1350 #endif
1351
1352         if (sk_filter(sk, skb))
1353                 goto discard;
1354
1355         /*
1356          *      socket locking is here for SMP purposes as backlog rcv
1357          *      is currently called with bh processing disabled.
1358          */
1359
1360         /* Do Stevens' IPV6_PKTOPTIONS.
1361
1362            Yes, guys, it is the only place in our code, where we
1363            may make it not affecting IPv4.
1364            The rest of code is protocol independent,
1365            and I do not like idea to uglify IPv4.
1366
1367            Actually, all the idea behind IPV6_PKTOPTIONS
1368            looks not very well thought. For now we latch
1369            options, received in the last packet, enqueued
1370            by tcp. Feel free to propose better solution.
1371                                                --ANK (980728)
1372          */
1373         if (np->rxopt.all)
1374                 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1375
1376         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1377                 struct dst_entry *dst = sk->sk_rx_dst;
1378
1379                 sock_rps_save_rxhash(sk, skb);
1380                 if (dst) {
1381                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1382                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1383                                 dst_release(dst);
1384                                 sk->sk_rx_dst = NULL;
1385                         }
1386                 }
1387
1388                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1389                 if (opt_skb)
1390                         goto ipv6_pktoptions;
1391                 return 0;
1392         }
1393
1394         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1395                 goto csum_err;
1396
1397         if (sk->sk_state == TCP_LISTEN) {
1398                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1399                 if (!nsk)
1400                         goto discard;
1401
1402                 /*
1403                  * Queue it on the new socket if the new socket is active,
1404                  * otherwise we just shortcircuit this and continue with
1405                  * the new socket..
1406                  */
1407                 if (nsk != sk) {
1408                         sock_rps_save_rxhash(nsk, skb);
1409                         if (tcp_child_process(sk, nsk, skb))
1410                                 goto reset;
1411                         if (opt_skb)
1412                                 __kfree_skb(opt_skb);
1413                         return 0;
1414                 }
1415         } else
1416                 sock_rps_save_rxhash(sk, skb);
1417
1418         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1419                 goto reset;
1420         if (opt_skb)
1421                 goto ipv6_pktoptions;
1422         return 0;
1423
1424 reset:
1425         tcp_v6_send_reset(sk, skb);
1426 discard:
1427         if (opt_skb)
1428                 __kfree_skb(opt_skb);
1429         kfree_skb(skb);
1430         return 0;
1431 csum_err:
1432         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1433         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1434         goto discard;
1435
1436
1437 ipv6_pktoptions:
1438         /* Do you ask, what is it?
1439
1440            1. skb was enqueued by tcp.
1441            2. skb is added to tail of read queue, rather than out of order.
1442            3. socket is not in passive state.
1443            4. Finally, it really contains options, which user wants to receive.
1444          */
1445         tp = tcp_sk(sk);
1446         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1447             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1448                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1449                         np->mcast_oif = inet6_iif(opt_skb);
1450                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1451                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1452                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1453                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1454                 if (np->repflow)
1455                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1456                 if (ipv6_opt_accepted(sk, opt_skb)) {
1457                         skb_set_owner_r(opt_skb, sk);
1458                         opt_skb = xchg(&np->pktoptions, opt_skb);
1459                 } else {
1460                         __kfree_skb(opt_skb);
1461                         opt_skb = xchg(&np->pktoptions, NULL);
1462                 }
1463         }
1464
1465         kfree_skb(opt_skb);
1466         return 0;
1467 }
1468
1469 static int tcp_v6_rcv(struct sk_buff *skb)
1470 {
1471         const struct tcphdr *th;
1472         const struct ipv6hdr *hdr;
1473         struct sock *sk;
1474         int ret;
1475         struct net *net = dev_net(skb->dev);
1476
1477         if (skb->pkt_type != PACKET_HOST)
1478                 goto discard_it;
1479
1480         /*
1481          *      Count it even if it's bad.
1482          */
1483         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1484
1485         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1486                 goto discard_it;
1487
1488         th = tcp_hdr(skb);
1489
1490         if (th->doff < sizeof(struct tcphdr)/4)
1491                 goto bad_packet;
1492         if (!pskb_may_pull(skb, th->doff*4))
1493                 goto discard_it;
1494
1495         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1496                 goto csum_error;
1497
1498         th = tcp_hdr(skb);
1499         hdr = ipv6_hdr(skb);
1500         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1501         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1502                                     skb->len - th->doff*4);
1503         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1504         TCP_SKB_CB(skb)->when = 0;
1505         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1506         TCP_SKB_CB(skb)->sacked = 0;
1507
1508         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1509         if (!sk)
1510                 goto no_tcp_socket;
1511
1512 process:
1513         if (sk->sk_state == TCP_TIME_WAIT)
1514                 goto do_time_wait;
1515
1516         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1517                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1518                 goto discard_and_relse;
1519         }
1520
1521         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1522                 goto discard_and_relse;
1523
1524         if (sk_filter(sk, skb))
1525                 goto discard_and_relse;
1526
1527         sk_mark_napi_id(sk, skb);
1528         skb->dev = NULL;
1529
1530         bh_lock_sock_nested(sk);
1531         ret = 0;
1532         if (!sock_owned_by_user(sk)) {
1533 #ifdef CONFIG_NET_DMA
1534                 struct tcp_sock *tp = tcp_sk(sk);
1535                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1536                         tp->ucopy.dma_chan = net_dma_find_channel();
1537                 if (tp->ucopy.dma_chan)
1538                         ret = tcp_v6_do_rcv(sk, skb);
1539                 else
1540 #endif
1541                 {
1542                         if (!tcp_prequeue(sk, skb))
1543                                 ret = tcp_v6_do_rcv(sk, skb);
1544                 }
1545         } else if (unlikely(sk_add_backlog(sk, skb,
1546                                            sk->sk_rcvbuf + sk->sk_sndbuf))) {
1547                 bh_unlock_sock(sk);
1548                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1549                 goto discard_and_relse;
1550         }
1551         bh_unlock_sock(sk);
1552
1553         sock_put(sk);
1554         return ret ? -1 : 0;
1555
1556 no_tcp_socket:
1557         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1558                 goto discard_it;
1559
1560         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1561 csum_error:
1562                 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1563 bad_packet:
1564                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1565         } else {
1566                 tcp_v6_send_reset(NULL, skb);
1567         }
1568
1569 discard_it:
1570         kfree_skb(skb);
1571         return 0;
1572
1573 discard_and_relse:
1574         sock_put(sk);
1575         goto discard_it;
1576
1577 do_time_wait:
1578         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1579                 inet_twsk_put(inet_twsk(sk));
1580                 goto discard_it;
1581         }
1582
1583         if (skb->len < (th->doff<<2)) {
1584                 inet_twsk_put(inet_twsk(sk));
1585                 goto bad_packet;
1586         }
1587         if (tcp_checksum_complete(skb)) {
1588                 inet_twsk_put(inet_twsk(sk));
1589                 goto csum_error;
1590         }
1591
1592         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1593         case TCP_TW_SYN:
1594         {
1595                 struct sock *sk2;
1596
1597                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1598                                             &ipv6_hdr(skb)->saddr, th->source,
1599                                             &ipv6_hdr(skb)->daddr,
1600                                             ntohs(th->dest), inet6_iif(skb));
1601                 if (sk2 != NULL) {
1602                         struct inet_timewait_sock *tw = inet_twsk(sk);
1603                         inet_twsk_deschedule(tw, &tcp_death_row);
1604                         inet_twsk_put(tw);
1605                         sk = sk2;
1606                         goto process;
1607                 }
1608                 /* Fall through to ACK */
1609         }
1610         case TCP_TW_ACK:
1611                 tcp_v6_timewait_ack(sk, skb);
1612                 break;
1613         case TCP_TW_RST:
1614                 goto no_tcp_socket;
1615         case TCP_TW_SUCCESS:
1616                 ;
1617         }
1618         goto discard_it;
1619 }
1620
1621 static void tcp_v6_early_demux(struct sk_buff *skb)
1622 {
1623         const struct ipv6hdr *hdr;
1624         const struct tcphdr *th;
1625         struct sock *sk;
1626
1627         if (skb->pkt_type != PACKET_HOST)
1628                 return;
1629
1630         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1631                 return;
1632
1633         hdr = ipv6_hdr(skb);
1634         th = tcp_hdr(skb);
1635
1636         if (th->doff < sizeof(struct tcphdr) / 4)
1637                 return;
1638
1639         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1640                                         &hdr->saddr, th->source,
1641                                         &hdr->daddr, ntohs(th->dest),
1642                                         inet6_iif(skb));
1643         if (sk) {
1644                 skb->sk = sk;
1645                 skb->destructor = sock_edemux;
1646                 if (sk->sk_state != TCP_TIME_WAIT) {
1647                         struct dst_entry *dst = sk->sk_rx_dst;
1648
1649                         if (dst)
1650                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1651                         if (dst &&
1652                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1653                                 skb_dst_set_noref(skb, dst);
1654                 }
1655         }
1656 }
1657
1658 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1659         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1660         .twsk_unique    = tcp_twsk_unique,
1661         .twsk_destructor = tcp_twsk_destructor,
1662 };
1663
1664 static const struct inet_connection_sock_af_ops ipv6_specific = {
1665         .queue_xmit        = inet6_csk_xmit,
1666         .send_check        = tcp_v6_send_check,
1667         .rebuild_header    = inet6_sk_rebuild_header,
1668         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1669         .conn_request      = tcp_v6_conn_request,
1670         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1671         .net_header_len    = sizeof(struct ipv6hdr),
1672         .net_frag_header_len = sizeof(struct frag_hdr),
1673         .setsockopt        = ipv6_setsockopt,
1674         .getsockopt        = ipv6_getsockopt,
1675         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1676         .sockaddr_len      = sizeof(struct sockaddr_in6),
1677         .bind_conflict     = inet6_csk_bind_conflict,
1678 #ifdef CONFIG_COMPAT
1679         .compat_setsockopt = compat_ipv6_setsockopt,
1680         .compat_getsockopt = compat_ipv6_getsockopt,
1681 #endif
1682 };
1683
1684 #ifdef CONFIG_TCP_MD5SIG
1685 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1686         .md5_lookup     =       tcp_v6_md5_lookup,
1687         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1688         .md5_parse      =       tcp_v6_parse_md5_keys,
1689 };
1690 #endif
1691
1692 /*
1693  *      TCP over IPv4 via INET6 API
1694  */
1695 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1696         .queue_xmit        = ip_queue_xmit,
1697         .send_check        = tcp_v4_send_check,
1698         .rebuild_header    = inet_sk_rebuild_header,
1699         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1700         .conn_request      = tcp_v6_conn_request,
1701         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1702         .net_header_len    = sizeof(struct iphdr),
1703         .setsockopt        = ipv6_setsockopt,
1704         .getsockopt        = ipv6_getsockopt,
1705         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1706         .sockaddr_len      = sizeof(struct sockaddr_in6),
1707         .bind_conflict     = inet6_csk_bind_conflict,
1708 #ifdef CONFIG_COMPAT
1709         .compat_setsockopt = compat_ipv6_setsockopt,
1710         .compat_getsockopt = compat_ipv6_getsockopt,
1711 #endif
1712 };
1713
1714 #ifdef CONFIG_TCP_MD5SIG
1715 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1716         .md5_lookup     =       tcp_v4_md5_lookup,
1717         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1718         .md5_parse      =       tcp_v6_parse_md5_keys,
1719 };
1720 #endif
1721
1722 /* NOTE: A lot of things set to zero explicitly by call to
1723  *       sk_alloc() so need not be done here.
1724  */
1725 static int tcp_v6_init_sock(struct sock *sk)
1726 {
1727         struct inet_connection_sock *icsk = inet_csk(sk);
1728
1729         tcp_init_sock(sk);
1730
1731         icsk->icsk_af_ops = &ipv6_specific;
1732
1733 #ifdef CONFIG_TCP_MD5SIG
1734         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1735 #endif
1736
1737         return 0;
1738 }
1739
1740 static void tcp_v6_destroy_sock(struct sock *sk)
1741 {
1742         tcp_v4_destroy_sock(sk);
1743         inet6_destroy_sock(sk);
1744 }
1745
1746 #ifdef CONFIG_PROC_FS
1747 /* Proc filesystem TCPv6 sock list dumping. */
1748 static void get_openreq6(struct seq_file *seq,
1749                          const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1750 {
1751         int ttd = req->expires - jiffies;
1752         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1753         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1754
1755         if (ttd < 0)
1756                 ttd = 0;
1757
1758         seq_printf(seq,
1759                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1760                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1761                    i,
1762                    src->s6_addr32[0], src->s6_addr32[1],
1763                    src->s6_addr32[2], src->s6_addr32[3],
1764                    inet_rsk(req)->ir_num,
1765                    dest->s6_addr32[0], dest->s6_addr32[1],
1766                    dest->s6_addr32[2], dest->s6_addr32[3],
1767                    ntohs(inet_rsk(req)->ir_rmt_port),
1768                    TCP_SYN_RECV,
1769                    0, 0, /* could print option size, but that is af dependent. */
1770                    1,   /* timers active (only the expire timer) */
1771                    jiffies_to_clock_t(ttd),
1772                    req->num_timeout,
1773                    from_kuid_munged(seq_user_ns(seq), uid),
1774                    0,  /* non standard timer */
1775                    0, /* open_requests have no inode */
1776                    0, req);
1777 }
1778
1779 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1780 {
1781         const struct in6_addr *dest, *src;
1782         __u16 destp, srcp;
1783         int timer_active;
1784         unsigned long timer_expires;
1785         const struct inet_sock *inet = inet_sk(sp);
1786         const struct tcp_sock *tp = tcp_sk(sp);
1787         const struct inet_connection_sock *icsk = inet_csk(sp);
1788         struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
1789
1790         dest  = &sp->sk_v6_daddr;
1791         src   = &sp->sk_v6_rcv_saddr;
1792         destp = ntohs(inet->inet_dport);
1793         srcp  = ntohs(inet->inet_sport);
1794
1795         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1796                 timer_active    = 1;
1797                 timer_expires   = icsk->icsk_timeout;
1798         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1799                 timer_active    = 4;
1800                 timer_expires   = icsk->icsk_timeout;
1801         } else if (timer_pending(&sp->sk_timer)) {
1802                 timer_active    = 2;
1803                 timer_expires   = sp->sk_timer.expires;
1804         } else {
1805                 timer_active    = 0;
1806                 timer_expires = jiffies;
1807         }
1808
1809         seq_printf(seq,
1810                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1811                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1812                    i,
1813                    src->s6_addr32[0], src->s6_addr32[1],
1814                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1815                    dest->s6_addr32[0], dest->s6_addr32[1],
1816                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1817                    sp->sk_state,
1818                    tp->write_seq-tp->snd_una,
1819                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1820                    timer_active,
1821                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1822                    icsk->icsk_retransmits,
1823                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1824                    icsk->icsk_probes_out,
1825                    sock_i_ino(sp),
1826                    atomic_read(&sp->sk_refcnt), sp,
1827                    jiffies_to_clock_t(icsk->icsk_rto),
1828                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1829                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1830                    tp->snd_cwnd,
1831                    sp->sk_state == TCP_LISTEN ?
1832                         (fastopenq ? fastopenq->max_qlen : 0) :
1833                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1834                    );
1835 }
1836
1837 static void get_timewait6_sock(struct seq_file *seq,
1838                                struct inet_timewait_sock *tw, int i)
1839 {
1840         const struct in6_addr *dest, *src;
1841         __u16 destp, srcp;
1842         s32 delta = tw->tw_ttd - inet_tw_time_stamp();
1843
1844         dest = &tw->tw_v6_daddr;
1845         src  = &tw->tw_v6_rcv_saddr;
1846         destp = ntohs(tw->tw_dport);
1847         srcp  = ntohs(tw->tw_sport);
1848
1849         seq_printf(seq,
1850                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1851                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1852                    i,
1853                    src->s6_addr32[0], src->s6_addr32[1],
1854                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1855                    dest->s6_addr32[0], dest->s6_addr32[1],
1856                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1857                    tw->tw_substate, 0, 0,
1858                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1859                    atomic_read(&tw->tw_refcnt), tw);
1860 }
1861
1862 static int tcp6_seq_show(struct seq_file *seq, void *v)
1863 {
1864         struct tcp_iter_state *st;
1865         struct sock *sk = v;
1866
1867         if (v == SEQ_START_TOKEN) {
1868                 seq_puts(seq,
1869                          "  sl  "
1870                          "local_address                         "
1871                          "remote_address                        "
1872                          "st tx_queue rx_queue tr tm->when retrnsmt"
1873                          "   uid  timeout inode\n");
1874                 goto out;
1875         }
1876         st = seq->private;
1877
1878         switch (st->state) {
1879         case TCP_SEQ_STATE_LISTENING:
1880         case TCP_SEQ_STATE_ESTABLISHED:
1881                 if (sk->sk_state == TCP_TIME_WAIT)
1882                         get_timewait6_sock(seq, v, st->num);
1883                 else
1884                         get_tcp6_sock(seq, v, st->num);
1885                 break;
1886         case TCP_SEQ_STATE_OPENREQ:
1887                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1888                 break;
1889         }
1890 out:
1891         return 0;
1892 }
1893
1894 static const struct file_operations tcp6_afinfo_seq_fops = {
1895         .owner   = THIS_MODULE,
1896         .open    = tcp_seq_open,
1897         .read    = seq_read,
1898         .llseek  = seq_lseek,
1899         .release = seq_release_net
1900 };
1901
1902 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1903         .name           = "tcp6",
1904         .family         = AF_INET6,
1905         .seq_fops       = &tcp6_afinfo_seq_fops,
1906         .seq_ops        = {
1907                 .show           = tcp6_seq_show,
1908         },
1909 };
1910
1911 int __net_init tcp6_proc_init(struct net *net)
1912 {
1913         return tcp_proc_register(net, &tcp6_seq_afinfo);
1914 }
1915
1916 void tcp6_proc_exit(struct net *net)
1917 {
1918         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1919 }
1920 #endif
1921
1922 static void tcp_v6_clear_sk(struct sock *sk, int size)
1923 {
1924         struct inet_sock *inet = inet_sk(sk);
1925
1926         /* we do not want to clear pinet6 field, because of RCU lookups */
1927         sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1928
1929         size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1930         memset(&inet->pinet6 + 1, 0, size);
1931 }
1932
1933 struct proto tcpv6_prot = {
1934         .name                   = "TCPv6",
1935         .owner                  = THIS_MODULE,
1936         .close                  = tcp_close,
1937         .connect                = tcp_v6_connect,
1938         .disconnect             = tcp_disconnect,
1939         .accept                 = inet_csk_accept,
1940         .ioctl                  = tcp_ioctl,
1941         .init                   = tcp_v6_init_sock,
1942         .destroy                = tcp_v6_destroy_sock,
1943         .shutdown               = tcp_shutdown,
1944         .setsockopt             = tcp_setsockopt,
1945         .getsockopt             = tcp_getsockopt,
1946         .recvmsg                = tcp_recvmsg,
1947         .sendmsg                = tcp_sendmsg,
1948         .sendpage               = tcp_sendpage,
1949         .backlog_rcv            = tcp_v6_do_rcv,
1950         .release_cb             = tcp_release_cb,
1951         .mtu_reduced            = tcp_v6_mtu_reduced,
1952         .hash                   = tcp_v6_hash,
1953         .unhash                 = inet_unhash,
1954         .get_port               = inet_csk_get_port,
1955         .enter_memory_pressure  = tcp_enter_memory_pressure,
1956         .stream_memory_free     = tcp_stream_memory_free,
1957         .sockets_allocated      = &tcp_sockets_allocated,
1958         .memory_allocated       = &tcp_memory_allocated,
1959         .memory_pressure        = &tcp_memory_pressure,
1960         .orphan_count           = &tcp_orphan_count,
1961         .sysctl_mem             = sysctl_tcp_mem,
1962         .sysctl_wmem            = sysctl_tcp_wmem,
1963         .sysctl_rmem            = sysctl_tcp_rmem,
1964         .max_header             = MAX_TCP_HEADER,
1965         .obj_size               = sizeof(struct tcp6_sock),
1966         .slab_flags             = SLAB_DESTROY_BY_RCU,
1967         .twsk_prot              = &tcp6_timewait_sock_ops,
1968         .rsk_prot               = &tcp6_request_sock_ops,
1969         .h.hashinfo             = &tcp_hashinfo,
1970         .no_autobind            = true,
1971 #ifdef CONFIG_COMPAT
1972         .compat_setsockopt      = compat_tcp_setsockopt,
1973         .compat_getsockopt      = compat_tcp_getsockopt,
1974 #endif
1975 #ifdef CONFIG_MEMCG_KMEM
1976         .proto_cgroup           = tcp_proto_cgroup,
1977 #endif
1978         .clear_sk               = tcp_v6_clear_sk,
1979 };
1980
1981 static const struct inet6_protocol tcpv6_protocol = {
1982         .early_demux    =       tcp_v6_early_demux,
1983         .handler        =       tcp_v6_rcv,
1984         .err_handler    =       tcp_v6_err,
1985         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1986 };
1987
1988 static struct inet_protosw tcpv6_protosw = {
1989         .type           =       SOCK_STREAM,
1990         .protocol       =       IPPROTO_TCP,
1991         .prot           =       &tcpv6_prot,
1992         .ops            =       &inet6_stream_ops,
1993         .no_check       =       0,
1994         .flags          =       INET_PROTOSW_PERMANENT |
1995                                 INET_PROTOSW_ICSK,
1996 };
1997
1998 static int __net_init tcpv6_net_init(struct net *net)
1999 {
2000         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2001                                     SOCK_RAW, IPPROTO_TCP, net);
2002 }
2003
2004 static void __net_exit tcpv6_net_exit(struct net *net)
2005 {
2006         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2007 }
2008
2009 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2010 {
2011         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2012 }
2013
2014 static struct pernet_operations tcpv6_net_ops = {
2015         .init       = tcpv6_net_init,
2016         .exit       = tcpv6_net_exit,
2017         .exit_batch = tcpv6_net_exit_batch,
2018 };
2019
2020 int __init tcpv6_init(void)
2021 {
2022         int ret;
2023
2024         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2025         if (ret)
2026                 goto out;
2027
2028         /* register inet6 protocol */
2029         ret = inet6_register_protosw(&tcpv6_protosw);
2030         if (ret)
2031                 goto out_tcpv6_protocol;
2032
2033         ret = register_pernet_subsys(&tcpv6_net_ops);
2034         if (ret)
2035                 goto out_tcpv6_protosw;
2036 out:
2037         return ret;
2038
2039 out_tcpv6_protosw:
2040         inet6_unregister_protosw(&tcpv6_protosw);
2041 out_tcpv6_protocol:
2042         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2043         goto out;
2044 }
2045
2046 void tcpv6_exit(void)
2047 {
2048         unregister_pernet_subsys(&tcpv6_net_ops);
2049         inet6_unregister_protosw(&tcpv6_protosw);
2050         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2051 }