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/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.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
 816         fl6.fl6_dport = t1->dest;
 817         fl6.fl6_sport = t1->source;
 818         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 819
 820         /* Pass a socket to ip6_dst_lookup either it is for RST
 821          * Underlying function will use this to retrieve the network
 822          * namespace
 823          */
 824         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 825         if (!IS_ERR(dst)) {
 826                 skb_dst_set(buff, dst);
 827                 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
 828                 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 829                 if (rst)
 830                         TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
 831                 return;
 832         }
 833
 834         kfree_skb(buff);
 835 }
 836
 837 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
 838 {
 839         const struct tcphdr *th = tcp_hdr(skb);
 840         u32 seq = 0, ack_seq = 0;
 841         struct tcp_md5sig_key *key = NULL;
 842 #ifdef CONFIG_TCP_MD5SIG
 843         const __u8 *hash_location = NULL;
 844         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 845         unsigned char newhash[16];
 846         int genhash;
 847         struct sock *sk1 = NULL;
 848 #endif
 849         int oif;
 850
 851         if (th->rst)
 852                 return;
 853
 854         if (!ipv6_unicast_destination(skb))
 855                 return;
 856
 857 #ifdef CONFIG_TCP_MD5SIG
 858         hash_location = tcp_parse_md5sig_option(th);
 859         if (!sk && hash_location) {
 860                 /*
 861                  * active side is lost. Try to find listening socket through
 862                  * source port, and then find md5 key through listening socket.
 863                  * we are not loose security here:
 864                  * Incoming packet is checked with md5 hash with finding key,
 865                  * no RST generated if md5 hash doesn't match.
 866                  */
 867                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
 868                                            &tcp_hashinfo, &ipv6h->saddr,
 869                                            th->source, &ipv6h->daddr,
 870                                            ntohs(th->source), inet6_iif(skb));
 871                 if (!sk1)
 872                         return;
 873
 874                 rcu_read_lock();
 875                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 876                 if (!key)
 877                         goto release_sk1;
 878
 879                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
 880                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
 881                         goto release_sk1;
 882         } else {
 883                 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
 884         }
 885 #endif
 886
 887         if (th->ack)
 888                 seq = ntohl(th->ack_seq);
 889         else
 890                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
 891                           (th->doff << 2);
 892
 893         oif = sk ? sk->sk_bound_dev_if : 0;
 894         tcp_v6_send_response(skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
 895
 896 #ifdef CONFIG_TCP_MD5SIG
 897 release_sk1:
 898         if (sk1) {
 899                 rcu_read_unlock();
 900                 sock_put(sk1);
 901         }
 902 #endif
 903 }
 904
 905 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
 906                             u32 win, u32 tsval, u32 tsecr, int oif,
 907                             struct tcp_md5sig_key *key, u8 tclass,
 908                             u32 label)
 909 {
 910         tcp_v6_send_response(skb, seq, ack, win, tsval, tsecr, oif, key, 0, tclass,
 911                              label);
 912 }
 913
 914 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 915 {
 916         struct inet_timewait_sock *tw = inet_twsk(sk);
 917         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 918
 919         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 920                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 921                         tcp_time_stamp + tcptw->tw_ts_offset,
 922                         tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
 923                         tw->tw_tclass, (tw->tw_flowlabel << 12));
 924
 925         inet_twsk_put(tw);
 926 }
 927
 928 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
 929                                   struct request_sock *req)
 930 {
 931         /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 932          * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 933          */
 934         tcp_v6_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
 935                         tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
 936                         tcp_rsk(req)->rcv_nxt,
 937                         req->rcv_wnd, tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
 938                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
 939                         0, 0);
 940 }
 941
 942
 943 static struct sock *tcp_v6_hnd_req(struct sock *sk, struct sk_buff *skb)
 944 {
 945         struct request_sock *req, **prev;
 946         const struct tcphdr *th = tcp_hdr(skb);
 947         struct sock *nsk;
 948
 949         /* Find possible connection requests. */
 950         req = inet6_csk_search_req(sk, &prev, th->source,
 951                                    &ipv6_hdr(skb)->saddr,
 952                                    &ipv6_hdr(skb)->daddr, inet6_iif(skb));
 953         if (req)
 954                 return tcp_check_req(sk, skb, req, prev, false);
 955
 956         nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
 957                         &ipv6_hdr(skb)->saddr, th->source,
 958                         &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
 959
 960         if (nsk) {
 961                 if (nsk->sk_state != TCP_TIME_WAIT) {
 962                         bh_lock_sock(nsk);
 963                         return nsk;
 964                 }
 965                 inet_twsk_put(inet_twsk(nsk));
 966                 return NULL;
 967         }
 968
 969 #ifdef CONFIG_SYN_COOKIES
 970         if (!th->syn)
 971                 sk = cookie_v6_check(sk, skb);
 972 #endif
 973         return sk;
 974 }
 975
 976 /* FIXME: this is substantially similar to the ipv4 code.
 977  * Can some kind of merge be done? -- erics
 978  */
 979 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 980 {
 981         struct tcp_options_received tmp_opt;
 982         struct request_sock *req;
 983         struct inet_request_sock *ireq;
 984         struct ipv6_pinfo *np = inet6_sk(sk);
 985         struct tcp_sock *tp = tcp_sk(sk);
 986         __u32 isn = TCP_SKB_CB(skb)->when;
 987         struct dst_entry *dst = NULL;
 988         struct tcp_fastopen_cookie foc = { .len = -1 };
 989         bool want_cookie = false, fastopen;
 990         struct flowi6 fl6;
 991         int err;
 992
 993         if (skb->protocol == htons(ETH_P_IP))
 994                 return tcp_v4_conn_request(sk, skb);
 995
 996         if (!ipv6_unicast_destination(skb))
 997                 goto drop;
 998
 999         if ((sysctl_tcp_syncookies == 2 ||
1000              inet_csk_reqsk_queue_is_full(sk)) && !isn) {
1001                 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
1002                 if (!want_cookie)
1003                         goto drop;
1004         }
1005
1006         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
1007                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1008                 goto drop;
1009         }
1010
1011         req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1012         if (req == NULL)
1013                 goto drop;
1014
1015 #ifdef CONFIG_TCP_MD5SIG
1016         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1017 #endif
1018
1019         tcp_clear_options(&tmp_opt);
1020         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1021         tmp_opt.user_mss = tp->rx_opt.user_mss;
1022         tcp_parse_options(skb, &tmp_opt, 0, want_cookie ? NULL : &foc);
1023
1024         if (want_cookie && !tmp_opt.saw_tstamp)
1025                 tcp_clear_options(&tmp_opt);
1026
1027         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1028         tcp_openreq_init(req, &tmp_opt, skb);
1029
1030         ireq = inet_rsk(req);
1031         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
1032         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
1033         if (!want_cookie || tmp_opt.tstamp_ok)
1034                 TCP_ECN_create_request(req, skb, sock_net(sk));
1035
1036         ireq->ir_iif = sk->sk_bound_dev_if;
1037         ireq->ir_mark = inet_request_mark(sk, skb);
1038
1039         /* So that link locals have meaning */
1040         if (!sk->sk_bound_dev_if &&
1041             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
1042                 ireq->ir_iif = inet6_iif(skb);
1043
1044         if (!isn) {
1045                 if (ipv6_opt_accepted(sk, skb) ||
1046                     np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1047                     np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim ||
1048                     np->repflow) {
1049                         atomic_inc(&skb->users);
1050                         ireq->pktopts = skb;
1051                 }
1052
1053                 if (want_cookie) {
1054                         isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1055                         req->cookie_ts = tmp_opt.tstamp_ok;
1056                         goto have_isn;
1057                 }
1058
1059                 /* VJ's idea. We save last timestamp seen
1060                  * from the destination in peer table, when entering
1061                  * state TIME-WAIT, and check against it before
1062                  * accepting new connection request.
1063                  *
1064                  * If "isn" is not zero, this request hit alive
1065                  * timewait bucket, so that all the necessary checks
1066                  * are made in the function processing timewait state.
1067                  */
1068                 if (tmp_opt.saw_tstamp &&
1069                     tcp_death_row.sysctl_tw_recycle &&
1070                     (dst = inet6_csk_route_req(sk, &fl6, req)) != NULL) {
1071                         if (!tcp_peer_is_proven(req, dst, true)) {
1072                                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1073                                 goto drop_and_release;
1074                         }
1075                 }
1076                 /* Kill the following clause, if you dislike this way. */
1077                 else if (!sysctl_tcp_syncookies &&
1078                          (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1079                           (sysctl_max_syn_backlog >> 2)) &&
1080                          !tcp_peer_is_proven(req, dst, false)) {
1081                         /* Without syncookies last quarter of
1082                          * backlog is filled with destinations,
1083                          * proven to be alive.
1084                          * It means that we continue to communicate
1085                          * to destinations, already remembered
1086                          * to the moment of synflood.
1087                          */
1088                         LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1089                                        &ireq->ir_v6_rmt_addr, ntohs(tcp_hdr(skb)->source));
1090                         goto drop_and_release;
1091                 }
1092
1093                 isn = tcp_v6_init_sequence(skb);
1094         }
1095 have_isn:
1096
1097         if (security_inet_conn_request(sk, skb, req))
1098                 goto drop_and_release;
1099
1100         if (!dst && (dst = inet6_csk_route_req(sk, &fl6, req)) == NULL)
1101                 goto drop_and_free;
1102
1103         tcp_rsk(req)->snt_isn = isn;
1104         tcp_rsk(req)->snt_synack = tcp_time_stamp;
1105         tcp_openreq_init_rwin(req, sk, dst);
1106         fastopen = !want_cookie &&
1107                    tcp_try_fastopen(sk, skb, req, &foc, dst);
1108         err = tcp_v6_send_synack(sk, dst, &fl6, req,
1109                                  skb_get_queue_mapping(skb), &foc);
1110         if (!fastopen) {
1111                 if (err || want_cookie)
1112                         goto drop_and_free;
1113
1114                 tcp_rsk(req)->listener = NULL;
1115                 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1116         }
1117         return 0;
1118
1119 drop_and_release:
1120         dst_release(dst);
1121 drop_and_free:
1122         reqsk_free(req);
1123 drop:
1124         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1125         return 0; /* don't send reset */
1126 }
1127
1128 static struct sock *tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1129                                          struct request_sock *req,
1130                                          struct dst_entry *dst)
1131 {
1132         struct inet_request_sock *ireq;
1133         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1134         struct tcp6_sock *newtcp6sk;
1135         struct inet_sock *newinet;
1136         struct tcp_sock *newtp;
1137         struct sock *newsk;
1138 #ifdef CONFIG_TCP_MD5SIG
1139         struct tcp_md5sig_key *key;
1140 #endif
1141         struct flowi6 fl6;
1142
1143         if (skb->protocol == htons(ETH_P_IP)) {
1144                 /*
1145                  *      v6 mapped
1146                  */
1147
1148                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1149
1150                 if (newsk == NULL)
1151                         return NULL;
1152
1153                 newtcp6sk = (struct tcp6_sock *)newsk;
1154                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1155
1156                 newinet = inet_sk(newsk);
1157                 newnp = inet6_sk(newsk);
1158                 newtp = tcp_sk(newsk);
1159
1160                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1161
1162                 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newsk->sk_v6_daddr);
1163
1164                 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1165
1166                 newsk->sk_v6_rcv_saddr = newnp->saddr;
1167
1168                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1169                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1170 #ifdef CONFIG_TCP_MD5SIG
1171                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1172 #endif
1173
1174                 newnp->ipv6_ac_list = NULL;
1175                 newnp->ipv6_fl_list = NULL;
1176                 newnp->pktoptions  = NULL;
1177                 newnp->opt         = NULL;
1178                 newnp->mcast_oif   = inet6_iif(skb);
1179                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1180                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1181                 if (np->repflow)
1182                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1183
1184                 /*
1185                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1186                  * here, tcp_create_openreq_child now does this for us, see the comment in
1187                  * that function for the gory details. -acme
1188                  */
1189
1190                 /* It is tricky place. Until this moment IPv4 tcp
1191                    worked with IPv6 icsk.icsk_af_ops.
1192                    Sync it now.
1193                  */
1194                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1195
1196                 return newsk;
1197         }
1198
1199         ireq = inet_rsk(req);
1200
1201         if (sk_acceptq_is_full(sk))
1202                 goto out_overflow;
1203
1204         if (!dst) {
1205                 dst = inet6_csk_route_req(sk, &fl6, req);
1206                 if (!dst)
1207                         goto out;
1208         }
1209
1210         newsk = tcp_create_openreq_child(sk, req, skb);
1211         if (newsk == NULL)
1212                 goto out_nonewsk;
1213
1214         /*
1215          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1216          * count here, tcp_create_openreq_child now does this for us, see the
1217          * comment in that function for the gory details. -acme
1218          */
1219
1220         newsk->sk_gso_type = SKB_GSO_TCPV6;
1221         __ip6_dst_store(newsk, dst, NULL, NULL);
1222         inet6_sk_rx_dst_set(newsk, skb);
1223
1224         newtcp6sk = (struct tcp6_sock *)newsk;
1225         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1226
1227         newtp = tcp_sk(newsk);
1228         newinet = inet_sk(newsk);
1229         newnp = inet6_sk(newsk);
1230
1231         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1232
1233         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1234         newnp->saddr = ireq->ir_v6_loc_addr;
1235         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1236         newsk->sk_bound_dev_if = ireq->ir_iif;
1237
1238         /* Now IPv6 options...
1239
1240            First: no IPv4 options.
1241          */
1242         newinet->inet_opt = NULL;
1243         newnp->ipv6_ac_list = NULL;
1244         newnp->ipv6_fl_list = NULL;
1245
1246         /* Clone RX bits */
1247         newnp->rxopt.all = np->rxopt.all;
1248
1249         /* Clone pktoptions received with SYN */
1250         newnp->pktoptions = NULL;
1251         if (ireq->pktopts != NULL) {
1252                 newnp->pktoptions = skb_clone(ireq->pktopts,
1253                                               sk_gfp_atomic(sk, GFP_ATOMIC));
1254                 consume_skb(ireq->pktopts);
1255                 ireq->pktopts = NULL;
1256                 if (newnp->pktoptions)
1257                         skb_set_owner_r(newnp->pktoptions, newsk);
1258         }
1259         newnp->opt        = NULL;
1260         newnp->mcast_oif  = inet6_iif(skb);
1261         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1262         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1263         if (np->repflow)
1264                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1265
1266         /* Clone native IPv6 options from listening socket (if any)
1267
1268            Yes, keeping reference count would be much more clever,
1269            but we make one more one thing there: reattach optmem
1270            to newsk.
1271          */
1272         if (np->opt)
1273                 newnp->opt = ipv6_dup_options(newsk, np->opt);
1274
1275         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1276         if (newnp->opt)
1277                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1278                                                      newnp->opt->opt_flen);
1279
1280         tcp_sync_mss(newsk, dst_mtu(dst));
1281         newtp->advmss = dst_metric_advmss(dst);
1282         if (tcp_sk(sk)->rx_opt.user_mss &&
1283             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1284                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1285
1286         tcp_initialize_rcv_mss(newsk);
1287
1288         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1289         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1290
1291 #ifdef CONFIG_TCP_MD5SIG
1292         /* Copy over the MD5 key from the original socket */
1293         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1294         if (key != NULL) {
1295                 /* We're using one, so create a matching key
1296                  * on the newsk structure. If we fail to get
1297                  * memory, then we end up not copying the key
1298                  * across. Shucks.
1299                  */
1300                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1301                                AF_INET6, key->key, key->keylen,
1302                                sk_gfp_atomic(sk, GFP_ATOMIC));
1303         }
1304 #endif
1305
1306         if (__inet_inherit_port(sk, newsk) < 0) {
1307                 inet_csk_prepare_forced_close(newsk);
1308                 tcp_done(newsk);
1309                 goto out;
1310         }
1311         __inet6_hash(newsk, NULL);
1312
1313         return newsk;
1314
1315 out_overflow:
1316         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1317 out_nonewsk:
1318         dst_release(dst);
1319 out:
1320         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1321         return NULL;
1322 }
1323
1324 /* The socket must have it's spinlock held when we get
1325  * here.
1326  *
1327  * We have a potential double-lock case here, so even when
1328  * doing backlog processing we use the BH locking scheme.
1329  * This is because we cannot sleep with the original spinlock
1330  * held.
1331  */
1332 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1333 {
1334         struct ipv6_pinfo *np = inet6_sk(sk);
1335         struct tcp_sock *tp;
1336         struct sk_buff *opt_skb = NULL;
1337
1338         /* Imagine: socket is IPv6. IPv4 packet arrives,
1339            goes to IPv4 receive handler and backlogged.
1340            From backlog it always goes here. Kerboom...
1341            Fortunately, tcp_rcv_established and rcv_established
1342            handle them correctly, but it is not case with
1343            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1344          */
1345
1346         if (skb->protocol == htons(ETH_P_IP))
1347                 return tcp_v4_do_rcv(sk, skb);
1348
1349 #ifdef CONFIG_TCP_MD5SIG
1350         if (tcp_v6_inbound_md5_hash(sk, skb))
1351                 goto discard;
1352 #endif
1353
1354         if (sk_filter(sk, skb))
1355                 goto discard;
1356
1357         /*
1358          *      socket locking is here for SMP purposes as backlog rcv
1359          *      is currently called with bh processing disabled.
1360          */
1361
1362         /* Do Stevens' IPV6_PKTOPTIONS.
1363
1364            Yes, guys, it is the only place in our code, where we
1365            may make it not affecting IPv4.
1366            The rest of code is protocol independent,
1367            and I do not like idea to uglify IPv4.
1368
1369            Actually, all the idea behind IPV6_PKTOPTIONS
1370            looks not very well thought. For now we latch
1371            options, received in the last packet, enqueued
1372            by tcp. Feel free to propose better solution.
1373                                                --ANK (980728)
1374          */
1375         if (np->rxopt.all)
1376                 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1377
1378         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1379                 struct dst_entry *dst = sk->sk_rx_dst;
1380
1381                 sock_rps_save_rxhash(sk, skb);
1382                 if (dst) {
1383                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1384                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1385                                 dst_release(dst);
1386                                 sk->sk_rx_dst = NULL;
1387                         }
1388                 }
1389
1390                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1391                 if (opt_skb)
1392                         goto ipv6_pktoptions;
1393                 return 0;
1394         }
1395
1396         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1397                 goto csum_err;
1398
1399         if (sk->sk_state == TCP_LISTEN) {
1400                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1401                 if (!nsk)
1402                         goto discard;
1403
1404                 /*
1405                  * Queue it on the new socket if the new socket is active,
1406                  * otherwise we just shortcircuit this and continue with
1407                  * the new socket..
1408                  */
1409                 if (nsk != sk) {
1410                         sock_rps_save_rxhash(nsk, skb);
1411                         if (tcp_child_process(sk, nsk, skb))
1412                                 goto reset;
1413                         if (opt_skb)
1414                                 __kfree_skb(opt_skb);
1415                         return 0;
1416                 }
1417         } else
1418                 sock_rps_save_rxhash(sk, skb);
1419
1420         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1421                 goto reset;
1422         if (opt_skb)
1423                 goto ipv6_pktoptions;
1424         return 0;
1425
1426 reset:
1427         tcp_v6_send_reset(sk, skb);
1428 discard:
1429         if (opt_skb)
1430                 __kfree_skb(opt_skb);
1431         kfree_skb(skb);
1432         return 0;
1433 csum_err:
1434         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1435         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1436         goto discard;
1437
1438
1439 ipv6_pktoptions:
1440         /* Do you ask, what is it?
1441
1442            1. skb was enqueued by tcp.
1443            2. skb is added to tail of read queue, rather than out of order.
1444            3. socket is not in passive state.
1445            4. Finally, it really contains options, which user wants to receive.
1446          */
1447         tp = tcp_sk(sk);
1448         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1449             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1450                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1451                         np->mcast_oif = inet6_iif(opt_skb);
1452                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1453                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1454                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1455                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1456                 if (np->repflow)
1457                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1458                 if (ipv6_opt_accepted(sk, opt_skb)) {
1459                         skb_set_owner_r(opt_skb, sk);
1460                         opt_skb = xchg(&np->pktoptions, opt_skb);
1461                 } else {
1462                         __kfree_skb(opt_skb);
1463                         opt_skb = xchg(&np->pktoptions, NULL);
1464                 }
1465         }
1466
1467         kfree_skb(opt_skb);
1468         return 0;
1469 }
1470
1471 static int tcp_v6_rcv(struct sk_buff *skb)
1472 {
1473         const struct tcphdr *th;
1474         const struct ipv6hdr *hdr;
1475         struct sock *sk;
1476         int ret;
1477         struct net *net = dev_net(skb->dev);
1478
1479         if (skb->pkt_type != PACKET_HOST)
1480                 goto discard_it;
1481
1482         /*
1483          *      Count it even if it's bad.
1484          */
1485         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1486
1487         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1488                 goto discard_it;
1489
1490         th = tcp_hdr(skb);
1491
1492         if (th->doff < sizeof(struct tcphdr)/4)
1493                 goto bad_packet;
1494         if (!pskb_may_pull(skb, th->doff*4))
1495                 goto discard_it;
1496
1497         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1498                 goto csum_error;
1499
1500         th = tcp_hdr(skb);
1501         hdr = ipv6_hdr(skb);
1502         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1503         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1504                                     skb->len - th->doff*4);
1505         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1506         TCP_SKB_CB(skb)->when = 0;
1507         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1508         TCP_SKB_CB(skb)->sacked = 0;
1509
1510         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1511         if (!sk)
1512                 goto no_tcp_socket;
1513
1514 process:
1515         if (sk->sk_state == TCP_TIME_WAIT)
1516                 goto do_time_wait;
1517
1518         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1519                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1520                 goto discard_and_relse;
1521         }
1522
1523         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1524                 goto discard_and_relse;
1525
1526         if (sk_filter(sk, skb))
1527                 goto discard_and_relse;
1528
1529         sk_mark_napi_id(sk, skb);
1530         skb->dev = NULL;
1531
1532         bh_lock_sock_nested(sk);
1533         ret = 0;
1534         if (!sock_owned_by_user(sk)) {
1535 #ifdef CONFIG_NET_DMA
1536                 struct tcp_sock *tp = tcp_sk(sk);
1537                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1538                         tp->ucopy.dma_chan = net_dma_find_channel();
1539                 if (tp->ucopy.dma_chan)
1540                         ret = tcp_v6_do_rcv(sk, skb);
1541                 else
1542 #endif
1543                 {
1544                         if (!tcp_prequeue(sk, skb))
1545                                 ret = tcp_v6_do_rcv(sk, skb);
1546                 }
1547         } else if (unlikely(sk_add_backlog(sk, skb,
1548                                            sk->sk_rcvbuf + sk->sk_sndbuf))) {
1549                 bh_unlock_sock(sk);
1550                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1551                 goto discard_and_relse;
1552         }
1553         bh_unlock_sock(sk);
1554
1555         sock_put(sk);
1556         return ret ? -1 : 0;
1557
1558 no_tcp_socket:
1559         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1560                 goto discard_it;
1561
1562         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1563 csum_error:
1564                 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1565 bad_packet:
1566                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1567         } else {
1568                 tcp_v6_send_reset(NULL, skb);
1569         }
1570
1571 discard_it:
1572         kfree_skb(skb);
1573         return 0;
1574
1575 discard_and_relse:
1576         sock_put(sk);
1577         goto discard_it;
1578
1579 do_time_wait:
1580         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1581                 inet_twsk_put(inet_twsk(sk));
1582                 goto discard_it;
1583         }
1584
1585         if (skb->len < (th->doff<<2)) {
1586                 inet_twsk_put(inet_twsk(sk));
1587                 goto bad_packet;
1588         }
1589         if (tcp_checksum_complete(skb)) {
1590                 inet_twsk_put(inet_twsk(sk));
1591                 goto csum_error;
1592         }
1593
1594         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1595         case TCP_TW_SYN:
1596         {
1597                 struct sock *sk2;
1598
1599                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1600                                             &ipv6_hdr(skb)->saddr, th->source,
1601                                             &ipv6_hdr(skb)->daddr,
1602                                             ntohs(th->dest), inet6_iif(skb));
1603                 if (sk2 != NULL) {
1604                         struct inet_timewait_sock *tw = inet_twsk(sk);
1605                         inet_twsk_deschedule(tw, &tcp_death_row);
1606                         inet_twsk_put(tw);
1607                         sk = sk2;
1608                         goto process;
1609                 }
1610                 /* Fall through to ACK */
1611         }
1612         case TCP_TW_ACK:
1613                 tcp_v6_timewait_ack(sk, skb);
1614                 break;
1615         case TCP_TW_RST:
1616                 goto no_tcp_socket;
1617         case TCP_TW_SUCCESS:
1618                 ;
1619         }
1620         goto discard_it;
1621 }
1622
1623 static void tcp_v6_early_demux(struct sk_buff *skb)
1624 {
1625         const struct ipv6hdr *hdr;
1626         const struct tcphdr *th;
1627         struct sock *sk;
1628
1629         if (skb->pkt_type != PACKET_HOST)
1630                 return;
1631
1632         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1633                 return;
1634
1635         hdr = ipv6_hdr(skb);
1636         th = tcp_hdr(skb);
1637
1638         if (th->doff < sizeof(struct tcphdr) / 4)
1639                 return;
1640
1641         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1642                                         &hdr->saddr, th->source,
1643                                         &hdr->daddr, ntohs(th->dest),
1644                                         inet6_iif(skb));
1645         if (sk) {
1646                 skb->sk = sk;
1647                 skb->destructor = sock_edemux;
1648                 if (sk->sk_state != TCP_TIME_WAIT) {
1649                         struct dst_entry *dst = sk->sk_rx_dst;
1650
1651                         if (dst)
1652                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1653                         if (dst &&
1654                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1655                                 skb_dst_set_noref(skb, dst);
1656                 }
1657         }
1658 }
1659
1660 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1661         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1662         .twsk_unique    = tcp_twsk_unique,
1663         .twsk_destructor = tcp_twsk_destructor,
1664 };
1665
1666 static const struct inet_connection_sock_af_ops ipv6_specific = {
1667         .queue_xmit        = inet6_csk_xmit,
1668         .send_check        = tcp_v6_send_check,
1669         .rebuild_header    = inet6_sk_rebuild_header,
1670         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1671         .conn_request      = tcp_v6_conn_request,
1672         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1673         .net_header_len    = sizeof(struct ipv6hdr),
1674         .net_frag_header_len = sizeof(struct frag_hdr),
1675         .setsockopt        = ipv6_setsockopt,
1676         .getsockopt        = ipv6_getsockopt,
1677         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1678         .sockaddr_len      = sizeof(struct sockaddr_in6),
1679         .bind_conflict     = inet6_csk_bind_conflict,
1680 #ifdef CONFIG_COMPAT
1681         .compat_setsockopt = compat_ipv6_setsockopt,
1682         .compat_getsockopt = compat_ipv6_getsockopt,
1683 #endif
1684 };
1685
1686 #ifdef CONFIG_TCP_MD5SIG
1687 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1688         .md5_lookup     =       tcp_v6_md5_lookup,
1689         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1690         .md5_parse      =       tcp_v6_parse_md5_keys,
1691 };
1692 #endif
1693
1694 /*
1695  *      TCP over IPv4 via INET6 API
1696  */
1697 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1698         .queue_xmit        = ip_queue_xmit,
1699         .send_check        = tcp_v4_send_check,
1700         .rebuild_header    = inet_sk_rebuild_header,
1701         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1702         .conn_request      = tcp_v6_conn_request,
1703         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1704         .net_header_len    = sizeof(struct iphdr),
1705         .setsockopt        = ipv6_setsockopt,
1706         .getsockopt        = ipv6_getsockopt,
1707         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1708         .sockaddr_len      = sizeof(struct sockaddr_in6),
1709         .bind_conflict     = inet6_csk_bind_conflict,
1710 #ifdef CONFIG_COMPAT
1711         .compat_setsockopt = compat_ipv6_setsockopt,
1712         .compat_getsockopt = compat_ipv6_getsockopt,
1713 #endif
1714 };
1715
1716 #ifdef CONFIG_TCP_MD5SIG
1717 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1718         .md5_lookup     =       tcp_v4_md5_lookup,
1719         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1720         .md5_parse      =       tcp_v6_parse_md5_keys,
1721 };
1722 #endif
1723
1724 /* NOTE: A lot of things set to zero explicitly by call to
1725  *       sk_alloc() so need not be done here.
1726  */
1727 static int tcp_v6_init_sock(struct sock *sk)
1728 {
1729         struct inet_connection_sock *icsk = inet_csk(sk);
1730
1731         tcp_init_sock(sk);
1732
1733         icsk->icsk_af_ops = &ipv6_specific;
1734
1735 #ifdef CONFIG_TCP_MD5SIG
1736         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1737 #endif
1738
1739         return 0;
1740 }
1741
1742 static void tcp_v6_destroy_sock(struct sock *sk)
1743 {
1744         tcp_v4_destroy_sock(sk);
1745         inet6_destroy_sock(sk);
1746 }
1747
1748 #ifdef CONFIG_PROC_FS
1749 /* Proc filesystem TCPv6 sock list dumping. */
1750 static void get_openreq6(struct seq_file *seq,
1751                          const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1752 {
1753         int ttd = req->expires - jiffies;
1754         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1755         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1756
1757         if (ttd < 0)
1758                 ttd = 0;
1759
1760         seq_printf(seq,
1761                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1762                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1763                    i,
1764                    src->s6_addr32[0], src->s6_addr32[1],
1765                    src->s6_addr32[2], src->s6_addr32[3],
1766                    inet_rsk(req)->ir_num,
1767                    dest->s6_addr32[0], dest->s6_addr32[1],
1768                    dest->s6_addr32[2], dest->s6_addr32[3],
1769                    ntohs(inet_rsk(req)->ir_rmt_port),
1770                    TCP_SYN_RECV,
1771                    0, 0, /* could print option size, but that is af dependent. */
1772                    1,   /* timers active (only the expire timer) */
1773                    jiffies_to_clock_t(ttd),
1774                    req->num_timeout,
1775                    from_kuid_munged(seq_user_ns(seq), uid),
1776                    0,  /* non standard timer */
1777                    0, /* open_requests have no inode */
1778                    0, req);
1779 }
1780
1781 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1782 {
1783         const struct in6_addr *dest, *src;
1784         __u16 destp, srcp;
1785         int timer_active;
1786         unsigned long timer_expires;
1787         const struct inet_sock *inet = inet_sk(sp);
1788         const struct tcp_sock *tp = tcp_sk(sp);
1789         const struct inet_connection_sock *icsk = inet_csk(sp);
1790         struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
1791
1792         dest  = &sp->sk_v6_daddr;
1793         src   = &sp->sk_v6_rcv_saddr;
1794         destp = ntohs(inet->inet_dport);
1795         srcp  = ntohs(inet->inet_sport);
1796
1797         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1798                 timer_active    = 1;
1799                 timer_expires   = icsk->icsk_timeout;
1800         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1801                 timer_active    = 4;
1802                 timer_expires   = icsk->icsk_timeout;
1803         } else if (timer_pending(&sp->sk_timer)) {
1804                 timer_active    = 2;
1805                 timer_expires   = sp->sk_timer.expires;
1806         } else {
1807                 timer_active    = 0;
1808                 timer_expires = jiffies;
1809         }
1810
1811         seq_printf(seq,
1812                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1813                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1814                    i,
1815                    src->s6_addr32[0], src->s6_addr32[1],
1816                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1817                    dest->s6_addr32[0], dest->s6_addr32[1],
1818                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1819                    sp->sk_state,
1820                    tp->write_seq-tp->snd_una,
1821                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1822                    timer_active,
1823                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1824                    icsk->icsk_retransmits,
1825                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1826                    icsk->icsk_probes_out,
1827                    sock_i_ino(sp),
1828                    atomic_read(&sp->sk_refcnt), sp,
1829                    jiffies_to_clock_t(icsk->icsk_rto),
1830                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1831                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1832                    tp->snd_cwnd,
1833                    sp->sk_state == TCP_LISTEN ?
1834                         (fastopenq ? fastopenq->max_qlen : 0) :
1835                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1836                    );
1837 }
1838
1839 static void get_timewait6_sock(struct seq_file *seq,
1840                                struct inet_timewait_sock *tw, int i)
1841 {
1842         const struct in6_addr *dest, *src;
1843         __u16 destp, srcp;
1844         s32 delta = tw->tw_ttd - inet_tw_time_stamp();
1845
1846         dest = &tw->tw_v6_daddr;
1847         src  = &tw->tw_v6_rcv_saddr;
1848         destp = ntohs(tw->tw_dport);
1849         srcp  = ntohs(tw->tw_sport);
1850
1851         seq_printf(seq,
1852                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1853                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1854                    i,
1855                    src->s6_addr32[0], src->s6_addr32[1],
1856                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1857                    dest->s6_addr32[0], dest->s6_addr32[1],
1858                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1859                    tw->tw_substate, 0, 0,
1860                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1861                    atomic_read(&tw->tw_refcnt), tw);
1862 }
1863
1864 static int tcp6_seq_show(struct seq_file *seq, void *v)
1865 {
1866         struct tcp_iter_state *st;
1867         struct sock *sk = v;
1868
1869         if (v == SEQ_START_TOKEN) {
1870                 seq_puts(seq,
1871                          "  sl  "
1872                          "local_address                         "
1873                          "remote_address                        "
1874                          "st tx_queue rx_queue tr tm->when retrnsmt"
1875                          "   uid  timeout inode\n");
1876                 goto out;
1877         }
1878         st = seq->private;
1879
1880         switch (st->state) {
1881         case TCP_SEQ_STATE_LISTENING:
1882         case TCP_SEQ_STATE_ESTABLISHED:
1883                 if (sk->sk_state == TCP_TIME_WAIT)
1884                         get_timewait6_sock(seq, v, st->num);
1885                 else
1886                         get_tcp6_sock(seq, v, st->num);
1887                 break;
1888         case TCP_SEQ_STATE_OPENREQ:
1889                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1890                 break;
1891         }
1892 out:
1893         return 0;
1894 }
1895
1896 static const struct file_operations tcp6_afinfo_seq_fops = {
1897         .owner   = THIS_MODULE,
1898         .open    = tcp_seq_open,
1899         .read    = seq_read,
1900         .llseek  = seq_lseek,
1901         .release = seq_release_net
1902 };
1903
1904 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1905         .name           = "tcp6",
1906         .family         = AF_INET6,
1907         .seq_fops       = &tcp6_afinfo_seq_fops,
1908         .seq_ops        = {
1909                 .show           = tcp6_seq_show,
1910         },
1911 };
1912
1913 int __net_init tcp6_proc_init(struct net *net)
1914 {
1915         return tcp_proc_register(net, &tcp6_seq_afinfo);
1916 }
1917
1918 void tcp6_proc_exit(struct net *net)
1919 {
1920         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1921 }
1922 #endif
1923
1924 static void tcp_v6_clear_sk(struct sock *sk, int size)
1925 {
1926         struct inet_sock *inet = inet_sk(sk);
1927
1928         /* we do not want to clear pinet6 field, because of RCU lookups */
1929         sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1930
1931         size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1932         memset(&inet->pinet6 + 1, 0, size);
1933 }
1934
1935 struct proto tcpv6_prot = {
1936         .name                   = "TCPv6",
1937         .owner                  = THIS_MODULE,
1938         .close                  = tcp_close,
1939         .connect                = tcp_v6_connect,
1940         .disconnect             = tcp_disconnect,
1941         .accept                 = inet_csk_accept,
1942         .ioctl                  = tcp_ioctl,
1943         .init                   = tcp_v6_init_sock,
1944         .destroy                = tcp_v6_destroy_sock,
1945         .shutdown               = tcp_shutdown,
1946         .setsockopt             = tcp_setsockopt,
1947         .getsockopt             = tcp_getsockopt,
1948         .recvmsg                = tcp_recvmsg,
1949         .sendmsg                = tcp_sendmsg,
1950         .sendpage               = tcp_sendpage,
1951         .backlog_rcv            = tcp_v6_do_rcv,
1952         .release_cb             = tcp_release_cb,
1953         .mtu_reduced            = tcp_v6_mtu_reduced,
1954         .hash                   = tcp_v6_hash,
1955         .unhash                 = inet_unhash,
1956         .get_port               = inet_csk_get_port,
1957         .enter_memory_pressure  = tcp_enter_memory_pressure,
1958         .stream_memory_free     = tcp_stream_memory_free,
1959         .sockets_allocated      = &tcp_sockets_allocated,
1960         .memory_allocated       = &tcp_memory_allocated,
1961         .memory_pressure        = &tcp_memory_pressure,
1962         .orphan_count           = &tcp_orphan_count,
1963         .sysctl_mem             = sysctl_tcp_mem,
1964         .sysctl_wmem            = sysctl_tcp_wmem,
1965         .sysctl_rmem            = sysctl_tcp_rmem,
1966         .max_header             = MAX_TCP_HEADER,
1967         .obj_size               = sizeof(struct tcp6_sock),
1968         .slab_flags             = SLAB_DESTROY_BY_RCU,
1969         .twsk_prot              = &tcp6_timewait_sock_ops,
1970         .rsk_prot               = &tcp6_request_sock_ops,
1971         .h.hashinfo             = &tcp_hashinfo,
1972         .no_autobind            = true,
1973 #ifdef CONFIG_COMPAT
1974         .compat_setsockopt      = compat_tcp_setsockopt,
1975         .compat_getsockopt      = compat_tcp_getsockopt,
1976 #endif
1977 #ifdef CONFIG_MEMCG_KMEM
1978         .proto_cgroup           = tcp_proto_cgroup,
1979 #endif
1980         .clear_sk               = tcp_v6_clear_sk,
1981 };
1982
1983 static const struct inet6_protocol tcpv6_protocol = {
1984         .early_demux    =       tcp_v6_early_demux,
1985         .handler        =       tcp_v6_rcv,
1986         .err_handler    =       tcp_v6_err,
1987         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1988 };
1989
1990 static struct inet_protosw tcpv6_protosw = {
1991         .type           =       SOCK_STREAM,
1992         .protocol       =       IPPROTO_TCP,
1993         .prot           =       &tcpv6_prot,
1994         .ops            =       &inet6_stream_ops,
1995         .flags          =       INET_PROTOSW_PERMANENT |
1996                                 INET_PROTOSW_ICSK,
1997 };
1998
1999 static int __net_init tcpv6_net_init(struct net *net)
2000 {
2001         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2002                                     SOCK_RAW, IPPROTO_TCP, net);
2003 }
2004
2005 static void __net_exit tcpv6_net_exit(struct net *net)
2006 {
2007         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2008 }
2009
2010 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2011 {
2012         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2013 }
2014
2015 static struct pernet_operations tcpv6_net_ops = {
2016         .init       = tcpv6_net_init,
2017         .exit       = tcpv6_net_exit,
2018         .exit_batch = tcpv6_net_exit_batch,
2019 };
2020
2021 int __init tcpv6_init(void)
2022 {
2023         int ret;
2024
2025         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2026         if (ret)
2027                 goto out;
2028
2029         /* register inet6 protocol */
2030         ret = inet6_register_protosw(&tcpv6_protosw);
2031         if (ret)
2032                 goto out_tcpv6_protocol;
2033
2034         ret = register_pernet_subsys(&tcpv6_net_ops);
2035         if (ret)
2036                 goto out_tcpv6_protosw;
2037 out:
2038         return ret;
2039
2040 out_tcpv6_protosw:
2041         inet6_unregister_protosw(&tcpv6_protosw);
2042 out_tcpv6_protocol:
2043         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2044         goto out;
2045 }
2046
2047 void tcpv6_exit(void)
2048 {
2049         unregister_pernet_subsys(&tcpv6_net_ops);
2050         inet6_unregister_protosw(&tcpv6_protosw);
2051         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2052 }