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/inet_common.h>
  63 #include <net/secure_seq.h>
  64 #include <net/tcp_memcontrol.h>
  65 #include <net/busy_poll.h>
  66
  67 #include <linux/proc_fs.h>
  68 #include <linux/seq_file.h>
  69
  70 #include <linux/crypto.h>
  71 #include <linux/scatterlist.h>
  72
  73 static void     tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
  74 static void     tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
  75                                       struct request_sock *req);
  76
  77 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  78
  79 static const struct inet_connection_sock_af_ops ipv6_mapped;
  80 static const struct inet_connection_sock_af_ops ipv6_specific;
  81 #ifdef CONFIG_TCP_MD5SIG
  82 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
  83 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
  84 #else
  85 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
  86                                                    const struct in6_addr *addr)
  87 {
  88         return NULL;
  89 }
  90 #endif
  91
  92 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
  93 {
  94         struct dst_entry *dst = skb_dst(skb);
  95
  96         if (dst) {
  97                 const struct rt6_info *rt = (const struct rt6_info *)dst;
  98
  99                 dst_hold(dst);
 100                 sk->sk_rx_dst = dst;
 101                 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
 102                 if (rt->rt6i_node)
 103                         inet6_sk(sk)->rx_dst_cookie = rt->rt6i_node->fn_sernum;
 104         }
 105 }
 106
 107 static void tcp_v6_hash(struct sock *sk)
 108 {
 109         if (sk->sk_state != TCP_CLOSE) {
 110                 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
 111                         tcp_prot.hash(sk);
 112                         return;
 113                 }
 114                 local_bh_disable();
 115                 __inet6_hash(sk, NULL);
 116                 local_bh_enable();
 117         }
 118 }
 119
 120 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
 121 {
 122         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
 123                                             ipv6_hdr(skb)->saddr.s6_addr32,
 124                                             tcp_hdr(skb)->dest,
 125                                             tcp_hdr(skb)->source);
 126 }
 127
 128 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 129                           int addr_len)
 130 {
 131         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 132         struct inet_sock *inet = inet_sk(sk);
 133         struct inet_connection_sock *icsk = inet_csk(sk);
 134         struct ipv6_pinfo *np = inet6_sk(sk);
 135         struct tcp_sock *tp = tcp_sk(sk);
 136         struct in6_addr *saddr = NULL, *final_p, final;
 137         struct rt6_info *rt;
 138         struct flowi6 fl6;
 139         struct dst_entry *dst;
 140         int addr_type;
 141         int err;
 142
 143         if (addr_len < SIN6_LEN_RFC2133)
 144                 return -EINVAL;
 145
 146         if (usin->sin6_family != AF_INET6)
 147                 return -EAFNOSUPPORT;
 148
 149         memset(&fl6, 0, sizeof(fl6));
 150
 151         if (np->sndflow) {
 152                 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 153                 IP6_ECN_flow_init(fl6.flowlabel);
 154                 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
 155                         struct ip6_flowlabel *flowlabel;
 156                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 157                         if (flowlabel == NULL)
 158                                 return -EINVAL;
 159                         fl6_sock_release(flowlabel);
 160                 }
 161         }
 162
 163         /*
 164          *      connect() to INADDR_ANY means loopback (BSD'ism).
 165          */
 166
 167         if (ipv6_addr_any(&usin->sin6_addr))
 168                 usin->sin6_addr.s6_addr[15] = 0x1;
 169
 170         addr_type = ipv6_addr_type(&usin->sin6_addr);
 171
 172         if (addr_type & IPV6_ADDR_MULTICAST)
 173                 return -ENETUNREACH;
 174
 175         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 176                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 177                     usin->sin6_scope_id) {
 178                         /* If interface is set while binding, indices
 179                          * must coincide.
 180                          */
 181                         if (sk->sk_bound_dev_if &&
 182                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 183                                 return -EINVAL;
 184
 185                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 186                 }
 187
 188                 /* Connect to link-local address requires an interface */
 189                 if (!sk->sk_bound_dev_if)
 190                         return -EINVAL;
 191         }
 192
 193         if (tp->rx_opt.ts_recent_stamp &&
 194             !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
 195                 tp->rx_opt.ts_recent = 0;
 196                 tp->rx_opt.ts_recent_stamp = 0;
 197                 tp->write_seq = 0;
 198         }
 199
 200         sk->sk_v6_daddr = usin->sin6_addr;
 201         np->flow_label = fl6.flowlabel;
 202
 203         ip6_set_txhash(sk);
 204
 205         /*
 206          *      TCP over IPv4
 207          */
 208
 209         if (addr_type == IPV6_ADDR_MAPPED) {
 210                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 211                 struct sockaddr_in sin;
 212
 213                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 214
 215                 if (__ipv6_only_sock(sk))
 216                         return -ENETUNREACH;
 217
 218                 sin.sin_family = AF_INET;
 219                 sin.sin_port = usin->sin6_port;
 220                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 221
 222                 icsk->icsk_af_ops = &ipv6_mapped;
 223                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 224 #ifdef CONFIG_TCP_MD5SIG
 225                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
 226 #endif
 227
 228                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 229
 230                 if (err) {
 231                         icsk->icsk_ext_hdr_len = exthdrlen;
 232                         icsk->icsk_af_ops = &ipv6_specific;
 233                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 234 #ifdef CONFIG_TCP_MD5SIG
 235                         tp->af_specific = &tcp_sock_ipv6_specific;
 236 #endif
 237                         goto failure;
 238                 } else {
 239                         ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
 240                         ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
 241                                                &sk->sk_v6_rcv_saddr);
 242                 }
 243
 244                 return err;
 245         }
 246
 247         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 248                 saddr = &sk->sk_v6_rcv_saddr;
 249
 250         fl6.flowi6_proto = IPPROTO_TCP;
 251         fl6.daddr = sk->sk_v6_daddr;
 252         fl6.saddr = saddr ? *saddr : np->saddr;
 253         fl6.flowi6_oif = sk->sk_bound_dev_if;
 254         fl6.flowi6_mark = sk->sk_mark;
 255         fl6.fl6_dport = usin->sin6_port;
 256         fl6.fl6_sport = inet->inet_sport;
 257
 258         final_p = fl6_update_dst(&fl6, np->opt, &final);
 259
 260         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 261
 262         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 263         if (IS_ERR(dst)) {
 264                 err = PTR_ERR(dst);
 265                 goto failure;
 266         }
 267
 268         if (saddr == NULL) {
 269                 saddr = &fl6.saddr;
 270                 sk->sk_v6_rcv_saddr = *saddr;
 271         }
 272
 273         /* set the source address */
 274         np->saddr = *saddr;
 275         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 276
 277         sk->sk_gso_type = SKB_GSO_TCPV6;
 278         __ip6_dst_store(sk, dst, NULL, NULL);
 279
 280         rt = (struct rt6_info *) dst;
 281         if (tcp_death_row.sysctl_tw_recycle &&
 282             !tp->rx_opt.ts_recent_stamp &&
 283             ipv6_addr_equal(&rt->rt6i_dst.addr, &sk->sk_v6_daddr))
 284                 tcp_fetch_timewait_stamp(sk, dst);
 285
 286         icsk->icsk_ext_hdr_len = 0;
 287         if (np->opt)
 288                 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
 289                                           np->opt->opt_nflen);
 290
 291         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 292
 293         inet->inet_dport = usin->sin6_port;
 294
 295         tcp_set_state(sk, TCP_SYN_SENT);
 296         err = inet6_hash_connect(&tcp_death_row, sk);
 297         if (err)
 298                 goto late_failure;
 299
 300         if (!tp->write_seq && likely(!tp->repair))
 301                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 302                                                              sk->sk_v6_daddr.s6_addr32,
 303                                                              inet->inet_sport,
 304                                                              inet->inet_dport);
 305
 306         err = tcp_connect(sk);
 307         if (err)
 308                 goto late_failure;
 309
 310         return 0;
 311
 312 late_failure:
 313         tcp_set_state(sk, TCP_CLOSE);
 314         __sk_dst_reset(sk);
 315 failure:
 316         inet->inet_dport = 0;
 317         sk->sk_route_caps = 0;
 318         return err;
 319 }
 320
 321 static void tcp_v6_mtu_reduced(struct sock *sk)
 322 {
 323         struct dst_entry *dst;
 324
 325         if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 326                 return;
 327
 328         dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
 329         if (!dst)
 330                 return;
 331
 332         if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
 333                 tcp_sync_mss(sk, dst_mtu(dst));
 334                 tcp_simple_retransmit(sk);
 335         }
 336 }
 337
 338 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 339                 u8 type, u8 code, int offset, __be32 info)
 340 {
 341         const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
 342         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 343         struct ipv6_pinfo *np;
 344         struct sock *sk;
 345         int err;
 346         struct tcp_sock *tp;
 347         struct request_sock *fastopen;
 348         __u32 seq, snd_una;
 349         struct net *net = dev_net(skb->dev);
 350
 351         sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
 352                         th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
 353
 354         if (sk == NULL) {
 355                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 356                                    ICMP6_MIB_INERRORS);
 357                 return;
 358         }
 359
 360         if (sk->sk_state == TCP_TIME_WAIT) {
 361                 inet_twsk_put(inet_twsk(sk));
 362                 return;
 363         }
 364
 365         bh_lock_sock(sk);
 366         if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
 367                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 368
 369         if (sk->sk_state == TCP_CLOSE)
 370                 goto out;
 371
 372         if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
 373                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
 374                 goto out;
 375         }
 376
 377         tp = tcp_sk(sk);
 378         seq = ntohl(th->seq);
 379         /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 380         fastopen = tp->fastopen_rsk;
 381         snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 382         if (sk->sk_state != TCP_LISTEN &&
 383             !between(seq, snd_una, tp->snd_nxt)) {
 384                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 385                 goto out;
 386         }
 387
 388         np = inet6_sk(sk);
 389
 390         if (type == NDISC_REDIRECT) {
 391                 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 392
 393                 if (dst)
 394                         dst->ops->redirect(dst, sk, skb);
 395                 goto out;
 396         }
 397
 398         if (type == ICMPV6_PKT_TOOBIG) {
 399                 /* We are not interested in TCP_LISTEN and open_requests
 400                  * (SYN-ACKs send out by Linux are always <576bytes so
 401                  * they should go through unfragmented).
 402                  */
 403                 if (sk->sk_state == TCP_LISTEN)
 404                         goto out;
 405
 406                 if (!ip6_sk_accept_pmtu(sk))
 407                         goto out;
 408
 409                 tp->mtu_info = ntohl(info);
 410                 if (!sock_owned_by_user(sk))
 411                         tcp_v6_mtu_reduced(sk);
 412                 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
 413                                            &tp->tsq_flags))
 414                         sock_hold(sk);
 415                 goto out;
 416         }
 417
 418         icmpv6_err_convert(type, code, &err);
 419
 420         /* Might be for an request_sock */
 421         switch (sk->sk_state) {
 422                 struct request_sock *req, **prev;
 423         case TCP_LISTEN:
 424                 if (sock_owned_by_user(sk))
 425                         goto out;
 426
 427                 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
 428                 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
 429                                            &hdr->saddr, inet6_iif(skb));
 430                 if (!req)
 431                         goto out;
 432
 433                 /* ICMPs are not backlogged, hence we cannot get
 434                  * an established socket here.
 435                  */
 436                 WARN_ON(req->sk != NULL);
 437
 438                 if (seq != tcp_rsk(req)->snt_isn) {
 439                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 440                         goto out;
 441                 }
 442
 443                 inet_csk_reqsk_queue_drop(sk, req, prev);
 444                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 445                 goto out;
 446
 447         case TCP_SYN_SENT:
 448         case TCP_SYN_RECV:
 449                 /* Only in fast or simultaneous open. If a fast open socket is
 450                  * is already accepted it is treated as a connected one below.
 451                  */
 452                 if (fastopen && fastopen->sk == NULL)
 453                         break;
 454
 455                 if (!sock_owned_by_user(sk)) {
 456                         sk->sk_err = err;
 457                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 458
 459                         tcp_done(sk);
 460                 } else
 461                         sk->sk_err_soft = err;
 462                 goto out;
 463         }
 464
 465         if (!sock_owned_by_user(sk) && np->recverr) {
 466                 sk->sk_err = err;
 467                 sk->sk_error_report(sk);
 468         } else
 469                 sk->sk_err_soft = err;
 470
 471 out:
 472         bh_unlock_sock(sk);
 473         sock_put(sk);
 474 }
 475
 476
 477 static int tcp_v6_send_synack(struct sock *sk, struct dst_entry *dst,
 478                               struct flowi *fl,
 479                               struct request_sock *req,
 480                               u16 queue_mapping,
 481                               struct tcp_fastopen_cookie *foc)
 482 {
 483         struct inet_request_sock *ireq = inet_rsk(req);
 484         struct ipv6_pinfo *np = inet6_sk(sk);
 485         struct flowi6 *fl6 = &fl->u.ip6;
 486         struct sk_buff *skb;
 487         int err = -ENOMEM;
 488
 489         /* First, grab a route. */
 490         if (!dst && (dst = inet6_csk_route_req(sk, fl6, req)) == NULL)
 491                 goto done;
 492
 493         skb = tcp_make_synack(sk, dst, req, foc);
 494
 495         if (skb) {
 496                 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
 497                                     &ireq->ir_v6_rmt_addr);
 498
 499                 fl6->daddr = ireq->ir_v6_rmt_addr;
 500                 if (np->repflow && (ireq->pktopts != NULL))
 501                         fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
 502
 503                 skb_set_queue_mapping(skb, queue_mapping);
 504                 err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass);
 505                 err = net_xmit_eval(err);
 506         }
 507
 508 done:
 509         return err;
 510 }
 511
 512
 513 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 514 {
 515         kfree_skb(inet_rsk(req)->pktopts);
 516 }
 517
 518 #ifdef CONFIG_TCP_MD5SIG
 519 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
 520                                                    const struct in6_addr *addr)
 521 {
 522         return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
 523 }
 524
 525 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
 526                                                 struct sock *addr_sk)
 527 {
 528         return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
 529 }
 530
 531 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
 532                                                       struct request_sock *req)
 533 {
 534         return tcp_v6_md5_do_lookup(sk, &inet_rsk(req)->ir_v6_rmt_addr);
 535 }
 536
 537 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
 538                                  int optlen)
 539 {
 540         struct tcp_md5sig cmd;
 541         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 542
 543         if (optlen < sizeof(cmd))
 544                 return -EINVAL;
 545
 546         if (copy_from_user(&cmd, optval, sizeof(cmd)))
 547                 return -EFAULT;
 548
 549         if (sin6->sin6_family != AF_INET6)
 550                 return -EINVAL;
 551
 552         if (!cmd.tcpm_keylen) {
 553                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 554                         return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 555                                               AF_INET);
 556                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 557                                       AF_INET6);
 558         }
 559
 560         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 561                 return -EINVAL;
 562
 563         if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 564                 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 565                                       AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 566
 567         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 568                               AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 569 }
 570
 571 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
 572                                         const struct in6_addr *daddr,
 573                                         const struct in6_addr *saddr, int nbytes)
 574 {
 575         struct tcp6_pseudohdr *bp;
 576         struct scatterlist sg;
 577
 578         bp = &hp->md5_blk.ip6;
 579         /* 1. TCP pseudo-header (RFC2460) */
 580         bp->saddr = *saddr;
 581         bp->daddr = *daddr;
 582         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 583         bp->len = cpu_to_be32(nbytes);
 584
 585         sg_init_one(&sg, bp, sizeof(*bp));
 586         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
 587 }
 588
 589 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
 590                                const struct in6_addr *daddr, struct in6_addr *saddr,
 591                                const struct tcphdr *th)
 592 {
 593         struct tcp_md5sig_pool *hp;
 594         struct hash_desc *desc;
 595
 596         hp = tcp_get_md5sig_pool();
 597         if (!hp)
 598                 goto clear_hash_noput;
 599         desc = &hp->md5_desc;
 600
 601         if (crypto_hash_init(desc))
 602                 goto clear_hash;
 603         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
 604                 goto clear_hash;
 605         if (tcp_md5_hash_header(hp, th))
 606                 goto clear_hash;
 607         if (tcp_md5_hash_key(hp, key))
 608                 goto clear_hash;
 609         if (crypto_hash_final(desc, md5_hash))
 610                 goto clear_hash;
 611
 612         tcp_put_md5sig_pool();
 613         return 0;
 614
 615 clear_hash:
 616         tcp_put_md5sig_pool();
 617 clear_hash_noput:
 618         memset(md5_hash, 0, 16);
 619         return 1;
 620 }
 621
 622 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
 623                                const struct sock *sk,
 624                                const struct request_sock *req,
 625                                const struct sk_buff *skb)
 626 {
 627         const struct in6_addr *saddr, *daddr;
 628         struct tcp_md5sig_pool *hp;
 629         struct hash_desc *desc;
 630         const struct tcphdr *th = tcp_hdr(skb);
 631
 632         if (sk) {
 633                 saddr = &inet6_sk(sk)->saddr;
 634                 daddr = &sk->sk_v6_daddr;
 635         } else if (req) {
 636                 saddr = &inet_rsk(req)->ir_v6_loc_addr;
 637                 daddr = &inet_rsk(req)->ir_v6_rmt_addr;
 638         } else {
 639                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 640                 saddr = &ip6h->saddr;
 641                 daddr = &ip6h->daddr;
 642         }
 643
 644         hp = tcp_get_md5sig_pool();
 645         if (!hp)
 646                 goto clear_hash_noput;
 647         desc = &hp->md5_desc;
 648
 649         if (crypto_hash_init(desc))
 650                 goto clear_hash;
 651
 652         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
 653                 goto clear_hash;
 654         if (tcp_md5_hash_header(hp, th))
 655                 goto clear_hash;
 656         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 657                 goto clear_hash;
 658         if (tcp_md5_hash_key(hp, key))
 659                 goto clear_hash;
 660         if (crypto_hash_final(desc, md5_hash))
 661                 goto clear_hash;
 662
 663         tcp_put_md5sig_pool();
 664         return 0;
 665
 666 clear_hash:
 667         tcp_put_md5sig_pool();
 668 clear_hash_noput:
 669         memset(md5_hash, 0, 16);
 670         return 1;
 671 }
 672
 673 static int __tcp_v6_inbound_md5_hash(struct sock *sk,
 674                                      const struct sk_buff *skb)
 675 {
 676         const __u8 *hash_location = NULL;
 677         struct tcp_md5sig_key *hash_expected;
 678         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 679         const struct tcphdr *th = tcp_hdr(skb);
 680         int genhash;
 681         u8 newhash[16];
 682
 683         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 684         hash_location = tcp_parse_md5sig_option(th);
 685
 686         /* We've parsed the options - do we have a hash? */
 687         if (!hash_expected && !hash_location)
 688                 return 0;
 689
 690         if (hash_expected && !hash_location) {
 691                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 692                 return 1;
 693         }
 694
 695         if (!hash_expected && hash_location) {
 696                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 697                 return 1;
 698         }
 699
 700         /* check the signature */
 701         genhash = tcp_v6_md5_hash_skb(newhash,
 702                                       hash_expected,
 703                                       NULL, NULL, skb);
 704
 705         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 706                 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
 707                                      genhash ? "failed" : "mismatch",
 708                                      &ip6h->saddr, ntohs(th->source),
 709                                      &ip6h->daddr, ntohs(th->dest));
 710                 return 1;
 711         }
 712         return 0;
 713 }
 714
 715 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
 716 {
 717         int ret;
 718
 719         rcu_read_lock();
 720         ret = __tcp_v6_inbound_md5_hash(sk, skb);
 721         rcu_read_unlock();
 722
 723         return ret;
 724 }
 725
 726 #endif
 727
 728 static void tcp_v6_init_req(struct request_sock *req, struct sock *sk,
 729                             struct sk_buff *skb)
 730 {
 731         struct inet_request_sock *ireq = inet_rsk(req);
 732         struct ipv6_pinfo *np = inet6_sk(sk);
 733
 734         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 735         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 736
 737         ireq->ir_iif = sk->sk_bound_dev_if;
 738
 739         /* So that link locals have meaning */
 740         if (!sk->sk_bound_dev_if &&
 741             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 742                 ireq->ir_iif = tcp_v6_iif(skb);
 743
 744         if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
 745             (ipv6_opt_accepted(sk, skb, &TCP_SKB_CB(skb)->header.h6) ||
 746              np->rxopt.bits.rxinfo ||
 747              np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
 748              np->rxopt.bits.rxohlim || np->repflow)) {
 749                 atomic_inc(&skb->users);
 750                 ireq->pktopts = skb;
 751         }
 752 }
 753
 754 static struct dst_entry *tcp_v6_route_req(struct sock *sk, struct flowi *fl,
 755                                           const struct request_sock *req,
 756                                           bool *strict)
 757 {
 758         if (strict)
 759                 *strict = true;
 760         return inet6_csk_route_req(sk, &fl->u.ip6, req);
 761 }
 762
 763 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 764         .family         =       AF_INET6,
 765         .obj_size       =       sizeof(struct tcp6_request_sock),
 766         .rtx_syn_ack    =       tcp_rtx_synack,
 767         .send_ack       =       tcp_v6_reqsk_send_ack,
 768         .destructor     =       tcp_v6_reqsk_destructor,
 769         .send_reset     =       tcp_v6_send_reset,
 770         .syn_ack_timeout =      tcp_syn_ack_timeout,
 771 };
 772
 773 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 774         .mss_clamp      =       IPV6_MIN_MTU - sizeof(struct tcphdr) -
 775                                 sizeof(struct ipv6hdr),
 776 #ifdef CONFIG_TCP_MD5SIG
 777         .md5_lookup     =       tcp_v6_reqsk_md5_lookup,
 778         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 779 #endif
 780         .init_req       =       tcp_v6_init_req,
 781 #ifdef CONFIG_SYN_COOKIES
 782         .cookie_init_seq =      cookie_v6_init_sequence,
 783 #endif
 784         .route_req      =       tcp_v6_route_req,
 785         .init_seq       =       tcp_v6_init_sequence,
 786         .send_synack    =       tcp_v6_send_synack,
 787         .queue_hash_add =       inet6_csk_reqsk_queue_hash_add,
 788 };
 789
 790 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
 791                                  u32 tsval, u32 tsecr, int oif,
 792                                  struct tcp_md5sig_key *key, int rst, u8 tclass,
 793                                  u32 label)
 794 {
 795         const struct tcphdr *th = tcp_hdr(skb);
 796         struct tcphdr *t1;
 797         struct sk_buff *buff;
 798         struct flowi6 fl6;
 799         struct net *net = dev_net(skb_dst(skb)->dev);
 800         struct sock *ctl_sk = net->ipv6.tcp_sk;
 801         unsigned int tot_len = sizeof(struct tcphdr);
 802         struct dst_entry *dst;
 803         __be32 *topt;
 804
 805         if (tsecr)
 806                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 807 #ifdef CONFIG_TCP_MD5SIG
 808         if (key)
 809                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 810 #endif
 811
 812         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 813                          GFP_ATOMIC);
 814         if (buff == NULL)
 815                 return;
 816
 817         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 818
 819         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 820         skb_reset_transport_header(buff);
 821
 822         /* Swap the send and the receive. */
 823         memset(t1, 0, sizeof(*t1));
 824         t1->dest = th->source;
 825         t1->source = th->dest;
 826         t1->doff = tot_len / 4;
 827         t1->seq = htonl(seq);
 828         t1->ack_seq = htonl(ack);
 829         t1->ack = !rst || !th->ack;
 830         t1->rst = rst;
 831         t1->window = htons(win);
 832
 833         topt = (__be32 *)(t1 + 1);
 834
 835         if (tsecr) {
 836                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 837                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 838                 *topt++ = htonl(tsval);
 839                 *topt++ = htonl(tsecr);
 840         }
 841
 842 #ifdef CONFIG_TCP_MD5SIG
 843         if (key) {
 844                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 845                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 846                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 847                                     &ipv6_hdr(skb)->saddr,
 848                                     &ipv6_hdr(skb)->daddr, t1);
 849         }
 850 #endif
 851
 852         memset(&fl6, 0, sizeof(fl6));
 853         fl6.daddr = ipv6_hdr(skb)->saddr;
 854         fl6.saddr = ipv6_hdr(skb)->daddr;
 855         fl6.flowlabel = label;
 856
 857         buff->ip_summed = CHECKSUM_PARTIAL;
 858         buff->csum = 0;
 859
 860         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
 861
 862         fl6.flowi6_proto = IPPROTO_TCP;
 863         if (rt6_need_strict(&fl6.daddr) && !oif)
 864                 fl6.flowi6_oif = tcp_v6_iif(skb);
 865         else
 866                 fl6.flowi6_oif = oif;
 867         fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
 868         fl6.fl6_dport = t1->dest;
 869         fl6.fl6_sport = t1->source;
 870         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 871
 872         /* Pass a socket to ip6_dst_lookup either it is for RST
 873          * Underlying function will use this to retrieve the network
 874          * namespace
 875          */
 876         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 877         if (!IS_ERR(dst)) {
 878                 skb_dst_set(buff, dst);
 879                 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
 880                 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 881                 if (rst)
 882                         TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
 883                 return;
 884         }
 885
 886         kfree_skb(buff);
 887 }
 888
 889 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
 890 {
 891         const struct tcphdr *th = tcp_hdr(skb);
 892         u32 seq = 0, ack_seq = 0;
 893         struct tcp_md5sig_key *key = NULL;
 894 #ifdef CONFIG_TCP_MD5SIG
 895         const __u8 *hash_location = NULL;
 896         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 897         unsigned char newhash[16];
 898         int genhash;
 899         struct sock *sk1 = NULL;
 900 #endif
 901         int oif;
 902
 903         if (th->rst)
 904                 return;
 905
 906         if (!ipv6_unicast_destination(skb))
 907                 return;
 908
 909 #ifdef CONFIG_TCP_MD5SIG
 910         hash_location = tcp_parse_md5sig_option(th);
 911         if (!sk && hash_location) {
 912                 /*
 913                  * active side is lost. Try to find listening socket through
 914                  * source port, and then find md5 key through listening socket.
 915                  * we are not loose security here:
 916                  * Incoming packet is checked with md5 hash with finding key,
 917                  * no RST generated if md5 hash doesn't match.
 918                  */
 919                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
 920                                            &tcp_hashinfo, &ipv6h->saddr,
 921                                            th->source, &ipv6h->daddr,
 922                                            ntohs(th->source), tcp_v6_iif(skb));
 923                 if (!sk1)
 924                         return;
 925
 926                 rcu_read_lock();
 927                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 928                 if (!key)
 929                         goto release_sk1;
 930
 931                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
 932                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
 933                         goto release_sk1;
 934         } else {
 935                 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
 936         }
 937 #endif
 938
 939         if (th->ack)
 940                 seq = ntohl(th->ack_seq);
 941         else
 942                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
 943                           (th->doff << 2);
 944
 945         oif = sk ? sk->sk_bound_dev_if : 0;
 946         tcp_v6_send_response(skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
 947
 948 #ifdef CONFIG_TCP_MD5SIG
 949 release_sk1:
 950         if (sk1) {
 951                 rcu_read_unlock();
 952                 sock_put(sk1);
 953         }
 954 #endif
 955 }
 956
 957 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
 958                             u32 win, u32 tsval, u32 tsecr, int oif,
 959                             struct tcp_md5sig_key *key, u8 tclass,
 960                             u32 label)
 961 {
 962         tcp_v6_send_response(skb, seq, ack, win, tsval, tsecr, oif, key, 0, tclass,
 963                              label);
 964 }
 965
 966 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 967 {
 968         struct inet_timewait_sock *tw = inet_twsk(sk);
 969         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 970
 971         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 972                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 973                         tcp_time_stamp + tcptw->tw_ts_offset,
 974                         tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
 975                         tw->tw_tclass, (tw->tw_flowlabel << 12));
 976
 977         inet_twsk_put(tw);
 978 }
 979
 980 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
 981                                   struct request_sock *req)
 982 {
 983         /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 984          * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 985          */
 986         tcp_v6_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
 987                         tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
 988                         tcp_rsk(req)->rcv_nxt,
 989                         req->rcv_wnd, tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
 990                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
 991                         0, 0);
 992 }
 993
 994
 995 static struct sock *tcp_v6_hnd_req(struct sock *sk, struct sk_buff *skb)
 996 {
 997         struct request_sock *req, **prev;
 998         const struct tcphdr *th = tcp_hdr(skb);
 999         struct sock *nsk;
1000
1001         /* Find possible connection requests. */
1002         req = inet6_csk_search_req(sk, &prev, th->source,
1003                                    &ipv6_hdr(skb)->saddr,
1004                                    &ipv6_hdr(skb)->daddr, tcp_v6_iif(skb));
1005         if (req)
1006                 return tcp_check_req(sk, skb, req, prev, false);
1007
1008         nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1009                                          &ipv6_hdr(skb)->saddr, th->source,
1010                                          &ipv6_hdr(skb)->daddr, ntohs(th->dest),
1011                                          tcp_v6_iif(skb));
1012
1013         if (nsk) {
1014                 if (nsk->sk_state != TCP_TIME_WAIT) {
1015                         bh_lock_sock(nsk);
1016                         return nsk;
1017                 }
1018                 inet_twsk_put(inet_twsk(nsk));
1019                 return NULL;
1020         }
1021
1022 #ifdef CONFIG_SYN_COOKIES
1023         if (!th->syn)
1024                 sk = cookie_v6_check(sk, skb);
1025 #endif
1026         return sk;
1027 }
1028
1029 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1030 {
1031         if (skb->protocol == htons(ETH_P_IP))
1032                 return tcp_v4_conn_request(sk, skb);
1033
1034         if (!ipv6_unicast_destination(skb))
1035                 goto drop;
1036
1037         return tcp_conn_request(&tcp6_request_sock_ops,
1038                                 &tcp_request_sock_ipv6_ops, sk, skb);
1039
1040 drop:
1041         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1042         return 0; /* don't send reset */
1043 }
1044
1045 static struct sock *tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1046                                          struct request_sock *req,
1047                                          struct dst_entry *dst)
1048 {
1049         struct inet_request_sock *ireq;
1050         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1051         struct tcp6_sock *newtcp6sk;
1052         struct inet_sock *newinet;
1053         struct tcp_sock *newtp;
1054         struct sock *newsk;
1055 #ifdef CONFIG_TCP_MD5SIG
1056         struct tcp_md5sig_key *key;
1057 #endif
1058         struct flowi6 fl6;
1059
1060         if (skb->protocol == htons(ETH_P_IP)) {
1061                 /*
1062                  *      v6 mapped
1063                  */
1064
1065                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1066
1067                 if (newsk == NULL)
1068                         return NULL;
1069
1070                 newtcp6sk = (struct tcp6_sock *)newsk;
1071                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1072
1073                 newinet = inet_sk(newsk);
1074                 newnp = inet6_sk(newsk);
1075                 newtp = tcp_sk(newsk);
1076
1077                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1078
1079                 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newsk->sk_v6_daddr);
1080
1081                 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1082
1083                 newsk->sk_v6_rcv_saddr = newnp->saddr;
1084
1085                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1086                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1087 #ifdef CONFIG_TCP_MD5SIG
1088                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1089 #endif
1090
1091                 newnp->ipv6_ac_list = NULL;
1092                 newnp->ipv6_fl_list = NULL;
1093                 newnp->pktoptions  = NULL;
1094                 newnp->opt         = NULL;
1095                 newnp->mcast_oif   = tcp_v6_iif(skb);
1096                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1097                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1098                 if (np->repflow)
1099                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1100
1101                 /*
1102                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1103                  * here, tcp_create_openreq_child now does this for us, see the comment in
1104                  * that function for the gory details. -acme
1105                  */
1106
1107                 /* It is tricky place. Until this moment IPv4 tcp
1108                    worked with IPv6 icsk.icsk_af_ops.
1109                    Sync it now.
1110                  */
1111                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1112
1113                 return newsk;
1114         }
1115
1116         ireq = inet_rsk(req);
1117
1118         if (sk_acceptq_is_full(sk))
1119                 goto out_overflow;
1120
1121         if (!dst) {
1122                 dst = inet6_csk_route_req(sk, &fl6, req);
1123                 if (!dst)
1124                         goto out;
1125         }
1126
1127         newsk = tcp_create_openreq_child(sk, req, skb);
1128         if (newsk == NULL)
1129                 goto out_nonewsk;
1130
1131         /*
1132          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1133          * count here, tcp_create_openreq_child now does this for us, see the
1134          * comment in that function for the gory details. -acme
1135          */
1136
1137         newsk->sk_gso_type = SKB_GSO_TCPV6;
1138         __ip6_dst_store(newsk, dst, NULL, NULL);
1139         inet6_sk_rx_dst_set(newsk, skb);
1140
1141         newtcp6sk = (struct tcp6_sock *)newsk;
1142         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1143
1144         newtp = tcp_sk(newsk);
1145         newinet = inet_sk(newsk);
1146         newnp = inet6_sk(newsk);
1147
1148         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1149
1150         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1151         newnp->saddr = ireq->ir_v6_loc_addr;
1152         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1153         newsk->sk_bound_dev_if = ireq->ir_iif;
1154
1155         ip6_set_txhash(newsk);
1156
1157         /* Now IPv6 options...
1158
1159            First: no IPv4 options.
1160          */
1161         newinet->inet_opt = NULL;
1162         newnp->ipv6_ac_list = NULL;
1163         newnp->ipv6_fl_list = NULL;
1164
1165         /* Clone RX bits */
1166         newnp->rxopt.all = np->rxopt.all;
1167
1168         /* Clone pktoptions received with SYN */
1169         newnp->pktoptions = NULL;
1170         if (ireq->pktopts != NULL) {
1171                 newnp->pktoptions = skb_clone(ireq->pktopts,
1172                                               sk_gfp_atomic(sk, GFP_ATOMIC));
1173                 consume_skb(ireq->pktopts);
1174                 ireq->pktopts = NULL;
1175                 if (newnp->pktoptions)
1176                         skb_set_owner_r(newnp->pktoptions, newsk);
1177         }
1178         newnp->opt        = NULL;
1179         newnp->mcast_oif  = tcp_v6_iif(skb);
1180         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1181         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1182         if (np->repflow)
1183                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1184
1185         /* Clone native IPv6 options from listening socket (if any)
1186
1187            Yes, keeping reference count would be much more clever,
1188            but we make one more one thing there: reattach optmem
1189            to newsk.
1190          */
1191         if (np->opt)
1192                 newnp->opt = ipv6_dup_options(newsk, np->opt);
1193
1194         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1195         if (newnp->opt)
1196                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1197                                                      newnp->opt->opt_flen);
1198
1199         tcp_sync_mss(newsk, dst_mtu(dst));
1200         newtp->advmss = dst_metric_advmss(dst);
1201         if (tcp_sk(sk)->rx_opt.user_mss &&
1202             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1203                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1204
1205         tcp_initialize_rcv_mss(newsk);
1206
1207         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1208         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1209
1210 #ifdef CONFIG_TCP_MD5SIG
1211         /* Copy over the MD5 key from the original socket */
1212         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1213         if (key != NULL) {
1214                 /* We're using one, so create a matching key
1215                  * on the newsk structure. If we fail to get
1216                  * memory, then we end up not copying the key
1217                  * across. Shucks.
1218                  */
1219                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1220                                AF_INET6, key->key, key->keylen,
1221                                sk_gfp_atomic(sk, GFP_ATOMIC));
1222         }
1223 #endif
1224
1225         if (__inet_inherit_port(sk, newsk) < 0) {
1226                 inet_csk_prepare_forced_close(newsk);
1227                 tcp_done(newsk);
1228                 goto out;
1229         }
1230         __inet6_hash(newsk, NULL);
1231
1232         return newsk;
1233
1234 out_overflow:
1235         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1236 out_nonewsk:
1237         dst_release(dst);
1238 out:
1239         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1240         return NULL;
1241 }
1242
1243 /* The socket must have it's spinlock held when we get
1244  * here.
1245  *
1246  * We have a potential double-lock case here, so even when
1247  * doing backlog processing we use the BH locking scheme.
1248  * This is because we cannot sleep with the original spinlock
1249  * held.
1250  */
1251 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1252 {
1253         struct ipv6_pinfo *np = inet6_sk(sk);
1254         struct tcp_sock *tp;
1255         struct sk_buff *opt_skb = NULL;
1256
1257         /* Imagine: socket is IPv6. IPv4 packet arrives,
1258            goes to IPv4 receive handler and backlogged.
1259            From backlog it always goes here. Kerboom...
1260            Fortunately, tcp_rcv_established and rcv_established
1261            handle them correctly, but it is not case with
1262            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1263          */
1264
1265         if (skb->protocol == htons(ETH_P_IP))
1266                 return tcp_v4_do_rcv(sk, skb);
1267
1268         if (sk_filter(sk, skb))
1269                 goto discard;
1270
1271         /*
1272          *      socket locking is here for SMP purposes as backlog rcv
1273          *      is currently called with bh processing disabled.
1274          */
1275
1276         /* Do Stevens' IPV6_PKTOPTIONS.
1277
1278            Yes, guys, it is the only place in our code, where we
1279            may make it not affecting IPv4.
1280            The rest of code is protocol independent,
1281            and I do not like idea to uglify IPv4.
1282
1283            Actually, all the idea behind IPV6_PKTOPTIONS
1284            looks not very well thought. For now we latch
1285            options, received in the last packet, enqueued
1286            by tcp. Feel free to propose better solution.
1287                                                --ANK (980728)
1288          */
1289         if (np->rxopt.all)
1290                 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1291
1292         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1293                 struct dst_entry *dst = sk->sk_rx_dst;
1294
1295                 sock_rps_save_rxhash(sk, skb);
1296                 if (dst) {
1297                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1298                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1299                                 dst_release(dst);
1300                                 sk->sk_rx_dst = NULL;
1301                         }
1302                 }
1303
1304                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1305                 if (opt_skb)
1306                         goto ipv6_pktoptions;
1307                 return 0;
1308         }
1309
1310         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1311                 goto csum_err;
1312
1313         if (sk->sk_state == TCP_LISTEN) {
1314                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1315                 if (!nsk)
1316                         goto discard;
1317
1318                 /*
1319                  * Queue it on the new socket if the new socket is active,
1320                  * otherwise we just shortcircuit this and continue with
1321                  * the new socket..
1322                  */
1323                 if (nsk != sk) {
1324                         sock_rps_save_rxhash(nsk, skb);
1325                         if (tcp_child_process(sk, nsk, skb))
1326                                 goto reset;
1327                         if (opt_skb)
1328                                 __kfree_skb(opt_skb);
1329                         return 0;
1330                 }
1331         } else
1332                 sock_rps_save_rxhash(sk, skb);
1333
1334         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1335                 goto reset;
1336         if (opt_skb)
1337                 goto ipv6_pktoptions;
1338         return 0;
1339
1340 reset:
1341         tcp_v6_send_reset(sk, skb);
1342 discard:
1343         if (opt_skb)
1344                 __kfree_skb(opt_skb);
1345         kfree_skb(skb);
1346         return 0;
1347 csum_err:
1348         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1349         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1350         goto discard;
1351
1352
1353 ipv6_pktoptions:
1354         /* Do you ask, what is it?
1355
1356            1. skb was enqueued by tcp.
1357            2. skb is added to tail of read queue, rather than out of order.
1358            3. socket is not in passive state.
1359            4. Finally, it really contains options, which user wants to receive.
1360          */
1361         tp = tcp_sk(sk);
1362         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1363             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1364                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1365                         np->mcast_oif = tcp_v6_iif(opt_skb);
1366                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1367                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1368                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1369                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1370                 if (np->repflow)
1371                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1372                 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1373                         skb_set_owner_r(opt_skb, sk);
1374                         opt_skb = xchg(&np->pktoptions, opt_skb);
1375                 } else {
1376                         __kfree_skb(opt_skb);
1377                         opt_skb = xchg(&np->pktoptions, NULL);
1378                 }
1379         }
1380
1381         kfree_skb(opt_skb);
1382         return 0;
1383 }
1384
1385 static int tcp_v6_rcv(struct sk_buff *skb)
1386 {
1387         const struct tcphdr *th;
1388         const struct ipv6hdr *hdr;
1389         struct sock *sk;
1390         int ret;
1391         struct net *net = dev_net(skb->dev);
1392
1393         if (skb->pkt_type != PACKET_HOST)
1394                 goto discard_it;
1395
1396         /*
1397          *      Count it even if it's bad.
1398          */
1399         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1400
1401         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1402                 goto discard_it;
1403
1404         th = tcp_hdr(skb);
1405
1406         if (th->doff < sizeof(struct tcphdr)/4)
1407                 goto bad_packet;
1408         if (!pskb_may_pull(skb, th->doff*4))
1409                 goto discard_it;
1410
1411         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1412                 goto csum_error;
1413
1414         th = tcp_hdr(skb);
1415         hdr = ipv6_hdr(skb);
1416         /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1417          * barrier() makes sure compiler wont play fool^Waliasing games.
1418          */
1419         memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1420                 sizeof(struct inet6_skb_parm));
1421         barrier();
1422
1423         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1424         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1425                                     skb->len - th->doff*4);
1426         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1427         TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1428         TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1429         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1430         TCP_SKB_CB(skb)->sacked = 0;
1431
1432         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest,
1433                                 tcp_v6_iif(skb));
1434         if (!sk)
1435                 goto no_tcp_socket;
1436
1437 process:
1438         if (sk->sk_state == TCP_TIME_WAIT)
1439                 goto do_time_wait;
1440
1441         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1442                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1443                 goto discard_and_relse;
1444         }
1445
1446         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1447                 goto discard_and_relse;
1448
1449 #ifdef CONFIG_TCP_MD5SIG
1450         if (tcp_v6_inbound_md5_hash(sk, skb))
1451                 goto discard_and_relse;
1452 #endif
1453
1454         if (sk_filter(sk, skb))
1455                 goto discard_and_relse;
1456
1457         sk_mark_napi_id(sk, skb);
1458         skb->dev = NULL;
1459
1460         bh_lock_sock_nested(sk);
1461         ret = 0;
1462         if (!sock_owned_by_user(sk)) {
1463                 if (!tcp_prequeue(sk, skb))
1464                         ret = tcp_v6_do_rcv(sk, skb);
1465         } else if (unlikely(sk_add_backlog(sk, skb,
1466                                            sk->sk_rcvbuf + sk->sk_sndbuf))) {
1467                 bh_unlock_sock(sk);
1468                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1469                 goto discard_and_relse;
1470         }
1471         bh_unlock_sock(sk);
1472
1473         sock_put(sk);
1474         return ret ? -1 : 0;
1475
1476 no_tcp_socket:
1477         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1478                 goto discard_it;
1479
1480         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1481 csum_error:
1482                 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1483 bad_packet:
1484                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1485         } else {
1486                 tcp_v6_send_reset(NULL, skb);
1487         }
1488
1489 discard_it:
1490         kfree_skb(skb);
1491         return 0;
1492
1493 discard_and_relse:
1494         sock_put(sk);
1495         goto discard_it;
1496
1497 do_time_wait:
1498         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1499                 inet_twsk_put(inet_twsk(sk));
1500                 goto discard_it;
1501         }
1502
1503         if (skb->len < (th->doff<<2)) {
1504                 inet_twsk_put(inet_twsk(sk));
1505                 goto bad_packet;
1506         }
1507         if (tcp_checksum_complete(skb)) {
1508                 inet_twsk_put(inet_twsk(sk));
1509                 goto csum_error;
1510         }
1511
1512         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1513         case TCP_TW_SYN:
1514         {
1515                 struct sock *sk2;
1516
1517                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1518                                             &ipv6_hdr(skb)->saddr, th->source,
1519                                             &ipv6_hdr(skb)->daddr,
1520                                             ntohs(th->dest), tcp_v6_iif(skb));
1521                 if (sk2 != NULL) {
1522                         struct inet_timewait_sock *tw = inet_twsk(sk);
1523                         inet_twsk_deschedule(tw, &tcp_death_row);
1524                         inet_twsk_put(tw);
1525                         sk = sk2;
1526                         goto process;
1527                 }
1528                 /* Fall through to ACK */
1529         }
1530         case TCP_TW_ACK:
1531                 tcp_v6_timewait_ack(sk, skb);
1532                 break;
1533         case TCP_TW_RST:
1534                 goto no_tcp_socket;
1535         case TCP_TW_SUCCESS:
1536                 ;
1537         }
1538         goto discard_it;
1539 }
1540
1541 static void tcp_v6_early_demux(struct sk_buff *skb)
1542 {
1543         const struct ipv6hdr *hdr;
1544         const struct tcphdr *th;
1545         struct sock *sk;
1546
1547         if (skb->pkt_type != PACKET_HOST)
1548                 return;
1549
1550         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1551                 return;
1552
1553         hdr = ipv6_hdr(skb);
1554         th = tcp_hdr(skb);
1555
1556         if (th->doff < sizeof(struct tcphdr) / 4)
1557                 return;
1558
1559         /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1560         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1561                                         &hdr->saddr, th->source,
1562                                         &hdr->daddr, ntohs(th->dest),
1563                                         inet6_iif(skb));
1564         if (sk) {
1565                 skb->sk = sk;
1566                 skb->destructor = sock_edemux;
1567                 if (sk->sk_state != TCP_TIME_WAIT) {
1568                         struct dst_entry *dst = sk->sk_rx_dst;
1569
1570                         if (dst)
1571                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1572                         if (dst &&
1573                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1574                                 skb_dst_set_noref(skb, dst);
1575                 }
1576         }
1577 }
1578
1579 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1580         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1581         .twsk_unique    = tcp_twsk_unique,
1582         .twsk_destructor = tcp_twsk_destructor,
1583 };
1584
1585 static const struct inet_connection_sock_af_ops ipv6_specific = {
1586         .queue_xmit        = inet6_csk_xmit,
1587         .send_check        = tcp_v6_send_check,
1588         .rebuild_header    = inet6_sk_rebuild_header,
1589         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1590         .conn_request      = tcp_v6_conn_request,
1591         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1592         .net_header_len    = sizeof(struct ipv6hdr),
1593         .net_frag_header_len = sizeof(struct frag_hdr),
1594         .setsockopt        = ipv6_setsockopt,
1595         .getsockopt        = ipv6_getsockopt,
1596         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1597         .sockaddr_len      = sizeof(struct sockaddr_in6),
1598         .bind_conflict     = inet6_csk_bind_conflict,
1599 #ifdef CONFIG_COMPAT
1600         .compat_setsockopt = compat_ipv6_setsockopt,
1601         .compat_getsockopt = compat_ipv6_getsockopt,
1602 #endif
1603         .mtu_reduced       = tcp_v6_mtu_reduced,
1604 };
1605
1606 #ifdef CONFIG_TCP_MD5SIG
1607 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1608         .md5_lookup     =       tcp_v6_md5_lookup,
1609         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1610         .md5_parse      =       tcp_v6_parse_md5_keys,
1611 };
1612 #endif
1613
1614 /*
1615  *      TCP over IPv4 via INET6 API
1616  */
1617 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1618         .queue_xmit        = ip_queue_xmit,
1619         .send_check        = tcp_v4_send_check,
1620         .rebuild_header    = inet_sk_rebuild_header,
1621         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1622         .conn_request      = tcp_v6_conn_request,
1623         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1624         .net_header_len    = sizeof(struct iphdr),
1625         .setsockopt        = ipv6_setsockopt,
1626         .getsockopt        = ipv6_getsockopt,
1627         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1628         .sockaddr_len      = sizeof(struct sockaddr_in6),
1629         .bind_conflict     = inet6_csk_bind_conflict,
1630 #ifdef CONFIG_COMPAT
1631         .compat_setsockopt = compat_ipv6_setsockopt,
1632         .compat_getsockopt = compat_ipv6_getsockopt,
1633 #endif
1634         .mtu_reduced       = tcp_v4_mtu_reduced,
1635 };
1636
1637 #ifdef CONFIG_TCP_MD5SIG
1638 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1639         .md5_lookup     =       tcp_v4_md5_lookup,
1640         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1641         .md5_parse      =       tcp_v6_parse_md5_keys,
1642 };
1643 #endif
1644
1645 /* NOTE: A lot of things set to zero explicitly by call to
1646  *       sk_alloc() so need not be done here.
1647  */
1648 static int tcp_v6_init_sock(struct sock *sk)
1649 {
1650         struct inet_connection_sock *icsk = inet_csk(sk);
1651
1652         tcp_init_sock(sk);
1653
1654         icsk->icsk_af_ops = &ipv6_specific;
1655
1656 #ifdef CONFIG_TCP_MD5SIG
1657         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1658 #endif
1659
1660         return 0;
1661 }
1662
1663 static void tcp_v6_destroy_sock(struct sock *sk)
1664 {
1665         tcp_v4_destroy_sock(sk);
1666         inet6_destroy_sock(sk);
1667 }
1668
1669 #ifdef CONFIG_PROC_FS
1670 /* Proc filesystem TCPv6 sock list dumping. */
1671 static void get_openreq6(struct seq_file *seq,
1672                          const struct sock *sk, struct request_sock *req, int i, kuid_t uid)
1673 {
1674         int ttd = req->expires - jiffies;
1675         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1676         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1677
1678         if (ttd < 0)
1679                 ttd = 0;
1680
1681         seq_printf(seq,
1682                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1683                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1684                    i,
1685                    src->s6_addr32[0], src->s6_addr32[1],
1686                    src->s6_addr32[2], src->s6_addr32[3],
1687                    inet_rsk(req)->ir_num,
1688                    dest->s6_addr32[0], dest->s6_addr32[1],
1689                    dest->s6_addr32[2], dest->s6_addr32[3],
1690                    ntohs(inet_rsk(req)->ir_rmt_port),
1691                    TCP_SYN_RECV,
1692                    0, 0, /* could print option size, but that is af dependent. */
1693                    1,   /* timers active (only the expire timer) */
1694                    jiffies_to_clock_t(ttd),
1695                    req->num_timeout,
1696                    from_kuid_munged(seq_user_ns(seq), uid),
1697                    0,  /* non standard timer */
1698                    0, /* open_requests have no inode */
1699                    0, req);
1700 }
1701
1702 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1703 {
1704         const struct in6_addr *dest, *src;
1705         __u16 destp, srcp;
1706         int timer_active;
1707         unsigned long timer_expires;
1708         const struct inet_sock *inet = inet_sk(sp);
1709         const struct tcp_sock *tp = tcp_sk(sp);
1710         const struct inet_connection_sock *icsk = inet_csk(sp);
1711         struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
1712
1713         dest  = &sp->sk_v6_daddr;
1714         src   = &sp->sk_v6_rcv_saddr;
1715         destp = ntohs(inet->inet_dport);
1716         srcp  = ntohs(inet->inet_sport);
1717
1718         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1719                 timer_active    = 1;
1720                 timer_expires   = icsk->icsk_timeout;
1721         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1722                 timer_active    = 4;
1723                 timer_expires   = icsk->icsk_timeout;
1724         } else if (timer_pending(&sp->sk_timer)) {
1725                 timer_active    = 2;
1726                 timer_expires   = sp->sk_timer.expires;
1727         } else {
1728                 timer_active    = 0;
1729                 timer_expires = jiffies;
1730         }
1731
1732         seq_printf(seq,
1733                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1734                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1735                    i,
1736                    src->s6_addr32[0], src->s6_addr32[1],
1737                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1738                    dest->s6_addr32[0], dest->s6_addr32[1],
1739                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1740                    sp->sk_state,
1741                    tp->write_seq-tp->snd_una,
1742                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1743                    timer_active,
1744                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1745                    icsk->icsk_retransmits,
1746                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1747                    icsk->icsk_probes_out,
1748                    sock_i_ino(sp),
1749                    atomic_read(&sp->sk_refcnt), sp,
1750                    jiffies_to_clock_t(icsk->icsk_rto),
1751                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1752                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1753                    tp->snd_cwnd,
1754                    sp->sk_state == TCP_LISTEN ?
1755                         (fastopenq ? fastopenq->max_qlen : 0) :
1756                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1757                    );
1758 }
1759
1760 static void get_timewait6_sock(struct seq_file *seq,
1761                                struct inet_timewait_sock *tw, int i)
1762 {
1763         const struct in6_addr *dest, *src;
1764         __u16 destp, srcp;
1765         s32 delta = tw->tw_ttd - inet_tw_time_stamp();
1766
1767         dest = &tw->tw_v6_daddr;
1768         src  = &tw->tw_v6_rcv_saddr;
1769         destp = ntohs(tw->tw_dport);
1770         srcp  = ntohs(tw->tw_sport);
1771
1772         seq_printf(seq,
1773                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1774                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1775                    i,
1776                    src->s6_addr32[0], src->s6_addr32[1],
1777                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1778                    dest->s6_addr32[0], dest->s6_addr32[1],
1779                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1780                    tw->tw_substate, 0, 0,
1781                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1782                    atomic_read(&tw->tw_refcnt), tw);
1783 }
1784
1785 static int tcp6_seq_show(struct seq_file *seq, void *v)
1786 {
1787         struct tcp_iter_state *st;
1788         struct sock *sk = v;
1789
1790         if (v == SEQ_START_TOKEN) {
1791                 seq_puts(seq,
1792                          "  sl  "
1793                          "local_address                         "
1794                          "remote_address                        "
1795                          "st tx_queue rx_queue tr tm->when retrnsmt"
1796                          "   uid  timeout inode\n");
1797                 goto out;
1798         }
1799         st = seq->private;
1800
1801         switch (st->state) {
1802         case TCP_SEQ_STATE_LISTENING:
1803         case TCP_SEQ_STATE_ESTABLISHED:
1804                 if (sk->sk_state == TCP_TIME_WAIT)
1805                         get_timewait6_sock(seq, v, st->num);
1806                 else
1807                         get_tcp6_sock(seq, v, st->num);
1808                 break;
1809         case TCP_SEQ_STATE_OPENREQ:
1810                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
1811                 break;
1812         }
1813 out:
1814         return 0;
1815 }
1816
1817 static const struct file_operations tcp6_afinfo_seq_fops = {
1818         .owner   = THIS_MODULE,
1819         .open    = tcp_seq_open,
1820         .read    = seq_read,
1821         .llseek  = seq_lseek,
1822         .release = seq_release_net
1823 };
1824
1825 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1826         .name           = "tcp6",
1827         .family         = AF_INET6,
1828         .seq_fops       = &tcp6_afinfo_seq_fops,
1829         .seq_ops        = {
1830                 .show           = tcp6_seq_show,
1831         },
1832 };
1833
1834 int __net_init tcp6_proc_init(struct net *net)
1835 {
1836         return tcp_proc_register(net, &tcp6_seq_afinfo);
1837 }
1838
1839 void tcp6_proc_exit(struct net *net)
1840 {
1841         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1842 }
1843 #endif
1844
1845 static void tcp_v6_clear_sk(struct sock *sk, int size)
1846 {
1847         struct inet_sock *inet = inet_sk(sk);
1848
1849         /* we do not want to clear pinet6 field, because of RCU lookups */
1850         sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1851
1852         size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1853         memset(&inet->pinet6 + 1, 0, size);
1854 }
1855
1856 struct proto tcpv6_prot = {
1857         .name                   = "TCPv6",
1858         .owner                  = THIS_MODULE,
1859         .close                  = tcp_close,
1860         .connect                = tcp_v6_connect,
1861         .disconnect             = tcp_disconnect,
1862         .accept                 = inet_csk_accept,
1863         .ioctl                  = tcp_ioctl,
1864         .init                   = tcp_v6_init_sock,
1865         .destroy                = tcp_v6_destroy_sock,
1866         .shutdown               = tcp_shutdown,
1867         .setsockopt             = tcp_setsockopt,
1868         .getsockopt             = tcp_getsockopt,
1869         .recvmsg                = tcp_recvmsg,
1870         .sendmsg                = tcp_sendmsg,
1871         .sendpage               = tcp_sendpage,
1872         .backlog_rcv            = tcp_v6_do_rcv,
1873         .release_cb             = tcp_release_cb,
1874         .hash                   = tcp_v6_hash,
1875         .unhash                 = inet_unhash,
1876         .get_port               = inet_csk_get_port,
1877         .enter_memory_pressure  = tcp_enter_memory_pressure,
1878         .stream_memory_free     = tcp_stream_memory_free,
1879         .sockets_allocated      = &tcp_sockets_allocated,
1880         .memory_allocated       = &tcp_memory_allocated,
1881         .memory_pressure        = &tcp_memory_pressure,
1882         .orphan_count           = &tcp_orphan_count,
1883         .sysctl_mem             = sysctl_tcp_mem,
1884         .sysctl_wmem            = sysctl_tcp_wmem,
1885         .sysctl_rmem            = sysctl_tcp_rmem,
1886         .max_header             = MAX_TCP_HEADER,
1887         .obj_size               = sizeof(struct tcp6_sock),
1888         .slab_flags             = SLAB_DESTROY_BY_RCU,
1889         .twsk_prot              = &tcp6_timewait_sock_ops,
1890         .rsk_prot               = &tcp6_request_sock_ops,
1891         .h.hashinfo             = &tcp_hashinfo,
1892         .no_autobind            = true,
1893 #ifdef CONFIG_COMPAT
1894         .compat_setsockopt      = compat_tcp_setsockopt,
1895         .compat_getsockopt      = compat_tcp_getsockopt,
1896 #endif
1897 #ifdef CONFIG_MEMCG_KMEM
1898         .proto_cgroup           = tcp_proto_cgroup,
1899 #endif
1900         .clear_sk               = tcp_v6_clear_sk,
1901 };
1902
1903 static const struct inet6_protocol tcpv6_protocol = {
1904         .early_demux    =       tcp_v6_early_demux,
1905         .handler        =       tcp_v6_rcv,
1906         .err_handler    =       tcp_v6_err,
1907         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1908 };
1909
1910 static struct inet_protosw tcpv6_protosw = {
1911         .type           =       SOCK_STREAM,
1912         .protocol       =       IPPROTO_TCP,
1913         .prot           =       &tcpv6_prot,
1914         .ops            =       &inet6_stream_ops,
1915         .flags          =       INET_PROTOSW_PERMANENT |
1916                                 INET_PROTOSW_ICSK,
1917 };
1918
1919 static int __net_init tcpv6_net_init(struct net *net)
1920 {
1921         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1922                                     SOCK_RAW, IPPROTO_TCP, net);
1923 }
1924
1925 static void __net_exit tcpv6_net_exit(struct net *net)
1926 {
1927         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1928 }
1929
1930 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1931 {
1932         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1933 }
1934
1935 static struct pernet_operations tcpv6_net_ops = {
1936         .init       = tcpv6_net_init,
1937         .exit       = tcpv6_net_exit,
1938         .exit_batch = tcpv6_net_exit_batch,
1939 };
1940
1941 int __init tcpv6_init(void)
1942 {
1943         int ret;
1944
1945         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1946         if (ret)
1947                 goto out;
1948
1949         /* register inet6 protocol */
1950         ret = inet6_register_protosw(&tcpv6_protosw);
1951         if (ret)
1952                 goto out_tcpv6_protocol;
1953
1954         ret = register_pernet_subsys(&tcpv6_net_ops);
1955         if (ret)
1956                 goto out_tcpv6_protosw;
1957 out:
1958         return ret;
1959
1960 out_tcpv6_protosw:
1961         inet6_unregister_protosw(&tcpv6_protosw);
1962 out_tcpv6_protocol:
1963         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1964         goto out;
1965 }
1966
1967 void tcpv6_exit(void)
1968 {
1969         unregister_pernet_subsys(&tcpv6_net_ops);
1970         inet6_unregister_protosw(&tcpv6_protosw);
1971         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1972 }