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/busy_poll.h>
  65
  66 #include <linux/proc_fs.h>
  67 #include <linux/seq_file.h>
  68
  69 #include <linux/crypto.h>
  70 #include <linux/scatterlist.h>
  71
  72 static void     tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
  73 static void     tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
  74                                       struct request_sock *req);
  75
  76 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  77
  78 static const struct inet_connection_sock_af_ops ipv6_mapped;
  79 static const struct inet_connection_sock_af_ops ipv6_specific;
  80 #ifdef CONFIG_TCP_MD5SIG
  81 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
  82 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
  83 #else
  84 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
  85                                                    const struct in6_addr *addr)
  86 {
  87         return NULL;
  88 }
  89 #endif
  90
  91 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
  92 {
  93         struct dst_entry *dst = skb_dst(skb);
  94
  95         if (dst && dst_hold_safe(dst)) {
  96                 const struct rt6_info *rt = (const struct rt6_info *)dst;
  97
  98                 sk->sk_rx_dst = dst;
  99                 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
 100                 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
 101         }
 102 }
 103
 104 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
 105 {
 106         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
 107                                             ipv6_hdr(skb)->saddr.s6_addr32,
 108                                             tcp_hdr(skb)->dest,
 109                                             tcp_hdr(skb)->source);
 110 }
 111
 112 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 113                           int addr_len)
 114 {
 115         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 116         struct inet_sock *inet = inet_sk(sk);
 117         struct inet_connection_sock *icsk = inet_csk(sk);
 118         struct ipv6_pinfo *np = inet6_sk(sk);
 119         struct tcp_sock *tp = tcp_sk(sk);
 120         struct in6_addr *saddr = NULL, *final_p, final;
 121         struct ipv6_txoptions *opt;
 122         struct flowi6 fl6;
 123         struct dst_entry *dst;
 124         int addr_type;
 125         int err;
 126
 127         if (addr_len < SIN6_LEN_RFC2133)
 128                 return -EINVAL;
 129
 130         if (usin->sin6_family != AF_INET6)
 131                 return -EAFNOSUPPORT;
 132
 133         memset(&fl6, 0, sizeof(fl6));
 134
 135         if (np->sndflow) {
 136                 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 137                 IP6_ECN_flow_init(fl6.flowlabel);
 138                 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
 139                         struct ip6_flowlabel *flowlabel;
 140                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 141                         if (!flowlabel)
 142                                 return -EINVAL;
 143                         fl6_sock_release(flowlabel);
 144                 }
 145         }
 146
 147         /*
 148          *      connect() to INADDR_ANY means loopback (BSD'ism).
 149          */
 150
 151         if (ipv6_addr_any(&usin->sin6_addr))
 152                 usin->sin6_addr.s6_addr[15] = 0x1;
 153
 154         addr_type = ipv6_addr_type(&usin->sin6_addr);
 155
 156         if (addr_type & IPV6_ADDR_MULTICAST)
 157                 return -ENETUNREACH;
 158
 159         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 160                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 161                     usin->sin6_scope_id) {
 162                         /* If interface is set while binding, indices
 163                          * must coincide.
 164                          */
 165                         if (sk->sk_bound_dev_if &&
 166                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 167                                 return -EINVAL;
 168
 169                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 170                 }
 171
 172                 /* Connect to link-local address requires an interface */
 173                 if (!sk->sk_bound_dev_if)
 174                         return -EINVAL;
 175         }
 176
 177         if (tp->rx_opt.ts_recent_stamp &&
 178             !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
 179                 tp->rx_opt.ts_recent = 0;
 180                 tp->rx_opt.ts_recent_stamp = 0;
 181                 tp->write_seq = 0;
 182         }
 183
 184         sk->sk_v6_daddr = usin->sin6_addr;
 185         np->flow_label = fl6.flowlabel;
 186
 187         /*
 188          *      TCP over IPv4
 189          */
 190
 191         if (addr_type == IPV6_ADDR_MAPPED) {
 192                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 193                 struct sockaddr_in sin;
 194
 195                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 196
 197                 if (__ipv6_only_sock(sk))
 198                         return -ENETUNREACH;
 199
 200                 sin.sin_family = AF_INET;
 201                 sin.sin_port = usin->sin6_port;
 202                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 203
 204                 icsk->icsk_af_ops = &ipv6_mapped;
 205                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 206 #ifdef CONFIG_TCP_MD5SIG
 207                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
 208 #endif
 209
 210                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 211
 212                 if (err) {
 213                         icsk->icsk_ext_hdr_len = exthdrlen;
 214                         icsk->icsk_af_ops = &ipv6_specific;
 215                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 216 #ifdef CONFIG_TCP_MD5SIG
 217                         tp->af_specific = &tcp_sock_ipv6_specific;
 218 #endif
 219                         goto failure;
 220                 }
 221                 np->saddr = sk->sk_v6_rcv_saddr;
 222
 223                 return err;
 224         }
 225
 226         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 227                 saddr = &sk->sk_v6_rcv_saddr;
 228
 229         fl6.flowi6_proto = IPPROTO_TCP;
 230         fl6.daddr = sk->sk_v6_daddr;
 231         fl6.saddr = saddr ? *saddr : np->saddr;
 232         fl6.flowi6_oif = sk->sk_bound_dev_if;
 233         fl6.flowi6_mark = sk->sk_mark;
 234         fl6.fl6_dport = usin->sin6_port;
 235         fl6.fl6_sport = inet->inet_sport;
 236
 237         opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
 238         final_p = fl6_update_dst(&fl6, opt, &final);
 239
 240         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 241
 242         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 243         if (IS_ERR(dst)) {
 244                 err = PTR_ERR(dst);
 245                 goto failure;
 246         }
 247
 248         if (!saddr) {
 249                 saddr = &fl6.saddr;
 250                 sk->sk_v6_rcv_saddr = *saddr;
 251         }
 252
 253         /* set the source address */
 254         np->saddr = *saddr;
 255         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 256
 257         sk->sk_gso_type = SKB_GSO_TCPV6;
 258         ip6_dst_store(sk, dst, NULL, NULL);
 259
 260         if (tcp_death_row.sysctl_tw_recycle &&
 261             !tp->rx_opt.ts_recent_stamp &&
 262             ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
 263                 tcp_fetch_timewait_stamp(sk, dst);
 264
 265         icsk->icsk_ext_hdr_len = 0;
 266         if (opt)
 267                 icsk->icsk_ext_hdr_len = opt->opt_flen +
 268                                          opt->opt_nflen;
 269
 270         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 271
 272         inet->inet_dport = usin->sin6_port;
 273
 274         tcp_set_state(sk, TCP_SYN_SENT);
 275         err = inet6_hash_connect(&tcp_death_row, sk);
 276         if (err)
 277                 goto late_failure;
 278
 279         sk_set_txhash(sk);
 280
 281         if (!tp->write_seq && likely(!tp->repair))
 282                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 283                                                              sk->sk_v6_daddr.s6_addr32,
 284                                                              inet->inet_sport,
 285                                                              inet->inet_dport);
 286
 287         err = tcp_connect(sk);
 288         if (err)
 289                 goto late_failure;
 290
 291         return 0;
 292
 293 late_failure:
 294         tcp_set_state(sk, TCP_CLOSE);
 295         __sk_dst_reset(sk);
 296 failure:
 297         inet->inet_dport = 0;
 298         sk->sk_route_caps = 0;
 299         return err;
 300 }
 301
 302 static void tcp_v6_mtu_reduced(struct sock *sk)
 303 {
 304         struct dst_entry *dst;
 305
 306         if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 307                 return;
 308
 309         dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
 310         if (!dst)
 311                 return;
 312
 313         if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
 314                 tcp_sync_mss(sk, dst_mtu(dst));
 315                 tcp_simple_retransmit(sk);
 316         }
 317 }
 318
 319 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 320                 u8 type, u8 code, int offset, __be32 info)
 321 {
 322         const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
 323         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 324         struct net *net = dev_net(skb->dev);
 325         struct request_sock *fastopen;
 326         struct ipv6_pinfo *np;
 327         struct tcp_sock *tp;
 328         __u32 seq, snd_una;
 329         struct sock *sk;
 330         int err;
 331
 332         sk = __inet6_lookup_established(net, &tcp_hashinfo,
 333                                         &hdr->daddr, th->dest,
 334                                         &hdr->saddr, ntohs(th->source),
 335                                         skb->dev->ifindex);
 336
 337         if (!sk) {
 338                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 339                                    ICMP6_MIB_INERRORS);
 340                 return;
 341         }
 342
 343         if (sk->sk_state == TCP_TIME_WAIT) {
 344                 inet_twsk_put(inet_twsk(sk));
 345                 return;
 346         }
 347         seq = ntohl(th->seq);
 348         if (sk->sk_state == TCP_NEW_SYN_RECV)
 349                 return tcp_req_err(sk, seq);
 350
 351         bh_lock_sock(sk);
 352         if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
 353                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 354
 355         if (sk->sk_state == TCP_CLOSE)
 356                 goto out;
 357
 358         if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
 359                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
 360                 goto out;
 361         }
 362
 363         tp = tcp_sk(sk);
 364         /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 365         fastopen = tp->fastopen_rsk;
 366         snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 367         if (sk->sk_state != TCP_LISTEN &&
 368             !between(seq, snd_una, tp->snd_nxt)) {
 369                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 370                 goto out;
 371         }
 372
 373         np = inet6_sk(sk);
 374
 375         if (type == NDISC_REDIRECT) {
 376                 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 377
 378                 if (dst)
 379                         dst->ops->redirect(dst, sk, skb);
 380                 goto out;
 381         }
 382
 383         if (type == ICMPV6_PKT_TOOBIG) {
 384                 /* We are not interested in TCP_LISTEN and open_requests
 385                  * (SYN-ACKs send out by Linux are always <576bytes so
 386                  * they should go through unfragmented).
 387                  */
 388                 if (sk->sk_state == TCP_LISTEN)
 389                         goto out;
 390
 391                 if (!ip6_sk_accept_pmtu(sk))
 392                         goto out;
 393
 394                 tp->mtu_info = ntohl(info);
 395                 if (!sock_owned_by_user(sk))
 396                         tcp_v6_mtu_reduced(sk);
 397                 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
 398                                            &tp->tsq_flags))
 399                         sock_hold(sk);
 400                 goto out;
 401         }
 402
 403         icmpv6_err_convert(type, code, &err);
 404
 405         /* Might be for an request_sock */
 406         switch (sk->sk_state) {
 407         case TCP_SYN_SENT:
 408         case TCP_SYN_RECV:
 409                 /* Only in fast or simultaneous open. If a fast open socket is
 410                  * is already accepted it is treated as a connected one below.
 411                  */
 412                 if (fastopen && !fastopen->sk)
 413                         break;
 414
 415                 if (!sock_owned_by_user(sk)) {
 416                         sk->sk_err = err;
 417                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 418
 419                         tcp_done(sk);
 420                 } else
 421                         sk->sk_err_soft = err;
 422                 goto out;
 423         }
 424
 425         if (!sock_owned_by_user(sk) && np->recverr) {
 426                 sk->sk_err = err;
 427                 sk->sk_error_report(sk);
 428         } else
 429                 sk->sk_err_soft = err;
 430
 431 out:
 432         bh_unlock_sock(sk);
 433         sock_put(sk);
 434 }
 435
 436
 437 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 438                               struct flowi *fl,
 439                               struct request_sock *req,
 440                               struct tcp_fastopen_cookie *foc,
 441                               bool attach_req)
 442 {
 443         struct inet_request_sock *ireq = inet_rsk(req);
 444         struct ipv6_pinfo *np = inet6_sk(sk);
 445         struct flowi6 *fl6 = &fl->u.ip6;
 446         struct sk_buff *skb;
 447         int err = -ENOMEM;
 448
 449         /* First, grab a route. */
 450         if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
 451                                                IPPROTO_TCP)) == NULL)
 452                 goto done;
 453
 454         skb = tcp_make_synack(sk, dst, req, foc, attach_req);
 455
 456         if (skb) {
 457                 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
 458                                     &ireq->ir_v6_rmt_addr);
 459
 460                 fl6->daddr = ireq->ir_v6_rmt_addr;
 461                 if (np->repflow && ireq->pktopts)
 462                         fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
 463
 464                 rcu_read_lock();
 465                 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
 466                                np->tclass);
 467                 rcu_read_unlock();
 468                 err = net_xmit_eval(err);
 469         }
 470
 471 done:
 472         return err;
 473 }
 474
 475
 476 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 477 {
 478         kfree_skb(inet_rsk(req)->pktopts);
 479 }
 480
 481 #ifdef CONFIG_TCP_MD5SIG
 482 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
 483                                                    const struct in6_addr *addr)
 484 {
 485         return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
 486 }
 487
 488 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
 489                                                 const struct sock *addr_sk)
 490 {
 491         return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
 492 }
 493
 494 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
 495                                  int optlen)
 496 {
 497         struct tcp_md5sig cmd;
 498         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 499
 500         if (optlen < sizeof(cmd))
 501                 return -EINVAL;
 502
 503         if (copy_from_user(&cmd, optval, sizeof(cmd)))
 504                 return -EFAULT;
 505
 506         if (sin6->sin6_family != AF_INET6)
 507                 return -EINVAL;
 508
 509         if (!cmd.tcpm_keylen) {
 510                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 511                         return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 512                                               AF_INET);
 513                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 514                                       AF_INET6);
 515         }
 516
 517         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 518                 return -EINVAL;
 519
 520         if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 521                 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 522                                       AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 523
 524         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 525                               AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 526 }
 527
 528 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
 529                                         const struct in6_addr *daddr,
 530                                         const struct in6_addr *saddr, int nbytes)
 531 {
 532         struct tcp6_pseudohdr *bp;
 533         struct scatterlist sg;
 534
 535         bp = &hp->md5_blk.ip6;
 536         /* 1. TCP pseudo-header (RFC2460) */
 537         bp->saddr = *saddr;
 538         bp->daddr = *daddr;
 539         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 540         bp->len = cpu_to_be32(nbytes);
 541
 542         sg_init_one(&sg, bp, sizeof(*bp));
 543         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
 544 }
 545
 546 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
 547                                const struct in6_addr *daddr, struct in6_addr *saddr,
 548                                const struct tcphdr *th)
 549 {
 550         struct tcp_md5sig_pool *hp;
 551         struct hash_desc *desc;
 552
 553         hp = tcp_get_md5sig_pool();
 554         if (!hp)
 555                 goto clear_hash_noput;
 556         desc = &hp->md5_desc;
 557
 558         if (crypto_hash_init(desc))
 559                 goto clear_hash;
 560         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
 561                 goto clear_hash;
 562         if (tcp_md5_hash_header(hp, th))
 563                 goto clear_hash;
 564         if (tcp_md5_hash_key(hp, key))
 565                 goto clear_hash;
 566         if (crypto_hash_final(desc, md5_hash))
 567                 goto clear_hash;
 568
 569         tcp_put_md5sig_pool();
 570         return 0;
 571
 572 clear_hash:
 573         tcp_put_md5sig_pool();
 574 clear_hash_noput:
 575         memset(md5_hash, 0, 16);
 576         return 1;
 577 }
 578
 579 static int tcp_v6_md5_hash_skb(char *md5_hash,
 580                                const struct tcp_md5sig_key *key,
 581                                const struct sock *sk,
 582                                const struct sk_buff *skb)
 583 {
 584         const struct in6_addr *saddr, *daddr;
 585         struct tcp_md5sig_pool *hp;
 586         struct hash_desc *desc;
 587         const struct tcphdr *th = tcp_hdr(skb);
 588
 589         if (sk) { /* valid for establish/request sockets */
 590                 saddr = &sk->sk_v6_rcv_saddr;
 591                 daddr = &sk->sk_v6_daddr;
 592         } else {
 593                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 594                 saddr = &ip6h->saddr;
 595                 daddr = &ip6h->daddr;
 596         }
 597
 598         hp = tcp_get_md5sig_pool();
 599         if (!hp)
 600                 goto clear_hash_noput;
 601         desc = &hp->md5_desc;
 602
 603         if (crypto_hash_init(desc))
 604                 goto clear_hash;
 605
 606         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
 607                 goto clear_hash;
 608         if (tcp_md5_hash_header(hp, th))
 609                 goto clear_hash;
 610         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 611                 goto clear_hash;
 612         if (tcp_md5_hash_key(hp, key))
 613                 goto clear_hash;
 614         if (crypto_hash_final(desc, md5_hash))
 615                 goto clear_hash;
 616
 617         tcp_put_md5sig_pool();
 618         return 0;
 619
 620 clear_hash:
 621         tcp_put_md5sig_pool();
 622 clear_hash_noput:
 623         memset(md5_hash, 0, 16);
 624         return 1;
 625 }
 626
 627 #endif
 628
 629 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
 630                                     const struct sk_buff *skb)
 631 {
 632 #ifdef CONFIG_TCP_MD5SIG
 633         const __u8 *hash_location = NULL;
 634         struct tcp_md5sig_key *hash_expected;
 635         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 636         const struct tcphdr *th = tcp_hdr(skb);
 637         int genhash;
 638         u8 newhash[16];
 639
 640         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 641         hash_location = tcp_parse_md5sig_option(th);
 642
 643         /* We've parsed the options - do we have a hash? */
 644         if (!hash_expected && !hash_location)
 645                 return false;
 646
 647         if (hash_expected && !hash_location) {
 648                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 649                 return true;
 650         }
 651
 652         if (!hash_expected && hash_location) {
 653                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 654                 return true;
 655         }
 656
 657         /* check the signature */
 658         genhash = tcp_v6_md5_hash_skb(newhash,
 659                                       hash_expected,
 660                                       NULL, skb);
 661
 662         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 663                 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
 664                                      genhash ? "failed" : "mismatch",
 665                                      &ip6h->saddr, ntohs(th->source),
 666                                      &ip6h->daddr, ntohs(th->dest));
 667                 return true;
 668         }
 669 #endif
 670         return false;
 671 }
 672
 673 static void tcp_v6_init_req(struct request_sock *req,
 674                             const struct sock *sk_listener,
 675                             struct sk_buff *skb)
 676 {
 677         struct inet_request_sock *ireq = inet_rsk(req);
 678         const struct ipv6_pinfo *np = inet6_sk(sk_listener);
 679
 680         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 681         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 682
 683         /* So that link locals have meaning */
 684         if (!sk_listener->sk_bound_dev_if &&
 685             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 686                 ireq->ir_iif = tcp_v6_iif(skb);
 687
 688         if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
 689             (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
 690              np->rxopt.bits.rxinfo ||
 691              np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
 692              np->rxopt.bits.rxohlim || np->repflow)) {
 693                 atomic_inc(&skb->users);
 694                 ireq->pktopts = skb;
 695         }
 696 }
 697
 698 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
 699                                           struct flowi *fl,
 700                                           const struct request_sock *req,
 701                                           bool *strict)
 702 {
 703         if (strict)
 704                 *strict = true;
 705         return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
 706 }
 707
 708 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 709         .family         =       AF_INET6,
 710         .obj_size       =       sizeof(struct tcp6_request_sock),
 711         .rtx_syn_ack    =       tcp_rtx_synack,
 712         .send_ack       =       tcp_v6_reqsk_send_ack,
 713         .destructor     =       tcp_v6_reqsk_destructor,
 714         .send_reset     =       tcp_v6_send_reset,
 715         .syn_ack_timeout =      tcp_syn_ack_timeout,
 716 };
 717
 718 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 719         .mss_clamp      =       IPV6_MIN_MTU - sizeof(struct tcphdr) -
 720                                 sizeof(struct ipv6hdr),
 721 #ifdef CONFIG_TCP_MD5SIG
 722         .req_md5_lookup =       tcp_v6_md5_lookup,
 723         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 724 #endif
 725         .init_req       =       tcp_v6_init_req,
 726 #ifdef CONFIG_SYN_COOKIES
 727         .cookie_init_seq =      cookie_v6_init_sequence,
 728 #endif
 729         .route_req      =       tcp_v6_route_req,
 730         .init_seq       =       tcp_v6_init_sequence,
 731         .send_synack    =       tcp_v6_send_synack,
 732 };
 733
 734 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
 735                                  u32 ack, u32 win, u32 tsval, u32 tsecr,
 736                                  int oif, struct tcp_md5sig_key *key, int rst,
 737                                  u8 tclass, u32 label)
 738 {
 739         const struct tcphdr *th = tcp_hdr(skb);
 740         struct tcphdr *t1;
 741         struct sk_buff *buff;
 742         struct flowi6 fl6;
 743         struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 744         struct sock *ctl_sk = net->ipv6.tcp_sk;
 745         unsigned int tot_len = sizeof(struct tcphdr);
 746         struct dst_entry *dst;
 747         __be32 *topt;
 748
 749         if (tsecr)
 750                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 751 #ifdef CONFIG_TCP_MD5SIG
 752         if (key)
 753                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 754 #endif
 755
 756         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 757                          GFP_ATOMIC);
 758         if (!buff)
 759                 return;
 760
 761         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 762
 763         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 764         skb_reset_transport_header(buff);
 765
 766         /* Swap the send and the receive. */
 767         memset(t1, 0, sizeof(*t1));
 768         t1->dest = th->source;
 769         t1->source = th->dest;
 770         t1->doff = tot_len / 4;
 771         t1->seq = htonl(seq);
 772         t1->ack_seq = htonl(ack);
 773         t1->ack = !rst || !th->ack;
 774         t1->rst = rst;
 775         t1->window = htons(win);
 776
 777         topt = (__be32 *)(t1 + 1);
 778
 779         if (tsecr) {
 780                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 781                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 782                 *topt++ = htonl(tsval);
 783                 *topt++ = htonl(tsecr);
 784         }
 785
 786 #ifdef CONFIG_TCP_MD5SIG
 787         if (key) {
 788                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 789                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 790                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 791                                     &ipv6_hdr(skb)->saddr,
 792                                     &ipv6_hdr(skb)->daddr, t1);
 793         }
 794 #endif
 795
 796         memset(&fl6, 0, sizeof(fl6));
 797         fl6.daddr = ipv6_hdr(skb)->saddr;
 798         fl6.saddr = ipv6_hdr(skb)->daddr;
 799         fl6.flowlabel = label;
 800
 801         buff->ip_summed = CHECKSUM_PARTIAL;
 802         buff->csum = 0;
 803
 804         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
 805
 806         fl6.flowi6_proto = IPPROTO_TCP;
 807         if (rt6_need_strict(&fl6.daddr) && !oif)
 808                 fl6.flowi6_oif = tcp_v6_iif(skb);
 809         else
 810                 fl6.flowi6_oif = oif;
 811         fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
 812         fl6.fl6_dport = t1->dest;
 813         fl6.fl6_sport = t1->source;
 814         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 815
 816         /* Pass a socket to ip6_dst_lookup either it is for RST
 817          * Underlying function will use this to retrieve the network
 818          * namespace
 819          */
 820         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 821         if (!IS_ERR(dst)) {
 822                 skb_dst_set(buff, dst);
 823                 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
 824                 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 825                 if (rst)
 826                         TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
 827                 return;
 828         }
 829
 830         kfree_skb(buff);
 831 }
 832
 833 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
 834 {
 835         const struct tcphdr *th = tcp_hdr(skb);
 836         u32 seq = 0, ack_seq = 0;
 837         struct tcp_md5sig_key *key = NULL;
 838 #ifdef CONFIG_TCP_MD5SIG
 839         const __u8 *hash_location = NULL;
 840         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 841         unsigned char newhash[16];
 842         int genhash;
 843         struct sock *sk1 = NULL;
 844 #endif
 845         int oif;
 846
 847         if (th->rst)
 848                 return;
 849
 850         /* If sk not NULL, it means we did a successful lookup and incoming
 851          * route had to be correct. prequeue might have dropped our dst.
 852          */
 853         if (!sk && !ipv6_unicast_destination(skb))
 854                 return;
 855
 856 #ifdef CONFIG_TCP_MD5SIG
 857         hash_location = tcp_parse_md5sig_option(th);
 858         if (sk && sk_fullsock(sk)) {
 859                 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
 860         } else if (hash_location) {
 861                 /*
 862                  * active side is lost. Try to find listening socket through
 863                  * source port, and then find md5 key through listening socket.
 864                  * we are not loose security here:
 865                  * Incoming packet is checked with md5 hash with finding key,
 866                  * no RST generated if md5 hash doesn't match.
 867                  */
 868                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
 869                                            &tcp_hashinfo, NULL, 0,
 870                                            &ipv6h->saddr,
 871                                            th->source, &ipv6h->daddr,
 872                                            ntohs(th->source), tcp_v6_iif(skb));
 873                 if (!sk1)
 874                         return;
 875
 876                 rcu_read_lock();
 877                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 878                 if (!key)
 879                         goto release_sk1;
 880
 881                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
 882                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
 883                         goto release_sk1;
 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(sk, 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(const struct sock *sk, struct sk_buff *skb, u32 seq,
 906                             u32 ack, 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(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
 911                              tclass, 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(sk, 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, cpu_to_be32(tw->tw_flowlabel));
 924
 925         inet_twsk_put(tw);
 926 }
 927
 928 static void tcp_v6_reqsk_send_ack(const 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(sk, skb, (sk->sk_state == TCP_LISTEN) ?
 935                         tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
 936                         tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
 937                         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_cookie_check(struct sock *sk, struct sk_buff *skb)
 944 {
 945 #ifdef CONFIG_SYN_COOKIES
 946         const struct tcphdr *th = tcp_hdr(skb);
 947
 948         if (!th->syn)
 949                 sk = cookie_v6_check(sk, skb);
 950 #endif
 951         return sk;
 952 }
 953
 954 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 955 {
 956         if (skb->protocol == htons(ETH_P_IP))
 957                 return tcp_v4_conn_request(sk, skb);
 958
 959         if (!ipv6_unicast_destination(skb))
 960                 goto drop;
 961
 962         return tcp_conn_request(&tcp6_request_sock_ops,
 963                                 &tcp_request_sock_ipv6_ops, sk, skb);
 964
 965 drop:
 966         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 967         return 0; /* don't send reset */
 968 }
 969
 970 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
 971                                          struct request_sock *req,
 972                                          struct dst_entry *dst,
 973                                          struct request_sock *req_unhash,
 974                                          bool *own_req)
 975 {
 976         struct inet_request_sock *ireq;
 977         struct ipv6_pinfo *newnp;
 978         const struct ipv6_pinfo *np = inet6_sk(sk);
 979         struct ipv6_txoptions *opt;
 980         struct tcp6_sock *newtcp6sk;
 981         struct inet_sock *newinet;
 982         struct tcp_sock *newtp;
 983         struct sock *newsk;
 984 #ifdef CONFIG_TCP_MD5SIG
 985         struct tcp_md5sig_key *key;
 986 #endif
 987         struct flowi6 fl6;
 988
 989         if (skb->protocol == htons(ETH_P_IP)) {
 990                 /*
 991                  *      v6 mapped
 992                  */
 993
 994                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
 995                                              req_unhash, own_req);
 996
 997                 if (!newsk)
 998                         return NULL;
 999
1000                 newtcp6sk = (struct tcp6_sock *)newsk;
1001                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1002
1003                 newinet = inet_sk(newsk);
1004                 newnp = inet6_sk(newsk);
1005                 newtp = tcp_sk(newsk);
1006
1007                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1008
1009                 newnp->saddr = newsk->sk_v6_rcv_saddr;
1010
1011                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1012                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1013 #ifdef CONFIG_TCP_MD5SIG
1014                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1015 #endif
1016
1017                 newnp->ipv6_ac_list = NULL;
1018                 newnp->ipv6_fl_list = NULL;
1019                 newnp->pktoptions  = NULL;
1020                 newnp->opt         = NULL;
1021                 newnp->mcast_oif   = tcp_v6_iif(skb);
1022                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1023                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1024                 if (np->repflow)
1025                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1026
1027                 /*
1028                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1029                  * here, tcp_create_openreq_child now does this for us, see the comment in
1030                  * that function for the gory details. -acme
1031                  */
1032
1033                 /* It is tricky place. Until this moment IPv4 tcp
1034                    worked with IPv6 icsk.icsk_af_ops.
1035                    Sync it now.
1036                  */
1037                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1038
1039                 return newsk;
1040         }
1041
1042         ireq = inet_rsk(req);
1043
1044         if (sk_acceptq_is_full(sk))
1045                 goto out_overflow;
1046
1047         if (!dst) {
1048                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1049                 if (!dst)
1050                         goto out;
1051         }
1052
1053         newsk = tcp_create_openreq_child(sk, req, skb);
1054         if (!newsk)
1055                 goto out_nonewsk;
1056
1057         /*
1058          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1059          * count here, tcp_create_openreq_child now does this for us, see the
1060          * comment in that function for the gory details. -acme
1061          */
1062
1063         newsk->sk_gso_type = SKB_GSO_TCPV6;
1064         ip6_dst_store(newsk, dst, NULL, NULL);
1065         inet6_sk_rx_dst_set(newsk, skb);
1066
1067         newtcp6sk = (struct tcp6_sock *)newsk;
1068         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1069
1070         newtp = tcp_sk(newsk);
1071         newinet = inet_sk(newsk);
1072         newnp = inet6_sk(newsk);
1073
1074         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1075
1076         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1077         newnp->saddr = ireq->ir_v6_loc_addr;
1078         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1079         newsk->sk_bound_dev_if = ireq->ir_iif;
1080
1081         /* Now IPv6 options...
1082
1083            First: no IPv4 options.
1084          */
1085         newinet->inet_opt = NULL;
1086         newnp->ipv6_ac_list = NULL;
1087         newnp->ipv6_fl_list = NULL;
1088
1089         /* Clone RX bits */
1090         newnp->rxopt.all = np->rxopt.all;
1091
1092         newnp->pktoptions = NULL;
1093         newnp->opt        = NULL;
1094         newnp->mcast_oif  = tcp_v6_iif(skb);
1095         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1096         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1097         if (np->repflow)
1098                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1099
1100         /* Clone native IPv6 options from listening socket (if any)
1101
1102            Yes, keeping reference count would be much more clever,
1103            but we make one more one thing there: reattach optmem
1104            to newsk.
1105          */
1106         opt = rcu_dereference(np->opt);
1107         if (opt) {
1108                 opt = ipv6_dup_options(newsk, opt);
1109                 RCU_INIT_POINTER(newnp->opt, opt);
1110         }
1111         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1112         if (opt)
1113                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1114                                                     opt->opt_flen;
1115
1116         tcp_ca_openreq_child(newsk, dst);
1117
1118         tcp_sync_mss(newsk, dst_mtu(dst));
1119         newtp->advmss = dst_metric_advmss(dst);
1120         if (tcp_sk(sk)->rx_opt.user_mss &&
1121             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1122                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1123
1124         tcp_initialize_rcv_mss(newsk);
1125
1126         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1127         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1128
1129 #ifdef CONFIG_TCP_MD5SIG
1130         /* Copy over the MD5 key from the original socket */
1131         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1132         if (key) {
1133                 /* We're using one, so create a matching key
1134                  * on the newsk structure. If we fail to get
1135                  * memory, then we end up not copying the key
1136                  * across. Shucks.
1137                  */
1138                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1139                                AF_INET6, key->key, key->keylen,
1140                                sk_gfp_mask(sk, GFP_ATOMIC));
1141         }
1142 #endif
1143
1144         if (__inet_inherit_port(sk, newsk) < 0) {
1145                 inet_csk_prepare_forced_close(newsk);
1146                 tcp_done(newsk);
1147                 goto out;
1148         }
1149         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1150         if (*own_req) {
1151                 tcp_move_syn(newtp, req);
1152
1153                 /* Clone pktoptions received with SYN, if we own the req */
1154                 if (ireq->pktopts) {
1155                         newnp->pktoptions = skb_clone(ireq->pktopts,
1156                                                       sk_gfp_mask(sk, GFP_ATOMIC));
1157                         consume_skb(ireq->pktopts);
1158                         ireq->pktopts = NULL;
1159                         if (newnp->pktoptions)
1160                                 skb_set_owner_r(newnp->pktoptions, newsk);
1161                 }
1162         }
1163
1164         return newsk;
1165
1166 out_overflow:
1167         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1168 out_nonewsk:
1169         dst_release(dst);
1170 out:
1171         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1172         return NULL;
1173 }
1174
1175 /* The socket must have it's spinlock held when we get
1176  * here, unless it is a TCP_LISTEN socket.
1177  *
1178  * We have a potential double-lock case here, so even when
1179  * doing backlog processing we use the BH locking scheme.
1180  * This is because we cannot sleep with the original spinlock
1181  * held.
1182  */
1183 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1184 {
1185         struct ipv6_pinfo *np = inet6_sk(sk);
1186         struct tcp_sock *tp;
1187         struct sk_buff *opt_skb = NULL;
1188
1189         /* Imagine: socket is IPv6. IPv4 packet arrives,
1190            goes to IPv4 receive handler and backlogged.
1191            From backlog it always goes here. Kerboom...
1192            Fortunately, tcp_rcv_established and rcv_established
1193            handle them correctly, but it is not case with
1194            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1195          */
1196
1197         if (skb->protocol == htons(ETH_P_IP))
1198                 return tcp_v4_do_rcv(sk, skb);
1199
1200         if (sk_filter(sk, skb))
1201                 goto discard;
1202
1203         /*
1204          *      socket locking is here for SMP purposes as backlog rcv
1205          *      is currently called with bh processing disabled.
1206          */
1207
1208         /* Do Stevens' IPV6_PKTOPTIONS.
1209
1210            Yes, guys, it is the only place in our code, where we
1211            may make it not affecting IPv4.
1212            The rest of code is protocol independent,
1213            and I do not like idea to uglify IPv4.
1214
1215            Actually, all the idea behind IPV6_PKTOPTIONS
1216            looks not very well thought. For now we latch
1217            options, received in the last packet, enqueued
1218            by tcp. Feel free to propose better solution.
1219                                                --ANK (980728)
1220          */
1221         if (np->rxopt.all)
1222                 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1223
1224         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1225                 struct dst_entry *dst = sk->sk_rx_dst;
1226
1227                 sock_rps_save_rxhash(sk, skb);
1228                 sk_mark_napi_id(sk, skb);
1229                 if (dst) {
1230                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1231                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1232                                 dst_release(dst);
1233                                 sk->sk_rx_dst = NULL;
1234                         }
1235                 }
1236
1237                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1238                 if (opt_skb)
1239                         goto ipv6_pktoptions;
1240                 return 0;
1241         }
1242
1243         if (tcp_checksum_complete(skb))
1244                 goto csum_err;
1245
1246         if (sk->sk_state == TCP_LISTEN) {
1247                 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1248
1249                 if (!nsk)
1250                         goto discard;
1251
1252                 if (nsk != sk) {
1253                         sock_rps_save_rxhash(nsk, skb);
1254                         sk_mark_napi_id(nsk, skb);
1255                         if (tcp_child_process(sk, nsk, skb))
1256                                 goto reset;
1257                         if (opt_skb)
1258                                 __kfree_skb(opt_skb);
1259                         return 0;
1260                 }
1261         } else
1262                 sock_rps_save_rxhash(sk, skb);
1263
1264         if (tcp_rcv_state_process(sk, skb))
1265                 goto reset;
1266         if (opt_skb)
1267                 goto ipv6_pktoptions;
1268         return 0;
1269
1270 reset:
1271         tcp_v6_send_reset(sk, skb);
1272 discard:
1273         if (opt_skb)
1274                 __kfree_skb(opt_skb);
1275         kfree_skb(skb);
1276         return 0;
1277 csum_err:
1278         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1279         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1280         goto discard;
1281
1282
1283 ipv6_pktoptions:
1284         /* Do you ask, what is it?
1285
1286            1. skb was enqueued by tcp.
1287            2. skb is added to tail of read queue, rather than out of order.
1288            3. socket is not in passive state.
1289            4. Finally, it really contains options, which user wants to receive.
1290          */
1291         tp = tcp_sk(sk);
1292         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1293             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1294                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1295                         np->mcast_oif = tcp_v6_iif(opt_skb);
1296                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1297                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1298                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1299                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1300                 if (np->repflow)
1301                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1302                 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1303                         skb_set_owner_r(opt_skb, sk);
1304                         opt_skb = xchg(&np->pktoptions, opt_skb);
1305                 } else {
1306                         __kfree_skb(opt_skb);
1307                         opt_skb = xchg(&np->pktoptions, NULL);
1308                 }
1309         }
1310
1311         kfree_skb(opt_skb);
1312         return 0;
1313 }
1314
1315 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1316                            const struct tcphdr *th)
1317 {
1318         /* This is tricky: we move IP6CB at its correct location into
1319          * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1320          * _decode_session6() uses IP6CB().
1321          * barrier() makes sure compiler won't play aliasing games.
1322          */
1323         memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1324                 sizeof(struct inet6_skb_parm));
1325         barrier();
1326
1327         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1328         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1329                                     skb->len - th->doff*4);
1330         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1331         TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1332         TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1333         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1334         TCP_SKB_CB(skb)->sacked = 0;
1335 }
1336
1337 static void tcp_v6_restore_cb(struct sk_buff *skb)
1338 {
1339         /* We need to move header back to the beginning if xfrm6_policy_check()
1340          * and tcp_v6_fill_cb() are going to be called again.
1341          */
1342         memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1343                 sizeof(struct inet6_skb_parm));
1344 }
1345
1346 static int tcp_v6_rcv(struct sk_buff *skb)
1347 {
1348         const struct tcphdr *th;
1349         const struct ipv6hdr *hdr;
1350         struct sock *sk;
1351         int ret;
1352         struct net *net = dev_net(skb->dev);
1353
1354         if (skb->pkt_type != PACKET_HOST)
1355                 goto discard_it;
1356
1357         /*
1358          *      Count it even if it's bad.
1359          */
1360         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1361
1362         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1363                 goto discard_it;
1364
1365         th = tcp_hdr(skb);
1366
1367         if (th->doff < sizeof(struct tcphdr)/4)
1368                 goto bad_packet;
1369         if (!pskb_may_pull(skb, th->doff*4))
1370                 goto discard_it;
1371
1372         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1373                 goto csum_error;
1374
1375         th = tcp_hdr(skb);
1376         hdr = ipv6_hdr(skb);
1377
1378 lookup:
1379         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1380                                 th->source, th->dest, inet6_iif(skb));
1381         if (!sk)
1382                 goto no_tcp_socket;
1383
1384 process:
1385         if (sk->sk_state == TCP_TIME_WAIT)
1386                 goto do_time_wait;
1387
1388         if (sk->sk_state == TCP_NEW_SYN_RECV) {
1389                 struct request_sock *req = inet_reqsk(sk);
1390                 struct sock *nsk = NULL;
1391
1392                 sk = req->rsk_listener;
1393                 tcp_v6_fill_cb(skb, hdr, th);
1394                 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1395                         reqsk_put(req);
1396                         goto discard_it;
1397                 }
1398                 if (likely(sk->sk_state == TCP_LISTEN)) {
1399                         nsk = tcp_check_req(sk, skb, req, false);
1400                 } else {
1401                         inet_csk_reqsk_queue_drop_and_put(sk, req);
1402                         goto lookup;
1403                 }
1404                 if (!nsk) {
1405                         reqsk_put(req);
1406                         goto discard_it;
1407                 }
1408                 if (nsk == sk) {
1409                         sock_hold(sk);
1410                         reqsk_put(req);
1411                         tcp_v6_restore_cb(skb);
1412                 } else if (tcp_child_process(sk, nsk, skb)) {
1413                         tcp_v6_send_reset(nsk, skb);
1414                         goto discard_it;
1415                 } else {
1416                         return 0;
1417                 }
1418         }
1419         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1420                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1421                 goto discard_and_relse;
1422         }
1423
1424         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1425                 goto discard_and_relse;
1426
1427         tcp_v6_fill_cb(skb, hdr, th);
1428
1429         if (tcp_v6_inbound_md5_hash(sk, skb))
1430                 goto discard_and_relse;
1431
1432         if (sk_filter(sk, skb))
1433                 goto discard_and_relse;
1434
1435         skb->dev = NULL;
1436
1437         if (sk->sk_state == TCP_LISTEN) {
1438                 ret = tcp_v6_do_rcv(sk, skb);
1439                 goto put_and_return;
1440         }
1441
1442         sk_incoming_cpu_update(sk);
1443
1444         bh_lock_sock_nested(sk);
1445         tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1446         ret = 0;
1447         if (!sock_owned_by_user(sk)) {
1448                 if (!tcp_prequeue(sk, skb))
1449                         ret = tcp_v6_do_rcv(sk, skb);
1450         } else if (unlikely(sk_add_backlog(sk, skb,
1451                                            sk->sk_rcvbuf + sk->sk_sndbuf))) {
1452                 bh_unlock_sock(sk);
1453                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1454                 goto discard_and_relse;
1455         }
1456         bh_unlock_sock(sk);
1457
1458 put_and_return:
1459         sock_put(sk);
1460         return ret ? -1 : 0;
1461
1462 no_tcp_socket:
1463         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1464                 goto discard_it;
1465
1466         tcp_v6_fill_cb(skb, hdr, th);
1467
1468         if (tcp_checksum_complete(skb)) {
1469 csum_error:
1470                 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1471 bad_packet:
1472                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1473         } else {
1474                 tcp_v6_send_reset(NULL, skb);
1475         }
1476
1477 discard_it:
1478         kfree_skb(skb);
1479         return 0;
1480
1481 discard_and_relse:
1482         sock_put(sk);
1483         goto discard_it;
1484
1485 do_time_wait:
1486         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1487                 inet_twsk_put(inet_twsk(sk));
1488                 goto discard_it;
1489         }
1490
1491         tcp_v6_fill_cb(skb, hdr, th);
1492
1493         if (tcp_checksum_complete(skb)) {
1494                 inet_twsk_put(inet_twsk(sk));
1495                 goto csum_error;
1496         }
1497
1498         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1499         case TCP_TW_SYN:
1500         {
1501                 struct sock *sk2;
1502
1503                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1504                                             skb, __tcp_hdrlen(th),
1505                                             &ipv6_hdr(skb)->saddr, th->source,
1506                                             &ipv6_hdr(skb)->daddr,
1507                                             ntohs(th->dest), tcp_v6_iif(skb));
1508                 if (sk2) {
1509                         struct inet_timewait_sock *tw = inet_twsk(sk);
1510                         inet_twsk_deschedule_put(tw);
1511                         sk = sk2;
1512                         tcp_v6_restore_cb(skb);
1513                         goto process;
1514                 }
1515                 /* Fall through to ACK */
1516         }
1517         case TCP_TW_ACK:
1518                 tcp_v6_timewait_ack(sk, skb);
1519                 break;
1520         case TCP_TW_RST:
1521                 tcp_v6_restore_cb(skb);
1522                 tcp_v6_send_reset(sk, skb);
1523                 inet_twsk_deschedule_put(inet_twsk(sk));
1524                 goto discard_it;
1525         case TCP_TW_SUCCESS:
1526                 ;
1527         }
1528         goto discard_it;
1529 }
1530
1531 static void tcp_v6_early_demux(struct sk_buff *skb)
1532 {
1533         const struct ipv6hdr *hdr;
1534         const struct tcphdr *th;
1535         struct sock *sk;
1536
1537         if (skb->pkt_type != PACKET_HOST)
1538                 return;
1539
1540         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1541                 return;
1542
1543         hdr = ipv6_hdr(skb);
1544         th = tcp_hdr(skb);
1545
1546         if (th->doff < sizeof(struct tcphdr) / 4)
1547                 return;
1548
1549         /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1550         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1551                                         &hdr->saddr, th->source,
1552                                         &hdr->daddr, ntohs(th->dest),
1553                                         inet6_iif(skb));
1554         if (sk) {
1555                 skb->sk = sk;
1556                 skb->destructor = sock_edemux;
1557                 if (sk_fullsock(sk)) {
1558                         struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1559
1560                         if (dst)
1561                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1562                         if (dst &&
1563                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1564                                 skb_dst_set_noref(skb, dst);
1565                 }
1566         }
1567 }
1568
1569 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1570         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1571         .twsk_unique    = tcp_twsk_unique,
1572         .twsk_destructor = tcp_twsk_destructor,
1573 };
1574
1575 static const struct inet_connection_sock_af_ops ipv6_specific = {
1576         .queue_xmit        = inet6_csk_xmit,
1577         .send_check        = tcp_v6_send_check,
1578         .rebuild_header    = inet6_sk_rebuild_header,
1579         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1580         .conn_request      = tcp_v6_conn_request,
1581         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1582         .net_header_len    = sizeof(struct ipv6hdr),
1583         .net_frag_header_len = sizeof(struct frag_hdr),
1584         .setsockopt        = ipv6_setsockopt,
1585         .getsockopt        = ipv6_getsockopt,
1586         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1587         .sockaddr_len      = sizeof(struct sockaddr_in6),
1588         .bind_conflict     = inet6_csk_bind_conflict,
1589 #ifdef CONFIG_COMPAT
1590         .compat_setsockopt = compat_ipv6_setsockopt,
1591         .compat_getsockopt = compat_ipv6_getsockopt,
1592 #endif
1593         .mtu_reduced       = tcp_v6_mtu_reduced,
1594 };
1595
1596 #ifdef CONFIG_TCP_MD5SIG
1597 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1598         .md5_lookup     =       tcp_v6_md5_lookup,
1599         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1600         .md5_parse      =       tcp_v6_parse_md5_keys,
1601 };
1602 #endif
1603
1604 /*
1605  *      TCP over IPv4 via INET6 API
1606  */
1607 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1608         .queue_xmit        = ip_queue_xmit,
1609         .send_check        = tcp_v4_send_check,
1610         .rebuild_header    = inet_sk_rebuild_header,
1611         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1612         .conn_request      = tcp_v6_conn_request,
1613         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1614         .net_header_len    = sizeof(struct iphdr),
1615         .setsockopt        = ipv6_setsockopt,
1616         .getsockopt        = ipv6_getsockopt,
1617         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1618         .sockaddr_len      = sizeof(struct sockaddr_in6),
1619         .bind_conflict     = inet6_csk_bind_conflict,
1620 #ifdef CONFIG_COMPAT
1621         .compat_setsockopt = compat_ipv6_setsockopt,
1622         .compat_getsockopt = compat_ipv6_getsockopt,
1623 #endif
1624         .mtu_reduced       = tcp_v4_mtu_reduced,
1625 };
1626
1627 #ifdef CONFIG_TCP_MD5SIG
1628 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1629         .md5_lookup     =       tcp_v4_md5_lookup,
1630         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1631         .md5_parse      =       tcp_v6_parse_md5_keys,
1632 };
1633 #endif
1634
1635 /* NOTE: A lot of things set to zero explicitly by call to
1636  *       sk_alloc() so need not be done here.
1637  */
1638 static int tcp_v6_init_sock(struct sock *sk)
1639 {
1640         struct inet_connection_sock *icsk = inet_csk(sk);
1641
1642         tcp_init_sock(sk);
1643
1644         icsk->icsk_af_ops = &ipv6_specific;
1645
1646 #ifdef CONFIG_TCP_MD5SIG
1647         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1648 #endif
1649
1650         return 0;
1651 }
1652
1653 static void tcp_v6_destroy_sock(struct sock *sk)
1654 {
1655         tcp_v4_destroy_sock(sk);
1656         inet6_destroy_sock(sk);
1657 }
1658
1659 #ifdef CONFIG_PROC_FS
1660 /* Proc filesystem TCPv6 sock list dumping. */
1661 static void get_openreq6(struct seq_file *seq,
1662                          const struct request_sock *req, int i)
1663 {
1664         long ttd = req->rsk_timer.expires - jiffies;
1665         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1666         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1667
1668         if (ttd < 0)
1669                 ttd = 0;
1670
1671         seq_printf(seq,
1672                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1673                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1674                    i,
1675                    src->s6_addr32[0], src->s6_addr32[1],
1676                    src->s6_addr32[2], src->s6_addr32[3],
1677                    inet_rsk(req)->ir_num,
1678                    dest->s6_addr32[0], dest->s6_addr32[1],
1679                    dest->s6_addr32[2], dest->s6_addr32[3],
1680                    ntohs(inet_rsk(req)->ir_rmt_port),
1681                    TCP_SYN_RECV,
1682                    0, 0, /* could print option size, but that is af dependent. */
1683                    1,   /* timers active (only the expire timer) */
1684                    jiffies_to_clock_t(ttd),
1685                    req->num_timeout,
1686                    from_kuid_munged(seq_user_ns(seq),
1687                                     sock_i_uid(req->rsk_listener)),
1688                    0,  /* non standard timer */
1689                    0, /* open_requests have no inode */
1690                    0, req);
1691 }
1692
1693 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1694 {
1695         const struct in6_addr *dest, *src;
1696         __u16 destp, srcp;
1697         int timer_active;
1698         unsigned long timer_expires;
1699         const struct inet_sock *inet = inet_sk(sp);
1700         const struct tcp_sock *tp = tcp_sk(sp);
1701         const struct inet_connection_sock *icsk = inet_csk(sp);
1702         const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1703         int rx_queue;
1704         int state;
1705
1706         dest  = &sp->sk_v6_daddr;
1707         src   = &sp->sk_v6_rcv_saddr;
1708         destp = ntohs(inet->inet_dport);
1709         srcp  = ntohs(inet->inet_sport);
1710
1711         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1712                 timer_active    = 1;
1713                 timer_expires   = icsk->icsk_timeout;
1714         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1715                 timer_active    = 4;
1716                 timer_expires   = icsk->icsk_timeout;
1717         } else if (timer_pending(&sp->sk_timer)) {
1718                 timer_active    = 2;
1719                 timer_expires   = sp->sk_timer.expires;
1720         } else {
1721                 timer_active    = 0;
1722                 timer_expires = jiffies;
1723         }
1724
1725         state = sk_state_load(sp);
1726         if (state == TCP_LISTEN)
1727                 rx_queue = sp->sk_ack_backlog;
1728         else
1729                 /* Because we don't lock the socket,
1730                  * we might find a transient negative value.
1731                  */
1732                 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1733
1734         seq_printf(seq,
1735                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1736                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1737                    i,
1738                    src->s6_addr32[0], src->s6_addr32[1],
1739                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1740                    dest->s6_addr32[0], dest->s6_addr32[1],
1741                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1742                    state,
1743                    tp->write_seq - tp->snd_una,
1744                    rx_queue,
1745                    timer_active,
1746                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1747                    icsk->icsk_retransmits,
1748                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1749                    icsk->icsk_probes_out,
1750                    sock_i_ino(sp),
1751                    atomic_read(&sp->sk_refcnt), sp,
1752                    jiffies_to_clock_t(icsk->icsk_rto),
1753                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1754                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1755                    tp->snd_cwnd,
1756                    state == TCP_LISTEN ?
1757                         fastopenq->max_qlen :
1758                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1759                    );
1760 }
1761
1762 static void get_timewait6_sock(struct seq_file *seq,
1763                                struct inet_timewait_sock *tw, int i)
1764 {
1765         long delta = tw->tw_timer.expires - jiffies;
1766         const struct in6_addr *dest, *src;
1767         __u16 destp, srcp;
1768
1769         dest = &tw->tw_v6_daddr;
1770         src  = &tw->tw_v6_rcv_saddr;
1771         destp = ntohs(tw->tw_dport);
1772         srcp  = ntohs(tw->tw_sport);
1773
1774         seq_printf(seq,
1775                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1776                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1777                    i,
1778                    src->s6_addr32[0], src->s6_addr32[1],
1779                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1780                    dest->s6_addr32[0], dest->s6_addr32[1],
1781                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1782                    tw->tw_substate, 0, 0,
1783                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1784                    atomic_read(&tw->tw_refcnt), tw);
1785 }
1786
1787 static int tcp6_seq_show(struct seq_file *seq, void *v)
1788 {
1789         struct tcp_iter_state *st;
1790         struct sock *sk = v;
1791
1792         if (v == SEQ_START_TOKEN) {
1793                 seq_puts(seq,
1794                          "  sl  "
1795                          "local_address                         "
1796                          "remote_address                        "
1797                          "st tx_queue rx_queue tr tm->when retrnsmt"
1798                          "   uid  timeout inode\n");
1799                 goto out;
1800         }
1801         st = seq->private;
1802
1803         if (sk->sk_state == TCP_TIME_WAIT)
1804                 get_timewait6_sock(seq, v, st->num);
1805         else if (sk->sk_state == TCP_NEW_SYN_RECV)
1806                 get_openreq6(seq, v, st->num);
1807         else
1808                 get_tcp6_sock(seq, v, st->num);
1809 out:
1810         return 0;
1811 }
1812
1813 static const struct file_operations tcp6_afinfo_seq_fops = {
1814         .owner   = THIS_MODULE,
1815         .open    = tcp_seq_open,
1816         .read    = seq_read,
1817         .llseek  = seq_lseek,
1818         .release = seq_release_net
1819 };
1820
1821 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1822         .name           = "tcp6",
1823         .family         = AF_INET6,
1824         .seq_fops       = &tcp6_afinfo_seq_fops,
1825         .seq_ops        = {
1826                 .show           = tcp6_seq_show,
1827         },
1828 };
1829
1830 int __net_init tcp6_proc_init(struct net *net)
1831 {
1832         return tcp_proc_register(net, &tcp6_seq_afinfo);
1833 }
1834
1835 void tcp6_proc_exit(struct net *net)
1836 {
1837         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1838 }
1839 #endif
1840
1841 static void tcp_v6_clear_sk(struct sock *sk, int size)
1842 {
1843         struct inet_sock *inet = inet_sk(sk);
1844
1845         /* we do not want to clear pinet6 field, because of RCU lookups */
1846         sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1847
1848         size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1849         memset(&inet->pinet6 + 1, 0, size);
1850 }
1851
1852 struct proto tcpv6_prot = {
1853         .name                   = "TCPv6",
1854         .owner                  = THIS_MODULE,
1855         .close                  = tcp_close,
1856         .connect                = tcp_v6_connect,
1857         .disconnect             = tcp_disconnect,
1858         .accept                 = inet_csk_accept,
1859         .ioctl                  = tcp_ioctl,
1860         .init                   = tcp_v6_init_sock,
1861         .destroy                = tcp_v6_destroy_sock,
1862         .shutdown               = tcp_shutdown,
1863         .setsockopt             = tcp_setsockopt,
1864         .getsockopt             = tcp_getsockopt,
1865         .recvmsg                = tcp_recvmsg,
1866         .sendmsg                = tcp_sendmsg,
1867         .sendpage               = tcp_sendpage,
1868         .backlog_rcv            = tcp_v6_do_rcv,
1869         .release_cb             = tcp_release_cb,
1870         .hash                   = inet6_hash,
1871         .unhash                 = inet_unhash,
1872         .get_port               = inet_csk_get_port,
1873         .enter_memory_pressure  = tcp_enter_memory_pressure,
1874         .stream_memory_free     = tcp_stream_memory_free,
1875         .sockets_allocated      = &tcp_sockets_allocated,
1876         .memory_allocated       = &tcp_memory_allocated,
1877         .memory_pressure        = &tcp_memory_pressure,
1878         .orphan_count           = &tcp_orphan_count,
1879         .sysctl_mem             = sysctl_tcp_mem,
1880         .sysctl_wmem            = sysctl_tcp_wmem,
1881         .sysctl_rmem            = sysctl_tcp_rmem,
1882         .max_header             = MAX_TCP_HEADER,
1883         .obj_size               = sizeof(struct tcp6_sock),
1884         .slab_flags             = SLAB_DESTROY_BY_RCU,
1885         .twsk_prot              = &tcp6_timewait_sock_ops,
1886         .rsk_prot               = &tcp6_request_sock_ops,
1887         .h.hashinfo             = &tcp_hashinfo,
1888         .no_autobind            = true,
1889 #ifdef CONFIG_COMPAT
1890         .compat_setsockopt      = compat_tcp_setsockopt,
1891         .compat_getsockopt      = compat_tcp_getsockopt,
1892 #endif
1893         .clear_sk               = tcp_v6_clear_sk,
1894         .diag_destroy           = tcp_abort,
1895 };
1896
1897 static const struct inet6_protocol tcpv6_protocol = {
1898         .early_demux    =       tcp_v6_early_demux,
1899         .handler        =       tcp_v6_rcv,
1900         .err_handler    =       tcp_v6_err,
1901         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1902 };
1903
1904 static struct inet_protosw tcpv6_protosw = {
1905         .type           =       SOCK_STREAM,
1906         .protocol       =       IPPROTO_TCP,
1907         .prot           =       &tcpv6_prot,
1908         .ops            =       &inet6_stream_ops,
1909         .flags          =       INET_PROTOSW_PERMANENT |
1910                                 INET_PROTOSW_ICSK,
1911 };
1912
1913 static int __net_init tcpv6_net_init(struct net *net)
1914 {
1915         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1916                                     SOCK_RAW, IPPROTO_TCP, net);
1917 }
1918
1919 static void __net_exit tcpv6_net_exit(struct net *net)
1920 {
1921         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1922 }
1923
1924 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1925 {
1926         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1927 }
1928
1929 static struct pernet_operations tcpv6_net_ops = {
1930         .init       = tcpv6_net_init,
1931         .exit       = tcpv6_net_exit,
1932         .exit_batch = tcpv6_net_exit_batch,
1933 };
1934
1935 int __init tcpv6_init(void)
1936 {
1937         int ret;
1938
1939         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1940         if (ret)
1941                 goto out;
1942
1943         /* register inet6 protocol */
1944         ret = inet6_register_protosw(&tcpv6_protosw);
1945         if (ret)
1946                 goto out_tcpv6_protocol;
1947
1948         ret = register_pernet_subsys(&tcpv6_net_ops);
1949         if (ret)
1950                 goto out_tcpv6_protosw;
1951 out:
1952         return ret;
1953
1954 out_tcpv6_protosw:
1955         inet6_unregister_protosw(&tcpv6_protosw);
1956 out_tcpv6_protocol:
1957         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1958         goto out;
1959 }
1960
1961 void tcpv6_exit(void)
1962 {
1963         unregister_pernet_subsys(&tcpv6_net_ops);
1964         inet6_unregister_protosw(&tcpv6_protosw);
1965         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1966 }