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         bool fatal;
 331         int err;
 332
 333         sk = __inet6_lookup_established(net, &tcp_hashinfo,
 334                                         &hdr->daddr, th->dest,
 335                                         &hdr->saddr, ntohs(th->source),
 336                                         skb->dev->ifindex);
 337
 338         if (!sk) {
 339                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 340                                    ICMP6_MIB_INERRORS);
 341                 return;
 342         }
 343
 344         if (sk->sk_state == TCP_TIME_WAIT) {
 345                 inet_twsk_put(inet_twsk(sk));
 346                 return;
 347         }
 348         seq = ntohl(th->seq);
 349         fatal = icmpv6_err_convert(type, code, &err);
 350         if (sk->sk_state == TCP_NEW_SYN_RECV)
 351                 return tcp_req_err(sk, seq, fatal);
 352
 353         bh_lock_sock(sk);
 354         if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
 355                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 356
 357         if (sk->sk_state == TCP_CLOSE)
 358                 goto out;
 359
 360         if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
 361                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
 362                 goto out;
 363         }
 364
 365         tp = tcp_sk(sk);
 366         /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 367         fastopen = tp->fastopen_rsk;
 368         snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 369         if (sk->sk_state != TCP_LISTEN &&
 370             !between(seq, snd_una, tp->snd_nxt)) {
 371                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 372                 goto out;
 373         }
 374
 375         np = inet6_sk(sk);
 376
 377         if (type == NDISC_REDIRECT) {
 378                 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 379
 380                 if (dst)
 381                         dst->ops->redirect(dst, sk, skb);
 382                 goto out;
 383         }
 384
 385         if (type == ICMPV6_PKT_TOOBIG) {
 386                 /* We are not interested in TCP_LISTEN and open_requests
 387                  * (SYN-ACKs send out by Linux are always <576bytes so
 388                  * they should go through unfragmented).
 389                  */
 390                 if (sk->sk_state == TCP_LISTEN)
 391                         goto out;
 392
 393                 if (!ip6_sk_accept_pmtu(sk))
 394                         goto out;
 395
 396                 tp->mtu_info = ntohl(info);
 397                 if (!sock_owned_by_user(sk))
 398                         tcp_v6_mtu_reduced(sk);
 399                 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
 400                                            &tp->tsq_flags))
 401                         sock_hold(sk);
 402                 goto out;
 403         }
 404
 405
 406         /* Might be for an request_sock */
 407         switch (sk->sk_state) {
 408         case TCP_SYN_SENT:
 409         case TCP_SYN_RECV:
 410                 /* Only in fast or simultaneous open. If a fast open socket is
 411                  * is already accepted it is treated as a connected one below.
 412                  */
 413                 if (fastopen && !fastopen->sk)
 414                         break;
 415
 416                 if (!sock_owned_by_user(sk)) {
 417                         sk->sk_err = err;
 418                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 419
 420                         tcp_done(sk);
 421                 } else
 422                         sk->sk_err_soft = err;
 423                 goto out;
 424         }
 425
 426         if (!sock_owned_by_user(sk) && np->recverr) {
 427                 sk->sk_err = err;
 428                 sk->sk_error_report(sk);
 429         } else
 430                 sk->sk_err_soft = err;
 431
 432 out:
 433         bh_unlock_sock(sk);
 434         sock_put(sk);
 435 }
 436
 437
 438 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 439                               struct flowi *fl,
 440                               struct request_sock *req,
 441                               struct tcp_fastopen_cookie *foc,
 442                               bool attach_req)
 443 {
 444         struct inet_request_sock *ireq = inet_rsk(req);
 445         struct ipv6_pinfo *np = inet6_sk(sk);
 446         struct flowi6 *fl6 = &fl->u.ip6;
 447         struct sk_buff *skb;
 448         int err = -ENOMEM;
 449
 450         /* First, grab a route. */
 451         if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
 452                                                IPPROTO_TCP)) == NULL)
 453                 goto done;
 454
 455         skb = tcp_make_synack(sk, dst, req, foc, attach_req);
 456
 457         if (skb) {
 458                 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
 459                                     &ireq->ir_v6_rmt_addr);
 460
 461                 fl6->daddr = ireq->ir_v6_rmt_addr;
 462                 if (np->repflow && ireq->pktopts)
 463                         fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
 464
 465                 rcu_read_lock();
 466                 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
 467                                np->tclass);
 468                 rcu_read_unlock();
 469                 err = net_xmit_eval(err);
 470         }
 471
 472 done:
 473         return err;
 474 }
 475
 476
 477 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 478 {
 479         kfree_skb(inet_rsk(req)->pktopts);
 480 }
 481
 482 #ifdef CONFIG_TCP_MD5SIG
 483 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
 484                                                    const struct in6_addr *addr)
 485 {
 486         return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
 487 }
 488
 489 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
 490                                                 const struct sock *addr_sk)
 491 {
 492         return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
 493 }
 494
 495 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
 496                                  int optlen)
 497 {
 498         struct tcp_md5sig cmd;
 499         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 500
 501         if (optlen < sizeof(cmd))
 502                 return -EINVAL;
 503
 504         if (copy_from_user(&cmd, optval, sizeof(cmd)))
 505                 return -EFAULT;
 506
 507         if (sin6->sin6_family != AF_INET6)
 508                 return -EINVAL;
 509
 510         if (!cmd.tcpm_keylen) {
 511                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 512                         return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 513                                               AF_INET);
 514                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 515                                       AF_INET6);
 516         }
 517
 518         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 519                 return -EINVAL;
 520
 521         if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 522                 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 523                                       AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 524
 525         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 526                               AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 527 }
 528
 529 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
 530                                         const struct in6_addr *daddr,
 531                                         const struct in6_addr *saddr, int nbytes)
 532 {
 533         struct tcp6_pseudohdr *bp;
 534         struct scatterlist sg;
 535
 536         bp = &hp->md5_blk.ip6;
 537         /* 1. TCP pseudo-header (RFC2460) */
 538         bp->saddr = *saddr;
 539         bp->daddr = *daddr;
 540         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 541         bp->len = cpu_to_be32(nbytes);
 542
 543         sg_init_one(&sg, bp, sizeof(*bp));
 544         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
 545 }
 546
 547 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
 548                                const struct in6_addr *daddr, struct in6_addr *saddr,
 549                                const struct tcphdr *th)
 550 {
 551         struct tcp_md5sig_pool *hp;
 552         struct hash_desc *desc;
 553
 554         hp = tcp_get_md5sig_pool();
 555         if (!hp)
 556                 goto clear_hash_noput;
 557         desc = &hp->md5_desc;
 558
 559         if (crypto_hash_init(desc))
 560                 goto clear_hash;
 561         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
 562                 goto clear_hash;
 563         if (tcp_md5_hash_header(hp, th))
 564                 goto clear_hash;
 565         if (tcp_md5_hash_key(hp, key))
 566                 goto clear_hash;
 567         if (crypto_hash_final(desc, md5_hash))
 568                 goto clear_hash;
 569
 570         tcp_put_md5sig_pool();
 571         return 0;
 572
 573 clear_hash:
 574         tcp_put_md5sig_pool();
 575 clear_hash_noput:
 576         memset(md5_hash, 0, 16);
 577         return 1;
 578 }
 579
 580 static int tcp_v6_md5_hash_skb(char *md5_hash,
 581                                const struct tcp_md5sig_key *key,
 582                                const struct sock *sk,
 583                                const struct sk_buff *skb)
 584 {
 585         const struct in6_addr *saddr, *daddr;
 586         struct tcp_md5sig_pool *hp;
 587         struct hash_desc *desc;
 588         const struct tcphdr *th = tcp_hdr(skb);
 589
 590         if (sk) { /* valid for establish/request sockets */
 591                 saddr = &sk->sk_v6_rcv_saddr;
 592                 daddr = &sk->sk_v6_daddr;
 593         } else {
 594                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 595                 saddr = &ip6h->saddr;
 596                 daddr = &ip6h->daddr;
 597         }
 598
 599         hp = tcp_get_md5sig_pool();
 600         if (!hp)
 601                 goto clear_hash_noput;
 602         desc = &hp->md5_desc;
 603
 604         if (crypto_hash_init(desc))
 605                 goto clear_hash;
 606
 607         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
 608                 goto clear_hash;
 609         if (tcp_md5_hash_header(hp, th))
 610                 goto clear_hash;
 611         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 612                 goto clear_hash;
 613         if (tcp_md5_hash_key(hp, key))
 614                 goto clear_hash;
 615         if (crypto_hash_final(desc, md5_hash))
 616                 goto clear_hash;
 617
 618         tcp_put_md5sig_pool();
 619         return 0;
 620
 621 clear_hash:
 622         tcp_put_md5sig_pool();
 623 clear_hash_noput:
 624         memset(md5_hash, 0, 16);
 625         return 1;
 626 }
 627
 628 #endif
 629
 630 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
 631                                     const struct sk_buff *skb)
 632 {
 633 #ifdef CONFIG_TCP_MD5SIG
 634         const __u8 *hash_location = NULL;
 635         struct tcp_md5sig_key *hash_expected;
 636         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 637         const struct tcphdr *th = tcp_hdr(skb);
 638         int genhash;
 639         u8 newhash[16];
 640
 641         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 642         hash_location = tcp_parse_md5sig_option(th);
 643
 644         /* We've parsed the options - do we have a hash? */
 645         if (!hash_expected && !hash_location)
 646                 return false;
 647
 648         if (hash_expected && !hash_location) {
 649                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 650                 return true;
 651         }
 652
 653         if (!hash_expected && hash_location) {
 654                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 655                 return true;
 656         }
 657
 658         /* check the signature */
 659         genhash = tcp_v6_md5_hash_skb(newhash,
 660                                       hash_expected,
 661                                       NULL, skb);
 662
 663         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 664                 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
 665                                      genhash ? "failed" : "mismatch",
 666                                      &ip6h->saddr, ntohs(th->source),
 667                                      &ip6h->daddr, ntohs(th->dest));
 668                 return true;
 669         }
 670 #endif
 671         return false;
 672 }
 673
 674 static void tcp_v6_init_req(struct request_sock *req,
 675                             const struct sock *sk_listener,
 676                             struct sk_buff *skb)
 677 {
 678         struct inet_request_sock *ireq = inet_rsk(req);
 679         const struct ipv6_pinfo *np = inet6_sk(sk_listener);
 680
 681         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 682         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 683
 684         /* So that link locals have meaning */
 685         if (!sk_listener->sk_bound_dev_if &&
 686             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 687                 ireq->ir_iif = tcp_v6_iif(skb);
 688
 689         if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
 690             (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
 691              np->rxopt.bits.rxinfo ||
 692              np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
 693              np->rxopt.bits.rxohlim || np->repflow)) {
 694                 atomic_inc(&skb->users);
 695                 ireq->pktopts = skb;
 696         }
 697 }
 698
 699 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
 700                                           struct flowi *fl,
 701                                           const struct request_sock *req,
 702                                           bool *strict)
 703 {
 704         if (strict)
 705                 *strict = true;
 706         return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
 707 }
 708
 709 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 710         .family         =       AF_INET6,
 711         .obj_size       =       sizeof(struct tcp6_request_sock),
 712         .rtx_syn_ack    =       tcp_rtx_synack,
 713         .send_ack       =       tcp_v6_reqsk_send_ack,
 714         .destructor     =       tcp_v6_reqsk_destructor,
 715         .send_reset     =       tcp_v6_send_reset,
 716         .syn_ack_timeout =      tcp_syn_ack_timeout,
 717 };
 718
 719 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 720         .mss_clamp      =       IPV6_MIN_MTU - sizeof(struct tcphdr) -
 721                                 sizeof(struct ipv6hdr),
 722 #ifdef CONFIG_TCP_MD5SIG
 723         .req_md5_lookup =       tcp_v6_md5_lookup,
 724         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 725 #endif
 726         .init_req       =       tcp_v6_init_req,
 727 #ifdef CONFIG_SYN_COOKIES
 728         .cookie_init_seq =      cookie_v6_init_sequence,
 729 #endif
 730         .route_req      =       tcp_v6_route_req,
 731         .init_seq       =       tcp_v6_init_sequence,
 732         .send_synack    =       tcp_v6_send_synack,
 733 };
 734
 735 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
 736                                  u32 ack, u32 win, u32 tsval, u32 tsecr,
 737                                  int oif, struct tcp_md5sig_key *key, int rst,
 738                                  u8 tclass, u32 label)
 739 {
 740         const struct tcphdr *th = tcp_hdr(skb);
 741         struct tcphdr *t1;
 742         struct sk_buff *buff;
 743         struct flowi6 fl6;
 744         struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 745         struct sock *ctl_sk = net->ipv6.tcp_sk;
 746         unsigned int tot_len = sizeof(struct tcphdr);
 747         struct dst_entry *dst;
 748         __be32 *topt;
 749
 750         if (tsecr)
 751                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 752 #ifdef CONFIG_TCP_MD5SIG
 753         if (key)
 754                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 755 #endif
 756
 757         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 758                          GFP_ATOMIC);
 759         if (!buff)
 760                 return;
 761
 762         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 763
 764         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 765         skb_reset_transport_header(buff);
 766
 767         /* Swap the send and the receive. */
 768         memset(t1, 0, sizeof(*t1));
 769         t1->dest = th->source;
 770         t1->source = th->dest;
 771         t1->doff = tot_len / 4;
 772         t1->seq = htonl(seq);
 773         t1->ack_seq = htonl(ack);
 774         t1->ack = !rst || !th->ack;
 775         t1->rst = rst;
 776         t1->window = htons(win);
 777
 778         topt = (__be32 *)(t1 + 1);
 779
 780         if (tsecr) {
 781                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 782                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 783                 *topt++ = htonl(tsval);
 784                 *topt++ = htonl(tsecr);
 785         }
 786
 787 #ifdef CONFIG_TCP_MD5SIG
 788         if (key) {
 789                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 790                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 791                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 792                                     &ipv6_hdr(skb)->saddr,
 793                                     &ipv6_hdr(skb)->daddr, t1);
 794         }
 795 #endif
 796
 797         memset(&fl6, 0, sizeof(fl6));
 798         fl6.daddr = ipv6_hdr(skb)->saddr;
 799         fl6.saddr = ipv6_hdr(skb)->daddr;
 800         fl6.flowlabel = label;
 801
 802         buff->ip_summed = CHECKSUM_PARTIAL;
 803         buff->csum = 0;
 804
 805         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
 806
 807         fl6.flowi6_proto = IPPROTO_TCP;
 808         if (rt6_need_strict(&fl6.daddr) && !oif)
 809                 fl6.flowi6_oif = tcp_v6_iif(skb);
 810         else
 811                 fl6.flowi6_oif = oif;
 812         fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
 813         fl6.fl6_dport = t1->dest;
 814         fl6.fl6_sport = t1->source;
 815         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 816
 817         /* Pass a socket to ip6_dst_lookup either it is for RST
 818          * Underlying function will use this to retrieve the network
 819          * namespace
 820          */
 821         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 822         if (!IS_ERR(dst)) {
 823                 skb_dst_set(buff, dst);
 824                 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
 825                 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 826                 if (rst)
 827                         TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
 828                 return;
 829         }
 830
 831         kfree_skb(buff);
 832 }
 833
 834 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
 835 {
 836         const struct tcphdr *th = tcp_hdr(skb);
 837         u32 seq = 0, ack_seq = 0;
 838         struct tcp_md5sig_key *key = NULL;
 839 #ifdef CONFIG_TCP_MD5SIG
 840         const __u8 *hash_location = NULL;
 841         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 842         unsigned char newhash[16];
 843         int genhash;
 844         struct sock *sk1 = NULL;
 845 #endif
 846         int oif;
 847
 848         if (th->rst)
 849                 return;
 850
 851         /* If sk not NULL, it means we did a successful lookup and incoming
 852          * route had to be correct. prequeue might have dropped our dst.
 853          */
 854         if (!sk && !ipv6_unicast_destination(skb))
 855                 return;
 856
 857 #ifdef CONFIG_TCP_MD5SIG
 858         hash_location = tcp_parse_md5sig_option(th);
 859         if (sk && sk_fullsock(sk)) {
 860                 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
 861         } else if (hash_location) {
 862                 /*
 863                  * active side is lost. Try to find listening socket through
 864                  * source port, and then find md5 key through listening socket.
 865                  * we are not loose security here:
 866                  * Incoming packet is checked with md5 hash with finding key,
 867                  * no RST generated if md5 hash doesn't match.
 868                  */
 869                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
 870                                            &tcp_hashinfo, NULL, 0,
 871                                            &ipv6h->saddr,
 872                                            th->source, &ipv6h->daddr,
 873                                            ntohs(th->source), tcp_v6_iif(skb));
 874                 if (!sk1)
 875                         return;
 876
 877                 rcu_read_lock();
 878                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 879                 if (!key)
 880                         goto release_sk1;
 881
 882                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
 883                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
 884                         goto release_sk1;
 885         }
 886 #endif
 887
 888         if (th->ack)
 889                 seq = ntohl(th->ack_seq);
 890         else
 891                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
 892                           (th->doff << 2);
 893
 894         oif = sk ? sk->sk_bound_dev_if : 0;
 895         tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
 896
 897 #ifdef CONFIG_TCP_MD5SIG
 898 release_sk1:
 899         if (sk1) {
 900                 rcu_read_unlock();
 901                 sock_put(sk1);
 902         }
 903 #endif
 904 }
 905
 906 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
 907                             u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
 908                             struct tcp_md5sig_key *key, u8 tclass,
 909                             u32 label)
 910 {
 911         tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
 912                              tclass, label);
 913 }
 914
 915 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 916 {
 917         struct inet_timewait_sock *tw = inet_twsk(sk);
 918         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 919
 920         tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 921                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 922                         tcp_time_stamp + tcptw->tw_ts_offset,
 923                         tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
 924                         tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
 925
 926         inet_twsk_put(tw);
 927 }
 928
 929 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 930                                   struct request_sock *req)
 931 {
 932         /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 933          * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 934          */
 935         tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
 936                         tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
 937                         tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
 938                         tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
 939                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
 940                         0, 0);
 941 }
 942
 943
 944 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
 945 {
 946 #ifdef CONFIG_SYN_COOKIES
 947         const struct tcphdr *th = tcp_hdr(skb);
 948
 949         if (!th->syn)
 950                 sk = cookie_v6_check(sk, skb);
 951 #endif
 952         return sk;
 953 }
 954
 955 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 956 {
 957         if (skb->protocol == htons(ETH_P_IP))
 958                 return tcp_v4_conn_request(sk, skb);
 959
 960         if (!ipv6_unicast_destination(skb))
 961                 goto drop;
 962
 963         return tcp_conn_request(&tcp6_request_sock_ops,
 964                                 &tcp_request_sock_ipv6_ops, sk, skb);
 965
 966 drop:
 967         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 968         return 0; /* don't send reset */
 969 }
 970
 971 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
 972                                          struct request_sock *req,
 973                                          struct dst_entry *dst,
 974                                          struct request_sock *req_unhash,
 975                                          bool *own_req)
 976 {
 977         struct inet_request_sock *ireq;
 978         struct ipv6_pinfo *newnp;
 979         const struct ipv6_pinfo *np = inet6_sk(sk);
 980         struct ipv6_txoptions *opt;
 981         struct tcp6_sock *newtcp6sk;
 982         struct inet_sock *newinet;
 983         struct tcp_sock *newtp;
 984         struct sock *newsk;
 985 #ifdef CONFIG_TCP_MD5SIG
 986         struct tcp_md5sig_key *key;
 987 #endif
 988         struct flowi6 fl6;
 989
 990         if (skb->protocol == htons(ETH_P_IP)) {
 991                 /*
 992                  *      v6 mapped
 993                  */
 994
 995                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
 996                                              req_unhash, own_req);
 997
 998                 if (!newsk)
 999                         return NULL;
1000
1001                 newtcp6sk = (struct tcp6_sock *)newsk;
1002                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1003
1004                 newinet = inet_sk(newsk);
1005                 newnp = inet6_sk(newsk);
1006                 newtp = tcp_sk(newsk);
1007
1008                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1009
1010                 newnp->saddr = newsk->sk_v6_rcv_saddr;
1011
1012                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1013                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1014 #ifdef CONFIG_TCP_MD5SIG
1015                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1016 #endif
1017
1018                 newnp->ipv6_ac_list = NULL;
1019                 newnp->ipv6_fl_list = NULL;
1020                 newnp->pktoptions  = NULL;
1021                 newnp->opt         = NULL;
1022                 newnp->mcast_oif   = tcp_v6_iif(skb);
1023                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1024                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1025                 if (np->repflow)
1026                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1027
1028                 /*
1029                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1030                  * here, tcp_create_openreq_child now does this for us, see the comment in
1031                  * that function for the gory details. -acme
1032                  */
1033
1034                 /* It is tricky place. Until this moment IPv4 tcp
1035                    worked with IPv6 icsk.icsk_af_ops.
1036                    Sync it now.
1037                  */
1038                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1039
1040                 return newsk;
1041         }
1042
1043         ireq = inet_rsk(req);
1044
1045         if (sk_acceptq_is_full(sk))
1046                 goto out_overflow;
1047
1048         if (!dst) {
1049                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1050                 if (!dst)
1051                         goto out;
1052         }
1053
1054         newsk = tcp_create_openreq_child(sk, req, skb);
1055         if (!newsk)
1056                 goto out_nonewsk;
1057
1058         /*
1059          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1060          * count here, tcp_create_openreq_child now does this for us, see the
1061          * comment in that function for the gory details. -acme
1062          */
1063
1064         newsk->sk_gso_type = SKB_GSO_TCPV6;
1065         ip6_dst_store(newsk, dst, NULL, NULL);
1066         inet6_sk_rx_dst_set(newsk, skb);
1067
1068         newtcp6sk = (struct tcp6_sock *)newsk;
1069         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1070
1071         newtp = tcp_sk(newsk);
1072         newinet = inet_sk(newsk);
1073         newnp = inet6_sk(newsk);
1074
1075         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1076
1077         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1078         newnp->saddr = ireq->ir_v6_loc_addr;
1079         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1080         newsk->sk_bound_dev_if = ireq->ir_iif;
1081
1082         /* Now IPv6 options...
1083
1084            First: no IPv4 options.
1085          */
1086         newinet->inet_opt = NULL;
1087         newnp->ipv6_ac_list = NULL;
1088         newnp->ipv6_fl_list = NULL;
1089
1090         /* Clone RX bits */
1091         newnp->rxopt.all = np->rxopt.all;
1092
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         /* Clone native IPv6 options from listening socket (if any)
1102
1103            Yes, keeping reference count would be much more clever,
1104            but we make one more one thing there: reattach optmem
1105            to newsk.
1106          */
1107         opt = rcu_dereference(np->opt);
1108         if (opt) {
1109                 opt = ipv6_dup_options(newsk, opt);
1110                 RCU_INIT_POINTER(newnp->opt, opt);
1111         }
1112         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1113         if (opt)
1114                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1115                                                     opt->opt_flen;
1116
1117         tcp_ca_openreq_child(newsk, dst);
1118
1119         tcp_sync_mss(newsk, dst_mtu(dst));
1120         newtp->advmss = dst_metric_advmss(dst);
1121         if (tcp_sk(sk)->rx_opt.user_mss &&
1122             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1123                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1124
1125         tcp_initialize_rcv_mss(newsk);
1126
1127         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1128         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1129
1130 #ifdef CONFIG_TCP_MD5SIG
1131         /* Copy over the MD5 key from the original socket */
1132         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1133         if (key) {
1134                 /* We're using one, so create a matching key
1135                  * on the newsk structure. If we fail to get
1136                  * memory, then we end up not copying the key
1137                  * across. Shucks.
1138                  */
1139                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1140                                AF_INET6, key->key, key->keylen,
1141                                sk_gfp_mask(sk, GFP_ATOMIC));
1142         }
1143 #endif
1144
1145         if (__inet_inherit_port(sk, newsk) < 0) {
1146                 inet_csk_prepare_forced_close(newsk);
1147                 tcp_done(newsk);
1148                 goto out;
1149         }
1150         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1151         if (*own_req) {
1152                 tcp_move_syn(newtp, req);
1153
1154                 /* Clone pktoptions received with SYN, if we own the req */
1155                 if (ireq->pktopts) {
1156                         newnp->pktoptions = skb_clone(ireq->pktopts,
1157                                                       sk_gfp_mask(sk, GFP_ATOMIC));
1158                         consume_skb(ireq->pktopts);
1159                         ireq->pktopts = NULL;
1160                         if (newnp->pktoptions)
1161                                 skb_set_owner_r(newnp->pktoptions, newsk);
1162                 }
1163         }
1164
1165         return newsk;
1166
1167 out_overflow:
1168         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1169 out_nonewsk:
1170         dst_release(dst);
1171 out:
1172         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1173         return NULL;
1174 }
1175
1176 /* The socket must have it's spinlock held when we get
1177  * here, unless it is a TCP_LISTEN socket.
1178  *
1179  * We have a potential double-lock case here, so even when
1180  * doing backlog processing we use the BH locking scheme.
1181  * This is because we cannot sleep with the original spinlock
1182  * held.
1183  */
1184 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1185 {
1186         struct ipv6_pinfo *np = inet6_sk(sk);
1187         struct tcp_sock *tp;
1188         struct sk_buff *opt_skb = NULL;
1189
1190         /* Imagine: socket is IPv6. IPv4 packet arrives,
1191            goes to IPv4 receive handler and backlogged.
1192            From backlog it always goes here. Kerboom...
1193            Fortunately, tcp_rcv_established and rcv_established
1194            handle them correctly, but it is not case with
1195            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1196          */
1197
1198         if (skb->protocol == htons(ETH_P_IP))
1199                 return tcp_v4_do_rcv(sk, skb);
1200
1201         if (sk_filter(sk, skb))
1202                 goto discard;
1203
1204         /*
1205          *      socket locking is here for SMP purposes as backlog rcv
1206          *      is currently called with bh processing disabled.
1207          */
1208
1209         /* Do Stevens' IPV6_PKTOPTIONS.
1210
1211            Yes, guys, it is the only place in our code, where we
1212            may make it not affecting IPv4.
1213            The rest of code is protocol independent,
1214            and I do not like idea to uglify IPv4.
1215
1216            Actually, all the idea behind IPV6_PKTOPTIONS
1217            looks not very well thought. For now we latch
1218            options, received in the last packet, enqueued
1219            by tcp. Feel free to propose better solution.
1220                                                --ANK (980728)
1221          */
1222         if (np->rxopt.all)
1223                 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1224
1225         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1226                 struct dst_entry *dst = sk->sk_rx_dst;
1227
1228                 sock_rps_save_rxhash(sk, skb);
1229                 sk_mark_napi_id(sk, skb);
1230                 if (dst) {
1231                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1232                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1233                                 dst_release(dst);
1234                                 sk->sk_rx_dst = NULL;
1235                         }
1236                 }
1237
1238                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1239                 if (opt_skb)
1240                         goto ipv6_pktoptions;
1241                 return 0;
1242         }
1243
1244         if (tcp_checksum_complete(skb))
1245                 goto csum_err;
1246
1247         if (sk->sk_state == TCP_LISTEN) {
1248                 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1249
1250                 if (!nsk)
1251                         goto discard;
1252
1253                 if (nsk != sk) {
1254                         sock_rps_save_rxhash(nsk, skb);
1255                         sk_mark_napi_id(nsk, skb);
1256                         if (tcp_child_process(sk, nsk, skb))
1257                                 goto reset;
1258                         if (opt_skb)
1259                                 __kfree_skb(opt_skb);
1260                         return 0;
1261                 }
1262         } else
1263                 sock_rps_save_rxhash(sk, skb);
1264
1265         if (tcp_rcv_state_process(sk, skb))
1266                 goto reset;
1267         if (opt_skb)
1268                 goto ipv6_pktoptions;
1269         return 0;
1270
1271 reset:
1272         tcp_v6_send_reset(sk, skb);
1273 discard:
1274         if (opt_skb)
1275                 __kfree_skb(opt_skb);
1276         kfree_skb(skb);
1277         return 0;
1278 csum_err:
1279         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1280         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1281         goto discard;
1282
1283
1284 ipv6_pktoptions:
1285         /* Do you ask, what is it?
1286
1287            1. skb was enqueued by tcp.
1288            2. skb is added to tail of read queue, rather than out of order.
1289            3. socket is not in passive state.
1290            4. Finally, it really contains options, which user wants to receive.
1291          */
1292         tp = tcp_sk(sk);
1293         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1294             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1295                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1296                         np->mcast_oif = tcp_v6_iif(opt_skb);
1297                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1298                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1299                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1300                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1301                 if (np->repflow)
1302                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1303                 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1304                         skb_set_owner_r(opt_skb, sk);
1305                         opt_skb = xchg(&np->pktoptions, opt_skb);
1306                 } else {
1307                         __kfree_skb(opt_skb);
1308                         opt_skb = xchg(&np->pktoptions, NULL);
1309                 }
1310         }
1311
1312         kfree_skb(opt_skb);
1313         return 0;
1314 }
1315
1316 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1317                            const struct tcphdr *th)
1318 {
1319         /* This is tricky: we move IP6CB at its correct location into
1320          * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1321          * _decode_session6() uses IP6CB().
1322          * barrier() makes sure compiler won't play aliasing games.
1323          */
1324         memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1325                 sizeof(struct inet6_skb_parm));
1326         barrier();
1327
1328         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1329         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1330                                     skb->len - th->doff*4);
1331         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1332         TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1333         TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1334         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1335         TCP_SKB_CB(skb)->sacked = 0;
1336 }
1337
1338 static void tcp_v6_restore_cb(struct sk_buff *skb)
1339 {
1340         /* We need to move header back to the beginning if xfrm6_policy_check()
1341          * and tcp_v6_fill_cb() are going to be called again.
1342          */
1343         memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1344                 sizeof(struct inet6_skb_parm));
1345 }
1346
1347 static int tcp_v6_rcv(struct sk_buff *skb)
1348 {
1349         const struct tcphdr *th;
1350         const struct ipv6hdr *hdr;
1351         struct sock *sk;
1352         int ret;
1353         struct net *net = dev_net(skb->dev);
1354
1355         if (skb->pkt_type != PACKET_HOST)
1356                 goto discard_it;
1357
1358         /*
1359          *      Count it even if it's bad.
1360          */
1361         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1362
1363         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1364                 goto discard_it;
1365
1366         th = tcp_hdr(skb);
1367
1368         if (th->doff < sizeof(struct tcphdr)/4)
1369                 goto bad_packet;
1370         if (!pskb_may_pull(skb, th->doff*4))
1371                 goto discard_it;
1372
1373         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1374                 goto csum_error;
1375
1376         th = tcp_hdr(skb);
1377         hdr = ipv6_hdr(skb);
1378
1379 lookup:
1380         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1381                                 th->source, th->dest, inet6_iif(skb));
1382         if (!sk)
1383                 goto no_tcp_socket;
1384
1385 process:
1386         if (sk->sk_state == TCP_TIME_WAIT)
1387                 goto do_time_wait;
1388
1389         if (sk->sk_state == TCP_NEW_SYN_RECV) {
1390                 struct request_sock *req = inet_reqsk(sk);
1391                 struct sock *nsk;
1392
1393                 sk = req->rsk_listener;
1394                 tcp_v6_fill_cb(skb, hdr, th);
1395                 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1396                         reqsk_put(req);
1397                         goto discard_it;
1398                 }
1399                 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1400                         inet_csk_reqsk_queue_drop_and_put(sk, req);
1401                         goto lookup;
1402                 }
1403                 sock_hold(sk);
1404                 nsk = tcp_check_req(sk, skb, req, false);
1405                 if (!nsk) {
1406                         reqsk_put(req);
1407                         goto discard_and_relse;
1408                 }
1409                 if (nsk == 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_and_relse;
1415                 } else {
1416                         sock_put(sk);
1417                         return 0;
1418                 }
1419         }
1420         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1421                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1422                 goto discard_and_relse;
1423         }
1424
1425         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1426                 goto discard_and_relse;
1427
1428         tcp_v6_fill_cb(skb, hdr, th);
1429
1430         if (tcp_v6_inbound_md5_hash(sk, skb))
1431                 goto discard_and_relse;
1432
1433         if (sk_filter(sk, skb))
1434                 goto discard_and_relse;
1435
1436         skb->dev = NULL;
1437
1438         if (sk->sk_state == TCP_LISTEN) {
1439                 ret = tcp_v6_do_rcv(sk, skb);
1440                 goto put_and_return;
1441         }
1442
1443         sk_incoming_cpu_update(sk);
1444
1445         bh_lock_sock_nested(sk);
1446         tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1447         ret = 0;
1448         if (!sock_owned_by_user(sk)) {
1449                 if (!tcp_prequeue(sk, skb))
1450                         ret = tcp_v6_do_rcv(sk, skb);
1451         } else if (unlikely(sk_add_backlog(sk, skb,
1452                                            sk->sk_rcvbuf + sk->sk_sndbuf))) {
1453                 bh_unlock_sock(sk);
1454                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1455                 goto discard_and_relse;
1456         }
1457         bh_unlock_sock(sk);
1458
1459 put_and_return:
1460         sock_put(sk);
1461         return ret ? -1 : 0;
1462
1463 no_tcp_socket:
1464         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1465                 goto discard_it;
1466
1467         tcp_v6_fill_cb(skb, hdr, th);
1468
1469         if (tcp_checksum_complete(skb)) {
1470 csum_error:
1471                 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1472 bad_packet:
1473                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1474         } else {
1475                 tcp_v6_send_reset(NULL, skb);
1476         }
1477
1478 discard_it:
1479         kfree_skb(skb);
1480         return 0;
1481
1482 discard_and_relse:
1483         sock_put(sk);
1484         goto discard_it;
1485
1486 do_time_wait:
1487         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1488                 inet_twsk_put(inet_twsk(sk));
1489                 goto discard_it;
1490         }
1491
1492         tcp_v6_fill_cb(skb, hdr, th);
1493
1494         if (tcp_checksum_complete(skb)) {
1495                 inet_twsk_put(inet_twsk(sk));
1496                 goto csum_error;
1497         }
1498
1499         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1500         case TCP_TW_SYN:
1501         {
1502                 struct sock *sk2;
1503
1504                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1505                                             skb, __tcp_hdrlen(th),
1506                                             &ipv6_hdr(skb)->saddr, th->source,
1507                                             &ipv6_hdr(skb)->daddr,
1508                                             ntohs(th->dest), tcp_v6_iif(skb));
1509                 if (sk2) {
1510                         struct inet_timewait_sock *tw = inet_twsk(sk);
1511                         inet_twsk_deschedule_put(tw);
1512                         sk = sk2;
1513                         tcp_v6_restore_cb(skb);
1514                         goto process;
1515                 }
1516                 /* Fall through to ACK */
1517         }
1518         case TCP_TW_ACK:
1519                 tcp_v6_timewait_ack(sk, skb);
1520                 break;
1521         case TCP_TW_RST:
1522                 tcp_v6_restore_cb(skb);
1523                 tcp_v6_send_reset(sk, skb);
1524                 inet_twsk_deschedule_put(inet_twsk(sk));
1525                 goto discard_it;
1526         case TCP_TW_SUCCESS:
1527                 ;
1528         }
1529         goto discard_it;
1530 }
1531
1532 static void tcp_v6_early_demux(struct sk_buff *skb)
1533 {
1534         const struct ipv6hdr *hdr;
1535         const struct tcphdr *th;
1536         struct sock *sk;
1537
1538         if (skb->pkt_type != PACKET_HOST)
1539                 return;
1540
1541         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1542                 return;
1543
1544         hdr = ipv6_hdr(skb);
1545         th = tcp_hdr(skb);
1546
1547         if (th->doff < sizeof(struct tcphdr) / 4)
1548                 return;
1549
1550         /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1551         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1552                                         &hdr->saddr, th->source,
1553                                         &hdr->daddr, ntohs(th->dest),
1554                                         inet6_iif(skb));
1555         if (sk) {
1556                 skb->sk = sk;
1557                 skb->destructor = sock_edemux;
1558                 if (sk_fullsock(sk)) {
1559                         struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1560
1561                         if (dst)
1562                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1563                         if (dst &&
1564                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1565                                 skb_dst_set_noref(skb, dst);
1566                 }
1567         }
1568 }
1569
1570 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1571         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1572         .twsk_unique    = tcp_twsk_unique,
1573         .twsk_destructor = tcp_twsk_destructor,
1574 };
1575
1576 static const struct inet_connection_sock_af_ops ipv6_specific = {
1577         .queue_xmit        = inet6_csk_xmit,
1578         .send_check        = tcp_v6_send_check,
1579         .rebuild_header    = inet6_sk_rebuild_header,
1580         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1581         .conn_request      = tcp_v6_conn_request,
1582         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1583         .net_header_len    = sizeof(struct ipv6hdr),
1584         .net_frag_header_len = sizeof(struct frag_hdr),
1585         .setsockopt        = ipv6_setsockopt,
1586         .getsockopt        = ipv6_getsockopt,
1587         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1588         .sockaddr_len      = sizeof(struct sockaddr_in6),
1589         .bind_conflict     = inet6_csk_bind_conflict,
1590 #ifdef CONFIG_COMPAT
1591         .compat_setsockopt = compat_ipv6_setsockopt,
1592         .compat_getsockopt = compat_ipv6_getsockopt,
1593 #endif
1594         .mtu_reduced       = tcp_v6_mtu_reduced,
1595 };
1596
1597 #ifdef CONFIG_TCP_MD5SIG
1598 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1599         .md5_lookup     =       tcp_v6_md5_lookup,
1600         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1601         .md5_parse      =       tcp_v6_parse_md5_keys,
1602 };
1603 #endif
1604
1605 /*
1606  *      TCP over IPv4 via INET6 API
1607  */
1608 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1609         .queue_xmit        = ip_queue_xmit,
1610         .send_check        = tcp_v4_send_check,
1611         .rebuild_header    = inet_sk_rebuild_header,
1612         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1613         .conn_request      = tcp_v6_conn_request,
1614         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1615         .net_header_len    = sizeof(struct iphdr),
1616         .setsockopt        = ipv6_setsockopt,
1617         .getsockopt        = ipv6_getsockopt,
1618         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1619         .sockaddr_len      = sizeof(struct sockaddr_in6),
1620         .bind_conflict     = inet6_csk_bind_conflict,
1621 #ifdef CONFIG_COMPAT
1622         .compat_setsockopt = compat_ipv6_setsockopt,
1623         .compat_getsockopt = compat_ipv6_getsockopt,
1624 #endif
1625         .mtu_reduced       = tcp_v4_mtu_reduced,
1626 };
1627
1628 #ifdef CONFIG_TCP_MD5SIG
1629 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1630         .md5_lookup     =       tcp_v4_md5_lookup,
1631         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1632         .md5_parse      =       tcp_v6_parse_md5_keys,
1633 };
1634 #endif
1635
1636 /* NOTE: A lot of things set to zero explicitly by call to
1637  *       sk_alloc() so need not be done here.
1638  */
1639 static int tcp_v6_init_sock(struct sock *sk)
1640 {
1641         struct inet_connection_sock *icsk = inet_csk(sk);
1642
1643         tcp_init_sock(sk);
1644
1645         icsk->icsk_af_ops = &ipv6_specific;
1646
1647 #ifdef CONFIG_TCP_MD5SIG
1648         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1649 #endif
1650
1651         return 0;
1652 }
1653
1654 static void tcp_v6_destroy_sock(struct sock *sk)
1655 {
1656         tcp_v4_destroy_sock(sk);
1657         inet6_destroy_sock(sk);
1658 }
1659
1660 #ifdef CONFIG_PROC_FS
1661 /* Proc filesystem TCPv6 sock list dumping. */
1662 static void get_openreq6(struct seq_file *seq,
1663                          const struct request_sock *req, int i)
1664 {
1665         long ttd = req->rsk_timer.expires - jiffies;
1666         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1667         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1668
1669         if (ttd < 0)
1670                 ttd = 0;
1671
1672         seq_printf(seq,
1673                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1674                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1675                    i,
1676                    src->s6_addr32[0], src->s6_addr32[1],
1677                    src->s6_addr32[2], src->s6_addr32[3],
1678                    inet_rsk(req)->ir_num,
1679                    dest->s6_addr32[0], dest->s6_addr32[1],
1680                    dest->s6_addr32[2], dest->s6_addr32[3],
1681                    ntohs(inet_rsk(req)->ir_rmt_port),
1682                    TCP_SYN_RECV,
1683                    0, 0, /* could print option size, but that is af dependent. */
1684                    1,   /* timers active (only the expire timer) */
1685                    jiffies_to_clock_t(ttd),
1686                    req->num_timeout,
1687                    from_kuid_munged(seq_user_ns(seq),
1688                                     sock_i_uid(req->rsk_listener)),
1689                    0,  /* non standard timer */
1690                    0, /* open_requests have no inode */
1691                    0, req);
1692 }
1693
1694 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1695 {
1696         const struct in6_addr *dest, *src;
1697         __u16 destp, srcp;
1698         int timer_active;
1699         unsigned long timer_expires;
1700         const struct inet_sock *inet = inet_sk(sp);
1701         const struct tcp_sock *tp = tcp_sk(sp);
1702         const struct inet_connection_sock *icsk = inet_csk(sp);
1703         const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1704         int rx_queue;
1705         int state;
1706
1707         dest  = &sp->sk_v6_daddr;
1708         src   = &sp->sk_v6_rcv_saddr;
1709         destp = ntohs(inet->inet_dport);
1710         srcp  = ntohs(inet->inet_sport);
1711
1712         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1713                 timer_active    = 1;
1714                 timer_expires   = icsk->icsk_timeout;
1715         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1716                 timer_active    = 4;
1717                 timer_expires   = icsk->icsk_timeout;
1718         } else if (timer_pending(&sp->sk_timer)) {
1719                 timer_active    = 2;
1720                 timer_expires   = sp->sk_timer.expires;
1721         } else {
1722                 timer_active    = 0;
1723                 timer_expires = jiffies;
1724         }
1725
1726         state = sk_state_load(sp);
1727         if (state == TCP_LISTEN)
1728                 rx_queue = sp->sk_ack_backlog;
1729         else
1730                 /* Because we don't lock the socket,
1731                  * we might find a transient negative value.
1732                  */
1733                 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1734
1735         seq_printf(seq,
1736                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1737                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1738                    i,
1739                    src->s6_addr32[0], src->s6_addr32[1],
1740                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1741                    dest->s6_addr32[0], dest->s6_addr32[1],
1742                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1743                    state,
1744                    tp->write_seq - tp->snd_una,
1745                    rx_queue,
1746                    timer_active,
1747                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1748                    icsk->icsk_retransmits,
1749                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1750                    icsk->icsk_probes_out,
1751                    sock_i_ino(sp),
1752                    atomic_read(&sp->sk_refcnt), sp,
1753                    jiffies_to_clock_t(icsk->icsk_rto),
1754                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1755                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1756                    tp->snd_cwnd,
1757                    state == TCP_LISTEN ?
1758                         fastopenq->max_qlen :
1759                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1760                    );
1761 }
1762
1763 static void get_timewait6_sock(struct seq_file *seq,
1764                                struct inet_timewait_sock *tw, int i)
1765 {
1766         long delta = tw->tw_timer.expires - jiffies;
1767         const struct in6_addr *dest, *src;
1768         __u16 destp, srcp;
1769
1770         dest = &tw->tw_v6_daddr;
1771         src  = &tw->tw_v6_rcv_saddr;
1772         destp = ntohs(tw->tw_dport);
1773         srcp  = ntohs(tw->tw_sport);
1774
1775         seq_printf(seq,
1776                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1777                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1778                    i,
1779                    src->s6_addr32[0], src->s6_addr32[1],
1780                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1781                    dest->s6_addr32[0], dest->s6_addr32[1],
1782                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1783                    tw->tw_substate, 0, 0,
1784                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1785                    atomic_read(&tw->tw_refcnt), tw);
1786 }
1787
1788 static int tcp6_seq_show(struct seq_file *seq, void *v)
1789 {
1790         struct tcp_iter_state *st;
1791         struct sock *sk = v;
1792
1793         if (v == SEQ_START_TOKEN) {
1794                 seq_puts(seq,
1795                          "  sl  "
1796                          "local_address                         "
1797                          "remote_address                        "
1798                          "st tx_queue rx_queue tr tm->when retrnsmt"
1799                          "   uid  timeout inode\n");
1800                 goto out;
1801         }
1802         st = seq->private;
1803
1804         if (sk->sk_state == TCP_TIME_WAIT)
1805                 get_timewait6_sock(seq, v, st->num);
1806         else if (sk->sk_state == TCP_NEW_SYN_RECV)
1807                 get_openreq6(seq, v, st->num);
1808         else
1809                 get_tcp6_sock(seq, v, st->num);
1810 out:
1811         return 0;
1812 }
1813
1814 static const struct file_operations tcp6_afinfo_seq_fops = {
1815         .owner   = THIS_MODULE,
1816         .open    = tcp_seq_open,
1817         .read    = seq_read,
1818         .llseek  = seq_lseek,
1819         .release = seq_release_net
1820 };
1821
1822 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1823         .name           = "tcp6",
1824         .family         = AF_INET6,
1825         .seq_fops       = &tcp6_afinfo_seq_fops,
1826         .seq_ops        = {
1827                 .show           = tcp6_seq_show,
1828         },
1829 };
1830
1831 int __net_init tcp6_proc_init(struct net *net)
1832 {
1833         return tcp_proc_register(net, &tcp6_seq_afinfo);
1834 }
1835
1836 void tcp6_proc_exit(struct net *net)
1837 {
1838         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1839 }
1840 #endif
1841
1842 static void tcp_v6_clear_sk(struct sock *sk, int size)
1843 {
1844         struct inet_sock *inet = inet_sk(sk);
1845
1846         /* we do not want to clear pinet6 field, because of RCU lookups */
1847         sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1848
1849         size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1850         memset(&inet->pinet6 + 1, 0, size);
1851 }
1852
1853 struct proto tcpv6_prot = {
1854         .name                   = "TCPv6",
1855         .owner                  = THIS_MODULE,
1856         .close                  = tcp_close,
1857         .connect                = tcp_v6_connect,
1858         .disconnect             = tcp_disconnect,
1859         .accept                 = inet_csk_accept,
1860         .ioctl                  = tcp_ioctl,
1861         .init                   = tcp_v6_init_sock,
1862         .destroy                = tcp_v6_destroy_sock,
1863         .shutdown               = tcp_shutdown,
1864         .setsockopt             = tcp_setsockopt,
1865         .getsockopt             = tcp_getsockopt,
1866         .recvmsg                = tcp_recvmsg,
1867         .sendmsg                = tcp_sendmsg,
1868         .sendpage               = tcp_sendpage,
1869         .backlog_rcv            = tcp_v6_do_rcv,
1870         .release_cb             = tcp_release_cb,
1871         .hash                   = inet6_hash,
1872         .unhash                 = inet_unhash,
1873         .get_port               = inet_csk_get_port,
1874         .enter_memory_pressure  = tcp_enter_memory_pressure,
1875         .stream_memory_free     = tcp_stream_memory_free,
1876         .sockets_allocated      = &tcp_sockets_allocated,
1877         .memory_allocated       = &tcp_memory_allocated,
1878         .memory_pressure        = &tcp_memory_pressure,
1879         .orphan_count           = &tcp_orphan_count,
1880         .sysctl_mem             = sysctl_tcp_mem,
1881         .sysctl_wmem            = sysctl_tcp_wmem,
1882         .sysctl_rmem            = sysctl_tcp_rmem,
1883         .max_header             = MAX_TCP_HEADER,
1884         .obj_size               = sizeof(struct tcp6_sock),
1885         .slab_flags             = SLAB_DESTROY_BY_RCU,
1886         .twsk_prot              = &tcp6_timewait_sock_ops,
1887         .rsk_prot               = &tcp6_request_sock_ops,
1888         .h.hashinfo             = &tcp_hashinfo,
1889         .no_autobind            = true,
1890 #ifdef CONFIG_COMPAT
1891         .compat_setsockopt      = compat_tcp_setsockopt,
1892         .compat_getsockopt      = compat_tcp_getsockopt,
1893 #endif
1894         .clear_sk               = tcp_v6_clear_sk,
1895         .diag_destroy           = tcp_abort,
1896 };
1897
1898 static const struct inet6_protocol tcpv6_protocol = {
1899         .early_demux    =       tcp_v6_early_demux,
1900         .handler        =       tcp_v6_rcv,
1901         .err_handler    =       tcp_v6_err,
1902         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1903 };
1904
1905 static struct inet_protosw tcpv6_protosw = {
1906         .type           =       SOCK_STREAM,
1907         .protocol       =       IPPROTO_TCP,
1908         .prot           =       &tcpv6_prot,
1909         .ops            =       &inet6_stream_ops,
1910         .flags          =       INET_PROTOSW_PERMANENT |
1911                                 INET_PROTOSW_ICSK,
1912 };
1913
1914 static int __net_init tcpv6_net_init(struct net *net)
1915 {
1916         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1917                                     SOCK_RAW, IPPROTO_TCP, net);
1918 }
1919
1920 static void __net_exit tcpv6_net_exit(struct net *net)
1921 {
1922         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1923 }
1924
1925 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1926 {
1927         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1928 }
1929
1930 static struct pernet_operations tcpv6_net_ops = {
1931         .init       = tcpv6_net_init,
1932         .exit       = tcpv6_net_exit,
1933         .exit_batch = tcpv6_net_exit_batch,
1934 };
1935
1936 int __init tcpv6_init(void)
1937 {
1938         int ret;
1939
1940         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1941         if (ret)
1942                 goto out;
1943
1944         /* register inet6 protocol */
1945         ret = inet6_register_protosw(&tcpv6_protosw);
1946         if (ret)
1947                 goto out_tcpv6_protocol;
1948
1949         ret = register_pernet_subsys(&tcpv6_net_ops);
1950         if (ret)
1951                 goto out_tcpv6_protosw;
1952 out:
1953         return ret;
1954
1955 out_tcpv6_protosw:
1956         inet6_unregister_protosw(&tcpv6_protosw);
1957 out_tcpv6_protocol:
1958         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1959         goto out;
1960 }
1961
1962 void tcpv6_exit(void)
1963 {
1964         unregister_pernet_subsys(&tcpv6_net_ops);
1965         inet6_unregister_protosw(&tcpv6_protosw);
1966         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1967 }