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 <crypto/hash.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, lockdep_sock_is_held(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(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(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(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(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                               enum tcp_synack_type synack_type)
 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, synack_type);
 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         ahash_request_set_crypt(hp->md5_req, &sg, NULL, sizeof(*bp));
 545         return crypto_ahash_update(hp->md5_req);
 546 }
 547
 548 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
 549                                const struct in6_addr *daddr, struct in6_addr *saddr,
 550                                const struct tcphdr *th)
 551 {
 552         struct tcp_md5sig_pool *hp;
 553         struct ahash_request *req;
 554
 555         hp = tcp_get_md5sig_pool();
 556         if (!hp)
 557                 goto clear_hash_noput;
 558         req = hp->md5_req;
 559
 560         if (crypto_ahash_init(req))
 561                 goto clear_hash;
 562         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
 563                 goto clear_hash;
 564         if (tcp_md5_hash_header(hp, th))
 565                 goto clear_hash;
 566         if (tcp_md5_hash_key(hp, key))
 567                 goto clear_hash;
 568         ahash_request_set_crypt(req, NULL, md5_hash, 0);
 569         if (crypto_ahash_final(req))
 570                 goto clear_hash;
 571
 572         tcp_put_md5sig_pool();
 573         return 0;
 574
 575 clear_hash:
 576         tcp_put_md5sig_pool();
 577 clear_hash_noput:
 578         memset(md5_hash, 0, 16);
 579         return 1;
 580 }
 581
 582 static int tcp_v6_md5_hash_skb(char *md5_hash,
 583                                const struct tcp_md5sig_key *key,
 584                                const struct sock *sk,
 585                                const struct sk_buff *skb)
 586 {
 587         const struct in6_addr *saddr, *daddr;
 588         struct tcp_md5sig_pool *hp;
 589         struct ahash_request *req;
 590         const struct tcphdr *th = tcp_hdr(skb);
 591
 592         if (sk) { /* valid for establish/request sockets */
 593                 saddr = &sk->sk_v6_rcv_saddr;
 594                 daddr = &sk->sk_v6_daddr;
 595         } else {
 596                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 597                 saddr = &ip6h->saddr;
 598                 daddr = &ip6h->daddr;
 599         }
 600
 601         hp = tcp_get_md5sig_pool();
 602         if (!hp)
 603                 goto clear_hash_noput;
 604         req = hp->md5_req;
 605
 606         if (crypto_ahash_init(req))
 607                 goto clear_hash;
 608
 609         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
 610                 goto clear_hash;
 611         if (tcp_md5_hash_header(hp, th))
 612                 goto clear_hash;
 613         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 614                 goto clear_hash;
 615         if (tcp_md5_hash_key(hp, key))
 616                 goto clear_hash;
 617         ahash_request_set_crypt(req, NULL, md5_hash, 0);
 618         if (crypto_ahash_final(req))
 619                 goto clear_hash;
 620
 621         tcp_put_md5sig_pool();
 622         return 0;
 623
 624 clear_hash:
 625         tcp_put_md5sig_pool();
 626 clear_hash_noput:
 627         memset(md5_hash, 0, 16);
 628         return 1;
 629 }
 630
 631 #endif
 632
 633 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
 634                                     const struct sk_buff *skb)
 635 {
 636 #ifdef CONFIG_TCP_MD5SIG
 637         const __u8 *hash_location = NULL;
 638         struct tcp_md5sig_key *hash_expected;
 639         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 640         const struct tcphdr *th = tcp_hdr(skb);
 641         int genhash;
 642         u8 newhash[16];
 643
 644         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 645         hash_location = tcp_parse_md5sig_option(th);
 646
 647         /* We've parsed the options - do we have a hash? */
 648         if (!hash_expected && !hash_location)
 649                 return false;
 650
 651         if (hash_expected && !hash_location) {
 652                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 653                 return true;
 654         }
 655
 656         if (!hash_expected && hash_location) {
 657                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 658                 return true;
 659         }
 660
 661         /* check the signature */
 662         genhash = tcp_v6_md5_hash_skb(newhash,
 663                                       hash_expected,
 664                                       NULL, skb);
 665
 666         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 667                 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
 668                                      genhash ? "failed" : "mismatch",
 669                                      &ip6h->saddr, ntohs(th->source),
 670                                      &ip6h->daddr, ntohs(th->dest));
 671                 return true;
 672         }
 673 #endif
 674         return false;
 675 }
 676
 677 static void tcp_v6_init_req(struct request_sock *req,
 678                             const struct sock *sk_listener,
 679                             struct sk_buff *skb)
 680 {
 681         struct inet_request_sock *ireq = inet_rsk(req);
 682         const struct ipv6_pinfo *np = inet6_sk(sk_listener);
 683
 684         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 685         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 686
 687         /* So that link locals have meaning */
 688         if (!sk_listener->sk_bound_dev_if &&
 689             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 690                 ireq->ir_iif = tcp_v6_iif(skb);
 691
 692         if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
 693             (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
 694              np->rxopt.bits.rxinfo ||
 695              np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
 696              np->rxopt.bits.rxohlim || np->repflow)) {
 697                 atomic_inc(&skb->users);
 698                 ireq->pktopts = skb;
 699         }
 700 }
 701
 702 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
 703                                           struct flowi *fl,
 704                                           const struct request_sock *req,
 705                                           bool *strict)
 706 {
 707         if (strict)
 708                 *strict = true;
 709         return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
 710 }
 711
 712 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 713         .family         =       AF_INET6,
 714         .obj_size       =       sizeof(struct tcp6_request_sock),
 715         .rtx_syn_ack    =       tcp_rtx_synack,
 716         .send_ack       =       tcp_v6_reqsk_send_ack,
 717         .destructor     =       tcp_v6_reqsk_destructor,
 718         .send_reset     =       tcp_v6_send_reset,
 719         .syn_ack_timeout =      tcp_syn_ack_timeout,
 720 };
 721
 722 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 723         .mss_clamp      =       IPV6_MIN_MTU - sizeof(struct tcphdr) -
 724                                 sizeof(struct ipv6hdr),
 725 #ifdef CONFIG_TCP_MD5SIG
 726         .req_md5_lookup =       tcp_v6_md5_lookup,
 727         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 728 #endif
 729         .init_req       =       tcp_v6_init_req,
 730 #ifdef CONFIG_SYN_COOKIES
 731         .cookie_init_seq =      cookie_v6_init_sequence,
 732 #endif
 733         .route_req      =       tcp_v6_route_req,
 734         .init_seq       =       tcp_v6_init_sequence,
 735         .send_synack    =       tcp_v6_send_synack,
 736 };
 737
 738 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
 739                                  u32 ack, u32 win, u32 tsval, u32 tsecr,
 740                                  int oif, struct tcp_md5sig_key *key, int rst,
 741                                  u8 tclass, u32 label)
 742 {
 743         const struct tcphdr *th = tcp_hdr(skb);
 744         struct tcphdr *t1;
 745         struct sk_buff *buff;
 746         struct flowi6 fl6;
 747         struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 748         struct sock *ctl_sk = net->ipv6.tcp_sk;
 749         unsigned int tot_len = sizeof(struct tcphdr);
 750         struct dst_entry *dst;
 751         __be32 *topt;
 752
 753         if (tsecr)
 754                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 755 #ifdef CONFIG_TCP_MD5SIG
 756         if (key)
 757                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 758 #endif
 759
 760         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 761                          GFP_ATOMIC);
 762         if (!buff)
 763                 return;
 764
 765         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 766
 767         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 768         skb_reset_transport_header(buff);
 769
 770         /* Swap the send and the receive. */
 771         memset(t1, 0, sizeof(*t1));
 772         t1->dest = th->source;
 773         t1->source = th->dest;
 774         t1->doff = tot_len / 4;
 775         t1->seq = htonl(seq);
 776         t1->ack_seq = htonl(ack);
 777         t1->ack = !rst || !th->ack;
 778         t1->rst = rst;
 779         t1->window = htons(win);
 780
 781         topt = (__be32 *)(t1 + 1);
 782
 783         if (tsecr) {
 784                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 785                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 786                 *topt++ = htonl(tsval);
 787                 *topt++ = htonl(tsecr);
 788         }
 789
 790 #ifdef CONFIG_TCP_MD5SIG
 791         if (key) {
 792                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 793                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 794                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 795                                     &ipv6_hdr(skb)->saddr,
 796                                     &ipv6_hdr(skb)->daddr, t1);
 797         }
 798 #endif
 799
 800         memset(&fl6, 0, sizeof(fl6));
 801         fl6.daddr = ipv6_hdr(skb)->saddr;
 802         fl6.saddr = ipv6_hdr(skb)->daddr;
 803         fl6.flowlabel = label;
 804
 805         buff->ip_summed = CHECKSUM_PARTIAL;
 806         buff->csum = 0;
 807
 808         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
 809
 810         fl6.flowi6_proto = IPPROTO_TCP;
 811         if (rt6_need_strict(&fl6.daddr) && !oif)
 812                 fl6.flowi6_oif = tcp_v6_iif(skb);
 813         else {
 814                 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
 815                         oif = skb->skb_iif;
 816
 817                 fl6.flowi6_oif = oif;
 818         }
 819
 820         fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
 821         fl6.fl6_dport = t1->dest;
 822         fl6.fl6_sport = t1->source;
 823         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 824
 825         /* Pass a socket to ip6_dst_lookup either it is for RST
 826          * Underlying function will use this to retrieve the network
 827          * namespace
 828          */
 829         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 830         if (!IS_ERR(dst)) {
 831                 skb_dst_set(buff, dst);
 832                 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
 833                 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 834                 if (rst)
 835                         TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
 836                 return;
 837         }
 838
 839         kfree_skb(buff);
 840 }
 841
 842 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
 843 {
 844         const struct tcphdr *th = tcp_hdr(skb);
 845         u32 seq = 0, ack_seq = 0;
 846         struct tcp_md5sig_key *key = NULL;
 847 #ifdef CONFIG_TCP_MD5SIG
 848         const __u8 *hash_location = NULL;
 849         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 850         unsigned char newhash[16];
 851         int genhash;
 852         struct sock *sk1 = NULL;
 853 #endif
 854         int oif;
 855
 856         if (th->rst)
 857                 return;
 858
 859         /* If sk not NULL, it means we did a successful lookup and incoming
 860          * route had to be correct. prequeue might have dropped our dst.
 861          */
 862         if (!sk && !ipv6_unicast_destination(skb))
 863                 return;
 864
 865 #ifdef CONFIG_TCP_MD5SIG
 866         rcu_read_lock();
 867         hash_location = tcp_parse_md5sig_option(th);
 868         if (sk && sk_fullsock(sk)) {
 869                 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
 870         } else if (hash_location) {
 871                 /*
 872                  * active side is lost. Try to find listening socket through
 873                  * source port, and then find md5 key through listening socket.
 874                  * we are not loose security here:
 875                  * Incoming packet is checked with md5 hash with finding key,
 876                  * no RST generated if md5 hash doesn't match.
 877                  */
 878                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
 879                                            &tcp_hashinfo, NULL, 0,
 880                                            &ipv6h->saddr,
 881                                            th->source, &ipv6h->daddr,
 882                                            ntohs(th->source), tcp_v6_iif(skb));
 883                 if (!sk1)
 884                         goto out;
 885
 886                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 887                 if (!key)
 888                         goto out;
 889
 890                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
 891                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
 892                         goto out;
 893         }
 894 #endif
 895
 896         if (th->ack)
 897                 seq = ntohl(th->ack_seq);
 898         else
 899                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
 900                           (th->doff << 2);
 901
 902         oif = sk ? sk->sk_bound_dev_if : 0;
 903         tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
 904
 905 #ifdef CONFIG_TCP_MD5SIG
 906 out:
 907         rcu_read_unlock();
 908 #endif
 909 }
 910
 911 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
 912                             u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
 913                             struct tcp_md5sig_key *key, u8 tclass,
 914                             u32 label)
 915 {
 916         tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
 917                              tclass, label);
 918 }
 919
 920 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 921 {
 922         struct inet_timewait_sock *tw = inet_twsk(sk);
 923         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 924
 925         tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 926                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 927                         tcp_time_stamp + tcptw->tw_ts_offset,
 928                         tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
 929                         tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
 930
 931         inet_twsk_put(tw);
 932 }
 933
 934 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 935                                   struct request_sock *req)
 936 {
 937         /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 938          * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 939          */
 940         tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
 941                         tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
 942                         tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
 943                         tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
 944                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
 945                         0, 0);
 946 }
 947
 948
 949 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
 950 {
 951 #ifdef CONFIG_SYN_COOKIES
 952         const struct tcphdr *th = tcp_hdr(skb);
 953
 954         if (!th->syn)
 955                 sk = cookie_v6_check(sk, skb);
 956 #endif
 957         return sk;
 958 }
 959
 960 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 961 {
 962         if (skb->protocol == htons(ETH_P_IP))
 963                 return tcp_v4_conn_request(sk, skb);
 964
 965         if (!ipv6_unicast_destination(skb))
 966                 goto drop;
 967
 968         return tcp_conn_request(&tcp6_request_sock_ops,
 969                                 &tcp_request_sock_ipv6_ops, sk, skb);
 970
 971 drop:
 972         tcp_listendrop(sk);
 973         return 0; /* don't send reset */
 974 }
 975
 976 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
 977                                          struct request_sock *req,
 978                                          struct dst_entry *dst,
 979                                          struct request_sock *req_unhash,
 980                                          bool *own_req)
 981 {
 982         struct inet_request_sock *ireq;
 983         struct ipv6_pinfo *newnp;
 984         const struct ipv6_pinfo *np = inet6_sk(sk);
 985         struct ipv6_txoptions *opt;
 986         struct tcp6_sock *newtcp6sk;
 987         struct inet_sock *newinet;
 988         struct tcp_sock *newtp;
 989         struct sock *newsk;
 990 #ifdef CONFIG_TCP_MD5SIG
 991         struct tcp_md5sig_key *key;
 992 #endif
 993         struct flowi6 fl6;
 994
 995         if (skb->protocol == htons(ETH_P_IP)) {
 996                 /*
 997                  *      v6 mapped
 998                  */
 999
1000                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1001                                              req_unhash, own_req);
1002
1003                 if (!newsk)
1004                         return NULL;
1005
1006                 newtcp6sk = (struct tcp6_sock *)newsk;
1007                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1008
1009                 newinet = inet_sk(newsk);
1010                 newnp = inet6_sk(newsk);
1011                 newtp = tcp_sk(newsk);
1012
1013                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1014
1015                 newnp->saddr = newsk->sk_v6_rcv_saddr;
1016
1017                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1018                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1019 #ifdef CONFIG_TCP_MD5SIG
1020                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1021 #endif
1022
1023                 newnp->ipv6_ac_list = NULL;
1024                 newnp->ipv6_fl_list = NULL;
1025                 newnp->pktoptions  = NULL;
1026                 newnp->opt         = NULL;
1027                 newnp->mcast_oif   = tcp_v6_iif(skb);
1028                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1029                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1030                 if (np->repflow)
1031                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1032
1033                 /*
1034                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1035                  * here, tcp_create_openreq_child now does this for us, see the comment in
1036                  * that function for the gory details. -acme
1037                  */
1038
1039                 /* It is tricky place. Until this moment IPv4 tcp
1040                    worked with IPv6 icsk.icsk_af_ops.
1041                    Sync it now.
1042                  */
1043                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1044
1045                 return newsk;
1046         }
1047
1048         ireq = inet_rsk(req);
1049
1050         if (sk_acceptq_is_full(sk))
1051                 goto out_overflow;
1052
1053         if (!dst) {
1054                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1055                 if (!dst)
1056                         goto out;
1057         }
1058
1059         newsk = tcp_create_openreq_child(sk, req, skb);
1060         if (!newsk)
1061                 goto out_nonewsk;
1062
1063         /*
1064          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1065          * count here, tcp_create_openreq_child now does this for us, see the
1066          * comment in that function for the gory details. -acme
1067          */
1068
1069         newsk->sk_gso_type = SKB_GSO_TCPV6;
1070         ip6_dst_store(newsk, dst, NULL, NULL);
1071         inet6_sk_rx_dst_set(newsk, skb);
1072
1073         newtcp6sk = (struct tcp6_sock *)newsk;
1074         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1075
1076         newtp = tcp_sk(newsk);
1077         newinet = inet_sk(newsk);
1078         newnp = inet6_sk(newsk);
1079
1080         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1081
1082         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1083         newnp->saddr = ireq->ir_v6_loc_addr;
1084         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1085         newsk->sk_bound_dev_if = ireq->ir_iif;
1086
1087         /* Now IPv6 options...
1088
1089            First: no IPv4 options.
1090          */
1091         newinet->inet_opt = NULL;
1092         newnp->ipv6_ac_list = NULL;
1093         newnp->ipv6_fl_list = NULL;
1094
1095         /* Clone RX bits */
1096         newnp->rxopt.all = np->rxopt.all;
1097
1098         newnp->pktoptions = NULL;
1099         newnp->opt        = NULL;
1100         newnp->mcast_oif  = tcp_v6_iif(skb);
1101         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1102         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1103         if (np->repflow)
1104                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1105
1106         /* Clone native IPv6 options from listening socket (if any)
1107
1108            Yes, keeping reference count would be much more clever,
1109            but we make one more one thing there: reattach optmem
1110            to newsk.
1111          */
1112         opt = rcu_dereference(np->opt);
1113         if (opt) {
1114                 opt = ipv6_dup_options(newsk, opt);
1115                 RCU_INIT_POINTER(newnp->opt, opt);
1116         }
1117         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1118         if (opt)
1119                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1120                                                     opt->opt_flen;
1121
1122         tcp_ca_openreq_child(newsk, dst);
1123
1124         tcp_sync_mss(newsk, dst_mtu(dst));
1125         newtp->advmss = dst_metric_advmss(dst);
1126         if (tcp_sk(sk)->rx_opt.user_mss &&
1127             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1128                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1129
1130         tcp_initialize_rcv_mss(newsk);
1131
1132         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1133         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1134
1135 #ifdef CONFIG_TCP_MD5SIG
1136         /* Copy over the MD5 key from the original socket */
1137         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1138         if (key) {
1139                 /* We're using one, so create a matching key
1140                  * on the newsk structure. If we fail to get
1141                  * memory, then we end up not copying the key
1142                  * across. Shucks.
1143                  */
1144                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1145                                AF_INET6, key->key, key->keylen,
1146                                sk_gfp_mask(sk, GFP_ATOMIC));
1147         }
1148 #endif
1149
1150         if (__inet_inherit_port(sk, newsk) < 0) {
1151                 inet_csk_prepare_forced_close(newsk);
1152                 tcp_done(newsk);
1153                 goto out;
1154         }
1155         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1156         if (*own_req) {
1157                 tcp_move_syn(newtp, req);
1158
1159                 /* Clone pktoptions received with SYN, if we own the req */
1160                 if (ireq->pktopts) {
1161                         newnp->pktoptions = skb_clone(ireq->pktopts,
1162                                                       sk_gfp_mask(sk, GFP_ATOMIC));
1163                         consume_skb(ireq->pktopts);
1164                         ireq->pktopts = NULL;
1165                         if (newnp->pktoptions)
1166                                 skb_set_owner_r(newnp->pktoptions, newsk);
1167                 }
1168         }
1169
1170         return newsk;
1171
1172 out_overflow:
1173         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1174 out_nonewsk:
1175         dst_release(dst);
1176 out:
1177         tcp_listendrop(sk);
1178         return NULL;
1179 }
1180
1181 /* The socket must have it's spinlock held when we get
1182  * here, unless it is a TCP_LISTEN socket.
1183  *
1184  * We have a potential double-lock case here, so even when
1185  * doing backlog processing we use the BH locking scheme.
1186  * This is because we cannot sleep with the original spinlock
1187  * held.
1188  */
1189 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1190 {
1191         struct ipv6_pinfo *np = inet6_sk(sk);
1192         struct tcp_sock *tp;
1193         struct sk_buff *opt_skb = NULL;
1194
1195         /* Imagine: socket is IPv6. IPv4 packet arrives,
1196            goes to IPv4 receive handler and backlogged.
1197            From backlog it always goes here. Kerboom...
1198            Fortunately, tcp_rcv_established and rcv_established
1199            handle them correctly, but it is not case with
1200            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1201          */
1202
1203         if (skb->protocol == htons(ETH_P_IP))
1204                 return tcp_v4_do_rcv(sk, skb);
1205
1206         if (sk_filter(sk, skb))
1207                 goto discard;
1208
1209         /*
1210          *      socket locking is here for SMP purposes as backlog rcv
1211          *      is currently called with bh processing disabled.
1212          */
1213
1214         /* Do Stevens' IPV6_PKTOPTIONS.
1215
1216            Yes, guys, it is the only place in our code, where we
1217            may make it not affecting IPv4.
1218            The rest of code is protocol independent,
1219            and I do not like idea to uglify IPv4.
1220
1221            Actually, all the idea behind IPV6_PKTOPTIONS
1222            looks not very well thought. For now we latch
1223            options, received in the last packet, enqueued
1224            by tcp. Feel free to propose better solution.
1225                                                --ANK (980728)
1226          */
1227         if (np->rxopt.all)
1228                 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1229
1230         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1231                 struct dst_entry *dst = sk->sk_rx_dst;
1232
1233                 sock_rps_save_rxhash(sk, skb);
1234                 sk_mark_napi_id(sk, skb);
1235                 if (dst) {
1236                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1237                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1238                                 dst_release(dst);
1239                                 sk->sk_rx_dst = NULL;
1240                         }
1241                 }
1242
1243                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1244                 if (opt_skb)
1245                         goto ipv6_pktoptions;
1246                 return 0;
1247         }
1248
1249         if (tcp_checksum_complete(skb))
1250                 goto csum_err;
1251
1252         if (sk->sk_state == TCP_LISTEN) {
1253                 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1254
1255                 if (!nsk)
1256                         goto discard;
1257
1258                 if (nsk != sk) {
1259                         sock_rps_save_rxhash(nsk, skb);
1260                         sk_mark_napi_id(nsk, skb);
1261                         if (tcp_child_process(sk, nsk, skb))
1262                                 goto reset;
1263                         if (opt_skb)
1264                                 __kfree_skb(opt_skb);
1265                         return 0;
1266                 }
1267         } else
1268                 sock_rps_save_rxhash(sk, skb);
1269
1270         if (tcp_rcv_state_process(sk, skb))
1271                 goto reset;
1272         if (opt_skb)
1273                 goto ipv6_pktoptions;
1274         return 0;
1275
1276 reset:
1277         tcp_v6_send_reset(sk, skb);
1278 discard:
1279         if (opt_skb)
1280                 __kfree_skb(opt_skb);
1281         kfree_skb(skb);
1282         return 0;
1283 csum_err:
1284         TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1285         TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1286         goto discard;
1287
1288
1289 ipv6_pktoptions:
1290         /* Do you ask, what is it?
1291
1292            1. skb was enqueued by tcp.
1293            2. skb is added to tail of read queue, rather than out of order.
1294            3. socket is not in passive state.
1295            4. Finally, it really contains options, which user wants to receive.
1296          */
1297         tp = tcp_sk(sk);
1298         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1299             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1300                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1301                         np->mcast_oif = tcp_v6_iif(opt_skb);
1302                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1303                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1304                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1305                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1306                 if (np->repflow)
1307                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1308                 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1309                         skb_set_owner_r(opt_skb, sk);
1310                         opt_skb = xchg(&np->pktoptions, opt_skb);
1311                 } else {
1312                         __kfree_skb(opt_skb);
1313                         opt_skb = xchg(&np->pktoptions, NULL);
1314                 }
1315         }
1316
1317         kfree_skb(opt_skb);
1318         return 0;
1319 }
1320
1321 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1322                            const struct tcphdr *th)
1323 {
1324         /* This is tricky: we move IP6CB at its correct location into
1325          * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1326          * _decode_session6() uses IP6CB().
1327          * barrier() makes sure compiler won't play aliasing games.
1328          */
1329         memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1330                 sizeof(struct inet6_skb_parm));
1331         barrier();
1332
1333         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1334         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1335                                     skb->len - th->doff*4);
1336         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1337         TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1338         TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1339         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1340         TCP_SKB_CB(skb)->sacked = 0;
1341 }
1342
1343 static void tcp_v6_restore_cb(struct sk_buff *skb)
1344 {
1345         /* We need to move header back to the beginning if xfrm6_policy_check()
1346          * and tcp_v6_fill_cb() are going to be called again.
1347          */
1348         memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1349                 sizeof(struct inet6_skb_parm));
1350 }
1351
1352 static int tcp_v6_rcv(struct sk_buff *skb)
1353 {
1354         const struct tcphdr *th;
1355         const struct ipv6hdr *hdr;
1356         bool refcounted;
1357         struct sock *sk;
1358         int ret;
1359         struct net *net = dev_net(skb->dev);
1360
1361         if (skb->pkt_type != PACKET_HOST)
1362                 goto discard_it;
1363
1364         /*
1365          *      Count it even if it's bad.
1366          */
1367         __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1368
1369         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1370                 goto discard_it;
1371
1372         th = (const struct tcphdr *)skb->data;
1373
1374         if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1375                 goto bad_packet;
1376         if (!pskb_may_pull(skb, th->doff*4))
1377                 goto discard_it;
1378
1379         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1380                 goto csum_error;
1381
1382         th = (const struct tcphdr *)skb->data;
1383         hdr = ipv6_hdr(skb);
1384
1385 lookup:
1386         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1387                                 th->source, th->dest, inet6_iif(skb),
1388                                 &refcounted);
1389         if (!sk)
1390                 goto no_tcp_socket;
1391
1392 process:
1393         if (sk->sk_state == TCP_TIME_WAIT)
1394                 goto do_time_wait;
1395
1396         if (sk->sk_state == TCP_NEW_SYN_RECV) {
1397                 struct request_sock *req = inet_reqsk(sk);
1398                 struct sock *nsk;
1399
1400                 sk = req->rsk_listener;
1401                 tcp_v6_fill_cb(skb, hdr, th);
1402                 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1403                         reqsk_put(req);
1404                         goto discard_it;
1405                 }
1406                 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1407                         inet_csk_reqsk_queue_drop_and_put(sk, req);
1408                         goto lookup;
1409                 }
1410                 sock_hold(sk);
1411                 refcounted = true;
1412                 nsk = tcp_check_req(sk, skb, req, false);
1413                 if (!nsk) {
1414                         reqsk_put(req);
1415                         goto discard_and_relse;
1416                 }
1417                 if (nsk == sk) {
1418                         reqsk_put(req);
1419                         tcp_v6_restore_cb(skb);
1420                 } else if (tcp_child_process(sk, nsk, skb)) {
1421                         tcp_v6_send_reset(nsk, skb);
1422                         goto discard_and_relse;
1423                 } else {
1424                         sock_put(sk);
1425                         return 0;
1426                 }
1427         }
1428         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1429                 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1430                 goto discard_and_relse;
1431         }
1432
1433         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1434                 goto discard_and_relse;
1435
1436         tcp_v6_fill_cb(skb, hdr, th);
1437
1438         if (tcp_v6_inbound_md5_hash(sk, skb))
1439                 goto discard_and_relse;
1440
1441         if (sk_filter(sk, skb))
1442                 goto discard_and_relse;
1443
1444         skb->dev = NULL;
1445
1446         if (sk->sk_state == TCP_LISTEN) {
1447                 ret = tcp_v6_do_rcv(sk, skb);
1448                 goto put_and_return;
1449         }
1450
1451         sk_incoming_cpu_update(sk);
1452
1453         bh_lock_sock_nested(sk);
1454         tcp_segs_in(tcp_sk(sk), skb);
1455         ret = 0;
1456         if (!sock_owned_by_user(sk)) {
1457                 if (!tcp_prequeue(sk, skb))
1458                         ret = tcp_v6_do_rcv(sk, skb);
1459         } else if (unlikely(sk_add_backlog(sk, skb,
1460                                            sk->sk_rcvbuf + sk->sk_sndbuf))) {
1461                 bh_unlock_sock(sk);
1462                 __NET_INC_STATS(net, LINUX_MIB_TCPBACKLOGDROP);
1463                 goto discard_and_relse;
1464         }
1465         bh_unlock_sock(sk);
1466
1467 put_and_return:
1468         if (refcounted)
1469                 sock_put(sk);
1470         return ret ? -1 : 0;
1471
1472 no_tcp_socket:
1473         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1474                 goto discard_it;
1475
1476         tcp_v6_fill_cb(skb, hdr, th);
1477
1478         if (tcp_checksum_complete(skb)) {
1479 csum_error:
1480                 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1481 bad_packet:
1482                 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1483         } else {
1484                 tcp_v6_send_reset(NULL, skb);
1485         }
1486
1487 discard_it:
1488         kfree_skb(skb);
1489         return 0;
1490
1491 discard_and_relse:
1492         sk_drops_add(sk, skb);
1493         if (refcounted)
1494                 sock_put(sk);
1495         goto discard_it;
1496
1497 do_time_wait:
1498         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1499                 inet_twsk_put(inet_twsk(sk));
1500                 goto discard_it;
1501         }
1502
1503         tcp_v6_fill_cb(skb, hdr, th);
1504
1505         if (tcp_checksum_complete(skb)) {
1506                 inet_twsk_put(inet_twsk(sk));
1507                 goto csum_error;
1508         }
1509
1510         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1511         case TCP_TW_SYN:
1512         {
1513                 struct sock *sk2;
1514
1515                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1516                                             skb, __tcp_hdrlen(th),
1517                                             &ipv6_hdr(skb)->saddr, th->source,
1518                                             &ipv6_hdr(skb)->daddr,
1519                                             ntohs(th->dest), tcp_v6_iif(skb));
1520                 if (sk2) {
1521                         struct inet_timewait_sock *tw = inet_twsk(sk);
1522                         inet_twsk_deschedule_put(tw);
1523                         sk = sk2;
1524                         tcp_v6_restore_cb(skb);
1525                         refcounted = false;
1526                         goto process;
1527                 }
1528                 /* Fall through to ACK */
1529         }
1530         case TCP_TW_ACK:
1531                 tcp_v6_timewait_ack(sk, skb);
1532                 break;
1533         case TCP_TW_RST:
1534                 tcp_v6_restore_cb(skb);
1535                 tcp_v6_send_reset(sk, skb);
1536                 inet_twsk_deschedule_put(inet_twsk(sk));
1537                 goto discard_it;
1538         case TCP_TW_SUCCESS:
1539                 ;
1540         }
1541         goto discard_it;
1542 }
1543
1544 static void tcp_v6_early_demux(struct sk_buff *skb)
1545 {
1546         const struct ipv6hdr *hdr;
1547         const struct tcphdr *th;
1548         struct sock *sk;
1549
1550         if (skb->pkt_type != PACKET_HOST)
1551                 return;
1552
1553         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1554                 return;
1555
1556         hdr = ipv6_hdr(skb);
1557         th = tcp_hdr(skb);
1558
1559         if (th->doff < sizeof(struct tcphdr) / 4)
1560                 return;
1561
1562         /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1563         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1564                                         &hdr->saddr, th->source,
1565                                         &hdr->daddr, ntohs(th->dest),
1566                                         inet6_iif(skb));
1567         if (sk) {
1568                 skb->sk = sk;
1569                 skb->destructor = sock_edemux;
1570                 if (sk_fullsock(sk)) {
1571                         struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1572
1573                         if (dst)
1574                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1575                         if (dst &&
1576                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1577                                 skb_dst_set_noref(skb, dst);
1578                 }
1579         }
1580 }
1581
1582 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1583         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1584         .twsk_unique    = tcp_twsk_unique,
1585         .twsk_destructor = tcp_twsk_destructor,
1586 };
1587
1588 static const struct inet_connection_sock_af_ops ipv6_specific = {
1589         .queue_xmit        = inet6_csk_xmit,
1590         .send_check        = tcp_v6_send_check,
1591         .rebuild_header    = inet6_sk_rebuild_header,
1592         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1593         .conn_request      = tcp_v6_conn_request,
1594         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1595         .net_header_len    = sizeof(struct ipv6hdr),
1596         .net_frag_header_len = sizeof(struct frag_hdr),
1597         .setsockopt        = ipv6_setsockopt,
1598         .getsockopt        = ipv6_getsockopt,
1599         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1600         .sockaddr_len      = sizeof(struct sockaddr_in6),
1601         .bind_conflict     = inet6_csk_bind_conflict,
1602 #ifdef CONFIG_COMPAT
1603         .compat_setsockopt = compat_ipv6_setsockopt,
1604         .compat_getsockopt = compat_ipv6_getsockopt,
1605 #endif
1606         .mtu_reduced       = tcp_v6_mtu_reduced,
1607 };
1608
1609 #ifdef CONFIG_TCP_MD5SIG
1610 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1611         .md5_lookup     =       tcp_v6_md5_lookup,
1612         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1613         .md5_parse      =       tcp_v6_parse_md5_keys,
1614 };
1615 #endif
1616
1617 /*
1618  *      TCP over IPv4 via INET6 API
1619  */
1620 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1621         .queue_xmit        = ip_queue_xmit,
1622         .send_check        = tcp_v4_send_check,
1623         .rebuild_header    = inet_sk_rebuild_header,
1624         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1625         .conn_request      = tcp_v6_conn_request,
1626         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1627         .net_header_len    = sizeof(struct iphdr),
1628         .setsockopt        = ipv6_setsockopt,
1629         .getsockopt        = ipv6_getsockopt,
1630         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1631         .sockaddr_len      = sizeof(struct sockaddr_in6),
1632         .bind_conflict     = inet6_csk_bind_conflict,
1633 #ifdef CONFIG_COMPAT
1634         .compat_setsockopt = compat_ipv6_setsockopt,
1635         .compat_getsockopt = compat_ipv6_getsockopt,
1636 #endif
1637         .mtu_reduced       = tcp_v4_mtu_reduced,
1638 };
1639
1640 #ifdef CONFIG_TCP_MD5SIG
1641 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1642         .md5_lookup     =       tcp_v4_md5_lookup,
1643         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1644         .md5_parse      =       tcp_v6_parse_md5_keys,
1645 };
1646 #endif
1647
1648 /* NOTE: A lot of things set to zero explicitly by call to
1649  *       sk_alloc() so need not be done here.
1650  */
1651 static int tcp_v6_init_sock(struct sock *sk)
1652 {
1653         struct inet_connection_sock *icsk = inet_csk(sk);
1654
1655         tcp_init_sock(sk);
1656
1657         icsk->icsk_af_ops = &ipv6_specific;
1658
1659 #ifdef CONFIG_TCP_MD5SIG
1660         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1661 #endif
1662
1663         return 0;
1664 }
1665
1666 static void tcp_v6_destroy_sock(struct sock *sk)
1667 {
1668         tcp_v4_destroy_sock(sk);
1669         inet6_destroy_sock(sk);
1670 }
1671
1672 #ifdef CONFIG_PROC_FS
1673 /* Proc filesystem TCPv6 sock list dumping. */
1674 static void get_openreq6(struct seq_file *seq,
1675                          const struct request_sock *req, int i)
1676 {
1677         long ttd = req->rsk_timer.expires - jiffies;
1678         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1679         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1680
1681         if (ttd < 0)
1682                 ttd = 0;
1683
1684         seq_printf(seq,
1685                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1686                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1687                    i,
1688                    src->s6_addr32[0], src->s6_addr32[1],
1689                    src->s6_addr32[2], src->s6_addr32[3],
1690                    inet_rsk(req)->ir_num,
1691                    dest->s6_addr32[0], dest->s6_addr32[1],
1692                    dest->s6_addr32[2], dest->s6_addr32[3],
1693                    ntohs(inet_rsk(req)->ir_rmt_port),
1694                    TCP_SYN_RECV,
1695                    0, 0, /* could print option size, but that is af dependent. */
1696                    1,   /* timers active (only the expire timer) */
1697                    jiffies_to_clock_t(ttd),
1698                    req->num_timeout,
1699                    from_kuid_munged(seq_user_ns(seq),
1700                                     sock_i_uid(req->rsk_listener)),
1701                    0,  /* non standard timer */
1702                    0, /* open_requests have no inode */
1703                    0, req);
1704 }
1705
1706 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1707 {
1708         const struct in6_addr *dest, *src;
1709         __u16 destp, srcp;
1710         int timer_active;
1711         unsigned long timer_expires;
1712         const struct inet_sock *inet = inet_sk(sp);
1713         const struct tcp_sock *tp = tcp_sk(sp);
1714         const struct inet_connection_sock *icsk = inet_csk(sp);
1715         const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1716         int rx_queue;
1717         int state;
1718
1719         dest  = &sp->sk_v6_daddr;
1720         src   = &sp->sk_v6_rcv_saddr;
1721         destp = ntohs(inet->inet_dport);
1722         srcp  = ntohs(inet->inet_sport);
1723
1724         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1725                 timer_active    = 1;
1726                 timer_expires   = icsk->icsk_timeout;
1727         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1728                 timer_active    = 4;
1729                 timer_expires   = icsk->icsk_timeout;
1730         } else if (timer_pending(&sp->sk_timer)) {
1731                 timer_active    = 2;
1732                 timer_expires   = sp->sk_timer.expires;
1733         } else {
1734                 timer_active    = 0;
1735                 timer_expires = jiffies;
1736         }
1737
1738         state = sk_state_load(sp);
1739         if (state == TCP_LISTEN)
1740                 rx_queue = sp->sk_ack_backlog;
1741         else
1742                 /* Because we don't lock the socket,
1743                  * we might find a transient negative value.
1744                  */
1745                 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1746
1747         seq_printf(seq,
1748                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1749                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1750                    i,
1751                    src->s6_addr32[0], src->s6_addr32[1],
1752                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1753                    dest->s6_addr32[0], dest->s6_addr32[1],
1754                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1755                    state,
1756                    tp->write_seq - tp->snd_una,
1757                    rx_queue,
1758                    timer_active,
1759                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1760                    icsk->icsk_retransmits,
1761                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1762                    icsk->icsk_probes_out,
1763                    sock_i_ino(sp),
1764                    atomic_read(&sp->sk_refcnt), sp,
1765                    jiffies_to_clock_t(icsk->icsk_rto),
1766                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1767                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1768                    tp->snd_cwnd,
1769                    state == TCP_LISTEN ?
1770                         fastopenq->max_qlen :
1771                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1772                    );
1773 }
1774
1775 static void get_timewait6_sock(struct seq_file *seq,
1776                                struct inet_timewait_sock *tw, int i)
1777 {
1778         long delta = tw->tw_timer.expires - jiffies;
1779         const struct in6_addr *dest, *src;
1780         __u16 destp, srcp;
1781
1782         dest = &tw->tw_v6_daddr;
1783         src  = &tw->tw_v6_rcv_saddr;
1784         destp = ntohs(tw->tw_dport);
1785         srcp  = ntohs(tw->tw_sport);
1786
1787         seq_printf(seq,
1788                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1789                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1790                    i,
1791                    src->s6_addr32[0], src->s6_addr32[1],
1792                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1793                    dest->s6_addr32[0], dest->s6_addr32[1],
1794                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1795                    tw->tw_substate, 0, 0,
1796                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1797                    atomic_read(&tw->tw_refcnt), tw);
1798 }
1799
1800 static int tcp6_seq_show(struct seq_file *seq, void *v)
1801 {
1802         struct tcp_iter_state *st;
1803         struct sock *sk = v;
1804
1805         if (v == SEQ_START_TOKEN) {
1806                 seq_puts(seq,
1807                          "  sl  "
1808                          "local_address                         "
1809                          "remote_address                        "
1810                          "st tx_queue rx_queue tr tm->when retrnsmt"
1811                          "   uid  timeout inode\n");
1812                 goto out;
1813         }
1814         st = seq->private;
1815
1816         if (sk->sk_state == TCP_TIME_WAIT)
1817                 get_timewait6_sock(seq, v, st->num);
1818         else if (sk->sk_state == TCP_NEW_SYN_RECV)
1819                 get_openreq6(seq, v, st->num);
1820         else
1821                 get_tcp6_sock(seq, v, st->num);
1822 out:
1823         return 0;
1824 }
1825
1826 static const struct file_operations tcp6_afinfo_seq_fops = {
1827         .owner   = THIS_MODULE,
1828         .open    = tcp_seq_open,
1829         .read    = seq_read,
1830         .llseek  = seq_lseek,
1831         .release = seq_release_net
1832 };
1833
1834 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1835         .name           = "tcp6",
1836         .family         = AF_INET6,
1837         .seq_fops       = &tcp6_afinfo_seq_fops,
1838         .seq_ops        = {
1839                 .show           = tcp6_seq_show,
1840         },
1841 };
1842
1843 int __net_init tcp6_proc_init(struct net *net)
1844 {
1845         return tcp_proc_register(net, &tcp6_seq_afinfo);
1846 }
1847
1848 void tcp6_proc_exit(struct net *net)
1849 {
1850         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1851 }
1852 #endif
1853
1854 static void tcp_v6_clear_sk(struct sock *sk, int size)
1855 {
1856         struct inet_sock *inet = inet_sk(sk);
1857
1858         /* we do not want to clear pinet6 field, because of RCU lookups */
1859         sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1860
1861         size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1862         memset(&inet->pinet6 + 1, 0, size);
1863 }
1864
1865 struct proto tcpv6_prot = {
1866         .name                   = "TCPv6",
1867         .owner                  = THIS_MODULE,
1868         .close                  = tcp_close,
1869         .connect                = tcp_v6_connect,
1870         .disconnect             = tcp_disconnect,
1871         .accept                 = inet_csk_accept,
1872         .ioctl                  = tcp_ioctl,
1873         .init                   = tcp_v6_init_sock,
1874         .destroy                = tcp_v6_destroy_sock,
1875         .shutdown               = tcp_shutdown,
1876         .setsockopt             = tcp_setsockopt,
1877         .getsockopt             = tcp_getsockopt,
1878         .recvmsg                = tcp_recvmsg,
1879         .sendmsg                = tcp_sendmsg,
1880         .sendpage               = tcp_sendpage,
1881         .backlog_rcv            = tcp_v6_do_rcv,
1882         .release_cb             = tcp_release_cb,
1883         .hash                   = inet6_hash,
1884         .unhash                 = inet_unhash,
1885         .get_port               = inet_csk_get_port,
1886         .enter_memory_pressure  = tcp_enter_memory_pressure,
1887         .stream_memory_free     = tcp_stream_memory_free,
1888         .sockets_allocated      = &tcp_sockets_allocated,
1889         .memory_allocated       = &tcp_memory_allocated,
1890         .memory_pressure        = &tcp_memory_pressure,
1891         .orphan_count           = &tcp_orphan_count,
1892         .sysctl_mem             = sysctl_tcp_mem,
1893         .sysctl_wmem            = sysctl_tcp_wmem,
1894         .sysctl_rmem            = sysctl_tcp_rmem,
1895         .max_header             = MAX_TCP_HEADER,
1896         .obj_size               = sizeof(struct tcp6_sock),
1897         .slab_flags             = SLAB_DESTROY_BY_RCU,
1898         .twsk_prot              = &tcp6_timewait_sock_ops,
1899         .rsk_prot               = &tcp6_request_sock_ops,
1900         .h.hashinfo             = &tcp_hashinfo,
1901         .no_autobind            = true,
1902 #ifdef CONFIG_COMPAT
1903         .compat_setsockopt      = compat_tcp_setsockopt,
1904         .compat_getsockopt      = compat_tcp_getsockopt,
1905 #endif
1906         .clear_sk               = tcp_v6_clear_sk,
1907         .diag_destroy           = tcp_abort,
1908 };
1909
1910 static const struct inet6_protocol tcpv6_protocol = {
1911         .early_demux    =       tcp_v6_early_demux,
1912         .handler        =       tcp_v6_rcv,
1913         .err_handler    =       tcp_v6_err,
1914         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1915 };
1916
1917 static struct inet_protosw tcpv6_protosw = {
1918         .type           =       SOCK_STREAM,
1919         .protocol       =       IPPROTO_TCP,
1920         .prot           =       &tcpv6_prot,
1921         .ops            =       &inet6_stream_ops,
1922         .flags          =       INET_PROTOSW_PERMANENT |
1923                                 INET_PROTOSW_ICSK,
1924 };
1925
1926 static int __net_init tcpv6_net_init(struct net *net)
1927 {
1928         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1929                                     SOCK_RAW, IPPROTO_TCP, net);
1930 }
1931
1932 static void __net_exit tcpv6_net_exit(struct net *net)
1933 {
1934         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1935 }
1936
1937 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1938 {
1939         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1940 }
1941
1942 static struct pernet_operations tcpv6_net_ops = {
1943         .init       = tcpv6_net_init,
1944         .exit       = tcpv6_net_exit,
1945         .exit_batch = tcpv6_net_exit_batch,
1946 };
1947
1948 int __init tcpv6_init(void)
1949 {
1950         int ret;
1951
1952         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1953         if (ret)
1954                 goto out;
1955
1956         /* register inet6 protocol */
1957         ret = inet6_register_protosw(&tcpv6_protosw);
1958         if (ret)
1959                 goto out_tcpv6_protocol;
1960
1961         ret = register_pernet_subsys(&tcpv6_net_ops);
1962         if (ret)
1963                 goto out_tcpv6_protosw;
1964 out:
1965         return ret;
1966
1967 out_tcpv6_protosw:
1968         inet6_unregister_protosw(&tcpv6_protosw);
1969 out_tcpv6_protocol:
1970         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1971         goto out;
1972 }
1973
1974 void tcpv6_exit(void)
1975 {
1976         unregister_pernet_subsys(&tcpv6_net_ops);
1977         inet6_unregister_protosw(&tcpv6_protosw);
1978         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1979 }