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_seq(const struct sk_buff *skb)
 105 {
 106         return secure_tcpv6_seq(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 u32 tcp_v6_init_ts_off(const struct sk_buff *skb)
 113 {
 114         return secure_tcpv6_ts_off(ipv6_hdr(skb)->daddr.s6_addr32,
 115                                    ipv6_hdr(skb)->saddr.s6_addr32);
 116 }
 117
 118 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 119                           int addr_len)
 120 {
 121         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 122         struct inet_sock *inet = inet_sk(sk);
 123         struct inet_connection_sock *icsk = inet_csk(sk);
 124         struct ipv6_pinfo *np = inet6_sk(sk);
 125         struct tcp_sock *tp = tcp_sk(sk);
 126         struct in6_addr *saddr = NULL, *final_p, final;
 127         struct ipv6_txoptions *opt;
 128         struct flowi6 fl6;
 129         struct dst_entry *dst;
 130         int addr_type;
 131         int err;
 132         struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
 133
 134         if (addr_len < SIN6_LEN_RFC2133)
 135                 return -EINVAL;
 136
 137         if (usin->sin6_family != AF_INET6)
 138                 return -EAFNOSUPPORT;
 139
 140         memset(&fl6, 0, sizeof(fl6));
 141
 142         if (np->sndflow) {
 143                 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 144                 IP6_ECN_flow_init(fl6.flowlabel);
 145                 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
 146                         struct ip6_flowlabel *flowlabel;
 147                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 148                         if (!flowlabel)
 149                                 return -EINVAL;
 150                         fl6_sock_release(flowlabel);
 151                 }
 152         }
 153
 154         /*
 155          *      connect() to INADDR_ANY means loopback (BSD'ism).
 156          */
 157
 158         if (ipv6_addr_any(&usin->sin6_addr)) {
 159                 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
 160                         ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
 161                                                &usin->sin6_addr);
 162                 else
 163                         usin->sin6_addr = in6addr_loopback;
 164         }
 165
 166         addr_type = ipv6_addr_type(&usin->sin6_addr);
 167
 168         if (addr_type & IPV6_ADDR_MULTICAST)
 169                 return -ENETUNREACH;
 170
 171         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 172                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 173                     usin->sin6_scope_id) {
 174                         /* If interface is set while binding, indices
 175                          * must coincide.
 176                          */
 177                         if (sk->sk_bound_dev_if &&
 178                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 179                                 return -EINVAL;
 180
 181                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 182                 }
 183
 184                 /* Connect to link-local address requires an interface */
 185                 if (!sk->sk_bound_dev_if)
 186                         return -EINVAL;
 187         }
 188
 189         if (tp->rx_opt.ts_recent_stamp &&
 190             !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
 191                 tp->rx_opt.ts_recent = 0;
 192                 tp->rx_opt.ts_recent_stamp = 0;
 193                 tp->write_seq = 0;
 194         }
 195
 196         sk->sk_v6_daddr = usin->sin6_addr;
 197         np->flow_label = fl6.flowlabel;
 198
 199         /*
 200          *      TCP over IPv4
 201          */
 202
 203         if (addr_type & IPV6_ADDR_MAPPED) {
 204                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 205                 struct sockaddr_in sin;
 206
 207                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 208
 209                 if (__ipv6_only_sock(sk))
 210                         return -ENETUNREACH;
 211
 212                 sin.sin_family = AF_INET;
 213                 sin.sin_port = usin->sin6_port;
 214                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 215
 216                 icsk->icsk_af_ops = &ipv6_mapped;
 217                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 218 #ifdef CONFIG_TCP_MD5SIG
 219                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
 220 #endif
 221
 222                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 223
 224                 if (err) {
 225                         icsk->icsk_ext_hdr_len = exthdrlen;
 226                         icsk->icsk_af_ops = &ipv6_specific;
 227                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 228 #ifdef CONFIG_TCP_MD5SIG
 229                         tp->af_specific = &tcp_sock_ipv6_specific;
 230 #endif
 231                         goto failure;
 232                 }
 233                 np->saddr = sk->sk_v6_rcv_saddr;
 234
 235                 return err;
 236         }
 237
 238         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 239                 saddr = &sk->sk_v6_rcv_saddr;
 240
 241         fl6.flowi6_proto = IPPROTO_TCP;
 242         fl6.daddr = sk->sk_v6_daddr;
 243         fl6.saddr = saddr ? *saddr : np->saddr;
 244         fl6.flowi6_oif = sk->sk_bound_dev_if;
 245         fl6.flowi6_mark = sk->sk_mark;
 246         fl6.fl6_dport = usin->sin6_port;
 247         fl6.fl6_sport = inet->inet_sport;
 248         fl6.flowi6_uid = sk->sk_uid;
 249
 250         opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
 251         final_p = fl6_update_dst(&fl6, opt, &final);
 252
 253         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 254
 255         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 256         if (IS_ERR(dst)) {
 257                 err = PTR_ERR(dst);
 258                 goto failure;
 259         }
 260
 261         if (!saddr) {
 262                 saddr = &fl6.saddr;
 263                 sk->sk_v6_rcv_saddr = *saddr;
 264         }
 265
 266         /* set the source address */
 267         np->saddr = *saddr;
 268         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 269
 270         sk->sk_gso_type = SKB_GSO_TCPV6;
 271         ip6_dst_store(sk, dst, NULL, NULL);
 272
 273         icsk->icsk_ext_hdr_len = 0;
 274         if (opt)
 275                 icsk->icsk_ext_hdr_len = opt->opt_flen +
 276                                          opt->opt_nflen;
 277
 278         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 279
 280         inet->inet_dport = usin->sin6_port;
 281
 282         tcp_set_state(sk, TCP_SYN_SENT);
 283         err = inet6_hash_connect(tcp_death_row, sk);
 284         if (err)
 285                 goto late_failure;
 286
 287         sk_set_txhash(sk);
 288
 289         if (likely(!tp->repair)) {
 290                 if (!tp->write_seq)
 291                         tp->write_seq = secure_tcpv6_seq(np->saddr.s6_addr32,
 292                                                          sk->sk_v6_daddr.s6_addr32,
 293                                                          inet->inet_sport,
 294                                                          inet->inet_dport);
 295                 tp->tsoffset = secure_tcpv6_ts_off(np->saddr.s6_addr32,
 296                                                    sk->sk_v6_daddr.s6_addr32);
 297         }
 298
 299         if (tcp_fastopen_defer_connect(sk, &err))
 300                 return err;
 301         if (err)
 302                 goto late_failure;
 303
 304         err = tcp_connect(sk);
 305         if (err)
 306                 goto late_failure;
 307
 308         return 0;
 309
 310 late_failure:
 311         tcp_set_state(sk, TCP_CLOSE);
 312 failure:
 313         inet->inet_dport = 0;
 314         sk->sk_route_caps = 0;
 315         return err;
 316 }
 317
 318 static void tcp_v6_mtu_reduced(struct sock *sk)
 319 {
 320         struct dst_entry *dst;
 321
 322         if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 323                 return;
 324
 325         dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
 326         if (!dst)
 327                 return;
 328
 329         if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
 330                 tcp_sync_mss(sk, dst_mtu(dst));
 331                 tcp_simple_retransmit(sk);
 332         }
 333 }
 334
 335 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 336                 u8 type, u8 code, int offset, __be32 info)
 337 {
 338         const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
 339         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 340         struct net *net = dev_net(skb->dev);
 341         struct request_sock *fastopen;
 342         struct ipv6_pinfo *np;
 343         struct tcp_sock *tp;
 344         __u32 seq, snd_una;
 345         struct sock *sk;
 346         bool fatal;
 347         int err;
 348
 349         sk = __inet6_lookup_established(net, &tcp_hashinfo,
 350                                         &hdr->daddr, th->dest,
 351                                         &hdr->saddr, ntohs(th->source),
 352                                         skb->dev->ifindex);
 353
 354         if (!sk) {
 355                 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
 356                                   ICMP6_MIB_INERRORS);
 357                 return;
 358         }
 359
 360         if (sk->sk_state == TCP_TIME_WAIT) {
 361                 inet_twsk_put(inet_twsk(sk));
 362                 return;
 363         }
 364         seq = ntohl(th->seq);
 365         fatal = icmpv6_err_convert(type, code, &err);
 366         if (sk->sk_state == TCP_NEW_SYN_RECV)
 367                 return tcp_req_err(sk, seq, fatal);
 368
 369         bh_lock_sock(sk);
 370         if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
 371                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 372
 373         if (sk->sk_state == TCP_CLOSE)
 374                 goto out;
 375
 376         if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
 377                 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
 378                 goto out;
 379         }
 380
 381         tp = tcp_sk(sk);
 382         /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 383         fastopen = tp->fastopen_rsk;
 384         snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 385         if (sk->sk_state != TCP_LISTEN &&
 386             !between(seq, snd_una, tp->snd_nxt)) {
 387                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 388                 goto out;
 389         }
 390
 391         np = inet6_sk(sk);
 392
 393         if (type == NDISC_REDIRECT) {
 394                 if (!sock_owned_by_user(sk)) {
 395                         struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 396
 397                         if (dst)
 398                                 dst->ops->redirect(dst, sk, skb);
 399                 }
 400                 goto out;
 401         }
 402
 403         if (type == ICMPV6_PKT_TOOBIG) {
 404                 /* We are not interested in TCP_LISTEN and open_requests
 405                  * (SYN-ACKs send out by Linux are always <576bytes so
 406                  * they should go through unfragmented).
 407                  */
 408                 if (sk->sk_state == TCP_LISTEN)
 409                         goto out;
 410
 411                 if (!ip6_sk_accept_pmtu(sk))
 412                         goto out;
 413
 414                 tp->mtu_info = ntohl(info);
 415                 if (!sock_owned_by_user(sk))
 416                         tcp_v6_mtu_reduced(sk);
 417                 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
 418                                            &sk->sk_tsq_flags))
 419                         sock_hold(sk);
 420                 goto out;
 421         }
 422
 423
 424         /* Might be for an request_sock */
 425         switch (sk->sk_state) {
 426         case TCP_SYN_SENT:
 427         case TCP_SYN_RECV:
 428                 /* Only in fast or simultaneous open. If a fast open socket is
 429                  * is already accepted it is treated as a connected one below.
 430                  */
 431                 if (fastopen && !fastopen->sk)
 432                         break;
 433
 434                 if (!sock_owned_by_user(sk)) {
 435                         sk->sk_err = err;
 436                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 437
 438                         tcp_done(sk);
 439                 } else
 440                         sk->sk_err_soft = err;
 441                 goto out;
 442         }
 443
 444         if (!sock_owned_by_user(sk) && np->recverr) {
 445                 sk->sk_err = err;
 446                 sk->sk_error_report(sk);
 447         } else
 448                 sk->sk_err_soft = err;
 449
 450 out:
 451         bh_unlock_sock(sk);
 452         sock_put(sk);
 453 }
 454
 455
 456 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 457                               struct flowi *fl,
 458                               struct request_sock *req,
 459                               struct tcp_fastopen_cookie *foc,
 460                               enum tcp_synack_type synack_type)
 461 {
 462         struct inet_request_sock *ireq = inet_rsk(req);
 463         struct ipv6_pinfo *np = inet6_sk(sk);
 464         struct ipv6_txoptions *opt;
 465         struct flowi6 *fl6 = &fl->u.ip6;
 466         struct sk_buff *skb;
 467         int err = -ENOMEM;
 468
 469         /* First, grab a route. */
 470         if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
 471                                                IPPROTO_TCP)) == NULL)
 472                 goto done;
 473
 474         skb = tcp_make_synack(sk, dst, req, foc, synack_type);
 475
 476         if (skb) {
 477                 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
 478                                     &ireq->ir_v6_rmt_addr);
 479
 480                 fl6->daddr = ireq->ir_v6_rmt_addr;
 481                 if (np->repflow && ireq->pktopts)
 482                         fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
 483
 484                 rcu_read_lock();
 485                 opt = ireq->ipv6_opt;
 486                 if (!opt)
 487                         opt = rcu_dereference(np->opt);
 488                 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt, np->tclass);
 489                 rcu_read_unlock();
 490                 err = net_xmit_eval(err);
 491         }
 492
 493 done:
 494         return err;
 495 }
 496
 497
 498 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 499 {
 500         kfree(inet_rsk(req)->ipv6_opt);
 501         kfree_skb(inet_rsk(req)->pktopts);
 502 }
 503
 504 #ifdef CONFIG_TCP_MD5SIG
 505 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
 506                                                    const struct in6_addr *addr)
 507 {
 508         return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
 509 }
 510
 511 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
 512                                                 const struct sock *addr_sk)
 513 {
 514         return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
 515 }
 516
 517 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
 518                                  int optlen)
 519 {
 520         struct tcp_md5sig cmd;
 521         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 522
 523         if (optlen < sizeof(cmd))
 524                 return -EINVAL;
 525
 526         if (copy_from_user(&cmd, optval, sizeof(cmd)))
 527                 return -EFAULT;
 528
 529         if (sin6->sin6_family != AF_INET6)
 530                 return -EINVAL;
 531
 532         if (!cmd.tcpm_keylen) {
 533                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 534                         return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 535                                               AF_INET);
 536                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 537                                       AF_INET6);
 538         }
 539
 540         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 541                 return -EINVAL;
 542
 543         if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 544                 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 545                                       AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 546
 547         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 548                               AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 549 }
 550
 551 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
 552                                    const struct in6_addr *daddr,
 553                                    const struct in6_addr *saddr,
 554                                    const struct tcphdr *th, int nbytes)
 555 {
 556         struct tcp6_pseudohdr *bp;
 557         struct scatterlist sg;
 558         struct tcphdr *_th;
 559
 560         bp = hp->scratch;
 561         /* 1. TCP pseudo-header (RFC2460) */
 562         bp->saddr = *saddr;
 563         bp->daddr = *daddr;
 564         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 565         bp->len = cpu_to_be32(nbytes);
 566
 567         _th = (struct tcphdr *)(bp + 1);
 568         memcpy(_th, th, sizeof(*th));
 569         _th->check = 0;
 570
 571         sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
 572         ahash_request_set_crypt(hp->md5_req, &sg, NULL,
 573                                 sizeof(*bp) + sizeof(*th));
 574         return crypto_ahash_update(hp->md5_req);
 575 }
 576
 577 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
 578                                const struct in6_addr *daddr, struct in6_addr *saddr,
 579                                const struct tcphdr *th)
 580 {
 581         struct tcp_md5sig_pool *hp;
 582         struct ahash_request *req;
 583
 584         hp = tcp_get_md5sig_pool();
 585         if (!hp)
 586                 goto clear_hash_noput;
 587         req = hp->md5_req;
 588
 589         if (crypto_ahash_init(req))
 590                 goto clear_hash;
 591         if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
 592                 goto clear_hash;
 593         if (tcp_md5_hash_key(hp, key))
 594                 goto clear_hash;
 595         ahash_request_set_crypt(req, NULL, md5_hash, 0);
 596         if (crypto_ahash_final(req))
 597                 goto clear_hash;
 598
 599         tcp_put_md5sig_pool();
 600         return 0;
 601
 602 clear_hash:
 603         tcp_put_md5sig_pool();
 604 clear_hash_noput:
 605         memset(md5_hash, 0, 16);
 606         return 1;
 607 }
 608
 609 static int tcp_v6_md5_hash_skb(char *md5_hash,
 610                                const struct tcp_md5sig_key *key,
 611                                const struct sock *sk,
 612                                const struct sk_buff *skb)
 613 {
 614         const struct in6_addr *saddr, *daddr;
 615         struct tcp_md5sig_pool *hp;
 616         struct ahash_request *req;
 617         const struct tcphdr *th = tcp_hdr(skb);
 618
 619         if (sk) { /* valid for establish/request sockets */
 620                 saddr = &sk->sk_v6_rcv_saddr;
 621                 daddr = &sk->sk_v6_daddr;
 622         } else {
 623                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 624                 saddr = &ip6h->saddr;
 625                 daddr = &ip6h->daddr;
 626         }
 627
 628         hp = tcp_get_md5sig_pool();
 629         if (!hp)
 630                 goto clear_hash_noput;
 631         req = hp->md5_req;
 632
 633         if (crypto_ahash_init(req))
 634                 goto clear_hash;
 635
 636         if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
 637                 goto clear_hash;
 638         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 639                 goto clear_hash;
 640         if (tcp_md5_hash_key(hp, key))
 641                 goto clear_hash;
 642         ahash_request_set_crypt(req, NULL, md5_hash, 0);
 643         if (crypto_ahash_final(req))
 644                 goto clear_hash;
 645
 646         tcp_put_md5sig_pool();
 647         return 0;
 648
 649 clear_hash:
 650         tcp_put_md5sig_pool();
 651 clear_hash_noput:
 652         memset(md5_hash, 0, 16);
 653         return 1;
 654 }
 655
 656 #endif
 657
 658 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
 659                                     const struct sk_buff *skb)
 660 {
 661 #ifdef CONFIG_TCP_MD5SIG
 662         const __u8 *hash_location = NULL;
 663         struct tcp_md5sig_key *hash_expected;
 664         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 665         const struct tcphdr *th = tcp_hdr(skb);
 666         int genhash;
 667         u8 newhash[16];
 668
 669         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 670         hash_location = tcp_parse_md5sig_option(th);
 671
 672         /* We've parsed the options - do we have a hash? */
 673         if (!hash_expected && !hash_location)
 674                 return false;
 675
 676         if (hash_expected && !hash_location) {
 677                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 678                 return true;
 679         }
 680
 681         if (!hash_expected && hash_location) {
 682                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 683                 return true;
 684         }
 685
 686         /* check the signature */
 687         genhash = tcp_v6_md5_hash_skb(newhash,
 688                                       hash_expected,
 689                                       NULL, skb);
 690
 691         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 692                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
 693                 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
 694                                      genhash ? "failed" : "mismatch",
 695                                      &ip6h->saddr, ntohs(th->source),
 696                                      &ip6h->daddr, ntohs(th->dest));
 697                 return true;
 698         }
 699 #endif
 700         return false;
 701 }
 702
 703 static void tcp_v6_init_req(struct request_sock *req,
 704                             const struct sock *sk_listener,
 705                             struct sk_buff *skb)
 706 {
 707         struct inet_request_sock *ireq = inet_rsk(req);
 708         const struct ipv6_pinfo *np = inet6_sk(sk_listener);
 709
 710         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 711         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 712
 713         /* So that link locals have meaning */
 714         if (!sk_listener->sk_bound_dev_if &&
 715             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 716                 ireq->ir_iif = tcp_v6_iif(skb);
 717
 718         if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
 719             (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
 720              np->rxopt.bits.rxinfo ||
 721              np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
 722              np->rxopt.bits.rxohlim || np->repflow)) {
 723                 atomic_inc(&skb->users);
 724                 ireq->pktopts = skb;
 725         }
 726 }
 727
 728 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
 729                                           struct flowi *fl,
 730                                           const struct request_sock *req)
 731 {
 732         return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
 733 }
 734
 735 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 736         .family         =       AF_INET6,
 737         .obj_size       =       sizeof(struct tcp6_request_sock),
 738         .rtx_syn_ack    =       tcp_rtx_synack,
 739         .send_ack       =       tcp_v6_reqsk_send_ack,
 740         .destructor     =       tcp_v6_reqsk_destructor,
 741         .send_reset     =       tcp_v6_send_reset,
 742         .syn_ack_timeout =      tcp_syn_ack_timeout,
 743 };
 744
 745 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 746         .mss_clamp      =       IPV6_MIN_MTU - sizeof(struct tcphdr) -
 747                                 sizeof(struct ipv6hdr),
 748 #ifdef CONFIG_TCP_MD5SIG
 749         .req_md5_lookup =       tcp_v6_md5_lookup,
 750         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 751 #endif
 752         .init_req       =       tcp_v6_init_req,
 753 #ifdef CONFIG_SYN_COOKIES
 754         .cookie_init_seq =      cookie_v6_init_sequence,
 755 #endif
 756         .route_req      =       tcp_v6_route_req,
 757         .init_seq       =       tcp_v6_init_seq,
 758         .init_ts_off    =       tcp_v6_init_ts_off,
 759         .send_synack    =       tcp_v6_send_synack,
 760 };
 761
 762 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
 763                                  u32 ack, u32 win, u32 tsval, u32 tsecr,
 764                                  int oif, struct tcp_md5sig_key *key, int rst,
 765                                  u8 tclass, __be32 label)
 766 {
 767         const struct tcphdr *th = tcp_hdr(skb);
 768         struct tcphdr *t1;
 769         struct sk_buff *buff;
 770         struct flowi6 fl6;
 771         struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 772         struct sock *ctl_sk = net->ipv6.tcp_sk;
 773         unsigned int tot_len = sizeof(struct tcphdr);
 774         struct dst_entry *dst;
 775         __be32 *topt;
 776
 777         if (tsecr)
 778                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 779 #ifdef CONFIG_TCP_MD5SIG
 780         if (key)
 781                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 782 #endif
 783
 784         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 785                          GFP_ATOMIC);
 786         if (!buff)
 787                 return;
 788
 789         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 790
 791         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 792         skb_reset_transport_header(buff);
 793
 794         /* Swap the send and the receive. */
 795         memset(t1, 0, sizeof(*t1));
 796         t1->dest = th->source;
 797         t1->source = th->dest;
 798         t1->doff = tot_len / 4;
 799         t1->seq = htonl(seq);
 800         t1->ack_seq = htonl(ack);
 801         t1->ack = !rst || !th->ack;
 802         t1->rst = rst;
 803         t1->window = htons(win);
 804
 805         topt = (__be32 *)(t1 + 1);
 806
 807         if (tsecr) {
 808                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 809                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 810                 *topt++ = htonl(tsval);
 811                 *topt++ = htonl(tsecr);
 812         }
 813
 814 #ifdef CONFIG_TCP_MD5SIG
 815         if (key) {
 816                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 817                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 818                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 819                                     &ipv6_hdr(skb)->saddr,
 820                                     &ipv6_hdr(skb)->daddr, t1);
 821         }
 822 #endif
 823
 824         memset(&fl6, 0, sizeof(fl6));
 825         fl6.daddr = ipv6_hdr(skb)->saddr;
 826         fl6.saddr = ipv6_hdr(skb)->daddr;
 827         fl6.flowlabel = label;
 828
 829         buff->ip_summed = CHECKSUM_PARTIAL;
 830         buff->csum = 0;
 831
 832         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
 833
 834         fl6.flowi6_proto = IPPROTO_TCP;
 835         if (rt6_need_strict(&fl6.daddr) && !oif)
 836                 fl6.flowi6_oif = tcp_v6_iif(skb);
 837         else {
 838                 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
 839                         oif = skb->skb_iif;
 840
 841                 fl6.flowi6_oif = oif;
 842         }
 843
 844         fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
 845         fl6.fl6_dport = t1->dest;
 846         fl6.fl6_sport = t1->source;
 847         fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
 848         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 849
 850         /* Pass a socket to ip6_dst_lookup either it is for RST
 851          * Underlying function will use this to retrieve the network
 852          * namespace
 853          */
 854         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 855         if (!IS_ERR(dst)) {
 856                 skb_dst_set(buff, dst);
 857                 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL, tclass);
 858                 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 859                 if (rst)
 860                         TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
 861                 return;
 862         }
 863
 864         kfree_skb(buff);
 865 }
 866
 867 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
 868 {
 869         const struct tcphdr *th = tcp_hdr(skb);
 870         u32 seq = 0, ack_seq = 0;
 871         struct tcp_md5sig_key *key = NULL;
 872 #ifdef CONFIG_TCP_MD5SIG
 873         const __u8 *hash_location = NULL;
 874         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 875         unsigned char newhash[16];
 876         int genhash;
 877         struct sock *sk1 = NULL;
 878 #endif
 879         int oif;
 880
 881         if (th->rst)
 882                 return;
 883
 884         /* If sk not NULL, it means we did a successful lookup and incoming
 885          * route had to be correct. prequeue might have dropped our dst.
 886          */
 887         if (!sk && !ipv6_unicast_destination(skb))
 888                 return;
 889
 890 #ifdef CONFIG_TCP_MD5SIG
 891         rcu_read_lock();
 892         hash_location = tcp_parse_md5sig_option(th);
 893         if (sk && sk_fullsock(sk)) {
 894                 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr);
 895         } else if (hash_location) {
 896                 /*
 897                  * active side is lost. Try to find listening socket through
 898                  * source port, and then find md5 key through listening socket.
 899                  * we are not loose security here:
 900                  * Incoming packet is checked with md5 hash with finding key,
 901                  * no RST generated if md5 hash doesn't match.
 902                  */
 903                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
 904                                            &tcp_hashinfo, NULL, 0,
 905                                            &ipv6h->saddr,
 906                                            th->source, &ipv6h->daddr,
 907                                            ntohs(th->source), tcp_v6_iif(skb));
 908                 if (!sk1)
 909                         goto out;
 910
 911                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 912                 if (!key)
 913                         goto out;
 914
 915                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
 916                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
 917                         goto out;
 918         }
 919 #endif
 920
 921         if (th->ack)
 922                 seq = ntohl(th->ack_seq);
 923         else
 924                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
 925                           (th->doff << 2);
 926
 927         oif = sk ? sk->sk_bound_dev_if : 0;
 928         tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
 929
 930 #ifdef CONFIG_TCP_MD5SIG
 931 out:
 932         rcu_read_unlock();
 933 #endif
 934 }
 935
 936 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
 937                             u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
 938                             struct tcp_md5sig_key *key, u8 tclass,
 939                             __be32 label)
 940 {
 941         tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
 942                              tclass, label);
 943 }
 944
 945 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 946 {
 947         struct inet_timewait_sock *tw = inet_twsk(sk);
 948         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 949
 950         tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 951                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 952                         tcp_time_stamp + tcptw->tw_ts_offset,
 953                         tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
 954                         tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
 955
 956         inet_twsk_put(tw);
 957 }
 958
 959 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 960                                   struct request_sock *req)
 961 {
 962         /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 963          * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 964          */
 965         /* RFC 7323 2.3
 966          * The window field (SEG.WND) of every outgoing segment, with the
 967          * exception of <SYN> segments, MUST be right-shifted by
 968          * Rcv.Wind.Shift bits:
 969          */
 970         tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
 971                         tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
 972                         tcp_rsk(req)->rcv_nxt,
 973                         req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
 974                         tcp_time_stamp + tcp_rsk(req)->ts_off,
 975                         req->ts_recent, sk->sk_bound_dev_if,
 976                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
 977                         0, 0);
 978 }
 979
 980
 981 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
 982 {
 983 #ifdef CONFIG_SYN_COOKIES
 984         const struct tcphdr *th = tcp_hdr(skb);
 985
 986         if (!th->syn)
 987                 sk = cookie_v6_check(sk, skb);
 988 #endif
 989         return sk;
 990 }
 991
 992 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 993 {
 994         if (skb->protocol == htons(ETH_P_IP))
 995                 return tcp_v4_conn_request(sk, skb);
 996
 997         if (!ipv6_unicast_destination(skb))
 998                 goto drop;
 999
1000         return tcp_conn_request(&tcp6_request_sock_ops,
1001                                 &tcp_request_sock_ipv6_ops, sk, skb);
1002
1003 drop:
1004         tcp_listendrop(sk);
1005         return 0; /* don't send reset */
1006 }
1007
1008 static void tcp_v6_restore_cb(struct sk_buff *skb)
1009 {
1010         /* We need to move header back to the beginning if xfrm6_policy_check()
1011          * and tcp_v6_fill_cb() are going to be called again.
1012          * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1013          */
1014         memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1015                 sizeof(struct inet6_skb_parm));
1016 }
1017
1018 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1019                                          struct request_sock *req,
1020                                          struct dst_entry *dst,
1021                                          struct request_sock *req_unhash,
1022                                          bool *own_req)
1023 {
1024         struct inet_request_sock *ireq;
1025         struct ipv6_pinfo *newnp;
1026         const struct ipv6_pinfo *np = inet6_sk(sk);
1027         struct ipv6_txoptions *opt;
1028         struct tcp6_sock *newtcp6sk;
1029         struct inet_sock *newinet;
1030         struct tcp_sock *newtp;
1031         struct sock *newsk;
1032 #ifdef CONFIG_TCP_MD5SIG
1033         struct tcp_md5sig_key *key;
1034 #endif
1035         struct flowi6 fl6;
1036
1037         if (skb->protocol == htons(ETH_P_IP)) {
1038                 /*
1039                  *      v6 mapped
1040                  */
1041
1042                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1043                                              req_unhash, own_req);
1044
1045                 if (!newsk)
1046                         return NULL;
1047
1048                 newtcp6sk = (struct tcp6_sock *)newsk;
1049                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1050
1051                 newinet = inet_sk(newsk);
1052                 newnp = inet6_sk(newsk);
1053                 newtp = tcp_sk(newsk);
1054
1055                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1056
1057                 newnp->saddr = newsk->sk_v6_rcv_saddr;
1058
1059                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1060                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1061 #ifdef CONFIG_TCP_MD5SIG
1062                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1063 #endif
1064
1065                 newnp->ipv6_mc_list = NULL;
1066                 newnp->ipv6_ac_list = NULL;
1067                 newnp->ipv6_fl_list = NULL;
1068                 newnp->pktoptions  = NULL;
1069                 newnp->opt         = NULL;
1070                 newnp->mcast_oif   = tcp_v6_iif(skb);
1071                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1072                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1073                 if (np->repflow)
1074                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1075
1076                 /*
1077                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1078                  * here, tcp_create_openreq_child now does this for us, see the comment in
1079                  * that function for the gory details. -acme
1080                  */
1081
1082                 /* It is tricky place. Until this moment IPv4 tcp
1083                    worked with IPv6 icsk.icsk_af_ops.
1084                    Sync it now.
1085                  */
1086                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1087
1088                 return newsk;
1089         }
1090
1091         ireq = inet_rsk(req);
1092
1093         if (sk_acceptq_is_full(sk))
1094                 goto out_overflow;
1095
1096         if (!dst) {
1097                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1098                 if (!dst)
1099                         goto out;
1100         }
1101
1102         newsk = tcp_create_openreq_child(sk, req, skb);
1103         if (!newsk)
1104                 goto out_nonewsk;
1105
1106         /*
1107          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1108          * count here, tcp_create_openreq_child now does this for us, see the
1109          * comment in that function for the gory details. -acme
1110          */
1111
1112         newsk->sk_gso_type = SKB_GSO_TCPV6;
1113         ip6_dst_store(newsk, dst, NULL, NULL);
1114         inet6_sk_rx_dst_set(newsk, skb);
1115
1116         newtcp6sk = (struct tcp6_sock *)newsk;
1117         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1118
1119         newtp = tcp_sk(newsk);
1120         newinet = inet_sk(newsk);
1121         newnp = inet6_sk(newsk);
1122
1123         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1124
1125         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1126         newnp->saddr = ireq->ir_v6_loc_addr;
1127         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1128         newsk->sk_bound_dev_if = ireq->ir_iif;
1129
1130         /* Now IPv6 options...
1131
1132            First: no IPv4 options.
1133          */
1134         newinet->inet_opt = NULL;
1135         newnp->ipv6_mc_list = NULL;
1136         newnp->ipv6_ac_list = NULL;
1137         newnp->ipv6_fl_list = NULL;
1138
1139         /* Clone RX bits */
1140         newnp->rxopt.all = np->rxopt.all;
1141
1142         newnp->pktoptions = NULL;
1143         newnp->opt        = NULL;
1144         newnp->mcast_oif  = tcp_v6_iif(skb);
1145         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1146         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1147         if (np->repflow)
1148                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1149
1150         /* Clone native IPv6 options from listening socket (if any)
1151
1152            Yes, keeping reference count would be much more clever,
1153            but we make one more one thing there: reattach optmem
1154            to newsk.
1155          */
1156         opt = ireq->ipv6_opt;
1157         if (!opt)
1158                 opt = rcu_dereference(np->opt);
1159         if (opt) {
1160                 opt = ipv6_dup_options(newsk, opt);
1161                 RCU_INIT_POINTER(newnp->opt, opt);
1162         }
1163         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1164         if (opt)
1165                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1166                                                     opt->opt_flen;
1167
1168         tcp_ca_openreq_child(newsk, dst);
1169
1170         tcp_sync_mss(newsk, dst_mtu(dst));
1171         newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1172
1173         tcp_initialize_rcv_mss(newsk);
1174
1175         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1176         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1177
1178 #ifdef CONFIG_TCP_MD5SIG
1179         /* Copy over the MD5 key from the original socket */
1180         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1181         if (key) {
1182                 /* We're using one, so create a matching key
1183                  * on the newsk structure. If we fail to get
1184                  * memory, then we end up not copying the key
1185                  * across. Shucks.
1186                  */
1187                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1188                                AF_INET6, key->key, key->keylen,
1189                                sk_gfp_mask(sk, GFP_ATOMIC));
1190         }
1191 #endif
1192
1193         if (__inet_inherit_port(sk, newsk) < 0) {
1194                 inet_csk_prepare_forced_close(newsk);
1195                 tcp_done(newsk);
1196                 goto out;
1197         }
1198         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1199         if (*own_req) {
1200                 tcp_move_syn(newtp, req);
1201
1202                 /* Clone pktoptions received with SYN, if we own the req */
1203                 if (ireq->pktopts) {
1204                         newnp->pktoptions = skb_clone(ireq->pktopts,
1205                                                       sk_gfp_mask(sk, GFP_ATOMIC));
1206                         consume_skb(ireq->pktopts);
1207                         ireq->pktopts = NULL;
1208                         if (newnp->pktoptions) {
1209                                 tcp_v6_restore_cb(newnp->pktoptions);
1210                                 skb_set_owner_r(newnp->pktoptions, newsk);
1211                         }
1212                 }
1213         }
1214
1215         return newsk;
1216
1217 out_overflow:
1218         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1219 out_nonewsk:
1220         dst_release(dst);
1221 out:
1222         tcp_listendrop(sk);
1223         return NULL;
1224 }
1225
1226 /* The socket must have it's spinlock held when we get
1227  * here, unless it is a TCP_LISTEN socket.
1228  *
1229  * We have a potential double-lock case here, so even when
1230  * doing backlog processing we use the BH locking scheme.
1231  * This is because we cannot sleep with the original spinlock
1232  * held.
1233  */
1234 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1235 {
1236         struct ipv6_pinfo *np = inet6_sk(sk);
1237         struct tcp_sock *tp;
1238         struct sk_buff *opt_skb = NULL;
1239
1240         /* Imagine: socket is IPv6. IPv4 packet arrives,
1241            goes to IPv4 receive handler and backlogged.
1242            From backlog it always goes here. Kerboom...
1243            Fortunately, tcp_rcv_established and rcv_established
1244            handle them correctly, but it is not case with
1245            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1246          */
1247
1248         if (skb->protocol == htons(ETH_P_IP))
1249                 return tcp_v4_do_rcv(sk, skb);
1250
1251         if (tcp_filter(sk, skb))
1252                 goto discard;
1253
1254         /*
1255          *      socket locking is here for SMP purposes as backlog rcv
1256          *      is currently called with bh processing disabled.
1257          */
1258
1259         /* Do Stevens' IPV6_PKTOPTIONS.
1260
1261            Yes, guys, it is the only place in our code, where we
1262            may make it not affecting IPv4.
1263            The rest of code is protocol independent,
1264            and I do not like idea to uglify IPv4.
1265
1266            Actually, all the idea behind IPV6_PKTOPTIONS
1267            looks not very well thought. For now we latch
1268            options, received in the last packet, enqueued
1269            by tcp. Feel free to propose better solution.
1270                                                --ANK (980728)
1271          */
1272         if (np->rxopt.all)
1273                 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1274
1275         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1276                 struct dst_entry *dst = sk->sk_rx_dst;
1277
1278                 sock_rps_save_rxhash(sk, skb);
1279                 sk_mark_napi_id(sk, skb);
1280                 if (dst) {
1281                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1282                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1283                                 dst_release(dst);
1284                                 sk->sk_rx_dst = NULL;
1285                         }
1286                 }
1287
1288                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1289                 if (opt_skb)
1290                         goto ipv6_pktoptions;
1291                 return 0;
1292         }
1293
1294         if (tcp_checksum_complete(skb))
1295                 goto csum_err;
1296
1297         if (sk->sk_state == TCP_LISTEN) {
1298                 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1299
1300                 if (!nsk)
1301                         goto discard;
1302
1303                 if (nsk != sk) {
1304                         if (tcp_child_process(sk, nsk, skb))
1305                                 goto reset;
1306                         if (opt_skb)
1307                                 __kfree_skb(opt_skb);
1308                         return 0;
1309                 }
1310         } else
1311                 sock_rps_save_rxhash(sk, skb);
1312
1313         if (tcp_rcv_state_process(sk, skb))
1314                 goto reset;
1315         if (opt_skb)
1316                 goto ipv6_pktoptions;
1317         return 0;
1318
1319 reset:
1320         tcp_v6_send_reset(sk, skb);
1321 discard:
1322         if (opt_skb)
1323                 __kfree_skb(opt_skb);
1324         kfree_skb(skb);
1325         return 0;
1326 csum_err:
1327         TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1328         TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1329         goto discard;
1330
1331
1332 ipv6_pktoptions:
1333         /* Do you ask, what is it?
1334
1335            1. skb was enqueued by tcp.
1336            2. skb is added to tail of read queue, rather than out of order.
1337            3. socket is not in passive state.
1338            4. Finally, it really contains options, which user wants to receive.
1339          */
1340         tp = tcp_sk(sk);
1341         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1342             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1343                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1344                         np->mcast_oif = tcp_v6_iif(opt_skb);
1345                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1346                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1347                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1348                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1349                 if (np->repflow)
1350                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1351                 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1352                         skb_set_owner_r(opt_skb, sk);
1353                         tcp_v6_restore_cb(opt_skb);
1354                         opt_skb = xchg(&np->pktoptions, opt_skb);
1355                 } else {
1356                         __kfree_skb(opt_skb);
1357                         opt_skb = xchg(&np->pktoptions, NULL);
1358                 }
1359         }
1360
1361         kfree_skb(opt_skb);
1362         return 0;
1363 }
1364
1365 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1366                            const struct tcphdr *th)
1367 {
1368         /* This is tricky: we move IP6CB at its correct location into
1369          * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1370          * _decode_session6() uses IP6CB().
1371          * barrier() makes sure compiler won't play aliasing games.
1372          */
1373         memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1374                 sizeof(struct inet6_skb_parm));
1375         barrier();
1376
1377         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1378         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1379                                     skb->len - th->doff*4);
1380         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1381         TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1382         TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1383         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1384         TCP_SKB_CB(skb)->sacked = 0;
1385 }
1386
1387 static int tcp_v6_rcv(struct sk_buff *skb)
1388 {
1389         const struct tcphdr *th;
1390         const struct ipv6hdr *hdr;
1391         bool refcounted;
1392         struct sock *sk;
1393         int ret;
1394         struct net *net = dev_net(skb->dev);
1395
1396         if (skb->pkt_type != PACKET_HOST)
1397                 goto discard_it;
1398
1399         /*
1400          *      Count it even if it's bad.
1401          */
1402         __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1403
1404         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1405                 goto discard_it;
1406
1407         th = (const struct tcphdr *)skb->data;
1408
1409         if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1410                 goto bad_packet;
1411         if (!pskb_may_pull(skb, th->doff*4))
1412                 goto discard_it;
1413
1414         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1415                 goto csum_error;
1416
1417         th = (const struct tcphdr *)skb->data;
1418         hdr = ipv6_hdr(skb);
1419
1420 lookup:
1421         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1422                                 th->source, th->dest, inet6_iif(skb),
1423                                 &refcounted);
1424         if (!sk)
1425                 goto no_tcp_socket;
1426
1427 process:
1428         if (sk->sk_state == TCP_TIME_WAIT)
1429                 goto do_time_wait;
1430
1431         if (sk->sk_state == TCP_NEW_SYN_RECV) {
1432                 struct request_sock *req = inet_reqsk(sk);
1433                 struct sock *nsk;
1434
1435                 sk = req->rsk_listener;
1436                 tcp_v6_fill_cb(skb, hdr, th);
1437                 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1438                         sk_drops_add(sk, skb);
1439                         reqsk_put(req);
1440                         goto discard_it;
1441                 }
1442                 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1443                         inet_csk_reqsk_queue_drop_and_put(sk, req);
1444                         goto lookup;
1445                 }
1446                 sock_hold(sk);
1447                 refcounted = true;
1448                 nsk = tcp_check_req(sk, skb, req, false);
1449                 if (!nsk) {
1450                         reqsk_put(req);
1451                         goto discard_and_relse;
1452                 }
1453                 if (nsk == sk) {
1454                         reqsk_put(req);
1455                         tcp_v6_restore_cb(skb);
1456                 } else if (tcp_child_process(sk, nsk, skb)) {
1457                         tcp_v6_send_reset(nsk, skb);
1458                         goto discard_and_relse;
1459                 } else {
1460                         sock_put(sk);
1461                         return 0;
1462                 }
1463         }
1464         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1465                 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1466                 goto discard_and_relse;
1467         }
1468
1469         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1470                 goto discard_and_relse;
1471
1472         tcp_v6_fill_cb(skb, hdr, th);
1473
1474         if (tcp_v6_inbound_md5_hash(sk, skb))
1475                 goto discard_and_relse;
1476
1477         if (tcp_filter(sk, skb))
1478                 goto discard_and_relse;
1479         th = (const struct tcphdr *)skb->data;
1480         hdr = ipv6_hdr(skb);
1481
1482         skb->dev = NULL;
1483
1484         if (sk->sk_state == TCP_LISTEN) {
1485                 ret = tcp_v6_do_rcv(sk, skb);
1486                 goto put_and_return;
1487         }
1488
1489         sk_incoming_cpu_update(sk);
1490
1491         bh_lock_sock_nested(sk);
1492         tcp_segs_in(tcp_sk(sk), skb);
1493         ret = 0;
1494         if (!sock_owned_by_user(sk)) {
1495                 if (!tcp_prequeue(sk, skb))
1496                         ret = tcp_v6_do_rcv(sk, skb);
1497         } else if (tcp_add_backlog(sk, skb)) {
1498                 goto discard_and_relse;
1499         }
1500         bh_unlock_sock(sk);
1501
1502 put_and_return:
1503         if (refcounted)
1504                 sock_put(sk);
1505         return ret ? -1 : 0;
1506
1507 no_tcp_socket:
1508         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1509                 goto discard_it;
1510
1511         tcp_v6_fill_cb(skb, hdr, th);
1512
1513         if (tcp_checksum_complete(skb)) {
1514 csum_error:
1515                 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1516 bad_packet:
1517                 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1518         } else {
1519                 tcp_v6_send_reset(NULL, skb);
1520         }
1521
1522 discard_it:
1523         kfree_skb(skb);
1524         return 0;
1525
1526 discard_and_relse:
1527         sk_drops_add(sk, skb);
1528         if (refcounted)
1529                 sock_put(sk);
1530         goto discard_it;
1531
1532 do_time_wait:
1533         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1534                 inet_twsk_put(inet_twsk(sk));
1535                 goto discard_it;
1536         }
1537
1538         tcp_v6_fill_cb(skb, hdr, th);
1539
1540         if (tcp_checksum_complete(skb)) {
1541                 inet_twsk_put(inet_twsk(sk));
1542                 goto csum_error;
1543         }
1544
1545         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1546         case TCP_TW_SYN:
1547         {
1548                 struct sock *sk2;
1549
1550                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1551                                             skb, __tcp_hdrlen(th),
1552                                             &ipv6_hdr(skb)->saddr, th->source,
1553                                             &ipv6_hdr(skb)->daddr,
1554                                             ntohs(th->dest), tcp_v6_iif(skb));
1555                 if (sk2) {
1556                         struct inet_timewait_sock *tw = inet_twsk(sk);
1557                         inet_twsk_deschedule_put(tw);
1558                         sk = sk2;
1559                         tcp_v6_restore_cb(skb);
1560                         refcounted = false;
1561                         goto process;
1562                 }
1563                 /* Fall through to ACK */
1564         }
1565         case TCP_TW_ACK:
1566                 tcp_v6_timewait_ack(sk, skb);
1567                 break;
1568         case TCP_TW_RST:
1569                 tcp_v6_restore_cb(skb);
1570                 tcp_v6_send_reset(sk, skb);
1571                 inet_twsk_deschedule_put(inet_twsk(sk));
1572                 goto discard_it;
1573         case TCP_TW_SUCCESS:
1574                 ;
1575         }
1576         goto discard_it;
1577 }
1578
1579 static void tcp_v6_early_demux(struct sk_buff *skb)
1580 {
1581         const struct ipv6hdr *hdr;
1582         const struct tcphdr *th;
1583         struct sock *sk;
1584
1585         if (skb->pkt_type != PACKET_HOST)
1586                 return;
1587
1588         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1589                 return;
1590
1591         hdr = ipv6_hdr(skb);
1592         th = tcp_hdr(skb);
1593
1594         if (th->doff < sizeof(struct tcphdr) / 4)
1595                 return;
1596
1597         /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1598         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1599                                         &hdr->saddr, th->source,
1600                                         &hdr->daddr, ntohs(th->dest),
1601                                         inet6_iif(skb));
1602         if (sk) {
1603                 skb->sk = sk;
1604                 skb->destructor = sock_edemux;
1605                 if (sk_fullsock(sk)) {
1606                         struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1607
1608                         if (dst)
1609                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1610                         if (dst &&
1611                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1612                                 skb_dst_set_noref(skb, dst);
1613                 }
1614         }
1615 }
1616
1617 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1618         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1619         .twsk_unique    = tcp_twsk_unique,
1620         .twsk_destructor = tcp_twsk_destructor,
1621 };
1622
1623 static const struct inet_connection_sock_af_ops ipv6_specific = {
1624         .queue_xmit        = inet6_csk_xmit,
1625         .send_check        = tcp_v6_send_check,
1626         .rebuild_header    = inet6_sk_rebuild_header,
1627         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1628         .conn_request      = tcp_v6_conn_request,
1629         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1630         .net_header_len    = sizeof(struct ipv6hdr),
1631         .net_frag_header_len = sizeof(struct frag_hdr),
1632         .setsockopt        = ipv6_setsockopt,
1633         .getsockopt        = ipv6_getsockopt,
1634         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1635         .sockaddr_len      = sizeof(struct sockaddr_in6),
1636 #ifdef CONFIG_COMPAT
1637         .compat_setsockopt = compat_ipv6_setsockopt,
1638         .compat_getsockopt = compat_ipv6_getsockopt,
1639 #endif
1640         .mtu_reduced       = tcp_v6_mtu_reduced,
1641 };
1642
1643 #ifdef CONFIG_TCP_MD5SIG
1644 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1645         .md5_lookup     =       tcp_v6_md5_lookup,
1646         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1647         .md5_parse      =       tcp_v6_parse_md5_keys,
1648 };
1649 #endif
1650
1651 /*
1652  *      TCP over IPv4 via INET6 API
1653  */
1654 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1655         .queue_xmit        = ip_queue_xmit,
1656         .send_check        = tcp_v4_send_check,
1657         .rebuild_header    = inet_sk_rebuild_header,
1658         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1659         .conn_request      = tcp_v6_conn_request,
1660         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1661         .net_header_len    = sizeof(struct iphdr),
1662         .setsockopt        = ipv6_setsockopt,
1663         .getsockopt        = ipv6_getsockopt,
1664         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1665         .sockaddr_len      = sizeof(struct sockaddr_in6),
1666 #ifdef CONFIG_COMPAT
1667         .compat_setsockopt = compat_ipv6_setsockopt,
1668         .compat_getsockopt = compat_ipv6_getsockopt,
1669 #endif
1670         .mtu_reduced       = tcp_v4_mtu_reduced,
1671 };
1672
1673 #ifdef CONFIG_TCP_MD5SIG
1674 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1675         .md5_lookup     =       tcp_v4_md5_lookup,
1676         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1677         .md5_parse      =       tcp_v6_parse_md5_keys,
1678 };
1679 #endif
1680
1681 /* NOTE: A lot of things set to zero explicitly by call to
1682  *       sk_alloc() so need not be done here.
1683  */
1684 static int tcp_v6_init_sock(struct sock *sk)
1685 {
1686         struct inet_connection_sock *icsk = inet_csk(sk);
1687
1688         tcp_init_sock(sk);
1689
1690         icsk->icsk_af_ops = &ipv6_specific;
1691
1692 #ifdef CONFIG_TCP_MD5SIG
1693         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1694 #endif
1695
1696         return 0;
1697 }
1698
1699 static void tcp_v6_destroy_sock(struct sock *sk)
1700 {
1701         tcp_v4_destroy_sock(sk);
1702         inet6_destroy_sock(sk);
1703 }
1704
1705 #ifdef CONFIG_PROC_FS
1706 /* Proc filesystem TCPv6 sock list dumping. */
1707 static void get_openreq6(struct seq_file *seq,
1708                          const struct request_sock *req, int i)
1709 {
1710         long ttd = req->rsk_timer.expires - jiffies;
1711         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1712         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1713
1714         if (ttd < 0)
1715                 ttd = 0;
1716
1717         seq_printf(seq,
1718                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1719                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1720                    i,
1721                    src->s6_addr32[0], src->s6_addr32[1],
1722                    src->s6_addr32[2], src->s6_addr32[3],
1723                    inet_rsk(req)->ir_num,
1724                    dest->s6_addr32[0], dest->s6_addr32[1],
1725                    dest->s6_addr32[2], dest->s6_addr32[3],
1726                    ntohs(inet_rsk(req)->ir_rmt_port),
1727                    TCP_SYN_RECV,
1728                    0, 0, /* could print option size, but that is af dependent. */
1729                    1,   /* timers active (only the expire timer) */
1730                    jiffies_to_clock_t(ttd),
1731                    req->num_timeout,
1732                    from_kuid_munged(seq_user_ns(seq),
1733                                     sock_i_uid(req->rsk_listener)),
1734                    0,  /* non standard timer */
1735                    0, /* open_requests have no inode */
1736                    0, req);
1737 }
1738
1739 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1740 {
1741         const struct in6_addr *dest, *src;
1742         __u16 destp, srcp;
1743         int timer_active;
1744         unsigned long timer_expires;
1745         const struct inet_sock *inet = inet_sk(sp);
1746         const struct tcp_sock *tp = tcp_sk(sp);
1747         const struct inet_connection_sock *icsk = inet_csk(sp);
1748         const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1749         int rx_queue;
1750         int state;
1751
1752         dest  = &sp->sk_v6_daddr;
1753         src   = &sp->sk_v6_rcv_saddr;
1754         destp = ntohs(inet->inet_dport);
1755         srcp  = ntohs(inet->inet_sport);
1756
1757         if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1758             icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
1759             icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1760                 timer_active    = 1;
1761                 timer_expires   = icsk->icsk_timeout;
1762         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1763                 timer_active    = 4;
1764                 timer_expires   = icsk->icsk_timeout;
1765         } else if (timer_pending(&sp->sk_timer)) {
1766                 timer_active    = 2;
1767                 timer_expires   = sp->sk_timer.expires;
1768         } else {
1769                 timer_active    = 0;
1770                 timer_expires = jiffies;
1771         }
1772
1773         state = sk_state_load(sp);
1774         if (state == TCP_LISTEN)
1775                 rx_queue = sp->sk_ack_backlog;
1776         else
1777                 /* Because we don't lock the socket,
1778                  * we might find a transient negative value.
1779                  */
1780                 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1781
1782         seq_printf(seq,
1783                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1784                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1785                    i,
1786                    src->s6_addr32[0], src->s6_addr32[1],
1787                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1788                    dest->s6_addr32[0], dest->s6_addr32[1],
1789                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1790                    state,
1791                    tp->write_seq - tp->snd_una,
1792                    rx_queue,
1793                    timer_active,
1794                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1795                    icsk->icsk_retransmits,
1796                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1797                    icsk->icsk_probes_out,
1798                    sock_i_ino(sp),
1799                    atomic_read(&sp->sk_refcnt), sp,
1800                    jiffies_to_clock_t(icsk->icsk_rto),
1801                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1802                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1803                    tp->snd_cwnd,
1804                    state == TCP_LISTEN ?
1805                         fastopenq->max_qlen :
1806                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1807                    );
1808 }
1809
1810 static void get_timewait6_sock(struct seq_file *seq,
1811                                struct inet_timewait_sock *tw, int i)
1812 {
1813         long delta = tw->tw_timer.expires - jiffies;
1814         const struct in6_addr *dest, *src;
1815         __u16 destp, srcp;
1816
1817         dest = &tw->tw_v6_daddr;
1818         src  = &tw->tw_v6_rcv_saddr;
1819         destp = ntohs(tw->tw_dport);
1820         srcp  = ntohs(tw->tw_sport);
1821
1822         seq_printf(seq,
1823                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1824                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1825                    i,
1826                    src->s6_addr32[0], src->s6_addr32[1],
1827                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1828                    dest->s6_addr32[0], dest->s6_addr32[1],
1829                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1830                    tw->tw_substate, 0, 0,
1831                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1832                    atomic_read(&tw->tw_refcnt), tw);
1833 }
1834
1835 static int tcp6_seq_show(struct seq_file *seq, void *v)
1836 {
1837         struct tcp_iter_state *st;
1838         struct sock *sk = v;
1839
1840         if (v == SEQ_START_TOKEN) {
1841                 seq_puts(seq,
1842                          "  sl  "
1843                          "local_address                         "
1844                          "remote_address                        "
1845                          "st tx_queue rx_queue tr tm->when retrnsmt"
1846                          "   uid  timeout inode\n");
1847                 goto out;
1848         }
1849         st = seq->private;
1850
1851         if (sk->sk_state == TCP_TIME_WAIT)
1852                 get_timewait6_sock(seq, v, st->num);
1853         else if (sk->sk_state == TCP_NEW_SYN_RECV)
1854                 get_openreq6(seq, v, st->num);
1855         else
1856                 get_tcp6_sock(seq, v, st->num);
1857 out:
1858         return 0;
1859 }
1860
1861 static const struct file_operations tcp6_afinfo_seq_fops = {
1862         .owner   = THIS_MODULE,
1863         .open    = tcp_seq_open,
1864         .read    = seq_read,
1865         .llseek  = seq_lseek,
1866         .release = seq_release_net
1867 };
1868
1869 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1870         .name           = "tcp6",
1871         .family         = AF_INET6,
1872         .seq_fops       = &tcp6_afinfo_seq_fops,
1873         .seq_ops        = {
1874                 .show           = tcp6_seq_show,
1875         },
1876 };
1877
1878 int __net_init tcp6_proc_init(struct net *net)
1879 {
1880         return tcp_proc_register(net, &tcp6_seq_afinfo);
1881 }
1882
1883 void tcp6_proc_exit(struct net *net)
1884 {
1885         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1886 }
1887 #endif
1888
1889 struct proto tcpv6_prot = {
1890         .name                   = "TCPv6",
1891         .owner                  = THIS_MODULE,
1892         .close                  = tcp_close,
1893         .connect                = tcp_v6_connect,
1894         .disconnect             = tcp_disconnect,
1895         .accept                 = inet_csk_accept,
1896         .ioctl                  = tcp_ioctl,
1897         .init                   = tcp_v6_init_sock,
1898         .destroy                = tcp_v6_destroy_sock,
1899         .shutdown               = tcp_shutdown,
1900         .setsockopt             = tcp_setsockopt,
1901         .getsockopt             = tcp_getsockopt,
1902         .keepalive              = tcp_set_keepalive,
1903         .recvmsg                = tcp_recvmsg,
1904         .sendmsg                = tcp_sendmsg,
1905         .sendpage               = tcp_sendpage,
1906         .backlog_rcv            = tcp_v6_do_rcv,
1907         .release_cb             = tcp_release_cb,
1908         .hash                   = inet6_hash,
1909         .unhash                 = inet_unhash,
1910         .get_port               = inet_csk_get_port,
1911         .enter_memory_pressure  = tcp_enter_memory_pressure,
1912         .stream_memory_free     = tcp_stream_memory_free,
1913         .sockets_allocated      = &tcp_sockets_allocated,
1914         .memory_allocated       = &tcp_memory_allocated,
1915         .memory_pressure        = &tcp_memory_pressure,
1916         .orphan_count           = &tcp_orphan_count,
1917         .sysctl_mem             = sysctl_tcp_mem,
1918         .sysctl_wmem            = sysctl_tcp_wmem,
1919         .sysctl_rmem            = sysctl_tcp_rmem,
1920         .max_header             = MAX_TCP_HEADER,
1921         .obj_size               = sizeof(struct tcp6_sock),
1922         .slab_flags             = SLAB_DESTROY_BY_RCU,
1923         .twsk_prot              = &tcp6_timewait_sock_ops,
1924         .rsk_prot               = &tcp6_request_sock_ops,
1925         .h.hashinfo             = &tcp_hashinfo,
1926         .no_autobind            = true,
1927 #ifdef CONFIG_COMPAT
1928         .compat_setsockopt      = compat_tcp_setsockopt,
1929         .compat_getsockopt      = compat_tcp_getsockopt,
1930 #endif
1931         .diag_destroy           = tcp_abort,
1932 };
1933
1934 static struct inet6_protocol tcpv6_protocol = {
1935         .early_demux    =       tcp_v6_early_demux,
1936         .early_demux_handler =  tcp_v6_early_demux,
1937         .handler        =       tcp_v6_rcv,
1938         .err_handler    =       tcp_v6_err,
1939         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1940 };
1941
1942 static struct inet_protosw tcpv6_protosw = {
1943         .type           =       SOCK_STREAM,
1944         .protocol       =       IPPROTO_TCP,
1945         .prot           =       &tcpv6_prot,
1946         .ops            =       &inet6_stream_ops,
1947         .flags          =       INET_PROTOSW_PERMANENT |
1948                                 INET_PROTOSW_ICSK,
1949 };
1950
1951 static int __net_init tcpv6_net_init(struct net *net)
1952 {
1953         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1954                                     SOCK_RAW, IPPROTO_TCP, net);
1955 }
1956
1957 static void __net_exit tcpv6_net_exit(struct net *net)
1958 {
1959         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1960 }
1961
1962 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1963 {
1964         inet_twsk_purge(&tcp_hashinfo, AF_INET6);
1965 }
1966
1967 static struct pernet_operations tcpv6_net_ops = {
1968         .init       = tcpv6_net_init,
1969         .exit       = tcpv6_net_exit,
1970         .exit_batch = tcpv6_net_exit_batch,
1971 };
1972
1973 int __init tcpv6_init(void)
1974 {
1975         int ret;
1976
1977         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1978         if (ret)
1979                 goto out;
1980
1981         /* register inet6 protocol */
1982         ret = inet6_register_protosw(&tcpv6_protosw);
1983         if (ret)
1984                 goto out_tcpv6_protocol;
1985
1986         ret = register_pernet_subsys(&tcpv6_net_ops);
1987         if (ret)
1988                 goto out_tcpv6_protosw;
1989 out:
1990         return ret;
1991
1992 out_tcpv6_protosw:
1993         inet6_unregister_protosw(&tcpv6_protosw);
1994 out_tcpv6_protocol:
1995         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1996         goto out;
1997 }
1998
1999 void tcpv6_exit(void)
2000 {
2001         unregister_pernet_subsys(&tcpv6_net_ops);
2002         inet6_unregister_protosw(&tcpv6_protosw);
2003         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2004 }