kernel/net/ipv6/tcp_ipv6.c

   1 /*
   2  *      TCP over IPv6
   3  *      Linux INET6 implementation
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      Based on:
   9  *      linux/net/ipv4/tcp.c
  10  *      linux/net/ipv4/tcp_input.c
  11  *      linux/net/ipv4/tcp_output.c
  12  *
  13  *      Fixes:
  14  *      Hideaki YOSHIFUJI       :       sin6_scope_id support
  15  *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
  16  *      Alexey Kuznetsov                allow both IPv4 and IPv6 sockets to bind
  17  *                                      a single port at the same time.
  18  *      YOSHIFUJI Hideaki @USAGI:       convert /proc/net/tcp6 to seq_file.
  19  *
  20  *      This program is free software; you can redistribute it and/or
  21  *      modify it under the terms of the GNU General Public License
  22  *      as published by the Free Software Foundation; either version
  23  *      2 of the License, or (at your option) any later version.
  24  */
  25
  26 #include <linux/bottom_half.h>
  27 #include <linux/module.h>
  28 #include <linux/errno.h>
  29 #include <linux/types.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/jiffies.h>
  34 #include <linux/in.h>
  35 #include <linux/in6.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/init.h>
  38 #include <linux/jhash.h>
  39 #include <linux/ipsec.h>
  40 #include <linux/times.h>
  41 #include <linux/slab.h>
  42 #include <linux/uaccess.h>
  43 #include <linux/ipv6.h>
  44 #include <linux/icmpv6.h>
  45 #include <linux/random.h>
  46
  47 #include <net/tcp.h>
  48 #include <net/ndisc.h>
  49 #include <net/inet6_hashtables.h>
  50 #include <net/inet6_connection_sock.h>
  51 #include <net/ipv6.h>
  52 #include <net/transp_v6.h>
  53 #include <net/addrconf.h>
  54 #include <net/ip6_route.h>
  55 #include <net/ip6_checksum.h>
  56 #include <net/inet_ecn.h>
  57 #include <net/protocol.h>
  58 #include <net/xfrm.h>
  59 #include <net/snmp.h>
  60 #include <net/dsfield.h>
  61 #include <net/timewait_sock.h>
  62 #include <net/inet_common.h>
  63 #include <net/secure_seq.h>
  64 #include <net/tcp_memcontrol.h>
  65 #include <net/busy_poll.h>
  66
  67 #include <linux/proc_fs.h>
  68 #include <linux/seq_file.h>
  69
  70 #include <linux/crypto.h>
  71 #include <linux/scatterlist.h>
  72
  73 static void     tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
  74 static void     tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
  75                                       struct request_sock *req);
  76
  77 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  78
  79 static const struct inet_connection_sock_af_ops ipv6_mapped;
  80 static const struct inet_connection_sock_af_ops ipv6_specific;
  81 #ifdef CONFIG_TCP_MD5SIG
  82 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
  83 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
  84 #else
  85 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
  86                                                    const struct in6_addr *addr)
  87 {
  88         return NULL;
  89 }
  90 #endif
  91
  92 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
  93 {
  94         struct dst_entry *dst = skb_dst(skb);
  95
  96         if (dst && dst_hold_safe(dst)) {
  97                 const struct rt6_info *rt = (const struct rt6_info *)dst;
  98
  99                 sk->sk_rx_dst = dst;
 100                 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
 101                 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
 102         }
 103 }
 104
 105 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
 106 {
 107         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
 108                                             ipv6_hdr(skb)->saddr.s6_addr32,
 109                                             tcp_hdr(skb)->dest,
 110                                             tcp_hdr(skb)->source);
 111 }
 112
 113 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 114                           int addr_len)
 115 {
 116         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 117         struct inet_sock *inet = inet_sk(sk);
 118         struct inet_connection_sock *icsk = inet_csk(sk);
 119         struct ipv6_pinfo *np = inet6_sk(sk);
 120         struct tcp_sock *tp = tcp_sk(sk);
 121         struct in6_addr *saddr = NULL, *final_p, final;
 122         struct ipv6_txoptions *opt;
 123         struct flowi6 fl6;
 124         struct dst_entry *dst;
 125         int addr_type;
 126         int err;
 127
 128         if (addr_len < SIN6_LEN_RFC2133)
 129                 return -EINVAL;
 130
 131         if (usin->sin6_family != AF_INET6)
 132                 return -EAFNOSUPPORT;
 133
 134         memset(&fl6, 0, sizeof(fl6));
 135
 136         if (np->sndflow) {
 137                 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 138                 IP6_ECN_flow_init(fl6.flowlabel);
 139                 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
 140                         struct ip6_flowlabel *flowlabel;
 141                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 142                         if (!flowlabel)
 143                                 return -EINVAL;
 144                         fl6_sock_release(flowlabel);
 145                 }
 146         }
 147
 148         /*
 149          *      connect() to INADDR_ANY means loopback (BSD'ism).
 150          */
 151
 152         if (ipv6_addr_any(&usin->sin6_addr))
 153                 usin->sin6_addr.s6_addr[15] = 0x1;
 154
 155         addr_type = ipv6_addr_type(&usin->sin6_addr);
 156
 157         if (addr_type & IPV6_ADDR_MULTICAST)
 158                 return -ENETUNREACH;
 159
 160         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 161                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 162                     usin->sin6_scope_id) {
 163                         /* If interface is set while binding, indices
 164                          * must coincide.
 165                          */
 166                         if (sk->sk_bound_dev_if &&
 167                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 168                                 return -EINVAL;
 169
 170                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 171                 }
 172
 173                 /* Connect to link-local address requires an interface */
 174                 if (!sk->sk_bound_dev_if)
 175                         return -EINVAL;
 176         }
 177
 178         if (tp->rx_opt.ts_recent_stamp &&
 179             !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
 180                 tp->rx_opt.ts_recent = 0;
 181                 tp->rx_opt.ts_recent_stamp = 0;
 182                 tp->write_seq = 0;
 183         }
 184
 185         sk->sk_v6_daddr = usin->sin6_addr;
 186         np->flow_label = fl6.flowlabel;
 187
 188         /*
 189          *      TCP over IPv4
 190          */
 191
 192         if (addr_type == IPV6_ADDR_MAPPED) {
 193                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 194                 struct sockaddr_in sin;
 195
 196                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 197
 198                 if (__ipv6_only_sock(sk))
 199                         return -ENETUNREACH;
 200
 201                 sin.sin_family = AF_INET;
 202                 sin.sin_port = usin->sin6_port;
 203                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 204
 205                 icsk->icsk_af_ops = &ipv6_mapped;
 206                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 207 #ifdef CONFIG_TCP_MD5SIG
 208                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
 209 #endif
 210
 211                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 212
 213                 if (err) {
 214                         icsk->icsk_ext_hdr_len = exthdrlen;
 215                         icsk->icsk_af_ops = &ipv6_specific;
 216                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 217 #ifdef CONFIG_TCP_MD5SIG
 218                         tp->af_specific = &tcp_sock_ipv6_specific;
 219 #endif
 220                         goto failure;
 221                 }
 222                 np->saddr = sk->sk_v6_rcv_saddr;
 223
 224                 return err;
 225         }
 226
 227         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 228                 saddr = &sk->sk_v6_rcv_saddr;
 229
 230         fl6.flowi6_proto = IPPROTO_TCP;
 231         fl6.daddr = sk->sk_v6_daddr;
 232         fl6.saddr = saddr ? *saddr : np->saddr;
 233         fl6.flowi6_oif = sk->sk_bound_dev_if;
 234         fl6.flowi6_mark = sk->sk_mark;
 235         fl6.fl6_dport = usin->sin6_port;
 236         fl6.fl6_sport = inet->inet_sport;
 237
 238         opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
 239         final_p = fl6_update_dst(&fl6, opt, &final);
 240
 241         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 242
 243         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 244         if (IS_ERR(dst)) {
 245                 err = PTR_ERR(dst);
 246                 goto failure;
 247         }
 248
 249         if (!saddr) {
 250                 saddr = &fl6.saddr;
 251                 sk->sk_v6_rcv_saddr = *saddr;
 252         }
 253
 254         /* set the source address */
 255         np->saddr = *saddr;
 256         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 257
 258         sk->sk_gso_type = SKB_GSO_TCPV6;
 259         ip6_dst_store(sk, dst, NULL, NULL);
 260
 261         if (tcp_death_row.sysctl_tw_recycle &&
 262             !tp->rx_opt.ts_recent_stamp &&
 263             ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr))
 264                 tcp_fetch_timewait_stamp(sk, dst);
 265
 266         icsk->icsk_ext_hdr_len = 0;
 267         if (opt)
 268                 icsk->icsk_ext_hdr_len = opt->opt_flen +
 269                                          opt->opt_nflen;
 270
 271         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 272
 273         inet->inet_dport = usin->sin6_port;
 274
 275         tcp_set_state(sk, TCP_SYN_SENT);
 276         err = inet6_hash_connect(&tcp_death_row, sk);
 277         if (err)
 278                 goto late_failure;
 279
 280         sk_set_txhash(sk);
 281
 282         if (!tp->write_seq && likely(!tp->repair))
 283                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 284                                                              sk->sk_v6_daddr.s6_addr32,
 285                                                              inet->inet_sport,
 286                                                              inet->inet_dport);
 287
 288         err = tcp_connect(sk);
 289         if (err)
 290                 goto late_failure;
 291
 292         return 0;
 293
 294 late_failure:
 295         tcp_set_state(sk, TCP_CLOSE);
 296         __sk_dst_reset(sk);
 297 failure:
 298         inet->inet_dport = 0;
 299         sk->sk_route_caps = 0;
 300         return err;
 301 }
 302
 303 static void tcp_v6_mtu_reduced(struct sock *sk)
 304 {
 305         struct dst_entry *dst;
 306
 307         if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 308                 return;
 309
 310         dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
 311         if (!dst)
 312                 return;
 313
 314         if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
 315                 tcp_sync_mss(sk, dst_mtu(dst));
 316                 tcp_simple_retransmit(sk);
 317         }
 318 }
 319
 320 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 321                 u8 type, u8 code, int offset, __be32 info)
 322 {
 323         const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
 324         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 325         struct net *net = dev_net(skb->dev);
 326         struct request_sock *fastopen;
 327         struct ipv6_pinfo *np;
 328         struct tcp_sock *tp;
 329         __u32 seq, snd_una;
 330         struct sock *sk;
 331         bool fatal;
 332         int err;
 333
 334         sk = __inet6_lookup_established(net, &tcp_hashinfo,
 335                                         &hdr->daddr, th->dest,
 336                                         &hdr->saddr, ntohs(th->source),
 337                                         skb->dev->ifindex);
 338
 339         if (!sk) {
 340                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 341                                    ICMP6_MIB_INERRORS);
 342                 return;
 343         }
 344
 345         if (sk->sk_state == TCP_TIME_WAIT) {
 346                 inet_twsk_put(inet_twsk(sk));
 347                 return;
 348         }
 349         seq = ntohl(th->seq);
 350         fatal = icmpv6_err_convert(type, code, &err);
 351         if (sk->sk_state == TCP_NEW_SYN_RECV)
 352                 return tcp_req_err(sk, seq, fatal);
 353
 354         bh_lock_sock(sk);
 355         if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
 356                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 357
 358         if (sk->sk_state == TCP_CLOSE)
 359                 goto out;
 360
 361         if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
 362                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
 363                 goto out;
 364         }
 365
 366         tp = tcp_sk(sk);
 367         /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 368         fastopen = tp->fastopen_rsk;
 369         snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 370         if (sk->sk_state != TCP_LISTEN &&
 371             !between(seq, snd_una, tp->snd_nxt)) {
 372                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 373                 goto out;
 374         }
 375
 376         np = inet6_sk(sk);
 377
 378         if (type == NDISC_REDIRECT) {
 379                 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 380
 381                 if (dst)
 382                         dst->ops->redirect(dst, sk, skb);
 383                 goto out;
 384         }
 385
 386         if (type == ICMPV6_PKT_TOOBIG) {
 387                 /* We are not interested in TCP_LISTEN and open_requests
 388                  * (SYN-ACKs send out by Linux are always <576bytes so
 389                  * they should go through unfragmented).
 390                  */
 391                 if (sk->sk_state == TCP_LISTEN)
 392                         goto out;
 393
 394                 if (!ip6_sk_accept_pmtu(sk))
 395                         goto out;
 396
 397                 tp->mtu_info = ntohl(info);
 398                 if (!sock_owned_by_user(sk))
 399                         tcp_v6_mtu_reduced(sk);
 400                 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
 401                                            &tp->tsq_flags))
 402                         sock_hold(sk);
 403                 goto out;
 404         }
 405
 406
 407         /* Might be for an request_sock */
 408         switch (sk->sk_state) {
 409         case TCP_SYN_SENT:
 410         case TCP_SYN_RECV:
 411                 /* Only in fast or simultaneous open. If a fast open socket is
 412                  * is already accepted it is treated as a connected one below.
 413                  */
 414                 if (fastopen && !fastopen->sk)
 415                         break;
 416
 417                 if (!sock_owned_by_user(sk)) {
 418                         sk->sk_err = err;
 419                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 420
 421                         tcp_done(sk);
 422                 } else
 423                         sk->sk_err_soft = err;
 424                 goto out;
 425         }
 426
 427         if (!sock_owned_by_user(sk) && np->recverr) {
 428                 sk->sk_err = err;
 429                 sk->sk_error_report(sk);
 430         } else
 431                 sk->sk_err_soft = err;
 432
 433 out:
 434         bh_unlock_sock(sk);
 435         sock_put(sk);
 436 }
 437
 438
 439 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 440                               struct flowi *fl,
 441                               struct request_sock *req,
 442                               struct tcp_fastopen_cookie *foc,
 443                               bool attach_req)
 444 {
 445         struct inet_request_sock *ireq = inet_rsk(req);
 446         struct ipv6_pinfo *np = inet6_sk(sk);
 447         struct flowi6 *fl6 = &fl->u.ip6;
 448         struct sk_buff *skb;
 449         int err = -ENOMEM;
 450
 451         /* First, grab a route. */
 452         if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
 453                                                IPPROTO_TCP)) == NULL)
 454                 goto done;
 455
 456         skb = tcp_make_synack(sk, dst, req, foc, attach_req);
 457
 458         if (skb) {
 459                 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
 460                                     &ireq->ir_v6_rmt_addr);
 461
 462                 fl6->daddr = ireq->ir_v6_rmt_addr;
 463                 if (np->repflow && ireq->pktopts)
 464                         fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
 465
 466                 rcu_read_lock();
 467                 err = ip6_xmit(sk, skb, fl6, rcu_dereference(np->opt),
 468                                np->tclass);
 469                 rcu_read_unlock();
 470                 err = net_xmit_eval(err);
 471         }
 472
 473 done:
 474         return err;
 475 }
 476
 477
 478 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 479 {
 480         kfree_skb(inet_rsk(req)->pktopts);
 481 }
 482
 483 #ifdef CONFIG_TCP_MD5SIG
 484 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
 485                                                    const struct in6_addr *addr)
 486 {
 487         return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
 488 }
 489
 490 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
 491                                                 const struct sock *addr_sk)
 492 {
 493         return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr);
 494 }
 495
 496 static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval,
 497                                  int optlen)
 498 {
 499         struct tcp_md5sig cmd;
 500         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 501
 502         if (optlen < sizeof(cmd))
 503                 return -EINVAL;
 504
 505         if (copy_from_user(&cmd, optval, sizeof(cmd)))
 506                 return -EFAULT;
 507
 508         if (sin6->sin6_family != AF_INET6)
 509                 return -EINVAL;
 510
 511         if (!cmd.tcpm_keylen) {
 512                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 513                         return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 514                                               AF_INET);
 515                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 516                                       AF_INET6);
 517         }
 518
 519         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 520                 return -EINVAL;
 521
 522         if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 523                 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 524                                       AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 525
 526         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 527                               AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 528 }
 529
 530 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
 531                                         const struct in6_addr *daddr,
 532                                         const struct in6_addr *saddr, int nbytes)
 533 {
 534         struct tcp6_pseudohdr *bp;
 535         struct scatterlist sg;
 536
 537         bp = &hp->md5_blk.ip6;
 538         /* 1. TCP pseudo-header (RFC2460) */
 539         bp->saddr = *saddr;
 540         bp->daddr = *daddr;
 541         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 542         bp->len = cpu_to_be32(nbytes);
 543
 544         sg_init_one(&sg, bp, sizeof(*bp));
 545         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
 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 hash_desc *desc;
 554
 555         hp = tcp_get_md5sig_pool();
 556         if (!hp)
 557                 goto clear_hash_noput;
 558         desc = &hp->md5_desc;
 559
 560         if (crypto_hash_init(desc))
 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         if (crypto_hash_final(desc, md5_hash))
 569                 goto clear_hash;
 570
 571         tcp_put_md5sig_pool();
 572         return 0;
 573
 574 clear_hash:
 575         tcp_put_md5sig_pool();
 576 clear_hash_noput:
 577         memset(md5_hash, 0, 16);
 578         return 1;
 579 }
 580
 581 static int tcp_v6_md5_hash_skb(char *md5_hash,
 582                                const struct tcp_md5sig_key *key,
 583                                const struct sock *sk,
 584                                const struct sk_buff *skb)
 585 {
 586         const struct in6_addr *saddr, *daddr;
 587         struct tcp_md5sig_pool *hp;
 588         struct hash_desc *desc;
 589         const struct tcphdr *th = tcp_hdr(skb);
 590
 591         if (sk) { /* valid for establish/request sockets */
 592                 saddr = &sk->sk_v6_rcv_saddr;
 593                 daddr = &sk->sk_v6_daddr;
 594         } else {
 595                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 596                 saddr = &ip6h->saddr;
 597                 daddr = &ip6h->daddr;
 598         }
 599
 600         hp = tcp_get_md5sig_pool();
 601         if (!hp)
 602                 goto clear_hash_noput;
 603         desc = &hp->md5_desc;
 604
 605         if (crypto_hash_init(desc))
 606                 goto clear_hash;
 607
 608         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
 609                 goto clear_hash;
 610         if (tcp_md5_hash_header(hp, th))
 611                 goto clear_hash;
 612         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 613                 goto clear_hash;
 614         if (tcp_md5_hash_key(hp, key))
 615                 goto clear_hash;
 616         if (crypto_hash_final(desc, md5_hash))
 617                 goto clear_hash;
 618
 619         tcp_put_md5sig_pool();
 620         return 0;
 621
 622 clear_hash:
 623         tcp_put_md5sig_pool();
 624 clear_hash_noput:
 625         memset(md5_hash, 0, 16);
 626         return 1;
 627 }
 628
 629 #endif
 630
 631 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
 632                                     const struct sk_buff *skb)
 633 {
 634 #ifdef CONFIG_TCP_MD5SIG
 635         const __u8 *hash_location = NULL;
 636         struct tcp_md5sig_key *hash_expected;
 637         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 638         const struct tcphdr *th = tcp_hdr(skb);
 639         int genhash;
 640         u8 newhash[16];
 641
 642         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 643         hash_location = tcp_parse_md5sig_option(th);
 644
 645         /* We've parsed the options - do we have a hash? */
 646         if (!hash_expected && !hash_location)
 647                 return false;
 648
 649         if (hash_expected && !hash_location) {
 650                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 651                 return true;
 652         }
 653
 654         if (!hash_expected && hash_location) {
 655                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 656                 return true;
 657         }
 658
 659         /* check the signature */
 660         genhash = tcp_v6_md5_hash_skb(newhash,
 661                                       hash_expected,
 662                                       NULL, skb);
 663
 664         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 665                 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
 666                                      genhash ? "failed" : "mismatch",
 667                                      &ip6h->saddr, ntohs(th->source),
 668                                      &ip6h->daddr, ntohs(th->dest));
 669                 return true;
 670         }
 671 #endif
 672         return false;
 673 }
 674
 675 static void tcp_v6_init_req(struct request_sock *req,
 676                             const struct sock *sk_listener,
 677                             struct sk_buff *skb)
 678 {
 679         struct inet_request_sock *ireq = inet_rsk(req);
 680         const struct ipv6_pinfo *np = inet6_sk(sk_listener);
 681
 682         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 683         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 684
 685         /* So that link locals have meaning */
 686         if (!sk_listener->sk_bound_dev_if &&
 687             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 688                 ireq->ir_iif = tcp_v6_iif(skb);
 689
 690         if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
 691             (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
 692              np->rxopt.bits.rxinfo ||
 693              np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
 694              np->rxopt.bits.rxohlim || np->repflow)) {
 695                 atomic_inc(&skb->users);
 696                 ireq->pktopts = skb;
 697         }
 698 }
 699
 700 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
 701                                           struct flowi *fl,
 702                                           const struct request_sock *req,
 703                                           bool *strict)
 704 {
 705         if (strict)
 706                 *strict = true;
 707         return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
 708 }
 709
 710 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 711         .family         =       AF_INET6,
 712         .obj_size       =       sizeof(struct tcp6_request_sock),
 713         .rtx_syn_ack    =       tcp_rtx_synack,
 714         .send_ack       =       tcp_v6_reqsk_send_ack,
 715         .destructor     =       tcp_v6_reqsk_destructor,
 716         .send_reset     =       tcp_v6_send_reset,
 717         .syn_ack_timeout =      tcp_syn_ack_timeout,
 718 };
 719
 720 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 721         .mss_clamp      =       IPV6_MIN_MTU - sizeof(struct tcphdr) -
 722                                 sizeof(struct ipv6hdr),
 723 #ifdef CONFIG_TCP_MD5SIG
 724         .req_md5_lookup =       tcp_v6_md5_lookup,
 725         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 726 #endif
 727         .init_req       =       tcp_v6_init_req,
 728 #ifdef CONFIG_SYN_COOKIES
 729         .cookie_init_seq =      cookie_v6_init_sequence,
 730 #endif
 731         .route_req      =       tcp_v6_route_req,
 732         .init_seq       =       tcp_v6_init_sequence,
 733         .send_synack    =       tcp_v6_send_synack,
 734 };
 735
 736 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
 737                                  u32 ack, u32 win, u32 tsval, u32 tsecr,
 738                                  int oif, struct tcp_md5sig_key *key, int rst,
 739                                  u8 tclass, u32 label)
 740 {
 741         const struct tcphdr *th = tcp_hdr(skb);
 742         struct tcphdr *t1;
 743         struct sk_buff *buff;
 744         struct flowi6 fl6;
 745         struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 746         struct sock *ctl_sk = net->ipv6.tcp_sk;
 747         unsigned int tot_len = sizeof(struct tcphdr);
 748         struct dst_entry *dst;
 749         __be32 *topt;
 750
 751         if (tsecr)
 752                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 753 #ifdef CONFIG_TCP_MD5SIG
 754         if (key)
 755                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 756 #endif
 757
 758         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 759                          GFP_ATOMIC);
 760         if (!buff)
 761                 return;
 762
 763         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 764
 765         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 766         skb_reset_transport_header(buff);
 767
 768         /* Swap the send and the receive. */
 769         memset(t1, 0, sizeof(*t1));
 770         t1->dest = th->source;
 771         t1->source = th->dest;
 772         t1->doff = tot_len / 4;
 773         t1->seq = htonl(seq);
 774         t1->ack_seq = htonl(ack);
 775         t1->ack = !rst || !th->ack;
 776         t1->rst = rst;
 777         t1->window = htons(win);
 778
 779         topt = (__be32 *)(t1 + 1);
 780
 781         if (tsecr) {
 782                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 783                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 784                 *topt++ = htonl(tsval);
 785                 *topt++ = htonl(tsecr);
 786         }
 787
 788 #ifdef CONFIG_TCP_MD5SIG
 789         if (key) {
 790                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 791                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 792                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 793                                     &ipv6_hdr(skb)->saddr,
 794                                     &ipv6_hdr(skb)->daddr, t1);
 795         }
 796 #endif
 797
 798         memset(&fl6, 0, sizeof(fl6));
 799         fl6.daddr = ipv6_hdr(skb)->saddr;
 800         fl6.saddr = ipv6_hdr(skb)->daddr;
 801         fl6.flowlabel = label;
 802
 803         buff->ip_summed = CHECKSUM_PARTIAL;
 804         buff->csum = 0;
 805
 806         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
 807
 808         fl6.flowi6_proto = IPPROTO_TCP;
 809         if (rt6_need_strict(&fl6.daddr) && !oif)
 810                 fl6.flowi6_oif = tcp_v6_iif(skb);
 811         else
 812                 fl6.flowi6_oif = oif;
 813         fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark);
 814         fl6.fl6_dport = t1->dest;
 815         fl6.fl6_sport = t1->source;
 816         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 817
 818         /* Pass a socket to ip6_dst_lookup either it is for RST
 819          * Underlying function will use this to retrieve the network
 820          * namespace
 821          */
 822         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 823         if (!IS_ERR(dst)) {
 824                 skb_dst_set(buff, dst);
 825                 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
 826                 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 827                 if (rst)
 828                         TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
 829                 return;
 830         }
 831
 832         kfree_skb(buff);
 833 }
 834
 835 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
 836 {
 837         const struct tcphdr *th = tcp_hdr(skb);
 838         u32 seq = 0, ack_seq = 0;
 839         struct tcp_md5sig_key *key = NULL;
 840 #ifdef CONFIG_TCP_MD5SIG
 841         const __u8 *hash_location = NULL;
 842         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 843         unsigned char newhash[16];
 844         int genhash;
 845         struct sock *sk1 = NULL;
 846 #endif
 847         int oif;
 848
 849         if (th->rst)
 850                 return;
 851
 852         /* If sk not NULL, it means we did a successful lookup and incoming
 853          * route had to be correct. prequeue might have dropped our dst.
 854          */
 855         if (!sk && !ipv6_unicast_destination(skb))
 856                 return;
 857
 858 #ifdef CONFIG_TCP_MD5SIG
 859         hash_location = tcp_parse_md5sig_option(th);
 860         if (!sk && hash_location) {
 861                 /*
 862                  * active side is lost. Try to find listening socket through
 863                  * source port, and then find md5 key through listening socket.
 864                  * we are not loose security here:
 865                  * Incoming packet is checked with md5 hash with finding key,
 866                  * no RST generated if md5 hash doesn't match.
 867                  */
 868                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
 869                                            &tcp_hashinfo, &ipv6h->saddr,
 870                                            th->source, &ipv6h->daddr,
 871                                            ntohs(th->source), tcp_v6_iif(skb));
 872                 if (!sk1)
 873                         return;
 874
 875                 rcu_read_lock();
 876                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 877                 if (!key)
 878                         goto release_sk1;
 879
 880                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
 881                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
 882                         goto release_sk1;
 883         } else {
 884                 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
 885         }
 886 #endif
 887
 888         if (th->ack)
 889                 seq = ntohl(th->ack_seq);
 890         else
 891                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
 892                           (th->doff << 2);
 893
 894         oif = sk ? sk->sk_bound_dev_if : 0;
 895         tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
 896
 897 #ifdef CONFIG_TCP_MD5SIG
 898 release_sk1:
 899         if (sk1) {
 900                 rcu_read_unlock();
 901                 sock_put(sk1);
 902         }
 903 #endif
 904 }
 905
 906 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
 907                             u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
 908                             struct tcp_md5sig_key *key, u8 tclass,
 909                             u32 label)
 910 {
 911         tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
 912                              tclass, label);
 913 }
 914
 915 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 916 {
 917         struct inet_timewait_sock *tw = inet_twsk(sk);
 918         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 919
 920         tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 921                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 922                         tcp_time_stamp + tcptw->tw_ts_offset,
 923                         tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
 924                         tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel));
 925
 926         inet_twsk_put(tw);
 927 }
 928
 929 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 930                                   struct request_sock *req)
 931 {
 932         /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 933          * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 934          */
 935         /* RFC 7323 2.3
 936          * The window field (SEG.WND) of every outgoing segment, with the
 937          * exception of <SYN> segments, MUST be right-shifted by
 938          * Rcv.Wind.Shift bits:
 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,
 943                         req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
 944                         tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if,
 945                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr),
 946                         0, 0);
 947 }
 948
 949
 950 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
 951 {
 952 #ifdef CONFIG_SYN_COOKIES
 953         const struct tcphdr *th = tcp_hdr(skb);
 954
 955         if (!th->syn)
 956                 sk = cookie_v6_check(sk, skb);
 957 #endif
 958         return sk;
 959 }
 960
 961 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 962 {
 963         if (skb->protocol == htons(ETH_P_IP))
 964                 return tcp_v4_conn_request(sk, skb);
 965
 966         if (!ipv6_unicast_destination(skb))
 967                 goto drop;
 968
 969         return tcp_conn_request(&tcp6_request_sock_ops,
 970                                 &tcp_request_sock_ipv6_ops, sk, skb);
 971
 972 drop:
 973         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 974         return 0; /* don't send reset */
 975 }
 976
 977 static void tcp_v6_restore_cb(struct sk_buff *skb)
 978 {
 979         /* We need to move header back to the beginning if xfrm6_policy_check()
 980          * and tcp_v6_fill_cb() are going to be called again.
 981          * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
 982          */
 983         memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
 984                 sizeof(struct inet6_skb_parm));
 985 }
 986
 987 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
 988                                          struct request_sock *req,
 989                                          struct dst_entry *dst,
 990                                          struct request_sock *req_unhash,
 991                                          bool *own_req)
 992 {
 993         struct inet_request_sock *ireq;
 994         struct ipv6_pinfo *newnp;
 995         const struct ipv6_pinfo *np = inet6_sk(sk);
 996         struct ipv6_txoptions *opt;
 997         struct tcp6_sock *newtcp6sk;
 998         struct inet_sock *newinet;
 999         struct tcp_sock *newtp;
1000         struct sock *newsk;
1001 #ifdef CONFIG_TCP_MD5SIG
1002         struct tcp_md5sig_key *key;
1003 #endif
1004         struct flowi6 fl6;
1005
1006         if (skb->protocol == htons(ETH_P_IP)) {
1007                 /*
1008                  *      v6 mapped
1009                  */
1010
1011                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1012                                              req_unhash, own_req);
1013
1014                 if (!newsk)
1015                         return NULL;
1016
1017                 newtcp6sk = (struct tcp6_sock *)newsk;
1018                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1019
1020                 newinet = inet_sk(newsk);
1021                 newnp = inet6_sk(newsk);
1022                 newtp = tcp_sk(newsk);
1023
1024                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1025
1026                 newnp->saddr = newsk->sk_v6_rcv_saddr;
1027
1028                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1029                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1030 #ifdef CONFIG_TCP_MD5SIG
1031                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1032 #endif
1033
1034                 newnp->ipv6_ac_list = NULL;
1035                 newnp->ipv6_fl_list = NULL;
1036                 newnp->pktoptions  = NULL;
1037                 newnp->opt         = NULL;
1038                 newnp->mcast_oif   = tcp_v6_iif(skb);
1039                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1040                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1041                 if (np->repflow)
1042                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1043
1044                 /*
1045                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1046                  * here, tcp_create_openreq_child now does this for us, see the comment in
1047                  * that function for the gory details. -acme
1048                  */
1049
1050                 /* It is tricky place. Until this moment IPv4 tcp
1051                    worked with IPv6 icsk.icsk_af_ops.
1052                    Sync it now.
1053                  */
1054                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1055
1056                 return newsk;
1057         }
1058
1059         ireq = inet_rsk(req);
1060
1061         if (sk_acceptq_is_full(sk))
1062                 goto out_overflow;
1063
1064         if (!dst) {
1065                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1066                 if (!dst)
1067                         goto out;
1068         }
1069
1070         newsk = tcp_create_openreq_child(sk, req, skb);
1071         if (!newsk)
1072                 goto out_nonewsk;
1073
1074         /*
1075          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1076          * count here, tcp_create_openreq_child now does this for us, see the
1077          * comment in that function for the gory details. -acme
1078          */
1079
1080         newsk->sk_gso_type = SKB_GSO_TCPV6;
1081         ip6_dst_store(newsk, dst, NULL, NULL);
1082         inet6_sk_rx_dst_set(newsk, skb);
1083
1084         newtcp6sk = (struct tcp6_sock *)newsk;
1085         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1086
1087         newtp = tcp_sk(newsk);
1088         newinet = inet_sk(newsk);
1089         newnp = inet6_sk(newsk);
1090
1091         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1092
1093         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1094         newnp->saddr = ireq->ir_v6_loc_addr;
1095         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1096         newsk->sk_bound_dev_if = ireq->ir_iif;
1097
1098         /* Now IPv6 options...
1099
1100            First: no IPv4 options.
1101          */
1102         newinet->inet_opt = NULL;
1103         newnp->ipv6_ac_list = NULL;
1104         newnp->ipv6_fl_list = NULL;
1105
1106         /* Clone RX bits */
1107         newnp->rxopt.all = np->rxopt.all;
1108
1109         newnp->pktoptions = NULL;
1110         newnp->opt        = NULL;
1111         newnp->mcast_oif  = tcp_v6_iif(skb);
1112         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1113         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1114         if (np->repflow)
1115                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1116
1117         /* Clone native IPv6 options from listening socket (if any)
1118
1119            Yes, keeping reference count would be much more clever,
1120            but we make one more one thing there: reattach optmem
1121            to newsk.
1122          */
1123         opt = rcu_dereference(np->opt);
1124         if (opt) {
1125                 opt = ipv6_dup_options(newsk, opt);
1126                 RCU_INIT_POINTER(newnp->opt, opt);
1127         }
1128         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1129         if (opt)
1130                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1131                                                     opt->opt_flen;
1132
1133         tcp_ca_openreq_child(newsk, dst);
1134
1135         tcp_sync_mss(newsk, dst_mtu(dst));
1136         newtp->advmss = dst_metric_advmss(dst);
1137         if (tcp_sk(sk)->rx_opt.user_mss &&
1138             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1139                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1140
1141         tcp_initialize_rcv_mss(newsk);
1142
1143         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1144         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1145
1146 #ifdef CONFIG_TCP_MD5SIG
1147         /* Copy over the MD5 key from the original socket */
1148         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1149         if (key) {
1150                 /* We're using one, so create a matching key
1151                  * on the newsk structure. If we fail to get
1152                  * memory, then we end up not copying the key
1153                  * across. Shucks.
1154                  */
1155                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1156                                AF_INET6, key->key, key->keylen,
1157                                sk_gfp_atomic(sk, GFP_ATOMIC));
1158         }
1159 #endif
1160
1161         if (__inet_inherit_port(sk, newsk) < 0) {
1162                 inet_csk_prepare_forced_close(newsk);
1163                 tcp_done(newsk);
1164                 goto out;
1165         }
1166         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1167         if (*own_req) {
1168                 tcp_move_syn(newtp, req);
1169
1170                 /* Clone pktoptions received with SYN, if we own the req */
1171                 if (ireq->pktopts) {
1172                         newnp->pktoptions = skb_clone(ireq->pktopts,
1173                                                       sk_gfp_atomic(sk, GFP_ATOMIC));
1174                         consume_skb(ireq->pktopts);
1175                         ireq->pktopts = NULL;
1176                         if (newnp->pktoptions) {
1177                                 tcp_v6_restore_cb(newnp->pktoptions);
1178                                 skb_set_owner_r(newnp->pktoptions, newsk);
1179                         }
1180                 }
1181         }
1182
1183         return newsk;
1184
1185 out_overflow:
1186         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1187 out_nonewsk:
1188         dst_release(dst);
1189 out:
1190         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1191         return NULL;
1192 }
1193
1194 /* The socket must have it's spinlock held when we get
1195  * here, unless it is a TCP_LISTEN socket.
1196  *
1197  * We have a potential double-lock case here, so even when
1198  * doing backlog processing we use the BH locking scheme.
1199  * This is because we cannot sleep with the original spinlock
1200  * held.
1201  */
1202 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1203 {
1204         struct ipv6_pinfo *np = inet6_sk(sk);
1205         struct tcp_sock *tp;
1206         struct sk_buff *opt_skb = NULL;
1207
1208         /* Imagine: socket is IPv6. IPv4 packet arrives,
1209            goes to IPv4 receive handler and backlogged.
1210            From backlog it always goes here. Kerboom...
1211            Fortunately, tcp_rcv_established and rcv_established
1212            handle them correctly, but it is not case with
1213            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1214          */
1215
1216         if (skb->protocol == htons(ETH_P_IP))
1217                 return tcp_v4_do_rcv(sk, skb);
1218
1219         if (tcp_filter(sk, skb))
1220                 goto discard;
1221
1222         /*
1223          *      socket locking is here for SMP purposes as backlog rcv
1224          *      is currently called with bh processing disabled.
1225          */
1226
1227         /* Do Stevens' IPV6_PKTOPTIONS.
1228
1229            Yes, guys, it is the only place in our code, where we
1230            may make it not affecting IPv4.
1231            The rest of code is protocol independent,
1232            and I do not like idea to uglify IPv4.
1233
1234            Actually, all the idea behind IPV6_PKTOPTIONS
1235            looks not very well thought. For now we latch
1236            options, received in the last packet, enqueued
1237            by tcp. Feel free to propose better solution.
1238                                                --ANK (980728)
1239          */
1240         if (np->rxopt.all)
1241                 opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC));
1242
1243         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1244                 struct dst_entry *dst = sk->sk_rx_dst;
1245
1246                 sock_rps_save_rxhash(sk, skb);
1247                 sk_mark_napi_id(sk, skb);
1248                 if (dst) {
1249                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1250                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1251                                 dst_release(dst);
1252                                 sk->sk_rx_dst = NULL;
1253                         }
1254                 }
1255
1256                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1257                 if (opt_skb)
1258                         goto ipv6_pktoptions;
1259                 return 0;
1260         }
1261
1262         if (tcp_checksum_complete(skb))
1263                 goto csum_err;
1264
1265         if (sk->sk_state == TCP_LISTEN) {
1266                 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1267
1268                 if (!nsk)
1269                         goto discard;
1270
1271                 if (nsk != sk) {
1272                         sock_rps_save_rxhash(nsk, skb);
1273                         sk_mark_napi_id(nsk, skb);
1274                         if (tcp_child_process(sk, nsk, skb))
1275                                 goto reset;
1276                         if (opt_skb)
1277                                 __kfree_skb(opt_skb);
1278                         return 0;
1279                 }
1280         } else
1281                 sock_rps_save_rxhash(sk, skb);
1282
1283         if (tcp_rcv_state_process(sk, skb))
1284                 goto reset;
1285         if (opt_skb)
1286                 goto ipv6_pktoptions;
1287         return 0;
1288
1289 reset:
1290         tcp_v6_send_reset(sk, skb);
1291 discard:
1292         if (opt_skb)
1293                 __kfree_skb(opt_skb);
1294         kfree_skb(skb);
1295         return 0;
1296 csum_err:
1297         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1298         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1299         goto discard;
1300
1301
1302 ipv6_pktoptions:
1303         /* Do you ask, what is it?
1304
1305            1. skb was enqueued by tcp.
1306            2. skb is added to tail of read queue, rather than out of order.
1307            3. socket is not in passive state.
1308            4. Finally, it really contains options, which user wants to receive.
1309          */
1310         tp = tcp_sk(sk);
1311         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1312             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1313                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1314                         np->mcast_oif = tcp_v6_iif(opt_skb);
1315                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1316                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1317                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1318                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1319                 if (np->repflow)
1320                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1321                 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1322                         skb_set_owner_r(opt_skb, sk);
1323                         tcp_v6_restore_cb(opt_skb);
1324                         opt_skb = xchg(&np->pktoptions, opt_skb);
1325                 } else {
1326                         __kfree_skb(opt_skb);
1327                         opt_skb = xchg(&np->pktoptions, NULL);
1328                 }
1329         }
1330
1331         kfree_skb(opt_skb);
1332         return 0;
1333 }
1334
1335 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1336                            const struct tcphdr *th)
1337 {
1338         /* This is tricky: we move IP6CB at its correct location into
1339          * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1340          * _decode_session6() uses IP6CB().
1341          * barrier() makes sure compiler won't play aliasing games.
1342          */
1343         memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1344                 sizeof(struct inet6_skb_parm));
1345         barrier();
1346
1347         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1348         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1349                                     skb->len - th->doff*4);
1350         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1351         TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1352         TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1353         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1354         TCP_SKB_CB(skb)->sacked = 0;
1355 }
1356
1357 static int tcp_v6_rcv(struct sk_buff *skb)
1358 {
1359         const struct tcphdr *th;
1360         const struct ipv6hdr *hdr;
1361         struct sock *sk;
1362         int ret;
1363         struct net *net = dev_net(skb->dev);
1364
1365         if (skb->pkt_type != PACKET_HOST)
1366                 goto discard_it;
1367
1368         /*
1369          *      Count it even if it's bad.
1370          */
1371         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1372
1373         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1374                 goto discard_it;
1375
1376         th = tcp_hdr(skb);
1377
1378         if (th->doff < sizeof(struct tcphdr)/4)
1379                 goto bad_packet;
1380         if (!pskb_may_pull(skb, th->doff*4))
1381                 goto discard_it;
1382
1383         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1384                 goto csum_error;
1385
1386         th = tcp_hdr(skb);
1387         hdr = ipv6_hdr(skb);
1388
1389 lookup:
1390         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest,
1391                                 inet6_iif(skb));
1392         if (!sk)
1393                 goto no_tcp_socket;
1394
1395 process:
1396         if (sk->sk_state == TCP_TIME_WAIT)
1397                 goto do_time_wait;
1398
1399         if (sk->sk_state == TCP_NEW_SYN_RECV) {
1400                 struct request_sock *req = inet_reqsk(sk);
1401                 struct sock *nsk;
1402
1403                 sk = req->rsk_listener;
1404                 tcp_v6_fill_cb(skb, hdr, th);
1405                 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1406                         reqsk_put(req);
1407                         goto discard_it;
1408                 }
1409                 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1410                         inet_csk_reqsk_queue_drop_and_put(sk, req);
1411                         goto lookup;
1412                 }
1413                 sock_hold(sk);
1414                 nsk = tcp_check_req(sk, skb, req, false);
1415                 if (!nsk) {
1416                         reqsk_put(req);
1417                         goto discard_and_relse;
1418                 }
1419                 if (nsk == sk) {
1420                         reqsk_put(req);
1421                         tcp_v6_restore_cb(skb);
1422                 } else if (tcp_child_process(sk, nsk, skb)) {
1423                         tcp_v6_send_reset(nsk, skb);
1424                         goto discard_and_relse;
1425                 } else {
1426                         sock_put(sk);
1427                         return 0;
1428                 }
1429         }
1430         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1431                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1432                 goto discard_and_relse;
1433         }
1434
1435         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1436                 goto discard_and_relse;
1437
1438         tcp_v6_fill_cb(skb, hdr, th);
1439
1440         if (tcp_v6_inbound_md5_hash(sk, skb))
1441                 goto discard_and_relse;
1442
1443         if (tcp_filter(sk, skb))
1444                 goto discard_and_relse;
1445         th = (const struct tcphdr *)skb->data;
1446         hdr = ipv6_hdr(skb);
1447
1448         skb->dev = NULL;
1449
1450         if (sk->sk_state == TCP_LISTEN) {
1451                 ret = tcp_v6_do_rcv(sk, skb);
1452                 goto put_and_return;
1453         }
1454
1455         sk_incoming_cpu_update(sk);
1456
1457         bh_lock_sock_nested(sk);
1458         tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1459         ret = 0;
1460         if (!sock_owned_by_user(sk)) {
1461                 if (!tcp_prequeue(sk, skb))
1462                         ret = tcp_v6_do_rcv(sk, skb);
1463         } else if (unlikely(sk_add_backlog(sk, skb,
1464                                            sk->sk_rcvbuf + sk->sk_sndbuf))) {
1465                 bh_unlock_sock(sk);
1466                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1467                 goto discard_and_relse;
1468         }
1469         bh_unlock_sock(sk);
1470
1471 put_and_return:
1472         sock_put(sk);
1473         return ret ? -1 : 0;
1474
1475 no_tcp_socket:
1476         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1477                 goto discard_it;
1478
1479         tcp_v6_fill_cb(skb, hdr, th);
1480
1481         if (tcp_checksum_complete(skb)) {
1482 csum_error:
1483                 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1484 bad_packet:
1485                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1486         } else {
1487                 tcp_v6_send_reset(NULL, skb);
1488         }
1489
1490 discard_it:
1491         kfree_skb(skb);
1492         return 0;
1493
1494 discard_and_relse:
1495         sock_put(sk);
1496         goto discard_it;
1497
1498 do_time_wait:
1499         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1500                 inet_twsk_put(inet_twsk(sk));
1501                 goto discard_it;
1502         }
1503
1504         tcp_v6_fill_cb(skb, hdr, th);
1505
1506         if (tcp_checksum_complete(skb)) {
1507                 inet_twsk_put(inet_twsk(sk));
1508                 goto csum_error;
1509         }
1510
1511         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1512         case TCP_TW_SYN:
1513         {
1514                 struct sock *sk2;
1515
1516                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
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                         goto process;
1526                 }
1527                 /* Fall through to ACK */
1528         }
1529         case TCP_TW_ACK:
1530                 tcp_v6_timewait_ack(sk, skb);
1531                 break;
1532         case TCP_TW_RST:
1533                 tcp_v6_restore_cb(skb);
1534                 goto no_tcp_socket;
1535         case TCP_TW_SUCCESS:
1536                 ;
1537         }
1538         goto discard_it;
1539 }
1540
1541 static void tcp_v6_early_demux(struct sk_buff *skb)
1542 {
1543         const struct ipv6hdr *hdr;
1544         const struct tcphdr *th;
1545         struct sock *sk;
1546
1547         if (skb->pkt_type != PACKET_HOST)
1548                 return;
1549
1550         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1551                 return;
1552
1553         hdr = ipv6_hdr(skb);
1554         th = tcp_hdr(skb);
1555
1556         if (th->doff < sizeof(struct tcphdr) / 4)
1557                 return;
1558
1559         /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1560         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1561                                         &hdr->saddr, th->source,
1562                                         &hdr->daddr, ntohs(th->dest),
1563                                         inet6_iif(skb));
1564         if (sk) {
1565                 skb->sk = sk;
1566                 skb->destructor = sock_edemux;
1567                 if (sk_fullsock(sk)) {
1568                         struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1569
1570                         if (dst)
1571                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1572                         if (dst &&
1573                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1574                                 skb_dst_set_noref(skb, dst);
1575                 }
1576         }
1577 }
1578
1579 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1580         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1581         .twsk_unique    = tcp_twsk_unique,
1582         .twsk_destructor = tcp_twsk_destructor,
1583 };
1584
1585 static const struct inet_connection_sock_af_ops ipv6_specific = {
1586         .queue_xmit        = inet6_csk_xmit,
1587         .send_check        = tcp_v6_send_check,
1588         .rebuild_header    = inet6_sk_rebuild_header,
1589         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1590         .conn_request      = tcp_v6_conn_request,
1591         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1592         .net_header_len    = sizeof(struct ipv6hdr),
1593         .net_frag_header_len = sizeof(struct frag_hdr),
1594         .setsockopt        = ipv6_setsockopt,
1595         .getsockopt        = ipv6_getsockopt,
1596         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1597         .sockaddr_len      = sizeof(struct sockaddr_in6),
1598         .bind_conflict     = inet6_csk_bind_conflict,
1599 #ifdef CONFIG_COMPAT
1600         .compat_setsockopt = compat_ipv6_setsockopt,
1601         .compat_getsockopt = compat_ipv6_getsockopt,
1602 #endif
1603         .mtu_reduced       = tcp_v6_mtu_reduced,
1604 };
1605
1606 #ifdef CONFIG_TCP_MD5SIG
1607 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1608         .md5_lookup     =       tcp_v6_md5_lookup,
1609         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1610         .md5_parse      =       tcp_v6_parse_md5_keys,
1611 };
1612 #endif
1613
1614 /*
1615  *      TCP over IPv4 via INET6 API
1616  */
1617 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1618         .queue_xmit        = ip_queue_xmit,
1619         .send_check        = tcp_v4_send_check,
1620         .rebuild_header    = inet_sk_rebuild_header,
1621         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1622         .conn_request      = tcp_v6_conn_request,
1623         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1624         .net_header_len    = sizeof(struct iphdr),
1625         .setsockopt        = ipv6_setsockopt,
1626         .getsockopt        = ipv6_getsockopt,
1627         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1628         .sockaddr_len      = sizeof(struct sockaddr_in6),
1629         .bind_conflict     = inet6_csk_bind_conflict,
1630 #ifdef CONFIG_COMPAT
1631         .compat_setsockopt = compat_ipv6_setsockopt,
1632         .compat_getsockopt = compat_ipv6_getsockopt,
1633 #endif
1634         .mtu_reduced       = tcp_v4_mtu_reduced,
1635 };
1636
1637 #ifdef CONFIG_TCP_MD5SIG
1638 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1639         .md5_lookup     =       tcp_v4_md5_lookup,
1640         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1641         .md5_parse      =       tcp_v6_parse_md5_keys,
1642 };
1643 #endif
1644
1645 /* NOTE: A lot of things set to zero explicitly by call to
1646  *       sk_alloc() so need not be done here.
1647  */
1648 static int tcp_v6_init_sock(struct sock *sk)
1649 {
1650         struct inet_connection_sock *icsk = inet_csk(sk);
1651
1652         tcp_init_sock(sk);
1653
1654         icsk->icsk_af_ops = &ipv6_specific;
1655
1656 #ifdef CONFIG_TCP_MD5SIG
1657         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1658 #endif
1659
1660         return 0;
1661 }
1662
1663 static void tcp_v6_destroy_sock(struct sock *sk)
1664 {
1665         tcp_v4_destroy_sock(sk);
1666         inet6_destroy_sock(sk);
1667 }
1668
1669 #ifdef CONFIG_PROC_FS
1670 /* Proc filesystem TCPv6 sock list dumping. */
1671 static void get_openreq6(struct seq_file *seq,
1672                          const struct request_sock *req, int i)
1673 {
1674         long ttd = req->rsk_timer.expires - jiffies;
1675         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1676         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1677
1678         if (ttd < 0)
1679                 ttd = 0;
1680
1681         seq_printf(seq,
1682                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1683                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1684                    i,
1685                    src->s6_addr32[0], src->s6_addr32[1],
1686                    src->s6_addr32[2], src->s6_addr32[3],
1687                    inet_rsk(req)->ir_num,
1688                    dest->s6_addr32[0], dest->s6_addr32[1],
1689                    dest->s6_addr32[2], dest->s6_addr32[3],
1690                    ntohs(inet_rsk(req)->ir_rmt_port),
1691                    TCP_SYN_RECV,
1692                    0, 0, /* could print option size, but that is af dependent. */
1693                    1,   /* timers active (only the expire timer) */
1694                    jiffies_to_clock_t(ttd),
1695                    req->num_timeout,
1696                    from_kuid_munged(seq_user_ns(seq),
1697                                     sock_i_uid(req->rsk_listener)),
1698                    0,  /* non standard timer */
1699                    0, /* open_requests have no inode */
1700                    0, req);
1701 }
1702
1703 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1704 {
1705         const struct in6_addr *dest, *src;
1706         __u16 destp, srcp;
1707         int timer_active;
1708         unsigned long timer_expires;
1709         const struct inet_sock *inet = inet_sk(sp);
1710         const struct tcp_sock *tp = tcp_sk(sp);
1711         const struct inet_connection_sock *icsk = inet_csk(sp);
1712         const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1713         int rx_queue;
1714         int state;
1715
1716         dest  = &sp->sk_v6_daddr;
1717         src   = &sp->sk_v6_rcv_saddr;
1718         destp = ntohs(inet->inet_dport);
1719         srcp  = ntohs(inet->inet_sport);
1720
1721         if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1722             icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
1723             icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1724                 timer_active    = 1;
1725                 timer_expires   = icsk->icsk_timeout;
1726         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1727                 timer_active    = 4;
1728                 timer_expires   = icsk->icsk_timeout;
1729         } else if (timer_pending(&sp->sk_timer)) {
1730                 timer_active    = 2;
1731                 timer_expires   = sp->sk_timer.expires;
1732         } else {
1733                 timer_active    = 0;
1734                 timer_expires = jiffies;
1735         }
1736
1737         state = sk_state_load(sp);
1738         if (state == TCP_LISTEN)
1739                 rx_queue = sp->sk_ack_backlog;
1740         else
1741                 /* Because we don't lock the socket,
1742                  * we might find a transient negative value.
1743                  */
1744                 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1745
1746         seq_printf(seq,
1747                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1748                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1749                    i,
1750                    src->s6_addr32[0], src->s6_addr32[1],
1751                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1752                    dest->s6_addr32[0], dest->s6_addr32[1],
1753                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1754                    state,
1755                    tp->write_seq - tp->snd_una,
1756                    rx_queue,
1757                    timer_active,
1758                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1759                    icsk->icsk_retransmits,
1760                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1761                    icsk->icsk_probes_out,
1762                    sock_i_ino(sp),
1763                    atomic_read(&sp->sk_refcnt), sp,
1764                    jiffies_to_clock_t(icsk->icsk_rto),
1765                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1766                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1767                    tp->snd_cwnd,
1768                    state == TCP_LISTEN ?
1769                         fastopenq->max_qlen :
1770                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1771                    );
1772 }
1773
1774 static void get_timewait6_sock(struct seq_file *seq,
1775                                struct inet_timewait_sock *tw, int i)
1776 {
1777         long delta = tw->tw_timer.expires - jiffies;
1778         const struct in6_addr *dest, *src;
1779         __u16 destp, srcp;
1780
1781         dest = &tw->tw_v6_daddr;
1782         src  = &tw->tw_v6_rcv_saddr;
1783         destp = ntohs(tw->tw_dport);
1784         srcp  = ntohs(tw->tw_sport);
1785
1786         seq_printf(seq,
1787                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1788                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1789                    i,
1790                    src->s6_addr32[0], src->s6_addr32[1],
1791                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1792                    dest->s6_addr32[0], dest->s6_addr32[1],
1793                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1794                    tw->tw_substate, 0, 0,
1795                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1796                    atomic_read(&tw->tw_refcnt), tw);
1797 }
1798
1799 static int tcp6_seq_show(struct seq_file *seq, void *v)
1800 {
1801         struct tcp_iter_state *st;
1802         struct sock *sk = v;
1803
1804         if (v == SEQ_START_TOKEN) {
1805                 seq_puts(seq,
1806                          "  sl  "
1807                          "local_address                         "
1808                          "remote_address                        "
1809                          "st tx_queue rx_queue tr tm->when retrnsmt"
1810                          "   uid  timeout inode\n");
1811                 goto out;
1812         }
1813         st = seq->private;
1814
1815         if (sk->sk_state == TCP_TIME_WAIT)
1816                 get_timewait6_sock(seq, v, st->num);
1817         else if (sk->sk_state == TCP_NEW_SYN_RECV)
1818                 get_openreq6(seq, v, st->num);
1819         else
1820                 get_tcp6_sock(seq, v, st->num);
1821 out:
1822         return 0;
1823 }
1824
1825 static const struct file_operations tcp6_afinfo_seq_fops = {
1826         .owner   = THIS_MODULE,
1827         .open    = tcp_seq_open,
1828         .read    = seq_read,
1829         .llseek  = seq_lseek,
1830         .release = seq_release_net
1831 };
1832
1833 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1834         .name           = "tcp6",
1835         .family         = AF_INET6,
1836         .seq_fops       = &tcp6_afinfo_seq_fops,
1837         .seq_ops        = {
1838                 .show           = tcp6_seq_show,
1839         },
1840 };
1841
1842 int __net_init tcp6_proc_init(struct net *net)
1843 {
1844         return tcp_proc_register(net, &tcp6_seq_afinfo);
1845 }
1846
1847 void tcp6_proc_exit(struct net *net)
1848 {
1849         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1850 }
1851 #endif
1852
1853 static void tcp_v6_clear_sk(struct sock *sk, int size)
1854 {
1855         struct inet_sock *inet = inet_sk(sk);
1856
1857         /* we do not want to clear pinet6 field, because of RCU lookups */
1858         sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
1859
1860         size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
1861         memset(&inet->pinet6 + 1, 0, size);
1862 }
1863
1864 struct proto tcpv6_prot = {
1865         .name                   = "TCPv6",
1866         .owner                  = THIS_MODULE,
1867         .close                  = tcp_close,
1868         .connect                = tcp_v6_connect,
1869         .disconnect             = tcp_disconnect,
1870         .accept                 = inet_csk_accept,
1871         .ioctl                  = tcp_ioctl,
1872         .init                   = tcp_v6_init_sock,
1873         .destroy                = tcp_v6_destroy_sock,
1874         .shutdown               = tcp_shutdown,
1875         .setsockopt             = tcp_setsockopt,
1876         .getsockopt             = tcp_getsockopt,
1877         .recvmsg                = tcp_recvmsg,
1878         .sendmsg                = tcp_sendmsg,
1879         .sendpage               = tcp_sendpage,
1880         .backlog_rcv            = tcp_v6_do_rcv,
1881         .release_cb             = tcp_release_cb,
1882         .hash                   = inet_hash,
1883         .unhash                 = inet_unhash,
1884         .get_port               = inet_csk_get_port,
1885         .enter_memory_pressure  = tcp_enter_memory_pressure,
1886         .stream_memory_free     = tcp_stream_memory_free,
1887         .sockets_allocated      = &tcp_sockets_allocated,
1888         .memory_allocated       = &tcp_memory_allocated,
1889         .memory_pressure        = &tcp_memory_pressure,
1890         .orphan_count           = &tcp_orphan_count,
1891         .sysctl_mem             = sysctl_tcp_mem,
1892         .sysctl_wmem            = sysctl_tcp_wmem,
1893         .sysctl_rmem            = sysctl_tcp_rmem,
1894         .max_header             = MAX_TCP_HEADER,
1895         .obj_size               = sizeof(struct tcp6_sock),
1896         .slab_flags             = SLAB_DESTROY_BY_RCU,
1897         .twsk_prot              = &tcp6_timewait_sock_ops,
1898         .rsk_prot               = &tcp6_request_sock_ops,
1899         .h.hashinfo             = &tcp_hashinfo,
1900         .no_autobind            = true,
1901 #ifdef CONFIG_COMPAT
1902         .compat_setsockopt      = compat_tcp_setsockopt,
1903         .compat_getsockopt      = compat_tcp_getsockopt,
1904 #endif
1905 #ifdef CONFIG_MEMCG_KMEM
1906         .proto_cgroup           = tcp_proto_cgroup,
1907 #endif
1908         .clear_sk               = tcp_v6_clear_sk,
1909 };
1910
1911 static const struct inet6_protocol tcpv6_protocol = {
1912         .early_demux    =       tcp_v6_early_demux,
1913         .handler        =       tcp_v6_rcv,
1914         .err_handler    =       tcp_v6_err,
1915         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1916 };
1917
1918 static struct inet_protosw tcpv6_protosw = {
1919         .type           =       SOCK_STREAM,
1920         .protocol       =       IPPROTO_TCP,
1921         .prot           =       &tcpv6_prot,
1922         .ops            =       &inet6_stream_ops,
1923         .flags          =       INET_PROTOSW_PERMANENT |
1924                                 INET_PROTOSW_ICSK,
1925 };
1926
1927 static int __net_init tcpv6_net_init(struct net *net)
1928 {
1929         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1930                                     SOCK_RAW, IPPROTO_TCP, net);
1931 }
1932
1933 static void __net_exit tcpv6_net_exit(struct net *net)
1934 {
1935         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1936 }
1937
1938 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1939 {
1940         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
1941 }
1942
1943 static struct pernet_operations tcpv6_net_ops = {
1944         .init       = tcpv6_net_init,
1945         .exit       = tcpv6_net_exit,
1946         .exit_batch = tcpv6_net_exit_batch,
1947 };
1948
1949 int __init tcpv6_init(void)
1950 {
1951         int ret;
1952
1953         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1954         if (ret)
1955                 goto out;
1956
1957         /* register inet6 protocol */
1958         ret = inet6_register_protosw(&tcpv6_protosw);
1959         if (ret)
1960                 goto out_tcpv6_protocol;
1961
1962         ret = register_pernet_subsys(&tcpv6_net_ops);
1963         if (ret)
1964                 goto out_tcpv6_protosw;
1965 out:
1966         return ret;
1967
1968 out_tcpv6_protosw:
1969         inet6_unregister_protosw(&tcpv6_protosw);
1970 out_tcpv6_protocol:
1971         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1972         goto out;
1973 }
1974
1975 void tcpv6_exit(void)
1976 {
1977         unregister_pernet_subsys(&tcpv6_net_ops);
1978         inet6_unregister_protosw(&tcpv6_protosw);
1979         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1980 }