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_ac_list = NULL;
1066                 newnp->ipv6_fl_list = NULL;
1067                 newnp->pktoptions  = NULL;
1068                 newnp->opt         = NULL;
1069                 newnp->mcast_oif   = tcp_v6_iif(skb);
1070                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1071                 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1072                 if (np->repflow)
1073                         newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1074
1075                 /*
1076                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1077                  * here, tcp_create_openreq_child now does this for us, see the comment in
1078                  * that function for the gory details. -acme
1079                  */
1080
1081                 /* It is tricky place. Until this moment IPv4 tcp
1082                    worked with IPv6 icsk.icsk_af_ops.
1083                    Sync it now.
1084                  */
1085                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1086
1087                 return newsk;
1088         }
1089
1090         ireq = inet_rsk(req);
1091
1092         if (sk_acceptq_is_full(sk))
1093                 goto out_overflow;
1094
1095         if (!dst) {
1096                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1097                 if (!dst)
1098                         goto out;
1099         }
1100
1101         newsk = tcp_create_openreq_child(sk, req, skb);
1102         if (!newsk)
1103                 goto out_nonewsk;
1104
1105         /*
1106          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1107          * count here, tcp_create_openreq_child now does this for us, see the
1108          * comment in that function for the gory details. -acme
1109          */
1110
1111         newsk->sk_gso_type = SKB_GSO_TCPV6;
1112         ip6_dst_store(newsk, dst, NULL, NULL);
1113         inet6_sk_rx_dst_set(newsk, skb);
1114
1115         newtcp6sk = (struct tcp6_sock *)newsk;
1116         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1117
1118         newtp = tcp_sk(newsk);
1119         newinet = inet_sk(newsk);
1120         newnp = inet6_sk(newsk);
1121
1122         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1123
1124         newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1125         newnp->saddr = ireq->ir_v6_loc_addr;
1126         newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1127         newsk->sk_bound_dev_if = ireq->ir_iif;
1128
1129         /* Now IPv6 options...
1130
1131            First: no IPv4 options.
1132          */
1133         newinet->inet_opt = NULL;
1134         newnp->ipv6_ac_list = NULL;
1135         newnp->ipv6_fl_list = NULL;
1136
1137         /* Clone RX bits */
1138         newnp->rxopt.all = np->rxopt.all;
1139
1140         newnp->pktoptions = NULL;
1141         newnp->opt        = NULL;
1142         newnp->mcast_oif  = tcp_v6_iif(skb);
1143         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1144         newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1145         if (np->repflow)
1146                 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1147
1148         /* Clone native IPv6 options from listening socket (if any)
1149
1150            Yes, keeping reference count would be much more clever,
1151            but we make one more one thing there: reattach optmem
1152            to newsk.
1153          */
1154         opt = ireq->ipv6_opt;
1155         if (!opt)
1156                 opt = rcu_dereference(np->opt);
1157         if (opt) {
1158                 opt = ipv6_dup_options(newsk, opt);
1159                 RCU_INIT_POINTER(newnp->opt, opt);
1160         }
1161         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1162         if (opt)
1163                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1164                                                     opt->opt_flen;
1165
1166         tcp_ca_openreq_child(newsk, dst);
1167
1168         tcp_sync_mss(newsk, dst_mtu(dst));
1169         newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1170
1171         tcp_initialize_rcv_mss(newsk);
1172
1173         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1174         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1175
1176 #ifdef CONFIG_TCP_MD5SIG
1177         /* Copy over the MD5 key from the original socket */
1178         key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr);
1179         if (key) {
1180                 /* We're using one, so create a matching key
1181                  * on the newsk structure. If we fail to get
1182                  * memory, then we end up not copying the key
1183                  * across. Shucks.
1184                  */
1185                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1186                                AF_INET6, key->key, key->keylen,
1187                                sk_gfp_mask(sk, GFP_ATOMIC));
1188         }
1189 #endif
1190
1191         if (__inet_inherit_port(sk, newsk) < 0) {
1192                 inet_csk_prepare_forced_close(newsk);
1193                 tcp_done(newsk);
1194                 goto out;
1195         }
1196         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1197         if (*own_req) {
1198                 tcp_move_syn(newtp, req);
1199
1200                 /* Clone pktoptions received with SYN, if we own the req */
1201                 if (ireq->pktopts) {
1202                         newnp->pktoptions = skb_clone(ireq->pktopts,
1203                                                       sk_gfp_mask(sk, GFP_ATOMIC));
1204                         consume_skb(ireq->pktopts);
1205                         ireq->pktopts = NULL;
1206                         if (newnp->pktoptions) {
1207                                 tcp_v6_restore_cb(newnp->pktoptions);
1208                                 skb_set_owner_r(newnp->pktoptions, newsk);
1209                         }
1210                 }
1211         }
1212
1213         return newsk;
1214
1215 out_overflow:
1216         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1217 out_nonewsk:
1218         dst_release(dst);
1219 out:
1220         tcp_listendrop(sk);
1221         return NULL;
1222 }
1223
1224 /* The socket must have it's spinlock held when we get
1225  * here, unless it is a TCP_LISTEN socket.
1226  *
1227  * We have a potential double-lock case here, so even when
1228  * doing backlog processing we use the BH locking scheme.
1229  * This is because we cannot sleep with the original spinlock
1230  * held.
1231  */
1232 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1233 {
1234         struct ipv6_pinfo *np = inet6_sk(sk);
1235         struct tcp_sock *tp;
1236         struct sk_buff *opt_skb = NULL;
1237
1238         /* Imagine: socket is IPv6. IPv4 packet arrives,
1239            goes to IPv4 receive handler and backlogged.
1240            From backlog it always goes here. Kerboom...
1241            Fortunately, tcp_rcv_established and rcv_established
1242            handle them correctly, but it is not case with
1243            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1244          */
1245
1246         if (skb->protocol == htons(ETH_P_IP))
1247                 return tcp_v4_do_rcv(sk, skb);
1248
1249         if (tcp_filter(sk, skb))
1250                 goto discard;
1251
1252         /*
1253          *      socket locking is here for SMP purposes as backlog rcv
1254          *      is currently called with bh processing disabled.
1255          */
1256
1257         /* Do Stevens' IPV6_PKTOPTIONS.
1258
1259            Yes, guys, it is the only place in our code, where we
1260            may make it not affecting IPv4.
1261            The rest of code is protocol independent,
1262            and I do not like idea to uglify IPv4.
1263
1264            Actually, all the idea behind IPV6_PKTOPTIONS
1265            looks not very well thought. For now we latch
1266            options, received in the last packet, enqueued
1267            by tcp. Feel free to propose better solution.
1268                                                --ANK (980728)
1269          */
1270         if (np->rxopt.all)
1271                 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1272
1273         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1274                 struct dst_entry *dst = sk->sk_rx_dst;
1275
1276                 sock_rps_save_rxhash(sk, skb);
1277                 sk_mark_napi_id(sk, skb);
1278                 if (dst) {
1279                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1280                             dst->ops->check(dst, np->rx_dst_cookie) == NULL) {
1281                                 dst_release(dst);
1282                                 sk->sk_rx_dst = NULL;
1283                         }
1284                 }
1285
1286                 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1287                 if (opt_skb)
1288                         goto ipv6_pktoptions;
1289                 return 0;
1290         }
1291
1292         if (tcp_checksum_complete(skb))
1293                 goto csum_err;
1294
1295         if (sk->sk_state == TCP_LISTEN) {
1296                 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1297
1298                 if (!nsk)
1299                         goto discard;
1300
1301                 if (nsk != sk) {
1302                         if (tcp_child_process(sk, nsk, skb))
1303                                 goto reset;
1304                         if (opt_skb)
1305                                 __kfree_skb(opt_skb);
1306                         return 0;
1307                 }
1308         } else
1309                 sock_rps_save_rxhash(sk, skb);
1310
1311         if (tcp_rcv_state_process(sk, skb))
1312                 goto reset;
1313         if (opt_skb)
1314                 goto ipv6_pktoptions;
1315         return 0;
1316
1317 reset:
1318         tcp_v6_send_reset(sk, skb);
1319 discard:
1320         if (opt_skb)
1321                 __kfree_skb(opt_skb);
1322         kfree_skb(skb);
1323         return 0;
1324 csum_err:
1325         TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1326         TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1327         goto discard;
1328
1329
1330 ipv6_pktoptions:
1331         /* Do you ask, what is it?
1332
1333            1. skb was enqueued by tcp.
1334            2. skb is added to tail of read queue, rather than out of order.
1335            3. socket is not in passive state.
1336            4. Finally, it really contains options, which user wants to receive.
1337          */
1338         tp = tcp_sk(sk);
1339         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1340             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1341                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1342                         np->mcast_oif = tcp_v6_iif(opt_skb);
1343                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1344                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1345                 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1346                         np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1347                 if (np->repflow)
1348                         np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1349                 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1350                         skb_set_owner_r(opt_skb, sk);
1351                         tcp_v6_restore_cb(opt_skb);
1352                         opt_skb = xchg(&np->pktoptions, opt_skb);
1353                 } else {
1354                         __kfree_skb(opt_skb);
1355                         opt_skb = xchg(&np->pktoptions, NULL);
1356                 }
1357         }
1358
1359         kfree_skb(opt_skb);
1360         return 0;
1361 }
1362
1363 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1364                            const struct tcphdr *th)
1365 {
1366         /* This is tricky: we move IP6CB at its correct location into
1367          * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1368          * _decode_session6() uses IP6CB().
1369          * barrier() makes sure compiler won't play aliasing games.
1370          */
1371         memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1372                 sizeof(struct inet6_skb_parm));
1373         barrier();
1374
1375         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1376         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1377                                     skb->len - th->doff*4);
1378         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1379         TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1380         TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1381         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1382         TCP_SKB_CB(skb)->sacked = 0;
1383 }
1384
1385 static int tcp_v6_rcv(struct sk_buff *skb)
1386 {
1387         const struct tcphdr *th;
1388         const struct ipv6hdr *hdr;
1389         bool refcounted;
1390         struct sock *sk;
1391         int ret;
1392         struct net *net = dev_net(skb->dev);
1393
1394         if (skb->pkt_type != PACKET_HOST)
1395                 goto discard_it;
1396
1397         /*
1398          *      Count it even if it's bad.
1399          */
1400         __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1401
1402         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1403                 goto discard_it;
1404
1405         th = (const struct tcphdr *)skb->data;
1406
1407         if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1408                 goto bad_packet;
1409         if (!pskb_may_pull(skb, th->doff*4))
1410                 goto discard_it;
1411
1412         if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1413                 goto csum_error;
1414
1415         th = (const struct tcphdr *)skb->data;
1416         hdr = ipv6_hdr(skb);
1417
1418 lookup:
1419         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1420                                 th->source, th->dest, inet6_iif(skb),
1421                                 &refcounted);
1422         if (!sk)
1423                 goto no_tcp_socket;
1424
1425 process:
1426         if (sk->sk_state == TCP_TIME_WAIT)
1427                 goto do_time_wait;
1428
1429         if (sk->sk_state == TCP_NEW_SYN_RECV) {
1430                 struct request_sock *req = inet_reqsk(sk);
1431                 struct sock *nsk;
1432
1433                 sk = req->rsk_listener;
1434                 tcp_v6_fill_cb(skb, hdr, th);
1435                 if (tcp_v6_inbound_md5_hash(sk, skb)) {
1436                         sk_drops_add(sk, skb);
1437                         reqsk_put(req);
1438                         goto discard_it;
1439                 }
1440                 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1441                         inet_csk_reqsk_queue_drop_and_put(sk, req);
1442                         goto lookup;
1443                 }
1444                 sock_hold(sk);
1445                 refcounted = true;
1446                 nsk = tcp_check_req(sk, skb, req, false);
1447                 if (!nsk) {
1448                         reqsk_put(req);
1449                         goto discard_and_relse;
1450                 }
1451                 if (nsk == sk) {
1452                         reqsk_put(req);
1453                         tcp_v6_restore_cb(skb);
1454                 } else if (tcp_child_process(sk, nsk, skb)) {
1455                         tcp_v6_send_reset(nsk, skb);
1456                         goto discard_and_relse;
1457                 } else {
1458                         sock_put(sk);
1459                         return 0;
1460                 }
1461         }
1462         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1463                 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1464                 goto discard_and_relse;
1465         }
1466
1467         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1468                 goto discard_and_relse;
1469
1470         tcp_v6_fill_cb(skb, hdr, th);
1471
1472         if (tcp_v6_inbound_md5_hash(sk, skb))
1473                 goto discard_and_relse;
1474
1475         if (tcp_filter(sk, skb))
1476                 goto discard_and_relse;
1477         th = (const struct tcphdr *)skb->data;
1478         hdr = ipv6_hdr(skb);
1479
1480         skb->dev = NULL;
1481
1482         if (sk->sk_state == TCP_LISTEN) {
1483                 ret = tcp_v6_do_rcv(sk, skb);
1484                 goto put_and_return;
1485         }
1486
1487         sk_incoming_cpu_update(sk);
1488
1489         bh_lock_sock_nested(sk);
1490         tcp_segs_in(tcp_sk(sk), skb);
1491         ret = 0;
1492         if (!sock_owned_by_user(sk)) {
1493                 if (!tcp_prequeue(sk, skb))
1494                         ret = tcp_v6_do_rcv(sk, skb);
1495         } else if (tcp_add_backlog(sk, skb)) {
1496                 goto discard_and_relse;
1497         }
1498         bh_unlock_sock(sk);
1499
1500 put_and_return:
1501         if (refcounted)
1502                 sock_put(sk);
1503         return ret ? -1 : 0;
1504
1505 no_tcp_socket:
1506         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1507                 goto discard_it;
1508
1509         tcp_v6_fill_cb(skb, hdr, th);
1510
1511         if (tcp_checksum_complete(skb)) {
1512 csum_error:
1513                 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1514 bad_packet:
1515                 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1516         } else {
1517                 tcp_v6_send_reset(NULL, skb);
1518         }
1519
1520 discard_it:
1521         kfree_skb(skb);
1522         return 0;
1523
1524 discard_and_relse:
1525         sk_drops_add(sk, skb);
1526         if (refcounted)
1527                 sock_put(sk);
1528         goto discard_it;
1529
1530 do_time_wait:
1531         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1532                 inet_twsk_put(inet_twsk(sk));
1533                 goto discard_it;
1534         }
1535
1536         tcp_v6_fill_cb(skb, hdr, th);
1537
1538         if (tcp_checksum_complete(skb)) {
1539                 inet_twsk_put(inet_twsk(sk));
1540                 goto csum_error;
1541         }
1542
1543         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1544         case TCP_TW_SYN:
1545         {
1546                 struct sock *sk2;
1547
1548                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1549                                             skb, __tcp_hdrlen(th),
1550                                             &ipv6_hdr(skb)->saddr, th->source,
1551                                             &ipv6_hdr(skb)->daddr,
1552                                             ntohs(th->dest), tcp_v6_iif(skb));
1553                 if (sk2) {
1554                         struct inet_timewait_sock *tw = inet_twsk(sk);
1555                         inet_twsk_deschedule_put(tw);
1556                         sk = sk2;
1557                         tcp_v6_restore_cb(skb);
1558                         refcounted = false;
1559                         goto process;
1560                 }
1561                 /* Fall through to ACK */
1562         }
1563         case TCP_TW_ACK:
1564                 tcp_v6_timewait_ack(sk, skb);
1565                 break;
1566         case TCP_TW_RST:
1567                 tcp_v6_restore_cb(skb);
1568                 tcp_v6_send_reset(sk, skb);
1569                 inet_twsk_deschedule_put(inet_twsk(sk));
1570                 goto discard_it;
1571         case TCP_TW_SUCCESS:
1572                 ;
1573         }
1574         goto discard_it;
1575 }
1576
1577 static void tcp_v6_early_demux(struct sk_buff *skb)
1578 {
1579         const struct ipv6hdr *hdr;
1580         const struct tcphdr *th;
1581         struct sock *sk;
1582
1583         if (skb->pkt_type != PACKET_HOST)
1584                 return;
1585
1586         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1587                 return;
1588
1589         hdr = ipv6_hdr(skb);
1590         th = tcp_hdr(skb);
1591
1592         if (th->doff < sizeof(struct tcphdr) / 4)
1593                 return;
1594
1595         /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1596         sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1597                                         &hdr->saddr, th->source,
1598                                         &hdr->daddr, ntohs(th->dest),
1599                                         inet6_iif(skb));
1600         if (sk) {
1601                 skb->sk = sk;
1602                 skb->destructor = sock_edemux;
1603                 if (sk_fullsock(sk)) {
1604                         struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1605
1606                         if (dst)
1607                                 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1608                         if (dst &&
1609                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1610                                 skb_dst_set_noref(skb, dst);
1611                 }
1612         }
1613 }
1614
1615 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1616         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1617         .twsk_unique    = tcp_twsk_unique,
1618         .twsk_destructor = tcp_twsk_destructor,
1619 };
1620
1621 static const struct inet_connection_sock_af_ops ipv6_specific = {
1622         .queue_xmit        = inet6_csk_xmit,
1623         .send_check        = tcp_v6_send_check,
1624         .rebuild_header    = inet6_sk_rebuild_header,
1625         .sk_rx_dst_set     = inet6_sk_rx_dst_set,
1626         .conn_request      = tcp_v6_conn_request,
1627         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1628         .net_header_len    = sizeof(struct ipv6hdr),
1629         .net_frag_header_len = sizeof(struct frag_hdr),
1630         .setsockopt        = ipv6_setsockopt,
1631         .getsockopt        = ipv6_getsockopt,
1632         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1633         .sockaddr_len      = sizeof(struct sockaddr_in6),
1634 #ifdef CONFIG_COMPAT
1635         .compat_setsockopt = compat_ipv6_setsockopt,
1636         .compat_getsockopt = compat_ipv6_getsockopt,
1637 #endif
1638         .mtu_reduced       = tcp_v6_mtu_reduced,
1639 };
1640
1641 #ifdef CONFIG_TCP_MD5SIG
1642 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1643         .md5_lookup     =       tcp_v6_md5_lookup,
1644         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1645         .md5_parse      =       tcp_v6_parse_md5_keys,
1646 };
1647 #endif
1648
1649 /*
1650  *      TCP over IPv4 via INET6 API
1651  */
1652 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1653         .queue_xmit        = ip_queue_xmit,
1654         .send_check        = tcp_v4_send_check,
1655         .rebuild_header    = inet_sk_rebuild_header,
1656         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1657         .conn_request      = tcp_v6_conn_request,
1658         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1659         .net_header_len    = sizeof(struct iphdr),
1660         .setsockopt        = ipv6_setsockopt,
1661         .getsockopt        = ipv6_getsockopt,
1662         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1663         .sockaddr_len      = sizeof(struct sockaddr_in6),
1664 #ifdef CONFIG_COMPAT
1665         .compat_setsockopt = compat_ipv6_setsockopt,
1666         .compat_getsockopt = compat_ipv6_getsockopt,
1667 #endif
1668         .mtu_reduced       = tcp_v4_mtu_reduced,
1669 };
1670
1671 #ifdef CONFIG_TCP_MD5SIG
1672 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1673         .md5_lookup     =       tcp_v4_md5_lookup,
1674         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1675         .md5_parse      =       tcp_v6_parse_md5_keys,
1676 };
1677 #endif
1678
1679 /* NOTE: A lot of things set to zero explicitly by call to
1680  *       sk_alloc() so need not be done here.
1681  */
1682 static int tcp_v6_init_sock(struct sock *sk)
1683 {
1684         struct inet_connection_sock *icsk = inet_csk(sk);
1685
1686         tcp_init_sock(sk);
1687
1688         icsk->icsk_af_ops = &ipv6_specific;
1689
1690 #ifdef CONFIG_TCP_MD5SIG
1691         tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1692 #endif
1693
1694         return 0;
1695 }
1696
1697 static void tcp_v6_destroy_sock(struct sock *sk)
1698 {
1699         tcp_v4_destroy_sock(sk);
1700         inet6_destroy_sock(sk);
1701 }
1702
1703 #ifdef CONFIG_PROC_FS
1704 /* Proc filesystem TCPv6 sock list dumping. */
1705 static void get_openreq6(struct seq_file *seq,
1706                          const struct request_sock *req, int i)
1707 {
1708         long ttd = req->rsk_timer.expires - jiffies;
1709         const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1710         const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1711
1712         if (ttd < 0)
1713                 ttd = 0;
1714
1715         seq_printf(seq,
1716                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1717                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1718                    i,
1719                    src->s6_addr32[0], src->s6_addr32[1],
1720                    src->s6_addr32[2], src->s6_addr32[3],
1721                    inet_rsk(req)->ir_num,
1722                    dest->s6_addr32[0], dest->s6_addr32[1],
1723                    dest->s6_addr32[2], dest->s6_addr32[3],
1724                    ntohs(inet_rsk(req)->ir_rmt_port),
1725                    TCP_SYN_RECV,
1726                    0, 0, /* could print option size, but that is af dependent. */
1727                    1,   /* timers active (only the expire timer) */
1728                    jiffies_to_clock_t(ttd),
1729                    req->num_timeout,
1730                    from_kuid_munged(seq_user_ns(seq),
1731                                     sock_i_uid(req->rsk_listener)),
1732                    0,  /* non standard timer */
1733                    0, /* open_requests have no inode */
1734                    0, req);
1735 }
1736
1737 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1738 {
1739         const struct in6_addr *dest, *src;
1740         __u16 destp, srcp;
1741         int timer_active;
1742         unsigned long timer_expires;
1743         const struct inet_sock *inet = inet_sk(sp);
1744         const struct tcp_sock *tp = tcp_sk(sp);
1745         const struct inet_connection_sock *icsk = inet_csk(sp);
1746         const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1747         int rx_queue;
1748         int state;
1749
1750         dest  = &sp->sk_v6_daddr;
1751         src   = &sp->sk_v6_rcv_saddr;
1752         destp = ntohs(inet->inet_dport);
1753         srcp  = ntohs(inet->inet_sport);
1754
1755         if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
1756             icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
1757             icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
1758                 timer_active    = 1;
1759                 timer_expires   = icsk->icsk_timeout;
1760         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1761                 timer_active    = 4;
1762                 timer_expires   = icsk->icsk_timeout;
1763         } else if (timer_pending(&sp->sk_timer)) {
1764                 timer_active    = 2;
1765                 timer_expires   = sp->sk_timer.expires;
1766         } else {
1767                 timer_active    = 0;
1768                 timer_expires = jiffies;
1769         }
1770
1771         state = sk_state_load(sp);
1772         if (state == TCP_LISTEN)
1773                 rx_queue = sp->sk_ack_backlog;
1774         else
1775                 /* Because we don't lock the socket,
1776                  * we might find a transient negative value.
1777                  */
1778                 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
1779
1780         seq_printf(seq,
1781                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1782                    "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
1783                    i,
1784                    src->s6_addr32[0], src->s6_addr32[1],
1785                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1786                    dest->s6_addr32[0], dest->s6_addr32[1],
1787                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1788                    state,
1789                    tp->write_seq - tp->snd_una,
1790                    rx_queue,
1791                    timer_active,
1792                    jiffies_delta_to_clock_t(timer_expires - jiffies),
1793                    icsk->icsk_retransmits,
1794                    from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1795                    icsk->icsk_probes_out,
1796                    sock_i_ino(sp),
1797                    atomic_read(&sp->sk_refcnt), sp,
1798                    jiffies_to_clock_t(icsk->icsk_rto),
1799                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1800                    (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1801                    tp->snd_cwnd,
1802                    state == TCP_LISTEN ?
1803                         fastopenq->max_qlen :
1804                         (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
1805                    );
1806 }
1807
1808 static void get_timewait6_sock(struct seq_file *seq,
1809                                struct inet_timewait_sock *tw, int i)
1810 {
1811         long delta = tw->tw_timer.expires - jiffies;
1812         const struct in6_addr *dest, *src;
1813         __u16 destp, srcp;
1814
1815         dest = &tw->tw_v6_daddr;
1816         src  = &tw->tw_v6_rcv_saddr;
1817         destp = ntohs(tw->tw_dport);
1818         srcp  = ntohs(tw->tw_sport);
1819
1820         seq_printf(seq,
1821                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1822                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1823                    i,
1824                    src->s6_addr32[0], src->s6_addr32[1],
1825                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1826                    dest->s6_addr32[0], dest->s6_addr32[1],
1827                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1828                    tw->tw_substate, 0, 0,
1829                    3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
1830                    atomic_read(&tw->tw_refcnt), tw);
1831 }
1832
1833 static int tcp6_seq_show(struct seq_file *seq, void *v)
1834 {
1835         struct tcp_iter_state *st;
1836         struct sock *sk = v;
1837
1838         if (v == SEQ_START_TOKEN) {
1839                 seq_puts(seq,
1840                          "  sl  "
1841                          "local_address                         "
1842                          "remote_address                        "
1843                          "st tx_queue rx_queue tr tm->when retrnsmt"
1844                          "   uid  timeout inode\n");
1845                 goto out;
1846         }
1847         st = seq->private;
1848
1849         if (sk->sk_state == TCP_TIME_WAIT)
1850                 get_timewait6_sock(seq, v, st->num);
1851         else if (sk->sk_state == TCP_NEW_SYN_RECV)
1852                 get_openreq6(seq, v, st->num);
1853         else
1854                 get_tcp6_sock(seq, v, st->num);
1855 out:
1856         return 0;
1857 }
1858
1859 static const struct file_operations tcp6_afinfo_seq_fops = {
1860         .owner   = THIS_MODULE,
1861         .open    = tcp_seq_open,
1862         .read    = seq_read,
1863         .llseek  = seq_lseek,
1864         .release = seq_release_net
1865 };
1866
1867 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
1868         .name           = "tcp6",
1869         .family         = AF_INET6,
1870         .seq_fops       = &tcp6_afinfo_seq_fops,
1871         .seq_ops        = {
1872                 .show           = tcp6_seq_show,
1873         },
1874 };
1875
1876 int __net_init tcp6_proc_init(struct net *net)
1877 {
1878         return tcp_proc_register(net, &tcp6_seq_afinfo);
1879 }
1880
1881 void tcp6_proc_exit(struct net *net)
1882 {
1883         tcp_proc_unregister(net, &tcp6_seq_afinfo);
1884 }
1885 #endif
1886
1887 struct proto tcpv6_prot = {
1888         .name                   = "TCPv6",
1889         .owner                  = THIS_MODULE,
1890         .close                  = tcp_close,
1891         .connect                = tcp_v6_connect,
1892         .disconnect             = tcp_disconnect,
1893         .accept                 = inet_csk_accept,
1894         .ioctl                  = tcp_ioctl,
1895         .init                   = tcp_v6_init_sock,
1896         .destroy                = tcp_v6_destroy_sock,
1897         .shutdown               = tcp_shutdown,
1898         .setsockopt             = tcp_setsockopt,
1899         .getsockopt             = tcp_getsockopt,
1900         .keepalive              = tcp_set_keepalive,
1901         .recvmsg                = tcp_recvmsg,
1902         .sendmsg                = tcp_sendmsg,
1903         .sendpage               = tcp_sendpage,
1904         .backlog_rcv            = tcp_v6_do_rcv,
1905         .release_cb             = tcp_release_cb,
1906         .hash                   = inet6_hash,
1907         .unhash                 = inet_unhash,
1908         .get_port               = inet_csk_get_port,
1909         .enter_memory_pressure  = tcp_enter_memory_pressure,
1910         .stream_memory_free     = tcp_stream_memory_free,
1911         .sockets_allocated      = &tcp_sockets_allocated,
1912         .memory_allocated       = &tcp_memory_allocated,
1913         .memory_pressure        = &tcp_memory_pressure,
1914         .orphan_count           = &tcp_orphan_count,
1915         .sysctl_mem             = sysctl_tcp_mem,
1916         .sysctl_wmem            = sysctl_tcp_wmem,
1917         .sysctl_rmem            = sysctl_tcp_rmem,
1918         .max_header             = MAX_TCP_HEADER,
1919         .obj_size               = sizeof(struct tcp6_sock),
1920         .slab_flags             = SLAB_TYPESAFE_BY_RCU,
1921         .twsk_prot              = &tcp6_timewait_sock_ops,
1922         .rsk_prot               = &tcp6_request_sock_ops,
1923         .h.hashinfo             = &tcp_hashinfo,
1924         .no_autobind            = true,
1925 #ifdef CONFIG_COMPAT
1926         .compat_setsockopt      = compat_tcp_setsockopt,
1927         .compat_getsockopt      = compat_tcp_getsockopt,
1928 #endif
1929         .diag_destroy           = tcp_abort,
1930 };
1931
1932 static struct inet6_protocol tcpv6_protocol = {
1933         .early_demux    =       tcp_v6_early_demux,
1934         .early_demux_handler =  tcp_v6_early_demux,
1935         .handler        =       tcp_v6_rcv,
1936         .err_handler    =       tcp_v6_err,
1937         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1938 };
1939
1940 static struct inet_protosw tcpv6_protosw = {
1941         .type           =       SOCK_STREAM,
1942         .protocol       =       IPPROTO_TCP,
1943         .prot           =       &tcpv6_prot,
1944         .ops            =       &inet6_stream_ops,
1945         .flags          =       INET_PROTOSW_PERMANENT |
1946                                 INET_PROTOSW_ICSK,
1947 };
1948
1949 static int __net_init tcpv6_net_init(struct net *net)
1950 {
1951         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
1952                                     SOCK_RAW, IPPROTO_TCP, net);
1953 }
1954
1955 static void __net_exit tcpv6_net_exit(struct net *net)
1956 {
1957         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
1958 }
1959
1960 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
1961 {
1962         inet_twsk_purge(&tcp_hashinfo, AF_INET6);
1963 }
1964
1965 static struct pernet_operations tcpv6_net_ops = {
1966         .init       = tcpv6_net_init,
1967         .exit       = tcpv6_net_exit,
1968         .exit_batch = tcpv6_net_exit_batch,
1969 };
1970
1971 int __init tcpv6_init(void)
1972 {
1973         int ret;
1974
1975         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
1976         if (ret)
1977                 goto out;
1978
1979         /* register inet6 protocol */
1980         ret = inet6_register_protosw(&tcpv6_protosw);
1981         if (ret)
1982                 goto out_tcpv6_protocol;
1983
1984         ret = register_pernet_subsys(&tcpv6_net_ops);
1985         if (ret)
1986                 goto out_tcpv6_protosw;
1987 out:
1988         return ret;
1989
1990 out_tcpv6_protosw:
1991         inet6_unregister_protosw(&tcpv6_protosw);
1992 out_tcpv6_protocol:
1993         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
1994         goto out;
1995 }
1996
1997 void tcpv6_exit(void)
1998 {
1999         unregister_pernet_subsys(&tcpv6_net_ops);
2000         inet6_unregister_protosw(&tcpv6_protosw);
2001         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2002 }