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
  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/netdma.h>
  63 #include <net/inet_common.h>
  64 #include <net/secure_seq.h>
  65 #include <net/tcp_memcontrol.h>
  66
  67 #include <asm/uaccess.h>
  68
  69 #include <linux/proc_fs.h>
  70 #include <linux/seq_file.h>
  71
  72 #include <linux/crypto.h>
  73 #include <linux/scatterlist.h>
  74
  75 static void     tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
  76 static void     tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
  77                                       struct request_sock *req);
  78
  79 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  80 static void     __tcp_v6_send_check(struct sk_buff *skb,
  81                                     const struct in6_addr *saddr,
  82                                     const struct in6_addr *daddr);
  83
  84 static const struct inet_connection_sock_af_ops ipv6_mapped;
  85 static const struct inet_connection_sock_af_ops ipv6_specific;
  86 #ifdef CONFIG_TCP_MD5SIG
  87 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
  88 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
  89 #else
  90 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
  91                                                    const struct in6_addr *addr)
  92 {
  93         return NULL;
  94 }
  95 #endif
  96
  97 static void tcp_v6_hash(struct sock *sk)
  98 {
  99         if (sk->sk_state != TCP_CLOSE) {
 100                 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
 101                         tcp_prot.hash(sk);
 102                         return;
 103                 }
 104                 local_bh_disable();
 105                 __inet6_hash(sk, NULL);
 106                 local_bh_enable();
 107         }
 108 }
 109
 110 static __inline__ __sum16 tcp_v6_check(int len,
 111                                    const struct in6_addr *saddr,
 112                                    const struct in6_addr *daddr,
 113                                    __wsum base)
 114 {
 115         return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
 116 }
 117
 118 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
 119 {
 120         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
 121                                             ipv6_hdr(skb)->saddr.s6_addr32,
 122                                             tcp_hdr(skb)->dest,
 123                                             tcp_hdr(skb)->source);
 124 }
 125
 126 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 127                           int addr_len)
 128 {
 129         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 130         struct inet_sock *inet = inet_sk(sk);
 131         struct inet_connection_sock *icsk = inet_csk(sk);
 132         struct ipv6_pinfo *np = inet6_sk(sk);
 133         struct tcp_sock *tp = tcp_sk(sk);
 134         struct in6_addr *saddr = NULL, *final_p, final;
 135         struct rt6_info *rt;
 136         struct flowi6 fl6;
 137         struct dst_entry *dst;
 138         int addr_type;
 139         int err;
 140
 141         if (addr_len < SIN6_LEN_RFC2133)
 142                 return -EINVAL;
 143
 144         if (usin->sin6_family != AF_INET6)
 145                 return -EAFNOSUPPORT;
 146
 147         memset(&fl6, 0, sizeof(fl6));
 148
 149         if (np->sndflow) {
 150                 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 151                 IP6_ECN_flow_init(fl6.flowlabel);
 152                 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
 153                         struct ip6_flowlabel *flowlabel;
 154                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 155                         if (flowlabel == NULL)
 156                                 return -EINVAL;
 157                         usin->sin6_addr = flowlabel->dst;
 158                         fl6_sock_release(flowlabel);
 159                 }
 160         }
 161
 162         /*
 163          *      connect() to INADDR_ANY means loopback (BSD'ism).
 164          */
 165
 166         if(ipv6_addr_any(&usin->sin6_addr))
 167                 usin->sin6_addr.s6_addr[15] = 0x1;
 168
 169         addr_type = ipv6_addr_type(&usin->sin6_addr);
 170
 171         if(addr_type & IPV6_ADDR_MULTICAST)
 172                 return -ENETUNREACH;
 173
 174         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 175                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 176                     usin->sin6_scope_id) {
 177                         /* If interface is set while binding, indices
 178                          * must coincide.
 179                          */
 180                         if (sk->sk_bound_dev_if &&
 181                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 182                                 return -EINVAL;
 183
 184                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 185                 }
 186
 187                 /* Connect to link-local address requires an interface */
 188                 if (!sk->sk_bound_dev_if)
 189                         return -EINVAL;
 190         }
 191
 192         if (tp->rx_opt.ts_recent_stamp &&
 193             !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
 194                 tp->rx_opt.ts_recent = 0;
 195                 tp->rx_opt.ts_recent_stamp = 0;
 196                 tp->write_seq = 0;
 197         }
 198
 199         np->daddr = usin->sin6_addr;
 200         np->flow_label = fl6.flowlabel;
 201
 202         /*
 203          *      TCP over IPv4
 204          */
 205
 206         if (addr_type == IPV6_ADDR_MAPPED) {
 207                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 208                 struct sockaddr_in sin;
 209
 210                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 211
 212                 if (__ipv6_only_sock(sk))
 213                         return -ENETUNREACH;
 214
 215                 sin.sin_family = AF_INET;
 216                 sin.sin_port = usin->sin6_port;
 217                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 218
 219                 icsk->icsk_af_ops = &ipv6_mapped;
 220                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 221 #ifdef CONFIG_TCP_MD5SIG
 222                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
 223 #endif
 224
 225                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 226
 227                 if (err) {
 228                         icsk->icsk_ext_hdr_len = exthdrlen;
 229                         icsk->icsk_af_ops = &ipv6_specific;
 230                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 231 #ifdef CONFIG_TCP_MD5SIG
 232                         tp->af_specific = &tcp_sock_ipv6_specific;
 233 #endif
 234                         goto failure;
 235                 } else {
 236                         ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
 237                         ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
 238                                                &np->rcv_saddr);
 239                 }
 240
 241                 return err;
 242         }
 243
 244         if (!ipv6_addr_any(&np->rcv_saddr))
 245                 saddr = &np->rcv_saddr;
 246
 247         fl6.flowi6_proto = IPPROTO_TCP;
 248         fl6.daddr = np->daddr;
 249         fl6.saddr = saddr ? *saddr : np->saddr;
 250         fl6.flowi6_oif = sk->sk_bound_dev_if;
 251         fl6.flowi6_mark = sk->sk_mark;
 252         fl6.fl6_dport = usin->sin6_port;
 253         fl6.fl6_sport = inet->inet_sport;
 254
 255         final_p = fl6_update_dst(&fl6, np->opt, &final);
 256
 257         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 258
 259         dst = ip6_dst_lookup_flow(sk, &fl6, final_p, true);
 260         if (IS_ERR(dst)) {
 261                 err = PTR_ERR(dst);
 262                 goto failure;
 263         }
 264
 265         if (saddr == NULL) {
 266                 saddr = &fl6.saddr;
 267                 np->rcv_saddr = *saddr;
 268         }
 269
 270         /* set the source address */
 271         np->saddr = *saddr;
 272         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 273
 274         sk->sk_gso_type = SKB_GSO_TCPV6;
 275         __ip6_dst_store(sk, dst, NULL, NULL);
 276
 277         rt = (struct rt6_info *) dst;
 278         if (tcp_death_row.sysctl_tw_recycle &&
 279             !tp->rx_opt.ts_recent_stamp &&
 280             ipv6_addr_equal(&rt->rt6i_dst.addr, &np->daddr)) {
 281                 struct inet_peer *peer = rt6_get_peer(rt);
 282                 /*
 283                  * VJ's idea. We save last timestamp seen from
 284                  * the destination in peer table, when entering state
 285                  * TIME-WAIT * and initialize rx_opt.ts_recent from it,
 286                  * when trying new connection.
 287                  */
 288                 if (peer) {
 289                         inet_peer_refcheck(peer);
 290                         if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
 291                                 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
 292                                 tp->rx_opt.ts_recent = peer->tcp_ts;
 293                         }
 294                 }
 295         }
 296
 297         icsk->icsk_ext_hdr_len = 0;
 298         if (np->opt)
 299                 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
 300                                           np->opt->opt_nflen);
 301
 302         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 303
 304         inet->inet_dport = usin->sin6_port;
 305
 306         tcp_set_state(sk, TCP_SYN_SENT);
 307         err = inet6_hash_connect(&tcp_death_row, sk);
 308         if (err)
 309                 goto late_failure;
 310
 311         if (!tp->write_seq)
 312                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 313                                                              np->daddr.s6_addr32,
 314                                                              inet->inet_sport,
 315                                                              inet->inet_dport);
 316
 317         err = tcp_connect(sk);
 318         if (err)
 319                 goto late_failure;
 320
 321         return 0;
 322
 323 late_failure:
 324         tcp_set_state(sk, TCP_CLOSE);
 325         __sk_dst_reset(sk);
 326 failure:
 327         inet->inet_dport = 0;
 328         sk->sk_route_caps = 0;
 329         return err;
 330 }
 331
 332 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 333                 u8 type, u8 code, int offset, __be32 info)
 334 {
 335         const struct ipv6hdr *hdr = (const struct ipv6hdr*)skb->data;
 336         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 337         struct ipv6_pinfo *np;
 338         struct sock *sk;
 339         int err;
 340         struct tcp_sock *tp;
 341         __u32 seq;
 342         struct net *net = dev_net(skb->dev);
 343
 344         sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
 345                         th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
 346
 347         if (sk == NULL) {
 348                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 349                                    ICMP6_MIB_INERRORS);
 350                 return;
 351         }
 352
 353         if (sk->sk_state == TCP_TIME_WAIT) {
 354                 inet_twsk_put(inet_twsk(sk));
 355                 return;
 356         }
 357
 358         bh_lock_sock(sk);
 359         if (sock_owned_by_user(sk))
 360                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 361
 362         if (sk->sk_state == TCP_CLOSE)
 363                 goto out;
 364
 365         if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
 366                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
 367                 goto out;
 368         }
 369
 370         tp = tcp_sk(sk);
 371         seq = ntohl(th->seq);
 372         if (sk->sk_state != TCP_LISTEN &&
 373             !between(seq, tp->snd_una, tp->snd_nxt)) {
 374                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 375                 goto out;
 376         }
 377
 378         np = inet6_sk(sk);
 379
 380         if (type == ICMPV6_PKT_TOOBIG) {
 381                 struct dst_entry *dst;
 382
 383                 if (sock_owned_by_user(sk))
 384                         goto out;
 385                 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 386                         goto out;
 387
 388                 /* icmp should have updated the destination cache entry */
 389                 dst = __sk_dst_check(sk, np->dst_cookie);
 390
 391                 if (dst == NULL) {
 392                         struct inet_sock *inet = inet_sk(sk);
 393                         struct flowi6 fl6;
 394
 395                         /* BUGGG_FUTURE: Again, it is not clear how
 396                            to handle rthdr case. Ignore this complexity
 397                            for now.
 398                          */
 399                         memset(&fl6, 0, sizeof(fl6));
 400                         fl6.flowi6_proto = IPPROTO_TCP;
 401                         fl6.daddr = np->daddr;
 402                         fl6.saddr = np->saddr;
 403                         fl6.flowi6_oif = sk->sk_bound_dev_if;
 404                         fl6.flowi6_mark = sk->sk_mark;
 405                         fl6.fl6_dport = inet->inet_dport;
 406                         fl6.fl6_sport = inet->inet_sport;
 407                         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 408
 409                         dst = ip6_dst_lookup_flow(sk, &fl6, NULL, false);
 410                         if (IS_ERR(dst)) {
 411                                 sk->sk_err_soft = -PTR_ERR(dst);
 412                                 goto out;
 413                         }
 414
 415                 } else
 416                         dst_hold(dst);
 417
 418                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
 419                         tcp_sync_mss(sk, dst_mtu(dst));
 420                         tcp_simple_retransmit(sk);
 421                 } /* else let the usual retransmit timer handle it */
 422                 dst_release(dst);
 423                 goto out;
 424         }
 425
 426         icmpv6_err_convert(type, code, &err);
 427
 428         /* Might be for an request_sock */
 429         switch (sk->sk_state) {
 430                 struct request_sock *req, **prev;
 431         case TCP_LISTEN:
 432                 if (sock_owned_by_user(sk))
 433                         goto out;
 434
 435                 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
 436                                            &hdr->saddr, inet6_iif(skb));
 437                 if (!req)
 438                         goto out;
 439
 440                 /* ICMPs are not backlogged, hence we cannot get
 441                  * an established socket here.
 442                  */
 443                 WARN_ON(req->sk != NULL);
 444
 445                 if (seq != tcp_rsk(req)->snt_isn) {
 446                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 447                         goto out;
 448                 }
 449
 450                 inet_csk_reqsk_queue_drop(sk, req, prev);
 451                 goto out;
 452
 453         case TCP_SYN_SENT:
 454         case TCP_SYN_RECV:  /* Cannot happen.
 455                                It can, it SYNs are crossed. --ANK */
 456                 if (!sock_owned_by_user(sk)) {
 457                         sk->sk_err = err;
 458                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 459
 460                         tcp_done(sk);
 461                 } else
 462                         sk->sk_err_soft = err;
 463                 goto out;
 464         }
 465
 466         if (!sock_owned_by_user(sk) && np->recverr) {
 467                 sk->sk_err = err;
 468                 sk->sk_error_report(sk);
 469         } else
 470                 sk->sk_err_soft = err;
 471
 472 out:
 473         bh_unlock_sock(sk);
 474         sock_put(sk);
 475 }
 476
 477
 478 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
 479                               struct request_values *rvp)
 480 {
 481         struct inet6_request_sock *treq = inet6_rsk(req);
 482         struct ipv6_pinfo *np = inet6_sk(sk);
 483         struct sk_buff * skb;
 484         struct ipv6_txoptions *opt = NULL;
 485         struct in6_addr * final_p, final;
 486         struct flowi6 fl6;
 487         struct dst_entry *dst;
 488         int err;
 489
 490         memset(&fl6, 0, sizeof(fl6));
 491         fl6.flowi6_proto = IPPROTO_TCP;
 492         fl6.daddr = treq->rmt_addr;
 493         fl6.saddr = treq->loc_addr;
 494         fl6.flowlabel = 0;
 495         fl6.flowi6_oif = treq->iif;
 496         fl6.flowi6_mark = sk->sk_mark;
 497         fl6.fl6_dport = inet_rsk(req)->rmt_port;
 498         fl6.fl6_sport = inet_rsk(req)->loc_port;
 499         security_req_classify_flow(req, flowi6_to_flowi(&fl6));
 500
 501         opt = np->opt;
 502         final_p = fl6_update_dst(&fl6, opt, &final);
 503
 504         dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false);
 505         if (IS_ERR(dst)) {
 506                 err = PTR_ERR(dst);
 507                 dst = NULL;
 508                 goto done;
 509         }
 510         skb = tcp_make_synack(sk, dst, req, rvp);
 511         err = -ENOMEM;
 512         if (skb) {
 513                 __tcp_v6_send_check(skb, &treq->loc_addr, &treq->rmt_addr);
 514
 515                 fl6.daddr = treq->rmt_addr;
 516                 err = ip6_xmit(sk, skb, &fl6, opt, np->tclass);
 517                 err = net_xmit_eval(err);
 518         }
 519
 520 done:
 521         if (opt && opt != np->opt)
 522                 sock_kfree_s(sk, opt, opt->tot_len);
 523         dst_release(dst);
 524         return err;
 525 }
 526
 527 static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req,
 528                              struct request_values *rvp)
 529 {
 530         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
 531         return tcp_v6_send_synack(sk, req, rvp);
 532 }
 533
 534 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 535 {
 536         kfree_skb(inet6_rsk(req)->pktopts);
 537 }
 538
 539 #ifdef CONFIG_TCP_MD5SIG
 540 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
 541                                                    const struct in6_addr *addr)
 542 {
 543         return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
 544 }
 545
 546 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
 547                                                 struct sock *addr_sk)
 548 {
 549         return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
 550 }
 551
 552 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
 553                                                       struct request_sock *req)
 554 {
 555         return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
 556 }
 557
 558 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
 559                                   int optlen)
 560 {
 561         struct tcp_md5sig cmd;
 562         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 563
 564         if (optlen < sizeof(cmd))
 565                 return -EINVAL;
 566
 567         if (copy_from_user(&cmd, optval, sizeof(cmd)))
 568                 return -EFAULT;
 569
 570         if (sin6->sin6_family != AF_INET6)
 571                 return -EINVAL;
 572
 573         if (!cmd.tcpm_keylen) {
 574                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 575                         return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 576                                               AF_INET);
 577                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 578                                       AF_INET6);
 579         }
 580
 581         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 582                 return -EINVAL;
 583
 584         if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 585                 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
 586                                       AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 587
 588         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
 589                               AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 590 }
 591
 592 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
 593                                         const struct in6_addr *daddr,
 594                                         const struct in6_addr *saddr, int nbytes)
 595 {
 596         struct tcp6_pseudohdr *bp;
 597         struct scatterlist sg;
 598
 599         bp = &hp->md5_blk.ip6;
 600         /* 1. TCP pseudo-header (RFC2460) */
 601         bp->saddr = *saddr;
 602         bp->daddr = *daddr;
 603         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 604         bp->len = cpu_to_be32(nbytes);
 605
 606         sg_init_one(&sg, bp, sizeof(*bp));
 607         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
 608 }
 609
 610 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
 611                                const struct in6_addr *daddr, struct in6_addr *saddr,
 612                                const struct tcphdr *th)
 613 {
 614         struct tcp_md5sig_pool *hp;
 615         struct hash_desc *desc;
 616
 617         hp = tcp_get_md5sig_pool();
 618         if (!hp)
 619                 goto clear_hash_noput;
 620         desc = &hp->md5_desc;
 621
 622         if (crypto_hash_init(desc))
 623                 goto clear_hash;
 624         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
 625                 goto clear_hash;
 626         if (tcp_md5_hash_header(hp, th))
 627                 goto clear_hash;
 628         if (tcp_md5_hash_key(hp, key))
 629                 goto clear_hash;
 630         if (crypto_hash_final(desc, md5_hash))
 631                 goto clear_hash;
 632
 633         tcp_put_md5sig_pool();
 634         return 0;
 635
 636 clear_hash:
 637         tcp_put_md5sig_pool();
 638 clear_hash_noput:
 639         memset(md5_hash, 0, 16);
 640         return 1;
 641 }
 642
 643 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
 644                                const struct sock *sk,
 645                                const struct request_sock *req,
 646                                const struct sk_buff *skb)
 647 {
 648         const struct in6_addr *saddr, *daddr;
 649         struct tcp_md5sig_pool *hp;
 650         struct hash_desc *desc;
 651         const struct tcphdr *th = tcp_hdr(skb);
 652
 653         if (sk) {
 654                 saddr = &inet6_sk(sk)->saddr;
 655                 daddr = &inet6_sk(sk)->daddr;
 656         } else if (req) {
 657                 saddr = &inet6_rsk(req)->loc_addr;
 658                 daddr = &inet6_rsk(req)->rmt_addr;
 659         } else {
 660                 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 661                 saddr = &ip6h->saddr;
 662                 daddr = &ip6h->daddr;
 663         }
 664
 665         hp = tcp_get_md5sig_pool();
 666         if (!hp)
 667                 goto clear_hash_noput;
 668         desc = &hp->md5_desc;
 669
 670         if (crypto_hash_init(desc))
 671                 goto clear_hash;
 672
 673         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
 674                 goto clear_hash;
 675         if (tcp_md5_hash_header(hp, th))
 676                 goto clear_hash;
 677         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 678                 goto clear_hash;
 679         if (tcp_md5_hash_key(hp, key))
 680                 goto clear_hash;
 681         if (crypto_hash_final(desc, md5_hash))
 682                 goto clear_hash;
 683
 684         tcp_put_md5sig_pool();
 685         return 0;
 686
 687 clear_hash:
 688         tcp_put_md5sig_pool();
 689 clear_hash_noput:
 690         memset(md5_hash, 0, 16);
 691         return 1;
 692 }
 693
 694 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
 695 {
 696         const __u8 *hash_location = NULL;
 697         struct tcp_md5sig_key *hash_expected;
 698         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 699         const struct tcphdr *th = tcp_hdr(skb);
 700         int genhash;
 701         u8 newhash[16];
 702
 703         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 704         hash_location = tcp_parse_md5sig_option(th);
 705
 706         /* We've parsed the options - do we have a hash? */
 707         if (!hash_expected && !hash_location)
 708                 return 0;
 709
 710         if (hash_expected && !hash_location) {
 711                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 712                 return 1;
 713         }
 714
 715         if (!hash_expected && hash_location) {
 716                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 717                 return 1;
 718         }
 719
 720         /* check the signature */
 721         genhash = tcp_v6_md5_hash_skb(newhash,
 722                                       hash_expected,
 723                                       NULL, NULL, skb);
 724
 725         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 726                 if (net_ratelimit()) {
 727                         printk(KERN_INFO "MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
 728                                genhash ? "failed" : "mismatch",
 729                                &ip6h->saddr, ntohs(th->source),
 730                                &ip6h->daddr, ntohs(th->dest));
 731                 }
 732                 return 1;
 733         }
 734         return 0;
 735 }
 736 #endif
 737
 738 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 739         .family         =       AF_INET6,
 740         .obj_size       =       sizeof(struct tcp6_request_sock),
 741         .rtx_syn_ack    =       tcp_v6_rtx_synack,
 742         .send_ack       =       tcp_v6_reqsk_send_ack,
 743         .destructor     =       tcp_v6_reqsk_destructor,
 744         .send_reset     =       tcp_v6_send_reset,
 745         .syn_ack_timeout =      tcp_syn_ack_timeout,
 746 };
 747
 748 #ifdef CONFIG_TCP_MD5SIG
 749 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 750         .md5_lookup     =       tcp_v6_reqsk_md5_lookup,
 751         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 752 };
 753 #endif
 754
 755 static void __tcp_v6_send_check(struct sk_buff *skb,
 756                                 const struct in6_addr *saddr, const struct in6_addr *daddr)
 757 {
 758         struct tcphdr *th = tcp_hdr(skb);
 759
 760         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 761                 th->check = ~tcp_v6_check(skb->len, saddr, daddr, 0);
 762                 skb->csum_start = skb_transport_header(skb) - skb->head;
 763                 skb->csum_offset = offsetof(struct tcphdr, check);
 764         } else {
 765                 th->check = tcp_v6_check(skb->len, saddr, daddr,
 766                                          csum_partial(th, th->doff << 2,
 767                                                       skb->csum));
 768         }
 769 }
 770
 771 static void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
 772 {
 773         struct ipv6_pinfo *np = inet6_sk(sk);
 774
 775         __tcp_v6_send_check(skb, &np->saddr, &np->daddr);
 776 }
 777
 778 static int tcp_v6_gso_send_check(struct sk_buff *skb)
 779 {
 780         const struct ipv6hdr *ipv6h;
 781         struct tcphdr *th;
 782
 783         if (!pskb_may_pull(skb, sizeof(*th)))
 784                 return -EINVAL;
 785
 786         ipv6h = ipv6_hdr(skb);
 787         th = tcp_hdr(skb);
 788
 789         th->check = 0;
 790         skb->ip_summed = CHECKSUM_PARTIAL;
 791         __tcp_v6_send_check(skb, &ipv6h->saddr, &ipv6h->daddr);
 792         return 0;
 793 }
 794
 795 static struct sk_buff **tcp6_gro_receive(struct sk_buff **head,
 796                                          struct sk_buff *skb)
 797 {
 798         const struct ipv6hdr *iph = skb_gro_network_header(skb);
 799
 800         switch (skb->ip_summed) {
 801         case CHECKSUM_COMPLETE:
 802                 if (!tcp_v6_check(skb_gro_len(skb), &iph->saddr, &iph->daddr,
 803                                   skb->csum)) {
 804                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 805                         break;
 806                 }
 807
 808                 /* fall through */
 809         case CHECKSUM_NONE:
 810                 NAPI_GRO_CB(skb)->flush = 1;
 811                 return NULL;
 812         }
 813
 814         return tcp_gro_receive(head, skb);
 815 }
 816
 817 static int tcp6_gro_complete(struct sk_buff *skb)
 818 {
 819         const struct ipv6hdr *iph = ipv6_hdr(skb);
 820         struct tcphdr *th = tcp_hdr(skb);
 821
 822         th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
 823                                   &iph->saddr, &iph->daddr, 0);
 824         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 825
 826         return tcp_gro_complete(skb);
 827 }
 828
 829 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
 830                                  u32 ts, struct tcp_md5sig_key *key, int rst, u8 tclass)
 831 {
 832         const struct tcphdr *th = tcp_hdr(skb);
 833         struct tcphdr *t1;
 834         struct sk_buff *buff;
 835         struct flowi6 fl6;
 836         struct net *net = dev_net(skb_dst(skb)->dev);
 837         struct sock *ctl_sk = net->ipv6.tcp_sk;
 838         unsigned int tot_len = sizeof(struct tcphdr);
 839         struct dst_entry *dst;
 840         __be32 *topt;
 841
 842         if (ts)
 843                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
 844 #ifdef CONFIG_TCP_MD5SIG
 845         if (key)
 846                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
 847 #endif
 848
 849         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
 850                          GFP_ATOMIC);
 851         if (buff == NULL)
 852                 return;
 853
 854         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
 855
 856         t1 = (struct tcphdr *) skb_push(buff, tot_len);
 857         skb_reset_transport_header(buff);
 858
 859         /* Swap the send and the receive. */
 860         memset(t1, 0, sizeof(*t1));
 861         t1->dest = th->source;
 862         t1->source = th->dest;
 863         t1->doff = tot_len / 4;
 864         t1->seq = htonl(seq);
 865         t1->ack_seq = htonl(ack);
 866         t1->ack = !rst || !th->ack;
 867         t1->rst = rst;
 868         t1->window = htons(win);
 869
 870         topt = (__be32 *)(t1 + 1);
 871
 872         if (ts) {
 873                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 874                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
 875                 *topt++ = htonl(tcp_time_stamp);
 876                 *topt++ = htonl(ts);
 877         }
 878
 879 #ifdef CONFIG_TCP_MD5SIG
 880         if (key) {
 881                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 882                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 883                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
 884                                     &ipv6_hdr(skb)->saddr,
 885                                     &ipv6_hdr(skb)->daddr, t1);
 886         }
 887 #endif
 888
 889         memset(&fl6, 0, sizeof(fl6));
 890         fl6.daddr = ipv6_hdr(skb)->saddr;
 891         fl6.saddr = ipv6_hdr(skb)->daddr;
 892
 893         buff->ip_summed = CHECKSUM_PARTIAL;
 894         buff->csum = 0;
 895
 896         __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
 897
 898         fl6.flowi6_proto = IPPROTO_TCP;
 899         fl6.flowi6_oif = inet6_iif(skb);
 900         fl6.fl6_dport = t1->dest;
 901         fl6.fl6_sport = t1->source;
 902         security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
 903
 904         /* Pass a socket to ip6_dst_lookup either it is for RST
 905          * Underlying function will use this to retrieve the network
 906          * namespace
 907          */
 908         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false);
 909         if (!IS_ERR(dst)) {
 910                 skb_dst_set(buff, dst);
 911                 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
 912                 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 913                 if (rst)
 914                         TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
 915                 return;
 916         }
 917
 918         kfree_skb(buff);
 919 }
 920
 921 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
 922 {
 923         const struct tcphdr *th = tcp_hdr(skb);
 924         u32 seq = 0, ack_seq = 0;
 925         struct tcp_md5sig_key *key = NULL;
 926 #ifdef CONFIG_TCP_MD5SIG
 927         const __u8 *hash_location = NULL;
 928         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 929         unsigned char newhash[16];
 930         int genhash;
 931         struct sock *sk1 = NULL;
 932 #endif
 933
 934         if (th->rst)
 935                 return;
 936
 937         if (!ipv6_unicast_destination(skb))
 938                 return;
 939
 940 #ifdef CONFIG_TCP_MD5SIG
 941         hash_location = tcp_parse_md5sig_option(th);
 942         if (!sk && hash_location) {
 943                 /*
 944                  * active side is lost. Try to find listening socket through
 945                  * source port, and then find md5 key through listening socket.
 946                  * we are not loose security here:
 947                  * Incoming packet is checked with md5 hash with finding key,
 948                  * no RST generated if md5 hash doesn't match.
 949                  */
 950                 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
 951                                            &tcp_hashinfo, &ipv6h->daddr,
 952                                            ntohs(th->source), inet6_iif(skb));
 953                 if (!sk1)
 954                         return;
 955
 956                 rcu_read_lock();
 957                 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
 958                 if (!key)
 959                         goto release_sk1;
 960
 961                 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
 962                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
 963                         goto release_sk1;
 964         } else {
 965                 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
 966         }
 967 #endif
 968
 969         if (th->ack)
 970                 seq = ntohl(th->ack_seq);
 971         else
 972                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
 973                           (th->doff << 2);
 974
 975         tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1, 0);
 976
 977 #ifdef CONFIG_TCP_MD5SIG
 978 release_sk1:
 979         if (sk1) {
 980                 rcu_read_unlock();
 981                 sock_put(sk1);
 982         }
 983 #endif
 984 }
 985
 986 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
 987                             struct tcp_md5sig_key *key, u8 tclass)
 988 {
 989         tcp_v6_send_response(skb, seq, ack, win, ts, key, 0, tclass);
 990 }
 991
 992 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
 993 {
 994         struct inet_timewait_sock *tw = inet_twsk(sk);
 995         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 996
 997         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 998                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 999                         tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw),
1000                         tw->tw_tclass);
1001
1002         inet_twsk_put(tw);
1003 }
1004
1005 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1006                                   struct request_sock *req)
1007 {
1008         tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1009                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr), 0);
1010 }
1011
1012
1013 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1014 {
1015         struct request_sock *req, **prev;
1016         const struct tcphdr *th = tcp_hdr(skb);
1017         struct sock *nsk;
1018
1019         /* Find possible connection requests. */
1020         req = inet6_csk_search_req(sk, &prev, th->source,
1021                                    &ipv6_hdr(skb)->saddr,
1022                                    &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1023         if (req)
1024                 return tcp_check_req(sk, skb, req, prev);
1025
1026         nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1027                         &ipv6_hdr(skb)->saddr, th->source,
1028                         &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1029
1030         if (nsk) {
1031                 if (nsk->sk_state != TCP_TIME_WAIT) {
1032                         bh_lock_sock(nsk);
1033                         return nsk;
1034                 }
1035                 inet_twsk_put(inet_twsk(nsk));
1036                 return NULL;
1037         }
1038
1039 #ifdef CONFIG_SYN_COOKIES
1040         if (!th->syn)
1041                 sk = cookie_v6_check(sk, skb);
1042 #endif
1043         return sk;
1044 }
1045
1046 /* FIXME: this is substantially similar to the ipv4 code.
1047  * Can some kind of merge be done? -- erics
1048  */
1049 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1050 {
1051         struct tcp_extend_values tmp_ext;
1052         struct tcp_options_received tmp_opt;
1053         const u8 *hash_location;
1054         struct request_sock *req;
1055         struct inet6_request_sock *treq;
1056         struct ipv6_pinfo *np = inet6_sk(sk);
1057         struct tcp_sock *tp = tcp_sk(sk);
1058         __u32 isn = TCP_SKB_CB(skb)->when;
1059         struct dst_entry *dst = NULL;
1060         int want_cookie = 0;
1061
1062         if (skb->protocol == htons(ETH_P_IP))
1063                 return tcp_v4_conn_request(sk, skb);
1064
1065         if (!ipv6_unicast_destination(skb))
1066                 goto drop;
1067
1068         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1069                 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
1070                 if (!want_cookie)
1071                         goto drop;
1072         }
1073
1074         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1075                 goto drop;
1076
1077         req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1078         if (req == NULL)
1079                 goto drop;
1080
1081 #ifdef CONFIG_TCP_MD5SIG
1082         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1083 #endif
1084
1085         tcp_clear_options(&tmp_opt);
1086         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1087         tmp_opt.user_mss = tp->rx_opt.user_mss;
1088         tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1089
1090         if (tmp_opt.cookie_plus > 0 &&
1091             tmp_opt.saw_tstamp &&
1092             !tp->rx_opt.cookie_out_never &&
1093             (sysctl_tcp_cookie_size > 0 ||
1094              (tp->cookie_values != NULL &&
1095               tp->cookie_values->cookie_desired > 0))) {
1096                 u8 *c;
1097                 u32 *d;
1098                 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1099                 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1100
1101                 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1102                         goto drop_and_free;
1103
1104                 /* Secret recipe starts with IP addresses */
1105                 d = (__force u32 *)&ipv6_hdr(skb)->daddr.s6_addr32[0];
1106                 *mess++ ^= *d++;
1107                 *mess++ ^= *d++;
1108                 *mess++ ^= *d++;
1109                 *mess++ ^= *d++;
1110                 d = (__force u32 *)&ipv6_hdr(skb)->saddr.s6_addr32[0];
1111                 *mess++ ^= *d++;
1112                 *mess++ ^= *d++;
1113                 *mess++ ^= *d++;
1114                 *mess++ ^= *d++;
1115
1116                 /* plus variable length Initiator Cookie */
1117                 c = (u8 *)mess;
1118                 while (l-- > 0)
1119                         *c++ ^= *hash_location++;
1120
1121                 want_cookie = 0;        /* not our kind of cookie */
1122                 tmp_ext.cookie_out_never = 0; /* false */
1123                 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1124         } else if (!tp->rx_opt.cookie_in_always) {
1125                 /* redundant indications, but ensure initialization. */
1126                 tmp_ext.cookie_out_never = 1; /* true */
1127                 tmp_ext.cookie_plus = 0;
1128         } else {
1129                 goto drop_and_free;
1130         }
1131         tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1132
1133         if (want_cookie && !tmp_opt.saw_tstamp)
1134                 tcp_clear_options(&tmp_opt);
1135
1136         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1137         tcp_openreq_init(req, &tmp_opt, skb);
1138
1139         treq = inet6_rsk(req);
1140         treq->rmt_addr = ipv6_hdr(skb)->saddr;
1141         treq->loc_addr = ipv6_hdr(skb)->daddr;
1142         if (!want_cookie || tmp_opt.tstamp_ok)
1143                 TCP_ECN_create_request(req, tcp_hdr(skb));
1144
1145         treq->iif = sk->sk_bound_dev_if;
1146
1147         /* So that link locals have meaning */
1148         if (!sk->sk_bound_dev_if &&
1149             ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1150                 treq->iif = inet6_iif(skb);
1151
1152         if (!isn) {
1153                 struct inet_peer *peer = NULL;
1154
1155                 if (ipv6_opt_accepted(sk, skb) ||
1156                     np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1157                     np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1158                         atomic_inc(&skb->users);
1159                         treq->pktopts = skb;
1160                 }
1161
1162                 if (want_cookie) {
1163                         isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1164                         req->cookie_ts = tmp_opt.tstamp_ok;
1165                         goto have_isn;
1166                 }
1167
1168                 /* VJ's idea. We save last timestamp seen
1169                  * from the destination in peer table, when entering
1170                  * state TIME-WAIT, and check against it before
1171                  * accepting new connection request.
1172                  *
1173                  * If "isn" is not zero, this request hit alive
1174                  * timewait bucket, so that all the necessary checks
1175                  * are made in the function processing timewait state.
1176                  */
1177                 if (tmp_opt.saw_tstamp &&
1178                     tcp_death_row.sysctl_tw_recycle &&
1179                     (dst = inet6_csk_route_req(sk, req)) != NULL &&
1180                     (peer = rt6_get_peer((struct rt6_info *)dst)) != NULL &&
1181                     ipv6_addr_equal((struct in6_addr *)peer->daddr.addr.a6,
1182                                     &treq->rmt_addr)) {
1183                         inet_peer_refcheck(peer);
1184                         if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1185                             (s32)(peer->tcp_ts - req->ts_recent) >
1186                                                         TCP_PAWS_WINDOW) {
1187                                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1188                                 goto drop_and_release;
1189                         }
1190                 }
1191                 /* Kill the following clause, if you dislike this way. */
1192                 else if (!sysctl_tcp_syncookies &&
1193                          (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1194                           (sysctl_max_syn_backlog >> 2)) &&
1195                          (!peer || !peer->tcp_ts_stamp) &&
1196                          (!dst || !dst_metric(dst, RTAX_RTT))) {
1197                         /* Without syncookies last quarter of
1198                          * backlog is filled with destinations,
1199                          * proven to be alive.
1200                          * It means that we continue to communicate
1201                          * to destinations, already remembered
1202                          * to the moment of synflood.
1203                          */
1204                         LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1205                                        &treq->rmt_addr, ntohs(tcp_hdr(skb)->source));
1206                         goto drop_and_release;
1207                 }
1208
1209                 isn = tcp_v6_init_sequence(skb);
1210         }
1211 have_isn:
1212         tcp_rsk(req)->snt_isn = isn;
1213         tcp_rsk(req)->snt_synack = tcp_time_stamp;
1214
1215         security_inet_conn_request(sk, skb, req);
1216
1217         if (tcp_v6_send_synack(sk, req,
1218                                (struct request_values *)&tmp_ext) ||
1219             want_cookie)
1220                 goto drop_and_free;
1221
1222         inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1223         return 0;
1224
1225 drop_and_release:
1226         dst_release(dst);
1227 drop_and_free:
1228         reqsk_free(req);
1229 drop:
1230         return 0; /* don't send reset */
1231 }
1232
1233 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1234                                           struct request_sock *req,
1235                                           struct dst_entry *dst)
1236 {
1237         struct inet6_request_sock *treq;
1238         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1239         struct tcp6_sock *newtcp6sk;
1240         struct inet_sock *newinet;
1241         struct tcp_sock *newtp;
1242         struct sock *newsk;
1243         struct ipv6_txoptions *opt;
1244 #ifdef CONFIG_TCP_MD5SIG
1245         struct tcp_md5sig_key *key;
1246 #endif
1247
1248         if (skb->protocol == htons(ETH_P_IP)) {
1249                 /*
1250                  *      v6 mapped
1251                  */
1252
1253                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1254
1255                 if (newsk == NULL)
1256                         return NULL;
1257
1258                 newtcp6sk = (struct tcp6_sock *)newsk;
1259                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1260
1261                 newinet = inet_sk(newsk);
1262                 newnp = inet6_sk(newsk);
1263                 newtp = tcp_sk(newsk);
1264
1265                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1266
1267                 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr);
1268
1269                 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1270
1271                 newnp->rcv_saddr = newnp->saddr;
1272
1273                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1274                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1275 #ifdef CONFIG_TCP_MD5SIG
1276                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1277 #endif
1278
1279                 newnp->ipv6_ac_list = NULL;
1280                 newnp->ipv6_fl_list = NULL;
1281                 newnp->pktoptions  = NULL;
1282                 newnp->opt         = NULL;
1283                 newnp->mcast_oif   = inet6_iif(skb);
1284                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1285
1286                 /*
1287                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1288                  * here, tcp_create_openreq_child now does this for us, see the comment in
1289                  * that function for the gory details. -acme
1290                  */
1291
1292                 /* It is tricky place. Until this moment IPv4 tcp
1293                    worked with IPv6 icsk.icsk_af_ops.
1294                    Sync it now.
1295                  */
1296                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1297
1298                 return newsk;
1299         }
1300
1301         treq = inet6_rsk(req);
1302         opt = np->opt;
1303
1304         if (sk_acceptq_is_full(sk))
1305                 goto out_overflow;
1306
1307         if (!dst) {
1308                 dst = inet6_csk_route_req(sk, req);
1309                 if (!dst)
1310                         goto out;
1311         }
1312
1313         newsk = tcp_create_openreq_child(sk, req, skb);
1314         if (newsk == NULL)
1315                 goto out_nonewsk;
1316
1317         /*
1318          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1319          * count here, tcp_create_openreq_child now does this for us, see the
1320          * comment in that function for the gory details. -acme
1321          */
1322
1323         newsk->sk_gso_type = SKB_GSO_TCPV6;
1324         __ip6_dst_store(newsk, dst, NULL, NULL);
1325
1326         newtcp6sk = (struct tcp6_sock *)newsk;
1327         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1328
1329         newtp = tcp_sk(newsk);
1330         newinet = inet_sk(newsk);
1331         newnp = inet6_sk(newsk);
1332
1333         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1334
1335         newnp->daddr = treq->rmt_addr;
1336         newnp->saddr = treq->loc_addr;
1337         newnp->rcv_saddr = treq->loc_addr;
1338         newsk->sk_bound_dev_if = treq->iif;
1339
1340         /* Now IPv6 options...
1341
1342            First: no IPv4 options.
1343          */
1344         newinet->inet_opt = NULL;
1345         newnp->ipv6_ac_list = NULL;
1346         newnp->ipv6_fl_list = NULL;
1347
1348         /* Clone RX bits */
1349         newnp->rxopt.all = np->rxopt.all;
1350
1351         /* Clone pktoptions received with SYN */
1352         newnp->pktoptions = NULL;
1353         if (treq->pktopts != NULL) {
1354                 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1355                 kfree_skb(treq->pktopts);
1356                 treq->pktopts = NULL;
1357                 if (newnp->pktoptions)
1358                         skb_set_owner_r(newnp->pktoptions, newsk);
1359         }
1360         newnp->opt        = NULL;
1361         newnp->mcast_oif  = inet6_iif(skb);
1362         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1363
1364         /* Clone native IPv6 options from listening socket (if any)
1365
1366            Yes, keeping reference count would be much more clever,
1367            but we make one more one thing there: reattach optmem
1368            to newsk.
1369          */
1370         if (opt) {
1371                 newnp->opt = ipv6_dup_options(newsk, opt);
1372                 if (opt != np->opt)
1373                         sock_kfree_s(sk, opt, opt->tot_len);
1374         }
1375
1376         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1377         if (newnp->opt)
1378                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1379                                                      newnp->opt->opt_flen);
1380
1381         tcp_mtup_init(newsk);
1382         tcp_sync_mss(newsk, dst_mtu(dst));
1383         newtp->advmss = dst_metric_advmss(dst);
1384         tcp_initialize_rcv_mss(newsk);
1385         if (tcp_rsk(req)->snt_synack)
1386                 tcp_valid_rtt_meas(newsk,
1387                     tcp_time_stamp - tcp_rsk(req)->snt_synack);
1388         newtp->total_retrans = req->retrans;
1389
1390         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1391         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1392
1393 #ifdef CONFIG_TCP_MD5SIG
1394         /* Copy over the MD5 key from the original socket */
1395         if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1396                 /* We're using one, so create a matching key
1397                  * on the newsk structure. If we fail to get
1398                  * memory, then we end up not copying the key
1399                  * across. Shucks.
1400                  */
1401                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newnp->daddr,
1402                                AF_INET6, key->key, key->keylen, GFP_ATOMIC);
1403         }
1404 #endif
1405
1406         if (__inet_inherit_port(sk, newsk) < 0) {
1407                 sock_put(newsk);
1408                 goto out;
1409         }
1410         __inet6_hash(newsk, NULL);
1411
1412         return newsk;
1413
1414 out_overflow:
1415         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1416 out_nonewsk:
1417         if (opt && opt != np->opt)
1418                 sock_kfree_s(sk, opt, opt->tot_len);
1419         dst_release(dst);
1420 out:
1421         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1422         return NULL;
1423 }
1424
1425 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1426 {
1427         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1428                 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1429                                   &ipv6_hdr(skb)->daddr, skb->csum)) {
1430                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1431                         return 0;
1432                 }
1433         }
1434
1435         skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1436                                               &ipv6_hdr(skb)->saddr,
1437                                               &ipv6_hdr(skb)->daddr, 0));
1438
1439         if (skb->len <= 76) {
1440                 return __skb_checksum_complete(skb);
1441         }
1442         return 0;
1443 }
1444
1445 /* The socket must have it's spinlock held when we get
1446  * here.
1447  *
1448  * We have a potential double-lock case here, so even when
1449  * doing backlog processing we use the BH locking scheme.
1450  * This is because we cannot sleep with the original spinlock
1451  * held.
1452  */
1453 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1454 {
1455         struct ipv6_pinfo *np = inet6_sk(sk);
1456         struct tcp_sock *tp;
1457         struct sk_buff *opt_skb = NULL;
1458
1459         /* Imagine: socket is IPv6. IPv4 packet arrives,
1460            goes to IPv4 receive handler and backlogged.
1461            From backlog it always goes here. Kerboom...
1462            Fortunately, tcp_rcv_established and rcv_established
1463            handle them correctly, but it is not case with
1464            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1465          */
1466
1467         if (skb->protocol == htons(ETH_P_IP))
1468                 return tcp_v4_do_rcv(sk, skb);
1469
1470 #ifdef CONFIG_TCP_MD5SIG
1471         if (tcp_v6_inbound_md5_hash (sk, skb))
1472                 goto discard;
1473 #endif
1474
1475         if (sk_filter(sk, skb))
1476                 goto discard;
1477
1478         /*
1479          *      socket locking is here for SMP purposes as backlog rcv
1480          *      is currently called with bh processing disabled.
1481          */
1482
1483         /* Do Stevens' IPV6_PKTOPTIONS.
1484
1485            Yes, guys, it is the only place in our code, where we
1486            may make it not affecting IPv4.
1487            The rest of code is protocol independent,
1488            and I do not like idea to uglify IPv4.
1489
1490            Actually, all the idea behind IPV6_PKTOPTIONS
1491            looks not very well thought. For now we latch
1492            options, received in the last packet, enqueued
1493            by tcp. Feel free to propose better solution.
1494                                                --ANK (980728)
1495          */
1496         if (np->rxopt.all)
1497                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1498
1499         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1500                 sock_rps_save_rxhash(sk, skb);
1501                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1502                         goto reset;
1503                 if (opt_skb)
1504                         goto ipv6_pktoptions;
1505                 return 0;
1506         }
1507
1508         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1509                 goto csum_err;
1510
1511         if (sk->sk_state == TCP_LISTEN) {
1512                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1513                 if (!nsk)
1514                         goto discard;
1515
1516                 /*
1517                  * Queue it on the new socket if the new socket is active,
1518                  * otherwise we just shortcircuit this and continue with
1519                  * the new socket..
1520                  */
1521                 if(nsk != sk) {
1522                         sock_rps_save_rxhash(nsk, skb);
1523                         if (tcp_child_process(sk, nsk, skb))
1524                                 goto reset;
1525                         if (opt_skb)
1526                                 __kfree_skb(opt_skb);
1527                         return 0;
1528                 }
1529         } else
1530                 sock_rps_save_rxhash(sk, skb);
1531
1532         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1533                 goto reset;
1534         if (opt_skb)
1535                 goto ipv6_pktoptions;
1536         return 0;
1537
1538 reset:
1539         tcp_v6_send_reset(sk, skb);
1540 discard:
1541         if (opt_skb)
1542                 __kfree_skb(opt_skb);
1543         kfree_skb(skb);
1544         return 0;
1545 csum_err:
1546         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1547         goto discard;
1548
1549
1550 ipv6_pktoptions:
1551         /* Do you ask, what is it?
1552
1553            1. skb was enqueued by tcp.
1554            2. skb is added to tail of read queue, rather than out of order.
1555            3. socket is not in passive state.
1556            4. Finally, it really contains options, which user wants to receive.
1557          */
1558         tp = tcp_sk(sk);
1559         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1560             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1561                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1562                         np->mcast_oif = inet6_iif(opt_skb);
1563                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1564                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1565                 if (ipv6_opt_accepted(sk, opt_skb)) {
1566                         skb_set_owner_r(opt_skb, sk);
1567                         opt_skb = xchg(&np->pktoptions, opt_skb);
1568                 } else {
1569                         __kfree_skb(opt_skb);
1570                         opt_skb = xchg(&np->pktoptions, NULL);
1571                 }
1572         }
1573
1574         kfree_skb(opt_skb);
1575         return 0;
1576 }
1577
1578 static int tcp_v6_rcv(struct sk_buff *skb)
1579 {
1580         const struct tcphdr *th;
1581         const struct ipv6hdr *hdr;
1582         struct sock *sk;
1583         int ret;
1584         struct net *net = dev_net(skb->dev);
1585
1586         if (skb->pkt_type != PACKET_HOST)
1587                 goto discard_it;
1588
1589         /*
1590          *      Count it even if it's bad.
1591          */
1592         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1593
1594         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1595                 goto discard_it;
1596
1597         th = tcp_hdr(skb);
1598
1599         if (th->doff < sizeof(struct tcphdr)/4)
1600                 goto bad_packet;
1601         if (!pskb_may_pull(skb, th->doff*4))
1602                 goto discard_it;
1603
1604         if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1605                 goto bad_packet;
1606
1607         th = tcp_hdr(skb);
1608         hdr = ipv6_hdr(skb);
1609         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1610         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1611                                     skb->len - th->doff*4);
1612         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1613         TCP_SKB_CB(skb)->when = 0;
1614         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1615         TCP_SKB_CB(skb)->sacked = 0;
1616
1617         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1618         if (!sk)
1619                 goto no_tcp_socket;
1620
1621 process:
1622         if (sk->sk_state == TCP_TIME_WAIT)
1623                 goto do_time_wait;
1624
1625         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1626                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1627                 goto discard_and_relse;
1628         }
1629
1630         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1631                 goto discard_and_relse;
1632
1633         if (sk_filter(sk, skb))
1634                 goto discard_and_relse;
1635
1636         skb->dev = NULL;
1637
1638         bh_lock_sock_nested(sk);
1639         ret = 0;
1640         if (!sock_owned_by_user(sk)) {
1641 #ifdef CONFIG_NET_DMA
1642                 struct tcp_sock *tp = tcp_sk(sk);
1643                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1644                         tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1645                 if (tp->ucopy.dma_chan)
1646                         ret = tcp_v6_do_rcv(sk, skb);
1647                 else
1648 #endif
1649                 {
1650                         if (!tcp_prequeue(sk, skb))
1651                                 ret = tcp_v6_do_rcv(sk, skb);
1652                 }
1653         } else if (unlikely(sk_add_backlog(sk, skb))) {
1654                 bh_unlock_sock(sk);
1655                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1656                 goto discard_and_relse;
1657         }
1658         bh_unlock_sock(sk);
1659
1660         sock_put(sk);
1661         return ret ? -1 : 0;
1662
1663 no_tcp_socket:
1664         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1665                 goto discard_it;
1666
1667         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1668 bad_packet:
1669                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1670         } else {
1671                 tcp_v6_send_reset(NULL, skb);
1672         }
1673
1674 discard_it:
1675
1676         /*
1677          *      Discard frame
1678          */
1679
1680         kfree_skb(skb);
1681         return 0;
1682
1683 discard_and_relse:
1684         sock_put(sk);
1685         goto discard_it;
1686
1687 do_time_wait:
1688         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1689                 inet_twsk_put(inet_twsk(sk));
1690                 goto discard_it;
1691         }
1692
1693         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1694                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1695                 inet_twsk_put(inet_twsk(sk));
1696                 goto discard_it;
1697         }
1698
1699         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1700         case TCP_TW_SYN:
1701         {
1702                 struct sock *sk2;
1703
1704                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1705                                             &ipv6_hdr(skb)->daddr,
1706                                             ntohs(th->dest), inet6_iif(skb));
1707                 if (sk2 != NULL) {
1708                         struct inet_timewait_sock *tw = inet_twsk(sk);
1709                         inet_twsk_deschedule(tw, &tcp_death_row);
1710                         inet_twsk_put(tw);
1711                         sk = sk2;
1712                         goto process;
1713                 }
1714                 /* Fall through to ACK */
1715         }
1716         case TCP_TW_ACK:
1717                 tcp_v6_timewait_ack(sk, skb);
1718                 break;
1719         case TCP_TW_RST:
1720                 goto no_tcp_socket;
1721         case TCP_TW_SUCCESS:;
1722         }
1723         goto discard_it;
1724 }
1725
1726 static struct inet_peer *tcp_v6_get_peer(struct sock *sk, bool *release_it)
1727 {
1728         struct rt6_info *rt = (struct rt6_info *) __sk_dst_get(sk);
1729         struct ipv6_pinfo *np = inet6_sk(sk);
1730         struct inet_peer *peer;
1731
1732         if (!rt ||
1733             !ipv6_addr_equal(&np->daddr, &rt->rt6i_dst.addr)) {
1734                 peer = inet_getpeer_v6(&np->daddr, 1);
1735                 *release_it = true;
1736         } else {
1737                 if (!rt->rt6i_peer)
1738                         rt6_bind_peer(rt, 1);
1739                 peer = rt->rt6i_peer;
1740                 *release_it = false;
1741         }
1742
1743         return peer;
1744 }
1745
1746 static void *tcp_v6_tw_get_peer(struct sock *sk)
1747 {
1748         const struct inet6_timewait_sock *tw6 = inet6_twsk(sk);
1749         const struct inet_timewait_sock *tw = inet_twsk(sk);
1750
1751         if (tw->tw_family == AF_INET)
1752                 return tcp_v4_tw_get_peer(sk);
1753
1754         return inet_getpeer_v6(&tw6->tw_v6_daddr, 1);
1755 }
1756
1757 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1758         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1759         .twsk_unique    = tcp_twsk_unique,
1760         .twsk_destructor= tcp_twsk_destructor,
1761         .twsk_getpeer   = tcp_v6_tw_get_peer,
1762 };
1763
1764 static const struct inet_connection_sock_af_ops ipv6_specific = {
1765         .queue_xmit        = inet6_csk_xmit,
1766         .send_check        = tcp_v6_send_check,
1767         .rebuild_header    = inet6_sk_rebuild_header,
1768         .conn_request      = tcp_v6_conn_request,
1769         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1770         .get_peer          = tcp_v6_get_peer,
1771         .net_header_len    = sizeof(struct ipv6hdr),
1772         .setsockopt        = ipv6_setsockopt,
1773         .getsockopt        = ipv6_getsockopt,
1774         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1775         .sockaddr_len      = sizeof(struct sockaddr_in6),
1776         .bind_conflict     = inet6_csk_bind_conflict,
1777 #ifdef CONFIG_COMPAT
1778         .compat_setsockopt = compat_ipv6_setsockopt,
1779         .compat_getsockopt = compat_ipv6_getsockopt,
1780 #endif
1781 };
1782
1783 #ifdef CONFIG_TCP_MD5SIG
1784 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1785         .md5_lookup     =       tcp_v6_md5_lookup,
1786         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1787         .md5_parse      =       tcp_v6_parse_md5_keys,
1788 };
1789 #endif
1790
1791 /*
1792  *      TCP over IPv4 via INET6 API
1793  */
1794
1795 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1796         .queue_xmit        = ip_queue_xmit,
1797         .send_check        = tcp_v4_send_check,
1798         .rebuild_header    = inet_sk_rebuild_header,
1799         .conn_request      = tcp_v6_conn_request,
1800         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1801         .get_peer          = tcp_v4_get_peer,
1802         .net_header_len    = sizeof(struct iphdr),
1803         .setsockopt        = ipv6_setsockopt,
1804         .getsockopt        = ipv6_getsockopt,
1805         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1806         .sockaddr_len      = sizeof(struct sockaddr_in6),
1807         .bind_conflict     = inet6_csk_bind_conflict,
1808 #ifdef CONFIG_COMPAT
1809         .compat_setsockopt = compat_ipv6_setsockopt,
1810         .compat_getsockopt = compat_ipv6_getsockopt,
1811 #endif
1812 };
1813
1814 #ifdef CONFIG_TCP_MD5SIG
1815 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1816         .md5_lookup     =       tcp_v4_md5_lookup,
1817         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1818         .md5_parse      =       tcp_v6_parse_md5_keys,
1819 };
1820 #endif
1821
1822 /* NOTE: A lot of things set to zero explicitly by call to
1823  *       sk_alloc() so need not be done here.
1824  */
1825 static int tcp_v6_init_sock(struct sock *sk)
1826 {
1827         struct inet_connection_sock *icsk = inet_csk(sk);
1828         struct tcp_sock *tp = tcp_sk(sk);
1829
1830         skb_queue_head_init(&tp->out_of_order_queue);
1831         tcp_init_xmit_timers(sk);
1832         tcp_prequeue_init(tp);
1833
1834         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1835         tp->mdev = TCP_TIMEOUT_INIT;
1836
1837         /* So many TCP implementations out there (incorrectly) count the
1838          * initial SYN frame in their delayed-ACK and congestion control
1839          * algorithms that we must have the following bandaid to talk
1840          * efficiently to them.  -DaveM
1841          */
1842         tp->snd_cwnd = 2;
1843
1844         /* See draft-stevens-tcpca-spec-01 for discussion of the
1845          * initialization of these values.
1846          */
1847         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1848         tp->snd_cwnd_clamp = ~0;
1849         tp->mss_cache = TCP_MSS_DEFAULT;
1850
1851         tp->reordering = sysctl_tcp_reordering;
1852
1853         sk->sk_state = TCP_CLOSE;
1854
1855         icsk->icsk_af_ops = &ipv6_specific;
1856         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1857         icsk->icsk_sync_mss = tcp_sync_mss;
1858         sk->sk_write_space = sk_stream_write_space;
1859         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1860
1861 #ifdef CONFIG_TCP_MD5SIG
1862         tp->af_specific = &tcp_sock_ipv6_specific;
1863 #endif
1864
1865         /* TCP Cookie Transactions */
1866         if (sysctl_tcp_cookie_size > 0) {
1867                 /* Default, cookies without s_data_payload. */
1868                 tp->cookie_values =
1869                         kzalloc(sizeof(*tp->cookie_values),
1870                                 sk->sk_allocation);
1871                 if (tp->cookie_values != NULL)
1872                         kref_init(&tp->cookie_values->kref);
1873         }
1874         /* Presumed zeroed, in order of appearance:
1875          *      cookie_in_always, cookie_out_never,
1876          *      s_data_constant, s_data_in, s_data_out
1877          */
1878         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1879         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1880
1881         local_bh_disable();
1882         sock_update_memcg(sk);
1883         sk_sockets_allocated_inc(sk);
1884         local_bh_enable();
1885
1886         return 0;
1887 }
1888
1889 static void tcp_v6_destroy_sock(struct sock *sk)
1890 {
1891         tcp_v4_destroy_sock(sk);
1892         inet6_destroy_sock(sk);
1893 }
1894
1895 #ifdef CONFIG_PROC_FS
1896 /* Proc filesystem TCPv6 sock list dumping. */
1897 static void get_openreq6(struct seq_file *seq,
1898                          const struct sock *sk, struct request_sock *req, int i, int uid)
1899 {
1900         int ttd = req->expires - jiffies;
1901         const struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1902         const struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1903
1904         if (ttd < 0)
1905                 ttd = 0;
1906
1907         seq_printf(seq,
1908                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1909                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1910                    i,
1911                    src->s6_addr32[0], src->s6_addr32[1],
1912                    src->s6_addr32[2], src->s6_addr32[3],
1913                    ntohs(inet_rsk(req)->loc_port),
1914                    dest->s6_addr32[0], dest->s6_addr32[1],
1915                    dest->s6_addr32[2], dest->s6_addr32[3],
1916                    ntohs(inet_rsk(req)->rmt_port),
1917                    TCP_SYN_RECV,
1918                    0,0, /* could print option size, but that is af dependent. */
1919                    1,   /* timers active (only the expire timer) */
1920                    jiffies_to_clock_t(ttd),
1921                    req->retrans,
1922                    uid,
1923                    0,  /* non standard timer */
1924                    0, /* open_requests have no inode */
1925                    0, req);
1926 }
1927
1928 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1929 {
1930         const struct in6_addr *dest, *src;
1931         __u16 destp, srcp;
1932         int timer_active;
1933         unsigned long timer_expires;
1934         const struct inet_sock *inet = inet_sk(sp);
1935         const struct tcp_sock *tp = tcp_sk(sp);
1936         const struct inet_connection_sock *icsk = inet_csk(sp);
1937         const struct ipv6_pinfo *np = inet6_sk(sp);
1938
1939         dest  = &np->daddr;
1940         src   = &np->rcv_saddr;
1941         destp = ntohs(inet->inet_dport);
1942         srcp  = ntohs(inet->inet_sport);
1943
1944         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1945                 timer_active    = 1;
1946                 timer_expires   = icsk->icsk_timeout;
1947         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1948                 timer_active    = 4;
1949                 timer_expires   = icsk->icsk_timeout;
1950         } else if (timer_pending(&sp->sk_timer)) {
1951                 timer_active    = 2;
1952                 timer_expires   = sp->sk_timer.expires;
1953         } else {
1954                 timer_active    = 0;
1955                 timer_expires = jiffies;
1956         }
1957
1958         seq_printf(seq,
1959                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1960                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %lu %lu %u %u %d\n",
1961                    i,
1962                    src->s6_addr32[0], src->s6_addr32[1],
1963                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1964                    dest->s6_addr32[0], dest->s6_addr32[1],
1965                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1966                    sp->sk_state,
1967                    tp->write_seq-tp->snd_una,
1968                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1969                    timer_active,
1970                    jiffies_to_clock_t(timer_expires - jiffies),
1971                    icsk->icsk_retransmits,
1972                    sock_i_uid(sp),
1973                    icsk->icsk_probes_out,
1974                    sock_i_ino(sp),
1975                    atomic_read(&sp->sk_refcnt), sp,
1976                    jiffies_to_clock_t(icsk->icsk_rto),
1977                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1978                    (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1979                    tp->snd_cwnd,
1980                    tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
1981                    );
1982 }
1983
1984 static void get_timewait6_sock(struct seq_file *seq,
1985                                struct inet_timewait_sock *tw, int i)
1986 {
1987         const struct in6_addr *dest, *src;
1988         __u16 destp, srcp;
1989         const struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
1990         int ttd = tw->tw_ttd - jiffies;
1991
1992         if (ttd < 0)
1993                 ttd = 0;
1994
1995         dest = &tw6->tw_v6_daddr;
1996         src  = &tw6->tw_v6_rcv_saddr;
1997         destp = ntohs(tw->tw_dport);
1998         srcp  = ntohs(tw->tw_sport);
1999
2000         seq_printf(seq,
2001                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2002                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2003                    i,
2004                    src->s6_addr32[0], src->s6_addr32[1],
2005                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2006                    dest->s6_addr32[0], dest->s6_addr32[1],
2007                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2008                    tw->tw_substate, 0, 0,
2009                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2010                    atomic_read(&tw->tw_refcnt), tw);
2011 }
2012
2013 static int tcp6_seq_show(struct seq_file *seq, void *v)
2014 {
2015         struct tcp_iter_state *st;
2016
2017         if (v == SEQ_START_TOKEN) {
2018                 seq_puts(seq,
2019                          "  sl  "
2020                          "local_address                         "
2021                          "remote_address                        "
2022                          "st tx_queue rx_queue tr tm->when retrnsmt"
2023                          "   uid  timeout inode\n");
2024                 goto out;
2025         }
2026         st = seq->private;
2027
2028         switch (st->state) {
2029         case TCP_SEQ_STATE_LISTENING:
2030         case TCP_SEQ_STATE_ESTABLISHED:
2031                 get_tcp6_sock(seq, v, st->num);
2032                 break;
2033         case TCP_SEQ_STATE_OPENREQ:
2034                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2035                 break;
2036         case TCP_SEQ_STATE_TIME_WAIT:
2037                 get_timewait6_sock(seq, v, st->num);
2038                 break;
2039         }
2040 out:
2041         return 0;
2042 }
2043
2044 static const struct file_operations tcp6_afinfo_seq_fops = {
2045         .owner   = THIS_MODULE,
2046         .open    = tcp_seq_open,
2047         .read    = seq_read,
2048         .llseek  = seq_lseek,
2049         .release = seq_release_net
2050 };
2051
2052 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2053         .name           = "tcp6",
2054         .family         = AF_INET6,
2055         .seq_fops       = &tcp6_afinfo_seq_fops,
2056         .seq_ops        = {
2057                 .show           = tcp6_seq_show,
2058         },
2059 };
2060
2061 int __net_init tcp6_proc_init(struct net *net)
2062 {
2063         return tcp_proc_register(net, &tcp6_seq_afinfo);
2064 }
2065
2066 void tcp6_proc_exit(struct net *net)
2067 {
2068         tcp_proc_unregister(net, &tcp6_seq_afinfo);
2069 }
2070 #endif
2071
2072 struct proto tcpv6_prot = {
2073         .name                   = "TCPv6",
2074         .owner                  = THIS_MODULE,
2075         .close                  = tcp_close,
2076         .connect                = tcp_v6_connect,
2077         .disconnect             = tcp_disconnect,
2078         .accept                 = inet_csk_accept,
2079         .ioctl                  = tcp_ioctl,
2080         .init                   = tcp_v6_init_sock,
2081         .destroy                = tcp_v6_destroy_sock,
2082         .shutdown               = tcp_shutdown,
2083         .setsockopt             = tcp_setsockopt,
2084         .getsockopt             = tcp_getsockopt,
2085         .recvmsg                = tcp_recvmsg,
2086         .sendmsg                = tcp_sendmsg,
2087         .sendpage               = tcp_sendpage,
2088         .backlog_rcv            = tcp_v6_do_rcv,
2089         .hash                   = tcp_v6_hash,
2090         .unhash                 = inet_unhash,
2091         .get_port               = inet_csk_get_port,
2092         .enter_memory_pressure  = tcp_enter_memory_pressure,
2093         .sockets_allocated      = &tcp_sockets_allocated,
2094         .memory_allocated       = &tcp_memory_allocated,
2095         .memory_pressure        = &tcp_memory_pressure,
2096         .orphan_count           = &tcp_orphan_count,
2097         .sysctl_wmem            = sysctl_tcp_wmem,
2098         .sysctl_rmem            = sysctl_tcp_rmem,
2099         .max_header             = MAX_TCP_HEADER,
2100         .obj_size               = sizeof(struct tcp6_sock),
2101         .slab_flags             = SLAB_DESTROY_BY_RCU,
2102         .twsk_prot              = &tcp6_timewait_sock_ops,
2103         .rsk_prot               = &tcp6_request_sock_ops,
2104         .h.hashinfo             = &tcp_hashinfo,
2105         .no_autobind            = true,
2106 #ifdef CONFIG_COMPAT
2107         .compat_setsockopt      = compat_tcp_setsockopt,
2108         .compat_getsockopt      = compat_tcp_getsockopt,
2109 #endif
2110 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
2111         .proto_cgroup           = tcp_proto_cgroup,
2112 #endif
2113 };
2114
2115 static const struct inet6_protocol tcpv6_protocol = {
2116         .handler        =       tcp_v6_rcv,
2117         .err_handler    =       tcp_v6_err,
2118         .gso_send_check =       tcp_v6_gso_send_check,
2119         .gso_segment    =       tcp_tso_segment,
2120         .gro_receive    =       tcp6_gro_receive,
2121         .gro_complete   =       tcp6_gro_complete,
2122         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2123 };
2124
2125 static struct inet_protosw tcpv6_protosw = {
2126         .type           =       SOCK_STREAM,
2127         .protocol       =       IPPROTO_TCP,
2128         .prot           =       &tcpv6_prot,
2129         .ops            =       &inet6_stream_ops,
2130         .no_check       =       0,
2131         .flags          =       INET_PROTOSW_PERMANENT |
2132                                 INET_PROTOSW_ICSK,
2133 };
2134
2135 static int __net_init tcpv6_net_init(struct net *net)
2136 {
2137         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2138                                     SOCK_RAW, IPPROTO_TCP, net);
2139 }
2140
2141 static void __net_exit tcpv6_net_exit(struct net *net)
2142 {
2143         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2144 }
2145
2146 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2147 {
2148         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2149 }
2150
2151 static struct pernet_operations tcpv6_net_ops = {
2152         .init       = tcpv6_net_init,
2153         .exit       = tcpv6_net_exit,
2154         .exit_batch = tcpv6_net_exit_batch,
2155 };
2156
2157 int __init tcpv6_init(void)
2158 {
2159         int ret;
2160
2161         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2162         if (ret)
2163                 goto out;
2164
2165         /* register inet6 protocol */
2166         ret = inet6_register_protosw(&tcpv6_protosw);
2167         if (ret)
2168                 goto out_tcpv6_protocol;
2169
2170         ret = register_pernet_subsys(&tcpv6_net_ops);
2171         if (ret)
2172                 goto out_tcpv6_protosw;
2173 out:
2174         return ret;
2175
2176 out_tcpv6_protocol:
2177         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2178 out_tcpv6_protosw:
2179         inet6_unregister_protosw(&tcpv6_protosw);
2180         goto out;
2181 }
2182
2183 void tcpv6_exit(void)
2184 {
2185         unregister_pernet_subsys(&tcpv6_net_ops);
2186         inet6_unregister_protosw(&tcpv6_protosw);
2187         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2188 }