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                 newnp->rcv_tclass  = ipv6_tclass(ipv6_hdr(skb));
1286
1287                 /*
1288                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1289                  * here, tcp_create_openreq_child now does this for us, see the comment in
1290                  * that function for the gory details. -acme
1291                  */
1292
1293                 /* It is tricky place. Until this moment IPv4 tcp
1294                    worked with IPv6 icsk.icsk_af_ops.
1295                    Sync it now.
1296                  */
1297                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1298
1299                 return newsk;
1300         }
1301
1302         treq = inet6_rsk(req);
1303         opt = np->opt;
1304
1305         if (sk_acceptq_is_full(sk))
1306                 goto out_overflow;
1307
1308         if (!dst) {
1309                 dst = inet6_csk_route_req(sk, req);
1310                 if (!dst)
1311                         goto out;
1312         }
1313
1314         newsk = tcp_create_openreq_child(sk, req, skb);
1315         if (newsk == NULL)
1316                 goto out_nonewsk;
1317
1318         /*
1319          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1320          * count here, tcp_create_openreq_child now does this for us, see the
1321          * comment in that function for the gory details. -acme
1322          */
1323
1324         newsk->sk_gso_type = SKB_GSO_TCPV6;
1325         __ip6_dst_store(newsk, dst, NULL, NULL);
1326
1327         newtcp6sk = (struct tcp6_sock *)newsk;
1328         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1329
1330         newtp = tcp_sk(newsk);
1331         newinet = inet_sk(newsk);
1332         newnp = inet6_sk(newsk);
1333
1334         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1335
1336         newnp->daddr = treq->rmt_addr;
1337         newnp->saddr = treq->loc_addr;
1338         newnp->rcv_saddr = treq->loc_addr;
1339         newsk->sk_bound_dev_if = treq->iif;
1340
1341         /* Now IPv6 options...
1342
1343            First: no IPv4 options.
1344          */
1345         newinet->inet_opt = NULL;
1346         newnp->ipv6_ac_list = NULL;
1347         newnp->ipv6_fl_list = NULL;
1348
1349         /* Clone RX bits */
1350         newnp->rxopt.all = np->rxopt.all;
1351
1352         /* Clone pktoptions received with SYN */
1353         newnp->pktoptions = NULL;
1354         if (treq->pktopts != NULL) {
1355                 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1356                 kfree_skb(treq->pktopts);
1357                 treq->pktopts = NULL;
1358                 if (newnp->pktoptions)
1359                         skb_set_owner_r(newnp->pktoptions, newsk);
1360         }
1361         newnp->opt        = NULL;
1362         newnp->mcast_oif  = inet6_iif(skb);
1363         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1364         newnp->rcv_tclass = ipv6_tclass(ipv6_hdr(skb));
1365
1366         /* Clone native IPv6 options from listening socket (if any)
1367
1368            Yes, keeping reference count would be much more clever,
1369            but we make one more one thing there: reattach optmem
1370            to newsk.
1371          */
1372         if (opt) {
1373                 newnp->opt = ipv6_dup_options(newsk, opt);
1374                 if (opt != np->opt)
1375                         sock_kfree_s(sk, opt, opt->tot_len);
1376         }
1377
1378         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1379         if (newnp->opt)
1380                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1381                                                      newnp->opt->opt_flen);
1382
1383         tcp_mtup_init(newsk);
1384         tcp_sync_mss(newsk, dst_mtu(dst));
1385         newtp->advmss = dst_metric_advmss(dst);
1386         tcp_initialize_rcv_mss(newsk);
1387         if (tcp_rsk(req)->snt_synack)
1388                 tcp_valid_rtt_meas(newsk,
1389                     tcp_time_stamp - tcp_rsk(req)->snt_synack);
1390         newtp->total_retrans = req->retrans;
1391
1392         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1393         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1394
1395 #ifdef CONFIG_TCP_MD5SIG
1396         /* Copy over the MD5 key from the original socket */
1397         if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1398                 /* We're using one, so create a matching key
1399                  * on the newsk structure. If we fail to get
1400                  * memory, then we end up not copying the key
1401                  * across. Shucks.
1402                  */
1403                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newnp->daddr,
1404                                AF_INET6, key->key, key->keylen, GFP_ATOMIC);
1405         }
1406 #endif
1407
1408         if (__inet_inherit_port(sk, newsk) < 0) {
1409                 sock_put(newsk);
1410                 goto out;
1411         }
1412         __inet6_hash(newsk, NULL);
1413
1414         return newsk;
1415
1416 out_overflow:
1417         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1418 out_nonewsk:
1419         if (opt && opt != np->opt)
1420                 sock_kfree_s(sk, opt, opt->tot_len);
1421         dst_release(dst);
1422 out:
1423         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1424         return NULL;
1425 }
1426
1427 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1428 {
1429         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1430                 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1431                                   &ipv6_hdr(skb)->daddr, skb->csum)) {
1432                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1433                         return 0;
1434                 }
1435         }
1436
1437         skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1438                                               &ipv6_hdr(skb)->saddr,
1439                                               &ipv6_hdr(skb)->daddr, 0));
1440
1441         if (skb->len <= 76) {
1442                 return __skb_checksum_complete(skb);
1443         }
1444         return 0;
1445 }
1446
1447 /* The socket must have it's spinlock held when we get
1448  * here.
1449  *
1450  * We have a potential double-lock case here, so even when
1451  * doing backlog processing we use the BH locking scheme.
1452  * This is because we cannot sleep with the original spinlock
1453  * held.
1454  */
1455 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1456 {
1457         struct ipv6_pinfo *np = inet6_sk(sk);
1458         struct tcp_sock *tp;
1459         struct sk_buff *opt_skb = NULL;
1460
1461         /* Imagine: socket is IPv6. IPv4 packet arrives,
1462            goes to IPv4 receive handler and backlogged.
1463            From backlog it always goes here. Kerboom...
1464            Fortunately, tcp_rcv_established and rcv_established
1465            handle them correctly, but it is not case with
1466            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1467          */
1468
1469         if (skb->protocol == htons(ETH_P_IP))
1470                 return tcp_v4_do_rcv(sk, skb);
1471
1472 #ifdef CONFIG_TCP_MD5SIG
1473         if (tcp_v6_inbound_md5_hash (sk, skb))
1474                 goto discard;
1475 #endif
1476
1477         if (sk_filter(sk, skb))
1478                 goto discard;
1479
1480         /*
1481          *      socket locking is here for SMP purposes as backlog rcv
1482          *      is currently called with bh processing disabled.
1483          */
1484
1485         /* Do Stevens' IPV6_PKTOPTIONS.
1486
1487            Yes, guys, it is the only place in our code, where we
1488            may make it not affecting IPv4.
1489            The rest of code is protocol independent,
1490            and I do not like idea to uglify IPv4.
1491
1492            Actually, all the idea behind IPV6_PKTOPTIONS
1493            looks not very well thought. For now we latch
1494            options, received in the last packet, enqueued
1495            by tcp. Feel free to propose better solution.
1496                                                --ANK (980728)
1497          */
1498         if (np->rxopt.all)
1499                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1500
1501         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1502                 sock_rps_save_rxhash(sk, skb);
1503                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1504                         goto reset;
1505                 if (opt_skb)
1506                         goto ipv6_pktoptions;
1507                 return 0;
1508         }
1509
1510         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1511                 goto csum_err;
1512
1513         if (sk->sk_state == TCP_LISTEN) {
1514                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1515                 if (!nsk)
1516                         goto discard;
1517
1518                 /*
1519                  * Queue it on the new socket if the new socket is active,
1520                  * otherwise we just shortcircuit this and continue with
1521                  * the new socket..
1522                  */
1523                 if(nsk != sk) {
1524                         sock_rps_save_rxhash(nsk, skb);
1525                         if (tcp_child_process(sk, nsk, skb))
1526                                 goto reset;
1527                         if (opt_skb)
1528                                 __kfree_skb(opt_skb);
1529                         return 0;
1530                 }
1531         } else
1532                 sock_rps_save_rxhash(sk, skb);
1533
1534         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1535                 goto reset;
1536         if (opt_skb)
1537                 goto ipv6_pktoptions;
1538         return 0;
1539
1540 reset:
1541         tcp_v6_send_reset(sk, skb);
1542 discard:
1543         if (opt_skb)
1544                 __kfree_skb(opt_skb);
1545         kfree_skb(skb);
1546         return 0;
1547 csum_err:
1548         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1549         goto discard;
1550
1551
1552 ipv6_pktoptions:
1553         /* Do you ask, what is it?
1554
1555            1. skb was enqueued by tcp.
1556            2. skb is added to tail of read queue, rather than out of order.
1557            3. socket is not in passive state.
1558            4. Finally, it really contains options, which user wants to receive.
1559          */
1560         tp = tcp_sk(sk);
1561         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1562             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1563                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1564                         np->mcast_oif = inet6_iif(opt_skb);
1565                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1566                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1567                 if (np->rxopt.bits.rxtclass)
1568                         np->rcv_tclass = ipv6_tclass(ipv6_hdr(skb));
1569                 if (ipv6_opt_accepted(sk, opt_skb)) {
1570                         skb_set_owner_r(opt_skb, sk);
1571                         opt_skb = xchg(&np->pktoptions, opt_skb);
1572                 } else {
1573                         __kfree_skb(opt_skb);
1574                         opt_skb = xchg(&np->pktoptions, NULL);
1575                 }
1576         }
1577
1578         kfree_skb(opt_skb);
1579         return 0;
1580 }
1581
1582 static int tcp_v6_rcv(struct sk_buff *skb)
1583 {
1584         const struct tcphdr *th;
1585         const struct ipv6hdr *hdr;
1586         struct sock *sk;
1587         int ret;
1588         struct net *net = dev_net(skb->dev);
1589
1590         if (skb->pkt_type != PACKET_HOST)
1591                 goto discard_it;
1592
1593         /*
1594          *      Count it even if it's bad.
1595          */
1596         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1597
1598         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1599                 goto discard_it;
1600
1601         th = tcp_hdr(skb);
1602
1603         if (th->doff < sizeof(struct tcphdr)/4)
1604                 goto bad_packet;
1605         if (!pskb_may_pull(skb, th->doff*4))
1606                 goto discard_it;
1607
1608         if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1609                 goto bad_packet;
1610
1611         th = tcp_hdr(skb);
1612         hdr = ipv6_hdr(skb);
1613         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1614         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1615                                     skb->len - th->doff*4);
1616         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1617         TCP_SKB_CB(skb)->when = 0;
1618         TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1619         TCP_SKB_CB(skb)->sacked = 0;
1620
1621         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1622         if (!sk)
1623                 goto no_tcp_socket;
1624
1625 process:
1626         if (sk->sk_state == TCP_TIME_WAIT)
1627                 goto do_time_wait;
1628
1629         if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1630                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1631                 goto discard_and_relse;
1632         }
1633
1634         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1635                 goto discard_and_relse;
1636
1637         if (sk_filter(sk, skb))
1638                 goto discard_and_relse;
1639
1640         skb->dev = NULL;
1641
1642         bh_lock_sock_nested(sk);
1643         ret = 0;
1644         if (!sock_owned_by_user(sk)) {
1645 #ifdef CONFIG_NET_DMA
1646                 struct tcp_sock *tp = tcp_sk(sk);
1647                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1648                         tp->ucopy.dma_chan = net_dma_find_channel();
1649                 if (tp->ucopy.dma_chan)
1650                         ret = tcp_v6_do_rcv(sk, skb);
1651                 else
1652 #endif
1653                 {
1654                         if (!tcp_prequeue(sk, skb))
1655                                 ret = tcp_v6_do_rcv(sk, skb);
1656                 }
1657         } else if (unlikely(sk_add_backlog(sk, skb))) {
1658                 bh_unlock_sock(sk);
1659                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1660                 goto discard_and_relse;
1661         }
1662         bh_unlock_sock(sk);
1663
1664         sock_put(sk);
1665         return ret ? -1 : 0;
1666
1667 no_tcp_socket:
1668         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1669                 goto discard_it;
1670
1671         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1672 bad_packet:
1673                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1674         } else {
1675                 tcp_v6_send_reset(NULL, skb);
1676         }
1677
1678 discard_it:
1679
1680         /*
1681          *      Discard frame
1682          */
1683
1684         kfree_skb(skb);
1685         return 0;
1686
1687 discard_and_relse:
1688         sock_put(sk);
1689         goto discard_it;
1690
1691 do_time_wait:
1692         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1693                 inet_twsk_put(inet_twsk(sk));
1694                 goto discard_it;
1695         }
1696
1697         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1698                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1699                 inet_twsk_put(inet_twsk(sk));
1700                 goto discard_it;
1701         }
1702
1703         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1704         case TCP_TW_SYN:
1705         {
1706                 struct sock *sk2;
1707
1708                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1709                                             &ipv6_hdr(skb)->daddr,
1710                                             ntohs(th->dest), inet6_iif(skb));
1711                 if (sk2 != NULL) {
1712                         struct inet_timewait_sock *tw = inet_twsk(sk);
1713                         inet_twsk_deschedule(tw, &tcp_death_row);
1714                         inet_twsk_put(tw);
1715                         sk = sk2;
1716                         goto process;
1717                 }
1718                 /* Fall through to ACK */
1719         }
1720         case TCP_TW_ACK:
1721                 tcp_v6_timewait_ack(sk, skb);
1722                 break;
1723         case TCP_TW_RST:
1724                 goto no_tcp_socket;
1725         case TCP_TW_SUCCESS:;
1726         }
1727         goto discard_it;
1728 }
1729
1730 static struct inet_peer *tcp_v6_get_peer(struct sock *sk, bool *release_it)
1731 {
1732         struct rt6_info *rt = (struct rt6_info *) __sk_dst_get(sk);
1733         struct ipv6_pinfo *np = inet6_sk(sk);
1734         struct inet_peer *peer;
1735
1736         if (!rt ||
1737             !ipv6_addr_equal(&np->daddr, &rt->rt6i_dst.addr)) {
1738                 peer = inet_getpeer_v6(&np->daddr, 1);
1739                 *release_it = true;
1740         } else {
1741                 if (!rt->rt6i_peer)
1742                         rt6_bind_peer(rt, 1);
1743                 peer = rt->rt6i_peer;
1744                 *release_it = false;
1745         }
1746
1747         return peer;
1748 }
1749
1750 static void *tcp_v6_tw_get_peer(struct sock *sk)
1751 {
1752         const struct inet6_timewait_sock *tw6 = inet6_twsk(sk);
1753         const struct inet_timewait_sock *tw = inet_twsk(sk);
1754
1755         if (tw->tw_family == AF_INET)
1756                 return tcp_v4_tw_get_peer(sk);
1757
1758         return inet_getpeer_v6(&tw6->tw_v6_daddr, 1);
1759 }
1760
1761 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1762         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
1763         .twsk_unique    = tcp_twsk_unique,
1764         .twsk_destructor= tcp_twsk_destructor,
1765         .twsk_getpeer   = tcp_v6_tw_get_peer,
1766 };
1767
1768 static const struct inet_connection_sock_af_ops ipv6_specific = {
1769         .queue_xmit        = inet6_csk_xmit,
1770         .send_check        = tcp_v6_send_check,
1771         .rebuild_header    = inet6_sk_rebuild_header,
1772         .conn_request      = tcp_v6_conn_request,
1773         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1774         .get_peer          = tcp_v6_get_peer,
1775         .net_header_len    = sizeof(struct ipv6hdr),
1776         .setsockopt        = ipv6_setsockopt,
1777         .getsockopt        = ipv6_getsockopt,
1778         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1779         .sockaddr_len      = sizeof(struct sockaddr_in6),
1780         .bind_conflict     = inet6_csk_bind_conflict,
1781 #ifdef CONFIG_COMPAT
1782         .compat_setsockopt = compat_ipv6_setsockopt,
1783         .compat_getsockopt = compat_ipv6_getsockopt,
1784 #endif
1785 };
1786
1787 #ifdef CONFIG_TCP_MD5SIG
1788 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1789         .md5_lookup     =       tcp_v6_md5_lookup,
1790         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1791         .md5_parse      =       tcp_v6_parse_md5_keys,
1792 };
1793 #endif
1794
1795 /*
1796  *      TCP over IPv4 via INET6 API
1797  */
1798
1799 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1800         .queue_xmit        = ip_queue_xmit,
1801         .send_check        = tcp_v4_send_check,
1802         .rebuild_header    = inet_sk_rebuild_header,
1803         .conn_request      = tcp_v6_conn_request,
1804         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1805         .get_peer          = tcp_v4_get_peer,
1806         .net_header_len    = sizeof(struct iphdr),
1807         .setsockopt        = ipv6_setsockopt,
1808         .getsockopt        = ipv6_getsockopt,
1809         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1810         .sockaddr_len      = sizeof(struct sockaddr_in6),
1811         .bind_conflict     = inet6_csk_bind_conflict,
1812 #ifdef CONFIG_COMPAT
1813         .compat_setsockopt = compat_ipv6_setsockopt,
1814         .compat_getsockopt = compat_ipv6_getsockopt,
1815 #endif
1816 };
1817
1818 #ifdef CONFIG_TCP_MD5SIG
1819 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1820         .md5_lookup     =       tcp_v4_md5_lookup,
1821         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1822         .md5_parse      =       tcp_v6_parse_md5_keys,
1823 };
1824 #endif
1825
1826 /* NOTE: A lot of things set to zero explicitly by call to
1827  *       sk_alloc() so need not be done here.
1828  */
1829 static int tcp_v6_init_sock(struct sock *sk)
1830 {
1831         struct inet_connection_sock *icsk = inet_csk(sk);
1832         struct tcp_sock *tp = tcp_sk(sk);
1833
1834         skb_queue_head_init(&tp->out_of_order_queue);
1835         tcp_init_xmit_timers(sk);
1836         tcp_prequeue_init(tp);
1837
1838         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1839         tp->mdev = TCP_TIMEOUT_INIT;
1840
1841         /* So many TCP implementations out there (incorrectly) count the
1842          * initial SYN frame in their delayed-ACK and congestion control
1843          * algorithms that we must have the following bandaid to talk
1844          * efficiently to them.  -DaveM
1845          */
1846         tp->snd_cwnd = 2;
1847
1848         /* See draft-stevens-tcpca-spec-01 for discussion of the
1849          * initialization of these values.
1850          */
1851         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1852         tp->snd_cwnd_clamp = ~0;
1853         tp->mss_cache = TCP_MSS_DEFAULT;
1854
1855         tp->reordering = sysctl_tcp_reordering;
1856
1857         sk->sk_state = TCP_CLOSE;
1858
1859         icsk->icsk_af_ops = &ipv6_specific;
1860         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1861         icsk->icsk_sync_mss = tcp_sync_mss;
1862         sk->sk_write_space = sk_stream_write_space;
1863         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1864
1865 #ifdef CONFIG_TCP_MD5SIG
1866         tp->af_specific = &tcp_sock_ipv6_specific;
1867 #endif
1868
1869         /* TCP Cookie Transactions */
1870         if (sysctl_tcp_cookie_size > 0) {
1871                 /* Default, cookies without s_data_payload. */
1872                 tp->cookie_values =
1873                         kzalloc(sizeof(*tp->cookie_values),
1874                                 sk->sk_allocation);
1875                 if (tp->cookie_values != NULL)
1876                         kref_init(&tp->cookie_values->kref);
1877         }
1878         /* Presumed zeroed, in order of appearance:
1879          *      cookie_in_always, cookie_out_never,
1880          *      s_data_constant, s_data_in, s_data_out
1881          */
1882         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1883         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1884
1885         local_bh_disable();
1886         sock_update_memcg(sk);
1887         sk_sockets_allocated_inc(sk);
1888         local_bh_enable();
1889
1890         return 0;
1891 }
1892
1893 static void tcp_v6_destroy_sock(struct sock *sk)
1894 {
1895         tcp_v4_destroy_sock(sk);
1896         inet6_destroy_sock(sk);
1897 }
1898
1899 #ifdef CONFIG_PROC_FS
1900 /* Proc filesystem TCPv6 sock list dumping. */
1901 static void get_openreq6(struct seq_file *seq,
1902                          const struct sock *sk, struct request_sock *req, int i, int uid)
1903 {
1904         int ttd = req->expires - jiffies;
1905         const struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1906         const struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1907
1908         if (ttd < 0)
1909                 ttd = 0;
1910
1911         seq_printf(seq,
1912                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1913                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1914                    i,
1915                    src->s6_addr32[0], src->s6_addr32[1],
1916                    src->s6_addr32[2], src->s6_addr32[3],
1917                    ntohs(inet_rsk(req)->loc_port),
1918                    dest->s6_addr32[0], dest->s6_addr32[1],
1919                    dest->s6_addr32[2], dest->s6_addr32[3],
1920                    ntohs(inet_rsk(req)->rmt_port),
1921                    TCP_SYN_RECV,
1922                    0,0, /* could print option size, but that is af dependent. */
1923                    1,   /* timers active (only the expire timer) */
1924                    jiffies_to_clock_t(ttd),
1925                    req->retrans,
1926                    uid,
1927                    0,  /* non standard timer */
1928                    0, /* open_requests have no inode */
1929                    0, req);
1930 }
1931
1932 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1933 {
1934         const struct in6_addr *dest, *src;
1935         __u16 destp, srcp;
1936         int timer_active;
1937         unsigned long timer_expires;
1938         const struct inet_sock *inet = inet_sk(sp);
1939         const struct tcp_sock *tp = tcp_sk(sp);
1940         const struct inet_connection_sock *icsk = inet_csk(sp);
1941         const struct ipv6_pinfo *np = inet6_sk(sp);
1942
1943         dest  = &np->daddr;
1944         src   = &np->rcv_saddr;
1945         destp = ntohs(inet->inet_dport);
1946         srcp  = ntohs(inet->inet_sport);
1947
1948         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1949                 timer_active    = 1;
1950                 timer_expires   = icsk->icsk_timeout;
1951         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1952                 timer_active    = 4;
1953                 timer_expires   = icsk->icsk_timeout;
1954         } else if (timer_pending(&sp->sk_timer)) {
1955                 timer_active    = 2;
1956                 timer_expires   = sp->sk_timer.expires;
1957         } else {
1958                 timer_active    = 0;
1959                 timer_expires = jiffies;
1960         }
1961
1962         seq_printf(seq,
1963                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1964                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %lu %lu %u %u %d\n",
1965                    i,
1966                    src->s6_addr32[0], src->s6_addr32[1],
1967                    src->s6_addr32[2], src->s6_addr32[3], srcp,
1968                    dest->s6_addr32[0], dest->s6_addr32[1],
1969                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
1970                    sp->sk_state,
1971                    tp->write_seq-tp->snd_una,
1972                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1973                    timer_active,
1974                    jiffies_to_clock_t(timer_expires - jiffies),
1975                    icsk->icsk_retransmits,
1976                    sock_i_uid(sp),
1977                    icsk->icsk_probes_out,
1978                    sock_i_ino(sp),
1979                    atomic_read(&sp->sk_refcnt), sp,
1980                    jiffies_to_clock_t(icsk->icsk_rto),
1981                    jiffies_to_clock_t(icsk->icsk_ack.ato),
1982                    (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1983                    tp->snd_cwnd,
1984                    tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
1985                    );
1986 }
1987
1988 static void get_timewait6_sock(struct seq_file *seq,
1989                                struct inet_timewait_sock *tw, int i)
1990 {
1991         const struct in6_addr *dest, *src;
1992         __u16 destp, srcp;
1993         const struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
1994         int ttd = tw->tw_ttd - jiffies;
1995
1996         if (ttd < 0)
1997                 ttd = 0;
1998
1999         dest = &tw6->tw_v6_daddr;
2000         src  = &tw6->tw_v6_rcv_saddr;
2001         destp = ntohs(tw->tw_dport);
2002         srcp  = ntohs(tw->tw_sport);
2003
2004         seq_printf(seq,
2005                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2006                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2007                    i,
2008                    src->s6_addr32[0], src->s6_addr32[1],
2009                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2010                    dest->s6_addr32[0], dest->s6_addr32[1],
2011                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2012                    tw->tw_substate, 0, 0,
2013                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2014                    atomic_read(&tw->tw_refcnt), tw);
2015 }
2016
2017 static int tcp6_seq_show(struct seq_file *seq, void *v)
2018 {
2019         struct tcp_iter_state *st;
2020
2021         if (v == SEQ_START_TOKEN) {
2022                 seq_puts(seq,
2023                          "  sl  "
2024                          "local_address                         "
2025                          "remote_address                        "
2026                          "st tx_queue rx_queue tr tm->when retrnsmt"
2027                          "   uid  timeout inode\n");
2028                 goto out;
2029         }
2030         st = seq->private;
2031
2032         switch (st->state) {
2033         case TCP_SEQ_STATE_LISTENING:
2034         case TCP_SEQ_STATE_ESTABLISHED:
2035                 get_tcp6_sock(seq, v, st->num);
2036                 break;
2037         case TCP_SEQ_STATE_OPENREQ:
2038                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2039                 break;
2040         case TCP_SEQ_STATE_TIME_WAIT:
2041                 get_timewait6_sock(seq, v, st->num);
2042                 break;
2043         }
2044 out:
2045         return 0;
2046 }
2047
2048 static const struct file_operations tcp6_afinfo_seq_fops = {
2049         .owner   = THIS_MODULE,
2050         .open    = tcp_seq_open,
2051         .read    = seq_read,
2052         .llseek  = seq_lseek,
2053         .release = seq_release_net
2054 };
2055
2056 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2057         .name           = "tcp6",
2058         .family         = AF_INET6,
2059         .seq_fops       = &tcp6_afinfo_seq_fops,
2060         .seq_ops        = {
2061                 .show           = tcp6_seq_show,
2062         },
2063 };
2064
2065 int __net_init tcp6_proc_init(struct net *net)
2066 {
2067         return tcp_proc_register(net, &tcp6_seq_afinfo);
2068 }
2069
2070 void tcp6_proc_exit(struct net *net)
2071 {
2072         tcp_proc_unregister(net, &tcp6_seq_afinfo);
2073 }
2074 #endif
2075
2076 struct proto tcpv6_prot = {
2077         .name                   = "TCPv6",
2078         .owner                  = THIS_MODULE,
2079         .close                  = tcp_close,
2080         .connect                = tcp_v6_connect,
2081         .disconnect             = tcp_disconnect,
2082         .accept                 = inet_csk_accept,
2083         .ioctl                  = tcp_ioctl,
2084         .init                   = tcp_v6_init_sock,
2085         .destroy                = tcp_v6_destroy_sock,
2086         .shutdown               = tcp_shutdown,
2087         .setsockopt             = tcp_setsockopt,
2088         .getsockopt             = tcp_getsockopt,
2089         .recvmsg                = tcp_recvmsg,
2090         .sendmsg                = tcp_sendmsg,
2091         .sendpage               = tcp_sendpage,
2092         .backlog_rcv            = tcp_v6_do_rcv,
2093         .hash                   = tcp_v6_hash,
2094         .unhash                 = inet_unhash,
2095         .get_port               = inet_csk_get_port,
2096         .enter_memory_pressure  = tcp_enter_memory_pressure,
2097         .sockets_allocated      = &tcp_sockets_allocated,
2098         .memory_allocated       = &tcp_memory_allocated,
2099         .memory_pressure        = &tcp_memory_pressure,
2100         .orphan_count           = &tcp_orphan_count,
2101         .sysctl_wmem            = sysctl_tcp_wmem,
2102         .sysctl_rmem            = sysctl_tcp_rmem,
2103         .max_header             = MAX_TCP_HEADER,
2104         .obj_size               = sizeof(struct tcp6_sock),
2105         .slab_flags             = SLAB_DESTROY_BY_RCU,
2106         .twsk_prot              = &tcp6_timewait_sock_ops,
2107         .rsk_prot               = &tcp6_request_sock_ops,
2108         .h.hashinfo             = &tcp_hashinfo,
2109         .no_autobind            = true,
2110 #ifdef CONFIG_COMPAT
2111         .compat_setsockopt      = compat_tcp_setsockopt,
2112         .compat_getsockopt      = compat_tcp_getsockopt,
2113 #endif
2114 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
2115         .proto_cgroup           = tcp_proto_cgroup,
2116 #endif
2117 };
2118
2119 static const struct inet6_protocol tcpv6_protocol = {
2120         .handler        =       tcp_v6_rcv,
2121         .err_handler    =       tcp_v6_err,
2122         .gso_send_check =       tcp_v6_gso_send_check,
2123         .gso_segment    =       tcp_tso_segment,
2124         .gro_receive    =       tcp6_gro_receive,
2125         .gro_complete   =       tcp6_gro_complete,
2126         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2127 };
2128
2129 static struct inet_protosw tcpv6_protosw = {
2130         .type           =       SOCK_STREAM,
2131         .protocol       =       IPPROTO_TCP,
2132         .prot           =       &tcpv6_prot,
2133         .ops            =       &inet6_stream_ops,
2134         .no_check       =       0,
2135         .flags          =       INET_PROTOSW_PERMANENT |
2136                                 INET_PROTOSW_ICSK,
2137 };
2138
2139 static int __net_init tcpv6_net_init(struct net *net)
2140 {
2141         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2142                                     SOCK_RAW, IPPROTO_TCP, net);
2143 }
2144
2145 static void __net_exit tcpv6_net_exit(struct net *net)
2146 {
2147         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2148 }
2149
2150 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2151 {
2152         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2153 }
2154
2155 static struct pernet_operations tcpv6_net_ops = {
2156         .init       = tcpv6_net_init,
2157         .exit       = tcpv6_net_exit,
2158         .exit_batch = tcpv6_net_exit_batch,
2159 };
2160
2161 int __init tcpv6_init(void)
2162 {
2163         int ret;
2164
2165         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2166         if (ret)
2167                 goto out;
2168
2169         /* register inet6 protocol */
2170         ret = inet6_register_protosw(&tcpv6_protosw);
2171         if (ret)
2172                 goto out_tcpv6_protocol;
2173
2174         ret = register_pernet_subsys(&tcpv6_net_ops);
2175         if (ret)
2176                 goto out_tcpv6_protosw;
2177 out:
2178         return ret;
2179
2180 out_tcpv6_protocol:
2181         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2182 out_tcpv6_protosw:
2183         inet6_unregister_protosw(&tcpv6_protosw);
2184         goto out;
2185 }
2186
2187 void tcpv6_exit(void)
2188 {
2189         unregister_pernet_subsys(&tcpv6_net_ops);
2190         inet6_unregister_protosw(&tcpv6_protosw);
2191         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2192 }