SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
  65 #include <asm/uaccess.h>
  66
  67 #include <linux/proc_fs.h>
  68 #include <linux/seq_file.h>
  69
  70 #include <linux/crypto.h>
  71 #include <linux/scatterlist.h>
  72
  73 static void     tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
  74 static void     tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
  75                                       struct request_sock *req);
  76
  77 static int      tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
  78 static void     __tcp_v6_send_check(struct sk_buff *skb,
  79                                     struct in6_addr *saddr,
  80                                     struct in6_addr *daddr);
  81
  82 static const struct inet_connection_sock_af_ops ipv6_mapped;
  83 static const struct inet_connection_sock_af_ops ipv6_specific;
  84 #ifdef CONFIG_TCP_MD5SIG
  85 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
  86 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
  87 #else
  88 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
  89                                                    struct in6_addr *addr)
  90 {
  91         return NULL;
  92 }
  93 #endif
  94
  95 static void tcp_v6_hash(struct sock *sk)
  96 {
  97         if (sk->sk_state != TCP_CLOSE) {
  98                 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
  99                         tcp_prot.hash(sk);
 100                         return;
 101                 }
 102                 local_bh_disable();
 103                 __inet6_hash(sk, NULL);
 104                 local_bh_enable();
 105         }
 106 }
 107
 108 static __inline__ __sum16 tcp_v6_check(int len,
 109                                    struct in6_addr *saddr,
 110                                    struct in6_addr *daddr,
 111                                    __wsum base)
 112 {
 113         return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
 114 }
 115
 116 static __u32 tcp_v6_init_sequence(struct sk_buff *skb)
 117 {
 118         return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
 119                                             ipv6_hdr(skb)->saddr.s6_addr32,
 120                                             tcp_hdr(skb)->dest,
 121                                             tcp_hdr(skb)->source);
 122 }
 123
 124 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 125                           int addr_len)
 126 {
 127         struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
 128         struct inet_sock *inet = inet_sk(sk);
 129         struct inet_connection_sock *icsk = inet_csk(sk);
 130         struct ipv6_pinfo *np = inet6_sk(sk);
 131         struct tcp_sock *tp = tcp_sk(sk);
 132         struct in6_addr *saddr = NULL, *final_p = NULL, final;
 133         struct flowi fl;
 134         struct dst_entry *dst;
 135         int addr_type;
 136         int err;
 137
 138         if (addr_len < SIN6_LEN_RFC2133)
 139                 return -EINVAL;
 140
 141         if (usin->sin6_family != AF_INET6)
 142                 return(-EAFNOSUPPORT);
 143
 144         memset(&fl, 0, sizeof(fl));
 145
 146         if (np->sndflow) {
 147                 fl.fl6_flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 148                 IP6_ECN_flow_init(fl.fl6_flowlabel);
 149                 if (fl.fl6_flowlabel&IPV6_FLOWLABEL_MASK) {
 150                         struct ip6_flowlabel *flowlabel;
 151                         flowlabel = fl6_sock_lookup(sk, fl.fl6_flowlabel);
 152                         if (flowlabel == NULL)
 153                                 return -EINVAL;
 154                         ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
 155                         fl6_sock_release(flowlabel);
 156                 }
 157         }
 158
 159         /*
 160          *      connect() to INADDR_ANY means loopback (BSD'ism).
 161          */
 162
 163         if(ipv6_addr_any(&usin->sin6_addr))
 164                 usin->sin6_addr.s6_addr[15] = 0x1;
 165
 166         addr_type = ipv6_addr_type(&usin->sin6_addr);
 167
 168         if(addr_type & IPV6_ADDR_MULTICAST)
 169                 return -ENETUNREACH;
 170
 171         if (addr_type&IPV6_ADDR_LINKLOCAL) {
 172                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 173                     usin->sin6_scope_id) {
 174                         /* If interface is set while binding, indices
 175                          * must coincide.
 176                          */
 177                         if (sk->sk_bound_dev_if &&
 178                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 179                                 return -EINVAL;
 180
 181                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 182                 }
 183
 184                 /* Connect to link-local address requires an interface */
 185                 if (!sk->sk_bound_dev_if)
 186                         return -EINVAL;
 187         }
 188
 189         if (tp->rx_opt.ts_recent_stamp &&
 190             !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
 191                 tp->rx_opt.ts_recent = 0;
 192                 tp->rx_opt.ts_recent_stamp = 0;
 193                 tp->write_seq = 0;
 194         }
 195
 196         ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
 197         np->flow_label = fl.fl6_flowlabel;
 198
 199         /*
 200          *      TCP over IPv4
 201          */
 202
 203         if (addr_type == IPV6_ADDR_MAPPED) {
 204                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 205                 struct sockaddr_in sin;
 206
 207                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 208
 209                 if (__ipv6_only_sock(sk))
 210                         return -ENETUNREACH;
 211
 212                 sin.sin_family = AF_INET;
 213                 sin.sin_port = usin->sin6_port;
 214                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 215
 216                 icsk->icsk_af_ops = &ipv6_mapped;
 217                 sk->sk_backlog_rcv = tcp_v4_do_rcv;
 218 #ifdef CONFIG_TCP_MD5SIG
 219                 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
 220 #endif
 221
 222                 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 223
 224                 if (err) {
 225                         icsk->icsk_ext_hdr_len = exthdrlen;
 226                         icsk->icsk_af_ops = &ipv6_specific;
 227                         sk->sk_backlog_rcv = tcp_v6_do_rcv;
 228 #ifdef CONFIG_TCP_MD5SIG
 229                         tp->af_specific = &tcp_sock_ipv6_specific;
 230 #endif
 231                         goto failure;
 232                 } else {
 233                         ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
 234                         ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
 235                                                &np->rcv_saddr);
 236                 }
 237
 238                 return err;
 239         }
 240
 241         if (!ipv6_addr_any(&np->rcv_saddr))
 242                 saddr = &np->rcv_saddr;
 243
 244         fl.proto = IPPROTO_TCP;
 245         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 246         ipv6_addr_copy(&fl.fl6_src,
 247                        (saddr ? saddr : &np->saddr));
 248         fl.oif = sk->sk_bound_dev_if;
 249         fl.mark = sk->sk_mark;
 250         fl.fl_ip_dport = usin->sin6_port;
 251         fl.fl_ip_sport = inet->inet_sport;
 252
 253         if (np->opt && np->opt->srcrt) {
 254                 struct rt0_hdr *rt0 = (struct rt0_hdr *)np->opt->srcrt;
 255                 ipv6_addr_copy(&final, &fl.fl6_dst);
 256                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 257                 final_p = &final;
 258         }
 259
 260         security_sk_classify_flow(sk, &fl);
 261
 262         err = ip6_dst_lookup(sk, &dst, &fl);
 263         if (err)
 264                 goto failure;
 265         if (final_p)
 266                 ipv6_addr_copy(&fl.fl6_dst, final_p);
 267
 268         err = __xfrm_lookup(sock_net(sk), &dst, &fl, sk, XFRM_LOOKUP_WAIT);
 269         if (err < 0) {
 270                 if (err == -EREMOTE)
 271                         err = ip6_dst_blackhole(sk, &dst, &fl);
 272                 if (err < 0)
 273                         goto failure;
 274         }
 275
 276         if (saddr == NULL) {
 277                 saddr = &fl.fl6_src;
 278                 ipv6_addr_copy(&np->rcv_saddr, saddr);
 279         }
 280
 281         /* set the source address */
 282         ipv6_addr_copy(&np->saddr, saddr);
 283         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 284
 285         sk->sk_gso_type = SKB_GSO_TCPV6;
 286         __ip6_dst_store(sk, dst, NULL, NULL);
 287
 288         icsk->icsk_ext_hdr_len = 0;
 289         if (np->opt)
 290                 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
 291                                           np->opt->opt_nflen);
 292
 293         tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 294
 295         inet->inet_dport = usin->sin6_port;
 296
 297         tcp_set_state(sk, TCP_SYN_SENT);
 298         err = inet6_hash_connect(&tcp_death_row, sk);
 299         if (err)
 300                 goto late_failure;
 301
 302         if (!tp->write_seq)
 303                 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
 304                                                              np->daddr.s6_addr32,
 305                                                              inet->inet_sport,
 306                                                              inet->inet_dport);
 307
 308         err = tcp_connect(sk);
 309         if (err)
 310                 goto late_failure;
 311
 312         return 0;
 313
 314 late_failure:
 315         tcp_set_state(sk, TCP_CLOSE);
 316         __sk_dst_reset(sk);
 317 failure:
 318         inet->inet_dport = 0;
 319         sk->sk_route_caps = 0;
 320         return err;
 321 }
 322
 323 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 324                 u8 type, u8 code, int offset, __be32 info)
 325 {
 326         struct ipv6hdr *hdr = (struct ipv6hdr*)skb->data;
 327         const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
 328         struct ipv6_pinfo *np;
 329         struct sock *sk;
 330         int err;
 331         struct tcp_sock *tp;
 332         __u32 seq;
 333         struct net *net = dev_net(skb->dev);
 334
 335         sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
 336                         th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
 337
 338         if (sk == NULL) {
 339                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
 340                                    ICMP6_MIB_INERRORS);
 341                 return;
 342         }
 343
 344         if (sk->sk_state == TCP_TIME_WAIT) {
 345                 inet_twsk_put(inet_twsk(sk));
 346                 return;
 347         }
 348
 349         bh_lock_sock(sk);
 350         if (sock_owned_by_user(sk))
 351                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 352
 353         if (sk->sk_state == TCP_CLOSE)
 354                 goto out;
 355
 356         tp = tcp_sk(sk);
 357         seq = ntohl(th->seq);
 358         if (sk->sk_state != TCP_LISTEN &&
 359             !between(seq, tp->snd_una, tp->snd_nxt)) {
 360                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 361                 goto out;
 362         }
 363
 364         np = inet6_sk(sk);
 365
 366         if (type == ICMPV6_PKT_TOOBIG) {
 367                 struct dst_entry *dst = NULL;
 368
 369                 if (sock_owned_by_user(sk))
 370                         goto out;
 371                 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 372                         goto out;
 373
 374                 /* icmp should have updated the destination cache entry */
 375                 dst = __sk_dst_check(sk, np->dst_cookie);
 376
 377                 if (dst == NULL) {
 378                         struct inet_sock *inet = inet_sk(sk);
 379                         struct flowi fl;
 380
 381                         /* BUGGG_FUTURE: Again, it is not clear how
 382                            to handle rthdr case. Ignore this complexity
 383                            for now.
 384                          */
 385                         memset(&fl, 0, sizeof(fl));
 386                         fl.proto = IPPROTO_TCP;
 387                         ipv6_addr_copy(&fl.fl6_dst, &np->daddr);
 388                         ipv6_addr_copy(&fl.fl6_src, &np->saddr);
 389                         fl.oif = sk->sk_bound_dev_if;
 390                         fl.mark = sk->sk_mark;
 391                         fl.fl_ip_dport = inet->inet_dport;
 392                         fl.fl_ip_sport = inet->inet_sport;
 393                         security_skb_classify_flow(skb, &fl);
 394
 395                         if ((err = ip6_dst_lookup(sk, &dst, &fl))) {
 396                                 sk->sk_err_soft = -err;
 397                                 goto out;
 398                         }
 399
 400                         if ((err = xfrm_lookup(net, &dst, &fl, sk, 0)) < 0) {
 401                                 sk->sk_err_soft = -err;
 402                                 goto out;
 403                         }
 404
 405                 } else
 406                         dst_hold(dst);
 407
 408                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
 409                         tcp_sync_mss(sk, dst_mtu(dst));
 410                         tcp_simple_retransmit(sk);
 411                 } /* else let the usual retransmit timer handle it */
 412                 dst_release(dst);
 413                 goto out;
 414         }
 415
 416         icmpv6_err_convert(type, code, &err);
 417
 418         /* Might be for an request_sock */
 419         switch (sk->sk_state) {
 420                 struct request_sock *req, **prev;
 421         case TCP_LISTEN:
 422                 if (sock_owned_by_user(sk))
 423                         goto out;
 424
 425                 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
 426                                            &hdr->saddr, inet6_iif(skb));
 427                 if (!req)
 428                         goto out;
 429
 430                 /* ICMPs are not backlogged, hence we cannot get
 431                  * an established socket here.
 432                  */
 433                 WARN_ON(req->sk != NULL);
 434
 435                 if (seq != tcp_rsk(req)->snt_isn) {
 436                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 437                         goto out;
 438                 }
 439
 440                 inet_csk_reqsk_queue_drop(sk, req, prev);
 441                 goto out;
 442
 443         case TCP_SYN_SENT:
 444         case TCP_SYN_RECV:  /* Cannot happen.
 445                                It can, it SYNs are crossed. --ANK */
 446                 if (!sock_owned_by_user(sk)) {
 447                         sk->sk_err = err;
 448                         sk->sk_error_report(sk);                /* Wake people up to see the error (see connect in sock.c) */
 449
 450                         tcp_done(sk);
 451                 } else
 452                         sk->sk_err_soft = err;
 453                 goto out;
 454         }
 455
 456         if (!sock_owned_by_user(sk) && np->recverr) {
 457                 sk->sk_err = err;
 458                 sk->sk_error_report(sk);
 459         } else
 460                 sk->sk_err_soft = err;
 461
 462 out:
 463         bh_unlock_sock(sk);
 464         sock_put(sk);
 465 }
 466
 467
 468 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
 469                               struct request_values *rvp)
 470 {
 471         struct inet6_request_sock *treq = inet6_rsk(req);
 472         struct ipv6_pinfo *np = inet6_sk(sk);
 473         struct sk_buff * skb;
 474         struct ipv6_txoptions *opt = NULL;
 475         struct in6_addr * final_p = NULL, final;
 476         struct flowi fl;
 477         struct dst_entry *dst;
 478         int err = -1;
 479
 480         memset(&fl, 0, sizeof(fl));
 481         fl.proto = IPPROTO_TCP;
 482         ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
 483         ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
 484         fl.fl6_flowlabel = 0;
 485         fl.oif = treq->iif;
 486         fl.mark = sk->sk_mark;
 487         fl.fl_ip_dport = inet_rsk(req)->rmt_port;
 488         fl.fl_ip_sport = inet_rsk(req)->loc_port;
 489         security_req_classify_flow(req, &fl);
 490
 491         opt = np->opt;
 492         if (opt && opt->srcrt) {
 493                 struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
 494                 ipv6_addr_copy(&final, &fl.fl6_dst);
 495                 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
 496                 final_p = &final;
 497         }
 498
 499         err = ip6_dst_lookup(sk, &dst, &fl);
 500         if (err)
 501                 goto done;
 502         if (final_p)
 503                 ipv6_addr_copy(&fl.fl6_dst, final_p);
 504         if ((err = xfrm_lookup(sock_net(sk), &dst, &fl, sk, 0)) < 0)
 505                 goto done;
 506
 507         skb = tcp_make_synack(sk, dst, req, rvp);
 508         if (skb) {
 509                 __tcp_v6_send_check(skb, &treq->loc_addr, &treq->rmt_addr);
 510
 511                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
 512                 err = ip6_xmit(sk, skb, &fl, opt);
 513                 err = net_xmit_eval(err);
 514         }
 515
 516 done:
 517         if (opt && opt != np->opt)
 518                 sock_kfree_s(sk, opt, opt->tot_len);
 519         dst_release(dst);
 520         return err;
 521 }
 522
 523 static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req,
 524                              struct request_values *rvp)
 525 {
 526         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
 527         return tcp_v6_send_synack(sk, req, rvp);
 528 }
 529
 530 static inline void syn_flood_warning(struct sk_buff *skb)
 531 {
 532 #ifdef CONFIG_SYN_COOKIES
 533         if (sysctl_tcp_syncookies)
 534                 printk(KERN_INFO
 535                        "TCPv6: Possible SYN flooding on port %d. "
 536                        "Sending cookies.\n", ntohs(tcp_hdr(skb)->dest));
 537         else
 538 #endif
 539                 printk(KERN_INFO
 540                        "TCPv6: Possible SYN flooding on port %d. "
 541                        "Dropping request.\n", ntohs(tcp_hdr(skb)->dest));
 542 }
 543
 544 static void tcp_v6_reqsk_destructor(struct request_sock *req)
 545 {
 546         kfree_skb(inet6_rsk(req)->pktopts);
 547 }
 548
 549 #ifdef CONFIG_TCP_MD5SIG
 550 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
 551                                                    struct in6_addr *addr)
 552 {
 553         struct tcp_sock *tp = tcp_sk(sk);
 554         int i;
 555
 556         BUG_ON(tp == NULL);
 557
 558         if (!tp->md5sig_info || !tp->md5sig_info->entries6)
 559                 return NULL;
 560
 561         for (i = 0; i < tp->md5sig_info->entries6; i++) {
 562                 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, addr))
 563                         return &tp->md5sig_info->keys6[i].base;
 564         }
 565         return NULL;
 566 }
 567
 568 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
 569                                                 struct sock *addr_sk)
 570 {
 571         return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
 572 }
 573
 574 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
 575                                                       struct request_sock *req)
 576 {
 577         return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
 578 }
 579
 580 static int tcp_v6_md5_do_add(struct sock *sk, struct in6_addr *peer,
 581                              char *newkey, u8 newkeylen)
 582 {
 583         /* Add key to the list */
 584         struct tcp_md5sig_key *key;
 585         struct tcp_sock *tp = tcp_sk(sk);
 586         struct tcp6_md5sig_key *keys;
 587
 588         key = tcp_v6_md5_do_lookup(sk, peer);
 589         if (key) {
 590                 /* modify existing entry - just update that one */
 591                 kfree(key->key);
 592                 key->key = newkey;
 593                 key->keylen = newkeylen;
 594         } else {
 595                 /* reallocate new list if current one is full. */
 596                 if (!tp->md5sig_info) {
 597                         tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
 598                         if (!tp->md5sig_info) {
 599                                 kfree(newkey);
 600                                 return -ENOMEM;
 601                         }
 602                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
 603                 }
 604                 if (tcp_alloc_md5sig_pool(sk) == NULL) {
 605                         kfree(newkey);
 606                         return -ENOMEM;
 607                 }
 608                 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
 609                         keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
 610                                        (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
 611
 612                         if (!keys) {
 613                                 tcp_free_md5sig_pool();
 614                                 kfree(newkey);
 615                                 return -ENOMEM;
 616                         }
 617
 618                         if (tp->md5sig_info->entries6)
 619                                 memmove(keys, tp->md5sig_info->keys6,
 620                                         (sizeof (tp->md5sig_info->keys6[0]) *
 621                                          tp->md5sig_info->entries6));
 622
 623                         kfree(tp->md5sig_info->keys6);
 624                         tp->md5sig_info->keys6 = keys;
 625                         tp->md5sig_info->alloced6++;
 626                 }
 627
 628                 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
 629                                peer);
 630                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
 631                 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
 632
 633                 tp->md5sig_info->entries6++;
 634         }
 635         return 0;
 636 }
 637
 638 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
 639                                u8 *newkey, __u8 newkeylen)
 640 {
 641         return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
 642                                  newkey, newkeylen);
 643 }
 644
 645 static int tcp_v6_md5_do_del(struct sock *sk, struct in6_addr *peer)
 646 {
 647         struct tcp_sock *tp = tcp_sk(sk);
 648         int i;
 649
 650         for (i = 0; i < tp->md5sig_info->entries6; i++) {
 651                 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) {
 652                         /* Free the key */
 653                         kfree(tp->md5sig_info->keys6[i].base.key);
 654                         tp->md5sig_info->entries6--;
 655
 656                         if (tp->md5sig_info->entries6 == 0) {
 657                                 kfree(tp->md5sig_info->keys6);
 658                                 tp->md5sig_info->keys6 = NULL;
 659                                 tp->md5sig_info->alloced6 = 0;
 660                         } else {
 661                                 /* shrink the database */
 662                                 if (tp->md5sig_info->entries6 != i)
 663                                         memmove(&tp->md5sig_info->keys6[i],
 664                                                 &tp->md5sig_info->keys6[i+1],
 665                                                 (tp->md5sig_info->entries6 - i)
 666                                                 * sizeof (tp->md5sig_info->keys6[0]));
 667                         }
 668                         tcp_free_md5sig_pool();
 669                         return 0;
 670                 }
 671         }
 672         return -ENOENT;
 673 }
 674
 675 static void tcp_v6_clear_md5_list (struct sock *sk)
 676 {
 677         struct tcp_sock *tp = tcp_sk(sk);
 678         int i;
 679
 680         if (tp->md5sig_info->entries6) {
 681                 for (i = 0; i < tp->md5sig_info->entries6; i++)
 682                         kfree(tp->md5sig_info->keys6[i].base.key);
 683                 tp->md5sig_info->entries6 = 0;
 684                 tcp_free_md5sig_pool();
 685         }
 686
 687         kfree(tp->md5sig_info->keys6);
 688         tp->md5sig_info->keys6 = NULL;
 689         tp->md5sig_info->alloced6 = 0;
 690
 691         if (tp->md5sig_info->entries4) {
 692                 for (i = 0; i < tp->md5sig_info->entries4; i++)
 693                         kfree(tp->md5sig_info->keys4[i].base.key);
 694                 tp->md5sig_info->entries4 = 0;
 695                 tcp_free_md5sig_pool();
 696         }
 697
 698         kfree(tp->md5sig_info->keys4);
 699         tp->md5sig_info->keys4 = NULL;
 700         tp->md5sig_info->alloced4 = 0;
 701 }
 702
 703 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
 704                                   int optlen)
 705 {
 706         struct tcp_md5sig cmd;
 707         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
 708         u8 *newkey;
 709
 710         if (optlen < sizeof(cmd))
 711                 return -EINVAL;
 712
 713         if (copy_from_user(&cmd, optval, sizeof(cmd)))
 714                 return -EFAULT;
 715
 716         if (sin6->sin6_family != AF_INET6)
 717                 return -EINVAL;
 718
 719         if (!cmd.tcpm_keylen) {
 720                 if (!tcp_sk(sk)->md5sig_info)
 721                         return -ENOENT;
 722                 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
 723                         return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
 724                 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
 725         }
 726
 727         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
 728                 return -EINVAL;
 729
 730         if (!tcp_sk(sk)->md5sig_info) {
 731                 struct tcp_sock *tp = tcp_sk(sk);
 732                 struct tcp_md5sig_info *p;
 733
 734                 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
 735                 if (!p)
 736                         return -ENOMEM;
 737
 738                 tp->md5sig_info = p;
 739                 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
 740         }
 741
 742         newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 743         if (!newkey)
 744                 return -ENOMEM;
 745         if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
 746                 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
 747                                          newkey, cmd.tcpm_keylen);
 748         }
 749         return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
 750 }
 751
 752 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
 753                                         struct in6_addr *daddr,
 754                                         struct in6_addr *saddr, int nbytes)
 755 {
 756         struct tcp6_pseudohdr *bp;
 757         struct scatterlist sg;
 758
 759         bp = &hp->md5_blk.ip6;
 760         /* 1. TCP pseudo-header (RFC2460) */
 761         ipv6_addr_copy(&bp->saddr, saddr);
 762         ipv6_addr_copy(&bp->daddr, daddr);
 763         bp->protocol = cpu_to_be32(IPPROTO_TCP);
 764         bp->len = cpu_to_be32(nbytes);
 765
 766         sg_init_one(&sg, bp, sizeof(*bp));
 767         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
 768 }
 769
 770 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
 771                                struct in6_addr *daddr, struct in6_addr *saddr,
 772                                struct tcphdr *th)
 773 {
 774         struct tcp_md5sig_pool *hp;
 775         struct hash_desc *desc;
 776
 777         hp = tcp_get_md5sig_pool();
 778         if (!hp)
 779                 goto clear_hash_noput;
 780         desc = &hp->md5_desc;
 781
 782         if (crypto_hash_init(desc))
 783                 goto clear_hash;
 784         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
 785                 goto clear_hash;
 786         if (tcp_md5_hash_header(hp, th))
 787                 goto clear_hash;
 788         if (tcp_md5_hash_key(hp, key))
 789                 goto clear_hash;
 790         if (crypto_hash_final(desc, md5_hash))
 791                 goto clear_hash;
 792
 793         tcp_put_md5sig_pool();
 794         return 0;
 795
 796 clear_hash:
 797         tcp_put_md5sig_pool();
 798 clear_hash_noput:
 799         memset(md5_hash, 0, 16);
 800         return 1;
 801 }
 802
 803 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
 804                                struct sock *sk, struct request_sock *req,
 805                                struct sk_buff *skb)
 806 {
 807         struct in6_addr *saddr, *daddr;
 808         struct tcp_md5sig_pool *hp;
 809         struct hash_desc *desc;
 810         struct tcphdr *th = tcp_hdr(skb);
 811
 812         if (sk) {
 813                 saddr = &inet6_sk(sk)->saddr;
 814                 daddr = &inet6_sk(sk)->daddr;
 815         } else if (req) {
 816                 saddr = &inet6_rsk(req)->loc_addr;
 817                 daddr = &inet6_rsk(req)->rmt_addr;
 818         } else {
 819                 struct ipv6hdr *ip6h = ipv6_hdr(skb);
 820                 saddr = &ip6h->saddr;
 821                 daddr = &ip6h->daddr;
 822         }
 823
 824         hp = tcp_get_md5sig_pool();
 825         if (!hp)
 826                 goto clear_hash_noput;
 827         desc = &hp->md5_desc;
 828
 829         if (crypto_hash_init(desc))
 830                 goto clear_hash;
 831
 832         if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
 833                 goto clear_hash;
 834         if (tcp_md5_hash_header(hp, th))
 835                 goto clear_hash;
 836         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
 837                 goto clear_hash;
 838         if (tcp_md5_hash_key(hp, key))
 839                 goto clear_hash;
 840         if (crypto_hash_final(desc, md5_hash))
 841                 goto clear_hash;
 842
 843         tcp_put_md5sig_pool();
 844         return 0;
 845
 846 clear_hash:
 847         tcp_put_md5sig_pool();
 848 clear_hash_noput:
 849         memset(md5_hash, 0, 16);
 850         return 1;
 851 }
 852
 853 static int tcp_v6_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
 854 {
 855         __u8 *hash_location = NULL;
 856         struct tcp_md5sig_key *hash_expected;
 857         struct ipv6hdr *ip6h = ipv6_hdr(skb);
 858         struct tcphdr *th = tcp_hdr(skb);
 859         int genhash;
 860         u8 newhash[16];
 861
 862         hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
 863         hash_location = tcp_parse_md5sig_option(th);
 864
 865         /* We've parsed the options - do we have a hash? */
 866         if (!hash_expected && !hash_location)
 867                 return 0;
 868
 869         if (hash_expected && !hash_location) {
 870                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
 871                 return 1;
 872         }
 873
 874         if (!hash_expected && hash_location) {
 875                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
 876                 return 1;
 877         }
 878
 879         /* check the signature */
 880         genhash = tcp_v6_md5_hash_skb(newhash,
 881                                       hash_expected,
 882                                       NULL, NULL, skb);
 883
 884         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 885                 if (net_ratelimit()) {
 886                         printk(KERN_INFO "MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
 887                                genhash ? "failed" : "mismatch",
 888                                &ip6h->saddr, ntohs(th->source),
 889                                &ip6h->daddr, ntohs(th->dest));
 890                 }
 891                 return 1;
 892         }
 893         return 0;
 894 }
 895 #endif
 896
 897 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
 898         .family         =       AF_INET6,
 899         .obj_size       =       sizeof(struct tcp6_request_sock),
 900         .rtx_syn_ack    =       tcp_v6_rtx_synack,
 901         .send_ack       =       tcp_v6_reqsk_send_ack,
 902         .destructor     =       tcp_v6_reqsk_destructor,
 903         .send_reset     =       tcp_v6_send_reset,
 904         .syn_ack_timeout =      tcp_syn_ack_timeout,
 905 };
 906
 907 #ifdef CONFIG_TCP_MD5SIG
 908 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 909         .md5_lookup     =       tcp_v6_reqsk_md5_lookup,
 910         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
 911 };
 912 #endif
 913
 914 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
 915         .twsk_obj_size  = sizeof(struct tcp6_timewait_sock),
 916         .twsk_unique    = tcp_twsk_unique,
 917         .twsk_destructor= tcp_twsk_destructor,
 918 };
 919
 920 static void __tcp_v6_send_check(struct sk_buff *skb,
 921                                 struct in6_addr *saddr, struct in6_addr *daddr)
 922 {
 923         struct tcphdr *th = tcp_hdr(skb);
 924
 925         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 926                 th->check = ~tcp_v6_check(skb->len, saddr, daddr, 0);
 927                 skb->csum_start = skb_transport_header(skb) - skb->head;
 928                 skb->csum_offset = offsetof(struct tcphdr, check);
 929         } else {
 930                 th->check = tcp_v6_check(skb->len, saddr, daddr,
 931                                          csum_partial(th, th->doff << 2,
 932                                                       skb->csum));
 933         }
 934 }
 935
 936 static void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
 937 {
 938         struct ipv6_pinfo *np = inet6_sk(sk);
 939
 940         __tcp_v6_send_check(skb, &np->saddr, &np->daddr);
 941 }
 942
 943 static int tcp_v6_gso_send_check(struct sk_buff *skb)
 944 {
 945         struct ipv6hdr *ipv6h;
 946         struct tcphdr *th;
 947
 948         if (!pskb_may_pull(skb, sizeof(*th)))
 949                 return -EINVAL;
 950
 951         ipv6h = ipv6_hdr(skb);
 952         th = tcp_hdr(skb);
 953
 954         th->check = 0;
 955         skb->ip_summed = CHECKSUM_PARTIAL;
 956         __tcp_v6_send_check(skb, &ipv6h->saddr, &ipv6h->daddr);
 957         return 0;
 958 }
 959
 960 static struct sk_buff **tcp6_gro_receive(struct sk_buff **head,
 961                                          struct sk_buff *skb)
 962 {
 963         struct ipv6hdr *iph = skb_gro_network_header(skb);
 964
 965         switch (skb->ip_summed) {
 966         case CHECKSUM_COMPLETE:
 967                 if (!tcp_v6_check(skb_gro_len(skb), &iph->saddr, &iph->daddr,
 968                                   skb->csum)) {
 969                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 970                         break;
 971                 }
 972
 973                 /* fall through */
 974         case CHECKSUM_NONE:
 975                 NAPI_GRO_CB(skb)->flush = 1;
 976                 return NULL;
 977         }
 978
 979         return tcp_gro_receive(head, skb);
 980 }
 981
 982 static int tcp6_gro_complete(struct sk_buff *skb)
 983 {
 984         struct ipv6hdr *iph = ipv6_hdr(skb);
 985         struct tcphdr *th = tcp_hdr(skb);
 986
 987         th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
 988                                   &iph->saddr, &iph->daddr, 0);
 989         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 990
 991         return tcp_gro_complete(skb);
 992 }
 993
 994 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
 995                                  u32 ts, struct tcp_md5sig_key *key, int rst)
 996 {
 997         struct tcphdr *th = tcp_hdr(skb), *t1;
 998         struct sk_buff *buff;
 999         struct flowi fl;
1000         struct net *net = dev_net(skb_dst(skb)->dev);
1001         struct sock *ctl_sk = net->ipv6.tcp_sk;
1002         unsigned int tot_len = sizeof(struct tcphdr);
1003         struct dst_entry *dst;
1004         __be32 *topt;
1005
1006         if (ts)
1007                 tot_len += TCPOLEN_TSTAMP_ALIGNED;
1008 #ifdef CONFIG_TCP_MD5SIG
1009         if (key)
1010                 tot_len += TCPOLEN_MD5SIG_ALIGNED;
1011 #endif
1012
1013         buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1014                          GFP_ATOMIC);
1015         if (buff == NULL)
1016                 return;
1017
1018         skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1019
1020         t1 = (struct tcphdr *) skb_push(buff, tot_len);
1021         skb_reset_transport_header(buff);
1022
1023         /* Swap the send and the receive. */
1024         memset(t1, 0, sizeof(*t1));
1025         t1->dest = th->source;
1026         t1->source = th->dest;
1027         t1->doff = tot_len / 4;
1028         t1->seq = htonl(seq);
1029         t1->ack_seq = htonl(ack);
1030         t1->ack = !rst || !th->ack;
1031         t1->rst = rst;
1032         t1->window = htons(win);
1033
1034         topt = (__be32 *)(t1 + 1);
1035
1036         if (ts) {
1037                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1038                                 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1039                 *topt++ = htonl(tcp_time_stamp);
1040                 *topt++ = htonl(ts);
1041         }
1042
1043 #ifdef CONFIG_TCP_MD5SIG
1044         if (key) {
1045                 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1046                                 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1047                 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
1048                                     &ipv6_hdr(skb)->saddr,
1049                                     &ipv6_hdr(skb)->daddr, t1);
1050         }
1051 #endif
1052
1053         memset(&fl, 0, sizeof(fl));
1054         ipv6_addr_copy(&fl.fl6_dst, &ipv6_hdr(skb)->saddr);
1055         ipv6_addr_copy(&fl.fl6_src, &ipv6_hdr(skb)->daddr);
1056
1057         __tcp_v6_send_check(buff, &fl.fl6_src, &fl.fl6_dst);
1058
1059         fl.proto = IPPROTO_TCP;
1060         fl.oif = inet6_iif(skb);
1061         fl.fl_ip_dport = t1->dest;
1062         fl.fl_ip_sport = t1->source;
1063         security_skb_classify_flow(skb, &fl);
1064
1065         /* Pass a socket to ip6_dst_lookup either it is for RST
1066          * Underlying function will use this to retrieve the network
1067          * namespace
1068          */
1069         if (!ip6_dst_lookup(ctl_sk, &dst, &fl)) {
1070                 if (xfrm_lookup(net, &dst, &fl, NULL, 0) >= 0) {
1071                         skb_dst_set(buff, dst);
1072                         ip6_xmit(ctl_sk, buff, &fl, NULL);
1073                         TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1074                         if (rst)
1075                                 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1076                         return;
1077                 }
1078         }
1079
1080         kfree_skb(buff);
1081 }
1082
1083 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
1084 {
1085         struct tcphdr *th = tcp_hdr(skb);
1086         u32 seq = 0, ack_seq = 0;
1087         struct tcp_md5sig_key *key = NULL;
1088
1089         if (th->rst)
1090                 return;
1091
1092         if (!ipv6_unicast_destination(skb))
1093                 return;
1094
1095 #ifdef CONFIG_TCP_MD5SIG
1096         if (sk)
1097                 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1098 #endif
1099
1100         if (th->ack)
1101                 seq = ntohl(th->ack_seq);
1102         else
1103                 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1104                           (th->doff << 2);
1105
1106         tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1);
1107 }
1108
1109 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1110                             struct tcp_md5sig_key *key)
1111 {
1112         tcp_v6_send_response(skb, seq, ack, win, ts, key, 0);
1113 }
1114
1115 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1116 {
1117         struct inet_timewait_sock *tw = inet_twsk(sk);
1118         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1119
1120         tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1121                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1122                         tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw));
1123
1124         inet_twsk_put(tw);
1125 }
1126
1127 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1128                                   struct request_sock *req)
1129 {
1130         tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1131                         tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr));
1132 }
1133
1134
1135 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1136 {
1137         struct request_sock *req, **prev;
1138         const struct tcphdr *th = tcp_hdr(skb);
1139         struct sock *nsk;
1140
1141         /* Find possible connection requests. */
1142         req = inet6_csk_search_req(sk, &prev, th->source,
1143                                    &ipv6_hdr(skb)->saddr,
1144                                    &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1145         if (req)
1146                 return tcp_check_req(sk, skb, req, prev);
1147
1148         nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1149                         &ipv6_hdr(skb)->saddr, th->source,
1150                         &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1151
1152         if (nsk) {
1153                 if (nsk->sk_state != TCP_TIME_WAIT) {
1154                         bh_lock_sock(nsk);
1155                         return nsk;
1156                 }
1157                 inet_twsk_put(inet_twsk(nsk));
1158                 return NULL;
1159         }
1160
1161 #ifdef CONFIG_SYN_COOKIES
1162         if (!th->rst && !th->syn && th->ack)
1163                 sk = cookie_v6_check(sk, skb);
1164 #endif
1165         return sk;
1166 }
1167
1168 /* FIXME: this is substantially similar to the ipv4 code.
1169  * Can some kind of merge be done? -- erics
1170  */
1171 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1172 {
1173         struct tcp_extend_values tmp_ext;
1174         struct tcp_options_received tmp_opt;
1175         u8 *hash_location;
1176         struct request_sock *req;
1177         struct inet6_request_sock *treq;
1178         struct ipv6_pinfo *np = inet6_sk(sk);
1179         struct tcp_sock *tp = tcp_sk(sk);
1180         __u32 isn = TCP_SKB_CB(skb)->when;
1181 #ifdef CONFIG_SYN_COOKIES
1182         int want_cookie = 0;
1183 #else
1184 #define want_cookie 0
1185 #endif
1186
1187         if (skb->protocol == htons(ETH_P_IP))
1188                 return tcp_v4_conn_request(sk, skb);
1189
1190         if (!ipv6_unicast_destination(skb))
1191                 goto drop;
1192
1193         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1194                 if (net_ratelimit())
1195                         syn_flood_warning(skb);
1196 #ifdef CONFIG_SYN_COOKIES
1197                 if (sysctl_tcp_syncookies)
1198                         want_cookie = 1;
1199                 else
1200 #endif
1201                 goto drop;
1202         }
1203
1204         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1205                 goto drop;
1206
1207         req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1208         if (req == NULL)
1209                 goto drop;
1210
1211 #ifdef CONFIG_TCP_MD5SIG
1212         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1213 #endif
1214
1215         tcp_clear_options(&tmp_opt);
1216         tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1217         tmp_opt.user_mss = tp->rx_opt.user_mss;
1218         tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1219
1220         if (tmp_opt.cookie_plus > 0 &&
1221             tmp_opt.saw_tstamp &&
1222             !tp->rx_opt.cookie_out_never &&
1223             (sysctl_tcp_cookie_size > 0 ||
1224              (tp->cookie_values != NULL &&
1225               tp->cookie_values->cookie_desired > 0))) {
1226                 u8 *c;
1227                 u32 *d;
1228                 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1229                 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1230
1231                 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1232                         goto drop_and_free;
1233
1234                 /* Secret recipe starts with IP addresses */
1235                 d = (__force u32 *)&ipv6_hdr(skb)->daddr.s6_addr32[0];
1236                 *mess++ ^= *d++;
1237                 *mess++ ^= *d++;
1238                 *mess++ ^= *d++;
1239                 *mess++ ^= *d++;
1240                 d = (__force u32 *)&ipv6_hdr(skb)->saddr.s6_addr32[0];
1241                 *mess++ ^= *d++;
1242                 *mess++ ^= *d++;
1243                 *mess++ ^= *d++;
1244                 *mess++ ^= *d++;
1245
1246                 /* plus variable length Initiator Cookie */
1247                 c = (u8 *)mess;
1248                 while (l-- > 0)
1249                         *c++ ^= *hash_location++;
1250
1251 #ifdef CONFIG_SYN_COOKIES
1252                 want_cookie = 0;        /* not our kind of cookie */
1253 #endif
1254                 tmp_ext.cookie_out_never = 0; /* false */
1255                 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1256         } else if (!tp->rx_opt.cookie_in_always) {
1257                 /* redundant indications, but ensure initialization. */
1258                 tmp_ext.cookie_out_never = 1; /* true */
1259                 tmp_ext.cookie_plus = 0;
1260         } else {
1261                 goto drop_and_free;
1262         }
1263         tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1264
1265         if (want_cookie && !tmp_opt.saw_tstamp)
1266                 tcp_clear_options(&tmp_opt);
1267
1268         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1269         tcp_openreq_init(req, &tmp_opt, skb);
1270
1271         treq = inet6_rsk(req);
1272         ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1273         ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1274         if (!want_cookie)
1275                 TCP_ECN_create_request(req, tcp_hdr(skb));
1276
1277         if (want_cookie) {
1278                 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1279                 req->cookie_ts = tmp_opt.tstamp_ok;
1280         } else if (!isn) {
1281                 if (ipv6_opt_accepted(sk, skb) ||
1282                     np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1283                     np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1284                         atomic_inc(&skb->users);
1285                         treq->pktopts = skb;
1286                 }
1287                 treq->iif = sk->sk_bound_dev_if;
1288
1289                 /* So that link locals have meaning */
1290                 if (!sk->sk_bound_dev_if &&
1291                     ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1292                         treq->iif = inet6_iif(skb);
1293
1294                 isn = tcp_v6_init_sequence(skb);
1295         }
1296         tcp_rsk(req)->snt_isn = isn;
1297
1298         security_inet_conn_request(sk, skb, req);
1299
1300         if (tcp_v6_send_synack(sk, req,
1301                                (struct request_values *)&tmp_ext) ||
1302             want_cookie)
1303                 goto drop_and_free;
1304
1305         inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1306         return 0;
1307
1308 drop_and_free:
1309         reqsk_free(req);
1310 drop:
1311         return 0; /* don't send reset */
1312 }
1313
1314 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1315                                           struct request_sock *req,
1316                                           struct dst_entry *dst)
1317 {
1318         struct inet6_request_sock *treq;
1319         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1320         struct tcp6_sock *newtcp6sk;
1321         struct inet_sock *newinet;
1322         struct tcp_sock *newtp;
1323         struct sock *newsk;
1324         struct ipv6_txoptions *opt;
1325 #ifdef CONFIG_TCP_MD5SIG
1326         struct tcp_md5sig_key *key;
1327 #endif
1328
1329         if (skb->protocol == htons(ETH_P_IP)) {
1330                 /*
1331                  *      v6 mapped
1332                  */
1333
1334                 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1335
1336                 if (newsk == NULL)
1337                         return NULL;
1338
1339                 newtcp6sk = (struct tcp6_sock *)newsk;
1340                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1341
1342                 newinet = inet_sk(newsk);
1343                 newnp = inet6_sk(newsk);
1344                 newtp = tcp_sk(newsk);
1345
1346                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1347
1348                 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr);
1349
1350                 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1351
1352                 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1353
1354                 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1355                 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1356 #ifdef CONFIG_TCP_MD5SIG
1357                 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1358 #endif
1359
1360                 newnp->pktoptions  = NULL;
1361                 newnp->opt         = NULL;
1362                 newnp->mcast_oif   = inet6_iif(skb);
1363                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
1364
1365                 /*
1366                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1367                  * here, tcp_create_openreq_child now does this for us, see the comment in
1368                  * that function for the gory details. -acme
1369                  */
1370
1371                 /* It is tricky place. Until this moment IPv4 tcp
1372                    worked with IPv6 icsk.icsk_af_ops.
1373                    Sync it now.
1374                  */
1375                 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1376
1377                 return newsk;
1378         }
1379
1380         treq = inet6_rsk(req);
1381         opt = np->opt;
1382
1383         if (sk_acceptq_is_full(sk))
1384                 goto out_overflow;
1385
1386         if (dst == NULL) {
1387                 struct in6_addr *final_p = NULL, final;
1388                 struct flowi fl;
1389
1390                 memset(&fl, 0, sizeof(fl));
1391                 fl.proto = IPPROTO_TCP;
1392                 ipv6_addr_copy(&fl.fl6_dst, &treq->rmt_addr);
1393                 if (opt && opt->srcrt) {
1394                         struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
1395                         ipv6_addr_copy(&final, &fl.fl6_dst);
1396                         ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
1397                         final_p = &final;
1398                 }
1399                 ipv6_addr_copy(&fl.fl6_src, &treq->loc_addr);
1400                 fl.oif = sk->sk_bound_dev_if;
1401                 fl.mark = sk->sk_mark;
1402                 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
1403                 fl.fl_ip_sport = inet_rsk(req)->loc_port;
1404                 security_req_classify_flow(req, &fl);
1405
1406                 if (ip6_dst_lookup(sk, &dst, &fl))
1407                         goto out;
1408
1409                 if (final_p)
1410                         ipv6_addr_copy(&fl.fl6_dst, final_p);
1411
1412                 if ((xfrm_lookup(sock_net(sk), &dst, &fl, sk, 0)) < 0)
1413                         goto out;
1414         }
1415
1416         newsk = tcp_create_openreq_child(sk, req, skb);
1417         if (newsk == NULL)
1418                 goto out;
1419
1420         /*
1421          * No need to charge this sock to the relevant IPv6 refcnt debug socks
1422          * count here, tcp_create_openreq_child now does this for us, see the
1423          * comment in that function for the gory details. -acme
1424          */
1425
1426         newsk->sk_gso_type = SKB_GSO_TCPV6;
1427         __ip6_dst_store(newsk, dst, NULL, NULL);
1428
1429         newtcp6sk = (struct tcp6_sock *)newsk;
1430         inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1431
1432         newtp = tcp_sk(newsk);
1433         newinet = inet_sk(newsk);
1434         newnp = inet6_sk(newsk);
1435
1436         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1437
1438         ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1439         ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1440         ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1441         newsk->sk_bound_dev_if = treq->iif;
1442
1443         /* Now IPv6 options...
1444
1445            First: no IPv4 options.
1446          */
1447         newinet->opt = NULL;
1448         newnp->ipv6_fl_list = NULL;
1449
1450         /* Clone RX bits */
1451         newnp->rxopt.all = np->rxopt.all;
1452
1453         /* Clone pktoptions received with SYN */
1454         newnp->pktoptions = NULL;
1455         if (treq->pktopts != NULL) {
1456                 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1457                 kfree_skb(treq->pktopts);
1458                 treq->pktopts = NULL;
1459                 if (newnp->pktoptions)
1460                         skb_set_owner_r(newnp->pktoptions, newsk);
1461         }
1462         newnp->opt        = NULL;
1463         newnp->mcast_oif  = inet6_iif(skb);
1464         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1465
1466         /* Clone native IPv6 options from listening socket (if any)
1467
1468            Yes, keeping reference count would be much more clever,
1469            but we make one more one thing there: reattach optmem
1470            to newsk.
1471          */
1472         if (opt) {
1473                 newnp->opt = ipv6_dup_options(newsk, opt);
1474                 if (opt != np->opt)
1475                         sock_kfree_s(sk, opt, opt->tot_len);
1476         }
1477
1478         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1479         if (newnp->opt)
1480                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1481                                                      newnp->opt->opt_flen);
1482
1483         tcp_mtup_init(newsk);
1484         tcp_sync_mss(newsk, dst_mtu(dst));
1485         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1486         tcp_initialize_rcv_mss(newsk);
1487
1488         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1489         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1490
1491 #ifdef CONFIG_TCP_MD5SIG
1492         /* Copy over the MD5 key from the original socket */
1493         if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1494                 /* We're using one, so create a matching key
1495                  * on the newsk structure. If we fail to get
1496                  * memory, then we end up not copying the key
1497                  * across. Shucks.
1498                  */
1499                 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1500                 if (newkey != NULL)
1501                         tcp_v6_md5_do_add(newsk, &newnp->daddr,
1502                                           newkey, key->keylen);
1503         }
1504 #endif
1505
1506         __inet6_hash(newsk, NULL);
1507         __inet_inherit_port(sk, newsk);
1508
1509         return newsk;
1510
1511 out_overflow:
1512         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1513 out:
1514         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1515         if (opt && opt != np->opt)
1516                 sock_kfree_s(sk, opt, opt->tot_len);
1517         dst_release(dst);
1518         return NULL;
1519 }
1520
1521 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1522 {
1523         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1524                 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1525                                   &ipv6_hdr(skb)->daddr, skb->csum)) {
1526                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1527                         return 0;
1528                 }
1529         }
1530
1531         skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1532                                               &ipv6_hdr(skb)->saddr,
1533                                               &ipv6_hdr(skb)->daddr, 0));
1534
1535         if (skb->len <= 76) {
1536                 return __skb_checksum_complete(skb);
1537         }
1538         return 0;
1539 }
1540
1541 /* The socket must have it's spinlock held when we get
1542  * here.
1543  *
1544  * We have a potential double-lock case here, so even when
1545  * doing backlog processing we use the BH locking scheme.
1546  * This is because we cannot sleep with the original spinlock
1547  * held.
1548  */
1549 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1550 {
1551         struct ipv6_pinfo *np = inet6_sk(sk);
1552         struct tcp_sock *tp;
1553         struct sk_buff *opt_skb = NULL;
1554
1555         /* Imagine: socket is IPv6. IPv4 packet arrives,
1556            goes to IPv4 receive handler and backlogged.
1557            From backlog it always goes here. Kerboom...
1558            Fortunately, tcp_rcv_established and rcv_established
1559            handle them correctly, but it is not case with
1560            tcp_v6_hnd_req and tcp_v6_send_reset().   --ANK
1561          */
1562
1563         if (skb->protocol == htons(ETH_P_IP))
1564                 return tcp_v4_do_rcv(sk, skb);
1565
1566 #ifdef CONFIG_TCP_MD5SIG
1567         if (tcp_v6_inbound_md5_hash (sk, skb))
1568                 goto discard;
1569 #endif
1570
1571         if (sk_filter(sk, skb))
1572                 goto discard;
1573
1574         /*
1575          *      socket locking is here for SMP purposes as backlog rcv
1576          *      is currently called with bh processing disabled.
1577          */
1578
1579         /* Do Stevens' IPV6_PKTOPTIONS.
1580
1581            Yes, guys, it is the only place in our code, where we
1582            may make it not affecting IPv4.
1583            The rest of code is protocol independent,
1584            and I do not like idea to uglify IPv4.
1585
1586            Actually, all the idea behind IPV6_PKTOPTIONS
1587            looks not very well thought. For now we latch
1588            options, received in the last packet, enqueued
1589            by tcp. Feel free to propose better solution.
1590                                                --ANK (980728)
1591          */
1592         if (np->rxopt.all)
1593                 opt_skb = skb_clone(skb, GFP_ATOMIC);
1594
1595         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1596                 TCP_CHECK_TIMER(sk);
1597                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1598                         goto reset;
1599                 TCP_CHECK_TIMER(sk);
1600                 if (opt_skb)
1601                         goto ipv6_pktoptions;
1602                 return 0;
1603         }
1604
1605         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1606                 goto csum_err;
1607
1608         if (sk->sk_state == TCP_LISTEN) {
1609                 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1610                 if (!nsk)
1611                         goto discard;
1612
1613                 /*
1614                  * Queue it on the new socket if the new socket is active,
1615                  * otherwise we just shortcircuit this and continue with
1616                  * the new socket..
1617                  */
1618                 if(nsk != sk) {
1619                         if (tcp_child_process(sk, nsk, skb))
1620                                 goto reset;
1621                         if (opt_skb)
1622                                 __kfree_skb(opt_skb);
1623                         return 0;
1624                 }
1625         }
1626
1627         TCP_CHECK_TIMER(sk);
1628         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1629                 goto reset;
1630         TCP_CHECK_TIMER(sk);
1631         if (opt_skb)
1632                 goto ipv6_pktoptions;
1633         return 0;
1634
1635 reset:
1636         tcp_v6_send_reset(sk, skb);
1637 discard:
1638         if (opt_skb)
1639                 __kfree_skb(opt_skb);
1640         kfree_skb(skb);
1641         return 0;
1642 csum_err:
1643         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1644         goto discard;
1645
1646
1647 ipv6_pktoptions:
1648         /* Do you ask, what is it?
1649
1650            1. skb was enqueued by tcp.
1651            2. skb is added to tail of read queue, rather than out of order.
1652            3. socket is not in passive state.
1653            4. Finally, it really contains options, which user wants to receive.
1654          */
1655         tp = tcp_sk(sk);
1656         if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1657             !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1658                 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1659                         np->mcast_oif = inet6_iif(opt_skb);
1660                 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1661                         np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1662                 if (ipv6_opt_accepted(sk, opt_skb)) {
1663                         skb_set_owner_r(opt_skb, sk);
1664                         opt_skb = xchg(&np->pktoptions, opt_skb);
1665                 } else {
1666                         __kfree_skb(opt_skb);
1667                         opt_skb = xchg(&np->pktoptions, NULL);
1668                 }
1669         }
1670
1671         kfree_skb(opt_skb);
1672         return 0;
1673 }
1674
1675 static int tcp_v6_rcv(struct sk_buff *skb)
1676 {
1677         struct tcphdr *th;
1678         struct sock *sk;
1679         int ret;
1680         struct net *net = dev_net(skb->dev);
1681
1682         if (skb->pkt_type != PACKET_HOST)
1683                 goto discard_it;
1684
1685         /*
1686          *      Count it even if it's bad.
1687          */
1688         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1689
1690         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1691                 goto discard_it;
1692
1693         th = tcp_hdr(skb);
1694
1695         if (th->doff < sizeof(struct tcphdr)/4)
1696                 goto bad_packet;
1697         if (!pskb_may_pull(skb, th->doff*4))
1698                 goto discard_it;
1699
1700         if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1701                 goto bad_packet;
1702
1703         th = tcp_hdr(skb);
1704         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1705         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1706                                     skb->len - th->doff*4);
1707         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1708         TCP_SKB_CB(skb)->when = 0;
1709         TCP_SKB_CB(skb)->flags = ipv6_get_dsfield(ipv6_hdr(skb));
1710         TCP_SKB_CB(skb)->sacked = 0;
1711
1712         sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1713         if (!sk)
1714                 goto no_tcp_socket;
1715
1716 process:
1717         if (sk->sk_state == TCP_TIME_WAIT)
1718                 goto do_time_wait;
1719
1720         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1721                 goto discard_and_relse;
1722
1723         if (sk_filter(sk, skb))
1724                 goto discard_and_relse;
1725
1726         skb->dev = NULL;
1727
1728         bh_lock_sock_nested(sk);
1729         ret = 0;
1730         if (!sock_owned_by_user(sk)) {
1731 #ifdef CONFIG_NET_DMA
1732                 struct tcp_sock *tp = tcp_sk(sk);
1733                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1734                         tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1735                 if (tp->ucopy.dma_chan)
1736                         ret = tcp_v6_do_rcv(sk, skb);
1737                 else
1738 #endif
1739                 {
1740                         if (!tcp_prequeue(sk, skb))
1741                                 ret = tcp_v6_do_rcv(sk, skb);
1742                 }
1743         } else if (unlikely(sk_add_backlog(sk, skb))) {
1744                 bh_unlock_sock(sk);
1745                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1746                 goto discard_and_relse;
1747         }
1748         bh_unlock_sock(sk);
1749
1750         sock_put(sk);
1751         return ret ? -1 : 0;
1752
1753 no_tcp_socket:
1754         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1755                 goto discard_it;
1756
1757         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1758 bad_packet:
1759                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1760         } else {
1761                 tcp_v6_send_reset(NULL, skb);
1762         }
1763
1764 discard_it:
1765
1766         /*
1767          *      Discard frame
1768          */
1769
1770         kfree_skb(skb);
1771         return 0;
1772
1773 discard_and_relse:
1774         sock_put(sk);
1775         goto discard_it;
1776
1777 do_time_wait:
1778         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1779                 inet_twsk_put(inet_twsk(sk));
1780                 goto discard_it;
1781         }
1782
1783         if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1784                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1785                 inet_twsk_put(inet_twsk(sk));
1786                 goto discard_it;
1787         }
1788
1789         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1790         case TCP_TW_SYN:
1791         {
1792                 struct sock *sk2;
1793
1794                 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1795                                             &ipv6_hdr(skb)->daddr,
1796                                             ntohs(th->dest), inet6_iif(skb));
1797                 if (sk2 != NULL) {
1798                         struct inet_timewait_sock *tw = inet_twsk(sk);
1799                         inet_twsk_deschedule(tw, &tcp_death_row);
1800                         inet_twsk_put(tw);
1801                         sk = sk2;
1802                         goto process;
1803                 }
1804                 /* Fall through to ACK */
1805         }
1806         case TCP_TW_ACK:
1807                 tcp_v6_timewait_ack(sk, skb);
1808                 break;
1809         case TCP_TW_RST:
1810                 goto no_tcp_socket;
1811         case TCP_TW_SUCCESS:;
1812         }
1813         goto discard_it;
1814 }
1815
1816 static int tcp_v6_remember_stamp(struct sock *sk)
1817 {
1818         /* Alas, not yet... */
1819         return 0;
1820 }
1821
1822 static const struct inet_connection_sock_af_ops ipv6_specific = {
1823         .queue_xmit        = inet6_csk_xmit,
1824         .send_check        = tcp_v6_send_check,
1825         .rebuild_header    = inet6_sk_rebuild_header,
1826         .conn_request      = tcp_v6_conn_request,
1827         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1828         .remember_stamp    = tcp_v6_remember_stamp,
1829         .net_header_len    = sizeof(struct ipv6hdr),
1830         .setsockopt        = ipv6_setsockopt,
1831         .getsockopt        = ipv6_getsockopt,
1832         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1833         .sockaddr_len      = sizeof(struct sockaddr_in6),
1834         .bind_conflict     = inet6_csk_bind_conflict,
1835 #ifdef CONFIG_COMPAT
1836         .compat_setsockopt = compat_ipv6_setsockopt,
1837         .compat_getsockopt = compat_ipv6_getsockopt,
1838 #endif
1839 };
1840
1841 #ifdef CONFIG_TCP_MD5SIG
1842 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1843         .md5_lookup     =       tcp_v6_md5_lookup,
1844         .calc_md5_hash  =       tcp_v6_md5_hash_skb,
1845         .md5_add        =       tcp_v6_md5_add_func,
1846         .md5_parse      =       tcp_v6_parse_md5_keys,
1847 };
1848 #endif
1849
1850 /*
1851  *      TCP over IPv4 via INET6 API
1852  */
1853
1854 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1855         .queue_xmit        = ip_queue_xmit,
1856         .send_check        = tcp_v4_send_check,
1857         .rebuild_header    = inet_sk_rebuild_header,
1858         .conn_request      = tcp_v6_conn_request,
1859         .syn_recv_sock     = tcp_v6_syn_recv_sock,
1860         .remember_stamp    = tcp_v4_remember_stamp,
1861         .net_header_len    = sizeof(struct iphdr),
1862         .setsockopt        = ipv6_setsockopt,
1863         .getsockopt        = ipv6_getsockopt,
1864         .addr2sockaddr     = inet6_csk_addr2sockaddr,
1865         .sockaddr_len      = sizeof(struct sockaddr_in6),
1866         .bind_conflict     = inet6_csk_bind_conflict,
1867 #ifdef CONFIG_COMPAT
1868         .compat_setsockopt = compat_ipv6_setsockopt,
1869         .compat_getsockopt = compat_ipv6_getsockopt,
1870 #endif
1871 };
1872
1873 #ifdef CONFIG_TCP_MD5SIG
1874 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1875         .md5_lookup     =       tcp_v4_md5_lookup,
1876         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1877         .md5_add        =       tcp_v6_md5_add_func,
1878         .md5_parse      =       tcp_v6_parse_md5_keys,
1879 };
1880 #endif
1881
1882 /* NOTE: A lot of things set to zero explicitly by call to
1883  *       sk_alloc() so need not be done here.
1884  */
1885 static int tcp_v6_init_sock(struct sock *sk)
1886 {
1887         struct inet_connection_sock *icsk = inet_csk(sk);
1888         struct tcp_sock *tp = tcp_sk(sk);
1889
1890         skb_queue_head_init(&tp->out_of_order_queue);
1891         tcp_init_xmit_timers(sk);
1892         tcp_prequeue_init(tp);
1893
1894         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1895         tp->mdev = TCP_TIMEOUT_INIT;
1896
1897         /* So many TCP implementations out there (incorrectly) count the
1898          * initial SYN frame in their delayed-ACK and congestion control
1899          * algorithms that we must have the following bandaid to talk
1900          * efficiently to them.  -DaveM
1901          */
1902         tp->snd_cwnd = 2;
1903
1904         /* See draft-stevens-tcpca-spec-01 for discussion of the
1905          * initialization of these values.
1906          */
1907         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1908         tp->snd_cwnd_clamp = ~0;
1909         tp->mss_cache = TCP_MSS_DEFAULT;
1910
1911         tp->reordering = sysctl_tcp_reordering;
1912
1913         sk->sk_state = TCP_CLOSE;
1914
1915         icsk->icsk_af_ops = &ipv6_specific;
1916         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1917         icsk->icsk_sync_mss = tcp_sync_mss;
1918         sk->sk_write_space = sk_stream_write_space;
1919         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1920
1921 #ifdef CONFIG_TCP_MD5SIG
1922         tp->af_specific = &tcp_sock_ipv6_specific;
1923 #endif
1924
1925         /* TCP Cookie Transactions */
1926         if (sysctl_tcp_cookie_size > 0) {
1927                 /* Default, cookies without s_data_payload. */
1928                 tp->cookie_values =
1929                         kzalloc(sizeof(*tp->cookie_values),
1930                                 sk->sk_allocation);
1931                 if (tp->cookie_values != NULL)
1932                         kref_init(&tp->cookie_values->kref);
1933         }
1934         /* Presumed zeroed, in order of appearance:
1935          *      cookie_in_always, cookie_out_never,
1936          *      s_data_constant, s_data_in, s_data_out
1937          */
1938         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1939         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1940
1941         local_bh_disable();
1942         percpu_counter_inc(&tcp_sockets_allocated);
1943         local_bh_enable();
1944
1945         return 0;
1946 }
1947
1948 static void tcp_v6_destroy_sock(struct sock *sk)
1949 {
1950 #ifdef CONFIG_TCP_MD5SIG
1951         /* Clean up the MD5 key list */
1952         if (tcp_sk(sk)->md5sig_info)
1953                 tcp_v6_clear_md5_list(sk);
1954 #endif
1955         tcp_v4_destroy_sock(sk);
1956         inet6_destroy_sock(sk);
1957 }
1958
1959 #ifdef CONFIG_PROC_FS
1960 /* Proc filesystem TCPv6 sock list dumping. */
1961 static void get_openreq6(struct seq_file *seq,
1962                          struct sock *sk, struct request_sock *req, int i, int uid)
1963 {
1964         int ttd = req->expires - jiffies;
1965         struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1966         struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1967
1968         if (ttd < 0)
1969                 ttd = 0;
1970
1971         seq_printf(seq,
1972                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1973                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
1974                    i,
1975                    src->s6_addr32[0], src->s6_addr32[1],
1976                    src->s6_addr32[2], src->s6_addr32[3],
1977                    ntohs(inet_rsk(req)->loc_port),
1978                    dest->s6_addr32[0], dest->s6_addr32[1],
1979                    dest->s6_addr32[2], dest->s6_addr32[3],
1980                    ntohs(inet_rsk(req)->rmt_port),
1981                    TCP_SYN_RECV,
1982                    0,0, /* could print option size, but that is af dependent. */
1983                    1,   /* timers active (only the expire timer) */
1984                    jiffies_to_clock_t(ttd),
1985                    req->retrans,
1986                    uid,
1987                    0,  /* non standard timer */
1988                    0, /* open_requests have no inode */
1989                    0, req);
1990 }
1991
1992 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1993 {
1994         struct in6_addr *dest, *src;
1995         __u16 destp, srcp;
1996         int timer_active;
1997         unsigned long timer_expires;
1998         struct inet_sock *inet = inet_sk(sp);
1999         struct tcp_sock *tp = tcp_sk(sp);
2000         const struct inet_connection_sock *icsk = inet_csk(sp);
2001         struct ipv6_pinfo *np = inet6_sk(sp);
2002
2003         dest  = &np->daddr;
2004         src   = &np->rcv_saddr;
2005         destp = ntohs(inet->inet_dport);
2006         srcp  = ntohs(inet->inet_sport);
2007
2008         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2009                 timer_active    = 1;
2010                 timer_expires   = icsk->icsk_timeout;
2011         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2012                 timer_active    = 4;
2013                 timer_expires   = icsk->icsk_timeout;
2014         } else if (timer_pending(&sp->sk_timer)) {
2015                 timer_active    = 2;
2016                 timer_expires   = sp->sk_timer.expires;
2017         } else {
2018                 timer_active    = 0;
2019                 timer_expires = jiffies;
2020         }
2021
2022         seq_printf(seq,
2023                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2024                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %lu %lu %u %u %d\n",
2025                    i,
2026                    src->s6_addr32[0], src->s6_addr32[1],
2027                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2028                    dest->s6_addr32[0], dest->s6_addr32[1],
2029                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2030                    sp->sk_state,
2031                    tp->write_seq-tp->snd_una,
2032                    (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2033                    timer_active,
2034                    jiffies_to_clock_t(timer_expires - jiffies),
2035                    icsk->icsk_retransmits,
2036                    sock_i_uid(sp),
2037                    icsk->icsk_probes_out,
2038                    sock_i_ino(sp),
2039                    atomic_read(&sp->sk_refcnt), sp,
2040                    jiffies_to_clock_t(icsk->icsk_rto),
2041                    jiffies_to_clock_t(icsk->icsk_ack.ato),
2042                    (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2043                    tp->snd_cwnd,
2044                    tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
2045                    );
2046 }
2047
2048 static void get_timewait6_sock(struct seq_file *seq,
2049                                struct inet_timewait_sock *tw, int i)
2050 {
2051         struct in6_addr *dest, *src;
2052         __u16 destp, srcp;
2053         struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2054         int ttd = tw->tw_ttd - jiffies;
2055
2056         if (ttd < 0)
2057                 ttd = 0;
2058
2059         dest = &tw6->tw_v6_daddr;
2060         src  = &tw6->tw_v6_rcv_saddr;
2061         destp = ntohs(tw->tw_dport);
2062         srcp  = ntohs(tw->tw_sport);
2063
2064         seq_printf(seq,
2065                    "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2066                    "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p\n",
2067                    i,
2068                    src->s6_addr32[0], src->s6_addr32[1],
2069                    src->s6_addr32[2], src->s6_addr32[3], srcp,
2070                    dest->s6_addr32[0], dest->s6_addr32[1],
2071                    dest->s6_addr32[2], dest->s6_addr32[3], destp,
2072                    tw->tw_substate, 0, 0,
2073                    3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2074                    atomic_read(&tw->tw_refcnt), tw);
2075 }
2076
2077 static int tcp6_seq_show(struct seq_file *seq, void *v)
2078 {
2079         struct tcp_iter_state *st;
2080
2081         if (v == SEQ_START_TOKEN) {
2082                 seq_puts(seq,
2083                          "  sl  "
2084                          "local_address                         "
2085                          "remote_address                        "
2086                          "st tx_queue rx_queue tr tm->when retrnsmt"
2087                          "   uid  timeout inode\n");
2088                 goto out;
2089         }
2090         st = seq->private;
2091
2092         switch (st->state) {
2093         case TCP_SEQ_STATE_LISTENING:
2094         case TCP_SEQ_STATE_ESTABLISHED:
2095                 get_tcp6_sock(seq, v, st->num);
2096                 break;
2097         case TCP_SEQ_STATE_OPENREQ:
2098                 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2099                 break;
2100         case TCP_SEQ_STATE_TIME_WAIT:
2101                 get_timewait6_sock(seq, v, st->num);
2102                 break;
2103         }
2104 out:
2105         return 0;
2106 }
2107
2108 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2109         .name           = "tcp6",
2110         .family         = AF_INET6,
2111         .seq_fops       = {
2112                 .owner          = THIS_MODULE,
2113         },
2114         .seq_ops        = {
2115                 .show           = tcp6_seq_show,
2116         },
2117 };
2118
2119 int __net_init tcp6_proc_init(struct net *net)
2120 {
2121         return tcp_proc_register(net, &tcp6_seq_afinfo);
2122 }
2123
2124 void tcp6_proc_exit(struct net *net)
2125 {
2126         tcp_proc_unregister(net, &tcp6_seq_afinfo);
2127 }
2128 #endif
2129
2130 struct proto tcpv6_prot = {
2131         .name                   = "TCPv6",
2132         .owner                  = THIS_MODULE,
2133         .close                  = tcp_close,
2134         .connect                = tcp_v6_connect,
2135         .disconnect             = tcp_disconnect,
2136         .accept                 = inet_csk_accept,
2137         .ioctl                  = tcp_ioctl,
2138         .init                   = tcp_v6_init_sock,
2139         .destroy                = tcp_v6_destroy_sock,
2140         .shutdown               = tcp_shutdown,
2141         .setsockopt             = tcp_setsockopt,
2142         .getsockopt             = tcp_getsockopt,
2143         .recvmsg                = tcp_recvmsg,
2144         .backlog_rcv            = tcp_v6_do_rcv,
2145         .hash                   = tcp_v6_hash,
2146         .unhash                 = inet_unhash,
2147         .get_port               = inet_csk_get_port,
2148         .enter_memory_pressure  = tcp_enter_memory_pressure,
2149         .sockets_allocated      = &tcp_sockets_allocated,
2150         .memory_allocated       = &tcp_memory_allocated,
2151         .memory_pressure        = &tcp_memory_pressure,
2152         .orphan_count           = &tcp_orphan_count,
2153         .sysctl_mem             = sysctl_tcp_mem,
2154         .sysctl_wmem            = sysctl_tcp_wmem,
2155         .sysctl_rmem            = sysctl_tcp_rmem,
2156         .max_header             = MAX_TCP_HEADER,
2157         .obj_size               = sizeof(struct tcp6_sock),
2158         .slab_flags             = SLAB_DESTROY_BY_RCU,
2159         .twsk_prot              = &tcp6_timewait_sock_ops,
2160         .rsk_prot               = &tcp6_request_sock_ops,
2161         .h.hashinfo             = &tcp_hashinfo,
2162 #ifdef CONFIG_COMPAT
2163         .compat_setsockopt      = compat_tcp_setsockopt,
2164         .compat_getsockopt      = compat_tcp_getsockopt,
2165 #endif
2166 };
2167
2168 static const struct inet6_protocol tcpv6_protocol = {
2169         .handler        =       tcp_v6_rcv,
2170         .err_handler    =       tcp_v6_err,
2171         .gso_send_check =       tcp_v6_gso_send_check,
2172         .gso_segment    =       tcp_tso_segment,
2173         .gro_receive    =       tcp6_gro_receive,
2174         .gro_complete   =       tcp6_gro_complete,
2175         .flags          =       INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2176 };
2177
2178 static struct inet_protosw tcpv6_protosw = {
2179         .type           =       SOCK_STREAM,
2180         .protocol       =       IPPROTO_TCP,
2181         .prot           =       &tcpv6_prot,
2182         .ops            =       &inet6_stream_ops,
2183         .no_check       =       0,
2184         .flags          =       INET_PROTOSW_PERMANENT |
2185                                 INET_PROTOSW_ICSK,
2186 };
2187
2188 static int __net_init tcpv6_net_init(struct net *net)
2189 {
2190         return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2191                                     SOCK_RAW, IPPROTO_TCP, net);
2192 }
2193
2194 static void __net_exit tcpv6_net_exit(struct net *net)
2195 {
2196         inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2197 }
2198
2199 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2200 {
2201         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2202 }
2203
2204 static struct pernet_operations tcpv6_net_ops = {
2205         .init       = tcpv6_net_init,
2206         .exit       = tcpv6_net_exit,
2207         .exit_batch = tcpv6_net_exit_batch,
2208 };
2209
2210 int __init tcpv6_init(void)
2211 {
2212         int ret;
2213
2214         ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2215         if (ret)
2216                 goto out;
2217
2218         /* register inet6 protocol */
2219         ret = inet6_register_protosw(&tcpv6_protosw);
2220         if (ret)
2221                 goto out_tcpv6_protocol;
2222
2223         ret = register_pernet_subsys(&tcpv6_net_ops);
2224         if (ret)
2225                 goto out_tcpv6_protosw;
2226 out:
2227         return ret;
2228
2229 out_tcpv6_protocol:
2230         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2231 out_tcpv6_protosw:
2232         inet6_unregister_protosw(&tcpv6_protosw);
2233         goto out;
2234 }
2235
2236 void tcpv6_exit(void)
2237 {
2238         unregister_pernet_subsys(&tcpv6_net_ops);
2239         inet6_unregister_protosw(&tcpv6_protosw);
2240         inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2241 }