net/ipv4/netfilter/nf_nat_core.c

   1 /* NAT for netfilter; shared with compatibility layer. */
   2
   3 /* (C) 1999-2001 Paul `Rusty' Russell
   4  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/types.h>
  13 #include <linux/timer.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/vmalloc.h>
  16 #include <net/checksum.h>
  17 #include <net/icmp.h>
  18 #include <net/ip.h>
  19 #include <net/tcp.h>  /* For tcp_prot in getorigdst */
  20 #include <linux/icmp.h>
  21 #include <linux/udp.h>
  22 #include <linux/jhash.h>
  23
  24 #include <linux/netfilter_ipv4.h>
  25 #include <net/netfilter/nf_conntrack.h>
  26 #include <net/netfilter/nf_conntrack_core.h>
  27 #include <net/netfilter/nf_nat.h>
  28 #include <net/netfilter/nf_nat_protocol.h>
  29 #include <net/netfilter/nf_nat_core.h>
  30 #include <net/netfilter/nf_nat_helper.h>
  31 #include <net/netfilter/nf_conntrack_helper.h>
  32 #include <net/netfilter/nf_conntrack_l3proto.h>
  33 #include <net/netfilter/nf_conntrack_l4proto.h>
  34
  35 #if 0
  36 #define DEBUGP printk
  37 #else
  38 #define DEBUGP(format, args...)
  39 #endif
  40
  41 static DEFINE_RWLOCK(nf_nat_lock);
  42
  43 static struct nf_conntrack_l3proto *l3proto = NULL;
  44
  45 /* Calculated at init based on memory size */
  46 static unsigned int nf_nat_htable_size;
  47
  48 static struct list_head *bysource;
  49
  50 #define MAX_IP_NAT_PROTO 256
  51 static struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO];
  52
  53 static inline struct nf_nat_protocol *
  54 __nf_nat_proto_find(u_int8_t protonum)
  55 {
  56         return rcu_dereference(nf_nat_protos[protonum]);
  57 }
  58
  59 struct nf_nat_protocol *
  60 nf_nat_proto_find_get(u_int8_t protonum)
  61 {
  62         struct nf_nat_protocol *p;
  63
  64         rcu_read_lock();
  65         p = __nf_nat_proto_find(protonum);
  66         if (!try_module_get(p->me))
  67                 p = &nf_nat_unknown_protocol;
  68         rcu_read_unlock();
  69
  70         return p;
  71 }
  72 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
  73
  74 void
  75 nf_nat_proto_put(struct nf_nat_protocol *p)
  76 {
  77         module_put(p->me);
  78 }
  79 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
  80
  81 /* We keep an extra hash for each conntrack, for fast searching. */
  82 static inline unsigned int
  83 hash_by_src(const struct nf_conntrack_tuple *tuple)
  84 {
  85         /* Original src, to ensure we map it consistently if poss. */
  86         return jhash_3words((__force u32)tuple->src.u3.ip, tuple->src.u.all,
  87                             tuple->dst.protonum, 0) % nf_nat_htable_size;
  88 }
  89
  90 /* Is this tuple already taken? (not by us) */
  91 int
  92 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
  93                   const struct nf_conn *ignored_conntrack)
  94 {
  95         /* Conntrack tracking doesn't keep track of outgoing tuples; only
  96            incoming ones.  NAT means they don't have a fixed mapping,
  97            so we invert the tuple and look for the incoming reply.
  98
  99            We could keep a separate hash if this proves too slow. */
 100         struct nf_conntrack_tuple reply;
 101
 102         nf_ct_invert_tuplepr(&reply, tuple);
 103         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 104 }
 105 EXPORT_SYMBOL(nf_nat_used_tuple);
 106
 107 /* If we source map this tuple so reply looks like reply_tuple, will
 108  * that meet the constraints of range. */
 109 static int
 110 in_range(const struct nf_conntrack_tuple *tuple,
 111          const struct nf_nat_range *range)
 112 {
 113         struct nf_nat_protocol *proto;
 114         int ret = 0;
 115
 116         /* If we are supposed to map IPs, then we must be in the
 117            range specified, otherwise let this drag us onto a new src IP. */
 118         if (range->flags & IP_NAT_RANGE_MAP_IPS) {
 119                 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
 120                     ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
 121                         return 0;
 122         }
 123
 124         rcu_read_lock();
 125         proto = __nf_nat_proto_find(tuple->dst.protonum);
 126         if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 127             proto->in_range(tuple, IP_NAT_MANIP_SRC,
 128                             &range->min, &range->max))
 129                 ret = 1;
 130         rcu_read_unlock();
 131
 132         return ret;
 133 }
 134
 135 static inline int
 136 same_src(const struct nf_conn *ct,
 137          const struct nf_conntrack_tuple *tuple)
 138 {
 139         const struct nf_conntrack_tuple *t;
 140
 141         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 142         return (t->dst.protonum == tuple->dst.protonum &&
 143                 t->src.u3.ip == tuple->src.u3.ip &&
 144                 t->src.u.all == tuple->src.u.all);
 145 }
 146
 147 /* Only called for SRC manip */
 148 static int
 149 find_appropriate_src(const struct nf_conntrack_tuple *tuple,
 150                      struct nf_conntrack_tuple *result,
 151                      const struct nf_nat_range *range)
 152 {
 153         unsigned int h = hash_by_src(tuple);
 154         struct nf_conn_nat *nat;
 155         struct nf_conn *ct;
 156
 157         read_lock_bh(&nf_nat_lock);
 158         list_for_each_entry(nat, &bysource[h], info.bysource) {
 159                 ct = nat->info.ct;
 160                 if (same_src(ct, tuple)) {
 161                         /* Copy source part from reply tuple. */
 162                         nf_ct_invert_tuplepr(result,
 163                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 164                         result->dst = tuple->dst;
 165
 166                         if (in_range(result, range)) {
 167                                 read_unlock_bh(&nf_nat_lock);
 168                                 return 1;
 169                         }
 170                 }
 171         }
 172         read_unlock_bh(&nf_nat_lock);
 173         return 0;
 174 }
 175
 176 /* For [FUTURE] fragmentation handling, we want the least-used
 177    src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 178    if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 179    1-65535, we don't do pro-rata allocation based on ports; we choose
 180    the ip with the lowest src-ip/dst-ip/proto usage.
 181 */
 182 static void
 183 find_best_ips_proto(struct nf_conntrack_tuple *tuple,
 184                     const struct nf_nat_range *range,
 185                     const struct nf_conn *ct,
 186                     enum nf_nat_manip_type maniptype)
 187 {
 188         __be32 *var_ipp;
 189         /* Host order */
 190         u_int32_t minip, maxip, j;
 191
 192         /* No IP mapping?  Do nothing. */
 193         if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
 194                 return;
 195
 196         if (maniptype == IP_NAT_MANIP_SRC)
 197                 var_ipp = &tuple->src.u3.ip;
 198         else
 199                 var_ipp = &tuple->dst.u3.ip;
 200
 201         /* Fast path: only one choice. */
 202         if (range->min_ip == range->max_ip) {
 203                 *var_ipp = range->min_ip;
 204                 return;
 205         }
 206
 207         /* Hashing source and destination IPs gives a fairly even
 208          * spread in practice (if there are a small number of IPs
 209          * involved, there usually aren't that many connections
 210          * anyway).  The consistency means that servers see the same
 211          * client coming from the same IP (some Internet Banking sites
 212          * like this), even across reboots. */
 213         minip = ntohl(range->min_ip);
 214         maxip = ntohl(range->max_ip);
 215         j = jhash_2words((__force u32)tuple->src.u3.ip,
 216                          (__force u32)tuple->dst.u3.ip, 0);
 217         *var_ipp = htonl(minip + j % (maxip - minip + 1));
 218 }
 219
 220 /* Manipulate the tuple into the range given.  For NF_IP_POST_ROUTING,
 221  * we change the source to map into the range.  For NF_IP_PRE_ROUTING
 222  * and NF_IP_LOCAL_OUT, we change the destination to map into the
 223  * range.  It might not be possible to get a unique tuple, but we try.
 224  * At worst (or if we race), we will end up with a final duplicate in
 225  * __ip_conntrack_confirm and drop the packet. */
 226 static void
 227 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 228                  const struct nf_conntrack_tuple *orig_tuple,
 229                  const struct nf_nat_range *range,
 230                  struct nf_conn *ct,
 231                  enum nf_nat_manip_type maniptype)
 232 {
 233         struct nf_nat_protocol *proto;
 234
 235         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 236            and that same mapping gives a unique tuple within the given
 237            range, use that.
 238
 239            This is only required for source (ie. NAT/masq) mappings.
 240            So far, we don't do local source mappings, so multiple
 241            manips not an issue.  */
 242         if (maniptype == IP_NAT_MANIP_SRC) {
 243                 if (find_appropriate_src(orig_tuple, tuple, range)) {
 244                         DEBUGP("get_unique_tuple: Found current src map\n");
 245                         if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM))
 246                                 if (!nf_nat_used_tuple(tuple, ct))
 247                                         return;
 248                 }
 249         }
 250
 251         /* 2) Select the least-used IP/proto combination in the given
 252            range. */
 253         *tuple = *orig_tuple;
 254         find_best_ips_proto(tuple, range, ct, maniptype);
 255
 256         /* 3) The per-protocol part of the manip is made to map into
 257            the range to make a unique tuple. */
 258
 259         rcu_read_lock();
 260         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
 261
 262         /* Change protocol info to have some randomization */
 263         if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) {
 264                 proto->unique_tuple(tuple, range, maniptype, ct);
 265                 goto out;
 266         }
 267
 268         /* Only bother mapping if it's not already in range and unique */
 269         if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 270              proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
 271             !nf_nat_used_tuple(tuple, ct))
 272                 goto out;
 273
 274         /* Last change: get protocol to try to obtain unique tuple. */
 275         proto->unique_tuple(tuple, range, maniptype, ct);
 276 out:
 277         rcu_read_unlock();
 278 }
 279
 280 unsigned int
 281 nf_nat_setup_info(struct nf_conn *ct,
 282                   const struct nf_nat_range *range,
 283                   unsigned int hooknum)
 284 {
 285         struct nf_conntrack_tuple curr_tuple, new_tuple;
 286         struct nf_conn_nat *nat;
 287         struct nf_nat_info *info;
 288         int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
 289         enum nf_nat_manip_type maniptype = HOOK2MANIP(hooknum);
 290
 291         /* nat helper or nfctnetlink also setup binding */
 292         nat = nfct_nat(ct);
 293         if (!nat) {
 294                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 295                 if (nat == NULL) {
 296                         DEBUGP("failed to add NAT extension\n");
 297                         return NF_ACCEPT;
 298                 }
 299         }
 300
 301         NF_CT_ASSERT(hooknum == NF_IP_PRE_ROUTING ||
 302                      hooknum == NF_IP_POST_ROUTING ||
 303                      hooknum == NF_IP_LOCAL_IN ||
 304                      hooknum == NF_IP_LOCAL_OUT);
 305         BUG_ON(nf_nat_initialized(ct, maniptype));
 306
 307         /* What we've got will look like inverse of reply. Normally
 308            this is what is in the conntrack, except for prior
 309            manipulations (future optimization: if num_manips == 0,
 310            orig_tp =
 311            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
 312         nf_ct_invert_tuplepr(&curr_tuple,
 313                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 314
 315         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 316
 317         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 318                 struct nf_conntrack_tuple reply;
 319
 320                 /* Alter conntrack table so will recognize replies. */
 321                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 322                 nf_conntrack_alter_reply(ct, &reply);
 323
 324                 /* Non-atomic: we own this at the moment. */
 325                 if (maniptype == IP_NAT_MANIP_SRC)
 326                         ct->status |= IPS_SRC_NAT;
 327                 else
 328                         ct->status |= IPS_DST_NAT;
 329         }
 330
 331         /* Place in source hash if this is the first time. */
 332         if (have_to_hash) {
 333                 unsigned int srchash;
 334
 335                 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 336                 write_lock_bh(&nf_nat_lock);
 337                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
 338                 info = &nfct_nat(ct)->info;
 339                 info->ct = ct;
 340                 list_add(&info->bysource, &bysource[srchash]);
 341                 write_unlock_bh(&nf_nat_lock);
 342         }
 343
 344         /* It's done. */
 345         if (maniptype == IP_NAT_MANIP_DST)
 346                 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
 347         else
 348                 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 349
 350         return NF_ACCEPT;
 351 }
 352 EXPORT_SYMBOL(nf_nat_setup_info);
 353
 354 /* Returns true if succeeded. */
 355 static int
 356 manip_pkt(u_int16_t proto,
 357           struct sk_buff **pskb,
 358           unsigned int iphdroff,
 359           const struct nf_conntrack_tuple *target,
 360           enum nf_nat_manip_type maniptype)
 361 {
 362         struct iphdr *iph;
 363         struct nf_nat_protocol *p;
 364
 365         if (!skb_make_writable(pskb, iphdroff + sizeof(*iph)))
 366                 return 0;
 367
 368         iph = (void *)(*pskb)->data + iphdroff;
 369
 370         /* Manipulate protcol part. */
 371
 372         /* rcu_read_lock()ed by nf_hook_slow */
 373         p = __nf_nat_proto_find(proto);
 374         if (!p->manip_pkt(pskb, iphdroff, target, maniptype))
 375                 return 0;
 376
 377         iph = (void *)(*pskb)->data + iphdroff;
 378
 379         if (maniptype == IP_NAT_MANIP_SRC) {
 380                 nf_csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 381                 iph->saddr = target->src.u3.ip;
 382         } else {
 383                 nf_csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 384                 iph->daddr = target->dst.u3.ip;
 385         }
 386         return 1;
 387 }
 388
 389 /* Do packet manipulations according to nf_nat_setup_info. */
 390 unsigned int nf_nat_packet(struct nf_conn *ct,
 391                            enum ip_conntrack_info ctinfo,
 392                            unsigned int hooknum,
 393                            struct sk_buff **pskb)
 394 {
 395         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 396         unsigned long statusbit;
 397         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 398
 399         if (mtype == IP_NAT_MANIP_SRC)
 400                 statusbit = IPS_SRC_NAT;
 401         else
 402                 statusbit = IPS_DST_NAT;
 403
 404         /* Invert if this is reply dir. */
 405         if (dir == IP_CT_DIR_REPLY)
 406                 statusbit ^= IPS_NAT_MASK;
 407
 408         /* Non-atomic: these bits don't change. */
 409         if (ct->status & statusbit) {
 410                 struct nf_conntrack_tuple target;
 411
 412                 /* We are aiming to look like inverse of other direction. */
 413                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 414
 415                 if (!manip_pkt(target.dst.protonum, pskb, 0, &target, mtype))
 416                         return NF_DROP;
 417         }
 418         return NF_ACCEPT;
 419 }
 420 EXPORT_SYMBOL_GPL(nf_nat_packet);
 421
 422 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
 423 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 424                                   enum ip_conntrack_info ctinfo,
 425                                   unsigned int hooknum,
 426                                   struct sk_buff **pskb)
 427 {
 428         struct {
 429                 struct icmphdr icmp;
 430                 struct iphdr ip;
 431         } *inside;
 432         struct nf_conntrack_l4proto *l4proto;
 433         struct nf_conntrack_tuple inner, target;
 434         int hdrlen = ip_hdrlen(*pskb);
 435         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 436         unsigned long statusbit;
 437         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 438
 439         if (!skb_make_writable(pskb, hdrlen + sizeof(*inside)))
 440                 return 0;
 441
 442         inside = (void *)(*pskb)->data + ip_hdrlen(*pskb);
 443
 444         /* We're actually going to mangle it beyond trivial checksum
 445            adjustment, so make sure the current checksum is correct. */
 446         if (nf_ip_checksum(*pskb, hooknum, hdrlen, 0))
 447                 return 0;
 448
 449         /* Must be RELATED */
 450         NF_CT_ASSERT((*pskb)->nfctinfo == IP_CT_RELATED ||
 451                      (*pskb)->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
 452
 453         /* Redirects on non-null nats must be dropped, else they'll
 454            start talking to each other without our translation, and be
 455            confused... --RR */
 456         if (inside->icmp.type == ICMP_REDIRECT) {
 457                 /* If NAT isn't finished, assume it and drop. */
 458                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 459                         return 0;
 460
 461                 if (ct->status & IPS_NAT_MASK)
 462                         return 0;
 463         }
 464
 465         DEBUGP("icmp_reply_translation: translating error %p manp %u dir %s\n",
 466                *pskb, manip, dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 467
 468         /* rcu_read_lock()ed by nf_hook_slow */
 469         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
 470
 471         if (!nf_ct_get_tuple(*pskb,
 472                              ip_hdrlen(*pskb) + sizeof(struct icmphdr),
 473                              (ip_hdrlen(*pskb) +
 474                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 475                              (u_int16_t)AF_INET,
 476                              inside->ip.protocol,
 477                              &inner, l3proto, l4proto))
 478                 return 0;
 479
 480         /* Change inner back to look like incoming packet.  We do the
 481            opposite manip on this hook to normal, because it might not
 482            pass all hooks (locally-generated ICMP).  Consider incoming
 483            packet: PREROUTING (DST manip), routing produces ICMP, goes
 484            through POSTROUTING (which must correct the DST manip). */
 485         if (!manip_pkt(inside->ip.protocol, pskb,
 486                        ip_hdrlen(*pskb) + sizeof(inside->icmp),
 487                        &ct->tuplehash[!dir].tuple,
 488                        !manip))
 489                 return 0;
 490
 491         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 492                 /* Reloading "inside" here since manip_pkt inner. */
 493                 inside = (void *)(*pskb)->data + ip_hdrlen(*pskb);
 494                 inside->icmp.checksum = 0;
 495                 inside->icmp.checksum =
 496                         csum_fold(skb_checksum(*pskb, hdrlen,
 497                                                (*pskb)->len - hdrlen, 0));
 498         }
 499
 500         /* Change outer to look the reply to an incoming packet
 501          * (proto 0 means don't invert per-proto part). */
 502         if (manip == IP_NAT_MANIP_SRC)
 503                 statusbit = IPS_SRC_NAT;
 504         else
 505                 statusbit = IPS_DST_NAT;
 506
 507         /* Invert if this is reply dir. */
 508         if (dir == IP_CT_DIR_REPLY)
 509                 statusbit ^= IPS_NAT_MASK;
 510
 511         if (ct->status & statusbit) {
 512                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 513                 if (!manip_pkt(0, pskb, 0, &target, manip))
 514                         return 0;
 515         }
 516
 517         return 1;
 518 }
 519 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 520
 521 /* Protocol registration. */
 522 int nf_nat_protocol_register(struct nf_nat_protocol *proto)
 523 {
 524         int ret = 0;
 525
 526         write_lock_bh(&nf_nat_lock);
 527         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
 528                 ret = -EBUSY;
 529                 goto out;
 530         }
 531         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 532  out:
 533         write_unlock_bh(&nf_nat_lock);
 534         return ret;
 535 }
 536 EXPORT_SYMBOL(nf_nat_protocol_register);
 537
 538 /* Noone stores the protocol anywhere; simply delete it. */
 539 void nf_nat_protocol_unregister(struct nf_nat_protocol *proto)
 540 {
 541         write_lock_bh(&nf_nat_lock);
 542         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 543                            &nf_nat_unknown_protocol);
 544         write_unlock_bh(&nf_nat_lock);
 545         synchronize_rcu();
 546 }
 547 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 548
 549 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 550 int
 551 nf_nat_port_range_to_nfattr(struct sk_buff *skb,
 552                             const struct nf_nat_range *range)
 553 {
 554         NFA_PUT(skb, CTA_PROTONAT_PORT_MIN, sizeof(__be16),
 555                 &range->min.tcp.port);
 556         NFA_PUT(skb, CTA_PROTONAT_PORT_MAX, sizeof(__be16),
 557                 &range->max.tcp.port);
 558
 559         return 0;
 560
 561 nfattr_failure:
 562         return -1;
 563 }
 564 EXPORT_SYMBOL_GPL(nf_nat_port_nfattr_to_range);
 565
 566 int
 567 nf_nat_port_nfattr_to_range(struct nfattr *tb[], struct nf_nat_range *range)
 568 {
 569         int ret = 0;
 570
 571         /* we have to return whether we actually parsed something or not */
 572
 573         if (tb[CTA_PROTONAT_PORT_MIN-1]) {
 574                 ret = 1;
 575                 range->min.tcp.port =
 576                         *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MIN-1]);
 577         }
 578
 579         if (!tb[CTA_PROTONAT_PORT_MAX-1]) {
 580                 if (ret)
 581                         range->max.tcp.port = range->min.tcp.port;
 582         } else {
 583                 ret = 1;
 584                 range->max.tcp.port =
 585                         *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MAX-1]);
 586         }
 587
 588         return ret;
 589 }
 590 EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nfattr);
 591 #endif
 592
 593 /* Noone using conntrack by the time this called. */
 594 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 595 {
 596         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 597
 598         if (nat == NULL || nat->info.ct == NULL)
 599                 return;
 600
 601         NF_CT_ASSERT(nat->info.ct->status & IPS_NAT_DONE_MASK);
 602
 603         write_lock_bh(&nf_nat_lock);
 604         list_del(&nat->info.bysource);
 605         nat->info.ct = NULL;
 606         write_unlock_bh(&nf_nat_lock);
 607 }
 608
 609 static void nf_nat_move_storage(struct nf_conn *conntrack, void *old)
 610 {
 611         struct nf_conn_nat *new_nat = nf_ct_ext_find(conntrack, NF_CT_EXT_NAT);
 612         struct nf_conn_nat *old_nat = (struct nf_conn_nat *)old;
 613         struct nf_conn *ct = old_nat->info.ct;
 614         unsigned int srchash;
 615
 616         if (!(ct->status & IPS_NAT_DONE_MASK))
 617                 return;
 618
 619         srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 620
 621         write_lock_bh(&nf_nat_lock);
 622         list_replace(&old_nat->info.bysource, &new_nat->info.bysource);
 623         new_nat->info.ct = ct;
 624         write_unlock_bh(&nf_nat_lock);
 625 }
 626
 627 struct nf_ct_ext_type nat_extend = {
 628         .len            = sizeof(struct nf_conn_nat),
 629         .align          = __alignof__(struct nf_conn_nat),
 630         .destroy        = nf_nat_cleanup_conntrack,
 631         .move           = nf_nat_move_storage,
 632         .id             = NF_CT_EXT_NAT,
 633         .flags          = NF_CT_EXT_F_PREALLOC,
 634 };
 635
 636 static int __init nf_nat_init(void)
 637 {
 638         size_t i;
 639         int ret;
 640
 641         ret = nf_ct_extend_register(&nat_extend);
 642         if (ret < 0) {
 643                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 644                 return ret;
 645         }
 646
 647         /* Leave them the same for the moment. */
 648         nf_nat_htable_size = nf_conntrack_htable_size;
 649
 650         /* One vmalloc for both hash tables */
 651         bysource = vmalloc(sizeof(struct list_head) * nf_nat_htable_size);
 652         if (!bysource) {
 653                 ret = -ENOMEM;
 654                 goto cleanup_extend;
 655         }
 656
 657         /* Sew in builtin protocols. */
 658         write_lock_bh(&nf_nat_lock);
 659         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 660                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
 661         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 662         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 663         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 664         write_unlock_bh(&nf_nat_lock);
 665
 666         for (i = 0; i < nf_nat_htable_size; i++) {
 667                 INIT_LIST_HEAD(&bysource[i]);
 668         }
 669
 670         /* Initialize fake conntrack so that NAT will skip it */
 671         nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
 672
 673         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 674         return 0;
 675
 676  cleanup_extend:
 677         nf_ct_extend_unregister(&nat_extend);
 678         return ret;
 679 }
 680
 681 /* Clear NAT section of all conntracks, in case we're loaded again. */
 682 static int clean_nat(struct nf_conn *i, void *data)
 683 {
 684         struct nf_conn_nat *nat = nfct_nat(i);
 685
 686         if (!nat)
 687                 return 0;
 688         memset(nat, 0, sizeof(nat));
 689         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
 690         return 0;
 691 }
 692
 693 static void __exit nf_nat_cleanup(void)
 694 {
 695         nf_ct_iterate_cleanup(&clean_nat, NULL);
 696         synchronize_rcu();
 697         vfree(bysource);
 698         nf_ct_l3proto_put(l3proto);
 699         nf_ct_extend_unregister(&nat_extend);
 700 }
 701
 702 MODULE_LICENSE("GPL");
 703
 704 module_init(nf_nat_init);
 705 module_exit(nf_nat_cleanup);