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