err.no Git - linux-2.6/blob - 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 <net/checksum.h>
  16 #include <net/icmp.h>
  17 #include <net/ip.h>
  18 #include <net/tcp.h>  /* For tcp_prot in getorigdst */
  19 #include <linux/icmp.h>
  20 #include <linux/udp.h>
  21 #include <linux/jhash.h>
  22
  23 #include <linux/netfilter_ipv4.h>
  24 #include <net/netfilter/nf_conntrack.h>
  25 #include <net/netfilter/nf_conntrack_core.h>
  26 #include <net/netfilter/nf_nat.h>
  27 #include <net/netfilter/nf_nat_protocol.h>
  28 #include <net/netfilter/nf_nat_core.h>
  29 #include <net/netfilter/nf_nat_helper.h>
  30 #include <net/netfilter/nf_conntrack_helper.h>
  31 #include <net/netfilter/nf_conntrack_l3proto.h>
  32 #include <net/netfilter/nf_conntrack_l4proto.h>
  33
  34 static DEFINE_RWLOCK(nf_nat_lock);
  35
  36 static struct nf_conntrack_l3proto *l3proto __read_mostly;
  37
  38 /* Calculated at init based on memory size */
  39 static unsigned int nf_nat_htable_size __read_mostly;
  40 static int nf_nat_vmalloced;
  41
  42 static struct hlist_head *bysource __read_mostly;
  43
  44 #define MAX_IP_NAT_PROTO 256
  45 static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
  46                                                 __read_mostly;
  47
  48 static inline const struct nf_nat_protocol *
  49 __nf_nat_proto_find(u_int8_t protonum)
  50 {
  51         return rcu_dereference(nf_nat_protos[protonum]);
  52 }
  53
  54 const struct nf_nat_protocol *
  55 nf_nat_proto_find_get(u_int8_t protonum)
  56 {
  57         const struct nf_nat_protocol *p;
  58
  59         rcu_read_lock();
  60         p = __nf_nat_proto_find(protonum);
  61         if (!try_module_get(p->me))
  62                 p = &nf_nat_unknown_protocol;
  63         rcu_read_unlock();
  64
  65         return p;
  66 }
  67 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
  68
  69 void
  70 nf_nat_proto_put(const struct nf_nat_protocol *p)
  71 {
  72         module_put(p->me);
  73 }
  74 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
  75
  76 /* We keep an extra hash for each conntrack, for fast searching. */
  77 static inline unsigned int
  78 hash_by_src(const struct nf_conntrack_tuple *tuple)
  79 {
  80         /* Original src, to ensure we map it consistently if poss. */
  81         return jhash_3words((__force u32)tuple->src.u3.ip,
  82                             (__force u32)tuple->src.u.all,
  83                             tuple->dst.protonum, 0) % nf_nat_htable_size;
  84 }
  85
  86 /* Is this tuple already taken? (not by us) */
  87 int
  88 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
  89                   const struct nf_conn *ignored_conntrack)
  90 {
  91         /* Conntrack tracking doesn't keep track of outgoing tuples; only
  92            incoming ones.  NAT means they don't have a fixed mapping,
  93            so we invert the tuple and look for the incoming reply.
  94
  95            We could keep a separate hash if this proves too slow. */
  96         struct nf_conntrack_tuple reply;
  97
  98         nf_ct_invert_tuplepr(&reply, tuple);
  99         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 100 }
 101 EXPORT_SYMBOL(nf_nat_used_tuple);
 102
 103 /* If we source map this tuple so reply looks like reply_tuple, will
 104  * that meet the constraints of range. */
 105 static int
 106 in_range(const struct nf_conntrack_tuple *tuple,
 107          const struct nf_nat_range *range)
 108 {
 109         const struct nf_nat_protocol *proto;
 110         int ret = 0;
 111
 112         /* If we are supposed to map IPs, then we must be in the
 113            range specified, otherwise let this drag us onto a new src IP. */
 114         if (range->flags & IP_NAT_RANGE_MAP_IPS) {
 115                 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
 116                     ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
 117                         return 0;
 118         }
 119
 120         rcu_read_lock();
 121         proto = __nf_nat_proto_find(tuple->dst.protonum);
 122         if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 123             proto->in_range(tuple, IP_NAT_MANIP_SRC,
 124                             &range->min, &range->max))
 125                 ret = 1;
 126         rcu_read_unlock();
 127
 128         return ret;
 129 }
 130
 131 static inline int
 132 same_src(const struct nf_conn *ct,
 133          const struct nf_conntrack_tuple *tuple)
 134 {
 135         const struct nf_conntrack_tuple *t;
 136
 137         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 138         return (t->dst.protonum == tuple->dst.protonum &&
 139                 t->src.u3.ip == tuple->src.u3.ip &&
 140                 t->src.u.all == tuple->src.u.all);
 141 }
 142
 143 /* Only called for SRC manip */
 144 static int
 145 find_appropriate_src(const struct nf_conntrack_tuple *tuple,
 146                      struct nf_conntrack_tuple *result,
 147                      const struct nf_nat_range *range)
 148 {
 149         unsigned int h = hash_by_src(tuple);
 150         struct nf_conn_nat *nat;
 151         struct nf_conn *ct;
 152         struct hlist_node *n;
 153
 154         read_lock_bh(&nf_nat_lock);
 155         hlist_for_each_entry(nat, n, &bysource[h], bysource) {
 156                 ct = nat->ct;
 157                 if (same_src(ct, tuple)) {
 158                         /* Copy source part from reply tuple. */
 159                         nf_ct_invert_tuplepr(result,
 160                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 161                         result->dst = tuple->dst;
 162
 163                         if (in_range(result, range)) {
 164                                 read_unlock_bh(&nf_nat_lock);
 165                                 return 1;
 166                         }
 167                 }
 168         }
 169         read_unlock_bh(&nf_nat_lock);
 170         return 0;
 171 }
 172
 173 /* For [FUTURE] fragmentation handling, we want the least-used
 174    src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 175    if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 176    1-65535, we don't do pro-rata allocation based on ports; we choose
 177    the ip with the lowest src-ip/dst-ip/proto usage.
 178 */
 179 static void
 180 find_best_ips_proto(struct nf_conntrack_tuple *tuple,
 181                     const struct nf_nat_range *range,
 182                     const struct nf_conn *ct,
 183                     enum nf_nat_manip_type maniptype)
 184 {
 185         __be32 *var_ipp;
 186         /* Host order */
 187         u_int32_t minip, maxip, j;
 188
 189         /* No IP mapping?  Do nothing. */
 190         if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
 191                 return;
 192
 193         if (maniptype == IP_NAT_MANIP_SRC)
 194                 var_ipp = &tuple->src.u3.ip;
 195         else
 196                 var_ipp = &tuple->dst.u3.ip;
 197
 198         /* Fast path: only one choice. */
 199         if (range->min_ip == range->max_ip) {
 200                 *var_ipp = range->min_ip;
 201                 return;
 202         }
 203
 204         /* Hashing source and destination IPs gives a fairly even
 205          * spread in practice (if there are a small number of IPs
 206          * involved, there usually aren't that many connections
 207          * anyway).  The consistency means that servers see the same
 208          * client coming from the same IP (some Internet Banking sites
 209          * like this), even across reboots. */
 210         minip = ntohl(range->min_ip);
 211         maxip = ntohl(range->max_ip);
 212         j = jhash_2words((__force u32)tuple->src.u3.ip,
 213                          (__force u32)tuple->dst.u3.ip, 0);
 214         *var_ipp = htonl(minip + j % (maxip - minip + 1));
 215 }
 216
 217 /* Manipulate the tuple into the range given.  For NF_INET_POST_ROUTING,
 218  * we change the source to map into the range.  For NF_INET_PRE_ROUTING
 219  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 220  * range.  It might not be possible to get a unique tuple, but we try.
 221  * At worst (or if we race), we will end up with a final duplicate in
 222  * __ip_conntrack_confirm and drop the packet. */
 223 static void
 224 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 225                  const struct nf_conntrack_tuple *orig_tuple,
 226                  const struct nf_nat_range *range,
 227                  struct nf_conn *ct,
 228                  enum nf_nat_manip_type maniptype)
 229 {
 230         const struct nf_nat_protocol *proto;
 231
 232         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 233            and that same mapping gives a unique tuple within the given
 234            range, use that.
 235
 236            This is only required for source (ie. NAT/masq) mappings.
 237            So far, we don't do local source mappings, so multiple
 238            manips not an issue.  */
 239         if (maniptype == IP_NAT_MANIP_SRC) {
 240                 if (find_appropriate_src(orig_tuple, tuple, range)) {
 241                         pr_debug("get_unique_tuple: Found current src map\n");
 242                         if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM))
 243                                 if (!nf_nat_used_tuple(tuple, ct))
 244                                         return;
 245                 }
 246         }
 247
 248         /* 2) Select the least-used IP/proto combination in the given
 249            range. */
 250         *tuple = *orig_tuple;
 251         find_best_ips_proto(tuple, range, ct, maniptype);
 252
 253         /* 3) The per-protocol part of the manip is made to map into
 254            the range to make a unique tuple. */
 255
 256         rcu_read_lock();
 257         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
 258
 259         /* Change protocol info to have some randomization */
 260         if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) {
 261                 proto->unique_tuple(tuple, range, maniptype, ct);
 262                 goto out;
 263         }
 264
 265         /* Only bother mapping if it's not already in range and unique */
 266         if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 267              proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
 268             !nf_nat_used_tuple(tuple, ct))
 269                 goto out;
 270
 271         /* Last change: get protocol to try to obtain unique tuple. */
 272         proto->unique_tuple(tuple, range, maniptype, ct);
 273 out:
 274         rcu_read_unlock();
 275 }
 276
 277 unsigned int
 278 nf_nat_setup_info(struct nf_conn *ct,
 279                   const struct nf_nat_range *range,
 280                   enum nf_nat_manip_type maniptype)
 281 {
 282         struct nf_conntrack_tuple curr_tuple, new_tuple;
 283         struct nf_conn_nat *nat;
 284         int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
 285
 286         /* nat helper or nfctnetlink also setup binding */
 287         nat = nfct_nat(ct);
 288         if (!nat) {
 289                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 290                 if (nat == NULL) {
 291                         pr_debug("failed to add NAT extension\n");
 292                         return NF_ACCEPT;
 293                 }
 294         }
 295
 296         NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC ||
 297                      maniptype == IP_NAT_MANIP_DST);
 298         BUG_ON(nf_nat_initialized(ct, maniptype));
 299
 300         /* What we've got will look like inverse of reply. Normally
 301            this is what is in the conntrack, except for prior
 302            manipulations (future optimization: if num_manips == 0,
 303            orig_tp =
 304            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
 305         nf_ct_invert_tuplepr(&curr_tuple,
 306                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 307
 308         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 309
 310         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 311                 struct nf_conntrack_tuple reply;
 312
 313                 /* Alter conntrack table so will recognize replies. */
 314                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 315                 nf_conntrack_alter_reply(ct, &reply);
 316
 317                 /* Non-atomic: we own this at the moment. */
 318                 if (maniptype == IP_NAT_MANIP_SRC)
 319                         ct->status |= IPS_SRC_NAT;
 320                 else
 321                         ct->status |= IPS_DST_NAT;
 322         }
 323
 324         /* Place in source hash if this is the first time. */
 325         if (have_to_hash) {
 326                 unsigned int srchash;
 327
 328                 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 329                 write_lock_bh(&nf_nat_lock);
 330                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
 331                 nat = nfct_nat(ct);
 332                 nat->ct = ct;
 333                 hlist_add_head(&nat->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 *skb,
 351           unsigned int iphdroff,
 352           const struct nf_conntrack_tuple *target,
 353           enum nf_nat_manip_type maniptype)
 354 {
 355         struct iphdr *iph;
 356         const struct nf_nat_protocol *p;
 357
 358         if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
 359                 return 0;
 360
 361         iph = (void *)skb->data + iphdroff;
 362
 363         /* Manipulate protcol part. */
 364
 365         /* rcu_read_lock()ed by nf_hook_slow */
 366         p = __nf_nat_proto_find(proto);
 367         if (!p->manip_pkt(skb, iphdroff, target, maniptype))
 368                 return 0;
 369
 370         iph = (void *)skb->data + iphdroff;
 371
 372         if (maniptype == IP_NAT_MANIP_SRC) {
 373                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 374                 iph->saddr = target->src.u3.ip;
 375         } else {
 376                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 377                 iph->daddr = target->dst.u3.ip;
 378         }
 379         return 1;
 380 }
 381
 382 /* Do packet manipulations according to nf_nat_setup_info. */
 383 unsigned int nf_nat_packet(struct nf_conn *ct,
 384                            enum ip_conntrack_info ctinfo,
 385                            unsigned int hooknum,
 386                            struct sk_buff *skb)
 387 {
 388         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 389         unsigned long statusbit;
 390         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 391
 392         if (mtype == IP_NAT_MANIP_SRC)
 393                 statusbit = IPS_SRC_NAT;
 394         else
 395                 statusbit = IPS_DST_NAT;
 396
 397         /* Invert if this is reply dir. */
 398         if (dir == IP_CT_DIR_REPLY)
 399                 statusbit ^= IPS_NAT_MASK;
 400
 401         /* Non-atomic: these bits don't change. */
 402         if (ct->status & statusbit) {
 403                 struct nf_conntrack_tuple target;
 404
 405                 /* We are aiming to look like inverse of other direction. */
 406                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 407
 408                 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
 409                         return NF_DROP;
 410         }
 411         return NF_ACCEPT;
 412 }
 413 EXPORT_SYMBOL_GPL(nf_nat_packet);
 414
 415 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
 416 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 417                                   enum ip_conntrack_info ctinfo,
 418                                   unsigned int hooknum,
 419                                   struct sk_buff *skb)
 420 {
 421         struct {
 422                 struct icmphdr icmp;
 423                 struct iphdr ip;
 424         } *inside;
 425         struct nf_conntrack_l4proto *l4proto;
 426         struct nf_conntrack_tuple inner, target;
 427         int hdrlen = ip_hdrlen(skb);
 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(skb, hdrlen + sizeof(*inside)))
 433                 return 0;
 434
 435         inside = (void *)skb->data + ip_hdrlen(skb);
 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(skb, hooknum, hdrlen, 0))
 440                 return 0;
 441
 442         /* Must be RELATED */
 443         NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
 444                      skb->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         pr_debug("icmp_reply_translation: translating error %p manip %u "
 459                  "dir %s\n", skb, manip,
 460                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 461
 462         /* rcu_read_lock()ed by nf_hook_slow */
 463         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
 464
 465         if (!nf_ct_get_tuple(skb,
 466                              ip_hdrlen(skb) + sizeof(struct icmphdr),
 467                              (ip_hdrlen(skb) +
 468                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 469                              (u_int16_t)AF_INET,
 470                              inside->ip.protocol,
 471                              &inner, l3proto, l4proto))
 472                 return 0;
 473
 474         /* Change inner back to look like incoming packet.  We do the
 475            opposite manip on this hook to normal, because it might not
 476            pass all hooks (locally-generated ICMP).  Consider incoming
 477            packet: PREROUTING (DST manip), routing produces ICMP, goes
 478            through POSTROUTING (which must correct the DST manip). */
 479         if (!manip_pkt(inside->ip.protocol, skb,
 480                        ip_hdrlen(skb) + sizeof(inside->icmp),
 481                        &ct->tuplehash[!dir].tuple,
 482                        !manip))
 483                 return 0;
 484
 485         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 486                 /* Reloading "inside" here since manip_pkt inner. */
 487                 inside = (void *)skb->data + ip_hdrlen(skb);
 488                 inside->icmp.checksum = 0;
 489                 inside->icmp.checksum =
 490                         csum_fold(skb_checksum(skb, hdrlen,
 491                                                skb->len - hdrlen, 0));
 492         }
 493
 494         /* Change outer to look the reply to an incoming packet
 495          * (proto 0 means don't invert per-proto part). */
 496         if (manip == IP_NAT_MANIP_SRC)
 497                 statusbit = IPS_SRC_NAT;
 498         else
 499                 statusbit = IPS_DST_NAT;
 500
 501         /* Invert if this is reply dir. */
 502         if (dir == IP_CT_DIR_REPLY)
 503                 statusbit ^= IPS_NAT_MASK;
 504
 505         if (ct->status & statusbit) {
 506                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 507                 if (!manip_pkt(0, skb, 0, &target, manip))
 508                         return 0;
 509         }
 510
 511         return 1;
 512 }
 513 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 514
 515 /* Protocol registration. */
 516 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 517 {
 518         int ret = 0;
 519
 520         write_lock_bh(&nf_nat_lock);
 521         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
 522                 ret = -EBUSY;
 523                 goto out;
 524         }
 525         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 526  out:
 527         write_unlock_bh(&nf_nat_lock);
 528         return ret;
 529 }
 530 EXPORT_SYMBOL(nf_nat_protocol_register);
 531
 532 /* Noone stores the protocol anywhere; simply delete it. */
 533 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 534 {
 535         write_lock_bh(&nf_nat_lock);
 536         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 537                            &nf_nat_unknown_protocol);
 538         write_unlock_bh(&nf_nat_lock);
 539         synchronize_rcu();
 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_nlattr(struct sk_buff *skb,
 546                             const struct nf_nat_range *range)
 547 {
 548         NLA_PUT_BE16(skb, CTA_PROTONAT_PORT_MIN, range->min.tcp.port);
 549         NLA_PUT_BE16(skb, CTA_PROTONAT_PORT_MAX, range->max.tcp.port);
 550
 551         return 0;
 552
 553 nla_put_failure:
 554         return -1;
 555 }
 556 EXPORT_SYMBOL_GPL(nf_nat_port_nlattr_to_range);
 557
 558 int
 559 nf_nat_port_nlattr_to_range(struct nlattr *tb[], struct nf_nat_range *range)
 560 {
 561         int ret = 0;
 562
 563         /* we have to return whether we actually parsed something or not */
 564
 565         if (tb[CTA_PROTONAT_PORT_MIN]) {
 566                 ret = 1;
 567                 range->min.tcp.port = nla_get_be16(tb[CTA_PROTONAT_PORT_MIN]);
 568         }
 569
 570         if (!tb[CTA_PROTONAT_PORT_MAX]) {
 571                 if (ret)
 572                         range->max.tcp.port = range->min.tcp.port;
 573         } else {
 574                 ret = 1;
 575                 range->max.tcp.port = nla_get_be16(tb[CTA_PROTONAT_PORT_MAX]);
 576         }
 577
 578         return ret;
 579 }
 580 EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nlattr);
 581 #endif
 582
 583 /* Noone using conntrack by the time this called. */
 584 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 585 {
 586         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 587
 588         if (nat == NULL || nat->ct == NULL)
 589                 return;
 590
 591         NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
 592
 593         write_lock_bh(&nf_nat_lock);
 594         hlist_del(&nat->bysource);
 595         nat->ct = NULL;
 596         write_unlock_bh(&nf_nat_lock);
 597 }
 598
 599 static void nf_nat_move_storage(struct nf_conn *conntrack, void *old)
 600 {
 601         struct nf_conn_nat *new_nat = nf_ct_ext_find(conntrack, NF_CT_EXT_NAT);
 602         struct nf_conn_nat *old_nat = (struct nf_conn_nat *)old;
 603         struct nf_conn *ct = old_nat->ct;
 604
 605         if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
 606                 return;
 607
 608         write_lock_bh(&nf_nat_lock);
 609         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 610         new_nat->ct = ct;
 611         write_unlock_bh(&nf_nat_lock);
 612 }
 613
 614 static struct nf_ct_ext_type nat_extend __read_mostly = {
 615         .len            = sizeof(struct nf_conn_nat),
 616         .align          = __alignof__(struct nf_conn_nat),
 617         .destroy        = nf_nat_cleanup_conntrack,
 618         .move           = nf_nat_move_storage,
 619         .id             = NF_CT_EXT_NAT,
 620         .flags          = NF_CT_EXT_F_PREALLOC,
 621 };
 622
 623 static int __init nf_nat_init(void)
 624 {
 625         size_t i;
 626         int ret;
 627
 628         ret = nf_ct_extend_register(&nat_extend);
 629         if (ret < 0) {
 630                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 631                 return ret;
 632         }
 633
 634         /* Leave them the same for the moment. */
 635         nf_nat_htable_size = nf_conntrack_htable_size;
 636
 637         bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
 638                                          &nf_nat_vmalloced);
 639         if (!bysource) {
 640                 ret = -ENOMEM;
 641                 goto cleanup_extend;
 642         }
 643
 644         /* Sew in builtin protocols. */
 645         write_lock_bh(&nf_nat_lock);
 646         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 647                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
 648         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 649         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 650         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 651         write_unlock_bh(&nf_nat_lock);
 652
 653         for (i = 0; i < nf_nat_htable_size; i++) {
 654                 INIT_HLIST_HEAD(&bysource[i]);
 655         }
 656
 657         /* Initialize fake conntrack so that NAT will skip it */
 658         nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
 659
 660         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 661         return 0;
 662
 663  cleanup_extend:
 664         nf_ct_extend_unregister(&nat_extend);
 665         return ret;
 666 }
 667
 668 /* Clear NAT section of all conntracks, in case we're loaded again. */
 669 static int clean_nat(struct nf_conn *i, void *data)
 670 {
 671         struct nf_conn_nat *nat = nfct_nat(i);
 672
 673         if (!nat)
 674                 return 0;
 675         memset(nat, 0, sizeof(*nat));
 676         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
 677         return 0;
 678 }
 679
 680 static void __exit nf_nat_cleanup(void)
 681 {
 682         nf_ct_iterate_cleanup(&clean_nat, NULL);
 683         synchronize_rcu();
 684         nf_ct_free_hashtable(bysource, nf_nat_vmalloced, nf_nat_htable_size);
 685         nf_ct_l3proto_put(l3proto);
 686         nf_ct_extend_unregister(&nat_extend);
 687 }
 688
 689 MODULE_LICENSE("GPL");
 690
 691 module_init(nf_nat_init);
 692 module_exit(nf_nat_cleanup);