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 = NULL;
  37
  38 /* Calculated at init based on memory size */
  39 static unsigned int nf_nat_htable_size;
  40 static int nf_nat_vmalloced;
  41
  42 static struct hlist_head *bysource;
  43
  44 #define MAX_IP_NAT_PROTO 256
  45 static struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO];
  46
  47 static inline struct nf_nat_protocol *
  48 __nf_nat_proto_find(u_int8_t protonum)
  49 {
  50         return rcu_dereference(nf_nat_protos[protonum]);
  51 }
  52
  53 struct nf_nat_protocol *
  54 nf_nat_proto_find_get(u_int8_t protonum)
  55 {
  56         struct nf_nat_protocol *p;
  57
  58         rcu_read_lock();
  59         p = __nf_nat_proto_find(protonum);
  60         if (!try_module_get(p->me))
  61                 p = &nf_nat_unknown_protocol;
  62         rcu_read_unlock();
  63
  64         return p;
  65 }
  66 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
  67
  68 void
  69 nf_nat_proto_put(struct nf_nat_protocol *p)
  70 {
  71         module_put(p->me);
  72 }
  73 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
  74
  75 /* We keep an extra hash for each conntrack, for fast searching. */
  76 static inline unsigned int
  77 hash_by_src(const struct nf_conntrack_tuple *tuple)
  78 {
  79         /* Original src, to ensure we map it consistently if poss. */
  80         return jhash_3words((__force u32)tuple->src.u3.ip, tuple->src.u.all,
  81                             tuple->dst.protonum, 0) % nf_nat_htable_size;
  82 }
  83
  84 /* Is this tuple already taken? (not by us) */
  85 int
  86 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
  87                   const struct nf_conn *ignored_conntrack)
  88 {
  89         /* Conntrack tracking doesn't keep track of outgoing tuples; only
  90            incoming ones.  NAT means they don't have a fixed mapping,
  91            so we invert the tuple and look for the incoming reply.
  92
  93            We could keep a separate hash if this proves too slow. */
  94         struct nf_conntrack_tuple reply;
  95
  96         nf_ct_invert_tuplepr(&reply, tuple);
  97         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
  98 }
  99 EXPORT_SYMBOL(nf_nat_used_tuple);
 100
 101 /* If we source map this tuple so reply looks like reply_tuple, will
 102  * that meet the constraints of range. */
 103 static int
 104 in_range(const struct nf_conntrack_tuple *tuple,
 105          const struct nf_nat_range *range)
 106 {
 107         struct nf_nat_protocol *proto;
 108         int ret = 0;
 109
 110         /* If we are supposed to map IPs, then we must be in the
 111            range specified, otherwise let this drag us onto a new src IP. */
 112         if (range->flags & IP_NAT_RANGE_MAP_IPS) {
 113                 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
 114                     ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
 115                         return 0;
 116         }
 117
 118         rcu_read_lock();
 119         proto = __nf_nat_proto_find(tuple->dst.protonum);
 120         if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 121             proto->in_range(tuple, IP_NAT_MANIP_SRC,
 122                             &range->min, &range->max))
 123                 ret = 1;
 124         rcu_read_unlock();
 125
 126         return ret;
 127 }
 128
 129 static inline int
 130 same_src(const struct nf_conn *ct,
 131          const struct nf_conntrack_tuple *tuple)
 132 {
 133         const struct nf_conntrack_tuple *t;
 134
 135         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 136         return (t->dst.protonum == tuple->dst.protonum &&
 137                 t->src.u3.ip == tuple->src.u3.ip &&
 138                 t->src.u.all == tuple->src.u.all);
 139 }
 140
 141 /* Only called for SRC manip */
 142 static int
 143 find_appropriate_src(const struct nf_conntrack_tuple *tuple,
 144                      struct nf_conntrack_tuple *result,
 145                      const struct nf_nat_range *range)
 146 {
 147         unsigned int h = hash_by_src(tuple);
 148         struct nf_conn_nat *nat;
 149         struct nf_conn *ct;
 150         struct hlist_node *n;
 151
 152         read_lock_bh(&nf_nat_lock);
 153         hlist_for_each_entry(nat, n, &bysource[h], bysource) {
 154                 ct = nat->ct;
 155                 if (same_src(ct, tuple)) {
 156                         /* Copy source part from reply tuple. */
 157                         nf_ct_invert_tuplepr(result,
 158                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 159                         result->dst = tuple->dst;
 160
 161                         if (in_range(result, range)) {
 162                                 read_unlock_bh(&nf_nat_lock);
 163                                 return 1;
 164                         }
 165                 }
 166         }
 167         read_unlock_bh(&nf_nat_lock);
 168         return 0;
 169 }
 170
 171 /* For [FUTURE] fragmentation handling, we want the least-used
 172    src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 173    if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 174    1-65535, we don't do pro-rata allocation based on ports; we choose
 175    the ip with the lowest src-ip/dst-ip/proto usage.
 176 */
 177 static void
 178 find_best_ips_proto(struct nf_conntrack_tuple *tuple,
 179                     const struct nf_nat_range *range,
 180                     const struct nf_conn *ct,
 181                     enum nf_nat_manip_type maniptype)
 182 {
 183         __be32 *var_ipp;
 184         /* Host order */
 185         u_int32_t minip, maxip, j;
 186
 187         /* No IP mapping?  Do nothing. */
 188         if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
 189                 return;
 190
 191         if (maniptype == IP_NAT_MANIP_SRC)
 192                 var_ipp = &tuple->src.u3.ip;
 193         else
 194                 var_ipp = &tuple->dst.u3.ip;
 195
 196         /* Fast path: only one choice. */
 197         if (range->min_ip == range->max_ip) {
 198                 *var_ipp = range->min_ip;
 199                 return;
 200         }
 201
 202         /* Hashing source and destination IPs gives a fairly even
 203          * spread in practice (if there are a small number of IPs
 204          * involved, there usually aren't that many connections
 205          * anyway).  The consistency means that servers see the same
 206          * client coming from the same IP (some Internet Banking sites
 207          * like this), even across reboots. */
 208         minip = ntohl(range->min_ip);
 209         maxip = ntohl(range->max_ip);
 210         j = jhash_2words((__force u32)tuple->src.u3.ip,
 211                          (__force u32)tuple->dst.u3.ip, 0);
 212         *var_ipp = htonl(minip + j % (maxip - minip + 1));
 213 }
 214
 215 /* Manipulate the tuple into the range given.  For NF_IP_POST_ROUTING,
 216  * we change the source to map into the range.  For NF_IP_PRE_ROUTING
 217  * and NF_IP_LOCAL_OUT, we change the destination to map into the
 218  * range.  It might not be possible to get a unique tuple, but we try.
 219  * At worst (or if we race), we will end up with a final duplicate in
 220  * __ip_conntrack_confirm and drop the packet. */
 221 static void
 222 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 223                  const struct nf_conntrack_tuple *orig_tuple,
 224                  const struct nf_nat_range *range,
 225                  struct nf_conn *ct,
 226                  enum nf_nat_manip_type maniptype)
 227 {
 228         struct nf_nat_protocol *proto;
 229
 230         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 231            and that same mapping gives a unique tuple within the given
 232            range, use that.
 233
 234            This is only required for source (ie. NAT/masq) mappings.
 235            So far, we don't do local source mappings, so multiple
 236            manips not an issue.  */
 237         if (maniptype == IP_NAT_MANIP_SRC) {
 238                 if (find_appropriate_src(orig_tuple, tuple, range)) {
 239                         pr_debug("get_unique_tuple: Found current src map\n");
 240                         if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM))
 241                                 if (!nf_nat_used_tuple(tuple, ct))
 242                                         return;
 243                 }
 244         }
 245
 246         /* 2) Select the least-used IP/proto combination in the given
 247            range. */
 248         *tuple = *orig_tuple;
 249         find_best_ips_proto(tuple, range, ct, maniptype);
 250
 251         /* 3) The per-protocol part of the manip is made to map into
 252            the range to make a unique tuple. */
 253
 254         rcu_read_lock();
 255         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
 256
 257         /* Change protocol info to have some randomization */
 258         if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) {
 259                 proto->unique_tuple(tuple, range, maniptype, ct);
 260                 goto out;
 261         }
 262
 263         /* Only bother mapping if it's not already in range and unique */
 264         if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 265              proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
 266             !nf_nat_used_tuple(tuple, ct))
 267                 goto out;
 268
 269         /* Last change: get protocol to try to obtain unique tuple. */
 270         proto->unique_tuple(tuple, range, maniptype, ct);
 271 out:
 272         rcu_read_unlock();
 273 }
 274
 275 unsigned int
 276 nf_nat_setup_info(struct nf_conn *ct,
 277                   const struct nf_nat_range *range,
 278                   unsigned int hooknum)
 279 {
 280         struct nf_conntrack_tuple curr_tuple, new_tuple;
 281         struct nf_conn_nat *nat;
 282         int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
 283         enum nf_nat_manip_type maniptype = HOOK2MANIP(hooknum);
 284
 285         /* nat helper or nfctnetlink also setup binding */
 286         nat = nfct_nat(ct);
 287         if (!nat) {
 288                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 289                 if (nat == NULL) {
 290                         pr_debug("failed to add NAT extension\n");
 291                         return NF_ACCEPT;
 292                 }
 293         }
 294
 295         NF_CT_ASSERT(hooknum == NF_IP_PRE_ROUTING ||
 296                      hooknum == NF_IP_POST_ROUTING ||
 297                      hooknum == NF_IP_LOCAL_IN ||
 298                      hooknum == NF_IP_LOCAL_OUT);
 299         BUG_ON(nf_nat_initialized(ct, maniptype));
 300
 301         /* What we've got will look like inverse of reply. Normally
 302            this is what is in the conntrack, except for prior
 303            manipulations (future optimization: if num_manips == 0,
 304            orig_tp =
 305            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
 306         nf_ct_invert_tuplepr(&curr_tuple,
 307                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 308
 309         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 310
 311         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 312                 struct nf_conntrack_tuple reply;
 313
 314                 /* Alter conntrack table so will recognize replies. */
 315                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 316                 nf_conntrack_alter_reply(ct, &reply);
 317
 318                 /* Non-atomic: we own this at the moment. */
 319                 if (maniptype == IP_NAT_MANIP_SRC)
 320                         ct->status |= IPS_SRC_NAT;
 321                 else
 322                         ct->status |= IPS_DST_NAT;
 323         }
 324
 325         /* Place in source hash if this is the first time. */
 326         if (have_to_hash) {
 327                 unsigned int srchash;
 328
 329                 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 330                 write_lock_bh(&nf_nat_lock);
 331                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
 332                 nat = nfct_nat(ct);
 333                 nat->ct = ct;
 334                 hlist_add_head(&nat->bysource, &bysource[srchash]);
 335                 write_unlock_bh(&nf_nat_lock);
 336         }
 337
 338         /* It's done. */
 339         if (maniptype == IP_NAT_MANIP_DST)
 340                 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
 341         else
 342                 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 343
 344         return NF_ACCEPT;
 345 }
 346 EXPORT_SYMBOL(nf_nat_setup_info);
 347
 348 /* Returns true if succeeded. */
 349 static int
 350 manip_pkt(u_int16_t proto,
 351           struct sk_buff **pskb,
 352           unsigned int iphdroff,
 353           const struct nf_conntrack_tuple *target,
 354           enum nf_nat_manip_type maniptype)
 355 {
 356         struct iphdr *iph;
 357         struct nf_nat_protocol *p;
 358
 359         if (!skb_make_writable(pskb, iphdroff + sizeof(*iph)))
 360                 return 0;
 361
 362         iph = (void *)(*pskb)->data + iphdroff;
 363
 364         /* Manipulate protcol part. */
 365
 366         /* rcu_read_lock()ed by nf_hook_slow */
 367         p = __nf_nat_proto_find(proto);
 368         if (!p->manip_pkt(pskb, iphdroff, target, maniptype))
 369                 return 0;
 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_l4proto *l4proto;
 427         struct nf_conntrack_tuple inner, target;
 428         int hdrlen = ip_hdrlen(*pskb);
 429         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 430         unsigned long statusbit;
 431         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 432
 433         if (!skb_make_writable(pskb, hdrlen + sizeof(*inside)))
 434                 return 0;
 435
 436         inside = (void *)(*pskb)->data + ip_hdrlen(*pskb);
 437
 438         /* We're actually going to mangle it beyond trivial checksum
 439            adjustment, so make sure the current checksum is correct. */
 440         if (nf_ip_checksum(*pskb, hooknum, hdrlen, 0))
 441                 return 0;
 442
 443         /* Must be RELATED */
 444         NF_CT_ASSERT((*pskb)->nfctinfo == IP_CT_RELATED ||
 445                      (*pskb)->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
 446
 447         /* Redirects on non-null nats must be dropped, else they'll
 448            start talking to each other without our translation, and be
 449            confused... --RR */
 450         if (inside->icmp.type == ICMP_REDIRECT) {
 451                 /* If NAT isn't finished, assume it and drop. */
 452                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 453                         return 0;
 454
 455                 if (ct->status & IPS_NAT_MASK)
 456                         return 0;
 457         }
 458
 459         pr_debug("icmp_reply_translation: translating error %p manip %u "
 460                  "dir %s\n", *pskb, manip,
 461                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 462
 463         /* rcu_read_lock()ed by nf_hook_slow */
 464         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
 465
 466         if (!nf_ct_get_tuple(*pskb,
 467                              ip_hdrlen(*pskb) + sizeof(struct icmphdr),
 468                              (ip_hdrlen(*pskb) +
 469                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 470                              (u_int16_t)AF_INET,
 471                              inside->ip.protocol,
 472                              &inner, l3proto, l4proto))
 473                 return 0;
 474
 475         /* Change inner back to look like incoming packet.  We do the
 476            opposite manip on this hook to normal, because it might not
 477            pass all hooks (locally-generated ICMP).  Consider incoming
 478            packet: PREROUTING (DST manip), routing produces ICMP, goes
 479            through POSTROUTING (which must correct the DST manip). */
 480         if (!manip_pkt(inside->ip.protocol, pskb,
 481                        ip_hdrlen(*pskb) + sizeof(inside->icmp),
 482                        &ct->tuplehash[!dir].tuple,
 483                        !manip))
 484                 return 0;
 485
 486         if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
 487                 /* Reloading "inside" here since manip_pkt inner. */
 488                 inside = (void *)(*pskb)->data + ip_hdrlen(*pskb);
 489                 inside->icmp.checksum = 0;
 490                 inside->icmp.checksum =
 491                         csum_fold(skb_checksum(*pskb, hdrlen,
 492                                                (*pskb)->len - hdrlen, 0));
 493         }
 494
 495         /* Change outer to look the reply to an incoming packet
 496          * (proto 0 means don't invert per-proto part). */
 497         if (manip == IP_NAT_MANIP_SRC)
 498                 statusbit = IPS_SRC_NAT;
 499         else
 500                 statusbit = IPS_DST_NAT;
 501
 502         /* Invert if this is reply dir. */
 503         if (dir == IP_CT_DIR_REPLY)
 504                 statusbit ^= IPS_NAT_MASK;
 505
 506         if (ct->status & statusbit) {
 507                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 508                 if (!manip_pkt(0, pskb, 0, &target, manip))
 509                         return 0;
 510         }
 511
 512         return 1;
 513 }
 514 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 515
 516 /* Protocol registration. */
 517 int nf_nat_protocol_register(struct nf_nat_protocol *proto)
 518 {
 519         int ret = 0;
 520
 521         write_lock_bh(&nf_nat_lock);
 522         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
 523                 ret = -EBUSY;
 524                 goto out;
 525         }
 526         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 527  out:
 528         write_unlock_bh(&nf_nat_lock);
 529         return ret;
 530 }
 531 EXPORT_SYMBOL(nf_nat_protocol_register);
 532
 533 /* Noone stores the protocol anywhere; simply delete it. */
 534 void nf_nat_protocol_unregister(struct nf_nat_protocol *proto)
 535 {
 536         write_lock_bh(&nf_nat_lock);
 537         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 538                            &nf_nat_unknown_protocol);
 539         write_unlock_bh(&nf_nat_lock);
 540         synchronize_rcu();
 541 }
 542 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 543
 544 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 545 int
 546 nf_nat_port_range_to_nfattr(struct sk_buff *skb,
 547                             const struct nf_nat_range *range)
 548 {
 549         NFA_PUT(skb, CTA_PROTONAT_PORT_MIN, sizeof(__be16),
 550                 &range->min.tcp.port);
 551         NFA_PUT(skb, CTA_PROTONAT_PORT_MAX, sizeof(__be16),
 552                 &range->max.tcp.port);
 553
 554         return 0;
 555
 556 nfattr_failure:
 557         return -1;
 558 }
 559 EXPORT_SYMBOL_GPL(nf_nat_port_nfattr_to_range);
 560
 561 int
 562 nf_nat_port_nfattr_to_range(struct nfattr *tb[], struct nf_nat_range *range)
 563 {
 564         int ret = 0;
 565
 566         /* we have to return whether we actually parsed something or not */
 567
 568         if (tb[CTA_PROTONAT_PORT_MIN-1]) {
 569                 ret = 1;
 570                 range->min.tcp.port =
 571                         *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MIN-1]);
 572         }
 573
 574         if (!tb[CTA_PROTONAT_PORT_MAX-1]) {
 575                 if (ret)
 576                         range->max.tcp.port = range->min.tcp.port;
 577         } else {
 578                 ret = 1;
 579                 range->max.tcp.port =
 580                         *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MAX-1]);
 581         }
 582
 583         return ret;
 584 }
 585 EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nfattr);
 586 #endif
 587
 588 /* Noone using conntrack by the time this called. */
 589 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 590 {
 591         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 592
 593         if (nat == NULL || nat->ct == NULL)
 594                 return;
 595
 596         NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
 597
 598         write_lock_bh(&nf_nat_lock);
 599         hlist_del(&nat->bysource);
 600         nat->ct = NULL;
 601         write_unlock_bh(&nf_nat_lock);
 602 }
 603
 604 static void nf_nat_move_storage(struct nf_conn *conntrack, void *old)
 605 {
 606         struct nf_conn_nat *new_nat = nf_ct_ext_find(conntrack, NF_CT_EXT_NAT);
 607         struct nf_conn_nat *old_nat = (struct nf_conn_nat *)old;
 608         struct nf_conn *ct = old_nat->ct;
 609         unsigned int srchash;
 610
 611         if (!(ct->status & IPS_NAT_DONE_MASK))
 612                 return;
 613
 614         srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 615
 616         write_lock_bh(&nf_nat_lock);
 617         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 618         new_nat->ct = ct;
 619         write_unlock_bh(&nf_nat_lock);
 620 }
 621
 622 static struct nf_ct_ext_type nat_extend __read_mostly = {
 623         .len            = sizeof(struct nf_conn_nat),
 624         .align          = __alignof__(struct nf_conn_nat),
 625         .destroy        = nf_nat_cleanup_conntrack,
 626         .move           = nf_nat_move_storage,
 627         .id             = NF_CT_EXT_NAT,
 628         .flags          = NF_CT_EXT_F_PREALLOC,
 629 };
 630
 631 static int __init nf_nat_init(void)
 632 {
 633         size_t i;
 634         int ret;
 635
 636         ret = nf_ct_extend_register(&nat_extend);
 637         if (ret < 0) {
 638                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 639                 return ret;
 640         }
 641
 642         /* Leave them the same for the moment. */
 643         nf_nat_htable_size = nf_conntrack_htable_size;
 644
 645         bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
 646                                          &nf_nat_vmalloced);
 647         if (!bysource) {
 648                 ret = -ENOMEM;
 649                 goto cleanup_extend;
 650         }
 651
 652         /* Sew in builtin protocols. */
 653         write_lock_bh(&nf_nat_lock);
 654         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 655                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
 656         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 657         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 658         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 659         write_unlock_bh(&nf_nat_lock);
 660
 661         for (i = 0; i < nf_nat_htable_size; i++) {
 662                 INIT_HLIST_HEAD(&bysource[i]);
 663         }
 664
 665         /* Initialize fake conntrack so that NAT will skip it */
 666         nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
 667
 668         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 669         return 0;
 670
 671  cleanup_extend:
 672         nf_ct_extend_unregister(&nat_extend);
 673         return ret;
 674 }
 675
 676 /* Clear NAT section of all conntracks, in case we're loaded again. */
 677 static int clean_nat(struct nf_conn *i, void *data)
 678 {
 679         struct nf_conn_nat *nat = nfct_nat(i);
 680
 681         if (!nat)
 682                 return 0;
 683         memset(nat, 0, sizeof(nat));
 684         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
 685         return 0;
 686 }
 687
 688 static void __exit nf_nat_cleanup(void)
 689 {
 690         nf_ct_iterate_cleanup(&clean_nat, NULL);
 691         synchronize_rcu();
 692         nf_ct_free_hashtable(bysource, nf_nat_vmalloced, nf_nat_htable_size);
 693         nf_ct_l3proto_put(l3proto);
 694         nf_ct_extend_unregister(&nat_extend);
 695 }
 696
 697 MODULE_LICENSE("GPL");
 698
 699 module_init(nf_nat_init);
 700 module_exit(nf_nat_cleanup);