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_SPINLOCK(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
  41 #define MAX_IP_NAT_PROTO 256
  42 static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
  43                                                 __read_mostly;
  44
  45 static inline const struct nf_nat_protocol *
  46 __nf_nat_proto_find(u_int8_t protonum)
  47 {
  48         return rcu_dereference(nf_nat_protos[protonum]);
  49 }
  50
  51 const struct nf_nat_protocol *
  52 nf_nat_proto_find_get(u_int8_t protonum)
  53 {
  54         const struct nf_nat_protocol *p;
  55
  56         rcu_read_lock();
  57         p = __nf_nat_proto_find(protonum);
  58         if (!try_module_get(p->me))
  59                 p = &nf_nat_unknown_protocol;
  60         rcu_read_unlock();
  61
  62         return p;
  63 }
  64 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
  65
  66 void
  67 nf_nat_proto_put(const struct nf_nat_protocol *p)
  68 {
  69         module_put(p->me);
  70 }
  71 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
  72
  73 /* We keep an extra hash for each conntrack, for fast searching. */
  74 static inline unsigned int
  75 hash_by_src(const struct nf_conntrack_tuple *tuple)
  76 {
  77         unsigned int hash;
  78
  79         /* Original src, to ensure we map it consistently if poss. */
  80         hash = jhash_3words((__force u32)tuple->src.u3.ip,
  81                             (__force u32)tuple->src.u.all,
  82                             tuple->dst.protonum, 0);
  83         return ((u64)hash * nf_nat_htable_size) >> 32;
  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(struct net *net,
 146                      const struct nf_conntrack_tuple *tuple,
 147                      struct nf_conntrack_tuple *result,
 148                      const struct nf_nat_range *range)
 149 {
 150         unsigned int h = hash_by_src(tuple);
 151         const struct nf_conn_nat *nat;
 152         const struct nf_conn *ct;
 153         const struct hlist_node *n;
 154
 155         rcu_read_lock();
 156         hlist_for_each_entry_rcu(nat, n, &net->ipv4.nat_bysource[h], bysource) {
 157                 ct = nat->ct;
 158                 if (same_src(ct, tuple)) {
 159                         /* Copy source part from reply tuple. */
 160                         nf_ct_invert_tuplepr(result,
 161                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 162                         result->dst = tuple->dst;
 163
 164                         if (in_range(result, range)) {
 165                                 rcu_read_unlock();
 166                                 return 1;
 167                         }
 168                 }
 169         }
 170         rcu_read_unlock();
 171         return 0;
 172 }
 173
 174 /* For [FUTURE] fragmentation handling, we want the least-used
 175    src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 176    if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 177    1-65535, we don't do pro-rata allocation based on ports; we choose
 178    the ip with the lowest src-ip/dst-ip/proto usage.
 179 */
 180 static void
 181 find_best_ips_proto(struct nf_conntrack_tuple *tuple,
 182                     const struct nf_nat_range *range,
 183                     const struct nf_conn *ct,
 184                     enum nf_nat_manip_type maniptype)
 185 {
 186         __be32 *var_ipp;
 187         /* Host order */
 188         u_int32_t minip, maxip, j;
 189
 190         /* No IP mapping?  Do nothing. */
 191         if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
 192                 return;
 193
 194         if (maniptype == IP_NAT_MANIP_SRC)
 195                 var_ipp = &tuple->src.u3.ip;
 196         else
 197                 var_ipp = &tuple->dst.u3.ip;
 198
 199         /* Fast path: only one choice. */
 200         if (range->min_ip == range->max_ip) {
 201                 *var_ipp = range->min_ip;
 202                 return;
 203         }
 204
 205         /* Hashing source and destination IPs gives a fairly even
 206          * spread in practice (if there are a small number of IPs
 207          * involved, there usually aren't that many connections
 208          * anyway).  The consistency means that servers see the same
 209          * client coming from the same IP (some Internet Banking sites
 210          * like this), even across reboots. */
 211         minip = ntohl(range->min_ip);
 212         maxip = ntohl(range->max_ip);
 213         j = jhash_2words((__force u32)tuple->src.u3.ip,
 214                          (__force u32)tuple->dst.u3.ip, 0);
 215         j = ((u64)j * (maxip - minip + 1)) >> 32;
 216         *var_ipp = htonl(minip + j);
 217 }
 218
 219 /* Manipulate the tuple into the range given.  For NF_INET_POST_ROUTING,
 220  * we change the source to map into the range.  For NF_INET_PRE_ROUTING
 221  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 222  * range.  It might not be possible to get a unique tuple, but we try.
 223  * At worst (or if we race), we will end up with a final duplicate in
 224  * __ip_conntrack_confirm and drop the packet. */
 225 static void
 226 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 227                  const struct nf_conntrack_tuple *orig_tuple,
 228                  const struct nf_nat_range *range,
 229                  struct nf_conn *ct,
 230                  enum nf_nat_manip_type maniptype)
 231 {
 232         struct net *net = nf_ct_net(ct);
 233         const 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             !(range->flags & IP_NAT_RANGE_PROTO_RANDOM)) {
 244                 if (find_appropriate_src(net, orig_tuple, tuple, range)) {
 245                         pr_debug("get_unique_tuple: Found current src map\n");
 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                   enum nf_nat_manip_type maniptype)
 284 {
 285         struct net *net = nf_ct_net(ct);
 286         struct nf_conntrack_tuple curr_tuple, new_tuple;
 287         struct nf_conn_nat *nat;
 288         int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
 289
 290         /* nat helper or nfctnetlink also setup binding */
 291         nat = nfct_nat(ct);
 292         if (!nat) {
 293                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 294                 if (nat == NULL) {
 295                         pr_debug("failed to add NAT extension\n");
 296                         return NF_ACCEPT;
 297                 }
 298         }
 299
 300         NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC ||
 301                      maniptype == IP_NAT_MANIP_DST);
 302         BUG_ON(nf_nat_initialized(ct, maniptype));
 303
 304         /* What we've got will look like inverse of reply. Normally
 305            this is what is in the conntrack, except for prior
 306            manipulations (future optimization: if num_manips == 0,
 307            orig_tp =
 308            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
 309         nf_ct_invert_tuplepr(&curr_tuple,
 310                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 311
 312         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 313
 314         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 315                 struct nf_conntrack_tuple reply;
 316
 317                 /* Alter conntrack table so will recognize replies. */
 318                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 319                 nf_conntrack_alter_reply(ct, &reply);
 320
 321                 /* Non-atomic: we own this at the moment. */
 322                 if (maniptype == IP_NAT_MANIP_SRC)
 323                         ct->status |= IPS_SRC_NAT;
 324                 else
 325                         ct->status |= IPS_DST_NAT;
 326         }
 327
 328         /* Place in source hash if this is the first time. */
 329         if (have_to_hash) {
 330                 unsigned int srchash;
 331
 332                 srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 333                 spin_lock_bh(&nf_nat_lock);
 334                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
 335                 nat = nfct_nat(ct);
 336                 nat->ct = ct;
 337                 hlist_add_head_rcu(&nat->bysource,
 338                                    &net->ipv4.nat_bysource[srchash]);
 339                 spin_unlock_bh(&nf_nat_lock);
 340         }
 341
 342         /* It's done. */
 343         if (maniptype == IP_NAT_MANIP_DST)
 344                 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
 345         else
 346                 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 347
 348         return NF_ACCEPT;
 349 }
 350 EXPORT_SYMBOL(nf_nat_setup_info);
 351
 352 /* Returns true if succeeded. */
 353 static bool
 354 manip_pkt(u_int16_t proto,
 355           struct sk_buff *skb,
 356           unsigned int iphdroff,
 357           const struct nf_conntrack_tuple *target,
 358           enum nf_nat_manip_type maniptype)
 359 {
 360         struct iphdr *iph;
 361         const struct nf_nat_protocol *p;
 362
 363         if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
 364                 return false;
 365
 366         iph = (void *)skb->data + iphdroff;
 367
 368         /* Manipulate protcol part. */
 369
 370         /* rcu_read_lock()ed by nf_hook_slow */
 371         p = __nf_nat_proto_find(proto);
 372         if (!p->manip_pkt(skb, iphdroff, target, maniptype))
 373                 return false;
 374
 375         iph = (void *)skb->data + iphdroff;
 376
 377         if (maniptype == IP_NAT_MANIP_SRC) {
 378                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 379                 iph->saddr = target->src.u3.ip;
 380         } else {
 381                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 382                 iph->daddr = target->dst.u3.ip;
 383         }
 384         return true;
 385 }
 386
 387 /* Do packet manipulations according to nf_nat_setup_info. */
 388 unsigned int nf_nat_packet(struct nf_conn *ct,
 389                            enum ip_conntrack_info ctinfo,
 390                            unsigned int hooknum,
 391                            struct sk_buff *skb)
 392 {
 393         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 394         unsigned long statusbit;
 395         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 396
 397         if (mtype == IP_NAT_MANIP_SRC)
 398                 statusbit = IPS_SRC_NAT;
 399         else
 400                 statusbit = IPS_DST_NAT;
 401
 402         /* Invert if this is reply dir. */
 403         if (dir == IP_CT_DIR_REPLY)
 404                 statusbit ^= IPS_NAT_MASK;
 405
 406         /* Non-atomic: these bits don't change. */
 407         if (ct->status & statusbit) {
 408                 struct nf_conntrack_tuple target;
 409
 410                 /* We are aiming to look like inverse of other direction. */
 411                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 412
 413                 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
 414                         return NF_DROP;
 415         }
 416         return NF_ACCEPT;
 417 }
 418 EXPORT_SYMBOL_GPL(nf_nat_packet);
 419
 420 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
 421 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 422                                   enum ip_conntrack_info ctinfo,
 423                                   unsigned int hooknum,
 424                                   struct sk_buff *skb)
 425 {
 426         struct {
 427                 struct icmphdr icmp;
 428                 struct iphdr ip;
 429         } *inside;
 430         const struct nf_conntrack_l4proto *l4proto;
 431         struct nf_conntrack_tuple inner, target;
 432         int hdrlen = ip_hdrlen(skb);
 433         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 434         unsigned long statusbit;
 435         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 436
 437         if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
 438                 return 0;
 439
 440         inside = (void *)skb->data + ip_hdrlen(skb);
 441
 442         /* We're actually going to mangle it beyond trivial checksum
 443            adjustment, so make sure the current checksum is correct. */
 444         if (nf_ip_checksum(skb, hooknum, hdrlen, 0))
 445                 return 0;
 446
 447         /* Must be RELATED */
 448         NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
 449                      skb->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
 450
 451         /* Redirects on non-null nats must be dropped, else they'll
 452            start talking to each other without our translation, and be
 453            confused... --RR */
 454         if (inside->icmp.type == ICMP_REDIRECT) {
 455                 /* If NAT isn't finished, assume it and drop. */
 456                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 457                         return 0;
 458
 459                 if (ct->status & IPS_NAT_MASK)
 460                         return 0;
 461         }
 462
 463         pr_debug("icmp_reply_translation: translating error %p manip %u "
 464                  "dir %s\n", skb, manip,
 465                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 466
 467         /* rcu_read_lock()ed by nf_hook_slow */
 468         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
 469
 470         if (!nf_ct_get_tuple(skb,
 471                              ip_hdrlen(skb) + sizeof(struct icmphdr),
 472                              (ip_hdrlen(skb) +
 473                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 474                              (u_int16_t)AF_INET,
 475                              inside->ip.protocol,
 476                              &inner, l3proto, l4proto))
 477                 return 0;
 478
 479         /* Change inner back to look like incoming packet.  We do the
 480            opposite manip on this hook to normal, because it might not
 481            pass all hooks (locally-generated ICMP).  Consider incoming
 482            packet: PREROUTING (DST manip), routing produces ICMP, goes
 483            through POSTROUTING (which must correct the DST manip). */
 484         if (!manip_pkt(inside->ip.protocol, skb,
 485                        ip_hdrlen(skb) + sizeof(inside->icmp),
 486                        &ct->tuplehash[!dir].tuple,
 487                        !manip))
 488                 return 0;
 489
 490         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 491                 /* Reloading "inside" here since manip_pkt inner. */
 492                 inside = (void *)skb->data + ip_hdrlen(skb);
 493                 inside->icmp.checksum = 0;
 494                 inside->icmp.checksum =
 495                         csum_fold(skb_checksum(skb, hdrlen,
 496                                                skb->len - hdrlen, 0));
 497         }
 498
 499         /* Change outer to look the reply to an incoming packet
 500          * (proto 0 means don't invert per-proto part). */
 501         if (manip == IP_NAT_MANIP_SRC)
 502                 statusbit = IPS_SRC_NAT;
 503         else
 504                 statusbit = IPS_DST_NAT;
 505
 506         /* Invert if this is reply dir. */
 507         if (dir == IP_CT_DIR_REPLY)
 508                 statusbit ^= IPS_NAT_MASK;
 509
 510         if (ct->status & statusbit) {
 511                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 512                 if (!manip_pkt(0, skb, 0, &target, manip))
 513                         return 0;
 514         }
 515
 516         return 1;
 517 }
 518 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 519
 520 /* Protocol registration. */
 521 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 522 {
 523         int ret = 0;
 524
 525         spin_lock_bh(&nf_nat_lock);
 526         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
 527                 ret = -EBUSY;
 528                 goto out;
 529         }
 530         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 531  out:
 532         spin_unlock_bh(&nf_nat_lock);
 533         return ret;
 534 }
 535 EXPORT_SYMBOL(nf_nat_protocol_register);
 536
 537 /* Noone stores the protocol anywhere; simply delete it. */
 538 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 539 {
 540         spin_lock_bh(&nf_nat_lock);
 541         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 542                            &nf_nat_unknown_protocol);
 543         spin_unlock_bh(&nf_nat_lock);
 544         synchronize_rcu();
 545 }
 546 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 547
 548 /* Noone using conntrack by the time this called. */
 549 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 550 {
 551         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 552
 553         if (nat == NULL || nat->ct == NULL)
 554                 return;
 555
 556         NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
 557
 558         spin_lock_bh(&nf_nat_lock);
 559         hlist_del_rcu(&nat->bysource);
 560         spin_unlock_bh(&nf_nat_lock);
 561 }
 562
 563 static void nf_nat_move_storage(void *new, void *old)
 564 {
 565         struct nf_conn_nat *new_nat = new;
 566         struct nf_conn_nat *old_nat = old;
 567         struct nf_conn *ct = old_nat->ct;
 568
 569         if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
 570                 return;
 571
 572         spin_lock_bh(&nf_nat_lock);
 573         new_nat->ct = ct;
 574         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 575         spin_unlock_bh(&nf_nat_lock);
 576 }
 577
 578 static struct nf_ct_ext_type nat_extend __read_mostly = {
 579         .len            = sizeof(struct nf_conn_nat),
 580         .align          = __alignof__(struct nf_conn_nat),
 581         .destroy        = nf_nat_cleanup_conntrack,
 582         .move           = nf_nat_move_storage,
 583         .id             = NF_CT_EXT_NAT,
 584         .flags          = NF_CT_EXT_F_PREALLOC,
 585 };
 586
 587 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 588
 589 #include <linux/netfilter/nfnetlink.h>
 590 #include <linux/netfilter/nfnetlink_conntrack.h>
 591
 592 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 593         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 594         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 595 };
 596
 597 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 598                                      const struct nf_conn *ct,
 599                                      struct nf_nat_range *range)
 600 {
 601         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 602         const struct nf_nat_protocol *npt;
 603         int err;
 604
 605         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 606         if (err < 0)
 607                 return err;
 608
 609         npt = nf_nat_proto_find_get(nf_ct_protonum(ct));
 610         if (npt->nlattr_to_range)
 611                 err = npt->nlattr_to_range(tb, range);
 612         nf_nat_proto_put(npt);
 613         return err;
 614 }
 615
 616 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 617         [CTA_NAT_MINIP]         = { .type = NLA_U32 },
 618         [CTA_NAT_MAXIP]         = { .type = NLA_U32 },
 619 };
 620
 621 static int
 622 nfnetlink_parse_nat(struct nlattr *nat,
 623                     const struct nf_conn *ct, struct nf_nat_range *range)
 624 {
 625         struct nlattr *tb[CTA_NAT_MAX+1];
 626         int err;
 627
 628         memset(range, 0, sizeof(*range));
 629
 630         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 631         if (err < 0)
 632                 return err;
 633
 634         if (tb[CTA_NAT_MINIP])
 635                 range->min_ip = nla_get_be32(tb[CTA_NAT_MINIP]);
 636
 637         if (!tb[CTA_NAT_MAXIP])
 638                 range->max_ip = range->min_ip;
 639         else
 640                 range->max_ip = nla_get_be32(tb[CTA_NAT_MAXIP]);
 641
 642         if (range->min_ip)
 643                 range->flags |= IP_NAT_RANGE_MAP_IPS;
 644
 645         if (!tb[CTA_NAT_PROTO])
 646                 return 0;
 647
 648         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 649         if (err < 0)
 650                 return err;
 651
 652         return 0;
 653 }
 654
 655 static int
 656 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 657                           enum nf_nat_manip_type manip,
 658                           struct nlattr *attr)
 659 {
 660         struct nf_nat_range range;
 661
 662         if (nfnetlink_parse_nat(attr, ct, &range) < 0)
 663                 return -EINVAL;
 664         if (nf_nat_initialized(ct, manip))
 665                 return -EEXIST;
 666
 667         return nf_nat_setup_info(ct, &range, manip);
 668 }
 669 #else
 670 static int
 671 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 672                           enum nf_nat_manip_type manip,
 673                           struct nlattr *attr)
 674 {
 675         return -EOPNOTSUPP;
 676 }
 677 #endif
 678
 679 static int __net_init nf_nat_net_init(struct net *net)
 680 {
 681         net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
 682                                                       &net->ipv4.nat_vmalloced);
 683         if (!net->ipv4.nat_bysource)
 684                 return -ENOMEM;
 685         return 0;
 686 }
 687
 688 /* Clear NAT section of all conntracks, in case we're loaded again. */
 689 static int clean_nat(struct nf_conn *i, void *data)
 690 {
 691         struct nf_conn_nat *nat = nfct_nat(i);
 692
 693         if (!nat)
 694                 return 0;
 695         memset(nat, 0, sizeof(*nat));
 696         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
 697         return 0;
 698 }
 699
 700 static void __net_exit nf_nat_net_exit(struct net *net)
 701 {
 702         nf_ct_iterate_cleanup(net, &clean_nat, NULL);
 703         synchronize_rcu();
 704         nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_vmalloced,
 705                              nf_nat_htable_size);
 706 }
 707
 708 static struct pernet_operations nf_nat_net_ops = {
 709         .init = nf_nat_net_init,
 710         .exit = nf_nat_net_exit,
 711 };
 712
 713 static int __init nf_nat_init(void)
 714 {
 715         size_t i;
 716         int ret;
 717
 718         need_ipv4_conntrack();
 719
 720         ret = nf_ct_extend_register(&nat_extend);
 721         if (ret < 0) {
 722                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 723                 return ret;
 724         }
 725
 726         /* Leave them the same for the moment. */
 727         nf_nat_htable_size = nf_conntrack_htable_size;
 728
 729         ret = register_pernet_subsys(&nf_nat_net_ops);
 730         if (ret < 0)
 731                 goto cleanup_extend;
 732
 733         /* Sew in builtin protocols. */
 734         spin_lock_bh(&nf_nat_lock);
 735         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 736                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
 737         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 738         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 739         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 740         spin_unlock_bh(&nf_nat_lock);
 741
 742         /* Initialize fake conntrack so that NAT will skip it */
 743         nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
 744
 745         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 746
 747         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 748         rcu_assign_pointer(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 749         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 750         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook,
 751                            nfnetlink_parse_nat_setup);
 752         return 0;
 753
 754  cleanup_extend:
 755         nf_ct_extend_unregister(&nat_extend);
 756         return ret;
 757 }
 758
 759 static void __exit nf_nat_cleanup(void)
 760 {
 761         unregister_pernet_subsys(&nf_nat_net_ops);
 762         nf_ct_l3proto_put(l3proto);
 763         nf_ct_extend_unregister(&nat_extend);
 764         rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL);
 765         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, NULL);
 766         synchronize_net();
 767 }
 768
 769 MODULE_LICENSE("GPL");
 770 MODULE_ALIAS("nf-nat-ipv4");
 771
 772 module_init(nf_nat_init);
 773 module_exit(nf_nat_cleanup);