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 #define MAX_IP_NAT_PROTO 256
  39 static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
  40                                                 __read_mostly;
  41
  42 static inline const struct nf_nat_protocol *
  43 __nf_nat_proto_find(u_int8_t protonum)
  44 {
  45         return rcu_dereference(nf_nat_protos[protonum]);
  46 }
  47
  48 const struct nf_nat_protocol *
  49 nf_nat_proto_find_get(u_int8_t protonum)
  50 {
  51         const struct nf_nat_protocol *p;
  52
  53         rcu_read_lock();
  54         p = __nf_nat_proto_find(protonum);
  55         if (!try_module_get(p->me))
  56                 p = &nf_nat_unknown_protocol;
  57         rcu_read_unlock();
  58
  59         return p;
  60 }
  61 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
  62
  63 void
  64 nf_nat_proto_put(const struct nf_nat_protocol *p)
  65 {
  66         module_put(p->me);
  67 }
  68 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
  69
  70 /* We keep an extra hash for each conntrack, for fast searching. */
  71 static inline unsigned int
  72 hash_by_src(const struct net *net, const struct nf_conntrack_tuple *tuple)
  73 {
  74         unsigned int hash;
  75
  76         /* Original src, to ensure we map it consistently if poss. */
  77         hash = jhash_3words((__force u32)tuple->src.u3.ip,
  78                             (__force u32)tuple->src.u.all,
  79                             tuple->dst.protonum, 0);
  80         return ((u64)hash * net->ipv4.nat_htable_size) >> 32;
  81 }
  82
  83 /* Is this tuple already taken? (not by us) */
  84 int
  85 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
  86                   const struct nf_conn *ignored_conntrack)
  87 {
  88         /* Conntrack tracking doesn't keep track of outgoing tuples; only
  89            incoming ones.  NAT means they don't have a fixed mapping,
  90            so we invert the tuple and look for the incoming reply.
  91
  92            We could keep a separate hash if this proves too slow. */
  93         struct nf_conntrack_tuple reply;
  94
  95         nf_ct_invert_tuplepr(&reply, tuple);
  96         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
  97 }
  98 EXPORT_SYMBOL(nf_nat_used_tuple);
  99
 100 /* If we source map this tuple so reply looks like reply_tuple, will
 101  * that meet the constraints of range. */
 102 static int
 103 in_range(const struct nf_conntrack_tuple *tuple,
 104          const struct nf_nat_range *range)
 105 {
 106         const struct nf_nat_protocol *proto;
 107         int ret = 0;
 108
 109         /* If we are supposed to map IPs, then we must be in the
 110            range specified, otherwise let this drag us onto a new src IP. */
 111         if (range->flags & IP_NAT_RANGE_MAP_IPS) {
 112                 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
 113                     ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
 114                         return 0;
 115         }
 116
 117         rcu_read_lock();
 118         proto = __nf_nat_proto_find(tuple->dst.protonum);
 119         if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 120             proto->in_range(tuple, IP_NAT_MANIP_SRC,
 121                             &range->min, &range->max))
 122                 ret = 1;
 123         rcu_read_unlock();
 124
 125         return ret;
 126 }
 127
 128 static inline int
 129 same_src(const struct nf_conn *ct,
 130          const struct nf_conntrack_tuple *tuple)
 131 {
 132         const struct nf_conntrack_tuple *t;
 133
 134         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 135         return (t->dst.protonum == tuple->dst.protonum &&
 136                 t->src.u3.ip == tuple->src.u3.ip &&
 137                 t->src.u.all == tuple->src.u.all);
 138 }
 139
 140 /* Only called for SRC manip */
 141 static int
 142 find_appropriate_src(struct net *net,
 143                      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(net, tuple);
 148         const struct nf_conn_nat *nat;
 149         const struct nf_conn *ct;
 150         const struct hlist_node *n;
 151
 152         rcu_read_lock();
 153         hlist_for_each_entry_rcu(nat, n, &net->ipv4.nat_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                                 rcu_read_unlock();
 163                                 return 1;
 164                         }
 165                 }
 166         }
 167         rcu_read_unlock();
 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                          range->flags & IP_NAT_RANGE_PERSISTENT ?
 212                                 0 : (__force u32)tuple->dst.u3.ip, 0);
 213         j = ((u64)j * (maxip - minip + 1)) >> 32;
 214         *var_ipp = htonl(minip + j);
 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         struct net *net = nf_ct_net(ct);
 231         const struct nf_nat_protocol *proto;
 232
 233         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 234            and that same mapping gives a unique tuple within the given
 235            range, use that.
 236
 237            This is only required for source (ie. NAT/masq) mappings.
 238            So far, we don't do local source mappings, so multiple
 239            manips not an issue.  */
 240         if (maniptype == IP_NAT_MANIP_SRC &&
 241             !(range->flags & IP_NAT_RANGE_PROTO_RANDOM)) {
 242                 if (find_appropriate_src(net, orig_tuple, tuple, range)) {
 243                         pr_debug("get_unique_tuple: Found current src map\n");
 244                         if (!nf_nat_used_tuple(tuple, ct))
 245                                 return;
 246                 }
 247         }
 248
 249         /* 2) Select the least-used IP/proto combination in the given
 250            range. */
 251         *tuple = *orig_tuple;
 252         find_best_ips_proto(tuple, range, ct, maniptype);
 253
 254         /* 3) The per-protocol part of the manip is made to map into
 255            the range to make a unique tuple. */
 256
 257         rcu_read_lock();
 258         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
 259
 260         /* Change protocol info to have some randomization */
 261         if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) {
 262                 proto->unique_tuple(tuple, range, maniptype, ct);
 263                 goto out;
 264         }
 265
 266         /* Only bother mapping if it's not already in range and unique */
 267         if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 268              proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
 269             !nf_nat_used_tuple(tuple, ct))
 270                 goto out;
 271
 272         /* Last change: get protocol to try to obtain unique tuple. */
 273         proto->unique_tuple(tuple, range, maniptype, ct);
 274 out:
 275         rcu_read_unlock();
 276 }
 277
 278 unsigned int
 279 nf_nat_setup_info(struct nf_conn *ct,
 280                   const struct nf_nat_range *range,
 281                   enum nf_nat_manip_type maniptype)
 282 {
 283         struct net *net = nf_ct_net(ct);
 284         struct nf_conntrack_tuple curr_tuple, new_tuple;
 285         struct nf_conn_nat *nat;
 286         int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
 287
 288         /* nat helper or nfctnetlink also setup binding */
 289         nat = nfct_nat(ct);
 290         if (!nat) {
 291                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 292                 if (nat == NULL) {
 293                         pr_debug("failed to add NAT extension\n");
 294                         return NF_ACCEPT;
 295                 }
 296         }
 297
 298         NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC ||
 299                      maniptype == IP_NAT_MANIP_DST);
 300         BUG_ON(nf_nat_initialized(ct, maniptype));
 301
 302         /* What we've got will look like inverse of reply. Normally
 303            this is what is in the conntrack, except for prior
 304            manipulations (future optimization: if num_manips == 0,
 305            orig_tp =
 306            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
 307         nf_ct_invert_tuplepr(&curr_tuple,
 308                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 309
 310         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 311
 312         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 313                 struct nf_conntrack_tuple reply;
 314
 315                 /* Alter conntrack table so will recognize replies. */
 316                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 317                 nf_conntrack_alter_reply(ct, &reply);
 318
 319                 /* Non-atomic: we own this at the moment. */
 320                 if (maniptype == IP_NAT_MANIP_SRC)
 321                         ct->status |= IPS_SRC_NAT;
 322                 else
 323                         ct->status |= IPS_DST_NAT;
 324         }
 325
 326         /* Place in source hash if this is the first time. */
 327         if (have_to_hash) {
 328                 unsigned int srchash;
 329
 330                 srchash = hash_by_src(net, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 331                 spin_lock_bh(&nf_nat_lock);
 332                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
 333                 nat = nfct_nat(ct);
 334                 nat->ct = ct;
 335                 hlist_add_head_rcu(&nat->bysource,
 336                                    &net->ipv4.nat_bysource[srchash]);
 337                 spin_unlock_bh(&nf_nat_lock);
 338         }
 339
 340         /* It's done. */
 341         if (maniptype == IP_NAT_MANIP_DST)
 342                 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
 343         else
 344                 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 345
 346         return NF_ACCEPT;
 347 }
 348 EXPORT_SYMBOL(nf_nat_setup_info);
 349
 350 /* Returns true if succeeded. */
 351 static bool
 352 manip_pkt(u_int16_t proto,
 353           struct sk_buff *skb,
 354           unsigned int iphdroff,
 355           const struct nf_conntrack_tuple *target,
 356           enum nf_nat_manip_type maniptype)
 357 {
 358         struct iphdr *iph;
 359         const struct nf_nat_protocol *p;
 360
 361         if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
 362                 return false;
 363
 364         iph = (void *)skb->data + iphdroff;
 365
 366         /* Manipulate protcol part. */
 367
 368         /* rcu_read_lock()ed by nf_hook_slow */
 369         p = __nf_nat_proto_find(proto);
 370         if (!p->manip_pkt(skb, iphdroff, target, maniptype))
 371                 return false;
 372
 373         iph = (void *)skb->data + iphdroff;
 374
 375         if (maniptype == IP_NAT_MANIP_SRC) {
 376                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 377                 iph->saddr = target->src.u3.ip;
 378         } else {
 379                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 380                 iph->daddr = target->dst.u3.ip;
 381         }
 382         return true;
 383 }
 384
 385 /* Do packet manipulations according to nf_nat_setup_info. */
 386 unsigned int nf_nat_packet(struct nf_conn *ct,
 387                            enum ip_conntrack_info ctinfo,
 388                            unsigned int hooknum,
 389                            struct sk_buff *skb)
 390 {
 391         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 392         unsigned long statusbit;
 393         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 394
 395         if (mtype == IP_NAT_MANIP_SRC)
 396                 statusbit = IPS_SRC_NAT;
 397         else
 398                 statusbit = IPS_DST_NAT;
 399
 400         /* Invert if this is reply dir. */
 401         if (dir == IP_CT_DIR_REPLY)
 402                 statusbit ^= IPS_NAT_MASK;
 403
 404         /* Non-atomic: these bits don't change. */
 405         if (ct->status & statusbit) {
 406                 struct nf_conntrack_tuple target;
 407
 408                 /* We are aiming to look like inverse of other direction. */
 409                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 410
 411                 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
 412                         return NF_DROP;
 413         }
 414         return NF_ACCEPT;
 415 }
 416 EXPORT_SYMBOL_GPL(nf_nat_packet);
 417
 418 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
 419 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 420                                   enum ip_conntrack_info ctinfo,
 421                                   unsigned int hooknum,
 422                                   struct sk_buff *skb)
 423 {
 424         struct {
 425                 struct icmphdr icmp;
 426                 struct iphdr ip;
 427         } *inside;
 428         const struct nf_conntrack_l4proto *l4proto;
 429         struct nf_conntrack_tuple inner, target;
 430         int hdrlen = ip_hdrlen(skb);
 431         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 432         unsigned long statusbit;
 433         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 434
 435         if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
 436                 return 0;
 437
 438         inside = (void *)skb->data + ip_hdrlen(skb);
 439
 440         /* We're actually going to mangle it beyond trivial checksum
 441            adjustment, so make sure the current checksum is correct. */
 442         if (nf_ip_checksum(skb, hooknum, hdrlen, 0))
 443                 return 0;
 444
 445         /* Must be RELATED */
 446         NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
 447                      skb->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
 448
 449         /* Redirects on non-null nats must be dropped, else they'll
 450            start talking to each other without our translation, and be
 451            confused... --RR */
 452         if (inside->icmp.type == ICMP_REDIRECT) {
 453                 /* If NAT isn't finished, assume it and drop. */
 454                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 455                         return 0;
 456
 457                 if (ct->status & IPS_NAT_MASK)
 458                         return 0;
 459         }
 460
 461         pr_debug("icmp_reply_translation: translating error %p manip %u "
 462                  "dir %s\n", skb, manip,
 463                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 464
 465         /* rcu_read_lock()ed by nf_hook_slow */
 466         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
 467
 468         if (!nf_ct_get_tuple(skb,
 469                              ip_hdrlen(skb) + sizeof(struct icmphdr),
 470                              (ip_hdrlen(skb) +
 471                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 472                              (u_int16_t)AF_INET,
 473                              inside->ip.protocol,
 474                              &inner, l3proto, l4proto))
 475                 return 0;
 476
 477         /* Change inner back to look like incoming packet.  We do the
 478            opposite manip on this hook to normal, because it might not
 479            pass all hooks (locally-generated ICMP).  Consider incoming
 480            packet: PREROUTING (DST manip), routing produces ICMP, goes
 481            through POSTROUTING (which must correct the DST manip). */
 482         if (!manip_pkt(inside->ip.protocol, skb,
 483                        ip_hdrlen(skb) + sizeof(inside->icmp),
 484                        &ct->tuplehash[!dir].tuple,
 485                        !manip))
 486                 return 0;
 487
 488         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 489                 /* Reloading "inside" here since manip_pkt inner. */
 490                 inside = (void *)skb->data + ip_hdrlen(skb);
 491                 inside->icmp.checksum = 0;
 492                 inside->icmp.checksum =
 493                         csum_fold(skb_checksum(skb, hdrlen,
 494                                                skb->len - hdrlen, 0));
 495         }
 496
 497         /* Change outer to look the reply to an incoming packet
 498          * (proto 0 means don't invert per-proto part). */
 499         if (manip == IP_NAT_MANIP_SRC)
 500                 statusbit = IPS_SRC_NAT;
 501         else
 502                 statusbit = IPS_DST_NAT;
 503
 504         /* Invert if this is reply dir. */
 505         if (dir == IP_CT_DIR_REPLY)
 506                 statusbit ^= IPS_NAT_MASK;
 507
 508         if (ct->status & statusbit) {
 509                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 510                 if (!manip_pkt(0, skb, 0, &target, manip))
 511                         return 0;
 512         }
 513
 514         return 1;
 515 }
 516 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 517
 518 /* Protocol registration. */
 519 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 520 {
 521         int ret = 0;
 522
 523         spin_lock_bh(&nf_nat_lock);
 524         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
 525                 ret = -EBUSY;
 526                 goto out;
 527         }
 528         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 529  out:
 530         spin_unlock_bh(&nf_nat_lock);
 531         return ret;
 532 }
 533 EXPORT_SYMBOL(nf_nat_protocol_register);
 534
 535 /* Noone stores the protocol anywhere; simply delete it. */
 536 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 537 {
 538         spin_lock_bh(&nf_nat_lock);
 539         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 540                            &nf_nat_unknown_protocol);
 541         spin_unlock_bh(&nf_nat_lock);
 542         synchronize_rcu();
 543 }
 544 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 545
 546 /* Noone using conntrack by the time this called. */
 547 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 548 {
 549         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 550
 551         if (nat == NULL || nat->ct == NULL)
 552                 return;
 553
 554         NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
 555
 556         spin_lock_bh(&nf_nat_lock);
 557         hlist_del_rcu(&nat->bysource);
 558         spin_unlock_bh(&nf_nat_lock);
 559 }
 560
 561 static void nf_nat_move_storage(void *new, void *old)
 562 {
 563         struct nf_conn_nat *new_nat = new;
 564         struct nf_conn_nat *old_nat = old;
 565         struct nf_conn *ct = old_nat->ct;
 566
 567         if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
 568                 return;
 569
 570         spin_lock_bh(&nf_nat_lock);
 571         new_nat->ct = ct;
 572         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 573         spin_unlock_bh(&nf_nat_lock);
 574 }
 575
 576 static struct nf_ct_ext_type nat_extend __read_mostly = {
 577         .len            = sizeof(struct nf_conn_nat),
 578         .align          = __alignof__(struct nf_conn_nat),
 579         .destroy        = nf_nat_cleanup_conntrack,
 580         .move           = nf_nat_move_storage,
 581         .id             = NF_CT_EXT_NAT,
 582         .flags          = NF_CT_EXT_F_PREALLOC,
 583 };
 584
 585 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 586
 587 #include <linux/netfilter/nfnetlink.h>
 588 #include <linux/netfilter/nfnetlink_conntrack.h>
 589
 590 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 591         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 592         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 593 };
 594
 595 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 596                                      const struct nf_conn *ct,
 597                                      struct nf_nat_range *range)
 598 {
 599         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 600         const struct nf_nat_protocol *npt;
 601         int err;
 602
 603         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 604         if (err < 0)
 605                 return err;
 606
 607         npt = nf_nat_proto_find_get(nf_ct_protonum(ct));
 608         if (npt->nlattr_to_range)
 609                 err = npt->nlattr_to_range(tb, range);
 610         nf_nat_proto_put(npt);
 611         return err;
 612 }
 613
 614 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 615         [CTA_NAT_MINIP]         = { .type = NLA_U32 },
 616         [CTA_NAT_MAXIP]         = { .type = NLA_U32 },
 617 };
 618
 619 static int
 620 nfnetlink_parse_nat(const struct nlattr *nat,
 621                     const struct nf_conn *ct, struct nf_nat_range *range)
 622 {
 623         struct nlattr *tb[CTA_NAT_MAX+1];
 624         int err;
 625
 626         memset(range, 0, sizeof(*range));
 627
 628         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 629         if (err < 0)
 630                 return err;
 631
 632         if (tb[CTA_NAT_MINIP])
 633                 range->min_ip = nla_get_be32(tb[CTA_NAT_MINIP]);
 634
 635         if (!tb[CTA_NAT_MAXIP])
 636                 range->max_ip = range->min_ip;
 637         else
 638                 range->max_ip = nla_get_be32(tb[CTA_NAT_MAXIP]);
 639
 640         if (range->min_ip)
 641                 range->flags |= IP_NAT_RANGE_MAP_IPS;
 642
 643         if (!tb[CTA_NAT_PROTO])
 644                 return 0;
 645
 646         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 647         if (err < 0)
 648                 return err;
 649
 650         return 0;
 651 }
 652
 653 static int
 654 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 655                           enum nf_nat_manip_type manip,
 656                           const struct nlattr *attr)
 657 {
 658         struct nf_nat_range range;
 659
 660         if (nfnetlink_parse_nat(attr, ct, &range) < 0)
 661                 return -EINVAL;
 662         if (nf_nat_initialized(ct, manip))
 663                 return -EEXIST;
 664
 665         return nf_nat_setup_info(ct, &range, manip);
 666 }
 667 #else
 668 static int
 669 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 670                           enum nf_nat_manip_type manip,
 671                           const struct nlattr *attr)
 672 {
 673         return -EOPNOTSUPP;
 674 }
 675 #endif
 676
 677 static int __net_init nf_nat_net_init(struct net *net)
 678 {
 679         /* Leave them the same for the moment. */
 680         net->ipv4.nat_htable_size = net->ct.htable_size;
 681         net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&net->ipv4.nat_htable_size,
 682                                                        &net->ipv4.nat_vmalloced, 0);
 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                              net->ipv4.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         ret = register_pernet_subsys(&nf_nat_net_ops);
 727         if (ret < 0)
 728                 goto cleanup_extend;
 729
 730         /* Sew in builtin protocols. */
 731         spin_lock_bh(&nf_nat_lock);
 732         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 733                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
 734         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 735         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 736         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 737         spin_unlock_bh(&nf_nat_lock);
 738
 739         /* Initialize fake conntrack so that NAT will skip it */
 740         nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
 741
 742         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 743
 744         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 745         rcu_assign_pointer(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 746         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 747         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook,
 748                            nfnetlink_parse_nat_setup);
 749         BUG_ON(nf_ct_nat_offset != NULL);
 750         rcu_assign_pointer(nf_ct_nat_offset, nf_nat_get_offset);
 751         return 0;
 752
 753  cleanup_extend:
 754         nf_ct_extend_unregister(&nat_extend);
 755         return ret;
 756 }
 757
 758 static void __exit nf_nat_cleanup(void)
 759 {
 760         unregister_pernet_subsys(&nf_nat_net_ops);
 761         nf_ct_l3proto_put(l3proto);
 762         nf_ct_extend_unregister(&nat_extend);
 763         rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL);
 764         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, NULL);
 765         rcu_assign_pointer(nf_ct_nat_offset, 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);