net/ipv4/netfilter/nf_nat_core.c

   1 /* NAT for netfilter; shared with compatibility layer. */
   2
   3 /* (C) 1999-2001 Paul `Rusty' Russell
   4  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/types.h>
  13 #include <linux/timer.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/gfp.h>
  16 #include <net/checksum.h>
  17 #include <net/icmp.h>
  18 #include <net/ip.h>
  19 #include <net/tcp.h>  /* For tcp_prot in getorigdst */
  20 #include <linux/icmp.h>
  21 #include <linux/udp.h>
  22 #include <linux/jhash.h>
  23
  24 #include <linux/netfilter_ipv4.h>
  25 #include <net/netfilter/nf_conntrack.h>
  26 #include <net/netfilter/nf_conntrack_core.h>
  27 #include <net/netfilter/nf_nat.h>
  28 #include <net/netfilter/nf_nat_protocol.h>
  29 #include <net/netfilter/nf_nat_core.h>
  30 #include <net/netfilter/nf_nat_helper.h>
  31 #include <net/netfilter/nf_conntrack_helper.h>
  32 #include <net/netfilter/nf_conntrack_l3proto.h>
  33 #include <net/netfilter/nf_conntrack_l4proto.h>
  34 #include <net/netfilter/nf_conntrack_zones.h>
  35
  36 static DEFINE_SPINLOCK(nf_nat_lock);
  37
  38 static struct nf_conntrack_l3proto *l3proto __read_mostly;
  39
  40 #define MAX_IP_NAT_PROTO 256
  41 static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
  42                                                 __read_mostly;
  43
  44 static inline const struct nf_nat_protocol *
  45 __nf_nat_proto_find(u_int8_t protonum)
  46 {
  47         return rcu_dereference(nf_nat_protos[protonum]);
  48 }
  49
  50 const struct nf_nat_protocol *
  51 nf_nat_proto_find_get(u_int8_t protonum)
  52 {
  53         const struct nf_nat_protocol *p;
  54
  55         rcu_read_lock();
  56         p = __nf_nat_proto_find(protonum);
  57         if (!try_module_get(p->me))
  58                 p = &nf_nat_unknown_protocol;
  59         rcu_read_unlock();
  60
  61         return p;
  62 }
  63 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
  64
  65 void
  66 nf_nat_proto_put(const struct nf_nat_protocol *p)
  67 {
  68         module_put(p->me);
  69 }
  70 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
  71
  72 /* We keep an extra hash for each conntrack, for fast searching. */
  73 static inline unsigned int
  74 hash_by_src(const struct net *net, u16 zone,
  75             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 ^ zone,
  82                             tuple->dst.protonum, 0);
  83         return ((u64)hash * net->ipv4.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, u16 zone,
 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(net, zone, 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) && nf_ct_zone(ct) == zone) {
 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(u16 zone, 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                          range->flags & IP_NAT_RANGE_PERSISTENT ?
 215                                 0 : (__force u32)tuple->dst.u3.ip ^ zone, 0);
 216         j = ((u64)j * (maxip - minip + 1)) >> 32;
 217         *var_ipp = htonl(minip + j);
 218 }
 219
 220 /* Manipulate the tuple into the range given.  For NF_INET_POST_ROUTING,
 221  * we change the source to map into the range.  For NF_INET_PRE_ROUTING
 222  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 223  * range.  It might not be possible to get a unique tuple, but we try.
 224  * At worst (or if we race), we will end up with a final duplicate in
 225  * __ip_conntrack_confirm and drop the packet. */
 226 static void
 227 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 228                  const struct nf_conntrack_tuple *orig_tuple,
 229                  const struct nf_nat_range *range,
 230                  struct nf_conn *ct,
 231                  enum nf_nat_manip_type maniptype)
 232 {
 233         struct net *net = nf_ct_net(ct);
 234         const struct nf_nat_protocol *proto;
 235         u16 zone = nf_ct_zone(ct);
 236
 237         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 238            and that same mapping gives a unique tuple within the given
 239            range, use that.
 240
 241            This is only required for source (ie. NAT/masq) mappings.
 242            So far, we don't do local source mappings, so multiple
 243            manips not an issue.  */
 244         if (maniptype == IP_NAT_MANIP_SRC &&
 245             !(range->flags & IP_NAT_RANGE_PROTO_RANDOM)) {
 246                 if (find_appropriate_src(net, zone, orig_tuple, tuple, range)) {
 247                         pr_debug("get_unique_tuple: Found current src map\n");
 248                         if (!nf_nat_used_tuple(tuple, ct))
 249                                 return;
 250                 }
 251         }
 252
 253         /* 2) Select the least-used IP/proto combination in the given
 254            range. */
 255         *tuple = *orig_tuple;
 256         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 257
 258         /* 3) The per-protocol part of the manip is made to map into
 259            the range to make a unique tuple. */
 260
 261         rcu_read_lock();
 262         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
 263
 264         /* Only bother mapping if it's not already in range and unique */
 265         if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM) &&
 266             (!(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 net *net = nf_ct_net(ct);
 283         struct nf_conntrack_tuple curr_tuple, new_tuple;
 284         struct nf_conn_nat *nat;
 285         int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
 286
 287         /* nat helper or nfctnetlink also setup binding */
 288         nat = nfct_nat(ct);
 289         if (!nat) {
 290                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 291                 if (nat == NULL) {
 292                         pr_debug("failed to add NAT extension\n");
 293                         return NF_ACCEPT;
 294                 }
 295         }
 296
 297         NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC ||
 298                      maniptype == IP_NAT_MANIP_DST);
 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(net, nf_ct_zone(ct),
 330                                       &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 + hdrlen;
 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, hdrlen + sizeof(struct icmphdr),
 469                              (hdrlen +
 470                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 471                              (u_int16_t)AF_INET, 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, skb, hdrlen + sizeof(inside->icmp),
 481                        &ct->tuplehash[!dir].tuple, !manip))
 482                 return 0;
 483
 484         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 485                 /* Reloading "inside" here since manip_pkt inner. */
 486                 inside = (void *)skb->data + hdrlen;
 487                 inside->icmp.checksum = 0;
 488                 inside->icmp.checksum =
 489                         csum_fold(skb_checksum(skb, hdrlen,
 490                                                skb->len - hdrlen, 0));
 491         }
 492
 493         /* Change outer to look the reply to an incoming packet
 494          * (proto 0 means don't invert per-proto part). */
 495         if (manip == IP_NAT_MANIP_SRC)
 496                 statusbit = IPS_SRC_NAT;
 497         else
 498                 statusbit = IPS_DST_NAT;
 499
 500         /* Invert if this is reply dir. */
 501         if (dir == IP_CT_DIR_REPLY)
 502                 statusbit ^= IPS_NAT_MASK;
 503
 504         if (ct->status & statusbit) {
 505                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 506                 if (!manip_pkt(0, skb, 0, &target, manip))
 507                         return 0;
 508         }
 509
 510         return 1;
 511 }
 512 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 513
 514 /* Protocol registration. */
 515 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 516 {
 517         int ret = 0;
 518
 519         spin_lock_bh(&nf_nat_lock);
 520         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
 521                 ret = -EBUSY;
 522                 goto out;
 523         }
 524         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 525  out:
 526         spin_unlock_bh(&nf_nat_lock);
 527         return ret;
 528 }
 529 EXPORT_SYMBOL(nf_nat_protocol_register);
 530
 531 /* Noone stores the protocol anywhere; simply delete it. */
 532 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 533 {
 534         spin_lock_bh(&nf_nat_lock);
 535         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 536                            &nf_nat_unknown_protocol);
 537         spin_unlock_bh(&nf_nat_lock);
 538         synchronize_rcu();
 539 }
 540 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 541
 542 /* Noone using conntrack by the time this called. */
 543 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 544 {
 545         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 546
 547         if (nat == NULL || nat->ct == NULL)
 548                 return;
 549
 550         NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
 551
 552         spin_lock_bh(&nf_nat_lock);
 553         hlist_del_rcu(&nat->bysource);
 554         spin_unlock_bh(&nf_nat_lock);
 555 }
 556
 557 static void nf_nat_move_storage(void *new, void *old)
 558 {
 559         struct nf_conn_nat *new_nat = new;
 560         struct nf_conn_nat *old_nat = old;
 561         struct nf_conn *ct = old_nat->ct;
 562
 563         if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
 564                 return;
 565
 566         spin_lock_bh(&nf_nat_lock);
 567         new_nat->ct = ct;
 568         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 569         spin_unlock_bh(&nf_nat_lock);
 570 }
 571
 572 static struct nf_ct_ext_type nat_extend __read_mostly = {
 573         .len            = sizeof(struct nf_conn_nat),
 574         .align          = __alignof__(struct nf_conn_nat),
 575         .destroy        = nf_nat_cleanup_conntrack,
 576         .move           = nf_nat_move_storage,
 577         .id             = NF_CT_EXT_NAT,
 578         .flags          = NF_CT_EXT_F_PREALLOC,
 579 };
 580
 581 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 582
 583 #include <linux/netfilter/nfnetlink.h>
 584 #include <linux/netfilter/nfnetlink_conntrack.h>
 585
 586 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 587         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 588         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 589 };
 590
 591 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 592                                      const struct nf_conn *ct,
 593                                      struct nf_nat_range *range)
 594 {
 595         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 596         const struct nf_nat_protocol *npt;
 597         int err;
 598
 599         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 600         if (err < 0)
 601                 return err;
 602
 603         npt = nf_nat_proto_find_get(nf_ct_protonum(ct));
 604         if (npt->nlattr_to_range)
 605                 err = npt->nlattr_to_range(tb, range);
 606         nf_nat_proto_put(npt);
 607         return err;
 608 }
 609
 610 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 611         [CTA_NAT_MINIP]         = { .type = NLA_U32 },
 612         [CTA_NAT_MAXIP]         = { .type = NLA_U32 },
 613 };
 614
 615 static int
 616 nfnetlink_parse_nat(const struct nlattr *nat,
 617                     const struct nf_conn *ct, struct nf_nat_range *range)
 618 {
 619         struct nlattr *tb[CTA_NAT_MAX+1];
 620         int err;
 621
 622         memset(range, 0, sizeof(*range));
 623
 624         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 625         if (err < 0)
 626                 return err;
 627
 628         if (tb[CTA_NAT_MINIP])
 629                 range->min_ip = nla_get_be32(tb[CTA_NAT_MINIP]);
 630
 631         if (!tb[CTA_NAT_MAXIP])
 632                 range->max_ip = range->min_ip;
 633         else
 634                 range->max_ip = nla_get_be32(tb[CTA_NAT_MAXIP]);
 635
 636         if (range->min_ip)
 637                 range->flags |= IP_NAT_RANGE_MAP_IPS;
 638
 639         if (!tb[CTA_NAT_PROTO])
 640                 return 0;
 641
 642         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 643         if (err < 0)
 644                 return err;
 645
 646         return 0;
 647 }
 648
 649 static int
 650 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 651                           enum nf_nat_manip_type manip,
 652                           const struct nlattr *attr)
 653 {
 654         struct nf_nat_range range;
 655
 656         if (nfnetlink_parse_nat(attr, ct, &range) < 0)
 657                 return -EINVAL;
 658         if (nf_nat_initialized(ct, manip))
 659                 return -EEXIST;
 660
 661         return nf_nat_setup_info(ct, &range, manip);
 662 }
 663 #else
 664 static int
 665 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 666                           enum nf_nat_manip_type manip,
 667                           const struct nlattr *attr)
 668 {
 669         return -EOPNOTSUPP;
 670 }
 671 #endif
 672
 673 static int __net_init nf_nat_net_init(struct net *net)
 674 {
 675         /* Leave them the same for the moment. */
 676         net->ipv4.nat_htable_size = net->ct.htable_size;
 677         net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&net->ipv4.nat_htable_size,
 678                                                        &net->ipv4.nat_vmalloced, 0);
 679         if (!net->ipv4.nat_bysource)
 680                 return -ENOMEM;
 681         return 0;
 682 }
 683
 684 /* Clear NAT section of all conntracks, in case we're loaded again. */
 685 static int clean_nat(struct nf_conn *i, void *data)
 686 {
 687         struct nf_conn_nat *nat = nfct_nat(i);
 688
 689         if (!nat)
 690                 return 0;
 691         memset(nat, 0, sizeof(*nat));
 692         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
 693         return 0;
 694 }
 695
 696 static void __net_exit nf_nat_net_exit(struct net *net)
 697 {
 698         nf_ct_iterate_cleanup(net, &clean_nat, NULL);
 699         synchronize_rcu();
 700         nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_vmalloced,
 701                              net->ipv4.nat_htable_size);
 702 }
 703
 704 static struct pernet_operations nf_nat_net_ops = {
 705         .init = nf_nat_net_init,
 706         .exit = nf_nat_net_exit,
 707 };
 708
 709 static int __init nf_nat_init(void)
 710 {
 711         size_t i;
 712         int ret;
 713
 714         need_ipv4_conntrack();
 715
 716         ret = nf_ct_extend_register(&nat_extend);
 717         if (ret < 0) {
 718                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 719                 return ret;
 720         }
 721
 722         ret = register_pernet_subsys(&nf_nat_net_ops);
 723         if (ret < 0)
 724                 goto cleanup_extend;
 725
 726         /* Sew in builtin protocols. */
 727         spin_lock_bh(&nf_nat_lock);
 728         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 729                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
 730         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 731         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 732         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 733         spin_unlock_bh(&nf_nat_lock);
 734
 735         /* Initialize fake conntrack so that NAT will skip it */
 736         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 737
 738         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 739
 740         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 741         rcu_assign_pointer(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 742         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 743         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook,
 744                            nfnetlink_parse_nat_setup);
 745         BUG_ON(nf_ct_nat_offset != NULL);
 746         rcu_assign_pointer(nf_ct_nat_offset, nf_nat_get_offset);
 747         return 0;
 748
 749  cleanup_extend:
 750         nf_ct_extend_unregister(&nat_extend);
 751         return ret;
 752 }
 753
 754 static void __exit nf_nat_cleanup(void)
 755 {
 756         unregister_pernet_subsys(&nf_nat_net_ops);
 757         nf_ct_l3proto_put(l3proto);
 758         nf_ct_extend_unregister(&nat_extend);
 759         rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL);
 760         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, NULL);
 761         rcu_assign_pointer(nf_ct_nat_offset, NULL);
 762         synchronize_net();
 763 }
 764
 765 MODULE_LICENSE("GPL");
 766 MODULE_ALIAS("nf-nat-ipv4");
 767
 768 module_init(nf_nat_init);
 769 module_exit(nf_nat_cleanup);