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 #include <net/netfilter/nf_conntrack_zones.h>
  34
  35 static DEFINE_SPINLOCK(nf_nat_lock);
  36
  37 static struct nf_conntrack_l3proto *l3proto __read_mostly;
  38
  39 #define MAX_IP_NAT_PROTO 256
  40 static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
  41                                                 __read_mostly;
  42
  43 static inline const struct nf_nat_protocol *
  44 __nf_nat_proto_find(u_int8_t protonum)
  45 {
  46         return rcu_dereference(nf_nat_protos[protonum]);
  47 }
  48
  49 const struct nf_nat_protocol *
  50 nf_nat_proto_find_get(u_int8_t protonum)
  51 {
  52         const struct nf_nat_protocol *p;
  53
  54         rcu_read_lock();
  55         p = __nf_nat_proto_find(protonum);
  56         if (!try_module_get(p->me))
  57                 p = &nf_nat_unknown_protocol;
  58         rcu_read_unlock();
  59
  60         return p;
  61 }
  62 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
  63
  64 void
  65 nf_nat_proto_put(const struct nf_nat_protocol *p)
  66 {
  67         module_put(p->me);
  68 }
  69 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
  70
  71 /* We keep an extra hash for each conntrack, for fast searching. */
  72 static inline unsigned int
  73 hash_by_src(const struct net *net, u16 zone,
  74             const struct nf_conntrack_tuple *tuple)
  75 {
  76         unsigned int hash;
  77
  78         /* Original src, to ensure we map it consistently if poss. */
  79         hash = jhash_3words((__force u32)tuple->src.u3.ip,
  80                             (__force u32)tuple->src.u.all ^ zone,
  81                             tuple->dst.protonum, 0);
  82         return ((u64)hash * net->ipv4.nat_htable_size) >> 32;
  83 }
  84
  85 /* Is this tuple already taken? (not by us) */
  86 int
  87 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
  88                   const struct nf_conn *ignored_conntrack)
  89 {
  90         /* Conntrack tracking doesn't keep track of outgoing tuples; only
  91            incoming ones.  NAT means they don't have a fixed mapping,
  92            so we invert the tuple and look for the incoming reply.
  93
  94            We could keep a separate hash if this proves too slow. */
  95         struct nf_conntrack_tuple reply;
  96
  97         nf_ct_invert_tuplepr(&reply, tuple);
  98         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
  99 }
 100 EXPORT_SYMBOL(nf_nat_used_tuple);
 101
 102 /* If we source map this tuple so reply looks like reply_tuple, will
 103  * that meet the constraints of range. */
 104 static int
 105 in_range(const struct nf_conntrack_tuple *tuple,
 106          const struct nf_nat_range *range)
 107 {
 108         const struct nf_nat_protocol *proto;
 109         int ret = 0;
 110
 111         /* If we are supposed to map IPs, then we must be in the
 112            range specified, otherwise let this drag us onto a new src IP. */
 113         if (range->flags & IP_NAT_RANGE_MAP_IPS) {
 114                 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
 115                     ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
 116                         return 0;
 117         }
 118
 119         rcu_read_lock();
 120         proto = __nf_nat_proto_find(tuple->dst.protonum);
 121         if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 122             proto->in_range(tuple, IP_NAT_MANIP_SRC,
 123                             &range->min, &range->max))
 124                 ret = 1;
 125         rcu_read_unlock();
 126
 127         return ret;
 128 }
 129
 130 static inline int
 131 same_src(const struct nf_conn *ct,
 132          const struct nf_conntrack_tuple *tuple)
 133 {
 134         const struct nf_conntrack_tuple *t;
 135
 136         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 137         return (t->dst.protonum == tuple->dst.protonum &&
 138                 t->src.u3.ip == tuple->src.u3.ip &&
 139                 t->src.u.all == tuple->src.u.all);
 140 }
 141
 142 /* Only called for SRC manip */
 143 static int
 144 find_appropriate_src(struct net *net, u16 zone,
 145                      const struct nf_conntrack_tuple *tuple,
 146                      struct nf_conntrack_tuple *result,
 147                      const struct nf_nat_range *range)
 148 {
 149         unsigned int h = hash_by_src(net, zone, tuple);
 150         const struct nf_conn_nat *nat;
 151         const struct nf_conn *ct;
 152         const struct hlist_node *n;
 153
 154         rcu_read_lock();
 155         hlist_for_each_entry_rcu(nat, n, &net->ipv4.nat_bysource[h], bysource) {
 156                 ct = nat->ct;
 157                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
 158                         /* Copy source part from reply tuple. */
 159                         nf_ct_invert_tuplepr(result,
 160                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 161                         result->dst = tuple->dst;
 162
 163                         if (in_range(result, range)) {
 164                                 rcu_read_unlock();
 165                                 return 1;
 166                         }
 167                 }
 168         }
 169         rcu_read_unlock();
 170         return 0;
 171 }
 172
 173 /* For [FUTURE] fragmentation handling, we want the least-used
 174    src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 175    if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 176    1-65535, we don't do pro-rata allocation based on ports; we choose
 177    the ip with the lowest src-ip/dst-ip/proto usage.
 178 */
 179 static void
 180 find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple,
 181                     const struct nf_nat_range *range,
 182                     const struct nf_conn *ct,
 183                     enum nf_nat_manip_type maniptype)
 184 {
 185         __be32 *var_ipp;
 186         /* Host order */
 187         u_int32_t minip, maxip, j;
 188
 189         /* No IP mapping?  Do nothing. */
 190         if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
 191                 return;
 192
 193         if (maniptype == IP_NAT_MANIP_SRC)
 194                 var_ipp = &tuple->src.u3.ip;
 195         else
 196                 var_ipp = &tuple->dst.u3.ip;
 197
 198         /* Fast path: only one choice. */
 199         if (range->min_ip == range->max_ip) {
 200                 *var_ipp = range->min_ip;
 201                 return;
 202         }
 203
 204         /* Hashing source and destination IPs gives a fairly even
 205          * spread in practice (if there are a small number of IPs
 206          * involved, there usually aren't that many connections
 207          * anyway).  The consistency means that servers see the same
 208          * client coming from the same IP (some Internet Banking sites
 209          * like this), even across reboots. */
 210         minip = ntohl(range->min_ip);
 211         maxip = ntohl(range->max_ip);
 212         j = jhash_2words((__force u32)tuple->src.u3.ip,
 213                          range->flags & IP_NAT_RANGE_PERSISTENT ?
 214                                 0 : (__force u32)tuple->dst.u3.ip ^ zone, 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         u16 zone = nf_ct_zone(ct);
 235
 236         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 237            and that same mapping gives a unique tuple within the given
 238            range, use that.
 239
 240            This is only required for source (ie. NAT/masq) mappings.
 241            So far, we don't do local source mappings, so multiple
 242            manips not an issue.  */
 243         if (maniptype == IP_NAT_MANIP_SRC &&
 244             !(range->flags & IP_NAT_RANGE_PROTO_RANDOM)) {
 245                 if (find_appropriate_src(net, zone, orig_tuple, tuple, range)) {
 246                         pr_debug("get_unique_tuple: Found current src map\n");
 247                         if (!nf_nat_used_tuple(tuple, ct))
 248                                 return;
 249                 }
 250         }
 251
 252         /* 2) Select the least-used IP/proto combination in the given
 253            range. */
 254         *tuple = *orig_tuple;
 255         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 256
 257         /* 3) The per-protocol part of the manip is made to map into
 258            the range to make a unique tuple. */
 259
 260         rcu_read_lock();
 261         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
 262
 263         /* Change protocol info to have some randomization */
 264         if (range->flags & IP_NAT_RANGE_PROTO_RANDOM) {
 265                 proto->unique_tuple(tuple, range, maniptype, ct);
 266                 goto out;
 267         }
 268
 269         /* Only bother mapping if it's not already in range and unique */
 270         if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 271              proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
 272             !nf_nat_used_tuple(tuple, ct))
 273                 goto out;
 274
 275         /* Last change: get protocol to try to obtain unique tuple. */
 276         proto->unique_tuple(tuple, range, maniptype, ct);
 277 out:
 278         rcu_read_unlock();
 279 }
 280
 281 unsigned int
 282 nf_nat_setup_info(struct nf_conn *ct,
 283                   const struct nf_nat_range *range,
 284                   enum nf_nat_manip_type maniptype)
 285 {
 286         struct net *net = nf_ct_net(ct);
 287         struct nf_conntrack_tuple curr_tuple, new_tuple;
 288         struct nf_conn_nat *nat;
 289         int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
 290
 291         /* nat helper or nfctnetlink also setup binding */
 292         nat = nfct_nat(ct);
 293         if (!nat) {
 294                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 295                 if (nat == NULL) {
 296                         pr_debug("failed to add NAT extension\n");
 297                         return NF_ACCEPT;
 298                 }
 299         }
 300
 301         NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC ||
 302                      maniptype == IP_NAT_MANIP_DST);
 303         BUG_ON(nf_nat_initialized(ct, maniptype));
 304
 305         /* What we've got will look like inverse of reply. Normally
 306            this is what is in the conntrack, except for prior
 307            manipulations (future optimization: if num_manips == 0,
 308            orig_tp =
 309            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
 310         nf_ct_invert_tuplepr(&curr_tuple,
 311                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 312
 313         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 314
 315         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 316                 struct nf_conntrack_tuple reply;
 317
 318                 /* Alter conntrack table so will recognize replies. */
 319                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 320                 nf_conntrack_alter_reply(ct, &reply);
 321
 322                 /* Non-atomic: we own this at the moment. */
 323                 if (maniptype == IP_NAT_MANIP_SRC)
 324                         ct->status |= IPS_SRC_NAT;
 325                 else
 326                         ct->status |= IPS_DST_NAT;
 327         }
 328
 329         /* Place in source hash if this is the first time. */
 330         if (have_to_hash) {
 331                 unsigned int srchash;
 332
 333                 srchash = hash_by_src(net, nf_ct_zone(ct),
 334                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 335                 spin_lock_bh(&nf_nat_lock);
 336                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
 337                 nat = nfct_nat(ct);
 338                 nat->ct = ct;
 339                 hlist_add_head_rcu(&nat->bysource,
 340                                    &net->ipv4.nat_bysource[srchash]);
 341                 spin_unlock_bh(&nf_nat_lock);
 342         }
 343
 344         /* It's done. */
 345         if (maniptype == IP_NAT_MANIP_DST)
 346                 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
 347         else
 348                 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 349
 350         return NF_ACCEPT;
 351 }
 352 EXPORT_SYMBOL(nf_nat_setup_info);
 353
 354 /* Returns true if succeeded. */
 355 static bool
 356 manip_pkt(u_int16_t proto,
 357           struct sk_buff *skb,
 358           unsigned int iphdroff,
 359           const struct nf_conntrack_tuple *target,
 360           enum nf_nat_manip_type maniptype)
 361 {
 362         struct iphdr *iph;
 363         const struct nf_nat_protocol *p;
 364
 365         if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
 366                 return false;
 367
 368         iph = (void *)skb->data + iphdroff;
 369
 370         /* Manipulate protcol part. */
 371
 372         /* rcu_read_lock()ed by nf_hook_slow */
 373         p = __nf_nat_proto_find(proto);
 374         if (!p->manip_pkt(skb, iphdroff, target, maniptype))
 375                 return false;
 376
 377         iph = (void *)skb->data + iphdroff;
 378
 379         if (maniptype == IP_NAT_MANIP_SRC) {
 380                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 381                 iph->saddr = target->src.u3.ip;
 382         } else {
 383                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 384                 iph->daddr = target->dst.u3.ip;
 385         }
 386         return true;
 387 }
 388
 389 /* Do packet manipulations according to nf_nat_setup_info. */
 390 unsigned int nf_nat_packet(struct nf_conn *ct,
 391                            enum ip_conntrack_info ctinfo,
 392                            unsigned int hooknum,
 393                            struct sk_buff *skb)
 394 {
 395         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 396         unsigned long statusbit;
 397         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 398
 399         if (mtype == IP_NAT_MANIP_SRC)
 400                 statusbit = IPS_SRC_NAT;
 401         else
 402                 statusbit = IPS_DST_NAT;
 403
 404         /* Invert if this is reply dir. */
 405         if (dir == IP_CT_DIR_REPLY)
 406                 statusbit ^= IPS_NAT_MASK;
 407
 408         /* Non-atomic: these bits don't change. */
 409         if (ct->status & statusbit) {
 410                 struct nf_conntrack_tuple target;
 411
 412                 /* We are aiming to look like inverse of other direction. */
 413                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 414
 415                 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
 416                         return NF_DROP;
 417         }
 418         return NF_ACCEPT;
 419 }
 420 EXPORT_SYMBOL_GPL(nf_nat_packet);
 421
 422 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
 423 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 424                                   enum ip_conntrack_info ctinfo,
 425                                   unsigned int hooknum,
 426                                   struct sk_buff *skb)
 427 {
 428         struct {
 429                 struct icmphdr icmp;
 430                 struct iphdr ip;
 431         } *inside;
 432         const struct nf_conntrack_l4proto *l4proto;
 433         struct nf_conntrack_tuple inner, target;
 434         int hdrlen = ip_hdrlen(skb);
 435         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 436         unsigned long statusbit;
 437         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 438
 439         if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
 440                 return 0;
 441
 442         inside = (void *)skb->data + ip_hdrlen(skb);
 443
 444         /* We're actually going to mangle it beyond trivial checksum
 445            adjustment, so make sure the current checksum is correct. */
 446         if (nf_ip_checksum(skb, hooknum, hdrlen, 0))
 447                 return 0;
 448
 449         /* Must be RELATED */
 450         NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
 451                      skb->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
 452
 453         /* Redirects on non-null nats must be dropped, else they'll
 454            start talking to each other without our translation, and be
 455            confused... --RR */
 456         if (inside->icmp.type == ICMP_REDIRECT) {
 457                 /* If NAT isn't finished, assume it and drop. */
 458                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 459                         return 0;
 460
 461                 if (ct->status & IPS_NAT_MASK)
 462                         return 0;
 463         }
 464
 465         pr_debug("icmp_reply_translation: translating error %p manip %u "
 466                  "dir %s\n", skb, manip,
 467                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 468
 469         /* rcu_read_lock()ed by nf_hook_slow */
 470         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
 471
 472         if (!nf_ct_get_tuple(skb,
 473                              ip_hdrlen(skb) + sizeof(struct icmphdr),
 474                              (ip_hdrlen(skb) +
 475                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 476                              (u_int16_t)AF_INET,
 477                              inside->ip.protocol,
 478                              &inner, l3proto, l4proto))
 479                 return 0;
 480
 481         /* Change inner back to look like incoming packet.  We do the
 482            opposite manip on this hook to normal, because it might not
 483            pass all hooks (locally-generated ICMP).  Consider incoming
 484            packet: PREROUTING (DST manip), routing produces ICMP, goes
 485            through POSTROUTING (which must correct the DST manip). */
 486         if (!manip_pkt(inside->ip.protocol, skb,
 487                        ip_hdrlen(skb) + sizeof(inside->icmp),
 488                        &ct->tuplehash[!dir].tuple,
 489                        !manip))
 490                 return 0;
 491
 492         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 493                 /* Reloading "inside" here since manip_pkt inner. */
 494                 inside = (void *)skb->data + ip_hdrlen(skb);
 495                 inside->icmp.checksum = 0;
 496                 inside->icmp.checksum =
 497                         csum_fold(skb_checksum(skb, hdrlen,
 498                                                skb->len - hdrlen, 0));
 499         }
 500
 501         /* Change outer to look the reply to an incoming packet
 502          * (proto 0 means don't invert per-proto part). */
 503         if (manip == IP_NAT_MANIP_SRC)
 504                 statusbit = IPS_SRC_NAT;
 505         else
 506                 statusbit = IPS_DST_NAT;
 507
 508         /* Invert if this is reply dir. */
 509         if (dir == IP_CT_DIR_REPLY)
 510                 statusbit ^= IPS_NAT_MASK;
 511
 512         if (ct->status & statusbit) {
 513                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 514                 if (!manip_pkt(0, skb, 0, &target, manip))
 515                         return 0;
 516         }
 517
 518         return 1;
 519 }
 520 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 521
 522 /* Protocol registration. */
 523 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 524 {
 525         int ret = 0;
 526
 527         spin_lock_bh(&nf_nat_lock);
 528         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
 529                 ret = -EBUSY;
 530                 goto out;
 531         }
 532         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 533  out:
 534         spin_unlock_bh(&nf_nat_lock);
 535         return ret;
 536 }
 537 EXPORT_SYMBOL(nf_nat_protocol_register);
 538
 539 /* Noone stores the protocol anywhere; simply delete it. */
 540 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 541 {
 542         spin_lock_bh(&nf_nat_lock);
 543         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 544                            &nf_nat_unknown_protocol);
 545         spin_unlock_bh(&nf_nat_lock);
 546         synchronize_rcu();
 547 }
 548 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 549
 550 /* Noone using conntrack by the time this called. */
 551 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 552 {
 553         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 554
 555         if (nat == NULL || nat->ct == NULL)
 556                 return;
 557
 558         NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
 559
 560         spin_lock_bh(&nf_nat_lock);
 561         hlist_del_rcu(&nat->bysource);
 562         spin_unlock_bh(&nf_nat_lock);
 563 }
 564
 565 static void nf_nat_move_storage(void *new, void *old)
 566 {
 567         struct nf_conn_nat *new_nat = new;
 568         struct nf_conn_nat *old_nat = old;
 569         struct nf_conn *ct = old_nat->ct;
 570
 571         if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
 572                 return;
 573
 574         spin_lock_bh(&nf_nat_lock);
 575         new_nat->ct = ct;
 576         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 577         spin_unlock_bh(&nf_nat_lock);
 578 }
 579
 580 static struct nf_ct_ext_type nat_extend __read_mostly = {
 581         .len            = sizeof(struct nf_conn_nat),
 582         .align          = __alignof__(struct nf_conn_nat),
 583         .destroy        = nf_nat_cleanup_conntrack,
 584         .move           = nf_nat_move_storage,
 585         .id             = NF_CT_EXT_NAT,
 586         .flags          = NF_CT_EXT_F_PREALLOC,
 587 };
 588
 589 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 590
 591 #include <linux/netfilter/nfnetlink.h>
 592 #include <linux/netfilter/nfnetlink_conntrack.h>
 593
 594 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 595         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 596         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 597 };
 598
 599 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 600                                      const struct nf_conn *ct,
 601                                      struct nf_nat_range *range)
 602 {
 603         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 604         const struct nf_nat_protocol *npt;
 605         int err;
 606
 607         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 608         if (err < 0)
 609                 return err;
 610
 611         npt = nf_nat_proto_find_get(nf_ct_protonum(ct));
 612         if (npt->nlattr_to_range)
 613                 err = npt->nlattr_to_range(tb, range);
 614         nf_nat_proto_put(npt);
 615         return err;
 616 }
 617
 618 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 619         [CTA_NAT_MINIP]         = { .type = NLA_U32 },
 620         [CTA_NAT_MAXIP]         = { .type = NLA_U32 },
 621 };
 622
 623 static int
 624 nfnetlink_parse_nat(const struct nlattr *nat,
 625                     const struct nf_conn *ct, struct nf_nat_range *range)
 626 {
 627         struct nlattr *tb[CTA_NAT_MAX+1];
 628         int err;
 629
 630         memset(range, 0, sizeof(*range));
 631
 632         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 633         if (err < 0)
 634                 return err;
 635
 636         if (tb[CTA_NAT_MINIP])
 637                 range->min_ip = nla_get_be32(tb[CTA_NAT_MINIP]);
 638
 639         if (!tb[CTA_NAT_MAXIP])
 640                 range->max_ip = range->min_ip;
 641         else
 642                 range->max_ip = nla_get_be32(tb[CTA_NAT_MAXIP]);
 643
 644         if (range->min_ip)
 645                 range->flags |= IP_NAT_RANGE_MAP_IPS;
 646
 647         if (!tb[CTA_NAT_PROTO])
 648                 return 0;
 649
 650         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 651         if (err < 0)
 652                 return err;
 653
 654         return 0;
 655 }
 656
 657 static int
 658 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 659                           enum nf_nat_manip_type manip,
 660                           const struct nlattr *attr)
 661 {
 662         struct nf_nat_range range;
 663
 664         if (nfnetlink_parse_nat(attr, ct, &range) < 0)
 665                 return -EINVAL;
 666         if (nf_nat_initialized(ct, manip))
 667                 return -EEXIST;
 668
 669         return nf_nat_setup_info(ct, &range, manip);
 670 }
 671 #else
 672 static int
 673 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 674                           enum nf_nat_manip_type manip,
 675                           const struct nlattr *attr)
 676 {
 677         return -EOPNOTSUPP;
 678 }
 679 #endif
 680
 681 static int __net_init nf_nat_net_init(struct net *net)
 682 {
 683         /* Leave them the same for the moment. */
 684         net->ipv4.nat_htable_size = net->ct.htable_size;
 685         net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&net->ipv4.nat_htable_size,
 686                                                        &net->ipv4.nat_vmalloced, 0);
 687         if (!net->ipv4.nat_bysource)
 688                 return -ENOMEM;
 689         return 0;
 690 }
 691
 692 /* Clear NAT section of all conntracks, in case we're loaded again. */
 693 static int clean_nat(struct nf_conn *i, void *data)
 694 {
 695         struct nf_conn_nat *nat = nfct_nat(i);
 696
 697         if (!nat)
 698                 return 0;
 699         memset(nat, 0, sizeof(*nat));
 700         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
 701         return 0;
 702 }
 703
 704 static void __net_exit nf_nat_net_exit(struct net *net)
 705 {
 706         nf_ct_iterate_cleanup(net, &clean_nat, NULL);
 707         synchronize_rcu();
 708         nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_vmalloced,
 709                              net->ipv4.nat_htable_size);
 710 }
 711
 712 static struct pernet_operations nf_nat_net_ops = {
 713         .init = nf_nat_net_init,
 714         .exit = nf_nat_net_exit,
 715 };
 716
 717 static int __init nf_nat_init(void)
 718 {
 719         size_t i;
 720         int ret;
 721
 722         need_ipv4_conntrack();
 723
 724         ret = nf_ct_extend_register(&nat_extend);
 725         if (ret < 0) {
 726                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 727                 return ret;
 728         }
 729
 730         ret = register_pernet_subsys(&nf_nat_net_ops);
 731         if (ret < 0)
 732                 goto cleanup_extend;
 733
 734         /* Sew in builtin protocols. */
 735         spin_lock_bh(&nf_nat_lock);
 736         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 737                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
 738         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 739         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 740         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 741         spin_unlock_bh(&nf_nat_lock);
 742
 743         /* Initialize fake conntrack so that NAT will skip it */
 744         nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
 745
 746         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 747
 748         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 749         rcu_assign_pointer(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 750         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 751         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook,
 752                            nfnetlink_parse_nat_setup);
 753         BUG_ON(nf_ct_nat_offset != NULL);
 754         rcu_assign_pointer(nf_ct_nat_offset, nf_nat_get_offset);
 755         return 0;
 756
 757  cleanup_extend:
 758         nf_ct_extend_unregister(&nat_extend);
 759         return ret;
 760 }
 761
 762 static void __exit nf_nat_cleanup(void)
 763 {
 764         unregister_pernet_subsys(&nf_nat_net_ops);
 765         nf_ct_l3proto_put(l3proto);
 766         nf_ct_extend_unregister(&nat_extend);
 767         rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL);
 768         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, NULL);
 769         rcu_assign_pointer(nf_ct_nat_offset, NULL);
 770         synchronize_net();
 771 }
 772
 773 MODULE_LICENSE("GPL");
 774 MODULE_ALIAS("nf-nat-ipv4");
 775
 776 module_init(nf_nat_init);
 777 module_exit(nf_nat_cleanup);