SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/ipv4/netfilter/nf_nat_core.c

   1 /* NAT for netfilter; shared with compatibility layer. */
   2
   3 /* (C) 1999-2001 Paul `Rusty' Russell
   4  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/types.h>
  13 #include <linux/timer.h>
  14 #include <linux/skbuff.h>
  15 #include <net/checksum.h>
  16 #include <net/icmp.h>
  17 #include <net/ip.h>
  18 #include <net/tcp.h>  /* For tcp_prot in getorigdst */
  19 #include <linux/icmp.h>
  20 #include <linux/udp.h>
  21 #include <linux/jhash.h>
  22
  23 #include <linux/netfilter_ipv4.h>
  24 #include <net/netfilter/nf_conntrack.h>
  25 #include <net/netfilter/nf_conntrack_core.h>
  26 #include <net/netfilter/nf_nat.h>
  27 #include <net/netfilter/nf_nat_protocol.h>
  28 #include <net/netfilter/nf_nat_core.h>
  29 #include <net/netfilter/nf_nat_helper.h>
  30 #include <net/netfilter/nf_conntrack_helper.h>
  31 #include <net/netfilter/nf_conntrack_l3proto.h>
  32 #include <net/netfilter/nf_conntrack_l4proto.h>
  33
  34 static DEFINE_SPINLOCK(nf_nat_lock);
  35
  36 static struct nf_conntrack_l3proto *l3proto __read_mostly;
  37
  38 /* Calculated at init based on memory size */
  39 static unsigned int nf_nat_htable_size __read_mostly;
  40
  41 #define MAX_IP_NAT_PROTO 256
  42 static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
  43                                                 __read_mostly;
  44
  45 static inline const struct nf_nat_protocol *
  46 __nf_nat_proto_find(u_int8_t protonum)
  47 {
  48         return rcu_dereference(nf_nat_protos[protonum]);
  49 }
  50
  51 const struct nf_nat_protocol *
  52 nf_nat_proto_find_get(u_int8_t protonum)
  53 {
  54         const struct nf_nat_protocol *p;
  55
  56         rcu_read_lock();
  57         p = __nf_nat_proto_find(protonum);
  58         if (!try_module_get(p->me))
  59                 p = &nf_nat_unknown_protocol;
  60         rcu_read_unlock();
  61
  62         return p;
  63 }
  64 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
  65
  66 void
  67 nf_nat_proto_put(const struct nf_nat_protocol *p)
  68 {
  69         module_put(p->me);
  70 }
  71 EXPORT_SYMBOL_GPL(nf_nat_proto_put);
  72
  73 /* We keep an extra hash for each conntrack, for fast searching. */
  74 static inline unsigned int
  75 hash_by_src(const struct nf_conntrack_tuple *tuple)
  76 {
  77         unsigned int hash;
  78
  79         /* Original src, to ensure we map it consistently if poss. */
  80         hash = jhash_3words((__force u32)tuple->src.u3.ip,
  81                             (__force u32)tuple->src.u.all,
  82                             tuple->dst.protonum, 0);
  83         return ((u64)hash * nf_nat_htable_size) >> 32;
  84 }
  85
  86 /* Is this tuple already taken? (not by us) */
  87 int
  88 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
  89                   const struct nf_conn *ignored_conntrack)
  90 {
  91         /* Conntrack tracking doesn't keep track of outgoing tuples; only
  92            incoming ones.  NAT means they don't have a fixed mapping,
  93            so we invert the tuple and look for the incoming reply.
  94
  95            We could keep a separate hash if this proves too slow. */
  96         struct nf_conntrack_tuple reply;
  97
  98         nf_ct_invert_tuplepr(&reply, tuple);
  99         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 100 }
 101 EXPORT_SYMBOL(nf_nat_used_tuple);
 102
 103 /* If we source map this tuple so reply looks like reply_tuple, will
 104  * that meet the constraints of range. */
 105 static int
 106 in_range(const struct nf_conntrack_tuple *tuple,
 107          const struct nf_nat_range *range)
 108 {
 109         const struct nf_nat_protocol *proto;
 110         int ret = 0;
 111
 112         /* If we are supposed to map IPs, then we must be in the
 113            range specified, otherwise let this drag us onto a new src IP. */
 114         if (range->flags & IP_NAT_RANGE_MAP_IPS) {
 115                 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
 116                     ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
 117                         return 0;
 118         }
 119
 120         rcu_read_lock();
 121         proto = __nf_nat_proto_find(tuple->dst.protonum);
 122         if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 123             proto->in_range(tuple, IP_NAT_MANIP_SRC,
 124                             &range->min, &range->max))
 125                 ret = 1;
 126         rcu_read_unlock();
 127
 128         return ret;
 129 }
 130
 131 static inline int
 132 same_src(const struct nf_conn *ct,
 133          const struct nf_conntrack_tuple *tuple)
 134 {
 135         const struct nf_conntrack_tuple *t;
 136
 137         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 138         return (t->dst.protonum == tuple->dst.protonum &&
 139                 t->src.u3.ip == tuple->src.u3.ip &&
 140                 t->src.u.all == tuple->src.u.all);
 141 }
 142
 143 /* Only called for SRC manip */
 144 static int
 145 find_appropriate_src(struct net *net,
 146                      const struct nf_conntrack_tuple *tuple,
 147                      struct nf_conntrack_tuple *result,
 148                      const struct nf_nat_range *range)
 149 {
 150         unsigned int h = hash_by_src(tuple);
 151         const struct nf_conn_nat *nat;
 152         const struct nf_conn *ct;
 153         const struct hlist_node *n;
 154
 155         rcu_read_lock();
 156         hlist_for_each_entry_rcu(nat, n, &net->ipv4.nat_bysource[h], bysource) {
 157                 ct = nat->ct;
 158                 if (same_src(ct, tuple)) {
 159                         /* Copy source part from reply tuple. */
 160                         nf_ct_invert_tuplepr(result,
 161                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 162                         result->dst = tuple->dst;
 163
 164                         if (in_range(result, range)) {
 165                                 rcu_read_unlock();
 166                                 return 1;
 167                         }
 168                 }
 169         }
 170         rcu_read_unlock();
 171         return 0;
 172 }
 173
 174 /* For [FUTURE] fragmentation handling, we want the least-used
 175    src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 176    if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 177    1-65535, we don't do pro-rata allocation based on ports; we choose
 178    the ip with the lowest src-ip/dst-ip/proto usage.
 179 */
 180 static void
 181 find_best_ips_proto(struct nf_conntrack_tuple *tuple,
 182                     const struct nf_nat_range *range,
 183                     const struct nf_conn *ct,
 184                     enum nf_nat_manip_type maniptype)
 185 {
 186         __be32 *var_ipp;
 187         /* Host order */
 188         u_int32_t minip, maxip, j;
 189
 190         /* No IP mapping?  Do nothing. */
 191         if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
 192                 return;
 193
 194         if (maniptype == IP_NAT_MANIP_SRC)
 195                 var_ipp = &tuple->src.u3.ip;
 196         else
 197                 var_ipp = &tuple->dst.u3.ip;
 198
 199         /* Fast path: only one choice. */
 200         if (range->min_ip == range->max_ip) {
 201                 *var_ipp = range->min_ip;
 202                 return;
 203         }
 204
 205         /* Hashing source and destination IPs gives a fairly even
 206          * spread in practice (if there are a small number of IPs
 207          * involved, there usually aren't that many connections
 208          * anyway).  The consistency means that servers see the same
 209          * client coming from the same IP (some Internet Banking sites
 210          * like this), even across reboots. */
 211         minip = ntohl(range->min_ip);
 212         maxip = ntohl(range->max_ip);
 213         j = jhash_2words((__force u32)tuple->src.u3.ip,
 214                          range->flags & IP_NAT_RANGE_PERSISTENT ?
 215                                 (__force u32)tuple->dst.u3.ip : 0, 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
 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, 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(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(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 334                 spin_lock_bh(&nf_nat_lock);
 335                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
 336                 nat = nfct_nat(ct);
 337                 nat->ct = ct;
 338                 hlist_add_head_rcu(&nat->bysource,
 339                                    &net->ipv4.nat_bysource[srchash]);
 340                 spin_unlock_bh(&nf_nat_lock);
 341         }
 342
 343         /* It's done. */
 344         if (maniptype == IP_NAT_MANIP_DST)
 345                 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
 346         else
 347                 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 348
 349         return NF_ACCEPT;
 350 }
 351 EXPORT_SYMBOL(nf_nat_setup_info);
 352
 353 /* Returns true if succeeded. */
 354 static bool
 355 manip_pkt(u_int16_t proto,
 356           struct sk_buff *skb,
 357           unsigned int iphdroff,
 358           const struct nf_conntrack_tuple *target,
 359           enum nf_nat_manip_type maniptype)
 360 {
 361         struct iphdr *iph;
 362         const struct nf_nat_protocol *p;
 363
 364         if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
 365                 return false;
 366
 367         iph = (void *)skb->data + iphdroff;
 368
 369         /* Manipulate protcol part. */
 370
 371         /* rcu_read_lock()ed by nf_hook_slow */
 372         p = __nf_nat_proto_find(proto);
 373         if (!p->manip_pkt(skb, iphdroff, target, maniptype))
 374                 return false;
 375
 376         iph = (void *)skb->data + iphdroff;
 377
 378         if (maniptype == IP_NAT_MANIP_SRC) {
 379                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 380                 iph->saddr = target->src.u3.ip;
 381         } else {
 382                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 383                 iph->daddr = target->dst.u3.ip;
 384         }
 385         return true;
 386 }
 387
 388 /* Do packet manipulations according to nf_nat_setup_info. */
 389 unsigned int nf_nat_packet(struct nf_conn *ct,
 390                            enum ip_conntrack_info ctinfo,
 391                            unsigned int hooknum,
 392                            struct sk_buff *skb)
 393 {
 394         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 395         unsigned long statusbit;
 396         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 397
 398         if (mtype == IP_NAT_MANIP_SRC)
 399                 statusbit = IPS_SRC_NAT;
 400         else
 401                 statusbit = IPS_DST_NAT;
 402
 403         /* Invert if this is reply dir. */
 404         if (dir == IP_CT_DIR_REPLY)
 405                 statusbit ^= IPS_NAT_MASK;
 406
 407         /* Non-atomic: these bits don't change. */
 408         if (ct->status & statusbit) {
 409                 struct nf_conntrack_tuple target;
 410
 411                 /* We are aiming to look like inverse of other direction. */
 412                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 413
 414                 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
 415                         return NF_DROP;
 416         }
 417         return NF_ACCEPT;
 418 }
 419 EXPORT_SYMBOL_GPL(nf_nat_packet);
 420
 421 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
 422 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 423                                   enum ip_conntrack_info ctinfo,
 424                                   unsigned int hooknum,
 425                                   struct sk_buff *skb)
 426 {
 427         struct {
 428                 struct icmphdr icmp;
 429                 struct iphdr ip;
 430         } *inside;
 431         const struct nf_conntrack_l4proto *l4proto;
 432         struct nf_conntrack_tuple inner, target;
 433         int hdrlen = ip_hdrlen(skb);
 434         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 435         unsigned long statusbit;
 436         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 437
 438         if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
 439                 return 0;
 440
 441         inside = (void *)skb->data + ip_hdrlen(skb);
 442
 443         /* We're actually going to mangle it beyond trivial checksum
 444            adjustment, so make sure the current checksum is correct. */
 445         if (nf_ip_checksum(skb, hooknum, hdrlen, 0))
 446                 return 0;
 447
 448         /* Must be RELATED */
 449         NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
 450                      skb->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
 451
 452         /* Redirects on non-null nats must be dropped, else they'll
 453            start talking to each other without our translation, and be
 454            confused... --RR */
 455         if (inside->icmp.type == ICMP_REDIRECT) {
 456                 /* If NAT isn't finished, assume it and drop. */
 457                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 458                         return 0;
 459
 460                 if (ct->status & IPS_NAT_MASK)
 461                         return 0;
 462         }
 463
 464         pr_debug("icmp_reply_translation: translating error %p manip %u "
 465                  "dir %s\n", skb, manip,
 466                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 467
 468         /* rcu_read_lock()ed by nf_hook_slow */
 469         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
 470
 471         if (!nf_ct_get_tuple(skb,
 472                              ip_hdrlen(skb) + sizeof(struct icmphdr),
 473                              (ip_hdrlen(skb) +
 474                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 475                              (u_int16_t)AF_INET,
 476                              inside->ip.protocol,
 477                              &inner, l3proto, l4proto))
 478                 return 0;
 479
 480         /* Change inner back to look like incoming packet.  We do the
 481            opposite manip on this hook to normal, because it might not
 482            pass all hooks (locally-generated ICMP).  Consider incoming
 483            packet: PREROUTING (DST manip), routing produces ICMP, goes
 484            through POSTROUTING (which must correct the DST manip). */
 485         if (!manip_pkt(inside->ip.protocol, skb,
 486                        ip_hdrlen(skb) + sizeof(inside->icmp),
 487                        &ct->tuplehash[!dir].tuple,
 488                        !manip))
 489                 return 0;
 490
 491         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 492                 /* Reloading "inside" here since manip_pkt inner. */
 493                 inside = (void *)skb->data + ip_hdrlen(skb);
 494                 inside->icmp.checksum = 0;
 495                 inside->icmp.checksum =
 496                         csum_fold(skb_checksum(skb, hdrlen,
 497                                                skb->len - hdrlen, 0));
 498         }
 499
 500         /* Change outer to look the reply to an incoming packet
 501          * (proto 0 means don't invert per-proto part). */
 502         if (manip == IP_NAT_MANIP_SRC)
 503                 statusbit = IPS_SRC_NAT;
 504         else
 505                 statusbit = IPS_DST_NAT;
 506
 507         /* Invert if this is reply dir. */
 508         if (dir == IP_CT_DIR_REPLY)
 509                 statusbit ^= IPS_NAT_MASK;
 510
 511         if (ct->status & statusbit) {
 512                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 513                 if (!manip_pkt(0, skb, 0, &target, manip))
 514                         return 0;
 515         }
 516
 517         return 1;
 518 }
 519 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 520
 521 /* Protocol registration. */
 522 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 523 {
 524         int ret = 0;
 525
 526         spin_lock_bh(&nf_nat_lock);
 527         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
 528                 ret = -EBUSY;
 529                 goto out;
 530         }
 531         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 532  out:
 533         spin_unlock_bh(&nf_nat_lock);
 534         return ret;
 535 }
 536 EXPORT_SYMBOL(nf_nat_protocol_register);
 537
 538 /* Noone stores the protocol anywhere; simply delete it. */
 539 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 540 {
 541         spin_lock_bh(&nf_nat_lock);
 542         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 543                            &nf_nat_unknown_protocol);
 544         spin_unlock_bh(&nf_nat_lock);
 545         synchronize_rcu();
 546 }
 547 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 548
 549 /* Noone using conntrack by the time this called. */
 550 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 551 {
 552         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 553
 554         if (nat == NULL || nat->ct == NULL)
 555                 return;
 556
 557         NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
 558
 559         spin_lock_bh(&nf_nat_lock);
 560         hlist_del_rcu(&nat->bysource);
 561         spin_unlock_bh(&nf_nat_lock);
 562 }
 563
 564 static void nf_nat_move_storage(void *new, void *old)
 565 {
 566         struct nf_conn_nat *new_nat = new;
 567         struct nf_conn_nat *old_nat = old;
 568         struct nf_conn *ct = old_nat->ct;
 569
 570         if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
 571                 return;
 572
 573         spin_lock_bh(&nf_nat_lock);
 574         new_nat->ct = ct;
 575         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 576         spin_unlock_bh(&nf_nat_lock);
 577 }
 578
 579 static struct nf_ct_ext_type nat_extend __read_mostly = {
 580         .len            = sizeof(struct nf_conn_nat),
 581         .align          = __alignof__(struct nf_conn_nat),
 582         .destroy        = nf_nat_cleanup_conntrack,
 583         .move           = nf_nat_move_storage,
 584         .id             = NF_CT_EXT_NAT,
 585         .flags          = NF_CT_EXT_F_PREALLOC,
 586 };
 587
 588 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 589
 590 #include <linux/netfilter/nfnetlink.h>
 591 #include <linux/netfilter/nfnetlink_conntrack.h>
 592
 593 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 594         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 595         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 596 };
 597
 598 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 599                                      const struct nf_conn *ct,
 600                                      struct nf_nat_range *range)
 601 {
 602         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 603         const struct nf_nat_protocol *npt;
 604         int err;
 605
 606         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 607         if (err < 0)
 608                 return err;
 609
 610         npt = nf_nat_proto_find_get(nf_ct_protonum(ct));
 611         if (npt->nlattr_to_range)
 612                 err = npt->nlattr_to_range(tb, range);
 613         nf_nat_proto_put(npt);
 614         return err;
 615 }
 616
 617 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 618         [CTA_NAT_MINIP]         = { .type = NLA_U32 },
 619         [CTA_NAT_MAXIP]         = { .type = NLA_U32 },
 620 };
 621
 622 static int
 623 nfnetlink_parse_nat(struct nlattr *nat,
 624                     const struct nf_conn *ct, struct nf_nat_range *range)
 625 {
 626         struct nlattr *tb[CTA_NAT_MAX+1];
 627         int err;
 628
 629         memset(range, 0, sizeof(*range));
 630
 631         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 632         if (err < 0)
 633                 return err;
 634
 635         if (tb[CTA_NAT_MINIP])
 636                 range->min_ip = nla_get_be32(tb[CTA_NAT_MINIP]);
 637
 638         if (!tb[CTA_NAT_MAXIP])
 639                 range->max_ip = range->min_ip;
 640         else
 641                 range->max_ip = nla_get_be32(tb[CTA_NAT_MAXIP]);
 642
 643         if (range->min_ip)
 644                 range->flags |= IP_NAT_RANGE_MAP_IPS;
 645
 646         if (!tb[CTA_NAT_PROTO])
 647                 return 0;
 648
 649         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 650         if (err < 0)
 651                 return err;
 652
 653         return 0;
 654 }
 655
 656 static int
 657 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 658                           enum nf_nat_manip_type manip,
 659                           struct nlattr *attr)
 660 {
 661         struct nf_nat_range range;
 662
 663         if (nfnetlink_parse_nat(attr, ct, &range) < 0)
 664                 return -EINVAL;
 665         if (nf_nat_initialized(ct, manip))
 666                 return -EEXIST;
 667
 668         return nf_nat_setup_info(ct, &range, manip);
 669 }
 670 #else
 671 static int
 672 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 673                           enum nf_nat_manip_type manip,
 674                           struct nlattr *attr)
 675 {
 676         return -EOPNOTSUPP;
 677 }
 678 #endif
 679
 680 static int __net_init nf_nat_net_init(struct net *net)
 681 {
 682         net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
 683                                                       &net->ipv4.nat_vmalloced, 0);
 684         if (!net->ipv4.nat_bysource)
 685                 return -ENOMEM;
 686         return 0;
 687 }
 688
 689 /* Clear NAT section of all conntracks, in case we're loaded again. */
 690 static int clean_nat(struct nf_conn *i, void *data)
 691 {
 692         struct nf_conn_nat *nat = nfct_nat(i);
 693
 694         if (!nat)
 695                 return 0;
 696         memset(nat, 0, sizeof(*nat));
 697         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
 698         return 0;
 699 }
 700
 701 static void __net_exit nf_nat_net_exit(struct net *net)
 702 {
 703         nf_ct_iterate_cleanup(net, &clean_nat, NULL);
 704         synchronize_rcu();
 705         nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_vmalloced,
 706                              nf_nat_htable_size);
 707 }
 708
 709 static struct pernet_operations nf_nat_net_ops = {
 710         .init = nf_nat_net_init,
 711         .exit = nf_nat_net_exit,
 712 };
 713
 714 static int __init nf_nat_init(void)
 715 {
 716         size_t i;
 717         int ret;
 718
 719         need_ipv4_conntrack();
 720
 721         ret = nf_ct_extend_register(&nat_extend);
 722         if (ret < 0) {
 723                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 724                 return ret;
 725         }
 726
 727         /* Leave them the same for the moment. */
 728         nf_nat_htable_size = nf_conntrack_htable_size;
 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         return 0;
 754
 755  cleanup_extend:
 756         nf_ct_extend_unregister(&nat_extend);
 757         return ret;
 758 }
 759
 760 static void __exit nf_nat_cleanup(void)
 761 {
 762         unregister_pernet_subsys(&nf_nat_net_ops);
 763         nf_ct_l3proto_put(l3proto);
 764         nf_ct_extend_unregister(&nat_extend);
 765         rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL);
 766         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, NULL);
 767         synchronize_net();
 768 }
 769
 770 MODULE_LICENSE("GPL");
 771 MODULE_ALIAS("nf-nat-ipv4");
 772
 773 module_init(nf_nat_init);
 774 module_exit(nf_nat_cleanup);