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 __rcu *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 /* We keep an extra hash for each conntrack, for fast searching. */
  51 static inline unsigned int
  52 hash_by_src(const struct net *net, u16 zone,
  53             const struct nf_conntrack_tuple *tuple)
  54 {
  55         unsigned int hash;
  56
  57         /* Original src, to ensure we map it consistently if poss. */
  58         hash = jhash_3words((__force u32)tuple->src.u3.ip,
  59                             (__force u32)tuple->src.u.all ^ zone,
  60                             tuple->dst.protonum, 0);
  61         return ((u64)hash * net->ipv4.nat_htable_size) >> 32;
  62 }
  63
  64 /* Is this tuple already taken? (not by us) */
  65 int
  66 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
  67                   const struct nf_conn *ignored_conntrack)
  68 {
  69         /* Conntrack tracking doesn't keep track of outgoing tuples; only
  70            incoming ones.  NAT means they don't have a fixed mapping,
  71            so we invert the tuple and look for the incoming reply.
  72
  73            We could keep a separate hash if this proves too slow. */
  74         struct nf_conntrack_tuple reply;
  75
  76         nf_ct_invert_tuplepr(&reply, tuple);
  77         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
  78 }
  79 EXPORT_SYMBOL(nf_nat_used_tuple);
  80
  81 /* If we source map this tuple so reply looks like reply_tuple, will
  82  * that meet the constraints of range. */
  83 static int
  84 in_range(const struct nf_conntrack_tuple *tuple,
  85          const struct nf_nat_range *range)
  86 {
  87         const struct nf_nat_protocol *proto;
  88         int ret = 0;
  89
  90         /* If we are supposed to map IPs, then we must be in the
  91            range specified, otherwise let this drag us onto a new src IP. */
  92         if (range->flags & IP_NAT_RANGE_MAP_IPS) {
  93                 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
  94                     ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
  95                         return 0;
  96         }
  97
  98         rcu_read_lock();
  99         proto = __nf_nat_proto_find(tuple->dst.protonum);
 100         if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
 101             proto->in_range(tuple, IP_NAT_MANIP_SRC,
 102                             &range->min, &range->max))
 103                 ret = 1;
 104         rcu_read_unlock();
 105
 106         return ret;
 107 }
 108
 109 static inline int
 110 same_src(const struct nf_conn *ct,
 111          const struct nf_conntrack_tuple *tuple)
 112 {
 113         const struct nf_conntrack_tuple *t;
 114
 115         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 116         return (t->dst.protonum == tuple->dst.protonum &&
 117                 t->src.u3.ip == tuple->src.u3.ip &&
 118                 t->src.u.all == tuple->src.u.all);
 119 }
 120
 121 /* Only called for SRC manip */
 122 static int
 123 find_appropriate_src(struct net *net, u16 zone,
 124                      const struct nf_conntrack_tuple *tuple,
 125                      struct nf_conntrack_tuple *result,
 126                      const struct nf_nat_range *range)
 127 {
 128         unsigned int h = hash_by_src(net, zone, tuple);
 129         const struct nf_conn_nat *nat;
 130         const struct nf_conn *ct;
 131         const struct hlist_node *n;
 132
 133         rcu_read_lock();
 134         hlist_for_each_entry_rcu(nat, n, &net->ipv4.nat_bysource[h], bysource) {
 135                 ct = nat->ct;
 136                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
 137                         /* Copy source part from reply tuple. */
 138                         nf_ct_invert_tuplepr(result,
 139                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 140                         result->dst = tuple->dst;
 141
 142                         if (in_range(result, range)) {
 143                                 rcu_read_unlock();
 144                                 return 1;
 145                         }
 146                 }
 147         }
 148         rcu_read_unlock();
 149         return 0;
 150 }
 151
 152 /* For [FUTURE] fragmentation handling, we want the least-used
 153    src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 154    if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 155    1-65535, we don't do pro-rata allocation based on ports; we choose
 156    the ip with the lowest src-ip/dst-ip/proto usage.
 157 */
 158 static void
 159 find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple,
 160                     const struct nf_nat_range *range,
 161                     const struct nf_conn *ct,
 162                     enum nf_nat_manip_type maniptype)
 163 {
 164         __be32 *var_ipp;
 165         /* Host order */
 166         u_int32_t minip, maxip, j;
 167
 168         /* No IP mapping?  Do nothing. */
 169         if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
 170                 return;
 171
 172         if (maniptype == IP_NAT_MANIP_SRC)
 173                 var_ipp = &tuple->src.u3.ip;
 174         else
 175                 var_ipp = &tuple->dst.u3.ip;
 176
 177         /* Fast path: only one choice. */
 178         if (range->min_ip == range->max_ip) {
 179                 *var_ipp = range->min_ip;
 180                 return;
 181         }
 182
 183         /* Hashing source and destination IPs gives a fairly even
 184          * spread in practice (if there are a small number of IPs
 185          * involved, there usually aren't that many connections
 186          * anyway).  The consistency means that servers see the same
 187          * client coming from the same IP (some Internet Banking sites
 188          * like this), even across reboots. */
 189         minip = ntohl(range->min_ip);
 190         maxip = ntohl(range->max_ip);
 191         j = jhash_2words((__force u32)tuple->src.u3.ip,
 192                          range->flags & IP_NAT_RANGE_PERSISTENT ?
 193                                 0 : (__force u32)tuple->dst.u3.ip ^ zone, 0);
 194         j = ((u64)j * (maxip - minip + 1)) >> 32;
 195         *var_ipp = htonl(minip + j);
 196 }
 197
 198 /* Manipulate the tuple into the range given.  For NF_INET_POST_ROUTING,
 199  * we change the source to map into the range.  For NF_INET_PRE_ROUTING
 200  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 201  * range.  It might not be possible to get a unique tuple, but we try.
 202  * At worst (or if we race), we will end up with a final duplicate in
 203  * __ip_conntrack_confirm and drop the packet. */
 204 static void
 205 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 206                  const struct nf_conntrack_tuple *orig_tuple,
 207                  const struct nf_nat_range *range,
 208                  struct nf_conn *ct,
 209                  enum nf_nat_manip_type maniptype)
 210 {
 211         struct net *net = nf_ct_net(ct);
 212         const struct nf_nat_protocol *proto;
 213         u16 zone = nf_ct_zone(ct);
 214
 215         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 216            and that same mapping gives a unique tuple within the given
 217            range, use that.
 218
 219            This is only required for source (ie. NAT/masq) mappings.
 220            So far, we don't do local source mappings, so multiple
 221            manips not an issue.  */
 222         if (maniptype == IP_NAT_MANIP_SRC &&
 223             !(range->flags & IP_NAT_RANGE_PROTO_RANDOM)) {
 224                 /* try the original tuple first */
 225                 if (in_range(orig_tuple, range)) {
 226                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 227                                 *tuple = *orig_tuple;
 228                                 return;
 229                         }
 230                 } else if (find_appropriate_src(net, zone, orig_tuple, tuple,
 231                            range)) {
 232                         pr_debug("get_unique_tuple: Found current src map\n");
 233                         if (!nf_nat_used_tuple(tuple, ct))
 234                                 return;
 235                 }
 236         }
 237
 238         /* 2) Select the least-used IP/proto combination in the given
 239            range. */
 240         *tuple = *orig_tuple;
 241         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 242
 243         /* 3) The per-protocol part of the manip is made to map into
 244            the range to make a unique tuple. */
 245
 246         rcu_read_lock();
 247         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
 248
 249         /* Only bother mapping if it's not already in range and unique */
 250         if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM)) {
 251                 if (range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) {
 252                         if (proto->in_range(tuple, maniptype, &range->min,
 253                                             &range->max) &&
 254                             (range->min.all == range->max.all ||
 255                              !nf_nat_used_tuple(tuple, ct)))
 256                                 goto out;
 257                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 258                         goto out;
 259                 }
 260         }
 261
 262         /* Last change: get protocol to try to obtain unique tuple. */
 263         proto->unique_tuple(tuple, range, maniptype, ct);
 264 out:
 265         rcu_read_unlock();
 266 }
 267
 268 unsigned int
 269 nf_nat_setup_info(struct nf_conn *ct,
 270                   const struct nf_nat_range *range,
 271                   enum nf_nat_manip_type maniptype)
 272 {
 273         struct net *net = nf_ct_net(ct);
 274         struct nf_conntrack_tuple curr_tuple, new_tuple;
 275         struct nf_conn_nat *nat;
 276
 277         /* nat helper or nfctnetlink also setup binding */
 278         nat = nfct_nat(ct);
 279         if (!nat) {
 280                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 281                 if (nat == NULL) {
 282                         pr_debug("failed to add NAT extension\n");
 283                         return NF_ACCEPT;
 284                 }
 285         }
 286
 287         NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC ||
 288                      maniptype == IP_NAT_MANIP_DST);
 289         BUG_ON(nf_nat_initialized(ct, maniptype));
 290
 291         /* What we've got will look like inverse of reply. Normally
 292            this is what is in the conntrack, except for prior
 293            manipulations (future optimization: if num_manips == 0,
 294            orig_tp =
 295            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
 296         nf_ct_invert_tuplepr(&curr_tuple,
 297                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 298
 299         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 300
 301         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 302                 struct nf_conntrack_tuple reply;
 303
 304                 /* Alter conntrack table so will recognize replies. */
 305                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 306                 nf_conntrack_alter_reply(ct, &reply);
 307
 308                 /* Non-atomic: we own this at the moment. */
 309                 if (maniptype == IP_NAT_MANIP_SRC)
 310                         ct->status |= IPS_SRC_NAT;
 311                 else
 312                         ct->status |= IPS_DST_NAT;
 313         }
 314
 315         if (maniptype == IP_NAT_MANIP_SRC) {
 316                 unsigned int srchash;
 317
 318                 srchash = hash_by_src(net, nf_ct_zone(ct),
 319                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 320                 spin_lock_bh(&nf_nat_lock);
 321                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
 322                 nat = nfct_nat(ct);
 323                 nat->ct = ct;
 324                 hlist_add_head_rcu(&nat->bysource,
 325                                    &net->ipv4.nat_bysource[srchash]);
 326                 spin_unlock_bh(&nf_nat_lock);
 327         }
 328
 329         /* It's done. */
 330         if (maniptype == IP_NAT_MANIP_DST)
 331                 ct->status |= IPS_DST_NAT_DONE;
 332         else
 333                 ct->status |= IPS_SRC_NAT_DONE;
 334
 335         return NF_ACCEPT;
 336 }
 337 EXPORT_SYMBOL(nf_nat_setup_info);
 338
 339 /* Returns true if succeeded. */
 340 static bool
 341 manip_pkt(u_int16_t proto,
 342           struct sk_buff *skb,
 343           unsigned int iphdroff,
 344           const struct nf_conntrack_tuple *target,
 345           enum nf_nat_manip_type maniptype)
 346 {
 347         struct iphdr *iph;
 348         const struct nf_nat_protocol *p;
 349
 350         if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
 351                 return false;
 352
 353         iph = (void *)skb->data + iphdroff;
 354
 355         /* Manipulate protcol part. */
 356
 357         /* rcu_read_lock()ed by nf_hook_slow */
 358         p = __nf_nat_proto_find(proto);
 359         if (!p->manip_pkt(skb, iphdroff, target, maniptype))
 360                 return false;
 361
 362         iph = (void *)skb->data + iphdroff;
 363
 364         if (maniptype == IP_NAT_MANIP_SRC) {
 365                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 366                 iph->saddr = target->src.u3.ip;
 367         } else {
 368                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 369                 iph->daddr = target->dst.u3.ip;
 370         }
 371         return true;
 372 }
 373
 374 /* Do packet manipulations according to nf_nat_setup_info. */
 375 unsigned int nf_nat_packet(struct nf_conn *ct,
 376                            enum ip_conntrack_info ctinfo,
 377                            unsigned int hooknum,
 378                            struct sk_buff *skb)
 379 {
 380         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 381         unsigned long statusbit;
 382         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 383
 384         if (mtype == IP_NAT_MANIP_SRC)
 385                 statusbit = IPS_SRC_NAT;
 386         else
 387                 statusbit = IPS_DST_NAT;
 388
 389         /* Invert if this is reply dir. */
 390         if (dir == IP_CT_DIR_REPLY)
 391                 statusbit ^= IPS_NAT_MASK;
 392
 393         /* Non-atomic: these bits don't change. */
 394         if (ct->status & statusbit) {
 395                 struct nf_conntrack_tuple target;
 396
 397                 /* We are aiming to look like inverse of other direction. */
 398                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 399
 400                 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
 401                         return NF_DROP;
 402         }
 403         return NF_ACCEPT;
 404 }
 405 EXPORT_SYMBOL_GPL(nf_nat_packet);
 406
 407 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
 408 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 409                                   enum ip_conntrack_info ctinfo,
 410                                   unsigned int hooknum,
 411                                   struct sk_buff *skb)
 412 {
 413         struct {
 414                 struct icmphdr icmp;
 415                 struct iphdr ip;
 416         } *inside;
 417         const struct nf_conntrack_l4proto *l4proto;
 418         struct nf_conntrack_tuple inner, target;
 419         int hdrlen = ip_hdrlen(skb);
 420         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 421         unsigned long statusbit;
 422         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 423
 424         if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
 425                 return 0;
 426
 427         inside = (void *)skb->data + hdrlen;
 428
 429         /* We're actually going to mangle it beyond trivial checksum
 430            adjustment, so make sure the current checksum is correct. */
 431         if (nf_ip_checksum(skb, hooknum, hdrlen, 0))
 432                 return 0;
 433
 434         /* Must be RELATED */
 435         NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
 436                      skb->nfctinfo == IP_CT_RELATED_REPLY);
 437
 438         /* Redirects on non-null nats must be dropped, else they'll
 439            start talking to each other without our translation, and be
 440            confused... --RR */
 441         if (inside->icmp.type == ICMP_REDIRECT) {
 442                 /* If NAT isn't finished, assume it and drop. */
 443                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 444                         return 0;
 445
 446                 if (ct->status & IPS_NAT_MASK)
 447                         return 0;
 448         }
 449
 450         if (manip == IP_NAT_MANIP_SRC)
 451                 statusbit = IPS_SRC_NAT;
 452         else
 453                 statusbit = IPS_DST_NAT;
 454
 455         /* Invert if this is reply dir. */
 456         if (dir == IP_CT_DIR_REPLY)
 457                 statusbit ^= IPS_NAT_MASK;
 458
 459         if (!(ct->status & statusbit))
 460                 return 1;
 461
 462         pr_debug("icmp_reply_translation: translating error %p manip %u "
 463                  "dir %s\n", skb, manip,
 464                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 465
 466         /* rcu_read_lock()ed by nf_hook_slow */
 467         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
 468
 469         if (!nf_ct_get_tuple(skb, hdrlen + sizeof(struct icmphdr),
 470                              (hdrlen +
 471                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 472                              (u_int16_t)AF_INET, inside->ip.protocol,
 473                              &inner, l3proto, l4proto))
 474                 return 0;
 475
 476         /* Change inner back to look like incoming packet.  We do the
 477            opposite manip on this hook to normal, because it might not
 478            pass all hooks (locally-generated ICMP).  Consider incoming
 479            packet: PREROUTING (DST manip), routing produces ICMP, goes
 480            through POSTROUTING (which must correct the DST manip). */
 481         if (!manip_pkt(inside->ip.protocol, skb, hdrlen + sizeof(inside->icmp),
 482                        &ct->tuplehash[!dir].tuple, !manip))
 483                 return 0;
 484
 485         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 486                 /* Reloading "inside" here since manip_pkt inner. */
 487                 inside = (void *)skb->data + hdrlen;
 488                 inside->icmp.checksum = 0;
 489                 inside->icmp.checksum =
 490                         csum_fold(skb_checksum(skb, hdrlen,
 491                                                skb->len - hdrlen, 0));
 492         }
 493
 494         /* Change outer to look the reply to an incoming packet
 495          * (proto 0 means don't invert per-proto part). */
 496         nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 497         if (!manip_pkt(0, skb, 0, &target, manip))
 498                 return 0;
 499
 500         return 1;
 501 }
 502 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 503
 504 /* Protocol registration. */
 505 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 506 {
 507         int ret = 0;
 508
 509         spin_lock_bh(&nf_nat_lock);
 510         if (rcu_dereference_protected(
 511                         nf_nat_protos[proto->protonum],
 512                         lockdep_is_held(&nf_nat_lock)
 513                         ) != &nf_nat_unknown_protocol) {
 514                 ret = -EBUSY;
 515                 goto out;
 516         }
 517         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 518  out:
 519         spin_unlock_bh(&nf_nat_lock);
 520         return ret;
 521 }
 522 EXPORT_SYMBOL(nf_nat_protocol_register);
 523
 524 /* No one stores the protocol anywhere; simply delete it. */
 525 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 526 {
 527         spin_lock_bh(&nf_nat_lock);
 528         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 529                            &nf_nat_unknown_protocol);
 530         spin_unlock_bh(&nf_nat_lock);
 531         synchronize_rcu();
 532 }
 533 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 534
 535 /* No one using conntrack by the time this called. */
 536 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 537 {
 538         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 539
 540         if (nat == NULL || nat->ct == NULL)
 541                 return;
 542
 543         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 544
 545         spin_lock_bh(&nf_nat_lock);
 546         hlist_del_rcu(&nat->bysource);
 547         spin_unlock_bh(&nf_nat_lock);
 548 }
 549
 550 static void nf_nat_move_storage(void *new, void *old)
 551 {
 552         struct nf_conn_nat *new_nat = new;
 553         struct nf_conn_nat *old_nat = old;
 554         struct nf_conn *ct = old_nat->ct;
 555
 556         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 557                 return;
 558
 559         spin_lock_bh(&nf_nat_lock);
 560         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 561         spin_unlock_bh(&nf_nat_lock);
 562 }
 563
 564 static struct nf_ct_ext_type nat_extend __read_mostly = {
 565         .len            = sizeof(struct nf_conn_nat),
 566         .align          = __alignof__(struct nf_conn_nat),
 567         .destroy        = nf_nat_cleanup_conntrack,
 568         .move           = nf_nat_move_storage,
 569         .id             = NF_CT_EXT_NAT,
 570         .flags          = NF_CT_EXT_F_PREALLOC,
 571 };
 572
 573 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 574
 575 #include <linux/netfilter/nfnetlink.h>
 576 #include <linux/netfilter/nfnetlink_conntrack.h>
 577
 578 static const struct nf_nat_protocol *
 579 nf_nat_proto_find_get(u_int8_t protonum)
 580 {
 581         const struct nf_nat_protocol *p;
 582
 583         rcu_read_lock();
 584         p = __nf_nat_proto_find(protonum);
 585         if (!try_module_get(p->me))
 586                 p = &nf_nat_unknown_protocol;
 587         rcu_read_unlock();
 588
 589         return p;
 590 }
 591
 592 static void
 593 nf_nat_proto_put(const struct nf_nat_protocol *p)
 594 {
 595         module_put(p->me);
 596 }
 597
 598 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 599         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 600         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 601 };
 602
 603 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 604                                      const struct nf_conn *ct,
 605                                      struct nf_nat_range *range)
 606 {
 607         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 608         const struct nf_nat_protocol *npt;
 609         int err;
 610
 611         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 612         if (err < 0)
 613                 return err;
 614
 615         npt = nf_nat_proto_find_get(nf_ct_protonum(ct));
 616         if (npt->nlattr_to_range)
 617                 err = npt->nlattr_to_range(tb, range);
 618         nf_nat_proto_put(npt);
 619         return err;
 620 }
 621
 622 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 623         [CTA_NAT_MINIP]         = { .type = NLA_U32 },
 624         [CTA_NAT_MAXIP]         = { .type = NLA_U32 },
 625 };
 626
 627 static int
 628 nfnetlink_parse_nat(const struct nlattr *nat,
 629                     const struct nf_conn *ct, struct nf_nat_range *range)
 630 {
 631         struct nlattr *tb[CTA_NAT_MAX+1];
 632         int err;
 633
 634         memset(range, 0, sizeof(*range));
 635
 636         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 637         if (err < 0)
 638                 return err;
 639
 640         if (tb[CTA_NAT_MINIP])
 641                 range->min_ip = nla_get_be32(tb[CTA_NAT_MINIP]);
 642
 643         if (!tb[CTA_NAT_MAXIP])
 644                 range->max_ip = range->min_ip;
 645         else
 646                 range->max_ip = nla_get_be32(tb[CTA_NAT_MAXIP]);
 647
 648         if (range->min_ip)
 649                 range->flags |= IP_NAT_RANGE_MAP_IPS;
 650
 651         if (!tb[CTA_NAT_PROTO])
 652                 return 0;
 653
 654         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 655         if (err < 0)
 656                 return err;
 657
 658         return 0;
 659 }
 660
 661 static int
 662 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 663                           enum nf_nat_manip_type manip,
 664                           const struct nlattr *attr)
 665 {
 666         struct nf_nat_range range;
 667
 668         if (nfnetlink_parse_nat(attr, ct, &range) < 0)
 669                 return -EINVAL;
 670         if (nf_nat_initialized(ct, manip))
 671                 return -EEXIST;
 672
 673         return nf_nat_setup_info(ct, &range, manip);
 674 }
 675 #else
 676 static int
 677 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 678                           enum nf_nat_manip_type manip,
 679                           const struct nlattr *attr)
 680 {
 681         return -EOPNOTSUPP;
 682 }
 683 #endif
 684
 685 static int __net_init nf_nat_net_init(struct net *net)
 686 {
 687         /* Leave them the same for the moment. */
 688         net->ipv4.nat_htable_size = net->ct.htable_size;
 689         net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&net->ipv4.nat_htable_size, 0);
 690         if (!net->ipv4.nat_bysource)
 691                 return -ENOMEM;
 692         return 0;
 693 }
 694
 695 /* Clear NAT section of all conntracks, in case we're loaded again. */
 696 static int clean_nat(struct nf_conn *i, void *data)
 697 {
 698         struct nf_conn_nat *nat = nfct_nat(i);
 699
 700         if (!nat)
 701                 return 0;
 702         memset(nat, 0, sizeof(*nat));
 703         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
 704         return 0;
 705 }
 706
 707 static void __net_exit nf_nat_net_exit(struct net *net)
 708 {
 709         nf_ct_iterate_cleanup(net, &clean_nat, NULL);
 710         synchronize_rcu();
 711         nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_htable_size);
 712 }
 713
 714 static struct pernet_operations nf_nat_net_ops = {
 715         .init = nf_nat_net_init,
 716         .exit = nf_nat_net_exit,
 717 };
 718
 719 static int __init nf_nat_init(void)
 720 {
 721         size_t i;
 722         int ret;
 723
 724         need_ipv4_conntrack();
 725
 726         ret = nf_ct_extend_register(&nat_extend);
 727         if (ret < 0) {
 728                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 729                 return ret;
 730         }
 731
 732         ret = register_pernet_subsys(&nf_nat_net_ops);
 733         if (ret < 0)
 734                 goto cleanup_extend;
 735
 736         /* Sew in builtin protocols. */
 737         spin_lock_bh(&nf_nat_lock);
 738         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 739                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
 740         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 741         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 742         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 743         spin_unlock_bh(&nf_nat_lock);
 744
 745         /* Initialize fake conntrack so that NAT will skip it */
 746         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 747
 748         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 749
 750         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 751         rcu_assign_pointer(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 752         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 753         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook,
 754                            nfnetlink_parse_nat_setup);
 755         BUG_ON(nf_ct_nat_offset != NULL);
 756         rcu_assign_pointer(nf_ct_nat_offset, nf_nat_get_offset);
 757         return 0;
 758
 759  cleanup_extend:
 760         nf_ct_extend_unregister(&nat_extend);
 761         return ret;
 762 }
 763
 764 static void __exit nf_nat_cleanup(void)
 765 {
 766         unregister_pernet_subsys(&nf_nat_net_ops);
 767         nf_ct_l3proto_put(l3proto);
 768         nf_ct_extend_unregister(&nat_extend);
 769         rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL);
 770         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, NULL);
 771         rcu_assign_pointer(nf_ct_nat_offset, NULL);
 772         synchronize_net();
 773 }
 774
 775 MODULE_LICENSE("GPL");
 776 MODULE_ALIAS("nf-nat-ipv4");
 777
 778 module_init(nf_nat_init);
 779 module_exit(nf_nat_cleanup);