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                 if (find_appropriate_src(net, zone, orig_tuple, tuple, range)) {
 225                         pr_debug("get_unique_tuple: Found current src map\n");
 226                         if (!nf_nat_used_tuple(tuple, ct))
 227                                 return;
 228                 }
 229         }
 230
 231         /* 2) Select the least-used IP/proto combination in the given
 232            range. */
 233         *tuple = *orig_tuple;
 234         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 235
 236         /* 3) The per-protocol part of the manip is made to map into
 237            the range to make a unique tuple. */
 238
 239         rcu_read_lock();
 240         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
 241
 242         /* Only bother mapping if it's not already in range and unique */
 243         if (!(range->flags & IP_NAT_RANGE_PROTO_RANDOM)) {
 244                 if (range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) {
 245                         if (proto->in_range(tuple, maniptype, &range->min,
 246                                             &range->max) &&
 247                             (range->min.all == range->max.all ||
 248                              !nf_nat_used_tuple(tuple, ct)))
 249                                 goto out;
 250                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 251                         goto out;
 252                 }
 253         }
 254
 255         /* Last change: get protocol to try to obtain unique tuple. */
 256         proto->unique_tuple(tuple, range, maniptype, ct);
 257 out:
 258         rcu_read_unlock();
 259 }
 260
 261 unsigned int
 262 nf_nat_setup_info(struct nf_conn *ct,
 263                   const struct nf_nat_range *range,
 264                   enum nf_nat_manip_type maniptype)
 265 {
 266         struct net *net = nf_ct_net(ct);
 267         struct nf_conntrack_tuple curr_tuple, new_tuple;
 268         struct nf_conn_nat *nat;
 269         int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
 270
 271         /* nat helper or nfctnetlink also setup binding */
 272         nat = nfct_nat(ct);
 273         if (!nat) {
 274                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 275                 if (nat == NULL) {
 276                         pr_debug("failed to add NAT extension\n");
 277                         return NF_ACCEPT;
 278                 }
 279         }
 280
 281         NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC ||
 282                      maniptype == IP_NAT_MANIP_DST);
 283         BUG_ON(nf_nat_initialized(ct, maniptype));
 284
 285         /* What we've got will look like inverse of reply. Normally
 286            this is what is in the conntrack, except for prior
 287            manipulations (future optimization: if num_manips == 0,
 288            orig_tp =
 289            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
 290         nf_ct_invert_tuplepr(&curr_tuple,
 291                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 292
 293         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 294
 295         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 296                 struct nf_conntrack_tuple reply;
 297
 298                 /* Alter conntrack table so will recognize replies. */
 299                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 300                 nf_conntrack_alter_reply(ct, &reply);
 301
 302                 /* Non-atomic: we own this at the moment. */
 303                 if (maniptype == IP_NAT_MANIP_SRC)
 304                         ct->status |= IPS_SRC_NAT;
 305                 else
 306                         ct->status |= IPS_DST_NAT;
 307         }
 308
 309         /* Place in source hash if this is the first time. */
 310         if (have_to_hash) {
 311                 unsigned int srchash;
 312
 313                 srchash = hash_by_src(net, nf_ct_zone(ct),
 314                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 315                 spin_lock_bh(&nf_nat_lock);
 316                 /* nf_conntrack_alter_reply might re-allocate exntension aera */
 317                 nat = nfct_nat(ct);
 318                 nat->ct = ct;
 319                 hlist_add_head_rcu(&nat->bysource,
 320                                    &net->ipv4.nat_bysource[srchash]);
 321                 spin_unlock_bh(&nf_nat_lock);
 322         }
 323
 324         /* It's done. */
 325         if (maniptype == IP_NAT_MANIP_DST)
 326                 set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
 327         else
 328                 set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 329
 330         return NF_ACCEPT;
 331 }
 332 EXPORT_SYMBOL(nf_nat_setup_info);
 333
 334 /* Returns true if succeeded. */
 335 static bool
 336 manip_pkt(u_int16_t proto,
 337           struct sk_buff *skb,
 338           unsigned int iphdroff,
 339           const struct nf_conntrack_tuple *target,
 340           enum nf_nat_manip_type maniptype)
 341 {
 342         struct iphdr *iph;
 343         const struct nf_nat_protocol *p;
 344
 345         if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
 346                 return false;
 347
 348         iph = (void *)skb->data + iphdroff;
 349
 350         /* Manipulate protcol part. */
 351
 352         /* rcu_read_lock()ed by nf_hook_slow */
 353         p = __nf_nat_proto_find(proto);
 354         if (!p->manip_pkt(skb, iphdroff, target, maniptype))
 355                 return false;
 356
 357         iph = (void *)skb->data + iphdroff;
 358
 359         if (maniptype == IP_NAT_MANIP_SRC) {
 360                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 361                 iph->saddr = target->src.u3.ip;
 362         } else {
 363                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 364                 iph->daddr = target->dst.u3.ip;
 365         }
 366         return true;
 367 }
 368
 369 /* Do packet manipulations according to nf_nat_setup_info. */
 370 unsigned int nf_nat_packet(struct nf_conn *ct,
 371                            enum ip_conntrack_info ctinfo,
 372                            unsigned int hooknum,
 373                            struct sk_buff *skb)
 374 {
 375         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 376         unsigned long statusbit;
 377         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 378
 379         if (mtype == IP_NAT_MANIP_SRC)
 380                 statusbit = IPS_SRC_NAT;
 381         else
 382                 statusbit = IPS_DST_NAT;
 383
 384         /* Invert if this is reply dir. */
 385         if (dir == IP_CT_DIR_REPLY)
 386                 statusbit ^= IPS_NAT_MASK;
 387
 388         /* Non-atomic: these bits don't change. */
 389         if (ct->status & statusbit) {
 390                 struct nf_conntrack_tuple target;
 391
 392                 /* We are aiming to look like inverse of other direction. */
 393                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 394
 395                 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
 396                         return NF_DROP;
 397         }
 398         return NF_ACCEPT;
 399 }
 400 EXPORT_SYMBOL_GPL(nf_nat_packet);
 401
 402 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
 403 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 404                                   enum ip_conntrack_info ctinfo,
 405                                   unsigned int hooknum,
 406                                   struct sk_buff *skb)
 407 {
 408         struct {
 409                 struct icmphdr icmp;
 410                 struct iphdr ip;
 411         } *inside;
 412         const struct nf_conntrack_l4proto *l4proto;
 413         struct nf_conntrack_tuple inner, target;
 414         int hdrlen = ip_hdrlen(skb);
 415         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 416         unsigned long statusbit;
 417         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 418
 419         if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
 420                 return 0;
 421
 422         inside = (void *)skb->data + hdrlen;
 423
 424         /* We're actually going to mangle it beyond trivial checksum
 425            adjustment, so make sure the current checksum is correct. */
 426         if (nf_ip_checksum(skb, hooknum, hdrlen, 0))
 427                 return 0;
 428
 429         /* Must be RELATED */
 430         NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
 431                      skb->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
 432
 433         /* Redirects on non-null nats must be dropped, else they'll
 434            start talking to each other without our translation, and be
 435            confused... --RR */
 436         if (inside->icmp.type == ICMP_REDIRECT) {
 437                 /* If NAT isn't finished, assume it and drop. */
 438                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 439                         return 0;
 440
 441                 if (ct->status & IPS_NAT_MASK)
 442                         return 0;
 443         }
 444
 445         if (manip == IP_NAT_MANIP_SRC)
 446                 statusbit = IPS_SRC_NAT;
 447         else
 448                 statusbit = IPS_DST_NAT;
 449
 450         /* Invert if this is reply dir. */
 451         if (dir == IP_CT_DIR_REPLY)
 452                 statusbit ^= IPS_NAT_MASK;
 453
 454         if (!(ct->status & statusbit))
 455                 return 1;
 456
 457         pr_debug("icmp_reply_translation: translating error %p manip %u "
 458                  "dir %s\n", skb, manip,
 459                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 460
 461         /* rcu_read_lock()ed by nf_hook_slow */
 462         l4proto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
 463
 464         if (!nf_ct_get_tuple(skb, hdrlen + sizeof(struct icmphdr),
 465                              (hdrlen +
 466                               sizeof(struct icmphdr) + inside->ip.ihl * 4),
 467                              (u_int16_t)AF_INET, inside->ip.protocol,
 468                              &inner, l3proto, l4proto))
 469                 return 0;
 470
 471         /* Change inner back to look like incoming packet.  We do the
 472            opposite manip on this hook to normal, because it might not
 473            pass all hooks (locally-generated ICMP).  Consider incoming
 474            packet: PREROUTING (DST manip), routing produces ICMP, goes
 475            through POSTROUTING (which must correct the DST manip). */
 476         if (!manip_pkt(inside->ip.protocol, skb, hdrlen + sizeof(inside->icmp),
 477                        &ct->tuplehash[!dir].tuple, !manip))
 478                 return 0;
 479
 480         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 481                 /* Reloading "inside" here since manip_pkt inner. */
 482                 inside = (void *)skb->data + hdrlen;
 483                 inside->icmp.checksum = 0;
 484                 inside->icmp.checksum =
 485                         csum_fold(skb_checksum(skb, hdrlen,
 486                                                skb->len - hdrlen, 0));
 487         }
 488
 489         /* Change outer to look the reply to an incoming packet
 490          * (proto 0 means don't invert per-proto part). */
 491         nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 492         if (!manip_pkt(0, skb, 0, &target, manip))
 493                 return 0;
 494
 495         return 1;
 496 }
 497 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 498
 499 /* Protocol registration. */
 500 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 501 {
 502         int ret = 0;
 503
 504         spin_lock_bh(&nf_nat_lock);
 505         if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
 506                 ret = -EBUSY;
 507                 goto out;
 508         }
 509         rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
 510  out:
 511         spin_unlock_bh(&nf_nat_lock);
 512         return ret;
 513 }
 514 EXPORT_SYMBOL(nf_nat_protocol_register);
 515
 516 /* Noone stores the protocol anywhere; simply delete it. */
 517 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 518 {
 519         spin_lock_bh(&nf_nat_lock);
 520         rcu_assign_pointer(nf_nat_protos[proto->protonum],
 521                            &nf_nat_unknown_protocol);
 522         spin_unlock_bh(&nf_nat_lock);
 523         synchronize_rcu();
 524 }
 525 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 526
 527 /* Noone using conntrack by the time this called. */
 528 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 529 {
 530         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 531
 532         if (nat == NULL || nat->ct == NULL)
 533                 return;
 534
 535         NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
 536
 537         spin_lock_bh(&nf_nat_lock);
 538         hlist_del_rcu(&nat->bysource);
 539         spin_unlock_bh(&nf_nat_lock);
 540 }
 541
 542 static void nf_nat_move_storage(void *new, void *old)
 543 {
 544         struct nf_conn_nat *new_nat = new;
 545         struct nf_conn_nat *old_nat = old;
 546         struct nf_conn *ct = old_nat->ct;
 547
 548         if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
 549                 return;
 550
 551         spin_lock_bh(&nf_nat_lock);
 552         new_nat->ct = ct;
 553         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 554         spin_unlock_bh(&nf_nat_lock);
 555 }
 556
 557 static struct nf_ct_ext_type nat_extend __read_mostly = {
 558         .len            = sizeof(struct nf_conn_nat),
 559         .align          = __alignof__(struct nf_conn_nat),
 560         .destroy        = nf_nat_cleanup_conntrack,
 561         .move           = nf_nat_move_storage,
 562         .id             = NF_CT_EXT_NAT,
 563         .flags          = NF_CT_EXT_F_PREALLOC,
 564 };
 565
 566 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 567
 568 #include <linux/netfilter/nfnetlink.h>
 569 #include <linux/netfilter/nfnetlink_conntrack.h>
 570
 571 static const struct nf_nat_protocol *
 572 nf_nat_proto_find_get(u_int8_t protonum)
 573 {
 574         const struct nf_nat_protocol *p;
 575
 576         rcu_read_lock();
 577         p = __nf_nat_proto_find(protonum);
 578         if (!try_module_get(p->me))
 579                 p = &nf_nat_unknown_protocol;
 580         rcu_read_unlock();
 581
 582         return p;
 583 }
 584
 585 static void
 586 nf_nat_proto_put(const struct nf_nat_protocol *p)
 587 {
 588         module_put(p->me);
 589 }
 590
 591 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 592         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 593         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 594 };
 595
 596 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 597                                      const struct nf_conn *ct,
 598                                      struct nf_nat_range *range)
 599 {
 600         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 601         const struct nf_nat_protocol *npt;
 602         int err;
 603
 604         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 605         if (err < 0)
 606                 return err;
 607
 608         npt = nf_nat_proto_find_get(nf_ct_protonum(ct));
 609         if (npt->nlattr_to_range)
 610                 err = npt->nlattr_to_range(tb, range);
 611         nf_nat_proto_put(npt);
 612         return err;
 613 }
 614
 615 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 616         [CTA_NAT_MINIP]         = { .type = NLA_U32 },
 617         [CTA_NAT_MAXIP]         = { .type = NLA_U32 },
 618 };
 619
 620 static int
 621 nfnetlink_parse_nat(const struct nlattr *nat,
 622                     const struct nf_conn *ct, struct nf_nat_range *range)
 623 {
 624         struct nlattr *tb[CTA_NAT_MAX+1];
 625         int err;
 626
 627         memset(range, 0, sizeof(*range));
 628
 629         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 630         if (err < 0)
 631                 return err;
 632
 633         if (tb[CTA_NAT_MINIP])
 634                 range->min_ip = nla_get_be32(tb[CTA_NAT_MINIP]);
 635
 636         if (!tb[CTA_NAT_MAXIP])
 637                 range->max_ip = range->min_ip;
 638         else
 639                 range->max_ip = nla_get_be32(tb[CTA_NAT_MAXIP]);
 640
 641         if (range->min_ip)
 642                 range->flags |= IP_NAT_RANGE_MAP_IPS;
 643
 644         if (!tb[CTA_NAT_PROTO])
 645                 return 0;
 646
 647         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 648         if (err < 0)
 649                 return err;
 650
 651         return 0;
 652 }
 653
 654 static int
 655 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 656                           enum nf_nat_manip_type manip,
 657                           const struct nlattr *attr)
 658 {
 659         struct nf_nat_range range;
 660
 661         if (nfnetlink_parse_nat(attr, ct, &range) < 0)
 662                 return -EINVAL;
 663         if (nf_nat_initialized(ct, manip))
 664                 return -EEXIST;
 665
 666         return nf_nat_setup_info(ct, &range, manip);
 667 }
 668 #else
 669 static int
 670 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 671                           enum nf_nat_manip_type manip,
 672                           const struct nlattr *attr)
 673 {
 674         return -EOPNOTSUPP;
 675 }
 676 #endif
 677
 678 static int __net_init nf_nat_net_init(struct net *net)
 679 {
 680         /* Leave them the same for the moment. */
 681         net->ipv4.nat_htable_size = net->ct.htable_size;
 682         net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&net->ipv4.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                              net->ipv4.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         ret = register_pernet_subsys(&nf_nat_net_ops);
 728         if (ret < 0)
 729                 goto cleanup_extend;
 730
 731         /* Sew in builtin protocols. */
 732         spin_lock_bh(&nf_nat_lock);
 733         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 734                 rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
 735         rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 736         rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 737         rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 738         spin_unlock_bh(&nf_nat_lock);
 739
 740         /* Initialize fake conntrack so that NAT will skip it */
 741         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 742
 743         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 744
 745         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 746         rcu_assign_pointer(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 747         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 748         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook,
 749                            nfnetlink_parse_nat_setup);
 750         BUG_ON(nf_ct_nat_offset != NULL);
 751         rcu_assign_pointer(nf_ct_nat_offset, nf_nat_get_offset);
 752         return 0;
 753
 754  cleanup_extend:
 755         nf_ct_extend_unregister(&nat_extend);
 756         return ret;
 757 }
 758
 759 static void __exit nf_nat_cleanup(void)
 760 {
 761         unregister_pernet_subsys(&nf_nat_net_ops);
 762         nf_ct_l3proto_put(l3proto);
 763         nf_ct_extend_unregister(&nat_extend);
 764         rcu_assign_pointer(nf_nat_seq_adjust_hook, NULL);
 765         rcu_assign_pointer(nfnetlink_parse_nat_setup_hook, NULL);
 766         rcu_assign_pointer(nf_ct_nat_offset, 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);