Pileus Git - ~andy/linux/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 <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_zones.h>
  34
  35 static DEFINE_SPINLOCK(nf_nat_lock);
  36
  37 static struct nf_conntrack_l3proto *l3proto __read_mostly;
  38
  39 #define MAX_IP_NAT_PROTO 256
  40 static const struct nf_nat_protocol __rcu *nf_nat_protos[MAX_IP_NAT_PROTO]
  41                                                 __read_mostly;
  42
  43 static inline const struct nf_nat_protocol *
  44 __nf_nat_proto_find(u_int8_t protonum)
  45 {
  46         return rcu_dereference(nf_nat_protos[protonum]);
  47 }
  48
  49 /* We keep an extra hash for each conntrack, for fast searching. */
  50 static inline unsigned int
  51 hash_by_src(const struct net *net, u16 zone,
  52             const struct nf_conntrack_tuple *tuple)
  53 {
  54         unsigned int hash;
  55
  56         /* Original src, to ensure we map it consistently if poss. */
  57         hash = jhash_3words((__force u32)tuple->src.u3.ip,
  58                             (__force u32)tuple->src.u.all ^ zone,
  59                             tuple->dst.protonum, nf_conntrack_hash_rnd);
  60         return ((u64)hash * net->ipv4.nat_htable_size) >> 32;
  61 }
  62
  63 /* Is this tuple already taken? (not by us) */
  64 int
  65 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
  66                   const struct nf_conn *ignored_conntrack)
  67 {
  68         /* Conntrack tracking doesn't keep track of outgoing tuples; only
  69            incoming ones.  NAT means they don't have a fixed mapping,
  70            so we invert the tuple and look for the incoming reply.
  71
  72            We could keep a separate hash if this proves too slow. */
  73         struct nf_conntrack_tuple reply;
  74
  75         nf_ct_invert_tuplepr(&reply, tuple);
  76         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
  77 }
  78 EXPORT_SYMBOL(nf_nat_used_tuple);
  79
  80 /* If we source map this tuple so reply looks like reply_tuple, will
  81  * that meet the constraints of range. */
  82 static int
  83 in_range(const struct nf_conntrack_tuple *tuple,
  84          const struct nf_nat_ipv4_range *range)
  85 {
  86         const struct nf_nat_protocol *proto;
  87         int ret = 0;
  88
  89         /* If we are supposed to map IPs, then we must be in the
  90            range specified, otherwise let this drag us onto a new src IP. */
  91         if (range->flags & NF_NAT_RANGE_MAP_IPS) {
  92                 if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
  93                     ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
  94                         return 0;
  95         }
  96
  97         rcu_read_lock();
  98         proto = __nf_nat_proto_find(tuple->dst.protonum);
  99         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 100             proto->in_range(tuple, NF_NAT_MANIP_SRC,
 101                             &range->min, &range->max))
 102                 ret = 1;
 103         rcu_read_unlock();
 104
 105         return ret;
 106 }
 107
 108 static inline int
 109 same_src(const struct nf_conn *ct,
 110          const struct nf_conntrack_tuple *tuple)
 111 {
 112         const struct nf_conntrack_tuple *t;
 113
 114         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 115         return (t->dst.protonum == tuple->dst.protonum &&
 116                 t->src.u3.ip == tuple->src.u3.ip &&
 117                 t->src.u.all == tuple->src.u.all);
 118 }
 119
 120 /* Only called for SRC manip */
 121 static int
 122 find_appropriate_src(struct net *net, u16 zone,
 123                      const struct nf_conntrack_tuple *tuple,
 124                      struct nf_conntrack_tuple *result,
 125                      const struct nf_nat_ipv4_range *range)
 126 {
 127         unsigned int h = hash_by_src(net, zone, tuple);
 128         const struct nf_conn_nat *nat;
 129         const struct nf_conn *ct;
 130         const struct hlist_node *n;
 131
 132         rcu_read_lock();
 133         hlist_for_each_entry_rcu(nat, n, &net->ipv4.nat_bysource[h], bysource) {
 134                 ct = nat->ct;
 135                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
 136                         /* Copy source part from reply tuple. */
 137                         nf_ct_invert_tuplepr(result,
 138                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 139                         result->dst = tuple->dst;
 140
 141                         if (in_range(result, range)) {
 142                                 rcu_read_unlock();
 143                                 return 1;
 144                         }
 145                 }
 146         }
 147         rcu_read_unlock();
 148         return 0;
 149 }
 150
 151 /* For [FUTURE] fragmentation handling, we want the least-used
 152    src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 153    if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 154    1-65535, we don't do pro-rata allocation based on ports; we choose
 155    the ip with the lowest src-ip/dst-ip/proto usage.
 156 */
 157 static void
 158 find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple,
 159                     const struct nf_nat_ipv4_range *range,
 160                     const struct nf_conn *ct,
 161                     enum nf_nat_manip_type maniptype)
 162 {
 163         __be32 *var_ipp;
 164         /* Host order */
 165         u_int32_t minip, maxip, j;
 166
 167         /* No IP mapping?  Do nothing. */
 168         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 169                 return;
 170
 171         if (maniptype == NF_NAT_MANIP_SRC)
 172                 var_ipp = &tuple->src.u3.ip;
 173         else
 174                 var_ipp = &tuple->dst.u3.ip;
 175
 176         /* Fast path: only one choice. */
 177         if (range->min_ip == range->max_ip) {
 178                 *var_ipp = range->min_ip;
 179                 return;
 180         }
 181
 182         /* Hashing source and destination IPs gives a fairly even
 183          * spread in practice (if there are a small number of IPs
 184          * involved, there usually aren't that many connections
 185          * anyway).  The consistency means that servers see the same
 186          * client coming from the same IP (some Internet Banking sites
 187          * like this), even across reboots. */
 188         minip = ntohl(range->min_ip);
 189         maxip = ntohl(range->max_ip);
 190         j = jhash_2words((__force u32)tuple->src.u3.ip,
 191                          range->flags & NF_NAT_RANGE_PERSISTENT ?
 192                                 0 : (__force u32)tuple->dst.u3.ip ^ zone, 0);
 193         j = ((u64)j * (maxip - minip + 1)) >> 32;
 194         *var_ipp = htonl(minip + j);
 195 }
 196
 197 /* Manipulate the tuple into the range given.  For NF_INET_POST_ROUTING,
 198  * we change the source to map into the range.  For NF_INET_PRE_ROUTING
 199  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 200  * range.  It might not be possible to get a unique tuple, but we try.
 201  * At worst (or if we race), we will end up with a final duplicate in
 202  * __ip_conntrack_confirm and drop the packet. */
 203 static void
 204 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 205                  const struct nf_conntrack_tuple *orig_tuple,
 206                  const struct nf_nat_ipv4_range *range,
 207                  struct nf_conn *ct,
 208                  enum nf_nat_manip_type maniptype)
 209 {
 210         struct net *net = nf_ct_net(ct);
 211         const struct nf_nat_protocol *proto;
 212         u16 zone = nf_ct_zone(ct);
 213
 214         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 215            and that same mapping gives a unique tuple within the given
 216            range, use that.
 217
 218            This is only required for source (ie. NAT/masq) mappings.
 219            So far, we don't do local source mappings, so multiple
 220            manips not an issue.  */
 221         if (maniptype == NF_NAT_MANIP_SRC &&
 222             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 223                 /* try the original tuple first */
 224                 if (in_range(orig_tuple, range)) {
 225                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 226                                 *tuple = *orig_tuple;
 227                                 return;
 228                         }
 229                 } else if (find_appropriate_src(net, zone, orig_tuple, tuple,
 230                            range)) {
 231                         pr_debug("get_unique_tuple: Found current src map\n");
 232                         if (!nf_nat_used_tuple(tuple, ct))
 233                                 return;
 234                 }
 235         }
 236
 237         /* 2) Select the least-used IP/proto combination in the given
 238            range. */
 239         *tuple = *orig_tuple;
 240         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 241
 242         /* 3) The per-protocol part of the manip is made to map into
 243            the range to make a unique tuple. */
 244
 245         rcu_read_lock();
 246         proto = __nf_nat_proto_find(orig_tuple->dst.protonum);
 247
 248         /* Only bother mapping if it's not already in range and unique */
 249         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 250                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 251                         if (proto->in_range(tuple, maniptype, &range->min,
 252                                             &range->max) &&
 253                             (range->min.all == range->max.all ||
 254                              !nf_nat_used_tuple(tuple, ct)))
 255                                 goto out;
 256                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 257                         goto out;
 258                 }
 259         }
 260
 261         /* Last change: get protocol to try to obtain unique tuple. */
 262         proto->unique_tuple(tuple, range, maniptype, ct);
 263 out:
 264         rcu_read_unlock();
 265 }
 266
 267 unsigned int
 268 nf_nat_setup_info(struct nf_conn *ct,
 269                   const struct nf_nat_ipv4_range *range,
 270                   enum nf_nat_manip_type maniptype)
 271 {
 272         struct net *net = nf_ct_net(ct);
 273         struct nf_conntrack_tuple curr_tuple, new_tuple;
 274         struct nf_conn_nat *nat;
 275
 276         /* nat helper or nfctnetlink also setup binding */
 277         nat = nfct_nat(ct);
 278         if (!nat) {
 279                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 280                 if (nat == NULL) {
 281                         pr_debug("failed to add NAT extension\n");
 282                         return NF_ACCEPT;
 283                 }
 284         }
 285
 286         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 287                      maniptype == NF_NAT_MANIP_DST);
 288         BUG_ON(nf_nat_initialized(ct, maniptype));
 289
 290         /* What we've got will look like inverse of reply. Normally
 291            this is what is in the conntrack, except for prior
 292            manipulations (future optimization: if num_manips == 0,
 293            orig_tp =
 294            conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
 295         nf_ct_invert_tuplepr(&curr_tuple,
 296                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 297
 298         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 299
 300         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 301                 struct nf_conntrack_tuple reply;
 302
 303                 /* Alter conntrack table so will recognize replies. */
 304                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 305                 nf_conntrack_alter_reply(ct, &reply);
 306
 307                 /* Non-atomic: we own this at the moment. */
 308                 if (maniptype == NF_NAT_MANIP_SRC)
 309                         ct->status |= IPS_SRC_NAT;
 310                 else
 311                         ct->status |= IPS_DST_NAT;
 312         }
 313
 314         if (maniptype == NF_NAT_MANIP_SRC) {
 315                 unsigned int srchash;
 316
 317                 srchash = hash_by_src(net, nf_ct_zone(ct),
 318                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 319                 spin_lock_bh(&nf_nat_lock);
 320                 /* nf_conntrack_alter_reply might re-allocate extension area */
 321                 nat = nfct_nat(ct);
 322                 nat->ct = ct;
 323                 hlist_add_head_rcu(&nat->bysource,
 324                                    &net->ipv4.nat_bysource[srchash]);
 325                 spin_unlock_bh(&nf_nat_lock);
 326         }
 327
 328         /* It's done. */
 329         if (maniptype == NF_NAT_MANIP_DST)
 330                 ct->status |= IPS_DST_NAT_DONE;
 331         else
 332                 ct->status |= IPS_SRC_NAT_DONE;
 333
 334         return NF_ACCEPT;
 335 }
 336 EXPORT_SYMBOL(nf_nat_setup_info);
 337
 338 /* Returns true if succeeded. */
 339 static bool
 340 manip_pkt(u_int16_t proto,
 341           struct sk_buff *skb,
 342           unsigned int iphdroff,
 343           const struct nf_conntrack_tuple *target,
 344           enum nf_nat_manip_type maniptype)
 345 {
 346         struct iphdr *iph;
 347         const struct nf_nat_protocol *p;
 348
 349         if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
 350                 return false;
 351
 352         iph = (void *)skb->data + iphdroff;
 353
 354         /* Manipulate protcol part. */
 355
 356         /* rcu_read_lock()ed by nf_hook_slow */
 357         p = __nf_nat_proto_find(proto);
 358         if (!p->manip_pkt(skb, iphdroff, target, maniptype))
 359                 return false;
 360
 361         iph = (void *)skb->data + iphdroff;
 362
 363         if (maniptype == NF_NAT_MANIP_SRC) {
 364                 csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
 365                 iph->saddr = target->src.u3.ip;
 366         } else {
 367                 csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
 368                 iph->daddr = target->dst.u3.ip;
 369         }
 370         return true;
 371 }
 372
 373 /* Do packet manipulations according to nf_nat_setup_info. */
 374 unsigned int nf_nat_packet(struct nf_conn *ct,
 375                            enum ip_conntrack_info ctinfo,
 376                            unsigned int hooknum,
 377                            struct sk_buff *skb)
 378 {
 379         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 380         unsigned long statusbit;
 381         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 382
 383         if (mtype == NF_NAT_MANIP_SRC)
 384                 statusbit = IPS_SRC_NAT;
 385         else
 386                 statusbit = IPS_DST_NAT;
 387
 388         /* Invert if this is reply dir. */
 389         if (dir == IP_CT_DIR_REPLY)
 390                 statusbit ^= IPS_NAT_MASK;
 391
 392         /* Non-atomic: these bits don't change. */
 393         if (ct->status & statusbit) {
 394                 struct nf_conntrack_tuple target;
 395
 396                 /* We are aiming to look like inverse of other direction. */
 397                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 398
 399                 if (!manip_pkt(target.dst.protonum, skb, 0, &target, mtype))
 400                         return NF_DROP;
 401         }
 402         return NF_ACCEPT;
 403 }
 404 EXPORT_SYMBOL_GPL(nf_nat_packet);
 405
 406 /* Dir is direction ICMP is coming from (opposite to packet it contains) */
 407 int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 408                                   enum ip_conntrack_info ctinfo,
 409                                   unsigned int hooknum,
 410                                   struct sk_buff *skb)
 411 {
 412         struct {
 413                 struct icmphdr icmp;
 414                 struct iphdr ip;
 415         } *inside;
 416         struct nf_conntrack_tuple target;
 417         int hdrlen = ip_hdrlen(skb);
 418         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 419         unsigned long statusbit;
 420         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 421
 422         if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
 423                 return 0;
 424
 425         inside = (void *)skb->data + hdrlen;
 426
 427         /* We're actually going to mangle it beyond trivial checksum
 428            adjustment, so make sure the current checksum is correct. */
 429         if (nf_ip_checksum(skb, hooknum, hdrlen, 0))
 430                 return 0;
 431
 432         /* Must be RELATED */
 433         NF_CT_ASSERT(skb->nfctinfo == IP_CT_RELATED ||
 434                      skb->nfctinfo == IP_CT_RELATED_REPLY);
 435
 436         /* Redirects on non-null nats must be dropped, else they'll
 437            start talking to each other without our translation, and be
 438            confused... --RR */
 439         if (inside->icmp.type == ICMP_REDIRECT) {
 440                 /* If NAT isn't finished, assume it and drop. */
 441                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 442                         return 0;
 443
 444                 if (ct->status & IPS_NAT_MASK)
 445                         return 0;
 446         }
 447
 448         if (manip == NF_NAT_MANIP_SRC)
 449                 statusbit = IPS_SRC_NAT;
 450         else
 451                 statusbit = IPS_DST_NAT;
 452
 453         /* Invert if this is reply dir. */
 454         if (dir == IP_CT_DIR_REPLY)
 455                 statusbit ^= IPS_NAT_MASK;
 456
 457         if (!(ct->status & statusbit))
 458                 return 1;
 459
 460         pr_debug("icmp_reply_translation: translating error %p manip %u "
 461                  "dir %s\n", skb, manip,
 462                  dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
 463
 464         /* Change inner back to look like incoming packet.  We do the
 465            opposite manip on this hook to normal, because it might not
 466            pass all hooks (locally-generated ICMP).  Consider incoming
 467            packet: PREROUTING (DST manip), routing produces ICMP, goes
 468            through POSTROUTING (which must correct the DST manip). */
 469         if (!manip_pkt(inside->ip.protocol, skb, hdrlen + sizeof(inside->icmp),
 470                        &ct->tuplehash[!dir].tuple, !manip))
 471                 return 0;
 472
 473         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 474                 /* Reloading "inside" here since manip_pkt inner. */
 475                 inside = (void *)skb->data + hdrlen;
 476                 inside->icmp.checksum = 0;
 477                 inside->icmp.checksum =
 478                         csum_fold(skb_checksum(skb, hdrlen,
 479                                                skb->len - hdrlen, 0));
 480         }
 481
 482         /* Change outer to look the reply to an incoming packet
 483          * (proto 0 means don't invert per-proto part). */
 484         nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 485         if (!manip_pkt(0, skb, 0, &target, manip))
 486                 return 0;
 487
 488         return 1;
 489 }
 490 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 491
 492 /* Protocol registration. */
 493 int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 494 {
 495         int ret = 0;
 496
 497         spin_lock_bh(&nf_nat_lock);
 498         if (rcu_dereference_protected(
 499                         nf_nat_protos[proto->protonum],
 500                         lockdep_is_held(&nf_nat_lock)
 501                         ) != &nf_nat_unknown_protocol) {
 502                 ret = -EBUSY;
 503                 goto out;
 504         }
 505         RCU_INIT_POINTER(nf_nat_protos[proto->protonum], proto);
 506  out:
 507         spin_unlock_bh(&nf_nat_lock);
 508         return ret;
 509 }
 510 EXPORT_SYMBOL(nf_nat_protocol_register);
 511
 512 /* No one stores the protocol anywhere; simply delete it. */
 513 void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 514 {
 515         spin_lock_bh(&nf_nat_lock);
 516         RCU_INIT_POINTER(nf_nat_protos[proto->protonum],
 517                            &nf_nat_unknown_protocol);
 518         spin_unlock_bh(&nf_nat_lock);
 519         synchronize_rcu();
 520 }
 521 EXPORT_SYMBOL(nf_nat_protocol_unregister);
 522
 523 /* No one using conntrack by the time this called. */
 524 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 525 {
 526         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 527
 528         if (nat == NULL || nat->ct == NULL)
 529                 return;
 530
 531         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 532
 533         spin_lock_bh(&nf_nat_lock);
 534         hlist_del_rcu(&nat->bysource);
 535         spin_unlock_bh(&nf_nat_lock);
 536 }
 537
 538 static void nf_nat_move_storage(void *new, void *old)
 539 {
 540         struct nf_conn_nat *new_nat = new;
 541         struct nf_conn_nat *old_nat = old;
 542         struct nf_conn *ct = old_nat->ct;
 543
 544         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 545                 return;
 546
 547         spin_lock_bh(&nf_nat_lock);
 548         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 549         spin_unlock_bh(&nf_nat_lock);
 550 }
 551
 552 static struct nf_ct_ext_type nat_extend __read_mostly = {
 553         .len            = sizeof(struct nf_conn_nat),
 554         .align          = __alignof__(struct nf_conn_nat),
 555         .destroy        = nf_nat_cleanup_conntrack,
 556         .move           = nf_nat_move_storage,
 557         .id             = NF_CT_EXT_NAT,
 558         .flags          = NF_CT_EXT_F_PREALLOC,
 559 };
 560
 561 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 562
 563 #include <linux/netfilter/nfnetlink.h>
 564 #include <linux/netfilter/nfnetlink_conntrack.h>
 565
 566 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 567         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 568         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 569 };
 570
 571 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 572                                      const struct nf_conn *ct,
 573                                      struct nf_nat_ipv4_range *range)
 574 {
 575         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 576         const struct nf_nat_protocol *npt;
 577         int err;
 578
 579         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 580         if (err < 0)
 581                 return err;
 582
 583         rcu_read_lock();
 584         npt = __nf_nat_proto_find(nf_ct_protonum(ct));
 585         if (npt->nlattr_to_range)
 586                 err = npt->nlattr_to_range(tb, range);
 587         rcu_read_unlock();
 588         return err;
 589 }
 590
 591 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 592         [CTA_NAT_MINIP]         = { .type = NLA_U32 },
 593         [CTA_NAT_MAXIP]         = { .type = NLA_U32 },
 594         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 595 };
 596
 597 static int
 598 nfnetlink_parse_nat(const struct nlattr *nat,
 599                     const struct nf_conn *ct, struct nf_nat_ipv4_range *range)
 600 {
 601         struct nlattr *tb[CTA_NAT_MAX+1];
 602         int err;
 603
 604         memset(range, 0, sizeof(*range));
 605
 606         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 607         if (err < 0)
 608                 return err;
 609
 610         if (tb[CTA_NAT_MINIP])
 611                 range->min_ip = nla_get_be32(tb[CTA_NAT_MINIP]);
 612
 613         if (!tb[CTA_NAT_MAXIP])
 614                 range->max_ip = range->min_ip;
 615         else
 616                 range->max_ip = nla_get_be32(tb[CTA_NAT_MAXIP]);
 617
 618         if (range->min_ip)
 619                 range->flags |= NF_NAT_RANGE_MAP_IPS;
 620
 621         if (!tb[CTA_NAT_PROTO])
 622                 return 0;
 623
 624         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 625         if (err < 0)
 626                 return err;
 627
 628         return 0;
 629 }
 630
 631 static int
 632 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 633                           enum nf_nat_manip_type manip,
 634                           const struct nlattr *attr)
 635 {
 636         struct nf_nat_ipv4_range range;
 637
 638         if (nfnetlink_parse_nat(attr, ct, &range) < 0)
 639                 return -EINVAL;
 640         if (nf_nat_initialized(ct, manip))
 641                 return -EEXIST;
 642
 643         return nf_nat_setup_info(ct, &range, manip);
 644 }
 645 #else
 646 static int
 647 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 648                           enum nf_nat_manip_type manip,
 649                           const struct nlattr *attr)
 650 {
 651         return -EOPNOTSUPP;
 652 }
 653 #endif
 654
 655 static int __net_init nf_nat_net_init(struct net *net)
 656 {
 657         /* Leave them the same for the moment. */
 658         net->ipv4.nat_htable_size = net->ct.htable_size;
 659         net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&net->ipv4.nat_htable_size, 0);
 660         if (!net->ipv4.nat_bysource)
 661                 return -ENOMEM;
 662         return 0;
 663 }
 664
 665 /* Clear NAT section of all conntracks, in case we're loaded again. */
 666 static int clean_nat(struct nf_conn *i, void *data)
 667 {
 668         struct nf_conn_nat *nat = nfct_nat(i);
 669
 670         if (!nat)
 671                 return 0;
 672         memset(nat, 0, sizeof(*nat));
 673         i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
 674         return 0;
 675 }
 676
 677 static void __net_exit nf_nat_net_exit(struct net *net)
 678 {
 679         nf_ct_iterate_cleanup(net, &clean_nat, NULL);
 680         synchronize_rcu();
 681         nf_ct_free_hashtable(net->ipv4.nat_bysource, net->ipv4.nat_htable_size);
 682 }
 683
 684 static struct pernet_operations nf_nat_net_ops = {
 685         .init = nf_nat_net_init,
 686         .exit = nf_nat_net_exit,
 687 };
 688
 689 static struct nf_ct_helper_expectfn follow_master_nat = {
 690         .name           = "nat-follow-master",
 691         .expectfn       = nf_nat_follow_master,
 692 };
 693
 694 static int __init nf_nat_init(void)
 695 {
 696         size_t i;
 697         int ret;
 698
 699         need_ipv4_conntrack();
 700
 701         ret = nf_ct_extend_register(&nat_extend);
 702         if (ret < 0) {
 703                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 704                 return ret;
 705         }
 706
 707         ret = register_pernet_subsys(&nf_nat_net_ops);
 708         if (ret < 0)
 709                 goto cleanup_extend;
 710
 711         /* Sew in builtin protocols. */
 712         spin_lock_bh(&nf_nat_lock);
 713         for (i = 0; i < MAX_IP_NAT_PROTO; i++)
 714                 RCU_INIT_POINTER(nf_nat_protos[i], &nf_nat_unknown_protocol);
 715         RCU_INIT_POINTER(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
 716         RCU_INIT_POINTER(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
 717         RCU_INIT_POINTER(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
 718         spin_unlock_bh(&nf_nat_lock);
 719
 720         /* Initialize fake conntrack so that NAT will skip it */
 721         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 722
 723         l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
 724
 725         nf_ct_helper_expectfn_register(&follow_master_nat);
 726
 727         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 728         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 729         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 730         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 731                            nfnetlink_parse_nat_setup);
 732         BUG_ON(nf_ct_nat_offset != NULL);
 733         RCU_INIT_POINTER(nf_ct_nat_offset, nf_nat_get_offset);
 734         return 0;
 735
 736  cleanup_extend:
 737         nf_ct_extend_unregister(&nat_extend);
 738         return ret;
 739 }
 740
 741 static void __exit nf_nat_cleanup(void)
 742 {
 743         unregister_pernet_subsys(&nf_nat_net_ops);
 744         nf_ct_l3proto_put(l3proto);
 745         nf_ct_extend_unregister(&nat_extend);
 746         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 747         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, NULL);
 748         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 749         RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
 750         synchronize_net();
 751 }
 752
 753 MODULE_LICENSE("GPL");
 754 MODULE_ALIAS("nf-nat-ipv4");
 755
 756 module_init(nf_nat_init);
 757 module_exit(nf_nat_cleanup);