net/netfilter/nf_nat_core.c

   1 /*
   2  * (C) 1999-2001 Paul `Rusty' Russell
   3  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   4  * (C) 2011 Patrick McHardy <kaber@trash.net>
   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/xfrm.h>
  17 #include <linux/jhash.h>
  18 #include <linux/rtnetlink.h>
  19
  20 #include <net/netfilter/nf_conntrack.h>
  21 #include <net/netfilter/nf_conntrack_core.h>
  22 #include <net/netfilter/nf_nat.h>
  23 #include <net/netfilter/nf_nat_l3proto.h>
  24 #include <net/netfilter/nf_nat_l4proto.h>
  25 #include <net/netfilter/nf_nat_core.h>
  26 #include <net/netfilter/nf_nat_helper.h>
  27 #include <net/netfilter/nf_conntrack_helper.h>
  28 #include <net/netfilter/nf_conntrack_l3proto.h>
  29 #include <net/netfilter/nf_conntrack_zones.h>
  30 #include <linux/netfilter/nf_nat.h>
  31
  32 static DEFINE_SPINLOCK(nf_nat_lock);
  33
  34 static DEFINE_MUTEX(nf_nat_proto_mutex);
  35 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
  36                                                 __read_mostly;
  37 static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO]
  38                                                 __read_mostly;
  39
  40
  41 inline const struct nf_nat_l3proto *
  42 __nf_nat_l3proto_find(u8 family)
  43 {
  44         return rcu_dereference(nf_nat_l3protos[family]);
  45 }
  46
  47 inline const struct nf_nat_l4proto *
  48 __nf_nat_l4proto_find(u8 family, u8 protonum)
  49 {
  50         return rcu_dereference(nf_nat_l4protos[family][protonum]);
  51 }
  52 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
  53
  54 #ifdef CONFIG_XFRM
  55 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  56 {
  57         const struct nf_nat_l3proto *l3proto;
  58         const struct nf_conn *ct;
  59         enum ip_conntrack_info ctinfo;
  60         enum ip_conntrack_dir dir;
  61         unsigned  long statusbit;
  62         u8 family;
  63
  64         ct = nf_ct_get(skb, &ctinfo);
  65         if (ct == NULL)
  66                 return;
  67
  68         family = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.l3num;
  69         rcu_read_lock();
  70         l3proto = __nf_nat_l3proto_find(family);
  71         if (l3proto == NULL)
  72                 goto out;
  73
  74         dir = CTINFO2DIR(ctinfo);
  75         if (dir == IP_CT_DIR_ORIGINAL)
  76                 statusbit = IPS_DST_NAT;
  77         else
  78                 statusbit = IPS_SRC_NAT;
  79
  80         l3proto->decode_session(skb, ct, dir, statusbit, fl);
  81 out:
  82         rcu_read_unlock();
  83 }
  84
  85 int nf_xfrm_me_harder(struct sk_buff *skb, unsigned int family)
  86 {
  87         struct flowi fl;
  88         unsigned int hh_len;
  89         struct dst_entry *dst;
  90
  91         if (xfrm_decode_session(skb, &fl, family) < 0)
  92                 return -1;
  93
  94         dst = skb_dst(skb);
  95         if (dst->xfrm)
  96                 dst = ((struct xfrm_dst *)dst)->route;
  97         dst_hold(dst);
  98
  99         dst = xfrm_lookup(dev_net(dst->dev), dst, &fl, skb->sk, 0);
 100         if (IS_ERR(dst))
 101                 return -1;
 102
 103         skb_dst_drop(skb);
 104         skb_dst_set(skb, dst);
 105
 106         /* Change in oif may mean change in hh_len. */
 107         hh_len = skb_dst(skb)->dev->hard_header_len;
 108         if (skb_headroom(skb) < hh_len &&
 109             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 110                 return -1;
 111         return 0;
 112 }
 113 EXPORT_SYMBOL(nf_xfrm_me_harder);
 114 #endif /* CONFIG_XFRM */
 115
 116 /* We keep an extra hash for each conntrack, for fast searching. */
 117 static inline unsigned int
 118 hash_by_src(const struct net *net, u16 zone,
 119             const struct nf_conntrack_tuple *tuple)
 120 {
 121         unsigned int hash;
 122
 123         /* Original src, to ensure we map it consistently if poss. */
 124         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 125                       tuple->dst.protonum ^ zone ^ nf_conntrack_hash_rnd);
 126         return ((u64)hash * net->ct.nat_htable_size) >> 32;
 127 }
 128
 129 /* Is this tuple already taken? (not by us) */
 130 int
 131 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 132                   const struct nf_conn *ignored_conntrack)
 133 {
 134         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 135          * incoming ones.  NAT means they don't have a fixed mapping,
 136          * so we invert the tuple and look for the incoming reply.
 137          *
 138          * We could keep a separate hash if this proves too slow.
 139          */
 140         struct nf_conntrack_tuple reply;
 141
 142         nf_ct_invert_tuplepr(&reply, tuple);
 143         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 144 }
 145 EXPORT_SYMBOL(nf_nat_used_tuple);
 146
 147 /* If we source map this tuple so reply looks like reply_tuple, will
 148  * that meet the constraints of range.
 149  */
 150 static int in_range(const struct nf_nat_l3proto *l3proto,
 151                     const struct nf_nat_l4proto *l4proto,
 152                     const struct nf_conntrack_tuple *tuple,
 153                     const struct nf_nat_range *range)
 154 {
 155         /* If we are supposed to map IPs, then we must be in the
 156          * range specified, otherwise let this drag us onto a new src IP.
 157          */
 158         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 159             !l3proto->in_range(tuple, range))
 160                 return 0;
 161
 162         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 163             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 164                               &range->min_proto, &range->max_proto))
 165                 return 1;
 166
 167         return 0;
 168 }
 169
 170 static inline int
 171 same_src(const struct nf_conn *ct,
 172          const struct nf_conntrack_tuple *tuple)
 173 {
 174         const struct nf_conntrack_tuple *t;
 175
 176         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 177         return (t->dst.protonum == tuple->dst.protonum &&
 178                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 179                 t->src.u.all == tuple->src.u.all);
 180 }
 181
 182 /* Only called for SRC manip */
 183 static int
 184 find_appropriate_src(struct net *net, u16 zone,
 185                      const struct nf_nat_l3proto *l3proto,
 186                      const struct nf_nat_l4proto *l4proto,
 187                      const struct nf_conntrack_tuple *tuple,
 188                      struct nf_conntrack_tuple *result,
 189                      const struct nf_nat_range *range)
 190 {
 191         unsigned int h = hash_by_src(net, zone, tuple);
 192         const struct nf_conn_nat *nat;
 193         const struct nf_conn *ct;
 194
 195         hlist_for_each_entry_rcu(nat, &net->ct.nat_bysource[h], bysource) {
 196                 ct = nat->ct;
 197                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
 198                         /* Copy source part from reply tuple. */
 199                         nf_ct_invert_tuplepr(result,
 200                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 201                         result->dst = tuple->dst;
 202
 203                         if (in_range(l3proto, l4proto, result, range))
 204                                 return 1;
 205                 }
 206         }
 207         return 0;
 208 }
 209
 210 /* For [FUTURE] fragmentation handling, we want the least-used
 211  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 212  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 213  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 214  * the ip with the lowest src-ip/dst-ip/proto usage.
 215  */
 216 static void
 217 find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple,
 218                     const struct nf_nat_range *range,
 219                     const struct nf_conn *ct,
 220                     enum nf_nat_manip_type maniptype)
 221 {
 222         union nf_inet_addr *var_ipp;
 223         unsigned int i, max;
 224         /* Host order */
 225         u32 minip, maxip, j, dist;
 226         bool full_range;
 227
 228         /* No IP mapping?  Do nothing. */
 229         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 230                 return;
 231
 232         if (maniptype == NF_NAT_MANIP_SRC)
 233                 var_ipp = &tuple->src.u3;
 234         else
 235                 var_ipp = &tuple->dst.u3;
 236
 237         /* Fast path: only one choice. */
 238         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 239                 *var_ipp = range->min_addr;
 240                 return;
 241         }
 242
 243         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 244                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 245         else
 246                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 247
 248         /* Hashing source and destination IPs gives a fairly even
 249          * spread in practice (if there are a small number of IPs
 250          * involved, there usually aren't that many connections
 251          * anyway).  The consistency means that servers see the same
 252          * client coming from the same IP (some Internet Banking sites
 253          * like this), even across reboots.
 254          */
 255         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 256                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 257                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone);
 258
 259         full_range = false;
 260         for (i = 0; i <= max; i++) {
 261                 /* If first bytes of the address are at the maximum, use the
 262                  * distance. Otherwise use the full range.
 263                  */
 264                 if (!full_range) {
 265                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 266                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 267                         dist  = maxip - minip + 1;
 268                 } else {
 269                         minip = 0;
 270                         dist  = ~0;
 271                 }
 272
 273                 var_ipp->all[i] = (__force __u32)
 274                         htonl(minip + (((u64)j * dist) >> 32));
 275                 if (var_ipp->all[i] != range->max_addr.all[i])
 276                         full_range = true;
 277
 278                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 279                         j ^= (__force u32)tuple->dst.u3.all[i];
 280         }
 281 }
 282
 283 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 284  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 285  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 286  * range. It might not be possible to get a unique tuple, but we try.
 287  * At worst (or if we race), we will end up with a final duplicate in
 288  * __ip_conntrack_confirm and drop the packet. */
 289 static void
 290 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 291                  const struct nf_conntrack_tuple *orig_tuple,
 292                  const struct nf_nat_range *range,
 293                  struct nf_conn *ct,
 294                  enum nf_nat_manip_type maniptype)
 295 {
 296         const struct nf_nat_l3proto *l3proto;
 297         const struct nf_nat_l4proto *l4proto;
 298         struct net *net = nf_ct_net(ct);
 299         u16 zone = nf_ct_zone(ct);
 300
 301         rcu_read_lock();
 302         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 303         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 304                                         orig_tuple->dst.protonum);
 305
 306         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 307          * and that same mapping gives a unique tuple within the given
 308          * range, use that.
 309          *
 310          * This is only required for source (ie. NAT/masq) mappings.
 311          * So far, we don't do local source mappings, so multiple
 312          * manips not an issue.
 313          */
 314         if (maniptype == NF_NAT_MANIP_SRC &&
 315             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 316                 /* try the original tuple first */
 317                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 318                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 319                                 *tuple = *orig_tuple;
 320                                 goto out;
 321                         }
 322                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 323                                                 orig_tuple, tuple, range)) {
 324                         pr_debug("get_unique_tuple: Found current src map\n");
 325                         if (!nf_nat_used_tuple(tuple, ct))
 326                                 goto out;
 327                 }
 328         }
 329
 330         /* 2) Select the least-used IP/proto combination in the given range */
 331         *tuple = *orig_tuple;
 332         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 333
 334         /* 3) The per-protocol part of the manip is made to map into
 335          * the range to make a unique tuple.
 336          */
 337
 338         /* Only bother mapping if it's not already in range and unique */
 339         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 340                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 341                         if (l4proto->in_range(tuple, maniptype,
 342                                               &range->min_proto,
 343                                               &range->max_proto) &&
 344                             (range->min_proto.all == range->max_proto.all ||
 345                              !nf_nat_used_tuple(tuple, ct)))
 346                                 goto out;
 347                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 348                         goto out;
 349                 }
 350         }
 351
 352         /* Last change: get protocol to try to obtain unique tuple. */
 353         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 354 out:
 355         rcu_read_unlock();
 356 }
 357
 358 unsigned int
 359 nf_nat_setup_info(struct nf_conn *ct,
 360                   const struct nf_nat_range *range,
 361                   enum nf_nat_manip_type maniptype)
 362 {
 363         struct net *net = nf_ct_net(ct);
 364         struct nf_conntrack_tuple curr_tuple, new_tuple;
 365         struct nf_conn_nat *nat;
 366
 367         /* nat helper or nfctnetlink also setup binding */
 368         nat = nfct_nat(ct);
 369         if (!nat) {
 370                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 371                 if (nat == NULL) {
 372                         pr_debug("failed to add NAT extension\n");
 373                         return NF_ACCEPT;
 374                 }
 375         }
 376
 377         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 378                      maniptype == NF_NAT_MANIP_DST);
 379         BUG_ON(nf_nat_initialized(ct, maniptype));
 380
 381         /* What we've got will look like inverse of reply. Normally
 382          * this is what is in the conntrack, except for prior
 383          * manipulations (future optimization: if num_manips == 0,
 384          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 385          */
 386         nf_ct_invert_tuplepr(&curr_tuple,
 387                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 388
 389         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 390
 391         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 392                 struct nf_conntrack_tuple reply;
 393
 394                 /* Alter conntrack table so will recognize replies. */
 395                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 396                 nf_conntrack_alter_reply(ct, &reply);
 397
 398                 /* Non-atomic: we own this at the moment. */
 399                 if (maniptype == NF_NAT_MANIP_SRC)
 400                         ct->status |= IPS_SRC_NAT;
 401                 else
 402                         ct->status |= IPS_DST_NAT;
 403         }
 404
 405         if (maniptype == NF_NAT_MANIP_SRC) {
 406                 unsigned int srchash;
 407
 408                 srchash = hash_by_src(net, nf_ct_zone(ct),
 409                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 410                 spin_lock_bh(&nf_nat_lock);
 411                 /* nf_conntrack_alter_reply might re-allocate extension aera */
 412                 nat = nfct_nat(ct);
 413                 nat->ct = ct;
 414                 hlist_add_head_rcu(&nat->bysource,
 415                                    &net->ct.nat_bysource[srchash]);
 416                 spin_unlock_bh(&nf_nat_lock);
 417         }
 418
 419         /* It's done. */
 420         if (maniptype == NF_NAT_MANIP_DST)
 421                 ct->status |= IPS_DST_NAT_DONE;
 422         else
 423                 ct->status |= IPS_SRC_NAT_DONE;
 424
 425         return NF_ACCEPT;
 426 }
 427 EXPORT_SYMBOL(nf_nat_setup_info);
 428
 429 /* Do packet manipulations according to nf_nat_setup_info. */
 430 unsigned int nf_nat_packet(struct nf_conn *ct,
 431                            enum ip_conntrack_info ctinfo,
 432                            unsigned int hooknum,
 433                            struct sk_buff *skb)
 434 {
 435         const struct nf_nat_l3proto *l3proto;
 436         const struct nf_nat_l4proto *l4proto;
 437         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 438         unsigned long statusbit;
 439         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 440
 441         if (mtype == NF_NAT_MANIP_SRC)
 442                 statusbit = IPS_SRC_NAT;
 443         else
 444                 statusbit = IPS_DST_NAT;
 445
 446         /* Invert if this is reply dir. */
 447         if (dir == IP_CT_DIR_REPLY)
 448                 statusbit ^= IPS_NAT_MASK;
 449
 450         /* Non-atomic: these bits don't change. */
 451         if (ct->status & statusbit) {
 452                 struct nf_conntrack_tuple target;
 453
 454                 /* We are aiming to look like inverse of other direction. */
 455                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 456
 457                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 458                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 459                                                 target.dst.protonum);
 460                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 461                         return NF_DROP;
 462         }
 463         return NF_ACCEPT;
 464 }
 465 EXPORT_SYMBOL_GPL(nf_nat_packet);
 466
 467 struct nf_nat_proto_clean {
 468         u8      l3proto;
 469         u8      l4proto;
 470 };
 471
 472 /* kill conntracks with affected NAT section */
 473 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 474 {
 475         const struct nf_nat_proto_clean *clean = data;
 476         struct nf_conn_nat *nat = nfct_nat(i);
 477
 478         if (!nat)
 479                 return 0;
 480
 481         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 482             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 483                 return 0;
 484
 485         return i->status & IPS_NAT_MASK ? 1 : 0;
 486 }
 487
 488 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 489 {
 490         struct nf_nat_proto_clean clean = {
 491                 .l3proto = l3proto,
 492                 .l4proto = l4proto,
 493         };
 494         struct net *net;
 495
 496         rtnl_lock();
 497         for_each_net(net)
 498                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean);
 499         rtnl_unlock();
 500 }
 501
 502 static void nf_nat_l3proto_clean(u8 l3proto)
 503 {
 504         struct nf_nat_proto_clean clean = {
 505                 .l3proto = l3proto,
 506         };
 507         struct net *net;
 508
 509         rtnl_lock();
 510
 511         for_each_net(net)
 512                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean);
 513         rtnl_unlock();
 514 }
 515
 516 /* Protocol registration. */
 517 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 518 {
 519         const struct nf_nat_l4proto **l4protos;
 520         unsigned int i;
 521         int ret = 0;
 522
 523         mutex_lock(&nf_nat_proto_mutex);
 524         if (nf_nat_l4protos[l3proto] == NULL) {
 525                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 526                                    GFP_KERNEL);
 527                 if (l4protos == NULL) {
 528                         ret = -ENOMEM;
 529                         goto out;
 530                 }
 531
 532                 for (i = 0; i < IPPROTO_MAX; i++)
 533                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 534
 535                 /* Before making proto_array visible to lockless readers,
 536                  * we must make sure its content is committed to memory.
 537                  */
 538                 smp_wmb();
 539
 540                 nf_nat_l4protos[l3proto] = l4protos;
 541         }
 542
 543         if (rcu_dereference_protected(
 544                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 545                         lockdep_is_held(&nf_nat_proto_mutex)
 546                         ) != &nf_nat_l4proto_unknown) {
 547                 ret = -EBUSY;
 548                 goto out;
 549         }
 550         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 551  out:
 552         mutex_unlock(&nf_nat_proto_mutex);
 553         return ret;
 554 }
 555 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 556
 557 /* No one stores the protocol anywhere; simply delete it. */
 558 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 559 {
 560         mutex_lock(&nf_nat_proto_mutex);
 561         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 562                          &nf_nat_l4proto_unknown);
 563         mutex_unlock(&nf_nat_proto_mutex);
 564         synchronize_rcu();
 565
 566         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 567 }
 568 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 569
 570 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 571 {
 572         int err;
 573
 574         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 575         if (err < 0)
 576                 return err;
 577
 578         mutex_lock(&nf_nat_proto_mutex);
 579         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 580                          &nf_nat_l4proto_tcp);
 581         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 582                          &nf_nat_l4proto_udp);
 583         mutex_unlock(&nf_nat_proto_mutex);
 584
 585         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 586         return 0;
 587 }
 588 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 589
 590 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 591 {
 592         mutex_lock(&nf_nat_proto_mutex);
 593         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 594         mutex_unlock(&nf_nat_proto_mutex);
 595         synchronize_rcu();
 596
 597         nf_nat_l3proto_clean(l3proto->l3proto);
 598         nf_ct_l3proto_module_put(l3proto->l3proto);
 599 }
 600 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 601
 602 /* No one using conntrack by the time this called. */
 603 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 604 {
 605         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 606
 607         if (nat == NULL || nat->ct == NULL)
 608                 return;
 609
 610         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 611
 612         spin_lock_bh(&nf_nat_lock);
 613         hlist_del_rcu(&nat->bysource);
 614         spin_unlock_bh(&nf_nat_lock);
 615 }
 616
 617 static void nf_nat_move_storage(void *new, void *old)
 618 {
 619         struct nf_conn_nat *new_nat = new;
 620         struct nf_conn_nat *old_nat = old;
 621         struct nf_conn *ct = old_nat->ct;
 622
 623         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 624                 return;
 625
 626         spin_lock_bh(&nf_nat_lock);
 627         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 628         spin_unlock_bh(&nf_nat_lock);
 629 }
 630
 631 static struct nf_ct_ext_type nat_extend __read_mostly = {
 632         .len            = sizeof(struct nf_conn_nat),
 633         .align          = __alignof__(struct nf_conn_nat),
 634         .destroy        = nf_nat_cleanup_conntrack,
 635         .move           = nf_nat_move_storage,
 636         .id             = NF_CT_EXT_NAT,
 637         .flags          = NF_CT_EXT_F_PREALLOC,
 638 };
 639
 640 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 641
 642 #include <linux/netfilter/nfnetlink.h>
 643 #include <linux/netfilter/nfnetlink_conntrack.h>
 644
 645 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 646         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 647         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 648 };
 649
 650 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 651                                      const struct nf_conn *ct,
 652                                      struct nf_nat_range *range)
 653 {
 654         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 655         const struct nf_nat_l4proto *l4proto;
 656         int err;
 657
 658         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 659         if (err < 0)
 660                 return err;
 661
 662         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 663         if (l4proto->nlattr_to_range)
 664                 err = l4proto->nlattr_to_range(tb, range);
 665
 666         return err;
 667 }
 668
 669 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 670         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 671         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 672         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 673         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 674         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 675 };
 676
 677 static int
 678 nfnetlink_parse_nat(const struct nlattr *nat,
 679                     const struct nf_conn *ct, struct nf_nat_range *range)
 680 {
 681         const struct nf_nat_l3proto *l3proto;
 682         struct nlattr *tb[CTA_NAT_MAX+1];
 683         int err;
 684
 685         memset(range, 0, sizeof(*range));
 686
 687         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 688         if (err < 0)
 689                 return err;
 690
 691         rcu_read_lock();
 692         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 693         if (l3proto == NULL) {
 694                 err = -EAGAIN;
 695                 goto out;
 696         }
 697         err = l3proto->nlattr_to_range(tb, range);
 698         if (err < 0)
 699                 goto out;
 700
 701         if (!tb[CTA_NAT_PROTO])
 702                 goto out;
 703
 704         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 705 out:
 706         rcu_read_unlock();
 707         return err;
 708 }
 709
 710 static int
 711 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 712                           enum nf_nat_manip_type manip,
 713                           const struct nlattr *attr)
 714 {
 715         struct nf_nat_range range;
 716         int err;
 717
 718         err = nfnetlink_parse_nat(attr, ct, &range);
 719         if (err < 0)
 720                 return err;
 721         if (nf_nat_initialized(ct, manip))
 722                 return -EEXIST;
 723
 724         return nf_nat_setup_info(ct, &range, manip);
 725 }
 726 #else
 727 static int
 728 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 729                           enum nf_nat_manip_type manip,
 730                           const struct nlattr *attr)
 731 {
 732         return -EOPNOTSUPP;
 733 }
 734 #endif
 735
 736 static int __net_init nf_nat_net_init(struct net *net)
 737 {
 738         /* Leave them the same for the moment. */
 739         net->ct.nat_htable_size = net->ct.htable_size;
 740         net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0);
 741         if (!net->ct.nat_bysource)
 742                 return -ENOMEM;
 743         return 0;
 744 }
 745
 746 static void __net_exit nf_nat_net_exit(struct net *net)
 747 {
 748         struct nf_nat_proto_clean clean = {};
 749
 750         nf_ct_iterate_cleanup(net, &nf_nat_proto_remove, &clean);
 751         synchronize_rcu();
 752         nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size);
 753 }
 754
 755 static struct pernet_operations nf_nat_net_ops = {
 756         .init = nf_nat_net_init,
 757         .exit = nf_nat_net_exit,
 758 };
 759
 760 static struct nf_ct_helper_expectfn follow_master_nat = {
 761         .name           = "nat-follow-master",
 762         .expectfn       = nf_nat_follow_master,
 763 };
 764
 765 static struct nfq_ct_nat_hook nfq_ct_nat = {
 766         .seq_adjust     = nf_nat_tcp_seq_adjust,
 767 };
 768
 769 static int __init nf_nat_init(void)
 770 {
 771         int ret;
 772
 773         ret = nf_ct_extend_register(&nat_extend);
 774         if (ret < 0) {
 775                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 776                 return ret;
 777         }
 778
 779         ret = register_pernet_subsys(&nf_nat_net_ops);
 780         if (ret < 0)
 781                 goto cleanup_extend;
 782
 783         nf_ct_helper_expectfn_register(&follow_master_nat);
 784
 785         /* Initialize fake conntrack so that NAT will skip it */
 786         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 787
 788         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 789         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 790         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 791         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 792                            nfnetlink_parse_nat_setup);
 793         BUG_ON(nf_ct_nat_offset != NULL);
 794         RCU_INIT_POINTER(nf_ct_nat_offset, nf_nat_get_offset);
 795         RCU_INIT_POINTER(nfq_ct_nat_hook, &nfq_ct_nat);
 796 #ifdef CONFIG_XFRM
 797         BUG_ON(nf_nat_decode_session_hook != NULL);
 798         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 799 #endif
 800         return 0;
 801
 802  cleanup_extend:
 803         nf_ct_extend_unregister(&nat_extend);
 804         return ret;
 805 }
 806
 807 static void __exit nf_nat_cleanup(void)
 808 {
 809         unsigned int i;
 810
 811         unregister_pernet_subsys(&nf_nat_net_ops);
 812         nf_ct_extend_unregister(&nat_extend);
 813         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 814         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, NULL);
 815         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 816         RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
 817         RCU_INIT_POINTER(nfq_ct_nat_hook, NULL);
 818 #ifdef CONFIG_XFRM
 819         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 820 #endif
 821         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 822                 kfree(nf_nat_l4protos[i]);
 823         synchronize_net();
 824 }
 825
 826 MODULE_LICENSE("GPL");
 827
 828 module_init(nf_nat_init);
 829 module_exit(nf_nat_cleanup);