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         const struct hlist_node *n;
 195
 196         hlist_for_each_entry_rcu(nat, n, &net->ct.nat_bysource[h], bysource) {
 197                 ct = nat->ct;
 198                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
 199                         /* Copy source part from reply tuple. */
 200                         nf_ct_invert_tuplepr(result,
 201                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 202                         result->dst = tuple->dst;
 203
 204                         if (in_range(l3proto, l4proto, result, range))
 205                                 return 1;
 206                 }
 207         }
 208         return 0;
 209 }
 210
 211 /* For [FUTURE] fragmentation handling, we want the least-used
 212  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 213  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 214  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 215  * the ip with the lowest src-ip/dst-ip/proto usage.
 216  */
 217 static void
 218 find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple,
 219                     const struct nf_nat_range *range,
 220                     const struct nf_conn *ct,
 221                     enum nf_nat_manip_type maniptype)
 222 {
 223         union nf_inet_addr *var_ipp;
 224         unsigned int i, max;
 225         /* Host order */
 226         u32 minip, maxip, j, dist;
 227         bool full_range;
 228
 229         /* No IP mapping?  Do nothing. */
 230         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 231                 return;
 232
 233         if (maniptype == NF_NAT_MANIP_SRC)
 234                 var_ipp = &tuple->src.u3;
 235         else
 236                 var_ipp = &tuple->dst.u3;
 237
 238         /* Fast path: only one choice. */
 239         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 240                 *var_ipp = range->min_addr;
 241                 return;
 242         }
 243
 244         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 245                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 246         else
 247                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 248
 249         /* Hashing source and destination IPs gives a fairly even
 250          * spread in practice (if there are a small number of IPs
 251          * involved, there usually aren't that many connections
 252          * anyway).  The consistency means that servers see the same
 253          * client coming from the same IP (some Internet Banking sites
 254          * like this), even across reboots.
 255          */
 256         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 257                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 258                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone);
 259
 260         full_range = false;
 261         for (i = 0; i <= max; i++) {
 262                 /* If first bytes of the address are at the maximum, use the
 263                  * distance. Otherwise use the full range.
 264                  */
 265                 if (!full_range) {
 266                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 267                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 268                         dist  = maxip - minip + 1;
 269                 } else {
 270                         minip = 0;
 271                         dist  = ~0;
 272                 }
 273
 274                 var_ipp->all[i] = (__force __u32)
 275                         htonl(minip + (((u64)j * dist) >> 32));
 276                 if (var_ipp->all[i] != range->max_addr.all[i])
 277                         full_range = true;
 278
 279                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 280                         j ^= (__force u32)tuple->dst.u3.all[i];
 281         }
 282 }
 283
 284 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 285  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 286  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 287  * range. It might not be possible to get a unique tuple, but we try.
 288  * At worst (or if we race), we will end up with a final duplicate in
 289  * __ip_conntrack_confirm and drop the packet. */
 290 static void
 291 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 292                  const struct nf_conntrack_tuple *orig_tuple,
 293                  const struct nf_nat_range *range,
 294                  struct nf_conn *ct,
 295                  enum nf_nat_manip_type maniptype)
 296 {
 297         const struct nf_nat_l3proto *l3proto;
 298         const struct nf_nat_l4proto *l4proto;
 299         struct net *net = nf_ct_net(ct);
 300         u16 zone = nf_ct_zone(ct);
 301
 302         rcu_read_lock();
 303         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 304         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 305                                         orig_tuple->dst.protonum);
 306
 307         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 308          * and that same mapping gives a unique tuple within the given
 309          * range, use that.
 310          *
 311          * This is only required for source (ie. NAT/masq) mappings.
 312          * So far, we don't do local source mappings, so multiple
 313          * manips not an issue.
 314          */
 315         if (maniptype == NF_NAT_MANIP_SRC &&
 316             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 317                 /* try the original tuple first */
 318                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 319                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 320                                 *tuple = *orig_tuple;
 321                                 goto out;
 322                         }
 323                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 324                                                 orig_tuple, tuple, range)) {
 325                         pr_debug("get_unique_tuple: Found current src map\n");
 326                         if (!nf_nat_used_tuple(tuple, ct))
 327                                 goto out;
 328                 }
 329         }
 330
 331         /* 2) Select the least-used IP/proto combination in the given range */
 332         *tuple = *orig_tuple;
 333         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 334
 335         /* 3) The per-protocol part of the manip is made to map into
 336          * the range to make a unique tuple.
 337          */
 338
 339         /* Only bother mapping if it's not already in range and unique */
 340         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 341                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 342                         if (l4proto->in_range(tuple, maniptype,
 343                                               &range->min_proto,
 344                                               &range->max_proto) &&
 345                             (range->min_proto.all == range->max_proto.all ||
 346                              !nf_nat_used_tuple(tuple, ct)))
 347                                 goto out;
 348                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 349                         goto out;
 350                 }
 351         }
 352
 353         /* Last change: get protocol to try to obtain unique tuple. */
 354         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 355 out:
 356         rcu_read_unlock();
 357 }
 358
 359 unsigned int
 360 nf_nat_setup_info(struct nf_conn *ct,
 361                   const struct nf_nat_range *range,
 362                   enum nf_nat_manip_type maniptype)
 363 {
 364         struct net *net = nf_ct_net(ct);
 365         struct nf_conntrack_tuple curr_tuple, new_tuple;
 366         struct nf_conn_nat *nat;
 367
 368         /* nat helper or nfctnetlink also setup binding */
 369         nat = nfct_nat(ct);
 370         if (!nat) {
 371                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 372                 if (nat == NULL) {
 373                         pr_debug("failed to add NAT extension\n");
 374                         return NF_ACCEPT;
 375                 }
 376         }
 377
 378         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 379                      maniptype == NF_NAT_MANIP_DST);
 380         BUG_ON(nf_nat_initialized(ct, maniptype));
 381
 382         /* What we've got will look like inverse of reply. Normally
 383          * this is what is in the conntrack, except for prior
 384          * manipulations (future optimization: if num_manips == 0,
 385          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 386          */
 387         nf_ct_invert_tuplepr(&curr_tuple,
 388                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 389
 390         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 391
 392         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 393                 struct nf_conntrack_tuple reply;
 394
 395                 /* Alter conntrack table so will recognize replies. */
 396                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 397                 nf_conntrack_alter_reply(ct, &reply);
 398
 399                 /* Non-atomic: we own this at the moment. */
 400                 if (maniptype == NF_NAT_MANIP_SRC)
 401                         ct->status |= IPS_SRC_NAT;
 402                 else
 403                         ct->status |= IPS_DST_NAT;
 404         }
 405
 406         if (maniptype == NF_NAT_MANIP_SRC) {
 407                 unsigned int srchash;
 408
 409                 srchash = hash_by_src(net, nf_ct_zone(ct),
 410                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 411                 spin_lock_bh(&nf_nat_lock);
 412                 /* nf_conntrack_alter_reply might re-allocate extension aera */
 413                 nat = nfct_nat(ct);
 414                 nat->ct = ct;
 415                 hlist_add_head_rcu(&nat->bysource,
 416                                    &net->ct.nat_bysource[srchash]);
 417                 spin_unlock_bh(&nf_nat_lock);
 418         }
 419
 420         /* It's done. */
 421         if (maniptype == NF_NAT_MANIP_DST)
 422                 ct->status |= IPS_DST_NAT_DONE;
 423         else
 424                 ct->status |= IPS_SRC_NAT_DONE;
 425
 426         return NF_ACCEPT;
 427 }
 428 EXPORT_SYMBOL(nf_nat_setup_info);
 429
 430 /* Do packet manipulations according to nf_nat_setup_info. */
 431 unsigned int nf_nat_packet(struct nf_conn *ct,
 432                            enum ip_conntrack_info ctinfo,
 433                            unsigned int hooknum,
 434                            struct sk_buff *skb)
 435 {
 436         const struct nf_nat_l3proto *l3proto;
 437         const struct nf_nat_l4proto *l4proto;
 438         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 439         unsigned long statusbit;
 440         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 441
 442         if (mtype == NF_NAT_MANIP_SRC)
 443                 statusbit = IPS_SRC_NAT;
 444         else
 445                 statusbit = IPS_DST_NAT;
 446
 447         /* Invert if this is reply dir. */
 448         if (dir == IP_CT_DIR_REPLY)
 449                 statusbit ^= IPS_NAT_MASK;
 450
 451         /* Non-atomic: these bits don't change. */
 452         if (ct->status & statusbit) {
 453                 struct nf_conntrack_tuple target;
 454
 455                 /* We are aiming to look like inverse of other direction. */
 456                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 457
 458                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 459                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 460                                                 target.dst.protonum);
 461                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 462                         return NF_DROP;
 463         }
 464         return NF_ACCEPT;
 465 }
 466 EXPORT_SYMBOL_GPL(nf_nat_packet);
 467
 468 struct nf_nat_proto_clean {
 469         u8      l3proto;
 470         u8      l4proto;
 471         bool    hash;
 472 };
 473
 474 /* Clear NAT section of all conntracks, in case we're loaded again. */
 475 static int nf_nat_proto_clean(struct nf_conn *i, void *data)
 476 {
 477         const struct nf_nat_proto_clean *clean = data;
 478         struct nf_conn_nat *nat = nfct_nat(i);
 479
 480         if (!nat)
 481                 return 0;
 482         if (!(i->status & IPS_SRC_NAT_DONE))
 483                 return 0;
 484         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 485             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 486                 return 0;
 487
 488         if (clean->hash) {
 489                 spin_lock_bh(&nf_nat_lock);
 490                 hlist_del_rcu(&nat->bysource);
 491                 spin_unlock_bh(&nf_nat_lock);
 492         } else {
 493                 memset(nat, 0, sizeof(*nat));
 494                 i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK |
 495                                IPS_SEQ_ADJUST);
 496         }
 497         return 0;
 498 }
 499
 500 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 501 {
 502         struct nf_nat_proto_clean clean = {
 503                 .l3proto = l3proto,
 504                 .l4proto = l4proto,
 505         };
 506         struct net *net;
 507
 508         rtnl_lock();
 509         /* Step 1 - remove from bysource hash */
 510         clean.hash = true;
 511         for_each_net(net)
 512                 nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean);
 513         synchronize_rcu();
 514
 515         /* Step 2 - clean NAT section */
 516         clean.hash = false;
 517         for_each_net(net)
 518                 nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean);
 519         rtnl_unlock();
 520 }
 521
 522 static void nf_nat_l3proto_clean(u8 l3proto)
 523 {
 524         struct nf_nat_proto_clean clean = {
 525                 .l3proto = l3proto,
 526         };
 527         struct net *net;
 528
 529         rtnl_lock();
 530         /* Step 1 - remove from bysource hash */
 531         clean.hash = true;
 532         for_each_net(net)
 533                 nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean);
 534         synchronize_rcu();
 535
 536         /* Step 2 - clean NAT section */
 537         clean.hash = false;
 538         for_each_net(net)
 539                 nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean);
 540         rtnl_unlock();
 541 }
 542
 543 /* Protocol registration. */
 544 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 545 {
 546         const struct nf_nat_l4proto **l4protos;
 547         unsigned int i;
 548         int ret = 0;
 549
 550         mutex_lock(&nf_nat_proto_mutex);
 551         if (nf_nat_l4protos[l3proto] == NULL) {
 552                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 553                                    GFP_KERNEL);
 554                 if (l4protos == NULL) {
 555                         ret = -ENOMEM;
 556                         goto out;
 557                 }
 558
 559                 for (i = 0; i < IPPROTO_MAX; i++)
 560                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 561
 562                 /* Before making proto_array visible to lockless readers,
 563                  * we must make sure its content is committed to memory.
 564                  */
 565                 smp_wmb();
 566
 567                 nf_nat_l4protos[l3proto] = l4protos;
 568         }
 569
 570         if (rcu_dereference_protected(
 571                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 572                         lockdep_is_held(&nf_nat_proto_mutex)
 573                         ) != &nf_nat_l4proto_unknown) {
 574                 ret = -EBUSY;
 575                 goto out;
 576         }
 577         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 578  out:
 579         mutex_unlock(&nf_nat_proto_mutex);
 580         return ret;
 581 }
 582 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 583
 584 /* No one stores the protocol anywhere; simply delete it. */
 585 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 586 {
 587         mutex_lock(&nf_nat_proto_mutex);
 588         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 589                          &nf_nat_l4proto_unknown);
 590         mutex_unlock(&nf_nat_proto_mutex);
 591         synchronize_rcu();
 592
 593         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 594 }
 595 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 596
 597 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 598 {
 599         int err;
 600
 601         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 602         if (err < 0)
 603                 return err;
 604
 605         mutex_lock(&nf_nat_proto_mutex);
 606         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 607                          &nf_nat_l4proto_tcp);
 608         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 609                          &nf_nat_l4proto_udp);
 610         mutex_unlock(&nf_nat_proto_mutex);
 611
 612         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 613         return 0;
 614 }
 615 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 616
 617 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 618 {
 619         mutex_lock(&nf_nat_proto_mutex);
 620         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 621         mutex_unlock(&nf_nat_proto_mutex);
 622         synchronize_rcu();
 623
 624         nf_nat_l3proto_clean(l3proto->l3proto);
 625         nf_ct_l3proto_module_put(l3proto->l3proto);
 626 }
 627 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 628
 629 /* No one using conntrack by the time this called. */
 630 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 631 {
 632         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 633
 634         if (nat == NULL || nat->ct == NULL)
 635                 return;
 636
 637         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 638
 639         spin_lock_bh(&nf_nat_lock);
 640         hlist_del_rcu(&nat->bysource);
 641         spin_unlock_bh(&nf_nat_lock);
 642 }
 643
 644 static void nf_nat_move_storage(void *new, void *old)
 645 {
 646         struct nf_conn_nat *new_nat = new;
 647         struct nf_conn_nat *old_nat = old;
 648         struct nf_conn *ct = old_nat->ct;
 649
 650         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 651                 return;
 652
 653         spin_lock_bh(&nf_nat_lock);
 654         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 655         spin_unlock_bh(&nf_nat_lock);
 656 }
 657
 658 static struct nf_ct_ext_type nat_extend __read_mostly = {
 659         .len            = sizeof(struct nf_conn_nat),
 660         .align          = __alignof__(struct nf_conn_nat),
 661         .destroy        = nf_nat_cleanup_conntrack,
 662         .move           = nf_nat_move_storage,
 663         .id             = NF_CT_EXT_NAT,
 664         .flags          = NF_CT_EXT_F_PREALLOC,
 665 };
 666
 667 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 668
 669 #include <linux/netfilter/nfnetlink.h>
 670 #include <linux/netfilter/nfnetlink_conntrack.h>
 671
 672 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 673         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 674         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 675 };
 676
 677 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 678                                      const struct nf_conn *ct,
 679                                      struct nf_nat_range *range)
 680 {
 681         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 682         const struct nf_nat_l4proto *l4proto;
 683         int err;
 684
 685         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 686         if (err < 0)
 687                 return err;
 688
 689         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 690         if (l4proto->nlattr_to_range)
 691                 err = l4proto->nlattr_to_range(tb, range);
 692
 693         return err;
 694 }
 695
 696 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 697         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 698         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 699         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 700         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 701         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 702 };
 703
 704 static int
 705 nfnetlink_parse_nat(const struct nlattr *nat,
 706                     const struct nf_conn *ct, struct nf_nat_range *range)
 707 {
 708         const struct nf_nat_l3proto *l3proto;
 709         struct nlattr *tb[CTA_NAT_MAX+1];
 710         int err;
 711
 712         memset(range, 0, sizeof(*range));
 713
 714         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 715         if (err < 0)
 716                 return err;
 717
 718         rcu_read_lock();
 719         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 720         if (l3proto == NULL) {
 721                 err = -EAGAIN;
 722                 goto out;
 723         }
 724         err = l3proto->nlattr_to_range(tb, range);
 725         if (err < 0)
 726                 goto out;
 727
 728         if (!tb[CTA_NAT_PROTO])
 729                 goto out;
 730
 731         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 732 out:
 733         rcu_read_unlock();
 734         return err;
 735 }
 736
 737 static int
 738 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 739                           enum nf_nat_manip_type manip,
 740                           const struct nlattr *attr)
 741 {
 742         struct nf_nat_range range;
 743         int err;
 744
 745         err = nfnetlink_parse_nat(attr, ct, &range);
 746         if (err < 0)
 747                 return err;
 748         if (nf_nat_initialized(ct, manip))
 749                 return -EEXIST;
 750
 751         return nf_nat_setup_info(ct, &range, manip);
 752 }
 753 #else
 754 static int
 755 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 756                           enum nf_nat_manip_type manip,
 757                           const struct nlattr *attr)
 758 {
 759         return -EOPNOTSUPP;
 760 }
 761 #endif
 762
 763 static int __net_init nf_nat_net_init(struct net *net)
 764 {
 765         /* Leave them the same for the moment. */
 766         net->ct.nat_htable_size = net->ct.htable_size;
 767         net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0);
 768         if (!net->ct.nat_bysource)
 769                 return -ENOMEM;
 770         return 0;
 771 }
 772
 773 static void __net_exit nf_nat_net_exit(struct net *net)
 774 {
 775         struct nf_nat_proto_clean clean = {};
 776
 777         nf_ct_iterate_cleanup(net, &nf_nat_proto_clean, &clean);
 778         synchronize_rcu();
 779         nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size);
 780 }
 781
 782 static struct pernet_operations nf_nat_net_ops = {
 783         .init = nf_nat_net_init,
 784         .exit = nf_nat_net_exit,
 785 };
 786
 787 static struct nf_ct_helper_expectfn follow_master_nat = {
 788         .name           = "nat-follow-master",
 789         .expectfn       = nf_nat_follow_master,
 790 };
 791
 792 static struct nfq_ct_nat_hook nfq_ct_nat = {
 793         .seq_adjust     = nf_nat_tcp_seq_adjust,
 794 };
 795
 796 static int __init nf_nat_init(void)
 797 {
 798         int ret;
 799
 800         ret = nf_ct_extend_register(&nat_extend);
 801         if (ret < 0) {
 802                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 803                 return ret;
 804         }
 805
 806         ret = register_pernet_subsys(&nf_nat_net_ops);
 807         if (ret < 0)
 808                 goto cleanup_extend;
 809
 810         nf_ct_helper_expectfn_register(&follow_master_nat);
 811
 812         /* Initialize fake conntrack so that NAT will skip it */
 813         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 814
 815         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 816         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 817         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 818         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 819                            nfnetlink_parse_nat_setup);
 820         BUG_ON(nf_ct_nat_offset != NULL);
 821         RCU_INIT_POINTER(nf_ct_nat_offset, nf_nat_get_offset);
 822         RCU_INIT_POINTER(nfq_ct_nat_hook, &nfq_ct_nat);
 823 #ifdef CONFIG_XFRM
 824         BUG_ON(nf_nat_decode_session_hook != NULL);
 825         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 826 #endif
 827         return 0;
 828
 829  cleanup_extend:
 830         nf_ct_extend_unregister(&nat_extend);
 831         return ret;
 832 }
 833
 834 static void __exit nf_nat_cleanup(void)
 835 {
 836         unsigned int i;
 837
 838         unregister_pernet_subsys(&nf_nat_net_ops);
 839         nf_ct_extend_unregister(&nat_extend);
 840         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 841         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, NULL);
 842         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 843         RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
 844         RCU_INIT_POINTER(nfq_ct_nat_hook, NULL);
 845 #ifdef CONFIG_XFRM
 846         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 847 #endif
 848         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 849                 kfree(nf_nat_l4protos[i]);
 850         synchronize_net();
 851 }
 852
 853 MODULE_LICENSE("GPL");
 854
 855 module_init(nf_nat_init);
 856 module_exit(nf_nat_cleanup);