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