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         int err;
  91
  92         err = xfrm_decode_session(skb, &fl, family);
  93         if (err < 0)
  94                 return err;
  95
  96         dst = skb_dst(skb);
  97         if (dst->xfrm)
  98                 dst = ((struct xfrm_dst *)dst)->route;
  99         dst_hold(dst);
 100
 101         dst = xfrm_lookup(dev_net(dst->dev), dst, &fl, skb->sk, 0);
 102         if (IS_ERR(dst))
 103                 return PTR_ERR(dst);
 104
 105         skb_dst_drop(skb);
 106         skb_dst_set(skb, dst);
 107
 108         /* Change in oif may mean change in hh_len. */
 109         hh_len = skb_dst(skb)->dev->hard_header_len;
 110         if (skb_headroom(skb) < hh_len &&
 111             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 112                 return -ENOMEM;
 113         return 0;
 114 }
 115 EXPORT_SYMBOL(nf_xfrm_me_harder);
 116 #endif /* CONFIG_XFRM */
 117
 118 /* We keep an extra hash for each conntrack, for fast searching. */
 119 static inline unsigned int
 120 hash_by_src(const struct net *net, u16 zone,
 121             const struct nf_conntrack_tuple *tuple)
 122 {
 123         unsigned int hash;
 124
 125         /* Original src, to ensure we map it consistently if poss. */
 126         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 127                       tuple->dst.protonum ^ zone ^ nf_conntrack_hash_rnd);
 128         return ((u64)hash * net->ct.nat_htable_size) >> 32;
 129 }
 130
 131 /* Is this tuple already taken? (not by us) */
 132 int
 133 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 134                   const struct nf_conn *ignored_conntrack)
 135 {
 136         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 137          * incoming ones.  NAT means they don't have a fixed mapping,
 138          * so we invert the tuple and look for the incoming reply.
 139          *
 140          * We could keep a separate hash if this proves too slow.
 141          */
 142         struct nf_conntrack_tuple reply;
 143
 144         nf_ct_invert_tuplepr(&reply, tuple);
 145         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 146 }
 147 EXPORT_SYMBOL(nf_nat_used_tuple);
 148
 149 /* If we source map this tuple so reply looks like reply_tuple, will
 150  * that meet the constraints of range.
 151  */
 152 static int in_range(const struct nf_nat_l3proto *l3proto,
 153                     const struct nf_nat_l4proto *l4proto,
 154                     const struct nf_conntrack_tuple *tuple,
 155                     const struct nf_nat_range *range)
 156 {
 157         /* If we are supposed to map IPs, then we must be in the
 158          * range specified, otherwise let this drag us onto a new src IP.
 159          */
 160         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 161             !l3proto->in_range(tuple, range))
 162                 return 0;
 163
 164         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 165             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 166                               &range->min_proto, &range->max_proto))
 167                 return 1;
 168
 169         return 0;
 170 }
 171
 172 static inline int
 173 same_src(const struct nf_conn *ct,
 174          const struct nf_conntrack_tuple *tuple)
 175 {
 176         const struct nf_conntrack_tuple *t;
 177
 178         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 179         return (t->dst.protonum == tuple->dst.protonum &&
 180                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 181                 t->src.u.all == tuple->src.u.all);
 182 }
 183
 184 /* Only called for SRC manip */
 185 static int
 186 find_appropriate_src(struct net *net, u16 zone,
 187                      const struct nf_nat_l3proto *l3proto,
 188                      const struct nf_nat_l4proto *l4proto,
 189                      const struct nf_conntrack_tuple *tuple,
 190                      struct nf_conntrack_tuple *result,
 191                      const struct nf_nat_range *range)
 192 {
 193         unsigned int h = hash_by_src(net, zone, tuple);
 194         const struct nf_conn_nat *nat;
 195         const struct nf_conn *ct;
 196
 197         hlist_for_each_entry_rcu(nat, &net->ct.nat_bysource[h], bysource) {
 198                 ct = nat->ct;
 199                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
 200                         /* Copy source part from reply tuple. */
 201                         nf_ct_invert_tuplepr(result,
 202                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 203                         result->dst = tuple->dst;
 204
 205                         if (in_range(l3proto, l4proto, result, range))
 206                                 return 1;
 207                 }
 208         }
 209         return 0;
 210 }
 211
 212 /* For [FUTURE] fragmentation handling, we want the least-used
 213  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 214  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 215  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 216  * the ip with the lowest src-ip/dst-ip/proto usage.
 217  */
 218 static void
 219 find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple,
 220                     const struct nf_nat_range *range,
 221                     const struct nf_conn *ct,
 222                     enum nf_nat_manip_type maniptype)
 223 {
 224         union nf_inet_addr *var_ipp;
 225         unsigned int i, max;
 226         /* Host order */
 227         u32 minip, maxip, j, dist;
 228         bool full_range;
 229
 230         /* No IP mapping?  Do nothing. */
 231         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 232                 return;
 233
 234         if (maniptype == NF_NAT_MANIP_SRC)
 235                 var_ipp = &tuple->src.u3;
 236         else
 237                 var_ipp = &tuple->dst.u3;
 238
 239         /* Fast path: only one choice. */
 240         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 241                 *var_ipp = range->min_addr;
 242                 return;
 243         }
 244
 245         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 246                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 247         else
 248                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 249
 250         /* Hashing source and destination IPs gives a fairly even
 251          * spread in practice (if there are a small number of IPs
 252          * involved, there usually aren't that many connections
 253          * anyway).  The consistency means that servers see the same
 254          * client coming from the same IP (some Internet Banking sites
 255          * like this), even across reboots.
 256          */
 257         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 258                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 259                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone);
 260
 261         full_range = false;
 262         for (i = 0; i <= max; i++) {
 263                 /* If first bytes of the address are at the maximum, use the
 264                  * distance. Otherwise use the full range.
 265                  */
 266                 if (!full_range) {
 267                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 268                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 269                         dist  = maxip - minip + 1;
 270                 } else {
 271                         minip = 0;
 272                         dist  = ~0;
 273                 }
 274
 275                 var_ipp->all[i] = (__force __u32)
 276                         htonl(minip + (((u64)j * dist) >> 32));
 277                 if (var_ipp->all[i] != range->max_addr.all[i])
 278                         full_range = true;
 279
 280                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 281                         j ^= (__force u32)tuple->dst.u3.all[i];
 282         }
 283 }
 284
 285 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 286  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 287  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 288  * range. It might not be possible to get a unique tuple, but we try.
 289  * At worst (or if we race), we will end up with a final duplicate in
 290  * __ip_conntrack_confirm and drop the packet. */
 291 static void
 292 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 293                  const struct nf_conntrack_tuple *orig_tuple,
 294                  const struct nf_nat_range *range,
 295                  struct nf_conn *ct,
 296                  enum nf_nat_manip_type maniptype)
 297 {
 298         const struct nf_nat_l3proto *l3proto;
 299         const struct nf_nat_l4proto *l4proto;
 300         struct net *net = nf_ct_net(ct);
 301         u16 zone = nf_ct_zone(ct);
 302
 303         rcu_read_lock();
 304         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 305         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 306                                         orig_tuple->dst.protonum);
 307
 308         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 309          * and that same mapping gives a unique tuple within the given
 310          * range, use that.
 311          *
 312          * This is only required for source (ie. NAT/masq) mappings.
 313          * So far, we don't do local source mappings, so multiple
 314          * manips not an issue.
 315          */
 316         if (maniptype == NF_NAT_MANIP_SRC &&
 317             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 318                 /* try the original tuple first */
 319                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 320                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 321                                 *tuple = *orig_tuple;
 322                                 goto out;
 323                         }
 324                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 325                                                 orig_tuple, tuple, range)) {
 326                         pr_debug("get_unique_tuple: Found current src map\n");
 327                         if (!nf_nat_used_tuple(tuple, ct))
 328                                 goto out;
 329                 }
 330         }
 331
 332         /* 2) Select the least-used IP/proto combination in the given range */
 333         *tuple = *orig_tuple;
 334         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 335
 336         /* 3) The per-protocol part of the manip is made to map into
 337          * the range to make a unique tuple.
 338          */
 339
 340         /* Only bother mapping if it's not already in range and unique */
 341         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM)) {
 342                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 343                         if (l4proto->in_range(tuple, maniptype,
 344                                               &range->min_proto,
 345                                               &range->max_proto) &&
 346                             (range->min_proto.all == range->max_proto.all ||
 347                              !nf_nat_used_tuple(tuple, ct)))
 348                                 goto out;
 349                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 350                         goto out;
 351                 }
 352         }
 353
 354         /* Last change: get protocol to try to obtain unique tuple. */
 355         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 356 out:
 357         rcu_read_unlock();
 358 }
 359
 360 unsigned int
 361 nf_nat_setup_info(struct nf_conn *ct,
 362                   const struct nf_nat_range *range,
 363                   enum nf_nat_manip_type maniptype)
 364 {
 365         struct net *net = nf_ct_net(ct);
 366         struct nf_conntrack_tuple curr_tuple, new_tuple;
 367         struct nf_conn_nat *nat;
 368
 369         /* nat helper or nfctnetlink also setup binding */
 370         nat = nfct_nat(ct);
 371         if (!nat) {
 372                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 373                 if (nat == NULL) {
 374                         pr_debug("failed to add NAT extension\n");
 375                         return NF_ACCEPT;
 376                 }
 377         }
 378
 379         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 380                      maniptype == NF_NAT_MANIP_DST);
 381         BUG_ON(nf_nat_initialized(ct, maniptype));
 382
 383         /* What we've got will look like inverse of reply. Normally
 384          * this is what is in the conntrack, except for prior
 385          * manipulations (future optimization: if num_manips == 0,
 386          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 387          */
 388         nf_ct_invert_tuplepr(&curr_tuple,
 389                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 390
 391         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 392
 393         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 394                 struct nf_conntrack_tuple reply;
 395
 396                 /* Alter conntrack table so will recognize replies. */
 397                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 398                 nf_conntrack_alter_reply(ct, &reply);
 399
 400                 /* Non-atomic: we own this at the moment. */
 401                 if (maniptype == NF_NAT_MANIP_SRC)
 402                         ct->status |= IPS_SRC_NAT;
 403                 else
 404                         ct->status |= IPS_DST_NAT;
 405         }
 406
 407         if (maniptype == NF_NAT_MANIP_SRC) {
 408                 unsigned int srchash;
 409
 410                 srchash = hash_by_src(net, nf_ct_zone(ct),
 411                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 412                 spin_lock_bh(&nf_nat_lock);
 413                 /* nf_conntrack_alter_reply might re-allocate extension aera */
 414                 nat = nfct_nat(ct);
 415                 nat->ct = ct;
 416                 hlist_add_head_rcu(&nat->bysource,
 417                                    &net->ct.nat_bysource[srchash]);
 418                 spin_unlock_bh(&nf_nat_lock);
 419         }
 420
 421         /* It's done. */
 422         if (maniptype == NF_NAT_MANIP_DST)
 423                 ct->status |= IPS_DST_NAT_DONE;
 424         else
 425                 ct->status |= IPS_SRC_NAT_DONE;
 426
 427         return NF_ACCEPT;
 428 }
 429 EXPORT_SYMBOL(nf_nat_setup_info);
 430
 431 /* Do packet manipulations according to nf_nat_setup_info. */
 432 unsigned int nf_nat_packet(struct nf_conn *ct,
 433                            enum ip_conntrack_info ctinfo,
 434                            unsigned int hooknum,
 435                            struct sk_buff *skb)
 436 {
 437         const struct nf_nat_l3proto *l3proto;
 438         const struct nf_nat_l4proto *l4proto;
 439         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 440         unsigned long statusbit;
 441         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 442
 443         if (mtype == NF_NAT_MANIP_SRC)
 444                 statusbit = IPS_SRC_NAT;
 445         else
 446                 statusbit = IPS_DST_NAT;
 447
 448         /* Invert if this is reply dir. */
 449         if (dir == IP_CT_DIR_REPLY)
 450                 statusbit ^= IPS_NAT_MASK;
 451
 452         /* Non-atomic: these bits don't change. */
 453         if (ct->status & statusbit) {
 454                 struct nf_conntrack_tuple target;
 455
 456                 /* We are aiming to look like inverse of other direction. */
 457                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 458
 459                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 460                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 461                                                 target.dst.protonum);
 462                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 463                         return NF_DROP;
 464         }
 465         return NF_ACCEPT;
 466 }
 467 EXPORT_SYMBOL_GPL(nf_nat_packet);
 468
 469 struct nf_nat_proto_clean {
 470         u8      l3proto;
 471         u8      l4proto;
 472 };
 473
 474 /* kill conntracks with affected NAT section */
 475 static int nf_nat_proto_remove(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
 483         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 484             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 485                 return 0;
 486
 487         return i->status & IPS_NAT_MASK ? 1 : 0;
 488 }
 489
 490 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 491 {
 492         struct nf_conn_nat *nat = nfct_nat(ct);
 493
 494         if (nf_nat_proto_remove(ct, data))
 495                 return 1;
 496
 497         if (!nat || !nat->ct)
 498                 return 0;
 499
 500         /* This netns is being destroyed, and conntrack has nat null binding.
 501          * Remove it from bysource hash, as the table will be freed soon.
 502          *
 503          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 504          * will delete entry from already-freed table.
 505          */
 506         if (!del_timer(&ct->timeout))
 507                 return 1;
 508
 509         spin_lock_bh(&nf_nat_lock);
 510         hlist_del_rcu(&nat->bysource);
 511         ct->status &= ~IPS_NAT_DONE_MASK;
 512         nat->ct = NULL;
 513         spin_unlock_bh(&nf_nat_lock);
 514
 515         add_timer(&ct->timeout);
 516
 517         /* don't delete conntrack.  Although that would make things a lot
 518          * simpler, we'd end up flushing all conntracks on nat rmmod.
 519          */
 520         return 0;
 521 }
 522
 523 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 524 {
 525         struct nf_nat_proto_clean clean = {
 526                 .l3proto = l3proto,
 527                 .l4proto = l4proto,
 528         };
 529         struct net *net;
 530
 531         rtnl_lock();
 532         for_each_net(net)
 533                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean);
 534         rtnl_unlock();
 535 }
 536
 537 static void nf_nat_l3proto_clean(u8 l3proto)
 538 {
 539         struct nf_nat_proto_clean clean = {
 540                 .l3proto = l3proto,
 541         };
 542         struct net *net;
 543
 544         rtnl_lock();
 545
 546         for_each_net(net)
 547                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean);
 548         rtnl_unlock();
 549 }
 550
 551 /* Protocol registration. */
 552 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 553 {
 554         const struct nf_nat_l4proto **l4protos;
 555         unsigned int i;
 556         int ret = 0;
 557
 558         mutex_lock(&nf_nat_proto_mutex);
 559         if (nf_nat_l4protos[l3proto] == NULL) {
 560                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 561                                    GFP_KERNEL);
 562                 if (l4protos == NULL) {
 563                         ret = -ENOMEM;
 564                         goto out;
 565                 }
 566
 567                 for (i = 0; i < IPPROTO_MAX; i++)
 568                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 569
 570                 /* Before making proto_array visible to lockless readers,
 571                  * we must make sure its content is committed to memory.
 572                  */
 573                 smp_wmb();
 574
 575                 nf_nat_l4protos[l3proto] = l4protos;
 576         }
 577
 578         if (rcu_dereference_protected(
 579                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 580                         lockdep_is_held(&nf_nat_proto_mutex)
 581                         ) != &nf_nat_l4proto_unknown) {
 582                 ret = -EBUSY;
 583                 goto out;
 584         }
 585         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 586  out:
 587         mutex_unlock(&nf_nat_proto_mutex);
 588         return ret;
 589 }
 590 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 591
 592 /* No one stores the protocol anywhere; simply delete it. */
 593 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 594 {
 595         mutex_lock(&nf_nat_proto_mutex);
 596         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 597                          &nf_nat_l4proto_unknown);
 598         mutex_unlock(&nf_nat_proto_mutex);
 599         synchronize_rcu();
 600
 601         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 602 }
 603 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 604
 605 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 606 {
 607         int err;
 608
 609         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 610         if (err < 0)
 611                 return err;
 612
 613         mutex_lock(&nf_nat_proto_mutex);
 614         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 615                          &nf_nat_l4proto_tcp);
 616         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 617                          &nf_nat_l4proto_udp);
 618         mutex_unlock(&nf_nat_proto_mutex);
 619
 620         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 621         return 0;
 622 }
 623 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 624
 625 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 626 {
 627         mutex_lock(&nf_nat_proto_mutex);
 628         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 629         mutex_unlock(&nf_nat_proto_mutex);
 630         synchronize_rcu();
 631
 632         nf_nat_l3proto_clean(l3proto->l3proto);
 633         nf_ct_l3proto_module_put(l3proto->l3proto);
 634 }
 635 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 636
 637 /* No one using conntrack by the time this called. */
 638 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 639 {
 640         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 641
 642         if (nat == NULL || nat->ct == NULL)
 643                 return;
 644
 645         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 646
 647         spin_lock_bh(&nf_nat_lock);
 648         hlist_del_rcu(&nat->bysource);
 649         spin_unlock_bh(&nf_nat_lock);
 650 }
 651
 652 static void nf_nat_move_storage(void *new, void *old)
 653 {
 654         struct nf_conn_nat *new_nat = new;
 655         struct nf_conn_nat *old_nat = old;
 656         struct nf_conn *ct = old_nat->ct;
 657
 658         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 659                 return;
 660
 661         spin_lock_bh(&nf_nat_lock);
 662         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 663         spin_unlock_bh(&nf_nat_lock);
 664 }
 665
 666 static struct nf_ct_ext_type nat_extend __read_mostly = {
 667         .len            = sizeof(struct nf_conn_nat),
 668         .align          = __alignof__(struct nf_conn_nat),
 669         .destroy        = nf_nat_cleanup_conntrack,
 670         .move           = nf_nat_move_storage,
 671         .id             = NF_CT_EXT_NAT,
 672         .flags          = NF_CT_EXT_F_PREALLOC,
 673 };
 674
 675 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 676
 677 #include <linux/netfilter/nfnetlink.h>
 678 #include <linux/netfilter/nfnetlink_conntrack.h>
 679
 680 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 681         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 682         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 683 };
 684
 685 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 686                                      const struct nf_conn *ct,
 687                                      struct nf_nat_range *range)
 688 {
 689         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 690         const struct nf_nat_l4proto *l4proto;
 691         int err;
 692
 693         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 694         if (err < 0)
 695                 return err;
 696
 697         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 698         if (l4proto->nlattr_to_range)
 699                 err = l4proto->nlattr_to_range(tb, range);
 700
 701         return err;
 702 }
 703
 704 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 705         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 706         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 707         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 708         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 709         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 710 };
 711
 712 static int
 713 nfnetlink_parse_nat(const struct nlattr *nat,
 714                     const struct nf_conn *ct, struct nf_nat_range *range)
 715 {
 716         const struct nf_nat_l3proto *l3proto;
 717         struct nlattr *tb[CTA_NAT_MAX+1];
 718         int err;
 719
 720         memset(range, 0, sizeof(*range));
 721
 722         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 723         if (err < 0)
 724                 return err;
 725
 726         rcu_read_lock();
 727         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 728         if (l3proto == NULL) {
 729                 err = -EAGAIN;
 730                 goto out;
 731         }
 732         err = l3proto->nlattr_to_range(tb, range);
 733         if (err < 0)
 734                 goto out;
 735
 736         if (!tb[CTA_NAT_PROTO])
 737                 goto out;
 738
 739         err = nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 740 out:
 741         rcu_read_unlock();
 742         return err;
 743 }
 744
 745 static int
 746 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 747                           enum nf_nat_manip_type manip,
 748                           const struct nlattr *attr)
 749 {
 750         struct nf_nat_range range;
 751         int err;
 752
 753         err = nfnetlink_parse_nat(attr, ct, &range);
 754         if (err < 0)
 755                 return err;
 756         if (nf_nat_initialized(ct, manip))
 757                 return -EEXIST;
 758
 759         return nf_nat_setup_info(ct, &range, manip);
 760 }
 761 #else
 762 static int
 763 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 764                           enum nf_nat_manip_type manip,
 765                           const struct nlattr *attr)
 766 {
 767         return -EOPNOTSUPP;
 768 }
 769 #endif
 770
 771 static int __net_init nf_nat_net_init(struct net *net)
 772 {
 773         /* Leave them the same for the moment. */
 774         net->ct.nat_htable_size = net->ct.htable_size;
 775         net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0);
 776         if (!net->ct.nat_bysource)
 777                 return -ENOMEM;
 778         return 0;
 779 }
 780
 781 static void __net_exit nf_nat_net_exit(struct net *net)
 782 {
 783         struct nf_nat_proto_clean clean = {};
 784
 785         nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean);
 786         synchronize_rcu();
 787         nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size);
 788 }
 789
 790 static struct pernet_operations nf_nat_net_ops = {
 791         .init = nf_nat_net_init,
 792         .exit = nf_nat_net_exit,
 793 };
 794
 795 static struct nf_ct_helper_expectfn follow_master_nat = {
 796         .name           = "nat-follow-master",
 797         .expectfn       = nf_nat_follow_master,
 798 };
 799
 800 static struct nfq_ct_nat_hook nfq_ct_nat = {
 801         .seq_adjust     = nf_nat_tcp_seq_adjust,
 802 };
 803
 804 static int __init nf_nat_init(void)
 805 {
 806         int ret;
 807
 808         ret = nf_ct_extend_register(&nat_extend);
 809         if (ret < 0) {
 810                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 811                 return ret;
 812         }
 813
 814         ret = register_pernet_subsys(&nf_nat_net_ops);
 815         if (ret < 0)
 816                 goto cleanup_extend;
 817
 818         nf_ct_helper_expectfn_register(&follow_master_nat);
 819
 820         /* Initialize fake conntrack so that NAT will skip it */
 821         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 822
 823         BUG_ON(nf_nat_seq_adjust_hook != NULL);
 824         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, nf_nat_seq_adjust);
 825         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 826         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 827                            nfnetlink_parse_nat_setup);
 828         BUG_ON(nf_ct_nat_offset != NULL);
 829         RCU_INIT_POINTER(nf_ct_nat_offset, nf_nat_get_offset);
 830         RCU_INIT_POINTER(nfq_ct_nat_hook, &nfq_ct_nat);
 831 #ifdef CONFIG_XFRM
 832         BUG_ON(nf_nat_decode_session_hook != NULL);
 833         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 834 #endif
 835         return 0;
 836
 837  cleanup_extend:
 838         nf_ct_extend_unregister(&nat_extend);
 839         return ret;
 840 }
 841
 842 static void __exit nf_nat_cleanup(void)
 843 {
 844         unsigned int i;
 845
 846         unregister_pernet_subsys(&nf_nat_net_ops);
 847         nf_ct_extend_unregister(&nat_extend);
 848         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 849         RCU_INIT_POINTER(nf_nat_seq_adjust_hook, NULL);
 850         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 851         RCU_INIT_POINTER(nf_ct_nat_offset, NULL);
 852         RCU_INIT_POINTER(nfq_ct_nat_hook, NULL);
 853 #ifdef CONFIG_XFRM
 854         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 855 #endif
 856         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 857                 kfree(nf_nat_l4protos[i]);
 858         synchronize_net();
 859 }
 860
 861 MODULE_LICENSE("GPL");
 862
 863 module_init(nf_nat_init);
 864 module_exit(nf_nat_cleanup);