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_seqadj.h>
  29 #include <net/netfilter/nf_conntrack_l3proto.h>
  30 #include <net/netfilter/nf_conntrack_zones.h>
  31 #include <linux/netfilter/nf_nat.h>
  32
  33 static DEFINE_SPINLOCK(nf_nat_lock);
  34
  35 static DEFINE_MUTEX(nf_nat_proto_mutex);
  36 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
  37                                                 __read_mostly;
  38 static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO]
  39                                                 __read_mostly;
  40
  41
  42 inline const struct nf_nat_l3proto *
  43 __nf_nat_l3proto_find(u8 family)
  44 {
  45         return rcu_dereference(nf_nat_l3protos[family]);
  46 }
  47
  48 inline const struct nf_nat_l4proto *
  49 __nf_nat_l4proto_find(u8 family, u8 protonum)
  50 {
  51         return rcu_dereference(nf_nat_l4protos[family][protonum]);
  52 }
  53 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
  54
  55 #ifdef CONFIG_XFRM
  56 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  57 {
  58         const struct nf_nat_l3proto *l3proto;
  59         const struct nf_conn *ct;
  60         enum ip_conntrack_info ctinfo;
  61         enum ip_conntrack_dir dir;
  62         unsigned  long statusbit;
  63         u8 family;
  64
  65         ct = nf_ct_get(skb, &ctinfo);
  66         if (ct == NULL)
  67                 return;
  68
  69         family = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.l3num;
  70         rcu_read_lock();
  71         l3proto = __nf_nat_l3proto_find(family);
  72         if (l3proto == NULL)
  73                 goto out;
  74
  75         dir = CTINFO2DIR(ctinfo);
  76         if (dir == IP_CT_DIR_ORIGINAL)
  77                 statusbit = IPS_DST_NAT;
  78         else
  79                 statusbit = IPS_SRC_NAT;
  80
  81         l3proto->decode_session(skb, ct, dir, statusbit, fl);
  82 out:
  83         rcu_read_unlock();
  84 }
  85
  86 int nf_xfrm_me_harder(struct sk_buff *skb, unsigned int family)
  87 {
  88         struct flowi fl;
  89         unsigned int hh_len;
  90         struct dst_entry *dst;
  91         int err;
  92
  93         err = xfrm_decode_session(skb, &fl, family);
  94         if (err < 0)
  95                 return err;
  96
  97         dst = skb_dst(skb);
  98         if (dst->xfrm)
  99                 dst = ((struct xfrm_dst *)dst)->route;
 100         dst_hold(dst);
 101
 102         dst = xfrm_lookup(dev_net(dst->dev), dst, &fl, skb->sk, 0);
 103         if (IS_ERR(dst))
 104                 return PTR_ERR(dst);
 105
 106         skb_dst_drop(skb);
 107         skb_dst_set(skb, dst);
 108
 109         /* Change in oif may mean change in hh_len. */
 110         hh_len = skb_dst(skb)->dev->hard_header_len;
 111         if (skb_headroom(skb) < hh_len &&
 112             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 113                 return -ENOMEM;
 114         return 0;
 115 }
 116 EXPORT_SYMBOL(nf_xfrm_me_harder);
 117 #endif /* CONFIG_XFRM */
 118
 119 /* We keep an extra hash for each conntrack, for fast searching. */
 120 static inline unsigned int
 121 hash_by_src(const struct net *net, u16 zone,
 122             const struct nf_conntrack_tuple *tuple)
 123 {
 124         unsigned int hash;
 125
 126         /* Original src, to ensure we map it consistently if poss. */
 127         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 128                       tuple->dst.protonum ^ zone ^ nf_conntrack_hash_rnd);
 129         return ((u64)hash * net->ct.nat_htable_size) >> 32;
 130 }
 131
 132 /* Is this tuple already taken? (not by us) */
 133 int
 134 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 135                   const struct nf_conn *ignored_conntrack)
 136 {
 137         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 138          * incoming ones.  NAT means they don't have a fixed mapping,
 139          * so we invert the tuple and look for the incoming reply.
 140          *
 141          * We could keep a separate hash if this proves too slow.
 142          */
 143         struct nf_conntrack_tuple reply;
 144
 145         nf_ct_invert_tuplepr(&reply, tuple);
 146         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 147 }
 148 EXPORT_SYMBOL(nf_nat_used_tuple);
 149
 150 /* If we source map this tuple so reply looks like reply_tuple, will
 151  * that meet the constraints of range.
 152  */
 153 static int in_range(const struct nf_nat_l3proto *l3proto,
 154                     const struct nf_nat_l4proto *l4proto,
 155                     const struct nf_conntrack_tuple *tuple,
 156                     const struct nf_nat_range *range)
 157 {
 158         /* If we are supposed to map IPs, then we must be in the
 159          * range specified, otherwise let this drag us onto a new src IP.
 160          */
 161         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 162             !l3proto->in_range(tuple, range))
 163                 return 0;
 164
 165         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 166             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 167                               &range->min_proto, &range->max_proto))
 168                 return 1;
 169
 170         return 0;
 171 }
 172
 173 static inline int
 174 same_src(const struct nf_conn *ct,
 175          const struct nf_conntrack_tuple *tuple)
 176 {
 177         const struct nf_conntrack_tuple *t;
 178
 179         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 180         return (t->dst.protonum == tuple->dst.protonum &&
 181                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 182                 t->src.u.all == tuple->src.u.all);
 183 }
 184
 185 /* Only called for SRC manip */
 186 static int
 187 find_appropriate_src(struct net *net, u16 zone,
 188                      const struct nf_nat_l3proto *l3proto,
 189                      const struct nf_nat_l4proto *l4proto,
 190                      const struct nf_conntrack_tuple *tuple,
 191                      struct nf_conntrack_tuple *result,
 192                      const struct nf_nat_range *range)
 193 {
 194         unsigned int h = hash_by_src(net, zone, tuple);
 195         const struct nf_conn_nat *nat;
 196         const struct nf_conn *ct;
 197
 198         hlist_for_each_entry_rcu(nat, &net->ct.nat_bysource[h], bysource) {
 199                 ct = nat->ct;
 200                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
 201                         /* Copy source part from reply tuple. */
 202                         nf_ct_invert_tuplepr(result,
 203                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 204                         result->dst = tuple->dst;
 205
 206                         if (in_range(l3proto, l4proto, result, range))
 207                                 return 1;
 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_ALL)) {
 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_ALL)) {
 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                 if (nfct_help(ct))
 408                         nfct_seqadj_ext_add(ct);
 409         }
 410
 411         if (maniptype == NF_NAT_MANIP_SRC) {
 412                 unsigned int srchash;
 413
 414                 srchash = hash_by_src(net, nf_ct_zone(ct),
 415                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 416                 spin_lock_bh(&nf_nat_lock);
 417                 /* nf_conntrack_alter_reply might re-allocate extension aera */
 418                 nat = nfct_nat(ct);
 419                 nat->ct = ct;
 420                 hlist_add_head_rcu(&nat->bysource,
 421                                    &net->ct.nat_bysource[srchash]);
 422                 spin_unlock_bh(&nf_nat_lock);
 423         }
 424
 425         /* It's done. */
 426         if (maniptype == NF_NAT_MANIP_DST)
 427                 ct->status |= IPS_DST_NAT_DONE;
 428         else
 429                 ct->status |= IPS_SRC_NAT_DONE;
 430
 431         return NF_ACCEPT;
 432 }
 433 EXPORT_SYMBOL(nf_nat_setup_info);
 434
 435 static unsigned int
 436 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 437 {
 438         /* Force range to this IP; let proto decide mapping for
 439          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 440          * Use reply in case it's already been mangled (eg local packet).
 441          */
 442         union nf_inet_addr ip =
 443                 (manip == NF_NAT_MANIP_SRC ?
 444                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 445                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 446         struct nf_nat_range range = {
 447                 .flags          = NF_NAT_RANGE_MAP_IPS,
 448                 .min_addr       = ip,
 449                 .max_addr       = ip,
 450         };
 451         return nf_nat_setup_info(ct, &range, manip);
 452 }
 453
 454 unsigned int
 455 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 456 {
 457         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 458 }
 459 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 460
 461 /* Do packet manipulations according to nf_nat_setup_info. */
 462 unsigned int nf_nat_packet(struct nf_conn *ct,
 463                            enum ip_conntrack_info ctinfo,
 464                            unsigned int hooknum,
 465                            struct sk_buff *skb)
 466 {
 467         const struct nf_nat_l3proto *l3proto;
 468         const struct nf_nat_l4proto *l4proto;
 469         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 470         unsigned long statusbit;
 471         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 472
 473         if (mtype == NF_NAT_MANIP_SRC)
 474                 statusbit = IPS_SRC_NAT;
 475         else
 476                 statusbit = IPS_DST_NAT;
 477
 478         /* Invert if this is reply dir. */
 479         if (dir == IP_CT_DIR_REPLY)
 480                 statusbit ^= IPS_NAT_MASK;
 481
 482         /* Non-atomic: these bits don't change. */
 483         if (ct->status & statusbit) {
 484                 struct nf_conntrack_tuple target;
 485
 486                 /* We are aiming to look like inverse of other direction. */
 487                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 488
 489                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 490                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 491                                                 target.dst.protonum);
 492                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 493                         return NF_DROP;
 494         }
 495         return NF_ACCEPT;
 496 }
 497 EXPORT_SYMBOL_GPL(nf_nat_packet);
 498
 499 struct nf_nat_proto_clean {
 500         u8      l3proto;
 501         u8      l4proto;
 502 };
 503
 504 /* kill conntracks with affected NAT section */
 505 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 506 {
 507         const struct nf_nat_proto_clean *clean = data;
 508         struct nf_conn_nat *nat = nfct_nat(i);
 509
 510         if (!nat)
 511                 return 0;
 512
 513         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 514             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 515                 return 0;
 516
 517         return i->status & IPS_NAT_MASK ? 1 : 0;
 518 }
 519
 520 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 521 {
 522         struct nf_nat_proto_clean clean = {
 523                 .l3proto = l3proto,
 524                 .l4proto = l4proto,
 525         };
 526         struct net *net;
 527
 528         rtnl_lock();
 529         for_each_net(net)
 530                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 531         rtnl_unlock();
 532 }
 533
 534 static void nf_nat_l3proto_clean(u8 l3proto)
 535 {
 536         struct nf_nat_proto_clean clean = {
 537                 .l3proto = l3proto,
 538         };
 539         struct net *net;
 540
 541         rtnl_lock();
 542
 543         for_each_net(net)
 544                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 545         rtnl_unlock();
 546 }
 547
 548 /* Protocol registration. */
 549 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 550 {
 551         const struct nf_nat_l4proto **l4protos;
 552         unsigned int i;
 553         int ret = 0;
 554
 555         mutex_lock(&nf_nat_proto_mutex);
 556         if (nf_nat_l4protos[l3proto] == NULL) {
 557                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 558                                    GFP_KERNEL);
 559                 if (l4protos == NULL) {
 560                         ret = -ENOMEM;
 561                         goto out;
 562                 }
 563
 564                 for (i = 0; i < IPPROTO_MAX; i++)
 565                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 566
 567                 /* Before making proto_array visible to lockless readers,
 568                  * we must make sure its content is committed to memory.
 569                  */
 570                 smp_wmb();
 571
 572                 nf_nat_l4protos[l3proto] = l4protos;
 573         }
 574
 575         if (rcu_dereference_protected(
 576                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 577                         lockdep_is_held(&nf_nat_proto_mutex)
 578                         ) != &nf_nat_l4proto_unknown) {
 579                 ret = -EBUSY;
 580                 goto out;
 581         }
 582         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 583  out:
 584         mutex_unlock(&nf_nat_proto_mutex);
 585         return ret;
 586 }
 587 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 588
 589 /* No one stores the protocol anywhere; simply delete it. */
 590 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 591 {
 592         mutex_lock(&nf_nat_proto_mutex);
 593         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 594                          &nf_nat_l4proto_unknown);
 595         mutex_unlock(&nf_nat_proto_mutex);
 596         synchronize_rcu();
 597
 598         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 599 }
 600 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 601
 602 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 603 {
 604         int err;
 605
 606         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 607         if (err < 0)
 608                 return err;
 609
 610         mutex_lock(&nf_nat_proto_mutex);
 611         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 612                          &nf_nat_l4proto_tcp);
 613         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 614                          &nf_nat_l4proto_udp);
 615         mutex_unlock(&nf_nat_proto_mutex);
 616
 617         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 618         return 0;
 619 }
 620 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 621
 622 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 623 {
 624         mutex_lock(&nf_nat_proto_mutex);
 625         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 626         mutex_unlock(&nf_nat_proto_mutex);
 627         synchronize_rcu();
 628
 629         nf_nat_l3proto_clean(l3proto->l3proto);
 630         nf_ct_l3proto_module_put(l3proto->l3proto);
 631 }
 632 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 633
 634 /* No one using conntrack by the time this called. */
 635 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 636 {
 637         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 638
 639         if (nat == NULL || nat->ct == NULL)
 640                 return;
 641
 642         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 643
 644         spin_lock_bh(&nf_nat_lock);
 645         hlist_del_rcu(&nat->bysource);
 646         spin_unlock_bh(&nf_nat_lock);
 647 }
 648
 649 static void nf_nat_move_storage(void *new, void *old)
 650 {
 651         struct nf_conn_nat *new_nat = new;
 652         struct nf_conn_nat *old_nat = old;
 653         struct nf_conn *ct = old_nat->ct;
 654
 655         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 656                 return;
 657
 658         spin_lock_bh(&nf_nat_lock);
 659         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 660         spin_unlock_bh(&nf_nat_lock);
 661 }
 662
 663 static struct nf_ct_ext_type nat_extend __read_mostly = {
 664         .len            = sizeof(struct nf_conn_nat),
 665         .align          = __alignof__(struct nf_conn_nat),
 666         .destroy        = nf_nat_cleanup_conntrack,
 667         .move           = nf_nat_move_storage,
 668         .id             = NF_CT_EXT_NAT,
 669         .flags          = NF_CT_EXT_F_PREALLOC,
 670 };
 671
 672 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 673
 674 #include <linux/netfilter/nfnetlink.h>
 675 #include <linux/netfilter/nfnetlink_conntrack.h>
 676
 677 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 678         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 679         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 680 };
 681
 682 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 683                                      const struct nf_conn *ct,
 684                                      struct nf_nat_range *range)
 685 {
 686         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 687         const struct nf_nat_l4proto *l4proto;
 688         int err;
 689
 690         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 691         if (err < 0)
 692                 return err;
 693
 694         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 695         if (l4proto->nlattr_to_range)
 696                 err = l4proto->nlattr_to_range(tb, range);
 697
 698         return err;
 699 }
 700
 701 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 702         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 703         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 704         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 705         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 706         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 707 };
 708
 709 static int
 710 nfnetlink_parse_nat(const struct nlattr *nat,
 711                     const struct nf_conn *ct, struct nf_nat_range *range,
 712                     const struct nf_nat_l3proto *l3proto)
 713 {
 714         struct nlattr *tb[CTA_NAT_MAX+1];
 715         int err;
 716
 717         memset(range, 0, sizeof(*range));
 718
 719         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 720         if (err < 0)
 721                 return err;
 722
 723         err = l3proto->nlattr_to_range(tb, range);
 724         if (err < 0)
 725                 return err;
 726
 727         if (!tb[CTA_NAT_PROTO])
 728                 return 0;
 729
 730         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 731 }
 732
 733 /* This function is called under rcu_read_lock() */
 734 static int
 735 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 736                           enum nf_nat_manip_type manip,
 737                           const struct nlattr *attr)
 738 {
 739         struct nf_nat_range range;
 740         const struct nf_nat_l3proto *l3proto;
 741         int err;
 742
 743         /* Should not happen, restricted to creating new conntracks
 744          * via ctnetlink.
 745          */
 746         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 747                 return -EEXIST;
 748
 749         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 750          * attach the null binding, otherwise this may oops.
 751          */
 752         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 753         if (l3proto == NULL)
 754                 return -EAGAIN;
 755
 756         /* No NAT information has been passed, allocate the null-binding */
 757         if (attr == NULL)
 758                 return __nf_nat_alloc_null_binding(ct, manip);
 759
 760         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 761         if (err < 0)
 762                 return err;
 763
 764         return nf_nat_setup_info(ct, &range, manip);
 765 }
 766 #else
 767 static int
 768 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 769                           enum nf_nat_manip_type manip,
 770                           const struct nlattr *attr)
 771 {
 772         return -EOPNOTSUPP;
 773 }
 774 #endif
 775
 776 static int __net_init nf_nat_net_init(struct net *net)
 777 {
 778         /* Leave them the same for the moment. */
 779         net->ct.nat_htable_size = net->ct.htable_size;
 780         net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0);
 781         if (!net->ct.nat_bysource)
 782                 return -ENOMEM;
 783         return 0;
 784 }
 785
 786 static void __net_exit nf_nat_net_exit(struct net *net)
 787 {
 788         struct nf_nat_proto_clean clean = {};
 789
 790         nf_ct_iterate_cleanup(net, &nf_nat_proto_remove, &clean, 0, 0);
 791         synchronize_rcu();
 792         nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size);
 793 }
 794
 795 static struct pernet_operations nf_nat_net_ops = {
 796         .init = nf_nat_net_init,
 797         .exit = nf_nat_net_exit,
 798 };
 799
 800 static struct nf_ct_helper_expectfn follow_master_nat = {
 801         .name           = "nat-follow-master",
 802         .expectfn       = nf_nat_follow_master,
 803 };
 804
 805 static int __init nf_nat_init(void)
 806 {
 807         int ret;
 808
 809         ret = nf_ct_extend_register(&nat_extend);
 810         if (ret < 0) {
 811                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 812                 return ret;
 813         }
 814
 815         ret = register_pernet_subsys(&nf_nat_net_ops);
 816         if (ret < 0)
 817                 goto cleanup_extend;
 818
 819         nf_ct_helper_expectfn_register(&follow_master_nat);
 820
 821         /* Initialize fake conntrack so that NAT will skip it */
 822         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 823
 824         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 825         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 826                            nfnetlink_parse_nat_setup);
 827 #ifdef CONFIG_XFRM
 828         BUG_ON(nf_nat_decode_session_hook != NULL);
 829         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 830 #endif
 831         return 0;
 832
 833  cleanup_extend:
 834         nf_ct_extend_unregister(&nat_extend);
 835         return ret;
 836 }
 837
 838 static void __exit nf_nat_cleanup(void)
 839 {
 840         unsigned int i;
 841
 842         unregister_pernet_subsys(&nf_nat_net_ops);
 843         nf_ct_extend_unregister(&nat_extend);
 844         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 845         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 846 #ifdef CONFIG_XFRM
 847         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 848 #endif
 849         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 850                 kfree(nf_nat_l4protos[i]);
 851         synchronize_net();
 852 }
 853
 854 MODULE_LICENSE("GPL");
 855
 856 module_init(nf_nat_init);
 857 module_exit(nf_nat_cleanup);