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 spinlock_t nf_nat_locks[CONNTRACK_LOCKS];
  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 static struct hlist_head *nf_nat_bysource __read_mostly;
  42 static unsigned int nf_nat_htable_size __read_mostly;
  43 static unsigned int nf_nat_hash_rnd __read_mostly;
  44
  45 inline const struct nf_nat_l3proto *
  46 __nf_nat_l3proto_find(u8 family)
  47 {
  48         return rcu_dereference(nf_nat_l3protos[family]);
  49 }
  50
  51 inline const struct nf_nat_l4proto *
  52 __nf_nat_l4proto_find(u8 family, u8 protonum)
  53 {
  54         return rcu_dereference(nf_nat_l4protos[family][protonum]);
  55 }
  56 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
  57
  58 #ifdef CONFIG_XFRM
  59 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  60 {
  61         const struct nf_nat_l3proto *l3proto;
  62         const struct nf_conn *ct;
  63         enum ip_conntrack_info ctinfo;
  64         enum ip_conntrack_dir dir;
  65         unsigned  long statusbit;
  66         u8 family;
  67
  68         ct = nf_ct_get(skb, &ctinfo);
  69         if (ct == NULL)
  70                 return;
  71
  72         family = nf_ct_l3num(ct);
  73         l3proto = __nf_nat_l3proto_find(family);
  74         if (l3proto == NULL)
  75                 return;
  76
  77         dir = CTINFO2DIR(ctinfo);
  78         if (dir == IP_CT_DIR_ORIGINAL)
  79                 statusbit = IPS_DST_NAT;
  80         else
  81                 statusbit = IPS_SRC_NAT;
  82
  83         l3proto->decode_session(skb, ct, dir, statusbit, fl);
  84 }
  85
  86 int nf_xfrm_me_harder(struct net *net, 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(net, 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 unsigned int
 121 hash_by_src(const struct net *n, const struct nf_conntrack_tuple *tuple)
 122 {
 123         unsigned int hash;
 124
 125         get_random_once(&nf_nat_hash_rnd, sizeof(nf_nat_hash_rnd));
 126
 127         /* Original src, to ensure we map it consistently if poss. */
 128         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 129                       tuple->dst.protonum ^ nf_nat_hash_rnd ^ net_hash_mix(n));
 130
 131         return reciprocal_scale(hash, nf_nat_htable_size);
 132 }
 133
 134 /* Is this tuple already taken? (not by us) */
 135 int
 136 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 137                   const struct nf_conn *ignored_conntrack)
 138 {
 139         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 140          * incoming ones.  NAT means they don't have a fixed mapping,
 141          * so we invert the tuple and look for the incoming reply.
 142          *
 143          * We could keep a separate hash if this proves too slow.
 144          */
 145         struct nf_conntrack_tuple reply;
 146
 147         nf_ct_invert_tuplepr(&reply, tuple);
 148         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 149 }
 150 EXPORT_SYMBOL(nf_nat_used_tuple);
 151
 152 /* If we source map this tuple so reply looks like reply_tuple, will
 153  * that meet the constraints of range.
 154  */
 155 static int in_range(const struct nf_nat_l3proto *l3proto,
 156                     const struct nf_nat_l4proto *l4proto,
 157                     const struct nf_conntrack_tuple *tuple,
 158                     const struct nf_nat_range *range)
 159 {
 160         /* If we are supposed to map IPs, then we must be in the
 161          * range specified, otherwise let this drag us onto a new src IP.
 162          */
 163         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 164             !l3proto->in_range(tuple, range))
 165                 return 0;
 166
 167         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 168             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 169                               &range->min_proto, &range->max_proto))
 170                 return 1;
 171
 172         return 0;
 173 }
 174
 175 static inline int
 176 same_src(const struct nf_conn *ct,
 177          const struct nf_conntrack_tuple *tuple)
 178 {
 179         const struct nf_conntrack_tuple *t;
 180
 181         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 182         return (t->dst.protonum == tuple->dst.protonum &&
 183                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 184                 t->src.u.all == tuple->src.u.all);
 185 }
 186
 187 /* Only called for SRC manip */
 188 static int
 189 find_appropriate_src(struct net *net,
 190                      const struct nf_conntrack_zone *zone,
 191                      const struct nf_nat_l3proto *l3proto,
 192                      const struct nf_nat_l4proto *l4proto,
 193                      const struct nf_conntrack_tuple *tuple,
 194                      struct nf_conntrack_tuple *result,
 195                      const struct nf_nat_range *range)
 196 {
 197         unsigned int h = hash_by_src(net, tuple);
 198         const struct nf_conn *ct;
 199
 200         hlist_for_each_entry_rcu(ct, &nf_nat_bysource[h], nat_bysource) {
 201                 if (same_src(ct, tuple) &&
 202                     net_eq(net, nf_ct_net(ct)) &&
 203                     nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) {
 204                         /* Copy source part from reply tuple. */
 205                         nf_ct_invert_tuplepr(result,
 206                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 207                         result->dst = tuple->dst;
 208
 209                         if (in_range(l3proto, l4proto, result, range))
 210                                 return 1;
 211                 }
 212         }
 213         return 0;
 214 }
 215
 216 /* For [FUTURE] fragmentation handling, we want the least-used
 217  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 218  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 219  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 220  * the ip with the lowest src-ip/dst-ip/proto usage.
 221  */
 222 static void
 223 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 224                     struct nf_conntrack_tuple *tuple,
 225                     const struct nf_nat_range *range,
 226                     const struct nf_conn *ct,
 227                     enum nf_nat_manip_type maniptype)
 228 {
 229         union nf_inet_addr *var_ipp;
 230         unsigned int i, max;
 231         /* Host order */
 232         u32 minip, maxip, j, dist;
 233         bool full_range;
 234
 235         /* No IP mapping?  Do nothing. */
 236         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 237                 return;
 238
 239         if (maniptype == NF_NAT_MANIP_SRC)
 240                 var_ipp = &tuple->src.u3;
 241         else
 242                 var_ipp = &tuple->dst.u3;
 243
 244         /* Fast path: only one choice. */
 245         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 246                 *var_ipp = range->min_addr;
 247                 return;
 248         }
 249
 250         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 251                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 252         else
 253                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 254
 255         /* Hashing source and destination IPs gives a fairly even
 256          * spread in practice (if there are a small number of IPs
 257          * involved, there usually aren't that many connections
 258          * anyway).  The consistency means that servers see the same
 259          * client coming from the same IP (some Internet Banking sites
 260          * like this), even across reboots.
 261          */
 262         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 263                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 264                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 265
 266         full_range = false;
 267         for (i = 0; i <= max; i++) {
 268                 /* If first bytes of the address are at the maximum, use the
 269                  * distance. Otherwise use the full range.
 270                  */
 271                 if (!full_range) {
 272                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 273                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 274                         dist  = maxip - minip + 1;
 275                 } else {
 276                         minip = 0;
 277                         dist  = ~0;
 278                 }
 279
 280                 var_ipp->all[i] = (__force __u32)
 281                         htonl(minip + reciprocal_scale(j, dist));
 282                 if (var_ipp->all[i] != range->max_addr.all[i])
 283                         full_range = true;
 284
 285                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 286                         j ^= (__force u32)tuple->dst.u3.all[i];
 287         }
 288 }
 289
 290 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 291  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 292  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 293  * range. It might not be possible to get a unique tuple, but we try.
 294  * At worst (or if we race), we will end up with a final duplicate in
 295  * __ip_conntrack_confirm and drop the packet. */
 296 static void
 297 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 298                  const struct nf_conntrack_tuple *orig_tuple,
 299                  const struct nf_nat_range *range,
 300                  struct nf_conn *ct,
 301                  enum nf_nat_manip_type maniptype)
 302 {
 303         const struct nf_conntrack_zone *zone;
 304         const struct nf_nat_l3proto *l3proto;
 305         const struct nf_nat_l4proto *l4proto;
 306         struct net *net = nf_ct_net(ct);
 307
 308         zone = nf_ct_zone(ct);
 309
 310         rcu_read_lock();
 311         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 312         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 313                                         orig_tuple->dst.protonum);
 314
 315         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 316          * and that same mapping gives a unique tuple within the given
 317          * range, use that.
 318          *
 319          * This is only required for source (ie. NAT/masq) mappings.
 320          * So far, we don't do local source mappings, so multiple
 321          * manips not an issue.
 322          */
 323         if (maniptype == NF_NAT_MANIP_SRC &&
 324             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 325                 /* try the original tuple first */
 326                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 327                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 328                                 *tuple = *orig_tuple;
 329                                 goto out;
 330                         }
 331                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 332                                                 orig_tuple, tuple, range)) {
 333                         pr_debug("get_unique_tuple: Found current src map\n");
 334                         if (!nf_nat_used_tuple(tuple, ct))
 335                                 goto out;
 336                 }
 337         }
 338
 339         /* 2) Select the least-used IP/proto combination in the given range */
 340         *tuple = *orig_tuple;
 341         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 342
 343         /* 3) The per-protocol part of the manip is made to map into
 344          * the range to make a unique tuple.
 345          */
 346
 347         /* Only bother mapping if it's not already in range and unique */
 348         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 349                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 350                         if (l4proto->in_range(tuple, maniptype,
 351                                               &range->min_proto,
 352                                               &range->max_proto) &&
 353                             (range->min_proto.all == range->max_proto.all ||
 354                              !nf_nat_used_tuple(tuple, ct)))
 355                                 goto out;
 356                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 357                         goto out;
 358                 }
 359         }
 360
 361         /* Last change: get protocol to try to obtain unique tuple. */
 362         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 363 out:
 364         rcu_read_unlock();
 365 }
 366
 367 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 368 {
 369         struct nf_conn_nat *nat = nfct_nat(ct);
 370         if (nat)
 371                 return nat;
 372
 373         if (!nf_ct_is_confirmed(ct))
 374                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 375
 376         return nat;
 377 }
 378 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 379
 380 unsigned int
 381 nf_nat_setup_info(struct nf_conn *ct,
 382                   const struct nf_nat_range *range,
 383                   enum nf_nat_manip_type maniptype)
 384 {
 385         struct net *net = nf_ct_net(ct);
 386         struct nf_conntrack_tuple curr_tuple, new_tuple;
 387
 388         /* Can't setup nat info for confirmed ct. */
 389         if (nf_ct_is_confirmed(ct))
 390                 return NF_ACCEPT;
 391
 392         WARN_ON(maniptype != NF_NAT_MANIP_SRC &&
 393                 maniptype != NF_NAT_MANIP_DST);
 394
 395         if (WARN_ON(nf_nat_initialized(ct, maniptype)))
 396                 return NF_DROP;
 397
 398         /* What we've got will look like inverse of reply. Normally
 399          * this is what is in the conntrack, except for prior
 400          * manipulations (future optimization: if num_manips == 0,
 401          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 402          */
 403         nf_ct_invert_tuplepr(&curr_tuple,
 404                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 405
 406         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 407
 408         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 409                 struct nf_conntrack_tuple reply;
 410
 411                 /* Alter conntrack table so will recognize replies. */
 412                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 413                 nf_conntrack_alter_reply(ct, &reply);
 414
 415                 /* Non-atomic: we own this at the moment. */
 416                 if (maniptype == NF_NAT_MANIP_SRC)
 417                         ct->status |= IPS_SRC_NAT;
 418                 else
 419                         ct->status |= IPS_DST_NAT;
 420
 421                 if (nfct_help(ct) && !nfct_seqadj(ct))
 422                         if (!nfct_seqadj_ext_add(ct))
 423                                 return NF_DROP;
 424         }
 425
 426         if (maniptype == NF_NAT_MANIP_SRC) {
 427                 unsigned int srchash;
 428                 spinlock_t *lock;
 429
 430                 srchash = hash_by_src(net,
 431                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 432                 lock = &nf_nat_locks[srchash % ARRAY_SIZE(nf_nat_locks)];
 433                 spin_lock_bh(lock);
 434                 hlist_add_head_rcu(&ct->nat_bysource,
 435                                    &nf_nat_bysource[srchash]);
 436                 spin_unlock_bh(lock);
 437         }
 438
 439         /* It's done. */
 440         if (maniptype == NF_NAT_MANIP_DST)
 441                 ct->status |= IPS_DST_NAT_DONE;
 442         else
 443                 ct->status |= IPS_SRC_NAT_DONE;
 444
 445         return NF_ACCEPT;
 446 }
 447 EXPORT_SYMBOL(nf_nat_setup_info);
 448
 449 static unsigned int
 450 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 451 {
 452         /* Force range to this IP; let proto decide mapping for
 453          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 454          * Use reply in case it's already been mangled (eg local packet).
 455          */
 456         union nf_inet_addr ip =
 457                 (manip == NF_NAT_MANIP_SRC ?
 458                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 459                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 460         struct nf_nat_range range = {
 461                 .flags          = NF_NAT_RANGE_MAP_IPS,
 462                 .min_addr       = ip,
 463                 .max_addr       = ip,
 464         };
 465         return nf_nat_setup_info(ct, &range, manip);
 466 }
 467
 468 unsigned int
 469 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 470 {
 471         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 472 }
 473 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 474
 475 /* Do packet manipulations according to nf_nat_setup_info. */
 476 unsigned int nf_nat_packet(struct nf_conn *ct,
 477                            enum ip_conntrack_info ctinfo,
 478                            unsigned int hooknum,
 479                            struct sk_buff *skb)
 480 {
 481         const struct nf_nat_l3proto *l3proto;
 482         const struct nf_nat_l4proto *l4proto;
 483         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 484         unsigned long statusbit;
 485         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 486
 487         if (mtype == NF_NAT_MANIP_SRC)
 488                 statusbit = IPS_SRC_NAT;
 489         else
 490                 statusbit = IPS_DST_NAT;
 491
 492         /* Invert if this is reply dir. */
 493         if (dir == IP_CT_DIR_REPLY)
 494                 statusbit ^= IPS_NAT_MASK;
 495
 496         /* Non-atomic: these bits don't change. */
 497         if (ct->status & statusbit) {
 498                 struct nf_conntrack_tuple target;
 499
 500                 /* We are aiming to look like inverse of other direction. */
 501                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 502
 503                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 504                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 505                                                 target.dst.protonum);
 506                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 507                         return NF_DROP;
 508         }
 509         return NF_ACCEPT;
 510 }
 511 EXPORT_SYMBOL_GPL(nf_nat_packet);
 512
 513 struct nf_nat_proto_clean {
 514         u8      l3proto;
 515         u8      l4proto;
 516 };
 517
 518 /* kill conntracks with affected NAT section */
 519 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 520 {
 521         const struct nf_nat_proto_clean *clean = data;
 522
 523         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 524             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 525                 return 0;
 526
 527         return i->status & IPS_NAT_MASK ? 1 : 0;
 528 }
 529
 530 static void __nf_nat_cleanup_conntrack(struct nf_conn *ct)
 531 {
 532         unsigned int h;
 533
 534         h = hash_by_src(nf_ct_net(ct), &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 535         spin_lock_bh(&nf_nat_locks[h % ARRAY_SIZE(nf_nat_locks)]);
 536         hlist_del_rcu(&ct->nat_bysource);
 537         spin_unlock_bh(&nf_nat_locks[h % ARRAY_SIZE(nf_nat_locks)]);
 538 }
 539
 540 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 541 {
 542         if (nf_nat_proto_remove(ct, data))
 543                 return 1;
 544
 545         if ((ct->status & IPS_SRC_NAT_DONE) == 0)
 546                 return 0;
 547
 548         /* This netns is being destroyed, and conntrack has nat null binding.
 549          * Remove it from bysource hash, as the table will be freed soon.
 550          *
 551          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 552          * will delete entry from already-freed table.
 553          */
 554         clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 555         __nf_nat_cleanup_conntrack(ct);
 556
 557         /* don't delete conntrack.  Although that would make things a lot
 558          * simpler, we'd end up flushing all conntracks on nat rmmod.
 559          */
 560         return 0;
 561 }
 562
 563 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 564 {
 565         struct nf_nat_proto_clean clean = {
 566                 .l3proto = l3proto,
 567                 .l4proto = l4proto,
 568         };
 569
 570         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 571 }
 572
 573 static void nf_nat_l3proto_clean(u8 l3proto)
 574 {
 575         struct nf_nat_proto_clean clean = {
 576                 .l3proto = l3proto,
 577         };
 578
 579         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 580 }
 581
 582 /* Protocol registration. */
 583 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 584 {
 585         const struct nf_nat_l4proto **l4protos;
 586         unsigned int i;
 587         int ret = 0;
 588
 589         mutex_lock(&nf_nat_proto_mutex);
 590         if (nf_nat_l4protos[l3proto] == NULL) {
 591                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 592                                    GFP_KERNEL);
 593                 if (l4protos == NULL) {
 594                         ret = -ENOMEM;
 595                         goto out;
 596                 }
 597
 598                 for (i = 0; i < IPPROTO_MAX; i++)
 599                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 600
 601                 /* Before making proto_array visible to lockless readers,
 602                  * we must make sure its content is committed to memory.
 603                  */
 604                 smp_wmb();
 605
 606                 nf_nat_l4protos[l3proto] = l4protos;
 607         }
 608
 609         if (rcu_dereference_protected(
 610                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 611                         lockdep_is_held(&nf_nat_proto_mutex)
 612                         ) != &nf_nat_l4proto_unknown) {
 613                 ret = -EBUSY;
 614                 goto out;
 615         }
 616         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 617  out:
 618         mutex_unlock(&nf_nat_proto_mutex);
 619         return ret;
 620 }
 621 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 622
 623 /* No one stores the protocol anywhere; simply delete it. */
 624 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 625 {
 626         mutex_lock(&nf_nat_proto_mutex);
 627         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 628                          &nf_nat_l4proto_unknown);
 629         mutex_unlock(&nf_nat_proto_mutex);
 630         synchronize_rcu();
 631
 632         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 633 }
 634 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 635
 636 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 637 {
 638         int err;
 639
 640         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 641         if (err < 0)
 642                 return err;
 643
 644         mutex_lock(&nf_nat_proto_mutex);
 645         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 646                          &nf_nat_l4proto_tcp);
 647         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 648                          &nf_nat_l4proto_udp);
 649 #ifdef CONFIG_NF_NAT_PROTO_DCCP
 650         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_DCCP],
 651                          &nf_nat_l4proto_dccp);
 652 #endif
 653 #ifdef CONFIG_NF_NAT_PROTO_SCTP
 654         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_SCTP],
 655                          &nf_nat_l4proto_sctp);
 656 #endif
 657 #ifdef CONFIG_NF_NAT_PROTO_UDPLITE
 658         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDPLITE],
 659                          &nf_nat_l4proto_udplite);
 660 #endif
 661         mutex_unlock(&nf_nat_proto_mutex);
 662
 663         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 664         return 0;
 665 }
 666 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 667
 668 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 669 {
 670         mutex_lock(&nf_nat_proto_mutex);
 671         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 672         mutex_unlock(&nf_nat_proto_mutex);
 673         synchronize_rcu();
 674
 675         nf_nat_l3proto_clean(l3proto->l3proto);
 676         nf_ct_l3proto_module_put(l3proto->l3proto);
 677 }
 678 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 679
 680 /* No one using conntrack by the time this called. */
 681 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 682 {
 683         if (ct->status & IPS_SRC_NAT_DONE)
 684                 __nf_nat_cleanup_conntrack(ct);
 685 }
 686
 687 static struct nf_ct_ext_type nat_extend __read_mostly = {
 688         .len            = sizeof(struct nf_conn_nat),
 689         .align          = __alignof__(struct nf_conn_nat),
 690         .destroy        = nf_nat_cleanup_conntrack,
 691         .id             = NF_CT_EXT_NAT,
 692 };
 693
 694 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 695
 696 #include <linux/netfilter/nfnetlink.h>
 697 #include <linux/netfilter/nfnetlink_conntrack.h>
 698
 699 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 700         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 701         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 702 };
 703
 704 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 705                                      const struct nf_conn *ct,
 706                                      struct nf_nat_range *range)
 707 {
 708         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 709         const struct nf_nat_l4proto *l4proto;
 710         int err;
 711
 712         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr,
 713                                protonat_nla_policy, NULL);
 714         if (err < 0)
 715                 return err;
 716
 717         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 718         if (l4proto->nlattr_to_range)
 719                 err = l4proto->nlattr_to_range(tb, range);
 720
 721         return err;
 722 }
 723
 724 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 725         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 726         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 727         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 728         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 729         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 730 };
 731
 732 static int
 733 nfnetlink_parse_nat(const struct nlattr *nat,
 734                     const struct nf_conn *ct, struct nf_nat_range *range,
 735                     const struct nf_nat_l3proto *l3proto)
 736 {
 737         struct nlattr *tb[CTA_NAT_MAX+1];
 738         int err;
 739
 740         memset(range, 0, sizeof(*range));
 741
 742         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy, NULL);
 743         if (err < 0)
 744                 return err;
 745
 746         err = l3proto->nlattr_to_range(tb, range);
 747         if (err < 0)
 748                 return err;
 749
 750         if (!tb[CTA_NAT_PROTO])
 751                 return 0;
 752
 753         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 754 }
 755
 756 /* This function is called under rcu_read_lock() */
 757 static int
 758 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 759                           enum nf_nat_manip_type manip,
 760                           const struct nlattr *attr)
 761 {
 762         struct nf_nat_range range;
 763         const struct nf_nat_l3proto *l3proto;
 764         int err;
 765
 766         /* Should not happen, restricted to creating new conntracks
 767          * via ctnetlink.
 768          */
 769         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 770                 return -EEXIST;
 771
 772         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 773          * attach the null binding, otherwise this may oops.
 774          */
 775         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 776         if (l3proto == NULL)
 777                 return -EAGAIN;
 778
 779         /* No NAT information has been passed, allocate the null-binding */
 780         if (attr == NULL)
 781                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
 782
 783         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 784         if (err < 0)
 785                 return err;
 786
 787         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 788 }
 789 #else
 790 static int
 791 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 792                           enum nf_nat_manip_type manip,
 793                           const struct nlattr *attr)
 794 {
 795         return -EOPNOTSUPP;
 796 }
 797 #endif
 798
 799 static struct nf_ct_helper_expectfn follow_master_nat = {
 800         .name           = "nat-follow-master",
 801         .expectfn       = nf_nat_follow_master,
 802 };
 803
 804 static int __init nf_nat_init(void)
 805 {
 806         int ret, i;
 807
 808         /* Leave them the same for the moment. */
 809         nf_nat_htable_size = nf_conntrack_htable_size;
 810         if (nf_nat_htable_size < ARRAY_SIZE(nf_nat_locks))
 811                 nf_nat_htable_size = ARRAY_SIZE(nf_nat_locks);
 812
 813         nf_nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, 0);
 814         if (!nf_nat_bysource)
 815                 return -ENOMEM;
 816
 817         ret = nf_ct_extend_register(&nat_extend);
 818         if (ret < 0) {
 819                 nf_ct_free_hashtable(nf_nat_bysource, nf_nat_htable_size);
 820                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 821                 return ret;
 822         }
 823
 824         for (i = 0; i < ARRAY_SIZE(nf_nat_locks); i++)
 825                 spin_lock_init(&nf_nat_locks[i]);
 826
 827         nf_ct_helper_expectfn_register(&follow_master_nat);
 828
 829         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 830         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 831                            nfnetlink_parse_nat_setup);
 832 #ifdef CONFIG_XFRM
 833         BUG_ON(nf_nat_decode_session_hook != NULL);
 834         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 835 #endif
 836         return 0;
 837 }
 838
 839 static void __exit nf_nat_cleanup(void)
 840 {
 841         struct nf_nat_proto_clean clean = {};
 842         unsigned int i;
 843
 844         nf_ct_iterate_destroy(nf_nat_proto_clean, &clean);
 845
 846         nf_ct_extend_unregister(&nat_extend);
 847         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 848         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 849 #ifdef CONFIG_XFRM
 850         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 851 #endif
 852         synchronize_rcu();
 853
 854         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 855                 kfree(nf_nat_l4protos[i]);
 856         synchronize_net();
 857         nf_ct_free_hashtable(nf_nat_bysource, nf_nat_htable_size);
 858 }
 859
 860 MODULE_LICENSE("GPL");
 861
 862 module_init(nf_nat_init);
 863 module_exit(nf_nat_cleanup);