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         if (!dst_hold_safe(dst))
 101                 return -EHOSTUNREACH;
 102
 103         dst = xfrm_lookup(net, dst, &fl, skb->sk, 0);
 104         if (IS_ERR(dst))
 105                 return PTR_ERR(dst);
 106
 107         skb_dst_drop(skb);
 108         skb_dst_set(skb, dst);
 109
 110         /* Change in oif may mean change in hh_len. */
 111         hh_len = skb_dst(skb)->dev->hard_header_len;
 112         if (skb_headroom(skb) < hh_len &&
 113             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 114                 return -ENOMEM;
 115         return 0;
 116 }
 117 EXPORT_SYMBOL(nf_xfrm_me_harder);
 118 #endif /* CONFIG_XFRM */
 119
 120 /* We keep an extra hash for each conntrack, for fast searching. */
 121 static unsigned int
 122 hash_by_src(const struct net *n, const struct nf_conntrack_tuple *tuple)
 123 {
 124         unsigned int hash;
 125
 126         get_random_once(&nf_nat_hash_rnd, sizeof(nf_nat_hash_rnd));
 127
 128         /* Original src, to ensure we map it consistently if poss. */
 129         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 130                       tuple->dst.protonum ^ nf_nat_hash_rnd ^ net_hash_mix(n));
 131
 132         return reciprocal_scale(hash, nf_nat_htable_size);
 133 }
 134
 135 /* Is this tuple already taken? (not by us) */
 136 int
 137 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 138                   const struct nf_conn *ignored_conntrack)
 139 {
 140         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 141          * incoming ones.  NAT means they don't have a fixed mapping,
 142          * so we invert the tuple and look for the incoming reply.
 143          *
 144          * We could keep a separate hash if this proves too slow.
 145          */
 146         struct nf_conntrack_tuple reply;
 147
 148         nf_ct_invert_tuplepr(&reply, tuple);
 149         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 150 }
 151 EXPORT_SYMBOL(nf_nat_used_tuple);
 152
 153 /* If we source map this tuple so reply looks like reply_tuple, will
 154  * that meet the constraints of range.
 155  */
 156 static int in_range(const struct nf_nat_l3proto *l3proto,
 157                     const struct nf_nat_l4proto *l4proto,
 158                     const struct nf_conntrack_tuple *tuple,
 159                     const struct nf_nat_range *range)
 160 {
 161         /* If we are supposed to map IPs, then we must be in the
 162          * range specified, otherwise let this drag us onto a new src IP.
 163          */
 164         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 165             !l3proto->in_range(tuple, range))
 166                 return 0;
 167
 168         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 169             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 170                               &range->min_proto, &range->max_proto))
 171                 return 1;
 172
 173         return 0;
 174 }
 175
 176 static inline int
 177 same_src(const struct nf_conn *ct,
 178          const struct nf_conntrack_tuple *tuple)
 179 {
 180         const struct nf_conntrack_tuple *t;
 181
 182         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 183         return (t->dst.protonum == tuple->dst.protonum &&
 184                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 185                 t->src.u.all == tuple->src.u.all);
 186 }
 187
 188 /* Only called for SRC manip */
 189 static int
 190 find_appropriate_src(struct net *net,
 191                      const struct nf_conntrack_zone *zone,
 192                      const struct nf_nat_l3proto *l3proto,
 193                      const struct nf_nat_l4proto *l4proto,
 194                      const struct nf_conntrack_tuple *tuple,
 195                      struct nf_conntrack_tuple *result,
 196                      const struct nf_nat_range *range)
 197 {
 198         unsigned int h = hash_by_src(net, tuple);
 199         const struct nf_conn *ct;
 200
 201         hlist_for_each_entry_rcu(ct, &nf_nat_bysource[h], nat_bysource) {
 202                 if (same_src(ct, tuple) &&
 203                     net_eq(net, nf_ct_net(ct)) &&
 204                     nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) {
 205                         /* Copy source part from reply tuple. */
 206                         nf_ct_invert_tuplepr(result,
 207                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 208                         result->dst = tuple->dst;
 209
 210                         if (in_range(l3proto, l4proto, result, range))
 211                                 return 1;
 212                 }
 213         }
 214         return 0;
 215 }
 216
 217 /* For [FUTURE] fragmentation handling, we want the least-used
 218  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 219  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 220  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 221  * the ip with the lowest src-ip/dst-ip/proto usage.
 222  */
 223 static void
 224 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 225                     struct nf_conntrack_tuple *tuple,
 226                     const struct nf_nat_range *range,
 227                     const struct nf_conn *ct,
 228                     enum nf_nat_manip_type maniptype)
 229 {
 230         union nf_inet_addr *var_ipp;
 231         unsigned int i, max;
 232         /* Host order */
 233         u32 minip, maxip, j, dist;
 234         bool full_range;
 235
 236         /* No IP mapping?  Do nothing. */
 237         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 238                 return;
 239
 240         if (maniptype == NF_NAT_MANIP_SRC)
 241                 var_ipp = &tuple->src.u3;
 242         else
 243                 var_ipp = &tuple->dst.u3;
 244
 245         /* Fast path: only one choice. */
 246         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 247                 *var_ipp = range->min_addr;
 248                 return;
 249         }
 250
 251         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 252                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 253         else
 254                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 255
 256         /* Hashing source and destination IPs gives a fairly even
 257          * spread in practice (if there are a small number of IPs
 258          * involved, there usually aren't that many connections
 259          * anyway).  The consistency means that servers see the same
 260          * client coming from the same IP (some Internet Banking sites
 261          * like this), even across reboots.
 262          */
 263         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 264                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 265                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 266
 267         full_range = false;
 268         for (i = 0; i <= max; i++) {
 269                 /* If first bytes of the address are at the maximum, use the
 270                  * distance. Otherwise use the full range.
 271                  */
 272                 if (!full_range) {
 273                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 274                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 275                         dist  = maxip - minip + 1;
 276                 } else {
 277                         minip = 0;
 278                         dist  = ~0;
 279                 }
 280
 281                 var_ipp->all[i] = (__force __u32)
 282                         htonl(minip + reciprocal_scale(j, dist));
 283                 if (var_ipp->all[i] != range->max_addr.all[i])
 284                         full_range = true;
 285
 286                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 287                         j ^= (__force u32)tuple->dst.u3.all[i];
 288         }
 289 }
 290
 291 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 292  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 293  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 294  * range. It might not be possible to get a unique tuple, but we try.
 295  * At worst (or if we race), we will end up with a final duplicate in
 296  * __ip_conntrack_confirm and drop the packet. */
 297 static void
 298 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 299                  const struct nf_conntrack_tuple *orig_tuple,
 300                  const struct nf_nat_range *range,
 301                  struct nf_conn *ct,
 302                  enum nf_nat_manip_type maniptype)
 303 {
 304         const struct nf_conntrack_zone *zone;
 305         const struct nf_nat_l3proto *l3proto;
 306         const struct nf_nat_l4proto *l4proto;
 307         struct net *net = nf_ct_net(ct);
 308
 309         zone = nf_ct_zone(ct);
 310
 311         rcu_read_lock();
 312         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 313         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 314                                         orig_tuple->dst.protonum);
 315
 316         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 317          * and that same mapping gives a unique tuple within the given
 318          * range, use that.
 319          *
 320          * This is only required for source (ie. NAT/masq) mappings.
 321          * So far, we don't do local source mappings, so multiple
 322          * manips not an issue.
 323          */
 324         if (maniptype == NF_NAT_MANIP_SRC &&
 325             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 326                 /* try the original tuple first */
 327                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 328                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 329                                 *tuple = *orig_tuple;
 330                                 goto out;
 331                         }
 332                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 333                                                 orig_tuple, tuple, range)) {
 334                         pr_debug("get_unique_tuple: Found current src map\n");
 335                         if (!nf_nat_used_tuple(tuple, ct))
 336                                 goto out;
 337                 }
 338         }
 339
 340         /* 2) Select the least-used IP/proto combination in the given range */
 341         *tuple = *orig_tuple;
 342         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 343
 344         /* 3) The per-protocol part of the manip is made to map into
 345          * the range to make a unique tuple.
 346          */
 347
 348         /* Only bother mapping if it's not already in range and unique */
 349         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 350                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 351                         if (l4proto->in_range(tuple, maniptype,
 352                                               &range->min_proto,
 353                                               &range->max_proto) &&
 354                             (range->min_proto.all == range->max_proto.all ||
 355                              !nf_nat_used_tuple(tuple, ct)))
 356                                 goto out;
 357                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 358                         goto out;
 359                 }
 360         }
 361
 362         /* Last change: get protocol to try to obtain unique tuple. */
 363         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 364 out:
 365         rcu_read_unlock();
 366 }
 367
 368 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 369 {
 370         struct nf_conn_nat *nat = nfct_nat(ct);
 371         if (nat)
 372                 return nat;
 373
 374         if (!nf_ct_is_confirmed(ct))
 375                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 376
 377         return nat;
 378 }
 379 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 380
 381 unsigned int
 382 nf_nat_setup_info(struct nf_conn *ct,
 383                   const struct nf_nat_range *range,
 384                   enum nf_nat_manip_type maniptype)
 385 {
 386         struct net *net = nf_ct_net(ct);
 387         struct nf_conntrack_tuple curr_tuple, new_tuple;
 388
 389         /* Can't setup nat info for confirmed ct. */
 390         if (nf_ct_is_confirmed(ct))
 391                 return NF_ACCEPT;
 392
 393         WARN_ON(maniptype != NF_NAT_MANIP_SRC &&
 394                 maniptype != NF_NAT_MANIP_DST);
 395
 396         if (WARN_ON(nf_nat_initialized(ct, maniptype)))
 397                 return NF_DROP;
 398
 399         /* What we've got will look like inverse of reply. Normally
 400          * this is what is in the conntrack, except for prior
 401          * manipulations (future optimization: if num_manips == 0,
 402          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 403          */
 404         nf_ct_invert_tuplepr(&curr_tuple,
 405                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 406
 407         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 408
 409         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 410                 struct nf_conntrack_tuple reply;
 411
 412                 /* Alter conntrack table so will recognize replies. */
 413                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 414                 nf_conntrack_alter_reply(ct, &reply);
 415
 416                 /* Non-atomic: we own this at the moment. */
 417                 if (maniptype == NF_NAT_MANIP_SRC)
 418                         ct->status |= IPS_SRC_NAT;
 419                 else
 420                         ct->status |= IPS_DST_NAT;
 421
 422                 if (nfct_help(ct) && !nfct_seqadj(ct))
 423                         if (!nfct_seqadj_ext_add(ct))
 424                                 return NF_DROP;
 425         }
 426
 427         if (maniptype == NF_NAT_MANIP_SRC) {
 428                 unsigned int srchash;
 429                 spinlock_t *lock;
 430
 431                 srchash = hash_by_src(net,
 432                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 433                 lock = &nf_nat_locks[srchash % CONNTRACK_LOCKS];
 434                 spin_lock_bh(lock);
 435                 hlist_add_head_rcu(&ct->nat_bysource,
 436                                    &nf_nat_bysource[srchash]);
 437                 spin_unlock_bh(lock);
 438         }
 439
 440         /* It's done. */
 441         if (maniptype == NF_NAT_MANIP_DST)
 442                 ct->status |= IPS_DST_NAT_DONE;
 443         else
 444                 ct->status |= IPS_SRC_NAT_DONE;
 445
 446         return NF_ACCEPT;
 447 }
 448 EXPORT_SYMBOL(nf_nat_setup_info);
 449
 450 static unsigned int
 451 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 452 {
 453         /* Force range to this IP; let proto decide mapping for
 454          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 455          * Use reply in case it's already been mangled (eg local packet).
 456          */
 457         union nf_inet_addr ip =
 458                 (manip == NF_NAT_MANIP_SRC ?
 459                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 460                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 461         struct nf_nat_range range = {
 462                 .flags          = NF_NAT_RANGE_MAP_IPS,
 463                 .min_addr       = ip,
 464                 .max_addr       = ip,
 465         };
 466         return nf_nat_setup_info(ct, &range, manip);
 467 }
 468
 469 unsigned int
 470 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 471 {
 472         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 473 }
 474 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 475
 476 /* Do packet manipulations according to nf_nat_setup_info. */
 477 unsigned int nf_nat_packet(struct nf_conn *ct,
 478                            enum ip_conntrack_info ctinfo,
 479                            unsigned int hooknum,
 480                            struct sk_buff *skb)
 481 {
 482         const struct nf_nat_l3proto *l3proto;
 483         const struct nf_nat_l4proto *l4proto;
 484         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 485         unsigned long statusbit;
 486         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 487
 488         if (mtype == NF_NAT_MANIP_SRC)
 489                 statusbit = IPS_SRC_NAT;
 490         else
 491                 statusbit = IPS_DST_NAT;
 492
 493         /* Invert if this is reply dir. */
 494         if (dir == IP_CT_DIR_REPLY)
 495                 statusbit ^= IPS_NAT_MASK;
 496
 497         /* Non-atomic: these bits don't change. */
 498         if (ct->status & statusbit) {
 499                 struct nf_conntrack_tuple target;
 500
 501                 /* We are aiming to look like inverse of other direction. */
 502                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 503
 504                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 505                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 506                                                 target.dst.protonum);
 507                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 508                         return NF_DROP;
 509         }
 510         return NF_ACCEPT;
 511 }
 512 EXPORT_SYMBOL_GPL(nf_nat_packet);
 513
 514 struct nf_nat_proto_clean {
 515         u8      l3proto;
 516         u8      l4proto;
 517 };
 518
 519 /* kill conntracks with affected NAT section */
 520 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 521 {
 522         const struct nf_nat_proto_clean *clean = data;
 523
 524         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 525             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 526                 return 0;
 527
 528         return i->status & IPS_NAT_MASK ? 1 : 0;
 529 }
 530
 531 static void __nf_nat_cleanup_conntrack(struct nf_conn *ct)
 532 {
 533         unsigned int h;
 534
 535         h = hash_by_src(nf_ct_net(ct), &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 536         spin_lock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 537         hlist_del_rcu(&ct->nat_bysource);
 538         spin_unlock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 539 }
 540
 541 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 542 {
 543         if (nf_nat_proto_remove(ct, data))
 544                 return 1;
 545
 546         if ((ct->status & IPS_SRC_NAT_DONE) == 0)
 547                 return 0;
 548
 549         /* This netns is being destroyed, and conntrack has nat null binding.
 550          * Remove it from bysource hash, as the table will be freed soon.
 551          *
 552          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 553          * will delete entry from already-freed table.
 554          */
 555         clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 556         __nf_nat_cleanup_conntrack(ct);
 557
 558         /* don't delete conntrack.  Although that would make things a lot
 559          * simpler, we'd end up flushing all conntracks on nat rmmod.
 560          */
 561         return 0;
 562 }
 563
 564 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 565 {
 566         struct nf_nat_proto_clean clean = {
 567                 .l3proto = l3proto,
 568                 .l4proto = l4proto,
 569         };
 570
 571         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 572 }
 573
 574 static void nf_nat_l3proto_clean(u8 l3proto)
 575 {
 576         struct nf_nat_proto_clean clean = {
 577                 .l3proto = l3proto,
 578         };
 579
 580         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 581 }
 582
 583 /* Protocol registration. */
 584 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 585 {
 586         const struct nf_nat_l4proto **l4protos;
 587         unsigned int i;
 588         int ret = 0;
 589
 590         mutex_lock(&nf_nat_proto_mutex);
 591         if (nf_nat_l4protos[l3proto] == NULL) {
 592                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 593                                    GFP_KERNEL);
 594                 if (l4protos == NULL) {
 595                         ret = -ENOMEM;
 596                         goto out;
 597                 }
 598
 599                 for (i = 0; i < IPPROTO_MAX; i++)
 600                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 601
 602                 /* Before making proto_array visible to lockless readers,
 603                  * we must make sure its content is committed to memory.
 604                  */
 605                 smp_wmb();
 606
 607                 nf_nat_l4protos[l3proto] = l4protos;
 608         }
 609
 610         if (rcu_dereference_protected(
 611                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 612                         lockdep_is_held(&nf_nat_proto_mutex)
 613                         ) != &nf_nat_l4proto_unknown) {
 614                 ret = -EBUSY;
 615                 goto out;
 616         }
 617         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 618  out:
 619         mutex_unlock(&nf_nat_proto_mutex);
 620         return ret;
 621 }
 622 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 623
 624 /* No one stores the protocol anywhere; simply delete it. */
 625 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 626 {
 627         mutex_lock(&nf_nat_proto_mutex);
 628         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 629                          &nf_nat_l4proto_unknown);
 630         mutex_unlock(&nf_nat_proto_mutex);
 631         synchronize_rcu();
 632
 633         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 634 }
 635 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 636
 637 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 638 {
 639         int err;
 640
 641         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 642         if (err < 0)
 643                 return err;
 644
 645         mutex_lock(&nf_nat_proto_mutex);
 646         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 647                          &nf_nat_l4proto_tcp);
 648         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 649                          &nf_nat_l4proto_udp);
 650 #ifdef CONFIG_NF_NAT_PROTO_DCCP
 651         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_DCCP],
 652                          &nf_nat_l4proto_dccp);
 653 #endif
 654 #ifdef CONFIG_NF_NAT_PROTO_SCTP
 655         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_SCTP],
 656                          &nf_nat_l4proto_sctp);
 657 #endif
 658 #ifdef CONFIG_NF_NAT_PROTO_UDPLITE
 659         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDPLITE],
 660                          &nf_nat_l4proto_udplite);
 661 #endif
 662         mutex_unlock(&nf_nat_proto_mutex);
 663
 664         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 665         return 0;
 666 }
 667 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 668
 669 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 670 {
 671         mutex_lock(&nf_nat_proto_mutex);
 672         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 673         mutex_unlock(&nf_nat_proto_mutex);
 674         synchronize_rcu();
 675
 676         nf_nat_l3proto_clean(l3proto->l3proto);
 677         nf_ct_l3proto_module_put(l3proto->l3proto);
 678 }
 679 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 680
 681 /* No one using conntrack by the time this called. */
 682 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 683 {
 684         if (ct->status & IPS_SRC_NAT_DONE)
 685                 __nf_nat_cleanup_conntrack(ct);
 686 }
 687
 688 static struct nf_ct_ext_type nat_extend __read_mostly = {
 689         .len            = sizeof(struct nf_conn_nat),
 690         .align          = __alignof__(struct nf_conn_nat),
 691         .destroy        = nf_nat_cleanup_conntrack,
 692         .id             = NF_CT_EXT_NAT,
 693 };
 694
 695 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 696
 697 #include <linux/netfilter/nfnetlink.h>
 698 #include <linux/netfilter/nfnetlink_conntrack.h>
 699
 700 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 701         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 702         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 703 };
 704
 705 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 706                                      const struct nf_conn *ct,
 707                                      struct nf_nat_range *range)
 708 {
 709         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 710         const struct nf_nat_l4proto *l4proto;
 711         int err;
 712
 713         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr,
 714                                protonat_nla_policy, NULL);
 715         if (err < 0)
 716                 return err;
 717
 718         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 719         if (l4proto->nlattr_to_range)
 720                 err = l4proto->nlattr_to_range(tb, range);
 721
 722         return err;
 723 }
 724
 725 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 726         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 727         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 728         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 729         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 730         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 731 };
 732
 733 static int
 734 nfnetlink_parse_nat(const struct nlattr *nat,
 735                     const struct nf_conn *ct, struct nf_nat_range *range,
 736                     const struct nf_nat_l3proto *l3proto)
 737 {
 738         struct nlattr *tb[CTA_NAT_MAX+1];
 739         int err;
 740
 741         memset(range, 0, sizeof(*range));
 742
 743         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy, NULL);
 744         if (err < 0)
 745                 return err;
 746
 747         err = l3proto->nlattr_to_range(tb, range);
 748         if (err < 0)
 749                 return err;
 750
 751         if (!tb[CTA_NAT_PROTO])
 752                 return 0;
 753
 754         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 755 }
 756
 757 /* This function is called under rcu_read_lock() */
 758 static int
 759 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 760                           enum nf_nat_manip_type manip,
 761                           const struct nlattr *attr)
 762 {
 763         struct nf_nat_range range;
 764         const struct nf_nat_l3proto *l3proto;
 765         int err;
 766
 767         /* Should not happen, restricted to creating new conntracks
 768          * via ctnetlink.
 769          */
 770         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 771                 return -EEXIST;
 772
 773         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 774          * attach the null binding, otherwise this may oops.
 775          */
 776         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 777         if (l3proto == NULL)
 778                 return -EAGAIN;
 779
 780         /* No NAT information has been passed, allocate the null-binding */
 781         if (attr == NULL)
 782                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
 783
 784         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 785         if (err < 0)
 786                 return err;
 787
 788         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 789 }
 790 #else
 791 static int
 792 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 793                           enum nf_nat_manip_type manip,
 794                           const struct nlattr *attr)
 795 {
 796         return -EOPNOTSUPP;
 797 }
 798 #endif
 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, i;
 808
 809         /* Leave them the same for the moment. */
 810         nf_nat_htable_size = nf_conntrack_htable_size;
 811         if (nf_nat_htable_size < CONNTRACK_LOCKS)
 812                 nf_nat_htable_size = CONNTRACK_LOCKS;
 813
 814         nf_nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, 0);
 815         if (!nf_nat_bysource)
 816                 return -ENOMEM;
 817
 818         ret = nf_ct_extend_register(&nat_extend);
 819         if (ret < 0) {
 820                 nf_ct_free_hashtable(nf_nat_bysource, nf_nat_htable_size);
 821                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 822                 return ret;
 823         }
 824
 825         for (i = 0; i < CONNTRACK_LOCKS; i++)
 826                 spin_lock_init(&nf_nat_locks[i]);
 827
 828         nf_ct_helper_expectfn_register(&follow_master_nat);
 829
 830         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 831         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 832                            nfnetlink_parse_nat_setup);
 833 #ifdef CONFIG_XFRM
 834         BUG_ON(nf_nat_decode_session_hook != NULL);
 835         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 836 #endif
 837         return 0;
 838 }
 839
 840 static void __exit nf_nat_cleanup(void)
 841 {
 842         struct nf_nat_proto_clean clean = {};
 843         unsigned int i;
 844
 845         nf_ct_iterate_destroy(nf_nat_proto_clean, &clean);
 846
 847         nf_ct_extend_unregister(&nat_extend);
 848         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 849         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 850 #ifdef CONFIG_XFRM
 851         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 852 #endif
 853         synchronize_rcu();
 854
 855         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 856                 kfree(nf_nat_l4protos[i]);
 857         synchronize_net();
 858         nf_ct_free_hashtable(nf_nat_bysource, nf_nat_htable_size);
 859 }
 860
 861 MODULE_LICENSE("GPL");
 862
 863 module_init(nf_nat_init);
 864 module_exit(nf_nat_cleanup);