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 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 362 {
 363         struct nf_conn_nat *nat = nfct_nat(ct);
 364         if (nat)
 365                 return nat;
 366
 367         if (!nf_ct_is_confirmed(ct))
 368                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 369
 370         return nat;
 371 }
 372 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 373
 374 unsigned int
 375 nf_nat_setup_info(struct nf_conn *ct,
 376                   const struct nf_nat_range *range,
 377                   enum nf_nat_manip_type maniptype)
 378 {
 379         struct net *net = nf_ct_net(ct);
 380         struct nf_conntrack_tuple curr_tuple, new_tuple;
 381         struct nf_conn_nat *nat;
 382
 383         /* nat helper or nfctnetlink also setup binding */
 384         nat = nf_ct_nat_ext_add(ct);
 385         if (nat == NULL)
 386                 return NF_ACCEPT;
 387
 388         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 389                      maniptype == NF_NAT_MANIP_DST);
 390         BUG_ON(nf_nat_initialized(ct, maniptype));
 391
 392         /* What we've got will look like inverse of reply. Normally
 393          * this is what is in the conntrack, except for prior
 394          * manipulations (future optimization: if num_manips == 0,
 395          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 396          */
 397         nf_ct_invert_tuplepr(&curr_tuple,
 398                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 399
 400         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 401
 402         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 403                 struct nf_conntrack_tuple reply;
 404
 405                 /* Alter conntrack table so will recognize replies. */
 406                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 407                 nf_conntrack_alter_reply(ct, &reply);
 408
 409                 /* Non-atomic: we own this at the moment. */
 410                 if (maniptype == NF_NAT_MANIP_SRC)
 411                         ct->status |= IPS_SRC_NAT;
 412                 else
 413                         ct->status |= IPS_DST_NAT;
 414
 415                 if (nfct_help(ct))
 416                         nfct_seqadj_ext_add(ct);
 417         }
 418
 419         if (maniptype == NF_NAT_MANIP_SRC) {
 420                 unsigned int srchash;
 421
 422                 srchash = hash_by_src(net, nf_ct_zone(ct),
 423                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 424                 spin_lock_bh(&nf_nat_lock);
 425                 /* nf_conntrack_alter_reply might re-allocate extension aera */
 426                 nat = nfct_nat(ct);
 427                 nat->ct = ct;
 428                 hlist_add_head_rcu(&nat->bysource,
 429                                    &net->ct.nat_bysource[srchash]);
 430                 spin_unlock_bh(&nf_nat_lock);
 431         }
 432
 433         /* It's done. */
 434         if (maniptype == NF_NAT_MANIP_DST)
 435                 ct->status |= IPS_DST_NAT_DONE;
 436         else
 437                 ct->status |= IPS_SRC_NAT_DONE;
 438
 439         return NF_ACCEPT;
 440 }
 441 EXPORT_SYMBOL(nf_nat_setup_info);
 442
 443 static unsigned int
 444 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 445 {
 446         /* Force range to this IP; let proto decide mapping for
 447          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 448          * Use reply in case it's already been mangled (eg local packet).
 449          */
 450         union nf_inet_addr ip =
 451                 (manip == NF_NAT_MANIP_SRC ?
 452                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 453                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 454         struct nf_nat_range range = {
 455                 .flags          = NF_NAT_RANGE_MAP_IPS,
 456                 .min_addr       = ip,
 457                 .max_addr       = ip,
 458         };
 459         return nf_nat_setup_info(ct, &range, manip);
 460 }
 461
 462 unsigned int
 463 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 464 {
 465         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 466 }
 467 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 468
 469 /* Do packet manipulations according to nf_nat_setup_info. */
 470 unsigned int nf_nat_packet(struct nf_conn *ct,
 471                            enum ip_conntrack_info ctinfo,
 472                            unsigned int hooknum,
 473                            struct sk_buff *skb)
 474 {
 475         const struct nf_nat_l3proto *l3proto;
 476         const struct nf_nat_l4proto *l4proto;
 477         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 478         unsigned long statusbit;
 479         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 480
 481         if (mtype == NF_NAT_MANIP_SRC)
 482                 statusbit = IPS_SRC_NAT;
 483         else
 484                 statusbit = IPS_DST_NAT;
 485
 486         /* Invert if this is reply dir. */
 487         if (dir == IP_CT_DIR_REPLY)
 488                 statusbit ^= IPS_NAT_MASK;
 489
 490         /* Non-atomic: these bits don't change. */
 491         if (ct->status & statusbit) {
 492                 struct nf_conntrack_tuple target;
 493
 494                 /* We are aiming to look like inverse of other direction. */
 495                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 496
 497                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 498                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 499                                                 target.dst.protonum);
 500                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 501                         return NF_DROP;
 502         }
 503         return NF_ACCEPT;
 504 }
 505 EXPORT_SYMBOL_GPL(nf_nat_packet);
 506
 507 struct nf_nat_proto_clean {
 508         u8      l3proto;
 509         u8      l4proto;
 510 };
 511
 512 /* kill conntracks with affected NAT section */
 513 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 514 {
 515         const struct nf_nat_proto_clean *clean = data;
 516         struct nf_conn_nat *nat = nfct_nat(i);
 517
 518         if (!nat)
 519                 return 0;
 520
 521         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 522             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 523                 return 0;
 524
 525         return i->status & IPS_NAT_MASK ? 1 : 0;
 526 }
 527
 528 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 529 {
 530         struct nf_nat_proto_clean clean = {
 531                 .l3proto = l3proto,
 532                 .l4proto = l4proto,
 533         };
 534         struct net *net;
 535
 536         rtnl_lock();
 537         for_each_net(net)
 538                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 539         rtnl_unlock();
 540 }
 541
 542 static void nf_nat_l3proto_clean(u8 l3proto)
 543 {
 544         struct nf_nat_proto_clean clean = {
 545                 .l3proto = l3proto,
 546         };
 547         struct net *net;
 548
 549         rtnl_lock();
 550
 551         for_each_net(net)
 552                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 553         rtnl_unlock();
 554 }
 555
 556 /* Protocol registration. */
 557 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 558 {
 559         const struct nf_nat_l4proto **l4protos;
 560         unsigned int i;
 561         int ret = 0;
 562
 563         mutex_lock(&nf_nat_proto_mutex);
 564         if (nf_nat_l4protos[l3proto] == NULL) {
 565                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 566                                    GFP_KERNEL);
 567                 if (l4protos == NULL) {
 568                         ret = -ENOMEM;
 569                         goto out;
 570                 }
 571
 572                 for (i = 0; i < IPPROTO_MAX; i++)
 573                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 574
 575                 /* Before making proto_array visible to lockless readers,
 576                  * we must make sure its content is committed to memory.
 577                  */
 578                 smp_wmb();
 579
 580                 nf_nat_l4protos[l3proto] = l4protos;
 581         }
 582
 583         if (rcu_dereference_protected(
 584                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 585                         lockdep_is_held(&nf_nat_proto_mutex)
 586                         ) != &nf_nat_l4proto_unknown) {
 587                 ret = -EBUSY;
 588                 goto out;
 589         }
 590         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 591  out:
 592         mutex_unlock(&nf_nat_proto_mutex);
 593         return ret;
 594 }
 595 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 596
 597 /* No one stores the protocol anywhere; simply delete it. */
 598 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 599 {
 600         mutex_lock(&nf_nat_proto_mutex);
 601         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 602                          &nf_nat_l4proto_unknown);
 603         mutex_unlock(&nf_nat_proto_mutex);
 604         synchronize_rcu();
 605
 606         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 607 }
 608 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 609
 610 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 611 {
 612         int err;
 613
 614         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 615         if (err < 0)
 616                 return err;
 617
 618         mutex_lock(&nf_nat_proto_mutex);
 619         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 620                          &nf_nat_l4proto_tcp);
 621         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 622                          &nf_nat_l4proto_udp);
 623         mutex_unlock(&nf_nat_proto_mutex);
 624
 625         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 626         return 0;
 627 }
 628 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 629
 630 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 631 {
 632         mutex_lock(&nf_nat_proto_mutex);
 633         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 634         mutex_unlock(&nf_nat_proto_mutex);
 635         synchronize_rcu();
 636
 637         nf_nat_l3proto_clean(l3proto->l3proto);
 638         nf_ct_l3proto_module_put(l3proto->l3proto);
 639 }
 640 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 641
 642 /* No one using conntrack by the time this called. */
 643 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 644 {
 645         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 646
 647         if (nat == NULL || nat->ct == NULL)
 648                 return;
 649
 650         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 651
 652         spin_lock_bh(&nf_nat_lock);
 653         hlist_del_rcu(&nat->bysource);
 654         spin_unlock_bh(&nf_nat_lock);
 655 }
 656
 657 static void nf_nat_move_storage(void *new, void *old)
 658 {
 659         struct nf_conn_nat *new_nat = new;
 660         struct nf_conn_nat *old_nat = old;
 661         struct nf_conn *ct = old_nat->ct;
 662
 663         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 664                 return;
 665
 666         spin_lock_bh(&nf_nat_lock);
 667         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 668         spin_unlock_bh(&nf_nat_lock);
 669 }
 670
 671 static struct nf_ct_ext_type nat_extend __read_mostly = {
 672         .len            = sizeof(struct nf_conn_nat),
 673         .align          = __alignof__(struct nf_conn_nat),
 674         .destroy        = nf_nat_cleanup_conntrack,
 675         .move           = nf_nat_move_storage,
 676         .id             = NF_CT_EXT_NAT,
 677         .flags          = NF_CT_EXT_F_PREALLOC,
 678 };
 679
 680 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 681
 682 #include <linux/netfilter/nfnetlink.h>
 683 #include <linux/netfilter/nfnetlink_conntrack.h>
 684
 685 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 686         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 687         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 688 };
 689
 690 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 691                                      const struct nf_conn *ct,
 692                                      struct nf_nat_range *range)
 693 {
 694         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 695         const struct nf_nat_l4proto *l4proto;
 696         int err;
 697
 698         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 699         if (err < 0)
 700                 return err;
 701
 702         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 703         if (l4proto->nlattr_to_range)
 704                 err = l4proto->nlattr_to_range(tb, range);
 705
 706         return err;
 707 }
 708
 709 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 710         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 711         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 712         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 713         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 714         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 715 };
 716
 717 static int
 718 nfnetlink_parse_nat(const struct nlattr *nat,
 719                     const struct nf_conn *ct, struct nf_nat_range *range,
 720                     const struct nf_nat_l3proto *l3proto)
 721 {
 722         struct nlattr *tb[CTA_NAT_MAX+1];
 723         int err;
 724
 725         memset(range, 0, sizeof(*range));
 726
 727         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 728         if (err < 0)
 729                 return err;
 730
 731         err = l3proto->nlattr_to_range(tb, range);
 732         if (err < 0)
 733                 return err;
 734
 735         if (!tb[CTA_NAT_PROTO])
 736                 return 0;
 737
 738         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 739 }
 740
 741 /* This function is called under rcu_read_lock() */
 742 static int
 743 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 744                           enum nf_nat_manip_type manip,
 745                           const struct nlattr *attr)
 746 {
 747         struct nf_nat_range range;
 748         const struct nf_nat_l3proto *l3proto;
 749         int err;
 750
 751         /* Should not happen, restricted to creating new conntracks
 752          * via ctnetlink.
 753          */
 754         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 755                 return -EEXIST;
 756
 757         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 758          * attach the null binding, otherwise this may oops.
 759          */
 760         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 761         if (l3proto == NULL)
 762                 return -EAGAIN;
 763
 764         /* No NAT information has been passed, allocate the null-binding */
 765         if (attr == NULL)
 766                 return __nf_nat_alloc_null_binding(ct, manip);
 767
 768         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 769         if (err < 0)
 770                 return err;
 771
 772         return nf_nat_setup_info(ct, &range, manip);
 773 }
 774 #else
 775 static int
 776 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 777                           enum nf_nat_manip_type manip,
 778                           const struct nlattr *attr)
 779 {
 780         return -EOPNOTSUPP;
 781 }
 782 #endif
 783
 784 static int __net_init nf_nat_net_init(struct net *net)
 785 {
 786         /* Leave them the same for the moment. */
 787         net->ct.nat_htable_size = net->ct.htable_size;
 788         net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0);
 789         if (!net->ct.nat_bysource)
 790                 return -ENOMEM;
 791         return 0;
 792 }
 793
 794 static void __net_exit nf_nat_net_exit(struct net *net)
 795 {
 796         struct nf_nat_proto_clean clean = {};
 797
 798         nf_ct_iterate_cleanup(net, &nf_nat_proto_remove, &clean, 0, 0);
 799         synchronize_rcu();
 800         nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size);
 801 }
 802
 803 static struct pernet_operations nf_nat_net_ops = {
 804         .init = nf_nat_net_init,
 805         .exit = nf_nat_net_exit,
 806 };
 807
 808 static struct nf_ct_helper_expectfn follow_master_nat = {
 809         .name           = "nat-follow-master",
 810         .expectfn       = nf_nat_follow_master,
 811 };
 812
 813 static int __init nf_nat_init(void)
 814 {
 815         int ret;
 816
 817         ret = nf_ct_extend_register(&nat_extend);
 818         if (ret < 0) {
 819                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 820                 return ret;
 821         }
 822
 823         ret = register_pernet_subsys(&nf_nat_net_ops);
 824         if (ret < 0)
 825                 goto cleanup_extend;
 826
 827         nf_ct_helper_expectfn_register(&follow_master_nat);
 828
 829         /* Initialize fake conntrack so that NAT will skip it */
 830         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 831
 832         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 833         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 834                            nfnetlink_parse_nat_setup);
 835 #ifdef CONFIG_XFRM
 836         BUG_ON(nf_nat_decode_session_hook != NULL);
 837         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 838 #endif
 839         return 0;
 840
 841  cleanup_extend:
 842         nf_ct_extend_unregister(&nat_extend);
 843         return ret;
 844 }
 845
 846 static void __exit nf_nat_cleanup(void)
 847 {
 848         unsigned int i;
 849
 850         unregister_pernet_subsys(&nf_nat_net_ops);
 851         nf_ct_extend_unregister(&nat_extend);
 852         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 853         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 854 #ifdef CONFIG_XFRM
 855         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 856 #endif
 857         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 858                 kfree(nf_nat_l4protos[i]);
 859         synchronize_net();
 860 }
 861
 862 MODULE_LICENSE("GPL");
 863
 864 module_init(nf_nat_init);
 865 module_exit(nf_nat_cleanup);