kernel/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 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 inline unsigned int
 121 hash_by_src(const struct net *net, const struct nf_conntrack_tuple *tuple)
 122 {
 123         unsigned int hash;
 124
 125         /* Original src, to ensure we map it consistently if poss. */
 126         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 127                       tuple->dst.protonum ^ nf_conntrack_hash_rnd);
 128
 129         return reciprocal_scale(hash, net->ct.nat_htable_size);
 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,
 188                      const struct nf_conntrack_zone *zone,
 189                      const struct nf_nat_l3proto *l3proto,
 190                      const struct nf_nat_l4proto *l4proto,
 191                      const struct nf_conntrack_tuple *tuple,
 192                      struct nf_conntrack_tuple *result,
 193                      const struct nf_nat_range *range)
 194 {
 195         unsigned int h = hash_by_src(net, tuple);
 196         const struct nf_conn_nat *nat;
 197         const struct nf_conn *ct;
 198
 199         hlist_for_each_entry_rcu(nat, &net->ct.nat_bysource[h], bysource) {
 200                 ct = nat->ct;
 201                 if (same_src(ct, tuple) &&
 202                     nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) {
 203                         /* Copy source part from reply tuple. */
 204                         nf_ct_invert_tuplepr(result,
 205                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 206                         result->dst = tuple->dst;
 207
 208                         if (in_range(l3proto, l4proto, result, range))
 209                                 return 1;
 210                 }
 211         }
 212         return 0;
 213 }
 214
 215 /* For [FUTURE] fragmentation handling, we want the least-used
 216  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 217  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 218  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 219  * the ip with the lowest src-ip/dst-ip/proto usage.
 220  */
 221 static void
 222 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 223                     struct nf_conntrack_tuple *tuple,
 224                     const struct nf_nat_range *range,
 225                     const struct nf_conn *ct,
 226                     enum nf_nat_manip_type maniptype)
 227 {
 228         union nf_inet_addr *var_ipp;
 229         unsigned int i, max;
 230         /* Host order */
 231         u32 minip, maxip, j, dist;
 232         bool full_range;
 233
 234         /* No IP mapping?  Do nothing. */
 235         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 236                 return;
 237
 238         if (maniptype == NF_NAT_MANIP_SRC)
 239                 var_ipp = &tuple->src.u3;
 240         else
 241                 var_ipp = &tuple->dst.u3;
 242
 243         /* Fast path: only one choice. */
 244         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 245                 *var_ipp = range->min_addr;
 246                 return;
 247         }
 248
 249         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 250                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 251         else
 252                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 253
 254         /* Hashing source and destination IPs gives a fairly even
 255          * spread in practice (if there are a small number of IPs
 256          * involved, there usually aren't that many connections
 257          * anyway).  The consistency means that servers see the same
 258          * client coming from the same IP (some Internet Banking sites
 259          * like this), even across reboots.
 260          */
 261         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 262                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 263                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 264
 265         full_range = false;
 266         for (i = 0; i <= max; i++) {
 267                 /* If first bytes of the address are at the maximum, use the
 268                  * distance. Otherwise use the full range.
 269                  */
 270                 if (!full_range) {
 271                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 272                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 273                         dist  = maxip - minip + 1;
 274                 } else {
 275                         minip = 0;
 276                         dist  = ~0;
 277                 }
 278
 279                 var_ipp->all[i] = (__force __u32)
 280                         htonl(minip + reciprocal_scale(j, dist));
 281                 if (var_ipp->all[i] != range->max_addr.all[i])
 282                         full_range = true;
 283
 284                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 285                         j ^= (__force u32)tuple->dst.u3.all[i];
 286         }
 287 }
 288
 289 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 290  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 291  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 292  * range. It might not be possible to get a unique tuple, but we try.
 293  * At worst (or if we race), we will end up with a final duplicate in
 294  * __ip_conntrack_confirm and drop the packet. */
 295 static void
 296 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 297                  const struct nf_conntrack_tuple *orig_tuple,
 298                  const struct nf_nat_range *range,
 299                  struct nf_conn *ct,
 300                  enum nf_nat_manip_type maniptype)
 301 {
 302         const struct nf_conntrack_zone *zone;
 303         const struct nf_nat_l3proto *l3proto;
 304         const struct nf_nat_l4proto *l4proto;
 305         struct net *net = nf_ct_net(ct);
 306
 307         zone = nf_ct_zone(ct);
 308
 309         rcu_read_lock();
 310         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 311         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 312                                         orig_tuple->dst.protonum);
 313
 314         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 315          * and that same mapping gives a unique tuple within the given
 316          * range, use that.
 317          *
 318          * This is only required for source (ie. NAT/masq) mappings.
 319          * So far, we don't do local source mappings, so multiple
 320          * manips not an issue.
 321          */
 322         if (maniptype == NF_NAT_MANIP_SRC &&
 323             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 324                 /* try the original tuple first */
 325                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 326                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 327                                 *tuple = *orig_tuple;
 328                                 goto out;
 329                         }
 330                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 331                                                 orig_tuple, tuple, range)) {
 332                         pr_debug("get_unique_tuple: Found current src map\n");
 333                         if (!nf_nat_used_tuple(tuple, ct))
 334                                 goto out;
 335                 }
 336         }
 337
 338         /* 2) Select the least-used IP/proto combination in the given range */
 339         *tuple = *orig_tuple;
 340         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 341
 342         /* 3) The per-protocol part of the manip is made to map into
 343          * the range to make a unique tuple.
 344          */
 345
 346         /* Only bother mapping if it's not already in range and unique */
 347         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 348                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 349                         if (l4proto->in_range(tuple, maniptype,
 350                                               &range->min_proto,
 351                                               &range->max_proto) &&
 352                             (range->min_proto.all == range->max_proto.all ||
 353                              !nf_nat_used_tuple(tuple, ct)))
 354                                 goto out;
 355                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 356                         goto out;
 357                 }
 358         }
 359
 360         /* Last change: get protocol to try to obtain unique tuple. */
 361         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 362 out:
 363         rcu_read_unlock();
 364 }
 365
 366 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 367 {
 368         struct nf_conn_nat *nat = nfct_nat(ct);
 369         if (nat)
 370                 return nat;
 371
 372         if (!nf_ct_is_confirmed(ct))
 373                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 374
 375         return nat;
 376 }
 377 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 378
 379 unsigned int
 380 nf_nat_setup_info(struct nf_conn *ct,
 381                   const struct nf_nat_range *range,
 382                   enum nf_nat_manip_type maniptype)
 383 {
 384         struct net *net = nf_ct_net(ct);
 385         struct nf_conntrack_tuple curr_tuple, new_tuple;
 386         struct nf_conn_nat *nat;
 387
 388         /* nat helper or nfctnetlink also setup binding */
 389         nat = nf_ct_nat_ext_add(ct);
 390         if (nat == NULL)
 391                 return NF_ACCEPT;
 392
 393         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 394                      maniptype == NF_NAT_MANIP_DST);
 395         BUG_ON(nf_nat_initialized(ct, maniptype));
 396
 397         /* What we've got will look like inverse of reply. Normally
 398          * this is what is in the conntrack, except for prior
 399          * manipulations (future optimization: if num_manips == 0,
 400          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 401          */
 402         nf_ct_invert_tuplepr(&curr_tuple,
 403                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 404
 405         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 406
 407         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 408                 struct nf_conntrack_tuple reply;
 409
 410                 /* Alter conntrack table so will recognize replies. */
 411                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 412                 nf_conntrack_alter_reply(ct, &reply);
 413
 414                 /* Non-atomic: we own this at the moment. */
 415                 if (maniptype == NF_NAT_MANIP_SRC)
 416                         ct->status |= IPS_SRC_NAT;
 417                 else
 418                         ct->status |= IPS_DST_NAT;
 419
 420                 if (nfct_help(ct))
 421                         nfct_seqadj_ext_add(ct);
 422         }
 423
 424         if (maniptype == NF_NAT_MANIP_SRC) {
 425                 unsigned int srchash;
 426
 427                 srchash = hash_by_src(net,
 428                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 429                 spin_lock_bh(&nf_nat_lock);
 430                 /* nf_conntrack_alter_reply might re-allocate extension aera */
 431                 nat = nfct_nat(ct);
 432                 nat->ct = ct;
 433                 hlist_add_head_rcu(&nat->bysource,
 434                                    &net->ct.nat_bysource[srchash]);
 435                 spin_unlock_bh(&nf_nat_lock);
 436         }
 437
 438         /* It's done. */
 439         if (maniptype == NF_NAT_MANIP_DST)
 440                 ct->status |= IPS_DST_NAT_DONE;
 441         else
 442                 ct->status |= IPS_SRC_NAT_DONE;
 443
 444         return NF_ACCEPT;
 445 }
 446 EXPORT_SYMBOL(nf_nat_setup_info);
 447
 448 static unsigned int
 449 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 450 {
 451         /* Force range to this IP; let proto decide mapping for
 452          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 453          * Use reply in case it's already been mangled (eg local packet).
 454          */
 455         union nf_inet_addr ip =
 456                 (manip == NF_NAT_MANIP_SRC ?
 457                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 458                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 459         struct nf_nat_range range = {
 460                 .flags          = NF_NAT_RANGE_MAP_IPS,
 461                 .min_addr       = ip,
 462                 .max_addr       = ip,
 463         };
 464         return nf_nat_setup_info(ct, &range, manip);
 465 }
 466
 467 unsigned int
 468 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 469 {
 470         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 471 }
 472 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 473
 474 /* Do packet manipulations according to nf_nat_setup_info. */
 475 unsigned int nf_nat_packet(struct nf_conn *ct,
 476                            enum ip_conntrack_info ctinfo,
 477                            unsigned int hooknum,
 478                            struct sk_buff *skb)
 479 {
 480         const struct nf_nat_l3proto *l3proto;
 481         const struct nf_nat_l4proto *l4proto;
 482         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 483         unsigned long statusbit;
 484         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 485
 486         if (mtype == NF_NAT_MANIP_SRC)
 487                 statusbit = IPS_SRC_NAT;
 488         else
 489                 statusbit = IPS_DST_NAT;
 490
 491         /* Invert if this is reply dir. */
 492         if (dir == IP_CT_DIR_REPLY)
 493                 statusbit ^= IPS_NAT_MASK;
 494
 495         /* Non-atomic: these bits don't change. */
 496         if (ct->status & statusbit) {
 497                 struct nf_conntrack_tuple target;
 498
 499                 /* We are aiming to look like inverse of other direction. */
 500                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 501
 502                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 503                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 504                                                 target.dst.protonum);
 505                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 506                         return NF_DROP;
 507         }
 508         return NF_ACCEPT;
 509 }
 510 EXPORT_SYMBOL_GPL(nf_nat_packet);
 511
 512 struct nf_nat_proto_clean {
 513         u8      l3proto;
 514         u8      l4proto;
 515 };
 516
 517 /* kill conntracks with affected NAT section */
 518 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 519 {
 520         const struct nf_nat_proto_clean *clean = data;
 521         struct nf_conn_nat *nat = nfct_nat(i);
 522
 523         if (!nat)
 524                 return 0;
 525
 526         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 527             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 528                 return 0;
 529
 530         return i->status & IPS_NAT_MASK ? 1 : 0;
 531 }
 532
 533 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 534 {
 535         struct nf_conn_nat *nat = nfct_nat(ct);
 536
 537         if (nf_nat_proto_remove(ct, data))
 538                 return 1;
 539
 540         if (!nat || !nat->ct)
 541                 return 0;
 542
 543         /* This netns is being destroyed, and conntrack has nat null binding.
 544          * Remove it from bysource hash, as the table will be freed soon.
 545          *
 546          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 547          * will delete entry from already-freed table.
 548          */
 549         if (!del_timer(&ct->timeout))
 550                 return 1;
 551
 552         spin_lock_bh(&nf_nat_lock);
 553         hlist_del_rcu(&nat->bysource);
 554         ct->status &= ~IPS_NAT_DONE_MASK;
 555         nat->ct = NULL;
 556         spin_unlock_bh(&nf_nat_lock);
 557
 558         add_timer(&ct->timeout);
 559
 560         /* don't delete conntrack.  Although that would make things a lot
 561          * simpler, we'd end up flushing all conntracks on nat rmmod.
 562          */
 563         return 0;
 564 }
 565
 566 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 567 {
 568         struct nf_nat_proto_clean clean = {
 569                 .l3proto = l3proto,
 570                 .l4proto = l4proto,
 571         };
 572         struct net *net;
 573
 574         rtnl_lock();
 575         for_each_net(net)
 576                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 577         rtnl_unlock();
 578 }
 579
 580 static void nf_nat_l3proto_clean(u8 l3proto)
 581 {
 582         struct nf_nat_proto_clean clean = {
 583                 .l3proto = l3proto,
 584         };
 585         struct net *net;
 586
 587         rtnl_lock();
 588
 589         for_each_net(net)
 590                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 591         rtnl_unlock();
 592 }
 593
 594 /* Protocol registration. */
 595 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 596 {
 597         const struct nf_nat_l4proto **l4protos;
 598         unsigned int i;
 599         int ret = 0;
 600
 601         mutex_lock(&nf_nat_proto_mutex);
 602         if (nf_nat_l4protos[l3proto] == NULL) {
 603                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 604                                    GFP_KERNEL);
 605                 if (l4protos == NULL) {
 606                         ret = -ENOMEM;
 607                         goto out;
 608                 }
 609
 610                 for (i = 0; i < IPPROTO_MAX; i++)
 611                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 612
 613                 /* Before making proto_array visible to lockless readers,
 614                  * we must make sure its content is committed to memory.
 615                  */
 616                 smp_wmb();
 617
 618                 nf_nat_l4protos[l3proto] = l4protos;
 619         }
 620
 621         if (rcu_dereference_protected(
 622                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 623                         lockdep_is_held(&nf_nat_proto_mutex)
 624                         ) != &nf_nat_l4proto_unknown) {
 625                 ret = -EBUSY;
 626                 goto out;
 627         }
 628         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 629  out:
 630         mutex_unlock(&nf_nat_proto_mutex);
 631         return ret;
 632 }
 633 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 634
 635 /* No one stores the protocol anywhere; simply delete it. */
 636 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 637 {
 638         mutex_lock(&nf_nat_proto_mutex);
 639         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 640                          &nf_nat_l4proto_unknown);
 641         mutex_unlock(&nf_nat_proto_mutex);
 642         synchronize_rcu();
 643
 644         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 645 }
 646 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 647
 648 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 649 {
 650         int err;
 651
 652         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 653         if (err < 0)
 654                 return err;
 655
 656         mutex_lock(&nf_nat_proto_mutex);
 657         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 658                          &nf_nat_l4proto_tcp);
 659         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 660                          &nf_nat_l4proto_udp);
 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         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 684
 685         if (nat == NULL || nat->ct == NULL)
 686                 return;
 687
 688         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 689
 690         spin_lock_bh(&nf_nat_lock);
 691         hlist_del_rcu(&nat->bysource);
 692         spin_unlock_bh(&nf_nat_lock);
 693 }
 694
 695 static void nf_nat_move_storage(void *new, void *old)
 696 {
 697         struct nf_conn_nat *new_nat = new;
 698         struct nf_conn_nat *old_nat = old;
 699         struct nf_conn *ct = old_nat->ct;
 700
 701         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 702                 return;
 703
 704         spin_lock_bh(&nf_nat_lock);
 705         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 706         spin_unlock_bh(&nf_nat_lock);
 707 }
 708
 709 static struct nf_ct_ext_type nat_extend __read_mostly = {
 710         .len            = sizeof(struct nf_conn_nat),
 711         .align          = __alignof__(struct nf_conn_nat),
 712         .destroy        = nf_nat_cleanup_conntrack,
 713         .move           = nf_nat_move_storage,
 714         .id             = NF_CT_EXT_NAT,
 715         .flags          = NF_CT_EXT_F_PREALLOC,
 716 };
 717
 718 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 719
 720 #include <linux/netfilter/nfnetlink.h>
 721 #include <linux/netfilter/nfnetlink_conntrack.h>
 722
 723 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 724         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 725         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 726 };
 727
 728 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 729                                      const struct nf_conn *ct,
 730                                      struct nf_nat_range *range)
 731 {
 732         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 733         const struct nf_nat_l4proto *l4proto;
 734         int err;
 735
 736         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 737         if (err < 0)
 738                 return err;
 739
 740         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 741         if (l4proto->nlattr_to_range)
 742                 err = l4proto->nlattr_to_range(tb, range);
 743
 744         return err;
 745 }
 746
 747 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 748         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 749         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 750         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 751         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 752         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 753 };
 754
 755 static int
 756 nfnetlink_parse_nat(const struct nlattr *nat,
 757                     const struct nf_conn *ct, struct nf_nat_range *range,
 758                     const struct nf_nat_l3proto *l3proto)
 759 {
 760         struct nlattr *tb[CTA_NAT_MAX+1];
 761         int err;
 762
 763         memset(range, 0, sizeof(*range));
 764
 765         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 766         if (err < 0)
 767                 return err;
 768
 769         err = l3proto->nlattr_to_range(tb, range);
 770         if (err < 0)
 771                 return err;
 772
 773         if (!tb[CTA_NAT_PROTO])
 774                 return 0;
 775
 776         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 777 }
 778
 779 /* This function is called under rcu_read_lock() */
 780 static int
 781 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 782                           enum nf_nat_manip_type manip,
 783                           const struct nlattr *attr)
 784 {
 785         struct nf_nat_range range;
 786         const struct nf_nat_l3proto *l3proto;
 787         int err;
 788
 789         /* Should not happen, restricted to creating new conntracks
 790          * via ctnetlink.
 791          */
 792         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 793                 return -EEXIST;
 794
 795         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 796          * attach the null binding, otherwise this may oops.
 797          */
 798         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 799         if (l3proto == NULL)
 800                 return -EAGAIN;
 801
 802         /* No NAT information has been passed, allocate the null-binding */
 803         if (attr == NULL)
 804                 return __nf_nat_alloc_null_binding(ct, manip);
 805
 806         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 807         if (err < 0)
 808                 return err;
 809
 810         return nf_nat_setup_info(ct, &range, manip);
 811 }
 812 #else
 813 static int
 814 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 815                           enum nf_nat_manip_type manip,
 816                           const struct nlattr *attr)
 817 {
 818         return -EOPNOTSUPP;
 819 }
 820 #endif
 821
 822 static int __net_init nf_nat_net_init(struct net *net)
 823 {
 824         /* Leave them the same for the moment. */
 825         net->ct.nat_htable_size = net->ct.htable_size;
 826         net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0);
 827         if (!net->ct.nat_bysource)
 828                 return -ENOMEM;
 829         return 0;
 830 }
 831
 832 static void __net_exit nf_nat_net_exit(struct net *net)
 833 {
 834         struct nf_nat_proto_clean clean = {};
 835
 836         nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean, 0, 0);
 837         synchronize_rcu();
 838         nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size);
 839 }
 840
 841 static struct pernet_operations nf_nat_net_ops = {
 842         .init = nf_nat_net_init,
 843         .exit = nf_nat_net_exit,
 844 };
 845
 846 static struct nf_ct_helper_expectfn follow_master_nat = {
 847         .name           = "nat-follow-master",
 848         .expectfn       = nf_nat_follow_master,
 849 };
 850
 851 static int __init nf_nat_init(void)
 852 {
 853         int ret;
 854
 855         ret = nf_ct_extend_register(&nat_extend);
 856         if (ret < 0) {
 857                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 858                 return ret;
 859         }
 860
 861         ret = register_pernet_subsys(&nf_nat_net_ops);
 862         if (ret < 0)
 863                 goto cleanup_extend;
 864
 865         nf_ct_helper_expectfn_register(&follow_master_nat);
 866
 867         /* Initialize fake conntrack so that NAT will skip it */
 868         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 869
 870         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 871         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 872                            nfnetlink_parse_nat_setup);
 873 #ifdef CONFIG_XFRM
 874         BUG_ON(nf_nat_decode_session_hook != NULL);
 875         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 876 #endif
 877         return 0;
 878
 879  cleanup_extend:
 880         nf_ct_extend_unregister(&nat_extend);
 881         return ret;
 882 }
 883
 884 static void __exit nf_nat_cleanup(void)
 885 {
 886         unsigned int i;
 887
 888         unregister_pernet_subsys(&nf_nat_net_ops);
 889         nf_ct_extend_unregister(&nat_extend);
 890         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 891         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 892 #ifdef CONFIG_XFRM
 893         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 894 #endif
 895         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 896                 kfree(nf_nat_l4protos[i]);
 897         synchronize_net();
 898 }
 899
 900 MODULE_LICENSE("GPL");
 901
 902 module_init(nf_nat_init);
 903 module_exit(nf_nat_cleanup);