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_MUTEX(nf_nat_proto_mutex);
  34 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
  35                                                 __read_mostly;
  36 static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO]
  37                                                 __read_mostly;
  38
  39 struct nf_nat_conn_key {
  40         const struct net *net;
  41         const struct nf_conntrack_tuple *tuple;
  42         const struct nf_conntrack_zone *zone;
  43 };
  44
  45 static struct rhltable nf_nat_bysource_table;
  46
  47 inline const struct nf_nat_l3proto *
  48 __nf_nat_l3proto_find(u8 family)
  49 {
  50         return rcu_dereference(nf_nat_l3protos[family]);
  51 }
  52
  53 inline const struct nf_nat_l4proto *
  54 __nf_nat_l4proto_find(u8 family, u8 protonum)
  55 {
  56         return rcu_dereference(nf_nat_l4protos[family][protonum]);
  57 }
  58 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
  59
  60 #ifdef CONFIG_XFRM
  61 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  62 {
  63         const struct nf_nat_l3proto *l3proto;
  64         const struct nf_conn *ct;
  65         enum ip_conntrack_info ctinfo;
  66         enum ip_conntrack_dir dir;
  67         unsigned  long statusbit;
  68         u8 family;
  69
  70         ct = nf_ct_get(skb, &ctinfo);
  71         if (ct == NULL)
  72                 return;
  73
  74         family = nf_ct_l3num(ct);
  75         l3proto = __nf_nat_l3proto_find(family);
  76         if (l3proto == NULL)
  77                 return;
  78
  79         dir = CTINFO2DIR(ctinfo);
  80         if (dir == IP_CT_DIR_ORIGINAL)
  81                 statusbit = IPS_DST_NAT;
  82         else
  83                 statusbit = IPS_SRC_NAT;
  84
  85         l3proto->decode_session(skb, ct, dir, statusbit, fl);
  86 }
  87
  88 int nf_xfrm_me_harder(struct net *net, struct sk_buff *skb, unsigned int family)
  89 {
  90         struct flowi fl;
  91         unsigned int hh_len;
  92         struct dst_entry *dst;
  93         int err;
  94
  95         err = xfrm_decode_session(skb, &fl, family);
  96         if (err < 0)
  97                 return err;
  98
  99         dst = skb_dst(skb);
 100         if (dst->xfrm)
 101                 dst = ((struct xfrm_dst *)dst)->route;
 102         dst_hold(dst);
 103
 104         dst = xfrm_lookup(net, dst, &fl, skb->sk, 0);
 105         if (IS_ERR(dst))
 106                 return PTR_ERR(dst);
 107
 108         skb_dst_drop(skb);
 109         skb_dst_set(skb, dst);
 110
 111         /* Change in oif may mean change in hh_len. */
 112         hh_len = skb_dst(skb)->dev->hard_header_len;
 113         if (skb_headroom(skb) < hh_len &&
 114             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 115                 return -ENOMEM;
 116         return 0;
 117 }
 118 EXPORT_SYMBOL(nf_xfrm_me_harder);
 119 #endif /* CONFIG_XFRM */
 120
 121 static u32 nf_nat_bysource_hash(const void *data, u32 len, u32 seed)
 122 {
 123         const struct nf_conntrack_tuple *t;
 124         const struct nf_conn *ct = data;
 125
 126         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 127         /* Original src, to ensure we map it consistently if poss. */
 128
 129         seed ^= net_hash_mix(nf_ct_net(ct));
 130         return jhash2((const u32 *)&t->src, sizeof(t->src) / sizeof(u32),
 131                       t->dst.protonum ^ seed);
 132 }
 133
 134 /* Is this tuple already taken? (not by us) */
 135 int
 136 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 137                   const struct nf_conn *ignored_conntrack)
 138 {
 139         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 140          * incoming ones.  NAT means they don't have a fixed mapping,
 141          * so we invert the tuple and look for the incoming reply.
 142          *
 143          * We could keep a separate hash if this proves too slow.
 144          */
 145         struct nf_conntrack_tuple reply;
 146
 147         nf_ct_invert_tuplepr(&reply, tuple);
 148         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 149 }
 150 EXPORT_SYMBOL(nf_nat_used_tuple);
 151
 152 /* If we source map this tuple so reply looks like reply_tuple, will
 153  * that meet the constraints of range.
 154  */
 155 static int in_range(const struct nf_nat_l3proto *l3proto,
 156                     const struct nf_nat_l4proto *l4proto,
 157                     const struct nf_conntrack_tuple *tuple,
 158                     const struct nf_nat_range *range)
 159 {
 160         /* If we are supposed to map IPs, then we must be in the
 161          * range specified, otherwise let this drag us onto a new src IP.
 162          */
 163         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 164             !l3proto->in_range(tuple, range))
 165                 return 0;
 166
 167         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 168             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 169                               &range->min_proto, &range->max_proto))
 170                 return 1;
 171
 172         return 0;
 173 }
 174
 175 static inline int
 176 same_src(const struct nf_conn *ct,
 177          const struct nf_conntrack_tuple *tuple)
 178 {
 179         const struct nf_conntrack_tuple *t;
 180
 181         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 182         return (t->dst.protonum == tuple->dst.protonum &&
 183                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 184                 t->src.u.all == tuple->src.u.all);
 185 }
 186
 187 static int nf_nat_bysource_cmp(struct rhashtable_compare_arg *arg,
 188                                const void *obj)
 189 {
 190         const struct nf_nat_conn_key *key = arg->key;
 191         const struct nf_conn *ct = obj;
 192
 193         if (!same_src(ct, key->tuple) ||
 194             !net_eq(nf_ct_net(ct), key->net) ||
 195             !nf_ct_zone_equal(ct, key->zone, IP_CT_DIR_ORIGINAL))
 196                 return 1;
 197
 198         return 0;
 199 }
 200
 201 static struct rhashtable_params nf_nat_bysource_params = {
 202         .head_offset = offsetof(struct nf_conn, nat_bysource),
 203         .obj_hashfn = nf_nat_bysource_hash,
 204         .obj_cmpfn = nf_nat_bysource_cmp,
 205         .nelem_hint = 256,
 206         .min_size = 1024,
 207 };
 208
 209 /* Only called for SRC manip */
 210 static int
 211 find_appropriate_src(struct net *net,
 212                      const struct nf_conntrack_zone *zone,
 213                      const struct nf_nat_l3proto *l3proto,
 214                      const struct nf_nat_l4proto *l4proto,
 215                      const struct nf_conntrack_tuple *tuple,
 216                      struct nf_conntrack_tuple *result,
 217                      const struct nf_nat_range *range)
 218 {
 219         const struct nf_conn *ct;
 220         struct nf_nat_conn_key key = {
 221                 .net = net,
 222                 .tuple = tuple,
 223                 .zone = zone
 224         };
 225         struct rhlist_head *hl, *h;
 226
 227         hl = rhltable_lookup(&nf_nat_bysource_table, &key,
 228                              nf_nat_bysource_params);
 229
 230         rhl_for_each_entry_rcu(ct, h, hl, nat_bysource) {
 231                 nf_ct_invert_tuplepr(result,
 232                                      &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 233                 result->dst = tuple->dst;
 234
 235                 if (in_range(l3proto, l4proto, result, range))
 236                         return 1;
 237         }
 238
 239         return 0;
 240 }
 241
 242 /* For [FUTURE] fragmentation handling, we want the least-used
 243  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 244  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 245  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 246  * the ip with the lowest src-ip/dst-ip/proto usage.
 247  */
 248 static void
 249 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 250                     struct nf_conntrack_tuple *tuple,
 251                     const struct nf_nat_range *range,
 252                     const struct nf_conn *ct,
 253                     enum nf_nat_manip_type maniptype)
 254 {
 255         union nf_inet_addr *var_ipp;
 256         unsigned int i, max;
 257         /* Host order */
 258         u32 minip, maxip, j, dist;
 259         bool full_range;
 260
 261         /* No IP mapping?  Do nothing. */
 262         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 263                 return;
 264
 265         if (maniptype == NF_NAT_MANIP_SRC)
 266                 var_ipp = &tuple->src.u3;
 267         else
 268                 var_ipp = &tuple->dst.u3;
 269
 270         /* Fast path: only one choice. */
 271         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 272                 *var_ipp = range->min_addr;
 273                 return;
 274         }
 275
 276         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 277                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 278         else
 279                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 280
 281         /* Hashing source and destination IPs gives a fairly even
 282          * spread in practice (if there are a small number of IPs
 283          * involved, there usually aren't that many connections
 284          * anyway).  The consistency means that servers see the same
 285          * client coming from the same IP (some Internet Banking sites
 286          * like this), even across reboots.
 287          */
 288         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 289                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 290                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 291
 292         full_range = false;
 293         for (i = 0; i <= max; i++) {
 294                 /* If first bytes of the address are at the maximum, use the
 295                  * distance. Otherwise use the full range.
 296                  */
 297                 if (!full_range) {
 298                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 299                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 300                         dist  = maxip - minip + 1;
 301                 } else {
 302                         minip = 0;
 303                         dist  = ~0;
 304                 }
 305
 306                 var_ipp->all[i] = (__force __u32)
 307                         htonl(minip + reciprocal_scale(j, dist));
 308                 if (var_ipp->all[i] != range->max_addr.all[i])
 309                         full_range = true;
 310
 311                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 312                         j ^= (__force u32)tuple->dst.u3.all[i];
 313         }
 314 }
 315
 316 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 317  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 318  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 319  * range. It might not be possible to get a unique tuple, but we try.
 320  * At worst (or if we race), we will end up with a final duplicate in
 321  * __ip_conntrack_confirm and drop the packet. */
 322 static void
 323 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 324                  const struct nf_conntrack_tuple *orig_tuple,
 325                  const struct nf_nat_range *range,
 326                  struct nf_conn *ct,
 327                  enum nf_nat_manip_type maniptype)
 328 {
 329         const struct nf_conntrack_zone *zone;
 330         const struct nf_nat_l3proto *l3proto;
 331         const struct nf_nat_l4proto *l4proto;
 332         struct net *net = nf_ct_net(ct);
 333
 334         zone = nf_ct_zone(ct);
 335
 336         rcu_read_lock();
 337         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 338         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 339                                         orig_tuple->dst.protonum);
 340
 341         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 342          * and that same mapping gives a unique tuple within the given
 343          * range, use that.
 344          *
 345          * This is only required for source (ie. NAT/masq) mappings.
 346          * So far, we don't do local source mappings, so multiple
 347          * manips not an issue.
 348          */
 349         if (maniptype == NF_NAT_MANIP_SRC &&
 350             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 351                 /* try the original tuple first */
 352                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 353                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 354                                 *tuple = *orig_tuple;
 355                                 goto out;
 356                         }
 357                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 358                                                 orig_tuple, tuple, range)) {
 359                         pr_debug("get_unique_tuple: Found current src map\n");
 360                         if (!nf_nat_used_tuple(tuple, ct))
 361                                 goto out;
 362                 }
 363         }
 364
 365         /* 2) Select the least-used IP/proto combination in the given range */
 366         *tuple = *orig_tuple;
 367         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 368
 369         /* 3) The per-protocol part of the manip is made to map into
 370          * the range to make a unique tuple.
 371          */
 372
 373         /* Only bother mapping if it's not already in range and unique */
 374         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 375                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 376                         if (l4proto->in_range(tuple, maniptype,
 377                                               &range->min_proto,
 378                                               &range->max_proto) &&
 379                             (range->min_proto.all == range->max_proto.all ||
 380                              !nf_nat_used_tuple(tuple, ct)))
 381                                 goto out;
 382                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 383                         goto out;
 384                 }
 385         }
 386
 387         /* Last change: get protocol to try to obtain unique tuple. */
 388         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 389 out:
 390         rcu_read_unlock();
 391 }
 392
 393 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 394 {
 395         struct nf_conn_nat *nat = nfct_nat(ct);
 396         if (nat)
 397                 return nat;
 398
 399         if (!nf_ct_is_confirmed(ct))
 400                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 401
 402         return nat;
 403 }
 404 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 405
 406 unsigned int
 407 nf_nat_setup_info(struct nf_conn *ct,
 408                   const struct nf_nat_range *range,
 409                   enum nf_nat_manip_type maniptype)
 410 {
 411         struct nf_conntrack_tuple curr_tuple, new_tuple;
 412
 413         /* Can't setup nat info for confirmed ct. */
 414         if (nf_ct_is_confirmed(ct))
 415                 return NF_ACCEPT;
 416
 417         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 418                      maniptype == NF_NAT_MANIP_DST);
 419         BUG_ON(nf_nat_initialized(ct, maniptype));
 420
 421         /* What we've got will look like inverse of reply. Normally
 422          * this is what is in the conntrack, except for prior
 423          * manipulations (future optimization: if num_manips == 0,
 424          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 425          */
 426         nf_ct_invert_tuplepr(&curr_tuple,
 427                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 428
 429         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 430
 431         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 432                 struct nf_conntrack_tuple reply;
 433
 434                 /* Alter conntrack table so will recognize replies. */
 435                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 436                 nf_conntrack_alter_reply(ct, &reply);
 437
 438                 /* Non-atomic: we own this at the moment. */
 439                 if (maniptype == NF_NAT_MANIP_SRC)
 440                         ct->status |= IPS_SRC_NAT;
 441                 else
 442                         ct->status |= IPS_DST_NAT;
 443
 444                 if (nfct_help(ct))
 445                         if (!nfct_seqadj_ext_add(ct))
 446                                 return NF_DROP;
 447         }
 448
 449         if (maniptype == NF_NAT_MANIP_SRC) {
 450                 struct nf_nat_conn_key key = {
 451                         .net = nf_ct_net(ct),
 452                         .tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 453                         .zone = nf_ct_zone(ct),
 454                 };
 455                 int err;
 456
 457                 err = rhltable_insert_key(&nf_nat_bysource_table,
 458                                           &key,
 459                                           &ct->nat_bysource,
 460                                           nf_nat_bysource_params);
 461                 if (err)
 462                         return NF_DROP;
 463         }
 464
 465         /* It's done. */
 466         if (maniptype == NF_NAT_MANIP_DST)
 467                 ct->status |= IPS_DST_NAT_DONE;
 468         else
 469                 ct->status |= IPS_SRC_NAT_DONE;
 470
 471         return NF_ACCEPT;
 472 }
 473 EXPORT_SYMBOL(nf_nat_setup_info);
 474
 475 static unsigned int
 476 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 477 {
 478         /* Force range to this IP; let proto decide mapping for
 479          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 480          * Use reply in case it's already been mangled (eg local packet).
 481          */
 482         union nf_inet_addr ip =
 483                 (manip == NF_NAT_MANIP_SRC ?
 484                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 485                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 486         struct nf_nat_range range = {
 487                 .flags          = NF_NAT_RANGE_MAP_IPS,
 488                 .min_addr       = ip,
 489                 .max_addr       = ip,
 490         };
 491         return nf_nat_setup_info(ct, &range, manip);
 492 }
 493
 494 unsigned int
 495 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 496 {
 497         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 498 }
 499 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 500
 501 /* Do packet manipulations according to nf_nat_setup_info. */
 502 unsigned int nf_nat_packet(struct nf_conn *ct,
 503                            enum ip_conntrack_info ctinfo,
 504                            unsigned int hooknum,
 505                            struct sk_buff *skb)
 506 {
 507         const struct nf_nat_l3proto *l3proto;
 508         const struct nf_nat_l4proto *l4proto;
 509         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 510         unsigned long statusbit;
 511         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 512
 513         if (mtype == NF_NAT_MANIP_SRC)
 514                 statusbit = IPS_SRC_NAT;
 515         else
 516                 statusbit = IPS_DST_NAT;
 517
 518         /* Invert if this is reply dir. */
 519         if (dir == IP_CT_DIR_REPLY)
 520                 statusbit ^= IPS_NAT_MASK;
 521
 522         /* Non-atomic: these bits don't change. */
 523         if (ct->status & statusbit) {
 524                 struct nf_conntrack_tuple target;
 525
 526                 /* We are aiming to look like inverse of other direction. */
 527                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 528
 529                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 530                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 531                                                 target.dst.protonum);
 532                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 533                         return NF_DROP;
 534         }
 535         return NF_ACCEPT;
 536 }
 537 EXPORT_SYMBOL_GPL(nf_nat_packet);
 538
 539 struct nf_nat_proto_clean {
 540         u8      l3proto;
 541         u8      l4proto;
 542 };
 543
 544 /* kill conntracks with affected NAT section */
 545 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 546 {
 547         const struct nf_nat_proto_clean *clean = data;
 548
 549         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 550             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 551                 return 0;
 552
 553         return i->status & IPS_NAT_MASK ? 1 : 0;
 554 }
 555
 556 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 557 {
 558         if (nf_nat_proto_remove(ct, data))
 559                 return 1;
 560
 561         if ((ct->status & IPS_SRC_NAT_DONE) == 0)
 562                 return 0;
 563
 564         /* This netns is being destroyed, and conntrack has nat null binding.
 565          * Remove it from bysource hash, as the table will be freed soon.
 566          *
 567          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 568          * will delete entry from already-freed table.
 569          */
 570         clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 571         rhltable_remove(&nf_nat_bysource_table, &ct->nat_bysource,
 572                         nf_nat_bysource_params);
 573
 574         /* don't delete conntrack.  Although that would make things a lot
 575          * simpler, we'd end up flushing all conntracks on nat rmmod.
 576          */
 577         return 0;
 578 }
 579
 580 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 581 {
 582         struct nf_nat_proto_clean clean = {
 583                 .l3proto = l3proto,
 584                 .l4proto = l4proto,
 585         };
 586
 587         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 588 }
 589
 590 static void nf_nat_l3proto_clean(u8 l3proto)
 591 {
 592         struct nf_nat_proto_clean clean = {
 593                 .l3proto = l3proto,
 594         };
 595
 596         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 597 }
 598
 599 /* Protocol registration. */
 600 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 601 {
 602         const struct nf_nat_l4proto **l4protos;
 603         unsigned int i;
 604         int ret = 0;
 605
 606         mutex_lock(&nf_nat_proto_mutex);
 607         if (nf_nat_l4protos[l3proto] == NULL) {
 608                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 609                                    GFP_KERNEL);
 610                 if (l4protos == NULL) {
 611                         ret = -ENOMEM;
 612                         goto out;
 613                 }
 614
 615                 for (i = 0; i < IPPROTO_MAX; i++)
 616                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 617
 618                 /* Before making proto_array visible to lockless readers,
 619                  * we must make sure its content is committed to memory.
 620                  */
 621                 smp_wmb();
 622
 623                 nf_nat_l4protos[l3proto] = l4protos;
 624         }
 625
 626         if (rcu_dereference_protected(
 627                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 628                         lockdep_is_held(&nf_nat_proto_mutex)
 629                         ) != &nf_nat_l4proto_unknown) {
 630                 ret = -EBUSY;
 631                 goto out;
 632         }
 633         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 634  out:
 635         mutex_unlock(&nf_nat_proto_mutex);
 636         return ret;
 637 }
 638 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 639
 640 /* No one stores the protocol anywhere; simply delete it. */
 641 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 642 {
 643         mutex_lock(&nf_nat_proto_mutex);
 644         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 645                          &nf_nat_l4proto_unknown);
 646         mutex_unlock(&nf_nat_proto_mutex);
 647         synchronize_rcu();
 648
 649         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 650 }
 651 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 652
 653 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 654 {
 655         int err;
 656
 657         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 658         if (err < 0)
 659                 return err;
 660
 661         mutex_lock(&nf_nat_proto_mutex);
 662         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 663                          &nf_nat_l4proto_tcp);
 664         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 665                          &nf_nat_l4proto_udp);
 666 #ifdef CONFIG_NF_NAT_PROTO_DCCP
 667         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_DCCP],
 668                          &nf_nat_l4proto_dccp);
 669 #endif
 670 #ifdef CONFIG_NF_NAT_PROTO_SCTP
 671         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_SCTP],
 672                          &nf_nat_l4proto_sctp);
 673 #endif
 674 #ifdef CONFIG_NF_NAT_PROTO_UDPLITE
 675         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDPLITE],
 676                          &nf_nat_l4proto_udplite);
 677 #endif
 678         mutex_unlock(&nf_nat_proto_mutex);
 679
 680         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 681         return 0;
 682 }
 683 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 684
 685 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 686 {
 687         mutex_lock(&nf_nat_proto_mutex);
 688         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 689         mutex_unlock(&nf_nat_proto_mutex);
 690         synchronize_rcu();
 691
 692         nf_nat_l3proto_clean(l3proto->l3proto);
 693         nf_ct_l3proto_module_put(l3proto->l3proto);
 694 }
 695 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 696
 697 /* No one using conntrack by the time this called. */
 698 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 699 {
 700         if (ct->status & IPS_SRC_NAT_DONE)
 701                 rhltable_remove(&nf_nat_bysource_table, &ct->nat_bysource,
 702                                 nf_nat_bysource_params);
 703 }
 704
 705 static struct nf_ct_ext_type nat_extend __read_mostly = {
 706         .len            = sizeof(struct nf_conn_nat),
 707         .align          = __alignof__(struct nf_conn_nat),
 708         .destroy        = nf_nat_cleanup_conntrack,
 709         .id             = NF_CT_EXT_NAT,
 710 };
 711
 712 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 713
 714 #include <linux/netfilter/nfnetlink.h>
 715 #include <linux/netfilter/nfnetlink_conntrack.h>
 716
 717 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 718         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 719         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 720 };
 721
 722 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 723                                      const struct nf_conn *ct,
 724                                      struct nf_nat_range *range)
 725 {
 726         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 727         const struct nf_nat_l4proto *l4proto;
 728         int err;
 729
 730         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr,
 731                                protonat_nla_policy, NULL);
 732         if (err < 0)
 733                 return err;
 734
 735         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 736         if (l4proto->nlattr_to_range)
 737                 err = l4proto->nlattr_to_range(tb, range);
 738
 739         return err;
 740 }
 741
 742 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 743         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 744         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 745         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 746         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 747         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 748 };
 749
 750 static int
 751 nfnetlink_parse_nat(const struct nlattr *nat,
 752                     const struct nf_conn *ct, struct nf_nat_range *range,
 753                     const struct nf_nat_l3proto *l3proto)
 754 {
 755         struct nlattr *tb[CTA_NAT_MAX+1];
 756         int err;
 757
 758         memset(range, 0, sizeof(*range));
 759
 760         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy, NULL);
 761         if (err < 0)
 762                 return err;
 763
 764         err = l3proto->nlattr_to_range(tb, range);
 765         if (err < 0)
 766                 return err;
 767
 768         if (!tb[CTA_NAT_PROTO])
 769                 return 0;
 770
 771         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 772 }
 773
 774 /* This function is called under rcu_read_lock() */
 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         struct nf_nat_range range;
 781         const struct nf_nat_l3proto *l3proto;
 782         int err;
 783
 784         /* Should not happen, restricted to creating new conntracks
 785          * via ctnetlink.
 786          */
 787         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 788                 return -EEXIST;
 789
 790         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 791          * attach the null binding, otherwise this may oops.
 792          */
 793         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 794         if (l3proto == NULL)
 795                 return -EAGAIN;
 796
 797         /* No NAT information has been passed, allocate the null-binding */
 798         if (attr == NULL)
 799                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
 800
 801         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 802         if (err < 0)
 803                 return err;
 804
 805         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 806 }
 807 #else
 808 static int
 809 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 810                           enum nf_nat_manip_type manip,
 811                           const struct nlattr *attr)
 812 {
 813         return -EOPNOTSUPP;
 814 }
 815 #endif
 816
 817 static struct nf_ct_helper_expectfn follow_master_nat = {
 818         .name           = "nat-follow-master",
 819         .expectfn       = nf_nat_follow_master,
 820 };
 821
 822 static int __init nf_nat_init(void)
 823 {
 824         int ret;
 825
 826         ret = rhltable_init(&nf_nat_bysource_table, &nf_nat_bysource_params);
 827         if (ret)
 828                 return ret;
 829
 830         ret = nf_ct_extend_register(&nat_extend);
 831         if (ret < 0) {
 832                 rhltable_destroy(&nf_nat_bysource_table);
 833                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 834                 return ret;
 835         }
 836
 837         nf_ct_helper_expectfn_register(&follow_master_nat);
 838
 839         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 840         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 841                            nfnetlink_parse_nat_setup);
 842 #ifdef CONFIG_XFRM
 843         BUG_ON(nf_nat_decode_session_hook != NULL);
 844         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 845 #endif
 846         return 0;
 847 }
 848
 849 static void __exit nf_nat_cleanup(void)
 850 {
 851         struct nf_nat_proto_clean clean = {};
 852         unsigned int i;
 853
 854         nf_ct_iterate_destroy(nf_nat_proto_clean, &clean);
 855
 856         nf_ct_extend_unregister(&nat_extend);
 857         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 858         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 859 #ifdef CONFIG_XFRM
 860         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 861 #endif
 862         synchronize_rcu();
 863
 864         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 865                 kfree(nf_nat_l4protos[i]);
 866
 867         rhltable_destroy(&nf_nat_bysource_table);
 868 }
 869
 870 MODULE_LICENSE("GPL");
 871
 872 module_init(nf_nat_init);
 873 module_exit(nf_nat_cleanup);