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 rhashtable 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 = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.l3num;
  75         rcu_read_lock();
  76         l3proto = __nf_nat_l3proto_find(family);
  77         if (l3proto == NULL)
  78                 goto out;
  79
  80         dir = CTINFO2DIR(ctinfo);
  81         if (dir == IP_CT_DIR_ORIGINAL)
  82                 statusbit = IPS_DST_NAT;
  83         else
  84                 statusbit = IPS_SRC_NAT;
  85
  86         l3proto->decode_session(skb, ct, dir, statusbit, fl);
  87 out:
  88         rcu_read_unlock();
  89 }
  90
  91 int nf_xfrm_me_harder(struct net *net, struct sk_buff *skb, unsigned int family)
  92 {
  93         struct flowi fl;
  94         unsigned int hh_len;
  95         struct dst_entry *dst;
  96         int err;
  97
  98         err = xfrm_decode_session(skb, &fl, family);
  99         if (err < 0)
 100                 return err;
 101
 102         dst = skb_dst(skb);
 103         if (dst->xfrm)
 104                 dst = ((struct xfrm_dst *)dst)->route;
 105         dst_hold(dst);
 106
 107         dst = xfrm_lookup(net, dst, &fl, skb->sk, 0);
 108         if (IS_ERR(dst))
 109                 return PTR_ERR(dst);
 110
 111         skb_dst_drop(skb);
 112         skb_dst_set(skb, dst);
 113
 114         /* Change in oif may mean change in hh_len. */
 115         hh_len = skb_dst(skb)->dev->hard_header_len;
 116         if (skb_headroom(skb) < hh_len &&
 117             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 118                 return -ENOMEM;
 119         return 0;
 120 }
 121 EXPORT_SYMBOL(nf_xfrm_me_harder);
 122 #endif /* CONFIG_XFRM */
 123
 124 static u32 nf_nat_bysource_hash(const void *data, u32 len, u32 seed)
 125 {
 126         const struct nf_conntrack_tuple *t;
 127         const struct nf_conn *ct = data;
 128
 129         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 130         /* Original src, to ensure we map it consistently if poss. */
 131
 132         seed ^= net_hash_mix(nf_ct_net(ct));
 133         return jhash2((const u32 *)&t->src, sizeof(t->src) / sizeof(u32),
 134                       t->dst.protonum ^ seed);
 135 }
 136
 137 /* Is this tuple already taken? (not by us) */
 138 int
 139 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 140                   const struct nf_conn *ignored_conntrack)
 141 {
 142         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 143          * incoming ones.  NAT means they don't have a fixed mapping,
 144          * so we invert the tuple and look for the incoming reply.
 145          *
 146          * We could keep a separate hash if this proves too slow.
 147          */
 148         struct nf_conntrack_tuple reply;
 149
 150         nf_ct_invert_tuplepr(&reply, tuple);
 151         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 152 }
 153 EXPORT_SYMBOL(nf_nat_used_tuple);
 154
 155 /* If we source map this tuple so reply looks like reply_tuple, will
 156  * that meet the constraints of range.
 157  */
 158 static int in_range(const struct nf_nat_l3proto *l3proto,
 159                     const struct nf_nat_l4proto *l4proto,
 160                     const struct nf_conntrack_tuple *tuple,
 161                     const struct nf_nat_range *range)
 162 {
 163         /* If we are supposed to map IPs, then we must be in the
 164          * range specified, otherwise let this drag us onto a new src IP.
 165          */
 166         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 167             !l3proto->in_range(tuple, range))
 168                 return 0;
 169
 170         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 171             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 172                               &range->min_proto, &range->max_proto))
 173                 return 1;
 174
 175         return 0;
 176 }
 177
 178 static inline int
 179 same_src(const struct nf_conn *ct,
 180          const struct nf_conntrack_tuple *tuple)
 181 {
 182         const struct nf_conntrack_tuple *t;
 183
 184         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 185         return (t->dst.protonum == tuple->dst.protonum &&
 186                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 187                 t->src.u.all == tuple->src.u.all);
 188 }
 189
 190 static int nf_nat_bysource_cmp(struct rhashtable_compare_arg *arg,
 191                                const void *obj)
 192 {
 193         const struct nf_nat_conn_key *key = arg->key;
 194         const struct nf_conn *ct = obj;
 195
 196         return same_src(ct, key->tuple) &&
 197                net_eq(nf_ct_net(ct), key->net) &&
 198                nf_ct_zone_equal(ct, key->zone, IP_CT_DIR_ORIGINAL);
 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         .nulls_base = (1U << RHT_BASE_SHIFT),
 208 };
 209
 210 /* Only called for SRC manip */
 211 static int
 212 find_appropriate_src(struct net *net,
 213                      const struct nf_conntrack_zone *zone,
 214                      const struct nf_nat_l3proto *l3proto,
 215                      const struct nf_nat_l4proto *l4proto,
 216                      const struct nf_conntrack_tuple *tuple,
 217                      struct nf_conntrack_tuple *result,
 218                      const struct nf_nat_range *range)
 219 {
 220         const struct nf_conn *ct;
 221         struct nf_nat_conn_key key = {
 222                 .net = net,
 223                 .tuple = tuple,
 224                 .zone = zone
 225         };
 226
 227         ct = rhashtable_lookup_fast(&nf_nat_bysource_table, &key,
 228                                     nf_nat_bysource_params);
 229         if (!ct)
 230                 return 0;
 231
 232         nf_ct_invert_tuplepr(result,
 233                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 234         result->dst = tuple->dst;
 235
 236         return in_range(l3proto, l4proto, result, range);
 237 }
 238
 239 /* For [FUTURE] fragmentation handling, we want the least-used
 240  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 241  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 242  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 243  * the ip with the lowest src-ip/dst-ip/proto usage.
 244  */
 245 static void
 246 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 247                     struct nf_conntrack_tuple *tuple,
 248                     const struct nf_nat_range *range,
 249                     const struct nf_conn *ct,
 250                     enum nf_nat_manip_type maniptype)
 251 {
 252         union nf_inet_addr *var_ipp;
 253         unsigned int i, max;
 254         /* Host order */
 255         u32 minip, maxip, j, dist;
 256         bool full_range;
 257
 258         /* No IP mapping?  Do nothing. */
 259         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 260                 return;
 261
 262         if (maniptype == NF_NAT_MANIP_SRC)
 263                 var_ipp = &tuple->src.u3;
 264         else
 265                 var_ipp = &tuple->dst.u3;
 266
 267         /* Fast path: only one choice. */
 268         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 269                 *var_ipp = range->min_addr;
 270                 return;
 271         }
 272
 273         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 274                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 275         else
 276                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 277
 278         /* Hashing source and destination IPs gives a fairly even
 279          * spread in practice (if there are a small number of IPs
 280          * involved, there usually aren't that many connections
 281          * anyway).  The consistency means that servers see the same
 282          * client coming from the same IP (some Internet Banking sites
 283          * like this), even across reboots.
 284          */
 285         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 286                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 287                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 288
 289         full_range = false;
 290         for (i = 0; i <= max; i++) {
 291                 /* If first bytes of the address are at the maximum, use the
 292                  * distance. Otherwise use the full range.
 293                  */
 294                 if (!full_range) {
 295                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 296                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 297                         dist  = maxip - minip + 1;
 298                 } else {
 299                         minip = 0;
 300                         dist  = ~0;
 301                 }
 302
 303                 var_ipp->all[i] = (__force __u32)
 304                         htonl(minip + reciprocal_scale(j, dist));
 305                 if (var_ipp->all[i] != range->max_addr.all[i])
 306                         full_range = true;
 307
 308                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 309                         j ^= (__force u32)tuple->dst.u3.all[i];
 310         }
 311 }
 312
 313 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 314  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 315  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 316  * range. It might not be possible to get a unique tuple, but we try.
 317  * At worst (or if we race), we will end up with a final duplicate in
 318  * __ip_conntrack_confirm and drop the packet. */
 319 static void
 320 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 321                  const struct nf_conntrack_tuple *orig_tuple,
 322                  const struct nf_nat_range *range,
 323                  struct nf_conn *ct,
 324                  enum nf_nat_manip_type maniptype)
 325 {
 326         const struct nf_conntrack_zone *zone;
 327         const struct nf_nat_l3proto *l3proto;
 328         const struct nf_nat_l4proto *l4proto;
 329         struct net *net = nf_ct_net(ct);
 330
 331         zone = nf_ct_zone(ct);
 332
 333         rcu_read_lock();
 334         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 335         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 336                                         orig_tuple->dst.protonum);
 337
 338         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 339          * and that same mapping gives a unique tuple within the given
 340          * range, use that.
 341          *
 342          * This is only required for source (ie. NAT/masq) mappings.
 343          * So far, we don't do local source mappings, so multiple
 344          * manips not an issue.
 345          */
 346         if (maniptype == NF_NAT_MANIP_SRC &&
 347             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 348                 /* try the original tuple first */
 349                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 350                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 351                                 *tuple = *orig_tuple;
 352                                 goto out;
 353                         }
 354                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 355                                                 orig_tuple, tuple, range)) {
 356                         pr_debug("get_unique_tuple: Found current src map\n");
 357                         if (!nf_nat_used_tuple(tuple, ct))
 358                                 goto out;
 359                 }
 360         }
 361
 362         /* 2) Select the least-used IP/proto combination in the given range */
 363         *tuple = *orig_tuple;
 364         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 365
 366         /* 3) The per-protocol part of the manip is made to map into
 367          * the range to make a unique tuple.
 368          */
 369
 370         /* Only bother mapping if it's not already in range and unique */
 371         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 372                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 373                         if (l4proto->in_range(tuple, maniptype,
 374                                               &range->min_proto,
 375                                               &range->max_proto) &&
 376                             (range->min_proto.all == range->max_proto.all ||
 377                              !nf_nat_used_tuple(tuple, ct)))
 378                                 goto out;
 379                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 380                         goto out;
 381                 }
 382         }
 383
 384         /* Last change: get protocol to try to obtain unique tuple. */
 385         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 386 out:
 387         rcu_read_unlock();
 388 }
 389
 390 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 391 {
 392         struct nf_conn_nat *nat = nfct_nat(ct);
 393         if (nat)
 394                 return nat;
 395
 396         if (!nf_ct_is_confirmed(ct))
 397                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 398
 399         return nat;
 400 }
 401 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 402
 403 unsigned int
 404 nf_nat_setup_info(struct nf_conn *ct,
 405                   const struct nf_nat_range *range,
 406                   enum nf_nat_manip_type maniptype)
 407 {
 408         struct nf_conntrack_tuple curr_tuple, new_tuple;
 409         struct nf_conn_nat *nat;
 410
 411         /* nat helper or nfctnetlink also setup binding */
 412         nat = nf_ct_nat_ext_add(ct);
 413         if (nat == NULL)
 414                 return NF_ACCEPT;
 415
 416         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 417                      maniptype == NF_NAT_MANIP_DST);
 418         BUG_ON(nf_nat_initialized(ct, maniptype));
 419
 420         /* What we've got will look like inverse of reply. Normally
 421          * this is what is in the conntrack, except for prior
 422          * manipulations (future optimization: if num_manips == 0,
 423          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 424          */
 425         nf_ct_invert_tuplepr(&curr_tuple,
 426                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 427
 428         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 429
 430         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 431                 struct nf_conntrack_tuple reply;
 432
 433                 /* Alter conntrack table so will recognize replies. */
 434                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 435                 nf_conntrack_alter_reply(ct, &reply);
 436
 437                 /* Non-atomic: we own this at the moment. */
 438                 if (maniptype == NF_NAT_MANIP_SRC)
 439                         ct->status |= IPS_SRC_NAT;
 440                 else
 441                         ct->status |= IPS_DST_NAT;
 442
 443                 if (nfct_help(ct))
 444                         if (!nfct_seqadj_ext_add(ct))
 445                                 return NF_DROP;
 446         }
 447
 448         if (maniptype == NF_NAT_MANIP_SRC) {
 449                 int err;
 450
 451                 err = rhashtable_insert_fast(&nf_nat_bysource_table,
 452                                              &ct->nat_bysource,
 453                                              nf_nat_bysource_params);
 454                 if (err)
 455                         return NF_DROP;
 456         }
 457
 458         /* It's done. */
 459         if (maniptype == NF_NAT_MANIP_DST)
 460                 ct->status |= IPS_DST_NAT_DONE;
 461         else
 462                 ct->status |= IPS_SRC_NAT_DONE;
 463
 464         return NF_ACCEPT;
 465 }
 466 EXPORT_SYMBOL(nf_nat_setup_info);
 467
 468 static unsigned int
 469 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 470 {
 471         /* Force range to this IP; let proto decide mapping for
 472          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 473          * Use reply in case it's already been mangled (eg local packet).
 474          */
 475         union nf_inet_addr ip =
 476                 (manip == NF_NAT_MANIP_SRC ?
 477                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 478                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 479         struct nf_nat_range range = {
 480                 .flags          = NF_NAT_RANGE_MAP_IPS,
 481                 .min_addr       = ip,
 482                 .max_addr       = ip,
 483         };
 484         return nf_nat_setup_info(ct, &range, manip);
 485 }
 486
 487 unsigned int
 488 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 489 {
 490         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 491 }
 492 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 493
 494 /* Do packet manipulations according to nf_nat_setup_info. */
 495 unsigned int nf_nat_packet(struct nf_conn *ct,
 496                            enum ip_conntrack_info ctinfo,
 497                            unsigned int hooknum,
 498                            struct sk_buff *skb)
 499 {
 500         const struct nf_nat_l3proto *l3proto;
 501         const struct nf_nat_l4proto *l4proto;
 502         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 503         unsigned long statusbit;
 504         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 505
 506         if (mtype == NF_NAT_MANIP_SRC)
 507                 statusbit = IPS_SRC_NAT;
 508         else
 509                 statusbit = IPS_DST_NAT;
 510
 511         /* Invert if this is reply dir. */
 512         if (dir == IP_CT_DIR_REPLY)
 513                 statusbit ^= IPS_NAT_MASK;
 514
 515         /* Non-atomic: these bits don't change. */
 516         if (ct->status & statusbit) {
 517                 struct nf_conntrack_tuple target;
 518
 519                 /* We are aiming to look like inverse of other direction. */
 520                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 521
 522                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 523                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 524                                                 target.dst.protonum);
 525                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 526                         return NF_DROP;
 527         }
 528         return NF_ACCEPT;
 529 }
 530 EXPORT_SYMBOL_GPL(nf_nat_packet);
 531
 532 struct nf_nat_proto_clean {
 533         u8      l3proto;
 534         u8      l4proto;
 535 };
 536
 537 /* kill conntracks with affected NAT section */
 538 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 539 {
 540         const struct nf_nat_proto_clean *clean = data;
 541         struct nf_conn_nat *nat = nfct_nat(i);
 542
 543         if (!nat)
 544                 return 0;
 545
 546         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 547             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 548                 return 0;
 549
 550         return i->status & IPS_NAT_MASK ? 1 : 0;
 551 }
 552
 553 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 554 {
 555         struct nf_conn_nat *nat = nfct_nat(ct);
 556
 557         if (nf_nat_proto_remove(ct, data))
 558                 return 1;
 559
 560         if (!nat)
 561                 return 0;
 562
 563         /* This netns is being destroyed, and conntrack has nat null binding.
 564          * Remove it from bysource hash, as the table will be freed soon.
 565          *
 566          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 567          * will delete entry from already-freed table.
 568          */
 569         ct->status &= ~IPS_NAT_DONE_MASK;
 570         rhashtable_remove_fast(&nf_nat_bysource_table, &ct->nat_bysource,
 571                                nf_nat_bysource_params);
 572
 573         /* don't delete conntrack.  Although that would make things a lot
 574          * simpler, we'd end up flushing all conntracks on nat rmmod.
 575          */
 576         return 0;
 577 }
 578
 579 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 580 {
 581         struct nf_nat_proto_clean clean = {
 582                 .l3proto = l3proto,
 583                 .l4proto = l4proto,
 584         };
 585         struct net *net;
 586
 587         rtnl_lock();
 588         for_each_net(net)
 589                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 590         rtnl_unlock();
 591 }
 592
 593 static void nf_nat_l3proto_clean(u8 l3proto)
 594 {
 595         struct nf_nat_proto_clean clean = {
 596                 .l3proto = l3proto,
 597         };
 598         struct net *net;
 599
 600         rtnl_lock();
 601
 602         for_each_net(net)
 603                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 604         rtnl_unlock();
 605 }
 606
 607 /* Protocol registration. */
 608 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 609 {
 610         const struct nf_nat_l4proto **l4protos;
 611         unsigned int i;
 612         int ret = 0;
 613
 614         mutex_lock(&nf_nat_proto_mutex);
 615         if (nf_nat_l4protos[l3proto] == NULL) {
 616                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 617                                    GFP_KERNEL);
 618                 if (l4protos == NULL) {
 619                         ret = -ENOMEM;
 620                         goto out;
 621                 }
 622
 623                 for (i = 0; i < IPPROTO_MAX; i++)
 624                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 625
 626                 /* Before making proto_array visible to lockless readers,
 627                  * we must make sure its content is committed to memory.
 628                  */
 629                 smp_wmb();
 630
 631                 nf_nat_l4protos[l3proto] = l4protos;
 632         }
 633
 634         if (rcu_dereference_protected(
 635                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 636                         lockdep_is_held(&nf_nat_proto_mutex)
 637                         ) != &nf_nat_l4proto_unknown) {
 638                 ret = -EBUSY;
 639                 goto out;
 640         }
 641         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 642  out:
 643         mutex_unlock(&nf_nat_proto_mutex);
 644         return ret;
 645 }
 646 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 647
 648 /* No one stores the protocol anywhere; simply delete it. */
 649 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 650 {
 651         mutex_lock(&nf_nat_proto_mutex);
 652         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 653                          &nf_nat_l4proto_unknown);
 654         mutex_unlock(&nf_nat_proto_mutex);
 655         synchronize_rcu();
 656
 657         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 658 }
 659 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 660
 661 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 662 {
 663         int err;
 664
 665         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 666         if (err < 0)
 667                 return err;
 668
 669         mutex_lock(&nf_nat_proto_mutex);
 670         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 671                          &nf_nat_l4proto_tcp);
 672         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 673                          &nf_nat_l4proto_udp);
 674         mutex_unlock(&nf_nat_proto_mutex);
 675
 676         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 677         return 0;
 678 }
 679 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 680
 681 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 682 {
 683         mutex_lock(&nf_nat_proto_mutex);
 684         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 685         mutex_unlock(&nf_nat_proto_mutex);
 686         synchronize_rcu();
 687
 688         nf_nat_l3proto_clean(l3proto->l3proto);
 689         nf_ct_l3proto_module_put(l3proto->l3proto);
 690 }
 691 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 692
 693 /* No one using conntrack by the time this called. */
 694 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 695 {
 696         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 697
 698         if (!nat)
 699                 return;
 700
 701         rhashtable_remove_fast(&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         .flags          = NF_CT_EXT_F_PREALLOC,
 711 };
 712
 713 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 714
 715 #include <linux/netfilter/nfnetlink.h>
 716 #include <linux/netfilter/nfnetlink_conntrack.h>
 717
 718 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 719         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 720         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 721 };
 722
 723 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 724                                      const struct nf_conn *ct,
 725                                      struct nf_nat_range *range)
 726 {
 727         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 728         const struct nf_nat_l4proto *l4proto;
 729         int err;
 730
 731         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 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);
 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);
 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 void __net_exit nf_nat_net_exit(struct net *net)
 818 {
 819         struct nf_nat_proto_clean clean = {};
 820
 821         nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean, 0, 0);
 822 }
 823
 824 static struct pernet_operations nf_nat_net_ops = {
 825         .exit = nf_nat_net_exit,
 826 };
 827
 828 static struct nf_ct_helper_expectfn follow_master_nat = {
 829         .name           = "nat-follow-master",
 830         .expectfn       = nf_nat_follow_master,
 831 };
 832
 833 static int __init nf_nat_init(void)
 834 {
 835         int ret;
 836
 837         ret = rhashtable_init(&nf_nat_bysource_table, &nf_nat_bysource_params);
 838         if (ret)
 839                 return ret;
 840
 841         ret = nf_ct_extend_register(&nat_extend);
 842         if (ret < 0) {
 843                 rhashtable_destroy(&nf_nat_bysource_table);
 844                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 845                 return ret;
 846         }
 847
 848         ret = register_pernet_subsys(&nf_nat_net_ops);
 849         if (ret < 0)
 850                 goto cleanup_extend;
 851
 852         nf_ct_helper_expectfn_register(&follow_master_nat);
 853
 854         /* Initialize fake conntrack so that NAT will skip it */
 855         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 856
 857         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 858         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 859                            nfnetlink_parse_nat_setup);
 860 #ifdef CONFIG_XFRM
 861         BUG_ON(nf_nat_decode_session_hook != NULL);
 862         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 863 #endif
 864         return 0;
 865
 866  cleanup_extend:
 867         rhashtable_destroy(&nf_nat_bysource_table);
 868         nf_ct_extend_unregister(&nat_extend);
 869         return ret;
 870 }
 871
 872 static void __exit nf_nat_cleanup(void)
 873 {
 874         unsigned int i;
 875
 876         unregister_pernet_subsys(&nf_nat_net_ops);
 877         nf_ct_extend_unregister(&nat_extend);
 878         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 879         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 880 #ifdef CONFIG_XFRM
 881         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 882 #endif
 883         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 884                 kfree(nf_nat_l4protos[i]);
 885
 886         rhashtable_destroy(&nf_nat_bysource_table);
 887 }
 888
 889 MODULE_LICENSE("GPL");
 890
 891 module_init(nf_nat_init);
 892 module_exit(nf_nat_cleanup);