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;
 226
 227         hl = rhltable_lookup(&nf_nat_bysource_table, &key,
 228                              nf_nat_bysource_params);
 229         if (!hl)
 230                 return 0;
 231
 232         ct = container_of(hl, typeof(*ct), nat_bysource);
 233
 234         nf_ct_invert_tuplepr(result,
 235                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 236         result->dst = tuple->dst;
 237
 238         return in_range(l3proto, l4proto, result, range);
 239 }
 240
 241 /* For [FUTURE] fragmentation handling, we want the least-used
 242  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 243  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 244  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 245  * the ip with the lowest src-ip/dst-ip/proto usage.
 246  */
 247 static void
 248 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 249                     struct nf_conntrack_tuple *tuple,
 250                     const struct nf_nat_range *range,
 251                     const struct nf_conn *ct,
 252                     enum nf_nat_manip_type maniptype)
 253 {
 254         union nf_inet_addr *var_ipp;
 255         unsigned int i, max;
 256         /* Host order */
 257         u32 minip, maxip, j, dist;
 258         bool full_range;
 259
 260         /* No IP mapping?  Do nothing. */
 261         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 262                 return;
 263
 264         if (maniptype == NF_NAT_MANIP_SRC)
 265                 var_ipp = &tuple->src.u3;
 266         else
 267                 var_ipp = &tuple->dst.u3;
 268
 269         /* Fast path: only one choice. */
 270         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 271                 *var_ipp = range->min_addr;
 272                 return;
 273         }
 274
 275         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 276                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 277         else
 278                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 279
 280         /* Hashing source and destination IPs gives a fairly even
 281          * spread in practice (if there are a small number of IPs
 282          * involved, there usually aren't that many connections
 283          * anyway).  The consistency means that servers see the same
 284          * client coming from the same IP (some Internet Banking sites
 285          * like this), even across reboots.
 286          */
 287         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 288                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 289                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 290
 291         full_range = false;
 292         for (i = 0; i <= max; i++) {
 293                 /* If first bytes of the address are at the maximum, use the
 294                  * distance. Otherwise use the full range.
 295                  */
 296                 if (!full_range) {
 297                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 298                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 299                         dist  = maxip - minip + 1;
 300                 } else {
 301                         minip = 0;
 302                         dist  = ~0;
 303                 }
 304
 305                 var_ipp->all[i] = (__force __u32)
 306                         htonl(minip + reciprocal_scale(j, dist));
 307                 if (var_ipp->all[i] != range->max_addr.all[i])
 308                         full_range = true;
 309
 310                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 311                         j ^= (__force u32)tuple->dst.u3.all[i];
 312         }
 313 }
 314
 315 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 316  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 317  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 318  * range. It might not be possible to get a unique tuple, but we try.
 319  * At worst (or if we race), we will end up with a final duplicate in
 320  * __ip_conntrack_confirm and drop the packet. */
 321 static void
 322 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 323                  const struct nf_conntrack_tuple *orig_tuple,
 324                  const struct nf_nat_range *range,
 325                  struct nf_conn *ct,
 326                  enum nf_nat_manip_type maniptype)
 327 {
 328         const struct nf_conntrack_zone *zone;
 329         const struct nf_nat_l3proto *l3proto;
 330         const struct nf_nat_l4proto *l4proto;
 331         struct net *net = nf_ct_net(ct);
 332
 333         zone = nf_ct_zone(ct);
 334
 335         rcu_read_lock();
 336         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 337         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 338                                         orig_tuple->dst.protonum);
 339
 340         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 341          * and that same mapping gives a unique tuple within the given
 342          * range, use that.
 343          *
 344          * This is only required for source (ie. NAT/masq) mappings.
 345          * So far, we don't do local source mappings, so multiple
 346          * manips not an issue.
 347          */
 348         if (maniptype == NF_NAT_MANIP_SRC &&
 349             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 350                 /* try the original tuple first */
 351                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 352                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 353                                 *tuple = *orig_tuple;
 354                                 goto out;
 355                         }
 356                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 357                                                 orig_tuple, tuple, range)) {
 358                         pr_debug("get_unique_tuple: Found current src map\n");
 359                         if (!nf_nat_used_tuple(tuple, ct))
 360                                 goto out;
 361                 }
 362         }
 363
 364         /* 2) Select the least-used IP/proto combination in the given range */
 365         *tuple = *orig_tuple;
 366         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 367
 368         /* 3) The per-protocol part of the manip is made to map into
 369          * the range to make a unique tuple.
 370          */
 371
 372         /* Only bother mapping if it's not already in range and unique */
 373         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 374                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 375                         if (l4proto->in_range(tuple, maniptype,
 376                                               &range->min_proto,
 377                                               &range->max_proto) &&
 378                             (range->min_proto.all == range->max_proto.all ||
 379                              !nf_nat_used_tuple(tuple, ct)))
 380                                 goto out;
 381                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 382                         goto out;
 383                 }
 384         }
 385
 386         /* Last change: get protocol to try to obtain unique tuple. */
 387         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 388 out:
 389         rcu_read_unlock();
 390 }
 391
 392 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 393 {
 394         struct nf_conn_nat *nat = nfct_nat(ct);
 395         if (nat)
 396                 return nat;
 397
 398         if (!nf_ct_is_confirmed(ct))
 399                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 400
 401         return nat;
 402 }
 403 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 404
 405 unsigned int
 406 nf_nat_setup_info(struct nf_conn *ct,
 407                   const struct nf_nat_range *range,
 408                   enum nf_nat_manip_type maniptype)
 409 {
 410         struct nf_conntrack_tuple curr_tuple, new_tuple;
 411
 412         /* Can't setup nat info for confirmed ct. */
 413         if (nf_ct_is_confirmed(ct))
 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                 struct nf_nat_conn_key key = {
 450                         .net = nf_ct_net(ct),
 451                         .tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 452                         .zone = nf_ct_zone(ct),
 453                 };
 454                 int err;
 455
 456                 err = rhltable_insert_key(&nf_nat_bysource_table,
 457                                           &key,
 458                                           &ct->nat_bysource,
 459                                           nf_nat_bysource_params);
 460                 if (err)
 461                         return NF_DROP;
 462         }
 463
 464         /* It's done. */
 465         if (maniptype == NF_NAT_MANIP_DST)
 466                 ct->status |= IPS_DST_NAT_DONE;
 467         else
 468                 ct->status |= IPS_SRC_NAT_DONE;
 469
 470         return NF_ACCEPT;
 471 }
 472 EXPORT_SYMBOL(nf_nat_setup_info);
 473
 474 static unsigned int
 475 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 476 {
 477         /* Force range to this IP; let proto decide mapping for
 478          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 479          * Use reply in case it's already been mangled (eg local packet).
 480          */
 481         union nf_inet_addr ip =
 482                 (manip == NF_NAT_MANIP_SRC ?
 483                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 484                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 485         struct nf_nat_range range = {
 486                 .flags          = NF_NAT_RANGE_MAP_IPS,
 487                 .min_addr       = ip,
 488                 .max_addr       = ip,
 489         };
 490         return nf_nat_setup_info(ct, &range, manip);
 491 }
 492
 493 unsigned int
 494 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 495 {
 496         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 497 }
 498 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 499
 500 /* Do packet manipulations according to nf_nat_setup_info. */
 501 unsigned int nf_nat_packet(struct nf_conn *ct,
 502                            enum ip_conntrack_info ctinfo,
 503                            unsigned int hooknum,
 504                            struct sk_buff *skb)
 505 {
 506         const struct nf_nat_l3proto *l3proto;
 507         const struct nf_nat_l4proto *l4proto;
 508         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 509         unsigned long statusbit;
 510         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 511
 512         if (mtype == NF_NAT_MANIP_SRC)
 513                 statusbit = IPS_SRC_NAT;
 514         else
 515                 statusbit = IPS_DST_NAT;
 516
 517         /* Invert if this is reply dir. */
 518         if (dir == IP_CT_DIR_REPLY)
 519                 statusbit ^= IPS_NAT_MASK;
 520
 521         /* Non-atomic: these bits don't change. */
 522         if (ct->status & statusbit) {
 523                 struct nf_conntrack_tuple target;
 524
 525                 /* We are aiming to look like inverse of other direction. */
 526                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 527
 528                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 529                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 530                                                 target.dst.protonum);
 531                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 532                         return NF_DROP;
 533         }
 534         return NF_ACCEPT;
 535 }
 536 EXPORT_SYMBOL_GPL(nf_nat_packet);
 537
 538 struct nf_nat_proto_clean {
 539         u8      l3proto;
 540         u8      l4proto;
 541 };
 542
 543 /* kill conntracks with affected NAT section */
 544 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 545 {
 546         const struct nf_nat_proto_clean *clean = data;
 547
 548         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 549             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 550                 return 0;
 551
 552         return i->status & IPS_NAT_MASK ? 1 : 0;
 553 }
 554
 555 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 556 {
 557         if (nf_nat_proto_remove(ct, data))
 558                 return 1;
 559
 560         if ((ct->status & IPS_SRC_NAT_DONE) == 0)
 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         clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
 570         rhltable_remove(&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 #ifdef CONFIG_NF_NAT_PROTO_DCCP
 675         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_DCCP],
 676                          &nf_nat_l4proto_dccp);
 677 #endif
 678 #ifdef CONFIG_NF_NAT_PROTO_SCTP
 679         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_SCTP],
 680                          &nf_nat_l4proto_sctp);
 681 #endif
 682 #ifdef CONFIG_NF_NAT_PROTO_UDPLITE
 683         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDPLITE],
 684                          &nf_nat_l4proto_udplite);
 685 #endif
 686         mutex_unlock(&nf_nat_proto_mutex);
 687
 688         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 689         return 0;
 690 }
 691 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 692
 693 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 694 {
 695         mutex_lock(&nf_nat_proto_mutex);
 696         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 697         mutex_unlock(&nf_nat_proto_mutex);
 698         synchronize_rcu();
 699
 700         nf_nat_l3proto_clean(l3proto->l3proto);
 701         nf_ct_l3proto_module_put(l3proto->l3proto);
 702 }
 703 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 704
 705 /* No one using conntrack by the time this called. */
 706 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 707 {
 708         if (ct->status & IPS_SRC_NAT_DONE)
 709                 rhltable_remove(&nf_nat_bysource_table, &ct->nat_bysource,
 710                                 nf_nat_bysource_params);
 711 }
 712
 713 static struct nf_ct_ext_type nat_extend __read_mostly = {
 714         .len            = sizeof(struct nf_conn_nat),
 715         .align          = __alignof__(struct nf_conn_nat),
 716         .destroy        = nf_nat_cleanup_conntrack,
 717         .id             = NF_CT_EXT_NAT,
 718 };
 719
 720 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 721
 722 #include <linux/netfilter/nfnetlink.h>
 723 #include <linux/netfilter/nfnetlink_conntrack.h>
 724
 725 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 726         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 727         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 728 };
 729
 730 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 731                                      const struct nf_conn *ct,
 732                                      struct nf_nat_range *range)
 733 {
 734         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 735         const struct nf_nat_l4proto *l4proto;
 736         int err;
 737
 738         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr,
 739                                protonat_nla_policy, NULL);
 740         if (err < 0)
 741                 return err;
 742
 743         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 744         if (l4proto->nlattr_to_range)
 745                 err = l4proto->nlattr_to_range(tb, range);
 746
 747         return err;
 748 }
 749
 750 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 751         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 752         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 753         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 754         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 755         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 756 };
 757
 758 static int
 759 nfnetlink_parse_nat(const struct nlattr *nat,
 760                     const struct nf_conn *ct, struct nf_nat_range *range,
 761                     const struct nf_nat_l3proto *l3proto)
 762 {
 763         struct nlattr *tb[CTA_NAT_MAX+1];
 764         int err;
 765
 766         memset(range, 0, sizeof(*range));
 767
 768         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy, NULL);
 769         if (err < 0)
 770                 return err;
 771
 772         err = l3proto->nlattr_to_range(tb, range);
 773         if (err < 0)
 774                 return err;
 775
 776         if (!tb[CTA_NAT_PROTO])
 777                 return 0;
 778
 779         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 780 }
 781
 782 /* This function is called under rcu_read_lock() */
 783 static int
 784 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 785                           enum nf_nat_manip_type manip,
 786                           const struct nlattr *attr)
 787 {
 788         struct nf_nat_range range;
 789         const struct nf_nat_l3proto *l3proto;
 790         int err;
 791
 792         /* Should not happen, restricted to creating new conntracks
 793          * via ctnetlink.
 794          */
 795         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 796                 return -EEXIST;
 797
 798         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 799          * attach the null binding, otherwise this may oops.
 800          */
 801         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 802         if (l3proto == NULL)
 803                 return -EAGAIN;
 804
 805         /* No NAT information has been passed, allocate the null-binding */
 806         if (attr == NULL)
 807                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
 808
 809         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 810         if (err < 0)
 811                 return err;
 812
 813         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 814 }
 815 #else
 816 static int
 817 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 818                           enum nf_nat_manip_type manip,
 819                           const struct nlattr *attr)
 820 {
 821         return -EOPNOTSUPP;
 822 }
 823 #endif
 824
 825 static void __net_exit nf_nat_net_exit(struct net *net)
 826 {
 827         struct nf_nat_proto_clean clean = {};
 828
 829         nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean, 0, 0);
 830 }
 831
 832 static struct pernet_operations nf_nat_net_ops = {
 833         .exit = nf_nat_net_exit,
 834 };
 835
 836 static struct nf_ct_helper_expectfn follow_master_nat = {
 837         .name           = "nat-follow-master",
 838         .expectfn       = nf_nat_follow_master,
 839 };
 840
 841 static int __init nf_nat_init(void)
 842 {
 843         int ret;
 844
 845         ret = rhltable_init(&nf_nat_bysource_table, &nf_nat_bysource_params);
 846         if (ret)
 847                 return ret;
 848
 849         ret = nf_ct_extend_register(&nat_extend);
 850         if (ret < 0) {
 851                 rhltable_destroy(&nf_nat_bysource_table);
 852                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 853                 return ret;
 854         }
 855
 856         ret = register_pernet_subsys(&nf_nat_net_ops);
 857         if (ret < 0)
 858                 goto cleanup_extend;
 859
 860         nf_ct_helper_expectfn_register(&follow_master_nat);
 861
 862         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 863         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 864                            nfnetlink_parse_nat_setup);
 865 #ifdef CONFIG_XFRM
 866         BUG_ON(nf_nat_decode_session_hook != NULL);
 867         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 868 #endif
 869         return 0;
 870
 871  cleanup_extend:
 872         rhltable_destroy(&nf_nat_bysource_table);
 873         nf_ct_extend_unregister(&nat_extend);
 874         return ret;
 875 }
 876
 877 static void __exit nf_nat_cleanup(void)
 878 {
 879         unsigned int i;
 880
 881         unregister_pernet_subsys(&nf_nat_net_ops);
 882         nf_ct_extend_unregister(&nat_extend);
 883         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 884         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 885 #ifdef CONFIG_XFRM
 886         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 887 #endif
 888         synchronize_rcu();
 889
 890         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 891                 kfree(nf_nat_l4protos[i]);
 892
 893         rhltable_destroy(&nf_nat_bysource_table);
 894 }
 895
 896 MODULE_LICENSE("GPL");
 897
 898 module_init(nf_nat_init);
 899 module_exit(nf_nat_cleanup);