kernel/net/netfilter/nf_nat_core.c

   1 /*
   2  * (C) 1999-2001 Paul `Rusty' Russell
   3  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   4  * (C) 2011 Patrick McHardy <kaber@trash.net>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/types.h>
  13 #include <linux/timer.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/gfp.h>
  16 #include <net/xfrm.h>
  17 #include <linux/jhash.h>
  18 #include <linux/rtnetlink.h>
  19
  20 #include <net/netfilter/nf_conntrack.h>
  21 #include <net/netfilter/nf_conntrack_core.h>
  22 #include <net/netfilter/nf_nat.h>
  23 #include <net/netfilter/nf_nat_l3proto.h>
  24 #include <net/netfilter/nf_nat_l4proto.h>
  25 #include <net/netfilter/nf_nat_core.h>
  26 #include <net/netfilter/nf_nat_helper.h>
  27 #include <net/netfilter/nf_conntrack_helper.h>
  28 #include <net/netfilter/nf_conntrack_seqadj.h>
  29 #include <net/netfilter/nf_conntrack_l3proto.h>
  30 #include <net/netfilter/nf_conntrack_zones.h>
  31 #include <linux/netfilter/nf_nat.h>
  32
  33 static DEFINE_SPINLOCK(nf_nat_lock);
  34
  35 static DEFINE_MUTEX(nf_nat_proto_mutex);
  36 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
  37                                                 __read_mostly;
  38 static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO]
  39                                                 __read_mostly;
  40
  41
  42 inline const struct nf_nat_l3proto *
  43 __nf_nat_l3proto_find(u8 family)
  44 {
  45         return rcu_dereference(nf_nat_l3protos[family]);
  46 }
  47
  48 inline const struct nf_nat_l4proto *
  49 __nf_nat_l4proto_find(u8 family, u8 protonum)
  50 {
  51         return rcu_dereference(nf_nat_l4protos[family][protonum]);
  52 }
  53 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
  54
  55 #ifdef CONFIG_XFRM
  56 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  57 {
  58         const struct nf_nat_l3proto *l3proto;
  59         const struct nf_conn *ct;
  60         enum ip_conntrack_info ctinfo;
  61         enum ip_conntrack_dir dir;
  62         unsigned  long statusbit;
  63         u8 family;
  64
  65         ct = nf_ct_get(skb, &ctinfo);
  66         if (ct == NULL)
  67                 return;
  68
  69         family = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.l3num;
  70         rcu_read_lock();
  71         l3proto = __nf_nat_l3proto_find(family);
  72         if (l3proto == NULL)
  73                 goto out;
  74
  75         dir = CTINFO2DIR(ctinfo);
  76         if (dir == IP_CT_DIR_ORIGINAL)
  77                 statusbit = IPS_DST_NAT;
  78         else
  79                 statusbit = IPS_SRC_NAT;
  80
  81         l3proto->decode_session(skb, ct, dir, statusbit, fl);
  82 out:
  83         rcu_read_unlock();
  84 }
  85
  86 int nf_xfrm_me_harder(struct sk_buff *skb, unsigned int family)
  87 {
  88         struct flowi fl;
  89         unsigned int hh_len;
  90         struct dst_entry *dst;
  91         int err;
  92
  93         err = xfrm_decode_session(skb, &fl, family);
  94         if (err < 0)
  95                 return err;
  96
  97         dst = skb_dst(skb);
  98         if (dst->xfrm)
  99                 dst = ((struct xfrm_dst *)dst)->route;
 100         dst_hold(dst);
 101
 102         dst = xfrm_lookup(dev_net(dst->dev), dst, &fl, skb->sk, 0);
 103         if (IS_ERR(dst))
 104                 return PTR_ERR(dst);
 105
 106         skb_dst_drop(skb);
 107         skb_dst_set(skb, dst);
 108
 109         /* Change in oif may mean change in hh_len. */
 110         hh_len = skb_dst(skb)->dev->hard_header_len;
 111         if (skb_headroom(skb) < hh_len &&
 112             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 113                 return -ENOMEM;
 114         return 0;
 115 }
 116 EXPORT_SYMBOL(nf_xfrm_me_harder);
 117 #endif /* CONFIG_XFRM */
 118
 119 /* We keep an extra hash for each conntrack, for fast searching. */
 120 static inline unsigned int
 121 hash_by_src(const struct net *net, u16 zone,
 122             const struct nf_conntrack_tuple *tuple)
 123 {
 124         unsigned int hash;
 125
 126         /* Original src, to ensure we map it consistently if poss. */
 127         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 128                       tuple->dst.protonum ^ zone ^ nf_conntrack_hash_rnd);
 129
 130         return reciprocal_scale(hash, net->ct.nat_htable_size);
 131 }
 132
 133 /* Is this tuple already taken? (not by us) */
 134 int
 135 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 136                   const struct nf_conn *ignored_conntrack)
 137 {
 138         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 139          * incoming ones.  NAT means they don't have a fixed mapping,
 140          * so we invert the tuple and look for the incoming reply.
 141          *
 142          * We could keep a separate hash if this proves too slow.
 143          */
 144         struct nf_conntrack_tuple reply;
 145
 146         nf_ct_invert_tuplepr(&reply, tuple);
 147         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 148 }
 149 EXPORT_SYMBOL(nf_nat_used_tuple);
 150
 151 /* If we source map this tuple so reply looks like reply_tuple, will
 152  * that meet the constraints of range.
 153  */
 154 static int in_range(const struct nf_nat_l3proto *l3proto,
 155                     const struct nf_nat_l4proto *l4proto,
 156                     const struct nf_conntrack_tuple *tuple,
 157                     const struct nf_nat_range *range)
 158 {
 159         /* If we are supposed to map IPs, then we must be in the
 160          * range specified, otherwise let this drag us onto a new src IP.
 161          */
 162         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 163             !l3proto->in_range(tuple, range))
 164                 return 0;
 165
 166         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 167             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 168                               &range->min_proto, &range->max_proto))
 169                 return 1;
 170
 171         return 0;
 172 }
 173
 174 static inline int
 175 same_src(const struct nf_conn *ct,
 176          const struct nf_conntrack_tuple *tuple)
 177 {
 178         const struct nf_conntrack_tuple *t;
 179
 180         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 181         return (t->dst.protonum == tuple->dst.protonum &&
 182                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 183                 t->src.u.all == tuple->src.u.all);
 184 }
 185
 186 /* Only called for SRC manip */
 187 static int
 188 find_appropriate_src(struct net *net, u16 zone,
 189                      const struct nf_nat_l3proto *l3proto,
 190                      const struct nf_nat_l4proto *l4proto,
 191                      const struct nf_conntrack_tuple *tuple,
 192                      struct nf_conntrack_tuple *result,
 193                      const struct nf_nat_range *range)
 194 {
 195         unsigned int h = hash_by_src(net, zone, tuple);
 196         const struct nf_conn_nat *nat;
 197         const struct nf_conn *ct;
 198
 199         hlist_for_each_entry_rcu(nat, &net->ct.nat_bysource[h], bysource) {
 200                 ct = nat->ct;
 201                 if (same_src(ct, tuple) && nf_ct_zone(ct) == zone) {
 202                         /* Copy source part from reply tuple. */
 203                         nf_ct_invert_tuplepr(result,
 204                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 205                         result->dst = tuple->dst;
 206
 207                         if (in_range(l3proto, l4proto, result, range))
 208                                 return 1;
 209                 }
 210         }
 211         return 0;
 212 }
 213
 214 /* For [FUTURE] fragmentation handling, we want the least-used
 215  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 216  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 217  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 218  * the ip with the lowest src-ip/dst-ip/proto usage.
 219  */
 220 static void
 221 find_best_ips_proto(u16 zone, struct nf_conntrack_tuple *tuple,
 222                     const struct nf_nat_range *range,
 223                     const struct nf_conn *ct,
 224                     enum nf_nat_manip_type maniptype)
 225 {
 226         union nf_inet_addr *var_ipp;
 227         unsigned int i, max;
 228         /* Host order */
 229         u32 minip, maxip, j, dist;
 230         bool full_range;
 231
 232         /* No IP mapping?  Do nothing. */
 233         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 234                 return;
 235
 236         if (maniptype == NF_NAT_MANIP_SRC)
 237                 var_ipp = &tuple->src.u3;
 238         else
 239                 var_ipp = &tuple->dst.u3;
 240
 241         /* Fast path: only one choice. */
 242         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 243                 *var_ipp = range->min_addr;
 244                 return;
 245         }
 246
 247         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 248                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 249         else
 250                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 251
 252         /* Hashing source and destination IPs gives a fairly even
 253          * spread in practice (if there are a small number of IPs
 254          * involved, there usually aren't that many connections
 255          * anyway).  The consistency means that servers see the same
 256          * client coming from the same IP (some Internet Banking sites
 257          * like this), even across reboots.
 258          */
 259         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 260                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 261                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone);
 262
 263         full_range = false;
 264         for (i = 0; i <= max; i++) {
 265                 /* If first bytes of the address are at the maximum, use the
 266                  * distance. Otherwise use the full range.
 267                  */
 268                 if (!full_range) {
 269                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 270                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 271                         dist  = maxip - minip + 1;
 272                 } else {
 273                         minip = 0;
 274                         dist  = ~0;
 275                 }
 276
 277                 var_ipp->all[i] = (__force __u32)
 278                         htonl(minip + reciprocal_scale(j, dist));
 279                 if (var_ipp->all[i] != range->max_addr.all[i])
 280                         full_range = true;
 281
 282                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 283                         j ^= (__force u32)tuple->dst.u3.all[i];
 284         }
 285 }
 286
 287 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 288  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 289  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 290  * range. It might not be possible to get a unique tuple, but we try.
 291  * At worst (or if we race), we will end up with a final duplicate in
 292  * __ip_conntrack_confirm and drop the packet. */
 293 static void
 294 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 295                  const struct nf_conntrack_tuple *orig_tuple,
 296                  const struct nf_nat_range *range,
 297                  struct nf_conn *ct,
 298                  enum nf_nat_manip_type maniptype)
 299 {
 300         const struct nf_nat_l3proto *l3proto;
 301         const struct nf_nat_l4proto *l4proto;
 302         struct net *net = nf_ct_net(ct);
 303         u16 zone = nf_ct_zone(ct);
 304
 305         rcu_read_lock();
 306         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 307         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 308                                         orig_tuple->dst.protonum);
 309
 310         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 311          * and that same mapping gives a unique tuple within the given
 312          * range, use that.
 313          *
 314          * This is only required for source (ie. NAT/masq) mappings.
 315          * So far, we don't do local source mappings, so multiple
 316          * manips not an issue.
 317          */
 318         if (maniptype == NF_NAT_MANIP_SRC &&
 319             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 320                 /* try the original tuple first */
 321                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 322                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 323                                 *tuple = *orig_tuple;
 324                                 goto out;
 325                         }
 326                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 327                                                 orig_tuple, tuple, range)) {
 328                         pr_debug("get_unique_tuple: Found current src map\n");
 329                         if (!nf_nat_used_tuple(tuple, ct))
 330                                 goto out;
 331                 }
 332         }
 333
 334         /* 2) Select the least-used IP/proto combination in the given range */
 335         *tuple = *orig_tuple;
 336         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 337
 338         /* 3) The per-protocol part of the manip is made to map into
 339          * the range to make a unique tuple.
 340          */
 341
 342         /* Only bother mapping if it's not already in range and unique */
 343         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 344                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 345                         if (l4proto->in_range(tuple, maniptype,
 346                                               &range->min_proto,
 347                                               &range->max_proto) &&
 348                             (range->min_proto.all == range->max_proto.all ||
 349                              !nf_nat_used_tuple(tuple, ct)))
 350                                 goto out;
 351                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 352                         goto out;
 353                 }
 354         }
 355
 356         /* Last change: get protocol to try to obtain unique tuple. */
 357         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 358 out:
 359         rcu_read_unlock();
 360 }
 361
 362 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 363 {
 364         struct nf_conn_nat *nat = nfct_nat(ct);
 365         if (nat)
 366                 return nat;
 367
 368         if (!nf_ct_is_confirmed(ct))
 369                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 370
 371         return nat;
 372 }
 373 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 374
 375 unsigned int
 376 nf_nat_setup_info(struct nf_conn *ct,
 377                   const struct nf_nat_range *range,
 378                   enum nf_nat_manip_type maniptype)
 379 {
 380         struct net *net = nf_ct_net(ct);
 381         struct nf_conntrack_tuple curr_tuple, new_tuple;
 382         struct nf_conn_nat *nat;
 383
 384         /* nat helper or nfctnetlink also setup binding */
 385         nat = nf_ct_nat_ext_add(ct);
 386         if (nat == NULL)
 387                 return NF_ACCEPT;
 388
 389         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 390                      maniptype == NF_NAT_MANIP_DST);
 391         BUG_ON(nf_nat_initialized(ct, maniptype));
 392
 393         /* What we've got will look like inverse of reply. Normally
 394          * this is what is in the conntrack, except for prior
 395          * manipulations (future optimization: if num_manips == 0,
 396          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 397          */
 398         nf_ct_invert_tuplepr(&curr_tuple,
 399                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 400
 401         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 402
 403         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 404                 struct nf_conntrack_tuple reply;
 405
 406                 /* Alter conntrack table so will recognize replies. */
 407                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 408                 nf_conntrack_alter_reply(ct, &reply);
 409
 410                 /* Non-atomic: we own this at the moment. */
 411                 if (maniptype == NF_NAT_MANIP_SRC)
 412                         ct->status |= IPS_SRC_NAT;
 413                 else
 414                         ct->status |= IPS_DST_NAT;
 415
 416                 if (nfct_help(ct))
 417                         nfct_seqadj_ext_add(ct);
 418         }
 419
 420         if (maniptype == NF_NAT_MANIP_SRC) {
 421                 unsigned int srchash;
 422
 423                 srchash = hash_by_src(net, nf_ct_zone(ct),
 424                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 425                 spin_lock_bh(&nf_nat_lock);
 426                 /* nf_conntrack_alter_reply might re-allocate extension aera */
 427                 nat = nfct_nat(ct);
 428                 nat->ct = ct;
 429                 hlist_add_head_rcu(&nat->bysource,
 430                                    &net->ct.nat_bysource[srchash]);
 431                 spin_unlock_bh(&nf_nat_lock);
 432         }
 433
 434         /* It's done. */
 435         if (maniptype == NF_NAT_MANIP_DST)
 436                 ct->status |= IPS_DST_NAT_DONE;
 437         else
 438                 ct->status |= IPS_SRC_NAT_DONE;
 439
 440         return NF_ACCEPT;
 441 }
 442 EXPORT_SYMBOL(nf_nat_setup_info);
 443
 444 static unsigned int
 445 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 446 {
 447         /* Force range to this IP; let proto decide mapping for
 448          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 449          * Use reply in case it's already been mangled (eg local packet).
 450          */
 451         union nf_inet_addr ip =
 452                 (manip == NF_NAT_MANIP_SRC ?
 453                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 454                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 455         struct nf_nat_range range = {
 456                 .flags          = NF_NAT_RANGE_MAP_IPS,
 457                 .min_addr       = ip,
 458                 .max_addr       = ip,
 459         };
 460         return nf_nat_setup_info(ct, &range, manip);
 461 }
 462
 463 unsigned int
 464 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 465 {
 466         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 467 }
 468 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 469
 470 /* Do packet manipulations according to nf_nat_setup_info. */
 471 unsigned int nf_nat_packet(struct nf_conn *ct,
 472                            enum ip_conntrack_info ctinfo,
 473                            unsigned int hooknum,
 474                            struct sk_buff *skb)
 475 {
 476         const struct nf_nat_l3proto *l3proto;
 477         const struct nf_nat_l4proto *l4proto;
 478         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 479         unsigned long statusbit;
 480         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 481
 482         if (mtype == NF_NAT_MANIP_SRC)
 483                 statusbit = IPS_SRC_NAT;
 484         else
 485                 statusbit = IPS_DST_NAT;
 486
 487         /* Invert if this is reply dir. */
 488         if (dir == IP_CT_DIR_REPLY)
 489                 statusbit ^= IPS_NAT_MASK;
 490
 491         /* Non-atomic: these bits don't change. */
 492         if (ct->status & statusbit) {
 493                 struct nf_conntrack_tuple target;
 494
 495                 /* We are aiming to look like inverse of other direction. */
 496                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 497
 498                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 499                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 500                                                 target.dst.protonum);
 501                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 502                         return NF_DROP;
 503         }
 504         return NF_ACCEPT;
 505 }
 506 EXPORT_SYMBOL_GPL(nf_nat_packet);
 507
 508 struct nf_nat_proto_clean {
 509         u8      l3proto;
 510         u8      l4proto;
 511 };
 512
 513 /* kill conntracks with affected NAT section */
 514 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 515 {
 516         const struct nf_nat_proto_clean *clean = data;
 517         struct nf_conn_nat *nat = nfct_nat(i);
 518
 519         if (!nat)
 520                 return 0;
 521
 522         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 523             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 524                 return 0;
 525
 526         return i->status & IPS_NAT_MASK ? 1 : 0;
 527 }
 528
 529 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 530 {
 531         struct nf_conn_nat *nat = nfct_nat(ct);
 532
 533         if (nf_nat_proto_remove(ct, data))
 534                 return 1;
 535
 536         if (!nat || !nat->ct)
 537                 return 0;
 538
 539         /* This netns is being destroyed, and conntrack has nat null binding.
 540          * Remove it from bysource hash, as the table will be freed soon.
 541          *
 542          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 543          * will delete entry from already-freed table.
 544          */
 545         if (!del_timer(&ct->timeout))
 546                 return 1;
 547
 548         spin_lock_bh(&nf_nat_lock);
 549         hlist_del_rcu(&nat->bysource);
 550         ct->status &= ~IPS_NAT_DONE_MASK;
 551         nat->ct = NULL;
 552         spin_unlock_bh(&nf_nat_lock);
 553
 554         add_timer(&ct->timeout);
 555
 556         /* don't delete conntrack.  Although that would make things a lot
 557          * simpler, we'd end up flushing all conntracks on nat rmmod.
 558          */
 559         return 0;
 560 }
 561
 562 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 563 {
 564         struct nf_nat_proto_clean clean = {
 565                 .l3proto = l3proto,
 566                 .l4proto = l4proto,
 567         };
 568         struct net *net;
 569
 570         rtnl_lock();
 571         for_each_net(net)
 572                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 573         rtnl_unlock();
 574 }
 575
 576 static void nf_nat_l3proto_clean(u8 l3proto)
 577 {
 578         struct nf_nat_proto_clean clean = {
 579                 .l3proto = l3proto,
 580         };
 581         struct net *net;
 582
 583         rtnl_lock();
 584
 585         for_each_net(net)
 586                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 587         rtnl_unlock();
 588 }
 589
 590 /* Protocol registration. */
 591 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 592 {
 593         const struct nf_nat_l4proto **l4protos;
 594         unsigned int i;
 595         int ret = 0;
 596
 597         mutex_lock(&nf_nat_proto_mutex);
 598         if (nf_nat_l4protos[l3proto] == NULL) {
 599                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 600                                    GFP_KERNEL);
 601                 if (l4protos == NULL) {
 602                         ret = -ENOMEM;
 603                         goto out;
 604                 }
 605
 606                 for (i = 0; i < IPPROTO_MAX; i++)
 607                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 608
 609                 /* Before making proto_array visible to lockless readers,
 610                  * we must make sure its content is committed to memory.
 611                  */
 612                 smp_wmb();
 613
 614                 nf_nat_l4protos[l3proto] = l4protos;
 615         }
 616
 617         if (rcu_dereference_protected(
 618                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 619                         lockdep_is_held(&nf_nat_proto_mutex)
 620                         ) != &nf_nat_l4proto_unknown) {
 621                 ret = -EBUSY;
 622                 goto out;
 623         }
 624         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 625  out:
 626         mutex_unlock(&nf_nat_proto_mutex);
 627         return ret;
 628 }
 629 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 630
 631 /* No one stores the protocol anywhere; simply delete it. */
 632 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 633 {
 634         mutex_lock(&nf_nat_proto_mutex);
 635         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 636                          &nf_nat_l4proto_unknown);
 637         mutex_unlock(&nf_nat_proto_mutex);
 638         synchronize_rcu();
 639
 640         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 641 }
 642 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 643
 644 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 645 {
 646         int err;
 647
 648         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 649         if (err < 0)
 650                 return err;
 651
 652         mutex_lock(&nf_nat_proto_mutex);
 653         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 654                          &nf_nat_l4proto_tcp);
 655         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 656                          &nf_nat_l4proto_udp);
 657         mutex_unlock(&nf_nat_proto_mutex);
 658
 659         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 660         return 0;
 661 }
 662 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 663
 664 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 665 {
 666         mutex_lock(&nf_nat_proto_mutex);
 667         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 668         mutex_unlock(&nf_nat_proto_mutex);
 669         synchronize_rcu();
 670
 671         nf_nat_l3proto_clean(l3proto->l3proto);
 672         nf_ct_l3proto_module_put(l3proto->l3proto);
 673 }
 674 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 675
 676 /* No one using conntrack by the time this called. */
 677 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 678 {
 679         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 680
 681         if (nat == NULL || nat->ct == NULL)
 682                 return;
 683
 684         NF_CT_ASSERT(nat->ct->status & IPS_SRC_NAT_DONE);
 685
 686         spin_lock_bh(&nf_nat_lock);
 687         hlist_del_rcu(&nat->bysource);
 688         spin_unlock_bh(&nf_nat_lock);
 689 }
 690
 691 static void nf_nat_move_storage(void *new, void *old)
 692 {
 693         struct nf_conn_nat *new_nat = new;
 694         struct nf_conn_nat *old_nat = old;
 695         struct nf_conn *ct = old_nat->ct;
 696
 697         if (!ct || !(ct->status & IPS_SRC_NAT_DONE))
 698                 return;
 699
 700         spin_lock_bh(&nf_nat_lock);
 701         hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
 702         spin_unlock_bh(&nf_nat_lock);
 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         .move           = nf_nat_move_storage,
 710         .id             = NF_CT_EXT_NAT,
 711         .flags          = NF_CT_EXT_F_PREALLOC,
 712 };
 713
 714 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 715
 716 #include <linux/netfilter/nfnetlink.h>
 717 #include <linux/netfilter/nfnetlink_conntrack.h>
 718
 719 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 720         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 721         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 722 };
 723
 724 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 725                                      const struct nf_conn *ct,
 726                                      struct nf_nat_range *range)
 727 {
 728         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 729         const struct nf_nat_l4proto *l4proto;
 730         int err;
 731
 732         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 733         if (err < 0)
 734                 return err;
 735
 736         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 737         if (l4proto->nlattr_to_range)
 738                 err = l4proto->nlattr_to_range(tb, range);
 739
 740         return err;
 741 }
 742
 743 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 744         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 745         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 746         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 747         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 748         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 749 };
 750
 751 static int
 752 nfnetlink_parse_nat(const struct nlattr *nat,
 753                     const struct nf_conn *ct, struct nf_nat_range *range,
 754                     const struct nf_nat_l3proto *l3proto)
 755 {
 756         struct nlattr *tb[CTA_NAT_MAX+1];
 757         int err;
 758
 759         memset(range, 0, sizeof(*range));
 760
 761         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 762         if (err < 0)
 763                 return err;
 764
 765         err = l3proto->nlattr_to_range(tb, range);
 766         if (err < 0)
 767                 return err;
 768
 769         if (!tb[CTA_NAT_PROTO])
 770                 return 0;
 771
 772         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 773 }
 774
 775 /* This function is called under rcu_read_lock() */
 776 static int
 777 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 778                           enum nf_nat_manip_type manip,
 779                           const struct nlattr *attr)
 780 {
 781         struct nf_nat_range range;
 782         const struct nf_nat_l3proto *l3proto;
 783         int err;
 784
 785         /* Should not happen, restricted to creating new conntracks
 786          * via ctnetlink.
 787          */
 788         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 789                 return -EEXIST;
 790
 791         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 792          * attach the null binding, otherwise this may oops.
 793          */
 794         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 795         if (l3proto == NULL)
 796                 return -EAGAIN;
 797
 798         /* No NAT information has been passed, allocate the null-binding */
 799         if (attr == NULL)
 800                 return __nf_nat_alloc_null_binding(ct, manip);
 801
 802         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 803         if (err < 0)
 804                 return err;
 805
 806         return nf_nat_setup_info(ct, &range, manip);
 807 }
 808 #else
 809 static int
 810 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 811                           enum nf_nat_manip_type manip,
 812                           const struct nlattr *attr)
 813 {
 814         return -EOPNOTSUPP;
 815 }
 816 #endif
 817
 818 static int __net_init nf_nat_net_init(struct net *net)
 819 {
 820         /* Leave them the same for the moment. */
 821         net->ct.nat_htable_size = net->ct.htable_size;
 822         net->ct.nat_bysource = nf_ct_alloc_hashtable(&net->ct.nat_htable_size, 0);
 823         if (!net->ct.nat_bysource)
 824                 return -ENOMEM;
 825         return 0;
 826 }
 827
 828 static void __net_exit nf_nat_net_exit(struct net *net)
 829 {
 830         struct nf_nat_proto_clean clean = {};
 831
 832         nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean, 0, 0);
 833         synchronize_rcu();
 834         nf_ct_free_hashtable(net->ct.nat_bysource, net->ct.nat_htable_size);
 835 }
 836
 837 static struct pernet_operations nf_nat_net_ops = {
 838         .init = nf_nat_net_init,
 839         .exit = nf_nat_net_exit,
 840 };
 841
 842 static struct nf_ct_helper_expectfn follow_master_nat = {
 843         .name           = "nat-follow-master",
 844         .expectfn       = nf_nat_follow_master,
 845 };
 846
 847 static int __init nf_nat_init(void)
 848 {
 849         int ret;
 850
 851         ret = nf_ct_extend_register(&nat_extend);
 852         if (ret < 0) {
 853                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 854                 return ret;
 855         }
 856
 857         ret = register_pernet_subsys(&nf_nat_net_ops);
 858         if (ret < 0)
 859                 goto cleanup_extend;
 860
 861         nf_ct_helper_expectfn_register(&follow_master_nat);
 862
 863         /* Initialize fake conntrack so that NAT will skip it */
 864         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 865
 866         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 867         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 868                            nfnetlink_parse_nat_setup);
 869 #ifdef CONFIG_XFRM
 870         BUG_ON(nf_nat_decode_session_hook != NULL);
 871         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 872 #endif
 873         return 0;
 874
 875  cleanup_extend:
 876         nf_ct_extend_unregister(&nat_extend);
 877         return ret;
 878 }
 879
 880 static void __exit nf_nat_cleanup(void)
 881 {
 882         unsigned int i;
 883
 884         unregister_pernet_subsys(&nf_nat_net_ops);
 885         nf_ct_extend_unregister(&nat_extend);
 886         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 887         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 888 #ifdef CONFIG_XFRM
 889         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 890 #endif
 891         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 892                 kfree(nf_nat_l4protos[i]);
 893         synchronize_net();
 894 }
 895
 896 MODULE_LICENSE("GPL");
 897
 898 module_init(nf_nat_init);
 899 module_exit(nf_nat_cleanup);