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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/module.h>
  14 #include <linux/types.h>
  15 #include <linux/timer.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/gfp.h>
  18 #include <net/xfrm.h>
  19 #include <linux/jhash.h>
  20 #include <linux/rtnetlink.h>
  21
  22 #include <net/netfilter/nf_conntrack.h>
  23 #include <net/netfilter/nf_conntrack_core.h>
  24 #include <net/netfilter/nf_nat.h>
  25 #include <net/netfilter/nf_nat_l3proto.h>
  26 #include <net/netfilter/nf_nat_l4proto.h>
  27 #include <net/netfilter/nf_nat_core.h>
  28 #include <net/netfilter/nf_nat_helper.h>
  29 #include <net/netfilter/nf_conntrack_helper.h>
  30 #include <net/netfilter/nf_conntrack_seqadj.h>
  31 #include <net/netfilter/nf_conntrack_zones.h>
  32 #include <linux/netfilter/nf_nat.h>
  33
  34 #include "nf_internals.h"
  35
  36 static spinlock_t nf_nat_locks[CONNTRACK_LOCKS];
  37
  38 static DEFINE_MUTEX(nf_nat_proto_mutex);
  39 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
  40                                                 __read_mostly;
  41 static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO]
  42                                                 __read_mostly;
  43 static unsigned int nat_net_id __read_mostly;
  44
  45 static struct hlist_head *nf_nat_bysource __read_mostly;
  46 static unsigned int nf_nat_htable_size __read_mostly;
  47 static unsigned int nf_nat_hash_rnd __read_mostly;
  48
  49 struct nf_nat_lookup_hook_priv {
  50         struct nf_hook_entries __rcu *entries;
  51
  52         struct rcu_head rcu_head;
  53 };
  54
  55 struct nf_nat_hooks_net {
  56         struct nf_hook_ops *nat_hook_ops;
  57         unsigned int users;
  58 };
  59
  60 struct nat_net {
  61         struct nf_nat_hooks_net nat_proto_net[NFPROTO_NUMPROTO];
  62 };
  63
  64 inline const struct nf_nat_l3proto *
  65 __nf_nat_l3proto_find(u8 family)
  66 {
  67         return rcu_dereference(nf_nat_l3protos[family]);
  68 }
  69
  70 inline const struct nf_nat_l4proto *
  71 __nf_nat_l4proto_find(u8 family, u8 protonum)
  72 {
  73         return rcu_dereference(nf_nat_l4protos[family][protonum]);
  74 }
  75 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
  76
  77 #ifdef CONFIG_XFRM
  78 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  79 {
  80         const struct nf_nat_l3proto *l3proto;
  81         const struct nf_conn *ct;
  82         enum ip_conntrack_info ctinfo;
  83         enum ip_conntrack_dir dir;
  84         unsigned  long statusbit;
  85         u8 family;
  86
  87         ct = nf_ct_get(skb, &ctinfo);
  88         if (ct == NULL)
  89                 return;
  90
  91         family = nf_ct_l3num(ct);
  92         l3proto = __nf_nat_l3proto_find(family);
  93         if (l3proto == NULL)
  94                 return;
  95
  96         dir = CTINFO2DIR(ctinfo);
  97         if (dir == IP_CT_DIR_ORIGINAL)
  98                 statusbit = IPS_DST_NAT;
  99         else
 100                 statusbit = IPS_SRC_NAT;
 101
 102         l3proto->decode_session(skb, ct, dir, statusbit, fl);
 103 }
 104
 105 int nf_xfrm_me_harder(struct net *net, struct sk_buff *skb, unsigned int family)
 106 {
 107         struct flowi fl;
 108         unsigned int hh_len;
 109         struct dst_entry *dst;
 110         struct sock *sk = skb->sk;
 111         int err;
 112
 113         err = xfrm_decode_session(skb, &fl, family);
 114         if (err < 0)
 115                 return err;
 116
 117         dst = skb_dst(skb);
 118         if (dst->xfrm)
 119                 dst = ((struct xfrm_dst *)dst)->route;
 120         dst_hold(dst);
 121
 122         if (sk && !net_eq(net, sock_net(sk)))
 123                 sk = NULL;
 124
 125         dst = xfrm_lookup(net, dst, &fl, sk, 0);
 126         if (IS_ERR(dst))
 127                 return PTR_ERR(dst);
 128
 129         skb_dst_drop(skb);
 130         skb_dst_set(skb, dst);
 131
 132         /* Change in oif may mean change in hh_len. */
 133         hh_len = skb_dst(skb)->dev->hard_header_len;
 134         if (skb_headroom(skb) < hh_len &&
 135             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 136                 return -ENOMEM;
 137         return 0;
 138 }
 139 EXPORT_SYMBOL(nf_xfrm_me_harder);
 140 #endif /* CONFIG_XFRM */
 141
 142 /* We keep an extra hash for each conntrack, for fast searching. */
 143 static unsigned int
 144 hash_by_src(const struct net *n, const struct nf_conntrack_tuple *tuple)
 145 {
 146         unsigned int hash;
 147
 148         get_random_once(&nf_nat_hash_rnd, sizeof(nf_nat_hash_rnd));
 149
 150         /* Original src, to ensure we map it consistently if poss. */
 151         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 152                       tuple->dst.protonum ^ nf_nat_hash_rnd ^ net_hash_mix(n));
 153
 154         return reciprocal_scale(hash, nf_nat_htable_size);
 155 }
 156
 157 /* Is this tuple already taken? (not by us) */
 158 int
 159 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 160                   const struct nf_conn *ignored_conntrack)
 161 {
 162         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 163          * incoming ones.  NAT means they don't have a fixed mapping,
 164          * so we invert the tuple and look for the incoming reply.
 165          *
 166          * We could keep a separate hash if this proves too slow.
 167          */
 168         struct nf_conntrack_tuple reply;
 169
 170         nf_ct_invert_tuplepr(&reply, tuple);
 171         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 172 }
 173 EXPORT_SYMBOL(nf_nat_used_tuple);
 174
 175 /* If we source map this tuple so reply looks like reply_tuple, will
 176  * that meet the constraints of range.
 177  */
 178 static int in_range(const struct nf_nat_l3proto *l3proto,
 179                     const struct nf_nat_l4proto *l4proto,
 180                     const struct nf_conntrack_tuple *tuple,
 181                     const struct nf_nat_range2 *range)
 182 {
 183         /* If we are supposed to map IPs, then we must be in the
 184          * range specified, otherwise let this drag us onto a new src IP.
 185          */
 186         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 187             !l3proto->in_range(tuple, range))
 188                 return 0;
 189
 190         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 191             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 192                               &range->min_proto, &range->max_proto))
 193                 return 1;
 194
 195         return 0;
 196 }
 197
 198 static inline int
 199 same_src(const struct nf_conn *ct,
 200          const struct nf_conntrack_tuple *tuple)
 201 {
 202         const struct nf_conntrack_tuple *t;
 203
 204         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 205         return (t->dst.protonum == tuple->dst.protonum &&
 206                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 207                 t->src.u.all == tuple->src.u.all);
 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_range2 *range)
 219 {
 220         unsigned int h = hash_by_src(net, tuple);
 221         const struct nf_conn *ct;
 222
 223         hlist_for_each_entry_rcu(ct, &nf_nat_bysource[h], nat_bysource) {
 224                 if (same_src(ct, tuple) &&
 225                     net_eq(net, nf_ct_net(ct)) &&
 226                     nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) {
 227                         /* Copy source part from reply tuple. */
 228                         nf_ct_invert_tuplepr(result,
 229                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 230                         result->dst = tuple->dst;
 231
 232                         if (in_range(l3proto, l4proto, result, range))
 233                                 return 1;
 234                 }
 235         }
 236         return 0;
 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_range2 *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_range2 *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 (!(range->flags & NF_NAT_RANGE_PROTO_OFFSET) &&
 374                             l4proto->in_range(tuple, maniptype,
 375                                   &range->min_proto,
 376                                   &range->max_proto) &&
 377                             (range->min_proto.all == range->max_proto.all ||
 378                              !nf_nat_used_tuple(tuple, ct)))
 379                                 goto out;
 380                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 381                         goto out;
 382                 }
 383         }
 384
 385         /* Last chance: get protocol to try to obtain unique tuple. */
 386         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 387 out:
 388         rcu_read_unlock();
 389 }
 390
 391 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 392 {
 393         struct nf_conn_nat *nat = nfct_nat(ct);
 394         if (nat)
 395                 return nat;
 396
 397         if (!nf_ct_is_confirmed(ct))
 398                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 399
 400         return nat;
 401 }
 402 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 403
 404 unsigned int
 405 nf_nat_setup_info(struct nf_conn *ct,
 406                   const struct nf_nat_range2 *range,
 407                   enum nf_nat_manip_type maniptype)
 408 {
 409         struct net *net = nf_ct_net(ct);
 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         WARN_ON(maniptype != NF_NAT_MANIP_SRC &&
 417                 maniptype != NF_NAT_MANIP_DST);
 418
 419         if (WARN_ON(nf_nat_initialized(ct, maniptype)))
 420                 return NF_DROP;
 421
 422         /* What we've got will look like inverse of reply. Normally
 423          * this is what is in the conntrack, except for prior
 424          * manipulations (future optimization: if num_manips == 0,
 425          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 426          */
 427         nf_ct_invert_tuplepr(&curr_tuple,
 428                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 429
 430         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 431
 432         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 433                 struct nf_conntrack_tuple reply;
 434
 435                 /* Alter conntrack table so will recognize replies. */
 436                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 437                 nf_conntrack_alter_reply(ct, &reply);
 438
 439                 /* Non-atomic: we own this at the moment. */
 440                 if (maniptype == NF_NAT_MANIP_SRC)
 441                         ct->status |= IPS_SRC_NAT;
 442                 else
 443                         ct->status |= IPS_DST_NAT;
 444
 445                 if (nfct_help(ct) && !nfct_seqadj(ct))
 446                         if (!nfct_seqadj_ext_add(ct))
 447                                 return NF_DROP;
 448         }
 449
 450         if (maniptype == NF_NAT_MANIP_SRC) {
 451                 unsigned int srchash;
 452                 spinlock_t *lock;
 453
 454                 srchash = hash_by_src(net,
 455                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 456                 lock = &nf_nat_locks[srchash % CONNTRACK_LOCKS];
 457                 spin_lock_bh(lock);
 458                 hlist_add_head_rcu(&ct->nat_bysource,
 459                                    &nf_nat_bysource[srchash]);
 460                 spin_unlock_bh(lock);
 461         }
 462
 463         /* It's done. */
 464         if (maniptype == NF_NAT_MANIP_DST)
 465                 ct->status |= IPS_DST_NAT_DONE;
 466         else
 467                 ct->status |= IPS_SRC_NAT_DONE;
 468
 469         return NF_ACCEPT;
 470 }
 471 EXPORT_SYMBOL(nf_nat_setup_info);
 472
 473 static unsigned int
 474 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 475 {
 476         /* Force range to this IP; let proto decide mapping for
 477          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 478          * Use reply in case it's already been mangled (eg local packet).
 479          */
 480         union nf_inet_addr ip =
 481                 (manip == NF_NAT_MANIP_SRC ?
 482                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 483                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 484         struct nf_nat_range2 range = {
 485                 .flags          = NF_NAT_RANGE_MAP_IPS,
 486                 .min_addr       = ip,
 487                 .max_addr       = ip,
 488         };
 489         return nf_nat_setup_info(ct, &range, manip);
 490 }
 491
 492 unsigned int
 493 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 494 {
 495         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 496 }
 497 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 498
 499 static unsigned int nf_nat_manip_pkt(struct sk_buff *skb, struct nf_conn *ct,
 500                                      enum nf_nat_manip_type mtype,
 501                                      enum ip_conntrack_dir dir)
 502 {
 503         const struct nf_nat_l3proto *l3proto;
 504         const struct nf_nat_l4proto *l4proto;
 505         struct nf_conntrack_tuple target;
 506
 507         /* We are aiming to look like inverse of other direction. */
 508         nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 509
 510         l3proto = __nf_nat_l3proto_find(target.src.l3num);
 511         l4proto = __nf_nat_l4proto_find(target.src.l3num,
 512                                         target.dst.protonum);
 513         if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 514                 return NF_DROP;
 515
 516         return NF_ACCEPT;
 517 }
 518
 519 /* Do packet manipulations according to nf_nat_setup_info. */
 520 unsigned int nf_nat_packet(struct nf_conn *ct,
 521                            enum ip_conntrack_info ctinfo,
 522                            unsigned int hooknum,
 523                            struct sk_buff *skb)
 524 {
 525         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 526         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 527         unsigned int verdict = NF_ACCEPT;
 528         unsigned long statusbit;
 529
 530         if (mtype == NF_NAT_MANIP_SRC)
 531                 statusbit = IPS_SRC_NAT;
 532         else
 533                 statusbit = IPS_DST_NAT;
 534
 535         /* Invert if this is reply dir. */
 536         if (dir == IP_CT_DIR_REPLY)
 537                 statusbit ^= IPS_NAT_MASK;
 538
 539         /* Non-atomic: these bits don't change. */
 540         if (ct->status & statusbit)
 541                 verdict = nf_nat_manip_pkt(skb, ct, mtype, dir);
 542
 543         return verdict;
 544 }
 545 EXPORT_SYMBOL_GPL(nf_nat_packet);
 546
 547 unsigned int
 548 nf_nat_inet_fn(void *priv, struct sk_buff *skb,
 549                const struct nf_hook_state *state)
 550 {
 551         struct nf_conn *ct;
 552         enum ip_conntrack_info ctinfo;
 553         struct nf_conn_nat *nat;
 554         /* maniptype == SRC for postrouting. */
 555         enum nf_nat_manip_type maniptype = HOOK2MANIP(state->hook);
 556
 557         ct = nf_ct_get(skb, &ctinfo);
 558         /* Can't track?  It's not due to stress, or conntrack would
 559          * have dropped it.  Hence it's the user's responsibilty to
 560          * packet filter it out, or implement conntrack/NAT for that
 561          * protocol. 8) --RR
 562          */
 563         if (!ct)
 564                 return NF_ACCEPT;
 565
 566         nat = nfct_nat(ct);
 567
 568         switch (ctinfo) {
 569         case IP_CT_RELATED:
 570         case IP_CT_RELATED_REPLY:
 571                 /* Only ICMPs can be IP_CT_IS_REPLY.  Fallthrough */
 572         case IP_CT_NEW:
 573                 /* Seen it before?  This can happen for loopback, retrans,
 574                  * or local packets.
 575                  */
 576                 if (!nf_nat_initialized(ct, maniptype)) {
 577                         struct nf_nat_lookup_hook_priv *lpriv = priv;
 578                         struct nf_hook_entries *e = rcu_dereference(lpriv->entries);
 579                         unsigned int ret;
 580                         int i;
 581
 582                         if (!e)
 583                                 goto null_bind;
 584
 585                         for (i = 0; i < e->num_hook_entries; i++) {
 586                                 ret = e->hooks[i].hook(e->hooks[i].priv, skb,
 587                                                        state);
 588                                 if (ret != NF_ACCEPT)
 589                                         return ret;
 590                                 if (nf_nat_initialized(ct, maniptype))
 591                                         goto do_nat;
 592                         }
 593 null_bind:
 594                         ret = nf_nat_alloc_null_binding(ct, state->hook);
 595                         if (ret != NF_ACCEPT)
 596                                 return ret;
 597                 } else {
 598                         pr_debug("Already setup manip %s for ct %p (status bits 0x%lx)\n",
 599                                  maniptype == NF_NAT_MANIP_SRC ? "SRC" : "DST",
 600                                  ct, ct->status);
 601                         if (nf_nat_oif_changed(state->hook, ctinfo, nat,
 602                                                state->out))
 603                                 goto oif_changed;
 604                 }
 605                 break;
 606         default:
 607                 /* ESTABLISHED */
 608                 WARN_ON(ctinfo != IP_CT_ESTABLISHED &&
 609                         ctinfo != IP_CT_ESTABLISHED_REPLY);
 610                 if (nf_nat_oif_changed(state->hook, ctinfo, nat, state->out))
 611                         goto oif_changed;
 612         }
 613 do_nat:
 614         return nf_nat_packet(ct, ctinfo, state->hook, skb);
 615
 616 oif_changed:
 617         nf_ct_kill_acct(ct, ctinfo, skb);
 618         return NF_DROP;
 619 }
 620 EXPORT_SYMBOL_GPL(nf_nat_inet_fn);
 621
 622 struct nf_nat_proto_clean {
 623         u8      l3proto;
 624         u8      l4proto;
 625 };
 626
 627 /* kill conntracks with affected NAT section */
 628 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 629 {
 630         const struct nf_nat_proto_clean *clean = data;
 631
 632         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 633             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 634                 return 0;
 635
 636         return i->status & IPS_NAT_MASK ? 1 : 0;
 637 }
 638
 639 static void __nf_nat_cleanup_conntrack(struct nf_conn *ct)
 640 {
 641         unsigned int h;
 642
 643         h = hash_by_src(nf_ct_net(ct), &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 644         spin_lock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 645         hlist_del_rcu(&ct->nat_bysource);
 646         spin_unlock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 647 }
 648
 649 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 650 {
 651         if (nf_nat_proto_remove(ct, data))
 652                 return 1;
 653
 654         /* This module is being removed and conntrack has nat null binding.
 655          * Remove it from bysource hash, as the table will be freed soon.
 656          *
 657          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 658          * will delete entry from already-freed table.
 659          */
 660         if (test_and_clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status))
 661                 __nf_nat_cleanup_conntrack(ct);
 662
 663         /* don't delete conntrack.  Although that would make things a lot
 664          * simpler, we'd end up flushing all conntracks on nat rmmod.
 665          */
 666         return 0;
 667 }
 668
 669 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 670 {
 671         struct nf_nat_proto_clean clean = {
 672                 .l3proto = l3proto,
 673                 .l4proto = l4proto,
 674         };
 675
 676         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 677 }
 678
 679 static void nf_nat_l3proto_clean(u8 l3proto)
 680 {
 681         struct nf_nat_proto_clean clean = {
 682                 .l3proto = l3proto,
 683         };
 684
 685         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 686 }
 687
 688 /* Protocol registration. */
 689 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 690 {
 691         const struct nf_nat_l4proto **l4protos;
 692         unsigned int i;
 693         int ret = 0;
 694
 695         mutex_lock(&nf_nat_proto_mutex);
 696         if (nf_nat_l4protos[l3proto] == NULL) {
 697                 l4protos = kmalloc_array(IPPROTO_MAX,
 698                                          sizeof(struct nf_nat_l4proto *),
 699                                          GFP_KERNEL);
 700                 if (l4protos == NULL) {
 701                         ret = -ENOMEM;
 702                         goto out;
 703                 }
 704
 705                 for (i = 0; i < IPPROTO_MAX; i++)
 706                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 707
 708                 /* Before making proto_array visible to lockless readers,
 709                  * we must make sure its content is committed to memory.
 710                  */
 711                 smp_wmb();
 712
 713                 nf_nat_l4protos[l3proto] = l4protos;
 714         }
 715
 716         if (rcu_dereference_protected(
 717                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 718                         lockdep_is_held(&nf_nat_proto_mutex)
 719                         ) != &nf_nat_l4proto_unknown) {
 720                 ret = -EBUSY;
 721                 goto out;
 722         }
 723         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 724  out:
 725         mutex_unlock(&nf_nat_proto_mutex);
 726         return ret;
 727 }
 728 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 729
 730 /* No one stores the protocol anywhere; simply delete it. */
 731 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 732 {
 733         mutex_lock(&nf_nat_proto_mutex);
 734         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 735                          &nf_nat_l4proto_unknown);
 736         mutex_unlock(&nf_nat_proto_mutex);
 737         synchronize_rcu();
 738
 739         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 740 }
 741 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 742
 743 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 744 {
 745         mutex_lock(&nf_nat_proto_mutex);
 746         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 747                          &nf_nat_l4proto_tcp);
 748         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 749                          &nf_nat_l4proto_udp);
 750 #ifdef CONFIG_NF_NAT_PROTO_DCCP
 751         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_DCCP],
 752                          &nf_nat_l4proto_dccp);
 753 #endif
 754 #ifdef CONFIG_NF_NAT_PROTO_SCTP
 755         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_SCTP],
 756                          &nf_nat_l4proto_sctp);
 757 #endif
 758 #ifdef CONFIG_NF_NAT_PROTO_UDPLITE
 759         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDPLITE],
 760                          &nf_nat_l4proto_udplite);
 761 #endif
 762         mutex_unlock(&nf_nat_proto_mutex);
 763
 764         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 765         return 0;
 766 }
 767 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 768
 769 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 770 {
 771         mutex_lock(&nf_nat_proto_mutex);
 772         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 773         mutex_unlock(&nf_nat_proto_mutex);
 774         synchronize_rcu();
 775
 776         nf_nat_l3proto_clean(l3proto->l3proto);
 777 }
 778 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 779
 780 /* No one using conntrack by the time this called. */
 781 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 782 {
 783         if (ct->status & IPS_SRC_NAT_DONE)
 784                 __nf_nat_cleanup_conntrack(ct);
 785 }
 786
 787 static struct nf_ct_ext_type nat_extend __read_mostly = {
 788         .len            = sizeof(struct nf_conn_nat),
 789         .align          = __alignof__(struct nf_conn_nat),
 790         .destroy        = nf_nat_cleanup_conntrack,
 791         .id             = NF_CT_EXT_NAT,
 792 };
 793
 794 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 795
 796 #include <linux/netfilter/nfnetlink.h>
 797 #include <linux/netfilter/nfnetlink_conntrack.h>
 798
 799 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 800         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 801         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 802 };
 803
 804 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 805                                      const struct nf_conn *ct,
 806                                      struct nf_nat_range2 *range)
 807 {
 808         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 809         const struct nf_nat_l4proto *l4proto;
 810         int err;
 811
 812         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr,
 813                                protonat_nla_policy, NULL);
 814         if (err < 0)
 815                 return err;
 816
 817         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 818         if (l4proto->nlattr_to_range)
 819                 err = l4proto->nlattr_to_range(tb, range);
 820
 821         return err;
 822 }
 823
 824 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 825         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 826         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 827         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 828         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 829         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 830 };
 831
 832 static int
 833 nfnetlink_parse_nat(const struct nlattr *nat,
 834                     const struct nf_conn *ct, struct nf_nat_range2 *range,
 835                     const struct nf_nat_l3proto *l3proto)
 836 {
 837         struct nlattr *tb[CTA_NAT_MAX+1];
 838         int err;
 839
 840         memset(range, 0, sizeof(*range));
 841
 842         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy, NULL);
 843         if (err < 0)
 844                 return err;
 845
 846         err = l3proto->nlattr_to_range(tb, range);
 847         if (err < 0)
 848                 return err;
 849
 850         if (!tb[CTA_NAT_PROTO])
 851                 return 0;
 852
 853         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 854 }
 855
 856 /* This function is called under rcu_read_lock() */
 857 static int
 858 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 859                           enum nf_nat_manip_type manip,
 860                           const struct nlattr *attr)
 861 {
 862         struct nf_nat_range2 range;
 863         const struct nf_nat_l3proto *l3proto;
 864         int err;
 865
 866         /* Should not happen, restricted to creating new conntracks
 867          * via ctnetlink.
 868          */
 869         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 870                 return -EEXIST;
 871
 872         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 873          * attach the null binding, otherwise this may oops.
 874          */
 875         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 876         if (l3proto == NULL)
 877                 return -EAGAIN;
 878
 879         /* No NAT information has been passed, allocate the null-binding */
 880         if (attr == NULL)
 881                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
 882
 883         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 884         if (err < 0)
 885                 return err;
 886
 887         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 888 }
 889 #else
 890 static int
 891 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 892                           enum nf_nat_manip_type manip,
 893                           const struct nlattr *attr)
 894 {
 895         return -EOPNOTSUPP;
 896 }
 897 #endif
 898
 899 static struct nf_ct_helper_expectfn follow_master_nat = {
 900         .name           = "nat-follow-master",
 901         .expectfn       = nf_nat_follow_master,
 902 };
 903
 904 int nf_nat_register_fn(struct net *net, const struct nf_hook_ops *ops,
 905                        const struct nf_hook_ops *orig_nat_ops, unsigned int ops_count)
 906 {
 907         struct nat_net *nat_net = net_generic(net, nat_net_id);
 908         struct nf_nat_hooks_net *nat_proto_net;
 909         struct nf_nat_lookup_hook_priv *priv;
 910         unsigned int hooknum = ops->hooknum;
 911         struct nf_hook_ops *nat_ops;
 912         int i, ret;
 913
 914         if (WARN_ON_ONCE(ops->pf >= ARRAY_SIZE(nat_net->nat_proto_net)))
 915                 return -EINVAL;
 916
 917         nat_proto_net = &nat_net->nat_proto_net[ops->pf];
 918
 919         for (i = 0; i < ops_count; i++) {
 920                 if (WARN_ON(orig_nat_ops[i].pf != ops->pf))
 921                         return -EINVAL;
 922                 if (orig_nat_ops[i].hooknum == hooknum) {
 923                         hooknum = i;
 924                         break;
 925                 }
 926         }
 927
 928         if (WARN_ON_ONCE(i == ops_count))
 929                 return -EINVAL;
 930
 931         mutex_lock(&nf_nat_proto_mutex);
 932         if (!nat_proto_net->nat_hook_ops) {
 933                 WARN_ON(nat_proto_net->users != 0);
 934
 935                 nat_ops = kmemdup(orig_nat_ops, sizeof(*orig_nat_ops) * ops_count, GFP_KERNEL);
 936                 if (!nat_ops) {
 937                         mutex_unlock(&nf_nat_proto_mutex);
 938                         return -ENOMEM;
 939                 }
 940
 941                 for (i = 0; i < ops_count; i++) {
 942                         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 943                         if (priv) {
 944                                 nat_ops[i].priv = priv;
 945                                 continue;
 946                         }
 947                         mutex_unlock(&nf_nat_proto_mutex);
 948                         while (i)
 949                                 kfree(nat_ops[--i].priv);
 950                         kfree(nat_ops);
 951                         return -ENOMEM;
 952                 }
 953
 954                 ret = nf_register_net_hooks(net, nat_ops, ops_count);
 955                 if (ret < 0) {
 956                         mutex_unlock(&nf_nat_proto_mutex);
 957                         for (i = 0; i < ops_count; i++)
 958                                 kfree(nat_ops[i].priv);
 959                         kfree(nat_ops);
 960                         return ret;
 961                 }
 962
 963                 nat_proto_net->nat_hook_ops = nat_ops;
 964         }
 965
 966         nat_ops = nat_proto_net->nat_hook_ops;
 967         priv = nat_ops[hooknum].priv;
 968         if (WARN_ON_ONCE(!priv)) {
 969                 mutex_unlock(&nf_nat_proto_mutex);
 970                 return -EOPNOTSUPP;
 971         }
 972
 973         ret = nf_hook_entries_insert_raw(&priv->entries, ops);
 974         if (ret == 0)
 975                 nat_proto_net->users++;
 976
 977         mutex_unlock(&nf_nat_proto_mutex);
 978         return ret;
 979 }
 980 EXPORT_SYMBOL_GPL(nf_nat_register_fn);
 981
 982 void nf_nat_unregister_fn(struct net *net, const struct nf_hook_ops *ops,
 983                           unsigned int ops_count)
 984 {
 985         struct nat_net *nat_net = net_generic(net, nat_net_id);
 986         struct nf_nat_hooks_net *nat_proto_net;
 987         struct nf_nat_lookup_hook_priv *priv;
 988         struct nf_hook_ops *nat_ops;
 989         int hooknum = ops->hooknum;
 990         int i;
 991
 992         if (ops->pf >= ARRAY_SIZE(nat_net->nat_proto_net))
 993                 return;
 994
 995         nat_proto_net = &nat_net->nat_proto_net[ops->pf];
 996
 997         mutex_lock(&nf_nat_proto_mutex);
 998         if (WARN_ON(nat_proto_net->users == 0))
 999                 goto unlock;
1000
1001         nat_proto_net->users--;
1002
1003         nat_ops = nat_proto_net->nat_hook_ops;
1004         for (i = 0; i < ops_count; i++) {
1005                 if (nat_ops[i].hooknum == hooknum) {
1006                         hooknum = i;
1007                         break;
1008                 }
1009         }
1010         if (WARN_ON_ONCE(i == ops_count))
1011                 goto unlock;
1012         priv = nat_ops[hooknum].priv;
1013         nf_hook_entries_delete_raw(&priv->entries, ops);
1014
1015         if (nat_proto_net->users == 0) {
1016                 nf_unregister_net_hooks(net, nat_ops, ops_count);
1017
1018                 for (i = 0; i < ops_count; i++) {
1019                         priv = nat_ops[i].priv;
1020                         kfree_rcu(priv, rcu_head);
1021                 }
1022
1023                 nat_proto_net->nat_hook_ops = NULL;
1024                 kfree(nat_ops);
1025         }
1026 unlock:
1027         mutex_unlock(&nf_nat_proto_mutex);
1028 }
1029 EXPORT_SYMBOL_GPL(nf_nat_unregister_fn);
1030
1031 static struct pernet_operations nat_net_ops = {
1032         .id = &nat_net_id,
1033         .size = sizeof(struct nat_net),
1034 };
1035
1036 static struct nf_nat_hook nat_hook = {
1037         .parse_nat_setup        = nfnetlink_parse_nat_setup,
1038 #ifdef CONFIG_XFRM
1039         .decode_session         = __nf_nat_decode_session,
1040 #endif
1041         .manip_pkt              = nf_nat_manip_pkt,
1042 };
1043
1044 static int __init nf_nat_init(void)
1045 {
1046         int ret, i;
1047
1048         /* Leave them the same for the moment. */
1049         nf_nat_htable_size = nf_conntrack_htable_size;
1050         if (nf_nat_htable_size < CONNTRACK_LOCKS)
1051                 nf_nat_htable_size = CONNTRACK_LOCKS;
1052
1053         nf_nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, 0);
1054         if (!nf_nat_bysource)
1055                 return -ENOMEM;
1056
1057         ret = nf_ct_extend_register(&nat_extend);
1058         if (ret < 0) {
1059                 kvfree(nf_nat_bysource);
1060                 pr_err("Unable to register extension\n");
1061                 return ret;
1062         }
1063
1064         for (i = 0; i < CONNTRACK_LOCKS; i++)
1065                 spin_lock_init(&nf_nat_locks[i]);
1066
1067         ret = register_pernet_subsys(&nat_net_ops);
1068         if (ret < 0) {
1069                 nf_ct_extend_unregister(&nat_extend);
1070                 return ret;
1071         }
1072
1073         nf_ct_helper_expectfn_register(&follow_master_nat);
1074
1075         WARN_ON(nf_nat_hook != NULL);
1076         RCU_INIT_POINTER(nf_nat_hook, &nat_hook);
1077
1078         return 0;
1079 }
1080
1081 static void __exit nf_nat_cleanup(void)
1082 {
1083         struct nf_nat_proto_clean clean = {};
1084         unsigned int i;
1085
1086         nf_ct_iterate_destroy(nf_nat_proto_clean, &clean);
1087
1088         nf_ct_extend_unregister(&nat_extend);
1089         nf_ct_helper_expectfn_unregister(&follow_master_nat);
1090         RCU_INIT_POINTER(nf_nat_hook, NULL);
1091
1092         synchronize_rcu();
1093
1094         for (i = 0; i < NFPROTO_NUMPROTO; i++)
1095                 kfree(nf_nat_l4protos[i]);
1096         synchronize_net();
1097         kvfree(nf_nat_bysource);
1098         unregister_pernet_subsys(&nat_net_ops);
1099 }
1100
1101 MODULE_LICENSE("GPL");
1102
1103 module_init(nf_nat_init);
1104 module_exit(nf_nat_cleanup);