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         if (!dst_hold_safe(dst))
 121                 return -EHOSTUNREACH;
 122
 123         if (sk && !net_eq(net, sock_net(sk)))
 124                 sk = NULL;
 125
 126         dst = xfrm_lookup(net, dst, &fl, sk, 0);
 127         if (IS_ERR(dst))
 128                 return PTR_ERR(dst);
 129
 130         skb_dst_drop(skb);
 131         skb_dst_set(skb, dst);
 132
 133         /* Change in oif may mean change in hh_len. */
 134         hh_len = skb_dst(skb)->dev->hard_header_len;
 135         if (skb_headroom(skb) < hh_len &&
 136             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 137                 return -ENOMEM;
 138         return 0;
 139 }
 140 EXPORT_SYMBOL(nf_xfrm_me_harder);
 141 #endif /* CONFIG_XFRM */
 142
 143 /* We keep an extra hash for each conntrack, for fast searching. */
 144 static unsigned int
 145 hash_by_src(const struct net *n, const struct nf_conntrack_tuple *tuple)
 146 {
 147         unsigned int hash;
 148
 149         get_random_once(&nf_nat_hash_rnd, sizeof(nf_nat_hash_rnd));
 150
 151         /* Original src, to ensure we map it consistently if poss. */
 152         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 153                       tuple->dst.protonum ^ nf_nat_hash_rnd ^ net_hash_mix(n));
 154
 155         return reciprocal_scale(hash, nf_nat_htable_size);
 156 }
 157
 158 /* Is this tuple already taken? (not by us) */
 159 int
 160 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 161                   const struct nf_conn *ignored_conntrack)
 162 {
 163         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 164          * incoming ones.  NAT means they don't have a fixed mapping,
 165          * so we invert the tuple and look for the incoming reply.
 166          *
 167          * We could keep a separate hash if this proves too slow.
 168          */
 169         struct nf_conntrack_tuple reply;
 170
 171         nf_ct_invert_tuplepr(&reply, tuple);
 172         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 173 }
 174 EXPORT_SYMBOL(nf_nat_used_tuple);
 175
 176 /* If we source map this tuple so reply looks like reply_tuple, will
 177  * that meet the constraints of range.
 178  */
 179 static int in_range(const struct nf_nat_l3proto *l3proto,
 180                     const struct nf_nat_l4proto *l4proto,
 181                     const struct nf_conntrack_tuple *tuple,
 182                     const struct nf_nat_range2 *range)
 183 {
 184         /* If we are supposed to map IPs, then we must be in the
 185          * range specified, otherwise let this drag us onto a new src IP.
 186          */
 187         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 188             !l3proto->in_range(tuple, range))
 189                 return 0;
 190
 191         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 192             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 193                               &range->min_proto, &range->max_proto))
 194                 return 1;
 195
 196         return 0;
 197 }
 198
 199 static inline int
 200 same_src(const struct nf_conn *ct,
 201          const struct nf_conntrack_tuple *tuple)
 202 {
 203         const struct nf_conntrack_tuple *t;
 204
 205         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 206         return (t->dst.protonum == tuple->dst.protonum &&
 207                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 208                 t->src.u.all == tuple->src.u.all);
 209 }
 210
 211 /* Only called for SRC manip */
 212 static int
 213 find_appropriate_src(struct net *net,
 214                      const struct nf_conntrack_zone *zone,
 215                      const struct nf_nat_l3proto *l3proto,
 216                      const struct nf_nat_l4proto *l4proto,
 217                      const struct nf_conntrack_tuple *tuple,
 218                      struct nf_conntrack_tuple *result,
 219                      const struct nf_nat_range2 *range)
 220 {
 221         unsigned int h = hash_by_src(net, tuple);
 222         const struct nf_conn *ct;
 223
 224         hlist_for_each_entry_rcu(ct, &nf_nat_bysource[h], nat_bysource) {
 225                 if (same_src(ct, tuple) &&
 226                     net_eq(net, nf_ct_net(ct)) &&
 227                     nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) {
 228                         /* Copy source part from reply tuple. */
 229                         nf_ct_invert_tuplepr(result,
 230                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 231                         result->dst = tuple->dst;
 232
 233                         if (in_range(l3proto, l4proto, result, range))
 234                                 return 1;
 235                 }
 236         }
 237         return 0;
 238 }
 239
 240 /* For [FUTURE] fragmentation handling, we want the least-used
 241  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 242  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 243  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 244  * the ip with the lowest src-ip/dst-ip/proto usage.
 245  */
 246 static void
 247 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 248                     struct nf_conntrack_tuple *tuple,
 249                     const struct nf_nat_range2 *range,
 250                     const struct nf_conn *ct,
 251                     enum nf_nat_manip_type maniptype)
 252 {
 253         union nf_inet_addr *var_ipp;
 254         unsigned int i, max;
 255         /* Host order */
 256         u32 minip, maxip, j, dist;
 257         bool full_range;
 258
 259         /* No IP mapping?  Do nothing. */
 260         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 261                 return;
 262
 263         if (maniptype == NF_NAT_MANIP_SRC)
 264                 var_ipp = &tuple->src.u3;
 265         else
 266                 var_ipp = &tuple->dst.u3;
 267
 268         /* Fast path: only one choice. */
 269         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 270                 *var_ipp = range->min_addr;
 271                 return;
 272         }
 273
 274         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 275                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 276         else
 277                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 278
 279         /* Hashing source and destination IPs gives a fairly even
 280          * spread in practice (if there are a small number of IPs
 281          * involved, there usually aren't that many connections
 282          * anyway).  The consistency means that servers see the same
 283          * client coming from the same IP (some Internet Banking sites
 284          * like this), even across reboots.
 285          */
 286         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 287                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 288                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 289
 290         full_range = false;
 291         for (i = 0; i <= max; i++) {
 292                 /* If first bytes of the address are at the maximum, use the
 293                  * distance. Otherwise use the full range.
 294                  */
 295                 if (!full_range) {
 296                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 297                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 298                         dist  = maxip - minip + 1;
 299                 } else {
 300                         minip = 0;
 301                         dist  = ~0;
 302                 }
 303
 304                 var_ipp->all[i] = (__force __u32)
 305                         htonl(minip + reciprocal_scale(j, dist));
 306                 if (var_ipp->all[i] != range->max_addr.all[i])
 307                         full_range = true;
 308
 309                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 310                         j ^= (__force u32)tuple->dst.u3.all[i];
 311         }
 312 }
 313
 314 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 315  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 316  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 317  * range. It might not be possible to get a unique tuple, but we try.
 318  * At worst (or if we race), we will end up with a final duplicate in
 319  * __ip_conntrack_confirm and drop the packet. */
 320 static void
 321 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 322                  const struct nf_conntrack_tuple *orig_tuple,
 323                  const struct nf_nat_range2 *range,
 324                  struct nf_conn *ct,
 325                  enum nf_nat_manip_type maniptype)
 326 {
 327         const struct nf_conntrack_zone *zone;
 328         const struct nf_nat_l3proto *l3proto;
 329         const struct nf_nat_l4proto *l4proto;
 330         struct net *net = nf_ct_net(ct);
 331
 332         zone = nf_ct_zone(ct);
 333
 334         rcu_read_lock();
 335         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 336         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 337                                         orig_tuple->dst.protonum);
 338
 339         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 340          * and that same mapping gives a unique tuple within the given
 341          * range, use that.
 342          *
 343          * This is only required for source (ie. NAT/masq) mappings.
 344          * So far, we don't do local source mappings, so multiple
 345          * manips not an issue.
 346          */
 347         if (maniptype == NF_NAT_MANIP_SRC &&
 348             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 349                 /* try the original tuple first */
 350                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 351                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 352                                 *tuple = *orig_tuple;
 353                                 goto out;
 354                         }
 355                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 356                                                 orig_tuple, tuple, range)) {
 357                         pr_debug("get_unique_tuple: Found current src map\n");
 358                         if (!nf_nat_used_tuple(tuple, ct))
 359                                 goto out;
 360                 }
 361         }
 362
 363         /* 2) Select the least-used IP/proto combination in the given range */
 364         *tuple = *orig_tuple;
 365         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 366
 367         /* 3) The per-protocol part of the manip is made to map into
 368          * the range to make a unique tuple.
 369          */
 370
 371         /* Only bother mapping if it's not already in range and unique */
 372         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 373                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 374                         if (!(range->flags & NF_NAT_RANGE_PROTO_OFFSET) &&
 375                             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 chance: 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_range2 *range,
 408                   enum nf_nat_manip_type maniptype)
 409 {
 410         struct net *net = nf_ct_net(ct);
 411         struct nf_conntrack_tuple curr_tuple, new_tuple;
 412
 413         /* Can't setup nat info for confirmed ct. */
 414         if (nf_ct_is_confirmed(ct))
 415                 return NF_ACCEPT;
 416
 417         WARN_ON(maniptype != NF_NAT_MANIP_SRC &&
 418                 maniptype != NF_NAT_MANIP_DST);
 419
 420         if (WARN_ON(nf_nat_initialized(ct, maniptype)))
 421                 return NF_DROP;
 422
 423         /* What we've got will look like inverse of reply. Normally
 424          * this is what is in the conntrack, except for prior
 425          * manipulations (future optimization: if num_manips == 0,
 426          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 427          */
 428         nf_ct_invert_tuplepr(&curr_tuple,
 429                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 430
 431         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 432
 433         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 434                 struct nf_conntrack_tuple reply;
 435
 436                 /* Alter conntrack table so will recognize replies. */
 437                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 438                 nf_conntrack_alter_reply(ct, &reply);
 439
 440                 /* Non-atomic: we own this at the moment. */
 441                 if (maniptype == NF_NAT_MANIP_SRC)
 442                         ct->status |= IPS_SRC_NAT;
 443                 else
 444                         ct->status |= IPS_DST_NAT;
 445
 446                 if (nfct_help(ct) && !nfct_seqadj(ct))
 447                         if (!nfct_seqadj_ext_add(ct))
 448                                 return NF_DROP;
 449         }
 450
 451         if (maniptype == NF_NAT_MANIP_SRC) {
 452                 unsigned int srchash;
 453                 spinlock_t *lock;
 454
 455                 srchash = hash_by_src(net,
 456                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 457                 lock = &nf_nat_locks[srchash % CONNTRACK_LOCKS];
 458                 spin_lock_bh(lock);
 459                 hlist_add_head_rcu(&ct->nat_bysource,
 460                                    &nf_nat_bysource[srchash]);
 461                 spin_unlock_bh(lock);
 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_range2 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 static unsigned int nf_nat_manip_pkt(struct sk_buff *skb, struct nf_conn *ct,
 501                                      enum nf_nat_manip_type mtype,
 502                                      enum ip_conntrack_dir dir)
 503 {
 504         const struct nf_nat_l3proto *l3proto;
 505         const struct nf_nat_l4proto *l4proto;
 506         struct nf_conntrack_tuple target;
 507
 508         /* We are aiming to look like inverse of other direction. */
 509         nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 510
 511         l3proto = __nf_nat_l3proto_find(target.src.l3num);
 512         l4proto = __nf_nat_l4proto_find(target.src.l3num,
 513                                         target.dst.protonum);
 514         if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 515                 return NF_DROP;
 516
 517         return NF_ACCEPT;
 518 }
 519
 520 /* Do packet manipulations according to nf_nat_setup_info. */
 521 unsigned int nf_nat_packet(struct nf_conn *ct,
 522                            enum ip_conntrack_info ctinfo,
 523                            unsigned int hooknum,
 524                            struct sk_buff *skb)
 525 {
 526         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 527         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 528         unsigned int verdict = NF_ACCEPT;
 529         unsigned long statusbit;
 530
 531         if (mtype == NF_NAT_MANIP_SRC)
 532                 statusbit = IPS_SRC_NAT;
 533         else
 534                 statusbit = IPS_DST_NAT;
 535
 536         /* Invert if this is reply dir. */
 537         if (dir == IP_CT_DIR_REPLY)
 538                 statusbit ^= IPS_NAT_MASK;
 539
 540         /* Non-atomic: these bits don't change. */
 541         if (ct->status & statusbit)
 542                 verdict = nf_nat_manip_pkt(skb, ct, mtype, dir);
 543
 544         return verdict;
 545 }
 546 EXPORT_SYMBOL_GPL(nf_nat_packet);
 547
 548 unsigned int
 549 nf_nat_inet_fn(void *priv, struct sk_buff *skb,
 550                const struct nf_hook_state *state)
 551 {
 552         struct nf_conn *ct;
 553         enum ip_conntrack_info ctinfo;
 554         struct nf_conn_nat *nat;
 555         /* maniptype == SRC for postrouting. */
 556         enum nf_nat_manip_type maniptype = HOOK2MANIP(state->hook);
 557
 558         ct = nf_ct_get(skb, &ctinfo);
 559         /* Can't track?  It's not due to stress, or conntrack would
 560          * have dropped it.  Hence it's the user's responsibilty to
 561          * packet filter it out, or implement conntrack/NAT for that
 562          * protocol. 8) --RR
 563          */
 564         if (!ct)
 565                 return NF_ACCEPT;
 566
 567         nat = nfct_nat(ct);
 568
 569         switch (ctinfo) {
 570         case IP_CT_RELATED:
 571         case IP_CT_RELATED_REPLY:
 572                 /* Only ICMPs can be IP_CT_IS_REPLY.  Fallthrough */
 573         case IP_CT_NEW:
 574                 /* Seen it before?  This can happen for loopback, retrans,
 575                  * or local packets.
 576                  */
 577                 if (!nf_nat_initialized(ct, maniptype)) {
 578                         struct nf_nat_lookup_hook_priv *lpriv = priv;
 579                         struct nf_hook_entries *e = rcu_dereference(lpriv->entries);
 580                         unsigned int ret;
 581                         int i;
 582
 583                         if (!e)
 584                                 goto null_bind;
 585
 586                         for (i = 0; i < e->num_hook_entries; i++) {
 587                                 ret = e->hooks[i].hook(e->hooks[i].priv, skb,
 588                                                        state);
 589                                 if (ret != NF_ACCEPT)
 590                                         return ret;
 591                                 if (nf_nat_initialized(ct, maniptype))
 592                                         goto do_nat;
 593                         }
 594 null_bind:
 595                         ret = nf_nat_alloc_null_binding(ct, state->hook);
 596                         if (ret != NF_ACCEPT)
 597                                 return ret;
 598                 } else {
 599                         pr_debug("Already setup manip %s for ct %p (status bits 0x%lx)\n",
 600                                  maniptype == NF_NAT_MANIP_SRC ? "SRC" : "DST",
 601                                  ct, ct->status);
 602                         if (nf_nat_oif_changed(state->hook, ctinfo, nat,
 603                                                state->out))
 604                                 goto oif_changed;
 605                 }
 606                 break;
 607         default:
 608                 /* ESTABLISHED */
 609                 WARN_ON(ctinfo != IP_CT_ESTABLISHED &&
 610                         ctinfo != IP_CT_ESTABLISHED_REPLY);
 611                 if (nf_nat_oif_changed(state->hook, ctinfo, nat, state->out))
 612                         goto oif_changed;
 613         }
 614 do_nat:
 615         return nf_nat_packet(ct, ctinfo, state->hook, skb);
 616
 617 oif_changed:
 618         nf_ct_kill_acct(ct, ctinfo, skb);
 619         return NF_DROP;
 620 }
 621 EXPORT_SYMBOL_GPL(nf_nat_inet_fn);
 622
 623 struct nf_nat_proto_clean {
 624         u8      l3proto;
 625         u8      l4proto;
 626 };
 627
 628 /* kill conntracks with affected NAT section */
 629 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 630 {
 631         const struct nf_nat_proto_clean *clean = data;
 632
 633         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 634             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 635                 return 0;
 636
 637         return i->status & IPS_NAT_MASK ? 1 : 0;
 638 }
 639
 640 static void __nf_nat_cleanup_conntrack(struct nf_conn *ct)
 641 {
 642         unsigned int h;
 643
 644         h = hash_by_src(nf_ct_net(ct), &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 645         spin_lock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 646         hlist_del_rcu(&ct->nat_bysource);
 647         spin_unlock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 648 }
 649
 650 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 651 {
 652         if (nf_nat_proto_remove(ct, data))
 653                 return 1;
 654
 655         /* This module is being removed and conntrack has nat null binding.
 656          * Remove it from bysource hash, as the table will be freed soon.
 657          *
 658          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 659          * will delete entry from already-freed table.
 660          */
 661         if (test_and_clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status))
 662                 __nf_nat_cleanup_conntrack(ct);
 663
 664         /* don't delete conntrack.  Although that would make things a lot
 665          * simpler, we'd end up flushing all conntracks on nat rmmod.
 666          */
 667         return 0;
 668 }
 669
 670 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 671 {
 672         struct nf_nat_proto_clean clean = {
 673                 .l3proto = l3proto,
 674                 .l4proto = l4proto,
 675         };
 676
 677         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 678 }
 679
 680 static void nf_nat_l3proto_clean(u8 l3proto)
 681 {
 682         struct nf_nat_proto_clean clean = {
 683                 .l3proto = l3proto,
 684         };
 685
 686         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 687 }
 688
 689 /* Protocol registration. */
 690 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 691 {
 692         const struct nf_nat_l4proto **l4protos;
 693         unsigned int i;
 694         int ret = 0;
 695
 696         mutex_lock(&nf_nat_proto_mutex);
 697         if (nf_nat_l4protos[l3proto] == NULL) {
 698                 l4protos = kmalloc_array(IPPROTO_MAX,
 699                                          sizeof(struct nf_nat_l4proto *),
 700                                          GFP_KERNEL);
 701                 if (l4protos == NULL) {
 702                         ret = -ENOMEM;
 703                         goto out;
 704                 }
 705
 706                 for (i = 0; i < IPPROTO_MAX; i++)
 707                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 708
 709                 /* Before making proto_array visible to lockless readers,
 710                  * we must make sure its content is committed to memory.
 711                  */
 712                 smp_wmb();
 713
 714                 nf_nat_l4protos[l3proto] = l4protos;
 715         }
 716
 717         if (rcu_dereference_protected(
 718                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 719                         lockdep_is_held(&nf_nat_proto_mutex)
 720                         ) != &nf_nat_l4proto_unknown) {
 721                 ret = -EBUSY;
 722                 goto out;
 723         }
 724         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 725  out:
 726         mutex_unlock(&nf_nat_proto_mutex);
 727         return ret;
 728 }
 729 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 730
 731 /* No one stores the protocol anywhere; simply delete it. */
 732 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 733 {
 734         mutex_lock(&nf_nat_proto_mutex);
 735         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 736                          &nf_nat_l4proto_unknown);
 737         mutex_unlock(&nf_nat_proto_mutex);
 738         synchronize_rcu();
 739
 740         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 741 }
 742 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 743
 744 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 745 {
 746         mutex_lock(&nf_nat_proto_mutex);
 747         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 748                          &nf_nat_l4proto_tcp);
 749         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 750                          &nf_nat_l4proto_udp);
 751 #ifdef CONFIG_NF_NAT_PROTO_DCCP
 752         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_DCCP],
 753                          &nf_nat_l4proto_dccp);
 754 #endif
 755 #ifdef CONFIG_NF_NAT_PROTO_SCTP
 756         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_SCTP],
 757                          &nf_nat_l4proto_sctp);
 758 #endif
 759 #ifdef CONFIG_NF_NAT_PROTO_UDPLITE
 760         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDPLITE],
 761                          &nf_nat_l4proto_udplite);
 762 #endif
 763         mutex_unlock(&nf_nat_proto_mutex);
 764
 765         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 766         return 0;
 767 }
 768 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 769
 770 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 771 {
 772         mutex_lock(&nf_nat_proto_mutex);
 773         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 774         mutex_unlock(&nf_nat_proto_mutex);
 775         synchronize_rcu();
 776
 777         nf_nat_l3proto_clean(l3proto->l3proto);
 778 }
 779 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 780
 781 /* No one using conntrack by the time this called. */
 782 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 783 {
 784         if (ct->status & IPS_SRC_NAT_DONE)
 785                 __nf_nat_cleanup_conntrack(ct);
 786 }
 787
 788 static struct nf_ct_ext_type nat_extend __read_mostly = {
 789         .len            = sizeof(struct nf_conn_nat),
 790         .align          = __alignof__(struct nf_conn_nat),
 791         .destroy        = nf_nat_cleanup_conntrack,
 792         .id             = NF_CT_EXT_NAT,
 793 };
 794
 795 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 796
 797 #include <linux/netfilter/nfnetlink.h>
 798 #include <linux/netfilter/nfnetlink_conntrack.h>
 799
 800 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 801         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 802         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 803 };
 804
 805 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 806                                      const struct nf_conn *ct,
 807                                      struct nf_nat_range2 *range)
 808 {
 809         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 810         const struct nf_nat_l4proto *l4proto;
 811         int err;
 812
 813         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr,
 814                                protonat_nla_policy, NULL);
 815         if (err < 0)
 816                 return err;
 817
 818         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 819         if (l4proto->nlattr_to_range)
 820                 err = l4proto->nlattr_to_range(tb, range);
 821
 822         return err;
 823 }
 824
 825 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 826         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 827         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 828         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 829         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 830         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 831 };
 832
 833 static int
 834 nfnetlink_parse_nat(const struct nlattr *nat,
 835                     const struct nf_conn *ct, struct nf_nat_range2 *range,
 836                     const struct nf_nat_l3proto *l3proto)
 837 {
 838         struct nlattr *tb[CTA_NAT_MAX+1];
 839         int err;
 840
 841         memset(range, 0, sizeof(*range));
 842
 843         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy, NULL);
 844         if (err < 0)
 845                 return err;
 846
 847         err = l3proto->nlattr_to_range(tb, range);
 848         if (err < 0)
 849                 return err;
 850
 851         if (!tb[CTA_NAT_PROTO])
 852                 return 0;
 853
 854         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 855 }
 856
 857 /* This function is called under rcu_read_lock() */
 858 static int
 859 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 860                           enum nf_nat_manip_type manip,
 861                           const struct nlattr *attr)
 862 {
 863         struct nf_nat_range2 range;
 864         const struct nf_nat_l3proto *l3proto;
 865         int err;
 866
 867         /* Should not happen, restricted to creating new conntracks
 868          * via ctnetlink.
 869          */
 870         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 871                 return -EEXIST;
 872
 873         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 874          * attach the null binding, otherwise this may oops.
 875          */
 876         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 877         if (l3proto == NULL)
 878                 return -EAGAIN;
 879
 880         /* No NAT information has been passed, allocate the null-binding */
 881         if (attr == NULL)
 882                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
 883
 884         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 885         if (err < 0)
 886                 return err;
 887
 888         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 889 }
 890 #else
 891 static int
 892 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 893                           enum nf_nat_manip_type manip,
 894                           const struct nlattr *attr)
 895 {
 896         return -EOPNOTSUPP;
 897 }
 898 #endif
 899
 900 static struct nf_ct_helper_expectfn follow_master_nat = {
 901         .name           = "nat-follow-master",
 902         .expectfn       = nf_nat_follow_master,
 903 };
 904
 905 int nf_nat_register_fn(struct net *net, const struct nf_hook_ops *ops,
 906                        const struct nf_hook_ops *orig_nat_ops, unsigned int ops_count)
 907 {
 908         struct nat_net *nat_net = net_generic(net, nat_net_id);
 909         struct nf_nat_hooks_net *nat_proto_net;
 910         struct nf_nat_lookup_hook_priv *priv;
 911         unsigned int hooknum = ops->hooknum;
 912         struct nf_hook_ops *nat_ops;
 913         int i, ret;
 914
 915         if (WARN_ON_ONCE(ops->pf >= ARRAY_SIZE(nat_net->nat_proto_net)))
 916                 return -EINVAL;
 917
 918         nat_proto_net = &nat_net->nat_proto_net[ops->pf];
 919
 920         for (i = 0; i < ops_count; i++) {
 921                 if (WARN_ON(orig_nat_ops[i].pf != ops->pf))
 922                         return -EINVAL;
 923                 if (orig_nat_ops[i].hooknum == hooknum) {
 924                         hooknum = i;
 925                         break;
 926                 }
 927         }
 928
 929         if (WARN_ON_ONCE(i == ops_count))
 930                 return -EINVAL;
 931
 932         mutex_lock(&nf_nat_proto_mutex);
 933         if (!nat_proto_net->nat_hook_ops) {
 934                 WARN_ON(nat_proto_net->users != 0);
 935
 936                 nat_ops = kmemdup(orig_nat_ops, sizeof(*orig_nat_ops) * ops_count, GFP_KERNEL);
 937                 if (!nat_ops) {
 938                         mutex_unlock(&nf_nat_proto_mutex);
 939                         return -ENOMEM;
 940                 }
 941
 942                 for (i = 0; i < ops_count; i++) {
 943                         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 944                         if (priv) {
 945                                 nat_ops[i].priv = priv;
 946                                 continue;
 947                         }
 948                         mutex_unlock(&nf_nat_proto_mutex);
 949                         while (i)
 950                                 kfree(nat_ops[--i].priv);
 951                         kfree(nat_ops);
 952                         return -ENOMEM;
 953                 }
 954
 955                 ret = nf_register_net_hooks(net, nat_ops, ops_count);
 956                 if (ret < 0) {
 957                         mutex_unlock(&nf_nat_proto_mutex);
 958                         for (i = 0; i < ops_count; i++)
 959                                 kfree(nat_ops[i].priv);
 960                         kfree(nat_ops);
 961                         return ret;
 962                 }
 963
 964                 nat_proto_net->nat_hook_ops = nat_ops;
 965         }
 966
 967         nat_ops = nat_proto_net->nat_hook_ops;
 968         priv = nat_ops[hooknum].priv;
 969         if (WARN_ON_ONCE(!priv)) {
 970                 mutex_unlock(&nf_nat_proto_mutex);
 971                 return -EOPNOTSUPP;
 972         }
 973
 974         ret = nf_hook_entries_insert_raw(&priv->entries, ops);
 975         if (ret == 0)
 976                 nat_proto_net->users++;
 977
 978         mutex_unlock(&nf_nat_proto_mutex);
 979         return ret;
 980 }
 981 EXPORT_SYMBOL_GPL(nf_nat_register_fn);
 982
 983 void nf_nat_unregister_fn(struct net *net, const struct nf_hook_ops *ops,
 984                           unsigned int ops_count)
 985 {
 986         struct nat_net *nat_net = net_generic(net, nat_net_id);
 987         struct nf_nat_hooks_net *nat_proto_net;
 988         struct nf_nat_lookup_hook_priv *priv;
 989         struct nf_hook_ops *nat_ops;
 990         int hooknum = ops->hooknum;
 991         int i;
 992
 993         if (ops->pf >= ARRAY_SIZE(nat_net->nat_proto_net))
 994                 return;
 995
 996         nat_proto_net = &nat_net->nat_proto_net[ops->pf];
 997
 998         mutex_lock(&nf_nat_proto_mutex);
 999         if (WARN_ON(nat_proto_net->users == 0))
1000                 goto unlock;
1001
1002         nat_proto_net->users--;
1003
1004         nat_ops = nat_proto_net->nat_hook_ops;
1005         for (i = 0; i < ops_count; i++) {
1006                 if (nat_ops[i].hooknum == hooknum) {
1007                         hooknum = i;
1008                         break;
1009                 }
1010         }
1011         if (WARN_ON_ONCE(i == ops_count))
1012                 goto unlock;
1013         priv = nat_ops[hooknum].priv;
1014         nf_hook_entries_delete_raw(&priv->entries, ops);
1015
1016         if (nat_proto_net->users == 0) {
1017                 nf_unregister_net_hooks(net, nat_ops, ops_count);
1018
1019                 for (i = 0; i < ops_count; i++) {
1020                         priv = nat_ops[i].priv;
1021                         kfree_rcu(priv, rcu_head);
1022                 }
1023
1024                 nat_proto_net->nat_hook_ops = NULL;
1025                 kfree(nat_ops);
1026         }
1027 unlock:
1028         mutex_unlock(&nf_nat_proto_mutex);
1029 }
1030 EXPORT_SYMBOL_GPL(nf_nat_unregister_fn);
1031
1032 static struct pernet_operations nat_net_ops = {
1033         .id = &nat_net_id,
1034         .size = sizeof(struct nat_net),
1035 };
1036
1037 static struct nf_nat_hook nat_hook = {
1038         .parse_nat_setup        = nfnetlink_parse_nat_setup,
1039 #ifdef CONFIG_XFRM
1040         .decode_session         = __nf_nat_decode_session,
1041 #endif
1042         .manip_pkt              = nf_nat_manip_pkt,
1043 };
1044
1045 static int __init nf_nat_init(void)
1046 {
1047         int ret, i;
1048
1049         /* Leave them the same for the moment. */
1050         nf_nat_htable_size = nf_conntrack_htable_size;
1051         if (nf_nat_htable_size < CONNTRACK_LOCKS)
1052                 nf_nat_htable_size = CONNTRACK_LOCKS;
1053
1054         nf_nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, 0);
1055         if (!nf_nat_bysource)
1056                 return -ENOMEM;
1057
1058         ret = nf_ct_extend_register(&nat_extend);
1059         if (ret < 0) {
1060                 kvfree(nf_nat_bysource);
1061                 pr_err("Unable to register extension\n");
1062                 return ret;
1063         }
1064
1065         for (i = 0; i < CONNTRACK_LOCKS; i++)
1066                 spin_lock_init(&nf_nat_locks[i]);
1067
1068         ret = register_pernet_subsys(&nat_net_ops);
1069         if (ret < 0) {
1070                 nf_ct_extend_unregister(&nat_extend);
1071                 return ret;
1072         }
1073
1074         nf_ct_helper_expectfn_register(&follow_master_nat);
1075
1076         WARN_ON(nf_nat_hook != NULL);
1077         RCU_INIT_POINTER(nf_nat_hook, &nat_hook);
1078
1079         return 0;
1080 }
1081
1082 static void __exit nf_nat_cleanup(void)
1083 {
1084         struct nf_nat_proto_clean clean = {};
1085         unsigned int i;
1086
1087         nf_ct_iterate_destroy(nf_nat_proto_clean, &clean);
1088
1089         nf_ct_extend_unregister(&nat_extend);
1090         nf_ct_helper_expectfn_unregister(&follow_master_nat);
1091         RCU_INIT_POINTER(nf_nat_hook, NULL);
1092
1093         synchronize_rcu();
1094
1095         for (i = 0; i < NFPROTO_NUMPROTO; i++)
1096                 kfree(nf_nat_l4protos[i]);
1097         synchronize_net();
1098         kvfree(nf_nat_bysource);
1099         unregister_pernet_subsys(&nat_net_ops);
1100 }
1101
1102 MODULE_LICENSE("GPL");
1103
1104 module_init(nf_nat_init);
1105 module_exit(nf_nat_cleanup);