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_l3proto.h>
  32 #include <net/netfilter/nf_conntrack_zones.h>
  33 #include <linux/netfilter/nf_nat.h>
  34
  35 #include "nf_internals.h"
  36
  37 static spinlock_t nf_nat_locks[CONNTRACK_LOCKS];
  38
  39 static DEFINE_MUTEX(nf_nat_proto_mutex);
  40 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
  41                                                 __read_mostly;
  42 static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO]
  43                                                 __read_mostly;
  44 static unsigned int nat_net_id __read_mostly;
  45
  46 static struct hlist_head *nf_nat_bysource __read_mostly;
  47 static unsigned int nf_nat_htable_size __read_mostly;
  48 static unsigned int nf_nat_hash_rnd __read_mostly;
  49
  50 struct nf_nat_lookup_hook_priv {
  51         struct nf_hook_entries __rcu *entries;
  52
  53         struct rcu_head rcu_head;
  54 };
  55
  56 struct nf_nat_hooks_net {
  57         struct nf_hook_ops *nat_hook_ops;
  58         unsigned int users;
  59 };
  60
  61 struct nat_net {
  62         struct nf_nat_hooks_net nat_proto_net[NFPROTO_NUMPROTO];
  63 };
  64
  65 inline const struct nf_nat_l3proto *
  66 __nf_nat_l3proto_find(u8 family)
  67 {
  68         return rcu_dereference(nf_nat_l3protos[family]);
  69 }
  70
  71 inline const struct nf_nat_l4proto *
  72 __nf_nat_l4proto_find(u8 family, u8 protonum)
  73 {
  74         return rcu_dereference(nf_nat_l4protos[family][protonum]);
  75 }
  76 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
  77
  78 #ifdef CONFIG_XFRM
  79 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  80 {
  81         const struct nf_nat_l3proto *l3proto;
  82         const struct nf_conn *ct;
  83         enum ip_conntrack_info ctinfo;
  84         enum ip_conntrack_dir dir;
  85         unsigned  long statusbit;
  86         u8 family;
  87
  88         ct = nf_ct_get(skb, &ctinfo);
  89         if (ct == NULL)
  90                 return;
  91
  92         family = nf_ct_l3num(ct);
  93         l3proto = __nf_nat_l3proto_find(family);
  94         if (l3proto == NULL)
  95                 return;
  96
  97         dir = CTINFO2DIR(ctinfo);
  98         if (dir == IP_CT_DIR_ORIGINAL)
  99                 statusbit = IPS_DST_NAT;
 100         else
 101                 statusbit = IPS_SRC_NAT;
 102
 103         l3proto->decode_session(skb, ct, dir, statusbit, fl);
 104 }
 105
 106 int nf_xfrm_me_harder(struct net *net, struct sk_buff *skb, unsigned int family)
 107 {
 108         struct flowi fl;
 109         unsigned int hh_len;
 110         struct dst_entry *dst;
 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         dst = xfrm_lookup(net, dst, &fl, skb->sk, 0);
 123         if (IS_ERR(dst))
 124                 return PTR_ERR(dst);
 125
 126         skb_dst_drop(skb);
 127         skb_dst_set(skb, dst);
 128
 129         /* Change in oif may mean change in hh_len. */
 130         hh_len = skb_dst(skb)->dev->hard_header_len;
 131         if (skb_headroom(skb) < hh_len &&
 132             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 133                 return -ENOMEM;
 134         return 0;
 135 }
 136 EXPORT_SYMBOL(nf_xfrm_me_harder);
 137 #endif /* CONFIG_XFRM */
 138
 139 /* We keep an extra hash for each conntrack, for fast searching. */
 140 static unsigned int
 141 hash_by_src(const struct net *n, const struct nf_conntrack_tuple *tuple)
 142 {
 143         unsigned int hash;
 144
 145         get_random_once(&nf_nat_hash_rnd, sizeof(nf_nat_hash_rnd));
 146
 147         /* Original src, to ensure we map it consistently if poss. */
 148         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 149                       tuple->dst.protonum ^ nf_nat_hash_rnd ^ net_hash_mix(n));
 150
 151         return reciprocal_scale(hash, nf_nat_htable_size);
 152 }
 153
 154 /* Is this tuple already taken? (not by us) */
 155 int
 156 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 157                   const struct nf_conn *ignored_conntrack)
 158 {
 159         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 160          * incoming ones.  NAT means they don't have a fixed mapping,
 161          * so we invert the tuple and look for the incoming reply.
 162          *
 163          * We could keep a separate hash if this proves too slow.
 164          */
 165         struct nf_conntrack_tuple reply;
 166
 167         nf_ct_invert_tuplepr(&reply, tuple);
 168         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 169 }
 170 EXPORT_SYMBOL(nf_nat_used_tuple);
 171
 172 /* If we source map this tuple so reply looks like reply_tuple, will
 173  * that meet the constraints of range.
 174  */
 175 static int in_range(const struct nf_nat_l3proto *l3proto,
 176                     const struct nf_nat_l4proto *l4proto,
 177                     const struct nf_conntrack_tuple *tuple,
 178                     const struct nf_nat_range2 *range)
 179 {
 180         /* If we are supposed to map IPs, then we must be in the
 181          * range specified, otherwise let this drag us onto a new src IP.
 182          */
 183         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 184             !l3proto->in_range(tuple, range))
 185                 return 0;
 186
 187         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 188             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 189                               &range->min_proto, &range->max_proto))
 190                 return 1;
 191
 192         return 0;
 193 }
 194
 195 static inline int
 196 same_src(const struct nf_conn *ct,
 197          const struct nf_conntrack_tuple *tuple)
 198 {
 199         const struct nf_conntrack_tuple *t;
 200
 201         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 202         return (t->dst.protonum == tuple->dst.protonum &&
 203                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 204                 t->src.u.all == tuple->src.u.all);
 205 }
 206
 207 /* Only called for SRC manip */
 208 static int
 209 find_appropriate_src(struct net *net,
 210                      const struct nf_conntrack_zone *zone,
 211                      const struct nf_nat_l3proto *l3proto,
 212                      const struct nf_nat_l4proto *l4proto,
 213                      const struct nf_conntrack_tuple *tuple,
 214                      struct nf_conntrack_tuple *result,
 215                      const struct nf_nat_range2 *range)
 216 {
 217         unsigned int h = hash_by_src(net, tuple);
 218         const struct nf_conn *ct;
 219
 220         hlist_for_each_entry_rcu(ct, &nf_nat_bysource[h], nat_bysource) {
 221                 if (same_src(ct, tuple) &&
 222                     net_eq(net, nf_ct_net(ct)) &&
 223                     nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) {
 224                         /* Copy source part from reply tuple. */
 225                         nf_ct_invert_tuplepr(result,
 226                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 227                         result->dst = tuple->dst;
 228
 229                         if (in_range(l3proto, l4proto, result, range))
 230                                 return 1;
 231                 }
 232         }
 233         return 0;
 234 }
 235
 236 /* For [FUTURE] fragmentation handling, we want the least-used
 237  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 238  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 239  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 240  * the ip with the lowest src-ip/dst-ip/proto usage.
 241  */
 242 static void
 243 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 244                     struct nf_conntrack_tuple *tuple,
 245                     const struct nf_nat_range2 *range,
 246                     const struct nf_conn *ct,
 247                     enum nf_nat_manip_type maniptype)
 248 {
 249         union nf_inet_addr *var_ipp;
 250         unsigned int i, max;
 251         /* Host order */
 252         u32 minip, maxip, j, dist;
 253         bool full_range;
 254
 255         /* No IP mapping?  Do nothing. */
 256         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 257                 return;
 258
 259         if (maniptype == NF_NAT_MANIP_SRC)
 260                 var_ipp = &tuple->src.u3;
 261         else
 262                 var_ipp = &tuple->dst.u3;
 263
 264         /* Fast path: only one choice. */
 265         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 266                 *var_ipp = range->min_addr;
 267                 return;
 268         }
 269
 270         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 271                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 272         else
 273                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 274
 275         /* Hashing source and destination IPs gives a fairly even
 276          * spread in practice (if there are a small number of IPs
 277          * involved, there usually aren't that many connections
 278          * anyway).  The consistency means that servers see the same
 279          * client coming from the same IP (some Internet Banking sites
 280          * like this), even across reboots.
 281          */
 282         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 283                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 284                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 285
 286         full_range = false;
 287         for (i = 0; i <= max; i++) {
 288                 /* If first bytes of the address are at the maximum, use the
 289                  * distance. Otherwise use the full range.
 290                  */
 291                 if (!full_range) {
 292                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 293                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 294                         dist  = maxip - minip + 1;
 295                 } else {
 296                         minip = 0;
 297                         dist  = ~0;
 298                 }
 299
 300                 var_ipp->all[i] = (__force __u32)
 301                         htonl(minip + reciprocal_scale(j, dist));
 302                 if (var_ipp->all[i] != range->max_addr.all[i])
 303                         full_range = true;
 304
 305                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 306                         j ^= (__force u32)tuple->dst.u3.all[i];
 307         }
 308 }
 309
 310 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 311  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 312  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 313  * range. It might not be possible to get a unique tuple, but we try.
 314  * At worst (or if we race), we will end up with a final duplicate in
 315  * __ip_conntrack_confirm and drop the packet. */
 316 static void
 317 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 318                  const struct nf_conntrack_tuple *orig_tuple,
 319                  const struct nf_nat_range2 *range,
 320                  struct nf_conn *ct,
 321                  enum nf_nat_manip_type maniptype)
 322 {
 323         const struct nf_conntrack_zone *zone;
 324         const struct nf_nat_l3proto *l3proto;
 325         const struct nf_nat_l4proto *l4proto;
 326         struct net *net = nf_ct_net(ct);
 327
 328         zone = nf_ct_zone(ct);
 329
 330         rcu_read_lock();
 331         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 332         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 333                                         orig_tuple->dst.protonum);
 334
 335         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 336          * and that same mapping gives a unique tuple within the given
 337          * range, use that.
 338          *
 339          * This is only required for source (ie. NAT/masq) mappings.
 340          * So far, we don't do local source mappings, so multiple
 341          * manips not an issue.
 342          */
 343         if (maniptype == NF_NAT_MANIP_SRC &&
 344             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 345                 /* try the original tuple first */
 346                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 347                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 348                                 *tuple = *orig_tuple;
 349                                 goto out;
 350                         }
 351                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 352                                                 orig_tuple, tuple, range)) {
 353                         pr_debug("get_unique_tuple: Found current src map\n");
 354                         if (!nf_nat_used_tuple(tuple, ct))
 355                                 goto out;
 356                 }
 357         }
 358
 359         /* 2) Select the least-used IP/proto combination in the given range */
 360         *tuple = *orig_tuple;
 361         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 362
 363         /* 3) The per-protocol part of the manip is made to map into
 364          * the range to make a unique tuple.
 365          */
 366
 367         /* Only bother mapping if it's not already in range and unique */
 368         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 369                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 370                         if (!(range->flags & NF_NAT_RANGE_PROTO_OFFSET) &&
 371                             l4proto->in_range(tuple, maniptype,
 372                                   &range->min_proto,
 373                                   &range->max_proto) &&
 374                             (range->min_proto.all == range->max_proto.all ||
 375                              !nf_nat_used_tuple(tuple, ct)))
 376                                 goto out;
 377                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 378                         goto out;
 379                 }
 380         }
 381
 382         /* Last chance: get protocol to try to obtain unique tuple. */
 383         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 384 out:
 385         rcu_read_unlock();
 386 }
 387
 388 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 389 {
 390         struct nf_conn_nat *nat = nfct_nat(ct);
 391         if (nat)
 392                 return nat;
 393
 394         if (!nf_ct_is_confirmed(ct))
 395                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 396
 397         return nat;
 398 }
 399 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 400
 401 unsigned int
 402 nf_nat_setup_info(struct nf_conn *ct,
 403                   const struct nf_nat_range2 *range,
 404                   enum nf_nat_manip_type maniptype)
 405 {
 406         struct net *net = nf_ct_net(ct);
 407         struct nf_conntrack_tuple curr_tuple, new_tuple;
 408
 409         /* Can't setup nat info for confirmed ct. */
 410         if (nf_ct_is_confirmed(ct))
 411                 return NF_ACCEPT;
 412
 413         WARN_ON(maniptype != NF_NAT_MANIP_SRC &&
 414                 maniptype != NF_NAT_MANIP_DST);
 415
 416         if (WARN_ON(nf_nat_initialized(ct, maniptype)))
 417                 return NF_DROP;
 418
 419         /* What we've got will look like inverse of reply. Normally
 420          * this is what is in the conntrack, except for prior
 421          * manipulations (future optimization: if num_manips == 0,
 422          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 423          */
 424         nf_ct_invert_tuplepr(&curr_tuple,
 425                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 426
 427         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 428
 429         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 430                 struct nf_conntrack_tuple reply;
 431
 432                 /* Alter conntrack table so will recognize replies. */
 433                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 434                 nf_conntrack_alter_reply(ct, &reply);
 435
 436                 /* Non-atomic: we own this at the moment. */
 437                 if (maniptype == NF_NAT_MANIP_SRC)
 438                         ct->status |= IPS_SRC_NAT;
 439                 else
 440                         ct->status |= IPS_DST_NAT;
 441
 442                 if (nfct_help(ct) && !nfct_seqadj(ct))
 443                         if (!nfct_seqadj_ext_add(ct))
 444                                 return NF_DROP;
 445         }
 446
 447         if (maniptype == NF_NAT_MANIP_SRC) {
 448                 unsigned int srchash;
 449                 spinlock_t *lock;
 450
 451                 srchash = hash_by_src(net,
 452                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 453                 lock = &nf_nat_locks[srchash % CONNTRACK_LOCKS];
 454                 spin_lock_bh(lock);
 455                 hlist_add_head_rcu(&ct->nat_bysource,
 456                                    &nf_nat_bysource[srchash]);
 457                 spin_unlock_bh(lock);
 458         }
 459
 460         /* It's done. */
 461         if (maniptype == NF_NAT_MANIP_DST)
 462                 ct->status |= IPS_DST_NAT_DONE;
 463         else
 464                 ct->status |= IPS_SRC_NAT_DONE;
 465
 466         return NF_ACCEPT;
 467 }
 468 EXPORT_SYMBOL(nf_nat_setup_info);
 469
 470 static unsigned int
 471 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 472 {
 473         /* Force range to this IP; let proto decide mapping for
 474          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 475          * Use reply in case it's already been mangled (eg local packet).
 476          */
 477         union nf_inet_addr ip =
 478                 (manip == NF_NAT_MANIP_SRC ?
 479                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 480                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 481         struct nf_nat_range2 range = {
 482                 .flags          = NF_NAT_RANGE_MAP_IPS,
 483                 .min_addr       = ip,
 484                 .max_addr       = ip,
 485         };
 486         return nf_nat_setup_info(ct, &range, manip);
 487 }
 488
 489 unsigned int
 490 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 491 {
 492         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 493 }
 494 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 495
 496 static unsigned int nf_nat_manip_pkt(struct sk_buff *skb, struct nf_conn *ct,
 497                                      enum nf_nat_manip_type mtype,
 498                                      enum ip_conntrack_dir dir)
 499 {
 500         const struct nf_nat_l3proto *l3proto;
 501         const struct nf_nat_l4proto *l4proto;
 502         struct nf_conntrack_tuple target;
 503
 504         /* We are aiming to look like inverse of other direction. */
 505         nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 506
 507         l3proto = __nf_nat_l3proto_find(target.src.l3num);
 508         l4proto = __nf_nat_l4proto_find(target.src.l3num,
 509                                         target.dst.protonum);
 510         if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 511                 return NF_DROP;
 512
 513         return NF_ACCEPT;
 514 }
 515
 516 /* Do packet manipulations according to nf_nat_setup_info. */
 517 unsigned int nf_nat_packet(struct nf_conn *ct,
 518                            enum ip_conntrack_info ctinfo,
 519                            unsigned int hooknum,
 520                            struct sk_buff *skb)
 521 {
 522         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 523         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 524         unsigned int verdict = NF_ACCEPT;
 525         unsigned long statusbit;
 526
 527         if (mtype == NF_NAT_MANIP_SRC)
 528                 statusbit = IPS_SRC_NAT;
 529         else
 530                 statusbit = IPS_DST_NAT;
 531
 532         /* Invert if this is reply dir. */
 533         if (dir == IP_CT_DIR_REPLY)
 534                 statusbit ^= IPS_NAT_MASK;
 535
 536         /* Non-atomic: these bits don't change. */
 537         if (ct->status & statusbit)
 538                 verdict = nf_nat_manip_pkt(skb, ct, mtype, dir);
 539
 540         return verdict;
 541 }
 542 EXPORT_SYMBOL_GPL(nf_nat_packet);
 543
 544 unsigned int
 545 nf_nat_inet_fn(void *priv, struct sk_buff *skb,
 546                const struct nf_hook_state *state)
 547 {
 548         struct nf_conn *ct;
 549         enum ip_conntrack_info ctinfo;
 550         struct nf_conn_nat *nat;
 551         /* maniptype == SRC for postrouting. */
 552         enum nf_nat_manip_type maniptype = HOOK2MANIP(state->hook);
 553
 554         ct = nf_ct_get(skb, &ctinfo);
 555         /* Can't track?  It's not due to stress, or conntrack would
 556          * have dropped it.  Hence it's the user's responsibilty to
 557          * packet filter it out, or implement conntrack/NAT for that
 558          * protocol. 8) --RR
 559          */
 560         if (!ct)
 561                 return NF_ACCEPT;
 562
 563         nat = nfct_nat(ct);
 564
 565         switch (ctinfo) {
 566         case IP_CT_RELATED:
 567         case IP_CT_RELATED_REPLY:
 568                 /* Only ICMPs can be IP_CT_IS_REPLY.  Fallthrough */
 569         case IP_CT_NEW:
 570                 /* Seen it before?  This can happen for loopback, retrans,
 571                  * or local packets.
 572                  */
 573                 if (!nf_nat_initialized(ct, maniptype)) {
 574                         struct nf_nat_lookup_hook_priv *lpriv = priv;
 575                         struct nf_hook_entries *e = rcu_dereference(lpriv->entries);
 576                         unsigned int ret;
 577                         int i;
 578
 579                         if (!e)
 580                                 goto null_bind;
 581
 582                         for (i = 0; i < e->num_hook_entries; i++) {
 583                                 ret = e->hooks[i].hook(e->hooks[i].priv, skb,
 584                                                        state);
 585                                 if (ret != NF_ACCEPT)
 586                                         return ret;
 587                                 if (nf_nat_initialized(ct, maniptype))
 588                                         goto do_nat;
 589                         }
 590 null_bind:
 591                         ret = nf_nat_alloc_null_binding(ct, state->hook);
 592                         if (ret != NF_ACCEPT)
 593                                 return ret;
 594                 } else {
 595                         pr_debug("Already setup manip %s for ct %p (status bits 0x%lx)\n",
 596                                  maniptype == NF_NAT_MANIP_SRC ? "SRC" : "DST",
 597                                  ct, ct->status);
 598                         if (nf_nat_oif_changed(state->hook, ctinfo, nat,
 599                                                state->out))
 600                                 goto oif_changed;
 601                 }
 602                 break;
 603         default:
 604                 /* ESTABLISHED */
 605                 WARN_ON(ctinfo != IP_CT_ESTABLISHED &&
 606                         ctinfo != IP_CT_ESTABLISHED_REPLY);
 607                 if (nf_nat_oif_changed(state->hook, ctinfo, nat, state->out))
 608                         goto oif_changed;
 609         }
 610 do_nat:
 611         return nf_nat_packet(ct, ctinfo, state->hook, skb);
 612
 613 oif_changed:
 614         nf_ct_kill_acct(ct, ctinfo, skb);
 615         return NF_DROP;
 616 }
 617 EXPORT_SYMBOL_GPL(nf_nat_inet_fn);
 618
 619 struct nf_nat_proto_clean {
 620         u8      l3proto;
 621         u8      l4proto;
 622 };
 623
 624 /* kill conntracks with affected NAT section */
 625 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 626 {
 627         const struct nf_nat_proto_clean *clean = data;
 628
 629         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 630             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 631                 return 0;
 632
 633         return i->status & IPS_NAT_MASK ? 1 : 0;
 634 }
 635
 636 static void __nf_nat_cleanup_conntrack(struct nf_conn *ct)
 637 {
 638         unsigned int h;
 639
 640         h = hash_by_src(nf_ct_net(ct), &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 641         spin_lock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 642         hlist_del_rcu(&ct->nat_bysource);
 643         spin_unlock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 644 }
 645
 646 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 647 {
 648         if (nf_nat_proto_remove(ct, data))
 649                 return 1;
 650
 651         /* This module is being removed and conntrack has nat null binding.
 652          * Remove it from bysource hash, as the table will be freed soon.
 653          *
 654          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 655          * will delete entry from already-freed table.
 656          */
 657         if (test_and_clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status))
 658                 __nf_nat_cleanup_conntrack(ct);
 659
 660         /* don't delete conntrack.  Although that would make things a lot
 661          * simpler, we'd end up flushing all conntracks on nat rmmod.
 662          */
 663         return 0;
 664 }
 665
 666 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 667 {
 668         struct nf_nat_proto_clean clean = {
 669                 .l3proto = l3proto,
 670                 .l4proto = l4proto,
 671         };
 672
 673         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 674 }
 675
 676 static void nf_nat_l3proto_clean(u8 l3proto)
 677 {
 678         struct nf_nat_proto_clean clean = {
 679                 .l3proto = l3proto,
 680         };
 681
 682         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 683 }
 684
 685 /* Protocol registration. */
 686 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 687 {
 688         const struct nf_nat_l4proto **l4protos;
 689         unsigned int i;
 690         int ret = 0;
 691
 692         mutex_lock(&nf_nat_proto_mutex);
 693         if (nf_nat_l4protos[l3proto] == NULL) {
 694                 l4protos = kmalloc_array(IPPROTO_MAX,
 695                                          sizeof(struct nf_nat_l4proto *),
 696                                          GFP_KERNEL);
 697                 if (l4protos == NULL) {
 698                         ret = -ENOMEM;
 699                         goto out;
 700                 }
 701
 702                 for (i = 0; i < IPPROTO_MAX; i++)
 703                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 704
 705                 /* Before making proto_array visible to lockless readers,
 706                  * we must make sure its content is committed to memory.
 707                  */
 708                 smp_wmb();
 709
 710                 nf_nat_l4protos[l3proto] = l4protos;
 711         }
 712
 713         if (rcu_dereference_protected(
 714                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 715                         lockdep_is_held(&nf_nat_proto_mutex)
 716                         ) != &nf_nat_l4proto_unknown) {
 717                 ret = -EBUSY;
 718                 goto out;
 719         }
 720         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 721  out:
 722         mutex_unlock(&nf_nat_proto_mutex);
 723         return ret;
 724 }
 725 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 726
 727 /* No one stores the protocol anywhere; simply delete it. */
 728 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 729 {
 730         mutex_lock(&nf_nat_proto_mutex);
 731         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 732                          &nf_nat_l4proto_unknown);
 733         mutex_unlock(&nf_nat_proto_mutex);
 734         synchronize_rcu();
 735
 736         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 737 }
 738 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 739
 740 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 741 {
 742         int err;
 743
 744         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 745         if (err < 0)
 746                 return err;
 747
 748         mutex_lock(&nf_nat_proto_mutex);
 749         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 750                          &nf_nat_l4proto_tcp);
 751         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 752                          &nf_nat_l4proto_udp);
 753 #ifdef CONFIG_NF_NAT_PROTO_DCCP
 754         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_DCCP],
 755                          &nf_nat_l4proto_dccp);
 756 #endif
 757 #ifdef CONFIG_NF_NAT_PROTO_SCTP
 758         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_SCTP],
 759                          &nf_nat_l4proto_sctp);
 760 #endif
 761 #ifdef CONFIG_NF_NAT_PROTO_UDPLITE
 762         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDPLITE],
 763                          &nf_nat_l4proto_udplite);
 764 #endif
 765         mutex_unlock(&nf_nat_proto_mutex);
 766
 767         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 768         return 0;
 769 }
 770 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 771
 772 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 773 {
 774         mutex_lock(&nf_nat_proto_mutex);
 775         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 776         mutex_unlock(&nf_nat_proto_mutex);
 777         synchronize_rcu();
 778
 779         nf_nat_l3proto_clean(l3proto->l3proto);
 780         nf_ct_l3proto_module_put(l3proto->l3proto);
 781 }
 782 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 783
 784 /* No one using conntrack by the time this called. */
 785 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 786 {
 787         if (ct->status & IPS_SRC_NAT_DONE)
 788                 __nf_nat_cleanup_conntrack(ct);
 789 }
 790
 791 static struct nf_ct_ext_type nat_extend __read_mostly = {
 792         .len            = sizeof(struct nf_conn_nat),
 793         .align          = __alignof__(struct nf_conn_nat),
 794         .destroy        = nf_nat_cleanup_conntrack,
 795         .id             = NF_CT_EXT_NAT,
 796 };
 797
 798 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 799
 800 #include <linux/netfilter/nfnetlink.h>
 801 #include <linux/netfilter/nfnetlink_conntrack.h>
 802
 803 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 804         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 805         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 806 };
 807
 808 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 809                                      const struct nf_conn *ct,
 810                                      struct nf_nat_range2 *range)
 811 {
 812         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 813         const struct nf_nat_l4proto *l4proto;
 814         int err;
 815
 816         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr,
 817                                protonat_nla_policy, NULL);
 818         if (err < 0)
 819                 return err;
 820
 821         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 822         if (l4proto->nlattr_to_range)
 823                 err = l4proto->nlattr_to_range(tb, range);
 824
 825         return err;
 826 }
 827
 828 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 829         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 830         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 831         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 832         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 833         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 834 };
 835
 836 static int
 837 nfnetlink_parse_nat(const struct nlattr *nat,
 838                     const struct nf_conn *ct, struct nf_nat_range2 *range,
 839                     const struct nf_nat_l3proto *l3proto)
 840 {
 841         struct nlattr *tb[CTA_NAT_MAX+1];
 842         int err;
 843
 844         memset(range, 0, sizeof(*range));
 845
 846         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy, NULL);
 847         if (err < 0)
 848                 return err;
 849
 850         err = l3proto->nlattr_to_range(tb, range);
 851         if (err < 0)
 852                 return err;
 853
 854         if (!tb[CTA_NAT_PROTO])
 855                 return 0;
 856
 857         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 858 }
 859
 860 /* This function is called under rcu_read_lock() */
 861 static int
 862 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 863                           enum nf_nat_manip_type manip,
 864                           const struct nlattr *attr)
 865 {
 866         struct nf_nat_range2 range;
 867         const struct nf_nat_l3proto *l3proto;
 868         int err;
 869
 870         /* Should not happen, restricted to creating new conntracks
 871          * via ctnetlink.
 872          */
 873         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 874                 return -EEXIST;
 875
 876         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 877          * attach the null binding, otherwise this may oops.
 878          */
 879         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 880         if (l3proto == NULL)
 881                 return -EAGAIN;
 882
 883         /* No NAT information has been passed, allocate the null-binding */
 884         if (attr == NULL)
 885                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
 886
 887         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 888         if (err < 0)
 889                 return err;
 890
 891         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 892 }
 893 #else
 894 static int
 895 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 896                           enum nf_nat_manip_type manip,
 897                           const struct nlattr *attr)
 898 {
 899         return -EOPNOTSUPP;
 900 }
 901 #endif
 902
 903 static struct nf_ct_helper_expectfn follow_master_nat = {
 904         .name           = "nat-follow-master",
 905         .expectfn       = nf_nat_follow_master,
 906 };
 907
 908 int nf_nat_register_fn(struct net *net, const struct nf_hook_ops *ops,
 909                        const struct nf_hook_ops *orig_nat_ops, unsigned int ops_count)
 910 {
 911         struct nat_net *nat_net = net_generic(net, nat_net_id);
 912         struct nf_nat_hooks_net *nat_proto_net;
 913         struct nf_nat_lookup_hook_priv *priv;
 914         unsigned int hooknum = ops->hooknum;
 915         struct nf_hook_ops *nat_ops;
 916         int i, ret;
 917
 918         if (WARN_ON_ONCE(ops->pf >= ARRAY_SIZE(nat_net->nat_proto_net)))
 919                 return -EINVAL;
 920
 921         nat_proto_net = &nat_net->nat_proto_net[ops->pf];
 922
 923         for (i = 0; i < ops_count; i++) {
 924                 if (WARN_ON(orig_nat_ops[i].pf != ops->pf))
 925                         return -EINVAL;
 926                 if (orig_nat_ops[i].hooknum == hooknum) {
 927                         hooknum = i;
 928                         break;
 929                 }
 930         }
 931
 932         if (WARN_ON_ONCE(i == ops_count))
 933                 return -EINVAL;
 934
 935         mutex_lock(&nf_nat_proto_mutex);
 936         if (!nat_proto_net->nat_hook_ops) {
 937                 WARN_ON(nat_proto_net->users != 0);
 938
 939                 nat_ops = kmemdup(orig_nat_ops, sizeof(*orig_nat_ops) * ops_count, GFP_KERNEL);
 940                 if (!nat_ops) {
 941                         mutex_unlock(&nf_nat_proto_mutex);
 942                         return -ENOMEM;
 943                 }
 944
 945                 for (i = 0; i < ops_count; i++) {
 946                         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 947                         if (priv) {
 948                                 nat_ops[i].priv = priv;
 949                                 continue;
 950                         }
 951                         mutex_unlock(&nf_nat_proto_mutex);
 952                         while (i)
 953                                 kfree(nat_ops[--i].priv);
 954                         kfree(nat_ops);
 955                         return -ENOMEM;
 956                 }
 957
 958                 ret = nf_register_net_hooks(net, nat_ops, ops_count);
 959                 if (ret < 0) {
 960                         mutex_unlock(&nf_nat_proto_mutex);
 961                         for (i = 0; i < ops_count; i++)
 962                                 kfree(nat_ops[i].priv);
 963                         kfree(nat_ops);
 964                         return ret;
 965                 }
 966
 967                 nat_proto_net->nat_hook_ops = nat_ops;
 968         }
 969
 970         nat_ops = nat_proto_net->nat_hook_ops;
 971         priv = nat_ops[hooknum].priv;
 972         if (WARN_ON_ONCE(!priv)) {
 973                 mutex_unlock(&nf_nat_proto_mutex);
 974                 return -EOPNOTSUPP;
 975         }
 976
 977         ret = nf_hook_entries_insert_raw(&priv->entries, ops);
 978         if (ret == 0)
 979                 nat_proto_net->users++;
 980
 981         mutex_unlock(&nf_nat_proto_mutex);
 982         return ret;
 983 }
 984 EXPORT_SYMBOL_GPL(nf_nat_register_fn);
 985
 986 void nf_nat_unregister_fn(struct net *net, const struct nf_hook_ops *ops,
 987                           unsigned int ops_count)
 988 {
 989         struct nat_net *nat_net = net_generic(net, nat_net_id);
 990         struct nf_nat_hooks_net *nat_proto_net;
 991         struct nf_nat_lookup_hook_priv *priv;
 992         struct nf_hook_ops *nat_ops;
 993         int hooknum = ops->hooknum;
 994         int i;
 995
 996         if (ops->pf >= ARRAY_SIZE(nat_net->nat_proto_net))
 997                 return;
 998
 999         nat_proto_net = &nat_net->nat_proto_net[ops->pf];
1000
1001         mutex_lock(&nf_nat_proto_mutex);
1002         if (WARN_ON(nat_proto_net->users == 0))
1003                 goto unlock;
1004
1005         nat_proto_net->users--;
1006
1007         nat_ops = nat_proto_net->nat_hook_ops;
1008         for (i = 0; i < ops_count; i++) {
1009                 if (nat_ops[i].hooknum == hooknum) {
1010                         hooknum = i;
1011                         break;
1012                 }
1013         }
1014         if (WARN_ON_ONCE(i == ops_count))
1015                 goto unlock;
1016         priv = nat_ops[hooknum].priv;
1017         nf_hook_entries_delete_raw(&priv->entries, ops);
1018
1019         if (nat_proto_net->users == 0) {
1020                 nf_unregister_net_hooks(net, nat_ops, ops_count);
1021
1022                 for (i = 0; i < ops_count; i++) {
1023                         priv = nat_ops[i].priv;
1024                         kfree_rcu(priv, rcu_head);
1025                 }
1026
1027                 nat_proto_net->nat_hook_ops = NULL;
1028                 kfree(nat_ops);
1029         }
1030 unlock:
1031         mutex_unlock(&nf_nat_proto_mutex);
1032 }
1033 EXPORT_SYMBOL_GPL(nf_nat_unregister_fn);
1034
1035 static struct pernet_operations nat_net_ops = {
1036         .id = &nat_net_id,
1037         .size = sizeof(struct nat_net),
1038 };
1039
1040 static struct nf_nat_hook nat_hook = {
1041         .parse_nat_setup        = nfnetlink_parse_nat_setup,
1042 #ifdef CONFIG_XFRM
1043         .decode_session         = __nf_nat_decode_session,
1044 #endif
1045         .manip_pkt              = nf_nat_manip_pkt,
1046 };
1047
1048 static int __init nf_nat_init(void)
1049 {
1050         int ret, i;
1051
1052         /* Leave them the same for the moment. */
1053         nf_nat_htable_size = nf_conntrack_htable_size;
1054         if (nf_nat_htable_size < CONNTRACK_LOCKS)
1055                 nf_nat_htable_size = CONNTRACK_LOCKS;
1056
1057         nf_nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, 0);
1058         if (!nf_nat_bysource)
1059                 return -ENOMEM;
1060
1061         ret = nf_ct_extend_register(&nat_extend);
1062         if (ret < 0) {
1063                 nf_ct_free_hashtable(nf_nat_bysource, nf_nat_htable_size);
1064                 pr_err("Unable to register extension\n");
1065                 return ret;
1066         }
1067
1068         for (i = 0; i < CONNTRACK_LOCKS; i++)
1069                 spin_lock_init(&nf_nat_locks[i]);
1070
1071         ret = register_pernet_subsys(&nat_net_ops);
1072         if (ret < 0) {
1073                 nf_ct_extend_unregister(&nat_extend);
1074                 return ret;
1075         }
1076
1077         nf_ct_helper_expectfn_register(&follow_master_nat);
1078
1079         WARN_ON(nf_nat_hook != NULL);
1080         RCU_INIT_POINTER(nf_nat_hook, &nat_hook);
1081
1082         return 0;
1083 }
1084
1085 static void __exit nf_nat_cleanup(void)
1086 {
1087         struct nf_nat_proto_clean clean = {};
1088         unsigned int i;
1089
1090         nf_ct_iterate_destroy(nf_nat_proto_clean, &clean);
1091
1092         nf_ct_extend_unregister(&nat_extend);
1093         nf_ct_helper_expectfn_unregister(&follow_master_nat);
1094         RCU_INIT_POINTER(nf_nat_hook, NULL);
1095
1096         synchronize_rcu();
1097
1098         for (i = 0; i < NFPROTO_NUMPROTO; i++)
1099                 kfree(nf_nat_l4protos[i]);
1100         synchronize_net();
1101         nf_ct_free_hashtable(nf_nat_bysource, nf_nat_htable_size);
1102         unregister_pernet_subsys(&nat_net_ops);
1103 }
1104
1105 MODULE_LICENSE("GPL");
1106
1107 module_init(nf_nat_init);
1108 module_exit(nf_nat_cleanup);