net/netfilter/nf_nat_core.c

   1 /*
   2  * (C) 1999-2001 Paul `Rusty' Russell
   3  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   4  * (C) 2011 Patrick McHardy <kaber@trash.net>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/types.h>
  13 #include <linux/timer.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/gfp.h>
  16 #include <net/xfrm.h>
  17 #include <linux/jhash.h>
  18 #include <linux/rtnetlink.h>
  19
  20 #include <net/netfilter/nf_conntrack.h>
  21 #include <net/netfilter/nf_conntrack_core.h>
  22 #include <net/netfilter/nf_nat.h>
  23 #include <net/netfilter/nf_nat_l3proto.h>
  24 #include <net/netfilter/nf_nat_l4proto.h>
  25 #include <net/netfilter/nf_nat_core.h>
  26 #include <net/netfilter/nf_nat_helper.h>
  27 #include <net/netfilter/nf_conntrack_helper.h>
  28 #include <net/netfilter/nf_conntrack_seqadj.h>
  29 #include <net/netfilter/nf_conntrack_l3proto.h>
  30 #include <net/netfilter/nf_conntrack_zones.h>
  31 #include <linux/netfilter/nf_nat.h>
  32
  33 static DEFINE_MUTEX(nf_nat_proto_mutex);
  34 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
  35                                                 __read_mostly;
  36 static const struct nf_nat_l4proto __rcu **nf_nat_l4protos[NFPROTO_NUMPROTO]
  37                                                 __read_mostly;
  38
  39 struct nf_nat_conn_key {
  40         const struct net *net;
  41         const struct nf_conntrack_tuple *tuple;
  42         const struct nf_conntrack_zone *zone;
  43 };
  44
  45 static struct rhltable nf_nat_bysource_table;
  46
  47 inline const struct nf_nat_l3proto *
  48 __nf_nat_l3proto_find(u8 family)
  49 {
  50         return rcu_dereference(nf_nat_l3protos[family]);
  51 }
  52
  53 inline const struct nf_nat_l4proto *
  54 __nf_nat_l4proto_find(u8 family, u8 protonum)
  55 {
  56         return rcu_dereference(nf_nat_l4protos[family][protonum]);
  57 }
  58 EXPORT_SYMBOL_GPL(__nf_nat_l4proto_find);
  59
  60 #ifdef CONFIG_XFRM
  61 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  62 {
  63         const struct nf_nat_l3proto *l3proto;
  64         const struct nf_conn *ct;
  65         enum ip_conntrack_info ctinfo;
  66         enum ip_conntrack_dir dir;
  67         unsigned  long statusbit;
  68         u8 family;
  69
  70         ct = nf_ct_get(skb, &ctinfo);
  71         if (ct == NULL)
  72                 return;
  73
  74         family = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.l3num;
  75         rcu_read_lock();
  76         l3proto = __nf_nat_l3proto_find(family);
  77         if (l3proto == NULL)
  78                 goto out;
  79
  80         dir = CTINFO2DIR(ctinfo);
  81         if (dir == IP_CT_DIR_ORIGINAL)
  82                 statusbit = IPS_DST_NAT;
  83         else
  84                 statusbit = IPS_SRC_NAT;
  85
  86         l3proto->decode_session(skb, ct, dir, statusbit, fl);
  87 out:
  88         rcu_read_unlock();
  89 }
  90
  91 int nf_xfrm_me_harder(struct net *net, struct sk_buff *skb, unsigned int family)
  92 {
  93         struct flowi fl;
  94         unsigned int hh_len;
  95         struct dst_entry *dst;
  96         int err;
  97
  98         err = xfrm_decode_session(skb, &fl, family);
  99         if (err < 0)
 100                 return err;
 101
 102         dst = skb_dst(skb);
 103         if (dst->xfrm)
 104                 dst = ((struct xfrm_dst *)dst)->route;
 105         dst_hold(dst);
 106
 107         dst = xfrm_lookup(net, dst, &fl, skb->sk, 0);
 108         if (IS_ERR(dst))
 109                 return PTR_ERR(dst);
 110
 111         skb_dst_drop(skb);
 112         skb_dst_set(skb, dst);
 113
 114         /* Change in oif may mean change in hh_len. */
 115         hh_len = skb_dst(skb)->dev->hard_header_len;
 116         if (skb_headroom(skb) < hh_len &&
 117             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 118                 return -ENOMEM;
 119         return 0;
 120 }
 121 EXPORT_SYMBOL(nf_xfrm_me_harder);
 122 #endif /* CONFIG_XFRM */
 123
 124 static u32 nf_nat_bysource_hash(const void *data, u32 len, u32 seed)
 125 {
 126         const struct nf_conntrack_tuple *t;
 127         const struct nf_conn *ct = data;
 128
 129         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 130         /* Original src, to ensure we map it consistently if poss. */
 131
 132         seed ^= net_hash_mix(nf_ct_net(ct));
 133         return jhash2((const u32 *)&t->src, sizeof(t->src) / sizeof(u32),
 134                       t->dst.protonum ^ seed);
 135 }
 136
 137 /* Is this tuple already taken? (not by us) */
 138 int
 139 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 140                   const struct nf_conn *ignored_conntrack)
 141 {
 142         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 143          * incoming ones.  NAT means they don't have a fixed mapping,
 144          * so we invert the tuple and look for the incoming reply.
 145          *
 146          * We could keep a separate hash if this proves too slow.
 147          */
 148         struct nf_conntrack_tuple reply;
 149
 150         nf_ct_invert_tuplepr(&reply, tuple);
 151         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 152 }
 153 EXPORT_SYMBOL(nf_nat_used_tuple);
 154
 155 /* If we source map this tuple so reply looks like reply_tuple, will
 156  * that meet the constraints of range.
 157  */
 158 static int in_range(const struct nf_nat_l3proto *l3proto,
 159                     const struct nf_nat_l4proto *l4proto,
 160                     const struct nf_conntrack_tuple *tuple,
 161                     const struct nf_nat_range *range)
 162 {
 163         /* If we are supposed to map IPs, then we must be in the
 164          * range specified, otherwise let this drag us onto a new src IP.
 165          */
 166         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 167             !l3proto->in_range(tuple, range))
 168                 return 0;
 169
 170         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) ||
 171             l4proto->in_range(tuple, NF_NAT_MANIP_SRC,
 172                               &range->min_proto, &range->max_proto))
 173                 return 1;
 174
 175         return 0;
 176 }
 177
 178 static inline int
 179 same_src(const struct nf_conn *ct,
 180          const struct nf_conntrack_tuple *tuple)
 181 {
 182         const struct nf_conntrack_tuple *t;
 183
 184         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 185         return (t->dst.protonum == tuple->dst.protonum &&
 186                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 187                 t->src.u.all == tuple->src.u.all);
 188 }
 189
 190 static int nf_nat_bysource_cmp(struct rhashtable_compare_arg *arg,
 191                                const void *obj)
 192 {
 193         const struct nf_nat_conn_key *key = arg->key;
 194         const struct nf_conn *ct = obj;
 195
 196         if (!same_src(ct, key->tuple) ||
 197             !net_eq(nf_ct_net(ct), key->net) ||
 198             !nf_ct_zone_equal(ct, key->zone, IP_CT_DIR_ORIGINAL))
 199                 return 1;
 200
 201         return 0;
 202 }
 203
 204 static struct rhashtable_params nf_nat_bysource_params = {
 205         .head_offset = offsetof(struct nf_conn, nat_bysource),
 206         .obj_hashfn = nf_nat_bysource_hash,
 207         .obj_cmpfn = nf_nat_bysource_cmp,
 208         .nelem_hint = 256,
 209         .min_size = 1024,
 210 };
 211
 212 /* Only called for SRC manip */
 213 static int
 214 find_appropriate_src(struct net *net,
 215                      const struct nf_conntrack_zone *zone,
 216                      const struct nf_nat_l3proto *l3proto,
 217                      const struct nf_nat_l4proto *l4proto,
 218                      const struct nf_conntrack_tuple *tuple,
 219                      struct nf_conntrack_tuple *result,
 220                      const struct nf_nat_range *range)
 221 {
 222         const struct nf_conn *ct;
 223         struct nf_nat_conn_key key = {
 224                 .net = net,
 225                 .tuple = tuple,
 226                 .zone = zone
 227         };
 228         struct rhlist_head *hl;
 229
 230         hl = rhltable_lookup(&nf_nat_bysource_table, &key,
 231                              nf_nat_bysource_params);
 232         if (!hl)
 233                 return 0;
 234
 235         ct = container_of(hl, typeof(*ct), nat_bysource);
 236
 237         nf_ct_invert_tuplepr(result,
 238                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 239         result->dst = tuple->dst;
 240
 241         return in_range(l3proto, l4proto, result, range);
 242 }
 243
 244 /* For [FUTURE] fragmentation handling, we want the least-used
 245  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 246  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 247  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 248  * the ip with the lowest src-ip/dst-ip/proto usage.
 249  */
 250 static void
 251 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 252                     struct nf_conntrack_tuple *tuple,
 253                     const struct nf_nat_range *range,
 254                     const struct nf_conn *ct,
 255                     enum nf_nat_manip_type maniptype)
 256 {
 257         union nf_inet_addr *var_ipp;
 258         unsigned int i, max;
 259         /* Host order */
 260         u32 minip, maxip, j, dist;
 261         bool full_range;
 262
 263         /* No IP mapping?  Do nothing. */
 264         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 265                 return;
 266
 267         if (maniptype == NF_NAT_MANIP_SRC)
 268                 var_ipp = &tuple->src.u3;
 269         else
 270                 var_ipp = &tuple->dst.u3;
 271
 272         /* Fast path: only one choice. */
 273         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 274                 *var_ipp = range->min_addr;
 275                 return;
 276         }
 277
 278         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 279                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 280         else
 281                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 282
 283         /* Hashing source and destination IPs gives a fairly even
 284          * spread in practice (if there are a small number of IPs
 285          * involved, there usually aren't that many connections
 286          * anyway).  The consistency means that servers see the same
 287          * client coming from the same IP (some Internet Banking sites
 288          * like this), even across reboots.
 289          */
 290         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 291                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 292                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 293
 294         full_range = false;
 295         for (i = 0; i <= max; i++) {
 296                 /* If first bytes of the address are at the maximum, use the
 297                  * distance. Otherwise use the full range.
 298                  */
 299                 if (!full_range) {
 300                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 301                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 302                         dist  = maxip - minip + 1;
 303                 } else {
 304                         minip = 0;
 305                         dist  = ~0;
 306                 }
 307
 308                 var_ipp->all[i] = (__force __u32)
 309                         htonl(minip + reciprocal_scale(j, dist));
 310                 if (var_ipp->all[i] != range->max_addr.all[i])
 311                         full_range = true;
 312
 313                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 314                         j ^= (__force u32)tuple->dst.u3.all[i];
 315         }
 316 }
 317
 318 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 319  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 320  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 321  * range. It might not be possible to get a unique tuple, but we try.
 322  * At worst (or if we race), we will end up with a final duplicate in
 323  * __ip_conntrack_confirm and drop the packet. */
 324 static void
 325 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 326                  const struct nf_conntrack_tuple *orig_tuple,
 327                  const struct nf_nat_range *range,
 328                  struct nf_conn *ct,
 329                  enum nf_nat_manip_type maniptype)
 330 {
 331         const struct nf_conntrack_zone *zone;
 332         const struct nf_nat_l3proto *l3proto;
 333         const struct nf_nat_l4proto *l4proto;
 334         struct net *net = nf_ct_net(ct);
 335
 336         zone = nf_ct_zone(ct);
 337
 338         rcu_read_lock();
 339         l3proto = __nf_nat_l3proto_find(orig_tuple->src.l3num);
 340         l4proto = __nf_nat_l4proto_find(orig_tuple->src.l3num,
 341                                         orig_tuple->dst.protonum);
 342
 343         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 344          * and that same mapping gives a unique tuple within the given
 345          * range, use that.
 346          *
 347          * This is only required for source (ie. NAT/masq) mappings.
 348          * So far, we don't do local source mappings, so multiple
 349          * manips not an issue.
 350          */
 351         if (maniptype == NF_NAT_MANIP_SRC &&
 352             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 353                 /* try the original tuple first */
 354                 if (in_range(l3proto, l4proto, orig_tuple, range)) {
 355                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 356                                 *tuple = *orig_tuple;
 357                                 goto out;
 358                         }
 359                 } else if (find_appropriate_src(net, zone, l3proto, l4proto,
 360                                                 orig_tuple, tuple, range)) {
 361                         pr_debug("get_unique_tuple: Found current src map\n");
 362                         if (!nf_nat_used_tuple(tuple, ct))
 363                                 goto out;
 364                 }
 365         }
 366
 367         /* 2) Select the least-used IP/proto combination in the given range */
 368         *tuple = *orig_tuple;
 369         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 370
 371         /* 3) The per-protocol part of the manip is made to map into
 372          * the range to make a unique tuple.
 373          */
 374
 375         /* Only bother mapping if it's not already in range and unique */
 376         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 377                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 378                         if (l4proto->in_range(tuple, maniptype,
 379                                               &range->min_proto,
 380                                               &range->max_proto) &&
 381                             (range->min_proto.all == range->max_proto.all ||
 382                              !nf_nat_used_tuple(tuple, ct)))
 383                                 goto out;
 384                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 385                         goto out;
 386                 }
 387         }
 388
 389         /* Last change: get protocol to try to obtain unique tuple. */
 390         l4proto->unique_tuple(l3proto, tuple, range, maniptype, ct);
 391 out:
 392         rcu_read_unlock();
 393 }
 394
 395 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 396 {
 397         struct nf_conn_nat *nat = nfct_nat(ct);
 398         if (nat)
 399                 return nat;
 400
 401         if (!nf_ct_is_confirmed(ct))
 402                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 403
 404         return nat;
 405 }
 406 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 407
 408 unsigned int
 409 nf_nat_setup_info(struct nf_conn *ct,
 410                   const struct nf_nat_range *range,
 411                   enum nf_nat_manip_type maniptype)
 412 {
 413         struct nf_conntrack_tuple curr_tuple, new_tuple;
 414         struct nf_conn_nat *nat;
 415
 416         /* nat helper or nfctnetlink also setup binding */
 417         nat = nf_ct_nat_ext_add(ct);
 418         if (nat == NULL)
 419                 return NF_ACCEPT;
 420
 421         NF_CT_ASSERT(maniptype == NF_NAT_MANIP_SRC ||
 422                      maniptype == NF_NAT_MANIP_DST);
 423         BUG_ON(nf_nat_initialized(ct, maniptype));
 424
 425         /* What we've got will look like inverse of reply. Normally
 426          * this is what is in the conntrack, except for prior
 427          * manipulations (future optimization: if num_manips == 0,
 428          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 429          */
 430         nf_ct_invert_tuplepr(&curr_tuple,
 431                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 432
 433         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 434
 435         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 436                 struct nf_conntrack_tuple reply;
 437
 438                 /* Alter conntrack table so will recognize replies. */
 439                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 440                 nf_conntrack_alter_reply(ct, &reply);
 441
 442                 /* Non-atomic: we own this at the moment. */
 443                 if (maniptype == NF_NAT_MANIP_SRC)
 444                         ct->status |= IPS_SRC_NAT;
 445                 else
 446                         ct->status |= IPS_DST_NAT;
 447
 448                 if (nfct_help(ct))
 449                         if (!nfct_seqadj_ext_add(ct))
 450                                 return NF_DROP;
 451         }
 452
 453         if (maniptype == NF_NAT_MANIP_SRC) {
 454                 struct nf_nat_conn_key key = {
 455                         .net = nf_ct_net(ct),
 456                         .tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 457                         .zone = nf_ct_zone(ct),
 458                 };
 459                 int err;
 460
 461                 err = rhltable_insert_key(&nf_nat_bysource_table,
 462                                           &key,
 463                                           &ct->nat_bysource,
 464                                           nf_nat_bysource_params);
 465                 if (err)
 466                         return NF_DROP;
 467         }
 468
 469         /* It's done. */
 470         if (maniptype == NF_NAT_MANIP_DST)
 471                 ct->status |= IPS_DST_NAT_DONE;
 472         else
 473                 ct->status |= IPS_SRC_NAT_DONE;
 474
 475         return NF_ACCEPT;
 476 }
 477 EXPORT_SYMBOL(nf_nat_setup_info);
 478
 479 static unsigned int
 480 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 481 {
 482         /* Force range to this IP; let proto decide mapping for
 483          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 484          * Use reply in case it's already been mangled (eg local packet).
 485          */
 486         union nf_inet_addr ip =
 487                 (manip == NF_NAT_MANIP_SRC ?
 488                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 489                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 490         struct nf_nat_range range = {
 491                 .flags          = NF_NAT_RANGE_MAP_IPS,
 492                 .min_addr       = ip,
 493                 .max_addr       = ip,
 494         };
 495         return nf_nat_setup_info(ct, &range, manip);
 496 }
 497
 498 unsigned int
 499 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 500 {
 501         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 502 }
 503 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 504
 505 /* Do packet manipulations according to nf_nat_setup_info. */
 506 unsigned int nf_nat_packet(struct nf_conn *ct,
 507                            enum ip_conntrack_info ctinfo,
 508                            unsigned int hooknum,
 509                            struct sk_buff *skb)
 510 {
 511         const struct nf_nat_l3proto *l3proto;
 512         const struct nf_nat_l4proto *l4proto;
 513         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 514         unsigned long statusbit;
 515         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 516
 517         if (mtype == NF_NAT_MANIP_SRC)
 518                 statusbit = IPS_SRC_NAT;
 519         else
 520                 statusbit = IPS_DST_NAT;
 521
 522         /* Invert if this is reply dir. */
 523         if (dir == IP_CT_DIR_REPLY)
 524                 statusbit ^= IPS_NAT_MASK;
 525
 526         /* Non-atomic: these bits don't change. */
 527         if (ct->status & statusbit) {
 528                 struct nf_conntrack_tuple target;
 529
 530                 /* We are aiming to look like inverse of other direction. */
 531                 nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 532
 533                 l3proto = __nf_nat_l3proto_find(target.src.l3num);
 534                 l4proto = __nf_nat_l4proto_find(target.src.l3num,
 535                                                 target.dst.protonum);
 536                 if (!l3proto->manip_pkt(skb, 0, l4proto, &target, mtype))
 537                         return NF_DROP;
 538         }
 539         return NF_ACCEPT;
 540 }
 541 EXPORT_SYMBOL_GPL(nf_nat_packet);
 542
 543 struct nf_nat_proto_clean {
 544         u8      l3proto;
 545         u8      l4proto;
 546 };
 547
 548 /* kill conntracks with affected NAT section */
 549 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 550 {
 551         const struct nf_nat_proto_clean *clean = data;
 552         struct nf_conn_nat *nat = nfct_nat(i);
 553
 554         if (!nat)
 555                 return 0;
 556
 557         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 558             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 559                 return 0;
 560
 561         return i->status & IPS_NAT_MASK ? 1 : 0;
 562 }
 563
 564 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 565 {
 566         struct nf_conn_nat *nat = nfct_nat(ct);
 567
 568         if (nf_nat_proto_remove(ct, data))
 569                 return 1;
 570
 571         if (!nat)
 572                 return 0;
 573
 574         /* This netns is being destroyed, and conntrack has nat null binding.
 575          * Remove it from bysource hash, as the table will be freed soon.
 576          *
 577          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 578          * will delete entry from already-freed table.
 579          */
 580         ct->status &= ~IPS_NAT_DONE_MASK;
 581         rhltable_remove(&nf_nat_bysource_table, &ct->nat_bysource,
 582                         nf_nat_bysource_params);
 583
 584         /* don't delete conntrack.  Although that would make things a lot
 585          * simpler, we'd end up flushing all conntracks on nat rmmod.
 586          */
 587         return 0;
 588 }
 589
 590 static void nf_nat_l4proto_clean(u8 l3proto, u8 l4proto)
 591 {
 592         struct nf_nat_proto_clean clean = {
 593                 .l3proto = l3proto,
 594                 .l4proto = l4proto,
 595         };
 596         struct net *net;
 597
 598         rtnl_lock();
 599         for_each_net(net)
 600                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 601         rtnl_unlock();
 602 }
 603
 604 static void nf_nat_l3proto_clean(u8 l3proto)
 605 {
 606         struct nf_nat_proto_clean clean = {
 607                 .l3proto = l3proto,
 608         };
 609         struct net *net;
 610
 611         rtnl_lock();
 612
 613         for_each_net(net)
 614                 nf_ct_iterate_cleanup(net, nf_nat_proto_remove, &clean, 0, 0);
 615         rtnl_unlock();
 616 }
 617
 618 /* Protocol registration. */
 619 int nf_nat_l4proto_register(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 620 {
 621         const struct nf_nat_l4proto **l4protos;
 622         unsigned int i;
 623         int ret = 0;
 624
 625         mutex_lock(&nf_nat_proto_mutex);
 626         if (nf_nat_l4protos[l3proto] == NULL) {
 627                 l4protos = kmalloc(IPPROTO_MAX * sizeof(struct nf_nat_l4proto *),
 628                                    GFP_KERNEL);
 629                 if (l4protos == NULL) {
 630                         ret = -ENOMEM;
 631                         goto out;
 632                 }
 633
 634                 for (i = 0; i < IPPROTO_MAX; i++)
 635                         RCU_INIT_POINTER(l4protos[i], &nf_nat_l4proto_unknown);
 636
 637                 /* Before making proto_array visible to lockless readers,
 638                  * we must make sure its content is committed to memory.
 639                  */
 640                 smp_wmb();
 641
 642                 nf_nat_l4protos[l3proto] = l4protos;
 643         }
 644
 645         if (rcu_dereference_protected(
 646                         nf_nat_l4protos[l3proto][l4proto->l4proto],
 647                         lockdep_is_held(&nf_nat_proto_mutex)
 648                         ) != &nf_nat_l4proto_unknown) {
 649                 ret = -EBUSY;
 650                 goto out;
 651         }
 652         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto], l4proto);
 653  out:
 654         mutex_unlock(&nf_nat_proto_mutex);
 655         return ret;
 656 }
 657 EXPORT_SYMBOL_GPL(nf_nat_l4proto_register);
 658
 659 /* No one stores the protocol anywhere; simply delete it. */
 660 void nf_nat_l4proto_unregister(u8 l3proto, const struct nf_nat_l4proto *l4proto)
 661 {
 662         mutex_lock(&nf_nat_proto_mutex);
 663         RCU_INIT_POINTER(nf_nat_l4protos[l3proto][l4proto->l4proto],
 664                          &nf_nat_l4proto_unknown);
 665         mutex_unlock(&nf_nat_proto_mutex);
 666         synchronize_rcu();
 667
 668         nf_nat_l4proto_clean(l3proto, l4proto->l4proto);
 669 }
 670 EXPORT_SYMBOL_GPL(nf_nat_l4proto_unregister);
 671
 672 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 673 {
 674         int err;
 675
 676         err = nf_ct_l3proto_try_module_get(l3proto->l3proto);
 677         if (err < 0)
 678                 return err;
 679
 680         mutex_lock(&nf_nat_proto_mutex);
 681         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_TCP],
 682                          &nf_nat_l4proto_tcp);
 683         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDP],
 684                          &nf_nat_l4proto_udp);
 685 #ifdef CONFIG_NF_NAT_PROTO_DCCP
 686         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_DCCP],
 687                          &nf_nat_l4proto_dccp);
 688 #endif
 689 #ifdef CONFIG_NF_NAT_PROTO_SCTP
 690         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_SCTP],
 691                          &nf_nat_l4proto_sctp);
 692 #endif
 693 #ifdef CONFIG_NF_NAT_PROTO_UDPLITE
 694         RCU_INIT_POINTER(nf_nat_l4protos[l3proto->l3proto][IPPROTO_UDPLITE],
 695                          &nf_nat_l4proto_udplite);
 696 #endif
 697         mutex_unlock(&nf_nat_proto_mutex);
 698
 699         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 700         return 0;
 701 }
 702 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 703
 704 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 705 {
 706         mutex_lock(&nf_nat_proto_mutex);
 707         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 708         mutex_unlock(&nf_nat_proto_mutex);
 709         synchronize_rcu();
 710
 711         nf_nat_l3proto_clean(l3proto->l3proto);
 712         nf_ct_l3proto_module_put(l3proto->l3proto);
 713 }
 714 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 715
 716 /* No one using conntrack by the time this called. */
 717 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 718 {
 719         struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
 720
 721         if (!nat)
 722                 return;
 723
 724         rhltable_remove(&nf_nat_bysource_table, &ct->nat_bysource,
 725                         nf_nat_bysource_params);
 726 }
 727
 728 static struct nf_ct_ext_type nat_extend __read_mostly = {
 729         .len            = sizeof(struct nf_conn_nat),
 730         .align          = __alignof__(struct nf_conn_nat),
 731         .destroy        = nf_nat_cleanup_conntrack,
 732         .id             = NF_CT_EXT_NAT,
 733         .flags          = NF_CT_EXT_F_PREALLOC,
 734 };
 735
 736 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 737
 738 #include <linux/netfilter/nfnetlink.h>
 739 #include <linux/netfilter/nfnetlink_conntrack.h>
 740
 741 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 742         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 743         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 744 };
 745
 746 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 747                                      const struct nf_conn *ct,
 748                                      struct nf_nat_range *range)
 749 {
 750         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 751         const struct nf_nat_l4proto *l4proto;
 752         int err;
 753
 754         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr, protonat_nla_policy);
 755         if (err < 0)
 756                 return err;
 757
 758         l4proto = __nf_nat_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 759         if (l4proto->nlattr_to_range)
 760                 err = l4proto->nlattr_to_range(tb, range);
 761
 762         return err;
 763 }
 764
 765 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 766         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 767         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 768         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 769         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 770         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 771 };
 772
 773 static int
 774 nfnetlink_parse_nat(const struct nlattr *nat,
 775                     const struct nf_conn *ct, struct nf_nat_range *range,
 776                     const struct nf_nat_l3proto *l3proto)
 777 {
 778         struct nlattr *tb[CTA_NAT_MAX+1];
 779         int err;
 780
 781         memset(range, 0, sizeof(*range));
 782
 783         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy);
 784         if (err < 0)
 785                 return err;
 786
 787         err = l3proto->nlattr_to_range(tb, range);
 788         if (err < 0)
 789                 return err;
 790
 791         if (!tb[CTA_NAT_PROTO])
 792                 return 0;
 793
 794         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 795 }
 796
 797 /* This function is called under rcu_read_lock() */
 798 static int
 799 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 800                           enum nf_nat_manip_type manip,
 801                           const struct nlattr *attr)
 802 {
 803         struct nf_nat_range range;
 804         const struct nf_nat_l3proto *l3proto;
 805         int err;
 806
 807         /* Should not happen, restricted to creating new conntracks
 808          * via ctnetlink.
 809          */
 810         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 811                 return -EEXIST;
 812
 813         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 814          * attach the null binding, otherwise this may oops.
 815          */
 816         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 817         if (l3proto == NULL)
 818                 return -EAGAIN;
 819
 820         /* No NAT information has been passed, allocate the null-binding */
 821         if (attr == NULL)
 822                 return __nf_nat_alloc_null_binding(ct, manip);
 823
 824         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 825         if (err < 0)
 826                 return err;
 827
 828         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 829 }
 830 #else
 831 static int
 832 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 833                           enum nf_nat_manip_type manip,
 834                           const struct nlattr *attr)
 835 {
 836         return -EOPNOTSUPP;
 837 }
 838 #endif
 839
 840 static void __net_exit nf_nat_net_exit(struct net *net)
 841 {
 842         struct nf_nat_proto_clean clean = {};
 843
 844         nf_ct_iterate_cleanup(net, nf_nat_proto_clean, &clean, 0, 0);
 845 }
 846
 847 static struct pernet_operations nf_nat_net_ops = {
 848         .exit = nf_nat_net_exit,
 849 };
 850
 851 static struct nf_ct_helper_expectfn follow_master_nat = {
 852         .name           = "nat-follow-master",
 853         .expectfn       = nf_nat_follow_master,
 854 };
 855
 856 static int __init nf_nat_init(void)
 857 {
 858         int ret;
 859
 860         ret = rhltable_init(&nf_nat_bysource_table, &nf_nat_bysource_params);
 861         if (ret)
 862                 return ret;
 863
 864         ret = nf_ct_extend_register(&nat_extend);
 865         if (ret < 0) {
 866                 rhltable_destroy(&nf_nat_bysource_table);
 867                 printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
 868                 return ret;
 869         }
 870
 871         ret = register_pernet_subsys(&nf_nat_net_ops);
 872         if (ret < 0)
 873                 goto cleanup_extend;
 874
 875         nf_ct_helper_expectfn_register(&follow_master_nat);
 876
 877         /* Initialize fake conntrack so that NAT will skip it */
 878         nf_ct_untracked_status_or(IPS_NAT_DONE_MASK);
 879
 880         BUG_ON(nfnetlink_parse_nat_setup_hook != NULL);
 881         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook,
 882                            nfnetlink_parse_nat_setup);
 883 #ifdef CONFIG_XFRM
 884         BUG_ON(nf_nat_decode_session_hook != NULL);
 885         RCU_INIT_POINTER(nf_nat_decode_session_hook, __nf_nat_decode_session);
 886 #endif
 887         return 0;
 888
 889  cleanup_extend:
 890         rhltable_destroy(&nf_nat_bysource_table);
 891         nf_ct_extend_unregister(&nat_extend);
 892         return ret;
 893 }
 894
 895 static void __exit nf_nat_cleanup(void)
 896 {
 897         unsigned int i;
 898
 899         unregister_pernet_subsys(&nf_nat_net_ops);
 900         nf_ct_extend_unregister(&nat_extend);
 901         nf_ct_helper_expectfn_unregister(&follow_master_nat);
 902         RCU_INIT_POINTER(nfnetlink_parse_nat_setup_hook, NULL);
 903 #ifdef CONFIG_XFRM
 904         RCU_INIT_POINTER(nf_nat_decode_session_hook, NULL);
 905 #endif
 906         for (i = 0; i < NFPROTO_NUMPROTO; i++)
 907                 kfree(nf_nat_l4protos[i]);
 908
 909         rhltable_destroy(&nf_nat_bysource_table);
 910 }
 911
 912 MODULE_LICENSE("GPL");
 913
 914 module_init(nf_nat_init);
 915 module_exit(nf_nat_cleanup);