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_core.h>
  27 #include <net/netfilter/nf_nat_helper.h>
  28 #include <net/netfilter/nf_conntrack_helper.h>
  29 #include <net/netfilter/nf_conntrack_seqadj.h>
  30 #include <net/netfilter/nf_conntrack_zones.h>
  31 #include <linux/netfilter/nf_nat.h>
  32
  33 #include "nf_internals.h"
  34
  35 static spinlock_t nf_nat_locks[CONNTRACK_LOCKS];
  36
  37 static DEFINE_MUTEX(nf_nat_proto_mutex);
  38 static const struct nf_nat_l3proto __rcu *nf_nat_l3protos[NFPROTO_NUMPROTO]
  39                                                 __read_mostly;
  40 static unsigned int nat_net_id __read_mostly;
  41
  42 static struct hlist_head *nf_nat_bysource __read_mostly;
  43 static unsigned int nf_nat_htable_size __read_mostly;
  44 static unsigned int nf_nat_hash_rnd __read_mostly;
  45
  46 struct nf_nat_lookup_hook_priv {
  47         struct nf_hook_entries __rcu *entries;
  48
  49         struct rcu_head rcu_head;
  50 };
  51
  52 struct nf_nat_hooks_net {
  53         struct nf_hook_ops *nat_hook_ops;
  54         unsigned int users;
  55 };
  56
  57 struct nat_net {
  58         struct nf_nat_hooks_net nat_proto_net[NFPROTO_NUMPROTO];
  59 };
  60
  61 inline const struct nf_nat_l3proto *
  62 __nf_nat_l3proto_find(u8 family)
  63 {
  64         return rcu_dereference(nf_nat_l3protos[family]);
  65 }
  66
  67 #ifdef CONFIG_XFRM
  68 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
  69 {
  70         const struct nf_nat_l3proto *l3proto;
  71         const struct nf_conn *ct;
  72         enum ip_conntrack_info ctinfo;
  73         enum ip_conntrack_dir dir;
  74         unsigned  long statusbit;
  75         u8 family;
  76
  77         ct = nf_ct_get(skb, &ctinfo);
  78         if (ct == NULL)
  79                 return;
  80
  81         family = nf_ct_l3num(ct);
  82         l3proto = __nf_nat_l3proto_find(family);
  83         if (l3proto == NULL)
  84                 return;
  85
  86         dir = CTINFO2DIR(ctinfo);
  87         if (dir == IP_CT_DIR_ORIGINAL)
  88                 statusbit = IPS_DST_NAT;
  89         else
  90                 statusbit = IPS_SRC_NAT;
  91
  92         l3proto->decode_session(skb, ct, dir, statusbit, fl);
  93 }
  94
  95 int nf_xfrm_me_harder(struct net *net, struct sk_buff *skb, unsigned int family)
  96 {
  97         struct flowi fl;
  98         unsigned int hh_len;
  99         struct dst_entry *dst;
 100         struct sock *sk = skb->sk;
 101         int err;
 102
 103         err = xfrm_decode_session(skb, &fl, family);
 104         if (err < 0)
 105                 return err;
 106
 107         dst = skb_dst(skb);
 108         if (dst->xfrm)
 109                 dst = ((struct xfrm_dst *)dst)->route;
 110         dst_hold(dst);
 111
 112         if (sk && !net_eq(net, sock_net(sk)))
 113                 sk = NULL;
 114
 115         dst = xfrm_lookup(net, dst, &fl, sk, 0);
 116         if (IS_ERR(dst))
 117                 return PTR_ERR(dst);
 118
 119         skb_dst_drop(skb);
 120         skb_dst_set(skb, dst);
 121
 122         /* Change in oif may mean change in hh_len. */
 123         hh_len = skb_dst(skb)->dev->hard_header_len;
 124         if (skb_headroom(skb) < hh_len &&
 125             pskb_expand_head(skb, hh_len - skb_headroom(skb), 0, GFP_ATOMIC))
 126                 return -ENOMEM;
 127         return 0;
 128 }
 129 EXPORT_SYMBOL(nf_xfrm_me_harder);
 130 #endif /* CONFIG_XFRM */
 131
 132 /* We keep an extra hash for each conntrack, for fast searching. */
 133 static unsigned int
 134 hash_by_src(const struct net *n, const struct nf_conntrack_tuple *tuple)
 135 {
 136         unsigned int hash;
 137
 138         get_random_once(&nf_nat_hash_rnd, sizeof(nf_nat_hash_rnd));
 139
 140         /* Original src, to ensure we map it consistently if poss. */
 141         hash = jhash2((u32 *)&tuple->src, sizeof(tuple->src) / sizeof(u32),
 142                       tuple->dst.protonum ^ nf_nat_hash_rnd ^ net_hash_mix(n));
 143
 144         return reciprocal_scale(hash, nf_nat_htable_size);
 145 }
 146
 147 /* Is this tuple already taken? (not by us) */
 148 int
 149 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 150                   const struct nf_conn *ignored_conntrack)
 151 {
 152         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 153          * incoming ones.  NAT means they don't have a fixed mapping,
 154          * so we invert the tuple and look for the incoming reply.
 155          *
 156          * We could keep a separate hash if this proves too slow.
 157          */
 158         struct nf_conntrack_tuple reply;
 159
 160         nf_ct_invert_tuplepr(&reply, tuple);
 161         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 162 }
 163 EXPORT_SYMBOL(nf_nat_used_tuple);
 164
 165 static bool nf_nat_inet_in_range(const struct nf_conntrack_tuple *t,
 166                                  const struct nf_nat_range2 *range)
 167 {
 168         if (t->src.l3num == NFPROTO_IPV4)
 169                 return ntohl(t->src.u3.ip) >= ntohl(range->min_addr.ip) &&
 170                        ntohl(t->src.u3.ip) <= ntohl(range->max_addr.ip);
 171
 172         return ipv6_addr_cmp(&t->src.u3.in6, &range->min_addr.in6) >= 0 &&
 173                ipv6_addr_cmp(&t->src.u3.in6, &range->max_addr.in6) <= 0;
 174 }
 175
 176 /* Is the manipable part of the tuple between min and max incl? */
 177 static bool l4proto_in_range(const struct nf_conntrack_tuple *tuple,
 178                              enum nf_nat_manip_type maniptype,
 179                              const union nf_conntrack_man_proto *min,
 180                              const union nf_conntrack_man_proto *max)
 181 {
 182         __be16 port;
 183
 184         switch (tuple->dst.protonum) {
 185         case IPPROTO_ICMP: /* fallthrough */
 186         case IPPROTO_ICMPV6:
 187                 return ntohs(tuple->src.u.icmp.id) >= ntohs(min->icmp.id) &&
 188                        ntohs(tuple->src.u.icmp.id) <= ntohs(max->icmp.id);
 189         case IPPROTO_GRE: /* all fall though */
 190         case IPPROTO_TCP:
 191         case IPPROTO_UDP:
 192         case IPPROTO_UDPLITE:
 193         case IPPROTO_DCCP:
 194         case IPPROTO_SCTP:
 195                 if (maniptype == NF_NAT_MANIP_SRC)
 196                         port = tuple->src.u.all;
 197                 else
 198                         port = tuple->dst.u.all;
 199
 200                 return ntohs(port) >= ntohs(min->all) &&
 201                        ntohs(port) <= ntohs(max->all);
 202         default:
 203                 return true;
 204         }
 205 }
 206
 207 /* If we source map this tuple so reply looks like reply_tuple, will
 208  * that meet the constraints of range.
 209  */
 210 static int in_range(const struct nf_conntrack_tuple *tuple,
 211                     const struct nf_nat_range2 *range)
 212 {
 213         /* If we are supposed to map IPs, then we must be in the
 214          * range specified, otherwise let this drag us onto a new src IP.
 215          */
 216         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 217             !nf_nat_inet_in_range(tuple, range))
 218                 return 0;
 219
 220         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED))
 221                 return 1;
 222
 223         return l4proto_in_range(tuple, NF_NAT_MANIP_SRC,
 224                                 &range->min_proto, &range->max_proto);
 225 }
 226
 227 static inline int
 228 same_src(const struct nf_conn *ct,
 229          const struct nf_conntrack_tuple *tuple)
 230 {
 231         const struct nf_conntrack_tuple *t;
 232
 233         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 234         return (t->dst.protonum == tuple->dst.protonum &&
 235                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 236                 t->src.u.all == tuple->src.u.all);
 237 }
 238
 239 /* Only called for SRC manip */
 240 static int
 241 find_appropriate_src(struct net *net,
 242                      const struct nf_conntrack_zone *zone,
 243                      const struct nf_conntrack_tuple *tuple,
 244                      struct nf_conntrack_tuple *result,
 245                      const struct nf_nat_range2 *range)
 246 {
 247         unsigned int h = hash_by_src(net, tuple);
 248         const struct nf_conn *ct;
 249
 250         hlist_for_each_entry_rcu(ct, &nf_nat_bysource[h], nat_bysource) {
 251                 if (same_src(ct, tuple) &&
 252                     net_eq(net, nf_ct_net(ct)) &&
 253                     nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) {
 254                         /* Copy source part from reply tuple. */
 255                         nf_ct_invert_tuplepr(result,
 256                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 257                         result->dst = tuple->dst;
 258
 259                         if (in_range(result, range))
 260                                 return 1;
 261                 }
 262         }
 263         return 0;
 264 }
 265
 266 /* For [FUTURE] fragmentation handling, we want the least-used
 267  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 268  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 269  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 270  * the ip with the lowest src-ip/dst-ip/proto usage.
 271  */
 272 static void
 273 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 274                     struct nf_conntrack_tuple *tuple,
 275                     const struct nf_nat_range2 *range,
 276                     const struct nf_conn *ct,
 277                     enum nf_nat_manip_type maniptype)
 278 {
 279         union nf_inet_addr *var_ipp;
 280         unsigned int i, max;
 281         /* Host order */
 282         u32 minip, maxip, j, dist;
 283         bool full_range;
 284
 285         /* No IP mapping?  Do nothing. */
 286         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 287                 return;
 288
 289         if (maniptype == NF_NAT_MANIP_SRC)
 290                 var_ipp = &tuple->src.u3;
 291         else
 292                 var_ipp = &tuple->dst.u3;
 293
 294         /* Fast path: only one choice. */
 295         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 296                 *var_ipp = range->min_addr;
 297                 return;
 298         }
 299
 300         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 301                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 302         else
 303                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 304
 305         /* Hashing source and destination IPs gives a fairly even
 306          * spread in practice (if there are a small number of IPs
 307          * involved, there usually aren't that many connections
 308          * anyway).  The consistency means that servers see the same
 309          * client coming from the same IP (some Internet Banking sites
 310          * like this), even across reboots.
 311          */
 312         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 313                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 314                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 315
 316         full_range = false;
 317         for (i = 0; i <= max; i++) {
 318                 /* If first bytes of the address are at the maximum, use the
 319                  * distance. Otherwise use the full range.
 320                  */
 321                 if (!full_range) {
 322                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 323                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 324                         dist  = maxip - minip + 1;
 325                 } else {
 326                         minip = 0;
 327                         dist  = ~0;
 328                 }
 329
 330                 var_ipp->all[i] = (__force __u32)
 331                         htonl(minip + reciprocal_scale(j, dist));
 332                 if (var_ipp->all[i] != range->max_addr.all[i])
 333                         full_range = true;
 334
 335                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 336                         j ^= (__force u32)tuple->dst.u3.all[i];
 337         }
 338 }
 339
 340 /* Alter the per-proto part of the tuple (depending on maniptype), to
 341  * give a unique tuple in the given range if possible.
 342  *
 343  * Per-protocol part of tuple is initialized to the incoming packet.
 344  */
 345 static void nf_nat_l4proto_unique_tuple(struct nf_conntrack_tuple *tuple,
 346                                         const struct nf_nat_range2 *range,
 347                                         enum nf_nat_manip_type maniptype,
 348                                         const struct nf_conn *ct)
 349 {
 350         unsigned int range_size, min, max, i, attempts;
 351         __be16 *keyptr;
 352         u16 off;
 353         static const unsigned int max_attempts = 128;
 354
 355         switch (tuple->dst.protonum) {
 356         case IPPROTO_ICMP: /* fallthrough */
 357         case IPPROTO_ICMPV6:
 358                 /* id is same for either direction... */
 359                 keyptr = &tuple->src.u.icmp.id;
 360                 min = range->min_proto.icmp.id;
 361                 range_size = ntohs(range->max_proto.icmp.id) -
 362                              ntohs(range->min_proto.icmp.id) + 1;
 363                 goto find_free_id;
 364 #if IS_ENABLED(CONFIG_NF_CT_PROTO_GRE)
 365         case IPPROTO_GRE:
 366                 /* If there is no master conntrack we are not PPTP,
 367                    do not change tuples */
 368                 if (!ct->master)
 369                         return;
 370
 371                 if (maniptype == NF_NAT_MANIP_SRC)
 372                         keyptr = &tuple->src.u.gre.key;
 373                 else
 374                         keyptr = &tuple->dst.u.gre.key;
 375
 376                 if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED)) {
 377                         min = 1;
 378                         range_size = 65535;
 379                 } else {
 380                         min = ntohs(range->min_proto.gre.key);
 381                         range_size = ntohs(range->max_proto.gre.key) - min + 1;
 382                 }
 383                 goto find_free_id;
 384 #endif
 385         case IPPROTO_UDP:       /* fallthrough */
 386         case IPPROTO_UDPLITE:   /* fallthrough */
 387         case IPPROTO_TCP:       /* fallthrough */
 388         case IPPROTO_SCTP:      /* fallthrough */
 389         case IPPROTO_DCCP:      /* fallthrough */
 390                 if (maniptype == NF_NAT_MANIP_SRC)
 391                         keyptr = &tuple->src.u.all;
 392                 else
 393                         keyptr = &tuple->dst.u.all;
 394
 395                 break;
 396         default:
 397                 return;
 398         }
 399
 400         /* If no range specified... */
 401         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED)) {
 402                 /* If it's dst rewrite, can't change port */
 403                 if (maniptype == NF_NAT_MANIP_DST)
 404                         return;
 405
 406                 if (ntohs(*keyptr) < 1024) {
 407                         /* Loose convention: >> 512 is credential passing */
 408                         if (ntohs(*keyptr) < 512) {
 409                                 min = 1;
 410                                 range_size = 511 - min + 1;
 411                         } else {
 412                                 min = 600;
 413                                 range_size = 1023 - min + 1;
 414                         }
 415                 } else {
 416                         min = 1024;
 417                         range_size = 65535 - 1024 + 1;
 418                 }
 419         } else {
 420                 min = ntohs(range->min_proto.all);
 421                 max = ntohs(range->max_proto.all);
 422                 if (unlikely(max < min))
 423                         swap(max, min);
 424                 range_size = max - min + 1;
 425         }
 426
 427 find_free_id:
 428         if (range->flags & NF_NAT_RANGE_PROTO_OFFSET)
 429                 off = (ntohs(*keyptr) - ntohs(range->base_proto.all));
 430         else
 431                 off = prandom_u32();
 432
 433         attempts = range_size;
 434         if (attempts > max_attempts)
 435                 attempts = max_attempts;
 436
 437         /* We are in softirq; doing a search of the entire range risks
 438          * soft lockup when all tuples are already used.
 439          *
 440          * If we can't find any free port from first offset, pick a new
 441          * one and try again, with ever smaller search window.
 442          */
 443 another_round:
 444         for (i = 0; i < attempts; i++, off++) {
 445                 *keyptr = htons(min + off % range_size);
 446                 if (!nf_nat_used_tuple(tuple, ct))
 447                         return;
 448         }
 449
 450         if (attempts >= range_size || attempts < 16)
 451                 return;
 452         attempts /= 2;
 453         off = prandom_u32();
 454         goto another_round;
 455 }
 456
 457 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 458  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 459  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 460  * range. It might not be possible to get a unique tuple, but we try.
 461  * At worst (or if we race), we will end up with a final duplicate in
 462  * __ip_conntrack_confirm and drop the packet. */
 463 static void
 464 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 465                  const struct nf_conntrack_tuple *orig_tuple,
 466                  const struct nf_nat_range2 *range,
 467                  struct nf_conn *ct,
 468                  enum nf_nat_manip_type maniptype)
 469 {
 470         const struct nf_conntrack_zone *zone;
 471         struct net *net = nf_ct_net(ct);
 472
 473         zone = nf_ct_zone(ct);
 474
 475         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 476          * and that same mapping gives a unique tuple within the given
 477          * range, use that.
 478          *
 479          * This is only required for source (ie. NAT/masq) mappings.
 480          * So far, we don't do local source mappings, so multiple
 481          * manips not an issue.
 482          */
 483         if (maniptype == NF_NAT_MANIP_SRC &&
 484             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 485                 /* try the original tuple first */
 486                 if (in_range(orig_tuple, range)) {
 487                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 488                                 *tuple = *orig_tuple;
 489                                 return;
 490                         }
 491                 } else if (find_appropriate_src(net, zone,
 492                                                 orig_tuple, tuple, range)) {
 493                         pr_debug("get_unique_tuple: Found current src map\n");
 494                         if (!nf_nat_used_tuple(tuple, ct))
 495                                 return;
 496                 }
 497         }
 498
 499         /* 2) Select the least-used IP/proto combination in the given range */
 500         *tuple = *orig_tuple;
 501         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 502
 503         /* 3) The per-protocol part of the manip is made to map into
 504          * the range to make a unique tuple.
 505          */
 506
 507         /* Only bother mapping if it's not already in range and unique */
 508         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 509                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 510                         if (!(range->flags & NF_NAT_RANGE_PROTO_OFFSET) &&
 511                             l4proto_in_range(tuple, maniptype,
 512                                   &range->min_proto,
 513                                   &range->max_proto) &&
 514                             (range->min_proto.all == range->max_proto.all ||
 515                              !nf_nat_used_tuple(tuple, ct)))
 516                                 return;
 517                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 518                         return;
 519                 }
 520         }
 521
 522         /* Last chance: get protocol to try to obtain unique tuple. */
 523         nf_nat_l4proto_unique_tuple(tuple, range, maniptype, ct);
 524 }
 525
 526 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 527 {
 528         struct nf_conn_nat *nat = nfct_nat(ct);
 529         if (nat)
 530                 return nat;
 531
 532         if (!nf_ct_is_confirmed(ct))
 533                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 534
 535         return nat;
 536 }
 537 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 538
 539 unsigned int
 540 nf_nat_setup_info(struct nf_conn *ct,
 541                   const struct nf_nat_range2 *range,
 542                   enum nf_nat_manip_type maniptype)
 543 {
 544         struct net *net = nf_ct_net(ct);
 545         struct nf_conntrack_tuple curr_tuple, new_tuple;
 546
 547         /* Can't setup nat info for confirmed ct. */
 548         if (nf_ct_is_confirmed(ct))
 549                 return NF_ACCEPT;
 550
 551         WARN_ON(maniptype != NF_NAT_MANIP_SRC &&
 552                 maniptype != NF_NAT_MANIP_DST);
 553
 554         if (WARN_ON(nf_nat_initialized(ct, maniptype)))
 555                 return NF_DROP;
 556
 557         /* What we've got will look like inverse of reply. Normally
 558          * this is what is in the conntrack, except for prior
 559          * manipulations (future optimization: if num_manips == 0,
 560          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 561          */
 562         nf_ct_invert_tuplepr(&curr_tuple,
 563                              &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 564
 565         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 566
 567         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 568                 struct nf_conntrack_tuple reply;
 569
 570                 /* Alter conntrack table so will recognize replies. */
 571                 nf_ct_invert_tuplepr(&reply, &new_tuple);
 572                 nf_conntrack_alter_reply(ct, &reply);
 573
 574                 /* Non-atomic: we own this at the moment. */
 575                 if (maniptype == NF_NAT_MANIP_SRC)
 576                         ct->status |= IPS_SRC_NAT;
 577                 else
 578                         ct->status |= IPS_DST_NAT;
 579
 580                 if (nfct_help(ct) && !nfct_seqadj(ct))
 581                         if (!nfct_seqadj_ext_add(ct))
 582                                 return NF_DROP;
 583         }
 584
 585         if (maniptype == NF_NAT_MANIP_SRC) {
 586                 unsigned int srchash;
 587                 spinlock_t *lock;
 588
 589                 srchash = hash_by_src(net,
 590                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 591                 lock = &nf_nat_locks[srchash % CONNTRACK_LOCKS];
 592                 spin_lock_bh(lock);
 593                 hlist_add_head_rcu(&ct->nat_bysource,
 594                                    &nf_nat_bysource[srchash]);
 595                 spin_unlock_bh(lock);
 596         }
 597
 598         /* It's done. */
 599         if (maniptype == NF_NAT_MANIP_DST)
 600                 ct->status |= IPS_DST_NAT_DONE;
 601         else
 602                 ct->status |= IPS_SRC_NAT_DONE;
 603
 604         return NF_ACCEPT;
 605 }
 606 EXPORT_SYMBOL(nf_nat_setup_info);
 607
 608 static unsigned int
 609 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 610 {
 611         /* Force range to this IP; let proto decide mapping for
 612          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 613          * Use reply in case it's already been mangled (eg local packet).
 614          */
 615         union nf_inet_addr ip =
 616                 (manip == NF_NAT_MANIP_SRC ?
 617                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 618                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 619         struct nf_nat_range2 range = {
 620                 .flags          = NF_NAT_RANGE_MAP_IPS,
 621                 .min_addr       = ip,
 622                 .max_addr       = ip,
 623         };
 624         return nf_nat_setup_info(ct, &range, manip);
 625 }
 626
 627 unsigned int
 628 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 629 {
 630         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 631 }
 632 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 633
 634 static unsigned int nf_nat_manip_pkt(struct sk_buff *skb, struct nf_conn *ct,
 635                                      enum nf_nat_manip_type mtype,
 636                                      enum ip_conntrack_dir dir)
 637 {
 638         const struct nf_nat_l3proto *l3proto;
 639         struct nf_conntrack_tuple target;
 640
 641         /* We are aiming to look like inverse of other direction. */
 642         nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 643
 644         l3proto = __nf_nat_l3proto_find(target.src.l3num);
 645         if (!l3proto->manip_pkt(skb, 0, &target, mtype))
 646                 return NF_DROP;
 647
 648         return NF_ACCEPT;
 649 }
 650
 651 /* Do packet manipulations according to nf_nat_setup_info. */
 652 unsigned int nf_nat_packet(struct nf_conn *ct,
 653                            enum ip_conntrack_info ctinfo,
 654                            unsigned int hooknum,
 655                            struct sk_buff *skb)
 656 {
 657         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 658         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 659         unsigned int verdict = NF_ACCEPT;
 660         unsigned long statusbit;
 661
 662         if (mtype == NF_NAT_MANIP_SRC)
 663                 statusbit = IPS_SRC_NAT;
 664         else
 665                 statusbit = IPS_DST_NAT;
 666
 667         /* Invert if this is reply dir. */
 668         if (dir == IP_CT_DIR_REPLY)
 669                 statusbit ^= IPS_NAT_MASK;
 670
 671         /* Non-atomic: these bits don't change. */
 672         if (ct->status & statusbit)
 673                 verdict = nf_nat_manip_pkt(skb, ct, mtype, dir);
 674
 675         return verdict;
 676 }
 677 EXPORT_SYMBOL_GPL(nf_nat_packet);
 678
 679 unsigned int
 680 nf_nat_inet_fn(void *priv, struct sk_buff *skb,
 681                const struct nf_hook_state *state)
 682 {
 683         struct nf_conn *ct;
 684         enum ip_conntrack_info ctinfo;
 685         struct nf_conn_nat *nat;
 686         /* maniptype == SRC for postrouting. */
 687         enum nf_nat_manip_type maniptype = HOOK2MANIP(state->hook);
 688
 689         ct = nf_ct_get(skb, &ctinfo);
 690         /* Can't track?  It's not due to stress, or conntrack would
 691          * have dropped it.  Hence it's the user's responsibilty to
 692          * packet filter it out, or implement conntrack/NAT for that
 693          * protocol. 8) --RR
 694          */
 695         if (!ct)
 696                 return NF_ACCEPT;
 697
 698         nat = nfct_nat(ct);
 699
 700         switch (ctinfo) {
 701         case IP_CT_RELATED:
 702         case IP_CT_RELATED_REPLY:
 703                 /* Only ICMPs can be IP_CT_IS_REPLY.  Fallthrough */
 704         case IP_CT_NEW:
 705                 /* Seen it before?  This can happen for loopback, retrans,
 706                  * or local packets.
 707                  */
 708                 if (!nf_nat_initialized(ct, maniptype)) {
 709                         struct nf_nat_lookup_hook_priv *lpriv = priv;
 710                         struct nf_hook_entries *e = rcu_dereference(lpriv->entries);
 711                         unsigned int ret;
 712                         int i;
 713
 714                         if (!e)
 715                                 goto null_bind;
 716
 717                         for (i = 0; i < e->num_hook_entries; i++) {
 718                                 ret = e->hooks[i].hook(e->hooks[i].priv, skb,
 719                                                        state);
 720                                 if (ret != NF_ACCEPT)
 721                                         return ret;
 722                                 if (nf_nat_initialized(ct, maniptype))
 723                                         goto do_nat;
 724                         }
 725 null_bind:
 726                         ret = nf_nat_alloc_null_binding(ct, state->hook);
 727                         if (ret != NF_ACCEPT)
 728                                 return ret;
 729                 } else {
 730                         pr_debug("Already setup manip %s for ct %p (status bits 0x%lx)\n",
 731                                  maniptype == NF_NAT_MANIP_SRC ? "SRC" : "DST",
 732                                  ct, ct->status);
 733                         if (nf_nat_oif_changed(state->hook, ctinfo, nat,
 734                                                state->out))
 735                                 goto oif_changed;
 736                 }
 737                 break;
 738         default:
 739                 /* ESTABLISHED */
 740                 WARN_ON(ctinfo != IP_CT_ESTABLISHED &&
 741                         ctinfo != IP_CT_ESTABLISHED_REPLY);
 742                 if (nf_nat_oif_changed(state->hook, ctinfo, nat, state->out))
 743                         goto oif_changed;
 744         }
 745 do_nat:
 746         return nf_nat_packet(ct, ctinfo, state->hook, skb);
 747
 748 oif_changed:
 749         nf_ct_kill_acct(ct, ctinfo, skb);
 750         return NF_DROP;
 751 }
 752 EXPORT_SYMBOL_GPL(nf_nat_inet_fn);
 753
 754 struct nf_nat_proto_clean {
 755         u8      l3proto;
 756         u8      l4proto;
 757 };
 758
 759 /* kill conntracks with affected NAT section */
 760 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 761 {
 762         const struct nf_nat_proto_clean *clean = data;
 763
 764         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 765             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 766                 return 0;
 767
 768         return i->status & IPS_NAT_MASK ? 1 : 0;
 769 }
 770
 771 static void __nf_nat_cleanup_conntrack(struct nf_conn *ct)
 772 {
 773         unsigned int h;
 774
 775         h = hash_by_src(nf_ct_net(ct), &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 776         spin_lock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 777         hlist_del_rcu(&ct->nat_bysource);
 778         spin_unlock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 779 }
 780
 781 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 782 {
 783         if (nf_nat_proto_remove(ct, data))
 784                 return 1;
 785
 786         /* This module is being removed and conntrack has nat null binding.
 787          * Remove it from bysource hash, as the table will be freed soon.
 788          *
 789          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 790          * will delete entry from already-freed table.
 791          */
 792         if (test_and_clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status))
 793                 __nf_nat_cleanup_conntrack(ct);
 794
 795         /* don't delete conntrack.  Although that would make things a lot
 796          * simpler, we'd end up flushing all conntracks on nat rmmod.
 797          */
 798         return 0;
 799 }
 800
 801 static void nf_nat_l3proto_clean(u8 l3proto)
 802 {
 803         struct nf_nat_proto_clean clean = {
 804                 .l3proto = l3proto,
 805         };
 806
 807         nf_ct_iterate_destroy(nf_nat_proto_remove, &clean);
 808 }
 809
 810 int nf_nat_l3proto_register(const struct nf_nat_l3proto *l3proto)
 811 {
 812         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], l3proto);
 813         return 0;
 814 }
 815 EXPORT_SYMBOL_GPL(nf_nat_l3proto_register);
 816
 817 void nf_nat_l3proto_unregister(const struct nf_nat_l3proto *l3proto)
 818 {
 819         mutex_lock(&nf_nat_proto_mutex);
 820         RCU_INIT_POINTER(nf_nat_l3protos[l3proto->l3proto], NULL);
 821         mutex_unlock(&nf_nat_proto_mutex);
 822         synchronize_rcu();
 823
 824         nf_nat_l3proto_clean(l3proto->l3proto);
 825 }
 826 EXPORT_SYMBOL_GPL(nf_nat_l3proto_unregister);
 827
 828 /* No one using conntrack by the time this called. */
 829 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 830 {
 831         if (ct->status & IPS_SRC_NAT_DONE)
 832                 __nf_nat_cleanup_conntrack(ct);
 833 }
 834
 835 static struct nf_ct_ext_type nat_extend __read_mostly = {
 836         .len            = sizeof(struct nf_conn_nat),
 837         .align          = __alignof__(struct nf_conn_nat),
 838         .destroy        = nf_nat_cleanup_conntrack,
 839         .id             = NF_CT_EXT_NAT,
 840 };
 841
 842 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 843
 844 #include <linux/netfilter/nfnetlink.h>
 845 #include <linux/netfilter/nfnetlink_conntrack.h>
 846
 847 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 848         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 849         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 850 };
 851
 852 static int nf_nat_l4proto_nlattr_to_range(struct nlattr *tb[],
 853                                           struct nf_nat_range2 *range)
 854 {
 855         if (tb[CTA_PROTONAT_PORT_MIN]) {
 856                 range->min_proto.all = nla_get_be16(tb[CTA_PROTONAT_PORT_MIN]);
 857                 range->max_proto.all = range->min_proto.all;
 858                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
 859         }
 860         if (tb[CTA_PROTONAT_PORT_MAX]) {
 861                 range->max_proto.all = nla_get_be16(tb[CTA_PROTONAT_PORT_MAX]);
 862                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
 863         }
 864         return 0;
 865 }
 866
 867 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 868                                      const struct nf_conn *ct,
 869                                      struct nf_nat_range2 *range)
 870 {
 871         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 872         int err;
 873
 874         err = nla_parse_nested(tb, CTA_PROTONAT_MAX, attr,
 875                                protonat_nla_policy, NULL);
 876         if (err < 0)
 877                 return err;
 878
 879         return nf_nat_l4proto_nlattr_to_range(tb, range);
 880 }
 881
 882 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 883         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 884         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 885         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 886         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 887         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 888 };
 889
 890 static int
 891 nfnetlink_parse_nat(const struct nlattr *nat,
 892                     const struct nf_conn *ct, struct nf_nat_range2 *range,
 893                     const struct nf_nat_l3proto *l3proto)
 894 {
 895         struct nlattr *tb[CTA_NAT_MAX+1];
 896         int err;
 897
 898         memset(range, 0, sizeof(*range));
 899
 900         err = nla_parse_nested(tb, CTA_NAT_MAX, nat, nat_nla_policy, NULL);
 901         if (err < 0)
 902                 return err;
 903
 904         err = l3proto->nlattr_to_range(tb, range);
 905         if (err < 0)
 906                 return err;
 907
 908         if (!tb[CTA_NAT_PROTO])
 909                 return 0;
 910
 911         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 912 }
 913
 914 /* This function is called under rcu_read_lock() */
 915 static int
 916 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 917                           enum nf_nat_manip_type manip,
 918                           const struct nlattr *attr)
 919 {
 920         struct nf_nat_range2 range;
 921         const struct nf_nat_l3proto *l3proto;
 922         int err;
 923
 924         /* Should not happen, restricted to creating new conntracks
 925          * via ctnetlink.
 926          */
 927         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 928                 return -EEXIST;
 929
 930         /* Make sure that L3 NAT is there by when we call nf_nat_setup_info to
 931          * attach the null binding, otherwise this may oops.
 932          */
 933         l3proto = __nf_nat_l3proto_find(nf_ct_l3num(ct));
 934         if (l3proto == NULL)
 935                 return -EAGAIN;
 936
 937         /* No NAT information has been passed, allocate the null-binding */
 938         if (attr == NULL)
 939                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
 940
 941         err = nfnetlink_parse_nat(attr, ct, &range, l3proto);
 942         if (err < 0)
 943                 return err;
 944
 945         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 946 }
 947 #else
 948 static int
 949 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 950                           enum nf_nat_manip_type manip,
 951                           const struct nlattr *attr)
 952 {
 953         return -EOPNOTSUPP;
 954 }
 955 #endif
 956
 957 static struct nf_ct_helper_expectfn follow_master_nat = {
 958         .name           = "nat-follow-master",
 959         .expectfn       = nf_nat_follow_master,
 960 };
 961
 962 int nf_nat_register_fn(struct net *net, const struct nf_hook_ops *ops,
 963                        const struct nf_hook_ops *orig_nat_ops, unsigned int ops_count)
 964 {
 965         struct nat_net *nat_net = net_generic(net, nat_net_id);
 966         struct nf_nat_hooks_net *nat_proto_net;
 967         struct nf_nat_lookup_hook_priv *priv;
 968         unsigned int hooknum = ops->hooknum;
 969         struct nf_hook_ops *nat_ops;
 970         int i, ret;
 971
 972         if (WARN_ON_ONCE(ops->pf >= ARRAY_SIZE(nat_net->nat_proto_net)))
 973                 return -EINVAL;
 974
 975         nat_proto_net = &nat_net->nat_proto_net[ops->pf];
 976
 977         for (i = 0; i < ops_count; i++) {
 978                 if (WARN_ON(orig_nat_ops[i].pf != ops->pf))
 979                         return -EINVAL;
 980                 if (orig_nat_ops[i].hooknum == hooknum) {
 981                         hooknum = i;
 982                         break;
 983                 }
 984         }
 985
 986         if (WARN_ON_ONCE(i == ops_count))
 987                 return -EINVAL;
 988
 989         mutex_lock(&nf_nat_proto_mutex);
 990         if (!nat_proto_net->nat_hook_ops) {
 991                 WARN_ON(nat_proto_net->users != 0);
 992
 993                 nat_ops = kmemdup(orig_nat_ops, sizeof(*orig_nat_ops) * ops_count, GFP_KERNEL);
 994                 if (!nat_ops) {
 995                         mutex_unlock(&nf_nat_proto_mutex);
 996                         return -ENOMEM;
 997                 }
 998
 999                 for (i = 0; i < ops_count; i++) {
1000                         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1001                         if (priv) {
1002                                 nat_ops[i].priv = priv;
1003                                 continue;
1004                         }
1005                         mutex_unlock(&nf_nat_proto_mutex);
1006                         while (i)
1007                                 kfree(nat_ops[--i].priv);
1008                         kfree(nat_ops);
1009                         return -ENOMEM;
1010                 }
1011
1012                 ret = nf_register_net_hooks(net, nat_ops, ops_count);
1013                 if (ret < 0) {
1014                         mutex_unlock(&nf_nat_proto_mutex);
1015                         for (i = 0; i < ops_count; i++)
1016                                 kfree(nat_ops[i].priv);
1017                         kfree(nat_ops);
1018                         return ret;
1019                 }
1020
1021                 nat_proto_net->nat_hook_ops = nat_ops;
1022         }
1023
1024         nat_ops = nat_proto_net->nat_hook_ops;
1025         priv = nat_ops[hooknum].priv;
1026         if (WARN_ON_ONCE(!priv)) {
1027                 mutex_unlock(&nf_nat_proto_mutex);
1028                 return -EOPNOTSUPP;
1029         }
1030
1031         ret = nf_hook_entries_insert_raw(&priv->entries, ops);
1032         if (ret == 0)
1033                 nat_proto_net->users++;
1034
1035         mutex_unlock(&nf_nat_proto_mutex);
1036         return ret;
1037 }
1038 EXPORT_SYMBOL_GPL(nf_nat_register_fn);
1039
1040 void nf_nat_unregister_fn(struct net *net, const struct nf_hook_ops *ops,
1041                           unsigned int ops_count)
1042 {
1043         struct nat_net *nat_net = net_generic(net, nat_net_id);
1044         struct nf_nat_hooks_net *nat_proto_net;
1045         struct nf_nat_lookup_hook_priv *priv;
1046         struct nf_hook_ops *nat_ops;
1047         int hooknum = ops->hooknum;
1048         int i;
1049
1050         if (ops->pf >= ARRAY_SIZE(nat_net->nat_proto_net))
1051                 return;
1052
1053         nat_proto_net = &nat_net->nat_proto_net[ops->pf];
1054
1055         mutex_lock(&nf_nat_proto_mutex);
1056         if (WARN_ON(nat_proto_net->users == 0))
1057                 goto unlock;
1058
1059         nat_proto_net->users--;
1060
1061         nat_ops = nat_proto_net->nat_hook_ops;
1062         for (i = 0; i < ops_count; i++) {
1063                 if (nat_ops[i].hooknum == hooknum) {
1064                         hooknum = i;
1065                         break;
1066                 }
1067         }
1068         if (WARN_ON_ONCE(i == ops_count))
1069                 goto unlock;
1070         priv = nat_ops[hooknum].priv;
1071         nf_hook_entries_delete_raw(&priv->entries, ops);
1072
1073         if (nat_proto_net->users == 0) {
1074                 nf_unregister_net_hooks(net, nat_ops, ops_count);
1075
1076                 for (i = 0; i < ops_count; i++) {
1077                         priv = nat_ops[i].priv;
1078                         kfree_rcu(priv, rcu_head);
1079                 }
1080
1081                 nat_proto_net->nat_hook_ops = NULL;
1082                 kfree(nat_ops);
1083         }
1084 unlock:
1085         mutex_unlock(&nf_nat_proto_mutex);
1086 }
1087 EXPORT_SYMBOL_GPL(nf_nat_unregister_fn);
1088
1089 static struct pernet_operations nat_net_ops = {
1090         .id = &nat_net_id,
1091         .size = sizeof(struct nat_net),
1092 };
1093
1094 static struct nf_nat_hook nat_hook = {
1095         .parse_nat_setup        = nfnetlink_parse_nat_setup,
1096 #ifdef CONFIG_XFRM
1097         .decode_session         = __nf_nat_decode_session,
1098 #endif
1099         .manip_pkt              = nf_nat_manip_pkt,
1100 };
1101
1102 static int __init nf_nat_init(void)
1103 {
1104         int ret, i;
1105
1106         /* Leave them the same for the moment. */
1107         nf_nat_htable_size = nf_conntrack_htable_size;
1108         if (nf_nat_htable_size < CONNTRACK_LOCKS)
1109                 nf_nat_htable_size = CONNTRACK_LOCKS;
1110
1111         nf_nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, 0);
1112         if (!nf_nat_bysource)
1113                 return -ENOMEM;
1114
1115         ret = nf_ct_extend_register(&nat_extend);
1116         if (ret < 0) {
1117                 kvfree(nf_nat_bysource);
1118                 pr_err("Unable to register extension\n");
1119                 return ret;
1120         }
1121
1122         for (i = 0; i < CONNTRACK_LOCKS; i++)
1123                 spin_lock_init(&nf_nat_locks[i]);
1124
1125         ret = register_pernet_subsys(&nat_net_ops);
1126         if (ret < 0) {
1127                 nf_ct_extend_unregister(&nat_extend);
1128                 return ret;
1129         }
1130
1131         nf_ct_helper_expectfn_register(&follow_master_nat);
1132
1133         WARN_ON(nf_nat_hook != NULL);
1134         RCU_INIT_POINTER(nf_nat_hook, &nat_hook);
1135
1136         return 0;
1137 }
1138
1139 static void __exit nf_nat_cleanup(void)
1140 {
1141         struct nf_nat_proto_clean clean = {};
1142
1143         nf_ct_iterate_destroy(nf_nat_proto_clean, &clean);
1144
1145         nf_ct_extend_unregister(&nat_extend);
1146         nf_ct_helper_expectfn_unregister(&follow_master_nat);
1147         RCU_INIT_POINTER(nf_nat_hook, NULL);
1148
1149         synchronize_net();
1150         kvfree(nf_nat_bysource);
1151         unregister_pernet_subsys(&nat_net_ops);
1152 }
1153
1154 MODULE_LICENSE("GPL");
1155
1156 module_init(nf_nat_init);
1157 module_exit(nf_nat_cleanup);