net/netfilter/nf_nat_core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * (C) 1999-2001 Paul `Rusty' Russell
   4  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   5  * (C) 2011 Patrick McHardy <kaber@trash.net>
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/module.h>
  11 #include <linux/types.h>
  12 #include <linux/timer.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/gfp.h>
  15 #include <net/xfrm.h>
  16 #include <linux/siphash.h>
  17 #include <linux/rtnetlink.h>
  18
  19 #include <net/netfilter/nf_conntrack.h>
  20 #include <net/netfilter/nf_conntrack_core.h>
  21 #include <net/netfilter/nf_conntrack_helper.h>
  22 #include <net/netfilter/nf_conntrack_seqadj.h>
  23 #include <net/netfilter/nf_conntrack_zones.h>
  24 #include <net/netfilter/nf_nat.h>
  25 #include <net/netfilter/nf_nat_helper.h>
  26 #include <uapi/linux/netfilter/nf_nat.h>
  27
  28 #include "nf_internals.h"
  29
  30 static spinlock_t nf_nat_locks[CONNTRACK_LOCKS];
  31
  32 static DEFINE_MUTEX(nf_nat_proto_mutex);
  33 static unsigned int nat_net_id __read_mostly;
  34
  35 static struct hlist_head *nf_nat_bysource __read_mostly;
  36 static unsigned int nf_nat_htable_size __read_mostly;
  37 static siphash_key_t nf_nat_hash_rnd __read_mostly;
  38
  39 struct nf_nat_lookup_hook_priv {
  40         struct nf_hook_entries __rcu *entries;
  41
  42         struct rcu_head rcu_head;
  43 };
  44
  45 struct nf_nat_hooks_net {
  46         struct nf_hook_ops *nat_hook_ops;
  47         unsigned int users;
  48 };
  49
  50 struct nat_net {
  51         struct nf_nat_hooks_net nat_proto_net[NFPROTO_NUMPROTO];
  52 };
  53
  54 #ifdef CONFIG_XFRM
  55 static void nf_nat_ipv4_decode_session(struct sk_buff *skb,
  56                                        const struct nf_conn *ct,
  57                                        enum ip_conntrack_dir dir,
  58                                        unsigned long statusbit,
  59                                        struct flowi *fl)
  60 {
  61         const struct nf_conntrack_tuple *t = &ct->tuplehash[dir].tuple;
  62         struct flowi4 *fl4 = &fl->u.ip4;
  63
  64         if (ct->status & statusbit) {
  65                 fl4->daddr = t->dst.u3.ip;
  66                 if (t->dst.protonum == IPPROTO_TCP ||
  67                     t->dst.protonum == IPPROTO_UDP ||
  68                     t->dst.protonum == IPPROTO_UDPLITE ||
  69                     t->dst.protonum == IPPROTO_DCCP ||
  70                     t->dst.protonum == IPPROTO_SCTP)
  71                         fl4->fl4_dport = t->dst.u.all;
  72         }
  73
  74         statusbit ^= IPS_NAT_MASK;
  75
  76         if (ct->status & statusbit) {
  77                 fl4->saddr = t->src.u3.ip;
  78                 if (t->dst.protonum == IPPROTO_TCP ||
  79                     t->dst.protonum == IPPROTO_UDP ||
  80                     t->dst.protonum == IPPROTO_UDPLITE ||
  81                     t->dst.protonum == IPPROTO_DCCP ||
  82                     t->dst.protonum == IPPROTO_SCTP)
  83                         fl4->fl4_sport = t->src.u.all;
  84         }
  85 }
  86
  87 static void nf_nat_ipv6_decode_session(struct sk_buff *skb,
  88                                        const struct nf_conn *ct,
  89                                        enum ip_conntrack_dir dir,
  90                                        unsigned long statusbit,
  91                                        struct flowi *fl)
  92 {
  93 #if IS_ENABLED(CONFIG_IPV6)
  94         const struct nf_conntrack_tuple *t = &ct->tuplehash[dir].tuple;
  95         struct flowi6 *fl6 = &fl->u.ip6;
  96
  97         if (ct->status & statusbit) {
  98                 fl6->daddr = t->dst.u3.in6;
  99                 if (t->dst.protonum == IPPROTO_TCP ||
 100                     t->dst.protonum == IPPROTO_UDP ||
 101                     t->dst.protonum == IPPROTO_UDPLITE ||
 102                     t->dst.protonum == IPPROTO_DCCP ||
 103                     t->dst.protonum == IPPROTO_SCTP)
 104                         fl6->fl6_dport = t->dst.u.all;
 105         }
 106
 107         statusbit ^= IPS_NAT_MASK;
 108
 109         if (ct->status & statusbit) {
 110                 fl6->saddr = t->src.u3.in6;
 111                 if (t->dst.protonum == IPPROTO_TCP ||
 112                     t->dst.protonum == IPPROTO_UDP ||
 113                     t->dst.protonum == IPPROTO_UDPLITE ||
 114                     t->dst.protonum == IPPROTO_DCCP ||
 115                     t->dst.protonum == IPPROTO_SCTP)
 116                         fl6->fl6_sport = t->src.u.all;
 117         }
 118 #endif
 119 }
 120
 121 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
 122 {
 123         const struct nf_conn *ct;
 124         enum ip_conntrack_info ctinfo;
 125         enum ip_conntrack_dir dir;
 126         unsigned  long statusbit;
 127         u8 family;
 128
 129         ct = nf_ct_get(skb, &ctinfo);
 130         if (ct == NULL)
 131                 return;
 132
 133         family = nf_ct_l3num(ct);
 134         dir = CTINFO2DIR(ctinfo);
 135         if (dir == IP_CT_DIR_ORIGINAL)
 136                 statusbit = IPS_DST_NAT;
 137         else
 138                 statusbit = IPS_SRC_NAT;
 139
 140         switch (family) {
 141         case NFPROTO_IPV4:
 142                 nf_nat_ipv4_decode_session(skb, ct, dir, statusbit, fl);
 143                 return;
 144         case NFPROTO_IPV6:
 145                 nf_nat_ipv6_decode_session(skb, ct, dir, statusbit, fl);
 146                 return;
 147         }
 148 }
 149 #endif /* CONFIG_XFRM */
 150
 151 /* We keep an extra hash for each conntrack, for fast searching. */
 152 static unsigned int
 153 hash_by_src(const struct net *n, const struct nf_conntrack_tuple *tuple)
 154 {
 155         unsigned int hash;
 156         struct {
 157                 struct nf_conntrack_man src;
 158                 u32 net_mix;
 159                 u32 protonum;
 160         } __aligned(SIPHASH_ALIGNMENT) combined;
 161
 162         get_random_once(&nf_nat_hash_rnd, sizeof(nf_nat_hash_rnd));
 163
 164         memset(&combined, 0, sizeof(combined));
 165
 166         /* Original src, to ensure we map it consistently if poss. */
 167         combined.src = tuple->src;
 168         combined.net_mix = net_hash_mix(n);
 169         combined.protonum = tuple->dst.protonum;
 170
 171         hash = siphash(&combined, sizeof(combined), &nf_nat_hash_rnd);
 172
 173         return reciprocal_scale(hash, nf_nat_htable_size);
 174 }
 175
 176 /* Is this tuple already taken? (not by us) */
 177 static int
 178 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 179                   const struct nf_conn *ignored_conntrack)
 180 {
 181         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 182          * incoming ones.  NAT means they don't have a fixed mapping,
 183          * so we invert the tuple and look for the incoming reply.
 184          *
 185          * We could keep a separate hash if this proves too slow.
 186          */
 187         struct nf_conntrack_tuple reply;
 188
 189         nf_ct_invert_tuple(&reply, tuple);
 190         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 191 }
 192
 193 static bool nf_nat_inet_in_range(const struct nf_conntrack_tuple *t,
 194                                  const struct nf_nat_range2 *range)
 195 {
 196         if (t->src.l3num == NFPROTO_IPV4)
 197                 return ntohl(t->src.u3.ip) >= ntohl(range->min_addr.ip) &&
 198                        ntohl(t->src.u3.ip) <= ntohl(range->max_addr.ip);
 199
 200         return ipv6_addr_cmp(&t->src.u3.in6, &range->min_addr.in6) >= 0 &&
 201                ipv6_addr_cmp(&t->src.u3.in6, &range->max_addr.in6) <= 0;
 202 }
 203
 204 /* Is the manipable part of the tuple between min and max incl? */
 205 static bool l4proto_in_range(const struct nf_conntrack_tuple *tuple,
 206                              enum nf_nat_manip_type maniptype,
 207                              const union nf_conntrack_man_proto *min,
 208                              const union nf_conntrack_man_proto *max)
 209 {
 210         __be16 port;
 211
 212         switch (tuple->dst.protonum) {
 213         case IPPROTO_ICMP:
 214         case IPPROTO_ICMPV6:
 215                 return ntohs(tuple->src.u.icmp.id) >= ntohs(min->icmp.id) &&
 216                        ntohs(tuple->src.u.icmp.id) <= ntohs(max->icmp.id);
 217         case IPPROTO_GRE: /* all fall though */
 218         case IPPROTO_TCP:
 219         case IPPROTO_UDP:
 220         case IPPROTO_UDPLITE:
 221         case IPPROTO_DCCP:
 222         case IPPROTO_SCTP:
 223                 if (maniptype == NF_NAT_MANIP_SRC)
 224                         port = tuple->src.u.all;
 225                 else
 226                         port = tuple->dst.u.all;
 227
 228                 return ntohs(port) >= ntohs(min->all) &&
 229                        ntohs(port) <= ntohs(max->all);
 230         default:
 231                 return true;
 232         }
 233 }
 234
 235 /* If we source map this tuple so reply looks like reply_tuple, will
 236  * that meet the constraints of range.
 237  */
 238 static int in_range(const struct nf_conntrack_tuple *tuple,
 239                     const struct nf_nat_range2 *range)
 240 {
 241         /* If we are supposed to map IPs, then we must be in the
 242          * range specified, otherwise let this drag us onto a new src IP.
 243          */
 244         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 245             !nf_nat_inet_in_range(tuple, range))
 246                 return 0;
 247
 248         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED))
 249                 return 1;
 250
 251         return l4proto_in_range(tuple, NF_NAT_MANIP_SRC,
 252                                 &range->min_proto, &range->max_proto);
 253 }
 254
 255 static inline int
 256 same_src(const struct nf_conn *ct,
 257          const struct nf_conntrack_tuple *tuple)
 258 {
 259         const struct nf_conntrack_tuple *t;
 260
 261         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 262         return (t->dst.protonum == tuple->dst.protonum &&
 263                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 264                 t->src.u.all == tuple->src.u.all);
 265 }
 266
 267 /* Only called for SRC manip */
 268 static int
 269 find_appropriate_src(struct net *net,
 270                      const struct nf_conntrack_zone *zone,
 271                      const struct nf_conntrack_tuple *tuple,
 272                      struct nf_conntrack_tuple *result,
 273                      const struct nf_nat_range2 *range)
 274 {
 275         unsigned int h = hash_by_src(net, tuple);
 276         const struct nf_conn *ct;
 277
 278         hlist_for_each_entry_rcu(ct, &nf_nat_bysource[h], nat_bysource) {
 279                 if (same_src(ct, tuple) &&
 280                     net_eq(net, nf_ct_net(ct)) &&
 281                     nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) {
 282                         /* Copy source part from reply tuple. */
 283                         nf_ct_invert_tuple(result,
 284                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 285                         result->dst = tuple->dst;
 286
 287                         if (in_range(result, range))
 288                                 return 1;
 289                 }
 290         }
 291         return 0;
 292 }
 293
 294 /* For [FUTURE] fragmentation handling, we want the least-used
 295  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 296  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 297  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 298  * the ip with the lowest src-ip/dst-ip/proto usage.
 299  */
 300 static void
 301 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 302                     struct nf_conntrack_tuple *tuple,
 303                     const struct nf_nat_range2 *range,
 304                     const struct nf_conn *ct,
 305                     enum nf_nat_manip_type maniptype)
 306 {
 307         union nf_inet_addr *var_ipp;
 308         unsigned int i, max;
 309         /* Host order */
 310         u32 minip, maxip, j, dist;
 311         bool full_range;
 312
 313         /* No IP mapping?  Do nothing. */
 314         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 315                 return;
 316
 317         if (maniptype == NF_NAT_MANIP_SRC)
 318                 var_ipp = &tuple->src.u3;
 319         else
 320                 var_ipp = &tuple->dst.u3;
 321
 322         /* Fast path: only one choice. */
 323         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 324                 *var_ipp = range->min_addr;
 325                 return;
 326         }
 327
 328         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 329                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 330         else
 331                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 332
 333         /* Hashing source and destination IPs gives a fairly even
 334          * spread in practice (if there are a small number of IPs
 335          * involved, there usually aren't that many connections
 336          * anyway).  The consistency means that servers see the same
 337          * client coming from the same IP (some Internet Banking sites
 338          * like this), even across reboots.
 339          */
 340         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 341                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 342                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 343
 344         full_range = false;
 345         for (i = 0; i <= max; i++) {
 346                 /* If first bytes of the address are at the maximum, use the
 347                  * distance. Otherwise use the full range.
 348                  */
 349                 if (!full_range) {
 350                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 351                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 352                         dist  = maxip - minip + 1;
 353                 } else {
 354                         minip = 0;
 355                         dist  = ~0;
 356                 }
 357
 358                 var_ipp->all[i] = (__force __u32)
 359                         htonl(minip + reciprocal_scale(j, dist));
 360                 if (var_ipp->all[i] != range->max_addr.all[i])
 361                         full_range = true;
 362
 363                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 364                         j ^= (__force u32)tuple->dst.u3.all[i];
 365         }
 366 }
 367
 368 /* Alter the per-proto part of the tuple (depending on maniptype), to
 369  * give a unique tuple in the given range if possible.
 370  *
 371  * Per-protocol part of tuple is initialized to the incoming packet.
 372  */
 373 static void nf_nat_l4proto_unique_tuple(struct nf_conntrack_tuple *tuple,
 374                                         const struct nf_nat_range2 *range,
 375                                         enum nf_nat_manip_type maniptype,
 376                                         const struct nf_conn *ct)
 377 {
 378         unsigned int range_size, min, max, i, attempts;
 379         __be16 *keyptr;
 380         u16 off;
 381         static const unsigned int max_attempts = 128;
 382
 383         switch (tuple->dst.protonum) {
 384         case IPPROTO_ICMP:
 385         case IPPROTO_ICMPV6:
 386                 /* id is same for either direction... */
 387                 keyptr = &tuple->src.u.icmp.id;
 388                 if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED)) {
 389                         min = 0;
 390                         range_size = 65536;
 391                 } else {
 392                         min = ntohs(range->min_proto.icmp.id);
 393                         range_size = ntohs(range->max_proto.icmp.id) -
 394                                      ntohs(range->min_proto.icmp.id) + 1;
 395                 }
 396                 goto find_free_id;
 397 #if IS_ENABLED(CONFIG_NF_CT_PROTO_GRE)
 398         case IPPROTO_GRE:
 399                 /* If there is no master conntrack we are not PPTP,
 400                    do not change tuples */
 401                 if (!ct->master)
 402                         return;
 403
 404                 if (maniptype == NF_NAT_MANIP_SRC)
 405                         keyptr = &tuple->src.u.gre.key;
 406                 else
 407                         keyptr = &tuple->dst.u.gre.key;
 408
 409                 if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED)) {
 410                         min = 1;
 411                         range_size = 65535;
 412                 } else {
 413                         min = ntohs(range->min_proto.gre.key);
 414                         range_size = ntohs(range->max_proto.gre.key) - min + 1;
 415                 }
 416                 goto find_free_id;
 417 #endif
 418         case IPPROTO_UDP:
 419         case IPPROTO_UDPLITE:
 420         case IPPROTO_TCP:
 421         case IPPROTO_SCTP:
 422         case IPPROTO_DCCP:
 423                 if (maniptype == NF_NAT_MANIP_SRC)
 424                         keyptr = &tuple->src.u.all;
 425                 else
 426                         keyptr = &tuple->dst.u.all;
 427
 428                 break;
 429         default:
 430                 return;
 431         }
 432
 433         /* If no range specified... */
 434         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED)) {
 435                 /* If it's dst rewrite, can't change port */
 436                 if (maniptype == NF_NAT_MANIP_DST)
 437                         return;
 438
 439                 if (ntohs(*keyptr) < 1024) {
 440                         /* Loose convention: >> 512 is credential passing */
 441                         if (ntohs(*keyptr) < 512) {
 442                                 min = 1;
 443                                 range_size = 511 - min + 1;
 444                         } else {
 445                                 min = 600;
 446                                 range_size = 1023 - min + 1;
 447                         }
 448                 } else {
 449                         min = 1024;
 450                         range_size = 65535 - 1024 + 1;
 451                 }
 452         } else {
 453                 min = ntohs(range->min_proto.all);
 454                 max = ntohs(range->max_proto.all);
 455                 if (unlikely(max < min))
 456                         swap(max, min);
 457                 range_size = max - min + 1;
 458         }
 459
 460 find_free_id:
 461         if (range->flags & NF_NAT_RANGE_PROTO_OFFSET)
 462                 off = (ntohs(*keyptr) - ntohs(range->base_proto.all));
 463         else
 464                 off = prandom_u32();
 465
 466         attempts = range_size;
 467         if (attempts > max_attempts)
 468                 attempts = max_attempts;
 469
 470         /* We are in softirq; doing a search of the entire range risks
 471          * soft lockup when all tuples are already used.
 472          *
 473          * If we can't find any free port from first offset, pick a new
 474          * one and try again, with ever smaller search window.
 475          */
 476 another_round:
 477         for (i = 0; i < attempts; i++, off++) {
 478                 *keyptr = htons(min + off % range_size);
 479                 if (!nf_nat_used_tuple(tuple, ct))
 480                         return;
 481         }
 482
 483         if (attempts >= range_size || attempts < 16)
 484                 return;
 485         attempts /= 2;
 486         off = prandom_u32();
 487         goto another_round;
 488 }
 489
 490 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 491  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 492  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 493  * range. It might not be possible to get a unique tuple, but we try.
 494  * At worst (or if we race), we will end up with a final duplicate in
 495  * __nf_conntrack_confirm and drop the packet. */
 496 static void
 497 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 498                  const struct nf_conntrack_tuple *orig_tuple,
 499                  const struct nf_nat_range2 *range,
 500                  struct nf_conn *ct,
 501                  enum nf_nat_manip_type maniptype)
 502 {
 503         const struct nf_conntrack_zone *zone;
 504         struct net *net = nf_ct_net(ct);
 505
 506         zone = nf_ct_zone(ct);
 507
 508         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 509          * and that same mapping gives a unique tuple within the given
 510          * range, use that.
 511          *
 512          * This is only required for source (ie. NAT/masq) mappings.
 513          * So far, we don't do local source mappings, so multiple
 514          * manips not an issue.
 515          */
 516         if (maniptype == NF_NAT_MANIP_SRC &&
 517             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 518                 /* try the original tuple first */
 519                 if (in_range(orig_tuple, range)) {
 520                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 521                                 *tuple = *orig_tuple;
 522                                 return;
 523                         }
 524                 } else if (find_appropriate_src(net, zone,
 525                                                 orig_tuple, tuple, range)) {
 526                         pr_debug("get_unique_tuple: Found current src map\n");
 527                         if (!nf_nat_used_tuple(tuple, ct))
 528                                 return;
 529                 }
 530         }
 531
 532         /* 2) Select the least-used IP/proto combination in the given range */
 533         *tuple = *orig_tuple;
 534         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 535
 536         /* 3) The per-protocol part of the manip is made to map into
 537          * the range to make a unique tuple.
 538          */
 539
 540         /* Only bother mapping if it's not already in range and unique */
 541         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 542                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 543                         if (!(range->flags & NF_NAT_RANGE_PROTO_OFFSET) &&
 544                             l4proto_in_range(tuple, maniptype,
 545                                   &range->min_proto,
 546                                   &range->max_proto) &&
 547                             (range->min_proto.all == range->max_proto.all ||
 548                              !nf_nat_used_tuple(tuple, ct)))
 549                                 return;
 550                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 551                         return;
 552                 }
 553         }
 554
 555         /* Last chance: get protocol to try to obtain unique tuple. */
 556         nf_nat_l4proto_unique_tuple(tuple, range, maniptype, ct);
 557 }
 558
 559 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 560 {
 561         struct nf_conn_nat *nat = nfct_nat(ct);
 562         if (nat)
 563                 return nat;
 564
 565         if (!nf_ct_is_confirmed(ct))
 566                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 567
 568         return nat;
 569 }
 570 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 571
 572 unsigned int
 573 nf_nat_setup_info(struct nf_conn *ct,
 574                   const struct nf_nat_range2 *range,
 575                   enum nf_nat_manip_type maniptype)
 576 {
 577         struct net *net = nf_ct_net(ct);
 578         struct nf_conntrack_tuple curr_tuple, new_tuple;
 579
 580         /* Can't setup nat info for confirmed ct. */
 581         if (nf_ct_is_confirmed(ct))
 582                 return NF_ACCEPT;
 583
 584         WARN_ON(maniptype != NF_NAT_MANIP_SRC &&
 585                 maniptype != NF_NAT_MANIP_DST);
 586
 587         if (WARN_ON(nf_nat_initialized(ct, maniptype)))
 588                 return NF_DROP;
 589
 590         /* What we've got will look like inverse of reply. Normally
 591          * this is what is in the conntrack, except for prior
 592          * manipulations (future optimization: if num_manips == 0,
 593          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 594          */
 595         nf_ct_invert_tuple(&curr_tuple,
 596                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 597
 598         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 599
 600         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 601                 struct nf_conntrack_tuple reply;
 602
 603                 /* Alter conntrack table so will recognize replies. */
 604                 nf_ct_invert_tuple(&reply, &new_tuple);
 605                 nf_conntrack_alter_reply(ct, &reply);
 606
 607                 /* Non-atomic: we own this at the moment. */
 608                 if (maniptype == NF_NAT_MANIP_SRC)
 609                         ct->status |= IPS_SRC_NAT;
 610                 else
 611                         ct->status |= IPS_DST_NAT;
 612
 613                 if (nfct_help(ct) && !nfct_seqadj(ct))
 614                         if (!nfct_seqadj_ext_add(ct))
 615                                 return NF_DROP;
 616         }
 617
 618         if (maniptype == NF_NAT_MANIP_SRC) {
 619                 unsigned int srchash;
 620                 spinlock_t *lock;
 621
 622                 srchash = hash_by_src(net,
 623                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 624                 lock = &nf_nat_locks[srchash % CONNTRACK_LOCKS];
 625                 spin_lock_bh(lock);
 626                 hlist_add_head_rcu(&ct->nat_bysource,
 627                                    &nf_nat_bysource[srchash]);
 628                 spin_unlock_bh(lock);
 629         }
 630
 631         /* It's done. */
 632         if (maniptype == NF_NAT_MANIP_DST)
 633                 ct->status |= IPS_DST_NAT_DONE;
 634         else
 635                 ct->status |= IPS_SRC_NAT_DONE;
 636
 637         return NF_ACCEPT;
 638 }
 639 EXPORT_SYMBOL(nf_nat_setup_info);
 640
 641 static unsigned int
 642 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 643 {
 644         /* Force range to this IP; let proto decide mapping for
 645          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 646          * Use reply in case it's already been mangled (eg local packet).
 647          */
 648         union nf_inet_addr ip =
 649                 (manip == NF_NAT_MANIP_SRC ?
 650                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 651                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 652         struct nf_nat_range2 range = {
 653                 .flags          = NF_NAT_RANGE_MAP_IPS,
 654                 .min_addr       = ip,
 655                 .max_addr       = ip,
 656         };
 657         return nf_nat_setup_info(ct, &range, manip);
 658 }
 659
 660 unsigned int
 661 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 662 {
 663         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 664 }
 665 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 666
 667 /* Do packet manipulations according to nf_nat_setup_info. */
 668 unsigned int nf_nat_packet(struct nf_conn *ct,
 669                            enum ip_conntrack_info ctinfo,
 670                            unsigned int hooknum,
 671                            struct sk_buff *skb)
 672 {
 673         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 674         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 675         unsigned int verdict = NF_ACCEPT;
 676         unsigned long statusbit;
 677
 678         if (mtype == NF_NAT_MANIP_SRC)
 679                 statusbit = IPS_SRC_NAT;
 680         else
 681                 statusbit = IPS_DST_NAT;
 682
 683         /* Invert if this is reply dir. */
 684         if (dir == IP_CT_DIR_REPLY)
 685                 statusbit ^= IPS_NAT_MASK;
 686
 687         /* Non-atomic: these bits don't change. */
 688         if (ct->status & statusbit)
 689                 verdict = nf_nat_manip_pkt(skb, ct, mtype, dir);
 690
 691         return verdict;
 692 }
 693 EXPORT_SYMBOL_GPL(nf_nat_packet);
 694
 695 unsigned int
 696 nf_nat_inet_fn(void *priv, struct sk_buff *skb,
 697                const struct nf_hook_state *state)
 698 {
 699         struct nf_conn *ct;
 700         enum ip_conntrack_info ctinfo;
 701         struct nf_conn_nat *nat;
 702         /* maniptype == SRC for postrouting. */
 703         enum nf_nat_manip_type maniptype = HOOK2MANIP(state->hook);
 704
 705         ct = nf_ct_get(skb, &ctinfo);
 706         /* Can't track?  It's not due to stress, or conntrack would
 707          * have dropped it.  Hence it's the user's responsibilty to
 708          * packet filter it out, or implement conntrack/NAT for that
 709          * protocol. 8) --RR
 710          */
 711         if (!ct)
 712                 return NF_ACCEPT;
 713
 714         nat = nfct_nat(ct);
 715
 716         switch (ctinfo) {
 717         case IP_CT_RELATED:
 718         case IP_CT_RELATED_REPLY:
 719                 /* Only ICMPs can be IP_CT_IS_REPLY.  Fallthrough */
 720         case IP_CT_NEW:
 721                 /* Seen it before?  This can happen for loopback, retrans,
 722                  * or local packets.
 723                  */
 724                 if (!nf_nat_initialized(ct, maniptype)) {
 725                         struct nf_nat_lookup_hook_priv *lpriv = priv;
 726                         struct nf_hook_entries *e = rcu_dereference(lpriv->entries);
 727                         unsigned int ret;
 728                         int i;
 729
 730                         if (!e)
 731                                 goto null_bind;
 732
 733                         for (i = 0; i < e->num_hook_entries; i++) {
 734                                 ret = e->hooks[i].hook(e->hooks[i].priv, skb,
 735                                                        state);
 736                                 if (ret != NF_ACCEPT)
 737                                         return ret;
 738                                 if (nf_nat_initialized(ct, maniptype))
 739                                         goto do_nat;
 740                         }
 741 null_bind:
 742                         ret = nf_nat_alloc_null_binding(ct, state->hook);
 743                         if (ret != NF_ACCEPT)
 744                                 return ret;
 745                 } else {
 746                         pr_debug("Already setup manip %s for ct %p (status bits 0x%lx)\n",
 747                                  maniptype == NF_NAT_MANIP_SRC ? "SRC" : "DST",
 748                                  ct, ct->status);
 749                         if (nf_nat_oif_changed(state->hook, ctinfo, nat,
 750                                                state->out))
 751                                 goto oif_changed;
 752                 }
 753                 break;
 754         default:
 755                 /* ESTABLISHED */
 756                 WARN_ON(ctinfo != IP_CT_ESTABLISHED &&
 757                         ctinfo != IP_CT_ESTABLISHED_REPLY);
 758                 if (nf_nat_oif_changed(state->hook, ctinfo, nat, state->out))
 759                         goto oif_changed;
 760         }
 761 do_nat:
 762         return nf_nat_packet(ct, ctinfo, state->hook, skb);
 763
 764 oif_changed:
 765         nf_ct_kill_acct(ct, ctinfo, skb);
 766         return NF_DROP;
 767 }
 768 EXPORT_SYMBOL_GPL(nf_nat_inet_fn);
 769
 770 struct nf_nat_proto_clean {
 771         u8      l3proto;
 772         u8      l4proto;
 773 };
 774
 775 /* kill conntracks with affected NAT section */
 776 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 777 {
 778         const struct nf_nat_proto_clean *clean = data;
 779
 780         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 781             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 782                 return 0;
 783
 784         return i->status & IPS_NAT_MASK ? 1 : 0;
 785 }
 786
 787 static void __nf_nat_cleanup_conntrack(struct nf_conn *ct)
 788 {
 789         unsigned int h;
 790
 791         h = hash_by_src(nf_ct_net(ct), &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 792         spin_lock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 793         hlist_del_rcu(&ct->nat_bysource);
 794         spin_unlock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 795 }
 796
 797 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 798 {
 799         if (nf_nat_proto_remove(ct, data))
 800                 return 1;
 801
 802         /* This module is being removed and conntrack has nat null binding.
 803          * Remove it from bysource hash, as the table will be freed soon.
 804          *
 805          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 806          * will delete entry from already-freed table.
 807          */
 808         if (test_and_clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status))
 809                 __nf_nat_cleanup_conntrack(ct);
 810
 811         /* don't delete conntrack.  Although that would make things a lot
 812          * simpler, we'd end up flushing all conntracks on nat rmmod.
 813          */
 814         return 0;
 815 }
 816
 817 /* No one using conntrack by the time this called. */
 818 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 819 {
 820         if (ct->status & IPS_SRC_NAT_DONE)
 821                 __nf_nat_cleanup_conntrack(ct);
 822 }
 823
 824 static struct nf_ct_ext_type nat_extend __read_mostly = {
 825         .len            = sizeof(struct nf_conn_nat),
 826         .align          = __alignof__(struct nf_conn_nat),
 827         .destroy        = nf_nat_cleanup_conntrack,
 828         .id             = NF_CT_EXT_NAT,
 829 };
 830
 831 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 832
 833 #include <linux/netfilter/nfnetlink.h>
 834 #include <linux/netfilter/nfnetlink_conntrack.h>
 835
 836 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 837         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 838         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 839 };
 840
 841 static int nf_nat_l4proto_nlattr_to_range(struct nlattr *tb[],
 842                                           struct nf_nat_range2 *range)
 843 {
 844         if (tb[CTA_PROTONAT_PORT_MIN]) {
 845                 range->min_proto.all = nla_get_be16(tb[CTA_PROTONAT_PORT_MIN]);
 846                 range->max_proto.all = range->min_proto.all;
 847                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
 848         }
 849         if (tb[CTA_PROTONAT_PORT_MAX]) {
 850                 range->max_proto.all = nla_get_be16(tb[CTA_PROTONAT_PORT_MAX]);
 851                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
 852         }
 853         return 0;
 854 }
 855
 856 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 857                                      const struct nf_conn *ct,
 858                                      struct nf_nat_range2 *range)
 859 {
 860         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 861         int err;
 862
 863         err = nla_parse_nested_deprecated(tb, CTA_PROTONAT_MAX, attr,
 864                                           protonat_nla_policy, NULL);
 865         if (err < 0)
 866                 return err;
 867
 868         return nf_nat_l4proto_nlattr_to_range(tb, range);
 869 }
 870
 871 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 872         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 873         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 874         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 875         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 876         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 877 };
 878
 879 static int nf_nat_ipv4_nlattr_to_range(struct nlattr *tb[],
 880                                        struct nf_nat_range2 *range)
 881 {
 882         if (tb[CTA_NAT_V4_MINIP]) {
 883                 range->min_addr.ip = nla_get_be32(tb[CTA_NAT_V4_MINIP]);
 884                 range->flags |= NF_NAT_RANGE_MAP_IPS;
 885         }
 886
 887         if (tb[CTA_NAT_V4_MAXIP])
 888                 range->max_addr.ip = nla_get_be32(tb[CTA_NAT_V4_MAXIP]);
 889         else
 890                 range->max_addr.ip = range->min_addr.ip;
 891
 892         return 0;
 893 }
 894
 895 static int nf_nat_ipv6_nlattr_to_range(struct nlattr *tb[],
 896                                        struct nf_nat_range2 *range)
 897 {
 898         if (tb[CTA_NAT_V6_MINIP]) {
 899                 nla_memcpy(&range->min_addr.ip6, tb[CTA_NAT_V6_MINIP],
 900                            sizeof(struct in6_addr));
 901                 range->flags |= NF_NAT_RANGE_MAP_IPS;
 902         }
 903
 904         if (tb[CTA_NAT_V6_MAXIP])
 905                 nla_memcpy(&range->max_addr.ip6, tb[CTA_NAT_V6_MAXIP],
 906                            sizeof(struct in6_addr));
 907         else
 908                 range->max_addr = range->min_addr;
 909
 910         return 0;
 911 }
 912
 913 static int
 914 nfnetlink_parse_nat(const struct nlattr *nat,
 915                     const struct nf_conn *ct, struct nf_nat_range2 *range)
 916 {
 917         struct nlattr *tb[CTA_NAT_MAX+1];
 918         int err;
 919
 920         memset(range, 0, sizeof(*range));
 921
 922         err = nla_parse_nested_deprecated(tb, CTA_NAT_MAX, nat,
 923                                           nat_nla_policy, NULL);
 924         if (err < 0)
 925                 return err;
 926
 927         switch (nf_ct_l3num(ct)) {
 928         case NFPROTO_IPV4:
 929                 err = nf_nat_ipv4_nlattr_to_range(tb, range);
 930                 break;
 931         case NFPROTO_IPV6:
 932                 err = nf_nat_ipv6_nlattr_to_range(tb, range);
 933                 break;
 934         default:
 935                 err = -EPROTONOSUPPORT;
 936                 break;
 937         }
 938
 939         if (err)
 940                 return err;
 941
 942         if (!tb[CTA_NAT_PROTO])
 943                 return 0;
 944
 945         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
 946 }
 947
 948 /* This function is called under rcu_read_lock() */
 949 static int
 950 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 951                           enum nf_nat_manip_type manip,
 952                           const struct nlattr *attr)
 953 {
 954         struct nf_nat_range2 range;
 955         int err;
 956
 957         /* Should not happen, restricted to creating new conntracks
 958          * via ctnetlink.
 959          */
 960         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
 961                 return -EEXIST;
 962
 963         /* No NAT information has been passed, allocate the null-binding */
 964         if (attr == NULL)
 965                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
 966
 967         err = nfnetlink_parse_nat(attr, ct, &range);
 968         if (err < 0)
 969                 return err;
 970
 971         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
 972 }
 973 #else
 974 static int
 975 nfnetlink_parse_nat_setup(struct nf_conn *ct,
 976                           enum nf_nat_manip_type manip,
 977                           const struct nlattr *attr)
 978 {
 979         return -EOPNOTSUPP;
 980 }
 981 #endif
 982
 983 static struct nf_ct_helper_expectfn follow_master_nat = {
 984         .name           = "nat-follow-master",
 985         .expectfn       = nf_nat_follow_master,
 986 };
 987
 988 int nf_nat_register_fn(struct net *net, u8 pf, const struct nf_hook_ops *ops,
 989                        const struct nf_hook_ops *orig_nat_ops, unsigned int ops_count)
 990 {
 991         struct nat_net *nat_net = net_generic(net, nat_net_id);
 992         struct nf_nat_hooks_net *nat_proto_net;
 993         struct nf_nat_lookup_hook_priv *priv;
 994         unsigned int hooknum = ops->hooknum;
 995         struct nf_hook_ops *nat_ops;
 996         int i, ret;
 997
 998         if (WARN_ON_ONCE(pf >= ARRAY_SIZE(nat_net->nat_proto_net)))
 999                 return -EINVAL;
1000
1001         nat_proto_net = &nat_net->nat_proto_net[pf];
1002
1003         for (i = 0; i < ops_count; i++) {
1004                 if (orig_nat_ops[i].hooknum == hooknum) {
1005                         hooknum = i;
1006                         break;
1007                 }
1008         }
1009
1010         if (WARN_ON_ONCE(i == ops_count))
1011                 return -EINVAL;
1012
1013         mutex_lock(&nf_nat_proto_mutex);
1014         if (!nat_proto_net->nat_hook_ops) {
1015                 WARN_ON(nat_proto_net->users != 0);
1016
1017                 nat_ops = kmemdup(orig_nat_ops, sizeof(*orig_nat_ops) * ops_count, GFP_KERNEL);
1018                 if (!nat_ops) {
1019                         mutex_unlock(&nf_nat_proto_mutex);
1020                         return -ENOMEM;
1021                 }
1022
1023                 for (i = 0; i < ops_count; i++) {
1024                         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1025                         if (priv) {
1026                                 nat_ops[i].priv = priv;
1027                                 continue;
1028                         }
1029                         mutex_unlock(&nf_nat_proto_mutex);
1030                         while (i)
1031                                 kfree(nat_ops[--i].priv);
1032                         kfree(nat_ops);
1033                         return -ENOMEM;
1034                 }
1035
1036                 ret = nf_register_net_hooks(net, nat_ops, ops_count);
1037                 if (ret < 0) {
1038                         mutex_unlock(&nf_nat_proto_mutex);
1039                         for (i = 0; i < ops_count; i++)
1040                                 kfree(nat_ops[i].priv);
1041                         kfree(nat_ops);
1042                         return ret;
1043                 }
1044
1045                 nat_proto_net->nat_hook_ops = nat_ops;
1046         }
1047
1048         nat_ops = nat_proto_net->nat_hook_ops;
1049         priv = nat_ops[hooknum].priv;
1050         if (WARN_ON_ONCE(!priv)) {
1051                 mutex_unlock(&nf_nat_proto_mutex);
1052                 return -EOPNOTSUPP;
1053         }
1054
1055         ret = nf_hook_entries_insert_raw(&priv->entries, ops);
1056         if (ret == 0)
1057                 nat_proto_net->users++;
1058
1059         mutex_unlock(&nf_nat_proto_mutex);
1060         return ret;
1061 }
1062
1063 void nf_nat_unregister_fn(struct net *net, u8 pf, const struct nf_hook_ops *ops,
1064                           unsigned int ops_count)
1065 {
1066         struct nat_net *nat_net = net_generic(net, nat_net_id);
1067         struct nf_nat_hooks_net *nat_proto_net;
1068         struct nf_nat_lookup_hook_priv *priv;
1069         struct nf_hook_ops *nat_ops;
1070         int hooknum = ops->hooknum;
1071         int i;
1072
1073         if (pf >= ARRAY_SIZE(nat_net->nat_proto_net))
1074                 return;
1075
1076         nat_proto_net = &nat_net->nat_proto_net[pf];
1077
1078         mutex_lock(&nf_nat_proto_mutex);
1079         if (WARN_ON(nat_proto_net->users == 0))
1080                 goto unlock;
1081
1082         nat_proto_net->users--;
1083
1084         nat_ops = nat_proto_net->nat_hook_ops;
1085         for (i = 0; i < ops_count; i++) {
1086                 if (nat_ops[i].hooknum == hooknum) {
1087                         hooknum = i;
1088                         break;
1089                 }
1090         }
1091         if (WARN_ON_ONCE(i == ops_count))
1092                 goto unlock;
1093         priv = nat_ops[hooknum].priv;
1094         nf_hook_entries_delete_raw(&priv->entries, ops);
1095
1096         if (nat_proto_net->users == 0) {
1097                 nf_unregister_net_hooks(net, nat_ops, ops_count);
1098
1099                 for (i = 0; i < ops_count; i++) {
1100                         priv = nat_ops[i].priv;
1101                         kfree_rcu(priv, rcu_head);
1102                 }
1103
1104                 nat_proto_net->nat_hook_ops = NULL;
1105                 kfree(nat_ops);
1106         }
1107 unlock:
1108         mutex_unlock(&nf_nat_proto_mutex);
1109 }
1110
1111 static struct pernet_operations nat_net_ops = {
1112         .id = &nat_net_id,
1113         .size = sizeof(struct nat_net),
1114 };
1115
1116 static struct nf_nat_hook nat_hook = {
1117         .parse_nat_setup        = nfnetlink_parse_nat_setup,
1118 #ifdef CONFIG_XFRM
1119         .decode_session         = __nf_nat_decode_session,
1120 #endif
1121         .manip_pkt              = nf_nat_manip_pkt,
1122 };
1123
1124 static int __init nf_nat_init(void)
1125 {
1126         int ret, i;
1127
1128         /* Leave them the same for the moment. */
1129         nf_nat_htable_size = nf_conntrack_htable_size;
1130         if (nf_nat_htable_size < CONNTRACK_LOCKS)
1131                 nf_nat_htable_size = CONNTRACK_LOCKS;
1132
1133         nf_nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, 0);
1134         if (!nf_nat_bysource)
1135                 return -ENOMEM;
1136
1137         ret = nf_ct_extend_register(&nat_extend);
1138         if (ret < 0) {
1139                 kvfree(nf_nat_bysource);
1140                 pr_err("Unable to register extension\n");
1141                 return ret;
1142         }
1143
1144         for (i = 0; i < CONNTRACK_LOCKS; i++)
1145                 spin_lock_init(&nf_nat_locks[i]);
1146
1147         ret = register_pernet_subsys(&nat_net_ops);
1148         if (ret < 0) {
1149                 nf_ct_extend_unregister(&nat_extend);
1150                 kvfree(nf_nat_bysource);
1151                 return ret;
1152         }
1153
1154         nf_ct_helper_expectfn_register(&follow_master_nat);
1155
1156         WARN_ON(nf_nat_hook != NULL);
1157         RCU_INIT_POINTER(nf_nat_hook, &nat_hook);
1158
1159         return 0;
1160 }
1161
1162 static void __exit nf_nat_cleanup(void)
1163 {
1164         struct nf_nat_proto_clean clean = {};
1165
1166         nf_ct_iterate_destroy(nf_nat_proto_clean, &clean);
1167
1168         nf_ct_extend_unregister(&nat_extend);
1169         nf_ct_helper_expectfn_unregister(&follow_master_nat);
1170         RCU_INIT_POINTER(nf_nat_hook, NULL);
1171
1172         synchronize_net();
1173         kvfree(nf_nat_bysource);
1174         unregister_pernet_subsys(&nat_net_ops);
1175 }
1176
1177 MODULE_LICENSE("GPL");
1178
1179 module_init(nf_nat_init);
1180 module_exit(nf_nat_cleanup);