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_bpf.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 #define NF_NAT_MAX_ATTEMPTS     128
  31 #define NF_NAT_HARDER_THRESH    (NF_NAT_MAX_ATTEMPTS / 4)
  32
  33 static spinlock_t nf_nat_locks[CONNTRACK_LOCKS];
  34
  35 static DEFINE_MUTEX(nf_nat_proto_mutex);
  36 static unsigned int nat_net_id __read_mostly;
  37
  38 static struct hlist_head *nf_nat_bysource __read_mostly;
  39 static unsigned int nf_nat_htable_size __read_mostly;
  40 static siphash_aligned_key_t nf_nat_hash_rnd;
  41
  42 struct nf_nat_lookup_hook_priv {
  43         struct nf_hook_entries __rcu *entries;
  44
  45         struct rcu_head rcu_head;
  46 };
  47
  48 struct nf_nat_hooks_net {
  49         struct nf_hook_ops *nat_hook_ops;
  50         unsigned int users;
  51 };
  52
  53 struct nat_net {
  54         struct nf_nat_hooks_net nat_proto_net[NFPROTO_NUMPROTO];
  55 };
  56
  57 #ifdef CONFIG_XFRM
  58 static void nf_nat_ipv4_decode_session(struct sk_buff *skb,
  59                                        const struct nf_conn *ct,
  60                                        enum ip_conntrack_dir dir,
  61                                        unsigned long statusbit,
  62                                        struct flowi *fl)
  63 {
  64         const struct nf_conntrack_tuple *t = &ct->tuplehash[dir].tuple;
  65         struct flowi4 *fl4 = &fl->u.ip4;
  66
  67         if (ct->status & statusbit) {
  68                 fl4->daddr = t->dst.u3.ip;
  69                 if (t->dst.protonum == IPPROTO_TCP ||
  70                     t->dst.protonum == IPPROTO_UDP ||
  71                     t->dst.protonum == IPPROTO_UDPLITE ||
  72                     t->dst.protonum == IPPROTO_DCCP ||
  73                     t->dst.protonum == IPPROTO_SCTP)
  74                         fl4->fl4_dport = t->dst.u.all;
  75         }
  76
  77         statusbit ^= IPS_NAT_MASK;
  78
  79         if (ct->status & statusbit) {
  80                 fl4->saddr = t->src.u3.ip;
  81                 if (t->dst.protonum == IPPROTO_TCP ||
  82                     t->dst.protonum == IPPROTO_UDP ||
  83                     t->dst.protonum == IPPROTO_UDPLITE ||
  84                     t->dst.protonum == IPPROTO_DCCP ||
  85                     t->dst.protonum == IPPROTO_SCTP)
  86                         fl4->fl4_sport = t->src.u.all;
  87         }
  88 }
  89
  90 static void nf_nat_ipv6_decode_session(struct sk_buff *skb,
  91                                        const struct nf_conn *ct,
  92                                        enum ip_conntrack_dir dir,
  93                                        unsigned long statusbit,
  94                                        struct flowi *fl)
  95 {
  96 #if IS_ENABLED(CONFIG_IPV6)
  97         const struct nf_conntrack_tuple *t = &ct->tuplehash[dir].tuple;
  98         struct flowi6 *fl6 = &fl->u.ip6;
  99
 100         if (ct->status & statusbit) {
 101                 fl6->daddr = t->dst.u3.in6;
 102                 if (t->dst.protonum == IPPROTO_TCP ||
 103                     t->dst.protonum == IPPROTO_UDP ||
 104                     t->dst.protonum == IPPROTO_UDPLITE ||
 105                     t->dst.protonum == IPPROTO_DCCP ||
 106                     t->dst.protonum == IPPROTO_SCTP)
 107                         fl6->fl6_dport = t->dst.u.all;
 108         }
 109
 110         statusbit ^= IPS_NAT_MASK;
 111
 112         if (ct->status & statusbit) {
 113                 fl6->saddr = t->src.u3.in6;
 114                 if (t->dst.protonum == IPPROTO_TCP ||
 115                     t->dst.protonum == IPPROTO_UDP ||
 116                     t->dst.protonum == IPPROTO_UDPLITE ||
 117                     t->dst.protonum == IPPROTO_DCCP ||
 118                     t->dst.protonum == IPPROTO_SCTP)
 119                         fl6->fl6_sport = t->src.u.all;
 120         }
 121 #endif
 122 }
 123
 124 static void __nf_nat_decode_session(struct sk_buff *skb, struct flowi *fl)
 125 {
 126         const struct nf_conn *ct;
 127         enum ip_conntrack_info ctinfo;
 128         enum ip_conntrack_dir dir;
 129         unsigned  long statusbit;
 130         u8 family;
 131
 132         ct = nf_ct_get(skb, &ctinfo);
 133         if (ct == NULL)
 134                 return;
 135
 136         family = nf_ct_l3num(ct);
 137         dir = CTINFO2DIR(ctinfo);
 138         if (dir == IP_CT_DIR_ORIGINAL)
 139                 statusbit = IPS_DST_NAT;
 140         else
 141                 statusbit = IPS_SRC_NAT;
 142
 143         switch (family) {
 144         case NFPROTO_IPV4:
 145                 nf_nat_ipv4_decode_session(skb, ct, dir, statusbit, fl);
 146                 return;
 147         case NFPROTO_IPV6:
 148                 nf_nat_ipv6_decode_session(skb, ct, dir, statusbit, fl);
 149                 return;
 150         }
 151 }
 152 #endif /* CONFIG_XFRM */
 153
 154 /* We keep an extra hash for each conntrack, for fast searching. */
 155 static unsigned int
 156 hash_by_src(const struct net *net,
 157             const struct nf_conntrack_zone *zone,
 158             const struct nf_conntrack_tuple *tuple)
 159 {
 160         unsigned int hash;
 161         struct {
 162                 struct nf_conntrack_man src;
 163                 u32 net_mix;
 164                 u32 protonum;
 165                 u32 zone;
 166         } __aligned(SIPHASH_ALIGNMENT) combined;
 167
 168         get_random_once(&nf_nat_hash_rnd, sizeof(nf_nat_hash_rnd));
 169
 170         memset(&combined, 0, sizeof(combined));
 171
 172         /* Original src, to ensure we map it consistently if poss. */
 173         combined.src = tuple->src;
 174         combined.net_mix = net_hash_mix(net);
 175         combined.protonum = tuple->dst.protonum;
 176
 177         /* Zone ID can be used provided its valid for both directions */
 178         if (zone->dir == NF_CT_DEFAULT_ZONE_DIR)
 179                 combined.zone = zone->id;
 180
 181         hash = siphash(&combined, sizeof(combined), &nf_nat_hash_rnd);
 182
 183         return reciprocal_scale(hash, nf_nat_htable_size);
 184 }
 185
 186 /* Is this tuple already taken? (not by us) */
 187 static int
 188 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 189                   const struct nf_conn *ignored_conntrack)
 190 {
 191         /* Conntrack tracking doesn't keep track of outgoing tuples; only
 192          * incoming ones.  NAT means they don't have a fixed mapping,
 193          * so we invert the tuple and look for the incoming reply.
 194          *
 195          * We could keep a separate hash if this proves too slow.
 196          */
 197         struct nf_conntrack_tuple reply;
 198
 199         nf_ct_invert_tuple(&reply, tuple);
 200         return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 201 }
 202
 203 static bool nf_nat_may_kill(struct nf_conn *ct, unsigned long flags)
 204 {
 205         static const unsigned long flags_refuse = IPS_FIXED_TIMEOUT |
 206                                                   IPS_DYING;
 207         static const unsigned long flags_needed = IPS_SRC_NAT;
 208         enum tcp_conntrack old_state;
 209
 210         old_state = READ_ONCE(ct->proto.tcp.state);
 211         if (old_state < TCP_CONNTRACK_TIME_WAIT)
 212                 return false;
 213
 214         if (flags & flags_refuse)
 215                 return false;
 216
 217         return (flags & flags_needed) == flags_needed;
 218 }
 219
 220 /* reverse direction will send packets to new source, so
 221  * make sure such packets are invalid.
 222  */
 223 static bool nf_seq_has_advanced(const struct nf_conn *old, const struct nf_conn *new)
 224 {
 225         return (__s32)(new->proto.tcp.seen[0].td_end -
 226                        old->proto.tcp.seen[0].td_end) > 0;
 227 }
 228
 229 static int
 230 nf_nat_used_tuple_harder(const struct nf_conntrack_tuple *tuple,
 231                          const struct nf_conn *ignored_conntrack,
 232                          unsigned int attempts_left)
 233 {
 234         static const unsigned long flags_offload = IPS_OFFLOAD | IPS_HW_OFFLOAD;
 235         struct nf_conntrack_tuple_hash *thash;
 236         const struct nf_conntrack_zone *zone;
 237         struct nf_conntrack_tuple reply;
 238         unsigned long flags;
 239         struct nf_conn *ct;
 240         bool taken = true;
 241         struct net *net;
 242
 243         nf_ct_invert_tuple(&reply, tuple);
 244
 245         if (attempts_left > NF_NAT_HARDER_THRESH ||
 246             tuple->dst.protonum != IPPROTO_TCP ||
 247             ignored_conntrack->proto.tcp.state != TCP_CONNTRACK_SYN_SENT)
 248                 return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 249
 250         /* :ast few attempts to find a free tcp port. Destructive
 251          * action: evict colliding if its in timewait state and the
 252          * tcp sequence number has advanced past the one used by the
 253          * old entry.
 254          */
 255         net = nf_ct_net(ignored_conntrack);
 256         zone = nf_ct_zone(ignored_conntrack);
 257
 258         thash = nf_conntrack_find_get(net, zone, &reply);
 259         if (!thash)
 260                 return false;
 261
 262         ct = nf_ct_tuplehash_to_ctrack(thash);
 263
 264         if (thash->tuple.dst.dir == IP_CT_DIR_ORIGINAL)
 265                 goto out;
 266
 267         if (WARN_ON_ONCE(ct == ignored_conntrack))
 268                 goto out;
 269
 270         flags = READ_ONCE(ct->status);
 271         if (!nf_nat_may_kill(ct, flags))
 272                 goto out;
 273
 274         if (!nf_seq_has_advanced(ct, ignored_conntrack))
 275                 goto out;
 276
 277         /* Even if we can evict do not reuse if entry is offloaded. */
 278         if (nf_ct_kill(ct))
 279                 taken = flags & flags_offload;
 280 out:
 281         nf_ct_put(ct);
 282         return taken;
 283 }
 284
 285 static bool nf_nat_inet_in_range(const struct nf_conntrack_tuple *t,
 286                                  const struct nf_nat_range2 *range)
 287 {
 288         if (t->src.l3num == NFPROTO_IPV4)
 289                 return ntohl(t->src.u3.ip) >= ntohl(range->min_addr.ip) &&
 290                        ntohl(t->src.u3.ip) <= ntohl(range->max_addr.ip);
 291
 292         return ipv6_addr_cmp(&t->src.u3.in6, &range->min_addr.in6) >= 0 &&
 293                ipv6_addr_cmp(&t->src.u3.in6, &range->max_addr.in6) <= 0;
 294 }
 295
 296 /* Is the manipable part of the tuple between min and max incl? */
 297 static bool l4proto_in_range(const struct nf_conntrack_tuple *tuple,
 298                              enum nf_nat_manip_type maniptype,
 299                              const union nf_conntrack_man_proto *min,
 300                              const union nf_conntrack_man_proto *max)
 301 {
 302         __be16 port;
 303
 304         switch (tuple->dst.protonum) {
 305         case IPPROTO_ICMP:
 306         case IPPROTO_ICMPV6:
 307                 return ntohs(tuple->src.u.icmp.id) >= ntohs(min->icmp.id) &&
 308                        ntohs(tuple->src.u.icmp.id) <= ntohs(max->icmp.id);
 309         case IPPROTO_GRE: /* all fall though */
 310         case IPPROTO_TCP:
 311         case IPPROTO_UDP:
 312         case IPPROTO_UDPLITE:
 313         case IPPROTO_DCCP:
 314         case IPPROTO_SCTP:
 315                 if (maniptype == NF_NAT_MANIP_SRC)
 316                         port = tuple->src.u.all;
 317                 else
 318                         port = tuple->dst.u.all;
 319
 320                 return ntohs(port) >= ntohs(min->all) &&
 321                        ntohs(port) <= ntohs(max->all);
 322         default:
 323                 return true;
 324         }
 325 }
 326
 327 /* If we source map this tuple so reply looks like reply_tuple, will
 328  * that meet the constraints of range.
 329  */
 330 static int nf_in_range(const struct nf_conntrack_tuple *tuple,
 331                     const struct nf_nat_range2 *range)
 332 {
 333         /* If we are supposed to map IPs, then we must be in the
 334          * range specified, otherwise let this drag us onto a new src IP.
 335          */
 336         if (range->flags & NF_NAT_RANGE_MAP_IPS &&
 337             !nf_nat_inet_in_range(tuple, range))
 338                 return 0;
 339
 340         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED))
 341                 return 1;
 342
 343         return l4proto_in_range(tuple, NF_NAT_MANIP_SRC,
 344                                 &range->min_proto, &range->max_proto);
 345 }
 346
 347 static inline int
 348 same_src(const struct nf_conn *ct,
 349          const struct nf_conntrack_tuple *tuple)
 350 {
 351         const struct nf_conntrack_tuple *t;
 352
 353         t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 354         return (t->dst.protonum == tuple->dst.protonum &&
 355                 nf_inet_addr_cmp(&t->src.u3, &tuple->src.u3) &&
 356                 t->src.u.all == tuple->src.u.all);
 357 }
 358
 359 /* Only called for SRC manip */
 360 static int
 361 find_appropriate_src(struct net *net,
 362                      const struct nf_conntrack_zone *zone,
 363                      const struct nf_conntrack_tuple *tuple,
 364                      struct nf_conntrack_tuple *result,
 365                      const struct nf_nat_range2 *range)
 366 {
 367         unsigned int h = hash_by_src(net, zone, tuple);
 368         const struct nf_conn *ct;
 369
 370         hlist_for_each_entry_rcu(ct, &nf_nat_bysource[h], nat_bysource) {
 371                 if (same_src(ct, tuple) &&
 372                     net_eq(net, nf_ct_net(ct)) &&
 373                     nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL)) {
 374                         /* Copy source part from reply tuple. */
 375                         nf_ct_invert_tuple(result,
 376                                        &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 377                         result->dst = tuple->dst;
 378
 379                         if (nf_in_range(result, range))
 380                                 return 1;
 381                 }
 382         }
 383         return 0;
 384 }
 385
 386 /* For [FUTURE] fragmentation handling, we want the least-used
 387  * src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
 388  * if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
 389  * 1-65535, we don't do pro-rata allocation based on ports; we choose
 390  * the ip with the lowest src-ip/dst-ip/proto usage.
 391  */
 392 static void
 393 find_best_ips_proto(const struct nf_conntrack_zone *zone,
 394                     struct nf_conntrack_tuple *tuple,
 395                     const struct nf_nat_range2 *range,
 396                     const struct nf_conn *ct,
 397                     enum nf_nat_manip_type maniptype)
 398 {
 399         union nf_inet_addr *var_ipp;
 400         unsigned int i, max;
 401         /* Host order */
 402         u32 minip, maxip, j, dist;
 403         bool full_range;
 404
 405         /* No IP mapping?  Do nothing. */
 406         if (!(range->flags & NF_NAT_RANGE_MAP_IPS))
 407                 return;
 408
 409         if (maniptype == NF_NAT_MANIP_SRC)
 410                 var_ipp = &tuple->src.u3;
 411         else
 412                 var_ipp = &tuple->dst.u3;
 413
 414         /* Fast path: only one choice. */
 415         if (nf_inet_addr_cmp(&range->min_addr, &range->max_addr)) {
 416                 *var_ipp = range->min_addr;
 417                 return;
 418         }
 419
 420         if (nf_ct_l3num(ct) == NFPROTO_IPV4)
 421                 max = sizeof(var_ipp->ip) / sizeof(u32) - 1;
 422         else
 423                 max = sizeof(var_ipp->ip6) / sizeof(u32) - 1;
 424
 425         /* Hashing source and destination IPs gives a fairly even
 426          * spread in practice (if there are a small number of IPs
 427          * involved, there usually aren't that many connections
 428          * anyway).  The consistency means that servers see the same
 429          * client coming from the same IP (some Internet Banking sites
 430          * like this), even across reboots.
 431          */
 432         j = jhash2((u32 *)&tuple->src.u3, sizeof(tuple->src.u3) / sizeof(u32),
 433                    range->flags & NF_NAT_RANGE_PERSISTENT ?
 434                         0 : (__force u32)tuple->dst.u3.all[max] ^ zone->id);
 435
 436         full_range = false;
 437         for (i = 0; i <= max; i++) {
 438                 /* If first bytes of the address are at the maximum, use the
 439                  * distance. Otherwise use the full range.
 440                  */
 441                 if (!full_range) {
 442                         minip = ntohl((__force __be32)range->min_addr.all[i]);
 443                         maxip = ntohl((__force __be32)range->max_addr.all[i]);
 444                         dist  = maxip - minip + 1;
 445                 } else {
 446                         minip = 0;
 447                         dist  = ~0;
 448                 }
 449
 450                 var_ipp->all[i] = (__force __u32)
 451                         htonl(minip + reciprocal_scale(j, dist));
 452                 if (var_ipp->all[i] != range->max_addr.all[i])
 453                         full_range = true;
 454
 455                 if (!(range->flags & NF_NAT_RANGE_PERSISTENT))
 456                         j ^= (__force u32)tuple->dst.u3.all[i];
 457         }
 458 }
 459
 460 /* Alter the per-proto part of the tuple (depending on maniptype), to
 461  * give a unique tuple in the given range if possible.
 462  *
 463  * Per-protocol part of tuple is initialized to the incoming packet.
 464  */
 465 static void nf_nat_l4proto_unique_tuple(struct nf_conntrack_tuple *tuple,
 466                                         const struct nf_nat_range2 *range,
 467                                         enum nf_nat_manip_type maniptype,
 468                                         const struct nf_conn *ct)
 469 {
 470         unsigned int range_size, min, max, i, attempts;
 471         __be16 *keyptr;
 472         u16 off;
 473
 474         switch (tuple->dst.protonum) {
 475         case IPPROTO_ICMP:
 476         case IPPROTO_ICMPV6:
 477                 /* id is same for either direction... */
 478                 keyptr = &tuple->src.u.icmp.id;
 479                 if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED)) {
 480                         min = 0;
 481                         range_size = 65536;
 482                 } else {
 483                         min = ntohs(range->min_proto.icmp.id);
 484                         range_size = ntohs(range->max_proto.icmp.id) -
 485                                      ntohs(range->min_proto.icmp.id) + 1;
 486                 }
 487                 goto find_free_id;
 488 #if IS_ENABLED(CONFIG_NF_CT_PROTO_GRE)
 489         case IPPROTO_GRE:
 490                 /* If there is no master conntrack we are not PPTP,
 491                    do not change tuples */
 492                 if (!ct->master)
 493                         return;
 494
 495                 if (maniptype == NF_NAT_MANIP_SRC)
 496                         keyptr = &tuple->src.u.gre.key;
 497                 else
 498                         keyptr = &tuple->dst.u.gre.key;
 499
 500                 if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED)) {
 501                         min = 1;
 502                         range_size = 65535;
 503                 } else {
 504                         min = ntohs(range->min_proto.gre.key);
 505                         range_size = ntohs(range->max_proto.gre.key) - min + 1;
 506                 }
 507                 goto find_free_id;
 508 #endif
 509         case IPPROTO_UDP:
 510         case IPPROTO_UDPLITE:
 511         case IPPROTO_TCP:
 512         case IPPROTO_SCTP:
 513         case IPPROTO_DCCP:
 514                 if (maniptype == NF_NAT_MANIP_SRC)
 515                         keyptr = &tuple->src.u.all;
 516                 else
 517                         keyptr = &tuple->dst.u.all;
 518
 519                 break;
 520         default:
 521                 return;
 522         }
 523
 524         /* If no range specified... */
 525         if (!(range->flags & NF_NAT_RANGE_PROTO_SPECIFIED)) {
 526                 /* If it's dst rewrite, can't change port */
 527                 if (maniptype == NF_NAT_MANIP_DST)
 528                         return;
 529
 530                 if (ntohs(*keyptr) < 1024) {
 531                         /* Loose convention: >> 512 is credential passing */
 532                         if (ntohs(*keyptr) < 512) {
 533                                 min = 1;
 534                                 range_size = 511 - min + 1;
 535                         } else {
 536                                 min = 600;
 537                                 range_size = 1023 - min + 1;
 538                         }
 539                 } else {
 540                         min = 1024;
 541                         range_size = 65535 - 1024 + 1;
 542                 }
 543         } else {
 544                 min = ntohs(range->min_proto.all);
 545                 max = ntohs(range->max_proto.all);
 546                 if (unlikely(max < min))
 547                         swap(max, min);
 548                 range_size = max - min + 1;
 549         }
 550
 551 find_free_id:
 552         if (range->flags & NF_NAT_RANGE_PROTO_OFFSET)
 553                 off = (ntohs(*keyptr) - ntohs(range->base_proto.all));
 554         else if ((range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL) ||
 555                  maniptype != NF_NAT_MANIP_DST)
 556                 off = get_random_u16();
 557         else
 558                 off = 0;
 559
 560         attempts = range_size;
 561         if (attempts > NF_NAT_MAX_ATTEMPTS)
 562                 attempts = NF_NAT_MAX_ATTEMPTS;
 563
 564         /* We are in softirq; doing a search of the entire range risks
 565          * soft lockup when all tuples are already used.
 566          *
 567          * If we can't find any free port from first offset, pick a new
 568          * one and try again, with ever smaller search window.
 569          */
 570 another_round:
 571         for (i = 0; i < attempts; i++, off++) {
 572                 *keyptr = htons(min + off % range_size);
 573                 if (!nf_nat_used_tuple_harder(tuple, ct, attempts - i))
 574                         return;
 575         }
 576
 577         if (attempts >= range_size || attempts < 16)
 578                 return;
 579         attempts /= 2;
 580         off = get_random_u16();
 581         goto another_round;
 582 }
 583
 584 /* Manipulate the tuple into the range given. For NF_INET_POST_ROUTING,
 585  * we change the source to map into the range. For NF_INET_PRE_ROUTING
 586  * and NF_INET_LOCAL_OUT, we change the destination to map into the
 587  * range. It might not be possible to get a unique tuple, but we try.
 588  * At worst (or if we race), we will end up with a final duplicate in
 589  * __nf_conntrack_confirm and drop the packet. */
 590 static void
 591 get_unique_tuple(struct nf_conntrack_tuple *tuple,
 592                  const struct nf_conntrack_tuple *orig_tuple,
 593                  const struct nf_nat_range2 *range,
 594                  struct nf_conn *ct,
 595                  enum nf_nat_manip_type maniptype)
 596 {
 597         const struct nf_conntrack_zone *zone;
 598         struct net *net = nf_ct_net(ct);
 599
 600         zone = nf_ct_zone(ct);
 601
 602         /* 1) If this srcip/proto/src-proto-part is currently mapped,
 603          * and that same mapping gives a unique tuple within the given
 604          * range, use that.
 605          *
 606          * This is only required for source (ie. NAT/masq) mappings.
 607          * So far, we don't do local source mappings, so multiple
 608          * manips not an issue.
 609          */
 610         if (maniptype == NF_NAT_MANIP_SRC &&
 611             !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 612                 /* try the original tuple first */
 613                 if (nf_in_range(orig_tuple, range)) {
 614                         if (!nf_nat_used_tuple(orig_tuple, ct)) {
 615                                 *tuple = *orig_tuple;
 616                                 return;
 617                         }
 618                 } else if (find_appropriate_src(net, zone,
 619                                                 orig_tuple, tuple, range)) {
 620                         pr_debug("get_unique_tuple: Found current src map\n");
 621                         if (!nf_nat_used_tuple(tuple, ct))
 622                                 return;
 623                 }
 624         }
 625
 626         /* 2) Select the least-used IP/proto combination in the given range */
 627         *tuple = *orig_tuple;
 628         find_best_ips_proto(zone, tuple, range, ct, maniptype);
 629
 630         /* 3) The per-protocol part of the manip is made to map into
 631          * the range to make a unique tuple.
 632          */
 633
 634         /* Only bother mapping if it's not already in range and unique */
 635         if (!(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 636                 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
 637                         if (!(range->flags & NF_NAT_RANGE_PROTO_OFFSET) &&
 638                             l4proto_in_range(tuple, maniptype,
 639                                              &range->min_proto,
 640                                              &range->max_proto) &&
 641                             (range->min_proto.all == range->max_proto.all ||
 642                              !nf_nat_used_tuple(tuple, ct)))
 643                                 return;
 644                 } else if (!nf_nat_used_tuple(tuple, ct)) {
 645                         return;
 646                 }
 647         }
 648
 649         /* Last chance: get protocol to try to obtain unique tuple. */
 650         nf_nat_l4proto_unique_tuple(tuple, range, maniptype, ct);
 651 }
 652
 653 struct nf_conn_nat *nf_ct_nat_ext_add(struct nf_conn *ct)
 654 {
 655         struct nf_conn_nat *nat = nfct_nat(ct);
 656         if (nat)
 657                 return nat;
 658
 659         if (!nf_ct_is_confirmed(ct))
 660                 nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
 661
 662         return nat;
 663 }
 664 EXPORT_SYMBOL_GPL(nf_ct_nat_ext_add);
 665
 666 unsigned int
 667 nf_nat_setup_info(struct nf_conn *ct,
 668                   const struct nf_nat_range2 *range,
 669                   enum nf_nat_manip_type maniptype)
 670 {
 671         struct net *net = nf_ct_net(ct);
 672         struct nf_conntrack_tuple curr_tuple, new_tuple;
 673
 674         /* Can't setup nat info for confirmed ct. */
 675         if (nf_ct_is_confirmed(ct))
 676                 return NF_ACCEPT;
 677
 678         WARN_ON(maniptype != NF_NAT_MANIP_SRC &&
 679                 maniptype != NF_NAT_MANIP_DST);
 680
 681         if (WARN_ON(nf_nat_initialized(ct, maniptype)))
 682                 return NF_DROP;
 683
 684         /* What we've got will look like inverse of reply. Normally
 685          * this is what is in the conntrack, except for prior
 686          * manipulations (future optimization: if num_manips == 0,
 687          * orig_tp = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
 688          */
 689         nf_ct_invert_tuple(&curr_tuple,
 690                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 691
 692         get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
 693
 694         if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
 695                 struct nf_conntrack_tuple reply;
 696
 697                 /* Alter conntrack table so will recognize replies. */
 698                 nf_ct_invert_tuple(&reply, &new_tuple);
 699                 nf_conntrack_alter_reply(ct, &reply);
 700
 701                 /* Non-atomic: we own this at the moment. */
 702                 if (maniptype == NF_NAT_MANIP_SRC)
 703                         ct->status |= IPS_SRC_NAT;
 704                 else
 705                         ct->status |= IPS_DST_NAT;
 706
 707                 if (nfct_help(ct) && !nfct_seqadj(ct))
 708                         if (!nfct_seqadj_ext_add(ct))
 709                                 return NF_DROP;
 710         }
 711
 712         if (maniptype == NF_NAT_MANIP_SRC) {
 713                 unsigned int srchash;
 714                 spinlock_t *lock;
 715
 716                 srchash = hash_by_src(net, nf_ct_zone(ct),
 717                                       &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 718                 lock = &nf_nat_locks[srchash % CONNTRACK_LOCKS];
 719                 spin_lock_bh(lock);
 720                 hlist_add_head_rcu(&ct->nat_bysource,
 721                                    &nf_nat_bysource[srchash]);
 722                 spin_unlock_bh(lock);
 723         }
 724
 725         /* It's done. */
 726         if (maniptype == NF_NAT_MANIP_DST)
 727                 ct->status |= IPS_DST_NAT_DONE;
 728         else
 729                 ct->status |= IPS_SRC_NAT_DONE;
 730
 731         return NF_ACCEPT;
 732 }
 733 EXPORT_SYMBOL(nf_nat_setup_info);
 734
 735 static unsigned int
 736 __nf_nat_alloc_null_binding(struct nf_conn *ct, enum nf_nat_manip_type manip)
 737 {
 738         /* Force range to this IP; let proto decide mapping for
 739          * per-proto parts (hence not IP_NAT_RANGE_PROTO_SPECIFIED).
 740          * Use reply in case it's already been mangled (eg local packet).
 741          */
 742         union nf_inet_addr ip =
 743                 (manip == NF_NAT_MANIP_SRC ?
 744                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3 :
 745                 ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3);
 746         struct nf_nat_range2 range = {
 747                 .flags          = NF_NAT_RANGE_MAP_IPS,
 748                 .min_addr       = ip,
 749                 .max_addr       = ip,
 750         };
 751         return nf_nat_setup_info(ct, &range, manip);
 752 }
 753
 754 unsigned int
 755 nf_nat_alloc_null_binding(struct nf_conn *ct, unsigned int hooknum)
 756 {
 757         return __nf_nat_alloc_null_binding(ct, HOOK2MANIP(hooknum));
 758 }
 759 EXPORT_SYMBOL_GPL(nf_nat_alloc_null_binding);
 760
 761 /* Do packet manipulations according to nf_nat_setup_info. */
 762 unsigned int nf_nat_packet(struct nf_conn *ct,
 763                            enum ip_conntrack_info ctinfo,
 764                            unsigned int hooknum,
 765                            struct sk_buff *skb)
 766 {
 767         enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
 768         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 769         unsigned int verdict = NF_ACCEPT;
 770         unsigned long statusbit;
 771
 772         if (mtype == NF_NAT_MANIP_SRC)
 773                 statusbit = IPS_SRC_NAT;
 774         else
 775                 statusbit = IPS_DST_NAT;
 776
 777         /* Invert if this is reply dir. */
 778         if (dir == IP_CT_DIR_REPLY)
 779                 statusbit ^= IPS_NAT_MASK;
 780
 781         /* Non-atomic: these bits don't change. */
 782         if (ct->status & statusbit)
 783                 verdict = nf_nat_manip_pkt(skb, ct, mtype, dir);
 784
 785         return verdict;
 786 }
 787 EXPORT_SYMBOL_GPL(nf_nat_packet);
 788
 789 static bool in_vrf_postrouting(const struct nf_hook_state *state)
 790 {
 791 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
 792         if (state->hook == NF_INET_POST_ROUTING &&
 793             netif_is_l3_master(state->out))
 794                 return true;
 795 #endif
 796         return false;
 797 }
 798
 799 unsigned int
 800 nf_nat_inet_fn(void *priv, struct sk_buff *skb,
 801                const struct nf_hook_state *state)
 802 {
 803         struct nf_conn *ct;
 804         enum ip_conntrack_info ctinfo;
 805         struct nf_conn_nat *nat;
 806         /* maniptype == SRC for postrouting. */
 807         enum nf_nat_manip_type maniptype = HOOK2MANIP(state->hook);
 808
 809         ct = nf_ct_get(skb, &ctinfo);
 810         /* Can't track?  It's not due to stress, or conntrack would
 811          * have dropped it.  Hence it's the user's responsibilty to
 812          * packet filter it out, or implement conntrack/NAT for that
 813          * protocol. 8) --RR
 814          */
 815         if (!ct || in_vrf_postrouting(state))
 816                 return NF_ACCEPT;
 817
 818         nat = nfct_nat(ct);
 819
 820         switch (ctinfo) {
 821         case IP_CT_RELATED:
 822         case IP_CT_RELATED_REPLY:
 823                 /* Only ICMPs can be IP_CT_IS_REPLY.  Fallthrough */
 824         case IP_CT_NEW:
 825                 /* Seen it before?  This can happen for loopback, retrans,
 826                  * or local packets.
 827                  */
 828                 if (!nf_nat_initialized(ct, maniptype)) {
 829                         struct nf_nat_lookup_hook_priv *lpriv = priv;
 830                         struct nf_hook_entries *e = rcu_dereference(lpriv->entries);
 831                         unsigned int ret;
 832                         int i;
 833
 834                         if (!e)
 835                                 goto null_bind;
 836
 837                         for (i = 0; i < e->num_hook_entries; i++) {
 838                                 ret = e->hooks[i].hook(e->hooks[i].priv, skb,
 839                                                        state);
 840                                 if (ret != NF_ACCEPT)
 841                                         return ret;
 842                                 if (nf_nat_initialized(ct, maniptype))
 843                                         goto do_nat;
 844                         }
 845 null_bind:
 846                         ret = nf_nat_alloc_null_binding(ct, state->hook);
 847                         if (ret != NF_ACCEPT)
 848                                 return ret;
 849                 } else {
 850                         pr_debug("Already setup manip %s for ct %p (status bits 0x%lx)\n",
 851                                  maniptype == NF_NAT_MANIP_SRC ? "SRC" : "DST",
 852                                  ct, ct->status);
 853                         if (nf_nat_oif_changed(state->hook, ctinfo, nat,
 854                                                state->out))
 855                                 goto oif_changed;
 856                 }
 857                 break;
 858         default:
 859                 /* ESTABLISHED */
 860                 WARN_ON(ctinfo != IP_CT_ESTABLISHED &&
 861                         ctinfo != IP_CT_ESTABLISHED_REPLY);
 862                 if (nf_nat_oif_changed(state->hook, ctinfo, nat, state->out))
 863                         goto oif_changed;
 864         }
 865 do_nat:
 866         return nf_nat_packet(ct, ctinfo, state->hook, skb);
 867
 868 oif_changed:
 869         nf_ct_kill_acct(ct, ctinfo, skb);
 870         return NF_DROP;
 871 }
 872 EXPORT_SYMBOL_GPL(nf_nat_inet_fn);
 873
 874 struct nf_nat_proto_clean {
 875         u8      l3proto;
 876         u8      l4proto;
 877 };
 878
 879 /* kill conntracks with affected NAT section */
 880 static int nf_nat_proto_remove(struct nf_conn *i, void *data)
 881 {
 882         const struct nf_nat_proto_clean *clean = data;
 883
 884         if ((clean->l3proto && nf_ct_l3num(i) != clean->l3proto) ||
 885             (clean->l4proto && nf_ct_protonum(i) != clean->l4proto))
 886                 return 0;
 887
 888         return i->status & IPS_NAT_MASK ? 1 : 0;
 889 }
 890
 891 static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 892 {
 893         unsigned int h;
 894
 895         h = hash_by_src(nf_ct_net(ct), nf_ct_zone(ct), &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 896         spin_lock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 897         hlist_del_rcu(&ct->nat_bysource);
 898         spin_unlock_bh(&nf_nat_locks[h % CONNTRACK_LOCKS]);
 899 }
 900
 901 static int nf_nat_proto_clean(struct nf_conn *ct, void *data)
 902 {
 903         if (nf_nat_proto_remove(ct, data))
 904                 return 1;
 905
 906         /* This module is being removed and conntrack has nat null binding.
 907          * Remove it from bysource hash, as the table will be freed soon.
 908          *
 909          * Else, when the conntrack is destoyed, nf_nat_cleanup_conntrack()
 910          * will delete entry from already-freed table.
 911          */
 912         if (test_and_clear_bit(IPS_SRC_NAT_DONE_BIT, &ct->status))
 913                 nf_nat_cleanup_conntrack(ct);
 914
 915         /* don't delete conntrack.  Although that would make things a lot
 916          * simpler, we'd end up flushing all conntracks on nat rmmod.
 917          */
 918         return 0;
 919 }
 920
 921 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 922
 923 #include <linux/netfilter/nfnetlink.h>
 924 #include <linux/netfilter/nfnetlink_conntrack.h>
 925
 926 static const struct nla_policy protonat_nla_policy[CTA_PROTONAT_MAX+1] = {
 927         [CTA_PROTONAT_PORT_MIN] = { .type = NLA_U16 },
 928         [CTA_PROTONAT_PORT_MAX] = { .type = NLA_U16 },
 929 };
 930
 931 static int nf_nat_l4proto_nlattr_to_range(struct nlattr *tb[],
 932                                           struct nf_nat_range2 *range)
 933 {
 934         if (tb[CTA_PROTONAT_PORT_MIN]) {
 935                 range->min_proto.all = nla_get_be16(tb[CTA_PROTONAT_PORT_MIN]);
 936                 range->max_proto.all = range->min_proto.all;
 937                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
 938         }
 939         if (tb[CTA_PROTONAT_PORT_MAX]) {
 940                 range->max_proto.all = nla_get_be16(tb[CTA_PROTONAT_PORT_MAX]);
 941                 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
 942         }
 943         return 0;
 944 }
 945
 946 static int nfnetlink_parse_nat_proto(struct nlattr *attr,
 947                                      const struct nf_conn *ct,
 948                                      struct nf_nat_range2 *range)
 949 {
 950         struct nlattr *tb[CTA_PROTONAT_MAX+1];
 951         int err;
 952
 953         err = nla_parse_nested_deprecated(tb, CTA_PROTONAT_MAX, attr,
 954                                           protonat_nla_policy, NULL);
 955         if (err < 0)
 956                 return err;
 957
 958         return nf_nat_l4proto_nlattr_to_range(tb, range);
 959 }
 960
 961 static const struct nla_policy nat_nla_policy[CTA_NAT_MAX+1] = {
 962         [CTA_NAT_V4_MINIP]      = { .type = NLA_U32 },
 963         [CTA_NAT_V4_MAXIP]      = { .type = NLA_U32 },
 964         [CTA_NAT_V6_MINIP]      = { .len = sizeof(struct in6_addr) },
 965         [CTA_NAT_V6_MAXIP]      = { .len = sizeof(struct in6_addr) },
 966         [CTA_NAT_PROTO]         = { .type = NLA_NESTED },
 967 };
 968
 969 static int nf_nat_ipv4_nlattr_to_range(struct nlattr *tb[],
 970                                        struct nf_nat_range2 *range)
 971 {
 972         if (tb[CTA_NAT_V4_MINIP]) {
 973                 range->min_addr.ip = nla_get_be32(tb[CTA_NAT_V4_MINIP]);
 974                 range->flags |= NF_NAT_RANGE_MAP_IPS;
 975         }
 976
 977         if (tb[CTA_NAT_V4_MAXIP])
 978                 range->max_addr.ip = nla_get_be32(tb[CTA_NAT_V4_MAXIP]);
 979         else
 980                 range->max_addr.ip = range->min_addr.ip;
 981
 982         return 0;
 983 }
 984
 985 static int nf_nat_ipv6_nlattr_to_range(struct nlattr *tb[],
 986                                        struct nf_nat_range2 *range)
 987 {
 988         if (tb[CTA_NAT_V6_MINIP]) {
 989                 nla_memcpy(&range->min_addr.ip6, tb[CTA_NAT_V6_MINIP],
 990                            sizeof(struct in6_addr));
 991                 range->flags |= NF_NAT_RANGE_MAP_IPS;
 992         }
 993
 994         if (tb[CTA_NAT_V6_MAXIP])
 995                 nla_memcpy(&range->max_addr.ip6, tb[CTA_NAT_V6_MAXIP],
 996                            sizeof(struct in6_addr));
 997         else
 998                 range->max_addr = range->min_addr;
 999
1000         return 0;
1001 }
1002
1003 static int
1004 nfnetlink_parse_nat(const struct nlattr *nat,
1005                     const struct nf_conn *ct, struct nf_nat_range2 *range)
1006 {
1007         struct nlattr *tb[CTA_NAT_MAX+1];
1008         int err;
1009
1010         memset(range, 0, sizeof(*range));
1011
1012         err = nla_parse_nested_deprecated(tb, CTA_NAT_MAX, nat,
1013                                           nat_nla_policy, NULL);
1014         if (err < 0)
1015                 return err;
1016
1017         switch (nf_ct_l3num(ct)) {
1018         case NFPROTO_IPV4:
1019                 err = nf_nat_ipv4_nlattr_to_range(tb, range);
1020                 break;
1021         case NFPROTO_IPV6:
1022                 err = nf_nat_ipv6_nlattr_to_range(tb, range);
1023                 break;
1024         default:
1025                 err = -EPROTONOSUPPORT;
1026                 break;
1027         }
1028
1029         if (err)
1030                 return err;
1031
1032         if (!tb[CTA_NAT_PROTO])
1033                 return 0;
1034
1035         return nfnetlink_parse_nat_proto(tb[CTA_NAT_PROTO], ct, range);
1036 }
1037
1038 /* This function is called under rcu_read_lock() */
1039 static int
1040 nfnetlink_parse_nat_setup(struct nf_conn *ct,
1041                           enum nf_nat_manip_type manip,
1042                           const struct nlattr *attr)
1043 {
1044         struct nf_nat_range2 range;
1045         int err;
1046
1047         /* Should not happen, restricted to creating new conntracks
1048          * via ctnetlink.
1049          */
1050         if (WARN_ON_ONCE(nf_nat_initialized(ct, manip)))
1051                 return -EEXIST;
1052
1053         /* No NAT information has been passed, allocate the null-binding */
1054         if (attr == NULL)
1055                 return __nf_nat_alloc_null_binding(ct, manip) == NF_DROP ? -ENOMEM : 0;
1056
1057         err = nfnetlink_parse_nat(attr, ct, &range);
1058         if (err < 0)
1059                 return err;
1060
1061         return nf_nat_setup_info(ct, &range, manip) == NF_DROP ? -ENOMEM : 0;
1062 }
1063 #else
1064 static int
1065 nfnetlink_parse_nat_setup(struct nf_conn *ct,
1066                           enum nf_nat_manip_type manip,
1067                           const struct nlattr *attr)
1068 {
1069         return -EOPNOTSUPP;
1070 }
1071 #endif
1072
1073 static struct nf_ct_helper_expectfn follow_master_nat = {
1074         .name           = "nat-follow-master",
1075         .expectfn       = nf_nat_follow_master,
1076 };
1077
1078 int nf_nat_register_fn(struct net *net, u8 pf, const struct nf_hook_ops *ops,
1079                        const struct nf_hook_ops *orig_nat_ops, unsigned int ops_count)
1080 {
1081         struct nat_net *nat_net = net_generic(net, nat_net_id);
1082         struct nf_nat_hooks_net *nat_proto_net;
1083         struct nf_nat_lookup_hook_priv *priv;
1084         unsigned int hooknum = ops->hooknum;
1085         struct nf_hook_ops *nat_ops;
1086         int i, ret;
1087
1088         if (WARN_ON_ONCE(pf >= ARRAY_SIZE(nat_net->nat_proto_net)))
1089                 return -EINVAL;
1090
1091         nat_proto_net = &nat_net->nat_proto_net[pf];
1092
1093         for (i = 0; i < ops_count; i++) {
1094                 if (orig_nat_ops[i].hooknum == hooknum) {
1095                         hooknum = i;
1096                         break;
1097                 }
1098         }
1099
1100         if (WARN_ON_ONCE(i == ops_count))
1101                 return -EINVAL;
1102
1103         mutex_lock(&nf_nat_proto_mutex);
1104         if (!nat_proto_net->nat_hook_ops) {
1105                 WARN_ON(nat_proto_net->users != 0);
1106
1107                 nat_ops = kmemdup(orig_nat_ops, sizeof(*orig_nat_ops) * ops_count, GFP_KERNEL);
1108                 if (!nat_ops) {
1109                         mutex_unlock(&nf_nat_proto_mutex);
1110                         return -ENOMEM;
1111                 }
1112
1113                 for (i = 0; i < ops_count; i++) {
1114                         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1115                         if (priv) {
1116                                 nat_ops[i].priv = priv;
1117                                 continue;
1118                         }
1119                         mutex_unlock(&nf_nat_proto_mutex);
1120                         while (i)
1121                                 kfree(nat_ops[--i].priv);
1122                         kfree(nat_ops);
1123                         return -ENOMEM;
1124                 }
1125
1126                 ret = nf_register_net_hooks(net, nat_ops, ops_count);
1127                 if (ret < 0) {
1128                         mutex_unlock(&nf_nat_proto_mutex);
1129                         for (i = 0; i < ops_count; i++)
1130                                 kfree(nat_ops[i].priv);
1131                         kfree(nat_ops);
1132                         return ret;
1133                 }
1134
1135                 nat_proto_net->nat_hook_ops = nat_ops;
1136         }
1137
1138         nat_ops = nat_proto_net->nat_hook_ops;
1139         priv = nat_ops[hooknum].priv;
1140         if (WARN_ON_ONCE(!priv)) {
1141                 mutex_unlock(&nf_nat_proto_mutex);
1142                 return -EOPNOTSUPP;
1143         }
1144
1145         ret = nf_hook_entries_insert_raw(&priv->entries, ops);
1146         if (ret == 0)
1147                 nat_proto_net->users++;
1148
1149         mutex_unlock(&nf_nat_proto_mutex);
1150         return ret;
1151 }
1152
1153 void nf_nat_unregister_fn(struct net *net, u8 pf, const struct nf_hook_ops *ops,
1154                           unsigned int ops_count)
1155 {
1156         struct nat_net *nat_net = net_generic(net, nat_net_id);
1157         struct nf_nat_hooks_net *nat_proto_net;
1158         struct nf_nat_lookup_hook_priv *priv;
1159         struct nf_hook_ops *nat_ops;
1160         int hooknum = ops->hooknum;
1161         int i;
1162
1163         if (pf >= ARRAY_SIZE(nat_net->nat_proto_net))
1164                 return;
1165
1166         nat_proto_net = &nat_net->nat_proto_net[pf];
1167
1168         mutex_lock(&nf_nat_proto_mutex);
1169         if (WARN_ON(nat_proto_net->users == 0))
1170                 goto unlock;
1171
1172         nat_proto_net->users--;
1173
1174         nat_ops = nat_proto_net->nat_hook_ops;
1175         for (i = 0; i < ops_count; i++) {
1176                 if (nat_ops[i].hooknum == hooknum) {
1177                         hooknum = i;
1178                         break;
1179                 }
1180         }
1181         if (WARN_ON_ONCE(i == ops_count))
1182                 goto unlock;
1183         priv = nat_ops[hooknum].priv;
1184         nf_hook_entries_delete_raw(&priv->entries, ops);
1185
1186         if (nat_proto_net->users == 0) {
1187                 nf_unregister_net_hooks(net, nat_ops, ops_count);
1188
1189                 for (i = 0; i < ops_count; i++) {
1190                         priv = nat_ops[i].priv;
1191                         kfree_rcu(priv, rcu_head);
1192                 }
1193
1194                 nat_proto_net->nat_hook_ops = NULL;
1195                 kfree(nat_ops);
1196         }
1197 unlock:
1198         mutex_unlock(&nf_nat_proto_mutex);
1199 }
1200
1201 static struct pernet_operations nat_net_ops = {
1202         .id = &nat_net_id,
1203         .size = sizeof(struct nat_net),
1204 };
1205
1206 static const struct nf_nat_hook nat_hook = {
1207         .parse_nat_setup        = nfnetlink_parse_nat_setup,
1208 #ifdef CONFIG_XFRM
1209         .decode_session         = __nf_nat_decode_session,
1210 #endif
1211         .manip_pkt              = nf_nat_manip_pkt,
1212         .remove_nat_bysrc       = nf_nat_cleanup_conntrack,
1213 };
1214
1215 static int __init nf_nat_init(void)
1216 {
1217         int ret, i;
1218
1219         /* Leave them the same for the moment. */
1220         nf_nat_htable_size = nf_conntrack_htable_size;
1221         if (nf_nat_htable_size < CONNTRACK_LOCKS)
1222                 nf_nat_htable_size = CONNTRACK_LOCKS;
1223
1224         nf_nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size, 0);
1225         if (!nf_nat_bysource)
1226                 return -ENOMEM;
1227
1228         for (i = 0; i < CONNTRACK_LOCKS; i++)
1229                 spin_lock_init(&nf_nat_locks[i]);
1230
1231         ret = register_pernet_subsys(&nat_net_ops);
1232         if (ret < 0) {
1233                 kvfree(nf_nat_bysource);
1234                 return ret;
1235         }
1236
1237         nf_ct_helper_expectfn_register(&follow_master_nat);
1238
1239         WARN_ON(nf_nat_hook != NULL);
1240         RCU_INIT_POINTER(nf_nat_hook, &nat_hook);
1241
1242         ret = register_nf_nat_bpf();
1243         if (ret < 0) {
1244                 RCU_INIT_POINTER(nf_nat_hook, NULL);
1245                 nf_ct_helper_expectfn_unregister(&follow_master_nat);
1246                 synchronize_net();
1247                 unregister_pernet_subsys(&nat_net_ops);
1248                 kvfree(nf_nat_bysource);
1249         }
1250
1251         return ret;
1252 }
1253
1254 static void __exit nf_nat_cleanup(void)
1255 {
1256         struct nf_nat_proto_clean clean = {};
1257
1258         nf_ct_iterate_destroy(nf_nat_proto_clean, &clean);
1259
1260         nf_ct_helper_expectfn_unregister(&follow_master_nat);
1261         RCU_INIT_POINTER(nf_nat_hook, NULL);
1262
1263         synchronize_net();
1264         kvfree(nf_nat_bysource);
1265         unregister_pernet_subsys(&nat_net_ops);
1266 }
1267
1268 MODULE_LICENSE("GPL");
1269 MODULE_DESCRIPTION("Network address translation core");
1270
1271 module_init(nf_nat_init);
1272 module_exit(nf_nat_cleanup);