net/netfilter/core.c

   1 /* netfilter.c: look after the filters for various protocols.
   2  * Heavily influenced by the old firewall.c by David Bonn and Alan Cox.
   3  *
   4  * Thanks to Rob `CmdrTaco' Malda for not influencing this code in any
   5  * way.
   6  *
   7  * This code is GPL.
   8  */
   9 #include <linux/kernel.h>
  10 #include <linux/netfilter.h>
  11 #include <net/protocol.h>
  12 #include <linux/init.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/wait.h>
  15 #include <linux/module.h>
  16 #include <linux/interrupt.h>
  17 #include <linux/if.h>
  18 #include <linux/netdevice.h>
  19 #include <linux/netfilter_ipv6.h>
  20 #include <linux/inetdevice.h>
  21 #include <linux/proc_fs.h>
  22 #include <linux/mutex.h>
  23 #include <linux/mm.h>
  24 #include <linux/rcupdate.h>
  25 #include <net/net_namespace.h>
  26 #include <net/netfilter/nf_queue.h>
  27 #include <net/sock.h>
  28
  29 #include "nf_internals.h"
  30
  31 const struct nf_ipv6_ops __rcu *nf_ipv6_ops __read_mostly;
  32 EXPORT_SYMBOL_GPL(nf_ipv6_ops);
  33
  34 DEFINE_PER_CPU(bool, nf_skb_duplicated);
  35 EXPORT_SYMBOL_GPL(nf_skb_duplicated);
  36
  37 #ifdef CONFIG_JUMP_LABEL
  38 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
  39 EXPORT_SYMBOL(nf_hooks_needed);
  40 #endif
  41
  42 static DEFINE_MUTEX(nf_hook_mutex);
  43
  44 /* max hooks per family/hooknum */
  45 #define MAX_HOOK_COUNT          1024
  46
  47 #define nf_entry_dereference(e) \
  48         rcu_dereference_protected(e, lockdep_is_held(&nf_hook_mutex))
  49
  50 static struct nf_hook_entries *allocate_hook_entries_size(u16 num)
  51 {
  52         struct nf_hook_entries *e;
  53         size_t alloc = sizeof(*e) +
  54                        sizeof(struct nf_hook_entry) * num +
  55                        sizeof(struct nf_hook_ops *) * num +
  56                        sizeof(struct nf_hook_entries_rcu_head);
  57
  58         if (num == 0)
  59                 return NULL;
  60
  61         e = kvzalloc(alloc, GFP_KERNEL);
  62         if (e)
  63                 e->num_hook_entries = num;
  64         return e;
  65 }
  66
  67 static void __nf_hook_entries_free(struct rcu_head *h)
  68 {
  69         struct nf_hook_entries_rcu_head *head;
  70
  71         head = container_of(h, struct nf_hook_entries_rcu_head, head);
  72         kvfree(head->allocation);
  73 }
  74
  75 static void nf_hook_entries_free(struct nf_hook_entries *e)
  76 {
  77         struct nf_hook_entries_rcu_head *head;
  78         struct nf_hook_ops **ops;
  79         unsigned int num;
  80
  81         if (!e)
  82                 return;
  83
  84         num = e->num_hook_entries;
  85         ops = nf_hook_entries_get_hook_ops(e);
  86         head = (void *)&ops[num];
  87         head->allocation = e;
  88         call_rcu(&head->head, __nf_hook_entries_free);
  89 }
  90
  91 static unsigned int accept_all(void *priv,
  92                                struct sk_buff *skb,
  93                                const struct nf_hook_state *state)
  94 {
  95         return NF_ACCEPT; /* ACCEPT makes nf_hook_slow call next hook */
  96 }
  97
  98 static const struct nf_hook_ops dummy_ops = {
  99         .hook = accept_all,
 100         .priority = INT_MIN,
 101 };
 102
 103 static struct nf_hook_entries *
 104 nf_hook_entries_grow(const struct nf_hook_entries *old,
 105                      const struct nf_hook_ops *reg)
 106 {
 107         unsigned int i, alloc_entries, nhooks, old_entries;
 108         struct nf_hook_ops **orig_ops = NULL;
 109         struct nf_hook_ops **new_ops;
 110         struct nf_hook_entries *new;
 111         bool inserted = false;
 112
 113         alloc_entries = 1;
 114         old_entries = old ? old->num_hook_entries : 0;
 115
 116         if (old) {
 117                 orig_ops = nf_hook_entries_get_hook_ops(old);
 118
 119                 for (i = 0; i < old_entries; i++) {
 120                         if (orig_ops[i] != &dummy_ops)
 121                                 alloc_entries++;
 122                 }
 123         }
 124
 125         if (alloc_entries > MAX_HOOK_COUNT)
 126                 return ERR_PTR(-E2BIG);
 127
 128         new = allocate_hook_entries_size(alloc_entries);
 129         if (!new)
 130                 return ERR_PTR(-ENOMEM);
 131
 132         new_ops = nf_hook_entries_get_hook_ops(new);
 133
 134         i = 0;
 135         nhooks = 0;
 136         while (i < old_entries) {
 137                 if (orig_ops[i] == &dummy_ops) {
 138                         ++i;
 139                         continue;
 140                 }
 141
 142                 if (inserted || reg->priority > orig_ops[i]->priority) {
 143                         new_ops[nhooks] = (void *)orig_ops[i];
 144                         new->hooks[nhooks] = old->hooks[i];
 145                         i++;
 146                 } else {
 147                         new_ops[nhooks] = (void *)reg;
 148                         new->hooks[nhooks].hook = reg->hook;
 149                         new->hooks[nhooks].priv = reg->priv;
 150                         inserted = true;
 151                 }
 152                 nhooks++;
 153         }
 154
 155         if (!inserted) {
 156                 new_ops[nhooks] = (void *)reg;
 157                 new->hooks[nhooks].hook = reg->hook;
 158                 new->hooks[nhooks].priv = reg->priv;
 159         }
 160
 161         return new;
 162 }
 163
 164 static void hooks_validate(const struct nf_hook_entries *hooks)
 165 {
 166 #ifdef CONFIG_DEBUG_MISC
 167         struct nf_hook_ops **orig_ops;
 168         int prio = INT_MIN;
 169         size_t i = 0;
 170
 171         orig_ops = nf_hook_entries_get_hook_ops(hooks);
 172
 173         for (i = 0; i < hooks->num_hook_entries; i++) {
 174                 if (orig_ops[i] == &dummy_ops)
 175                         continue;
 176
 177                 WARN_ON(orig_ops[i]->priority < prio);
 178
 179                 if (orig_ops[i]->priority > prio)
 180                         prio = orig_ops[i]->priority;
 181         }
 182 #endif
 183 }
 184
 185 int nf_hook_entries_insert_raw(struct nf_hook_entries __rcu **pp,
 186                                 const struct nf_hook_ops *reg)
 187 {
 188         struct nf_hook_entries *new_hooks;
 189         struct nf_hook_entries *p;
 190
 191         p = rcu_dereference_raw(*pp);
 192         new_hooks = nf_hook_entries_grow(p, reg);
 193         if (IS_ERR(new_hooks))
 194                 return PTR_ERR(new_hooks);
 195
 196         hooks_validate(new_hooks);
 197
 198         rcu_assign_pointer(*pp, new_hooks);
 199
 200         BUG_ON(p == new_hooks);
 201         nf_hook_entries_free(p);
 202         return 0;
 203 }
 204 EXPORT_SYMBOL_GPL(nf_hook_entries_insert_raw);
 205
 206 /*
 207  * __nf_hook_entries_try_shrink - try to shrink hook array
 208  *
 209  * @old -- current hook blob at @pp
 210  * @pp -- location of hook blob
 211  *
 212  * Hook unregistration must always succeed, so to-be-removed hooks
 213  * are replaced by a dummy one that will just move to next hook.
 214  *
 215  * This counts the current dummy hooks, attempts to allocate new blob,
 216  * copies the live hooks, then replaces and discards old one.
 217  *
 218  * return values:
 219  *
 220  * Returns address to free, or NULL.
 221  */
 222 static void *__nf_hook_entries_try_shrink(struct nf_hook_entries *old,
 223                                           struct nf_hook_entries __rcu **pp)
 224 {
 225         unsigned int i, j, skip = 0, hook_entries;
 226         struct nf_hook_entries *new = NULL;
 227         struct nf_hook_ops **orig_ops;
 228         struct nf_hook_ops **new_ops;
 229
 230         if (WARN_ON_ONCE(!old))
 231                 return NULL;
 232
 233         orig_ops = nf_hook_entries_get_hook_ops(old);
 234         for (i = 0; i < old->num_hook_entries; i++) {
 235                 if (orig_ops[i] == &dummy_ops)
 236                         skip++;
 237         }
 238
 239         /* if skip == hook_entries all hooks have been removed */
 240         hook_entries = old->num_hook_entries;
 241         if (skip == hook_entries)
 242                 goto out_assign;
 243
 244         if (skip == 0)
 245                 return NULL;
 246
 247         hook_entries -= skip;
 248         new = allocate_hook_entries_size(hook_entries);
 249         if (!new)
 250                 return NULL;
 251
 252         new_ops = nf_hook_entries_get_hook_ops(new);
 253         for (i = 0, j = 0; i < old->num_hook_entries; i++) {
 254                 if (orig_ops[i] == &dummy_ops)
 255                         continue;
 256                 new->hooks[j] = old->hooks[i];
 257                 new_ops[j] = (void *)orig_ops[i];
 258                 j++;
 259         }
 260         hooks_validate(new);
 261 out_assign:
 262         rcu_assign_pointer(*pp, new);
 263         return old;
 264 }
 265
 266 static struct nf_hook_entries __rcu **
 267 nf_hook_entry_head(struct net *net, int pf, unsigned int hooknum,
 268                    struct net_device *dev)
 269 {
 270         switch (pf) {
 271         case NFPROTO_NETDEV:
 272                 break;
 273 #ifdef CONFIG_NETFILTER_FAMILY_ARP
 274         case NFPROTO_ARP:
 275                 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_arp) <= hooknum))
 276                         return NULL;
 277                 return net->nf.hooks_arp + hooknum;
 278 #endif
 279 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 280         case NFPROTO_BRIDGE:
 281                 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_bridge) <= hooknum))
 282                         return NULL;
 283                 return net->nf.hooks_bridge + hooknum;
 284 #endif
 285 #ifdef CONFIG_NETFILTER_INGRESS
 286         case NFPROTO_INET:
 287                 if (WARN_ON_ONCE(hooknum != NF_INET_INGRESS))
 288                         return NULL;
 289                 if (!dev || dev_net(dev) != net) {
 290                         WARN_ON_ONCE(1);
 291                         return NULL;
 292                 }
 293                 return &dev->nf_hooks_ingress;
 294 #endif
 295         case NFPROTO_IPV4:
 296                 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_ipv4) <= hooknum))
 297                         return NULL;
 298                 return net->nf.hooks_ipv4 + hooknum;
 299         case NFPROTO_IPV6:
 300                 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_ipv6) <= hooknum))
 301                         return NULL;
 302                 return net->nf.hooks_ipv6 + hooknum;
 303 #if IS_ENABLED(CONFIG_DECNET)
 304         case NFPROTO_DECNET:
 305                 if (WARN_ON_ONCE(ARRAY_SIZE(net->nf.hooks_decnet) <= hooknum))
 306                         return NULL;
 307                 return net->nf.hooks_decnet + hooknum;
 308 #endif
 309         default:
 310                 WARN_ON_ONCE(1);
 311                 return NULL;
 312         }
 313
 314 #ifdef CONFIG_NETFILTER_INGRESS
 315         if (hooknum == NF_NETDEV_INGRESS) {
 316                 if (dev && dev_net(dev) == net)
 317                         return &dev->nf_hooks_ingress;
 318         }
 319 #endif
 320 #ifdef CONFIG_NETFILTER_EGRESS
 321         if (hooknum == NF_NETDEV_EGRESS) {
 322                 if (dev && dev_net(dev) == net)
 323                         return &dev->nf_hooks_egress;
 324         }
 325 #endif
 326         WARN_ON_ONCE(1);
 327         return NULL;
 328 }
 329
 330 static int nf_ingress_check(struct net *net, const struct nf_hook_ops *reg,
 331                             int hooknum)
 332 {
 333 #ifndef CONFIG_NETFILTER_INGRESS
 334         if (reg->hooknum == hooknum)
 335                 return -EOPNOTSUPP;
 336 #endif
 337         if (reg->hooknum != hooknum ||
 338             !reg->dev || dev_net(reg->dev) != net)
 339                 return -EINVAL;
 340
 341         return 0;
 342 }
 343
 344 static inline bool __maybe_unused nf_ingress_hook(const struct nf_hook_ops *reg,
 345                                                   int pf)
 346 {
 347         if ((pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS) ||
 348             (pf == NFPROTO_INET && reg->hooknum == NF_INET_INGRESS))
 349                 return true;
 350
 351         return false;
 352 }
 353
 354 static inline bool __maybe_unused nf_egress_hook(const struct nf_hook_ops *reg,
 355                                                  int pf)
 356 {
 357         return pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_EGRESS;
 358 }
 359
 360 static void nf_static_key_inc(const struct nf_hook_ops *reg, int pf)
 361 {
 362 #ifdef CONFIG_JUMP_LABEL
 363         int hooknum;
 364
 365         if (pf == NFPROTO_INET && reg->hooknum == NF_INET_INGRESS) {
 366                 pf = NFPROTO_NETDEV;
 367                 hooknum = NF_NETDEV_INGRESS;
 368         } else {
 369                 hooknum = reg->hooknum;
 370         }
 371         static_key_slow_inc(&nf_hooks_needed[pf][hooknum]);
 372 #endif
 373 }
 374
 375 static void nf_static_key_dec(const struct nf_hook_ops *reg, int pf)
 376 {
 377 #ifdef CONFIG_JUMP_LABEL
 378         int hooknum;
 379
 380         if (pf == NFPROTO_INET && reg->hooknum == NF_INET_INGRESS) {
 381                 pf = NFPROTO_NETDEV;
 382                 hooknum = NF_NETDEV_INGRESS;
 383         } else {
 384                 hooknum = reg->hooknum;
 385         }
 386         static_key_slow_dec(&nf_hooks_needed[pf][hooknum]);
 387 #endif
 388 }
 389
 390 static int __nf_register_net_hook(struct net *net, int pf,
 391                                   const struct nf_hook_ops *reg)
 392 {
 393         struct nf_hook_entries *p, *new_hooks;
 394         struct nf_hook_entries __rcu **pp;
 395         int err;
 396
 397         switch (pf) {
 398         case NFPROTO_NETDEV:
 399 #ifndef CONFIG_NETFILTER_INGRESS
 400                 if (reg->hooknum == NF_NETDEV_INGRESS)
 401                         return -EOPNOTSUPP;
 402 #endif
 403 #ifndef CONFIG_NETFILTER_EGRESS
 404                 if (reg->hooknum == NF_NETDEV_EGRESS)
 405                         return -EOPNOTSUPP;
 406 #endif
 407                 if ((reg->hooknum != NF_NETDEV_INGRESS &&
 408                      reg->hooknum != NF_NETDEV_EGRESS) ||
 409                     !reg->dev || dev_net(reg->dev) != net)
 410                         return -EINVAL;
 411                 break;
 412         case NFPROTO_INET:
 413                 if (reg->hooknum != NF_INET_INGRESS)
 414                         break;
 415
 416                 err = nf_ingress_check(net, reg, NF_INET_INGRESS);
 417                 if (err < 0)
 418                         return err;
 419                 break;
 420         }
 421
 422         pp = nf_hook_entry_head(net, pf, reg->hooknum, reg->dev);
 423         if (!pp)
 424                 return -EINVAL;
 425
 426         mutex_lock(&nf_hook_mutex);
 427
 428         p = nf_entry_dereference(*pp);
 429         new_hooks = nf_hook_entries_grow(p, reg);
 430
 431         if (!IS_ERR(new_hooks))
 432                 rcu_assign_pointer(*pp, new_hooks);
 433
 434         mutex_unlock(&nf_hook_mutex);
 435         if (IS_ERR(new_hooks))
 436                 return PTR_ERR(new_hooks);
 437
 438         hooks_validate(new_hooks);
 439 #ifdef CONFIG_NETFILTER_INGRESS
 440         if (nf_ingress_hook(reg, pf))
 441                 net_inc_ingress_queue();
 442 #endif
 443 #ifdef CONFIG_NETFILTER_EGRESS
 444         if (nf_egress_hook(reg, pf))
 445                 net_inc_egress_queue();
 446 #endif
 447         nf_static_key_inc(reg, pf);
 448
 449         BUG_ON(p == new_hooks);
 450         nf_hook_entries_free(p);
 451         return 0;
 452 }
 453
 454 /*
 455  * nf_remove_net_hook - remove a hook from blob
 456  *
 457  * @oldp: current address of hook blob
 458  * @unreg: hook to unregister
 459  *
 460  * This cannot fail, hook unregistration must always succeed.
 461  * Therefore replace the to-be-removed hook with a dummy hook.
 462  */
 463 static bool nf_remove_net_hook(struct nf_hook_entries *old,
 464                                const struct nf_hook_ops *unreg)
 465 {
 466         struct nf_hook_ops **orig_ops;
 467         unsigned int i;
 468
 469         orig_ops = nf_hook_entries_get_hook_ops(old);
 470         for (i = 0; i < old->num_hook_entries; i++) {
 471                 if (orig_ops[i] != unreg)
 472                         continue;
 473                 WRITE_ONCE(old->hooks[i].hook, accept_all);
 474                 WRITE_ONCE(orig_ops[i], (void *)&dummy_ops);
 475                 return true;
 476         }
 477
 478         return false;
 479 }
 480
 481 static void __nf_unregister_net_hook(struct net *net, int pf,
 482                                      const struct nf_hook_ops *reg)
 483 {
 484         struct nf_hook_entries __rcu **pp;
 485         struct nf_hook_entries *p;
 486
 487         pp = nf_hook_entry_head(net, pf, reg->hooknum, reg->dev);
 488         if (!pp)
 489                 return;
 490
 491         mutex_lock(&nf_hook_mutex);
 492
 493         p = nf_entry_dereference(*pp);
 494         if (WARN_ON_ONCE(!p)) {
 495                 mutex_unlock(&nf_hook_mutex);
 496                 return;
 497         }
 498
 499         if (nf_remove_net_hook(p, reg)) {
 500 #ifdef CONFIG_NETFILTER_INGRESS
 501                 if (nf_ingress_hook(reg, pf))
 502                         net_dec_ingress_queue();
 503 #endif
 504 #ifdef CONFIG_NETFILTER_EGRESS
 505                 if (nf_egress_hook(reg, pf))
 506                         net_dec_egress_queue();
 507 #endif
 508                 nf_static_key_dec(reg, pf);
 509         } else {
 510                 WARN_ONCE(1, "hook not found, pf %d num %d", pf, reg->hooknum);
 511         }
 512
 513         p = __nf_hook_entries_try_shrink(p, pp);
 514         mutex_unlock(&nf_hook_mutex);
 515         if (!p)
 516                 return;
 517
 518         nf_queue_nf_hook_drop(net);
 519         nf_hook_entries_free(p);
 520 }
 521
 522 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 523 {
 524         if (reg->pf == NFPROTO_INET) {
 525                 if (reg->hooknum == NF_INET_INGRESS) {
 526                         __nf_unregister_net_hook(net, NFPROTO_INET, reg);
 527                 } else {
 528                         __nf_unregister_net_hook(net, NFPROTO_IPV4, reg);
 529                         __nf_unregister_net_hook(net, NFPROTO_IPV6, reg);
 530                 }
 531         } else {
 532                 __nf_unregister_net_hook(net, reg->pf, reg);
 533         }
 534 }
 535 EXPORT_SYMBOL(nf_unregister_net_hook);
 536
 537 void nf_hook_entries_delete_raw(struct nf_hook_entries __rcu **pp,
 538                                 const struct nf_hook_ops *reg)
 539 {
 540         struct nf_hook_entries *p;
 541
 542         p = rcu_dereference_raw(*pp);
 543         if (nf_remove_net_hook(p, reg)) {
 544                 p = __nf_hook_entries_try_shrink(p, pp);
 545                 nf_hook_entries_free(p);
 546         }
 547 }
 548 EXPORT_SYMBOL_GPL(nf_hook_entries_delete_raw);
 549
 550 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
 551 {
 552         int err;
 553
 554         if (reg->pf == NFPROTO_INET) {
 555                 if (reg->hooknum == NF_INET_INGRESS) {
 556                         err = __nf_register_net_hook(net, NFPROTO_INET, reg);
 557                         if (err < 0)
 558                                 return err;
 559                 } else {
 560                         err = __nf_register_net_hook(net, NFPROTO_IPV4, reg);
 561                         if (err < 0)
 562                                 return err;
 563
 564                         err = __nf_register_net_hook(net, NFPROTO_IPV6, reg);
 565                         if (err < 0) {
 566                                 __nf_unregister_net_hook(net, NFPROTO_IPV4, reg);
 567                                 return err;
 568                         }
 569                 }
 570         } else {
 571                 err = __nf_register_net_hook(net, reg->pf, reg);
 572                 if (err < 0)
 573                         return err;
 574         }
 575
 576         return 0;
 577 }
 578 EXPORT_SYMBOL(nf_register_net_hook);
 579
 580 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 581                           unsigned int n)
 582 {
 583         unsigned int i;
 584         int err = 0;
 585
 586         for (i = 0; i < n; i++) {
 587                 err = nf_register_net_hook(net, &reg[i]);
 588                 if (err)
 589                         goto err;
 590         }
 591         return err;
 592
 593 err:
 594         if (i > 0)
 595                 nf_unregister_net_hooks(net, reg, i);
 596         return err;
 597 }
 598 EXPORT_SYMBOL(nf_register_net_hooks);
 599
 600 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 601                              unsigned int hookcount)
 602 {
 603         unsigned int i;
 604
 605         for (i = 0; i < hookcount; i++)
 606                 nf_unregister_net_hook(net, &reg[i]);
 607 }
 608 EXPORT_SYMBOL(nf_unregister_net_hooks);
 609
 610 /* Returns 1 if okfn() needs to be executed by the caller,
 611  * -EPERM for NF_DROP, 0 otherwise.  Caller must hold rcu_read_lock. */
 612 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
 613                  const struct nf_hook_entries *e, unsigned int s)
 614 {
 615         unsigned int verdict;
 616         int ret;
 617
 618         for (; s < e->num_hook_entries; s++) {
 619                 verdict = nf_hook_entry_hookfn(&e->hooks[s], skb, state);
 620                 switch (verdict & NF_VERDICT_MASK) {
 621                 case NF_ACCEPT:
 622                         break;
 623                 case NF_DROP:
 624                         kfree_skb(skb);
 625                         ret = NF_DROP_GETERR(verdict);
 626                         if (ret == 0)
 627                                 ret = -EPERM;
 628                         return ret;
 629                 case NF_QUEUE:
 630                         ret = nf_queue(skb, state, s, verdict);
 631                         if (ret == 1)
 632                                 continue;
 633                         return ret;
 634                 default:
 635                         /* Implicit handling for NF_STOLEN, as well as any other
 636                          * non conventional verdicts.
 637                          */
 638                         return 0;
 639                 }
 640         }
 641
 642         return 1;
 643 }
 644 EXPORT_SYMBOL(nf_hook_slow);
 645
 646 void nf_hook_slow_list(struct list_head *head, struct nf_hook_state *state,
 647                        const struct nf_hook_entries *e)
 648 {
 649         struct sk_buff *skb, *next;
 650         struct list_head sublist;
 651         int ret;
 652
 653         INIT_LIST_HEAD(&sublist);
 654
 655         list_for_each_entry_safe(skb, next, head, list) {
 656                 skb_list_del_init(skb);
 657                 ret = nf_hook_slow(skb, state, e, 0);
 658                 if (ret == 1)
 659                         list_add_tail(&skb->list, &sublist);
 660         }
 661         /* Put passed packets back on main list */
 662         list_splice(&sublist, head);
 663 }
 664 EXPORT_SYMBOL(nf_hook_slow_list);
 665
 666 /* This needs to be compiled in any case to avoid dependencies between the
 667  * nfnetlink_queue code and nf_conntrack.
 668  */
 669 const struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly;
 670 EXPORT_SYMBOL_GPL(nfnl_ct_hook);
 671
 672 const struct nf_ct_hook __rcu *nf_ct_hook __read_mostly;
 673 EXPORT_SYMBOL_GPL(nf_ct_hook);
 674
 675 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 676 const struct nf_nat_hook __rcu *nf_nat_hook __read_mostly;
 677 EXPORT_SYMBOL_GPL(nf_nat_hook);
 678
 679 /* This does not belong here, but locally generated errors need it if connection
 680  * tracking in use: without this, connection may not be in hash table, and hence
 681  * manufactured ICMP or RST packets will not be associated with it.
 682  */
 683 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 684 {
 685         const struct nf_ct_hook *ct_hook;
 686
 687         if (skb->_nfct) {
 688                 rcu_read_lock();
 689                 ct_hook = rcu_dereference(nf_ct_hook);
 690                 if (ct_hook)
 691                         ct_hook->attach(new, skb);
 692                 rcu_read_unlock();
 693         }
 694 }
 695 EXPORT_SYMBOL(nf_ct_attach);
 696
 697 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 698 {
 699         const struct nf_ct_hook *ct_hook;
 700
 701         rcu_read_lock();
 702         ct_hook = rcu_dereference(nf_ct_hook);
 703         BUG_ON(ct_hook == NULL);
 704         ct_hook->destroy(nfct);
 705         rcu_read_unlock();
 706 }
 707 EXPORT_SYMBOL(nf_conntrack_destroy);
 708
 709 bool nf_ct_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
 710                          const struct sk_buff *skb)
 711 {
 712         const struct nf_ct_hook *ct_hook;
 713         bool ret = false;
 714
 715         rcu_read_lock();
 716         ct_hook = rcu_dereference(nf_ct_hook);
 717         if (ct_hook)
 718                 ret = ct_hook->get_tuple_skb(dst_tuple, skb);
 719         rcu_read_unlock();
 720         return ret;
 721 }
 722 EXPORT_SYMBOL(nf_ct_get_tuple_skb);
 723
 724 /* Built-in default zone used e.g. by modules. */
 725 const struct nf_conntrack_zone nf_ct_zone_dflt = {
 726         .id     = NF_CT_DEFAULT_ZONE_ID,
 727         .dir    = NF_CT_DEFAULT_ZONE_DIR,
 728 };
 729 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt);
 730 #endif /* CONFIG_NF_CONNTRACK */
 731
 732 static void __net_init
 733 __netfilter_net_init(struct nf_hook_entries __rcu **e, int max)
 734 {
 735         int h;
 736
 737         for (h = 0; h < max; h++)
 738                 RCU_INIT_POINTER(e[h], NULL);
 739 }
 740
 741 static int __net_init netfilter_net_init(struct net *net)
 742 {
 743         __netfilter_net_init(net->nf.hooks_ipv4, ARRAY_SIZE(net->nf.hooks_ipv4));
 744         __netfilter_net_init(net->nf.hooks_ipv6, ARRAY_SIZE(net->nf.hooks_ipv6));
 745 #ifdef CONFIG_NETFILTER_FAMILY_ARP
 746         __netfilter_net_init(net->nf.hooks_arp, ARRAY_SIZE(net->nf.hooks_arp));
 747 #endif
 748 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 749         __netfilter_net_init(net->nf.hooks_bridge, ARRAY_SIZE(net->nf.hooks_bridge));
 750 #endif
 751 #if IS_ENABLED(CONFIG_DECNET)
 752         __netfilter_net_init(net->nf.hooks_decnet, ARRAY_SIZE(net->nf.hooks_decnet));
 753 #endif
 754
 755 #ifdef CONFIG_PROC_FS
 756         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 757                                                 net->proc_net);
 758         if (!net->nf.proc_netfilter) {
 759                 if (!net_eq(net, &init_net))
 760                         pr_err("cannot create netfilter proc entry");
 761
 762                 return -ENOMEM;
 763         }
 764 #endif
 765
 766         return 0;
 767 }
 768
 769 static void __net_exit netfilter_net_exit(struct net *net)
 770 {
 771         remove_proc_entry("netfilter", net->proc_net);
 772 }
 773
 774 static struct pernet_operations netfilter_net_ops = {
 775         .init = netfilter_net_init,
 776         .exit = netfilter_net_exit,
 777 };
 778
 779 int __init netfilter_init(void)
 780 {
 781         int ret;
 782
 783         ret = register_pernet_subsys(&netfilter_net_ops);
 784         if (ret < 0)
 785                 goto err;
 786
 787         ret = netfilter_log_init();
 788         if (ret < 0)
 789                 goto err_pernet;
 790
 791         return 0;
 792 err_pernet:
 793         unregister_pernet_subsys(&netfilter_net_ops);
 794 err:
 795         return ret;
 796 }