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  * Rusty Russell (C)2000 -- This code is GPL.
   8  * Patrick McHardy (c) 2006-2012
   9  */
  10 #include <linux/kernel.h>
  11 #include <linux/netfilter.h>
  12 #include <net/protocol.h>
  13 #include <linux/init.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/wait.h>
  16 #include <linux/module.h>
  17 #include <linux/interrupt.h>
  18 #include <linux/if.h>
  19 #include <linux/netdevice.h>
  20 #include <linux/netfilter_ipv6.h>
  21 #include <linux/inetdevice.h>
  22 #include <linux/proc_fs.h>
  23 #include <linux/mutex.h>
  24 #include <linux/slab.h>
  25 #include <linux/rcupdate.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28
  29 #include "nf_internals.h"
  30
  31 static DEFINE_MUTEX(afinfo_mutex);
  32
  33 const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO] __read_mostly;
  34 EXPORT_SYMBOL(nf_afinfo);
  35 const struct nf_ipv6_ops __rcu *nf_ipv6_ops __read_mostly;
  36 EXPORT_SYMBOL_GPL(nf_ipv6_ops);
  37
  38 DEFINE_PER_CPU(bool, nf_skb_duplicated);
  39 EXPORT_SYMBOL_GPL(nf_skb_duplicated);
  40
  41 int nf_register_afinfo(const struct nf_afinfo *afinfo)
  42 {
  43         mutex_lock(&afinfo_mutex);
  44         RCU_INIT_POINTER(nf_afinfo[afinfo->family], afinfo);
  45         mutex_unlock(&afinfo_mutex);
  46         return 0;
  47 }
  48 EXPORT_SYMBOL_GPL(nf_register_afinfo);
  49
  50 void nf_unregister_afinfo(const struct nf_afinfo *afinfo)
  51 {
  52         mutex_lock(&afinfo_mutex);
  53         RCU_INIT_POINTER(nf_afinfo[afinfo->family], NULL);
  54         mutex_unlock(&afinfo_mutex);
  55         synchronize_rcu();
  56 }
  57 EXPORT_SYMBOL_GPL(nf_unregister_afinfo);
  58
  59 #ifdef HAVE_JUMP_LABEL
  60 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
  61 EXPORT_SYMBOL(nf_hooks_needed);
  62 #endif
  63
  64 static DEFINE_MUTEX(nf_hook_mutex);
  65 #define nf_entry_dereference(e) \
  66         rcu_dereference_protected(e, lockdep_is_held(&nf_hook_mutex))
  67
  68 static struct nf_hook_entry *nf_hook_entry_head(struct net *net,
  69                                                 const struct nf_hook_ops *reg)
  70 {
  71         struct nf_hook_entry *hook_head = NULL;
  72
  73         if (reg->pf != NFPROTO_NETDEV)
  74                 hook_head = nf_entry_dereference(net->nf.hooks[reg->pf]
  75                                                  [reg->hooknum]);
  76         else if (reg->hooknum == NF_NETDEV_INGRESS) {
  77 #ifdef CONFIG_NETFILTER_INGRESS
  78                 if (reg->dev && dev_net(reg->dev) == net)
  79                         hook_head =
  80                                 nf_entry_dereference(
  81                                         reg->dev->nf_hooks_ingress);
  82 #endif
  83         }
  84         return hook_head;
  85 }
  86
  87 /* must hold nf_hook_mutex */
  88 static void nf_set_hooks_head(struct net *net, const struct nf_hook_ops *reg,
  89                               struct nf_hook_entry *entry)
  90 {
  91         switch (reg->pf) {
  92         case NFPROTO_NETDEV:
  93 #ifdef CONFIG_NETFILTER_INGRESS
  94                 /* We already checked in nf_register_net_hook() that this is
  95                  * used from ingress.
  96                  */
  97                 rcu_assign_pointer(reg->dev->nf_hooks_ingress, entry);
  98 #endif
  99                 break;
 100         default:
 101                 rcu_assign_pointer(net->nf.hooks[reg->pf][reg->hooknum],
 102                                    entry);
 103                 break;
 104         }
 105 }
 106
 107 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
 108 {
 109         struct nf_hook_entry *hooks_entry;
 110         struct nf_hook_entry *entry;
 111
 112         if (reg->pf == NFPROTO_NETDEV) {
 113 #ifndef CONFIG_NETFILTER_INGRESS
 114                 if (reg->hooknum == NF_NETDEV_INGRESS)
 115                         return -EOPNOTSUPP;
 116 #endif
 117                 if (reg->hooknum != NF_NETDEV_INGRESS ||
 118                     !reg->dev || dev_net(reg->dev) != net)
 119                         return -EINVAL;
 120         }
 121
 122         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 123         if (!entry)
 124                 return -ENOMEM;
 125
 126         entry->orig_ops = reg;
 127         entry->ops      = *reg;
 128         entry->next     = NULL;
 129
 130         mutex_lock(&nf_hook_mutex);
 131         hooks_entry = nf_hook_entry_head(net, reg);
 132
 133         if (hooks_entry && hooks_entry->orig_ops->priority > reg->priority) {
 134                 /* This is the case where we need to insert at the head */
 135                 entry->next = hooks_entry;
 136                 hooks_entry = NULL;
 137         }
 138
 139         while (hooks_entry &&
 140                 reg->priority >= hooks_entry->orig_ops->priority &&
 141                 nf_entry_dereference(hooks_entry->next)) {
 142                 hooks_entry = nf_entry_dereference(hooks_entry->next);
 143         }
 144
 145         if (hooks_entry) {
 146                 entry->next = nf_entry_dereference(hooks_entry->next);
 147                 rcu_assign_pointer(hooks_entry->next, entry);
 148         } else {
 149                 nf_set_hooks_head(net, reg, entry);
 150         }
 151
 152         mutex_unlock(&nf_hook_mutex);
 153 #ifdef CONFIG_NETFILTER_INGRESS
 154         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 155                 net_inc_ingress_queue();
 156 #endif
 157 #ifdef HAVE_JUMP_LABEL
 158         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
 159 #endif
 160         return 0;
 161 }
 162 EXPORT_SYMBOL(nf_register_net_hook);
 163
 164 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 165 {
 166         struct nf_hook_entry *hooks_entry;
 167
 168         mutex_lock(&nf_hook_mutex);
 169         hooks_entry = nf_hook_entry_head(net, reg);
 170         if (hooks_entry && hooks_entry->orig_ops == reg) {
 171                 nf_set_hooks_head(net, reg,
 172                                   nf_entry_dereference(hooks_entry->next));
 173                 goto unlock;
 174         }
 175         while (hooks_entry && nf_entry_dereference(hooks_entry->next)) {
 176                 struct nf_hook_entry *next =
 177                         nf_entry_dereference(hooks_entry->next);
 178                 struct nf_hook_entry *nnext;
 179
 180                 if (next->orig_ops != reg) {
 181                         hooks_entry = next;
 182                         continue;
 183                 }
 184                 nnext = nf_entry_dereference(next->next);
 185                 rcu_assign_pointer(hooks_entry->next, nnext);
 186                 hooks_entry = next;
 187                 break;
 188         }
 189
 190 unlock:
 191         mutex_unlock(&nf_hook_mutex);
 192         if (!hooks_entry) {
 193                 WARN(1, "nf_unregister_net_hook: hook not found!\n");
 194                 return;
 195         }
 196 #ifdef CONFIG_NETFILTER_INGRESS
 197         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 198                 net_dec_ingress_queue();
 199 #endif
 200 #ifdef HAVE_JUMP_LABEL
 201         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
 202 #endif
 203         synchronize_net();
 204         nf_queue_nf_hook_drop(net, hooks_entry);
 205         /* other cpu might still process nfqueue verdict that used reg */
 206         synchronize_net();
 207         kfree(hooks_entry);
 208 }
 209 EXPORT_SYMBOL(nf_unregister_net_hook);
 210
 211 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 212                           unsigned int n)
 213 {
 214         unsigned int i;
 215         int err = 0;
 216
 217         for (i = 0; i < n; i++) {
 218                 err = nf_register_net_hook(net, &reg[i]);
 219                 if (err)
 220                         goto err;
 221         }
 222         return err;
 223
 224 err:
 225         if (i > 0)
 226                 nf_unregister_net_hooks(net, reg, i);
 227         return err;
 228 }
 229 EXPORT_SYMBOL(nf_register_net_hooks);
 230
 231 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 232                              unsigned int n)
 233 {
 234         while (n-- > 0)
 235                 nf_unregister_net_hook(net, &reg[n]);
 236 }
 237 EXPORT_SYMBOL(nf_unregister_net_hooks);
 238
 239 static LIST_HEAD(nf_hook_list);
 240
 241 static int _nf_register_hook(struct nf_hook_ops *reg)
 242 {
 243         struct net *net, *last;
 244         int ret;
 245
 246         for_each_net(net) {
 247                 ret = nf_register_net_hook(net, reg);
 248                 if (ret && ret != -ENOENT)
 249                         goto rollback;
 250         }
 251         list_add_tail(&reg->list, &nf_hook_list);
 252
 253         return 0;
 254 rollback:
 255         last = net;
 256         for_each_net(net) {
 257                 if (net == last)
 258                         break;
 259                 nf_unregister_net_hook(net, reg);
 260         }
 261         return ret;
 262 }
 263
 264 int nf_register_hook(struct nf_hook_ops *reg)
 265 {
 266         int ret;
 267
 268         rtnl_lock();
 269         ret = _nf_register_hook(reg);
 270         rtnl_unlock();
 271
 272         return ret;
 273 }
 274 EXPORT_SYMBOL(nf_register_hook);
 275
 276 static void _nf_unregister_hook(struct nf_hook_ops *reg)
 277 {
 278         struct net *net;
 279
 280         list_del(&reg->list);
 281         for_each_net(net)
 282                 nf_unregister_net_hook(net, reg);
 283 }
 284
 285 void nf_unregister_hook(struct nf_hook_ops *reg)
 286 {
 287         rtnl_lock();
 288         _nf_unregister_hook(reg);
 289         rtnl_unlock();
 290 }
 291 EXPORT_SYMBOL(nf_unregister_hook);
 292
 293 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 294 {
 295         unsigned int i;
 296         int err = 0;
 297
 298         for (i = 0; i < n; i++) {
 299                 err = nf_register_hook(&reg[i]);
 300                 if (err)
 301                         goto err;
 302         }
 303         return err;
 304
 305 err:
 306         if (i > 0)
 307                 nf_unregister_hooks(reg, i);
 308         return err;
 309 }
 310 EXPORT_SYMBOL(nf_register_hooks);
 311
 312 /* Caller MUST take rtnl_lock() */
 313 int _nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 314 {
 315         unsigned int i;
 316         int err = 0;
 317
 318         for (i = 0; i < n; i++) {
 319                 err = _nf_register_hook(&reg[i]);
 320                 if (err)
 321                         goto err;
 322         }
 323         return err;
 324
 325 err:
 326         if (i > 0)
 327                 _nf_unregister_hooks(reg, i);
 328         return err;
 329 }
 330 EXPORT_SYMBOL(_nf_register_hooks);
 331
 332 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 333 {
 334         while (n-- > 0)
 335                 nf_unregister_hook(&reg[n]);
 336 }
 337 EXPORT_SYMBOL(nf_unregister_hooks);
 338
 339 /* Caller MUST take rtnl_lock */
 340 void _nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 341 {
 342         while (n-- > 0)
 343                 _nf_unregister_hook(&reg[n]);
 344 }
 345 EXPORT_SYMBOL(_nf_unregister_hooks);
 346
 347 unsigned int nf_iterate(struct sk_buff *skb,
 348                         struct nf_hook_state *state,
 349                         struct nf_hook_entry **entryp)
 350 {
 351         unsigned int verdict;
 352
 353         /*
 354          * The caller must not block between calls to this
 355          * function because of risk of continuing from deleted element.
 356          */
 357         while (*entryp) {
 358                 if (state->thresh > (*entryp)->ops.priority) {
 359                         *entryp = rcu_dereference((*entryp)->next);
 360                         continue;
 361                 }
 362
 363                 /* Optimization: we don't need to hold module
 364                    reference here, since function can't sleep. --RR */
 365 repeat:
 366                 verdict = (*entryp)->ops.hook((*entryp)->ops.priv, skb, state);
 367                 if (verdict != NF_ACCEPT) {
 368 #ifdef CONFIG_NETFILTER_DEBUG
 369                         if (unlikely((verdict & NF_VERDICT_MASK)
 370                                                         > NF_MAX_VERDICT)) {
 371                                 NFDEBUG("Evil return from %p(%u).\n",
 372                                         (*entryp)->ops.hook, state->hook);
 373                                 *entryp = rcu_dereference((*entryp)->next);
 374                                 continue;
 375                         }
 376 #endif
 377                         if (verdict != NF_REPEAT)
 378                                 return verdict;
 379                         goto repeat;
 380                 }
 381                 *entryp = rcu_dereference((*entryp)->next);
 382         }
 383         return NF_ACCEPT;
 384 }
 385
 386
 387 /* Returns 1 if okfn() needs to be executed by the caller,
 388  * -EPERM for NF_DROP, 0 otherwise.  Caller must hold rcu_read_lock. */
 389 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state)
 390 {
 391         struct nf_hook_entry *entry;
 392         unsigned int verdict;
 393         int ret = 0;
 394
 395         entry = rcu_dereference(state->hook_entries);
 396 next_hook:
 397         verdict = nf_iterate(skb, state, &entry);
 398         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 399                 ret = 1;
 400         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 401                 kfree_skb(skb);
 402                 ret = NF_DROP_GETERR(verdict);
 403                 if (ret == 0)
 404                         ret = -EPERM;
 405         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 406                 int err;
 407
 408                 RCU_INIT_POINTER(state->hook_entries, entry);
 409                 err = nf_queue(skb, state, verdict >> NF_VERDICT_QBITS);
 410                 if (err < 0) {
 411                         if (err == -ESRCH &&
 412                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 413                                 goto next_hook;
 414                         kfree_skb(skb);
 415                 }
 416         }
 417         return ret;
 418 }
 419 EXPORT_SYMBOL(nf_hook_slow);
 420
 421
 422 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 423 {
 424         if (writable_len > skb->len)
 425                 return 0;
 426
 427         /* Not exclusive use of packet?  Must copy. */
 428         if (!skb_cloned(skb)) {
 429                 if (writable_len <= skb_headlen(skb))
 430                         return 1;
 431         } else if (skb_clone_writable(skb, writable_len))
 432                 return 1;
 433
 434         if (writable_len <= skb_headlen(skb))
 435                 writable_len = 0;
 436         else
 437                 writable_len -= skb_headlen(skb);
 438
 439         return !!__pskb_pull_tail(skb, writable_len);
 440 }
 441 EXPORT_SYMBOL(skb_make_writable);
 442
 443 /* This needs to be compiled in any case to avoid dependencies between the
 444  * nfnetlink_queue code and nf_conntrack.
 445  */
 446 struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly;
 447 EXPORT_SYMBOL_GPL(nfnl_ct_hook);
 448
 449 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 450 /* This does not belong here, but locally generated errors need it if connection
 451    tracking in use: without this, connection may not be in hash table, and hence
 452    manufactured ICMP or RST packets will not be associated with it. */
 453 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 454                 __rcu __read_mostly;
 455 EXPORT_SYMBOL(ip_ct_attach);
 456
 457 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 458 {
 459         void (*attach)(struct sk_buff *, const struct sk_buff *);
 460
 461         if (skb->nfct) {
 462                 rcu_read_lock();
 463                 attach = rcu_dereference(ip_ct_attach);
 464                 if (attach)
 465                         attach(new, skb);
 466                 rcu_read_unlock();
 467         }
 468 }
 469 EXPORT_SYMBOL(nf_ct_attach);
 470
 471 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 472 EXPORT_SYMBOL(nf_ct_destroy);
 473
 474 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 475 {
 476         void (*destroy)(struct nf_conntrack *);
 477
 478         rcu_read_lock();
 479         destroy = rcu_dereference(nf_ct_destroy);
 480         BUG_ON(destroy == NULL);
 481         destroy(nfct);
 482         rcu_read_unlock();
 483 }
 484 EXPORT_SYMBOL(nf_conntrack_destroy);
 485
 486 /* Built-in default zone used e.g. by modules. */
 487 const struct nf_conntrack_zone nf_ct_zone_dflt = {
 488         .id     = NF_CT_DEFAULT_ZONE_ID,
 489         .dir    = NF_CT_DEFAULT_ZONE_DIR,
 490 };
 491 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt);
 492 #endif /* CONFIG_NF_CONNTRACK */
 493
 494 #ifdef CONFIG_NF_NAT_NEEDED
 495 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 496 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 497 #endif
 498
 499 static int nf_register_hook_list(struct net *net)
 500 {
 501         struct nf_hook_ops *elem;
 502         int ret;
 503
 504         rtnl_lock();
 505         list_for_each_entry(elem, &nf_hook_list, list) {
 506                 ret = nf_register_net_hook(net, elem);
 507                 if (ret && ret != -ENOENT)
 508                         goto out_undo;
 509         }
 510         rtnl_unlock();
 511         return 0;
 512
 513 out_undo:
 514         list_for_each_entry_continue_reverse(elem, &nf_hook_list, list)
 515                 nf_unregister_net_hook(net, elem);
 516         rtnl_unlock();
 517         return ret;
 518 }
 519
 520 static void nf_unregister_hook_list(struct net *net)
 521 {
 522         struct nf_hook_ops *elem;
 523
 524         rtnl_lock();
 525         list_for_each_entry(elem, &nf_hook_list, list)
 526                 nf_unregister_net_hook(net, elem);
 527         rtnl_unlock();
 528 }
 529
 530 static int __net_init netfilter_net_init(struct net *net)
 531 {
 532         int i, h, ret;
 533
 534         for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
 535                 for (h = 0; h < NF_MAX_HOOKS; h++)
 536                         RCU_INIT_POINTER(net->nf.hooks[i][h], NULL);
 537         }
 538
 539 #ifdef CONFIG_PROC_FS
 540         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 541                                                 net->proc_net);
 542         if (!net->nf.proc_netfilter) {
 543                 if (!net_eq(net, &init_net))
 544                         pr_err("cannot create netfilter proc entry");
 545
 546                 return -ENOMEM;
 547         }
 548 #endif
 549         ret = nf_register_hook_list(net);
 550         if (ret)
 551                 remove_proc_entry("netfilter", net->proc_net);
 552
 553         return ret;
 554 }
 555
 556 static void __net_exit netfilter_net_exit(struct net *net)
 557 {
 558         nf_unregister_hook_list(net);
 559         remove_proc_entry("netfilter", net->proc_net);
 560 }
 561
 562 static struct pernet_operations netfilter_net_ops = {
 563         .init = netfilter_net_init,
 564         .exit = netfilter_net_exit,
 565 };
 566
 567 int __init netfilter_init(void)
 568 {
 569         int ret;
 570
 571         ret = register_pernet_subsys(&netfilter_net_ops);
 572         if (ret < 0)
 573                 goto err;
 574
 575         ret = netfilter_log_init();
 576         if (ret < 0)
 577                 goto err_pernet;
 578
 579         return 0;
 580 err_pernet:
 581         unregister_pernet_subsys(&netfilter_net_ops);
 582 err:
 583         return ret;
 584 }