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