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