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         nf_hook_entry_init(entry, reg);
 106
 107         mutex_lock(&nf_hook_mutex);
 108
 109         /* Find the spot in the list */
 110         for (; (p = nf_entry_dereference(*pp)) != NULL; pp = &p->next) {
 111                 if (reg->priority < nf_hook_entry_priority(p))
 112                         break;
 113         }
 114         rcu_assign_pointer(entry->next, p);
 115         rcu_assign_pointer(*pp, entry);
 116
 117         mutex_unlock(&nf_hook_mutex);
 118 #ifdef CONFIG_NETFILTER_INGRESS
 119         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 120                 net_inc_ingress_queue();
 121 #endif
 122 #ifdef HAVE_JUMP_LABEL
 123         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
 124 #endif
 125         return 0;
 126 }
 127 EXPORT_SYMBOL(nf_register_net_hook);
 128
 129 static struct nf_hook_entry *
 130 __nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 131 {
 132         struct nf_hook_entry __rcu **pp;
 133         struct nf_hook_entry *p;
 134
 135         pp = nf_hook_entry_head(net, reg);
 136         if (WARN_ON_ONCE(!pp))
 137                 return NULL;
 138
 139         mutex_lock(&nf_hook_mutex);
 140         for (; (p = nf_entry_dereference(*pp)) != NULL; pp = &p->next) {
 141                 if (nf_hook_entry_ops(p) == reg) {
 142                         rcu_assign_pointer(*pp, p->next);
 143                         break;
 144                 }
 145         }
 146         mutex_unlock(&nf_hook_mutex);
 147         if (!p) {
 148                 WARN(1, "nf_unregister_net_hook: hook not found!\n");
 149                 return NULL;
 150         }
 151 #ifdef CONFIG_NETFILTER_INGRESS
 152         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 153                 net_dec_ingress_queue();
 154 #endif
 155 #ifdef HAVE_JUMP_LABEL
 156         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
 157 #endif
 158
 159         return p;
 160 }
 161
 162 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 163 {
 164         struct nf_hook_entry *p = __nf_unregister_net_hook(net, reg);
 165         unsigned int nfq;
 166
 167         if (!p)
 168                 return;
 169
 170         synchronize_net();
 171
 172         /* other cpu might still process nfqueue verdict that used reg */
 173         nfq = nf_queue_nf_hook_drop(net);
 174         if (nfq)
 175                 synchronize_net();
 176         kfree(p);
 177 }
 178 EXPORT_SYMBOL(nf_unregister_net_hook);
 179
 180 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 181                           unsigned int n)
 182 {
 183         unsigned int i;
 184         int err = 0;
 185
 186         for (i = 0; i < n; i++) {
 187                 err = nf_register_net_hook(net, &reg[i]);
 188                 if (err)
 189                         goto err;
 190         }
 191         return err;
 192
 193 err:
 194         if (i > 0)
 195                 nf_unregister_net_hooks(net, reg, i);
 196         return err;
 197 }
 198 EXPORT_SYMBOL(nf_register_net_hooks);
 199
 200 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 201                              unsigned int hookcount)
 202 {
 203         struct nf_hook_entry *to_free[16];
 204         unsigned int i, n, nfq;
 205
 206         do {
 207                 n = min_t(unsigned int, hookcount, ARRAY_SIZE(to_free));
 208
 209                 for (i = 0; i < n; i++)
 210                         to_free[i] = __nf_unregister_net_hook(net, &reg[i]);
 211
 212                 synchronize_net();
 213
 214                 /* need 2nd synchronize_net() if nfqueue is used, skb
 215                  * can get reinjected right before nf_queue_hook_drop()
 216                  */
 217                 nfq = nf_queue_nf_hook_drop(net);
 218                 if (nfq)
 219                         synchronize_net();
 220
 221                 for (i = 0; i < n; i++)
 222                         kfree(to_free[i]);
 223
 224                 reg += n;
 225                 hookcount -= n;
 226         } while (hookcount > 0);
 227 }
 228 EXPORT_SYMBOL(nf_unregister_net_hooks);
 229
 230 static LIST_HEAD(nf_hook_list);
 231
 232 static int _nf_register_hook(struct nf_hook_ops *reg)
 233 {
 234         struct net *net, *last;
 235         int ret;
 236
 237         for_each_net(net) {
 238                 ret = nf_register_net_hook(net, reg);
 239                 if (ret && ret != -ENOENT)
 240                         goto rollback;
 241         }
 242         list_add_tail(&reg->list, &nf_hook_list);
 243
 244         return 0;
 245 rollback:
 246         last = net;
 247         for_each_net(net) {
 248                 if (net == last)
 249                         break;
 250                 nf_unregister_net_hook(net, reg);
 251         }
 252         return ret;
 253 }
 254
 255 int nf_register_hook(struct nf_hook_ops *reg)
 256 {
 257         int ret;
 258
 259         rtnl_lock();
 260         ret = _nf_register_hook(reg);
 261         rtnl_unlock();
 262
 263         return ret;
 264 }
 265 EXPORT_SYMBOL(nf_register_hook);
 266
 267 static void _nf_unregister_hook(struct nf_hook_ops *reg)
 268 {
 269         struct net *net;
 270
 271         list_del(&reg->list);
 272         for_each_net(net)
 273                 nf_unregister_net_hook(net, reg);
 274 }
 275
 276 void nf_unregister_hook(struct nf_hook_ops *reg)
 277 {
 278         rtnl_lock();
 279         _nf_unregister_hook(reg);
 280         rtnl_unlock();
 281 }
 282 EXPORT_SYMBOL(nf_unregister_hook);
 283
 284 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 285 {
 286         unsigned int i;
 287         int err = 0;
 288
 289         for (i = 0; i < n; i++) {
 290                 err = nf_register_hook(&reg[i]);
 291                 if (err)
 292                         goto err;
 293         }
 294         return err;
 295
 296 err:
 297         if (i > 0)
 298                 nf_unregister_hooks(reg, i);
 299         return err;
 300 }
 301 EXPORT_SYMBOL(nf_register_hooks);
 302
 303 /* Caller MUST take rtnl_lock() */
 304 int _nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 305 {
 306         unsigned int i;
 307         int err = 0;
 308
 309         for (i = 0; i < n; i++) {
 310                 err = _nf_register_hook(&reg[i]);
 311                 if (err)
 312                         goto err;
 313         }
 314         return err;
 315
 316 err:
 317         if (i > 0)
 318                 _nf_unregister_hooks(reg, i);
 319         return err;
 320 }
 321 EXPORT_SYMBOL(_nf_register_hooks);
 322
 323 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 324 {
 325         while (n-- > 0)
 326                 nf_unregister_hook(&reg[n]);
 327 }
 328 EXPORT_SYMBOL(nf_unregister_hooks);
 329
 330 /* Caller MUST take rtnl_lock */
 331 void _nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 332 {
 333         while (n-- > 0)
 334                 _nf_unregister_hook(&reg[n]);
 335 }
 336 EXPORT_SYMBOL(_nf_unregister_hooks);
 337
 338 /* Returns 1 if okfn() needs to be executed by the caller,
 339  * -EPERM for NF_DROP, 0 otherwise.  Caller must hold rcu_read_lock. */
 340 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
 341                  struct nf_hook_entry *entry)
 342 {
 343         unsigned int verdict;
 344         int ret;
 345
 346         do {
 347                 verdict = nf_hook_entry_hookfn(entry, skb, state);
 348                 switch (verdict & NF_VERDICT_MASK) {
 349                 case NF_ACCEPT:
 350                         entry = rcu_dereference(entry->next);
 351                         break;
 352                 case NF_DROP:
 353                         kfree_skb(skb);
 354                         ret = NF_DROP_GETERR(verdict);
 355                         if (ret == 0)
 356                                 ret = -EPERM;
 357                         return ret;
 358                 case NF_QUEUE:
 359                         ret = nf_queue(skb, state, &entry, verdict);
 360                         if (ret == 1 && entry)
 361                                 continue;
 362                         return ret;
 363                 default:
 364                         /* Implicit handling for NF_STOLEN, as well as any other
 365                          * non conventional verdicts.
 366                          */
 367                         return 0;
 368                 }
 369         } while (entry);
 370
 371         return 1;
 372 }
 373 EXPORT_SYMBOL(nf_hook_slow);
 374
 375
 376 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 377 {
 378         if (writable_len > skb->len)
 379                 return 0;
 380
 381         /* Not exclusive use of packet?  Must copy. */
 382         if (!skb_cloned(skb)) {
 383                 if (writable_len <= skb_headlen(skb))
 384                         return 1;
 385         } else if (skb_clone_writable(skb, writable_len))
 386                 return 1;
 387
 388         if (writable_len <= skb_headlen(skb))
 389                 writable_len = 0;
 390         else
 391                 writable_len -= skb_headlen(skb);
 392
 393         return !!__pskb_pull_tail(skb, writable_len);
 394 }
 395 EXPORT_SYMBOL(skb_make_writable);
 396
 397 /* This needs to be compiled in any case to avoid dependencies between the
 398  * nfnetlink_queue code and nf_conntrack.
 399  */
 400 struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly;
 401 EXPORT_SYMBOL_GPL(nfnl_ct_hook);
 402
 403 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 404 /* This does not belong here, but locally generated errors need it if connection
 405    tracking in use: without this, connection may not be in hash table, and hence
 406    manufactured ICMP or RST packets will not be associated with it. */
 407 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 408                 __rcu __read_mostly;
 409 EXPORT_SYMBOL(ip_ct_attach);
 410
 411 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 412 {
 413         void (*attach)(struct sk_buff *, const struct sk_buff *);
 414
 415         if (skb->_nfct) {
 416                 rcu_read_lock();
 417                 attach = rcu_dereference(ip_ct_attach);
 418                 if (attach)
 419                         attach(new, skb);
 420                 rcu_read_unlock();
 421         }
 422 }
 423 EXPORT_SYMBOL(nf_ct_attach);
 424
 425 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 426 EXPORT_SYMBOL(nf_ct_destroy);
 427
 428 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 429 {
 430         void (*destroy)(struct nf_conntrack *);
 431
 432         rcu_read_lock();
 433         destroy = rcu_dereference(nf_ct_destroy);
 434         BUG_ON(destroy == NULL);
 435         destroy(nfct);
 436         rcu_read_unlock();
 437 }
 438 EXPORT_SYMBOL(nf_conntrack_destroy);
 439
 440 /* Built-in default zone used e.g. by modules. */
 441 const struct nf_conntrack_zone nf_ct_zone_dflt = {
 442         .id     = NF_CT_DEFAULT_ZONE_ID,
 443         .dir    = NF_CT_DEFAULT_ZONE_DIR,
 444 };
 445 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt);
 446 #endif /* CONFIG_NF_CONNTRACK */
 447
 448 #ifdef CONFIG_NF_NAT_NEEDED
 449 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 450 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 451 #endif
 452
 453 static int nf_register_hook_list(struct net *net)
 454 {
 455         struct nf_hook_ops *elem;
 456         int ret;
 457
 458         rtnl_lock();
 459         list_for_each_entry(elem, &nf_hook_list, list) {
 460                 ret = nf_register_net_hook(net, elem);
 461                 if (ret && ret != -ENOENT)
 462                         goto out_undo;
 463         }
 464         rtnl_unlock();
 465         return 0;
 466
 467 out_undo:
 468         list_for_each_entry_continue_reverse(elem, &nf_hook_list, list)
 469                 nf_unregister_net_hook(net, elem);
 470         rtnl_unlock();
 471         return ret;
 472 }
 473
 474 static void nf_unregister_hook_list(struct net *net)
 475 {
 476         struct nf_hook_ops *elem;
 477
 478         rtnl_lock();
 479         list_for_each_entry(elem, &nf_hook_list, list)
 480                 nf_unregister_net_hook(net, elem);
 481         rtnl_unlock();
 482 }
 483
 484 static int __net_init netfilter_net_init(struct net *net)
 485 {
 486         int i, h, ret;
 487
 488         for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
 489                 for (h = 0; h < NF_MAX_HOOKS; h++)
 490                         RCU_INIT_POINTER(net->nf.hooks[i][h], NULL);
 491         }
 492
 493 #ifdef CONFIG_PROC_FS
 494         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 495                                                 net->proc_net);
 496         if (!net->nf.proc_netfilter) {
 497                 if (!net_eq(net, &init_net))
 498                         pr_err("cannot create netfilter proc entry");
 499
 500                 return -ENOMEM;
 501         }
 502 #endif
 503         ret = nf_register_hook_list(net);
 504         if (ret)
 505                 remove_proc_entry("netfilter", net->proc_net);
 506
 507         return ret;
 508 }
 509
 510 static void __net_exit netfilter_net_exit(struct net *net)
 511 {
 512         nf_unregister_hook_list(net);
 513         remove_proc_entry("netfilter", net->proc_net);
 514 }
 515
 516 static struct pernet_operations netfilter_net_ops = {
 517         .init = netfilter_net_init,
 518         .exit = netfilter_net_exit,
 519 };
 520
 521 int __init netfilter_init(void)
 522 {
 523         int ret;
 524
 525         ret = register_pernet_subsys(&netfilter_net_ops);
 526         if (ret < 0)
 527                 goto err;
 528
 529         ret = netfilter_log_init();
 530         if (ret < 0)
 531                 goto err_pernet;
 532
 533         return 0;
 534 err_pernet:
 535         unregister_pernet_subsys(&netfilter_net_ops);
 536 err:
 537         return ret;
 538 }