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