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