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 int nf_register_hook(struct nf_hook_ops *reg)
 192 {
 193         struct net *net, *last;
 194         int ret;
 195
 196         rtnl_lock();
 197         for_each_net(net) {
 198                 ret = nf_register_net_hook(net, reg);
 199                 if (ret && ret != -ENOENT)
 200                         goto rollback;
 201         }
 202         list_add_tail(&reg->list, &nf_hook_list);
 203         rtnl_unlock();
 204
 205         return 0;
 206 rollback:
 207         last = net;
 208         for_each_net(net) {
 209                 if (net == last)
 210                         break;
 211                 nf_unregister_net_hook(net, reg);
 212         }
 213         rtnl_unlock();
 214         return ret;
 215 }
 216 EXPORT_SYMBOL(nf_register_hook);
 217
 218 void nf_unregister_hook(struct nf_hook_ops *reg)
 219 {
 220         struct net *net;
 221
 222         rtnl_lock();
 223         list_del(&reg->list);
 224         for_each_net(net)
 225                 nf_unregister_net_hook(net, reg);
 226         rtnl_unlock();
 227 }
 228 EXPORT_SYMBOL(nf_unregister_hook);
 229
 230 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 231 {
 232         unsigned int i;
 233         int err = 0;
 234
 235         for (i = 0; i < n; i++) {
 236                 err = nf_register_hook(&reg[i]);
 237                 if (err)
 238                         goto err;
 239         }
 240         return err;
 241
 242 err:
 243         if (i > 0)
 244                 nf_unregister_hooks(reg, i);
 245         return err;
 246 }
 247 EXPORT_SYMBOL(nf_register_hooks);
 248
 249 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 250 {
 251         while (n-- > 0)
 252                 nf_unregister_hook(&reg[n]);
 253 }
 254 EXPORT_SYMBOL(nf_unregister_hooks);
 255
 256 unsigned int nf_iterate(struct list_head *head,
 257                         struct sk_buff *skb,
 258                         struct nf_hook_state *state,
 259                         struct nf_hook_ops **elemp)
 260 {
 261         unsigned int verdict;
 262
 263         /*
 264          * The caller must not block between calls to this
 265          * function because of risk of continuing from deleted element.
 266          */
 267         list_for_each_entry_continue_rcu((*elemp), head, list) {
 268                 if (state->thresh > (*elemp)->priority)
 269                         continue;
 270
 271                 /* Optimization: we don't need to hold module
 272                    reference here, since function can't sleep. --RR */
 273 repeat:
 274                 verdict = (*elemp)->hook((*elemp)->priv, skb, state);
 275                 if (verdict != NF_ACCEPT) {
 276 #ifdef CONFIG_NETFILTER_DEBUG
 277                         if (unlikely((verdict & NF_VERDICT_MASK)
 278                                                         > NF_MAX_VERDICT)) {
 279                                 NFDEBUG("Evil return from %p(%u).\n",
 280                                         (*elemp)->hook, state->hook);
 281                                 continue;
 282                         }
 283 #endif
 284                         if (verdict != NF_REPEAT)
 285                                 return verdict;
 286                         goto repeat;
 287                 }
 288         }
 289         return NF_ACCEPT;
 290 }
 291
 292
 293 /* Returns 1 if okfn() needs to be executed by the caller,
 294  * -EPERM for NF_DROP, 0 otherwise. */
 295 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state)
 296 {
 297         struct nf_hook_ops *elem;
 298         unsigned int verdict;
 299         int ret = 0;
 300
 301         /* We may already have this, but read-locks nest anyway */
 302         rcu_read_lock();
 303
 304         elem = list_entry_rcu(state->hook_list, struct nf_hook_ops, list);
 305 next_hook:
 306         verdict = nf_iterate(state->hook_list, skb, state, &elem);
 307         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 308                 ret = 1;
 309         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 310                 kfree_skb(skb);
 311                 ret = NF_DROP_GETERR(verdict);
 312                 if (ret == 0)
 313                         ret = -EPERM;
 314         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 315                 int err = nf_queue(skb, elem, state,
 316                                    verdict >> NF_VERDICT_QBITS);
 317                 if (err < 0) {
 318                         if (err == -ESRCH &&
 319                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 320                                 goto next_hook;
 321                         kfree_skb(skb);
 322                 }
 323         }
 324         rcu_read_unlock();
 325         return ret;
 326 }
 327 EXPORT_SYMBOL(nf_hook_slow);
 328
 329
 330 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 331 {
 332         if (writable_len > skb->len)
 333                 return 0;
 334
 335         /* Not exclusive use of packet?  Must copy. */
 336         if (!skb_cloned(skb)) {
 337                 if (writable_len <= skb_headlen(skb))
 338                         return 1;
 339         } else if (skb_clone_writable(skb, writable_len))
 340                 return 1;
 341
 342         if (writable_len <= skb_headlen(skb))
 343                 writable_len = 0;
 344         else
 345                 writable_len -= skb_headlen(skb);
 346
 347         return !!__pskb_pull_tail(skb, writable_len);
 348 }
 349 EXPORT_SYMBOL(skb_make_writable);
 350
 351 /* This needs to be compiled in any case to avoid dependencies between the
 352  * nfnetlink_queue code and nf_conntrack.
 353  */
 354 struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly;
 355 EXPORT_SYMBOL_GPL(nfnl_ct_hook);
 356
 357 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 358 /* This does not belong here, but locally generated errors need it if connection
 359    tracking in use: without this, connection may not be in hash table, and hence
 360    manufactured ICMP or RST packets will not be associated with it. */
 361 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 362                 __rcu __read_mostly;
 363 EXPORT_SYMBOL(ip_ct_attach);
 364
 365 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 366 {
 367         void (*attach)(struct sk_buff *, const struct sk_buff *);
 368
 369         if (skb->nfct) {
 370                 rcu_read_lock();
 371                 attach = rcu_dereference(ip_ct_attach);
 372                 if (attach)
 373                         attach(new, skb);
 374                 rcu_read_unlock();
 375         }
 376 }
 377 EXPORT_SYMBOL(nf_ct_attach);
 378
 379 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 380 EXPORT_SYMBOL(nf_ct_destroy);
 381
 382 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 383 {
 384         void (*destroy)(struct nf_conntrack *);
 385
 386         rcu_read_lock();
 387         destroy = rcu_dereference(nf_ct_destroy);
 388         BUG_ON(destroy == NULL);
 389         destroy(nfct);
 390         rcu_read_unlock();
 391 }
 392 EXPORT_SYMBOL(nf_conntrack_destroy);
 393
 394 /* Built-in default zone used e.g. by modules. */
 395 const struct nf_conntrack_zone nf_ct_zone_dflt = {
 396         .id     = NF_CT_DEFAULT_ZONE_ID,
 397         .dir    = NF_CT_DEFAULT_ZONE_DIR,
 398 };
 399 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt);
 400 #endif /* CONFIG_NF_CONNTRACK */
 401
 402 #ifdef CONFIG_NF_NAT_NEEDED
 403 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 404 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 405 #endif
 406
 407 static int nf_register_hook_list(struct net *net)
 408 {
 409         struct nf_hook_ops *elem;
 410         int ret;
 411
 412         rtnl_lock();
 413         list_for_each_entry(elem, &nf_hook_list, list) {
 414                 ret = nf_register_net_hook(net, elem);
 415                 if (ret && ret != -ENOENT)
 416                         goto out_undo;
 417         }
 418         rtnl_unlock();
 419         return 0;
 420
 421 out_undo:
 422         list_for_each_entry_continue_reverse(elem, &nf_hook_list, list)
 423                 nf_unregister_net_hook(net, elem);
 424         rtnl_unlock();
 425         return ret;
 426 }
 427
 428 static void nf_unregister_hook_list(struct net *net)
 429 {
 430         struct nf_hook_ops *elem;
 431
 432         rtnl_lock();
 433         list_for_each_entry(elem, &nf_hook_list, list)
 434                 nf_unregister_net_hook(net, elem);
 435         rtnl_unlock();
 436 }
 437
 438 static int __net_init netfilter_net_init(struct net *net)
 439 {
 440         int i, h, ret;
 441
 442         for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
 443                 for (h = 0; h < NF_MAX_HOOKS; h++)
 444                         INIT_LIST_HEAD(&net->nf.hooks[i][h]);
 445         }
 446
 447 #ifdef CONFIG_PROC_FS
 448         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 449                                                 net->proc_net);
 450         if (!net->nf.proc_netfilter) {
 451                 if (!net_eq(net, &init_net))
 452                         pr_err("cannot create netfilter proc entry");
 453
 454                 return -ENOMEM;
 455         }
 456 #endif
 457         ret = nf_register_hook_list(net);
 458         if (ret)
 459                 remove_proc_entry("netfilter", net->proc_net);
 460
 461         return ret;
 462 }
 463
 464 static void __net_exit netfilter_net_exit(struct net *net)
 465 {
 466         nf_unregister_hook_list(net);
 467         remove_proc_entry("netfilter", net->proc_net);
 468 }
 469
 470 static struct pernet_operations netfilter_net_ops = {
 471         .init = netfilter_net_init,
 472         .exit = netfilter_net_exit,
 473 };
 474
 475 int __init netfilter_init(void)
 476 {
 477         int ret;
 478
 479         ret = register_pernet_subsys(&netfilter_net_ops);
 480         if (ret < 0)
 481                 goto err;
 482
 483         ret = netfilter_log_init();
 484         if (ret < 0)
 485                 goto err_pernet;
 486
 487         return 0;
 488 err_pernet:
 489         unregister_pernet_subsys(&netfilter_net_ops);
 490 err:
 491         return ret;
 492 }