kernel/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/locallock.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28
  29 #include "nf_internals.h"
  30
  31 #ifdef CONFIG_PREEMPT_RT_BASE
  32 DEFINE_LOCAL_IRQ_LOCK(xt_write_lock);
  33 EXPORT_PER_CPU_SYMBOL(xt_write_lock);
  34 #endif
  35
  36 static DEFINE_MUTEX(afinfo_mutex);
  37
  38 const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO] __read_mostly;
  39 EXPORT_SYMBOL(nf_afinfo);
  40 const struct nf_ipv6_ops __rcu *nf_ipv6_ops __read_mostly;
  41 EXPORT_SYMBOL_GPL(nf_ipv6_ops);
  42
  43 DEFINE_PER_CPU(bool, nf_skb_duplicated);
  44 EXPORT_SYMBOL_GPL(nf_skb_duplicated);
  45
  46 int nf_register_afinfo(const struct nf_afinfo *afinfo)
  47 {
  48         mutex_lock(&afinfo_mutex);
  49         RCU_INIT_POINTER(nf_afinfo[afinfo->family], afinfo);
  50         mutex_unlock(&afinfo_mutex);
  51         return 0;
  52 }
  53 EXPORT_SYMBOL_GPL(nf_register_afinfo);
  54
  55 void nf_unregister_afinfo(const struct nf_afinfo *afinfo)
  56 {
  57         mutex_lock(&afinfo_mutex);
  58         RCU_INIT_POINTER(nf_afinfo[afinfo->family], NULL);
  59         mutex_unlock(&afinfo_mutex);
  60         synchronize_rcu();
  61 }
  62 EXPORT_SYMBOL_GPL(nf_unregister_afinfo);
  63
  64 #ifdef HAVE_JUMP_LABEL
  65 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
  66 EXPORT_SYMBOL(nf_hooks_needed);
  67 #endif
  68
  69 static DEFINE_MUTEX(nf_hook_mutex);
  70
  71 static struct list_head *nf_find_hook_list(struct net *net,
  72                                            const struct nf_hook_ops *reg)
  73 {
  74         struct list_head *hook_list = NULL;
  75
  76         if (reg->pf != NFPROTO_NETDEV)
  77                 hook_list = &net->nf.hooks[reg->pf][reg->hooknum];
  78         else if (reg->hooknum == NF_NETDEV_INGRESS) {
  79 #ifdef CONFIG_NETFILTER_INGRESS
  80                 if (reg->dev && dev_net(reg->dev) == net)
  81                         hook_list = &reg->dev->nf_hooks_ingress;
  82 #endif
  83         }
  84         return hook_list;
  85 }
  86
  87 struct nf_hook_entry {
  88         const struct nf_hook_ops        *orig_ops;
  89         struct nf_hook_ops              ops;
  90 };
  91
  92 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
  93 {
  94         struct list_head *hook_list;
  95         struct nf_hook_entry *entry;
  96         struct nf_hook_ops *elem;
  97
  98         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
  99         if (!entry)
 100                 return -ENOMEM;
 101
 102         entry->orig_ops = reg;
 103         entry->ops      = *reg;
 104
 105         hook_list = nf_find_hook_list(net, reg);
 106         if (!hook_list) {
 107                 kfree(entry);
 108                 return -ENOENT;
 109         }
 110
 111         mutex_lock(&nf_hook_mutex);
 112         list_for_each_entry(elem, hook_list, list) {
 113                 if (reg->priority < elem->priority)
 114                         break;
 115         }
 116         list_add_rcu(&entry->ops.list, elem->list.prev);
 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 list_head *hook_list;
 132         struct nf_hook_entry *entry;
 133         struct nf_hook_ops *elem;
 134
 135         hook_list = nf_find_hook_list(net, reg);
 136         if (!hook_list)
 137                 return;
 138
 139         mutex_lock(&nf_hook_mutex);
 140         list_for_each_entry(elem, hook_list, list) {
 141                 entry = container_of(elem, struct nf_hook_entry, ops);
 142                 if (entry->orig_ops == reg) {
 143                         list_del_rcu(&entry->ops.list);
 144                         break;
 145                 }
 146         }
 147         mutex_unlock(&nf_hook_mutex);
 148         if (&elem->list == hook_list) {
 149                 WARN(1, "nf_unregister_net_hook: hook not found!\n");
 150                 return;
 151         }
 152 #ifdef CONFIG_NETFILTER_INGRESS
 153         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 154                 net_dec_ingress_queue();
 155 #endif
 156 #ifdef HAVE_JUMP_LABEL
 157         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
 158 #endif
 159         synchronize_net();
 160         nf_queue_nf_hook_drop(net, &entry->ops);
 161         /* other cpu might still process nfqueue verdict that used reg */
 162         synchronize_net();
 163         kfree(entry);
 164 }
 165 EXPORT_SYMBOL(nf_unregister_net_hook);
 166
 167 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 168                           unsigned int n)
 169 {
 170         unsigned int i;
 171         int err = 0;
 172
 173         for (i = 0; i < n; i++) {
 174                 err = nf_register_net_hook(net, &reg[i]);
 175                 if (err)
 176                         goto err;
 177         }
 178         return err;
 179
 180 err:
 181         if (i > 0)
 182                 nf_unregister_net_hooks(net, reg, i);
 183         return err;
 184 }
 185 EXPORT_SYMBOL(nf_register_net_hooks);
 186
 187 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 188                              unsigned int n)
 189 {
 190         while (n-- > 0)
 191                 nf_unregister_net_hook(net, &reg[n]);
 192 }
 193 EXPORT_SYMBOL(nf_unregister_net_hooks);
 194
 195 static LIST_HEAD(nf_hook_list);
 196
 197 int nf_register_hook(struct nf_hook_ops *reg)
 198 {
 199         struct net *net, *last;
 200         int ret;
 201
 202         rtnl_lock();
 203         for_each_net(net) {
 204                 ret = nf_register_net_hook(net, reg);
 205                 if (ret && ret != -ENOENT)
 206                         goto rollback;
 207         }
 208         list_add_tail(&reg->list, &nf_hook_list);
 209         rtnl_unlock();
 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         rtnl_unlock();
 220         return ret;
 221 }
 222 EXPORT_SYMBOL(nf_register_hook);
 223
 224 void nf_unregister_hook(struct nf_hook_ops *reg)
 225 {
 226         struct net *net;
 227
 228         rtnl_lock();
 229         list_del(&reg->list);
 230         for_each_net(net)
 231                 nf_unregister_net_hook(net, reg);
 232         rtnl_unlock();
 233 }
 234 EXPORT_SYMBOL(nf_unregister_hook);
 235
 236 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 237 {
 238         unsigned int i;
 239         int err = 0;
 240
 241         for (i = 0; i < n; i++) {
 242                 err = nf_register_hook(&reg[i]);
 243                 if (err)
 244                         goto err;
 245         }
 246         return err;
 247
 248 err:
 249         if (i > 0)
 250                 nf_unregister_hooks(reg, i);
 251         return err;
 252 }
 253 EXPORT_SYMBOL(nf_register_hooks);
 254
 255 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 256 {
 257         while (n-- > 0)
 258                 nf_unregister_hook(&reg[n]);
 259 }
 260 EXPORT_SYMBOL(nf_unregister_hooks);
 261
 262 unsigned int nf_iterate(struct list_head *head,
 263                         struct sk_buff *skb,
 264                         struct nf_hook_state *state,
 265                         struct nf_hook_ops **elemp)
 266 {
 267         unsigned int verdict;
 268
 269         /*
 270          * The caller must not block between calls to this
 271          * function because of risk of continuing from deleted element.
 272          */
 273         list_for_each_entry_continue_rcu((*elemp), head, list) {
 274                 if (state->thresh > (*elemp)->priority)
 275                         continue;
 276
 277                 /* Optimization: we don't need to hold module
 278                    reference here, since function can't sleep. --RR */
 279 repeat:
 280                 verdict = (*elemp)->hook((*elemp)->priv, skb, state);
 281                 if (verdict != NF_ACCEPT) {
 282 #ifdef CONFIG_NETFILTER_DEBUG
 283                         if (unlikely((verdict & NF_VERDICT_MASK)
 284                                                         > NF_MAX_VERDICT)) {
 285                                 NFDEBUG("Evil return from %p(%u).\n",
 286                                         (*elemp)->hook, state->hook);
 287                                 continue;
 288                         }
 289 #endif
 290                         if (verdict != NF_REPEAT)
 291                                 return verdict;
 292                         goto repeat;
 293                 }
 294         }
 295         return NF_ACCEPT;
 296 }
 297
 298
 299 /* Returns 1 if okfn() needs to be executed by the caller,
 300  * -EPERM for NF_DROP, 0 otherwise. */
 301 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state)
 302 {
 303         struct nf_hook_ops *elem;
 304         unsigned int verdict;
 305         int ret = 0;
 306
 307         /* We may already have this, but read-locks nest anyway */
 308         rcu_read_lock();
 309
 310         elem = list_entry_rcu(state->hook_list, struct nf_hook_ops, list);
 311 next_hook:
 312         verdict = nf_iterate(state->hook_list, skb, state, &elem);
 313         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 314                 ret = 1;
 315         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 316                 kfree_skb(skb);
 317                 ret = NF_DROP_GETERR(verdict);
 318                 if (ret == 0)
 319                         ret = -EPERM;
 320         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 321                 int err = nf_queue(skb, elem, state,
 322                                    verdict >> NF_VERDICT_QBITS);
 323                 if (err < 0) {
 324                         if (err == -ESRCH &&
 325                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 326                                 goto next_hook;
 327                         kfree_skb(skb);
 328                 }
 329         }
 330         rcu_read_unlock();
 331         return ret;
 332 }
 333 EXPORT_SYMBOL(nf_hook_slow);
 334
 335
 336 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 337 {
 338         if (writable_len > skb->len)
 339                 return 0;
 340
 341         /* Not exclusive use of packet?  Must copy. */
 342         if (!skb_cloned(skb)) {
 343                 if (writable_len <= skb_headlen(skb))
 344                         return 1;
 345         } else if (skb_clone_writable(skb, writable_len))
 346                 return 1;
 347
 348         if (writable_len <= skb_headlen(skb))
 349                 writable_len = 0;
 350         else
 351                 writable_len -= skb_headlen(skb);
 352
 353         return !!__pskb_pull_tail(skb, writable_len);
 354 }
 355 EXPORT_SYMBOL(skb_make_writable);
 356
 357 /* This needs to be compiled in any case to avoid dependencies between the
 358  * nfnetlink_queue code and nf_conntrack.
 359  */
 360 struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly;
 361 EXPORT_SYMBOL_GPL(nfnl_ct_hook);
 362
 363 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 364 /* This does not belong here, but locally generated errors need it if connection
 365    tracking in use: without this, connection may not be in hash table, and hence
 366    manufactured ICMP or RST packets will not be associated with it. */
 367 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 368                 __rcu __read_mostly;
 369 EXPORT_SYMBOL(ip_ct_attach);
 370
 371 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 372 {
 373         void (*attach)(struct sk_buff *, const struct sk_buff *);
 374
 375         if (skb->nfct) {
 376                 rcu_read_lock();
 377                 attach = rcu_dereference(ip_ct_attach);
 378                 if (attach)
 379                         attach(new, skb);
 380                 rcu_read_unlock();
 381         }
 382 }
 383 EXPORT_SYMBOL(nf_ct_attach);
 384
 385 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 386 EXPORT_SYMBOL(nf_ct_destroy);
 387
 388 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 389 {
 390         void (*destroy)(struct nf_conntrack *);
 391
 392         rcu_read_lock();
 393         destroy = rcu_dereference(nf_ct_destroy);
 394         BUG_ON(destroy == NULL);
 395         destroy(nfct);
 396         rcu_read_unlock();
 397 }
 398 EXPORT_SYMBOL(nf_conntrack_destroy);
 399
 400 /* Built-in default zone used e.g. by modules. */
 401 const struct nf_conntrack_zone nf_ct_zone_dflt = {
 402         .id     = NF_CT_DEFAULT_ZONE_ID,
 403         .dir    = NF_CT_DEFAULT_ZONE_DIR,
 404 };
 405 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt);
 406 #endif /* CONFIG_NF_CONNTRACK */
 407
 408 #ifdef CONFIG_NF_NAT_NEEDED
 409 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 410 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 411 #endif
 412
 413 static int nf_register_hook_list(struct net *net)
 414 {
 415         struct nf_hook_ops *elem;
 416         int ret;
 417
 418         rtnl_lock();
 419         list_for_each_entry(elem, &nf_hook_list, list) {
 420                 ret = nf_register_net_hook(net, elem);
 421                 if (ret && ret != -ENOENT)
 422                         goto out_undo;
 423         }
 424         rtnl_unlock();
 425         return 0;
 426
 427 out_undo:
 428         list_for_each_entry_continue_reverse(elem, &nf_hook_list, list)
 429                 nf_unregister_net_hook(net, elem);
 430         rtnl_unlock();
 431         return ret;
 432 }
 433
 434 static void nf_unregister_hook_list(struct net *net)
 435 {
 436         struct nf_hook_ops *elem;
 437
 438         rtnl_lock();
 439         list_for_each_entry(elem, &nf_hook_list, list)
 440                 nf_unregister_net_hook(net, elem);
 441         rtnl_unlock();
 442 }
 443
 444 static int __net_init netfilter_net_init(struct net *net)
 445 {
 446         int i, h, ret;
 447
 448         for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
 449                 for (h = 0; h < NF_MAX_HOOKS; h++)
 450                         INIT_LIST_HEAD(&net->nf.hooks[i][h]);
 451         }
 452
 453 #ifdef CONFIG_PROC_FS
 454         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 455                                                 net->proc_net);
 456         if (!net->nf.proc_netfilter) {
 457                 if (!net_eq(net, &init_net))
 458                         pr_err("cannot create netfilter proc entry");
 459
 460                 return -ENOMEM;
 461         }
 462 #endif
 463         ret = nf_register_hook_list(net);
 464         if (ret)
 465                 remove_proc_entry("netfilter", net->proc_net);
 466
 467         return ret;
 468 }
 469
 470 static void __net_exit netfilter_net_exit(struct net *net)
 471 {
 472         nf_unregister_hook_list(net);
 473         remove_proc_entry("netfilter", net->proc_net);
 474 }
 475
 476 static struct pernet_operations netfilter_net_ops = {
 477         .init = netfilter_net_init,
 478         .exit = netfilter_net_exit,
 479 };
 480
 481 int __init netfilter_init(void)
 482 {
 483         int ret;
 484
 485         ret = register_pernet_subsys(&netfilter_net_ops);
 486         if (ret < 0)
 487                 goto err;
 488
 489         ret = netfilter_log_init();
 490         if (ret < 0)
 491                 goto err_pernet;
 492
 493         return 0;
 494 err_pernet:
 495         unregister_pernet_subsys(&netfilter_net_ops);
 496 err:
 497         return ret;
 498 }