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         if (reg->pf == NFPROTO_NETDEV &&
  93             (reg->hooknum != NF_NETDEV_INGRESS ||
  94              !reg->dev || dev_net(reg->dev) != net))
  95                 return -EINVAL;
  96
  97         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
  98         if (!entry)
  99                 return -ENOMEM;
 100
 101         entry->orig_ops = reg;
 102         entry->ops      = *reg;
 103
 104         hook_list = nf_find_hook_list(net, reg);
 105         if (!hook_list) {
 106                 kfree(entry);
 107                 return -ENOENT;
 108         }
 109
 110         mutex_lock(&nf_hook_mutex);
 111         list_for_each_entry(elem, hook_list, list) {
 112                 if (reg->priority < elem->priority)
 113                         break;
 114         }
 115         list_add_rcu(&entry->ops.list, elem->list.prev);
 116         mutex_unlock(&nf_hook_mutex);
 117 #ifdef CONFIG_NETFILTER_INGRESS
 118         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 119                 net_inc_ingress_queue();
 120 #endif
 121 #ifdef HAVE_JUMP_LABEL
 122         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
 123 #endif
 124         return 0;
 125 }
 126 EXPORT_SYMBOL(nf_register_net_hook);
 127
 128 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 129 {
 130         struct list_head *hook_list;
 131         struct nf_hook_entry *entry;
 132         struct nf_hook_ops *elem;
 133
 134         hook_list = nf_find_hook_list(net, reg);
 135         if (!hook_list)
 136                 return;
 137
 138         mutex_lock(&nf_hook_mutex);
 139         list_for_each_entry(elem, hook_list, list) {
 140                 entry = container_of(elem, struct nf_hook_entry, ops);
 141                 if (entry->orig_ops == reg) {
 142                         list_del_rcu(&entry->ops.list);
 143                         break;
 144                 }
 145         }
 146         mutex_unlock(&nf_hook_mutex);
 147         if (&elem->list == hook_list) {
 148                 WARN(1, "nf_unregister_net_hook: hook not found!\n");
 149                 return;
 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         synchronize_net();
 159         nf_queue_nf_hook_drop(net, &entry->ops);
 160         /* other cpu might still process nfqueue verdict that used reg */
 161         synchronize_net();
 162         kfree(entry);
 163 }
 164 EXPORT_SYMBOL(nf_unregister_net_hook);
 165
 166 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 167                           unsigned int n)
 168 {
 169         unsigned int i;
 170         int err = 0;
 171
 172         for (i = 0; i < n; i++) {
 173                 err = nf_register_net_hook(net, &reg[i]);
 174                 if (err)
 175                         goto err;
 176         }
 177         return err;
 178
 179 err:
 180         if (i > 0)
 181                 nf_unregister_net_hooks(net, reg, i);
 182         return err;
 183 }
 184 EXPORT_SYMBOL(nf_register_net_hooks);
 185
 186 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 187                              unsigned int n)
 188 {
 189         while (n-- > 0)
 190                 nf_unregister_net_hook(net, &reg[n]);
 191 }
 192 EXPORT_SYMBOL(nf_unregister_net_hooks);
 193
 194 static LIST_HEAD(nf_hook_list);
 195
 196 int nf_register_hook(struct nf_hook_ops *reg)
 197 {
 198         struct net *net, *last;
 199         int ret;
 200
 201         rtnl_lock();
 202         for_each_net(net) {
 203                 ret = nf_register_net_hook(net, reg);
 204                 if (ret && ret != -ENOENT)
 205                         goto rollback;
 206         }
 207         list_add_tail(&reg->list, &nf_hook_list);
 208         rtnl_unlock();
 209
 210         return 0;
 211 rollback:
 212         last = net;
 213         for_each_net(net) {
 214                 if (net == last)
 215                         break;
 216                 nf_unregister_net_hook(net, reg);
 217         }
 218         rtnl_unlock();
 219         return ret;
 220 }
 221 EXPORT_SYMBOL(nf_register_hook);
 222
 223 void nf_unregister_hook(struct nf_hook_ops *reg)
 224 {
 225         struct net *net;
 226
 227         rtnl_lock();
 228         list_del(&reg->list);
 229         for_each_net(net)
 230                 nf_unregister_net_hook(net, reg);
 231         rtnl_unlock();
 232 }
 233 EXPORT_SYMBOL(nf_unregister_hook);
 234
 235 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 236 {
 237         unsigned int i;
 238         int err = 0;
 239
 240         for (i = 0; i < n; i++) {
 241                 err = nf_register_hook(&reg[i]);
 242                 if (err)
 243                         goto err;
 244         }
 245         return err;
 246
 247 err:
 248         if (i > 0)
 249                 nf_unregister_hooks(reg, i);
 250         return err;
 251 }
 252 EXPORT_SYMBOL(nf_register_hooks);
 253
 254 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 255 {
 256         while (n-- > 0)
 257                 nf_unregister_hook(&reg[n]);
 258 }
 259 EXPORT_SYMBOL(nf_unregister_hooks);
 260
 261 unsigned int nf_iterate(struct list_head *head,
 262                         struct sk_buff *skb,
 263                         struct nf_hook_state *state,
 264                         struct nf_hook_ops **elemp)
 265 {
 266         unsigned int verdict;
 267
 268         /*
 269          * The caller must not block between calls to this
 270          * function because of risk of continuing from deleted element.
 271          */
 272         list_for_each_entry_continue_rcu((*elemp), head, list) {
 273                 if (state->thresh > (*elemp)->priority)
 274                         continue;
 275
 276                 /* Optimization: we don't need to hold module
 277                    reference here, since function can't sleep. --RR */
 278 repeat:
 279                 verdict = (*elemp)->hook((*elemp)->priv, skb, state);
 280                 if (verdict != NF_ACCEPT) {
 281 #ifdef CONFIG_NETFILTER_DEBUG
 282                         if (unlikely((verdict & NF_VERDICT_MASK)
 283                                                         > NF_MAX_VERDICT)) {
 284                                 NFDEBUG("Evil return from %p(%u).\n",
 285                                         (*elemp)->hook, state->hook);
 286                                 continue;
 287                         }
 288 #endif
 289                         if (verdict != NF_REPEAT)
 290                                 return verdict;
 291                         goto repeat;
 292                 }
 293         }
 294         return NF_ACCEPT;
 295 }
 296
 297
 298 /* Returns 1 if okfn() needs to be executed by the caller,
 299  * -EPERM for NF_DROP, 0 otherwise.  Caller must hold rcu_read_lock. */
 300 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state)
 301 {
 302         struct nf_hook_ops *elem;
 303         unsigned int verdict;
 304         int ret = 0;
 305
 306         elem = list_entry_rcu(state->hook_list, struct nf_hook_ops, list);
 307 next_hook:
 308         verdict = nf_iterate(state->hook_list, skb, state, &elem);
 309         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 310                 ret = 1;
 311         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 312                 kfree_skb(skb);
 313                 ret = NF_DROP_GETERR(verdict);
 314                 if (ret == 0)
 315                         ret = -EPERM;
 316         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 317                 int err = nf_queue(skb, elem, state,
 318                                    verdict >> NF_VERDICT_QBITS);
 319                 if (err < 0) {
 320                         if (err == -ESRCH &&
 321                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 322                                 goto next_hook;
 323                         kfree_skb(skb);
 324                 }
 325         }
 326         return ret;
 327 }
 328 EXPORT_SYMBOL(nf_hook_slow);
 329
 330
 331 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 332 {
 333         if (writable_len > skb->len)
 334                 return 0;
 335
 336         /* Not exclusive use of packet?  Must copy. */
 337         if (!skb_cloned(skb)) {
 338                 if (writable_len <= skb_headlen(skb))
 339                         return 1;
 340         } else if (skb_clone_writable(skb, writable_len))
 341                 return 1;
 342
 343         if (writable_len <= skb_headlen(skb))
 344                 writable_len = 0;
 345         else
 346                 writable_len -= skb_headlen(skb);
 347
 348         return !!__pskb_pull_tail(skb, writable_len);
 349 }
 350 EXPORT_SYMBOL(skb_make_writable);
 351
 352 /* This needs to be compiled in any case to avoid dependencies between the
 353  * nfnetlink_queue code and nf_conntrack.
 354  */
 355 struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly;
 356 EXPORT_SYMBOL_GPL(nfnl_ct_hook);
 357
 358 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 359 /* This does not belong here, but locally generated errors need it if connection
 360    tracking in use: without this, connection may not be in hash table, and hence
 361    manufactured ICMP or RST packets will not be associated with it. */
 362 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 363                 __rcu __read_mostly;
 364 EXPORT_SYMBOL(ip_ct_attach);
 365
 366 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 367 {
 368         void (*attach)(struct sk_buff *, const struct sk_buff *);
 369
 370         if (skb->nfct) {
 371                 rcu_read_lock();
 372                 attach = rcu_dereference(ip_ct_attach);
 373                 if (attach)
 374                         attach(new, skb);
 375                 rcu_read_unlock();
 376         }
 377 }
 378 EXPORT_SYMBOL(nf_ct_attach);
 379
 380 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 381 EXPORT_SYMBOL(nf_ct_destroy);
 382
 383 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 384 {
 385         void (*destroy)(struct nf_conntrack *);
 386
 387         rcu_read_lock();
 388         destroy = rcu_dereference(nf_ct_destroy);
 389         BUG_ON(destroy == NULL);
 390         destroy(nfct);
 391         rcu_read_unlock();
 392 }
 393 EXPORT_SYMBOL(nf_conntrack_destroy);
 394
 395 /* Built-in default zone used e.g. by modules. */
 396 const struct nf_conntrack_zone nf_ct_zone_dflt = {
 397         .id     = NF_CT_DEFAULT_ZONE_ID,
 398         .dir    = NF_CT_DEFAULT_ZONE_DIR,
 399 };
 400 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt);
 401 #endif /* CONFIG_NF_CONNTRACK */
 402
 403 #ifdef CONFIG_NF_NAT_NEEDED
 404 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 405 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 406 #endif
 407
 408 static int nf_register_hook_list(struct net *net)
 409 {
 410         struct nf_hook_ops *elem;
 411         int ret;
 412
 413         rtnl_lock();
 414         list_for_each_entry(elem, &nf_hook_list, list) {
 415                 ret = nf_register_net_hook(net, elem);
 416                 if (ret && ret != -ENOENT)
 417                         goto out_undo;
 418         }
 419         rtnl_unlock();
 420         return 0;
 421
 422 out_undo:
 423         list_for_each_entry_continue_reverse(elem, &nf_hook_list, list)
 424                 nf_unregister_net_hook(net, elem);
 425         rtnl_unlock();
 426         return ret;
 427 }
 428
 429 static void nf_unregister_hook_list(struct net *net)
 430 {
 431         struct nf_hook_ops *elem;
 432
 433         rtnl_lock();
 434         list_for_each_entry(elem, &nf_hook_list, list)
 435                 nf_unregister_net_hook(net, elem);
 436         rtnl_unlock();
 437 }
 438
 439 static int __net_init netfilter_net_init(struct net *net)
 440 {
 441         int i, h, ret;
 442
 443         for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
 444                 for (h = 0; h < NF_MAX_HOOKS; h++)
 445                         INIT_LIST_HEAD(&net->nf.hooks[i][h]);
 446         }
 447
 448 #ifdef CONFIG_PROC_FS
 449         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 450                                                 net->proc_net);
 451         if (!net->nf.proc_netfilter) {
 452                 if (!net_eq(net, &init_net))
 453                         pr_err("cannot create netfilter proc entry");
 454
 455                 return -ENOMEM;
 456         }
 457 #endif
 458         ret = nf_register_hook_list(net);
 459         if (ret)
 460                 remove_proc_entry("netfilter", net->proc_net);
 461
 462         return ret;
 463 }
 464
 465 static void __net_exit netfilter_net_exit(struct net *net)
 466 {
 467         nf_unregister_hook_list(net);
 468         remove_proc_entry("netfilter", net->proc_net);
 469 }
 470
 471 static struct pernet_operations netfilter_net_ops = {
 472         .init = netfilter_net_init,
 473         .exit = netfilter_net_exit,
 474 };
 475
 476 int __init netfilter_init(void)
 477 {
 478         int ret;
 479
 480         ret = register_pernet_subsys(&netfilter_net_ops);
 481         if (ret < 0)
 482                 goto err;
 483
 484         ret = netfilter_log_init();
 485         if (ret < 0)
 486                 goto err_pernet;
 487
 488         return 0;
 489 err_pernet:
 490         unregister_pernet_subsys(&netfilter_net_ops);
 491 err:
 492         return ret;
 493 }