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 int nf_register_afinfo(const struct nf_afinfo *afinfo)
  38 {
  39         mutex_lock(&afinfo_mutex);
  40         RCU_INIT_POINTER(nf_afinfo[afinfo->family], afinfo);
  41         mutex_unlock(&afinfo_mutex);
  42         return 0;
  43 }
  44 EXPORT_SYMBOL_GPL(nf_register_afinfo);
  45
  46 void nf_unregister_afinfo(const struct nf_afinfo *afinfo)
  47 {
  48         mutex_lock(&afinfo_mutex);
  49         RCU_INIT_POINTER(nf_afinfo[afinfo->family], NULL);
  50         mutex_unlock(&afinfo_mutex);
  51         synchronize_rcu();
  52 }
  53 EXPORT_SYMBOL_GPL(nf_unregister_afinfo);
  54
  55 #ifdef HAVE_JUMP_LABEL
  56 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
  57 EXPORT_SYMBOL(nf_hooks_needed);
  58 #endif
  59
  60 static DEFINE_MUTEX(nf_hook_mutex);
  61
  62 static struct list_head *find_nf_hook_list(struct net *net,
  63                                            const struct nf_hook_ops *reg)
  64 {
  65         struct list_head *nf_hook_list = NULL;
  66
  67         if (reg->pf != NFPROTO_NETDEV)
  68                 nf_hook_list = &net->nf.hooks[reg->pf][reg->hooknum];
  69         else if (reg->hooknum == NF_NETDEV_INGRESS) {
  70 #ifdef CONFIG_NETFILTER_INGRESS
  71                 if (reg->dev && dev_net(reg->dev) == net)
  72                         nf_hook_list = &reg->dev->nf_hooks_ingress;
  73 #endif
  74         }
  75         return nf_hook_list;
  76 }
  77
  78 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
  79 {
  80         struct list_head *nf_hook_list;
  81         struct nf_hook_ops *elem, *new;
  82
  83         new = kzalloc(sizeof(*new), GFP_KERNEL);
  84         if (!new)
  85                 return -ENOMEM;
  86
  87         new->hook     = reg->hook;
  88         new->dev      = reg->dev;
  89         new->owner    = reg->owner;
  90         new->priv     = reg->priv;
  91         new->pf       = reg->pf;
  92         new->hooknum  = reg->hooknum;
  93         new->priority = reg->priority;
  94
  95         nf_hook_list = find_nf_hook_list(net, reg);
  96         if (!nf_hook_list)
  97                 return -ENOENT;
  98
  99         mutex_lock(&nf_hook_mutex);
 100         list_for_each_entry(elem, nf_hook_list, list) {
 101                 if (reg->priority < elem->priority)
 102                         break;
 103         }
 104         list_add_rcu(&new->list, elem->list.prev);
 105         mutex_unlock(&nf_hook_mutex);
 106 #ifdef CONFIG_NETFILTER_INGRESS
 107         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 108                 net_inc_ingress_queue();
 109 #endif
 110 #ifdef HAVE_JUMP_LABEL
 111         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
 112 #endif
 113         return 0;
 114 }
 115 EXPORT_SYMBOL(nf_register_net_hook);
 116
 117 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 118 {
 119         struct list_head *nf_hook_list;
 120         struct nf_hook_ops *elem;
 121
 122         nf_hook_list = find_nf_hook_list(net, reg);
 123         if (!nf_hook_list)
 124                 return;
 125
 126         mutex_lock(&nf_hook_mutex);
 127         list_for_each_entry(elem, nf_hook_list, list) {
 128                 if ((reg->hook     == elem->hook) &&
 129                     (reg->dev      == elem->dev) &&
 130                     (reg->owner    == elem->owner) &&
 131                     (reg->priv     == elem->priv) &&
 132                     (reg->pf       == elem->pf) &&
 133                     (reg->hooknum  == elem->hooknum) &&
 134                     (reg->priority == elem->priority)) {
 135                         list_del_rcu(&elem->list);
 136                         break;
 137                 }
 138         }
 139         mutex_unlock(&nf_hook_mutex);
 140         if (&elem->list == nf_hook_list) {
 141                 WARN(1, "nf_unregister_net_hook: hook not found!\n");
 142                 return;
 143         }
 144 #ifdef CONFIG_NETFILTER_INGRESS
 145         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 146                 net_dec_ingress_queue();
 147 #endif
 148 #ifdef HAVE_JUMP_LABEL
 149         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
 150 #endif
 151         synchronize_net();
 152         nf_queue_nf_hook_drop(elem);
 153         kfree(elem);
 154 }
 155 EXPORT_SYMBOL(nf_unregister_net_hook);
 156
 157 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 158                           unsigned int n)
 159 {
 160         unsigned int i;
 161         int err = 0;
 162
 163         for (i = 0; i < n; i++) {
 164                 err = nf_register_net_hook(net, &reg[i]);
 165                 if (err)
 166                         goto err;
 167         }
 168         return err;
 169
 170 err:
 171         if (i > 0)
 172                 nf_unregister_net_hooks(net, reg, i);
 173         return err;
 174 }
 175 EXPORT_SYMBOL(nf_register_net_hooks);
 176
 177 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 178                              unsigned int n)
 179 {
 180         while (n-- > 0)
 181                 nf_unregister_net_hook(net, &reg[n]);
 182 }
 183 EXPORT_SYMBOL(nf_unregister_net_hooks);
 184
 185 static LIST_HEAD(nf_hook_list);
 186
 187 int nf_register_hook(struct nf_hook_ops *reg)
 188 {
 189         struct net *net, *last;
 190         int ret;
 191
 192         rtnl_lock();
 193         for_each_net(net) {
 194                 ret = nf_register_net_hook(net, reg);
 195                 if (ret && ret != -ENOENT)
 196                         goto rollback;
 197         }
 198         list_add_tail(&reg->list, &nf_hook_list);
 199         rtnl_unlock();
 200
 201         return 0;
 202 rollback:
 203         last = net;
 204         for_each_net(net) {
 205                 if (net == last)
 206                         break;
 207                 nf_unregister_net_hook(net, reg);
 208         }
 209         rtnl_unlock();
 210         return ret;
 211 }
 212 EXPORT_SYMBOL(nf_register_hook);
 213
 214 void nf_unregister_hook(struct nf_hook_ops *reg)
 215 {
 216         struct net *net;
 217
 218         rtnl_lock();
 219         list_del(&reg->list);
 220         for_each_net(net)
 221                 nf_unregister_net_hook(net, reg);
 222         rtnl_unlock();
 223 }
 224 EXPORT_SYMBOL(nf_unregister_hook);
 225
 226 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 227 {
 228         unsigned int i;
 229         int err = 0;
 230
 231         for (i = 0; i < n; i++) {
 232                 err = nf_register_hook(&reg[i]);
 233                 if (err)
 234                         goto err;
 235         }
 236         return err;
 237
 238 err:
 239         if (i > 0)
 240                 nf_unregister_hooks(reg, i);
 241         return err;
 242 }
 243 EXPORT_SYMBOL(nf_register_hooks);
 244
 245 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 246 {
 247         while (n-- > 0)
 248                 nf_unregister_hook(&reg[n]);
 249 }
 250 EXPORT_SYMBOL(nf_unregister_hooks);
 251
 252 unsigned int nf_iterate(struct list_head *head,
 253                         struct sk_buff *skb,
 254                         struct nf_hook_state *state,
 255                         struct nf_hook_ops **elemp)
 256 {
 257         unsigned int verdict;
 258
 259         /*
 260          * The caller must not block between calls to this
 261          * function because of risk of continuing from deleted element.
 262          */
 263         list_for_each_entry_continue_rcu((*elemp), head, list) {
 264                 if (state->thresh > (*elemp)->priority)
 265                         continue;
 266
 267                 /* Optimization: we don't need to hold module
 268                    reference here, since function can't sleep. --RR */
 269 repeat:
 270                 verdict = (*elemp)->hook(*elemp, skb, state);
 271                 if (verdict != NF_ACCEPT) {
 272 #ifdef CONFIG_NETFILTER_DEBUG
 273                         if (unlikely((verdict & NF_VERDICT_MASK)
 274                                                         > NF_MAX_VERDICT)) {
 275                                 NFDEBUG("Evil return from %p(%u).\n",
 276                                         (*elemp)->hook, state->hook);
 277                                 continue;
 278                         }
 279 #endif
 280                         if (verdict != NF_REPEAT)
 281                                 return verdict;
 282                         goto repeat;
 283                 }
 284         }
 285         return NF_ACCEPT;
 286 }
 287
 288
 289 /* Returns 1 if okfn() needs to be executed by the caller,
 290  * -EPERM for NF_DROP, 0 otherwise. */
 291 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state)
 292 {
 293         struct nf_hook_ops *elem;
 294         unsigned int verdict;
 295         int ret = 0;
 296
 297         /* We may already have this, but read-locks nest anyway */
 298         rcu_read_lock();
 299
 300         elem = list_entry_rcu(state->hook_list, struct nf_hook_ops, list);
 301 next_hook:
 302         verdict = nf_iterate(state->hook_list, skb, state, &elem);
 303         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 304                 ret = 1;
 305         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 306                 kfree_skb(skb);
 307                 ret = NF_DROP_GETERR(verdict);
 308                 if (ret == 0)
 309                         ret = -EPERM;
 310         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 311                 int err = nf_queue(skb, elem, state,
 312                                    verdict >> NF_VERDICT_QBITS);
 313                 if (err < 0) {
 314                         if (err == -ECANCELED)
 315                                 goto next_hook;
 316                         if (err == -ESRCH &&
 317                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 318                                 goto next_hook;
 319                         kfree_skb(skb);
 320                 }
 321         }
 322         rcu_read_unlock();
 323         return ret;
 324 }
 325 EXPORT_SYMBOL(nf_hook_slow);
 326
 327
 328 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 329 {
 330         if (writable_len > skb->len)
 331                 return 0;
 332
 333         /* Not exclusive use of packet?  Must copy. */
 334         if (!skb_cloned(skb)) {
 335                 if (writable_len <= skb_headlen(skb))
 336                         return 1;
 337         } else if (skb_clone_writable(skb, writable_len))
 338                 return 1;
 339
 340         if (writable_len <= skb_headlen(skb))
 341                 writable_len = 0;
 342         else
 343                 writable_len -= skb_headlen(skb);
 344
 345         return !!__pskb_pull_tail(skb, writable_len);
 346 }
 347 EXPORT_SYMBOL(skb_make_writable);
 348
 349 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 350 /* This does not belong here, but locally generated errors need it if connection
 351    tracking in use: without this, connection may not be in hash table, and hence
 352    manufactured ICMP or RST packets will not be associated with it. */
 353 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 354                 __rcu __read_mostly;
 355 EXPORT_SYMBOL(ip_ct_attach);
 356
 357 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 358 {
 359         void (*attach)(struct sk_buff *, const struct sk_buff *);
 360
 361         if (skb->nfct) {
 362                 rcu_read_lock();
 363                 attach = rcu_dereference(ip_ct_attach);
 364                 if (attach)
 365                         attach(new, skb);
 366                 rcu_read_unlock();
 367         }
 368 }
 369 EXPORT_SYMBOL(nf_ct_attach);
 370
 371 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 372 EXPORT_SYMBOL(nf_ct_destroy);
 373
 374 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 375 {
 376         void (*destroy)(struct nf_conntrack *);
 377
 378         rcu_read_lock();
 379         destroy = rcu_dereference(nf_ct_destroy);
 380         BUG_ON(destroy == NULL);
 381         destroy(nfct);
 382         rcu_read_unlock();
 383 }
 384 EXPORT_SYMBOL(nf_conntrack_destroy);
 385
 386 struct nfq_ct_hook __rcu *nfq_ct_hook __read_mostly;
 387 EXPORT_SYMBOL_GPL(nfq_ct_hook);
 388
 389 struct nfq_ct_nat_hook __rcu *nfq_ct_nat_hook __read_mostly;
 390 EXPORT_SYMBOL_GPL(nfq_ct_nat_hook);
 391
 392 #endif /* CONFIG_NF_CONNTRACK */
 393
 394 #ifdef CONFIG_NF_NAT_NEEDED
 395 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 396 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 397 #endif
 398
 399 static int nf_register_hook_list(struct net *net)
 400 {
 401         struct nf_hook_ops *elem;
 402         int ret;
 403
 404         rtnl_lock();
 405         list_for_each_entry(elem, &nf_hook_list, list) {
 406                 ret = nf_register_net_hook(net, elem);
 407                 if (ret && ret != -ENOENT)
 408                         goto out_undo;
 409         }
 410         rtnl_unlock();
 411         return 0;
 412
 413 out_undo:
 414         list_for_each_entry_continue_reverse(elem, &nf_hook_list, list)
 415                 nf_unregister_net_hook(net, elem);
 416         rtnl_unlock();
 417         return ret;
 418 }
 419
 420 static void nf_unregister_hook_list(struct net *net)
 421 {
 422         struct nf_hook_ops *elem;
 423
 424         rtnl_lock();
 425         list_for_each_entry(elem, &nf_hook_list, list)
 426                 nf_unregister_net_hook(net, elem);
 427         rtnl_unlock();
 428 }
 429
 430 static int __net_init netfilter_net_init(struct net *net)
 431 {
 432         int i, h, ret;
 433
 434         for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
 435                 for (h = 0; h < NF_MAX_HOOKS; h++)
 436                         INIT_LIST_HEAD(&net->nf.hooks[i][h]);
 437         }
 438
 439 #ifdef CONFIG_PROC_FS
 440         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 441                                                 net->proc_net);
 442         if (!net->nf.proc_netfilter) {
 443                 if (!net_eq(net, &init_net))
 444                         pr_err("cannot create netfilter proc entry");
 445
 446                 return -ENOMEM;
 447         }
 448 #endif
 449         ret = nf_register_hook_list(net);
 450         if (ret)
 451                 remove_proc_entry("netfilter", net->proc_net);
 452
 453         return ret;
 454 }
 455
 456 static void __net_exit netfilter_net_exit(struct net *net)
 457 {
 458         nf_unregister_hook_list(net);
 459         remove_proc_entry("netfilter", net->proc_net);
 460 }
 461
 462 static struct pernet_operations netfilter_net_ops = {
 463         .init = netfilter_net_init,
 464         .exit = netfilter_net_exit,
 465 };
 466
 467 int __init netfilter_init(void)
 468 {
 469         int ret;
 470
 471         ret = register_pernet_subsys(&netfilter_net_ops);
 472         if (ret < 0)
 473                 goto err;
 474
 475         ret = netfilter_log_init();
 476         if (ret < 0)
 477                 goto err_pernet;
 478
 479         return 0;
 480 err_pernet:
 481         unregister_pernet_subsys(&netfilter_net_ops);
 482 err:
 483         return ret;
 484 }