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 *find_nf_hook_list(struct net *net,
  66                                            const struct nf_hook_ops *reg)
  67 {
  68         struct list_head *nf_hook_list = NULL;
  69
  70         if (reg->pf != NFPROTO_NETDEV)
  71                 nf_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                         nf_hook_list = &reg->dev->nf_hooks_ingress;
  76 #endif
  77         }
  78         return nf_hook_list;
  79 }
  80
  81 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
  82 {
  83         struct list_head *nf_hook_list;
  84         struct nf_hook_ops *elem, *new;
  85
  86         new = kzalloc(sizeof(*new), GFP_KERNEL);
  87         if (!new)
  88                 return -ENOMEM;
  89
  90         new->hook     = reg->hook;
  91         new->dev      = reg->dev;
  92         new->owner    = reg->owner;
  93         new->priv     = reg->priv;
  94         new->pf       = reg->pf;
  95         new->hooknum  = reg->hooknum;
  96         new->priority = reg->priority;
  97
  98         nf_hook_list = find_nf_hook_list(net, reg);
  99         if (!nf_hook_list) {
 100                 kfree(new);
 101                 return -ENOENT;
 102         }
 103
 104         mutex_lock(&nf_hook_mutex);
 105         list_for_each_entry(elem, nf_hook_list, list) {
 106                 if (reg->priority < elem->priority)
 107                         break;
 108         }
 109         list_add_rcu(&new->list, elem->list.prev);
 110         mutex_unlock(&nf_hook_mutex);
 111 #ifdef CONFIG_NETFILTER_INGRESS
 112         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 113                 net_inc_ingress_queue();
 114 #endif
 115 #ifdef HAVE_JUMP_LABEL
 116         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
 117 #endif
 118         return 0;
 119 }
 120 EXPORT_SYMBOL(nf_register_net_hook);
 121
 122 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 123 {
 124         struct list_head *nf_hook_list;
 125         struct nf_hook_ops *elem;
 126
 127         nf_hook_list = find_nf_hook_list(net, reg);
 128         if (!nf_hook_list)
 129                 return;
 130
 131         mutex_lock(&nf_hook_mutex);
 132         list_for_each_entry(elem, nf_hook_list, list) {
 133                 if ((reg->hook     == elem->hook) &&
 134                     (reg->dev      == elem->dev) &&
 135                     (reg->owner    == elem->owner) &&
 136                     (reg->priv     == elem->priv) &&
 137                     (reg->pf       == elem->pf) &&
 138                     (reg->hooknum  == elem->hooknum) &&
 139                     (reg->priority == elem->priority)) {
 140                         list_del_rcu(&elem->list);
 141                         break;
 142                 }
 143         }
 144         mutex_unlock(&nf_hook_mutex);
 145         if (&elem->list == nf_hook_list) {
 146                 WARN(1, "nf_unregister_net_hook: hook not found!\n");
 147                 return;
 148         }
 149 #ifdef CONFIG_NETFILTER_INGRESS
 150         if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 151                 net_dec_ingress_queue();
 152 #endif
 153 #ifdef HAVE_JUMP_LABEL
 154         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
 155 #endif
 156         synchronize_net();
 157         nf_queue_nf_hook_drop(elem);
 158         kfree(elem);
 159 }
 160 EXPORT_SYMBOL(nf_unregister_net_hook);
 161
 162 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 163                           unsigned int n)
 164 {
 165         unsigned int i;
 166         int err = 0;
 167
 168         for (i = 0; i < n; i++) {
 169                 err = nf_register_net_hook(net, &reg[i]);
 170                 if (err)
 171                         goto err;
 172         }
 173         return err;
 174
 175 err:
 176         if (i > 0)
 177                 nf_unregister_net_hooks(net, reg, i);
 178         return err;
 179 }
 180 EXPORT_SYMBOL(nf_register_net_hooks);
 181
 182 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 183                              unsigned int n)
 184 {
 185         while (n-- > 0)
 186                 nf_unregister_net_hook(net, &reg[n]);
 187 }
 188 EXPORT_SYMBOL(nf_unregister_net_hooks);
 189
 190 static LIST_HEAD(nf_hook_list);
 191
 192 int nf_register_hook(struct nf_hook_ops *reg)
 193 {
 194         struct net *net, *last;
 195         int ret;
 196
 197         rtnl_lock();
 198         for_each_net(net) {
 199                 ret = nf_register_net_hook(net, reg);
 200                 if (ret && ret != -ENOENT)
 201                         goto rollback;
 202         }
 203         list_add_tail(&reg->list, &nf_hook_list);
 204         rtnl_unlock();
 205
 206         return 0;
 207 rollback:
 208         last = net;
 209         for_each_net(net) {
 210                 if (net == last)
 211                         break;
 212                 nf_unregister_net_hook(net, reg);
 213         }
 214         rtnl_unlock();
 215         return ret;
 216 }
 217 EXPORT_SYMBOL(nf_register_hook);
 218
 219 void nf_unregister_hook(struct nf_hook_ops *reg)
 220 {
 221         struct net *net;
 222
 223         rtnl_lock();
 224         list_del(&reg->list);
 225         for_each_net(net)
 226                 nf_unregister_net_hook(net, reg);
 227         rtnl_unlock();
 228 }
 229 EXPORT_SYMBOL(nf_unregister_hook);
 230
 231 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 232 {
 233         unsigned int i;
 234         int err = 0;
 235
 236         for (i = 0; i < n; i++) {
 237                 err = nf_register_hook(&reg[i]);
 238                 if (err)
 239                         goto err;
 240         }
 241         return err;
 242
 243 err:
 244         if (i > 0)
 245                 nf_unregister_hooks(reg, i);
 246         return err;
 247 }
 248 EXPORT_SYMBOL(nf_register_hooks);
 249
 250 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 251 {
 252         while (n-- > 0)
 253                 nf_unregister_hook(&reg[n]);
 254 }
 255 EXPORT_SYMBOL(nf_unregister_hooks);
 256
 257 unsigned int nf_iterate(struct list_head *head,
 258                         struct sk_buff *skb,
 259                         struct nf_hook_state *state,
 260                         struct nf_hook_ops **elemp)
 261 {
 262         unsigned int verdict;
 263
 264         /*
 265          * The caller must not block between calls to this
 266          * function because of risk of continuing from deleted element.
 267          */
 268         list_for_each_entry_continue_rcu((*elemp), head, list) {
 269                 if (state->thresh > (*elemp)->priority)
 270                         continue;
 271
 272                 /* Optimization: we don't need to hold module
 273                    reference here, since function can't sleep. --RR */
 274 repeat:
 275                 verdict = (*elemp)->hook(*elemp, skb, state);
 276                 if (verdict != NF_ACCEPT) {
 277 #ifdef CONFIG_NETFILTER_DEBUG
 278                         if (unlikely((verdict & NF_VERDICT_MASK)
 279                                                         > NF_MAX_VERDICT)) {
 280                                 NFDEBUG("Evil return from %p(%u).\n",
 281                                         (*elemp)->hook, state->hook);
 282                                 continue;
 283                         }
 284 #endif
 285                         if (verdict != NF_REPEAT)
 286                                 return verdict;
 287                         goto repeat;
 288                 }
 289         }
 290         return NF_ACCEPT;
 291 }
 292
 293
 294 /* Returns 1 if okfn() needs to be executed by the caller,
 295  * -EPERM for NF_DROP, 0 otherwise. */
 296 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state)
 297 {
 298         struct nf_hook_ops *elem;
 299         unsigned int verdict;
 300         int ret = 0;
 301
 302         /* We may already have this, but read-locks nest anyway */
 303         rcu_read_lock();
 304
 305         elem = list_entry_rcu(state->hook_list, struct nf_hook_ops, list);
 306 next_hook:
 307         verdict = nf_iterate(state->hook_list, skb, state, &elem);
 308         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 309                 ret = 1;
 310         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 311                 kfree_skb(skb);
 312                 ret = NF_DROP_GETERR(verdict);
 313                 if (ret == 0)
 314                         ret = -EPERM;
 315         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 316                 int err = nf_queue(skb, elem, state,
 317                                    verdict >> NF_VERDICT_QBITS);
 318                 if (err < 0) {
 319                         if (err == -ECANCELED)
 320                                 goto next_hook;
 321                         if (err == -ESRCH &&
 322                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 323                                 goto next_hook;
 324                         kfree_skb(skb);
 325                 }
 326         }
 327         rcu_read_unlock();
 328         return ret;
 329 }
 330 EXPORT_SYMBOL(nf_hook_slow);
 331
 332
 333 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 334 {
 335         if (writable_len > skb->len)
 336                 return 0;
 337
 338         /* Not exclusive use of packet?  Must copy. */
 339         if (!skb_cloned(skb)) {
 340                 if (writable_len <= skb_headlen(skb))
 341                         return 1;
 342         } else if (skb_clone_writable(skb, writable_len))
 343                 return 1;
 344
 345         if (writable_len <= skb_headlen(skb))
 346                 writable_len = 0;
 347         else
 348                 writable_len -= skb_headlen(skb);
 349
 350         return !!__pskb_pull_tail(skb, writable_len);
 351 }
 352 EXPORT_SYMBOL(skb_make_writable);
 353
 354 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 355 /* This does not belong here, but locally generated errors need it if connection
 356    tracking in use: without this, connection may not be in hash table, and hence
 357    manufactured ICMP or RST packets will not be associated with it. */
 358 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 359                 __rcu __read_mostly;
 360 EXPORT_SYMBOL(ip_ct_attach);
 361
 362 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 363 {
 364         void (*attach)(struct sk_buff *, const struct sk_buff *);
 365
 366         if (skb->nfct) {
 367                 rcu_read_lock();
 368                 attach = rcu_dereference(ip_ct_attach);
 369                 if (attach)
 370                         attach(new, skb);
 371                 rcu_read_unlock();
 372         }
 373 }
 374 EXPORT_SYMBOL(nf_ct_attach);
 375
 376 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 377 EXPORT_SYMBOL(nf_ct_destroy);
 378
 379 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 380 {
 381         void (*destroy)(struct nf_conntrack *);
 382
 383         rcu_read_lock();
 384         destroy = rcu_dereference(nf_ct_destroy);
 385         BUG_ON(destroy == NULL);
 386         destroy(nfct);
 387         rcu_read_unlock();
 388 }
 389 EXPORT_SYMBOL(nf_conntrack_destroy);
 390
 391 struct nfq_ct_hook __rcu *nfq_ct_hook __read_mostly;
 392 EXPORT_SYMBOL_GPL(nfq_ct_hook);
 393
 394 struct nfq_ct_nat_hook __rcu *nfq_ct_nat_hook __read_mostly;
 395 EXPORT_SYMBOL_GPL(nfq_ct_nat_hook);
 396
 397 #endif /* CONFIG_NF_CONNTRACK */
 398
 399 #ifdef CONFIG_NF_NAT_NEEDED
 400 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 401 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 402 #endif
 403
 404 static int nf_register_hook_list(struct net *net)
 405 {
 406         struct nf_hook_ops *elem;
 407         int ret;
 408
 409         rtnl_lock();
 410         list_for_each_entry(elem, &nf_hook_list, list) {
 411                 ret = nf_register_net_hook(net, elem);
 412                 if (ret && ret != -ENOENT)
 413                         goto out_undo;
 414         }
 415         rtnl_unlock();
 416         return 0;
 417
 418 out_undo:
 419         list_for_each_entry_continue_reverse(elem, &nf_hook_list, list)
 420                 nf_unregister_net_hook(net, elem);
 421         rtnl_unlock();
 422         return ret;
 423 }
 424
 425 static void nf_unregister_hook_list(struct net *net)
 426 {
 427         struct nf_hook_ops *elem;
 428
 429         rtnl_lock();
 430         list_for_each_entry(elem, &nf_hook_list, list)
 431                 nf_unregister_net_hook(net, elem);
 432         rtnl_unlock();
 433 }
 434
 435 static int __net_init netfilter_net_init(struct net *net)
 436 {
 437         int i, h, ret;
 438
 439         for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
 440                 for (h = 0; h < NF_MAX_HOOKS; h++)
 441                         INIT_LIST_HEAD(&net->nf.hooks[i][h]);
 442         }
 443
 444 #ifdef CONFIG_PROC_FS
 445         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 446                                                 net->proc_net);
 447         if (!net->nf.proc_netfilter) {
 448                 if (!net_eq(net, &init_net))
 449                         pr_err("cannot create netfilter proc entry");
 450
 451                 return -ENOMEM;
 452         }
 453 #endif
 454         ret = nf_register_hook_list(net);
 455         if (ret)
 456                 remove_proc_entry("netfilter", net->proc_net);
 457
 458         return ret;
 459 }
 460
 461 static void __net_exit netfilter_net_exit(struct net *net)
 462 {
 463         nf_unregister_hook_list(net);
 464         remove_proc_entry("netfilter", net->proc_net);
 465 }
 466
 467 static struct pernet_operations netfilter_net_ops = {
 468         .init = netfilter_net_init,
 469         .exit = netfilter_net_exit,
 470 };
 471
 472 int __init netfilter_init(void)
 473 {
 474         int ret;
 475
 476         ret = register_pernet_subsys(&netfilter_net_ops);
 477         if (ret < 0)
 478                 goto err;
 479
 480         ret = netfilter_log_init();
 481         if (ret < 0)
 482                 goto err_pernet;
 483
 484         return 0;
 485 err_pernet:
 486         unregister_pernet_subsys(&netfilter_net_ops);
 487 err:
 488         return ret;
 489 }