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