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/rcupdate.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28
  29 #include "nf_internals.h"
  30
  31 static DEFINE_MUTEX(afinfo_mutex);
  32
  33 const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO] __read_mostly;
  34 EXPORT_SYMBOL(nf_afinfo);
  35 const struct nf_ipv6_ops __rcu *nf_ipv6_ops __read_mostly;
  36 EXPORT_SYMBOL_GPL(nf_ipv6_ops);
  37
  38 DEFINE_PER_CPU(bool, nf_skb_duplicated);
  39 EXPORT_SYMBOL_GPL(nf_skb_duplicated);
  40
  41 int nf_register_afinfo(const struct nf_afinfo *afinfo)
  42 {
  43         mutex_lock(&afinfo_mutex);
  44         RCU_INIT_POINTER(nf_afinfo[afinfo->family], afinfo);
  45         mutex_unlock(&afinfo_mutex);
  46         return 0;
  47 }
  48 EXPORT_SYMBOL_GPL(nf_register_afinfo);
  49
  50 void nf_unregister_afinfo(const struct nf_afinfo *afinfo)
  51 {
  52         mutex_lock(&afinfo_mutex);
  53         RCU_INIT_POINTER(nf_afinfo[afinfo->family], NULL);
  54         mutex_unlock(&afinfo_mutex);
  55         synchronize_rcu();
  56 }
  57 EXPORT_SYMBOL_GPL(nf_unregister_afinfo);
  58
  59 #ifdef HAVE_JUMP_LABEL
  60 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
  61 EXPORT_SYMBOL(nf_hooks_needed);
  62 #endif
  63
  64 static DEFINE_MUTEX(nf_hook_mutex);
  65 #define nf_entry_dereference(e) \
  66         rcu_dereference_protected(e, lockdep_is_held(&nf_hook_mutex))
  67
  68 static struct nf_hook_entry __rcu **nf_hook_entry_head(struct net *net, const struct nf_hook_ops *reg)
  69 {
  70         if (reg->pf != NFPROTO_NETDEV)
  71                 return net->nf.hooks[reg->pf]+reg->hooknum;
  72
  73 #ifdef CONFIG_NETFILTER_INGRESS
  74         if (reg->hooknum == NF_NETDEV_INGRESS) {
  75                 if (reg->dev && dev_net(reg->dev) == net)
  76                         return &reg->dev->nf_hooks_ingress;
  77         }
  78 #endif
  79         return NULL;
  80 }
  81
  82 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
  83 {
  84         struct nf_hook_entry __rcu **pp;
  85         struct nf_hook_entry *entry, *p;
  86
  87         if (reg->pf == NFPROTO_NETDEV) {
  88 #ifndef CONFIG_NETFILTER_INGRESS
  89                 if (reg->hooknum == NF_NETDEV_INGRESS)
  90                         return -EOPNOTSUPP;
  91 #endif
  92                 if (reg->hooknum != NF_NETDEV_INGRESS ||
  93                     !reg->dev || dev_net(reg->dev) != net)
  94                         return -EINVAL;
  95         }
  96
  97         pp = nf_hook_entry_head(net, reg);
  98         if (!pp)
  99                 return -EINVAL;
 100
 101         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 102         if (!entry)
 103                 return -ENOMEM;
 104
 105         nf_hook_entry_init(entry, reg);
 106
 107         mutex_lock(&nf_hook_mutex);
 108
 109         /* Find the spot in the list */
 110         for (; (p = nf_entry_dereference(*pp)) != NULL; pp = &p->next) {
 111                 if (reg->priority < nf_hook_entry_priority(p))
 112                         break;
 113         }
 114         rcu_assign_pointer(entry->next, p);
 115         rcu_assign_pointer(*pp, entry);
 116
 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 static struct nf_hook_entry *
 130 __nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 131 {
 132         struct nf_hook_entry __rcu **pp;
 133         struct nf_hook_entry *p;
 134
 135         pp = nf_hook_entry_head(net, reg);
 136         if (WARN_ON_ONCE(!pp))
 137                 return NULL;
 138
 139         mutex_lock(&nf_hook_mutex);
 140         for (; (p = nf_entry_dereference(*pp)) != NULL; pp = &p->next) {
 141                 if (nf_hook_entry_ops(p) == reg) {
 142                         rcu_assign_pointer(*pp, p->next);
 143                         break;
 144                 }
 145         }
 146         mutex_unlock(&nf_hook_mutex);
 147         if (!p) {
 148                 WARN(1, "nf_unregister_net_hook: hook not found!\n");
 149                 return NULL;
 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
 159         return p;
 160 }
 161
 162 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 163 {
 164         struct nf_hook_entry *p = __nf_unregister_net_hook(net, reg);
 165         unsigned int nfq;
 166
 167         if (!p)
 168                 return;
 169
 170         synchronize_net();
 171
 172         /* other cpu might still process nfqueue verdict that used reg */
 173         nfq = nf_queue_nf_hook_drop(net);
 174         if (nfq)
 175                 synchronize_net();
 176         kfree(p);
 177 }
 178 EXPORT_SYMBOL(nf_unregister_net_hook);
 179
 180 int nf_register_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 181                           unsigned int n)
 182 {
 183         unsigned int i;
 184         int err = 0;
 185
 186         for (i = 0; i < n; i++) {
 187                 err = nf_register_net_hook(net, &reg[i]);
 188                 if (err)
 189                         goto err;
 190         }
 191         return err;
 192
 193 err:
 194         if (i > 0)
 195                 nf_unregister_net_hooks(net, reg, i);
 196         return err;
 197 }
 198 EXPORT_SYMBOL(nf_register_net_hooks);
 199
 200 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 201                              unsigned int hookcount)
 202 {
 203         struct nf_hook_entry *to_free[16];
 204         unsigned int i, n, nfq;
 205
 206         do {
 207                 n = min_t(unsigned int, hookcount, ARRAY_SIZE(to_free));
 208
 209                 for (i = 0; i < n; i++)
 210                         to_free[i] = __nf_unregister_net_hook(net, &reg[i]);
 211
 212                 synchronize_net();
 213
 214                 /* need 2nd synchronize_net() if nfqueue is used, skb
 215                  * can get reinjected right before nf_queue_hook_drop()
 216                  */
 217                 nfq = nf_queue_nf_hook_drop(net);
 218                 if (nfq)
 219                         synchronize_net();
 220
 221                 for (i = 0; i < n; i++)
 222                         kfree(to_free[i]);
 223
 224                 reg += n;
 225                 hookcount -= n;
 226         } while (hookcount > 0);
 227 }
 228 EXPORT_SYMBOL(nf_unregister_net_hooks);
 229
 230 /* Returns 1 if okfn() needs to be executed by the caller,
 231  * -EPERM for NF_DROP, 0 otherwise.  Caller must hold rcu_read_lock. */
 232 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
 233                  struct nf_hook_entry *entry)
 234 {
 235         unsigned int verdict;
 236         int ret;
 237
 238         do {
 239                 verdict = nf_hook_entry_hookfn(entry, skb, state);
 240                 switch (verdict & NF_VERDICT_MASK) {
 241                 case NF_ACCEPT:
 242                         entry = rcu_dereference(entry->next);
 243                         break;
 244                 case NF_DROP:
 245                         kfree_skb(skb);
 246                         ret = NF_DROP_GETERR(verdict);
 247                         if (ret == 0)
 248                                 ret = -EPERM;
 249                         return ret;
 250                 case NF_QUEUE:
 251                         ret = nf_queue(skb, state, &entry, verdict);
 252                         if (ret == 1 && entry)
 253                                 continue;
 254                         return ret;
 255                 default:
 256                         /* Implicit handling for NF_STOLEN, as well as any other
 257                          * non conventional verdicts.
 258                          */
 259                         return 0;
 260                 }
 261         } while (entry);
 262
 263         return 1;
 264 }
 265 EXPORT_SYMBOL(nf_hook_slow);
 266
 267
 268 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 269 {
 270         if (writable_len > skb->len)
 271                 return 0;
 272
 273         /* Not exclusive use of packet?  Must copy. */
 274         if (!skb_cloned(skb)) {
 275                 if (writable_len <= skb_headlen(skb))
 276                         return 1;
 277         } else if (skb_clone_writable(skb, writable_len))
 278                 return 1;
 279
 280         if (writable_len <= skb_headlen(skb))
 281                 writable_len = 0;
 282         else
 283                 writable_len -= skb_headlen(skb);
 284
 285         return !!__pskb_pull_tail(skb, writable_len);
 286 }
 287 EXPORT_SYMBOL(skb_make_writable);
 288
 289 /* This needs to be compiled in any case to avoid dependencies between the
 290  * nfnetlink_queue code and nf_conntrack.
 291  */
 292 struct nfnl_ct_hook __rcu *nfnl_ct_hook __read_mostly;
 293 EXPORT_SYMBOL_GPL(nfnl_ct_hook);
 294
 295 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 296 /* This does not belong here, but locally generated errors need it if connection
 297    tracking in use: without this, connection may not be in hash table, and hence
 298    manufactured ICMP or RST packets will not be associated with it. */
 299 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 300                 __rcu __read_mostly;
 301 EXPORT_SYMBOL(ip_ct_attach);
 302
 303 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 304 {
 305         void (*attach)(struct sk_buff *, const struct sk_buff *);
 306
 307         if (skb->_nfct) {
 308                 rcu_read_lock();
 309                 attach = rcu_dereference(ip_ct_attach);
 310                 if (attach)
 311                         attach(new, skb);
 312                 rcu_read_unlock();
 313         }
 314 }
 315 EXPORT_SYMBOL(nf_ct_attach);
 316
 317 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 318 EXPORT_SYMBOL(nf_ct_destroy);
 319
 320 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 321 {
 322         void (*destroy)(struct nf_conntrack *);
 323
 324         rcu_read_lock();
 325         destroy = rcu_dereference(nf_ct_destroy);
 326         BUG_ON(destroy == NULL);
 327         destroy(nfct);
 328         rcu_read_unlock();
 329 }
 330 EXPORT_SYMBOL(nf_conntrack_destroy);
 331
 332 /* Built-in default zone used e.g. by modules. */
 333 const struct nf_conntrack_zone nf_ct_zone_dflt = {
 334         .id     = NF_CT_DEFAULT_ZONE_ID,
 335         .dir    = NF_CT_DEFAULT_ZONE_DIR,
 336 };
 337 EXPORT_SYMBOL_GPL(nf_ct_zone_dflt);
 338 #endif /* CONFIG_NF_CONNTRACK */
 339
 340 #ifdef CONFIG_NF_NAT_NEEDED
 341 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 342 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 343 #endif
 344
 345 static int __net_init netfilter_net_init(struct net *net)
 346 {
 347         int i, h;
 348
 349         for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
 350                 for (h = 0; h < NF_MAX_HOOKS; h++)
 351                         RCU_INIT_POINTER(net->nf.hooks[i][h], NULL);
 352         }
 353
 354 #ifdef CONFIG_PROC_FS
 355         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 356                                                 net->proc_net);
 357         if (!net->nf.proc_netfilter) {
 358                 if (!net_eq(net, &init_net))
 359                         pr_err("cannot create netfilter proc entry");
 360
 361                 return -ENOMEM;
 362         }
 363 #endif
 364
 365         return 0;
 366 }
 367
 368 static void __net_exit netfilter_net_exit(struct net *net)
 369 {
 370         remove_proc_entry("netfilter", net->proc_net);
 371 }
 372
 373 static struct pernet_operations netfilter_net_ops = {
 374         .init = netfilter_net_init,
 375         .exit = netfilter_net_exit,
 376 };
 377
 378 int __init netfilter_init(void)
 379 {
 380         int ret;
 381
 382         ret = register_pernet_subsys(&netfilter_net_ops);
 383         if (ret < 0)
 384                 goto err;
 385
 386         ret = netfilter_log_init();
 387         if (ret < 0)
 388                 goto err_pernet;
 389
 390         return 0;
 391 err_pernet:
 392         unregister_pernet_subsys(&netfilter_net_ops);
 393 err:
 394         return ret;
 395 }