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 struct list_head nf_hooks[NFPROTO_NUMPROTO][NF_MAX_HOOKS] __read_mostly;
  56 EXPORT_SYMBOL(nf_hooks);
  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 int nf_register_hook(struct nf_hook_ops *reg)
  66 {
  67         struct list_head *nf_hook_list;
  68         struct nf_hook_ops *elem;
  69
  70         mutex_lock(&nf_hook_mutex);
  71         switch (reg->pf) {
  72         case NFPROTO_NETDEV:
  73 #ifdef CONFIG_NETFILTER_INGRESS
  74                 if (reg->hooknum == NF_NETDEV_INGRESS) {
  75                         BUG_ON(reg->dev == NULL);
  76                         nf_hook_list = &reg->dev->nf_hooks_ingress;
  77                         net_inc_ingress_queue();
  78                         break;
  79                 }
  80 #endif
  81                 /* Fall through. */
  82         default:
  83                 nf_hook_list = &nf_hooks[reg->pf][reg->hooknum];
  84                 break;
  85         }
  86
  87         list_for_each_entry(elem, nf_hook_list, list) {
  88                 if (reg->priority < elem->priority)
  89                         break;
  90         }
  91         list_add_rcu(&reg->list, elem->list.prev);
  92         mutex_unlock(&nf_hook_mutex);
  93 #ifdef HAVE_JUMP_LABEL
  94         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
  95 #endif
  96         return 0;
  97 }
  98 EXPORT_SYMBOL(nf_register_hook);
  99
 100 void nf_unregister_hook(struct nf_hook_ops *reg)
 101 {
 102         mutex_lock(&nf_hook_mutex);
 103         list_del_rcu(&reg->list);
 104         mutex_unlock(&nf_hook_mutex);
 105         switch (reg->pf) {
 106         case NFPROTO_NETDEV:
 107 #ifdef CONFIG_NETFILTER_INGRESS
 108                 if (reg->hooknum == NF_NETDEV_INGRESS) {
 109                         net_dec_ingress_queue();
 110                         break;
 111                 }
 112                 break;
 113 #endif
 114         default:
 115                 break;
 116         }
 117 #ifdef HAVE_JUMP_LABEL
 118         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
 119 #endif
 120         synchronize_net();
 121         nf_queue_nf_hook_drop(reg);
 122 }
 123 EXPORT_SYMBOL(nf_unregister_hook);
 124
 125 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 126 {
 127         unsigned int i;
 128         int err = 0;
 129
 130         for (i = 0; i < n; i++) {
 131                 err = nf_register_hook(&reg[i]);
 132                 if (err)
 133                         goto err;
 134         }
 135         return err;
 136
 137 err:
 138         if (i > 0)
 139                 nf_unregister_hooks(reg, i);
 140         return err;
 141 }
 142 EXPORT_SYMBOL(nf_register_hooks);
 143
 144 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 145 {
 146         while (n-- > 0)
 147                 nf_unregister_hook(&reg[n]);
 148 }
 149 EXPORT_SYMBOL(nf_unregister_hooks);
 150
 151 unsigned int nf_iterate(struct list_head *head,
 152                         struct sk_buff *skb,
 153                         struct nf_hook_state *state,
 154                         struct nf_hook_ops **elemp)
 155 {
 156         unsigned int verdict;
 157
 158         /*
 159          * The caller must not block between calls to this
 160          * function because of risk of continuing from deleted element.
 161          */
 162         list_for_each_entry_continue_rcu((*elemp), head, list) {
 163                 if (state->thresh > (*elemp)->priority)
 164                         continue;
 165
 166                 /* Optimization: we don't need to hold module
 167                    reference here, since function can't sleep. --RR */
 168 repeat:
 169                 verdict = (*elemp)->hook(*elemp, skb, state);
 170                 if (verdict != NF_ACCEPT) {
 171 #ifdef CONFIG_NETFILTER_DEBUG
 172                         if (unlikely((verdict & NF_VERDICT_MASK)
 173                                                         > NF_MAX_VERDICT)) {
 174                                 NFDEBUG("Evil return from %p(%u).\n",
 175                                         (*elemp)->hook, state->hook);
 176                                 continue;
 177                         }
 178 #endif
 179                         if (verdict != NF_REPEAT)
 180                                 return verdict;
 181                         goto repeat;
 182                 }
 183         }
 184         return NF_ACCEPT;
 185 }
 186
 187
 188 /* Returns 1 if okfn() needs to be executed by the caller,
 189  * -EPERM for NF_DROP, 0 otherwise. */
 190 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state)
 191 {
 192         struct nf_hook_ops *elem;
 193         unsigned int verdict;
 194         int ret = 0;
 195
 196         /* We may already have this, but read-locks nest anyway */
 197         rcu_read_lock();
 198
 199         elem = list_entry_rcu(state->hook_list, struct nf_hook_ops, list);
 200 next_hook:
 201         verdict = nf_iterate(state->hook_list, skb, state, &elem);
 202         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 203                 ret = 1;
 204         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 205                 kfree_skb(skb);
 206                 ret = NF_DROP_GETERR(verdict);
 207                 if (ret == 0)
 208                         ret = -EPERM;
 209         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 210                 int err = nf_queue(skb, elem, state,
 211                                    verdict >> NF_VERDICT_QBITS);
 212                 if (err < 0) {
 213                         if (err == -ECANCELED)
 214                                 goto next_hook;
 215                         if (err == -ESRCH &&
 216                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 217                                 goto next_hook;
 218                         kfree_skb(skb);
 219                 }
 220         }
 221         rcu_read_unlock();
 222         return ret;
 223 }
 224 EXPORT_SYMBOL(nf_hook_slow);
 225
 226
 227 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 228 {
 229         if (writable_len > skb->len)
 230                 return 0;
 231
 232         /* Not exclusive use of packet?  Must copy. */
 233         if (!skb_cloned(skb)) {
 234                 if (writable_len <= skb_headlen(skb))
 235                         return 1;
 236         } else if (skb_clone_writable(skb, writable_len))
 237                 return 1;
 238
 239         if (writable_len <= skb_headlen(skb))
 240                 writable_len = 0;
 241         else
 242                 writable_len -= skb_headlen(skb);
 243
 244         return !!__pskb_pull_tail(skb, writable_len);
 245 }
 246 EXPORT_SYMBOL(skb_make_writable);
 247
 248 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 249 /* This does not belong here, but locally generated errors need it if connection
 250    tracking in use: without this, connection may not be in hash table, and hence
 251    manufactured ICMP or RST packets will not be associated with it. */
 252 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 253                 __rcu __read_mostly;
 254 EXPORT_SYMBOL(ip_ct_attach);
 255
 256 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 257 {
 258         void (*attach)(struct sk_buff *, const struct sk_buff *);
 259
 260         if (skb->nfct) {
 261                 rcu_read_lock();
 262                 attach = rcu_dereference(ip_ct_attach);
 263                 if (attach)
 264                         attach(new, skb);
 265                 rcu_read_unlock();
 266         }
 267 }
 268 EXPORT_SYMBOL(nf_ct_attach);
 269
 270 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 271 EXPORT_SYMBOL(nf_ct_destroy);
 272
 273 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 274 {
 275         void (*destroy)(struct nf_conntrack *);
 276
 277         rcu_read_lock();
 278         destroy = rcu_dereference(nf_ct_destroy);
 279         BUG_ON(destroy == NULL);
 280         destroy(nfct);
 281         rcu_read_unlock();
 282 }
 283 EXPORT_SYMBOL(nf_conntrack_destroy);
 284
 285 struct nfq_ct_hook __rcu *nfq_ct_hook __read_mostly;
 286 EXPORT_SYMBOL_GPL(nfq_ct_hook);
 287
 288 struct nfq_ct_nat_hook __rcu *nfq_ct_nat_hook __read_mostly;
 289 EXPORT_SYMBOL_GPL(nfq_ct_nat_hook);
 290
 291 #endif /* CONFIG_NF_CONNTRACK */
 292
 293 #ifdef CONFIG_NF_NAT_NEEDED
 294 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 295 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 296 #endif
 297
 298 static int __net_init netfilter_net_init(struct net *net)
 299 {
 300 #ifdef CONFIG_PROC_FS
 301         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 302                                                 net->proc_net);
 303         if (!net->nf.proc_netfilter) {
 304                 if (!net_eq(net, &init_net))
 305                         pr_err("cannot create netfilter proc entry");
 306
 307                 return -ENOMEM;
 308         }
 309 #endif
 310         return 0;
 311 }
 312
 313 static void __net_exit netfilter_net_exit(struct net *net)
 314 {
 315         remove_proc_entry("netfilter", net->proc_net);
 316 }
 317
 318 static struct pernet_operations netfilter_net_ops = {
 319         .init = netfilter_net_init,
 320         .exit = netfilter_net_exit,
 321 };
 322
 323 int __init netfilter_init(void)
 324 {
 325         int i, h, ret;
 326
 327         for (i = 0; i < ARRAY_SIZE(nf_hooks); i++) {
 328                 for (h = 0; h < NF_MAX_HOOKS; h++)
 329                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 330         }
 331
 332         ret = register_pernet_subsys(&netfilter_net_ops);
 333         if (ret < 0)
 334                 goto err;
 335
 336         ret = netfilter_log_init();
 337         if (ret < 0)
 338                 goto err_pernet;
 339
 340         return 0;
 341 err_pernet:
 342         unregister_pernet_subsys(&netfilter_net_ops);
 343 err:
 344         return ret;
 345 }