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/inetdevice.h>
  21 #include <linux/proc_fs.h>
  22 #include <linux/mutex.h>
  23 #include <linux/slab.h>
  24 #include <net/net_namespace.h>
  25 #include <net/sock.h>
  26
  27 #include "nf_internals.h"
  28
  29 static DEFINE_MUTEX(afinfo_mutex);
  30
  31 const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO] __read_mostly;
  32 EXPORT_SYMBOL(nf_afinfo);
  33 const struct nf_ipv6_ops __rcu *nf_ipv6_ops __read_mostly;
  34 EXPORT_SYMBOL_GPL(nf_ipv6_ops);
  35
  36 int nf_register_afinfo(const struct nf_afinfo *afinfo)
  37 {
  38         int err;
  39
  40         err = mutex_lock_interruptible(&afinfo_mutex);
  41         if (err < 0)
  42                 return err;
  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 struct list_head nf_hooks[NFPROTO_NUMPROTO][NF_MAX_HOOKS] __read_mostly;
  59 EXPORT_SYMBOL(nf_hooks);
  60
  61 #if defined(CONFIG_JUMP_LABEL)
  62 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
  63 EXPORT_SYMBOL(nf_hooks_needed);
  64 #endif
  65
  66 static DEFINE_MUTEX(nf_hook_mutex);
  67
  68 int nf_register_hook(struct nf_hook_ops *reg)
  69 {
  70         struct nf_hook_ops *elem;
  71         int err;
  72
  73         err = mutex_lock_interruptible(&nf_hook_mutex);
  74         if (err < 0)
  75                 return err;
  76         list_for_each_entry(elem, &nf_hooks[reg->pf][reg->hooknum], list) {
  77                 if (reg->priority < elem->priority)
  78                         break;
  79         }
  80         list_add_rcu(&reg->list, elem->list.prev);
  81         mutex_unlock(&nf_hook_mutex);
  82 #if defined(CONFIG_JUMP_LABEL)
  83         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
  84 #endif
  85         return 0;
  86 }
  87 EXPORT_SYMBOL(nf_register_hook);
  88
  89 void nf_unregister_hook(struct nf_hook_ops *reg)
  90 {
  91         mutex_lock(&nf_hook_mutex);
  92         list_del_rcu(&reg->list);
  93         mutex_unlock(&nf_hook_mutex);
  94 #if defined(CONFIG_JUMP_LABEL)
  95         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
  96 #endif
  97         synchronize_net();
  98 }
  99 EXPORT_SYMBOL(nf_unregister_hook);
 100
 101 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 102 {
 103         unsigned int i;
 104         int err = 0;
 105
 106         for (i = 0; i < n; i++) {
 107                 err = nf_register_hook(&reg[i]);
 108                 if (err)
 109                         goto err;
 110         }
 111         return err;
 112
 113 err:
 114         if (i > 0)
 115                 nf_unregister_hooks(reg, i);
 116         return err;
 117 }
 118 EXPORT_SYMBOL(nf_register_hooks);
 119
 120 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 121 {
 122         while (n-- > 0)
 123                 nf_unregister_hook(&reg[n]);
 124 }
 125 EXPORT_SYMBOL(nf_unregister_hooks);
 126
 127 unsigned int nf_iterate(struct list_head *head,
 128                         struct sk_buff *skb,
 129                         unsigned int hook,
 130                         const struct net_device *indev,
 131                         const struct net_device *outdev,
 132                         struct nf_hook_ops **elemp,
 133                         int (*okfn)(struct sk_buff *),
 134                         int hook_thresh)
 135 {
 136         unsigned int verdict;
 137
 138         /*
 139          * The caller must not block between calls to this
 140          * function because of risk of continuing from deleted element.
 141          */
 142         list_for_each_entry_continue_rcu((*elemp), head, list) {
 143                 if (hook_thresh > (*elemp)->priority)
 144                         continue;
 145
 146                 /* Optimization: we don't need to hold module
 147                    reference here, since function can't sleep. --RR */
 148 repeat:
 149                 verdict = (*elemp)->hook(*elemp, skb, indev, outdev, okfn);
 150                 if (verdict != NF_ACCEPT) {
 151 #ifdef CONFIG_NETFILTER_DEBUG
 152                         if (unlikely((verdict & NF_VERDICT_MASK)
 153                                                         > NF_MAX_VERDICT)) {
 154                                 NFDEBUG("Evil return from %p(%u).\n",
 155                                         (*elemp)->hook, hook);
 156                                 continue;
 157                         }
 158 #endif
 159                         if (verdict != NF_REPEAT)
 160                                 return verdict;
 161                         goto repeat;
 162                 }
 163         }
 164         return NF_ACCEPT;
 165 }
 166
 167
 168 /* Returns 1 if okfn() needs to be executed by the caller,
 169  * -EPERM for NF_DROP, 0 otherwise. */
 170 int nf_hook_slow(u_int8_t pf, unsigned int hook, struct sk_buff *skb,
 171                  struct net_device *indev,
 172                  struct net_device *outdev,
 173                  int (*okfn)(struct sk_buff *),
 174                  int hook_thresh)
 175 {
 176         struct nf_hook_ops *elem;
 177         unsigned int verdict;
 178         int ret = 0;
 179
 180         /* We may already have this, but read-locks nest anyway */
 181         rcu_read_lock();
 182
 183         elem = list_entry_rcu(&nf_hooks[pf][hook], struct nf_hook_ops, list);
 184 next_hook:
 185         verdict = nf_iterate(&nf_hooks[pf][hook], skb, hook, indev,
 186                              outdev, &elem, okfn, hook_thresh);
 187         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 188                 ret = 1;
 189         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 190                 kfree_skb(skb);
 191                 ret = NF_DROP_GETERR(verdict);
 192                 if (ret == 0)
 193                         ret = -EPERM;
 194         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 195                 int err = nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
 196                                                 verdict >> NF_VERDICT_QBITS);
 197                 if (err < 0) {
 198                         if (err == -ECANCELED)
 199                                 goto next_hook;
 200                         if (err == -ESRCH &&
 201                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 202                                 goto next_hook;
 203                         kfree_skb(skb);
 204                 }
 205         }
 206         rcu_read_unlock();
 207         return ret;
 208 }
 209 EXPORT_SYMBOL(nf_hook_slow);
 210
 211
 212 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 213 {
 214         if (writable_len > skb->len)
 215                 return 0;
 216
 217         /* Not exclusive use of packet?  Must copy. */
 218         if (!skb_cloned(skb)) {
 219                 if (writable_len <= skb_headlen(skb))
 220                         return 1;
 221         } else if (skb_clone_writable(skb, writable_len))
 222                 return 1;
 223
 224         if (writable_len <= skb_headlen(skb))
 225                 writable_len = 0;
 226         else
 227                 writable_len -= skb_headlen(skb);
 228
 229         return !!__pskb_pull_tail(skb, writable_len);
 230 }
 231 EXPORT_SYMBOL(skb_make_writable);
 232
 233 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 234 /* This does not belong here, but locally generated errors need it if connection
 235    tracking in use: without this, connection may not be in hash table, and hence
 236    manufactured ICMP or RST packets will not be associated with it. */
 237 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 238                 __rcu __read_mostly;
 239 EXPORT_SYMBOL(ip_ct_attach);
 240
 241 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 242 {
 243         void (*attach)(struct sk_buff *, const struct sk_buff *);
 244
 245         if (skb->nfct) {
 246                 rcu_read_lock();
 247                 attach = rcu_dereference(ip_ct_attach);
 248                 if (attach)
 249                         attach(new, skb);
 250                 rcu_read_unlock();
 251         }
 252 }
 253 EXPORT_SYMBOL(nf_ct_attach);
 254
 255 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 256 EXPORT_SYMBOL(nf_ct_destroy);
 257
 258 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 259 {
 260         void (*destroy)(struct nf_conntrack *);
 261
 262         rcu_read_lock();
 263         destroy = rcu_dereference(nf_ct_destroy);
 264         BUG_ON(destroy == NULL);
 265         destroy(nfct);
 266         rcu_read_unlock();
 267 }
 268 EXPORT_SYMBOL(nf_conntrack_destroy);
 269
 270 struct nfq_ct_hook __rcu *nfq_ct_hook __read_mostly;
 271 EXPORT_SYMBOL_GPL(nfq_ct_hook);
 272
 273 struct nfq_ct_nat_hook __rcu *nfq_ct_nat_hook __read_mostly;
 274 EXPORT_SYMBOL_GPL(nfq_ct_nat_hook);
 275
 276 #endif /* CONFIG_NF_CONNTRACK */
 277
 278 #ifdef CONFIG_NF_NAT_NEEDED
 279 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 280 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 281 #endif
 282
 283 static int __net_init netfilter_net_init(struct net *net)
 284 {
 285 #ifdef CONFIG_PROC_FS
 286         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 287                                                 net->proc_net);
 288         if (!net->nf.proc_netfilter) {
 289                 if (!net_eq(net, &init_net))
 290                         pr_err("cannot create netfilter proc entry");
 291
 292                 return -ENOMEM;
 293         }
 294 #endif
 295         return 0;
 296 }
 297
 298 static void __net_exit netfilter_net_exit(struct net *net)
 299 {
 300         remove_proc_entry("netfilter", net->proc_net);
 301 }
 302
 303 static struct pernet_operations netfilter_net_ops = {
 304         .init = netfilter_net_init,
 305         .exit = netfilter_net_exit,
 306 };
 307
 308 int __init netfilter_init(void)
 309 {
 310         int i, h, ret;
 311
 312         for (i = 0; i < ARRAY_SIZE(nf_hooks); i++) {
 313                 for (h = 0; h < NF_MAX_HOOKS; h++)
 314                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 315         }
 316
 317         ret = register_pernet_subsys(&netfilter_net_ops);
 318         if (ret < 0)
 319                 goto err;
 320
 321         ret = netfilter_log_init();
 322         if (ret < 0)
 323                 goto err_pernet;
 324
 325         return 0;
 326 err_pernet:
 327         unregister_pernet_subsys(&netfilter_net_ops);
 328 err:
 329         return ret;
 330 }