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
  34 int nf_register_afinfo(const struct nf_afinfo *afinfo)
  35 {
  36         int err;
  37
  38         err = mutex_lock_interruptible(&afinfo_mutex);
  39         if (err < 0)
  40                 return err;
  41         RCU_INIT_POINTER(nf_afinfo[afinfo->family], afinfo);
  42         mutex_unlock(&afinfo_mutex);
  43         return 0;
  44 }
  45 EXPORT_SYMBOL_GPL(nf_register_afinfo);
  46
  47 void nf_unregister_afinfo(const struct nf_afinfo *afinfo)
  48 {
  49         mutex_lock(&afinfo_mutex);
  50         RCU_INIT_POINTER(nf_afinfo[afinfo->family], NULL);
  51         mutex_unlock(&afinfo_mutex);
  52         synchronize_rcu();
  53 }
  54 EXPORT_SYMBOL_GPL(nf_unregister_afinfo);
  55
  56 struct list_head nf_hooks[NFPROTO_NUMPROTO][NF_MAX_HOOKS] __read_mostly;
  57 EXPORT_SYMBOL(nf_hooks);
  58
  59 #if defined(CONFIG_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
  66 int nf_register_hook(struct nf_hook_ops *reg)
  67 {
  68         struct nf_hook_ops *elem;
  69         int err;
  70
  71         err = mutex_lock_interruptible(&nf_hook_mutex);
  72         if (err < 0)
  73                 return err;
  74         list_for_each_entry(elem, &nf_hooks[reg->pf][reg->hooknum], list) {
  75                 if (reg->priority < elem->priority)
  76                         break;
  77         }
  78         list_add_rcu(&reg->list, elem->list.prev);
  79         mutex_unlock(&nf_hook_mutex);
  80 #if defined(CONFIG_JUMP_LABEL)
  81         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
  82 #endif
  83         return 0;
  84 }
  85 EXPORT_SYMBOL(nf_register_hook);
  86
  87 void nf_unregister_hook(struct nf_hook_ops *reg)
  88 {
  89         mutex_lock(&nf_hook_mutex);
  90         list_del_rcu(&reg->list);
  91         mutex_unlock(&nf_hook_mutex);
  92 #if defined(CONFIG_JUMP_LABEL)
  93         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
  94 #endif
  95         synchronize_net();
  96 }
  97 EXPORT_SYMBOL(nf_unregister_hook);
  98
  99 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 100 {
 101         unsigned int i;
 102         int err = 0;
 103
 104         for (i = 0; i < n; i++) {
 105                 err = nf_register_hook(&reg[i]);
 106                 if (err)
 107                         goto err;
 108         }
 109         return err;
 110
 111 err:
 112         if (i > 0)
 113                 nf_unregister_hooks(reg, i);
 114         return err;
 115 }
 116 EXPORT_SYMBOL(nf_register_hooks);
 117
 118 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 119 {
 120         while (n-- > 0)
 121                 nf_unregister_hook(&reg[n]);
 122 }
 123 EXPORT_SYMBOL(nf_unregister_hooks);
 124
 125 unsigned int nf_iterate(struct list_head *head,
 126                         struct sk_buff *skb,
 127                         unsigned int hook,
 128                         const struct net_device *indev,
 129                         const struct net_device *outdev,
 130                         struct nf_hook_ops **elemp,
 131                         int (*okfn)(struct sk_buff *),
 132                         int hook_thresh)
 133 {
 134         unsigned int verdict;
 135
 136         /*
 137          * The caller must not block between calls to this
 138          * function because of risk of continuing from deleted element.
 139          */
 140         list_for_each_entry_continue_rcu((*elemp), head, list) {
 141                 if (hook_thresh > (*elemp)->priority)
 142                         continue;
 143
 144                 /* Optimization: we don't need to hold module
 145                    reference here, since function can't sleep. --RR */
 146 repeat:
 147                 verdict = (*elemp)->hook(hook, skb, indev, outdev, okfn);
 148                 if (verdict != NF_ACCEPT) {
 149 #ifdef CONFIG_NETFILTER_DEBUG
 150                         if (unlikely((verdict & NF_VERDICT_MASK)
 151                                                         > NF_MAX_VERDICT)) {
 152                                 NFDEBUG("Evil return from %p(%u).\n",
 153                                         (*elemp)->hook, hook);
 154                                 continue;
 155                         }
 156 #endif
 157                         if (verdict != NF_REPEAT)
 158                                 return verdict;
 159                         goto repeat;
 160                 }
 161         }
 162         return NF_ACCEPT;
 163 }
 164
 165
 166 /* Returns 1 if okfn() needs to be executed by the caller,
 167  * -EPERM for NF_DROP, 0 otherwise. */
 168 int nf_hook_slow(u_int8_t pf, unsigned int hook, struct sk_buff *skb,
 169                  struct net_device *indev,
 170                  struct net_device *outdev,
 171                  int (*okfn)(struct sk_buff *),
 172                  int hook_thresh)
 173 {
 174         struct nf_hook_ops *elem;
 175         unsigned int verdict;
 176         int ret = 0;
 177
 178         /* We may already have this, but read-locks nest anyway */
 179         rcu_read_lock();
 180
 181         elem = list_entry_rcu(&nf_hooks[pf][hook], struct nf_hook_ops, list);
 182 next_hook:
 183         verdict = nf_iterate(&nf_hooks[pf][hook], skb, hook, indev,
 184                              outdev, &elem, okfn, hook_thresh);
 185         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 186                 ret = 1;
 187         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 188                 kfree_skb(skb);
 189                 ret = NF_DROP_GETERR(verdict);
 190                 if (ret == 0)
 191                         ret = -EPERM;
 192         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 193                 int err = nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
 194                                                 verdict >> NF_VERDICT_QBITS);
 195                 if (err < 0) {
 196                         if (err == -ECANCELED)
 197                                 goto next_hook;
 198                         if (err == -ESRCH &&
 199                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 200                                 goto next_hook;
 201                         kfree_skb(skb);
 202                 }
 203         }
 204         rcu_read_unlock();
 205         return ret;
 206 }
 207 EXPORT_SYMBOL(nf_hook_slow);
 208
 209
 210 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 211 {
 212         if (writable_len > skb->len)
 213                 return 0;
 214
 215         /* Not exclusive use of packet?  Must copy. */
 216         if (!skb_cloned(skb)) {
 217                 if (writable_len <= skb_headlen(skb))
 218                         return 1;
 219         } else if (skb_clone_writable(skb, writable_len))
 220                 return 1;
 221
 222         if (writable_len <= skb_headlen(skb))
 223                 writable_len = 0;
 224         else
 225                 writable_len -= skb_headlen(skb);
 226
 227         return !!__pskb_pull_tail(skb, writable_len);
 228 }
 229 EXPORT_SYMBOL(skb_make_writable);
 230
 231 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 232 /* This does not belong here, but locally generated errors need it if connection
 233    tracking in use: without this, connection may not be in hash table, and hence
 234    manufactured ICMP or RST packets will not be associated with it. */
 235 void (*ip_ct_attach)(struct sk_buff *, struct sk_buff *) __rcu __read_mostly;
 236 EXPORT_SYMBOL(ip_ct_attach);
 237
 238 void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb)
 239 {
 240         void (*attach)(struct sk_buff *, struct sk_buff *);
 241
 242         if (skb->nfct) {
 243                 rcu_read_lock();
 244                 attach = rcu_dereference(ip_ct_attach);
 245                 if (attach)
 246                         attach(new, skb);
 247                 rcu_read_unlock();
 248         }
 249 }
 250 EXPORT_SYMBOL(nf_ct_attach);
 251
 252 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 253 EXPORT_SYMBOL(nf_ct_destroy);
 254
 255 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 256 {
 257         void (*destroy)(struct nf_conntrack *);
 258
 259         rcu_read_lock();
 260         destroy = rcu_dereference(nf_ct_destroy);
 261         BUG_ON(destroy == NULL);
 262         destroy(nfct);
 263         rcu_read_unlock();
 264 }
 265 EXPORT_SYMBOL(nf_conntrack_destroy);
 266
 267 struct nfq_ct_hook __rcu *nfq_ct_hook __read_mostly;
 268 EXPORT_SYMBOL_GPL(nfq_ct_hook);
 269
 270 struct nfq_ct_nat_hook __rcu *nfq_ct_nat_hook __read_mostly;
 271 EXPORT_SYMBOL_GPL(nfq_ct_nat_hook);
 272
 273 #endif /* CONFIG_NF_CONNTRACK */
 274
 275 #ifdef CONFIG_NF_NAT_NEEDED
 276 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 277 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 278 #endif
 279
 280 static int __net_init netfilter_net_init(struct net *net)
 281 {
 282 #ifdef CONFIG_PROC_FS
 283         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 284                                                 net->proc_net);
 285         if (!net->nf.proc_netfilter) {
 286                 if (!net_eq(net, &init_net))
 287                         pr_err("cannot create netfilter proc entry");
 288
 289                 return -ENOMEM;
 290         }
 291 #endif
 292         return 0;
 293 }
 294
 295 static void __net_exit netfilter_net_exit(struct net *net)
 296 {
 297         remove_proc_entry("netfilter", net->proc_net);
 298 }
 299
 300 static struct pernet_operations netfilter_net_ops = {
 301         .init = netfilter_net_init,
 302         .exit = netfilter_net_exit,
 303 };
 304
 305 void __init netfilter_init(void)
 306 {
 307         int i, h;
 308         for (i = 0; i < ARRAY_SIZE(nf_hooks); i++) {
 309                 for (h = 0; h < NF_MAX_HOOKS; h++)
 310                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 311         }
 312
 313         if (register_pernet_subsys(&netfilter_net_ops) < 0)
 314                 panic("cannot create netfilter proc entry");
 315
 316         if (netfilter_log_init() < 0)
 317                 panic("cannot initialize nf_log");
 318 }