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  */
   9 #include <linux/kernel.h>
  10 #include <linux/netfilter.h>
  11 #include <net/protocol.h>
  12 #include <linux/init.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/wait.h>
  15 #include <linux/module.h>
  16 #include <linux/interrupt.h>
  17 #include <linux/if.h>
  18 #include <linux/netdevice.h>
  19 #include <linux/inetdevice.h>
  20 #include <linux/proc_fs.h>
  21 #include <linux/mutex.h>
  22 #include <linux/slab.h>
  23 #include <net/net_namespace.h>
  24 #include <net/sock.h>
  25
  26 #include "nf_internals.h"
  27
  28 static DEFINE_MUTEX(afinfo_mutex);
  29
  30 const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO] __read_mostly;
  31 EXPORT_SYMBOL(nf_afinfo);
  32
  33 int nf_register_afinfo(const struct nf_afinfo *afinfo)
  34 {
  35         int err;
  36
  37         err = mutex_lock_interruptible(&afinfo_mutex);
  38         if (err < 0)
  39                 return err;
  40         rcu_assign_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_assign_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 static DEFINE_MUTEX(nf_hook_mutex);
  58
  59 int nf_register_hook(struct nf_hook_ops *reg)
  60 {
  61         struct nf_hook_ops *elem;
  62         int err;
  63
  64         err = mutex_lock_interruptible(&nf_hook_mutex);
  65         if (err < 0)
  66                 return err;
  67         list_for_each_entry(elem, &nf_hooks[reg->pf][reg->hooknum], list) {
  68                 if (reg->priority < elem->priority)
  69                         break;
  70         }
  71         list_add_rcu(&reg->list, elem->list.prev);
  72         mutex_unlock(&nf_hook_mutex);
  73         return 0;
  74 }
  75 EXPORT_SYMBOL(nf_register_hook);
  76
  77 void nf_unregister_hook(struct nf_hook_ops *reg)
  78 {
  79         mutex_lock(&nf_hook_mutex);
  80         list_del_rcu(&reg->list);
  81         mutex_unlock(&nf_hook_mutex);
  82
  83         synchronize_net();
  84 }
  85 EXPORT_SYMBOL(nf_unregister_hook);
  86
  87 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
  88 {
  89         unsigned int i;
  90         int err = 0;
  91
  92         for (i = 0; i < n; i++) {
  93                 err = nf_register_hook(&reg[i]);
  94                 if (err)
  95                         goto err;
  96         }
  97         return err;
  98
  99 err:
 100         if (i > 0)
 101                 nf_unregister_hooks(reg, i);
 102         return err;
 103 }
 104 EXPORT_SYMBOL(nf_register_hooks);
 105
 106 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 107 {
 108         while (n-- > 0)
 109                 nf_unregister_hook(&reg[n]);
 110 }
 111 EXPORT_SYMBOL(nf_unregister_hooks);
 112
 113 unsigned int nf_iterate(struct list_head *head,
 114                         struct sk_buff *skb,
 115                         unsigned int hook,
 116                         const struct net_device *indev,
 117                         const struct net_device *outdev,
 118                         struct list_head **i,
 119                         int (*okfn)(struct sk_buff *),
 120                         int hook_thresh)
 121 {
 122         unsigned int verdict;
 123
 124         /*
 125          * The caller must not block between calls to this
 126          * function because of risk of continuing from deleted element.
 127          */
 128         list_for_each_continue_rcu(*i, head) {
 129                 struct nf_hook_ops *elem = (struct nf_hook_ops *)*i;
 130
 131                 if (hook_thresh > elem->priority)
 132                         continue;
 133
 134                 /* Optimization: we don't need to hold module
 135                    reference here, since function can't sleep. --RR */
 136                 verdict = elem->hook(hook, skb, indev, outdev, okfn);
 137                 if (verdict != NF_ACCEPT) {
 138 #ifdef CONFIG_NETFILTER_DEBUG
 139                         if (unlikely((verdict & NF_VERDICT_MASK)
 140                                                         > NF_MAX_VERDICT)) {
 141                                 NFDEBUG("Evil return from %p(%u).\n",
 142                                         elem->hook, hook);
 143                                 continue;
 144                         }
 145 #endif
 146                         if (verdict != NF_REPEAT)
 147                                 return verdict;
 148                         *i = (*i)->prev;
 149                 }
 150         }
 151         return NF_ACCEPT;
 152 }
 153
 154
 155 /* Returns 1 if okfn() needs to be executed by the caller,
 156  * -EPERM for NF_DROP, 0 otherwise. */
 157 int nf_hook_slow(u_int8_t pf, unsigned int hook, struct sk_buff *skb,
 158                  struct net_device *indev,
 159                  struct net_device *outdev,
 160                  int (*okfn)(struct sk_buff *),
 161                  int hook_thresh)
 162 {
 163         struct list_head *elem;
 164         unsigned int verdict;
 165         int ret = 0;
 166
 167         /* We may already have this, but read-locks nest anyway */
 168         rcu_read_lock();
 169
 170         elem = &nf_hooks[pf][hook];
 171 next_hook:
 172         verdict = nf_iterate(&nf_hooks[pf][hook], skb, hook, indev,
 173                              outdev, &elem, okfn, hook_thresh);
 174         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 175                 ret = 1;
 176         } else if (verdict == NF_DROP) {
 177                 kfree_skb(skb);
 178                 ret = -EPERM;
 179         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 180                 if (!nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
 181                               verdict >> NF_VERDICT_BITS))
 182                         goto next_hook;
 183         }
 184         rcu_read_unlock();
 185         return ret;
 186 }
 187 EXPORT_SYMBOL(nf_hook_slow);
 188
 189
 190 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 191 {
 192         if (writable_len > skb->len)
 193                 return 0;
 194
 195         /* Not exclusive use of packet?  Must copy. */
 196         if (!skb_cloned(skb)) {
 197                 if (writable_len <= skb_headlen(skb))
 198                         return 1;
 199         } else if (skb_clone_writable(skb, writable_len))
 200                 return 1;
 201
 202         if (writable_len <= skb_headlen(skb))
 203                 writable_len = 0;
 204         else
 205                 writable_len -= skb_headlen(skb);
 206
 207         return !!__pskb_pull_tail(skb, writable_len);
 208 }
 209 EXPORT_SYMBOL(skb_make_writable);
 210
 211 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 212 /* This does not belong here, but locally generated errors need it if connection
 213    tracking in use: without this, connection may not be in hash table, and hence
 214    manufactured ICMP or RST packets will not be associated with it. */
 215 void (*ip_ct_attach)(struct sk_buff *, struct sk_buff *) __rcu __read_mostly;
 216 EXPORT_SYMBOL(ip_ct_attach);
 217
 218 void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb)
 219 {
 220         void (*attach)(struct sk_buff *, struct sk_buff *);
 221
 222         if (skb->nfct) {
 223                 rcu_read_lock();
 224                 attach = rcu_dereference(ip_ct_attach);
 225                 if (attach)
 226                         attach(new, skb);
 227                 rcu_read_unlock();
 228         }
 229 }
 230 EXPORT_SYMBOL(nf_ct_attach);
 231
 232 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 233 EXPORT_SYMBOL(nf_ct_destroy);
 234
 235 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 236 {
 237         void (*destroy)(struct nf_conntrack *);
 238
 239         rcu_read_lock();
 240         destroy = rcu_dereference(nf_ct_destroy);
 241         BUG_ON(destroy == NULL);
 242         destroy(nfct);
 243         rcu_read_unlock();
 244 }
 245 EXPORT_SYMBOL(nf_conntrack_destroy);
 246 #endif /* CONFIG_NF_CONNTRACK */
 247
 248 #ifdef CONFIG_PROC_FS
 249 struct proc_dir_entry *proc_net_netfilter;
 250 EXPORT_SYMBOL(proc_net_netfilter);
 251 #endif
 252
 253 void __init netfilter_init(void)
 254 {
 255         int i, h;
 256         for (i = 0; i < ARRAY_SIZE(nf_hooks); i++) {
 257                 for (h = 0; h < NF_MAX_HOOKS; h++)
 258                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 259         }
 260
 261 #ifdef CONFIG_PROC_FS
 262         proc_net_netfilter = proc_mkdir("netfilter", init_net.proc_net);
 263         if (!proc_net_netfilter)
 264                 panic("cannot create netfilter proc entry");
 265 #endif
 266
 267         if (netfilter_queue_init() < 0)
 268                 panic("cannot initialize nf_queue");
 269         if (netfilter_log_init() < 0)
 270                 panic("cannot initialize nf_log");
 271 }
 272
 273 #ifdef CONFIG_SYSCTL
 274 struct ctl_path nf_net_netfilter_sysctl_path[] = {
 275         { .procname = "net", },
 276         { .procname = "netfilter", },
 277         { }
 278 };
 279 EXPORT_SYMBOL_GPL(nf_net_netfilter_sysctl_path);
 280 #endif /* CONFIG_SYSCTL */