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  * February 2000: Modified by James Morris to have 1 queue per protocol.
  10  * 15-Mar-2000:   Added NF_REPEAT --RR.
  11  * 08-May-2003:   Internal logging interface added by Jozsef Kadlecsik.
  12  */
  13 #include <linux/kernel.h>
  14 #include <linux/netfilter.h>
  15 #include <net/protocol.h>
  16 #include <linux/init.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/wait.h>
  19 #include <linux/module.h>
  20 #include <linux/interrupt.h>
  21 #include <linux/if.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/inetdevice.h>
  24 #include <linux/proc_fs.h>
  25 #include <net/sock.h>
  26
  27 #include "nf_internals.h"
  28
  29 static DEFINE_SPINLOCK(afinfo_lock);
  30
  31 struct nf_afinfo *nf_afinfo[NPROTO] __read_mostly;
  32 EXPORT_SYMBOL(nf_afinfo);
  33
  34 int nf_register_afinfo(struct nf_afinfo *afinfo)
  35 {
  36         spin_lock(&afinfo_lock);
  37         rcu_assign_pointer(nf_afinfo[afinfo->family], afinfo);
  38         spin_unlock(&afinfo_lock);
  39         return 0;
  40 }
  41 EXPORT_SYMBOL_GPL(nf_register_afinfo);
  42
  43 void nf_unregister_afinfo(struct nf_afinfo *afinfo)
  44 {
  45         spin_lock(&afinfo_lock);
  46         rcu_assign_pointer(nf_afinfo[afinfo->family], NULL);
  47         spin_unlock(&afinfo_lock);
  48         synchronize_rcu();
  49 }
  50 EXPORT_SYMBOL_GPL(nf_unregister_afinfo);
  51
  52 /* In this code, we can be waiting indefinitely for userspace to
  53  * service a packet if a hook returns NF_QUEUE.  We could keep a count
  54  * of skbuffs queued for userspace, and not deregister a hook unless
  55  * this is zero, but that sucks.  Now, we simply check when the
  56  * packets come back: if the hook is gone, the packet is discarded. */
  57 struct list_head nf_hooks[NPROTO][NF_MAX_HOOKS] __read_mostly;
  58 EXPORT_SYMBOL(nf_hooks);
  59 static DEFINE_SPINLOCK(nf_hook_lock);
  60
  61 int nf_register_hook(struct nf_hook_ops *reg)
  62 {
  63         struct list_head *i;
  64
  65         spin_lock_bh(&nf_hook_lock);
  66         list_for_each(i, &nf_hooks[reg->pf][reg->hooknum]) {
  67                 if (reg->priority < ((struct nf_hook_ops *)i)->priority)
  68                         break;
  69         }
  70         list_add_rcu(&reg->list, i->prev);
  71         spin_unlock_bh(&nf_hook_lock);
  72
  73         synchronize_net();
  74         return 0;
  75 }
  76 EXPORT_SYMBOL(nf_register_hook);
  77
  78 void nf_unregister_hook(struct nf_hook_ops *reg)
  79 {
  80         spin_lock_bh(&nf_hook_lock);
  81         list_del_rcu(&reg->list);
  82         spin_unlock_bh(&nf_hook_lock);
  83
  84         synchronize_net();
  85 }
  86 EXPORT_SYMBOL(nf_unregister_hook);
  87
  88 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
  89 {
  90         unsigned int i;
  91         int err = 0;
  92
  93         for (i = 0; i < n; i++) {
  94                 err = nf_register_hook(&reg[i]);
  95                 if (err)
  96                         goto err;
  97         }
  98         return err;
  99
 100 err:
 101         if (i > 0)
 102                 nf_unregister_hooks(reg, i);
 103         return err;
 104 }
 105 EXPORT_SYMBOL(nf_register_hooks);
 106
 107 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 108 {
 109         unsigned int i;
 110
 111         for (i = 0; i < n; i++)
 112                 nf_unregister_hook(&reg[i]);
 113 }
 114 EXPORT_SYMBOL(nf_unregister_hooks);
 115
 116 unsigned int nf_iterate(struct list_head *head,
 117                         struct sk_buff **skb,
 118                         int hook,
 119                         const struct net_device *indev,
 120                         const struct net_device *outdev,
 121                         struct list_head **i,
 122                         int (*okfn)(struct sk_buff *),
 123                         int hook_thresh)
 124 {
 125         unsigned int verdict;
 126
 127         /*
 128          * The caller must not block between calls to this
 129          * function because of risk of continuing from deleted element.
 130          */
 131         list_for_each_continue_rcu(*i, head) {
 132                 struct nf_hook_ops *elem = (struct nf_hook_ops *)*i;
 133
 134                 if (hook_thresh > elem->priority)
 135                         continue;
 136
 137                 /* Optimization: we don't need to hold module
 138                    reference here, since function can't sleep. --RR */
 139                 verdict = elem->hook(hook, skb, indev, outdev, okfn);
 140                 if (verdict != NF_ACCEPT) {
 141 #ifdef CONFIG_NETFILTER_DEBUG
 142                         if (unlikely((verdict & NF_VERDICT_MASK)
 143                                                         > NF_MAX_VERDICT)) {
 144                                 NFDEBUG("Evil return from %p(%u).\n",
 145                                         elem->hook, hook);
 146                                 continue;
 147                         }
 148 #endif
 149                         if (verdict != NF_REPEAT)
 150                                 return verdict;
 151                         *i = (*i)->prev;
 152                 }
 153         }
 154         return NF_ACCEPT;
 155 }
 156
 157
 158 /* Returns 1 if okfn() needs to be executed by the caller,
 159  * -EPERM for NF_DROP, 0 otherwise. */
 160 int nf_hook_slow(int pf, unsigned int hook, struct sk_buff **pskb,
 161                  struct net_device *indev,
 162                  struct net_device *outdev,
 163                  int (*okfn)(struct sk_buff *),
 164                  int hook_thresh)
 165 {
 166         struct list_head *elem;
 167         unsigned int verdict;
 168         int ret = 0;
 169
 170         /* We may already have this, but read-locks nest anyway */
 171         rcu_read_lock();
 172
 173         elem = &nf_hooks[pf][hook];
 174 next_hook:
 175         verdict = nf_iterate(&nf_hooks[pf][hook], pskb, hook, indev,
 176                              outdev, &elem, okfn, hook_thresh);
 177         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 178                 ret = 1;
 179                 goto unlock;
 180         } else if (verdict == NF_DROP) {
 181                 kfree_skb(*pskb);
 182                 ret = -EPERM;
 183         } else if ((verdict & NF_VERDICT_MASK)  == NF_QUEUE) {
 184                 NFDEBUG("nf_hook: Verdict = QUEUE.\n");
 185                 if (!nf_queue(*pskb, elem, pf, hook, indev, outdev, okfn,
 186                               verdict >> NF_VERDICT_BITS))
 187                         goto next_hook;
 188         }
 189 unlock:
 190         rcu_read_unlock();
 191         return ret;
 192 }
 193 EXPORT_SYMBOL(nf_hook_slow);
 194
 195
 196 int skb_make_writable(struct sk_buff **pskb, unsigned int writable_len)
 197 {
 198         struct sk_buff *nskb;
 199
 200         if (writable_len > (*pskb)->len)
 201                 return 0;
 202
 203         /* Not exclusive use of packet?  Must copy. */
 204         if (skb_shared(*pskb) || skb_cloned(*pskb))
 205                 goto copy_skb;
 206
 207         return pskb_may_pull(*pskb, writable_len);
 208
 209 copy_skb:
 210         nskb = skb_copy(*pskb, GFP_ATOMIC);
 211         if (!nskb)
 212                 return 0;
 213         BUG_ON(skb_is_nonlinear(nskb));
 214
 215         /* Rest of kernel will get very unhappy if we pass it a
 216            suddenly-orphaned skbuff */
 217         if ((*pskb)->sk)
 218                 skb_set_owner_w(nskb, (*pskb)->sk);
 219         kfree_skb(*pskb);
 220         *pskb = nskb;
 221         return 1;
 222 }
 223 EXPORT_SYMBOL(skb_make_writable);
 224
 225 void nf_proto_csum_replace4(__sum16 *sum, struct sk_buff *skb,
 226                             __be32 from, __be32 to, int pseudohdr)
 227 {
 228         __be32 diff[] = { ~from, to };
 229         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 230                 *sum = csum_fold(csum_partial((char *)diff, sizeof(diff),
 231                                 ~csum_unfold(*sum)));
 232                 if (skb->ip_summed == CHECKSUM_COMPLETE && pseudohdr)
 233                         skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 234                                                 ~skb->csum);
 235         } else if (pseudohdr)
 236                 *sum = ~csum_fold(csum_partial((char *)diff, sizeof(diff),
 237                                 csum_unfold(*sum)));
 238 }
 239 EXPORT_SYMBOL(nf_proto_csum_replace4);
 240
 241 /* This does not belong here, but locally generated errors need it if connection
 242    tracking in use: without this, connection may not be in hash table, and hence
 243    manufactured ICMP or RST packets will not be associated with it. */
 244 void (*ip_ct_attach)(struct sk_buff *, struct sk_buff *);
 245 EXPORT_SYMBOL(ip_ct_attach);
 246
 247 void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb)
 248 {
 249         void (*attach)(struct sk_buff *, struct sk_buff *);
 250
 251         if (skb->nfct && (attach = ip_ct_attach) != NULL) {
 252                 mb(); /* Just to be sure: must be read before executing this */
 253                 attach(new, skb);
 254         }
 255 }
 256 EXPORT_SYMBOL(nf_ct_attach);
 257
 258 #ifdef CONFIG_PROC_FS
 259 struct proc_dir_entry *proc_net_netfilter;
 260 EXPORT_SYMBOL(proc_net_netfilter);
 261 #endif
 262
 263 void __init netfilter_init(void)
 264 {
 265         int i, h;
 266         for (i = 0; i < NPROTO; i++) {
 267                 for (h = 0; h < NF_MAX_HOOKS; h++)
 268                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 269         }
 270
 271 #ifdef CONFIG_PROC_FS
 272         proc_net_netfilter = proc_mkdir("netfilter", proc_net);
 273         if (!proc_net_netfilter)
 274                 panic("cannot create netfilter proc entry");
 275 #endif
 276
 277         if (netfilter_queue_init() < 0)
 278                 panic("cannot initialize nf_queue");
 279         if (netfilter_log_init() < 0)
 280                 panic("cannot initialize nf_log");
 281 }