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