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         return 0;
  73 }
  74 EXPORT_SYMBOL(nf_register_hook);
  75
  76 void nf_unregister_hook(struct nf_hook_ops *reg)
  77 {
  78         spin_lock_bh(&nf_hook_lock);
  79         list_del_rcu(&reg->list);
  80         spin_unlock_bh(&nf_hook_lock);
  81
  82         synchronize_net();
  83 }
  84 EXPORT_SYMBOL(nf_unregister_hook);
  85
  86 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
  87 {
  88         unsigned int i;
  89         int err = 0;
  90
  91         for (i = 0; i < n; i++) {
  92                 err = nf_register_hook(&reg[i]);
  93                 if (err)
  94                         goto err;
  95         }
  96         return err;
  97
  98 err:
  99         if (i > 0)
 100                 nf_unregister_hooks(reg, i);
 101         return err;
 102 }
 103 EXPORT_SYMBOL(nf_register_hooks);
 104
 105 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 106 {
 107         unsigned int i;
 108
 109         for (i = 0; i < n; i++)
 110                 nf_unregister_hook(&reg[i]);
 111 }
 112 EXPORT_SYMBOL(nf_unregister_hooks);
 113
 114 unsigned int nf_iterate(struct list_head *head,
 115                         struct sk_buff **skb,
 116                         int hook,
 117                         const struct net_device *indev,
 118                         const struct net_device *outdev,
 119                         struct list_head **i,
 120                         int (*okfn)(struct sk_buff *),
 121                         int hook_thresh)
 122 {
 123         unsigned int verdict;
 124
 125         /*
 126          * The caller must not block between calls to this
 127          * function because of risk of continuing from deleted element.
 128          */
 129         list_for_each_continue_rcu(*i, head) {
 130                 struct nf_hook_ops *elem = (struct nf_hook_ops *)*i;
 131
 132                 if (hook_thresh > elem->priority)
 133                         continue;
 134
 135                 /* Optimization: we don't need to hold module
 136                    reference here, since function can't sleep. --RR */
 137                 verdict = elem->hook(hook, skb, indev, outdev, okfn);
 138                 if (verdict != NF_ACCEPT) {
 139 #ifdef CONFIG_NETFILTER_DEBUG
 140                         if (unlikely((verdict & NF_VERDICT_MASK)
 141                                                         > NF_MAX_VERDICT)) {
 142                                 NFDEBUG("Evil return from %p(%u).\n",
 143                                         elem->hook, hook);
 144                                 continue;
 145                         }
 146 #endif
 147                         if (verdict != NF_REPEAT)
 148                                 return verdict;
 149                         *i = (*i)->prev;
 150                 }
 151         }
 152         return NF_ACCEPT;
 153 }
 154
 155
 156 /* Returns 1 if okfn() needs to be executed by the caller,
 157  * -EPERM for NF_DROP, 0 otherwise. */
 158 int nf_hook_slow(int pf, unsigned int hook, struct sk_buff **pskb,
 159                  struct net_device *indev,
 160                  struct net_device *outdev,
 161                  int (*okfn)(struct sk_buff *),
 162                  int hook_thresh)
 163 {
 164         struct list_head *elem;
 165         unsigned int verdict;
 166         int ret = 0;
 167
 168         /* We may already have this, but read-locks nest anyway */
 169         rcu_read_lock();
 170
 171         elem = &nf_hooks[pf][hook];
 172 next_hook:
 173         verdict = nf_iterate(&nf_hooks[pf][hook], pskb, hook, indev,
 174                              outdev, &elem, okfn, hook_thresh);
 175         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 176                 ret = 1;
 177                 goto unlock;
 178         } else if (verdict == NF_DROP) {
 179                 kfree_skb(*pskb);
 180                 ret = -EPERM;
 181         } else if ((verdict & NF_VERDICT_MASK)  == NF_QUEUE) {
 182                 NFDEBUG("nf_hook: Verdict = QUEUE.\n");
 183                 if (!nf_queue(*pskb, elem, pf, hook, indev, outdev, okfn,
 184                               verdict >> NF_VERDICT_BITS))
 185                         goto next_hook;
 186         }
 187 unlock:
 188         rcu_read_unlock();
 189         return ret;
 190 }
 191 EXPORT_SYMBOL(nf_hook_slow);
 192
 193
 194 int skb_make_writable(struct sk_buff **pskb, unsigned int writable_len)
 195 {
 196         struct sk_buff *nskb;
 197
 198         if (writable_len > (*pskb)->len)
 199                 return 0;
 200
 201         /* Not exclusive use of packet?  Must copy. */
 202         if (skb_shared(*pskb) || skb_cloned(*pskb))
 203                 goto copy_skb;
 204
 205         return pskb_may_pull(*pskb, writable_len);
 206
 207 copy_skb:
 208         nskb = skb_copy(*pskb, GFP_ATOMIC);
 209         if (!nskb)
 210                 return 0;
 211         BUG_ON(skb_is_nonlinear(nskb));
 212
 213         /* Rest of kernel will get very unhappy if we pass it a
 214            suddenly-orphaned skbuff */
 215         if ((*pskb)->sk)
 216                 skb_set_owner_w(nskb, (*pskb)->sk);
 217         kfree_skb(*pskb);
 218         *pskb = nskb;
 219         return 1;
 220 }
 221 EXPORT_SYMBOL(skb_make_writable);
 222
 223 void nf_proto_csum_replace4(__sum16 *sum, struct sk_buff *skb,
 224                             __be32 from, __be32 to, int pseudohdr)
 225 {
 226         __be32 diff[] = { ~from, to };
 227         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 228                 *sum = csum_fold(csum_partial((char *)diff, sizeof(diff),
 229                                 ~csum_unfold(*sum)));
 230                 if (skb->ip_summed == CHECKSUM_COMPLETE && pseudohdr)
 231                         skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 232                                                 ~skb->csum);
 233         } else if (pseudohdr)
 234                 *sum = ~csum_fold(csum_partial((char *)diff, sizeof(diff),
 235                                 csum_unfold(*sum)));
 236 }
 237 EXPORT_SYMBOL(nf_proto_csum_replace4);
 238
 239 /* This does not belong here, but locally generated errors need it if connection
 240    tracking in use: without this, connection may not be in hash table, and hence
 241    manufactured ICMP or RST packets will not be associated with it. */
 242 void (*ip_ct_attach)(struct sk_buff *, struct sk_buff *);
 243 EXPORT_SYMBOL(ip_ct_attach);
 244
 245 void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb)
 246 {
 247         void (*attach)(struct sk_buff *, struct sk_buff *);
 248
 249         if (skb->nfct) {
 250                 rcu_read_lock();
 251                 attach = rcu_dereference(ip_ct_attach);
 252                 if (attach)
 253                         attach(new, skb);
 254                 rcu_read_unlock();
 255         }
 256 }
 257 EXPORT_SYMBOL(nf_ct_attach);
 258
 259 #ifdef CONFIG_PROC_FS
 260 struct proc_dir_entry *proc_net_netfilter;
 261 EXPORT_SYMBOL(proc_net_netfilter);
 262 #endif
 263
 264 void __init netfilter_init(void)
 265 {
 266         int i, h;
 267         for (i = 0; i < NPROTO; i++) {
 268                 for (h = 0; h < NF_MAX_HOOKS; h++)
 269                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 270         }
 271
 272 #ifdef CONFIG_PROC_FS
 273         proc_net_netfilter = proc_mkdir("netfilter", proc_net);
 274         if (!proc_net_netfilter)
 275                 panic("cannot create netfilter proc entry");
 276 #endif
 277
 278         if (netfilter_queue_init() < 0)
 279                 panic("cannot initialize nf_queue");
 280         if (netfilter_log_init() < 0)
 281                 panic("cannot initialize nf_log");
 282 }