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_INIT_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_INIT_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
  58 #if defined(CONFIG_JUMP_LABEL)
  59 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
  60 EXPORT_SYMBOL(nf_hooks_needed);
  61 #endif
  62
  63 static DEFINE_MUTEX(nf_hook_mutex);
  64
  65 int nf_register_hook(struct nf_hook_ops *reg)
  66 {
  67         struct nf_hook_ops *elem;
  68         int err;
  69
  70         err = mutex_lock_interruptible(&nf_hook_mutex);
  71         if (err < 0)
  72                 return err;
  73         list_for_each_entry(elem, &nf_hooks[reg->pf][reg->hooknum], list) {
  74                 if (reg->priority < elem->priority)
  75                         break;
  76         }
  77         list_add_rcu(&reg->list, elem->list.prev);
  78         mutex_unlock(&nf_hook_mutex);
  79 #if defined(CONFIG_JUMP_LABEL)
  80         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
  81 #endif
  82         return 0;
  83 }
  84 EXPORT_SYMBOL(nf_register_hook);
  85
  86 void nf_unregister_hook(struct nf_hook_ops *reg)
  87 {
  88         mutex_lock(&nf_hook_mutex);
  89         list_del_rcu(&reg->list);
  90         mutex_unlock(&nf_hook_mutex);
  91 #if defined(CONFIG_JUMP_LABEL)
  92         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
  93 #endif
  94         synchronize_net();
  95 }
  96 EXPORT_SYMBOL(nf_unregister_hook);
  97
  98 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
  99 {
 100         unsigned int i;
 101         int err = 0;
 102
 103         for (i = 0; i < n; i++) {
 104                 err = nf_register_hook(&reg[i]);
 105                 if (err)
 106                         goto err;
 107         }
 108         return err;
 109
 110 err:
 111         if (i > 0)
 112                 nf_unregister_hooks(reg, i);
 113         return err;
 114 }
 115 EXPORT_SYMBOL(nf_register_hooks);
 116
 117 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 118 {
 119         while (n-- > 0)
 120                 nf_unregister_hook(&reg[n]);
 121 }
 122 EXPORT_SYMBOL(nf_unregister_hooks);
 123
 124 unsigned int nf_iterate(struct list_head *head,
 125                         struct sk_buff *skb,
 126                         unsigned int hook,
 127                         const struct net_device *indev,
 128                         const struct net_device *outdev,
 129                         struct list_head **i,
 130                         int (*okfn)(struct sk_buff *),
 131                         int hook_thresh)
 132 {
 133         unsigned int verdict;
 134
 135         /*
 136          * The caller must not block between calls to this
 137          * function because of risk of continuing from deleted element.
 138          */
 139         list_for_each_continue_rcu(*i, head) {
 140                 struct nf_hook_ops *elem = (struct nf_hook_ops *)*i;
 141
 142                 if (hook_thresh > elem->priority)
 143                         continue;
 144
 145                 /* Optimization: we don't need to hold module
 146                    reference here, since function can't sleep. --RR */
 147 repeat:
 148                 verdict = elem->hook(hook, skb, indev, outdev, okfn);
 149                 if (verdict != NF_ACCEPT) {
 150 #ifdef CONFIG_NETFILTER_DEBUG
 151                         if (unlikely((verdict & NF_VERDICT_MASK)
 152                                                         > NF_MAX_VERDICT)) {
 153                                 NFDEBUG("Evil return from %p(%u).\n",
 154                                         elem->hook, hook);
 155                                 continue;
 156                         }
 157 #endif
 158                         if (verdict != NF_REPEAT)
 159                                 return verdict;
 160                         goto repeat;
 161                 }
 162         }
 163         return NF_ACCEPT;
 164 }
 165
 166
 167 /* Returns 1 if okfn() needs to be executed by the caller,
 168  * -EPERM for NF_DROP, 0 otherwise. */
 169 int nf_hook_slow(u_int8_t pf, unsigned int hook, struct sk_buff *skb,
 170                  struct net_device *indev,
 171                  struct net_device *outdev,
 172                  int (*okfn)(struct sk_buff *),
 173                  int hook_thresh)
 174 {
 175         struct list_head *elem;
 176         unsigned int verdict;
 177         int ret = 0;
 178
 179         /* We may already have this, but read-locks nest anyway */
 180         rcu_read_lock();
 181
 182         elem = &nf_hooks[pf][hook];
 183 next_hook:
 184         verdict = nf_iterate(&nf_hooks[pf][hook], skb, hook, indev,
 185                              outdev, &elem, okfn, hook_thresh);
 186         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 187                 ret = 1;
 188         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 189                 kfree_skb(skb);
 190                 ret = NF_DROP_GETERR(verdict);
 191                 if (ret == 0)
 192                         ret = -EPERM;
 193         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 194                 int err = nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
 195                                                 verdict >> NF_VERDICT_QBITS);
 196                 if (err < 0) {
 197                         if (err == -ECANCELED)
 198                                 goto next_hook;
 199                         if (err == -ESRCH &&
 200                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 201                                 goto next_hook;
 202                         kfree_skb(skb);
 203                 }
 204         }
 205         rcu_read_unlock();
 206         return ret;
 207 }
 208 EXPORT_SYMBOL(nf_hook_slow);
 209
 210
 211 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 212 {
 213         if (writable_len > skb->len)
 214                 return 0;
 215
 216         /* Not exclusive use of packet?  Must copy. */
 217         if (!skb_cloned(skb)) {
 218                 if (writable_len <= skb_headlen(skb))
 219                         return 1;
 220         } else if (skb_clone_writable(skb, writable_len))
 221                 return 1;
 222
 223         if (writable_len <= skb_headlen(skb))
 224                 writable_len = 0;
 225         else
 226                 writable_len -= skb_headlen(skb);
 227
 228         return !!__pskb_pull_tail(skb, writable_len);
 229 }
 230 EXPORT_SYMBOL(skb_make_writable);
 231
 232 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 233 /* This does not belong here, but locally generated errors need it if connection
 234    tracking in use: without this, connection may not be in hash table, and hence
 235    manufactured ICMP or RST packets will not be associated with it. */
 236 void (*ip_ct_attach)(struct sk_buff *, struct sk_buff *) __rcu __read_mostly;
 237 EXPORT_SYMBOL(ip_ct_attach);
 238
 239 void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb)
 240 {
 241         void (*attach)(struct sk_buff *, struct sk_buff *);
 242
 243         if (skb->nfct) {
 244                 rcu_read_lock();
 245                 attach = rcu_dereference(ip_ct_attach);
 246                 if (attach)
 247                         attach(new, skb);
 248                 rcu_read_unlock();
 249         }
 250 }
 251 EXPORT_SYMBOL(nf_ct_attach);
 252
 253 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 254 EXPORT_SYMBOL(nf_ct_destroy);
 255
 256 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 257 {
 258         void (*destroy)(struct nf_conntrack *);
 259
 260         rcu_read_lock();
 261         destroy = rcu_dereference(nf_ct_destroy);
 262         BUG_ON(destroy == NULL);
 263         destroy(nfct);
 264         rcu_read_unlock();
 265 }
 266 EXPORT_SYMBOL(nf_conntrack_destroy);
 267 #endif /* CONFIG_NF_CONNTRACK */
 268
 269 #ifdef CONFIG_PROC_FS
 270 struct proc_dir_entry *proc_net_netfilter;
 271 EXPORT_SYMBOL(proc_net_netfilter);
 272 #endif
 273
 274 void __init netfilter_init(void)
 275 {
 276         int i, h;
 277         for (i = 0; i < ARRAY_SIZE(nf_hooks); i++) {
 278                 for (h = 0; h < NF_MAX_HOOKS; h++)
 279                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 280         }
 281
 282 #ifdef CONFIG_PROC_FS
 283         proc_net_netfilter = proc_mkdir("netfilter", init_net.proc_net);
 284         if (!proc_net_netfilter)
 285                 panic("cannot create netfilter proc entry");
 286 #endif
 287
 288         if (netfilter_queue_init() < 0)
 289                 panic("cannot initialize nf_queue");
 290         if (netfilter_log_init() < 0)
 291                 panic("cannot initialize nf_log");
 292 }
 293
 294 #ifdef CONFIG_SYSCTL
 295 struct ctl_path nf_net_netfilter_sysctl_path[] = {
 296         { .procname = "net", },
 297         { .procname = "netfilter", },
 298         { }
 299 };
 300 EXPORT_SYMBOL_GPL(nf_net_netfilter_sysctl_path);
 301 #endif /* CONFIG_SYSCTL */