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  * Patrick McHardy (c) 2006-2012
   9  */
  10 #include <linux/kernel.h>
  11 #include <linux/netfilter.h>
  12 #include <net/protocol.h>
  13 #include <linux/init.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/wait.h>
  16 #include <linux/module.h>
  17 #include <linux/interrupt.h>
  18 #include <linux/if.h>
  19 #include <linux/netdevice.h>
  20 #include <linux/netfilter_ipv6.h>
  21 #include <linux/inetdevice.h>
  22 #include <linux/proc_fs.h>
  23 #include <linux/mutex.h>
  24 #include <linux/slab.h>
  25 #include <net/net_namespace.h>
  26 #include <net/sock.h>
  27
  28 #include "nf_internals.h"
  29
  30 static DEFINE_MUTEX(afinfo_mutex);
  31
  32 const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO] __read_mostly;
  33 EXPORT_SYMBOL(nf_afinfo);
  34 const struct nf_ipv6_ops __rcu *nf_ipv6_ops __read_mostly;
  35 EXPORT_SYMBOL_GPL(nf_ipv6_ops);
  36
  37 int nf_register_afinfo(const struct nf_afinfo *afinfo)
  38 {
  39         mutex_lock(&afinfo_mutex);
  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 #ifdef HAVE_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
  69         mutex_lock(&nf_hook_mutex);
  70         list_for_each_entry(elem, &nf_hooks[reg->pf][reg->hooknum], list) {
  71                 if (reg->priority < elem->priority)
  72                         break;
  73         }
  74         list_add_rcu(&reg->list, elem->list.prev);
  75         mutex_unlock(&nf_hook_mutex);
  76 #ifdef HAVE_JUMP_LABEL
  77         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
  78 #endif
  79         return 0;
  80 }
  81 EXPORT_SYMBOL(nf_register_hook);
  82
  83 void nf_unregister_hook(struct nf_hook_ops *reg)
  84 {
  85         mutex_lock(&nf_hook_mutex);
  86         list_del_rcu(&reg->list);
  87         mutex_unlock(&nf_hook_mutex);
  88 #ifdef HAVE_JUMP_LABEL
  89         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
  90 #endif
  91         synchronize_net();
  92 }
  93 EXPORT_SYMBOL(nf_unregister_hook);
  94
  95 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
  96 {
  97         unsigned int i;
  98         int err = 0;
  99
 100         for (i = 0; i < n; i++) {
 101                 err = nf_register_hook(&reg[i]);
 102                 if (err)
 103                         goto err;
 104         }
 105         return err;
 106
 107 err:
 108         if (i > 0)
 109                 nf_unregister_hooks(reg, i);
 110         return err;
 111 }
 112 EXPORT_SYMBOL(nf_register_hooks);
 113
 114 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 115 {
 116         while (n-- > 0)
 117                 nf_unregister_hook(&reg[n]);
 118 }
 119 EXPORT_SYMBOL(nf_unregister_hooks);
 120
 121 unsigned int nf_iterate(struct list_head *head,
 122                         struct sk_buff *skb,
 123                         unsigned int hook,
 124                         const struct net_device *indev,
 125                         const struct net_device *outdev,
 126                         struct nf_hook_ops **elemp,
 127                         int (*okfn)(struct sk_buff *),
 128                         int hook_thresh)
 129 {
 130         unsigned int verdict;
 131
 132         /*
 133          * The caller must not block between calls to this
 134          * function because of risk of continuing from deleted element.
 135          */
 136         list_for_each_entry_continue_rcu((*elemp), head, list) {
 137                 if (hook_thresh > (*elemp)->priority)
 138                         continue;
 139
 140                 /* Optimization: we don't need to hold module
 141                    reference here, since function can't sleep. --RR */
 142 repeat:
 143                 verdict = (*elemp)->hook(*elemp, skb, indev, outdev, okfn);
 144                 if (verdict != NF_ACCEPT) {
 145 #ifdef CONFIG_NETFILTER_DEBUG
 146                         if (unlikely((verdict & NF_VERDICT_MASK)
 147                                                         > NF_MAX_VERDICT)) {
 148                                 NFDEBUG("Evil return from %p(%u).\n",
 149                                         (*elemp)->hook, hook);
 150                                 continue;
 151                         }
 152 #endif
 153                         if (verdict != NF_REPEAT)
 154                                 return verdict;
 155                         goto repeat;
 156                 }
 157         }
 158         return NF_ACCEPT;
 159 }
 160
 161
 162 /* Returns 1 if okfn() needs to be executed by the caller,
 163  * -EPERM for NF_DROP, 0 otherwise. */
 164 int nf_hook_slow(u_int8_t pf, unsigned int hook, struct sk_buff *skb,
 165                  struct net_device *indev,
 166                  struct net_device *outdev,
 167                  int (*okfn)(struct sk_buff *),
 168                  int hook_thresh)
 169 {
 170         struct nf_hook_ops *elem;
 171         unsigned int verdict;
 172         int ret = 0;
 173
 174         /* We may already have this, but read-locks nest anyway */
 175         rcu_read_lock();
 176
 177         elem = list_entry_rcu(&nf_hooks[pf][hook], struct nf_hook_ops, list);
 178 next_hook:
 179         verdict = nf_iterate(&nf_hooks[pf][hook], skb, hook, indev,
 180                              outdev, &elem, okfn, hook_thresh);
 181         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 182                 ret = 1;
 183         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 184                 kfree_skb(skb);
 185                 ret = NF_DROP_GETERR(verdict);
 186                 if (ret == 0)
 187                         ret = -EPERM;
 188         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 189                 int err = nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
 190                                                 verdict >> NF_VERDICT_QBITS);
 191                 if (err < 0) {
 192                         if (err == -ECANCELED)
 193                                 goto next_hook;
 194                         if (err == -ESRCH &&
 195                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 196                                 goto next_hook;
 197                         kfree_skb(skb);
 198                 }
 199         }
 200         rcu_read_unlock();
 201         return ret;
 202 }
 203 EXPORT_SYMBOL(nf_hook_slow);
 204
 205
 206 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 207 {
 208         if (writable_len > skb->len)
 209                 return 0;
 210
 211         /* Not exclusive use of packet?  Must copy. */
 212         if (!skb_cloned(skb)) {
 213                 if (writable_len <= skb_headlen(skb))
 214                         return 1;
 215         } else if (skb_clone_writable(skb, writable_len))
 216                 return 1;
 217
 218         if (writable_len <= skb_headlen(skb))
 219                 writable_len = 0;
 220         else
 221                 writable_len -= skb_headlen(skb);
 222
 223         return !!__pskb_pull_tail(skb, writable_len);
 224 }
 225 EXPORT_SYMBOL(skb_make_writable);
 226
 227 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 228 /* This does not belong here, but locally generated errors need it if connection
 229    tracking in use: without this, connection may not be in hash table, and hence
 230    manufactured ICMP or RST packets will not be associated with it. */
 231 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 232                 __rcu __read_mostly;
 233 EXPORT_SYMBOL(ip_ct_attach);
 234
 235 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 236 {
 237         void (*attach)(struct sk_buff *, const struct sk_buff *);
 238
 239         if (skb->nfct) {
 240                 rcu_read_lock();
 241                 attach = rcu_dereference(ip_ct_attach);
 242                 if (attach)
 243                         attach(new, skb);
 244                 rcu_read_unlock();
 245         }
 246 }
 247 EXPORT_SYMBOL(nf_ct_attach);
 248
 249 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 250 EXPORT_SYMBOL(nf_ct_destroy);
 251
 252 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 253 {
 254         void (*destroy)(struct nf_conntrack *);
 255
 256         rcu_read_lock();
 257         destroy = rcu_dereference(nf_ct_destroy);
 258         BUG_ON(destroy == NULL);
 259         destroy(nfct);
 260         rcu_read_unlock();
 261 }
 262 EXPORT_SYMBOL(nf_conntrack_destroy);
 263
 264 struct nfq_ct_hook __rcu *nfq_ct_hook __read_mostly;
 265 EXPORT_SYMBOL_GPL(nfq_ct_hook);
 266
 267 struct nfq_ct_nat_hook __rcu *nfq_ct_nat_hook __read_mostly;
 268 EXPORT_SYMBOL_GPL(nfq_ct_nat_hook);
 269
 270 #endif /* CONFIG_NF_CONNTRACK */
 271
 272 #ifdef CONFIG_NF_NAT_NEEDED
 273 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 274 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 275 #endif
 276
 277 static int __net_init netfilter_net_init(struct net *net)
 278 {
 279 #ifdef CONFIG_PROC_FS
 280         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 281                                                 net->proc_net);
 282         if (!net->nf.proc_netfilter) {
 283                 if (!net_eq(net, &init_net))
 284                         pr_err("cannot create netfilter proc entry");
 285
 286                 return -ENOMEM;
 287         }
 288 #endif
 289         return 0;
 290 }
 291
 292 static void __net_exit netfilter_net_exit(struct net *net)
 293 {
 294         remove_proc_entry("netfilter", net->proc_net);
 295 }
 296
 297 static struct pernet_operations netfilter_net_ops = {
 298         .init = netfilter_net_init,
 299         .exit = netfilter_net_exit,
 300 };
 301
 302 int __init netfilter_init(void)
 303 {
 304         int i, h, ret;
 305
 306         for (i = 0; i < ARRAY_SIZE(nf_hooks); i++) {
 307                 for (h = 0; h < NF_MAX_HOOKS; h++)
 308                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 309         }
 310
 311         ret = register_pernet_subsys(&netfilter_net_ops);
 312         if (ret < 0)
 313                 goto err;
 314
 315         ret = netfilter_log_init();
 316         if (ret < 0)
 317                 goto err_pernet;
 318
 319         return 0;
 320 err_pernet:
 321         unregister_pernet_subsys(&netfilter_net_ops);
 322 err:
 323         return ret;
 324 }