kernel/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 <linux/locallock.h>
  26 #include <net/net_namespace.h>
  27 #include <net/sock.h>
  28
  29 #include "nf_internals.h"
  30
  31 #ifdef CONFIG_PREEMPT_RT_BASE
  32 DEFINE_LOCAL_IRQ_LOCK(xt_write_lock);
  33 EXPORT_PER_CPU_SYMBOL(xt_write_lock);
  34 #endif
  35
  36 static DEFINE_MUTEX(afinfo_mutex);
  37
  38 const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO] __read_mostly;
  39 EXPORT_SYMBOL(nf_afinfo);
  40 const struct nf_ipv6_ops __rcu *nf_ipv6_ops __read_mostly;
  41 EXPORT_SYMBOL_GPL(nf_ipv6_ops);
  42
  43 int nf_register_afinfo(const struct nf_afinfo *afinfo)
  44 {
  45         mutex_lock(&afinfo_mutex);
  46         RCU_INIT_POINTER(nf_afinfo[afinfo->family], afinfo);
  47         mutex_unlock(&afinfo_mutex);
  48         return 0;
  49 }
  50 EXPORT_SYMBOL_GPL(nf_register_afinfo);
  51
  52 void nf_unregister_afinfo(const struct nf_afinfo *afinfo)
  53 {
  54         mutex_lock(&afinfo_mutex);
  55         RCU_INIT_POINTER(nf_afinfo[afinfo->family], NULL);
  56         mutex_unlock(&afinfo_mutex);
  57         synchronize_rcu();
  58 }
  59 EXPORT_SYMBOL_GPL(nf_unregister_afinfo);
  60
  61 struct list_head nf_hooks[NFPROTO_NUMPROTO][NF_MAX_HOOKS] __read_mostly;
  62 EXPORT_SYMBOL(nf_hooks);
  63
  64 #ifdef HAVE_JUMP_LABEL
  65 struct static_key nf_hooks_needed[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
  66 EXPORT_SYMBOL(nf_hooks_needed);
  67 #endif
  68
  69 static DEFINE_MUTEX(nf_hook_mutex);
  70
  71 int nf_register_hook(struct nf_hook_ops *reg)
  72 {
  73         struct nf_hook_ops *elem;
  74
  75         mutex_lock(&nf_hook_mutex);
  76         list_for_each_entry(elem, &nf_hooks[reg->pf][reg->hooknum], list) {
  77                 if (reg->priority < elem->priority)
  78                         break;
  79         }
  80         list_add_rcu(&reg->list, elem->list.prev);
  81         mutex_unlock(&nf_hook_mutex);
  82 #ifdef HAVE_JUMP_LABEL
  83         static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
  84 #endif
  85         return 0;
  86 }
  87 EXPORT_SYMBOL(nf_register_hook);
  88
  89 void nf_unregister_hook(struct nf_hook_ops *reg)
  90 {
  91         mutex_lock(&nf_hook_mutex);
  92         list_del_rcu(&reg->list);
  93         mutex_unlock(&nf_hook_mutex);
  94 #ifdef HAVE_JUMP_LABEL
  95         static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
  96 #endif
  97         synchronize_net();
  98 }
  99 EXPORT_SYMBOL(nf_unregister_hook);
 100
 101 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
 102 {
 103         unsigned int i;
 104         int err = 0;
 105
 106         for (i = 0; i < n; i++) {
 107                 err = nf_register_hook(&reg[i]);
 108                 if (err)
 109                         goto err;
 110         }
 111         return err;
 112
 113 err:
 114         if (i > 0)
 115                 nf_unregister_hooks(reg, i);
 116         return err;
 117 }
 118 EXPORT_SYMBOL(nf_register_hooks);
 119
 120 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 121 {
 122         while (n-- > 0)
 123                 nf_unregister_hook(&reg[n]);
 124 }
 125 EXPORT_SYMBOL(nf_unregister_hooks);
 126
 127 unsigned int nf_iterate(struct list_head *head,
 128                         struct sk_buff *skb,
 129                         struct nf_hook_state *state,
 130                         struct nf_hook_ops **elemp)
 131 {
 132         unsigned int verdict;
 133
 134         /*
 135          * The caller must not block between calls to this
 136          * function because of risk of continuing from deleted element.
 137          */
 138         list_for_each_entry_continue_rcu((*elemp), head, list) {
 139                 if (state->thresh > (*elemp)->priority)
 140                         continue;
 141
 142                 /* Optimization: we don't need to hold module
 143                    reference here, since function can't sleep. --RR */
 144 repeat:
 145                 verdict = (*elemp)->hook(*elemp, skb, state);
 146                 if (verdict != NF_ACCEPT) {
 147 #ifdef CONFIG_NETFILTER_DEBUG
 148                         if (unlikely((verdict & NF_VERDICT_MASK)
 149                                                         > NF_MAX_VERDICT)) {
 150                                 NFDEBUG("Evil return from %p(%u).\n",
 151                                         (*elemp)->hook, state->hook);
 152                                 continue;
 153                         }
 154 #endif
 155                         if (verdict != NF_REPEAT)
 156                                 return verdict;
 157                         goto repeat;
 158                 }
 159         }
 160         return NF_ACCEPT;
 161 }
 162
 163
 164 /* Returns 1 if okfn() needs to be executed by the caller,
 165  * -EPERM for NF_DROP, 0 otherwise. */
 166 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state)
 167 {
 168         struct nf_hook_ops *elem;
 169         unsigned int verdict;
 170         int ret = 0;
 171
 172         /* We may already have this, but read-locks nest anyway */
 173         rcu_read_lock();
 174
 175         elem = list_entry_rcu(&nf_hooks[state->pf][state->hook],
 176                               struct nf_hook_ops, list);
 177 next_hook:
 178         verdict = nf_iterate(&nf_hooks[state->pf][state->hook], skb, state,
 179                              &elem);
 180         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
 181                 ret = 1;
 182         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
 183                 kfree_skb(skb);
 184                 ret = NF_DROP_GETERR(verdict);
 185                 if (ret == 0)
 186                         ret = -EPERM;
 187         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
 188                 int err = nf_queue(skb, elem, state,
 189                                    verdict >> NF_VERDICT_QBITS);
 190                 if (err < 0) {
 191                         if (err == -ECANCELED)
 192                                 goto next_hook;
 193                         if (err == -ESRCH &&
 194                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 195                                 goto next_hook;
 196                         kfree_skb(skb);
 197                 }
 198         }
 199         rcu_read_unlock();
 200         return ret;
 201 }
 202 EXPORT_SYMBOL(nf_hook_slow);
 203
 204
 205 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
 206 {
 207         if (writable_len > skb->len)
 208                 return 0;
 209
 210         /* Not exclusive use of packet?  Must copy. */
 211         if (!skb_cloned(skb)) {
 212                 if (writable_len <= skb_headlen(skb))
 213                         return 1;
 214         } else if (skb_clone_writable(skb, writable_len))
 215                 return 1;
 216
 217         if (writable_len <= skb_headlen(skb))
 218                 writable_len = 0;
 219         else
 220                 writable_len -= skb_headlen(skb);
 221
 222         return !!__pskb_pull_tail(skb, writable_len);
 223 }
 224 EXPORT_SYMBOL(skb_make_writable);
 225
 226 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 227 /* This does not belong here, but locally generated errors need it if connection
 228    tracking in use: without this, connection may not be in hash table, and hence
 229    manufactured ICMP or RST packets will not be associated with it. */
 230 void (*ip_ct_attach)(struct sk_buff *, const struct sk_buff *)
 231                 __rcu __read_mostly;
 232 EXPORT_SYMBOL(ip_ct_attach);
 233
 234 void nf_ct_attach(struct sk_buff *new, const struct sk_buff *skb)
 235 {
 236         void (*attach)(struct sk_buff *, const struct sk_buff *);
 237
 238         if (skb->nfct) {
 239                 rcu_read_lock();
 240                 attach = rcu_dereference(ip_ct_attach);
 241                 if (attach)
 242                         attach(new, skb);
 243                 rcu_read_unlock();
 244         }
 245 }
 246 EXPORT_SYMBOL(nf_ct_attach);
 247
 248 void (*nf_ct_destroy)(struct nf_conntrack *) __rcu __read_mostly;
 249 EXPORT_SYMBOL(nf_ct_destroy);
 250
 251 void nf_conntrack_destroy(struct nf_conntrack *nfct)
 252 {
 253         void (*destroy)(struct nf_conntrack *);
 254
 255         rcu_read_lock();
 256         destroy = rcu_dereference(nf_ct_destroy);
 257         BUG_ON(destroy == NULL);
 258         destroy(nfct);
 259         rcu_read_unlock();
 260 }
 261 EXPORT_SYMBOL(nf_conntrack_destroy);
 262
 263 struct nfq_ct_hook __rcu *nfq_ct_hook __read_mostly;
 264 EXPORT_SYMBOL_GPL(nfq_ct_hook);
 265
 266 struct nfq_ct_nat_hook __rcu *nfq_ct_nat_hook __read_mostly;
 267 EXPORT_SYMBOL_GPL(nfq_ct_nat_hook);
 268
 269 #endif /* CONFIG_NF_CONNTRACK */
 270
 271 #ifdef CONFIG_NF_NAT_NEEDED
 272 void (*nf_nat_decode_session_hook)(struct sk_buff *, struct flowi *);
 273 EXPORT_SYMBOL(nf_nat_decode_session_hook);
 274 #endif
 275
 276 static int __net_init netfilter_net_init(struct net *net)
 277 {
 278 #ifdef CONFIG_PROC_FS
 279         net->nf.proc_netfilter = proc_net_mkdir(net, "netfilter",
 280                                                 net->proc_net);
 281         if (!net->nf.proc_netfilter) {
 282                 if (!net_eq(net, &init_net))
 283                         pr_err("cannot create netfilter proc entry");
 284
 285                 return -ENOMEM;
 286         }
 287 #endif
 288         return 0;
 289 }
 290
 291 static void __net_exit netfilter_net_exit(struct net *net)
 292 {
 293         remove_proc_entry("netfilter", net->proc_net);
 294 }
 295
 296 static struct pernet_operations netfilter_net_ops = {
 297         .init = netfilter_net_init,
 298         .exit = netfilter_net_exit,
 299 };
 300
 301 int __init netfilter_init(void)
 302 {
 303         int i, h, ret;
 304
 305         for (i = 0; i < ARRAY_SIZE(nf_hooks); i++) {
 306                 for (h = 0; h < NF_MAX_HOOKS; h++)
 307                         INIT_LIST_HEAD(&nf_hooks[i][h]);
 308         }
 309
 310         ret = register_pernet_subsys(&netfilter_net_ops);
 311         if (ret < 0)
 312                 goto err;
 313
 314         ret = netfilter_log_init();
 315         if (ret < 0)
 316                 goto err_pernet;
 317
 318         return 0;
 319 err_pernet:
 320         unregister_pernet_subsys(&netfilter_net_ops);
 321 err:
 322         return ret;
 323 }