Merge branches 'stable/ia64', 'stable/blkfront-cleanup' and 'stable/cleanup' of git...
[pandora-kernel.git] / 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_assign_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_assign_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 static DEFINE_MUTEX(nf_hook_mutex);
58
59 int nf_register_hook(struct nf_hook_ops *reg)
60 {
61         struct nf_hook_ops *elem;
62         int err;
63
64         err = mutex_lock_interruptible(&nf_hook_mutex);
65         if (err < 0)
66                 return err;
67         list_for_each_entry(elem, &nf_hooks[reg->pf][reg->hooknum], list) {
68                 if (reg->priority < elem->priority)
69                         break;
70         }
71         list_add_rcu(&reg->list, elem->list.prev);
72         mutex_unlock(&nf_hook_mutex);
73         return 0;
74 }
75 EXPORT_SYMBOL(nf_register_hook);
76
77 void nf_unregister_hook(struct nf_hook_ops *reg)
78 {
79         mutex_lock(&nf_hook_mutex);
80         list_del_rcu(&reg->list);
81         mutex_unlock(&nf_hook_mutex);
82
83         synchronize_net();
84 }
85 EXPORT_SYMBOL(nf_unregister_hook);
86
87 int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n)
88 {
89         unsigned int i;
90         int err = 0;
91
92         for (i = 0; i < n; i++) {
93                 err = nf_register_hook(&reg[i]);
94                 if (err)
95                         goto err;
96         }
97         return err;
98
99 err:
100         if (i > 0)
101                 nf_unregister_hooks(reg, i);
102         return err;
103 }
104 EXPORT_SYMBOL(nf_register_hooks);
105
106 void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
107 {
108         while (n-- > 0)
109                 nf_unregister_hook(&reg[n]);
110 }
111 EXPORT_SYMBOL(nf_unregister_hooks);
112
113 unsigned int nf_iterate(struct list_head *head,
114                         struct sk_buff *skb,
115                         unsigned int hook,
116                         const struct net_device *indev,
117                         const struct net_device *outdev,
118                         struct list_head **i,
119                         int (*okfn)(struct sk_buff *),
120                         int hook_thresh)
121 {
122         unsigned int verdict;
123
124         /*
125          * The caller must not block between calls to this
126          * function because of risk of continuing from deleted element.
127          */
128         list_for_each_continue_rcu(*i, head) {
129                 struct nf_hook_ops *elem = (struct nf_hook_ops *)*i;
130
131                 if (hook_thresh > elem->priority)
132                         continue;
133
134                 /* Optimization: we don't need to hold module
135                    reference here, since function can't sleep. --RR */
136 repeat:
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                         goto repeat;
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(u_int8_t pf, unsigned int hook, struct sk_buff *skb,
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], skb, hook, indev,
174                              outdev, &elem, okfn, hook_thresh);
175         if (verdict == NF_ACCEPT || verdict == NF_STOP) {
176                 ret = 1;
177         } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
178                 kfree_skb(skb);
179                 ret = -(verdict >> NF_VERDICT_BITS);
180                 if (ret == 0)
181                         ret = -EPERM;
182         } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
183                 if (!nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
184                               verdict >> NF_VERDICT_BITS))
185                         goto next_hook;
186         }
187         rcu_read_unlock();
188         return ret;
189 }
190 EXPORT_SYMBOL(nf_hook_slow);
191
192
193 int skb_make_writable(struct sk_buff *skb, unsigned int writable_len)
194 {
195         if (writable_len > skb->len)
196                 return 0;
197
198         /* Not exclusive use of packet?  Must copy. */
199         if (!skb_cloned(skb)) {
200                 if (writable_len <= skb_headlen(skb))
201                         return 1;
202         } else if (skb_clone_writable(skb, writable_len))
203                 return 1;
204
205         if (writable_len <= skb_headlen(skb))
206                 writable_len = 0;
207         else
208                 writable_len -= skb_headlen(skb);
209
210         return !!__pskb_pull_tail(skb, writable_len);
211 }
212 EXPORT_SYMBOL(skb_make_writable);
213
214 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
215 /* This does not belong here, but locally generated errors need it if connection
216    tracking in use: without this, connection may not be in hash table, and hence
217    manufactured ICMP or RST packets will not be associated with it. */
218 void (*ip_ct_attach)(struct sk_buff *, struct sk_buff *);
219 EXPORT_SYMBOL(ip_ct_attach);
220
221 void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb)
222 {
223         void (*attach)(struct sk_buff *, struct sk_buff *);
224
225         if (skb->nfct) {
226                 rcu_read_lock();
227                 attach = rcu_dereference(ip_ct_attach);
228                 if (attach)
229                         attach(new, skb);
230                 rcu_read_unlock();
231         }
232 }
233 EXPORT_SYMBOL(nf_ct_attach);
234
235 void (*nf_ct_destroy)(struct nf_conntrack *);
236 EXPORT_SYMBOL(nf_ct_destroy);
237
238 void nf_conntrack_destroy(struct nf_conntrack *nfct)
239 {
240         void (*destroy)(struct nf_conntrack *);
241
242         rcu_read_lock();
243         destroy = rcu_dereference(nf_ct_destroy);
244         BUG_ON(destroy == NULL);
245         destroy(nfct);
246         rcu_read_unlock();
247 }
248 EXPORT_SYMBOL(nf_conntrack_destroy);
249 #endif /* CONFIG_NF_CONNTRACK */
250
251 #ifdef CONFIG_PROC_FS
252 struct proc_dir_entry *proc_net_netfilter;
253 EXPORT_SYMBOL(proc_net_netfilter);
254 #endif
255
256 void __init netfilter_init(void)
257 {
258         int i, h;
259         for (i = 0; i < ARRAY_SIZE(nf_hooks); i++) {
260                 for (h = 0; h < NF_MAX_HOOKS; h++)
261                         INIT_LIST_HEAD(&nf_hooks[i][h]);
262         }
263
264 #ifdef CONFIG_PROC_FS
265         proc_net_netfilter = proc_mkdir("netfilter", init_net.proc_net);
266         if (!proc_net_netfilter)
267                 panic("cannot create netfilter proc entry");
268 #endif
269
270         if (netfilter_queue_init() < 0)
271                 panic("cannot initialize nf_queue");
272         if (netfilter_log_init() < 0)
273                 panic("cannot initialize nf_log");
274 }
275
276 #ifdef CONFIG_SYSCTL
277 struct ctl_path nf_net_netfilter_sysctl_path[] = {
278         { .procname = "net", },
279         { .procname = "netfilter", },
280         { }
281 };
282 EXPORT_SYMBOL_GPL(nf_net_netfilter_sysctl_path);
283 #endif /* CONFIG_SYSCTL */