Commit | Line | Data |
---|---|---|
601e68e1 | 1 | /* netfilter.c: look after the filters for various protocols. |
f6ebe77f HW |
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. | |
f6ebe77f | 8 | */ |
f6ebe77f HW |
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> | |
d486dd1f | 21 | #include <linux/mutex.h> |
457c4cbc | 22 | #include <net/net_namespace.h> |
f6ebe77f HW |
23 | #include <net/sock.h> |
24 | ||
25 | #include "nf_internals.h" | |
26 | ||
d486dd1f | 27 | static DEFINE_MUTEX(afinfo_mutex); |
bce8032e | 28 | |
e2b7606c | 29 | struct nf_afinfo *nf_afinfo[NPROTO] __read_mostly; |
bce8032e PM |
30 | EXPORT_SYMBOL(nf_afinfo); |
31 | ||
32 | int nf_register_afinfo(struct nf_afinfo *afinfo) | |
33 | { | |
d486dd1f PM |
34 | int err; |
35 | ||
36 | err = mutex_lock_interruptible(&afinfo_mutex); | |
37 | if (err < 0) | |
38 | return err; | |
bce8032e | 39 | rcu_assign_pointer(nf_afinfo[afinfo->family], afinfo); |
d486dd1f | 40 | mutex_unlock(&afinfo_mutex); |
bce8032e PM |
41 | return 0; |
42 | } | |
43 | EXPORT_SYMBOL_GPL(nf_register_afinfo); | |
44 | ||
45 | void nf_unregister_afinfo(struct nf_afinfo *afinfo) | |
46 | { | |
d486dd1f | 47 | mutex_lock(&afinfo_mutex); |
bce8032e | 48 | rcu_assign_pointer(nf_afinfo[afinfo->family], NULL); |
d486dd1f | 49 | mutex_unlock(&afinfo_mutex); |
bce8032e PM |
50 | synchronize_rcu(); |
51 | } | |
52 | EXPORT_SYMBOL_GPL(nf_unregister_afinfo); | |
53 | ||
f6ebe77f HW |
54 | /* In this code, we can be waiting indefinitely for userspace to |
55 | * service a packet if a hook returns NF_QUEUE. We could keep a count | |
56 | * of skbuffs queued for userspace, and not deregister a hook unless | |
57 | * this is zero, but that sucks. Now, we simply check when the | |
58 | * packets come back: if the hook is gone, the packet is discarded. */ | |
e2b7606c | 59 | struct list_head nf_hooks[NPROTO][NF_MAX_HOOKS] __read_mostly; |
f6ebe77f | 60 | EXPORT_SYMBOL(nf_hooks); |
fd706d69 | 61 | static DEFINE_MUTEX(nf_hook_mutex); |
f6ebe77f HW |
62 | |
63 | int nf_register_hook(struct nf_hook_ops *reg) | |
64 | { | |
65 | struct list_head *i; | |
fd706d69 | 66 | int err; |
f6ebe77f | 67 | |
fd706d69 PM |
68 | err = mutex_lock_interruptible(&nf_hook_mutex); |
69 | if (err < 0) | |
70 | return err; | |
f6ebe77f HW |
71 | list_for_each(i, &nf_hooks[reg->pf][reg->hooknum]) { |
72 | if (reg->priority < ((struct nf_hook_ops *)i)->priority) | |
73 | break; | |
74 | } | |
75 | list_add_rcu(®->list, i->prev); | |
fd706d69 | 76 | mutex_unlock(&nf_hook_mutex); |
f6ebe77f HW |
77 | return 0; |
78 | } | |
79 | EXPORT_SYMBOL(nf_register_hook); | |
80 | ||
81 | void nf_unregister_hook(struct nf_hook_ops *reg) | |
82 | { | |
fd706d69 | 83 | mutex_lock(&nf_hook_mutex); |
f6ebe77f | 84 | list_del_rcu(®->list); |
fd706d69 | 85 | mutex_unlock(&nf_hook_mutex); |
f6ebe77f HW |
86 | |
87 | synchronize_net(); | |
88 | } | |
89 | EXPORT_SYMBOL(nf_unregister_hook); | |
90 | ||
972d1cb1 PM |
91 | int nf_register_hooks(struct nf_hook_ops *reg, unsigned int n) |
92 | { | |
93 | unsigned int i; | |
94 | int err = 0; | |
95 | ||
96 | for (i = 0; i < n; i++) { | |
97 | err = nf_register_hook(®[i]); | |
98 | if (err) | |
99 | goto err; | |
100 | } | |
101 | return err; | |
102 | ||
103 | err: | |
104 | if (i > 0) | |
105 | nf_unregister_hooks(reg, i); | |
106 | return err; | |
107 | } | |
108 | EXPORT_SYMBOL(nf_register_hooks); | |
109 | ||
110 | void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n) | |
111 | { | |
112 | unsigned int i; | |
113 | ||
114 | for (i = 0; i < n; i++) | |
115 | nf_unregister_hook(®[i]); | |
116 | } | |
117 | EXPORT_SYMBOL(nf_unregister_hooks); | |
118 | ||
f6ebe77f HW |
119 | unsigned int nf_iterate(struct list_head *head, |
120 | struct sk_buff **skb, | |
121 | int hook, | |
122 | const struct net_device *indev, | |
123 | const struct net_device *outdev, | |
124 | struct list_head **i, | |
125 | int (*okfn)(struct sk_buff *), | |
126 | int hook_thresh) | |
127 | { | |
128 | unsigned int verdict; | |
129 | ||
130 | /* | |
131 | * The caller must not block between calls to this | |
132 | * function because of risk of continuing from deleted element. | |
133 | */ | |
134 | list_for_each_continue_rcu(*i, head) { | |
135 | struct nf_hook_ops *elem = (struct nf_hook_ops *)*i; | |
136 | ||
137 | if (hook_thresh > elem->priority) | |
138 | continue; | |
139 | ||
140 | /* Optimization: we don't need to hold module | |
601e68e1 | 141 | reference here, since function can't sleep. --RR */ |
f6ebe77f HW |
142 | verdict = elem->hook(hook, skb, indev, outdev, okfn); |
143 | if (verdict != NF_ACCEPT) { | |
144 | #ifdef CONFIG_NETFILTER_DEBUG | |
145 | if (unlikely((verdict & NF_VERDICT_MASK) | |
146 | > NF_MAX_VERDICT)) { | |
147 | NFDEBUG("Evil return from %p(%u).\n", | |
601e68e1 | 148 | elem->hook, hook); |
f6ebe77f HW |
149 | continue; |
150 | } | |
151 | #endif | |
152 | if (verdict != NF_REPEAT) | |
153 | return verdict; | |
154 | *i = (*i)->prev; | |
155 | } | |
156 | } | |
157 | return NF_ACCEPT; | |
158 | } | |
159 | ||
160 | ||
161 | /* Returns 1 if okfn() needs to be executed by the caller, | |
162 | * -EPERM for NF_DROP, 0 otherwise. */ | |
163 | int nf_hook_slow(int pf, unsigned int hook, struct sk_buff **pskb, | |
164 | struct net_device *indev, | |
165 | struct net_device *outdev, | |
166 | int (*okfn)(struct sk_buff *), | |
167 | int hook_thresh) | |
168 | { | |
169 | struct list_head *elem; | |
170 | unsigned int verdict; | |
171 | int ret = 0; | |
172 | ||
173 | /* We may already have this, but read-locks nest anyway */ | |
174 | rcu_read_lock(); | |
175 | ||
176 | elem = &nf_hooks[pf][hook]; | |
177 | next_hook: | |
178 | verdict = nf_iterate(&nf_hooks[pf][hook], pskb, hook, indev, | |
179 | outdev, &elem, okfn, hook_thresh); | |
180 | if (verdict == NF_ACCEPT || verdict == NF_STOP) { | |
181 | ret = 1; | |
182 | goto unlock; | |
183 | } else if (verdict == NF_DROP) { | |
184 | kfree_skb(*pskb); | |
185 | ret = -EPERM; | |
186 | } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) { | |
187 | NFDEBUG("nf_hook: Verdict = QUEUE.\n"); | |
394f545d | 188 | if (!nf_queue(*pskb, elem, pf, hook, indev, outdev, okfn, |
f6ebe77f HW |
189 | verdict >> NF_VERDICT_BITS)) |
190 | goto next_hook; | |
191 | } | |
192 | unlock: | |
193 | rcu_read_unlock(); | |
194 | return ret; | |
195 | } | |
196 | EXPORT_SYMBOL(nf_hook_slow); | |
197 | ||
198 | ||
199 | int skb_make_writable(struct sk_buff **pskb, unsigned int writable_len) | |
200 | { | |
201 | struct sk_buff *nskb; | |
202 | ||
203 | if (writable_len > (*pskb)->len) | |
204 | return 0; | |
205 | ||
206 | /* Not exclusive use of packet? Must copy. */ | |
334a8132 PM |
207 | if (skb_cloned(*pskb) && !skb_clone_writable(*pskb, writable_len)) |
208 | goto copy_skb; | |
209 | if (skb_shared(*pskb)) | |
f6ebe77f HW |
210 | goto copy_skb; |
211 | ||
212 | return pskb_may_pull(*pskb, writable_len); | |
213 | ||
214 | copy_skb: | |
215 | nskb = skb_copy(*pskb, GFP_ATOMIC); | |
216 | if (!nskb) | |
217 | return 0; | |
218 | BUG_ON(skb_is_nonlinear(nskb)); | |
219 | ||
220 | /* Rest of kernel will get very unhappy if we pass it a | |
221 | suddenly-orphaned skbuff */ | |
222 | if ((*pskb)->sk) | |
223 | skb_set_owner_w(nskb, (*pskb)->sk); | |
224 | kfree_skb(*pskb); | |
225 | *pskb = nskb; | |
226 | return 1; | |
227 | } | |
228 | EXPORT_SYMBOL(skb_make_writable); | |
229 | ||
43bc0ca7 AV |
230 | void nf_proto_csum_replace4(__sum16 *sum, struct sk_buff *skb, |
231 | __be32 from, __be32 to, int pseudohdr) | |
4cf411de | 232 | { |
43bc0ca7 | 233 | __be32 diff[] = { ~from, to }; |
4cf411de | 234 | if (skb->ip_summed != CHECKSUM_PARTIAL) { |
a47362a2 | 235 | *sum = csum_fold(csum_partial(diff, sizeof(diff), |
43bc0ca7 | 236 | ~csum_unfold(*sum))); |
4cf411de | 237 | if (skb->ip_summed == CHECKSUM_COMPLETE && pseudohdr) |
a47362a2 | 238 | skb->csum = ~csum_partial(diff, sizeof(diff), |
43bc0ca7 | 239 | ~skb->csum); |
4cf411de | 240 | } else if (pseudohdr) |
a47362a2 | 241 | *sum = ~csum_fold(csum_partial(diff, sizeof(diff), |
43bc0ca7 | 242 | csum_unfold(*sum))); |
4cf411de | 243 | } |
43bc0ca7 | 244 | EXPORT_SYMBOL(nf_proto_csum_replace4); |
f6ebe77f | 245 | |
5f79e0f9 | 246 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) |
f6ebe77f HW |
247 | /* This does not belong here, but locally generated errors need it if connection |
248 | tracking in use: without this, connection may not be in hash table, and hence | |
249 | manufactured ICMP or RST packets will not be associated with it. */ | |
250 | void (*ip_ct_attach)(struct sk_buff *, struct sk_buff *); | |
251 | EXPORT_SYMBOL(ip_ct_attach); | |
252 | ||
253 | void nf_ct_attach(struct sk_buff *new, struct sk_buff *skb) | |
254 | { | |
255 | void (*attach)(struct sk_buff *, struct sk_buff *); | |
256 | ||
c3a47ab3 PM |
257 | if (skb->nfct) { |
258 | rcu_read_lock(); | |
259 | attach = rcu_dereference(ip_ct_attach); | |
260 | if (attach) | |
261 | attach(new, skb); | |
262 | rcu_read_unlock(); | |
f6ebe77f HW |
263 | } |
264 | } | |
265 | EXPORT_SYMBOL(nf_ct_attach); | |
de6e05c4 YK |
266 | |
267 | void (*nf_ct_destroy)(struct nf_conntrack *); | |
268 | EXPORT_SYMBOL(nf_ct_destroy); | |
269 | ||
270 | void nf_conntrack_destroy(struct nf_conntrack *nfct) | |
271 | { | |
272 | void (*destroy)(struct nf_conntrack *); | |
273 | ||
274 | rcu_read_lock(); | |
275 | destroy = rcu_dereference(nf_ct_destroy); | |
276 | BUG_ON(destroy == NULL); | |
277 | destroy(nfct); | |
278 | rcu_read_unlock(); | |
279 | } | |
280 | EXPORT_SYMBOL(nf_conntrack_destroy); | |
281 | #endif /* CONFIG_NF_CONNTRACK */ | |
f6ebe77f HW |
282 | |
283 | #ifdef CONFIG_PROC_FS | |
284 | struct proc_dir_entry *proc_net_netfilter; | |
285 | EXPORT_SYMBOL(proc_net_netfilter); | |
286 | #endif | |
287 | ||
288 | void __init netfilter_init(void) | |
289 | { | |
290 | int i, h; | |
291 | for (i = 0; i < NPROTO; i++) { | |
292 | for (h = 0; h < NF_MAX_HOOKS; h++) | |
293 | INIT_LIST_HEAD(&nf_hooks[i][h]); | |
294 | } | |
295 | ||
296 | #ifdef CONFIG_PROC_FS | |
457c4cbc | 297 | proc_net_netfilter = proc_mkdir("netfilter", init_net.proc_net); |
f6ebe77f HW |
298 | if (!proc_net_netfilter) |
299 | panic("cannot create netfilter proc entry"); | |
300 | #endif | |
301 | ||
302 | if (netfilter_queue_init() < 0) | |
303 | panic("cannot initialize nf_queue"); | |
304 | if (netfilter_log_init() < 0) | |
305 | panic("cannot initialize nf_log"); | |
306 | } |