netfilter: allow NFQUEUE bypass if no listener is available
[linux-2.6.git] / net / netfilter / nf_queue.c
1 #include <linux/kernel.h>
2 #include <linux/slab.h>
3 #include <linux/init.h>
4 #include <linux/module.h>
5 #include <linux/proc_fs.h>
6 #include <linux/skbuff.h>
7 #include <linux/netfilter.h>
8 #include <linux/seq_file.h>
9 #include <linux/rcupdate.h>
10 #include <net/protocol.h>
11 #include <net/netfilter/nf_queue.h>
12 #include <net/dst.h>
13
14 #include "nf_internals.h"
15
16 /*
17  * A queue handler may be registered for each protocol.  Each is protected by
18  * long term mutex.  The handler must provide an an outfn() to accept packets
19  * for queueing and must reinject all packets it receives, no matter what.
20  */
21 static const struct nf_queue_handler __rcu *queue_handler[NFPROTO_NUMPROTO] __read_mostly;
22
23 static DEFINE_MUTEX(queue_handler_mutex);
24
25 /* return EBUSY when somebody else is registered, return EEXIST if the
26  * same handler is registered, return 0 in case of success. */
27 int nf_register_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh)
28 {
29         int ret;
30         const struct nf_queue_handler *old;
31
32         if (pf >= ARRAY_SIZE(queue_handler))
33                 return -EINVAL;
34
35         mutex_lock(&queue_handler_mutex);
36         old = rcu_dereference_protected(queue_handler[pf],
37                                         lockdep_is_held(&queue_handler_mutex));
38         if (old == qh)
39                 ret = -EEXIST;
40         else if (old)
41                 ret = -EBUSY;
42         else {
43                 rcu_assign_pointer(queue_handler[pf], qh);
44                 ret = 0;
45         }
46         mutex_unlock(&queue_handler_mutex);
47
48         return ret;
49 }
50 EXPORT_SYMBOL(nf_register_queue_handler);
51
52 /* The caller must flush their queue before this */
53 int nf_unregister_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh)
54 {
55         const struct nf_queue_handler *old;
56
57         if (pf >= ARRAY_SIZE(queue_handler))
58                 return -EINVAL;
59
60         mutex_lock(&queue_handler_mutex);
61         old = rcu_dereference_protected(queue_handler[pf],
62                                         lockdep_is_held(&queue_handler_mutex));
63         if (old && old != qh) {
64                 mutex_unlock(&queue_handler_mutex);
65                 return -EINVAL;
66         }
67
68         rcu_assign_pointer(queue_handler[pf], NULL);
69         mutex_unlock(&queue_handler_mutex);
70
71         synchronize_rcu();
72
73         return 0;
74 }
75 EXPORT_SYMBOL(nf_unregister_queue_handler);
76
77 void nf_unregister_queue_handlers(const struct nf_queue_handler *qh)
78 {
79         u_int8_t pf;
80
81         mutex_lock(&queue_handler_mutex);
82         for (pf = 0; pf < ARRAY_SIZE(queue_handler); pf++)  {
83                 if (rcu_dereference_protected(
84                                 queue_handler[pf],
85                                 lockdep_is_held(&queue_handler_mutex)
86                                 ) == qh)
87                         rcu_assign_pointer(queue_handler[pf], NULL);
88         }
89         mutex_unlock(&queue_handler_mutex);
90
91         synchronize_rcu();
92 }
93 EXPORT_SYMBOL_GPL(nf_unregister_queue_handlers);
94
95 static void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
96 {
97         /* Release those devices we held, or Alexey will kill me. */
98         if (entry->indev)
99                 dev_put(entry->indev);
100         if (entry->outdev)
101                 dev_put(entry->outdev);
102 #ifdef CONFIG_BRIDGE_NETFILTER
103         if (entry->skb->nf_bridge) {
104                 struct nf_bridge_info *nf_bridge = entry->skb->nf_bridge;
105
106                 if (nf_bridge->physindev)
107                         dev_put(nf_bridge->physindev);
108                 if (nf_bridge->physoutdev)
109                         dev_put(nf_bridge->physoutdev);
110         }
111 #endif
112         /* Drop reference to owner of hook which queued us. */
113         module_put(entry->elem->owner);
114 }
115
116 /*
117  * Any packet that leaves via this function must come back
118  * through nf_reinject().
119  */
120 static int __nf_queue(struct sk_buff *skb,
121                       struct list_head *elem,
122                       u_int8_t pf, unsigned int hook,
123                       struct net_device *indev,
124                       struct net_device *outdev,
125                       int (*okfn)(struct sk_buff *),
126                       unsigned int queuenum)
127 {
128         int status = -ENOENT;
129         struct nf_queue_entry *entry = NULL;
130 #ifdef CONFIG_BRIDGE_NETFILTER
131         struct net_device *physindev;
132         struct net_device *physoutdev;
133 #endif
134         const struct nf_afinfo *afinfo;
135         const struct nf_queue_handler *qh;
136
137         /* QUEUE == DROP if noone is waiting, to be safe. */
138         rcu_read_lock();
139
140         qh = rcu_dereference(queue_handler[pf]);
141         if (!qh) {
142                 status = -ESRCH;
143                 goto err_unlock;
144         }
145
146         afinfo = nf_get_afinfo(pf);
147         if (!afinfo)
148                 goto err_unlock;
149
150         entry = kmalloc(sizeof(*entry) + afinfo->route_key_size, GFP_ATOMIC);
151         if (!entry) {
152                 status = -ENOMEM;
153                 goto err_unlock;
154         }
155
156         *entry = (struct nf_queue_entry) {
157                 .skb    = skb,
158                 .elem   = list_entry(elem, struct nf_hook_ops, list),
159                 .pf     = pf,
160                 .hook   = hook,
161                 .indev  = indev,
162                 .outdev = outdev,
163                 .okfn   = okfn,
164         };
165
166         /* If it's going away, ignore hook. */
167         if (!try_module_get(entry->elem->owner)) {
168                 status = -ECANCELED;
169                 goto err_unlock;
170         }
171         /* Bump dev refs so they don't vanish while packet is out */
172         if (indev)
173                 dev_hold(indev);
174         if (outdev)
175                 dev_hold(outdev);
176 #ifdef CONFIG_BRIDGE_NETFILTER
177         if (skb->nf_bridge) {
178                 physindev = skb->nf_bridge->physindev;
179                 if (physindev)
180                         dev_hold(physindev);
181                 physoutdev = skb->nf_bridge->physoutdev;
182                 if (physoutdev)
183                         dev_hold(physoutdev);
184         }
185 #endif
186         skb_dst_force(skb);
187         afinfo->saveroute(skb, entry);
188         status = qh->outfn(entry, queuenum);
189
190         rcu_read_unlock();
191
192         if (status < 0) {
193                 nf_queue_entry_release_refs(entry);
194                 goto err;
195         }
196
197         return 0;
198
199 err_unlock:
200         rcu_read_unlock();
201 err:
202         kfree(entry);
203         return status;
204 }
205
206 int nf_queue(struct sk_buff *skb,
207              struct list_head *elem,
208              u_int8_t pf, unsigned int hook,
209              struct net_device *indev,
210              struct net_device *outdev,
211              int (*okfn)(struct sk_buff *),
212              unsigned int queuenum)
213 {
214         struct sk_buff *segs;
215         int err;
216         unsigned int queued;
217
218         if (!skb_is_gso(skb))
219                 return __nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
220                                   queuenum);
221
222         switch (pf) {
223         case NFPROTO_IPV4:
224                 skb->protocol = htons(ETH_P_IP);
225                 break;
226         case NFPROTO_IPV6:
227                 skb->protocol = htons(ETH_P_IPV6);
228                 break;
229         }
230
231         segs = skb_gso_segment(skb, 0);
232         /* Does not use PTR_ERR to limit the number of error codes that can be
233          * returned by nf_queue.  For instance, callers rely on -ECANCELED to mean
234          * 'ignore this hook'.
235          */
236         if (IS_ERR(segs))
237                 return -EINVAL;
238
239         queued = 0;
240         err = 0;
241         do {
242                 struct sk_buff *nskb = segs->next;
243
244                 segs->next = NULL;
245                 if (err == 0)
246                         err = __nf_queue(segs, elem, pf, hook, indev,
247                                            outdev, okfn, queuenum);
248                 if (err == 0)
249                         queued++;
250                 else
251                         kfree_skb(segs);
252                 segs = nskb;
253         } while (segs);
254
255         /* also free orig skb if only some segments were queued */
256         if (unlikely(err && queued))
257                 err = 0;
258         if (err == 0)
259                 kfree_skb(skb);
260         return err;
261 }
262
263 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
264 {
265         struct sk_buff *skb = entry->skb;
266         struct list_head *elem = &entry->elem->list;
267         const struct nf_afinfo *afinfo;
268         int err;
269
270         rcu_read_lock();
271
272         nf_queue_entry_release_refs(entry);
273
274         /* Continue traversal iff userspace said ok... */
275         if (verdict == NF_REPEAT) {
276                 elem = elem->prev;
277                 verdict = NF_ACCEPT;
278         }
279
280         if (verdict == NF_ACCEPT) {
281                 afinfo = nf_get_afinfo(entry->pf);
282                 if (!afinfo || afinfo->reroute(skb, entry) < 0)
283                         verdict = NF_DROP;
284         }
285
286         if (verdict == NF_ACCEPT) {
287         next_hook:
288                 verdict = nf_iterate(&nf_hooks[entry->pf][entry->hook],
289                                      skb, entry->hook,
290                                      entry->indev, entry->outdev, &elem,
291                                      entry->okfn, INT_MIN);
292         }
293
294         switch (verdict & NF_VERDICT_MASK) {
295         case NF_ACCEPT:
296         case NF_STOP:
297                 local_bh_disable();
298                 entry->okfn(skb);
299                 local_bh_enable();
300                 break;
301         case NF_QUEUE:
302                 err = __nf_queue(skb, elem, entry->pf, entry->hook,
303                                  entry->indev, entry->outdev, entry->okfn,
304                                  verdict >> NF_VERDICT_QBITS);
305                 if (err < 0) {
306                         if (err == -ECANCELED)
307                                 goto next_hook;
308                         if (err == -ESRCH &&
309                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
310                                 goto next_hook;
311                         kfree_skb(skb);
312                 }
313                 break;
314         case NF_STOLEN:
315         default:
316                 kfree_skb(skb);
317         }
318         rcu_read_unlock();
319         kfree(entry);
320 }
321 EXPORT_SYMBOL(nf_reinject);
322
323 #ifdef CONFIG_PROC_FS
324 static void *seq_start(struct seq_file *seq, loff_t *pos)
325 {
326         if (*pos >= ARRAY_SIZE(queue_handler))
327                 return NULL;
328
329         return pos;
330 }
331
332 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
333 {
334         (*pos)++;
335
336         if (*pos >= ARRAY_SIZE(queue_handler))
337                 return NULL;
338
339         return pos;
340 }
341
342 static void seq_stop(struct seq_file *s, void *v)
343 {
344
345 }
346
347 static int seq_show(struct seq_file *s, void *v)
348 {
349         int ret;
350         loff_t *pos = v;
351         const struct nf_queue_handler *qh;
352
353         rcu_read_lock();
354         qh = rcu_dereference(queue_handler[*pos]);
355         if (!qh)
356                 ret = seq_printf(s, "%2lld NONE\n", *pos);
357         else
358                 ret = seq_printf(s, "%2lld %s\n", *pos, qh->name);
359         rcu_read_unlock();
360
361         return ret;
362 }
363
364 static const struct seq_operations nfqueue_seq_ops = {
365         .start  = seq_start,
366         .next   = seq_next,
367         .stop   = seq_stop,
368         .show   = seq_show,
369 };
370
371 static int nfqueue_open(struct inode *inode, struct file *file)
372 {
373         return seq_open(file, &nfqueue_seq_ops);
374 }
375
376 static const struct file_operations nfqueue_file_ops = {
377         .owner   = THIS_MODULE,
378         .open    = nfqueue_open,
379         .read    = seq_read,
380         .llseek  = seq_lseek,
381         .release = seq_release,
382 };
383 #endif /* PROC_FS */
384
385
386 int __init netfilter_queue_init(void)
387 {
388 #ifdef CONFIG_PROC_FS
389         if (!proc_create("nf_queue", S_IRUGO,
390                          proc_net_netfilter, &nfqueue_file_ops))
391                 return -1;
392 #endif
393         return 0;
394 }
395