1 /* Event cache for netfilter. */
2 
3 /*
4  * (C) 2005 Harald Welte <laforge@gnumonks.org>
5  * (C) 2005 Patrick McHardy <kaber@trash.net>
6  * (C) 2005-2006 Netfilter Core Team <coreteam@netfilter.org>
7  * (C) 2005 USAGI/WIDE Project <http://www.linux-ipv6.org>
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  */
13 
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/skbuff.h>
17 #include <linux/vmalloc.h>
18 #include <linux/stddef.h>
19 #include <linux/err.h>
20 #include <linux/percpu.h>
21 #include <linux/kernel.h>
22 #include <linux/netdevice.h>
23 #include <linux/slab.h>
24 #include <linux/export.h>
25 
26 #include <net/netfilter/nf_conntrack.h>
27 #include <net/netfilter/nf_conntrack_core.h>
28 #include <net/netfilter/nf_conntrack_extend.h>
29 
30 static DEFINE_MUTEX(nf_ct_ecache_mutex);
31 
32 #define ECACHE_RETRY_WAIT (HZ/10)
33 
34 enum retry_state {
35 	STATE_CONGESTED,
36 	STATE_RESTART,
37 	STATE_DONE,
38 };
39 
40 static enum retry_state ecache_work_evict_list(struct ct_pcpu *pcpu)
41 {
42 	struct nf_conn *refs[16];
43 	struct nf_conntrack_tuple_hash *h;
44 	struct hlist_nulls_node *n;
45 	unsigned int evicted = 0;
46 	enum retry_state ret = STATE_DONE;
47 
48 	spin_lock(&pcpu->lock);
49 
50 	hlist_nulls_for_each_entry(h, n, &pcpu->dying, hnnode) {
51 		struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
52 
53 		if (nf_ct_is_dying(ct))
54 			continue;
55 
56 		if (nf_conntrack_event(IPCT_DESTROY, ct)) {
57 			ret = STATE_CONGESTED;
58 			break;
59 		}
60 
61 		/* we've got the event delivered, now it's dying */
62 		set_bit(IPS_DYING_BIT, &ct->status);
63 		refs[evicted] = ct;
64 
65 		if (++evicted >= ARRAY_SIZE(refs)) {
66 			ret = STATE_RESTART;
67 			break;
68 		}
69 	}
70 
71 	spin_unlock(&pcpu->lock);
72 
73 	/* can't _put while holding lock */
74 	while (evicted)
75 		nf_ct_put(refs[--evicted]);
76 
77 	return ret;
78 }
79 
80 static void ecache_work(struct work_struct *work)
81 {
82 	struct netns_ct *ctnet =
83 		container_of(work, struct netns_ct, ecache_dwork.work);
84 	int cpu, delay = -1;
85 	struct ct_pcpu *pcpu;
86 
87 	local_bh_disable();
88 
89 	for_each_possible_cpu(cpu) {
90 		enum retry_state ret;
91 
92 		pcpu = per_cpu_ptr(ctnet->pcpu_lists, cpu);
93 
94 		ret = ecache_work_evict_list(pcpu);
95 
96 		switch (ret) {
97 		case STATE_CONGESTED:
98 			delay = ECACHE_RETRY_WAIT;
99 			goto out;
100 		case STATE_RESTART:
101 			delay = 0;
102 			break;
103 		case STATE_DONE:
104 			break;
105 		}
106 	}
107 
108  out:
109 	local_bh_enable();
110 
111 	ctnet->ecache_dwork_pending = delay > 0;
112 	if (delay >= 0)
113 		schedule_delayed_work(&ctnet->ecache_dwork, delay);
114 }
115 
116 int nf_conntrack_eventmask_report(unsigned int eventmask, struct nf_conn *ct,
117 				  u32 portid, int report)
118 {
119 	int ret = 0;
120 	struct net *net = nf_ct_net(ct);
121 	struct nf_ct_event_notifier *notify;
122 	struct nf_conntrack_ecache *e;
123 
124 	rcu_read_lock();
125 	notify = rcu_dereference(net->ct.nf_conntrack_event_cb);
126 	if (!notify)
127 		goto out_unlock;
128 
129 	e = nf_ct_ecache_find(ct);
130 	if (!e)
131 		goto out_unlock;
132 
133 	if (nf_ct_is_confirmed(ct) && !nf_ct_is_dying(ct)) {
134 		struct nf_ct_event item = {
135 			.ct	= ct,
136 			.portid	= e->portid ? e->portid : portid,
137 			.report = report
138 		};
139 		/* This is a resent of a destroy event? If so, skip missed */
140 		unsigned long missed = e->portid ? 0 : e->missed;
141 
142 		if (!((eventmask | missed) & e->ctmask))
143 			goto out_unlock;
144 
145 		ret = notify->fcn(eventmask | missed, &item);
146 		if (unlikely(ret < 0 || missed)) {
147 			spin_lock_bh(&ct->lock);
148 			if (ret < 0) {
149 				/* This is a destroy event that has been
150 				 * triggered by a process, we store the PORTID
151 				 * to include it in the retransmission.
152 				 */
153 				if (eventmask & (1 << IPCT_DESTROY) &&
154 				    e->portid == 0 && portid != 0)
155 					e->portid = portid;
156 				else
157 					e->missed |= eventmask;
158 			} else {
159 				e->missed &= ~missed;
160 			}
161 			spin_unlock_bh(&ct->lock);
162 		}
163 	}
164 out_unlock:
165 	rcu_read_unlock();
166 	return ret;
167 }
168 EXPORT_SYMBOL_GPL(nf_conntrack_eventmask_report);
169 
170 /* deliver cached events and clear cache entry - must be called with locally
171  * disabled softirqs */
172 void nf_ct_deliver_cached_events(struct nf_conn *ct)
173 {
174 	struct net *net = nf_ct_net(ct);
175 	unsigned long events, missed;
176 	struct nf_ct_event_notifier *notify;
177 	struct nf_conntrack_ecache *e;
178 	struct nf_ct_event item;
179 	int ret;
180 
181 	rcu_read_lock();
182 	notify = rcu_dereference(net->ct.nf_conntrack_event_cb);
183 	if (notify == NULL)
184 		goto out_unlock;
185 
186 	e = nf_ct_ecache_find(ct);
187 	if (e == NULL)
188 		goto out_unlock;
189 
190 	events = xchg(&e->cache, 0);
191 
192 	if (!nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct) || !events)
193 		goto out_unlock;
194 
195 	/* We make a copy of the missed event cache without taking
196 	 * the lock, thus we may send missed events twice. However,
197 	 * this does not harm and it happens very rarely. */
198 	missed = e->missed;
199 
200 	if (!((events | missed) & e->ctmask))
201 		goto out_unlock;
202 
203 	item.ct = ct;
204 	item.portid = 0;
205 	item.report = 0;
206 
207 	ret = notify->fcn(events | missed, &item);
208 
209 	if (likely(ret >= 0 && !missed))
210 		goto out_unlock;
211 
212 	spin_lock_bh(&ct->lock);
213 	if (ret < 0)
214 		e->missed |= events;
215 	else
216 		e->missed &= ~missed;
217 	spin_unlock_bh(&ct->lock);
218 
219 out_unlock:
220 	rcu_read_unlock();
221 }
222 EXPORT_SYMBOL_GPL(nf_ct_deliver_cached_events);
223 
224 void nf_ct_expect_event_report(enum ip_conntrack_expect_events event,
225 			       struct nf_conntrack_expect *exp,
226 			       u32 portid, int report)
227 
228 {
229 	struct net *net = nf_ct_exp_net(exp);
230 	struct nf_exp_event_notifier *notify;
231 	struct nf_conntrack_ecache *e;
232 
233 	rcu_read_lock();
234 	notify = rcu_dereference(net->ct.nf_expect_event_cb);
235 	if (!notify)
236 		goto out_unlock;
237 
238 	e = nf_ct_ecache_find(exp->master);
239 	if (!e)
240 		goto out_unlock;
241 
242 	if (e->expmask & (1 << event)) {
243 		struct nf_exp_event item = {
244 			.exp	= exp,
245 			.portid	= portid,
246 			.report = report
247 		};
248 		notify->fcn(1 << event, &item);
249 	}
250 out_unlock:
251 	rcu_read_unlock();
252 }
253 
254 int nf_conntrack_register_notifier(struct net *net,
255 				   struct nf_ct_event_notifier *new)
256 {
257 	int ret;
258 	struct nf_ct_event_notifier *notify;
259 
260 	mutex_lock(&nf_ct_ecache_mutex);
261 	notify = rcu_dereference_protected(net->ct.nf_conntrack_event_cb,
262 					   lockdep_is_held(&nf_ct_ecache_mutex));
263 	if (notify != NULL) {
264 		ret = -EBUSY;
265 		goto out_unlock;
266 	}
267 	rcu_assign_pointer(net->ct.nf_conntrack_event_cb, new);
268 	ret = 0;
269 
270 out_unlock:
271 	mutex_unlock(&nf_ct_ecache_mutex);
272 	return ret;
273 }
274 EXPORT_SYMBOL_GPL(nf_conntrack_register_notifier);
275 
276 void nf_conntrack_unregister_notifier(struct net *net,
277 				      struct nf_ct_event_notifier *new)
278 {
279 	struct nf_ct_event_notifier *notify;
280 
281 	mutex_lock(&nf_ct_ecache_mutex);
282 	notify = rcu_dereference_protected(net->ct.nf_conntrack_event_cb,
283 					   lockdep_is_held(&nf_ct_ecache_mutex));
284 	BUG_ON(notify != new);
285 	RCU_INIT_POINTER(net->ct.nf_conntrack_event_cb, NULL);
286 	mutex_unlock(&nf_ct_ecache_mutex);
287 }
288 EXPORT_SYMBOL_GPL(nf_conntrack_unregister_notifier);
289 
290 int nf_ct_expect_register_notifier(struct net *net,
291 				   struct nf_exp_event_notifier *new)
292 {
293 	int ret;
294 	struct nf_exp_event_notifier *notify;
295 
296 	mutex_lock(&nf_ct_ecache_mutex);
297 	notify = rcu_dereference_protected(net->ct.nf_expect_event_cb,
298 					   lockdep_is_held(&nf_ct_ecache_mutex));
299 	if (notify != NULL) {
300 		ret = -EBUSY;
301 		goto out_unlock;
302 	}
303 	rcu_assign_pointer(net->ct.nf_expect_event_cb, new);
304 	ret = 0;
305 
306 out_unlock:
307 	mutex_unlock(&nf_ct_ecache_mutex);
308 	return ret;
309 }
310 EXPORT_SYMBOL_GPL(nf_ct_expect_register_notifier);
311 
312 void nf_ct_expect_unregister_notifier(struct net *net,
313 				      struct nf_exp_event_notifier *new)
314 {
315 	struct nf_exp_event_notifier *notify;
316 
317 	mutex_lock(&nf_ct_ecache_mutex);
318 	notify = rcu_dereference_protected(net->ct.nf_expect_event_cb,
319 					   lockdep_is_held(&nf_ct_ecache_mutex));
320 	BUG_ON(notify != new);
321 	RCU_INIT_POINTER(net->ct.nf_expect_event_cb, NULL);
322 	mutex_unlock(&nf_ct_ecache_mutex);
323 }
324 EXPORT_SYMBOL_GPL(nf_ct_expect_unregister_notifier);
325 
326 #define NF_CT_EVENTS_DEFAULT 1
327 static int nf_ct_events __read_mostly = NF_CT_EVENTS_DEFAULT;
328 
329 #ifdef CONFIG_SYSCTL
330 static struct ctl_table event_sysctl_table[] = {
331 	{
332 		.procname	= "nf_conntrack_events",
333 		.data		= &init_net.ct.sysctl_events,
334 		.maxlen		= sizeof(unsigned int),
335 		.mode		= 0644,
336 		.proc_handler	= proc_dointvec,
337 	},
338 	{}
339 };
340 #endif /* CONFIG_SYSCTL */
341 
342 static struct nf_ct_ext_type event_extend __read_mostly = {
343 	.len	= sizeof(struct nf_conntrack_ecache),
344 	.align	= __alignof__(struct nf_conntrack_ecache),
345 	.id	= NF_CT_EXT_ECACHE,
346 };
347 
348 #ifdef CONFIG_SYSCTL
349 static int nf_conntrack_event_init_sysctl(struct net *net)
350 {
351 	struct ctl_table *table;
352 
353 	table = kmemdup(event_sysctl_table, sizeof(event_sysctl_table),
354 			GFP_KERNEL);
355 	if (!table)
356 		goto out;
357 
358 	table[0].data = &net->ct.sysctl_events;
359 
360 	/* Don't export sysctls to unprivileged users */
361 	if (net->user_ns != &init_user_ns)
362 		table[0].procname = NULL;
363 
364 	net->ct.event_sysctl_header =
365 		register_net_sysctl(net, "net/netfilter", table);
366 	if (!net->ct.event_sysctl_header) {
367 		printk(KERN_ERR "nf_ct_event: can't register to sysctl.\n");
368 		goto out_register;
369 	}
370 	return 0;
371 
372 out_register:
373 	kfree(table);
374 out:
375 	return -ENOMEM;
376 }
377 
378 static void nf_conntrack_event_fini_sysctl(struct net *net)
379 {
380 	struct ctl_table *table;
381 
382 	table = net->ct.event_sysctl_header->ctl_table_arg;
383 	unregister_net_sysctl_table(net->ct.event_sysctl_header);
384 	kfree(table);
385 }
386 #else
387 static int nf_conntrack_event_init_sysctl(struct net *net)
388 {
389 	return 0;
390 }
391 
392 static void nf_conntrack_event_fini_sysctl(struct net *net)
393 {
394 }
395 #endif /* CONFIG_SYSCTL */
396 
397 int nf_conntrack_ecache_pernet_init(struct net *net)
398 {
399 	net->ct.sysctl_events = nf_ct_events;
400 	INIT_DELAYED_WORK(&net->ct.ecache_dwork, ecache_work);
401 	return nf_conntrack_event_init_sysctl(net);
402 }
403 
404 void nf_conntrack_ecache_pernet_fini(struct net *net)
405 {
406 	cancel_delayed_work_sync(&net->ct.ecache_dwork);
407 	nf_conntrack_event_fini_sysctl(net);
408 }
409 
410 int nf_conntrack_ecache_init(void)
411 {
412 	int ret = nf_ct_extend_register(&event_extend);
413 	if (ret < 0)
414 		pr_err("nf_ct_event: Unable to register event extension.\n");
415 	return ret;
416 }
417 
418 void nf_conntrack_ecache_fini(void)
419 {
420 	nf_ct_extend_unregister(&event_extend);
421 }
422