1 /* L3/L4 protocol support for nf_conntrack. */
2 
3 /* (C) 1999-2001 Paul `Rusty' Russell
4  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
5  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
6  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  */
12 
13 #include <linux/types.h>
14 #include <linux/netfilter.h>
15 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <linux/mutex.h>
18 #include <linux/vmalloc.h>
19 #include <linux/stddef.h>
20 #include <linux/err.h>
21 #include <linux/percpu.h>
22 #include <linux/notifier.h>
23 #include <linux/kernel.h>
24 #include <linux/netdevice.h>
25 
26 #include <net/netfilter/nf_conntrack.h>
27 #include <net/netfilter/nf_conntrack_l3proto.h>
28 #include <net/netfilter/nf_conntrack_l4proto.h>
29 #include <net/netfilter/nf_conntrack_core.h>
30 #include <net/netfilter/nf_log.h>
31 
32 static struct nf_conntrack_l4proto __rcu **nf_ct_protos[NFPROTO_NUMPROTO] __read_mostly;
33 struct nf_conntrack_l3proto __rcu *nf_ct_l3protos[NFPROTO_NUMPROTO] __read_mostly;
34 EXPORT_SYMBOL_GPL(nf_ct_l3protos);
35 
36 static DEFINE_MUTEX(nf_ct_proto_mutex);
37 
38 #ifdef CONFIG_SYSCTL
39 static int
40 nf_ct_register_sysctl(struct net *net,
41 		      struct ctl_table_header **header,
42 		      const char *path,
43 		      struct ctl_table *table)
44 {
45 	if (*header == NULL) {
46 		*header = register_net_sysctl(net, path, table);
47 		if (*header == NULL)
48 			return -ENOMEM;
49 	}
50 
51 	return 0;
52 }
53 
54 static void
55 nf_ct_unregister_sysctl(struct ctl_table_header **header,
56 			struct ctl_table **table,
57 			unsigned int users)
58 {
59 	if (users > 0)
60 		return;
61 
62 	unregister_net_sysctl_table(*header);
63 	kfree(*table);
64 	*header = NULL;
65 	*table = NULL;
66 }
67 
68 __printf(5, 6)
69 void nf_l4proto_log_invalid(const struct sk_buff *skb,
70 			    struct net *net,
71 			    u16 pf, u8 protonum,
72 			    const char *fmt, ...)
73 {
74 	struct va_format vaf;
75 	va_list args;
76 
77 	if (net->ct.sysctl_log_invalid != protonum ||
78 	    net->ct.sysctl_log_invalid != IPPROTO_RAW)
79 		return;
80 
81 	va_start(args, fmt);
82 	vaf.fmt = fmt;
83 	vaf.va = &args;
84 
85 	nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
86 		      "nf_ct_proto_%d: %pV ", protonum, &vaf);
87 	va_end(args);
88 }
89 EXPORT_SYMBOL_GPL(nf_l4proto_log_invalid);
90 
91 __printf(3, 4)
92 void nf_ct_l4proto_log_invalid(const struct sk_buff *skb,
93 			       const struct nf_conn *ct,
94 			       const char *fmt, ...)
95 {
96 	struct va_format vaf;
97 	struct net *net;
98 	va_list args;
99 
100 	net = nf_ct_net(ct);
101 	if (likely(net->ct.sysctl_log_invalid == 0))
102 		return;
103 
104 	va_start(args, fmt);
105 	vaf.fmt = fmt;
106 	vaf.va = &args;
107 
108 	nf_l4proto_log_invalid(skb, net, nf_ct_l3num(ct),
109 			       nf_ct_protonum(ct), "%pV", &vaf);
110 	va_end(args);
111 }
112 EXPORT_SYMBOL_GPL(nf_ct_l4proto_log_invalid);
113 #endif
114 
115 const struct nf_conntrack_l4proto *
116 __nf_ct_l4proto_find(u_int16_t l3proto, u_int8_t l4proto)
117 {
118 	if (unlikely(l3proto >= NFPROTO_NUMPROTO || nf_ct_protos[l3proto] == NULL))
119 		return &nf_conntrack_l4proto_generic;
120 
121 	return rcu_dereference(nf_ct_protos[l3proto][l4proto]);
122 }
123 EXPORT_SYMBOL_GPL(__nf_ct_l4proto_find);
124 
125 /* this is guaranteed to always return a valid protocol helper, since
126  * it falls back to generic_protocol */
127 const struct nf_conntrack_l3proto *
128 nf_ct_l3proto_find_get(u_int16_t l3proto)
129 {
130 	struct nf_conntrack_l3proto *p;
131 
132 	rcu_read_lock();
133 	p = __nf_ct_l3proto_find(l3proto);
134 	if (!try_module_get(p->me))
135 		p = &nf_conntrack_l3proto_generic;
136 	rcu_read_unlock();
137 
138 	return p;
139 }
140 EXPORT_SYMBOL_GPL(nf_ct_l3proto_find_get);
141 
142 int
143 nf_ct_l3proto_try_module_get(unsigned short l3proto)
144 {
145 	const struct nf_conntrack_l3proto *p;
146 	int ret;
147 
148 retry:	p = nf_ct_l3proto_find_get(l3proto);
149 	if (p == &nf_conntrack_l3proto_generic) {
150 		ret = request_module("nf_conntrack-%d", l3proto);
151 		if (!ret)
152 			goto retry;
153 
154 		return -EPROTOTYPE;
155 	}
156 
157 	return 0;
158 }
159 EXPORT_SYMBOL_GPL(nf_ct_l3proto_try_module_get);
160 
161 void nf_ct_l3proto_module_put(unsigned short l3proto)
162 {
163 	struct nf_conntrack_l3proto *p;
164 
165 	/* rcu_read_lock not necessary since the caller holds a reference, but
166 	 * taken anyways to avoid lockdep warnings in __nf_ct_l3proto_find()
167 	 */
168 	rcu_read_lock();
169 	p = __nf_ct_l3proto_find(l3proto);
170 	module_put(p->me);
171 	rcu_read_unlock();
172 }
173 EXPORT_SYMBOL_GPL(nf_ct_l3proto_module_put);
174 
175 static int nf_ct_netns_do_get(struct net *net, u8 nfproto)
176 {
177 	const struct nf_conntrack_l3proto *l3proto;
178 	int ret;
179 
180 	might_sleep();
181 
182 	ret = nf_ct_l3proto_try_module_get(nfproto);
183 	if (ret < 0)
184 		return ret;
185 
186 	/* we already have a reference, can't fail */
187 	rcu_read_lock();
188 	l3proto = __nf_ct_l3proto_find(nfproto);
189 	rcu_read_unlock();
190 
191 	if (!l3proto->net_ns_get)
192 		return 0;
193 
194 	ret = l3proto->net_ns_get(net);
195 	if (ret < 0)
196 		nf_ct_l3proto_module_put(nfproto);
197 
198 	return ret;
199 }
200 
201 int nf_ct_netns_get(struct net *net, u8 nfproto)
202 {
203 	int err;
204 
205 	if (nfproto == NFPROTO_INET) {
206 		err = nf_ct_netns_do_get(net, NFPROTO_IPV4);
207 		if (err < 0)
208 			goto err1;
209 		err = nf_ct_netns_do_get(net, NFPROTO_IPV6);
210 		if (err < 0)
211 			goto err2;
212 	} else {
213 		err = nf_ct_netns_do_get(net, nfproto);
214 		if (err < 0)
215 			goto err1;
216 	}
217 	return 0;
218 
219 err2:
220 	nf_ct_netns_put(net, NFPROTO_IPV4);
221 err1:
222 	return err;
223 }
224 EXPORT_SYMBOL_GPL(nf_ct_netns_get);
225 
226 static void nf_ct_netns_do_put(struct net *net, u8 nfproto)
227 {
228 	const struct nf_conntrack_l3proto *l3proto;
229 
230 	might_sleep();
231 
232 	/* same as nf_conntrack_netns_get(), reference assumed */
233 	rcu_read_lock();
234 	l3proto = __nf_ct_l3proto_find(nfproto);
235 	rcu_read_unlock();
236 
237 	if (WARN_ON(!l3proto))
238 		return;
239 
240 	if (l3proto->net_ns_put)
241 		l3proto->net_ns_put(net);
242 
243 	nf_ct_l3proto_module_put(nfproto);
244 }
245 
246 void nf_ct_netns_put(struct net *net, uint8_t nfproto)
247 {
248 	if (nfproto == NFPROTO_INET) {
249 		nf_ct_netns_do_put(net, NFPROTO_IPV4);
250 		nf_ct_netns_do_put(net, NFPROTO_IPV6);
251 	} else
252 		nf_ct_netns_do_put(net, nfproto);
253 }
254 EXPORT_SYMBOL_GPL(nf_ct_netns_put);
255 
256 const struct nf_conntrack_l4proto *
257 nf_ct_l4proto_find_get(u_int16_t l3num, u_int8_t l4num)
258 {
259 	const struct nf_conntrack_l4proto *p;
260 
261 	rcu_read_lock();
262 	p = __nf_ct_l4proto_find(l3num, l4num);
263 	if (!try_module_get(p->me))
264 		p = &nf_conntrack_l4proto_generic;
265 	rcu_read_unlock();
266 
267 	return p;
268 }
269 EXPORT_SYMBOL_GPL(nf_ct_l4proto_find_get);
270 
271 void nf_ct_l4proto_put(const struct nf_conntrack_l4proto *p)
272 {
273 	module_put(p->me);
274 }
275 EXPORT_SYMBOL_GPL(nf_ct_l4proto_put);
276 
277 static int kill_l3proto(struct nf_conn *i, void *data)
278 {
279 	return nf_ct_l3num(i) == ((const struct nf_conntrack_l3proto *)data)->l3proto;
280 }
281 
282 static int kill_l4proto(struct nf_conn *i, void *data)
283 {
284 	const struct nf_conntrack_l4proto *l4proto;
285 	l4proto = data;
286 	return nf_ct_protonum(i) == l4proto->l4proto &&
287 	       nf_ct_l3num(i) == l4proto->l3proto;
288 }
289 
290 int nf_ct_l3proto_register(const struct nf_conntrack_l3proto *proto)
291 {
292 	int ret = 0;
293 	struct nf_conntrack_l3proto *old;
294 
295 	if (proto->l3proto >= NFPROTO_NUMPROTO)
296 		return -EBUSY;
297 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
298 	if (proto->tuple_to_nlattr && proto->nla_size == 0)
299 		return -EINVAL;
300 #endif
301 	mutex_lock(&nf_ct_proto_mutex);
302 	old = rcu_dereference_protected(nf_ct_l3protos[proto->l3proto],
303 					lockdep_is_held(&nf_ct_proto_mutex));
304 	if (old != &nf_conntrack_l3proto_generic) {
305 		ret = -EBUSY;
306 		goto out_unlock;
307 	}
308 
309 	rcu_assign_pointer(nf_ct_l3protos[proto->l3proto], proto);
310 
311 out_unlock:
312 	mutex_unlock(&nf_ct_proto_mutex);
313 	return ret;
314 
315 }
316 EXPORT_SYMBOL_GPL(nf_ct_l3proto_register);
317 
318 void nf_ct_l3proto_unregister(const struct nf_conntrack_l3proto *proto)
319 {
320 	BUG_ON(proto->l3proto >= NFPROTO_NUMPROTO);
321 
322 	mutex_lock(&nf_ct_proto_mutex);
323 	BUG_ON(rcu_dereference_protected(nf_ct_l3protos[proto->l3proto],
324 					 lockdep_is_held(&nf_ct_proto_mutex)
325 					 ) != proto);
326 	rcu_assign_pointer(nf_ct_l3protos[proto->l3proto],
327 			   &nf_conntrack_l3proto_generic);
328 	mutex_unlock(&nf_ct_proto_mutex);
329 
330 	synchronize_rcu();
331 	/* Remove all contrack entries for this protocol */
332 	nf_ct_iterate_destroy(kill_l3proto, (void*)proto);
333 }
334 EXPORT_SYMBOL_GPL(nf_ct_l3proto_unregister);
335 
336 static struct nf_proto_net *nf_ct_l4proto_net(struct net *net,
337 				const struct nf_conntrack_l4proto *l4proto)
338 {
339 	if (l4proto->get_net_proto) {
340 		/* statically built-in protocols use static per-net */
341 		return l4proto->get_net_proto(net);
342 	} else if (l4proto->net_id) {
343 		/* ... and loadable protocols use dynamic per-net */
344 		return net_generic(net, *l4proto->net_id);
345 	}
346 	return NULL;
347 }
348 
349 static
350 int nf_ct_l4proto_register_sysctl(struct net *net,
351 				  struct nf_proto_net *pn,
352 				  const struct nf_conntrack_l4proto *l4proto)
353 {
354 	int err = 0;
355 
356 #ifdef CONFIG_SYSCTL
357 	if (pn->ctl_table != NULL) {
358 		err = nf_ct_register_sysctl(net,
359 					    &pn->ctl_table_header,
360 					    "net/netfilter",
361 					    pn->ctl_table);
362 		if (err < 0) {
363 			if (!pn->users) {
364 				kfree(pn->ctl_table);
365 				pn->ctl_table = NULL;
366 			}
367 		}
368 	}
369 #endif /* CONFIG_SYSCTL */
370 	return err;
371 }
372 
373 static
374 void nf_ct_l4proto_unregister_sysctl(struct net *net,
375 				struct nf_proto_net *pn,
376 				const struct nf_conntrack_l4proto *l4proto)
377 {
378 #ifdef CONFIG_SYSCTL
379 	if (pn->ctl_table_header != NULL)
380 		nf_ct_unregister_sysctl(&pn->ctl_table_header,
381 					&pn->ctl_table,
382 					pn->users);
383 #endif /* CONFIG_SYSCTL */
384 }
385 
386 /* FIXME: Allow NULL functions and sub in pointers to generic for
387    them. --RR */
388 int nf_ct_l4proto_register_one(const struct nf_conntrack_l4proto *l4proto)
389 {
390 	int ret = 0;
391 
392 	if (l4proto->l3proto >= ARRAY_SIZE(nf_ct_protos))
393 		return -EBUSY;
394 
395 	if ((l4proto->to_nlattr && l4proto->nlattr_size == 0) ||
396 	    (l4proto->tuple_to_nlattr && !l4proto->nlattr_tuple_size))
397 		return -EINVAL;
398 
399 	mutex_lock(&nf_ct_proto_mutex);
400 	if (!nf_ct_protos[l4proto->l3proto]) {
401 		/* l3proto may be loaded latter. */
402 		struct nf_conntrack_l4proto __rcu **proto_array;
403 		int i;
404 
405 		proto_array =
406 			kmalloc_array(MAX_NF_CT_PROTO,
407 				      sizeof(struct nf_conntrack_l4proto *),
408 				      GFP_KERNEL);
409 		if (proto_array == NULL) {
410 			ret = -ENOMEM;
411 			goto out_unlock;
412 		}
413 
414 		for (i = 0; i < MAX_NF_CT_PROTO; i++)
415 			RCU_INIT_POINTER(proto_array[i],
416 					 &nf_conntrack_l4proto_generic);
417 
418 		/* Before making proto_array visible to lockless readers,
419 		 * we must make sure its content is committed to memory.
420 		 */
421 		smp_wmb();
422 
423 		nf_ct_protos[l4proto->l3proto] = proto_array;
424 	} else if (rcu_dereference_protected(
425 			nf_ct_protos[l4proto->l3proto][l4proto->l4proto],
426 			lockdep_is_held(&nf_ct_proto_mutex)
427 			) != &nf_conntrack_l4proto_generic) {
428 		ret = -EBUSY;
429 		goto out_unlock;
430 	}
431 
432 	rcu_assign_pointer(nf_ct_protos[l4proto->l3proto][l4proto->l4proto],
433 			   l4proto);
434 out_unlock:
435 	mutex_unlock(&nf_ct_proto_mutex);
436 	return ret;
437 }
438 EXPORT_SYMBOL_GPL(nf_ct_l4proto_register_one);
439 
440 int nf_ct_l4proto_pernet_register_one(struct net *net,
441 				const struct nf_conntrack_l4proto *l4proto)
442 {
443 	int ret = 0;
444 	struct nf_proto_net *pn = NULL;
445 
446 	if (l4proto->init_net) {
447 		ret = l4proto->init_net(net, l4proto->l3proto);
448 		if (ret < 0)
449 			goto out;
450 	}
451 
452 	pn = nf_ct_l4proto_net(net, l4proto);
453 	if (pn == NULL)
454 		goto out;
455 
456 	ret = nf_ct_l4proto_register_sysctl(net, pn, l4proto);
457 	if (ret < 0)
458 		goto out;
459 
460 	pn->users++;
461 out:
462 	return ret;
463 }
464 EXPORT_SYMBOL_GPL(nf_ct_l4proto_pernet_register_one);
465 
466 static void __nf_ct_l4proto_unregister_one(const struct nf_conntrack_l4proto *l4proto)
467 
468 {
469 	BUG_ON(l4proto->l3proto >= ARRAY_SIZE(nf_ct_protos));
470 
471 	BUG_ON(rcu_dereference_protected(
472 			nf_ct_protos[l4proto->l3proto][l4proto->l4proto],
473 			lockdep_is_held(&nf_ct_proto_mutex)
474 			) != l4proto);
475 	rcu_assign_pointer(nf_ct_protos[l4proto->l3proto][l4proto->l4proto],
476 			   &nf_conntrack_l4proto_generic);
477 }
478 
479 void nf_ct_l4proto_unregister_one(const struct nf_conntrack_l4proto *l4proto)
480 {
481 	mutex_lock(&nf_ct_proto_mutex);
482 	__nf_ct_l4proto_unregister_one(l4proto);
483 	mutex_unlock(&nf_ct_proto_mutex);
484 
485 	synchronize_rcu();
486 }
487 EXPORT_SYMBOL_GPL(nf_ct_l4proto_unregister_one);
488 
489 void nf_ct_l4proto_pernet_unregister_one(struct net *net,
490 				const struct nf_conntrack_l4proto *l4proto)
491 {
492 	struct nf_proto_net *pn = nf_ct_l4proto_net(net, l4proto);
493 
494 	if (pn == NULL)
495 		return;
496 
497 	pn->users--;
498 	nf_ct_l4proto_unregister_sysctl(net, pn, l4proto);
499 }
500 EXPORT_SYMBOL_GPL(nf_ct_l4proto_pernet_unregister_one);
501 
502 int nf_ct_l4proto_register(const struct nf_conntrack_l4proto * const l4proto[],
503 			   unsigned int num_proto)
504 {
505 	int ret = -EINVAL, ver;
506 	unsigned int i;
507 
508 	for (i = 0; i < num_proto; i++) {
509 		ret = nf_ct_l4proto_register_one(l4proto[i]);
510 		if (ret < 0)
511 			break;
512 	}
513 	if (i != num_proto) {
514 		ver = l4proto[i]->l3proto == PF_INET6 ? 6 : 4;
515 		pr_err("nf_conntrack_ipv%d: can't register l4 %d proto.\n",
516 		       ver, l4proto[i]->l4proto);
517 		nf_ct_l4proto_unregister(l4proto, i);
518 	}
519 	return ret;
520 }
521 EXPORT_SYMBOL_GPL(nf_ct_l4proto_register);
522 
523 int nf_ct_l4proto_pernet_register(struct net *net,
524 				  const struct nf_conntrack_l4proto *const l4proto[],
525 				  unsigned int num_proto)
526 {
527 	int ret = -EINVAL;
528 	unsigned int i;
529 
530 	for (i = 0; i < num_proto; i++) {
531 		ret = nf_ct_l4proto_pernet_register_one(net, l4proto[i]);
532 		if (ret < 0)
533 			break;
534 	}
535 	if (i != num_proto) {
536 		pr_err("nf_conntrack_proto_%d %d: pernet registration failed\n",
537 		       l4proto[i]->l4proto,
538 		       l4proto[i]->l3proto == PF_INET6 ? 6 : 4);
539 		nf_ct_l4proto_pernet_unregister(net, l4proto, i);
540 	}
541 	return ret;
542 }
543 EXPORT_SYMBOL_GPL(nf_ct_l4proto_pernet_register);
544 
545 void nf_ct_l4proto_unregister(const struct nf_conntrack_l4proto * const l4proto[],
546 			      unsigned int num_proto)
547 {
548 	mutex_lock(&nf_ct_proto_mutex);
549 	while (num_proto-- != 0)
550 		__nf_ct_l4proto_unregister_one(l4proto[num_proto]);
551 	mutex_unlock(&nf_ct_proto_mutex);
552 
553 	synchronize_net();
554 	/* Remove all contrack entries for this protocol */
555 	nf_ct_iterate_destroy(kill_l4proto, (void *)l4proto);
556 }
557 EXPORT_SYMBOL_GPL(nf_ct_l4proto_unregister);
558 
559 void nf_ct_l4proto_pernet_unregister(struct net *net,
560 				const struct nf_conntrack_l4proto *const l4proto[],
561 				unsigned int num_proto)
562 {
563 	while (num_proto-- != 0)
564 		nf_ct_l4proto_pernet_unregister_one(net, l4proto[num_proto]);
565 }
566 EXPORT_SYMBOL_GPL(nf_ct_l4proto_pernet_unregister);
567 
568 int nf_conntrack_proto_pernet_init(struct net *net)
569 {
570 	int err;
571 	struct nf_proto_net *pn = nf_ct_l4proto_net(net,
572 					&nf_conntrack_l4proto_generic);
573 
574 	err = nf_conntrack_l4proto_generic.init_net(net,
575 					nf_conntrack_l4proto_generic.l3proto);
576 	if (err < 0)
577 		return err;
578 	err = nf_ct_l4proto_register_sysctl(net,
579 					    pn,
580 					    &nf_conntrack_l4proto_generic);
581 	if (err < 0)
582 		return err;
583 
584 	pn->users++;
585 	return 0;
586 }
587 
588 void nf_conntrack_proto_pernet_fini(struct net *net)
589 {
590 	struct nf_proto_net *pn = nf_ct_l4proto_net(net,
591 					&nf_conntrack_l4proto_generic);
592 
593 	pn->users--;
594 	nf_ct_l4proto_unregister_sysctl(net,
595 					pn,
596 					&nf_conntrack_l4proto_generic);
597 }
598 
599 int nf_conntrack_proto_init(void)
600 {
601 	unsigned int i;
602 	for (i = 0; i < NFPROTO_NUMPROTO; i++)
603 		rcu_assign_pointer(nf_ct_l3protos[i],
604 				   &nf_conntrack_l3proto_generic);
605 	return 0;
606 }
607 
608 void nf_conntrack_proto_fini(void)
609 {
610 	unsigned int i;
611 	/* free l3proto protocol tables */
612 	for (i = 0; i < ARRAY_SIZE(nf_ct_protos); i++)
613 		kfree(nf_ct_protos[i]);
614 }
615