1/*
2 * (C) 2012 by Pablo Neira Ayuso <pablo@netfilter.org>
3 * (C) 2012 by Vyatta Inc. <http://www.vyatta.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation (or any later at your option).
8 */
9#include <linux/init.h>
10#include <linux/module.h>
11#include <linux/kernel.h>
12#include <linux/rculist.h>
13#include <linux/rculist_nulls.h>
14#include <linux/types.h>
15#include <linux/timer.h>
16#include <linux/security.h>
17#include <linux/skbuff.h>
18#include <linux/errno.h>
19#include <linux/netlink.h>
20#include <linux/spinlock.h>
21#include <linux/interrupt.h>
22#include <linux/slab.h>
23
24#include <linux/netfilter.h>
25#include <net/netlink.h>
26#include <net/sock.h>
27#include <net/netfilter/nf_conntrack.h>
28#include <net/netfilter/nf_conntrack_core.h>
29#include <net/netfilter/nf_conntrack_l3proto.h>
30#include <net/netfilter/nf_conntrack_l4proto.h>
31#include <net/netfilter/nf_conntrack_tuple.h>
32#include <net/netfilter/nf_conntrack_timeout.h>
33
34#include <linux/netfilter/nfnetlink.h>
35#include <linux/netfilter/nfnetlink_cttimeout.h>
36
37MODULE_LICENSE("GPL");
38MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
39MODULE_DESCRIPTION("cttimeout: Extended Netfilter Connection Tracking timeout tuning");
40
41static LIST_HEAD(cttimeout_list);
42
43static const struct nla_policy cttimeout_nla_policy[CTA_TIMEOUT_MAX+1] = {
44	[CTA_TIMEOUT_NAME]	= { .type = NLA_NUL_STRING,
45				    .len  = CTNL_TIMEOUT_NAME_MAX - 1},
46	[CTA_TIMEOUT_L3PROTO]	= { .type = NLA_U16 },
47	[CTA_TIMEOUT_L4PROTO]	= { .type = NLA_U8 },
48	[CTA_TIMEOUT_DATA]	= { .type = NLA_NESTED },
49};
50
51static int
52ctnl_timeout_parse_policy(void *timeouts, struct nf_conntrack_l4proto *l4proto,
53			  struct net *net, const struct nlattr *attr)
54{
55	int ret = 0;
56
57	if (likely(l4proto->ctnl_timeout.nlattr_to_obj)) {
58		struct nlattr *tb[l4proto->ctnl_timeout.nlattr_max+1];
59
60		ret = nla_parse_nested(tb, l4proto->ctnl_timeout.nlattr_max,
61				       attr, l4proto->ctnl_timeout.nla_policy);
62		if (ret < 0)
63			return ret;
64
65		ret = l4proto->ctnl_timeout.nlattr_to_obj(tb, net, timeouts);
66	}
67	return ret;
68}
69
70static int
71cttimeout_new_timeout(struct sock *ctnl, struct sk_buff *skb,
72		      const struct nlmsghdr *nlh,
73		      const struct nlattr * const cda[])
74{
75	__u16 l3num;
76	__u8 l4num;
77	struct nf_conntrack_l4proto *l4proto;
78	struct ctnl_timeout *timeout, *matching = NULL;
79	struct net *net = sock_net(skb->sk);
80	char *name;
81	int ret;
82
83	if (!cda[CTA_TIMEOUT_NAME] ||
84	    !cda[CTA_TIMEOUT_L3PROTO] ||
85	    !cda[CTA_TIMEOUT_L4PROTO] ||
86	    !cda[CTA_TIMEOUT_DATA])
87		return -EINVAL;
88
89	name = nla_data(cda[CTA_TIMEOUT_NAME]);
90	l3num = ntohs(nla_get_be16(cda[CTA_TIMEOUT_L3PROTO]));
91	l4num = nla_get_u8(cda[CTA_TIMEOUT_L4PROTO]);
92
93	list_for_each_entry(timeout, &cttimeout_list, head) {
94		if (strncmp(timeout->name, name, CTNL_TIMEOUT_NAME_MAX) != 0)
95			continue;
96
97		if (nlh->nlmsg_flags & NLM_F_EXCL)
98			return -EEXIST;
99
100		matching = timeout;
101		break;
102	}
103
104	l4proto = nf_ct_l4proto_find_get(l3num, l4num);
105
106	/* This protocol is not supportted, skip. */
107	if (l4proto->l4proto != l4num) {
108		ret = -EOPNOTSUPP;
109		goto err_proto_put;
110	}
111
112	if (matching) {
113		if (nlh->nlmsg_flags & NLM_F_REPLACE) {
114			/* You cannot replace one timeout policy by another of
115			 * different kind, sorry.
116			 */
117			if (matching->l3num != l3num ||
118			    matching->l4proto->l4proto != l4num) {
119				ret = -EINVAL;
120				goto err_proto_put;
121			}
122
123			ret = ctnl_timeout_parse_policy(&matching->data,
124							l4proto, net,
125							cda[CTA_TIMEOUT_DATA]);
126			return ret;
127		}
128		ret = -EBUSY;
129		goto err_proto_put;
130	}
131
132	timeout = kzalloc(sizeof(struct ctnl_timeout) +
133			  l4proto->ctnl_timeout.obj_size, GFP_KERNEL);
134	if (timeout == NULL) {
135		ret = -ENOMEM;
136		goto err_proto_put;
137	}
138
139	ret = ctnl_timeout_parse_policy(&timeout->data, l4proto, net,
140					cda[CTA_TIMEOUT_DATA]);
141	if (ret < 0)
142		goto err;
143
144	strcpy(timeout->name, nla_data(cda[CTA_TIMEOUT_NAME]));
145	timeout->l3num = l3num;
146	timeout->l4proto = l4proto;
147	atomic_set(&timeout->refcnt, 1);
148	list_add_tail_rcu(&timeout->head, &cttimeout_list);
149
150	return 0;
151err:
152	kfree(timeout);
153err_proto_put:
154	nf_ct_l4proto_put(l4proto);
155	return ret;
156}
157
158static int
159ctnl_timeout_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
160		       int event, struct ctnl_timeout *timeout)
161{
162	struct nlmsghdr *nlh;
163	struct nfgenmsg *nfmsg;
164	unsigned int flags = portid ? NLM_F_MULTI : 0;
165	struct nf_conntrack_l4proto *l4proto = timeout->l4proto;
166
167	event |= NFNL_SUBSYS_CTNETLINK_TIMEOUT << 8;
168	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
169	if (nlh == NULL)
170		goto nlmsg_failure;
171
172	nfmsg = nlmsg_data(nlh);
173	nfmsg->nfgen_family = AF_UNSPEC;
174	nfmsg->version = NFNETLINK_V0;
175	nfmsg->res_id = 0;
176
177	if (nla_put_string(skb, CTA_TIMEOUT_NAME, timeout->name) ||
178	    nla_put_be16(skb, CTA_TIMEOUT_L3PROTO, htons(timeout->l3num)) ||
179	    nla_put_u8(skb, CTA_TIMEOUT_L4PROTO, timeout->l4proto->l4proto) ||
180	    nla_put_be32(skb, CTA_TIMEOUT_USE,
181			 htonl(atomic_read(&timeout->refcnt))))
182		goto nla_put_failure;
183
184	if (likely(l4proto->ctnl_timeout.obj_to_nlattr)) {
185		struct nlattr *nest_parms;
186		int ret;
187
188		nest_parms = nla_nest_start(skb,
189					    CTA_TIMEOUT_DATA | NLA_F_NESTED);
190		if (!nest_parms)
191			goto nla_put_failure;
192
193		ret = l4proto->ctnl_timeout.obj_to_nlattr(skb, &timeout->data);
194		if (ret < 0)
195			goto nla_put_failure;
196
197		nla_nest_end(skb, nest_parms);
198	}
199
200	nlmsg_end(skb, nlh);
201	return skb->len;
202
203nlmsg_failure:
204nla_put_failure:
205	nlmsg_cancel(skb, nlh);
206	return -1;
207}
208
209static int
210ctnl_timeout_dump(struct sk_buff *skb, struct netlink_callback *cb)
211{
212	struct ctnl_timeout *cur, *last;
213
214	if (cb->args[2])
215		return 0;
216
217	last = (struct ctnl_timeout *)cb->args[1];
218	if (cb->args[1])
219		cb->args[1] = 0;
220
221	rcu_read_lock();
222	list_for_each_entry_rcu(cur, &cttimeout_list, head) {
223		if (last) {
224			if (cur != last)
225				continue;
226
227			last = NULL;
228		}
229		if (ctnl_timeout_fill_info(skb, NETLINK_CB(cb->skb).portid,
230					   cb->nlh->nlmsg_seq,
231					   NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
232					   IPCTNL_MSG_TIMEOUT_NEW, cur) < 0) {
233			cb->args[1] = (unsigned long)cur;
234			break;
235		}
236	}
237	if (!cb->args[1])
238		cb->args[2] = 1;
239	rcu_read_unlock();
240	return skb->len;
241}
242
243static int
244cttimeout_get_timeout(struct sock *ctnl, struct sk_buff *skb,
245		      const struct nlmsghdr *nlh,
246		      const struct nlattr * const cda[])
247{
248	int ret = -ENOENT;
249	char *name;
250	struct ctnl_timeout *cur;
251
252	if (nlh->nlmsg_flags & NLM_F_DUMP) {
253		struct netlink_dump_control c = {
254			.dump = ctnl_timeout_dump,
255		};
256		return netlink_dump_start(ctnl, skb, nlh, &c);
257	}
258
259	if (!cda[CTA_TIMEOUT_NAME])
260		return -EINVAL;
261	name = nla_data(cda[CTA_TIMEOUT_NAME]);
262
263	list_for_each_entry(cur, &cttimeout_list, head) {
264		struct sk_buff *skb2;
265
266		if (strncmp(cur->name, name, CTNL_TIMEOUT_NAME_MAX) != 0)
267			continue;
268
269		skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
270		if (skb2 == NULL) {
271			ret = -ENOMEM;
272			break;
273		}
274
275		ret = ctnl_timeout_fill_info(skb2, NETLINK_CB(skb).portid,
276					     nlh->nlmsg_seq,
277					     NFNL_MSG_TYPE(nlh->nlmsg_type),
278					     IPCTNL_MSG_TIMEOUT_NEW, cur);
279		if (ret <= 0) {
280			kfree_skb(skb2);
281			break;
282		}
283		ret = netlink_unicast(ctnl, skb2, NETLINK_CB(skb).portid,
284					MSG_DONTWAIT);
285		if (ret > 0)
286			ret = 0;
287
288		/* this avoids a loop in nfnetlink. */
289		return ret == -EAGAIN ? -ENOBUFS : ret;
290	}
291	return ret;
292}
293
294/* try to delete object, fail if it is still in use. */
295static int ctnl_timeout_try_del(struct ctnl_timeout *timeout)
296{
297	int ret = 0;
298
299	/* we want to avoid races with nf_ct_timeout_find_get. */
300	if (atomic_dec_and_test(&timeout->refcnt)) {
301		/* We are protected by nfnl mutex. */
302		list_del_rcu(&timeout->head);
303		nf_ct_l4proto_put(timeout->l4proto);
304		kfree_rcu(timeout, rcu_head);
305	} else {
306		/* still in use, restore reference counter. */
307		atomic_inc(&timeout->refcnt);
308		ret = -EBUSY;
309	}
310	return ret;
311}
312
313static int
314cttimeout_del_timeout(struct sock *ctnl, struct sk_buff *skb,
315		      const struct nlmsghdr *nlh,
316		      const struct nlattr * const cda[])
317{
318	char *name;
319	struct ctnl_timeout *cur;
320	int ret = -ENOENT;
321
322	if (!cda[CTA_TIMEOUT_NAME]) {
323		list_for_each_entry(cur, &cttimeout_list, head)
324			ctnl_timeout_try_del(cur);
325
326		return 0;
327	}
328	name = nla_data(cda[CTA_TIMEOUT_NAME]);
329
330	list_for_each_entry(cur, &cttimeout_list, head) {
331		if (strncmp(cur->name, name, CTNL_TIMEOUT_NAME_MAX) != 0)
332			continue;
333
334		ret = ctnl_timeout_try_del(cur);
335		if (ret < 0)
336			return ret;
337
338		break;
339	}
340	return ret;
341}
342
343static int
344cttimeout_default_set(struct sock *ctnl, struct sk_buff *skb,
345		      const struct nlmsghdr *nlh,
346		      const struct nlattr * const cda[])
347{
348	__u16 l3num;
349	__u8 l4num;
350	struct nf_conntrack_l4proto *l4proto;
351	struct net *net = sock_net(skb->sk);
352	unsigned int *timeouts;
353	int ret;
354
355	if (!cda[CTA_TIMEOUT_L3PROTO] ||
356	    !cda[CTA_TIMEOUT_L4PROTO] ||
357	    !cda[CTA_TIMEOUT_DATA])
358		return -EINVAL;
359
360	l3num = ntohs(nla_get_be16(cda[CTA_TIMEOUT_L3PROTO]));
361	l4num = nla_get_u8(cda[CTA_TIMEOUT_L4PROTO]);
362	l4proto = nf_ct_l4proto_find_get(l3num, l4num);
363
364	/* This protocol is not supported, skip. */
365	if (l4proto->l4proto != l4num) {
366		ret = -EOPNOTSUPP;
367		goto err;
368	}
369
370	timeouts = l4proto->get_timeouts(net);
371
372	ret = ctnl_timeout_parse_policy(timeouts, l4proto, net,
373					cda[CTA_TIMEOUT_DATA]);
374	if (ret < 0)
375		goto err;
376
377	nf_ct_l4proto_put(l4proto);
378	return 0;
379err:
380	nf_ct_l4proto_put(l4proto);
381	return ret;
382}
383
384static int
385cttimeout_default_fill_info(struct net *net, struct sk_buff *skb, u32 portid,
386			    u32 seq, u32 type, int event,
387			    struct nf_conntrack_l4proto *l4proto)
388{
389	struct nlmsghdr *nlh;
390	struct nfgenmsg *nfmsg;
391	unsigned int flags = portid ? NLM_F_MULTI : 0;
392
393	event |= NFNL_SUBSYS_CTNETLINK_TIMEOUT << 8;
394	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
395	if (nlh == NULL)
396		goto nlmsg_failure;
397
398	nfmsg = nlmsg_data(nlh);
399	nfmsg->nfgen_family = AF_UNSPEC;
400	nfmsg->version = NFNETLINK_V0;
401	nfmsg->res_id = 0;
402
403	if (nla_put_be16(skb, CTA_TIMEOUT_L3PROTO, htons(l4proto->l3proto)) ||
404	    nla_put_u8(skb, CTA_TIMEOUT_L4PROTO, l4proto->l4proto))
405		goto nla_put_failure;
406
407	if (likely(l4proto->ctnl_timeout.obj_to_nlattr)) {
408		struct nlattr *nest_parms;
409		unsigned int *timeouts = l4proto->get_timeouts(net);
410		int ret;
411
412		nest_parms = nla_nest_start(skb,
413					    CTA_TIMEOUT_DATA | NLA_F_NESTED);
414		if (!nest_parms)
415			goto nla_put_failure;
416
417		ret = l4proto->ctnl_timeout.obj_to_nlattr(skb, timeouts);
418		if (ret < 0)
419			goto nla_put_failure;
420
421		nla_nest_end(skb, nest_parms);
422	}
423
424	nlmsg_end(skb, nlh);
425	return skb->len;
426
427nlmsg_failure:
428nla_put_failure:
429	nlmsg_cancel(skb, nlh);
430	return -1;
431}
432
433static int cttimeout_default_get(struct sock *ctnl, struct sk_buff *skb,
434				 const struct nlmsghdr *nlh,
435				 const struct nlattr * const cda[])
436{
437	__u16 l3num;
438	__u8 l4num;
439	struct nf_conntrack_l4proto *l4proto;
440	struct net *net = sock_net(skb->sk);
441	struct sk_buff *skb2;
442	int ret, err;
443
444	if (!cda[CTA_TIMEOUT_L3PROTO] || !cda[CTA_TIMEOUT_L4PROTO])
445		return -EINVAL;
446
447	l3num = ntohs(nla_get_be16(cda[CTA_TIMEOUT_L3PROTO]));
448	l4num = nla_get_u8(cda[CTA_TIMEOUT_L4PROTO]);
449	l4proto = nf_ct_l4proto_find_get(l3num, l4num);
450
451	/* This protocol is not supported, skip. */
452	if (l4proto->l4proto != l4num) {
453		err = -EOPNOTSUPP;
454		goto err;
455	}
456
457	skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
458	if (skb2 == NULL) {
459		err = -ENOMEM;
460		goto err;
461	}
462
463	ret = cttimeout_default_fill_info(net, skb2, NETLINK_CB(skb).portid,
464					  nlh->nlmsg_seq,
465					  NFNL_MSG_TYPE(nlh->nlmsg_type),
466					  IPCTNL_MSG_TIMEOUT_DEFAULT_SET,
467					  l4proto);
468	if (ret <= 0) {
469		kfree_skb(skb2);
470		err = -ENOMEM;
471		goto err;
472	}
473	ret = netlink_unicast(ctnl, skb2, NETLINK_CB(skb).portid, MSG_DONTWAIT);
474	if (ret > 0)
475		ret = 0;
476
477	/* this avoids a loop in nfnetlink. */
478	return ret == -EAGAIN ? -ENOBUFS : ret;
479err:
480	nf_ct_l4proto_put(l4proto);
481	return err;
482}
483
484#ifdef CONFIG_NF_CONNTRACK_TIMEOUT
485static struct ctnl_timeout *ctnl_timeout_find_get(const char *name)
486{
487	struct ctnl_timeout *timeout, *matching = NULL;
488
489	rcu_read_lock();
490	list_for_each_entry_rcu(timeout, &cttimeout_list, head) {
491		if (strncmp(timeout->name, name, CTNL_TIMEOUT_NAME_MAX) != 0)
492			continue;
493
494		if (!try_module_get(THIS_MODULE))
495			goto err;
496
497		if (!atomic_inc_not_zero(&timeout->refcnt)) {
498			module_put(THIS_MODULE);
499			goto err;
500		}
501		matching = timeout;
502		break;
503	}
504err:
505	rcu_read_unlock();
506	return matching;
507}
508
509static void ctnl_timeout_put(struct ctnl_timeout *timeout)
510{
511	atomic_dec(&timeout->refcnt);
512	module_put(THIS_MODULE);
513}
514#endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
515
516static const struct nfnl_callback cttimeout_cb[IPCTNL_MSG_TIMEOUT_MAX] = {
517	[IPCTNL_MSG_TIMEOUT_NEW]	= { .call = cttimeout_new_timeout,
518					    .attr_count = CTA_TIMEOUT_MAX,
519					    .policy = cttimeout_nla_policy },
520	[IPCTNL_MSG_TIMEOUT_GET]	= { .call = cttimeout_get_timeout,
521					    .attr_count = CTA_TIMEOUT_MAX,
522					    .policy = cttimeout_nla_policy },
523	[IPCTNL_MSG_TIMEOUT_DELETE]	= { .call = cttimeout_del_timeout,
524					    .attr_count = CTA_TIMEOUT_MAX,
525					    .policy = cttimeout_nla_policy },
526	[IPCTNL_MSG_TIMEOUT_DEFAULT_SET]= { .call = cttimeout_default_set,
527					    .attr_count = CTA_TIMEOUT_MAX,
528					    .policy = cttimeout_nla_policy },
529	[IPCTNL_MSG_TIMEOUT_DEFAULT_GET]= { .call = cttimeout_default_get,
530					    .attr_count = CTA_TIMEOUT_MAX,
531					    .policy = cttimeout_nla_policy },
532};
533
534static const struct nfnetlink_subsystem cttimeout_subsys = {
535	.name				= "conntrack_timeout",
536	.subsys_id			= NFNL_SUBSYS_CTNETLINK_TIMEOUT,
537	.cb_count			= IPCTNL_MSG_TIMEOUT_MAX,
538	.cb				= cttimeout_cb,
539};
540
541MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_CTNETLINK_TIMEOUT);
542
543static int __init cttimeout_init(void)
544{
545	int ret;
546
547	ret = nfnetlink_subsys_register(&cttimeout_subsys);
548	if (ret < 0) {
549		pr_err("cttimeout_init: cannot register cttimeout with "
550			"nfnetlink.\n");
551		goto err_out;
552	}
553#ifdef CONFIG_NF_CONNTRACK_TIMEOUT
554	RCU_INIT_POINTER(nf_ct_timeout_find_get_hook, ctnl_timeout_find_get);
555	RCU_INIT_POINTER(nf_ct_timeout_put_hook, ctnl_timeout_put);
556#endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
557	return 0;
558
559err_out:
560	return ret;
561}
562
563static void __exit cttimeout_exit(void)
564{
565	struct ctnl_timeout *cur, *tmp;
566
567	pr_info("cttimeout: unregistering from nfnetlink.\n");
568
569	nfnetlink_subsys_unregister(&cttimeout_subsys);
570	list_for_each_entry_safe(cur, tmp, &cttimeout_list, head) {
571		list_del_rcu(&cur->head);
572		/* We are sure that our objects have no clients at this point,
573		 * it's safe to release them all without checking refcnt.
574		 */
575		nf_ct_l4proto_put(cur->l4proto);
576		kfree_rcu(cur, rcu_head);
577	}
578#ifdef CONFIG_NF_CONNTRACK_TIMEOUT
579	RCU_INIT_POINTER(nf_ct_timeout_find_get_hook, NULL);
580	RCU_INIT_POINTER(nf_ct_timeout_put_hook, NULL);
581#endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
582}
583
584module_init(cttimeout_init);
585module_exit(cttimeout_exit);
586