1/* Netfilter messages via netlink socket. Allows for user space
2 * protocol helpers and general trouble making from userspace.
3 *
4 * (C) 2001 by Jay Schulist <jschlst@samba.org>,
5 * (C) 2002-2005 by Harald Welte <laforge@gnumonks.org>
6 * (C) 2005,2007 by Pablo Neira Ayuso <pablo@netfilter.org>
7 *
8 * Initial netfilter messages via netlink development funded and
9 * generally made possible by Network Robots, Inc. (www.networkrobots.com)
10 *
11 * Further development of this code funded by Astaro AG (http://www.astaro.com)
12 *
13 * This software may be used and distributed according to the terms
14 * of the GNU General Public License, incorporated herein by reference.
15 */
16
17#include <linux/module.h>
18#include <linux/types.h>
19#include <linux/socket.h>
20#include <linux/kernel.h>
21#include <linux/string.h>
22#include <linux/sockios.h>
23#include <linux/net.h>
24#include <linux/skbuff.h>
25#include <asm/uaccess.h>
26#include <net/sock.h>
27#include <linux/init.h>
28
29#include <net/netlink.h>
30#include <linux/netfilter/nfnetlink.h>
31
32MODULE_LICENSE("GPL");
33MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
34MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
35
36static char __initdata nfversion[] = "0.30";
37
38static struct {
39	struct mutex				mutex;
40	const struct nfnetlink_subsystem __rcu	*subsys;
41} table[NFNL_SUBSYS_COUNT];
42
43static const int nfnl_group2type[NFNLGRP_MAX+1] = {
44	[NFNLGRP_CONNTRACK_NEW]		= NFNL_SUBSYS_CTNETLINK,
45	[NFNLGRP_CONNTRACK_UPDATE]	= NFNL_SUBSYS_CTNETLINK,
46	[NFNLGRP_CONNTRACK_DESTROY]	= NFNL_SUBSYS_CTNETLINK,
47	[NFNLGRP_CONNTRACK_EXP_NEW]	= NFNL_SUBSYS_CTNETLINK_EXP,
48	[NFNLGRP_CONNTRACK_EXP_UPDATE]	= NFNL_SUBSYS_CTNETLINK_EXP,
49	[NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
50	[NFNLGRP_NFTABLES]		= NFNL_SUBSYS_NFTABLES,
51	[NFNLGRP_ACCT_QUOTA]		= NFNL_SUBSYS_ACCT,
52};
53
54void nfnl_lock(__u8 subsys_id)
55{
56	mutex_lock(&table[subsys_id].mutex);
57}
58EXPORT_SYMBOL_GPL(nfnl_lock);
59
60void nfnl_unlock(__u8 subsys_id)
61{
62	mutex_unlock(&table[subsys_id].mutex);
63}
64EXPORT_SYMBOL_GPL(nfnl_unlock);
65
66#ifdef CONFIG_PROVE_LOCKING
67int lockdep_nfnl_is_held(u8 subsys_id)
68{
69	return lockdep_is_held(&table[subsys_id].mutex);
70}
71EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
72#endif
73
74int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
75{
76	nfnl_lock(n->subsys_id);
77	if (table[n->subsys_id].subsys) {
78		nfnl_unlock(n->subsys_id);
79		return -EBUSY;
80	}
81	rcu_assign_pointer(table[n->subsys_id].subsys, n);
82	nfnl_unlock(n->subsys_id);
83
84	return 0;
85}
86EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
87
88int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
89{
90	nfnl_lock(n->subsys_id);
91	table[n->subsys_id].subsys = NULL;
92	nfnl_unlock(n->subsys_id);
93	synchronize_rcu();
94	return 0;
95}
96EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
97
98static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u_int16_t type)
99{
100	u_int8_t subsys_id = NFNL_SUBSYS_ID(type);
101
102	if (subsys_id >= NFNL_SUBSYS_COUNT)
103		return NULL;
104
105	return rcu_dereference(table[subsys_id].subsys);
106}
107
108static inline const struct nfnl_callback *
109nfnetlink_find_client(u_int16_t type, const struct nfnetlink_subsystem *ss)
110{
111	u_int8_t cb_id = NFNL_MSG_TYPE(type);
112
113	if (cb_id >= ss->cb_count)
114		return NULL;
115
116	return &ss->cb[cb_id];
117}
118
119int nfnetlink_has_listeners(struct net *net, unsigned int group)
120{
121	return netlink_has_listeners(net->nfnl, group);
122}
123EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
124
125struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
126				    u32 dst_portid, gfp_t gfp_mask)
127{
128	return netlink_alloc_skb(net->nfnl, size, dst_portid, gfp_mask);
129}
130EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb);
131
132int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
133		   unsigned int group, int echo, gfp_t flags)
134{
135	return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
136}
137EXPORT_SYMBOL_GPL(nfnetlink_send);
138
139int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
140{
141	return netlink_set_err(net->nfnl, portid, group, error);
142}
143EXPORT_SYMBOL_GPL(nfnetlink_set_err);
144
145int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
146		      int flags)
147{
148	return netlink_unicast(net->nfnl, skb, portid, flags);
149}
150EXPORT_SYMBOL_GPL(nfnetlink_unicast);
151
152/* Process one complete nfnetlink message. */
153static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
154{
155	struct net *net = sock_net(skb->sk);
156	const struct nfnl_callback *nc;
157	const struct nfnetlink_subsystem *ss;
158	int type, err;
159
160	/* All the messages must at least contain nfgenmsg */
161	if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
162		return 0;
163
164	type = nlh->nlmsg_type;
165replay:
166	rcu_read_lock();
167	ss = nfnetlink_get_subsys(type);
168	if (!ss) {
169#ifdef CONFIG_MODULES
170		rcu_read_unlock();
171		request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
172		rcu_read_lock();
173		ss = nfnetlink_get_subsys(type);
174		if (!ss)
175#endif
176		{
177			rcu_read_unlock();
178			return -EINVAL;
179		}
180	}
181
182	nc = nfnetlink_find_client(type, ss);
183	if (!nc) {
184		rcu_read_unlock();
185		return -EINVAL;
186	}
187
188	{
189		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
190		u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
191		struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
192		struct nlattr *attr = (void *)nlh + min_len;
193		int attrlen = nlh->nlmsg_len - min_len;
194		__u8 subsys_id = NFNL_SUBSYS_ID(type);
195
196		err = nla_parse(cda, ss->cb[cb_id].attr_count,
197				attr, attrlen, ss->cb[cb_id].policy);
198		if (err < 0) {
199			rcu_read_unlock();
200			return err;
201		}
202
203		if (nc->call_rcu) {
204			err = nc->call_rcu(net->nfnl, skb, nlh,
205					   (const struct nlattr **)cda);
206			rcu_read_unlock();
207		} else {
208			rcu_read_unlock();
209			nfnl_lock(subsys_id);
210			if (rcu_dereference_protected(table[subsys_id].subsys,
211				lockdep_is_held(&table[subsys_id].mutex)) != ss ||
212			    nfnetlink_find_client(type, ss) != nc)
213				err = -EAGAIN;
214			else if (nc->call)
215				err = nc->call(net->nfnl, skb, nlh,
216						   (const struct nlattr **)cda);
217			else
218				err = -EINVAL;
219			nfnl_unlock(subsys_id);
220		}
221		if (err == -EAGAIN)
222			goto replay;
223		return err;
224	}
225}
226
227struct nfnl_err {
228	struct list_head	head;
229	struct nlmsghdr		*nlh;
230	int			err;
231};
232
233static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err)
234{
235	struct nfnl_err *nfnl_err;
236
237	nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
238	if (nfnl_err == NULL)
239		return -ENOMEM;
240
241	nfnl_err->nlh = nlh;
242	nfnl_err->err = err;
243	list_add_tail(&nfnl_err->head, list);
244
245	return 0;
246}
247
248static void nfnl_err_del(struct nfnl_err *nfnl_err)
249{
250	list_del(&nfnl_err->head);
251	kfree(nfnl_err);
252}
253
254static void nfnl_err_reset(struct list_head *err_list)
255{
256	struct nfnl_err *nfnl_err, *next;
257
258	list_for_each_entry_safe(nfnl_err, next, err_list, head)
259		nfnl_err_del(nfnl_err);
260}
261
262static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
263{
264	struct nfnl_err *nfnl_err, *next;
265
266	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
267		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err);
268		nfnl_err_del(nfnl_err);
269	}
270}
271
272static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
273				u_int16_t subsys_id)
274{
275	struct sk_buff *oskb = skb;
276	struct net *net = sock_net(skb->sk);
277	const struct nfnetlink_subsystem *ss;
278	const struct nfnl_callback *nc;
279	bool success = true, done = false;
280	static LIST_HEAD(err_list);
281	int err;
282
283	if (subsys_id >= NFNL_SUBSYS_COUNT)
284		return netlink_ack(skb, nlh, -EINVAL);
285replay:
286	skb = netlink_skb_clone(oskb, GFP_KERNEL);
287	if (!skb)
288		return netlink_ack(oskb, nlh, -ENOMEM);
289
290	skb->sk = oskb->sk;
291
292	nfnl_lock(subsys_id);
293	ss = rcu_dereference_protected(table[subsys_id].subsys,
294				       lockdep_is_held(&table[subsys_id].mutex));
295	if (!ss) {
296#ifdef CONFIG_MODULES
297		nfnl_unlock(subsys_id);
298		request_module("nfnetlink-subsys-%d", subsys_id);
299		nfnl_lock(subsys_id);
300		ss = rcu_dereference_protected(table[subsys_id].subsys,
301					       lockdep_is_held(&table[subsys_id].mutex));
302		if (!ss)
303#endif
304		{
305			nfnl_unlock(subsys_id);
306			netlink_ack(skb, nlh, -EOPNOTSUPP);
307			return kfree_skb(skb);
308		}
309	}
310
311	if (!ss->commit || !ss->abort) {
312		nfnl_unlock(subsys_id);
313		netlink_ack(skb, nlh, -EOPNOTSUPP);
314		return kfree_skb(skb);
315	}
316
317	while (skb->len >= nlmsg_total_size(0)) {
318		int msglen, type;
319
320		nlh = nlmsg_hdr(skb);
321		err = 0;
322
323		if (nlmsg_len(nlh) < sizeof(struct nfgenmsg) ||
324		    skb->len < nlh->nlmsg_len) {
325			err = -EINVAL;
326			goto ack;
327		}
328
329		/* Only requests are handled by the kernel */
330		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
331			err = -EINVAL;
332			goto ack;
333		}
334
335		type = nlh->nlmsg_type;
336		if (type == NFNL_MSG_BATCH_BEGIN) {
337			/* Malformed: Batch begin twice */
338			nfnl_err_reset(&err_list);
339			success = false;
340			goto done;
341		} else if (type == NFNL_MSG_BATCH_END) {
342			done = true;
343			goto done;
344		} else if (type < NLMSG_MIN_TYPE) {
345			err = -EINVAL;
346			goto ack;
347		}
348
349		/* We only accept a batch with messages for the same
350		 * subsystem.
351		 */
352		if (NFNL_SUBSYS_ID(type) != subsys_id) {
353			err = -EINVAL;
354			goto ack;
355		}
356
357		nc = nfnetlink_find_client(type, ss);
358		if (!nc) {
359			err = -EINVAL;
360			goto ack;
361		}
362
363		{
364			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
365			u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
366			struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
367			struct nlattr *attr = (void *)nlh + min_len;
368			int attrlen = nlh->nlmsg_len - min_len;
369
370			err = nla_parse(cda, ss->cb[cb_id].attr_count,
371					attr, attrlen, ss->cb[cb_id].policy);
372			if (err < 0)
373				goto ack;
374
375			if (nc->call_batch) {
376				err = nc->call_batch(net->nfnl, skb, nlh,
377						     (const struct nlattr **)cda);
378			}
379
380			/* The lock was released to autoload some module, we
381			 * have to abort and start from scratch using the
382			 * original skb.
383			 */
384			if (err == -EAGAIN) {
385				nfnl_err_reset(&err_list);
386				ss->abort(oskb);
387				nfnl_unlock(subsys_id);
388				kfree_skb(skb);
389				goto replay;
390			}
391		}
392ack:
393		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
394			/* Errors are delivered once the full batch has been
395			 * processed, this avoids that the same error is
396			 * reported several times when replaying the batch.
397			 */
398			if (nfnl_err_add(&err_list, nlh, err) < 0) {
399				/* We failed to enqueue an error, reset the
400				 * list of errors and send OOM to userspace
401				 * pointing to the batch header.
402				 */
403				nfnl_err_reset(&err_list);
404				netlink_ack(skb, nlmsg_hdr(oskb), -ENOMEM);
405				success = false;
406				goto done;
407			}
408			/* We don't stop processing the batch on errors, thus,
409			 * userspace gets all the errors that the batch
410			 * triggers.
411			 */
412			if (err)
413				success = false;
414		}
415
416		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
417		if (msglen > skb->len)
418			msglen = skb->len;
419		skb_pull(skb, msglen);
420	}
421done:
422	if (success && done)
423		ss->commit(oskb);
424	else
425		ss->abort(oskb);
426
427	nfnl_err_deliver(&err_list, oskb);
428	nfnl_unlock(subsys_id);
429	kfree_skb(skb);
430}
431
432static void nfnetlink_rcv(struct sk_buff *skb)
433{
434	struct nlmsghdr *nlh = nlmsg_hdr(skb);
435	u_int16_t res_id;
436	int msglen;
437
438	if (nlh->nlmsg_len < NLMSG_HDRLEN ||
439	    skb->len < nlh->nlmsg_len)
440		return;
441
442	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
443		netlink_ack(skb, nlh, -EPERM);
444		return;
445	}
446
447	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
448		struct nfgenmsg *nfgenmsg;
449
450		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
451		if (msglen > skb->len)
452			msglen = skb->len;
453
454		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
455		    skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
456			return;
457
458		nfgenmsg = nlmsg_data(nlh);
459		skb_pull(skb, msglen);
460		/* Work around old nft using host byte order */
461		if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
462			res_id = NFNL_SUBSYS_NFTABLES;
463		else
464			res_id = ntohs(nfgenmsg->res_id);
465		nfnetlink_rcv_batch(skb, nlh, res_id);
466	} else {
467		netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
468	}
469}
470
471#ifdef CONFIG_MODULES
472static int nfnetlink_bind(struct net *net, int group)
473{
474	const struct nfnetlink_subsystem *ss;
475	int type;
476
477	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
478		return 0;
479
480	type = nfnl_group2type[group];
481
482	rcu_read_lock();
483	ss = nfnetlink_get_subsys(type);
484	rcu_read_unlock();
485	if (!ss)
486		request_module("nfnetlink-subsys-%d", type);
487	return 0;
488}
489#endif
490
491static int __net_init nfnetlink_net_init(struct net *net)
492{
493	struct sock *nfnl;
494	struct netlink_kernel_cfg cfg = {
495		.groups	= NFNLGRP_MAX,
496		.input	= nfnetlink_rcv,
497#ifdef CONFIG_MODULES
498		.bind	= nfnetlink_bind,
499#endif
500	};
501
502	nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
503	if (!nfnl)
504		return -ENOMEM;
505	net->nfnl_stash = nfnl;
506	rcu_assign_pointer(net->nfnl, nfnl);
507	return 0;
508}
509
510static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
511{
512	struct net *net;
513
514	list_for_each_entry(net, net_exit_list, exit_list)
515		RCU_INIT_POINTER(net->nfnl, NULL);
516	synchronize_net();
517	list_for_each_entry(net, net_exit_list, exit_list)
518		netlink_kernel_release(net->nfnl_stash);
519}
520
521static struct pernet_operations nfnetlink_net_ops = {
522	.init		= nfnetlink_net_init,
523	.exit_batch	= nfnetlink_net_exit_batch,
524};
525
526static int __init nfnetlink_init(void)
527{
528	int i;
529
530	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
531		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
532
533	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
534		mutex_init(&table[i].mutex);
535
536	pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
537	return register_pernet_subsys(&nfnetlink_net_ops);
538}
539
540static void __exit nfnetlink_exit(void)
541{
542	pr_info("Removing netfilter NETLINK layer.\n");
543	unregister_pernet_subsys(&nfnetlink_net_ops);
544}
545module_init(nfnetlink_init);
546module_exit(nfnetlink_exit);
547