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
67bool 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
272enum {
273	NFNL_BATCH_FAILURE	= (1 << 0),
274	NFNL_BATCH_DONE		= (1 << 1),
275	NFNL_BATCH_REPLAY	= (1 << 2),
276};
277
278static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
279				u_int16_t subsys_id)
280{
281	struct sk_buff *oskb = skb;
282	struct net *net = sock_net(skb->sk);
283	const struct nfnetlink_subsystem *ss;
284	const struct nfnl_callback *nc;
285	static LIST_HEAD(err_list);
286	u32 status;
287	int err;
288
289	if (subsys_id >= NFNL_SUBSYS_COUNT)
290		return netlink_ack(skb, nlh, -EINVAL);
291replay:
292	status = 0;
293
294	skb = netlink_skb_clone(oskb, GFP_KERNEL);
295	if (!skb)
296		return netlink_ack(oskb, nlh, -ENOMEM);
297
298	nfnl_lock(subsys_id);
299	ss = rcu_dereference_protected(table[subsys_id].subsys,
300				       lockdep_is_held(&table[subsys_id].mutex));
301	if (!ss) {
302#ifdef CONFIG_MODULES
303		nfnl_unlock(subsys_id);
304		request_module("nfnetlink-subsys-%d", subsys_id);
305		nfnl_lock(subsys_id);
306		ss = rcu_dereference_protected(table[subsys_id].subsys,
307					       lockdep_is_held(&table[subsys_id].mutex));
308		if (!ss)
309#endif
310		{
311			nfnl_unlock(subsys_id);
312			netlink_ack(skb, nlh, -EOPNOTSUPP);
313			return kfree_skb(skb);
314		}
315	}
316
317	if (!ss->commit || !ss->abort) {
318		nfnl_unlock(subsys_id);
319		netlink_ack(skb, nlh, -EOPNOTSUPP);
320		return kfree_skb(skb);
321	}
322
323	while (skb->len >= nlmsg_total_size(0)) {
324		int msglen, type;
325
326		nlh = nlmsg_hdr(skb);
327		err = 0;
328
329		if (nlmsg_len(nlh) < sizeof(struct nfgenmsg) ||
330		    skb->len < nlh->nlmsg_len) {
331			err = -EINVAL;
332			goto ack;
333		}
334
335		/* Only requests are handled by the kernel */
336		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
337			err = -EINVAL;
338			goto ack;
339		}
340
341		type = nlh->nlmsg_type;
342		if (type == NFNL_MSG_BATCH_BEGIN) {
343			/* Malformed: Batch begin twice */
344			nfnl_err_reset(&err_list);
345			status |= NFNL_BATCH_FAILURE;
346			goto done;
347		} else if (type == NFNL_MSG_BATCH_END) {
348			status |= NFNL_BATCH_DONE;
349			goto done;
350		} else if (type < NLMSG_MIN_TYPE) {
351			err = -EINVAL;
352			goto ack;
353		}
354
355		/* We only accept a batch with messages for the same
356		 * subsystem.
357		 */
358		if (NFNL_SUBSYS_ID(type) != subsys_id) {
359			err = -EINVAL;
360			goto ack;
361		}
362
363		nc = nfnetlink_find_client(type, ss);
364		if (!nc) {
365			err = -EINVAL;
366			goto ack;
367		}
368
369		{
370			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
371			u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
372			struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
373			struct nlattr *attr = (void *)nlh + min_len;
374			int attrlen = nlh->nlmsg_len - min_len;
375
376			err = nla_parse(cda, ss->cb[cb_id].attr_count,
377					attr, attrlen, ss->cb[cb_id].policy);
378			if (err < 0)
379				goto ack;
380
381			if (nc->call_batch) {
382				err = nc->call_batch(net, net->nfnl, skb, nlh,
383						     (const struct nlattr **)cda);
384			}
385
386			/* The lock was released to autoload some module, we
387			 * have to abort and start from scratch using the
388			 * original skb.
389			 */
390			if (err == -EAGAIN) {
391				status |= NFNL_BATCH_REPLAY;
392				goto next;
393			}
394		}
395ack:
396		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
397			/* Errors are delivered once the full batch has been
398			 * processed, this avoids that the same error is
399			 * reported several times when replaying the batch.
400			 */
401			if (nfnl_err_add(&err_list, nlh, err) < 0) {
402				/* We failed to enqueue an error, reset the
403				 * list of errors and send OOM to userspace
404				 * pointing to the batch header.
405				 */
406				nfnl_err_reset(&err_list);
407				netlink_ack(skb, nlmsg_hdr(oskb), -ENOMEM);
408				status |= NFNL_BATCH_FAILURE;
409				goto done;
410			}
411			/* We don't stop processing the batch on errors, thus,
412			 * userspace gets all the errors that the batch
413			 * triggers.
414			 */
415			if (err)
416				status |= NFNL_BATCH_FAILURE;
417		}
418next:
419		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
420		if (msglen > skb->len)
421			msglen = skb->len;
422		skb_pull(skb, msglen);
423	}
424done:
425	if (status & NFNL_BATCH_REPLAY) {
426		ss->abort(oskb);
427		nfnl_err_reset(&err_list);
428		nfnl_unlock(subsys_id);
429		kfree_skb(skb);
430		goto replay;
431	} else if (status == NFNL_BATCH_DONE) {
432		ss->commit(oskb);
433	} else {
434		ss->abort(oskb);
435	}
436
437	nfnl_err_deliver(&err_list, oskb);
438	nfnl_unlock(subsys_id);
439	kfree_skb(skb);
440}
441
442static void nfnetlink_rcv(struct sk_buff *skb)
443{
444	struct nlmsghdr *nlh = nlmsg_hdr(skb);
445	u_int16_t res_id;
446	int msglen;
447
448	if (nlh->nlmsg_len < NLMSG_HDRLEN ||
449	    skb->len < nlh->nlmsg_len)
450		return;
451
452	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
453		netlink_ack(skb, nlh, -EPERM);
454		return;
455	}
456
457	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
458		struct nfgenmsg *nfgenmsg;
459
460		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
461		if (msglen > skb->len)
462			msglen = skb->len;
463
464		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
465		    skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
466			return;
467
468		nfgenmsg = nlmsg_data(nlh);
469		skb_pull(skb, msglen);
470		/* Work around old nft using host byte order */
471		if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
472			res_id = NFNL_SUBSYS_NFTABLES;
473		else
474			res_id = ntohs(nfgenmsg->res_id);
475		nfnetlink_rcv_batch(skb, nlh, res_id);
476	} else {
477		netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
478	}
479}
480
481#ifdef CONFIG_MODULES
482static int nfnetlink_bind(struct net *net, int group)
483{
484	const struct nfnetlink_subsystem *ss;
485	int type;
486
487	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
488		return 0;
489
490	type = nfnl_group2type[group];
491
492	rcu_read_lock();
493	ss = nfnetlink_get_subsys(type << 8);
494	rcu_read_unlock();
495	if (!ss)
496		request_module("nfnetlink-subsys-%d", type);
497	return 0;
498}
499#endif
500
501static int __net_init nfnetlink_net_init(struct net *net)
502{
503	struct sock *nfnl;
504	struct netlink_kernel_cfg cfg = {
505		.groups	= NFNLGRP_MAX,
506		.input	= nfnetlink_rcv,
507#ifdef CONFIG_MODULES
508		.bind	= nfnetlink_bind,
509#endif
510	};
511
512	nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
513	if (!nfnl)
514		return -ENOMEM;
515	net->nfnl_stash = nfnl;
516	rcu_assign_pointer(net->nfnl, nfnl);
517	return 0;
518}
519
520static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
521{
522	struct net *net;
523
524	list_for_each_entry(net, net_exit_list, exit_list)
525		RCU_INIT_POINTER(net->nfnl, NULL);
526	synchronize_net();
527	list_for_each_entry(net, net_exit_list, exit_list)
528		netlink_kernel_release(net->nfnl_stash);
529}
530
531static struct pernet_operations nfnetlink_net_ops = {
532	.init		= nfnetlink_net_init,
533	.exit_batch	= nfnetlink_net_exit_batch,
534};
535
536static int __init nfnetlink_init(void)
537{
538	int i;
539
540	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
541		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
542
543	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
544		mutex_init(&table[i].mutex);
545
546	pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
547	return register_pernet_subsys(&nfnetlink_net_ops);
548}
549
550static void __exit nfnetlink_exit(void)
551{
552	pr_info("Removing netfilter NETLINK layer.\n");
553	unregister_pernet_subsys(&nfnetlink_net_ops);
554}
555module_init(nfnetlink_init);
556module_exit(nfnetlink_exit);
557