nfnetlink.c

  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) 2005-2017 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 <linux/uaccess.h>
 26#include <net/sock.h>
 27#include <linux/init.h>
 28#include <linux/sched/signal.h>
 29
 30#include <net/netlink.h>
 31#include <net/netns/generic.h>
 32#include <linux/netfilter.h>
 33#include <linux/netfilter/nfnetlink.h>
 34
 35MODULE_LICENSE("GPL");
 36MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
 37MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
 38MODULE_DESCRIPTION("Netfilter messages via netlink socket");
 39
 40#define nfnl_dereference_protected(id) \
 41	rcu_dereference_protected(table[(id)].subsys, \
 42				  lockdep_nfnl_is_held((id)))
 43
 44#define NFNL_MAX_ATTR_COUNT	32
 45
 46static unsigned int nfnetlink_pernet_id __read_mostly;
 47
 48#ifdef CONFIG_NF_CONNTRACK_EVENTS
 49static DEFINE_SPINLOCK(nfnl_grp_active_lock);
 50#endif
 51
 52struct nfnl_net {
 53	struct sock *nfnl;
 54};
 55
 56static struct {
 57	struct mutex				mutex;
 58	const struct nfnetlink_subsystem __rcu	*subsys;
 59} table[NFNL_SUBSYS_COUNT];
 60
 61static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
 62
 63static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = {
 64	[NFNL_SUBSYS_NONE] = "nfnl_subsys_none",
 65	[NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink",
 66	[NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp",
 67	[NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue",
 68	[NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog",
 69	[NFNL_SUBSYS_OSF] = "nfnl_subsys_osf",
 70	[NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset",
 71	[NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct",
 72	[NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout",
 73	[NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper",
 74	[NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables",
 75	[NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat",
 76	[NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook",
 77};
 78
 79static const int nfnl_group2type[NFNLGRP_MAX+1] = {
 80	[NFNLGRP_CONNTRACK_NEW]		= NFNL_SUBSYS_CTNETLINK,
 81	[NFNLGRP_CONNTRACK_UPDATE]	= NFNL_SUBSYS_CTNETLINK,
 82	[NFNLGRP_CONNTRACK_DESTROY]	= NFNL_SUBSYS_CTNETLINK,
 83	[NFNLGRP_CONNTRACK_EXP_NEW]	= NFNL_SUBSYS_CTNETLINK_EXP,
 84	[NFNLGRP_CONNTRACK_EXP_UPDATE]	= NFNL_SUBSYS_CTNETLINK_EXP,
 85	[NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
 86	[NFNLGRP_NFTABLES]		= NFNL_SUBSYS_NFTABLES,
 87	[NFNLGRP_ACCT_QUOTA]		= NFNL_SUBSYS_ACCT,
 88	[NFNLGRP_NFTRACE]		= NFNL_SUBSYS_NFTABLES,
 89};
 90
 91static struct nfnl_net *nfnl_pernet(struct net *net)
 92{
 93	return net_generic(net, nfnetlink_pernet_id);
 94}
 95
 96void nfnl_lock(__u8 subsys_id)
 97{
 98	mutex_lock(&table[subsys_id].mutex);
 99}
100EXPORT_SYMBOL_GPL(nfnl_lock);
101
102void nfnl_unlock(__u8 subsys_id)
103{
104	mutex_unlock(&table[subsys_id].mutex);
105}
106EXPORT_SYMBOL_GPL(nfnl_unlock);
107
108#ifdef CONFIG_PROVE_LOCKING
109bool lockdep_nfnl_is_held(u8 subsys_id)
110{
111	return lockdep_is_held(&table[subsys_id].mutex);
112}
113EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
114#endif
115
116int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
117{
118	u8 cb_id;
119
120	/* Sanity-check attr_count size to avoid stack buffer overflow. */
121	for (cb_id = 0; cb_id < n->cb_count; cb_id++)
122		if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
123			return -EINVAL;
124
125	nfnl_lock(n->subsys_id);
126	if (table[n->subsys_id].subsys) {
127		nfnl_unlock(n->subsys_id);
128		return -EBUSY;
129	}
130	rcu_assign_pointer(table[n->subsys_id].subsys, n);
131	nfnl_unlock(n->subsys_id);
132
133	return 0;
134}
135EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
136
137int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
138{
139	nfnl_lock(n->subsys_id);
140	table[n->subsys_id].subsys = NULL;
141	nfnl_unlock(n->subsys_id);
142	synchronize_rcu();
143	return 0;
144}
145EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
146
147static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
148{
149	u8 subsys_id = NFNL_SUBSYS_ID(type);
150
151	if (subsys_id >= NFNL_SUBSYS_COUNT)
152		return NULL;
153
154	return rcu_dereference(table[subsys_id].subsys);
155}
156
157static inline const struct nfnl_callback *
158nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
159{
160	u8 cb_id = NFNL_MSG_TYPE(type);
161
162	if (cb_id >= ss->cb_count)
163		return NULL;
164
165	return &ss->cb[cb_id];
166}
167
168int nfnetlink_has_listeners(struct net *net, unsigned int group)
169{
170	struct nfnl_net *nfnlnet = nfnl_pernet(net);
171
172	return netlink_has_listeners(nfnlnet->nfnl, group);
173}
174EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
175
176int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
177		   unsigned int group, int echo, gfp_t flags)
178{
179	struct nfnl_net *nfnlnet = nfnl_pernet(net);
180
181	return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
182}
183EXPORT_SYMBOL_GPL(nfnetlink_send);
184
185int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
186{
187	struct nfnl_net *nfnlnet = nfnl_pernet(net);
188
189	return netlink_set_err(nfnlnet->nfnl, portid, group, error);
190}
191EXPORT_SYMBOL_GPL(nfnetlink_set_err);
192
193int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
194{
195	struct nfnl_net *nfnlnet = nfnl_pernet(net);
196	int err;
197
198	err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
199	if (err == -EAGAIN)
200		err = -ENOBUFS;
201
202	return err;
203}
204EXPORT_SYMBOL_GPL(nfnetlink_unicast);
205
206void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
207			 __u32 group, gfp_t allocation)
208{
209	struct nfnl_net *nfnlnet = nfnl_pernet(net);
210
211	netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
212}
213EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
214
215/* Process one complete nfnetlink message. */
216static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
217			     struct netlink_ext_ack *extack)
218{
219	struct net *net = sock_net(skb->sk);
220	const struct nfnl_callback *nc;
221	const struct nfnetlink_subsystem *ss;
222	int type, err;
223
224	/* All the messages must at least contain nfgenmsg */
225	if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
226		return 0;
227
228	type = nlh->nlmsg_type;
229replay:
230	rcu_read_lock();
231
232	ss = nfnetlink_get_subsys(type);
233	if (!ss) {
234#ifdef CONFIG_MODULES
235		rcu_read_unlock();
236		request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
237		rcu_read_lock();
238		ss = nfnetlink_get_subsys(type);
239		if (!ss)
240#endif
241		{
242			rcu_read_unlock();
243			return -EINVAL;
244		}
245	}
246
247	nc = nfnetlink_find_client(type, ss);
248	if (!nc) {
249		rcu_read_unlock();
250		return -EINVAL;
251	}
252
253	{
254		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
255		struct nfnl_net *nfnlnet = nfnl_pernet(net);
256		u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
257		struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
258		struct nlattr *attr = (void *)nlh + min_len;
259		int attrlen = nlh->nlmsg_len - min_len;
260		__u8 subsys_id = NFNL_SUBSYS_ID(type);
261		struct nfnl_info info = {
262			.net	= net,
263			.sk	= nfnlnet->nfnl,
264			.nlh	= nlh,
265			.nfmsg	= nlmsg_data(nlh),
266			.extack	= extack,
267		};
268
269		/* Sanity-check NFNL_MAX_ATTR_COUNT */
270		if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
271			rcu_read_unlock();
272			return -ENOMEM;
273		}
274
275		err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
276					   attr, attrlen,
277					   ss->cb[cb_id].policy, extack);
278		if (err < 0) {
279			rcu_read_unlock();
280			return err;
281		}
282
283		if (!nc->call) {
284			rcu_read_unlock();
285			return -EINVAL;
286		}
287
288		switch (nc->type) {
289		case NFNL_CB_RCU:
290			err = nc->call(skb, &info, (const struct nlattr **)cda);
291			rcu_read_unlock();
292			break;
293		case NFNL_CB_MUTEX:
294			rcu_read_unlock();
295			nfnl_lock(subsys_id);
296			if (nfnl_dereference_protected(subsys_id) != ss ||
297			    nfnetlink_find_client(type, ss) != nc) {
298				nfnl_unlock(subsys_id);
299				err = -EAGAIN;
300				break;
301			}
302			err = nc->call(skb, &info, (const struct nlattr **)cda);
303			nfnl_unlock(subsys_id);
304			break;
305		default:
306			rcu_read_unlock();
307			err = -EINVAL;
308			break;
309		}
310		if (err == -EAGAIN)
311			goto replay;
312		return err;
313	}
314}
315
316struct nfnl_err {
317	struct list_head	head;
318	struct nlmsghdr		*nlh;
319	int			err;
320	struct netlink_ext_ack	extack;
321};
322
323static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
324			const struct netlink_ext_ack *extack)
325{
326	struct nfnl_err *nfnl_err;
327
328	nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
329	if (nfnl_err == NULL)
330		return -ENOMEM;
331
332	nfnl_err->nlh = nlh;
333	nfnl_err->err = err;
334	nfnl_err->extack = *extack;
335	list_add_tail(&nfnl_err->head, list);
336
337	return 0;
338}
339
340static void nfnl_err_del(struct nfnl_err *nfnl_err)
341{
342	list_del(&nfnl_err->head);
343	kfree(nfnl_err);
344}
345
346static void nfnl_err_reset(struct list_head *err_list)
347{
348	struct nfnl_err *nfnl_err, *next;
349
350	list_for_each_entry_safe(nfnl_err, next, err_list, head)
351		nfnl_err_del(nfnl_err);
352}
353
354static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
355{
356	struct nfnl_err *nfnl_err, *next;
357
358	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
359		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
360			    &nfnl_err->extack);
361		nfnl_err_del(nfnl_err);
362	}
363}
364
365enum {
366	NFNL_BATCH_FAILURE	= (1 << 0),
367	NFNL_BATCH_DONE		= (1 << 1),
368	NFNL_BATCH_REPLAY	= (1 << 2),
369};
370
371static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
372				u16 subsys_id, u32 genid)
373{
374	struct sk_buff *oskb = skb;
375	struct net *net = sock_net(skb->sk);
376	const struct nfnetlink_subsystem *ss;
377	const struct nfnl_callback *nc;
378	struct netlink_ext_ack extack;
379	struct nlmsghdr *onlh = nlh;
380	LIST_HEAD(err_list);
381	u32 status;
382	int err;
383
384	if (subsys_id >= NFNL_SUBSYS_COUNT)
385		return netlink_ack(skb, nlh, -EINVAL, NULL);
386replay:
387	status = 0;
388replay_abort:
389	skb = netlink_skb_clone(oskb, GFP_KERNEL);
390	nlh = onlh;
391	if (!skb)
392		return netlink_ack(oskb, nlh, -ENOMEM, NULL);
393
394	nfnl_lock(subsys_id);
395	ss = nfnl_dereference_protected(subsys_id);
396	if (!ss) {
397#ifdef CONFIG_MODULES
398		nfnl_unlock(subsys_id);
399		request_module("nfnetlink-subsys-%d", subsys_id);
400		nfnl_lock(subsys_id);
401		ss = nfnl_dereference_protected(subsys_id);
402		if (!ss)
403#endif
404		{
405			nfnl_unlock(subsys_id);
406			netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
407			return consume_skb(skb);
408		}
409	}
410
411	if (!ss->valid_genid || !ss->commit || !ss->abort) {
412		nfnl_unlock(subsys_id);
413		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
414		return consume_skb(skb);
415	}
416
417	if (!try_module_get(ss->owner)) {
418		nfnl_unlock(subsys_id);
419		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
420		return consume_skb(skb);
421	}
422
423	if (!ss->valid_genid(net, genid)) {
424		module_put(ss->owner);
425		nfnl_unlock(subsys_id);
426		netlink_ack(oskb, nlh, -ERESTART, NULL);
427		return consume_skb(skb);
428	}
429
430	nfnl_unlock(subsys_id);
431
432	if (nlh->nlmsg_flags & NLM_F_ACK) {
433		memset(&extack, 0, sizeof(extack));
434		nfnl_err_add(&err_list, nlh, 0, &extack);
435	}
436
437	while (skb->len >= nlmsg_total_size(0)) {
438		int msglen, type;
439
440		if (fatal_signal_pending(current)) {
441			nfnl_err_reset(&err_list);
442			err = -EINTR;
443			status = NFNL_BATCH_FAILURE;
444			goto done;
445		}
446
447		memset(&extack, 0, sizeof(extack));
448		nlh = nlmsg_hdr(skb);
449		err = 0;
450
451		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
452		    skb->len < nlh->nlmsg_len ||
453		    nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
454			nfnl_err_reset(&err_list);
455			status |= NFNL_BATCH_FAILURE;
456			goto done;
457		}
458
459		/* Only requests are handled by the kernel */
460		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
461			err = -EINVAL;
462			goto ack;
463		}
464
465		type = nlh->nlmsg_type;
466		if (type == NFNL_MSG_BATCH_BEGIN) {
467			/* Malformed: Batch begin twice */
468			nfnl_err_reset(&err_list);
469			status |= NFNL_BATCH_FAILURE;
470			goto done;
471		} else if (type == NFNL_MSG_BATCH_END) {
472			status |= NFNL_BATCH_DONE;
473			goto done;
474		} else if (type < NLMSG_MIN_TYPE) {
475			err = -EINVAL;
476			goto ack;
477		}
478
479		/* We only accept a batch with messages for the same
480		 * subsystem.
481		 */
482		if (NFNL_SUBSYS_ID(type) != subsys_id) {
483			err = -EINVAL;
484			goto ack;
485		}
486
487		nc = nfnetlink_find_client(type, ss);
488		if (!nc) {
489			err = -EINVAL;
490			goto ack;
491		}
492
493		if (nc->type != NFNL_CB_BATCH) {
494			err = -EINVAL;
495			goto ack;
496		}
497
498		{
499			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
500			struct nfnl_net *nfnlnet = nfnl_pernet(net);
501			struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
502			struct nlattr *attr = (void *)nlh + min_len;
503			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
504			int attrlen = nlh->nlmsg_len - min_len;
505			struct nfnl_info info = {
506				.net	= net,
507				.sk	= nfnlnet->nfnl,
508				.nlh	= nlh,
509				.nfmsg	= nlmsg_data(nlh),
510				.extack	= &extack,
511			};
512
513			/* Sanity-check NFTA_MAX_ATTR */
514			if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
515				err = -ENOMEM;
516				goto ack;
517			}
518
519			err = nla_parse_deprecated(cda,
520						   ss->cb[cb_id].attr_count,
521						   attr, attrlen,
522						   ss->cb[cb_id].policy, &extack);
523			if (err < 0)
524				goto ack;
525
526			err = nc->call(skb, &info, (const struct nlattr **)cda);
527
528			/* The lock was released to autoload some module, we
529			 * have to abort and start from scratch using the
530			 * original skb.
531			 */
532			if (err == -EAGAIN) {
533				status |= NFNL_BATCH_REPLAY;
534				goto done;
535			}
536		}
537ack:
538		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
539			/* Errors are delivered once the full batch has been
540			 * processed, this avoids that the same error is
541			 * reported several times when replaying the batch.
542			 */
543			if (err == -ENOMEM ||
544			    nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
545				/* We failed to enqueue an error, reset the
546				 * list of errors and send OOM to userspace
547				 * pointing to the batch header.
548				 */
549				nfnl_err_reset(&err_list);
550				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
551					    NULL);
552				status |= NFNL_BATCH_FAILURE;
553				goto done;
554			}
555			/* We don't stop processing the batch on errors, thus,
556			 * userspace gets all the errors that the batch
557			 * triggers.
558			 */
559			if (err)
560				status |= NFNL_BATCH_FAILURE;
561		}
562
563		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
564		if (msglen > skb->len)
565			msglen = skb->len;
566		skb_pull(skb, msglen);
567	}
568done:
569	if (status & NFNL_BATCH_REPLAY) {
570		ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
571		nfnl_err_reset(&err_list);
572		consume_skb(skb);
573		module_put(ss->owner);
574		goto replay;
575	} else if (status == NFNL_BATCH_DONE) {
576		err = ss->commit(net, oskb);
577		if (err == -EAGAIN) {
578			status |= NFNL_BATCH_REPLAY;
579			goto done;
580		} else if (err) {
581			ss->abort(net, oskb, NFNL_ABORT_NONE);
582			netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
583		} else if (nlh->nlmsg_flags & NLM_F_ACK) {
584			memset(&extack, 0, sizeof(extack));
585			nfnl_err_add(&err_list, nlh, 0, &extack);
586		}
587	} else {
588		enum nfnl_abort_action abort_action;
589
590		if (status & NFNL_BATCH_FAILURE)
591			abort_action = NFNL_ABORT_NONE;
592		else
593			abort_action = NFNL_ABORT_VALIDATE;
594
595		err = ss->abort(net, oskb, abort_action);
596		if (err == -EAGAIN) {
597			nfnl_err_reset(&err_list);
598			consume_skb(skb);
599			module_put(ss->owner);
600			status |= NFNL_BATCH_FAILURE;
601			goto replay_abort;
602		}
603	}
604
605	nfnl_err_deliver(&err_list, oskb);
606	consume_skb(skb);
607	module_put(ss->owner);
608}
609
610static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
611	[NFNL_BATCH_GENID]	= { .type = NLA_U32 },
612};
613
614static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
615{
616	int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
617	struct nlattr *attr = (void *)nlh + min_len;
618	struct nlattr *cda[NFNL_BATCH_MAX + 1];
619	int attrlen = nlh->nlmsg_len - min_len;
620	struct nfgenmsg *nfgenmsg;
621	int msglen, err;
622	u32 gen_id = 0;
623	u16 res_id;
624
625	msglen = NLMSG_ALIGN(nlh->nlmsg_len);
626	if (msglen > skb->len)
627		msglen = skb->len;
628
629	if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
630		return;
631
632	err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
633				   nfnl_batch_policy, NULL);
634	if (err < 0) {
635		netlink_ack(skb, nlh, err, NULL);
636		return;
637	}
638	if (cda[NFNL_BATCH_GENID])
639		gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
640
641	nfgenmsg = nlmsg_data(nlh);
642	skb_pull(skb, msglen);
643	/* Work around old nft using host byte order */
644	if (nfgenmsg->res_id == (__force __be16)NFNL_SUBSYS_NFTABLES)
645		res_id = NFNL_SUBSYS_NFTABLES;
646	else
647		res_id = ntohs(nfgenmsg->res_id);
648
649	nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
650}
651
652static void nfnetlink_rcv(struct sk_buff *skb)
653{
654	struct nlmsghdr *nlh = nlmsg_hdr(skb);
655
656	if (skb->len < NLMSG_HDRLEN ||
657	    nlh->nlmsg_len < NLMSG_HDRLEN ||
658	    skb->len < nlh->nlmsg_len)
659		return;
660
661	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
662		netlink_ack(skb, nlh, -EPERM, NULL);
663		return;
664	}
665
666	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
667		nfnetlink_rcv_skb_batch(skb, nlh);
668	else
669		netlink_rcv_skb(skb, nfnetlink_rcv_msg);
670}
671
672static void nfnetlink_bind_event(struct net *net, unsigned int group)
673{
674#ifdef CONFIG_NF_CONNTRACK_EVENTS
675	int type, group_bit;
676	u8 v;
677
678	/* All NFNLGRP_CONNTRACK_* group bits fit into u8.
679	 * The other groups are not relevant and can be ignored.
680	 */
681	if (group >= 8)
682		return;
683
684	type = nfnl_group2type[group];
685
686	switch (type) {
687	case NFNL_SUBSYS_CTNETLINK:
688		break;
689	case NFNL_SUBSYS_CTNETLINK_EXP:
690		break;
691	default:
692		return;
693	}
694
695	group_bit = (1 << group);
696
697	spin_lock(&nfnl_grp_active_lock);
698	v = READ_ONCE(nf_ctnetlink_has_listener);
699	if ((v & group_bit) == 0) {
700		v |= group_bit;
701
702		/* read concurrently without nfnl_grp_active_lock held. */
703		WRITE_ONCE(nf_ctnetlink_has_listener, v);
704	}
705
706	spin_unlock(&nfnl_grp_active_lock);
707#endif
708}
709
710static int nfnetlink_bind(struct net *net, int group)
711{
712	const struct nfnetlink_subsystem *ss;
713	int type;
714
715	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
716		return 0;
717
718	type = nfnl_group2type[group];
719
720	rcu_read_lock();
721	ss = nfnetlink_get_subsys(type << 8);
722	rcu_read_unlock();
723	if (!ss)
724		request_module_nowait("nfnetlink-subsys-%d", type);
725
726	nfnetlink_bind_event(net, group);
727	return 0;
728}
729
730static void nfnetlink_unbind(struct net *net, int group)
731{
732#ifdef CONFIG_NF_CONNTRACK_EVENTS
733	int type, group_bit;
734
735	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
736		return;
737
738	type = nfnl_group2type[group];
739
740	switch (type) {
741	case NFNL_SUBSYS_CTNETLINK:
742		break;
743	case NFNL_SUBSYS_CTNETLINK_EXP:
744		break;
745	default:
746		return;
747	}
748
749	/* ctnetlink_has_listener is u8 */
750	if (group >= 8)
751		return;
752
753	group_bit = (1 << group);
754
755	spin_lock(&nfnl_grp_active_lock);
756	if (!nfnetlink_has_listeners(net, group)) {
757		u8 v = READ_ONCE(nf_ctnetlink_has_listener);
758
759		v &= ~group_bit;
760
761		/* read concurrently without nfnl_grp_active_lock held. */
762		WRITE_ONCE(nf_ctnetlink_has_listener, v);
763	}
764	spin_unlock(&nfnl_grp_active_lock);
765#endif
766}
767
768static int __net_init nfnetlink_net_init(struct net *net)
769{
770	struct nfnl_net *nfnlnet = nfnl_pernet(net);
771	struct netlink_kernel_cfg cfg = {
772		.groups	= NFNLGRP_MAX,
773		.input	= nfnetlink_rcv,
774		.bind	= nfnetlink_bind,
775		.unbind	= nfnetlink_unbind,
776	};
777
778	nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
779	if (!nfnlnet->nfnl)
780		return -ENOMEM;
781	return 0;
782}
783
784static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
785{
786	struct nfnl_net *nfnlnet;
787	struct net *net;
788
789	list_for_each_entry(net, net_exit_list, exit_list) {
790		nfnlnet = nfnl_pernet(net);
791
792		netlink_kernel_release(nfnlnet->nfnl);
793	}
794}
795
796static struct pernet_operations nfnetlink_net_ops = {
797	.init		= nfnetlink_net_init,
798	.exit_batch	= nfnetlink_net_exit_batch,
799	.id		= &nfnetlink_pernet_id,
800	.size		= sizeof(struct nfnl_net),
801};
802
803static int __init nfnetlink_init(void)
804{
805	int i;
806
807	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
808		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
809
810	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
811		__mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
812
813	return register_pernet_subsys(&nfnetlink_net_ops);
814}
815
816static void __exit nfnetlink_exit(void)
817{
818	unregister_pernet_subsys(&nfnetlink_net_ops);
819}
820module_init(nfnetlink_init);
821module_exit(nfnetlink_exit);