Linux Audio

Check our new training course

Loading...
v5.14.15
  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/nfnetlink.h>
 33
 34MODULE_LICENSE("GPL");
 35MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
 36MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
 37MODULE_DESCRIPTION("Netfilter messages via netlink socket");
 38
 39#define nfnl_dereference_protected(id) \
 40	rcu_dereference_protected(table[(id)].subsys, \
 41				  lockdep_nfnl_is_held((id)))
 42
 43#define NFNL_MAX_ATTR_COUNT	32
 44
 45static unsigned int nfnetlink_pernet_id __read_mostly;
 46
 
 
 
 
 47struct nfnl_net {
 48	struct sock *nfnl;
 49};
 50
 51static struct {
 52	struct mutex				mutex;
 53	const struct nfnetlink_subsystem __rcu	*subsys;
 54} table[NFNL_SUBSYS_COUNT];
 55
 56static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
 57
 58static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = {
 59	[NFNL_SUBSYS_NONE] = "nfnl_subsys_none",
 60	[NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink",
 61	[NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp",
 62	[NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue",
 63	[NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog",
 64	[NFNL_SUBSYS_OSF] = "nfnl_subsys_osf",
 65	[NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset",
 66	[NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct",
 67	[NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout",
 68	[NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper",
 69	[NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables",
 70	[NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat",
 71	[NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook",
 72};
 73
 74static const int nfnl_group2type[NFNLGRP_MAX+1] = {
 75	[NFNLGRP_CONNTRACK_NEW]		= NFNL_SUBSYS_CTNETLINK,
 76	[NFNLGRP_CONNTRACK_UPDATE]	= NFNL_SUBSYS_CTNETLINK,
 77	[NFNLGRP_CONNTRACK_DESTROY]	= NFNL_SUBSYS_CTNETLINK,
 78	[NFNLGRP_CONNTRACK_EXP_NEW]	= NFNL_SUBSYS_CTNETLINK_EXP,
 79	[NFNLGRP_CONNTRACK_EXP_UPDATE]	= NFNL_SUBSYS_CTNETLINK_EXP,
 80	[NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
 81	[NFNLGRP_NFTABLES]		= NFNL_SUBSYS_NFTABLES,
 82	[NFNLGRP_ACCT_QUOTA]		= NFNL_SUBSYS_ACCT,
 83	[NFNLGRP_NFTRACE]		= NFNL_SUBSYS_NFTABLES,
 84};
 85
 86static struct nfnl_net *nfnl_pernet(struct net *net)
 87{
 88	return net_generic(net, nfnetlink_pernet_id);
 89}
 90
 91void nfnl_lock(__u8 subsys_id)
 92{
 93	mutex_lock(&table[subsys_id].mutex);
 94}
 95EXPORT_SYMBOL_GPL(nfnl_lock);
 96
 97void nfnl_unlock(__u8 subsys_id)
 98{
 99	mutex_unlock(&table[subsys_id].mutex);
100}
101EXPORT_SYMBOL_GPL(nfnl_unlock);
102
103#ifdef CONFIG_PROVE_LOCKING
104bool lockdep_nfnl_is_held(u8 subsys_id)
105{
106	return lockdep_is_held(&table[subsys_id].mutex);
107}
108EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
109#endif
110
111int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
112{
113	u8 cb_id;
114
115	/* Sanity-check attr_count size to avoid stack buffer overflow. */
116	for (cb_id = 0; cb_id < n->cb_count; cb_id++)
117		if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
118			return -EINVAL;
119
120	nfnl_lock(n->subsys_id);
121	if (table[n->subsys_id].subsys) {
122		nfnl_unlock(n->subsys_id);
123		return -EBUSY;
124	}
125	rcu_assign_pointer(table[n->subsys_id].subsys, n);
126	nfnl_unlock(n->subsys_id);
127
128	return 0;
129}
130EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
131
132int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
133{
134	nfnl_lock(n->subsys_id);
135	table[n->subsys_id].subsys = NULL;
136	nfnl_unlock(n->subsys_id);
137	synchronize_rcu();
138	return 0;
139}
140EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
141
142static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
143{
144	u8 subsys_id = NFNL_SUBSYS_ID(type);
145
146	if (subsys_id >= NFNL_SUBSYS_COUNT)
147		return NULL;
148
149	return rcu_dereference(table[subsys_id].subsys);
150}
151
152static inline const struct nfnl_callback *
153nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
154{
155	u8 cb_id = NFNL_MSG_TYPE(type);
156
157	if (cb_id >= ss->cb_count)
158		return NULL;
159
160	return &ss->cb[cb_id];
161}
162
163int nfnetlink_has_listeners(struct net *net, unsigned int group)
164{
165	struct nfnl_net *nfnlnet = nfnl_pernet(net);
166
167	return netlink_has_listeners(nfnlnet->nfnl, group);
168}
169EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
170
171int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
172		   unsigned int group, int echo, gfp_t flags)
173{
174	struct nfnl_net *nfnlnet = nfnl_pernet(net);
175
176	return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
177}
178EXPORT_SYMBOL_GPL(nfnetlink_send);
179
180int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
181{
182	struct nfnl_net *nfnlnet = nfnl_pernet(net);
183
184	return netlink_set_err(nfnlnet->nfnl, portid, group, error);
185}
186EXPORT_SYMBOL_GPL(nfnetlink_set_err);
187
188int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
189{
190	struct nfnl_net *nfnlnet = nfnl_pernet(net);
191	int err;
192
193	err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
194	if (err == -EAGAIN)
195		err = -ENOBUFS;
196
197	return err;
198}
199EXPORT_SYMBOL_GPL(nfnetlink_unicast);
200
201void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
202			 __u32 group, gfp_t allocation)
203{
204	struct nfnl_net *nfnlnet = nfnl_pernet(net);
205
206	netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
207}
208EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
209
210/* Process one complete nfnetlink message. */
211static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
212			     struct netlink_ext_ack *extack)
213{
214	struct net *net = sock_net(skb->sk);
215	const struct nfnl_callback *nc;
216	const struct nfnetlink_subsystem *ss;
217	int type, err;
218
219	/* All the messages must at least contain nfgenmsg */
220	if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
221		return 0;
222
223	type = nlh->nlmsg_type;
224replay:
225	rcu_read_lock();
226
227	ss = nfnetlink_get_subsys(type);
228	if (!ss) {
229#ifdef CONFIG_MODULES
230		rcu_read_unlock();
231		request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
232		rcu_read_lock();
233		ss = nfnetlink_get_subsys(type);
234		if (!ss)
235#endif
236		{
237			rcu_read_unlock();
238			return -EINVAL;
239		}
240	}
241
242	nc = nfnetlink_find_client(type, ss);
243	if (!nc) {
244		rcu_read_unlock();
245		return -EINVAL;
246	}
247
248	{
249		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
250		struct nfnl_net *nfnlnet = nfnl_pernet(net);
251		u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
252		struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
253		struct nlattr *attr = (void *)nlh + min_len;
254		int attrlen = nlh->nlmsg_len - min_len;
255		__u8 subsys_id = NFNL_SUBSYS_ID(type);
256		struct nfnl_info info = {
257			.net	= net,
258			.sk	= nfnlnet->nfnl,
259			.nlh	= nlh,
260			.nfmsg	= nlmsg_data(nlh),
261			.extack	= extack,
262		};
263
264		/* Sanity-check NFNL_MAX_ATTR_COUNT */
265		if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
266			rcu_read_unlock();
267			return -ENOMEM;
268		}
269
270		err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
271					   attr, attrlen,
272					   ss->cb[cb_id].policy, extack);
273		if (err < 0) {
274			rcu_read_unlock();
275			return err;
276		}
277
278		if (!nc->call) {
279			rcu_read_unlock();
280			return -EINVAL;
281		}
282
283		switch (nc->type) {
284		case NFNL_CB_RCU:
285			err = nc->call(skb, &info, (const struct nlattr **)cda);
286			rcu_read_unlock();
287			break;
288		case NFNL_CB_MUTEX:
289			rcu_read_unlock();
290			nfnl_lock(subsys_id);
291			if (nfnl_dereference_protected(subsys_id) != ss ||
292			    nfnetlink_find_client(type, ss) != nc) {
 
293				err = -EAGAIN;
294				break;
295			}
296			err = nc->call(skb, &info, (const struct nlattr **)cda);
297			nfnl_unlock(subsys_id);
298			break;
299		default:
300			rcu_read_unlock();
301			err = -EINVAL;
302			break;
303		}
304		if (err == -EAGAIN)
305			goto replay;
306		return err;
307	}
308}
309
310struct nfnl_err {
311	struct list_head	head;
312	struct nlmsghdr		*nlh;
313	int			err;
314	struct netlink_ext_ack	extack;
315};
316
317static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
318			const struct netlink_ext_ack *extack)
319{
320	struct nfnl_err *nfnl_err;
321
322	nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
323	if (nfnl_err == NULL)
324		return -ENOMEM;
325
326	nfnl_err->nlh = nlh;
327	nfnl_err->err = err;
328	nfnl_err->extack = *extack;
329	list_add_tail(&nfnl_err->head, list);
330
331	return 0;
332}
333
334static void nfnl_err_del(struct nfnl_err *nfnl_err)
335{
336	list_del(&nfnl_err->head);
337	kfree(nfnl_err);
338}
339
340static void nfnl_err_reset(struct list_head *err_list)
341{
342	struct nfnl_err *nfnl_err, *next;
343
344	list_for_each_entry_safe(nfnl_err, next, err_list, head)
345		nfnl_err_del(nfnl_err);
346}
347
348static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
349{
350	struct nfnl_err *nfnl_err, *next;
351
352	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
353		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
354			    &nfnl_err->extack);
355		nfnl_err_del(nfnl_err);
356	}
357}
358
359enum {
360	NFNL_BATCH_FAILURE	= (1 << 0),
361	NFNL_BATCH_DONE		= (1 << 1),
362	NFNL_BATCH_REPLAY	= (1 << 2),
363};
364
365static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
366				u16 subsys_id, u32 genid)
367{
368	struct sk_buff *oskb = skb;
369	struct net *net = sock_net(skb->sk);
370	const struct nfnetlink_subsystem *ss;
371	const struct nfnl_callback *nc;
372	struct netlink_ext_ack extack;
373	LIST_HEAD(err_list);
374	u32 status;
375	int err;
376
377	if (subsys_id >= NFNL_SUBSYS_COUNT)
378		return netlink_ack(skb, nlh, -EINVAL, NULL);
379replay:
380	status = 0;
381replay_abort:
382	skb = netlink_skb_clone(oskb, GFP_KERNEL);
383	if (!skb)
384		return netlink_ack(oskb, nlh, -ENOMEM, NULL);
385
386	nfnl_lock(subsys_id);
387	ss = nfnl_dereference_protected(subsys_id);
388	if (!ss) {
389#ifdef CONFIG_MODULES
390		nfnl_unlock(subsys_id);
391		request_module("nfnetlink-subsys-%d", subsys_id);
392		nfnl_lock(subsys_id);
393		ss = nfnl_dereference_protected(subsys_id);
394		if (!ss)
395#endif
396		{
397			nfnl_unlock(subsys_id);
398			netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
399			return kfree_skb(skb);
400		}
401	}
402
403	if (!ss->valid_genid || !ss->commit || !ss->abort) {
404		nfnl_unlock(subsys_id);
405		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
406		return kfree_skb(skb);
407	}
408
409	if (!try_module_get(ss->owner)) {
410		nfnl_unlock(subsys_id);
411		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
412		return kfree_skb(skb);
413	}
414
415	if (!ss->valid_genid(net, genid)) {
416		module_put(ss->owner);
417		nfnl_unlock(subsys_id);
418		netlink_ack(oskb, nlh, -ERESTART, NULL);
419		return kfree_skb(skb);
420	}
421
422	nfnl_unlock(subsys_id);
423
424	while (skb->len >= nlmsg_total_size(0)) {
425		int msglen, type;
426
427		if (fatal_signal_pending(current)) {
428			nfnl_err_reset(&err_list);
429			err = -EINTR;
430			status = NFNL_BATCH_FAILURE;
431			goto done;
432		}
433
434		memset(&extack, 0, sizeof(extack));
435		nlh = nlmsg_hdr(skb);
436		err = 0;
437
438		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
439		    skb->len < nlh->nlmsg_len ||
440		    nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
441			nfnl_err_reset(&err_list);
442			status |= NFNL_BATCH_FAILURE;
443			goto done;
444		}
445
446		/* Only requests are handled by the kernel */
447		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
448			err = -EINVAL;
449			goto ack;
450		}
451
452		type = nlh->nlmsg_type;
453		if (type == NFNL_MSG_BATCH_BEGIN) {
454			/* Malformed: Batch begin twice */
455			nfnl_err_reset(&err_list);
456			status |= NFNL_BATCH_FAILURE;
457			goto done;
458		} else if (type == NFNL_MSG_BATCH_END) {
459			status |= NFNL_BATCH_DONE;
460			goto done;
461		} else if (type < NLMSG_MIN_TYPE) {
462			err = -EINVAL;
463			goto ack;
464		}
465
466		/* We only accept a batch with messages for the same
467		 * subsystem.
468		 */
469		if (NFNL_SUBSYS_ID(type) != subsys_id) {
470			err = -EINVAL;
471			goto ack;
472		}
473
474		nc = nfnetlink_find_client(type, ss);
475		if (!nc) {
476			err = -EINVAL;
477			goto ack;
478		}
479
480		if (nc->type != NFNL_CB_BATCH) {
481			err = -EINVAL;
482			goto ack;
483		}
484
485		{
486			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
487			struct nfnl_net *nfnlnet = nfnl_pernet(net);
488			struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
489			struct nlattr *attr = (void *)nlh + min_len;
490			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
491			int attrlen = nlh->nlmsg_len - min_len;
492			struct nfnl_info info = {
493				.net	= net,
494				.sk	= nfnlnet->nfnl,
495				.nlh	= nlh,
496				.nfmsg	= nlmsg_data(nlh),
497				.extack	= &extack,
498			};
499
500			/* Sanity-check NFTA_MAX_ATTR */
501			if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
502				err = -ENOMEM;
503				goto ack;
504			}
505
506			err = nla_parse_deprecated(cda,
507						   ss->cb[cb_id].attr_count,
508						   attr, attrlen,
509						   ss->cb[cb_id].policy, NULL);
510			if (err < 0)
511				goto ack;
512
513			err = nc->call(skb, &info, (const struct nlattr **)cda);
514
515			/* The lock was released to autoload some module, we
516			 * have to abort and start from scratch using the
517			 * original skb.
518			 */
519			if (err == -EAGAIN) {
520				status |= NFNL_BATCH_REPLAY;
521				goto done;
522			}
523		}
524ack:
525		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
526			/* Errors are delivered once the full batch has been
527			 * processed, this avoids that the same error is
528			 * reported several times when replaying the batch.
529			 */
530			if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
 
531				/* We failed to enqueue an error, reset the
532				 * list of errors and send OOM to userspace
533				 * pointing to the batch header.
534				 */
535				nfnl_err_reset(&err_list);
536				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
537					    NULL);
538				status |= NFNL_BATCH_FAILURE;
539				goto done;
540			}
541			/* We don't stop processing the batch on errors, thus,
542			 * userspace gets all the errors that the batch
543			 * triggers.
544			 */
545			if (err)
546				status |= NFNL_BATCH_FAILURE;
547		}
548
549		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
550		if (msglen > skb->len)
551			msglen = skb->len;
552		skb_pull(skb, msglen);
553	}
554done:
555	if (status & NFNL_BATCH_REPLAY) {
556		ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
557		nfnl_err_reset(&err_list);
558		kfree_skb(skb);
559		module_put(ss->owner);
560		goto replay;
561	} else if (status == NFNL_BATCH_DONE) {
562		err = ss->commit(net, oskb);
563		if (err == -EAGAIN) {
564			status |= NFNL_BATCH_REPLAY;
565			goto done;
566		} else if (err) {
567			ss->abort(net, oskb, NFNL_ABORT_NONE);
568			netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
569		}
570	} else {
571		enum nfnl_abort_action abort_action;
572
573		if (status & NFNL_BATCH_FAILURE)
574			abort_action = NFNL_ABORT_NONE;
575		else
576			abort_action = NFNL_ABORT_VALIDATE;
577
578		err = ss->abort(net, oskb, abort_action);
579		if (err == -EAGAIN) {
580			nfnl_err_reset(&err_list);
581			kfree_skb(skb);
582			module_put(ss->owner);
583			status |= NFNL_BATCH_FAILURE;
584			goto replay_abort;
585		}
586	}
587	if (ss->cleanup)
588		ss->cleanup(net);
589
590	nfnl_err_deliver(&err_list, oskb);
591	kfree_skb(skb);
592	module_put(ss->owner);
593}
594
595static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
596	[NFNL_BATCH_GENID]	= { .type = NLA_U32 },
597};
598
599static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
600{
601	int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
602	struct nlattr *attr = (void *)nlh + min_len;
603	struct nlattr *cda[NFNL_BATCH_MAX + 1];
604	int attrlen = nlh->nlmsg_len - min_len;
605	struct nfgenmsg *nfgenmsg;
606	int msglen, err;
607	u32 gen_id = 0;
608	u16 res_id;
609
610	msglen = NLMSG_ALIGN(nlh->nlmsg_len);
611	if (msglen > skb->len)
612		msglen = skb->len;
613
614	if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
615		return;
616
617	err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
618				   nfnl_batch_policy, NULL);
619	if (err < 0) {
620		netlink_ack(skb, nlh, err, NULL);
621		return;
622	}
623	if (cda[NFNL_BATCH_GENID])
624		gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
625
626	nfgenmsg = nlmsg_data(nlh);
627	skb_pull(skb, msglen);
628	/* Work around old nft using host byte order */
629	if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
630		res_id = NFNL_SUBSYS_NFTABLES;
631	else
632		res_id = ntohs(nfgenmsg->res_id);
633
634	nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
635}
636
637static void nfnetlink_rcv(struct sk_buff *skb)
638{
639	struct nlmsghdr *nlh = nlmsg_hdr(skb);
640
641	if (skb->len < NLMSG_HDRLEN ||
642	    nlh->nlmsg_len < NLMSG_HDRLEN ||
643	    skb->len < nlh->nlmsg_len)
644		return;
645
646	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
647		netlink_ack(skb, nlh, -EPERM, NULL);
648		return;
649	}
650
651	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
652		nfnetlink_rcv_skb_batch(skb, nlh);
653	else
654		netlink_rcv_skb(skb, nfnetlink_rcv_msg);
655}
656
657#ifdef CONFIG_MODULES
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
658static int nfnetlink_bind(struct net *net, int group)
659{
660	const struct nfnetlink_subsystem *ss;
661	int type;
662
663	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
664		return 0;
665
666	type = nfnl_group2type[group];
667
668	rcu_read_lock();
669	ss = nfnetlink_get_subsys(type << 8);
670	rcu_read_unlock();
671	if (!ss)
672		request_module_nowait("nfnetlink-subsys-%d", type);
 
 
673	return 0;
674}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
675#endif
 
676
677static int __net_init nfnetlink_net_init(struct net *net)
678{
679	struct nfnl_net *nfnlnet = nfnl_pernet(net);
680	struct netlink_kernel_cfg cfg = {
681		.groups	= NFNLGRP_MAX,
682		.input	= nfnetlink_rcv,
683#ifdef CONFIG_MODULES
684		.bind	= nfnetlink_bind,
685#endif
686	};
687
688	nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
689	if (!nfnlnet->nfnl)
690		return -ENOMEM;
691	return 0;
692}
693
694static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
695{
696	struct nfnl_net *nfnlnet;
697	struct net *net;
698
699	list_for_each_entry(net, net_exit_list, exit_list) {
700		nfnlnet = nfnl_pernet(net);
701
702		netlink_kernel_release(nfnlnet->nfnl);
703	}
704}
705
706static struct pernet_operations nfnetlink_net_ops = {
707	.init		= nfnetlink_net_init,
708	.exit_batch	= nfnetlink_net_exit_batch,
709	.id		= &nfnetlink_pernet_id,
710	.size		= sizeof(struct nfnl_net),
711};
712
713static int __init nfnetlink_init(void)
714{
715	int i;
716
717	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
718		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
719
720	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
721		__mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
722
723	return register_pernet_subsys(&nfnetlink_net_ops);
724}
725
726static void __exit nfnetlink_exit(void)
727{
728	unregister_pernet_subsys(&nfnetlink_net_ops);
729}
730module_init(nfnetlink_init);
731module_exit(nfnetlink_exit);
v6.9.4
  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	LIST_HEAD(err_list);
380	u32 status;
381	int err;
382
383	if (subsys_id >= NFNL_SUBSYS_COUNT)
384		return netlink_ack(skb, nlh, -EINVAL, NULL);
385replay:
386	status = 0;
387replay_abort:
388	skb = netlink_skb_clone(oskb, GFP_KERNEL);
389	if (!skb)
390		return netlink_ack(oskb, nlh, -ENOMEM, NULL);
391
392	nfnl_lock(subsys_id);
393	ss = nfnl_dereference_protected(subsys_id);
394	if (!ss) {
395#ifdef CONFIG_MODULES
396		nfnl_unlock(subsys_id);
397		request_module("nfnetlink-subsys-%d", subsys_id);
398		nfnl_lock(subsys_id);
399		ss = nfnl_dereference_protected(subsys_id);
400		if (!ss)
401#endif
402		{
403			nfnl_unlock(subsys_id);
404			netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
405			return kfree_skb(skb);
406		}
407	}
408
409	if (!ss->valid_genid || !ss->commit || !ss->abort) {
410		nfnl_unlock(subsys_id);
411		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
412		return kfree_skb(skb);
413	}
414
415	if (!try_module_get(ss->owner)) {
416		nfnl_unlock(subsys_id);
417		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
418		return kfree_skb(skb);
419	}
420
421	if (!ss->valid_genid(net, genid)) {
422		module_put(ss->owner);
423		nfnl_unlock(subsys_id);
424		netlink_ack(oskb, nlh, -ERESTART, NULL);
425		return kfree_skb(skb);
426	}
427
428	nfnl_unlock(subsys_id);
429
430	while (skb->len >= nlmsg_total_size(0)) {
431		int msglen, type;
432
433		if (fatal_signal_pending(current)) {
434			nfnl_err_reset(&err_list);
435			err = -EINTR;
436			status = NFNL_BATCH_FAILURE;
437			goto done;
438		}
439
440		memset(&extack, 0, sizeof(extack));
441		nlh = nlmsg_hdr(skb);
442		err = 0;
443
444		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
445		    skb->len < nlh->nlmsg_len ||
446		    nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
447			nfnl_err_reset(&err_list);
448			status |= NFNL_BATCH_FAILURE;
449			goto done;
450		}
451
452		/* Only requests are handled by the kernel */
453		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
454			err = -EINVAL;
455			goto ack;
456		}
457
458		type = nlh->nlmsg_type;
459		if (type == NFNL_MSG_BATCH_BEGIN) {
460			/* Malformed: Batch begin twice */
461			nfnl_err_reset(&err_list);
462			status |= NFNL_BATCH_FAILURE;
463			goto done;
464		} else if (type == NFNL_MSG_BATCH_END) {
465			status |= NFNL_BATCH_DONE;
466			goto done;
467		} else if (type < NLMSG_MIN_TYPE) {
468			err = -EINVAL;
469			goto ack;
470		}
471
472		/* We only accept a batch with messages for the same
473		 * subsystem.
474		 */
475		if (NFNL_SUBSYS_ID(type) != subsys_id) {
476			err = -EINVAL;
477			goto ack;
478		}
479
480		nc = nfnetlink_find_client(type, ss);
481		if (!nc) {
482			err = -EINVAL;
483			goto ack;
484		}
485
486		if (nc->type != NFNL_CB_BATCH) {
487			err = -EINVAL;
488			goto ack;
489		}
490
491		{
492			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
493			struct nfnl_net *nfnlnet = nfnl_pernet(net);
494			struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
495			struct nlattr *attr = (void *)nlh + min_len;
496			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
497			int attrlen = nlh->nlmsg_len - min_len;
498			struct nfnl_info info = {
499				.net	= net,
500				.sk	= nfnlnet->nfnl,
501				.nlh	= nlh,
502				.nfmsg	= nlmsg_data(nlh),
503				.extack	= &extack,
504			};
505
506			/* Sanity-check NFTA_MAX_ATTR */
507			if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
508				err = -ENOMEM;
509				goto ack;
510			}
511
512			err = nla_parse_deprecated(cda,
513						   ss->cb[cb_id].attr_count,
514						   attr, attrlen,
515						   ss->cb[cb_id].policy, NULL);
516			if (err < 0)
517				goto ack;
518
519			err = nc->call(skb, &info, (const struct nlattr **)cda);
520
521			/* The lock was released to autoload some module, we
522			 * have to abort and start from scratch using the
523			 * original skb.
524			 */
525			if (err == -EAGAIN) {
526				status |= NFNL_BATCH_REPLAY;
527				goto done;
528			}
529		}
530ack:
531		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
532			/* Errors are delivered once the full batch has been
533			 * processed, this avoids that the same error is
534			 * reported several times when replaying the batch.
535			 */
536			if (err == -ENOMEM ||
537			    nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
538				/* We failed to enqueue an error, reset the
539				 * list of errors and send OOM to userspace
540				 * pointing to the batch header.
541				 */
542				nfnl_err_reset(&err_list);
543				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
544					    NULL);
545				status |= NFNL_BATCH_FAILURE;
546				goto done;
547			}
548			/* We don't stop processing the batch on errors, thus,
549			 * userspace gets all the errors that the batch
550			 * triggers.
551			 */
552			if (err)
553				status |= NFNL_BATCH_FAILURE;
554		}
555
556		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
557		if (msglen > skb->len)
558			msglen = skb->len;
559		skb_pull(skb, msglen);
560	}
561done:
562	if (status & NFNL_BATCH_REPLAY) {
563		ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
564		nfnl_err_reset(&err_list);
565		kfree_skb(skb);
566		module_put(ss->owner);
567		goto replay;
568	} else if (status == NFNL_BATCH_DONE) {
569		err = ss->commit(net, oskb);
570		if (err == -EAGAIN) {
571			status |= NFNL_BATCH_REPLAY;
572			goto done;
573		} else if (err) {
574			ss->abort(net, oskb, NFNL_ABORT_NONE);
575			netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
576		}
577	} else {
578		enum nfnl_abort_action abort_action;
579
580		if (status & NFNL_BATCH_FAILURE)
581			abort_action = NFNL_ABORT_NONE;
582		else
583			abort_action = NFNL_ABORT_VALIDATE;
584
585		err = ss->abort(net, oskb, abort_action);
586		if (err == -EAGAIN) {
587			nfnl_err_reset(&err_list);
588			kfree_skb(skb);
589			module_put(ss->owner);
590			status |= NFNL_BATCH_FAILURE;
591			goto replay_abort;
592		}
593	}
 
 
594
595	nfnl_err_deliver(&err_list, oskb);
596	kfree_skb(skb);
597	module_put(ss->owner);
598}
599
600static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
601	[NFNL_BATCH_GENID]	= { .type = NLA_U32 },
602};
603
604static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
605{
606	int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
607	struct nlattr *attr = (void *)nlh + min_len;
608	struct nlattr *cda[NFNL_BATCH_MAX + 1];
609	int attrlen = nlh->nlmsg_len - min_len;
610	struct nfgenmsg *nfgenmsg;
611	int msglen, err;
612	u32 gen_id = 0;
613	u16 res_id;
614
615	msglen = NLMSG_ALIGN(nlh->nlmsg_len);
616	if (msglen > skb->len)
617		msglen = skb->len;
618
619	if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
620		return;
621
622	err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
623				   nfnl_batch_policy, NULL);
624	if (err < 0) {
625		netlink_ack(skb, nlh, err, NULL);
626		return;
627	}
628	if (cda[NFNL_BATCH_GENID])
629		gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
630
631	nfgenmsg = nlmsg_data(nlh);
632	skb_pull(skb, msglen);
633	/* Work around old nft using host byte order */
634	if (nfgenmsg->res_id == (__force __be16)NFNL_SUBSYS_NFTABLES)
635		res_id = NFNL_SUBSYS_NFTABLES;
636	else
637		res_id = ntohs(nfgenmsg->res_id);
638
639	nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
640}
641
642static void nfnetlink_rcv(struct sk_buff *skb)
643{
644	struct nlmsghdr *nlh = nlmsg_hdr(skb);
645
646	if (skb->len < NLMSG_HDRLEN ||
647	    nlh->nlmsg_len < NLMSG_HDRLEN ||
648	    skb->len < nlh->nlmsg_len)
649		return;
650
651	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
652		netlink_ack(skb, nlh, -EPERM, NULL);
653		return;
654	}
655
656	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
657		nfnetlink_rcv_skb_batch(skb, nlh);
658	else
659		netlink_rcv_skb(skb, nfnetlink_rcv_msg);
660}
661
662static void nfnetlink_bind_event(struct net *net, unsigned int group)
663{
664#ifdef CONFIG_NF_CONNTRACK_EVENTS
665	int type, group_bit;
666	u8 v;
667
668	/* All NFNLGRP_CONNTRACK_* group bits fit into u8.
669	 * The other groups are not relevant and can be ignored.
670	 */
671	if (group >= 8)
672		return;
673
674	type = nfnl_group2type[group];
675
676	switch (type) {
677	case NFNL_SUBSYS_CTNETLINK:
678		break;
679	case NFNL_SUBSYS_CTNETLINK_EXP:
680		break;
681	default:
682		return;
683	}
684
685	group_bit = (1 << group);
686
687	spin_lock(&nfnl_grp_active_lock);
688	v = READ_ONCE(nf_ctnetlink_has_listener);
689	if ((v & group_bit) == 0) {
690		v |= group_bit;
691
692		/* read concurrently without nfnl_grp_active_lock held. */
693		WRITE_ONCE(nf_ctnetlink_has_listener, v);
694	}
695
696	spin_unlock(&nfnl_grp_active_lock);
697#endif
698}
699
700static int nfnetlink_bind(struct net *net, int group)
701{
702	const struct nfnetlink_subsystem *ss;
703	int type;
704
705	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
706		return 0;
707
708	type = nfnl_group2type[group];
709
710	rcu_read_lock();
711	ss = nfnetlink_get_subsys(type << 8);
712	rcu_read_unlock();
713	if (!ss)
714		request_module_nowait("nfnetlink-subsys-%d", type);
715
716	nfnetlink_bind_event(net, group);
717	return 0;
718}
719
720static void nfnetlink_unbind(struct net *net, int group)
721{
722#ifdef CONFIG_NF_CONNTRACK_EVENTS
723	int type, group_bit;
724
725	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
726		return;
727
728	type = nfnl_group2type[group];
729
730	switch (type) {
731	case NFNL_SUBSYS_CTNETLINK:
732		break;
733	case NFNL_SUBSYS_CTNETLINK_EXP:
734		break;
735	default:
736		return;
737	}
738
739	/* ctnetlink_has_listener is u8 */
740	if (group >= 8)
741		return;
742
743	group_bit = (1 << group);
744
745	spin_lock(&nfnl_grp_active_lock);
746	if (!nfnetlink_has_listeners(net, group)) {
747		u8 v = READ_ONCE(nf_ctnetlink_has_listener);
748
749		v &= ~group_bit;
750
751		/* read concurrently without nfnl_grp_active_lock held. */
752		WRITE_ONCE(nf_ctnetlink_has_listener, v);
753	}
754	spin_unlock(&nfnl_grp_active_lock);
755#endif
756}
757
758static int __net_init nfnetlink_net_init(struct net *net)
759{
760	struct nfnl_net *nfnlnet = nfnl_pernet(net);
761	struct netlink_kernel_cfg cfg = {
762		.groups	= NFNLGRP_MAX,
763		.input	= nfnetlink_rcv,
 
764		.bind	= nfnetlink_bind,
765		.unbind	= nfnetlink_unbind,
766	};
767
768	nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
769	if (!nfnlnet->nfnl)
770		return -ENOMEM;
771	return 0;
772}
773
774static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
775{
776	struct nfnl_net *nfnlnet;
777	struct net *net;
778
779	list_for_each_entry(net, net_exit_list, exit_list) {
780		nfnlnet = nfnl_pernet(net);
781
782		netlink_kernel_release(nfnlnet->nfnl);
783	}
784}
785
786static struct pernet_operations nfnetlink_net_ops = {
787	.init		= nfnetlink_net_init,
788	.exit_batch	= nfnetlink_net_exit_batch,
789	.id		= &nfnetlink_pernet_id,
790	.size		= sizeof(struct nfnl_net),
791};
792
793static int __init nfnetlink_init(void)
794{
795	int i;
796
797	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
798		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
799
800	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
801		__mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
802
803	return register_pernet_subsys(&nfnetlink_net_ops);
804}
805
806static void __exit nfnetlink_exit(void)
807{
808	unregister_pernet_subsys(&nfnetlink_net_ops);
809}
810module_init(nfnetlink_init);
811module_exit(nfnetlink_exit);