xref: /openbmc/linux/net/netfilter/nfnetlink.c (revision 9b12f050) 
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 
35 MODULE_LICENSE("GPL");
36 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
37 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
38 MODULE_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 
46 static unsigned int nfnetlink_pernet_id __read_mostly;
47 
48 #ifdef CONFIG_NF_CONNTRACK_EVENTS
49 static DEFINE_SPINLOCK(nfnl_grp_active_lock);
50 #endif
51 
52 struct nfnl_net {
53 	struct sock *nfnl;
54 };
55 
56 static struct {
57 	struct mutex				mutex;
58 	const struct nfnetlink_subsystem __rcu	*subsys;
59 } table[NFNL_SUBSYS_COUNT];
60 
61 static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
62 
63 static 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 
79 static 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 
91 static struct nfnl_net *nfnl_pernet(struct net *net)
92 {
93 	return net_generic(net, nfnetlink_pernet_id);
94 }
95 
96 void nfnl_lock(__u8 subsys_id)
97 {
98 	mutex_lock(&table[subsys_id].mutex);
99 }
100 EXPORT_SYMBOL_GPL(nfnl_lock);
101 
102 void nfnl_unlock(__u8 subsys_id)
103 {
104 	mutex_unlock(&table[subsys_id].mutex);
105 }
106 EXPORT_SYMBOL_GPL(nfnl_unlock);
107 
108 #ifdef CONFIG_PROVE_LOCKING
109 bool lockdep_nfnl_is_held(u8 subsys_id)
110 {
111 	return lockdep_is_held(&table[subsys_id].mutex);
112 }
113 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
114 #endif
115 
116 int 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 }
135 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
136 
137 int 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 }
145 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
146 
147 static 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 
157 static inline const struct nfnl_callback *
158 nfnetlink_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 
168 int 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 }
174 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
175 
176 int 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 }
183 EXPORT_SYMBOL_GPL(nfnetlink_send);
184 
185 int 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 }
191 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
192 
193 int 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 }
204 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
205 
206 void 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 }
213 EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
214 
215 /* Process one complete nfnetlink message. */
216 static 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;
229 replay:
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 
316 struct nfnl_err {
317 	struct list_head	head;
318 	struct nlmsghdr		*nlh;
319 	int			err;
320 	struct netlink_ext_ack	extack;
321 };
322 
323 static 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 
340 static void nfnl_err_del(struct nfnl_err *nfnl_err)
341 {
342 	list_del(&nfnl_err->head);
343 	kfree(nfnl_err);
344 }
345 
346 static 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 
354 static 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 
365 enum {
366 	NFNL_BATCH_FAILURE	= (1 << 0),
367 	NFNL_BATCH_DONE		= (1 << 1),
368 	NFNL_BATCH_REPLAY	= (1 << 2),
369 };
370 
371 static 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);
385 replay:
386 	status = 0;
387 replay_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 		}
530 ack:
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 (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
537 				/* We failed to enqueue an error, reset the
538 				 * list of errors and send OOM to userspace
539 				 * pointing to the batch header.
540 				 */
541 				nfnl_err_reset(&err_list);
542 				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
543 					    NULL);
544 				status |= NFNL_BATCH_FAILURE;
545 				goto done;
546 			}
547 			/* We don't stop processing the batch on errors, thus,
548 			 * userspace gets all the errors that the batch
549 			 * triggers.
550 			 */
551 			if (err)
552 				status |= NFNL_BATCH_FAILURE;
553 		}
554 
555 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
556 		if (msglen > skb->len)
557 			msglen = skb->len;
558 		skb_pull(skb, msglen);
559 	}
560 done:
561 	if (status & NFNL_BATCH_REPLAY) {
562 		ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
563 		nfnl_err_reset(&err_list);
564 		kfree_skb(skb);
565 		module_put(ss->owner);
566 		goto replay;
567 	} else if (status == NFNL_BATCH_DONE) {
568 		err = ss->commit(net, oskb);
569 		if (err == -EAGAIN) {
570 			status |= NFNL_BATCH_REPLAY;
571 			goto done;
572 		} else if (err) {
573 			ss->abort(net, oskb, NFNL_ABORT_NONE);
574 			netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
575 		}
576 	} else {
577 		enum nfnl_abort_action abort_action;
578 
579 		if (status & NFNL_BATCH_FAILURE)
580 			abort_action = NFNL_ABORT_NONE;
581 		else
582 			abort_action = NFNL_ABORT_VALIDATE;
583 
584 		err = ss->abort(net, oskb, abort_action);
585 		if (err == -EAGAIN) {
586 			nfnl_err_reset(&err_list);
587 			kfree_skb(skb);
588 			module_put(ss->owner);
589 			status |= NFNL_BATCH_FAILURE;
590 			goto replay_abort;
591 		}
592 	}
593 	if (ss->cleanup)
594 		ss->cleanup(net);
595 
596 	nfnl_err_deliver(&err_list, oskb);
597 	kfree_skb(skb);
598 	module_put(ss->owner);
599 }
600 
601 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
602 	[NFNL_BATCH_GENID]	= { .type = NLA_U32 },
603 };
604 
605 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
606 {
607 	int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
608 	struct nlattr *attr = (void *)nlh + min_len;
609 	struct nlattr *cda[NFNL_BATCH_MAX + 1];
610 	int attrlen = nlh->nlmsg_len - min_len;
611 	struct nfgenmsg *nfgenmsg;
612 	int msglen, err;
613 	u32 gen_id = 0;
614 	u16 res_id;
615 
616 	msglen = NLMSG_ALIGN(nlh->nlmsg_len);
617 	if (msglen > skb->len)
618 		msglen = skb->len;
619 
620 	if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
621 		return;
622 
623 	err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
624 				   nfnl_batch_policy, NULL);
625 	if (err < 0) {
626 		netlink_ack(skb, nlh, err, NULL);
627 		return;
628 	}
629 	if (cda[NFNL_BATCH_GENID])
630 		gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
631 
632 	nfgenmsg = nlmsg_data(nlh);
633 	skb_pull(skb, msglen);
634 	/* Work around old nft using host byte order */
635 	if (nfgenmsg->res_id == (__force __be16)NFNL_SUBSYS_NFTABLES)
636 		res_id = NFNL_SUBSYS_NFTABLES;
637 	else
638 		res_id = ntohs(nfgenmsg->res_id);
639 
640 	nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
641 }
642 
643 static void nfnetlink_rcv(struct sk_buff *skb)
644 {
645 	struct nlmsghdr *nlh = nlmsg_hdr(skb);
646 
647 	if (skb->len < NLMSG_HDRLEN ||
648 	    nlh->nlmsg_len < NLMSG_HDRLEN ||
649 	    skb->len < nlh->nlmsg_len)
650 		return;
651 
652 	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
653 		netlink_ack(skb, nlh, -EPERM, NULL);
654 		return;
655 	}
656 
657 	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
658 		nfnetlink_rcv_skb_batch(skb, nlh);
659 	else
660 		netlink_rcv_skb(skb, nfnetlink_rcv_msg);
661 }
662 
663 static void nfnetlink_bind_event(struct net *net, unsigned int group)
664 {
665 #ifdef CONFIG_NF_CONNTRACK_EVENTS
666 	int type, group_bit;
667 	u8 v;
668 
669 	/* All NFNLGRP_CONNTRACK_* group bits fit into u8.
670 	 * The other groups are not relevant and can be ignored.
671 	 */
672 	if (group >= 8)
673 		return;
674 
675 	type = nfnl_group2type[group];
676 
677 	switch (type) {
678 	case NFNL_SUBSYS_CTNETLINK:
679 		break;
680 	case NFNL_SUBSYS_CTNETLINK_EXP:
681 		break;
682 	default:
683 		return;
684 	}
685 
686 	group_bit = (1 << group);
687 
688 	spin_lock(&nfnl_grp_active_lock);
689 	v = READ_ONCE(nf_ctnetlink_has_listener);
690 	if ((v & group_bit) == 0) {
691 		v |= group_bit;
692 
693 		/* read concurrently without nfnl_grp_active_lock held. */
694 		WRITE_ONCE(nf_ctnetlink_has_listener, v);
695 	}
696 
697 	spin_unlock(&nfnl_grp_active_lock);
698 #endif
699 }
700 
701 static int nfnetlink_bind(struct net *net, int group)
702 {
703 	const struct nfnetlink_subsystem *ss;
704 	int type;
705 
706 	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
707 		return 0;
708 
709 	type = nfnl_group2type[group];
710 
711 	rcu_read_lock();
712 	ss = nfnetlink_get_subsys(type << 8);
713 	rcu_read_unlock();
714 	if (!ss)
715 		request_module_nowait("nfnetlink-subsys-%d", type);
716 
717 	nfnetlink_bind_event(net, group);
718 	return 0;
719 }
720 
721 static void nfnetlink_unbind(struct net *net, int group)
722 {
723 #ifdef CONFIG_NF_CONNTRACK_EVENTS
724 	int type, group_bit;
725 
726 	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
727 		return;
728 
729 	type = nfnl_group2type[group];
730 
731 	switch (type) {
732 	case NFNL_SUBSYS_CTNETLINK:
733 		break;
734 	case NFNL_SUBSYS_CTNETLINK_EXP:
735 		break;
736 	default:
737 		return;
738 	}
739 
740 	/* ctnetlink_has_listener is u8 */
741 	if (group >= 8)
742 		return;
743 
744 	group_bit = (1 << group);
745 
746 	spin_lock(&nfnl_grp_active_lock);
747 	if (!nfnetlink_has_listeners(net, group)) {
748 		u8 v = READ_ONCE(nf_ctnetlink_has_listener);
749 
750 		v &= ~group_bit;
751 
752 		/* read concurrently without nfnl_grp_active_lock held. */
753 		WRITE_ONCE(nf_ctnetlink_has_listener, v);
754 	}
755 	spin_unlock(&nfnl_grp_active_lock);
756 #endif
757 }
758 
759 static int __net_init nfnetlink_net_init(struct net *net)
760 {
761 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
762 	struct netlink_kernel_cfg cfg = {
763 		.groups	= NFNLGRP_MAX,
764 		.input	= nfnetlink_rcv,
765 		.bind	= nfnetlink_bind,
766 		.unbind	= nfnetlink_unbind,
767 	};
768 
769 	nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
770 	if (!nfnlnet->nfnl)
771 		return -ENOMEM;
772 	return 0;
773 }
774 
775 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
776 {
777 	struct nfnl_net *nfnlnet;
778 	struct net *net;
779 
780 	list_for_each_entry(net, net_exit_list, exit_list) {
781 		nfnlnet = nfnl_pernet(net);
782 
783 		netlink_kernel_release(nfnlnet->nfnl);
784 	}
785 }
786 
787 static struct pernet_operations nfnetlink_net_ops = {
788 	.init		= nfnetlink_net_init,
789 	.exit_batch	= nfnetlink_net_exit_batch,
790 	.id		= &nfnetlink_pernet_id,
791 	.size		= sizeof(struct nfnl_net),
792 };
793 
794 static int __init nfnetlink_init(void)
795 {
796 	int i;
797 
798 	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
799 		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
800 
801 	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
802 		__mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
803 
804 	return register_pernet_subsys(&nfnetlink_net_ops);
805 }
806 
807 static void __exit nfnetlink_exit(void)
808 {
809 	unregister_pernet_subsys(&nfnetlink_net_ops);
810 }
811 module_init(nfnetlink_init);
812 module_exit(nfnetlink_exit);
813