xref: /openbmc/linux/net/netfilter/nfnetlink.c (revision a4e1d0b7) 
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 
34 MODULE_LICENSE("GPL");
35 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
36 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
37 MODULE_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 
45 static unsigned int nfnetlink_pernet_id __read_mostly;
46 
47 #ifdef CONFIG_NF_CONNTRACK_EVENTS
48 static DEFINE_SPINLOCK(nfnl_grp_active_lock);
49 #endif
50 
51 struct nfnl_net {
52 	struct sock *nfnl;
53 };
54 
55 static struct {
56 	struct mutex				mutex;
57 	const struct nfnetlink_subsystem __rcu	*subsys;
58 } table[NFNL_SUBSYS_COUNT];
59 
60 static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
61 
62 static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = {
63 	[NFNL_SUBSYS_NONE] = "nfnl_subsys_none",
64 	[NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink",
65 	[NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp",
66 	[NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue",
67 	[NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog",
68 	[NFNL_SUBSYS_OSF] = "nfnl_subsys_osf",
69 	[NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset",
70 	[NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct",
71 	[NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout",
72 	[NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper",
73 	[NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables",
74 	[NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat",
75 	[NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook",
76 };
77 
78 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
79 	[NFNLGRP_CONNTRACK_NEW]		= NFNL_SUBSYS_CTNETLINK,
80 	[NFNLGRP_CONNTRACK_UPDATE]	= NFNL_SUBSYS_CTNETLINK,
81 	[NFNLGRP_CONNTRACK_DESTROY]	= NFNL_SUBSYS_CTNETLINK,
82 	[NFNLGRP_CONNTRACK_EXP_NEW]	= NFNL_SUBSYS_CTNETLINK_EXP,
83 	[NFNLGRP_CONNTRACK_EXP_UPDATE]	= NFNL_SUBSYS_CTNETLINK_EXP,
84 	[NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
85 	[NFNLGRP_NFTABLES]		= NFNL_SUBSYS_NFTABLES,
86 	[NFNLGRP_ACCT_QUOTA]		= NFNL_SUBSYS_ACCT,
87 	[NFNLGRP_NFTRACE]		= NFNL_SUBSYS_NFTABLES,
88 };
89 
90 static struct nfnl_net *nfnl_pernet(struct net *net)
91 {
92 	return net_generic(net, nfnetlink_pernet_id);
93 }
94 
95 void nfnl_lock(__u8 subsys_id)
96 {
97 	mutex_lock(&table[subsys_id].mutex);
98 }
99 EXPORT_SYMBOL_GPL(nfnl_lock);
100 
101 void nfnl_unlock(__u8 subsys_id)
102 {
103 	mutex_unlock(&table[subsys_id].mutex);
104 }
105 EXPORT_SYMBOL_GPL(nfnl_unlock);
106 
107 #ifdef CONFIG_PROVE_LOCKING
108 bool lockdep_nfnl_is_held(u8 subsys_id)
109 {
110 	return lockdep_is_held(&table[subsys_id].mutex);
111 }
112 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
113 #endif
114 
115 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
116 {
117 	u8 cb_id;
118 
119 	/* Sanity-check attr_count size to avoid stack buffer overflow. */
120 	for (cb_id = 0; cb_id < n->cb_count; cb_id++)
121 		if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
122 			return -EINVAL;
123 
124 	nfnl_lock(n->subsys_id);
125 	if (table[n->subsys_id].subsys) {
126 		nfnl_unlock(n->subsys_id);
127 		return -EBUSY;
128 	}
129 	rcu_assign_pointer(table[n->subsys_id].subsys, n);
130 	nfnl_unlock(n->subsys_id);
131 
132 	return 0;
133 }
134 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
135 
136 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
137 {
138 	nfnl_lock(n->subsys_id);
139 	table[n->subsys_id].subsys = NULL;
140 	nfnl_unlock(n->subsys_id);
141 	synchronize_rcu();
142 	return 0;
143 }
144 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
145 
146 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
147 {
148 	u8 subsys_id = NFNL_SUBSYS_ID(type);
149 
150 	if (subsys_id >= NFNL_SUBSYS_COUNT)
151 		return NULL;
152 
153 	return rcu_dereference(table[subsys_id].subsys);
154 }
155 
156 static inline const struct nfnl_callback *
157 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
158 {
159 	u8 cb_id = NFNL_MSG_TYPE(type);
160 
161 	if (cb_id >= ss->cb_count)
162 		return NULL;
163 
164 	return &ss->cb[cb_id];
165 }
166 
167 int nfnetlink_has_listeners(struct net *net, unsigned int group)
168 {
169 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
170 
171 	return netlink_has_listeners(nfnlnet->nfnl, group);
172 }
173 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
174 
175 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
176 		   unsigned int group, int echo, gfp_t flags)
177 {
178 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
179 
180 	return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
181 }
182 EXPORT_SYMBOL_GPL(nfnetlink_send);
183 
184 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
185 {
186 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
187 
188 	return netlink_set_err(nfnlnet->nfnl, portid, group, error);
189 }
190 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
191 
192 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
193 {
194 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
195 	int err;
196 
197 	err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
198 	if (err == -EAGAIN)
199 		err = -ENOBUFS;
200 
201 	return err;
202 }
203 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
204 
205 void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
206 			 __u32 group, gfp_t allocation)
207 {
208 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
209 
210 	netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
211 }
212 EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
213 
214 /* Process one complete nfnetlink message. */
215 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
216 			     struct netlink_ext_ack *extack)
217 {
218 	struct net *net = sock_net(skb->sk);
219 	const struct nfnl_callback *nc;
220 	const struct nfnetlink_subsystem *ss;
221 	int type, err;
222 
223 	/* All the messages must at least contain nfgenmsg */
224 	if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
225 		return 0;
226 
227 	type = nlh->nlmsg_type;
228 replay:
229 	rcu_read_lock();
230 
231 	ss = nfnetlink_get_subsys(type);
232 	if (!ss) {
233 #ifdef CONFIG_MODULES
234 		rcu_read_unlock();
235 		request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
236 		rcu_read_lock();
237 		ss = nfnetlink_get_subsys(type);
238 		if (!ss)
239 #endif
240 		{
241 			rcu_read_unlock();
242 			return -EINVAL;
243 		}
244 	}
245 
246 	nc = nfnetlink_find_client(type, ss);
247 	if (!nc) {
248 		rcu_read_unlock();
249 		return -EINVAL;
250 	}
251 
252 	{
253 		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
254 		struct nfnl_net *nfnlnet = nfnl_pernet(net);
255 		u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
256 		struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
257 		struct nlattr *attr = (void *)nlh + min_len;
258 		int attrlen = nlh->nlmsg_len - min_len;
259 		__u8 subsys_id = NFNL_SUBSYS_ID(type);
260 		struct nfnl_info info = {
261 			.net	= net,
262 			.sk	= nfnlnet->nfnl,
263 			.nlh	= nlh,
264 			.nfmsg	= nlmsg_data(nlh),
265 			.extack	= extack,
266 		};
267 
268 		/* Sanity-check NFNL_MAX_ATTR_COUNT */
269 		if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
270 			rcu_read_unlock();
271 			return -ENOMEM;
272 		}
273 
274 		err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
275 					   attr, attrlen,
276 					   ss->cb[cb_id].policy, extack);
277 		if (err < 0) {
278 			rcu_read_unlock();
279 			return err;
280 		}
281 
282 		if (!nc->call) {
283 			rcu_read_unlock();
284 			return -EINVAL;
285 		}
286 
287 		switch (nc->type) {
288 		case NFNL_CB_RCU:
289 			err = nc->call(skb, &info, (const struct nlattr **)cda);
290 			rcu_read_unlock();
291 			break;
292 		case NFNL_CB_MUTEX:
293 			rcu_read_unlock();
294 			nfnl_lock(subsys_id);
295 			if (nfnl_dereference_protected(subsys_id) != ss ||
296 			    nfnetlink_find_client(type, ss) != nc) {
297 				err = -EAGAIN;
298 				break;
299 			}
300 			err = nc->call(skb, &info, (const struct nlattr **)cda);
301 			nfnl_unlock(subsys_id);
302 			break;
303 		default:
304 			rcu_read_unlock();
305 			err = -EINVAL;
306 			break;
307 		}
308 		if (err == -EAGAIN)
309 			goto replay;
310 		return err;
311 	}
312 }
313 
314 struct nfnl_err {
315 	struct list_head	head;
316 	struct nlmsghdr		*nlh;
317 	int			err;
318 	struct netlink_ext_ack	extack;
319 };
320 
321 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
322 			const struct netlink_ext_ack *extack)
323 {
324 	struct nfnl_err *nfnl_err;
325 
326 	nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
327 	if (nfnl_err == NULL)
328 		return -ENOMEM;
329 
330 	nfnl_err->nlh = nlh;
331 	nfnl_err->err = err;
332 	nfnl_err->extack = *extack;
333 	list_add_tail(&nfnl_err->head, list);
334 
335 	return 0;
336 }
337 
338 static void nfnl_err_del(struct nfnl_err *nfnl_err)
339 {
340 	list_del(&nfnl_err->head);
341 	kfree(nfnl_err);
342 }
343 
344 static void nfnl_err_reset(struct list_head *err_list)
345 {
346 	struct nfnl_err *nfnl_err, *next;
347 
348 	list_for_each_entry_safe(nfnl_err, next, err_list, head)
349 		nfnl_err_del(nfnl_err);
350 }
351 
352 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
353 {
354 	struct nfnl_err *nfnl_err, *next;
355 
356 	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
357 		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
358 			    &nfnl_err->extack);
359 		nfnl_err_del(nfnl_err);
360 	}
361 }
362 
363 enum {
364 	NFNL_BATCH_FAILURE	= (1 << 0),
365 	NFNL_BATCH_DONE		= (1 << 1),
366 	NFNL_BATCH_REPLAY	= (1 << 2),
367 };
368 
369 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
370 				u16 subsys_id, u32 genid)
371 {
372 	struct sk_buff *oskb = skb;
373 	struct net *net = sock_net(skb->sk);
374 	const struct nfnetlink_subsystem *ss;
375 	const struct nfnl_callback *nc;
376 	struct netlink_ext_ack extack;
377 	LIST_HEAD(err_list);
378 	u32 status;
379 	int err;
380 
381 	if (subsys_id >= NFNL_SUBSYS_COUNT)
382 		return netlink_ack(skb, nlh, -EINVAL, NULL);
383 replay:
384 	status = 0;
385 replay_abort:
386 	skb = netlink_skb_clone(oskb, GFP_KERNEL);
387 	if (!skb)
388 		return netlink_ack(oskb, nlh, -ENOMEM, NULL);
389 
390 	nfnl_lock(subsys_id);
391 	ss = nfnl_dereference_protected(subsys_id);
392 	if (!ss) {
393 #ifdef CONFIG_MODULES
394 		nfnl_unlock(subsys_id);
395 		request_module("nfnetlink-subsys-%d", subsys_id);
396 		nfnl_lock(subsys_id);
397 		ss = nfnl_dereference_protected(subsys_id);
398 		if (!ss)
399 #endif
400 		{
401 			nfnl_unlock(subsys_id);
402 			netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
403 			return kfree_skb(skb);
404 		}
405 	}
406 
407 	if (!ss->valid_genid || !ss->commit || !ss->abort) {
408 		nfnl_unlock(subsys_id);
409 		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
410 		return kfree_skb(skb);
411 	}
412 
413 	if (!try_module_get(ss->owner)) {
414 		nfnl_unlock(subsys_id);
415 		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
416 		return kfree_skb(skb);
417 	}
418 
419 	if (!ss->valid_genid(net, genid)) {
420 		module_put(ss->owner);
421 		nfnl_unlock(subsys_id);
422 		netlink_ack(oskb, nlh, -ERESTART, NULL);
423 		return kfree_skb(skb);
424 	}
425 
426 	nfnl_unlock(subsys_id);
427 
428 	while (skb->len >= nlmsg_total_size(0)) {
429 		int msglen, type;
430 
431 		if (fatal_signal_pending(current)) {
432 			nfnl_err_reset(&err_list);
433 			err = -EINTR;
434 			status = NFNL_BATCH_FAILURE;
435 			goto done;
436 		}
437 
438 		memset(&extack, 0, sizeof(extack));
439 		nlh = nlmsg_hdr(skb);
440 		err = 0;
441 
442 		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
443 		    skb->len < nlh->nlmsg_len ||
444 		    nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
445 			nfnl_err_reset(&err_list);
446 			status |= NFNL_BATCH_FAILURE;
447 			goto done;
448 		}
449 
450 		/* Only requests are handled by the kernel */
451 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
452 			err = -EINVAL;
453 			goto ack;
454 		}
455 
456 		type = nlh->nlmsg_type;
457 		if (type == NFNL_MSG_BATCH_BEGIN) {
458 			/* Malformed: Batch begin twice */
459 			nfnl_err_reset(&err_list);
460 			status |= NFNL_BATCH_FAILURE;
461 			goto done;
462 		} else if (type == NFNL_MSG_BATCH_END) {
463 			status |= NFNL_BATCH_DONE;
464 			goto done;
465 		} else if (type < NLMSG_MIN_TYPE) {
466 			err = -EINVAL;
467 			goto ack;
468 		}
469 
470 		/* We only accept a batch with messages for the same
471 		 * subsystem.
472 		 */
473 		if (NFNL_SUBSYS_ID(type) != subsys_id) {
474 			err = -EINVAL;
475 			goto ack;
476 		}
477 
478 		nc = nfnetlink_find_client(type, ss);
479 		if (!nc) {
480 			err = -EINVAL;
481 			goto ack;
482 		}
483 
484 		if (nc->type != NFNL_CB_BATCH) {
485 			err = -EINVAL;
486 			goto ack;
487 		}
488 
489 		{
490 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
491 			struct nfnl_net *nfnlnet = nfnl_pernet(net);
492 			struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
493 			struct nlattr *attr = (void *)nlh + min_len;
494 			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
495 			int attrlen = nlh->nlmsg_len - min_len;
496 			struct nfnl_info info = {
497 				.net	= net,
498 				.sk	= nfnlnet->nfnl,
499 				.nlh	= nlh,
500 				.nfmsg	= nlmsg_data(nlh),
501 				.extack	= &extack,
502 			};
503 
504 			/* Sanity-check NFTA_MAX_ATTR */
505 			if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
506 				err = -ENOMEM;
507 				goto ack;
508 			}
509 
510 			err = nla_parse_deprecated(cda,
511 						   ss->cb[cb_id].attr_count,
512 						   attr, attrlen,
513 						   ss->cb[cb_id].policy, NULL);
514 			if (err < 0)
515 				goto ack;
516 
517 			err = nc->call(skb, &info, (const struct nlattr **)cda);
518 
519 			/* The lock was released to autoload some module, we
520 			 * have to abort and start from scratch using the
521 			 * original skb.
522 			 */
523 			if (err == -EAGAIN) {
524 				status |= NFNL_BATCH_REPLAY;
525 				goto done;
526 			}
527 		}
528 ack:
529 		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
530 			/* Errors are delivered once the full batch has been
531 			 * processed, this avoids that the same error is
532 			 * reported several times when replaying the batch.
533 			 */
534 			if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
535 				/* We failed to enqueue an error, reset the
536 				 * list of errors and send OOM to userspace
537 				 * pointing to the batch header.
538 				 */
539 				nfnl_err_reset(&err_list);
540 				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
541 					    NULL);
542 				status |= NFNL_BATCH_FAILURE;
543 				goto done;
544 			}
545 			/* We don't stop processing the batch on errors, thus,
546 			 * userspace gets all the errors that the batch
547 			 * triggers.
548 			 */
549 			if (err)
550 				status |= NFNL_BATCH_FAILURE;
551 		}
552 
553 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
554 		if (msglen > skb->len)
555 			msglen = skb->len;
556 		skb_pull(skb, msglen);
557 	}
558 done:
559 	if (status & NFNL_BATCH_REPLAY) {
560 		ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
561 		nfnl_err_reset(&err_list);
562 		kfree_skb(skb);
563 		module_put(ss->owner);
564 		goto replay;
565 	} else if (status == NFNL_BATCH_DONE) {
566 		err = ss->commit(net, oskb);
567 		if (err == -EAGAIN) {
568 			status |= NFNL_BATCH_REPLAY;
569 			goto done;
570 		} else if (err) {
571 			ss->abort(net, oskb, NFNL_ABORT_NONE);
572 			netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
573 		}
574 	} else {
575 		enum nfnl_abort_action abort_action;
576 
577 		if (status & NFNL_BATCH_FAILURE)
578 			abort_action = NFNL_ABORT_NONE;
579 		else
580 			abort_action = NFNL_ABORT_VALIDATE;
581 
582 		err = ss->abort(net, oskb, abort_action);
583 		if (err == -EAGAIN) {
584 			nfnl_err_reset(&err_list);
585 			kfree_skb(skb);
586 			module_put(ss->owner);
587 			status |= NFNL_BATCH_FAILURE;
588 			goto replay_abort;
589 		}
590 	}
591 	if (ss->cleanup)
592 		ss->cleanup(net);
593 
594 	nfnl_err_deliver(&err_list, oskb);
595 	kfree_skb(skb);
596 	module_put(ss->owner);
597 }
598 
599 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
600 	[NFNL_BATCH_GENID]	= { .type = NLA_U32 },
601 };
602 
603 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
604 {
605 	int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
606 	struct nlattr *attr = (void *)nlh + min_len;
607 	struct nlattr *cda[NFNL_BATCH_MAX + 1];
608 	int attrlen = nlh->nlmsg_len - min_len;
609 	struct nfgenmsg *nfgenmsg;
610 	int msglen, err;
611 	u32 gen_id = 0;
612 	u16 res_id;
613 
614 	msglen = NLMSG_ALIGN(nlh->nlmsg_len);
615 	if (msglen > skb->len)
616 		msglen = skb->len;
617 
618 	if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
619 		return;
620 
621 	err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
622 				   nfnl_batch_policy, NULL);
623 	if (err < 0) {
624 		netlink_ack(skb, nlh, err, NULL);
625 		return;
626 	}
627 	if (cda[NFNL_BATCH_GENID])
628 		gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
629 
630 	nfgenmsg = nlmsg_data(nlh);
631 	skb_pull(skb, msglen);
632 	/* Work around old nft using host byte order */
633 	if (nfgenmsg->res_id == (__force __be16)NFNL_SUBSYS_NFTABLES)
634 		res_id = NFNL_SUBSYS_NFTABLES;
635 	else
636 		res_id = ntohs(nfgenmsg->res_id);
637 
638 	nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
639 }
640 
641 static void nfnetlink_rcv(struct sk_buff *skb)
642 {
643 	struct nlmsghdr *nlh = nlmsg_hdr(skb);
644 
645 	if (skb->len < NLMSG_HDRLEN ||
646 	    nlh->nlmsg_len < NLMSG_HDRLEN ||
647 	    skb->len < nlh->nlmsg_len)
648 		return;
649 
650 	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
651 		netlink_ack(skb, nlh, -EPERM, NULL);
652 		return;
653 	}
654 
655 	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
656 		nfnetlink_rcv_skb_batch(skb, nlh);
657 	else
658 		netlink_rcv_skb(skb, nfnetlink_rcv_msg);
659 }
660 
661 static void nfnetlink_bind_event(struct net *net, unsigned int group)
662 {
663 #ifdef CONFIG_NF_CONNTRACK_EVENTS
664 	int type, group_bit;
665 	u8 v;
666 
667 	/* All NFNLGRP_CONNTRACK_* group bits fit into u8.
668 	 * The other groups are not relevant and can be ignored.
669 	 */
670 	if (group >= 8)
671 		return;
672 
673 	type = nfnl_group2type[group];
674 
675 	switch (type) {
676 	case NFNL_SUBSYS_CTNETLINK:
677 		break;
678 	case NFNL_SUBSYS_CTNETLINK_EXP:
679 		break;
680 	default:
681 		return;
682 	}
683 
684 	group_bit = (1 << group);
685 
686 	spin_lock(&nfnl_grp_active_lock);
687 	v = READ_ONCE(net->ct.ctnetlink_has_listener);
688 	if ((v & group_bit) == 0) {
689 		v |= group_bit;
690 
691 		/* read concurrently without nfnl_grp_active_lock held. */
692 		WRITE_ONCE(net->ct.ctnetlink_has_listener, v);
693 	}
694 
695 	spin_unlock(&nfnl_grp_active_lock);
696 #endif
697 }
698 
699 static int nfnetlink_bind(struct net *net, int group)
700 {
701 	const struct nfnetlink_subsystem *ss;
702 	int type;
703 
704 	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
705 		return 0;
706 
707 	type = nfnl_group2type[group];
708 
709 	rcu_read_lock();
710 	ss = nfnetlink_get_subsys(type << 8);
711 	rcu_read_unlock();
712 	if (!ss)
713 		request_module_nowait("nfnetlink-subsys-%d", type);
714 
715 	nfnetlink_bind_event(net, group);
716 	return 0;
717 }
718 
719 static void nfnetlink_unbind(struct net *net, int group)
720 {
721 #ifdef CONFIG_NF_CONNTRACK_EVENTS
722 	int type, group_bit;
723 
724 	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
725 		return;
726 
727 	type = nfnl_group2type[group];
728 
729 	switch (type) {
730 	case NFNL_SUBSYS_CTNETLINK:
731 		break;
732 	case NFNL_SUBSYS_CTNETLINK_EXP:
733 		break;
734 	default:
735 		return;
736 	}
737 
738 	/* ctnetlink_has_listener is u8 */
739 	if (group >= 8)
740 		return;
741 
742 	group_bit = (1 << group);
743 
744 	spin_lock(&nfnl_grp_active_lock);
745 	if (!nfnetlink_has_listeners(net, group)) {
746 		u8 v = READ_ONCE(net->ct.ctnetlink_has_listener);
747 
748 		v &= ~group_bit;
749 
750 		/* read concurrently without nfnl_grp_active_lock held. */
751 		WRITE_ONCE(net->ct.ctnetlink_has_listener, v);
752 	}
753 	spin_unlock(&nfnl_grp_active_lock);
754 #endif
755 }
756 
757 static int __net_init nfnetlink_net_init(struct net *net)
758 {
759 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
760 	struct netlink_kernel_cfg cfg = {
761 		.groups	= NFNLGRP_MAX,
762 		.input	= nfnetlink_rcv,
763 		.bind	= nfnetlink_bind,
764 		.unbind	= nfnetlink_unbind,
765 	};
766 
767 	nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
768 	if (!nfnlnet->nfnl)
769 		return -ENOMEM;
770 	return 0;
771 }
772 
773 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
774 {
775 	struct nfnl_net *nfnlnet;
776 	struct net *net;
777 
778 	list_for_each_entry(net, net_exit_list, exit_list) {
779 		nfnlnet = nfnl_pernet(net);
780 
781 		netlink_kernel_release(nfnlnet->nfnl);
782 	}
783 }
784 
785 static struct pernet_operations nfnetlink_net_ops = {
786 	.init		= nfnetlink_net_init,
787 	.exit_batch	= nfnetlink_net_exit_batch,
788 	.id		= &nfnetlink_pernet_id,
789 	.size		= sizeof(struct nfnl_net),
790 };
791 
792 static int __init nfnetlink_init(void)
793 {
794 	int i;
795 
796 	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
797 		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
798 
799 	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
800 		__mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
801 
802 	return register_pernet_subsys(&nfnetlink_net_ops);
803 }
804 
805 static void __exit nfnetlink_exit(void)
806 {
807 	unregister_pernet_subsys(&nfnetlink_net_ops);
808 }
809 module_init(nfnetlink_init);
810 module_exit(nfnetlink_exit);
811