xref: /openbmc/linux/net/netfilter/nfnetlink.c (revision 2794cdb0) 
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 struct nfnl_net {
48 	unsigned int ctnetlink_listeners;
49 	struct sock *nfnl;
50 };
51 
52 static struct {
53 	struct mutex				mutex;
54 	const struct nfnetlink_subsystem __rcu	*subsys;
55 } table[NFNL_SUBSYS_COUNT];
56 
57 static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
58 
59 static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = {
60 	[NFNL_SUBSYS_NONE] = "nfnl_subsys_none",
61 	[NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink",
62 	[NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp",
63 	[NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue",
64 	[NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog",
65 	[NFNL_SUBSYS_OSF] = "nfnl_subsys_osf",
66 	[NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset",
67 	[NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct",
68 	[NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout",
69 	[NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper",
70 	[NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables",
71 	[NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat",
72 	[NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook",
73 };
74 
75 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
76 	[NFNLGRP_CONNTRACK_NEW]		= NFNL_SUBSYS_CTNETLINK,
77 	[NFNLGRP_CONNTRACK_UPDATE]	= NFNL_SUBSYS_CTNETLINK,
78 	[NFNLGRP_CONNTRACK_DESTROY]	= NFNL_SUBSYS_CTNETLINK,
79 	[NFNLGRP_CONNTRACK_EXP_NEW]	= NFNL_SUBSYS_CTNETLINK_EXP,
80 	[NFNLGRP_CONNTRACK_EXP_UPDATE]	= NFNL_SUBSYS_CTNETLINK_EXP,
81 	[NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
82 	[NFNLGRP_NFTABLES]		= NFNL_SUBSYS_NFTABLES,
83 	[NFNLGRP_ACCT_QUOTA]		= NFNL_SUBSYS_ACCT,
84 	[NFNLGRP_NFTRACE]		= NFNL_SUBSYS_NFTABLES,
85 };
86 
87 static struct nfnl_net *nfnl_pernet(struct net *net)
88 {
89 	return net_generic(net, nfnetlink_pernet_id);
90 }
91 
92 void nfnl_lock(__u8 subsys_id)
93 {
94 	mutex_lock(&table[subsys_id].mutex);
95 }
96 EXPORT_SYMBOL_GPL(nfnl_lock);
97 
98 void nfnl_unlock(__u8 subsys_id)
99 {
100 	mutex_unlock(&table[subsys_id].mutex);
101 }
102 EXPORT_SYMBOL_GPL(nfnl_unlock);
103 
104 #ifdef CONFIG_PROVE_LOCKING
105 bool lockdep_nfnl_is_held(u8 subsys_id)
106 {
107 	return lockdep_is_held(&table[subsys_id].mutex);
108 }
109 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
110 #endif
111 
112 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
113 {
114 	u8 cb_id;
115 
116 	/* Sanity-check attr_count size to avoid stack buffer overflow. */
117 	for (cb_id = 0; cb_id < n->cb_count; cb_id++)
118 		if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
119 			return -EINVAL;
120 
121 	nfnl_lock(n->subsys_id);
122 	if (table[n->subsys_id].subsys) {
123 		nfnl_unlock(n->subsys_id);
124 		return -EBUSY;
125 	}
126 	rcu_assign_pointer(table[n->subsys_id].subsys, n);
127 	nfnl_unlock(n->subsys_id);
128 
129 	return 0;
130 }
131 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
132 
133 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
134 {
135 	nfnl_lock(n->subsys_id);
136 	table[n->subsys_id].subsys = NULL;
137 	nfnl_unlock(n->subsys_id);
138 	synchronize_rcu();
139 	return 0;
140 }
141 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
142 
143 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
144 {
145 	u8 subsys_id = NFNL_SUBSYS_ID(type);
146 
147 	if (subsys_id >= NFNL_SUBSYS_COUNT)
148 		return NULL;
149 
150 	return rcu_dereference(table[subsys_id].subsys);
151 }
152 
153 static inline const struct nfnl_callback *
154 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
155 {
156 	u8 cb_id = NFNL_MSG_TYPE(type);
157 
158 	if (cb_id >= ss->cb_count)
159 		return NULL;
160 
161 	return &ss->cb[cb_id];
162 }
163 
164 int nfnetlink_has_listeners(struct net *net, unsigned int group)
165 {
166 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
167 
168 	return netlink_has_listeners(nfnlnet->nfnl, group);
169 }
170 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
171 
172 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
173 		   unsigned int group, int echo, gfp_t flags)
174 {
175 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
176 
177 	return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
178 }
179 EXPORT_SYMBOL_GPL(nfnetlink_send);
180 
181 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
182 {
183 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
184 
185 	return netlink_set_err(nfnlnet->nfnl, portid, group, error);
186 }
187 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
188 
189 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
190 {
191 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
192 	int err;
193 
194 	err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
195 	if (err == -EAGAIN)
196 		err = -ENOBUFS;
197 
198 	return err;
199 }
200 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
201 
202 void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
203 			 __u32 group, gfp_t allocation)
204 {
205 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
206 
207 	netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
208 }
209 EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
210 
211 /* Process one complete nfnetlink message. */
212 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
213 			     struct netlink_ext_ack *extack)
214 {
215 	struct net *net = sock_net(skb->sk);
216 	const struct nfnl_callback *nc;
217 	const struct nfnetlink_subsystem *ss;
218 	int type, err;
219 
220 	/* All the messages must at least contain nfgenmsg */
221 	if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
222 		return 0;
223 
224 	type = nlh->nlmsg_type;
225 replay:
226 	rcu_read_lock();
227 
228 	ss = nfnetlink_get_subsys(type);
229 	if (!ss) {
230 #ifdef CONFIG_MODULES
231 		rcu_read_unlock();
232 		request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
233 		rcu_read_lock();
234 		ss = nfnetlink_get_subsys(type);
235 		if (!ss)
236 #endif
237 		{
238 			rcu_read_unlock();
239 			return -EINVAL;
240 		}
241 	}
242 
243 	nc = nfnetlink_find_client(type, ss);
244 	if (!nc) {
245 		rcu_read_unlock();
246 		return -EINVAL;
247 	}
248 
249 	{
250 		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
251 		struct nfnl_net *nfnlnet = nfnl_pernet(net);
252 		u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
253 		struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
254 		struct nlattr *attr = (void *)nlh + min_len;
255 		int attrlen = nlh->nlmsg_len - min_len;
256 		__u8 subsys_id = NFNL_SUBSYS_ID(type);
257 		struct nfnl_info info = {
258 			.net	= net,
259 			.sk	= nfnlnet->nfnl,
260 			.nlh	= nlh,
261 			.nfmsg	= nlmsg_data(nlh),
262 			.extack	= extack,
263 		};
264 
265 		/* Sanity-check NFNL_MAX_ATTR_COUNT */
266 		if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
267 			rcu_read_unlock();
268 			return -ENOMEM;
269 		}
270 
271 		err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
272 					   attr, attrlen,
273 					   ss->cb[cb_id].policy, extack);
274 		if (err < 0) {
275 			rcu_read_unlock();
276 			return err;
277 		}
278 
279 		if (!nc->call) {
280 			rcu_read_unlock();
281 			return -EINVAL;
282 		}
283 
284 		switch (nc->type) {
285 		case NFNL_CB_RCU:
286 			err = nc->call(skb, &info, (const struct nlattr **)cda);
287 			rcu_read_unlock();
288 			break;
289 		case NFNL_CB_MUTEX:
290 			rcu_read_unlock();
291 			nfnl_lock(subsys_id);
292 			if (nfnl_dereference_protected(subsys_id) != ss ||
293 			    nfnetlink_find_client(type, ss) != nc) {
294 				err = -EAGAIN;
295 				break;
296 			}
297 			err = nc->call(skb, &info, (const struct nlattr **)cda);
298 			nfnl_unlock(subsys_id);
299 			break;
300 		default:
301 			rcu_read_unlock();
302 			err = -EINVAL;
303 			break;
304 		}
305 		if (err == -EAGAIN)
306 			goto replay;
307 		return err;
308 	}
309 }
310 
311 struct nfnl_err {
312 	struct list_head	head;
313 	struct nlmsghdr		*nlh;
314 	int			err;
315 	struct netlink_ext_ack	extack;
316 };
317 
318 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
319 			const struct netlink_ext_ack *extack)
320 {
321 	struct nfnl_err *nfnl_err;
322 
323 	nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
324 	if (nfnl_err == NULL)
325 		return -ENOMEM;
326 
327 	nfnl_err->nlh = nlh;
328 	nfnl_err->err = err;
329 	nfnl_err->extack = *extack;
330 	list_add_tail(&nfnl_err->head, list);
331 
332 	return 0;
333 }
334 
335 static void nfnl_err_del(struct nfnl_err *nfnl_err)
336 {
337 	list_del(&nfnl_err->head);
338 	kfree(nfnl_err);
339 }
340 
341 static void nfnl_err_reset(struct list_head *err_list)
342 {
343 	struct nfnl_err *nfnl_err, *next;
344 
345 	list_for_each_entry_safe(nfnl_err, next, err_list, head)
346 		nfnl_err_del(nfnl_err);
347 }
348 
349 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
350 {
351 	struct nfnl_err *nfnl_err, *next;
352 
353 	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
354 		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
355 			    &nfnl_err->extack);
356 		nfnl_err_del(nfnl_err);
357 	}
358 }
359 
360 enum {
361 	NFNL_BATCH_FAILURE	= (1 << 0),
362 	NFNL_BATCH_DONE		= (1 << 1),
363 	NFNL_BATCH_REPLAY	= (1 << 2),
364 };
365 
366 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
367 				u16 subsys_id, u32 genid)
368 {
369 	struct sk_buff *oskb = skb;
370 	struct net *net = sock_net(skb->sk);
371 	const struct nfnetlink_subsystem *ss;
372 	const struct nfnl_callback *nc;
373 	struct netlink_ext_ack extack;
374 	LIST_HEAD(err_list);
375 	u32 status;
376 	int err;
377 
378 	if (subsys_id >= NFNL_SUBSYS_COUNT)
379 		return netlink_ack(skb, nlh, -EINVAL, NULL);
380 replay:
381 	status = 0;
382 replay_abort:
383 	skb = netlink_skb_clone(oskb, GFP_KERNEL);
384 	if (!skb)
385 		return netlink_ack(oskb, nlh, -ENOMEM, NULL);
386 
387 	nfnl_lock(subsys_id);
388 	ss = nfnl_dereference_protected(subsys_id);
389 	if (!ss) {
390 #ifdef CONFIG_MODULES
391 		nfnl_unlock(subsys_id);
392 		request_module("nfnetlink-subsys-%d", subsys_id);
393 		nfnl_lock(subsys_id);
394 		ss = nfnl_dereference_protected(subsys_id);
395 		if (!ss)
396 #endif
397 		{
398 			nfnl_unlock(subsys_id);
399 			netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
400 			return kfree_skb(skb);
401 		}
402 	}
403 
404 	if (!ss->valid_genid || !ss->commit || !ss->abort) {
405 		nfnl_unlock(subsys_id);
406 		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
407 		return kfree_skb(skb);
408 	}
409 
410 	if (!try_module_get(ss->owner)) {
411 		nfnl_unlock(subsys_id);
412 		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
413 		return kfree_skb(skb);
414 	}
415 
416 	if (!ss->valid_genid(net, genid)) {
417 		module_put(ss->owner);
418 		nfnl_unlock(subsys_id);
419 		netlink_ack(oskb, nlh, -ERESTART, NULL);
420 		return kfree_skb(skb);
421 	}
422 
423 	nfnl_unlock(subsys_id);
424 
425 	while (skb->len >= nlmsg_total_size(0)) {
426 		int msglen, type;
427 
428 		if (fatal_signal_pending(current)) {
429 			nfnl_err_reset(&err_list);
430 			err = -EINTR;
431 			status = NFNL_BATCH_FAILURE;
432 			goto done;
433 		}
434 
435 		memset(&extack, 0, sizeof(extack));
436 		nlh = nlmsg_hdr(skb);
437 		err = 0;
438 
439 		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
440 		    skb->len < nlh->nlmsg_len ||
441 		    nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
442 			nfnl_err_reset(&err_list);
443 			status |= NFNL_BATCH_FAILURE;
444 			goto done;
445 		}
446 
447 		/* Only requests are handled by the kernel */
448 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
449 			err = -EINVAL;
450 			goto ack;
451 		}
452 
453 		type = nlh->nlmsg_type;
454 		if (type == NFNL_MSG_BATCH_BEGIN) {
455 			/* Malformed: Batch begin twice */
456 			nfnl_err_reset(&err_list);
457 			status |= NFNL_BATCH_FAILURE;
458 			goto done;
459 		} else if (type == NFNL_MSG_BATCH_END) {
460 			status |= NFNL_BATCH_DONE;
461 			goto done;
462 		} else if (type < NLMSG_MIN_TYPE) {
463 			err = -EINVAL;
464 			goto ack;
465 		}
466 
467 		/* We only accept a batch with messages for the same
468 		 * subsystem.
469 		 */
470 		if (NFNL_SUBSYS_ID(type) != subsys_id) {
471 			err = -EINVAL;
472 			goto ack;
473 		}
474 
475 		nc = nfnetlink_find_client(type, ss);
476 		if (!nc) {
477 			err = -EINVAL;
478 			goto ack;
479 		}
480 
481 		if (nc->type != NFNL_CB_BATCH) {
482 			err = -EINVAL;
483 			goto ack;
484 		}
485 
486 		{
487 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
488 			struct nfnl_net *nfnlnet = nfnl_pernet(net);
489 			struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
490 			struct nlattr *attr = (void *)nlh + min_len;
491 			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
492 			int attrlen = nlh->nlmsg_len - min_len;
493 			struct nfnl_info info = {
494 				.net	= net,
495 				.sk	= nfnlnet->nfnl,
496 				.nlh	= nlh,
497 				.nfmsg	= nlmsg_data(nlh),
498 				.extack	= &extack,
499 			};
500 
501 			/* Sanity-check NFTA_MAX_ATTR */
502 			if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
503 				err = -ENOMEM;
504 				goto ack;
505 			}
506 
507 			err = nla_parse_deprecated(cda,
508 						   ss->cb[cb_id].attr_count,
509 						   attr, attrlen,
510 						   ss->cb[cb_id].policy, NULL);
511 			if (err < 0)
512 				goto ack;
513 
514 			err = nc->call(skb, &info, (const struct nlattr **)cda);
515 
516 			/* The lock was released to autoload some module, we
517 			 * have to abort and start from scratch using the
518 			 * original skb.
519 			 */
520 			if (err == -EAGAIN) {
521 				status |= NFNL_BATCH_REPLAY;
522 				goto done;
523 			}
524 		}
525 ack:
526 		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
527 			/* Errors are delivered once the full batch has been
528 			 * processed, this avoids that the same error is
529 			 * reported several times when replaying the batch.
530 			 */
531 			if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
532 				/* We failed to enqueue an error, reset the
533 				 * list of errors and send OOM to userspace
534 				 * pointing to the batch header.
535 				 */
536 				nfnl_err_reset(&err_list);
537 				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
538 					    NULL);
539 				status |= NFNL_BATCH_FAILURE;
540 				goto done;
541 			}
542 			/* We don't stop processing the batch on errors, thus,
543 			 * userspace gets all the errors that the batch
544 			 * triggers.
545 			 */
546 			if (err)
547 				status |= NFNL_BATCH_FAILURE;
548 		}
549 
550 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
551 		if (msglen > skb->len)
552 			msglen = skb->len;
553 		skb_pull(skb, msglen);
554 	}
555 done:
556 	if (status & NFNL_BATCH_REPLAY) {
557 		ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
558 		nfnl_err_reset(&err_list);
559 		kfree_skb(skb);
560 		module_put(ss->owner);
561 		goto replay;
562 	} else if (status == NFNL_BATCH_DONE) {
563 		err = ss->commit(net, oskb);
564 		if (err == -EAGAIN) {
565 			status |= NFNL_BATCH_REPLAY;
566 			goto done;
567 		} else if (err) {
568 			ss->abort(net, oskb, NFNL_ABORT_NONE);
569 			netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
570 		}
571 	} else {
572 		enum nfnl_abort_action abort_action;
573 
574 		if (status & NFNL_BATCH_FAILURE)
575 			abort_action = NFNL_ABORT_NONE;
576 		else
577 			abort_action = NFNL_ABORT_VALIDATE;
578 
579 		err = ss->abort(net, oskb, abort_action);
580 		if (err == -EAGAIN) {
581 			nfnl_err_reset(&err_list);
582 			kfree_skb(skb);
583 			module_put(ss->owner);
584 			status |= NFNL_BATCH_FAILURE;
585 			goto replay_abort;
586 		}
587 	}
588 	if (ss->cleanup)
589 		ss->cleanup(net);
590 
591 	nfnl_err_deliver(&err_list, oskb);
592 	kfree_skb(skb);
593 	module_put(ss->owner);
594 }
595 
596 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
597 	[NFNL_BATCH_GENID]	= { .type = NLA_U32 },
598 };
599 
600 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
601 {
602 	int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
603 	struct nlattr *attr = (void *)nlh + min_len;
604 	struct nlattr *cda[NFNL_BATCH_MAX + 1];
605 	int attrlen = nlh->nlmsg_len - min_len;
606 	struct nfgenmsg *nfgenmsg;
607 	int msglen, err;
608 	u32 gen_id = 0;
609 	u16 res_id;
610 
611 	msglen = NLMSG_ALIGN(nlh->nlmsg_len);
612 	if (msglen > skb->len)
613 		msglen = skb->len;
614 
615 	if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
616 		return;
617 
618 	err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
619 				   nfnl_batch_policy, NULL);
620 	if (err < 0) {
621 		netlink_ack(skb, nlh, err, NULL);
622 		return;
623 	}
624 	if (cda[NFNL_BATCH_GENID])
625 		gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
626 
627 	nfgenmsg = nlmsg_data(nlh);
628 	skb_pull(skb, msglen);
629 	/* Work around old nft using host byte order */
630 	if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
631 		res_id = NFNL_SUBSYS_NFTABLES;
632 	else
633 		res_id = ntohs(nfgenmsg->res_id);
634 
635 	nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
636 }
637 
638 static void nfnetlink_rcv(struct sk_buff *skb)
639 {
640 	struct nlmsghdr *nlh = nlmsg_hdr(skb);
641 
642 	if (skb->len < NLMSG_HDRLEN ||
643 	    nlh->nlmsg_len < NLMSG_HDRLEN ||
644 	    skb->len < nlh->nlmsg_len)
645 		return;
646 
647 	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
648 		netlink_ack(skb, nlh, -EPERM, NULL);
649 		return;
650 	}
651 
652 	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
653 		nfnetlink_rcv_skb_batch(skb, nlh);
654 	else
655 		netlink_rcv_skb(skb, nfnetlink_rcv_msg);
656 }
657 
658 static 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 
674 #ifdef CONFIG_NF_CONNTRACK_EVENTS
675 	if (type == NFNL_SUBSYS_CTNETLINK) {
676 		struct nfnl_net *nfnlnet = nfnl_pernet(net);
677 
678 		nfnl_lock(NFNL_SUBSYS_CTNETLINK);
679 
680 		if (WARN_ON_ONCE(nfnlnet->ctnetlink_listeners == UINT_MAX)) {
681 			nfnl_unlock(NFNL_SUBSYS_CTNETLINK);
682 			return -EOVERFLOW;
683 		}
684 
685 		nfnlnet->ctnetlink_listeners++;
686 		if (nfnlnet->ctnetlink_listeners == 1)
687 			WRITE_ONCE(net->ct.ctnetlink_has_listener, true);
688 		nfnl_unlock(NFNL_SUBSYS_CTNETLINK);
689 	}
690 #endif
691 	return 0;
692 }
693 
694 static void nfnetlink_unbind(struct net *net, int group)
695 {
696 #ifdef CONFIG_NF_CONNTRACK_EVENTS
697 	int type = nfnl_group2type[group];
698 
699 	if (type == NFNL_SUBSYS_CTNETLINK) {
700 		struct nfnl_net *nfnlnet = nfnl_pernet(net);
701 
702 		nfnl_lock(NFNL_SUBSYS_CTNETLINK);
703 		WARN_ON_ONCE(nfnlnet->ctnetlink_listeners == 0);
704 		nfnlnet->ctnetlink_listeners--;
705 		if (nfnlnet->ctnetlink_listeners == 0)
706 			WRITE_ONCE(net->ct.ctnetlink_has_listener, false);
707 		nfnl_unlock(NFNL_SUBSYS_CTNETLINK);
708 	}
709 #endif
710 }
711 
712 static int __net_init nfnetlink_net_init(struct net *net)
713 {
714 	struct nfnl_net *nfnlnet = nfnl_pernet(net);
715 	struct netlink_kernel_cfg cfg = {
716 		.groups	= NFNLGRP_MAX,
717 		.input	= nfnetlink_rcv,
718 		.bind	= nfnetlink_bind,
719 		.unbind	= nfnetlink_unbind,
720 	};
721 
722 	nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
723 	if (!nfnlnet->nfnl)
724 		return -ENOMEM;
725 	return 0;
726 }
727 
728 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
729 {
730 	struct nfnl_net *nfnlnet;
731 	struct net *net;
732 
733 	list_for_each_entry(net, net_exit_list, exit_list) {
734 		nfnlnet = nfnl_pernet(net);
735 
736 		netlink_kernel_release(nfnlnet->nfnl);
737 	}
738 }
739 
740 static struct pernet_operations nfnetlink_net_ops = {
741 	.init		= nfnetlink_net_init,
742 	.exit_batch	= nfnetlink_net_exit_batch,
743 	.id		= &nfnetlink_pernet_id,
744 	.size		= sizeof(struct nfnl_net),
745 };
746 
747 static int __init nfnetlink_init(void)
748 {
749 	int i;
750 
751 	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
752 		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
753 
754 	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
755 		__mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
756 
757 	return register_pernet_subsys(&nfnetlink_net_ops);
758 }
759 
760 static void __exit nfnetlink_exit(void)
761 {
762 	unregister_pernet_subsys(&nfnetlink_net_ops);
763 }
764 module_init(nfnetlink_init);
765 module_exit(nfnetlink_exit);
766