xref: /openbmc/linux/net/netfilter/nfnetlink.c (revision f0009360)
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 				nfnl_unlock(subsys_id);
298 				err = -EAGAIN;
299 				break;
300 			}
301 			err = nc->call(skb, &info, (const struct nlattr **)cda);
302 			nfnl_unlock(subsys_id);
303 			break;
304 		default:
305 			rcu_read_unlock();
306 			err = -EINVAL;
307 			break;
308 		}
309 		if (err == -EAGAIN)
310 			goto replay;
311 		return err;
312 	}
313 }
314 
315 struct nfnl_err {
316 	struct list_head	head;
317 	struct nlmsghdr		*nlh;
318 	int			err;
319 	struct netlink_ext_ack	extack;
320 };
321 
322 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
323 			const struct netlink_ext_ack *extack)
324 {
325 	struct nfnl_err *nfnl_err;
326 
327 	nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
328 	if (nfnl_err == NULL)
329 		return -ENOMEM;
330 
331 	nfnl_err->nlh = nlh;
332 	nfnl_err->err = err;
333 	nfnl_err->extack = *extack;
334 	list_add_tail(&nfnl_err->head, list);
335 
336 	return 0;
337 }
338 
339 static void nfnl_err_del(struct nfnl_err *nfnl_err)
340 {
341 	list_del(&nfnl_err->head);
342 	kfree(nfnl_err);
343 }
344 
345 static void nfnl_err_reset(struct list_head *err_list)
346 {
347 	struct nfnl_err *nfnl_err, *next;
348 
349 	list_for_each_entry_safe(nfnl_err, next, err_list, head)
350 		nfnl_err_del(nfnl_err);
351 }
352 
353 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
354 {
355 	struct nfnl_err *nfnl_err, *next;
356 
357 	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
358 		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
359 			    &nfnl_err->extack);
360 		nfnl_err_del(nfnl_err);
361 	}
362 }
363 
364 enum {
365 	NFNL_BATCH_FAILURE	= (1 << 0),
366 	NFNL_BATCH_DONE		= (1 << 1),
367 	NFNL_BATCH_REPLAY	= (1 << 2),
368 };
369 
370 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
371 				u16 subsys_id, u32 genid)
372 {
373 	struct sk_buff *oskb = skb;
374 	struct net *net = sock_net(skb->sk);
375 	const struct nfnetlink_subsystem *ss;
376 	const struct nfnl_callback *nc;
377 	struct netlink_ext_ack extack;
378 	LIST_HEAD(err_list);
379 	u32 status;
380 	int err;
381 
382 	if (subsys_id >= NFNL_SUBSYS_COUNT)
383 		return netlink_ack(skb, nlh, -EINVAL, NULL);
384 replay:
385 	status = 0;
386 replay_abort:
387 	skb = netlink_skb_clone(oskb, GFP_KERNEL);
388 	if (!skb)
389 		return netlink_ack(oskb, nlh, -ENOMEM, NULL);
390 
391 	nfnl_lock(subsys_id);
392 	ss = nfnl_dereference_protected(subsys_id);
393 	if (!ss) {
394 #ifdef CONFIG_MODULES
395 		nfnl_unlock(subsys_id);
396 		request_module("nfnetlink-subsys-%d", subsys_id);
397 		nfnl_lock(subsys_id);
398 		ss = nfnl_dereference_protected(subsys_id);
399 		if (!ss)
400 #endif
401 		{
402 			nfnl_unlock(subsys_id);
403 			netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
404 			return kfree_skb(skb);
405 		}
406 	}
407 
408 	if (!ss->valid_genid || !ss->commit || !ss->abort) {
409 		nfnl_unlock(subsys_id);
410 		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
411 		return kfree_skb(skb);
412 	}
413 
414 	if (!try_module_get(ss->owner)) {
415 		nfnl_unlock(subsys_id);
416 		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
417 		return kfree_skb(skb);
418 	}
419 
420 	if (!ss->valid_genid(net, genid)) {
421 		module_put(ss->owner);
422 		nfnl_unlock(subsys_id);
423 		netlink_ack(oskb, nlh, -ERESTART, NULL);
424 		return kfree_skb(skb);
425 	}
426 
427 	nfnl_unlock(subsys_id);
428 
429 	while (skb->len >= nlmsg_total_size(0)) {
430 		int msglen, type;
431 
432 		if (fatal_signal_pending(current)) {
433 			nfnl_err_reset(&err_list);
434 			err = -EINTR;
435 			status = NFNL_BATCH_FAILURE;
436 			goto done;
437 		}
438 
439 		memset(&extack, 0, sizeof(extack));
440 		nlh = nlmsg_hdr(skb);
441 		err = 0;
442 
443 		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
444 		    skb->len < nlh->nlmsg_len ||
445 		    nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
446 			nfnl_err_reset(&err_list);
447 			status |= NFNL_BATCH_FAILURE;
448 			goto done;
449 		}
450 
451 		/* Only requests are handled by the kernel */
452 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
453 			err = -EINVAL;
454 			goto ack;
455 		}
456 
457 		type = nlh->nlmsg_type;
458 		if (type == NFNL_MSG_BATCH_BEGIN) {
459 			/* Malformed: Batch begin twice */
460 			nfnl_err_reset(&err_list);
461 			status |= NFNL_BATCH_FAILURE;
462 			goto done;
463 		} else if (type == NFNL_MSG_BATCH_END) {
464 			status |= NFNL_BATCH_DONE;
465 			goto done;
466 		} else if (type < NLMSG_MIN_TYPE) {
467 			err = -EINVAL;
468 			goto ack;
469 		}
470 
471 		/* We only accept a batch with messages for the same
472 		 * subsystem.
473 		 */
474 		if (NFNL_SUBSYS_ID(type) != subsys_id) {
475 			err = -EINVAL;
476 			goto ack;
477 		}
478 
479 		nc = nfnetlink_find_client(type, ss);
480 		if (!nc) {
481 			err = -EINVAL;
482 			goto ack;
483 		}
484 
485 		if (nc->type != NFNL_CB_BATCH) {
486 			err = -EINVAL;
487 			goto ack;
488 		}
489 
490 		{
491 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
492 			struct nfnl_net *nfnlnet = nfnl_pernet(net);
493 			struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
494 			struct nlattr *attr = (void *)nlh + min_len;
495 			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
496 			int attrlen = nlh->nlmsg_len - min_len;
497 			struct nfnl_info info = {
498 				.net	= net,
499 				.sk	= nfnlnet->nfnl,
500 				.nlh	= nlh,
501 				.nfmsg	= nlmsg_data(nlh),
502 				.extack	= &extack,
503 			};
504 
505 			/* Sanity-check NFTA_MAX_ATTR */
506 			if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
507 				err = -ENOMEM;
508 				goto ack;
509 			}
510 
511 			err = nla_parse_deprecated(cda,
512 						   ss->cb[cb_id].attr_count,
513 						   attr, attrlen,
514 						   ss->cb[cb_id].policy, NULL);
515 			if (err < 0)
516 				goto ack;
517 
518 			err = nc->call(skb, &info, (const struct nlattr **)cda);
519 
520 			/* The lock was released to autoload some module, we
521 			 * have to abort and start from scratch using the
522 			 * original skb.
523 			 */
524 			if (err == -EAGAIN) {
525 				status |= NFNL_BATCH_REPLAY;
526 				goto done;
527 			}
528 		}
529 ack:
530 		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
531 			/* Errors are delivered once the full batch has been
532 			 * processed, this avoids that the same error is
533 			 * reported several times when replaying the batch.
534 			 */
535 			if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
536 				/* We failed to enqueue an error, reset the
537 				 * list of errors and send OOM to userspace
538 				 * pointing to the batch header.
539 				 */
540 				nfnl_err_reset(&err_list);
541 				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
542 					    NULL);
543 				status |= NFNL_BATCH_FAILURE;
544 				goto done;
545 			}
546 			/* We don't stop processing the batch on errors, thus,
547 			 * userspace gets all the errors that the batch
548 			 * triggers.
549 			 */
550 			if (err)
551 				status |= NFNL_BATCH_FAILURE;
552 		}
553 
554 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
555 		if (msglen > skb->len)
556 			msglen = skb->len;
557 		skb_pull(skb, msglen);
558 	}
559 done:
560 	if (status & NFNL_BATCH_REPLAY) {
561 		ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
562 		nfnl_err_reset(&err_list);
563 		kfree_skb(skb);
564 		module_put(ss->owner);
565 		goto replay;
566 	} else if (status == NFNL_BATCH_DONE) {
567 		err = ss->commit(net, oskb);
568 		if (err == -EAGAIN) {
569 			status |= NFNL_BATCH_REPLAY;
570 			goto done;
571 		} else if (err) {
572 			ss->abort(net, oskb, NFNL_ABORT_NONE);
573 			netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
574 		}
575 	} else {
576 		enum nfnl_abort_action abort_action;
577 
578 		if (status & NFNL_BATCH_FAILURE)
579 			abort_action = NFNL_ABORT_NONE;
580 		else
581 			abort_action = NFNL_ABORT_VALIDATE;
582 
583 		err = ss->abort(net, oskb, abort_action);
584 		if (err == -EAGAIN) {
585 			nfnl_err_reset(&err_list);
586 			kfree_skb(skb);
587 			module_put(ss->owner);
588 			status |= NFNL_BATCH_FAILURE;
589 			goto replay_abort;
590 		}
591 	}
592 	if (ss->cleanup)
593 		ss->cleanup(net);
594 
595 	nfnl_err_deliver(&err_list, oskb);
596 	kfree_skb(skb);
597 	module_put(ss->owner);
598 }
599 
600 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
601 	[NFNL_BATCH_GENID]	= { .type = NLA_U32 },
602 };
603 
604 static 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 
642 static 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 
662 static 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(net->ct.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(net->ct.ctnetlink_has_listener, v);
694 	}
695 
696 	spin_unlock(&nfnl_grp_active_lock);
697 #endif
698 }
699 
700 static 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 
720 static 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(net->ct.ctnetlink_has_listener);
748 
749 		v &= ~group_bit;
750 
751 		/* read concurrently without nfnl_grp_active_lock held. */
752 		WRITE_ONCE(net->ct.ctnetlink_has_listener, v);
753 	}
754 	spin_unlock(&nfnl_grp_active_lock);
755 #endif
756 }
757 
758 static 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 
774 static 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 
786 static 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 
793 static 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 
806 static void __exit nfnetlink_exit(void)
807 {
808 	unregister_pernet_subsys(&nfnetlink_net_ops);
809 }
810 module_init(nfnetlink_init);
811 module_exit(nfnetlink_exit);
812