xref: /linux/net/netfilter/nfnetlink.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa) 
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 
29 #include <net/netlink.h>
30 #include <linux/netfilter/nfnetlink.h>
31 
32 MODULE_LICENSE("GPL");
33 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
34 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
35 
36 #define nfnl_dereference_protected(id) \
37 	rcu_dereference_protected(table[(id)].subsys, \
38 				  lockdep_nfnl_is_held((id)))
39 
40 static char __initdata nfversion[] = "0.30";
41 
42 static struct {
43 	struct mutex				mutex;
44 	const struct nfnetlink_subsystem __rcu	*subsys;
45 } table[NFNL_SUBSYS_COUNT];
46 
47 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
48 	[NFNLGRP_CONNTRACK_NEW]		= NFNL_SUBSYS_CTNETLINK,
49 	[NFNLGRP_CONNTRACK_UPDATE]	= NFNL_SUBSYS_CTNETLINK,
50 	[NFNLGRP_CONNTRACK_DESTROY]	= NFNL_SUBSYS_CTNETLINK,
51 	[NFNLGRP_CONNTRACK_EXP_NEW]	= NFNL_SUBSYS_CTNETLINK_EXP,
52 	[NFNLGRP_CONNTRACK_EXP_UPDATE]	= NFNL_SUBSYS_CTNETLINK_EXP,
53 	[NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
54 	[NFNLGRP_NFTABLES]		= NFNL_SUBSYS_NFTABLES,
55 	[NFNLGRP_ACCT_QUOTA]		= NFNL_SUBSYS_ACCT,
56 	[NFNLGRP_NFTRACE]		= NFNL_SUBSYS_NFTABLES,
57 };
58 
59 void nfnl_lock(__u8 subsys_id)
60 {
61 	mutex_lock(&table[subsys_id].mutex);
62 }
63 EXPORT_SYMBOL_GPL(nfnl_lock);
64 
65 void nfnl_unlock(__u8 subsys_id)
66 {
67 	mutex_unlock(&table[subsys_id].mutex);
68 }
69 EXPORT_SYMBOL_GPL(nfnl_unlock);
70 
71 #ifdef CONFIG_PROVE_LOCKING
72 bool lockdep_nfnl_is_held(u8 subsys_id)
73 {
74 	return lockdep_is_held(&table[subsys_id].mutex);
75 }
76 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
77 #endif
78 
79 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
80 {
81 	nfnl_lock(n->subsys_id);
82 	if (table[n->subsys_id].subsys) {
83 		nfnl_unlock(n->subsys_id);
84 		return -EBUSY;
85 	}
86 	rcu_assign_pointer(table[n->subsys_id].subsys, n);
87 	nfnl_unlock(n->subsys_id);
88 
89 	return 0;
90 }
91 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
92 
93 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
94 {
95 	nfnl_lock(n->subsys_id);
96 	table[n->subsys_id].subsys = NULL;
97 	nfnl_unlock(n->subsys_id);
98 	synchronize_rcu();
99 	return 0;
100 }
101 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
102 
103 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
104 {
105 	u8 subsys_id = NFNL_SUBSYS_ID(type);
106 
107 	if (subsys_id >= NFNL_SUBSYS_COUNT)
108 		return NULL;
109 
110 	return rcu_dereference(table[subsys_id].subsys);
111 }
112 
113 static inline const struct nfnl_callback *
114 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
115 {
116 	u8 cb_id = NFNL_MSG_TYPE(type);
117 
118 	if (cb_id >= ss->cb_count)
119 		return NULL;
120 
121 	return &ss->cb[cb_id];
122 }
123 
124 int nfnetlink_has_listeners(struct net *net, unsigned int group)
125 {
126 	return netlink_has_listeners(net->nfnl, group);
127 }
128 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
129 
130 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
131 		   unsigned int group, int echo, gfp_t flags)
132 {
133 	return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
134 }
135 EXPORT_SYMBOL_GPL(nfnetlink_send);
136 
137 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
138 {
139 	return netlink_set_err(net->nfnl, portid, group, error);
140 }
141 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
142 
143 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
144 		      int flags)
145 {
146 	return netlink_unicast(net->nfnl, skb, portid, flags);
147 }
148 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
149 
150 /* Process one complete nfnetlink message. */
151 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
152 			     struct netlink_ext_ack *extack)
153 {
154 	struct net *net = sock_net(skb->sk);
155 	const struct nfnl_callback *nc;
156 	const struct nfnetlink_subsystem *ss;
157 	int type, err;
158 
159 	/* All the messages must at least contain nfgenmsg */
160 	if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
161 		return 0;
162 
163 	type = nlh->nlmsg_type;
164 replay:
165 	rcu_read_lock();
166 	ss = nfnetlink_get_subsys(type);
167 	if (!ss) {
168 #ifdef CONFIG_MODULES
169 		rcu_read_unlock();
170 		request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
171 		rcu_read_lock();
172 		ss = nfnetlink_get_subsys(type);
173 		if (!ss)
174 #endif
175 		{
176 			rcu_read_unlock();
177 			return -EINVAL;
178 		}
179 	}
180 
181 	nc = nfnetlink_find_client(type, ss);
182 	if (!nc) {
183 		rcu_read_unlock();
184 		return -EINVAL;
185 	}
186 
187 	{
188 		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
189 		u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
190 		struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
191 		struct nlattr *attr = (void *)nlh + min_len;
192 		int attrlen = nlh->nlmsg_len - min_len;
193 		__u8 subsys_id = NFNL_SUBSYS_ID(type);
194 
195 		err = nla_parse(cda, ss->cb[cb_id].attr_count, attr, attrlen,
196 				ss->cb[cb_id].policy, extack);
197 		if (err < 0) {
198 			rcu_read_unlock();
199 			return err;
200 		}
201 
202 		if (nc->call_rcu) {
203 			err = nc->call_rcu(net, net->nfnl, skb, nlh,
204 					   (const struct nlattr **)cda);
205 			rcu_read_unlock();
206 		} else {
207 			rcu_read_unlock();
208 			nfnl_lock(subsys_id);
209 			if (nfnl_dereference_protected(subsys_id) != ss ||
210 			    nfnetlink_find_client(type, ss) != nc)
211 				err = -EAGAIN;
212 			else if (nc->call)
213 				err = nc->call(net, net->nfnl, skb, nlh,
214 					       (const struct nlattr **)cda);
215 			else
216 				err = -EINVAL;
217 			nfnl_unlock(subsys_id);
218 		}
219 		if (err == -EAGAIN)
220 			goto replay;
221 		return err;
222 	}
223 }
224 
225 struct nfnl_err {
226 	struct list_head	head;
227 	struct nlmsghdr		*nlh;
228 	int			err;
229 };
230 
231 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err)
232 {
233 	struct nfnl_err *nfnl_err;
234 
235 	nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
236 	if (nfnl_err == NULL)
237 		return -ENOMEM;
238 
239 	nfnl_err->nlh = nlh;
240 	nfnl_err->err = err;
241 	list_add_tail(&nfnl_err->head, list);
242 
243 	return 0;
244 }
245 
246 static void nfnl_err_del(struct nfnl_err *nfnl_err)
247 {
248 	list_del(&nfnl_err->head);
249 	kfree(nfnl_err);
250 }
251 
252 static void nfnl_err_reset(struct list_head *err_list)
253 {
254 	struct nfnl_err *nfnl_err, *next;
255 
256 	list_for_each_entry_safe(nfnl_err, next, err_list, head)
257 		nfnl_err_del(nfnl_err);
258 }
259 
260 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
261 {
262 	struct nfnl_err *nfnl_err, *next;
263 
264 	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
265 		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err, NULL);
266 		nfnl_err_del(nfnl_err);
267 	}
268 }
269 
270 enum {
271 	NFNL_BATCH_FAILURE	= (1 << 0),
272 	NFNL_BATCH_DONE		= (1 << 1),
273 	NFNL_BATCH_REPLAY	= (1 << 2),
274 };
275 
276 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
277 				u16 subsys_id, u32 genid)
278 {
279 	struct sk_buff *oskb = skb;
280 	struct net *net = sock_net(skb->sk);
281 	const struct nfnetlink_subsystem *ss;
282 	const struct nfnl_callback *nc;
283 	LIST_HEAD(err_list);
284 	u32 status;
285 	int err;
286 
287 	if (subsys_id >= NFNL_SUBSYS_COUNT)
288 		return netlink_ack(skb, nlh, -EINVAL, NULL);
289 replay:
290 	status = 0;
291 
292 	skb = netlink_skb_clone(oskb, GFP_KERNEL);
293 	if (!skb)
294 		return netlink_ack(oskb, nlh, -ENOMEM, NULL);
295 
296 	nfnl_lock(subsys_id);
297 	ss = nfnl_dereference_protected(subsys_id);
298 	if (!ss) {
299 #ifdef CONFIG_MODULES
300 		nfnl_unlock(subsys_id);
301 		request_module("nfnetlink-subsys-%d", subsys_id);
302 		nfnl_lock(subsys_id);
303 		ss = nfnl_dereference_protected(subsys_id);
304 		if (!ss)
305 #endif
306 		{
307 			nfnl_unlock(subsys_id);
308 			netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
309 			return kfree_skb(skb);
310 		}
311 	}
312 
313 	if (!ss->commit || !ss->abort) {
314 		nfnl_unlock(subsys_id);
315 		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
316 		return kfree_skb(skb);
317 	}
318 
319 	if (genid && ss->valid_genid && !ss->valid_genid(net, genid)) {
320 		nfnl_unlock(subsys_id);
321 		netlink_ack(oskb, nlh, -ERESTART, NULL);
322 		return kfree_skb(skb);
323 	}
324 
325 	while (skb->len >= nlmsg_total_size(0)) {
326 		int msglen, type;
327 
328 		nlh = nlmsg_hdr(skb);
329 		err = 0;
330 
331 		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
332 		    skb->len < nlh->nlmsg_len ||
333 		    nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
334 			nfnl_err_reset(&err_list);
335 			status |= NFNL_BATCH_FAILURE;
336 			goto done;
337 		}
338 
339 		/* Only requests are handled by the kernel */
340 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
341 			err = -EINVAL;
342 			goto ack;
343 		}
344 
345 		type = nlh->nlmsg_type;
346 		if (type == NFNL_MSG_BATCH_BEGIN) {
347 			/* Malformed: Batch begin twice */
348 			nfnl_err_reset(&err_list);
349 			status |= NFNL_BATCH_FAILURE;
350 			goto done;
351 		} else if (type == NFNL_MSG_BATCH_END) {
352 			status |= NFNL_BATCH_DONE;
353 			goto done;
354 		} else if (type < NLMSG_MIN_TYPE) {
355 			err = -EINVAL;
356 			goto ack;
357 		}
358 
359 		/* We only accept a batch with messages for the same
360 		 * subsystem.
361 		 */
362 		if (NFNL_SUBSYS_ID(type) != subsys_id) {
363 			err = -EINVAL;
364 			goto ack;
365 		}
366 
367 		nc = nfnetlink_find_client(type, ss);
368 		if (!nc) {
369 			err = -EINVAL;
370 			goto ack;
371 		}
372 
373 		{
374 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
375 			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
376 			struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
377 			struct nlattr *attr = (void *)nlh + min_len;
378 			int attrlen = nlh->nlmsg_len - min_len;
379 
380 			err = nla_parse(cda, ss->cb[cb_id].attr_count, attr,
381 					attrlen, ss->cb[cb_id].policy, NULL);
382 			if (err < 0)
383 				goto ack;
384 
385 			if (nc->call_batch) {
386 				err = nc->call_batch(net, net->nfnl, skb, nlh,
387 						     (const struct nlattr **)cda);
388 			}
389 
390 			/* The lock was released to autoload some module, we
391 			 * have to abort and start from scratch using the
392 			 * original skb.
393 			 */
394 			if (err == -EAGAIN) {
395 				status |= NFNL_BATCH_REPLAY;
396 				goto next;
397 			}
398 		}
399 ack:
400 		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
401 			/* Errors are delivered once the full batch has been
402 			 * processed, this avoids that the same error is
403 			 * reported several times when replaying the batch.
404 			 */
405 			if (nfnl_err_add(&err_list, nlh, err) < 0) {
406 				/* We failed to enqueue an error, reset the
407 				 * list of errors and send OOM to userspace
408 				 * pointing to the batch header.
409 				 */
410 				nfnl_err_reset(&err_list);
411 				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
412 					    NULL);
413 				status |= NFNL_BATCH_FAILURE;
414 				goto done;
415 			}
416 			/* We don't stop processing the batch on errors, thus,
417 			 * userspace gets all the errors that the batch
418 			 * triggers.
419 			 */
420 			if (err)
421 				status |= NFNL_BATCH_FAILURE;
422 		}
423 next:
424 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
425 		if (msglen > skb->len)
426 			msglen = skb->len;
427 		skb_pull(skb, msglen);
428 	}
429 done:
430 	if (status & NFNL_BATCH_REPLAY) {
431 		ss->abort(net, oskb);
432 		nfnl_err_reset(&err_list);
433 		nfnl_unlock(subsys_id);
434 		kfree_skb(skb);
435 		goto replay;
436 	} else if (status == NFNL_BATCH_DONE) {
437 		ss->commit(net, oskb);
438 	} else {
439 		ss->abort(net, oskb);
440 	}
441 
442 	nfnl_err_deliver(&err_list, oskb);
443 	nfnl_unlock(subsys_id);
444 	kfree_skb(skb);
445 }
446 
447 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
448 	[NFNL_BATCH_GENID]	= { .type = NLA_U32 },
449 };
450 
451 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
452 {
453 	int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
454 	struct nlattr *attr = (void *)nlh + min_len;
455 	struct nlattr *cda[NFNL_BATCH_MAX + 1];
456 	int attrlen = nlh->nlmsg_len - min_len;
457 	struct nfgenmsg *nfgenmsg;
458 	int msglen, err;
459 	u32 gen_id = 0;
460 	u16 res_id;
461 
462 	msglen = NLMSG_ALIGN(nlh->nlmsg_len);
463 	if (msglen > skb->len)
464 		msglen = skb->len;
465 
466 	if (nlh->nlmsg_len < NLMSG_HDRLEN ||
467 	    skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
468 		return;
469 
470 	err = nla_parse(cda, NFNL_BATCH_MAX, attr, attrlen, nfnl_batch_policy,
471 			NULL);
472 	if (err < 0) {
473 		netlink_ack(skb, nlh, err, NULL);
474 		return;
475 	}
476 	if (cda[NFNL_BATCH_GENID])
477 		gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
478 
479 	nfgenmsg = nlmsg_data(nlh);
480 	skb_pull(skb, msglen);
481 	/* Work around old nft using host byte order */
482 	if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
483 		res_id = NFNL_SUBSYS_NFTABLES;
484 	else
485 		res_id = ntohs(nfgenmsg->res_id);
486 
487 	nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
488 }
489 
490 static void nfnetlink_rcv(struct sk_buff *skb)
491 {
492 	struct nlmsghdr *nlh = nlmsg_hdr(skb);
493 
494 	if (nlh->nlmsg_len < NLMSG_HDRLEN ||
495 	    skb->len < nlh->nlmsg_len)
496 		return;
497 
498 	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
499 		netlink_ack(skb, nlh, -EPERM, NULL);
500 		return;
501 	}
502 
503 	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
504 		nfnetlink_rcv_skb_batch(skb, nlh);
505 	else
506 		netlink_rcv_skb(skb, nfnetlink_rcv_msg);
507 }
508 
509 #ifdef CONFIG_MODULES
510 static int nfnetlink_bind(struct net *net, int group)
511 {
512 	const struct nfnetlink_subsystem *ss;
513 	int type;
514 
515 	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
516 		return 0;
517 
518 	type = nfnl_group2type[group];
519 
520 	rcu_read_lock();
521 	ss = nfnetlink_get_subsys(type << 8);
522 	rcu_read_unlock();
523 	if (!ss)
524 		request_module("nfnetlink-subsys-%d", type);
525 	return 0;
526 }
527 #endif
528 
529 static int __net_init nfnetlink_net_init(struct net *net)
530 {
531 	struct sock *nfnl;
532 	struct netlink_kernel_cfg cfg = {
533 		.groups	= NFNLGRP_MAX,
534 		.input	= nfnetlink_rcv,
535 #ifdef CONFIG_MODULES
536 		.bind	= nfnetlink_bind,
537 #endif
538 	};
539 
540 	nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
541 	if (!nfnl)
542 		return -ENOMEM;
543 	net->nfnl_stash = nfnl;
544 	rcu_assign_pointer(net->nfnl, nfnl);
545 	return 0;
546 }
547 
548 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
549 {
550 	struct net *net;
551 
552 	list_for_each_entry(net, net_exit_list, exit_list)
553 		RCU_INIT_POINTER(net->nfnl, NULL);
554 	synchronize_net();
555 	list_for_each_entry(net, net_exit_list, exit_list)
556 		netlink_kernel_release(net->nfnl_stash);
557 }
558 
559 static struct pernet_operations nfnetlink_net_ops = {
560 	.init		= nfnetlink_net_init,
561 	.exit_batch	= nfnetlink_net_exit_batch,
562 };
563 
564 static int __init nfnetlink_init(void)
565 {
566 	int i;
567 
568 	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
569 		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
570 
571 	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
572 		mutex_init(&table[i].mutex);
573 
574 	pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
575 	return register_pernet_subsys(&nfnetlink_net_ops);
576 }
577 
578 static void __exit nfnetlink_exit(void)
579 {
580 	pr_info("Removing netfilter NETLINK layer.\n");
581 	unregister_pernet_subsys(&nfnetlink_net_ops);
582 }
583 module_init(nfnetlink_init);
584 module_exit(nfnetlink_exit);
585