xref: /openbmc/linux/net/netfilter/nfnetlink.c (revision bc5aa3a0) 
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) 2005,2007 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 <asm/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(u_int16_t type)
104 {
105 	u_int8_t 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(u_int16_t type, const struct nfnetlink_subsystem *ss)
115 {
116 	u_int8_t 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 {
153 	struct net *net = sock_net(skb->sk);
154 	const struct nfnl_callback *nc;
155 	const struct nfnetlink_subsystem *ss;
156 	int type, err;
157 
158 	/* All the messages must at least contain nfgenmsg */
159 	if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
160 		return 0;
161 
162 	type = nlh->nlmsg_type;
163 replay:
164 	rcu_read_lock();
165 	ss = nfnetlink_get_subsys(type);
166 	if (!ss) {
167 #ifdef CONFIG_MODULES
168 		rcu_read_unlock();
169 		request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
170 		rcu_read_lock();
171 		ss = nfnetlink_get_subsys(type);
172 		if (!ss)
173 #endif
174 		{
175 			rcu_read_unlock();
176 			return -EINVAL;
177 		}
178 	}
179 
180 	nc = nfnetlink_find_client(type, ss);
181 	if (!nc) {
182 		rcu_read_unlock();
183 		return -EINVAL;
184 	}
185 
186 	{
187 		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
188 		u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
189 		struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
190 		struct nlattr *attr = (void *)nlh + min_len;
191 		int attrlen = nlh->nlmsg_len - min_len;
192 		__u8 subsys_id = NFNL_SUBSYS_ID(type);
193 
194 		err = nla_parse(cda, ss->cb[cb_id].attr_count,
195 				attr, attrlen, ss->cb[cb_id].policy);
196 		if (err < 0) {
197 			rcu_read_unlock();
198 			return err;
199 		}
200 
201 		if (nc->call_rcu) {
202 			err = nc->call_rcu(net, net->nfnl, skb, nlh,
203 					   (const struct nlattr **)cda);
204 			rcu_read_unlock();
205 		} else {
206 			rcu_read_unlock();
207 			nfnl_lock(subsys_id);
208 			if (nfnl_dereference_protected(subsys_id) != ss ||
209 			    nfnetlink_find_client(type, ss) != nc)
210 				err = -EAGAIN;
211 			else if (nc->call)
212 				err = nc->call(net, net->nfnl, skb, nlh,
213 					       (const struct nlattr **)cda);
214 			else
215 				err = -EINVAL;
216 			nfnl_unlock(subsys_id);
217 		}
218 		if (err == -EAGAIN)
219 			goto replay;
220 		return err;
221 	}
222 }
223 
224 struct nfnl_err {
225 	struct list_head	head;
226 	struct nlmsghdr		*nlh;
227 	int			err;
228 };
229 
230 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err)
231 {
232 	struct nfnl_err *nfnl_err;
233 
234 	nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
235 	if (nfnl_err == NULL)
236 		return -ENOMEM;
237 
238 	nfnl_err->nlh = nlh;
239 	nfnl_err->err = err;
240 	list_add_tail(&nfnl_err->head, list);
241 
242 	return 0;
243 }
244 
245 static void nfnl_err_del(struct nfnl_err *nfnl_err)
246 {
247 	list_del(&nfnl_err->head);
248 	kfree(nfnl_err);
249 }
250 
251 static void nfnl_err_reset(struct list_head *err_list)
252 {
253 	struct nfnl_err *nfnl_err, *next;
254 
255 	list_for_each_entry_safe(nfnl_err, next, err_list, head)
256 		nfnl_err_del(nfnl_err);
257 }
258 
259 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
260 {
261 	struct nfnl_err *nfnl_err, *next;
262 
263 	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
264 		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err);
265 		nfnl_err_del(nfnl_err);
266 	}
267 }
268 
269 enum {
270 	NFNL_BATCH_FAILURE	= (1 << 0),
271 	NFNL_BATCH_DONE		= (1 << 1),
272 	NFNL_BATCH_REPLAY	= (1 << 2),
273 };
274 
275 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
276 				u_int16_t subsys_id)
277 {
278 	struct sk_buff *oskb = skb;
279 	struct net *net = sock_net(skb->sk);
280 	const struct nfnetlink_subsystem *ss;
281 	const struct nfnl_callback *nc;
282 	static LIST_HEAD(err_list);
283 	u32 status;
284 	int err;
285 
286 	if (subsys_id >= NFNL_SUBSYS_COUNT)
287 		return netlink_ack(skb, nlh, -EINVAL);
288 replay:
289 	status = 0;
290 
291 	skb = netlink_skb_clone(oskb, GFP_KERNEL);
292 	if (!skb)
293 		return netlink_ack(oskb, nlh, -ENOMEM);
294 
295 	nfnl_lock(subsys_id);
296 	ss = nfnl_dereference_protected(subsys_id);
297 	if (!ss) {
298 #ifdef CONFIG_MODULES
299 		nfnl_unlock(subsys_id);
300 		request_module("nfnetlink-subsys-%d", subsys_id);
301 		nfnl_lock(subsys_id);
302 		ss = nfnl_dereference_protected(subsys_id);
303 		if (!ss)
304 #endif
305 		{
306 			nfnl_unlock(subsys_id);
307 			netlink_ack(oskb, nlh, -EOPNOTSUPP);
308 			return kfree_skb(skb);
309 		}
310 	}
311 
312 	if (!ss->commit || !ss->abort) {
313 		nfnl_unlock(subsys_id);
314 		netlink_ack(oskb, nlh, -EOPNOTSUPP);
315 		return kfree_skb(skb);
316 	}
317 
318 	while (skb->len >= nlmsg_total_size(0)) {
319 		int msglen, type;
320 
321 		nlh = nlmsg_hdr(skb);
322 		err = 0;
323 
324 		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
325 		    skb->len < nlh->nlmsg_len ||
326 		    nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
327 			nfnl_err_reset(&err_list);
328 			status |= NFNL_BATCH_FAILURE;
329 			goto done;
330 		}
331 
332 		/* Only requests are handled by the kernel */
333 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
334 			err = -EINVAL;
335 			goto ack;
336 		}
337 
338 		type = nlh->nlmsg_type;
339 		if (type == NFNL_MSG_BATCH_BEGIN) {
340 			/* Malformed: Batch begin twice */
341 			nfnl_err_reset(&err_list);
342 			status |= NFNL_BATCH_FAILURE;
343 			goto done;
344 		} else if (type == NFNL_MSG_BATCH_END) {
345 			status |= NFNL_BATCH_DONE;
346 			goto done;
347 		} else if (type < NLMSG_MIN_TYPE) {
348 			err = -EINVAL;
349 			goto ack;
350 		}
351 
352 		/* We only accept a batch with messages for the same
353 		 * subsystem.
354 		 */
355 		if (NFNL_SUBSYS_ID(type) != subsys_id) {
356 			err = -EINVAL;
357 			goto ack;
358 		}
359 
360 		nc = nfnetlink_find_client(type, ss);
361 		if (!nc) {
362 			err = -EINVAL;
363 			goto ack;
364 		}
365 
366 		{
367 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
368 			u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
369 			struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
370 			struct nlattr *attr = (void *)nlh + min_len;
371 			int attrlen = nlh->nlmsg_len - min_len;
372 
373 			err = nla_parse(cda, ss->cb[cb_id].attr_count,
374 					attr, attrlen, ss->cb[cb_id].policy);
375 			if (err < 0)
376 				goto ack;
377 
378 			if (nc->call_batch) {
379 				err = nc->call_batch(net, net->nfnl, skb, nlh,
380 						     (const struct nlattr **)cda);
381 			}
382 
383 			/* The lock was released to autoload some module, we
384 			 * have to abort and start from scratch using the
385 			 * original skb.
386 			 */
387 			if (err == -EAGAIN) {
388 				status |= NFNL_BATCH_REPLAY;
389 				goto next;
390 			}
391 		}
392 ack:
393 		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
394 			/* Errors are delivered once the full batch has been
395 			 * processed, this avoids that the same error is
396 			 * reported several times when replaying the batch.
397 			 */
398 			if (nfnl_err_add(&err_list, nlh, err) < 0) {
399 				/* We failed to enqueue an error, reset the
400 				 * list of errors and send OOM to userspace
401 				 * pointing to the batch header.
402 				 */
403 				nfnl_err_reset(&err_list);
404 				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM);
405 				status |= NFNL_BATCH_FAILURE;
406 				goto done;
407 			}
408 			/* We don't stop processing the batch on errors, thus,
409 			 * userspace gets all the errors that the batch
410 			 * triggers.
411 			 */
412 			if (err)
413 				status |= NFNL_BATCH_FAILURE;
414 		}
415 next:
416 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
417 		if (msglen > skb->len)
418 			msglen = skb->len;
419 		skb_pull(skb, msglen);
420 	}
421 done:
422 	if (status & NFNL_BATCH_REPLAY) {
423 		ss->abort(net, oskb);
424 		nfnl_err_reset(&err_list);
425 		nfnl_unlock(subsys_id);
426 		kfree_skb(skb);
427 		goto replay;
428 	} else if (status == NFNL_BATCH_DONE) {
429 		ss->commit(net, oskb);
430 	} else {
431 		ss->abort(net, oskb);
432 	}
433 
434 	nfnl_err_deliver(&err_list, oskb);
435 	nfnl_unlock(subsys_id);
436 	kfree_skb(skb);
437 }
438 
439 static void nfnetlink_rcv(struct sk_buff *skb)
440 {
441 	struct nlmsghdr *nlh = nlmsg_hdr(skb);
442 	u_int16_t res_id;
443 	int msglen;
444 
445 	if (nlh->nlmsg_len < NLMSG_HDRLEN ||
446 	    skb->len < nlh->nlmsg_len)
447 		return;
448 
449 	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
450 		netlink_ack(skb, nlh, -EPERM);
451 		return;
452 	}
453 
454 	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
455 		struct nfgenmsg *nfgenmsg;
456 
457 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
458 		if (msglen > skb->len)
459 			msglen = skb->len;
460 
461 		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
462 		    skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
463 			return;
464 
465 		nfgenmsg = nlmsg_data(nlh);
466 		skb_pull(skb, msglen);
467 		/* Work around old nft using host byte order */
468 		if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
469 			res_id = NFNL_SUBSYS_NFTABLES;
470 		else
471 			res_id = ntohs(nfgenmsg->res_id);
472 		nfnetlink_rcv_batch(skb, nlh, res_id);
473 	} else {
474 		netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
475 	}
476 }
477 
478 #ifdef CONFIG_MODULES
479 static int nfnetlink_bind(struct net *net, int group)
480 {
481 	const struct nfnetlink_subsystem *ss;
482 	int type;
483 
484 	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
485 		return 0;
486 
487 	type = nfnl_group2type[group];
488 
489 	rcu_read_lock();
490 	ss = nfnetlink_get_subsys(type << 8);
491 	rcu_read_unlock();
492 	if (!ss)
493 		request_module("nfnetlink-subsys-%d", type);
494 	return 0;
495 }
496 #endif
497 
498 static int __net_init nfnetlink_net_init(struct net *net)
499 {
500 	struct sock *nfnl;
501 	struct netlink_kernel_cfg cfg = {
502 		.groups	= NFNLGRP_MAX,
503 		.input	= nfnetlink_rcv,
504 #ifdef CONFIG_MODULES
505 		.bind	= nfnetlink_bind,
506 #endif
507 	};
508 
509 	nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
510 	if (!nfnl)
511 		return -ENOMEM;
512 	net->nfnl_stash = nfnl;
513 	rcu_assign_pointer(net->nfnl, nfnl);
514 	return 0;
515 }
516 
517 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
518 {
519 	struct net *net;
520 
521 	list_for_each_entry(net, net_exit_list, exit_list)
522 		RCU_INIT_POINTER(net->nfnl, NULL);
523 	synchronize_net();
524 	list_for_each_entry(net, net_exit_list, exit_list)
525 		netlink_kernel_release(net->nfnl_stash);
526 }
527 
528 static struct pernet_operations nfnetlink_net_ops = {
529 	.init		= nfnetlink_net_init,
530 	.exit_batch	= nfnetlink_net_exit_batch,
531 };
532 
533 static int __init nfnetlink_init(void)
534 {
535 	int i;
536 
537 	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
538 		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
539 
540 	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
541 		mutex_init(&table[i].mutex);
542 
543 	pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
544 	return register_pernet_subsys(&nfnetlink_net_ops);
545 }
546 
547 static void __exit nfnetlink_exit(void)
548 {
549 	pr_info("Removing netfilter NETLINK layer.\n");
550 	unregister_pernet_subsys(&nfnetlink_net_ops);
551 }
552 module_init(nfnetlink_init);
553 module_exit(nfnetlink_exit);
554