xref: /openbmc/linux/net/netfilter/nfnetlink.c (revision eb3fcf00) 
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 static char __initdata nfversion[] = "0.30";
37 
38 static struct {
39 	struct mutex				mutex;
40 	const struct nfnetlink_subsystem __rcu	*subsys;
41 } table[NFNL_SUBSYS_COUNT];
42 
43 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
44 	[NFNLGRP_CONNTRACK_NEW]		= NFNL_SUBSYS_CTNETLINK,
45 	[NFNLGRP_CONNTRACK_UPDATE]	= NFNL_SUBSYS_CTNETLINK,
46 	[NFNLGRP_CONNTRACK_DESTROY]	= NFNL_SUBSYS_CTNETLINK,
47 	[NFNLGRP_CONNTRACK_EXP_NEW]	= NFNL_SUBSYS_CTNETLINK_EXP,
48 	[NFNLGRP_CONNTRACK_EXP_UPDATE]	= NFNL_SUBSYS_CTNETLINK_EXP,
49 	[NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
50 	[NFNLGRP_NFTABLES]		= NFNL_SUBSYS_NFTABLES,
51 	[NFNLGRP_ACCT_QUOTA]		= NFNL_SUBSYS_ACCT,
52 };
53 
54 void nfnl_lock(__u8 subsys_id)
55 {
56 	mutex_lock(&table[subsys_id].mutex);
57 }
58 EXPORT_SYMBOL_GPL(nfnl_lock);
59 
60 void nfnl_unlock(__u8 subsys_id)
61 {
62 	mutex_unlock(&table[subsys_id].mutex);
63 }
64 EXPORT_SYMBOL_GPL(nfnl_unlock);
65 
66 #ifdef CONFIG_PROVE_LOCKING
67 int lockdep_nfnl_is_held(u8 subsys_id)
68 {
69 	return lockdep_is_held(&table[subsys_id].mutex);
70 }
71 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
72 #endif
73 
74 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
75 {
76 	nfnl_lock(n->subsys_id);
77 	if (table[n->subsys_id].subsys) {
78 		nfnl_unlock(n->subsys_id);
79 		return -EBUSY;
80 	}
81 	rcu_assign_pointer(table[n->subsys_id].subsys, n);
82 	nfnl_unlock(n->subsys_id);
83 
84 	return 0;
85 }
86 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
87 
88 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
89 {
90 	nfnl_lock(n->subsys_id);
91 	table[n->subsys_id].subsys = NULL;
92 	nfnl_unlock(n->subsys_id);
93 	synchronize_rcu();
94 	return 0;
95 }
96 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
97 
98 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u_int16_t type)
99 {
100 	u_int8_t subsys_id = NFNL_SUBSYS_ID(type);
101 
102 	if (subsys_id >= NFNL_SUBSYS_COUNT)
103 		return NULL;
104 
105 	return rcu_dereference(table[subsys_id].subsys);
106 }
107 
108 static inline const struct nfnl_callback *
109 nfnetlink_find_client(u_int16_t type, const struct nfnetlink_subsystem *ss)
110 {
111 	u_int8_t cb_id = NFNL_MSG_TYPE(type);
112 
113 	if (cb_id >= ss->cb_count)
114 		return NULL;
115 
116 	return &ss->cb[cb_id];
117 }
118 
119 int nfnetlink_has_listeners(struct net *net, unsigned int group)
120 {
121 	return netlink_has_listeners(net->nfnl, group);
122 }
123 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
124 
125 struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
126 				    u32 dst_portid, gfp_t gfp_mask)
127 {
128 	return netlink_alloc_skb(net->nfnl, size, dst_portid, gfp_mask);
129 }
130 EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb);
131 
132 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
133 		   unsigned int group, int echo, gfp_t flags)
134 {
135 	return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
136 }
137 EXPORT_SYMBOL_GPL(nfnetlink_send);
138 
139 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
140 {
141 	return netlink_set_err(net->nfnl, portid, group, error);
142 }
143 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
144 
145 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
146 		      int flags)
147 {
148 	return netlink_unicast(net->nfnl, skb, portid, flags);
149 }
150 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
151 
152 /* Process one complete nfnetlink message. */
153 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
154 {
155 	struct net *net = sock_net(skb->sk);
156 	const struct nfnl_callback *nc;
157 	const struct nfnetlink_subsystem *ss;
158 	int type, err;
159 
160 	/* All the messages must at least contain nfgenmsg */
161 	if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
162 		return 0;
163 
164 	type = nlh->nlmsg_type;
165 replay:
166 	rcu_read_lock();
167 	ss = nfnetlink_get_subsys(type);
168 	if (!ss) {
169 #ifdef CONFIG_MODULES
170 		rcu_read_unlock();
171 		request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
172 		rcu_read_lock();
173 		ss = nfnetlink_get_subsys(type);
174 		if (!ss)
175 #endif
176 		{
177 			rcu_read_unlock();
178 			return -EINVAL;
179 		}
180 	}
181 
182 	nc = nfnetlink_find_client(type, ss);
183 	if (!nc) {
184 		rcu_read_unlock();
185 		return -EINVAL;
186 	}
187 
188 	{
189 		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
190 		u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
191 		struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
192 		struct nlattr *attr = (void *)nlh + min_len;
193 		int attrlen = nlh->nlmsg_len - min_len;
194 		__u8 subsys_id = NFNL_SUBSYS_ID(type);
195 
196 		err = nla_parse(cda, ss->cb[cb_id].attr_count,
197 				attr, attrlen, ss->cb[cb_id].policy);
198 		if (err < 0) {
199 			rcu_read_unlock();
200 			return err;
201 		}
202 
203 		if (nc->call_rcu) {
204 			err = nc->call_rcu(net->nfnl, skb, nlh,
205 					   (const struct nlattr **)cda);
206 			rcu_read_unlock();
207 		} else {
208 			rcu_read_unlock();
209 			nfnl_lock(subsys_id);
210 			if (rcu_dereference_protected(table[subsys_id].subsys,
211 				lockdep_is_held(&table[subsys_id].mutex)) != ss ||
212 			    nfnetlink_find_client(type, ss) != nc)
213 				err = -EAGAIN;
214 			else if (nc->call)
215 				err = nc->call(net->nfnl, skb, nlh,
216 						   (const struct nlattr **)cda);
217 			else
218 				err = -EINVAL;
219 			nfnl_unlock(subsys_id);
220 		}
221 		if (err == -EAGAIN)
222 			goto replay;
223 		return err;
224 	}
225 }
226 
227 struct nfnl_err {
228 	struct list_head	head;
229 	struct nlmsghdr		*nlh;
230 	int			err;
231 };
232 
233 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err)
234 {
235 	struct nfnl_err *nfnl_err;
236 
237 	nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
238 	if (nfnl_err == NULL)
239 		return -ENOMEM;
240 
241 	nfnl_err->nlh = nlh;
242 	nfnl_err->err = err;
243 	list_add_tail(&nfnl_err->head, list);
244 
245 	return 0;
246 }
247 
248 static void nfnl_err_del(struct nfnl_err *nfnl_err)
249 {
250 	list_del(&nfnl_err->head);
251 	kfree(nfnl_err);
252 }
253 
254 static void nfnl_err_reset(struct list_head *err_list)
255 {
256 	struct nfnl_err *nfnl_err, *next;
257 
258 	list_for_each_entry_safe(nfnl_err, next, err_list, head)
259 		nfnl_err_del(nfnl_err);
260 }
261 
262 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
263 {
264 	struct nfnl_err *nfnl_err, *next;
265 
266 	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
267 		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err);
268 		nfnl_err_del(nfnl_err);
269 	}
270 }
271 
272 enum {
273 	NFNL_BATCH_FAILURE	= (1 << 0),
274 	NFNL_BATCH_DONE		= (1 << 1),
275 	NFNL_BATCH_REPLAY	= (1 << 2),
276 };
277 
278 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
279 				u_int16_t subsys_id)
280 {
281 	struct sk_buff *oskb = skb;
282 	struct net *net = sock_net(skb->sk);
283 	const struct nfnetlink_subsystem *ss;
284 	const struct nfnl_callback *nc;
285 	static LIST_HEAD(err_list);
286 	u32 status;
287 	int err;
288 
289 	if (subsys_id >= NFNL_SUBSYS_COUNT)
290 		return netlink_ack(skb, nlh, -EINVAL);
291 replay:
292 	status = 0;
293 
294 	skb = netlink_skb_clone(oskb, GFP_KERNEL);
295 	if (!skb)
296 		return netlink_ack(oskb, nlh, -ENOMEM);
297 
298 	skb->sk = oskb->sk;
299 
300 	nfnl_lock(subsys_id);
301 	ss = rcu_dereference_protected(table[subsys_id].subsys,
302 				       lockdep_is_held(&table[subsys_id].mutex));
303 	if (!ss) {
304 #ifdef CONFIG_MODULES
305 		nfnl_unlock(subsys_id);
306 		request_module("nfnetlink-subsys-%d", subsys_id);
307 		nfnl_lock(subsys_id);
308 		ss = rcu_dereference_protected(table[subsys_id].subsys,
309 					       lockdep_is_held(&table[subsys_id].mutex));
310 		if (!ss)
311 #endif
312 		{
313 			nfnl_unlock(subsys_id);
314 			netlink_ack(skb, nlh, -EOPNOTSUPP);
315 			return kfree_skb(skb);
316 		}
317 	}
318 
319 	if (!ss->commit || !ss->abort) {
320 		nfnl_unlock(subsys_id);
321 		netlink_ack(skb, nlh, -EOPNOTSUPP);
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 (nlmsg_len(nlh) < sizeof(struct nfgenmsg) ||
332 		    skb->len < nlh->nlmsg_len) {
333 			err = -EINVAL;
334 			goto ack;
335 		}
336 
337 		/* Only requests are handled by the kernel */
338 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
339 			err = -EINVAL;
340 			goto ack;
341 		}
342 
343 		type = nlh->nlmsg_type;
344 		if (type == NFNL_MSG_BATCH_BEGIN) {
345 			/* Malformed: Batch begin twice */
346 			nfnl_err_reset(&err_list);
347 			status |= NFNL_BATCH_FAILURE;
348 			goto done;
349 		} else if (type == NFNL_MSG_BATCH_END) {
350 			status |= NFNL_BATCH_DONE;
351 			goto done;
352 		} else if (type < NLMSG_MIN_TYPE) {
353 			err = -EINVAL;
354 			goto ack;
355 		}
356 
357 		/* We only accept a batch with messages for the same
358 		 * subsystem.
359 		 */
360 		if (NFNL_SUBSYS_ID(type) != subsys_id) {
361 			err = -EINVAL;
362 			goto ack;
363 		}
364 
365 		nc = nfnetlink_find_client(type, ss);
366 		if (!nc) {
367 			err = -EINVAL;
368 			goto ack;
369 		}
370 
371 		{
372 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
373 			u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
374 			struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
375 			struct nlattr *attr = (void *)nlh + min_len;
376 			int attrlen = nlh->nlmsg_len - min_len;
377 
378 			err = nla_parse(cda, ss->cb[cb_id].attr_count,
379 					attr, attrlen, ss->cb[cb_id].policy);
380 			if (err < 0)
381 				goto ack;
382 
383 			if (nc->call_batch) {
384 				err = nc->call_batch(net->nfnl, skb, nlh,
385 						     (const struct nlattr **)cda);
386 			}
387 
388 			/* The lock was released to autoload some module, we
389 			 * have to abort and start from scratch using the
390 			 * original skb.
391 			 */
392 			if (err == -EAGAIN) {
393 				status |= NFNL_BATCH_REPLAY;
394 				goto next;
395 			}
396 		}
397 ack:
398 		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
399 			/* Errors are delivered once the full batch has been
400 			 * processed, this avoids that the same error is
401 			 * reported several times when replaying the batch.
402 			 */
403 			if (nfnl_err_add(&err_list, nlh, err) < 0) {
404 				/* We failed to enqueue an error, reset the
405 				 * list of errors and send OOM to userspace
406 				 * pointing to the batch header.
407 				 */
408 				nfnl_err_reset(&err_list);
409 				netlink_ack(skb, nlmsg_hdr(oskb), -ENOMEM);
410 				status |= NFNL_BATCH_FAILURE;
411 				goto done;
412 			}
413 			/* We don't stop processing the batch on errors, thus,
414 			 * userspace gets all the errors that the batch
415 			 * triggers.
416 			 */
417 			if (err)
418 				status |= NFNL_BATCH_FAILURE;
419 		}
420 next:
421 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
422 		if (msglen > skb->len)
423 			msglen = skb->len;
424 		skb_pull(skb, msglen);
425 	}
426 done:
427 	if (status & NFNL_BATCH_REPLAY) {
428 		ss->abort(oskb);
429 		nfnl_err_reset(&err_list);
430 		nfnl_unlock(subsys_id);
431 		kfree_skb(skb);
432 		goto replay;
433 	} else if (status == NFNL_BATCH_DONE) {
434 		ss->commit(oskb);
435 	} else {
436 		ss->abort(oskb);
437 	}
438 
439 	nfnl_err_deliver(&err_list, oskb);
440 	nfnl_unlock(subsys_id);
441 	kfree_skb(skb);
442 }
443 
444 static void nfnetlink_rcv(struct sk_buff *skb)
445 {
446 	struct nlmsghdr *nlh = nlmsg_hdr(skb);
447 	u_int16_t res_id;
448 	int msglen;
449 
450 	if (nlh->nlmsg_len < NLMSG_HDRLEN ||
451 	    skb->len < nlh->nlmsg_len)
452 		return;
453 
454 	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
455 		netlink_ack(skb, nlh, -EPERM);
456 		return;
457 	}
458 
459 	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
460 		struct nfgenmsg *nfgenmsg;
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 		nfgenmsg = nlmsg_data(nlh);
471 		skb_pull(skb, msglen);
472 		/* Work around old nft using host byte order */
473 		if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
474 			res_id = NFNL_SUBSYS_NFTABLES;
475 		else
476 			res_id = ntohs(nfgenmsg->res_id);
477 		nfnetlink_rcv_batch(skb, nlh, res_id);
478 	} else {
479 		netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
480 	}
481 }
482 
483 #ifdef CONFIG_MODULES
484 static int nfnetlink_bind(struct net *net, int group)
485 {
486 	const struct nfnetlink_subsystem *ss;
487 	int type;
488 
489 	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
490 		return 0;
491 
492 	type = nfnl_group2type[group];
493 
494 	rcu_read_lock();
495 	ss = nfnetlink_get_subsys(type);
496 	rcu_read_unlock();
497 	if (!ss)
498 		request_module("nfnetlink-subsys-%d", type);
499 	return 0;
500 }
501 #endif
502 
503 static int __net_init nfnetlink_net_init(struct net *net)
504 {
505 	struct sock *nfnl;
506 	struct netlink_kernel_cfg cfg = {
507 		.groups	= NFNLGRP_MAX,
508 		.input	= nfnetlink_rcv,
509 #ifdef CONFIG_MODULES
510 		.bind	= nfnetlink_bind,
511 #endif
512 	};
513 
514 	nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
515 	if (!nfnl)
516 		return -ENOMEM;
517 	net->nfnl_stash = nfnl;
518 	rcu_assign_pointer(net->nfnl, nfnl);
519 	return 0;
520 }
521 
522 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
523 {
524 	struct net *net;
525 
526 	list_for_each_entry(net, net_exit_list, exit_list)
527 		RCU_INIT_POINTER(net->nfnl, NULL);
528 	synchronize_net();
529 	list_for_each_entry(net, net_exit_list, exit_list)
530 		netlink_kernel_release(net->nfnl_stash);
531 }
532 
533 static struct pernet_operations nfnetlink_net_ops = {
534 	.init		= nfnetlink_net_init,
535 	.exit_batch	= nfnetlink_net_exit_batch,
536 };
537 
538 static int __init nfnetlink_init(void)
539 {
540 	int i;
541 
542 	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
543 		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
544 
545 	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
546 		mutex_init(&table[i].mutex);
547 
548 	pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
549 	return register_pernet_subsys(&nfnetlink_net_ops);
550 }
551 
552 static void __exit nfnetlink_exit(void)
553 {
554 	pr_info("Removing netfilter NETLINK layer.\n");
555 	unregister_pernet_subsys(&nfnetlink_net_ops);
556 }
557 module_init(nfnetlink_init);
558 module_exit(nfnetlink_exit);
559