xref: /openbmc/linux/net/netfilter/nfnetlink.c (revision b34e08d5) 
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 };
51 
52 void nfnl_lock(__u8 subsys_id)
53 {
54 	mutex_lock(&table[subsys_id].mutex);
55 }
56 EXPORT_SYMBOL_GPL(nfnl_lock);
57 
58 void nfnl_unlock(__u8 subsys_id)
59 {
60 	mutex_unlock(&table[subsys_id].mutex);
61 }
62 EXPORT_SYMBOL_GPL(nfnl_unlock);
63 
64 #ifdef CONFIG_PROVE_LOCKING
65 int lockdep_nfnl_is_held(u8 subsys_id)
66 {
67 	return lockdep_is_held(&table[subsys_id].mutex);
68 }
69 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
70 #endif
71 
72 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
73 {
74 	nfnl_lock(n->subsys_id);
75 	if (table[n->subsys_id].subsys) {
76 		nfnl_unlock(n->subsys_id);
77 		return -EBUSY;
78 	}
79 	rcu_assign_pointer(table[n->subsys_id].subsys, n);
80 	nfnl_unlock(n->subsys_id);
81 
82 	return 0;
83 }
84 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
85 
86 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
87 {
88 	nfnl_lock(n->subsys_id);
89 	table[n->subsys_id].subsys = NULL;
90 	nfnl_unlock(n->subsys_id);
91 	synchronize_rcu();
92 	return 0;
93 }
94 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
95 
96 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u_int16_t type)
97 {
98 	u_int8_t subsys_id = NFNL_SUBSYS_ID(type);
99 
100 	if (subsys_id >= NFNL_SUBSYS_COUNT)
101 		return NULL;
102 
103 	return rcu_dereference(table[subsys_id].subsys);
104 }
105 
106 static inline const struct nfnl_callback *
107 nfnetlink_find_client(u_int16_t type, const struct nfnetlink_subsystem *ss)
108 {
109 	u_int8_t cb_id = NFNL_MSG_TYPE(type);
110 
111 	if (cb_id >= ss->cb_count)
112 		return NULL;
113 
114 	return &ss->cb[cb_id];
115 }
116 
117 int nfnetlink_has_listeners(struct net *net, unsigned int group)
118 {
119 	return netlink_has_listeners(net->nfnl, group);
120 }
121 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
122 
123 struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
124 				    u32 dst_portid, gfp_t gfp_mask)
125 {
126 	return netlink_alloc_skb(net->nfnl, size, dst_portid, gfp_mask);
127 }
128 EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb);
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->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 (rcu_dereference_protected(table[subsys_id].subsys,
209 				lockdep_is_held(&table[subsys_id].mutex)) != ss ||
210 			    nfnetlink_find_client(type, ss) != nc)
211 				err = -EAGAIN;
212 			else if (nc->call)
213 				err = nc->call(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 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
226 				u_int16_t subsys_id)
227 {
228 	struct sk_buff *nskb, *oskb = skb;
229 	struct net *net = sock_net(skb->sk);
230 	const struct nfnetlink_subsystem *ss;
231 	const struct nfnl_callback *nc;
232 	bool success = true, done = false;
233 	int err;
234 
235 	if (subsys_id >= NFNL_SUBSYS_COUNT)
236 		return netlink_ack(skb, nlh, -EINVAL);
237 replay:
238 	nskb = netlink_skb_clone(oskb, GFP_KERNEL);
239 	if (!nskb)
240 		return netlink_ack(oskb, nlh, -ENOMEM);
241 
242 	nskb->sk = oskb->sk;
243 	skb = nskb;
244 
245 	nfnl_lock(subsys_id);
246 	ss = rcu_dereference_protected(table[subsys_id].subsys,
247 				       lockdep_is_held(&table[subsys_id].mutex));
248 	if (!ss) {
249 #ifdef CONFIG_MODULES
250 		nfnl_unlock(subsys_id);
251 		request_module("nfnetlink-subsys-%d", subsys_id);
252 		nfnl_lock(subsys_id);
253 		ss = rcu_dereference_protected(table[subsys_id].subsys,
254 					       lockdep_is_held(&table[subsys_id].mutex));
255 		if (!ss)
256 #endif
257 		{
258 			nfnl_unlock(subsys_id);
259 			kfree_skb(nskb);
260 			return netlink_ack(skb, nlh, -EOPNOTSUPP);
261 		}
262 	}
263 
264 	if (!ss->commit || !ss->abort) {
265 		nfnl_unlock(subsys_id);
266 		kfree_skb(nskb);
267 		return netlink_ack(skb, nlh, -EOPNOTSUPP);
268 	}
269 
270 	while (skb->len >= nlmsg_total_size(0)) {
271 		int msglen, type;
272 
273 		nlh = nlmsg_hdr(skb);
274 		err = 0;
275 
276 		if (nlh->nlmsg_len < NLMSG_HDRLEN) {
277 			err = -EINVAL;
278 			goto ack;
279 		}
280 
281 		/* Only requests are handled by the kernel */
282 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
283 			err = -EINVAL;
284 			goto ack;
285 		}
286 
287 		type = nlh->nlmsg_type;
288 		if (type == NFNL_MSG_BATCH_BEGIN) {
289 			/* Malformed: Batch begin twice */
290 			success = false;
291 			goto done;
292 		} else if (type == NFNL_MSG_BATCH_END) {
293 			done = true;
294 			goto done;
295 		} else if (type < NLMSG_MIN_TYPE) {
296 			err = -EINVAL;
297 			goto ack;
298 		}
299 
300 		/* We only accept a batch with messages for the same
301 		 * subsystem.
302 		 */
303 		if (NFNL_SUBSYS_ID(type) != subsys_id) {
304 			err = -EINVAL;
305 			goto ack;
306 		}
307 
308 		nc = nfnetlink_find_client(type, ss);
309 		if (!nc) {
310 			err = -EINVAL;
311 			goto ack;
312 		}
313 
314 		{
315 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
316 			u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
317 			struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
318 			struct nlattr *attr = (void *)nlh + min_len;
319 			int attrlen = nlh->nlmsg_len - min_len;
320 
321 			err = nla_parse(cda, ss->cb[cb_id].attr_count,
322 					attr, attrlen, ss->cb[cb_id].policy);
323 			if (err < 0)
324 				goto ack;
325 
326 			if (nc->call_batch) {
327 				err = nc->call_batch(net->nfnl, skb, nlh,
328 						     (const struct nlattr **)cda);
329 			}
330 
331 			/* The lock was released to autoload some module, we
332 			 * have to abort and start from scratch using the
333 			 * original skb.
334 			 */
335 			if (err == -EAGAIN) {
336 				ss->abort(skb);
337 				nfnl_unlock(subsys_id);
338 				kfree_skb(nskb);
339 				goto replay;
340 			}
341 		}
342 ack:
343 		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
344 			/* We don't stop processing the batch on errors, thus,
345 			 * userspace gets all the errors that the batch
346 			 * triggers.
347 			 */
348 			netlink_ack(skb, nlh, err);
349 			if (err)
350 				success = false;
351 		}
352 
353 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
354 		if (msglen > skb->len)
355 			msglen = skb->len;
356 		skb_pull(skb, msglen);
357 	}
358 done:
359 	if (success && done)
360 		ss->commit(skb);
361 	else
362 		ss->abort(skb);
363 
364 	nfnl_unlock(subsys_id);
365 	kfree_skb(nskb);
366 }
367 
368 static void nfnetlink_rcv(struct sk_buff *skb)
369 {
370 	struct nlmsghdr *nlh = nlmsg_hdr(skb);
371 	struct net *net = sock_net(skb->sk);
372 	int msglen;
373 
374 	if (nlh->nlmsg_len < NLMSG_HDRLEN ||
375 	    skb->len < nlh->nlmsg_len)
376 		return;
377 
378 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) {
379 		netlink_ack(skb, nlh, -EPERM);
380 		return;
381 	}
382 
383 	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
384 		struct nfgenmsg *nfgenmsg;
385 
386 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
387 		if (msglen > skb->len)
388 			msglen = skb->len;
389 
390 		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
391 		    skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
392 			return;
393 
394 		nfgenmsg = nlmsg_data(nlh);
395 		skb_pull(skb, msglen);
396 		nfnetlink_rcv_batch(skb, nlh, nfgenmsg->res_id);
397 	} else {
398 		netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
399 	}
400 }
401 
402 #ifdef CONFIG_MODULES
403 static void nfnetlink_bind(int group)
404 {
405 	const struct nfnetlink_subsystem *ss;
406 	int type = nfnl_group2type[group];
407 
408 	rcu_read_lock();
409 	ss = nfnetlink_get_subsys(type);
410 	if (!ss) {
411 		rcu_read_unlock();
412 		request_module("nfnetlink-subsys-%d", type);
413 		return;
414 	}
415 	rcu_read_unlock();
416 }
417 #endif
418 
419 static int __net_init nfnetlink_net_init(struct net *net)
420 {
421 	struct sock *nfnl;
422 	struct netlink_kernel_cfg cfg = {
423 		.groups	= NFNLGRP_MAX,
424 		.input	= nfnetlink_rcv,
425 #ifdef CONFIG_MODULES
426 		.bind	= nfnetlink_bind,
427 #endif
428 	};
429 
430 	nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
431 	if (!nfnl)
432 		return -ENOMEM;
433 	net->nfnl_stash = nfnl;
434 	rcu_assign_pointer(net->nfnl, nfnl);
435 	return 0;
436 }
437 
438 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
439 {
440 	struct net *net;
441 
442 	list_for_each_entry(net, net_exit_list, exit_list)
443 		RCU_INIT_POINTER(net->nfnl, NULL);
444 	synchronize_net();
445 	list_for_each_entry(net, net_exit_list, exit_list)
446 		netlink_kernel_release(net->nfnl_stash);
447 }
448 
449 static struct pernet_operations nfnetlink_net_ops = {
450 	.init		= nfnetlink_net_init,
451 	.exit_batch	= nfnetlink_net_exit_batch,
452 };
453 
454 static int __init nfnetlink_init(void)
455 {
456 	int i;
457 
458 	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
459 		mutex_init(&table[i].mutex);
460 
461 	pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
462 	return register_pernet_subsys(&nfnetlink_net_ops);
463 }
464 
465 static void __exit nfnetlink_exit(void)
466 {
467 	pr_info("Removing netfilter NETLINK layer.\n");
468 	unregister_pernet_subsys(&nfnetlink_net_ops);
469 }
470 module_init(nfnetlink_init);
471 module_exit(nfnetlink_exit);
472