xref: /openbmc/linux/net/netfilter/nfnetlink.c (revision e2c75e76) 
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 struct {
41 	struct mutex				mutex;
42 	const struct nfnetlink_subsystem __rcu	*subsys;
43 } table[NFNL_SUBSYS_COUNT];
44 
45 static const int nfnl_group2type[NFNLGRP_MAX+1] = {
46 	[NFNLGRP_CONNTRACK_NEW]		= NFNL_SUBSYS_CTNETLINK,
47 	[NFNLGRP_CONNTRACK_UPDATE]	= NFNL_SUBSYS_CTNETLINK,
48 	[NFNLGRP_CONNTRACK_DESTROY]	= NFNL_SUBSYS_CTNETLINK,
49 	[NFNLGRP_CONNTRACK_EXP_NEW]	= NFNL_SUBSYS_CTNETLINK_EXP,
50 	[NFNLGRP_CONNTRACK_EXP_UPDATE]	= NFNL_SUBSYS_CTNETLINK_EXP,
51 	[NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
52 	[NFNLGRP_NFTABLES]		= NFNL_SUBSYS_NFTABLES,
53 	[NFNLGRP_ACCT_QUOTA]		= NFNL_SUBSYS_ACCT,
54 	[NFNLGRP_NFTRACE]		= NFNL_SUBSYS_NFTABLES,
55 };
56 
57 void nfnl_lock(__u8 subsys_id)
58 {
59 	mutex_lock(&table[subsys_id].mutex);
60 }
61 EXPORT_SYMBOL_GPL(nfnl_lock);
62 
63 void nfnl_unlock(__u8 subsys_id)
64 {
65 	mutex_unlock(&table[subsys_id].mutex);
66 }
67 EXPORT_SYMBOL_GPL(nfnl_unlock);
68 
69 #ifdef CONFIG_PROVE_LOCKING
70 bool lockdep_nfnl_is_held(u8 subsys_id)
71 {
72 	return lockdep_is_held(&table[subsys_id].mutex);
73 }
74 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
75 #endif
76 
77 int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
78 {
79 	nfnl_lock(n->subsys_id);
80 	if (table[n->subsys_id].subsys) {
81 		nfnl_unlock(n->subsys_id);
82 		return -EBUSY;
83 	}
84 	rcu_assign_pointer(table[n->subsys_id].subsys, n);
85 	nfnl_unlock(n->subsys_id);
86 
87 	return 0;
88 }
89 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
90 
91 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
92 {
93 	nfnl_lock(n->subsys_id);
94 	table[n->subsys_id].subsys = NULL;
95 	nfnl_unlock(n->subsys_id);
96 	synchronize_rcu();
97 	return 0;
98 }
99 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
100 
101 static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
102 {
103 	u8 subsys_id = NFNL_SUBSYS_ID(type);
104 
105 	if (subsys_id >= NFNL_SUBSYS_COUNT)
106 		return NULL;
107 
108 	return rcu_dereference(table[subsys_id].subsys);
109 }
110 
111 static inline const struct nfnl_callback *
112 nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
113 {
114 	u8 cb_id = NFNL_MSG_TYPE(type);
115 
116 	if (cb_id >= ss->cb_count)
117 		return NULL;
118 
119 	return &ss->cb[cb_id];
120 }
121 
122 int nfnetlink_has_listeners(struct net *net, unsigned int group)
123 {
124 	return netlink_has_listeners(net->nfnl, group);
125 }
126 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
127 
128 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
129 		   unsigned int group, int echo, gfp_t flags)
130 {
131 	return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
132 }
133 EXPORT_SYMBOL_GPL(nfnetlink_send);
134 
135 int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
136 {
137 	return netlink_set_err(net->nfnl, portid, group, error);
138 }
139 EXPORT_SYMBOL_GPL(nfnetlink_set_err);
140 
141 int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
142 		      int flags)
143 {
144 	return netlink_unicast(net->nfnl, skb, portid, flags);
145 }
146 EXPORT_SYMBOL_GPL(nfnetlink_unicast);
147 
148 /* Process one complete nfnetlink message. */
149 static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
150 			     struct netlink_ext_ack *extack)
151 {
152 	struct net *net = sock_net(skb->sk);
153 	const struct nfnl_callback *nc;
154 	const struct nfnetlink_subsystem *ss;
155 	int type, err;
156 
157 	/* All the messages must at least contain nfgenmsg */
158 	if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
159 		return 0;
160 
161 	type = nlh->nlmsg_type;
162 replay:
163 	rcu_read_lock();
164 	ss = nfnetlink_get_subsys(type);
165 	if (!ss) {
166 #ifdef CONFIG_MODULES
167 		rcu_read_unlock();
168 		request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
169 		rcu_read_lock();
170 		ss = nfnetlink_get_subsys(type);
171 		if (!ss)
172 #endif
173 		{
174 			rcu_read_unlock();
175 			return -EINVAL;
176 		}
177 	}
178 
179 	nc = nfnetlink_find_client(type, ss);
180 	if (!nc) {
181 		rcu_read_unlock();
182 		return -EINVAL;
183 	}
184 
185 	{
186 		int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
187 		u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
188 		struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
189 		struct nlattr *attr = (void *)nlh + min_len;
190 		int attrlen = nlh->nlmsg_len - min_len;
191 		__u8 subsys_id = NFNL_SUBSYS_ID(type);
192 
193 		err = nla_parse(cda, ss->cb[cb_id].attr_count, attr, attrlen,
194 				ss->cb[cb_id].policy, extack);
195 		if (err < 0) {
196 			rcu_read_unlock();
197 			return err;
198 		}
199 
200 		if (nc->call_rcu) {
201 			err = nc->call_rcu(net, net->nfnl, skb, nlh,
202 					   (const struct nlattr **)cda,
203 					   extack);
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 					       extack);
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 	struct netlink_ext_ack	extack;
230 };
231 
232 static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
233 			const struct netlink_ext_ack *extack)
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 	nfnl_err->extack = *extack;
244 	list_add_tail(&nfnl_err->head, list);
245 
246 	return 0;
247 }
248 
249 static void nfnl_err_del(struct nfnl_err *nfnl_err)
250 {
251 	list_del(&nfnl_err->head);
252 	kfree(nfnl_err);
253 }
254 
255 static void nfnl_err_reset(struct list_head *err_list)
256 {
257 	struct nfnl_err *nfnl_err, *next;
258 
259 	list_for_each_entry_safe(nfnl_err, next, err_list, head)
260 		nfnl_err_del(nfnl_err);
261 }
262 
263 static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
264 {
265 	struct nfnl_err *nfnl_err, *next;
266 
267 	list_for_each_entry_safe(nfnl_err, next, err_list, head) {
268 		netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
269 			    &nfnl_err->extack);
270 		nfnl_err_del(nfnl_err);
271 	}
272 }
273 
274 enum {
275 	NFNL_BATCH_FAILURE	= (1 << 0),
276 	NFNL_BATCH_DONE		= (1 << 1),
277 	NFNL_BATCH_REPLAY	= (1 << 2),
278 };
279 
280 static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
281 				u16 subsys_id, u32 genid)
282 {
283 	struct sk_buff *oskb = skb;
284 	struct net *net = sock_net(skb->sk);
285 	const struct nfnetlink_subsystem *ss;
286 	const struct nfnl_callback *nc;
287 	struct netlink_ext_ack extack;
288 	LIST_HEAD(err_list);
289 	u32 status;
290 	int err;
291 
292 	if (subsys_id >= NFNL_SUBSYS_COUNT)
293 		return netlink_ack(skb, nlh, -EINVAL, NULL);
294 replay:
295 	status = 0;
296 
297 	skb = netlink_skb_clone(oskb, GFP_KERNEL);
298 	if (!skb)
299 		return netlink_ack(oskb, nlh, -ENOMEM, NULL);
300 
301 	nfnl_lock(subsys_id);
302 	ss = nfnl_dereference_protected(subsys_id);
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 = nfnl_dereference_protected(subsys_id);
309 		if (!ss)
310 #endif
311 		{
312 			nfnl_unlock(subsys_id);
313 			netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
314 			return kfree_skb(skb);
315 		}
316 	}
317 
318 	if (!ss->commit || !ss->abort) {
319 		nfnl_unlock(subsys_id);
320 		netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
321 		return kfree_skb(skb);
322 	}
323 
324 	if (genid && ss->valid_genid && !ss->valid_genid(net, genid)) {
325 		nfnl_unlock(subsys_id);
326 		netlink_ack(oskb, nlh, -ERESTART, NULL);
327 		return kfree_skb(skb);
328 	}
329 
330 	while (skb->len >= nlmsg_total_size(0)) {
331 		int msglen, type;
332 
333 		memset(&extack, 0, sizeof(extack));
334 		nlh = nlmsg_hdr(skb);
335 		err = 0;
336 
337 		if (nlh->nlmsg_len < NLMSG_HDRLEN ||
338 		    skb->len < nlh->nlmsg_len ||
339 		    nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
340 			nfnl_err_reset(&err_list);
341 			status |= NFNL_BATCH_FAILURE;
342 			goto done;
343 		}
344 
345 		/* Only requests are handled by the kernel */
346 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
347 			err = -EINVAL;
348 			goto ack;
349 		}
350 
351 		type = nlh->nlmsg_type;
352 		if (type == NFNL_MSG_BATCH_BEGIN) {
353 			/* Malformed: Batch begin twice */
354 			nfnl_err_reset(&err_list);
355 			status |= NFNL_BATCH_FAILURE;
356 			goto done;
357 		} else if (type == NFNL_MSG_BATCH_END) {
358 			status |= NFNL_BATCH_DONE;
359 			goto done;
360 		} else if (type < NLMSG_MIN_TYPE) {
361 			err = -EINVAL;
362 			goto ack;
363 		}
364 
365 		/* We only accept a batch with messages for the same
366 		 * subsystem.
367 		 */
368 		if (NFNL_SUBSYS_ID(type) != subsys_id) {
369 			err = -EINVAL;
370 			goto ack;
371 		}
372 
373 		nc = nfnetlink_find_client(type, ss);
374 		if (!nc) {
375 			err = -EINVAL;
376 			goto ack;
377 		}
378 
379 		{
380 			int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
381 			u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
382 			struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
383 			struct nlattr *attr = (void *)nlh + min_len;
384 			int attrlen = nlh->nlmsg_len - min_len;
385 
386 			err = nla_parse(cda, ss->cb[cb_id].attr_count, attr,
387 					attrlen, ss->cb[cb_id].policy, NULL);
388 			if (err < 0)
389 				goto ack;
390 
391 			if (nc->call_batch) {
392 				err = nc->call_batch(net, net->nfnl, skb, nlh,
393 						     (const struct nlattr **)cda,
394 						     &extack);
395 			}
396 
397 			/* The lock was released to autoload some module, we
398 			 * have to abort and start from scratch using the
399 			 * original skb.
400 			 */
401 			if (err == -EAGAIN) {
402 				status |= NFNL_BATCH_REPLAY;
403 				goto next;
404 			}
405 		}
406 ack:
407 		if (nlh->nlmsg_flags & NLM_F_ACK || err) {
408 			/* Errors are delivered once the full batch has been
409 			 * processed, this avoids that the same error is
410 			 * reported several times when replaying the batch.
411 			 */
412 			if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
413 				/* We failed to enqueue an error, reset the
414 				 * list of errors and send OOM to userspace
415 				 * pointing to the batch header.
416 				 */
417 				nfnl_err_reset(&err_list);
418 				netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
419 					    NULL);
420 				status |= NFNL_BATCH_FAILURE;
421 				goto done;
422 			}
423 			/* We don't stop processing the batch on errors, thus,
424 			 * userspace gets all the errors that the batch
425 			 * triggers.
426 			 */
427 			if (err)
428 				status |= NFNL_BATCH_FAILURE;
429 		}
430 next:
431 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
432 		if (msglen > skb->len)
433 			msglen = skb->len;
434 		skb_pull(skb, msglen);
435 	}
436 done:
437 	if (status & NFNL_BATCH_REPLAY) {
438 		ss->abort(net, oskb);
439 		nfnl_err_reset(&err_list);
440 		nfnl_unlock(subsys_id);
441 		kfree_skb(skb);
442 		goto replay;
443 	} else if (status == NFNL_BATCH_DONE) {
444 		ss->commit(net, oskb);
445 	} else {
446 		ss->abort(net, oskb);
447 	}
448 
449 	nfnl_err_deliver(&err_list, oskb);
450 	nfnl_unlock(subsys_id);
451 	kfree_skb(skb);
452 }
453 
454 static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
455 	[NFNL_BATCH_GENID]	= { .type = NLA_U32 },
456 };
457 
458 static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
459 {
460 	int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
461 	struct nlattr *attr = (void *)nlh + min_len;
462 	struct nlattr *cda[NFNL_BATCH_MAX + 1];
463 	int attrlen = nlh->nlmsg_len - min_len;
464 	struct nfgenmsg *nfgenmsg;
465 	int msglen, err;
466 	u32 gen_id = 0;
467 	u16 res_id;
468 
469 	msglen = NLMSG_ALIGN(nlh->nlmsg_len);
470 	if (msglen > skb->len)
471 		msglen = skb->len;
472 
473 	if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
474 		return;
475 
476 	err = nla_parse(cda, NFNL_BATCH_MAX, attr, attrlen, nfnl_batch_policy,
477 			NULL);
478 	if (err < 0) {
479 		netlink_ack(skb, nlh, err, NULL);
480 		return;
481 	}
482 	if (cda[NFNL_BATCH_GENID])
483 		gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
484 
485 	nfgenmsg = nlmsg_data(nlh);
486 	skb_pull(skb, msglen);
487 	/* Work around old nft using host byte order */
488 	if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
489 		res_id = NFNL_SUBSYS_NFTABLES;
490 	else
491 		res_id = ntohs(nfgenmsg->res_id);
492 
493 	nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
494 }
495 
496 static void nfnetlink_rcv(struct sk_buff *skb)
497 {
498 	struct nlmsghdr *nlh = nlmsg_hdr(skb);
499 
500 	if (skb->len < NLMSG_HDRLEN ||
501 	    nlh->nlmsg_len < NLMSG_HDRLEN ||
502 	    skb->len < nlh->nlmsg_len)
503 		return;
504 
505 	if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
506 		netlink_ack(skb, nlh, -EPERM, NULL);
507 		return;
508 	}
509 
510 	if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
511 		nfnetlink_rcv_skb_batch(skb, nlh);
512 	else
513 		netlink_rcv_skb(skb, nfnetlink_rcv_msg);
514 }
515 
516 #ifdef CONFIG_MODULES
517 static int nfnetlink_bind(struct net *net, int group)
518 {
519 	const struct nfnetlink_subsystem *ss;
520 	int type;
521 
522 	if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
523 		return 0;
524 
525 	type = nfnl_group2type[group];
526 
527 	rcu_read_lock();
528 	ss = nfnetlink_get_subsys(type << 8);
529 	rcu_read_unlock();
530 	if (!ss)
531 		request_module("nfnetlink-subsys-%d", type);
532 	return 0;
533 }
534 #endif
535 
536 static int __net_init nfnetlink_net_init(struct net *net)
537 {
538 	struct sock *nfnl;
539 	struct netlink_kernel_cfg cfg = {
540 		.groups	= NFNLGRP_MAX,
541 		.input	= nfnetlink_rcv,
542 #ifdef CONFIG_MODULES
543 		.bind	= nfnetlink_bind,
544 #endif
545 	};
546 
547 	nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
548 	if (!nfnl)
549 		return -ENOMEM;
550 	net->nfnl_stash = nfnl;
551 	rcu_assign_pointer(net->nfnl, nfnl);
552 	return 0;
553 }
554 
555 static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
556 {
557 	struct net *net;
558 
559 	list_for_each_entry(net, net_exit_list, exit_list)
560 		RCU_INIT_POINTER(net->nfnl, NULL);
561 	synchronize_net();
562 	list_for_each_entry(net, net_exit_list, exit_list)
563 		netlink_kernel_release(net->nfnl_stash);
564 }
565 
566 static struct pernet_operations nfnetlink_net_ops = {
567 	.init		= nfnetlink_net_init,
568 	.exit_batch	= nfnetlink_net_exit_batch,
569 };
570 
571 static int __init nfnetlink_init(void)
572 {
573 	int i;
574 
575 	for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
576 		BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
577 
578 	for (i=0; i<NFNL_SUBSYS_COUNT; i++)
579 		mutex_init(&table[i].mutex);
580 
581 	return register_pernet_subsys(&nfnetlink_net_ops);
582 }
583 
584 static void __exit nfnetlink_exit(void)
585 {
586 	unregister_pernet_subsys(&nfnetlink_net_ops);
587 }
588 module_init(nfnetlink_init);
589 module_exit(nfnetlink_exit);
590