Loading...
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#include <linux/sched/signal.h>
29
30#include <net/netlink.h>
31#include <net/netns/generic.h>
32#include <linux/netfilter/nfnetlink.h>
33
34MODULE_LICENSE("GPL");
35MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
36MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
37MODULE_DESCRIPTION("Netfilter messages via netlink socket");
38
39#define nfnl_dereference_protected(id) \
40 rcu_dereference_protected(table[(id)].subsys, \
41 lockdep_nfnl_is_held((id)))
42
43#define NFNL_MAX_ATTR_COUNT 32
44
45static unsigned int nfnetlink_pernet_id __read_mostly;
46
47struct nfnl_net {
48 struct sock *nfnl;
49};
50
51static struct {
52 struct mutex mutex;
53 const struct nfnetlink_subsystem __rcu *subsys;
54} table[NFNL_SUBSYS_COUNT];
55
56static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
57
58static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = {
59 [NFNL_SUBSYS_NONE] = "nfnl_subsys_none",
60 [NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink",
61 [NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp",
62 [NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue",
63 [NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog",
64 [NFNL_SUBSYS_OSF] = "nfnl_subsys_osf",
65 [NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset",
66 [NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct",
67 [NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout",
68 [NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper",
69 [NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables",
70 [NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat",
71 [NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook",
72};
73
74static const int nfnl_group2type[NFNLGRP_MAX+1] = {
75 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
76 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
77 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
78 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
79 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
80 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
81 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
82 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
83 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
84};
85
86static struct nfnl_net *nfnl_pernet(struct net *net)
87{
88 return net_generic(net, nfnetlink_pernet_id);
89}
90
91void nfnl_lock(__u8 subsys_id)
92{
93 mutex_lock(&table[subsys_id].mutex);
94}
95EXPORT_SYMBOL_GPL(nfnl_lock);
96
97void nfnl_unlock(__u8 subsys_id)
98{
99 mutex_unlock(&table[subsys_id].mutex);
100}
101EXPORT_SYMBOL_GPL(nfnl_unlock);
102
103#ifdef CONFIG_PROVE_LOCKING
104bool lockdep_nfnl_is_held(u8 subsys_id)
105{
106 return lockdep_is_held(&table[subsys_id].mutex);
107}
108EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
109#endif
110
111int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
112{
113 u8 cb_id;
114
115 /* Sanity-check attr_count size to avoid stack buffer overflow. */
116 for (cb_id = 0; cb_id < n->cb_count; cb_id++)
117 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
118 return -EINVAL;
119
120 nfnl_lock(n->subsys_id);
121 if (table[n->subsys_id].subsys) {
122 nfnl_unlock(n->subsys_id);
123 return -EBUSY;
124 }
125 rcu_assign_pointer(table[n->subsys_id].subsys, n);
126 nfnl_unlock(n->subsys_id);
127
128 return 0;
129}
130EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
131
132int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
133{
134 nfnl_lock(n->subsys_id);
135 table[n->subsys_id].subsys = NULL;
136 nfnl_unlock(n->subsys_id);
137 synchronize_rcu();
138 return 0;
139}
140EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
141
142static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
143{
144 u8 subsys_id = NFNL_SUBSYS_ID(type);
145
146 if (subsys_id >= NFNL_SUBSYS_COUNT)
147 return NULL;
148
149 return rcu_dereference(table[subsys_id].subsys);
150}
151
152static inline const struct nfnl_callback *
153nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
154{
155 u8 cb_id = NFNL_MSG_TYPE(type);
156
157 if (cb_id >= ss->cb_count)
158 return NULL;
159
160 return &ss->cb[cb_id];
161}
162
163int nfnetlink_has_listeners(struct net *net, unsigned int group)
164{
165 struct nfnl_net *nfnlnet = nfnl_pernet(net);
166
167 return netlink_has_listeners(nfnlnet->nfnl, group);
168}
169EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
170
171int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
172 unsigned int group, int echo, gfp_t flags)
173{
174 struct nfnl_net *nfnlnet = nfnl_pernet(net);
175
176 return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
177}
178EXPORT_SYMBOL_GPL(nfnetlink_send);
179
180int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
181{
182 struct nfnl_net *nfnlnet = nfnl_pernet(net);
183
184 return netlink_set_err(nfnlnet->nfnl, portid, group, error);
185}
186EXPORT_SYMBOL_GPL(nfnetlink_set_err);
187
188int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
189{
190 struct nfnl_net *nfnlnet = nfnl_pernet(net);
191 int err;
192
193 err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
194 if (err == -EAGAIN)
195 err = -ENOBUFS;
196
197 return err;
198}
199EXPORT_SYMBOL_GPL(nfnetlink_unicast);
200
201void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
202 __u32 group, gfp_t allocation)
203{
204 struct nfnl_net *nfnlnet = nfnl_pernet(net);
205
206 netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
207}
208EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
209
210/* Process one complete nfnetlink message. */
211static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
212 struct netlink_ext_ack *extack)
213{
214 struct net *net = sock_net(skb->sk);
215 const struct nfnl_callback *nc;
216 const struct nfnetlink_subsystem *ss;
217 int type, err;
218
219 /* All the messages must at least contain nfgenmsg */
220 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
221 return 0;
222
223 type = nlh->nlmsg_type;
224replay:
225 rcu_read_lock();
226
227 ss = nfnetlink_get_subsys(type);
228 if (!ss) {
229#ifdef CONFIG_MODULES
230 rcu_read_unlock();
231 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
232 rcu_read_lock();
233 ss = nfnetlink_get_subsys(type);
234 if (!ss)
235#endif
236 {
237 rcu_read_unlock();
238 return -EINVAL;
239 }
240 }
241
242 nc = nfnetlink_find_client(type, ss);
243 if (!nc) {
244 rcu_read_unlock();
245 return -EINVAL;
246 }
247
248 {
249 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
250 struct nfnl_net *nfnlnet = nfnl_pernet(net);
251 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
252 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
253 struct nlattr *attr = (void *)nlh + min_len;
254 int attrlen = nlh->nlmsg_len - min_len;
255 __u8 subsys_id = NFNL_SUBSYS_ID(type);
256 struct nfnl_info info = {
257 .net = net,
258 .sk = nfnlnet->nfnl,
259 .nlh = nlh,
260 .nfmsg = nlmsg_data(nlh),
261 .extack = extack,
262 };
263
264 /* Sanity-check NFNL_MAX_ATTR_COUNT */
265 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
266 rcu_read_unlock();
267 return -ENOMEM;
268 }
269
270 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
271 attr, attrlen,
272 ss->cb[cb_id].policy, extack);
273 if (err < 0) {
274 rcu_read_unlock();
275 return err;
276 }
277
278 if (!nc->call) {
279 rcu_read_unlock();
280 return -EINVAL;
281 }
282
283 switch (nc->type) {
284 case NFNL_CB_RCU:
285 err = nc->call(skb, &info, (const struct nlattr **)cda);
286 rcu_read_unlock();
287 break;
288 case NFNL_CB_MUTEX:
289 rcu_read_unlock();
290 nfnl_lock(subsys_id);
291 if (nfnl_dereference_protected(subsys_id) != ss ||
292 nfnetlink_find_client(type, ss) != nc) {
293 err = -EAGAIN;
294 break;
295 }
296 err = nc->call(skb, &info, (const struct nlattr **)cda);
297 nfnl_unlock(subsys_id);
298 break;
299 default:
300 rcu_read_unlock();
301 err = -EINVAL;
302 break;
303 }
304 if (err == -EAGAIN)
305 goto replay;
306 return err;
307 }
308}
309
310struct nfnl_err {
311 struct list_head head;
312 struct nlmsghdr *nlh;
313 int err;
314 struct netlink_ext_ack extack;
315};
316
317static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
318 const struct netlink_ext_ack *extack)
319{
320 struct nfnl_err *nfnl_err;
321
322 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
323 if (nfnl_err == NULL)
324 return -ENOMEM;
325
326 nfnl_err->nlh = nlh;
327 nfnl_err->err = err;
328 nfnl_err->extack = *extack;
329 list_add_tail(&nfnl_err->head, list);
330
331 return 0;
332}
333
334static void nfnl_err_del(struct nfnl_err *nfnl_err)
335{
336 list_del(&nfnl_err->head);
337 kfree(nfnl_err);
338}
339
340static void nfnl_err_reset(struct list_head *err_list)
341{
342 struct nfnl_err *nfnl_err, *next;
343
344 list_for_each_entry_safe(nfnl_err, next, err_list, head)
345 nfnl_err_del(nfnl_err);
346}
347
348static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
349{
350 struct nfnl_err *nfnl_err, *next;
351
352 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
353 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
354 &nfnl_err->extack);
355 nfnl_err_del(nfnl_err);
356 }
357}
358
359enum {
360 NFNL_BATCH_FAILURE = (1 << 0),
361 NFNL_BATCH_DONE = (1 << 1),
362 NFNL_BATCH_REPLAY = (1 << 2),
363};
364
365static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
366 u16 subsys_id, u32 genid)
367{
368 struct sk_buff *oskb = skb;
369 struct net *net = sock_net(skb->sk);
370 const struct nfnetlink_subsystem *ss;
371 const struct nfnl_callback *nc;
372 struct netlink_ext_ack extack;
373 LIST_HEAD(err_list);
374 u32 status;
375 int err;
376
377 if (subsys_id >= NFNL_SUBSYS_COUNT)
378 return netlink_ack(skb, nlh, -EINVAL, NULL);
379replay:
380 status = 0;
381replay_abort:
382 skb = netlink_skb_clone(oskb, GFP_KERNEL);
383 if (!skb)
384 return netlink_ack(oskb, nlh, -ENOMEM, NULL);
385
386 nfnl_lock(subsys_id);
387 ss = nfnl_dereference_protected(subsys_id);
388 if (!ss) {
389#ifdef CONFIG_MODULES
390 nfnl_unlock(subsys_id);
391 request_module("nfnetlink-subsys-%d", subsys_id);
392 nfnl_lock(subsys_id);
393 ss = nfnl_dereference_protected(subsys_id);
394 if (!ss)
395#endif
396 {
397 nfnl_unlock(subsys_id);
398 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
399 return kfree_skb(skb);
400 }
401 }
402
403 if (!ss->valid_genid || !ss->commit || !ss->abort) {
404 nfnl_unlock(subsys_id);
405 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
406 return kfree_skb(skb);
407 }
408
409 if (!try_module_get(ss->owner)) {
410 nfnl_unlock(subsys_id);
411 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
412 return kfree_skb(skb);
413 }
414
415 if (!ss->valid_genid(net, genid)) {
416 module_put(ss->owner);
417 nfnl_unlock(subsys_id);
418 netlink_ack(oskb, nlh, -ERESTART, NULL);
419 return kfree_skb(skb);
420 }
421
422 nfnl_unlock(subsys_id);
423
424 while (skb->len >= nlmsg_total_size(0)) {
425 int msglen, type;
426
427 if (fatal_signal_pending(current)) {
428 nfnl_err_reset(&err_list);
429 err = -EINTR;
430 status = NFNL_BATCH_FAILURE;
431 goto done;
432 }
433
434 memset(&extack, 0, sizeof(extack));
435 nlh = nlmsg_hdr(skb);
436 err = 0;
437
438 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
439 skb->len < nlh->nlmsg_len ||
440 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
441 nfnl_err_reset(&err_list);
442 status |= NFNL_BATCH_FAILURE;
443 goto done;
444 }
445
446 /* Only requests are handled by the kernel */
447 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
448 err = -EINVAL;
449 goto ack;
450 }
451
452 type = nlh->nlmsg_type;
453 if (type == NFNL_MSG_BATCH_BEGIN) {
454 /* Malformed: Batch begin twice */
455 nfnl_err_reset(&err_list);
456 status |= NFNL_BATCH_FAILURE;
457 goto done;
458 } else if (type == NFNL_MSG_BATCH_END) {
459 status |= NFNL_BATCH_DONE;
460 goto done;
461 } else if (type < NLMSG_MIN_TYPE) {
462 err = -EINVAL;
463 goto ack;
464 }
465
466 /* We only accept a batch with messages for the same
467 * subsystem.
468 */
469 if (NFNL_SUBSYS_ID(type) != subsys_id) {
470 err = -EINVAL;
471 goto ack;
472 }
473
474 nc = nfnetlink_find_client(type, ss);
475 if (!nc) {
476 err = -EINVAL;
477 goto ack;
478 }
479
480 if (nc->type != NFNL_CB_BATCH) {
481 err = -EINVAL;
482 goto ack;
483 }
484
485 {
486 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
487 struct nfnl_net *nfnlnet = nfnl_pernet(net);
488 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
489 struct nlattr *attr = (void *)nlh + min_len;
490 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
491 int attrlen = nlh->nlmsg_len - min_len;
492 struct nfnl_info info = {
493 .net = net,
494 .sk = nfnlnet->nfnl,
495 .nlh = nlh,
496 .nfmsg = nlmsg_data(nlh),
497 .extack = &extack,
498 };
499
500 /* Sanity-check NFTA_MAX_ATTR */
501 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
502 err = -ENOMEM;
503 goto ack;
504 }
505
506 err = nla_parse_deprecated(cda,
507 ss->cb[cb_id].attr_count,
508 attr, attrlen,
509 ss->cb[cb_id].policy, NULL);
510 if (err < 0)
511 goto ack;
512
513 err = nc->call(skb, &info, (const struct nlattr **)cda);
514
515 /* The lock was released to autoload some module, we
516 * have to abort and start from scratch using the
517 * original skb.
518 */
519 if (err == -EAGAIN) {
520 status |= NFNL_BATCH_REPLAY;
521 goto done;
522 }
523 }
524ack:
525 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
526 /* Errors are delivered once the full batch has been
527 * processed, this avoids that the same error is
528 * reported several times when replaying the batch.
529 */
530 if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
531 /* We failed to enqueue an error, reset the
532 * list of errors and send OOM to userspace
533 * pointing to the batch header.
534 */
535 nfnl_err_reset(&err_list);
536 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
537 NULL);
538 status |= NFNL_BATCH_FAILURE;
539 goto done;
540 }
541 /* We don't stop processing the batch on errors, thus,
542 * userspace gets all the errors that the batch
543 * triggers.
544 */
545 if (err)
546 status |= NFNL_BATCH_FAILURE;
547 }
548
549 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
550 if (msglen > skb->len)
551 msglen = skb->len;
552 skb_pull(skb, msglen);
553 }
554done:
555 if (status & NFNL_BATCH_REPLAY) {
556 ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
557 nfnl_err_reset(&err_list);
558 kfree_skb(skb);
559 module_put(ss->owner);
560 goto replay;
561 } else if (status == NFNL_BATCH_DONE) {
562 err = ss->commit(net, oskb);
563 if (err == -EAGAIN) {
564 status |= NFNL_BATCH_REPLAY;
565 goto done;
566 } else if (err) {
567 ss->abort(net, oskb, NFNL_ABORT_NONE);
568 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
569 }
570 } else {
571 enum nfnl_abort_action abort_action;
572
573 if (status & NFNL_BATCH_FAILURE)
574 abort_action = NFNL_ABORT_NONE;
575 else
576 abort_action = NFNL_ABORT_VALIDATE;
577
578 err = ss->abort(net, oskb, abort_action);
579 if (err == -EAGAIN) {
580 nfnl_err_reset(&err_list);
581 kfree_skb(skb);
582 module_put(ss->owner);
583 status |= NFNL_BATCH_FAILURE;
584 goto replay_abort;
585 }
586 }
587 if (ss->cleanup)
588 ss->cleanup(net);
589
590 nfnl_err_deliver(&err_list, oskb);
591 kfree_skb(skb);
592 module_put(ss->owner);
593}
594
595static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
596 [NFNL_BATCH_GENID] = { .type = NLA_U32 },
597};
598
599static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
600{
601 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
602 struct nlattr *attr = (void *)nlh + min_len;
603 struct nlattr *cda[NFNL_BATCH_MAX + 1];
604 int attrlen = nlh->nlmsg_len - min_len;
605 struct nfgenmsg *nfgenmsg;
606 int msglen, err;
607 u32 gen_id = 0;
608 u16 res_id;
609
610 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
611 if (msglen > skb->len)
612 msglen = skb->len;
613
614 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
615 return;
616
617 err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
618 nfnl_batch_policy, NULL);
619 if (err < 0) {
620 netlink_ack(skb, nlh, err, NULL);
621 return;
622 }
623 if (cda[NFNL_BATCH_GENID])
624 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
625
626 nfgenmsg = nlmsg_data(nlh);
627 skb_pull(skb, msglen);
628 /* Work around old nft using host byte order */
629 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
630 res_id = NFNL_SUBSYS_NFTABLES;
631 else
632 res_id = ntohs(nfgenmsg->res_id);
633
634 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
635}
636
637static void nfnetlink_rcv(struct sk_buff *skb)
638{
639 struct nlmsghdr *nlh = nlmsg_hdr(skb);
640
641 if (skb->len < NLMSG_HDRLEN ||
642 nlh->nlmsg_len < NLMSG_HDRLEN ||
643 skb->len < nlh->nlmsg_len)
644 return;
645
646 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
647 netlink_ack(skb, nlh, -EPERM, NULL);
648 return;
649 }
650
651 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
652 nfnetlink_rcv_skb_batch(skb, nlh);
653 else
654 netlink_rcv_skb(skb, nfnetlink_rcv_msg);
655}
656
657#ifdef CONFIG_MODULES
658static int nfnetlink_bind(struct net *net, int group)
659{
660 const struct nfnetlink_subsystem *ss;
661 int type;
662
663 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
664 return 0;
665
666 type = nfnl_group2type[group];
667
668 rcu_read_lock();
669 ss = nfnetlink_get_subsys(type << 8);
670 rcu_read_unlock();
671 if (!ss)
672 request_module_nowait("nfnetlink-subsys-%d", type);
673 return 0;
674}
675#endif
676
677static int __net_init nfnetlink_net_init(struct net *net)
678{
679 struct nfnl_net *nfnlnet = nfnl_pernet(net);
680 struct netlink_kernel_cfg cfg = {
681 .groups = NFNLGRP_MAX,
682 .input = nfnetlink_rcv,
683#ifdef CONFIG_MODULES
684 .bind = nfnetlink_bind,
685#endif
686 };
687
688 nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
689 if (!nfnlnet->nfnl)
690 return -ENOMEM;
691 return 0;
692}
693
694static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
695{
696 struct nfnl_net *nfnlnet;
697 struct net *net;
698
699 list_for_each_entry(net, net_exit_list, exit_list) {
700 nfnlnet = nfnl_pernet(net);
701
702 netlink_kernel_release(nfnlnet->nfnl);
703 }
704}
705
706static struct pernet_operations nfnetlink_net_ops = {
707 .init = nfnetlink_net_init,
708 .exit_batch = nfnetlink_net_exit_batch,
709 .id = &nfnetlink_pernet_id,
710 .size = sizeof(struct nfnl_net),
711};
712
713static int __init nfnetlink_init(void)
714{
715 int i;
716
717 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
718 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
719
720 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
721 __mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
722
723 return register_pernet_subsys(&nfnetlink_net_ops);
724}
725
726static void __exit nfnetlink_exit(void)
727{
728 unregister_pernet_subsys(&nfnetlink_net_ops);
729}
730module_init(nfnetlink_init);
731module_exit(nfnetlink_exit);
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
32MODULE_LICENSE("GPL");
33MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
34MODULE_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
40static char __initdata nfversion[] = "0.30";
41
42static struct {
43 struct mutex mutex;
44 const struct nfnetlink_subsystem __rcu *subsys;
45} table[NFNL_SUBSYS_COUNT];
46
47static 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
59void nfnl_lock(__u8 subsys_id)
60{
61 mutex_lock(&table[subsys_id].mutex);
62}
63EXPORT_SYMBOL_GPL(nfnl_lock);
64
65void nfnl_unlock(__u8 subsys_id)
66{
67 mutex_unlock(&table[subsys_id].mutex);
68}
69EXPORT_SYMBOL_GPL(nfnl_unlock);
70
71#ifdef CONFIG_PROVE_LOCKING
72bool lockdep_nfnl_is_held(u8 subsys_id)
73{
74 return lockdep_is_held(&table[subsys_id].mutex);
75}
76EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
77#endif
78
79int 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}
91EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
92
93int 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}
101EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
102
103static 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
113static inline const struct nfnl_callback *
114nfnetlink_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
124int nfnetlink_has_listeners(struct net *net, unsigned int group)
125{
126 return netlink_has_listeners(net->nfnl, group);
127}
128EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
129
130int 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}
135EXPORT_SYMBOL_GPL(nfnetlink_send);
136
137int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
138{
139 return netlink_set_err(net->nfnl, portid, group, error);
140}
141EXPORT_SYMBOL_GPL(nfnetlink_set_err);
142
143int 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}
148EXPORT_SYMBOL_GPL(nfnetlink_unicast);
149
150/* Process one complete nfnetlink message. */
151static 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;
163replay:
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
224struct nfnl_err {
225 struct list_head head;
226 struct nlmsghdr *nlh;
227 int err;
228};
229
230static 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
245static void nfnl_err_del(struct nfnl_err *nfnl_err)
246{
247 list_del(&nfnl_err->head);
248 kfree(nfnl_err);
249}
250
251static 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
259static 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
269enum {
270 NFNL_BATCH_FAILURE = (1 << 0),
271 NFNL_BATCH_DONE = (1 << 1),
272 NFNL_BATCH_REPLAY = (1 << 2),
273};
274
275static 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);
288replay:
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 }
392ack:
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 }
415next:
416 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
417 if (msglen > skb->len)
418 msglen = skb->len;
419 skb_pull(skb, msglen);
420 }
421done:
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
439static 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
479static 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
498static 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
517static 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
528static struct pernet_operations nfnetlink_net_ops = {
529 .init = nfnetlink_net_init,
530 .exit_batch = nfnetlink_net_exit_batch,
531};
532
533static 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
547static void __exit nfnetlink_exit(void)
548{
549 pr_info("Removing netfilter NETLINK layer.\n");
550 unregister_pernet_subsys(&nfnetlink_net_ops);
551}
552module_init(nfnetlink_init);
553module_exit(nfnetlink_exit);