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) 2005-2017 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
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 struct {
41 struct mutex mutex;
42 const struct nfnetlink_subsystem __rcu *subsys;
43} table[NFNL_SUBSYS_COUNT];
44
45static 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
57void nfnl_lock(__u8 subsys_id)
58{
59 mutex_lock(&table[subsys_id].mutex);
60}
61EXPORT_SYMBOL_GPL(nfnl_lock);
62
63void nfnl_unlock(__u8 subsys_id)
64{
65 mutex_unlock(&table[subsys_id].mutex);
66}
67EXPORT_SYMBOL_GPL(nfnl_unlock);
68
69#ifdef CONFIG_PROVE_LOCKING
70bool lockdep_nfnl_is_held(u8 subsys_id)
71{
72 return lockdep_is_held(&table[subsys_id].mutex);
73}
74EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
75#endif
76
77int 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}
89EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
90
91int 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}
99EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
100
101static 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
111static inline const struct nfnl_callback *
112nfnetlink_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
122int nfnetlink_has_listeners(struct net *net, unsigned int group)
123{
124 return netlink_has_listeners(net->nfnl, group);
125}
126EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
127
128int 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}
133EXPORT_SYMBOL_GPL(nfnetlink_send);
134
135int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
136{
137 return netlink_set_err(net->nfnl, portid, group, error);
138}
139EXPORT_SYMBOL_GPL(nfnetlink_set_err);
140
141int 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}
146EXPORT_SYMBOL_GPL(nfnetlink_unicast);
147
148/* Process one complete nfnetlink message. */
149static 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;
162replay:
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
225struct nfnl_err {
226 struct list_head head;
227 struct nlmsghdr *nlh;
228 int err;
229 struct netlink_ext_ack extack;
230};
231
232static 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
249static void nfnl_err_del(struct nfnl_err *nfnl_err)
250{
251 list_del(&nfnl_err->head);
252 kfree(nfnl_err);
253}
254
255static 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
263static 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
274enum {
275 NFNL_BATCH_FAILURE = (1 << 0),
276 NFNL_BATCH_DONE = (1 << 1),
277 NFNL_BATCH_REPLAY = (1 << 2),
278};
279
280static 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);
294replay:
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 }
406ack:
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 }
430next:
431 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
432 if (msglen > skb->len)
433 msglen = skb->len;
434 skb_pull(skb, msglen);
435 }
436done:
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
454static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
455 [NFNL_BATCH_GENID] = { .type = NLA_U32 },
456};
457
458static 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
496static 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
517static 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
536static 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
555static 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
566static struct pernet_operations nfnetlink_net_ops = {
567 .init = nfnetlink_net_init,
568 .exit_batch = nfnetlink_net_exit_batch,
569};
570
571static 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
584static void __exit nfnetlink_exit(void)
585{
586 unregister_pernet_subsys(&nfnetlink_net_ops);
587}
588module_init(nfnetlink_init);
589module_exit(nfnetlink_exit);