projects/linux·blob master · refs

819 lines · 19.4 KiB

1// SPDX-License-Identifier: GPL-2.0
2/* Netfilter messages via netlink socket. Allows for user space
3 * protocol helpers and general trouble making from userspace.
4 *
5 * (C) 2001 by Jay Schulist <[email protected]>,
6 * (C) 2002-2005 by Harald Welte <[email protected]>
7 * (C) 2005-2017 by Pablo Neira Ayuso <[email protected]>
8 *
9 * Initial netfilter messages via netlink development funded and
10 * generally made possible by Network Robots, Inc. (www.networkrobots.com)
11 *
12 * Further development of this code funded by Astaro AG (http://www.astaro.com)
13 */
14
15#include <linux/module.h>
16#include <linux/types.h>
17#include <linux/socket.h>
18#include <linux/kernel.h>
19#include <linux/string.h>
20#include <linux/sockios.h>
21#include <linux/net.h>
22#include <linux/skbuff.h>
23#include <linux/uaccess.h>
24#include <net/sock.h>
25#include <linux/init.h>
26#include <linux/sched/signal.h>
27
28#include <net/netlink.h>
29#include <net/netns/generic.h>
30#include <linux/netfilter.h>
31#include <linux/netfilter/nfnetlink.h>
32
33MODULE_LICENSE("GPL");
34MODULE_AUTHOR("Harald Welte <[email protected]>");
35MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
36MODULE_DESCRIPTION("Netfilter messages via netlink socket");
37
38#define nfnl_dereference_protected(id) \
39 rcu_dereference_protected(table[(id)].subsys, \
40 lockdep_nfnl_is_held((id)))
41
42#define NFNL_MAX_ATTR_COUNT 32
43
44static unsigned int nfnetlink_pernet_id __read_mostly;
45
46#ifdef CONFIG_NF_CONNTRACK_EVENTS
47static DEFINE_SPINLOCK(nfnl_grp_active_lock);
48#endif
49
50struct nfnl_net {
51 struct sock *nfnl;
52};
53
54static struct {
55 struct mutex mutex;
56 const struct nfnetlink_subsystem __rcu *subsys;
57} table[NFNL_SUBSYS_COUNT];
58
59static struct lock_class_key nfnl_lockdep_keys[NFNL_SUBSYS_COUNT];
60
61static const char *const nfnl_lockdep_names[NFNL_SUBSYS_COUNT] = {
62 [NFNL_SUBSYS_NONE] = "nfnl_subsys_none",
63 [NFNL_SUBSYS_CTNETLINK] = "nfnl_subsys_ctnetlink",
64 [NFNL_SUBSYS_CTNETLINK_EXP] = "nfnl_subsys_ctnetlink_exp",
65 [NFNL_SUBSYS_QUEUE] = "nfnl_subsys_queue",
66 [NFNL_SUBSYS_ULOG] = "nfnl_subsys_ulog",
67 [NFNL_SUBSYS_OSF] = "nfnl_subsys_osf",
68 [NFNL_SUBSYS_IPSET] = "nfnl_subsys_ipset",
69 [NFNL_SUBSYS_ACCT] = "nfnl_subsys_acct",
70 [NFNL_SUBSYS_CTNETLINK_TIMEOUT] = "nfnl_subsys_cttimeout",
71 [NFNL_SUBSYS_CTHELPER] = "nfnl_subsys_cthelper",
72 [NFNL_SUBSYS_NFTABLES] = "nfnl_subsys_nftables",
73 [NFNL_SUBSYS_NFT_COMPAT] = "nfnl_subsys_nftcompat",
74 [NFNL_SUBSYS_HOOK] = "nfnl_subsys_hook",
75};
76
77static const int nfnl_group2type[NFNLGRP_MAX+1] = {
78 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
79 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
80 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
81 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
82 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
83 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
84 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
85 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
86 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
87};
88
89static struct nfnl_net *nfnl_pernet(struct net *net)
90{
91 return net_generic(net, nfnetlink_pernet_id);
92}
93
94void nfnl_lock(__u8 subsys_id)
95{
96 mutex_lock(&table[subsys_id].mutex);
97}
98EXPORT_SYMBOL_GPL(nfnl_lock);
99
100void nfnl_unlock(__u8 subsys_id)
101{
102 mutex_unlock(&table[subsys_id].mutex);
103}
104EXPORT_SYMBOL_GPL(nfnl_unlock);
105
106#ifdef CONFIG_PROVE_LOCKING
107bool lockdep_nfnl_is_held(u8 subsys_id)
108{
109 return lockdep_is_held(&table[subsys_id].mutex);
110}
111EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
112#endif
113
114int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
115{
116 u8 cb_id;
117
118 /* Sanity-check attr_count size to avoid stack buffer overflow. */
119 for (cb_id = 0; cb_id < n->cb_count; cb_id++)
120 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
121 return -EINVAL;
122
123 nfnl_lock(n->subsys_id);
124 if (table[n->subsys_id].subsys) {
125 nfnl_unlock(n->subsys_id);
126 return -EBUSY;
127 }
128 rcu_assign_pointer(table[n->subsys_id].subsys, n);
129 nfnl_unlock(n->subsys_id);
130
131 return 0;
132}
133EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
134
135int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
136{
137 nfnl_lock(n->subsys_id);
138 table[n->subsys_id].subsys = NULL;
139 nfnl_unlock(n->subsys_id);
140 synchronize_rcu();
141 return 0;
142}
143EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
144
145static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
146{
147 u8 subsys_id = NFNL_SUBSYS_ID(type);
148
149 if (subsys_id >= NFNL_SUBSYS_COUNT)
150 return NULL;
151
152 return rcu_dereference(table[subsys_id].subsys);
153}
154
155static inline const struct nfnl_callback *
156nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
157{
158 u8 cb_id = NFNL_MSG_TYPE(type);
159
160 if (cb_id >= ss->cb_count)
161 return NULL;
162
163 return &ss->cb[cb_id];
164}
165
166int nfnetlink_has_listeners(struct net *net, unsigned int group)
167{
168 struct nfnl_net *nfnlnet = nfnl_pernet(net);
169
170 return netlink_has_listeners(nfnlnet->nfnl, group);
171}
172EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
173
174int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
175 unsigned int group, int echo, gfp_t flags)
176{
177 struct nfnl_net *nfnlnet = nfnl_pernet(net);
178
179 return nlmsg_notify(nfnlnet->nfnl, skb, portid, group, echo, flags);
180}
181EXPORT_SYMBOL_GPL(nfnetlink_send);
182
183int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
184{
185 struct nfnl_net *nfnlnet = nfnl_pernet(net);
186
187 return netlink_set_err(nfnlnet->nfnl, portid, group, error);
188}
189EXPORT_SYMBOL_GPL(nfnetlink_set_err);
190
191int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
192{
193 struct nfnl_net *nfnlnet = nfnl_pernet(net);
194 int err;
195
196 err = nlmsg_unicast(nfnlnet->nfnl, skb, portid);
197 if (err == -EAGAIN)
198 err = -ENOBUFS;
199
200 return err;
201}
202EXPORT_SYMBOL_GPL(nfnetlink_unicast);
203
204void nfnetlink_broadcast(struct net *net, struct sk_buff *skb, __u32 portid,
205 __u32 group, gfp_t allocation)
206{
207 struct nfnl_net *nfnlnet = nfnl_pernet(net);
208
209 netlink_broadcast(nfnlnet->nfnl, skb, portid, group, allocation);
210}
211EXPORT_SYMBOL_GPL(nfnetlink_broadcast);
212
213/* Process one complete nfnetlink message. */
214static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
215 struct netlink_ext_ack *extack)
216{
217 struct net *net = sock_net(skb->sk);
218 const struct nfnl_callback *nc;
219 const struct nfnetlink_subsystem *ss;
220 int type, err;
221
222 /* All the messages must at least contain nfgenmsg */
223 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
224 return 0;
225
226 type = nlh->nlmsg_type;
227replay:
228 rcu_read_lock();
229
230 ss = nfnetlink_get_subsys(type);
231 if (!ss) {
232#ifdef CONFIG_MODULES
233 rcu_read_unlock();
234 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
235 rcu_read_lock();
236 ss = nfnetlink_get_subsys(type);
237 if (!ss)
238#endif
239 {
240 rcu_read_unlock();
241 return -EINVAL;
242 }
243 }
244
245 nc = nfnetlink_find_client(type, ss);
246 if (!nc) {
247 rcu_read_unlock();
248 return -EINVAL;
249 }
250
251 {
252 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
253 struct nfnl_net *nfnlnet = nfnl_pernet(net);
254 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
255 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
256 struct nlattr *attr = (void *)nlh + min_len;
257 int attrlen = nlh->nlmsg_len - min_len;
258 __u8 subsys_id = NFNL_SUBSYS_ID(type);
259 struct nfnl_info info = {
260 .net = net,
261 .sk = nfnlnet->nfnl,
262 .nlh = nlh,
263 .nfmsg = nlmsg_data(nlh),
264 .extack = extack,
265 };
266
267 /* Sanity-check NFNL_MAX_ATTR_COUNT */
268 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
269 rcu_read_unlock();
270 return -ENOMEM;
271 }
272
273 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
274 attr, attrlen,
275 ss->cb[cb_id].policy, extack);
276 if (err < 0) {
277 rcu_read_unlock();
278 return err;
279 }
280
281 if (!nc->call) {
282 rcu_read_unlock();
283 return -EINVAL;
284 }
285
286 switch (nc->type) {
287 case NFNL_CB_RCU:
288 err = nc->call(skb, &info, (const struct nlattr **)cda);
289 rcu_read_unlock();
290 break;
291 case NFNL_CB_MUTEX:
292 rcu_read_unlock();
293 nfnl_lock(subsys_id);
294 if (nfnl_dereference_protected(subsys_id) != ss ||
295 nfnetlink_find_client(type, ss) != nc) {
296 nfnl_unlock(subsys_id);
297 err = -EAGAIN;
298 break;
299 }
300 err = nc->call(skb, &info, (const struct nlattr **)cda);
301 nfnl_unlock(subsys_id);
302 break;
303 default:
304 rcu_read_unlock();
305 err = -EINVAL;
306 break;
307 }
308 if (err == -EAGAIN)
309 goto replay;
310 return err;
311 }
312}
313
314struct nfnl_err {
315 struct list_head head;
316 struct nlmsghdr *nlh;
317 int err;
318 struct netlink_ext_ack extack;
319};
320
321static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
322 const struct netlink_ext_ack *extack)
323{
324 struct nfnl_err *nfnl_err;
325
326 nfnl_err = kmalloc_obj(struct nfnl_err);
327 if (nfnl_err == NULL)
328 return -ENOMEM;
329
330 nfnl_err->nlh = nlh;
331 nfnl_err->err = err;
332 nfnl_err->extack = *extack;
333 list_add_tail(&nfnl_err->head, list);
334
335 return 0;
336}
337
338static void nfnl_err_del(struct nfnl_err *nfnl_err)
339{
340 list_del(&nfnl_err->head);
341 kfree(nfnl_err);
342}
343
344static void nfnl_err_reset(struct list_head *err_list)
345{
346 struct nfnl_err *nfnl_err, *next;
347
348 list_for_each_entry_safe(nfnl_err, next, err_list, head)
349 nfnl_err_del(nfnl_err);
350}
351
352static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
353{
354 struct nfnl_err *nfnl_err, *next;
355
356 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
357 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
358 &nfnl_err->extack);
359 nfnl_err_del(nfnl_err);
360 }
361}
362
363enum {
364 NFNL_BATCH_FAILURE = (1 << 0),
365 NFNL_BATCH_DONE = (1 << 1),
366 NFNL_BATCH_REPLAY = (1 << 2),
367};
368
369static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
370 u16 subsys_id, u32 genid)
371{
372 struct sk_buff *oskb = skb;
373 struct net *net = sock_net(skb->sk);
374 const struct nfnetlink_subsystem *ss;
375 const struct nfnl_callback *nc;
376 struct netlink_ext_ack extack;
377 struct nlmsghdr *onlh = nlh;
378 LIST_HEAD(err_list);
379 u32 status;
380 int err;
381
382 if (subsys_id >= NFNL_SUBSYS_COUNT)
383 return netlink_ack(skb, nlh, -EINVAL, NULL);
384replay:
385 status = 0;
386replay_abort:
387 skb = netlink_skb_clone(oskb, GFP_KERNEL);
388 nlh = onlh;
389 if (!skb)
390 return netlink_ack(oskb, nlh, -ENOMEM, NULL);
391
392 nfnl_lock(subsys_id);
393 ss = nfnl_dereference_protected(subsys_id);
394 if (!ss) {
395#ifdef CONFIG_MODULES
396 nfnl_unlock(subsys_id);
397 request_module("nfnetlink-subsys-%d", subsys_id);
398 nfnl_lock(subsys_id);
399 ss = nfnl_dereference_protected(subsys_id);
400 if (!ss)
401#endif
402 {
403 nfnl_unlock(subsys_id);
404 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
405 return consume_skb(skb);
406 }
407 }
408
409 if (!ss->valid_genid || !ss->commit || !ss->abort) {
410 nfnl_unlock(subsys_id);
411 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
412 return consume_skb(skb);
413 }
414
415 if (!try_module_get(ss->owner)) {
416 nfnl_unlock(subsys_id);
417 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
418 return consume_skb(skb);
419 }
420
421 if (!ss->valid_genid(net, genid)) {
422 module_put(ss->owner);
423 nfnl_unlock(subsys_id);
424 netlink_ack(oskb, nlh, -ERESTART, NULL);
425 return consume_skb(skb);
426 }
427
428 nfnl_unlock(subsys_id);
429
430 if (nlh->nlmsg_flags & NLM_F_ACK) {
431 memset(&extack, 0, sizeof(extack));
432 nfnl_err_add(&err_list, nlh, 0, &extack);
433 }
434
435 while (skb->len >= nlmsg_total_size(0)) {
436 int msglen, type;
437
438 if (fatal_signal_pending(current)) {
439 nfnl_err_reset(&err_list);
440 err = -EINTR;
441 status = NFNL_BATCH_FAILURE;
442 goto done;
443 }
444
445 memset(&extack, 0, sizeof(extack));
446 nlh = nlmsg_hdr(skb);
447 err = 0;
448
449 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
450 skb->len < nlh->nlmsg_len ||
451 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
452 nfnl_err_reset(&err_list);
453 status |= NFNL_BATCH_FAILURE;
454 goto done;
455 }
456
457 /* Only requests are handled by the kernel */
458 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
459 err = -EINVAL;
460 goto ack;
461 }
462
463 type = nlh->nlmsg_type;
464 if (type == NFNL_MSG_BATCH_BEGIN) {
465 /* Malformed: Batch begin twice */
466 nfnl_err_reset(&err_list);
467 status |= NFNL_BATCH_FAILURE;
468 goto done;
469 } else if (type == NFNL_MSG_BATCH_END) {
470 status |= NFNL_BATCH_DONE;
471 goto done;
472 } else if (type < NLMSG_MIN_TYPE) {
473 err = -EINVAL;
474 goto ack;
475 }
476
477 /* We only accept a batch with messages for the same
478 * subsystem.
479 */
480 if (NFNL_SUBSYS_ID(type) != subsys_id) {
481 err = -EINVAL;
482 goto ack;
483 }
484
485 nc = nfnetlink_find_client(type, ss);
486 if (!nc) {
487 err = -EINVAL;
488 goto ack;
489 }
490
491 if (nc->type != NFNL_CB_BATCH) {
492 err = -EINVAL;
493 goto ack;
494 }
495
496 {
497 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
498 struct nfnl_net *nfnlnet = nfnl_pernet(net);
499 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
500 struct nlattr *attr = (void *)nlh + min_len;
501 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
502 int attrlen = nlh->nlmsg_len - min_len;
503 struct nfnl_info info = {
504 .net = net,
505 .sk = nfnlnet->nfnl,
506 .nlh = nlh,
507 .nfmsg = nlmsg_data(nlh),
508 .extack = &extack,
509 };
510
511 /* Sanity-check NFTA_MAX_ATTR */
512 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
513 err = -ENOMEM;
514 goto ack;
515 }
516
517 err = nla_parse_deprecated(cda,
518 ss->cb[cb_id].attr_count,
519 attr, attrlen,
520 ss->cb[cb_id].policy, &extack);
521 if (err < 0)
522 goto ack;
523
524 err = nc->call(skb, &info, (const struct nlattr **)cda);
525
526 /* The lock was released to autoload some module, we
527 * have to abort and start from scratch using the
528 * original skb.
529 */
530 if (err == -EAGAIN) {
531 status |= NFNL_BATCH_REPLAY;
532 goto done;
533 }
534 }
535ack:
536 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
537 /* Errors are delivered once the full batch has been
538 * processed, this avoids that the same error is
539 * reported several times when replaying the batch.
540 */
541 if (err == -ENOMEM ||
542 nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
543 /* We failed to enqueue an error, reset the
544 * list of errors and send OOM to userspace
545 * pointing to the batch header.
546 */
547 nfnl_err_reset(&err_list);
548 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
549 NULL);
550 status |= NFNL_BATCH_FAILURE;
551 goto done;
552 }
553 /* We don't stop processing the batch on errors, thus,
554 * userspace gets all the errors that the batch
555 * triggers.
556 */
557 if (err)
558 status |= NFNL_BATCH_FAILURE;
559 }
560
561 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
562 if (msglen > skb->len)
563 msglen = skb->len;
564 skb_pull(skb, msglen);
565 }
566done:
567 if (status & NFNL_BATCH_REPLAY) {
568 ss->abort(net, oskb, NFNL_ABORT_AUTOLOAD);
569 nfnl_err_reset(&err_list);
570 consume_skb(skb);
571 module_put(ss->owner);
572 goto replay;
573 } else if (status == NFNL_BATCH_DONE) {
574 err = ss->commit(net, oskb);
575 if (err == -EAGAIN) {
576 status |= NFNL_BATCH_REPLAY;
577 goto done;
578 } else if (err) {
579 ss->abort(net, oskb, NFNL_ABORT_NONE);
580 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
581 } else if (nlh->nlmsg_flags & NLM_F_ACK) {
582 memset(&extack, 0, sizeof(extack));
583 nfnl_err_add(&err_list, nlh, 0, &extack);
584 }
585 } else {
586 enum nfnl_abort_action abort_action;
587
588 if (status & NFNL_BATCH_FAILURE)
589 abort_action = NFNL_ABORT_NONE;
590 else
591 abort_action = NFNL_ABORT_VALIDATE;
592
593 err = ss->abort(net, oskb, abort_action);
594 if (err == -EAGAIN) {
595 nfnl_err_reset(&err_list);
596 consume_skb(skb);
597 module_put(ss->owner);
598 status |= NFNL_BATCH_FAILURE;
599 goto replay_abort;
600 }
601 }
602
603 nfnl_err_deliver(&err_list, oskb);
604 consume_skb(skb);
605 module_put(ss->owner);
606}
607
608static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
609 [NFNL_BATCH_GENID] = { .type = NLA_U32 },
610};
611
612static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
613{
614 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
615 struct nlattr *attr = (void *)nlh + min_len;
616 struct nlattr *cda[NFNL_BATCH_MAX + 1];
617 int attrlen = nlh->nlmsg_len - min_len;
618 struct nfgenmsg *nfgenmsg;
619 int msglen, err;
620 u32 gen_id = 0;
621 u16 res_id;
622
623 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
624 if (msglen > skb->len)
625 msglen = skb->len;
626
627 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
628 return;
629
630 err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
631 nfnl_batch_policy, NULL);
632 if (err < 0) {
633 netlink_ack(skb, nlh, err, NULL);
634 return;
635 }
636 if (cda[NFNL_BATCH_GENID])
637 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
638
639 nfgenmsg = nlmsg_data(nlh);
640 skb_pull(skb, msglen);
641 /* Work around old nft using host byte order */
642 if (nfgenmsg->res_id == (__force __be16)NFNL_SUBSYS_NFTABLES)
643 res_id = NFNL_SUBSYS_NFTABLES;
644 else
645 res_id = ntohs(nfgenmsg->res_id);
646
647 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
648}
649
650static void nfnetlink_rcv(struct sk_buff *skb)
651{
652 struct nlmsghdr *nlh = nlmsg_hdr(skb);
653
654 if (skb->len < NLMSG_HDRLEN ||
655 nlh->nlmsg_len < NLMSG_HDRLEN ||
656 skb->len < nlh->nlmsg_len)
657 return;
658
659 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
660 netlink_ack(skb, nlh, -EPERM, NULL);
661 return;
662 }
663
664 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
665 nfnetlink_rcv_skb_batch(skb, nlh);
666 else
667 netlink_rcv_skb(skb, nfnetlink_rcv_msg);
668}
669
670static void nfnetlink_bind_event(struct net *net, unsigned int group)
671{
672#ifdef CONFIG_NF_CONNTRACK_EVENTS
673 int type, group_bit;
674 u8 v;
675
676 /* All NFNLGRP_CONNTRACK_* group bits fit into u8.
677 * The other groups are not relevant and can be ignored.
678 */
679 if (group >= 8)
680 return;
681
682 type = nfnl_group2type[group];
683
684 switch (type) {
685 case NFNL_SUBSYS_CTNETLINK:
686 break;
687 case NFNL_SUBSYS_CTNETLINK_EXP:
688 break;
689 default:
690 return;
691 }
692
693 group_bit = (1 << group);
694
695 spin_lock(&nfnl_grp_active_lock);
696 v = READ_ONCE(nf_ctnetlink_has_listener);
697 if ((v & group_bit) == 0) {
698 v |= group_bit;
699
700 /* read concurrently without nfnl_grp_active_lock held. */
701 WRITE_ONCE(nf_ctnetlink_has_listener, v);
702 }
703
704 spin_unlock(&nfnl_grp_active_lock);
705#endif
706}
707
708static int nfnetlink_bind(struct net *net, int group)
709{
710 const struct nfnetlink_subsystem *ss;
711 int type;
712
713 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
714 return 0;
715
716 type = nfnl_group2type[group];
717
718 rcu_read_lock();
719 ss = nfnetlink_get_subsys(type << 8);
720 rcu_read_unlock();
721 if (!ss)
722 request_module_nowait("nfnetlink-subsys-%d", type);
723
724 nfnetlink_bind_event(net, group);
725 return 0;
726}
727
728static void nfnetlink_unbind(struct net *net, int group)
729{
730#ifdef CONFIG_NF_CONNTRACK_EVENTS
731 int type, group_bit;
732
733 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
734 return;
735
736 type = nfnl_group2type[group];
737
738 switch (type) {
739 case NFNL_SUBSYS_CTNETLINK:
740 break;
741 case NFNL_SUBSYS_CTNETLINK_EXP:
742 break;
743 default:
744 return;
745 }
746
747 /* ctnetlink_has_listener is u8 */
748 if (group >= 8)
749 return;
750
751 group_bit = (1 << group);
752
753 spin_lock(&nfnl_grp_active_lock);
754 if (!nfnetlink_has_listeners(net, group)) {
755 u8 v = READ_ONCE(nf_ctnetlink_has_listener);
756
757 v &= ~group_bit;
758
759 /* read concurrently without nfnl_grp_active_lock held. */
760 WRITE_ONCE(nf_ctnetlink_has_listener, v);
761 }
762 spin_unlock(&nfnl_grp_active_lock);
763#endif
764}
765
766static int __net_init nfnetlink_net_init(struct net *net)
767{
768 struct nfnl_net *nfnlnet = nfnl_pernet(net);
769 struct netlink_kernel_cfg cfg = {
770 .groups = NFNLGRP_MAX,
771 .input = nfnetlink_rcv,
772 .bind = nfnetlink_bind,
773 .unbind = nfnetlink_unbind,
774 };
775
776 nfnlnet->nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
777 if (!nfnlnet->nfnl)
778 return -ENOMEM;
779 return 0;
780}
781
782static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
783{
784 struct nfnl_net *nfnlnet;
785 struct net *net;
786
787 list_for_each_entry(net, net_exit_list, exit_list) {
788 nfnlnet = nfnl_pernet(net);
789
790 netlink_kernel_release(nfnlnet->nfnl);
791 }
792}
793
794static struct pernet_operations nfnetlink_net_ops = {
795 .init = nfnetlink_net_init,
796 .exit_batch = nfnetlink_net_exit_batch,
797 .id = &nfnetlink_pernet_id,
798 .size = sizeof(struct nfnl_net),
799};
800
801static int __init nfnetlink_init(void)
802{
803 int i;
804
805 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
806 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
807
808 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
809 __mutex_init(&table[i].mutex, nfnl_lockdep_names[i], &nfnl_lockdep_keys[i]);
810
811 return register_pernet_subsys(&nfnetlink_net_ops);
812}
813
814static void __exit nfnetlink_exit(void)
815{
816 unregister_pernet_subsys(&nfnetlink_net_ops);
817}
818module_init(nfnetlink_init);
819module_exit(nfnetlink_exit);