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