?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 5.7.7
[linux/fpc-iii.git] / net / netlink / af_netlink.c
blob5ded01ca8b20ca8e707877923180d6498019bf73
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * NETLINK Kernel-user communication protocol.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
8 *
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
19 */
20
21 #include <linux/module.h>
22
23 #include <linux/capability.h>
24 #include <linux/kernel.h>
25 #include <linux/init.h>
26 #include <linux/signal.h>
27 #include <linux/sched.h>
28 #include <linux/errno.h>
29 #include <linux/string.h>
30 #include <linux/stat.h>
31 #include <linux/socket.h>
32 #include <linux/un.h>
33 #include <linux/fcntl.h>
34 #include <linux/termios.h>
35 #include <linux/sockios.h>
36 #include <linux/net.h>
37 #include <linux/fs.h>
38 #include <linux/slab.h>
39 #include <linux/uaccess.h>
40 #include <linux/skbuff.h>
41 #include <linux/netdevice.h>
42 #include <linux/rtnetlink.h>
43 #include <linux/proc_fs.h>
44 #include <linux/seq_file.h>
45 #include <linux/notifier.h>
46 #include <linux/security.h>
47 #include <linux/jhash.h>
48 #include <linux/jiffies.h>
49 #include <linux/random.h>
50 #include <linux/bitops.h>
51 #include <linux/mm.h>
52 #include <linux/types.h>
53 #include <linux/audit.h>
54 #include <linux/mutex.h>
55 #include <linux/vmalloc.h>
56 #include <linux/if_arp.h>
57 #include <linux/rhashtable.h>
58 #include <asm/cacheflush.h>
59 #include <linux/hash.h>
60 #include <linux/genetlink.h>
61 #include <linux/net_namespace.h>
62 #include <linux/nospec.h>
63
64 #include <net/net_namespace.h>
65 #include <net/netns/generic.h>
66 #include <net/sock.h>
67 #include <net/scm.h>
68 #include <net/netlink.h>
69
70 #include "af_netlink.h"
71
72 struct listeners {
73 struct rcu_head rcu;
74 unsigned long masks[];
75 };
76
77 /* state bits */
78 #define NETLINK_S_CONGESTED 0x0
79
80 static inline int netlink_is_kernel(struct sock *sk)
81 {
82 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
83 }
84
85 struct netlink_table *nl_table __read_mostly;
86 EXPORT_SYMBOL_GPL(nl_table);
87
88 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
89
90 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
91
92 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
93 "nlk_cb_mutex-ROUTE",
94 "nlk_cb_mutex-1",
95 "nlk_cb_mutex-USERSOCK",
96 "nlk_cb_mutex-FIREWALL",
97 "nlk_cb_mutex-SOCK_DIAG",
98 "nlk_cb_mutex-NFLOG",
99 "nlk_cb_mutex-XFRM",
100 "nlk_cb_mutex-SELINUX",
101 "nlk_cb_mutex-ISCSI",
102 "nlk_cb_mutex-AUDIT",
103 "nlk_cb_mutex-FIB_LOOKUP",
104 "nlk_cb_mutex-CONNECTOR",
105 "nlk_cb_mutex-NETFILTER",
106 "nlk_cb_mutex-IP6_FW",
107 "nlk_cb_mutex-DNRTMSG",
108 "nlk_cb_mutex-KOBJECT_UEVENT",
109 "nlk_cb_mutex-GENERIC",
110 "nlk_cb_mutex-17",
111 "nlk_cb_mutex-SCSITRANSPORT",
112 "nlk_cb_mutex-ECRYPTFS",
113 "nlk_cb_mutex-RDMA",
114 "nlk_cb_mutex-CRYPTO",
115 "nlk_cb_mutex-SMC",
116 "nlk_cb_mutex-23",
117 "nlk_cb_mutex-24",
118 "nlk_cb_mutex-25",
119 "nlk_cb_mutex-26",
120 "nlk_cb_mutex-27",
121 "nlk_cb_mutex-28",
122 "nlk_cb_mutex-29",
123 "nlk_cb_mutex-30",
124 "nlk_cb_mutex-31",
125 "nlk_cb_mutex-MAX_LINKS"
126 };
127
128 static int netlink_dump(struct sock *sk);
129
130 /* nl_table locking explained:
131 * Lookup and traversal are protected with an RCU read-side lock. Insertion
132 * and removal are protected with per bucket lock while using RCU list
133 * modification primitives and may run in parallel to RCU protected lookups.
134 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
135 * been acquired * either during or after the socket has been removed from
136 * the list and after an RCU grace period.
137 */
138 DEFINE_RWLOCK(nl_table_lock);
139 EXPORT_SYMBOL_GPL(nl_table_lock);
140 static atomic_t nl_table_users = ATOMIC_INIT(0);
141
142 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
143
144 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
145
146
147 static const struct rhashtable_params netlink_rhashtable_params;
148
149 static inline u32 netlink_group_mask(u32 group)
150 {
151 return group ? 1 << (group - 1) : 0;
152 }
153
154 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
155 gfp_t gfp_mask)
156 {
157 unsigned int len = skb_end_offset(skb);
158 struct sk_buff *new;
159
160 new = alloc_skb(len, gfp_mask);
161 if (new == NULL)
162 return NULL;
163
164 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
165 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
166 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
167
168 skb_put_data(new, skb->data, len);
169 return new;
170 }
171
172 static unsigned int netlink_tap_net_id;
173
174 struct netlink_tap_net {
175 struct list_head netlink_tap_all;
176 struct mutex netlink_tap_lock;
177 };
178
179 int netlink_add_tap(struct netlink_tap *nt)
180 {
181 struct net *net = dev_net(nt->dev);
182 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
183
184 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
185 return -EINVAL;
186
187 mutex_lock(&nn->netlink_tap_lock);
188 list_add_rcu(&nt->list, &nn->netlink_tap_all);
189 mutex_unlock(&nn->netlink_tap_lock);
190
191 __module_get(nt->module);
192
193 return 0;
194 }
195 EXPORT_SYMBOL_GPL(netlink_add_tap);
196
197 static int __netlink_remove_tap(struct netlink_tap *nt)
198 {
199 struct net *net = dev_net(nt->dev);
200 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
201 bool found = false;
202 struct netlink_tap *tmp;
203
204 mutex_lock(&nn->netlink_tap_lock);
205
206 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
207 if (nt == tmp) {
208 list_del_rcu(&nt->list);
209 found = true;
210 goto out;
211 }
212 }
213
214 pr_warn("__netlink_remove_tap: %p not found\n", nt);
215 out:
216 mutex_unlock(&nn->netlink_tap_lock);
217
218 if (found)
219 module_put(nt->module);
220
221 return found ? 0 : -ENODEV;
222 }
223
224 int netlink_remove_tap(struct netlink_tap *nt)
225 {
226 int ret;
227
228 ret = __netlink_remove_tap(nt);
229 synchronize_net();
230
231 return ret;
232 }
233 EXPORT_SYMBOL_GPL(netlink_remove_tap);
234
235 static __net_init int netlink_tap_init_net(struct net *net)
236 {
237 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
238
239 INIT_LIST_HEAD(&nn->netlink_tap_all);
240 mutex_init(&nn->netlink_tap_lock);
241 return 0;
242 }
243
244 static struct pernet_operations netlink_tap_net_ops = {
245 .init = netlink_tap_init_net,
246 .id = &netlink_tap_net_id,
247 .size = sizeof(struct netlink_tap_net),
248 };
249
250 static bool netlink_filter_tap(const struct sk_buff *skb)
251 {
252 struct sock *sk = skb->sk;
253
254 /* We take the more conservative approach and
255 * whitelist socket protocols that may pass.
256 */
257 switch (sk->sk_protocol) {
258 case NETLINK_ROUTE:
259 case NETLINK_USERSOCK:
260 case NETLINK_SOCK_DIAG:
261 case NETLINK_NFLOG:
262 case NETLINK_XFRM:
263 case NETLINK_FIB_LOOKUP:
264 case NETLINK_NETFILTER:
265 case NETLINK_GENERIC:
266 return true;
267 }
268
269 return false;
270 }
271
272 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
273 struct net_device *dev)
274 {
275 struct sk_buff *nskb;
276 struct sock *sk = skb->sk;
277 int ret = -ENOMEM;
278
279 if (!net_eq(dev_net(dev), sock_net(sk)))
280 return 0;
281
282 dev_hold(dev);
283
284 if (is_vmalloc_addr(skb->head))
285 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
286 else
287 nskb = skb_clone(skb, GFP_ATOMIC);
288 if (nskb) {
289 nskb->dev = dev;
290 nskb->protocol = htons((u16) sk->sk_protocol);
291 nskb->pkt_type = netlink_is_kernel(sk) ?
292 PACKET_KERNEL : PACKET_USER;
293 skb_reset_network_header(nskb);
294 ret = dev_queue_xmit(nskb);
295 if (unlikely(ret > 0))
296 ret = net_xmit_errno(ret);
297 }
298
299 dev_put(dev);
300 return ret;
301 }
302
303 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
304 {
305 int ret;
306 struct netlink_tap *tmp;
307
308 if (!netlink_filter_tap(skb))
309 return;
310
311 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
312 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
313 if (unlikely(ret))
314 break;
315 }
316 }
317
318 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
319 {
320 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
321
322 rcu_read_lock();
323
324 if (unlikely(!list_empty(&nn->netlink_tap_all)))
325 __netlink_deliver_tap(skb, nn);
326
327 rcu_read_unlock();
328 }
329
330 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
331 struct sk_buff *skb)
332 {
333 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
334 netlink_deliver_tap(sock_net(dst), skb);
335 }
336
337 static void netlink_overrun(struct sock *sk)
338 {
339 struct netlink_sock *nlk = nlk_sk(sk);
340
341 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
342 if (!test_and_set_bit(NETLINK_S_CONGESTED,
343 &nlk_sk(sk)->state)) {
344 sk->sk_err = ENOBUFS;
345 sk->sk_error_report(sk);
346 }
347 }
348 atomic_inc(&sk->sk_drops);
349 }
350
351 static void netlink_rcv_wake(struct sock *sk)
352 {
353 struct netlink_sock *nlk = nlk_sk(sk);
354
355 if (skb_queue_empty(&sk->sk_receive_queue))
356 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
357 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
358 wake_up_interruptible(&nlk->wait);
359 }
360
361 static void netlink_skb_destructor(struct sk_buff *skb)
362 {
363 if (is_vmalloc_addr(skb->head)) {
364 if (!skb->cloned ||
365 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
366 vfree(skb->head);
367
368 skb->head = NULL;
369 }
370 if (skb->sk != NULL)
371 sock_rfree(skb);
372 }
373
374 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
375 {
376 WARN_ON(skb->sk != NULL);
377 skb->sk = sk;
378 skb->destructor = netlink_skb_destructor;
379 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
380 sk_mem_charge(sk, skb->truesize);
381 }
382
383 static void netlink_sock_destruct(struct sock *sk)
384 {
385 struct netlink_sock *nlk = nlk_sk(sk);
386
387 if (nlk->cb_running) {
388 if (nlk->cb.done)
389 nlk->cb.done(&nlk->cb);
390 module_put(nlk->cb.module);
391 kfree_skb(nlk->cb.skb);
392 }
393
394 skb_queue_purge(&sk->sk_receive_queue);
395
396 if (!sock_flag(sk, SOCK_DEAD)) {
397 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
398 return;
399 }
400
401 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
402 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
403 WARN_ON(nlk_sk(sk)->groups);
404 }
405
406 static void netlink_sock_destruct_work(struct work_struct *work)
407 {
408 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
409 work);
410
411 sk_free(&nlk->sk);
412 }
413
414 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
415 * SMP. Look, when several writers sleep and reader wakes them up, all but one
416 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
417 * this, _but_ remember, it adds useless work on UP machines.
418 */
419
420 void netlink_table_grab(void)
421 __acquires(nl_table_lock)
422 {
423 might_sleep();
424
425 write_lock_irq(&nl_table_lock);
426
427 if (atomic_read(&nl_table_users)) {
428 DECLARE_WAITQUEUE(wait, current);
429
430 add_wait_queue_exclusive(&nl_table_wait, &wait);
431 for (;;) {
432 set_current_state(TASK_UNINTERRUPTIBLE);
433 if (atomic_read(&nl_table_users) == 0)
434 break;
435 write_unlock_irq(&nl_table_lock);
436 schedule();
437 write_lock_irq(&nl_table_lock);
438 }
439
440 __set_current_state(TASK_RUNNING);
441 remove_wait_queue(&nl_table_wait, &wait);
442 }
443 }
444
445 void netlink_table_ungrab(void)
446 __releases(nl_table_lock)
447 {
448 write_unlock_irq(&nl_table_lock);
449 wake_up(&nl_table_wait);
450 }
451
452 static inline void
453 netlink_lock_table(void)
454 {
455 /* read_lock() synchronizes us to netlink_table_grab */
456
457 read_lock(&nl_table_lock);
458 atomic_inc(&nl_table_users);
459 read_unlock(&nl_table_lock);
460 }
461
462 static inline void
463 netlink_unlock_table(void)
464 {
465 if (atomic_dec_and_test(&nl_table_users))
466 wake_up(&nl_table_wait);
467 }
468
469 struct netlink_compare_arg
470 {
471 possible_net_t pnet;
472 u32 portid;
473 };
474
475 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
476 #define netlink_compare_arg_len \
477 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
478
479 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
480 const void *ptr)
481 {
482 const struct netlink_compare_arg *x = arg->key;
483 const struct netlink_sock *nlk = ptr;
484
485 return nlk->portid != x->portid ||
486 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
487 }
488
489 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
490 struct net *net, u32 portid)
491 {
492 memset(arg, 0, sizeof(*arg));
493 write_pnet(&arg->pnet, net);
494 arg->portid = portid;
495 }
496
497 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
498 struct net *net)
499 {
500 struct netlink_compare_arg arg;
501
502 netlink_compare_arg_init(&arg, net, portid);
503 return rhashtable_lookup_fast(&table->hash, &arg,
504 netlink_rhashtable_params);
505 }
506
507 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
508 {
509 struct netlink_compare_arg arg;
510
511 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
512 return rhashtable_lookup_insert_key(&table->hash, &arg,
513 &nlk_sk(sk)->node,
514 netlink_rhashtable_params);
515 }
516
517 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
518 {
519 struct netlink_table *table = &nl_table[protocol];
520 struct sock *sk;
521
522 rcu_read_lock();
523 sk = __netlink_lookup(table, portid, net);
524 if (sk)
525 sock_hold(sk);
526 rcu_read_unlock();
527
528 return sk;
529 }
530
531 static const struct proto_ops netlink_ops;
532
533 static void
534 netlink_update_listeners(struct sock *sk)
535 {
536 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
537 unsigned long mask;
538 unsigned int i;
539 struct listeners *listeners;
540
541 listeners = nl_deref_protected(tbl->listeners);
542 if (!listeners)
543 return;
544
545 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
546 mask = 0;
547 sk_for_each_bound(sk, &tbl->mc_list) {
548 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
549 mask |= nlk_sk(sk)->groups[i];
550 }
551 listeners->masks[i] = mask;
552 }
553 /* this function is only called with the netlink table "grabbed", which
554 * makes sure updates are visible before bind or setsockopt return. */
555 }
556
557 static int netlink_insert(struct sock *sk, u32 portid)
558 {
559 struct netlink_table *table = &nl_table[sk->sk_protocol];
560 int err;
561
562 lock_sock(sk);
563
564 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
565 if (nlk_sk(sk)->bound)
566 goto err;
567
568 nlk_sk(sk)->portid = portid;
569 sock_hold(sk);
570
571 err = __netlink_insert(table, sk);
572 if (err) {
573 /* In case the hashtable backend returns with -EBUSY
574 * from here, it must not escape to the caller.
575 */
576 if (unlikely(err == -EBUSY))
577 err = -EOVERFLOW;
578 if (err == -EEXIST)
579 err = -EADDRINUSE;
580 sock_put(sk);
581 goto err;
582 }
583
584 /* We need to ensure that the socket is hashed and visible. */
585 smp_wmb();
586 nlk_sk(sk)->bound = portid;
587
588 err:
589 release_sock(sk);
590 return err;
591 }
592
593 static void netlink_remove(struct sock *sk)
594 {
595 struct netlink_table *table;
596
597 table = &nl_table[sk->sk_protocol];
598 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
599 netlink_rhashtable_params)) {
600 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
601 __sock_put(sk);
602 }
603
604 netlink_table_grab();
605 if (nlk_sk(sk)->subscriptions) {
606 __sk_del_bind_node(sk);
607 netlink_update_listeners(sk);
608 }
609 if (sk->sk_protocol == NETLINK_GENERIC)
610 atomic_inc(&genl_sk_destructing_cnt);
611 netlink_table_ungrab();
612 }
613
614 static struct proto netlink_proto = {
615 .name = "NETLINK",
616 .owner = THIS_MODULE,
617 .obj_size = sizeof(struct netlink_sock),
618 };
619
620 static int __netlink_create(struct net *net, struct socket *sock,
621 struct mutex *cb_mutex, int protocol,
622 int kern)
623 {
624 struct sock *sk;
625 struct netlink_sock *nlk;
626
627 sock->ops = &netlink_ops;
628
629 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
630 if (!sk)
631 return -ENOMEM;
632
633 sock_init_data(sock, sk);
634
635 nlk = nlk_sk(sk);
636 if (cb_mutex) {
637 nlk->cb_mutex = cb_mutex;
638 } else {
639 nlk->cb_mutex = &nlk->cb_def_mutex;
640 mutex_init(nlk->cb_mutex);
641 lockdep_set_class_and_name(nlk->cb_mutex,
642 nlk_cb_mutex_keys + protocol,
643 nlk_cb_mutex_key_strings[protocol]);
644 }
645 init_waitqueue_head(&nlk->wait);
646
647 sk->sk_destruct = netlink_sock_destruct;
648 sk->sk_protocol = protocol;
649 return 0;
650 }
651
652 static int netlink_create(struct net *net, struct socket *sock, int protocol,
653 int kern)
654 {
655 struct module *module = NULL;
656 struct mutex *cb_mutex;
657 struct netlink_sock *nlk;
658 int (*bind)(struct net *net, int group);
659 void (*unbind)(struct net *net, int group);
660 int err = 0;
661
662 sock->state = SS_UNCONNECTED;
663
664 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
665 return -ESOCKTNOSUPPORT;
666
667 if (protocol < 0 || protocol >= MAX_LINKS)
668 return -EPROTONOSUPPORT;
669 protocol = array_index_nospec(protocol, MAX_LINKS);
670
671 netlink_lock_table();
672 #ifdef CONFIG_MODULES
673 if (!nl_table[protocol].registered) {
674 netlink_unlock_table();
675 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
676 netlink_lock_table();
677 }
678 #endif
679 if (nl_table[protocol].registered &&
680 try_module_get(nl_table[protocol].module))
681 module = nl_table[protocol].module;
682 else
683 err = -EPROTONOSUPPORT;
684 cb_mutex = nl_table[protocol].cb_mutex;
685 bind = nl_table[protocol].bind;
686 unbind = nl_table[protocol].unbind;
687 netlink_unlock_table();
688
689 if (err < 0)
690 goto out;
691
692 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
693 if (err < 0)
694 goto out_module;
695
696 local_bh_disable();
697 sock_prot_inuse_add(net, &netlink_proto, 1);
698 local_bh_enable();
699
700 nlk = nlk_sk(sock->sk);
701 nlk->module = module;
702 nlk->netlink_bind = bind;
703 nlk->netlink_unbind = unbind;
704 out:
705 return err;
706
707 out_module:
708 module_put(module);
709 goto out;
710 }
711
712 static void deferred_put_nlk_sk(struct rcu_head *head)
713 {
714 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
715 struct sock *sk = &nlk->sk;
716
717 kfree(nlk->groups);
718 nlk->groups = NULL;
719
720 if (!refcount_dec_and_test(&sk->sk_refcnt))
721 return;
722
723 if (nlk->cb_running && nlk->cb.done) {
724 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
725 schedule_work(&nlk->work);
726 return;
727 }
728
729 sk_free(sk);
730 }
731
732 static int netlink_release(struct socket *sock)
733 {
734 struct sock *sk = sock->sk;
735 struct netlink_sock *nlk;
736
737 if (!sk)
738 return 0;
739
740 netlink_remove(sk);
741 sock_orphan(sk);
742 nlk = nlk_sk(sk);
743
744 /*
745 * OK. Socket is unlinked, any packets that arrive now
746 * will be purged.
747 */
748
749 /* must not acquire netlink_table_lock in any way again before unbind
750 * and notifying genetlink is done as otherwise it might deadlock
751 */
752 if (nlk->netlink_unbind) {
753 int i;
754
755 for (i = 0; i < nlk->ngroups; i++)
756 if (test_bit(i, nlk->groups))
757 nlk->netlink_unbind(sock_net(sk), i + 1);
758 }
759 if (sk->sk_protocol == NETLINK_GENERIC &&
760 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
761 wake_up(&genl_sk_destructing_waitq);
762
763 sock->sk = NULL;
764 wake_up_interruptible_all(&nlk->wait);
765
766 skb_queue_purge(&sk->sk_write_queue);
767
768 if (nlk->portid && nlk->bound) {
769 struct netlink_notify n = {
770 .net = sock_net(sk),
771 .protocol = sk->sk_protocol,
772 .portid = nlk->portid,
773 };
774 blocking_notifier_call_chain(&netlink_chain,
775 NETLINK_URELEASE, &n);
776 }
777
778 module_put(nlk->module);
779
780 if (netlink_is_kernel(sk)) {
781 netlink_table_grab();
782 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
783 if (--nl_table[sk->sk_protocol].registered == 0) {
784 struct listeners *old;
785
786 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
787 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
788 kfree_rcu(old, rcu);
789 nl_table[sk->sk_protocol].module = NULL;
790 nl_table[sk->sk_protocol].bind = NULL;
791 nl_table[sk->sk_protocol].unbind = NULL;
792 nl_table[sk->sk_protocol].flags = 0;
793 nl_table[sk->sk_protocol].registered = 0;
794 }
795 netlink_table_ungrab();
796 }
797
798 local_bh_disable();
799 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
800 local_bh_enable();
801 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
802 return 0;
803 }
804
805 static int netlink_autobind(struct socket *sock)
806 {
807 struct sock *sk = sock->sk;
808 struct net *net = sock_net(sk);
809 struct netlink_table *table = &nl_table[sk->sk_protocol];
810 s32 portid = task_tgid_vnr(current);
811 int err;
812 s32 rover = -4096;
813 bool ok;
814
815 retry:
816 cond_resched();
817 rcu_read_lock();
818 ok = !__netlink_lookup(table, portid, net);
819 rcu_read_unlock();
820 if (!ok) {
821 /* Bind collision, search negative portid values. */
822 if (rover == -4096)
823 /* rover will be in range [S32_MIN, -4097] */
824 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
825 else if (rover >= -4096)
826 rover = -4097;
827 portid = rover--;
828 goto retry;
829 }
830
831 err = netlink_insert(sk, portid);
832 if (err == -EADDRINUSE)
833 goto retry;
834
835 /* If 2 threads race to autobind, that is fine. */
836 if (err == -EBUSY)
837 err = 0;
838
839 return err;
840 }
841
842 /**
843 * __netlink_ns_capable - General netlink message capability test
844 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
845 * @user_ns: The user namespace of the capability to use
846 * @cap: The capability to use
847 *
848 * Test to see if the opener of the socket we received the message
849 * from had when the netlink socket was created and the sender of the
850 * message has has the capability @cap in the user namespace @user_ns.
851 */
852 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
853 struct user_namespace *user_ns, int cap)
854 {
855 return ((nsp->flags & NETLINK_SKB_DST) ||
856 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
857 ns_capable(user_ns, cap);
858 }
859 EXPORT_SYMBOL(__netlink_ns_capable);
860
861 /**
862 * netlink_ns_capable - General netlink message capability test
863 * @skb: socket buffer holding a netlink command from userspace
864 * @user_ns: The user namespace of the capability to use
865 * @cap: The capability to use
866 *
867 * Test to see if the opener of the socket we received the message
868 * from had when the netlink socket was created and the sender of the
869 * message has has the capability @cap in the user namespace @user_ns.
870 */
871 bool netlink_ns_capable(const struct sk_buff *skb,
872 struct user_namespace *user_ns, int cap)
873 {
874 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
875 }
876 EXPORT_SYMBOL(netlink_ns_capable);
877
878 /**
879 * netlink_capable - Netlink global message capability test
880 * @skb: socket buffer holding a netlink command from userspace
881 * @cap: The capability to use
882 *
883 * Test to see if the opener of the socket we received the message
884 * from had when the netlink socket was created and the sender of the
885 * message has has the capability @cap in all user namespaces.
886 */
887 bool netlink_capable(const struct sk_buff *skb, int cap)
888 {
889 return netlink_ns_capable(skb, &init_user_ns, cap);
890 }
891 EXPORT_SYMBOL(netlink_capable);
892
893 /**
894 * netlink_net_capable - Netlink network namespace message capability test
895 * @skb: socket buffer holding a netlink command from userspace
896 * @cap: The capability to use
897 *
898 * Test to see if the opener of the socket we received the message
899 * from had when the netlink socket was created and the sender of the
900 * message has has the capability @cap over the network namespace of
901 * the socket we received the message from.
902 */
903 bool netlink_net_capable(const struct sk_buff *skb, int cap)
904 {
905 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
906 }
907 EXPORT_SYMBOL(netlink_net_capable);
908
909 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
910 {
911 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
912 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
913 }
914
915 static void
916 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
917 {
918 struct netlink_sock *nlk = nlk_sk(sk);
919
920 if (nlk->subscriptions && !subscriptions)
921 __sk_del_bind_node(sk);
922 else if (!nlk->subscriptions && subscriptions)
923 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
924 nlk->subscriptions = subscriptions;
925 }
926
927 static int netlink_realloc_groups(struct sock *sk)
928 {
929 struct netlink_sock *nlk = nlk_sk(sk);
930 unsigned int groups;
931 unsigned long *new_groups;
932 int err = 0;
933
934 netlink_table_grab();
935
936 groups = nl_table[sk->sk_protocol].groups;
937 if (!nl_table[sk->sk_protocol].registered) {
938 err = -ENOENT;
939 goto out_unlock;
940 }
941
942 if (nlk->ngroups >= groups)
943 goto out_unlock;
944
945 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
946 if (new_groups == NULL) {
947 err = -ENOMEM;
948 goto out_unlock;
949 }
950 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
951 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
952
953 nlk->groups = new_groups;
954 nlk->ngroups = groups;
955 out_unlock:
956 netlink_table_ungrab();
957 return err;
958 }
959
960 static void netlink_undo_bind(int group, long unsigned int groups,
961 struct sock *sk)
962 {
963 struct netlink_sock *nlk = nlk_sk(sk);
964 int undo;
965
966 if (!nlk->netlink_unbind)
967 return;
968
969 for (undo = 0; undo < group; undo++)
970 if (test_bit(undo, &groups))
971 nlk->netlink_unbind(sock_net(sk), undo + 1);
972 }
973
974 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
975 int addr_len)
976 {
977 struct sock *sk = sock->sk;
978 struct net *net = sock_net(sk);
979 struct netlink_sock *nlk = nlk_sk(sk);
980 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
981 int err = 0;
982 unsigned long groups;
983 bool bound;
984
985 if (addr_len < sizeof(struct sockaddr_nl))
986 return -EINVAL;
987
988 if (nladdr->nl_family != AF_NETLINK)
989 return -EINVAL;
990 groups = nladdr->nl_groups;
991
992 /* Only superuser is allowed to listen multicasts */
993 if (groups) {
994 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
995 return -EPERM;
996 err = netlink_realloc_groups(sk);
997 if (err)
998 return err;
999 }
1000
1001 if (nlk->ngroups < BITS_PER_LONG)
1002 groups &= (1UL << nlk->ngroups) - 1;
1003
1004 bound = nlk->bound;
1005 if (bound) {
1006 /* Ensure nlk->portid is up-to-date. */
1007 smp_rmb();
1008
1009 if (nladdr->nl_pid != nlk->portid)
1010 return -EINVAL;
1011 }
1012
1013 netlink_lock_table();
1014 if (nlk->netlink_bind && groups) {
1015 int group;
1016
1017 /* nl_groups is a u32, so cap the maximum groups we can bind */
1018 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1019 if (!test_bit(group, &groups))
1020 continue;
1021 err = nlk->netlink_bind(net, group + 1);
1022 if (!err)
1023 continue;
1024 netlink_undo_bind(group, groups, sk);
1025 goto unlock;
1026 }
1027 }
1028
1029 /* No need for barriers here as we return to user-space without
1030 * using any of the bound attributes.
1031 */
1032 if (!bound) {
1033 err = nladdr->nl_pid ?
1034 netlink_insert(sk, nladdr->nl_pid) :
1035 netlink_autobind(sock);
1036 if (err) {
1037 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1038 goto unlock;
1039 }
1040 }
1041
1042 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1043 goto unlock;
1044 netlink_unlock_table();
1045
1046 netlink_table_grab();
1047 netlink_update_subscriptions(sk, nlk->subscriptions +
1048 hweight32(groups) -
1049 hweight32(nlk->groups[0]));
1050 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1051 netlink_update_listeners(sk);
1052 netlink_table_ungrab();
1053
1054 return 0;
1055
1056 unlock:
1057 netlink_unlock_table();
1058 return err;
1059 }
1060
1061 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1062 int alen, int flags)
1063 {
1064 int err = 0;
1065 struct sock *sk = sock->sk;
1066 struct netlink_sock *nlk = nlk_sk(sk);
1067 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1068
1069 if (alen < sizeof(addr->sa_family))
1070 return -EINVAL;
1071
1072 if (addr->sa_family == AF_UNSPEC) {
1073 sk->sk_state = NETLINK_UNCONNECTED;
1074 nlk->dst_portid = 0;
1075 nlk->dst_group = 0;
1076 return 0;
1077 }
1078 if (addr->sa_family != AF_NETLINK)
1079 return -EINVAL;
1080
1081 if (alen < sizeof(struct sockaddr_nl))
1082 return -EINVAL;
1083
1084 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1085 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1086 return -EPERM;
1087
1088 /* No need for barriers here as we return to user-space without
1089 * using any of the bound attributes.
1090 */
1091 if (!nlk->bound)
1092 err = netlink_autobind(sock);
1093
1094 if (err == 0) {
1095 sk->sk_state = NETLINK_CONNECTED;
1096 nlk->dst_portid = nladdr->nl_pid;
1097 nlk->dst_group = ffs(nladdr->nl_groups);
1098 }
1099
1100 return err;
1101 }
1102
1103 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1104 int peer)
1105 {
1106 struct sock *sk = sock->sk;
1107 struct netlink_sock *nlk = nlk_sk(sk);
1108 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1109
1110 nladdr->nl_family = AF_NETLINK;
1111 nladdr->nl_pad = 0;
1112
1113 if (peer) {
1114 nladdr->nl_pid = nlk->dst_portid;
1115 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1116 } else {
1117 nladdr->nl_pid = nlk->portid;
1118 netlink_lock_table();
1119 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1120 netlink_unlock_table();
1121 }
1122 return sizeof(*nladdr);
1123 }
1124
1125 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1126 unsigned long arg)
1127 {
1128 /* try to hand this ioctl down to the NIC drivers.
1129 */
1130 return -ENOIOCTLCMD;
1131 }
1132
1133 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1134 {
1135 struct sock *sock;
1136 struct netlink_sock *nlk;
1137
1138 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1139 if (!sock)
1140 return ERR_PTR(-ECONNREFUSED);
1141
1142 /* Don't bother queuing skb if kernel socket has no input function */
1143 nlk = nlk_sk(sock);
1144 if (sock->sk_state == NETLINK_CONNECTED &&
1145 nlk->dst_portid != nlk_sk(ssk)->portid) {
1146 sock_put(sock);
1147 return ERR_PTR(-ECONNREFUSED);
1148 }
1149 return sock;
1150 }
1151
1152 struct sock *netlink_getsockbyfilp(struct file *filp)
1153 {
1154 struct inode *inode = file_inode(filp);
1155 struct sock *sock;
1156
1157 if (!S_ISSOCK(inode->i_mode))
1158 return ERR_PTR(-ENOTSOCK);
1159
1160 sock = SOCKET_I(inode)->sk;
1161 if (sock->sk_family != AF_NETLINK)
1162 return ERR_PTR(-EINVAL);
1163
1164 sock_hold(sock);
1165 return sock;
1166 }
1167
1168 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1169 int broadcast)
1170 {
1171 struct sk_buff *skb;
1172 void *data;
1173
1174 if (size <= NLMSG_GOODSIZE || broadcast)
1175 return alloc_skb(size, GFP_KERNEL);
1176
1177 size = SKB_DATA_ALIGN(size) +
1178 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1179
1180 data = vmalloc(size);
1181 if (data == NULL)
1182 return NULL;
1183
1184 skb = __build_skb(data, size);
1185 if (skb == NULL)
1186 vfree(data);
1187 else
1188 skb->destructor = netlink_skb_destructor;
1189
1190 return skb;
1191 }
1192
1193 /*
1194 * Attach a skb to a netlink socket.
1195 * The caller must hold a reference to the destination socket. On error, the
1196 * reference is dropped. The skb is not send to the destination, just all
1197 * all error checks are performed and memory in the queue is reserved.
1198 * Return values:
1199 * < 0: error. skb freed, reference to sock dropped.
1200 * 0: continue
1201 * 1: repeat lookup - reference dropped while waiting for socket memory.
1202 */
1203 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1204 long *timeo, struct sock *ssk)
1205 {
1206 struct netlink_sock *nlk;
1207
1208 nlk = nlk_sk(sk);
1209
1210 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1211 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1212 DECLARE_WAITQUEUE(wait, current);
1213 if (!*timeo) {
1214 if (!ssk || netlink_is_kernel(ssk))
1215 netlink_overrun(sk);
1216 sock_put(sk);
1217 kfree_skb(skb);
1218 return -EAGAIN;
1219 }
1220
1221 __set_current_state(TASK_INTERRUPTIBLE);
1222 add_wait_queue(&nlk->wait, &wait);
1223
1224 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1225 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1226 !sock_flag(sk, SOCK_DEAD))
1227 *timeo = schedule_timeout(*timeo);
1228
1229 __set_current_state(TASK_RUNNING);
1230 remove_wait_queue(&nlk->wait, &wait);
1231 sock_put(sk);
1232
1233 if (signal_pending(current)) {
1234 kfree_skb(skb);
1235 return sock_intr_errno(*timeo);
1236 }
1237 return 1;
1238 }
1239 netlink_skb_set_owner_r(skb, sk);
1240 return 0;
1241 }
1242
1243 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1244 {
1245 int len = skb->len;
1246
1247 netlink_deliver_tap(sock_net(sk), skb);
1248
1249 skb_queue_tail(&sk->sk_receive_queue, skb);
1250 sk->sk_data_ready(sk);
1251 return len;
1252 }
1253
1254 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1255 {
1256 int len = __netlink_sendskb(sk, skb);
1257
1258 sock_put(sk);
1259 return len;
1260 }
1261
1262 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1263 {
1264 kfree_skb(skb);
1265 sock_put(sk);
1266 }
1267
1268 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1269 {
1270 int delta;
1271
1272 WARN_ON(skb->sk != NULL);
1273 delta = skb->end - skb->tail;
1274 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1275 return skb;
1276
1277 if (skb_shared(skb)) {
1278 struct sk_buff *nskb = skb_clone(skb, allocation);
1279 if (!nskb)
1280 return skb;
1281 consume_skb(skb);
1282 skb = nskb;
1283 }
1284
1285 pskb_expand_head(skb, 0, -delta,
1286 (allocation & ~__GFP_DIRECT_RECLAIM) |
1287 __GFP_NOWARN | __GFP_NORETRY);
1288 return skb;
1289 }
1290
1291 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1292 struct sock *ssk)
1293 {
1294 int ret;
1295 struct netlink_sock *nlk = nlk_sk(sk);
1296
1297 ret = -ECONNREFUSED;
1298 if (nlk->netlink_rcv != NULL) {
1299 ret = skb->len;
1300 netlink_skb_set_owner_r(skb, sk);
1301 NETLINK_CB(skb).sk = ssk;
1302 netlink_deliver_tap_kernel(sk, ssk, skb);
1303 nlk->netlink_rcv(skb);
1304 consume_skb(skb);
1305 } else {
1306 kfree_skb(skb);
1307 }
1308 sock_put(sk);
1309 return ret;
1310 }
1311
1312 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1313 u32 portid, int nonblock)
1314 {
1315 struct sock *sk;
1316 int err;
1317 long timeo;
1318
1319 skb = netlink_trim(skb, gfp_any());
1320
1321 timeo = sock_sndtimeo(ssk, nonblock);
1322 retry:
1323 sk = netlink_getsockbyportid(ssk, portid);
1324 if (IS_ERR(sk)) {
1325 kfree_skb(skb);
1326 return PTR_ERR(sk);
1327 }
1328 if (netlink_is_kernel(sk))
1329 return netlink_unicast_kernel(sk, skb, ssk);
1330
1331 if (sk_filter(sk, skb)) {
1332 err = skb->len;
1333 kfree_skb(skb);
1334 sock_put(sk);
1335 return err;
1336 }
1337
1338 err = netlink_attachskb(sk, skb, &timeo, ssk);
1339 if (err == 1)
1340 goto retry;
1341 if (err)
1342 return err;
1343
1344 return netlink_sendskb(sk, skb);
1345 }
1346 EXPORT_SYMBOL(netlink_unicast);
1347
1348 int netlink_has_listeners(struct sock *sk, unsigned int group)
1349 {
1350 int res = 0;
1351 struct listeners *listeners;
1352
1353 BUG_ON(!netlink_is_kernel(sk));
1354
1355 rcu_read_lock();
1356 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1357
1358 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1359 res = test_bit(group - 1, listeners->masks);
1360
1361 rcu_read_unlock();
1362
1363 return res;
1364 }
1365 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1366
1367 bool netlink_strict_get_check(struct sk_buff *skb)
1368 {
1369 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1370
1371 return nlk->flags & NETLINK_F_STRICT_CHK;
1372 }
1373 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1374
1375 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1376 {
1377 struct netlink_sock *nlk = nlk_sk(sk);
1378
1379 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1380 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1381 netlink_skb_set_owner_r(skb, sk);
1382 __netlink_sendskb(sk, skb);
1383 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1384 }
1385 return -1;
1386 }
1387
1388 struct netlink_broadcast_data {
1389 struct sock *exclude_sk;
1390 struct net *net;
1391 u32 portid;
1392 u32 group;
1393 int failure;
1394 int delivery_failure;
1395 int congested;
1396 int delivered;
1397 gfp_t allocation;
1398 struct sk_buff *skb, *skb2;
1399 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1400 void *tx_data;
1401 };
1402
1403 static void do_one_broadcast(struct sock *sk,
1404 struct netlink_broadcast_data *p)
1405 {
1406 struct netlink_sock *nlk = nlk_sk(sk);
1407 int val;
1408
1409 if (p->exclude_sk == sk)
1410 return;
1411
1412 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1413 !test_bit(p->group - 1, nlk->groups))
1414 return;
1415
1416 if (!net_eq(sock_net(sk), p->net)) {
1417 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1418 return;
1419
1420 if (!peernet_has_id(sock_net(sk), p->net))
1421 return;
1422
1423 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1424 CAP_NET_BROADCAST))
1425 return;
1426 }
1427
1428 if (p->failure) {
1429 netlink_overrun(sk);
1430 return;
1431 }
1432
1433 sock_hold(sk);
1434 if (p->skb2 == NULL) {
1435 if (skb_shared(p->skb)) {
1436 p->skb2 = skb_clone(p->skb, p->allocation);
1437 } else {
1438 p->skb2 = skb_get(p->skb);
1439 /*
1440 * skb ownership may have been set when
1441 * delivered to a previous socket.
1442 */
1443 skb_orphan(p->skb2);
1444 }
1445 }
1446 if (p->skb2 == NULL) {
1447 netlink_overrun(sk);
1448 /* Clone failed. Notify ALL listeners. */
1449 p->failure = 1;
1450 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1451 p->delivery_failure = 1;
1452 goto out;
1453 }
1454 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1455 kfree_skb(p->skb2);
1456 p->skb2 = NULL;
1457 goto out;
1458 }
1459 if (sk_filter(sk, p->skb2)) {
1460 kfree_skb(p->skb2);
1461 p->skb2 = NULL;
1462 goto out;
1463 }
1464 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1465 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1466 NETLINK_CB(p->skb2).nsid_is_set = true;
1467 val = netlink_broadcast_deliver(sk, p->skb2);
1468 if (val < 0) {
1469 netlink_overrun(sk);
1470 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1471 p->delivery_failure = 1;
1472 } else {
1473 p->congested |= val;
1474 p->delivered = 1;
1475 p->skb2 = NULL;
1476 }
1477 out:
1478 sock_put(sk);
1479 }
1480
1481 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1482 u32 group, gfp_t allocation,
1483 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1484 void *filter_data)
1485 {
1486 struct net *net = sock_net(ssk);
1487 struct netlink_broadcast_data info;
1488 struct sock *sk;
1489
1490 skb = netlink_trim(skb, allocation);
1491
1492 info.exclude_sk = ssk;
1493 info.net = net;
1494 info.portid = portid;
1495 info.group = group;
1496 info.failure = 0;
1497 info.delivery_failure = 0;
1498 info.congested = 0;
1499 info.delivered = 0;
1500 info.allocation = allocation;
1501 info.skb = skb;
1502 info.skb2 = NULL;
1503 info.tx_filter = filter;
1504 info.tx_data = filter_data;
1505
1506 /* While we sleep in clone, do not allow to change socket list */
1507
1508 netlink_lock_table();
1509
1510 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1511 do_one_broadcast(sk, &info);
1512
1513 consume_skb(skb);
1514
1515 netlink_unlock_table();
1516
1517 if (info.delivery_failure) {
1518 kfree_skb(info.skb2);
1519 return -ENOBUFS;
1520 }
1521 consume_skb(info.skb2);
1522
1523 if (info.delivered) {
1524 if (info.congested && gfpflags_allow_blocking(allocation))
1525 yield();
1526 return 0;
1527 }
1528 return -ESRCH;
1529 }
1530 EXPORT_SYMBOL(netlink_broadcast_filtered);
1531
1532 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1533 u32 group, gfp_t allocation)
1534 {
1535 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1536 NULL, NULL);
1537 }
1538 EXPORT_SYMBOL(netlink_broadcast);
1539
1540 struct netlink_set_err_data {
1541 struct sock *exclude_sk;
1542 u32 portid;
1543 u32 group;
1544 int code;
1545 };
1546
1547 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1548 {
1549 struct netlink_sock *nlk = nlk_sk(sk);
1550 int ret = 0;
1551
1552 if (sk == p->exclude_sk)
1553 goto out;
1554
1555 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1556 goto out;
1557
1558 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1559 !test_bit(p->group - 1, nlk->groups))
1560 goto out;
1561
1562 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1563 ret = 1;
1564 goto out;
1565 }
1566
1567 sk->sk_err = p->code;
1568 sk->sk_error_report(sk);
1569 out:
1570 return ret;
1571 }
1572
1573 /**
1574 * netlink_set_err - report error to broadcast listeners
1575 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1576 * @portid: the PORTID of a process that we want to skip (if any)
1577 * @group: the broadcast group that will notice the error
1578 * @code: error code, must be negative (as usual in kernelspace)
1579 *
1580 * This function returns the number of broadcast listeners that have set the
1581 * NETLINK_NO_ENOBUFS socket option.
1582 */
1583 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1584 {
1585 struct netlink_set_err_data info;
1586 struct sock *sk;
1587 int ret = 0;
1588
1589 info.exclude_sk = ssk;
1590 info.portid = portid;
1591 info.group = group;
1592 /* sk->sk_err wants a positive error value */
1593 info.code = -code;
1594
1595 read_lock(&nl_table_lock);
1596
1597 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1598 ret += do_one_set_err(sk, &info);
1599
1600 read_unlock(&nl_table_lock);
1601 return ret;
1602 }
1603 EXPORT_SYMBOL(netlink_set_err);
1604
1605 /* must be called with netlink table grabbed */
1606 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1607 unsigned int group,
1608 int is_new)
1609 {
1610 int old, new = !!is_new, subscriptions;
1611
1612 old = test_bit(group - 1, nlk->groups);
1613 subscriptions = nlk->subscriptions - old + new;
1614 if (new)
1615 __set_bit(group - 1, nlk->groups);
1616 else
1617 __clear_bit(group - 1, nlk->groups);
1618 netlink_update_subscriptions(&nlk->sk, subscriptions);
1619 netlink_update_listeners(&nlk->sk);
1620 }
1621
1622 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1623 char __user *optval, unsigned int optlen)
1624 {
1625 struct sock *sk = sock->sk;
1626 struct netlink_sock *nlk = nlk_sk(sk);
1627 unsigned int val = 0;
1628 int err;
1629
1630 if (level != SOL_NETLINK)
1631 return -ENOPROTOOPT;
1632
1633 if (optlen >= sizeof(int) &&
1634 get_user(val, (unsigned int __user *)optval))
1635 return -EFAULT;
1636
1637 switch (optname) {
1638 case NETLINK_PKTINFO:
1639 if (val)
1640 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1641 else
1642 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1643 err = 0;
1644 break;
1645 case NETLINK_ADD_MEMBERSHIP:
1646 case NETLINK_DROP_MEMBERSHIP: {
1647 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1648 return -EPERM;
1649 err = netlink_realloc_groups(sk);
1650 if (err)
1651 return err;
1652 if (!val || val - 1 >= nlk->ngroups)
1653 return -EINVAL;
1654 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1655 err = nlk->netlink_bind(sock_net(sk), val);
1656 if (err)
1657 return err;
1658 }
1659 netlink_table_grab();
1660 netlink_update_socket_mc(nlk, val,
1661 optname == NETLINK_ADD_MEMBERSHIP);
1662 netlink_table_ungrab();
1663 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1664 nlk->netlink_unbind(sock_net(sk), val);
1665
1666 err = 0;
1667 break;
1668 }
1669 case NETLINK_BROADCAST_ERROR:
1670 if (val)
1671 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1672 else
1673 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1674 err = 0;
1675 break;
1676 case NETLINK_NO_ENOBUFS:
1677 if (val) {
1678 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1679 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1680 wake_up_interruptible(&nlk->wait);
1681 } else {
1682 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1683 }
1684 err = 0;
1685 break;
1686 case NETLINK_LISTEN_ALL_NSID:
1687 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1688 return -EPERM;
1689
1690 if (val)
1691 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1692 else
1693 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1694 err = 0;
1695 break;
1696 case NETLINK_CAP_ACK:
1697 if (val)
1698 nlk->flags |= NETLINK_F_CAP_ACK;
1699 else
1700 nlk->flags &= ~NETLINK_F_CAP_ACK;
1701 err = 0;
1702 break;
1703 case NETLINK_EXT_ACK:
1704 if (val)
1705 nlk->flags |= NETLINK_F_EXT_ACK;
1706 else
1707 nlk->flags &= ~NETLINK_F_EXT_ACK;
1708 err = 0;
1709 break;
1710 case NETLINK_GET_STRICT_CHK:
1711 if (val)
1712 nlk->flags |= NETLINK_F_STRICT_CHK;
1713 else
1714 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1715 err = 0;
1716 break;
1717 default:
1718 err = -ENOPROTOOPT;
1719 }
1720 return err;
1721 }
1722
1723 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1724 char __user *optval, int __user *optlen)
1725 {
1726 struct sock *sk = sock->sk;
1727 struct netlink_sock *nlk = nlk_sk(sk);
1728 int len, val, err;
1729
1730 if (level != SOL_NETLINK)
1731 return -ENOPROTOOPT;
1732
1733 if (get_user(len, optlen))
1734 return -EFAULT;
1735 if (len < 0)
1736 return -EINVAL;
1737
1738 switch (optname) {
1739 case NETLINK_PKTINFO:
1740 if (len < sizeof(int))
1741 return -EINVAL;
1742 len = sizeof(int);
1743 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1744 if (put_user(len, optlen) ||
1745 put_user(val, optval))
1746 return -EFAULT;
1747 err = 0;
1748 break;
1749 case NETLINK_BROADCAST_ERROR:
1750 if (len < sizeof(int))
1751 return -EINVAL;
1752 len = sizeof(int);
1753 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1754 if (put_user(len, optlen) ||
1755 put_user(val, optval))
1756 return -EFAULT;
1757 err = 0;
1758 break;
1759 case NETLINK_NO_ENOBUFS:
1760 if (len < sizeof(int))
1761 return -EINVAL;
1762 len = sizeof(int);
1763 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1764 if (put_user(len, optlen) ||
1765 put_user(val, optval))
1766 return -EFAULT;
1767 err = 0;
1768 break;
1769 case NETLINK_LIST_MEMBERSHIPS: {
1770 int pos, idx, shift;
1771
1772 err = 0;
1773 netlink_lock_table();
1774 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1775 if (len - pos < sizeof(u32))
1776 break;
1777
1778 idx = pos / sizeof(unsigned long);
1779 shift = (pos % sizeof(unsigned long)) * 8;
1780 if (put_user((u32)(nlk->groups[idx] >> shift),
1781 (u32 __user *)(optval + pos))) {
1782 err = -EFAULT;
1783 break;
1784 }
1785 }
1786 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1787 err = -EFAULT;
1788 netlink_unlock_table();
1789 break;
1790 }
1791 case NETLINK_CAP_ACK:
1792 if (len < sizeof(int))
1793 return -EINVAL;
1794 len = sizeof(int);
1795 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1796 if (put_user(len, optlen) ||
1797 put_user(val, optval))
1798 return -EFAULT;
1799 err = 0;
1800 break;
1801 case NETLINK_EXT_ACK:
1802 if (len < sizeof(int))
1803 return -EINVAL;
1804 len = sizeof(int);
1805 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1806 if (put_user(len, optlen) || put_user(val, optval))
1807 return -EFAULT;
1808 err = 0;
1809 break;
1810 case NETLINK_GET_STRICT_CHK:
1811 if (len < sizeof(int))
1812 return -EINVAL;
1813 len = sizeof(int);
1814 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1815 if (put_user(len, optlen) || put_user(val, optval))
1816 return -EFAULT;
1817 err = 0;
1818 break;
1819 default:
1820 err = -ENOPROTOOPT;
1821 }
1822 return err;
1823 }
1824
1825 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1826 {
1827 struct nl_pktinfo info;
1828
1829 info.group = NETLINK_CB(skb).dst_group;
1830 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1831 }
1832
1833 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1834 struct sk_buff *skb)
1835 {
1836 if (!NETLINK_CB(skb).nsid_is_set)
1837 return;
1838
1839 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1840 &NETLINK_CB(skb).nsid);
1841 }
1842
1843 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1844 {
1845 struct sock *sk = sock->sk;
1846 struct netlink_sock *nlk = nlk_sk(sk);
1847 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1848 u32 dst_portid;
1849 u32 dst_group;
1850 struct sk_buff *skb;
1851 int err;
1852 struct scm_cookie scm;
1853 u32 netlink_skb_flags = 0;
1854
1855 if (msg->msg_flags&MSG_OOB)
1856 return -EOPNOTSUPP;
1857
1858 err = scm_send(sock, msg, &scm, true);
1859 if (err < 0)
1860 return err;
1861
1862 if (msg->msg_namelen) {
1863 err = -EINVAL;
1864 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1865 goto out;
1866 if (addr->nl_family != AF_NETLINK)
1867 goto out;
1868 dst_portid = addr->nl_pid;
1869 dst_group = ffs(addr->nl_groups);
1870 err = -EPERM;
1871 if ((dst_group || dst_portid) &&
1872 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1873 goto out;
1874 netlink_skb_flags |= NETLINK_SKB_DST;
1875 } else {
1876 dst_portid = nlk->dst_portid;
1877 dst_group = nlk->dst_group;
1878 }
1879
1880 if (!nlk->bound) {
1881 err = netlink_autobind(sock);
1882 if (err)
1883 goto out;
1884 } else {
1885 /* Ensure nlk is hashed and visible. */
1886 smp_rmb();
1887 }
1888
1889 err = -EMSGSIZE;
1890 if (len > sk->sk_sndbuf - 32)
1891 goto out;
1892 err = -ENOBUFS;
1893 skb = netlink_alloc_large_skb(len, dst_group);
1894 if (skb == NULL)
1895 goto out;
1896
1897 NETLINK_CB(skb).portid = nlk->portid;
1898 NETLINK_CB(skb).dst_group = dst_group;
1899 NETLINK_CB(skb).creds = scm.creds;
1900 NETLINK_CB(skb).flags = netlink_skb_flags;
1901
1902 err = -EFAULT;
1903 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1904 kfree_skb(skb);
1905 goto out;
1906 }
1907
1908 err = security_netlink_send(sk, skb);
1909 if (err) {
1910 kfree_skb(skb);
1911 goto out;
1912 }
1913
1914 if (dst_group) {
1915 refcount_inc(&skb->users);
1916 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1917 }
1918 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1919
1920 out:
1921 scm_destroy(&scm);
1922 return err;
1923 }
1924
1925 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1926 int flags)
1927 {
1928 struct scm_cookie scm;
1929 struct sock *sk = sock->sk;
1930 struct netlink_sock *nlk = nlk_sk(sk);
1931 int noblock = flags&MSG_DONTWAIT;
1932 size_t copied;
1933 struct sk_buff *skb, *data_skb;
1934 int err, ret;
1935
1936 if (flags&MSG_OOB)
1937 return -EOPNOTSUPP;
1938
1939 copied = 0;
1940
1941 skb = skb_recv_datagram(sk, flags, noblock, &err);
1942 if (skb == NULL)
1943 goto out;
1944
1945 data_skb = skb;
1946
1947 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1948 if (unlikely(skb_shinfo(skb)->frag_list)) {
1949 /*
1950 * If this skb has a frag_list, then here that means that we
1951 * will have to use the frag_list skb's data for compat tasks
1952 * and the regular skb's data for normal (non-compat) tasks.
1953 *
1954 * If we need to send the compat skb, assign it to the
1955 * 'data_skb' variable so that it will be used below for data
1956 * copying. We keep 'skb' for everything else, including
1957 * freeing both later.
1958 */
1959 if (flags & MSG_CMSG_COMPAT)
1960 data_skb = skb_shinfo(skb)->frag_list;
1961 }
1962 #endif
1963
1964 /* Record the max length of recvmsg() calls for future allocations */
1965 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1966 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1967 SKB_WITH_OVERHEAD(32768));
1968
1969 copied = data_skb->len;
1970 if (len < copied) {
1971 msg->msg_flags |= MSG_TRUNC;
1972 copied = len;
1973 }
1974
1975 skb_reset_transport_header(data_skb);
1976 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1977
1978 if (msg->msg_name) {
1979 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1980 addr->nl_family = AF_NETLINK;
1981 addr->nl_pad = 0;
1982 addr->nl_pid = NETLINK_CB(skb).portid;
1983 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1984 msg->msg_namelen = sizeof(*addr);
1985 }
1986
1987 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1988 netlink_cmsg_recv_pktinfo(msg, skb);
1989 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1990 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1991
1992 memset(&scm, 0, sizeof(scm));
1993 scm.creds = *NETLINK_CREDS(skb);
1994 if (flags & MSG_TRUNC)
1995 copied = data_skb->len;
1996
1997 skb_free_datagram(sk, skb);
1998
1999 if (nlk->cb_running &&
2000 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2001 ret = netlink_dump(sk);
2002 if (ret) {
2003 sk->sk_err = -ret;
2004 sk->sk_error_report(sk);
2005 }
2006 }
2007
2008 scm_recv(sock, msg, &scm, flags);
2009 out:
2010 netlink_rcv_wake(sk);
2011 return err ? : copied;
2012 }
2013
2014 static void netlink_data_ready(struct sock *sk)
2015 {
2016 BUG();
2017 }
2018
2019 /*
2020 * We export these functions to other modules. They provide a
2021 * complete set of kernel non-blocking support for message
2022 * queueing.
2023 */
2024
2025 struct sock *
2026 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2027 struct netlink_kernel_cfg *cfg)
2028 {
2029 struct socket *sock;
2030 struct sock *sk;
2031 struct netlink_sock *nlk;
2032 struct listeners *listeners = NULL;
2033 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2034 unsigned int groups;
2035
2036 BUG_ON(!nl_table);
2037
2038 if (unit < 0 || unit >= MAX_LINKS)
2039 return NULL;
2040
2041 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2042 return NULL;
2043
2044 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2045 goto out_sock_release_nosk;
2046
2047 sk = sock->sk;
2048
2049 if (!cfg || cfg->groups < 32)
2050 groups = 32;
2051 else
2052 groups = cfg->groups;
2053
2054 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2055 if (!listeners)
2056 goto out_sock_release;
2057
2058 sk->sk_data_ready = netlink_data_ready;
2059 if (cfg && cfg->input)
2060 nlk_sk(sk)->netlink_rcv = cfg->input;
2061
2062 if (netlink_insert(sk, 0))
2063 goto out_sock_release;
2064
2065 nlk = nlk_sk(sk);
2066 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2067
2068 netlink_table_grab();
2069 if (!nl_table[unit].registered) {
2070 nl_table[unit].groups = groups;
2071 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2072 nl_table[unit].cb_mutex = cb_mutex;
2073 nl_table[unit].module = module;
2074 if (cfg) {
2075 nl_table[unit].bind = cfg->bind;
2076 nl_table[unit].unbind = cfg->unbind;
2077 nl_table[unit].flags = cfg->flags;
2078 if (cfg->compare)
2079 nl_table[unit].compare = cfg->compare;
2080 }
2081 nl_table[unit].registered = 1;
2082 } else {
2083 kfree(listeners);
2084 nl_table[unit].registered++;
2085 }
2086 netlink_table_ungrab();
2087 return sk;
2088
2089 out_sock_release:
2090 kfree(listeners);
2091 netlink_kernel_release(sk);
2092 return NULL;
2093
2094 out_sock_release_nosk:
2095 sock_release(sock);
2096 return NULL;
2097 }
2098 EXPORT_SYMBOL(__netlink_kernel_create);
2099
2100 void
2101 netlink_kernel_release(struct sock *sk)
2102 {
2103 if (sk == NULL || sk->sk_socket == NULL)
2104 return;
2105
2106 sock_release(sk->sk_socket);
2107 }
2108 EXPORT_SYMBOL(netlink_kernel_release);
2109
2110 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2111 {
2112 struct listeners *new, *old;
2113 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2114
2115 if (groups < 32)
2116 groups = 32;
2117
2118 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2119 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2120 if (!new)
2121 return -ENOMEM;
2122 old = nl_deref_protected(tbl->listeners);
2123 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2124 rcu_assign_pointer(tbl->listeners, new);
2125
2126 kfree_rcu(old, rcu);
2127 }
2128 tbl->groups = groups;
2129
2130 return 0;
2131 }
2132
2133 /**
2134 * netlink_change_ngroups - change number of multicast groups
2135 *
2136 * This changes the number of multicast groups that are available
2137 * on a certain netlink family. Note that it is not possible to
2138 * change the number of groups to below 32. Also note that it does
2139 * not implicitly call netlink_clear_multicast_users() when the
2140 * number of groups is reduced.
2141 *
2142 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2143 * @groups: The new number of groups.
2144 */
2145 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2146 {
2147 int err;
2148
2149 netlink_table_grab();
2150 err = __netlink_change_ngroups(sk, groups);
2151 netlink_table_ungrab();
2152
2153 return err;
2154 }
2155
2156 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2157 {
2158 struct sock *sk;
2159 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2160
2161 sk_for_each_bound(sk, &tbl->mc_list)
2162 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2163 }
2164
2165 struct nlmsghdr *
2166 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2167 {
2168 struct nlmsghdr *nlh;
2169 int size = nlmsg_msg_size(len);
2170
2171 nlh = skb_put(skb, NLMSG_ALIGN(size));
2172 nlh->nlmsg_type = type;
2173 nlh->nlmsg_len = size;
2174 nlh->nlmsg_flags = flags;
2175 nlh->nlmsg_pid = portid;
2176 nlh->nlmsg_seq = seq;
2177 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2178 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2179 return nlh;
2180 }
2181 EXPORT_SYMBOL(__nlmsg_put);
2182
2183 /*
2184 * It looks a bit ugly.
2185 * It would be better to create kernel thread.
2186 */
2187
2188 static int netlink_dump(struct sock *sk)
2189 {
2190 struct netlink_sock *nlk = nlk_sk(sk);
2191 struct netlink_ext_ack extack = {};
2192 struct netlink_callback *cb;
2193 struct sk_buff *skb = NULL;
2194 struct nlmsghdr *nlh;
2195 struct module *module;
2196 int err = -ENOBUFS;
2197 int alloc_min_size;
2198 int alloc_size;
2199
2200 mutex_lock(nlk->cb_mutex);
2201 if (!nlk->cb_running) {
2202 err = -EINVAL;
2203 goto errout_skb;
2204 }
2205
2206 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2207 goto errout_skb;
2208
2209 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2210 * required, but it makes sense to _attempt_ a 16K bytes allocation
2211 * to reduce number of system calls on dump operations, if user
2212 * ever provided a big enough buffer.
2213 */
2214 cb = &nlk->cb;
2215 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2216
2217 if (alloc_min_size < nlk->max_recvmsg_len) {
2218 alloc_size = nlk->max_recvmsg_len;
2219 skb = alloc_skb(alloc_size,
2220 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2221 __GFP_NOWARN | __GFP_NORETRY);
2222 }
2223 if (!skb) {
2224 alloc_size = alloc_min_size;
2225 skb = alloc_skb(alloc_size, GFP_KERNEL);
2226 }
2227 if (!skb)
2228 goto errout_skb;
2229
2230 /* Trim skb to allocated size. User is expected to provide buffer as
2231 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2232 * netlink_recvmsg())). dump will pack as many smaller messages as
2233 * could fit within the allocated skb. skb is typically allocated
2234 * with larger space than required (could be as much as near 2x the
2235 * requested size with align to next power of 2 approach). Allowing
2236 * dump to use the excess space makes it difficult for a user to have a
2237 * reasonable static buffer based on the expected largest dump of a
2238 * single netdev. The outcome is MSG_TRUNC error.
2239 */
2240 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2241 netlink_skb_set_owner_r(skb, sk);
2242
2243 if (nlk->dump_done_errno > 0) {
2244 cb->extack = &extack;
2245 nlk->dump_done_errno = cb->dump(skb, cb);
2246 cb->extack = NULL;
2247 }
2248
2249 if (nlk->dump_done_errno > 0 ||
2250 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2251 mutex_unlock(nlk->cb_mutex);
2252
2253 if (sk_filter(sk, skb))
2254 kfree_skb(skb);
2255 else
2256 __netlink_sendskb(sk, skb);
2257 return 0;
2258 }
2259
2260 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2261 sizeof(nlk->dump_done_errno),
2262 NLM_F_MULTI | cb->answer_flags);
2263 if (WARN_ON(!nlh))
2264 goto errout_skb;
2265
2266 nl_dump_check_consistent(cb, nlh);
2267
2268 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2269 sizeof(nlk->dump_done_errno));
2270
2271 if (extack._msg && nlk->flags & NETLINK_F_EXT_ACK) {
2272 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2273 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack._msg))
2274 nlmsg_end(skb, nlh);
2275 }
2276
2277 if (sk_filter(sk, skb))
2278 kfree_skb(skb);
2279 else
2280 __netlink_sendskb(sk, skb);
2281
2282 if (cb->done)
2283 cb->done(cb);
2284
2285 nlk->cb_running = false;
2286 module = cb->module;
2287 skb = cb->skb;
2288 mutex_unlock(nlk->cb_mutex);
2289 module_put(module);
2290 consume_skb(skb);
2291 return 0;
2292
2293 errout_skb:
2294 mutex_unlock(nlk->cb_mutex);
2295 kfree_skb(skb);
2296 return err;
2297 }
2298
2299 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2300 const struct nlmsghdr *nlh,
2301 struct netlink_dump_control *control)
2302 {
2303 struct netlink_sock *nlk, *nlk2;
2304 struct netlink_callback *cb;
2305 struct sock *sk;
2306 int ret;
2307
2308 refcount_inc(&skb->users);
2309
2310 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2311 if (sk == NULL) {
2312 ret = -ECONNREFUSED;
2313 goto error_free;
2314 }
2315
2316 nlk = nlk_sk(sk);
2317 mutex_lock(nlk->cb_mutex);
2318 /* A dump is in progress... */
2319 if (nlk->cb_running) {
2320 ret = -EBUSY;
2321 goto error_unlock;
2322 }
2323 /* add reference of module which cb->dump belongs to */
2324 if (!try_module_get(control->module)) {
2325 ret = -EPROTONOSUPPORT;
2326 goto error_unlock;
2327 }
2328
2329 cb = &nlk->cb;
2330 memset(cb, 0, sizeof(*cb));
2331 cb->dump = control->dump;
2332 cb->done = control->done;
2333 cb->nlh = nlh;
2334 cb->data = control->data;
2335 cb->module = control->module;
2336 cb->min_dump_alloc = control->min_dump_alloc;
2337 cb->skb = skb;
2338
2339 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2340 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2341
2342 if (control->start) {
2343 ret = control->start(cb);
2344 if (ret)
2345 goto error_put;
2346 }
2347
2348 nlk->cb_running = true;
2349 nlk->dump_done_errno = INT_MAX;
2350
2351 mutex_unlock(nlk->cb_mutex);
2352
2353 ret = netlink_dump(sk);
2354
2355 sock_put(sk);
2356
2357 if (ret)
2358 return ret;
2359
2360 /* We successfully started a dump, by returning -EINTR we
2361 * signal not to send ACK even if it was requested.
2362 */
2363 return -EINTR;
2364
2365 error_put:
2366 module_put(control->module);
2367 error_unlock:
2368 sock_put(sk);
2369 mutex_unlock(nlk->cb_mutex);
2370 error_free:
2371 kfree_skb(skb);
2372 return ret;
2373 }
2374 EXPORT_SYMBOL(__netlink_dump_start);
2375
2376 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2377 const struct netlink_ext_ack *extack)
2378 {
2379 struct sk_buff *skb;
2380 struct nlmsghdr *rep;
2381 struct nlmsgerr *errmsg;
2382 size_t payload = sizeof(*errmsg);
2383 size_t tlvlen = 0;
2384 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2385 unsigned int flags = 0;
2386 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2387
2388 /* Error messages get the original request appened, unless the user
2389 * requests to cap the error message, and get extra error data if
2390 * requested.
2391 */
2392 if (nlk_has_extack && extack && extack->_msg)
2393 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2394
2395 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2396 payload += nlmsg_len(nlh);
2397 else
2398 flags |= NLM_F_CAPPED;
2399 if (err && nlk_has_extack && extack && extack->bad_attr)
2400 tlvlen += nla_total_size(sizeof(u32));
2401 if (nlk_has_extack && extack && extack->cookie_len)
2402 tlvlen += nla_total_size(extack->cookie_len);
2403
2404 if (tlvlen)
2405 flags |= NLM_F_ACK_TLVS;
2406
2407 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2408 if (!skb) {
2409 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2410 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2411 return;
2412 }
2413
2414 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2415 NLMSG_ERROR, payload, flags);
2416 errmsg = nlmsg_data(rep);
2417 errmsg->error = err;
2418 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2419
2420 if (nlk_has_extack && extack) {
2421 if (extack->_msg) {
2422 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2423 extack->_msg));
2424 }
2425 if (err && extack->bad_attr &&
2426 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2427 (u8 *)extack->bad_attr >= in_skb->data +
2428 in_skb->len))
2429 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2430 (u8 *)extack->bad_attr -
2431 (u8 *)nlh));
2432 if (extack->cookie_len)
2433 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2434 extack->cookie_len, extack->cookie));
2435 }
2436
2437 nlmsg_end(skb, rep);
2438
2439 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2440 }
2441 EXPORT_SYMBOL(netlink_ack);
2442
2443 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2444 struct nlmsghdr *,
2445 struct netlink_ext_ack *))
2446 {
2447 struct netlink_ext_ack extack;
2448 struct nlmsghdr *nlh;
2449 int err;
2450
2451 while (skb->len >= nlmsg_total_size(0)) {
2452 int msglen;
2453
2454 memset(&extack, 0, sizeof(extack));
2455 nlh = nlmsg_hdr(skb);
2456 err = 0;
2457
2458 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2459 return 0;
2460
2461 /* Only requests are handled by the kernel */
2462 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2463 goto ack;
2464
2465 /* Skip control messages */
2466 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2467 goto ack;
2468
2469 err = cb(skb, nlh, &extack);
2470 if (err == -EINTR)
2471 goto skip;
2472
2473 ack:
2474 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2475 netlink_ack(skb, nlh, err, &extack);
2476
2477 skip:
2478 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2479 if (msglen > skb->len)
2480 msglen = skb->len;
2481 skb_pull(skb, msglen);
2482 }
2483
2484 return 0;
2485 }
2486 EXPORT_SYMBOL(netlink_rcv_skb);
2487
2488 /**
2489 * nlmsg_notify - send a notification netlink message
2490 * @sk: netlink socket to use
2491 * @skb: notification message
2492 * @portid: destination netlink portid for reports or 0
2493 * @group: destination multicast group or 0
2494 * @report: 1 to report back, 0 to disable
2495 * @flags: allocation flags
2496 */
2497 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2498 unsigned int group, int report, gfp_t flags)
2499 {
2500 int err = 0;
2501
2502 if (group) {
2503 int exclude_portid = 0;
2504
2505 if (report) {
2506 refcount_inc(&skb->users);
2507 exclude_portid = portid;
2508 }
2509
2510 /* errors reported via destination sk->sk_err, but propagate
2511 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2512 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2513 }
2514
2515 if (report) {
2516 int err2;
2517
2518 err2 = nlmsg_unicast(sk, skb, portid);
2519 if (!err || err == -ESRCH)
2520 err = err2;
2521 }
2522
2523 return err;
2524 }
2525 EXPORT_SYMBOL(nlmsg_notify);
2526
2527 #ifdef CONFIG_PROC_FS
2528 struct nl_seq_iter {
2529 struct seq_net_private p;
2530 struct rhashtable_iter hti;
2531 int link;
2532 };
2533
2534 static void netlink_walk_start(struct nl_seq_iter *iter)
2535 {
2536 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2537 rhashtable_walk_start(&iter->hti);
2538 }
2539
2540 static void netlink_walk_stop(struct nl_seq_iter *iter)
2541 {
2542 rhashtable_walk_stop(&iter->hti);
2543 rhashtable_walk_exit(&iter->hti);
2544 }
2545
2546 static void *__netlink_seq_next(struct seq_file *seq)
2547 {
2548 struct nl_seq_iter *iter = seq->private;
2549 struct netlink_sock *nlk;
2550
2551 do {
2552 for (;;) {
2553 nlk = rhashtable_walk_next(&iter->hti);
2554
2555 if (IS_ERR(nlk)) {
2556 if (PTR_ERR(nlk) == -EAGAIN)
2557 continue;
2558
2559 return nlk;
2560 }
2561
2562 if (nlk)
2563 break;
2564
2565 netlink_walk_stop(iter);
2566 if (++iter->link >= MAX_LINKS)
2567 return NULL;
2568
2569 netlink_walk_start(iter);
2570 }
2571 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2572
2573 return nlk;
2574 }
2575
2576 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2577 __acquires(RCU)
2578 {
2579 struct nl_seq_iter *iter = seq->private;
2580 void *obj = SEQ_START_TOKEN;
2581 loff_t pos;
2582
2583 iter->link = 0;
2584
2585 netlink_walk_start(iter);
2586
2587 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2588 obj = __netlink_seq_next(seq);
2589
2590 return obj;
2591 }
2592
2593 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2594 {
2595 ++*pos;
2596 return __netlink_seq_next(seq);
2597 }
2598
2599 static void netlink_seq_stop(struct seq_file *seq, void *v)
2600 {
2601 struct nl_seq_iter *iter = seq->private;
2602
2603 if (iter->link >= MAX_LINKS)
2604 return;
2605
2606 netlink_walk_stop(iter);
2607 }
2608
2609
2610 static int netlink_seq_show(struct seq_file *seq, void *v)
2611 {
2612 if (v == SEQ_START_TOKEN) {
2613 seq_puts(seq,
2614 "sk Eth Pid Groups "
2615 "Rmem Wmem Dump Locks Drops Inode\n");
2616 } else {
2617 struct sock *s = v;
2618 struct netlink_sock *nlk = nlk_sk(s);
2619
2620 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2621 s,
2622 s->sk_protocol,
2623 nlk->portid,
2624 nlk->groups ? (u32)nlk->groups[0] : 0,
2625 sk_rmem_alloc_get(s),
2626 sk_wmem_alloc_get(s),
2627 nlk->cb_running,
2628 refcount_read(&s->sk_refcnt),
2629 atomic_read(&s->sk_drops),
2630 sock_i_ino(s)
2631 );
2632
2633 }
2634 return 0;
2635 }
2636
2637 static const struct seq_operations netlink_seq_ops = {
2638 .start = netlink_seq_start,
2639 .next = netlink_seq_next,
2640 .stop = netlink_seq_stop,
2641 .show = netlink_seq_show,
2642 };
2643 #endif
2644
2645 int netlink_register_notifier(struct notifier_block *nb)
2646 {
2647 return blocking_notifier_chain_register(&netlink_chain, nb);
2648 }
2649 EXPORT_SYMBOL(netlink_register_notifier);
2650
2651 int netlink_unregister_notifier(struct notifier_block *nb)
2652 {
2653 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2654 }
2655 EXPORT_SYMBOL(netlink_unregister_notifier);
2656
2657 static const struct proto_ops netlink_ops = {
2658 .family = PF_NETLINK,
2659 .owner = THIS_MODULE,
2660 .release = netlink_release,
2661 .bind = netlink_bind,
2662 .connect = netlink_connect,
2663 .socketpair = sock_no_socketpair,
2664 .accept = sock_no_accept,
2665 .getname = netlink_getname,
2666 .poll = datagram_poll,
2667 .ioctl = netlink_ioctl,
2668 .listen = sock_no_listen,
2669 .shutdown = sock_no_shutdown,
2670 .setsockopt = netlink_setsockopt,
2671 .getsockopt = netlink_getsockopt,
2672 .sendmsg = netlink_sendmsg,
2673 .recvmsg = netlink_recvmsg,
2674 .mmap = sock_no_mmap,
2675 .sendpage = sock_no_sendpage,
2676 };
2677
2678 static const struct net_proto_family netlink_family_ops = {
2679 .family = PF_NETLINK,
2680 .create = netlink_create,
2681 .owner = THIS_MODULE, /* for consistency 8) */
2682 };
2683
2684 static int __net_init netlink_net_init(struct net *net)
2685 {
2686 #ifdef CONFIG_PROC_FS
2687 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2688 sizeof(struct nl_seq_iter)))
2689 return -ENOMEM;
2690 #endif
2691 return 0;
2692 }
2693
2694 static void __net_exit netlink_net_exit(struct net *net)
2695 {
2696 #ifdef CONFIG_PROC_FS
2697 remove_proc_entry("netlink", net->proc_net);
2698 #endif
2699 }
2700
2701 static void __init netlink_add_usersock_entry(void)
2702 {
2703 struct listeners *listeners;
2704 int groups = 32;
2705
2706 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2707 if (!listeners)
2708 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2709
2710 netlink_table_grab();
2711
2712 nl_table[NETLINK_USERSOCK].groups = groups;
2713 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2714 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2715 nl_table[NETLINK_USERSOCK].registered = 1;
2716 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2717
2718 netlink_table_ungrab();
2719 }
2720
2721 static struct pernet_operations __net_initdata netlink_net_ops = {
2722 .init = netlink_net_init,
2723 .exit = netlink_net_exit,
2724 };
2725
2726 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2727 {
2728 const struct netlink_sock *nlk = data;
2729 struct netlink_compare_arg arg;
2730
2731 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2732 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2733 }
2734
2735 static const struct rhashtable_params netlink_rhashtable_params = {
2736 .head_offset = offsetof(struct netlink_sock, node),
2737 .key_len = netlink_compare_arg_len,
2738 .obj_hashfn = netlink_hash,
2739 .obj_cmpfn = netlink_compare,
2740 .automatic_shrinking = true,
2741 };
2742
2743 static int __init netlink_proto_init(void)
2744 {
2745 int i;
2746 int err = proto_register(&netlink_proto, 0);
2747
2748 if (err != 0)
2749 goto out;
2750
2751 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2752
2753 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2754 if (!nl_table)
2755 goto panic;
2756
2757 for (i = 0; i < MAX_LINKS; i++) {
2758 if (rhashtable_init(&nl_table[i].hash,
2759 &netlink_rhashtable_params) < 0) {
2760 while (--i > 0)
2761 rhashtable_destroy(&nl_table[i].hash);
2762 kfree(nl_table);
2763 goto panic;
2764 }
2765 }
2766
2767 netlink_add_usersock_entry();
2768
2769 sock_register(&netlink_family_ops);
2770 register_pernet_subsys(&netlink_net_ops);
2771 register_pernet_subsys(&netlink_tap_net_ops);
2772 /* The netlink device handler may be needed early. */
2773 rtnetlink_init();
2774 out:
2775 return err;
2776 panic:
2777 panic("netlink_init: Cannot allocate nl_table\n");
2778 }
2779
2780 core_initcall(netlink_proto_init);
Linux with added FPC-III support