xref: /linux/net/netlink/af_netlink.c (revision 95e9fd10f06cb5642028b6b851e32b8c8afb4571)
1 /*
2  * NETLINK      Kernel-user communication protocol.
3  *
4  * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
5  * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6  *
7  *		This program is free software; you can redistribute it and/or
8  *		modify it under the terms of the GNU General Public License
9  *		as published by the Free Software Foundation; either version
10  *		2 of the License, or (at your option) any later version.
11  *
12  * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13  *                               added netlink_proto_exit
14  * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15  * 				 use nlk_sk, as sk->protinfo is on a diet 8)
16  * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17  * 				 - inc module use count of module that owns
18  * 				   the kernel socket in case userspace opens
19  * 				   socket of same protocol
20  * 				 - remove all module support, since netlink is
21  * 				   mandatory if CONFIG_NET=y these days
22  */
23 
24 #include <linux/module.h>
25 
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/mutex.h>
58 
59 #include <net/net_namespace.h>
60 #include <net/sock.h>
61 #include <net/scm.h>
62 #include <net/netlink.h>
63 
64 #define NLGRPSZ(x)	(ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x)	(NLGRPSZ(x)/sizeof(unsigned long))
66 
67 struct netlink_sock {
68 	/* struct sock has to be the first member of netlink_sock */
69 	struct sock		sk;
70 	u32			pid;
71 	u32			dst_pid;
72 	u32			dst_group;
73 	u32			flags;
74 	u32			subscriptions;
75 	u32			ngroups;
76 	unsigned long		*groups;
77 	unsigned long		state;
78 	wait_queue_head_t	wait;
79 	struct netlink_callback	*cb;
80 	struct mutex		*cb_mutex;
81 	struct mutex		cb_def_mutex;
82 	void			(*netlink_rcv)(struct sk_buff *skb);
83 	void			(*netlink_bind)(int group);
84 	struct module		*module;
85 };
86 
87 struct listeners {
88 	struct rcu_head		rcu;
89 	unsigned long		masks[0];
90 };
91 
92 #define NETLINK_KERNEL_SOCKET	0x1
93 #define NETLINK_RECV_PKTINFO	0x2
94 #define NETLINK_BROADCAST_SEND_ERROR	0x4
95 #define NETLINK_RECV_NO_ENOBUFS	0x8
96 
97 static inline struct netlink_sock *nlk_sk(struct sock *sk)
98 {
99 	return container_of(sk, struct netlink_sock, sk);
100 }
101 
102 static inline int netlink_is_kernel(struct sock *sk)
103 {
104 	return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
105 }
106 
107 struct nl_pid_hash {
108 	struct hlist_head	*table;
109 	unsigned long		rehash_time;
110 
111 	unsigned int		mask;
112 	unsigned int		shift;
113 
114 	unsigned int		entries;
115 	unsigned int		max_shift;
116 
117 	u32			rnd;
118 };
119 
120 struct netlink_table {
121 	struct nl_pid_hash	hash;
122 	struct hlist_head	mc_list;
123 	struct listeners __rcu	*listeners;
124 	unsigned int		nl_nonroot;
125 	unsigned int		groups;
126 	struct mutex		*cb_mutex;
127 	struct module		*module;
128 	void			(*bind)(int group);
129 	int			registered;
130 };
131 
132 static struct netlink_table *nl_table;
133 
134 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
135 
136 static int netlink_dump(struct sock *sk);
137 
138 static DEFINE_RWLOCK(nl_table_lock);
139 static atomic_t nl_table_users = ATOMIC_INIT(0);
140 
141 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
142 
143 static inline u32 netlink_group_mask(u32 group)
144 {
145 	return group ? 1 << (group - 1) : 0;
146 }
147 
148 static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
149 {
150 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
151 }
152 
153 static void netlink_destroy_callback(struct netlink_callback *cb)
154 {
155 	kfree_skb(cb->skb);
156 	kfree(cb);
157 }
158 
159 static void netlink_consume_callback(struct netlink_callback *cb)
160 {
161 	consume_skb(cb->skb);
162 	kfree(cb);
163 }
164 
165 static void netlink_sock_destruct(struct sock *sk)
166 {
167 	struct netlink_sock *nlk = nlk_sk(sk);
168 
169 	if (nlk->cb) {
170 		if (nlk->cb->done)
171 			nlk->cb->done(nlk->cb);
172 		netlink_destroy_callback(nlk->cb);
173 	}
174 
175 	skb_queue_purge(&sk->sk_receive_queue);
176 
177 	if (!sock_flag(sk, SOCK_DEAD)) {
178 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
179 		return;
180 	}
181 
182 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
183 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
184 	WARN_ON(nlk_sk(sk)->groups);
185 }
186 
187 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
188  * SMP. Look, when several writers sleep and reader wakes them up, all but one
189  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
190  * this, _but_ remember, it adds useless work on UP machines.
191  */
192 
193 void netlink_table_grab(void)
194 	__acquires(nl_table_lock)
195 {
196 	might_sleep();
197 
198 	write_lock_irq(&nl_table_lock);
199 
200 	if (atomic_read(&nl_table_users)) {
201 		DECLARE_WAITQUEUE(wait, current);
202 
203 		add_wait_queue_exclusive(&nl_table_wait, &wait);
204 		for (;;) {
205 			set_current_state(TASK_UNINTERRUPTIBLE);
206 			if (atomic_read(&nl_table_users) == 0)
207 				break;
208 			write_unlock_irq(&nl_table_lock);
209 			schedule();
210 			write_lock_irq(&nl_table_lock);
211 		}
212 
213 		__set_current_state(TASK_RUNNING);
214 		remove_wait_queue(&nl_table_wait, &wait);
215 	}
216 }
217 
218 void netlink_table_ungrab(void)
219 	__releases(nl_table_lock)
220 {
221 	write_unlock_irq(&nl_table_lock);
222 	wake_up(&nl_table_wait);
223 }
224 
225 static inline void
226 netlink_lock_table(void)
227 {
228 	/* read_lock() synchronizes us to netlink_table_grab */
229 
230 	read_lock(&nl_table_lock);
231 	atomic_inc(&nl_table_users);
232 	read_unlock(&nl_table_lock);
233 }
234 
235 static inline void
236 netlink_unlock_table(void)
237 {
238 	if (atomic_dec_and_test(&nl_table_users))
239 		wake_up(&nl_table_wait);
240 }
241 
242 static struct sock *netlink_lookup(struct net *net, int protocol, u32 pid)
243 {
244 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
245 	struct hlist_head *head;
246 	struct sock *sk;
247 	struct hlist_node *node;
248 
249 	read_lock(&nl_table_lock);
250 	head = nl_pid_hashfn(hash, pid);
251 	sk_for_each(sk, node, head) {
252 		if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
253 			sock_hold(sk);
254 			goto found;
255 		}
256 	}
257 	sk = NULL;
258 found:
259 	read_unlock(&nl_table_lock);
260 	return sk;
261 }
262 
263 static struct hlist_head *nl_pid_hash_zalloc(size_t size)
264 {
265 	if (size <= PAGE_SIZE)
266 		return kzalloc(size, GFP_ATOMIC);
267 	else
268 		return (struct hlist_head *)
269 			__get_free_pages(GFP_ATOMIC | __GFP_ZERO,
270 					 get_order(size));
271 }
272 
273 static void nl_pid_hash_free(struct hlist_head *table, size_t size)
274 {
275 	if (size <= PAGE_SIZE)
276 		kfree(table);
277 	else
278 		free_pages((unsigned long)table, get_order(size));
279 }
280 
281 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
282 {
283 	unsigned int omask, mask, shift;
284 	size_t osize, size;
285 	struct hlist_head *otable, *table;
286 	int i;
287 
288 	omask = mask = hash->mask;
289 	osize = size = (mask + 1) * sizeof(*table);
290 	shift = hash->shift;
291 
292 	if (grow) {
293 		if (++shift > hash->max_shift)
294 			return 0;
295 		mask = mask * 2 + 1;
296 		size *= 2;
297 	}
298 
299 	table = nl_pid_hash_zalloc(size);
300 	if (!table)
301 		return 0;
302 
303 	otable = hash->table;
304 	hash->table = table;
305 	hash->mask = mask;
306 	hash->shift = shift;
307 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
308 
309 	for (i = 0; i <= omask; i++) {
310 		struct sock *sk;
311 		struct hlist_node *node, *tmp;
312 
313 		sk_for_each_safe(sk, node, tmp, &otable[i])
314 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
315 	}
316 
317 	nl_pid_hash_free(otable, osize);
318 	hash->rehash_time = jiffies + 10 * 60 * HZ;
319 	return 1;
320 }
321 
322 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
323 {
324 	int avg = hash->entries >> hash->shift;
325 
326 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
327 		return 1;
328 
329 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
330 		nl_pid_hash_rehash(hash, 0);
331 		return 1;
332 	}
333 
334 	return 0;
335 }
336 
337 static const struct proto_ops netlink_ops;
338 
339 static void
340 netlink_update_listeners(struct sock *sk)
341 {
342 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
343 	struct hlist_node *node;
344 	unsigned long mask;
345 	unsigned int i;
346 
347 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
348 		mask = 0;
349 		sk_for_each_bound(sk, node, &tbl->mc_list) {
350 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
351 				mask |= nlk_sk(sk)->groups[i];
352 		}
353 		tbl->listeners->masks[i] = mask;
354 	}
355 	/* this function is only called with the netlink table "grabbed", which
356 	 * makes sure updates are visible before bind or setsockopt return. */
357 }
358 
359 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
360 {
361 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
362 	struct hlist_head *head;
363 	int err = -EADDRINUSE;
364 	struct sock *osk;
365 	struct hlist_node *node;
366 	int len;
367 
368 	netlink_table_grab();
369 	head = nl_pid_hashfn(hash, pid);
370 	len = 0;
371 	sk_for_each(osk, node, head) {
372 		if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
373 			break;
374 		len++;
375 	}
376 	if (node)
377 		goto err;
378 
379 	err = -EBUSY;
380 	if (nlk_sk(sk)->pid)
381 		goto err;
382 
383 	err = -ENOMEM;
384 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
385 		goto err;
386 
387 	if (len && nl_pid_hash_dilute(hash, len))
388 		head = nl_pid_hashfn(hash, pid);
389 	hash->entries++;
390 	nlk_sk(sk)->pid = pid;
391 	sk_add_node(sk, head);
392 	err = 0;
393 
394 err:
395 	netlink_table_ungrab();
396 	return err;
397 }
398 
399 static void netlink_remove(struct sock *sk)
400 {
401 	netlink_table_grab();
402 	if (sk_del_node_init(sk))
403 		nl_table[sk->sk_protocol].hash.entries--;
404 	if (nlk_sk(sk)->subscriptions)
405 		__sk_del_bind_node(sk);
406 	netlink_table_ungrab();
407 }
408 
409 static struct proto netlink_proto = {
410 	.name	  = "NETLINK",
411 	.owner	  = THIS_MODULE,
412 	.obj_size = sizeof(struct netlink_sock),
413 };
414 
415 static int __netlink_create(struct net *net, struct socket *sock,
416 			    struct mutex *cb_mutex, int protocol)
417 {
418 	struct sock *sk;
419 	struct netlink_sock *nlk;
420 
421 	sock->ops = &netlink_ops;
422 
423 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
424 	if (!sk)
425 		return -ENOMEM;
426 
427 	sock_init_data(sock, sk);
428 
429 	nlk = nlk_sk(sk);
430 	if (cb_mutex) {
431 		nlk->cb_mutex = cb_mutex;
432 	} else {
433 		nlk->cb_mutex = &nlk->cb_def_mutex;
434 		mutex_init(nlk->cb_mutex);
435 	}
436 	init_waitqueue_head(&nlk->wait);
437 
438 	sk->sk_destruct = netlink_sock_destruct;
439 	sk->sk_protocol = protocol;
440 	return 0;
441 }
442 
443 static int netlink_create(struct net *net, struct socket *sock, int protocol,
444 			  int kern)
445 {
446 	struct module *module = NULL;
447 	struct mutex *cb_mutex;
448 	struct netlink_sock *nlk;
449 	void (*bind)(int group);
450 	int err = 0;
451 
452 	sock->state = SS_UNCONNECTED;
453 
454 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
455 		return -ESOCKTNOSUPPORT;
456 
457 	if (protocol < 0 || protocol >= MAX_LINKS)
458 		return -EPROTONOSUPPORT;
459 
460 	netlink_lock_table();
461 #ifdef CONFIG_MODULES
462 	if (!nl_table[protocol].registered) {
463 		netlink_unlock_table();
464 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
465 		netlink_lock_table();
466 	}
467 #endif
468 	if (nl_table[protocol].registered &&
469 	    try_module_get(nl_table[protocol].module))
470 		module = nl_table[protocol].module;
471 	else
472 		err = -EPROTONOSUPPORT;
473 	cb_mutex = nl_table[protocol].cb_mutex;
474 	bind = nl_table[protocol].bind;
475 	netlink_unlock_table();
476 
477 	if (err < 0)
478 		goto out;
479 
480 	err = __netlink_create(net, sock, cb_mutex, protocol);
481 	if (err < 0)
482 		goto out_module;
483 
484 	local_bh_disable();
485 	sock_prot_inuse_add(net, &netlink_proto, 1);
486 	local_bh_enable();
487 
488 	nlk = nlk_sk(sock->sk);
489 	nlk->module = module;
490 	nlk->netlink_bind = bind;
491 out:
492 	return err;
493 
494 out_module:
495 	module_put(module);
496 	goto out;
497 }
498 
499 static int netlink_release(struct socket *sock)
500 {
501 	struct sock *sk = sock->sk;
502 	struct netlink_sock *nlk;
503 
504 	if (!sk)
505 		return 0;
506 
507 	netlink_remove(sk);
508 	sock_orphan(sk);
509 	nlk = nlk_sk(sk);
510 
511 	/*
512 	 * OK. Socket is unlinked, any packets that arrive now
513 	 * will be purged.
514 	 */
515 
516 	sock->sk = NULL;
517 	wake_up_interruptible_all(&nlk->wait);
518 
519 	skb_queue_purge(&sk->sk_write_queue);
520 
521 	if (nlk->pid) {
522 		struct netlink_notify n = {
523 						.net = sock_net(sk),
524 						.protocol = sk->sk_protocol,
525 						.pid = nlk->pid,
526 					  };
527 		atomic_notifier_call_chain(&netlink_chain,
528 				NETLINK_URELEASE, &n);
529 	}
530 
531 	module_put(nlk->module);
532 
533 	netlink_table_grab();
534 	if (netlink_is_kernel(sk)) {
535 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
536 		if (--nl_table[sk->sk_protocol].registered == 0) {
537 			kfree(nl_table[sk->sk_protocol].listeners);
538 			nl_table[sk->sk_protocol].module = NULL;
539 			nl_table[sk->sk_protocol].registered = 0;
540 		}
541 	} else if (nlk->subscriptions) {
542 		netlink_update_listeners(sk);
543 	}
544 	netlink_table_ungrab();
545 
546 	kfree(nlk->groups);
547 	nlk->groups = NULL;
548 
549 	local_bh_disable();
550 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
551 	local_bh_enable();
552 	sock_put(sk);
553 	return 0;
554 }
555 
556 static int netlink_autobind(struct socket *sock)
557 {
558 	struct sock *sk = sock->sk;
559 	struct net *net = sock_net(sk);
560 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
561 	struct hlist_head *head;
562 	struct sock *osk;
563 	struct hlist_node *node;
564 	s32 pid = task_tgid_vnr(current);
565 	int err;
566 	static s32 rover = -4097;
567 
568 retry:
569 	cond_resched();
570 	netlink_table_grab();
571 	head = nl_pid_hashfn(hash, pid);
572 	sk_for_each(osk, node, head) {
573 		if (!net_eq(sock_net(osk), net))
574 			continue;
575 		if (nlk_sk(osk)->pid == pid) {
576 			/* Bind collision, search negative pid values. */
577 			pid = rover--;
578 			if (rover > -4097)
579 				rover = -4097;
580 			netlink_table_ungrab();
581 			goto retry;
582 		}
583 	}
584 	netlink_table_ungrab();
585 
586 	err = netlink_insert(sk, net, pid);
587 	if (err == -EADDRINUSE)
588 		goto retry;
589 
590 	/* If 2 threads race to autobind, that is fine.  */
591 	if (err == -EBUSY)
592 		err = 0;
593 
594 	return err;
595 }
596 
597 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
598 {
599 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
600 	       capable(CAP_NET_ADMIN);
601 }
602 
603 static void
604 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
605 {
606 	struct netlink_sock *nlk = nlk_sk(sk);
607 
608 	if (nlk->subscriptions && !subscriptions)
609 		__sk_del_bind_node(sk);
610 	else if (!nlk->subscriptions && subscriptions)
611 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
612 	nlk->subscriptions = subscriptions;
613 }
614 
615 static int netlink_realloc_groups(struct sock *sk)
616 {
617 	struct netlink_sock *nlk = nlk_sk(sk);
618 	unsigned int groups;
619 	unsigned long *new_groups;
620 	int err = 0;
621 
622 	netlink_table_grab();
623 
624 	groups = nl_table[sk->sk_protocol].groups;
625 	if (!nl_table[sk->sk_protocol].registered) {
626 		err = -ENOENT;
627 		goto out_unlock;
628 	}
629 
630 	if (nlk->ngroups >= groups)
631 		goto out_unlock;
632 
633 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
634 	if (new_groups == NULL) {
635 		err = -ENOMEM;
636 		goto out_unlock;
637 	}
638 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
639 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
640 
641 	nlk->groups = new_groups;
642 	nlk->ngroups = groups;
643  out_unlock:
644 	netlink_table_ungrab();
645 	return err;
646 }
647 
648 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
649 			int addr_len)
650 {
651 	struct sock *sk = sock->sk;
652 	struct net *net = sock_net(sk);
653 	struct netlink_sock *nlk = nlk_sk(sk);
654 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
655 	int err;
656 
657 	if (nladdr->nl_family != AF_NETLINK)
658 		return -EINVAL;
659 
660 	/* Only superuser is allowed to listen multicasts */
661 	if (nladdr->nl_groups) {
662 		if (!netlink_capable(sock, NL_NONROOT_RECV))
663 			return -EPERM;
664 		err = netlink_realloc_groups(sk);
665 		if (err)
666 			return err;
667 	}
668 
669 	if (nlk->pid) {
670 		if (nladdr->nl_pid != nlk->pid)
671 			return -EINVAL;
672 	} else {
673 		err = nladdr->nl_pid ?
674 			netlink_insert(sk, net, nladdr->nl_pid) :
675 			netlink_autobind(sock);
676 		if (err)
677 			return err;
678 	}
679 
680 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
681 		return 0;
682 
683 	netlink_table_grab();
684 	netlink_update_subscriptions(sk, nlk->subscriptions +
685 					 hweight32(nladdr->nl_groups) -
686 					 hweight32(nlk->groups[0]));
687 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
688 	netlink_update_listeners(sk);
689 	netlink_table_ungrab();
690 
691 	if (nlk->netlink_bind && nlk->groups[0]) {
692 		int i;
693 
694 		for (i=0; i<nlk->ngroups; i++) {
695 			if (test_bit(i, nlk->groups))
696 				nlk->netlink_bind(i);
697 		}
698 	}
699 
700 	return 0;
701 }
702 
703 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
704 			   int alen, int flags)
705 {
706 	int err = 0;
707 	struct sock *sk = sock->sk;
708 	struct netlink_sock *nlk = nlk_sk(sk);
709 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
710 
711 	if (alen < sizeof(addr->sa_family))
712 		return -EINVAL;
713 
714 	if (addr->sa_family == AF_UNSPEC) {
715 		sk->sk_state	= NETLINK_UNCONNECTED;
716 		nlk->dst_pid	= 0;
717 		nlk->dst_group  = 0;
718 		return 0;
719 	}
720 	if (addr->sa_family != AF_NETLINK)
721 		return -EINVAL;
722 
723 	/* Only superuser is allowed to send multicasts */
724 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
725 		return -EPERM;
726 
727 	if (!nlk->pid)
728 		err = netlink_autobind(sock);
729 
730 	if (err == 0) {
731 		sk->sk_state	= NETLINK_CONNECTED;
732 		nlk->dst_pid 	= nladdr->nl_pid;
733 		nlk->dst_group  = ffs(nladdr->nl_groups);
734 	}
735 
736 	return err;
737 }
738 
739 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
740 			   int *addr_len, int peer)
741 {
742 	struct sock *sk = sock->sk;
743 	struct netlink_sock *nlk = nlk_sk(sk);
744 	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
745 
746 	nladdr->nl_family = AF_NETLINK;
747 	nladdr->nl_pad = 0;
748 	*addr_len = sizeof(*nladdr);
749 
750 	if (peer) {
751 		nladdr->nl_pid = nlk->dst_pid;
752 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
753 	} else {
754 		nladdr->nl_pid = nlk->pid;
755 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
756 	}
757 	return 0;
758 }
759 
760 static void netlink_overrun(struct sock *sk)
761 {
762 	struct netlink_sock *nlk = nlk_sk(sk);
763 
764 	if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
765 		if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
766 			sk->sk_err = ENOBUFS;
767 			sk->sk_error_report(sk);
768 		}
769 	}
770 	atomic_inc(&sk->sk_drops);
771 }
772 
773 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
774 {
775 	struct sock *sock;
776 	struct netlink_sock *nlk;
777 
778 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
779 	if (!sock)
780 		return ERR_PTR(-ECONNREFUSED);
781 
782 	/* Don't bother queuing skb if kernel socket has no input function */
783 	nlk = nlk_sk(sock);
784 	if (sock->sk_state == NETLINK_CONNECTED &&
785 	    nlk->dst_pid != nlk_sk(ssk)->pid) {
786 		sock_put(sock);
787 		return ERR_PTR(-ECONNREFUSED);
788 	}
789 	return sock;
790 }
791 
792 struct sock *netlink_getsockbyfilp(struct file *filp)
793 {
794 	struct inode *inode = filp->f_path.dentry->d_inode;
795 	struct sock *sock;
796 
797 	if (!S_ISSOCK(inode->i_mode))
798 		return ERR_PTR(-ENOTSOCK);
799 
800 	sock = SOCKET_I(inode)->sk;
801 	if (sock->sk_family != AF_NETLINK)
802 		return ERR_PTR(-EINVAL);
803 
804 	sock_hold(sock);
805 	return sock;
806 }
807 
808 /*
809  * Attach a skb to a netlink socket.
810  * The caller must hold a reference to the destination socket. On error, the
811  * reference is dropped. The skb is not send to the destination, just all
812  * all error checks are performed and memory in the queue is reserved.
813  * Return values:
814  * < 0: error. skb freed, reference to sock dropped.
815  * 0: continue
816  * 1: repeat lookup - reference dropped while waiting for socket memory.
817  */
818 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
819 		      long *timeo, struct sock *ssk)
820 {
821 	struct netlink_sock *nlk;
822 
823 	nlk = nlk_sk(sk);
824 
825 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
826 	    test_bit(0, &nlk->state)) {
827 		DECLARE_WAITQUEUE(wait, current);
828 		if (!*timeo) {
829 			if (!ssk || netlink_is_kernel(ssk))
830 				netlink_overrun(sk);
831 			sock_put(sk);
832 			kfree_skb(skb);
833 			return -EAGAIN;
834 		}
835 
836 		__set_current_state(TASK_INTERRUPTIBLE);
837 		add_wait_queue(&nlk->wait, &wait);
838 
839 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
840 		     test_bit(0, &nlk->state)) &&
841 		    !sock_flag(sk, SOCK_DEAD))
842 			*timeo = schedule_timeout(*timeo);
843 
844 		__set_current_state(TASK_RUNNING);
845 		remove_wait_queue(&nlk->wait, &wait);
846 		sock_put(sk);
847 
848 		if (signal_pending(current)) {
849 			kfree_skb(skb);
850 			return sock_intr_errno(*timeo);
851 		}
852 		return 1;
853 	}
854 	skb_set_owner_r(skb, sk);
855 	return 0;
856 }
857 
858 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
859 {
860 	int len = skb->len;
861 
862 	skb_queue_tail(&sk->sk_receive_queue, skb);
863 	sk->sk_data_ready(sk, len);
864 	return len;
865 }
866 
867 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
868 {
869 	int len = __netlink_sendskb(sk, skb);
870 
871 	sock_put(sk);
872 	return len;
873 }
874 
875 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
876 {
877 	kfree_skb(skb);
878 	sock_put(sk);
879 }
880 
881 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
882 {
883 	int delta;
884 
885 	skb_orphan(skb);
886 
887 	delta = skb->end - skb->tail;
888 	if (delta * 2 < skb->truesize)
889 		return skb;
890 
891 	if (skb_shared(skb)) {
892 		struct sk_buff *nskb = skb_clone(skb, allocation);
893 		if (!nskb)
894 			return skb;
895 		consume_skb(skb);
896 		skb = nskb;
897 	}
898 
899 	if (!pskb_expand_head(skb, 0, -delta, allocation))
900 		skb->truesize -= delta;
901 
902 	return skb;
903 }
904 
905 static void netlink_rcv_wake(struct sock *sk)
906 {
907 	struct netlink_sock *nlk = nlk_sk(sk);
908 
909 	if (skb_queue_empty(&sk->sk_receive_queue))
910 		clear_bit(0, &nlk->state);
911 	if (!test_bit(0, &nlk->state))
912 		wake_up_interruptible(&nlk->wait);
913 }
914 
915 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
916 {
917 	int ret;
918 	struct netlink_sock *nlk = nlk_sk(sk);
919 
920 	ret = -ECONNREFUSED;
921 	if (nlk->netlink_rcv != NULL) {
922 		ret = skb->len;
923 		skb_set_owner_r(skb, sk);
924 		nlk->netlink_rcv(skb);
925 		consume_skb(skb);
926 	} else {
927 		kfree_skb(skb);
928 	}
929 	sock_put(sk);
930 	return ret;
931 }
932 
933 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
934 		    u32 pid, int nonblock)
935 {
936 	struct sock *sk;
937 	int err;
938 	long timeo;
939 
940 	skb = netlink_trim(skb, gfp_any());
941 
942 	timeo = sock_sndtimeo(ssk, nonblock);
943 retry:
944 	sk = netlink_getsockbypid(ssk, pid);
945 	if (IS_ERR(sk)) {
946 		kfree_skb(skb);
947 		return PTR_ERR(sk);
948 	}
949 	if (netlink_is_kernel(sk))
950 		return netlink_unicast_kernel(sk, skb);
951 
952 	if (sk_filter(sk, skb)) {
953 		err = skb->len;
954 		kfree_skb(skb);
955 		sock_put(sk);
956 		return err;
957 	}
958 
959 	err = netlink_attachskb(sk, skb, &timeo, ssk);
960 	if (err == 1)
961 		goto retry;
962 	if (err)
963 		return err;
964 
965 	return netlink_sendskb(sk, skb);
966 }
967 EXPORT_SYMBOL(netlink_unicast);
968 
969 int netlink_has_listeners(struct sock *sk, unsigned int group)
970 {
971 	int res = 0;
972 	struct listeners *listeners;
973 
974 	BUG_ON(!netlink_is_kernel(sk));
975 
976 	rcu_read_lock();
977 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
978 
979 	if (group - 1 < nl_table[sk->sk_protocol].groups)
980 		res = test_bit(group - 1, listeners->masks);
981 
982 	rcu_read_unlock();
983 
984 	return res;
985 }
986 EXPORT_SYMBOL_GPL(netlink_has_listeners);
987 
988 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
989 {
990 	struct netlink_sock *nlk = nlk_sk(sk);
991 
992 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
993 	    !test_bit(0, &nlk->state)) {
994 		skb_set_owner_r(skb, sk);
995 		__netlink_sendskb(sk, skb);
996 		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
997 	}
998 	return -1;
999 }
1000 
1001 struct netlink_broadcast_data {
1002 	struct sock *exclude_sk;
1003 	struct net *net;
1004 	u32 pid;
1005 	u32 group;
1006 	int failure;
1007 	int delivery_failure;
1008 	int congested;
1009 	int delivered;
1010 	gfp_t allocation;
1011 	struct sk_buff *skb, *skb2;
1012 	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1013 	void *tx_data;
1014 };
1015 
1016 static int do_one_broadcast(struct sock *sk,
1017 				   struct netlink_broadcast_data *p)
1018 {
1019 	struct netlink_sock *nlk = nlk_sk(sk);
1020 	int val;
1021 
1022 	if (p->exclude_sk == sk)
1023 		goto out;
1024 
1025 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1026 	    !test_bit(p->group - 1, nlk->groups))
1027 		goto out;
1028 
1029 	if (!net_eq(sock_net(sk), p->net))
1030 		goto out;
1031 
1032 	if (p->failure) {
1033 		netlink_overrun(sk);
1034 		goto out;
1035 	}
1036 
1037 	sock_hold(sk);
1038 	if (p->skb2 == NULL) {
1039 		if (skb_shared(p->skb)) {
1040 			p->skb2 = skb_clone(p->skb, p->allocation);
1041 		} else {
1042 			p->skb2 = skb_get(p->skb);
1043 			/*
1044 			 * skb ownership may have been set when
1045 			 * delivered to a previous socket.
1046 			 */
1047 			skb_orphan(p->skb2);
1048 		}
1049 	}
1050 	if (p->skb2 == NULL) {
1051 		netlink_overrun(sk);
1052 		/* Clone failed. Notify ALL listeners. */
1053 		p->failure = 1;
1054 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1055 			p->delivery_failure = 1;
1056 	} else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1057 		kfree_skb(p->skb2);
1058 		p->skb2 = NULL;
1059 	} else if (sk_filter(sk, p->skb2)) {
1060 		kfree_skb(p->skb2);
1061 		p->skb2 = NULL;
1062 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1063 		netlink_overrun(sk);
1064 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1065 			p->delivery_failure = 1;
1066 	} else {
1067 		p->congested |= val;
1068 		p->delivered = 1;
1069 		p->skb2 = NULL;
1070 	}
1071 	sock_put(sk);
1072 
1073 out:
1074 	return 0;
1075 }
1076 
1077 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1078 	u32 group, gfp_t allocation,
1079 	int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1080 	void *filter_data)
1081 {
1082 	struct net *net = sock_net(ssk);
1083 	struct netlink_broadcast_data info;
1084 	struct hlist_node *node;
1085 	struct sock *sk;
1086 
1087 	skb = netlink_trim(skb, allocation);
1088 
1089 	info.exclude_sk = ssk;
1090 	info.net = net;
1091 	info.pid = pid;
1092 	info.group = group;
1093 	info.failure = 0;
1094 	info.delivery_failure = 0;
1095 	info.congested = 0;
1096 	info.delivered = 0;
1097 	info.allocation = allocation;
1098 	info.skb = skb;
1099 	info.skb2 = NULL;
1100 	info.tx_filter = filter;
1101 	info.tx_data = filter_data;
1102 
1103 	/* While we sleep in clone, do not allow to change socket list */
1104 
1105 	netlink_lock_table();
1106 
1107 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1108 		do_one_broadcast(sk, &info);
1109 
1110 	consume_skb(skb);
1111 
1112 	netlink_unlock_table();
1113 
1114 	if (info.delivery_failure) {
1115 		kfree_skb(info.skb2);
1116 		return -ENOBUFS;
1117 	}
1118 	consume_skb(info.skb2);
1119 
1120 	if (info.delivered) {
1121 		if (info.congested && (allocation & __GFP_WAIT))
1122 			yield();
1123 		return 0;
1124 	}
1125 	return -ESRCH;
1126 }
1127 EXPORT_SYMBOL(netlink_broadcast_filtered);
1128 
1129 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1130 		      u32 group, gfp_t allocation)
1131 {
1132 	return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1133 		NULL, NULL);
1134 }
1135 EXPORT_SYMBOL(netlink_broadcast);
1136 
1137 struct netlink_set_err_data {
1138 	struct sock *exclude_sk;
1139 	u32 pid;
1140 	u32 group;
1141 	int code;
1142 };
1143 
1144 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1145 {
1146 	struct netlink_sock *nlk = nlk_sk(sk);
1147 	int ret = 0;
1148 
1149 	if (sk == p->exclude_sk)
1150 		goto out;
1151 
1152 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1153 		goto out;
1154 
1155 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1156 	    !test_bit(p->group - 1, nlk->groups))
1157 		goto out;
1158 
1159 	if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1160 		ret = 1;
1161 		goto out;
1162 	}
1163 
1164 	sk->sk_err = p->code;
1165 	sk->sk_error_report(sk);
1166 out:
1167 	return ret;
1168 }
1169 
1170 /**
1171  * netlink_set_err - report error to broadcast listeners
1172  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1173  * @pid: the PID of a process that we want to skip (if any)
1174  * @groups: the broadcast group that will notice the error
1175  * @code: error code, must be negative (as usual in kernelspace)
1176  *
1177  * This function returns the number of broadcast listeners that have set the
1178  * NETLINK_RECV_NO_ENOBUFS socket option.
1179  */
1180 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1181 {
1182 	struct netlink_set_err_data info;
1183 	struct hlist_node *node;
1184 	struct sock *sk;
1185 	int ret = 0;
1186 
1187 	info.exclude_sk = ssk;
1188 	info.pid = pid;
1189 	info.group = group;
1190 	/* sk->sk_err wants a positive error value */
1191 	info.code = -code;
1192 
1193 	read_lock(&nl_table_lock);
1194 
1195 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1196 		ret += do_one_set_err(sk, &info);
1197 
1198 	read_unlock(&nl_table_lock);
1199 	return ret;
1200 }
1201 EXPORT_SYMBOL(netlink_set_err);
1202 
1203 /* must be called with netlink table grabbed */
1204 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1205 				     unsigned int group,
1206 				     int is_new)
1207 {
1208 	int old, new = !!is_new, subscriptions;
1209 
1210 	old = test_bit(group - 1, nlk->groups);
1211 	subscriptions = nlk->subscriptions - old + new;
1212 	if (new)
1213 		__set_bit(group - 1, nlk->groups);
1214 	else
1215 		__clear_bit(group - 1, nlk->groups);
1216 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1217 	netlink_update_listeners(&nlk->sk);
1218 }
1219 
1220 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1221 			      char __user *optval, unsigned int optlen)
1222 {
1223 	struct sock *sk = sock->sk;
1224 	struct netlink_sock *nlk = nlk_sk(sk);
1225 	unsigned int val = 0;
1226 	int err;
1227 
1228 	if (level != SOL_NETLINK)
1229 		return -ENOPROTOOPT;
1230 
1231 	if (optlen >= sizeof(int) &&
1232 	    get_user(val, (unsigned int __user *)optval))
1233 		return -EFAULT;
1234 
1235 	switch (optname) {
1236 	case NETLINK_PKTINFO:
1237 		if (val)
1238 			nlk->flags |= NETLINK_RECV_PKTINFO;
1239 		else
1240 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1241 		err = 0;
1242 		break;
1243 	case NETLINK_ADD_MEMBERSHIP:
1244 	case NETLINK_DROP_MEMBERSHIP: {
1245 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1246 			return -EPERM;
1247 		err = netlink_realloc_groups(sk);
1248 		if (err)
1249 			return err;
1250 		if (!val || val - 1 >= nlk->ngroups)
1251 			return -EINVAL;
1252 		netlink_table_grab();
1253 		netlink_update_socket_mc(nlk, val,
1254 					 optname == NETLINK_ADD_MEMBERSHIP);
1255 		netlink_table_ungrab();
1256 
1257 		if (nlk->netlink_bind)
1258 			nlk->netlink_bind(val);
1259 
1260 		err = 0;
1261 		break;
1262 	}
1263 	case NETLINK_BROADCAST_ERROR:
1264 		if (val)
1265 			nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1266 		else
1267 			nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1268 		err = 0;
1269 		break;
1270 	case NETLINK_NO_ENOBUFS:
1271 		if (val) {
1272 			nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1273 			clear_bit(0, &nlk->state);
1274 			wake_up_interruptible(&nlk->wait);
1275 		} else {
1276 			nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1277 		}
1278 		err = 0;
1279 		break;
1280 	default:
1281 		err = -ENOPROTOOPT;
1282 	}
1283 	return err;
1284 }
1285 
1286 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1287 			      char __user *optval, int __user *optlen)
1288 {
1289 	struct sock *sk = sock->sk;
1290 	struct netlink_sock *nlk = nlk_sk(sk);
1291 	int len, val, err;
1292 
1293 	if (level != SOL_NETLINK)
1294 		return -ENOPROTOOPT;
1295 
1296 	if (get_user(len, optlen))
1297 		return -EFAULT;
1298 	if (len < 0)
1299 		return -EINVAL;
1300 
1301 	switch (optname) {
1302 	case NETLINK_PKTINFO:
1303 		if (len < sizeof(int))
1304 			return -EINVAL;
1305 		len = sizeof(int);
1306 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1307 		if (put_user(len, optlen) ||
1308 		    put_user(val, optval))
1309 			return -EFAULT;
1310 		err = 0;
1311 		break;
1312 	case NETLINK_BROADCAST_ERROR:
1313 		if (len < sizeof(int))
1314 			return -EINVAL;
1315 		len = sizeof(int);
1316 		val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1317 		if (put_user(len, optlen) ||
1318 		    put_user(val, optval))
1319 			return -EFAULT;
1320 		err = 0;
1321 		break;
1322 	case NETLINK_NO_ENOBUFS:
1323 		if (len < sizeof(int))
1324 			return -EINVAL;
1325 		len = sizeof(int);
1326 		val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1327 		if (put_user(len, optlen) ||
1328 		    put_user(val, optval))
1329 			return -EFAULT;
1330 		err = 0;
1331 		break;
1332 	default:
1333 		err = -ENOPROTOOPT;
1334 	}
1335 	return err;
1336 }
1337 
1338 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1339 {
1340 	struct nl_pktinfo info;
1341 
1342 	info.group = NETLINK_CB(skb).dst_group;
1343 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1344 }
1345 
1346 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1347 			   struct msghdr *msg, size_t len)
1348 {
1349 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1350 	struct sock *sk = sock->sk;
1351 	struct netlink_sock *nlk = nlk_sk(sk);
1352 	struct sockaddr_nl *addr = msg->msg_name;
1353 	u32 dst_pid;
1354 	u32 dst_group;
1355 	struct sk_buff *skb;
1356 	int err;
1357 	struct scm_cookie scm;
1358 
1359 	if (msg->msg_flags&MSG_OOB)
1360 		return -EOPNOTSUPP;
1361 
1362 	if (NULL == siocb->scm)
1363 		siocb->scm = &scm;
1364 
1365 	err = scm_send(sock, msg, siocb->scm, true);
1366 	if (err < 0)
1367 		return err;
1368 
1369 	if (msg->msg_namelen) {
1370 		err = -EINVAL;
1371 		if (addr->nl_family != AF_NETLINK)
1372 			goto out;
1373 		dst_pid = addr->nl_pid;
1374 		dst_group = ffs(addr->nl_groups);
1375 		err =  -EPERM;
1376 		if ((dst_group || dst_pid) &&
1377 		    !netlink_capable(sock, NL_NONROOT_SEND))
1378 			goto out;
1379 	} else {
1380 		dst_pid = nlk->dst_pid;
1381 		dst_group = nlk->dst_group;
1382 	}
1383 
1384 	if (!nlk->pid) {
1385 		err = netlink_autobind(sock);
1386 		if (err)
1387 			goto out;
1388 	}
1389 
1390 	err = -EMSGSIZE;
1391 	if (len > sk->sk_sndbuf - 32)
1392 		goto out;
1393 	err = -ENOBUFS;
1394 	skb = alloc_skb(len, GFP_KERNEL);
1395 	if (skb == NULL)
1396 		goto out;
1397 
1398 	NETLINK_CB(skb).pid	= nlk->pid;
1399 	NETLINK_CB(skb).dst_group = dst_group;
1400 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1401 
1402 	err = -EFAULT;
1403 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1404 		kfree_skb(skb);
1405 		goto out;
1406 	}
1407 
1408 	err = security_netlink_send(sk, skb);
1409 	if (err) {
1410 		kfree_skb(skb);
1411 		goto out;
1412 	}
1413 
1414 	if (dst_group) {
1415 		atomic_inc(&skb->users);
1416 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1417 	}
1418 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1419 
1420 out:
1421 	scm_destroy(siocb->scm);
1422 	return err;
1423 }
1424 
1425 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1426 			   struct msghdr *msg, size_t len,
1427 			   int flags)
1428 {
1429 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1430 	struct scm_cookie scm;
1431 	struct sock *sk = sock->sk;
1432 	struct netlink_sock *nlk = nlk_sk(sk);
1433 	int noblock = flags&MSG_DONTWAIT;
1434 	size_t copied;
1435 	struct sk_buff *skb, *data_skb;
1436 	int err, ret;
1437 
1438 	if (flags&MSG_OOB)
1439 		return -EOPNOTSUPP;
1440 
1441 	copied = 0;
1442 
1443 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1444 	if (skb == NULL)
1445 		goto out;
1446 
1447 	data_skb = skb;
1448 
1449 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1450 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1451 		/*
1452 		 * If this skb has a frag_list, then here that means that we
1453 		 * will have to use the frag_list skb's data for compat tasks
1454 		 * and the regular skb's data for normal (non-compat) tasks.
1455 		 *
1456 		 * If we need to send the compat skb, assign it to the
1457 		 * 'data_skb' variable so that it will be used below for data
1458 		 * copying. We keep 'skb' for everything else, including
1459 		 * freeing both later.
1460 		 */
1461 		if (flags & MSG_CMSG_COMPAT)
1462 			data_skb = skb_shinfo(skb)->frag_list;
1463 	}
1464 #endif
1465 
1466 	msg->msg_namelen = 0;
1467 
1468 	copied = data_skb->len;
1469 	if (len < copied) {
1470 		msg->msg_flags |= MSG_TRUNC;
1471 		copied = len;
1472 	}
1473 
1474 	skb_reset_transport_header(data_skb);
1475 	err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1476 
1477 	if (msg->msg_name) {
1478 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1479 		addr->nl_family = AF_NETLINK;
1480 		addr->nl_pad    = 0;
1481 		addr->nl_pid	= NETLINK_CB(skb).pid;
1482 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1483 		msg->msg_namelen = sizeof(*addr);
1484 	}
1485 
1486 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1487 		netlink_cmsg_recv_pktinfo(msg, skb);
1488 
1489 	if (NULL == siocb->scm) {
1490 		memset(&scm, 0, sizeof(scm));
1491 		siocb->scm = &scm;
1492 	}
1493 	siocb->scm->creds = *NETLINK_CREDS(skb);
1494 	if (flags & MSG_TRUNC)
1495 		copied = data_skb->len;
1496 
1497 	skb_free_datagram(sk, skb);
1498 
1499 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1500 		ret = netlink_dump(sk);
1501 		if (ret) {
1502 			sk->sk_err = ret;
1503 			sk->sk_error_report(sk);
1504 		}
1505 	}
1506 
1507 	scm_recv(sock, msg, siocb->scm, flags);
1508 out:
1509 	netlink_rcv_wake(sk);
1510 	return err ? : copied;
1511 }
1512 
1513 static void netlink_data_ready(struct sock *sk, int len)
1514 {
1515 	BUG();
1516 }
1517 
1518 /*
1519  *	We export these functions to other modules. They provide a
1520  *	complete set of kernel non-blocking support for message
1521  *	queueing.
1522  */
1523 
1524 struct sock *
1525 netlink_kernel_create(struct net *net, int unit,
1526 		      struct module *module,
1527 		      struct netlink_kernel_cfg *cfg)
1528 {
1529 	struct socket *sock;
1530 	struct sock *sk;
1531 	struct netlink_sock *nlk;
1532 	struct listeners *listeners = NULL;
1533 	struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1534 	unsigned int groups;
1535 
1536 	BUG_ON(!nl_table);
1537 
1538 	if (unit < 0 || unit >= MAX_LINKS)
1539 		return NULL;
1540 
1541 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1542 		return NULL;
1543 
1544 	/*
1545 	 * We have to just have a reference on the net from sk, but don't
1546 	 * get_net it. Besides, we cannot get and then put the net here.
1547 	 * So we create one inside init_net and the move it to net.
1548 	 */
1549 
1550 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1551 		goto out_sock_release_nosk;
1552 
1553 	sk = sock->sk;
1554 	sk_change_net(sk, net);
1555 
1556 	if (!cfg || cfg->groups < 32)
1557 		groups = 32;
1558 	else
1559 		groups = cfg->groups;
1560 
1561 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1562 	if (!listeners)
1563 		goto out_sock_release;
1564 
1565 	sk->sk_data_ready = netlink_data_ready;
1566 	if (cfg && cfg->input)
1567 		nlk_sk(sk)->netlink_rcv = cfg->input;
1568 
1569 	if (netlink_insert(sk, net, 0))
1570 		goto out_sock_release;
1571 
1572 	nlk = nlk_sk(sk);
1573 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1574 
1575 	netlink_table_grab();
1576 	if (!nl_table[unit].registered) {
1577 		nl_table[unit].groups = groups;
1578 		rcu_assign_pointer(nl_table[unit].listeners, listeners);
1579 		nl_table[unit].cb_mutex = cb_mutex;
1580 		nl_table[unit].module = module;
1581 		nl_table[unit].bind = cfg ? cfg->bind : NULL;
1582 		nl_table[unit].registered = 1;
1583 	} else {
1584 		kfree(listeners);
1585 		nl_table[unit].registered++;
1586 	}
1587 	netlink_table_ungrab();
1588 	return sk;
1589 
1590 out_sock_release:
1591 	kfree(listeners);
1592 	netlink_kernel_release(sk);
1593 	return NULL;
1594 
1595 out_sock_release_nosk:
1596 	sock_release(sock);
1597 	return NULL;
1598 }
1599 EXPORT_SYMBOL(netlink_kernel_create);
1600 
1601 
1602 void
1603 netlink_kernel_release(struct sock *sk)
1604 {
1605 	sk_release_kernel(sk);
1606 }
1607 EXPORT_SYMBOL(netlink_kernel_release);
1608 
1609 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1610 {
1611 	struct listeners *new, *old;
1612 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1613 
1614 	if (groups < 32)
1615 		groups = 32;
1616 
1617 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1618 		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1619 		if (!new)
1620 			return -ENOMEM;
1621 		old = rcu_dereference_protected(tbl->listeners, 1);
1622 		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1623 		rcu_assign_pointer(tbl->listeners, new);
1624 
1625 		kfree_rcu(old, rcu);
1626 	}
1627 	tbl->groups = groups;
1628 
1629 	return 0;
1630 }
1631 
1632 /**
1633  * netlink_change_ngroups - change number of multicast groups
1634  *
1635  * This changes the number of multicast groups that are available
1636  * on a certain netlink family. Note that it is not possible to
1637  * change the number of groups to below 32. Also note that it does
1638  * not implicitly call netlink_clear_multicast_users() when the
1639  * number of groups is reduced.
1640  *
1641  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1642  * @groups: The new number of groups.
1643  */
1644 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1645 {
1646 	int err;
1647 
1648 	netlink_table_grab();
1649 	err = __netlink_change_ngroups(sk, groups);
1650 	netlink_table_ungrab();
1651 
1652 	return err;
1653 }
1654 
1655 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1656 {
1657 	struct sock *sk;
1658 	struct hlist_node *node;
1659 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1660 
1661 	sk_for_each_bound(sk, node, &tbl->mc_list)
1662 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1663 }
1664 
1665 /**
1666  * netlink_clear_multicast_users - kick off multicast listeners
1667  *
1668  * This function removes all listeners from the given group.
1669  * @ksk: The kernel netlink socket, as returned by
1670  *	netlink_kernel_create().
1671  * @group: The multicast group to clear.
1672  */
1673 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1674 {
1675 	netlink_table_grab();
1676 	__netlink_clear_multicast_users(ksk, group);
1677 	netlink_table_ungrab();
1678 }
1679 
1680 void netlink_set_nonroot(int protocol, unsigned int flags)
1681 {
1682 	if ((unsigned int)protocol < MAX_LINKS)
1683 		nl_table[protocol].nl_nonroot = flags;
1684 }
1685 EXPORT_SYMBOL(netlink_set_nonroot);
1686 
1687 struct nlmsghdr *
1688 __nlmsg_put(struct sk_buff *skb, u32 pid, u32 seq, int type, int len, int flags)
1689 {
1690 	struct nlmsghdr *nlh;
1691 	int size = NLMSG_LENGTH(len);
1692 
1693 	nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
1694 	nlh->nlmsg_type = type;
1695 	nlh->nlmsg_len = size;
1696 	nlh->nlmsg_flags = flags;
1697 	nlh->nlmsg_pid = pid;
1698 	nlh->nlmsg_seq = seq;
1699 	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
1700 		memset(NLMSG_DATA(nlh) + len, 0, NLMSG_ALIGN(size) - size);
1701 	return nlh;
1702 }
1703 EXPORT_SYMBOL(__nlmsg_put);
1704 
1705 /*
1706  * It looks a bit ugly.
1707  * It would be better to create kernel thread.
1708  */
1709 
1710 static int netlink_dump(struct sock *sk)
1711 {
1712 	struct netlink_sock *nlk = nlk_sk(sk);
1713 	struct netlink_callback *cb;
1714 	struct sk_buff *skb = NULL;
1715 	struct nlmsghdr *nlh;
1716 	int len, err = -ENOBUFS;
1717 	int alloc_size;
1718 
1719 	mutex_lock(nlk->cb_mutex);
1720 
1721 	cb = nlk->cb;
1722 	if (cb == NULL) {
1723 		err = -EINVAL;
1724 		goto errout_skb;
1725 	}
1726 
1727 	alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
1728 
1729 	skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
1730 	if (!skb)
1731 		goto errout_skb;
1732 
1733 	len = cb->dump(skb, cb);
1734 
1735 	if (len > 0) {
1736 		mutex_unlock(nlk->cb_mutex);
1737 
1738 		if (sk_filter(sk, skb))
1739 			kfree_skb(skb);
1740 		else
1741 			__netlink_sendskb(sk, skb);
1742 		return 0;
1743 	}
1744 
1745 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1746 	if (!nlh)
1747 		goto errout_skb;
1748 
1749 	nl_dump_check_consistent(cb, nlh);
1750 
1751 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1752 
1753 	if (sk_filter(sk, skb))
1754 		kfree_skb(skb);
1755 	else
1756 		__netlink_sendskb(sk, skb);
1757 
1758 	if (cb->done)
1759 		cb->done(cb);
1760 	nlk->cb = NULL;
1761 	mutex_unlock(nlk->cb_mutex);
1762 
1763 	netlink_consume_callback(cb);
1764 	return 0;
1765 
1766 errout_skb:
1767 	mutex_unlock(nlk->cb_mutex);
1768 	kfree_skb(skb);
1769 	return err;
1770 }
1771 
1772 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1773 		       const struct nlmsghdr *nlh,
1774 		       struct netlink_dump_control *control)
1775 {
1776 	struct netlink_callback *cb;
1777 	struct sock *sk;
1778 	struct netlink_sock *nlk;
1779 	int ret;
1780 
1781 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1782 	if (cb == NULL)
1783 		return -ENOBUFS;
1784 
1785 	cb->dump = control->dump;
1786 	cb->done = control->done;
1787 	cb->nlh = nlh;
1788 	cb->data = control->data;
1789 	cb->min_dump_alloc = control->min_dump_alloc;
1790 	atomic_inc(&skb->users);
1791 	cb->skb = skb;
1792 
1793 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1794 	if (sk == NULL) {
1795 		netlink_destroy_callback(cb);
1796 		return -ECONNREFUSED;
1797 	}
1798 	nlk = nlk_sk(sk);
1799 	/* A dump is in progress... */
1800 	mutex_lock(nlk->cb_mutex);
1801 	if (nlk->cb) {
1802 		mutex_unlock(nlk->cb_mutex);
1803 		netlink_destroy_callback(cb);
1804 		sock_put(sk);
1805 		return -EBUSY;
1806 	}
1807 	nlk->cb = cb;
1808 	mutex_unlock(nlk->cb_mutex);
1809 
1810 	ret = netlink_dump(sk);
1811 
1812 	sock_put(sk);
1813 
1814 	if (ret)
1815 		return ret;
1816 
1817 	/* We successfully started a dump, by returning -EINTR we
1818 	 * signal not to send ACK even if it was requested.
1819 	 */
1820 	return -EINTR;
1821 }
1822 EXPORT_SYMBOL(netlink_dump_start);
1823 
1824 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1825 {
1826 	struct sk_buff *skb;
1827 	struct nlmsghdr *rep;
1828 	struct nlmsgerr *errmsg;
1829 	size_t payload = sizeof(*errmsg);
1830 
1831 	/* error messages get the original request appened */
1832 	if (err)
1833 		payload += nlmsg_len(nlh);
1834 
1835 	skb = nlmsg_new(payload, GFP_KERNEL);
1836 	if (!skb) {
1837 		struct sock *sk;
1838 
1839 		sk = netlink_lookup(sock_net(in_skb->sk),
1840 				    in_skb->sk->sk_protocol,
1841 				    NETLINK_CB(in_skb).pid);
1842 		if (sk) {
1843 			sk->sk_err = ENOBUFS;
1844 			sk->sk_error_report(sk);
1845 			sock_put(sk);
1846 		}
1847 		return;
1848 	}
1849 
1850 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1851 			  NLMSG_ERROR, payload, 0);
1852 	errmsg = nlmsg_data(rep);
1853 	errmsg->error = err;
1854 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1855 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1856 }
1857 EXPORT_SYMBOL(netlink_ack);
1858 
1859 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1860 						     struct nlmsghdr *))
1861 {
1862 	struct nlmsghdr *nlh;
1863 	int err;
1864 
1865 	while (skb->len >= nlmsg_total_size(0)) {
1866 		int msglen;
1867 
1868 		nlh = nlmsg_hdr(skb);
1869 		err = 0;
1870 
1871 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1872 			return 0;
1873 
1874 		/* Only requests are handled by the kernel */
1875 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1876 			goto ack;
1877 
1878 		/* Skip control messages */
1879 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1880 			goto ack;
1881 
1882 		err = cb(skb, nlh);
1883 		if (err == -EINTR)
1884 			goto skip;
1885 
1886 ack:
1887 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1888 			netlink_ack(skb, nlh, err);
1889 
1890 skip:
1891 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1892 		if (msglen > skb->len)
1893 			msglen = skb->len;
1894 		skb_pull(skb, msglen);
1895 	}
1896 
1897 	return 0;
1898 }
1899 EXPORT_SYMBOL(netlink_rcv_skb);
1900 
1901 /**
1902  * nlmsg_notify - send a notification netlink message
1903  * @sk: netlink socket to use
1904  * @skb: notification message
1905  * @pid: destination netlink pid for reports or 0
1906  * @group: destination multicast group or 0
1907  * @report: 1 to report back, 0 to disable
1908  * @flags: allocation flags
1909  */
1910 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1911 		 unsigned int group, int report, gfp_t flags)
1912 {
1913 	int err = 0;
1914 
1915 	if (group) {
1916 		int exclude_pid = 0;
1917 
1918 		if (report) {
1919 			atomic_inc(&skb->users);
1920 			exclude_pid = pid;
1921 		}
1922 
1923 		/* errors reported via destination sk->sk_err, but propagate
1924 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1925 		err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1926 	}
1927 
1928 	if (report) {
1929 		int err2;
1930 
1931 		err2 = nlmsg_unicast(sk, skb, pid);
1932 		if (!err || err == -ESRCH)
1933 			err = err2;
1934 	}
1935 
1936 	return err;
1937 }
1938 EXPORT_SYMBOL(nlmsg_notify);
1939 
1940 #ifdef CONFIG_PROC_FS
1941 struct nl_seq_iter {
1942 	struct seq_net_private p;
1943 	int link;
1944 	int hash_idx;
1945 };
1946 
1947 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1948 {
1949 	struct nl_seq_iter *iter = seq->private;
1950 	int i, j;
1951 	struct sock *s;
1952 	struct hlist_node *node;
1953 	loff_t off = 0;
1954 
1955 	for (i = 0; i < MAX_LINKS; i++) {
1956 		struct nl_pid_hash *hash = &nl_table[i].hash;
1957 
1958 		for (j = 0; j <= hash->mask; j++) {
1959 			sk_for_each(s, node, &hash->table[j]) {
1960 				if (sock_net(s) != seq_file_net(seq))
1961 					continue;
1962 				if (off == pos) {
1963 					iter->link = i;
1964 					iter->hash_idx = j;
1965 					return s;
1966 				}
1967 				++off;
1968 			}
1969 		}
1970 	}
1971 	return NULL;
1972 }
1973 
1974 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1975 	__acquires(nl_table_lock)
1976 {
1977 	read_lock(&nl_table_lock);
1978 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1979 }
1980 
1981 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1982 {
1983 	struct sock *s;
1984 	struct nl_seq_iter *iter;
1985 	int i, j;
1986 
1987 	++*pos;
1988 
1989 	if (v == SEQ_START_TOKEN)
1990 		return netlink_seq_socket_idx(seq, 0);
1991 
1992 	iter = seq->private;
1993 	s = v;
1994 	do {
1995 		s = sk_next(s);
1996 	} while (s && sock_net(s) != seq_file_net(seq));
1997 	if (s)
1998 		return s;
1999 
2000 	i = iter->link;
2001 	j = iter->hash_idx + 1;
2002 
2003 	do {
2004 		struct nl_pid_hash *hash = &nl_table[i].hash;
2005 
2006 		for (; j <= hash->mask; j++) {
2007 			s = sk_head(&hash->table[j]);
2008 			while (s && sock_net(s) != seq_file_net(seq))
2009 				s = sk_next(s);
2010 			if (s) {
2011 				iter->link = i;
2012 				iter->hash_idx = j;
2013 				return s;
2014 			}
2015 		}
2016 
2017 		j = 0;
2018 	} while (++i < MAX_LINKS);
2019 
2020 	return NULL;
2021 }
2022 
2023 static void netlink_seq_stop(struct seq_file *seq, void *v)
2024 	__releases(nl_table_lock)
2025 {
2026 	read_unlock(&nl_table_lock);
2027 }
2028 
2029 
2030 static int netlink_seq_show(struct seq_file *seq, void *v)
2031 {
2032 	if (v == SEQ_START_TOKEN) {
2033 		seq_puts(seq,
2034 			 "sk       Eth Pid    Groups   "
2035 			 "Rmem     Wmem     Dump     Locks     Drops     Inode\n");
2036 	} else {
2037 		struct sock *s = v;
2038 		struct netlink_sock *nlk = nlk_sk(s);
2039 
2040 		seq_printf(seq, "%pK %-3d %-6d %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2041 			   s,
2042 			   s->sk_protocol,
2043 			   nlk->pid,
2044 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2045 			   sk_rmem_alloc_get(s),
2046 			   sk_wmem_alloc_get(s),
2047 			   nlk->cb,
2048 			   atomic_read(&s->sk_refcnt),
2049 			   atomic_read(&s->sk_drops),
2050 			   sock_i_ino(s)
2051 			);
2052 
2053 	}
2054 	return 0;
2055 }
2056 
2057 static const struct seq_operations netlink_seq_ops = {
2058 	.start  = netlink_seq_start,
2059 	.next   = netlink_seq_next,
2060 	.stop   = netlink_seq_stop,
2061 	.show   = netlink_seq_show,
2062 };
2063 
2064 
2065 static int netlink_seq_open(struct inode *inode, struct file *file)
2066 {
2067 	return seq_open_net(inode, file, &netlink_seq_ops,
2068 				sizeof(struct nl_seq_iter));
2069 }
2070 
2071 static const struct file_operations netlink_seq_fops = {
2072 	.owner		= THIS_MODULE,
2073 	.open		= netlink_seq_open,
2074 	.read		= seq_read,
2075 	.llseek		= seq_lseek,
2076 	.release	= seq_release_net,
2077 };
2078 
2079 #endif
2080 
2081 int netlink_register_notifier(struct notifier_block *nb)
2082 {
2083 	return atomic_notifier_chain_register(&netlink_chain, nb);
2084 }
2085 EXPORT_SYMBOL(netlink_register_notifier);
2086 
2087 int netlink_unregister_notifier(struct notifier_block *nb)
2088 {
2089 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
2090 }
2091 EXPORT_SYMBOL(netlink_unregister_notifier);
2092 
2093 static const struct proto_ops netlink_ops = {
2094 	.family =	PF_NETLINK,
2095 	.owner =	THIS_MODULE,
2096 	.release =	netlink_release,
2097 	.bind =		netlink_bind,
2098 	.connect =	netlink_connect,
2099 	.socketpair =	sock_no_socketpair,
2100 	.accept =	sock_no_accept,
2101 	.getname =	netlink_getname,
2102 	.poll =		datagram_poll,
2103 	.ioctl =	sock_no_ioctl,
2104 	.listen =	sock_no_listen,
2105 	.shutdown =	sock_no_shutdown,
2106 	.setsockopt =	netlink_setsockopt,
2107 	.getsockopt =	netlink_getsockopt,
2108 	.sendmsg =	netlink_sendmsg,
2109 	.recvmsg =	netlink_recvmsg,
2110 	.mmap =		sock_no_mmap,
2111 	.sendpage =	sock_no_sendpage,
2112 };
2113 
2114 static const struct net_proto_family netlink_family_ops = {
2115 	.family = PF_NETLINK,
2116 	.create = netlink_create,
2117 	.owner	= THIS_MODULE,	/* for consistency 8) */
2118 };
2119 
2120 static int __net_init netlink_net_init(struct net *net)
2121 {
2122 #ifdef CONFIG_PROC_FS
2123 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2124 		return -ENOMEM;
2125 #endif
2126 	return 0;
2127 }
2128 
2129 static void __net_exit netlink_net_exit(struct net *net)
2130 {
2131 #ifdef CONFIG_PROC_FS
2132 	proc_net_remove(net, "netlink");
2133 #endif
2134 }
2135 
2136 static void __init netlink_add_usersock_entry(void)
2137 {
2138 	struct listeners *listeners;
2139 	int groups = 32;
2140 
2141 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2142 	if (!listeners)
2143 		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2144 
2145 	netlink_table_grab();
2146 
2147 	nl_table[NETLINK_USERSOCK].groups = groups;
2148 	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2149 	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2150 	nl_table[NETLINK_USERSOCK].registered = 1;
2151 	nl_table[NETLINK_USERSOCK].nl_nonroot = NL_NONROOT_SEND;
2152 
2153 	netlink_table_ungrab();
2154 }
2155 
2156 static struct pernet_operations __net_initdata netlink_net_ops = {
2157 	.init = netlink_net_init,
2158 	.exit = netlink_net_exit,
2159 };
2160 
2161 static int __init netlink_proto_init(void)
2162 {
2163 	struct sk_buff *dummy_skb;
2164 	int i;
2165 	unsigned long limit;
2166 	unsigned int order;
2167 	int err = proto_register(&netlink_proto, 0);
2168 
2169 	if (err != 0)
2170 		goto out;
2171 
2172 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2173 
2174 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2175 	if (!nl_table)
2176 		goto panic;
2177 
2178 	if (totalram_pages >= (128 * 1024))
2179 		limit = totalram_pages >> (21 - PAGE_SHIFT);
2180 	else
2181 		limit = totalram_pages >> (23 - PAGE_SHIFT);
2182 
2183 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2184 	limit = (1UL << order) / sizeof(struct hlist_head);
2185 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2186 
2187 	for (i = 0; i < MAX_LINKS; i++) {
2188 		struct nl_pid_hash *hash = &nl_table[i].hash;
2189 
2190 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2191 		if (!hash->table) {
2192 			while (i-- > 0)
2193 				nl_pid_hash_free(nl_table[i].hash.table,
2194 						 1 * sizeof(*hash->table));
2195 			kfree(nl_table);
2196 			goto panic;
2197 		}
2198 		hash->max_shift = order;
2199 		hash->shift = 0;
2200 		hash->mask = 0;
2201 		hash->rehash_time = jiffies;
2202 	}
2203 
2204 	netlink_add_usersock_entry();
2205 
2206 	sock_register(&netlink_family_ops);
2207 	register_pernet_subsys(&netlink_net_ops);
2208 	/* The netlink device handler may be needed early. */
2209 	rtnetlink_init();
2210 out:
2211 	return err;
2212 panic:
2213 	panic("netlink_init: Cannot allocate nl_table\n");
2214 }
2215 
2216 core_initcall(netlink_proto_init);
2217