xref: /linux/net/netlink/af_netlink.c (revision 9f6d3c4b76314c40c866a935d78c80fd284768bd)
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 	struct module		*module;
84 };
85 
86 struct listeners_rcu_head {
87 	struct rcu_head rcu_head;
88 	void *ptr;
89 };
90 
91 #define NETLINK_KERNEL_SOCKET	0x1
92 #define NETLINK_RECV_PKTINFO	0x2
93 #define NETLINK_BROADCAST_SEND_ERROR	0x4
94 #define NETLINK_RECV_NO_ENOBUFS	0x8
95 
96 static inline struct netlink_sock *nlk_sk(struct sock *sk)
97 {
98 	return container_of(sk, struct netlink_sock, sk);
99 }
100 
101 static inline int netlink_is_kernel(struct sock *sk)
102 {
103 	return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
104 }
105 
106 struct nl_pid_hash {
107 	struct hlist_head *table;
108 	unsigned long rehash_time;
109 
110 	unsigned int mask;
111 	unsigned int shift;
112 
113 	unsigned int entries;
114 	unsigned int max_shift;
115 
116 	u32 rnd;
117 };
118 
119 struct netlink_table {
120 	struct nl_pid_hash hash;
121 	struct hlist_head mc_list;
122 	unsigned long *listeners;
123 	unsigned int nl_nonroot;
124 	unsigned int groups;
125 	struct mutex *cb_mutex;
126 	struct module *module;
127 	int registered;
128 };
129 
130 static struct netlink_table *nl_table;
131 
132 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
133 
134 static int netlink_dump(struct sock *sk);
135 static void netlink_destroy_callback(struct netlink_callback *cb);
136 
137 static DEFINE_RWLOCK(nl_table_lock);
138 static atomic_t nl_table_users = ATOMIC_INIT(0);
139 
140 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
141 
142 static u32 netlink_group_mask(u32 group)
143 {
144 	return group ? 1 << (group - 1) : 0;
145 }
146 
147 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
148 {
149 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
150 }
151 
152 static void netlink_sock_destruct(struct sock *sk)
153 {
154 	struct netlink_sock *nlk = nlk_sk(sk);
155 
156 	if (nlk->cb) {
157 		if (nlk->cb->done)
158 			nlk->cb->done(nlk->cb);
159 		netlink_destroy_callback(nlk->cb);
160 	}
161 
162 	skb_queue_purge(&sk->sk_receive_queue);
163 
164 	if (!sock_flag(sk, SOCK_DEAD)) {
165 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
166 		return;
167 	}
168 
169 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
170 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
171 	WARN_ON(nlk_sk(sk)->groups);
172 }
173 
174 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
175  * SMP. Look, when several writers sleep and reader wakes them up, all but one
176  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
177  * this, _but_ remember, it adds useless work on UP machines.
178  */
179 
180 void netlink_table_grab(void)
181 	__acquires(nl_table_lock)
182 {
183 	might_sleep();
184 
185 	write_lock_irq(&nl_table_lock);
186 
187 	if (atomic_read(&nl_table_users)) {
188 		DECLARE_WAITQUEUE(wait, current);
189 
190 		add_wait_queue_exclusive(&nl_table_wait, &wait);
191 		for (;;) {
192 			set_current_state(TASK_UNINTERRUPTIBLE);
193 			if (atomic_read(&nl_table_users) == 0)
194 				break;
195 			write_unlock_irq(&nl_table_lock);
196 			schedule();
197 			write_lock_irq(&nl_table_lock);
198 		}
199 
200 		__set_current_state(TASK_RUNNING);
201 		remove_wait_queue(&nl_table_wait, &wait);
202 	}
203 }
204 
205 void netlink_table_ungrab(void)
206 	__releases(nl_table_lock)
207 {
208 	write_unlock_irq(&nl_table_lock);
209 	wake_up(&nl_table_wait);
210 }
211 
212 static inline void
213 netlink_lock_table(void)
214 {
215 	/* read_lock() synchronizes us to netlink_table_grab */
216 
217 	read_lock(&nl_table_lock);
218 	atomic_inc(&nl_table_users);
219 	read_unlock(&nl_table_lock);
220 }
221 
222 static inline void
223 netlink_unlock_table(void)
224 {
225 	if (atomic_dec_and_test(&nl_table_users))
226 		wake_up(&nl_table_wait);
227 }
228 
229 static inline struct sock *netlink_lookup(struct net *net, int protocol,
230 					  u32 pid)
231 {
232 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
233 	struct hlist_head *head;
234 	struct sock *sk;
235 	struct hlist_node *node;
236 
237 	read_lock(&nl_table_lock);
238 	head = nl_pid_hashfn(hash, pid);
239 	sk_for_each(sk, node, head) {
240 		if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
241 			sock_hold(sk);
242 			goto found;
243 		}
244 	}
245 	sk = NULL;
246 found:
247 	read_unlock(&nl_table_lock);
248 	return sk;
249 }
250 
251 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
252 {
253 	if (size <= PAGE_SIZE)
254 		return kzalloc(size, GFP_ATOMIC);
255 	else
256 		return (struct hlist_head *)
257 			__get_free_pages(GFP_ATOMIC | __GFP_ZERO,
258 					 get_order(size));
259 }
260 
261 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
262 {
263 	if (size <= PAGE_SIZE)
264 		kfree(table);
265 	else
266 		free_pages((unsigned long)table, get_order(size));
267 }
268 
269 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
270 {
271 	unsigned int omask, mask, shift;
272 	size_t osize, size;
273 	struct hlist_head *otable, *table;
274 	int i;
275 
276 	omask = mask = hash->mask;
277 	osize = size = (mask + 1) * sizeof(*table);
278 	shift = hash->shift;
279 
280 	if (grow) {
281 		if (++shift > hash->max_shift)
282 			return 0;
283 		mask = mask * 2 + 1;
284 		size *= 2;
285 	}
286 
287 	table = nl_pid_hash_zalloc(size);
288 	if (!table)
289 		return 0;
290 
291 	otable = hash->table;
292 	hash->table = table;
293 	hash->mask = mask;
294 	hash->shift = shift;
295 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
296 
297 	for (i = 0; i <= omask; i++) {
298 		struct sock *sk;
299 		struct hlist_node *node, *tmp;
300 
301 		sk_for_each_safe(sk, node, tmp, &otable[i])
302 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
303 	}
304 
305 	nl_pid_hash_free(otable, osize);
306 	hash->rehash_time = jiffies + 10 * 60 * HZ;
307 	return 1;
308 }
309 
310 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
311 {
312 	int avg = hash->entries >> hash->shift;
313 
314 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
315 		return 1;
316 
317 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
318 		nl_pid_hash_rehash(hash, 0);
319 		return 1;
320 	}
321 
322 	return 0;
323 }
324 
325 static const struct proto_ops netlink_ops;
326 
327 static void
328 netlink_update_listeners(struct sock *sk)
329 {
330 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
331 	struct hlist_node *node;
332 	unsigned long mask;
333 	unsigned int i;
334 
335 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
336 		mask = 0;
337 		sk_for_each_bound(sk, node, &tbl->mc_list) {
338 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
339 				mask |= nlk_sk(sk)->groups[i];
340 		}
341 		tbl->listeners[i] = mask;
342 	}
343 	/* this function is only called with the netlink table "grabbed", which
344 	 * makes sure updates are visible before bind or setsockopt return. */
345 }
346 
347 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
348 {
349 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
350 	struct hlist_head *head;
351 	int err = -EADDRINUSE;
352 	struct sock *osk;
353 	struct hlist_node *node;
354 	int len;
355 
356 	netlink_table_grab();
357 	head = nl_pid_hashfn(hash, pid);
358 	len = 0;
359 	sk_for_each(osk, node, head) {
360 		if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
361 			break;
362 		len++;
363 	}
364 	if (node)
365 		goto err;
366 
367 	err = -EBUSY;
368 	if (nlk_sk(sk)->pid)
369 		goto err;
370 
371 	err = -ENOMEM;
372 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
373 		goto err;
374 
375 	if (len && nl_pid_hash_dilute(hash, len))
376 		head = nl_pid_hashfn(hash, pid);
377 	hash->entries++;
378 	nlk_sk(sk)->pid = pid;
379 	sk_add_node(sk, head);
380 	err = 0;
381 
382 err:
383 	netlink_table_ungrab();
384 	return err;
385 }
386 
387 static void netlink_remove(struct sock *sk)
388 {
389 	netlink_table_grab();
390 	if (sk_del_node_init(sk))
391 		nl_table[sk->sk_protocol].hash.entries--;
392 	if (nlk_sk(sk)->subscriptions)
393 		__sk_del_bind_node(sk);
394 	netlink_table_ungrab();
395 }
396 
397 static struct proto netlink_proto = {
398 	.name	  = "NETLINK",
399 	.owner	  = THIS_MODULE,
400 	.obj_size = sizeof(struct netlink_sock),
401 };
402 
403 static int __netlink_create(struct net *net, struct socket *sock,
404 			    struct mutex *cb_mutex, int protocol)
405 {
406 	struct sock *sk;
407 	struct netlink_sock *nlk;
408 
409 	sock->ops = &netlink_ops;
410 
411 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
412 	if (!sk)
413 		return -ENOMEM;
414 
415 	sock_init_data(sock, sk);
416 
417 	nlk = nlk_sk(sk);
418 	if (cb_mutex)
419 		nlk->cb_mutex = cb_mutex;
420 	else {
421 		nlk->cb_mutex = &nlk->cb_def_mutex;
422 		mutex_init(nlk->cb_mutex);
423 	}
424 	init_waitqueue_head(&nlk->wait);
425 
426 	sk->sk_destruct = netlink_sock_destruct;
427 	sk->sk_protocol = protocol;
428 	return 0;
429 }
430 
431 static int netlink_create(struct net *net, struct socket *sock, int protocol,
432 			  int kern)
433 {
434 	struct module *module = NULL;
435 	struct mutex *cb_mutex;
436 	struct netlink_sock *nlk;
437 	int err = 0;
438 
439 	sock->state = SS_UNCONNECTED;
440 
441 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
442 		return -ESOCKTNOSUPPORT;
443 
444 	if (protocol < 0 || protocol >= MAX_LINKS)
445 		return -EPROTONOSUPPORT;
446 
447 	netlink_lock_table();
448 #ifdef CONFIG_MODULES
449 	if (!nl_table[protocol].registered) {
450 		netlink_unlock_table();
451 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
452 		netlink_lock_table();
453 	}
454 #endif
455 	if (nl_table[protocol].registered &&
456 	    try_module_get(nl_table[protocol].module))
457 		module = nl_table[protocol].module;
458 	else
459 		err = -EPROTONOSUPPORT;
460 	cb_mutex = nl_table[protocol].cb_mutex;
461 	netlink_unlock_table();
462 
463 	if (err < 0)
464 		goto out;
465 
466 	err = __netlink_create(net, sock, cb_mutex, protocol);
467 	if (err < 0)
468 		goto out_module;
469 
470 	local_bh_disable();
471 	sock_prot_inuse_add(net, &netlink_proto, 1);
472 	local_bh_enable();
473 
474 	nlk = nlk_sk(sock->sk);
475 	nlk->module = module;
476 out:
477 	return err;
478 
479 out_module:
480 	module_put(module);
481 	goto out;
482 }
483 
484 static int netlink_release(struct socket *sock)
485 {
486 	struct sock *sk = sock->sk;
487 	struct netlink_sock *nlk;
488 
489 	if (!sk)
490 		return 0;
491 
492 	netlink_remove(sk);
493 	sock_orphan(sk);
494 	nlk = nlk_sk(sk);
495 
496 	/*
497 	 * OK. Socket is unlinked, any packets that arrive now
498 	 * will be purged.
499 	 */
500 
501 	sock->sk = NULL;
502 	wake_up_interruptible_all(&nlk->wait);
503 
504 	skb_queue_purge(&sk->sk_write_queue);
505 
506 	if (nlk->pid) {
507 		struct netlink_notify n = {
508 						.net = sock_net(sk),
509 						.protocol = sk->sk_protocol,
510 						.pid = nlk->pid,
511 					  };
512 		atomic_notifier_call_chain(&netlink_chain,
513 				NETLINK_URELEASE, &n);
514 	}
515 
516 	module_put(nlk->module);
517 
518 	netlink_table_grab();
519 	if (netlink_is_kernel(sk)) {
520 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
521 		if (--nl_table[sk->sk_protocol].registered == 0) {
522 			kfree(nl_table[sk->sk_protocol].listeners);
523 			nl_table[sk->sk_protocol].module = NULL;
524 			nl_table[sk->sk_protocol].registered = 0;
525 		}
526 	} else if (nlk->subscriptions)
527 		netlink_update_listeners(sk);
528 	netlink_table_ungrab();
529 
530 	kfree(nlk->groups);
531 	nlk->groups = NULL;
532 
533 	local_bh_disable();
534 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
535 	local_bh_enable();
536 	sock_put(sk);
537 	return 0;
538 }
539 
540 static int netlink_autobind(struct socket *sock)
541 {
542 	struct sock *sk = sock->sk;
543 	struct net *net = sock_net(sk);
544 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
545 	struct hlist_head *head;
546 	struct sock *osk;
547 	struct hlist_node *node;
548 	s32 pid = current->tgid;
549 	int err;
550 	static s32 rover = -4097;
551 
552 retry:
553 	cond_resched();
554 	netlink_table_grab();
555 	head = nl_pid_hashfn(hash, pid);
556 	sk_for_each(osk, node, head) {
557 		if (!net_eq(sock_net(osk), net))
558 			continue;
559 		if (nlk_sk(osk)->pid == pid) {
560 			/* Bind collision, search negative pid values. */
561 			pid = rover--;
562 			if (rover > -4097)
563 				rover = -4097;
564 			netlink_table_ungrab();
565 			goto retry;
566 		}
567 	}
568 	netlink_table_ungrab();
569 
570 	err = netlink_insert(sk, net, pid);
571 	if (err == -EADDRINUSE)
572 		goto retry;
573 
574 	/* If 2 threads race to autobind, that is fine.  */
575 	if (err == -EBUSY)
576 		err = 0;
577 
578 	return err;
579 }
580 
581 static inline int netlink_capable(struct socket *sock, unsigned int flag)
582 {
583 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
584 	       capable(CAP_NET_ADMIN);
585 }
586 
587 static void
588 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
589 {
590 	struct netlink_sock *nlk = nlk_sk(sk);
591 
592 	if (nlk->subscriptions && !subscriptions)
593 		__sk_del_bind_node(sk);
594 	else if (!nlk->subscriptions && subscriptions)
595 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
596 	nlk->subscriptions = subscriptions;
597 }
598 
599 static int netlink_realloc_groups(struct sock *sk)
600 {
601 	struct netlink_sock *nlk = nlk_sk(sk);
602 	unsigned int groups;
603 	unsigned long *new_groups;
604 	int err = 0;
605 
606 	netlink_table_grab();
607 
608 	groups = nl_table[sk->sk_protocol].groups;
609 	if (!nl_table[sk->sk_protocol].registered) {
610 		err = -ENOENT;
611 		goto out_unlock;
612 	}
613 
614 	if (nlk->ngroups >= groups)
615 		goto out_unlock;
616 
617 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
618 	if (new_groups == NULL) {
619 		err = -ENOMEM;
620 		goto out_unlock;
621 	}
622 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
623 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
624 
625 	nlk->groups = new_groups;
626 	nlk->ngroups = groups;
627  out_unlock:
628 	netlink_table_ungrab();
629 	return err;
630 }
631 
632 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
633 			int addr_len)
634 {
635 	struct sock *sk = sock->sk;
636 	struct net *net = sock_net(sk);
637 	struct netlink_sock *nlk = nlk_sk(sk);
638 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
639 	int err;
640 
641 	if (nladdr->nl_family != AF_NETLINK)
642 		return -EINVAL;
643 
644 	/* Only superuser is allowed to listen multicasts */
645 	if (nladdr->nl_groups) {
646 		if (!netlink_capable(sock, NL_NONROOT_RECV))
647 			return -EPERM;
648 		err = netlink_realloc_groups(sk);
649 		if (err)
650 			return err;
651 	}
652 
653 	if (nlk->pid) {
654 		if (nladdr->nl_pid != nlk->pid)
655 			return -EINVAL;
656 	} else {
657 		err = nladdr->nl_pid ?
658 			netlink_insert(sk, net, nladdr->nl_pid) :
659 			netlink_autobind(sock);
660 		if (err)
661 			return err;
662 	}
663 
664 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
665 		return 0;
666 
667 	netlink_table_grab();
668 	netlink_update_subscriptions(sk, nlk->subscriptions +
669 					 hweight32(nladdr->nl_groups) -
670 					 hweight32(nlk->groups[0]));
671 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
672 	netlink_update_listeners(sk);
673 	netlink_table_ungrab();
674 
675 	return 0;
676 }
677 
678 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
679 			   int alen, int flags)
680 {
681 	int err = 0;
682 	struct sock *sk = sock->sk;
683 	struct netlink_sock *nlk = nlk_sk(sk);
684 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
685 
686 	if (alen < sizeof(addr->sa_family))
687 		return -EINVAL;
688 
689 	if (addr->sa_family == AF_UNSPEC) {
690 		sk->sk_state	= NETLINK_UNCONNECTED;
691 		nlk->dst_pid	= 0;
692 		nlk->dst_group  = 0;
693 		return 0;
694 	}
695 	if (addr->sa_family != AF_NETLINK)
696 		return -EINVAL;
697 
698 	/* Only superuser is allowed to send multicasts */
699 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
700 		return -EPERM;
701 
702 	if (!nlk->pid)
703 		err = netlink_autobind(sock);
704 
705 	if (err == 0) {
706 		sk->sk_state	= NETLINK_CONNECTED;
707 		nlk->dst_pid 	= nladdr->nl_pid;
708 		nlk->dst_group  = ffs(nladdr->nl_groups);
709 	}
710 
711 	return err;
712 }
713 
714 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
715 			   int *addr_len, int peer)
716 {
717 	struct sock *sk = sock->sk;
718 	struct netlink_sock *nlk = nlk_sk(sk);
719 	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
720 
721 	nladdr->nl_family = AF_NETLINK;
722 	nladdr->nl_pad = 0;
723 	*addr_len = sizeof(*nladdr);
724 
725 	if (peer) {
726 		nladdr->nl_pid = nlk->dst_pid;
727 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
728 	} else {
729 		nladdr->nl_pid = nlk->pid;
730 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
731 	}
732 	return 0;
733 }
734 
735 static void netlink_overrun(struct sock *sk)
736 {
737 	struct netlink_sock *nlk = nlk_sk(sk);
738 
739 	if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
740 		if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
741 			sk->sk_err = ENOBUFS;
742 			sk->sk_error_report(sk);
743 		}
744 	}
745 	atomic_inc(&sk->sk_drops);
746 }
747 
748 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
749 {
750 	struct sock *sock;
751 	struct netlink_sock *nlk;
752 
753 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
754 	if (!sock)
755 		return ERR_PTR(-ECONNREFUSED);
756 
757 	/* Don't bother queuing skb if kernel socket has no input function */
758 	nlk = nlk_sk(sock);
759 	if (sock->sk_state == NETLINK_CONNECTED &&
760 	    nlk->dst_pid != nlk_sk(ssk)->pid) {
761 		sock_put(sock);
762 		return ERR_PTR(-ECONNREFUSED);
763 	}
764 	return sock;
765 }
766 
767 struct sock *netlink_getsockbyfilp(struct file *filp)
768 {
769 	struct inode *inode = filp->f_path.dentry->d_inode;
770 	struct sock *sock;
771 
772 	if (!S_ISSOCK(inode->i_mode))
773 		return ERR_PTR(-ENOTSOCK);
774 
775 	sock = SOCKET_I(inode)->sk;
776 	if (sock->sk_family != AF_NETLINK)
777 		return ERR_PTR(-EINVAL);
778 
779 	sock_hold(sock);
780 	return sock;
781 }
782 
783 /*
784  * Attach a skb to a netlink socket.
785  * The caller must hold a reference to the destination socket. On error, the
786  * reference is dropped. The skb is not send to the destination, just all
787  * all error checks are performed and memory in the queue is reserved.
788  * Return values:
789  * < 0: error. skb freed, reference to sock dropped.
790  * 0: continue
791  * 1: repeat lookup - reference dropped while waiting for socket memory.
792  */
793 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
794 		      long *timeo, struct sock *ssk)
795 {
796 	struct netlink_sock *nlk;
797 
798 	nlk = nlk_sk(sk);
799 
800 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
801 	    test_bit(0, &nlk->state)) {
802 		DECLARE_WAITQUEUE(wait, current);
803 		if (!*timeo) {
804 			if (!ssk || netlink_is_kernel(ssk))
805 				netlink_overrun(sk);
806 			sock_put(sk);
807 			kfree_skb(skb);
808 			return -EAGAIN;
809 		}
810 
811 		__set_current_state(TASK_INTERRUPTIBLE);
812 		add_wait_queue(&nlk->wait, &wait);
813 
814 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
815 		     test_bit(0, &nlk->state)) &&
816 		    !sock_flag(sk, SOCK_DEAD))
817 			*timeo = schedule_timeout(*timeo);
818 
819 		__set_current_state(TASK_RUNNING);
820 		remove_wait_queue(&nlk->wait, &wait);
821 		sock_put(sk);
822 
823 		if (signal_pending(current)) {
824 			kfree_skb(skb);
825 			return sock_intr_errno(*timeo);
826 		}
827 		return 1;
828 	}
829 	skb_set_owner_r(skb, sk);
830 	return 0;
831 }
832 
833 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
834 {
835 	int len = skb->len;
836 
837 	skb_queue_tail(&sk->sk_receive_queue, skb);
838 	sk->sk_data_ready(sk, len);
839 	sock_put(sk);
840 	return len;
841 }
842 
843 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
844 {
845 	kfree_skb(skb);
846 	sock_put(sk);
847 }
848 
849 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
850 					   gfp_t allocation)
851 {
852 	int delta;
853 
854 	skb_orphan(skb);
855 
856 	delta = skb->end - skb->tail;
857 	if (delta * 2 < skb->truesize)
858 		return skb;
859 
860 	if (skb_shared(skb)) {
861 		struct sk_buff *nskb = skb_clone(skb, allocation);
862 		if (!nskb)
863 			return skb;
864 		kfree_skb(skb);
865 		skb = nskb;
866 	}
867 
868 	if (!pskb_expand_head(skb, 0, -delta, allocation))
869 		skb->truesize -= delta;
870 
871 	return skb;
872 }
873 
874 static inline void netlink_rcv_wake(struct sock *sk)
875 {
876 	struct netlink_sock *nlk = nlk_sk(sk);
877 
878 	if (skb_queue_empty(&sk->sk_receive_queue))
879 		clear_bit(0, &nlk->state);
880 	if (!test_bit(0, &nlk->state))
881 		wake_up_interruptible(&nlk->wait);
882 }
883 
884 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
885 {
886 	int ret;
887 	struct netlink_sock *nlk = nlk_sk(sk);
888 
889 	ret = -ECONNREFUSED;
890 	if (nlk->netlink_rcv != NULL) {
891 		ret = skb->len;
892 		skb_set_owner_r(skb, sk);
893 		nlk->netlink_rcv(skb);
894 	}
895 	kfree_skb(skb);
896 	sock_put(sk);
897 	return ret;
898 }
899 
900 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
901 		    u32 pid, int nonblock)
902 {
903 	struct sock *sk;
904 	int err;
905 	long timeo;
906 
907 	skb = netlink_trim(skb, gfp_any());
908 
909 	timeo = sock_sndtimeo(ssk, nonblock);
910 retry:
911 	sk = netlink_getsockbypid(ssk, pid);
912 	if (IS_ERR(sk)) {
913 		kfree_skb(skb);
914 		return PTR_ERR(sk);
915 	}
916 	if (netlink_is_kernel(sk))
917 		return netlink_unicast_kernel(sk, skb);
918 
919 	if (sk_filter(sk, skb)) {
920 		err = skb->len;
921 		kfree_skb(skb);
922 		sock_put(sk);
923 		return err;
924 	}
925 
926 	err = netlink_attachskb(sk, skb, &timeo, ssk);
927 	if (err == 1)
928 		goto retry;
929 	if (err)
930 		return err;
931 
932 	return netlink_sendskb(sk, skb);
933 }
934 EXPORT_SYMBOL(netlink_unicast);
935 
936 int netlink_has_listeners(struct sock *sk, unsigned int group)
937 {
938 	int res = 0;
939 	unsigned long *listeners;
940 
941 	BUG_ON(!netlink_is_kernel(sk));
942 
943 	rcu_read_lock();
944 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
945 
946 	if (group - 1 < nl_table[sk->sk_protocol].groups)
947 		res = test_bit(group - 1, listeners);
948 
949 	rcu_read_unlock();
950 
951 	return res;
952 }
953 EXPORT_SYMBOL_GPL(netlink_has_listeners);
954 
955 static inline int netlink_broadcast_deliver(struct sock *sk,
956 					    struct sk_buff *skb)
957 {
958 	struct netlink_sock *nlk = nlk_sk(sk);
959 
960 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
961 	    !test_bit(0, &nlk->state)) {
962 		skb_set_owner_r(skb, sk);
963 		skb_queue_tail(&sk->sk_receive_queue, skb);
964 		sk->sk_data_ready(sk, skb->len);
965 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
966 	}
967 	return -1;
968 }
969 
970 struct netlink_broadcast_data {
971 	struct sock *exclude_sk;
972 	struct net *net;
973 	u32 pid;
974 	u32 group;
975 	int failure;
976 	int delivery_failure;
977 	int congested;
978 	int delivered;
979 	gfp_t allocation;
980 	struct sk_buff *skb, *skb2;
981 };
982 
983 static inline int do_one_broadcast(struct sock *sk,
984 				   struct netlink_broadcast_data *p)
985 {
986 	struct netlink_sock *nlk = nlk_sk(sk);
987 	int val;
988 
989 	if (p->exclude_sk == sk)
990 		goto out;
991 
992 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
993 	    !test_bit(p->group - 1, nlk->groups))
994 		goto out;
995 
996 	if (!net_eq(sock_net(sk), p->net))
997 		goto out;
998 
999 	if (p->failure) {
1000 		netlink_overrun(sk);
1001 		goto out;
1002 	}
1003 
1004 	sock_hold(sk);
1005 	if (p->skb2 == NULL) {
1006 		if (skb_shared(p->skb)) {
1007 			p->skb2 = skb_clone(p->skb, p->allocation);
1008 		} else {
1009 			p->skb2 = skb_get(p->skb);
1010 			/*
1011 			 * skb ownership may have been set when
1012 			 * delivered to a previous socket.
1013 			 */
1014 			skb_orphan(p->skb2);
1015 		}
1016 	}
1017 	if (p->skb2 == NULL) {
1018 		netlink_overrun(sk);
1019 		/* Clone failed. Notify ALL listeners. */
1020 		p->failure = 1;
1021 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1022 			p->delivery_failure = 1;
1023 	} else if (sk_filter(sk, p->skb2)) {
1024 		kfree_skb(p->skb2);
1025 		p->skb2 = NULL;
1026 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1027 		netlink_overrun(sk);
1028 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1029 			p->delivery_failure = 1;
1030 	} else {
1031 		p->congested |= val;
1032 		p->delivered = 1;
1033 		p->skb2 = NULL;
1034 	}
1035 	sock_put(sk);
1036 
1037 out:
1038 	return 0;
1039 }
1040 
1041 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1042 		      u32 group, gfp_t allocation)
1043 {
1044 	struct net *net = sock_net(ssk);
1045 	struct netlink_broadcast_data info;
1046 	struct hlist_node *node;
1047 	struct sock *sk;
1048 
1049 	skb = netlink_trim(skb, allocation);
1050 
1051 	info.exclude_sk = ssk;
1052 	info.net = net;
1053 	info.pid = pid;
1054 	info.group = group;
1055 	info.failure = 0;
1056 	info.delivery_failure = 0;
1057 	info.congested = 0;
1058 	info.delivered = 0;
1059 	info.allocation = allocation;
1060 	info.skb = skb;
1061 	info.skb2 = NULL;
1062 
1063 	/* While we sleep in clone, do not allow to change socket list */
1064 
1065 	netlink_lock_table();
1066 
1067 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1068 		do_one_broadcast(sk, &info);
1069 
1070 	kfree_skb(skb);
1071 
1072 	netlink_unlock_table();
1073 
1074 	kfree_skb(info.skb2);
1075 
1076 	if (info.delivery_failure)
1077 		return -ENOBUFS;
1078 
1079 	if (info.delivered) {
1080 		if (info.congested && (allocation & __GFP_WAIT))
1081 			yield();
1082 		return 0;
1083 	}
1084 	return -ESRCH;
1085 }
1086 EXPORT_SYMBOL(netlink_broadcast);
1087 
1088 struct netlink_set_err_data {
1089 	struct sock *exclude_sk;
1090 	u32 pid;
1091 	u32 group;
1092 	int code;
1093 };
1094 
1095 static inline int do_one_set_err(struct sock *sk,
1096 				 struct netlink_set_err_data *p)
1097 {
1098 	struct netlink_sock *nlk = nlk_sk(sk);
1099 	int ret = 0;
1100 
1101 	if (sk == p->exclude_sk)
1102 		goto out;
1103 
1104 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1105 		goto out;
1106 
1107 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1108 	    !test_bit(p->group - 1, nlk->groups))
1109 		goto out;
1110 
1111 	if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1112 		ret = 1;
1113 		goto out;
1114 	}
1115 
1116 	sk->sk_err = p->code;
1117 	sk->sk_error_report(sk);
1118 out:
1119 	return ret;
1120 }
1121 
1122 /**
1123  * netlink_set_err - report error to broadcast listeners
1124  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1125  * @pid: the PID of a process that we want to skip (if any)
1126  * @groups: the broadcast group that will notice the error
1127  * @code: error code, must be negative (as usual in kernelspace)
1128  *
1129  * This function returns the number of broadcast listeners that have set the
1130  * NETLINK_RECV_NO_ENOBUFS socket option.
1131  */
1132 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1133 {
1134 	struct netlink_set_err_data info;
1135 	struct hlist_node *node;
1136 	struct sock *sk;
1137 	int ret = 0;
1138 
1139 	info.exclude_sk = ssk;
1140 	info.pid = pid;
1141 	info.group = group;
1142 	/* sk->sk_err wants a positive error value */
1143 	info.code = -code;
1144 
1145 	read_lock(&nl_table_lock);
1146 
1147 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1148 		ret += do_one_set_err(sk, &info);
1149 
1150 	read_unlock(&nl_table_lock);
1151 	return ret;
1152 }
1153 EXPORT_SYMBOL(netlink_set_err);
1154 
1155 /* must be called with netlink table grabbed */
1156 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1157 				     unsigned int group,
1158 				     int is_new)
1159 {
1160 	int old, new = !!is_new, subscriptions;
1161 
1162 	old = test_bit(group - 1, nlk->groups);
1163 	subscriptions = nlk->subscriptions - old + new;
1164 	if (new)
1165 		__set_bit(group - 1, nlk->groups);
1166 	else
1167 		__clear_bit(group - 1, nlk->groups);
1168 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1169 	netlink_update_listeners(&nlk->sk);
1170 }
1171 
1172 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1173 			      char __user *optval, unsigned int optlen)
1174 {
1175 	struct sock *sk = sock->sk;
1176 	struct netlink_sock *nlk = nlk_sk(sk);
1177 	unsigned int val = 0;
1178 	int err;
1179 
1180 	if (level != SOL_NETLINK)
1181 		return -ENOPROTOOPT;
1182 
1183 	if (optlen >= sizeof(int) &&
1184 	    get_user(val, (unsigned int __user *)optval))
1185 		return -EFAULT;
1186 
1187 	switch (optname) {
1188 	case NETLINK_PKTINFO:
1189 		if (val)
1190 			nlk->flags |= NETLINK_RECV_PKTINFO;
1191 		else
1192 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1193 		err = 0;
1194 		break;
1195 	case NETLINK_ADD_MEMBERSHIP:
1196 	case NETLINK_DROP_MEMBERSHIP: {
1197 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1198 			return -EPERM;
1199 		err = netlink_realloc_groups(sk);
1200 		if (err)
1201 			return err;
1202 		if (!val || val - 1 >= nlk->ngroups)
1203 			return -EINVAL;
1204 		netlink_table_grab();
1205 		netlink_update_socket_mc(nlk, val,
1206 					 optname == NETLINK_ADD_MEMBERSHIP);
1207 		netlink_table_ungrab();
1208 		err = 0;
1209 		break;
1210 	}
1211 	case NETLINK_BROADCAST_ERROR:
1212 		if (val)
1213 			nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1214 		else
1215 			nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1216 		err = 0;
1217 		break;
1218 	case NETLINK_NO_ENOBUFS:
1219 		if (val) {
1220 			nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1221 			clear_bit(0, &nlk->state);
1222 			wake_up_interruptible(&nlk->wait);
1223 		} else
1224 			nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1225 		err = 0;
1226 		break;
1227 	default:
1228 		err = -ENOPROTOOPT;
1229 	}
1230 	return err;
1231 }
1232 
1233 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1234 			      char __user *optval, int __user *optlen)
1235 {
1236 	struct sock *sk = sock->sk;
1237 	struct netlink_sock *nlk = nlk_sk(sk);
1238 	int len, val, err;
1239 
1240 	if (level != SOL_NETLINK)
1241 		return -ENOPROTOOPT;
1242 
1243 	if (get_user(len, optlen))
1244 		return -EFAULT;
1245 	if (len < 0)
1246 		return -EINVAL;
1247 
1248 	switch (optname) {
1249 	case NETLINK_PKTINFO:
1250 		if (len < sizeof(int))
1251 			return -EINVAL;
1252 		len = sizeof(int);
1253 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1254 		if (put_user(len, optlen) ||
1255 		    put_user(val, optval))
1256 			return -EFAULT;
1257 		err = 0;
1258 		break;
1259 	case NETLINK_BROADCAST_ERROR:
1260 		if (len < sizeof(int))
1261 			return -EINVAL;
1262 		len = sizeof(int);
1263 		val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1264 		if (put_user(len, optlen) ||
1265 		    put_user(val, optval))
1266 			return -EFAULT;
1267 		err = 0;
1268 		break;
1269 	case NETLINK_NO_ENOBUFS:
1270 		if (len < sizeof(int))
1271 			return -EINVAL;
1272 		len = sizeof(int);
1273 		val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1274 		if (put_user(len, optlen) ||
1275 		    put_user(val, optval))
1276 			return -EFAULT;
1277 		err = 0;
1278 		break;
1279 	default:
1280 		err = -ENOPROTOOPT;
1281 	}
1282 	return err;
1283 }
1284 
1285 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1286 {
1287 	struct nl_pktinfo info;
1288 
1289 	info.group = NETLINK_CB(skb).dst_group;
1290 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1291 }
1292 
1293 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1294 			   struct msghdr *msg, size_t len)
1295 {
1296 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1297 	struct sock *sk = sock->sk;
1298 	struct netlink_sock *nlk = nlk_sk(sk);
1299 	struct sockaddr_nl *addr = msg->msg_name;
1300 	u32 dst_pid;
1301 	u32 dst_group;
1302 	struct sk_buff *skb;
1303 	int err;
1304 	struct scm_cookie scm;
1305 
1306 	if (msg->msg_flags&MSG_OOB)
1307 		return -EOPNOTSUPP;
1308 
1309 	if (NULL == siocb->scm)
1310 		siocb->scm = &scm;
1311 	err = scm_send(sock, msg, siocb->scm);
1312 	if (err < 0)
1313 		return err;
1314 
1315 	if (msg->msg_namelen) {
1316 		if (addr->nl_family != AF_NETLINK)
1317 			return -EINVAL;
1318 		dst_pid = addr->nl_pid;
1319 		dst_group = ffs(addr->nl_groups);
1320 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1321 			return -EPERM;
1322 	} else {
1323 		dst_pid = nlk->dst_pid;
1324 		dst_group = nlk->dst_group;
1325 	}
1326 
1327 	if (!nlk->pid) {
1328 		err = netlink_autobind(sock);
1329 		if (err)
1330 			goto out;
1331 	}
1332 
1333 	err = -EMSGSIZE;
1334 	if (len > sk->sk_sndbuf - 32)
1335 		goto out;
1336 	err = -ENOBUFS;
1337 	skb = alloc_skb(len, GFP_KERNEL);
1338 	if (skb == NULL)
1339 		goto out;
1340 
1341 	NETLINK_CB(skb).pid	= nlk->pid;
1342 	NETLINK_CB(skb).dst_group = dst_group;
1343 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1344 	NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1345 	security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1346 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1347 
1348 	/* What can I do? Netlink is asynchronous, so that
1349 	   we will have to save current capabilities to
1350 	   check them, when this message will be delivered
1351 	   to corresponding kernel module.   --ANK (980802)
1352 	 */
1353 
1354 	err = -EFAULT;
1355 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1356 		kfree_skb(skb);
1357 		goto out;
1358 	}
1359 
1360 	err = security_netlink_send(sk, skb);
1361 	if (err) {
1362 		kfree_skb(skb);
1363 		goto out;
1364 	}
1365 
1366 	if (dst_group) {
1367 		atomic_inc(&skb->users);
1368 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1369 	}
1370 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1371 
1372 out:
1373 	return err;
1374 }
1375 
1376 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1377 			   struct msghdr *msg, size_t len,
1378 			   int flags)
1379 {
1380 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1381 	struct scm_cookie scm;
1382 	struct sock *sk = sock->sk;
1383 	struct netlink_sock *nlk = nlk_sk(sk);
1384 	int noblock = flags&MSG_DONTWAIT;
1385 	size_t copied;
1386 	struct sk_buff *skb, *frag __maybe_unused = NULL;
1387 	int err;
1388 
1389 	if (flags&MSG_OOB)
1390 		return -EOPNOTSUPP;
1391 
1392 	copied = 0;
1393 
1394 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1395 	if (skb == NULL)
1396 		goto out;
1397 
1398 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1399 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1400 		bool need_compat = !!(flags & MSG_CMSG_COMPAT);
1401 
1402 		/*
1403 		 * If this skb has a frag_list, then here that means that
1404 		 * we will have to use the frag_list skb for compat tasks
1405 		 * and the regular skb for non-compat tasks.
1406 		 *
1407 		 * The skb might (and likely will) be cloned, so we can't
1408 		 * just reset frag_list and go on with things -- we need to
1409 		 * keep that. For the compat case that's easy -- simply get
1410 		 * a reference to the compat skb and free the regular one
1411 		 * including the frag. For the non-compat case, we need to
1412 		 * avoid sending the frag to the user -- so assign NULL but
1413 		 * restore it below before freeing the skb.
1414 		 */
1415 		if (need_compat) {
1416 			struct sk_buff *compskb = skb_shinfo(skb)->frag_list;
1417 			skb_get(compskb);
1418 			kfree_skb(skb);
1419 			skb = compskb;
1420 		} else {
1421 			frag = skb_shinfo(skb)->frag_list;
1422 			skb_shinfo(skb)->frag_list = NULL;
1423 		}
1424 	}
1425 #endif
1426 
1427 	msg->msg_namelen = 0;
1428 
1429 	copied = skb->len;
1430 	if (len < copied) {
1431 		msg->msg_flags |= MSG_TRUNC;
1432 		copied = len;
1433 	}
1434 
1435 	skb_reset_transport_header(skb);
1436 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1437 
1438 	if (msg->msg_name) {
1439 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1440 		addr->nl_family = AF_NETLINK;
1441 		addr->nl_pad    = 0;
1442 		addr->nl_pid	= NETLINK_CB(skb).pid;
1443 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1444 		msg->msg_namelen = sizeof(*addr);
1445 	}
1446 
1447 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1448 		netlink_cmsg_recv_pktinfo(msg, skb);
1449 
1450 	if (NULL == siocb->scm) {
1451 		memset(&scm, 0, sizeof(scm));
1452 		siocb->scm = &scm;
1453 	}
1454 	siocb->scm->creds = *NETLINK_CREDS(skb);
1455 	if (flags & MSG_TRUNC)
1456 		copied = skb->len;
1457 
1458 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1459 	skb_shinfo(skb)->frag_list = frag;
1460 #endif
1461 
1462 	skb_free_datagram(sk, skb);
1463 
1464 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1465 		netlink_dump(sk);
1466 
1467 	scm_recv(sock, msg, siocb->scm, flags);
1468 out:
1469 	netlink_rcv_wake(sk);
1470 	return err ? : copied;
1471 }
1472 
1473 static void netlink_data_ready(struct sock *sk, int len)
1474 {
1475 	BUG();
1476 }
1477 
1478 /*
1479  *	We export these functions to other modules. They provide a
1480  *	complete set of kernel non-blocking support for message
1481  *	queueing.
1482  */
1483 
1484 struct sock *
1485 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1486 		      void (*input)(struct sk_buff *skb),
1487 		      struct mutex *cb_mutex, struct module *module)
1488 {
1489 	struct socket *sock;
1490 	struct sock *sk;
1491 	struct netlink_sock *nlk;
1492 	unsigned long *listeners = NULL;
1493 
1494 	BUG_ON(!nl_table);
1495 
1496 	if (unit < 0 || unit >= MAX_LINKS)
1497 		return NULL;
1498 
1499 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1500 		return NULL;
1501 
1502 	/*
1503 	 * We have to just have a reference on the net from sk, but don't
1504 	 * get_net it. Besides, we cannot get and then put the net here.
1505 	 * So we create one inside init_net and the move it to net.
1506 	 */
1507 
1508 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1509 		goto out_sock_release_nosk;
1510 
1511 	sk = sock->sk;
1512 	sk_change_net(sk, net);
1513 
1514 	if (groups < 32)
1515 		groups = 32;
1516 
1517 	listeners = kzalloc(NLGRPSZ(groups) + sizeof(struct listeners_rcu_head),
1518 			    GFP_KERNEL);
1519 	if (!listeners)
1520 		goto out_sock_release;
1521 
1522 	sk->sk_data_ready = netlink_data_ready;
1523 	if (input)
1524 		nlk_sk(sk)->netlink_rcv = input;
1525 
1526 	if (netlink_insert(sk, net, 0))
1527 		goto out_sock_release;
1528 
1529 	nlk = nlk_sk(sk);
1530 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1531 
1532 	netlink_table_grab();
1533 	if (!nl_table[unit].registered) {
1534 		nl_table[unit].groups = groups;
1535 		nl_table[unit].listeners = listeners;
1536 		nl_table[unit].cb_mutex = cb_mutex;
1537 		nl_table[unit].module = module;
1538 		nl_table[unit].registered = 1;
1539 	} else {
1540 		kfree(listeners);
1541 		nl_table[unit].registered++;
1542 	}
1543 	netlink_table_ungrab();
1544 	return sk;
1545 
1546 out_sock_release:
1547 	kfree(listeners);
1548 	netlink_kernel_release(sk);
1549 	return NULL;
1550 
1551 out_sock_release_nosk:
1552 	sock_release(sock);
1553 	return NULL;
1554 }
1555 EXPORT_SYMBOL(netlink_kernel_create);
1556 
1557 
1558 void
1559 netlink_kernel_release(struct sock *sk)
1560 {
1561 	sk_release_kernel(sk);
1562 }
1563 EXPORT_SYMBOL(netlink_kernel_release);
1564 
1565 
1566 static void netlink_free_old_listeners(struct rcu_head *rcu_head)
1567 {
1568 	struct listeners_rcu_head *lrh;
1569 
1570 	lrh = container_of(rcu_head, struct listeners_rcu_head, rcu_head);
1571 	kfree(lrh->ptr);
1572 }
1573 
1574 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1575 {
1576 	unsigned long *listeners, *old = NULL;
1577 	struct listeners_rcu_head *old_rcu_head;
1578 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1579 
1580 	if (groups < 32)
1581 		groups = 32;
1582 
1583 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1584 		listeners = kzalloc(NLGRPSZ(groups) +
1585 				    sizeof(struct listeners_rcu_head),
1586 				    GFP_ATOMIC);
1587 		if (!listeners)
1588 			return -ENOMEM;
1589 		old = tbl->listeners;
1590 		memcpy(listeners, old, NLGRPSZ(tbl->groups));
1591 		rcu_assign_pointer(tbl->listeners, listeners);
1592 		/*
1593 		 * Free the old memory after an RCU grace period so we
1594 		 * don't leak it. We use call_rcu() here in order to be
1595 		 * able to call this function from atomic contexts. The
1596 		 * allocation of this memory will have reserved enough
1597 		 * space for struct listeners_rcu_head at the end.
1598 		 */
1599 		old_rcu_head = (void *)(tbl->listeners +
1600 					NLGRPLONGS(tbl->groups));
1601 		old_rcu_head->ptr = old;
1602 		call_rcu(&old_rcu_head->rcu_head, netlink_free_old_listeners);
1603 	}
1604 	tbl->groups = groups;
1605 
1606 	return 0;
1607 }
1608 
1609 /**
1610  * netlink_change_ngroups - change number of multicast groups
1611  *
1612  * This changes the number of multicast groups that are available
1613  * on a certain netlink family. Note that it is not possible to
1614  * change the number of groups to below 32. Also note that it does
1615  * not implicitly call netlink_clear_multicast_users() when the
1616  * number of groups is reduced.
1617  *
1618  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1619  * @groups: The new number of groups.
1620  */
1621 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1622 {
1623 	int err;
1624 
1625 	netlink_table_grab();
1626 	err = __netlink_change_ngroups(sk, groups);
1627 	netlink_table_ungrab();
1628 
1629 	return err;
1630 }
1631 
1632 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1633 {
1634 	struct sock *sk;
1635 	struct hlist_node *node;
1636 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1637 
1638 	sk_for_each_bound(sk, node, &tbl->mc_list)
1639 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1640 }
1641 
1642 /**
1643  * netlink_clear_multicast_users - kick off multicast listeners
1644  *
1645  * This function removes all listeners from the given group.
1646  * @ksk: The kernel netlink socket, as returned by
1647  *	netlink_kernel_create().
1648  * @group: The multicast group to clear.
1649  */
1650 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1651 {
1652 	netlink_table_grab();
1653 	__netlink_clear_multicast_users(ksk, group);
1654 	netlink_table_ungrab();
1655 }
1656 
1657 void netlink_set_nonroot(int protocol, unsigned int flags)
1658 {
1659 	if ((unsigned int)protocol < MAX_LINKS)
1660 		nl_table[protocol].nl_nonroot = flags;
1661 }
1662 EXPORT_SYMBOL(netlink_set_nonroot);
1663 
1664 static void netlink_destroy_callback(struct netlink_callback *cb)
1665 {
1666 	kfree_skb(cb->skb);
1667 	kfree(cb);
1668 }
1669 
1670 /*
1671  * It looks a bit ugly.
1672  * It would be better to create kernel thread.
1673  */
1674 
1675 static int netlink_dump(struct sock *sk)
1676 {
1677 	struct netlink_sock *nlk = nlk_sk(sk);
1678 	struct netlink_callback *cb;
1679 	struct sk_buff *skb;
1680 	struct nlmsghdr *nlh;
1681 	int len, err = -ENOBUFS;
1682 
1683 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1684 	if (!skb)
1685 		goto errout;
1686 
1687 	mutex_lock(nlk->cb_mutex);
1688 
1689 	cb = nlk->cb;
1690 	if (cb == NULL) {
1691 		err = -EINVAL;
1692 		goto errout_skb;
1693 	}
1694 
1695 	len = cb->dump(skb, cb);
1696 
1697 	if (len > 0) {
1698 		mutex_unlock(nlk->cb_mutex);
1699 
1700 		if (sk_filter(sk, skb))
1701 			kfree_skb(skb);
1702 		else {
1703 			skb_queue_tail(&sk->sk_receive_queue, skb);
1704 			sk->sk_data_ready(sk, skb->len);
1705 		}
1706 		return 0;
1707 	}
1708 
1709 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1710 	if (!nlh)
1711 		goto errout_skb;
1712 
1713 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1714 
1715 	if (sk_filter(sk, skb))
1716 		kfree_skb(skb);
1717 	else {
1718 		skb_queue_tail(&sk->sk_receive_queue, skb);
1719 		sk->sk_data_ready(sk, skb->len);
1720 	}
1721 
1722 	if (cb->done)
1723 		cb->done(cb);
1724 	nlk->cb = NULL;
1725 	mutex_unlock(nlk->cb_mutex);
1726 
1727 	netlink_destroy_callback(cb);
1728 	return 0;
1729 
1730 errout_skb:
1731 	mutex_unlock(nlk->cb_mutex);
1732 	kfree_skb(skb);
1733 errout:
1734 	return err;
1735 }
1736 
1737 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1738 		       const struct nlmsghdr *nlh,
1739 		       int (*dump)(struct sk_buff *skb,
1740 				   struct netlink_callback *),
1741 		       int (*done)(struct netlink_callback *))
1742 {
1743 	struct netlink_callback *cb;
1744 	struct sock *sk;
1745 	struct netlink_sock *nlk;
1746 
1747 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1748 	if (cb == NULL)
1749 		return -ENOBUFS;
1750 
1751 	cb->dump = dump;
1752 	cb->done = done;
1753 	cb->nlh = nlh;
1754 	atomic_inc(&skb->users);
1755 	cb->skb = skb;
1756 
1757 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1758 	if (sk == NULL) {
1759 		netlink_destroy_callback(cb);
1760 		return -ECONNREFUSED;
1761 	}
1762 	nlk = nlk_sk(sk);
1763 	/* A dump is in progress... */
1764 	mutex_lock(nlk->cb_mutex);
1765 	if (nlk->cb) {
1766 		mutex_unlock(nlk->cb_mutex);
1767 		netlink_destroy_callback(cb);
1768 		sock_put(sk);
1769 		return -EBUSY;
1770 	}
1771 	nlk->cb = cb;
1772 	mutex_unlock(nlk->cb_mutex);
1773 
1774 	netlink_dump(sk);
1775 	sock_put(sk);
1776 
1777 	/* We successfully started a dump, by returning -EINTR we
1778 	 * signal not to send ACK even if it was requested.
1779 	 */
1780 	return -EINTR;
1781 }
1782 EXPORT_SYMBOL(netlink_dump_start);
1783 
1784 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1785 {
1786 	struct sk_buff *skb;
1787 	struct nlmsghdr *rep;
1788 	struct nlmsgerr *errmsg;
1789 	size_t payload = sizeof(*errmsg);
1790 
1791 	/* error messages get the original request appened */
1792 	if (err)
1793 		payload += nlmsg_len(nlh);
1794 
1795 	skb = nlmsg_new(payload, GFP_KERNEL);
1796 	if (!skb) {
1797 		struct sock *sk;
1798 
1799 		sk = netlink_lookup(sock_net(in_skb->sk),
1800 				    in_skb->sk->sk_protocol,
1801 				    NETLINK_CB(in_skb).pid);
1802 		if (sk) {
1803 			sk->sk_err = ENOBUFS;
1804 			sk->sk_error_report(sk);
1805 			sock_put(sk);
1806 		}
1807 		return;
1808 	}
1809 
1810 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1811 			  NLMSG_ERROR, payload, 0);
1812 	errmsg = nlmsg_data(rep);
1813 	errmsg->error = err;
1814 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1815 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1816 }
1817 EXPORT_SYMBOL(netlink_ack);
1818 
1819 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1820 						     struct nlmsghdr *))
1821 {
1822 	struct nlmsghdr *nlh;
1823 	int err;
1824 
1825 	while (skb->len >= nlmsg_total_size(0)) {
1826 		int msglen;
1827 
1828 		nlh = nlmsg_hdr(skb);
1829 		err = 0;
1830 
1831 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1832 			return 0;
1833 
1834 		/* Only requests are handled by the kernel */
1835 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1836 			goto ack;
1837 
1838 		/* Skip control messages */
1839 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1840 			goto ack;
1841 
1842 		err = cb(skb, nlh);
1843 		if (err == -EINTR)
1844 			goto skip;
1845 
1846 ack:
1847 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1848 			netlink_ack(skb, nlh, err);
1849 
1850 skip:
1851 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1852 		if (msglen > skb->len)
1853 			msglen = skb->len;
1854 		skb_pull(skb, msglen);
1855 	}
1856 
1857 	return 0;
1858 }
1859 EXPORT_SYMBOL(netlink_rcv_skb);
1860 
1861 /**
1862  * nlmsg_notify - send a notification netlink message
1863  * @sk: netlink socket to use
1864  * @skb: notification message
1865  * @pid: destination netlink pid for reports or 0
1866  * @group: destination multicast group or 0
1867  * @report: 1 to report back, 0 to disable
1868  * @flags: allocation flags
1869  */
1870 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1871 		 unsigned int group, int report, gfp_t flags)
1872 {
1873 	int err = 0;
1874 
1875 	if (group) {
1876 		int exclude_pid = 0;
1877 
1878 		if (report) {
1879 			atomic_inc(&skb->users);
1880 			exclude_pid = pid;
1881 		}
1882 
1883 		/* errors reported via destination sk->sk_err, but propagate
1884 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1885 		err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1886 	}
1887 
1888 	if (report) {
1889 		int err2;
1890 
1891 		err2 = nlmsg_unicast(sk, skb, pid);
1892 		if (!err || err == -ESRCH)
1893 			err = err2;
1894 	}
1895 
1896 	return err;
1897 }
1898 EXPORT_SYMBOL(nlmsg_notify);
1899 
1900 #ifdef CONFIG_PROC_FS
1901 struct nl_seq_iter {
1902 	struct seq_net_private p;
1903 	int link;
1904 	int hash_idx;
1905 };
1906 
1907 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1908 {
1909 	struct nl_seq_iter *iter = seq->private;
1910 	int i, j;
1911 	struct sock *s;
1912 	struct hlist_node *node;
1913 	loff_t off = 0;
1914 
1915 	for (i = 0; i < MAX_LINKS; i++) {
1916 		struct nl_pid_hash *hash = &nl_table[i].hash;
1917 
1918 		for (j = 0; j <= hash->mask; j++) {
1919 			sk_for_each(s, node, &hash->table[j]) {
1920 				if (sock_net(s) != seq_file_net(seq))
1921 					continue;
1922 				if (off == pos) {
1923 					iter->link = i;
1924 					iter->hash_idx = j;
1925 					return s;
1926 				}
1927 				++off;
1928 			}
1929 		}
1930 	}
1931 	return NULL;
1932 }
1933 
1934 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1935 	__acquires(nl_table_lock)
1936 {
1937 	read_lock(&nl_table_lock);
1938 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1939 }
1940 
1941 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1942 {
1943 	struct sock *s;
1944 	struct nl_seq_iter *iter;
1945 	int i, j;
1946 
1947 	++*pos;
1948 
1949 	if (v == SEQ_START_TOKEN)
1950 		return netlink_seq_socket_idx(seq, 0);
1951 
1952 	iter = seq->private;
1953 	s = v;
1954 	do {
1955 		s = sk_next(s);
1956 	} while (s && sock_net(s) != seq_file_net(seq));
1957 	if (s)
1958 		return s;
1959 
1960 	i = iter->link;
1961 	j = iter->hash_idx + 1;
1962 
1963 	do {
1964 		struct nl_pid_hash *hash = &nl_table[i].hash;
1965 
1966 		for (; j <= hash->mask; j++) {
1967 			s = sk_head(&hash->table[j]);
1968 			while (s && sock_net(s) != seq_file_net(seq))
1969 				s = sk_next(s);
1970 			if (s) {
1971 				iter->link = i;
1972 				iter->hash_idx = j;
1973 				return s;
1974 			}
1975 		}
1976 
1977 		j = 0;
1978 	} while (++i < MAX_LINKS);
1979 
1980 	return NULL;
1981 }
1982 
1983 static void netlink_seq_stop(struct seq_file *seq, void *v)
1984 	__releases(nl_table_lock)
1985 {
1986 	read_unlock(&nl_table_lock);
1987 }
1988 
1989 
1990 static int netlink_seq_show(struct seq_file *seq, void *v)
1991 {
1992 	if (v == SEQ_START_TOKEN)
1993 		seq_puts(seq,
1994 			 "sk       Eth Pid    Groups   "
1995 			 "Rmem     Wmem     Dump     Locks     Drops     Inode\n");
1996 	else {
1997 		struct sock *s = v;
1998 		struct netlink_sock *nlk = nlk_sk(s);
1999 
2000 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %-8d %-8d %-8lu\n",
2001 			   s,
2002 			   s->sk_protocol,
2003 			   nlk->pid,
2004 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2005 			   sk_rmem_alloc_get(s),
2006 			   sk_wmem_alloc_get(s),
2007 			   nlk->cb,
2008 			   atomic_read(&s->sk_refcnt),
2009 			   atomic_read(&s->sk_drops),
2010 			   sock_i_ino(s)
2011 			);
2012 
2013 	}
2014 	return 0;
2015 }
2016 
2017 static const struct seq_operations netlink_seq_ops = {
2018 	.start  = netlink_seq_start,
2019 	.next   = netlink_seq_next,
2020 	.stop   = netlink_seq_stop,
2021 	.show   = netlink_seq_show,
2022 };
2023 
2024 
2025 static int netlink_seq_open(struct inode *inode, struct file *file)
2026 {
2027 	return seq_open_net(inode, file, &netlink_seq_ops,
2028 				sizeof(struct nl_seq_iter));
2029 }
2030 
2031 static const struct file_operations netlink_seq_fops = {
2032 	.owner		= THIS_MODULE,
2033 	.open		= netlink_seq_open,
2034 	.read		= seq_read,
2035 	.llseek		= seq_lseek,
2036 	.release	= seq_release_net,
2037 };
2038 
2039 #endif
2040 
2041 int netlink_register_notifier(struct notifier_block *nb)
2042 {
2043 	return atomic_notifier_chain_register(&netlink_chain, nb);
2044 }
2045 EXPORT_SYMBOL(netlink_register_notifier);
2046 
2047 int netlink_unregister_notifier(struct notifier_block *nb)
2048 {
2049 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
2050 }
2051 EXPORT_SYMBOL(netlink_unregister_notifier);
2052 
2053 static const struct proto_ops netlink_ops = {
2054 	.family =	PF_NETLINK,
2055 	.owner =	THIS_MODULE,
2056 	.release =	netlink_release,
2057 	.bind =		netlink_bind,
2058 	.connect =	netlink_connect,
2059 	.socketpair =	sock_no_socketpair,
2060 	.accept =	sock_no_accept,
2061 	.getname =	netlink_getname,
2062 	.poll =		datagram_poll,
2063 	.ioctl =	sock_no_ioctl,
2064 	.listen =	sock_no_listen,
2065 	.shutdown =	sock_no_shutdown,
2066 	.setsockopt =	netlink_setsockopt,
2067 	.getsockopt =	netlink_getsockopt,
2068 	.sendmsg =	netlink_sendmsg,
2069 	.recvmsg =	netlink_recvmsg,
2070 	.mmap =		sock_no_mmap,
2071 	.sendpage =	sock_no_sendpage,
2072 };
2073 
2074 static const struct net_proto_family netlink_family_ops = {
2075 	.family = PF_NETLINK,
2076 	.create = netlink_create,
2077 	.owner	= THIS_MODULE,	/* for consistency 8) */
2078 };
2079 
2080 static int __net_init netlink_net_init(struct net *net)
2081 {
2082 #ifdef CONFIG_PROC_FS
2083 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2084 		return -ENOMEM;
2085 #endif
2086 	return 0;
2087 }
2088 
2089 static void __net_exit netlink_net_exit(struct net *net)
2090 {
2091 #ifdef CONFIG_PROC_FS
2092 	proc_net_remove(net, "netlink");
2093 #endif
2094 }
2095 
2096 static struct pernet_operations __net_initdata netlink_net_ops = {
2097 	.init = netlink_net_init,
2098 	.exit = netlink_net_exit,
2099 };
2100 
2101 static int __init netlink_proto_init(void)
2102 {
2103 	struct sk_buff *dummy_skb;
2104 	int i;
2105 	unsigned long limit;
2106 	unsigned int order;
2107 	int err = proto_register(&netlink_proto, 0);
2108 
2109 	if (err != 0)
2110 		goto out;
2111 
2112 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2113 
2114 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2115 	if (!nl_table)
2116 		goto panic;
2117 
2118 	if (totalram_pages >= (128 * 1024))
2119 		limit = totalram_pages >> (21 - PAGE_SHIFT);
2120 	else
2121 		limit = totalram_pages >> (23 - PAGE_SHIFT);
2122 
2123 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2124 	limit = (1UL << order) / sizeof(struct hlist_head);
2125 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2126 
2127 	for (i = 0; i < MAX_LINKS; i++) {
2128 		struct nl_pid_hash *hash = &nl_table[i].hash;
2129 
2130 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2131 		if (!hash->table) {
2132 			while (i-- > 0)
2133 				nl_pid_hash_free(nl_table[i].hash.table,
2134 						 1 * sizeof(*hash->table));
2135 			kfree(nl_table);
2136 			goto panic;
2137 		}
2138 		hash->max_shift = order;
2139 		hash->shift = 0;
2140 		hash->mask = 0;
2141 		hash->rehash_time = jiffies;
2142 	}
2143 
2144 	sock_register(&netlink_family_ops);
2145 	register_pernet_subsys(&netlink_net_ops);
2146 	/* The netlink device handler may be needed early. */
2147 	rtnetlink_init();
2148 out:
2149 	return err;
2150 panic:
2151 	panic("netlink_init: Cannot allocate nl_table\n");
2152 }
2153 
2154 core_initcall(netlink_proto_init);
2155