xref: /linux/net/netlink/af_netlink.c (revision a0b54e256d513ed99e456bea6e4e188ff92e7c46)
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 static void netlink_table_grab(void)
181 	__acquires(nl_table_lock)
182 {
183 	write_lock_irq(&nl_table_lock);
184 
185 	if (atomic_read(&nl_table_users)) {
186 		DECLARE_WAITQUEUE(wait, current);
187 
188 		add_wait_queue_exclusive(&nl_table_wait, &wait);
189 		for (;;) {
190 			set_current_state(TASK_UNINTERRUPTIBLE);
191 			if (atomic_read(&nl_table_users) == 0)
192 				break;
193 			write_unlock_irq(&nl_table_lock);
194 			schedule();
195 			write_lock_irq(&nl_table_lock);
196 		}
197 
198 		__set_current_state(TASK_RUNNING);
199 		remove_wait_queue(&nl_table_wait, &wait);
200 	}
201 }
202 
203 static void netlink_table_ungrab(void)
204 	__releases(nl_table_lock)
205 {
206 	write_unlock_irq(&nl_table_lock);
207 	wake_up(&nl_table_wait);
208 }
209 
210 static inline void
211 netlink_lock_table(void)
212 {
213 	/* read_lock() synchronizes us to netlink_table_grab */
214 
215 	read_lock(&nl_table_lock);
216 	atomic_inc(&nl_table_users);
217 	read_unlock(&nl_table_lock);
218 }
219 
220 static inline void
221 netlink_unlock_table(void)
222 {
223 	if (atomic_dec_and_test(&nl_table_users))
224 		wake_up(&nl_table_wait);
225 }
226 
227 static inline struct sock *netlink_lookup(struct net *net, int protocol,
228 					  u32 pid)
229 {
230 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
231 	struct hlist_head *head;
232 	struct sock *sk;
233 	struct hlist_node *node;
234 
235 	read_lock(&nl_table_lock);
236 	head = nl_pid_hashfn(hash, pid);
237 	sk_for_each(sk, node, head) {
238 		if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
239 			sock_hold(sk);
240 			goto found;
241 		}
242 	}
243 	sk = NULL;
244 found:
245 	read_unlock(&nl_table_lock);
246 	return sk;
247 }
248 
249 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
250 {
251 	if (size <= PAGE_SIZE)
252 		return kzalloc(size, GFP_ATOMIC);
253 	else
254 		return (struct hlist_head *)
255 			__get_free_pages(GFP_ATOMIC | __GFP_ZERO,
256 					 get_order(size));
257 }
258 
259 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
260 {
261 	if (size <= PAGE_SIZE)
262 		kfree(table);
263 	else
264 		free_pages((unsigned long)table, get_order(size));
265 }
266 
267 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
268 {
269 	unsigned int omask, mask, shift;
270 	size_t osize, size;
271 	struct hlist_head *otable, *table;
272 	int i;
273 
274 	omask = mask = hash->mask;
275 	osize = size = (mask + 1) * sizeof(*table);
276 	shift = hash->shift;
277 
278 	if (grow) {
279 		if (++shift > hash->max_shift)
280 			return 0;
281 		mask = mask * 2 + 1;
282 		size *= 2;
283 	}
284 
285 	table = nl_pid_hash_zalloc(size);
286 	if (!table)
287 		return 0;
288 
289 	otable = hash->table;
290 	hash->table = table;
291 	hash->mask = mask;
292 	hash->shift = shift;
293 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
294 
295 	for (i = 0; i <= omask; i++) {
296 		struct sock *sk;
297 		struct hlist_node *node, *tmp;
298 
299 		sk_for_each_safe(sk, node, tmp, &otable[i])
300 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
301 	}
302 
303 	nl_pid_hash_free(otable, osize);
304 	hash->rehash_time = jiffies + 10 * 60 * HZ;
305 	return 1;
306 }
307 
308 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
309 {
310 	int avg = hash->entries >> hash->shift;
311 
312 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
313 		return 1;
314 
315 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
316 		nl_pid_hash_rehash(hash, 0);
317 		return 1;
318 	}
319 
320 	return 0;
321 }
322 
323 static const struct proto_ops netlink_ops;
324 
325 static void
326 netlink_update_listeners(struct sock *sk)
327 {
328 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
329 	struct hlist_node *node;
330 	unsigned long mask;
331 	unsigned int i;
332 
333 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
334 		mask = 0;
335 		sk_for_each_bound(sk, node, &tbl->mc_list) {
336 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
337 				mask |= nlk_sk(sk)->groups[i];
338 		}
339 		tbl->listeners[i] = mask;
340 	}
341 	/* this function is only called with the netlink table "grabbed", which
342 	 * makes sure updates are visible before bind or setsockopt return. */
343 }
344 
345 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
346 {
347 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
348 	struct hlist_head *head;
349 	int err = -EADDRINUSE;
350 	struct sock *osk;
351 	struct hlist_node *node;
352 	int len;
353 
354 	netlink_table_grab();
355 	head = nl_pid_hashfn(hash, pid);
356 	len = 0;
357 	sk_for_each(osk, node, head) {
358 		if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
359 			break;
360 		len++;
361 	}
362 	if (node)
363 		goto err;
364 
365 	err = -EBUSY;
366 	if (nlk_sk(sk)->pid)
367 		goto err;
368 
369 	err = -ENOMEM;
370 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
371 		goto err;
372 
373 	if (len && nl_pid_hash_dilute(hash, len))
374 		head = nl_pid_hashfn(hash, pid);
375 	hash->entries++;
376 	nlk_sk(sk)->pid = pid;
377 	sk_add_node(sk, head);
378 	err = 0;
379 
380 err:
381 	netlink_table_ungrab();
382 	return err;
383 }
384 
385 static void netlink_remove(struct sock *sk)
386 {
387 	netlink_table_grab();
388 	if (sk_del_node_init(sk))
389 		nl_table[sk->sk_protocol].hash.entries--;
390 	if (nlk_sk(sk)->subscriptions)
391 		__sk_del_bind_node(sk);
392 	netlink_table_ungrab();
393 }
394 
395 static struct proto netlink_proto = {
396 	.name	  = "NETLINK",
397 	.owner	  = THIS_MODULE,
398 	.obj_size = sizeof(struct netlink_sock),
399 };
400 
401 static int __netlink_create(struct net *net, struct socket *sock,
402 			    struct mutex *cb_mutex, int protocol)
403 {
404 	struct sock *sk;
405 	struct netlink_sock *nlk;
406 
407 	sock->ops = &netlink_ops;
408 
409 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
410 	if (!sk)
411 		return -ENOMEM;
412 
413 	sock_init_data(sock, sk);
414 
415 	nlk = nlk_sk(sk);
416 	if (cb_mutex)
417 		nlk->cb_mutex = cb_mutex;
418 	else {
419 		nlk->cb_mutex = &nlk->cb_def_mutex;
420 		mutex_init(nlk->cb_mutex);
421 	}
422 	init_waitqueue_head(&nlk->wait);
423 
424 	sk->sk_destruct = netlink_sock_destruct;
425 	sk->sk_protocol = protocol;
426 	return 0;
427 }
428 
429 static int netlink_create(struct net *net, struct socket *sock, int protocol)
430 {
431 	struct module *module = NULL;
432 	struct mutex *cb_mutex;
433 	struct netlink_sock *nlk;
434 	int err = 0;
435 
436 	sock->state = SS_UNCONNECTED;
437 
438 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
439 		return -ESOCKTNOSUPPORT;
440 
441 	if (protocol < 0 || protocol >= MAX_LINKS)
442 		return -EPROTONOSUPPORT;
443 
444 	netlink_lock_table();
445 #ifdef CONFIG_MODULES
446 	if (!nl_table[protocol].registered) {
447 		netlink_unlock_table();
448 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
449 		netlink_lock_table();
450 	}
451 #endif
452 	if (nl_table[protocol].registered &&
453 	    try_module_get(nl_table[protocol].module))
454 		module = nl_table[protocol].module;
455 	cb_mutex = nl_table[protocol].cb_mutex;
456 	netlink_unlock_table();
457 
458 	err = __netlink_create(net, sock, cb_mutex, protocol);
459 	if (err < 0)
460 		goto out_module;
461 
462 	local_bh_disable();
463 	sock_prot_inuse_add(net, &netlink_proto, 1);
464 	local_bh_enable();
465 
466 	nlk = nlk_sk(sock->sk);
467 	nlk->module = module;
468 out:
469 	return err;
470 
471 out_module:
472 	module_put(module);
473 	goto out;
474 }
475 
476 static int netlink_release(struct socket *sock)
477 {
478 	struct sock *sk = sock->sk;
479 	struct netlink_sock *nlk;
480 
481 	if (!sk)
482 		return 0;
483 
484 	netlink_remove(sk);
485 	sock_orphan(sk);
486 	nlk = nlk_sk(sk);
487 
488 	/*
489 	 * OK. Socket is unlinked, any packets that arrive now
490 	 * will be purged.
491 	 */
492 
493 	sock->sk = NULL;
494 	wake_up_interruptible_all(&nlk->wait);
495 
496 	skb_queue_purge(&sk->sk_write_queue);
497 
498 	if (nlk->pid && !nlk->subscriptions) {
499 		struct netlink_notify n = {
500 						.net = sock_net(sk),
501 						.protocol = sk->sk_protocol,
502 						.pid = nlk->pid,
503 					  };
504 		atomic_notifier_call_chain(&netlink_chain,
505 				NETLINK_URELEASE, &n);
506 	}
507 
508 	module_put(nlk->module);
509 
510 	netlink_table_grab();
511 	if (netlink_is_kernel(sk)) {
512 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
513 		if (--nl_table[sk->sk_protocol].registered == 0) {
514 			kfree(nl_table[sk->sk_protocol].listeners);
515 			nl_table[sk->sk_protocol].module = NULL;
516 			nl_table[sk->sk_protocol].registered = 0;
517 		}
518 	} else if (nlk->subscriptions)
519 		netlink_update_listeners(sk);
520 	netlink_table_ungrab();
521 
522 	kfree(nlk->groups);
523 	nlk->groups = NULL;
524 
525 	local_bh_disable();
526 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
527 	local_bh_enable();
528 	sock_put(sk);
529 	return 0;
530 }
531 
532 static int netlink_autobind(struct socket *sock)
533 {
534 	struct sock *sk = sock->sk;
535 	struct net *net = sock_net(sk);
536 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
537 	struct hlist_head *head;
538 	struct sock *osk;
539 	struct hlist_node *node;
540 	s32 pid = current->tgid;
541 	int err;
542 	static s32 rover = -4097;
543 
544 retry:
545 	cond_resched();
546 	netlink_table_grab();
547 	head = nl_pid_hashfn(hash, pid);
548 	sk_for_each(osk, node, head) {
549 		if (!net_eq(sock_net(osk), net))
550 			continue;
551 		if (nlk_sk(osk)->pid == pid) {
552 			/* Bind collision, search negative pid values. */
553 			pid = rover--;
554 			if (rover > -4097)
555 				rover = -4097;
556 			netlink_table_ungrab();
557 			goto retry;
558 		}
559 	}
560 	netlink_table_ungrab();
561 
562 	err = netlink_insert(sk, net, pid);
563 	if (err == -EADDRINUSE)
564 		goto retry;
565 
566 	/* If 2 threads race to autobind, that is fine.  */
567 	if (err == -EBUSY)
568 		err = 0;
569 
570 	return err;
571 }
572 
573 static inline int netlink_capable(struct socket *sock, unsigned int flag)
574 {
575 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
576 	       capable(CAP_NET_ADMIN);
577 }
578 
579 static void
580 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
581 {
582 	struct netlink_sock *nlk = nlk_sk(sk);
583 
584 	if (nlk->subscriptions && !subscriptions)
585 		__sk_del_bind_node(sk);
586 	else if (!nlk->subscriptions && subscriptions)
587 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
588 	nlk->subscriptions = subscriptions;
589 }
590 
591 static int netlink_realloc_groups(struct sock *sk)
592 {
593 	struct netlink_sock *nlk = nlk_sk(sk);
594 	unsigned int groups;
595 	unsigned long *new_groups;
596 	int err = 0;
597 
598 	netlink_table_grab();
599 
600 	groups = nl_table[sk->sk_protocol].groups;
601 	if (!nl_table[sk->sk_protocol].registered) {
602 		err = -ENOENT;
603 		goto out_unlock;
604 	}
605 
606 	if (nlk->ngroups >= groups)
607 		goto out_unlock;
608 
609 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
610 	if (new_groups == NULL) {
611 		err = -ENOMEM;
612 		goto out_unlock;
613 	}
614 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
615 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
616 
617 	nlk->groups = new_groups;
618 	nlk->ngroups = groups;
619  out_unlock:
620 	netlink_table_ungrab();
621 	return err;
622 }
623 
624 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
625 			int addr_len)
626 {
627 	struct sock *sk = sock->sk;
628 	struct net *net = sock_net(sk);
629 	struct netlink_sock *nlk = nlk_sk(sk);
630 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
631 	int err;
632 
633 	if (nladdr->nl_family != AF_NETLINK)
634 		return -EINVAL;
635 
636 	/* Only superuser is allowed to listen multicasts */
637 	if (nladdr->nl_groups) {
638 		if (!netlink_capable(sock, NL_NONROOT_RECV))
639 			return -EPERM;
640 		err = netlink_realloc_groups(sk);
641 		if (err)
642 			return err;
643 	}
644 
645 	if (nlk->pid) {
646 		if (nladdr->nl_pid != nlk->pid)
647 			return -EINVAL;
648 	} else {
649 		err = nladdr->nl_pid ?
650 			netlink_insert(sk, net, nladdr->nl_pid) :
651 			netlink_autobind(sock);
652 		if (err)
653 			return err;
654 	}
655 
656 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
657 		return 0;
658 
659 	netlink_table_grab();
660 	netlink_update_subscriptions(sk, nlk->subscriptions +
661 					 hweight32(nladdr->nl_groups) -
662 					 hweight32(nlk->groups[0]));
663 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
664 	netlink_update_listeners(sk);
665 	netlink_table_ungrab();
666 
667 	return 0;
668 }
669 
670 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
671 			   int alen, int flags)
672 {
673 	int err = 0;
674 	struct sock *sk = sock->sk;
675 	struct netlink_sock *nlk = nlk_sk(sk);
676 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
677 
678 	if (addr->sa_family == AF_UNSPEC) {
679 		sk->sk_state	= NETLINK_UNCONNECTED;
680 		nlk->dst_pid	= 0;
681 		nlk->dst_group  = 0;
682 		return 0;
683 	}
684 	if (addr->sa_family != AF_NETLINK)
685 		return -EINVAL;
686 
687 	/* Only superuser is allowed to send multicasts */
688 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
689 		return -EPERM;
690 
691 	if (!nlk->pid)
692 		err = netlink_autobind(sock);
693 
694 	if (err == 0) {
695 		sk->sk_state	= NETLINK_CONNECTED;
696 		nlk->dst_pid 	= nladdr->nl_pid;
697 		nlk->dst_group  = ffs(nladdr->nl_groups);
698 	}
699 
700 	return err;
701 }
702 
703 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
704 			   int *addr_len, int peer)
705 {
706 	struct sock *sk = sock->sk;
707 	struct netlink_sock *nlk = nlk_sk(sk);
708 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
709 
710 	nladdr->nl_family = AF_NETLINK;
711 	nladdr->nl_pad = 0;
712 	*addr_len = sizeof(*nladdr);
713 
714 	if (peer) {
715 		nladdr->nl_pid = nlk->dst_pid;
716 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
717 	} else {
718 		nladdr->nl_pid = nlk->pid;
719 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
720 	}
721 	return 0;
722 }
723 
724 static void netlink_overrun(struct sock *sk)
725 {
726 	struct netlink_sock *nlk = nlk_sk(sk);
727 
728 	if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
729 		if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
730 			sk->sk_err = ENOBUFS;
731 			sk->sk_error_report(sk);
732 		}
733 	}
734 	atomic_inc(&sk->sk_drops);
735 }
736 
737 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
738 {
739 	struct sock *sock;
740 	struct netlink_sock *nlk;
741 
742 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
743 	if (!sock)
744 		return ERR_PTR(-ECONNREFUSED);
745 
746 	/* Don't bother queuing skb if kernel socket has no input function */
747 	nlk = nlk_sk(sock);
748 	if (sock->sk_state == NETLINK_CONNECTED &&
749 	    nlk->dst_pid != nlk_sk(ssk)->pid) {
750 		sock_put(sock);
751 		return ERR_PTR(-ECONNREFUSED);
752 	}
753 	return sock;
754 }
755 
756 struct sock *netlink_getsockbyfilp(struct file *filp)
757 {
758 	struct inode *inode = filp->f_path.dentry->d_inode;
759 	struct sock *sock;
760 
761 	if (!S_ISSOCK(inode->i_mode))
762 		return ERR_PTR(-ENOTSOCK);
763 
764 	sock = SOCKET_I(inode)->sk;
765 	if (sock->sk_family != AF_NETLINK)
766 		return ERR_PTR(-EINVAL);
767 
768 	sock_hold(sock);
769 	return sock;
770 }
771 
772 /*
773  * Attach a skb to a netlink socket.
774  * The caller must hold a reference to the destination socket. On error, the
775  * reference is dropped. The skb is not send to the destination, just all
776  * all error checks are performed and memory in the queue is reserved.
777  * Return values:
778  * < 0: error. skb freed, reference to sock dropped.
779  * 0: continue
780  * 1: repeat lookup - reference dropped while waiting for socket memory.
781  */
782 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
783 		      long *timeo, struct sock *ssk)
784 {
785 	struct netlink_sock *nlk;
786 
787 	nlk = nlk_sk(sk);
788 
789 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
790 	    test_bit(0, &nlk->state)) {
791 		DECLARE_WAITQUEUE(wait, current);
792 		if (!*timeo) {
793 			if (!ssk || netlink_is_kernel(ssk))
794 				netlink_overrun(sk);
795 			sock_put(sk);
796 			kfree_skb(skb);
797 			return -EAGAIN;
798 		}
799 
800 		__set_current_state(TASK_INTERRUPTIBLE);
801 		add_wait_queue(&nlk->wait, &wait);
802 
803 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
804 		     test_bit(0, &nlk->state)) &&
805 		    !sock_flag(sk, SOCK_DEAD))
806 			*timeo = schedule_timeout(*timeo);
807 
808 		__set_current_state(TASK_RUNNING);
809 		remove_wait_queue(&nlk->wait, &wait);
810 		sock_put(sk);
811 
812 		if (signal_pending(current)) {
813 			kfree_skb(skb);
814 			return sock_intr_errno(*timeo);
815 		}
816 		return 1;
817 	}
818 	skb_set_owner_r(skb, sk);
819 	return 0;
820 }
821 
822 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
823 {
824 	int len = skb->len;
825 
826 	skb_queue_tail(&sk->sk_receive_queue, skb);
827 	sk->sk_data_ready(sk, len);
828 	sock_put(sk);
829 	return len;
830 }
831 
832 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
833 {
834 	kfree_skb(skb);
835 	sock_put(sk);
836 }
837 
838 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
839 					   gfp_t allocation)
840 {
841 	int delta;
842 
843 	skb_orphan(skb);
844 
845 	delta = skb->end - skb->tail;
846 	if (delta * 2 < skb->truesize)
847 		return skb;
848 
849 	if (skb_shared(skb)) {
850 		struct sk_buff *nskb = skb_clone(skb, allocation);
851 		if (!nskb)
852 			return skb;
853 		kfree_skb(skb);
854 		skb = nskb;
855 	}
856 
857 	if (!pskb_expand_head(skb, 0, -delta, allocation))
858 		skb->truesize -= delta;
859 
860 	return skb;
861 }
862 
863 static inline void netlink_rcv_wake(struct sock *sk)
864 {
865 	struct netlink_sock *nlk = nlk_sk(sk);
866 
867 	if (skb_queue_empty(&sk->sk_receive_queue))
868 		clear_bit(0, &nlk->state);
869 	if (!test_bit(0, &nlk->state))
870 		wake_up_interruptible(&nlk->wait);
871 }
872 
873 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
874 {
875 	int ret;
876 	struct netlink_sock *nlk = nlk_sk(sk);
877 
878 	ret = -ECONNREFUSED;
879 	if (nlk->netlink_rcv != NULL) {
880 		ret = skb->len;
881 		skb_set_owner_r(skb, sk);
882 		nlk->netlink_rcv(skb);
883 	}
884 	kfree_skb(skb);
885 	sock_put(sk);
886 	return ret;
887 }
888 
889 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
890 		    u32 pid, int nonblock)
891 {
892 	struct sock *sk;
893 	int err;
894 	long timeo;
895 
896 	skb = netlink_trim(skb, gfp_any());
897 
898 	timeo = sock_sndtimeo(ssk, nonblock);
899 retry:
900 	sk = netlink_getsockbypid(ssk, pid);
901 	if (IS_ERR(sk)) {
902 		kfree_skb(skb);
903 		return PTR_ERR(sk);
904 	}
905 	if (netlink_is_kernel(sk))
906 		return netlink_unicast_kernel(sk, skb);
907 
908 	if (sk_filter(sk, skb)) {
909 		err = skb->len;
910 		kfree_skb(skb);
911 		sock_put(sk);
912 		return err;
913 	}
914 
915 	err = netlink_attachskb(sk, skb, &timeo, ssk);
916 	if (err == 1)
917 		goto retry;
918 	if (err)
919 		return err;
920 
921 	return netlink_sendskb(sk, skb);
922 }
923 EXPORT_SYMBOL(netlink_unicast);
924 
925 int netlink_has_listeners(struct sock *sk, unsigned int group)
926 {
927 	int res = 0;
928 	unsigned long *listeners;
929 
930 	BUG_ON(!netlink_is_kernel(sk));
931 
932 	rcu_read_lock();
933 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
934 
935 	if (group - 1 < nl_table[sk->sk_protocol].groups)
936 		res = test_bit(group - 1, listeners);
937 
938 	rcu_read_unlock();
939 
940 	return res;
941 }
942 EXPORT_SYMBOL_GPL(netlink_has_listeners);
943 
944 static inline int netlink_broadcast_deliver(struct sock *sk,
945 					    struct sk_buff *skb)
946 {
947 	struct netlink_sock *nlk = nlk_sk(sk);
948 
949 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
950 	    !test_bit(0, &nlk->state)) {
951 		skb_set_owner_r(skb, sk);
952 		skb_queue_tail(&sk->sk_receive_queue, skb);
953 		sk->sk_data_ready(sk, skb->len);
954 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
955 	}
956 	return -1;
957 }
958 
959 struct netlink_broadcast_data {
960 	struct sock *exclude_sk;
961 	struct net *net;
962 	u32 pid;
963 	u32 group;
964 	int failure;
965 	int delivery_failure;
966 	int congested;
967 	int delivered;
968 	gfp_t allocation;
969 	struct sk_buff *skb, *skb2;
970 };
971 
972 static inline int do_one_broadcast(struct sock *sk,
973 				   struct netlink_broadcast_data *p)
974 {
975 	struct netlink_sock *nlk = nlk_sk(sk);
976 	int val;
977 
978 	if (p->exclude_sk == sk)
979 		goto out;
980 
981 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
982 	    !test_bit(p->group - 1, nlk->groups))
983 		goto out;
984 
985 	if (!net_eq(sock_net(sk), p->net))
986 		goto out;
987 
988 	if (p->failure) {
989 		netlink_overrun(sk);
990 		goto out;
991 	}
992 
993 	sock_hold(sk);
994 	if (p->skb2 == NULL) {
995 		if (skb_shared(p->skb)) {
996 			p->skb2 = skb_clone(p->skb, p->allocation);
997 		} else {
998 			p->skb2 = skb_get(p->skb);
999 			/*
1000 			 * skb ownership may have been set when
1001 			 * delivered to a previous socket.
1002 			 */
1003 			skb_orphan(p->skb2);
1004 		}
1005 	}
1006 	if (p->skb2 == NULL) {
1007 		netlink_overrun(sk);
1008 		/* Clone failed. Notify ALL listeners. */
1009 		p->failure = 1;
1010 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1011 			p->delivery_failure = 1;
1012 	} else if (sk_filter(sk, p->skb2)) {
1013 		kfree_skb(p->skb2);
1014 		p->skb2 = NULL;
1015 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1016 		netlink_overrun(sk);
1017 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1018 			p->delivery_failure = 1;
1019 	} else {
1020 		p->congested |= val;
1021 		p->delivered = 1;
1022 		p->skb2 = NULL;
1023 	}
1024 	sock_put(sk);
1025 
1026 out:
1027 	return 0;
1028 }
1029 
1030 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1031 		      u32 group, gfp_t allocation)
1032 {
1033 	struct net *net = sock_net(ssk);
1034 	struct netlink_broadcast_data info;
1035 	struct hlist_node *node;
1036 	struct sock *sk;
1037 
1038 	skb = netlink_trim(skb, allocation);
1039 
1040 	info.exclude_sk = ssk;
1041 	info.net = net;
1042 	info.pid = pid;
1043 	info.group = group;
1044 	info.failure = 0;
1045 	info.delivery_failure = 0;
1046 	info.congested = 0;
1047 	info.delivered = 0;
1048 	info.allocation = allocation;
1049 	info.skb = skb;
1050 	info.skb2 = NULL;
1051 
1052 	/* While we sleep in clone, do not allow to change socket list */
1053 
1054 	netlink_lock_table();
1055 
1056 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1057 		do_one_broadcast(sk, &info);
1058 
1059 	kfree_skb(skb);
1060 
1061 	netlink_unlock_table();
1062 
1063 	kfree_skb(info.skb2);
1064 
1065 	if (info.delivery_failure)
1066 		return -ENOBUFS;
1067 
1068 	if (info.delivered) {
1069 		if (info.congested && (allocation & __GFP_WAIT))
1070 			yield();
1071 		return 0;
1072 	}
1073 	return -ESRCH;
1074 }
1075 EXPORT_SYMBOL(netlink_broadcast);
1076 
1077 struct netlink_set_err_data {
1078 	struct sock *exclude_sk;
1079 	u32 pid;
1080 	u32 group;
1081 	int code;
1082 };
1083 
1084 static inline int do_one_set_err(struct sock *sk,
1085 				 struct netlink_set_err_data *p)
1086 {
1087 	struct netlink_sock *nlk = nlk_sk(sk);
1088 
1089 	if (sk == p->exclude_sk)
1090 		goto out;
1091 
1092 	if (sock_net(sk) != sock_net(p->exclude_sk))
1093 		goto out;
1094 
1095 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1096 	    !test_bit(p->group - 1, nlk->groups))
1097 		goto out;
1098 
1099 	sk->sk_err = p->code;
1100 	sk->sk_error_report(sk);
1101 out:
1102 	return 0;
1103 }
1104 
1105 /**
1106  * netlink_set_err - report error to broadcast listeners
1107  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1108  * @pid: the PID of a process that we want to skip (if any)
1109  * @groups: the broadcast group that will notice the error
1110  * @code: error code, must be negative (as usual in kernelspace)
1111  */
1112 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1113 {
1114 	struct netlink_set_err_data info;
1115 	struct hlist_node *node;
1116 	struct sock *sk;
1117 
1118 	info.exclude_sk = ssk;
1119 	info.pid = pid;
1120 	info.group = group;
1121 	/* sk->sk_err wants a positive error value */
1122 	info.code = -code;
1123 
1124 	read_lock(&nl_table_lock);
1125 
1126 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1127 		do_one_set_err(sk, &info);
1128 
1129 	read_unlock(&nl_table_lock);
1130 }
1131 EXPORT_SYMBOL(netlink_set_err);
1132 
1133 /* must be called with netlink table grabbed */
1134 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1135 				     unsigned int group,
1136 				     int is_new)
1137 {
1138 	int old, new = !!is_new, subscriptions;
1139 
1140 	old = test_bit(group - 1, nlk->groups);
1141 	subscriptions = nlk->subscriptions - old + new;
1142 	if (new)
1143 		__set_bit(group - 1, nlk->groups);
1144 	else
1145 		__clear_bit(group - 1, nlk->groups);
1146 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1147 	netlink_update_listeners(&nlk->sk);
1148 }
1149 
1150 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1151 			      char __user *optval, int optlen)
1152 {
1153 	struct sock *sk = sock->sk;
1154 	struct netlink_sock *nlk = nlk_sk(sk);
1155 	unsigned int val = 0;
1156 	int err;
1157 
1158 	if (level != SOL_NETLINK)
1159 		return -ENOPROTOOPT;
1160 
1161 	if (optlen >= sizeof(int) &&
1162 	    get_user(val, (unsigned int __user *)optval))
1163 		return -EFAULT;
1164 
1165 	switch (optname) {
1166 	case NETLINK_PKTINFO:
1167 		if (val)
1168 			nlk->flags |= NETLINK_RECV_PKTINFO;
1169 		else
1170 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1171 		err = 0;
1172 		break;
1173 	case NETLINK_ADD_MEMBERSHIP:
1174 	case NETLINK_DROP_MEMBERSHIP: {
1175 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1176 			return -EPERM;
1177 		err = netlink_realloc_groups(sk);
1178 		if (err)
1179 			return err;
1180 		if (!val || val - 1 >= nlk->ngroups)
1181 			return -EINVAL;
1182 		netlink_table_grab();
1183 		netlink_update_socket_mc(nlk, val,
1184 					 optname == NETLINK_ADD_MEMBERSHIP);
1185 		netlink_table_ungrab();
1186 		err = 0;
1187 		break;
1188 	}
1189 	case NETLINK_BROADCAST_ERROR:
1190 		if (val)
1191 			nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1192 		else
1193 			nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1194 		err = 0;
1195 		break;
1196 	case NETLINK_NO_ENOBUFS:
1197 		if (val) {
1198 			nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1199 			clear_bit(0, &nlk->state);
1200 			wake_up_interruptible(&nlk->wait);
1201 		} else
1202 			nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1203 		err = 0;
1204 		break;
1205 	default:
1206 		err = -ENOPROTOOPT;
1207 	}
1208 	return err;
1209 }
1210 
1211 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1212 			      char __user *optval, int __user *optlen)
1213 {
1214 	struct sock *sk = sock->sk;
1215 	struct netlink_sock *nlk = nlk_sk(sk);
1216 	int len, val, err;
1217 
1218 	if (level != SOL_NETLINK)
1219 		return -ENOPROTOOPT;
1220 
1221 	if (get_user(len, optlen))
1222 		return -EFAULT;
1223 	if (len < 0)
1224 		return -EINVAL;
1225 
1226 	switch (optname) {
1227 	case NETLINK_PKTINFO:
1228 		if (len < sizeof(int))
1229 			return -EINVAL;
1230 		len = sizeof(int);
1231 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1232 		if (put_user(len, optlen) ||
1233 		    put_user(val, optval))
1234 			return -EFAULT;
1235 		err = 0;
1236 		break;
1237 	case NETLINK_BROADCAST_ERROR:
1238 		if (len < sizeof(int))
1239 			return -EINVAL;
1240 		len = sizeof(int);
1241 		val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1242 		if (put_user(len, optlen) ||
1243 		    put_user(val, optval))
1244 			return -EFAULT;
1245 		err = 0;
1246 		break;
1247 	case NETLINK_NO_ENOBUFS:
1248 		if (len < sizeof(int))
1249 			return -EINVAL;
1250 		len = sizeof(int);
1251 		val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1252 		if (put_user(len, optlen) ||
1253 		    put_user(val, optval))
1254 			return -EFAULT;
1255 		err = 0;
1256 		break;
1257 	default:
1258 		err = -ENOPROTOOPT;
1259 	}
1260 	return err;
1261 }
1262 
1263 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1264 {
1265 	struct nl_pktinfo info;
1266 
1267 	info.group = NETLINK_CB(skb).dst_group;
1268 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1269 }
1270 
1271 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1272 			   struct msghdr *msg, size_t len)
1273 {
1274 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1275 	struct sock *sk = sock->sk;
1276 	struct netlink_sock *nlk = nlk_sk(sk);
1277 	struct sockaddr_nl *addr = msg->msg_name;
1278 	u32 dst_pid;
1279 	u32 dst_group;
1280 	struct sk_buff *skb;
1281 	int err;
1282 	struct scm_cookie scm;
1283 
1284 	if (msg->msg_flags&MSG_OOB)
1285 		return -EOPNOTSUPP;
1286 
1287 	if (NULL == siocb->scm)
1288 		siocb->scm = &scm;
1289 	err = scm_send(sock, msg, siocb->scm);
1290 	if (err < 0)
1291 		return err;
1292 
1293 	if (msg->msg_namelen) {
1294 		if (addr->nl_family != AF_NETLINK)
1295 			return -EINVAL;
1296 		dst_pid = addr->nl_pid;
1297 		dst_group = ffs(addr->nl_groups);
1298 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1299 			return -EPERM;
1300 	} else {
1301 		dst_pid = nlk->dst_pid;
1302 		dst_group = nlk->dst_group;
1303 	}
1304 
1305 	if (!nlk->pid) {
1306 		err = netlink_autobind(sock);
1307 		if (err)
1308 			goto out;
1309 	}
1310 
1311 	err = -EMSGSIZE;
1312 	if (len > sk->sk_sndbuf - 32)
1313 		goto out;
1314 	err = -ENOBUFS;
1315 	skb = alloc_skb(len, GFP_KERNEL);
1316 	if (skb == NULL)
1317 		goto out;
1318 
1319 	NETLINK_CB(skb).pid	= nlk->pid;
1320 	NETLINK_CB(skb).dst_group = dst_group;
1321 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1322 	NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1323 	security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1324 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1325 
1326 	/* What can I do? Netlink is asynchronous, so that
1327 	   we will have to save current capabilities to
1328 	   check them, when this message will be delivered
1329 	   to corresponding kernel module.   --ANK (980802)
1330 	 */
1331 
1332 	err = -EFAULT;
1333 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1334 		kfree_skb(skb);
1335 		goto out;
1336 	}
1337 
1338 	err = security_netlink_send(sk, skb);
1339 	if (err) {
1340 		kfree_skb(skb);
1341 		goto out;
1342 	}
1343 
1344 	if (dst_group) {
1345 		atomic_inc(&skb->users);
1346 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1347 	}
1348 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1349 
1350 out:
1351 	return err;
1352 }
1353 
1354 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1355 			   struct msghdr *msg, size_t len,
1356 			   int flags)
1357 {
1358 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1359 	struct scm_cookie scm;
1360 	struct sock *sk = sock->sk;
1361 	struct netlink_sock *nlk = nlk_sk(sk);
1362 	int noblock = flags&MSG_DONTWAIT;
1363 	size_t copied;
1364 	struct sk_buff *skb, *frag __maybe_unused = NULL;
1365 	int err;
1366 
1367 	if (flags&MSG_OOB)
1368 		return -EOPNOTSUPP;
1369 
1370 	copied = 0;
1371 
1372 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1373 	if (skb == NULL)
1374 		goto out;
1375 
1376 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1377 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1378 		bool need_compat = !!(flags & MSG_CMSG_COMPAT);
1379 
1380 		/*
1381 		 * If this skb has a frag_list, then here that means that
1382 		 * we will have to use the frag_list skb for compat tasks
1383 		 * and the regular skb for non-compat tasks.
1384 		 *
1385 		 * The skb might (and likely will) be cloned, so we can't
1386 		 * just reset frag_list and go on with things -- we need to
1387 		 * keep that. For the compat case that's easy -- simply get
1388 		 * a reference to the compat skb and free the regular one
1389 		 * including the frag. For the non-compat case, we need to
1390 		 * avoid sending the frag to the user -- so assign NULL but
1391 		 * restore it below before freeing the skb.
1392 		 */
1393 		if (need_compat) {
1394 			struct sk_buff *compskb = skb_shinfo(skb)->frag_list;
1395 			skb_get(compskb);
1396 			kfree_skb(skb);
1397 			skb = compskb;
1398 		} else {
1399 			frag = skb_shinfo(skb)->frag_list;
1400 			skb_shinfo(skb)->frag_list = NULL;
1401 		}
1402 	}
1403 #endif
1404 
1405 	msg->msg_namelen = 0;
1406 
1407 	copied = skb->len;
1408 	if (len < copied) {
1409 		msg->msg_flags |= MSG_TRUNC;
1410 		copied = len;
1411 	}
1412 
1413 	skb_reset_transport_header(skb);
1414 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1415 
1416 	if (msg->msg_name) {
1417 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1418 		addr->nl_family = AF_NETLINK;
1419 		addr->nl_pad    = 0;
1420 		addr->nl_pid	= NETLINK_CB(skb).pid;
1421 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1422 		msg->msg_namelen = sizeof(*addr);
1423 	}
1424 
1425 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1426 		netlink_cmsg_recv_pktinfo(msg, skb);
1427 
1428 	if (NULL == siocb->scm) {
1429 		memset(&scm, 0, sizeof(scm));
1430 		siocb->scm = &scm;
1431 	}
1432 	siocb->scm->creds = *NETLINK_CREDS(skb);
1433 	if (flags & MSG_TRUNC)
1434 		copied = skb->len;
1435 
1436 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1437 	skb_shinfo(skb)->frag_list = frag;
1438 #endif
1439 
1440 	skb_free_datagram(sk, skb);
1441 
1442 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1443 		netlink_dump(sk);
1444 
1445 	scm_recv(sock, msg, siocb->scm, flags);
1446 out:
1447 	netlink_rcv_wake(sk);
1448 	return err ? : copied;
1449 }
1450 
1451 static void netlink_data_ready(struct sock *sk, int len)
1452 {
1453 	BUG();
1454 }
1455 
1456 /*
1457  *	We export these functions to other modules. They provide a
1458  *	complete set of kernel non-blocking support for message
1459  *	queueing.
1460  */
1461 
1462 struct sock *
1463 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1464 		      void (*input)(struct sk_buff *skb),
1465 		      struct mutex *cb_mutex, struct module *module)
1466 {
1467 	struct socket *sock;
1468 	struct sock *sk;
1469 	struct netlink_sock *nlk;
1470 	unsigned long *listeners = NULL;
1471 
1472 	BUG_ON(!nl_table);
1473 
1474 	if (unit < 0 || unit >= MAX_LINKS)
1475 		return NULL;
1476 
1477 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1478 		return NULL;
1479 
1480 	/*
1481 	 * We have to just have a reference on the net from sk, but don't
1482 	 * get_net it. Besides, we cannot get and then put the net here.
1483 	 * So we create one inside init_net and the move it to net.
1484 	 */
1485 
1486 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1487 		goto out_sock_release_nosk;
1488 
1489 	sk = sock->sk;
1490 	sk_change_net(sk, net);
1491 
1492 	if (groups < 32)
1493 		groups = 32;
1494 
1495 	listeners = kzalloc(NLGRPSZ(groups) + sizeof(struct listeners_rcu_head),
1496 			    GFP_KERNEL);
1497 	if (!listeners)
1498 		goto out_sock_release;
1499 
1500 	sk->sk_data_ready = netlink_data_ready;
1501 	if (input)
1502 		nlk_sk(sk)->netlink_rcv = input;
1503 
1504 	if (netlink_insert(sk, net, 0))
1505 		goto out_sock_release;
1506 
1507 	nlk = nlk_sk(sk);
1508 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1509 
1510 	netlink_table_grab();
1511 	if (!nl_table[unit].registered) {
1512 		nl_table[unit].groups = groups;
1513 		nl_table[unit].listeners = listeners;
1514 		nl_table[unit].cb_mutex = cb_mutex;
1515 		nl_table[unit].module = module;
1516 		nl_table[unit].registered = 1;
1517 	} else {
1518 		kfree(listeners);
1519 		nl_table[unit].registered++;
1520 	}
1521 	netlink_table_ungrab();
1522 	return sk;
1523 
1524 out_sock_release:
1525 	kfree(listeners);
1526 	netlink_kernel_release(sk);
1527 	return NULL;
1528 
1529 out_sock_release_nosk:
1530 	sock_release(sock);
1531 	return NULL;
1532 }
1533 EXPORT_SYMBOL(netlink_kernel_create);
1534 
1535 
1536 void
1537 netlink_kernel_release(struct sock *sk)
1538 {
1539 	sk_release_kernel(sk);
1540 }
1541 EXPORT_SYMBOL(netlink_kernel_release);
1542 
1543 
1544 static void netlink_free_old_listeners(struct rcu_head *rcu_head)
1545 {
1546 	struct listeners_rcu_head *lrh;
1547 
1548 	lrh = container_of(rcu_head, struct listeners_rcu_head, rcu_head);
1549 	kfree(lrh->ptr);
1550 }
1551 
1552 /**
1553  * netlink_change_ngroups - change number of multicast groups
1554  *
1555  * This changes the number of multicast groups that are available
1556  * on a certain netlink family. Note that it is not possible to
1557  * change the number of groups to below 32. Also note that it does
1558  * not implicitly call netlink_clear_multicast_users() when the
1559  * number of groups is reduced.
1560  *
1561  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1562  * @groups: The new number of groups.
1563  */
1564 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1565 {
1566 	unsigned long *listeners, *old = NULL;
1567 	struct listeners_rcu_head *old_rcu_head;
1568 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1569 	int err = 0;
1570 
1571 	if (groups < 32)
1572 		groups = 32;
1573 
1574 	netlink_table_grab();
1575 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1576 		listeners = kzalloc(NLGRPSZ(groups) +
1577 				    sizeof(struct listeners_rcu_head),
1578 				    GFP_ATOMIC);
1579 		if (!listeners) {
1580 			err = -ENOMEM;
1581 			goto out_ungrab;
1582 		}
1583 		old = tbl->listeners;
1584 		memcpy(listeners, old, NLGRPSZ(tbl->groups));
1585 		rcu_assign_pointer(tbl->listeners, listeners);
1586 		/*
1587 		 * Free the old memory after an RCU grace period so we
1588 		 * don't leak it. We use call_rcu() here in order to be
1589 		 * able to call this function from atomic contexts. The
1590 		 * allocation of this memory will have reserved enough
1591 		 * space for struct listeners_rcu_head at the end.
1592 		 */
1593 		old_rcu_head = (void *)(tbl->listeners +
1594 					NLGRPLONGS(tbl->groups));
1595 		old_rcu_head->ptr = old;
1596 		call_rcu(&old_rcu_head->rcu_head, netlink_free_old_listeners);
1597 	}
1598 	tbl->groups = groups;
1599 
1600  out_ungrab:
1601 	netlink_table_ungrab();
1602 	return err;
1603 }
1604 
1605 /**
1606  * netlink_clear_multicast_users - kick off multicast listeners
1607  *
1608  * This function removes all listeners from the given group.
1609  * @ksk: The kernel netlink socket, as returned by
1610  *	netlink_kernel_create().
1611  * @group: The multicast group to clear.
1612  */
1613 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1614 {
1615 	struct sock *sk;
1616 	struct hlist_node *node;
1617 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1618 
1619 	netlink_table_grab();
1620 
1621 	sk_for_each_bound(sk, node, &tbl->mc_list)
1622 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1623 
1624 	netlink_table_ungrab();
1625 }
1626 
1627 void netlink_set_nonroot(int protocol, unsigned int flags)
1628 {
1629 	if ((unsigned int)protocol < MAX_LINKS)
1630 		nl_table[protocol].nl_nonroot = flags;
1631 }
1632 EXPORT_SYMBOL(netlink_set_nonroot);
1633 
1634 static void netlink_destroy_callback(struct netlink_callback *cb)
1635 {
1636 	kfree_skb(cb->skb);
1637 	kfree(cb);
1638 }
1639 
1640 /*
1641  * It looks a bit ugly.
1642  * It would be better to create kernel thread.
1643  */
1644 
1645 static int netlink_dump(struct sock *sk)
1646 {
1647 	struct netlink_sock *nlk = nlk_sk(sk);
1648 	struct netlink_callback *cb;
1649 	struct sk_buff *skb;
1650 	struct nlmsghdr *nlh;
1651 	int len, err = -ENOBUFS;
1652 
1653 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1654 	if (!skb)
1655 		goto errout;
1656 
1657 	mutex_lock(nlk->cb_mutex);
1658 
1659 	cb = nlk->cb;
1660 	if (cb == NULL) {
1661 		err = -EINVAL;
1662 		goto errout_skb;
1663 	}
1664 
1665 	len = cb->dump(skb, cb);
1666 
1667 	if (len > 0) {
1668 		mutex_unlock(nlk->cb_mutex);
1669 
1670 		if (sk_filter(sk, skb))
1671 			kfree_skb(skb);
1672 		else {
1673 			skb_queue_tail(&sk->sk_receive_queue, skb);
1674 			sk->sk_data_ready(sk, skb->len);
1675 		}
1676 		return 0;
1677 	}
1678 
1679 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1680 	if (!nlh)
1681 		goto errout_skb;
1682 
1683 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1684 
1685 	if (sk_filter(sk, skb))
1686 		kfree_skb(skb);
1687 	else {
1688 		skb_queue_tail(&sk->sk_receive_queue, skb);
1689 		sk->sk_data_ready(sk, skb->len);
1690 	}
1691 
1692 	if (cb->done)
1693 		cb->done(cb);
1694 	nlk->cb = NULL;
1695 	mutex_unlock(nlk->cb_mutex);
1696 
1697 	netlink_destroy_callback(cb);
1698 	return 0;
1699 
1700 errout_skb:
1701 	mutex_unlock(nlk->cb_mutex);
1702 	kfree_skb(skb);
1703 errout:
1704 	return err;
1705 }
1706 
1707 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1708 		       const struct nlmsghdr *nlh,
1709 		       int (*dump)(struct sk_buff *skb,
1710 				   struct netlink_callback *),
1711 		       int (*done)(struct netlink_callback *))
1712 {
1713 	struct netlink_callback *cb;
1714 	struct sock *sk;
1715 	struct netlink_sock *nlk;
1716 
1717 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1718 	if (cb == NULL)
1719 		return -ENOBUFS;
1720 
1721 	cb->dump = dump;
1722 	cb->done = done;
1723 	cb->nlh = nlh;
1724 	atomic_inc(&skb->users);
1725 	cb->skb = skb;
1726 
1727 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1728 	if (sk == NULL) {
1729 		netlink_destroy_callback(cb);
1730 		return -ECONNREFUSED;
1731 	}
1732 	nlk = nlk_sk(sk);
1733 	/* A dump is in progress... */
1734 	mutex_lock(nlk->cb_mutex);
1735 	if (nlk->cb) {
1736 		mutex_unlock(nlk->cb_mutex);
1737 		netlink_destroy_callback(cb);
1738 		sock_put(sk);
1739 		return -EBUSY;
1740 	}
1741 	nlk->cb = cb;
1742 	mutex_unlock(nlk->cb_mutex);
1743 
1744 	netlink_dump(sk);
1745 	sock_put(sk);
1746 
1747 	/* We successfully started a dump, by returning -EINTR we
1748 	 * signal not to send ACK even if it was requested.
1749 	 */
1750 	return -EINTR;
1751 }
1752 EXPORT_SYMBOL(netlink_dump_start);
1753 
1754 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1755 {
1756 	struct sk_buff *skb;
1757 	struct nlmsghdr *rep;
1758 	struct nlmsgerr *errmsg;
1759 	size_t payload = sizeof(*errmsg);
1760 
1761 	/* error messages get the original request appened */
1762 	if (err)
1763 		payload += nlmsg_len(nlh);
1764 
1765 	skb = nlmsg_new(payload, GFP_KERNEL);
1766 	if (!skb) {
1767 		struct sock *sk;
1768 
1769 		sk = netlink_lookup(sock_net(in_skb->sk),
1770 				    in_skb->sk->sk_protocol,
1771 				    NETLINK_CB(in_skb).pid);
1772 		if (sk) {
1773 			sk->sk_err = ENOBUFS;
1774 			sk->sk_error_report(sk);
1775 			sock_put(sk);
1776 		}
1777 		return;
1778 	}
1779 
1780 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1781 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1782 	errmsg = nlmsg_data(rep);
1783 	errmsg->error = err;
1784 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1785 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1786 }
1787 EXPORT_SYMBOL(netlink_ack);
1788 
1789 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1790 						     struct nlmsghdr *))
1791 {
1792 	struct nlmsghdr *nlh;
1793 	int err;
1794 
1795 	while (skb->len >= nlmsg_total_size(0)) {
1796 		int msglen;
1797 
1798 		nlh = nlmsg_hdr(skb);
1799 		err = 0;
1800 
1801 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1802 			return 0;
1803 
1804 		/* Only requests are handled by the kernel */
1805 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1806 			goto ack;
1807 
1808 		/* Skip control messages */
1809 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1810 			goto ack;
1811 
1812 		err = cb(skb, nlh);
1813 		if (err == -EINTR)
1814 			goto skip;
1815 
1816 ack:
1817 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1818 			netlink_ack(skb, nlh, err);
1819 
1820 skip:
1821 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1822 		if (msglen > skb->len)
1823 			msglen = skb->len;
1824 		skb_pull(skb, msglen);
1825 	}
1826 
1827 	return 0;
1828 }
1829 EXPORT_SYMBOL(netlink_rcv_skb);
1830 
1831 /**
1832  * nlmsg_notify - send a notification netlink message
1833  * @sk: netlink socket to use
1834  * @skb: notification message
1835  * @pid: destination netlink pid for reports or 0
1836  * @group: destination multicast group or 0
1837  * @report: 1 to report back, 0 to disable
1838  * @flags: allocation flags
1839  */
1840 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1841 		 unsigned int group, int report, gfp_t flags)
1842 {
1843 	int err = 0;
1844 
1845 	if (group) {
1846 		int exclude_pid = 0;
1847 
1848 		if (report) {
1849 			atomic_inc(&skb->users);
1850 			exclude_pid = pid;
1851 		}
1852 
1853 		/* errors reported via destination sk->sk_err, but propagate
1854 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1855 		err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1856 	}
1857 
1858 	if (report) {
1859 		int err2;
1860 
1861 		err2 = nlmsg_unicast(sk, skb, pid);
1862 		if (!err || err == -ESRCH)
1863 			err = err2;
1864 	}
1865 
1866 	return err;
1867 }
1868 EXPORT_SYMBOL(nlmsg_notify);
1869 
1870 #ifdef CONFIG_PROC_FS
1871 struct nl_seq_iter {
1872 	struct seq_net_private p;
1873 	int link;
1874 	int hash_idx;
1875 };
1876 
1877 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1878 {
1879 	struct nl_seq_iter *iter = seq->private;
1880 	int i, j;
1881 	struct sock *s;
1882 	struct hlist_node *node;
1883 	loff_t off = 0;
1884 
1885 	for (i = 0; i < MAX_LINKS; i++) {
1886 		struct nl_pid_hash *hash = &nl_table[i].hash;
1887 
1888 		for (j = 0; j <= hash->mask; j++) {
1889 			sk_for_each(s, node, &hash->table[j]) {
1890 				if (sock_net(s) != seq_file_net(seq))
1891 					continue;
1892 				if (off == pos) {
1893 					iter->link = i;
1894 					iter->hash_idx = j;
1895 					return s;
1896 				}
1897 				++off;
1898 			}
1899 		}
1900 	}
1901 	return NULL;
1902 }
1903 
1904 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1905 	__acquires(nl_table_lock)
1906 {
1907 	read_lock(&nl_table_lock);
1908 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1909 }
1910 
1911 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1912 {
1913 	struct sock *s;
1914 	struct nl_seq_iter *iter;
1915 	int i, j;
1916 
1917 	++*pos;
1918 
1919 	if (v == SEQ_START_TOKEN)
1920 		return netlink_seq_socket_idx(seq, 0);
1921 
1922 	iter = seq->private;
1923 	s = v;
1924 	do {
1925 		s = sk_next(s);
1926 	} while (s && sock_net(s) != seq_file_net(seq));
1927 	if (s)
1928 		return s;
1929 
1930 	i = iter->link;
1931 	j = iter->hash_idx + 1;
1932 
1933 	do {
1934 		struct nl_pid_hash *hash = &nl_table[i].hash;
1935 
1936 		for (; j <= hash->mask; j++) {
1937 			s = sk_head(&hash->table[j]);
1938 			while (s && sock_net(s) != seq_file_net(seq))
1939 				s = sk_next(s);
1940 			if (s) {
1941 				iter->link = i;
1942 				iter->hash_idx = j;
1943 				return s;
1944 			}
1945 		}
1946 
1947 		j = 0;
1948 	} while (++i < MAX_LINKS);
1949 
1950 	return NULL;
1951 }
1952 
1953 static void netlink_seq_stop(struct seq_file *seq, void *v)
1954 	__releases(nl_table_lock)
1955 {
1956 	read_unlock(&nl_table_lock);
1957 }
1958 
1959 
1960 static int netlink_seq_show(struct seq_file *seq, void *v)
1961 {
1962 	if (v == SEQ_START_TOKEN)
1963 		seq_puts(seq,
1964 			 "sk       Eth Pid    Groups   "
1965 			 "Rmem     Wmem     Dump     Locks     Drops\n");
1966 	else {
1967 		struct sock *s = v;
1968 		struct netlink_sock *nlk = nlk_sk(s);
1969 
1970 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %-8d %-8d\n",
1971 			   s,
1972 			   s->sk_protocol,
1973 			   nlk->pid,
1974 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1975 			   sk_rmem_alloc_get(s),
1976 			   sk_wmem_alloc_get(s),
1977 			   nlk->cb,
1978 			   atomic_read(&s->sk_refcnt),
1979 			   atomic_read(&s->sk_drops)
1980 			);
1981 
1982 	}
1983 	return 0;
1984 }
1985 
1986 static const struct seq_operations netlink_seq_ops = {
1987 	.start  = netlink_seq_start,
1988 	.next   = netlink_seq_next,
1989 	.stop   = netlink_seq_stop,
1990 	.show   = netlink_seq_show,
1991 };
1992 
1993 
1994 static int netlink_seq_open(struct inode *inode, struct file *file)
1995 {
1996 	return seq_open_net(inode, file, &netlink_seq_ops,
1997 				sizeof(struct nl_seq_iter));
1998 }
1999 
2000 static const struct file_operations netlink_seq_fops = {
2001 	.owner		= THIS_MODULE,
2002 	.open		= netlink_seq_open,
2003 	.read		= seq_read,
2004 	.llseek		= seq_lseek,
2005 	.release	= seq_release_net,
2006 };
2007 
2008 #endif
2009 
2010 int netlink_register_notifier(struct notifier_block *nb)
2011 {
2012 	return atomic_notifier_chain_register(&netlink_chain, nb);
2013 }
2014 EXPORT_SYMBOL(netlink_register_notifier);
2015 
2016 int netlink_unregister_notifier(struct notifier_block *nb)
2017 {
2018 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
2019 }
2020 EXPORT_SYMBOL(netlink_unregister_notifier);
2021 
2022 static const struct proto_ops netlink_ops = {
2023 	.family =	PF_NETLINK,
2024 	.owner =	THIS_MODULE,
2025 	.release =	netlink_release,
2026 	.bind =		netlink_bind,
2027 	.connect =	netlink_connect,
2028 	.socketpair =	sock_no_socketpair,
2029 	.accept =	sock_no_accept,
2030 	.getname =	netlink_getname,
2031 	.poll =		datagram_poll,
2032 	.ioctl =	sock_no_ioctl,
2033 	.listen =	sock_no_listen,
2034 	.shutdown =	sock_no_shutdown,
2035 	.setsockopt =	netlink_setsockopt,
2036 	.getsockopt =	netlink_getsockopt,
2037 	.sendmsg =	netlink_sendmsg,
2038 	.recvmsg =	netlink_recvmsg,
2039 	.mmap =		sock_no_mmap,
2040 	.sendpage =	sock_no_sendpage,
2041 };
2042 
2043 static struct net_proto_family netlink_family_ops = {
2044 	.family = PF_NETLINK,
2045 	.create = netlink_create,
2046 	.owner	= THIS_MODULE,	/* for consistency 8) */
2047 };
2048 
2049 static int __net_init netlink_net_init(struct net *net)
2050 {
2051 #ifdef CONFIG_PROC_FS
2052 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2053 		return -ENOMEM;
2054 #endif
2055 	return 0;
2056 }
2057 
2058 static void __net_exit netlink_net_exit(struct net *net)
2059 {
2060 #ifdef CONFIG_PROC_FS
2061 	proc_net_remove(net, "netlink");
2062 #endif
2063 }
2064 
2065 static struct pernet_operations __net_initdata netlink_net_ops = {
2066 	.init = netlink_net_init,
2067 	.exit = netlink_net_exit,
2068 };
2069 
2070 static int __init netlink_proto_init(void)
2071 {
2072 	struct sk_buff *dummy_skb;
2073 	int i;
2074 	unsigned long limit;
2075 	unsigned int order;
2076 	int err = proto_register(&netlink_proto, 0);
2077 
2078 	if (err != 0)
2079 		goto out;
2080 
2081 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2082 
2083 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2084 	if (!nl_table)
2085 		goto panic;
2086 
2087 	if (num_physpages >= (128 * 1024))
2088 		limit = num_physpages >> (21 - PAGE_SHIFT);
2089 	else
2090 		limit = num_physpages >> (23 - PAGE_SHIFT);
2091 
2092 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2093 	limit = (1UL << order) / sizeof(struct hlist_head);
2094 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2095 
2096 	for (i = 0; i < MAX_LINKS; i++) {
2097 		struct nl_pid_hash *hash = &nl_table[i].hash;
2098 
2099 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2100 		if (!hash->table) {
2101 			while (i-- > 0)
2102 				nl_pid_hash_free(nl_table[i].hash.table,
2103 						 1 * sizeof(*hash->table));
2104 			kfree(nl_table);
2105 			goto panic;
2106 		}
2107 		hash->max_shift = order;
2108 		hash->shift = 0;
2109 		hash->mask = 0;
2110 		hash->rehash_time = jiffies;
2111 	}
2112 
2113 	sock_register(&netlink_family_ops);
2114 	register_pernet_subsys(&netlink_net_ops);
2115 	/* The netlink device handler may be needed early. */
2116 	rtnetlink_init();
2117 out:
2118 	return err;
2119 panic:
2120 	panic("netlink_init: Cannot allocate nl_table\n");
2121 }
2122 
2123 core_initcall(netlink_proto_init);
2124