xref: /openbmc/linux/net/netlink/af_netlink.c (revision 4800cd83)
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 {
87 	struct rcu_head		rcu;
88 	unsigned long		masks[0];
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 	struct listeners __rcu *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->masks[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 = task_tgid_vnr(current);
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 	struct listeners *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->masks);
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 	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
982 	void *tx_data;
983 };
984 
985 static inline int do_one_broadcast(struct sock *sk,
986 				   struct netlink_broadcast_data *p)
987 {
988 	struct netlink_sock *nlk = nlk_sk(sk);
989 	int val;
990 
991 	if (p->exclude_sk == sk)
992 		goto out;
993 
994 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
995 	    !test_bit(p->group - 1, nlk->groups))
996 		goto out;
997 
998 	if (!net_eq(sock_net(sk), p->net))
999 		goto out;
1000 
1001 	if (p->failure) {
1002 		netlink_overrun(sk);
1003 		goto out;
1004 	}
1005 
1006 	sock_hold(sk);
1007 	if (p->skb2 == NULL) {
1008 		if (skb_shared(p->skb)) {
1009 			p->skb2 = skb_clone(p->skb, p->allocation);
1010 		} else {
1011 			p->skb2 = skb_get(p->skb);
1012 			/*
1013 			 * skb ownership may have been set when
1014 			 * delivered to a previous socket.
1015 			 */
1016 			skb_orphan(p->skb2);
1017 		}
1018 	}
1019 	if (p->skb2 == NULL) {
1020 		netlink_overrun(sk);
1021 		/* Clone failed. Notify ALL listeners. */
1022 		p->failure = 1;
1023 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1024 			p->delivery_failure = 1;
1025 	} else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1026 		kfree_skb(p->skb2);
1027 		p->skb2 = NULL;
1028 	} else if (sk_filter(sk, p->skb2)) {
1029 		kfree_skb(p->skb2);
1030 		p->skb2 = NULL;
1031 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1032 		netlink_overrun(sk);
1033 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1034 			p->delivery_failure = 1;
1035 	} else {
1036 		p->congested |= val;
1037 		p->delivered = 1;
1038 		p->skb2 = NULL;
1039 	}
1040 	sock_put(sk);
1041 
1042 out:
1043 	return 0;
1044 }
1045 
1046 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1047 	u32 group, gfp_t allocation,
1048 	int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1049 	void *filter_data)
1050 {
1051 	struct net *net = sock_net(ssk);
1052 	struct netlink_broadcast_data info;
1053 	struct hlist_node *node;
1054 	struct sock *sk;
1055 
1056 	skb = netlink_trim(skb, allocation);
1057 
1058 	info.exclude_sk = ssk;
1059 	info.net = net;
1060 	info.pid = pid;
1061 	info.group = group;
1062 	info.failure = 0;
1063 	info.delivery_failure = 0;
1064 	info.congested = 0;
1065 	info.delivered = 0;
1066 	info.allocation = allocation;
1067 	info.skb = skb;
1068 	info.skb2 = NULL;
1069 	info.tx_filter = filter;
1070 	info.tx_data = filter_data;
1071 
1072 	/* While we sleep in clone, do not allow to change socket list */
1073 
1074 	netlink_lock_table();
1075 
1076 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1077 		do_one_broadcast(sk, &info);
1078 
1079 	consume_skb(skb);
1080 
1081 	netlink_unlock_table();
1082 
1083 	if (info.delivery_failure) {
1084 		kfree_skb(info.skb2);
1085 		return -ENOBUFS;
1086 	} else
1087 		consume_skb(info.skb2);
1088 
1089 	if (info.delivered) {
1090 		if (info.congested && (allocation & __GFP_WAIT))
1091 			yield();
1092 		return 0;
1093 	}
1094 	return -ESRCH;
1095 }
1096 EXPORT_SYMBOL(netlink_broadcast_filtered);
1097 
1098 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1099 		      u32 group, gfp_t allocation)
1100 {
1101 	return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1102 		NULL, NULL);
1103 }
1104 EXPORT_SYMBOL(netlink_broadcast);
1105 
1106 struct netlink_set_err_data {
1107 	struct sock *exclude_sk;
1108 	u32 pid;
1109 	u32 group;
1110 	int code;
1111 };
1112 
1113 static inline int do_one_set_err(struct sock *sk,
1114 				 struct netlink_set_err_data *p)
1115 {
1116 	struct netlink_sock *nlk = nlk_sk(sk);
1117 	int ret = 0;
1118 
1119 	if (sk == p->exclude_sk)
1120 		goto out;
1121 
1122 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1123 		goto out;
1124 
1125 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1126 	    !test_bit(p->group - 1, nlk->groups))
1127 		goto out;
1128 
1129 	if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1130 		ret = 1;
1131 		goto out;
1132 	}
1133 
1134 	sk->sk_err = p->code;
1135 	sk->sk_error_report(sk);
1136 out:
1137 	return ret;
1138 }
1139 
1140 /**
1141  * netlink_set_err - report error to broadcast listeners
1142  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1143  * @pid: the PID of a process that we want to skip (if any)
1144  * @groups: the broadcast group that will notice the error
1145  * @code: error code, must be negative (as usual in kernelspace)
1146  *
1147  * This function returns the number of broadcast listeners that have set the
1148  * NETLINK_RECV_NO_ENOBUFS socket option.
1149  */
1150 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1151 {
1152 	struct netlink_set_err_data info;
1153 	struct hlist_node *node;
1154 	struct sock *sk;
1155 	int ret = 0;
1156 
1157 	info.exclude_sk = ssk;
1158 	info.pid = pid;
1159 	info.group = group;
1160 	/* sk->sk_err wants a positive error value */
1161 	info.code = -code;
1162 
1163 	read_lock(&nl_table_lock);
1164 
1165 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1166 		ret += do_one_set_err(sk, &info);
1167 
1168 	read_unlock(&nl_table_lock);
1169 	return ret;
1170 }
1171 EXPORT_SYMBOL(netlink_set_err);
1172 
1173 /* must be called with netlink table grabbed */
1174 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1175 				     unsigned int group,
1176 				     int is_new)
1177 {
1178 	int old, new = !!is_new, subscriptions;
1179 
1180 	old = test_bit(group - 1, nlk->groups);
1181 	subscriptions = nlk->subscriptions - old + new;
1182 	if (new)
1183 		__set_bit(group - 1, nlk->groups);
1184 	else
1185 		__clear_bit(group - 1, nlk->groups);
1186 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1187 	netlink_update_listeners(&nlk->sk);
1188 }
1189 
1190 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1191 			      char __user *optval, unsigned int optlen)
1192 {
1193 	struct sock *sk = sock->sk;
1194 	struct netlink_sock *nlk = nlk_sk(sk);
1195 	unsigned int val = 0;
1196 	int err;
1197 
1198 	if (level != SOL_NETLINK)
1199 		return -ENOPROTOOPT;
1200 
1201 	if (optlen >= sizeof(int) &&
1202 	    get_user(val, (unsigned int __user *)optval))
1203 		return -EFAULT;
1204 
1205 	switch (optname) {
1206 	case NETLINK_PKTINFO:
1207 		if (val)
1208 			nlk->flags |= NETLINK_RECV_PKTINFO;
1209 		else
1210 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1211 		err = 0;
1212 		break;
1213 	case NETLINK_ADD_MEMBERSHIP:
1214 	case NETLINK_DROP_MEMBERSHIP: {
1215 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1216 			return -EPERM;
1217 		err = netlink_realloc_groups(sk);
1218 		if (err)
1219 			return err;
1220 		if (!val || val - 1 >= nlk->ngroups)
1221 			return -EINVAL;
1222 		netlink_table_grab();
1223 		netlink_update_socket_mc(nlk, val,
1224 					 optname == NETLINK_ADD_MEMBERSHIP);
1225 		netlink_table_ungrab();
1226 		err = 0;
1227 		break;
1228 	}
1229 	case NETLINK_BROADCAST_ERROR:
1230 		if (val)
1231 			nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1232 		else
1233 			nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1234 		err = 0;
1235 		break;
1236 	case NETLINK_NO_ENOBUFS:
1237 		if (val) {
1238 			nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1239 			clear_bit(0, &nlk->state);
1240 			wake_up_interruptible(&nlk->wait);
1241 		} else
1242 			nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1243 		err = 0;
1244 		break;
1245 	default:
1246 		err = -ENOPROTOOPT;
1247 	}
1248 	return err;
1249 }
1250 
1251 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1252 			      char __user *optval, int __user *optlen)
1253 {
1254 	struct sock *sk = sock->sk;
1255 	struct netlink_sock *nlk = nlk_sk(sk);
1256 	int len, val, err;
1257 
1258 	if (level != SOL_NETLINK)
1259 		return -ENOPROTOOPT;
1260 
1261 	if (get_user(len, optlen))
1262 		return -EFAULT;
1263 	if (len < 0)
1264 		return -EINVAL;
1265 
1266 	switch (optname) {
1267 	case NETLINK_PKTINFO:
1268 		if (len < sizeof(int))
1269 			return -EINVAL;
1270 		len = sizeof(int);
1271 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1272 		if (put_user(len, optlen) ||
1273 		    put_user(val, optval))
1274 			return -EFAULT;
1275 		err = 0;
1276 		break;
1277 	case NETLINK_BROADCAST_ERROR:
1278 		if (len < sizeof(int))
1279 			return -EINVAL;
1280 		len = sizeof(int);
1281 		val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1282 		if (put_user(len, optlen) ||
1283 		    put_user(val, optval))
1284 			return -EFAULT;
1285 		err = 0;
1286 		break;
1287 	case NETLINK_NO_ENOBUFS:
1288 		if (len < sizeof(int))
1289 			return -EINVAL;
1290 		len = sizeof(int);
1291 		val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1292 		if (put_user(len, optlen) ||
1293 		    put_user(val, optval))
1294 			return -EFAULT;
1295 		err = 0;
1296 		break;
1297 	default:
1298 		err = -ENOPROTOOPT;
1299 	}
1300 	return err;
1301 }
1302 
1303 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1304 {
1305 	struct nl_pktinfo info;
1306 
1307 	info.group = NETLINK_CB(skb).dst_group;
1308 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1309 }
1310 
1311 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1312 			   struct msghdr *msg, size_t len)
1313 {
1314 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1315 	struct sock *sk = sock->sk;
1316 	struct netlink_sock *nlk = nlk_sk(sk);
1317 	struct sockaddr_nl *addr = msg->msg_name;
1318 	u32 dst_pid;
1319 	u32 dst_group;
1320 	struct sk_buff *skb;
1321 	int err;
1322 	struct scm_cookie scm;
1323 
1324 	if (msg->msg_flags&MSG_OOB)
1325 		return -EOPNOTSUPP;
1326 
1327 	if (NULL == siocb->scm) {
1328 		siocb->scm = &scm;
1329 		memset(&scm, 0, sizeof(scm));
1330 	}
1331 	err = scm_send(sock, msg, siocb->scm);
1332 	if (err < 0)
1333 		return err;
1334 
1335 	if (msg->msg_namelen) {
1336 		err = -EINVAL;
1337 		if (addr->nl_family != AF_NETLINK)
1338 			goto out;
1339 		dst_pid = addr->nl_pid;
1340 		dst_group = ffs(addr->nl_groups);
1341 		err =  -EPERM;
1342 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1343 			goto out;
1344 	} else {
1345 		dst_pid = nlk->dst_pid;
1346 		dst_group = nlk->dst_group;
1347 	}
1348 
1349 	if (!nlk->pid) {
1350 		err = netlink_autobind(sock);
1351 		if (err)
1352 			goto out;
1353 	}
1354 
1355 	err = -EMSGSIZE;
1356 	if (len > sk->sk_sndbuf - 32)
1357 		goto out;
1358 	err = -ENOBUFS;
1359 	skb = alloc_skb(len, GFP_KERNEL);
1360 	if (skb == NULL)
1361 		goto out;
1362 
1363 	NETLINK_CB(skb).pid	= nlk->pid;
1364 	NETLINK_CB(skb).dst_group = dst_group;
1365 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1366 	NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1367 	security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1368 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1369 
1370 	/* What can I do? Netlink is asynchronous, so that
1371 	   we will have to save current capabilities to
1372 	   check them, when this message will be delivered
1373 	   to corresponding kernel module.   --ANK (980802)
1374 	 */
1375 
1376 	err = -EFAULT;
1377 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1378 		kfree_skb(skb);
1379 		goto out;
1380 	}
1381 
1382 	err = security_netlink_send(sk, skb);
1383 	if (err) {
1384 		kfree_skb(skb);
1385 		goto out;
1386 	}
1387 
1388 	if (dst_group) {
1389 		atomic_inc(&skb->users);
1390 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1391 	}
1392 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1393 
1394 out:
1395 	scm_destroy(siocb->scm);
1396 	return err;
1397 }
1398 
1399 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1400 			   struct msghdr *msg, size_t len,
1401 			   int flags)
1402 {
1403 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1404 	struct scm_cookie scm;
1405 	struct sock *sk = sock->sk;
1406 	struct netlink_sock *nlk = nlk_sk(sk);
1407 	int noblock = flags&MSG_DONTWAIT;
1408 	size_t copied;
1409 	struct sk_buff *skb, *data_skb;
1410 	int err, ret;
1411 
1412 	if (flags&MSG_OOB)
1413 		return -EOPNOTSUPP;
1414 
1415 	copied = 0;
1416 
1417 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1418 	if (skb == NULL)
1419 		goto out;
1420 
1421 	data_skb = skb;
1422 
1423 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1424 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1425 		/*
1426 		 * If this skb has a frag_list, then here that means that we
1427 		 * will have to use the frag_list skb's data for compat tasks
1428 		 * and the regular skb's data for normal (non-compat) tasks.
1429 		 *
1430 		 * If we need to send the compat skb, assign it to the
1431 		 * 'data_skb' variable so that it will be used below for data
1432 		 * copying. We keep 'skb' for everything else, including
1433 		 * freeing both later.
1434 		 */
1435 		if (flags & MSG_CMSG_COMPAT)
1436 			data_skb = skb_shinfo(skb)->frag_list;
1437 	}
1438 #endif
1439 
1440 	msg->msg_namelen = 0;
1441 
1442 	copied = data_skb->len;
1443 	if (len < copied) {
1444 		msg->msg_flags |= MSG_TRUNC;
1445 		copied = len;
1446 	}
1447 
1448 	skb_reset_transport_header(data_skb);
1449 	err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1450 
1451 	if (msg->msg_name) {
1452 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1453 		addr->nl_family = AF_NETLINK;
1454 		addr->nl_pad    = 0;
1455 		addr->nl_pid	= NETLINK_CB(skb).pid;
1456 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1457 		msg->msg_namelen = sizeof(*addr);
1458 	}
1459 
1460 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1461 		netlink_cmsg_recv_pktinfo(msg, skb);
1462 
1463 	if (NULL == siocb->scm) {
1464 		memset(&scm, 0, sizeof(scm));
1465 		siocb->scm = &scm;
1466 	}
1467 	siocb->scm->creds = *NETLINK_CREDS(skb);
1468 	if (flags & MSG_TRUNC)
1469 		copied = data_skb->len;
1470 
1471 	skb_free_datagram(sk, skb);
1472 
1473 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1474 		ret = netlink_dump(sk);
1475 		if (ret) {
1476 			sk->sk_err = ret;
1477 			sk->sk_error_report(sk);
1478 		}
1479 	}
1480 
1481 	scm_recv(sock, msg, siocb->scm, flags);
1482 out:
1483 	netlink_rcv_wake(sk);
1484 	return err ? : copied;
1485 }
1486 
1487 static void netlink_data_ready(struct sock *sk, int len)
1488 {
1489 	BUG();
1490 }
1491 
1492 /*
1493  *	We export these functions to other modules. They provide a
1494  *	complete set of kernel non-blocking support for message
1495  *	queueing.
1496  */
1497 
1498 struct sock *
1499 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1500 		      void (*input)(struct sk_buff *skb),
1501 		      struct mutex *cb_mutex, struct module *module)
1502 {
1503 	struct socket *sock;
1504 	struct sock *sk;
1505 	struct netlink_sock *nlk;
1506 	struct listeners *listeners = NULL;
1507 
1508 	BUG_ON(!nl_table);
1509 
1510 	if (unit < 0 || unit >= MAX_LINKS)
1511 		return NULL;
1512 
1513 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1514 		return NULL;
1515 
1516 	/*
1517 	 * We have to just have a reference on the net from sk, but don't
1518 	 * get_net it. Besides, we cannot get and then put the net here.
1519 	 * So we create one inside init_net and the move it to net.
1520 	 */
1521 
1522 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1523 		goto out_sock_release_nosk;
1524 
1525 	sk = sock->sk;
1526 	sk_change_net(sk, net);
1527 
1528 	if (groups < 32)
1529 		groups = 32;
1530 
1531 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1532 	if (!listeners)
1533 		goto out_sock_release;
1534 
1535 	sk->sk_data_ready = netlink_data_ready;
1536 	if (input)
1537 		nlk_sk(sk)->netlink_rcv = input;
1538 
1539 	if (netlink_insert(sk, net, 0))
1540 		goto out_sock_release;
1541 
1542 	nlk = nlk_sk(sk);
1543 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1544 
1545 	netlink_table_grab();
1546 	if (!nl_table[unit].registered) {
1547 		nl_table[unit].groups = groups;
1548 		rcu_assign_pointer(nl_table[unit].listeners, listeners);
1549 		nl_table[unit].cb_mutex = cb_mutex;
1550 		nl_table[unit].module = module;
1551 		nl_table[unit].registered = 1;
1552 	} else {
1553 		kfree(listeners);
1554 		nl_table[unit].registered++;
1555 	}
1556 	netlink_table_ungrab();
1557 	return sk;
1558 
1559 out_sock_release:
1560 	kfree(listeners);
1561 	netlink_kernel_release(sk);
1562 	return NULL;
1563 
1564 out_sock_release_nosk:
1565 	sock_release(sock);
1566 	return NULL;
1567 }
1568 EXPORT_SYMBOL(netlink_kernel_create);
1569 
1570 
1571 void
1572 netlink_kernel_release(struct sock *sk)
1573 {
1574 	sk_release_kernel(sk);
1575 }
1576 EXPORT_SYMBOL(netlink_kernel_release);
1577 
1578 
1579 static void listeners_free_rcu(struct rcu_head *head)
1580 {
1581 	kfree(container_of(head, struct listeners, rcu));
1582 }
1583 
1584 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1585 {
1586 	struct listeners *new, *old;
1587 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1588 
1589 	if (groups < 32)
1590 		groups = 32;
1591 
1592 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1593 		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1594 		if (!new)
1595 			return -ENOMEM;
1596 		old = rcu_dereference_raw(tbl->listeners);
1597 		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1598 		rcu_assign_pointer(tbl->listeners, new);
1599 
1600 		call_rcu(&old->rcu, listeners_free_rcu);
1601 	}
1602 	tbl->groups = groups;
1603 
1604 	return 0;
1605 }
1606 
1607 /**
1608  * netlink_change_ngroups - change number of multicast groups
1609  *
1610  * This changes the number of multicast groups that are available
1611  * on a certain netlink family. Note that it is not possible to
1612  * change the number of groups to below 32. Also note that it does
1613  * not implicitly call netlink_clear_multicast_users() when the
1614  * number of groups is reduced.
1615  *
1616  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1617  * @groups: The new number of groups.
1618  */
1619 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1620 {
1621 	int err;
1622 
1623 	netlink_table_grab();
1624 	err = __netlink_change_ngroups(sk, groups);
1625 	netlink_table_ungrab();
1626 
1627 	return err;
1628 }
1629 
1630 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1631 {
1632 	struct sock *sk;
1633 	struct hlist_node *node;
1634 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1635 
1636 	sk_for_each_bound(sk, node, &tbl->mc_list)
1637 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1638 }
1639 
1640 /**
1641  * netlink_clear_multicast_users - kick off multicast listeners
1642  *
1643  * This function removes all listeners from the given group.
1644  * @ksk: The kernel netlink socket, as returned by
1645  *	netlink_kernel_create().
1646  * @group: The multicast group to clear.
1647  */
1648 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1649 {
1650 	netlink_table_grab();
1651 	__netlink_clear_multicast_users(ksk, group);
1652 	netlink_table_ungrab();
1653 }
1654 
1655 void netlink_set_nonroot(int protocol, unsigned int flags)
1656 {
1657 	if ((unsigned int)protocol < MAX_LINKS)
1658 		nl_table[protocol].nl_nonroot = flags;
1659 }
1660 EXPORT_SYMBOL(netlink_set_nonroot);
1661 
1662 static void netlink_destroy_callback(struct netlink_callback *cb)
1663 {
1664 	kfree_skb(cb->skb);
1665 	kfree(cb);
1666 }
1667 
1668 /*
1669  * It looks a bit ugly.
1670  * It would be better to create kernel thread.
1671  */
1672 
1673 static int netlink_dump(struct sock *sk)
1674 {
1675 	struct netlink_sock *nlk = nlk_sk(sk);
1676 	struct netlink_callback *cb;
1677 	struct sk_buff *skb;
1678 	struct nlmsghdr *nlh;
1679 	int len, err = -ENOBUFS;
1680 
1681 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1682 	if (!skb)
1683 		goto errout;
1684 
1685 	mutex_lock(nlk->cb_mutex);
1686 
1687 	cb = nlk->cb;
1688 	if (cb == NULL) {
1689 		err = -EINVAL;
1690 		goto errout_skb;
1691 	}
1692 
1693 	len = cb->dump(skb, cb);
1694 
1695 	if (len > 0) {
1696 		mutex_unlock(nlk->cb_mutex);
1697 
1698 		if (sk_filter(sk, skb))
1699 			kfree_skb(skb);
1700 		else {
1701 			skb_queue_tail(&sk->sk_receive_queue, skb);
1702 			sk->sk_data_ready(sk, skb->len);
1703 		}
1704 		return 0;
1705 	}
1706 
1707 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1708 	if (!nlh)
1709 		goto errout_skb;
1710 
1711 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1712 
1713 	if (sk_filter(sk, skb))
1714 		kfree_skb(skb);
1715 	else {
1716 		skb_queue_tail(&sk->sk_receive_queue, skb);
1717 		sk->sk_data_ready(sk, skb->len);
1718 	}
1719 
1720 	if (cb->done)
1721 		cb->done(cb);
1722 	nlk->cb = NULL;
1723 	mutex_unlock(nlk->cb_mutex);
1724 
1725 	netlink_destroy_callback(cb);
1726 	return 0;
1727 
1728 errout_skb:
1729 	mutex_unlock(nlk->cb_mutex);
1730 	kfree_skb(skb);
1731 errout:
1732 	return err;
1733 }
1734 
1735 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1736 		       const struct nlmsghdr *nlh,
1737 		       int (*dump)(struct sk_buff *skb,
1738 				   struct netlink_callback *),
1739 		       int (*done)(struct netlink_callback *))
1740 {
1741 	struct netlink_callback *cb;
1742 	struct sock *sk;
1743 	struct netlink_sock *nlk;
1744 	int ret;
1745 
1746 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1747 	if (cb == NULL)
1748 		return -ENOBUFS;
1749 
1750 	cb->dump = dump;
1751 	cb->done = done;
1752 	cb->nlh = nlh;
1753 	atomic_inc(&skb->users);
1754 	cb->skb = skb;
1755 
1756 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1757 	if (sk == NULL) {
1758 		netlink_destroy_callback(cb);
1759 		return -ECONNREFUSED;
1760 	}
1761 	nlk = nlk_sk(sk);
1762 	/* A dump is in progress... */
1763 	mutex_lock(nlk->cb_mutex);
1764 	if (nlk->cb) {
1765 		mutex_unlock(nlk->cb_mutex);
1766 		netlink_destroy_callback(cb);
1767 		sock_put(sk);
1768 		return -EBUSY;
1769 	}
1770 	nlk->cb = cb;
1771 	mutex_unlock(nlk->cb_mutex);
1772 
1773 	ret = netlink_dump(sk);
1774 
1775 	sock_put(sk);
1776 
1777 	if (ret)
1778 		return ret;
1779 
1780 	/* We successfully started a dump, by returning -EINTR we
1781 	 * signal not to send ACK even if it was requested.
1782 	 */
1783 	return -EINTR;
1784 }
1785 EXPORT_SYMBOL(netlink_dump_start);
1786 
1787 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1788 {
1789 	struct sk_buff *skb;
1790 	struct nlmsghdr *rep;
1791 	struct nlmsgerr *errmsg;
1792 	size_t payload = sizeof(*errmsg);
1793 
1794 	/* error messages get the original request appened */
1795 	if (err)
1796 		payload += nlmsg_len(nlh);
1797 
1798 	skb = nlmsg_new(payload, GFP_KERNEL);
1799 	if (!skb) {
1800 		struct sock *sk;
1801 
1802 		sk = netlink_lookup(sock_net(in_skb->sk),
1803 				    in_skb->sk->sk_protocol,
1804 				    NETLINK_CB(in_skb).pid);
1805 		if (sk) {
1806 			sk->sk_err = ENOBUFS;
1807 			sk->sk_error_report(sk);
1808 			sock_put(sk);
1809 		}
1810 		return;
1811 	}
1812 
1813 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1814 			  NLMSG_ERROR, payload, 0);
1815 	errmsg = nlmsg_data(rep);
1816 	errmsg->error = err;
1817 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1818 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1819 }
1820 EXPORT_SYMBOL(netlink_ack);
1821 
1822 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1823 						     struct nlmsghdr *))
1824 {
1825 	struct nlmsghdr *nlh;
1826 	int err;
1827 
1828 	while (skb->len >= nlmsg_total_size(0)) {
1829 		int msglen;
1830 
1831 		nlh = nlmsg_hdr(skb);
1832 		err = 0;
1833 
1834 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1835 			return 0;
1836 
1837 		/* Only requests are handled by the kernel */
1838 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1839 			goto ack;
1840 
1841 		/* Skip control messages */
1842 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1843 			goto ack;
1844 
1845 		err = cb(skb, nlh);
1846 		if (err == -EINTR)
1847 			goto skip;
1848 
1849 ack:
1850 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1851 			netlink_ack(skb, nlh, err);
1852 
1853 skip:
1854 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1855 		if (msglen > skb->len)
1856 			msglen = skb->len;
1857 		skb_pull(skb, msglen);
1858 	}
1859 
1860 	return 0;
1861 }
1862 EXPORT_SYMBOL(netlink_rcv_skb);
1863 
1864 /**
1865  * nlmsg_notify - send a notification netlink message
1866  * @sk: netlink socket to use
1867  * @skb: notification message
1868  * @pid: destination netlink pid for reports or 0
1869  * @group: destination multicast group or 0
1870  * @report: 1 to report back, 0 to disable
1871  * @flags: allocation flags
1872  */
1873 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1874 		 unsigned int group, int report, gfp_t flags)
1875 {
1876 	int err = 0;
1877 
1878 	if (group) {
1879 		int exclude_pid = 0;
1880 
1881 		if (report) {
1882 			atomic_inc(&skb->users);
1883 			exclude_pid = pid;
1884 		}
1885 
1886 		/* errors reported via destination sk->sk_err, but propagate
1887 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1888 		err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1889 	}
1890 
1891 	if (report) {
1892 		int err2;
1893 
1894 		err2 = nlmsg_unicast(sk, skb, pid);
1895 		if (!err || err == -ESRCH)
1896 			err = err2;
1897 	}
1898 
1899 	return err;
1900 }
1901 EXPORT_SYMBOL(nlmsg_notify);
1902 
1903 #ifdef CONFIG_PROC_FS
1904 struct nl_seq_iter {
1905 	struct seq_net_private p;
1906 	int link;
1907 	int hash_idx;
1908 };
1909 
1910 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1911 {
1912 	struct nl_seq_iter *iter = seq->private;
1913 	int i, j;
1914 	struct sock *s;
1915 	struct hlist_node *node;
1916 	loff_t off = 0;
1917 
1918 	for (i = 0; i < MAX_LINKS; i++) {
1919 		struct nl_pid_hash *hash = &nl_table[i].hash;
1920 
1921 		for (j = 0; j <= hash->mask; j++) {
1922 			sk_for_each(s, node, &hash->table[j]) {
1923 				if (sock_net(s) != seq_file_net(seq))
1924 					continue;
1925 				if (off == pos) {
1926 					iter->link = i;
1927 					iter->hash_idx = j;
1928 					return s;
1929 				}
1930 				++off;
1931 			}
1932 		}
1933 	}
1934 	return NULL;
1935 }
1936 
1937 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1938 	__acquires(nl_table_lock)
1939 {
1940 	read_lock(&nl_table_lock);
1941 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1942 }
1943 
1944 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1945 {
1946 	struct sock *s;
1947 	struct nl_seq_iter *iter;
1948 	int i, j;
1949 
1950 	++*pos;
1951 
1952 	if (v == SEQ_START_TOKEN)
1953 		return netlink_seq_socket_idx(seq, 0);
1954 
1955 	iter = seq->private;
1956 	s = v;
1957 	do {
1958 		s = sk_next(s);
1959 	} while (s && sock_net(s) != seq_file_net(seq));
1960 	if (s)
1961 		return s;
1962 
1963 	i = iter->link;
1964 	j = iter->hash_idx + 1;
1965 
1966 	do {
1967 		struct nl_pid_hash *hash = &nl_table[i].hash;
1968 
1969 		for (; j <= hash->mask; j++) {
1970 			s = sk_head(&hash->table[j]);
1971 			while (s && sock_net(s) != seq_file_net(seq))
1972 				s = sk_next(s);
1973 			if (s) {
1974 				iter->link = i;
1975 				iter->hash_idx = j;
1976 				return s;
1977 			}
1978 		}
1979 
1980 		j = 0;
1981 	} while (++i < MAX_LINKS);
1982 
1983 	return NULL;
1984 }
1985 
1986 static void netlink_seq_stop(struct seq_file *seq, void *v)
1987 	__releases(nl_table_lock)
1988 {
1989 	read_unlock(&nl_table_lock);
1990 }
1991 
1992 
1993 static int netlink_seq_show(struct seq_file *seq, void *v)
1994 {
1995 	if (v == SEQ_START_TOKEN)
1996 		seq_puts(seq,
1997 			 "sk       Eth Pid    Groups   "
1998 			 "Rmem     Wmem     Dump     Locks     Drops     Inode\n");
1999 	else {
2000 		struct sock *s = v;
2001 		struct netlink_sock *nlk = nlk_sk(s);
2002 
2003 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %-8d %-8d %-8lu\n",
2004 			   s,
2005 			   s->sk_protocol,
2006 			   nlk->pid,
2007 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2008 			   sk_rmem_alloc_get(s),
2009 			   sk_wmem_alloc_get(s),
2010 			   nlk->cb,
2011 			   atomic_read(&s->sk_refcnt),
2012 			   atomic_read(&s->sk_drops),
2013 			   sock_i_ino(s)
2014 			);
2015 
2016 	}
2017 	return 0;
2018 }
2019 
2020 static const struct seq_operations netlink_seq_ops = {
2021 	.start  = netlink_seq_start,
2022 	.next   = netlink_seq_next,
2023 	.stop   = netlink_seq_stop,
2024 	.show   = netlink_seq_show,
2025 };
2026 
2027 
2028 static int netlink_seq_open(struct inode *inode, struct file *file)
2029 {
2030 	return seq_open_net(inode, file, &netlink_seq_ops,
2031 				sizeof(struct nl_seq_iter));
2032 }
2033 
2034 static const struct file_operations netlink_seq_fops = {
2035 	.owner		= THIS_MODULE,
2036 	.open		= netlink_seq_open,
2037 	.read		= seq_read,
2038 	.llseek		= seq_lseek,
2039 	.release	= seq_release_net,
2040 };
2041 
2042 #endif
2043 
2044 int netlink_register_notifier(struct notifier_block *nb)
2045 {
2046 	return atomic_notifier_chain_register(&netlink_chain, nb);
2047 }
2048 EXPORT_SYMBOL(netlink_register_notifier);
2049 
2050 int netlink_unregister_notifier(struct notifier_block *nb)
2051 {
2052 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
2053 }
2054 EXPORT_SYMBOL(netlink_unregister_notifier);
2055 
2056 static const struct proto_ops netlink_ops = {
2057 	.family =	PF_NETLINK,
2058 	.owner =	THIS_MODULE,
2059 	.release =	netlink_release,
2060 	.bind =		netlink_bind,
2061 	.connect =	netlink_connect,
2062 	.socketpair =	sock_no_socketpair,
2063 	.accept =	sock_no_accept,
2064 	.getname =	netlink_getname,
2065 	.poll =		datagram_poll,
2066 	.ioctl =	sock_no_ioctl,
2067 	.listen =	sock_no_listen,
2068 	.shutdown =	sock_no_shutdown,
2069 	.setsockopt =	netlink_setsockopt,
2070 	.getsockopt =	netlink_getsockopt,
2071 	.sendmsg =	netlink_sendmsg,
2072 	.recvmsg =	netlink_recvmsg,
2073 	.mmap =		sock_no_mmap,
2074 	.sendpage =	sock_no_sendpage,
2075 };
2076 
2077 static const struct net_proto_family netlink_family_ops = {
2078 	.family = PF_NETLINK,
2079 	.create = netlink_create,
2080 	.owner	= THIS_MODULE,	/* for consistency 8) */
2081 };
2082 
2083 static int __net_init netlink_net_init(struct net *net)
2084 {
2085 #ifdef CONFIG_PROC_FS
2086 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2087 		return -ENOMEM;
2088 #endif
2089 	return 0;
2090 }
2091 
2092 static void __net_exit netlink_net_exit(struct net *net)
2093 {
2094 #ifdef CONFIG_PROC_FS
2095 	proc_net_remove(net, "netlink");
2096 #endif
2097 }
2098 
2099 static void __init netlink_add_usersock_entry(void)
2100 {
2101 	struct listeners *listeners;
2102 	int groups = 32;
2103 
2104 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2105 	if (!listeners)
2106 		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2107 
2108 	netlink_table_grab();
2109 
2110 	nl_table[NETLINK_USERSOCK].groups = groups;
2111 	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2112 	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2113 	nl_table[NETLINK_USERSOCK].registered = 1;
2114 
2115 	netlink_table_ungrab();
2116 }
2117 
2118 static struct pernet_operations __net_initdata netlink_net_ops = {
2119 	.init = netlink_net_init,
2120 	.exit = netlink_net_exit,
2121 };
2122 
2123 static int __init netlink_proto_init(void)
2124 {
2125 	struct sk_buff *dummy_skb;
2126 	int i;
2127 	unsigned long limit;
2128 	unsigned int order;
2129 	int err = proto_register(&netlink_proto, 0);
2130 
2131 	if (err != 0)
2132 		goto out;
2133 
2134 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2135 
2136 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2137 	if (!nl_table)
2138 		goto panic;
2139 
2140 	if (totalram_pages >= (128 * 1024))
2141 		limit = totalram_pages >> (21 - PAGE_SHIFT);
2142 	else
2143 		limit = totalram_pages >> (23 - PAGE_SHIFT);
2144 
2145 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2146 	limit = (1UL << order) / sizeof(struct hlist_head);
2147 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2148 
2149 	for (i = 0; i < MAX_LINKS; i++) {
2150 		struct nl_pid_hash *hash = &nl_table[i].hash;
2151 
2152 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2153 		if (!hash->table) {
2154 			while (i-- > 0)
2155 				nl_pid_hash_free(nl_table[i].hash.table,
2156 						 1 * sizeof(*hash->table));
2157 			kfree(nl_table);
2158 			goto panic;
2159 		}
2160 		hash->max_shift = order;
2161 		hash->shift = 0;
2162 		hash->mask = 0;
2163 		hash->rehash_time = jiffies;
2164 	}
2165 
2166 	netlink_add_usersock_entry();
2167 
2168 	sock_register(&netlink_family_ops);
2169 	register_pernet_subsys(&netlink_net_ops);
2170 	/* The netlink device handler may be needed early. */
2171 	rtnetlink_init();
2172 out:
2173 	return err;
2174 panic:
2175 	panic("netlink_init: Cannot allocate nl_table\n");
2176 }
2177 
2178 core_initcall(netlink_proto_init);
2179