xref: /linux/net/netlink/af_netlink.c (revision b3b77c8caef1750ebeea1054e39e358550ea9f55)
1 /*
2  * NETLINK      Kernel-user communication protocol.
3  *
4  * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
5  * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6  *
7  *		This program is free software; you can redistribute it and/or
8  *		modify it under the terms of the GNU General Public License
9  *		as published by the Free Software Foundation; either version
10  *		2 of the License, or (at your option) any later version.
11  *
12  * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13  *                               added netlink_proto_exit
14  * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15  * 				 use nlk_sk, as sk->protinfo is on a diet 8)
16  * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17  * 				 - inc module use count of module that owns
18  * 				   the kernel socket in case userspace opens
19  * 				   socket of same protocol
20  * 				 - remove all module support, since netlink is
21  * 				   mandatory if CONFIG_NET=y these days
22  */
23 
24 #include <linux/module.h>
25 
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/mutex.h>
58 
59 #include <net/net_namespace.h>
60 #include <net/sock.h>
61 #include <net/scm.h>
62 #include <net/netlink.h>
63 
64 #define NLGRPSZ(x)	(ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x)	(NLGRPSZ(x)/sizeof(unsigned long))
66 
67 struct netlink_sock {
68 	/* struct sock has to be the first member of netlink_sock */
69 	struct sock		sk;
70 	u32			pid;
71 	u32			dst_pid;
72 	u32			dst_group;
73 	u32			flags;
74 	u32			subscriptions;
75 	u32			ngroups;
76 	unsigned long		*groups;
77 	unsigned long		state;
78 	wait_queue_head_t	wait;
79 	struct netlink_callback	*cb;
80 	struct mutex		*cb_mutex;
81 	struct mutex		cb_def_mutex;
82 	void			(*netlink_rcv)(struct sk_buff *skb);
83 	struct module		*module;
84 };
85 
86 struct listeners_rcu_head {
87 	struct rcu_head rcu_head;
88 	void *ptr;
89 };
90 
91 #define NETLINK_KERNEL_SOCKET	0x1
92 #define NETLINK_RECV_PKTINFO	0x2
93 #define NETLINK_BROADCAST_SEND_ERROR	0x4
94 #define NETLINK_RECV_NO_ENOBUFS	0x8
95 
96 static inline struct netlink_sock *nlk_sk(struct sock *sk)
97 {
98 	return container_of(sk, struct netlink_sock, sk);
99 }
100 
101 static inline int netlink_is_kernel(struct sock *sk)
102 {
103 	return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
104 }
105 
106 struct nl_pid_hash {
107 	struct hlist_head *table;
108 	unsigned long rehash_time;
109 
110 	unsigned int mask;
111 	unsigned int shift;
112 
113 	unsigned int entries;
114 	unsigned int max_shift;
115 
116 	u32 rnd;
117 };
118 
119 struct netlink_table {
120 	struct nl_pid_hash hash;
121 	struct hlist_head mc_list;
122 	unsigned long *listeners;
123 	unsigned int nl_nonroot;
124 	unsigned int groups;
125 	struct mutex *cb_mutex;
126 	struct module *module;
127 	int registered;
128 };
129 
130 static struct netlink_table *nl_table;
131 
132 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
133 
134 static int netlink_dump(struct sock *sk);
135 static void netlink_destroy_callback(struct netlink_callback *cb);
136 
137 static DEFINE_RWLOCK(nl_table_lock);
138 static atomic_t nl_table_users = ATOMIC_INIT(0);
139 
140 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
141 
142 static u32 netlink_group_mask(u32 group)
143 {
144 	return group ? 1 << (group - 1) : 0;
145 }
146 
147 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
148 {
149 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
150 }
151 
152 static void netlink_sock_destruct(struct sock *sk)
153 {
154 	struct netlink_sock *nlk = nlk_sk(sk);
155 
156 	if (nlk->cb) {
157 		if (nlk->cb->done)
158 			nlk->cb->done(nlk->cb);
159 		netlink_destroy_callback(nlk->cb);
160 	}
161 
162 	skb_queue_purge(&sk->sk_receive_queue);
163 
164 	if (!sock_flag(sk, SOCK_DEAD)) {
165 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
166 		return;
167 	}
168 
169 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
170 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
171 	WARN_ON(nlk_sk(sk)->groups);
172 }
173 
174 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
175  * SMP. Look, when several writers sleep and reader wakes them up, all but one
176  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
177  * this, _but_ remember, it adds useless work on UP machines.
178  */
179 
180 void netlink_table_grab(void)
181 	__acquires(nl_table_lock)
182 {
183 	might_sleep();
184 
185 	write_lock_irq(&nl_table_lock);
186 
187 	if (atomic_read(&nl_table_users)) {
188 		DECLARE_WAITQUEUE(wait, current);
189 
190 		add_wait_queue_exclusive(&nl_table_wait, &wait);
191 		for (;;) {
192 			set_current_state(TASK_UNINTERRUPTIBLE);
193 			if (atomic_read(&nl_table_users) == 0)
194 				break;
195 			write_unlock_irq(&nl_table_lock);
196 			schedule();
197 			write_lock_irq(&nl_table_lock);
198 		}
199 
200 		__set_current_state(TASK_RUNNING);
201 		remove_wait_queue(&nl_table_wait, &wait);
202 	}
203 }
204 
205 void netlink_table_ungrab(void)
206 	__releases(nl_table_lock)
207 {
208 	write_unlock_irq(&nl_table_lock);
209 	wake_up(&nl_table_wait);
210 }
211 
212 static inline void
213 netlink_lock_table(void)
214 {
215 	/* read_lock() synchronizes us to netlink_table_grab */
216 
217 	read_lock(&nl_table_lock);
218 	atomic_inc(&nl_table_users);
219 	read_unlock(&nl_table_lock);
220 }
221 
222 static inline void
223 netlink_unlock_table(void)
224 {
225 	if (atomic_dec_and_test(&nl_table_users))
226 		wake_up(&nl_table_wait);
227 }
228 
229 static inline struct sock *netlink_lookup(struct net *net, int protocol,
230 					  u32 pid)
231 {
232 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
233 	struct hlist_head *head;
234 	struct sock *sk;
235 	struct hlist_node *node;
236 
237 	read_lock(&nl_table_lock);
238 	head = nl_pid_hashfn(hash, pid);
239 	sk_for_each(sk, node, head) {
240 		if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
241 			sock_hold(sk);
242 			goto found;
243 		}
244 	}
245 	sk = NULL;
246 found:
247 	read_unlock(&nl_table_lock);
248 	return sk;
249 }
250 
251 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
252 {
253 	if (size <= PAGE_SIZE)
254 		return kzalloc(size, GFP_ATOMIC);
255 	else
256 		return (struct hlist_head *)
257 			__get_free_pages(GFP_ATOMIC | __GFP_ZERO,
258 					 get_order(size));
259 }
260 
261 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
262 {
263 	if (size <= PAGE_SIZE)
264 		kfree(table);
265 	else
266 		free_pages((unsigned long)table, get_order(size));
267 }
268 
269 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
270 {
271 	unsigned int omask, mask, shift;
272 	size_t osize, size;
273 	struct hlist_head *otable, *table;
274 	int i;
275 
276 	omask = mask = hash->mask;
277 	osize = size = (mask + 1) * sizeof(*table);
278 	shift = hash->shift;
279 
280 	if (grow) {
281 		if (++shift > hash->max_shift)
282 			return 0;
283 		mask = mask * 2 + 1;
284 		size *= 2;
285 	}
286 
287 	table = nl_pid_hash_zalloc(size);
288 	if (!table)
289 		return 0;
290 
291 	otable = hash->table;
292 	hash->table = table;
293 	hash->mask = mask;
294 	hash->shift = shift;
295 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
296 
297 	for (i = 0; i <= omask; i++) {
298 		struct sock *sk;
299 		struct hlist_node *node, *tmp;
300 
301 		sk_for_each_safe(sk, node, tmp, &otable[i])
302 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
303 	}
304 
305 	nl_pid_hash_free(otable, osize);
306 	hash->rehash_time = jiffies + 10 * 60 * HZ;
307 	return 1;
308 }
309 
310 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
311 {
312 	int avg = hash->entries >> hash->shift;
313 
314 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
315 		return 1;
316 
317 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
318 		nl_pid_hash_rehash(hash, 0);
319 		return 1;
320 	}
321 
322 	return 0;
323 }
324 
325 static const struct proto_ops netlink_ops;
326 
327 static void
328 netlink_update_listeners(struct sock *sk)
329 {
330 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
331 	struct hlist_node *node;
332 	unsigned long mask;
333 	unsigned int i;
334 
335 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
336 		mask = 0;
337 		sk_for_each_bound(sk, node, &tbl->mc_list) {
338 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
339 				mask |= nlk_sk(sk)->groups[i];
340 		}
341 		tbl->listeners[i] = mask;
342 	}
343 	/* this function is only called with the netlink table "grabbed", which
344 	 * makes sure updates are visible before bind or setsockopt return. */
345 }
346 
347 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
348 {
349 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
350 	struct hlist_head *head;
351 	int err = -EADDRINUSE;
352 	struct sock *osk;
353 	struct hlist_node *node;
354 	int len;
355 
356 	netlink_table_grab();
357 	head = nl_pid_hashfn(hash, pid);
358 	len = 0;
359 	sk_for_each(osk, node, head) {
360 		if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
361 			break;
362 		len++;
363 	}
364 	if (node)
365 		goto err;
366 
367 	err = -EBUSY;
368 	if (nlk_sk(sk)->pid)
369 		goto err;
370 
371 	err = -ENOMEM;
372 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
373 		goto err;
374 
375 	if (len && nl_pid_hash_dilute(hash, len))
376 		head = nl_pid_hashfn(hash, pid);
377 	hash->entries++;
378 	nlk_sk(sk)->pid = pid;
379 	sk_add_node(sk, head);
380 	err = 0;
381 
382 err:
383 	netlink_table_ungrab();
384 	return err;
385 }
386 
387 static void netlink_remove(struct sock *sk)
388 {
389 	netlink_table_grab();
390 	if (sk_del_node_init(sk))
391 		nl_table[sk->sk_protocol].hash.entries--;
392 	if (nlk_sk(sk)->subscriptions)
393 		__sk_del_bind_node(sk);
394 	netlink_table_ungrab();
395 }
396 
397 static struct proto netlink_proto = {
398 	.name	  = "NETLINK",
399 	.owner	  = THIS_MODULE,
400 	.obj_size = sizeof(struct netlink_sock),
401 };
402 
403 static int __netlink_create(struct net *net, struct socket *sock,
404 			    struct mutex *cb_mutex, int protocol)
405 {
406 	struct sock *sk;
407 	struct netlink_sock *nlk;
408 
409 	sock->ops = &netlink_ops;
410 
411 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
412 	if (!sk)
413 		return -ENOMEM;
414 
415 	sock_init_data(sock, sk);
416 
417 	nlk = nlk_sk(sk);
418 	if (cb_mutex)
419 		nlk->cb_mutex = cb_mutex;
420 	else {
421 		nlk->cb_mutex = &nlk->cb_def_mutex;
422 		mutex_init(nlk->cb_mutex);
423 	}
424 	init_waitqueue_head(&nlk->wait);
425 
426 	sk->sk_destruct = netlink_sock_destruct;
427 	sk->sk_protocol = protocol;
428 	return 0;
429 }
430 
431 static int netlink_create(struct net *net, struct socket *sock, int protocol,
432 			  int kern)
433 {
434 	struct module *module = NULL;
435 	struct mutex *cb_mutex;
436 	struct netlink_sock *nlk;
437 	int err = 0;
438 
439 	sock->state = SS_UNCONNECTED;
440 
441 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
442 		return -ESOCKTNOSUPPORT;
443 
444 	if (protocol < 0 || protocol >= MAX_LINKS)
445 		return -EPROTONOSUPPORT;
446 
447 	netlink_lock_table();
448 #ifdef CONFIG_MODULES
449 	if (!nl_table[protocol].registered) {
450 		netlink_unlock_table();
451 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
452 		netlink_lock_table();
453 	}
454 #endif
455 	if (nl_table[protocol].registered &&
456 	    try_module_get(nl_table[protocol].module))
457 		module = nl_table[protocol].module;
458 	else
459 		err = -EPROTONOSUPPORT;
460 	cb_mutex = nl_table[protocol].cb_mutex;
461 	netlink_unlock_table();
462 
463 	if (err < 0)
464 		goto out;
465 
466 	err = __netlink_create(net, sock, cb_mutex, protocol);
467 	if (err < 0)
468 		goto out_module;
469 
470 	local_bh_disable();
471 	sock_prot_inuse_add(net, &netlink_proto, 1);
472 	local_bh_enable();
473 
474 	nlk = nlk_sk(sock->sk);
475 	nlk->module = module;
476 out:
477 	return err;
478 
479 out_module:
480 	module_put(module);
481 	goto out;
482 }
483 
484 static int netlink_release(struct socket *sock)
485 {
486 	struct sock *sk = sock->sk;
487 	struct netlink_sock *nlk;
488 
489 	if (!sk)
490 		return 0;
491 
492 	netlink_remove(sk);
493 	sock_orphan(sk);
494 	nlk = nlk_sk(sk);
495 
496 	/*
497 	 * OK. Socket is unlinked, any packets that arrive now
498 	 * will be purged.
499 	 */
500 
501 	sock->sk = NULL;
502 	wake_up_interruptible_all(&nlk->wait);
503 
504 	skb_queue_purge(&sk->sk_write_queue);
505 
506 	if (nlk->pid) {
507 		struct netlink_notify n = {
508 						.net = sock_net(sk),
509 						.protocol = sk->sk_protocol,
510 						.pid = nlk->pid,
511 					  };
512 		atomic_notifier_call_chain(&netlink_chain,
513 				NETLINK_URELEASE, &n);
514 	}
515 
516 	module_put(nlk->module);
517 
518 	netlink_table_grab();
519 	if (netlink_is_kernel(sk)) {
520 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
521 		if (--nl_table[sk->sk_protocol].registered == 0) {
522 			kfree(nl_table[sk->sk_protocol].listeners);
523 			nl_table[sk->sk_protocol].module = NULL;
524 			nl_table[sk->sk_protocol].registered = 0;
525 		}
526 	} else if (nlk->subscriptions)
527 		netlink_update_listeners(sk);
528 	netlink_table_ungrab();
529 
530 	kfree(nlk->groups);
531 	nlk->groups = NULL;
532 
533 	local_bh_disable();
534 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
535 	local_bh_enable();
536 	sock_put(sk);
537 	return 0;
538 }
539 
540 static int netlink_autobind(struct socket *sock)
541 {
542 	struct sock *sk = sock->sk;
543 	struct net *net = sock_net(sk);
544 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
545 	struct hlist_head *head;
546 	struct sock *osk;
547 	struct hlist_node *node;
548 	s32 pid = 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 	unsigned long *listeners;
940 
941 	BUG_ON(!netlink_is_kernel(sk));
942 
943 	rcu_read_lock();
944 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
945 
946 	if (group - 1 < nl_table[sk->sk_protocol].groups)
947 		res = test_bit(group - 1, listeners);
948 
949 	rcu_read_unlock();
950 
951 	return res;
952 }
953 EXPORT_SYMBOL_GPL(netlink_has_listeners);
954 
955 static inline int netlink_broadcast_deliver(struct sock *sk,
956 					    struct sk_buff *skb)
957 {
958 	struct netlink_sock *nlk = nlk_sk(sk);
959 
960 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
961 	    !test_bit(0, &nlk->state)) {
962 		skb_set_owner_r(skb, sk);
963 		skb_queue_tail(&sk->sk_receive_queue, skb);
964 		sk->sk_data_ready(sk, skb->len);
965 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
966 	}
967 	return -1;
968 }
969 
970 struct netlink_broadcast_data {
971 	struct sock *exclude_sk;
972 	struct net *net;
973 	u32 pid;
974 	u32 group;
975 	int failure;
976 	int delivery_failure;
977 	int congested;
978 	int delivered;
979 	gfp_t allocation;
980 	struct sk_buff *skb, *skb2;
981 	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 	kfree_skb(skb);
1080 
1081 	netlink_unlock_table();
1082 
1083 	kfree_skb(info.skb2);
1084 
1085 	if (info.delivery_failure)
1086 		return -ENOBUFS;
1087 
1088 	if (info.delivered) {
1089 		if (info.congested && (allocation & __GFP_WAIT))
1090 			yield();
1091 		return 0;
1092 	}
1093 	return -ESRCH;
1094 }
1095 EXPORT_SYMBOL(netlink_broadcast_filtered);
1096 
1097 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1098 		      u32 group, gfp_t allocation)
1099 {
1100 	return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1101 		NULL, NULL);
1102 }
1103 EXPORT_SYMBOL(netlink_broadcast);
1104 
1105 struct netlink_set_err_data {
1106 	struct sock *exclude_sk;
1107 	u32 pid;
1108 	u32 group;
1109 	int code;
1110 };
1111 
1112 static inline int do_one_set_err(struct sock *sk,
1113 				 struct netlink_set_err_data *p)
1114 {
1115 	struct netlink_sock *nlk = nlk_sk(sk);
1116 	int ret = 0;
1117 
1118 	if (sk == p->exclude_sk)
1119 		goto out;
1120 
1121 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1122 		goto out;
1123 
1124 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1125 	    !test_bit(p->group - 1, nlk->groups))
1126 		goto out;
1127 
1128 	if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1129 		ret = 1;
1130 		goto out;
1131 	}
1132 
1133 	sk->sk_err = p->code;
1134 	sk->sk_error_report(sk);
1135 out:
1136 	return ret;
1137 }
1138 
1139 /**
1140  * netlink_set_err - report error to broadcast listeners
1141  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1142  * @pid: the PID of a process that we want to skip (if any)
1143  * @groups: the broadcast group that will notice the error
1144  * @code: error code, must be negative (as usual in kernelspace)
1145  *
1146  * This function returns the number of broadcast listeners that have set the
1147  * NETLINK_RECV_NO_ENOBUFS socket option.
1148  */
1149 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1150 {
1151 	struct netlink_set_err_data info;
1152 	struct hlist_node *node;
1153 	struct sock *sk;
1154 	int ret = 0;
1155 
1156 	info.exclude_sk = ssk;
1157 	info.pid = pid;
1158 	info.group = group;
1159 	/* sk->sk_err wants a positive error value */
1160 	info.code = -code;
1161 
1162 	read_lock(&nl_table_lock);
1163 
1164 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1165 		ret += do_one_set_err(sk, &info);
1166 
1167 	read_unlock(&nl_table_lock);
1168 	return ret;
1169 }
1170 EXPORT_SYMBOL(netlink_set_err);
1171 
1172 /* must be called with netlink table grabbed */
1173 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1174 				     unsigned int group,
1175 				     int is_new)
1176 {
1177 	int old, new = !!is_new, subscriptions;
1178 
1179 	old = test_bit(group - 1, nlk->groups);
1180 	subscriptions = nlk->subscriptions - old + new;
1181 	if (new)
1182 		__set_bit(group - 1, nlk->groups);
1183 	else
1184 		__clear_bit(group - 1, nlk->groups);
1185 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1186 	netlink_update_listeners(&nlk->sk);
1187 }
1188 
1189 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1190 			      char __user *optval, unsigned int optlen)
1191 {
1192 	struct sock *sk = sock->sk;
1193 	struct netlink_sock *nlk = nlk_sk(sk);
1194 	unsigned int val = 0;
1195 	int err;
1196 
1197 	if (level != SOL_NETLINK)
1198 		return -ENOPROTOOPT;
1199 
1200 	if (optlen >= sizeof(int) &&
1201 	    get_user(val, (unsigned int __user *)optval))
1202 		return -EFAULT;
1203 
1204 	switch (optname) {
1205 	case NETLINK_PKTINFO:
1206 		if (val)
1207 			nlk->flags |= NETLINK_RECV_PKTINFO;
1208 		else
1209 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1210 		err = 0;
1211 		break;
1212 	case NETLINK_ADD_MEMBERSHIP:
1213 	case NETLINK_DROP_MEMBERSHIP: {
1214 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1215 			return -EPERM;
1216 		err = netlink_realloc_groups(sk);
1217 		if (err)
1218 			return err;
1219 		if (!val || val - 1 >= nlk->ngroups)
1220 			return -EINVAL;
1221 		netlink_table_grab();
1222 		netlink_update_socket_mc(nlk, val,
1223 					 optname == NETLINK_ADD_MEMBERSHIP);
1224 		netlink_table_ungrab();
1225 		err = 0;
1226 		break;
1227 	}
1228 	case NETLINK_BROADCAST_ERROR:
1229 		if (val)
1230 			nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1231 		else
1232 			nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1233 		err = 0;
1234 		break;
1235 	case NETLINK_NO_ENOBUFS:
1236 		if (val) {
1237 			nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1238 			clear_bit(0, &nlk->state);
1239 			wake_up_interruptible(&nlk->wait);
1240 		} else
1241 			nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1242 		err = 0;
1243 		break;
1244 	default:
1245 		err = -ENOPROTOOPT;
1246 	}
1247 	return err;
1248 }
1249 
1250 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1251 			      char __user *optval, int __user *optlen)
1252 {
1253 	struct sock *sk = sock->sk;
1254 	struct netlink_sock *nlk = nlk_sk(sk);
1255 	int len, val, err;
1256 
1257 	if (level != SOL_NETLINK)
1258 		return -ENOPROTOOPT;
1259 
1260 	if (get_user(len, optlen))
1261 		return -EFAULT;
1262 	if (len < 0)
1263 		return -EINVAL;
1264 
1265 	switch (optname) {
1266 	case NETLINK_PKTINFO:
1267 		if (len < sizeof(int))
1268 			return -EINVAL;
1269 		len = sizeof(int);
1270 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1271 		if (put_user(len, optlen) ||
1272 		    put_user(val, optval))
1273 			return -EFAULT;
1274 		err = 0;
1275 		break;
1276 	case NETLINK_BROADCAST_ERROR:
1277 		if (len < sizeof(int))
1278 			return -EINVAL;
1279 		len = sizeof(int);
1280 		val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1281 		if (put_user(len, optlen) ||
1282 		    put_user(val, optval))
1283 			return -EFAULT;
1284 		err = 0;
1285 		break;
1286 	case NETLINK_NO_ENOBUFS:
1287 		if (len < sizeof(int))
1288 			return -EINVAL;
1289 		len = sizeof(int);
1290 		val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1291 		if (put_user(len, optlen) ||
1292 		    put_user(val, optval))
1293 			return -EFAULT;
1294 		err = 0;
1295 		break;
1296 	default:
1297 		err = -ENOPROTOOPT;
1298 	}
1299 	return err;
1300 }
1301 
1302 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1303 {
1304 	struct nl_pktinfo info;
1305 
1306 	info.group = NETLINK_CB(skb).dst_group;
1307 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1308 }
1309 
1310 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1311 			   struct msghdr *msg, size_t len)
1312 {
1313 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1314 	struct sock *sk = sock->sk;
1315 	struct netlink_sock *nlk = nlk_sk(sk);
1316 	struct sockaddr_nl *addr = msg->msg_name;
1317 	u32 dst_pid;
1318 	u32 dst_group;
1319 	struct sk_buff *skb;
1320 	int err;
1321 	struct scm_cookie scm;
1322 
1323 	if (msg->msg_flags&MSG_OOB)
1324 		return -EOPNOTSUPP;
1325 
1326 	if (NULL == siocb->scm)
1327 		siocb->scm = &scm;
1328 	err = scm_send(sock, msg, siocb->scm);
1329 	if (err < 0)
1330 		return err;
1331 
1332 	if (msg->msg_namelen) {
1333 		if (addr->nl_family != AF_NETLINK)
1334 			return -EINVAL;
1335 		dst_pid = addr->nl_pid;
1336 		dst_group = ffs(addr->nl_groups);
1337 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1338 			return -EPERM;
1339 	} else {
1340 		dst_pid = nlk->dst_pid;
1341 		dst_group = nlk->dst_group;
1342 	}
1343 
1344 	if (!nlk->pid) {
1345 		err = netlink_autobind(sock);
1346 		if (err)
1347 			goto out;
1348 	}
1349 
1350 	err = -EMSGSIZE;
1351 	if (len > sk->sk_sndbuf - 32)
1352 		goto out;
1353 	err = -ENOBUFS;
1354 	skb = alloc_skb(len, GFP_KERNEL);
1355 	if (skb == NULL)
1356 		goto out;
1357 
1358 	NETLINK_CB(skb).pid	= nlk->pid;
1359 	NETLINK_CB(skb).dst_group = dst_group;
1360 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1361 	NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1362 	security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1363 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1364 
1365 	/* What can I do? Netlink is asynchronous, so that
1366 	   we will have to save current capabilities to
1367 	   check them, when this message will be delivered
1368 	   to corresponding kernel module.   --ANK (980802)
1369 	 */
1370 
1371 	err = -EFAULT;
1372 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1373 		kfree_skb(skb);
1374 		goto out;
1375 	}
1376 
1377 	err = security_netlink_send(sk, skb);
1378 	if (err) {
1379 		kfree_skb(skb);
1380 		goto out;
1381 	}
1382 
1383 	if (dst_group) {
1384 		atomic_inc(&skb->users);
1385 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1386 	}
1387 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1388 
1389 out:
1390 	return err;
1391 }
1392 
1393 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1394 			   struct msghdr *msg, size_t len,
1395 			   int flags)
1396 {
1397 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1398 	struct scm_cookie scm;
1399 	struct sock *sk = sock->sk;
1400 	struct netlink_sock *nlk = nlk_sk(sk);
1401 	int noblock = flags&MSG_DONTWAIT;
1402 	size_t copied;
1403 	struct sk_buff *skb, *frag __maybe_unused = NULL;
1404 	int err;
1405 
1406 	if (flags&MSG_OOB)
1407 		return -EOPNOTSUPP;
1408 
1409 	copied = 0;
1410 
1411 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1412 	if (skb == NULL)
1413 		goto out;
1414 
1415 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1416 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1417 		bool need_compat = !!(flags & MSG_CMSG_COMPAT);
1418 
1419 		/*
1420 		 * If this skb has a frag_list, then here that means that
1421 		 * we will have to use the frag_list skb for compat tasks
1422 		 * and the regular skb for non-compat tasks.
1423 		 *
1424 		 * The skb might (and likely will) be cloned, so we can't
1425 		 * just reset frag_list and go on with things -- we need to
1426 		 * keep that. For the compat case that's easy -- simply get
1427 		 * a reference to the compat skb and free the regular one
1428 		 * including the frag. For the non-compat case, we need to
1429 		 * avoid sending the frag to the user -- so assign NULL but
1430 		 * restore it below before freeing the skb.
1431 		 */
1432 		if (need_compat) {
1433 			struct sk_buff *compskb = skb_shinfo(skb)->frag_list;
1434 			skb_get(compskb);
1435 			kfree_skb(skb);
1436 			skb = compskb;
1437 		} else {
1438 			frag = skb_shinfo(skb)->frag_list;
1439 			skb_shinfo(skb)->frag_list = NULL;
1440 		}
1441 	}
1442 #endif
1443 
1444 	msg->msg_namelen = 0;
1445 
1446 	copied = skb->len;
1447 	if (len < copied) {
1448 		msg->msg_flags |= MSG_TRUNC;
1449 		copied = len;
1450 	}
1451 
1452 	skb_reset_transport_header(skb);
1453 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1454 
1455 	if (msg->msg_name) {
1456 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1457 		addr->nl_family = AF_NETLINK;
1458 		addr->nl_pad    = 0;
1459 		addr->nl_pid	= NETLINK_CB(skb).pid;
1460 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1461 		msg->msg_namelen = sizeof(*addr);
1462 	}
1463 
1464 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1465 		netlink_cmsg_recv_pktinfo(msg, skb);
1466 
1467 	if (NULL == siocb->scm) {
1468 		memset(&scm, 0, sizeof(scm));
1469 		siocb->scm = &scm;
1470 	}
1471 	siocb->scm->creds = *NETLINK_CREDS(skb);
1472 	if (flags & MSG_TRUNC)
1473 		copied = skb->len;
1474 
1475 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1476 	skb_shinfo(skb)->frag_list = frag;
1477 #endif
1478 
1479 	skb_free_datagram(sk, skb);
1480 
1481 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1482 		netlink_dump(sk);
1483 
1484 	scm_recv(sock, msg, siocb->scm, flags);
1485 out:
1486 	netlink_rcv_wake(sk);
1487 	return err ? : copied;
1488 }
1489 
1490 static void netlink_data_ready(struct sock *sk, int len)
1491 {
1492 	BUG();
1493 }
1494 
1495 /*
1496  *	We export these functions to other modules. They provide a
1497  *	complete set of kernel non-blocking support for message
1498  *	queueing.
1499  */
1500 
1501 struct sock *
1502 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1503 		      void (*input)(struct sk_buff *skb),
1504 		      struct mutex *cb_mutex, struct module *module)
1505 {
1506 	struct socket *sock;
1507 	struct sock *sk;
1508 	struct netlink_sock *nlk;
1509 	unsigned long *listeners = NULL;
1510 
1511 	BUG_ON(!nl_table);
1512 
1513 	if (unit < 0 || unit >= MAX_LINKS)
1514 		return NULL;
1515 
1516 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1517 		return NULL;
1518 
1519 	/*
1520 	 * We have to just have a reference on the net from sk, but don't
1521 	 * get_net it. Besides, we cannot get and then put the net here.
1522 	 * So we create one inside init_net and the move it to net.
1523 	 */
1524 
1525 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1526 		goto out_sock_release_nosk;
1527 
1528 	sk = sock->sk;
1529 	sk_change_net(sk, net);
1530 
1531 	if (groups < 32)
1532 		groups = 32;
1533 
1534 	listeners = kzalloc(NLGRPSZ(groups) + sizeof(struct listeners_rcu_head),
1535 			    GFP_KERNEL);
1536 	if (!listeners)
1537 		goto out_sock_release;
1538 
1539 	sk->sk_data_ready = netlink_data_ready;
1540 	if (input)
1541 		nlk_sk(sk)->netlink_rcv = input;
1542 
1543 	if (netlink_insert(sk, net, 0))
1544 		goto out_sock_release;
1545 
1546 	nlk = nlk_sk(sk);
1547 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1548 
1549 	netlink_table_grab();
1550 	if (!nl_table[unit].registered) {
1551 		nl_table[unit].groups = groups;
1552 		nl_table[unit].listeners = listeners;
1553 		nl_table[unit].cb_mutex = cb_mutex;
1554 		nl_table[unit].module = module;
1555 		nl_table[unit].registered = 1;
1556 	} else {
1557 		kfree(listeners);
1558 		nl_table[unit].registered++;
1559 	}
1560 	netlink_table_ungrab();
1561 	return sk;
1562 
1563 out_sock_release:
1564 	kfree(listeners);
1565 	netlink_kernel_release(sk);
1566 	return NULL;
1567 
1568 out_sock_release_nosk:
1569 	sock_release(sock);
1570 	return NULL;
1571 }
1572 EXPORT_SYMBOL(netlink_kernel_create);
1573 
1574 
1575 void
1576 netlink_kernel_release(struct sock *sk)
1577 {
1578 	sk_release_kernel(sk);
1579 }
1580 EXPORT_SYMBOL(netlink_kernel_release);
1581 
1582 
1583 static void netlink_free_old_listeners(struct rcu_head *rcu_head)
1584 {
1585 	struct listeners_rcu_head *lrh;
1586 
1587 	lrh = container_of(rcu_head, struct listeners_rcu_head, rcu_head);
1588 	kfree(lrh->ptr);
1589 }
1590 
1591 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1592 {
1593 	unsigned long *listeners, *old = NULL;
1594 	struct listeners_rcu_head *old_rcu_head;
1595 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1596 
1597 	if (groups < 32)
1598 		groups = 32;
1599 
1600 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1601 		listeners = kzalloc(NLGRPSZ(groups) +
1602 				    sizeof(struct listeners_rcu_head),
1603 				    GFP_ATOMIC);
1604 		if (!listeners)
1605 			return -ENOMEM;
1606 		old = tbl->listeners;
1607 		memcpy(listeners, old, NLGRPSZ(tbl->groups));
1608 		rcu_assign_pointer(tbl->listeners, listeners);
1609 		/*
1610 		 * Free the old memory after an RCU grace period so we
1611 		 * don't leak it. We use call_rcu() here in order to be
1612 		 * able to call this function from atomic contexts. The
1613 		 * allocation of this memory will have reserved enough
1614 		 * space for struct listeners_rcu_head at the end.
1615 		 */
1616 		old_rcu_head = (void *)(tbl->listeners +
1617 					NLGRPLONGS(tbl->groups));
1618 		old_rcu_head->ptr = old;
1619 		call_rcu(&old_rcu_head->rcu_head, netlink_free_old_listeners);
1620 	}
1621 	tbl->groups = groups;
1622 
1623 	return 0;
1624 }
1625 
1626 /**
1627  * netlink_change_ngroups - change number of multicast groups
1628  *
1629  * This changes the number of multicast groups that are available
1630  * on a certain netlink family. Note that it is not possible to
1631  * change the number of groups to below 32. Also note that it does
1632  * not implicitly call netlink_clear_multicast_users() when the
1633  * number of groups is reduced.
1634  *
1635  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1636  * @groups: The new number of groups.
1637  */
1638 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1639 {
1640 	int err;
1641 
1642 	netlink_table_grab();
1643 	err = __netlink_change_ngroups(sk, groups);
1644 	netlink_table_ungrab();
1645 
1646 	return err;
1647 }
1648 
1649 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1650 {
1651 	struct sock *sk;
1652 	struct hlist_node *node;
1653 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1654 
1655 	sk_for_each_bound(sk, node, &tbl->mc_list)
1656 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1657 }
1658 
1659 /**
1660  * netlink_clear_multicast_users - kick off multicast listeners
1661  *
1662  * This function removes all listeners from the given group.
1663  * @ksk: The kernel netlink socket, as returned by
1664  *	netlink_kernel_create().
1665  * @group: The multicast group to clear.
1666  */
1667 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1668 {
1669 	netlink_table_grab();
1670 	__netlink_clear_multicast_users(ksk, group);
1671 	netlink_table_ungrab();
1672 }
1673 
1674 void netlink_set_nonroot(int protocol, unsigned int flags)
1675 {
1676 	if ((unsigned int)protocol < MAX_LINKS)
1677 		nl_table[protocol].nl_nonroot = flags;
1678 }
1679 EXPORT_SYMBOL(netlink_set_nonroot);
1680 
1681 static void netlink_destroy_callback(struct netlink_callback *cb)
1682 {
1683 	kfree_skb(cb->skb);
1684 	kfree(cb);
1685 }
1686 
1687 /*
1688  * It looks a bit ugly.
1689  * It would be better to create kernel thread.
1690  */
1691 
1692 static int netlink_dump(struct sock *sk)
1693 {
1694 	struct netlink_sock *nlk = nlk_sk(sk);
1695 	struct netlink_callback *cb;
1696 	struct sk_buff *skb;
1697 	struct nlmsghdr *nlh;
1698 	int len, err = -ENOBUFS;
1699 
1700 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1701 	if (!skb)
1702 		goto errout;
1703 
1704 	mutex_lock(nlk->cb_mutex);
1705 
1706 	cb = nlk->cb;
1707 	if (cb == NULL) {
1708 		err = -EINVAL;
1709 		goto errout_skb;
1710 	}
1711 
1712 	len = cb->dump(skb, cb);
1713 
1714 	if (len > 0) {
1715 		mutex_unlock(nlk->cb_mutex);
1716 
1717 		if (sk_filter(sk, skb))
1718 			kfree_skb(skb);
1719 		else {
1720 			skb_queue_tail(&sk->sk_receive_queue, skb);
1721 			sk->sk_data_ready(sk, skb->len);
1722 		}
1723 		return 0;
1724 	}
1725 
1726 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1727 	if (!nlh)
1728 		goto errout_skb;
1729 
1730 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1731 
1732 	if (sk_filter(sk, skb))
1733 		kfree_skb(skb);
1734 	else {
1735 		skb_queue_tail(&sk->sk_receive_queue, skb);
1736 		sk->sk_data_ready(sk, skb->len);
1737 	}
1738 
1739 	if (cb->done)
1740 		cb->done(cb);
1741 	nlk->cb = NULL;
1742 	mutex_unlock(nlk->cb_mutex);
1743 
1744 	netlink_destroy_callback(cb);
1745 	return 0;
1746 
1747 errout_skb:
1748 	mutex_unlock(nlk->cb_mutex);
1749 	kfree_skb(skb);
1750 errout:
1751 	return err;
1752 }
1753 
1754 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1755 		       const struct nlmsghdr *nlh,
1756 		       int (*dump)(struct sk_buff *skb,
1757 				   struct netlink_callback *),
1758 		       int (*done)(struct netlink_callback *))
1759 {
1760 	struct netlink_callback *cb;
1761 	struct sock *sk;
1762 	struct netlink_sock *nlk;
1763 
1764 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1765 	if (cb == NULL)
1766 		return -ENOBUFS;
1767 
1768 	cb->dump = dump;
1769 	cb->done = done;
1770 	cb->nlh = nlh;
1771 	atomic_inc(&skb->users);
1772 	cb->skb = skb;
1773 
1774 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1775 	if (sk == NULL) {
1776 		netlink_destroy_callback(cb);
1777 		return -ECONNREFUSED;
1778 	}
1779 	nlk = nlk_sk(sk);
1780 	/* A dump is in progress... */
1781 	mutex_lock(nlk->cb_mutex);
1782 	if (nlk->cb) {
1783 		mutex_unlock(nlk->cb_mutex);
1784 		netlink_destroy_callback(cb);
1785 		sock_put(sk);
1786 		return -EBUSY;
1787 	}
1788 	nlk->cb = cb;
1789 	mutex_unlock(nlk->cb_mutex);
1790 
1791 	netlink_dump(sk);
1792 	sock_put(sk);
1793 
1794 	/* We successfully started a dump, by returning -EINTR we
1795 	 * signal not to send ACK even if it was requested.
1796 	 */
1797 	return -EINTR;
1798 }
1799 EXPORT_SYMBOL(netlink_dump_start);
1800 
1801 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1802 {
1803 	struct sk_buff *skb;
1804 	struct nlmsghdr *rep;
1805 	struct nlmsgerr *errmsg;
1806 	size_t payload = sizeof(*errmsg);
1807 
1808 	/* error messages get the original request appened */
1809 	if (err)
1810 		payload += nlmsg_len(nlh);
1811 
1812 	skb = nlmsg_new(payload, GFP_KERNEL);
1813 	if (!skb) {
1814 		struct sock *sk;
1815 
1816 		sk = netlink_lookup(sock_net(in_skb->sk),
1817 				    in_skb->sk->sk_protocol,
1818 				    NETLINK_CB(in_skb).pid);
1819 		if (sk) {
1820 			sk->sk_err = ENOBUFS;
1821 			sk->sk_error_report(sk);
1822 			sock_put(sk);
1823 		}
1824 		return;
1825 	}
1826 
1827 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1828 			  NLMSG_ERROR, payload, 0);
1829 	errmsg = nlmsg_data(rep);
1830 	errmsg->error = err;
1831 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1832 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1833 }
1834 EXPORT_SYMBOL(netlink_ack);
1835 
1836 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1837 						     struct nlmsghdr *))
1838 {
1839 	struct nlmsghdr *nlh;
1840 	int err;
1841 
1842 	while (skb->len >= nlmsg_total_size(0)) {
1843 		int msglen;
1844 
1845 		nlh = nlmsg_hdr(skb);
1846 		err = 0;
1847 
1848 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1849 			return 0;
1850 
1851 		/* Only requests are handled by the kernel */
1852 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1853 			goto ack;
1854 
1855 		/* Skip control messages */
1856 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1857 			goto ack;
1858 
1859 		err = cb(skb, nlh);
1860 		if (err == -EINTR)
1861 			goto skip;
1862 
1863 ack:
1864 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1865 			netlink_ack(skb, nlh, err);
1866 
1867 skip:
1868 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1869 		if (msglen > skb->len)
1870 			msglen = skb->len;
1871 		skb_pull(skb, msglen);
1872 	}
1873 
1874 	return 0;
1875 }
1876 EXPORT_SYMBOL(netlink_rcv_skb);
1877 
1878 /**
1879  * nlmsg_notify - send a notification netlink message
1880  * @sk: netlink socket to use
1881  * @skb: notification message
1882  * @pid: destination netlink pid for reports or 0
1883  * @group: destination multicast group or 0
1884  * @report: 1 to report back, 0 to disable
1885  * @flags: allocation flags
1886  */
1887 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1888 		 unsigned int group, int report, gfp_t flags)
1889 {
1890 	int err = 0;
1891 
1892 	if (group) {
1893 		int exclude_pid = 0;
1894 
1895 		if (report) {
1896 			atomic_inc(&skb->users);
1897 			exclude_pid = pid;
1898 		}
1899 
1900 		/* errors reported via destination sk->sk_err, but propagate
1901 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1902 		err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1903 	}
1904 
1905 	if (report) {
1906 		int err2;
1907 
1908 		err2 = nlmsg_unicast(sk, skb, pid);
1909 		if (!err || err == -ESRCH)
1910 			err = err2;
1911 	}
1912 
1913 	return err;
1914 }
1915 EXPORT_SYMBOL(nlmsg_notify);
1916 
1917 #ifdef CONFIG_PROC_FS
1918 struct nl_seq_iter {
1919 	struct seq_net_private p;
1920 	int link;
1921 	int hash_idx;
1922 };
1923 
1924 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1925 {
1926 	struct nl_seq_iter *iter = seq->private;
1927 	int i, j;
1928 	struct sock *s;
1929 	struct hlist_node *node;
1930 	loff_t off = 0;
1931 
1932 	for (i = 0; i < MAX_LINKS; i++) {
1933 		struct nl_pid_hash *hash = &nl_table[i].hash;
1934 
1935 		for (j = 0; j <= hash->mask; j++) {
1936 			sk_for_each(s, node, &hash->table[j]) {
1937 				if (sock_net(s) != seq_file_net(seq))
1938 					continue;
1939 				if (off == pos) {
1940 					iter->link = i;
1941 					iter->hash_idx = j;
1942 					return s;
1943 				}
1944 				++off;
1945 			}
1946 		}
1947 	}
1948 	return NULL;
1949 }
1950 
1951 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1952 	__acquires(nl_table_lock)
1953 {
1954 	read_lock(&nl_table_lock);
1955 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1956 }
1957 
1958 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1959 {
1960 	struct sock *s;
1961 	struct nl_seq_iter *iter;
1962 	int i, j;
1963 
1964 	++*pos;
1965 
1966 	if (v == SEQ_START_TOKEN)
1967 		return netlink_seq_socket_idx(seq, 0);
1968 
1969 	iter = seq->private;
1970 	s = v;
1971 	do {
1972 		s = sk_next(s);
1973 	} while (s && sock_net(s) != seq_file_net(seq));
1974 	if (s)
1975 		return s;
1976 
1977 	i = iter->link;
1978 	j = iter->hash_idx + 1;
1979 
1980 	do {
1981 		struct nl_pid_hash *hash = &nl_table[i].hash;
1982 
1983 		for (; j <= hash->mask; j++) {
1984 			s = sk_head(&hash->table[j]);
1985 			while (s && sock_net(s) != seq_file_net(seq))
1986 				s = sk_next(s);
1987 			if (s) {
1988 				iter->link = i;
1989 				iter->hash_idx = j;
1990 				return s;
1991 			}
1992 		}
1993 
1994 		j = 0;
1995 	} while (++i < MAX_LINKS);
1996 
1997 	return NULL;
1998 }
1999 
2000 static void netlink_seq_stop(struct seq_file *seq, void *v)
2001 	__releases(nl_table_lock)
2002 {
2003 	read_unlock(&nl_table_lock);
2004 }
2005 
2006 
2007 static int netlink_seq_show(struct seq_file *seq, void *v)
2008 {
2009 	if (v == SEQ_START_TOKEN)
2010 		seq_puts(seq,
2011 			 "sk       Eth Pid    Groups   "
2012 			 "Rmem     Wmem     Dump     Locks     Drops     Inode\n");
2013 	else {
2014 		struct sock *s = v;
2015 		struct netlink_sock *nlk = nlk_sk(s);
2016 
2017 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %-8d %-8d %-8lu\n",
2018 			   s,
2019 			   s->sk_protocol,
2020 			   nlk->pid,
2021 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2022 			   sk_rmem_alloc_get(s),
2023 			   sk_wmem_alloc_get(s),
2024 			   nlk->cb,
2025 			   atomic_read(&s->sk_refcnt),
2026 			   atomic_read(&s->sk_drops),
2027 			   sock_i_ino(s)
2028 			);
2029 
2030 	}
2031 	return 0;
2032 }
2033 
2034 static const struct seq_operations netlink_seq_ops = {
2035 	.start  = netlink_seq_start,
2036 	.next   = netlink_seq_next,
2037 	.stop   = netlink_seq_stop,
2038 	.show   = netlink_seq_show,
2039 };
2040 
2041 
2042 static int netlink_seq_open(struct inode *inode, struct file *file)
2043 {
2044 	return seq_open_net(inode, file, &netlink_seq_ops,
2045 				sizeof(struct nl_seq_iter));
2046 }
2047 
2048 static const struct file_operations netlink_seq_fops = {
2049 	.owner		= THIS_MODULE,
2050 	.open		= netlink_seq_open,
2051 	.read		= seq_read,
2052 	.llseek		= seq_lseek,
2053 	.release	= seq_release_net,
2054 };
2055 
2056 #endif
2057 
2058 int netlink_register_notifier(struct notifier_block *nb)
2059 {
2060 	return atomic_notifier_chain_register(&netlink_chain, nb);
2061 }
2062 EXPORT_SYMBOL(netlink_register_notifier);
2063 
2064 int netlink_unregister_notifier(struct notifier_block *nb)
2065 {
2066 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
2067 }
2068 EXPORT_SYMBOL(netlink_unregister_notifier);
2069 
2070 static const struct proto_ops netlink_ops = {
2071 	.family =	PF_NETLINK,
2072 	.owner =	THIS_MODULE,
2073 	.release =	netlink_release,
2074 	.bind =		netlink_bind,
2075 	.connect =	netlink_connect,
2076 	.socketpair =	sock_no_socketpair,
2077 	.accept =	sock_no_accept,
2078 	.getname =	netlink_getname,
2079 	.poll =		datagram_poll,
2080 	.ioctl =	sock_no_ioctl,
2081 	.listen =	sock_no_listen,
2082 	.shutdown =	sock_no_shutdown,
2083 	.setsockopt =	netlink_setsockopt,
2084 	.getsockopt =	netlink_getsockopt,
2085 	.sendmsg =	netlink_sendmsg,
2086 	.recvmsg =	netlink_recvmsg,
2087 	.mmap =		sock_no_mmap,
2088 	.sendpage =	sock_no_sendpage,
2089 };
2090 
2091 static const struct net_proto_family netlink_family_ops = {
2092 	.family = PF_NETLINK,
2093 	.create = netlink_create,
2094 	.owner	= THIS_MODULE,	/* for consistency 8) */
2095 };
2096 
2097 static int __net_init netlink_net_init(struct net *net)
2098 {
2099 #ifdef CONFIG_PROC_FS
2100 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2101 		return -ENOMEM;
2102 #endif
2103 	return 0;
2104 }
2105 
2106 static void __net_exit netlink_net_exit(struct net *net)
2107 {
2108 #ifdef CONFIG_PROC_FS
2109 	proc_net_remove(net, "netlink");
2110 #endif
2111 }
2112 
2113 static struct pernet_operations __net_initdata netlink_net_ops = {
2114 	.init = netlink_net_init,
2115 	.exit = netlink_net_exit,
2116 };
2117 
2118 static int __init netlink_proto_init(void)
2119 {
2120 	struct sk_buff *dummy_skb;
2121 	int i;
2122 	unsigned long limit;
2123 	unsigned int order;
2124 	int err = proto_register(&netlink_proto, 0);
2125 
2126 	if (err != 0)
2127 		goto out;
2128 
2129 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2130 
2131 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2132 	if (!nl_table)
2133 		goto panic;
2134 
2135 	if (totalram_pages >= (128 * 1024))
2136 		limit = totalram_pages >> (21 - PAGE_SHIFT);
2137 	else
2138 		limit = totalram_pages >> (23 - PAGE_SHIFT);
2139 
2140 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2141 	limit = (1UL << order) / sizeof(struct hlist_head);
2142 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2143 
2144 	for (i = 0; i < MAX_LINKS; i++) {
2145 		struct nl_pid_hash *hash = &nl_table[i].hash;
2146 
2147 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2148 		if (!hash->table) {
2149 			while (i-- > 0)
2150 				nl_pid_hash_free(nl_table[i].hash.table,
2151 						 1 * sizeof(*hash->table));
2152 			kfree(nl_table);
2153 			goto panic;
2154 		}
2155 		hash->max_shift = order;
2156 		hash->shift = 0;
2157 		hash->mask = 0;
2158 		hash->rehash_time = jiffies;
2159 	}
2160 
2161 	sock_register(&netlink_family_ops);
2162 	register_pernet_subsys(&netlink_net_ops);
2163 	/* The netlink device handler may be needed early. */
2164 	rtnetlink_init();
2165 out:
2166 	return err;
2167 panic:
2168 	panic("netlink_init: Cannot allocate nl_table\n");
2169 }
2170 
2171 core_initcall(netlink_proto_init);
2172