xref: /openbmc/linux/net/netlink/af_netlink.c (revision b6dcefde)
1 /*
2  * NETLINK      Kernel-user communication protocol.
3  *
4  * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
5  * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6  *
7  *		This program is free software; you can redistribute it and/or
8  *		modify it under the terms of the GNU General Public License
9  *		as published by the Free Software Foundation; either version
10  *		2 of the License, or (at your option) any later version.
11  *
12  * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13  *                               added netlink_proto_exit
14  * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15  * 				 use nlk_sk, as sk->protinfo is on a diet 8)
16  * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17  * 				 - inc module use count of module that owns
18  * 				   the kernel socket in case userspace opens
19  * 				   socket of same protocol
20  * 				 - remove all module support, since netlink is
21  * 				   mandatory if CONFIG_NET=y these days
22  */
23 
24 #include <linux/module.h>
25 
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/mutex.h>
58 
59 #include <net/net_namespace.h>
60 #include <net/sock.h>
61 #include <net/scm.h>
62 #include <net/netlink.h>
63 
64 #define NLGRPSZ(x)	(ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x)	(NLGRPSZ(x)/sizeof(unsigned long))
66 
67 struct netlink_sock {
68 	/* struct sock has to be the first member of netlink_sock */
69 	struct sock		sk;
70 	u32			pid;
71 	u32			dst_pid;
72 	u32			dst_group;
73 	u32			flags;
74 	u32			subscriptions;
75 	u32			ngroups;
76 	unsigned long		*groups;
77 	unsigned long		state;
78 	wait_queue_head_t	wait;
79 	struct netlink_callback	*cb;
80 	struct mutex		*cb_mutex;
81 	struct mutex		cb_def_mutex;
82 	void			(*netlink_rcv)(struct sk_buff *skb);
83 	struct module		*module;
84 };
85 
86 struct listeners_rcu_head {
87 	struct rcu_head rcu_head;
88 	void *ptr;
89 };
90 
91 #define NETLINK_KERNEL_SOCKET	0x1
92 #define NETLINK_RECV_PKTINFO	0x2
93 #define NETLINK_BROADCAST_SEND_ERROR	0x4
94 #define NETLINK_RECV_NO_ENOBUFS	0x8
95 
96 static inline struct netlink_sock *nlk_sk(struct sock *sk)
97 {
98 	return container_of(sk, struct netlink_sock, sk);
99 }
100 
101 static inline int netlink_is_kernel(struct sock *sk)
102 {
103 	return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
104 }
105 
106 struct nl_pid_hash {
107 	struct hlist_head *table;
108 	unsigned long rehash_time;
109 
110 	unsigned int mask;
111 	unsigned int shift;
112 
113 	unsigned int entries;
114 	unsigned int max_shift;
115 
116 	u32 rnd;
117 };
118 
119 struct netlink_table {
120 	struct nl_pid_hash hash;
121 	struct hlist_head mc_list;
122 	unsigned long *listeners;
123 	unsigned int nl_nonroot;
124 	unsigned int groups;
125 	struct mutex *cb_mutex;
126 	struct module *module;
127 	int registered;
128 };
129 
130 static struct netlink_table *nl_table;
131 
132 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
133 
134 static int netlink_dump(struct sock *sk);
135 static void netlink_destroy_callback(struct netlink_callback *cb);
136 
137 static DEFINE_RWLOCK(nl_table_lock);
138 static atomic_t nl_table_users = ATOMIC_INIT(0);
139 
140 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
141 
142 static u32 netlink_group_mask(u32 group)
143 {
144 	return group ? 1 << (group - 1) : 0;
145 }
146 
147 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
148 {
149 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
150 }
151 
152 static void netlink_sock_destruct(struct sock *sk)
153 {
154 	struct netlink_sock *nlk = nlk_sk(sk);
155 
156 	if (nlk->cb) {
157 		if (nlk->cb->done)
158 			nlk->cb->done(nlk->cb);
159 		netlink_destroy_callback(nlk->cb);
160 	}
161 
162 	skb_queue_purge(&sk->sk_receive_queue);
163 
164 	if (!sock_flag(sk, SOCK_DEAD)) {
165 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
166 		return;
167 	}
168 
169 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
170 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
171 	WARN_ON(nlk_sk(sk)->groups);
172 }
173 
174 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
175  * SMP. Look, when several writers sleep and reader wakes them up, all but one
176  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
177  * this, _but_ remember, it adds useless work on UP machines.
178  */
179 
180 void netlink_table_grab(void)
181 	__acquires(nl_table_lock)
182 {
183 	might_sleep();
184 
185 	write_lock_irq(&nl_table_lock);
186 
187 	if (atomic_read(&nl_table_users)) {
188 		DECLARE_WAITQUEUE(wait, current);
189 
190 		add_wait_queue_exclusive(&nl_table_wait, &wait);
191 		for (;;) {
192 			set_current_state(TASK_UNINTERRUPTIBLE);
193 			if (atomic_read(&nl_table_users) == 0)
194 				break;
195 			write_unlock_irq(&nl_table_lock);
196 			schedule();
197 			write_lock_irq(&nl_table_lock);
198 		}
199 
200 		__set_current_state(TASK_RUNNING);
201 		remove_wait_queue(&nl_table_wait, &wait);
202 	}
203 }
204 
205 void netlink_table_ungrab(void)
206 	__releases(nl_table_lock)
207 {
208 	write_unlock_irq(&nl_table_lock);
209 	wake_up(&nl_table_wait);
210 }
211 
212 static inline void
213 netlink_lock_table(void)
214 {
215 	/* read_lock() synchronizes us to netlink_table_grab */
216 
217 	read_lock(&nl_table_lock);
218 	atomic_inc(&nl_table_users);
219 	read_unlock(&nl_table_lock);
220 }
221 
222 static inline void
223 netlink_unlock_table(void)
224 {
225 	if (atomic_dec_and_test(&nl_table_users))
226 		wake_up(&nl_table_wait);
227 }
228 
229 static inline struct sock *netlink_lookup(struct net *net, int protocol,
230 					  u32 pid)
231 {
232 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
233 	struct hlist_head *head;
234 	struct sock *sk;
235 	struct hlist_node *node;
236 
237 	read_lock(&nl_table_lock);
238 	head = nl_pid_hashfn(hash, pid);
239 	sk_for_each(sk, node, head) {
240 		if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
241 			sock_hold(sk);
242 			goto found;
243 		}
244 	}
245 	sk = NULL;
246 found:
247 	read_unlock(&nl_table_lock);
248 	return sk;
249 }
250 
251 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
252 {
253 	if (size <= PAGE_SIZE)
254 		return kzalloc(size, GFP_ATOMIC);
255 	else
256 		return (struct hlist_head *)
257 			__get_free_pages(GFP_ATOMIC | __GFP_ZERO,
258 					 get_order(size));
259 }
260 
261 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
262 {
263 	if (size <= PAGE_SIZE)
264 		kfree(table);
265 	else
266 		free_pages((unsigned long)table, get_order(size));
267 }
268 
269 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
270 {
271 	unsigned int omask, mask, shift;
272 	size_t osize, size;
273 	struct hlist_head *otable, *table;
274 	int i;
275 
276 	omask = mask = hash->mask;
277 	osize = size = (mask + 1) * sizeof(*table);
278 	shift = hash->shift;
279 
280 	if (grow) {
281 		if (++shift > hash->max_shift)
282 			return 0;
283 		mask = mask * 2 + 1;
284 		size *= 2;
285 	}
286 
287 	table = nl_pid_hash_zalloc(size);
288 	if (!table)
289 		return 0;
290 
291 	otable = hash->table;
292 	hash->table = table;
293 	hash->mask = mask;
294 	hash->shift = shift;
295 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
296 
297 	for (i = 0; i <= omask; i++) {
298 		struct sock *sk;
299 		struct hlist_node *node, *tmp;
300 
301 		sk_for_each_safe(sk, node, tmp, &otable[i])
302 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
303 	}
304 
305 	nl_pid_hash_free(otable, osize);
306 	hash->rehash_time = jiffies + 10 * 60 * HZ;
307 	return 1;
308 }
309 
310 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
311 {
312 	int avg = hash->entries >> hash->shift;
313 
314 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
315 		return 1;
316 
317 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
318 		nl_pid_hash_rehash(hash, 0);
319 		return 1;
320 	}
321 
322 	return 0;
323 }
324 
325 static const struct proto_ops netlink_ops;
326 
327 static void
328 netlink_update_listeners(struct sock *sk)
329 {
330 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
331 	struct hlist_node *node;
332 	unsigned long mask;
333 	unsigned int i;
334 
335 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
336 		mask = 0;
337 		sk_for_each_bound(sk, node, &tbl->mc_list) {
338 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
339 				mask |= nlk_sk(sk)->groups[i];
340 		}
341 		tbl->listeners[i] = mask;
342 	}
343 	/* this function is only called with the netlink table "grabbed", which
344 	 * makes sure updates are visible before bind or setsockopt return. */
345 }
346 
347 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
348 {
349 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
350 	struct hlist_head *head;
351 	int err = -EADDRINUSE;
352 	struct sock *osk;
353 	struct hlist_node *node;
354 	int len;
355 
356 	netlink_table_grab();
357 	head = nl_pid_hashfn(hash, pid);
358 	len = 0;
359 	sk_for_each(osk, node, head) {
360 		if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
361 			break;
362 		len++;
363 	}
364 	if (node)
365 		goto err;
366 
367 	err = -EBUSY;
368 	if (nlk_sk(sk)->pid)
369 		goto err;
370 
371 	err = -ENOMEM;
372 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
373 		goto err;
374 
375 	if (len && nl_pid_hash_dilute(hash, len))
376 		head = nl_pid_hashfn(hash, pid);
377 	hash->entries++;
378 	nlk_sk(sk)->pid = pid;
379 	sk_add_node(sk, head);
380 	err = 0;
381 
382 err:
383 	netlink_table_ungrab();
384 	return err;
385 }
386 
387 static void netlink_remove(struct sock *sk)
388 {
389 	netlink_table_grab();
390 	if (sk_del_node_init(sk))
391 		nl_table[sk->sk_protocol].hash.entries--;
392 	if (nlk_sk(sk)->subscriptions)
393 		__sk_del_bind_node(sk);
394 	netlink_table_ungrab();
395 }
396 
397 static struct proto netlink_proto = {
398 	.name	  = "NETLINK",
399 	.owner	  = THIS_MODULE,
400 	.obj_size = sizeof(struct netlink_sock),
401 };
402 
403 static int __netlink_create(struct net *net, struct socket *sock,
404 			    struct mutex *cb_mutex, int protocol)
405 {
406 	struct sock *sk;
407 	struct netlink_sock *nlk;
408 
409 	sock->ops = &netlink_ops;
410 
411 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
412 	if (!sk)
413 		return -ENOMEM;
414 
415 	sock_init_data(sock, sk);
416 
417 	nlk = nlk_sk(sk);
418 	if (cb_mutex)
419 		nlk->cb_mutex = cb_mutex;
420 	else {
421 		nlk->cb_mutex = &nlk->cb_def_mutex;
422 		mutex_init(nlk->cb_mutex);
423 	}
424 	init_waitqueue_head(&nlk->wait);
425 
426 	sk->sk_destruct = netlink_sock_destruct;
427 	sk->sk_protocol = protocol;
428 	return 0;
429 }
430 
431 static int netlink_create(struct net *net, struct socket *sock, int protocol,
432 			  int kern)
433 {
434 	struct module *module = NULL;
435 	struct mutex *cb_mutex;
436 	struct netlink_sock *nlk;
437 	int err = 0;
438 
439 	sock->state = SS_UNCONNECTED;
440 
441 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
442 		return -ESOCKTNOSUPPORT;
443 
444 	if (protocol < 0 || protocol >= MAX_LINKS)
445 		return -EPROTONOSUPPORT;
446 
447 	netlink_lock_table();
448 #ifdef CONFIG_MODULES
449 	if (!nl_table[protocol].registered) {
450 		netlink_unlock_table();
451 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
452 		netlink_lock_table();
453 	}
454 #endif
455 	if (nl_table[protocol].registered &&
456 	    try_module_get(nl_table[protocol].module))
457 		module = nl_table[protocol].module;
458 	else
459 		err = -EPROTONOSUPPORT;
460 	cb_mutex = nl_table[protocol].cb_mutex;
461 	netlink_unlock_table();
462 
463 	if (err < 0)
464 		goto out;
465 
466 	err = __netlink_create(net, sock, cb_mutex, protocol);
467 	if (err < 0)
468 		goto out_module;
469 
470 	local_bh_disable();
471 	sock_prot_inuse_add(net, &netlink_proto, 1);
472 	local_bh_enable();
473 
474 	nlk = nlk_sk(sock->sk);
475 	nlk->module = module;
476 out:
477 	return err;
478 
479 out_module:
480 	module_put(module);
481 	goto out;
482 }
483 
484 static int netlink_release(struct socket *sock)
485 {
486 	struct sock *sk = sock->sk;
487 	struct netlink_sock *nlk;
488 
489 	if (!sk)
490 		return 0;
491 
492 	netlink_remove(sk);
493 	sock_orphan(sk);
494 	nlk = nlk_sk(sk);
495 
496 	/*
497 	 * OK. Socket is unlinked, any packets that arrive now
498 	 * will be purged.
499 	 */
500 
501 	sock->sk = NULL;
502 	wake_up_interruptible_all(&nlk->wait);
503 
504 	skb_queue_purge(&sk->sk_write_queue);
505 
506 	if (nlk->pid) {
507 		struct netlink_notify n = {
508 						.net = sock_net(sk),
509 						.protocol = sk->sk_protocol,
510 						.pid = nlk->pid,
511 					  };
512 		atomic_notifier_call_chain(&netlink_chain,
513 				NETLINK_URELEASE, &n);
514 	}
515 
516 	module_put(nlk->module);
517 
518 	netlink_table_grab();
519 	if (netlink_is_kernel(sk)) {
520 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
521 		if (--nl_table[sk->sk_protocol].registered == 0) {
522 			kfree(nl_table[sk->sk_protocol].listeners);
523 			nl_table[sk->sk_protocol].module = NULL;
524 			nl_table[sk->sk_protocol].registered = 0;
525 		}
526 	} else if (nlk->subscriptions)
527 		netlink_update_listeners(sk);
528 	netlink_table_ungrab();
529 
530 	kfree(nlk->groups);
531 	nlk->groups = NULL;
532 
533 	local_bh_disable();
534 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
535 	local_bh_enable();
536 	sock_put(sk);
537 	return 0;
538 }
539 
540 static int netlink_autobind(struct socket *sock)
541 {
542 	struct sock *sk = sock->sk;
543 	struct net *net = sock_net(sk);
544 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
545 	struct hlist_head *head;
546 	struct sock *osk;
547 	struct hlist_node *node;
548 	s32 pid = current->tgid;
549 	int err;
550 	static s32 rover = -4097;
551 
552 retry:
553 	cond_resched();
554 	netlink_table_grab();
555 	head = nl_pid_hashfn(hash, pid);
556 	sk_for_each(osk, node, head) {
557 		if (!net_eq(sock_net(osk), net))
558 			continue;
559 		if (nlk_sk(osk)->pid == pid) {
560 			/* Bind collision, search negative pid values. */
561 			pid = rover--;
562 			if (rover > -4097)
563 				rover = -4097;
564 			netlink_table_ungrab();
565 			goto retry;
566 		}
567 	}
568 	netlink_table_ungrab();
569 
570 	err = netlink_insert(sk, net, pid);
571 	if (err == -EADDRINUSE)
572 		goto retry;
573 
574 	/* If 2 threads race to autobind, that is fine.  */
575 	if (err == -EBUSY)
576 		err = 0;
577 
578 	return err;
579 }
580 
581 static inline int netlink_capable(struct socket *sock, unsigned int flag)
582 {
583 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
584 	       capable(CAP_NET_ADMIN);
585 }
586 
587 static void
588 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
589 {
590 	struct netlink_sock *nlk = nlk_sk(sk);
591 
592 	if (nlk->subscriptions && !subscriptions)
593 		__sk_del_bind_node(sk);
594 	else if (!nlk->subscriptions && subscriptions)
595 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
596 	nlk->subscriptions = subscriptions;
597 }
598 
599 static int netlink_realloc_groups(struct sock *sk)
600 {
601 	struct netlink_sock *nlk = nlk_sk(sk);
602 	unsigned int groups;
603 	unsigned long *new_groups;
604 	int err = 0;
605 
606 	netlink_table_grab();
607 
608 	groups = nl_table[sk->sk_protocol].groups;
609 	if (!nl_table[sk->sk_protocol].registered) {
610 		err = -ENOENT;
611 		goto out_unlock;
612 	}
613 
614 	if (nlk->ngroups >= groups)
615 		goto out_unlock;
616 
617 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
618 	if (new_groups == NULL) {
619 		err = -ENOMEM;
620 		goto out_unlock;
621 	}
622 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
623 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
624 
625 	nlk->groups = new_groups;
626 	nlk->ngroups = groups;
627  out_unlock:
628 	netlink_table_ungrab();
629 	return err;
630 }
631 
632 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
633 			int addr_len)
634 {
635 	struct sock *sk = sock->sk;
636 	struct net *net = sock_net(sk);
637 	struct netlink_sock *nlk = nlk_sk(sk);
638 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
639 	int err;
640 
641 	if (nladdr->nl_family != AF_NETLINK)
642 		return -EINVAL;
643 
644 	/* Only superuser is allowed to listen multicasts */
645 	if (nladdr->nl_groups) {
646 		if (!netlink_capable(sock, NL_NONROOT_RECV))
647 			return -EPERM;
648 		err = netlink_realloc_groups(sk);
649 		if (err)
650 			return err;
651 	}
652 
653 	if (nlk->pid) {
654 		if (nladdr->nl_pid != nlk->pid)
655 			return -EINVAL;
656 	} else {
657 		err = nladdr->nl_pid ?
658 			netlink_insert(sk, net, nladdr->nl_pid) :
659 			netlink_autobind(sock);
660 		if (err)
661 			return err;
662 	}
663 
664 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
665 		return 0;
666 
667 	netlink_table_grab();
668 	netlink_update_subscriptions(sk, nlk->subscriptions +
669 					 hweight32(nladdr->nl_groups) -
670 					 hweight32(nlk->groups[0]));
671 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
672 	netlink_update_listeners(sk);
673 	netlink_table_ungrab();
674 
675 	return 0;
676 }
677 
678 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
679 			   int alen, int flags)
680 {
681 	int err = 0;
682 	struct sock *sk = sock->sk;
683 	struct netlink_sock *nlk = nlk_sk(sk);
684 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
685 
686 	if (addr->sa_family == AF_UNSPEC) {
687 		sk->sk_state	= NETLINK_UNCONNECTED;
688 		nlk->dst_pid	= 0;
689 		nlk->dst_group  = 0;
690 		return 0;
691 	}
692 	if (addr->sa_family != AF_NETLINK)
693 		return -EINVAL;
694 
695 	/* Only superuser is allowed to send multicasts */
696 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
697 		return -EPERM;
698 
699 	if (!nlk->pid)
700 		err = netlink_autobind(sock);
701 
702 	if (err == 0) {
703 		sk->sk_state	= NETLINK_CONNECTED;
704 		nlk->dst_pid 	= nladdr->nl_pid;
705 		nlk->dst_group  = ffs(nladdr->nl_groups);
706 	}
707 
708 	return err;
709 }
710 
711 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
712 			   int *addr_len, int peer)
713 {
714 	struct sock *sk = sock->sk;
715 	struct netlink_sock *nlk = nlk_sk(sk);
716 	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
717 
718 	nladdr->nl_family = AF_NETLINK;
719 	nladdr->nl_pad = 0;
720 	*addr_len = sizeof(*nladdr);
721 
722 	if (peer) {
723 		nladdr->nl_pid = nlk->dst_pid;
724 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
725 	} else {
726 		nladdr->nl_pid = nlk->pid;
727 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
728 	}
729 	return 0;
730 }
731 
732 static void netlink_overrun(struct sock *sk)
733 {
734 	struct netlink_sock *nlk = nlk_sk(sk);
735 
736 	if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
737 		if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
738 			sk->sk_err = ENOBUFS;
739 			sk->sk_error_report(sk);
740 		}
741 	}
742 	atomic_inc(&sk->sk_drops);
743 }
744 
745 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
746 {
747 	struct sock *sock;
748 	struct netlink_sock *nlk;
749 
750 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
751 	if (!sock)
752 		return ERR_PTR(-ECONNREFUSED);
753 
754 	/* Don't bother queuing skb if kernel socket has no input function */
755 	nlk = nlk_sk(sock);
756 	if (sock->sk_state == NETLINK_CONNECTED &&
757 	    nlk->dst_pid != nlk_sk(ssk)->pid) {
758 		sock_put(sock);
759 		return ERR_PTR(-ECONNREFUSED);
760 	}
761 	return sock;
762 }
763 
764 struct sock *netlink_getsockbyfilp(struct file *filp)
765 {
766 	struct inode *inode = filp->f_path.dentry->d_inode;
767 	struct sock *sock;
768 
769 	if (!S_ISSOCK(inode->i_mode))
770 		return ERR_PTR(-ENOTSOCK);
771 
772 	sock = SOCKET_I(inode)->sk;
773 	if (sock->sk_family != AF_NETLINK)
774 		return ERR_PTR(-EINVAL);
775 
776 	sock_hold(sock);
777 	return sock;
778 }
779 
780 /*
781  * Attach a skb to a netlink socket.
782  * The caller must hold a reference to the destination socket. On error, the
783  * reference is dropped. The skb is not send to the destination, just all
784  * all error checks are performed and memory in the queue is reserved.
785  * Return values:
786  * < 0: error. skb freed, reference to sock dropped.
787  * 0: continue
788  * 1: repeat lookup - reference dropped while waiting for socket memory.
789  */
790 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
791 		      long *timeo, struct sock *ssk)
792 {
793 	struct netlink_sock *nlk;
794 
795 	nlk = nlk_sk(sk);
796 
797 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
798 	    test_bit(0, &nlk->state)) {
799 		DECLARE_WAITQUEUE(wait, current);
800 		if (!*timeo) {
801 			if (!ssk || netlink_is_kernel(ssk))
802 				netlink_overrun(sk);
803 			sock_put(sk);
804 			kfree_skb(skb);
805 			return -EAGAIN;
806 		}
807 
808 		__set_current_state(TASK_INTERRUPTIBLE);
809 		add_wait_queue(&nlk->wait, &wait);
810 
811 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
812 		     test_bit(0, &nlk->state)) &&
813 		    !sock_flag(sk, SOCK_DEAD))
814 			*timeo = schedule_timeout(*timeo);
815 
816 		__set_current_state(TASK_RUNNING);
817 		remove_wait_queue(&nlk->wait, &wait);
818 		sock_put(sk);
819 
820 		if (signal_pending(current)) {
821 			kfree_skb(skb);
822 			return sock_intr_errno(*timeo);
823 		}
824 		return 1;
825 	}
826 	skb_set_owner_r(skb, sk);
827 	return 0;
828 }
829 
830 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
831 {
832 	int len = skb->len;
833 
834 	skb_queue_tail(&sk->sk_receive_queue, skb);
835 	sk->sk_data_ready(sk, len);
836 	sock_put(sk);
837 	return len;
838 }
839 
840 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
841 {
842 	kfree_skb(skb);
843 	sock_put(sk);
844 }
845 
846 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
847 					   gfp_t allocation)
848 {
849 	int delta;
850 
851 	skb_orphan(skb);
852 
853 	delta = skb->end - skb->tail;
854 	if (delta * 2 < skb->truesize)
855 		return skb;
856 
857 	if (skb_shared(skb)) {
858 		struct sk_buff *nskb = skb_clone(skb, allocation);
859 		if (!nskb)
860 			return skb;
861 		kfree_skb(skb);
862 		skb = nskb;
863 	}
864 
865 	if (!pskb_expand_head(skb, 0, -delta, allocation))
866 		skb->truesize -= delta;
867 
868 	return skb;
869 }
870 
871 static inline void netlink_rcv_wake(struct sock *sk)
872 {
873 	struct netlink_sock *nlk = nlk_sk(sk);
874 
875 	if (skb_queue_empty(&sk->sk_receive_queue))
876 		clear_bit(0, &nlk->state);
877 	if (!test_bit(0, &nlk->state))
878 		wake_up_interruptible(&nlk->wait);
879 }
880 
881 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
882 {
883 	int ret;
884 	struct netlink_sock *nlk = nlk_sk(sk);
885 
886 	ret = -ECONNREFUSED;
887 	if (nlk->netlink_rcv != NULL) {
888 		ret = skb->len;
889 		skb_set_owner_r(skb, sk);
890 		nlk->netlink_rcv(skb);
891 	}
892 	kfree_skb(skb);
893 	sock_put(sk);
894 	return ret;
895 }
896 
897 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
898 		    u32 pid, int nonblock)
899 {
900 	struct sock *sk;
901 	int err;
902 	long timeo;
903 
904 	skb = netlink_trim(skb, gfp_any());
905 
906 	timeo = sock_sndtimeo(ssk, nonblock);
907 retry:
908 	sk = netlink_getsockbypid(ssk, pid);
909 	if (IS_ERR(sk)) {
910 		kfree_skb(skb);
911 		return PTR_ERR(sk);
912 	}
913 	if (netlink_is_kernel(sk))
914 		return netlink_unicast_kernel(sk, skb);
915 
916 	if (sk_filter(sk, skb)) {
917 		err = skb->len;
918 		kfree_skb(skb);
919 		sock_put(sk);
920 		return err;
921 	}
922 
923 	err = netlink_attachskb(sk, skb, &timeo, ssk);
924 	if (err == 1)
925 		goto retry;
926 	if (err)
927 		return err;
928 
929 	return netlink_sendskb(sk, skb);
930 }
931 EXPORT_SYMBOL(netlink_unicast);
932 
933 int netlink_has_listeners(struct sock *sk, unsigned int group)
934 {
935 	int res = 0;
936 	unsigned long *listeners;
937 
938 	BUG_ON(!netlink_is_kernel(sk));
939 
940 	rcu_read_lock();
941 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
942 
943 	if (group - 1 < nl_table[sk->sk_protocol].groups)
944 		res = test_bit(group - 1, listeners);
945 
946 	rcu_read_unlock();
947 
948 	return res;
949 }
950 EXPORT_SYMBOL_GPL(netlink_has_listeners);
951 
952 static inline int netlink_broadcast_deliver(struct sock *sk,
953 					    struct sk_buff *skb)
954 {
955 	struct netlink_sock *nlk = nlk_sk(sk);
956 
957 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
958 	    !test_bit(0, &nlk->state)) {
959 		skb_set_owner_r(skb, sk);
960 		skb_queue_tail(&sk->sk_receive_queue, skb);
961 		sk->sk_data_ready(sk, skb->len);
962 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
963 	}
964 	return -1;
965 }
966 
967 struct netlink_broadcast_data {
968 	struct sock *exclude_sk;
969 	struct net *net;
970 	u32 pid;
971 	u32 group;
972 	int failure;
973 	int delivery_failure;
974 	int congested;
975 	int delivered;
976 	gfp_t allocation;
977 	struct sk_buff *skb, *skb2;
978 };
979 
980 static inline int do_one_broadcast(struct sock *sk,
981 				   struct netlink_broadcast_data *p)
982 {
983 	struct netlink_sock *nlk = nlk_sk(sk);
984 	int val;
985 
986 	if (p->exclude_sk == sk)
987 		goto out;
988 
989 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
990 	    !test_bit(p->group - 1, nlk->groups))
991 		goto out;
992 
993 	if (!net_eq(sock_net(sk), p->net))
994 		goto out;
995 
996 	if (p->failure) {
997 		netlink_overrun(sk);
998 		goto out;
999 	}
1000 
1001 	sock_hold(sk);
1002 	if (p->skb2 == NULL) {
1003 		if (skb_shared(p->skb)) {
1004 			p->skb2 = skb_clone(p->skb, p->allocation);
1005 		} else {
1006 			p->skb2 = skb_get(p->skb);
1007 			/*
1008 			 * skb ownership may have been set when
1009 			 * delivered to a previous socket.
1010 			 */
1011 			skb_orphan(p->skb2);
1012 		}
1013 	}
1014 	if (p->skb2 == NULL) {
1015 		netlink_overrun(sk);
1016 		/* Clone failed. Notify ALL listeners. */
1017 		p->failure = 1;
1018 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1019 			p->delivery_failure = 1;
1020 	} else if (sk_filter(sk, p->skb2)) {
1021 		kfree_skb(p->skb2);
1022 		p->skb2 = NULL;
1023 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1024 		netlink_overrun(sk);
1025 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1026 			p->delivery_failure = 1;
1027 	} else {
1028 		p->congested |= val;
1029 		p->delivered = 1;
1030 		p->skb2 = NULL;
1031 	}
1032 	sock_put(sk);
1033 
1034 out:
1035 	return 0;
1036 }
1037 
1038 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1039 		      u32 group, gfp_t allocation)
1040 {
1041 	struct net *net = sock_net(ssk);
1042 	struct netlink_broadcast_data info;
1043 	struct hlist_node *node;
1044 	struct sock *sk;
1045 
1046 	skb = netlink_trim(skb, allocation);
1047 
1048 	info.exclude_sk = ssk;
1049 	info.net = net;
1050 	info.pid = pid;
1051 	info.group = group;
1052 	info.failure = 0;
1053 	info.delivery_failure = 0;
1054 	info.congested = 0;
1055 	info.delivered = 0;
1056 	info.allocation = allocation;
1057 	info.skb = skb;
1058 	info.skb2 = NULL;
1059 
1060 	/* While we sleep in clone, do not allow to change socket list */
1061 
1062 	netlink_lock_table();
1063 
1064 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1065 		do_one_broadcast(sk, &info);
1066 
1067 	kfree_skb(skb);
1068 
1069 	netlink_unlock_table();
1070 
1071 	kfree_skb(info.skb2);
1072 
1073 	if (info.delivery_failure)
1074 		return -ENOBUFS;
1075 
1076 	if (info.delivered) {
1077 		if (info.congested && (allocation & __GFP_WAIT))
1078 			yield();
1079 		return 0;
1080 	}
1081 	return -ESRCH;
1082 }
1083 EXPORT_SYMBOL(netlink_broadcast);
1084 
1085 struct netlink_set_err_data {
1086 	struct sock *exclude_sk;
1087 	u32 pid;
1088 	u32 group;
1089 	int code;
1090 };
1091 
1092 static inline int do_one_set_err(struct sock *sk,
1093 				 struct netlink_set_err_data *p)
1094 {
1095 	struct netlink_sock *nlk = nlk_sk(sk);
1096 
1097 	if (sk == p->exclude_sk)
1098 		goto out;
1099 
1100 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1101 		goto out;
1102 
1103 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1104 	    !test_bit(p->group - 1, nlk->groups))
1105 		goto out;
1106 
1107 	sk->sk_err = p->code;
1108 	sk->sk_error_report(sk);
1109 out:
1110 	return 0;
1111 }
1112 
1113 /**
1114  * netlink_set_err - report error to broadcast listeners
1115  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1116  * @pid: the PID of a process that we want to skip (if any)
1117  * @groups: the broadcast group that will notice the error
1118  * @code: error code, must be negative (as usual in kernelspace)
1119  */
1120 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1121 {
1122 	struct netlink_set_err_data info;
1123 	struct hlist_node *node;
1124 	struct sock *sk;
1125 
1126 	info.exclude_sk = ssk;
1127 	info.pid = pid;
1128 	info.group = group;
1129 	/* sk->sk_err wants a positive error value */
1130 	info.code = -code;
1131 
1132 	read_lock(&nl_table_lock);
1133 
1134 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1135 		do_one_set_err(sk, &info);
1136 
1137 	read_unlock(&nl_table_lock);
1138 }
1139 EXPORT_SYMBOL(netlink_set_err);
1140 
1141 /* must be called with netlink table grabbed */
1142 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1143 				     unsigned int group,
1144 				     int is_new)
1145 {
1146 	int old, new = !!is_new, subscriptions;
1147 
1148 	old = test_bit(group - 1, nlk->groups);
1149 	subscriptions = nlk->subscriptions - old + new;
1150 	if (new)
1151 		__set_bit(group - 1, nlk->groups);
1152 	else
1153 		__clear_bit(group - 1, nlk->groups);
1154 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1155 	netlink_update_listeners(&nlk->sk);
1156 }
1157 
1158 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1159 			      char __user *optval, unsigned int optlen)
1160 {
1161 	struct sock *sk = sock->sk;
1162 	struct netlink_sock *nlk = nlk_sk(sk);
1163 	unsigned int val = 0;
1164 	int err;
1165 
1166 	if (level != SOL_NETLINK)
1167 		return -ENOPROTOOPT;
1168 
1169 	if (optlen >= sizeof(int) &&
1170 	    get_user(val, (unsigned int __user *)optval))
1171 		return -EFAULT;
1172 
1173 	switch (optname) {
1174 	case NETLINK_PKTINFO:
1175 		if (val)
1176 			nlk->flags |= NETLINK_RECV_PKTINFO;
1177 		else
1178 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1179 		err = 0;
1180 		break;
1181 	case NETLINK_ADD_MEMBERSHIP:
1182 	case NETLINK_DROP_MEMBERSHIP: {
1183 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1184 			return -EPERM;
1185 		err = netlink_realloc_groups(sk);
1186 		if (err)
1187 			return err;
1188 		if (!val || val - 1 >= nlk->ngroups)
1189 			return -EINVAL;
1190 		netlink_table_grab();
1191 		netlink_update_socket_mc(nlk, val,
1192 					 optname == NETLINK_ADD_MEMBERSHIP);
1193 		netlink_table_ungrab();
1194 		err = 0;
1195 		break;
1196 	}
1197 	case NETLINK_BROADCAST_ERROR:
1198 		if (val)
1199 			nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1200 		else
1201 			nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1202 		err = 0;
1203 		break;
1204 	case NETLINK_NO_ENOBUFS:
1205 		if (val) {
1206 			nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1207 			clear_bit(0, &nlk->state);
1208 			wake_up_interruptible(&nlk->wait);
1209 		} else
1210 			nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1211 		err = 0;
1212 		break;
1213 	default:
1214 		err = -ENOPROTOOPT;
1215 	}
1216 	return err;
1217 }
1218 
1219 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1220 			      char __user *optval, int __user *optlen)
1221 {
1222 	struct sock *sk = sock->sk;
1223 	struct netlink_sock *nlk = nlk_sk(sk);
1224 	int len, val, err;
1225 
1226 	if (level != SOL_NETLINK)
1227 		return -ENOPROTOOPT;
1228 
1229 	if (get_user(len, optlen))
1230 		return -EFAULT;
1231 	if (len < 0)
1232 		return -EINVAL;
1233 
1234 	switch (optname) {
1235 	case NETLINK_PKTINFO:
1236 		if (len < sizeof(int))
1237 			return -EINVAL;
1238 		len = sizeof(int);
1239 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1240 		if (put_user(len, optlen) ||
1241 		    put_user(val, optval))
1242 			return -EFAULT;
1243 		err = 0;
1244 		break;
1245 	case NETLINK_BROADCAST_ERROR:
1246 		if (len < sizeof(int))
1247 			return -EINVAL;
1248 		len = sizeof(int);
1249 		val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1250 		if (put_user(len, optlen) ||
1251 		    put_user(val, optval))
1252 			return -EFAULT;
1253 		err = 0;
1254 		break;
1255 	case NETLINK_NO_ENOBUFS:
1256 		if (len < sizeof(int))
1257 			return -EINVAL;
1258 		len = sizeof(int);
1259 		val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1260 		if (put_user(len, optlen) ||
1261 		    put_user(val, optval))
1262 			return -EFAULT;
1263 		err = 0;
1264 		break;
1265 	default:
1266 		err = -ENOPROTOOPT;
1267 	}
1268 	return err;
1269 }
1270 
1271 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1272 {
1273 	struct nl_pktinfo info;
1274 
1275 	info.group = NETLINK_CB(skb).dst_group;
1276 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1277 }
1278 
1279 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1280 			   struct msghdr *msg, size_t len)
1281 {
1282 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1283 	struct sock *sk = sock->sk;
1284 	struct netlink_sock *nlk = nlk_sk(sk);
1285 	struct sockaddr_nl *addr = msg->msg_name;
1286 	u32 dst_pid;
1287 	u32 dst_group;
1288 	struct sk_buff *skb;
1289 	int err;
1290 	struct scm_cookie scm;
1291 
1292 	if (msg->msg_flags&MSG_OOB)
1293 		return -EOPNOTSUPP;
1294 
1295 	if (NULL == siocb->scm)
1296 		siocb->scm = &scm;
1297 	err = scm_send(sock, msg, siocb->scm);
1298 	if (err < 0)
1299 		return err;
1300 
1301 	if (msg->msg_namelen) {
1302 		if (addr->nl_family != AF_NETLINK)
1303 			return -EINVAL;
1304 		dst_pid = addr->nl_pid;
1305 		dst_group = ffs(addr->nl_groups);
1306 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1307 			return -EPERM;
1308 	} else {
1309 		dst_pid = nlk->dst_pid;
1310 		dst_group = nlk->dst_group;
1311 	}
1312 
1313 	if (!nlk->pid) {
1314 		err = netlink_autobind(sock);
1315 		if (err)
1316 			goto out;
1317 	}
1318 
1319 	err = -EMSGSIZE;
1320 	if (len > sk->sk_sndbuf - 32)
1321 		goto out;
1322 	err = -ENOBUFS;
1323 	skb = alloc_skb(len, GFP_KERNEL);
1324 	if (skb == NULL)
1325 		goto out;
1326 
1327 	NETLINK_CB(skb).pid	= nlk->pid;
1328 	NETLINK_CB(skb).dst_group = dst_group;
1329 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1330 	NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1331 	security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1332 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1333 
1334 	/* What can I do? Netlink is asynchronous, so that
1335 	   we will have to save current capabilities to
1336 	   check them, when this message will be delivered
1337 	   to corresponding kernel module.   --ANK (980802)
1338 	 */
1339 
1340 	err = -EFAULT;
1341 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1342 		kfree_skb(skb);
1343 		goto out;
1344 	}
1345 
1346 	err = security_netlink_send(sk, skb);
1347 	if (err) {
1348 		kfree_skb(skb);
1349 		goto out;
1350 	}
1351 
1352 	if (dst_group) {
1353 		atomic_inc(&skb->users);
1354 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1355 	}
1356 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1357 
1358 out:
1359 	return err;
1360 }
1361 
1362 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1363 			   struct msghdr *msg, size_t len,
1364 			   int flags)
1365 {
1366 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1367 	struct scm_cookie scm;
1368 	struct sock *sk = sock->sk;
1369 	struct netlink_sock *nlk = nlk_sk(sk);
1370 	int noblock = flags&MSG_DONTWAIT;
1371 	size_t copied;
1372 	struct sk_buff *skb, *frag __maybe_unused = NULL;
1373 	int err;
1374 
1375 	if (flags&MSG_OOB)
1376 		return -EOPNOTSUPP;
1377 
1378 	copied = 0;
1379 
1380 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1381 	if (skb == NULL)
1382 		goto out;
1383 
1384 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1385 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1386 		bool need_compat = !!(flags & MSG_CMSG_COMPAT);
1387 
1388 		/*
1389 		 * If this skb has a frag_list, then here that means that
1390 		 * we will have to use the frag_list skb for compat tasks
1391 		 * and the regular skb for non-compat tasks.
1392 		 *
1393 		 * The skb might (and likely will) be cloned, so we can't
1394 		 * just reset frag_list and go on with things -- we need to
1395 		 * keep that. For the compat case that's easy -- simply get
1396 		 * a reference to the compat skb and free the regular one
1397 		 * including the frag. For the non-compat case, we need to
1398 		 * avoid sending the frag to the user -- so assign NULL but
1399 		 * restore it below before freeing the skb.
1400 		 */
1401 		if (need_compat) {
1402 			struct sk_buff *compskb = skb_shinfo(skb)->frag_list;
1403 			skb_get(compskb);
1404 			kfree_skb(skb);
1405 			skb = compskb;
1406 		} else {
1407 			frag = skb_shinfo(skb)->frag_list;
1408 			skb_shinfo(skb)->frag_list = NULL;
1409 		}
1410 	}
1411 #endif
1412 
1413 	msg->msg_namelen = 0;
1414 
1415 	copied = skb->len;
1416 	if (len < copied) {
1417 		msg->msg_flags |= MSG_TRUNC;
1418 		copied = len;
1419 	}
1420 
1421 	skb_reset_transport_header(skb);
1422 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1423 
1424 	if (msg->msg_name) {
1425 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1426 		addr->nl_family = AF_NETLINK;
1427 		addr->nl_pad    = 0;
1428 		addr->nl_pid	= NETLINK_CB(skb).pid;
1429 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1430 		msg->msg_namelen = sizeof(*addr);
1431 	}
1432 
1433 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1434 		netlink_cmsg_recv_pktinfo(msg, skb);
1435 
1436 	if (NULL == siocb->scm) {
1437 		memset(&scm, 0, sizeof(scm));
1438 		siocb->scm = &scm;
1439 	}
1440 	siocb->scm->creds = *NETLINK_CREDS(skb);
1441 	if (flags & MSG_TRUNC)
1442 		copied = skb->len;
1443 
1444 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1445 	skb_shinfo(skb)->frag_list = frag;
1446 #endif
1447 
1448 	skb_free_datagram(sk, skb);
1449 
1450 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1451 		netlink_dump(sk);
1452 
1453 	scm_recv(sock, msg, siocb->scm, flags);
1454 out:
1455 	netlink_rcv_wake(sk);
1456 	return err ? : copied;
1457 }
1458 
1459 static void netlink_data_ready(struct sock *sk, int len)
1460 {
1461 	BUG();
1462 }
1463 
1464 /*
1465  *	We export these functions to other modules. They provide a
1466  *	complete set of kernel non-blocking support for message
1467  *	queueing.
1468  */
1469 
1470 struct sock *
1471 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1472 		      void (*input)(struct sk_buff *skb),
1473 		      struct mutex *cb_mutex, struct module *module)
1474 {
1475 	struct socket *sock;
1476 	struct sock *sk;
1477 	struct netlink_sock *nlk;
1478 	unsigned long *listeners = NULL;
1479 
1480 	BUG_ON(!nl_table);
1481 
1482 	if (unit < 0 || unit >= MAX_LINKS)
1483 		return NULL;
1484 
1485 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1486 		return NULL;
1487 
1488 	/*
1489 	 * We have to just have a reference on the net from sk, but don't
1490 	 * get_net it. Besides, we cannot get and then put the net here.
1491 	 * So we create one inside init_net and the move it to net.
1492 	 */
1493 
1494 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1495 		goto out_sock_release_nosk;
1496 
1497 	sk = sock->sk;
1498 	sk_change_net(sk, net);
1499 
1500 	if (groups < 32)
1501 		groups = 32;
1502 
1503 	listeners = kzalloc(NLGRPSZ(groups) + sizeof(struct listeners_rcu_head),
1504 			    GFP_KERNEL);
1505 	if (!listeners)
1506 		goto out_sock_release;
1507 
1508 	sk->sk_data_ready = netlink_data_ready;
1509 	if (input)
1510 		nlk_sk(sk)->netlink_rcv = input;
1511 
1512 	if (netlink_insert(sk, net, 0))
1513 		goto out_sock_release;
1514 
1515 	nlk = nlk_sk(sk);
1516 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1517 
1518 	netlink_table_grab();
1519 	if (!nl_table[unit].registered) {
1520 		nl_table[unit].groups = groups;
1521 		nl_table[unit].listeners = listeners;
1522 		nl_table[unit].cb_mutex = cb_mutex;
1523 		nl_table[unit].module = module;
1524 		nl_table[unit].registered = 1;
1525 	} else {
1526 		kfree(listeners);
1527 		nl_table[unit].registered++;
1528 	}
1529 	netlink_table_ungrab();
1530 	return sk;
1531 
1532 out_sock_release:
1533 	kfree(listeners);
1534 	netlink_kernel_release(sk);
1535 	return NULL;
1536 
1537 out_sock_release_nosk:
1538 	sock_release(sock);
1539 	return NULL;
1540 }
1541 EXPORT_SYMBOL(netlink_kernel_create);
1542 
1543 
1544 void
1545 netlink_kernel_release(struct sock *sk)
1546 {
1547 	sk_release_kernel(sk);
1548 }
1549 EXPORT_SYMBOL(netlink_kernel_release);
1550 
1551 
1552 static void netlink_free_old_listeners(struct rcu_head *rcu_head)
1553 {
1554 	struct listeners_rcu_head *lrh;
1555 
1556 	lrh = container_of(rcu_head, struct listeners_rcu_head, rcu_head);
1557 	kfree(lrh->ptr);
1558 }
1559 
1560 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1561 {
1562 	unsigned long *listeners, *old = NULL;
1563 	struct listeners_rcu_head *old_rcu_head;
1564 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1565 
1566 	if (groups < 32)
1567 		groups = 32;
1568 
1569 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1570 		listeners = kzalloc(NLGRPSZ(groups) +
1571 				    sizeof(struct listeners_rcu_head),
1572 				    GFP_ATOMIC);
1573 		if (!listeners)
1574 			return -ENOMEM;
1575 		old = tbl->listeners;
1576 		memcpy(listeners, old, NLGRPSZ(tbl->groups));
1577 		rcu_assign_pointer(tbl->listeners, listeners);
1578 		/*
1579 		 * Free the old memory after an RCU grace period so we
1580 		 * don't leak it. We use call_rcu() here in order to be
1581 		 * able to call this function from atomic contexts. The
1582 		 * allocation of this memory will have reserved enough
1583 		 * space for struct listeners_rcu_head at the end.
1584 		 */
1585 		old_rcu_head = (void *)(tbl->listeners +
1586 					NLGRPLONGS(tbl->groups));
1587 		old_rcu_head->ptr = old;
1588 		call_rcu(&old_rcu_head->rcu_head, netlink_free_old_listeners);
1589 	}
1590 	tbl->groups = groups;
1591 
1592 	return 0;
1593 }
1594 
1595 /**
1596  * netlink_change_ngroups - change number of multicast groups
1597  *
1598  * This changes the number of multicast groups that are available
1599  * on a certain netlink family. Note that it is not possible to
1600  * change the number of groups to below 32. Also note that it does
1601  * not implicitly call netlink_clear_multicast_users() when the
1602  * number of groups is reduced.
1603  *
1604  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1605  * @groups: The new number of groups.
1606  */
1607 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1608 {
1609 	int err;
1610 
1611 	netlink_table_grab();
1612 	err = __netlink_change_ngroups(sk, groups);
1613 	netlink_table_ungrab();
1614 
1615 	return err;
1616 }
1617 
1618 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1619 {
1620 	struct sock *sk;
1621 	struct hlist_node *node;
1622 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1623 
1624 	sk_for_each_bound(sk, node, &tbl->mc_list)
1625 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1626 }
1627 
1628 /**
1629  * netlink_clear_multicast_users - kick off multicast listeners
1630  *
1631  * This function removes all listeners from the given group.
1632  * @ksk: The kernel netlink socket, as returned by
1633  *	netlink_kernel_create().
1634  * @group: The multicast group to clear.
1635  */
1636 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1637 {
1638 	netlink_table_grab();
1639 	__netlink_clear_multicast_users(ksk, group);
1640 	netlink_table_ungrab();
1641 }
1642 
1643 void netlink_set_nonroot(int protocol, unsigned int flags)
1644 {
1645 	if ((unsigned int)protocol < MAX_LINKS)
1646 		nl_table[protocol].nl_nonroot = flags;
1647 }
1648 EXPORT_SYMBOL(netlink_set_nonroot);
1649 
1650 static void netlink_destroy_callback(struct netlink_callback *cb)
1651 {
1652 	kfree_skb(cb->skb);
1653 	kfree(cb);
1654 }
1655 
1656 /*
1657  * It looks a bit ugly.
1658  * It would be better to create kernel thread.
1659  */
1660 
1661 static int netlink_dump(struct sock *sk)
1662 {
1663 	struct netlink_sock *nlk = nlk_sk(sk);
1664 	struct netlink_callback *cb;
1665 	struct sk_buff *skb;
1666 	struct nlmsghdr *nlh;
1667 	int len, err = -ENOBUFS;
1668 
1669 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1670 	if (!skb)
1671 		goto errout;
1672 
1673 	mutex_lock(nlk->cb_mutex);
1674 
1675 	cb = nlk->cb;
1676 	if (cb == NULL) {
1677 		err = -EINVAL;
1678 		goto errout_skb;
1679 	}
1680 
1681 	len = cb->dump(skb, cb);
1682 
1683 	if (len > 0) {
1684 		mutex_unlock(nlk->cb_mutex);
1685 
1686 		if (sk_filter(sk, skb))
1687 			kfree_skb(skb);
1688 		else {
1689 			skb_queue_tail(&sk->sk_receive_queue, skb);
1690 			sk->sk_data_ready(sk, skb->len);
1691 		}
1692 		return 0;
1693 	}
1694 
1695 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1696 	if (!nlh)
1697 		goto errout_skb;
1698 
1699 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1700 
1701 	if (sk_filter(sk, skb))
1702 		kfree_skb(skb);
1703 	else {
1704 		skb_queue_tail(&sk->sk_receive_queue, skb);
1705 		sk->sk_data_ready(sk, skb->len);
1706 	}
1707 
1708 	if (cb->done)
1709 		cb->done(cb);
1710 	nlk->cb = NULL;
1711 	mutex_unlock(nlk->cb_mutex);
1712 
1713 	netlink_destroy_callback(cb);
1714 	return 0;
1715 
1716 errout_skb:
1717 	mutex_unlock(nlk->cb_mutex);
1718 	kfree_skb(skb);
1719 errout:
1720 	return err;
1721 }
1722 
1723 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1724 		       const struct nlmsghdr *nlh,
1725 		       int (*dump)(struct sk_buff *skb,
1726 				   struct netlink_callback *),
1727 		       int (*done)(struct netlink_callback *))
1728 {
1729 	struct netlink_callback *cb;
1730 	struct sock *sk;
1731 	struct netlink_sock *nlk;
1732 
1733 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1734 	if (cb == NULL)
1735 		return -ENOBUFS;
1736 
1737 	cb->dump = dump;
1738 	cb->done = done;
1739 	cb->nlh = nlh;
1740 	atomic_inc(&skb->users);
1741 	cb->skb = skb;
1742 
1743 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1744 	if (sk == NULL) {
1745 		netlink_destroy_callback(cb);
1746 		return -ECONNREFUSED;
1747 	}
1748 	nlk = nlk_sk(sk);
1749 	/* A dump is in progress... */
1750 	mutex_lock(nlk->cb_mutex);
1751 	if (nlk->cb) {
1752 		mutex_unlock(nlk->cb_mutex);
1753 		netlink_destroy_callback(cb);
1754 		sock_put(sk);
1755 		return -EBUSY;
1756 	}
1757 	nlk->cb = cb;
1758 	mutex_unlock(nlk->cb_mutex);
1759 
1760 	netlink_dump(sk);
1761 	sock_put(sk);
1762 
1763 	/* We successfully started a dump, by returning -EINTR we
1764 	 * signal not to send ACK even if it was requested.
1765 	 */
1766 	return -EINTR;
1767 }
1768 EXPORT_SYMBOL(netlink_dump_start);
1769 
1770 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1771 {
1772 	struct sk_buff *skb;
1773 	struct nlmsghdr *rep;
1774 	struct nlmsgerr *errmsg;
1775 	size_t payload = sizeof(*errmsg);
1776 
1777 	/* error messages get the original request appened */
1778 	if (err)
1779 		payload += nlmsg_len(nlh);
1780 
1781 	skb = nlmsg_new(payload, GFP_KERNEL);
1782 	if (!skb) {
1783 		struct sock *sk;
1784 
1785 		sk = netlink_lookup(sock_net(in_skb->sk),
1786 				    in_skb->sk->sk_protocol,
1787 				    NETLINK_CB(in_skb).pid);
1788 		if (sk) {
1789 			sk->sk_err = ENOBUFS;
1790 			sk->sk_error_report(sk);
1791 			sock_put(sk);
1792 		}
1793 		return;
1794 	}
1795 
1796 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1797 			  NLMSG_ERROR, payload, 0);
1798 	errmsg = nlmsg_data(rep);
1799 	errmsg->error = err;
1800 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1801 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1802 }
1803 EXPORT_SYMBOL(netlink_ack);
1804 
1805 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1806 						     struct nlmsghdr *))
1807 {
1808 	struct nlmsghdr *nlh;
1809 	int err;
1810 
1811 	while (skb->len >= nlmsg_total_size(0)) {
1812 		int msglen;
1813 
1814 		nlh = nlmsg_hdr(skb);
1815 		err = 0;
1816 
1817 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1818 			return 0;
1819 
1820 		/* Only requests are handled by the kernel */
1821 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1822 			goto ack;
1823 
1824 		/* Skip control messages */
1825 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1826 			goto ack;
1827 
1828 		err = cb(skb, nlh);
1829 		if (err == -EINTR)
1830 			goto skip;
1831 
1832 ack:
1833 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1834 			netlink_ack(skb, nlh, err);
1835 
1836 skip:
1837 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1838 		if (msglen > skb->len)
1839 			msglen = skb->len;
1840 		skb_pull(skb, msglen);
1841 	}
1842 
1843 	return 0;
1844 }
1845 EXPORT_SYMBOL(netlink_rcv_skb);
1846 
1847 /**
1848  * nlmsg_notify - send a notification netlink message
1849  * @sk: netlink socket to use
1850  * @skb: notification message
1851  * @pid: destination netlink pid for reports or 0
1852  * @group: destination multicast group or 0
1853  * @report: 1 to report back, 0 to disable
1854  * @flags: allocation flags
1855  */
1856 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1857 		 unsigned int group, int report, gfp_t flags)
1858 {
1859 	int err = 0;
1860 
1861 	if (group) {
1862 		int exclude_pid = 0;
1863 
1864 		if (report) {
1865 			atomic_inc(&skb->users);
1866 			exclude_pid = pid;
1867 		}
1868 
1869 		/* errors reported via destination sk->sk_err, but propagate
1870 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1871 		err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1872 	}
1873 
1874 	if (report) {
1875 		int err2;
1876 
1877 		err2 = nlmsg_unicast(sk, skb, pid);
1878 		if (!err || err == -ESRCH)
1879 			err = err2;
1880 	}
1881 
1882 	return err;
1883 }
1884 EXPORT_SYMBOL(nlmsg_notify);
1885 
1886 #ifdef CONFIG_PROC_FS
1887 struct nl_seq_iter {
1888 	struct seq_net_private p;
1889 	int link;
1890 	int hash_idx;
1891 };
1892 
1893 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1894 {
1895 	struct nl_seq_iter *iter = seq->private;
1896 	int i, j;
1897 	struct sock *s;
1898 	struct hlist_node *node;
1899 	loff_t off = 0;
1900 
1901 	for (i = 0; i < MAX_LINKS; i++) {
1902 		struct nl_pid_hash *hash = &nl_table[i].hash;
1903 
1904 		for (j = 0; j <= hash->mask; j++) {
1905 			sk_for_each(s, node, &hash->table[j]) {
1906 				if (sock_net(s) != seq_file_net(seq))
1907 					continue;
1908 				if (off == pos) {
1909 					iter->link = i;
1910 					iter->hash_idx = j;
1911 					return s;
1912 				}
1913 				++off;
1914 			}
1915 		}
1916 	}
1917 	return NULL;
1918 }
1919 
1920 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1921 	__acquires(nl_table_lock)
1922 {
1923 	read_lock(&nl_table_lock);
1924 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1925 }
1926 
1927 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1928 {
1929 	struct sock *s;
1930 	struct nl_seq_iter *iter;
1931 	int i, j;
1932 
1933 	++*pos;
1934 
1935 	if (v == SEQ_START_TOKEN)
1936 		return netlink_seq_socket_idx(seq, 0);
1937 
1938 	iter = seq->private;
1939 	s = v;
1940 	do {
1941 		s = sk_next(s);
1942 	} while (s && sock_net(s) != seq_file_net(seq));
1943 	if (s)
1944 		return s;
1945 
1946 	i = iter->link;
1947 	j = iter->hash_idx + 1;
1948 
1949 	do {
1950 		struct nl_pid_hash *hash = &nl_table[i].hash;
1951 
1952 		for (; j <= hash->mask; j++) {
1953 			s = sk_head(&hash->table[j]);
1954 			while (s && sock_net(s) != seq_file_net(seq))
1955 				s = sk_next(s);
1956 			if (s) {
1957 				iter->link = i;
1958 				iter->hash_idx = j;
1959 				return s;
1960 			}
1961 		}
1962 
1963 		j = 0;
1964 	} while (++i < MAX_LINKS);
1965 
1966 	return NULL;
1967 }
1968 
1969 static void netlink_seq_stop(struct seq_file *seq, void *v)
1970 	__releases(nl_table_lock)
1971 {
1972 	read_unlock(&nl_table_lock);
1973 }
1974 
1975 
1976 static int netlink_seq_show(struct seq_file *seq, void *v)
1977 {
1978 	if (v == SEQ_START_TOKEN)
1979 		seq_puts(seq,
1980 			 "sk       Eth Pid    Groups   "
1981 			 "Rmem     Wmem     Dump     Locks     Drops\n");
1982 	else {
1983 		struct sock *s = v;
1984 		struct netlink_sock *nlk = nlk_sk(s);
1985 
1986 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %-8d %-8d\n",
1987 			   s,
1988 			   s->sk_protocol,
1989 			   nlk->pid,
1990 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1991 			   sk_rmem_alloc_get(s),
1992 			   sk_wmem_alloc_get(s),
1993 			   nlk->cb,
1994 			   atomic_read(&s->sk_refcnt),
1995 			   atomic_read(&s->sk_drops)
1996 			);
1997 
1998 	}
1999 	return 0;
2000 }
2001 
2002 static const struct seq_operations netlink_seq_ops = {
2003 	.start  = netlink_seq_start,
2004 	.next   = netlink_seq_next,
2005 	.stop   = netlink_seq_stop,
2006 	.show   = netlink_seq_show,
2007 };
2008 
2009 
2010 static int netlink_seq_open(struct inode *inode, struct file *file)
2011 {
2012 	return seq_open_net(inode, file, &netlink_seq_ops,
2013 				sizeof(struct nl_seq_iter));
2014 }
2015 
2016 static const struct file_operations netlink_seq_fops = {
2017 	.owner		= THIS_MODULE,
2018 	.open		= netlink_seq_open,
2019 	.read		= seq_read,
2020 	.llseek		= seq_lseek,
2021 	.release	= seq_release_net,
2022 };
2023 
2024 #endif
2025 
2026 int netlink_register_notifier(struct notifier_block *nb)
2027 {
2028 	return atomic_notifier_chain_register(&netlink_chain, nb);
2029 }
2030 EXPORT_SYMBOL(netlink_register_notifier);
2031 
2032 int netlink_unregister_notifier(struct notifier_block *nb)
2033 {
2034 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
2035 }
2036 EXPORT_SYMBOL(netlink_unregister_notifier);
2037 
2038 static const struct proto_ops netlink_ops = {
2039 	.family =	PF_NETLINK,
2040 	.owner =	THIS_MODULE,
2041 	.release =	netlink_release,
2042 	.bind =		netlink_bind,
2043 	.connect =	netlink_connect,
2044 	.socketpair =	sock_no_socketpair,
2045 	.accept =	sock_no_accept,
2046 	.getname =	netlink_getname,
2047 	.poll =		datagram_poll,
2048 	.ioctl =	sock_no_ioctl,
2049 	.listen =	sock_no_listen,
2050 	.shutdown =	sock_no_shutdown,
2051 	.setsockopt =	netlink_setsockopt,
2052 	.getsockopt =	netlink_getsockopt,
2053 	.sendmsg =	netlink_sendmsg,
2054 	.recvmsg =	netlink_recvmsg,
2055 	.mmap =		sock_no_mmap,
2056 	.sendpage =	sock_no_sendpage,
2057 };
2058 
2059 static const struct net_proto_family netlink_family_ops = {
2060 	.family = PF_NETLINK,
2061 	.create = netlink_create,
2062 	.owner	= THIS_MODULE,	/* for consistency 8) */
2063 };
2064 
2065 static int __net_init netlink_net_init(struct net *net)
2066 {
2067 #ifdef CONFIG_PROC_FS
2068 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2069 		return -ENOMEM;
2070 #endif
2071 	return 0;
2072 }
2073 
2074 static void __net_exit netlink_net_exit(struct net *net)
2075 {
2076 #ifdef CONFIG_PROC_FS
2077 	proc_net_remove(net, "netlink");
2078 #endif
2079 }
2080 
2081 static struct pernet_operations __net_initdata netlink_net_ops = {
2082 	.init = netlink_net_init,
2083 	.exit = netlink_net_exit,
2084 };
2085 
2086 static int __init netlink_proto_init(void)
2087 {
2088 	struct sk_buff *dummy_skb;
2089 	int i;
2090 	unsigned long limit;
2091 	unsigned int order;
2092 	int err = proto_register(&netlink_proto, 0);
2093 
2094 	if (err != 0)
2095 		goto out;
2096 
2097 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2098 
2099 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2100 	if (!nl_table)
2101 		goto panic;
2102 
2103 	if (totalram_pages >= (128 * 1024))
2104 		limit = totalram_pages >> (21 - PAGE_SHIFT);
2105 	else
2106 		limit = totalram_pages >> (23 - PAGE_SHIFT);
2107 
2108 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2109 	limit = (1UL << order) / sizeof(struct hlist_head);
2110 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2111 
2112 	for (i = 0; i < MAX_LINKS; i++) {
2113 		struct nl_pid_hash *hash = &nl_table[i].hash;
2114 
2115 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2116 		if (!hash->table) {
2117 			while (i-- > 0)
2118 				nl_pid_hash_free(nl_table[i].hash.table,
2119 						 1 * sizeof(*hash->table));
2120 			kfree(nl_table);
2121 			goto panic;
2122 		}
2123 		hash->max_shift = order;
2124 		hash->shift = 0;
2125 		hash->mask = 0;
2126 		hash->rehash_time = jiffies;
2127 	}
2128 
2129 	sock_register(&netlink_family_ops);
2130 	register_pernet_subsys(&netlink_net_ops);
2131 	/* The netlink device handler may be needed early. */
2132 	rtnetlink_init();
2133 out:
2134 	return err;
2135 panic:
2136 	panic("netlink_init: Cannot allocate nl_table\n");
2137 }
2138 
2139 core_initcall(netlink_proto_init);
2140