xref: /linux/net/netlink/af_netlink.c (revision b233b28eac0cc37d07c2d007ea08c86c778c5af4)
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 #define NETLINK_KERNEL_SOCKET	0x1
87 #define NETLINK_RECV_PKTINFO	0x2
88 
89 static inline struct netlink_sock *nlk_sk(struct sock *sk)
90 {
91 	return container_of(sk, struct netlink_sock, sk);
92 }
93 
94 static inline int netlink_is_kernel(struct sock *sk)
95 {
96 	return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
97 }
98 
99 struct nl_pid_hash {
100 	struct hlist_head *table;
101 	unsigned long rehash_time;
102 
103 	unsigned int mask;
104 	unsigned int shift;
105 
106 	unsigned int entries;
107 	unsigned int max_shift;
108 
109 	u32 rnd;
110 };
111 
112 struct netlink_table {
113 	struct nl_pid_hash hash;
114 	struct hlist_head mc_list;
115 	unsigned long *listeners;
116 	unsigned int nl_nonroot;
117 	unsigned int groups;
118 	struct mutex *cb_mutex;
119 	struct module *module;
120 	int registered;
121 };
122 
123 static struct netlink_table *nl_table;
124 
125 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
126 
127 static int netlink_dump(struct sock *sk);
128 static void netlink_destroy_callback(struct netlink_callback *cb);
129 
130 static DEFINE_RWLOCK(nl_table_lock);
131 static atomic_t nl_table_users = ATOMIC_INIT(0);
132 
133 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
134 
135 static u32 netlink_group_mask(u32 group)
136 {
137 	return group ? 1 << (group - 1) : 0;
138 }
139 
140 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
141 {
142 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
143 }
144 
145 static void netlink_sock_destruct(struct sock *sk)
146 {
147 	struct netlink_sock *nlk = nlk_sk(sk);
148 
149 	if (nlk->cb) {
150 		if (nlk->cb->done)
151 			nlk->cb->done(nlk->cb);
152 		netlink_destroy_callback(nlk->cb);
153 	}
154 
155 	skb_queue_purge(&sk->sk_receive_queue);
156 
157 	if (!sock_flag(sk, SOCK_DEAD)) {
158 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
159 		return;
160 	}
161 
162 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
163 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
164 	WARN_ON(nlk_sk(sk)->groups);
165 }
166 
167 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
168  * SMP. Look, when several writers sleep and reader wakes them up, all but one
169  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
170  * this, _but_ remember, it adds useless work on UP machines.
171  */
172 
173 static void netlink_table_grab(void)
174 	__acquires(nl_table_lock)
175 {
176 	write_lock_irq(&nl_table_lock);
177 
178 	if (atomic_read(&nl_table_users)) {
179 		DECLARE_WAITQUEUE(wait, current);
180 
181 		add_wait_queue_exclusive(&nl_table_wait, &wait);
182 		for (;;) {
183 			set_current_state(TASK_UNINTERRUPTIBLE);
184 			if (atomic_read(&nl_table_users) == 0)
185 				break;
186 			write_unlock_irq(&nl_table_lock);
187 			schedule();
188 			write_lock_irq(&nl_table_lock);
189 		}
190 
191 		__set_current_state(TASK_RUNNING);
192 		remove_wait_queue(&nl_table_wait, &wait);
193 	}
194 }
195 
196 static void netlink_table_ungrab(void)
197 	__releases(nl_table_lock)
198 {
199 	write_unlock_irq(&nl_table_lock);
200 	wake_up(&nl_table_wait);
201 }
202 
203 static inline void
204 netlink_lock_table(void)
205 {
206 	/* read_lock() synchronizes us to netlink_table_grab */
207 
208 	read_lock(&nl_table_lock);
209 	atomic_inc(&nl_table_users);
210 	read_unlock(&nl_table_lock);
211 }
212 
213 static inline void
214 netlink_unlock_table(void)
215 {
216 	if (atomic_dec_and_test(&nl_table_users))
217 		wake_up(&nl_table_wait);
218 }
219 
220 static inline struct sock *netlink_lookup(struct net *net, int protocol,
221 					  u32 pid)
222 {
223 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
224 	struct hlist_head *head;
225 	struct sock *sk;
226 	struct hlist_node *node;
227 
228 	read_lock(&nl_table_lock);
229 	head = nl_pid_hashfn(hash, pid);
230 	sk_for_each(sk, node, head) {
231 		if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
232 			sock_hold(sk);
233 			goto found;
234 		}
235 	}
236 	sk = NULL;
237 found:
238 	read_unlock(&nl_table_lock);
239 	return sk;
240 }
241 
242 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
243 {
244 	if (size <= PAGE_SIZE)
245 		return kzalloc(size, GFP_ATOMIC);
246 	else
247 		return (struct hlist_head *)
248 			__get_free_pages(GFP_ATOMIC | __GFP_ZERO,
249 					 get_order(size));
250 }
251 
252 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
253 {
254 	if (size <= PAGE_SIZE)
255 		kfree(table);
256 	else
257 		free_pages((unsigned long)table, get_order(size));
258 }
259 
260 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
261 {
262 	unsigned int omask, mask, shift;
263 	size_t osize, size;
264 	struct hlist_head *otable, *table;
265 	int i;
266 
267 	omask = mask = hash->mask;
268 	osize = size = (mask + 1) * sizeof(*table);
269 	shift = hash->shift;
270 
271 	if (grow) {
272 		if (++shift > hash->max_shift)
273 			return 0;
274 		mask = mask * 2 + 1;
275 		size *= 2;
276 	}
277 
278 	table = nl_pid_hash_zalloc(size);
279 	if (!table)
280 		return 0;
281 
282 	otable = hash->table;
283 	hash->table = table;
284 	hash->mask = mask;
285 	hash->shift = shift;
286 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
287 
288 	for (i = 0; i <= omask; i++) {
289 		struct sock *sk;
290 		struct hlist_node *node, *tmp;
291 
292 		sk_for_each_safe(sk, node, tmp, &otable[i])
293 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
294 	}
295 
296 	nl_pid_hash_free(otable, osize);
297 	hash->rehash_time = jiffies + 10 * 60 * HZ;
298 	return 1;
299 }
300 
301 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
302 {
303 	int avg = hash->entries >> hash->shift;
304 
305 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
306 		return 1;
307 
308 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
309 		nl_pid_hash_rehash(hash, 0);
310 		return 1;
311 	}
312 
313 	return 0;
314 }
315 
316 static const struct proto_ops netlink_ops;
317 
318 static void
319 netlink_update_listeners(struct sock *sk)
320 {
321 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
322 	struct hlist_node *node;
323 	unsigned long mask;
324 	unsigned int i;
325 
326 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
327 		mask = 0;
328 		sk_for_each_bound(sk, node, &tbl->mc_list) {
329 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
330 				mask |= nlk_sk(sk)->groups[i];
331 		}
332 		tbl->listeners[i] = mask;
333 	}
334 	/* this function is only called with the netlink table "grabbed", which
335 	 * makes sure updates are visible before bind or setsockopt return. */
336 }
337 
338 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
339 {
340 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
341 	struct hlist_head *head;
342 	int err = -EADDRINUSE;
343 	struct sock *osk;
344 	struct hlist_node *node;
345 	int len;
346 
347 	netlink_table_grab();
348 	head = nl_pid_hashfn(hash, pid);
349 	len = 0;
350 	sk_for_each(osk, node, head) {
351 		if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
352 			break;
353 		len++;
354 	}
355 	if (node)
356 		goto err;
357 
358 	err = -EBUSY;
359 	if (nlk_sk(sk)->pid)
360 		goto err;
361 
362 	err = -ENOMEM;
363 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
364 		goto err;
365 
366 	if (len && nl_pid_hash_dilute(hash, len))
367 		head = nl_pid_hashfn(hash, pid);
368 	hash->entries++;
369 	nlk_sk(sk)->pid = pid;
370 	sk_add_node(sk, head);
371 	err = 0;
372 
373 err:
374 	netlink_table_ungrab();
375 	return err;
376 }
377 
378 static void netlink_remove(struct sock *sk)
379 {
380 	netlink_table_grab();
381 	if (sk_del_node_init(sk))
382 		nl_table[sk->sk_protocol].hash.entries--;
383 	if (nlk_sk(sk)->subscriptions)
384 		__sk_del_bind_node(sk);
385 	netlink_table_ungrab();
386 }
387 
388 static struct proto netlink_proto = {
389 	.name	  = "NETLINK",
390 	.owner	  = THIS_MODULE,
391 	.obj_size = sizeof(struct netlink_sock),
392 };
393 
394 static int __netlink_create(struct net *net, struct socket *sock,
395 			    struct mutex *cb_mutex, int protocol)
396 {
397 	struct sock *sk;
398 	struct netlink_sock *nlk;
399 
400 	sock->ops = &netlink_ops;
401 
402 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
403 	if (!sk)
404 		return -ENOMEM;
405 
406 	sock_init_data(sock, sk);
407 
408 	nlk = nlk_sk(sk);
409 	if (cb_mutex)
410 		nlk->cb_mutex = cb_mutex;
411 	else {
412 		nlk->cb_mutex = &nlk->cb_def_mutex;
413 		mutex_init(nlk->cb_mutex);
414 	}
415 	init_waitqueue_head(&nlk->wait);
416 
417 	sk->sk_destruct = netlink_sock_destruct;
418 	sk->sk_protocol = protocol;
419 	return 0;
420 }
421 
422 static int netlink_create(struct net *net, struct socket *sock, int protocol)
423 {
424 	struct module *module = NULL;
425 	struct mutex *cb_mutex;
426 	struct netlink_sock *nlk;
427 	int err = 0;
428 
429 	sock->state = SS_UNCONNECTED;
430 
431 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
432 		return -ESOCKTNOSUPPORT;
433 
434 	if (protocol < 0 || protocol >= MAX_LINKS)
435 		return -EPROTONOSUPPORT;
436 
437 	netlink_lock_table();
438 #ifdef CONFIG_MODULES
439 	if (!nl_table[protocol].registered) {
440 		netlink_unlock_table();
441 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
442 		netlink_lock_table();
443 	}
444 #endif
445 	if (nl_table[protocol].registered &&
446 	    try_module_get(nl_table[protocol].module))
447 		module = nl_table[protocol].module;
448 	cb_mutex = nl_table[protocol].cb_mutex;
449 	netlink_unlock_table();
450 
451 	err = __netlink_create(net, sock, cb_mutex, protocol);
452 	if (err < 0)
453 		goto out_module;
454 
455 	local_bh_disable();
456 	sock_prot_inuse_add(net, &netlink_proto, 1);
457 	local_bh_enable();
458 
459 	nlk = nlk_sk(sock->sk);
460 	nlk->module = module;
461 out:
462 	return err;
463 
464 out_module:
465 	module_put(module);
466 	goto out;
467 }
468 
469 static int netlink_release(struct socket *sock)
470 {
471 	struct sock *sk = sock->sk;
472 	struct netlink_sock *nlk;
473 
474 	if (!sk)
475 		return 0;
476 
477 	netlink_remove(sk);
478 	sock_orphan(sk);
479 	nlk = nlk_sk(sk);
480 
481 	/*
482 	 * OK. Socket is unlinked, any packets that arrive now
483 	 * will be purged.
484 	 */
485 
486 	sock->sk = NULL;
487 	wake_up_interruptible_all(&nlk->wait);
488 
489 	skb_queue_purge(&sk->sk_write_queue);
490 
491 	if (nlk->pid && !nlk->subscriptions) {
492 		struct netlink_notify n = {
493 						.net = sock_net(sk),
494 						.protocol = sk->sk_protocol,
495 						.pid = nlk->pid,
496 					  };
497 		atomic_notifier_call_chain(&netlink_chain,
498 				NETLINK_URELEASE, &n);
499 	}
500 
501 	module_put(nlk->module);
502 
503 	netlink_table_grab();
504 	if (netlink_is_kernel(sk)) {
505 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
506 		if (--nl_table[sk->sk_protocol].registered == 0) {
507 			kfree(nl_table[sk->sk_protocol].listeners);
508 			nl_table[sk->sk_protocol].module = NULL;
509 			nl_table[sk->sk_protocol].registered = 0;
510 		}
511 	} else if (nlk->subscriptions)
512 		netlink_update_listeners(sk);
513 	netlink_table_ungrab();
514 
515 	kfree(nlk->groups);
516 	nlk->groups = NULL;
517 
518 	local_bh_disable();
519 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
520 	local_bh_enable();
521 	sock_put(sk);
522 	return 0;
523 }
524 
525 static int netlink_autobind(struct socket *sock)
526 {
527 	struct sock *sk = sock->sk;
528 	struct net *net = sock_net(sk);
529 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
530 	struct hlist_head *head;
531 	struct sock *osk;
532 	struct hlist_node *node;
533 	s32 pid = current->tgid;
534 	int err;
535 	static s32 rover = -4097;
536 
537 retry:
538 	cond_resched();
539 	netlink_table_grab();
540 	head = nl_pid_hashfn(hash, pid);
541 	sk_for_each(osk, node, head) {
542 		if (!net_eq(sock_net(osk), net))
543 			continue;
544 		if (nlk_sk(osk)->pid == pid) {
545 			/* Bind collision, search negative pid values. */
546 			pid = rover--;
547 			if (rover > -4097)
548 				rover = -4097;
549 			netlink_table_ungrab();
550 			goto retry;
551 		}
552 	}
553 	netlink_table_ungrab();
554 
555 	err = netlink_insert(sk, net, pid);
556 	if (err == -EADDRINUSE)
557 		goto retry;
558 
559 	/* If 2 threads race to autobind, that is fine.  */
560 	if (err == -EBUSY)
561 		err = 0;
562 
563 	return err;
564 }
565 
566 static inline int netlink_capable(struct socket *sock, unsigned int flag)
567 {
568 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
569 	       capable(CAP_NET_ADMIN);
570 }
571 
572 static void
573 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
574 {
575 	struct netlink_sock *nlk = nlk_sk(sk);
576 
577 	if (nlk->subscriptions && !subscriptions)
578 		__sk_del_bind_node(sk);
579 	else if (!nlk->subscriptions && subscriptions)
580 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
581 	nlk->subscriptions = subscriptions;
582 }
583 
584 static int netlink_realloc_groups(struct sock *sk)
585 {
586 	struct netlink_sock *nlk = nlk_sk(sk);
587 	unsigned int groups;
588 	unsigned long *new_groups;
589 	int err = 0;
590 
591 	netlink_table_grab();
592 
593 	groups = nl_table[sk->sk_protocol].groups;
594 	if (!nl_table[sk->sk_protocol].registered) {
595 		err = -ENOENT;
596 		goto out_unlock;
597 	}
598 
599 	if (nlk->ngroups >= groups)
600 		goto out_unlock;
601 
602 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
603 	if (new_groups == NULL) {
604 		err = -ENOMEM;
605 		goto out_unlock;
606 	}
607 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
608 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
609 
610 	nlk->groups = new_groups;
611 	nlk->ngroups = groups;
612  out_unlock:
613 	netlink_table_ungrab();
614 	return err;
615 }
616 
617 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
618 			int addr_len)
619 {
620 	struct sock *sk = sock->sk;
621 	struct net *net = sock_net(sk);
622 	struct netlink_sock *nlk = nlk_sk(sk);
623 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
624 	int err;
625 
626 	if (nladdr->nl_family != AF_NETLINK)
627 		return -EINVAL;
628 
629 	/* Only superuser is allowed to listen multicasts */
630 	if (nladdr->nl_groups) {
631 		if (!netlink_capable(sock, NL_NONROOT_RECV))
632 			return -EPERM;
633 		err = netlink_realloc_groups(sk);
634 		if (err)
635 			return err;
636 	}
637 
638 	if (nlk->pid) {
639 		if (nladdr->nl_pid != nlk->pid)
640 			return -EINVAL;
641 	} else {
642 		err = nladdr->nl_pid ?
643 			netlink_insert(sk, net, nladdr->nl_pid) :
644 			netlink_autobind(sock);
645 		if (err)
646 			return err;
647 	}
648 
649 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
650 		return 0;
651 
652 	netlink_table_grab();
653 	netlink_update_subscriptions(sk, nlk->subscriptions +
654 					 hweight32(nladdr->nl_groups) -
655 					 hweight32(nlk->groups[0]));
656 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
657 	netlink_update_listeners(sk);
658 	netlink_table_ungrab();
659 
660 	return 0;
661 }
662 
663 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
664 			   int alen, int flags)
665 {
666 	int err = 0;
667 	struct sock *sk = sock->sk;
668 	struct netlink_sock *nlk = nlk_sk(sk);
669 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
670 
671 	if (addr->sa_family == AF_UNSPEC) {
672 		sk->sk_state	= NETLINK_UNCONNECTED;
673 		nlk->dst_pid	= 0;
674 		nlk->dst_group  = 0;
675 		return 0;
676 	}
677 	if (addr->sa_family != AF_NETLINK)
678 		return -EINVAL;
679 
680 	/* Only superuser is allowed to send multicasts */
681 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
682 		return -EPERM;
683 
684 	if (!nlk->pid)
685 		err = netlink_autobind(sock);
686 
687 	if (err == 0) {
688 		sk->sk_state	= NETLINK_CONNECTED;
689 		nlk->dst_pid 	= nladdr->nl_pid;
690 		nlk->dst_group  = ffs(nladdr->nl_groups);
691 	}
692 
693 	return err;
694 }
695 
696 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
697 			   int *addr_len, int peer)
698 {
699 	struct sock *sk = sock->sk;
700 	struct netlink_sock *nlk = nlk_sk(sk);
701 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
702 
703 	nladdr->nl_family = AF_NETLINK;
704 	nladdr->nl_pad = 0;
705 	*addr_len = sizeof(*nladdr);
706 
707 	if (peer) {
708 		nladdr->nl_pid = nlk->dst_pid;
709 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
710 	} else {
711 		nladdr->nl_pid = nlk->pid;
712 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
713 	}
714 	return 0;
715 }
716 
717 static void netlink_overrun(struct sock *sk)
718 {
719 	if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
720 		sk->sk_err = ENOBUFS;
721 		sk->sk_error_report(sk);
722 	}
723 }
724 
725 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
726 {
727 	struct sock *sock;
728 	struct netlink_sock *nlk;
729 
730 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
731 	if (!sock)
732 		return ERR_PTR(-ECONNREFUSED);
733 
734 	/* Don't bother queuing skb if kernel socket has no input function */
735 	nlk = nlk_sk(sock);
736 	if (sock->sk_state == NETLINK_CONNECTED &&
737 	    nlk->dst_pid != nlk_sk(ssk)->pid) {
738 		sock_put(sock);
739 		return ERR_PTR(-ECONNREFUSED);
740 	}
741 	return sock;
742 }
743 
744 struct sock *netlink_getsockbyfilp(struct file *filp)
745 {
746 	struct inode *inode = filp->f_path.dentry->d_inode;
747 	struct sock *sock;
748 
749 	if (!S_ISSOCK(inode->i_mode))
750 		return ERR_PTR(-ENOTSOCK);
751 
752 	sock = SOCKET_I(inode)->sk;
753 	if (sock->sk_family != AF_NETLINK)
754 		return ERR_PTR(-EINVAL);
755 
756 	sock_hold(sock);
757 	return sock;
758 }
759 
760 /*
761  * Attach a skb to a netlink socket.
762  * The caller must hold a reference to the destination socket. On error, the
763  * reference is dropped. The skb is not send to the destination, just all
764  * all error checks are performed and memory in the queue is reserved.
765  * Return values:
766  * < 0: error. skb freed, reference to sock dropped.
767  * 0: continue
768  * 1: repeat lookup - reference dropped while waiting for socket memory.
769  */
770 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
771 		      long *timeo, struct sock *ssk)
772 {
773 	struct netlink_sock *nlk;
774 
775 	nlk = nlk_sk(sk);
776 
777 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
778 	    test_bit(0, &nlk->state)) {
779 		DECLARE_WAITQUEUE(wait, current);
780 		if (!*timeo) {
781 			if (!ssk || netlink_is_kernel(ssk))
782 				netlink_overrun(sk);
783 			sock_put(sk);
784 			kfree_skb(skb);
785 			return -EAGAIN;
786 		}
787 
788 		__set_current_state(TASK_INTERRUPTIBLE);
789 		add_wait_queue(&nlk->wait, &wait);
790 
791 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
792 		     test_bit(0, &nlk->state)) &&
793 		    !sock_flag(sk, SOCK_DEAD))
794 			*timeo = schedule_timeout(*timeo);
795 
796 		__set_current_state(TASK_RUNNING);
797 		remove_wait_queue(&nlk->wait, &wait);
798 		sock_put(sk);
799 
800 		if (signal_pending(current)) {
801 			kfree_skb(skb);
802 			return sock_intr_errno(*timeo);
803 		}
804 		return 1;
805 	}
806 	skb_set_owner_r(skb, sk);
807 	return 0;
808 }
809 
810 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
811 {
812 	int len = skb->len;
813 
814 	skb_queue_tail(&sk->sk_receive_queue, skb);
815 	sk->sk_data_ready(sk, len);
816 	sock_put(sk);
817 	return len;
818 }
819 
820 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
821 {
822 	kfree_skb(skb);
823 	sock_put(sk);
824 }
825 
826 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
827 					   gfp_t allocation)
828 {
829 	int delta;
830 
831 	skb_orphan(skb);
832 
833 	delta = skb->end - skb->tail;
834 	if (delta * 2 < skb->truesize)
835 		return skb;
836 
837 	if (skb_shared(skb)) {
838 		struct sk_buff *nskb = skb_clone(skb, allocation);
839 		if (!nskb)
840 			return skb;
841 		kfree_skb(skb);
842 		skb = nskb;
843 	}
844 
845 	if (!pskb_expand_head(skb, 0, -delta, allocation))
846 		skb->truesize -= delta;
847 
848 	return skb;
849 }
850 
851 static inline void netlink_rcv_wake(struct sock *sk)
852 {
853 	struct netlink_sock *nlk = nlk_sk(sk);
854 
855 	if (skb_queue_empty(&sk->sk_receive_queue))
856 		clear_bit(0, &nlk->state);
857 	if (!test_bit(0, &nlk->state))
858 		wake_up_interruptible(&nlk->wait);
859 }
860 
861 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
862 {
863 	int ret;
864 	struct netlink_sock *nlk = nlk_sk(sk);
865 
866 	ret = -ECONNREFUSED;
867 	if (nlk->netlink_rcv != NULL) {
868 		ret = skb->len;
869 		skb_set_owner_r(skb, sk);
870 		nlk->netlink_rcv(skb);
871 	}
872 	kfree_skb(skb);
873 	sock_put(sk);
874 	return ret;
875 }
876 
877 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
878 		    u32 pid, int nonblock)
879 {
880 	struct sock *sk;
881 	int err;
882 	long timeo;
883 
884 	skb = netlink_trim(skb, gfp_any());
885 
886 	timeo = sock_sndtimeo(ssk, nonblock);
887 retry:
888 	sk = netlink_getsockbypid(ssk, pid);
889 	if (IS_ERR(sk)) {
890 		kfree_skb(skb);
891 		return PTR_ERR(sk);
892 	}
893 	if (netlink_is_kernel(sk))
894 		return netlink_unicast_kernel(sk, skb);
895 
896 	if (sk_filter(sk, skb)) {
897 		err = skb->len;
898 		kfree_skb(skb);
899 		sock_put(sk);
900 		return err;
901 	}
902 
903 	err = netlink_attachskb(sk, skb, &timeo, ssk);
904 	if (err == 1)
905 		goto retry;
906 	if (err)
907 		return err;
908 
909 	return netlink_sendskb(sk, skb);
910 }
911 EXPORT_SYMBOL(netlink_unicast);
912 
913 int netlink_has_listeners(struct sock *sk, unsigned int group)
914 {
915 	int res = 0;
916 	unsigned long *listeners;
917 
918 	BUG_ON(!netlink_is_kernel(sk));
919 
920 	rcu_read_lock();
921 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
922 
923 	if (group - 1 < nl_table[sk->sk_protocol].groups)
924 		res = test_bit(group - 1, listeners);
925 
926 	rcu_read_unlock();
927 
928 	return res;
929 }
930 EXPORT_SYMBOL_GPL(netlink_has_listeners);
931 
932 static inline int netlink_broadcast_deliver(struct sock *sk,
933 					    struct sk_buff *skb)
934 {
935 	struct netlink_sock *nlk = nlk_sk(sk);
936 
937 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
938 	    !test_bit(0, &nlk->state)) {
939 		skb_set_owner_r(skb, sk);
940 		skb_queue_tail(&sk->sk_receive_queue, skb);
941 		sk->sk_data_ready(sk, skb->len);
942 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
943 	}
944 	return -1;
945 }
946 
947 struct netlink_broadcast_data {
948 	struct sock *exclude_sk;
949 	struct net *net;
950 	u32 pid;
951 	u32 group;
952 	int failure;
953 	int congested;
954 	int delivered;
955 	gfp_t allocation;
956 	struct sk_buff *skb, *skb2;
957 };
958 
959 static inline int do_one_broadcast(struct sock *sk,
960 				   struct netlink_broadcast_data *p)
961 {
962 	struct netlink_sock *nlk = nlk_sk(sk);
963 	int val;
964 
965 	if (p->exclude_sk == sk)
966 		goto out;
967 
968 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
969 	    !test_bit(p->group - 1, nlk->groups))
970 		goto out;
971 
972 	if (!net_eq(sock_net(sk), p->net))
973 		goto out;
974 
975 	if (p->failure) {
976 		netlink_overrun(sk);
977 		goto out;
978 	}
979 
980 	sock_hold(sk);
981 	if (p->skb2 == NULL) {
982 		if (skb_shared(p->skb)) {
983 			p->skb2 = skb_clone(p->skb, p->allocation);
984 		} else {
985 			p->skb2 = skb_get(p->skb);
986 			/*
987 			 * skb ownership may have been set when
988 			 * delivered to a previous socket.
989 			 */
990 			skb_orphan(p->skb2);
991 		}
992 	}
993 	if (p->skb2 == NULL) {
994 		netlink_overrun(sk);
995 		/* Clone failed. Notify ALL listeners. */
996 		p->failure = 1;
997 	} else if (sk_filter(sk, p->skb2)) {
998 		kfree_skb(p->skb2);
999 		p->skb2 = NULL;
1000 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1001 		netlink_overrun(sk);
1002 	} else {
1003 		p->congested |= val;
1004 		p->delivered = 1;
1005 		p->skb2 = NULL;
1006 	}
1007 	sock_put(sk);
1008 
1009 out:
1010 	return 0;
1011 }
1012 
1013 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1014 		      u32 group, gfp_t allocation)
1015 {
1016 	struct net *net = sock_net(ssk);
1017 	struct netlink_broadcast_data info;
1018 	struct hlist_node *node;
1019 	struct sock *sk;
1020 
1021 	skb = netlink_trim(skb, allocation);
1022 
1023 	info.exclude_sk = ssk;
1024 	info.net = net;
1025 	info.pid = pid;
1026 	info.group = group;
1027 	info.failure = 0;
1028 	info.congested = 0;
1029 	info.delivered = 0;
1030 	info.allocation = allocation;
1031 	info.skb = skb;
1032 	info.skb2 = NULL;
1033 
1034 	/* While we sleep in clone, do not allow to change socket list */
1035 
1036 	netlink_lock_table();
1037 
1038 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1039 		do_one_broadcast(sk, &info);
1040 
1041 	kfree_skb(skb);
1042 
1043 	netlink_unlock_table();
1044 
1045 	if (info.skb2)
1046 		kfree_skb(info.skb2);
1047 
1048 	if (info.delivered) {
1049 		if (info.congested && (allocation & __GFP_WAIT))
1050 			yield();
1051 		return 0;
1052 	}
1053 	if (info.failure)
1054 		return -ENOBUFS;
1055 	return -ESRCH;
1056 }
1057 EXPORT_SYMBOL(netlink_broadcast);
1058 
1059 struct netlink_set_err_data {
1060 	struct sock *exclude_sk;
1061 	u32 pid;
1062 	u32 group;
1063 	int code;
1064 };
1065 
1066 static inline int do_one_set_err(struct sock *sk,
1067 				 struct netlink_set_err_data *p)
1068 {
1069 	struct netlink_sock *nlk = nlk_sk(sk);
1070 
1071 	if (sk == p->exclude_sk)
1072 		goto out;
1073 
1074 	if (sock_net(sk) != sock_net(p->exclude_sk))
1075 		goto out;
1076 
1077 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1078 	    !test_bit(p->group - 1, nlk->groups))
1079 		goto out;
1080 
1081 	sk->sk_err = p->code;
1082 	sk->sk_error_report(sk);
1083 out:
1084 	return 0;
1085 }
1086 
1087 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1088 {
1089 	struct netlink_set_err_data info;
1090 	struct hlist_node *node;
1091 	struct sock *sk;
1092 
1093 	info.exclude_sk = ssk;
1094 	info.pid = pid;
1095 	info.group = group;
1096 	info.code = code;
1097 
1098 	read_lock(&nl_table_lock);
1099 
1100 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1101 		do_one_set_err(sk, &info);
1102 
1103 	read_unlock(&nl_table_lock);
1104 }
1105 
1106 /* must be called with netlink table grabbed */
1107 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1108 				     unsigned int group,
1109 				     int is_new)
1110 {
1111 	int old, new = !!is_new, subscriptions;
1112 
1113 	old = test_bit(group - 1, nlk->groups);
1114 	subscriptions = nlk->subscriptions - old + new;
1115 	if (new)
1116 		__set_bit(group - 1, nlk->groups);
1117 	else
1118 		__clear_bit(group - 1, nlk->groups);
1119 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1120 	netlink_update_listeners(&nlk->sk);
1121 }
1122 
1123 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1124 			      char __user *optval, int optlen)
1125 {
1126 	struct sock *sk = sock->sk;
1127 	struct netlink_sock *nlk = nlk_sk(sk);
1128 	unsigned int val = 0;
1129 	int err;
1130 
1131 	if (level != SOL_NETLINK)
1132 		return -ENOPROTOOPT;
1133 
1134 	if (optlen >= sizeof(int) &&
1135 	    get_user(val, (unsigned int __user *)optval))
1136 		return -EFAULT;
1137 
1138 	switch (optname) {
1139 	case NETLINK_PKTINFO:
1140 		if (val)
1141 			nlk->flags |= NETLINK_RECV_PKTINFO;
1142 		else
1143 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1144 		err = 0;
1145 		break;
1146 	case NETLINK_ADD_MEMBERSHIP:
1147 	case NETLINK_DROP_MEMBERSHIP: {
1148 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1149 			return -EPERM;
1150 		err = netlink_realloc_groups(sk);
1151 		if (err)
1152 			return err;
1153 		if (!val || val - 1 >= nlk->ngroups)
1154 			return -EINVAL;
1155 		netlink_table_grab();
1156 		netlink_update_socket_mc(nlk, val,
1157 					 optname == NETLINK_ADD_MEMBERSHIP);
1158 		netlink_table_ungrab();
1159 		err = 0;
1160 		break;
1161 	}
1162 	default:
1163 		err = -ENOPROTOOPT;
1164 	}
1165 	return err;
1166 }
1167 
1168 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1169 			      char __user *optval, int __user *optlen)
1170 {
1171 	struct sock *sk = sock->sk;
1172 	struct netlink_sock *nlk = nlk_sk(sk);
1173 	int len, val, err;
1174 
1175 	if (level != SOL_NETLINK)
1176 		return -ENOPROTOOPT;
1177 
1178 	if (get_user(len, optlen))
1179 		return -EFAULT;
1180 	if (len < 0)
1181 		return -EINVAL;
1182 
1183 	switch (optname) {
1184 	case NETLINK_PKTINFO:
1185 		if (len < sizeof(int))
1186 			return -EINVAL;
1187 		len = sizeof(int);
1188 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1189 		if (put_user(len, optlen) ||
1190 		    put_user(val, optval))
1191 			return -EFAULT;
1192 		err = 0;
1193 		break;
1194 	default:
1195 		err = -ENOPROTOOPT;
1196 	}
1197 	return err;
1198 }
1199 
1200 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1201 {
1202 	struct nl_pktinfo info;
1203 
1204 	info.group = NETLINK_CB(skb).dst_group;
1205 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1206 }
1207 
1208 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1209 			   struct msghdr *msg, size_t len)
1210 {
1211 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1212 	struct sock *sk = sock->sk;
1213 	struct netlink_sock *nlk = nlk_sk(sk);
1214 	struct sockaddr_nl *addr = msg->msg_name;
1215 	u32 dst_pid;
1216 	u32 dst_group;
1217 	struct sk_buff *skb;
1218 	int err;
1219 	struct scm_cookie scm;
1220 
1221 	if (msg->msg_flags&MSG_OOB)
1222 		return -EOPNOTSUPP;
1223 
1224 	if (NULL == siocb->scm)
1225 		siocb->scm = &scm;
1226 	err = scm_send(sock, msg, siocb->scm);
1227 	if (err < 0)
1228 		return err;
1229 
1230 	if (msg->msg_namelen) {
1231 		if (addr->nl_family != AF_NETLINK)
1232 			return -EINVAL;
1233 		dst_pid = addr->nl_pid;
1234 		dst_group = ffs(addr->nl_groups);
1235 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1236 			return -EPERM;
1237 	} else {
1238 		dst_pid = nlk->dst_pid;
1239 		dst_group = nlk->dst_group;
1240 	}
1241 
1242 	if (!nlk->pid) {
1243 		err = netlink_autobind(sock);
1244 		if (err)
1245 			goto out;
1246 	}
1247 
1248 	err = -EMSGSIZE;
1249 	if (len > sk->sk_sndbuf - 32)
1250 		goto out;
1251 	err = -ENOBUFS;
1252 	skb = alloc_skb(len, GFP_KERNEL);
1253 	if (skb == NULL)
1254 		goto out;
1255 
1256 	NETLINK_CB(skb).pid	= nlk->pid;
1257 	NETLINK_CB(skb).dst_group = dst_group;
1258 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1259 	NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1260 	security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1261 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1262 
1263 	/* What can I do? Netlink is asynchronous, so that
1264 	   we will have to save current capabilities to
1265 	   check them, when this message will be delivered
1266 	   to corresponding kernel module.   --ANK (980802)
1267 	 */
1268 
1269 	err = -EFAULT;
1270 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1271 		kfree_skb(skb);
1272 		goto out;
1273 	}
1274 
1275 	err = security_netlink_send(sk, skb);
1276 	if (err) {
1277 		kfree_skb(skb);
1278 		goto out;
1279 	}
1280 
1281 	if (dst_group) {
1282 		atomic_inc(&skb->users);
1283 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1284 	}
1285 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1286 
1287 out:
1288 	return err;
1289 }
1290 
1291 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1292 			   struct msghdr *msg, size_t len,
1293 			   int flags)
1294 {
1295 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1296 	struct scm_cookie scm;
1297 	struct sock *sk = sock->sk;
1298 	struct netlink_sock *nlk = nlk_sk(sk);
1299 	int noblock = flags&MSG_DONTWAIT;
1300 	size_t copied;
1301 	struct sk_buff *skb;
1302 	int err;
1303 
1304 	if (flags&MSG_OOB)
1305 		return -EOPNOTSUPP;
1306 
1307 	copied = 0;
1308 
1309 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1310 	if (skb == NULL)
1311 		goto out;
1312 
1313 	msg->msg_namelen = 0;
1314 
1315 	copied = skb->len;
1316 	if (len < copied) {
1317 		msg->msg_flags |= MSG_TRUNC;
1318 		copied = len;
1319 	}
1320 
1321 	skb_reset_transport_header(skb);
1322 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1323 
1324 	if (msg->msg_name) {
1325 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1326 		addr->nl_family = AF_NETLINK;
1327 		addr->nl_pad    = 0;
1328 		addr->nl_pid	= NETLINK_CB(skb).pid;
1329 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1330 		msg->msg_namelen = sizeof(*addr);
1331 	}
1332 
1333 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1334 		netlink_cmsg_recv_pktinfo(msg, skb);
1335 
1336 	if (NULL == siocb->scm) {
1337 		memset(&scm, 0, sizeof(scm));
1338 		siocb->scm = &scm;
1339 	}
1340 	siocb->scm->creds = *NETLINK_CREDS(skb);
1341 	if (flags & MSG_TRUNC)
1342 		copied = skb->len;
1343 	skb_free_datagram(sk, skb);
1344 
1345 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1346 		netlink_dump(sk);
1347 
1348 	scm_recv(sock, msg, siocb->scm, flags);
1349 out:
1350 	netlink_rcv_wake(sk);
1351 	return err ? : copied;
1352 }
1353 
1354 static void netlink_data_ready(struct sock *sk, int len)
1355 {
1356 	BUG();
1357 }
1358 
1359 /*
1360  *	We export these functions to other modules. They provide a
1361  *	complete set of kernel non-blocking support for message
1362  *	queueing.
1363  */
1364 
1365 struct sock *
1366 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1367 		      void (*input)(struct sk_buff *skb),
1368 		      struct mutex *cb_mutex, struct module *module)
1369 {
1370 	struct socket *sock;
1371 	struct sock *sk;
1372 	struct netlink_sock *nlk;
1373 	unsigned long *listeners = NULL;
1374 
1375 	BUG_ON(!nl_table);
1376 
1377 	if (unit < 0 || unit >= MAX_LINKS)
1378 		return NULL;
1379 
1380 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1381 		return NULL;
1382 
1383 	/*
1384 	 * We have to just have a reference on the net from sk, but don't
1385 	 * get_net it. Besides, we cannot get and then put the net here.
1386 	 * So we create one inside init_net and the move it to net.
1387 	 */
1388 
1389 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1390 		goto out_sock_release_nosk;
1391 
1392 	sk = sock->sk;
1393 	sk_change_net(sk, net);
1394 
1395 	if (groups < 32)
1396 		groups = 32;
1397 
1398 	listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1399 	if (!listeners)
1400 		goto out_sock_release;
1401 
1402 	sk->sk_data_ready = netlink_data_ready;
1403 	if (input)
1404 		nlk_sk(sk)->netlink_rcv = input;
1405 
1406 	if (netlink_insert(sk, net, 0))
1407 		goto out_sock_release;
1408 
1409 	nlk = nlk_sk(sk);
1410 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1411 
1412 	netlink_table_grab();
1413 	if (!nl_table[unit].registered) {
1414 		nl_table[unit].groups = groups;
1415 		nl_table[unit].listeners = listeners;
1416 		nl_table[unit].cb_mutex = cb_mutex;
1417 		nl_table[unit].module = module;
1418 		nl_table[unit].registered = 1;
1419 	} else {
1420 		kfree(listeners);
1421 		nl_table[unit].registered++;
1422 	}
1423 	netlink_table_ungrab();
1424 	return sk;
1425 
1426 out_sock_release:
1427 	kfree(listeners);
1428 	netlink_kernel_release(sk);
1429 	return NULL;
1430 
1431 out_sock_release_nosk:
1432 	sock_release(sock);
1433 	return NULL;
1434 }
1435 EXPORT_SYMBOL(netlink_kernel_create);
1436 
1437 
1438 void
1439 netlink_kernel_release(struct sock *sk)
1440 {
1441 	sk_release_kernel(sk);
1442 }
1443 EXPORT_SYMBOL(netlink_kernel_release);
1444 
1445 
1446 /**
1447  * netlink_change_ngroups - change number of multicast groups
1448  *
1449  * This changes the number of multicast groups that are available
1450  * on a certain netlink family. Note that it is not possible to
1451  * change the number of groups to below 32. Also note that it does
1452  * not implicitly call netlink_clear_multicast_users() when the
1453  * number of groups is reduced.
1454  *
1455  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1456  * @groups: The new number of groups.
1457  */
1458 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1459 {
1460 	unsigned long *listeners, *old = NULL;
1461 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1462 	int err = 0;
1463 
1464 	if (groups < 32)
1465 		groups = 32;
1466 
1467 	netlink_table_grab();
1468 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1469 		listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1470 		if (!listeners) {
1471 			err = -ENOMEM;
1472 			goto out_ungrab;
1473 		}
1474 		old = tbl->listeners;
1475 		memcpy(listeners, old, NLGRPSZ(tbl->groups));
1476 		rcu_assign_pointer(tbl->listeners, listeners);
1477 	}
1478 	tbl->groups = groups;
1479 
1480  out_ungrab:
1481 	netlink_table_ungrab();
1482 	synchronize_rcu();
1483 	kfree(old);
1484 	return err;
1485 }
1486 EXPORT_SYMBOL(netlink_change_ngroups);
1487 
1488 /**
1489  * netlink_clear_multicast_users - kick off multicast listeners
1490  *
1491  * This function removes all listeners from the given group.
1492  * @ksk: The kernel netlink socket, as returned by
1493  *	netlink_kernel_create().
1494  * @group: The multicast group to clear.
1495  */
1496 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1497 {
1498 	struct sock *sk;
1499 	struct hlist_node *node;
1500 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1501 
1502 	netlink_table_grab();
1503 
1504 	sk_for_each_bound(sk, node, &tbl->mc_list)
1505 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1506 
1507 	netlink_table_ungrab();
1508 }
1509 EXPORT_SYMBOL(netlink_clear_multicast_users);
1510 
1511 void netlink_set_nonroot(int protocol, unsigned int flags)
1512 {
1513 	if ((unsigned int)protocol < MAX_LINKS)
1514 		nl_table[protocol].nl_nonroot = flags;
1515 }
1516 EXPORT_SYMBOL(netlink_set_nonroot);
1517 
1518 static void netlink_destroy_callback(struct netlink_callback *cb)
1519 {
1520 	if (cb->skb)
1521 		kfree_skb(cb->skb);
1522 	kfree(cb);
1523 }
1524 
1525 /*
1526  * It looks a bit ugly.
1527  * It would be better to create kernel thread.
1528  */
1529 
1530 static int netlink_dump(struct sock *sk)
1531 {
1532 	struct netlink_sock *nlk = nlk_sk(sk);
1533 	struct netlink_callback *cb;
1534 	struct sk_buff *skb;
1535 	struct nlmsghdr *nlh;
1536 	int len, err = -ENOBUFS;
1537 
1538 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1539 	if (!skb)
1540 		goto errout;
1541 
1542 	mutex_lock(nlk->cb_mutex);
1543 
1544 	cb = nlk->cb;
1545 	if (cb == NULL) {
1546 		err = -EINVAL;
1547 		goto errout_skb;
1548 	}
1549 
1550 	len = cb->dump(skb, cb);
1551 
1552 	if (len > 0) {
1553 		mutex_unlock(nlk->cb_mutex);
1554 
1555 		if (sk_filter(sk, skb))
1556 			kfree_skb(skb);
1557 		else {
1558 			skb_queue_tail(&sk->sk_receive_queue, skb);
1559 			sk->sk_data_ready(sk, skb->len);
1560 		}
1561 		return 0;
1562 	}
1563 
1564 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1565 	if (!nlh)
1566 		goto errout_skb;
1567 
1568 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1569 
1570 	if (sk_filter(sk, skb))
1571 		kfree_skb(skb);
1572 	else {
1573 		skb_queue_tail(&sk->sk_receive_queue, skb);
1574 		sk->sk_data_ready(sk, skb->len);
1575 	}
1576 
1577 	if (cb->done)
1578 		cb->done(cb);
1579 	nlk->cb = NULL;
1580 	mutex_unlock(nlk->cb_mutex);
1581 
1582 	netlink_destroy_callback(cb);
1583 	return 0;
1584 
1585 errout_skb:
1586 	mutex_unlock(nlk->cb_mutex);
1587 	kfree_skb(skb);
1588 errout:
1589 	return err;
1590 }
1591 
1592 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1593 		       struct nlmsghdr *nlh,
1594 		       int (*dump)(struct sk_buff *skb,
1595 				   struct netlink_callback *),
1596 		       int (*done)(struct netlink_callback *))
1597 {
1598 	struct netlink_callback *cb;
1599 	struct sock *sk;
1600 	struct netlink_sock *nlk;
1601 
1602 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1603 	if (cb == NULL)
1604 		return -ENOBUFS;
1605 
1606 	cb->dump = dump;
1607 	cb->done = done;
1608 	cb->nlh = nlh;
1609 	atomic_inc(&skb->users);
1610 	cb->skb = skb;
1611 
1612 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1613 	if (sk == NULL) {
1614 		netlink_destroy_callback(cb);
1615 		return -ECONNREFUSED;
1616 	}
1617 	nlk = nlk_sk(sk);
1618 	/* A dump is in progress... */
1619 	mutex_lock(nlk->cb_mutex);
1620 	if (nlk->cb) {
1621 		mutex_unlock(nlk->cb_mutex);
1622 		netlink_destroy_callback(cb);
1623 		sock_put(sk);
1624 		return -EBUSY;
1625 	}
1626 	nlk->cb = cb;
1627 	mutex_unlock(nlk->cb_mutex);
1628 
1629 	netlink_dump(sk);
1630 	sock_put(sk);
1631 
1632 	/* We successfully started a dump, by returning -EINTR we
1633 	 * signal not to send ACK even if it was requested.
1634 	 */
1635 	return -EINTR;
1636 }
1637 EXPORT_SYMBOL(netlink_dump_start);
1638 
1639 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1640 {
1641 	struct sk_buff *skb;
1642 	struct nlmsghdr *rep;
1643 	struct nlmsgerr *errmsg;
1644 	size_t payload = sizeof(*errmsg);
1645 
1646 	/* error messages get the original request appened */
1647 	if (err)
1648 		payload += nlmsg_len(nlh);
1649 
1650 	skb = nlmsg_new(payload, GFP_KERNEL);
1651 	if (!skb) {
1652 		struct sock *sk;
1653 
1654 		sk = netlink_lookup(sock_net(in_skb->sk),
1655 				    in_skb->sk->sk_protocol,
1656 				    NETLINK_CB(in_skb).pid);
1657 		if (sk) {
1658 			sk->sk_err = ENOBUFS;
1659 			sk->sk_error_report(sk);
1660 			sock_put(sk);
1661 		}
1662 		return;
1663 	}
1664 
1665 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1666 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1667 	errmsg = nlmsg_data(rep);
1668 	errmsg->error = err;
1669 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1670 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1671 }
1672 EXPORT_SYMBOL(netlink_ack);
1673 
1674 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1675 						     struct nlmsghdr *))
1676 {
1677 	struct nlmsghdr *nlh;
1678 	int err;
1679 
1680 	while (skb->len >= nlmsg_total_size(0)) {
1681 		int msglen;
1682 
1683 		nlh = nlmsg_hdr(skb);
1684 		err = 0;
1685 
1686 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1687 			return 0;
1688 
1689 		/* Only requests are handled by the kernel */
1690 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1691 			goto ack;
1692 
1693 		/* Skip control messages */
1694 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1695 			goto ack;
1696 
1697 		err = cb(skb, nlh);
1698 		if (err == -EINTR)
1699 			goto skip;
1700 
1701 ack:
1702 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1703 			netlink_ack(skb, nlh, err);
1704 
1705 skip:
1706 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1707 		if (msglen > skb->len)
1708 			msglen = skb->len;
1709 		skb_pull(skb, msglen);
1710 	}
1711 
1712 	return 0;
1713 }
1714 EXPORT_SYMBOL(netlink_rcv_skb);
1715 
1716 /**
1717  * nlmsg_notify - send a notification netlink message
1718  * @sk: netlink socket to use
1719  * @skb: notification message
1720  * @pid: destination netlink pid for reports or 0
1721  * @group: destination multicast group or 0
1722  * @report: 1 to report back, 0 to disable
1723  * @flags: allocation flags
1724  */
1725 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1726 		 unsigned int group, int report, gfp_t flags)
1727 {
1728 	int err = 0;
1729 
1730 	if (group) {
1731 		int exclude_pid = 0;
1732 
1733 		if (report) {
1734 			atomic_inc(&skb->users);
1735 			exclude_pid = pid;
1736 		}
1737 
1738 		/* errors reported via destination sk->sk_err */
1739 		nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1740 	}
1741 
1742 	if (report)
1743 		err = nlmsg_unicast(sk, skb, pid);
1744 
1745 	return err;
1746 }
1747 EXPORT_SYMBOL(nlmsg_notify);
1748 
1749 #ifdef CONFIG_PROC_FS
1750 struct nl_seq_iter {
1751 	struct seq_net_private p;
1752 	int link;
1753 	int hash_idx;
1754 };
1755 
1756 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1757 {
1758 	struct nl_seq_iter *iter = seq->private;
1759 	int i, j;
1760 	struct sock *s;
1761 	struct hlist_node *node;
1762 	loff_t off = 0;
1763 
1764 	for (i = 0; i < MAX_LINKS; i++) {
1765 		struct nl_pid_hash *hash = &nl_table[i].hash;
1766 
1767 		for (j = 0; j <= hash->mask; j++) {
1768 			sk_for_each(s, node, &hash->table[j]) {
1769 				if (sock_net(s) != seq_file_net(seq))
1770 					continue;
1771 				if (off == pos) {
1772 					iter->link = i;
1773 					iter->hash_idx = j;
1774 					return s;
1775 				}
1776 				++off;
1777 			}
1778 		}
1779 	}
1780 	return NULL;
1781 }
1782 
1783 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1784 	__acquires(nl_table_lock)
1785 {
1786 	read_lock(&nl_table_lock);
1787 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1788 }
1789 
1790 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1791 {
1792 	struct sock *s;
1793 	struct nl_seq_iter *iter;
1794 	int i, j;
1795 
1796 	++*pos;
1797 
1798 	if (v == SEQ_START_TOKEN)
1799 		return netlink_seq_socket_idx(seq, 0);
1800 
1801 	iter = seq->private;
1802 	s = v;
1803 	do {
1804 		s = sk_next(s);
1805 	} while (s && sock_net(s) != seq_file_net(seq));
1806 	if (s)
1807 		return s;
1808 
1809 	i = iter->link;
1810 	j = iter->hash_idx + 1;
1811 
1812 	do {
1813 		struct nl_pid_hash *hash = &nl_table[i].hash;
1814 
1815 		for (; j <= hash->mask; j++) {
1816 			s = sk_head(&hash->table[j]);
1817 			while (s && sock_net(s) != seq_file_net(seq))
1818 				s = sk_next(s);
1819 			if (s) {
1820 				iter->link = i;
1821 				iter->hash_idx = j;
1822 				return s;
1823 			}
1824 		}
1825 
1826 		j = 0;
1827 	} while (++i < MAX_LINKS);
1828 
1829 	return NULL;
1830 }
1831 
1832 static void netlink_seq_stop(struct seq_file *seq, void *v)
1833 	__releases(nl_table_lock)
1834 {
1835 	read_unlock(&nl_table_lock);
1836 }
1837 
1838 
1839 static int netlink_seq_show(struct seq_file *seq, void *v)
1840 {
1841 	if (v == SEQ_START_TOKEN)
1842 		seq_puts(seq,
1843 			 "sk       Eth Pid    Groups   "
1844 			 "Rmem     Wmem     Dump     Locks\n");
1845 	else {
1846 		struct sock *s = v;
1847 		struct netlink_sock *nlk = nlk_sk(s);
1848 
1849 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1850 			   s,
1851 			   s->sk_protocol,
1852 			   nlk->pid,
1853 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1854 			   atomic_read(&s->sk_rmem_alloc),
1855 			   atomic_read(&s->sk_wmem_alloc),
1856 			   nlk->cb,
1857 			   atomic_read(&s->sk_refcnt)
1858 			);
1859 
1860 	}
1861 	return 0;
1862 }
1863 
1864 static const struct seq_operations netlink_seq_ops = {
1865 	.start  = netlink_seq_start,
1866 	.next   = netlink_seq_next,
1867 	.stop   = netlink_seq_stop,
1868 	.show   = netlink_seq_show,
1869 };
1870 
1871 
1872 static int netlink_seq_open(struct inode *inode, struct file *file)
1873 {
1874 	return seq_open_net(inode, file, &netlink_seq_ops,
1875 				sizeof(struct nl_seq_iter));
1876 }
1877 
1878 static const struct file_operations netlink_seq_fops = {
1879 	.owner		= THIS_MODULE,
1880 	.open		= netlink_seq_open,
1881 	.read		= seq_read,
1882 	.llseek		= seq_lseek,
1883 	.release	= seq_release_net,
1884 };
1885 
1886 #endif
1887 
1888 int netlink_register_notifier(struct notifier_block *nb)
1889 {
1890 	return atomic_notifier_chain_register(&netlink_chain, nb);
1891 }
1892 EXPORT_SYMBOL(netlink_register_notifier);
1893 
1894 int netlink_unregister_notifier(struct notifier_block *nb)
1895 {
1896 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
1897 }
1898 EXPORT_SYMBOL(netlink_unregister_notifier);
1899 
1900 static const struct proto_ops netlink_ops = {
1901 	.family =	PF_NETLINK,
1902 	.owner =	THIS_MODULE,
1903 	.release =	netlink_release,
1904 	.bind =		netlink_bind,
1905 	.connect =	netlink_connect,
1906 	.socketpair =	sock_no_socketpair,
1907 	.accept =	sock_no_accept,
1908 	.getname =	netlink_getname,
1909 	.poll =		datagram_poll,
1910 	.ioctl =	sock_no_ioctl,
1911 	.listen =	sock_no_listen,
1912 	.shutdown =	sock_no_shutdown,
1913 	.setsockopt =	netlink_setsockopt,
1914 	.getsockopt =	netlink_getsockopt,
1915 	.sendmsg =	netlink_sendmsg,
1916 	.recvmsg =	netlink_recvmsg,
1917 	.mmap =		sock_no_mmap,
1918 	.sendpage =	sock_no_sendpage,
1919 };
1920 
1921 static struct net_proto_family netlink_family_ops = {
1922 	.family = PF_NETLINK,
1923 	.create = netlink_create,
1924 	.owner	= THIS_MODULE,	/* for consistency 8) */
1925 };
1926 
1927 static int __net_init netlink_net_init(struct net *net)
1928 {
1929 #ifdef CONFIG_PROC_FS
1930 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1931 		return -ENOMEM;
1932 #endif
1933 	return 0;
1934 }
1935 
1936 static void __net_exit netlink_net_exit(struct net *net)
1937 {
1938 #ifdef CONFIG_PROC_FS
1939 	proc_net_remove(net, "netlink");
1940 #endif
1941 }
1942 
1943 static struct pernet_operations __net_initdata netlink_net_ops = {
1944 	.init = netlink_net_init,
1945 	.exit = netlink_net_exit,
1946 };
1947 
1948 static int __init netlink_proto_init(void)
1949 {
1950 	struct sk_buff *dummy_skb;
1951 	int i;
1952 	unsigned long limit;
1953 	unsigned int order;
1954 	int err = proto_register(&netlink_proto, 0);
1955 
1956 	if (err != 0)
1957 		goto out;
1958 
1959 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1960 
1961 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1962 	if (!nl_table)
1963 		goto panic;
1964 
1965 	if (num_physpages >= (128 * 1024))
1966 		limit = num_physpages >> (21 - PAGE_SHIFT);
1967 	else
1968 		limit = num_physpages >> (23 - PAGE_SHIFT);
1969 
1970 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1971 	limit = (1UL << order) / sizeof(struct hlist_head);
1972 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1973 
1974 	for (i = 0; i < MAX_LINKS; i++) {
1975 		struct nl_pid_hash *hash = &nl_table[i].hash;
1976 
1977 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1978 		if (!hash->table) {
1979 			while (i-- > 0)
1980 				nl_pid_hash_free(nl_table[i].hash.table,
1981 						 1 * sizeof(*hash->table));
1982 			kfree(nl_table);
1983 			goto panic;
1984 		}
1985 		hash->max_shift = order;
1986 		hash->shift = 0;
1987 		hash->mask = 0;
1988 		hash->rehash_time = jiffies;
1989 	}
1990 
1991 	sock_register(&netlink_family_ops);
1992 	register_pernet_subsys(&netlink_net_ops);
1993 	/* The netlink device handler may be needed early. */
1994 	rtnetlink_init();
1995 out:
1996 	return err;
1997 panic:
1998 	panic("netlink_init: Cannot allocate nl_table\n");
1999 }
2000 
2001 core_initcall(netlink_proto_init);
2002