xref: /openbmc/linux/net/netlink/af_netlink.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
1 /*
2  * NETLINK      Kernel-user communication protocol.
3  *
4  * 		Authors:	Alan Cox <alan@redhat.com>
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/selinux.h>
58 #include <linux/mutex.h>
59 
60 #include <net/net_namespace.h>
61 #include <net/sock.h>
62 #include <net/scm.h>
63 #include <net/netlink.h>
64 
65 #define NLGRPSZ(x)	(ALIGN(x, sizeof(unsigned long) * 8) / 8)
66 #define NLGRPLONGS(x)	(NLGRPSZ(x)/sizeof(unsigned long))
67 
68 struct netlink_sock {
69 	/* struct sock has to be the first member of netlink_sock */
70 	struct sock		sk;
71 	u32			pid;
72 	u32			dst_pid;
73 	u32			dst_group;
74 	u32			flags;
75 	u32			subscriptions;
76 	u32			ngroups;
77 	unsigned long		*groups;
78 	unsigned long		state;
79 	wait_queue_head_t	wait;
80 	struct netlink_callback	*cb;
81 	struct mutex		*cb_mutex;
82 	struct mutex		cb_def_mutex;
83 	void			(*netlink_rcv)(struct sk_buff *skb);
84 	struct module		*module;
85 };
86 
87 #define NETLINK_KERNEL_SOCKET	0x1
88 #define NETLINK_RECV_PKTINFO	0x2
89 
90 static inline struct netlink_sock *nlk_sk(struct sock *sk)
91 {
92 	return container_of(sk, struct netlink_sock, sk);
93 }
94 
95 static inline int netlink_is_kernel(struct sock *sk)
96 {
97 	return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
98 }
99 
100 struct nl_pid_hash {
101 	struct hlist_head *table;
102 	unsigned long rehash_time;
103 
104 	unsigned int mask;
105 	unsigned int shift;
106 
107 	unsigned int entries;
108 	unsigned int max_shift;
109 
110 	u32 rnd;
111 };
112 
113 struct netlink_table {
114 	struct nl_pid_hash hash;
115 	struct hlist_head mc_list;
116 	unsigned long *listeners;
117 	unsigned int nl_nonroot;
118 	unsigned int groups;
119 	struct mutex *cb_mutex;
120 	struct module *module;
121 	int registered;
122 };
123 
124 static struct netlink_table *nl_table;
125 
126 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
127 
128 static int netlink_dump(struct sock *sk);
129 static void netlink_destroy_callback(struct netlink_callback *cb);
130 
131 static DEFINE_RWLOCK(nl_table_lock);
132 static atomic_t nl_table_users = ATOMIC_INIT(0);
133 
134 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
135 
136 static u32 netlink_group_mask(u32 group)
137 {
138 	return group ? 1 << (group - 1) : 0;
139 }
140 
141 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
142 {
143 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
144 }
145 
146 static void netlink_sock_destruct(struct sock *sk)
147 {
148 	struct netlink_sock *nlk = nlk_sk(sk);
149 
150 	if (nlk->cb) {
151 		if (nlk->cb->done)
152 			nlk->cb->done(nlk->cb);
153 		netlink_destroy_callback(nlk->cb);
154 	}
155 
156 	skb_queue_purge(&sk->sk_receive_queue);
157 
158 	if (!sock_flag(sk, SOCK_DEAD)) {
159 		printk("Freeing alive netlink socket %p\n", sk);
160 		return;
161 	}
162 	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
163 	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
164 	BUG_TRAP(!nlk_sk(sk)->groups);
165 }
166 
167 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
168  * 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 {
175 	write_lock_irq(&nl_table_lock);
176 
177 	if (atomic_read(&nl_table_users)) {
178 		DECLARE_WAITQUEUE(wait, current);
179 
180 		add_wait_queue_exclusive(&nl_table_wait, &wait);
181 		for(;;) {
182 			set_current_state(TASK_UNINTERRUPTIBLE);
183 			if (atomic_read(&nl_table_users) == 0)
184 				break;
185 			write_unlock_irq(&nl_table_lock);
186 			schedule();
187 			write_lock_irq(&nl_table_lock);
188 		}
189 
190 		__set_current_state(TASK_RUNNING);
191 		remove_wait_queue(&nl_table_wait, &wait);
192 	}
193 }
194 
195 static __inline__ void netlink_table_ungrab(void)
196 {
197 	write_unlock_irq(&nl_table_lock);
198 	wake_up(&nl_table_wait);
199 }
200 
201 static __inline__ void
202 netlink_lock_table(void)
203 {
204 	/* read_lock() synchronizes us to netlink_table_grab */
205 
206 	read_lock(&nl_table_lock);
207 	atomic_inc(&nl_table_users);
208 	read_unlock(&nl_table_lock);
209 }
210 
211 static __inline__ void
212 netlink_unlock_table(void)
213 {
214 	if (atomic_dec_and_test(&nl_table_users))
215 		wake_up(&nl_table_wait);
216 }
217 
218 static __inline__ struct sock *netlink_lookup(struct net *net, int protocol, u32 pid)
219 {
220 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
221 	struct hlist_head *head;
222 	struct sock *sk;
223 	struct hlist_node *node;
224 
225 	read_lock(&nl_table_lock);
226 	head = nl_pid_hashfn(hash, pid);
227 	sk_for_each(sk, node, head) {
228 		if ((sk->sk_net == net) && (nlk_sk(sk)->pid == pid)) {
229 			sock_hold(sk);
230 			goto found;
231 		}
232 	}
233 	sk = NULL;
234 found:
235 	read_unlock(&nl_table_lock);
236 	return sk;
237 }
238 
239 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
240 {
241 	if (size <= PAGE_SIZE)
242 		return kmalloc(size, GFP_ATOMIC);
243 	else
244 		return (struct hlist_head *)
245 			__get_free_pages(GFP_ATOMIC, get_order(size));
246 }
247 
248 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
249 {
250 	if (size <= PAGE_SIZE)
251 		kfree(table);
252 	else
253 		free_pages((unsigned long)table, get_order(size));
254 }
255 
256 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
257 {
258 	unsigned int omask, mask, shift;
259 	size_t osize, size;
260 	struct hlist_head *otable, *table;
261 	int i;
262 
263 	omask = mask = hash->mask;
264 	osize = size = (mask + 1) * sizeof(*table);
265 	shift = hash->shift;
266 
267 	if (grow) {
268 		if (++shift > hash->max_shift)
269 			return 0;
270 		mask = mask * 2 + 1;
271 		size *= 2;
272 	}
273 
274 	table = nl_pid_hash_alloc(size);
275 	if (!table)
276 		return 0;
277 
278 	memset(table, 0, size);
279 	otable = hash->table;
280 	hash->table = table;
281 	hash->mask = mask;
282 	hash->shift = shift;
283 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
284 
285 	for (i = 0; i <= omask; i++) {
286 		struct sock *sk;
287 		struct hlist_node *node, *tmp;
288 
289 		sk_for_each_safe(sk, node, tmp, &otable[i])
290 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
291 	}
292 
293 	nl_pid_hash_free(otable, osize);
294 	hash->rehash_time = jiffies + 10 * 60 * HZ;
295 	return 1;
296 }
297 
298 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
299 {
300 	int avg = hash->entries >> hash->shift;
301 
302 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
303 		return 1;
304 
305 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
306 		nl_pid_hash_rehash(hash, 0);
307 		return 1;
308 	}
309 
310 	return 0;
311 }
312 
313 static const struct proto_ops netlink_ops;
314 
315 static void
316 netlink_update_listeners(struct sock *sk)
317 {
318 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
319 	struct hlist_node *node;
320 	unsigned long mask;
321 	unsigned int i;
322 
323 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
324 		mask = 0;
325 		sk_for_each_bound(sk, node, &tbl->mc_list) {
326 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
327 				mask |= nlk_sk(sk)->groups[i];
328 		}
329 		tbl->listeners[i] = mask;
330 	}
331 	/* this function is only called with the netlink table "grabbed", which
332 	 * makes sure updates are visible before bind or setsockopt return. */
333 }
334 
335 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
336 {
337 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
338 	struct hlist_head *head;
339 	int err = -EADDRINUSE;
340 	struct sock *osk;
341 	struct hlist_node *node;
342 	int len;
343 
344 	netlink_table_grab();
345 	head = nl_pid_hashfn(hash, pid);
346 	len = 0;
347 	sk_for_each(osk, node, head) {
348 		if ((osk->sk_net == net) && (nlk_sk(osk)->pid == pid))
349 			break;
350 		len++;
351 	}
352 	if (node)
353 		goto err;
354 
355 	err = -EBUSY;
356 	if (nlk_sk(sk)->pid)
357 		goto err;
358 
359 	err = -ENOMEM;
360 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
361 		goto err;
362 
363 	if (len && nl_pid_hash_dilute(hash, len))
364 		head = nl_pid_hashfn(hash, pid);
365 	hash->entries++;
366 	nlk_sk(sk)->pid = pid;
367 	sk_add_node(sk, head);
368 	err = 0;
369 
370 err:
371 	netlink_table_ungrab();
372 	return err;
373 }
374 
375 static void netlink_remove(struct sock *sk)
376 {
377 	netlink_table_grab();
378 	if (sk_del_node_init(sk))
379 		nl_table[sk->sk_protocol].hash.entries--;
380 	if (nlk_sk(sk)->subscriptions)
381 		__sk_del_bind_node(sk);
382 	netlink_table_ungrab();
383 }
384 
385 static struct proto netlink_proto = {
386 	.name	  = "NETLINK",
387 	.owner	  = THIS_MODULE,
388 	.obj_size = sizeof(struct netlink_sock),
389 };
390 
391 static int __netlink_create(struct net *net, struct socket *sock,
392 			    struct mutex *cb_mutex, int protocol)
393 {
394 	struct sock *sk;
395 	struct netlink_sock *nlk;
396 
397 	sock->ops = &netlink_ops;
398 
399 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
400 	if (!sk)
401 		return -ENOMEM;
402 
403 	sock_init_data(sock, sk);
404 
405 	nlk = nlk_sk(sk);
406 	if (cb_mutex)
407 		nlk->cb_mutex = cb_mutex;
408 	else {
409 		nlk->cb_mutex = &nlk->cb_def_mutex;
410 		mutex_init(nlk->cb_mutex);
411 	}
412 	init_waitqueue_head(&nlk->wait);
413 
414 	sk->sk_destruct = netlink_sock_destruct;
415 	sk->sk_protocol = protocol;
416 	return 0;
417 }
418 
419 static int netlink_create(struct net *net, struct socket *sock, int protocol)
420 {
421 	struct module *module = NULL;
422 	struct mutex *cb_mutex;
423 	struct netlink_sock *nlk;
424 	int err = 0;
425 
426 	sock->state = SS_UNCONNECTED;
427 
428 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
429 		return -ESOCKTNOSUPPORT;
430 
431 	if (protocol<0 || protocol >= MAX_LINKS)
432 		return -EPROTONOSUPPORT;
433 
434 	netlink_lock_table();
435 #ifdef CONFIG_KMOD
436 	if (!nl_table[protocol].registered) {
437 		netlink_unlock_table();
438 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
439 		netlink_lock_table();
440 	}
441 #endif
442 	if (nl_table[protocol].registered &&
443 	    try_module_get(nl_table[protocol].module))
444 		module = nl_table[protocol].module;
445 	cb_mutex = nl_table[protocol].cb_mutex;
446 	netlink_unlock_table();
447 
448 	if ((err = __netlink_create(net, sock, cb_mutex, protocol)) < 0)
449 		goto out_module;
450 
451 	nlk = nlk_sk(sock->sk);
452 	nlk->module = module;
453 out:
454 	return err;
455 
456 out_module:
457 	module_put(module);
458 	goto out;
459 }
460 
461 static int netlink_release(struct socket *sock)
462 {
463 	struct sock *sk = sock->sk;
464 	struct netlink_sock *nlk;
465 
466 	if (!sk)
467 		return 0;
468 
469 	netlink_remove(sk);
470 	sock_orphan(sk);
471 	nlk = nlk_sk(sk);
472 
473 	/*
474 	 * OK. Socket is unlinked, any packets that arrive now
475 	 * will be purged.
476 	 */
477 
478 	sock->sk = NULL;
479 	wake_up_interruptible_all(&nlk->wait);
480 
481 	skb_queue_purge(&sk->sk_write_queue);
482 
483 	if (nlk->pid && !nlk->subscriptions) {
484 		struct netlink_notify n = {
485 						.net = sk->sk_net,
486 						.protocol = sk->sk_protocol,
487 						.pid = nlk->pid,
488 					  };
489 		atomic_notifier_call_chain(&netlink_chain,
490 				NETLINK_URELEASE, &n);
491 	}
492 
493 	module_put(nlk->module);
494 
495 	netlink_table_grab();
496 	if (netlink_is_kernel(sk)) {
497 		kfree(nl_table[sk->sk_protocol].listeners);
498 		nl_table[sk->sk_protocol].module = NULL;
499 		nl_table[sk->sk_protocol].registered = 0;
500 	} else if (nlk->subscriptions)
501 		netlink_update_listeners(sk);
502 	netlink_table_ungrab();
503 
504 	kfree(nlk->groups);
505 	nlk->groups = NULL;
506 
507 	sock_put(sk);
508 	return 0;
509 }
510 
511 static int netlink_autobind(struct socket *sock)
512 {
513 	struct sock *sk = sock->sk;
514 	struct net *net = sk->sk_net;
515 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
516 	struct hlist_head *head;
517 	struct sock *osk;
518 	struct hlist_node *node;
519 	s32 pid = current->tgid;
520 	int err;
521 	static s32 rover = -4097;
522 
523 retry:
524 	cond_resched();
525 	netlink_table_grab();
526 	head = nl_pid_hashfn(hash, pid);
527 	sk_for_each(osk, node, head) {
528 		if ((osk->sk_net != net))
529 			continue;
530 		if (nlk_sk(osk)->pid == pid) {
531 			/* Bind collision, search negative pid values. */
532 			pid = rover--;
533 			if (rover > -4097)
534 				rover = -4097;
535 			netlink_table_ungrab();
536 			goto retry;
537 		}
538 	}
539 	netlink_table_ungrab();
540 
541 	err = netlink_insert(sk, net, pid);
542 	if (err == -EADDRINUSE)
543 		goto retry;
544 
545 	/* If 2 threads race to autobind, that is fine.  */
546 	if (err == -EBUSY)
547 		err = 0;
548 
549 	return err;
550 }
551 
552 static inline int netlink_capable(struct socket *sock, unsigned int flag)
553 {
554 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
555 	       capable(CAP_NET_ADMIN);
556 }
557 
558 static void
559 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
560 {
561 	struct netlink_sock *nlk = nlk_sk(sk);
562 
563 	if (nlk->subscriptions && !subscriptions)
564 		__sk_del_bind_node(sk);
565 	else if (!nlk->subscriptions && subscriptions)
566 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
567 	nlk->subscriptions = subscriptions;
568 }
569 
570 static int netlink_realloc_groups(struct sock *sk)
571 {
572 	struct netlink_sock *nlk = nlk_sk(sk);
573 	unsigned int groups;
574 	unsigned long *new_groups;
575 	int err = 0;
576 
577 	netlink_table_grab();
578 
579 	groups = nl_table[sk->sk_protocol].groups;
580 	if (!nl_table[sk->sk_protocol].registered) {
581 		err = -ENOENT;
582 		goto out_unlock;
583 	}
584 
585 	if (nlk->ngroups >= groups)
586 		goto out_unlock;
587 
588 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
589 	if (new_groups == NULL) {
590 		err = -ENOMEM;
591 		goto out_unlock;
592 	}
593 	memset((char*)new_groups + NLGRPSZ(nlk->ngroups), 0,
594 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
595 
596 	nlk->groups = new_groups;
597 	nlk->ngroups = groups;
598  out_unlock:
599 	netlink_table_ungrab();
600 	return err;
601 }
602 
603 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
604 {
605 	struct sock *sk = sock->sk;
606 	struct net *net = sk->sk_net;
607 	struct netlink_sock *nlk = nlk_sk(sk);
608 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
609 	int err;
610 
611 	if (nladdr->nl_family != AF_NETLINK)
612 		return -EINVAL;
613 
614 	/* Only superuser is allowed to listen multicasts */
615 	if (nladdr->nl_groups) {
616 		if (!netlink_capable(sock, NL_NONROOT_RECV))
617 			return -EPERM;
618 		err = netlink_realloc_groups(sk);
619 		if (err)
620 			return err;
621 	}
622 
623 	if (nlk->pid) {
624 		if (nladdr->nl_pid != nlk->pid)
625 			return -EINVAL;
626 	} else {
627 		err = nladdr->nl_pid ?
628 			netlink_insert(sk, net, nladdr->nl_pid) :
629 			netlink_autobind(sock);
630 		if (err)
631 			return err;
632 	}
633 
634 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
635 		return 0;
636 
637 	netlink_table_grab();
638 	netlink_update_subscriptions(sk, nlk->subscriptions +
639 					 hweight32(nladdr->nl_groups) -
640 					 hweight32(nlk->groups[0]));
641 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
642 	netlink_update_listeners(sk);
643 	netlink_table_ungrab();
644 
645 	return 0;
646 }
647 
648 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
649 			   int alen, int flags)
650 {
651 	int err = 0;
652 	struct sock *sk = sock->sk;
653 	struct netlink_sock *nlk = nlk_sk(sk);
654 	struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
655 
656 	if (addr->sa_family == AF_UNSPEC) {
657 		sk->sk_state	= NETLINK_UNCONNECTED;
658 		nlk->dst_pid	= 0;
659 		nlk->dst_group  = 0;
660 		return 0;
661 	}
662 	if (addr->sa_family != AF_NETLINK)
663 		return -EINVAL;
664 
665 	/* Only superuser is allowed to send multicasts */
666 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
667 		return -EPERM;
668 
669 	if (!nlk->pid)
670 		err = netlink_autobind(sock);
671 
672 	if (err == 0) {
673 		sk->sk_state	= NETLINK_CONNECTED;
674 		nlk->dst_pid 	= nladdr->nl_pid;
675 		nlk->dst_group  = ffs(nladdr->nl_groups);
676 	}
677 
678 	return err;
679 }
680 
681 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
682 {
683 	struct sock *sk = sock->sk;
684 	struct netlink_sock *nlk = nlk_sk(sk);
685 	struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
686 
687 	nladdr->nl_family = AF_NETLINK;
688 	nladdr->nl_pad = 0;
689 	*addr_len = sizeof(*nladdr);
690 
691 	if (peer) {
692 		nladdr->nl_pid = nlk->dst_pid;
693 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
694 	} else {
695 		nladdr->nl_pid = nlk->pid;
696 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
697 	}
698 	return 0;
699 }
700 
701 static void netlink_overrun(struct sock *sk)
702 {
703 	if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
704 		sk->sk_err = ENOBUFS;
705 		sk->sk_error_report(sk);
706 	}
707 }
708 
709 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
710 {
711 	struct sock *sock;
712 	struct netlink_sock *nlk;
713 
714 	sock = netlink_lookup(ssk->sk_net, ssk->sk_protocol, pid);
715 	if (!sock)
716 		return ERR_PTR(-ECONNREFUSED);
717 
718 	/* Don't bother queuing skb if kernel socket has no input function */
719 	nlk = nlk_sk(sock);
720 	if (sock->sk_state == NETLINK_CONNECTED &&
721 	    nlk->dst_pid != nlk_sk(ssk)->pid) {
722 		sock_put(sock);
723 		return ERR_PTR(-ECONNREFUSED);
724 	}
725 	return sock;
726 }
727 
728 struct sock *netlink_getsockbyfilp(struct file *filp)
729 {
730 	struct inode *inode = filp->f_path.dentry->d_inode;
731 	struct sock *sock;
732 
733 	if (!S_ISSOCK(inode->i_mode))
734 		return ERR_PTR(-ENOTSOCK);
735 
736 	sock = SOCKET_I(inode)->sk;
737 	if (sock->sk_family != AF_NETLINK)
738 		return ERR_PTR(-EINVAL);
739 
740 	sock_hold(sock);
741 	return sock;
742 }
743 
744 /*
745  * Attach a skb to a netlink socket.
746  * The caller must hold a reference to the destination socket. On error, the
747  * reference is dropped. The skb is not send to the destination, just all
748  * all error checks are performed and memory in the queue is reserved.
749  * Return values:
750  * < 0: error. skb freed, reference to sock dropped.
751  * 0: continue
752  * 1: repeat lookup - reference dropped while waiting for socket memory.
753  */
754 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
755 		long timeo, struct sock *ssk)
756 {
757 	struct netlink_sock *nlk;
758 
759 	nlk = nlk_sk(sk);
760 
761 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
762 	    test_bit(0, &nlk->state)) {
763 		DECLARE_WAITQUEUE(wait, current);
764 		if (!timeo) {
765 			if (!ssk || netlink_is_kernel(ssk))
766 				netlink_overrun(sk);
767 			sock_put(sk);
768 			kfree_skb(skb);
769 			return -EAGAIN;
770 		}
771 
772 		__set_current_state(TASK_INTERRUPTIBLE);
773 		add_wait_queue(&nlk->wait, &wait);
774 
775 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
776 		     test_bit(0, &nlk->state)) &&
777 		    !sock_flag(sk, SOCK_DEAD))
778 			timeo = schedule_timeout(timeo);
779 
780 		__set_current_state(TASK_RUNNING);
781 		remove_wait_queue(&nlk->wait, &wait);
782 		sock_put(sk);
783 
784 		if (signal_pending(current)) {
785 			kfree_skb(skb);
786 			return sock_intr_errno(timeo);
787 		}
788 		return 1;
789 	}
790 	skb_set_owner_r(skb, sk);
791 	return 0;
792 }
793 
794 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
795 {
796 	int len = skb->len;
797 
798 	skb_queue_tail(&sk->sk_receive_queue, skb);
799 	sk->sk_data_ready(sk, len);
800 	sock_put(sk);
801 	return len;
802 }
803 
804 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
805 {
806 	kfree_skb(skb);
807 	sock_put(sk);
808 }
809 
810 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
811 					   gfp_t allocation)
812 {
813 	int delta;
814 
815 	skb_orphan(skb);
816 
817 	delta = skb->end - skb->tail;
818 	if (delta * 2 < skb->truesize)
819 		return skb;
820 
821 	if (skb_shared(skb)) {
822 		struct sk_buff *nskb = skb_clone(skb, allocation);
823 		if (!nskb)
824 			return skb;
825 		kfree_skb(skb);
826 		skb = nskb;
827 	}
828 
829 	if (!pskb_expand_head(skb, 0, -delta, allocation))
830 		skb->truesize -= delta;
831 
832 	return skb;
833 }
834 
835 static inline void netlink_rcv_wake(struct sock *sk)
836 {
837 	struct netlink_sock *nlk = nlk_sk(sk);
838 
839 	if (skb_queue_empty(&sk->sk_receive_queue))
840 		clear_bit(0, &nlk->state);
841 	if (!test_bit(0, &nlk->state))
842 		wake_up_interruptible(&nlk->wait);
843 }
844 
845 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
846 {
847 	int ret;
848 	struct netlink_sock *nlk = nlk_sk(sk);
849 
850 	ret = -ECONNREFUSED;
851 	if (nlk->netlink_rcv != NULL) {
852 		ret = skb->len;
853 		skb_set_owner_r(skb, sk);
854 		nlk->netlink_rcv(skb);
855 	}
856 	kfree_skb(skb);
857 	sock_put(sk);
858 	return ret;
859 }
860 
861 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
862 		    u32 pid, int nonblock)
863 {
864 	struct sock *sk;
865 	int err;
866 	long timeo;
867 
868 	skb = netlink_trim(skb, gfp_any());
869 
870 	timeo = sock_sndtimeo(ssk, nonblock);
871 retry:
872 	sk = netlink_getsockbypid(ssk, pid);
873 	if (IS_ERR(sk)) {
874 		kfree_skb(skb);
875 		return PTR_ERR(sk);
876 	}
877 	if (netlink_is_kernel(sk))
878 		return netlink_unicast_kernel(sk, skb);
879 
880 	err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
881 	if (err == 1)
882 		goto retry;
883 	if (err)
884 		return err;
885 
886 	return netlink_sendskb(sk, skb);
887 }
888 
889 int netlink_has_listeners(struct sock *sk, unsigned int group)
890 {
891 	int res = 0;
892 	unsigned long *listeners;
893 
894 	BUG_ON(!netlink_is_kernel(sk));
895 
896 	rcu_read_lock();
897 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
898 
899 	if (group - 1 < nl_table[sk->sk_protocol].groups)
900 		res = test_bit(group - 1, listeners);
901 
902 	rcu_read_unlock();
903 
904 	return res;
905 }
906 EXPORT_SYMBOL_GPL(netlink_has_listeners);
907 
908 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
909 {
910 	struct netlink_sock *nlk = nlk_sk(sk);
911 
912 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
913 	    !test_bit(0, &nlk->state)) {
914 		skb_set_owner_r(skb, sk);
915 		skb_queue_tail(&sk->sk_receive_queue, skb);
916 		sk->sk_data_ready(sk, skb->len);
917 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
918 	}
919 	return -1;
920 }
921 
922 struct netlink_broadcast_data {
923 	struct sock *exclude_sk;
924 	struct net *net;
925 	u32 pid;
926 	u32 group;
927 	int failure;
928 	int congested;
929 	int delivered;
930 	gfp_t allocation;
931 	struct sk_buff *skb, *skb2;
932 };
933 
934 static inline int do_one_broadcast(struct sock *sk,
935 				   struct netlink_broadcast_data *p)
936 {
937 	struct netlink_sock *nlk = nlk_sk(sk);
938 	int val;
939 
940 	if (p->exclude_sk == sk)
941 		goto out;
942 
943 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
944 	    !test_bit(p->group - 1, nlk->groups))
945 		goto out;
946 
947 	if ((sk->sk_net != p->net))
948 		goto out;
949 
950 	if (p->failure) {
951 		netlink_overrun(sk);
952 		goto out;
953 	}
954 
955 	sock_hold(sk);
956 	if (p->skb2 == NULL) {
957 		if (skb_shared(p->skb)) {
958 			p->skb2 = skb_clone(p->skb, p->allocation);
959 		} else {
960 			p->skb2 = skb_get(p->skb);
961 			/*
962 			 * skb ownership may have been set when
963 			 * delivered to a previous socket.
964 			 */
965 			skb_orphan(p->skb2);
966 		}
967 	}
968 	if (p->skb2 == NULL) {
969 		netlink_overrun(sk);
970 		/* Clone failed. Notify ALL listeners. */
971 		p->failure = 1;
972 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
973 		netlink_overrun(sk);
974 	} else {
975 		p->congested |= val;
976 		p->delivered = 1;
977 		p->skb2 = NULL;
978 	}
979 	sock_put(sk);
980 
981 out:
982 	return 0;
983 }
984 
985 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
986 		      u32 group, gfp_t allocation)
987 {
988 	struct net *net = ssk->sk_net;
989 	struct netlink_broadcast_data info;
990 	struct hlist_node *node;
991 	struct sock *sk;
992 
993 	skb = netlink_trim(skb, allocation);
994 
995 	info.exclude_sk = ssk;
996 	info.net = net;
997 	info.pid = pid;
998 	info.group = group;
999 	info.failure = 0;
1000 	info.congested = 0;
1001 	info.delivered = 0;
1002 	info.allocation = allocation;
1003 	info.skb = skb;
1004 	info.skb2 = NULL;
1005 
1006 	/* While we sleep in clone, do not allow to change socket list */
1007 
1008 	netlink_lock_table();
1009 
1010 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1011 		do_one_broadcast(sk, &info);
1012 
1013 	kfree_skb(skb);
1014 
1015 	netlink_unlock_table();
1016 
1017 	if (info.skb2)
1018 		kfree_skb(info.skb2);
1019 
1020 	if (info.delivered) {
1021 		if (info.congested && (allocation & __GFP_WAIT))
1022 			yield();
1023 		return 0;
1024 	}
1025 	if (info.failure)
1026 		return -ENOBUFS;
1027 	return -ESRCH;
1028 }
1029 
1030 struct netlink_set_err_data {
1031 	struct sock *exclude_sk;
1032 	u32 pid;
1033 	u32 group;
1034 	int code;
1035 };
1036 
1037 static inline int do_one_set_err(struct sock *sk,
1038 				 struct netlink_set_err_data *p)
1039 {
1040 	struct netlink_sock *nlk = nlk_sk(sk);
1041 
1042 	if (sk == p->exclude_sk)
1043 		goto out;
1044 
1045 	if (sk->sk_net != p->exclude_sk->sk_net)
1046 		goto out;
1047 
1048 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1049 	    !test_bit(p->group - 1, nlk->groups))
1050 		goto out;
1051 
1052 	sk->sk_err = p->code;
1053 	sk->sk_error_report(sk);
1054 out:
1055 	return 0;
1056 }
1057 
1058 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1059 {
1060 	struct netlink_set_err_data info;
1061 	struct hlist_node *node;
1062 	struct sock *sk;
1063 
1064 	info.exclude_sk = ssk;
1065 	info.pid = pid;
1066 	info.group = group;
1067 	info.code = code;
1068 
1069 	read_lock(&nl_table_lock);
1070 
1071 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1072 		do_one_set_err(sk, &info);
1073 
1074 	read_unlock(&nl_table_lock);
1075 }
1076 
1077 /* must be called with netlink table grabbed */
1078 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1079 				     unsigned int group,
1080 				     int is_new)
1081 {
1082 	int old, new = !!is_new, subscriptions;
1083 
1084 	old = test_bit(group - 1, nlk->groups);
1085 	subscriptions = nlk->subscriptions - old + new;
1086 	if (new)
1087 		__set_bit(group - 1, nlk->groups);
1088 	else
1089 		__clear_bit(group - 1, nlk->groups);
1090 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1091 	netlink_update_listeners(&nlk->sk);
1092 }
1093 
1094 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1095 			      char __user *optval, int optlen)
1096 {
1097 	struct sock *sk = sock->sk;
1098 	struct netlink_sock *nlk = nlk_sk(sk);
1099 	unsigned int val = 0;
1100 	int err;
1101 
1102 	if (level != SOL_NETLINK)
1103 		return -ENOPROTOOPT;
1104 
1105 	if (optlen >= sizeof(int) &&
1106 	    get_user(val, (unsigned int __user *)optval))
1107 		return -EFAULT;
1108 
1109 	switch (optname) {
1110 	case NETLINK_PKTINFO:
1111 		if (val)
1112 			nlk->flags |= NETLINK_RECV_PKTINFO;
1113 		else
1114 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1115 		err = 0;
1116 		break;
1117 	case NETLINK_ADD_MEMBERSHIP:
1118 	case NETLINK_DROP_MEMBERSHIP: {
1119 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1120 			return -EPERM;
1121 		err = netlink_realloc_groups(sk);
1122 		if (err)
1123 			return err;
1124 		if (!val || val - 1 >= nlk->ngroups)
1125 			return -EINVAL;
1126 		netlink_table_grab();
1127 		netlink_update_socket_mc(nlk, val,
1128 					 optname == NETLINK_ADD_MEMBERSHIP);
1129 		netlink_table_ungrab();
1130 		err = 0;
1131 		break;
1132 	}
1133 	default:
1134 		err = -ENOPROTOOPT;
1135 	}
1136 	return err;
1137 }
1138 
1139 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1140 			      char __user *optval, int __user *optlen)
1141 {
1142 	struct sock *sk = sock->sk;
1143 	struct netlink_sock *nlk = nlk_sk(sk);
1144 	int len, val, err;
1145 
1146 	if (level != SOL_NETLINK)
1147 		return -ENOPROTOOPT;
1148 
1149 	if (get_user(len, optlen))
1150 		return -EFAULT;
1151 	if (len < 0)
1152 		return -EINVAL;
1153 
1154 	switch (optname) {
1155 	case NETLINK_PKTINFO:
1156 		if (len < sizeof(int))
1157 			return -EINVAL;
1158 		len = sizeof(int);
1159 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1160 		if (put_user(len, optlen) ||
1161 		    put_user(val, optval))
1162 			return -EFAULT;
1163 		err = 0;
1164 		break;
1165 	default:
1166 		err = -ENOPROTOOPT;
1167 	}
1168 	return err;
1169 }
1170 
1171 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1172 {
1173 	struct nl_pktinfo info;
1174 
1175 	info.group = NETLINK_CB(skb).dst_group;
1176 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1177 }
1178 
1179 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1180 			   struct msghdr *msg, size_t len)
1181 {
1182 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1183 	struct sock *sk = sock->sk;
1184 	struct netlink_sock *nlk = nlk_sk(sk);
1185 	struct sockaddr_nl *addr=msg->msg_name;
1186 	u32 dst_pid;
1187 	u32 dst_group;
1188 	struct sk_buff *skb;
1189 	int err;
1190 	struct scm_cookie scm;
1191 
1192 	if (msg->msg_flags&MSG_OOB)
1193 		return -EOPNOTSUPP;
1194 
1195 	if (NULL == siocb->scm)
1196 		siocb->scm = &scm;
1197 	err = scm_send(sock, msg, siocb->scm);
1198 	if (err < 0)
1199 		return err;
1200 
1201 	if (msg->msg_namelen) {
1202 		if (addr->nl_family != AF_NETLINK)
1203 			return -EINVAL;
1204 		dst_pid = addr->nl_pid;
1205 		dst_group = ffs(addr->nl_groups);
1206 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1207 			return -EPERM;
1208 	} else {
1209 		dst_pid = nlk->dst_pid;
1210 		dst_group = nlk->dst_group;
1211 	}
1212 
1213 	if (!nlk->pid) {
1214 		err = netlink_autobind(sock);
1215 		if (err)
1216 			goto out;
1217 	}
1218 
1219 	err = -EMSGSIZE;
1220 	if (len > sk->sk_sndbuf - 32)
1221 		goto out;
1222 	err = -ENOBUFS;
1223 	skb = alloc_skb(len, GFP_KERNEL);
1224 	if (skb==NULL)
1225 		goto out;
1226 
1227 	NETLINK_CB(skb).pid	= nlk->pid;
1228 	NETLINK_CB(skb).dst_group = dst_group;
1229 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1230 	selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1231 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1232 
1233 	/* What can I do? Netlink is asynchronous, so that
1234 	   we will have to save current capabilities to
1235 	   check them, when this message will be delivered
1236 	   to corresponding kernel module.   --ANK (980802)
1237 	 */
1238 
1239 	err = -EFAULT;
1240 	if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1241 		kfree_skb(skb);
1242 		goto out;
1243 	}
1244 
1245 	err = security_netlink_send(sk, skb);
1246 	if (err) {
1247 		kfree_skb(skb);
1248 		goto out;
1249 	}
1250 
1251 	if (dst_group) {
1252 		atomic_inc(&skb->users);
1253 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1254 	}
1255 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1256 
1257 out:
1258 	return err;
1259 }
1260 
1261 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1262 			   struct msghdr *msg, size_t len,
1263 			   int flags)
1264 {
1265 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1266 	struct scm_cookie scm;
1267 	struct sock *sk = sock->sk;
1268 	struct netlink_sock *nlk = nlk_sk(sk);
1269 	int noblock = flags&MSG_DONTWAIT;
1270 	size_t copied;
1271 	struct sk_buff *skb;
1272 	int err;
1273 
1274 	if (flags&MSG_OOB)
1275 		return -EOPNOTSUPP;
1276 
1277 	copied = 0;
1278 
1279 	skb = skb_recv_datagram(sk,flags,noblock,&err);
1280 	if (skb==NULL)
1281 		goto out;
1282 
1283 	msg->msg_namelen = 0;
1284 
1285 	copied = skb->len;
1286 	if (len < copied) {
1287 		msg->msg_flags |= MSG_TRUNC;
1288 		copied = len;
1289 	}
1290 
1291 	skb_reset_transport_header(skb);
1292 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1293 
1294 	if (msg->msg_name) {
1295 		struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1296 		addr->nl_family = AF_NETLINK;
1297 		addr->nl_pad    = 0;
1298 		addr->nl_pid	= NETLINK_CB(skb).pid;
1299 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1300 		msg->msg_namelen = sizeof(*addr);
1301 	}
1302 
1303 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1304 		netlink_cmsg_recv_pktinfo(msg, skb);
1305 
1306 	if (NULL == siocb->scm) {
1307 		memset(&scm, 0, sizeof(scm));
1308 		siocb->scm = &scm;
1309 	}
1310 	siocb->scm->creds = *NETLINK_CREDS(skb);
1311 	if (flags & MSG_TRUNC)
1312 		copied = skb->len;
1313 	skb_free_datagram(sk, skb);
1314 
1315 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1316 		netlink_dump(sk);
1317 
1318 	scm_recv(sock, msg, siocb->scm, flags);
1319 out:
1320 	netlink_rcv_wake(sk);
1321 	return err ? : copied;
1322 }
1323 
1324 static void netlink_data_ready(struct sock *sk, int len)
1325 {
1326 	BUG();
1327 }
1328 
1329 /*
1330  *	We export these functions to other modules. They provide a
1331  *	complete set of kernel non-blocking support for message
1332  *	queueing.
1333  */
1334 
1335 struct sock *
1336 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1337 		      void (*input)(struct sk_buff *skb),
1338 		      struct mutex *cb_mutex, struct module *module)
1339 {
1340 	struct socket *sock;
1341 	struct sock *sk;
1342 	struct netlink_sock *nlk;
1343 	unsigned long *listeners = NULL;
1344 
1345 	BUG_ON(!nl_table);
1346 
1347 	if (unit<0 || unit>=MAX_LINKS)
1348 		return NULL;
1349 
1350 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1351 		return NULL;
1352 
1353 	if (__netlink_create(net, sock, cb_mutex, unit) < 0)
1354 		goto out_sock_release;
1355 
1356 	if (groups < 32)
1357 		groups = 32;
1358 
1359 	listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1360 	if (!listeners)
1361 		goto out_sock_release;
1362 
1363 	sk = sock->sk;
1364 	sk->sk_data_ready = netlink_data_ready;
1365 	if (input)
1366 		nlk_sk(sk)->netlink_rcv = input;
1367 
1368 	if (netlink_insert(sk, net, 0))
1369 		goto out_sock_release;
1370 
1371 	nlk = nlk_sk(sk);
1372 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1373 
1374 	netlink_table_grab();
1375 	if (!nl_table[unit].registered) {
1376 		nl_table[unit].groups = groups;
1377 		nl_table[unit].listeners = listeners;
1378 		nl_table[unit].cb_mutex = cb_mutex;
1379 		nl_table[unit].module = module;
1380 		nl_table[unit].registered = 1;
1381 	} else {
1382 		kfree(listeners);
1383 	}
1384 	netlink_table_ungrab();
1385 
1386 	return sk;
1387 
1388 out_sock_release:
1389 	kfree(listeners);
1390 	sock_release(sock);
1391 	return NULL;
1392 }
1393 
1394 /**
1395  * netlink_change_ngroups - change number of multicast groups
1396  *
1397  * This changes the number of multicast groups that are available
1398  * on a certain netlink family. Note that it is not possible to
1399  * change the number of groups to below 32. Also note that it does
1400  * not implicitly call netlink_clear_multicast_users() when the
1401  * number of groups is reduced.
1402  *
1403  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1404  * @groups: The new number of groups.
1405  */
1406 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1407 {
1408 	unsigned long *listeners, *old = NULL;
1409 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1410 	int err = 0;
1411 
1412 	if (groups < 32)
1413 		groups = 32;
1414 
1415 	netlink_table_grab();
1416 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1417 		listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1418 		if (!listeners) {
1419 			err = -ENOMEM;
1420 			goto out_ungrab;
1421 		}
1422 		old = tbl->listeners;
1423 		memcpy(listeners, old, NLGRPSZ(tbl->groups));
1424 		rcu_assign_pointer(tbl->listeners, listeners);
1425 	}
1426 	tbl->groups = groups;
1427 
1428  out_ungrab:
1429 	netlink_table_ungrab();
1430 	synchronize_rcu();
1431 	kfree(old);
1432 	return err;
1433 }
1434 EXPORT_SYMBOL(netlink_change_ngroups);
1435 
1436 /**
1437  * netlink_clear_multicast_users - kick off multicast listeners
1438  *
1439  * This function removes all listeners from the given group.
1440  * @ksk: The kernel netlink socket, as returned by
1441  *	netlink_kernel_create().
1442  * @group: The multicast group to clear.
1443  */
1444 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1445 {
1446 	struct sock *sk;
1447 	struct hlist_node *node;
1448 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1449 
1450 	netlink_table_grab();
1451 
1452 	sk_for_each_bound(sk, node, &tbl->mc_list)
1453 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1454 
1455 	netlink_table_ungrab();
1456 }
1457 EXPORT_SYMBOL(netlink_clear_multicast_users);
1458 
1459 void netlink_set_nonroot(int protocol, unsigned int flags)
1460 {
1461 	if ((unsigned int)protocol < MAX_LINKS)
1462 		nl_table[protocol].nl_nonroot = flags;
1463 }
1464 
1465 static void netlink_destroy_callback(struct netlink_callback *cb)
1466 {
1467 	if (cb->skb)
1468 		kfree_skb(cb->skb);
1469 	kfree(cb);
1470 }
1471 
1472 /*
1473  * It looks a bit ugly.
1474  * It would be better to create kernel thread.
1475  */
1476 
1477 static int netlink_dump(struct sock *sk)
1478 {
1479 	struct netlink_sock *nlk = nlk_sk(sk);
1480 	struct netlink_callback *cb;
1481 	struct sk_buff *skb;
1482 	struct nlmsghdr *nlh;
1483 	int len, err = -ENOBUFS;
1484 
1485 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1486 	if (!skb)
1487 		goto errout;
1488 
1489 	mutex_lock(nlk->cb_mutex);
1490 
1491 	cb = nlk->cb;
1492 	if (cb == NULL) {
1493 		err = -EINVAL;
1494 		goto errout_skb;
1495 	}
1496 
1497 	len = cb->dump(skb, cb);
1498 
1499 	if (len > 0) {
1500 		mutex_unlock(nlk->cb_mutex);
1501 		skb_queue_tail(&sk->sk_receive_queue, skb);
1502 		sk->sk_data_ready(sk, len);
1503 		return 0;
1504 	}
1505 
1506 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1507 	if (!nlh)
1508 		goto errout_skb;
1509 
1510 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1511 
1512 	skb_queue_tail(&sk->sk_receive_queue, skb);
1513 	sk->sk_data_ready(sk, skb->len);
1514 
1515 	if (cb->done)
1516 		cb->done(cb);
1517 	nlk->cb = NULL;
1518 	mutex_unlock(nlk->cb_mutex);
1519 
1520 	netlink_destroy_callback(cb);
1521 	return 0;
1522 
1523 errout_skb:
1524 	mutex_unlock(nlk->cb_mutex);
1525 	kfree_skb(skb);
1526 errout:
1527 	return err;
1528 }
1529 
1530 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1531 		       struct nlmsghdr *nlh,
1532 		       int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1533 		       int (*done)(struct netlink_callback*))
1534 {
1535 	struct netlink_callback *cb;
1536 	struct sock *sk;
1537 	struct netlink_sock *nlk;
1538 
1539 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1540 	if (cb == NULL)
1541 		return -ENOBUFS;
1542 
1543 	cb->dump = dump;
1544 	cb->done = done;
1545 	cb->nlh = nlh;
1546 	atomic_inc(&skb->users);
1547 	cb->skb = skb;
1548 
1549 	sk = netlink_lookup(ssk->sk_net, ssk->sk_protocol, NETLINK_CB(skb).pid);
1550 	if (sk == NULL) {
1551 		netlink_destroy_callback(cb);
1552 		return -ECONNREFUSED;
1553 	}
1554 	nlk = nlk_sk(sk);
1555 	/* A dump is in progress... */
1556 	mutex_lock(nlk->cb_mutex);
1557 	if (nlk->cb) {
1558 		mutex_unlock(nlk->cb_mutex);
1559 		netlink_destroy_callback(cb);
1560 		sock_put(sk);
1561 		return -EBUSY;
1562 	}
1563 	nlk->cb = cb;
1564 	mutex_unlock(nlk->cb_mutex);
1565 
1566 	netlink_dump(sk);
1567 	sock_put(sk);
1568 	return 0;
1569 }
1570 
1571 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1572 {
1573 	struct sk_buff *skb;
1574 	struct nlmsghdr *rep;
1575 	struct nlmsgerr *errmsg;
1576 	size_t payload = sizeof(*errmsg);
1577 
1578 	/* error messages get the original request appened */
1579 	if (err)
1580 		payload += nlmsg_len(nlh);
1581 
1582 	skb = nlmsg_new(payload, GFP_KERNEL);
1583 	if (!skb) {
1584 		struct sock *sk;
1585 
1586 		sk = netlink_lookup(in_skb->sk->sk_net,
1587 				    in_skb->sk->sk_protocol,
1588 				    NETLINK_CB(in_skb).pid);
1589 		if (sk) {
1590 			sk->sk_err = ENOBUFS;
1591 			sk->sk_error_report(sk);
1592 			sock_put(sk);
1593 		}
1594 		return;
1595 	}
1596 
1597 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1598 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1599 	errmsg = nlmsg_data(rep);
1600 	errmsg->error = err;
1601 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1602 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1603 }
1604 
1605 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1606 						     struct nlmsghdr *))
1607 {
1608 	struct nlmsghdr *nlh;
1609 	int err;
1610 
1611 	while (skb->len >= nlmsg_total_size(0)) {
1612 		int msglen;
1613 
1614 		nlh = nlmsg_hdr(skb);
1615 		err = 0;
1616 
1617 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1618 			return 0;
1619 
1620 		/* Only requests are handled by the kernel */
1621 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1622 			goto skip;
1623 
1624 		/* Skip control messages */
1625 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1626 			goto skip;
1627 
1628 		err = cb(skb, nlh);
1629 skip:
1630 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1631 			netlink_ack(skb, nlh, err);
1632 
1633 	        msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1634 		if (msglen > skb->len)
1635 			msglen = skb->len;
1636 		skb_pull(skb, msglen);
1637 	}
1638 
1639 	return 0;
1640 }
1641 
1642 /**
1643  * nlmsg_notify - send a notification netlink message
1644  * @sk: netlink socket to use
1645  * @skb: notification message
1646  * @pid: destination netlink pid for reports or 0
1647  * @group: destination multicast group or 0
1648  * @report: 1 to report back, 0 to disable
1649  * @flags: allocation flags
1650  */
1651 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1652 		 unsigned int group, int report, gfp_t flags)
1653 {
1654 	int err = 0;
1655 
1656 	if (group) {
1657 		int exclude_pid = 0;
1658 
1659 		if (report) {
1660 			atomic_inc(&skb->users);
1661 			exclude_pid = pid;
1662 		}
1663 
1664 		/* errors reported via destination sk->sk_err */
1665 		nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1666 	}
1667 
1668 	if (report)
1669 		err = nlmsg_unicast(sk, skb, pid);
1670 
1671 	return err;
1672 }
1673 
1674 #ifdef CONFIG_PROC_FS
1675 struct nl_seq_iter {
1676 	struct net *net;
1677 	int link;
1678 	int hash_idx;
1679 };
1680 
1681 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1682 {
1683 	struct nl_seq_iter *iter = seq->private;
1684 	int i, j;
1685 	struct sock *s;
1686 	struct hlist_node *node;
1687 	loff_t off = 0;
1688 
1689 	for (i=0; i<MAX_LINKS; i++) {
1690 		struct nl_pid_hash *hash = &nl_table[i].hash;
1691 
1692 		for (j = 0; j <= hash->mask; j++) {
1693 			sk_for_each(s, node, &hash->table[j]) {
1694 				if (iter->net != s->sk_net)
1695 					continue;
1696 				if (off == pos) {
1697 					iter->link = i;
1698 					iter->hash_idx = j;
1699 					return s;
1700 				}
1701 				++off;
1702 			}
1703 		}
1704 	}
1705 	return NULL;
1706 }
1707 
1708 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1709 {
1710 	read_lock(&nl_table_lock);
1711 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1712 }
1713 
1714 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1715 {
1716 	struct sock *s;
1717 	struct nl_seq_iter *iter;
1718 	int i, j;
1719 
1720 	++*pos;
1721 
1722 	if (v == SEQ_START_TOKEN)
1723 		return netlink_seq_socket_idx(seq, 0);
1724 
1725 	iter = seq->private;
1726 	s = v;
1727 	do {
1728 		s = sk_next(s);
1729 	} while (s && (iter->net != s->sk_net));
1730 	if (s)
1731 		return s;
1732 
1733 	i = iter->link;
1734 	j = iter->hash_idx + 1;
1735 
1736 	do {
1737 		struct nl_pid_hash *hash = &nl_table[i].hash;
1738 
1739 		for (; j <= hash->mask; j++) {
1740 			s = sk_head(&hash->table[j]);
1741 			while (s && (iter->net != s->sk_net))
1742 				s = sk_next(s);
1743 			if (s) {
1744 				iter->link = i;
1745 				iter->hash_idx = j;
1746 				return s;
1747 			}
1748 		}
1749 
1750 		j = 0;
1751 	} while (++i < MAX_LINKS);
1752 
1753 	return NULL;
1754 }
1755 
1756 static void netlink_seq_stop(struct seq_file *seq, void *v)
1757 {
1758 	read_unlock(&nl_table_lock);
1759 }
1760 
1761 
1762 static int netlink_seq_show(struct seq_file *seq, void *v)
1763 {
1764 	if (v == SEQ_START_TOKEN)
1765 		seq_puts(seq,
1766 			 "sk       Eth Pid    Groups   "
1767 			 "Rmem     Wmem     Dump     Locks\n");
1768 	else {
1769 		struct sock *s = v;
1770 		struct netlink_sock *nlk = nlk_sk(s);
1771 
1772 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1773 			   s,
1774 			   s->sk_protocol,
1775 			   nlk->pid,
1776 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1777 			   atomic_read(&s->sk_rmem_alloc),
1778 			   atomic_read(&s->sk_wmem_alloc),
1779 			   nlk->cb,
1780 			   atomic_read(&s->sk_refcnt)
1781 			);
1782 
1783 	}
1784 	return 0;
1785 }
1786 
1787 static const struct seq_operations netlink_seq_ops = {
1788 	.start  = netlink_seq_start,
1789 	.next   = netlink_seq_next,
1790 	.stop   = netlink_seq_stop,
1791 	.show   = netlink_seq_show,
1792 };
1793 
1794 
1795 static int netlink_seq_open(struct inode *inode, struct file *file)
1796 {
1797 	struct nl_seq_iter *iter;
1798 
1799 	iter = __seq_open_private(file, &netlink_seq_ops, sizeof(*iter));
1800 	if (!iter)
1801 		return -ENOMEM;
1802 
1803 	iter->net = get_proc_net(inode);
1804 	if (!iter->net) {
1805 		seq_release_private(inode, file);
1806 		return -ENXIO;
1807 	}
1808 
1809 	return 0;
1810 }
1811 
1812 static int netlink_seq_release(struct inode *inode, struct file *file)
1813 {
1814 	struct seq_file *seq = file->private_data;
1815 	struct nl_seq_iter *iter = seq->private;
1816 	put_net(iter->net);
1817 	return seq_release_private(inode, file);
1818 }
1819 
1820 static const struct file_operations netlink_seq_fops = {
1821 	.owner		= THIS_MODULE,
1822 	.open		= netlink_seq_open,
1823 	.read		= seq_read,
1824 	.llseek		= seq_lseek,
1825 	.release	= netlink_seq_release,
1826 };
1827 
1828 #endif
1829 
1830 int netlink_register_notifier(struct notifier_block *nb)
1831 {
1832 	return atomic_notifier_chain_register(&netlink_chain, nb);
1833 }
1834 
1835 int netlink_unregister_notifier(struct notifier_block *nb)
1836 {
1837 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
1838 }
1839 
1840 static const struct proto_ops netlink_ops = {
1841 	.family =	PF_NETLINK,
1842 	.owner =	THIS_MODULE,
1843 	.release =	netlink_release,
1844 	.bind =		netlink_bind,
1845 	.connect =	netlink_connect,
1846 	.socketpair =	sock_no_socketpair,
1847 	.accept =	sock_no_accept,
1848 	.getname =	netlink_getname,
1849 	.poll =		datagram_poll,
1850 	.ioctl =	sock_no_ioctl,
1851 	.listen =	sock_no_listen,
1852 	.shutdown =	sock_no_shutdown,
1853 	.setsockopt =	netlink_setsockopt,
1854 	.getsockopt =	netlink_getsockopt,
1855 	.sendmsg =	netlink_sendmsg,
1856 	.recvmsg =	netlink_recvmsg,
1857 	.mmap =		sock_no_mmap,
1858 	.sendpage =	sock_no_sendpage,
1859 };
1860 
1861 static struct net_proto_family netlink_family_ops = {
1862 	.family = PF_NETLINK,
1863 	.create = netlink_create,
1864 	.owner	= THIS_MODULE,	/* for consistency 8) */
1865 };
1866 
1867 static int __net_init netlink_net_init(struct net *net)
1868 {
1869 #ifdef CONFIG_PROC_FS
1870 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1871 		return -ENOMEM;
1872 #endif
1873 	return 0;
1874 }
1875 
1876 static void __net_exit netlink_net_exit(struct net *net)
1877 {
1878 #ifdef CONFIG_PROC_FS
1879 	proc_net_remove(net, "netlink");
1880 #endif
1881 }
1882 
1883 static struct pernet_operations __net_initdata netlink_net_ops = {
1884 	.init = netlink_net_init,
1885 	.exit = netlink_net_exit,
1886 };
1887 
1888 static int __init netlink_proto_init(void)
1889 {
1890 	struct sk_buff *dummy_skb;
1891 	int i;
1892 	unsigned long limit;
1893 	unsigned int order;
1894 	int err = proto_register(&netlink_proto, 0);
1895 
1896 	if (err != 0)
1897 		goto out;
1898 
1899 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1900 
1901 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1902 	if (!nl_table)
1903 		goto panic;
1904 
1905 	if (num_physpages >= (128 * 1024))
1906 		limit = num_physpages >> (21 - PAGE_SHIFT);
1907 	else
1908 		limit = num_physpages >> (23 - PAGE_SHIFT);
1909 
1910 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1911 	limit = (1UL << order) / sizeof(struct hlist_head);
1912 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1913 
1914 	for (i = 0; i < MAX_LINKS; i++) {
1915 		struct nl_pid_hash *hash = &nl_table[i].hash;
1916 
1917 		hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1918 		if (!hash->table) {
1919 			while (i-- > 0)
1920 				nl_pid_hash_free(nl_table[i].hash.table,
1921 						 1 * sizeof(*hash->table));
1922 			kfree(nl_table);
1923 			goto panic;
1924 		}
1925 		memset(hash->table, 0, 1 * sizeof(*hash->table));
1926 		hash->max_shift = order;
1927 		hash->shift = 0;
1928 		hash->mask = 0;
1929 		hash->rehash_time = jiffies;
1930 	}
1931 
1932 	sock_register(&netlink_family_ops);
1933 	register_pernet_subsys(&netlink_net_ops);
1934 	/* The netlink device handler may be needed early. */
1935 	rtnetlink_init();
1936 out:
1937 	return err;
1938 panic:
1939 	panic("netlink_init: Cannot allocate nl_table\n");
1940 }
1941 
1942 core_initcall(netlink_proto_init);
1943 
1944 EXPORT_SYMBOL(netlink_ack);
1945 EXPORT_SYMBOL(netlink_rcv_skb);
1946 EXPORT_SYMBOL(netlink_broadcast);
1947 EXPORT_SYMBOL(netlink_dump_start);
1948 EXPORT_SYMBOL(netlink_kernel_create);
1949 EXPORT_SYMBOL(netlink_register_notifier);
1950 EXPORT_SYMBOL(netlink_set_nonroot);
1951 EXPORT_SYMBOL(netlink_unicast);
1952 EXPORT_SYMBOL(netlink_unregister_notifier);
1953 EXPORT_SYMBOL(nlmsg_notify);
1954