xref: /linux/net/netlink/af_netlink.c (revision 367b8112fe2ea5c39a7bb4d263dcdd9b612fae18)
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 	nlk = nlk_sk(sock->sk);
456 	nlk->module = module;
457 out:
458 	return err;
459 
460 out_module:
461 	module_put(module);
462 	goto out;
463 }
464 
465 static int netlink_release(struct socket *sock)
466 {
467 	struct sock *sk = sock->sk;
468 	struct netlink_sock *nlk;
469 
470 	if (!sk)
471 		return 0;
472 
473 	netlink_remove(sk);
474 	sock_orphan(sk);
475 	nlk = nlk_sk(sk);
476 
477 	/*
478 	 * OK. Socket is unlinked, any packets that arrive now
479 	 * will be purged.
480 	 */
481 
482 	sock->sk = NULL;
483 	wake_up_interruptible_all(&nlk->wait);
484 
485 	skb_queue_purge(&sk->sk_write_queue);
486 
487 	if (nlk->pid && !nlk->subscriptions) {
488 		struct netlink_notify n = {
489 						.net = sock_net(sk),
490 						.protocol = sk->sk_protocol,
491 						.pid = nlk->pid,
492 					  };
493 		atomic_notifier_call_chain(&netlink_chain,
494 				NETLINK_URELEASE, &n);
495 	}
496 
497 	module_put(nlk->module);
498 
499 	netlink_table_grab();
500 	if (netlink_is_kernel(sk)) {
501 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
502 		if (--nl_table[sk->sk_protocol].registered == 0) {
503 			kfree(nl_table[sk->sk_protocol].listeners);
504 			nl_table[sk->sk_protocol].module = NULL;
505 			nl_table[sk->sk_protocol].registered = 0;
506 		}
507 	} else if (nlk->subscriptions)
508 		netlink_update_listeners(sk);
509 	netlink_table_ungrab();
510 
511 	kfree(nlk->groups);
512 	nlk->groups = NULL;
513 
514 	sock_put(sk);
515 	return 0;
516 }
517 
518 static int netlink_autobind(struct socket *sock)
519 {
520 	struct sock *sk = sock->sk;
521 	struct net *net = sock_net(sk);
522 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
523 	struct hlist_head *head;
524 	struct sock *osk;
525 	struct hlist_node *node;
526 	s32 pid = current->tgid;
527 	int err;
528 	static s32 rover = -4097;
529 
530 retry:
531 	cond_resched();
532 	netlink_table_grab();
533 	head = nl_pid_hashfn(hash, pid);
534 	sk_for_each(osk, node, head) {
535 		if (!net_eq(sock_net(osk), net))
536 			continue;
537 		if (nlk_sk(osk)->pid == pid) {
538 			/* Bind collision, search negative pid values. */
539 			pid = rover--;
540 			if (rover > -4097)
541 				rover = -4097;
542 			netlink_table_ungrab();
543 			goto retry;
544 		}
545 	}
546 	netlink_table_ungrab();
547 
548 	err = netlink_insert(sk, net, pid);
549 	if (err == -EADDRINUSE)
550 		goto retry;
551 
552 	/* If 2 threads race to autobind, that is fine.  */
553 	if (err == -EBUSY)
554 		err = 0;
555 
556 	return err;
557 }
558 
559 static inline int netlink_capable(struct socket *sock, unsigned int flag)
560 {
561 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
562 	       capable(CAP_NET_ADMIN);
563 }
564 
565 static void
566 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
567 {
568 	struct netlink_sock *nlk = nlk_sk(sk);
569 
570 	if (nlk->subscriptions && !subscriptions)
571 		__sk_del_bind_node(sk);
572 	else if (!nlk->subscriptions && subscriptions)
573 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
574 	nlk->subscriptions = subscriptions;
575 }
576 
577 static int netlink_realloc_groups(struct sock *sk)
578 {
579 	struct netlink_sock *nlk = nlk_sk(sk);
580 	unsigned int groups;
581 	unsigned long *new_groups;
582 	int err = 0;
583 
584 	netlink_table_grab();
585 
586 	groups = nl_table[sk->sk_protocol].groups;
587 	if (!nl_table[sk->sk_protocol].registered) {
588 		err = -ENOENT;
589 		goto out_unlock;
590 	}
591 
592 	if (nlk->ngroups >= groups)
593 		goto out_unlock;
594 
595 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
596 	if (new_groups == NULL) {
597 		err = -ENOMEM;
598 		goto out_unlock;
599 	}
600 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
601 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
602 
603 	nlk->groups = new_groups;
604 	nlk->ngroups = groups;
605  out_unlock:
606 	netlink_table_ungrab();
607 	return err;
608 }
609 
610 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
611 			int addr_len)
612 {
613 	struct sock *sk = sock->sk;
614 	struct net *net = sock_net(sk);
615 	struct netlink_sock *nlk = nlk_sk(sk);
616 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
617 	int err;
618 
619 	if (nladdr->nl_family != AF_NETLINK)
620 		return -EINVAL;
621 
622 	/* Only superuser is allowed to listen multicasts */
623 	if (nladdr->nl_groups) {
624 		if (!netlink_capable(sock, NL_NONROOT_RECV))
625 			return -EPERM;
626 		err = netlink_realloc_groups(sk);
627 		if (err)
628 			return err;
629 	}
630 
631 	if (nlk->pid) {
632 		if (nladdr->nl_pid != nlk->pid)
633 			return -EINVAL;
634 	} else {
635 		err = nladdr->nl_pid ?
636 			netlink_insert(sk, net, nladdr->nl_pid) :
637 			netlink_autobind(sock);
638 		if (err)
639 			return err;
640 	}
641 
642 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
643 		return 0;
644 
645 	netlink_table_grab();
646 	netlink_update_subscriptions(sk, nlk->subscriptions +
647 					 hweight32(nladdr->nl_groups) -
648 					 hweight32(nlk->groups[0]));
649 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
650 	netlink_update_listeners(sk);
651 	netlink_table_ungrab();
652 
653 	return 0;
654 }
655 
656 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
657 			   int alen, int flags)
658 {
659 	int err = 0;
660 	struct sock *sk = sock->sk;
661 	struct netlink_sock *nlk = nlk_sk(sk);
662 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
663 
664 	if (addr->sa_family == AF_UNSPEC) {
665 		sk->sk_state	= NETLINK_UNCONNECTED;
666 		nlk->dst_pid	= 0;
667 		nlk->dst_group  = 0;
668 		return 0;
669 	}
670 	if (addr->sa_family != AF_NETLINK)
671 		return -EINVAL;
672 
673 	/* Only superuser is allowed to send multicasts */
674 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
675 		return -EPERM;
676 
677 	if (!nlk->pid)
678 		err = netlink_autobind(sock);
679 
680 	if (err == 0) {
681 		sk->sk_state	= NETLINK_CONNECTED;
682 		nlk->dst_pid 	= nladdr->nl_pid;
683 		nlk->dst_group  = ffs(nladdr->nl_groups);
684 	}
685 
686 	return err;
687 }
688 
689 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
690 			   int *addr_len, int peer)
691 {
692 	struct sock *sk = sock->sk;
693 	struct netlink_sock *nlk = nlk_sk(sk);
694 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
695 
696 	nladdr->nl_family = AF_NETLINK;
697 	nladdr->nl_pad = 0;
698 	*addr_len = sizeof(*nladdr);
699 
700 	if (peer) {
701 		nladdr->nl_pid = nlk->dst_pid;
702 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
703 	} else {
704 		nladdr->nl_pid = nlk->pid;
705 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
706 	}
707 	return 0;
708 }
709 
710 static void netlink_overrun(struct sock *sk)
711 {
712 	if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
713 		sk->sk_err = ENOBUFS;
714 		sk->sk_error_report(sk);
715 	}
716 }
717 
718 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
719 {
720 	struct sock *sock;
721 	struct netlink_sock *nlk;
722 
723 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
724 	if (!sock)
725 		return ERR_PTR(-ECONNREFUSED);
726 
727 	/* Don't bother queuing skb if kernel socket has no input function */
728 	nlk = nlk_sk(sock);
729 	if (sock->sk_state == NETLINK_CONNECTED &&
730 	    nlk->dst_pid != nlk_sk(ssk)->pid) {
731 		sock_put(sock);
732 		return ERR_PTR(-ECONNREFUSED);
733 	}
734 	return sock;
735 }
736 
737 struct sock *netlink_getsockbyfilp(struct file *filp)
738 {
739 	struct inode *inode = filp->f_path.dentry->d_inode;
740 	struct sock *sock;
741 
742 	if (!S_ISSOCK(inode->i_mode))
743 		return ERR_PTR(-ENOTSOCK);
744 
745 	sock = SOCKET_I(inode)->sk;
746 	if (sock->sk_family != AF_NETLINK)
747 		return ERR_PTR(-EINVAL);
748 
749 	sock_hold(sock);
750 	return sock;
751 }
752 
753 /*
754  * Attach a skb to a netlink socket.
755  * The caller must hold a reference to the destination socket. On error, the
756  * reference is dropped. The skb is not send to the destination, just all
757  * all error checks are performed and memory in the queue is reserved.
758  * Return values:
759  * < 0: error. skb freed, reference to sock dropped.
760  * 0: continue
761  * 1: repeat lookup - reference dropped while waiting for socket memory.
762  */
763 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
764 		      long *timeo, struct sock *ssk)
765 {
766 	struct netlink_sock *nlk;
767 
768 	nlk = nlk_sk(sk);
769 
770 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
771 	    test_bit(0, &nlk->state)) {
772 		DECLARE_WAITQUEUE(wait, current);
773 		if (!*timeo) {
774 			if (!ssk || netlink_is_kernel(ssk))
775 				netlink_overrun(sk);
776 			sock_put(sk);
777 			kfree_skb(skb);
778 			return -EAGAIN;
779 		}
780 
781 		__set_current_state(TASK_INTERRUPTIBLE);
782 		add_wait_queue(&nlk->wait, &wait);
783 
784 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
785 		     test_bit(0, &nlk->state)) &&
786 		    !sock_flag(sk, SOCK_DEAD))
787 			*timeo = schedule_timeout(*timeo);
788 
789 		__set_current_state(TASK_RUNNING);
790 		remove_wait_queue(&nlk->wait, &wait);
791 		sock_put(sk);
792 
793 		if (signal_pending(current)) {
794 			kfree_skb(skb);
795 			return sock_intr_errno(*timeo);
796 		}
797 		return 1;
798 	}
799 	skb_set_owner_r(skb, sk);
800 	return 0;
801 }
802 
803 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
804 {
805 	int len = skb->len;
806 
807 	skb_queue_tail(&sk->sk_receive_queue, skb);
808 	sk->sk_data_ready(sk, len);
809 	sock_put(sk);
810 	return len;
811 }
812 
813 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
814 {
815 	kfree_skb(skb);
816 	sock_put(sk);
817 }
818 
819 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
820 					   gfp_t allocation)
821 {
822 	int delta;
823 
824 	skb_orphan(skb);
825 
826 	delta = skb->end - skb->tail;
827 	if (delta * 2 < skb->truesize)
828 		return skb;
829 
830 	if (skb_shared(skb)) {
831 		struct sk_buff *nskb = skb_clone(skb, allocation);
832 		if (!nskb)
833 			return skb;
834 		kfree_skb(skb);
835 		skb = nskb;
836 	}
837 
838 	if (!pskb_expand_head(skb, 0, -delta, allocation))
839 		skb->truesize -= delta;
840 
841 	return skb;
842 }
843 
844 static inline void netlink_rcv_wake(struct sock *sk)
845 {
846 	struct netlink_sock *nlk = nlk_sk(sk);
847 
848 	if (skb_queue_empty(&sk->sk_receive_queue))
849 		clear_bit(0, &nlk->state);
850 	if (!test_bit(0, &nlk->state))
851 		wake_up_interruptible(&nlk->wait);
852 }
853 
854 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
855 {
856 	int ret;
857 	struct netlink_sock *nlk = nlk_sk(sk);
858 
859 	ret = -ECONNREFUSED;
860 	if (nlk->netlink_rcv != NULL) {
861 		ret = skb->len;
862 		skb_set_owner_r(skb, sk);
863 		nlk->netlink_rcv(skb);
864 	}
865 	kfree_skb(skb);
866 	sock_put(sk);
867 	return ret;
868 }
869 
870 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
871 		    u32 pid, int nonblock)
872 {
873 	struct sock *sk;
874 	int err;
875 	long timeo;
876 
877 	skb = netlink_trim(skb, gfp_any());
878 
879 	timeo = sock_sndtimeo(ssk, nonblock);
880 retry:
881 	sk = netlink_getsockbypid(ssk, pid);
882 	if (IS_ERR(sk)) {
883 		kfree_skb(skb);
884 		return PTR_ERR(sk);
885 	}
886 	if (netlink_is_kernel(sk))
887 		return netlink_unicast_kernel(sk, skb);
888 
889 	if (sk_filter(sk, skb)) {
890 		err = skb->len;
891 		kfree_skb(skb);
892 		sock_put(sk);
893 		return err;
894 	}
895 
896 	err = netlink_attachskb(sk, skb, &timeo, ssk);
897 	if (err == 1)
898 		goto retry;
899 	if (err)
900 		return err;
901 
902 	return netlink_sendskb(sk, skb);
903 }
904 EXPORT_SYMBOL(netlink_unicast);
905 
906 int netlink_has_listeners(struct sock *sk, unsigned int group)
907 {
908 	int res = 0;
909 	unsigned long *listeners;
910 
911 	BUG_ON(!netlink_is_kernel(sk));
912 
913 	rcu_read_lock();
914 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
915 
916 	if (group - 1 < nl_table[sk->sk_protocol].groups)
917 		res = test_bit(group - 1, listeners);
918 
919 	rcu_read_unlock();
920 
921 	return res;
922 }
923 EXPORT_SYMBOL_GPL(netlink_has_listeners);
924 
925 static inline int netlink_broadcast_deliver(struct sock *sk,
926 					    struct sk_buff *skb)
927 {
928 	struct netlink_sock *nlk = nlk_sk(sk);
929 
930 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
931 	    !test_bit(0, &nlk->state)) {
932 		skb_set_owner_r(skb, sk);
933 		skb_queue_tail(&sk->sk_receive_queue, skb);
934 		sk->sk_data_ready(sk, skb->len);
935 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
936 	}
937 	return -1;
938 }
939 
940 struct netlink_broadcast_data {
941 	struct sock *exclude_sk;
942 	struct net *net;
943 	u32 pid;
944 	u32 group;
945 	int failure;
946 	int congested;
947 	int delivered;
948 	gfp_t allocation;
949 	struct sk_buff *skb, *skb2;
950 };
951 
952 static inline int do_one_broadcast(struct sock *sk,
953 				   struct netlink_broadcast_data *p)
954 {
955 	struct netlink_sock *nlk = nlk_sk(sk);
956 	int val;
957 
958 	if (p->exclude_sk == sk)
959 		goto out;
960 
961 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
962 	    !test_bit(p->group - 1, nlk->groups))
963 		goto out;
964 
965 	if (!net_eq(sock_net(sk), p->net))
966 		goto out;
967 
968 	if (p->failure) {
969 		netlink_overrun(sk);
970 		goto out;
971 	}
972 
973 	sock_hold(sk);
974 	if (p->skb2 == NULL) {
975 		if (skb_shared(p->skb)) {
976 			p->skb2 = skb_clone(p->skb, p->allocation);
977 		} else {
978 			p->skb2 = skb_get(p->skb);
979 			/*
980 			 * skb ownership may have been set when
981 			 * delivered to a previous socket.
982 			 */
983 			skb_orphan(p->skb2);
984 		}
985 	}
986 	if (p->skb2 == NULL) {
987 		netlink_overrun(sk);
988 		/* Clone failed. Notify ALL listeners. */
989 		p->failure = 1;
990 	} else if (sk_filter(sk, p->skb2)) {
991 		kfree_skb(p->skb2);
992 		p->skb2 = NULL;
993 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
994 		netlink_overrun(sk);
995 	} else {
996 		p->congested |= val;
997 		p->delivered = 1;
998 		p->skb2 = NULL;
999 	}
1000 	sock_put(sk);
1001 
1002 out:
1003 	return 0;
1004 }
1005 
1006 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1007 		      u32 group, gfp_t allocation)
1008 {
1009 	struct net *net = sock_net(ssk);
1010 	struct netlink_broadcast_data info;
1011 	struct hlist_node *node;
1012 	struct sock *sk;
1013 
1014 	skb = netlink_trim(skb, allocation);
1015 
1016 	info.exclude_sk = ssk;
1017 	info.net = net;
1018 	info.pid = pid;
1019 	info.group = group;
1020 	info.failure = 0;
1021 	info.congested = 0;
1022 	info.delivered = 0;
1023 	info.allocation = allocation;
1024 	info.skb = skb;
1025 	info.skb2 = NULL;
1026 
1027 	/* While we sleep in clone, do not allow to change socket list */
1028 
1029 	netlink_lock_table();
1030 
1031 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1032 		do_one_broadcast(sk, &info);
1033 
1034 	kfree_skb(skb);
1035 
1036 	netlink_unlock_table();
1037 
1038 	if (info.skb2)
1039 		kfree_skb(info.skb2);
1040 
1041 	if (info.delivered) {
1042 		if (info.congested && (allocation & __GFP_WAIT))
1043 			yield();
1044 		return 0;
1045 	}
1046 	if (info.failure)
1047 		return -ENOBUFS;
1048 	return -ESRCH;
1049 }
1050 EXPORT_SYMBOL(netlink_broadcast);
1051 
1052 struct netlink_set_err_data {
1053 	struct sock *exclude_sk;
1054 	u32 pid;
1055 	u32 group;
1056 	int code;
1057 };
1058 
1059 static inline int do_one_set_err(struct sock *sk,
1060 				 struct netlink_set_err_data *p)
1061 {
1062 	struct netlink_sock *nlk = nlk_sk(sk);
1063 
1064 	if (sk == p->exclude_sk)
1065 		goto out;
1066 
1067 	if (sock_net(sk) != sock_net(p->exclude_sk))
1068 		goto out;
1069 
1070 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1071 	    !test_bit(p->group - 1, nlk->groups))
1072 		goto out;
1073 
1074 	sk->sk_err = p->code;
1075 	sk->sk_error_report(sk);
1076 out:
1077 	return 0;
1078 }
1079 
1080 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1081 {
1082 	struct netlink_set_err_data info;
1083 	struct hlist_node *node;
1084 	struct sock *sk;
1085 
1086 	info.exclude_sk = ssk;
1087 	info.pid = pid;
1088 	info.group = group;
1089 	info.code = code;
1090 
1091 	read_lock(&nl_table_lock);
1092 
1093 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1094 		do_one_set_err(sk, &info);
1095 
1096 	read_unlock(&nl_table_lock);
1097 }
1098 
1099 /* must be called with netlink table grabbed */
1100 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1101 				     unsigned int group,
1102 				     int is_new)
1103 {
1104 	int old, new = !!is_new, subscriptions;
1105 
1106 	old = test_bit(group - 1, nlk->groups);
1107 	subscriptions = nlk->subscriptions - old + new;
1108 	if (new)
1109 		__set_bit(group - 1, nlk->groups);
1110 	else
1111 		__clear_bit(group - 1, nlk->groups);
1112 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1113 	netlink_update_listeners(&nlk->sk);
1114 }
1115 
1116 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1117 			      char __user *optval, int optlen)
1118 {
1119 	struct sock *sk = sock->sk;
1120 	struct netlink_sock *nlk = nlk_sk(sk);
1121 	unsigned int val = 0;
1122 	int err;
1123 
1124 	if (level != SOL_NETLINK)
1125 		return -ENOPROTOOPT;
1126 
1127 	if (optlen >= sizeof(int) &&
1128 	    get_user(val, (unsigned int __user *)optval))
1129 		return -EFAULT;
1130 
1131 	switch (optname) {
1132 	case NETLINK_PKTINFO:
1133 		if (val)
1134 			nlk->flags |= NETLINK_RECV_PKTINFO;
1135 		else
1136 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1137 		err = 0;
1138 		break;
1139 	case NETLINK_ADD_MEMBERSHIP:
1140 	case NETLINK_DROP_MEMBERSHIP: {
1141 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1142 			return -EPERM;
1143 		err = netlink_realloc_groups(sk);
1144 		if (err)
1145 			return err;
1146 		if (!val || val - 1 >= nlk->ngroups)
1147 			return -EINVAL;
1148 		netlink_table_grab();
1149 		netlink_update_socket_mc(nlk, val,
1150 					 optname == NETLINK_ADD_MEMBERSHIP);
1151 		netlink_table_ungrab();
1152 		err = 0;
1153 		break;
1154 	}
1155 	default:
1156 		err = -ENOPROTOOPT;
1157 	}
1158 	return err;
1159 }
1160 
1161 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1162 			      char __user *optval, int __user *optlen)
1163 {
1164 	struct sock *sk = sock->sk;
1165 	struct netlink_sock *nlk = nlk_sk(sk);
1166 	int len, val, err;
1167 
1168 	if (level != SOL_NETLINK)
1169 		return -ENOPROTOOPT;
1170 
1171 	if (get_user(len, optlen))
1172 		return -EFAULT;
1173 	if (len < 0)
1174 		return -EINVAL;
1175 
1176 	switch (optname) {
1177 	case NETLINK_PKTINFO:
1178 		if (len < sizeof(int))
1179 			return -EINVAL;
1180 		len = sizeof(int);
1181 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1182 		if (put_user(len, optlen) ||
1183 		    put_user(val, optval))
1184 			return -EFAULT;
1185 		err = 0;
1186 		break;
1187 	default:
1188 		err = -ENOPROTOOPT;
1189 	}
1190 	return err;
1191 }
1192 
1193 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1194 {
1195 	struct nl_pktinfo info;
1196 
1197 	info.group = NETLINK_CB(skb).dst_group;
1198 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1199 }
1200 
1201 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1202 			   struct msghdr *msg, size_t len)
1203 {
1204 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1205 	struct sock *sk = sock->sk;
1206 	struct netlink_sock *nlk = nlk_sk(sk);
1207 	struct sockaddr_nl *addr = msg->msg_name;
1208 	u32 dst_pid;
1209 	u32 dst_group;
1210 	struct sk_buff *skb;
1211 	int err;
1212 	struct scm_cookie scm;
1213 
1214 	if (msg->msg_flags&MSG_OOB)
1215 		return -EOPNOTSUPP;
1216 
1217 	if (NULL == siocb->scm)
1218 		siocb->scm = &scm;
1219 	err = scm_send(sock, msg, siocb->scm);
1220 	if (err < 0)
1221 		return err;
1222 
1223 	if (msg->msg_namelen) {
1224 		if (addr->nl_family != AF_NETLINK)
1225 			return -EINVAL;
1226 		dst_pid = addr->nl_pid;
1227 		dst_group = ffs(addr->nl_groups);
1228 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1229 			return -EPERM;
1230 	} else {
1231 		dst_pid = nlk->dst_pid;
1232 		dst_group = nlk->dst_group;
1233 	}
1234 
1235 	if (!nlk->pid) {
1236 		err = netlink_autobind(sock);
1237 		if (err)
1238 			goto out;
1239 	}
1240 
1241 	err = -EMSGSIZE;
1242 	if (len > sk->sk_sndbuf - 32)
1243 		goto out;
1244 	err = -ENOBUFS;
1245 	skb = alloc_skb(len, GFP_KERNEL);
1246 	if (skb == NULL)
1247 		goto out;
1248 
1249 	NETLINK_CB(skb).pid	= nlk->pid;
1250 	NETLINK_CB(skb).dst_group = dst_group;
1251 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1252 	NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1253 	security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1254 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1255 
1256 	/* What can I do? Netlink is asynchronous, so that
1257 	   we will have to save current capabilities to
1258 	   check them, when this message will be delivered
1259 	   to corresponding kernel module.   --ANK (980802)
1260 	 */
1261 
1262 	err = -EFAULT;
1263 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1264 		kfree_skb(skb);
1265 		goto out;
1266 	}
1267 
1268 	err = security_netlink_send(sk, skb);
1269 	if (err) {
1270 		kfree_skb(skb);
1271 		goto out;
1272 	}
1273 
1274 	if (dst_group) {
1275 		atomic_inc(&skb->users);
1276 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1277 	}
1278 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1279 
1280 out:
1281 	return err;
1282 }
1283 
1284 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1285 			   struct msghdr *msg, size_t len,
1286 			   int flags)
1287 {
1288 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1289 	struct scm_cookie scm;
1290 	struct sock *sk = sock->sk;
1291 	struct netlink_sock *nlk = nlk_sk(sk);
1292 	int noblock = flags&MSG_DONTWAIT;
1293 	size_t copied;
1294 	struct sk_buff *skb;
1295 	int err;
1296 
1297 	if (flags&MSG_OOB)
1298 		return -EOPNOTSUPP;
1299 
1300 	copied = 0;
1301 
1302 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1303 	if (skb == NULL)
1304 		goto out;
1305 
1306 	msg->msg_namelen = 0;
1307 
1308 	copied = skb->len;
1309 	if (len < copied) {
1310 		msg->msg_flags |= MSG_TRUNC;
1311 		copied = len;
1312 	}
1313 
1314 	skb_reset_transport_header(skb);
1315 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1316 
1317 	if (msg->msg_name) {
1318 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1319 		addr->nl_family = AF_NETLINK;
1320 		addr->nl_pad    = 0;
1321 		addr->nl_pid	= NETLINK_CB(skb).pid;
1322 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1323 		msg->msg_namelen = sizeof(*addr);
1324 	}
1325 
1326 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1327 		netlink_cmsg_recv_pktinfo(msg, skb);
1328 
1329 	if (NULL == siocb->scm) {
1330 		memset(&scm, 0, sizeof(scm));
1331 		siocb->scm = &scm;
1332 	}
1333 	siocb->scm->creds = *NETLINK_CREDS(skb);
1334 	if (flags & MSG_TRUNC)
1335 		copied = skb->len;
1336 	skb_free_datagram(sk, skb);
1337 
1338 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1339 		netlink_dump(sk);
1340 
1341 	scm_recv(sock, msg, siocb->scm, flags);
1342 out:
1343 	netlink_rcv_wake(sk);
1344 	return err ? : copied;
1345 }
1346 
1347 static void netlink_data_ready(struct sock *sk, int len)
1348 {
1349 	BUG();
1350 }
1351 
1352 /*
1353  *	We export these functions to other modules. They provide a
1354  *	complete set of kernel non-blocking support for message
1355  *	queueing.
1356  */
1357 
1358 struct sock *
1359 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1360 		      void (*input)(struct sk_buff *skb),
1361 		      struct mutex *cb_mutex, struct module *module)
1362 {
1363 	struct socket *sock;
1364 	struct sock *sk;
1365 	struct netlink_sock *nlk;
1366 	unsigned long *listeners = NULL;
1367 
1368 	BUG_ON(!nl_table);
1369 
1370 	if (unit < 0 || unit >= MAX_LINKS)
1371 		return NULL;
1372 
1373 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1374 		return NULL;
1375 
1376 	/*
1377 	 * We have to just have a reference on the net from sk, but don't
1378 	 * get_net it. Besides, we cannot get and then put the net here.
1379 	 * So we create one inside init_net and the move it to net.
1380 	 */
1381 
1382 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1383 		goto out_sock_release_nosk;
1384 
1385 	sk = sock->sk;
1386 	sk_change_net(sk, net);
1387 
1388 	if (groups < 32)
1389 		groups = 32;
1390 
1391 	listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1392 	if (!listeners)
1393 		goto out_sock_release;
1394 
1395 	sk->sk_data_ready = netlink_data_ready;
1396 	if (input)
1397 		nlk_sk(sk)->netlink_rcv = input;
1398 
1399 	if (netlink_insert(sk, net, 0))
1400 		goto out_sock_release;
1401 
1402 	nlk = nlk_sk(sk);
1403 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1404 
1405 	netlink_table_grab();
1406 	if (!nl_table[unit].registered) {
1407 		nl_table[unit].groups = groups;
1408 		nl_table[unit].listeners = listeners;
1409 		nl_table[unit].cb_mutex = cb_mutex;
1410 		nl_table[unit].module = module;
1411 		nl_table[unit].registered = 1;
1412 	} else {
1413 		kfree(listeners);
1414 		nl_table[unit].registered++;
1415 	}
1416 	netlink_table_ungrab();
1417 	return sk;
1418 
1419 out_sock_release:
1420 	kfree(listeners);
1421 	netlink_kernel_release(sk);
1422 	return NULL;
1423 
1424 out_sock_release_nosk:
1425 	sock_release(sock);
1426 	return NULL;
1427 }
1428 EXPORT_SYMBOL(netlink_kernel_create);
1429 
1430 
1431 void
1432 netlink_kernel_release(struct sock *sk)
1433 {
1434 	sk_release_kernel(sk);
1435 }
1436 EXPORT_SYMBOL(netlink_kernel_release);
1437 
1438 
1439 /**
1440  * netlink_change_ngroups - change number of multicast groups
1441  *
1442  * This changes the number of multicast groups that are available
1443  * on a certain netlink family. Note that it is not possible to
1444  * change the number of groups to below 32. Also note that it does
1445  * not implicitly call netlink_clear_multicast_users() when the
1446  * number of groups is reduced.
1447  *
1448  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1449  * @groups: The new number of groups.
1450  */
1451 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1452 {
1453 	unsigned long *listeners, *old = NULL;
1454 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1455 	int err = 0;
1456 
1457 	if (groups < 32)
1458 		groups = 32;
1459 
1460 	netlink_table_grab();
1461 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1462 		listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1463 		if (!listeners) {
1464 			err = -ENOMEM;
1465 			goto out_ungrab;
1466 		}
1467 		old = tbl->listeners;
1468 		memcpy(listeners, old, NLGRPSZ(tbl->groups));
1469 		rcu_assign_pointer(tbl->listeners, listeners);
1470 	}
1471 	tbl->groups = groups;
1472 
1473  out_ungrab:
1474 	netlink_table_ungrab();
1475 	synchronize_rcu();
1476 	kfree(old);
1477 	return err;
1478 }
1479 EXPORT_SYMBOL(netlink_change_ngroups);
1480 
1481 /**
1482  * netlink_clear_multicast_users - kick off multicast listeners
1483  *
1484  * This function removes all listeners from the given group.
1485  * @ksk: The kernel netlink socket, as returned by
1486  *	netlink_kernel_create().
1487  * @group: The multicast group to clear.
1488  */
1489 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1490 {
1491 	struct sock *sk;
1492 	struct hlist_node *node;
1493 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1494 
1495 	netlink_table_grab();
1496 
1497 	sk_for_each_bound(sk, node, &tbl->mc_list)
1498 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1499 
1500 	netlink_table_ungrab();
1501 }
1502 EXPORT_SYMBOL(netlink_clear_multicast_users);
1503 
1504 void netlink_set_nonroot(int protocol, unsigned int flags)
1505 {
1506 	if ((unsigned int)protocol < MAX_LINKS)
1507 		nl_table[protocol].nl_nonroot = flags;
1508 }
1509 EXPORT_SYMBOL(netlink_set_nonroot);
1510 
1511 static void netlink_destroy_callback(struct netlink_callback *cb)
1512 {
1513 	if (cb->skb)
1514 		kfree_skb(cb->skb);
1515 	kfree(cb);
1516 }
1517 
1518 /*
1519  * It looks a bit ugly.
1520  * It would be better to create kernel thread.
1521  */
1522 
1523 static int netlink_dump(struct sock *sk)
1524 {
1525 	struct netlink_sock *nlk = nlk_sk(sk);
1526 	struct netlink_callback *cb;
1527 	struct sk_buff *skb;
1528 	struct nlmsghdr *nlh;
1529 	int len, err = -ENOBUFS;
1530 
1531 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1532 	if (!skb)
1533 		goto errout;
1534 
1535 	mutex_lock(nlk->cb_mutex);
1536 
1537 	cb = nlk->cb;
1538 	if (cb == NULL) {
1539 		err = -EINVAL;
1540 		goto errout_skb;
1541 	}
1542 
1543 	len = cb->dump(skb, cb);
1544 
1545 	if (len > 0) {
1546 		mutex_unlock(nlk->cb_mutex);
1547 
1548 		if (sk_filter(sk, skb))
1549 			kfree_skb(skb);
1550 		else {
1551 			skb_queue_tail(&sk->sk_receive_queue, skb);
1552 			sk->sk_data_ready(sk, skb->len);
1553 		}
1554 		return 0;
1555 	}
1556 
1557 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1558 	if (!nlh)
1559 		goto errout_skb;
1560 
1561 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1562 
1563 	if (sk_filter(sk, skb))
1564 		kfree_skb(skb);
1565 	else {
1566 		skb_queue_tail(&sk->sk_receive_queue, skb);
1567 		sk->sk_data_ready(sk, skb->len);
1568 	}
1569 
1570 	if (cb->done)
1571 		cb->done(cb);
1572 	nlk->cb = NULL;
1573 	mutex_unlock(nlk->cb_mutex);
1574 
1575 	netlink_destroy_callback(cb);
1576 	return 0;
1577 
1578 errout_skb:
1579 	mutex_unlock(nlk->cb_mutex);
1580 	kfree_skb(skb);
1581 errout:
1582 	return err;
1583 }
1584 
1585 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1586 		       struct nlmsghdr *nlh,
1587 		       int (*dump)(struct sk_buff *skb,
1588 				   struct netlink_callback *),
1589 		       int (*done)(struct netlink_callback *))
1590 {
1591 	struct netlink_callback *cb;
1592 	struct sock *sk;
1593 	struct netlink_sock *nlk;
1594 
1595 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1596 	if (cb == NULL)
1597 		return -ENOBUFS;
1598 
1599 	cb->dump = dump;
1600 	cb->done = done;
1601 	cb->nlh = nlh;
1602 	atomic_inc(&skb->users);
1603 	cb->skb = skb;
1604 
1605 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1606 	if (sk == NULL) {
1607 		netlink_destroy_callback(cb);
1608 		return -ECONNREFUSED;
1609 	}
1610 	nlk = nlk_sk(sk);
1611 	/* A dump is in progress... */
1612 	mutex_lock(nlk->cb_mutex);
1613 	if (nlk->cb) {
1614 		mutex_unlock(nlk->cb_mutex);
1615 		netlink_destroy_callback(cb);
1616 		sock_put(sk);
1617 		return -EBUSY;
1618 	}
1619 	nlk->cb = cb;
1620 	mutex_unlock(nlk->cb_mutex);
1621 
1622 	netlink_dump(sk);
1623 	sock_put(sk);
1624 
1625 	/* We successfully started a dump, by returning -EINTR we
1626 	 * signal not to send ACK even if it was requested.
1627 	 */
1628 	return -EINTR;
1629 }
1630 EXPORT_SYMBOL(netlink_dump_start);
1631 
1632 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1633 {
1634 	struct sk_buff *skb;
1635 	struct nlmsghdr *rep;
1636 	struct nlmsgerr *errmsg;
1637 	size_t payload = sizeof(*errmsg);
1638 
1639 	/* error messages get the original request appened */
1640 	if (err)
1641 		payload += nlmsg_len(nlh);
1642 
1643 	skb = nlmsg_new(payload, GFP_KERNEL);
1644 	if (!skb) {
1645 		struct sock *sk;
1646 
1647 		sk = netlink_lookup(sock_net(in_skb->sk),
1648 				    in_skb->sk->sk_protocol,
1649 				    NETLINK_CB(in_skb).pid);
1650 		if (sk) {
1651 			sk->sk_err = ENOBUFS;
1652 			sk->sk_error_report(sk);
1653 			sock_put(sk);
1654 		}
1655 		return;
1656 	}
1657 
1658 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1659 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1660 	errmsg = nlmsg_data(rep);
1661 	errmsg->error = err;
1662 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1663 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1664 }
1665 EXPORT_SYMBOL(netlink_ack);
1666 
1667 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1668 						     struct nlmsghdr *))
1669 {
1670 	struct nlmsghdr *nlh;
1671 	int err;
1672 
1673 	while (skb->len >= nlmsg_total_size(0)) {
1674 		int msglen;
1675 
1676 		nlh = nlmsg_hdr(skb);
1677 		err = 0;
1678 
1679 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1680 			return 0;
1681 
1682 		/* Only requests are handled by the kernel */
1683 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1684 			goto ack;
1685 
1686 		/* Skip control messages */
1687 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1688 			goto ack;
1689 
1690 		err = cb(skb, nlh);
1691 		if (err == -EINTR)
1692 			goto skip;
1693 
1694 ack:
1695 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1696 			netlink_ack(skb, nlh, err);
1697 
1698 skip:
1699 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1700 		if (msglen > skb->len)
1701 			msglen = skb->len;
1702 		skb_pull(skb, msglen);
1703 	}
1704 
1705 	return 0;
1706 }
1707 EXPORT_SYMBOL(netlink_rcv_skb);
1708 
1709 /**
1710  * nlmsg_notify - send a notification netlink message
1711  * @sk: netlink socket to use
1712  * @skb: notification message
1713  * @pid: destination netlink pid for reports or 0
1714  * @group: destination multicast group or 0
1715  * @report: 1 to report back, 0 to disable
1716  * @flags: allocation flags
1717  */
1718 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1719 		 unsigned int group, int report, gfp_t flags)
1720 {
1721 	int err = 0;
1722 
1723 	if (group) {
1724 		int exclude_pid = 0;
1725 
1726 		if (report) {
1727 			atomic_inc(&skb->users);
1728 			exclude_pid = pid;
1729 		}
1730 
1731 		/* errors reported via destination sk->sk_err */
1732 		nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1733 	}
1734 
1735 	if (report)
1736 		err = nlmsg_unicast(sk, skb, pid);
1737 
1738 	return err;
1739 }
1740 EXPORT_SYMBOL(nlmsg_notify);
1741 
1742 #ifdef CONFIG_PROC_FS
1743 struct nl_seq_iter {
1744 	struct seq_net_private p;
1745 	int link;
1746 	int hash_idx;
1747 };
1748 
1749 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1750 {
1751 	struct nl_seq_iter *iter = seq->private;
1752 	int i, j;
1753 	struct sock *s;
1754 	struct hlist_node *node;
1755 	loff_t off = 0;
1756 
1757 	for (i = 0; i < MAX_LINKS; i++) {
1758 		struct nl_pid_hash *hash = &nl_table[i].hash;
1759 
1760 		for (j = 0; j <= hash->mask; j++) {
1761 			sk_for_each(s, node, &hash->table[j]) {
1762 				if (sock_net(s) != seq_file_net(seq))
1763 					continue;
1764 				if (off == pos) {
1765 					iter->link = i;
1766 					iter->hash_idx = j;
1767 					return s;
1768 				}
1769 				++off;
1770 			}
1771 		}
1772 	}
1773 	return NULL;
1774 }
1775 
1776 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1777 	__acquires(nl_table_lock)
1778 {
1779 	read_lock(&nl_table_lock);
1780 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1781 }
1782 
1783 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1784 {
1785 	struct sock *s;
1786 	struct nl_seq_iter *iter;
1787 	int i, j;
1788 
1789 	++*pos;
1790 
1791 	if (v == SEQ_START_TOKEN)
1792 		return netlink_seq_socket_idx(seq, 0);
1793 
1794 	iter = seq->private;
1795 	s = v;
1796 	do {
1797 		s = sk_next(s);
1798 	} while (s && sock_net(s) != seq_file_net(seq));
1799 	if (s)
1800 		return s;
1801 
1802 	i = iter->link;
1803 	j = iter->hash_idx + 1;
1804 
1805 	do {
1806 		struct nl_pid_hash *hash = &nl_table[i].hash;
1807 
1808 		for (; j <= hash->mask; j++) {
1809 			s = sk_head(&hash->table[j]);
1810 			while (s && sock_net(s) != seq_file_net(seq))
1811 				s = sk_next(s);
1812 			if (s) {
1813 				iter->link = i;
1814 				iter->hash_idx = j;
1815 				return s;
1816 			}
1817 		}
1818 
1819 		j = 0;
1820 	} while (++i < MAX_LINKS);
1821 
1822 	return NULL;
1823 }
1824 
1825 static void netlink_seq_stop(struct seq_file *seq, void *v)
1826 	__releases(nl_table_lock)
1827 {
1828 	read_unlock(&nl_table_lock);
1829 }
1830 
1831 
1832 static int netlink_seq_show(struct seq_file *seq, void *v)
1833 {
1834 	if (v == SEQ_START_TOKEN)
1835 		seq_puts(seq,
1836 			 "sk       Eth Pid    Groups   "
1837 			 "Rmem     Wmem     Dump     Locks\n");
1838 	else {
1839 		struct sock *s = v;
1840 		struct netlink_sock *nlk = nlk_sk(s);
1841 
1842 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1843 			   s,
1844 			   s->sk_protocol,
1845 			   nlk->pid,
1846 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1847 			   atomic_read(&s->sk_rmem_alloc),
1848 			   atomic_read(&s->sk_wmem_alloc),
1849 			   nlk->cb,
1850 			   atomic_read(&s->sk_refcnt)
1851 			);
1852 
1853 	}
1854 	return 0;
1855 }
1856 
1857 static const struct seq_operations netlink_seq_ops = {
1858 	.start  = netlink_seq_start,
1859 	.next   = netlink_seq_next,
1860 	.stop   = netlink_seq_stop,
1861 	.show   = netlink_seq_show,
1862 };
1863 
1864 
1865 static int netlink_seq_open(struct inode *inode, struct file *file)
1866 {
1867 	return seq_open_net(inode, file, &netlink_seq_ops,
1868 				sizeof(struct nl_seq_iter));
1869 }
1870 
1871 static const struct file_operations netlink_seq_fops = {
1872 	.owner		= THIS_MODULE,
1873 	.open		= netlink_seq_open,
1874 	.read		= seq_read,
1875 	.llseek		= seq_lseek,
1876 	.release	= seq_release_net,
1877 };
1878 
1879 #endif
1880 
1881 int netlink_register_notifier(struct notifier_block *nb)
1882 {
1883 	return atomic_notifier_chain_register(&netlink_chain, nb);
1884 }
1885 EXPORT_SYMBOL(netlink_register_notifier);
1886 
1887 int netlink_unregister_notifier(struct notifier_block *nb)
1888 {
1889 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
1890 }
1891 EXPORT_SYMBOL(netlink_unregister_notifier);
1892 
1893 static const struct proto_ops netlink_ops = {
1894 	.family =	PF_NETLINK,
1895 	.owner =	THIS_MODULE,
1896 	.release =	netlink_release,
1897 	.bind =		netlink_bind,
1898 	.connect =	netlink_connect,
1899 	.socketpair =	sock_no_socketpair,
1900 	.accept =	sock_no_accept,
1901 	.getname =	netlink_getname,
1902 	.poll =		datagram_poll,
1903 	.ioctl =	sock_no_ioctl,
1904 	.listen =	sock_no_listen,
1905 	.shutdown =	sock_no_shutdown,
1906 	.setsockopt =	netlink_setsockopt,
1907 	.getsockopt =	netlink_getsockopt,
1908 	.sendmsg =	netlink_sendmsg,
1909 	.recvmsg =	netlink_recvmsg,
1910 	.mmap =		sock_no_mmap,
1911 	.sendpage =	sock_no_sendpage,
1912 };
1913 
1914 static struct net_proto_family netlink_family_ops = {
1915 	.family = PF_NETLINK,
1916 	.create = netlink_create,
1917 	.owner	= THIS_MODULE,	/* for consistency 8) */
1918 };
1919 
1920 static int __net_init netlink_net_init(struct net *net)
1921 {
1922 #ifdef CONFIG_PROC_FS
1923 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1924 		return -ENOMEM;
1925 #endif
1926 	return 0;
1927 }
1928 
1929 static void __net_exit netlink_net_exit(struct net *net)
1930 {
1931 #ifdef CONFIG_PROC_FS
1932 	proc_net_remove(net, "netlink");
1933 #endif
1934 }
1935 
1936 static struct pernet_operations __net_initdata netlink_net_ops = {
1937 	.init = netlink_net_init,
1938 	.exit = netlink_net_exit,
1939 };
1940 
1941 static int __init netlink_proto_init(void)
1942 {
1943 	struct sk_buff *dummy_skb;
1944 	int i;
1945 	unsigned long limit;
1946 	unsigned int order;
1947 	int err = proto_register(&netlink_proto, 0);
1948 
1949 	if (err != 0)
1950 		goto out;
1951 
1952 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1953 
1954 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1955 	if (!nl_table)
1956 		goto panic;
1957 
1958 	if (num_physpages >= (128 * 1024))
1959 		limit = num_physpages >> (21 - PAGE_SHIFT);
1960 	else
1961 		limit = num_physpages >> (23 - PAGE_SHIFT);
1962 
1963 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1964 	limit = (1UL << order) / sizeof(struct hlist_head);
1965 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1966 
1967 	for (i = 0; i < MAX_LINKS; i++) {
1968 		struct nl_pid_hash *hash = &nl_table[i].hash;
1969 
1970 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1971 		if (!hash->table) {
1972 			while (i-- > 0)
1973 				nl_pid_hash_free(nl_table[i].hash.table,
1974 						 1 * sizeof(*hash->table));
1975 			kfree(nl_table);
1976 			goto panic;
1977 		}
1978 		hash->max_shift = order;
1979 		hash->shift = 0;
1980 		hash->mask = 0;
1981 		hash->rehash_time = jiffies;
1982 	}
1983 
1984 	sock_register(&netlink_family_ops);
1985 	register_pernet_subsys(&netlink_net_ops);
1986 	/* The netlink device handler may be needed early. */
1987 	rtnetlink_init();
1988 out:
1989 	return err;
1990 panic:
1991 	panic("netlink_init: Cannot allocate nl_table\n");
1992 }
1993 
1994 core_initcall(netlink_proto_init);
1995