• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
nlk_sk(struct sock * sk)89 static inline struct netlink_sock *nlk_sk(struct sock *sk)
90 {
91 	return container_of(sk, struct netlink_sock, sk);
92 }
93 
netlink_is_kernel(struct sock * sk)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 
netlink_group_mask(u32 group)135 static u32 netlink_group_mask(u32 group)
136 {
137 	return group ? 1 << (group - 1) : 0;
138 }
139 
nl_pid_hashfn(struct nl_pid_hash * hash,u32 pid)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 
netlink_sock_destruct(struct sock * sk)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 
netlink_table_grab(void)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 
netlink_table_ungrab(void)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
netlink_lock_table(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
netlink_unlock_table(void)214 netlink_unlock_table(void)
215 {
216 	if (atomic_dec_and_test(&nl_table_users))
217 		wake_up(&nl_table_wait);
218 }
219 
netlink_lookup(struct net * net,int protocol,u32 pid)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 
nl_pid_hash_zalloc(size_t size)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 
nl_pid_hash_free(struct hlist_head * table,size_t size)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 
nl_pid_hash_rehash(struct nl_pid_hash * hash,int grow)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 
nl_pid_hash_dilute(struct nl_pid_hash * hash,int len)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
netlink_update_listeners(struct sock * sk)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 
netlink_insert(struct sock * sk,struct net * net,u32 pid)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 
netlink_remove(struct sock * sk)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 
__netlink_create(struct net * net,struct socket * sock,struct mutex * cb_mutex,int protocol)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 
netlink_create(struct net * net,struct socket * sock,int protocol)422 static int netlink_create(struct net *net, struct socket *sock, int protocol)
423 {
424 	struct module *module = NULL;
425 	struct mutex *cb_mutex;
426 	struct netlink_sock *nlk;
427 	int err = 0;
428 
429 	sock->state = SS_UNCONNECTED;
430 
431 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
432 		return -ESOCKTNOSUPPORT;
433 
434 	if (protocol < 0 || protocol >= MAX_LINKS)
435 		return -EPROTONOSUPPORT;
436 
437 	netlink_lock_table();
438 #ifdef CONFIG_MODULES
439 	if (!nl_table[protocol].registered) {
440 		netlink_unlock_table();
441 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
442 		netlink_lock_table();
443 	}
444 #endif
445 	if (nl_table[protocol].registered &&
446 	    try_module_get(nl_table[protocol].module))
447 		module = nl_table[protocol].module;
448 	cb_mutex = nl_table[protocol].cb_mutex;
449 	netlink_unlock_table();
450 
451 	err = __netlink_create(net, sock, cb_mutex, protocol);
452 	if (err < 0)
453 		goto out_module;
454 
455 	local_bh_disable();
456 	sock_prot_inuse_add(net, &netlink_proto, 1);
457 	local_bh_enable();
458 
459 	nlk = nlk_sk(sock->sk);
460 	nlk->module = module;
461 out:
462 	return err;
463 
464 out_module:
465 	module_put(module);
466 	goto out;
467 }
468 
netlink_release(struct socket * sock)469 static int netlink_release(struct socket *sock)
470 {
471 	struct sock *sk = sock->sk;
472 	struct netlink_sock *nlk;
473 
474 	if (!sk)
475 		return 0;
476 
477 	netlink_remove(sk);
478 	sock_orphan(sk);
479 	nlk = nlk_sk(sk);
480 
481 	/*
482 	 * OK. Socket is unlinked, any packets that arrive now
483 	 * will be purged.
484 	 */
485 
486 	sock->sk = NULL;
487 	wake_up_interruptible_all(&nlk->wait);
488 
489 	skb_queue_purge(&sk->sk_write_queue);
490 
491 	if (nlk->pid && !nlk->subscriptions) {
492 		struct netlink_notify n = {
493 						.net = sock_net(sk),
494 						.protocol = sk->sk_protocol,
495 						.pid = nlk->pid,
496 					  };
497 		atomic_notifier_call_chain(&netlink_chain,
498 				NETLINK_URELEASE, &n);
499 	}
500 
501 	module_put(nlk->module);
502 
503 	netlink_table_grab();
504 	if (netlink_is_kernel(sk)) {
505 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
506 		if (--nl_table[sk->sk_protocol].registered == 0) {
507 			kfree(nl_table[sk->sk_protocol].listeners);
508 			nl_table[sk->sk_protocol].module = NULL;
509 			nl_table[sk->sk_protocol].registered = 0;
510 		}
511 	} else if (nlk->subscriptions)
512 		netlink_update_listeners(sk);
513 	netlink_table_ungrab();
514 
515 	kfree(nlk->groups);
516 	nlk->groups = NULL;
517 
518 	local_bh_disable();
519 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
520 	local_bh_enable();
521 	sock_put(sk);
522 	return 0;
523 }
524 
netlink_autobind(struct socket * sock)525 static int netlink_autobind(struct socket *sock)
526 {
527 	struct sock *sk = sock->sk;
528 	struct net *net = sock_net(sk);
529 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
530 	struct hlist_head *head;
531 	struct sock *osk;
532 	struct hlist_node *node;
533 	s32 pid = current->tgid;
534 	int err;
535 	static s32 rover = -4097;
536 
537 retry:
538 	cond_resched();
539 	netlink_table_grab();
540 	head = nl_pid_hashfn(hash, pid);
541 	sk_for_each(osk, node, head) {
542 		if (!net_eq(sock_net(osk), net))
543 			continue;
544 		if (nlk_sk(osk)->pid == pid) {
545 			/* Bind collision, search negative pid values. */
546 			pid = rover--;
547 			if (rover > -4097)
548 				rover = -4097;
549 			netlink_table_ungrab();
550 			goto retry;
551 		}
552 	}
553 	netlink_table_ungrab();
554 
555 	err = netlink_insert(sk, net, pid);
556 	if (err == -EADDRINUSE)
557 		goto retry;
558 
559 	/* If 2 threads race to autobind, that is fine.  */
560 	if (err == -EBUSY)
561 		err = 0;
562 
563 	return err;
564 }
565 
netlink_capable(struct socket * sock,unsigned int flag)566 static inline int netlink_capable(struct socket *sock, unsigned int flag)
567 {
568 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
569 	       capable(CAP_NET_ADMIN);
570 }
571 
572 static void
netlink_update_subscriptions(struct sock * sk,unsigned int subscriptions)573 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
574 {
575 	struct netlink_sock *nlk = nlk_sk(sk);
576 
577 	if (nlk->subscriptions && !subscriptions)
578 		__sk_del_bind_node(sk);
579 	else if (!nlk->subscriptions && subscriptions)
580 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
581 	nlk->subscriptions = subscriptions;
582 }
583 
netlink_realloc_groups(struct sock * sk)584 static int netlink_realloc_groups(struct sock *sk)
585 {
586 	struct netlink_sock *nlk = nlk_sk(sk);
587 	unsigned int groups;
588 	unsigned long *new_groups;
589 	int err = 0;
590 
591 	netlink_table_grab();
592 
593 	groups = nl_table[sk->sk_protocol].groups;
594 	if (!nl_table[sk->sk_protocol].registered) {
595 		err = -ENOENT;
596 		goto out_unlock;
597 	}
598 
599 	if (nlk->ngroups >= groups)
600 		goto out_unlock;
601 
602 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
603 	if (new_groups == NULL) {
604 		err = -ENOMEM;
605 		goto out_unlock;
606 	}
607 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
608 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
609 
610 	nlk->groups = new_groups;
611 	nlk->ngroups = groups;
612  out_unlock:
613 	netlink_table_ungrab();
614 	return err;
615 }
616 
netlink_bind(struct socket * sock,struct sockaddr * addr,int addr_len)617 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
618 			int addr_len)
619 {
620 	struct sock *sk = sock->sk;
621 	struct net *net = sock_net(sk);
622 	struct netlink_sock *nlk = nlk_sk(sk);
623 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
624 	int err;
625 
626 	if (nladdr->nl_family != AF_NETLINK)
627 		return -EINVAL;
628 
629 	/* Only superuser is allowed to listen multicasts */
630 	if (nladdr->nl_groups) {
631 		if (!netlink_capable(sock, NL_NONROOT_RECV))
632 			return -EPERM;
633 		err = netlink_realloc_groups(sk);
634 		if (err)
635 			return err;
636 	}
637 
638 	if (nlk->pid) {
639 		if (nladdr->nl_pid != nlk->pid)
640 			return -EINVAL;
641 	} else {
642 		err = nladdr->nl_pid ?
643 			netlink_insert(sk, net, nladdr->nl_pid) :
644 			netlink_autobind(sock);
645 		if (err)
646 			return err;
647 	}
648 
649 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
650 		return 0;
651 
652 	netlink_table_grab();
653 	netlink_update_subscriptions(sk, nlk->subscriptions +
654 					 hweight32(nladdr->nl_groups) -
655 					 hweight32(nlk->groups[0]));
656 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
657 	netlink_update_listeners(sk);
658 	netlink_table_ungrab();
659 
660 	return 0;
661 }
662 
netlink_connect(struct socket * sock,struct sockaddr * addr,int alen,int flags)663 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
664 			   int alen, int flags)
665 {
666 	int err = 0;
667 	struct sock *sk = sock->sk;
668 	struct netlink_sock *nlk = nlk_sk(sk);
669 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
670 
671 	if (addr->sa_family == AF_UNSPEC) {
672 		sk->sk_state	= NETLINK_UNCONNECTED;
673 		nlk->dst_pid	= 0;
674 		nlk->dst_group  = 0;
675 		return 0;
676 	}
677 	if (addr->sa_family != AF_NETLINK)
678 		return -EINVAL;
679 
680 	/* Only superuser is allowed to send multicasts */
681 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
682 		return -EPERM;
683 
684 	if (!nlk->pid)
685 		err = netlink_autobind(sock);
686 
687 	if (err == 0) {
688 		sk->sk_state	= NETLINK_CONNECTED;
689 		nlk->dst_pid 	= nladdr->nl_pid;
690 		nlk->dst_group  = ffs(nladdr->nl_groups);
691 	}
692 
693 	return err;
694 }
695 
netlink_getname(struct socket * sock,struct sockaddr * addr,int * addr_len,int peer)696 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
697 			   int *addr_len, int peer)
698 {
699 	struct sock *sk = sock->sk;
700 	struct netlink_sock *nlk = nlk_sk(sk);
701 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
702 
703 	nladdr->nl_family = AF_NETLINK;
704 	nladdr->nl_pad = 0;
705 	*addr_len = sizeof(*nladdr);
706 
707 	if (peer) {
708 		nladdr->nl_pid = nlk->dst_pid;
709 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
710 	} else {
711 		nladdr->nl_pid = nlk->pid;
712 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
713 	}
714 	return 0;
715 }
716 
netlink_overrun(struct sock * sk)717 static void netlink_overrun(struct sock *sk)
718 {
719 	if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
720 		sk->sk_err = ENOBUFS;
721 		sk->sk_error_report(sk);
722 	}
723 }
724 
netlink_getsockbypid(struct sock * ssk,u32 pid)725 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
726 {
727 	struct sock *sock;
728 	struct netlink_sock *nlk;
729 
730 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
731 	if (!sock)
732 		return ERR_PTR(-ECONNREFUSED);
733 
734 	/* Don't bother queuing skb if kernel socket has no input function */
735 	nlk = nlk_sk(sock);
736 	if (sock->sk_state == NETLINK_CONNECTED &&
737 	    nlk->dst_pid != nlk_sk(ssk)->pid) {
738 		sock_put(sock);
739 		return ERR_PTR(-ECONNREFUSED);
740 	}
741 	return sock;
742 }
743 
netlink_getsockbyfilp(struct file * filp)744 struct sock *netlink_getsockbyfilp(struct file *filp)
745 {
746 	struct inode *inode = filp->f_path.dentry->d_inode;
747 	struct sock *sock;
748 
749 	if (!S_ISSOCK(inode->i_mode))
750 		return ERR_PTR(-ENOTSOCK);
751 
752 	sock = SOCKET_I(inode)->sk;
753 	if (sock->sk_family != AF_NETLINK)
754 		return ERR_PTR(-EINVAL);
755 
756 	sock_hold(sock);
757 	return sock;
758 }
759 
760 /*
761  * Attach a skb to a netlink socket.
762  * The caller must hold a reference to the destination socket. On error, the
763  * reference is dropped. The skb is not send to the destination, just all
764  * all error checks are performed and memory in the queue is reserved.
765  * Return values:
766  * < 0: error. skb freed, reference to sock dropped.
767  * 0: continue
768  * 1: repeat lookup - reference dropped while waiting for socket memory.
769  */
netlink_attachskb(struct sock * sk,struct sk_buff * skb,long * timeo,struct sock * ssk)770 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
771 		      long *timeo, struct sock *ssk)
772 {
773 	struct netlink_sock *nlk;
774 
775 	nlk = nlk_sk(sk);
776 
777 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
778 	    test_bit(0, &nlk->state)) {
779 		DECLARE_WAITQUEUE(wait, current);
780 		if (!*timeo) {
781 			if (!ssk || netlink_is_kernel(ssk))
782 				netlink_overrun(sk);
783 			sock_put(sk);
784 			kfree_skb(skb);
785 			return -EAGAIN;
786 		}
787 
788 		__set_current_state(TASK_INTERRUPTIBLE);
789 		add_wait_queue(&nlk->wait, &wait);
790 
791 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
792 		     test_bit(0, &nlk->state)) &&
793 		    !sock_flag(sk, SOCK_DEAD))
794 			*timeo = schedule_timeout(*timeo);
795 
796 		__set_current_state(TASK_RUNNING);
797 		remove_wait_queue(&nlk->wait, &wait);
798 		sock_put(sk);
799 
800 		if (signal_pending(current)) {
801 			kfree_skb(skb);
802 			return sock_intr_errno(*timeo);
803 		}
804 		return 1;
805 	}
806 	skb_set_owner_r(skb, sk);
807 	return 0;
808 }
809 
netlink_sendskb(struct sock * sk,struct sk_buff * skb)810 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
811 {
812 	int len = skb->len;
813 
814 	skb_queue_tail(&sk->sk_receive_queue, skb);
815 	sk->sk_data_ready(sk, len);
816 	sock_put(sk);
817 	return len;
818 }
819 
netlink_detachskb(struct sock * sk,struct sk_buff * skb)820 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
821 {
822 	kfree_skb(skb);
823 	sock_put(sk);
824 }
825 
netlink_trim(struct sk_buff * skb,gfp_t allocation)826 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
827 					   gfp_t allocation)
828 {
829 	int delta;
830 
831 	skb_orphan(skb);
832 
833 	delta = skb->end - skb->tail;
834 	if (delta * 2 < skb->truesize)
835 		return skb;
836 
837 	if (skb_shared(skb)) {
838 		struct sk_buff *nskb = skb_clone(skb, allocation);
839 		if (!nskb)
840 			return skb;
841 		kfree_skb(skb);
842 		skb = nskb;
843 	}
844 
845 	if (!pskb_expand_head(skb, 0, -delta, allocation))
846 		skb->truesize -= delta;
847 
848 	return skb;
849 }
850 
netlink_rcv_wake(struct sock * sk)851 static inline void netlink_rcv_wake(struct sock *sk)
852 {
853 	struct netlink_sock *nlk = nlk_sk(sk);
854 
855 	if (skb_queue_empty(&sk->sk_receive_queue))
856 		clear_bit(0, &nlk->state);
857 	if (!test_bit(0, &nlk->state))
858 		wake_up_interruptible(&nlk->wait);
859 }
860 
netlink_unicast_kernel(struct sock * sk,struct sk_buff * skb)861 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
862 {
863 	int ret;
864 	struct netlink_sock *nlk = nlk_sk(sk);
865 
866 	ret = -ECONNREFUSED;
867 	if (nlk->netlink_rcv != NULL) {
868 		ret = skb->len;
869 		skb_set_owner_r(skb, sk);
870 		nlk->netlink_rcv(skb);
871 	}
872 	kfree_skb(skb);
873 	sock_put(sk);
874 	return ret;
875 }
876 
netlink_unicast(struct sock * ssk,struct sk_buff * skb,u32 pid,int nonblock)877 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
878 		    u32 pid, int nonblock)
879 {
880 	struct sock *sk;
881 	int err;
882 	long timeo;
883 
884 	skb = netlink_trim(skb, gfp_any());
885 
886 	timeo = sock_sndtimeo(ssk, nonblock);
887 retry:
888 	sk = netlink_getsockbypid(ssk, pid);
889 	if (IS_ERR(sk)) {
890 		kfree_skb(skb);
891 		return PTR_ERR(sk);
892 	}
893 	if (netlink_is_kernel(sk))
894 		return netlink_unicast_kernel(sk, skb);
895 
896 	if (sk_filter(sk, skb)) {
897 		err = skb->len;
898 		kfree_skb(skb);
899 		sock_put(sk);
900 		return err;
901 	}
902 
903 	err = netlink_attachskb(sk, skb, &timeo, ssk);
904 	if (err == 1)
905 		goto retry;
906 	if (err)
907 		return err;
908 
909 	return netlink_sendskb(sk, skb);
910 }
911 EXPORT_SYMBOL(netlink_unicast);
912 
netlink_has_listeners(struct sock * sk,unsigned int group)913 int netlink_has_listeners(struct sock *sk, unsigned int group)
914 {
915 	int res = 0;
916 	unsigned long *listeners;
917 
918 	BUG_ON(!netlink_is_kernel(sk));
919 
920 	rcu_read_lock();
921 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
922 
923 	if (group - 1 < nl_table[sk->sk_protocol].groups)
924 		res = test_bit(group - 1, listeners);
925 
926 	rcu_read_unlock();
927 
928 	return res;
929 }
930 EXPORT_SYMBOL_GPL(netlink_has_listeners);
931 
netlink_broadcast_deliver(struct sock * sk,struct sk_buff * skb)932 static inline int netlink_broadcast_deliver(struct sock *sk,
933 					    struct sk_buff *skb)
934 {
935 	struct netlink_sock *nlk = nlk_sk(sk);
936 
937 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
938 	    !test_bit(0, &nlk->state)) {
939 		skb_set_owner_r(skb, sk);
940 		skb_queue_tail(&sk->sk_receive_queue, skb);
941 		sk->sk_data_ready(sk, skb->len);
942 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
943 	}
944 	return -1;
945 }
946 
947 struct netlink_broadcast_data {
948 	struct sock *exclude_sk;
949 	struct net *net;
950 	u32 pid;
951 	u32 group;
952 	int failure;
953 	int congested;
954 	int delivered;
955 	gfp_t allocation;
956 	struct sk_buff *skb, *skb2;
957 };
958 
do_one_broadcast(struct sock * sk,struct netlink_broadcast_data * p)959 static inline int do_one_broadcast(struct sock *sk,
960 				   struct netlink_broadcast_data *p)
961 {
962 	struct netlink_sock *nlk = nlk_sk(sk);
963 	int val;
964 
965 	if (p->exclude_sk == sk)
966 		goto out;
967 
968 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
969 	    !test_bit(p->group - 1, nlk->groups))
970 		goto out;
971 
972 	if (!net_eq(sock_net(sk), p->net))
973 		goto out;
974 
975 	if (p->failure) {
976 		netlink_overrun(sk);
977 		goto out;
978 	}
979 
980 	sock_hold(sk);
981 	if (p->skb2 == NULL) {
982 		if (skb_shared(p->skb)) {
983 			p->skb2 = skb_clone(p->skb, p->allocation);
984 		} else {
985 			p->skb2 = skb_get(p->skb);
986 			/*
987 			 * skb ownership may have been set when
988 			 * delivered to a previous socket.
989 			 */
990 			skb_orphan(p->skb2);
991 		}
992 	}
993 	if (p->skb2 == NULL) {
994 		netlink_overrun(sk);
995 		/* Clone failed. Notify ALL listeners. */
996 		p->failure = 1;
997 	} else if (sk_filter(sk, p->skb2)) {
998 		kfree_skb(p->skb2);
999 		p->skb2 = NULL;
1000 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1001 		netlink_overrun(sk);
1002 	} else {
1003 		p->congested |= val;
1004 		p->delivered = 1;
1005 		p->skb2 = NULL;
1006 	}
1007 	sock_put(sk);
1008 
1009 out:
1010 	return 0;
1011 }
1012 
netlink_broadcast(struct sock * ssk,struct sk_buff * skb,u32 pid,u32 group,gfp_t allocation)1013 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1014 		      u32 group, gfp_t allocation)
1015 {
1016 	struct net *net = sock_net(ssk);
1017 	struct netlink_broadcast_data info;
1018 	struct hlist_node *node;
1019 	struct sock *sk;
1020 
1021 	skb = netlink_trim(skb, allocation);
1022 
1023 	info.exclude_sk = ssk;
1024 	info.net = net;
1025 	info.pid = pid;
1026 	info.group = group;
1027 	info.failure = 0;
1028 	info.congested = 0;
1029 	info.delivered = 0;
1030 	info.allocation = allocation;
1031 	info.skb = skb;
1032 	info.skb2 = NULL;
1033 
1034 	/* While we sleep in clone, do not allow to change socket list */
1035 
1036 	netlink_lock_table();
1037 
1038 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1039 		do_one_broadcast(sk, &info);
1040 
1041 	kfree_skb(skb);
1042 
1043 	netlink_unlock_table();
1044 
1045 	if (info.skb2)
1046 		kfree_skb(info.skb2);
1047 
1048 	if (info.delivered) {
1049 		if (info.congested && (allocation & __GFP_WAIT))
1050 			yield();
1051 		return 0;
1052 	}
1053 	if (info.failure)
1054 		return -ENOBUFS;
1055 	return -ESRCH;
1056 }
1057 EXPORT_SYMBOL(netlink_broadcast);
1058 
1059 struct netlink_set_err_data {
1060 	struct sock *exclude_sk;
1061 	u32 pid;
1062 	u32 group;
1063 	int code;
1064 };
1065 
do_one_set_err(struct sock * sk,struct netlink_set_err_data * p)1066 static inline int do_one_set_err(struct sock *sk,
1067 				 struct netlink_set_err_data *p)
1068 {
1069 	struct netlink_sock *nlk = nlk_sk(sk);
1070 
1071 	if (sk == p->exclude_sk)
1072 		goto out;
1073 
1074 	if (sock_net(sk) != sock_net(p->exclude_sk))
1075 		goto out;
1076 
1077 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1078 	    !test_bit(p->group - 1, nlk->groups))
1079 		goto out;
1080 
1081 	sk->sk_err = p->code;
1082 	sk->sk_error_report(sk);
1083 out:
1084 	return 0;
1085 }
1086 
1087 /**
1088  * netlink_set_err - report error to broadcast listeners
1089  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1090  * @pid: the PID of a process that we want to skip (if any)
1091  * @groups: the broadcast group that will notice the error
1092  * @code: error code, must be negative (as usual in kernelspace)
1093  */
netlink_set_err(struct sock * ssk,u32 pid,u32 group,int code)1094 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1095 {
1096 	struct netlink_set_err_data info;
1097 	struct hlist_node *node;
1098 	struct sock *sk;
1099 
1100 	info.exclude_sk = ssk;
1101 	info.pid = pid;
1102 	info.group = group;
1103 	/* sk->sk_err wants a positive error value */
1104 	info.code = -code;
1105 
1106 	read_lock(&nl_table_lock);
1107 
1108 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1109 		do_one_set_err(sk, &info);
1110 
1111 	read_unlock(&nl_table_lock);
1112 }
1113 
1114 /* must be called with netlink table grabbed */
netlink_update_socket_mc(struct netlink_sock * nlk,unsigned int group,int is_new)1115 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1116 				     unsigned int group,
1117 				     int is_new)
1118 {
1119 	int old, new = !!is_new, subscriptions;
1120 
1121 	old = test_bit(group - 1, nlk->groups);
1122 	subscriptions = nlk->subscriptions - old + new;
1123 	if (new)
1124 		__set_bit(group - 1, nlk->groups);
1125 	else
1126 		__clear_bit(group - 1, nlk->groups);
1127 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1128 	netlink_update_listeners(&nlk->sk);
1129 }
1130 
netlink_setsockopt(struct socket * sock,int level,int optname,char __user * optval,int optlen)1131 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1132 			      char __user *optval, int optlen)
1133 {
1134 	struct sock *sk = sock->sk;
1135 	struct netlink_sock *nlk = nlk_sk(sk);
1136 	unsigned int val = 0;
1137 	int err;
1138 
1139 	if (level != SOL_NETLINK)
1140 		return -ENOPROTOOPT;
1141 
1142 	if (optlen >= sizeof(int) &&
1143 	    get_user(val, (unsigned int __user *)optval))
1144 		return -EFAULT;
1145 
1146 	switch (optname) {
1147 	case NETLINK_PKTINFO:
1148 		if (val)
1149 			nlk->flags |= NETLINK_RECV_PKTINFO;
1150 		else
1151 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1152 		err = 0;
1153 		break;
1154 	case NETLINK_ADD_MEMBERSHIP:
1155 	case NETLINK_DROP_MEMBERSHIP: {
1156 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1157 			return -EPERM;
1158 		err = netlink_realloc_groups(sk);
1159 		if (err)
1160 			return err;
1161 		if (!val || val - 1 >= nlk->ngroups)
1162 			return -EINVAL;
1163 		netlink_table_grab();
1164 		netlink_update_socket_mc(nlk, val,
1165 					 optname == NETLINK_ADD_MEMBERSHIP);
1166 		netlink_table_ungrab();
1167 		err = 0;
1168 		break;
1169 	}
1170 	default:
1171 		err = -ENOPROTOOPT;
1172 	}
1173 	return err;
1174 }
1175 
netlink_getsockopt(struct socket * sock,int level,int optname,char __user * optval,int __user * optlen)1176 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1177 			      char __user *optval, int __user *optlen)
1178 {
1179 	struct sock *sk = sock->sk;
1180 	struct netlink_sock *nlk = nlk_sk(sk);
1181 	int len, val, err;
1182 
1183 	if (level != SOL_NETLINK)
1184 		return -ENOPROTOOPT;
1185 
1186 	if (get_user(len, optlen))
1187 		return -EFAULT;
1188 	if (len < 0)
1189 		return -EINVAL;
1190 
1191 	switch (optname) {
1192 	case NETLINK_PKTINFO:
1193 		if (len < sizeof(int))
1194 			return -EINVAL;
1195 		len = sizeof(int);
1196 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1197 		if (put_user(len, optlen) ||
1198 		    put_user(val, optval))
1199 			return -EFAULT;
1200 		err = 0;
1201 		break;
1202 	default:
1203 		err = -ENOPROTOOPT;
1204 	}
1205 	return err;
1206 }
1207 
netlink_cmsg_recv_pktinfo(struct msghdr * msg,struct sk_buff * skb)1208 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1209 {
1210 	struct nl_pktinfo info;
1211 
1212 	info.group = NETLINK_CB(skb).dst_group;
1213 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1214 }
1215 
netlink_sendmsg(struct kiocb * kiocb,struct socket * sock,struct msghdr * msg,size_t len)1216 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1217 			   struct msghdr *msg, size_t len)
1218 {
1219 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1220 	struct sock *sk = sock->sk;
1221 	struct netlink_sock *nlk = nlk_sk(sk);
1222 	struct sockaddr_nl *addr = msg->msg_name;
1223 	u32 dst_pid;
1224 	u32 dst_group;
1225 	struct sk_buff *skb;
1226 	int err;
1227 	struct scm_cookie scm;
1228 
1229 	if (msg->msg_flags&MSG_OOB)
1230 		return -EOPNOTSUPP;
1231 
1232 	if (NULL == siocb->scm)
1233 		siocb->scm = &scm;
1234 	err = scm_send(sock, msg, siocb->scm);
1235 	if (err < 0)
1236 		return err;
1237 
1238 	if (msg->msg_namelen) {
1239 		if (addr->nl_family != AF_NETLINK)
1240 			return -EINVAL;
1241 		dst_pid = addr->nl_pid;
1242 		dst_group = ffs(addr->nl_groups);
1243 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1244 			return -EPERM;
1245 	} else {
1246 		dst_pid = nlk->dst_pid;
1247 		dst_group = nlk->dst_group;
1248 	}
1249 
1250 	if (!nlk->pid) {
1251 		err = netlink_autobind(sock);
1252 		if (err)
1253 			goto out;
1254 	}
1255 
1256 	err = -EMSGSIZE;
1257 	if (len > sk->sk_sndbuf - 32)
1258 		goto out;
1259 	err = -ENOBUFS;
1260 	skb = alloc_skb(len, GFP_KERNEL);
1261 	if (skb == NULL)
1262 		goto out;
1263 
1264 	NETLINK_CB(skb).pid	= nlk->pid;
1265 	NETLINK_CB(skb).dst_group = dst_group;
1266 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1267 	NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1268 	security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1269 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1270 
1271 	/* What can I do? Netlink is asynchronous, so that
1272 	   we will have to save current capabilities to
1273 	   check them, when this message will be delivered
1274 	   to corresponding kernel module.   --ANK (980802)
1275 	 */
1276 
1277 	err = -EFAULT;
1278 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1279 		kfree_skb(skb);
1280 		goto out;
1281 	}
1282 
1283 	err = security_netlink_send(sk, skb);
1284 	if (err) {
1285 		kfree_skb(skb);
1286 		goto out;
1287 	}
1288 
1289 	if (dst_group) {
1290 		atomic_inc(&skb->users);
1291 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1292 	}
1293 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1294 
1295 out:
1296 	return err;
1297 }
1298 
netlink_recvmsg(struct kiocb * kiocb,struct socket * sock,struct msghdr * msg,size_t len,int flags)1299 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1300 			   struct msghdr *msg, size_t len,
1301 			   int flags)
1302 {
1303 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1304 	struct scm_cookie scm;
1305 	struct sock *sk = sock->sk;
1306 	struct netlink_sock *nlk = nlk_sk(sk);
1307 	int noblock = flags&MSG_DONTWAIT;
1308 	size_t copied;
1309 	struct sk_buff *skb;
1310 	int err;
1311 
1312 	if (flags&MSG_OOB)
1313 		return -EOPNOTSUPP;
1314 
1315 	copied = 0;
1316 
1317 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1318 	if (skb == NULL)
1319 		goto out;
1320 
1321 	msg->msg_namelen = 0;
1322 
1323 	copied = skb->len;
1324 	if (len < copied) {
1325 		msg->msg_flags |= MSG_TRUNC;
1326 		copied = len;
1327 	}
1328 
1329 	skb_reset_transport_header(skb);
1330 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1331 
1332 	if (msg->msg_name) {
1333 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1334 		addr->nl_family = AF_NETLINK;
1335 		addr->nl_pad    = 0;
1336 		addr->nl_pid	= NETLINK_CB(skb).pid;
1337 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1338 		msg->msg_namelen = sizeof(*addr);
1339 	}
1340 
1341 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1342 		netlink_cmsg_recv_pktinfo(msg, skb);
1343 
1344 	if (NULL == siocb->scm) {
1345 		memset(&scm, 0, sizeof(scm));
1346 		siocb->scm = &scm;
1347 	}
1348 	siocb->scm->creds = *NETLINK_CREDS(skb);
1349 	if (flags & MSG_TRUNC)
1350 		copied = skb->len;
1351 	skb_free_datagram(sk, skb);
1352 
1353 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1354 		netlink_dump(sk);
1355 
1356 	scm_recv(sock, msg, siocb->scm, flags);
1357 out:
1358 	netlink_rcv_wake(sk);
1359 	return err ? : copied;
1360 }
1361 
netlink_data_ready(struct sock * sk,int len)1362 static void netlink_data_ready(struct sock *sk, int len)
1363 {
1364 	BUG();
1365 }
1366 
1367 /*
1368  *	We export these functions to other modules. They provide a
1369  *	complete set of kernel non-blocking support for message
1370  *	queueing.
1371  */
1372 
1373 struct sock *
netlink_kernel_create(struct net * net,int unit,unsigned int groups,void (* input)(struct sk_buff * skb),struct mutex * cb_mutex,struct module * module)1374 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1375 		      void (*input)(struct sk_buff *skb),
1376 		      struct mutex *cb_mutex, struct module *module)
1377 {
1378 	struct socket *sock;
1379 	struct sock *sk;
1380 	struct netlink_sock *nlk;
1381 	unsigned long *listeners = NULL;
1382 
1383 	BUG_ON(!nl_table);
1384 
1385 	if (unit < 0 || unit >= MAX_LINKS)
1386 		return NULL;
1387 
1388 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1389 		return NULL;
1390 
1391 	/*
1392 	 * We have to just have a reference on the net from sk, but don't
1393 	 * get_net it. Besides, we cannot get and then put the net here.
1394 	 * So we create one inside init_net and the move it to net.
1395 	 */
1396 
1397 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1398 		goto out_sock_release_nosk;
1399 
1400 	sk = sock->sk;
1401 	sk_change_net(sk, net);
1402 
1403 	if (groups < 32)
1404 		groups = 32;
1405 
1406 	listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1407 	if (!listeners)
1408 		goto out_sock_release;
1409 
1410 	sk->sk_data_ready = netlink_data_ready;
1411 	if (input)
1412 		nlk_sk(sk)->netlink_rcv = input;
1413 
1414 	if (netlink_insert(sk, net, 0))
1415 		goto out_sock_release;
1416 
1417 	nlk = nlk_sk(sk);
1418 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1419 
1420 	netlink_table_grab();
1421 	if (!nl_table[unit].registered) {
1422 		nl_table[unit].groups = groups;
1423 		nl_table[unit].listeners = listeners;
1424 		nl_table[unit].cb_mutex = cb_mutex;
1425 		nl_table[unit].module = module;
1426 		nl_table[unit].registered = 1;
1427 	} else {
1428 		kfree(listeners);
1429 		nl_table[unit].registered++;
1430 	}
1431 	netlink_table_ungrab();
1432 	return sk;
1433 
1434 out_sock_release:
1435 	kfree(listeners);
1436 	netlink_kernel_release(sk);
1437 	return NULL;
1438 
1439 out_sock_release_nosk:
1440 	sock_release(sock);
1441 	return NULL;
1442 }
1443 EXPORT_SYMBOL(netlink_kernel_create);
1444 
1445 
1446 void
netlink_kernel_release(struct sock * sk)1447 netlink_kernel_release(struct sock *sk)
1448 {
1449 	sk_release_kernel(sk);
1450 }
1451 EXPORT_SYMBOL(netlink_kernel_release);
1452 
1453 
1454 /**
1455  * netlink_change_ngroups - change number of multicast groups
1456  *
1457  * This changes the number of multicast groups that are available
1458  * on a certain netlink family. Note that it is not possible to
1459  * change the number of groups to below 32. Also note that it does
1460  * not implicitly call netlink_clear_multicast_users() when the
1461  * number of groups is reduced.
1462  *
1463  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1464  * @groups: The new number of groups.
1465  */
netlink_change_ngroups(struct sock * sk,unsigned int groups)1466 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1467 {
1468 	unsigned long *listeners, *old = NULL;
1469 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1470 	int err = 0;
1471 
1472 	if (groups < 32)
1473 		groups = 32;
1474 
1475 	netlink_table_grab();
1476 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1477 		listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1478 		if (!listeners) {
1479 			err = -ENOMEM;
1480 			goto out_ungrab;
1481 		}
1482 		old = tbl->listeners;
1483 		memcpy(listeners, old, NLGRPSZ(tbl->groups));
1484 		rcu_assign_pointer(tbl->listeners, listeners);
1485 	}
1486 	tbl->groups = groups;
1487 
1488  out_ungrab:
1489 	netlink_table_ungrab();
1490 	synchronize_rcu();
1491 	kfree(old);
1492 	return err;
1493 }
1494 EXPORT_SYMBOL(netlink_change_ngroups);
1495 
1496 /**
1497  * netlink_clear_multicast_users - kick off multicast listeners
1498  *
1499  * This function removes all listeners from the given group.
1500  * @ksk: The kernel netlink socket, as returned by
1501  *	netlink_kernel_create().
1502  * @group: The multicast group to clear.
1503  */
netlink_clear_multicast_users(struct sock * ksk,unsigned int group)1504 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1505 {
1506 	struct sock *sk;
1507 	struct hlist_node *node;
1508 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1509 
1510 	netlink_table_grab();
1511 
1512 	sk_for_each_bound(sk, node, &tbl->mc_list)
1513 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1514 
1515 	netlink_table_ungrab();
1516 }
1517 EXPORT_SYMBOL(netlink_clear_multicast_users);
1518 
netlink_set_nonroot(int protocol,unsigned int flags)1519 void netlink_set_nonroot(int protocol, unsigned int flags)
1520 {
1521 	if ((unsigned int)protocol < MAX_LINKS)
1522 		nl_table[protocol].nl_nonroot = flags;
1523 }
1524 EXPORT_SYMBOL(netlink_set_nonroot);
1525 
netlink_destroy_callback(struct netlink_callback * cb)1526 static void netlink_destroy_callback(struct netlink_callback *cb)
1527 {
1528 	if (cb->skb)
1529 		kfree_skb(cb->skb);
1530 	kfree(cb);
1531 }
1532 
1533 /*
1534  * It looks a bit ugly.
1535  * It would be better to create kernel thread.
1536  */
1537 
netlink_dump(struct sock * sk)1538 static int netlink_dump(struct sock *sk)
1539 {
1540 	struct netlink_sock *nlk = nlk_sk(sk);
1541 	struct netlink_callback *cb;
1542 	struct sk_buff *skb;
1543 	struct nlmsghdr *nlh;
1544 	int len, err = -ENOBUFS;
1545 
1546 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1547 	if (!skb)
1548 		goto errout;
1549 
1550 	mutex_lock(nlk->cb_mutex);
1551 
1552 	cb = nlk->cb;
1553 	if (cb == NULL) {
1554 		err = -EINVAL;
1555 		goto errout_skb;
1556 	}
1557 
1558 	len = cb->dump(skb, cb);
1559 
1560 	if (len > 0) {
1561 		mutex_unlock(nlk->cb_mutex);
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 		return 0;
1570 	}
1571 
1572 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1573 	if (!nlh)
1574 		goto errout_skb;
1575 
1576 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1577 
1578 	if (sk_filter(sk, skb))
1579 		kfree_skb(skb);
1580 	else {
1581 		skb_queue_tail(&sk->sk_receive_queue, skb);
1582 		sk->sk_data_ready(sk, skb->len);
1583 	}
1584 
1585 	if (cb->done)
1586 		cb->done(cb);
1587 	nlk->cb = NULL;
1588 	mutex_unlock(nlk->cb_mutex);
1589 
1590 	netlink_destroy_callback(cb);
1591 	return 0;
1592 
1593 errout_skb:
1594 	mutex_unlock(nlk->cb_mutex);
1595 	kfree_skb(skb);
1596 errout:
1597 	return err;
1598 }
1599 
netlink_dump_start(struct sock * ssk,struct sk_buff * skb,struct nlmsghdr * nlh,int (* dump)(struct sk_buff * skb,struct netlink_callback *),int (* done)(struct netlink_callback *))1600 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1601 		       struct nlmsghdr *nlh,
1602 		       int (*dump)(struct sk_buff *skb,
1603 				   struct netlink_callback *),
1604 		       int (*done)(struct netlink_callback *))
1605 {
1606 	struct netlink_callback *cb;
1607 	struct sock *sk;
1608 	struct netlink_sock *nlk;
1609 
1610 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1611 	if (cb == NULL)
1612 		return -ENOBUFS;
1613 
1614 	cb->dump = dump;
1615 	cb->done = done;
1616 	cb->nlh = nlh;
1617 	atomic_inc(&skb->users);
1618 	cb->skb = skb;
1619 
1620 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1621 	if (sk == NULL) {
1622 		netlink_destroy_callback(cb);
1623 		return -ECONNREFUSED;
1624 	}
1625 	nlk = nlk_sk(sk);
1626 	/* A dump is in progress... */
1627 	mutex_lock(nlk->cb_mutex);
1628 	if (nlk->cb) {
1629 		mutex_unlock(nlk->cb_mutex);
1630 		netlink_destroy_callback(cb);
1631 		sock_put(sk);
1632 		return -EBUSY;
1633 	}
1634 	nlk->cb = cb;
1635 	mutex_unlock(nlk->cb_mutex);
1636 
1637 	netlink_dump(sk);
1638 	sock_put(sk);
1639 
1640 	/* We successfully started a dump, by returning -EINTR we
1641 	 * signal not to send ACK even if it was requested.
1642 	 */
1643 	return -EINTR;
1644 }
1645 EXPORT_SYMBOL(netlink_dump_start);
1646 
netlink_ack(struct sk_buff * in_skb,struct nlmsghdr * nlh,int err)1647 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1648 {
1649 	struct sk_buff *skb;
1650 	struct nlmsghdr *rep;
1651 	struct nlmsgerr *errmsg;
1652 	size_t payload = sizeof(*errmsg);
1653 
1654 	/* error messages get the original request appened */
1655 	if (err)
1656 		payload += nlmsg_len(nlh);
1657 
1658 	skb = nlmsg_new(payload, GFP_KERNEL);
1659 	if (!skb) {
1660 		struct sock *sk;
1661 
1662 		sk = netlink_lookup(sock_net(in_skb->sk),
1663 				    in_skb->sk->sk_protocol,
1664 				    NETLINK_CB(in_skb).pid);
1665 		if (sk) {
1666 			sk->sk_err = ENOBUFS;
1667 			sk->sk_error_report(sk);
1668 			sock_put(sk);
1669 		}
1670 		return;
1671 	}
1672 
1673 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1674 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1675 	errmsg = nlmsg_data(rep);
1676 	errmsg->error = err;
1677 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1678 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1679 }
1680 EXPORT_SYMBOL(netlink_ack);
1681 
netlink_rcv_skb(struct sk_buff * skb,int (* cb)(struct sk_buff *,struct nlmsghdr *))1682 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1683 						     struct nlmsghdr *))
1684 {
1685 	struct nlmsghdr *nlh;
1686 	int err;
1687 
1688 	while (skb->len >= nlmsg_total_size(0)) {
1689 		int msglen;
1690 
1691 		nlh = nlmsg_hdr(skb);
1692 		err = 0;
1693 
1694 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1695 			return 0;
1696 
1697 		/* Only requests are handled by the kernel */
1698 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1699 			goto ack;
1700 
1701 		/* Skip control messages */
1702 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1703 			goto ack;
1704 
1705 		err = cb(skb, nlh);
1706 		if (err == -EINTR)
1707 			goto skip;
1708 
1709 ack:
1710 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1711 			netlink_ack(skb, nlh, err);
1712 
1713 skip:
1714 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1715 		if (msglen > skb->len)
1716 			msglen = skb->len;
1717 		skb_pull(skb, msglen);
1718 	}
1719 
1720 	return 0;
1721 }
1722 EXPORT_SYMBOL(netlink_rcv_skb);
1723 
1724 /**
1725  * nlmsg_notify - send a notification netlink message
1726  * @sk: netlink socket to use
1727  * @skb: notification message
1728  * @pid: destination netlink pid for reports or 0
1729  * @group: destination multicast group or 0
1730  * @report: 1 to report back, 0 to disable
1731  * @flags: allocation flags
1732  */
nlmsg_notify(struct sock * sk,struct sk_buff * skb,u32 pid,unsigned int group,int report,gfp_t flags)1733 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1734 		 unsigned int group, int report, gfp_t flags)
1735 {
1736 	int err = 0;
1737 
1738 	if (group) {
1739 		int exclude_pid = 0;
1740 
1741 		if (report) {
1742 			atomic_inc(&skb->users);
1743 			exclude_pid = pid;
1744 		}
1745 
1746 		/* errors reported via destination sk->sk_err */
1747 		nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1748 	}
1749 
1750 	if (report)
1751 		err = nlmsg_unicast(sk, skb, pid);
1752 
1753 	return err;
1754 }
1755 EXPORT_SYMBOL(nlmsg_notify);
1756 
1757 #ifdef CONFIG_PROC_FS
1758 struct nl_seq_iter {
1759 	struct seq_net_private p;
1760 	int link;
1761 	int hash_idx;
1762 };
1763 
netlink_seq_socket_idx(struct seq_file * seq,loff_t pos)1764 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1765 {
1766 	struct nl_seq_iter *iter = seq->private;
1767 	int i, j;
1768 	struct sock *s;
1769 	struct hlist_node *node;
1770 	loff_t off = 0;
1771 
1772 	for (i = 0; i < MAX_LINKS; i++) {
1773 		struct nl_pid_hash *hash = &nl_table[i].hash;
1774 
1775 		for (j = 0; j <= hash->mask; j++) {
1776 			sk_for_each(s, node, &hash->table[j]) {
1777 				if (sock_net(s) != seq_file_net(seq))
1778 					continue;
1779 				if (off == pos) {
1780 					iter->link = i;
1781 					iter->hash_idx = j;
1782 					return s;
1783 				}
1784 				++off;
1785 			}
1786 		}
1787 	}
1788 	return NULL;
1789 }
1790 
netlink_seq_start(struct seq_file * seq,loff_t * pos)1791 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1792 	__acquires(nl_table_lock)
1793 {
1794 	read_lock(&nl_table_lock);
1795 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1796 }
1797 
netlink_seq_next(struct seq_file * seq,void * v,loff_t * pos)1798 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1799 {
1800 	struct sock *s;
1801 	struct nl_seq_iter *iter;
1802 	int i, j;
1803 
1804 	++*pos;
1805 
1806 	if (v == SEQ_START_TOKEN)
1807 		return netlink_seq_socket_idx(seq, 0);
1808 
1809 	iter = seq->private;
1810 	s = v;
1811 	do {
1812 		s = sk_next(s);
1813 	} while (s && sock_net(s) != seq_file_net(seq));
1814 	if (s)
1815 		return s;
1816 
1817 	i = iter->link;
1818 	j = iter->hash_idx + 1;
1819 
1820 	do {
1821 		struct nl_pid_hash *hash = &nl_table[i].hash;
1822 
1823 		for (; j <= hash->mask; j++) {
1824 			s = sk_head(&hash->table[j]);
1825 			while (s && sock_net(s) != seq_file_net(seq))
1826 				s = sk_next(s);
1827 			if (s) {
1828 				iter->link = i;
1829 				iter->hash_idx = j;
1830 				return s;
1831 			}
1832 		}
1833 
1834 		j = 0;
1835 	} while (++i < MAX_LINKS);
1836 
1837 	return NULL;
1838 }
1839 
netlink_seq_stop(struct seq_file * seq,void * v)1840 static void netlink_seq_stop(struct seq_file *seq, void *v)
1841 	__releases(nl_table_lock)
1842 {
1843 	read_unlock(&nl_table_lock);
1844 }
1845 
1846 
netlink_seq_show(struct seq_file * seq,void * v)1847 static int netlink_seq_show(struct seq_file *seq, void *v)
1848 {
1849 	if (v == SEQ_START_TOKEN)
1850 		seq_puts(seq,
1851 			 "sk       Eth Pid    Groups   "
1852 			 "Rmem     Wmem     Dump     Locks\n");
1853 	else {
1854 		struct sock *s = v;
1855 		struct netlink_sock *nlk = nlk_sk(s);
1856 
1857 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1858 			   s,
1859 			   s->sk_protocol,
1860 			   nlk->pid,
1861 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1862 			   atomic_read(&s->sk_rmem_alloc),
1863 			   atomic_read(&s->sk_wmem_alloc),
1864 			   nlk->cb,
1865 			   atomic_read(&s->sk_refcnt)
1866 			);
1867 
1868 	}
1869 	return 0;
1870 }
1871 
1872 static const struct seq_operations netlink_seq_ops = {
1873 	.start  = netlink_seq_start,
1874 	.next   = netlink_seq_next,
1875 	.stop   = netlink_seq_stop,
1876 	.show   = netlink_seq_show,
1877 };
1878 
1879 
netlink_seq_open(struct inode * inode,struct file * file)1880 static int netlink_seq_open(struct inode *inode, struct file *file)
1881 {
1882 	return seq_open_net(inode, file, &netlink_seq_ops,
1883 				sizeof(struct nl_seq_iter));
1884 }
1885 
1886 static const struct file_operations netlink_seq_fops = {
1887 	.owner		= THIS_MODULE,
1888 	.open		= netlink_seq_open,
1889 	.read		= seq_read,
1890 	.llseek		= seq_lseek,
1891 	.release	= seq_release_net,
1892 };
1893 
1894 #endif
1895 
netlink_register_notifier(struct notifier_block * nb)1896 int netlink_register_notifier(struct notifier_block *nb)
1897 {
1898 	return atomic_notifier_chain_register(&netlink_chain, nb);
1899 }
1900 EXPORT_SYMBOL(netlink_register_notifier);
1901 
netlink_unregister_notifier(struct notifier_block * nb)1902 int netlink_unregister_notifier(struct notifier_block *nb)
1903 {
1904 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
1905 }
1906 EXPORT_SYMBOL(netlink_unregister_notifier);
1907 
1908 static const struct proto_ops netlink_ops = {
1909 	.family =	PF_NETLINK,
1910 	.owner =	THIS_MODULE,
1911 	.release =	netlink_release,
1912 	.bind =		netlink_bind,
1913 	.connect =	netlink_connect,
1914 	.socketpair =	sock_no_socketpair,
1915 	.accept =	sock_no_accept,
1916 	.getname =	netlink_getname,
1917 	.poll =		datagram_poll,
1918 	.ioctl =	sock_no_ioctl,
1919 	.listen =	sock_no_listen,
1920 	.shutdown =	sock_no_shutdown,
1921 	.setsockopt =	netlink_setsockopt,
1922 	.getsockopt =	netlink_getsockopt,
1923 	.sendmsg =	netlink_sendmsg,
1924 	.recvmsg =	netlink_recvmsg,
1925 	.mmap =		sock_no_mmap,
1926 	.sendpage =	sock_no_sendpage,
1927 };
1928 
1929 static struct net_proto_family netlink_family_ops = {
1930 	.family = PF_NETLINK,
1931 	.create = netlink_create,
1932 	.owner	= THIS_MODULE,	/* for consistency 8) */
1933 };
1934 
netlink_net_init(struct net * net)1935 static int __net_init netlink_net_init(struct net *net)
1936 {
1937 #ifdef CONFIG_PROC_FS
1938 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1939 		return -ENOMEM;
1940 #endif
1941 	return 0;
1942 }
1943 
netlink_net_exit(struct net * net)1944 static void __net_exit netlink_net_exit(struct net *net)
1945 {
1946 #ifdef CONFIG_PROC_FS
1947 	proc_net_remove(net, "netlink");
1948 #endif
1949 }
1950 
1951 static struct pernet_operations __net_initdata netlink_net_ops = {
1952 	.init = netlink_net_init,
1953 	.exit = netlink_net_exit,
1954 };
1955 
netlink_proto_init(void)1956 static int __init netlink_proto_init(void)
1957 {
1958 	struct sk_buff *dummy_skb;
1959 	int i;
1960 	unsigned long limit;
1961 	unsigned int order;
1962 	int err = proto_register(&netlink_proto, 0);
1963 
1964 	if (err != 0)
1965 		goto out;
1966 
1967 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1968 
1969 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1970 	if (!nl_table)
1971 		goto panic;
1972 
1973 	if (num_physpages >= (128 * 1024))
1974 		limit = num_physpages >> (21 - PAGE_SHIFT);
1975 	else
1976 		limit = num_physpages >> (23 - PAGE_SHIFT);
1977 
1978 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1979 	limit = (1UL << order) / sizeof(struct hlist_head);
1980 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1981 
1982 	for (i = 0; i < MAX_LINKS; i++) {
1983 		struct nl_pid_hash *hash = &nl_table[i].hash;
1984 
1985 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1986 		if (!hash->table) {
1987 			while (i-- > 0)
1988 				nl_pid_hash_free(nl_table[i].hash.table,
1989 						 1 * sizeof(*hash->table));
1990 			kfree(nl_table);
1991 			goto panic;
1992 		}
1993 		hash->max_shift = order;
1994 		hash->shift = 0;
1995 		hash->mask = 0;
1996 		hash->rehash_time = jiffies;
1997 	}
1998 
1999 	sock_register(&netlink_family_ops);
2000 	register_pernet_subsys(&netlink_net_ops);
2001 	/* The netlink device handler may be needed early. */
2002 	rtnetlink_init();
2003 out:
2004 	return err;
2005 panic:
2006 	panic("netlink_init: Cannot allocate nl_table\n");
2007 }
2008 
2009 core_initcall(netlink_proto_init);
2010