xref: /openbmc/linux/net/netlink/af_netlink.c (revision f42b3800)
1 /*
2  * NETLINK      Kernel-user communication protocol.
3  *
4  * 		Authors:	Alan Cox <alan@redhat.com>
5  * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6  *
7  *		This program is free software; you can redistribute it and/or
8  *		modify it under the terms of the GNU General Public License
9  *		as published by the Free Software Foundation; either version
10  *		2 of the License, or (at your option) any later version.
11  *
12  * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13  *                               added netlink_proto_exit
14  * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15  * 				 use nlk_sk, as sk->protinfo is on a diet 8)
16  * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17  * 				 - inc module use count of module that owns
18  * 				   the kernel socket in case userspace opens
19  * 				   socket of same protocol
20  * 				 - remove all module support, since netlink is
21  * 				   mandatory if CONFIG_NET=y these days
22  */
23 
24 #include <linux/module.h>
25 
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/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 	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
162 	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
163 	BUG_TRAP(!nlk_sk(sk)->groups);
164 }
165 
166 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
167  * SMP. Look, when several writers sleep and reader wakes them up, all but one
168  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
169  * this, _but_ remember, it adds useless work on UP machines.
170  */
171 
172 static void netlink_table_grab(void)
173 	__acquires(nl_table_lock)
174 {
175 	write_lock_irq(&nl_table_lock);
176 
177 	if (atomic_read(&nl_table_users)) {
178 		DECLARE_WAITQUEUE(wait, current);
179 
180 		add_wait_queue_exclusive(&nl_table_wait, &wait);
181 		for (;;) {
182 			set_current_state(TASK_UNINTERRUPTIBLE);
183 			if (atomic_read(&nl_table_users) == 0)
184 				break;
185 			write_unlock_irq(&nl_table_lock);
186 			schedule();
187 			write_lock_irq(&nl_table_lock);
188 		}
189 
190 		__set_current_state(TASK_RUNNING);
191 		remove_wait_queue(&nl_table_wait, &wait);
192 	}
193 }
194 
195 static void netlink_table_ungrab(void)
196 	__releases(nl_table_lock)
197 {
198 	write_unlock_irq(&nl_table_lock);
199 	wake_up(&nl_table_wait);
200 }
201 
202 static inline void
203 netlink_lock_table(void)
204 {
205 	/* read_lock() synchronizes us to netlink_table_grab */
206 
207 	read_lock(&nl_table_lock);
208 	atomic_inc(&nl_table_users);
209 	read_unlock(&nl_table_lock);
210 }
211 
212 static inline void
213 netlink_unlock_table(void)
214 {
215 	if (atomic_dec_and_test(&nl_table_users))
216 		wake_up(&nl_table_wait);
217 }
218 
219 static inline struct sock *netlink_lookup(struct net *net, int protocol,
220 					  u32 pid)
221 {
222 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
223 	struct hlist_head *head;
224 	struct sock *sk;
225 	struct hlist_node *node;
226 
227 	read_lock(&nl_table_lock);
228 	head = nl_pid_hashfn(hash, pid);
229 	sk_for_each(sk, node, head) {
230 		if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
231 			sock_hold(sk);
232 			goto found;
233 		}
234 	}
235 	sk = NULL;
236 found:
237 	read_unlock(&nl_table_lock);
238 	return sk;
239 }
240 
241 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
242 {
243 	if (size <= PAGE_SIZE)
244 		return kzalloc(size, GFP_ATOMIC);
245 	else
246 		return (struct hlist_head *)
247 			__get_free_pages(GFP_ATOMIC | __GFP_ZERO,
248 					 get_order(size));
249 }
250 
251 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
252 {
253 	if (size <= PAGE_SIZE)
254 		kfree(table);
255 	else
256 		free_pages((unsigned long)table, get_order(size));
257 }
258 
259 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
260 {
261 	unsigned int omask, mask, shift;
262 	size_t osize, size;
263 	struct hlist_head *otable, *table;
264 	int i;
265 
266 	omask = mask = hash->mask;
267 	osize = size = (mask + 1) * sizeof(*table);
268 	shift = hash->shift;
269 
270 	if (grow) {
271 		if (++shift > hash->max_shift)
272 			return 0;
273 		mask = mask * 2 + 1;
274 		size *= 2;
275 	}
276 
277 	table = nl_pid_hash_zalloc(size);
278 	if (!table)
279 		return 0;
280 
281 	otable = hash->table;
282 	hash->table = table;
283 	hash->mask = mask;
284 	hash->shift = shift;
285 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
286 
287 	for (i = 0; i <= omask; i++) {
288 		struct sock *sk;
289 		struct hlist_node *node, *tmp;
290 
291 		sk_for_each_safe(sk, node, tmp, &otable[i])
292 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
293 	}
294 
295 	nl_pid_hash_free(otable, osize);
296 	hash->rehash_time = jiffies + 10 * 60 * HZ;
297 	return 1;
298 }
299 
300 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
301 {
302 	int avg = hash->entries >> hash->shift;
303 
304 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
305 		return 1;
306 
307 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
308 		nl_pid_hash_rehash(hash, 0);
309 		return 1;
310 	}
311 
312 	return 0;
313 }
314 
315 static const struct proto_ops netlink_ops;
316 
317 static void
318 netlink_update_listeners(struct sock *sk)
319 {
320 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
321 	struct hlist_node *node;
322 	unsigned long mask;
323 	unsigned int i;
324 
325 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
326 		mask = 0;
327 		sk_for_each_bound(sk, node, &tbl->mc_list) {
328 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
329 				mask |= nlk_sk(sk)->groups[i];
330 		}
331 		tbl->listeners[i] = mask;
332 	}
333 	/* this function is only called with the netlink table "grabbed", which
334 	 * makes sure updates are visible before bind or setsockopt return. */
335 }
336 
337 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
338 {
339 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
340 	struct hlist_head *head;
341 	int err = -EADDRINUSE;
342 	struct sock *osk;
343 	struct hlist_node *node;
344 	int len;
345 
346 	netlink_table_grab();
347 	head = nl_pid_hashfn(hash, pid);
348 	len = 0;
349 	sk_for_each(osk, node, head) {
350 		if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
351 			break;
352 		len++;
353 	}
354 	if (node)
355 		goto err;
356 
357 	err = -EBUSY;
358 	if (nlk_sk(sk)->pid)
359 		goto err;
360 
361 	err = -ENOMEM;
362 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
363 		goto err;
364 
365 	if (len && nl_pid_hash_dilute(hash, len))
366 		head = nl_pid_hashfn(hash, pid);
367 	hash->entries++;
368 	nlk_sk(sk)->pid = pid;
369 	sk_add_node(sk, head);
370 	err = 0;
371 
372 err:
373 	netlink_table_ungrab();
374 	return err;
375 }
376 
377 static void netlink_remove(struct sock *sk)
378 {
379 	netlink_table_grab();
380 	if (sk_del_node_init(sk))
381 		nl_table[sk->sk_protocol].hash.entries--;
382 	if (nlk_sk(sk)->subscriptions)
383 		__sk_del_bind_node(sk);
384 	netlink_table_ungrab();
385 }
386 
387 static struct proto netlink_proto = {
388 	.name	  = "NETLINK",
389 	.owner	  = THIS_MODULE,
390 	.obj_size = sizeof(struct netlink_sock),
391 };
392 
393 static int __netlink_create(struct net *net, struct socket *sock,
394 			    struct mutex *cb_mutex, int protocol)
395 {
396 	struct sock *sk;
397 	struct netlink_sock *nlk;
398 
399 	sock->ops = &netlink_ops;
400 
401 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
402 	if (!sk)
403 		return -ENOMEM;
404 
405 	sock_init_data(sock, sk);
406 
407 	nlk = nlk_sk(sk);
408 	if (cb_mutex)
409 		nlk->cb_mutex = cb_mutex;
410 	else {
411 		nlk->cb_mutex = &nlk->cb_def_mutex;
412 		mutex_init(nlk->cb_mutex);
413 	}
414 	init_waitqueue_head(&nlk->wait);
415 
416 	sk->sk_destruct = netlink_sock_destruct;
417 	sk->sk_protocol = protocol;
418 	return 0;
419 }
420 
421 static int netlink_create(struct net *net, struct socket *sock, int protocol)
422 {
423 	struct module *module = NULL;
424 	struct mutex *cb_mutex;
425 	struct netlink_sock *nlk;
426 	int err = 0;
427 
428 	sock->state = SS_UNCONNECTED;
429 
430 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
431 		return -ESOCKTNOSUPPORT;
432 
433 	if (protocol < 0 || protocol >= MAX_LINKS)
434 		return -EPROTONOSUPPORT;
435 
436 	netlink_lock_table();
437 #ifdef CONFIG_KMOD
438 	if (!nl_table[protocol].registered) {
439 		netlink_unlock_table();
440 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
441 		netlink_lock_table();
442 	}
443 #endif
444 	if (nl_table[protocol].registered &&
445 	    try_module_get(nl_table[protocol].module))
446 		module = nl_table[protocol].module;
447 	cb_mutex = nl_table[protocol].cb_mutex;
448 	netlink_unlock_table();
449 
450 	err = __netlink_create(net, sock, cb_mutex, protocol);
451 	if (err < 0)
452 		goto out_module;
453 
454 	nlk = nlk_sk(sock->sk);
455 	nlk->module = module;
456 out:
457 	return err;
458 
459 out_module:
460 	module_put(module);
461 	goto out;
462 }
463 
464 static int netlink_release(struct socket *sock)
465 {
466 	struct sock *sk = sock->sk;
467 	struct netlink_sock *nlk;
468 
469 	if (!sk)
470 		return 0;
471 
472 	netlink_remove(sk);
473 	sock_orphan(sk);
474 	nlk = nlk_sk(sk);
475 
476 	/*
477 	 * OK. Socket is unlinked, any packets that arrive now
478 	 * will be purged.
479 	 */
480 
481 	sock->sk = NULL;
482 	wake_up_interruptible_all(&nlk->wait);
483 
484 	skb_queue_purge(&sk->sk_write_queue);
485 
486 	if (nlk->pid && !nlk->subscriptions) {
487 		struct netlink_notify n = {
488 						.net = sock_net(sk),
489 						.protocol = sk->sk_protocol,
490 						.pid = nlk->pid,
491 					  };
492 		atomic_notifier_call_chain(&netlink_chain,
493 				NETLINK_URELEASE, &n);
494 	}
495 
496 	module_put(nlk->module);
497 
498 	netlink_table_grab();
499 	if (netlink_is_kernel(sk)) {
500 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
501 		if (--nl_table[sk->sk_protocol].registered == 0) {
502 			kfree(nl_table[sk->sk_protocol].listeners);
503 			nl_table[sk->sk_protocol].module = NULL;
504 			nl_table[sk->sk_protocol].registered = 0;
505 		}
506 	} else if (nlk->subscriptions)
507 		netlink_update_listeners(sk);
508 	netlink_table_ungrab();
509 
510 	kfree(nlk->groups);
511 	nlk->groups = NULL;
512 
513 	sock_put(sk);
514 	return 0;
515 }
516 
517 static int netlink_autobind(struct socket *sock)
518 {
519 	struct sock *sk = sock->sk;
520 	struct net *net = sock_net(sk);
521 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
522 	struct hlist_head *head;
523 	struct sock *osk;
524 	struct hlist_node *node;
525 	s32 pid = current->tgid;
526 	int err;
527 	static s32 rover = -4097;
528 
529 retry:
530 	cond_resched();
531 	netlink_table_grab();
532 	head = nl_pid_hashfn(hash, pid);
533 	sk_for_each(osk, node, head) {
534 		if (!net_eq(sock_net(osk), net))
535 			continue;
536 		if (nlk_sk(osk)->pid == pid) {
537 			/* Bind collision, search negative pid values. */
538 			pid = rover--;
539 			if (rover > -4097)
540 				rover = -4097;
541 			netlink_table_ungrab();
542 			goto retry;
543 		}
544 	}
545 	netlink_table_ungrab();
546 
547 	err = netlink_insert(sk, net, pid);
548 	if (err == -EADDRINUSE)
549 		goto retry;
550 
551 	/* If 2 threads race to autobind, that is fine.  */
552 	if (err == -EBUSY)
553 		err = 0;
554 
555 	return err;
556 }
557 
558 static inline int netlink_capable(struct socket *sock, unsigned int flag)
559 {
560 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
561 	       capable(CAP_NET_ADMIN);
562 }
563 
564 static void
565 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
566 {
567 	struct netlink_sock *nlk = nlk_sk(sk);
568 
569 	if (nlk->subscriptions && !subscriptions)
570 		__sk_del_bind_node(sk);
571 	else if (!nlk->subscriptions && subscriptions)
572 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
573 	nlk->subscriptions = subscriptions;
574 }
575 
576 static int netlink_realloc_groups(struct sock *sk)
577 {
578 	struct netlink_sock *nlk = nlk_sk(sk);
579 	unsigned int groups;
580 	unsigned long *new_groups;
581 	int err = 0;
582 
583 	netlink_table_grab();
584 
585 	groups = nl_table[sk->sk_protocol].groups;
586 	if (!nl_table[sk->sk_protocol].registered) {
587 		err = -ENOENT;
588 		goto out_unlock;
589 	}
590 
591 	if (nlk->ngroups >= groups)
592 		goto out_unlock;
593 
594 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
595 	if (new_groups == NULL) {
596 		err = -ENOMEM;
597 		goto out_unlock;
598 	}
599 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
600 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
601 
602 	nlk->groups = new_groups;
603 	nlk->ngroups = groups;
604  out_unlock:
605 	netlink_table_ungrab();
606 	return err;
607 }
608 
609 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
610 			int addr_len)
611 {
612 	struct sock *sk = sock->sk;
613 	struct net *net = sock_net(sk);
614 	struct netlink_sock *nlk = nlk_sk(sk);
615 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
616 	int err;
617 
618 	if (nladdr->nl_family != AF_NETLINK)
619 		return -EINVAL;
620 
621 	/* Only superuser is allowed to listen multicasts */
622 	if (nladdr->nl_groups) {
623 		if (!netlink_capable(sock, NL_NONROOT_RECV))
624 			return -EPERM;
625 		err = netlink_realloc_groups(sk);
626 		if (err)
627 			return err;
628 	}
629 
630 	if (nlk->pid) {
631 		if (nladdr->nl_pid != nlk->pid)
632 			return -EINVAL;
633 	} else {
634 		err = nladdr->nl_pid ?
635 			netlink_insert(sk, net, nladdr->nl_pid) :
636 			netlink_autobind(sock);
637 		if (err)
638 			return err;
639 	}
640 
641 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
642 		return 0;
643 
644 	netlink_table_grab();
645 	netlink_update_subscriptions(sk, nlk->subscriptions +
646 					 hweight32(nladdr->nl_groups) -
647 					 hweight32(nlk->groups[0]));
648 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
649 	netlink_update_listeners(sk);
650 	netlink_table_ungrab();
651 
652 	return 0;
653 }
654 
655 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
656 			   int alen, int flags)
657 {
658 	int err = 0;
659 	struct sock *sk = sock->sk;
660 	struct netlink_sock *nlk = nlk_sk(sk);
661 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
662 
663 	if (addr->sa_family == AF_UNSPEC) {
664 		sk->sk_state	= NETLINK_UNCONNECTED;
665 		nlk->dst_pid	= 0;
666 		nlk->dst_group  = 0;
667 		return 0;
668 	}
669 	if (addr->sa_family != AF_NETLINK)
670 		return -EINVAL;
671 
672 	/* Only superuser is allowed to send multicasts */
673 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
674 		return -EPERM;
675 
676 	if (!nlk->pid)
677 		err = netlink_autobind(sock);
678 
679 	if (err == 0) {
680 		sk->sk_state	= NETLINK_CONNECTED;
681 		nlk->dst_pid 	= nladdr->nl_pid;
682 		nlk->dst_group  = ffs(nladdr->nl_groups);
683 	}
684 
685 	return err;
686 }
687 
688 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
689 			   int *addr_len, int peer)
690 {
691 	struct sock *sk = sock->sk;
692 	struct netlink_sock *nlk = nlk_sk(sk);
693 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
694 
695 	nladdr->nl_family = AF_NETLINK;
696 	nladdr->nl_pad = 0;
697 	*addr_len = sizeof(*nladdr);
698 
699 	if (peer) {
700 		nladdr->nl_pid = nlk->dst_pid;
701 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
702 	} else {
703 		nladdr->nl_pid = nlk->pid;
704 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
705 	}
706 	return 0;
707 }
708 
709 static void netlink_overrun(struct sock *sk)
710 {
711 	if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
712 		sk->sk_err = ENOBUFS;
713 		sk->sk_error_report(sk);
714 	}
715 }
716 
717 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
718 {
719 	struct sock *sock;
720 	struct netlink_sock *nlk;
721 
722 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
723 	if (!sock)
724 		return ERR_PTR(-ECONNREFUSED);
725 
726 	/* Don't bother queuing skb if kernel socket has no input function */
727 	nlk = nlk_sk(sock);
728 	if (sock->sk_state == NETLINK_CONNECTED &&
729 	    nlk->dst_pid != nlk_sk(ssk)->pid) {
730 		sock_put(sock);
731 		return ERR_PTR(-ECONNREFUSED);
732 	}
733 	return sock;
734 }
735 
736 struct sock *netlink_getsockbyfilp(struct file *filp)
737 {
738 	struct inode *inode = filp->f_path.dentry->d_inode;
739 	struct sock *sock;
740 
741 	if (!S_ISSOCK(inode->i_mode))
742 		return ERR_PTR(-ENOTSOCK);
743 
744 	sock = SOCKET_I(inode)->sk;
745 	if (sock->sk_family != AF_NETLINK)
746 		return ERR_PTR(-EINVAL);
747 
748 	sock_hold(sock);
749 	return sock;
750 }
751 
752 /*
753  * Attach a skb to a netlink socket.
754  * The caller must hold a reference to the destination socket. On error, the
755  * reference is dropped. The skb is not send to the destination, just all
756  * all error checks are performed and memory in the queue is reserved.
757  * Return values:
758  * < 0: error. skb freed, reference to sock dropped.
759  * 0: continue
760  * 1: repeat lookup - reference dropped while waiting for socket memory.
761  */
762 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
763 		      long *timeo, struct sock *ssk)
764 {
765 	struct netlink_sock *nlk;
766 
767 	nlk = nlk_sk(sk);
768 
769 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
770 	    test_bit(0, &nlk->state)) {
771 		DECLARE_WAITQUEUE(wait, current);
772 		if (!*timeo) {
773 			if (!ssk || netlink_is_kernel(ssk))
774 				netlink_overrun(sk);
775 			sock_put(sk);
776 			kfree_skb(skb);
777 			return -EAGAIN;
778 		}
779 
780 		__set_current_state(TASK_INTERRUPTIBLE);
781 		add_wait_queue(&nlk->wait, &wait);
782 
783 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
784 		     test_bit(0, &nlk->state)) &&
785 		    !sock_flag(sk, SOCK_DEAD))
786 			*timeo = schedule_timeout(*timeo);
787 
788 		__set_current_state(TASK_RUNNING);
789 		remove_wait_queue(&nlk->wait, &wait);
790 		sock_put(sk);
791 
792 		if (signal_pending(current)) {
793 			kfree_skb(skb);
794 			return sock_intr_errno(*timeo);
795 		}
796 		return 1;
797 	}
798 	skb_set_owner_r(skb, sk);
799 	return 0;
800 }
801 
802 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
803 {
804 	int len = skb->len;
805 
806 	skb_queue_tail(&sk->sk_receive_queue, skb);
807 	sk->sk_data_ready(sk, len);
808 	sock_put(sk);
809 	return len;
810 }
811 
812 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
813 {
814 	kfree_skb(skb);
815 	sock_put(sk);
816 }
817 
818 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
819 					   gfp_t allocation)
820 {
821 	int delta;
822 
823 	skb_orphan(skb);
824 
825 	delta = skb->end - skb->tail;
826 	if (delta * 2 < skb->truesize)
827 		return skb;
828 
829 	if (skb_shared(skb)) {
830 		struct sk_buff *nskb = skb_clone(skb, allocation);
831 		if (!nskb)
832 			return skb;
833 		kfree_skb(skb);
834 		skb = nskb;
835 	}
836 
837 	if (!pskb_expand_head(skb, 0, -delta, allocation))
838 		skb->truesize -= delta;
839 
840 	return skb;
841 }
842 
843 static inline void netlink_rcv_wake(struct sock *sk)
844 {
845 	struct netlink_sock *nlk = nlk_sk(sk);
846 
847 	if (skb_queue_empty(&sk->sk_receive_queue))
848 		clear_bit(0, &nlk->state);
849 	if (!test_bit(0, &nlk->state))
850 		wake_up_interruptible(&nlk->wait);
851 }
852 
853 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
854 {
855 	int ret;
856 	struct netlink_sock *nlk = nlk_sk(sk);
857 
858 	ret = -ECONNREFUSED;
859 	if (nlk->netlink_rcv != NULL) {
860 		ret = skb->len;
861 		skb_set_owner_r(skb, sk);
862 		nlk->netlink_rcv(skb);
863 	}
864 	kfree_skb(skb);
865 	sock_put(sk);
866 	return ret;
867 }
868 
869 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
870 		    u32 pid, int nonblock)
871 {
872 	struct sock *sk;
873 	int err;
874 	long timeo;
875 
876 	skb = netlink_trim(skb, gfp_any());
877 
878 	timeo = sock_sndtimeo(ssk, nonblock);
879 retry:
880 	sk = netlink_getsockbypid(ssk, pid);
881 	if (IS_ERR(sk)) {
882 		kfree_skb(skb);
883 		return PTR_ERR(sk);
884 	}
885 	if (netlink_is_kernel(sk))
886 		return netlink_unicast_kernel(sk, skb);
887 
888 	if (sk_filter(sk, skb)) {
889 		int err = skb->len;
890 		kfree_skb(skb);
891 		sock_put(sk);
892 		return err;
893 	}
894 
895 	err = netlink_attachskb(sk, skb, nonblock, &timeo, ssk);
896 	if (err == 1)
897 		goto retry;
898 	if (err)
899 		return err;
900 
901 	return netlink_sendskb(sk, skb);
902 }
903 EXPORT_SYMBOL(netlink_unicast);
904 
905 int netlink_has_listeners(struct sock *sk, unsigned int group)
906 {
907 	int res = 0;
908 	unsigned long *listeners;
909 
910 	BUG_ON(!netlink_is_kernel(sk));
911 
912 	rcu_read_lock();
913 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
914 
915 	if (group - 1 < nl_table[sk->sk_protocol].groups)
916 		res = test_bit(group - 1, listeners);
917 
918 	rcu_read_unlock();
919 
920 	return res;
921 }
922 EXPORT_SYMBOL_GPL(netlink_has_listeners);
923 
924 static inline int netlink_broadcast_deliver(struct sock *sk,
925 					    struct sk_buff *skb)
926 {
927 	struct netlink_sock *nlk = nlk_sk(sk);
928 
929 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
930 	    !test_bit(0, &nlk->state)) {
931 		skb_set_owner_r(skb, sk);
932 		skb_queue_tail(&sk->sk_receive_queue, skb);
933 		sk->sk_data_ready(sk, skb->len);
934 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
935 	}
936 	return -1;
937 }
938 
939 struct netlink_broadcast_data {
940 	struct sock *exclude_sk;
941 	struct net *net;
942 	u32 pid;
943 	u32 group;
944 	int failure;
945 	int congested;
946 	int delivered;
947 	gfp_t allocation;
948 	struct sk_buff *skb, *skb2;
949 };
950 
951 static inline int do_one_broadcast(struct sock *sk,
952 				   struct netlink_broadcast_data *p)
953 {
954 	struct netlink_sock *nlk = nlk_sk(sk);
955 	int val;
956 
957 	if (p->exclude_sk == sk)
958 		goto out;
959 
960 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
961 	    !test_bit(p->group - 1, nlk->groups))
962 		goto out;
963 
964 	if (!net_eq(sock_net(sk), p->net))
965 		goto out;
966 
967 	if (p->failure) {
968 		netlink_overrun(sk);
969 		goto out;
970 	}
971 
972 	sock_hold(sk);
973 	if (p->skb2 == NULL) {
974 		if (skb_shared(p->skb)) {
975 			p->skb2 = skb_clone(p->skb, p->allocation);
976 		} else {
977 			p->skb2 = skb_get(p->skb);
978 			/*
979 			 * skb ownership may have been set when
980 			 * delivered to a previous socket.
981 			 */
982 			skb_orphan(p->skb2);
983 		}
984 	}
985 	if (p->skb2 == NULL) {
986 		netlink_overrun(sk);
987 		/* Clone failed. Notify ALL listeners. */
988 		p->failure = 1;
989 	} else if (sk_filter(sk, p->skb2)) {
990 		kfree_skb(p->skb2);
991 		p->skb2 = NULL;
992 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
993 		netlink_overrun(sk);
994 	} else {
995 		p->congested |= val;
996 		p->delivered = 1;
997 		p->skb2 = NULL;
998 	}
999 	sock_put(sk);
1000 
1001 out:
1002 	return 0;
1003 }
1004 
1005 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1006 		      u32 group, gfp_t allocation)
1007 {
1008 	struct net *net = sock_net(ssk);
1009 	struct netlink_broadcast_data info;
1010 	struct hlist_node *node;
1011 	struct sock *sk;
1012 
1013 	skb = netlink_trim(skb, allocation);
1014 
1015 	info.exclude_sk = ssk;
1016 	info.net = net;
1017 	info.pid = pid;
1018 	info.group = group;
1019 	info.failure = 0;
1020 	info.congested = 0;
1021 	info.delivered = 0;
1022 	info.allocation = allocation;
1023 	info.skb = skb;
1024 	info.skb2 = NULL;
1025 
1026 	/* While we sleep in clone, do not allow to change socket list */
1027 
1028 	netlink_lock_table();
1029 
1030 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1031 		do_one_broadcast(sk, &info);
1032 
1033 	kfree_skb(skb);
1034 
1035 	netlink_unlock_table();
1036 
1037 	if (info.skb2)
1038 		kfree_skb(info.skb2);
1039 
1040 	if (info.delivered) {
1041 		if (info.congested && (allocation & __GFP_WAIT))
1042 			yield();
1043 		return 0;
1044 	}
1045 	if (info.failure)
1046 		return -ENOBUFS;
1047 	return -ESRCH;
1048 }
1049 EXPORT_SYMBOL(netlink_broadcast);
1050 
1051 struct netlink_set_err_data {
1052 	struct sock *exclude_sk;
1053 	u32 pid;
1054 	u32 group;
1055 	int code;
1056 };
1057 
1058 static inline int do_one_set_err(struct sock *sk,
1059 				 struct netlink_set_err_data *p)
1060 {
1061 	struct netlink_sock *nlk = nlk_sk(sk);
1062 
1063 	if (sk == p->exclude_sk)
1064 		goto out;
1065 
1066 	if (sock_net(sk) != sock_net(p->exclude_sk))
1067 		goto out;
1068 
1069 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1070 	    !test_bit(p->group - 1, nlk->groups))
1071 		goto out;
1072 
1073 	sk->sk_err = p->code;
1074 	sk->sk_error_report(sk);
1075 out:
1076 	return 0;
1077 }
1078 
1079 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1080 {
1081 	struct netlink_set_err_data info;
1082 	struct hlist_node *node;
1083 	struct sock *sk;
1084 
1085 	info.exclude_sk = ssk;
1086 	info.pid = pid;
1087 	info.group = group;
1088 	info.code = code;
1089 
1090 	read_lock(&nl_table_lock);
1091 
1092 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1093 		do_one_set_err(sk, &info);
1094 
1095 	read_unlock(&nl_table_lock);
1096 }
1097 
1098 /* must be called with netlink table grabbed */
1099 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1100 				     unsigned int group,
1101 				     int is_new)
1102 {
1103 	int old, new = !!is_new, subscriptions;
1104 
1105 	old = test_bit(group - 1, nlk->groups);
1106 	subscriptions = nlk->subscriptions - old + new;
1107 	if (new)
1108 		__set_bit(group - 1, nlk->groups);
1109 	else
1110 		__clear_bit(group - 1, nlk->groups);
1111 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1112 	netlink_update_listeners(&nlk->sk);
1113 }
1114 
1115 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1116 			      char __user *optval, int optlen)
1117 {
1118 	struct sock *sk = sock->sk;
1119 	struct netlink_sock *nlk = nlk_sk(sk);
1120 	unsigned int val = 0;
1121 	int err;
1122 
1123 	if (level != SOL_NETLINK)
1124 		return -ENOPROTOOPT;
1125 
1126 	if (optlen >= sizeof(int) &&
1127 	    get_user(val, (unsigned int __user *)optval))
1128 		return -EFAULT;
1129 
1130 	switch (optname) {
1131 	case NETLINK_PKTINFO:
1132 		if (val)
1133 			nlk->flags |= NETLINK_RECV_PKTINFO;
1134 		else
1135 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1136 		err = 0;
1137 		break;
1138 	case NETLINK_ADD_MEMBERSHIP:
1139 	case NETLINK_DROP_MEMBERSHIP: {
1140 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1141 			return -EPERM;
1142 		err = netlink_realloc_groups(sk);
1143 		if (err)
1144 			return err;
1145 		if (!val || val - 1 >= nlk->ngroups)
1146 			return -EINVAL;
1147 		netlink_table_grab();
1148 		netlink_update_socket_mc(nlk, val,
1149 					 optname == NETLINK_ADD_MEMBERSHIP);
1150 		netlink_table_ungrab();
1151 		err = 0;
1152 		break;
1153 	}
1154 	default:
1155 		err = -ENOPROTOOPT;
1156 	}
1157 	return err;
1158 }
1159 
1160 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1161 			      char __user *optval, int __user *optlen)
1162 {
1163 	struct sock *sk = sock->sk;
1164 	struct netlink_sock *nlk = nlk_sk(sk);
1165 	int len, val, err;
1166 
1167 	if (level != SOL_NETLINK)
1168 		return -ENOPROTOOPT;
1169 
1170 	if (get_user(len, optlen))
1171 		return -EFAULT;
1172 	if (len < 0)
1173 		return -EINVAL;
1174 
1175 	switch (optname) {
1176 	case NETLINK_PKTINFO:
1177 		if (len < sizeof(int))
1178 			return -EINVAL;
1179 		len = sizeof(int);
1180 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1181 		if (put_user(len, optlen) ||
1182 		    put_user(val, optval))
1183 			return -EFAULT;
1184 		err = 0;
1185 		break;
1186 	default:
1187 		err = -ENOPROTOOPT;
1188 	}
1189 	return err;
1190 }
1191 
1192 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1193 {
1194 	struct nl_pktinfo info;
1195 
1196 	info.group = NETLINK_CB(skb).dst_group;
1197 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1198 }
1199 
1200 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1201 			   struct msghdr *msg, size_t len)
1202 {
1203 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1204 	struct sock *sk = sock->sk;
1205 	struct netlink_sock *nlk = nlk_sk(sk);
1206 	struct sockaddr_nl *addr = msg->msg_name;
1207 	u32 dst_pid;
1208 	u32 dst_group;
1209 	struct sk_buff *skb;
1210 	int err;
1211 	struct scm_cookie scm;
1212 
1213 	if (msg->msg_flags&MSG_OOB)
1214 		return -EOPNOTSUPP;
1215 
1216 	if (NULL == siocb->scm)
1217 		siocb->scm = &scm;
1218 	err = scm_send(sock, msg, siocb->scm);
1219 	if (err < 0)
1220 		return err;
1221 
1222 	if (msg->msg_namelen) {
1223 		if (addr->nl_family != AF_NETLINK)
1224 			return -EINVAL;
1225 		dst_pid = addr->nl_pid;
1226 		dst_group = ffs(addr->nl_groups);
1227 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1228 			return -EPERM;
1229 	} else {
1230 		dst_pid = nlk->dst_pid;
1231 		dst_group = nlk->dst_group;
1232 	}
1233 
1234 	if (!nlk->pid) {
1235 		err = netlink_autobind(sock);
1236 		if (err)
1237 			goto out;
1238 	}
1239 
1240 	err = -EMSGSIZE;
1241 	if (len > sk->sk_sndbuf - 32)
1242 		goto out;
1243 	err = -ENOBUFS;
1244 	skb = alloc_skb(len, GFP_KERNEL);
1245 	if (skb == NULL)
1246 		goto out;
1247 
1248 	NETLINK_CB(skb).pid	= nlk->pid;
1249 	NETLINK_CB(skb).dst_group = dst_group;
1250 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1251 	security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1252 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1253 
1254 	/* What can I do? Netlink is asynchronous, so that
1255 	   we will have to save current capabilities to
1256 	   check them, when this message will be delivered
1257 	   to corresponding kernel module.   --ANK (980802)
1258 	 */
1259 
1260 	err = -EFAULT;
1261 	if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1262 		kfree_skb(skb);
1263 		goto out;
1264 	}
1265 
1266 	err = security_netlink_send(sk, skb);
1267 	if (err) {
1268 		kfree_skb(skb);
1269 		goto out;
1270 	}
1271 
1272 	if (dst_group) {
1273 		atomic_inc(&skb->users);
1274 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1275 	}
1276 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1277 
1278 out:
1279 	return err;
1280 }
1281 
1282 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1283 			   struct msghdr *msg, size_t len,
1284 			   int flags)
1285 {
1286 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1287 	struct scm_cookie scm;
1288 	struct sock *sk = sock->sk;
1289 	struct netlink_sock *nlk = nlk_sk(sk);
1290 	int noblock = flags&MSG_DONTWAIT;
1291 	size_t copied;
1292 	struct sk_buff *skb;
1293 	int err;
1294 
1295 	if (flags&MSG_OOB)
1296 		return -EOPNOTSUPP;
1297 
1298 	copied = 0;
1299 
1300 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1301 	if (skb == NULL)
1302 		goto out;
1303 
1304 	msg->msg_namelen = 0;
1305 
1306 	copied = skb->len;
1307 	if (len < copied) {
1308 		msg->msg_flags |= MSG_TRUNC;
1309 		copied = len;
1310 	}
1311 
1312 	skb_reset_transport_header(skb);
1313 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1314 
1315 	if (msg->msg_name) {
1316 		struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1317 		addr->nl_family = AF_NETLINK;
1318 		addr->nl_pad    = 0;
1319 		addr->nl_pid	= NETLINK_CB(skb).pid;
1320 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1321 		msg->msg_namelen = sizeof(*addr);
1322 	}
1323 
1324 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1325 		netlink_cmsg_recv_pktinfo(msg, skb);
1326 
1327 	if (NULL == siocb->scm) {
1328 		memset(&scm, 0, sizeof(scm));
1329 		siocb->scm = &scm;
1330 	}
1331 	siocb->scm->creds = *NETLINK_CREDS(skb);
1332 	if (flags & MSG_TRUNC)
1333 		copied = skb->len;
1334 	skb_free_datagram(sk, skb);
1335 
1336 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1337 		netlink_dump(sk);
1338 
1339 	scm_recv(sock, msg, siocb->scm, flags);
1340 out:
1341 	netlink_rcv_wake(sk);
1342 	return err ? : copied;
1343 }
1344 
1345 static void netlink_data_ready(struct sock *sk, int len)
1346 {
1347 	BUG();
1348 }
1349 
1350 /*
1351  *	We export these functions to other modules. They provide a
1352  *	complete set of kernel non-blocking support for message
1353  *	queueing.
1354  */
1355 
1356 struct sock *
1357 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1358 		      void (*input)(struct sk_buff *skb),
1359 		      struct mutex *cb_mutex, struct module *module)
1360 {
1361 	struct socket *sock;
1362 	struct sock *sk;
1363 	struct netlink_sock *nlk;
1364 	unsigned long *listeners = NULL;
1365 
1366 	BUG_ON(!nl_table);
1367 
1368 	if (unit < 0 || unit >= MAX_LINKS)
1369 		return NULL;
1370 
1371 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1372 		return NULL;
1373 
1374 	/*
1375 	 * We have to just have a reference on the net from sk, but don't
1376 	 * get_net it. Besides, we cannot get and then put the net here.
1377 	 * So we create one inside init_net and the move it to net.
1378 	 */
1379 
1380 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1381 		goto out_sock_release_nosk;
1382 
1383 	sk = sock->sk;
1384 	sk_change_net(sk, net);
1385 
1386 	if (groups < 32)
1387 		groups = 32;
1388 
1389 	listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1390 	if (!listeners)
1391 		goto out_sock_release;
1392 
1393 	sk->sk_data_ready = netlink_data_ready;
1394 	if (input)
1395 		nlk_sk(sk)->netlink_rcv = input;
1396 
1397 	if (netlink_insert(sk, net, 0))
1398 		goto out_sock_release;
1399 
1400 	nlk = nlk_sk(sk);
1401 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1402 
1403 	netlink_table_grab();
1404 	if (!nl_table[unit].registered) {
1405 		nl_table[unit].groups = groups;
1406 		nl_table[unit].listeners = listeners;
1407 		nl_table[unit].cb_mutex = cb_mutex;
1408 		nl_table[unit].module = module;
1409 		nl_table[unit].registered = 1;
1410 	} else {
1411 		kfree(listeners);
1412 		nl_table[unit].registered++;
1413 	}
1414 	netlink_table_ungrab();
1415 	return sk;
1416 
1417 out_sock_release:
1418 	kfree(listeners);
1419 	netlink_kernel_release(sk);
1420 	return NULL;
1421 
1422 out_sock_release_nosk:
1423 	sock_release(sock);
1424 	return NULL;
1425 }
1426 EXPORT_SYMBOL(netlink_kernel_create);
1427 
1428 
1429 void
1430 netlink_kernel_release(struct sock *sk)
1431 {
1432 	sk_release_kernel(sk);
1433 }
1434 EXPORT_SYMBOL(netlink_kernel_release);
1435 
1436 
1437 /**
1438  * netlink_change_ngroups - change number of multicast groups
1439  *
1440  * This changes the number of multicast groups that are available
1441  * on a certain netlink family. Note that it is not possible to
1442  * change the number of groups to below 32. Also note that it does
1443  * not implicitly call netlink_clear_multicast_users() when the
1444  * number of groups is reduced.
1445  *
1446  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1447  * @groups: The new number of groups.
1448  */
1449 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1450 {
1451 	unsigned long *listeners, *old = NULL;
1452 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1453 	int err = 0;
1454 
1455 	if (groups < 32)
1456 		groups = 32;
1457 
1458 	netlink_table_grab();
1459 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1460 		listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1461 		if (!listeners) {
1462 			err = -ENOMEM;
1463 			goto out_ungrab;
1464 		}
1465 		old = tbl->listeners;
1466 		memcpy(listeners, old, NLGRPSZ(tbl->groups));
1467 		rcu_assign_pointer(tbl->listeners, listeners);
1468 	}
1469 	tbl->groups = groups;
1470 
1471  out_ungrab:
1472 	netlink_table_ungrab();
1473 	synchronize_rcu();
1474 	kfree(old);
1475 	return err;
1476 }
1477 EXPORT_SYMBOL(netlink_change_ngroups);
1478 
1479 /**
1480  * netlink_clear_multicast_users - kick off multicast listeners
1481  *
1482  * This function removes all listeners from the given group.
1483  * @ksk: The kernel netlink socket, as returned by
1484  *	netlink_kernel_create().
1485  * @group: The multicast group to clear.
1486  */
1487 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1488 {
1489 	struct sock *sk;
1490 	struct hlist_node *node;
1491 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1492 
1493 	netlink_table_grab();
1494 
1495 	sk_for_each_bound(sk, node, &tbl->mc_list)
1496 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
1497 
1498 	netlink_table_ungrab();
1499 }
1500 EXPORT_SYMBOL(netlink_clear_multicast_users);
1501 
1502 void netlink_set_nonroot(int protocol, unsigned int flags)
1503 {
1504 	if ((unsigned int)protocol < MAX_LINKS)
1505 		nl_table[protocol].nl_nonroot = flags;
1506 }
1507 EXPORT_SYMBOL(netlink_set_nonroot);
1508 
1509 static void netlink_destroy_callback(struct netlink_callback *cb)
1510 {
1511 	if (cb->skb)
1512 		kfree_skb(cb->skb);
1513 	kfree(cb);
1514 }
1515 
1516 /*
1517  * It looks a bit ugly.
1518  * It would be better to create kernel thread.
1519  */
1520 
1521 static int netlink_dump(struct sock *sk)
1522 {
1523 	struct netlink_sock *nlk = nlk_sk(sk);
1524 	struct netlink_callback *cb;
1525 	struct sk_buff *skb;
1526 	struct nlmsghdr *nlh;
1527 	int len, err = -ENOBUFS;
1528 
1529 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1530 	if (!skb)
1531 		goto errout;
1532 
1533 	mutex_lock(nlk->cb_mutex);
1534 
1535 	cb = nlk->cb;
1536 	if (cb == NULL) {
1537 		err = -EINVAL;
1538 		goto errout_skb;
1539 	}
1540 
1541 	len = cb->dump(skb, cb);
1542 
1543 	if (len > 0) {
1544 		mutex_unlock(nlk->cb_mutex);
1545 
1546 		if (sk_filter(sk, skb))
1547 			kfree_skb(skb);
1548 		else {
1549 			skb_queue_tail(&sk->sk_receive_queue, skb);
1550 			sk->sk_data_ready(sk, skb->len);
1551 		}
1552 		return 0;
1553 	}
1554 
1555 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1556 	if (!nlh)
1557 		goto errout_skb;
1558 
1559 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1560 
1561 	if (sk_filter(sk, skb))
1562 		kfree_skb(skb);
1563 	else {
1564 		skb_queue_tail(&sk->sk_receive_queue, skb);
1565 		sk->sk_data_ready(sk, skb->len);
1566 	}
1567 
1568 	if (cb->done)
1569 		cb->done(cb);
1570 	nlk->cb = NULL;
1571 	mutex_unlock(nlk->cb_mutex);
1572 
1573 	netlink_destroy_callback(cb);
1574 	return 0;
1575 
1576 errout_skb:
1577 	mutex_unlock(nlk->cb_mutex);
1578 	kfree_skb(skb);
1579 errout:
1580 	return err;
1581 }
1582 
1583 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1584 		       struct nlmsghdr *nlh,
1585 		       int (*dump)(struct sk_buff *skb,
1586 				   struct netlink_callback *),
1587 		       int (*done)(struct netlink_callback *))
1588 {
1589 	struct netlink_callback *cb;
1590 	struct sock *sk;
1591 	struct netlink_sock *nlk;
1592 
1593 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1594 	if (cb == NULL)
1595 		return -ENOBUFS;
1596 
1597 	cb->dump = dump;
1598 	cb->done = done;
1599 	cb->nlh = nlh;
1600 	atomic_inc(&skb->users);
1601 	cb->skb = skb;
1602 
1603 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1604 	if (sk == NULL) {
1605 		netlink_destroy_callback(cb);
1606 		return -ECONNREFUSED;
1607 	}
1608 	nlk = nlk_sk(sk);
1609 	/* A dump is in progress... */
1610 	mutex_lock(nlk->cb_mutex);
1611 	if (nlk->cb) {
1612 		mutex_unlock(nlk->cb_mutex);
1613 		netlink_destroy_callback(cb);
1614 		sock_put(sk);
1615 		return -EBUSY;
1616 	}
1617 	nlk->cb = cb;
1618 	mutex_unlock(nlk->cb_mutex);
1619 
1620 	netlink_dump(sk);
1621 	sock_put(sk);
1622 
1623 	/* We successfully started a dump, by returning -EINTR we
1624 	 * signal not to send ACK even if it was requested.
1625 	 */
1626 	return -EINTR;
1627 }
1628 EXPORT_SYMBOL(netlink_dump_start);
1629 
1630 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1631 {
1632 	struct sk_buff *skb;
1633 	struct nlmsghdr *rep;
1634 	struct nlmsgerr *errmsg;
1635 	size_t payload = sizeof(*errmsg);
1636 
1637 	/* error messages get the original request appened */
1638 	if (err)
1639 		payload += nlmsg_len(nlh);
1640 
1641 	skb = nlmsg_new(payload, GFP_KERNEL);
1642 	if (!skb) {
1643 		struct sock *sk;
1644 
1645 		sk = netlink_lookup(sock_net(in_skb->sk),
1646 				    in_skb->sk->sk_protocol,
1647 				    NETLINK_CB(in_skb).pid);
1648 		if (sk) {
1649 			sk->sk_err = ENOBUFS;
1650 			sk->sk_error_report(sk);
1651 			sock_put(sk);
1652 		}
1653 		return;
1654 	}
1655 
1656 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1657 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1658 	errmsg = nlmsg_data(rep);
1659 	errmsg->error = err;
1660 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1661 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1662 }
1663 EXPORT_SYMBOL(netlink_ack);
1664 
1665 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1666 						     struct nlmsghdr *))
1667 {
1668 	struct nlmsghdr *nlh;
1669 	int err;
1670 
1671 	while (skb->len >= nlmsg_total_size(0)) {
1672 		int msglen;
1673 
1674 		nlh = nlmsg_hdr(skb);
1675 		err = 0;
1676 
1677 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1678 			return 0;
1679 
1680 		/* Only requests are handled by the kernel */
1681 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1682 			goto ack;
1683 
1684 		/* Skip control messages */
1685 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1686 			goto ack;
1687 
1688 		err = cb(skb, nlh);
1689 		if (err == -EINTR)
1690 			goto skip;
1691 
1692 ack:
1693 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1694 			netlink_ack(skb, nlh, err);
1695 
1696 skip:
1697 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1698 		if (msglen > skb->len)
1699 			msglen = skb->len;
1700 		skb_pull(skb, msglen);
1701 	}
1702 
1703 	return 0;
1704 }
1705 EXPORT_SYMBOL(netlink_rcv_skb);
1706 
1707 /**
1708  * nlmsg_notify - send a notification netlink message
1709  * @sk: netlink socket to use
1710  * @skb: notification message
1711  * @pid: destination netlink pid for reports or 0
1712  * @group: destination multicast group or 0
1713  * @report: 1 to report back, 0 to disable
1714  * @flags: allocation flags
1715  */
1716 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1717 		 unsigned int group, int report, gfp_t flags)
1718 {
1719 	int err = 0;
1720 
1721 	if (group) {
1722 		int exclude_pid = 0;
1723 
1724 		if (report) {
1725 			atomic_inc(&skb->users);
1726 			exclude_pid = pid;
1727 		}
1728 
1729 		/* errors reported via destination sk->sk_err */
1730 		nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1731 	}
1732 
1733 	if (report)
1734 		err = nlmsg_unicast(sk, skb, pid);
1735 
1736 	return err;
1737 }
1738 EXPORT_SYMBOL(nlmsg_notify);
1739 
1740 #ifdef CONFIG_PROC_FS
1741 struct nl_seq_iter {
1742 	struct seq_net_private p;
1743 	int link;
1744 	int hash_idx;
1745 };
1746 
1747 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1748 {
1749 	struct nl_seq_iter *iter = seq->private;
1750 	int i, j;
1751 	struct sock *s;
1752 	struct hlist_node *node;
1753 	loff_t off = 0;
1754 
1755 	for (i = 0; i < MAX_LINKS; i++) {
1756 		struct nl_pid_hash *hash = &nl_table[i].hash;
1757 
1758 		for (j = 0; j <= hash->mask; j++) {
1759 			sk_for_each(s, node, &hash->table[j]) {
1760 				if (sock_net(s) != seq_file_net(seq))
1761 					continue;
1762 				if (off == pos) {
1763 					iter->link = i;
1764 					iter->hash_idx = j;
1765 					return s;
1766 				}
1767 				++off;
1768 			}
1769 		}
1770 	}
1771 	return NULL;
1772 }
1773 
1774 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1775 	__acquires(nl_table_lock)
1776 {
1777 	read_lock(&nl_table_lock);
1778 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1779 }
1780 
1781 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1782 {
1783 	struct sock *s;
1784 	struct nl_seq_iter *iter;
1785 	int i, j;
1786 
1787 	++*pos;
1788 
1789 	if (v == SEQ_START_TOKEN)
1790 		return netlink_seq_socket_idx(seq, 0);
1791 
1792 	iter = seq->private;
1793 	s = v;
1794 	do {
1795 		s = sk_next(s);
1796 	} while (s && sock_net(s) != seq_file_net(seq));
1797 	if (s)
1798 		return s;
1799 
1800 	i = iter->link;
1801 	j = iter->hash_idx + 1;
1802 
1803 	do {
1804 		struct nl_pid_hash *hash = &nl_table[i].hash;
1805 
1806 		for (; j <= hash->mask; j++) {
1807 			s = sk_head(&hash->table[j]);
1808 			while (s && sock_net(s) != seq_file_net(seq))
1809 				s = sk_next(s);
1810 			if (s) {
1811 				iter->link = i;
1812 				iter->hash_idx = j;
1813 				return s;
1814 			}
1815 		}
1816 
1817 		j = 0;
1818 	} while (++i < MAX_LINKS);
1819 
1820 	return NULL;
1821 }
1822 
1823 static void netlink_seq_stop(struct seq_file *seq, void *v)
1824 	__releases(nl_table_lock)
1825 {
1826 	read_unlock(&nl_table_lock);
1827 }
1828 
1829 
1830 static int netlink_seq_show(struct seq_file *seq, void *v)
1831 {
1832 	if (v == SEQ_START_TOKEN)
1833 		seq_puts(seq,
1834 			 "sk       Eth Pid    Groups   "
1835 			 "Rmem     Wmem     Dump     Locks\n");
1836 	else {
1837 		struct sock *s = v;
1838 		struct netlink_sock *nlk = nlk_sk(s);
1839 
1840 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1841 			   s,
1842 			   s->sk_protocol,
1843 			   nlk->pid,
1844 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1845 			   atomic_read(&s->sk_rmem_alloc),
1846 			   atomic_read(&s->sk_wmem_alloc),
1847 			   nlk->cb,
1848 			   atomic_read(&s->sk_refcnt)
1849 			);
1850 
1851 	}
1852 	return 0;
1853 }
1854 
1855 static const struct seq_operations netlink_seq_ops = {
1856 	.start  = netlink_seq_start,
1857 	.next   = netlink_seq_next,
1858 	.stop   = netlink_seq_stop,
1859 	.show   = netlink_seq_show,
1860 };
1861 
1862 
1863 static int netlink_seq_open(struct inode *inode, struct file *file)
1864 {
1865 	return seq_open_net(inode, file, &netlink_seq_ops,
1866 				sizeof(struct nl_seq_iter));
1867 }
1868 
1869 static const struct file_operations netlink_seq_fops = {
1870 	.owner		= THIS_MODULE,
1871 	.open		= netlink_seq_open,
1872 	.read		= seq_read,
1873 	.llseek		= seq_lseek,
1874 	.release	= seq_release_net,
1875 };
1876 
1877 #endif
1878 
1879 int netlink_register_notifier(struct notifier_block *nb)
1880 {
1881 	return atomic_notifier_chain_register(&netlink_chain, nb);
1882 }
1883 EXPORT_SYMBOL(netlink_register_notifier);
1884 
1885 int netlink_unregister_notifier(struct notifier_block *nb)
1886 {
1887 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
1888 }
1889 EXPORT_SYMBOL(netlink_unregister_notifier);
1890 
1891 static const struct proto_ops netlink_ops = {
1892 	.family =	PF_NETLINK,
1893 	.owner =	THIS_MODULE,
1894 	.release =	netlink_release,
1895 	.bind =		netlink_bind,
1896 	.connect =	netlink_connect,
1897 	.socketpair =	sock_no_socketpair,
1898 	.accept =	sock_no_accept,
1899 	.getname =	netlink_getname,
1900 	.poll =		datagram_poll,
1901 	.ioctl =	sock_no_ioctl,
1902 	.listen =	sock_no_listen,
1903 	.shutdown =	sock_no_shutdown,
1904 	.setsockopt =	netlink_setsockopt,
1905 	.getsockopt =	netlink_getsockopt,
1906 	.sendmsg =	netlink_sendmsg,
1907 	.recvmsg =	netlink_recvmsg,
1908 	.mmap =		sock_no_mmap,
1909 	.sendpage =	sock_no_sendpage,
1910 };
1911 
1912 static struct net_proto_family netlink_family_ops = {
1913 	.family = PF_NETLINK,
1914 	.create = netlink_create,
1915 	.owner	= THIS_MODULE,	/* for consistency 8) */
1916 };
1917 
1918 static int __net_init netlink_net_init(struct net *net)
1919 {
1920 #ifdef CONFIG_PROC_FS
1921 	if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1922 		return -ENOMEM;
1923 #endif
1924 	return 0;
1925 }
1926 
1927 static void __net_exit netlink_net_exit(struct net *net)
1928 {
1929 #ifdef CONFIG_PROC_FS
1930 	proc_net_remove(net, "netlink");
1931 #endif
1932 }
1933 
1934 static struct pernet_operations __net_initdata netlink_net_ops = {
1935 	.init = netlink_net_init,
1936 	.exit = netlink_net_exit,
1937 };
1938 
1939 static int __init netlink_proto_init(void)
1940 {
1941 	struct sk_buff *dummy_skb;
1942 	int i;
1943 	unsigned long limit;
1944 	unsigned int order;
1945 	int err = proto_register(&netlink_proto, 0);
1946 
1947 	if (err != 0)
1948 		goto out;
1949 
1950 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1951 
1952 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1953 	if (!nl_table)
1954 		goto panic;
1955 
1956 	if (num_physpages >= (128 * 1024))
1957 		limit = num_physpages >> (21 - PAGE_SHIFT);
1958 	else
1959 		limit = num_physpages >> (23 - PAGE_SHIFT);
1960 
1961 	order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1962 	limit = (1UL << order) / sizeof(struct hlist_head);
1963 	order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1964 
1965 	for (i = 0; i < MAX_LINKS; i++) {
1966 		struct nl_pid_hash *hash = &nl_table[i].hash;
1967 
1968 		hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1969 		if (!hash->table) {
1970 			while (i-- > 0)
1971 				nl_pid_hash_free(nl_table[i].hash.table,
1972 						 1 * sizeof(*hash->table));
1973 			kfree(nl_table);
1974 			goto panic;
1975 		}
1976 		hash->max_shift = order;
1977 		hash->shift = 0;
1978 		hash->mask = 0;
1979 		hash->rehash_time = jiffies;
1980 	}
1981 
1982 	sock_register(&netlink_family_ops);
1983 	register_pernet_subsys(&netlink_net_ops);
1984 	/* The netlink device handler may be needed early. */
1985 	rtnetlink_init();
1986 out:
1987 	return err;
1988 panic:
1989 	panic("netlink_init: Cannot allocate nl_table\n");
1990 }
1991 
1992 core_initcall(netlink_proto_init);
1993