xref: /openbmc/linux/net/netlink/af_netlink.c (revision 64c70b1c)
1 /*
2  * NETLINK      Kernel-user communication protocol.
3  *
4  * 		Authors:	Alan Cox <alan@redhat.com>
5  * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6  *
7  *		This program is free software; you can redistribute it and/or
8  *		modify it under the terms of the GNU General Public License
9  *		as published by the Free Software Foundation; either version
10  *		2 of the License, or (at your option) any later version.
11  *
12  * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13  *                               added netlink_proto_exit
14  * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15  * 				 use nlk_sk, as sk->protinfo is on a diet 8)
16  * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17  * 				 - inc module use count of module that owns
18  * 				   the kernel socket in case userspace opens
19  * 				   socket of same protocol
20  * 				 - remove all module support, since netlink is
21  * 				   mandatory if CONFIG_NET=y these days
22  */
23 
24 #include <linux/module.h>
25 
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/selinux.h>
58 #include <linux/mutex.h>
59 
60 #include <net/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 
66 struct netlink_sock {
67 	/* struct sock has to be the first member of netlink_sock */
68 	struct sock		sk;
69 	u32			pid;
70 	u32			dst_pid;
71 	u32			dst_group;
72 	u32			flags;
73 	u32			subscriptions;
74 	u32			ngroups;
75 	unsigned long		*groups;
76 	unsigned long		state;
77 	wait_queue_head_t	wait;
78 	struct netlink_callback	*cb;
79 	struct mutex		*cb_mutex;
80 	struct mutex		cb_def_mutex;
81 	void			(*data_ready)(struct sock *sk, int bytes);
82 	struct module		*module;
83 };
84 
85 #define NETLINK_KERNEL_SOCKET	0x1
86 #define NETLINK_RECV_PKTINFO	0x2
87 
88 static inline struct netlink_sock *nlk_sk(struct sock *sk)
89 {
90 	return (struct netlink_sock *)sk;
91 }
92 
93 struct nl_pid_hash {
94 	struct hlist_head *table;
95 	unsigned long rehash_time;
96 
97 	unsigned int mask;
98 	unsigned int shift;
99 
100 	unsigned int entries;
101 	unsigned int max_shift;
102 
103 	u32 rnd;
104 };
105 
106 struct netlink_table {
107 	struct nl_pid_hash hash;
108 	struct hlist_head mc_list;
109 	unsigned long *listeners;
110 	unsigned int nl_nonroot;
111 	unsigned int groups;
112 	struct mutex *cb_mutex;
113 	struct module *module;
114 	int registered;
115 };
116 
117 static struct netlink_table *nl_table;
118 
119 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
120 
121 static int netlink_dump(struct sock *sk);
122 static void netlink_destroy_callback(struct netlink_callback *cb);
123 static void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb);
124 
125 static DEFINE_RWLOCK(nl_table_lock);
126 static atomic_t nl_table_users = ATOMIC_INIT(0);
127 
128 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
129 
130 static u32 netlink_group_mask(u32 group)
131 {
132 	return group ? 1 << (group - 1) : 0;
133 }
134 
135 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
136 {
137 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
138 }
139 
140 static void netlink_sock_destruct(struct sock *sk)
141 {
142 	struct netlink_sock *nlk = nlk_sk(sk);
143 
144 	if (nlk->cb) {
145 		if (nlk->cb->done)
146 			nlk->cb->done(nlk->cb);
147 		netlink_destroy_callback(nlk->cb);
148 	}
149 
150 	skb_queue_purge(&sk->sk_receive_queue);
151 
152 	if (!sock_flag(sk, SOCK_DEAD)) {
153 		printk("Freeing alive netlink socket %p\n", sk);
154 		return;
155 	}
156 	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
157 	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
158 	BUG_TRAP(!nlk_sk(sk)->groups);
159 }
160 
161 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
162  * Look, when several writers sleep and reader wakes them up, all but one
163  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
164  * this, _but_ remember, it adds useless work on UP machines.
165  */
166 
167 static void netlink_table_grab(void)
168 {
169 	write_lock_irq(&nl_table_lock);
170 
171 	if (atomic_read(&nl_table_users)) {
172 		DECLARE_WAITQUEUE(wait, current);
173 
174 		add_wait_queue_exclusive(&nl_table_wait, &wait);
175 		for(;;) {
176 			set_current_state(TASK_UNINTERRUPTIBLE);
177 			if (atomic_read(&nl_table_users) == 0)
178 				break;
179 			write_unlock_irq(&nl_table_lock);
180 			schedule();
181 			write_lock_irq(&nl_table_lock);
182 		}
183 
184 		__set_current_state(TASK_RUNNING);
185 		remove_wait_queue(&nl_table_wait, &wait);
186 	}
187 }
188 
189 static __inline__ void netlink_table_ungrab(void)
190 {
191 	write_unlock_irq(&nl_table_lock);
192 	wake_up(&nl_table_wait);
193 }
194 
195 static __inline__ void
196 netlink_lock_table(void)
197 {
198 	/* read_lock() synchronizes us to netlink_table_grab */
199 
200 	read_lock(&nl_table_lock);
201 	atomic_inc(&nl_table_users);
202 	read_unlock(&nl_table_lock);
203 }
204 
205 static __inline__ void
206 netlink_unlock_table(void)
207 {
208 	if (atomic_dec_and_test(&nl_table_users))
209 		wake_up(&nl_table_wait);
210 }
211 
212 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
213 {
214 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
215 	struct hlist_head *head;
216 	struct sock *sk;
217 	struct hlist_node *node;
218 
219 	read_lock(&nl_table_lock);
220 	head = nl_pid_hashfn(hash, pid);
221 	sk_for_each(sk, node, head) {
222 		if (nlk_sk(sk)->pid == pid) {
223 			sock_hold(sk);
224 			goto found;
225 		}
226 	}
227 	sk = NULL;
228 found:
229 	read_unlock(&nl_table_lock);
230 	return sk;
231 }
232 
233 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
234 {
235 	if (size <= PAGE_SIZE)
236 		return kmalloc(size, GFP_ATOMIC);
237 	else
238 		return (struct hlist_head *)
239 			__get_free_pages(GFP_ATOMIC, get_order(size));
240 }
241 
242 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
243 {
244 	if (size <= PAGE_SIZE)
245 		kfree(table);
246 	else
247 		free_pages((unsigned long)table, get_order(size));
248 }
249 
250 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
251 {
252 	unsigned int omask, mask, shift;
253 	size_t osize, size;
254 	struct hlist_head *otable, *table;
255 	int i;
256 
257 	omask = mask = hash->mask;
258 	osize = size = (mask + 1) * sizeof(*table);
259 	shift = hash->shift;
260 
261 	if (grow) {
262 		if (++shift > hash->max_shift)
263 			return 0;
264 		mask = mask * 2 + 1;
265 		size *= 2;
266 	}
267 
268 	table = nl_pid_hash_alloc(size);
269 	if (!table)
270 		return 0;
271 
272 	memset(table, 0, size);
273 	otable = hash->table;
274 	hash->table = table;
275 	hash->mask = mask;
276 	hash->shift = shift;
277 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
278 
279 	for (i = 0; i <= omask; i++) {
280 		struct sock *sk;
281 		struct hlist_node *node, *tmp;
282 
283 		sk_for_each_safe(sk, node, tmp, &otable[i])
284 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
285 	}
286 
287 	nl_pid_hash_free(otable, osize);
288 	hash->rehash_time = jiffies + 10 * 60 * HZ;
289 	return 1;
290 }
291 
292 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
293 {
294 	int avg = hash->entries >> hash->shift;
295 
296 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
297 		return 1;
298 
299 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
300 		nl_pid_hash_rehash(hash, 0);
301 		return 1;
302 	}
303 
304 	return 0;
305 }
306 
307 static const struct proto_ops netlink_ops;
308 
309 static void
310 netlink_update_listeners(struct sock *sk)
311 {
312 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
313 	struct hlist_node *node;
314 	unsigned long mask;
315 	unsigned int i;
316 
317 	for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
318 		mask = 0;
319 		sk_for_each_bound(sk, node, &tbl->mc_list)
320 			mask |= nlk_sk(sk)->groups[i];
321 		tbl->listeners[i] = mask;
322 	}
323 	/* this function is only called with the netlink table "grabbed", which
324 	 * makes sure updates are visible before bind or setsockopt return. */
325 }
326 
327 static int netlink_insert(struct sock *sk, u32 pid)
328 {
329 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
330 	struct hlist_head *head;
331 	int err = -EADDRINUSE;
332 	struct sock *osk;
333 	struct hlist_node *node;
334 	int len;
335 
336 	netlink_table_grab();
337 	head = nl_pid_hashfn(hash, pid);
338 	len = 0;
339 	sk_for_each(osk, node, head) {
340 		if (nlk_sk(osk)->pid == pid)
341 			break;
342 		len++;
343 	}
344 	if (node)
345 		goto err;
346 
347 	err = -EBUSY;
348 	if (nlk_sk(sk)->pid)
349 		goto err;
350 
351 	err = -ENOMEM;
352 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
353 		goto err;
354 
355 	if (len && nl_pid_hash_dilute(hash, len))
356 		head = nl_pid_hashfn(hash, pid);
357 	hash->entries++;
358 	nlk_sk(sk)->pid = pid;
359 	sk_add_node(sk, head);
360 	err = 0;
361 
362 err:
363 	netlink_table_ungrab();
364 	return err;
365 }
366 
367 static void netlink_remove(struct sock *sk)
368 {
369 	netlink_table_grab();
370 	if (sk_del_node_init(sk))
371 		nl_table[sk->sk_protocol].hash.entries--;
372 	if (nlk_sk(sk)->subscriptions)
373 		__sk_del_bind_node(sk);
374 	netlink_table_ungrab();
375 }
376 
377 static struct proto netlink_proto = {
378 	.name	  = "NETLINK",
379 	.owner	  = THIS_MODULE,
380 	.obj_size = sizeof(struct netlink_sock),
381 };
382 
383 static int __netlink_create(struct socket *sock, struct mutex *cb_mutex,
384 			    int protocol)
385 {
386 	struct sock *sk;
387 	struct netlink_sock *nlk;
388 
389 	sock->ops = &netlink_ops;
390 
391 	sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
392 	if (!sk)
393 		return -ENOMEM;
394 
395 	sock_init_data(sock, sk);
396 
397 	nlk = nlk_sk(sk);
398 	if (cb_mutex)
399 		nlk->cb_mutex = cb_mutex;
400 	else {
401 		nlk->cb_mutex = &nlk->cb_def_mutex;
402 		mutex_init(nlk->cb_mutex);
403 	}
404 	init_waitqueue_head(&nlk->wait);
405 
406 	sk->sk_destruct = netlink_sock_destruct;
407 	sk->sk_protocol = protocol;
408 	return 0;
409 }
410 
411 static int netlink_create(struct socket *sock, int protocol)
412 {
413 	struct module *module = NULL;
414 	struct mutex *cb_mutex;
415 	struct netlink_sock *nlk;
416 	int err = 0;
417 
418 	sock->state = SS_UNCONNECTED;
419 
420 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
421 		return -ESOCKTNOSUPPORT;
422 
423 	if (protocol<0 || protocol >= MAX_LINKS)
424 		return -EPROTONOSUPPORT;
425 
426 	netlink_lock_table();
427 #ifdef CONFIG_KMOD
428 	if (!nl_table[protocol].registered) {
429 		netlink_unlock_table();
430 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
431 		netlink_lock_table();
432 	}
433 #endif
434 	if (nl_table[protocol].registered &&
435 	    try_module_get(nl_table[protocol].module))
436 		module = nl_table[protocol].module;
437 	cb_mutex = nl_table[protocol].cb_mutex;
438 	netlink_unlock_table();
439 
440 	if ((err = __netlink_create(sock, cb_mutex, protocol)) < 0)
441 		goto out_module;
442 
443 	nlk = nlk_sk(sock->sk);
444 	nlk->module = module;
445 out:
446 	return err;
447 
448 out_module:
449 	module_put(module);
450 	goto out;
451 }
452 
453 static int netlink_release(struct socket *sock)
454 {
455 	struct sock *sk = sock->sk;
456 	struct netlink_sock *nlk;
457 
458 	if (!sk)
459 		return 0;
460 
461 	netlink_remove(sk);
462 	sock_orphan(sk);
463 	nlk = nlk_sk(sk);
464 
465 	/*
466 	 * OK. Socket is unlinked, any packets that arrive now
467 	 * will be purged.
468 	 */
469 
470 	sock->sk = NULL;
471 	wake_up_interruptible_all(&nlk->wait);
472 
473 	skb_queue_purge(&sk->sk_write_queue);
474 
475 	if (nlk->pid && !nlk->subscriptions) {
476 		struct netlink_notify n = {
477 						.protocol = sk->sk_protocol,
478 						.pid = nlk->pid,
479 					  };
480 		atomic_notifier_call_chain(&netlink_chain,
481 				NETLINK_URELEASE, &n);
482 	}
483 
484 	module_put(nlk->module);
485 
486 	netlink_table_grab();
487 	if (nlk->flags & NETLINK_KERNEL_SOCKET) {
488 		kfree(nl_table[sk->sk_protocol].listeners);
489 		nl_table[sk->sk_protocol].module = NULL;
490 		nl_table[sk->sk_protocol].registered = 0;
491 	} else if (nlk->subscriptions)
492 		netlink_update_listeners(sk);
493 	netlink_table_ungrab();
494 
495 	kfree(nlk->groups);
496 	nlk->groups = NULL;
497 
498 	sock_put(sk);
499 	return 0;
500 }
501 
502 static int netlink_autobind(struct socket *sock)
503 {
504 	struct sock *sk = sock->sk;
505 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
506 	struct hlist_head *head;
507 	struct sock *osk;
508 	struct hlist_node *node;
509 	s32 pid = current->tgid;
510 	int err;
511 	static s32 rover = -4097;
512 
513 retry:
514 	cond_resched();
515 	netlink_table_grab();
516 	head = nl_pid_hashfn(hash, pid);
517 	sk_for_each(osk, node, head) {
518 		if (nlk_sk(osk)->pid == pid) {
519 			/* Bind collision, search negative pid values. */
520 			pid = rover--;
521 			if (rover > -4097)
522 				rover = -4097;
523 			netlink_table_ungrab();
524 			goto retry;
525 		}
526 	}
527 	netlink_table_ungrab();
528 
529 	err = netlink_insert(sk, pid);
530 	if (err == -EADDRINUSE)
531 		goto retry;
532 
533 	/* If 2 threads race to autobind, that is fine.  */
534 	if (err == -EBUSY)
535 		err = 0;
536 
537 	return err;
538 }
539 
540 static inline int netlink_capable(struct socket *sock, unsigned int flag)
541 {
542 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
543 	       capable(CAP_NET_ADMIN);
544 }
545 
546 static void
547 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
548 {
549 	struct netlink_sock *nlk = nlk_sk(sk);
550 
551 	if (nlk->subscriptions && !subscriptions)
552 		__sk_del_bind_node(sk);
553 	else if (!nlk->subscriptions && subscriptions)
554 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
555 	nlk->subscriptions = subscriptions;
556 }
557 
558 static int netlink_alloc_groups(struct sock *sk)
559 {
560 	struct netlink_sock *nlk = nlk_sk(sk);
561 	unsigned int groups;
562 	int err = 0;
563 
564 	netlink_lock_table();
565 	groups = nl_table[sk->sk_protocol].groups;
566 	if (!nl_table[sk->sk_protocol].registered)
567 		err = -ENOENT;
568 	netlink_unlock_table();
569 
570 	if (err)
571 		return err;
572 
573 	nlk->groups = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
574 	if (nlk->groups == NULL)
575 		return -ENOMEM;
576 	nlk->ngroups = groups;
577 	return 0;
578 }
579 
580 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
581 {
582 	struct sock *sk = sock->sk;
583 	struct netlink_sock *nlk = nlk_sk(sk);
584 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
585 	int err;
586 
587 	if (nladdr->nl_family != AF_NETLINK)
588 		return -EINVAL;
589 
590 	/* Only superuser is allowed to listen multicasts */
591 	if (nladdr->nl_groups) {
592 		if (!netlink_capable(sock, NL_NONROOT_RECV))
593 			return -EPERM;
594 		if (nlk->groups == NULL) {
595 			err = netlink_alloc_groups(sk);
596 			if (err)
597 				return err;
598 		}
599 	}
600 
601 	if (nlk->pid) {
602 		if (nladdr->nl_pid != nlk->pid)
603 			return -EINVAL;
604 	} else {
605 		err = nladdr->nl_pid ?
606 			netlink_insert(sk, nladdr->nl_pid) :
607 			netlink_autobind(sock);
608 		if (err)
609 			return err;
610 	}
611 
612 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
613 		return 0;
614 
615 	netlink_table_grab();
616 	netlink_update_subscriptions(sk, nlk->subscriptions +
617 					 hweight32(nladdr->nl_groups) -
618 					 hweight32(nlk->groups[0]));
619 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
620 	netlink_update_listeners(sk);
621 	netlink_table_ungrab();
622 
623 	return 0;
624 }
625 
626 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
627 			   int alen, int flags)
628 {
629 	int err = 0;
630 	struct sock *sk = sock->sk;
631 	struct netlink_sock *nlk = nlk_sk(sk);
632 	struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
633 
634 	if (addr->sa_family == AF_UNSPEC) {
635 		sk->sk_state	= NETLINK_UNCONNECTED;
636 		nlk->dst_pid	= 0;
637 		nlk->dst_group  = 0;
638 		return 0;
639 	}
640 	if (addr->sa_family != AF_NETLINK)
641 		return -EINVAL;
642 
643 	/* Only superuser is allowed to send multicasts */
644 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
645 		return -EPERM;
646 
647 	if (!nlk->pid)
648 		err = netlink_autobind(sock);
649 
650 	if (err == 0) {
651 		sk->sk_state	= NETLINK_CONNECTED;
652 		nlk->dst_pid 	= nladdr->nl_pid;
653 		nlk->dst_group  = ffs(nladdr->nl_groups);
654 	}
655 
656 	return err;
657 }
658 
659 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
660 {
661 	struct sock *sk = sock->sk;
662 	struct netlink_sock *nlk = nlk_sk(sk);
663 	struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
664 
665 	nladdr->nl_family = AF_NETLINK;
666 	nladdr->nl_pad = 0;
667 	*addr_len = sizeof(*nladdr);
668 
669 	if (peer) {
670 		nladdr->nl_pid = nlk->dst_pid;
671 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
672 	} else {
673 		nladdr->nl_pid = nlk->pid;
674 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
675 	}
676 	return 0;
677 }
678 
679 static void netlink_overrun(struct sock *sk)
680 {
681 	if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
682 		sk->sk_err = ENOBUFS;
683 		sk->sk_error_report(sk);
684 	}
685 }
686 
687 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
688 {
689 	int protocol = ssk->sk_protocol;
690 	struct sock *sock;
691 	struct netlink_sock *nlk;
692 
693 	sock = netlink_lookup(protocol, pid);
694 	if (!sock)
695 		return ERR_PTR(-ECONNREFUSED);
696 
697 	/* Don't bother queuing skb if kernel socket has no input function */
698 	nlk = nlk_sk(sock);
699 	if ((nlk->pid == 0 && !nlk->data_ready) ||
700 	    (sock->sk_state == NETLINK_CONNECTED &&
701 	     nlk->dst_pid != nlk_sk(ssk)->pid)) {
702 		sock_put(sock);
703 		return ERR_PTR(-ECONNREFUSED);
704 	}
705 	return sock;
706 }
707 
708 struct sock *netlink_getsockbyfilp(struct file *filp)
709 {
710 	struct inode *inode = filp->f_path.dentry->d_inode;
711 	struct sock *sock;
712 
713 	if (!S_ISSOCK(inode->i_mode))
714 		return ERR_PTR(-ENOTSOCK);
715 
716 	sock = SOCKET_I(inode)->sk;
717 	if (sock->sk_family != AF_NETLINK)
718 		return ERR_PTR(-EINVAL);
719 
720 	sock_hold(sock);
721 	return sock;
722 }
723 
724 /*
725  * Attach a skb to a netlink socket.
726  * The caller must hold a reference to the destination socket. On error, the
727  * reference is dropped. The skb is not send to the destination, just all
728  * all error checks are performed and memory in the queue is reserved.
729  * Return values:
730  * < 0: error. skb freed, reference to sock dropped.
731  * 0: continue
732  * 1: repeat lookup - reference dropped while waiting for socket memory.
733  */
734 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
735 		long timeo, struct sock *ssk)
736 {
737 	struct netlink_sock *nlk;
738 
739 	nlk = nlk_sk(sk);
740 
741 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
742 	    test_bit(0, &nlk->state)) {
743 		DECLARE_WAITQUEUE(wait, current);
744 		if (!timeo) {
745 			if (!ssk || nlk_sk(ssk)->pid == 0)
746 				netlink_overrun(sk);
747 			sock_put(sk);
748 			kfree_skb(skb);
749 			return -EAGAIN;
750 		}
751 
752 		__set_current_state(TASK_INTERRUPTIBLE);
753 		add_wait_queue(&nlk->wait, &wait);
754 
755 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
756 		     test_bit(0, &nlk->state)) &&
757 		    !sock_flag(sk, SOCK_DEAD))
758 			timeo = schedule_timeout(timeo);
759 
760 		__set_current_state(TASK_RUNNING);
761 		remove_wait_queue(&nlk->wait, &wait);
762 		sock_put(sk);
763 
764 		if (signal_pending(current)) {
765 			kfree_skb(skb);
766 			return sock_intr_errno(timeo);
767 		}
768 		return 1;
769 	}
770 	skb_set_owner_r(skb, sk);
771 	return 0;
772 }
773 
774 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
775 {
776 	int len = skb->len;
777 
778 	skb_queue_tail(&sk->sk_receive_queue, skb);
779 	sk->sk_data_ready(sk, len);
780 	sock_put(sk);
781 	return len;
782 }
783 
784 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
785 {
786 	kfree_skb(skb);
787 	sock_put(sk);
788 }
789 
790 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
791 					   gfp_t allocation)
792 {
793 	int delta;
794 
795 	skb_orphan(skb);
796 
797 	delta = skb->end - skb->tail;
798 	if (delta * 2 < skb->truesize)
799 		return skb;
800 
801 	if (skb_shared(skb)) {
802 		struct sk_buff *nskb = skb_clone(skb, allocation);
803 		if (!nskb)
804 			return skb;
805 		kfree_skb(skb);
806 		skb = nskb;
807 	}
808 
809 	if (!pskb_expand_head(skb, 0, -delta, allocation))
810 		skb->truesize -= delta;
811 
812 	return skb;
813 }
814 
815 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
816 {
817 	struct sock *sk;
818 	int err;
819 	long timeo;
820 
821 	skb = netlink_trim(skb, gfp_any());
822 
823 	timeo = sock_sndtimeo(ssk, nonblock);
824 retry:
825 	sk = netlink_getsockbypid(ssk, pid);
826 	if (IS_ERR(sk)) {
827 		kfree_skb(skb);
828 		return PTR_ERR(sk);
829 	}
830 	err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
831 	if (err == 1)
832 		goto retry;
833 	if (err)
834 		return err;
835 
836 	return netlink_sendskb(sk, skb, ssk->sk_protocol);
837 }
838 
839 int netlink_has_listeners(struct sock *sk, unsigned int group)
840 {
841 	int res = 0;
842 
843 	BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
844 	if (group - 1 < nl_table[sk->sk_protocol].groups)
845 		res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
846 	return res;
847 }
848 EXPORT_SYMBOL_GPL(netlink_has_listeners);
849 
850 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
851 {
852 	struct netlink_sock *nlk = nlk_sk(sk);
853 
854 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
855 	    !test_bit(0, &nlk->state)) {
856 		skb_set_owner_r(skb, sk);
857 		skb_queue_tail(&sk->sk_receive_queue, skb);
858 		sk->sk_data_ready(sk, skb->len);
859 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
860 	}
861 	return -1;
862 }
863 
864 struct netlink_broadcast_data {
865 	struct sock *exclude_sk;
866 	u32 pid;
867 	u32 group;
868 	int failure;
869 	int congested;
870 	int delivered;
871 	gfp_t allocation;
872 	struct sk_buff *skb, *skb2;
873 };
874 
875 static inline int do_one_broadcast(struct sock *sk,
876 				   struct netlink_broadcast_data *p)
877 {
878 	struct netlink_sock *nlk = nlk_sk(sk);
879 	int val;
880 
881 	if (p->exclude_sk == sk)
882 		goto out;
883 
884 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
885 	    !test_bit(p->group - 1, nlk->groups))
886 		goto out;
887 
888 	if (p->failure) {
889 		netlink_overrun(sk);
890 		goto out;
891 	}
892 
893 	sock_hold(sk);
894 	if (p->skb2 == NULL) {
895 		if (skb_shared(p->skb)) {
896 			p->skb2 = skb_clone(p->skb, p->allocation);
897 		} else {
898 			p->skb2 = skb_get(p->skb);
899 			/*
900 			 * skb ownership may have been set when
901 			 * delivered to a previous socket.
902 			 */
903 			skb_orphan(p->skb2);
904 		}
905 	}
906 	if (p->skb2 == NULL) {
907 		netlink_overrun(sk);
908 		/* Clone failed. Notify ALL listeners. */
909 		p->failure = 1;
910 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
911 		netlink_overrun(sk);
912 	} else {
913 		p->congested |= val;
914 		p->delivered = 1;
915 		p->skb2 = NULL;
916 	}
917 	sock_put(sk);
918 
919 out:
920 	return 0;
921 }
922 
923 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
924 		      u32 group, gfp_t allocation)
925 {
926 	struct netlink_broadcast_data info;
927 	struct hlist_node *node;
928 	struct sock *sk;
929 
930 	skb = netlink_trim(skb, allocation);
931 
932 	info.exclude_sk = ssk;
933 	info.pid = pid;
934 	info.group = group;
935 	info.failure = 0;
936 	info.congested = 0;
937 	info.delivered = 0;
938 	info.allocation = allocation;
939 	info.skb = skb;
940 	info.skb2 = NULL;
941 
942 	/* While we sleep in clone, do not allow to change socket list */
943 
944 	netlink_lock_table();
945 
946 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
947 		do_one_broadcast(sk, &info);
948 
949 	kfree_skb(skb);
950 
951 	netlink_unlock_table();
952 
953 	if (info.skb2)
954 		kfree_skb(info.skb2);
955 
956 	if (info.delivered) {
957 		if (info.congested && (allocation & __GFP_WAIT))
958 			yield();
959 		return 0;
960 	}
961 	if (info.failure)
962 		return -ENOBUFS;
963 	return -ESRCH;
964 }
965 
966 struct netlink_set_err_data {
967 	struct sock *exclude_sk;
968 	u32 pid;
969 	u32 group;
970 	int code;
971 };
972 
973 static inline int do_one_set_err(struct sock *sk,
974 				 struct netlink_set_err_data *p)
975 {
976 	struct netlink_sock *nlk = nlk_sk(sk);
977 
978 	if (sk == p->exclude_sk)
979 		goto out;
980 
981 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
982 	    !test_bit(p->group - 1, nlk->groups))
983 		goto out;
984 
985 	sk->sk_err = p->code;
986 	sk->sk_error_report(sk);
987 out:
988 	return 0;
989 }
990 
991 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
992 {
993 	struct netlink_set_err_data info;
994 	struct hlist_node *node;
995 	struct sock *sk;
996 
997 	info.exclude_sk = ssk;
998 	info.pid = pid;
999 	info.group = group;
1000 	info.code = code;
1001 
1002 	read_lock(&nl_table_lock);
1003 
1004 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1005 		do_one_set_err(sk, &info);
1006 
1007 	read_unlock(&nl_table_lock);
1008 }
1009 
1010 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1011 			      char __user *optval, int optlen)
1012 {
1013 	struct sock *sk = sock->sk;
1014 	struct netlink_sock *nlk = nlk_sk(sk);
1015 	int val = 0, err;
1016 
1017 	if (level != SOL_NETLINK)
1018 		return -ENOPROTOOPT;
1019 
1020 	if (optlen >= sizeof(int) &&
1021 	    get_user(val, (int __user *)optval))
1022 		return -EFAULT;
1023 
1024 	switch (optname) {
1025 	case NETLINK_PKTINFO:
1026 		if (val)
1027 			nlk->flags |= NETLINK_RECV_PKTINFO;
1028 		else
1029 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1030 		err = 0;
1031 		break;
1032 	case NETLINK_ADD_MEMBERSHIP:
1033 	case NETLINK_DROP_MEMBERSHIP: {
1034 		unsigned int subscriptions;
1035 		int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1036 
1037 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1038 			return -EPERM;
1039 		if (nlk->groups == NULL) {
1040 			err = netlink_alloc_groups(sk);
1041 			if (err)
1042 				return err;
1043 		}
1044 		if (!val || val - 1 >= nlk->ngroups)
1045 			return -EINVAL;
1046 		netlink_table_grab();
1047 		old = test_bit(val - 1, nlk->groups);
1048 		subscriptions = nlk->subscriptions - old + new;
1049 		if (new)
1050 			__set_bit(val - 1, nlk->groups);
1051 		else
1052 			__clear_bit(val - 1, nlk->groups);
1053 		netlink_update_subscriptions(sk, subscriptions);
1054 		netlink_update_listeners(sk);
1055 		netlink_table_ungrab();
1056 		err = 0;
1057 		break;
1058 	}
1059 	default:
1060 		err = -ENOPROTOOPT;
1061 	}
1062 	return err;
1063 }
1064 
1065 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1066 			      char __user *optval, int __user *optlen)
1067 {
1068 	struct sock *sk = sock->sk;
1069 	struct netlink_sock *nlk = nlk_sk(sk);
1070 	int len, val, err;
1071 
1072 	if (level != SOL_NETLINK)
1073 		return -ENOPROTOOPT;
1074 
1075 	if (get_user(len, optlen))
1076 		return -EFAULT;
1077 	if (len < 0)
1078 		return -EINVAL;
1079 
1080 	switch (optname) {
1081 	case NETLINK_PKTINFO:
1082 		if (len < sizeof(int))
1083 			return -EINVAL;
1084 		len = sizeof(int);
1085 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1086 		if (put_user(len, optlen) ||
1087 		    put_user(val, optval))
1088 			return -EFAULT;
1089 		err = 0;
1090 		break;
1091 	default:
1092 		err = -ENOPROTOOPT;
1093 	}
1094 	return err;
1095 }
1096 
1097 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1098 {
1099 	struct nl_pktinfo info;
1100 
1101 	info.group = NETLINK_CB(skb).dst_group;
1102 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1103 }
1104 
1105 static inline void netlink_rcv_wake(struct sock *sk)
1106 {
1107 	struct netlink_sock *nlk = nlk_sk(sk);
1108 
1109 	if (skb_queue_empty(&sk->sk_receive_queue))
1110 		clear_bit(0, &nlk->state);
1111 	if (!test_bit(0, &nlk->state))
1112 		wake_up_interruptible(&nlk->wait);
1113 }
1114 
1115 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1116 			   struct msghdr *msg, size_t len)
1117 {
1118 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1119 	struct sock *sk = sock->sk;
1120 	struct netlink_sock *nlk = nlk_sk(sk);
1121 	struct sockaddr_nl *addr=msg->msg_name;
1122 	u32 dst_pid;
1123 	u32 dst_group;
1124 	struct sk_buff *skb;
1125 	int err;
1126 	struct scm_cookie scm;
1127 
1128 	if (msg->msg_flags&MSG_OOB)
1129 		return -EOPNOTSUPP;
1130 
1131 	if (NULL == siocb->scm)
1132 		siocb->scm = &scm;
1133 	err = scm_send(sock, msg, siocb->scm);
1134 	if (err < 0)
1135 		return err;
1136 
1137 	if (msg->msg_namelen) {
1138 		if (addr->nl_family != AF_NETLINK)
1139 			return -EINVAL;
1140 		dst_pid = addr->nl_pid;
1141 		dst_group = ffs(addr->nl_groups);
1142 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1143 			return -EPERM;
1144 	} else {
1145 		dst_pid = nlk->dst_pid;
1146 		dst_group = nlk->dst_group;
1147 	}
1148 
1149 	if (!nlk->pid) {
1150 		err = netlink_autobind(sock);
1151 		if (err)
1152 			goto out;
1153 	}
1154 
1155 	err = -EMSGSIZE;
1156 	if (len > sk->sk_sndbuf - 32)
1157 		goto out;
1158 	err = -ENOBUFS;
1159 	skb = alloc_skb(len, GFP_KERNEL);
1160 	if (skb==NULL)
1161 		goto out;
1162 
1163 	NETLINK_CB(skb).pid	= nlk->pid;
1164 	NETLINK_CB(skb).dst_group = dst_group;
1165 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1166 	selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1167 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1168 
1169 	/* What can I do? Netlink is asynchronous, so that
1170 	   we will have to save current capabilities to
1171 	   check them, when this message will be delivered
1172 	   to corresponding kernel module.   --ANK (980802)
1173 	 */
1174 
1175 	err = -EFAULT;
1176 	if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1177 		kfree_skb(skb);
1178 		goto out;
1179 	}
1180 
1181 	err = security_netlink_send(sk, skb);
1182 	if (err) {
1183 		kfree_skb(skb);
1184 		goto out;
1185 	}
1186 
1187 	if (dst_group) {
1188 		atomic_inc(&skb->users);
1189 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1190 	}
1191 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1192 
1193 out:
1194 	return err;
1195 }
1196 
1197 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1198 			   struct msghdr *msg, size_t len,
1199 			   int flags)
1200 {
1201 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1202 	struct scm_cookie scm;
1203 	struct sock *sk = sock->sk;
1204 	struct netlink_sock *nlk = nlk_sk(sk);
1205 	int noblock = flags&MSG_DONTWAIT;
1206 	size_t copied;
1207 	struct sk_buff *skb;
1208 	int err;
1209 
1210 	if (flags&MSG_OOB)
1211 		return -EOPNOTSUPP;
1212 
1213 	copied = 0;
1214 
1215 	skb = skb_recv_datagram(sk,flags,noblock,&err);
1216 	if (skb==NULL)
1217 		goto out;
1218 
1219 	msg->msg_namelen = 0;
1220 
1221 	copied = skb->len;
1222 	if (len < copied) {
1223 		msg->msg_flags |= MSG_TRUNC;
1224 		copied = len;
1225 	}
1226 
1227 	skb_reset_transport_header(skb);
1228 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1229 
1230 	if (msg->msg_name) {
1231 		struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1232 		addr->nl_family = AF_NETLINK;
1233 		addr->nl_pad    = 0;
1234 		addr->nl_pid	= NETLINK_CB(skb).pid;
1235 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1236 		msg->msg_namelen = sizeof(*addr);
1237 	}
1238 
1239 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1240 		netlink_cmsg_recv_pktinfo(msg, skb);
1241 
1242 	if (NULL == siocb->scm) {
1243 		memset(&scm, 0, sizeof(scm));
1244 		siocb->scm = &scm;
1245 	}
1246 	siocb->scm->creds = *NETLINK_CREDS(skb);
1247 	if (flags & MSG_TRUNC)
1248 		copied = skb->len;
1249 	skb_free_datagram(sk, skb);
1250 
1251 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1252 		netlink_dump(sk);
1253 
1254 	scm_recv(sock, msg, siocb->scm, flags);
1255 out:
1256 	netlink_rcv_wake(sk);
1257 	return err ? : copied;
1258 }
1259 
1260 static void netlink_data_ready(struct sock *sk, int len)
1261 {
1262 	struct netlink_sock *nlk = nlk_sk(sk);
1263 
1264 	if (nlk->data_ready)
1265 		nlk->data_ready(sk, len);
1266 	netlink_rcv_wake(sk);
1267 }
1268 
1269 /*
1270  *	We export these functions to other modules. They provide a
1271  *	complete set of kernel non-blocking support for message
1272  *	queueing.
1273  */
1274 
1275 struct sock *
1276 netlink_kernel_create(int unit, unsigned int groups,
1277 		      void (*input)(struct sock *sk, int len),
1278 		      struct mutex *cb_mutex, struct module *module)
1279 {
1280 	struct socket *sock;
1281 	struct sock *sk;
1282 	struct netlink_sock *nlk;
1283 	unsigned long *listeners = NULL;
1284 
1285 	BUG_ON(!nl_table);
1286 
1287 	if (unit<0 || unit>=MAX_LINKS)
1288 		return NULL;
1289 
1290 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1291 		return NULL;
1292 
1293 	if (__netlink_create(sock, cb_mutex, unit) < 0)
1294 		goto out_sock_release;
1295 
1296 	if (groups < 32)
1297 		groups = 32;
1298 
1299 	listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1300 	if (!listeners)
1301 		goto out_sock_release;
1302 
1303 	sk = sock->sk;
1304 	sk->sk_data_ready = netlink_data_ready;
1305 	if (input)
1306 		nlk_sk(sk)->data_ready = input;
1307 
1308 	if (netlink_insert(sk, 0))
1309 		goto out_sock_release;
1310 
1311 	nlk = nlk_sk(sk);
1312 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1313 
1314 	netlink_table_grab();
1315 	nl_table[unit].groups = groups;
1316 	nl_table[unit].listeners = listeners;
1317 	nl_table[unit].cb_mutex = cb_mutex;
1318 	nl_table[unit].module = module;
1319 	nl_table[unit].registered = 1;
1320 	netlink_table_ungrab();
1321 
1322 	return sk;
1323 
1324 out_sock_release:
1325 	kfree(listeners);
1326 	sock_release(sock);
1327 	return NULL;
1328 }
1329 
1330 void netlink_set_nonroot(int protocol, unsigned int flags)
1331 {
1332 	if ((unsigned int)protocol < MAX_LINKS)
1333 		nl_table[protocol].nl_nonroot = flags;
1334 }
1335 
1336 static void netlink_destroy_callback(struct netlink_callback *cb)
1337 {
1338 	if (cb->skb)
1339 		kfree_skb(cb->skb);
1340 	kfree(cb);
1341 }
1342 
1343 /*
1344  * It looks a bit ugly.
1345  * It would be better to create kernel thread.
1346  */
1347 
1348 static int netlink_dump(struct sock *sk)
1349 {
1350 	struct netlink_sock *nlk = nlk_sk(sk);
1351 	struct netlink_callback *cb;
1352 	struct sk_buff *skb;
1353 	struct nlmsghdr *nlh;
1354 	int len, err = -ENOBUFS;
1355 
1356 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1357 	if (!skb)
1358 		goto errout;
1359 
1360 	mutex_lock(nlk->cb_mutex);
1361 
1362 	cb = nlk->cb;
1363 	if (cb == NULL) {
1364 		err = -EINVAL;
1365 		goto errout_skb;
1366 	}
1367 
1368 	len = cb->dump(skb, cb);
1369 
1370 	if (len > 0) {
1371 		mutex_unlock(nlk->cb_mutex);
1372 		skb_queue_tail(&sk->sk_receive_queue, skb);
1373 		sk->sk_data_ready(sk, len);
1374 		return 0;
1375 	}
1376 
1377 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1378 	if (!nlh)
1379 		goto errout_skb;
1380 
1381 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1382 
1383 	skb_queue_tail(&sk->sk_receive_queue, skb);
1384 	sk->sk_data_ready(sk, skb->len);
1385 
1386 	if (cb->done)
1387 		cb->done(cb);
1388 	nlk->cb = NULL;
1389 	mutex_unlock(nlk->cb_mutex);
1390 
1391 	netlink_destroy_callback(cb);
1392 	return 0;
1393 
1394 errout_skb:
1395 	mutex_unlock(nlk->cb_mutex);
1396 	kfree_skb(skb);
1397 errout:
1398 	return err;
1399 }
1400 
1401 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1402 		       struct nlmsghdr *nlh,
1403 		       int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1404 		       int (*done)(struct netlink_callback*))
1405 {
1406 	struct netlink_callback *cb;
1407 	struct sock *sk;
1408 	struct netlink_sock *nlk;
1409 
1410 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1411 	if (cb == NULL)
1412 		return -ENOBUFS;
1413 
1414 	cb->dump = dump;
1415 	cb->done = done;
1416 	cb->nlh = nlh;
1417 	atomic_inc(&skb->users);
1418 	cb->skb = skb;
1419 
1420 	sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1421 	if (sk == NULL) {
1422 		netlink_destroy_callback(cb);
1423 		return -ECONNREFUSED;
1424 	}
1425 	nlk = nlk_sk(sk);
1426 	/* A dump is in progress... */
1427 	mutex_lock(nlk->cb_mutex);
1428 	if (nlk->cb) {
1429 		mutex_unlock(nlk->cb_mutex);
1430 		netlink_destroy_callback(cb);
1431 		sock_put(sk);
1432 		return -EBUSY;
1433 	}
1434 	nlk->cb = cb;
1435 	mutex_unlock(nlk->cb_mutex);
1436 
1437 	netlink_dump(sk);
1438 	sock_put(sk);
1439 
1440 	/* We successfully started a dump, by returning -EINTR we
1441 	 * signal the queue mangement to interrupt processing of
1442 	 * any netlink messages so userspace gets a chance to read
1443 	 * the results. */
1444 	return -EINTR;
1445 }
1446 
1447 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1448 {
1449 	struct sk_buff *skb;
1450 	struct nlmsghdr *rep;
1451 	struct nlmsgerr *errmsg;
1452 	size_t payload = sizeof(*errmsg);
1453 
1454 	/* error messages get the original request appened */
1455 	if (err)
1456 		payload += nlmsg_len(nlh);
1457 
1458 	skb = nlmsg_new(payload, GFP_KERNEL);
1459 	if (!skb) {
1460 		struct sock *sk;
1461 
1462 		sk = netlink_lookup(in_skb->sk->sk_protocol,
1463 				    NETLINK_CB(in_skb).pid);
1464 		if (sk) {
1465 			sk->sk_err = ENOBUFS;
1466 			sk->sk_error_report(sk);
1467 			sock_put(sk);
1468 		}
1469 		return;
1470 	}
1471 
1472 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1473 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1474 	errmsg = nlmsg_data(rep);
1475 	errmsg->error = err;
1476 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1477 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1478 }
1479 
1480 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1481 						     struct nlmsghdr *))
1482 {
1483 	struct nlmsghdr *nlh;
1484 	int err;
1485 
1486 	while (skb->len >= nlmsg_total_size(0)) {
1487 		nlh = nlmsg_hdr(skb);
1488 		err = 0;
1489 
1490 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1491 			return 0;
1492 
1493 		/* Only requests are handled by the kernel */
1494 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1495 			goto skip;
1496 
1497 		/* Skip control messages */
1498 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1499 			goto skip;
1500 
1501 		err = cb(skb, nlh);
1502 		if (err == -EINTR) {
1503 			/* Not an error, but we interrupt processing */
1504 			netlink_queue_skip(nlh, skb);
1505 			return err;
1506 		}
1507 skip:
1508 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
1509 			netlink_ack(skb, nlh, err);
1510 
1511 		netlink_queue_skip(nlh, skb);
1512 	}
1513 
1514 	return 0;
1515 }
1516 
1517 /**
1518  * nelink_run_queue - Process netlink receive queue.
1519  * @sk: Netlink socket containing the queue
1520  * @qlen: Place to store queue length upon entry
1521  * @cb: Callback function invoked for each netlink message found
1522  *
1523  * Processes as much as there was in the queue upon entry and invokes
1524  * a callback function for each netlink message found. The callback
1525  * function may refuse a message by returning a negative error code
1526  * but setting the error pointer to 0 in which case this function
1527  * returns with a qlen != 0.
1528  *
1529  * qlen must be initialized to 0 before the initial entry, afterwards
1530  * the function may be called repeatedly until qlen reaches 0.
1531  *
1532  * The callback function may return -EINTR to signal that processing
1533  * of netlink messages shall be interrupted. In this case the message
1534  * currently being processed will NOT be requeued onto the receive
1535  * queue.
1536  */
1537 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1538 		       int (*cb)(struct sk_buff *, struct nlmsghdr *))
1539 {
1540 	struct sk_buff *skb;
1541 
1542 	if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1543 		*qlen = skb_queue_len(&sk->sk_receive_queue);
1544 
1545 	for (; *qlen; (*qlen)--) {
1546 		skb = skb_dequeue(&sk->sk_receive_queue);
1547 		if (netlink_rcv_skb(skb, cb)) {
1548 			if (skb->len)
1549 				skb_queue_head(&sk->sk_receive_queue, skb);
1550 			else {
1551 				kfree_skb(skb);
1552 				(*qlen)--;
1553 			}
1554 			break;
1555 		}
1556 
1557 		kfree_skb(skb);
1558 	}
1559 }
1560 
1561 /**
1562  * netlink_queue_skip - Skip netlink message while processing queue.
1563  * @nlh: Netlink message to be skipped
1564  * @skb: Socket buffer containing the netlink messages.
1565  *
1566  * Pulls the given netlink message off the socket buffer so the next
1567  * call to netlink_queue_run() will not reconsider the message.
1568  */
1569 static void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1570 {
1571 	int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1572 
1573 	if (msglen > skb->len)
1574 		msglen = skb->len;
1575 
1576 	skb_pull(skb, msglen);
1577 }
1578 
1579 /**
1580  * nlmsg_notify - send a notification netlink message
1581  * @sk: netlink socket to use
1582  * @skb: notification message
1583  * @pid: destination netlink pid for reports or 0
1584  * @group: destination multicast group or 0
1585  * @report: 1 to report back, 0 to disable
1586  * @flags: allocation flags
1587  */
1588 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1589 		 unsigned int group, int report, gfp_t flags)
1590 {
1591 	int err = 0;
1592 
1593 	if (group) {
1594 		int exclude_pid = 0;
1595 
1596 		if (report) {
1597 			atomic_inc(&skb->users);
1598 			exclude_pid = pid;
1599 		}
1600 
1601 		/* errors reported via destination sk->sk_err */
1602 		nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1603 	}
1604 
1605 	if (report)
1606 		err = nlmsg_unicast(sk, skb, pid);
1607 
1608 	return err;
1609 }
1610 
1611 #ifdef CONFIG_PROC_FS
1612 struct nl_seq_iter {
1613 	int link;
1614 	int hash_idx;
1615 };
1616 
1617 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1618 {
1619 	struct nl_seq_iter *iter = seq->private;
1620 	int i, j;
1621 	struct sock *s;
1622 	struct hlist_node *node;
1623 	loff_t off = 0;
1624 
1625 	for (i=0; i<MAX_LINKS; i++) {
1626 		struct nl_pid_hash *hash = &nl_table[i].hash;
1627 
1628 		for (j = 0; j <= hash->mask; j++) {
1629 			sk_for_each(s, node, &hash->table[j]) {
1630 				if (off == pos) {
1631 					iter->link = i;
1632 					iter->hash_idx = j;
1633 					return s;
1634 				}
1635 				++off;
1636 			}
1637 		}
1638 	}
1639 	return NULL;
1640 }
1641 
1642 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1643 {
1644 	read_lock(&nl_table_lock);
1645 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1646 }
1647 
1648 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1649 {
1650 	struct sock *s;
1651 	struct nl_seq_iter *iter;
1652 	int i, j;
1653 
1654 	++*pos;
1655 
1656 	if (v == SEQ_START_TOKEN)
1657 		return netlink_seq_socket_idx(seq, 0);
1658 
1659 	s = sk_next(v);
1660 	if (s)
1661 		return s;
1662 
1663 	iter = seq->private;
1664 	i = iter->link;
1665 	j = iter->hash_idx + 1;
1666 
1667 	do {
1668 		struct nl_pid_hash *hash = &nl_table[i].hash;
1669 
1670 		for (; j <= hash->mask; j++) {
1671 			s = sk_head(&hash->table[j]);
1672 			if (s) {
1673 				iter->link = i;
1674 				iter->hash_idx = j;
1675 				return s;
1676 			}
1677 		}
1678 
1679 		j = 0;
1680 	} while (++i < MAX_LINKS);
1681 
1682 	return NULL;
1683 }
1684 
1685 static void netlink_seq_stop(struct seq_file *seq, void *v)
1686 {
1687 	read_unlock(&nl_table_lock);
1688 }
1689 
1690 
1691 static int netlink_seq_show(struct seq_file *seq, void *v)
1692 {
1693 	if (v == SEQ_START_TOKEN)
1694 		seq_puts(seq,
1695 			 "sk       Eth Pid    Groups   "
1696 			 "Rmem     Wmem     Dump     Locks\n");
1697 	else {
1698 		struct sock *s = v;
1699 		struct netlink_sock *nlk = nlk_sk(s);
1700 
1701 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1702 			   s,
1703 			   s->sk_protocol,
1704 			   nlk->pid,
1705 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1706 			   atomic_read(&s->sk_rmem_alloc),
1707 			   atomic_read(&s->sk_wmem_alloc),
1708 			   nlk->cb,
1709 			   atomic_read(&s->sk_refcnt)
1710 			);
1711 
1712 	}
1713 	return 0;
1714 }
1715 
1716 static struct seq_operations netlink_seq_ops = {
1717 	.start  = netlink_seq_start,
1718 	.next   = netlink_seq_next,
1719 	.stop   = netlink_seq_stop,
1720 	.show   = netlink_seq_show,
1721 };
1722 
1723 
1724 static int netlink_seq_open(struct inode *inode, struct file *file)
1725 {
1726 	struct seq_file *seq;
1727 	struct nl_seq_iter *iter;
1728 	int err;
1729 
1730 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1731 	if (!iter)
1732 		return -ENOMEM;
1733 
1734 	err = seq_open(file, &netlink_seq_ops);
1735 	if (err) {
1736 		kfree(iter);
1737 		return err;
1738 	}
1739 
1740 	seq = file->private_data;
1741 	seq->private = iter;
1742 	return 0;
1743 }
1744 
1745 static const struct file_operations netlink_seq_fops = {
1746 	.owner		= THIS_MODULE,
1747 	.open		= netlink_seq_open,
1748 	.read		= seq_read,
1749 	.llseek		= seq_lseek,
1750 	.release	= seq_release_private,
1751 };
1752 
1753 #endif
1754 
1755 int netlink_register_notifier(struct notifier_block *nb)
1756 {
1757 	return atomic_notifier_chain_register(&netlink_chain, nb);
1758 }
1759 
1760 int netlink_unregister_notifier(struct notifier_block *nb)
1761 {
1762 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
1763 }
1764 
1765 static const struct proto_ops netlink_ops = {
1766 	.family =	PF_NETLINK,
1767 	.owner =	THIS_MODULE,
1768 	.release =	netlink_release,
1769 	.bind =		netlink_bind,
1770 	.connect =	netlink_connect,
1771 	.socketpair =	sock_no_socketpair,
1772 	.accept =	sock_no_accept,
1773 	.getname =	netlink_getname,
1774 	.poll =		datagram_poll,
1775 	.ioctl =	sock_no_ioctl,
1776 	.listen =	sock_no_listen,
1777 	.shutdown =	sock_no_shutdown,
1778 	.setsockopt =	netlink_setsockopt,
1779 	.getsockopt =	netlink_getsockopt,
1780 	.sendmsg =	netlink_sendmsg,
1781 	.recvmsg =	netlink_recvmsg,
1782 	.mmap =		sock_no_mmap,
1783 	.sendpage =	sock_no_sendpage,
1784 };
1785 
1786 static struct net_proto_family netlink_family_ops = {
1787 	.family = PF_NETLINK,
1788 	.create = netlink_create,
1789 	.owner	= THIS_MODULE,	/* for consistency 8) */
1790 };
1791 
1792 static int __init netlink_proto_init(void)
1793 {
1794 	struct sk_buff *dummy_skb;
1795 	int i;
1796 	unsigned long max;
1797 	unsigned int order;
1798 	int err = proto_register(&netlink_proto, 0);
1799 
1800 	if (err != 0)
1801 		goto out;
1802 
1803 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1804 
1805 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1806 	if (!nl_table)
1807 		goto panic;
1808 
1809 	if (num_physpages >= (128 * 1024))
1810 		max = num_physpages >> (21 - PAGE_SHIFT);
1811 	else
1812 		max = num_physpages >> (23 - PAGE_SHIFT);
1813 
1814 	order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1815 	max = (1UL << order) / sizeof(struct hlist_head);
1816 	order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1817 
1818 	for (i = 0; i < MAX_LINKS; i++) {
1819 		struct nl_pid_hash *hash = &nl_table[i].hash;
1820 
1821 		hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1822 		if (!hash->table) {
1823 			while (i-- > 0)
1824 				nl_pid_hash_free(nl_table[i].hash.table,
1825 						 1 * sizeof(*hash->table));
1826 			kfree(nl_table);
1827 			goto panic;
1828 		}
1829 		memset(hash->table, 0, 1 * sizeof(*hash->table));
1830 		hash->max_shift = order;
1831 		hash->shift = 0;
1832 		hash->mask = 0;
1833 		hash->rehash_time = jiffies;
1834 	}
1835 
1836 	sock_register(&netlink_family_ops);
1837 #ifdef CONFIG_PROC_FS
1838 	proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1839 #endif
1840 	/* The netlink device handler may be needed early. */
1841 	rtnetlink_init();
1842 out:
1843 	return err;
1844 panic:
1845 	panic("netlink_init: Cannot allocate nl_table\n");
1846 }
1847 
1848 core_initcall(netlink_proto_init);
1849 
1850 EXPORT_SYMBOL(netlink_ack);
1851 EXPORT_SYMBOL(netlink_run_queue);
1852 EXPORT_SYMBOL(netlink_broadcast);
1853 EXPORT_SYMBOL(netlink_dump_start);
1854 EXPORT_SYMBOL(netlink_kernel_create);
1855 EXPORT_SYMBOL(netlink_register_notifier);
1856 EXPORT_SYMBOL(netlink_set_nonroot);
1857 EXPORT_SYMBOL(netlink_unicast);
1858 EXPORT_SYMBOL(netlink_unregister_notifier);
1859 EXPORT_SYMBOL(nlmsg_notify);
1860