xref: /linux/net/netlink/af_netlink.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
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/smp_lock.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
55 #include <linux/mm.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/selinux.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 	spinlock_t		cb_lock;
80 	void			(*data_ready)(struct sock *sk, int bytes);
81 	struct module		*module;
82 };
83 
84 #define NETLINK_KERNEL_SOCKET	0x1
85 #define NETLINK_RECV_PKTINFO	0x2
86 
87 static inline struct netlink_sock *nlk_sk(struct sock *sk)
88 {
89 	return (struct netlink_sock *)sk;
90 }
91 
92 struct nl_pid_hash {
93 	struct hlist_head *table;
94 	unsigned long rehash_time;
95 
96 	unsigned int mask;
97 	unsigned int shift;
98 
99 	unsigned int entries;
100 	unsigned int max_shift;
101 
102 	u32 rnd;
103 };
104 
105 struct netlink_table {
106 	struct nl_pid_hash hash;
107 	struct hlist_head mc_list;
108 	unsigned long *listeners;
109 	unsigned int nl_nonroot;
110 	unsigned int groups;
111 	struct module *module;
112 	int registered;
113 };
114 
115 static struct netlink_table *nl_table;
116 
117 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
118 
119 static int netlink_dump(struct sock *sk);
120 static void netlink_destroy_callback(struct netlink_callback *cb);
121 
122 static DEFINE_RWLOCK(nl_table_lock);
123 static atomic_t nl_table_users = ATOMIC_INIT(0);
124 
125 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
126 
127 static u32 netlink_group_mask(u32 group)
128 {
129 	return group ? 1 << (group - 1) : 0;
130 }
131 
132 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
133 {
134 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
135 }
136 
137 static void netlink_sock_destruct(struct sock *sk)
138 {
139 	skb_queue_purge(&sk->sk_receive_queue);
140 
141 	if (!sock_flag(sk, SOCK_DEAD)) {
142 		printk("Freeing alive netlink socket %p\n", sk);
143 		return;
144 	}
145 	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
146 	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
147 	BUG_TRAP(!nlk_sk(sk)->cb);
148 	BUG_TRAP(!nlk_sk(sk)->groups);
149 }
150 
151 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
152  * Look, when several writers sleep and reader wakes them up, all but one
153  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
154  * this, _but_ remember, it adds useless work on UP machines.
155  */
156 
157 static void netlink_table_grab(void)
158 {
159 	write_lock_irq(&nl_table_lock);
160 
161 	if (atomic_read(&nl_table_users)) {
162 		DECLARE_WAITQUEUE(wait, current);
163 
164 		add_wait_queue_exclusive(&nl_table_wait, &wait);
165 		for(;;) {
166 			set_current_state(TASK_UNINTERRUPTIBLE);
167 			if (atomic_read(&nl_table_users) == 0)
168 				break;
169 			write_unlock_irq(&nl_table_lock);
170 			schedule();
171 			write_lock_irq(&nl_table_lock);
172 		}
173 
174 		__set_current_state(TASK_RUNNING);
175 		remove_wait_queue(&nl_table_wait, &wait);
176 	}
177 }
178 
179 static __inline__ void netlink_table_ungrab(void)
180 {
181 	write_unlock_irq(&nl_table_lock);
182 	wake_up(&nl_table_wait);
183 }
184 
185 static __inline__ void
186 netlink_lock_table(void)
187 {
188 	/* read_lock() synchronizes us to netlink_table_grab */
189 
190 	read_lock(&nl_table_lock);
191 	atomic_inc(&nl_table_users);
192 	read_unlock(&nl_table_lock);
193 }
194 
195 static __inline__ void
196 netlink_unlock_table(void)
197 {
198 	if (atomic_dec_and_test(&nl_table_users))
199 		wake_up(&nl_table_wait);
200 }
201 
202 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
203 {
204 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
205 	struct hlist_head *head;
206 	struct sock *sk;
207 	struct hlist_node *node;
208 
209 	read_lock(&nl_table_lock);
210 	head = nl_pid_hashfn(hash, pid);
211 	sk_for_each(sk, node, head) {
212 		if (nlk_sk(sk)->pid == pid) {
213 			sock_hold(sk);
214 			goto found;
215 		}
216 	}
217 	sk = NULL;
218 found:
219 	read_unlock(&nl_table_lock);
220 	return sk;
221 }
222 
223 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
224 {
225 	if (size <= PAGE_SIZE)
226 		return kmalloc(size, GFP_ATOMIC);
227 	else
228 		return (struct hlist_head *)
229 			__get_free_pages(GFP_ATOMIC, get_order(size));
230 }
231 
232 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
233 {
234 	if (size <= PAGE_SIZE)
235 		kfree(table);
236 	else
237 		free_pages((unsigned long)table, get_order(size));
238 }
239 
240 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
241 {
242 	unsigned int omask, mask, shift;
243 	size_t osize, size;
244 	struct hlist_head *otable, *table;
245 	int i;
246 
247 	omask = mask = hash->mask;
248 	osize = size = (mask + 1) * sizeof(*table);
249 	shift = hash->shift;
250 
251 	if (grow) {
252 		if (++shift > hash->max_shift)
253 			return 0;
254 		mask = mask * 2 + 1;
255 		size *= 2;
256 	}
257 
258 	table = nl_pid_hash_alloc(size);
259 	if (!table)
260 		return 0;
261 
262 	memset(table, 0, size);
263 	otable = hash->table;
264 	hash->table = table;
265 	hash->mask = mask;
266 	hash->shift = shift;
267 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
268 
269 	for (i = 0; i <= omask; i++) {
270 		struct sock *sk;
271 		struct hlist_node *node, *tmp;
272 
273 		sk_for_each_safe(sk, node, tmp, &otable[i])
274 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
275 	}
276 
277 	nl_pid_hash_free(otable, osize);
278 	hash->rehash_time = jiffies + 10 * 60 * HZ;
279 	return 1;
280 }
281 
282 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
283 {
284 	int avg = hash->entries >> hash->shift;
285 
286 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
287 		return 1;
288 
289 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
290 		nl_pid_hash_rehash(hash, 0);
291 		return 1;
292 	}
293 
294 	return 0;
295 }
296 
297 static const struct proto_ops netlink_ops;
298 
299 static void
300 netlink_update_listeners(struct sock *sk)
301 {
302 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
303 	struct hlist_node *node;
304 	unsigned long mask;
305 	unsigned int i;
306 
307 	for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
308 		mask = 0;
309 		sk_for_each_bound(sk, node, &tbl->mc_list)
310 			mask |= nlk_sk(sk)->groups[i];
311 		tbl->listeners[i] = mask;
312 	}
313 	/* this function is only called with the netlink table "grabbed", which
314 	 * makes sure updates are visible before bind or setsockopt return. */
315 }
316 
317 static int netlink_insert(struct sock *sk, u32 pid)
318 {
319 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
320 	struct hlist_head *head;
321 	int err = -EADDRINUSE;
322 	struct sock *osk;
323 	struct hlist_node *node;
324 	int len;
325 
326 	netlink_table_grab();
327 	head = nl_pid_hashfn(hash, pid);
328 	len = 0;
329 	sk_for_each(osk, node, head) {
330 		if (nlk_sk(osk)->pid == pid)
331 			break;
332 		len++;
333 	}
334 	if (node)
335 		goto err;
336 
337 	err = -EBUSY;
338 	if (nlk_sk(sk)->pid)
339 		goto err;
340 
341 	err = -ENOMEM;
342 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
343 		goto err;
344 
345 	if (len && nl_pid_hash_dilute(hash, len))
346 		head = nl_pid_hashfn(hash, pid);
347 	hash->entries++;
348 	nlk_sk(sk)->pid = pid;
349 	sk_add_node(sk, head);
350 	err = 0;
351 
352 err:
353 	netlink_table_ungrab();
354 	return err;
355 }
356 
357 static void netlink_remove(struct sock *sk)
358 {
359 	netlink_table_grab();
360 	if (sk_del_node_init(sk))
361 		nl_table[sk->sk_protocol].hash.entries--;
362 	if (nlk_sk(sk)->subscriptions)
363 		__sk_del_bind_node(sk);
364 	netlink_table_ungrab();
365 }
366 
367 static struct proto netlink_proto = {
368 	.name	  = "NETLINK",
369 	.owner	  = THIS_MODULE,
370 	.obj_size = sizeof(struct netlink_sock),
371 };
372 
373 static int __netlink_create(struct socket *sock, int protocol)
374 {
375 	struct sock *sk;
376 	struct netlink_sock *nlk;
377 
378 	sock->ops = &netlink_ops;
379 
380 	sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
381 	if (!sk)
382 		return -ENOMEM;
383 
384 	sock_init_data(sock, sk);
385 
386 	nlk = nlk_sk(sk);
387 	spin_lock_init(&nlk->cb_lock);
388 	init_waitqueue_head(&nlk->wait);
389 
390 	sk->sk_destruct = netlink_sock_destruct;
391 	sk->sk_protocol = protocol;
392 	return 0;
393 }
394 
395 static int netlink_create(struct socket *sock, int protocol)
396 {
397 	struct module *module = NULL;
398 	struct netlink_sock *nlk;
399 	unsigned int groups;
400 	int err = 0;
401 
402 	sock->state = SS_UNCONNECTED;
403 
404 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
405 		return -ESOCKTNOSUPPORT;
406 
407 	if (protocol<0 || protocol >= MAX_LINKS)
408 		return -EPROTONOSUPPORT;
409 
410 	netlink_lock_table();
411 #ifdef CONFIG_KMOD
412 	if (!nl_table[protocol].registered) {
413 		netlink_unlock_table();
414 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
415 		netlink_lock_table();
416 	}
417 #endif
418 	if (nl_table[protocol].registered &&
419 	    try_module_get(nl_table[protocol].module))
420 		module = nl_table[protocol].module;
421 	groups = nl_table[protocol].groups;
422 	netlink_unlock_table();
423 
424 	if ((err = __netlink_create(sock, protocol)) < 0)
425 		goto out_module;
426 
427 	nlk = nlk_sk(sock->sk);
428 	nlk->module = module;
429 out:
430 	return err;
431 
432 out_module:
433 	module_put(module);
434 	goto out;
435 }
436 
437 static int netlink_release(struct socket *sock)
438 {
439 	struct sock *sk = sock->sk;
440 	struct netlink_sock *nlk;
441 
442 	if (!sk)
443 		return 0;
444 
445 	netlink_remove(sk);
446 	nlk = nlk_sk(sk);
447 
448 	spin_lock(&nlk->cb_lock);
449 	if (nlk->cb) {
450 		if (nlk->cb->done)
451 			nlk->cb->done(nlk->cb);
452 		netlink_destroy_callback(nlk->cb);
453 		nlk->cb = NULL;
454 	}
455 	spin_unlock(&nlk->cb_lock);
456 
457 	/* OK. Socket is unlinked, and, therefore,
458 	   no new packets will arrive */
459 
460 	sock_orphan(sk);
461 	sock->sk = NULL;
462 	wake_up_interruptible_all(&nlk->wait);
463 
464 	skb_queue_purge(&sk->sk_write_queue);
465 
466 	if (nlk->pid && !nlk->subscriptions) {
467 		struct netlink_notify n = {
468 						.protocol = sk->sk_protocol,
469 						.pid = nlk->pid,
470 					  };
471 		atomic_notifier_call_chain(&netlink_chain,
472 				NETLINK_URELEASE, &n);
473 	}
474 
475 	module_put(nlk->module);
476 
477 	netlink_table_grab();
478 	if (nlk->flags & NETLINK_KERNEL_SOCKET) {
479 		kfree(nl_table[sk->sk_protocol].listeners);
480 		nl_table[sk->sk_protocol].module = NULL;
481 		nl_table[sk->sk_protocol].registered = 0;
482 	} else if (nlk->subscriptions)
483 		netlink_update_listeners(sk);
484 	netlink_table_ungrab();
485 
486 	kfree(nlk->groups);
487 	nlk->groups = NULL;
488 
489 	sock_put(sk);
490 	return 0;
491 }
492 
493 static int netlink_autobind(struct socket *sock)
494 {
495 	struct sock *sk = sock->sk;
496 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
497 	struct hlist_head *head;
498 	struct sock *osk;
499 	struct hlist_node *node;
500 	s32 pid = current->tgid;
501 	int err;
502 	static s32 rover = -4097;
503 
504 retry:
505 	cond_resched();
506 	netlink_table_grab();
507 	head = nl_pid_hashfn(hash, pid);
508 	sk_for_each(osk, node, head) {
509 		if (nlk_sk(osk)->pid == pid) {
510 			/* Bind collision, search negative pid values. */
511 			pid = rover--;
512 			if (rover > -4097)
513 				rover = -4097;
514 			netlink_table_ungrab();
515 			goto retry;
516 		}
517 	}
518 	netlink_table_ungrab();
519 
520 	err = netlink_insert(sk, pid);
521 	if (err == -EADDRINUSE)
522 		goto retry;
523 
524 	/* If 2 threads race to autobind, that is fine.  */
525 	if (err == -EBUSY)
526 		err = 0;
527 
528 	return err;
529 }
530 
531 static inline int netlink_capable(struct socket *sock, unsigned int flag)
532 {
533 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
534 	       capable(CAP_NET_ADMIN);
535 }
536 
537 static void
538 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
539 {
540 	struct netlink_sock *nlk = nlk_sk(sk);
541 
542 	if (nlk->subscriptions && !subscriptions)
543 		__sk_del_bind_node(sk);
544 	else if (!nlk->subscriptions && subscriptions)
545 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
546 	nlk->subscriptions = subscriptions;
547 }
548 
549 static int netlink_alloc_groups(struct sock *sk)
550 {
551 	struct netlink_sock *nlk = nlk_sk(sk);
552 	unsigned int groups;
553 	int err = 0;
554 
555 	netlink_lock_table();
556 	groups = nl_table[sk->sk_protocol].groups;
557 	if (!nl_table[sk->sk_protocol].registered)
558 		err = -ENOENT;
559 	netlink_unlock_table();
560 
561 	if (err)
562 		return err;
563 
564 	nlk->groups = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
565 	if (nlk->groups == NULL)
566 		return -ENOMEM;
567 	nlk->ngroups = groups;
568 	return 0;
569 }
570 
571 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
572 {
573 	struct sock *sk = sock->sk;
574 	struct netlink_sock *nlk = nlk_sk(sk);
575 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
576 	int err;
577 
578 	if (nladdr->nl_family != AF_NETLINK)
579 		return -EINVAL;
580 
581 	/* Only superuser is allowed to listen multicasts */
582 	if (nladdr->nl_groups) {
583 		if (!netlink_capable(sock, NL_NONROOT_RECV))
584 			return -EPERM;
585 		if (nlk->groups == NULL) {
586 			err = netlink_alloc_groups(sk);
587 			if (err)
588 				return err;
589 		}
590 	}
591 
592 	if (nlk->pid) {
593 		if (nladdr->nl_pid != nlk->pid)
594 			return -EINVAL;
595 	} else {
596 		err = nladdr->nl_pid ?
597 			netlink_insert(sk, nladdr->nl_pid) :
598 			netlink_autobind(sock);
599 		if (err)
600 			return err;
601 	}
602 
603 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
604 		return 0;
605 
606 	netlink_table_grab();
607 	netlink_update_subscriptions(sk, nlk->subscriptions +
608 					 hweight32(nladdr->nl_groups) -
609 					 hweight32(nlk->groups[0]));
610 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
611 	netlink_update_listeners(sk);
612 	netlink_table_ungrab();
613 
614 	return 0;
615 }
616 
617 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
618 			   int alen, int flags)
619 {
620 	int err = 0;
621 	struct sock *sk = sock->sk;
622 	struct netlink_sock *nlk = nlk_sk(sk);
623 	struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
624 
625 	if (addr->sa_family == AF_UNSPEC) {
626 		sk->sk_state	= NETLINK_UNCONNECTED;
627 		nlk->dst_pid	= 0;
628 		nlk->dst_group  = 0;
629 		return 0;
630 	}
631 	if (addr->sa_family != AF_NETLINK)
632 		return -EINVAL;
633 
634 	/* Only superuser is allowed to send multicasts */
635 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
636 		return -EPERM;
637 
638 	if (!nlk->pid)
639 		err = netlink_autobind(sock);
640 
641 	if (err == 0) {
642 		sk->sk_state	= NETLINK_CONNECTED;
643 		nlk->dst_pid 	= nladdr->nl_pid;
644 		nlk->dst_group  = ffs(nladdr->nl_groups);
645 	}
646 
647 	return err;
648 }
649 
650 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
651 {
652 	struct sock *sk = sock->sk;
653 	struct netlink_sock *nlk = nlk_sk(sk);
654 	struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
655 
656 	nladdr->nl_family = AF_NETLINK;
657 	nladdr->nl_pad = 0;
658 	*addr_len = sizeof(*nladdr);
659 
660 	if (peer) {
661 		nladdr->nl_pid = nlk->dst_pid;
662 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
663 	} else {
664 		nladdr->nl_pid = nlk->pid;
665 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
666 	}
667 	return 0;
668 }
669 
670 static void netlink_overrun(struct sock *sk)
671 {
672 	if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
673 		sk->sk_err = ENOBUFS;
674 		sk->sk_error_report(sk);
675 	}
676 }
677 
678 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
679 {
680 	int protocol = ssk->sk_protocol;
681 	struct sock *sock;
682 	struct netlink_sock *nlk;
683 
684 	sock = netlink_lookup(protocol, pid);
685 	if (!sock)
686 		return ERR_PTR(-ECONNREFUSED);
687 
688 	/* Don't bother queuing skb if kernel socket has no input function */
689 	nlk = nlk_sk(sock);
690 	if ((nlk->pid == 0 && !nlk->data_ready) ||
691 	    (sock->sk_state == NETLINK_CONNECTED &&
692 	     nlk->dst_pid != nlk_sk(ssk)->pid)) {
693 		sock_put(sock);
694 		return ERR_PTR(-ECONNREFUSED);
695 	}
696 	return sock;
697 }
698 
699 struct sock *netlink_getsockbyfilp(struct file *filp)
700 {
701 	struct inode *inode = filp->f_path.dentry->d_inode;
702 	struct sock *sock;
703 
704 	if (!S_ISSOCK(inode->i_mode))
705 		return ERR_PTR(-ENOTSOCK);
706 
707 	sock = SOCKET_I(inode)->sk;
708 	if (sock->sk_family != AF_NETLINK)
709 		return ERR_PTR(-EINVAL);
710 
711 	sock_hold(sock);
712 	return sock;
713 }
714 
715 /*
716  * Attach a skb to a netlink socket.
717  * The caller must hold a reference to the destination socket. On error, the
718  * reference is dropped. The skb is not send to the destination, just all
719  * all error checks are performed and memory in the queue is reserved.
720  * Return values:
721  * < 0: error. skb freed, reference to sock dropped.
722  * 0: continue
723  * 1: repeat lookup - reference dropped while waiting for socket memory.
724  */
725 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
726 		long timeo, struct sock *ssk)
727 {
728 	struct netlink_sock *nlk;
729 
730 	nlk = nlk_sk(sk);
731 
732 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
733 	    test_bit(0, &nlk->state)) {
734 		DECLARE_WAITQUEUE(wait, current);
735 		if (!timeo) {
736 			if (!ssk || nlk_sk(ssk)->pid == 0)
737 				netlink_overrun(sk);
738 			sock_put(sk);
739 			kfree_skb(skb);
740 			return -EAGAIN;
741 		}
742 
743 		__set_current_state(TASK_INTERRUPTIBLE);
744 		add_wait_queue(&nlk->wait, &wait);
745 
746 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
747 		     test_bit(0, &nlk->state)) &&
748 		    !sock_flag(sk, SOCK_DEAD))
749 			timeo = schedule_timeout(timeo);
750 
751 		__set_current_state(TASK_RUNNING);
752 		remove_wait_queue(&nlk->wait, &wait);
753 		sock_put(sk);
754 
755 		if (signal_pending(current)) {
756 			kfree_skb(skb);
757 			return sock_intr_errno(timeo);
758 		}
759 		return 1;
760 	}
761 	skb_set_owner_r(skb, sk);
762 	return 0;
763 }
764 
765 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
766 {
767 	int len = skb->len;
768 
769 	skb_queue_tail(&sk->sk_receive_queue, skb);
770 	sk->sk_data_ready(sk, len);
771 	sock_put(sk);
772 	return len;
773 }
774 
775 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
776 {
777 	kfree_skb(skb);
778 	sock_put(sk);
779 }
780 
781 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
782 					   gfp_t allocation)
783 {
784 	int delta;
785 
786 	skb_orphan(skb);
787 
788 	delta = skb->end - skb->tail;
789 	if (delta * 2 < skb->truesize)
790 		return skb;
791 
792 	if (skb_shared(skb)) {
793 		struct sk_buff *nskb = skb_clone(skb, allocation);
794 		if (!nskb)
795 			return skb;
796 		kfree_skb(skb);
797 		skb = nskb;
798 	}
799 
800 	if (!pskb_expand_head(skb, 0, -delta, allocation))
801 		skb->truesize -= delta;
802 
803 	return skb;
804 }
805 
806 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
807 {
808 	struct sock *sk;
809 	int err;
810 	long timeo;
811 
812 	skb = netlink_trim(skb, gfp_any());
813 
814 	timeo = sock_sndtimeo(ssk, nonblock);
815 retry:
816 	sk = netlink_getsockbypid(ssk, pid);
817 	if (IS_ERR(sk)) {
818 		kfree_skb(skb);
819 		return PTR_ERR(sk);
820 	}
821 	err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
822 	if (err == 1)
823 		goto retry;
824 	if (err)
825 		return err;
826 
827 	return netlink_sendskb(sk, skb, ssk->sk_protocol);
828 }
829 
830 int netlink_has_listeners(struct sock *sk, unsigned int group)
831 {
832 	int res = 0;
833 
834 	BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
835 	if (group - 1 < nl_table[sk->sk_protocol].groups)
836 		res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
837 	return res;
838 }
839 EXPORT_SYMBOL_GPL(netlink_has_listeners);
840 
841 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
842 {
843 	struct netlink_sock *nlk = nlk_sk(sk);
844 
845 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
846 	    !test_bit(0, &nlk->state)) {
847 		skb_set_owner_r(skb, sk);
848 		skb_queue_tail(&sk->sk_receive_queue, skb);
849 		sk->sk_data_ready(sk, skb->len);
850 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
851 	}
852 	return -1;
853 }
854 
855 struct netlink_broadcast_data {
856 	struct sock *exclude_sk;
857 	u32 pid;
858 	u32 group;
859 	int failure;
860 	int congested;
861 	int delivered;
862 	gfp_t allocation;
863 	struct sk_buff *skb, *skb2;
864 };
865 
866 static inline int do_one_broadcast(struct sock *sk,
867 				   struct netlink_broadcast_data *p)
868 {
869 	struct netlink_sock *nlk = nlk_sk(sk);
870 	int val;
871 
872 	if (p->exclude_sk == sk)
873 		goto out;
874 
875 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
876 	    !test_bit(p->group - 1, nlk->groups))
877 		goto out;
878 
879 	if (p->failure) {
880 		netlink_overrun(sk);
881 		goto out;
882 	}
883 
884 	sock_hold(sk);
885 	if (p->skb2 == NULL) {
886 		if (skb_shared(p->skb)) {
887 			p->skb2 = skb_clone(p->skb, p->allocation);
888 		} else {
889 			p->skb2 = skb_get(p->skb);
890 			/*
891 			 * skb ownership may have been set when
892 			 * delivered to a previous socket.
893 			 */
894 			skb_orphan(p->skb2);
895 		}
896 	}
897 	if (p->skb2 == NULL) {
898 		netlink_overrun(sk);
899 		/* Clone failed. Notify ALL listeners. */
900 		p->failure = 1;
901 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
902 		netlink_overrun(sk);
903 	} else {
904 		p->congested |= val;
905 		p->delivered = 1;
906 		p->skb2 = NULL;
907 	}
908 	sock_put(sk);
909 
910 out:
911 	return 0;
912 }
913 
914 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
915 		      u32 group, gfp_t allocation)
916 {
917 	struct netlink_broadcast_data info;
918 	struct hlist_node *node;
919 	struct sock *sk;
920 
921 	skb = netlink_trim(skb, allocation);
922 
923 	info.exclude_sk = ssk;
924 	info.pid = pid;
925 	info.group = group;
926 	info.failure = 0;
927 	info.congested = 0;
928 	info.delivered = 0;
929 	info.allocation = allocation;
930 	info.skb = skb;
931 	info.skb2 = NULL;
932 
933 	/* While we sleep in clone, do not allow to change socket list */
934 
935 	netlink_lock_table();
936 
937 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
938 		do_one_broadcast(sk, &info);
939 
940 	kfree_skb(skb);
941 
942 	netlink_unlock_table();
943 
944 	if (info.skb2)
945 		kfree_skb(info.skb2);
946 
947 	if (info.delivered) {
948 		if (info.congested && (allocation & __GFP_WAIT))
949 			yield();
950 		return 0;
951 	}
952 	if (info.failure)
953 		return -ENOBUFS;
954 	return -ESRCH;
955 }
956 
957 struct netlink_set_err_data {
958 	struct sock *exclude_sk;
959 	u32 pid;
960 	u32 group;
961 	int code;
962 };
963 
964 static inline int do_one_set_err(struct sock *sk,
965 				 struct netlink_set_err_data *p)
966 {
967 	struct netlink_sock *nlk = nlk_sk(sk);
968 
969 	if (sk == p->exclude_sk)
970 		goto out;
971 
972 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
973 	    !test_bit(p->group - 1, nlk->groups))
974 		goto out;
975 
976 	sk->sk_err = p->code;
977 	sk->sk_error_report(sk);
978 out:
979 	return 0;
980 }
981 
982 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
983 {
984 	struct netlink_set_err_data info;
985 	struct hlist_node *node;
986 	struct sock *sk;
987 
988 	info.exclude_sk = ssk;
989 	info.pid = pid;
990 	info.group = group;
991 	info.code = code;
992 
993 	read_lock(&nl_table_lock);
994 
995 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
996 		do_one_set_err(sk, &info);
997 
998 	read_unlock(&nl_table_lock);
999 }
1000 
1001 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1002 			      char __user *optval, int optlen)
1003 {
1004 	struct sock *sk = sock->sk;
1005 	struct netlink_sock *nlk = nlk_sk(sk);
1006 	int val = 0, err;
1007 
1008 	if (level != SOL_NETLINK)
1009 		return -ENOPROTOOPT;
1010 
1011 	if (optlen >= sizeof(int) &&
1012 	    get_user(val, (int __user *)optval))
1013 		return -EFAULT;
1014 
1015 	switch (optname) {
1016 	case NETLINK_PKTINFO:
1017 		if (val)
1018 			nlk->flags |= NETLINK_RECV_PKTINFO;
1019 		else
1020 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
1021 		err = 0;
1022 		break;
1023 	case NETLINK_ADD_MEMBERSHIP:
1024 	case NETLINK_DROP_MEMBERSHIP: {
1025 		unsigned int subscriptions;
1026 		int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1027 
1028 		if (!netlink_capable(sock, NL_NONROOT_RECV))
1029 			return -EPERM;
1030 		if (nlk->groups == NULL) {
1031 			err = netlink_alloc_groups(sk);
1032 			if (err)
1033 				return err;
1034 		}
1035 		if (!val || val - 1 >= nlk->ngroups)
1036 			return -EINVAL;
1037 		netlink_table_grab();
1038 		old = test_bit(val - 1, nlk->groups);
1039 		subscriptions = nlk->subscriptions - old + new;
1040 		if (new)
1041 			__set_bit(val - 1, nlk->groups);
1042 		else
1043 			__clear_bit(val - 1, nlk->groups);
1044 		netlink_update_subscriptions(sk, subscriptions);
1045 		netlink_update_listeners(sk);
1046 		netlink_table_ungrab();
1047 		err = 0;
1048 		break;
1049 	}
1050 	default:
1051 		err = -ENOPROTOOPT;
1052 	}
1053 	return err;
1054 }
1055 
1056 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1057 			      char __user *optval, int __user *optlen)
1058 {
1059 	struct sock *sk = sock->sk;
1060 	struct netlink_sock *nlk = nlk_sk(sk);
1061 	int len, val, err;
1062 
1063 	if (level != SOL_NETLINK)
1064 		return -ENOPROTOOPT;
1065 
1066 	if (get_user(len, optlen))
1067 		return -EFAULT;
1068 	if (len < 0)
1069 		return -EINVAL;
1070 
1071 	switch (optname) {
1072 	case NETLINK_PKTINFO:
1073 		if (len < sizeof(int))
1074 			return -EINVAL;
1075 		len = sizeof(int);
1076 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1077 		if (put_user(len, optlen) ||
1078 		    put_user(val, optval))
1079 			return -EFAULT;
1080 		err = 0;
1081 		break;
1082 	default:
1083 		err = -ENOPROTOOPT;
1084 	}
1085 	return err;
1086 }
1087 
1088 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1089 {
1090 	struct nl_pktinfo info;
1091 
1092 	info.group = NETLINK_CB(skb).dst_group;
1093 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1094 }
1095 
1096 static inline void netlink_rcv_wake(struct sock *sk)
1097 {
1098 	struct netlink_sock *nlk = nlk_sk(sk);
1099 
1100 	if (skb_queue_empty(&sk->sk_receive_queue))
1101 		clear_bit(0, &nlk->state);
1102 	if (!test_bit(0, &nlk->state))
1103 		wake_up_interruptible(&nlk->wait);
1104 }
1105 
1106 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1107 			   struct msghdr *msg, size_t len)
1108 {
1109 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1110 	struct sock *sk = sock->sk;
1111 	struct netlink_sock *nlk = nlk_sk(sk);
1112 	struct sockaddr_nl *addr=msg->msg_name;
1113 	u32 dst_pid;
1114 	u32 dst_group;
1115 	struct sk_buff *skb;
1116 	int err;
1117 	struct scm_cookie scm;
1118 
1119 	if (msg->msg_flags&MSG_OOB)
1120 		return -EOPNOTSUPP;
1121 
1122 	if (NULL == siocb->scm)
1123 		siocb->scm = &scm;
1124 	err = scm_send(sock, msg, siocb->scm);
1125 	if (err < 0)
1126 		return err;
1127 
1128 	if (msg->msg_namelen) {
1129 		if (addr->nl_family != AF_NETLINK)
1130 			return -EINVAL;
1131 		dst_pid = addr->nl_pid;
1132 		dst_group = ffs(addr->nl_groups);
1133 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1134 			return -EPERM;
1135 	} else {
1136 		dst_pid = nlk->dst_pid;
1137 		dst_group = nlk->dst_group;
1138 	}
1139 
1140 	if (!nlk->pid) {
1141 		err = netlink_autobind(sock);
1142 		if (err)
1143 			goto out;
1144 	}
1145 
1146 	err = -EMSGSIZE;
1147 	if (len > sk->sk_sndbuf - 32)
1148 		goto out;
1149 	err = -ENOBUFS;
1150 	skb = alloc_skb(len, GFP_KERNEL);
1151 	if (skb==NULL)
1152 		goto out;
1153 
1154 	NETLINK_CB(skb).pid	= nlk->pid;
1155 	NETLINK_CB(skb).dst_group = dst_group;
1156 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1157 	selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1158 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1159 
1160 	/* What can I do? Netlink is asynchronous, so that
1161 	   we will have to save current capabilities to
1162 	   check them, when this message will be delivered
1163 	   to corresponding kernel module.   --ANK (980802)
1164 	 */
1165 
1166 	err = -EFAULT;
1167 	if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1168 		kfree_skb(skb);
1169 		goto out;
1170 	}
1171 
1172 	err = security_netlink_send(sk, skb);
1173 	if (err) {
1174 		kfree_skb(skb);
1175 		goto out;
1176 	}
1177 
1178 	if (dst_group) {
1179 		atomic_inc(&skb->users);
1180 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1181 	}
1182 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1183 
1184 out:
1185 	return err;
1186 }
1187 
1188 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1189 			   struct msghdr *msg, size_t len,
1190 			   int flags)
1191 {
1192 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1193 	struct scm_cookie scm;
1194 	struct sock *sk = sock->sk;
1195 	struct netlink_sock *nlk = nlk_sk(sk);
1196 	int noblock = flags&MSG_DONTWAIT;
1197 	size_t copied;
1198 	struct sk_buff *skb;
1199 	int err;
1200 
1201 	if (flags&MSG_OOB)
1202 		return -EOPNOTSUPP;
1203 
1204 	copied = 0;
1205 
1206 	skb = skb_recv_datagram(sk,flags,noblock,&err);
1207 	if (skb==NULL)
1208 		goto out;
1209 
1210 	msg->msg_namelen = 0;
1211 
1212 	copied = skb->len;
1213 	if (len < copied) {
1214 		msg->msg_flags |= MSG_TRUNC;
1215 		copied = len;
1216 	}
1217 
1218 	skb->h.raw = skb->data;
1219 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1220 
1221 	if (msg->msg_name) {
1222 		struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1223 		addr->nl_family = AF_NETLINK;
1224 		addr->nl_pad    = 0;
1225 		addr->nl_pid	= NETLINK_CB(skb).pid;
1226 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1227 		msg->msg_namelen = sizeof(*addr);
1228 	}
1229 
1230 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1231 		netlink_cmsg_recv_pktinfo(msg, skb);
1232 
1233 	if (NULL == siocb->scm) {
1234 		memset(&scm, 0, sizeof(scm));
1235 		siocb->scm = &scm;
1236 	}
1237 	siocb->scm->creds = *NETLINK_CREDS(skb);
1238 	skb_free_datagram(sk, skb);
1239 
1240 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1241 		netlink_dump(sk);
1242 
1243 	scm_recv(sock, msg, siocb->scm, flags);
1244 
1245 out:
1246 	netlink_rcv_wake(sk);
1247 	return err ? : copied;
1248 }
1249 
1250 static void netlink_data_ready(struct sock *sk, int len)
1251 {
1252 	struct netlink_sock *nlk = nlk_sk(sk);
1253 
1254 	if (nlk->data_ready)
1255 		nlk->data_ready(sk, len);
1256 	netlink_rcv_wake(sk);
1257 }
1258 
1259 /*
1260  *	We export these functions to other modules. They provide a
1261  *	complete set of kernel non-blocking support for message
1262  *	queueing.
1263  */
1264 
1265 struct sock *
1266 netlink_kernel_create(int unit, unsigned int groups,
1267 		      void (*input)(struct sock *sk, int len),
1268 		      struct module *module)
1269 {
1270 	struct socket *sock;
1271 	struct sock *sk;
1272 	struct netlink_sock *nlk;
1273 	unsigned long *listeners = NULL;
1274 
1275 	BUG_ON(!nl_table);
1276 
1277 	if (unit<0 || unit>=MAX_LINKS)
1278 		return NULL;
1279 
1280 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1281 		return NULL;
1282 
1283 	if (__netlink_create(sock, unit) < 0)
1284 		goto out_sock_release;
1285 
1286 	if (groups < 32)
1287 		groups = 32;
1288 
1289 	listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1290 	if (!listeners)
1291 		goto out_sock_release;
1292 
1293 	sk = sock->sk;
1294 	sk->sk_data_ready = netlink_data_ready;
1295 	if (input)
1296 		nlk_sk(sk)->data_ready = input;
1297 
1298 	if (netlink_insert(sk, 0))
1299 		goto out_sock_release;
1300 
1301 	nlk = nlk_sk(sk);
1302 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1303 
1304 	netlink_table_grab();
1305 	nl_table[unit].groups = groups;
1306 	nl_table[unit].listeners = listeners;
1307 	nl_table[unit].module = module;
1308 	nl_table[unit].registered = 1;
1309 	netlink_table_ungrab();
1310 
1311 	return sk;
1312 
1313 out_sock_release:
1314 	kfree(listeners);
1315 	sock_release(sock);
1316 	return NULL;
1317 }
1318 
1319 void netlink_set_nonroot(int protocol, unsigned int flags)
1320 {
1321 	if ((unsigned int)protocol < MAX_LINKS)
1322 		nl_table[protocol].nl_nonroot = flags;
1323 }
1324 
1325 static void netlink_destroy_callback(struct netlink_callback *cb)
1326 {
1327 	if (cb->skb)
1328 		kfree_skb(cb->skb);
1329 	kfree(cb);
1330 }
1331 
1332 /*
1333  * It looks a bit ugly.
1334  * It would be better to create kernel thread.
1335  */
1336 
1337 static int netlink_dump(struct sock *sk)
1338 {
1339 	struct netlink_sock *nlk = nlk_sk(sk);
1340 	struct netlink_callback *cb;
1341 	struct sk_buff *skb;
1342 	struct nlmsghdr *nlh;
1343 	int len, err = -ENOBUFS;
1344 
1345 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1346 	if (!skb)
1347 		goto errout;
1348 
1349 	spin_lock(&nlk->cb_lock);
1350 
1351 	cb = nlk->cb;
1352 	if (cb == NULL) {
1353 		err = -EINVAL;
1354 		goto errout_skb;
1355 	}
1356 
1357 	len = cb->dump(skb, cb);
1358 
1359 	if (len > 0) {
1360 		spin_unlock(&nlk->cb_lock);
1361 		skb_queue_tail(&sk->sk_receive_queue, skb);
1362 		sk->sk_data_ready(sk, len);
1363 		return 0;
1364 	}
1365 
1366 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1367 	if (!nlh)
1368 		goto errout_skb;
1369 
1370 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
1371 
1372 	skb_queue_tail(&sk->sk_receive_queue, skb);
1373 	sk->sk_data_ready(sk, skb->len);
1374 
1375 	if (cb->done)
1376 		cb->done(cb);
1377 	nlk->cb = NULL;
1378 	spin_unlock(&nlk->cb_lock);
1379 
1380 	netlink_destroy_callback(cb);
1381 	return 0;
1382 
1383 errout_skb:
1384 	spin_unlock(&nlk->cb_lock);
1385 	kfree_skb(skb);
1386 errout:
1387 	return err;
1388 }
1389 
1390 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1391 		       struct nlmsghdr *nlh,
1392 		       int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1393 		       int (*done)(struct netlink_callback*))
1394 {
1395 	struct netlink_callback *cb;
1396 	struct sock *sk;
1397 	struct netlink_sock *nlk;
1398 
1399 	cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1400 	if (cb == NULL)
1401 		return -ENOBUFS;
1402 
1403 	cb->dump = dump;
1404 	cb->done = done;
1405 	cb->nlh = nlh;
1406 	atomic_inc(&skb->users);
1407 	cb->skb = skb;
1408 
1409 	sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1410 	if (sk == NULL) {
1411 		netlink_destroy_callback(cb);
1412 		return -ECONNREFUSED;
1413 	}
1414 	nlk = nlk_sk(sk);
1415 	/* A dump is in progress... */
1416 	spin_lock(&nlk->cb_lock);
1417 	if (nlk->cb) {
1418 		spin_unlock(&nlk->cb_lock);
1419 		netlink_destroy_callback(cb);
1420 		sock_put(sk);
1421 		return -EBUSY;
1422 	}
1423 	nlk->cb = cb;
1424 	spin_unlock(&nlk->cb_lock);
1425 
1426 	netlink_dump(sk);
1427 	sock_put(sk);
1428 	return 0;
1429 }
1430 
1431 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1432 {
1433 	struct sk_buff *skb;
1434 	struct nlmsghdr *rep;
1435 	struct nlmsgerr *errmsg;
1436 	size_t payload = sizeof(*errmsg);
1437 
1438 	/* error messages get the original request appened */
1439 	if (err)
1440 		payload += nlmsg_len(nlh);
1441 
1442 	skb = nlmsg_new(payload, GFP_KERNEL);
1443 	if (!skb) {
1444 		struct sock *sk;
1445 
1446 		sk = netlink_lookup(in_skb->sk->sk_protocol,
1447 				    NETLINK_CB(in_skb).pid);
1448 		if (sk) {
1449 			sk->sk_err = ENOBUFS;
1450 			sk->sk_error_report(sk);
1451 			sock_put(sk);
1452 		}
1453 		return;
1454 	}
1455 
1456 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1457 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1458 	errmsg = nlmsg_data(rep);
1459 	errmsg->error = err;
1460 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1461 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1462 }
1463 
1464 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1465 						     struct nlmsghdr *, int *))
1466 {
1467 	struct nlmsghdr *nlh;
1468 	int err;
1469 
1470 	while (skb->len >= nlmsg_total_size(0)) {
1471 		nlh = (struct nlmsghdr *) skb->data;
1472 
1473 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1474 			return 0;
1475 
1476 		if (cb(skb, nlh, &err) < 0) {
1477 			/* Not an error, but we have to interrupt processing
1478 			 * here. Note: that in this case we do not pull
1479 			 * message from skb, it will be processed later.
1480 			 */
1481 			if (err == 0)
1482 				return -1;
1483 			netlink_ack(skb, nlh, err);
1484 		} else if (nlh->nlmsg_flags & NLM_F_ACK)
1485 			netlink_ack(skb, nlh, 0);
1486 
1487 		netlink_queue_skip(nlh, skb);
1488 	}
1489 
1490 	return 0;
1491 }
1492 
1493 /**
1494  * nelink_run_queue - Process netlink receive queue.
1495  * @sk: Netlink socket containing the queue
1496  * @qlen: Place to store queue length upon entry
1497  * @cb: Callback function invoked for each netlink message found
1498  *
1499  * Processes as much as there was in the queue upon entry and invokes
1500  * a callback function for each netlink message found. The callback
1501  * function may refuse a message by returning a negative error code
1502  * but setting the error pointer to 0 in which case this function
1503  * returns with a qlen != 0.
1504  *
1505  * qlen must be initialized to 0 before the initial entry, afterwards
1506  * the function may be called repeatedly until qlen reaches 0.
1507  */
1508 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1509 		       int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1510 {
1511 	struct sk_buff *skb;
1512 
1513 	if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1514 		*qlen = skb_queue_len(&sk->sk_receive_queue);
1515 
1516 	for (; *qlen; (*qlen)--) {
1517 		skb = skb_dequeue(&sk->sk_receive_queue);
1518 		if (netlink_rcv_skb(skb, cb)) {
1519 			if (skb->len)
1520 				skb_queue_head(&sk->sk_receive_queue, skb);
1521 			else {
1522 				kfree_skb(skb);
1523 				(*qlen)--;
1524 			}
1525 			break;
1526 		}
1527 
1528 		kfree_skb(skb);
1529 	}
1530 }
1531 
1532 /**
1533  * netlink_queue_skip - Skip netlink message while processing queue.
1534  * @nlh: Netlink message to be skipped
1535  * @skb: Socket buffer containing the netlink messages.
1536  *
1537  * Pulls the given netlink message off the socket buffer so the next
1538  * call to netlink_queue_run() will not reconsider the message.
1539  */
1540 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1541 {
1542 	int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1543 
1544 	if (msglen > skb->len)
1545 		msglen = skb->len;
1546 
1547 	skb_pull(skb, msglen);
1548 }
1549 
1550 /**
1551  * nlmsg_notify - send a notification netlink message
1552  * @sk: netlink socket to use
1553  * @skb: notification message
1554  * @pid: destination netlink pid for reports or 0
1555  * @group: destination multicast group or 0
1556  * @report: 1 to report back, 0 to disable
1557  * @flags: allocation flags
1558  */
1559 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1560 		 unsigned int group, int report, gfp_t flags)
1561 {
1562 	int err = 0;
1563 
1564 	if (group) {
1565 		int exclude_pid = 0;
1566 
1567 		if (report) {
1568 			atomic_inc(&skb->users);
1569 			exclude_pid = pid;
1570 		}
1571 
1572 		/* errors reported via destination sk->sk_err */
1573 		nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1574 	}
1575 
1576 	if (report)
1577 		err = nlmsg_unicast(sk, skb, pid);
1578 
1579 	return err;
1580 }
1581 
1582 #ifdef CONFIG_PROC_FS
1583 struct nl_seq_iter {
1584 	int link;
1585 	int hash_idx;
1586 };
1587 
1588 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1589 {
1590 	struct nl_seq_iter *iter = seq->private;
1591 	int i, j;
1592 	struct sock *s;
1593 	struct hlist_node *node;
1594 	loff_t off = 0;
1595 
1596 	for (i=0; i<MAX_LINKS; i++) {
1597 		struct nl_pid_hash *hash = &nl_table[i].hash;
1598 
1599 		for (j = 0; j <= hash->mask; j++) {
1600 			sk_for_each(s, node, &hash->table[j]) {
1601 				if (off == pos) {
1602 					iter->link = i;
1603 					iter->hash_idx = j;
1604 					return s;
1605 				}
1606 				++off;
1607 			}
1608 		}
1609 	}
1610 	return NULL;
1611 }
1612 
1613 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1614 {
1615 	read_lock(&nl_table_lock);
1616 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1617 }
1618 
1619 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1620 {
1621 	struct sock *s;
1622 	struct nl_seq_iter *iter;
1623 	int i, j;
1624 
1625 	++*pos;
1626 
1627 	if (v == SEQ_START_TOKEN)
1628 		return netlink_seq_socket_idx(seq, 0);
1629 
1630 	s = sk_next(v);
1631 	if (s)
1632 		return s;
1633 
1634 	iter = seq->private;
1635 	i = iter->link;
1636 	j = iter->hash_idx + 1;
1637 
1638 	do {
1639 		struct nl_pid_hash *hash = &nl_table[i].hash;
1640 
1641 		for (; j <= hash->mask; j++) {
1642 			s = sk_head(&hash->table[j]);
1643 			if (s) {
1644 				iter->link = i;
1645 				iter->hash_idx = j;
1646 				return s;
1647 			}
1648 		}
1649 
1650 		j = 0;
1651 	} while (++i < MAX_LINKS);
1652 
1653 	return NULL;
1654 }
1655 
1656 static void netlink_seq_stop(struct seq_file *seq, void *v)
1657 {
1658 	read_unlock(&nl_table_lock);
1659 }
1660 
1661 
1662 static int netlink_seq_show(struct seq_file *seq, void *v)
1663 {
1664 	if (v == SEQ_START_TOKEN)
1665 		seq_puts(seq,
1666 			 "sk       Eth Pid    Groups   "
1667 			 "Rmem     Wmem     Dump     Locks\n");
1668 	else {
1669 		struct sock *s = v;
1670 		struct netlink_sock *nlk = nlk_sk(s);
1671 
1672 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1673 			   s,
1674 			   s->sk_protocol,
1675 			   nlk->pid,
1676 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1677 			   atomic_read(&s->sk_rmem_alloc),
1678 			   atomic_read(&s->sk_wmem_alloc),
1679 			   nlk->cb,
1680 			   atomic_read(&s->sk_refcnt)
1681 			);
1682 
1683 	}
1684 	return 0;
1685 }
1686 
1687 static struct seq_operations netlink_seq_ops = {
1688 	.start  = netlink_seq_start,
1689 	.next   = netlink_seq_next,
1690 	.stop   = netlink_seq_stop,
1691 	.show   = netlink_seq_show,
1692 };
1693 
1694 
1695 static int netlink_seq_open(struct inode *inode, struct file *file)
1696 {
1697 	struct seq_file *seq;
1698 	struct nl_seq_iter *iter;
1699 	int err;
1700 
1701 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1702 	if (!iter)
1703 		return -ENOMEM;
1704 
1705 	err = seq_open(file, &netlink_seq_ops);
1706 	if (err) {
1707 		kfree(iter);
1708 		return err;
1709 	}
1710 
1711 	seq = file->private_data;
1712 	seq->private = iter;
1713 	return 0;
1714 }
1715 
1716 static const struct file_operations netlink_seq_fops = {
1717 	.owner		= THIS_MODULE,
1718 	.open		= netlink_seq_open,
1719 	.read		= seq_read,
1720 	.llseek		= seq_lseek,
1721 	.release	= seq_release_private,
1722 };
1723 
1724 #endif
1725 
1726 int netlink_register_notifier(struct notifier_block *nb)
1727 {
1728 	return atomic_notifier_chain_register(&netlink_chain, nb);
1729 }
1730 
1731 int netlink_unregister_notifier(struct notifier_block *nb)
1732 {
1733 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
1734 }
1735 
1736 static const struct proto_ops netlink_ops = {
1737 	.family =	PF_NETLINK,
1738 	.owner =	THIS_MODULE,
1739 	.release =	netlink_release,
1740 	.bind =		netlink_bind,
1741 	.connect =	netlink_connect,
1742 	.socketpair =	sock_no_socketpair,
1743 	.accept =	sock_no_accept,
1744 	.getname =	netlink_getname,
1745 	.poll =		datagram_poll,
1746 	.ioctl =	sock_no_ioctl,
1747 	.listen =	sock_no_listen,
1748 	.shutdown =	sock_no_shutdown,
1749 	.setsockopt =	netlink_setsockopt,
1750 	.getsockopt =	netlink_getsockopt,
1751 	.sendmsg =	netlink_sendmsg,
1752 	.recvmsg =	netlink_recvmsg,
1753 	.mmap =		sock_no_mmap,
1754 	.sendpage =	sock_no_sendpage,
1755 };
1756 
1757 static struct net_proto_family netlink_family_ops = {
1758 	.family = PF_NETLINK,
1759 	.create = netlink_create,
1760 	.owner	= THIS_MODULE,	/* for consistency 8) */
1761 };
1762 
1763 static int __init netlink_proto_init(void)
1764 {
1765 	struct sk_buff *dummy_skb;
1766 	int i;
1767 	unsigned long max;
1768 	unsigned int order;
1769 	int err = proto_register(&netlink_proto, 0);
1770 
1771 	if (err != 0)
1772 		goto out;
1773 
1774 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1775 
1776 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1777 	if (!nl_table)
1778 		goto panic;
1779 
1780 	if (num_physpages >= (128 * 1024))
1781 		max = num_physpages >> (21 - PAGE_SHIFT);
1782 	else
1783 		max = num_physpages >> (23 - PAGE_SHIFT);
1784 
1785 	order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1786 	max = (1UL << order) / sizeof(struct hlist_head);
1787 	order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1788 
1789 	for (i = 0; i < MAX_LINKS; i++) {
1790 		struct nl_pid_hash *hash = &nl_table[i].hash;
1791 
1792 		hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1793 		if (!hash->table) {
1794 			while (i-- > 0)
1795 				nl_pid_hash_free(nl_table[i].hash.table,
1796 						 1 * sizeof(*hash->table));
1797 			kfree(nl_table);
1798 			goto panic;
1799 		}
1800 		memset(hash->table, 0, 1 * sizeof(*hash->table));
1801 		hash->max_shift = order;
1802 		hash->shift = 0;
1803 		hash->mask = 0;
1804 		hash->rehash_time = jiffies;
1805 	}
1806 
1807 	sock_register(&netlink_family_ops);
1808 #ifdef CONFIG_PROC_FS
1809 	proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1810 #endif
1811 	/* The netlink device handler may be needed early. */
1812 	rtnetlink_init();
1813 out:
1814 	return err;
1815 panic:
1816 	panic("netlink_init: Cannot allocate nl_table\n");
1817 }
1818 
1819 core_initcall(netlink_proto_init);
1820 
1821 EXPORT_SYMBOL(netlink_ack);
1822 EXPORT_SYMBOL(netlink_run_queue);
1823 EXPORT_SYMBOL(netlink_queue_skip);
1824 EXPORT_SYMBOL(netlink_broadcast);
1825 EXPORT_SYMBOL(netlink_dump_start);
1826 EXPORT_SYMBOL(netlink_kernel_create);
1827 EXPORT_SYMBOL(netlink_register_notifier);
1828 EXPORT_SYMBOL(netlink_set_err);
1829 EXPORT_SYMBOL(netlink_set_nonroot);
1830 EXPORT_SYMBOL(netlink_unicast);
1831 EXPORT_SYMBOL(netlink_unregister_notifier);
1832 EXPORT_SYMBOL(nlmsg_notify);
1833