xref: /openbmc/linux/net/netlink/af_netlink.c (revision 87c2ce3b)
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/config.h>
25 #include <linux/module.h>
26 
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 
59 #include <net/sock.h>
60 #include <net/scm.h>
61 #include <net/netlink.h>
62 
63 #define Nprintk(a...)
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 int nl_nonroot;
109 	unsigned int groups;
110 	struct module *module;
111 	int registered;
112 };
113 
114 static struct netlink_table *nl_table;
115 
116 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
117 
118 static int netlink_dump(struct sock *sk);
119 static void netlink_destroy_callback(struct netlink_callback *cb);
120 
121 static DEFINE_RWLOCK(nl_table_lock);
122 static atomic_t nl_table_users = ATOMIC_INIT(0);
123 
124 static struct notifier_block *netlink_chain;
125 
126 static u32 netlink_group_mask(u32 group)
127 {
128 	return group ? 1 << (group - 1) : 0;
129 }
130 
131 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
132 {
133 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
134 }
135 
136 static void netlink_sock_destruct(struct sock *sk)
137 {
138 	skb_queue_purge(&sk->sk_receive_queue);
139 
140 	if (!sock_flag(sk, SOCK_DEAD)) {
141 		printk("Freeing alive netlink socket %p\n", sk);
142 		return;
143 	}
144 	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
145 	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
146 	BUG_TRAP(!nlk_sk(sk)->cb);
147 	BUG_TRAP(!nlk_sk(sk)->groups);
148 }
149 
150 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
151  * Look, when several writers sleep and reader wakes them up, all but one
152  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
153  * this, _but_ remember, it adds useless work on UP machines.
154  */
155 
156 static void netlink_table_grab(void)
157 {
158 	write_lock_bh(&nl_table_lock);
159 
160 	if (atomic_read(&nl_table_users)) {
161 		DECLARE_WAITQUEUE(wait, current);
162 
163 		add_wait_queue_exclusive(&nl_table_wait, &wait);
164 		for(;;) {
165 			set_current_state(TASK_UNINTERRUPTIBLE);
166 			if (atomic_read(&nl_table_users) == 0)
167 				break;
168 			write_unlock_bh(&nl_table_lock);
169 			schedule();
170 			write_lock_bh(&nl_table_lock);
171 		}
172 
173 		__set_current_state(TASK_RUNNING);
174 		remove_wait_queue(&nl_table_wait, &wait);
175 	}
176 }
177 
178 static __inline__ void netlink_table_ungrab(void)
179 {
180 	write_unlock_bh(&nl_table_lock);
181 	wake_up(&nl_table_wait);
182 }
183 
184 static __inline__ void
185 netlink_lock_table(void)
186 {
187 	/* read_lock() synchronizes us to netlink_table_grab */
188 
189 	read_lock(&nl_table_lock);
190 	atomic_inc(&nl_table_users);
191 	read_unlock(&nl_table_lock);
192 }
193 
194 static __inline__ void
195 netlink_unlock_table(void)
196 {
197 	if (atomic_dec_and_test(&nl_table_users))
198 		wake_up(&nl_table_wait);
199 }
200 
201 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
202 {
203 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
204 	struct hlist_head *head;
205 	struct sock *sk;
206 	struct hlist_node *node;
207 
208 	read_lock(&nl_table_lock);
209 	head = nl_pid_hashfn(hash, pid);
210 	sk_for_each(sk, node, head) {
211 		if (nlk_sk(sk)->pid == pid) {
212 			sock_hold(sk);
213 			goto found;
214 		}
215 	}
216 	sk = NULL;
217 found:
218 	read_unlock(&nl_table_lock);
219 	return sk;
220 }
221 
222 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
223 {
224 	if (size <= PAGE_SIZE)
225 		return kmalloc(size, GFP_ATOMIC);
226 	else
227 		return (struct hlist_head *)
228 			__get_free_pages(GFP_ATOMIC, get_order(size));
229 }
230 
231 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
232 {
233 	if (size <= PAGE_SIZE)
234 		kfree(table);
235 	else
236 		free_pages((unsigned long)table, get_order(size));
237 }
238 
239 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
240 {
241 	unsigned int omask, mask, shift;
242 	size_t osize, size;
243 	struct hlist_head *otable, *table;
244 	int i;
245 
246 	omask = mask = hash->mask;
247 	osize = size = (mask + 1) * sizeof(*table);
248 	shift = hash->shift;
249 
250 	if (grow) {
251 		if (++shift > hash->max_shift)
252 			return 0;
253 		mask = mask * 2 + 1;
254 		size *= 2;
255 	}
256 
257 	table = nl_pid_hash_alloc(size);
258 	if (!table)
259 		return 0;
260 
261 	memset(table, 0, size);
262 	otable = hash->table;
263 	hash->table = table;
264 	hash->mask = mask;
265 	hash->shift = shift;
266 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
267 
268 	for (i = 0; i <= omask; i++) {
269 		struct sock *sk;
270 		struct hlist_node *node, *tmp;
271 
272 		sk_for_each_safe(sk, node, tmp, &otable[i])
273 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
274 	}
275 
276 	nl_pid_hash_free(otable, osize);
277 	hash->rehash_time = jiffies + 10 * 60 * HZ;
278 	return 1;
279 }
280 
281 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
282 {
283 	int avg = hash->entries >> hash->shift;
284 
285 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
286 		return 1;
287 
288 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
289 		nl_pid_hash_rehash(hash, 0);
290 		return 1;
291 	}
292 
293 	return 0;
294 }
295 
296 static const struct proto_ops netlink_ops;
297 
298 static int netlink_insert(struct sock *sk, u32 pid)
299 {
300 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
301 	struct hlist_head *head;
302 	int err = -EADDRINUSE;
303 	struct sock *osk;
304 	struct hlist_node *node;
305 	int len;
306 
307 	netlink_table_grab();
308 	head = nl_pid_hashfn(hash, pid);
309 	len = 0;
310 	sk_for_each(osk, node, head) {
311 		if (nlk_sk(osk)->pid == pid)
312 			break;
313 		len++;
314 	}
315 	if (node)
316 		goto err;
317 
318 	err = -EBUSY;
319 	if (nlk_sk(sk)->pid)
320 		goto err;
321 
322 	err = -ENOMEM;
323 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
324 		goto err;
325 
326 	if (len && nl_pid_hash_dilute(hash, len))
327 		head = nl_pid_hashfn(hash, pid);
328 	hash->entries++;
329 	nlk_sk(sk)->pid = pid;
330 	sk_add_node(sk, head);
331 	err = 0;
332 
333 err:
334 	netlink_table_ungrab();
335 	return err;
336 }
337 
338 static void netlink_remove(struct sock *sk)
339 {
340 	netlink_table_grab();
341 	if (sk_del_node_init(sk))
342 		nl_table[sk->sk_protocol].hash.entries--;
343 	if (nlk_sk(sk)->subscriptions)
344 		__sk_del_bind_node(sk);
345 	netlink_table_ungrab();
346 }
347 
348 static struct proto netlink_proto = {
349 	.name	  = "NETLINK",
350 	.owner	  = THIS_MODULE,
351 	.obj_size = sizeof(struct netlink_sock),
352 };
353 
354 static int __netlink_create(struct socket *sock, int protocol)
355 {
356 	struct sock *sk;
357 	struct netlink_sock *nlk;
358 
359 	sock->ops = &netlink_ops;
360 
361 	sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
362 	if (!sk)
363 		return -ENOMEM;
364 
365 	sock_init_data(sock, sk);
366 
367 	nlk = nlk_sk(sk);
368 	spin_lock_init(&nlk->cb_lock);
369 	init_waitqueue_head(&nlk->wait);
370 
371 	sk->sk_destruct = netlink_sock_destruct;
372 	sk->sk_protocol = protocol;
373 	return 0;
374 }
375 
376 static int netlink_create(struct socket *sock, int protocol)
377 {
378 	struct module *module = NULL;
379 	struct netlink_sock *nlk;
380 	unsigned int groups;
381 	int err = 0;
382 
383 	sock->state = SS_UNCONNECTED;
384 
385 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
386 		return -ESOCKTNOSUPPORT;
387 
388 	if (protocol<0 || protocol >= MAX_LINKS)
389 		return -EPROTONOSUPPORT;
390 
391 	netlink_lock_table();
392 #ifdef CONFIG_KMOD
393 	if (!nl_table[protocol].registered) {
394 		netlink_unlock_table();
395 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
396 		netlink_lock_table();
397 	}
398 #endif
399 	if (nl_table[protocol].registered &&
400 	    try_module_get(nl_table[protocol].module))
401 		module = nl_table[protocol].module;
402 	groups = nl_table[protocol].groups;
403 	netlink_unlock_table();
404 
405 	if ((err = __netlink_create(sock, protocol)) < 0)
406 		goto out_module;
407 
408 	nlk = nlk_sk(sock->sk);
409 	nlk->module = module;
410 out:
411 	return err;
412 
413 out_module:
414 	module_put(module);
415 	goto out;
416 }
417 
418 static int netlink_release(struct socket *sock)
419 {
420 	struct sock *sk = sock->sk;
421 	struct netlink_sock *nlk;
422 
423 	if (!sk)
424 		return 0;
425 
426 	netlink_remove(sk);
427 	nlk = nlk_sk(sk);
428 
429 	spin_lock(&nlk->cb_lock);
430 	if (nlk->cb) {
431 		if (nlk->cb->done)
432 			nlk->cb->done(nlk->cb);
433 		netlink_destroy_callback(nlk->cb);
434 		nlk->cb = NULL;
435 	}
436 	spin_unlock(&nlk->cb_lock);
437 
438 	/* OK. Socket is unlinked, and, therefore,
439 	   no new packets will arrive */
440 
441 	sock_orphan(sk);
442 	sock->sk = NULL;
443 	wake_up_interruptible_all(&nlk->wait);
444 
445 	skb_queue_purge(&sk->sk_write_queue);
446 
447 	if (nlk->pid && !nlk->subscriptions) {
448 		struct netlink_notify n = {
449 						.protocol = sk->sk_protocol,
450 						.pid = nlk->pid,
451 					  };
452 		notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
453 	}
454 
455 	if (nlk->module)
456 		module_put(nlk->module);
457 
458 	if (nlk->flags & NETLINK_KERNEL_SOCKET) {
459 		netlink_table_grab();
460 		nl_table[sk->sk_protocol].module = NULL;
461 		nl_table[sk->sk_protocol].registered = 0;
462 		netlink_table_ungrab();
463 	}
464 
465 	kfree(nlk->groups);
466 	nlk->groups = NULL;
467 
468 	sock_put(sk);
469 	return 0;
470 }
471 
472 static int netlink_autobind(struct socket *sock)
473 {
474 	struct sock *sk = sock->sk;
475 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
476 	struct hlist_head *head;
477 	struct sock *osk;
478 	struct hlist_node *node;
479 	s32 pid = current->tgid;
480 	int err;
481 	static s32 rover = -4097;
482 
483 retry:
484 	cond_resched();
485 	netlink_table_grab();
486 	head = nl_pid_hashfn(hash, pid);
487 	sk_for_each(osk, node, head) {
488 		if (nlk_sk(osk)->pid == pid) {
489 			/* Bind collision, search negative pid values. */
490 			pid = rover--;
491 			if (rover > -4097)
492 				rover = -4097;
493 			netlink_table_ungrab();
494 			goto retry;
495 		}
496 	}
497 	netlink_table_ungrab();
498 
499 	err = netlink_insert(sk, pid);
500 	if (err == -EADDRINUSE)
501 		goto retry;
502 
503 	/* If 2 threads race to autobind, that is fine.  */
504 	if (err == -EBUSY)
505 		err = 0;
506 
507 	return err;
508 }
509 
510 static inline int netlink_capable(struct socket *sock, unsigned int flag)
511 {
512 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
513 	       capable(CAP_NET_ADMIN);
514 }
515 
516 static void
517 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
518 {
519 	struct netlink_sock *nlk = nlk_sk(sk);
520 
521 	if (nlk->subscriptions && !subscriptions)
522 		__sk_del_bind_node(sk);
523 	else if (!nlk->subscriptions && subscriptions)
524 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
525 	nlk->subscriptions = subscriptions;
526 }
527 
528 static int netlink_alloc_groups(struct sock *sk)
529 {
530 	struct netlink_sock *nlk = nlk_sk(sk);
531 	unsigned int groups;
532 	int err = 0;
533 
534 	netlink_lock_table();
535 	groups = nl_table[sk->sk_protocol].groups;
536 	if (!nl_table[sk->sk_protocol].registered)
537 		err = -ENOENT;
538 	netlink_unlock_table();
539 
540 	if (err)
541 		return err;
542 
543 	nlk->groups = kmalloc(NLGRPSZ(groups), GFP_KERNEL);
544 	if (nlk->groups == NULL)
545 		return -ENOMEM;
546 	memset(nlk->groups, 0, NLGRPSZ(groups));
547 	nlk->ngroups = groups;
548 	return 0;
549 }
550 
551 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
552 {
553 	struct sock *sk = sock->sk;
554 	struct netlink_sock *nlk = nlk_sk(sk);
555 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
556 	int err;
557 
558 	if (nladdr->nl_family != AF_NETLINK)
559 		return -EINVAL;
560 
561 	/* Only superuser is allowed to listen multicasts */
562 	if (nladdr->nl_groups) {
563 		if (!netlink_capable(sock, NL_NONROOT_RECV))
564 			return -EPERM;
565 		if (nlk->groups == NULL) {
566 			err = netlink_alloc_groups(sk);
567 			if (err)
568 				return err;
569 		}
570 	}
571 
572 	if (nlk->pid) {
573 		if (nladdr->nl_pid != nlk->pid)
574 			return -EINVAL;
575 	} else {
576 		err = nladdr->nl_pid ?
577 			netlink_insert(sk, nladdr->nl_pid) :
578 			netlink_autobind(sock);
579 		if (err)
580 			return err;
581 	}
582 
583 	if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
584 		return 0;
585 
586 	netlink_table_grab();
587 	netlink_update_subscriptions(sk, nlk->subscriptions +
588 	                                 hweight32(nladdr->nl_groups) -
589 	                                 hweight32(nlk->groups[0]));
590 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
591 	netlink_table_ungrab();
592 
593 	return 0;
594 }
595 
596 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
597 			   int alen, int flags)
598 {
599 	int err = 0;
600 	struct sock *sk = sock->sk;
601 	struct netlink_sock *nlk = nlk_sk(sk);
602 	struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
603 
604 	if (addr->sa_family == AF_UNSPEC) {
605 		sk->sk_state	= NETLINK_UNCONNECTED;
606 		nlk->dst_pid	= 0;
607 		nlk->dst_group  = 0;
608 		return 0;
609 	}
610 	if (addr->sa_family != AF_NETLINK)
611 		return -EINVAL;
612 
613 	/* Only superuser is allowed to send multicasts */
614 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
615 		return -EPERM;
616 
617 	if (!nlk->pid)
618 		err = netlink_autobind(sock);
619 
620 	if (err == 0) {
621 		sk->sk_state	= NETLINK_CONNECTED;
622 		nlk->dst_pid 	= nladdr->nl_pid;
623 		nlk->dst_group  = ffs(nladdr->nl_groups);
624 	}
625 
626 	return err;
627 }
628 
629 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
630 {
631 	struct sock *sk = sock->sk;
632 	struct netlink_sock *nlk = nlk_sk(sk);
633 	struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
634 
635 	nladdr->nl_family = AF_NETLINK;
636 	nladdr->nl_pad = 0;
637 	*addr_len = sizeof(*nladdr);
638 
639 	if (peer) {
640 		nladdr->nl_pid = nlk->dst_pid;
641 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
642 	} else {
643 		nladdr->nl_pid = nlk->pid;
644 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
645 	}
646 	return 0;
647 }
648 
649 static void netlink_overrun(struct sock *sk)
650 {
651 	if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
652 		sk->sk_err = ENOBUFS;
653 		sk->sk_error_report(sk);
654 	}
655 }
656 
657 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
658 {
659 	int protocol = ssk->sk_protocol;
660 	struct sock *sock;
661 	struct netlink_sock *nlk;
662 
663 	sock = netlink_lookup(protocol, pid);
664 	if (!sock)
665 		return ERR_PTR(-ECONNREFUSED);
666 
667 	/* Don't bother queuing skb if kernel socket has no input function */
668 	nlk = nlk_sk(sock);
669 	if ((nlk->pid == 0 && !nlk->data_ready) ||
670 	    (sock->sk_state == NETLINK_CONNECTED &&
671 	     nlk->dst_pid != nlk_sk(ssk)->pid)) {
672 		sock_put(sock);
673 		return ERR_PTR(-ECONNREFUSED);
674 	}
675 	return sock;
676 }
677 
678 struct sock *netlink_getsockbyfilp(struct file *filp)
679 {
680 	struct inode *inode = filp->f_dentry->d_inode;
681 	struct sock *sock;
682 
683 	if (!S_ISSOCK(inode->i_mode))
684 		return ERR_PTR(-ENOTSOCK);
685 
686 	sock = SOCKET_I(inode)->sk;
687 	if (sock->sk_family != AF_NETLINK)
688 		return ERR_PTR(-EINVAL);
689 
690 	sock_hold(sock);
691 	return sock;
692 }
693 
694 /*
695  * Attach a skb to a netlink socket.
696  * The caller must hold a reference to the destination socket. On error, the
697  * reference is dropped. The skb is not send to the destination, just all
698  * all error checks are performed and memory in the queue is reserved.
699  * Return values:
700  * < 0: error. skb freed, reference to sock dropped.
701  * 0: continue
702  * 1: repeat lookup - reference dropped while waiting for socket memory.
703  */
704 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo)
705 {
706 	struct netlink_sock *nlk;
707 
708 	nlk = nlk_sk(sk);
709 
710 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
711 	    test_bit(0, &nlk->state)) {
712 		DECLARE_WAITQUEUE(wait, current);
713 		if (!timeo) {
714 			if (!nlk->pid)
715 				netlink_overrun(sk);
716 			sock_put(sk);
717 			kfree_skb(skb);
718 			return -EAGAIN;
719 		}
720 
721 		__set_current_state(TASK_INTERRUPTIBLE);
722 		add_wait_queue(&nlk->wait, &wait);
723 
724 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
725 		     test_bit(0, &nlk->state)) &&
726 		    !sock_flag(sk, SOCK_DEAD))
727 			timeo = schedule_timeout(timeo);
728 
729 		__set_current_state(TASK_RUNNING);
730 		remove_wait_queue(&nlk->wait, &wait);
731 		sock_put(sk);
732 
733 		if (signal_pending(current)) {
734 			kfree_skb(skb);
735 			return sock_intr_errno(timeo);
736 		}
737 		return 1;
738 	}
739 	skb_set_owner_r(skb, sk);
740 	return 0;
741 }
742 
743 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
744 {
745 	int len = skb->len;
746 
747 	skb_queue_tail(&sk->sk_receive_queue, skb);
748 	sk->sk_data_ready(sk, len);
749 	sock_put(sk);
750 	return len;
751 }
752 
753 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
754 {
755 	kfree_skb(skb);
756 	sock_put(sk);
757 }
758 
759 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
760 					   gfp_t allocation)
761 {
762 	int delta;
763 
764 	skb_orphan(skb);
765 
766 	delta = skb->end - skb->tail;
767 	if (delta * 2 < skb->truesize)
768 		return skb;
769 
770 	if (skb_shared(skb)) {
771 		struct sk_buff *nskb = skb_clone(skb, allocation);
772 		if (!nskb)
773 			return skb;
774 		kfree_skb(skb);
775 		skb = nskb;
776 	}
777 
778 	if (!pskb_expand_head(skb, 0, -delta, allocation))
779 		skb->truesize -= delta;
780 
781 	return skb;
782 }
783 
784 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
785 {
786 	struct sock *sk;
787 	int err;
788 	long timeo;
789 
790 	skb = netlink_trim(skb, gfp_any());
791 
792 	timeo = sock_sndtimeo(ssk, nonblock);
793 retry:
794 	sk = netlink_getsockbypid(ssk, pid);
795 	if (IS_ERR(sk)) {
796 		kfree_skb(skb);
797 		return PTR_ERR(sk);
798 	}
799 	err = netlink_attachskb(sk, skb, nonblock, timeo);
800 	if (err == 1)
801 		goto retry;
802 	if (err)
803 		return err;
804 
805 	return netlink_sendskb(sk, skb, ssk->sk_protocol);
806 }
807 
808 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
809 {
810 	struct netlink_sock *nlk = nlk_sk(sk);
811 
812 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
813 	    !test_bit(0, &nlk->state)) {
814 		skb_set_owner_r(skb, sk);
815 		skb_queue_tail(&sk->sk_receive_queue, skb);
816 		sk->sk_data_ready(sk, skb->len);
817 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
818 	}
819 	return -1;
820 }
821 
822 struct netlink_broadcast_data {
823 	struct sock *exclude_sk;
824 	u32 pid;
825 	u32 group;
826 	int failure;
827 	int congested;
828 	int delivered;
829 	gfp_t allocation;
830 	struct sk_buff *skb, *skb2;
831 };
832 
833 static inline int do_one_broadcast(struct sock *sk,
834 				   struct netlink_broadcast_data *p)
835 {
836 	struct netlink_sock *nlk = nlk_sk(sk);
837 	int val;
838 
839 	if (p->exclude_sk == sk)
840 		goto out;
841 
842 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
843 	    !test_bit(p->group - 1, nlk->groups))
844 		goto out;
845 
846 	if (p->failure) {
847 		netlink_overrun(sk);
848 		goto out;
849 	}
850 
851 	sock_hold(sk);
852 	if (p->skb2 == NULL) {
853 		if (skb_shared(p->skb)) {
854 			p->skb2 = skb_clone(p->skb, p->allocation);
855 		} else {
856 			p->skb2 = skb_get(p->skb);
857 			/*
858 			 * skb ownership may have been set when
859 			 * delivered to a previous socket.
860 			 */
861 			skb_orphan(p->skb2);
862 		}
863 	}
864 	if (p->skb2 == NULL) {
865 		netlink_overrun(sk);
866 		/* Clone failed. Notify ALL listeners. */
867 		p->failure = 1;
868 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
869 		netlink_overrun(sk);
870 	} else {
871 		p->congested |= val;
872 		p->delivered = 1;
873 		p->skb2 = NULL;
874 	}
875 	sock_put(sk);
876 
877 out:
878 	return 0;
879 }
880 
881 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
882 		      u32 group, gfp_t allocation)
883 {
884 	struct netlink_broadcast_data info;
885 	struct hlist_node *node;
886 	struct sock *sk;
887 
888 	skb = netlink_trim(skb, allocation);
889 
890 	info.exclude_sk = ssk;
891 	info.pid = pid;
892 	info.group = group;
893 	info.failure = 0;
894 	info.congested = 0;
895 	info.delivered = 0;
896 	info.allocation = allocation;
897 	info.skb = skb;
898 	info.skb2 = NULL;
899 
900 	/* While we sleep in clone, do not allow to change socket list */
901 
902 	netlink_lock_table();
903 
904 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
905 		do_one_broadcast(sk, &info);
906 
907 	kfree_skb(skb);
908 
909 	netlink_unlock_table();
910 
911 	if (info.skb2)
912 		kfree_skb(info.skb2);
913 
914 	if (info.delivered) {
915 		if (info.congested && (allocation & __GFP_WAIT))
916 			yield();
917 		return 0;
918 	}
919 	if (info.failure)
920 		return -ENOBUFS;
921 	return -ESRCH;
922 }
923 
924 struct netlink_set_err_data {
925 	struct sock *exclude_sk;
926 	u32 pid;
927 	u32 group;
928 	int code;
929 };
930 
931 static inline int do_one_set_err(struct sock *sk,
932 				 struct netlink_set_err_data *p)
933 {
934 	struct netlink_sock *nlk = nlk_sk(sk);
935 
936 	if (sk == p->exclude_sk)
937 		goto out;
938 
939 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
940 	    !test_bit(p->group - 1, nlk->groups))
941 		goto out;
942 
943 	sk->sk_err = p->code;
944 	sk->sk_error_report(sk);
945 out:
946 	return 0;
947 }
948 
949 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
950 {
951 	struct netlink_set_err_data info;
952 	struct hlist_node *node;
953 	struct sock *sk;
954 
955 	info.exclude_sk = ssk;
956 	info.pid = pid;
957 	info.group = group;
958 	info.code = code;
959 
960 	read_lock(&nl_table_lock);
961 
962 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
963 		do_one_set_err(sk, &info);
964 
965 	read_unlock(&nl_table_lock);
966 }
967 
968 static int netlink_setsockopt(struct socket *sock, int level, int optname,
969                               char __user *optval, int optlen)
970 {
971 	struct sock *sk = sock->sk;
972 	struct netlink_sock *nlk = nlk_sk(sk);
973 	int val = 0, err;
974 
975 	if (level != SOL_NETLINK)
976 		return -ENOPROTOOPT;
977 
978 	if (optlen >= sizeof(int) &&
979 	    get_user(val, (int __user *)optval))
980 		return -EFAULT;
981 
982 	switch (optname) {
983 	case NETLINK_PKTINFO:
984 		if (val)
985 			nlk->flags |= NETLINK_RECV_PKTINFO;
986 		else
987 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
988 		err = 0;
989 		break;
990 	case NETLINK_ADD_MEMBERSHIP:
991 	case NETLINK_DROP_MEMBERSHIP: {
992 		unsigned int subscriptions;
993 		int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
994 
995 		if (!netlink_capable(sock, NL_NONROOT_RECV))
996 			return -EPERM;
997 		if (nlk->groups == NULL) {
998 			err = netlink_alloc_groups(sk);
999 			if (err)
1000 				return err;
1001 		}
1002 		if (!val || val - 1 >= nlk->ngroups)
1003 			return -EINVAL;
1004 		netlink_table_grab();
1005 		old = test_bit(val - 1, nlk->groups);
1006 		subscriptions = nlk->subscriptions - old + new;
1007 		if (new)
1008 			__set_bit(val - 1, nlk->groups);
1009 		else
1010 			__clear_bit(val - 1, nlk->groups);
1011 		netlink_update_subscriptions(sk, subscriptions);
1012 		netlink_table_ungrab();
1013 		err = 0;
1014 		break;
1015 	}
1016 	default:
1017 		err = -ENOPROTOOPT;
1018 	}
1019 	return err;
1020 }
1021 
1022 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1023                               char __user *optval, int __user *optlen)
1024 {
1025 	struct sock *sk = sock->sk;
1026 	struct netlink_sock *nlk = nlk_sk(sk);
1027 	int len, val, err;
1028 
1029 	if (level != SOL_NETLINK)
1030 		return -ENOPROTOOPT;
1031 
1032 	if (get_user(len, optlen))
1033 		return -EFAULT;
1034 	if (len < 0)
1035 		return -EINVAL;
1036 
1037 	switch (optname) {
1038 	case NETLINK_PKTINFO:
1039 		if (len < sizeof(int))
1040 			return -EINVAL;
1041 		len = sizeof(int);
1042 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1043 		put_user(len, optlen);
1044 		put_user(val, optval);
1045 		err = 0;
1046 		break;
1047 	default:
1048 		err = -ENOPROTOOPT;
1049 	}
1050 	return err;
1051 }
1052 
1053 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1054 {
1055 	struct nl_pktinfo info;
1056 
1057 	info.group = NETLINK_CB(skb).dst_group;
1058 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1059 }
1060 
1061 static inline void netlink_rcv_wake(struct sock *sk)
1062 {
1063 	struct netlink_sock *nlk = nlk_sk(sk);
1064 
1065 	if (skb_queue_empty(&sk->sk_receive_queue))
1066 		clear_bit(0, &nlk->state);
1067 	if (!test_bit(0, &nlk->state))
1068 		wake_up_interruptible(&nlk->wait);
1069 }
1070 
1071 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1072 			   struct msghdr *msg, size_t len)
1073 {
1074 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1075 	struct sock *sk = sock->sk;
1076 	struct netlink_sock *nlk = nlk_sk(sk);
1077 	struct sockaddr_nl *addr=msg->msg_name;
1078 	u32 dst_pid;
1079 	u32 dst_group;
1080 	struct sk_buff *skb;
1081 	int err;
1082 	struct scm_cookie scm;
1083 
1084 	if (msg->msg_flags&MSG_OOB)
1085 		return -EOPNOTSUPP;
1086 
1087 	if (NULL == siocb->scm)
1088 		siocb->scm = &scm;
1089 	err = scm_send(sock, msg, siocb->scm);
1090 	if (err < 0)
1091 		return err;
1092 
1093 	if (msg->msg_namelen) {
1094 		if (addr->nl_family != AF_NETLINK)
1095 			return -EINVAL;
1096 		dst_pid = addr->nl_pid;
1097 		dst_group = ffs(addr->nl_groups);
1098 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1099 			return -EPERM;
1100 	} else {
1101 		dst_pid = nlk->dst_pid;
1102 		dst_group = nlk->dst_group;
1103 	}
1104 
1105 	if (!nlk->pid) {
1106 		err = netlink_autobind(sock);
1107 		if (err)
1108 			goto out;
1109 	}
1110 
1111 	err = -EMSGSIZE;
1112 	if (len > sk->sk_sndbuf - 32)
1113 		goto out;
1114 	err = -ENOBUFS;
1115 	skb = alloc_skb(len, GFP_KERNEL);
1116 	if (skb==NULL)
1117 		goto out;
1118 
1119 	NETLINK_CB(skb).pid	= nlk->pid;
1120 	NETLINK_CB(skb).dst_pid = dst_pid;
1121 	NETLINK_CB(skb).dst_group = dst_group;
1122 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1123 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1124 
1125 	/* What can I do? Netlink is asynchronous, so that
1126 	   we will have to save current capabilities to
1127 	   check them, when this message will be delivered
1128 	   to corresponding kernel module.   --ANK (980802)
1129 	 */
1130 
1131 	err = -EFAULT;
1132 	if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1133 		kfree_skb(skb);
1134 		goto out;
1135 	}
1136 
1137 	err = security_netlink_send(sk, skb);
1138 	if (err) {
1139 		kfree_skb(skb);
1140 		goto out;
1141 	}
1142 
1143 	if (dst_group) {
1144 		atomic_inc(&skb->users);
1145 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1146 	}
1147 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1148 
1149 out:
1150 	return err;
1151 }
1152 
1153 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1154 			   struct msghdr *msg, size_t len,
1155 			   int flags)
1156 {
1157 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1158 	struct scm_cookie scm;
1159 	struct sock *sk = sock->sk;
1160 	struct netlink_sock *nlk = nlk_sk(sk);
1161 	int noblock = flags&MSG_DONTWAIT;
1162 	size_t copied;
1163 	struct sk_buff *skb;
1164 	int err;
1165 
1166 	if (flags&MSG_OOB)
1167 		return -EOPNOTSUPP;
1168 
1169 	copied = 0;
1170 
1171 	skb = skb_recv_datagram(sk,flags,noblock,&err);
1172 	if (skb==NULL)
1173 		goto out;
1174 
1175 	msg->msg_namelen = 0;
1176 
1177 	copied = skb->len;
1178 	if (len < copied) {
1179 		msg->msg_flags |= MSG_TRUNC;
1180 		copied = len;
1181 	}
1182 
1183 	skb->h.raw = skb->data;
1184 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1185 
1186 	if (msg->msg_name) {
1187 		struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1188 		addr->nl_family = AF_NETLINK;
1189 		addr->nl_pad    = 0;
1190 		addr->nl_pid	= NETLINK_CB(skb).pid;
1191 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1192 		msg->msg_namelen = sizeof(*addr);
1193 	}
1194 
1195 	if (NULL == siocb->scm) {
1196 		memset(&scm, 0, sizeof(scm));
1197 		siocb->scm = &scm;
1198 	}
1199 	siocb->scm->creds = *NETLINK_CREDS(skb);
1200 	skb_free_datagram(sk, skb);
1201 
1202 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1203 		netlink_dump(sk);
1204 
1205 	scm_recv(sock, msg, siocb->scm, flags);
1206 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1207 		netlink_cmsg_recv_pktinfo(msg, skb);
1208 
1209 out:
1210 	netlink_rcv_wake(sk);
1211 	return err ? : copied;
1212 }
1213 
1214 static void netlink_data_ready(struct sock *sk, int len)
1215 {
1216 	struct netlink_sock *nlk = nlk_sk(sk);
1217 
1218 	if (nlk->data_ready)
1219 		nlk->data_ready(sk, len);
1220 	netlink_rcv_wake(sk);
1221 }
1222 
1223 /*
1224  *	We export these functions to other modules. They provide a
1225  *	complete set of kernel non-blocking support for message
1226  *	queueing.
1227  */
1228 
1229 struct sock *
1230 netlink_kernel_create(int unit, unsigned int groups,
1231                       void (*input)(struct sock *sk, int len),
1232                       struct module *module)
1233 {
1234 	struct socket *sock;
1235 	struct sock *sk;
1236 	struct netlink_sock *nlk;
1237 
1238 	if (!nl_table)
1239 		return NULL;
1240 
1241 	if (unit<0 || unit>=MAX_LINKS)
1242 		return NULL;
1243 
1244 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1245 		return NULL;
1246 
1247 	if (__netlink_create(sock, unit) < 0)
1248 		goto out_sock_release;
1249 
1250 	sk = sock->sk;
1251 	sk->sk_data_ready = netlink_data_ready;
1252 	if (input)
1253 		nlk_sk(sk)->data_ready = input;
1254 
1255 	if (netlink_insert(sk, 0))
1256 		goto out_sock_release;
1257 
1258 	nlk = nlk_sk(sk);
1259 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1260 
1261 	netlink_table_grab();
1262 	nl_table[unit].groups = groups < 32 ? 32 : groups;
1263 	nl_table[unit].module = module;
1264 	nl_table[unit].registered = 1;
1265 	netlink_table_ungrab();
1266 
1267 	return sk;
1268 
1269 out_sock_release:
1270 	sock_release(sock);
1271 	return NULL;
1272 }
1273 
1274 void netlink_set_nonroot(int protocol, unsigned int flags)
1275 {
1276 	if ((unsigned int)protocol < MAX_LINKS)
1277 		nl_table[protocol].nl_nonroot = flags;
1278 }
1279 
1280 static void netlink_destroy_callback(struct netlink_callback *cb)
1281 {
1282 	if (cb->skb)
1283 		kfree_skb(cb->skb);
1284 	kfree(cb);
1285 }
1286 
1287 /*
1288  * It looks a bit ugly.
1289  * It would be better to create kernel thread.
1290  */
1291 
1292 static int netlink_dump(struct sock *sk)
1293 {
1294 	struct netlink_sock *nlk = nlk_sk(sk);
1295 	struct netlink_callback *cb;
1296 	struct sk_buff *skb;
1297 	struct nlmsghdr *nlh;
1298 	int len;
1299 
1300 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1301 	if (!skb)
1302 		return -ENOBUFS;
1303 
1304 	spin_lock(&nlk->cb_lock);
1305 
1306 	cb = nlk->cb;
1307 	if (cb == NULL) {
1308 		spin_unlock(&nlk->cb_lock);
1309 		kfree_skb(skb);
1310 		return -EINVAL;
1311 	}
1312 
1313 	len = cb->dump(skb, cb);
1314 
1315 	if (len > 0) {
1316 		spin_unlock(&nlk->cb_lock);
1317 		skb_queue_tail(&sk->sk_receive_queue, skb);
1318 		sk->sk_data_ready(sk, len);
1319 		return 0;
1320 	}
1321 
1322 	nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1323 	memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1324 	skb_queue_tail(&sk->sk_receive_queue, skb);
1325 	sk->sk_data_ready(sk, skb->len);
1326 
1327 	if (cb->done)
1328 		cb->done(cb);
1329 	nlk->cb = NULL;
1330 	spin_unlock(&nlk->cb_lock);
1331 
1332 	netlink_destroy_callback(cb);
1333 	return 0;
1334 
1335 nlmsg_failure:
1336 	return -ENOBUFS;
1337 }
1338 
1339 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1340 		       struct nlmsghdr *nlh,
1341 		       int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1342 		       int (*done)(struct netlink_callback*))
1343 {
1344 	struct netlink_callback *cb;
1345 	struct sock *sk;
1346 	struct netlink_sock *nlk;
1347 
1348 	cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1349 	if (cb == NULL)
1350 		return -ENOBUFS;
1351 
1352 	memset(cb, 0, sizeof(*cb));
1353 	cb->dump = dump;
1354 	cb->done = done;
1355 	cb->nlh = nlh;
1356 	atomic_inc(&skb->users);
1357 	cb->skb = skb;
1358 
1359 	sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1360 	if (sk == NULL) {
1361 		netlink_destroy_callback(cb);
1362 		return -ECONNREFUSED;
1363 	}
1364 	nlk = nlk_sk(sk);
1365 	/* A dump is in progress... */
1366 	spin_lock(&nlk->cb_lock);
1367 	if (nlk->cb) {
1368 		spin_unlock(&nlk->cb_lock);
1369 		netlink_destroy_callback(cb);
1370 		sock_put(sk);
1371 		return -EBUSY;
1372 	}
1373 	nlk->cb = cb;
1374 	spin_unlock(&nlk->cb_lock);
1375 
1376 	netlink_dump(sk);
1377 	sock_put(sk);
1378 	return 0;
1379 }
1380 
1381 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1382 {
1383 	struct sk_buff *skb;
1384 	struct nlmsghdr *rep;
1385 	struct nlmsgerr *errmsg;
1386 	int size;
1387 
1388 	if (err == 0)
1389 		size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1390 	else
1391 		size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1392 
1393 	skb = alloc_skb(size, GFP_KERNEL);
1394 	if (!skb) {
1395 		struct sock *sk;
1396 
1397 		sk = netlink_lookup(in_skb->sk->sk_protocol,
1398 				    NETLINK_CB(in_skb).pid);
1399 		if (sk) {
1400 			sk->sk_err = ENOBUFS;
1401 			sk->sk_error_report(sk);
1402 			sock_put(sk);
1403 		}
1404 		return;
1405 	}
1406 
1407 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1408 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1409 	errmsg = NLMSG_DATA(rep);
1410 	errmsg->error = err;
1411 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1412 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1413 }
1414 
1415 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1416 						     struct nlmsghdr *, int *))
1417 {
1418 	unsigned int total_len;
1419 	struct nlmsghdr *nlh;
1420 	int err;
1421 
1422 	while (skb->len >= nlmsg_total_size(0)) {
1423 		nlh = (struct nlmsghdr *) skb->data;
1424 
1425 		if (skb->len < nlh->nlmsg_len)
1426 			return 0;
1427 
1428 		total_len = min(NLMSG_ALIGN(nlh->nlmsg_len), skb->len);
1429 
1430 		if (cb(skb, nlh, &err) < 0) {
1431 			/* Not an error, but we have to interrupt processing
1432 			 * here. Note: that in this case we do not pull
1433 			 * message from skb, it will be processed later.
1434 			 */
1435 			if (err == 0)
1436 				return -1;
1437 			netlink_ack(skb, nlh, err);
1438 		} else if (nlh->nlmsg_flags & NLM_F_ACK)
1439 			netlink_ack(skb, nlh, 0);
1440 
1441 		skb_pull(skb, total_len);
1442 	}
1443 
1444 	return 0;
1445 }
1446 
1447 /**
1448  * nelink_run_queue - Process netlink receive queue.
1449  * @sk: Netlink socket containing the queue
1450  * @qlen: Place to store queue length upon entry
1451  * @cb: Callback function invoked for each netlink message found
1452  *
1453  * Processes as much as there was in the queue upon entry and invokes
1454  * a callback function for each netlink message found. The callback
1455  * function may refuse a message by returning a negative error code
1456  * but setting the error pointer to 0 in which case this function
1457  * returns with a qlen != 0.
1458  *
1459  * qlen must be initialized to 0 before the initial entry, afterwards
1460  * the function may be called repeatedly until qlen reaches 0.
1461  */
1462 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1463 		       int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1464 {
1465 	struct sk_buff *skb;
1466 
1467 	if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1468 		*qlen = skb_queue_len(&sk->sk_receive_queue);
1469 
1470 	for (; *qlen; (*qlen)--) {
1471 		skb = skb_dequeue(&sk->sk_receive_queue);
1472 		if (netlink_rcv_skb(skb, cb)) {
1473 			if (skb->len)
1474 				skb_queue_head(&sk->sk_receive_queue, skb);
1475 			else {
1476 				kfree_skb(skb);
1477 				(*qlen)--;
1478 			}
1479 			break;
1480 		}
1481 
1482 		kfree_skb(skb);
1483 	}
1484 }
1485 
1486 /**
1487  * netlink_queue_skip - Skip netlink message while processing queue.
1488  * @nlh: Netlink message to be skipped
1489  * @skb: Socket buffer containing the netlink messages.
1490  *
1491  * Pulls the given netlink message off the socket buffer so the next
1492  * call to netlink_queue_run() will not reconsider the message.
1493  */
1494 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1495 {
1496 	int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1497 
1498 	if (msglen > skb->len)
1499 		msglen = skb->len;
1500 
1501 	skb_pull(skb, msglen);
1502 }
1503 
1504 #ifdef CONFIG_PROC_FS
1505 struct nl_seq_iter {
1506 	int link;
1507 	int hash_idx;
1508 };
1509 
1510 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1511 {
1512 	struct nl_seq_iter *iter = seq->private;
1513 	int i, j;
1514 	struct sock *s;
1515 	struct hlist_node *node;
1516 	loff_t off = 0;
1517 
1518 	for (i=0; i<MAX_LINKS; i++) {
1519 		struct nl_pid_hash *hash = &nl_table[i].hash;
1520 
1521 		for (j = 0; j <= hash->mask; j++) {
1522 			sk_for_each(s, node, &hash->table[j]) {
1523 				if (off == pos) {
1524 					iter->link = i;
1525 					iter->hash_idx = j;
1526 					return s;
1527 				}
1528 				++off;
1529 			}
1530 		}
1531 	}
1532 	return NULL;
1533 }
1534 
1535 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1536 {
1537 	read_lock(&nl_table_lock);
1538 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1539 }
1540 
1541 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1542 {
1543 	struct sock *s;
1544 	struct nl_seq_iter *iter;
1545 	int i, j;
1546 
1547 	++*pos;
1548 
1549 	if (v == SEQ_START_TOKEN)
1550 		return netlink_seq_socket_idx(seq, 0);
1551 
1552 	s = sk_next(v);
1553 	if (s)
1554 		return s;
1555 
1556 	iter = seq->private;
1557 	i = iter->link;
1558 	j = iter->hash_idx + 1;
1559 
1560 	do {
1561 		struct nl_pid_hash *hash = &nl_table[i].hash;
1562 
1563 		for (; j <= hash->mask; j++) {
1564 			s = sk_head(&hash->table[j]);
1565 			if (s) {
1566 				iter->link = i;
1567 				iter->hash_idx = j;
1568 				return s;
1569 			}
1570 		}
1571 
1572 		j = 0;
1573 	} while (++i < MAX_LINKS);
1574 
1575 	return NULL;
1576 }
1577 
1578 static void netlink_seq_stop(struct seq_file *seq, void *v)
1579 {
1580 	read_unlock(&nl_table_lock);
1581 }
1582 
1583 
1584 static int netlink_seq_show(struct seq_file *seq, void *v)
1585 {
1586 	if (v == SEQ_START_TOKEN)
1587 		seq_puts(seq,
1588 			 "sk       Eth Pid    Groups   "
1589 			 "Rmem     Wmem     Dump     Locks\n");
1590 	else {
1591 		struct sock *s = v;
1592 		struct netlink_sock *nlk = nlk_sk(s);
1593 
1594 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1595 			   s,
1596 			   s->sk_protocol,
1597 			   nlk->pid,
1598 			   nlk->groups ? (u32)nlk->groups[0] : 0,
1599 			   atomic_read(&s->sk_rmem_alloc),
1600 			   atomic_read(&s->sk_wmem_alloc),
1601 			   nlk->cb,
1602 			   atomic_read(&s->sk_refcnt)
1603 			);
1604 
1605 	}
1606 	return 0;
1607 }
1608 
1609 static struct seq_operations netlink_seq_ops = {
1610 	.start  = netlink_seq_start,
1611 	.next   = netlink_seq_next,
1612 	.stop   = netlink_seq_stop,
1613 	.show   = netlink_seq_show,
1614 };
1615 
1616 
1617 static int netlink_seq_open(struct inode *inode, struct file *file)
1618 {
1619 	struct seq_file *seq;
1620 	struct nl_seq_iter *iter;
1621 	int err;
1622 
1623 	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1624 	if (!iter)
1625 		return -ENOMEM;
1626 
1627 	err = seq_open(file, &netlink_seq_ops);
1628 	if (err) {
1629 		kfree(iter);
1630 		return err;
1631 	}
1632 
1633 	memset(iter, 0, sizeof(*iter));
1634 	seq = file->private_data;
1635 	seq->private = iter;
1636 	return 0;
1637 }
1638 
1639 static struct file_operations netlink_seq_fops = {
1640 	.owner		= THIS_MODULE,
1641 	.open		= netlink_seq_open,
1642 	.read		= seq_read,
1643 	.llseek		= seq_lseek,
1644 	.release	= seq_release_private,
1645 };
1646 
1647 #endif
1648 
1649 int netlink_register_notifier(struct notifier_block *nb)
1650 {
1651 	return notifier_chain_register(&netlink_chain, nb);
1652 }
1653 
1654 int netlink_unregister_notifier(struct notifier_block *nb)
1655 {
1656 	return notifier_chain_unregister(&netlink_chain, nb);
1657 }
1658 
1659 static const struct proto_ops netlink_ops = {
1660 	.family =	PF_NETLINK,
1661 	.owner =	THIS_MODULE,
1662 	.release =	netlink_release,
1663 	.bind =		netlink_bind,
1664 	.connect =	netlink_connect,
1665 	.socketpair =	sock_no_socketpair,
1666 	.accept =	sock_no_accept,
1667 	.getname =	netlink_getname,
1668 	.poll =		datagram_poll,
1669 	.ioctl =	sock_no_ioctl,
1670 	.listen =	sock_no_listen,
1671 	.shutdown =	sock_no_shutdown,
1672 	.setsockopt =	netlink_setsockopt,
1673 	.getsockopt =	netlink_getsockopt,
1674 	.sendmsg =	netlink_sendmsg,
1675 	.recvmsg =	netlink_recvmsg,
1676 	.mmap =		sock_no_mmap,
1677 	.sendpage =	sock_no_sendpage,
1678 };
1679 
1680 static struct net_proto_family netlink_family_ops = {
1681 	.family = PF_NETLINK,
1682 	.create = netlink_create,
1683 	.owner	= THIS_MODULE,	/* for consistency 8) */
1684 };
1685 
1686 extern void netlink_skb_parms_too_large(void);
1687 
1688 static int __init netlink_proto_init(void)
1689 {
1690 	struct sk_buff *dummy_skb;
1691 	int i;
1692 	unsigned long max;
1693 	unsigned int order;
1694 	int err = proto_register(&netlink_proto, 0);
1695 
1696 	if (err != 0)
1697 		goto out;
1698 
1699 	if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1700 		netlink_skb_parms_too_large();
1701 
1702 	nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1703 	if (!nl_table) {
1704 enomem:
1705 		printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1706 		return -ENOMEM;
1707 	}
1708 
1709 	memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1710 
1711 	if (num_physpages >= (128 * 1024))
1712 		max = num_physpages >> (21 - PAGE_SHIFT);
1713 	else
1714 		max = num_physpages >> (23 - PAGE_SHIFT);
1715 
1716 	order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1717 	max = (1UL << order) / sizeof(struct hlist_head);
1718 	order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1719 
1720 	for (i = 0; i < MAX_LINKS; i++) {
1721 		struct nl_pid_hash *hash = &nl_table[i].hash;
1722 
1723 		hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1724 		if (!hash->table) {
1725 			while (i-- > 0)
1726 				nl_pid_hash_free(nl_table[i].hash.table,
1727 						 1 * sizeof(*hash->table));
1728 			kfree(nl_table);
1729 			goto enomem;
1730 		}
1731 		memset(hash->table, 0, 1 * sizeof(*hash->table));
1732 		hash->max_shift = order;
1733 		hash->shift = 0;
1734 		hash->mask = 0;
1735 		hash->rehash_time = jiffies;
1736 	}
1737 
1738 	sock_register(&netlink_family_ops);
1739 #ifdef CONFIG_PROC_FS
1740 	proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1741 #endif
1742 	/* The netlink device handler may be needed early. */
1743 	rtnetlink_init();
1744 out:
1745 	return err;
1746 }
1747 
1748 core_initcall(netlink_proto_init);
1749 
1750 EXPORT_SYMBOL(netlink_ack);
1751 EXPORT_SYMBOL(netlink_run_queue);
1752 EXPORT_SYMBOL(netlink_queue_skip);
1753 EXPORT_SYMBOL(netlink_broadcast);
1754 EXPORT_SYMBOL(netlink_dump_start);
1755 EXPORT_SYMBOL(netlink_kernel_create);
1756 EXPORT_SYMBOL(netlink_register_notifier);
1757 EXPORT_SYMBOL(netlink_set_err);
1758 EXPORT_SYMBOL(netlink_set_nonroot);
1759 EXPORT_SYMBOL(netlink_unicast);
1760 EXPORT_SYMBOL(netlink_unregister_notifier);
1761 
1762