xref: /linux/net/netlink/af_netlink.c (revision 13abf8130139c2ccd4962a7e5a8902be5e6cb5a7)
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 
62 #define Nprintk(a...)
63 #define NLGRPSZ(x)	(ALIGN(x, sizeof(unsigned long) * 8) / 8)
64 
65 struct netlink_sock {
66 	/* struct sock has to be the first member of netlink_sock */
67 	struct sock		sk;
68 	u32			pid;
69 	u32			dst_pid;
70 	u32			dst_group;
71 	u32			flags;
72 	u32			subscriptions;
73 	u32			ngroups;
74 	unsigned long		*groups;
75 	unsigned long		state;
76 	wait_queue_head_t	wait;
77 	struct netlink_callback	*cb;
78 	spinlock_t		cb_lock;
79 	void			(*data_ready)(struct sock *sk, int bytes);
80 	struct module		*module;
81 };
82 
83 #define NETLINK_KERNEL_SOCKET	0x1
84 #define NETLINK_RECV_PKTINFO	0x2
85 
86 static inline struct netlink_sock *nlk_sk(struct sock *sk)
87 {
88 	return (struct netlink_sock *)sk;
89 }
90 
91 struct nl_pid_hash {
92 	struct hlist_head *table;
93 	unsigned long rehash_time;
94 
95 	unsigned int mask;
96 	unsigned int shift;
97 
98 	unsigned int entries;
99 	unsigned int max_shift;
100 
101 	u32 rnd;
102 };
103 
104 struct netlink_table {
105 	struct nl_pid_hash hash;
106 	struct hlist_head mc_list;
107 	unsigned int nl_nonroot;
108 	unsigned int groups;
109 	struct module *module;
110 	int registered;
111 };
112 
113 static struct netlink_table *nl_table;
114 
115 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
116 
117 static int netlink_dump(struct sock *sk);
118 static void netlink_destroy_callback(struct netlink_callback *cb);
119 
120 static DEFINE_RWLOCK(nl_table_lock);
121 static atomic_t nl_table_users = ATOMIC_INIT(0);
122 
123 static struct notifier_block *netlink_chain;
124 
125 static u32 netlink_group_mask(u32 group)
126 {
127 	return group ? 1 << (group - 1) : 0;
128 }
129 
130 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
131 {
132 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
133 }
134 
135 static void netlink_sock_destruct(struct sock *sk)
136 {
137 	skb_queue_purge(&sk->sk_receive_queue);
138 
139 	if (!sock_flag(sk, SOCK_DEAD)) {
140 		printk("Freeing alive netlink socket %p\n", sk);
141 		return;
142 	}
143 	BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
144 	BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
145 	BUG_TRAP(!nlk_sk(sk)->cb);
146 	BUG_TRAP(!nlk_sk(sk)->groups);
147 }
148 
149 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
150  * Look, when several writers sleep and reader wakes them up, all but one
151  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
152  * this, _but_ remember, it adds useless work on UP machines.
153  */
154 
155 static void netlink_table_grab(void)
156 {
157 	write_lock_bh(&nl_table_lock);
158 
159 	if (atomic_read(&nl_table_users)) {
160 		DECLARE_WAITQUEUE(wait, current);
161 
162 		add_wait_queue_exclusive(&nl_table_wait, &wait);
163 		for(;;) {
164 			set_current_state(TASK_UNINTERRUPTIBLE);
165 			if (atomic_read(&nl_table_users) == 0)
166 				break;
167 			write_unlock_bh(&nl_table_lock);
168 			schedule();
169 			write_lock_bh(&nl_table_lock);
170 		}
171 
172 		__set_current_state(TASK_RUNNING);
173 		remove_wait_queue(&nl_table_wait, &wait);
174 	}
175 }
176 
177 static __inline__ void netlink_table_ungrab(void)
178 {
179 	write_unlock_bh(&nl_table_lock);
180 	wake_up(&nl_table_wait);
181 }
182 
183 static __inline__ void
184 netlink_lock_table(void)
185 {
186 	/* read_lock() synchronizes us to netlink_table_grab */
187 
188 	read_lock(&nl_table_lock);
189 	atomic_inc(&nl_table_users);
190 	read_unlock(&nl_table_lock);
191 }
192 
193 static __inline__ void
194 netlink_unlock_table(void)
195 {
196 	if (atomic_dec_and_test(&nl_table_users))
197 		wake_up(&nl_table_wait);
198 }
199 
200 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
201 {
202 	struct nl_pid_hash *hash = &nl_table[protocol].hash;
203 	struct hlist_head *head;
204 	struct sock *sk;
205 	struct hlist_node *node;
206 
207 	read_lock(&nl_table_lock);
208 	head = nl_pid_hashfn(hash, pid);
209 	sk_for_each(sk, node, head) {
210 		if (nlk_sk(sk)->pid == pid) {
211 			sock_hold(sk);
212 			goto found;
213 		}
214 	}
215 	sk = NULL;
216 found:
217 	read_unlock(&nl_table_lock);
218 	return sk;
219 }
220 
221 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
222 {
223 	if (size <= PAGE_SIZE)
224 		return kmalloc(size, GFP_ATOMIC);
225 	else
226 		return (struct hlist_head *)
227 			__get_free_pages(GFP_ATOMIC, get_order(size));
228 }
229 
230 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
231 {
232 	if (size <= PAGE_SIZE)
233 		kfree(table);
234 	else
235 		free_pages((unsigned long)table, get_order(size));
236 }
237 
238 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
239 {
240 	unsigned int omask, mask, shift;
241 	size_t osize, size;
242 	struct hlist_head *otable, *table;
243 	int i;
244 
245 	omask = mask = hash->mask;
246 	osize = size = (mask + 1) * sizeof(*table);
247 	shift = hash->shift;
248 
249 	if (grow) {
250 		if (++shift > hash->max_shift)
251 			return 0;
252 		mask = mask * 2 + 1;
253 		size *= 2;
254 	}
255 
256 	table = nl_pid_hash_alloc(size);
257 	if (!table)
258 		return 0;
259 
260 	memset(table, 0, size);
261 	otable = hash->table;
262 	hash->table = table;
263 	hash->mask = mask;
264 	hash->shift = shift;
265 	get_random_bytes(&hash->rnd, sizeof(hash->rnd));
266 
267 	for (i = 0; i <= omask; i++) {
268 		struct sock *sk;
269 		struct hlist_node *node, *tmp;
270 
271 		sk_for_each_safe(sk, node, tmp, &otable[i])
272 			__sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
273 	}
274 
275 	nl_pid_hash_free(otable, osize);
276 	hash->rehash_time = jiffies + 10 * 60 * HZ;
277 	return 1;
278 }
279 
280 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
281 {
282 	int avg = hash->entries >> hash->shift;
283 
284 	if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
285 		return 1;
286 
287 	if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
288 		nl_pid_hash_rehash(hash, 0);
289 		return 1;
290 	}
291 
292 	return 0;
293 }
294 
295 static struct proto_ops netlink_ops;
296 
297 static int netlink_insert(struct sock *sk, u32 pid)
298 {
299 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
300 	struct hlist_head *head;
301 	int err = -EADDRINUSE;
302 	struct sock *osk;
303 	struct hlist_node *node;
304 	int len;
305 
306 	netlink_table_grab();
307 	head = nl_pid_hashfn(hash, pid);
308 	len = 0;
309 	sk_for_each(osk, node, head) {
310 		if (nlk_sk(osk)->pid == pid)
311 			break;
312 		len++;
313 	}
314 	if (node)
315 		goto err;
316 
317 	err = -EBUSY;
318 	if (nlk_sk(sk)->pid)
319 		goto err;
320 
321 	err = -ENOMEM;
322 	if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
323 		goto err;
324 
325 	if (len && nl_pid_hash_dilute(hash, len))
326 		head = nl_pid_hashfn(hash, pid);
327 	hash->entries++;
328 	nlk_sk(sk)->pid = pid;
329 	sk_add_node(sk, head);
330 	err = 0;
331 
332 err:
333 	netlink_table_ungrab();
334 	return err;
335 }
336 
337 static void netlink_remove(struct sock *sk)
338 {
339 	netlink_table_grab();
340 	if (sk_del_node_init(sk))
341 		nl_table[sk->sk_protocol].hash.entries--;
342 	if (nlk_sk(sk)->subscriptions)
343 		__sk_del_bind_node(sk);
344 	netlink_table_ungrab();
345 }
346 
347 static struct proto netlink_proto = {
348 	.name	  = "NETLINK",
349 	.owner	  = THIS_MODULE,
350 	.obj_size = sizeof(struct netlink_sock),
351 };
352 
353 static int __netlink_create(struct socket *sock, int protocol)
354 {
355 	struct sock *sk;
356 	struct netlink_sock *nlk;
357 
358 	sock->ops = &netlink_ops;
359 
360 	sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
361 	if (!sk)
362 		return -ENOMEM;
363 
364 	sock_init_data(sock, sk);
365 
366 	nlk = nlk_sk(sk);
367 	spin_lock_init(&nlk->cb_lock);
368 	init_waitqueue_head(&nlk->wait);
369 
370 	sk->sk_destruct = netlink_sock_destruct;
371 	sk->sk_protocol = protocol;
372 	return 0;
373 }
374 
375 static int netlink_create(struct socket *sock, int protocol)
376 {
377 	struct module *module = NULL;
378 	struct netlink_sock *nlk;
379 	unsigned int groups;
380 	int err = 0;
381 
382 	sock->state = SS_UNCONNECTED;
383 
384 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
385 		return -ESOCKTNOSUPPORT;
386 
387 	if (protocol<0 || protocol >= MAX_LINKS)
388 		return -EPROTONOSUPPORT;
389 
390 	netlink_lock_table();
391 #ifdef CONFIG_KMOD
392 	if (!nl_table[protocol].registered) {
393 		netlink_unlock_table();
394 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
395 		netlink_lock_table();
396 	}
397 #endif
398 	if (nl_table[protocol].registered &&
399 	    try_module_get(nl_table[protocol].module))
400 		module = nl_table[protocol].module;
401 	else
402 		err = -EPROTONOSUPPORT;
403 	groups = nl_table[protocol].groups;
404 	netlink_unlock_table();
405 
406 	if (err || (err = __netlink_create(sock, protocol) < 0))
407 		goto out_module;
408 
409 	nlk = nlk_sk(sock->sk);
410 
411 	nlk->groups = kmalloc(NLGRPSZ(groups), GFP_KERNEL);
412 	if (nlk->groups == NULL) {
413 		err = -ENOMEM;
414 		goto out_module;
415 	}
416 	memset(nlk->groups, 0, NLGRPSZ(groups));
417 	nlk->ngroups = groups;
418 
419 	nlk->module = module;
420 out:
421 	return err;
422 
423 out_module:
424 	module_put(module);
425 	goto out;
426 }
427 
428 static int netlink_release(struct socket *sock)
429 {
430 	struct sock *sk = sock->sk;
431 	struct netlink_sock *nlk;
432 
433 	if (!sk)
434 		return 0;
435 
436 	netlink_remove(sk);
437 	nlk = nlk_sk(sk);
438 
439 	spin_lock(&nlk->cb_lock);
440 	if (nlk->cb) {
441 		nlk->cb->done(nlk->cb);
442 		netlink_destroy_callback(nlk->cb);
443 		nlk->cb = NULL;
444 	}
445 	spin_unlock(&nlk->cb_lock);
446 
447 	/* OK. Socket is unlinked, and, therefore,
448 	   no new packets will arrive */
449 
450 	sock_orphan(sk);
451 	sock->sk = NULL;
452 	wake_up_interruptible_all(&nlk->wait);
453 
454 	skb_queue_purge(&sk->sk_write_queue);
455 
456 	if (nlk->pid && !nlk->subscriptions) {
457 		struct netlink_notify n = {
458 						.protocol = sk->sk_protocol,
459 						.pid = nlk->pid,
460 					  };
461 		notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
462 	}
463 
464 	if (nlk->module)
465 		module_put(nlk->module);
466 
467 	if (nlk->flags & NETLINK_KERNEL_SOCKET) {
468 		netlink_table_grab();
469 		nl_table[sk->sk_protocol].module = NULL;
470 		nl_table[sk->sk_protocol].registered = 0;
471 		netlink_table_ungrab();
472 	}
473 
474 	kfree(nlk->groups);
475 	nlk->groups = NULL;
476 
477 	sock_put(sk);
478 	return 0;
479 }
480 
481 static int netlink_autobind(struct socket *sock)
482 {
483 	struct sock *sk = sock->sk;
484 	struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
485 	struct hlist_head *head;
486 	struct sock *osk;
487 	struct hlist_node *node;
488 	s32 pid = current->pid;
489 	int err;
490 	static s32 rover = -4097;
491 
492 retry:
493 	cond_resched();
494 	netlink_table_grab();
495 	head = nl_pid_hashfn(hash, pid);
496 	sk_for_each(osk, node, head) {
497 		if (nlk_sk(osk)->pid == pid) {
498 			/* Bind collision, search negative pid values. */
499 			pid = rover--;
500 			if (rover > -4097)
501 				rover = -4097;
502 			netlink_table_ungrab();
503 			goto retry;
504 		}
505 	}
506 	netlink_table_ungrab();
507 
508 	err = netlink_insert(sk, pid);
509 	if (err == -EADDRINUSE)
510 		goto retry;
511 
512 	/* If 2 threads race to autobind, that is fine.  */
513 	if (err == -EBUSY)
514 		err = 0;
515 
516 	return err;
517 }
518 
519 static inline int netlink_capable(struct socket *sock, unsigned int flag)
520 {
521 	return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
522 	       capable(CAP_NET_ADMIN);
523 }
524 
525 static void
526 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
527 {
528 	struct netlink_sock *nlk = nlk_sk(sk);
529 
530 	if (nlk->subscriptions && !subscriptions)
531 		__sk_del_bind_node(sk);
532 	else if (!nlk->subscriptions && subscriptions)
533 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
534 	nlk->subscriptions = subscriptions;
535 }
536 
537 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
538 {
539 	struct sock *sk = sock->sk;
540 	struct netlink_sock *nlk = nlk_sk(sk);
541 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
542 	int err;
543 
544 	if (nladdr->nl_family != AF_NETLINK)
545 		return -EINVAL;
546 
547 	/* Only superuser is allowed to listen multicasts */
548 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV))
549 		return -EPERM;
550 
551 	if (nlk->pid) {
552 		if (nladdr->nl_pid != nlk->pid)
553 			return -EINVAL;
554 	} else {
555 		err = nladdr->nl_pid ?
556 			netlink_insert(sk, nladdr->nl_pid) :
557 			netlink_autobind(sock);
558 		if (err)
559 			return err;
560 	}
561 
562 	if (!nladdr->nl_groups && !(u32)nlk->groups[0])
563 		return 0;
564 
565 	netlink_table_grab();
566 	netlink_update_subscriptions(sk, nlk->subscriptions +
567 	                                 hweight32(nladdr->nl_groups) -
568 	                                 hweight32(nlk->groups[0]));
569 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
570 	netlink_table_ungrab();
571 
572 	return 0;
573 }
574 
575 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
576 			   int alen, int flags)
577 {
578 	int err = 0;
579 	struct sock *sk = sock->sk;
580 	struct netlink_sock *nlk = nlk_sk(sk);
581 	struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
582 
583 	if (addr->sa_family == AF_UNSPEC) {
584 		sk->sk_state	= NETLINK_UNCONNECTED;
585 		nlk->dst_pid	= 0;
586 		nlk->dst_group  = 0;
587 		return 0;
588 	}
589 	if (addr->sa_family != AF_NETLINK)
590 		return -EINVAL;
591 
592 	/* Only superuser is allowed to send multicasts */
593 	if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
594 		return -EPERM;
595 
596 	if (!nlk->pid)
597 		err = netlink_autobind(sock);
598 
599 	if (err == 0) {
600 		sk->sk_state	= NETLINK_CONNECTED;
601 		nlk->dst_pid 	= nladdr->nl_pid;
602 		nlk->dst_group  = ffs(nladdr->nl_groups);
603 	}
604 
605 	return err;
606 }
607 
608 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
609 {
610 	struct sock *sk = sock->sk;
611 	struct netlink_sock *nlk = nlk_sk(sk);
612 	struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
613 
614 	nladdr->nl_family = AF_NETLINK;
615 	nladdr->nl_pad = 0;
616 	*addr_len = sizeof(*nladdr);
617 
618 	if (peer) {
619 		nladdr->nl_pid = nlk->dst_pid;
620 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
621 	} else {
622 		nladdr->nl_pid = nlk->pid;
623 		nladdr->nl_groups = nlk->groups[0];
624 	}
625 	return 0;
626 }
627 
628 static void netlink_overrun(struct sock *sk)
629 {
630 	if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
631 		sk->sk_err = ENOBUFS;
632 		sk->sk_error_report(sk);
633 	}
634 }
635 
636 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
637 {
638 	int protocol = ssk->sk_protocol;
639 	struct sock *sock;
640 	struct netlink_sock *nlk;
641 
642 	sock = netlink_lookup(protocol, pid);
643 	if (!sock)
644 		return ERR_PTR(-ECONNREFUSED);
645 
646 	/* Don't bother queuing skb if kernel socket has no input function */
647 	nlk = nlk_sk(sock);
648 	if ((nlk->pid == 0 && !nlk->data_ready) ||
649 	    (sock->sk_state == NETLINK_CONNECTED &&
650 	     nlk->dst_pid != nlk_sk(ssk)->pid)) {
651 		sock_put(sock);
652 		return ERR_PTR(-ECONNREFUSED);
653 	}
654 	return sock;
655 }
656 
657 struct sock *netlink_getsockbyfilp(struct file *filp)
658 {
659 	struct inode *inode = filp->f_dentry->d_inode;
660 	struct sock *sock;
661 
662 	if (!S_ISSOCK(inode->i_mode))
663 		return ERR_PTR(-ENOTSOCK);
664 
665 	sock = SOCKET_I(inode)->sk;
666 	if (sock->sk_family != AF_NETLINK)
667 		return ERR_PTR(-EINVAL);
668 
669 	sock_hold(sock);
670 	return sock;
671 }
672 
673 /*
674  * Attach a skb to a netlink socket.
675  * The caller must hold a reference to the destination socket. On error, the
676  * reference is dropped. The skb is not send to the destination, just all
677  * all error checks are performed and memory in the queue is reserved.
678  * Return values:
679  * < 0: error. skb freed, reference to sock dropped.
680  * 0: continue
681  * 1: repeat lookup - reference dropped while waiting for socket memory.
682  */
683 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo)
684 {
685 	struct netlink_sock *nlk;
686 
687 	nlk = nlk_sk(sk);
688 
689 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
690 	    test_bit(0, &nlk->state)) {
691 		DECLARE_WAITQUEUE(wait, current);
692 		if (!timeo) {
693 			if (!nlk->pid)
694 				netlink_overrun(sk);
695 			sock_put(sk);
696 			kfree_skb(skb);
697 			return -EAGAIN;
698 		}
699 
700 		__set_current_state(TASK_INTERRUPTIBLE);
701 		add_wait_queue(&nlk->wait, &wait);
702 
703 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
704 		     test_bit(0, &nlk->state)) &&
705 		    !sock_flag(sk, SOCK_DEAD))
706 			timeo = schedule_timeout(timeo);
707 
708 		__set_current_state(TASK_RUNNING);
709 		remove_wait_queue(&nlk->wait, &wait);
710 		sock_put(sk);
711 
712 		if (signal_pending(current)) {
713 			kfree_skb(skb);
714 			return sock_intr_errno(timeo);
715 		}
716 		return 1;
717 	}
718 	skb_set_owner_r(skb, sk);
719 	return 0;
720 }
721 
722 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
723 {
724 	struct netlink_sock *nlk;
725 	int len = skb->len;
726 
727 	nlk = nlk_sk(sk);
728 
729 	skb_queue_tail(&sk->sk_receive_queue, skb);
730 	sk->sk_data_ready(sk, len);
731 	sock_put(sk);
732 	return len;
733 }
734 
735 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
736 {
737 	kfree_skb(skb);
738 	sock_put(sk);
739 }
740 
741 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
742 					   unsigned int __nocast allocation)
743 {
744 	int delta;
745 
746 	skb_orphan(skb);
747 
748 	delta = skb->end - skb->tail;
749 	if (delta * 2 < skb->truesize)
750 		return skb;
751 
752 	if (skb_shared(skb)) {
753 		struct sk_buff *nskb = skb_clone(skb, allocation);
754 		if (!nskb)
755 			return skb;
756 		kfree_skb(skb);
757 		skb = nskb;
758 	}
759 
760 	if (!pskb_expand_head(skb, 0, -delta, allocation))
761 		skb->truesize -= delta;
762 
763 	return skb;
764 }
765 
766 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
767 {
768 	struct sock *sk;
769 	int err;
770 	long timeo;
771 
772 	skb = netlink_trim(skb, gfp_any());
773 
774 	timeo = sock_sndtimeo(ssk, nonblock);
775 retry:
776 	sk = netlink_getsockbypid(ssk, pid);
777 	if (IS_ERR(sk)) {
778 		kfree_skb(skb);
779 		return PTR_ERR(sk);
780 	}
781 	err = netlink_attachskb(sk, skb, nonblock, timeo);
782 	if (err == 1)
783 		goto retry;
784 	if (err)
785 		return err;
786 
787 	return netlink_sendskb(sk, skb, ssk->sk_protocol);
788 }
789 
790 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
791 {
792 	struct netlink_sock *nlk = nlk_sk(sk);
793 
794 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
795 	    !test_bit(0, &nlk->state)) {
796 		skb_set_owner_r(skb, sk);
797 		skb_queue_tail(&sk->sk_receive_queue, skb);
798 		sk->sk_data_ready(sk, skb->len);
799 		return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
800 	}
801 	return -1;
802 }
803 
804 struct netlink_broadcast_data {
805 	struct sock *exclude_sk;
806 	u32 pid;
807 	u32 group;
808 	int failure;
809 	int congested;
810 	int delivered;
811 	unsigned int allocation;
812 	struct sk_buff *skb, *skb2;
813 };
814 
815 static inline int do_one_broadcast(struct sock *sk,
816 				   struct netlink_broadcast_data *p)
817 {
818 	struct netlink_sock *nlk = nlk_sk(sk);
819 	int val;
820 
821 	if (p->exclude_sk == sk)
822 		goto out;
823 
824 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
825 	    !test_bit(p->group - 1, nlk->groups))
826 		goto out;
827 
828 	if (p->failure) {
829 		netlink_overrun(sk);
830 		goto out;
831 	}
832 
833 	sock_hold(sk);
834 	if (p->skb2 == NULL) {
835 		if (skb_shared(p->skb)) {
836 			p->skb2 = skb_clone(p->skb, p->allocation);
837 		} else {
838 			p->skb2 = skb_get(p->skb);
839 			/*
840 			 * skb ownership may have been set when
841 			 * delivered to a previous socket.
842 			 */
843 			skb_orphan(p->skb2);
844 		}
845 	}
846 	if (p->skb2 == NULL) {
847 		netlink_overrun(sk);
848 		/* Clone failed. Notify ALL listeners. */
849 		p->failure = 1;
850 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
851 		netlink_overrun(sk);
852 	} else {
853 		p->congested |= val;
854 		p->delivered = 1;
855 		p->skb2 = NULL;
856 	}
857 	sock_put(sk);
858 
859 out:
860 	return 0;
861 }
862 
863 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
864 		      u32 group, unsigned int __nocast allocation)
865 {
866 	struct netlink_broadcast_data info;
867 	struct hlist_node *node;
868 	struct sock *sk;
869 
870 	skb = netlink_trim(skb, allocation);
871 
872 	info.exclude_sk = ssk;
873 	info.pid = pid;
874 	info.group = group;
875 	info.failure = 0;
876 	info.congested = 0;
877 	info.delivered = 0;
878 	info.allocation = allocation;
879 	info.skb = skb;
880 	info.skb2 = NULL;
881 
882 	/* While we sleep in clone, do not allow to change socket list */
883 
884 	netlink_lock_table();
885 
886 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
887 		do_one_broadcast(sk, &info);
888 
889 	kfree_skb(skb);
890 
891 	netlink_unlock_table();
892 
893 	if (info.skb2)
894 		kfree_skb(info.skb2);
895 
896 	if (info.delivered) {
897 		if (info.congested && (allocation & __GFP_WAIT))
898 			yield();
899 		return 0;
900 	}
901 	if (info.failure)
902 		return -ENOBUFS;
903 	return -ESRCH;
904 }
905 
906 struct netlink_set_err_data {
907 	struct sock *exclude_sk;
908 	u32 pid;
909 	u32 group;
910 	int code;
911 };
912 
913 static inline int do_one_set_err(struct sock *sk,
914 				 struct netlink_set_err_data *p)
915 {
916 	struct netlink_sock *nlk = nlk_sk(sk);
917 
918 	if (sk == p->exclude_sk)
919 		goto out;
920 
921 	if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
922 	    !test_bit(p->group - 1, nlk->groups))
923 		goto out;
924 
925 	sk->sk_err = p->code;
926 	sk->sk_error_report(sk);
927 out:
928 	return 0;
929 }
930 
931 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
932 {
933 	struct netlink_set_err_data info;
934 	struct hlist_node *node;
935 	struct sock *sk;
936 
937 	info.exclude_sk = ssk;
938 	info.pid = pid;
939 	info.group = group;
940 	info.code = code;
941 
942 	read_lock(&nl_table_lock);
943 
944 	sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
945 		do_one_set_err(sk, &info);
946 
947 	read_unlock(&nl_table_lock);
948 }
949 
950 static int netlink_setsockopt(struct socket *sock, int level, int optname,
951                               char __user *optval, int optlen)
952 {
953 	struct sock *sk = sock->sk;
954 	struct netlink_sock *nlk = nlk_sk(sk);
955 	int val = 0, err;
956 
957 	if (level != SOL_NETLINK)
958 		return -ENOPROTOOPT;
959 
960 	if (optlen >= sizeof(int) &&
961 	    get_user(val, (int __user *)optval))
962 		return -EFAULT;
963 
964 	switch (optname) {
965 	case NETLINK_PKTINFO:
966 		if (val)
967 			nlk->flags |= NETLINK_RECV_PKTINFO;
968 		else
969 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
970 		err = 0;
971 		break;
972 	case NETLINK_ADD_MEMBERSHIP:
973 	case NETLINK_DROP_MEMBERSHIP: {
974 		unsigned int subscriptions;
975 		int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
976 
977 		if (!netlink_capable(sock, NL_NONROOT_RECV))
978 			return -EPERM;
979 		if (!val || val - 1 >= nlk->ngroups)
980 			return -EINVAL;
981 		netlink_table_grab();
982 		old = test_bit(val - 1, nlk->groups);
983 		subscriptions = nlk->subscriptions - old + new;
984 		if (new)
985 			__set_bit(val - 1, nlk->groups);
986 		else
987 			__clear_bit(val - 1, nlk->groups);
988 		netlink_update_subscriptions(sk, subscriptions);
989 		netlink_table_ungrab();
990 		err = 0;
991 		break;
992 	}
993 	default:
994 		err = -ENOPROTOOPT;
995 	}
996 	return err;
997 }
998 
999 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1000                               char __user *optval, int __user *optlen)
1001 {
1002 	struct sock *sk = sock->sk;
1003 	struct netlink_sock *nlk = nlk_sk(sk);
1004 	int len, val, err;
1005 
1006 	if (level != SOL_NETLINK)
1007 		return -ENOPROTOOPT;
1008 
1009 	if (get_user(len, optlen))
1010 		return -EFAULT;
1011 	if (len < 0)
1012 		return -EINVAL;
1013 
1014 	switch (optname) {
1015 	case NETLINK_PKTINFO:
1016 		if (len < sizeof(int))
1017 			return -EINVAL;
1018 		len = sizeof(int);
1019 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1020 		put_user(len, optlen);
1021 		put_user(val, optval);
1022 		err = 0;
1023 		break;
1024 	default:
1025 		err = -ENOPROTOOPT;
1026 	}
1027 	return err;
1028 }
1029 
1030 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1031 {
1032 	struct nl_pktinfo info;
1033 
1034 	info.group = NETLINK_CB(skb).dst_group;
1035 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1036 }
1037 
1038 static inline void netlink_rcv_wake(struct sock *sk)
1039 {
1040 	struct netlink_sock *nlk = nlk_sk(sk);
1041 
1042 	if (skb_queue_empty(&sk->sk_receive_queue))
1043 		clear_bit(0, &nlk->state);
1044 	if (!test_bit(0, &nlk->state))
1045 		wake_up_interruptible(&nlk->wait);
1046 }
1047 
1048 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1049 			   struct msghdr *msg, size_t len)
1050 {
1051 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1052 	struct sock *sk = sock->sk;
1053 	struct netlink_sock *nlk = nlk_sk(sk);
1054 	struct sockaddr_nl *addr=msg->msg_name;
1055 	u32 dst_pid;
1056 	u32 dst_group;
1057 	struct sk_buff *skb;
1058 	int err;
1059 	struct scm_cookie scm;
1060 
1061 	if (msg->msg_flags&MSG_OOB)
1062 		return -EOPNOTSUPP;
1063 
1064 	if (NULL == siocb->scm)
1065 		siocb->scm = &scm;
1066 	err = scm_send(sock, msg, siocb->scm);
1067 	if (err < 0)
1068 		return err;
1069 
1070 	if (msg->msg_namelen) {
1071 		if (addr->nl_family != AF_NETLINK)
1072 			return -EINVAL;
1073 		dst_pid = addr->nl_pid;
1074 		dst_group = ffs(addr->nl_groups);
1075 		if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1076 			return -EPERM;
1077 	} else {
1078 		dst_pid = nlk->dst_pid;
1079 		dst_group = nlk->dst_group;
1080 	}
1081 
1082 	if (!nlk->pid) {
1083 		err = netlink_autobind(sock);
1084 		if (err)
1085 			goto out;
1086 	}
1087 
1088 	err = -EMSGSIZE;
1089 	if (len > sk->sk_sndbuf - 32)
1090 		goto out;
1091 	err = -ENOBUFS;
1092 	skb = alloc_skb(len, GFP_KERNEL);
1093 	if (skb==NULL)
1094 		goto out;
1095 
1096 	NETLINK_CB(skb).pid	= nlk->pid;
1097 	NETLINK_CB(skb).dst_pid = dst_pid;
1098 	NETLINK_CB(skb).dst_group = dst_group;
1099 	NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1100 	memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1101 
1102 	/* What can I do? Netlink is asynchronous, so that
1103 	   we will have to save current capabilities to
1104 	   check them, when this message will be delivered
1105 	   to corresponding kernel module.   --ANK (980802)
1106 	 */
1107 
1108 	err = -EFAULT;
1109 	if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1110 		kfree_skb(skb);
1111 		goto out;
1112 	}
1113 
1114 	err = security_netlink_send(sk, skb);
1115 	if (err) {
1116 		kfree_skb(skb);
1117 		goto out;
1118 	}
1119 
1120 	if (dst_group) {
1121 		atomic_inc(&skb->users);
1122 		netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1123 	}
1124 	err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1125 
1126 out:
1127 	return err;
1128 }
1129 
1130 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1131 			   struct msghdr *msg, size_t len,
1132 			   int flags)
1133 {
1134 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1135 	struct scm_cookie scm;
1136 	struct sock *sk = sock->sk;
1137 	struct netlink_sock *nlk = nlk_sk(sk);
1138 	int noblock = flags&MSG_DONTWAIT;
1139 	size_t copied;
1140 	struct sk_buff *skb;
1141 	int err;
1142 
1143 	if (flags&MSG_OOB)
1144 		return -EOPNOTSUPP;
1145 
1146 	copied = 0;
1147 
1148 	skb = skb_recv_datagram(sk,flags,noblock,&err);
1149 	if (skb==NULL)
1150 		goto out;
1151 
1152 	msg->msg_namelen = 0;
1153 
1154 	copied = skb->len;
1155 	if (len < copied) {
1156 		msg->msg_flags |= MSG_TRUNC;
1157 		copied = len;
1158 	}
1159 
1160 	skb->h.raw = skb->data;
1161 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1162 
1163 	if (msg->msg_name) {
1164 		struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1165 		addr->nl_family = AF_NETLINK;
1166 		addr->nl_pad    = 0;
1167 		addr->nl_pid	= NETLINK_CB(skb).pid;
1168 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1169 		msg->msg_namelen = sizeof(*addr);
1170 	}
1171 
1172 	if (NULL == siocb->scm) {
1173 		memset(&scm, 0, sizeof(scm));
1174 		siocb->scm = &scm;
1175 	}
1176 	siocb->scm->creds = *NETLINK_CREDS(skb);
1177 	skb_free_datagram(sk, skb);
1178 
1179 	if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1180 		netlink_dump(sk);
1181 
1182 	scm_recv(sock, msg, siocb->scm, flags);
1183 	if (nlk->flags & NETLINK_RECV_PKTINFO)
1184 		netlink_cmsg_recv_pktinfo(msg, skb);
1185 
1186 out:
1187 	netlink_rcv_wake(sk);
1188 	return err ? : copied;
1189 }
1190 
1191 static void netlink_data_ready(struct sock *sk, int len)
1192 {
1193 	struct netlink_sock *nlk = nlk_sk(sk);
1194 
1195 	if (nlk->data_ready)
1196 		nlk->data_ready(sk, len);
1197 	netlink_rcv_wake(sk);
1198 }
1199 
1200 /*
1201  *	We export these functions to other modules. They provide a
1202  *	complete set of kernel non-blocking support for message
1203  *	queueing.
1204  */
1205 
1206 struct sock *
1207 netlink_kernel_create(int unit, unsigned int groups,
1208                       void (*input)(struct sock *sk, int len),
1209                       struct module *module)
1210 {
1211 	struct socket *sock;
1212 	struct sock *sk;
1213 	struct netlink_sock *nlk;
1214 
1215 	if (!nl_table)
1216 		return NULL;
1217 
1218 	if (unit<0 || unit>=MAX_LINKS)
1219 		return NULL;
1220 
1221 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1222 		return NULL;
1223 
1224 	if (__netlink_create(sock, unit) < 0)
1225 		goto out_sock_release;
1226 
1227 	sk = sock->sk;
1228 	sk->sk_data_ready = netlink_data_ready;
1229 	if (input)
1230 		nlk_sk(sk)->data_ready = input;
1231 
1232 	if (netlink_insert(sk, 0))
1233 		goto out_sock_release;
1234 
1235 	nlk = nlk_sk(sk);
1236 	nlk->flags |= NETLINK_KERNEL_SOCKET;
1237 
1238 	netlink_table_grab();
1239 	nl_table[unit].groups = groups < 32 ? 32 : groups;
1240 	nl_table[unit].module = module;
1241 	nl_table[unit].registered = 1;
1242 	netlink_table_ungrab();
1243 
1244 	return sk;
1245 
1246 out_sock_release:
1247 	sock_release(sock);
1248 	return NULL;
1249 }
1250 
1251 void netlink_set_nonroot(int protocol, unsigned int flags)
1252 {
1253 	if ((unsigned int)protocol < MAX_LINKS)
1254 		nl_table[protocol].nl_nonroot = flags;
1255 }
1256 
1257 static void netlink_destroy_callback(struct netlink_callback *cb)
1258 {
1259 	if (cb->skb)
1260 		kfree_skb(cb->skb);
1261 	kfree(cb);
1262 }
1263 
1264 /*
1265  * It looks a bit ugly.
1266  * It would be better to create kernel thread.
1267  */
1268 
1269 static int netlink_dump(struct sock *sk)
1270 {
1271 	struct netlink_sock *nlk = nlk_sk(sk);
1272 	struct netlink_callback *cb;
1273 	struct sk_buff *skb;
1274 	struct nlmsghdr *nlh;
1275 	int len;
1276 
1277 	skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1278 	if (!skb)
1279 		return -ENOBUFS;
1280 
1281 	spin_lock(&nlk->cb_lock);
1282 
1283 	cb = nlk->cb;
1284 	if (cb == NULL) {
1285 		spin_unlock(&nlk->cb_lock);
1286 		kfree_skb(skb);
1287 		return -EINVAL;
1288 	}
1289 
1290 	len = cb->dump(skb, cb);
1291 
1292 	if (len > 0) {
1293 		spin_unlock(&nlk->cb_lock);
1294 		skb_queue_tail(&sk->sk_receive_queue, skb);
1295 		sk->sk_data_ready(sk, len);
1296 		return 0;
1297 	}
1298 
1299 	nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1300 	memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1301 	skb_queue_tail(&sk->sk_receive_queue, skb);
1302 	sk->sk_data_ready(sk, skb->len);
1303 
1304 	cb->done(cb);
1305 	nlk->cb = NULL;
1306 	spin_unlock(&nlk->cb_lock);
1307 
1308 	netlink_destroy_callback(cb);
1309 	return 0;
1310 
1311 nlmsg_failure:
1312 	return -ENOBUFS;
1313 }
1314 
1315 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1316 		       struct nlmsghdr *nlh,
1317 		       int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1318 		       int (*done)(struct netlink_callback*))
1319 {
1320 	struct netlink_callback *cb;
1321 	struct sock *sk;
1322 	struct netlink_sock *nlk;
1323 
1324 	cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1325 	if (cb == NULL)
1326 		return -ENOBUFS;
1327 
1328 	memset(cb, 0, sizeof(*cb));
1329 	cb->dump = dump;
1330 	cb->done = done;
1331 	cb->nlh = nlh;
1332 	atomic_inc(&skb->users);
1333 	cb->skb = skb;
1334 
1335 	sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1336 	if (sk == NULL) {
1337 		netlink_destroy_callback(cb);
1338 		return -ECONNREFUSED;
1339 	}
1340 	nlk = nlk_sk(sk);
1341 	/* A dump is in progress... */
1342 	spin_lock(&nlk->cb_lock);
1343 	if (nlk->cb) {
1344 		spin_unlock(&nlk->cb_lock);
1345 		netlink_destroy_callback(cb);
1346 		sock_put(sk);
1347 		return -EBUSY;
1348 	}
1349 	nlk->cb = cb;
1350 	spin_unlock(&nlk->cb_lock);
1351 
1352 	netlink_dump(sk);
1353 	sock_put(sk);
1354 	return 0;
1355 }
1356 
1357 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1358 {
1359 	struct sk_buff *skb;
1360 	struct nlmsghdr *rep;
1361 	struct nlmsgerr *errmsg;
1362 	int size;
1363 
1364 	if (err == 0)
1365 		size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1366 	else
1367 		size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1368 
1369 	skb = alloc_skb(size, GFP_KERNEL);
1370 	if (!skb) {
1371 		struct sock *sk;
1372 
1373 		sk = netlink_lookup(in_skb->sk->sk_protocol,
1374 				    NETLINK_CB(in_skb).pid);
1375 		if (sk) {
1376 			sk->sk_err = ENOBUFS;
1377 			sk->sk_error_report(sk);
1378 			sock_put(sk);
1379 		}
1380 		return;
1381 	}
1382 
1383 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1384 			  NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1385 	errmsg = NLMSG_DATA(rep);
1386 	errmsg->error = err;
1387 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1388 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1389 }
1390 
1391 
1392 #ifdef CONFIG_PROC_FS
1393 struct nl_seq_iter {
1394 	int link;
1395 	int hash_idx;
1396 };
1397 
1398 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1399 {
1400 	struct nl_seq_iter *iter = seq->private;
1401 	int i, j;
1402 	struct sock *s;
1403 	struct hlist_node *node;
1404 	loff_t off = 0;
1405 
1406 	for (i=0; i<MAX_LINKS; i++) {
1407 		struct nl_pid_hash *hash = &nl_table[i].hash;
1408 
1409 		for (j = 0; j <= hash->mask; j++) {
1410 			sk_for_each(s, node, &hash->table[j]) {
1411 				if (off == pos) {
1412 					iter->link = i;
1413 					iter->hash_idx = j;
1414 					return s;
1415 				}
1416 				++off;
1417 			}
1418 		}
1419 	}
1420 	return NULL;
1421 }
1422 
1423 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1424 {
1425 	read_lock(&nl_table_lock);
1426 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1427 }
1428 
1429 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1430 {
1431 	struct sock *s;
1432 	struct nl_seq_iter *iter;
1433 	int i, j;
1434 
1435 	++*pos;
1436 
1437 	if (v == SEQ_START_TOKEN)
1438 		return netlink_seq_socket_idx(seq, 0);
1439 
1440 	s = sk_next(v);
1441 	if (s)
1442 		return s;
1443 
1444 	iter = seq->private;
1445 	i = iter->link;
1446 	j = iter->hash_idx + 1;
1447 
1448 	do {
1449 		struct nl_pid_hash *hash = &nl_table[i].hash;
1450 
1451 		for (; j <= hash->mask; j++) {
1452 			s = sk_head(&hash->table[j]);
1453 			if (s) {
1454 				iter->link = i;
1455 				iter->hash_idx = j;
1456 				return s;
1457 			}
1458 		}
1459 
1460 		j = 0;
1461 	} while (++i < MAX_LINKS);
1462 
1463 	return NULL;
1464 }
1465 
1466 static void netlink_seq_stop(struct seq_file *seq, void *v)
1467 {
1468 	read_unlock(&nl_table_lock);
1469 }
1470 
1471 
1472 static int netlink_seq_show(struct seq_file *seq, void *v)
1473 {
1474 	if (v == SEQ_START_TOKEN)
1475 		seq_puts(seq,
1476 			 "sk       Eth Pid    Groups   "
1477 			 "Rmem     Wmem     Dump     Locks\n");
1478 	else {
1479 		struct sock *s = v;
1480 		struct netlink_sock *nlk = nlk_sk(s);
1481 
1482 		seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1483 			   s,
1484 			   s->sk_protocol,
1485 			   nlk->pid,
1486 			   nlk->flags & NETLINK_KERNEL_SOCKET ?
1487 				0 : (unsigned int)nlk->groups[0],
1488 			   atomic_read(&s->sk_rmem_alloc),
1489 			   atomic_read(&s->sk_wmem_alloc),
1490 			   nlk->cb,
1491 			   atomic_read(&s->sk_refcnt)
1492 			);
1493 
1494 	}
1495 	return 0;
1496 }
1497 
1498 static struct seq_operations netlink_seq_ops = {
1499 	.start  = netlink_seq_start,
1500 	.next   = netlink_seq_next,
1501 	.stop   = netlink_seq_stop,
1502 	.show   = netlink_seq_show,
1503 };
1504 
1505 
1506 static int netlink_seq_open(struct inode *inode, struct file *file)
1507 {
1508 	struct seq_file *seq;
1509 	struct nl_seq_iter *iter;
1510 	int err;
1511 
1512 	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1513 	if (!iter)
1514 		return -ENOMEM;
1515 
1516 	err = seq_open(file, &netlink_seq_ops);
1517 	if (err) {
1518 		kfree(iter);
1519 		return err;
1520 	}
1521 
1522 	memset(iter, 0, sizeof(*iter));
1523 	seq = file->private_data;
1524 	seq->private = iter;
1525 	return 0;
1526 }
1527 
1528 static struct file_operations netlink_seq_fops = {
1529 	.owner		= THIS_MODULE,
1530 	.open		= netlink_seq_open,
1531 	.read		= seq_read,
1532 	.llseek		= seq_lseek,
1533 	.release	= seq_release_private,
1534 };
1535 
1536 #endif
1537 
1538 int netlink_register_notifier(struct notifier_block *nb)
1539 {
1540 	return notifier_chain_register(&netlink_chain, nb);
1541 }
1542 
1543 int netlink_unregister_notifier(struct notifier_block *nb)
1544 {
1545 	return notifier_chain_unregister(&netlink_chain, nb);
1546 }
1547 
1548 static struct proto_ops netlink_ops = {
1549 	.family =	PF_NETLINK,
1550 	.owner =	THIS_MODULE,
1551 	.release =	netlink_release,
1552 	.bind =		netlink_bind,
1553 	.connect =	netlink_connect,
1554 	.socketpair =	sock_no_socketpair,
1555 	.accept =	sock_no_accept,
1556 	.getname =	netlink_getname,
1557 	.poll =		datagram_poll,
1558 	.ioctl =	sock_no_ioctl,
1559 	.listen =	sock_no_listen,
1560 	.shutdown =	sock_no_shutdown,
1561 	.setsockopt =	netlink_setsockopt,
1562 	.getsockopt =	netlink_getsockopt,
1563 	.sendmsg =	netlink_sendmsg,
1564 	.recvmsg =	netlink_recvmsg,
1565 	.mmap =		sock_no_mmap,
1566 	.sendpage =	sock_no_sendpage,
1567 };
1568 
1569 static struct net_proto_family netlink_family_ops = {
1570 	.family = PF_NETLINK,
1571 	.create = netlink_create,
1572 	.owner	= THIS_MODULE,	/* for consistency 8) */
1573 };
1574 
1575 extern void netlink_skb_parms_too_large(void);
1576 
1577 static int __init netlink_proto_init(void)
1578 {
1579 	struct sk_buff *dummy_skb;
1580 	int i;
1581 	unsigned long max;
1582 	unsigned int order;
1583 	int err = proto_register(&netlink_proto, 0);
1584 
1585 	if (err != 0)
1586 		goto out;
1587 
1588 	if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1589 		netlink_skb_parms_too_large();
1590 
1591 	nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1592 	if (!nl_table) {
1593 enomem:
1594 		printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1595 		return -ENOMEM;
1596 	}
1597 
1598 	memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1599 
1600 	if (num_physpages >= (128 * 1024))
1601 		max = num_physpages >> (21 - PAGE_SHIFT);
1602 	else
1603 		max = num_physpages >> (23 - PAGE_SHIFT);
1604 
1605 	order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1606 	max = (1UL << order) / sizeof(struct hlist_head);
1607 	order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1608 
1609 	for (i = 0; i < MAX_LINKS; i++) {
1610 		struct nl_pid_hash *hash = &nl_table[i].hash;
1611 
1612 		hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1613 		if (!hash->table) {
1614 			while (i-- > 0)
1615 				nl_pid_hash_free(nl_table[i].hash.table,
1616 						 1 * sizeof(*hash->table));
1617 			kfree(nl_table);
1618 			goto enomem;
1619 		}
1620 		memset(hash->table, 0, 1 * sizeof(*hash->table));
1621 		hash->max_shift = order;
1622 		hash->shift = 0;
1623 		hash->mask = 0;
1624 		hash->rehash_time = jiffies;
1625 	}
1626 
1627 	sock_register(&netlink_family_ops);
1628 #ifdef CONFIG_PROC_FS
1629 	proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1630 #endif
1631 	/* The netlink device handler may be needed early. */
1632 	rtnetlink_init();
1633 out:
1634 	return err;
1635 }
1636 
1637 core_initcall(netlink_proto_init);
1638 
1639 EXPORT_SYMBOL(netlink_ack);
1640 EXPORT_SYMBOL(netlink_broadcast);
1641 EXPORT_SYMBOL(netlink_dump_start);
1642 EXPORT_SYMBOL(netlink_kernel_create);
1643 EXPORT_SYMBOL(netlink_register_notifier);
1644 EXPORT_SYMBOL(netlink_set_err);
1645 EXPORT_SYMBOL(netlink_set_nonroot);
1646 EXPORT_SYMBOL(netlink_unicast);
1647 EXPORT_SYMBOL(netlink_unregister_notifier);
1648 
1649