xref: /linux/net/netlink/af_netlink.c (revision b0d5c81e872ed21de1e56feb0fa6e4161da7be61)
1 /*
2  * NETLINK      Kernel-user communication protocol.
3  *
4  * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
5  * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6  * 				Patrick McHardy <kaber@trash.net>
7  *
8  *		This program is free software; you can redistribute it and/or
9  *		modify it under the terms of the GNU General Public License
10  *		as published by the Free Software Foundation; either version
11  *		2 of the License, or (at your option) any later version.
12  *
13  * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14  *                               added netlink_proto_exit
15  * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16  * 				 use nlk_sk, as sk->protinfo is on a diet 8)
17  * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18  * 				 - inc module use count of module that owns
19  * 				   the kernel socket in case userspace opens
20  * 				   socket of same protocol
21  * 				 - remove all module support, since netlink is
22  * 				   mandatory if CONFIG_NET=y these days
23  */
24 
25 #include <linux/module.h>
26 
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
36 #include <linux/un.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
41 #include <linux/fs.h>
42 #include <linux/slab.h>
43 #include <linux/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.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/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65 #include <linux/net_namespace.h>
66 
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/sock.h>
70 #include <net/scm.h>
71 #include <net/netlink.h>
72 
73 #include "af_netlink.h"
74 
75 struct listeners {
76 	struct rcu_head		rcu;
77 	unsigned long		masks[0];
78 };
79 
80 /* state bits */
81 #define NETLINK_S_CONGESTED		0x0
82 
83 static inline int netlink_is_kernel(struct sock *sk)
84 {
85 	return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
86 }
87 
88 struct netlink_table *nl_table __read_mostly;
89 EXPORT_SYMBOL_GPL(nl_table);
90 
91 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
92 
93 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
94 
95 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
96 	"nlk_cb_mutex-ROUTE",
97 	"nlk_cb_mutex-1",
98 	"nlk_cb_mutex-USERSOCK",
99 	"nlk_cb_mutex-FIREWALL",
100 	"nlk_cb_mutex-SOCK_DIAG",
101 	"nlk_cb_mutex-NFLOG",
102 	"nlk_cb_mutex-XFRM",
103 	"nlk_cb_mutex-SELINUX",
104 	"nlk_cb_mutex-ISCSI",
105 	"nlk_cb_mutex-AUDIT",
106 	"nlk_cb_mutex-FIB_LOOKUP",
107 	"nlk_cb_mutex-CONNECTOR",
108 	"nlk_cb_mutex-NETFILTER",
109 	"nlk_cb_mutex-IP6_FW",
110 	"nlk_cb_mutex-DNRTMSG",
111 	"nlk_cb_mutex-KOBJECT_UEVENT",
112 	"nlk_cb_mutex-GENERIC",
113 	"nlk_cb_mutex-17",
114 	"nlk_cb_mutex-SCSITRANSPORT",
115 	"nlk_cb_mutex-ECRYPTFS",
116 	"nlk_cb_mutex-RDMA",
117 	"nlk_cb_mutex-CRYPTO",
118 	"nlk_cb_mutex-SMC",
119 	"nlk_cb_mutex-23",
120 	"nlk_cb_mutex-24",
121 	"nlk_cb_mutex-25",
122 	"nlk_cb_mutex-26",
123 	"nlk_cb_mutex-27",
124 	"nlk_cb_mutex-28",
125 	"nlk_cb_mutex-29",
126 	"nlk_cb_mutex-30",
127 	"nlk_cb_mutex-31",
128 	"nlk_cb_mutex-MAX_LINKS"
129 };
130 
131 static int netlink_dump(struct sock *sk);
132 
133 /* nl_table locking explained:
134  * Lookup and traversal are protected with an RCU read-side lock. Insertion
135  * and removal are protected with per bucket lock while using RCU list
136  * modification primitives and may run in parallel to RCU protected lookups.
137  * Destruction of the Netlink socket may only occur *after* nl_table_lock has
138  * been acquired * either during or after the socket has been removed from
139  * the list and after an RCU grace period.
140  */
141 DEFINE_RWLOCK(nl_table_lock);
142 EXPORT_SYMBOL_GPL(nl_table_lock);
143 static atomic_t nl_table_users = ATOMIC_INIT(0);
144 
145 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
146 
147 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
148 
149 
150 static const struct rhashtable_params netlink_rhashtable_params;
151 
152 static inline u32 netlink_group_mask(u32 group)
153 {
154 	return group ? 1 << (group - 1) : 0;
155 }
156 
157 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
158 					   gfp_t gfp_mask)
159 {
160 	unsigned int len = skb_end_offset(skb);
161 	struct sk_buff *new;
162 
163 	new = alloc_skb(len, gfp_mask);
164 	if (new == NULL)
165 		return NULL;
166 
167 	NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
168 	NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
169 	NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
170 
171 	skb_put_data(new, skb->data, len);
172 	return new;
173 }
174 
175 static unsigned int netlink_tap_net_id;
176 
177 struct netlink_tap_net {
178 	struct list_head netlink_tap_all;
179 	struct mutex netlink_tap_lock;
180 };
181 
182 int netlink_add_tap(struct netlink_tap *nt)
183 {
184 	struct net *net = dev_net(nt->dev);
185 	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
186 
187 	if (unlikely(nt->dev->type != ARPHRD_NETLINK))
188 		return -EINVAL;
189 
190 	mutex_lock(&nn->netlink_tap_lock);
191 	list_add_rcu(&nt->list, &nn->netlink_tap_all);
192 	mutex_unlock(&nn->netlink_tap_lock);
193 
194 	__module_get(nt->module);
195 
196 	return 0;
197 }
198 EXPORT_SYMBOL_GPL(netlink_add_tap);
199 
200 static int __netlink_remove_tap(struct netlink_tap *nt)
201 {
202 	struct net *net = dev_net(nt->dev);
203 	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
204 	bool found = false;
205 	struct netlink_tap *tmp;
206 
207 	mutex_lock(&nn->netlink_tap_lock);
208 
209 	list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
210 		if (nt == tmp) {
211 			list_del_rcu(&nt->list);
212 			found = true;
213 			goto out;
214 		}
215 	}
216 
217 	pr_warn("__netlink_remove_tap: %p not found\n", nt);
218 out:
219 	mutex_unlock(&nn->netlink_tap_lock);
220 
221 	if (found)
222 		module_put(nt->module);
223 
224 	return found ? 0 : -ENODEV;
225 }
226 
227 int netlink_remove_tap(struct netlink_tap *nt)
228 {
229 	int ret;
230 
231 	ret = __netlink_remove_tap(nt);
232 	synchronize_net();
233 
234 	return ret;
235 }
236 EXPORT_SYMBOL_GPL(netlink_remove_tap);
237 
238 static __net_init int netlink_tap_init_net(struct net *net)
239 {
240 	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
241 
242 	INIT_LIST_HEAD(&nn->netlink_tap_all);
243 	mutex_init(&nn->netlink_tap_lock);
244 	return 0;
245 }
246 
247 static void __net_exit netlink_tap_exit_net(struct net *net)
248 {
249 }
250 
251 static struct pernet_operations netlink_tap_net_ops = {
252 	.init = netlink_tap_init_net,
253 	.exit = netlink_tap_exit_net,
254 	.id   = &netlink_tap_net_id,
255 	.size = sizeof(struct netlink_tap_net),
256 };
257 
258 static bool netlink_filter_tap(const struct sk_buff *skb)
259 {
260 	struct sock *sk = skb->sk;
261 
262 	/* We take the more conservative approach and
263 	 * whitelist socket protocols that may pass.
264 	 */
265 	switch (sk->sk_protocol) {
266 	case NETLINK_ROUTE:
267 	case NETLINK_USERSOCK:
268 	case NETLINK_SOCK_DIAG:
269 	case NETLINK_NFLOG:
270 	case NETLINK_XFRM:
271 	case NETLINK_FIB_LOOKUP:
272 	case NETLINK_NETFILTER:
273 	case NETLINK_GENERIC:
274 		return true;
275 	}
276 
277 	return false;
278 }
279 
280 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
281 				     struct net_device *dev)
282 {
283 	struct sk_buff *nskb;
284 	struct sock *sk = skb->sk;
285 	int ret = -ENOMEM;
286 
287 	if (!net_eq(dev_net(dev), sock_net(sk)))
288 		return 0;
289 
290 	dev_hold(dev);
291 
292 	if (is_vmalloc_addr(skb->head))
293 		nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
294 	else
295 		nskb = skb_clone(skb, GFP_ATOMIC);
296 	if (nskb) {
297 		nskb->dev = dev;
298 		nskb->protocol = htons((u16) sk->sk_protocol);
299 		nskb->pkt_type = netlink_is_kernel(sk) ?
300 				 PACKET_KERNEL : PACKET_USER;
301 		skb_reset_network_header(nskb);
302 		ret = dev_queue_xmit(nskb);
303 		if (unlikely(ret > 0))
304 			ret = net_xmit_errno(ret);
305 	}
306 
307 	dev_put(dev);
308 	return ret;
309 }
310 
311 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
312 {
313 	int ret;
314 	struct netlink_tap *tmp;
315 
316 	if (!netlink_filter_tap(skb))
317 		return;
318 
319 	list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
320 		ret = __netlink_deliver_tap_skb(skb, tmp->dev);
321 		if (unlikely(ret))
322 			break;
323 	}
324 }
325 
326 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
327 {
328 	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
329 
330 	rcu_read_lock();
331 
332 	if (unlikely(!list_empty(&nn->netlink_tap_all)))
333 		__netlink_deliver_tap(skb, nn);
334 
335 	rcu_read_unlock();
336 }
337 
338 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
339 				       struct sk_buff *skb)
340 {
341 	if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
342 		netlink_deliver_tap(sock_net(dst), skb);
343 }
344 
345 static void netlink_overrun(struct sock *sk)
346 {
347 	struct netlink_sock *nlk = nlk_sk(sk);
348 
349 	if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
350 		if (!test_and_set_bit(NETLINK_S_CONGESTED,
351 				      &nlk_sk(sk)->state)) {
352 			sk->sk_err = ENOBUFS;
353 			sk->sk_error_report(sk);
354 		}
355 	}
356 	atomic_inc(&sk->sk_drops);
357 }
358 
359 static void netlink_rcv_wake(struct sock *sk)
360 {
361 	struct netlink_sock *nlk = nlk_sk(sk);
362 
363 	if (skb_queue_empty(&sk->sk_receive_queue))
364 		clear_bit(NETLINK_S_CONGESTED, &nlk->state);
365 	if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
366 		wake_up_interruptible(&nlk->wait);
367 }
368 
369 static void netlink_skb_destructor(struct sk_buff *skb)
370 {
371 	if (is_vmalloc_addr(skb->head)) {
372 		if (!skb->cloned ||
373 		    !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
374 			vfree(skb->head);
375 
376 		skb->head = NULL;
377 	}
378 	if (skb->sk != NULL)
379 		sock_rfree(skb);
380 }
381 
382 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
383 {
384 	WARN_ON(skb->sk != NULL);
385 	skb->sk = sk;
386 	skb->destructor = netlink_skb_destructor;
387 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
388 	sk_mem_charge(sk, skb->truesize);
389 }
390 
391 static void netlink_sock_destruct(struct sock *sk)
392 {
393 	struct netlink_sock *nlk = nlk_sk(sk);
394 
395 	if (nlk->cb_running) {
396 		if (nlk->cb.done)
397 			nlk->cb.done(&nlk->cb);
398 		module_put(nlk->cb.module);
399 		kfree_skb(nlk->cb.skb);
400 	}
401 
402 	skb_queue_purge(&sk->sk_receive_queue);
403 
404 	if (!sock_flag(sk, SOCK_DEAD)) {
405 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
406 		return;
407 	}
408 
409 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
410 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
411 	WARN_ON(nlk_sk(sk)->groups);
412 }
413 
414 static void netlink_sock_destruct_work(struct work_struct *work)
415 {
416 	struct netlink_sock *nlk = container_of(work, struct netlink_sock,
417 						work);
418 
419 	sk_free(&nlk->sk);
420 }
421 
422 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
423  * SMP. Look, when several writers sleep and reader wakes them up, all but one
424  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
425  * this, _but_ remember, it adds useless work on UP machines.
426  */
427 
428 void netlink_table_grab(void)
429 	__acquires(nl_table_lock)
430 {
431 	might_sleep();
432 
433 	write_lock_irq(&nl_table_lock);
434 
435 	if (atomic_read(&nl_table_users)) {
436 		DECLARE_WAITQUEUE(wait, current);
437 
438 		add_wait_queue_exclusive(&nl_table_wait, &wait);
439 		for (;;) {
440 			set_current_state(TASK_UNINTERRUPTIBLE);
441 			if (atomic_read(&nl_table_users) == 0)
442 				break;
443 			write_unlock_irq(&nl_table_lock);
444 			schedule();
445 			write_lock_irq(&nl_table_lock);
446 		}
447 
448 		__set_current_state(TASK_RUNNING);
449 		remove_wait_queue(&nl_table_wait, &wait);
450 	}
451 }
452 
453 void netlink_table_ungrab(void)
454 	__releases(nl_table_lock)
455 {
456 	write_unlock_irq(&nl_table_lock);
457 	wake_up(&nl_table_wait);
458 }
459 
460 static inline void
461 netlink_lock_table(void)
462 {
463 	/* read_lock() synchronizes us to netlink_table_grab */
464 
465 	read_lock(&nl_table_lock);
466 	atomic_inc(&nl_table_users);
467 	read_unlock(&nl_table_lock);
468 }
469 
470 static inline void
471 netlink_unlock_table(void)
472 {
473 	if (atomic_dec_and_test(&nl_table_users))
474 		wake_up(&nl_table_wait);
475 }
476 
477 struct netlink_compare_arg
478 {
479 	possible_net_t pnet;
480 	u32 portid;
481 };
482 
483 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
484 #define netlink_compare_arg_len \
485 	(offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
486 
487 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
488 				  const void *ptr)
489 {
490 	const struct netlink_compare_arg *x = arg->key;
491 	const struct netlink_sock *nlk = ptr;
492 
493 	return nlk->portid != x->portid ||
494 	       !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
495 }
496 
497 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
498 				     struct net *net, u32 portid)
499 {
500 	memset(arg, 0, sizeof(*arg));
501 	write_pnet(&arg->pnet, net);
502 	arg->portid = portid;
503 }
504 
505 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
506 				     struct net *net)
507 {
508 	struct netlink_compare_arg arg;
509 
510 	netlink_compare_arg_init(&arg, net, portid);
511 	return rhashtable_lookup_fast(&table->hash, &arg,
512 				      netlink_rhashtable_params);
513 }
514 
515 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
516 {
517 	struct netlink_compare_arg arg;
518 
519 	netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
520 	return rhashtable_lookup_insert_key(&table->hash, &arg,
521 					    &nlk_sk(sk)->node,
522 					    netlink_rhashtable_params);
523 }
524 
525 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
526 {
527 	struct netlink_table *table = &nl_table[protocol];
528 	struct sock *sk;
529 
530 	rcu_read_lock();
531 	sk = __netlink_lookup(table, portid, net);
532 	if (sk)
533 		sock_hold(sk);
534 	rcu_read_unlock();
535 
536 	return sk;
537 }
538 
539 static const struct proto_ops netlink_ops;
540 
541 static void
542 netlink_update_listeners(struct sock *sk)
543 {
544 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
545 	unsigned long mask;
546 	unsigned int i;
547 	struct listeners *listeners;
548 
549 	listeners = nl_deref_protected(tbl->listeners);
550 	if (!listeners)
551 		return;
552 
553 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
554 		mask = 0;
555 		sk_for_each_bound(sk, &tbl->mc_list) {
556 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
557 				mask |= nlk_sk(sk)->groups[i];
558 		}
559 		listeners->masks[i] = mask;
560 	}
561 	/* this function is only called with the netlink table "grabbed", which
562 	 * makes sure updates are visible before bind or setsockopt return. */
563 }
564 
565 static int netlink_insert(struct sock *sk, u32 portid)
566 {
567 	struct netlink_table *table = &nl_table[sk->sk_protocol];
568 	int err;
569 
570 	lock_sock(sk);
571 
572 	err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
573 	if (nlk_sk(sk)->bound)
574 		goto err;
575 
576 	err = -ENOMEM;
577 	if (BITS_PER_LONG > 32 &&
578 	    unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
579 		goto err;
580 
581 	nlk_sk(sk)->portid = portid;
582 	sock_hold(sk);
583 
584 	err = __netlink_insert(table, sk);
585 	if (err) {
586 		/* In case the hashtable backend returns with -EBUSY
587 		 * from here, it must not escape to the caller.
588 		 */
589 		if (unlikely(err == -EBUSY))
590 			err = -EOVERFLOW;
591 		if (err == -EEXIST)
592 			err = -EADDRINUSE;
593 		sock_put(sk);
594 		goto err;
595 	}
596 
597 	/* We need to ensure that the socket is hashed and visible. */
598 	smp_wmb();
599 	nlk_sk(sk)->bound = portid;
600 
601 err:
602 	release_sock(sk);
603 	return err;
604 }
605 
606 static void netlink_remove(struct sock *sk)
607 {
608 	struct netlink_table *table;
609 
610 	table = &nl_table[sk->sk_protocol];
611 	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
612 				    netlink_rhashtable_params)) {
613 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
614 		__sock_put(sk);
615 	}
616 
617 	netlink_table_grab();
618 	if (nlk_sk(sk)->subscriptions) {
619 		__sk_del_bind_node(sk);
620 		netlink_update_listeners(sk);
621 	}
622 	if (sk->sk_protocol == NETLINK_GENERIC)
623 		atomic_inc(&genl_sk_destructing_cnt);
624 	netlink_table_ungrab();
625 }
626 
627 static struct proto netlink_proto = {
628 	.name	  = "NETLINK",
629 	.owner	  = THIS_MODULE,
630 	.obj_size = sizeof(struct netlink_sock),
631 };
632 
633 static int __netlink_create(struct net *net, struct socket *sock,
634 			    struct mutex *cb_mutex, int protocol,
635 			    int kern)
636 {
637 	struct sock *sk;
638 	struct netlink_sock *nlk;
639 
640 	sock->ops = &netlink_ops;
641 
642 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
643 	if (!sk)
644 		return -ENOMEM;
645 
646 	sock_init_data(sock, sk);
647 
648 	nlk = nlk_sk(sk);
649 	if (cb_mutex) {
650 		nlk->cb_mutex = cb_mutex;
651 	} else {
652 		nlk->cb_mutex = &nlk->cb_def_mutex;
653 		mutex_init(nlk->cb_mutex);
654 		lockdep_set_class_and_name(nlk->cb_mutex,
655 					   nlk_cb_mutex_keys + protocol,
656 					   nlk_cb_mutex_key_strings[protocol]);
657 	}
658 	init_waitqueue_head(&nlk->wait);
659 
660 	sk->sk_destruct = netlink_sock_destruct;
661 	sk->sk_protocol = protocol;
662 	return 0;
663 }
664 
665 static int netlink_create(struct net *net, struct socket *sock, int protocol,
666 			  int kern)
667 {
668 	struct module *module = NULL;
669 	struct mutex *cb_mutex;
670 	struct netlink_sock *nlk;
671 	int (*bind)(struct net *net, int group);
672 	void (*unbind)(struct net *net, int group);
673 	int err = 0;
674 
675 	sock->state = SS_UNCONNECTED;
676 
677 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
678 		return -ESOCKTNOSUPPORT;
679 
680 	if (protocol < 0 || protocol >= MAX_LINKS)
681 		return -EPROTONOSUPPORT;
682 
683 	netlink_lock_table();
684 #ifdef CONFIG_MODULES
685 	if (!nl_table[protocol].registered) {
686 		netlink_unlock_table();
687 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
688 		netlink_lock_table();
689 	}
690 #endif
691 	if (nl_table[protocol].registered &&
692 	    try_module_get(nl_table[protocol].module))
693 		module = nl_table[protocol].module;
694 	else
695 		err = -EPROTONOSUPPORT;
696 	cb_mutex = nl_table[protocol].cb_mutex;
697 	bind = nl_table[protocol].bind;
698 	unbind = nl_table[protocol].unbind;
699 	netlink_unlock_table();
700 
701 	if (err < 0)
702 		goto out;
703 
704 	err = __netlink_create(net, sock, cb_mutex, protocol, kern);
705 	if (err < 0)
706 		goto out_module;
707 
708 	local_bh_disable();
709 	sock_prot_inuse_add(net, &netlink_proto, 1);
710 	local_bh_enable();
711 
712 	nlk = nlk_sk(sock->sk);
713 	nlk->module = module;
714 	nlk->netlink_bind = bind;
715 	nlk->netlink_unbind = unbind;
716 out:
717 	return err;
718 
719 out_module:
720 	module_put(module);
721 	goto out;
722 }
723 
724 static void deferred_put_nlk_sk(struct rcu_head *head)
725 {
726 	struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
727 	struct sock *sk = &nlk->sk;
728 
729 	kfree(nlk->groups);
730 	nlk->groups = NULL;
731 
732 	if (!refcount_dec_and_test(&sk->sk_refcnt))
733 		return;
734 
735 	if (nlk->cb_running && nlk->cb.done) {
736 		INIT_WORK(&nlk->work, netlink_sock_destruct_work);
737 		schedule_work(&nlk->work);
738 		return;
739 	}
740 
741 	sk_free(sk);
742 }
743 
744 static int netlink_release(struct socket *sock)
745 {
746 	struct sock *sk = sock->sk;
747 	struct netlink_sock *nlk;
748 
749 	if (!sk)
750 		return 0;
751 
752 	netlink_remove(sk);
753 	sock_orphan(sk);
754 	nlk = nlk_sk(sk);
755 
756 	/*
757 	 * OK. Socket is unlinked, any packets that arrive now
758 	 * will be purged.
759 	 */
760 
761 	/* must not acquire netlink_table_lock in any way again before unbind
762 	 * and notifying genetlink is done as otherwise it might deadlock
763 	 */
764 	if (nlk->netlink_unbind) {
765 		int i;
766 
767 		for (i = 0; i < nlk->ngroups; i++)
768 			if (test_bit(i, nlk->groups))
769 				nlk->netlink_unbind(sock_net(sk), i + 1);
770 	}
771 	if (sk->sk_protocol == NETLINK_GENERIC &&
772 	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
773 		wake_up(&genl_sk_destructing_waitq);
774 
775 	sock->sk = NULL;
776 	wake_up_interruptible_all(&nlk->wait);
777 
778 	skb_queue_purge(&sk->sk_write_queue);
779 
780 	if (nlk->portid && nlk->bound) {
781 		struct netlink_notify n = {
782 						.net = sock_net(sk),
783 						.protocol = sk->sk_protocol,
784 						.portid = nlk->portid,
785 					  };
786 		blocking_notifier_call_chain(&netlink_chain,
787 				NETLINK_URELEASE, &n);
788 	}
789 
790 	module_put(nlk->module);
791 
792 	if (netlink_is_kernel(sk)) {
793 		netlink_table_grab();
794 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
795 		if (--nl_table[sk->sk_protocol].registered == 0) {
796 			struct listeners *old;
797 
798 			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
799 			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
800 			kfree_rcu(old, rcu);
801 			nl_table[sk->sk_protocol].module = NULL;
802 			nl_table[sk->sk_protocol].bind = NULL;
803 			nl_table[sk->sk_protocol].unbind = NULL;
804 			nl_table[sk->sk_protocol].flags = 0;
805 			nl_table[sk->sk_protocol].registered = 0;
806 		}
807 		netlink_table_ungrab();
808 	}
809 
810 	local_bh_disable();
811 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
812 	local_bh_enable();
813 	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
814 	return 0;
815 }
816 
817 static int netlink_autobind(struct socket *sock)
818 {
819 	struct sock *sk = sock->sk;
820 	struct net *net = sock_net(sk);
821 	struct netlink_table *table = &nl_table[sk->sk_protocol];
822 	s32 portid = task_tgid_vnr(current);
823 	int err;
824 	s32 rover = -4096;
825 	bool ok;
826 
827 retry:
828 	cond_resched();
829 	rcu_read_lock();
830 	ok = !__netlink_lookup(table, portid, net);
831 	rcu_read_unlock();
832 	if (!ok) {
833 		/* Bind collision, search negative portid values. */
834 		if (rover == -4096)
835 			/* rover will be in range [S32_MIN, -4097] */
836 			rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
837 		else if (rover >= -4096)
838 			rover = -4097;
839 		portid = rover--;
840 		goto retry;
841 	}
842 
843 	err = netlink_insert(sk, portid);
844 	if (err == -EADDRINUSE)
845 		goto retry;
846 
847 	/* If 2 threads race to autobind, that is fine.  */
848 	if (err == -EBUSY)
849 		err = 0;
850 
851 	return err;
852 }
853 
854 /**
855  * __netlink_ns_capable - General netlink message capability test
856  * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
857  * @user_ns: The user namespace of the capability to use
858  * @cap: The capability to use
859  *
860  * Test to see if the opener of the socket we received the message
861  * from had when the netlink socket was created and the sender of the
862  * message has has the capability @cap in the user namespace @user_ns.
863  */
864 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
865 			struct user_namespace *user_ns, int cap)
866 {
867 	return ((nsp->flags & NETLINK_SKB_DST) ||
868 		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
869 		ns_capable(user_ns, cap);
870 }
871 EXPORT_SYMBOL(__netlink_ns_capable);
872 
873 /**
874  * netlink_ns_capable - General netlink message capability test
875  * @skb: socket buffer holding a netlink command from userspace
876  * @user_ns: The user namespace of the capability to use
877  * @cap: The capability to use
878  *
879  * Test to see if the opener of the socket we received the message
880  * from had when the netlink socket was created and the sender of the
881  * message has has the capability @cap in the user namespace @user_ns.
882  */
883 bool netlink_ns_capable(const struct sk_buff *skb,
884 			struct user_namespace *user_ns, int cap)
885 {
886 	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
887 }
888 EXPORT_SYMBOL(netlink_ns_capable);
889 
890 /**
891  * netlink_capable - Netlink global message capability test
892  * @skb: socket buffer holding a netlink command from userspace
893  * @cap: The capability to use
894  *
895  * Test to see if the opener of the socket we received the message
896  * from had when the netlink socket was created and the sender of the
897  * message has has the capability @cap in all user namespaces.
898  */
899 bool netlink_capable(const struct sk_buff *skb, int cap)
900 {
901 	return netlink_ns_capable(skb, &init_user_ns, cap);
902 }
903 EXPORT_SYMBOL(netlink_capable);
904 
905 /**
906  * netlink_net_capable - Netlink network namespace message capability test
907  * @skb: socket buffer holding a netlink command from userspace
908  * @cap: The capability to use
909  *
910  * Test to see if the opener of the socket we received the message
911  * from had when the netlink socket was created and the sender of the
912  * message has has the capability @cap over the network namespace of
913  * the socket we received the message from.
914  */
915 bool netlink_net_capable(const struct sk_buff *skb, int cap)
916 {
917 	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
918 }
919 EXPORT_SYMBOL(netlink_net_capable);
920 
921 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
922 {
923 	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
924 		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
925 }
926 
927 static void
928 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
929 {
930 	struct netlink_sock *nlk = nlk_sk(sk);
931 
932 	if (nlk->subscriptions && !subscriptions)
933 		__sk_del_bind_node(sk);
934 	else if (!nlk->subscriptions && subscriptions)
935 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
936 	nlk->subscriptions = subscriptions;
937 }
938 
939 static int netlink_realloc_groups(struct sock *sk)
940 {
941 	struct netlink_sock *nlk = nlk_sk(sk);
942 	unsigned int groups;
943 	unsigned long *new_groups;
944 	int err = 0;
945 
946 	netlink_table_grab();
947 
948 	groups = nl_table[sk->sk_protocol].groups;
949 	if (!nl_table[sk->sk_protocol].registered) {
950 		err = -ENOENT;
951 		goto out_unlock;
952 	}
953 
954 	if (nlk->ngroups >= groups)
955 		goto out_unlock;
956 
957 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
958 	if (new_groups == NULL) {
959 		err = -ENOMEM;
960 		goto out_unlock;
961 	}
962 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
963 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
964 
965 	nlk->groups = new_groups;
966 	nlk->ngroups = groups;
967  out_unlock:
968 	netlink_table_ungrab();
969 	return err;
970 }
971 
972 static void netlink_undo_bind(int group, long unsigned int groups,
973 			      struct sock *sk)
974 {
975 	struct netlink_sock *nlk = nlk_sk(sk);
976 	int undo;
977 
978 	if (!nlk->netlink_unbind)
979 		return;
980 
981 	for (undo = 0; undo < group; undo++)
982 		if (test_bit(undo, &groups))
983 			nlk->netlink_unbind(sock_net(sk), undo + 1);
984 }
985 
986 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
987 			int addr_len)
988 {
989 	struct sock *sk = sock->sk;
990 	struct net *net = sock_net(sk);
991 	struct netlink_sock *nlk = nlk_sk(sk);
992 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
993 	int err = 0;
994 	long unsigned int groups = nladdr->nl_groups;
995 	bool bound;
996 
997 	if (addr_len < sizeof(struct sockaddr_nl))
998 		return -EINVAL;
999 
1000 	if (nladdr->nl_family != AF_NETLINK)
1001 		return -EINVAL;
1002 
1003 	/* Only superuser is allowed to listen multicasts */
1004 	if (groups) {
1005 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1006 			return -EPERM;
1007 		err = netlink_realloc_groups(sk);
1008 		if (err)
1009 			return err;
1010 	}
1011 
1012 	bound = nlk->bound;
1013 	if (bound) {
1014 		/* Ensure nlk->portid is up-to-date. */
1015 		smp_rmb();
1016 
1017 		if (nladdr->nl_pid != nlk->portid)
1018 			return -EINVAL;
1019 	}
1020 
1021 	netlink_lock_table();
1022 	if (nlk->netlink_bind && groups) {
1023 		int group;
1024 
1025 		for (group = 0; group < nlk->ngroups; group++) {
1026 			if (!test_bit(group, &groups))
1027 				continue;
1028 			err = nlk->netlink_bind(net, group + 1);
1029 			if (!err)
1030 				continue;
1031 			netlink_undo_bind(group, groups, sk);
1032 			goto unlock;
1033 		}
1034 	}
1035 
1036 	/* No need for barriers here as we return to user-space without
1037 	 * using any of the bound attributes.
1038 	 */
1039 	if (!bound) {
1040 		err = nladdr->nl_pid ?
1041 			netlink_insert(sk, nladdr->nl_pid) :
1042 			netlink_autobind(sock);
1043 		if (err) {
1044 			netlink_undo_bind(nlk->ngroups, groups, sk);
1045 			goto unlock;
1046 		}
1047 	}
1048 
1049 	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1050 		goto unlock;
1051 	netlink_unlock_table();
1052 
1053 	netlink_table_grab();
1054 	netlink_update_subscriptions(sk, nlk->subscriptions +
1055 					 hweight32(groups) -
1056 					 hweight32(nlk->groups[0]));
1057 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1058 	netlink_update_listeners(sk);
1059 	netlink_table_ungrab();
1060 
1061 	return 0;
1062 
1063 unlock:
1064 	netlink_unlock_table();
1065 	return err;
1066 }
1067 
1068 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1069 			   int alen, int flags)
1070 {
1071 	int err = 0;
1072 	struct sock *sk = sock->sk;
1073 	struct netlink_sock *nlk = nlk_sk(sk);
1074 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1075 
1076 	if (alen < sizeof(addr->sa_family))
1077 		return -EINVAL;
1078 
1079 	if (addr->sa_family == AF_UNSPEC) {
1080 		sk->sk_state	= NETLINK_UNCONNECTED;
1081 		nlk->dst_portid	= 0;
1082 		nlk->dst_group  = 0;
1083 		return 0;
1084 	}
1085 	if (addr->sa_family != AF_NETLINK)
1086 		return -EINVAL;
1087 
1088 	if (alen < sizeof(struct sockaddr_nl))
1089 		return -EINVAL;
1090 
1091 	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1092 	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1093 		return -EPERM;
1094 
1095 	/* No need for barriers here as we return to user-space without
1096 	 * using any of the bound attributes.
1097 	 */
1098 	if (!nlk->bound)
1099 		err = netlink_autobind(sock);
1100 
1101 	if (err == 0) {
1102 		sk->sk_state	= NETLINK_CONNECTED;
1103 		nlk->dst_portid = nladdr->nl_pid;
1104 		nlk->dst_group  = ffs(nladdr->nl_groups);
1105 	}
1106 
1107 	return err;
1108 }
1109 
1110 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1111 			   int *addr_len, int peer)
1112 {
1113 	struct sock *sk = sock->sk;
1114 	struct netlink_sock *nlk = nlk_sk(sk);
1115 	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1116 
1117 	nladdr->nl_family = AF_NETLINK;
1118 	nladdr->nl_pad = 0;
1119 	*addr_len = sizeof(*nladdr);
1120 
1121 	if (peer) {
1122 		nladdr->nl_pid = nlk->dst_portid;
1123 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1124 	} else {
1125 		nladdr->nl_pid = nlk->portid;
1126 		netlink_lock_table();
1127 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1128 		netlink_unlock_table();
1129 	}
1130 	return 0;
1131 }
1132 
1133 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1134 			 unsigned long arg)
1135 {
1136 	/* try to hand this ioctl down to the NIC drivers.
1137 	 */
1138 	return -ENOIOCTLCMD;
1139 }
1140 
1141 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1142 {
1143 	struct sock *sock;
1144 	struct netlink_sock *nlk;
1145 
1146 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1147 	if (!sock)
1148 		return ERR_PTR(-ECONNREFUSED);
1149 
1150 	/* Don't bother queuing skb if kernel socket has no input function */
1151 	nlk = nlk_sk(sock);
1152 	if (sock->sk_state == NETLINK_CONNECTED &&
1153 	    nlk->dst_portid != nlk_sk(ssk)->portid) {
1154 		sock_put(sock);
1155 		return ERR_PTR(-ECONNREFUSED);
1156 	}
1157 	return sock;
1158 }
1159 
1160 struct sock *netlink_getsockbyfilp(struct file *filp)
1161 {
1162 	struct inode *inode = file_inode(filp);
1163 	struct sock *sock;
1164 
1165 	if (!S_ISSOCK(inode->i_mode))
1166 		return ERR_PTR(-ENOTSOCK);
1167 
1168 	sock = SOCKET_I(inode)->sk;
1169 	if (sock->sk_family != AF_NETLINK)
1170 		return ERR_PTR(-EINVAL);
1171 
1172 	sock_hold(sock);
1173 	return sock;
1174 }
1175 
1176 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1177 					       int broadcast)
1178 {
1179 	struct sk_buff *skb;
1180 	void *data;
1181 
1182 	if (size <= NLMSG_GOODSIZE || broadcast)
1183 		return alloc_skb(size, GFP_KERNEL);
1184 
1185 	size = SKB_DATA_ALIGN(size) +
1186 	       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1187 
1188 	data = vmalloc(size);
1189 	if (data == NULL)
1190 		return NULL;
1191 
1192 	skb = __build_skb(data, size);
1193 	if (skb == NULL)
1194 		vfree(data);
1195 	else
1196 		skb->destructor = netlink_skb_destructor;
1197 
1198 	return skb;
1199 }
1200 
1201 /*
1202  * Attach a skb to a netlink socket.
1203  * The caller must hold a reference to the destination socket. On error, the
1204  * reference is dropped. The skb is not send to the destination, just all
1205  * all error checks are performed and memory in the queue is reserved.
1206  * Return values:
1207  * < 0: error. skb freed, reference to sock dropped.
1208  * 0: continue
1209  * 1: repeat lookup - reference dropped while waiting for socket memory.
1210  */
1211 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1212 		      long *timeo, struct sock *ssk)
1213 {
1214 	struct netlink_sock *nlk;
1215 
1216 	nlk = nlk_sk(sk);
1217 
1218 	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1219 	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1220 		DECLARE_WAITQUEUE(wait, current);
1221 		if (!*timeo) {
1222 			if (!ssk || netlink_is_kernel(ssk))
1223 				netlink_overrun(sk);
1224 			sock_put(sk);
1225 			kfree_skb(skb);
1226 			return -EAGAIN;
1227 		}
1228 
1229 		__set_current_state(TASK_INTERRUPTIBLE);
1230 		add_wait_queue(&nlk->wait, &wait);
1231 
1232 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1233 		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1234 		    !sock_flag(sk, SOCK_DEAD))
1235 			*timeo = schedule_timeout(*timeo);
1236 
1237 		__set_current_state(TASK_RUNNING);
1238 		remove_wait_queue(&nlk->wait, &wait);
1239 		sock_put(sk);
1240 
1241 		if (signal_pending(current)) {
1242 			kfree_skb(skb);
1243 			return sock_intr_errno(*timeo);
1244 		}
1245 		return 1;
1246 	}
1247 	netlink_skb_set_owner_r(skb, sk);
1248 	return 0;
1249 }
1250 
1251 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1252 {
1253 	int len = skb->len;
1254 
1255 	netlink_deliver_tap(sock_net(sk), skb);
1256 
1257 	skb_queue_tail(&sk->sk_receive_queue, skb);
1258 	sk->sk_data_ready(sk);
1259 	return len;
1260 }
1261 
1262 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1263 {
1264 	int len = __netlink_sendskb(sk, skb);
1265 
1266 	sock_put(sk);
1267 	return len;
1268 }
1269 
1270 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1271 {
1272 	kfree_skb(skb);
1273 	sock_put(sk);
1274 }
1275 
1276 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1277 {
1278 	int delta;
1279 
1280 	WARN_ON(skb->sk != NULL);
1281 	delta = skb->end - skb->tail;
1282 	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1283 		return skb;
1284 
1285 	if (skb_shared(skb)) {
1286 		struct sk_buff *nskb = skb_clone(skb, allocation);
1287 		if (!nskb)
1288 			return skb;
1289 		consume_skb(skb);
1290 		skb = nskb;
1291 	}
1292 
1293 	pskb_expand_head(skb, 0, -delta,
1294 			 (allocation & ~__GFP_DIRECT_RECLAIM) |
1295 			 __GFP_NOWARN | __GFP_NORETRY);
1296 	return skb;
1297 }
1298 
1299 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1300 				  struct sock *ssk)
1301 {
1302 	int ret;
1303 	struct netlink_sock *nlk = nlk_sk(sk);
1304 
1305 	ret = -ECONNREFUSED;
1306 	if (nlk->netlink_rcv != NULL) {
1307 		ret = skb->len;
1308 		netlink_skb_set_owner_r(skb, sk);
1309 		NETLINK_CB(skb).sk = ssk;
1310 		netlink_deliver_tap_kernel(sk, ssk, skb);
1311 		nlk->netlink_rcv(skb);
1312 		consume_skb(skb);
1313 	} else {
1314 		kfree_skb(skb);
1315 	}
1316 	sock_put(sk);
1317 	return ret;
1318 }
1319 
1320 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1321 		    u32 portid, int nonblock)
1322 {
1323 	struct sock *sk;
1324 	int err;
1325 	long timeo;
1326 
1327 	skb = netlink_trim(skb, gfp_any());
1328 
1329 	timeo = sock_sndtimeo(ssk, nonblock);
1330 retry:
1331 	sk = netlink_getsockbyportid(ssk, portid);
1332 	if (IS_ERR(sk)) {
1333 		kfree_skb(skb);
1334 		return PTR_ERR(sk);
1335 	}
1336 	if (netlink_is_kernel(sk))
1337 		return netlink_unicast_kernel(sk, skb, ssk);
1338 
1339 	if (sk_filter(sk, skb)) {
1340 		err = skb->len;
1341 		kfree_skb(skb);
1342 		sock_put(sk);
1343 		return err;
1344 	}
1345 
1346 	err = netlink_attachskb(sk, skb, &timeo, ssk);
1347 	if (err == 1)
1348 		goto retry;
1349 	if (err)
1350 		return err;
1351 
1352 	return netlink_sendskb(sk, skb);
1353 }
1354 EXPORT_SYMBOL(netlink_unicast);
1355 
1356 int netlink_has_listeners(struct sock *sk, unsigned int group)
1357 {
1358 	int res = 0;
1359 	struct listeners *listeners;
1360 
1361 	BUG_ON(!netlink_is_kernel(sk));
1362 
1363 	rcu_read_lock();
1364 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1365 
1366 	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1367 		res = test_bit(group - 1, listeners->masks);
1368 
1369 	rcu_read_unlock();
1370 
1371 	return res;
1372 }
1373 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1374 
1375 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1376 {
1377 	struct netlink_sock *nlk = nlk_sk(sk);
1378 
1379 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1380 	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1381 		netlink_skb_set_owner_r(skb, sk);
1382 		__netlink_sendskb(sk, skb);
1383 		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1384 	}
1385 	return -1;
1386 }
1387 
1388 struct netlink_broadcast_data {
1389 	struct sock *exclude_sk;
1390 	struct net *net;
1391 	u32 portid;
1392 	u32 group;
1393 	int failure;
1394 	int delivery_failure;
1395 	int congested;
1396 	int delivered;
1397 	gfp_t allocation;
1398 	struct sk_buff *skb, *skb2;
1399 	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1400 	void *tx_data;
1401 };
1402 
1403 static void do_one_broadcast(struct sock *sk,
1404 				    struct netlink_broadcast_data *p)
1405 {
1406 	struct netlink_sock *nlk = nlk_sk(sk);
1407 	int val;
1408 
1409 	if (p->exclude_sk == sk)
1410 		return;
1411 
1412 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1413 	    !test_bit(p->group - 1, nlk->groups))
1414 		return;
1415 
1416 	if (!net_eq(sock_net(sk), p->net)) {
1417 		if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1418 			return;
1419 
1420 		if (!peernet_has_id(sock_net(sk), p->net))
1421 			return;
1422 
1423 		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1424 				     CAP_NET_BROADCAST))
1425 			return;
1426 	}
1427 
1428 	if (p->failure) {
1429 		netlink_overrun(sk);
1430 		return;
1431 	}
1432 
1433 	sock_hold(sk);
1434 	if (p->skb2 == NULL) {
1435 		if (skb_shared(p->skb)) {
1436 			p->skb2 = skb_clone(p->skb, p->allocation);
1437 		} else {
1438 			p->skb2 = skb_get(p->skb);
1439 			/*
1440 			 * skb ownership may have been set when
1441 			 * delivered to a previous socket.
1442 			 */
1443 			skb_orphan(p->skb2);
1444 		}
1445 	}
1446 	if (p->skb2 == NULL) {
1447 		netlink_overrun(sk);
1448 		/* Clone failed. Notify ALL listeners. */
1449 		p->failure = 1;
1450 		if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1451 			p->delivery_failure = 1;
1452 		goto out;
1453 	}
1454 	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1455 		kfree_skb(p->skb2);
1456 		p->skb2 = NULL;
1457 		goto out;
1458 	}
1459 	if (sk_filter(sk, p->skb2)) {
1460 		kfree_skb(p->skb2);
1461 		p->skb2 = NULL;
1462 		goto out;
1463 	}
1464 	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1465 	if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1466 		NETLINK_CB(p->skb2).nsid_is_set = true;
1467 	val = netlink_broadcast_deliver(sk, p->skb2);
1468 	if (val < 0) {
1469 		netlink_overrun(sk);
1470 		if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1471 			p->delivery_failure = 1;
1472 	} else {
1473 		p->congested |= val;
1474 		p->delivered = 1;
1475 		p->skb2 = NULL;
1476 	}
1477 out:
1478 	sock_put(sk);
1479 }
1480 
1481 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1482 	u32 group, gfp_t allocation,
1483 	int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1484 	void *filter_data)
1485 {
1486 	struct net *net = sock_net(ssk);
1487 	struct netlink_broadcast_data info;
1488 	struct sock *sk;
1489 
1490 	skb = netlink_trim(skb, allocation);
1491 
1492 	info.exclude_sk = ssk;
1493 	info.net = net;
1494 	info.portid = portid;
1495 	info.group = group;
1496 	info.failure = 0;
1497 	info.delivery_failure = 0;
1498 	info.congested = 0;
1499 	info.delivered = 0;
1500 	info.allocation = allocation;
1501 	info.skb = skb;
1502 	info.skb2 = NULL;
1503 	info.tx_filter = filter;
1504 	info.tx_data = filter_data;
1505 
1506 	/* While we sleep in clone, do not allow to change socket list */
1507 
1508 	netlink_lock_table();
1509 
1510 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1511 		do_one_broadcast(sk, &info);
1512 
1513 	consume_skb(skb);
1514 
1515 	netlink_unlock_table();
1516 
1517 	if (info.delivery_failure) {
1518 		kfree_skb(info.skb2);
1519 		return -ENOBUFS;
1520 	}
1521 	consume_skb(info.skb2);
1522 
1523 	if (info.delivered) {
1524 		if (info.congested && gfpflags_allow_blocking(allocation))
1525 			yield();
1526 		return 0;
1527 	}
1528 	return -ESRCH;
1529 }
1530 EXPORT_SYMBOL(netlink_broadcast_filtered);
1531 
1532 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1533 		      u32 group, gfp_t allocation)
1534 {
1535 	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1536 		NULL, NULL);
1537 }
1538 EXPORT_SYMBOL(netlink_broadcast);
1539 
1540 struct netlink_set_err_data {
1541 	struct sock *exclude_sk;
1542 	u32 portid;
1543 	u32 group;
1544 	int code;
1545 };
1546 
1547 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1548 {
1549 	struct netlink_sock *nlk = nlk_sk(sk);
1550 	int ret = 0;
1551 
1552 	if (sk == p->exclude_sk)
1553 		goto out;
1554 
1555 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1556 		goto out;
1557 
1558 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1559 	    !test_bit(p->group - 1, nlk->groups))
1560 		goto out;
1561 
1562 	if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1563 		ret = 1;
1564 		goto out;
1565 	}
1566 
1567 	sk->sk_err = p->code;
1568 	sk->sk_error_report(sk);
1569 out:
1570 	return ret;
1571 }
1572 
1573 /**
1574  * netlink_set_err - report error to broadcast listeners
1575  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1576  * @portid: the PORTID of a process that we want to skip (if any)
1577  * @group: the broadcast group that will notice the error
1578  * @code: error code, must be negative (as usual in kernelspace)
1579  *
1580  * This function returns the number of broadcast listeners that have set the
1581  * NETLINK_NO_ENOBUFS socket option.
1582  */
1583 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1584 {
1585 	struct netlink_set_err_data info;
1586 	struct sock *sk;
1587 	int ret = 0;
1588 
1589 	info.exclude_sk = ssk;
1590 	info.portid = portid;
1591 	info.group = group;
1592 	/* sk->sk_err wants a positive error value */
1593 	info.code = -code;
1594 
1595 	read_lock(&nl_table_lock);
1596 
1597 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1598 		ret += do_one_set_err(sk, &info);
1599 
1600 	read_unlock(&nl_table_lock);
1601 	return ret;
1602 }
1603 EXPORT_SYMBOL(netlink_set_err);
1604 
1605 /* must be called with netlink table grabbed */
1606 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1607 				     unsigned int group,
1608 				     int is_new)
1609 {
1610 	int old, new = !!is_new, subscriptions;
1611 
1612 	old = test_bit(group - 1, nlk->groups);
1613 	subscriptions = nlk->subscriptions - old + new;
1614 	if (new)
1615 		__set_bit(group - 1, nlk->groups);
1616 	else
1617 		__clear_bit(group - 1, nlk->groups);
1618 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1619 	netlink_update_listeners(&nlk->sk);
1620 }
1621 
1622 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1623 			      char __user *optval, unsigned int optlen)
1624 {
1625 	struct sock *sk = sock->sk;
1626 	struct netlink_sock *nlk = nlk_sk(sk);
1627 	unsigned int val = 0;
1628 	int err;
1629 
1630 	if (level != SOL_NETLINK)
1631 		return -ENOPROTOOPT;
1632 
1633 	if (optlen >= sizeof(int) &&
1634 	    get_user(val, (unsigned int __user *)optval))
1635 		return -EFAULT;
1636 
1637 	switch (optname) {
1638 	case NETLINK_PKTINFO:
1639 		if (val)
1640 			nlk->flags |= NETLINK_F_RECV_PKTINFO;
1641 		else
1642 			nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1643 		err = 0;
1644 		break;
1645 	case NETLINK_ADD_MEMBERSHIP:
1646 	case NETLINK_DROP_MEMBERSHIP: {
1647 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1648 			return -EPERM;
1649 		err = netlink_realloc_groups(sk);
1650 		if (err)
1651 			return err;
1652 		if (!val || val - 1 >= nlk->ngroups)
1653 			return -EINVAL;
1654 		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1655 			err = nlk->netlink_bind(sock_net(sk), val);
1656 			if (err)
1657 				return err;
1658 		}
1659 		netlink_table_grab();
1660 		netlink_update_socket_mc(nlk, val,
1661 					 optname == NETLINK_ADD_MEMBERSHIP);
1662 		netlink_table_ungrab();
1663 		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1664 			nlk->netlink_unbind(sock_net(sk), val);
1665 
1666 		err = 0;
1667 		break;
1668 	}
1669 	case NETLINK_BROADCAST_ERROR:
1670 		if (val)
1671 			nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1672 		else
1673 			nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1674 		err = 0;
1675 		break;
1676 	case NETLINK_NO_ENOBUFS:
1677 		if (val) {
1678 			nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1679 			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1680 			wake_up_interruptible(&nlk->wait);
1681 		} else {
1682 			nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1683 		}
1684 		err = 0;
1685 		break;
1686 	case NETLINK_LISTEN_ALL_NSID:
1687 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1688 			return -EPERM;
1689 
1690 		if (val)
1691 			nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1692 		else
1693 			nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1694 		err = 0;
1695 		break;
1696 	case NETLINK_CAP_ACK:
1697 		if (val)
1698 			nlk->flags |= NETLINK_F_CAP_ACK;
1699 		else
1700 			nlk->flags &= ~NETLINK_F_CAP_ACK;
1701 		err = 0;
1702 		break;
1703 	case NETLINK_EXT_ACK:
1704 		if (val)
1705 			nlk->flags |= NETLINK_F_EXT_ACK;
1706 		else
1707 			nlk->flags &= ~NETLINK_F_EXT_ACK;
1708 		err = 0;
1709 		break;
1710 	default:
1711 		err = -ENOPROTOOPT;
1712 	}
1713 	return err;
1714 }
1715 
1716 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1717 			      char __user *optval, int __user *optlen)
1718 {
1719 	struct sock *sk = sock->sk;
1720 	struct netlink_sock *nlk = nlk_sk(sk);
1721 	int len, val, err;
1722 
1723 	if (level != SOL_NETLINK)
1724 		return -ENOPROTOOPT;
1725 
1726 	if (get_user(len, optlen))
1727 		return -EFAULT;
1728 	if (len < 0)
1729 		return -EINVAL;
1730 
1731 	switch (optname) {
1732 	case NETLINK_PKTINFO:
1733 		if (len < sizeof(int))
1734 			return -EINVAL;
1735 		len = sizeof(int);
1736 		val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1737 		if (put_user(len, optlen) ||
1738 		    put_user(val, optval))
1739 			return -EFAULT;
1740 		err = 0;
1741 		break;
1742 	case NETLINK_BROADCAST_ERROR:
1743 		if (len < sizeof(int))
1744 			return -EINVAL;
1745 		len = sizeof(int);
1746 		val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1747 		if (put_user(len, optlen) ||
1748 		    put_user(val, optval))
1749 			return -EFAULT;
1750 		err = 0;
1751 		break;
1752 	case NETLINK_NO_ENOBUFS:
1753 		if (len < sizeof(int))
1754 			return -EINVAL;
1755 		len = sizeof(int);
1756 		val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1757 		if (put_user(len, optlen) ||
1758 		    put_user(val, optval))
1759 			return -EFAULT;
1760 		err = 0;
1761 		break;
1762 	case NETLINK_LIST_MEMBERSHIPS: {
1763 		int pos, idx, shift;
1764 
1765 		err = 0;
1766 		netlink_lock_table();
1767 		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1768 			if (len - pos < sizeof(u32))
1769 				break;
1770 
1771 			idx = pos / sizeof(unsigned long);
1772 			shift = (pos % sizeof(unsigned long)) * 8;
1773 			if (put_user((u32)(nlk->groups[idx] >> shift),
1774 				     (u32 __user *)(optval + pos))) {
1775 				err = -EFAULT;
1776 				break;
1777 			}
1778 		}
1779 		if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1780 			err = -EFAULT;
1781 		netlink_unlock_table();
1782 		break;
1783 	}
1784 	case NETLINK_CAP_ACK:
1785 		if (len < sizeof(int))
1786 			return -EINVAL;
1787 		len = sizeof(int);
1788 		val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1789 		if (put_user(len, optlen) ||
1790 		    put_user(val, optval))
1791 			return -EFAULT;
1792 		err = 0;
1793 		break;
1794 	case NETLINK_EXT_ACK:
1795 		if (len < sizeof(int))
1796 			return -EINVAL;
1797 		len = sizeof(int);
1798 		val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1799 		if (put_user(len, optlen) || put_user(val, optval))
1800 			return -EFAULT;
1801 		err = 0;
1802 		break;
1803 	default:
1804 		err = -ENOPROTOOPT;
1805 	}
1806 	return err;
1807 }
1808 
1809 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1810 {
1811 	struct nl_pktinfo info;
1812 
1813 	info.group = NETLINK_CB(skb).dst_group;
1814 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1815 }
1816 
1817 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1818 					 struct sk_buff *skb)
1819 {
1820 	if (!NETLINK_CB(skb).nsid_is_set)
1821 		return;
1822 
1823 	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1824 		 &NETLINK_CB(skb).nsid);
1825 }
1826 
1827 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1828 {
1829 	struct sock *sk = sock->sk;
1830 	struct netlink_sock *nlk = nlk_sk(sk);
1831 	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1832 	u32 dst_portid;
1833 	u32 dst_group;
1834 	struct sk_buff *skb;
1835 	int err;
1836 	struct scm_cookie scm;
1837 	u32 netlink_skb_flags = 0;
1838 
1839 	if (msg->msg_flags&MSG_OOB)
1840 		return -EOPNOTSUPP;
1841 
1842 	err = scm_send(sock, msg, &scm, true);
1843 	if (err < 0)
1844 		return err;
1845 
1846 	if (msg->msg_namelen) {
1847 		err = -EINVAL;
1848 		if (addr->nl_family != AF_NETLINK)
1849 			goto out;
1850 		dst_portid = addr->nl_pid;
1851 		dst_group = ffs(addr->nl_groups);
1852 		err =  -EPERM;
1853 		if ((dst_group || dst_portid) &&
1854 		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1855 			goto out;
1856 		netlink_skb_flags |= NETLINK_SKB_DST;
1857 	} else {
1858 		dst_portid = nlk->dst_portid;
1859 		dst_group = nlk->dst_group;
1860 	}
1861 
1862 	if (!nlk->bound) {
1863 		err = netlink_autobind(sock);
1864 		if (err)
1865 			goto out;
1866 	} else {
1867 		/* Ensure nlk is hashed and visible. */
1868 		smp_rmb();
1869 	}
1870 
1871 	err = -EMSGSIZE;
1872 	if (len > sk->sk_sndbuf - 32)
1873 		goto out;
1874 	err = -ENOBUFS;
1875 	skb = netlink_alloc_large_skb(len, dst_group);
1876 	if (skb == NULL)
1877 		goto out;
1878 
1879 	NETLINK_CB(skb).portid	= nlk->portid;
1880 	NETLINK_CB(skb).dst_group = dst_group;
1881 	NETLINK_CB(skb).creds	= scm.creds;
1882 	NETLINK_CB(skb).flags	= netlink_skb_flags;
1883 
1884 	err = -EFAULT;
1885 	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1886 		kfree_skb(skb);
1887 		goto out;
1888 	}
1889 
1890 	err = security_netlink_send(sk, skb);
1891 	if (err) {
1892 		kfree_skb(skb);
1893 		goto out;
1894 	}
1895 
1896 	if (dst_group) {
1897 		refcount_inc(&skb->users);
1898 		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1899 	}
1900 	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1901 
1902 out:
1903 	scm_destroy(&scm);
1904 	return err;
1905 }
1906 
1907 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1908 			   int flags)
1909 {
1910 	struct scm_cookie scm;
1911 	struct sock *sk = sock->sk;
1912 	struct netlink_sock *nlk = nlk_sk(sk);
1913 	int noblock = flags&MSG_DONTWAIT;
1914 	size_t copied;
1915 	struct sk_buff *skb, *data_skb;
1916 	int err, ret;
1917 
1918 	if (flags&MSG_OOB)
1919 		return -EOPNOTSUPP;
1920 
1921 	copied = 0;
1922 
1923 	skb = skb_recv_datagram(sk, flags, noblock, &err);
1924 	if (skb == NULL)
1925 		goto out;
1926 
1927 	data_skb = skb;
1928 
1929 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1930 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1931 		/*
1932 		 * If this skb has a frag_list, then here that means that we
1933 		 * will have to use the frag_list skb's data for compat tasks
1934 		 * and the regular skb's data for normal (non-compat) tasks.
1935 		 *
1936 		 * If we need to send the compat skb, assign it to the
1937 		 * 'data_skb' variable so that it will be used below for data
1938 		 * copying. We keep 'skb' for everything else, including
1939 		 * freeing both later.
1940 		 */
1941 		if (flags & MSG_CMSG_COMPAT)
1942 			data_skb = skb_shinfo(skb)->frag_list;
1943 	}
1944 #endif
1945 
1946 	/* Record the max length of recvmsg() calls for future allocations */
1947 	nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1948 	nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1949 				     SKB_WITH_OVERHEAD(32768));
1950 
1951 	copied = data_skb->len;
1952 	if (len < copied) {
1953 		msg->msg_flags |= MSG_TRUNC;
1954 		copied = len;
1955 	}
1956 
1957 	skb_reset_transport_header(data_skb);
1958 	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1959 
1960 	if (msg->msg_name) {
1961 		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1962 		addr->nl_family = AF_NETLINK;
1963 		addr->nl_pad    = 0;
1964 		addr->nl_pid	= NETLINK_CB(skb).portid;
1965 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1966 		msg->msg_namelen = sizeof(*addr);
1967 	}
1968 
1969 	if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1970 		netlink_cmsg_recv_pktinfo(msg, skb);
1971 	if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1972 		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1973 
1974 	memset(&scm, 0, sizeof(scm));
1975 	scm.creds = *NETLINK_CREDS(skb);
1976 	if (flags & MSG_TRUNC)
1977 		copied = data_skb->len;
1978 
1979 	skb_free_datagram(sk, skb);
1980 
1981 	if (nlk->cb_running &&
1982 	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1983 		ret = netlink_dump(sk);
1984 		if (ret) {
1985 			sk->sk_err = -ret;
1986 			sk->sk_error_report(sk);
1987 		}
1988 	}
1989 
1990 	scm_recv(sock, msg, &scm, flags);
1991 out:
1992 	netlink_rcv_wake(sk);
1993 	return err ? : copied;
1994 }
1995 
1996 static void netlink_data_ready(struct sock *sk)
1997 {
1998 	BUG();
1999 }
2000 
2001 /*
2002  *	We export these functions to other modules. They provide a
2003  *	complete set of kernel non-blocking support for message
2004  *	queueing.
2005  */
2006 
2007 struct sock *
2008 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2009 			struct netlink_kernel_cfg *cfg)
2010 {
2011 	struct socket *sock;
2012 	struct sock *sk;
2013 	struct netlink_sock *nlk;
2014 	struct listeners *listeners = NULL;
2015 	struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2016 	unsigned int groups;
2017 
2018 	BUG_ON(!nl_table);
2019 
2020 	if (unit < 0 || unit >= MAX_LINKS)
2021 		return NULL;
2022 
2023 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2024 		return NULL;
2025 
2026 	if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2027 		goto out_sock_release_nosk;
2028 
2029 	sk = sock->sk;
2030 
2031 	if (!cfg || cfg->groups < 32)
2032 		groups = 32;
2033 	else
2034 		groups = cfg->groups;
2035 
2036 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2037 	if (!listeners)
2038 		goto out_sock_release;
2039 
2040 	sk->sk_data_ready = netlink_data_ready;
2041 	if (cfg && cfg->input)
2042 		nlk_sk(sk)->netlink_rcv = cfg->input;
2043 
2044 	if (netlink_insert(sk, 0))
2045 		goto out_sock_release;
2046 
2047 	nlk = nlk_sk(sk);
2048 	nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2049 
2050 	netlink_table_grab();
2051 	if (!nl_table[unit].registered) {
2052 		nl_table[unit].groups = groups;
2053 		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2054 		nl_table[unit].cb_mutex = cb_mutex;
2055 		nl_table[unit].module = module;
2056 		if (cfg) {
2057 			nl_table[unit].bind = cfg->bind;
2058 			nl_table[unit].unbind = cfg->unbind;
2059 			nl_table[unit].flags = cfg->flags;
2060 			if (cfg->compare)
2061 				nl_table[unit].compare = cfg->compare;
2062 		}
2063 		nl_table[unit].registered = 1;
2064 	} else {
2065 		kfree(listeners);
2066 		nl_table[unit].registered++;
2067 	}
2068 	netlink_table_ungrab();
2069 	return sk;
2070 
2071 out_sock_release:
2072 	kfree(listeners);
2073 	netlink_kernel_release(sk);
2074 	return NULL;
2075 
2076 out_sock_release_nosk:
2077 	sock_release(sock);
2078 	return NULL;
2079 }
2080 EXPORT_SYMBOL(__netlink_kernel_create);
2081 
2082 void
2083 netlink_kernel_release(struct sock *sk)
2084 {
2085 	if (sk == NULL || sk->sk_socket == NULL)
2086 		return;
2087 
2088 	sock_release(sk->sk_socket);
2089 }
2090 EXPORT_SYMBOL(netlink_kernel_release);
2091 
2092 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2093 {
2094 	struct listeners *new, *old;
2095 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2096 
2097 	if (groups < 32)
2098 		groups = 32;
2099 
2100 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2101 		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2102 		if (!new)
2103 			return -ENOMEM;
2104 		old = nl_deref_protected(tbl->listeners);
2105 		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2106 		rcu_assign_pointer(tbl->listeners, new);
2107 
2108 		kfree_rcu(old, rcu);
2109 	}
2110 	tbl->groups = groups;
2111 
2112 	return 0;
2113 }
2114 
2115 /**
2116  * netlink_change_ngroups - change number of multicast groups
2117  *
2118  * This changes the number of multicast groups that are available
2119  * on a certain netlink family. Note that it is not possible to
2120  * change the number of groups to below 32. Also note that it does
2121  * not implicitly call netlink_clear_multicast_users() when the
2122  * number of groups is reduced.
2123  *
2124  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2125  * @groups: The new number of groups.
2126  */
2127 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2128 {
2129 	int err;
2130 
2131 	netlink_table_grab();
2132 	err = __netlink_change_ngroups(sk, groups);
2133 	netlink_table_ungrab();
2134 
2135 	return err;
2136 }
2137 
2138 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2139 {
2140 	struct sock *sk;
2141 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2142 
2143 	sk_for_each_bound(sk, &tbl->mc_list)
2144 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2145 }
2146 
2147 struct nlmsghdr *
2148 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2149 {
2150 	struct nlmsghdr *nlh;
2151 	int size = nlmsg_msg_size(len);
2152 
2153 	nlh = skb_put(skb, NLMSG_ALIGN(size));
2154 	nlh->nlmsg_type = type;
2155 	nlh->nlmsg_len = size;
2156 	nlh->nlmsg_flags = flags;
2157 	nlh->nlmsg_pid = portid;
2158 	nlh->nlmsg_seq = seq;
2159 	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2160 		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2161 	return nlh;
2162 }
2163 EXPORT_SYMBOL(__nlmsg_put);
2164 
2165 /*
2166  * It looks a bit ugly.
2167  * It would be better to create kernel thread.
2168  */
2169 
2170 static int netlink_dump(struct sock *sk)
2171 {
2172 	struct netlink_sock *nlk = nlk_sk(sk);
2173 	struct netlink_callback *cb;
2174 	struct sk_buff *skb = NULL;
2175 	struct nlmsghdr *nlh;
2176 	struct module *module;
2177 	int err = -ENOBUFS;
2178 	int alloc_min_size;
2179 	int alloc_size;
2180 
2181 	mutex_lock(nlk->cb_mutex);
2182 	if (!nlk->cb_running) {
2183 		err = -EINVAL;
2184 		goto errout_skb;
2185 	}
2186 
2187 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2188 		goto errout_skb;
2189 
2190 	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2191 	 * required, but it makes sense to _attempt_ a 16K bytes allocation
2192 	 * to reduce number of system calls on dump operations, if user
2193 	 * ever provided a big enough buffer.
2194 	 */
2195 	cb = &nlk->cb;
2196 	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2197 
2198 	if (alloc_min_size < nlk->max_recvmsg_len) {
2199 		alloc_size = nlk->max_recvmsg_len;
2200 		skb = alloc_skb(alloc_size,
2201 				(GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2202 				__GFP_NOWARN | __GFP_NORETRY);
2203 	}
2204 	if (!skb) {
2205 		alloc_size = alloc_min_size;
2206 		skb = alloc_skb(alloc_size, GFP_KERNEL);
2207 	}
2208 	if (!skb)
2209 		goto errout_skb;
2210 
2211 	/* Trim skb to allocated size. User is expected to provide buffer as
2212 	 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2213 	 * netlink_recvmsg())). dump will pack as many smaller messages as
2214 	 * could fit within the allocated skb. skb is typically allocated
2215 	 * with larger space than required (could be as much as near 2x the
2216 	 * requested size with align to next power of 2 approach). Allowing
2217 	 * dump to use the excess space makes it difficult for a user to have a
2218 	 * reasonable static buffer based on the expected largest dump of a
2219 	 * single netdev. The outcome is MSG_TRUNC error.
2220 	 */
2221 	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2222 	netlink_skb_set_owner_r(skb, sk);
2223 
2224 	if (nlk->dump_done_errno > 0)
2225 		nlk->dump_done_errno = cb->dump(skb, cb);
2226 
2227 	if (nlk->dump_done_errno > 0 ||
2228 	    skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2229 		mutex_unlock(nlk->cb_mutex);
2230 
2231 		if (sk_filter(sk, skb))
2232 			kfree_skb(skb);
2233 		else
2234 			__netlink_sendskb(sk, skb);
2235 		return 0;
2236 	}
2237 
2238 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2239 			       sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2240 	if (WARN_ON(!nlh))
2241 		goto errout_skb;
2242 
2243 	nl_dump_check_consistent(cb, nlh);
2244 
2245 	memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2246 	       sizeof(nlk->dump_done_errno));
2247 
2248 	if (sk_filter(sk, skb))
2249 		kfree_skb(skb);
2250 	else
2251 		__netlink_sendskb(sk, skb);
2252 
2253 	if (cb->done)
2254 		cb->done(cb);
2255 
2256 	nlk->cb_running = false;
2257 	module = cb->module;
2258 	skb = cb->skb;
2259 	mutex_unlock(nlk->cb_mutex);
2260 	module_put(module);
2261 	consume_skb(skb);
2262 	return 0;
2263 
2264 errout_skb:
2265 	mutex_unlock(nlk->cb_mutex);
2266 	kfree_skb(skb);
2267 	return err;
2268 }
2269 
2270 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2271 			 const struct nlmsghdr *nlh,
2272 			 struct netlink_dump_control *control)
2273 {
2274 	struct netlink_callback *cb;
2275 	struct sock *sk;
2276 	struct netlink_sock *nlk;
2277 	int ret;
2278 
2279 	refcount_inc(&skb->users);
2280 
2281 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2282 	if (sk == NULL) {
2283 		ret = -ECONNREFUSED;
2284 		goto error_free;
2285 	}
2286 
2287 	nlk = nlk_sk(sk);
2288 	mutex_lock(nlk->cb_mutex);
2289 	/* A dump is in progress... */
2290 	if (nlk->cb_running) {
2291 		ret = -EBUSY;
2292 		goto error_unlock;
2293 	}
2294 	/* add reference of module which cb->dump belongs to */
2295 	if (!try_module_get(control->module)) {
2296 		ret = -EPROTONOSUPPORT;
2297 		goto error_unlock;
2298 	}
2299 
2300 	cb = &nlk->cb;
2301 	memset(cb, 0, sizeof(*cb));
2302 	cb->start = control->start;
2303 	cb->dump = control->dump;
2304 	cb->done = control->done;
2305 	cb->nlh = nlh;
2306 	cb->data = control->data;
2307 	cb->module = control->module;
2308 	cb->min_dump_alloc = control->min_dump_alloc;
2309 	cb->skb = skb;
2310 
2311 	if (cb->start) {
2312 		ret = cb->start(cb);
2313 		if (ret)
2314 			goto error_put;
2315 	}
2316 
2317 	nlk->cb_running = true;
2318 	nlk->dump_done_errno = INT_MAX;
2319 
2320 	mutex_unlock(nlk->cb_mutex);
2321 
2322 	ret = netlink_dump(sk);
2323 
2324 	sock_put(sk);
2325 
2326 	if (ret)
2327 		return ret;
2328 
2329 	/* We successfully started a dump, by returning -EINTR we
2330 	 * signal not to send ACK even if it was requested.
2331 	 */
2332 	return -EINTR;
2333 
2334 error_put:
2335 	module_put(control->module);
2336 error_unlock:
2337 	sock_put(sk);
2338 	mutex_unlock(nlk->cb_mutex);
2339 error_free:
2340 	kfree_skb(skb);
2341 	return ret;
2342 }
2343 EXPORT_SYMBOL(__netlink_dump_start);
2344 
2345 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2346 		 const struct netlink_ext_ack *extack)
2347 {
2348 	struct sk_buff *skb;
2349 	struct nlmsghdr *rep;
2350 	struct nlmsgerr *errmsg;
2351 	size_t payload = sizeof(*errmsg);
2352 	size_t tlvlen = 0;
2353 	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2354 	unsigned int flags = 0;
2355 	bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2356 
2357 	/* Error messages get the original request appened, unless the user
2358 	 * requests to cap the error message, and get extra error data if
2359 	 * requested.
2360 	 */
2361 	if (nlk_has_extack && extack && extack->_msg)
2362 		tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2363 
2364 	if (err) {
2365 		if (!(nlk->flags & NETLINK_F_CAP_ACK))
2366 			payload += nlmsg_len(nlh);
2367 		else
2368 			flags |= NLM_F_CAPPED;
2369 		if (nlk_has_extack && extack && extack->bad_attr)
2370 			tlvlen += nla_total_size(sizeof(u32));
2371 	} else {
2372 		flags |= NLM_F_CAPPED;
2373 
2374 		if (nlk_has_extack && extack && extack->cookie_len)
2375 			tlvlen += nla_total_size(extack->cookie_len);
2376 	}
2377 
2378 	if (tlvlen)
2379 		flags |= NLM_F_ACK_TLVS;
2380 
2381 	skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2382 	if (!skb) {
2383 		NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2384 		NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2385 		return;
2386 	}
2387 
2388 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2389 			  NLMSG_ERROR, payload, flags);
2390 	errmsg = nlmsg_data(rep);
2391 	errmsg->error = err;
2392 	memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2393 
2394 	if (nlk_has_extack && extack) {
2395 		if (extack->_msg) {
2396 			WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2397 					       extack->_msg));
2398 		}
2399 		if (err) {
2400 			if (extack->bad_attr &&
2401 			    !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2402 				     (u8 *)extack->bad_attr >= in_skb->data +
2403 							       in_skb->len))
2404 				WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2405 						    (u8 *)extack->bad_attr -
2406 						    in_skb->data));
2407 		} else {
2408 			if (extack->cookie_len)
2409 				WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2410 						extack->cookie_len,
2411 						extack->cookie));
2412 		}
2413 	}
2414 
2415 	nlmsg_end(skb, rep);
2416 
2417 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2418 }
2419 EXPORT_SYMBOL(netlink_ack);
2420 
2421 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2422 						   struct nlmsghdr *,
2423 						   struct netlink_ext_ack *))
2424 {
2425 	struct netlink_ext_ack extack;
2426 	struct nlmsghdr *nlh;
2427 	int err;
2428 
2429 	while (skb->len >= nlmsg_total_size(0)) {
2430 		int msglen;
2431 
2432 		memset(&extack, 0, sizeof(extack));
2433 		nlh = nlmsg_hdr(skb);
2434 		err = 0;
2435 
2436 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2437 			return 0;
2438 
2439 		/* Only requests are handled by the kernel */
2440 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2441 			goto ack;
2442 
2443 		/* Skip control messages */
2444 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2445 			goto ack;
2446 
2447 		err = cb(skb, nlh, &extack);
2448 		if (err == -EINTR)
2449 			goto skip;
2450 
2451 ack:
2452 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2453 			netlink_ack(skb, nlh, err, &extack);
2454 
2455 skip:
2456 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2457 		if (msglen > skb->len)
2458 			msglen = skb->len;
2459 		skb_pull(skb, msglen);
2460 	}
2461 
2462 	return 0;
2463 }
2464 EXPORT_SYMBOL(netlink_rcv_skb);
2465 
2466 /**
2467  * nlmsg_notify - send a notification netlink message
2468  * @sk: netlink socket to use
2469  * @skb: notification message
2470  * @portid: destination netlink portid for reports or 0
2471  * @group: destination multicast group or 0
2472  * @report: 1 to report back, 0 to disable
2473  * @flags: allocation flags
2474  */
2475 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2476 		 unsigned int group, int report, gfp_t flags)
2477 {
2478 	int err = 0;
2479 
2480 	if (group) {
2481 		int exclude_portid = 0;
2482 
2483 		if (report) {
2484 			refcount_inc(&skb->users);
2485 			exclude_portid = portid;
2486 		}
2487 
2488 		/* errors reported via destination sk->sk_err, but propagate
2489 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2490 		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2491 	}
2492 
2493 	if (report) {
2494 		int err2;
2495 
2496 		err2 = nlmsg_unicast(sk, skb, portid);
2497 		if (!err || err == -ESRCH)
2498 			err = err2;
2499 	}
2500 
2501 	return err;
2502 }
2503 EXPORT_SYMBOL(nlmsg_notify);
2504 
2505 #ifdef CONFIG_PROC_FS
2506 struct nl_seq_iter {
2507 	struct seq_net_private p;
2508 	struct rhashtable_iter hti;
2509 	int link;
2510 };
2511 
2512 static int netlink_walk_start(struct nl_seq_iter *iter)
2513 {
2514 	int err;
2515 
2516 	err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2517 				   GFP_KERNEL);
2518 	if (err) {
2519 		iter->link = MAX_LINKS;
2520 		return err;
2521 	}
2522 
2523 	rhashtable_walk_start(&iter->hti);
2524 
2525 	return 0;
2526 }
2527 
2528 static void netlink_walk_stop(struct nl_seq_iter *iter)
2529 {
2530 	rhashtable_walk_stop(&iter->hti);
2531 	rhashtable_walk_exit(&iter->hti);
2532 }
2533 
2534 static void *__netlink_seq_next(struct seq_file *seq)
2535 {
2536 	struct nl_seq_iter *iter = seq->private;
2537 	struct netlink_sock *nlk;
2538 
2539 	do {
2540 		for (;;) {
2541 			int err;
2542 
2543 			nlk = rhashtable_walk_next(&iter->hti);
2544 
2545 			if (IS_ERR(nlk)) {
2546 				if (PTR_ERR(nlk) == -EAGAIN)
2547 					continue;
2548 
2549 				return nlk;
2550 			}
2551 
2552 			if (nlk)
2553 				break;
2554 
2555 			netlink_walk_stop(iter);
2556 			if (++iter->link >= MAX_LINKS)
2557 				return NULL;
2558 
2559 			err = netlink_walk_start(iter);
2560 			if (err)
2561 				return ERR_PTR(err);
2562 		}
2563 	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2564 
2565 	return nlk;
2566 }
2567 
2568 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2569 {
2570 	struct nl_seq_iter *iter = seq->private;
2571 	void *obj = SEQ_START_TOKEN;
2572 	loff_t pos;
2573 	int err;
2574 
2575 	iter->link = 0;
2576 
2577 	err = netlink_walk_start(iter);
2578 	if (err)
2579 		return ERR_PTR(err);
2580 
2581 	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2582 		obj = __netlink_seq_next(seq);
2583 
2584 	return obj;
2585 }
2586 
2587 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2588 {
2589 	++*pos;
2590 	return __netlink_seq_next(seq);
2591 }
2592 
2593 static void netlink_seq_stop(struct seq_file *seq, void *v)
2594 {
2595 	struct nl_seq_iter *iter = seq->private;
2596 
2597 	if (iter->link >= MAX_LINKS)
2598 		return;
2599 
2600 	netlink_walk_stop(iter);
2601 }
2602 
2603 
2604 static int netlink_seq_show(struct seq_file *seq, void *v)
2605 {
2606 	if (v == SEQ_START_TOKEN) {
2607 		seq_puts(seq,
2608 			 "sk       Eth Pid    Groups   "
2609 			 "Rmem     Wmem     Dump     Locks     Drops     Inode\n");
2610 	} else {
2611 		struct sock *s = v;
2612 		struct netlink_sock *nlk = nlk_sk(s);
2613 
2614 		seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2615 			   s,
2616 			   s->sk_protocol,
2617 			   nlk->portid,
2618 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2619 			   sk_rmem_alloc_get(s),
2620 			   sk_wmem_alloc_get(s),
2621 			   nlk->cb_running,
2622 			   refcount_read(&s->sk_refcnt),
2623 			   atomic_read(&s->sk_drops),
2624 			   sock_i_ino(s)
2625 			);
2626 
2627 	}
2628 	return 0;
2629 }
2630 
2631 static const struct seq_operations netlink_seq_ops = {
2632 	.start  = netlink_seq_start,
2633 	.next   = netlink_seq_next,
2634 	.stop   = netlink_seq_stop,
2635 	.show   = netlink_seq_show,
2636 };
2637 
2638 
2639 static int netlink_seq_open(struct inode *inode, struct file *file)
2640 {
2641 	return seq_open_net(inode, file, &netlink_seq_ops,
2642 				sizeof(struct nl_seq_iter));
2643 }
2644 
2645 static const struct file_operations netlink_seq_fops = {
2646 	.open		= netlink_seq_open,
2647 	.read		= seq_read,
2648 	.llseek		= seq_lseek,
2649 	.release	= seq_release_net,
2650 };
2651 
2652 #endif
2653 
2654 int netlink_register_notifier(struct notifier_block *nb)
2655 {
2656 	return blocking_notifier_chain_register(&netlink_chain, nb);
2657 }
2658 EXPORT_SYMBOL(netlink_register_notifier);
2659 
2660 int netlink_unregister_notifier(struct notifier_block *nb)
2661 {
2662 	return blocking_notifier_chain_unregister(&netlink_chain, nb);
2663 }
2664 EXPORT_SYMBOL(netlink_unregister_notifier);
2665 
2666 static const struct proto_ops netlink_ops = {
2667 	.family =	PF_NETLINK,
2668 	.owner =	THIS_MODULE,
2669 	.release =	netlink_release,
2670 	.bind =		netlink_bind,
2671 	.connect =	netlink_connect,
2672 	.socketpair =	sock_no_socketpair,
2673 	.accept =	sock_no_accept,
2674 	.getname =	netlink_getname,
2675 	.poll =		datagram_poll,
2676 	.ioctl =	netlink_ioctl,
2677 	.listen =	sock_no_listen,
2678 	.shutdown =	sock_no_shutdown,
2679 	.setsockopt =	netlink_setsockopt,
2680 	.getsockopt =	netlink_getsockopt,
2681 	.sendmsg =	netlink_sendmsg,
2682 	.recvmsg =	netlink_recvmsg,
2683 	.mmap =		sock_no_mmap,
2684 	.sendpage =	sock_no_sendpage,
2685 };
2686 
2687 static const struct net_proto_family netlink_family_ops = {
2688 	.family = PF_NETLINK,
2689 	.create = netlink_create,
2690 	.owner	= THIS_MODULE,	/* for consistency 8) */
2691 };
2692 
2693 static int __net_init netlink_net_init(struct net *net)
2694 {
2695 #ifdef CONFIG_PROC_FS
2696 	if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2697 		return -ENOMEM;
2698 #endif
2699 	return 0;
2700 }
2701 
2702 static void __net_exit netlink_net_exit(struct net *net)
2703 {
2704 #ifdef CONFIG_PROC_FS
2705 	remove_proc_entry("netlink", net->proc_net);
2706 #endif
2707 }
2708 
2709 static void __init netlink_add_usersock_entry(void)
2710 {
2711 	struct listeners *listeners;
2712 	int groups = 32;
2713 
2714 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2715 	if (!listeners)
2716 		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2717 
2718 	netlink_table_grab();
2719 
2720 	nl_table[NETLINK_USERSOCK].groups = groups;
2721 	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2722 	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2723 	nl_table[NETLINK_USERSOCK].registered = 1;
2724 	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2725 
2726 	netlink_table_ungrab();
2727 }
2728 
2729 static struct pernet_operations __net_initdata netlink_net_ops = {
2730 	.init = netlink_net_init,
2731 	.exit = netlink_net_exit,
2732 };
2733 
2734 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2735 {
2736 	const struct netlink_sock *nlk = data;
2737 	struct netlink_compare_arg arg;
2738 
2739 	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2740 	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2741 }
2742 
2743 static const struct rhashtable_params netlink_rhashtable_params = {
2744 	.head_offset = offsetof(struct netlink_sock, node),
2745 	.key_len = netlink_compare_arg_len,
2746 	.obj_hashfn = netlink_hash,
2747 	.obj_cmpfn = netlink_compare,
2748 	.automatic_shrinking = true,
2749 };
2750 
2751 static int __init netlink_proto_init(void)
2752 {
2753 	int i;
2754 	int err = proto_register(&netlink_proto, 0);
2755 
2756 	if (err != 0)
2757 		goto out;
2758 
2759 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2760 
2761 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2762 	if (!nl_table)
2763 		goto panic;
2764 
2765 	for (i = 0; i < MAX_LINKS; i++) {
2766 		if (rhashtable_init(&nl_table[i].hash,
2767 				    &netlink_rhashtable_params) < 0) {
2768 			while (--i > 0)
2769 				rhashtable_destroy(&nl_table[i].hash);
2770 			kfree(nl_table);
2771 			goto panic;
2772 		}
2773 	}
2774 
2775 	netlink_add_usersock_entry();
2776 
2777 	sock_register(&netlink_family_ops);
2778 	register_pernet_subsys(&netlink_net_ops);
2779 	register_pernet_subsys(&netlink_tap_net_ops);
2780 	/* The netlink device handler may be needed early. */
2781 	rtnetlink_init();
2782 out:
2783 	return err;
2784 panic:
2785 	panic("netlink_init: Cannot allocate nl_table\n");
2786 }
2787 
2788 core_initcall(netlink_proto_init);
2789