xref: /openbmc/linux/net/netlink/af_netlink.c (revision e9b7b8b3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * NETLINK      Kernel-user communication protocol.
4  *
5  * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
6  * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7  * 				Patrick McHardy <kaber@trash.net>
8  *
9  * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10  *                               added netlink_proto_exit
11  * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12  * 				 use nlk_sk, as sk->protinfo is on a diet 8)
13  * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14  * 				 - inc module use count of module that owns
15  * 				   the kernel socket in case userspace opens
16  * 				   socket of same protocol
17  * 				 - remove all module support, since netlink is
18  * 				   mandatory if CONFIG_NET=y these days
19  */
20 
21 #include <linux/module.h>
22 
23 #include <linux/bpf.h>
24 #include <linux/capability.h>
25 #include <linux/kernel.h>
26 #include <linux/filter.h>
27 #include <linux/init.h>
28 #include <linux/signal.h>
29 #include <linux/sched.h>
30 #include <linux/errno.h>
31 #include <linux/string.h>
32 #include <linux/stat.h>
33 #include <linux/socket.h>
34 #include <linux/un.h>
35 #include <linux/fcntl.h>
36 #include <linux/termios.h>
37 #include <linux/sockios.h>
38 #include <linux/net.h>
39 #include <linux/fs.h>
40 #include <linux/slab.h>
41 #include <linux/uaccess.h>
42 #include <linux/skbuff.h>
43 #include <linux/netdevice.h>
44 #include <linux/rtnetlink.h>
45 #include <linux/proc_fs.h>
46 #include <linux/seq_file.h>
47 #include <linux/notifier.h>
48 #include <linux/security.h>
49 #include <linux/jhash.h>
50 #include <linux/jiffies.h>
51 #include <linux/random.h>
52 #include <linux/bitops.h>
53 #include <linux/mm.h>
54 #include <linux/types.h>
55 #include <linux/audit.h>
56 #include <linux/mutex.h>
57 #include <linux/vmalloc.h>
58 #include <linux/if_arp.h>
59 #include <linux/rhashtable.h>
60 #include <asm/cacheflush.h>
61 #include <linux/hash.h>
62 #include <linux/genetlink.h>
63 #include <linux/net_namespace.h>
64 #include <linux/nospec.h>
65 #include <linux/btf_ids.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 #define CREATE_TRACE_POINTS
73 #include <trace/events/netlink.h>
74 
75 #include "af_netlink.h"
76 
77 struct listeners {
78 	struct rcu_head		rcu;
79 	unsigned long		masks[];
80 };
81 
82 /* state bits */
83 #define NETLINK_S_CONGESTED		0x0
84 
85 static inline int netlink_is_kernel(struct sock *sk)
86 {
87 	return nlk_test_bit(KERNEL_SOCKET, sk);
88 }
89 
90 struct netlink_table *nl_table __read_mostly;
91 EXPORT_SYMBOL_GPL(nl_table);
92 
93 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94 
95 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96 
97 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
98 	"nlk_cb_mutex-ROUTE",
99 	"nlk_cb_mutex-1",
100 	"nlk_cb_mutex-USERSOCK",
101 	"nlk_cb_mutex-FIREWALL",
102 	"nlk_cb_mutex-SOCK_DIAG",
103 	"nlk_cb_mutex-NFLOG",
104 	"nlk_cb_mutex-XFRM",
105 	"nlk_cb_mutex-SELINUX",
106 	"nlk_cb_mutex-ISCSI",
107 	"nlk_cb_mutex-AUDIT",
108 	"nlk_cb_mutex-FIB_LOOKUP",
109 	"nlk_cb_mutex-CONNECTOR",
110 	"nlk_cb_mutex-NETFILTER",
111 	"nlk_cb_mutex-IP6_FW",
112 	"nlk_cb_mutex-DNRTMSG",
113 	"nlk_cb_mutex-KOBJECT_UEVENT",
114 	"nlk_cb_mutex-GENERIC",
115 	"nlk_cb_mutex-17",
116 	"nlk_cb_mutex-SCSITRANSPORT",
117 	"nlk_cb_mutex-ECRYPTFS",
118 	"nlk_cb_mutex-RDMA",
119 	"nlk_cb_mutex-CRYPTO",
120 	"nlk_cb_mutex-SMC",
121 	"nlk_cb_mutex-23",
122 	"nlk_cb_mutex-24",
123 	"nlk_cb_mutex-25",
124 	"nlk_cb_mutex-26",
125 	"nlk_cb_mutex-27",
126 	"nlk_cb_mutex-28",
127 	"nlk_cb_mutex-29",
128 	"nlk_cb_mutex-30",
129 	"nlk_cb_mutex-31",
130 	"nlk_cb_mutex-MAX_LINKS"
131 };
132 
133 static int netlink_dump(struct sock *sk);
134 
135 /* nl_table locking explained:
136  * Lookup and traversal are protected with an RCU read-side lock. Insertion
137  * and removal are protected with per bucket lock while using RCU list
138  * modification primitives and may run in parallel to RCU protected lookups.
139  * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140  * been acquired * either during or after the socket has been removed from
141  * the list and after an RCU grace period.
142  */
143 DEFINE_RWLOCK(nl_table_lock);
144 EXPORT_SYMBOL_GPL(nl_table_lock);
145 static atomic_t nl_table_users = ATOMIC_INIT(0);
146 
147 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
148 
149 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150 
151 
152 static const struct rhashtable_params netlink_rhashtable_params;
153 
154 void do_trace_netlink_extack(const char *msg)
155 {
156 	trace_netlink_extack(msg);
157 }
158 EXPORT_SYMBOL(do_trace_netlink_extack);
159 
160 static inline u32 netlink_group_mask(u32 group)
161 {
162 	if (group > 32)
163 		return 0;
164 	return group ? 1 << (group - 1) : 0;
165 }
166 
167 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
168 					   gfp_t gfp_mask)
169 {
170 	unsigned int len = skb_end_offset(skb);
171 	struct sk_buff *new;
172 
173 	new = alloc_skb(len, gfp_mask);
174 	if (new == NULL)
175 		return NULL;
176 
177 	NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
178 	NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
179 	NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
180 
181 	skb_put_data(new, skb->data, len);
182 	return new;
183 }
184 
185 static unsigned int netlink_tap_net_id;
186 
187 struct netlink_tap_net {
188 	struct list_head netlink_tap_all;
189 	struct mutex netlink_tap_lock;
190 };
191 
192 int netlink_add_tap(struct netlink_tap *nt)
193 {
194 	struct net *net = dev_net(nt->dev);
195 	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
196 
197 	if (unlikely(nt->dev->type != ARPHRD_NETLINK))
198 		return -EINVAL;
199 
200 	mutex_lock(&nn->netlink_tap_lock);
201 	list_add_rcu(&nt->list, &nn->netlink_tap_all);
202 	mutex_unlock(&nn->netlink_tap_lock);
203 
204 	__module_get(nt->module);
205 
206 	return 0;
207 }
208 EXPORT_SYMBOL_GPL(netlink_add_tap);
209 
210 static int __netlink_remove_tap(struct netlink_tap *nt)
211 {
212 	struct net *net = dev_net(nt->dev);
213 	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
214 	bool found = false;
215 	struct netlink_tap *tmp;
216 
217 	mutex_lock(&nn->netlink_tap_lock);
218 
219 	list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
220 		if (nt == tmp) {
221 			list_del_rcu(&nt->list);
222 			found = true;
223 			goto out;
224 		}
225 	}
226 
227 	pr_warn("__netlink_remove_tap: %p not found\n", nt);
228 out:
229 	mutex_unlock(&nn->netlink_tap_lock);
230 
231 	if (found)
232 		module_put(nt->module);
233 
234 	return found ? 0 : -ENODEV;
235 }
236 
237 int netlink_remove_tap(struct netlink_tap *nt)
238 {
239 	int ret;
240 
241 	ret = __netlink_remove_tap(nt);
242 	synchronize_net();
243 
244 	return ret;
245 }
246 EXPORT_SYMBOL_GPL(netlink_remove_tap);
247 
248 static __net_init int netlink_tap_init_net(struct net *net)
249 {
250 	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
251 
252 	INIT_LIST_HEAD(&nn->netlink_tap_all);
253 	mutex_init(&nn->netlink_tap_lock);
254 	return 0;
255 }
256 
257 static struct pernet_operations netlink_tap_net_ops = {
258 	.init = netlink_tap_init_net,
259 	.id   = &netlink_tap_net_id,
260 	.size = sizeof(struct netlink_tap_net),
261 };
262 
263 static bool netlink_filter_tap(const struct sk_buff *skb)
264 {
265 	struct sock *sk = skb->sk;
266 
267 	/* We take the more conservative approach and
268 	 * whitelist socket protocols that may pass.
269 	 */
270 	switch (sk->sk_protocol) {
271 	case NETLINK_ROUTE:
272 	case NETLINK_USERSOCK:
273 	case NETLINK_SOCK_DIAG:
274 	case NETLINK_NFLOG:
275 	case NETLINK_XFRM:
276 	case NETLINK_FIB_LOOKUP:
277 	case NETLINK_NETFILTER:
278 	case NETLINK_GENERIC:
279 		return true;
280 	}
281 
282 	return false;
283 }
284 
285 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
286 				     struct net_device *dev)
287 {
288 	struct sk_buff *nskb;
289 	struct sock *sk = skb->sk;
290 	int ret = -ENOMEM;
291 
292 	if (!net_eq(dev_net(dev), sock_net(sk)))
293 		return 0;
294 
295 	dev_hold(dev);
296 
297 	if (is_vmalloc_addr(skb->head))
298 		nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
299 	else
300 		nskb = skb_clone(skb, GFP_ATOMIC);
301 	if (nskb) {
302 		nskb->dev = dev;
303 		nskb->protocol = htons((u16) sk->sk_protocol);
304 		nskb->pkt_type = netlink_is_kernel(sk) ?
305 				 PACKET_KERNEL : PACKET_USER;
306 		skb_reset_network_header(nskb);
307 		ret = dev_queue_xmit(nskb);
308 		if (unlikely(ret > 0))
309 			ret = net_xmit_errno(ret);
310 	}
311 
312 	dev_put(dev);
313 	return ret;
314 }
315 
316 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
317 {
318 	int ret;
319 	struct netlink_tap *tmp;
320 
321 	if (!netlink_filter_tap(skb))
322 		return;
323 
324 	list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
325 		ret = __netlink_deliver_tap_skb(skb, tmp->dev);
326 		if (unlikely(ret))
327 			break;
328 	}
329 }
330 
331 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
332 {
333 	struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
334 
335 	rcu_read_lock();
336 
337 	if (unlikely(!list_empty(&nn->netlink_tap_all)))
338 		__netlink_deliver_tap(skb, nn);
339 
340 	rcu_read_unlock();
341 }
342 
343 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
344 				       struct sk_buff *skb)
345 {
346 	if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
347 		netlink_deliver_tap(sock_net(dst), skb);
348 }
349 
350 static void netlink_overrun(struct sock *sk)
351 {
352 	if (!nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
353 		if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 				      &nlk_sk(sk)->state)) {
355 			WRITE_ONCE(sk->sk_err, ENOBUFS);
356 			sk_error_report(sk);
357 		}
358 	}
359 	atomic_inc(&sk->sk_drops);
360 }
361 
362 static void netlink_rcv_wake(struct sock *sk)
363 {
364 	struct netlink_sock *nlk = nlk_sk(sk);
365 
366 	if (skb_queue_empty_lockless(&sk->sk_receive_queue))
367 		clear_bit(NETLINK_S_CONGESTED, &nlk->state);
368 	if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
369 		wake_up_interruptible(&nlk->wait);
370 }
371 
372 static void netlink_skb_destructor(struct sk_buff *skb)
373 {
374 	if (is_vmalloc_addr(skb->head)) {
375 		if (!skb->cloned ||
376 		    !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
377 			vfree(skb->head);
378 
379 		skb->head = NULL;
380 	}
381 	if (skb->sk != NULL)
382 		sock_rfree(skb);
383 }
384 
385 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
386 {
387 	WARN_ON(skb->sk != NULL);
388 	skb->sk = sk;
389 	skb->destructor = netlink_skb_destructor;
390 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
391 	sk_mem_charge(sk, skb->truesize);
392 }
393 
394 static void netlink_sock_destruct(struct sock *sk)
395 {
396 	struct netlink_sock *nlk = nlk_sk(sk);
397 
398 	if (nlk->cb_running) {
399 		if (nlk->cb.done)
400 			nlk->cb.done(&nlk->cb);
401 		module_put(nlk->cb.module);
402 		kfree_skb(nlk->cb.skb);
403 	}
404 
405 	skb_queue_purge(&sk->sk_receive_queue);
406 
407 	if (!sock_flag(sk, SOCK_DEAD)) {
408 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
409 		return;
410 	}
411 
412 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
413 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
414 	WARN_ON(nlk_sk(sk)->groups);
415 }
416 
417 static void netlink_sock_destruct_work(struct work_struct *work)
418 {
419 	struct netlink_sock *nlk = container_of(work, struct netlink_sock,
420 						work);
421 
422 	sk_free(&nlk->sk);
423 }
424 
425 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
426  * SMP. Look, when several writers sleep and reader wakes them up, all but one
427  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
428  * this, _but_ remember, it adds useless work on UP machines.
429  */
430 
431 void netlink_table_grab(void)
432 	__acquires(nl_table_lock)
433 {
434 	might_sleep();
435 
436 	write_lock_irq(&nl_table_lock);
437 
438 	if (atomic_read(&nl_table_users)) {
439 		DECLARE_WAITQUEUE(wait, current);
440 
441 		add_wait_queue_exclusive(&nl_table_wait, &wait);
442 		for (;;) {
443 			set_current_state(TASK_UNINTERRUPTIBLE);
444 			if (atomic_read(&nl_table_users) == 0)
445 				break;
446 			write_unlock_irq(&nl_table_lock);
447 			schedule();
448 			write_lock_irq(&nl_table_lock);
449 		}
450 
451 		__set_current_state(TASK_RUNNING);
452 		remove_wait_queue(&nl_table_wait, &wait);
453 	}
454 }
455 
456 void netlink_table_ungrab(void)
457 	__releases(nl_table_lock)
458 {
459 	write_unlock_irq(&nl_table_lock);
460 	wake_up(&nl_table_wait);
461 }
462 
463 static inline void
464 netlink_lock_table(void)
465 {
466 	unsigned long flags;
467 
468 	/* read_lock() synchronizes us to netlink_table_grab */
469 
470 	read_lock_irqsave(&nl_table_lock, flags);
471 	atomic_inc(&nl_table_users);
472 	read_unlock_irqrestore(&nl_table_lock, flags);
473 }
474 
475 static inline void
476 netlink_unlock_table(void)
477 {
478 	if (atomic_dec_and_test(&nl_table_users))
479 		wake_up(&nl_table_wait);
480 }
481 
482 struct netlink_compare_arg
483 {
484 	possible_net_t pnet;
485 	u32 portid;
486 };
487 
488 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
489 #define netlink_compare_arg_len \
490 	(offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
491 
492 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
493 				  const void *ptr)
494 {
495 	const struct netlink_compare_arg *x = arg->key;
496 	const struct netlink_sock *nlk = ptr;
497 
498 	return nlk->portid != x->portid ||
499 	       !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
500 }
501 
502 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
503 				     struct net *net, u32 portid)
504 {
505 	memset(arg, 0, sizeof(*arg));
506 	write_pnet(&arg->pnet, net);
507 	arg->portid = portid;
508 }
509 
510 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
511 				     struct net *net)
512 {
513 	struct netlink_compare_arg arg;
514 
515 	netlink_compare_arg_init(&arg, net, portid);
516 	return rhashtable_lookup_fast(&table->hash, &arg,
517 				      netlink_rhashtable_params);
518 }
519 
520 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
521 {
522 	struct netlink_compare_arg arg;
523 
524 	netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
525 	return rhashtable_lookup_insert_key(&table->hash, &arg,
526 					    &nlk_sk(sk)->node,
527 					    netlink_rhashtable_params);
528 }
529 
530 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
531 {
532 	struct netlink_table *table = &nl_table[protocol];
533 	struct sock *sk;
534 
535 	rcu_read_lock();
536 	sk = __netlink_lookup(table, portid, net);
537 	if (sk)
538 		sock_hold(sk);
539 	rcu_read_unlock();
540 
541 	return sk;
542 }
543 
544 static const struct proto_ops netlink_ops;
545 
546 static void
547 netlink_update_listeners(struct sock *sk)
548 {
549 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
550 	unsigned long mask;
551 	unsigned int i;
552 	struct listeners *listeners;
553 
554 	listeners = nl_deref_protected(tbl->listeners);
555 	if (!listeners)
556 		return;
557 
558 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
559 		mask = 0;
560 		sk_for_each_bound(sk, &tbl->mc_list) {
561 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
562 				mask |= nlk_sk(sk)->groups[i];
563 		}
564 		listeners->masks[i] = mask;
565 	}
566 	/* this function is only called with the netlink table "grabbed", which
567 	 * makes sure updates are visible before bind or setsockopt return. */
568 }
569 
570 static int netlink_insert(struct sock *sk, u32 portid)
571 {
572 	struct netlink_table *table = &nl_table[sk->sk_protocol];
573 	int err;
574 
575 	lock_sock(sk);
576 
577 	err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
578 	if (nlk_sk(sk)->bound)
579 		goto err;
580 
581 	/* portid can be read locklessly from netlink_getname(). */
582 	WRITE_ONCE(nlk_sk(sk)->portid, portid);
583 
584 	sock_hold(sk);
585 
586 	err = __netlink_insert(table, sk);
587 	if (err) {
588 		/* In case the hashtable backend returns with -EBUSY
589 		 * from here, it must not escape to the caller.
590 		 */
591 		if (unlikely(err == -EBUSY))
592 			err = -EOVERFLOW;
593 		if (err == -EEXIST)
594 			err = -EADDRINUSE;
595 		sock_put(sk);
596 		goto err;
597 	}
598 
599 	/* We need to ensure that the socket is hashed and visible. */
600 	smp_wmb();
601 	/* Paired with lockless reads from netlink_bind(),
602 	 * netlink_connect() and netlink_sendmsg().
603 	 */
604 	WRITE_ONCE(nlk_sk(sk)->bound, portid);
605 
606 err:
607 	release_sock(sk);
608 	return err;
609 }
610 
611 static void netlink_remove(struct sock *sk)
612 {
613 	struct netlink_table *table;
614 
615 	table = &nl_table[sk->sk_protocol];
616 	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
617 				    netlink_rhashtable_params)) {
618 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
619 		__sock_put(sk);
620 	}
621 
622 	netlink_table_grab();
623 	if (nlk_sk(sk)->subscriptions) {
624 		__sk_del_bind_node(sk);
625 		netlink_update_listeners(sk);
626 	}
627 	if (sk->sk_protocol == NETLINK_GENERIC)
628 		atomic_inc(&genl_sk_destructing_cnt);
629 	netlink_table_ungrab();
630 }
631 
632 static struct proto netlink_proto = {
633 	.name	  = "NETLINK",
634 	.owner	  = THIS_MODULE,
635 	.obj_size = sizeof(struct netlink_sock),
636 };
637 
638 static int __netlink_create(struct net *net, struct socket *sock,
639 			    struct mutex *cb_mutex, int protocol,
640 			    int kern)
641 {
642 	struct sock *sk;
643 	struct netlink_sock *nlk;
644 
645 	sock->ops = &netlink_ops;
646 
647 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
648 	if (!sk)
649 		return -ENOMEM;
650 
651 	sock_init_data(sock, sk);
652 
653 	nlk = nlk_sk(sk);
654 	if (cb_mutex) {
655 		nlk->cb_mutex = cb_mutex;
656 	} else {
657 		nlk->cb_mutex = &nlk->cb_def_mutex;
658 		mutex_init(nlk->cb_mutex);
659 		lockdep_set_class_and_name(nlk->cb_mutex,
660 					   nlk_cb_mutex_keys + protocol,
661 					   nlk_cb_mutex_key_strings[protocol]);
662 	}
663 	init_waitqueue_head(&nlk->wait);
664 
665 	sk->sk_destruct = netlink_sock_destruct;
666 	sk->sk_protocol = protocol;
667 	return 0;
668 }
669 
670 static int netlink_create(struct net *net, struct socket *sock, int protocol,
671 			  int kern)
672 {
673 	struct module *module = NULL;
674 	struct mutex *cb_mutex;
675 	struct netlink_sock *nlk;
676 	int (*bind)(struct net *net, int group);
677 	void (*unbind)(struct net *net, int group);
678 	void (*release)(struct sock *sock, unsigned long *groups);
679 	int err = 0;
680 
681 	sock->state = SS_UNCONNECTED;
682 
683 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
684 		return -ESOCKTNOSUPPORT;
685 
686 	if (protocol < 0 || protocol >= MAX_LINKS)
687 		return -EPROTONOSUPPORT;
688 	protocol = array_index_nospec(protocol, MAX_LINKS);
689 
690 	netlink_lock_table();
691 #ifdef CONFIG_MODULES
692 	if (!nl_table[protocol].registered) {
693 		netlink_unlock_table();
694 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
695 		netlink_lock_table();
696 	}
697 #endif
698 	if (nl_table[protocol].registered &&
699 	    try_module_get(nl_table[protocol].module))
700 		module = nl_table[protocol].module;
701 	else
702 		err = -EPROTONOSUPPORT;
703 	cb_mutex = nl_table[protocol].cb_mutex;
704 	bind = nl_table[protocol].bind;
705 	unbind = nl_table[protocol].unbind;
706 	release = nl_table[protocol].release;
707 	netlink_unlock_table();
708 
709 	if (err < 0)
710 		goto out;
711 
712 	err = __netlink_create(net, sock, cb_mutex, protocol, kern);
713 	if (err < 0)
714 		goto out_module;
715 
716 	sock_prot_inuse_add(net, &netlink_proto, 1);
717 
718 	nlk = nlk_sk(sock->sk);
719 	nlk->module = module;
720 	nlk->netlink_bind = bind;
721 	nlk->netlink_unbind = unbind;
722 	nlk->netlink_release = release;
723 out:
724 	return err;
725 
726 out_module:
727 	module_put(module);
728 	goto out;
729 }
730 
731 static void deferred_put_nlk_sk(struct rcu_head *head)
732 {
733 	struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
734 	struct sock *sk = &nlk->sk;
735 
736 	kfree(nlk->groups);
737 	nlk->groups = NULL;
738 
739 	if (!refcount_dec_and_test(&sk->sk_refcnt))
740 		return;
741 
742 	if (nlk->cb_running && nlk->cb.done) {
743 		INIT_WORK(&nlk->work, netlink_sock_destruct_work);
744 		schedule_work(&nlk->work);
745 		return;
746 	}
747 
748 	sk_free(sk);
749 }
750 
751 static int netlink_release(struct socket *sock)
752 {
753 	struct sock *sk = sock->sk;
754 	struct netlink_sock *nlk;
755 
756 	if (!sk)
757 		return 0;
758 
759 	netlink_remove(sk);
760 	sock_orphan(sk);
761 	nlk = nlk_sk(sk);
762 
763 	/*
764 	 * OK. Socket is unlinked, any packets that arrive now
765 	 * will be purged.
766 	 */
767 	if (nlk->netlink_release)
768 		nlk->netlink_release(sk, nlk->groups);
769 
770 	/* must not acquire netlink_table_lock in any way again before unbind
771 	 * and notifying genetlink is done as otherwise it might deadlock
772 	 */
773 	if (nlk->netlink_unbind) {
774 		int i;
775 
776 		for (i = 0; i < nlk->ngroups; i++)
777 			if (test_bit(i, nlk->groups))
778 				nlk->netlink_unbind(sock_net(sk), i + 1);
779 	}
780 	if (sk->sk_protocol == NETLINK_GENERIC &&
781 	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
782 		wake_up(&genl_sk_destructing_waitq);
783 
784 	sock->sk = NULL;
785 	wake_up_interruptible_all(&nlk->wait);
786 
787 	skb_queue_purge(&sk->sk_write_queue);
788 
789 	if (nlk->portid && nlk->bound) {
790 		struct netlink_notify n = {
791 						.net = sock_net(sk),
792 						.protocol = sk->sk_protocol,
793 						.portid = nlk->portid,
794 					  };
795 		blocking_notifier_call_chain(&netlink_chain,
796 				NETLINK_URELEASE, &n);
797 	}
798 
799 	module_put(nlk->module);
800 
801 	if (netlink_is_kernel(sk)) {
802 		netlink_table_grab();
803 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
804 		if (--nl_table[sk->sk_protocol].registered == 0) {
805 			struct listeners *old;
806 
807 			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
808 			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
809 			kfree_rcu(old, rcu);
810 			nl_table[sk->sk_protocol].module = NULL;
811 			nl_table[sk->sk_protocol].bind = NULL;
812 			nl_table[sk->sk_protocol].unbind = NULL;
813 			nl_table[sk->sk_protocol].flags = 0;
814 			nl_table[sk->sk_protocol].registered = 0;
815 		}
816 		netlink_table_ungrab();
817 	}
818 
819 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
820 
821 	/* Because struct net might disappear soon, do not keep a pointer. */
822 	if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
823 		__netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
824 		/* Because of deferred_put_nlk_sk and use of work queue,
825 		 * it is possible  netns will be freed before this socket.
826 		 */
827 		sock_net_set(sk, &init_net);
828 		__netns_tracker_alloc(&init_net, &sk->ns_tracker,
829 				      false, GFP_KERNEL);
830 	}
831 	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
832 	return 0;
833 }
834 
835 static int netlink_autobind(struct socket *sock)
836 {
837 	struct sock *sk = sock->sk;
838 	struct net *net = sock_net(sk);
839 	struct netlink_table *table = &nl_table[sk->sk_protocol];
840 	s32 portid = task_tgid_vnr(current);
841 	int err;
842 	s32 rover = -4096;
843 	bool ok;
844 
845 retry:
846 	cond_resched();
847 	rcu_read_lock();
848 	ok = !__netlink_lookup(table, portid, net);
849 	rcu_read_unlock();
850 	if (!ok) {
851 		/* Bind collision, search negative portid values. */
852 		if (rover == -4096)
853 			/* rover will be in range [S32_MIN, -4097] */
854 			rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
855 		else if (rover >= -4096)
856 			rover = -4097;
857 		portid = rover--;
858 		goto retry;
859 	}
860 
861 	err = netlink_insert(sk, portid);
862 	if (err == -EADDRINUSE)
863 		goto retry;
864 
865 	/* If 2 threads race to autobind, that is fine.  */
866 	if (err == -EBUSY)
867 		err = 0;
868 
869 	return err;
870 }
871 
872 /**
873  * __netlink_ns_capable - General netlink message capability test
874  * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
875  * @user_ns: The user namespace of the capability to use
876  * @cap: The capability to use
877  *
878  * Test to see if the opener of the socket we received the message
879  * from had when the netlink socket was created and the sender of the
880  * message has the capability @cap in the user namespace @user_ns.
881  */
882 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
883 			struct user_namespace *user_ns, int cap)
884 {
885 	return ((nsp->flags & NETLINK_SKB_DST) ||
886 		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
887 		ns_capable(user_ns, cap);
888 }
889 EXPORT_SYMBOL(__netlink_ns_capable);
890 
891 /**
892  * netlink_ns_capable - General netlink message capability test
893  * @skb: socket buffer holding a netlink command from userspace
894  * @user_ns: The user namespace of the capability to use
895  * @cap: The capability to use
896  *
897  * Test to see if the opener of the socket we received the message
898  * from had when the netlink socket was created and the sender of the
899  * message has the capability @cap in the user namespace @user_ns.
900  */
901 bool netlink_ns_capable(const struct sk_buff *skb,
902 			struct user_namespace *user_ns, int cap)
903 {
904 	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
905 }
906 EXPORT_SYMBOL(netlink_ns_capable);
907 
908 /**
909  * netlink_capable - Netlink global message capability test
910  * @skb: socket buffer holding a netlink command from userspace
911  * @cap: The capability to use
912  *
913  * Test to see if the opener of the socket we received the message
914  * from had when the netlink socket was created and the sender of the
915  * message has the capability @cap in all user namespaces.
916  */
917 bool netlink_capable(const struct sk_buff *skb, int cap)
918 {
919 	return netlink_ns_capable(skb, &init_user_ns, cap);
920 }
921 EXPORT_SYMBOL(netlink_capable);
922 
923 /**
924  * netlink_net_capable - Netlink network namespace message capability test
925  * @skb: socket buffer holding a netlink command from userspace
926  * @cap: The capability to use
927  *
928  * Test to see if the opener of the socket we received the message
929  * from had when the netlink socket was created and the sender of the
930  * message has the capability @cap over the network namespace of
931  * the socket we received the message from.
932  */
933 bool netlink_net_capable(const struct sk_buff *skb, int cap)
934 {
935 	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
936 }
937 EXPORT_SYMBOL(netlink_net_capable);
938 
939 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
940 {
941 	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
942 		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
943 }
944 
945 static void
946 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
947 {
948 	struct netlink_sock *nlk = nlk_sk(sk);
949 
950 	if (nlk->subscriptions && !subscriptions)
951 		__sk_del_bind_node(sk);
952 	else if (!nlk->subscriptions && subscriptions)
953 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
954 	nlk->subscriptions = subscriptions;
955 }
956 
957 static int netlink_realloc_groups(struct sock *sk)
958 {
959 	struct netlink_sock *nlk = nlk_sk(sk);
960 	unsigned int groups;
961 	unsigned long *new_groups;
962 	int err = 0;
963 
964 	netlink_table_grab();
965 
966 	groups = nl_table[sk->sk_protocol].groups;
967 	if (!nl_table[sk->sk_protocol].registered) {
968 		err = -ENOENT;
969 		goto out_unlock;
970 	}
971 
972 	if (nlk->ngroups >= groups)
973 		goto out_unlock;
974 
975 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
976 	if (new_groups == NULL) {
977 		err = -ENOMEM;
978 		goto out_unlock;
979 	}
980 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
981 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
982 
983 	nlk->groups = new_groups;
984 	nlk->ngroups = groups;
985  out_unlock:
986 	netlink_table_ungrab();
987 	return err;
988 }
989 
990 static void netlink_undo_bind(int group, long unsigned int groups,
991 			      struct sock *sk)
992 {
993 	struct netlink_sock *nlk = nlk_sk(sk);
994 	int undo;
995 
996 	if (!nlk->netlink_unbind)
997 		return;
998 
999 	for (undo = 0; undo < group; undo++)
1000 		if (test_bit(undo, &groups))
1001 			nlk->netlink_unbind(sock_net(sk), undo + 1);
1002 }
1003 
1004 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1005 			int addr_len)
1006 {
1007 	struct sock *sk = sock->sk;
1008 	struct net *net = sock_net(sk);
1009 	struct netlink_sock *nlk = nlk_sk(sk);
1010 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1011 	int err = 0;
1012 	unsigned long groups;
1013 	bool bound;
1014 
1015 	if (addr_len < sizeof(struct sockaddr_nl))
1016 		return -EINVAL;
1017 
1018 	if (nladdr->nl_family != AF_NETLINK)
1019 		return -EINVAL;
1020 	groups = nladdr->nl_groups;
1021 
1022 	/* Only superuser is allowed to listen multicasts */
1023 	if (groups) {
1024 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1025 			return -EPERM;
1026 		err = netlink_realloc_groups(sk);
1027 		if (err)
1028 			return err;
1029 	}
1030 
1031 	if (nlk->ngroups < BITS_PER_LONG)
1032 		groups &= (1UL << nlk->ngroups) - 1;
1033 
1034 	/* Paired with WRITE_ONCE() in netlink_insert() */
1035 	bound = READ_ONCE(nlk->bound);
1036 	if (bound) {
1037 		/* Ensure nlk->portid is up-to-date. */
1038 		smp_rmb();
1039 
1040 		if (nladdr->nl_pid != nlk->portid)
1041 			return -EINVAL;
1042 	}
1043 
1044 	if (nlk->netlink_bind && groups) {
1045 		int group;
1046 
1047 		/* nl_groups is a u32, so cap the maximum groups we can bind */
1048 		for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1049 			if (!test_bit(group, &groups))
1050 				continue;
1051 			err = nlk->netlink_bind(net, group + 1);
1052 			if (!err)
1053 				continue;
1054 			netlink_undo_bind(group, groups, sk);
1055 			return err;
1056 		}
1057 	}
1058 
1059 	/* No need for barriers here as we return to user-space without
1060 	 * using any of the bound attributes.
1061 	 */
1062 	netlink_lock_table();
1063 	if (!bound) {
1064 		err = nladdr->nl_pid ?
1065 			netlink_insert(sk, nladdr->nl_pid) :
1066 			netlink_autobind(sock);
1067 		if (err) {
1068 			netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1069 			goto unlock;
1070 		}
1071 	}
1072 
1073 	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1074 		goto unlock;
1075 	netlink_unlock_table();
1076 
1077 	netlink_table_grab();
1078 	netlink_update_subscriptions(sk, nlk->subscriptions +
1079 					 hweight32(groups) -
1080 					 hweight32(nlk->groups[0]));
1081 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1082 	netlink_update_listeners(sk);
1083 	netlink_table_ungrab();
1084 
1085 	return 0;
1086 
1087 unlock:
1088 	netlink_unlock_table();
1089 	return err;
1090 }
1091 
1092 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1093 			   int alen, int flags)
1094 {
1095 	int err = 0;
1096 	struct sock *sk = sock->sk;
1097 	struct netlink_sock *nlk = nlk_sk(sk);
1098 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1099 
1100 	if (alen < sizeof(addr->sa_family))
1101 		return -EINVAL;
1102 
1103 	if (addr->sa_family == AF_UNSPEC) {
1104 		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1105 		WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1106 		/* dst_portid and dst_group can be read locklessly */
1107 		WRITE_ONCE(nlk->dst_portid, 0);
1108 		WRITE_ONCE(nlk->dst_group, 0);
1109 		return 0;
1110 	}
1111 	if (addr->sa_family != AF_NETLINK)
1112 		return -EINVAL;
1113 
1114 	if (alen < sizeof(struct sockaddr_nl))
1115 		return -EINVAL;
1116 
1117 	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1118 	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1119 		return -EPERM;
1120 
1121 	/* No need for barriers here as we return to user-space without
1122 	 * using any of the bound attributes.
1123 	 * Paired with WRITE_ONCE() in netlink_insert().
1124 	 */
1125 	if (!READ_ONCE(nlk->bound))
1126 		err = netlink_autobind(sock);
1127 
1128 	if (err == 0) {
1129 		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1130 		WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1131 		/* dst_portid and dst_group can be read locklessly */
1132 		WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1133 		WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1134 	}
1135 
1136 	return err;
1137 }
1138 
1139 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1140 			   int peer)
1141 {
1142 	struct sock *sk = sock->sk;
1143 	struct netlink_sock *nlk = nlk_sk(sk);
1144 	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1145 
1146 	nladdr->nl_family = AF_NETLINK;
1147 	nladdr->nl_pad = 0;
1148 
1149 	if (peer) {
1150 		/* Paired with WRITE_ONCE() in netlink_connect() */
1151 		nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1152 		nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1153 	} else {
1154 		/* Paired with WRITE_ONCE() in netlink_insert() */
1155 		nladdr->nl_pid = READ_ONCE(nlk->portid);
1156 		netlink_lock_table();
1157 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1158 		netlink_unlock_table();
1159 	}
1160 	return sizeof(*nladdr);
1161 }
1162 
1163 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1164 			 unsigned long arg)
1165 {
1166 	/* try to hand this ioctl down to the NIC drivers.
1167 	 */
1168 	return -ENOIOCTLCMD;
1169 }
1170 
1171 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1172 {
1173 	struct sock *sock;
1174 	struct netlink_sock *nlk;
1175 
1176 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1177 	if (!sock)
1178 		return ERR_PTR(-ECONNREFUSED);
1179 
1180 	/* Don't bother queuing skb if kernel socket has no input function */
1181 	nlk = nlk_sk(sock);
1182 	/* dst_portid and sk_state can be changed in netlink_connect() */
1183 	if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1184 	    READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1185 		sock_put(sock);
1186 		return ERR_PTR(-ECONNREFUSED);
1187 	}
1188 	return sock;
1189 }
1190 
1191 struct sock *netlink_getsockbyfilp(struct file *filp)
1192 {
1193 	struct inode *inode = file_inode(filp);
1194 	struct sock *sock;
1195 
1196 	if (!S_ISSOCK(inode->i_mode))
1197 		return ERR_PTR(-ENOTSOCK);
1198 
1199 	sock = SOCKET_I(inode)->sk;
1200 	if (sock->sk_family != AF_NETLINK)
1201 		return ERR_PTR(-EINVAL);
1202 
1203 	sock_hold(sock);
1204 	return sock;
1205 }
1206 
1207 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1208 					       int broadcast)
1209 {
1210 	struct sk_buff *skb;
1211 	void *data;
1212 
1213 	if (size <= NLMSG_GOODSIZE || broadcast)
1214 		return alloc_skb(size, GFP_KERNEL);
1215 
1216 	size = SKB_DATA_ALIGN(size) +
1217 	       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1218 
1219 	data = vmalloc(size);
1220 	if (data == NULL)
1221 		return NULL;
1222 
1223 	skb = __build_skb(data, size);
1224 	if (skb == NULL)
1225 		vfree(data);
1226 	else
1227 		skb->destructor = netlink_skb_destructor;
1228 
1229 	return skb;
1230 }
1231 
1232 /*
1233  * Attach a skb to a netlink socket.
1234  * The caller must hold a reference to the destination socket. On error, the
1235  * reference is dropped. The skb is not send to the destination, just all
1236  * all error checks are performed and memory in the queue is reserved.
1237  * Return values:
1238  * < 0: error. skb freed, reference to sock dropped.
1239  * 0: continue
1240  * 1: repeat lookup - reference dropped while waiting for socket memory.
1241  */
1242 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1243 		      long *timeo, struct sock *ssk)
1244 {
1245 	struct netlink_sock *nlk;
1246 
1247 	nlk = nlk_sk(sk);
1248 
1249 	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1250 	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1251 		DECLARE_WAITQUEUE(wait, current);
1252 		if (!*timeo) {
1253 			if (!ssk || netlink_is_kernel(ssk))
1254 				netlink_overrun(sk);
1255 			sock_put(sk);
1256 			kfree_skb(skb);
1257 			return -EAGAIN;
1258 		}
1259 
1260 		__set_current_state(TASK_INTERRUPTIBLE);
1261 		add_wait_queue(&nlk->wait, &wait);
1262 
1263 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1264 		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1265 		    !sock_flag(sk, SOCK_DEAD))
1266 			*timeo = schedule_timeout(*timeo);
1267 
1268 		__set_current_state(TASK_RUNNING);
1269 		remove_wait_queue(&nlk->wait, &wait);
1270 		sock_put(sk);
1271 
1272 		if (signal_pending(current)) {
1273 			kfree_skb(skb);
1274 			return sock_intr_errno(*timeo);
1275 		}
1276 		return 1;
1277 	}
1278 	netlink_skb_set_owner_r(skb, sk);
1279 	return 0;
1280 }
1281 
1282 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1283 {
1284 	int len = skb->len;
1285 
1286 	netlink_deliver_tap(sock_net(sk), skb);
1287 
1288 	skb_queue_tail(&sk->sk_receive_queue, skb);
1289 	sk->sk_data_ready(sk);
1290 	return len;
1291 }
1292 
1293 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1294 {
1295 	int len = __netlink_sendskb(sk, skb);
1296 
1297 	sock_put(sk);
1298 	return len;
1299 }
1300 
1301 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1302 {
1303 	kfree_skb(skb);
1304 	sock_put(sk);
1305 }
1306 
1307 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1308 {
1309 	int delta;
1310 
1311 	WARN_ON(skb->sk != NULL);
1312 	delta = skb->end - skb->tail;
1313 	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1314 		return skb;
1315 
1316 	if (skb_shared(skb)) {
1317 		struct sk_buff *nskb = skb_clone(skb, allocation);
1318 		if (!nskb)
1319 			return skb;
1320 		consume_skb(skb);
1321 		skb = nskb;
1322 	}
1323 
1324 	pskb_expand_head(skb, 0, -delta,
1325 			 (allocation & ~__GFP_DIRECT_RECLAIM) |
1326 			 __GFP_NOWARN | __GFP_NORETRY);
1327 	return skb;
1328 }
1329 
1330 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1331 				  struct sock *ssk)
1332 {
1333 	int ret;
1334 	struct netlink_sock *nlk = nlk_sk(sk);
1335 
1336 	ret = -ECONNREFUSED;
1337 	if (nlk->netlink_rcv != NULL) {
1338 		ret = skb->len;
1339 		netlink_skb_set_owner_r(skb, sk);
1340 		NETLINK_CB(skb).sk = ssk;
1341 		netlink_deliver_tap_kernel(sk, ssk, skb);
1342 		nlk->netlink_rcv(skb);
1343 		consume_skb(skb);
1344 	} else {
1345 		kfree_skb(skb);
1346 	}
1347 	sock_put(sk);
1348 	return ret;
1349 }
1350 
1351 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1352 		    u32 portid, int nonblock)
1353 {
1354 	struct sock *sk;
1355 	int err;
1356 	long timeo;
1357 
1358 	skb = netlink_trim(skb, gfp_any());
1359 
1360 	timeo = sock_sndtimeo(ssk, nonblock);
1361 retry:
1362 	sk = netlink_getsockbyportid(ssk, portid);
1363 	if (IS_ERR(sk)) {
1364 		kfree_skb(skb);
1365 		return PTR_ERR(sk);
1366 	}
1367 	if (netlink_is_kernel(sk))
1368 		return netlink_unicast_kernel(sk, skb, ssk);
1369 
1370 	if (sk_filter(sk, skb)) {
1371 		err = skb->len;
1372 		kfree_skb(skb);
1373 		sock_put(sk);
1374 		return err;
1375 	}
1376 
1377 	err = netlink_attachskb(sk, skb, &timeo, ssk);
1378 	if (err == 1)
1379 		goto retry;
1380 	if (err)
1381 		return err;
1382 
1383 	return netlink_sendskb(sk, skb);
1384 }
1385 EXPORT_SYMBOL(netlink_unicast);
1386 
1387 int netlink_has_listeners(struct sock *sk, unsigned int group)
1388 {
1389 	int res = 0;
1390 	struct listeners *listeners;
1391 
1392 	BUG_ON(!netlink_is_kernel(sk));
1393 
1394 	rcu_read_lock();
1395 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1396 
1397 	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1398 		res = test_bit(group - 1, listeners->masks);
1399 
1400 	rcu_read_unlock();
1401 
1402 	return res;
1403 }
1404 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1405 
1406 bool netlink_strict_get_check(struct sk_buff *skb)
1407 {
1408 	return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1409 }
1410 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1411 
1412 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1413 {
1414 	struct netlink_sock *nlk = nlk_sk(sk);
1415 
1416 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1417 	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1418 		netlink_skb_set_owner_r(skb, sk);
1419 		__netlink_sendskb(sk, skb);
1420 		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1421 	}
1422 	return -1;
1423 }
1424 
1425 struct netlink_broadcast_data {
1426 	struct sock *exclude_sk;
1427 	struct net *net;
1428 	u32 portid;
1429 	u32 group;
1430 	int failure;
1431 	int delivery_failure;
1432 	int congested;
1433 	int delivered;
1434 	gfp_t allocation;
1435 	struct sk_buff *skb, *skb2;
1436 	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1437 	void *tx_data;
1438 };
1439 
1440 static void do_one_broadcast(struct sock *sk,
1441 				    struct netlink_broadcast_data *p)
1442 {
1443 	struct netlink_sock *nlk = nlk_sk(sk);
1444 	int val;
1445 
1446 	if (p->exclude_sk == sk)
1447 		return;
1448 
1449 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1450 	    !test_bit(p->group - 1, nlk->groups))
1451 		return;
1452 
1453 	if (!net_eq(sock_net(sk), p->net)) {
1454 		if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1455 			return;
1456 
1457 		if (!peernet_has_id(sock_net(sk), p->net))
1458 			return;
1459 
1460 		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1461 				     CAP_NET_BROADCAST))
1462 			return;
1463 	}
1464 
1465 	if (p->failure) {
1466 		netlink_overrun(sk);
1467 		return;
1468 	}
1469 
1470 	sock_hold(sk);
1471 	if (p->skb2 == NULL) {
1472 		if (skb_shared(p->skb)) {
1473 			p->skb2 = skb_clone(p->skb, p->allocation);
1474 		} else {
1475 			p->skb2 = skb_get(p->skb);
1476 			/*
1477 			 * skb ownership may have been set when
1478 			 * delivered to a previous socket.
1479 			 */
1480 			skb_orphan(p->skb2);
1481 		}
1482 	}
1483 	if (p->skb2 == NULL) {
1484 		netlink_overrun(sk);
1485 		/* Clone failed. Notify ALL listeners. */
1486 		p->failure = 1;
1487 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1488 			p->delivery_failure = 1;
1489 		goto out;
1490 	}
1491 
1492 	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1493 		kfree_skb(p->skb2);
1494 		p->skb2 = NULL;
1495 		goto out;
1496 	}
1497 
1498 	if (sk_filter(sk, p->skb2)) {
1499 		kfree_skb(p->skb2);
1500 		p->skb2 = NULL;
1501 		goto out;
1502 	}
1503 	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1504 	if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1505 		NETLINK_CB(p->skb2).nsid_is_set = true;
1506 	val = netlink_broadcast_deliver(sk, p->skb2);
1507 	if (val < 0) {
1508 		netlink_overrun(sk);
1509 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1510 			p->delivery_failure = 1;
1511 	} else {
1512 		p->congested |= val;
1513 		p->delivered = 1;
1514 		p->skb2 = NULL;
1515 	}
1516 out:
1517 	sock_put(sk);
1518 }
1519 
1520 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1521 			       u32 portid,
1522 			       u32 group, gfp_t allocation,
1523 			       int (*filter)(struct sock *dsk,
1524 					     struct sk_buff *skb, void *data),
1525 			       void *filter_data)
1526 {
1527 	struct net *net = sock_net(ssk);
1528 	struct netlink_broadcast_data info;
1529 	struct sock *sk;
1530 
1531 	skb = netlink_trim(skb, allocation);
1532 
1533 	info.exclude_sk = ssk;
1534 	info.net = net;
1535 	info.portid = portid;
1536 	info.group = group;
1537 	info.failure = 0;
1538 	info.delivery_failure = 0;
1539 	info.congested = 0;
1540 	info.delivered = 0;
1541 	info.allocation = allocation;
1542 	info.skb = skb;
1543 	info.skb2 = NULL;
1544 	info.tx_filter = filter;
1545 	info.tx_data = filter_data;
1546 
1547 	/* While we sleep in clone, do not allow to change socket list */
1548 
1549 	netlink_lock_table();
1550 
1551 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1552 		do_one_broadcast(sk, &info);
1553 
1554 	consume_skb(skb);
1555 
1556 	netlink_unlock_table();
1557 
1558 	if (info.delivery_failure) {
1559 		kfree_skb(info.skb2);
1560 		return -ENOBUFS;
1561 	}
1562 	consume_skb(info.skb2);
1563 
1564 	if (info.delivered) {
1565 		if (info.congested && gfpflags_allow_blocking(allocation))
1566 			yield();
1567 		return 0;
1568 	}
1569 	return -ESRCH;
1570 }
1571 EXPORT_SYMBOL(netlink_broadcast_filtered);
1572 
1573 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1574 		      u32 group, gfp_t allocation)
1575 {
1576 	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1577 					  NULL, NULL);
1578 }
1579 EXPORT_SYMBOL(netlink_broadcast);
1580 
1581 struct netlink_set_err_data {
1582 	struct sock *exclude_sk;
1583 	u32 portid;
1584 	u32 group;
1585 	int code;
1586 };
1587 
1588 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1589 {
1590 	struct netlink_sock *nlk = nlk_sk(sk);
1591 	int ret = 0;
1592 
1593 	if (sk == p->exclude_sk)
1594 		goto out;
1595 
1596 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1597 		goto out;
1598 
1599 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1600 	    !test_bit(p->group - 1, nlk->groups))
1601 		goto out;
1602 
1603 	if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1604 		ret = 1;
1605 		goto out;
1606 	}
1607 
1608 	WRITE_ONCE(sk->sk_err, p->code);
1609 	sk_error_report(sk);
1610 out:
1611 	return ret;
1612 }
1613 
1614 /**
1615  * netlink_set_err - report error to broadcast listeners
1616  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1617  * @portid: the PORTID of a process that we want to skip (if any)
1618  * @group: the broadcast group that will notice the error
1619  * @code: error code, must be negative (as usual in kernelspace)
1620  *
1621  * This function returns the number of broadcast listeners that have set the
1622  * NETLINK_NO_ENOBUFS socket option.
1623  */
1624 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1625 {
1626 	struct netlink_set_err_data info;
1627 	unsigned long flags;
1628 	struct sock *sk;
1629 	int ret = 0;
1630 
1631 	info.exclude_sk = ssk;
1632 	info.portid = portid;
1633 	info.group = group;
1634 	/* sk->sk_err wants a positive error value */
1635 	info.code = -code;
1636 
1637 	read_lock_irqsave(&nl_table_lock, flags);
1638 
1639 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1640 		ret += do_one_set_err(sk, &info);
1641 
1642 	read_unlock_irqrestore(&nl_table_lock, flags);
1643 	return ret;
1644 }
1645 EXPORT_SYMBOL(netlink_set_err);
1646 
1647 /* must be called with netlink table grabbed */
1648 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1649 				     unsigned int group,
1650 				     int is_new)
1651 {
1652 	int old, new = !!is_new, subscriptions;
1653 
1654 	old = test_bit(group - 1, nlk->groups);
1655 	subscriptions = nlk->subscriptions - old + new;
1656 	__assign_bit(group - 1, nlk->groups, new);
1657 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1658 	netlink_update_listeners(&nlk->sk);
1659 }
1660 
1661 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1662 			      sockptr_t optval, unsigned int optlen)
1663 {
1664 	struct sock *sk = sock->sk;
1665 	struct netlink_sock *nlk = nlk_sk(sk);
1666 	unsigned int val = 0;
1667 	int nr = -1;
1668 
1669 	if (level != SOL_NETLINK)
1670 		return -ENOPROTOOPT;
1671 
1672 	if (optlen >= sizeof(int) &&
1673 	    copy_from_sockptr(&val, optval, sizeof(val)))
1674 		return -EFAULT;
1675 
1676 	switch (optname) {
1677 	case NETLINK_PKTINFO:
1678 		nr = NETLINK_F_RECV_PKTINFO;
1679 		break;
1680 	case NETLINK_ADD_MEMBERSHIP:
1681 	case NETLINK_DROP_MEMBERSHIP: {
1682 		int err;
1683 
1684 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1685 			return -EPERM;
1686 		err = netlink_realloc_groups(sk);
1687 		if (err)
1688 			return err;
1689 		if (!val || val - 1 >= nlk->ngroups)
1690 			return -EINVAL;
1691 		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1692 			err = nlk->netlink_bind(sock_net(sk), val);
1693 			if (err)
1694 				return err;
1695 		}
1696 		netlink_table_grab();
1697 		netlink_update_socket_mc(nlk, val,
1698 					 optname == NETLINK_ADD_MEMBERSHIP);
1699 		netlink_table_ungrab();
1700 		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1701 			nlk->netlink_unbind(sock_net(sk), val);
1702 
1703 		break;
1704 	}
1705 	case NETLINK_BROADCAST_ERROR:
1706 		nr = NETLINK_F_BROADCAST_SEND_ERROR;
1707 		break;
1708 	case NETLINK_NO_ENOBUFS:
1709 		assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1710 		if (val) {
1711 			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1712 			wake_up_interruptible(&nlk->wait);
1713 		}
1714 		break;
1715 	case NETLINK_LISTEN_ALL_NSID:
1716 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1717 			return -EPERM;
1718 		nr = NETLINK_F_LISTEN_ALL_NSID;
1719 		break;
1720 	case NETLINK_CAP_ACK:
1721 		nr = NETLINK_F_CAP_ACK;
1722 		break;
1723 	case NETLINK_EXT_ACK:
1724 		nr = NETLINK_F_EXT_ACK;
1725 		break;
1726 	case NETLINK_GET_STRICT_CHK:
1727 		nr = NETLINK_F_STRICT_CHK;
1728 		break;
1729 	default:
1730 		return -ENOPROTOOPT;
1731 	}
1732 	if (nr >= 0)
1733 		assign_bit(nr, &nlk->flags, val);
1734 	return 0;
1735 }
1736 
1737 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1738 			      char __user *optval, int __user *optlen)
1739 {
1740 	struct sock *sk = sock->sk;
1741 	struct netlink_sock *nlk = nlk_sk(sk);
1742 	unsigned int flag;
1743 	int len, val;
1744 
1745 	if (level != SOL_NETLINK)
1746 		return -ENOPROTOOPT;
1747 
1748 	if (get_user(len, optlen))
1749 		return -EFAULT;
1750 	if (len < 0)
1751 		return -EINVAL;
1752 
1753 	switch (optname) {
1754 	case NETLINK_PKTINFO:
1755 		flag = NETLINK_F_RECV_PKTINFO;
1756 		break;
1757 	case NETLINK_BROADCAST_ERROR:
1758 		flag = NETLINK_F_BROADCAST_SEND_ERROR;
1759 		break;
1760 	case NETLINK_NO_ENOBUFS:
1761 		flag = NETLINK_F_RECV_NO_ENOBUFS;
1762 		break;
1763 	case NETLINK_LIST_MEMBERSHIPS: {
1764 		int pos, idx, shift, err = 0;
1765 
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(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1780 			err = -EFAULT;
1781 		netlink_unlock_table();
1782 		return err;
1783 	}
1784 	case NETLINK_CAP_ACK:
1785 		flag = NETLINK_F_CAP_ACK;
1786 		break;
1787 	case NETLINK_EXT_ACK:
1788 		flag = NETLINK_F_EXT_ACK;
1789 		break;
1790 	case NETLINK_GET_STRICT_CHK:
1791 		flag = NETLINK_F_STRICT_CHK;
1792 		break;
1793 	default:
1794 		return -ENOPROTOOPT;
1795 	}
1796 
1797 	if (len < sizeof(int))
1798 		return -EINVAL;
1799 
1800 	len = sizeof(int);
1801 	val = test_bit(flag, &nlk->flags);
1802 
1803 	if (put_user(len, optlen) ||
1804 	    copy_to_user(optval, &val, len))
1805 		return -EFAULT;
1806 
1807 	return 0;
1808 }
1809 
1810 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1811 {
1812 	struct nl_pktinfo info;
1813 
1814 	info.group = NETLINK_CB(skb).dst_group;
1815 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1816 }
1817 
1818 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1819 					 struct sk_buff *skb)
1820 {
1821 	if (!NETLINK_CB(skb).nsid_is_set)
1822 		return;
1823 
1824 	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1825 		 &NETLINK_CB(skb).nsid);
1826 }
1827 
1828 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1829 {
1830 	struct sock *sk = sock->sk;
1831 	struct netlink_sock *nlk = nlk_sk(sk);
1832 	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1833 	u32 dst_portid;
1834 	u32 dst_group;
1835 	struct sk_buff *skb;
1836 	int err;
1837 	struct scm_cookie scm;
1838 	u32 netlink_skb_flags = 0;
1839 
1840 	if (msg->msg_flags & MSG_OOB)
1841 		return -EOPNOTSUPP;
1842 
1843 	if (len == 0) {
1844 		pr_warn_once("Zero length message leads to an empty skb\n");
1845 		return -ENODATA;
1846 	}
1847 
1848 	err = scm_send(sock, msg, &scm, true);
1849 	if (err < 0)
1850 		return err;
1851 
1852 	if (msg->msg_namelen) {
1853 		err = -EINVAL;
1854 		if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1855 			goto out;
1856 		if (addr->nl_family != AF_NETLINK)
1857 			goto out;
1858 		dst_portid = addr->nl_pid;
1859 		dst_group = ffs(addr->nl_groups);
1860 		err =  -EPERM;
1861 		if ((dst_group || dst_portid) &&
1862 		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1863 			goto out;
1864 		netlink_skb_flags |= NETLINK_SKB_DST;
1865 	} else {
1866 		/* Paired with WRITE_ONCE() in netlink_connect() */
1867 		dst_portid = READ_ONCE(nlk->dst_portid);
1868 		dst_group = READ_ONCE(nlk->dst_group);
1869 	}
1870 
1871 	/* Paired with WRITE_ONCE() in netlink_insert() */
1872 	if (!READ_ONCE(nlk->bound)) {
1873 		err = netlink_autobind(sock);
1874 		if (err)
1875 			goto out;
1876 	} else {
1877 		/* Ensure nlk is hashed and visible. */
1878 		smp_rmb();
1879 	}
1880 
1881 	err = -EMSGSIZE;
1882 	if (len > sk->sk_sndbuf - 32)
1883 		goto out;
1884 	err = -ENOBUFS;
1885 	skb = netlink_alloc_large_skb(len, dst_group);
1886 	if (skb == NULL)
1887 		goto out;
1888 
1889 	NETLINK_CB(skb).portid	= nlk->portid;
1890 	NETLINK_CB(skb).dst_group = dst_group;
1891 	NETLINK_CB(skb).creds	= scm.creds;
1892 	NETLINK_CB(skb).flags	= netlink_skb_flags;
1893 
1894 	err = -EFAULT;
1895 	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1896 		kfree_skb(skb);
1897 		goto out;
1898 	}
1899 
1900 	err = security_netlink_send(sk, skb);
1901 	if (err) {
1902 		kfree_skb(skb);
1903 		goto out;
1904 	}
1905 
1906 	if (dst_group) {
1907 		refcount_inc(&skb->users);
1908 		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1909 	}
1910 	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1911 
1912 out:
1913 	scm_destroy(&scm);
1914 	return err;
1915 }
1916 
1917 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1918 			   int flags)
1919 {
1920 	struct scm_cookie scm;
1921 	struct sock *sk = sock->sk;
1922 	struct netlink_sock *nlk = nlk_sk(sk);
1923 	size_t copied, max_recvmsg_len;
1924 	struct sk_buff *skb, *data_skb;
1925 	int err, ret;
1926 
1927 	if (flags & MSG_OOB)
1928 		return -EOPNOTSUPP;
1929 
1930 	copied = 0;
1931 
1932 	skb = skb_recv_datagram(sk, flags, &err);
1933 	if (skb == NULL)
1934 		goto out;
1935 
1936 	data_skb = skb;
1937 
1938 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1939 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1940 		/*
1941 		 * If this skb has a frag_list, then here that means that we
1942 		 * will have to use the frag_list skb's data for compat tasks
1943 		 * and the regular skb's data for normal (non-compat) tasks.
1944 		 *
1945 		 * If we need to send the compat skb, assign it to the
1946 		 * 'data_skb' variable so that it will be used below for data
1947 		 * copying. We keep 'skb' for everything else, including
1948 		 * freeing both later.
1949 		 */
1950 		if (flags & MSG_CMSG_COMPAT)
1951 			data_skb = skb_shinfo(skb)->frag_list;
1952 	}
1953 #endif
1954 
1955 	/* Record the max length of recvmsg() calls for future allocations */
1956 	max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1957 	max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1958 				SKB_WITH_OVERHEAD(32768));
1959 	WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1960 
1961 	copied = data_skb->len;
1962 	if (len < copied) {
1963 		msg->msg_flags |= MSG_TRUNC;
1964 		copied = len;
1965 	}
1966 
1967 	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1968 
1969 	if (msg->msg_name) {
1970 		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1971 		addr->nl_family = AF_NETLINK;
1972 		addr->nl_pad    = 0;
1973 		addr->nl_pid	= NETLINK_CB(skb).portid;
1974 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1975 		msg->msg_namelen = sizeof(*addr);
1976 	}
1977 
1978 	if (nlk_test_bit(RECV_PKTINFO, sk))
1979 		netlink_cmsg_recv_pktinfo(msg, skb);
1980 	if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1981 		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1982 
1983 	memset(&scm, 0, sizeof(scm));
1984 	scm.creds = *NETLINK_CREDS(skb);
1985 	if (flags & MSG_TRUNC)
1986 		copied = data_skb->len;
1987 
1988 	skb_free_datagram(sk, skb);
1989 
1990 	if (READ_ONCE(nlk->cb_running) &&
1991 	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1992 		ret = netlink_dump(sk);
1993 		if (ret) {
1994 			WRITE_ONCE(sk->sk_err, -ret);
1995 			sk_error_report(sk);
1996 		}
1997 	}
1998 
1999 	scm_recv(sock, msg, &scm, flags);
2000 out:
2001 	netlink_rcv_wake(sk);
2002 	return err ? : copied;
2003 }
2004 
2005 static void netlink_data_ready(struct sock *sk)
2006 {
2007 	BUG();
2008 }
2009 
2010 /*
2011  *	We export these functions to other modules. They provide a
2012  *	complete set of kernel non-blocking support for message
2013  *	queueing.
2014  */
2015 
2016 struct sock *
2017 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2018 			struct netlink_kernel_cfg *cfg)
2019 {
2020 	struct socket *sock;
2021 	struct sock *sk;
2022 	struct netlink_sock *nlk;
2023 	struct listeners *listeners = NULL;
2024 	struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2025 	unsigned int groups;
2026 
2027 	BUG_ON(!nl_table);
2028 
2029 	if (unit < 0 || unit >= MAX_LINKS)
2030 		return NULL;
2031 
2032 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2033 		return NULL;
2034 
2035 	if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2036 		goto out_sock_release_nosk;
2037 
2038 	sk = sock->sk;
2039 
2040 	if (!cfg || cfg->groups < 32)
2041 		groups = 32;
2042 	else
2043 		groups = cfg->groups;
2044 
2045 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2046 	if (!listeners)
2047 		goto out_sock_release;
2048 
2049 	sk->sk_data_ready = netlink_data_ready;
2050 	if (cfg && cfg->input)
2051 		nlk_sk(sk)->netlink_rcv = cfg->input;
2052 
2053 	if (netlink_insert(sk, 0))
2054 		goto out_sock_release;
2055 
2056 	nlk = nlk_sk(sk);
2057 	set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2058 
2059 	netlink_table_grab();
2060 	if (!nl_table[unit].registered) {
2061 		nl_table[unit].groups = groups;
2062 		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2063 		nl_table[unit].cb_mutex = cb_mutex;
2064 		nl_table[unit].module = module;
2065 		if (cfg) {
2066 			nl_table[unit].bind = cfg->bind;
2067 			nl_table[unit].unbind = cfg->unbind;
2068 			nl_table[unit].release = cfg->release;
2069 			nl_table[unit].flags = cfg->flags;
2070 		}
2071 		nl_table[unit].registered = 1;
2072 	} else {
2073 		kfree(listeners);
2074 		nl_table[unit].registered++;
2075 	}
2076 	netlink_table_ungrab();
2077 	return sk;
2078 
2079 out_sock_release:
2080 	kfree(listeners);
2081 	netlink_kernel_release(sk);
2082 	return NULL;
2083 
2084 out_sock_release_nosk:
2085 	sock_release(sock);
2086 	return NULL;
2087 }
2088 EXPORT_SYMBOL(__netlink_kernel_create);
2089 
2090 void
2091 netlink_kernel_release(struct sock *sk)
2092 {
2093 	if (sk == NULL || sk->sk_socket == NULL)
2094 		return;
2095 
2096 	sock_release(sk->sk_socket);
2097 }
2098 EXPORT_SYMBOL(netlink_kernel_release);
2099 
2100 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2101 {
2102 	struct listeners *new, *old;
2103 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2104 
2105 	if (groups < 32)
2106 		groups = 32;
2107 
2108 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2109 		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2110 		if (!new)
2111 			return -ENOMEM;
2112 		old = nl_deref_protected(tbl->listeners);
2113 		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2114 		rcu_assign_pointer(tbl->listeners, new);
2115 
2116 		kfree_rcu(old, rcu);
2117 	}
2118 	tbl->groups = groups;
2119 
2120 	return 0;
2121 }
2122 
2123 /**
2124  * netlink_change_ngroups - change number of multicast groups
2125  *
2126  * This changes the number of multicast groups that are available
2127  * on a certain netlink family. Note that it is not possible to
2128  * change the number of groups to below 32. Also note that it does
2129  * not implicitly call netlink_clear_multicast_users() when the
2130  * number of groups is reduced.
2131  *
2132  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2133  * @groups: The new number of groups.
2134  */
2135 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2136 {
2137 	int err;
2138 
2139 	netlink_table_grab();
2140 	err = __netlink_change_ngroups(sk, groups);
2141 	netlink_table_ungrab();
2142 
2143 	return err;
2144 }
2145 
2146 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2147 {
2148 	struct sock *sk;
2149 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2150 
2151 	sk_for_each_bound(sk, &tbl->mc_list)
2152 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2153 }
2154 
2155 struct nlmsghdr *
2156 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2157 {
2158 	struct nlmsghdr *nlh;
2159 	int size = nlmsg_msg_size(len);
2160 
2161 	nlh = skb_put(skb, NLMSG_ALIGN(size));
2162 	nlh->nlmsg_type = type;
2163 	nlh->nlmsg_len = size;
2164 	nlh->nlmsg_flags = flags;
2165 	nlh->nlmsg_pid = portid;
2166 	nlh->nlmsg_seq = seq;
2167 	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2168 		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2169 	return nlh;
2170 }
2171 EXPORT_SYMBOL(__nlmsg_put);
2172 
2173 /*
2174  * It looks a bit ugly.
2175  * It would be better to create kernel thread.
2176  */
2177 
2178 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2179 			     struct netlink_callback *cb,
2180 			     struct netlink_ext_ack *extack)
2181 {
2182 	struct nlmsghdr *nlh;
2183 
2184 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2185 			       NLM_F_MULTI | cb->answer_flags);
2186 	if (WARN_ON(!nlh))
2187 		return -ENOBUFS;
2188 
2189 	nl_dump_check_consistent(cb, nlh);
2190 	memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2191 
2192 	if (extack->_msg && test_bit(NETLINK_F_EXT_ACK, &nlk->flags)) {
2193 		nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2194 		if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2195 			nlmsg_end(skb, nlh);
2196 	}
2197 
2198 	return 0;
2199 }
2200 
2201 static int netlink_dump(struct sock *sk)
2202 {
2203 	struct netlink_sock *nlk = nlk_sk(sk);
2204 	struct netlink_ext_ack extack = {};
2205 	struct netlink_callback *cb;
2206 	struct sk_buff *skb = NULL;
2207 	size_t max_recvmsg_len;
2208 	struct module *module;
2209 	int err = -ENOBUFS;
2210 	int alloc_min_size;
2211 	int alloc_size;
2212 
2213 	mutex_lock(nlk->cb_mutex);
2214 	if (!nlk->cb_running) {
2215 		err = -EINVAL;
2216 		goto errout_skb;
2217 	}
2218 
2219 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2220 		goto errout_skb;
2221 
2222 	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2223 	 * required, but it makes sense to _attempt_ a 16K bytes allocation
2224 	 * to reduce number of system calls on dump operations, if user
2225 	 * ever provided a big enough buffer.
2226 	 */
2227 	cb = &nlk->cb;
2228 	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2229 
2230 	max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2231 	if (alloc_min_size < max_recvmsg_len) {
2232 		alloc_size = max_recvmsg_len;
2233 		skb = alloc_skb(alloc_size,
2234 				(GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2235 				__GFP_NOWARN | __GFP_NORETRY);
2236 	}
2237 	if (!skb) {
2238 		alloc_size = alloc_min_size;
2239 		skb = alloc_skb(alloc_size, GFP_KERNEL);
2240 	}
2241 	if (!skb)
2242 		goto errout_skb;
2243 
2244 	/* Trim skb to allocated size. User is expected to provide buffer as
2245 	 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2246 	 * netlink_recvmsg())). dump will pack as many smaller messages as
2247 	 * could fit within the allocated skb. skb is typically allocated
2248 	 * with larger space than required (could be as much as near 2x the
2249 	 * requested size with align to next power of 2 approach). Allowing
2250 	 * dump to use the excess space makes it difficult for a user to have a
2251 	 * reasonable static buffer based on the expected largest dump of a
2252 	 * single netdev. The outcome is MSG_TRUNC error.
2253 	 */
2254 	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2255 
2256 	/* Make sure malicious BPF programs can not read unitialized memory
2257 	 * from skb->head -> skb->data
2258 	 */
2259 	skb_reset_network_header(skb);
2260 	skb_reset_mac_header(skb);
2261 
2262 	netlink_skb_set_owner_r(skb, sk);
2263 
2264 	if (nlk->dump_done_errno > 0) {
2265 		cb->extack = &extack;
2266 		nlk->dump_done_errno = cb->dump(skb, cb);
2267 		cb->extack = NULL;
2268 	}
2269 
2270 	if (nlk->dump_done_errno > 0 ||
2271 	    skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2272 		mutex_unlock(nlk->cb_mutex);
2273 
2274 		if (sk_filter(sk, skb))
2275 			kfree_skb(skb);
2276 		else
2277 			__netlink_sendskb(sk, skb);
2278 		return 0;
2279 	}
2280 
2281 	if (netlink_dump_done(nlk, skb, cb, &extack))
2282 		goto errout_skb;
2283 
2284 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2285 	/* frag_list skb's data is used for compat tasks
2286 	 * and the regular skb's data for normal (non-compat) tasks.
2287 	 * See netlink_recvmsg().
2288 	 */
2289 	if (unlikely(skb_shinfo(skb)->frag_list)) {
2290 		if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2291 			goto errout_skb;
2292 	}
2293 #endif
2294 
2295 	if (sk_filter(sk, skb))
2296 		kfree_skb(skb);
2297 	else
2298 		__netlink_sendskb(sk, skb);
2299 
2300 	if (cb->done)
2301 		cb->done(cb);
2302 
2303 	WRITE_ONCE(nlk->cb_running, false);
2304 	module = cb->module;
2305 	skb = cb->skb;
2306 	mutex_unlock(nlk->cb_mutex);
2307 	module_put(module);
2308 	consume_skb(skb);
2309 	return 0;
2310 
2311 errout_skb:
2312 	mutex_unlock(nlk->cb_mutex);
2313 	kfree_skb(skb);
2314 	return err;
2315 }
2316 
2317 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2318 			 const struct nlmsghdr *nlh,
2319 			 struct netlink_dump_control *control)
2320 {
2321 	struct netlink_callback *cb;
2322 	struct netlink_sock *nlk;
2323 	struct sock *sk;
2324 	int ret;
2325 
2326 	refcount_inc(&skb->users);
2327 
2328 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2329 	if (sk == NULL) {
2330 		ret = -ECONNREFUSED;
2331 		goto error_free;
2332 	}
2333 
2334 	nlk = nlk_sk(sk);
2335 	mutex_lock(nlk->cb_mutex);
2336 	/* A dump is in progress... */
2337 	if (nlk->cb_running) {
2338 		ret = -EBUSY;
2339 		goto error_unlock;
2340 	}
2341 	/* add reference of module which cb->dump belongs to */
2342 	if (!try_module_get(control->module)) {
2343 		ret = -EPROTONOSUPPORT;
2344 		goto error_unlock;
2345 	}
2346 
2347 	cb = &nlk->cb;
2348 	memset(cb, 0, sizeof(*cb));
2349 	cb->dump = control->dump;
2350 	cb->done = control->done;
2351 	cb->nlh = nlh;
2352 	cb->data = control->data;
2353 	cb->module = control->module;
2354 	cb->min_dump_alloc = control->min_dump_alloc;
2355 	cb->skb = skb;
2356 
2357 	cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2358 
2359 	if (control->start) {
2360 		cb->extack = control->extack;
2361 		ret = control->start(cb);
2362 		cb->extack = NULL;
2363 		if (ret)
2364 			goto error_put;
2365 	}
2366 
2367 	WRITE_ONCE(nlk->cb_running, true);
2368 	nlk->dump_done_errno = INT_MAX;
2369 
2370 	mutex_unlock(nlk->cb_mutex);
2371 
2372 	ret = netlink_dump(sk);
2373 
2374 	sock_put(sk);
2375 
2376 	if (ret)
2377 		return ret;
2378 
2379 	/* We successfully started a dump, by returning -EINTR we
2380 	 * signal not to send ACK even if it was requested.
2381 	 */
2382 	return -EINTR;
2383 
2384 error_put:
2385 	module_put(control->module);
2386 error_unlock:
2387 	sock_put(sk);
2388 	mutex_unlock(nlk->cb_mutex);
2389 error_free:
2390 	kfree_skb(skb);
2391 	return ret;
2392 }
2393 EXPORT_SYMBOL(__netlink_dump_start);
2394 
2395 static size_t
2396 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2397 		    const struct netlink_ext_ack *extack)
2398 {
2399 	size_t tlvlen;
2400 
2401 	if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2402 		return 0;
2403 
2404 	tlvlen = 0;
2405 	if (extack->_msg)
2406 		tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2407 	if (extack->cookie_len)
2408 		tlvlen += nla_total_size(extack->cookie_len);
2409 
2410 	/* Following attributes are only reported as error (not warning) */
2411 	if (!err)
2412 		return tlvlen;
2413 
2414 	if (extack->bad_attr)
2415 		tlvlen += nla_total_size(sizeof(u32));
2416 	if (extack->policy)
2417 		tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2418 	if (extack->miss_type)
2419 		tlvlen += nla_total_size(sizeof(u32));
2420 	if (extack->miss_nest)
2421 		tlvlen += nla_total_size(sizeof(u32));
2422 
2423 	return tlvlen;
2424 }
2425 
2426 static void
2427 netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2428 		     struct nlmsghdr *nlh, int err,
2429 		     const struct netlink_ext_ack *extack)
2430 {
2431 	if (extack->_msg)
2432 		WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2433 	if (extack->cookie_len)
2434 		WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2435 				extack->cookie_len, extack->cookie));
2436 
2437 	if (!err)
2438 		return;
2439 
2440 	if (extack->bad_attr &&
2441 	    !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2442 		     (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2443 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2444 				    (u8 *)extack->bad_attr - (u8 *)nlh));
2445 	if (extack->policy)
2446 		netlink_policy_dump_write_attr(skb, extack->policy,
2447 					       NLMSGERR_ATTR_POLICY);
2448 	if (extack->miss_type)
2449 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2450 				    extack->miss_type));
2451 	if (extack->miss_nest &&
2452 	    !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2453 		     (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2454 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2455 				    (u8 *)extack->miss_nest - (u8 *)nlh));
2456 }
2457 
2458 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2459 		 const struct netlink_ext_ack *extack)
2460 {
2461 	struct sk_buff *skb;
2462 	struct nlmsghdr *rep;
2463 	struct nlmsgerr *errmsg;
2464 	size_t payload = sizeof(*errmsg);
2465 	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2466 	unsigned int flags = 0;
2467 	size_t tlvlen;
2468 
2469 	/* Error messages get the original request appened, unless the user
2470 	 * requests to cap the error message, and get extra error data if
2471 	 * requested.
2472 	 */
2473 	if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2474 		payload += nlmsg_len(nlh);
2475 	else
2476 		flags |= NLM_F_CAPPED;
2477 
2478 	tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2479 	if (tlvlen)
2480 		flags |= NLM_F_ACK_TLVS;
2481 
2482 	skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2483 	if (!skb)
2484 		goto err_skb;
2485 
2486 	rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2487 			NLMSG_ERROR, sizeof(*errmsg), flags);
2488 	if (!rep)
2489 		goto err_bad_put;
2490 	errmsg = nlmsg_data(rep);
2491 	errmsg->error = err;
2492 	errmsg->msg = *nlh;
2493 
2494 	if (!(flags & NLM_F_CAPPED)) {
2495 		if (!nlmsg_append(skb, nlmsg_len(nlh)))
2496 			goto err_bad_put;
2497 
2498 		memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2499 		       nlmsg_len(nlh));
2500 	}
2501 
2502 	if (tlvlen)
2503 		netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2504 
2505 	nlmsg_end(skb, rep);
2506 
2507 	nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2508 
2509 	return;
2510 
2511 err_bad_put:
2512 	nlmsg_free(skb);
2513 err_skb:
2514 	WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2515 	sk_error_report(NETLINK_CB(in_skb).sk);
2516 }
2517 EXPORT_SYMBOL(netlink_ack);
2518 
2519 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2520 						   struct nlmsghdr *,
2521 						   struct netlink_ext_ack *))
2522 {
2523 	struct netlink_ext_ack extack;
2524 	struct nlmsghdr *nlh;
2525 	int err;
2526 
2527 	while (skb->len >= nlmsg_total_size(0)) {
2528 		int msglen;
2529 
2530 		memset(&extack, 0, sizeof(extack));
2531 		nlh = nlmsg_hdr(skb);
2532 		err = 0;
2533 
2534 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2535 			return 0;
2536 
2537 		/* Only requests are handled by the kernel */
2538 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2539 			goto ack;
2540 
2541 		/* Skip control messages */
2542 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2543 			goto ack;
2544 
2545 		err = cb(skb, nlh, &extack);
2546 		if (err == -EINTR)
2547 			goto skip;
2548 
2549 ack:
2550 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2551 			netlink_ack(skb, nlh, err, &extack);
2552 
2553 skip:
2554 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2555 		if (msglen > skb->len)
2556 			msglen = skb->len;
2557 		skb_pull(skb, msglen);
2558 	}
2559 
2560 	return 0;
2561 }
2562 EXPORT_SYMBOL(netlink_rcv_skb);
2563 
2564 /**
2565  * nlmsg_notify - send a notification netlink message
2566  * @sk: netlink socket to use
2567  * @skb: notification message
2568  * @portid: destination netlink portid for reports or 0
2569  * @group: destination multicast group or 0
2570  * @report: 1 to report back, 0 to disable
2571  * @flags: allocation flags
2572  */
2573 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2574 		 unsigned int group, int report, gfp_t flags)
2575 {
2576 	int err = 0;
2577 
2578 	if (group) {
2579 		int exclude_portid = 0;
2580 
2581 		if (report) {
2582 			refcount_inc(&skb->users);
2583 			exclude_portid = portid;
2584 		}
2585 
2586 		/* errors reported via destination sk->sk_err, but propagate
2587 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2588 		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2589 		if (err == -ESRCH)
2590 			err = 0;
2591 	}
2592 
2593 	if (report) {
2594 		int err2;
2595 
2596 		err2 = nlmsg_unicast(sk, skb, portid);
2597 		if (!err)
2598 			err = err2;
2599 	}
2600 
2601 	return err;
2602 }
2603 EXPORT_SYMBOL(nlmsg_notify);
2604 
2605 #ifdef CONFIG_PROC_FS
2606 struct nl_seq_iter {
2607 	struct seq_net_private p;
2608 	struct rhashtable_iter hti;
2609 	int link;
2610 };
2611 
2612 static void netlink_walk_start(struct nl_seq_iter *iter)
2613 {
2614 	rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2615 	rhashtable_walk_start(&iter->hti);
2616 }
2617 
2618 static void netlink_walk_stop(struct nl_seq_iter *iter)
2619 {
2620 	rhashtable_walk_stop(&iter->hti);
2621 	rhashtable_walk_exit(&iter->hti);
2622 }
2623 
2624 static void *__netlink_seq_next(struct seq_file *seq)
2625 {
2626 	struct nl_seq_iter *iter = seq->private;
2627 	struct netlink_sock *nlk;
2628 
2629 	do {
2630 		for (;;) {
2631 			nlk = rhashtable_walk_next(&iter->hti);
2632 
2633 			if (IS_ERR(nlk)) {
2634 				if (PTR_ERR(nlk) == -EAGAIN)
2635 					continue;
2636 
2637 				return nlk;
2638 			}
2639 
2640 			if (nlk)
2641 				break;
2642 
2643 			netlink_walk_stop(iter);
2644 			if (++iter->link >= MAX_LINKS)
2645 				return NULL;
2646 
2647 			netlink_walk_start(iter);
2648 		}
2649 	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2650 
2651 	return nlk;
2652 }
2653 
2654 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2655 	__acquires(RCU)
2656 {
2657 	struct nl_seq_iter *iter = seq->private;
2658 	void *obj = SEQ_START_TOKEN;
2659 	loff_t pos;
2660 
2661 	iter->link = 0;
2662 
2663 	netlink_walk_start(iter);
2664 
2665 	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2666 		obj = __netlink_seq_next(seq);
2667 
2668 	return obj;
2669 }
2670 
2671 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2672 {
2673 	++*pos;
2674 	return __netlink_seq_next(seq);
2675 }
2676 
2677 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2678 {
2679 	struct nl_seq_iter *iter = seq->private;
2680 
2681 	if (iter->link >= MAX_LINKS)
2682 		return;
2683 
2684 	netlink_walk_stop(iter);
2685 }
2686 
2687 
2688 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2689 {
2690 	if (v == SEQ_START_TOKEN) {
2691 		seq_puts(seq,
2692 			 "sk               Eth Pid        Groups   "
2693 			 "Rmem     Wmem     Dump  Locks    Drops    Inode\n");
2694 	} else {
2695 		struct sock *s = v;
2696 		struct netlink_sock *nlk = nlk_sk(s);
2697 
2698 		seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2699 			   s,
2700 			   s->sk_protocol,
2701 			   nlk->portid,
2702 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2703 			   sk_rmem_alloc_get(s),
2704 			   sk_wmem_alloc_get(s),
2705 			   READ_ONCE(nlk->cb_running),
2706 			   refcount_read(&s->sk_refcnt),
2707 			   atomic_read(&s->sk_drops),
2708 			   sock_i_ino(s)
2709 			);
2710 
2711 	}
2712 	return 0;
2713 }
2714 
2715 #ifdef CONFIG_BPF_SYSCALL
2716 struct bpf_iter__netlink {
2717 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2718 	__bpf_md_ptr(struct netlink_sock *, sk);
2719 };
2720 
2721 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2722 
2723 static int netlink_prog_seq_show(struct bpf_prog *prog,
2724 				  struct bpf_iter_meta *meta,
2725 				  void *v)
2726 {
2727 	struct bpf_iter__netlink ctx;
2728 
2729 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2730 	ctx.meta = meta;
2731 	ctx.sk = nlk_sk((struct sock *)v);
2732 	return bpf_iter_run_prog(prog, &ctx);
2733 }
2734 
2735 static int netlink_seq_show(struct seq_file *seq, void *v)
2736 {
2737 	struct bpf_iter_meta meta;
2738 	struct bpf_prog *prog;
2739 
2740 	meta.seq = seq;
2741 	prog = bpf_iter_get_info(&meta, false);
2742 	if (!prog)
2743 		return netlink_native_seq_show(seq, v);
2744 
2745 	if (v != SEQ_START_TOKEN)
2746 		return netlink_prog_seq_show(prog, &meta, v);
2747 
2748 	return 0;
2749 }
2750 
2751 static void netlink_seq_stop(struct seq_file *seq, void *v)
2752 {
2753 	struct bpf_iter_meta meta;
2754 	struct bpf_prog *prog;
2755 
2756 	if (!v) {
2757 		meta.seq = seq;
2758 		prog = bpf_iter_get_info(&meta, true);
2759 		if (prog)
2760 			(void)netlink_prog_seq_show(prog, &meta, v);
2761 	}
2762 
2763 	netlink_native_seq_stop(seq, v);
2764 }
2765 #else
2766 static int netlink_seq_show(struct seq_file *seq, void *v)
2767 {
2768 	return netlink_native_seq_show(seq, v);
2769 }
2770 
2771 static void netlink_seq_stop(struct seq_file *seq, void *v)
2772 {
2773 	netlink_native_seq_stop(seq, v);
2774 }
2775 #endif
2776 
2777 static const struct seq_operations netlink_seq_ops = {
2778 	.start  = netlink_seq_start,
2779 	.next   = netlink_seq_next,
2780 	.stop   = netlink_seq_stop,
2781 	.show   = netlink_seq_show,
2782 };
2783 #endif
2784 
2785 int netlink_register_notifier(struct notifier_block *nb)
2786 {
2787 	return blocking_notifier_chain_register(&netlink_chain, nb);
2788 }
2789 EXPORT_SYMBOL(netlink_register_notifier);
2790 
2791 int netlink_unregister_notifier(struct notifier_block *nb)
2792 {
2793 	return blocking_notifier_chain_unregister(&netlink_chain, nb);
2794 }
2795 EXPORT_SYMBOL(netlink_unregister_notifier);
2796 
2797 static const struct proto_ops netlink_ops = {
2798 	.family =	PF_NETLINK,
2799 	.owner =	THIS_MODULE,
2800 	.release =	netlink_release,
2801 	.bind =		netlink_bind,
2802 	.connect =	netlink_connect,
2803 	.socketpair =	sock_no_socketpair,
2804 	.accept =	sock_no_accept,
2805 	.getname =	netlink_getname,
2806 	.poll =		datagram_poll,
2807 	.ioctl =	netlink_ioctl,
2808 	.listen =	sock_no_listen,
2809 	.shutdown =	sock_no_shutdown,
2810 	.setsockopt =	netlink_setsockopt,
2811 	.getsockopt =	netlink_getsockopt,
2812 	.sendmsg =	netlink_sendmsg,
2813 	.recvmsg =	netlink_recvmsg,
2814 	.mmap =		sock_no_mmap,
2815 };
2816 
2817 static const struct net_proto_family netlink_family_ops = {
2818 	.family = PF_NETLINK,
2819 	.create = netlink_create,
2820 	.owner	= THIS_MODULE,	/* for consistency 8) */
2821 };
2822 
2823 static int __net_init netlink_net_init(struct net *net)
2824 {
2825 #ifdef CONFIG_PROC_FS
2826 	if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2827 			sizeof(struct nl_seq_iter)))
2828 		return -ENOMEM;
2829 #endif
2830 	return 0;
2831 }
2832 
2833 static void __net_exit netlink_net_exit(struct net *net)
2834 {
2835 #ifdef CONFIG_PROC_FS
2836 	remove_proc_entry("netlink", net->proc_net);
2837 #endif
2838 }
2839 
2840 static void __init netlink_add_usersock_entry(void)
2841 {
2842 	struct listeners *listeners;
2843 	int groups = 32;
2844 
2845 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2846 	if (!listeners)
2847 		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2848 
2849 	netlink_table_grab();
2850 
2851 	nl_table[NETLINK_USERSOCK].groups = groups;
2852 	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2853 	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2854 	nl_table[NETLINK_USERSOCK].registered = 1;
2855 	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2856 
2857 	netlink_table_ungrab();
2858 }
2859 
2860 static struct pernet_operations __net_initdata netlink_net_ops = {
2861 	.init = netlink_net_init,
2862 	.exit = netlink_net_exit,
2863 };
2864 
2865 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2866 {
2867 	const struct netlink_sock *nlk = data;
2868 	struct netlink_compare_arg arg;
2869 
2870 	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2871 	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2872 }
2873 
2874 static const struct rhashtable_params netlink_rhashtable_params = {
2875 	.head_offset = offsetof(struct netlink_sock, node),
2876 	.key_len = netlink_compare_arg_len,
2877 	.obj_hashfn = netlink_hash,
2878 	.obj_cmpfn = netlink_compare,
2879 	.automatic_shrinking = true,
2880 };
2881 
2882 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2883 BTF_ID_LIST(btf_netlink_sock_id)
2884 BTF_ID(struct, netlink_sock)
2885 
2886 static const struct bpf_iter_seq_info netlink_seq_info = {
2887 	.seq_ops		= &netlink_seq_ops,
2888 	.init_seq_private	= bpf_iter_init_seq_net,
2889 	.fini_seq_private	= bpf_iter_fini_seq_net,
2890 	.seq_priv_size		= sizeof(struct nl_seq_iter),
2891 };
2892 
2893 static struct bpf_iter_reg netlink_reg_info = {
2894 	.target			= "netlink",
2895 	.ctx_arg_info_size	= 1,
2896 	.ctx_arg_info		= {
2897 		{ offsetof(struct bpf_iter__netlink, sk),
2898 		  PTR_TO_BTF_ID_OR_NULL },
2899 	},
2900 	.seq_info		= &netlink_seq_info,
2901 };
2902 
2903 static int __init bpf_iter_register(void)
2904 {
2905 	netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2906 	return bpf_iter_reg_target(&netlink_reg_info);
2907 }
2908 #endif
2909 
2910 static int __init netlink_proto_init(void)
2911 {
2912 	int i;
2913 	int err = proto_register(&netlink_proto, 0);
2914 
2915 	if (err != 0)
2916 		goto out;
2917 
2918 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2919 	err = bpf_iter_register();
2920 	if (err)
2921 		goto out;
2922 #endif
2923 
2924 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2925 
2926 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2927 	if (!nl_table)
2928 		goto panic;
2929 
2930 	for (i = 0; i < MAX_LINKS; i++) {
2931 		if (rhashtable_init(&nl_table[i].hash,
2932 				    &netlink_rhashtable_params) < 0) {
2933 			while (--i > 0)
2934 				rhashtable_destroy(&nl_table[i].hash);
2935 			kfree(nl_table);
2936 			goto panic;
2937 		}
2938 	}
2939 
2940 	netlink_add_usersock_entry();
2941 
2942 	sock_register(&netlink_family_ops);
2943 	register_pernet_subsys(&netlink_net_ops);
2944 	register_pernet_subsys(&netlink_tap_net_ops);
2945 	/* The netlink device handler may be needed early. */
2946 	rtnetlink_init();
2947 out:
2948 	return err;
2949 panic:
2950 	panic("netlink_init: Cannot allocate nl_table\n");
2951 }
2952 
2953 core_initcall(netlink_proto_init);
2954