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