xref: /linux/net/netlink/af_netlink.c (revision 1a9239bb4253f9076b5b4b2a1a4e8d7defd77a95)
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/net_namespace.h>
63 #include <linux/nospec.h>
64 #include <linux/btf_ids.h>
65 
66 #include <net/net_namespace.h>
67 #include <net/netns/generic.h>
68 #include <net/sock.h>
69 #include <net/scm.h>
70 #include <net/netlink.h>
71 #define CREATE_TRACE_POINTS
72 #include <trace/events/netlink.h>
73 
74 #include "af_netlink.h"
75 #include "genetlink.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 
netlink_is_kernel(struct sock * sk)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, bool lock_taken);
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 
do_trace_netlink_extack(const char * msg)154 void do_trace_netlink_extack(const char *msg)
155 {
156 	trace_netlink_extack(msg);
157 }
158 EXPORT_SYMBOL(do_trace_netlink_extack);
159 
netlink_group_mask(u32 group)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 
netlink_to_full_skb(const struct sk_buff * skb,gfp_t gfp_mask)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->len;
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 
netlink_add_tap(struct netlink_tap * nt)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 
__netlink_remove_tap(struct netlink_tap * nt)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 
netlink_remove_tap(struct netlink_tap * nt)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 
netlink_tap_init_net(struct net * net)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 
netlink_filter_tap(const struct sk_buff * skb)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 
__netlink_deliver_tap_skb(struct sk_buff * skb,struct net_device * dev)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 
__netlink_deliver_tap(struct sk_buff * skb,struct netlink_tap_net * nn)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 
netlink_deliver_tap(struct net * net,struct sk_buff * skb)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 
netlink_deliver_tap_kernel(struct sock * dst,struct sock * src,struct sk_buff * skb)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 
netlink_overrun(struct sock * sk)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 
netlink_rcv_wake(struct sock * sk)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 
netlink_skb_destructor(struct sk_buff * skb)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_atomic(skb->head);
378 
379 		skb->head = NULL;
380 	}
381 	if (skb->sk != NULL)
382 		sock_rfree(skb);
383 }
384 
netlink_skb_set_owner_r(struct sk_buff * skb,struct sock * sk)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 
netlink_sock_destruct(struct sock * sk)394 static void netlink_sock_destruct(struct sock *sk)
395 {
396 	skb_queue_purge(&sk->sk_receive_queue);
397 
398 	if (!sock_flag(sk, SOCK_DEAD)) {
399 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
400 		return;
401 	}
402 
403 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
404 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
405 	WARN_ON(nlk_sk(sk)->groups);
406 }
407 
408 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
409  * SMP. Look, when several writers sleep and reader wakes them up, all but one
410  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
411  * this, _but_ remember, it adds useless work on UP machines.
412  */
413 
netlink_table_grab(void)414 void netlink_table_grab(void)
415 	__acquires(nl_table_lock)
416 {
417 	might_sleep();
418 
419 	write_lock_irq(&nl_table_lock);
420 
421 	if (atomic_read(&nl_table_users)) {
422 		DECLARE_WAITQUEUE(wait, current);
423 
424 		add_wait_queue_exclusive(&nl_table_wait, &wait);
425 		for (;;) {
426 			set_current_state(TASK_UNINTERRUPTIBLE);
427 			if (atomic_read(&nl_table_users) == 0)
428 				break;
429 			write_unlock_irq(&nl_table_lock);
430 			schedule();
431 			write_lock_irq(&nl_table_lock);
432 		}
433 
434 		__set_current_state(TASK_RUNNING);
435 		remove_wait_queue(&nl_table_wait, &wait);
436 	}
437 }
438 
netlink_table_ungrab(void)439 void netlink_table_ungrab(void)
440 	__releases(nl_table_lock)
441 {
442 	write_unlock_irq(&nl_table_lock);
443 	wake_up(&nl_table_wait);
444 }
445 
446 static inline void
netlink_lock_table(void)447 netlink_lock_table(void)
448 {
449 	unsigned long flags;
450 
451 	/* read_lock() synchronizes us to netlink_table_grab */
452 
453 	read_lock_irqsave(&nl_table_lock, flags);
454 	atomic_inc(&nl_table_users);
455 	read_unlock_irqrestore(&nl_table_lock, flags);
456 }
457 
458 static inline void
netlink_unlock_table(void)459 netlink_unlock_table(void)
460 {
461 	if (atomic_dec_and_test(&nl_table_users))
462 		wake_up(&nl_table_wait);
463 }
464 
465 struct netlink_compare_arg
466 {
467 	possible_net_t pnet;
468 	u32 portid;
469 };
470 
471 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
472 #define netlink_compare_arg_len \
473 	(offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
474 
netlink_compare(struct rhashtable_compare_arg * arg,const void * ptr)475 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
476 				  const void *ptr)
477 {
478 	const struct netlink_compare_arg *x = arg->key;
479 	const struct netlink_sock *nlk = ptr;
480 
481 	return nlk->portid != x->portid ||
482 	       !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
483 }
484 
netlink_compare_arg_init(struct netlink_compare_arg * arg,struct net * net,u32 portid)485 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
486 				     struct net *net, u32 portid)
487 {
488 	memset(arg, 0, sizeof(*arg));
489 	write_pnet(&arg->pnet, net);
490 	arg->portid = portid;
491 }
492 
__netlink_lookup(struct netlink_table * table,u32 portid,struct net * net)493 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
494 				     struct net *net)
495 {
496 	struct netlink_compare_arg arg;
497 
498 	netlink_compare_arg_init(&arg, net, portid);
499 	return rhashtable_lookup_fast(&table->hash, &arg,
500 				      netlink_rhashtable_params);
501 }
502 
__netlink_insert(struct netlink_table * table,struct sock * sk)503 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
504 {
505 	struct netlink_compare_arg arg;
506 
507 	netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
508 	return rhashtable_lookup_insert_key(&table->hash, &arg,
509 					    &nlk_sk(sk)->node,
510 					    netlink_rhashtable_params);
511 }
512 
netlink_lookup(struct net * net,int protocol,u32 portid)513 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
514 {
515 	struct netlink_table *table = &nl_table[protocol];
516 	struct sock *sk;
517 
518 	rcu_read_lock();
519 	sk = __netlink_lookup(table, portid, net);
520 	if (sk)
521 		sock_hold(sk);
522 	rcu_read_unlock();
523 
524 	return sk;
525 }
526 
527 static const struct proto_ops netlink_ops;
528 
529 static void
netlink_update_listeners(struct sock * sk)530 netlink_update_listeners(struct sock *sk)
531 {
532 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
533 	unsigned long mask;
534 	unsigned int i;
535 	struct listeners *listeners;
536 
537 	listeners = nl_deref_protected(tbl->listeners);
538 	if (!listeners)
539 		return;
540 
541 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
542 		mask = 0;
543 		sk_for_each_bound(sk, &tbl->mc_list) {
544 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
545 				mask |= nlk_sk(sk)->groups[i];
546 		}
547 		listeners->masks[i] = mask;
548 	}
549 	/* this function is only called with the netlink table "grabbed", which
550 	 * makes sure updates are visible before bind or setsockopt return. */
551 }
552 
netlink_insert(struct sock * sk,u32 portid)553 static int netlink_insert(struct sock *sk, u32 portid)
554 {
555 	struct netlink_table *table = &nl_table[sk->sk_protocol];
556 	int err;
557 
558 	lock_sock(sk);
559 
560 	err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
561 	if (nlk_sk(sk)->bound)
562 		goto err;
563 
564 	/* portid can be read locklessly from netlink_getname(). */
565 	WRITE_ONCE(nlk_sk(sk)->portid, portid);
566 
567 	sock_hold(sk);
568 
569 	err = __netlink_insert(table, sk);
570 	if (err) {
571 		/* In case the hashtable backend returns with -EBUSY
572 		 * from here, it must not escape to the caller.
573 		 */
574 		if (unlikely(err == -EBUSY))
575 			err = -EOVERFLOW;
576 		if (err == -EEXIST)
577 			err = -EADDRINUSE;
578 		sock_put(sk);
579 		goto err;
580 	}
581 
582 	/* We need to ensure that the socket is hashed and visible. */
583 	smp_wmb();
584 	/* Paired with lockless reads from netlink_bind(),
585 	 * netlink_connect() and netlink_sendmsg().
586 	 */
587 	WRITE_ONCE(nlk_sk(sk)->bound, portid);
588 
589 err:
590 	release_sock(sk);
591 	return err;
592 }
593 
netlink_remove(struct sock * sk)594 static void netlink_remove(struct sock *sk)
595 {
596 	struct netlink_table *table;
597 
598 	table = &nl_table[sk->sk_protocol];
599 	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
600 				    netlink_rhashtable_params)) {
601 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
602 		__sock_put(sk);
603 	}
604 
605 	netlink_table_grab();
606 	if (nlk_sk(sk)->subscriptions) {
607 		__sk_del_bind_node(sk);
608 		netlink_update_listeners(sk);
609 	}
610 	if (sk->sk_protocol == NETLINK_GENERIC)
611 		atomic_inc(&genl_sk_destructing_cnt);
612 	netlink_table_ungrab();
613 }
614 
615 static struct proto netlink_proto = {
616 	.name	  = "NETLINK",
617 	.owner	  = THIS_MODULE,
618 	.obj_size = sizeof(struct netlink_sock),
619 };
620 
__netlink_create(struct net * net,struct socket * sock,int protocol,int kern)621 static int __netlink_create(struct net *net, struct socket *sock,
622 			    int protocol, int kern)
623 {
624 	struct sock *sk;
625 	struct netlink_sock *nlk;
626 
627 	sock->ops = &netlink_ops;
628 
629 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
630 	if (!sk)
631 		return -ENOMEM;
632 
633 	sock_init_data(sock, sk);
634 
635 	nlk = nlk_sk(sk);
636 	mutex_init(&nlk->nl_cb_mutex);
637 	lockdep_set_class_and_name(&nlk->nl_cb_mutex,
638 					   nlk_cb_mutex_keys + protocol,
639 					   nlk_cb_mutex_key_strings[protocol]);
640 	init_waitqueue_head(&nlk->wait);
641 
642 	sk->sk_destruct = netlink_sock_destruct;
643 	sk->sk_protocol = protocol;
644 	return 0;
645 }
646 
netlink_create(struct net * net,struct socket * sock,int protocol,int kern)647 static int netlink_create(struct net *net, struct socket *sock, int protocol,
648 			  int kern)
649 {
650 	struct module *module = NULL;
651 	struct netlink_sock *nlk;
652 	int (*bind)(struct net *net, int group);
653 	void (*unbind)(struct net *net, int group);
654 	void (*release)(struct sock *sock, unsigned long *groups);
655 	int err = 0;
656 
657 	sock->state = SS_UNCONNECTED;
658 
659 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
660 		return -ESOCKTNOSUPPORT;
661 
662 	if (protocol < 0 || protocol >= MAX_LINKS)
663 		return -EPROTONOSUPPORT;
664 	protocol = array_index_nospec(protocol, MAX_LINKS);
665 
666 	netlink_lock_table();
667 #ifdef CONFIG_MODULES
668 	if (!nl_table[protocol].registered) {
669 		netlink_unlock_table();
670 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
671 		netlink_lock_table();
672 	}
673 #endif
674 	if (nl_table[protocol].registered &&
675 	    try_module_get(nl_table[protocol].module))
676 		module = nl_table[protocol].module;
677 	else
678 		err = -EPROTONOSUPPORT;
679 	bind = nl_table[protocol].bind;
680 	unbind = nl_table[protocol].unbind;
681 	release = nl_table[protocol].release;
682 	netlink_unlock_table();
683 
684 	if (err < 0)
685 		goto out;
686 
687 	err = __netlink_create(net, sock, protocol, kern);
688 	if (err < 0)
689 		goto out_module;
690 
691 	sock_prot_inuse_add(net, &netlink_proto, 1);
692 
693 	nlk = nlk_sk(sock->sk);
694 	nlk->module = module;
695 	nlk->netlink_bind = bind;
696 	nlk->netlink_unbind = unbind;
697 	nlk->netlink_release = release;
698 out:
699 	return err;
700 
701 out_module:
702 	module_put(module);
703 	goto out;
704 }
705 
deferred_put_nlk_sk(struct rcu_head * head)706 static void deferred_put_nlk_sk(struct rcu_head *head)
707 {
708 	struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
709 	struct sock *sk = &nlk->sk;
710 
711 	kfree(nlk->groups);
712 	nlk->groups = NULL;
713 
714 	if (!refcount_dec_and_test(&sk->sk_refcnt))
715 		return;
716 
717 	sk_free(sk);
718 }
719 
netlink_release(struct socket * sock)720 static int netlink_release(struct socket *sock)
721 {
722 	struct sock *sk = sock->sk;
723 	struct netlink_sock *nlk;
724 
725 	if (!sk)
726 		return 0;
727 
728 	netlink_remove(sk);
729 	sock_orphan(sk);
730 	nlk = nlk_sk(sk);
731 
732 	/*
733 	 * OK. Socket is unlinked, any packets that arrive now
734 	 * will be purged.
735 	 */
736 	if (nlk->netlink_release)
737 		nlk->netlink_release(sk, nlk->groups);
738 
739 	/* must not acquire netlink_table_lock in any way again before unbind
740 	 * and notifying genetlink is done as otherwise it might deadlock
741 	 */
742 	if (nlk->netlink_unbind) {
743 		int i;
744 
745 		for (i = 0; i < nlk->ngroups; i++)
746 			if (test_bit(i, nlk->groups))
747 				nlk->netlink_unbind(sock_net(sk), i + 1);
748 	}
749 	if (sk->sk_protocol == NETLINK_GENERIC &&
750 	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
751 		wake_up(&genl_sk_destructing_waitq);
752 
753 	sock->sk = NULL;
754 	wake_up_interruptible_all(&nlk->wait);
755 
756 	skb_queue_purge(&sk->sk_write_queue);
757 
758 	if (nlk->portid && nlk->bound) {
759 		struct netlink_notify n = {
760 						.net = sock_net(sk),
761 						.protocol = sk->sk_protocol,
762 						.portid = nlk->portid,
763 					  };
764 		blocking_notifier_call_chain(&netlink_chain,
765 				NETLINK_URELEASE, &n);
766 	}
767 
768 	/* Terminate any outstanding dump */
769 	if (nlk->cb_running) {
770 		if (nlk->cb.done)
771 			nlk->cb.done(&nlk->cb);
772 		module_put(nlk->cb.module);
773 		kfree_skb(nlk->cb.skb);
774 		WRITE_ONCE(nlk->cb_running, false);
775 	}
776 
777 	module_put(nlk->module);
778 
779 	if (netlink_is_kernel(sk)) {
780 		netlink_table_grab();
781 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
782 		if (--nl_table[sk->sk_protocol].registered == 0) {
783 			struct listeners *old;
784 
785 			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
786 			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
787 			kfree_rcu(old, rcu);
788 			nl_table[sk->sk_protocol].module = NULL;
789 			nl_table[sk->sk_protocol].bind = NULL;
790 			nl_table[sk->sk_protocol].unbind = NULL;
791 			nl_table[sk->sk_protocol].flags = 0;
792 			nl_table[sk->sk_protocol].registered = 0;
793 		}
794 		netlink_table_ungrab();
795 	}
796 
797 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
798 
799 	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
800 	return 0;
801 }
802 
netlink_autobind(struct socket * sock)803 static int netlink_autobind(struct socket *sock)
804 {
805 	struct sock *sk = sock->sk;
806 	struct net *net = sock_net(sk);
807 	struct netlink_table *table = &nl_table[sk->sk_protocol];
808 	s32 portid = task_tgid_vnr(current);
809 	int err;
810 	s32 rover = -4096;
811 	bool ok;
812 
813 retry:
814 	cond_resched();
815 	rcu_read_lock();
816 	ok = !__netlink_lookup(table, portid, net);
817 	rcu_read_unlock();
818 	if (!ok) {
819 		/* Bind collision, search negative portid values. */
820 		if (rover == -4096)
821 			/* rover will be in range [S32_MIN, -4097] */
822 			rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
823 		else if (rover >= -4096)
824 			rover = -4097;
825 		portid = rover--;
826 		goto retry;
827 	}
828 
829 	err = netlink_insert(sk, portid);
830 	if (err == -EADDRINUSE)
831 		goto retry;
832 
833 	/* If 2 threads race to autobind, that is fine.  */
834 	if (err == -EBUSY)
835 		err = 0;
836 
837 	return err;
838 }
839 
840 /**
841  * __netlink_ns_capable - General netlink message capability test
842  * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
843  * @user_ns: The user namespace of the capability to use
844  * @cap: The capability to use
845  *
846  * Test to see if the opener of the socket we received the message
847  * from had when the netlink socket was created and the sender of the
848  * message has the capability @cap in the user namespace @user_ns.
849  */
__netlink_ns_capable(const struct netlink_skb_parms * nsp,struct user_namespace * user_ns,int cap)850 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
851 			struct user_namespace *user_ns, int cap)
852 {
853 	return ((nsp->flags & NETLINK_SKB_DST) ||
854 		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
855 		ns_capable(user_ns, cap);
856 }
857 EXPORT_SYMBOL(__netlink_ns_capable);
858 
859 /**
860  * netlink_ns_capable - General netlink message capability test
861  * @skb: socket buffer holding a netlink command from userspace
862  * @user_ns: The user namespace of the capability to use
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 the capability @cap in the user namespace @user_ns.
868  */
netlink_ns_capable(const struct sk_buff * skb,struct user_namespace * user_ns,int cap)869 bool netlink_ns_capable(const struct sk_buff *skb,
870 			struct user_namespace *user_ns, int cap)
871 {
872 	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
873 }
874 EXPORT_SYMBOL(netlink_ns_capable);
875 
876 /**
877  * netlink_capable - Netlink global message capability test
878  * @skb: socket buffer holding a netlink command from userspace
879  * @cap: The capability to use
880  *
881  * Test to see if the opener of the socket we received the message
882  * from had when the netlink socket was created and the sender of the
883  * message has the capability @cap in all user namespaces.
884  */
netlink_capable(const struct sk_buff * skb,int cap)885 bool netlink_capable(const struct sk_buff *skb, int cap)
886 {
887 	return netlink_ns_capable(skb, &init_user_ns, cap);
888 }
889 EXPORT_SYMBOL(netlink_capable);
890 
891 /**
892  * netlink_net_capable - Netlink network namespace message capability test
893  * @skb: socket buffer holding a netlink command from userspace
894  * @cap: The capability to use
895  *
896  * Test to see if the opener of the socket we received the message
897  * from had when the netlink socket was created and the sender of the
898  * message has the capability @cap over the network namespace of
899  * the socket we received the message from.
900  */
netlink_net_capable(const struct sk_buff * skb,int cap)901 bool netlink_net_capable(const struct sk_buff *skb, int cap)
902 {
903 	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
904 }
905 EXPORT_SYMBOL(netlink_net_capable);
906 
netlink_allowed(const struct socket * sock,unsigned int flag)907 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
908 {
909 	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
910 		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
911 }
912 
913 static void
netlink_update_subscriptions(struct sock * sk,unsigned int subscriptions)914 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
915 {
916 	struct netlink_sock *nlk = nlk_sk(sk);
917 
918 	if (nlk->subscriptions && !subscriptions)
919 		__sk_del_bind_node(sk);
920 	else if (!nlk->subscriptions && subscriptions)
921 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
922 	nlk->subscriptions = subscriptions;
923 }
924 
netlink_realloc_groups(struct sock * sk)925 static int netlink_realloc_groups(struct sock *sk)
926 {
927 	struct netlink_sock *nlk = nlk_sk(sk);
928 	unsigned int groups;
929 	unsigned long *new_groups;
930 	int err = 0;
931 
932 	netlink_table_grab();
933 
934 	groups = nl_table[sk->sk_protocol].groups;
935 	if (!nl_table[sk->sk_protocol].registered) {
936 		err = -ENOENT;
937 		goto out_unlock;
938 	}
939 
940 	if (nlk->ngroups >= groups)
941 		goto out_unlock;
942 
943 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
944 	if (new_groups == NULL) {
945 		err = -ENOMEM;
946 		goto out_unlock;
947 	}
948 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
949 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
950 
951 	nlk->groups = new_groups;
952 	nlk->ngroups = groups;
953  out_unlock:
954 	netlink_table_ungrab();
955 	return err;
956 }
957 
netlink_undo_bind(int group,long unsigned int groups,struct sock * sk)958 static void netlink_undo_bind(int group, long unsigned int groups,
959 			      struct sock *sk)
960 {
961 	struct netlink_sock *nlk = nlk_sk(sk);
962 	int undo;
963 
964 	if (!nlk->netlink_unbind)
965 		return;
966 
967 	for (undo = 0; undo < group; undo++)
968 		if (test_bit(undo, &groups))
969 			nlk->netlink_unbind(sock_net(sk), undo + 1);
970 }
971 
netlink_bind(struct socket * sock,struct sockaddr * addr,int addr_len)972 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
973 			int addr_len)
974 {
975 	struct sock *sk = sock->sk;
976 	struct net *net = sock_net(sk);
977 	struct netlink_sock *nlk = nlk_sk(sk);
978 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
979 	int err = 0;
980 	unsigned long groups;
981 	bool bound;
982 
983 	if (addr_len < sizeof(struct sockaddr_nl))
984 		return -EINVAL;
985 
986 	if (nladdr->nl_family != AF_NETLINK)
987 		return -EINVAL;
988 	groups = nladdr->nl_groups;
989 
990 	/* Only superuser is allowed to listen multicasts */
991 	if (groups) {
992 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
993 			return -EPERM;
994 		err = netlink_realloc_groups(sk);
995 		if (err)
996 			return err;
997 	}
998 
999 	if (nlk->ngroups < BITS_PER_LONG)
1000 		groups &= (1UL << nlk->ngroups) - 1;
1001 
1002 	/* Paired with WRITE_ONCE() in netlink_insert() */
1003 	bound = READ_ONCE(nlk->bound);
1004 	if (bound) {
1005 		/* Ensure nlk->portid is up-to-date. */
1006 		smp_rmb();
1007 
1008 		if (nladdr->nl_pid != nlk->portid)
1009 			return -EINVAL;
1010 	}
1011 
1012 	if (nlk->netlink_bind && groups) {
1013 		int group;
1014 
1015 		/* nl_groups is a u32, so cap the maximum groups we can bind */
1016 		for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1017 			if (!test_bit(group, &groups))
1018 				continue;
1019 			err = nlk->netlink_bind(net, group + 1);
1020 			if (!err)
1021 				continue;
1022 			netlink_undo_bind(group, groups, sk);
1023 			return err;
1024 		}
1025 	}
1026 
1027 	/* No need for barriers here as we return to user-space without
1028 	 * using any of the bound attributes.
1029 	 */
1030 	netlink_lock_table();
1031 	if (!bound) {
1032 		err = nladdr->nl_pid ?
1033 			netlink_insert(sk, nladdr->nl_pid) :
1034 			netlink_autobind(sock);
1035 		if (err) {
1036 			netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1037 			goto unlock;
1038 		}
1039 	}
1040 
1041 	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1042 		goto unlock;
1043 	netlink_unlock_table();
1044 
1045 	netlink_table_grab();
1046 	netlink_update_subscriptions(sk, nlk->subscriptions +
1047 					 hweight32(groups) -
1048 					 hweight32(nlk->groups[0]));
1049 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1050 	netlink_update_listeners(sk);
1051 	netlink_table_ungrab();
1052 
1053 	return 0;
1054 
1055 unlock:
1056 	netlink_unlock_table();
1057 	return err;
1058 }
1059 
netlink_connect(struct socket * sock,struct sockaddr * addr,int alen,int flags)1060 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1061 			   int alen, int flags)
1062 {
1063 	int err = 0;
1064 	struct sock *sk = sock->sk;
1065 	struct netlink_sock *nlk = nlk_sk(sk);
1066 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1067 
1068 	if (alen < sizeof(addr->sa_family))
1069 		return -EINVAL;
1070 
1071 	if (addr->sa_family == AF_UNSPEC) {
1072 		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1073 		WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1074 		/* dst_portid and dst_group can be read locklessly */
1075 		WRITE_ONCE(nlk->dst_portid, 0);
1076 		WRITE_ONCE(nlk->dst_group, 0);
1077 		return 0;
1078 	}
1079 	if (addr->sa_family != AF_NETLINK)
1080 		return -EINVAL;
1081 
1082 	if (alen < sizeof(struct sockaddr_nl))
1083 		return -EINVAL;
1084 
1085 	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1086 	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1087 		return -EPERM;
1088 
1089 	/* No need for barriers here as we return to user-space without
1090 	 * using any of the bound attributes.
1091 	 * Paired with WRITE_ONCE() in netlink_insert().
1092 	 */
1093 	if (!READ_ONCE(nlk->bound))
1094 		err = netlink_autobind(sock);
1095 
1096 	if (err == 0) {
1097 		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1098 		WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1099 		/* dst_portid and dst_group can be read locklessly */
1100 		WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1101 		WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1102 	}
1103 
1104 	return err;
1105 }
1106 
netlink_getname(struct socket * sock,struct sockaddr * addr,int peer)1107 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1108 			   int peer)
1109 {
1110 	struct sock *sk = sock->sk;
1111 	struct netlink_sock *nlk = nlk_sk(sk);
1112 	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1113 
1114 	nladdr->nl_family = AF_NETLINK;
1115 	nladdr->nl_pad = 0;
1116 
1117 	if (peer) {
1118 		/* Paired with WRITE_ONCE() in netlink_connect() */
1119 		nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1120 		nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1121 	} else {
1122 		/* Paired with WRITE_ONCE() in netlink_insert() */
1123 		nladdr->nl_pid = READ_ONCE(nlk->portid);
1124 		netlink_lock_table();
1125 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1126 		netlink_unlock_table();
1127 	}
1128 	return sizeof(*nladdr);
1129 }
1130 
netlink_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1131 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1132 			 unsigned long arg)
1133 {
1134 	/* try to hand this ioctl down to the NIC drivers.
1135 	 */
1136 	return -ENOIOCTLCMD;
1137 }
1138 
netlink_getsockbyportid(struct sock * ssk,u32 portid)1139 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1140 {
1141 	struct sock *sock;
1142 	struct netlink_sock *nlk;
1143 
1144 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1145 	if (!sock)
1146 		return ERR_PTR(-ECONNREFUSED);
1147 
1148 	/* Don't bother queuing skb if kernel socket has no input function */
1149 	nlk = nlk_sk(sock);
1150 	/* dst_portid and sk_state can be changed in netlink_connect() */
1151 	if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1152 	    READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1153 		sock_put(sock);
1154 		return ERR_PTR(-ECONNREFUSED);
1155 	}
1156 	return sock;
1157 }
1158 
netlink_getsockbyfd(int fd)1159 struct sock *netlink_getsockbyfd(int fd)
1160 {
1161 	CLASS(fd, f)(fd);
1162 	struct inode *inode;
1163 	struct sock *sock;
1164 
1165 	if (fd_empty(f))
1166 		return ERR_PTR(-EBADF);
1167 
1168 	inode = file_inode(fd_file(f));
1169 	if (!S_ISSOCK(inode->i_mode))
1170 		return ERR_PTR(-ENOTSOCK);
1171 
1172 	sock = SOCKET_I(inode)->sk;
1173 	if (sock->sk_family != AF_NETLINK)
1174 		return ERR_PTR(-EINVAL);
1175 
1176 	sock_hold(sock);
1177 	return sock;
1178 }
1179 
netlink_alloc_large_skb(unsigned int size,int broadcast)1180 struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1181 {
1182 	size_t head_size = SKB_HEAD_ALIGN(size);
1183 	struct sk_buff *skb;
1184 	void *data;
1185 
1186 	if (head_size <= PAGE_SIZE || broadcast)
1187 		return alloc_skb(size, GFP_KERNEL);
1188 
1189 	data = kvmalloc(head_size, GFP_KERNEL);
1190 	if (!data)
1191 		return NULL;
1192 
1193 	skb = __build_skb(data, head_size);
1194 	if (!skb)
1195 		kvfree(data);
1196 	else if (is_vmalloc_addr(data))
1197 		skb->destructor = netlink_skb_destructor;
1198 
1199 	return skb;
1200 }
1201 
1202 /*
1203  * Attach a skb to a netlink socket.
1204  * The caller must hold a reference to the destination socket. On error, the
1205  * reference is dropped. The skb is not send to the destination, just all
1206  * all error checks are performed and memory in the queue is reserved.
1207  * Return values:
1208  * < 0: error. skb freed, reference to sock dropped.
1209  * 0: continue
1210  * 1: repeat lookup - reference dropped while waiting for socket memory.
1211  */
netlink_attachskb(struct sock * sk,struct sk_buff * skb,long * timeo,struct sock * ssk)1212 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1213 		      long *timeo, struct sock *ssk)
1214 {
1215 	struct netlink_sock *nlk;
1216 
1217 	nlk = nlk_sk(sk);
1218 
1219 	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1220 	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1221 		DECLARE_WAITQUEUE(wait, current);
1222 		if (!*timeo) {
1223 			if (!ssk || netlink_is_kernel(ssk))
1224 				netlink_overrun(sk);
1225 			sock_put(sk);
1226 			kfree_skb(skb);
1227 			return -EAGAIN;
1228 		}
1229 
1230 		__set_current_state(TASK_INTERRUPTIBLE);
1231 		add_wait_queue(&nlk->wait, &wait);
1232 
1233 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1234 		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1235 		    !sock_flag(sk, SOCK_DEAD))
1236 			*timeo = schedule_timeout(*timeo);
1237 
1238 		__set_current_state(TASK_RUNNING);
1239 		remove_wait_queue(&nlk->wait, &wait);
1240 		sock_put(sk);
1241 
1242 		if (signal_pending(current)) {
1243 			kfree_skb(skb);
1244 			return sock_intr_errno(*timeo);
1245 		}
1246 		return 1;
1247 	}
1248 	netlink_skb_set_owner_r(skb, sk);
1249 	return 0;
1250 }
1251 
__netlink_sendskb(struct sock * sk,struct sk_buff * skb)1252 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1253 {
1254 	int len = skb->len;
1255 
1256 	netlink_deliver_tap(sock_net(sk), skb);
1257 
1258 	skb_queue_tail(&sk->sk_receive_queue, skb);
1259 	sk->sk_data_ready(sk);
1260 	return len;
1261 }
1262 
netlink_sendskb(struct sock * sk,struct sk_buff * skb)1263 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1264 {
1265 	int len = __netlink_sendskb(sk, skb);
1266 
1267 	sock_put(sk);
1268 	return len;
1269 }
1270 
netlink_detachskb(struct sock * sk,struct sk_buff * skb)1271 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1272 {
1273 	kfree_skb(skb);
1274 	sock_put(sk);
1275 }
1276 
netlink_trim(struct sk_buff * skb,gfp_t allocation)1277 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1278 {
1279 	int delta;
1280 
1281 	skb_assert_len(skb);
1282 	WARN_ON(skb->sk != NULL);
1283 	delta = skb->end - skb->tail;
1284 	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1285 		return skb;
1286 
1287 	if (skb_shared(skb)) {
1288 		struct sk_buff *nskb = skb_clone(skb, allocation);
1289 		if (!nskb)
1290 			return skb;
1291 		consume_skb(skb);
1292 		skb = nskb;
1293 	}
1294 
1295 	pskb_expand_head(skb, 0, -delta,
1296 			 (allocation & ~__GFP_DIRECT_RECLAIM) |
1297 			 __GFP_NOWARN | __GFP_NORETRY);
1298 	return skb;
1299 }
1300 
netlink_unicast_kernel(struct sock * sk,struct sk_buff * skb,struct sock * ssk)1301 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1302 				  struct sock *ssk)
1303 {
1304 	int ret;
1305 	struct netlink_sock *nlk = nlk_sk(sk);
1306 
1307 	ret = -ECONNREFUSED;
1308 	if (nlk->netlink_rcv != NULL) {
1309 		ret = skb->len;
1310 		netlink_skb_set_owner_r(skb, sk);
1311 		NETLINK_CB(skb).sk = ssk;
1312 		netlink_deliver_tap_kernel(sk, ssk, skb);
1313 		nlk->netlink_rcv(skb);
1314 		consume_skb(skb);
1315 	} else {
1316 		kfree_skb(skb);
1317 	}
1318 	sock_put(sk);
1319 	return ret;
1320 }
1321 
netlink_unicast(struct sock * ssk,struct sk_buff * skb,u32 portid,int nonblock)1322 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1323 		    u32 portid, int nonblock)
1324 {
1325 	struct sock *sk;
1326 	int err;
1327 	long timeo;
1328 
1329 	skb = netlink_trim(skb, gfp_any());
1330 
1331 	timeo = sock_sndtimeo(ssk, nonblock);
1332 retry:
1333 	sk = netlink_getsockbyportid(ssk, portid);
1334 	if (IS_ERR(sk)) {
1335 		kfree_skb(skb);
1336 		return PTR_ERR(sk);
1337 	}
1338 	if (netlink_is_kernel(sk))
1339 		return netlink_unicast_kernel(sk, skb, ssk);
1340 
1341 	if (sk_filter(sk, skb)) {
1342 		err = skb->len;
1343 		kfree_skb(skb);
1344 		sock_put(sk);
1345 		return err;
1346 	}
1347 
1348 	err = netlink_attachskb(sk, skb, &timeo, ssk);
1349 	if (err == 1)
1350 		goto retry;
1351 	if (err)
1352 		return err;
1353 
1354 	return netlink_sendskb(sk, skb);
1355 }
1356 EXPORT_SYMBOL(netlink_unicast);
1357 
netlink_has_listeners(struct sock * sk,unsigned int group)1358 int netlink_has_listeners(struct sock *sk, unsigned int group)
1359 {
1360 	int res = 0;
1361 	struct listeners *listeners;
1362 
1363 	BUG_ON(!netlink_is_kernel(sk));
1364 
1365 	rcu_read_lock();
1366 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1367 
1368 	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1369 		res = test_bit(group - 1, listeners->masks);
1370 
1371 	rcu_read_unlock();
1372 
1373 	return res;
1374 }
1375 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1376 
netlink_strict_get_check(struct sk_buff * skb)1377 bool netlink_strict_get_check(struct sk_buff *skb)
1378 {
1379 	return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1380 }
1381 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1382 
netlink_broadcast_deliver(struct sock * sk,struct sk_buff * skb)1383 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1384 {
1385 	struct netlink_sock *nlk = nlk_sk(sk);
1386 
1387 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1388 	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1389 		netlink_skb_set_owner_r(skb, sk);
1390 		__netlink_sendskb(sk, skb);
1391 		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1392 	}
1393 	return -1;
1394 }
1395 
1396 struct netlink_broadcast_data {
1397 	struct sock *exclude_sk;
1398 	struct net *net;
1399 	u32 portid;
1400 	u32 group;
1401 	int failure;
1402 	int delivery_failure;
1403 	int congested;
1404 	int delivered;
1405 	gfp_t allocation;
1406 	struct sk_buff *skb, *skb2;
1407 	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1408 	void *tx_data;
1409 };
1410 
do_one_broadcast(struct sock * sk,struct netlink_broadcast_data * p)1411 static void do_one_broadcast(struct sock *sk,
1412 				    struct netlink_broadcast_data *p)
1413 {
1414 	struct netlink_sock *nlk = nlk_sk(sk);
1415 	int val;
1416 
1417 	if (p->exclude_sk == sk)
1418 		return;
1419 
1420 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1421 	    !test_bit(p->group - 1, nlk->groups))
1422 		return;
1423 
1424 	if (!net_eq(sock_net(sk), p->net)) {
1425 		if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1426 			return;
1427 
1428 		if (!peernet_has_id(sock_net(sk), p->net))
1429 			return;
1430 
1431 		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1432 				     CAP_NET_BROADCAST))
1433 			return;
1434 	}
1435 
1436 	if (p->failure) {
1437 		netlink_overrun(sk);
1438 		return;
1439 	}
1440 
1441 	sock_hold(sk);
1442 	if (p->skb2 == NULL) {
1443 		if (skb_shared(p->skb)) {
1444 			p->skb2 = skb_clone(p->skb, p->allocation);
1445 		} else {
1446 			p->skb2 = skb_get(p->skb);
1447 			/*
1448 			 * skb ownership may have been set when
1449 			 * delivered to a previous socket.
1450 			 */
1451 			skb_orphan(p->skb2);
1452 		}
1453 	}
1454 	if (p->skb2 == NULL) {
1455 		netlink_overrun(sk);
1456 		/* Clone failed. Notify ALL listeners. */
1457 		p->failure = 1;
1458 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1459 			p->delivery_failure = 1;
1460 		goto out;
1461 	}
1462 
1463 	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1464 		kfree_skb(p->skb2);
1465 		p->skb2 = NULL;
1466 		goto out;
1467 	}
1468 
1469 	if (sk_filter(sk, p->skb2)) {
1470 		kfree_skb(p->skb2);
1471 		p->skb2 = NULL;
1472 		goto out;
1473 	}
1474 	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1475 	if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1476 		NETLINK_CB(p->skb2).nsid_is_set = true;
1477 	val = netlink_broadcast_deliver(sk, p->skb2);
1478 	if (val < 0) {
1479 		netlink_overrun(sk);
1480 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1481 			p->delivery_failure = 1;
1482 	} else {
1483 		p->congested |= val;
1484 		p->delivered = 1;
1485 		p->skb2 = NULL;
1486 	}
1487 out:
1488 	sock_put(sk);
1489 }
1490 
netlink_broadcast_filtered(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation,netlink_filter_fn filter,void * filter_data)1491 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1492 			       u32 portid,
1493 			       u32 group, gfp_t allocation,
1494 			       netlink_filter_fn filter,
1495 			       void *filter_data)
1496 {
1497 	struct net *net = sock_net(ssk);
1498 	struct netlink_broadcast_data info;
1499 	struct sock *sk;
1500 
1501 	skb = netlink_trim(skb, allocation);
1502 
1503 	info.exclude_sk = ssk;
1504 	info.net = net;
1505 	info.portid = portid;
1506 	info.group = group;
1507 	info.failure = 0;
1508 	info.delivery_failure = 0;
1509 	info.congested = 0;
1510 	info.delivered = 0;
1511 	info.allocation = allocation;
1512 	info.skb = skb;
1513 	info.skb2 = NULL;
1514 	info.tx_filter = filter;
1515 	info.tx_data = filter_data;
1516 
1517 	/* While we sleep in clone, do not allow to change socket list */
1518 
1519 	netlink_lock_table();
1520 
1521 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1522 		do_one_broadcast(sk, &info);
1523 
1524 	consume_skb(skb);
1525 
1526 	netlink_unlock_table();
1527 
1528 	if (info.delivery_failure) {
1529 		kfree_skb(info.skb2);
1530 		return -ENOBUFS;
1531 	}
1532 	consume_skb(info.skb2);
1533 
1534 	if (info.delivered) {
1535 		if (info.congested && gfpflags_allow_blocking(allocation))
1536 			yield();
1537 		return 0;
1538 	}
1539 	return -ESRCH;
1540 }
1541 EXPORT_SYMBOL(netlink_broadcast_filtered);
1542 
netlink_broadcast(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation)1543 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1544 		      u32 group, gfp_t allocation)
1545 {
1546 	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1547 					  NULL, NULL);
1548 }
1549 EXPORT_SYMBOL(netlink_broadcast);
1550 
1551 struct netlink_set_err_data {
1552 	struct sock *exclude_sk;
1553 	u32 portid;
1554 	u32 group;
1555 	int code;
1556 };
1557 
do_one_set_err(struct sock * sk,struct netlink_set_err_data * p)1558 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1559 {
1560 	struct netlink_sock *nlk = nlk_sk(sk);
1561 	int ret = 0;
1562 
1563 	if (sk == p->exclude_sk)
1564 		goto out;
1565 
1566 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1567 		goto out;
1568 
1569 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1570 	    !test_bit(p->group - 1, nlk->groups))
1571 		goto out;
1572 
1573 	if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1574 		ret = 1;
1575 		goto out;
1576 	}
1577 
1578 	WRITE_ONCE(sk->sk_err, p->code);
1579 	sk_error_report(sk);
1580 out:
1581 	return ret;
1582 }
1583 
1584 /**
1585  * netlink_set_err - report error to broadcast listeners
1586  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1587  * @portid: the PORTID of a process that we want to skip (if any)
1588  * @group: the broadcast group that will notice the error
1589  * @code: error code, must be negative (as usual in kernelspace)
1590  *
1591  * This function returns the number of broadcast listeners that have set the
1592  * NETLINK_NO_ENOBUFS socket option.
1593  */
netlink_set_err(struct sock * ssk,u32 portid,u32 group,int code)1594 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1595 {
1596 	struct netlink_set_err_data info;
1597 	unsigned long flags;
1598 	struct sock *sk;
1599 	int ret = 0;
1600 
1601 	info.exclude_sk = ssk;
1602 	info.portid = portid;
1603 	info.group = group;
1604 	/* sk->sk_err wants a positive error value */
1605 	info.code = -code;
1606 
1607 	read_lock_irqsave(&nl_table_lock, flags);
1608 
1609 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1610 		ret += do_one_set_err(sk, &info);
1611 
1612 	read_unlock_irqrestore(&nl_table_lock, flags);
1613 	return ret;
1614 }
1615 EXPORT_SYMBOL(netlink_set_err);
1616 
1617 /* must be called with netlink table grabbed */
netlink_update_socket_mc(struct netlink_sock * nlk,unsigned int group,int is_new)1618 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1619 				     unsigned int group,
1620 				     int is_new)
1621 {
1622 	int old, new = !!is_new, subscriptions;
1623 
1624 	old = test_bit(group - 1, nlk->groups);
1625 	subscriptions = nlk->subscriptions - old + new;
1626 	__assign_bit(group - 1, nlk->groups, new);
1627 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1628 	netlink_update_listeners(&nlk->sk);
1629 }
1630 
netlink_setsockopt(struct socket * sock,int level,int optname,sockptr_t optval,unsigned int optlen)1631 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1632 			      sockptr_t optval, unsigned int optlen)
1633 {
1634 	struct sock *sk = sock->sk;
1635 	struct netlink_sock *nlk = nlk_sk(sk);
1636 	unsigned int val = 0;
1637 	int nr = -1;
1638 
1639 	if (level != SOL_NETLINK)
1640 		return -ENOPROTOOPT;
1641 
1642 	if (optlen >= sizeof(int) &&
1643 	    copy_from_sockptr(&val, optval, sizeof(val)))
1644 		return -EFAULT;
1645 
1646 	switch (optname) {
1647 	case NETLINK_PKTINFO:
1648 		nr = NETLINK_F_RECV_PKTINFO;
1649 		break;
1650 	case NETLINK_ADD_MEMBERSHIP:
1651 	case NETLINK_DROP_MEMBERSHIP: {
1652 		int err;
1653 
1654 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1655 			return -EPERM;
1656 		err = netlink_realloc_groups(sk);
1657 		if (err)
1658 			return err;
1659 		if (!val || val - 1 >= nlk->ngroups)
1660 			return -EINVAL;
1661 		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1662 			err = nlk->netlink_bind(sock_net(sk), val);
1663 			if (err)
1664 				return err;
1665 		}
1666 		netlink_table_grab();
1667 		netlink_update_socket_mc(nlk, val,
1668 					 optname == NETLINK_ADD_MEMBERSHIP);
1669 		netlink_table_ungrab();
1670 		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1671 			nlk->netlink_unbind(sock_net(sk), val);
1672 
1673 		break;
1674 	}
1675 	case NETLINK_BROADCAST_ERROR:
1676 		nr = NETLINK_F_BROADCAST_SEND_ERROR;
1677 		break;
1678 	case NETLINK_NO_ENOBUFS:
1679 		assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1680 		if (val) {
1681 			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1682 			wake_up_interruptible(&nlk->wait);
1683 		}
1684 		break;
1685 	case NETLINK_LISTEN_ALL_NSID:
1686 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1687 			return -EPERM;
1688 		nr = NETLINK_F_LISTEN_ALL_NSID;
1689 		break;
1690 	case NETLINK_CAP_ACK:
1691 		nr = NETLINK_F_CAP_ACK;
1692 		break;
1693 	case NETLINK_EXT_ACK:
1694 		nr = NETLINK_F_EXT_ACK;
1695 		break;
1696 	case NETLINK_GET_STRICT_CHK:
1697 		nr = NETLINK_F_STRICT_CHK;
1698 		break;
1699 	default:
1700 		return -ENOPROTOOPT;
1701 	}
1702 	if (nr >= 0)
1703 		assign_bit(nr, &nlk->flags, val);
1704 	return 0;
1705 }
1706 
netlink_getsockopt(struct socket * sock,int level,int optname,char __user * optval,int __user * optlen)1707 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1708 			      char __user *optval, int __user *optlen)
1709 {
1710 	struct sock *sk = sock->sk;
1711 	struct netlink_sock *nlk = nlk_sk(sk);
1712 	unsigned int flag;
1713 	int len, val;
1714 
1715 	if (level != SOL_NETLINK)
1716 		return -ENOPROTOOPT;
1717 
1718 	if (get_user(len, optlen))
1719 		return -EFAULT;
1720 	if (len < 0)
1721 		return -EINVAL;
1722 
1723 	switch (optname) {
1724 	case NETLINK_PKTINFO:
1725 		flag = NETLINK_F_RECV_PKTINFO;
1726 		break;
1727 	case NETLINK_BROADCAST_ERROR:
1728 		flag = NETLINK_F_BROADCAST_SEND_ERROR;
1729 		break;
1730 	case NETLINK_NO_ENOBUFS:
1731 		flag = NETLINK_F_RECV_NO_ENOBUFS;
1732 		break;
1733 	case NETLINK_LIST_MEMBERSHIPS: {
1734 		int pos, idx, shift, err = 0;
1735 
1736 		netlink_lock_table();
1737 		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1738 			if (len - pos < sizeof(u32))
1739 				break;
1740 
1741 			idx = pos / sizeof(unsigned long);
1742 			shift = (pos % sizeof(unsigned long)) * 8;
1743 			if (put_user((u32)(nlk->groups[idx] >> shift),
1744 				     (u32 __user *)(optval + pos))) {
1745 				err = -EFAULT;
1746 				break;
1747 			}
1748 		}
1749 		if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1750 			err = -EFAULT;
1751 		netlink_unlock_table();
1752 		return err;
1753 	}
1754 	case NETLINK_LISTEN_ALL_NSID:
1755 		flag = NETLINK_F_LISTEN_ALL_NSID;
1756 		break;
1757 	case NETLINK_CAP_ACK:
1758 		flag = NETLINK_F_CAP_ACK;
1759 		break;
1760 	case NETLINK_EXT_ACK:
1761 		flag = NETLINK_F_EXT_ACK;
1762 		break;
1763 	case NETLINK_GET_STRICT_CHK:
1764 		flag = NETLINK_F_STRICT_CHK;
1765 		break;
1766 	default:
1767 		return -ENOPROTOOPT;
1768 	}
1769 
1770 	if (len < sizeof(int))
1771 		return -EINVAL;
1772 
1773 	len = sizeof(int);
1774 	val = test_bit(flag, &nlk->flags);
1775 
1776 	if (put_user(len, optlen) ||
1777 	    copy_to_user(optval, &val, len))
1778 		return -EFAULT;
1779 
1780 	return 0;
1781 }
1782 
netlink_cmsg_recv_pktinfo(struct msghdr * msg,struct sk_buff * skb)1783 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1784 {
1785 	struct nl_pktinfo info;
1786 
1787 	info.group = NETLINK_CB(skb).dst_group;
1788 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1789 }
1790 
netlink_cmsg_listen_all_nsid(struct sock * sk,struct msghdr * msg,struct sk_buff * skb)1791 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1792 					 struct sk_buff *skb)
1793 {
1794 	if (!NETLINK_CB(skb).nsid_is_set)
1795 		return;
1796 
1797 	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1798 		 &NETLINK_CB(skb).nsid);
1799 }
1800 
netlink_sendmsg(struct socket * sock,struct msghdr * msg,size_t len)1801 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1802 {
1803 	struct sock *sk = sock->sk;
1804 	struct netlink_sock *nlk = nlk_sk(sk);
1805 	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1806 	u32 dst_portid;
1807 	u32 dst_group;
1808 	struct sk_buff *skb;
1809 	int err;
1810 	struct scm_cookie scm;
1811 	u32 netlink_skb_flags = 0;
1812 
1813 	if (msg->msg_flags & MSG_OOB)
1814 		return -EOPNOTSUPP;
1815 
1816 	if (len == 0) {
1817 		pr_warn_once("Zero length message leads to an empty skb\n");
1818 		return -ENODATA;
1819 	}
1820 
1821 	err = scm_send(sock, msg, &scm, true);
1822 	if (err < 0)
1823 		return err;
1824 
1825 	if (msg->msg_namelen) {
1826 		err = -EINVAL;
1827 		if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1828 			goto out;
1829 		if (addr->nl_family != AF_NETLINK)
1830 			goto out;
1831 		dst_portid = addr->nl_pid;
1832 		dst_group = ffs(addr->nl_groups);
1833 		err =  -EPERM;
1834 		if ((dst_group || dst_portid) &&
1835 		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1836 			goto out;
1837 		netlink_skb_flags |= NETLINK_SKB_DST;
1838 	} else {
1839 		/* Paired with WRITE_ONCE() in netlink_connect() */
1840 		dst_portid = READ_ONCE(nlk->dst_portid);
1841 		dst_group = READ_ONCE(nlk->dst_group);
1842 	}
1843 
1844 	/* Paired with WRITE_ONCE() in netlink_insert() */
1845 	if (!READ_ONCE(nlk->bound)) {
1846 		err = netlink_autobind(sock);
1847 		if (err)
1848 			goto out;
1849 	} else {
1850 		/* Ensure nlk is hashed and visible. */
1851 		smp_rmb();
1852 	}
1853 
1854 	err = -EMSGSIZE;
1855 	if (len > sk->sk_sndbuf - 32)
1856 		goto out;
1857 	err = -ENOBUFS;
1858 	skb = netlink_alloc_large_skb(len, dst_group);
1859 	if (skb == NULL)
1860 		goto out;
1861 
1862 	NETLINK_CB(skb).portid	= nlk->portid;
1863 	NETLINK_CB(skb).dst_group = dst_group;
1864 	NETLINK_CB(skb).creds	= scm.creds;
1865 	NETLINK_CB(skb).flags	= netlink_skb_flags;
1866 
1867 	err = -EFAULT;
1868 	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1869 		kfree_skb(skb);
1870 		goto out;
1871 	}
1872 
1873 	err = security_netlink_send(sk, skb);
1874 	if (err) {
1875 		kfree_skb(skb);
1876 		goto out;
1877 	}
1878 
1879 	if (dst_group) {
1880 		refcount_inc(&skb->users);
1881 		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1882 	}
1883 	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1884 
1885 out:
1886 	scm_destroy(&scm);
1887 	return err;
1888 }
1889 
netlink_recvmsg(struct socket * sock,struct msghdr * msg,size_t len,int flags)1890 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1891 			   int flags)
1892 {
1893 	struct scm_cookie scm;
1894 	struct sock *sk = sock->sk;
1895 	struct netlink_sock *nlk = nlk_sk(sk);
1896 	size_t copied, max_recvmsg_len;
1897 	struct sk_buff *skb, *data_skb;
1898 	int err, ret;
1899 
1900 	if (flags & MSG_OOB)
1901 		return -EOPNOTSUPP;
1902 
1903 	copied = 0;
1904 
1905 	skb = skb_recv_datagram(sk, flags, &err);
1906 	if (skb == NULL)
1907 		goto out;
1908 
1909 	data_skb = skb;
1910 
1911 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1912 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1913 		/*
1914 		 * If this skb has a frag_list, then here that means that we
1915 		 * will have to use the frag_list skb's data for compat tasks
1916 		 * and the regular skb's data for normal (non-compat) tasks.
1917 		 *
1918 		 * If we need to send the compat skb, assign it to the
1919 		 * 'data_skb' variable so that it will be used below for data
1920 		 * copying. We keep 'skb' for everything else, including
1921 		 * freeing both later.
1922 		 */
1923 		if (flags & MSG_CMSG_COMPAT)
1924 			data_skb = skb_shinfo(skb)->frag_list;
1925 	}
1926 #endif
1927 
1928 	/* Record the max length of recvmsg() calls for future allocations */
1929 	max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1930 	max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1931 				SKB_WITH_OVERHEAD(32768));
1932 	WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1933 
1934 	copied = data_skb->len;
1935 	if (len < copied) {
1936 		msg->msg_flags |= MSG_TRUNC;
1937 		copied = len;
1938 	}
1939 
1940 	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1941 
1942 	if (msg->msg_name) {
1943 		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1944 		addr->nl_family = AF_NETLINK;
1945 		addr->nl_pad    = 0;
1946 		addr->nl_pid	= NETLINK_CB(skb).portid;
1947 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1948 		msg->msg_namelen = sizeof(*addr);
1949 	}
1950 
1951 	if (nlk_test_bit(RECV_PKTINFO, sk))
1952 		netlink_cmsg_recv_pktinfo(msg, skb);
1953 	if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1954 		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1955 
1956 	memset(&scm, 0, sizeof(scm));
1957 	scm.creds = *NETLINK_CREDS(skb);
1958 	if (flags & MSG_TRUNC)
1959 		copied = data_skb->len;
1960 
1961 	skb_free_datagram(sk, skb);
1962 
1963 	if (READ_ONCE(nlk->cb_running) &&
1964 	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1965 		ret = netlink_dump(sk, false);
1966 		if (ret) {
1967 			WRITE_ONCE(sk->sk_err, -ret);
1968 			sk_error_report(sk);
1969 		}
1970 	}
1971 
1972 	scm_recv(sock, msg, &scm, flags);
1973 out:
1974 	netlink_rcv_wake(sk);
1975 	return err ? : copied;
1976 }
1977 
netlink_data_ready(struct sock * sk)1978 static void netlink_data_ready(struct sock *sk)
1979 {
1980 	BUG();
1981 }
1982 
1983 /*
1984  *	We export these functions to other modules. They provide a
1985  *	complete set of kernel non-blocking support for message
1986  *	queueing.
1987  */
1988 
1989 struct sock *
__netlink_kernel_create(struct net * net,int unit,struct module * module,struct netlink_kernel_cfg * cfg)1990 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1991 			struct netlink_kernel_cfg *cfg)
1992 {
1993 	struct socket *sock;
1994 	struct sock *sk;
1995 	struct netlink_sock *nlk;
1996 	struct listeners *listeners = NULL;
1997 	unsigned int groups;
1998 
1999 	BUG_ON(!nl_table);
2000 
2001 	if (unit < 0 || unit >= MAX_LINKS)
2002 		return NULL;
2003 
2004 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2005 		return NULL;
2006 
2007 	if (__netlink_create(net, sock, unit, 1) < 0)
2008 		goto out_sock_release_nosk;
2009 
2010 	sk = sock->sk;
2011 
2012 	if (!cfg || cfg->groups < 32)
2013 		groups = 32;
2014 	else
2015 		groups = cfg->groups;
2016 
2017 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2018 	if (!listeners)
2019 		goto out_sock_release;
2020 
2021 	sk->sk_data_ready = netlink_data_ready;
2022 	if (cfg && cfg->input)
2023 		nlk_sk(sk)->netlink_rcv = cfg->input;
2024 
2025 	if (netlink_insert(sk, 0))
2026 		goto out_sock_release;
2027 
2028 	nlk = nlk_sk(sk);
2029 	set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2030 
2031 	netlink_table_grab();
2032 	if (!nl_table[unit].registered) {
2033 		nl_table[unit].groups = groups;
2034 		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2035 		nl_table[unit].module = module;
2036 		if (cfg) {
2037 			nl_table[unit].bind = cfg->bind;
2038 			nl_table[unit].unbind = cfg->unbind;
2039 			nl_table[unit].release = cfg->release;
2040 			nl_table[unit].flags = cfg->flags;
2041 		}
2042 		nl_table[unit].registered = 1;
2043 	} else {
2044 		kfree(listeners);
2045 		nl_table[unit].registered++;
2046 	}
2047 	netlink_table_ungrab();
2048 	return sk;
2049 
2050 out_sock_release:
2051 	kfree(listeners);
2052 	netlink_kernel_release(sk);
2053 	return NULL;
2054 
2055 out_sock_release_nosk:
2056 	sock_release(sock);
2057 	return NULL;
2058 }
2059 EXPORT_SYMBOL(__netlink_kernel_create);
2060 
2061 void
netlink_kernel_release(struct sock * sk)2062 netlink_kernel_release(struct sock *sk)
2063 {
2064 	if (sk == NULL || sk->sk_socket == NULL)
2065 		return;
2066 
2067 	sock_release(sk->sk_socket);
2068 }
2069 EXPORT_SYMBOL(netlink_kernel_release);
2070 
__netlink_change_ngroups(struct sock * sk,unsigned int groups)2071 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2072 {
2073 	struct listeners *new, *old;
2074 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2075 
2076 	if (groups < 32)
2077 		groups = 32;
2078 
2079 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2080 		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2081 		if (!new)
2082 			return -ENOMEM;
2083 		old = nl_deref_protected(tbl->listeners);
2084 		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2085 		rcu_assign_pointer(tbl->listeners, new);
2086 
2087 		kfree_rcu(old, rcu);
2088 	}
2089 	tbl->groups = groups;
2090 
2091 	return 0;
2092 }
2093 
2094 /**
2095  * netlink_change_ngroups - change number of multicast groups
2096  *
2097  * This changes the number of multicast groups that are available
2098  * on a certain netlink family. Note that it is not possible to
2099  * change the number of groups to below 32. Also note that it does
2100  * not implicitly call netlink_clear_multicast_users() when the
2101  * number of groups is reduced.
2102  *
2103  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2104  * @groups: The new number of groups.
2105  */
netlink_change_ngroups(struct sock * sk,unsigned int groups)2106 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2107 {
2108 	int err;
2109 
2110 	netlink_table_grab();
2111 	err = __netlink_change_ngroups(sk, groups);
2112 	netlink_table_ungrab();
2113 
2114 	return err;
2115 }
2116 
__netlink_clear_multicast_users(struct sock * ksk,unsigned int group)2117 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2118 {
2119 	struct sock *sk;
2120 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2121 	struct hlist_node *tmp;
2122 
2123 	sk_for_each_bound_safe(sk, tmp, &tbl->mc_list)
2124 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2125 }
2126 
2127 struct nlmsghdr *
__nlmsg_put(struct sk_buff * skb,u32 portid,u32 seq,int type,int len,int flags)2128 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2129 {
2130 	struct nlmsghdr *nlh;
2131 	int size = nlmsg_msg_size(len);
2132 
2133 	nlh = skb_put(skb, NLMSG_ALIGN(size));
2134 	nlh->nlmsg_type = type;
2135 	nlh->nlmsg_len = size;
2136 	nlh->nlmsg_flags = flags;
2137 	nlh->nlmsg_pid = portid;
2138 	nlh->nlmsg_seq = seq;
2139 	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2140 		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2141 	return nlh;
2142 }
2143 EXPORT_SYMBOL(__nlmsg_put);
2144 
2145 static size_t
netlink_ack_tlv_len(struct netlink_sock * nlk,int err,const struct netlink_ext_ack * extack)2146 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2147 		    const struct netlink_ext_ack *extack)
2148 {
2149 	size_t tlvlen;
2150 
2151 	if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2152 		return 0;
2153 
2154 	tlvlen = 0;
2155 	if (extack->_msg)
2156 		tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2157 	if (extack->cookie_len)
2158 		tlvlen += nla_total_size(extack->cookie_len);
2159 
2160 	/* Following attributes are only reported as error (not warning) */
2161 	if (!err)
2162 		return tlvlen;
2163 
2164 	if (extack->bad_attr)
2165 		tlvlen += nla_total_size(sizeof(u32));
2166 	if (extack->policy)
2167 		tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2168 	if (extack->miss_type)
2169 		tlvlen += nla_total_size(sizeof(u32));
2170 	if (extack->miss_nest)
2171 		tlvlen += nla_total_size(sizeof(u32));
2172 
2173 	return tlvlen;
2174 }
2175 
nlmsg_check_in_payload(const struct nlmsghdr * nlh,const void * addr)2176 static bool nlmsg_check_in_payload(const struct nlmsghdr *nlh, const void *addr)
2177 {
2178 	return !WARN_ON(addr < nlmsg_data(nlh) ||
2179 			addr - (const void *) nlh >= nlh->nlmsg_len);
2180 }
2181 
2182 static void
netlink_ack_tlv_fill(struct sk_buff * skb,const struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2183 netlink_ack_tlv_fill(struct sk_buff *skb, const struct nlmsghdr *nlh, int err,
2184 		     const struct netlink_ext_ack *extack)
2185 {
2186 	if (extack->_msg)
2187 		WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2188 	if (extack->cookie_len)
2189 		WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2190 				extack->cookie_len, extack->cookie));
2191 
2192 	if (!err)
2193 		return;
2194 
2195 	if (extack->bad_attr && nlmsg_check_in_payload(nlh, extack->bad_attr))
2196 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2197 				    (u8 *)extack->bad_attr - (const u8 *)nlh));
2198 	if (extack->policy)
2199 		netlink_policy_dump_write_attr(skb, extack->policy,
2200 					       NLMSGERR_ATTR_POLICY);
2201 	if (extack->miss_type)
2202 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2203 				    extack->miss_type));
2204 	if (extack->miss_nest && nlmsg_check_in_payload(nlh, extack->miss_nest))
2205 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2206 				    (u8 *)extack->miss_nest - (const u8 *)nlh));
2207 }
2208 
2209 /*
2210  * It looks a bit ugly.
2211  * It would be better to create kernel thread.
2212  */
2213 
netlink_dump_done(struct netlink_sock * nlk,struct sk_buff * skb,struct netlink_callback * cb,struct netlink_ext_ack * extack)2214 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2215 			     struct netlink_callback *cb,
2216 			     struct netlink_ext_ack *extack)
2217 {
2218 	struct nlmsghdr *nlh;
2219 	size_t extack_len;
2220 
2221 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2222 			       NLM_F_MULTI | cb->answer_flags);
2223 	if (WARN_ON(!nlh))
2224 		return -ENOBUFS;
2225 
2226 	nl_dump_check_consistent(cb, nlh);
2227 	memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2228 
2229 	extack_len = netlink_ack_tlv_len(nlk, nlk->dump_done_errno, extack);
2230 	if (extack_len) {
2231 		nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2232 		if (skb_tailroom(skb) >= extack_len) {
2233 			netlink_ack_tlv_fill(skb, cb->nlh,
2234 					     nlk->dump_done_errno, extack);
2235 			nlmsg_end(skb, nlh);
2236 		}
2237 	}
2238 
2239 	return 0;
2240 }
2241 
netlink_dump(struct sock * sk,bool lock_taken)2242 static int netlink_dump(struct sock *sk, bool lock_taken)
2243 {
2244 	struct netlink_sock *nlk = nlk_sk(sk);
2245 	struct netlink_ext_ack extack = {};
2246 	struct netlink_callback *cb;
2247 	struct sk_buff *skb = NULL;
2248 	size_t max_recvmsg_len;
2249 	struct module *module;
2250 	int err = -ENOBUFS;
2251 	int alloc_min_size;
2252 	int alloc_size;
2253 
2254 	if (!lock_taken)
2255 		mutex_lock(&nlk->nl_cb_mutex);
2256 	if (!nlk->cb_running) {
2257 		err = -EINVAL;
2258 		goto errout_skb;
2259 	}
2260 
2261 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2262 		goto errout_skb;
2263 
2264 	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2265 	 * required, but it makes sense to _attempt_ a 32KiB allocation
2266 	 * to reduce number of system calls on dump operations, if user
2267 	 * ever provided a big enough buffer.
2268 	 */
2269 	cb = &nlk->cb;
2270 	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2271 
2272 	max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2273 	if (alloc_min_size < max_recvmsg_len) {
2274 		alloc_size = max_recvmsg_len;
2275 		skb = alloc_skb(alloc_size,
2276 				(GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2277 				__GFP_NOWARN | __GFP_NORETRY);
2278 	}
2279 	if (!skb) {
2280 		alloc_size = alloc_min_size;
2281 		skb = alloc_skb(alloc_size, GFP_KERNEL);
2282 	}
2283 	if (!skb)
2284 		goto errout_skb;
2285 
2286 	/* Trim skb to allocated size. User is expected to provide buffer as
2287 	 * large as max(min_dump_alloc, 32KiB (max_recvmsg_len capped at
2288 	 * netlink_recvmsg())). dump will pack as many smaller messages as
2289 	 * could fit within the allocated skb. skb is typically allocated
2290 	 * with larger space than required (could be as much as near 2x the
2291 	 * requested size with align to next power of 2 approach). Allowing
2292 	 * dump to use the excess space makes it difficult for a user to have a
2293 	 * reasonable static buffer based on the expected largest dump of a
2294 	 * single netdev. The outcome is MSG_TRUNC error.
2295 	 */
2296 	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2297 
2298 	/* Make sure malicious BPF programs can not read unitialized memory
2299 	 * from skb->head -> skb->data
2300 	 */
2301 	skb_reset_network_header(skb);
2302 	skb_reset_mac_header(skb);
2303 
2304 	netlink_skb_set_owner_r(skb, sk);
2305 
2306 	if (nlk->dump_done_errno > 0) {
2307 		cb->extack = &extack;
2308 
2309 		nlk->dump_done_errno = cb->dump(skb, cb);
2310 
2311 		/* EMSGSIZE plus something already in the skb means
2312 		 * that there's more to dump but current skb has filled up.
2313 		 * If the callback really wants to return EMSGSIZE to user space
2314 		 * it needs to do so again, on the next cb->dump() call,
2315 		 * without putting data in the skb.
2316 		 */
2317 		if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2318 			nlk->dump_done_errno = skb->len;
2319 
2320 		cb->extack = NULL;
2321 	}
2322 
2323 	if (nlk->dump_done_errno > 0 ||
2324 	    skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2325 		mutex_unlock(&nlk->nl_cb_mutex);
2326 
2327 		if (sk_filter(sk, skb))
2328 			kfree_skb(skb);
2329 		else
2330 			__netlink_sendskb(sk, skb);
2331 		return 0;
2332 	}
2333 
2334 	if (netlink_dump_done(nlk, skb, cb, &extack))
2335 		goto errout_skb;
2336 
2337 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2338 	/* frag_list skb's data is used for compat tasks
2339 	 * and the regular skb's data for normal (non-compat) tasks.
2340 	 * See netlink_recvmsg().
2341 	 */
2342 	if (unlikely(skb_shinfo(skb)->frag_list)) {
2343 		if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2344 			goto errout_skb;
2345 	}
2346 #endif
2347 
2348 	if (sk_filter(sk, skb))
2349 		kfree_skb(skb);
2350 	else
2351 		__netlink_sendskb(sk, skb);
2352 
2353 	if (cb->done)
2354 		cb->done(cb);
2355 
2356 	WRITE_ONCE(nlk->cb_running, false);
2357 	module = cb->module;
2358 	skb = cb->skb;
2359 	mutex_unlock(&nlk->nl_cb_mutex);
2360 	module_put(module);
2361 	consume_skb(skb);
2362 	return 0;
2363 
2364 errout_skb:
2365 	mutex_unlock(&nlk->nl_cb_mutex);
2366 	kfree_skb(skb);
2367 	return err;
2368 }
2369 
__netlink_dump_start(struct sock * ssk,struct sk_buff * skb,const struct nlmsghdr * nlh,struct netlink_dump_control * control)2370 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2371 			 const struct nlmsghdr *nlh,
2372 			 struct netlink_dump_control *control)
2373 {
2374 	struct netlink_callback *cb;
2375 	struct netlink_sock *nlk;
2376 	struct sock *sk;
2377 	int ret;
2378 
2379 	refcount_inc(&skb->users);
2380 
2381 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2382 	if (sk == NULL) {
2383 		ret = -ECONNREFUSED;
2384 		goto error_free;
2385 	}
2386 
2387 	nlk = nlk_sk(sk);
2388 	mutex_lock(&nlk->nl_cb_mutex);
2389 	/* A dump is in progress... */
2390 	if (nlk->cb_running) {
2391 		ret = -EBUSY;
2392 		goto error_unlock;
2393 	}
2394 	/* add reference of module which cb->dump belongs to */
2395 	if (!try_module_get(control->module)) {
2396 		ret = -EPROTONOSUPPORT;
2397 		goto error_unlock;
2398 	}
2399 
2400 	cb = &nlk->cb;
2401 	memset(cb, 0, sizeof(*cb));
2402 	cb->dump = control->dump;
2403 	cb->done = control->done;
2404 	cb->nlh = nlh;
2405 	cb->data = control->data;
2406 	cb->module = control->module;
2407 	cb->min_dump_alloc = control->min_dump_alloc;
2408 	cb->flags = control->flags;
2409 	cb->skb = skb;
2410 
2411 	cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2412 
2413 	if (control->start) {
2414 		cb->extack = control->extack;
2415 		ret = control->start(cb);
2416 		cb->extack = NULL;
2417 		if (ret)
2418 			goto error_put;
2419 	}
2420 
2421 	WRITE_ONCE(nlk->cb_running, true);
2422 	nlk->dump_done_errno = INT_MAX;
2423 
2424 	ret = netlink_dump(sk, true);
2425 
2426 	sock_put(sk);
2427 
2428 	if (ret)
2429 		return ret;
2430 
2431 	/* We successfully started a dump, by returning -EINTR we
2432 	 * signal not to send ACK even if it was requested.
2433 	 */
2434 	return -EINTR;
2435 
2436 error_put:
2437 	module_put(control->module);
2438 error_unlock:
2439 	sock_put(sk);
2440 	mutex_unlock(&nlk->nl_cb_mutex);
2441 error_free:
2442 	kfree_skb(skb);
2443 	return ret;
2444 }
2445 EXPORT_SYMBOL(__netlink_dump_start);
2446 
netlink_ack(struct sk_buff * in_skb,struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2447 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2448 		 const struct netlink_ext_ack *extack)
2449 {
2450 	struct sk_buff *skb;
2451 	struct nlmsghdr *rep;
2452 	struct nlmsgerr *errmsg;
2453 	size_t payload = sizeof(*errmsg);
2454 	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2455 	unsigned int flags = 0;
2456 	size_t tlvlen;
2457 
2458 	/* Error messages get the original request appened, unless the user
2459 	 * requests to cap the error message, and get extra error data if
2460 	 * requested.
2461 	 */
2462 	if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2463 		payload += nlmsg_len(nlh);
2464 	else
2465 		flags |= NLM_F_CAPPED;
2466 
2467 	tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2468 	if (tlvlen)
2469 		flags |= NLM_F_ACK_TLVS;
2470 
2471 	skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2472 	if (!skb)
2473 		goto err_skb;
2474 
2475 	rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2476 			NLMSG_ERROR, sizeof(*errmsg), flags);
2477 	if (!rep)
2478 		goto err_bad_put;
2479 	errmsg = nlmsg_data(rep);
2480 	errmsg->error = err;
2481 	errmsg->msg = *nlh;
2482 
2483 	if (!(flags & NLM_F_CAPPED)) {
2484 		if (!nlmsg_append(skb, nlmsg_len(nlh)))
2485 			goto err_bad_put;
2486 
2487 		memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2488 		       nlmsg_len(nlh));
2489 	}
2490 
2491 	if (tlvlen)
2492 		netlink_ack_tlv_fill(skb, nlh, err, extack);
2493 
2494 	nlmsg_end(skb, rep);
2495 
2496 	nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2497 
2498 	return;
2499 
2500 err_bad_put:
2501 	nlmsg_free(skb);
2502 err_skb:
2503 	WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2504 	sk_error_report(NETLINK_CB(in_skb).sk);
2505 }
2506 EXPORT_SYMBOL(netlink_ack);
2507 
netlink_rcv_skb(struct sk_buff * skb,int (* cb)(struct sk_buff *,struct nlmsghdr *,struct netlink_ext_ack *))2508 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2509 						   struct nlmsghdr *,
2510 						   struct netlink_ext_ack *))
2511 {
2512 	struct netlink_ext_ack extack;
2513 	struct nlmsghdr *nlh;
2514 	int err;
2515 
2516 	while (skb->len >= nlmsg_total_size(0)) {
2517 		int msglen;
2518 
2519 		memset(&extack, 0, sizeof(extack));
2520 		nlh = nlmsg_hdr(skb);
2521 		err = 0;
2522 
2523 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2524 			return 0;
2525 
2526 		/* Only requests are handled by the kernel */
2527 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2528 			goto ack;
2529 
2530 		/* Skip control messages */
2531 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2532 			goto ack;
2533 
2534 		err = cb(skb, nlh, &extack);
2535 		if (err == -EINTR)
2536 			goto skip;
2537 
2538 ack:
2539 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2540 			netlink_ack(skb, nlh, err, &extack);
2541 
2542 skip:
2543 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2544 		if (msglen > skb->len)
2545 			msglen = skb->len;
2546 		skb_pull(skb, msglen);
2547 	}
2548 
2549 	return 0;
2550 }
2551 EXPORT_SYMBOL(netlink_rcv_skb);
2552 
2553 /**
2554  * nlmsg_notify - send a notification netlink message
2555  * @sk: netlink socket to use
2556  * @skb: notification message
2557  * @portid: destination netlink portid for reports or 0
2558  * @group: destination multicast group or 0
2559  * @report: 1 to report back, 0 to disable
2560  * @flags: allocation flags
2561  */
nlmsg_notify(struct sock * sk,struct sk_buff * skb,u32 portid,unsigned int group,int report,gfp_t flags)2562 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2563 		 unsigned int group, int report, gfp_t flags)
2564 {
2565 	int err = 0;
2566 
2567 	if (group) {
2568 		int exclude_portid = 0;
2569 
2570 		if (report) {
2571 			refcount_inc(&skb->users);
2572 			exclude_portid = portid;
2573 		}
2574 
2575 		/* errors reported via destination sk->sk_err, but propagate
2576 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2577 		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2578 		if (err == -ESRCH)
2579 			err = 0;
2580 	}
2581 
2582 	if (report) {
2583 		int err2;
2584 
2585 		err2 = nlmsg_unicast(sk, skb, portid);
2586 		if (!err)
2587 			err = err2;
2588 	}
2589 
2590 	return err;
2591 }
2592 EXPORT_SYMBOL(nlmsg_notify);
2593 
2594 #ifdef CONFIG_PROC_FS
2595 struct nl_seq_iter {
2596 	struct seq_net_private p;
2597 	struct rhashtable_iter hti;
2598 	int link;
2599 };
2600 
netlink_walk_start(struct nl_seq_iter * iter)2601 static void netlink_walk_start(struct nl_seq_iter *iter)
2602 {
2603 	rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2604 	rhashtable_walk_start(&iter->hti);
2605 }
2606 
netlink_walk_stop(struct nl_seq_iter * iter)2607 static void netlink_walk_stop(struct nl_seq_iter *iter)
2608 {
2609 	rhashtable_walk_stop(&iter->hti);
2610 	rhashtable_walk_exit(&iter->hti);
2611 }
2612 
__netlink_seq_next(struct seq_file * seq)2613 static void *__netlink_seq_next(struct seq_file *seq)
2614 {
2615 	struct nl_seq_iter *iter = seq->private;
2616 	struct netlink_sock *nlk;
2617 
2618 	do {
2619 		for (;;) {
2620 			nlk = rhashtable_walk_next(&iter->hti);
2621 
2622 			if (IS_ERR(nlk)) {
2623 				if (PTR_ERR(nlk) == -EAGAIN)
2624 					continue;
2625 
2626 				return nlk;
2627 			}
2628 
2629 			if (nlk)
2630 				break;
2631 
2632 			netlink_walk_stop(iter);
2633 			if (++iter->link >= MAX_LINKS)
2634 				return NULL;
2635 
2636 			netlink_walk_start(iter);
2637 		}
2638 	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2639 
2640 	return nlk;
2641 }
2642 
netlink_seq_start(struct seq_file * seq,loff_t * posp)2643 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2644 	__acquires(RCU)
2645 {
2646 	struct nl_seq_iter *iter = seq->private;
2647 	void *obj = SEQ_START_TOKEN;
2648 	loff_t pos;
2649 
2650 	iter->link = 0;
2651 
2652 	netlink_walk_start(iter);
2653 
2654 	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2655 		obj = __netlink_seq_next(seq);
2656 
2657 	return obj;
2658 }
2659 
netlink_seq_next(struct seq_file * seq,void * v,loff_t * pos)2660 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2661 {
2662 	++*pos;
2663 	return __netlink_seq_next(seq);
2664 }
2665 
netlink_native_seq_stop(struct seq_file * seq,void * v)2666 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2667 {
2668 	struct nl_seq_iter *iter = seq->private;
2669 
2670 	if (iter->link >= MAX_LINKS)
2671 		return;
2672 
2673 	netlink_walk_stop(iter);
2674 }
2675 
2676 
netlink_native_seq_show(struct seq_file * seq,void * v)2677 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2678 {
2679 	if (v == SEQ_START_TOKEN) {
2680 		seq_puts(seq,
2681 			 "sk               Eth Pid        Groups   "
2682 			 "Rmem     Wmem     Dump  Locks    Drops    Inode\n");
2683 	} else {
2684 		struct sock *s = v;
2685 		struct netlink_sock *nlk = nlk_sk(s);
2686 
2687 		seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2688 			   s,
2689 			   s->sk_protocol,
2690 			   nlk->portid,
2691 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2692 			   sk_rmem_alloc_get(s),
2693 			   sk_wmem_alloc_get(s),
2694 			   READ_ONCE(nlk->cb_running),
2695 			   refcount_read(&s->sk_refcnt),
2696 			   atomic_read(&s->sk_drops),
2697 			   sock_i_ino(s)
2698 			);
2699 
2700 	}
2701 	return 0;
2702 }
2703 
2704 #ifdef CONFIG_BPF_SYSCALL
2705 struct bpf_iter__netlink {
2706 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2707 	__bpf_md_ptr(struct netlink_sock *, sk);
2708 };
2709 
DEFINE_BPF_ITER_FUNC(netlink,struct bpf_iter_meta * meta,struct netlink_sock * sk)2710 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2711 
2712 static int netlink_prog_seq_show(struct bpf_prog *prog,
2713 				  struct bpf_iter_meta *meta,
2714 				  void *v)
2715 {
2716 	struct bpf_iter__netlink ctx;
2717 
2718 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2719 	ctx.meta = meta;
2720 	ctx.sk = nlk_sk((struct sock *)v);
2721 	return bpf_iter_run_prog(prog, &ctx);
2722 }
2723 
netlink_seq_show(struct seq_file * seq,void * v)2724 static int netlink_seq_show(struct seq_file *seq, void *v)
2725 {
2726 	struct bpf_iter_meta meta;
2727 	struct bpf_prog *prog;
2728 
2729 	meta.seq = seq;
2730 	prog = bpf_iter_get_info(&meta, false);
2731 	if (!prog)
2732 		return netlink_native_seq_show(seq, v);
2733 
2734 	if (v != SEQ_START_TOKEN)
2735 		return netlink_prog_seq_show(prog, &meta, v);
2736 
2737 	return 0;
2738 }
2739 
netlink_seq_stop(struct seq_file * seq,void * v)2740 static void netlink_seq_stop(struct seq_file *seq, void *v)
2741 {
2742 	struct bpf_iter_meta meta;
2743 	struct bpf_prog *prog;
2744 
2745 	if (!v) {
2746 		meta.seq = seq;
2747 		prog = bpf_iter_get_info(&meta, true);
2748 		if (prog)
2749 			(void)netlink_prog_seq_show(prog, &meta, v);
2750 	}
2751 
2752 	netlink_native_seq_stop(seq, v);
2753 }
2754 #else
netlink_seq_show(struct seq_file * seq,void * v)2755 static int netlink_seq_show(struct seq_file *seq, void *v)
2756 {
2757 	return netlink_native_seq_show(seq, v);
2758 }
2759 
netlink_seq_stop(struct seq_file * seq,void * v)2760 static void netlink_seq_stop(struct seq_file *seq, void *v)
2761 {
2762 	netlink_native_seq_stop(seq, v);
2763 }
2764 #endif
2765 
2766 static const struct seq_operations netlink_seq_ops = {
2767 	.start  = netlink_seq_start,
2768 	.next   = netlink_seq_next,
2769 	.stop   = netlink_seq_stop,
2770 	.show   = netlink_seq_show,
2771 };
2772 #endif
2773 
netlink_register_notifier(struct notifier_block * nb)2774 int netlink_register_notifier(struct notifier_block *nb)
2775 {
2776 	return blocking_notifier_chain_register(&netlink_chain, nb);
2777 }
2778 EXPORT_SYMBOL(netlink_register_notifier);
2779 
netlink_unregister_notifier(struct notifier_block * nb)2780 int netlink_unregister_notifier(struct notifier_block *nb)
2781 {
2782 	return blocking_notifier_chain_unregister(&netlink_chain, nb);
2783 }
2784 EXPORT_SYMBOL(netlink_unregister_notifier);
2785 
2786 static const struct proto_ops netlink_ops = {
2787 	.family =	PF_NETLINK,
2788 	.owner =	THIS_MODULE,
2789 	.release =	netlink_release,
2790 	.bind =		netlink_bind,
2791 	.connect =	netlink_connect,
2792 	.socketpair =	sock_no_socketpair,
2793 	.accept =	sock_no_accept,
2794 	.getname =	netlink_getname,
2795 	.poll =		datagram_poll,
2796 	.ioctl =	netlink_ioctl,
2797 	.listen =	sock_no_listen,
2798 	.shutdown =	sock_no_shutdown,
2799 	.setsockopt =	netlink_setsockopt,
2800 	.getsockopt =	netlink_getsockopt,
2801 	.sendmsg =	netlink_sendmsg,
2802 	.recvmsg =	netlink_recvmsg,
2803 	.mmap =		sock_no_mmap,
2804 };
2805 
2806 static const struct net_proto_family netlink_family_ops = {
2807 	.family = PF_NETLINK,
2808 	.create = netlink_create,
2809 	.owner	= THIS_MODULE,	/* for consistency 8) */
2810 };
2811 
netlink_net_init(struct net * net)2812 static int __net_init netlink_net_init(struct net *net)
2813 {
2814 #ifdef CONFIG_PROC_FS
2815 	if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2816 			sizeof(struct nl_seq_iter)))
2817 		return -ENOMEM;
2818 #endif
2819 	return 0;
2820 }
2821 
netlink_net_exit(struct net * net)2822 static void __net_exit netlink_net_exit(struct net *net)
2823 {
2824 #ifdef CONFIG_PROC_FS
2825 	remove_proc_entry("netlink", net->proc_net);
2826 #endif
2827 }
2828 
netlink_add_usersock_entry(void)2829 static void __init netlink_add_usersock_entry(void)
2830 {
2831 	struct listeners *listeners;
2832 	int groups = 32;
2833 
2834 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2835 	if (!listeners)
2836 		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2837 
2838 	netlink_table_grab();
2839 
2840 	nl_table[NETLINK_USERSOCK].groups = groups;
2841 	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2842 	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2843 	nl_table[NETLINK_USERSOCK].registered = 1;
2844 	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2845 
2846 	netlink_table_ungrab();
2847 }
2848 
2849 static struct pernet_operations __net_initdata netlink_net_ops = {
2850 	.init = netlink_net_init,
2851 	.exit = netlink_net_exit,
2852 };
2853 
netlink_hash(const void * data,u32 len,u32 seed)2854 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2855 {
2856 	const struct netlink_sock *nlk = data;
2857 	struct netlink_compare_arg arg;
2858 
2859 	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2860 	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2861 }
2862 
2863 static const struct rhashtable_params netlink_rhashtable_params = {
2864 	.head_offset = offsetof(struct netlink_sock, node),
2865 	.key_len = netlink_compare_arg_len,
2866 	.obj_hashfn = netlink_hash,
2867 	.obj_cmpfn = netlink_compare,
2868 	.automatic_shrinking = true,
2869 };
2870 
2871 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2872 BTF_ID_LIST(btf_netlink_sock_id)
2873 BTF_ID(struct, netlink_sock)
2874 
2875 static const struct bpf_iter_seq_info netlink_seq_info = {
2876 	.seq_ops		= &netlink_seq_ops,
2877 	.init_seq_private	= bpf_iter_init_seq_net,
2878 	.fini_seq_private	= bpf_iter_fini_seq_net,
2879 	.seq_priv_size		= sizeof(struct nl_seq_iter),
2880 };
2881 
2882 static struct bpf_iter_reg netlink_reg_info = {
2883 	.target			= "netlink",
2884 	.ctx_arg_info_size	= 1,
2885 	.ctx_arg_info		= {
2886 		{ offsetof(struct bpf_iter__netlink, sk),
2887 		  PTR_TO_BTF_ID_OR_NULL },
2888 	},
2889 	.seq_info		= &netlink_seq_info,
2890 };
2891 
bpf_iter_register(void)2892 static int __init bpf_iter_register(void)
2893 {
2894 	netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2895 	return bpf_iter_reg_target(&netlink_reg_info);
2896 }
2897 #endif
2898 
netlink_proto_init(void)2899 static int __init netlink_proto_init(void)
2900 {
2901 	int i;
2902 	int err = proto_register(&netlink_proto, 0);
2903 
2904 	if (err != 0)
2905 		goto out;
2906 
2907 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2908 	err = bpf_iter_register();
2909 	if (err)
2910 		goto out;
2911 #endif
2912 
2913 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2914 
2915 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2916 	if (!nl_table)
2917 		goto panic;
2918 
2919 	for (i = 0; i < MAX_LINKS; i++) {
2920 		if (rhashtable_init(&nl_table[i].hash,
2921 				    &netlink_rhashtable_params) < 0)
2922 			goto panic;
2923 	}
2924 
2925 	netlink_add_usersock_entry();
2926 
2927 	sock_register(&netlink_family_ops);
2928 	register_pernet_subsys(&netlink_net_ops);
2929 	register_pernet_subsys(&netlink_tap_net_ops);
2930 	/* The netlink device handler may be needed early. */
2931 	rtnetlink_init();
2932 out:
2933 	return err;
2934 panic:
2935 	panic("netlink_init: Cannot allocate nl_table\n");
2936 }
2937 
2938 core_initcall(netlink_proto_init);
2939