xref: /linux/net/netlink/af_netlink.c (revision 65ae975e97d5aab3ee9dc5ec701b12090572ed43)
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 	}
775 
776 	module_put(nlk->module);
777 
778 	if (netlink_is_kernel(sk)) {
779 		netlink_table_grab();
780 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
781 		if (--nl_table[sk->sk_protocol].registered == 0) {
782 			struct listeners *old;
783 
784 			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
785 			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
786 			kfree_rcu(old, rcu);
787 			nl_table[sk->sk_protocol].module = NULL;
788 			nl_table[sk->sk_protocol].bind = NULL;
789 			nl_table[sk->sk_protocol].unbind = NULL;
790 			nl_table[sk->sk_protocol].flags = 0;
791 			nl_table[sk->sk_protocol].registered = 0;
792 		}
793 		netlink_table_ungrab();
794 	}
795 
796 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
797 
798 	/* Because struct net might disappear soon, do not keep a pointer. */
799 	if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
800 		__netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
801 		/* Because of deferred_put_nlk_sk and use of work queue,
802 		 * it is possible  netns will be freed before this socket.
803 		 */
804 		sock_net_set(sk, &init_net);
805 		__netns_tracker_alloc(&init_net, &sk->ns_tracker,
806 				      false, GFP_KERNEL);
807 	}
808 	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
809 	return 0;
810 }
811 
netlink_autobind(struct socket * sock)812 static int netlink_autobind(struct socket *sock)
813 {
814 	struct sock *sk = sock->sk;
815 	struct net *net = sock_net(sk);
816 	struct netlink_table *table = &nl_table[sk->sk_protocol];
817 	s32 portid = task_tgid_vnr(current);
818 	int err;
819 	s32 rover = -4096;
820 	bool ok;
821 
822 retry:
823 	cond_resched();
824 	rcu_read_lock();
825 	ok = !__netlink_lookup(table, portid, net);
826 	rcu_read_unlock();
827 	if (!ok) {
828 		/* Bind collision, search negative portid values. */
829 		if (rover == -4096)
830 			/* rover will be in range [S32_MIN, -4097] */
831 			rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
832 		else if (rover >= -4096)
833 			rover = -4097;
834 		portid = rover--;
835 		goto retry;
836 	}
837 
838 	err = netlink_insert(sk, portid);
839 	if (err == -EADDRINUSE)
840 		goto retry;
841 
842 	/* If 2 threads race to autobind, that is fine.  */
843 	if (err == -EBUSY)
844 		err = 0;
845 
846 	return err;
847 }
848 
849 /**
850  * __netlink_ns_capable - General netlink message capability test
851  * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
852  * @user_ns: The user namespace of the capability to use
853  * @cap: The capability to use
854  *
855  * Test to see if the opener of the socket we received the message
856  * from had when the netlink socket was created and the sender of the
857  * message has the capability @cap in the user namespace @user_ns.
858  */
__netlink_ns_capable(const struct netlink_skb_parms * nsp,struct user_namespace * user_ns,int cap)859 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
860 			struct user_namespace *user_ns, int cap)
861 {
862 	return ((nsp->flags & NETLINK_SKB_DST) ||
863 		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
864 		ns_capable(user_ns, cap);
865 }
866 EXPORT_SYMBOL(__netlink_ns_capable);
867 
868 /**
869  * netlink_ns_capable - General netlink message capability test
870  * @skb: socket buffer holding a netlink command from userspace
871  * @user_ns: The user namespace of the capability to use
872  * @cap: The capability to use
873  *
874  * Test to see if the opener of the socket we received the message
875  * from had when the netlink socket was created and the sender of the
876  * message has the capability @cap in the user namespace @user_ns.
877  */
netlink_ns_capable(const struct sk_buff * skb,struct user_namespace * user_ns,int cap)878 bool netlink_ns_capable(const struct sk_buff *skb,
879 			struct user_namespace *user_ns, int cap)
880 {
881 	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
882 }
883 EXPORT_SYMBOL(netlink_ns_capable);
884 
885 /**
886  * netlink_capable - Netlink global message capability test
887  * @skb: socket buffer holding a netlink command from userspace
888  * @cap: The capability to use
889  *
890  * Test to see if the opener of the socket we received the message
891  * from had when the netlink socket was created and the sender of the
892  * message has the capability @cap in all user namespaces.
893  */
netlink_capable(const struct sk_buff * skb,int cap)894 bool netlink_capable(const struct sk_buff *skb, int cap)
895 {
896 	return netlink_ns_capable(skb, &init_user_ns, cap);
897 }
898 EXPORT_SYMBOL(netlink_capable);
899 
900 /**
901  * netlink_net_capable - Netlink network namespace message capability test
902  * @skb: socket buffer holding a netlink command from userspace
903  * @cap: The capability to use
904  *
905  * Test to see if the opener of the socket we received the message
906  * from had when the netlink socket was created and the sender of the
907  * message has the capability @cap over the network namespace of
908  * the socket we received the message from.
909  */
netlink_net_capable(const struct sk_buff * skb,int cap)910 bool netlink_net_capable(const struct sk_buff *skb, int cap)
911 {
912 	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
913 }
914 EXPORT_SYMBOL(netlink_net_capable);
915 
netlink_allowed(const struct socket * sock,unsigned int flag)916 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
917 {
918 	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
919 		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
920 }
921 
922 static void
netlink_update_subscriptions(struct sock * sk,unsigned int subscriptions)923 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
924 {
925 	struct netlink_sock *nlk = nlk_sk(sk);
926 
927 	if (nlk->subscriptions && !subscriptions)
928 		__sk_del_bind_node(sk);
929 	else if (!nlk->subscriptions && subscriptions)
930 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
931 	nlk->subscriptions = subscriptions;
932 }
933 
netlink_realloc_groups(struct sock * sk)934 static int netlink_realloc_groups(struct sock *sk)
935 {
936 	struct netlink_sock *nlk = nlk_sk(sk);
937 	unsigned int groups;
938 	unsigned long *new_groups;
939 	int err = 0;
940 
941 	netlink_table_grab();
942 
943 	groups = nl_table[sk->sk_protocol].groups;
944 	if (!nl_table[sk->sk_protocol].registered) {
945 		err = -ENOENT;
946 		goto out_unlock;
947 	}
948 
949 	if (nlk->ngroups >= groups)
950 		goto out_unlock;
951 
952 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
953 	if (new_groups == NULL) {
954 		err = -ENOMEM;
955 		goto out_unlock;
956 	}
957 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
958 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
959 
960 	nlk->groups = new_groups;
961 	nlk->ngroups = groups;
962  out_unlock:
963 	netlink_table_ungrab();
964 	return err;
965 }
966 
netlink_undo_bind(int group,long unsigned int groups,struct sock * sk)967 static void netlink_undo_bind(int group, long unsigned int groups,
968 			      struct sock *sk)
969 {
970 	struct netlink_sock *nlk = nlk_sk(sk);
971 	int undo;
972 
973 	if (!nlk->netlink_unbind)
974 		return;
975 
976 	for (undo = 0; undo < group; undo++)
977 		if (test_bit(undo, &groups))
978 			nlk->netlink_unbind(sock_net(sk), undo + 1);
979 }
980 
netlink_bind(struct socket * sock,struct sockaddr * addr,int addr_len)981 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
982 			int addr_len)
983 {
984 	struct sock *sk = sock->sk;
985 	struct net *net = sock_net(sk);
986 	struct netlink_sock *nlk = nlk_sk(sk);
987 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
988 	int err = 0;
989 	unsigned long groups;
990 	bool bound;
991 
992 	if (addr_len < sizeof(struct sockaddr_nl))
993 		return -EINVAL;
994 
995 	if (nladdr->nl_family != AF_NETLINK)
996 		return -EINVAL;
997 	groups = nladdr->nl_groups;
998 
999 	/* Only superuser is allowed to listen multicasts */
1000 	if (groups) {
1001 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1002 			return -EPERM;
1003 		err = netlink_realloc_groups(sk);
1004 		if (err)
1005 			return err;
1006 	}
1007 
1008 	if (nlk->ngroups < BITS_PER_LONG)
1009 		groups &= (1UL << nlk->ngroups) - 1;
1010 
1011 	/* Paired with WRITE_ONCE() in netlink_insert() */
1012 	bound = READ_ONCE(nlk->bound);
1013 	if (bound) {
1014 		/* Ensure nlk->portid is up-to-date. */
1015 		smp_rmb();
1016 
1017 		if (nladdr->nl_pid != nlk->portid)
1018 			return -EINVAL;
1019 	}
1020 
1021 	if (nlk->netlink_bind && groups) {
1022 		int group;
1023 
1024 		/* nl_groups is a u32, so cap the maximum groups we can bind */
1025 		for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1026 			if (!test_bit(group, &groups))
1027 				continue;
1028 			err = nlk->netlink_bind(net, group + 1);
1029 			if (!err)
1030 				continue;
1031 			netlink_undo_bind(group, groups, sk);
1032 			return err;
1033 		}
1034 	}
1035 
1036 	/* No need for barriers here as we return to user-space without
1037 	 * using any of the bound attributes.
1038 	 */
1039 	netlink_lock_table();
1040 	if (!bound) {
1041 		err = nladdr->nl_pid ?
1042 			netlink_insert(sk, nladdr->nl_pid) :
1043 			netlink_autobind(sock);
1044 		if (err) {
1045 			netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1046 			goto unlock;
1047 		}
1048 	}
1049 
1050 	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1051 		goto unlock;
1052 	netlink_unlock_table();
1053 
1054 	netlink_table_grab();
1055 	netlink_update_subscriptions(sk, nlk->subscriptions +
1056 					 hweight32(groups) -
1057 					 hweight32(nlk->groups[0]));
1058 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1059 	netlink_update_listeners(sk);
1060 	netlink_table_ungrab();
1061 
1062 	return 0;
1063 
1064 unlock:
1065 	netlink_unlock_table();
1066 	return err;
1067 }
1068 
netlink_connect(struct socket * sock,struct sockaddr * addr,int alen,int flags)1069 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1070 			   int alen, int flags)
1071 {
1072 	int err = 0;
1073 	struct sock *sk = sock->sk;
1074 	struct netlink_sock *nlk = nlk_sk(sk);
1075 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1076 
1077 	if (alen < sizeof(addr->sa_family))
1078 		return -EINVAL;
1079 
1080 	if (addr->sa_family == AF_UNSPEC) {
1081 		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1082 		WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1083 		/* dst_portid and dst_group can be read locklessly */
1084 		WRITE_ONCE(nlk->dst_portid, 0);
1085 		WRITE_ONCE(nlk->dst_group, 0);
1086 		return 0;
1087 	}
1088 	if (addr->sa_family != AF_NETLINK)
1089 		return -EINVAL;
1090 
1091 	if (alen < sizeof(struct sockaddr_nl))
1092 		return -EINVAL;
1093 
1094 	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1095 	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1096 		return -EPERM;
1097 
1098 	/* No need for barriers here as we return to user-space without
1099 	 * using any of the bound attributes.
1100 	 * Paired with WRITE_ONCE() in netlink_insert().
1101 	 */
1102 	if (!READ_ONCE(nlk->bound))
1103 		err = netlink_autobind(sock);
1104 
1105 	if (err == 0) {
1106 		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1107 		WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1108 		/* dst_portid and dst_group can be read locklessly */
1109 		WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1110 		WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1111 	}
1112 
1113 	return err;
1114 }
1115 
netlink_getname(struct socket * sock,struct sockaddr * addr,int peer)1116 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1117 			   int peer)
1118 {
1119 	struct sock *sk = sock->sk;
1120 	struct netlink_sock *nlk = nlk_sk(sk);
1121 	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1122 
1123 	nladdr->nl_family = AF_NETLINK;
1124 	nladdr->nl_pad = 0;
1125 
1126 	if (peer) {
1127 		/* Paired with WRITE_ONCE() in netlink_connect() */
1128 		nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1129 		nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1130 	} else {
1131 		/* Paired with WRITE_ONCE() in netlink_insert() */
1132 		nladdr->nl_pid = READ_ONCE(nlk->portid);
1133 		netlink_lock_table();
1134 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1135 		netlink_unlock_table();
1136 	}
1137 	return sizeof(*nladdr);
1138 }
1139 
netlink_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1140 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1141 			 unsigned long arg)
1142 {
1143 	/* try to hand this ioctl down to the NIC drivers.
1144 	 */
1145 	return -ENOIOCTLCMD;
1146 }
1147 
netlink_getsockbyportid(struct sock * ssk,u32 portid)1148 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1149 {
1150 	struct sock *sock;
1151 	struct netlink_sock *nlk;
1152 
1153 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1154 	if (!sock)
1155 		return ERR_PTR(-ECONNREFUSED);
1156 
1157 	/* Don't bother queuing skb if kernel socket has no input function */
1158 	nlk = nlk_sk(sock);
1159 	/* dst_portid and sk_state can be changed in netlink_connect() */
1160 	if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1161 	    READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1162 		sock_put(sock);
1163 		return ERR_PTR(-ECONNREFUSED);
1164 	}
1165 	return sock;
1166 }
1167 
netlink_getsockbyfd(int fd)1168 struct sock *netlink_getsockbyfd(int fd)
1169 {
1170 	CLASS(fd, f)(fd);
1171 	struct inode *inode;
1172 	struct sock *sock;
1173 
1174 	if (fd_empty(f))
1175 		return ERR_PTR(-EBADF);
1176 
1177 	inode = file_inode(fd_file(f));
1178 	if (!S_ISSOCK(inode->i_mode))
1179 		return ERR_PTR(-ENOTSOCK);
1180 
1181 	sock = SOCKET_I(inode)->sk;
1182 	if (sock->sk_family != AF_NETLINK)
1183 		return ERR_PTR(-EINVAL);
1184 
1185 	sock_hold(sock);
1186 	return sock;
1187 }
1188 
netlink_alloc_large_skb(unsigned int size,int broadcast)1189 struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1190 {
1191 	size_t head_size = SKB_HEAD_ALIGN(size);
1192 	struct sk_buff *skb;
1193 	void *data;
1194 
1195 	if (head_size <= PAGE_SIZE || broadcast)
1196 		return alloc_skb(size, GFP_KERNEL);
1197 
1198 	data = kvmalloc(head_size, GFP_KERNEL);
1199 	if (!data)
1200 		return NULL;
1201 
1202 	skb = __build_skb(data, head_size);
1203 	if (!skb)
1204 		kvfree(data);
1205 	else if (is_vmalloc_addr(data))
1206 		skb->destructor = netlink_skb_destructor;
1207 
1208 	return skb;
1209 }
1210 
1211 /*
1212  * Attach a skb to a netlink socket.
1213  * The caller must hold a reference to the destination socket. On error, the
1214  * reference is dropped. The skb is not send to the destination, just all
1215  * all error checks are performed and memory in the queue is reserved.
1216  * Return values:
1217  * < 0: error. skb freed, reference to sock dropped.
1218  * 0: continue
1219  * 1: repeat lookup - reference dropped while waiting for socket memory.
1220  */
netlink_attachskb(struct sock * sk,struct sk_buff * skb,long * timeo,struct sock * ssk)1221 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1222 		      long *timeo, struct sock *ssk)
1223 {
1224 	struct netlink_sock *nlk;
1225 
1226 	nlk = nlk_sk(sk);
1227 
1228 	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1229 	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1230 		DECLARE_WAITQUEUE(wait, current);
1231 		if (!*timeo) {
1232 			if (!ssk || netlink_is_kernel(ssk))
1233 				netlink_overrun(sk);
1234 			sock_put(sk);
1235 			kfree_skb(skb);
1236 			return -EAGAIN;
1237 		}
1238 
1239 		__set_current_state(TASK_INTERRUPTIBLE);
1240 		add_wait_queue(&nlk->wait, &wait);
1241 
1242 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1243 		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1244 		    !sock_flag(sk, SOCK_DEAD))
1245 			*timeo = schedule_timeout(*timeo);
1246 
1247 		__set_current_state(TASK_RUNNING);
1248 		remove_wait_queue(&nlk->wait, &wait);
1249 		sock_put(sk);
1250 
1251 		if (signal_pending(current)) {
1252 			kfree_skb(skb);
1253 			return sock_intr_errno(*timeo);
1254 		}
1255 		return 1;
1256 	}
1257 	netlink_skb_set_owner_r(skb, sk);
1258 	return 0;
1259 }
1260 
__netlink_sendskb(struct sock * sk,struct sk_buff * skb)1261 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1262 {
1263 	int len = skb->len;
1264 
1265 	netlink_deliver_tap(sock_net(sk), skb);
1266 
1267 	skb_queue_tail(&sk->sk_receive_queue, skb);
1268 	sk->sk_data_ready(sk);
1269 	return len;
1270 }
1271 
netlink_sendskb(struct sock * sk,struct sk_buff * skb)1272 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1273 {
1274 	int len = __netlink_sendskb(sk, skb);
1275 
1276 	sock_put(sk);
1277 	return len;
1278 }
1279 
netlink_detachskb(struct sock * sk,struct sk_buff * skb)1280 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1281 {
1282 	kfree_skb(skb);
1283 	sock_put(sk);
1284 }
1285 
netlink_trim(struct sk_buff * skb,gfp_t allocation)1286 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1287 {
1288 	int delta;
1289 
1290 	WARN_ON(skb->sk != NULL);
1291 	delta = skb->end - skb->tail;
1292 	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1293 		return skb;
1294 
1295 	if (skb_shared(skb)) {
1296 		struct sk_buff *nskb = skb_clone(skb, allocation);
1297 		if (!nskb)
1298 			return skb;
1299 		consume_skb(skb);
1300 		skb = nskb;
1301 	}
1302 
1303 	pskb_expand_head(skb, 0, -delta,
1304 			 (allocation & ~__GFP_DIRECT_RECLAIM) |
1305 			 __GFP_NOWARN | __GFP_NORETRY);
1306 	return skb;
1307 }
1308 
netlink_unicast_kernel(struct sock * sk,struct sk_buff * skb,struct sock * ssk)1309 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1310 				  struct sock *ssk)
1311 {
1312 	int ret;
1313 	struct netlink_sock *nlk = nlk_sk(sk);
1314 
1315 	ret = -ECONNREFUSED;
1316 	if (nlk->netlink_rcv != NULL) {
1317 		ret = skb->len;
1318 		netlink_skb_set_owner_r(skb, sk);
1319 		NETLINK_CB(skb).sk = ssk;
1320 		netlink_deliver_tap_kernel(sk, ssk, skb);
1321 		nlk->netlink_rcv(skb);
1322 		consume_skb(skb);
1323 	} else {
1324 		kfree_skb(skb);
1325 	}
1326 	sock_put(sk);
1327 	return ret;
1328 }
1329 
netlink_unicast(struct sock * ssk,struct sk_buff * skb,u32 portid,int nonblock)1330 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1331 		    u32 portid, int nonblock)
1332 {
1333 	struct sock *sk;
1334 	int err;
1335 	long timeo;
1336 
1337 	skb = netlink_trim(skb, gfp_any());
1338 
1339 	timeo = sock_sndtimeo(ssk, nonblock);
1340 retry:
1341 	sk = netlink_getsockbyportid(ssk, portid);
1342 	if (IS_ERR(sk)) {
1343 		kfree_skb(skb);
1344 		return PTR_ERR(sk);
1345 	}
1346 	if (netlink_is_kernel(sk))
1347 		return netlink_unicast_kernel(sk, skb, ssk);
1348 
1349 	if (sk_filter(sk, skb)) {
1350 		err = skb->len;
1351 		kfree_skb(skb);
1352 		sock_put(sk);
1353 		return err;
1354 	}
1355 
1356 	err = netlink_attachskb(sk, skb, &timeo, ssk);
1357 	if (err == 1)
1358 		goto retry;
1359 	if (err)
1360 		return err;
1361 
1362 	return netlink_sendskb(sk, skb);
1363 }
1364 EXPORT_SYMBOL(netlink_unicast);
1365 
netlink_has_listeners(struct sock * sk,unsigned int group)1366 int netlink_has_listeners(struct sock *sk, unsigned int group)
1367 {
1368 	int res = 0;
1369 	struct listeners *listeners;
1370 
1371 	BUG_ON(!netlink_is_kernel(sk));
1372 
1373 	rcu_read_lock();
1374 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1375 
1376 	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1377 		res = test_bit(group - 1, listeners->masks);
1378 
1379 	rcu_read_unlock();
1380 
1381 	return res;
1382 }
1383 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1384 
netlink_strict_get_check(struct sk_buff * skb)1385 bool netlink_strict_get_check(struct sk_buff *skb)
1386 {
1387 	return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1388 }
1389 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1390 
netlink_broadcast_deliver(struct sock * sk,struct sk_buff * skb)1391 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1392 {
1393 	struct netlink_sock *nlk = nlk_sk(sk);
1394 
1395 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1396 	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1397 		netlink_skb_set_owner_r(skb, sk);
1398 		__netlink_sendskb(sk, skb);
1399 		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1400 	}
1401 	return -1;
1402 }
1403 
1404 struct netlink_broadcast_data {
1405 	struct sock *exclude_sk;
1406 	struct net *net;
1407 	u32 portid;
1408 	u32 group;
1409 	int failure;
1410 	int delivery_failure;
1411 	int congested;
1412 	int delivered;
1413 	gfp_t allocation;
1414 	struct sk_buff *skb, *skb2;
1415 	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1416 	void *tx_data;
1417 };
1418 
do_one_broadcast(struct sock * sk,struct netlink_broadcast_data * p)1419 static void do_one_broadcast(struct sock *sk,
1420 				    struct netlink_broadcast_data *p)
1421 {
1422 	struct netlink_sock *nlk = nlk_sk(sk);
1423 	int val;
1424 
1425 	if (p->exclude_sk == sk)
1426 		return;
1427 
1428 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1429 	    !test_bit(p->group - 1, nlk->groups))
1430 		return;
1431 
1432 	if (!net_eq(sock_net(sk), p->net)) {
1433 		if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1434 			return;
1435 
1436 		if (!peernet_has_id(sock_net(sk), p->net))
1437 			return;
1438 
1439 		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1440 				     CAP_NET_BROADCAST))
1441 			return;
1442 	}
1443 
1444 	if (p->failure) {
1445 		netlink_overrun(sk);
1446 		return;
1447 	}
1448 
1449 	sock_hold(sk);
1450 	if (p->skb2 == NULL) {
1451 		if (skb_shared(p->skb)) {
1452 			p->skb2 = skb_clone(p->skb, p->allocation);
1453 		} else {
1454 			p->skb2 = skb_get(p->skb);
1455 			/*
1456 			 * skb ownership may have been set when
1457 			 * delivered to a previous socket.
1458 			 */
1459 			skb_orphan(p->skb2);
1460 		}
1461 	}
1462 	if (p->skb2 == NULL) {
1463 		netlink_overrun(sk);
1464 		/* Clone failed. Notify ALL listeners. */
1465 		p->failure = 1;
1466 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1467 			p->delivery_failure = 1;
1468 		goto out;
1469 	}
1470 
1471 	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1472 		kfree_skb(p->skb2);
1473 		p->skb2 = NULL;
1474 		goto out;
1475 	}
1476 
1477 	if (sk_filter(sk, p->skb2)) {
1478 		kfree_skb(p->skb2);
1479 		p->skb2 = NULL;
1480 		goto out;
1481 	}
1482 	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1483 	if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1484 		NETLINK_CB(p->skb2).nsid_is_set = true;
1485 	val = netlink_broadcast_deliver(sk, p->skb2);
1486 	if (val < 0) {
1487 		netlink_overrun(sk);
1488 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1489 			p->delivery_failure = 1;
1490 	} else {
1491 		p->congested |= val;
1492 		p->delivered = 1;
1493 		p->skb2 = NULL;
1494 	}
1495 out:
1496 	sock_put(sk);
1497 }
1498 
netlink_broadcast_filtered(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation,netlink_filter_fn filter,void * filter_data)1499 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1500 			       u32 portid,
1501 			       u32 group, gfp_t allocation,
1502 			       netlink_filter_fn filter,
1503 			       void *filter_data)
1504 {
1505 	struct net *net = sock_net(ssk);
1506 	struct netlink_broadcast_data info;
1507 	struct sock *sk;
1508 
1509 	skb = netlink_trim(skb, allocation);
1510 
1511 	info.exclude_sk = ssk;
1512 	info.net = net;
1513 	info.portid = portid;
1514 	info.group = group;
1515 	info.failure = 0;
1516 	info.delivery_failure = 0;
1517 	info.congested = 0;
1518 	info.delivered = 0;
1519 	info.allocation = allocation;
1520 	info.skb = skb;
1521 	info.skb2 = NULL;
1522 	info.tx_filter = filter;
1523 	info.tx_data = filter_data;
1524 
1525 	/* While we sleep in clone, do not allow to change socket list */
1526 
1527 	netlink_lock_table();
1528 
1529 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1530 		do_one_broadcast(sk, &info);
1531 
1532 	consume_skb(skb);
1533 
1534 	netlink_unlock_table();
1535 
1536 	if (info.delivery_failure) {
1537 		kfree_skb(info.skb2);
1538 		return -ENOBUFS;
1539 	}
1540 	consume_skb(info.skb2);
1541 
1542 	if (info.delivered) {
1543 		if (info.congested && gfpflags_allow_blocking(allocation))
1544 			yield();
1545 		return 0;
1546 	}
1547 	return -ESRCH;
1548 }
1549 EXPORT_SYMBOL(netlink_broadcast_filtered);
1550 
netlink_broadcast(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation)1551 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1552 		      u32 group, gfp_t allocation)
1553 {
1554 	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1555 					  NULL, NULL);
1556 }
1557 EXPORT_SYMBOL(netlink_broadcast);
1558 
1559 struct netlink_set_err_data {
1560 	struct sock *exclude_sk;
1561 	u32 portid;
1562 	u32 group;
1563 	int code;
1564 };
1565 
do_one_set_err(struct sock * sk,struct netlink_set_err_data * p)1566 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1567 {
1568 	struct netlink_sock *nlk = nlk_sk(sk);
1569 	int ret = 0;
1570 
1571 	if (sk == p->exclude_sk)
1572 		goto out;
1573 
1574 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1575 		goto out;
1576 
1577 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1578 	    !test_bit(p->group - 1, nlk->groups))
1579 		goto out;
1580 
1581 	if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1582 		ret = 1;
1583 		goto out;
1584 	}
1585 
1586 	WRITE_ONCE(sk->sk_err, p->code);
1587 	sk_error_report(sk);
1588 out:
1589 	return ret;
1590 }
1591 
1592 /**
1593  * netlink_set_err - report error to broadcast listeners
1594  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1595  * @portid: the PORTID of a process that we want to skip (if any)
1596  * @group: the broadcast group that will notice the error
1597  * @code: error code, must be negative (as usual in kernelspace)
1598  *
1599  * This function returns the number of broadcast listeners that have set the
1600  * NETLINK_NO_ENOBUFS socket option.
1601  */
netlink_set_err(struct sock * ssk,u32 portid,u32 group,int code)1602 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1603 {
1604 	struct netlink_set_err_data info;
1605 	unsigned long flags;
1606 	struct sock *sk;
1607 	int ret = 0;
1608 
1609 	info.exclude_sk = ssk;
1610 	info.portid = portid;
1611 	info.group = group;
1612 	/* sk->sk_err wants a positive error value */
1613 	info.code = -code;
1614 
1615 	read_lock_irqsave(&nl_table_lock, flags);
1616 
1617 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1618 		ret += do_one_set_err(sk, &info);
1619 
1620 	read_unlock_irqrestore(&nl_table_lock, flags);
1621 	return ret;
1622 }
1623 EXPORT_SYMBOL(netlink_set_err);
1624 
1625 /* must be called with netlink table grabbed */
netlink_update_socket_mc(struct netlink_sock * nlk,unsigned int group,int is_new)1626 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1627 				     unsigned int group,
1628 				     int is_new)
1629 {
1630 	int old, new = !!is_new, subscriptions;
1631 
1632 	old = test_bit(group - 1, nlk->groups);
1633 	subscriptions = nlk->subscriptions - old + new;
1634 	__assign_bit(group - 1, nlk->groups, new);
1635 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1636 	netlink_update_listeners(&nlk->sk);
1637 }
1638 
netlink_setsockopt(struct socket * sock,int level,int optname,sockptr_t optval,unsigned int optlen)1639 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1640 			      sockptr_t optval, unsigned int optlen)
1641 {
1642 	struct sock *sk = sock->sk;
1643 	struct netlink_sock *nlk = nlk_sk(sk);
1644 	unsigned int val = 0;
1645 	int nr = -1;
1646 
1647 	if (level != SOL_NETLINK)
1648 		return -ENOPROTOOPT;
1649 
1650 	if (optlen >= sizeof(int) &&
1651 	    copy_from_sockptr(&val, optval, sizeof(val)))
1652 		return -EFAULT;
1653 
1654 	switch (optname) {
1655 	case NETLINK_PKTINFO:
1656 		nr = NETLINK_F_RECV_PKTINFO;
1657 		break;
1658 	case NETLINK_ADD_MEMBERSHIP:
1659 	case NETLINK_DROP_MEMBERSHIP: {
1660 		int err;
1661 
1662 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1663 			return -EPERM;
1664 		err = netlink_realloc_groups(sk);
1665 		if (err)
1666 			return err;
1667 		if (!val || val - 1 >= nlk->ngroups)
1668 			return -EINVAL;
1669 		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1670 			err = nlk->netlink_bind(sock_net(sk), val);
1671 			if (err)
1672 				return err;
1673 		}
1674 		netlink_table_grab();
1675 		netlink_update_socket_mc(nlk, val,
1676 					 optname == NETLINK_ADD_MEMBERSHIP);
1677 		netlink_table_ungrab();
1678 		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1679 			nlk->netlink_unbind(sock_net(sk), val);
1680 
1681 		break;
1682 	}
1683 	case NETLINK_BROADCAST_ERROR:
1684 		nr = NETLINK_F_BROADCAST_SEND_ERROR;
1685 		break;
1686 	case NETLINK_NO_ENOBUFS:
1687 		assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1688 		if (val) {
1689 			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1690 			wake_up_interruptible(&nlk->wait);
1691 		}
1692 		break;
1693 	case NETLINK_LISTEN_ALL_NSID:
1694 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1695 			return -EPERM;
1696 		nr = NETLINK_F_LISTEN_ALL_NSID;
1697 		break;
1698 	case NETLINK_CAP_ACK:
1699 		nr = NETLINK_F_CAP_ACK;
1700 		break;
1701 	case NETLINK_EXT_ACK:
1702 		nr = NETLINK_F_EXT_ACK;
1703 		break;
1704 	case NETLINK_GET_STRICT_CHK:
1705 		nr = NETLINK_F_STRICT_CHK;
1706 		break;
1707 	default:
1708 		return -ENOPROTOOPT;
1709 	}
1710 	if (nr >= 0)
1711 		assign_bit(nr, &nlk->flags, val);
1712 	return 0;
1713 }
1714 
netlink_getsockopt(struct socket * sock,int level,int optname,char __user * optval,int __user * optlen)1715 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1716 			      char __user *optval, int __user *optlen)
1717 {
1718 	struct sock *sk = sock->sk;
1719 	struct netlink_sock *nlk = nlk_sk(sk);
1720 	unsigned int flag;
1721 	int len, val;
1722 
1723 	if (level != SOL_NETLINK)
1724 		return -ENOPROTOOPT;
1725 
1726 	if (get_user(len, optlen))
1727 		return -EFAULT;
1728 	if (len < 0)
1729 		return -EINVAL;
1730 
1731 	switch (optname) {
1732 	case NETLINK_PKTINFO:
1733 		flag = NETLINK_F_RECV_PKTINFO;
1734 		break;
1735 	case NETLINK_BROADCAST_ERROR:
1736 		flag = NETLINK_F_BROADCAST_SEND_ERROR;
1737 		break;
1738 	case NETLINK_NO_ENOBUFS:
1739 		flag = NETLINK_F_RECV_NO_ENOBUFS;
1740 		break;
1741 	case NETLINK_LIST_MEMBERSHIPS: {
1742 		int pos, idx, shift, err = 0;
1743 
1744 		netlink_lock_table();
1745 		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1746 			if (len - pos < sizeof(u32))
1747 				break;
1748 
1749 			idx = pos / sizeof(unsigned long);
1750 			shift = (pos % sizeof(unsigned long)) * 8;
1751 			if (put_user((u32)(nlk->groups[idx] >> shift),
1752 				     (u32 __user *)(optval + pos))) {
1753 				err = -EFAULT;
1754 				break;
1755 			}
1756 		}
1757 		if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1758 			err = -EFAULT;
1759 		netlink_unlock_table();
1760 		return err;
1761 	}
1762 	case NETLINK_LISTEN_ALL_NSID:
1763 		flag = NETLINK_F_LISTEN_ALL_NSID;
1764 		break;
1765 	case NETLINK_CAP_ACK:
1766 		flag = NETLINK_F_CAP_ACK;
1767 		break;
1768 	case NETLINK_EXT_ACK:
1769 		flag = NETLINK_F_EXT_ACK;
1770 		break;
1771 	case NETLINK_GET_STRICT_CHK:
1772 		flag = NETLINK_F_STRICT_CHK;
1773 		break;
1774 	default:
1775 		return -ENOPROTOOPT;
1776 	}
1777 
1778 	if (len < sizeof(int))
1779 		return -EINVAL;
1780 
1781 	len = sizeof(int);
1782 	val = test_bit(flag, &nlk->flags);
1783 
1784 	if (put_user(len, optlen) ||
1785 	    copy_to_user(optval, &val, len))
1786 		return -EFAULT;
1787 
1788 	return 0;
1789 }
1790 
netlink_cmsg_recv_pktinfo(struct msghdr * msg,struct sk_buff * skb)1791 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1792 {
1793 	struct nl_pktinfo info;
1794 
1795 	info.group = NETLINK_CB(skb).dst_group;
1796 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1797 }
1798 
netlink_cmsg_listen_all_nsid(struct sock * sk,struct msghdr * msg,struct sk_buff * skb)1799 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1800 					 struct sk_buff *skb)
1801 {
1802 	if (!NETLINK_CB(skb).nsid_is_set)
1803 		return;
1804 
1805 	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1806 		 &NETLINK_CB(skb).nsid);
1807 }
1808 
netlink_sendmsg(struct socket * sock,struct msghdr * msg,size_t len)1809 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1810 {
1811 	struct sock *sk = sock->sk;
1812 	struct netlink_sock *nlk = nlk_sk(sk);
1813 	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1814 	u32 dst_portid;
1815 	u32 dst_group;
1816 	struct sk_buff *skb;
1817 	int err;
1818 	struct scm_cookie scm;
1819 	u32 netlink_skb_flags = 0;
1820 
1821 	if (msg->msg_flags & MSG_OOB)
1822 		return -EOPNOTSUPP;
1823 
1824 	if (len == 0) {
1825 		pr_warn_once("Zero length message leads to an empty skb\n");
1826 		return -ENODATA;
1827 	}
1828 
1829 	err = scm_send(sock, msg, &scm, true);
1830 	if (err < 0)
1831 		return err;
1832 
1833 	if (msg->msg_namelen) {
1834 		err = -EINVAL;
1835 		if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1836 			goto out;
1837 		if (addr->nl_family != AF_NETLINK)
1838 			goto out;
1839 		dst_portid = addr->nl_pid;
1840 		dst_group = ffs(addr->nl_groups);
1841 		err =  -EPERM;
1842 		if ((dst_group || dst_portid) &&
1843 		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1844 			goto out;
1845 		netlink_skb_flags |= NETLINK_SKB_DST;
1846 	} else {
1847 		/* Paired with WRITE_ONCE() in netlink_connect() */
1848 		dst_portid = READ_ONCE(nlk->dst_portid);
1849 		dst_group = READ_ONCE(nlk->dst_group);
1850 	}
1851 
1852 	/* Paired with WRITE_ONCE() in netlink_insert() */
1853 	if (!READ_ONCE(nlk->bound)) {
1854 		err = netlink_autobind(sock);
1855 		if (err)
1856 			goto out;
1857 	} else {
1858 		/* Ensure nlk is hashed and visible. */
1859 		smp_rmb();
1860 	}
1861 
1862 	err = -EMSGSIZE;
1863 	if (len > sk->sk_sndbuf - 32)
1864 		goto out;
1865 	err = -ENOBUFS;
1866 	skb = netlink_alloc_large_skb(len, dst_group);
1867 	if (skb == NULL)
1868 		goto out;
1869 
1870 	NETLINK_CB(skb).portid	= nlk->portid;
1871 	NETLINK_CB(skb).dst_group = dst_group;
1872 	NETLINK_CB(skb).creds	= scm.creds;
1873 	NETLINK_CB(skb).flags	= netlink_skb_flags;
1874 
1875 	err = -EFAULT;
1876 	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1877 		kfree_skb(skb);
1878 		goto out;
1879 	}
1880 
1881 	err = security_netlink_send(sk, skb);
1882 	if (err) {
1883 		kfree_skb(skb);
1884 		goto out;
1885 	}
1886 
1887 	if (dst_group) {
1888 		refcount_inc(&skb->users);
1889 		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1890 	}
1891 	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1892 
1893 out:
1894 	scm_destroy(&scm);
1895 	return err;
1896 }
1897 
netlink_recvmsg(struct socket * sock,struct msghdr * msg,size_t len,int flags)1898 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1899 			   int flags)
1900 {
1901 	struct scm_cookie scm;
1902 	struct sock *sk = sock->sk;
1903 	struct netlink_sock *nlk = nlk_sk(sk);
1904 	size_t copied, max_recvmsg_len;
1905 	struct sk_buff *skb, *data_skb;
1906 	int err, ret;
1907 
1908 	if (flags & MSG_OOB)
1909 		return -EOPNOTSUPP;
1910 
1911 	copied = 0;
1912 
1913 	skb = skb_recv_datagram(sk, flags, &err);
1914 	if (skb == NULL)
1915 		goto out;
1916 
1917 	data_skb = skb;
1918 
1919 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1920 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1921 		/*
1922 		 * If this skb has a frag_list, then here that means that we
1923 		 * will have to use the frag_list skb's data for compat tasks
1924 		 * and the regular skb's data for normal (non-compat) tasks.
1925 		 *
1926 		 * If we need to send the compat skb, assign it to the
1927 		 * 'data_skb' variable so that it will be used below for data
1928 		 * copying. We keep 'skb' for everything else, including
1929 		 * freeing both later.
1930 		 */
1931 		if (flags & MSG_CMSG_COMPAT)
1932 			data_skb = skb_shinfo(skb)->frag_list;
1933 	}
1934 #endif
1935 
1936 	/* Record the max length of recvmsg() calls for future allocations */
1937 	max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1938 	max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1939 				SKB_WITH_OVERHEAD(32768));
1940 	WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1941 
1942 	copied = data_skb->len;
1943 	if (len < copied) {
1944 		msg->msg_flags |= MSG_TRUNC;
1945 		copied = len;
1946 	}
1947 
1948 	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1949 
1950 	if (msg->msg_name) {
1951 		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1952 		addr->nl_family = AF_NETLINK;
1953 		addr->nl_pad    = 0;
1954 		addr->nl_pid	= NETLINK_CB(skb).portid;
1955 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1956 		msg->msg_namelen = sizeof(*addr);
1957 	}
1958 
1959 	if (nlk_test_bit(RECV_PKTINFO, sk))
1960 		netlink_cmsg_recv_pktinfo(msg, skb);
1961 	if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1962 		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1963 
1964 	memset(&scm, 0, sizeof(scm));
1965 	scm.creds = *NETLINK_CREDS(skb);
1966 	if (flags & MSG_TRUNC)
1967 		copied = data_skb->len;
1968 
1969 	skb_free_datagram(sk, skb);
1970 
1971 	if (READ_ONCE(nlk->cb_running) &&
1972 	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1973 		ret = netlink_dump(sk, false);
1974 		if (ret) {
1975 			WRITE_ONCE(sk->sk_err, -ret);
1976 			sk_error_report(sk);
1977 		}
1978 	}
1979 
1980 	scm_recv(sock, msg, &scm, flags);
1981 out:
1982 	netlink_rcv_wake(sk);
1983 	return err ? : copied;
1984 }
1985 
netlink_data_ready(struct sock * sk)1986 static void netlink_data_ready(struct sock *sk)
1987 {
1988 	BUG();
1989 }
1990 
1991 /*
1992  *	We export these functions to other modules. They provide a
1993  *	complete set of kernel non-blocking support for message
1994  *	queueing.
1995  */
1996 
1997 struct sock *
__netlink_kernel_create(struct net * net,int unit,struct module * module,struct netlink_kernel_cfg * cfg)1998 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1999 			struct netlink_kernel_cfg *cfg)
2000 {
2001 	struct socket *sock;
2002 	struct sock *sk;
2003 	struct netlink_sock *nlk;
2004 	struct listeners *listeners = NULL;
2005 	unsigned int groups;
2006 
2007 	BUG_ON(!nl_table);
2008 
2009 	if (unit < 0 || unit >= MAX_LINKS)
2010 		return NULL;
2011 
2012 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2013 		return NULL;
2014 
2015 	if (__netlink_create(net, sock, unit, 1) < 0)
2016 		goto out_sock_release_nosk;
2017 
2018 	sk = sock->sk;
2019 
2020 	if (!cfg || cfg->groups < 32)
2021 		groups = 32;
2022 	else
2023 		groups = cfg->groups;
2024 
2025 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2026 	if (!listeners)
2027 		goto out_sock_release;
2028 
2029 	sk->sk_data_ready = netlink_data_ready;
2030 	if (cfg && cfg->input)
2031 		nlk_sk(sk)->netlink_rcv = cfg->input;
2032 
2033 	if (netlink_insert(sk, 0))
2034 		goto out_sock_release;
2035 
2036 	nlk = nlk_sk(sk);
2037 	set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2038 
2039 	netlink_table_grab();
2040 	if (!nl_table[unit].registered) {
2041 		nl_table[unit].groups = groups;
2042 		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2043 		nl_table[unit].module = module;
2044 		if (cfg) {
2045 			nl_table[unit].bind = cfg->bind;
2046 			nl_table[unit].unbind = cfg->unbind;
2047 			nl_table[unit].release = cfg->release;
2048 			nl_table[unit].flags = cfg->flags;
2049 		}
2050 		nl_table[unit].registered = 1;
2051 	} else {
2052 		kfree(listeners);
2053 		nl_table[unit].registered++;
2054 	}
2055 	netlink_table_ungrab();
2056 	return sk;
2057 
2058 out_sock_release:
2059 	kfree(listeners);
2060 	netlink_kernel_release(sk);
2061 	return NULL;
2062 
2063 out_sock_release_nosk:
2064 	sock_release(sock);
2065 	return NULL;
2066 }
2067 EXPORT_SYMBOL(__netlink_kernel_create);
2068 
2069 void
netlink_kernel_release(struct sock * sk)2070 netlink_kernel_release(struct sock *sk)
2071 {
2072 	if (sk == NULL || sk->sk_socket == NULL)
2073 		return;
2074 
2075 	sock_release(sk->sk_socket);
2076 }
2077 EXPORT_SYMBOL(netlink_kernel_release);
2078 
__netlink_change_ngroups(struct sock * sk,unsigned int groups)2079 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2080 {
2081 	struct listeners *new, *old;
2082 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2083 
2084 	if (groups < 32)
2085 		groups = 32;
2086 
2087 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2088 		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2089 		if (!new)
2090 			return -ENOMEM;
2091 		old = nl_deref_protected(tbl->listeners);
2092 		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2093 		rcu_assign_pointer(tbl->listeners, new);
2094 
2095 		kfree_rcu(old, rcu);
2096 	}
2097 	tbl->groups = groups;
2098 
2099 	return 0;
2100 }
2101 
2102 /**
2103  * netlink_change_ngroups - change number of multicast groups
2104  *
2105  * This changes the number of multicast groups that are available
2106  * on a certain netlink family. Note that it is not possible to
2107  * change the number of groups to below 32. Also note that it does
2108  * not implicitly call netlink_clear_multicast_users() when the
2109  * number of groups is reduced.
2110  *
2111  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2112  * @groups: The new number of groups.
2113  */
netlink_change_ngroups(struct sock * sk,unsigned int groups)2114 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2115 {
2116 	int err;
2117 
2118 	netlink_table_grab();
2119 	err = __netlink_change_ngroups(sk, groups);
2120 	netlink_table_ungrab();
2121 
2122 	return err;
2123 }
2124 
__netlink_clear_multicast_users(struct sock * ksk,unsigned int group)2125 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2126 {
2127 	struct sock *sk;
2128 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2129 	struct hlist_node *tmp;
2130 
2131 	sk_for_each_bound_safe(sk, tmp, &tbl->mc_list)
2132 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2133 }
2134 
2135 struct nlmsghdr *
__nlmsg_put(struct sk_buff * skb,u32 portid,u32 seq,int type,int len,int flags)2136 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2137 {
2138 	struct nlmsghdr *nlh;
2139 	int size = nlmsg_msg_size(len);
2140 
2141 	nlh = skb_put(skb, NLMSG_ALIGN(size));
2142 	nlh->nlmsg_type = type;
2143 	nlh->nlmsg_len = size;
2144 	nlh->nlmsg_flags = flags;
2145 	nlh->nlmsg_pid = portid;
2146 	nlh->nlmsg_seq = seq;
2147 	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2148 		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2149 	return nlh;
2150 }
2151 EXPORT_SYMBOL(__nlmsg_put);
2152 
2153 static size_t
netlink_ack_tlv_len(struct netlink_sock * nlk,int err,const struct netlink_ext_ack * extack)2154 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2155 		    const struct netlink_ext_ack *extack)
2156 {
2157 	size_t tlvlen;
2158 
2159 	if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2160 		return 0;
2161 
2162 	tlvlen = 0;
2163 	if (extack->_msg)
2164 		tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2165 	if (extack->cookie_len)
2166 		tlvlen += nla_total_size(extack->cookie_len);
2167 
2168 	/* Following attributes are only reported as error (not warning) */
2169 	if (!err)
2170 		return tlvlen;
2171 
2172 	if (extack->bad_attr)
2173 		tlvlen += nla_total_size(sizeof(u32));
2174 	if (extack->policy)
2175 		tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2176 	if (extack->miss_type)
2177 		tlvlen += nla_total_size(sizeof(u32));
2178 	if (extack->miss_nest)
2179 		tlvlen += nla_total_size(sizeof(u32));
2180 
2181 	return tlvlen;
2182 }
2183 
nlmsg_check_in_payload(const struct nlmsghdr * nlh,const void * addr)2184 static bool nlmsg_check_in_payload(const struct nlmsghdr *nlh, const void *addr)
2185 {
2186 	return !WARN_ON(addr < nlmsg_data(nlh) ||
2187 			addr - (const void *) nlh >= nlh->nlmsg_len);
2188 }
2189 
2190 static void
netlink_ack_tlv_fill(struct sk_buff * skb,const struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2191 netlink_ack_tlv_fill(struct sk_buff *skb, const struct nlmsghdr *nlh, int err,
2192 		     const struct netlink_ext_ack *extack)
2193 {
2194 	if (extack->_msg)
2195 		WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2196 	if (extack->cookie_len)
2197 		WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2198 				extack->cookie_len, extack->cookie));
2199 
2200 	if (!err)
2201 		return;
2202 
2203 	if (extack->bad_attr && nlmsg_check_in_payload(nlh, extack->bad_attr))
2204 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2205 				    (u8 *)extack->bad_attr - (const u8 *)nlh));
2206 	if (extack->policy)
2207 		netlink_policy_dump_write_attr(skb, extack->policy,
2208 					       NLMSGERR_ATTR_POLICY);
2209 	if (extack->miss_type)
2210 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2211 				    extack->miss_type));
2212 	if (extack->miss_nest && nlmsg_check_in_payload(nlh, extack->miss_nest))
2213 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2214 				    (u8 *)extack->miss_nest - (const u8 *)nlh));
2215 }
2216 
2217 /*
2218  * It looks a bit ugly.
2219  * It would be better to create kernel thread.
2220  */
2221 
netlink_dump_done(struct netlink_sock * nlk,struct sk_buff * skb,struct netlink_callback * cb,struct netlink_ext_ack * extack)2222 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2223 			     struct netlink_callback *cb,
2224 			     struct netlink_ext_ack *extack)
2225 {
2226 	struct nlmsghdr *nlh;
2227 	size_t extack_len;
2228 
2229 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2230 			       NLM_F_MULTI | cb->answer_flags);
2231 	if (WARN_ON(!nlh))
2232 		return -ENOBUFS;
2233 
2234 	nl_dump_check_consistent(cb, nlh);
2235 	memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2236 
2237 	extack_len = netlink_ack_tlv_len(nlk, nlk->dump_done_errno, extack);
2238 	if (extack_len) {
2239 		nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2240 		if (skb_tailroom(skb) >= extack_len) {
2241 			netlink_ack_tlv_fill(skb, cb->nlh,
2242 					     nlk->dump_done_errno, extack);
2243 			nlmsg_end(skb, nlh);
2244 		}
2245 	}
2246 
2247 	return 0;
2248 }
2249 
netlink_dump(struct sock * sk,bool lock_taken)2250 static int netlink_dump(struct sock *sk, bool lock_taken)
2251 {
2252 	struct netlink_sock *nlk = nlk_sk(sk);
2253 	struct netlink_ext_ack extack = {};
2254 	struct netlink_callback *cb;
2255 	struct sk_buff *skb = NULL;
2256 	size_t max_recvmsg_len;
2257 	struct module *module;
2258 	int err = -ENOBUFS;
2259 	int alloc_min_size;
2260 	int alloc_size;
2261 
2262 	if (!lock_taken)
2263 		mutex_lock(&nlk->nl_cb_mutex);
2264 	if (!nlk->cb_running) {
2265 		err = -EINVAL;
2266 		goto errout_skb;
2267 	}
2268 
2269 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2270 		goto errout_skb;
2271 
2272 	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2273 	 * required, but it makes sense to _attempt_ a 32KiB allocation
2274 	 * to reduce number of system calls on dump operations, if user
2275 	 * ever provided a big enough buffer.
2276 	 */
2277 	cb = &nlk->cb;
2278 	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2279 
2280 	max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2281 	if (alloc_min_size < max_recvmsg_len) {
2282 		alloc_size = max_recvmsg_len;
2283 		skb = alloc_skb(alloc_size,
2284 				(GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2285 				__GFP_NOWARN | __GFP_NORETRY);
2286 	}
2287 	if (!skb) {
2288 		alloc_size = alloc_min_size;
2289 		skb = alloc_skb(alloc_size, GFP_KERNEL);
2290 	}
2291 	if (!skb)
2292 		goto errout_skb;
2293 
2294 	/* Trim skb to allocated size. User is expected to provide buffer as
2295 	 * large as max(min_dump_alloc, 32KiB (max_recvmsg_len capped at
2296 	 * netlink_recvmsg())). dump will pack as many smaller messages as
2297 	 * could fit within the allocated skb. skb is typically allocated
2298 	 * with larger space than required (could be as much as near 2x the
2299 	 * requested size with align to next power of 2 approach). Allowing
2300 	 * dump to use the excess space makes it difficult for a user to have a
2301 	 * reasonable static buffer based on the expected largest dump of a
2302 	 * single netdev. The outcome is MSG_TRUNC error.
2303 	 */
2304 	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2305 
2306 	/* Make sure malicious BPF programs can not read unitialized memory
2307 	 * from skb->head -> skb->data
2308 	 */
2309 	skb_reset_network_header(skb);
2310 	skb_reset_mac_header(skb);
2311 
2312 	netlink_skb_set_owner_r(skb, sk);
2313 
2314 	if (nlk->dump_done_errno > 0) {
2315 		cb->extack = &extack;
2316 
2317 		nlk->dump_done_errno = cb->dump(skb, cb);
2318 
2319 		/* EMSGSIZE plus something already in the skb means
2320 		 * that there's more to dump but current skb has filled up.
2321 		 * If the callback really wants to return EMSGSIZE to user space
2322 		 * it needs to do so again, on the next cb->dump() call,
2323 		 * without putting data in the skb.
2324 		 */
2325 		if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2326 			nlk->dump_done_errno = skb->len;
2327 
2328 		cb->extack = NULL;
2329 	}
2330 
2331 	if (nlk->dump_done_errno > 0 ||
2332 	    skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2333 		mutex_unlock(&nlk->nl_cb_mutex);
2334 
2335 		if (sk_filter(sk, skb))
2336 			kfree_skb(skb);
2337 		else
2338 			__netlink_sendskb(sk, skb);
2339 		return 0;
2340 	}
2341 
2342 	if (netlink_dump_done(nlk, skb, cb, &extack))
2343 		goto errout_skb;
2344 
2345 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2346 	/* frag_list skb's data is used for compat tasks
2347 	 * and the regular skb's data for normal (non-compat) tasks.
2348 	 * See netlink_recvmsg().
2349 	 */
2350 	if (unlikely(skb_shinfo(skb)->frag_list)) {
2351 		if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2352 			goto errout_skb;
2353 	}
2354 #endif
2355 
2356 	if (sk_filter(sk, skb))
2357 		kfree_skb(skb);
2358 	else
2359 		__netlink_sendskb(sk, skb);
2360 
2361 	if (cb->done)
2362 		cb->done(cb);
2363 
2364 	WRITE_ONCE(nlk->cb_running, false);
2365 	module = cb->module;
2366 	skb = cb->skb;
2367 	mutex_unlock(&nlk->nl_cb_mutex);
2368 	module_put(module);
2369 	consume_skb(skb);
2370 	return 0;
2371 
2372 errout_skb:
2373 	mutex_unlock(&nlk->nl_cb_mutex);
2374 	kfree_skb(skb);
2375 	return err;
2376 }
2377 
__netlink_dump_start(struct sock * ssk,struct sk_buff * skb,const struct nlmsghdr * nlh,struct netlink_dump_control * control)2378 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2379 			 const struct nlmsghdr *nlh,
2380 			 struct netlink_dump_control *control)
2381 {
2382 	struct netlink_callback *cb;
2383 	struct netlink_sock *nlk;
2384 	struct sock *sk;
2385 	int ret;
2386 
2387 	refcount_inc(&skb->users);
2388 
2389 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2390 	if (sk == NULL) {
2391 		ret = -ECONNREFUSED;
2392 		goto error_free;
2393 	}
2394 
2395 	nlk = nlk_sk(sk);
2396 	mutex_lock(&nlk->nl_cb_mutex);
2397 	/* A dump is in progress... */
2398 	if (nlk->cb_running) {
2399 		ret = -EBUSY;
2400 		goto error_unlock;
2401 	}
2402 	/* add reference of module which cb->dump belongs to */
2403 	if (!try_module_get(control->module)) {
2404 		ret = -EPROTONOSUPPORT;
2405 		goto error_unlock;
2406 	}
2407 
2408 	cb = &nlk->cb;
2409 	memset(cb, 0, sizeof(*cb));
2410 	cb->dump = control->dump;
2411 	cb->done = control->done;
2412 	cb->nlh = nlh;
2413 	cb->data = control->data;
2414 	cb->module = control->module;
2415 	cb->min_dump_alloc = control->min_dump_alloc;
2416 	cb->flags = control->flags;
2417 	cb->skb = skb;
2418 
2419 	cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2420 
2421 	if (control->start) {
2422 		cb->extack = control->extack;
2423 		ret = control->start(cb);
2424 		cb->extack = NULL;
2425 		if (ret)
2426 			goto error_put;
2427 	}
2428 
2429 	WRITE_ONCE(nlk->cb_running, true);
2430 	nlk->dump_done_errno = INT_MAX;
2431 
2432 	ret = netlink_dump(sk, true);
2433 
2434 	sock_put(sk);
2435 
2436 	if (ret)
2437 		return ret;
2438 
2439 	/* We successfully started a dump, by returning -EINTR we
2440 	 * signal not to send ACK even if it was requested.
2441 	 */
2442 	return -EINTR;
2443 
2444 error_put:
2445 	module_put(control->module);
2446 error_unlock:
2447 	sock_put(sk);
2448 	mutex_unlock(&nlk->nl_cb_mutex);
2449 error_free:
2450 	kfree_skb(skb);
2451 	return ret;
2452 }
2453 EXPORT_SYMBOL(__netlink_dump_start);
2454 
netlink_ack(struct sk_buff * in_skb,struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2455 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2456 		 const struct netlink_ext_ack *extack)
2457 {
2458 	struct sk_buff *skb;
2459 	struct nlmsghdr *rep;
2460 	struct nlmsgerr *errmsg;
2461 	size_t payload = sizeof(*errmsg);
2462 	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2463 	unsigned int flags = 0;
2464 	size_t tlvlen;
2465 
2466 	/* Error messages get the original request appened, unless the user
2467 	 * requests to cap the error message, and get extra error data if
2468 	 * requested.
2469 	 */
2470 	if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2471 		payload += nlmsg_len(nlh);
2472 	else
2473 		flags |= NLM_F_CAPPED;
2474 
2475 	tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2476 	if (tlvlen)
2477 		flags |= NLM_F_ACK_TLVS;
2478 
2479 	skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2480 	if (!skb)
2481 		goto err_skb;
2482 
2483 	rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2484 			NLMSG_ERROR, sizeof(*errmsg), flags);
2485 	if (!rep)
2486 		goto err_bad_put;
2487 	errmsg = nlmsg_data(rep);
2488 	errmsg->error = err;
2489 	errmsg->msg = *nlh;
2490 
2491 	if (!(flags & NLM_F_CAPPED)) {
2492 		if (!nlmsg_append(skb, nlmsg_len(nlh)))
2493 			goto err_bad_put;
2494 
2495 		memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2496 		       nlmsg_len(nlh));
2497 	}
2498 
2499 	if (tlvlen)
2500 		netlink_ack_tlv_fill(skb, nlh, err, extack);
2501 
2502 	nlmsg_end(skb, rep);
2503 
2504 	nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2505 
2506 	return;
2507 
2508 err_bad_put:
2509 	nlmsg_free(skb);
2510 err_skb:
2511 	WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2512 	sk_error_report(NETLINK_CB(in_skb).sk);
2513 }
2514 EXPORT_SYMBOL(netlink_ack);
2515 
netlink_rcv_skb(struct sk_buff * skb,int (* cb)(struct sk_buff *,struct nlmsghdr *,struct netlink_ext_ack *))2516 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2517 						   struct nlmsghdr *,
2518 						   struct netlink_ext_ack *))
2519 {
2520 	struct netlink_ext_ack extack;
2521 	struct nlmsghdr *nlh;
2522 	int err;
2523 
2524 	while (skb->len >= nlmsg_total_size(0)) {
2525 		int msglen;
2526 
2527 		memset(&extack, 0, sizeof(extack));
2528 		nlh = nlmsg_hdr(skb);
2529 		err = 0;
2530 
2531 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2532 			return 0;
2533 
2534 		/* Only requests are handled by the kernel */
2535 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2536 			goto ack;
2537 
2538 		/* Skip control messages */
2539 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2540 			goto ack;
2541 
2542 		err = cb(skb, nlh, &extack);
2543 		if (err == -EINTR)
2544 			goto skip;
2545 
2546 ack:
2547 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2548 			netlink_ack(skb, nlh, err, &extack);
2549 
2550 skip:
2551 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2552 		if (msglen > skb->len)
2553 			msglen = skb->len;
2554 		skb_pull(skb, msglen);
2555 	}
2556 
2557 	return 0;
2558 }
2559 EXPORT_SYMBOL(netlink_rcv_skb);
2560 
2561 /**
2562  * nlmsg_notify - send a notification netlink message
2563  * @sk: netlink socket to use
2564  * @skb: notification message
2565  * @portid: destination netlink portid for reports or 0
2566  * @group: destination multicast group or 0
2567  * @report: 1 to report back, 0 to disable
2568  * @flags: allocation flags
2569  */
nlmsg_notify(struct sock * sk,struct sk_buff * skb,u32 portid,unsigned int group,int report,gfp_t flags)2570 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2571 		 unsigned int group, int report, gfp_t flags)
2572 {
2573 	int err = 0;
2574 
2575 	if (group) {
2576 		int exclude_portid = 0;
2577 
2578 		if (report) {
2579 			refcount_inc(&skb->users);
2580 			exclude_portid = portid;
2581 		}
2582 
2583 		/* errors reported via destination sk->sk_err, but propagate
2584 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2585 		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2586 		if (err == -ESRCH)
2587 			err = 0;
2588 	}
2589 
2590 	if (report) {
2591 		int err2;
2592 
2593 		err2 = nlmsg_unicast(sk, skb, portid);
2594 		if (!err)
2595 			err = err2;
2596 	}
2597 
2598 	return err;
2599 }
2600 EXPORT_SYMBOL(nlmsg_notify);
2601 
2602 #ifdef CONFIG_PROC_FS
2603 struct nl_seq_iter {
2604 	struct seq_net_private p;
2605 	struct rhashtable_iter hti;
2606 	int link;
2607 };
2608 
netlink_walk_start(struct nl_seq_iter * iter)2609 static void netlink_walk_start(struct nl_seq_iter *iter)
2610 {
2611 	rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2612 	rhashtable_walk_start(&iter->hti);
2613 }
2614 
netlink_walk_stop(struct nl_seq_iter * iter)2615 static void netlink_walk_stop(struct nl_seq_iter *iter)
2616 {
2617 	rhashtable_walk_stop(&iter->hti);
2618 	rhashtable_walk_exit(&iter->hti);
2619 }
2620 
__netlink_seq_next(struct seq_file * seq)2621 static void *__netlink_seq_next(struct seq_file *seq)
2622 {
2623 	struct nl_seq_iter *iter = seq->private;
2624 	struct netlink_sock *nlk;
2625 
2626 	do {
2627 		for (;;) {
2628 			nlk = rhashtable_walk_next(&iter->hti);
2629 
2630 			if (IS_ERR(nlk)) {
2631 				if (PTR_ERR(nlk) == -EAGAIN)
2632 					continue;
2633 
2634 				return nlk;
2635 			}
2636 
2637 			if (nlk)
2638 				break;
2639 
2640 			netlink_walk_stop(iter);
2641 			if (++iter->link >= MAX_LINKS)
2642 				return NULL;
2643 
2644 			netlink_walk_start(iter);
2645 		}
2646 	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2647 
2648 	return nlk;
2649 }
2650 
netlink_seq_start(struct seq_file * seq,loff_t * posp)2651 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2652 	__acquires(RCU)
2653 {
2654 	struct nl_seq_iter *iter = seq->private;
2655 	void *obj = SEQ_START_TOKEN;
2656 	loff_t pos;
2657 
2658 	iter->link = 0;
2659 
2660 	netlink_walk_start(iter);
2661 
2662 	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2663 		obj = __netlink_seq_next(seq);
2664 
2665 	return obj;
2666 }
2667 
netlink_seq_next(struct seq_file * seq,void * v,loff_t * pos)2668 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2669 {
2670 	++*pos;
2671 	return __netlink_seq_next(seq);
2672 }
2673 
netlink_native_seq_stop(struct seq_file * seq,void * v)2674 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2675 {
2676 	struct nl_seq_iter *iter = seq->private;
2677 
2678 	if (iter->link >= MAX_LINKS)
2679 		return;
2680 
2681 	netlink_walk_stop(iter);
2682 }
2683 
2684 
netlink_native_seq_show(struct seq_file * seq,void * v)2685 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2686 {
2687 	if (v == SEQ_START_TOKEN) {
2688 		seq_puts(seq,
2689 			 "sk               Eth Pid        Groups   "
2690 			 "Rmem     Wmem     Dump  Locks    Drops    Inode\n");
2691 	} else {
2692 		struct sock *s = v;
2693 		struct netlink_sock *nlk = nlk_sk(s);
2694 
2695 		seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2696 			   s,
2697 			   s->sk_protocol,
2698 			   nlk->portid,
2699 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2700 			   sk_rmem_alloc_get(s),
2701 			   sk_wmem_alloc_get(s),
2702 			   READ_ONCE(nlk->cb_running),
2703 			   refcount_read(&s->sk_refcnt),
2704 			   atomic_read(&s->sk_drops),
2705 			   sock_i_ino(s)
2706 			);
2707 
2708 	}
2709 	return 0;
2710 }
2711 
2712 #ifdef CONFIG_BPF_SYSCALL
2713 struct bpf_iter__netlink {
2714 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2715 	__bpf_md_ptr(struct netlink_sock *, sk);
2716 };
2717 
DEFINE_BPF_ITER_FUNC(netlink,struct bpf_iter_meta * meta,struct netlink_sock * sk)2718 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2719 
2720 static int netlink_prog_seq_show(struct bpf_prog *prog,
2721 				  struct bpf_iter_meta *meta,
2722 				  void *v)
2723 {
2724 	struct bpf_iter__netlink ctx;
2725 
2726 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2727 	ctx.meta = meta;
2728 	ctx.sk = nlk_sk((struct sock *)v);
2729 	return bpf_iter_run_prog(prog, &ctx);
2730 }
2731 
netlink_seq_show(struct seq_file * seq,void * v)2732 static int netlink_seq_show(struct seq_file *seq, void *v)
2733 {
2734 	struct bpf_iter_meta meta;
2735 	struct bpf_prog *prog;
2736 
2737 	meta.seq = seq;
2738 	prog = bpf_iter_get_info(&meta, false);
2739 	if (!prog)
2740 		return netlink_native_seq_show(seq, v);
2741 
2742 	if (v != SEQ_START_TOKEN)
2743 		return netlink_prog_seq_show(prog, &meta, v);
2744 
2745 	return 0;
2746 }
2747 
netlink_seq_stop(struct seq_file * seq,void * v)2748 static void netlink_seq_stop(struct seq_file *seq, void *v)
2749 {
2750 	struct bpf_iter_meta meta;
2751 	struct bpf_prog *prog;
2752 
2753 	if (!v) {
2754 		meta.seq = seq;
2755 		prog = bpf_iter_get_info(&meta, true);
2756 		if (prog)
2757 			(void)netlink_prog_seq_show(prog, &meta, v);
2758 	}
2759 
2760 	netlink_native_seq_stop(seq, v);
2761 }
2762 #else
netlink_seq_show(struct seq_file * seq,void * v)2763 static int netlink_seq_show(struct seq_file *seq, void *v)
2764 {
2765 	return netlink_native_seq_show(seq, v);
2766 }
2767 
netlink_seq_stop(struct seq_file * seq,void * v)2768 static void netlink_seq_stop(struct seq_file *seq, void *v)
2769 {
2770 	netlink_native_seq_stop(seq, v);
2771 }
2772 #endif
2773 
2774 static const struct seq_operations netlink_seq_ops = {
2775 	.start  = netlink_seq_start,
2776 	.next   = netlink_seq_next,
2777 	.stop   = netlink_seq_stop,
2778 	.show   = netlink_seq_show,
2779 };
2780 #endif
2781 
netlink_register_notifier(struct notifier_block * nb)2782 int netlink_register_notifier(struct notifier_block *nb)
2783 {
2784 	return blocking_notifier_chain_register(&netlink_chain, nb);
2785 }
2786 EXPORT_SYMBOL(netlink_register_notifier);
2787 
netlink_unregister_notifier(struct notifier_block * nb)2788 int netlink_unregister_notifier(struct notifier_block *nb)
2789 {
2790 	return blocking_notifier_chain_unregister(&netlink_chain, nb);
2791 }
2792 EXPORT_SYMBOL(netlink_unregister_notifier);
2793 
2794 static const struct proto_ops netlink_ops = {
2795 	.family =	PF_NETLINK,
2796 	.owner =	THIS_MODULE,
2797 	.release =	netlink_release,
2798 	.bind =		netlink_bind,
2799 	.connect =	netlink_connect,
2800 	.socketpair =	sock_no_socketpair,
2801 	.accept =	sock_no_accept,
2802 	.getname =	netlink_getname,
2803 	.poll =		datagram_poll,
2804 	.ioctl =	netlink_ioctl,
2805 	.listen =	sock_no_listen,
2806 	.shutdown =	sock_no_shutdown,
2807 	.setsockopt =	netlink_setsockopt,
2808 	.getsockopt =	netlink_getsockopt,
2809 	.sendmsg =	netlink_sendmsg,
2810 	.recvmsg =	netlink_recvmsg,
2811 	.mmap =		sock_no_mmap,
2812 };
2813 
2814 static const struct net_proto_family netlink_family_ops = {
2815 	.family = PF_NETLINK,
2816 	.create = netlink_create,
2817 	.owner	= THIS_MODULE,	/* for consistency 8) */
2818 };
2819 
netlink_net_init(struct net * net)2820 static int __net_init netlink_net_init(struct net *net)
2821 {
2822 #ifdef CONFIG_PROC_FS
2823 	if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2824 			sizeof(struct nl_seq_iter)))
2825 		return -ENOMEM;
2826 #endif
2827 	return 0;
2828 }
2829 
netlink_net_exit(struct net * net)2830 static void __net_exit netlink_net_exit(struct net *net)
2831 {
2832 #ifdef CONFIG_PROC_FS
2833 	remove_proc_entry("netlink", net->proc_net);
2834 #endif
2835 }
2836 
netlink_add_usersock_entry(void)2837 static void __init netlink_add_usersock_entry(void)
2838 {
2839 	struct listeners *listeners;
2840 	int groups = 32;
2841 
2842 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2843 	if (!listeners)
2844 		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2845 
2846 	netlink_table_grab();
2847 
2848 	nl_table[NETLINK_USERSOCK].groups = groups;
2849 	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2850 	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2851 	nl_table[NETLINK_USERSOCK].registered = 1;
2852 	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2853 
2854 	netlink_table_ungrab();
2855 }
2856 
2857 static struct pernet_operations __net_initdata netlink_net_ops = {
2858 	.init = netlink_net_init,
2859 	.exit = netlink_net_exit,
2860 };
2861 
netlink_hash(const void * data,u32 len,u32 seed)2862 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2863 {
2864 	const struct netlink_sock *nlk = data;
2865 	struct netlink_compare_arg arg;
2866 
2867 	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2868 	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2869 }
2870 
2871 static const struct rhashtable_params netlink_rhashtable_params = {
2872 	.head_offset = offsetof(struct netlink_sock, node),
2873 	.key_len = netlink_compare_arg_len,
2874 	.obj_hashfn = netlink_hash,
2875 	.obj_cmpfn = netlink_compare,
2876 	.automatic_shrinking = true,
2877 };
2878 
2879 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2880 BTF_ID_LIST(btf_netlink_sock_id)
2881 BTF_ID(struct, netlink_sock)
2882 
2883 static const struct bpf_iter_seq_info netlink_seq_info = {
2884 	.seq_ops		= &netlink_seq_ops,
2885 	.init_seq_private	= bpf_iter_init_seq_net,
2886 	.fini_seq_private	= bpf_iter_fini_seq_net,
2887 	.seq_priv_size		= sizeof(struct nl_seq_iter),
2888 };
2889 
2890 static struct bpf_iter_reg netlink_reg_info = {
2891 	.target			= "netlink",
2892 	.ctx_arg_info_size	= 1,
2893 	.ctx_arg_info		= {
2894 		{ offsetof(struct bpf_iter__netlink, sk),
2895 		  PTR_TO_BTF_ID_OR_NULL },
2896 	},
2897 	.seq_info		= &netlink_seq_info,
2898 };
2899 
bpf_iter_register(void)2900 static int __init bpf_iter_register(void)
2901 {
2902 	netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2903 	return bpf_iter_reg_target(&netlink_reg_info);
2904 }
2905 #endif
2906 
netlink_proto_init(void)2907 static int __init netlink_proto_init(void)
2908 {
2909 	int i;
2910 	int err = proto_register(&netlink_proto, 0);
2911 
2912 	if (err != 0)
2913 		goto out;
2914 
2915 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2916 	err = bpf_iter_register();
2917 	if (err)
2918 		goto out;
2919 #endif
2920 
2921 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2922 
2923 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2924 	if (!nl_table)
2925 		goto panic;
2926 
2927 	for (i = 0; i < MAX_LINKS; i++) {
2928 		if (rhashtable_init(&nl_table[i].hash,
2929 				    &netlink_rhashtable_params) < 0)
2930 			goto panic;
2931 	}
2932 
2933 	netlink_add_usersock_entry();
2934 
2935 	sock_register(&netlink_family_ops);
2936 	register_pernet_subsys(&netlink_net_ops);
2937 	register_pernet_subsys(&netlink_tap_net_ops);
2938 	/* The netlink device handler may be needed early. */
2939 	rtnetlink_init();
2940 out:
2941 	return err;
2942 panic:
2943 	panic("netlink_init: Cannot allocate nl_table\n");
2944 }
2945 
2946 core_initcall(netlink_proto_init);
2947