xref: /linux/net/netlink/af_netlink.c (revision 1e15510b71c99c6e49134d756df91069f7d18141)
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 	call_rcu(&nlk->rcu, deferred_put_nlk_sk);
799 	return 0;
800 }
801 
netlink_autobind(struct socket * sock)802 static int netlink_autobind(struct socket *sock)
803 {
804 	struct sock *sk = sock->sk;
805 	struct net *net = sock_net(sk);
806 	struct netlink_table *table = &nl_table[sk->sk_protocol];
807 	s32 portid = task_tgid_vnr(current);
808 	int err;
809 	s32 rover = -4096;
810 	bool ok;
811 
812 retry:
813 	cond_resched();
814 	rcu_read_lock();
815 	ok = !__netlink_lookup(table, portid, net);
816 	rcu_read_unlock();
817 	if (!ok) {
818 		/* Bind collision, search negative portid values. */
819 		if (rover == -4096)
820 			/* rover will be in range [S32_MIN, -4097] */
821 			rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
822 		else if (rover >= -4096)
823 			rover = -4097;
824 		portid = rover--;
825 		goto retry;
826 	}
827 
828 	err = netlink_insert(sk, portid);
829 	if (err == -EADDRINUSE)
830 		goto retry;
831 
832 	/* If 2 threads race to autobind, that is fine.  */
833 	if (err == -EBUSY)
834 		err = 0;
835 
836 	return err;
837 }
838 
839 /**
840  * __netlink_ns_capable - General netlink message capability test
841  * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
842  * @user_ns: The user namespace of the capability to use
843  * @cap: The capability to use
844  *
845  * Test to see if the opener of the socket we received the message
846  * from had when the netlink socket was created and the sender of the
847  * message has the capability @cap in the user namespace @user_ns.
848  */
__netlink_ns_capable(const struct netlink_skb_parms * nsp,struct user_namespace * user_ns,int cap)849 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
850 			struct user_namespace *user_ns, int cap)
851 {
852 	return ((nsp->flags & NETLINK_SKB_DST) ||
853 		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
854 		ns_capable(user_ns, cap);
855 }
856 EXPORT_SYMBOL(__netlink_ns_capable);
857 
858 /**
859  * netlink_ns_capable - General netlink message capability test
860  * @skb: socket buffer holding a netlink command from userspace
861  * @user_ns: The user namespace of the capability to use
862  * @cap: The capability to use
863  *
864  * Test to see if the opener of the socket we received the message
865  * from had when the netlink socket was created and the sender of the
866  * message has the capability @cap in the user namespace @user_ns.
867  */
netlink_ns_capable(const struct sk_buff * skb,struct user_namespace * user_ns,int cap)868 bool netlink_ns_capable(const struct sk_buff *skb,
869 			struct user_namespace *user_ns, int cap)
870 {
871 	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
872 }
873 EXPORT_SYMBOL(netlink_ns_capable);
874 
875 /**
876  * netlink_capable - Netlink global message capability test
877  * @skb: socket buffer holding a netlink command from userspace
878  * @cap: The capability to use
879  *
880  * Test to see if the opener of the socket we received the message
881  * from had when the netlink socket was created and the sender of the
882  * message has the capability @cap in all user namespaces.
883  */
netlink_capable(const struct sk_buff * skb,int cap)884 bool netlink_capable(const struct sk_buff *skb, int cap)
885 {
886 	return netlink_ns_capable(skb, &init_user_ns, cap);
887 }
888 EXPORT_SYMBOL(netlink_capable);
889 
890 /**
891  * netlink_net_capable - Netlink network namespace message capability test
892  * @skb: socket buffer holding a netlink command from userspace
893  * @cap: The capability to use
894  *
895  * Test to see if the opener of the socket we received the message
896  * from had when the netlink socket was created and the sender of the
897  * message has the capability @cap over the network namespace of
898  * the socket we received the message from.
899  */
netlink_net_capable(const struct sk_buff * skb,int cap)900 bool netlink_net_capable(const struct sk_buff *skb, int cap)
901 {
902 	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
903 }
904 EXPORT_SYMBOL(netlink_net_capable);
905 
netlink_allowed(const struct socket * sock,unsigned int flag)906 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
907 {
908 	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
909 		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
910 }
911 
912 static void
netlink_update_subscriptions(struct sock * sk,unsigned int subscriptions)913 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
914 {
915 	struct netlink_sock *nlk = nlk_sk(sk);
916 
917 	if (nlk->subscriptions && !subscriptions)
918 		__sk_del_bind_node(sk);
919 	else if (!nlk->subscriptions && subscriptions)
920 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
921 	nlk->subscriptions = subscriptions;
922 }
923 
netlink_realloc_groups(struct sock * sk)924 static int netlink_realloc_groups(struct sock *sk)
925 {
926 	struct netlink_sock *nlk = nlk_sk(sk);
927 	unsigned int groups;
928 	unsigned long *new_groups;
929 	int err = 0;
930 
931 	netlink_table_grab();
932 
933 	groups = nl_table[sk->sk_protocol].groups;
934 	if (!nl_table[sk->sk_protocol].registered) {
935 		err = -ENOENT;
936 		goto out_unlock;
937 	}
938 
939 	if (nlk->ngroups >= groups)
940 		goto out_unlock;
941 
942 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
943 	if (new_groups == NULL) {
944 		err = -ENOMEM;
945 		goto out_unlock;
946 	}
947 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
948 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
949 
950 	nlk->groups = new_groups;
951 	nlk->ngroups = groups;
952  out_unlock:
953 	netlink_table_ungrab();
954 	return err;
955 }
956 
netlink_undo_bind(int group,long unsigned int groups,struct sock * sk)957 static void netlink_undo_bind(int group, long unsigned int groups,
958 			      struct sock *sk)
959 {
960 	struct netlink_sock *nlk = nlk_sk(sk);
961 	int undo;
962 
963 	if (!nlk->netlink_unbind)
964 		return;
965 
966 	for (undo = 0; undo < group; undo++)
967 		if (test_bit(undo, &groups))
968 			nlk->netlink_unbind(sock_net(sk), undo + 1);
969 }
970 
netlink_bind(struct socket * sock,struct sockaddr * addr,int addr_len)971 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
972 			int addr_len)
973 {
974 	struct sock *sk = sock->sk;
975 	struct net *net = sock_net(sk);
976 	struct netlink_sock *nlk = nlk_sk(sk);
977 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
978 	int err = 0;
979 	unsigned long groups;
980 	bool bound;
981 
982 	if (addr_len < sizeof(struct sockaddr_nl))
983 		return -EINVAL;
984 
985 	if (nladdr->nl_family != AF_NETLINK)
986 		return -EINVAL;
987 	groups = nladdr->nl_groups;
988 
989 	/* Only superuser is allowed to listen multicasts */
990 	if (groups) {
991 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
992 			return -EPERM;
993 		err = netlink_realloc_groups(sk);
994 		if (err)
995 			return err;
996 	}
997 
998 	if (nlk->ngroups < BITS_PER_LONG)
999 		groups &= (1UL << nlk->ngroups) - 1;
1000 
1001 	/* Paired with WRITE_ONCE() in netlink_insert() */
1002 	bound = READ_ONCE(nlk->bound);
1003 	if (bound) {
1004 		/* Ensure nlk->portid is up-to-date. */
1005 		smp_rmb();
1006 
1007 		if (nladdr->nl_pid != nlk->portid)
1008 			return -EINVAL;
1009 	}
1010 
1011 	if (nlk->netlink_bind && groups) {
1012 		int group;
1013 
1014 		/* nl_groups is a u32, so cap the maximum groups we can bind */
1015 		for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1016 			if (!test_bit(group, &groups))
1017 				continue;
1018 			err = nlk->netlink_bind(net, group + 1);
1019 			if (!err)
1020 				continue;
1021 			netlink_undo_bind(group, groups, sk);
1022 			return err;
1023 		}
1024 	}
1025 
1026 	/* No need for barriers here as we return to user-space without
1027 	 * using any of the bound attributes.
1028 	 */
1029 	netlink_lock_table();
1030 	if (!bound) {
1031 		err = nladdr->nl_pid ?
1032 			netlink_insert(sk, nladdr->nl_pid) :
1033 			netlink_autobind(sock);
1034 		if (err) {
1035 			netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1036 			goto unlock;
1037 		}
1038 	}
1039 
1040 	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1041 		goto unlock;
1042 	netlink_unlock_table();
1043 
1044 	netlink_table_grab();
1045 	netlink_update_subscriptions(sk, nlk->subscriptions +
1046 					 hweight32(groups) -
1047 					 hweight32(nlk->groups[0]));
1048 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1049 	netlink_update_listeners(sk);
1050 	netlink_table_ungrab();
1051 
1052 	return 0;
1053 
1054 unlock:
1055 	netlink_unlock_table();
1056 	return err;
1057 }
1058 
netlink_connect(struct socket * sock,struct sockaddr * addr,int alen,int flags)1059 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1060 			   int alen, int flags)
1061 {
1062 	int err = 0;
1063 	struct sock *sk = sock->sk;
1064 	struct netlink_sock *nlk = nlk_sk(sk);
1065 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1066 
1067 	if (alen < sizeof(addr->sa_family))
1068 		return -EINVAL;
1069 
1070 	if (addr->sa_family == AF_UNSPEC) {
1071 		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1072 		WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1073 		/* dst_portid and dst_group can be read locklessly */
1074 		WRITE_ONCE(nlk->dst_portid, 0);
1075 		WRITE_ONCE(nlk->dst_group, 0);
1076 		return 0;
1077 	}
1078 	if (addr->sa_family != AF_NETLINK)
1079 		return -EINVAL;
1080 
1081 	if (alen < sizeof(struct sockaddr_nl))
1082 		return -EINVAL;
1083 
1084 	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1085 	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1086 		return -EPERM;
1087 
1088 	/* No need for barriers here as we return to user-space without
1089 	 * using any of the bound attributes.
1090 	 * Paired with WRITE_ONCE() in netlink_insert().
1091 	 */
1092 	if (!READ_ONCE(nlk->bound))
1093 		err = netlink_autobind(sock);
1094 
1095 	if (err == 0) {
1096 		/* paired with READ_ONCE() in netlink_getsockbyportid() */
1097 		WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1098 		/* dst_portid and dst_group can be read locklessly */
1099 		WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1100 		WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1101 	}
1102 
1103 	return err;
1104 }
1105 
netlink_getname(struct socket * sock,struct sockaddr * addr,int peer)1106 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1107 			   int peer)
1108 {
1109 	struct sock *sk = sock->sk;
1110 	struct netlink_sock *nlk = nlk_sk(sk);
1111 	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1112 
1113 	nladdr->nl_family = AF_NETLINK;
1114 	nladdr->nl_pad = 0;
1115 
1116 	if (peer) {
1117 		/* Paired with WRITE_ONCE() in netlink_connect() */
1118 		nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1119 		nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1120 	} else {
1121 		/* Paired with WRITE_ONCE() in netlink_insert() */
1122 		nladdr->nl_pid = READ_ONCE(nlk->portid);
1123 		netlink_lock_table();
1124 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1125 		netlink_unlock_table();
1126 	}
1127 	return sizeof(*nladdr);
1128 }
1129 
netlink_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1130 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1131 			 unsigned long arg)
1132 {
1133 	/* try to hand this ioctl down to the NIC drivers.
1134 	 */
1135 	return -ENOIOCTLCMD;
1136 }
1137 
netlink_getsockbyportid(struct sock * ssk,u32 portid)1138 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1139 {
1140 	struct sock *sock;
1141 	struct netlink_sock *nlk;
1142 
1143 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1144 	if (!sock)
1145 		return ERR_PTR(-ECONNREFUSED);
1146 
1147 	/* Don't bother queuing skb if kernel socket has no input function */
1148 	nlk = nlk_sk(sock);
1149 	/* dst_portid and sk_state can be changed in netlink_connect() */
1150 	if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1151 	    READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1152 		sock_put(sock);
1153 		return ERR_PTR(-ECONNREFUSED);
1154 	}
1155 	return sock;
1156 }
1157 
netlink_getsockbyfd(int fd)1158 struct sock *netlink_getsockbyfd(int fd)
1159 {
1160 	CLASS(fd, f)(fd);
1161 	struct inode *inode;
1162 	struct sock *sock;
1163 
1164 	if (fd_empty(f))
1165 		return ERR_PTR(-EBADF);
1166 
1167 	inode = file_inode(fd_file(f));
1168 	if (!S_ISSOCK(inode->i_mode))
1169 		return ERR_PTR(-ENOTSOCK);
1170 
1171 	sock = SOCKET_I(inode)->sk;
1172 	if (sock->sk_family != AF_NETLINK)
1173 		return ERR_PTR(-EINVAL);
1174 
1175 	sock_hold(sock);
1176 	return sock;
1177 }
1178 
netlink_alloc_large_skb(unsigned int size,int broadcast)1179 struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1180 {
1181 	size_t head_size = SKB_HEAD_ALIGN(size);
1182 	struct sk_buff *skb;
1183 	void *data;
1184 
1185 	if (head_size <= PAGE_SIZE || broadcast)
1186 		return alloc_skb(size, GFP_KERNEL);
1187 
1188 	data = kvmalloc(head_size, GFP_KERNEL);
1189 	if (!data)
1190 		return NULL;
1191 
1192 	skb = __build_skb(data, head_size);
1193 	if (!skb)
1194 		kvfree(data);
1195 	else if (is_vmalloc_addr(data))
1196 		skb->destructor = netlink_skb_destructor;
1197 
1198 	return skb;
1199 }
1200 
1201 /*
1202  * Attach a skb to a netlink socket.
1203  * The caller must hold a reference to the destination socket. On error, the
1204  * reference is dropped. The skb is not send to the destination, just all
1205  * all error checks are performed and memory in the queue is reserved.
1206  * Return values:
1207  * < 0: error. skb freed, reference to sock dropped.
1208  * 0: continue
1209  * 1: repeat lookup - reference dropped while waiting for socket memory.
1210  */
netlink_attachskb(struct sock * sk,struct sk_buff * skb,long * timeo,struct sock * ssk)1211 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1212 		      long *timeo, struct sock *ssk)
1213 {
1214 	struct netlink_sock *nlk;
1215 
1216 	nlk = nlk_sk(sk);
1217 
1218 	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1219 	     test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1220 		DECLARE_WAITQUEUE(wait, current);
1221 		if (!*timeo) {
1222 			if (!ssk || netlink_is_kernel(ssk))
1223 				netlink_overrun(sk);
1224 			sock_put(sk);
1225 			kfree_skb(skb);
1226 			return -EAGAIN;
1227 		}
1228 
1229 		__set_current_state(TASK_INTERRUPTIBLE);
1230 		add_wait_queue(&nlk->wait, &wait);
1231 
1232 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1233 		     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1234 		    !sock_flag(sk, SOCK_DEAD))
1235 			*timeo = schedule_timeout(*timeo);
1236 
1237 		__set_current_state(TASK_RUNNING);
1238 		remove_wait_queue(&nlk->wait, &wait);
1239 		sock_put(sk);
1240 
1241 		if (signal_pending(current)) {
1242 			kfree_skb(skb);
1243 			return sock_intr_errno(*timeo);
1244 		}
1245 		return 1;
1246 	}
1247 	netlink_skb_set_owner_r(skb, sk);
1248 	return 0;
1249 }
1250 
__netlink_sendskb(struct sock * sk,struct sk_buff * skb)1251 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1252 {
1253 	int len = skb->len;
1254 
1255 	netlink_deliver_tap(sock_net(sk), skb);
1256 
1257 	skb_queue_tail(&sk->sk_receive_queue, skb);
1258 	sk->sk_data_ready(sk);
1259 	return len;
1260 }
1261 
netlink_sendskb(struct sock * sk,struct sk_buff * skb)1262 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1263 {
1264 	int len = __netlink_sendskb(sk, skb);
1265 
1266 	sock_put(sk);
1267 	return len;
1268 }
1269 
netlink_detachskb(struct sock * sk,struct sk_buff * skb)1270 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1271 {
1272 	kfree_skb(skb);
1273 	sock_put(sk);
1274 }
1275 
netlink_trim(struct sk_buff * skb,gfp_t allocation)1276 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1277 {
1278 	int delta;
1279 
1280 	skb_assert_len(skb);
1281 	WARN_ON(skb->sk != NULL);
1282 	delta = skb->end - skb->tail;
1283 	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1284 		return skb;
1285 
1286 	if (skb_shared(skb)) {
1287 		struct sk_buff *nskb = skb_clone(skb, allocation);
1288 		if (!nskb)
1289 			return skb;
1290 		consume_skb(skb);
1291 		skb = nskb;
1292 	}
1293 
1294 	pskb_expand_head(skb, 0, -delta,
1295 			 (allocation & ~__GFP_DIRECT_RECLAIM) |
1296 			 __GFP_NOWARN | __GFP_NORETRY);
1297 	return skb;
1298 }
1299 
netlink_unicast_kernel(struct sock * sk,struct sk_buff * skb,struct sock * ssk)1300 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1301 				  struct sock *ssk)
1302 {
1303 	int ret;
1304 	struct netlink_sock *nlk = nlk_sk(sk);
1305 
1306 	ret = -ECONNREFUSED;
1307 	if (nlk->netlink_rcv != NULL) {
1308 		ret = skb->len;
1309 		netlink_skb_set_owner_r(skb, sk);
1310 		NETLINK_CB(skb).sk = ssk;
1311 		netlink_deliver_tap_kernel(sk, ssk, skb);
1312 		nlk->netlink_rcv(skb);
1313 		consume_skb(skb);
1314 	} else {
1315 		kfree_skb(skb);
1316 	}
1317 	sock_put(sk);
1318 	return ret;
1319 }
1320 
netlink_unicast(struct sock * ssk,struct sk_buff * skb,u32 portid,int nonblock)1321 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1322 		    u32 portid, int nonblock)
1323 {
1324 	struct sock *sk;
1325 	int err;
1326 	long timeo;
1327 
1328 	skb = netlink_trim(skb, gfp_any());
1329 
1330 	timeo = sock_sndtimeo(ssk, nonblock);
1331 retry:
1332 	sk = netlink_getsockbyportid(ssk, portid);
1333 	if (IS_ERR(sk)) {
1334 		kfree_skb(skb);
1335 		return PTR_ERR(sk);
1336 	}
1337 	if (netlink_is_kernel(sk))
1338 		return netlink_unicast_kernel(sk, skb, ssk);
1339 
1340 	if (sk_filter(sk, skb)) {
1341 		err = skb->len;
1342 		kfree_skb(skb);
1343 		sock_put(sk);
1344 		return err;
1345 	}
1346 
1347 	err = netlink_attachskb(sk, skb, &timeo, ssk);
1348 	if (err == 1)
1349 		goto retry;
1350 	if (err)
1351 		return err;
1352 
1353 	return netlink_sendskb(sk, skb);
1354 }
1355 EXPORT_SYMBOL(netlink_unicast);
1356 
netlink_has_listeners(struct sock * sk,unsigned int group)1357 int netlink_has_listeners(struct sock *sk, unsigned int group)
1358 {
1359 	int res = 0;
1360 	struct listeners *listeners;
1361 
1362 	BUG_ON(!netlink_is_kernel(sk));
1363 
1364 	rcu_read_lock();
1365 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1366 
1367 	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1368 		res = test_bit(group - 1, listeners->masks);
1369 
1370 	rcu_read_unlock();
1371 
1372 	return res;
1373 }
1374 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1375 
netlink_strict_get_check(struct sk_buff * skb)1376 bool netlink_strict_get_check(struct sk_buff *skb)
1377 {
1378 	return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1379 }
1380 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1381 
netlink_broadcast_deliver(struct sock * sk,struct sk_buff * skb)1382 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1383 {
1384 	struct netlink_sock *nlk = nlk_sk(sk);
1385 
1386 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1387 	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1388 		netlink_skb_set_owner_r(skb, sk);
1389 		__netlink_sendskb(sk, skb);
1390 		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1391 	}
1392 	return -1;
1393 }
1394 
1395 struct netlink_broadcast_data {
1396 	struct sock *exclude_sk;
1397 	struct net *net;
1398 	u32 portid;
1399 	u32 group;
1400 	int failure;
1401 	int delivery_failure;
1402 	int congested;
1403 	int delivered;
1404 	gfp_t allocation;
1405 	struct sk_buff *skb, *skb2;
1406 	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1407 	void *tx_data;
1408 };
1409 
do_one_broadcast(struct sock * sk,struct netlink_broadcast_data * p)1410 static void do_one_broadcast(struct sock *sk,
1411 				    struct netlink_broadcast_data *p)
1412 {
1413 	struct netlink_sock *nlk = nlk_sk(sk);
1414 	int val;
1415 
1416 	if (p->exclude_sk == sk)
1417 		return;
1418 
1419 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1420 	    !test_bit(p->group - 1, nlk->groups))
1421 		return;
1422 
1423 	if (!net_eq(sock_net(sk), p->net)) {
1424 		if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1425 			return;
1426 
1427 		if (!peernet_has_id(sock_net(sk), p->net))
1428 			return;
1429 
1430 		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1431 				     CAP_NET_BROADCAST))
1432 			return;
1433 	}
1434 
1435 	if (p->failure) {
1436 		netlink_overrun(sk);
1437 		return;
1438 	}
1439 
1440 	sock_hold(sk);
1441 	if (p->skb2 == NULL) {
1442 		if (skb_shared(p->skb)) {
1443 			p->skb2 = skb_clone(p->skb, p->allocation);
1444 		} else {
1445 			p->skb2 = skb_get(p->skb);
1446 			/*
1447 			 * skb ownership may have been set when
1448 			 * delivered to a previous socket.
1449 			 */
1450 			skb_orphan(p->skb2);
1451 		}
1452 	}
1453 	if (p->skb2 == NULL) {
1454 		netlink_overrun(sk);
1455 		/* Clone failed. Notify ALL listeners. */
1456 		p->failure = 1;
1457 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1458 			p->delivery_failure = 1;
1459 		goto out;
1460 	}
1461 
1462 	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1463 		kfree_skb(p->skb2);
1464 		p->skb2 = NULL;
1465 		goto out;
1466 	}
1467 
1468 	if (sk_filter(sk, p->skb2)) {
1469 		kfree_skb(p->skb2);
1470 		p->skb2 = NULL;
1471 		goto out;
1472 	}
1473 	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1474 	if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1475 		NETLINK_CB(p->skb2).nsid_is_set = true;
1476 	val = netlink_broadcast_deliver(sk, p->skb2);
1477 	if (val < 0) {
1478 		netlink_overrun(sk);
1479 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1480 			p->delivery_failure = 1;
1481 	} else {
1482 		p->congested |= val;
1483 		p->delivered = 1;
1484 		p->skb2 = NULL;
1485 	}
1486 out:
1487 	sock_put(sk);
1488 }
1489 
netlink_broadcast_filtered(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation,netlink_filter_fn filter,void * filter_data)1490 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1491 			       u32 portid,
1492 			       u32 group, gfp_t allocation,
1493 			       netlink_filter_fn filter,
1494 			       void *filter_data)
1495 {
1496 	struct net *net = sock_net(ssk);
1497 	struct netlink_broadcast_data info;
1498 	struct sock *sk;
1499 
1500 	skb = netlink_trim(skb, allocation);
1501 
1502 	info.exclude_sk = ssk;
1503 	info.net = net;
1504 	info.portid = portid;
1505 	info.group = group;
1506 	info.failure = 0;
1507 	info.delivery_failure = 0;
1508 	info.congested = 0;
1509 	info.delivered = 0;
1510 	info.allocation = allocation;
1511 	info.skb = skb;
1512 	info.skb2 = NULL;
1513 	info.tx_filter = filter;
1514 	info.tx_data = filter_data;
1515 
1516 	/* While we sleep in clone, do not allow to change socket list */
1517 
1518 	netlink_lock_table();
1519 
1520 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1521 		do_one_broadcast(sk, &info);
1522 
1523 	consume_skb(skb);
1524 
1525 	netlink_unlock_table();
1526 
1527 	if (info.delivery_failure) {
1528 		kfree_skb(info.skb2);
1529 		return -ENOBUFS;
1530 	}
1531 	consume_skb(info.skb2);
1532 
1533 	if (info.delivered) {
1534 		if (info.congested && gfpflags_allow_blocking(allocation))
1535 			yield();
1536 		return 0;
1537 	}
1538 	return -ESRCH;
1539 }
1540 EXPORT_SYMBOL(netlink_broadcast_filtered);
1541 
netlink_broadcast(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation)1542 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1543 		      u32 group, gfp_t allocation)
1544 {
1545 	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1546 					  NULL, NULL);
1547 }
1548 EXPORT_SYMBOL(netlink_broadcast);
1549 
1550 struct netlink_set_err_data {
1551 	struct sock *exclude_sk;
1552 	u32 portid;
1553 	u32 group;
1554 	int code;
1555 };
1556 
do_one_set_err(struct sock * sk,struct netlink_set_err_data * p)1557 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1558 {
1559 	struct netlink_sock *nlk = nlk_sk(sk);
1560 	int ret = 0;
1561 
1562 	if (sk == p->exclude_sk)
1563 		goto out;
1564 
1565 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1566 		goto out;
1567 
1568 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1569 	    !test_bit(p->group - 1, nlk->groups))
1570 		goto out;
1571 
1572 	if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1573 		ret = 1;
1574 		goto out;
1575 	}
1576 
1577 	WRITE_ONCE(sk->sk_err, p->code);
1578 	sk_error_report(sk);
1579 out:
1580 	return ret;
1581 }
1582 
1583 /**
1584  * netlink_set_err - report error to broadcast listeners
1585  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1586  * @portid: the PORTID of a process that we want to skip (if any)
1587  * @group: the broadcast group that will notice the error
1588  * @code: error code, must be negative (as usual in kernelspace)
1589  *
1590  * This function returns the number of broadcast listeners that have set the
1591  * NETLINK_NO_ENOBUFS socket option.
1592  */
netlink_set_err(struct sock * ssk,u32 portid,u32 group,int code)1593 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1594 {
1595 	struct netlink_set_err_data info;
1596 	unsigned long flags;
1597 	struct sock *sk;
1598 	int ret = 0;
1599 
1600 	info.exclude_sk = ssk;
1601 	info.portid = portid;
1602 	info.group = group;
1603 	/* sk->sk_err wants a positive error value */
1604 	info.code = -code;
1605 
1606 	read_lock_irqsave(&nl_table_lock, flags);
1607 
1608 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1609 		ret += do_one_set_err(sk, &info);
1610 
1611 	read_unlock_irqrestore(&nl_table_lock, flags);
1612 	return ret;
1613 }
1614 EXPORT_SYMBOL(netlink_set_err);
1615 
1616 /* must be called with netlink table grabbed */
netlink_update_socket_mc(struct netlink_sock * nlk,unsigned int group,int is_new)1617 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1618 				     unsigned int group,
1619 				     int is_new)
1620 {
1621 	int old, new = !!is_new, subscriptions;
1622 
1623 	old = test_bit(group - 1, nlk->groups);
1624 	subscriptions = nlk->subscriptions - old + new;
1625 	__assign_bit(group - 1, nlk->groups, new);
1626 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1627 	netlink_update_listeners(&nlk->sk);
1628 }
1629 
netlink_setsockopt(struct socket * sock,int level,int optname,sockptr_t optval,unsigned int optlen)1630 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1631 			      sockptr_t optval, unsigned int optlen)
1632 {
1633 	struct sock *sk = sock->sk;
1634 	struct netlink_sock *nlk = nlk_sk(sk);
1635 	unsigned int val = 0;
1636 	int nr = -1;
1637 
1638 	if (level != SOL_NETLINK)
1639 		return -ENOPROTOOPT;
1640 
1641 	if (optlen >= sizeof(int) &&
1642 	    copy_from_sockptr(&val, optval, sizeof(val)))
1643 		return -EFAULT;
1644 
1645 	switch (optname) {
1646 	case NETLINK_PKTINFO:
1647 		nr = NETLINK_F_RECV_PKTINFO;
1648 		break;
1649 	case NETLINK_ADD_MEMBERSHIP:
1650 	case NETLINK_DROP_MEMBERSHIP: {
1651 		int err;
1652 
1653 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1654 			return -EPERM;
1655 		err = netlink_realloc_groups(sk);
1656 		if (err)
1657 			return err;
1658 		if (!val || val - 1 >= nlk->ngroups)
1659 			return -EINVAL;
1660 		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1661 			err = nlk->netlink_bind(sock_net(sk), val);
1662 			if (err)
1663 				return err;
1664 		}
1665 		netlink_table_grab();
1666 		netlink_update_socket_mc(nlk, val,
1667 					 optname == NETLINK_ADD_MEMBERSHIP);
1668 		netlink_table_ungrab();
1669 		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1670 			nlk->netlink_unbind(sock_net(sk), val);
1671 
1672 		break;
1673 	}
1674 	case NETLINK_BROADCAST_ERROR:
1675 		nr = NETLINK_F_BROADCAST_SEND_ERROR;
1676 		break;
1677 	case NETLINK_NO_ENOBUFS:
1678 		assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1679 		if (val) {
1680 			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1681 			wake_up_interruptible(&nlk->wait);
1682 		}
1683 		break;
1684 	case NETLINK_LISTEN_ALL_NSID:
1685 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1686 			return -EPERM;
1687 		nr = NETLINK_F_LISTEN_ALL_NSID;
1688 		break;
1689 	case NETLINK_CAP_ACK:
1690 		nr = NETLINK_F_CAP_ACK;
1691 		break;
1692 	case NETLINK_EXT_ACK:
1693 		nr = NETLINK_F_EXT_ACK;
1694 		break;
1695 	case NETLINK_GET_STRICT_CHK:
1696 		nr = NETLINK_F_STRICT_CHK;
1697 		break;
1698 	default:
1699 		return -ENOPROTOOPT;
1700 	}
1701 	if (nr >= 0)
1702 		assign_bit(nr, &nlk->flags, val);
1703 	return 0;
1704 }
1705 
netlink_getsockopt(struct socket * sock,int level,int optname,char __user * optval,int __user * optlen)1706 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1707 			      char __user *optval, int __user *optlen)
1708 {
1709 	struct sock *sk = sock->sk;
1710 	struct netlink_sock *nlk = nlk_sk(sk);
1711 	unsigned int flag;
1712 	int len, val;
1713 
1714 	if (level != SOL_NETLINK)
1715 		return -ENOPROTOOPT;
1716 
1717 	if (get_user(len, optlen))
1718 		return -EFAULT;
1719 	if (len < 0)
1720 		return -EINVAL;
1721 
1722 	switch (optname) {
1723 	case NETLINK_PKTINFO:
1724 		flag = NETLINK_F_RECV_PKTINFO;
1725 		break;
1726 	case NETLINK_BROADCAST_ERROR:
1727 		flag = NETLINK_F_BROADCAST_SEND_ERROR;
1728 		break;
1729 	case NETLINK_NO_ENOBUFS:
1730 		flag = NETLINK_F_RECV_NO_ENOBUFS;
1731 		break;
1732 	case NETLINK_LIST_MEMBERSHIPS: {
1733 		int pos, idx, shift, err = 0;
1734 
1735 		netlink_lock_table();
1736 		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1737 			if (len - pos < sizeof(u32))
1738 				break;
1739 
1740 			idx = pos / sizeof(unsigned long);
1741 			shift = (pos % sizeof(unsigned long)) * 8;
1742 			if (put_user((u32)(nlk->groups[idx] >> shift),
1743 				     (u32 __user *)(optval + pos))) {
1744 				err = -EFAULT;
1745 				break;
1746 			}
1747 		}
1748 		if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1749 			err = -EFAULT;
1750 		netlink_unlock_table();
1751 		return err;
1752 	}
1753 	case NETLINK_LISTEN_ALL_NSID:
1754 		flag = NETLINK_F_LISTEN_ALL_NSID;
1755 		break;
1756 	case NETLINK_CAP_ACK:
1757 		flag = NETLINK_F_CAP_ACK;
1758 		break;
1759 	case NETLINK_EXT_ACK:
1760 		flag = NETLINK_F_EXT_ACK;
1761 		break;
1762 	case NETLINK_GET_STRICT_CHK:
1763 		flag = NETLINK_F_STRICT_CHK;
1764 		break;
1765 	default:
1766 		return -ENOPROTOOPT;
1767 	}
1768 
1769 	if (len < sizeof(int))
1770 		return -EINVAL;
1771 
1772 	len = sizeof(int);
1773 	val = test_bit(flag, &nlk->flags);
1774 
1775 	if (put_user(len, optlen) ||
1776 	    copy_to_user(optval, &val, len))
1777 		return -EFAULT;
1778 
1779 	return 0;
1780 }
1781 
netlink_cmsg_recv_pktinfo(struct msghdr * msg,struct sk_buff * skb)1782 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1783 {
1784 	struct nl_pktinfo info;
1785 
1786 	info.group = NETLINK_CB(skb).dst_group;
1787 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1788 }
1789 
netlink_cmsg_listen_all_nsid(struct sock * sk,struct msghdr * msg,struct sk_buff * skb)1790 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1791 					 struct sk_buff *skb)
1792 {
1793 	if (!NETLINK_CB(skb).nsid_is_set)
1794 		return;
1795 
1796 	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1797 		 &NETLINK_CB(skb).nsid);
1798 }
1799 
netlink_sendmsg(struct socket * sock,struct msghdr * msg,size_t len)1800 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1801 {
1802 	struct sock *sk = sock->sk;
1803 	struct netlink_sock *nlk = nlk_sk(sk);
1804 	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1805 	u32 dst_portid;
1806 	u32 dst_group;
1807 	struct sk_buff *skb;
1808 	int err;
1809 	struct scm_cookie scm;
1810 	u32 netlink_skb_flags = 0;
1811 
1812 	if (msg->msg_flags & MSG_OOB)
1813 		return -EOPNOTSUPP;
1814 
1815 	if (len == 0) {
1816 		pr_warn_once("Zero length message leads to an empty skb\n");
1817 		return -ENODATA;
1818 	}
1819 
1820 	err = scm_send(sock, msg, &scm, true);
1821 	if (err < 0)
1822 		return err;
1823 
1824 	if (msg->msg_namelen) {
1825 		err = -EINVAL;
1826 		if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1827 			goto out;
1828 		if (addr->nl_family != AF_NETLINK)
1829 			goto out;
1830 		dst_portid = addr->nl_pid;
1831 		dst_group = ffs(addr->nl_groups);
1832 		err =  -EPERM;
1833 		if ((dst_group || dst_portid) &&
1834 		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1835 			goto out;
1836 		netlink_skb_flags |= NETLINK_SKB_DST;
1837 	} else {
1838 		/* Paired with WRITE_ONCE() in netlink_connect() */
1839 		dst_portid = READ_ONCE(nlk->dst_portid);
1840 		dst_group = READ_ONCE(nlk->dst_group);
1841 	}
1842 
1843 	/* Paired with WRITE_ONCE() in netlink_insert() */
1844 	if (!READ_ONCE(nlk->bound)) {
1845 		err = netlink_autobind(sock);
1846 		if (err)
1847 			goto out;
1848 	} else {
1849 		/* Ensure nlk is hashed and visible. */
1850 		smp_rmb();
1851 	}
1852 
1853 	err = -EMSGSIZE;
1854 	if (len > sk->sk_sndbuf - 32)
1855 		goto out;
1856 	err = -ENOBUFS;
1857 	skb = netlink_alloc_large_skb(len, dst_group);
1858 	if (skb == NULL)
1859 		goto out;
1860 
1861 	NETLINK_CB(skb).portid	= nlk->portid;
1862 	NETLINK_CB(skb).dst_group = dst_group;
1863 	NETLINK_CB(skb).creds	= scm.creds;
1864 	NETLINK_CB(skb).flags	= netlink_skb_flags;
1865 
1866 	err = -EFAULT;
1867 	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1868 		kfree_skb(skb);
1869 		goto out;
1870 	}
1871 
1872 	err = security_netlink_send(sk, skb);
1873 	if (err) {
1874 		kfree_skb(skb);
1875 		goto out;
1876 	}
1877 
1878 	if (dst_group) {
1879 		refcount_inc(&skb->users);
1880 		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1881 	}
1882 	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1883 
1884 out:
1885 	scm_destroy(&scm);
1886 	return err;
1887 }
1888 
netlink_recvmsg(struct socket * sock,struct msghdr * msg,size_t len,int flags)1889 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1890 			   int flags)
1891 {
1892 	struct scm_cookie scm;
1893 	struct sock *sk = sock->sk;
1894 	struct netlink_sock *nlk = nlk_sk(sk);
1895 	size_t copied, max_recvmsg_len;
1896 	struct sk_buff *skb, *data_skb;
1897 	int err, ret;
1898 
1899 	if (flags & MSG_OOB)
1900 		return -EOPNOTSUPP;
1901 
1902 	copied = 0;
1903 
1904 	skb = skb_recv_datagram(sk, flags, &err);
1905 	if (skb == NULL)
1906 		goto out;
1907 
1908 	data_skb = skb;
1909 
1910 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1911 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1912 		/*
1913 		 * If this skb has a frag_list, then here that means that we
1914 		 * will have to use the frag_list skb's data for compat tasks
1915 		 * and the regular skb's data for normal (non-compat) tasks.
1916 		 *
1917 		 * If we need to send the compat skb, assign it to the
1918 		 * 'data_skb' variable so that it will be used below for data
1919 		 * copying. We keep 'skb' for everything else, including
1920 		 * freeing both later.
1921 		 */
1922 		if (flags & MSG_CMSG_COMPAT)
1923 			data_skb = skb_shinfo(skb)->frag_list;
1924 	}
1925 #endif
1926 
1927 	/* Record the max length of recvmsg() calls for future allocations */
1928 	max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1929 	max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1930 				SKB_WITH_OVERHEAD(32768));
1931 	WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1932 
1933 	copied = data_skb->len;
1934 	if (len < copied) {
1935 		msg->msg_flags |= MSG_TRUNC;
1936 		copied = len;
1937 	}
1938 
1939 	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1940 
1941 	if (msg->msg_name) {
1942 		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1943 		addr->nl_family = AF_NETLINK;
1944 		addr->nl_pad    = 0;
1945 		addr->nl_pid	= NETLINK_CB(skb).portid;
1946 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1947 		msg->msg_namelen = sizeof(*addr);
1948 	}
1949 
1950 	if (nlk_test_bit(RECV_PKTINFO, sk))
1951 		netlink_cmsg_recv_pktinfo(msg, skb);
1952 	if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1953 		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1954 
1955 	memset(&scm, 0, sizeof(scm));
1956 	scm.creds = *NETLINK_CREDS(skb);
1957 	if (flags & MSG_TRUNC)
1958 		copied = data_skb->len;
1959 
1960 	skb_free_datagram(sk, skb);
1961 
1962 	if (READ_ONCE(nlk->cb_running) &&
1963 	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1964 		ret = netlink_dump(sk, false);
1965 		if (ret) {
1966 			WRITE_ONCE(sk->sk_err, -ret);
1967 			sk_error_report(sk);
1968 		}
1969 	}
1970 
1971 	scm_recv(sock, msg, &scm, flags);
1972 out:
1973 	netlink_rcv_wake(sk);
1974 	return err ? : copied;
1975 }
1976 
netlink_data_ready(struct sock * sk)1977 static void netlink_data_ready(struct sock *sk)
1978 {
1979 	BUG();
1980 }
1981 
1982 /*
1983  *	We export these functions to other modules. They provide a
1984  *	complete set of kernel non-blocking support for message
1985  *	queueing.
1986  */
1987 
1988 struct sock *
__netlink_kernel_create(struct net * net,int unit,struct module * module,struct netlink_kernel_cfg * cfg)1989 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1990 			struct netlink_kernel_cfg *cfg)
1991 {
1992 	struct socket *sock;
1993 	struct sock *sk;
1994 	struct netlink_sock *nlk;
1995 	struct listeners *listeners = NULL;
1996 	unsigned int groups;
1997 
1998 	BUG_ON(!nl_table);
1999 
2000 	if (unit < 0 || unit >= MAX_LINKS)
2001 		return NULL;
2002 
2003 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2004 		return NULL;
2005 
2006 	if (__netlink_create(net, sock, unit, 1) < 0)
2007 		goto out_sock_release_nosk;
2008 
2009 	sk = sock->sk;
2010 
2011 	if (!cfg || cfg->groups < 32)
2012 		groups = 32;
2013 	else
2014 		groups = cfg->groups;
2015 
2016 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2017 	if (!listeners)
2018 		goto out_sock_release;
2019 
2020 	sk->sk_data_ready = netlink_data_ready;
2021 	if (cfg && cfg->input)
2022 		nlk_sk(sk)->netlink_rcv = cfg->input;
2023 
2024 	if (netlink_insert(sk, 0))
2025 		goto out_sock_release;
2026 
2027 	nlk = nlk_sk(sk);
2028 	set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2029 
2030 	netlink_table_grab();
2031 	if (!nl_table[unit].registered) {
2032 		nl_table[unit].groups = groups;
2033 		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2034 		nl_table[unit].module = module;
2035 		if (cfg) {
2036 			nl_table[unit].bind = cfg->bind;
2037 			nl_table[unit].unbind = cfg->unbind;
2038 			nl_table[unit].release = cfg->release;
2039 			nl_table[unit].flags = cfg->flags;
2040 		}
2041 		nl_table[unit].registered = 1;
2042 	} else {
2043 		kfree(listeners);
2044 		nl_table[unit].registered++;
2045 	}
2046 	netlink_table_ungrab();
2047 	return sk;
2048 
2049 out_sock_release:
2050 	kfree(listeners);
2051 	netlink_kernel_release(sk);
2052 	return NULL;
2053 
2054 out_sock_release_nosk:
2055 	sock_release(sock);
2056 	return NULL;
2057 }
2058 EXPORT_SYMBOL(__netlink_kernel_create);
2059 
2060 void
netlink_kernel_release(struct sock * sk)2061 netlink_kernel_release(struct sock *sk)
2062 {
2063 	if (sk == NULL || sk->sk_socket == NULL)
2064 		return;
2065 
2066 	sock_release(sk->sk_socket);
2067 }
2068 EXPORT_SYMBOL(netlink_kernel_release);
2069 
__netlink_change_ngroups(struct sock * sk,unsigned int groups)2070 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2071 {
2072 	struct listeners *new, *old;
2073 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2074 
2075 	if (groups < 32)
2076 		groups = 32;
2077 
2078 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2079 		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2080 		if (!new)
2081 			return -ENOMEM;
2082 		old = nl_deref_protected(tbl->listeners);
2083 		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2084 		rcu_assign_pointer(tbl->listeners, new);
2085 
2086 		kfree_rcu(old, rcu);
2087 	}
2088 	tbl->groups = groups;
2089 
2090 	return 0;
2091 }
2092 
2093 /**
2094  * netlink_change_ngroups - change number of multicast groups
2095  *
2096  * This changes the number of multicast groups that are available
2097  * on a certain netlink family. Note that it is not possible to
2098  * change the number of groups to below 32. Also note that it does
2099  * not implicitly call netlink_clear_multicast_users() when the
2100  * number of groups is reduced.
2101  *
2102  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2103  * @groups: The new number of groups.
2104  */
netlink_change_ngroups(struct sock * sk,unsigned int groups)2105 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2106 {
2107 	int err;
2108 
2109 	netlink_table_grab();
2110 	err = __netlink_change_ngroups(sk, groups);
2111 	netlink_table_ungrab();
2112 
2113 	return err;
2114 }
2115 
__netlink_clear_multicast_users(struct sock * ksk,unsigned int group)2116 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2117 {
2118 	struct sock *sk;
2119 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2120 	struct hlist_node *tmp;
2121 
2122 	sk_for_each_bound_safe(sk, tmp, &tbl->mc_list)
2123 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2124 }
2125 
2126 struct nlmsghdr *
__nlmsg_put(struct sk_buff * skb,u32 portid,u32 seq,int type,int len,int flags)2127 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2128 {
2129 	struct nlmsghdr *nlh;
2130 	int size = nlmsg_msg_size(len);
2131 
2132 	nlh = skb_put(skb, NLMSG_ALIGN(size));
2133 	nlh->nlmsg_type = type;
2134 	nlh->nlmsg_len = size;
2135 	nlh->nlmsg_flags = flags;
2136 	nlh->nlmsg_pid = portid;
2137 	nlh->nlmsg_seq = seq;
2138 	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2139 		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2140 	return nlh;
2141 }
2142 EXPORT_SYMBOL(__nlmsg_put);
2143 
2144 static size_t
netlink_ack_tlv_len(struct netlink_sock * nlk,int err,const struct netlink_ext_ack * extack)2145 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2146 		    const struct netlink_ext_ack *extack)
2147 {
2148 	size_t tlvlen;
2149 
2150 	if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2151 		return 0;
2152 
2153 	tlvlen = 0;
2154 	if (extack->_msg)
2155 		tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2156 	if (extack->cookie_len)
2157 		tlvlen += nla_total_size(extack->cookie_len);
2158 
2159 	/* Following attributes are only reported as error (not warning) */
2160 	if (!err)
2161 		return tlvlen;
2162 
2163 	if (extack->bad_attr)
2164 		tlvlen += nla_total_size(sizeof(u32));
2165 	if (extack->policy)
2166 		tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2167 	if (extack->miss_type)
2168 		tlvlen += nla_total_size(sizeof(u32));
2169 	if (extack->miss_nest)
2170 		tlvlen += nla_total_size(sizeof(u32));
2171 
2172 	return tlvlen;
2173 }
2174 
nlmsg_check_in_payload(const struct nlmsghdr * nlh,const void * addr)2175 static bool nlmsg_check_in_payload(const struct nlmsghdr *nlh, const void *addr)
2176 {
2177 	return !WARN_ON(addr < nlmsg_data(nlh) ||
2178 			addr - (const void *) nlh >= nlh->nlmsg_len);
2179 }
2180 
2181 static void
netlink_ack_tlv_fill(struct sk_buff * skb,const struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2182 netlink_ack_tlv_fill(struct sk_buff *skb, const struct nlmsghdr *nlh, int err,
2183 		     const struct netlink_ext_ack *extack)
2184 {
2185 	if (extack->_msg)
2186 		WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2187 	if (extack->cookie_len)
2188 		WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2189 				extack->cookie_len, extack->cookie));
2190 
2191 	if (!err)
2192 		return;
2193 
2194 	if (extack->bad_attr && nlmsg_check_in_payload(nlh, extack->bad_attr))
2195 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2196 				    (u8 *)extack->bad_attr - (const u8 *)nlh));
2197 	if (extack->policy)
2198 		netlink_policy_dump_write_attr(skb, extack->policy,
2199 					       NLMSGERR_ATTR_POLICY);
2200 	if (extack->miss_type)
2201 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2202 				    extack->miss_type));
2203 	if (extack->miss_nest && nlmsg_check_in_payload(nlh, extack->miss_nest))
2204 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2205 				    (u8 *)extack->miss_nest - (const u8 *)nlh));
2206 }
2207 
2208 /*
2209  * It looks a bit ugly.
2210  * It would be better to create kernel thread.
2211  */
2212 
netlink_dump_done(struct netlink_sock * nlk,struct sk_buff * skb,struct netlink_callback * cb,struct netlink_ext_ack * extack)2213 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2214 			     struct netlink_callback *cb,
2215 			     struct netlink_ext_ack *extack)
2216 {
2217 	struct nlmsghdr *nlh;
2218 	size_t extack_len;
2219 
2220 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2221 			       NLM_F_MULTI | cb->answer_flags);
2222 	if (WARN_ON(!nlh))
2223 		return -ENOBUFS;
2224 
2225 	nl_dump_check_consistent(cb, nlh);
2226 	memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2227 
2228 	extack_len = netlink_ack_tlv_len(nlk, nlk->dump_done_errno, extack);
2229 	if (extack_len) {
2230 		nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2231 		if (skb_tailroom(skb) >= extack_len) {
2232 			netlink_ack_tlv_fill(skb, cb->nlh,
2233 					     nlk->dump_done_errno, extack);
2234 			nlmsg_end(skb, nlh);
2235 		}
2236 	}
2237 
2238 	return 0;
2239 }
2240 
netlink_dump(struct sock * sk,bool lock_taken)2241 static int netlink_dump(struct sock *sk, bool lock_taken)
2242 {
2243 	struct netlink_sock *nlk = nlk_sk(sk);
2244 	struct netlink_ext_ack extack = {};
2245 	struct netlink_callback *cb;
2246 	struct sk_buff *skb = NULL;
2247 	size_t max_recvmsg_len;
2248 	struct module *module;
2249 	int err = -ENOBUFS;
2250 	int alloc_min_size;
2251 	int alloc_size;
2252 
2253 	if (!lock_taken)
2254 		mutex_lock(&nlk->nl_cb_mutex);
2255 	if (!nlk->cb_running) {
2256 		err = -EINVAL;
2257 		goto errout_skb;
2258 	}
2259 
2260 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2261 		goto errout_skb;
2262 
2263 	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2264 	 * required, but it makes sense to _attempt_ a 32KiB allocation
2265 	 * to reduce number of system calls on dump operations, if user
2266 	 * ever provided a big enough buffer.
2267 	 */
2268 	cb = &nlk->cb;
2269 	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2270 
2271 	max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2272 	if (alloc_min_size < max_recvmsg_len) {
2273 		alloc_size = max_recvmsg_len;
2274 		skb = alloc_skb(alloc_size,
2275 				(GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2276 				__GFP_NOWARN | __GFP_NORETRY);
2277 	}
2278 	if (!skb) {
2279 		alloc_size = alloc_min_size;
2280 		skb = alloc_skb(alloc_size, GFP_KERNEL);
2281 	}
2282 	if (!skb)
2283 		goto errout_skb;
2284 
2285 	/* Trim skb to allocated size. User is expected to provide buffer as
2286 	 * large as max(min_dump_alloc, 32KiB (max_recvmsg_len capped at
2287 	 * netlink_recvmsg())). dump will pack as many smaller messages as
2288 	 * could fit within the allocated skb. skb is typically allocated
2289 	 * with larger space than required (could be as much as near 2x the
2290 	 * requested size with align to next power of 2 approach). Allowing
2291 	 * dump to use the excess space makes it difficult for a user to have a
2292 	 * reasonable static buffer based on the expected largest dump of a
2293 	 * single netdev. The outcome is MSG_TRUNC error.
2294 	 */
2295 	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2296 
2297 	/* Make sure malicious BPF programs can not read unitialized memory
2298 	 * from skb->head -> skb->data
2299 	 */
2300 	skb_reset_network_header(skb);
2301 	skb_reset_mac_header(skb);
2302 
2303 	netlink_skb_set_owner_r(skb, sk);
2304 
2305 	if (nlk->dump_done_errno > 0) {
2306 		cb->extack = &extack;
2307 
2308 		nlk->dump_done_errno = cb->dump(skb, cb);
2309 
2310 		/* EMSGSIZE plus something already in the skb means
2311 		 * that there's more to dump but current skb has filled up.
2312 		 * If the callback really wants to return EMSGSIZE to user space
2313 		 * it needs to do so again, on the next cb->dump() call,
2314 		 * without putting data in the skb.
2315 		 */
2316 		if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2317 			nlk->dump_done_errno = skb->len;
2318 
2319 		cb->extack = NULL;
2320 	}
2321 
2322 	if (nlk->dump_done_errno > 0 ||
2323 	    skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2324 		mutex_unlock(&nlk->nl_cb_mutex);
2325 
2326 		if (sk_filter(sk, skb))
2327 			kfree_skb(skb);
2328 		else
2329 			__netlink_sendskb(sk, skb);
2330 		return 0;
2331 	}
2332 
2333 	if (netlink_dump_done(nlk, skb, cb, &extack))
2334 		goto errout_skb;
2335 
2336 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2337 	/* frag_list skb's data is used for compat tasks
2338 	 * and the regular skb's data for normal (non-compat) tasks.
2339 	 * See netlink_recvmsg().
2340 	 */
2341 	if (unlikely(skb_shinfo(skb)->frag_list)) {
2342 		if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2343 			goto errout_skb;
2344 	}
2345 #endif
2346 
2347 	if (sk_filter(sk, skb))
2348 		kfree_skb(skb);
2349 	else
2350 		__netlink_sendskb(sk, skb);
2351 
2352 	if (cb->done)
2353 		cb->done(cb);
2354 
2355 	WRITE_ONCE(nlk->cb_running, false);
2356 	module = cb->module;
2357 	skb = cb->skb;
2358 	mutex_unlock(&nlk->nl_cb_mutex);
2359 	module_put(module);
2360 	consume_skb(skb);
2361 	return 0;
2362 
2363 errout_skb:
2364 	mutex_unlock(&nlk->nl_cb_mutex);
2365 	kfree_skb(skb);
2366 	return err;
2367 }
2368 
__netlink_dump_start(struct sock * ssk,struct sk_buff * skb,const struct nlmsghdr * nlh,struct netlink_dump_control * control)2369 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2370 			 const struct nlmsghdr *nlh,
2371 			 struct netlink_dump_control *control)
2372 {
2373 	struct netlink_callback *cb;
2374 	struct netlink_sock *nlk;
2375 	struct sock *sk;
2376 	int ret;
2377 
2378 	refcount_inc(&skb->users);
2379 
2380 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2381 	if (sk == NULL) {
2382 		ret = -ECONNREFUSED;
2383 		goto error_free;
2384 	}
2385 
2386 	nlk = nlk_sk(sk);
2387 	mutex_lock(&nlk->nl_cb_mutex);
2388 	/* A dump is in progress... */
2389 	if (nlk->cb_running) {
2390 		ret = -EBUSY;
2391 		goto error_unlock;
2392 	}
2393 	/* add reference of module which cb->dump belongs to */
2394 	if (!try_module_get(control->module)) {
2395 		ret = -EPROTONOSUPPORT;
2396 		goto error_unlock;
2397 	}
2398 
2399 	cb = &nlk->cb;
2400 	memset(cb, 0, sizeof(*cb));
2401 	cb->dump = control->dump;
2402 	cb->done = control->done;
2403 	cb->nlh = nlh;
2404 	cb->data = control->data;
2405 	cb->module = control->module;
2406 	cb->min_dump_alloc = control->min_dump_alloc;
2407 	cb->flags = control->flags;
2408 	cb->skb = skb;
2409 
2410 	cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2411 
2412 	if (control->start) {
2413 		cb->extack = control->extack;
2414 		ret = control->start(cb);
2415 		cb->extack = NULL;
2416 		if (ret)
2417 			goto error_put;
2418 	}
2419 
2420 	WRITE_ONCE(nlk->cb_running, true);
2421 	nlk->dump_done_errno = INT_MAX;
2422 
2423 	ret = netlink_dump(sk, true);
2424 
2425 	sock_put(sk);
2426 
2427 	if (ret)
2428 		return ret;
2429 
2430 	/* We successfully started a dump, by returning -EINTR we
2431 	 * signal not to send ACK even if it was requested.
2432 	 */
2433 	return -EINTR;
2434 
2435 error_put:
2436 	module_put(control->module);
2437 error_unlock:
2438 	sock_put(sk);
2439 	mutex_unlock(&nlk->nl_cb_mutex);
2440 error_free:
2441 	kfree_skb(skb);
2442 	return ret;
2443 }
2444 EXPORT_SYMBOL(__netlink_dump_start);
2445 
netlink_ack(struct sk_buff * in_skb,struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2446 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2447 		 const struct netlink_ext_ack *extack)
2448 {
2449 	struct sk_buff *skb;
2450 	struct nlmsghdr *rep;
2451 	struct nlmsgerr *errmsg;
2452 	size_t payload = sizeof(*errmsg);
2453 	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2454 	unsigned int flags = 0;
2455 	size_t tlvlen;
2456 
2457 	/* Error messages get the original request appened, unless the user
2458 	 * requests to cap the error message, and get extra error data if
2459 	 * requested.
2460 	 */
2461 	if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2462 		payload += nlmsg_len(nlh);
2463 	else
2464 		flags |= NLM_F_CAPPED;
2465 
2466 	tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2467 	if (tlvlen)
2468 		flags |= NLM_F_ACK_TLVS;
2469 
2470 	skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2471 	if (!skb)
2472 		goto err_skb;
2473 
2474 	rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2475 			NLMSG_ERROR, sizeof(*errmsg), flags);
2476 	if (!rep)
2477 		goto err_bad_put;
2478 	errmsg = nlmsg_data(rep);
2479 	errmsg->error = err;
2480 	errmsg->msg = *nlh;
2481 
2482 	if (!(flags & NLM_F_CAPPED)) {
2483 		if (!nlmsg_append(skb, nlmsg_len(nlh)))
2484 			goto err_bad_put;
2485 
2486 		memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2487 		       nlmsg_len(nlh));
2488 	}
2489 
2490 	if (tlvlen)
2491 		netlink_ack_tlv_fill(skb, nlh, err, extack);
2492 
2493 	nlmsg_end(skb, rep);
2494 
2495 	nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2496 
2497 	return;
2498 
2499 err_bad_put:
2500 	nlmsg_free(skb);
2501 err_skb:
2502 	WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2503 	sk_error_report(NETLINK_CB(in_skb).sk);
2504 }
2505 EXPORT_SYMBOL(netlink_ack);
2506 
netlink_rcv_skb(struct sk_buff * skb,int (* cb)(struct sk_buff *,struct nlmsghdr *,struct netlink_ext_ack *))2507 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2508 						   struct nlmsghdr *,
2509 						   struct netlink_ext_ack *))
2510 {
2511 	struct netlink_ext_ack extack;
2512 	struct nlmsghdr *nlh;
2513 	int err;
2514 
2515 	while (skb->len >= nlmsg_total_size(0)) {
2516 		int msglen;
2517 
2518 		memset(&extack, 0, sizeof(extack));
2519 		nlh = nlmsg_hdr(skb);
2520 		err = 0;
2521 
2522 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2523 			return 0;
2524 
2525 		/* Only requests are handled by the kernel */
2526 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2527 			goto ack;
2528 
2529 		/* Skip control messages */
2530 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2531 			goto ack;
2532 
2533 		err = cb(skb, nlh, &extack);
2534 		if (err == -EINTR)
2535 			goto skip;
2536 
2537 ack:
2538 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2539 			netlink_ack(skb, nlh, err, &extack);
2540 
2541 skip:
2542 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2543 		if (msglen > skb->len)
2544 			msglen = skb->len;
2545 		skb_pull(skb, msglen);
2546 	}
2547 
2548 	return 0;
2549 }
2550 EXPORT_SYMBOL(netlink_rcv_skb);
2551 
2552 /**
2553  * nlmsg_notify - send a notification netlink message
2554  * @sk: netlink socket to use
2555  * @skb: notification message
2556  * @portid: destination netlink portid for reports or 0
2557  * @group: destination multicast group or 0
2558  * @report: 1 to report back, 0 to disable
2559  * @flags: allocation flags
2560  */
nlmsg_notify(struct sock * sk,struct sk_buff * skb,u32 portid,unsigned int group,int report,gfp_t flags)2561 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2562 		 unsigned int group, int report, gfp_t flags)
2563 {
2564 	int err = 0;
2565 
2566 	if (group) {
2567 		int exclude_portid = 0;
2568 
2569 		if (report) {
2570 			refcount_inc(&skb->users);
2571 			exclude_portid = portid;
2572 		}
2573 
2574 		/* errors reported via destination sk->sk_err, but propagate
2575 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2576 		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2577 		if (err == -ESRCH)
2578 			err = 0;
2579 	}
2580 
2581 	if (report) {
2582 		int err2;
2583 
2584 		err2 = nlmsg_unicast(sk, skb, portid);
2585 		if (!err)
2586 			err = err2;
2587 	}
2588 
2589 	return err;
2590 }
2591 EXPORT_SYMBOL(nlmsg_notify);
2592 
2593 #ifdef CONFIG_PROC_FS
2594 struct nl_seq_iter {
2595 	struct seq_net_private p;
2596 	struct rhashtable_iter hti;
2597 	int link;
2598 };
2599 
netlink_walk_start(struct nl_seq_iter * iter)2600 static void netlink_walk_start(struct nl_seq_iter *iter)
2601 {
2602 	rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2603 	rhashtable_walk_start(&iter->hti);
2604 }
2605 
netlink_walk_stop(struct nl_seq_iter * iter)2606 static void netlink_walk_stop(struct nl_seq_iter *iter)
2607 {
2608 	rhashtable_walk_stop(&iter->hti);
2609 	rhashtable_walk_exit(&iter->hti);
2610 }
2611 
__netlink_seq_next(struct seq_file * seq)2612 static void *__netlink_seq_next(struct seq_file *seq)
2613 {
2614 	struct nl_seq_iter *iter = seq->private;
2615 	struct netlink_sock *nlk;
2616 
2617 	do {
2618 		for (;;) {
2619 			nlk = rhashtable_walk_next(&iter->hti);
2620 
2621 			if (IS_ERR(nlk)) {
2622 				if (PTR_ERR(nlk) == -EAGAIN)
2623 					continue;
2624 
2625 				return nlk;
2626 			}
2627 
2628 			if (nlk)
2629 				break;
2630 
2631 			netlink_walk_stop(iter);
2632 			if (++iter->link >= MAX_LINKS)
2633 				return NULL;
2634 
2635 			netlink_walk_start(iter);
2636 		}
2637 	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2638 
2639 	return nlk;
2640 }
2641 
netlink_seq_start(struct seq_file * seq,loff_t * posp)2642 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2643 	__acquires(RCU)
2644 {
2645 	struct nl_seq_iter *iter = seq->private;
2646 	void *obj = SEQ_START_TOKEN;
2647 	loff_t pos;
2648 
2649 	iter->link = 0;
2650 
2651 	netlink_walk_start(iter);
2652 
2653 	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2654 		obj = __netlink_seq_next(seq);
2655 
2656 	return obj;
2657 }
2658 
netlink_seq_next(struct seq_file * seq,void * v,loff_t * pos)2659 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2660 {
2661 	++*pos;
2662 	return __netlink_seq_next(seq);
2663 }
2664 
netlink_native_seq_stop(struct seq_file * seq,void * v)2665 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2666 {
2667 	struct nl_seq_iter *iter = seq->private;
2668 
2669 	if (iter->link >= MAX_LINKS)
2670 		return;
2671 
2672 	netlink_walk_stop(iter);
2673 }
2674 
2675 
netlink_native_seq_show(struct seq_file * seq,void * v)2676 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2677 {
2678 	if (v == SEQ_START_TOKEN) {
2679 		seq_puts(seq,
2680 			 "sk               Eth Pid        Groups   "
2681 			 "Rmem     Wmem     Dump  Locks    Drops    Inode\n");
2682 	} else {
2683 		struct sock *s = v;
2684 		struct netlink_sock *nlk = nlk_sk(s);
2685 
2686 		seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2687 			   s,
2688 			   s->sk_protocol,
2689 			   nlk->portid,
2690 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2691 			   sk_rmem_alloc_get(s),
2692 			   sk_wmem_alloc_get(s),
2693 			   READ_ONCE(nlk->cb_running),
2694 			   refcount_read(&s->sk_refcnt),
2695 			   atomic_read(&s->sk_drops),
2696 			   sock_i_ino(s)
2697 			);
2698 
2699 	}
2700 	return 0;
2701 }
2702 
2703 #ifdef CONFIG_BPF_SYSCALL
2704 struct bpf_iter__netlink {
2705 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2706 	__bpf_md_ptr(struct netlink_sock *, sk);
2707 };
2708 
DEFINE_BPF_ITER_FUNC(netlink,struct bpf_iter_meta * meta,struct netlink_sock * sk)2709 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2710 
2711 static int netlink_prog_seq_show(struct bpf_prog *prog,
2712 				  struct bpf_iter_meta *meta,
2713 				  void *v)
2714 {
2715 	struct bpf_iter__netlink ctx;
2716 
2717 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2718 	ctx.meta = meta;
2719 	ctx.sk = nlk_sk((struct sock *)v);
2720 	return bpf_iter_run_prog(prog, &ctx);
2721 }
2722 
netlink_seq_show(struct seq_file * seq,void * v)2723 static int netlink_seq_show(struct seq_file *seq, void *v)
2724 {
2725 	struct bpf_iter_meta meta;
2726 	struct bpf_prog *prog;
2727 
2728 	meta.seq = seq;
2729 	prog = bpf_iter_get_info(&meta, false);
2730 	if (!prog)
2731 		return netlink_native_seq_show(seq, v);
2732 
2733 	if (v != SEQ_START_TOKEN)
2734 		return netlink_prog_seq_show(prog, &meta, v);
2735 
2736 	return 0;
2737 }
2738 
netlink_seq_stop(struct seq_file * seq,void * v)2739 static void netlink_seq_stop(struct seq_file *seq, void *v)
2740 {
2741 	struct bpf_iter_meta meta;
2742 	struct bpf_prog *prog;
2743 
2744 	if (!v) {
2745 		meta.seq = seq;
2746 		prog = bpf_iter_get_info(&meta, true);
2747 		if (prog)
2748 			(void)netlink_prog_seq_show(prog, &meta, v);
2749 	}
2750 
2751 	netlink_native_seq_stop(seq, v);
2752 }
2753 #else
netlink_seq_show(struct seq_file * seq,void * v)2754 static int netlink_seq_show(struct seq_file *seq, void *v)
2755 {
2756 	return netlink_native_seq_show(seq, v);
2757 }
2758 
netlink_seq_stop(struct seq_file * seq,void * v)2759 static void netlink_seq_stop(struct seq_file *seq, void *v)
2760 {
2761 	netlink_native_seq_stop(seq, v);
2762 }
2763 #endif
2764 
2765 static const struct seq_operations netlink_seq_ops = {
2766 	.start  = netlink_seq_start,
2767 	.next   = netlink_seq_next,
2768 	.stop   = netlink_seq_stop,
2769 	.show   = netlink_seq_show,
2770 };
2771 #endif
2772 
netlink_register_notifier(struct notifier_block * nb)2773 int netlink_register_notifier(struct notifier_block *nb)
2774 {
2775 	return blocking_notifier_chain_register(&netlink_chain, nb);
2776 }
2777 EXPORT_SYMBOL(netlink_register_notifier);
2778 
netlink_unregister_notifier(struct notifier_block * nb)2779 int netlink_unregister_notifier(struct notifier_block *nb)
2780 {
2781 	return blocking_notifier_chain_unregister(&netlink_chain, nb);
2782 }
2783 EXPORT_SYMBOL(netlink_unregister_notifier);
2784 
2785 static const struct proto_ops netlink_ops = {
2786 	.family =	PF_NETLINK,
2787 	.owner =	THIS_MODULE,
2788 	.release =	netlink_release,
2789 	.bind =		netlink_bind,
2790 	.connect =	netlink_connect,
2791 	.socketpair =	sock_no_socketpair,
2792 	.accept =	sock_no_accept,
2793 	.getname =	netlink_getname,
2794 	.poll =		datagram_poll,
2795 	.ioctl =	netlink_ioctl,
2796 	.listen =	sock_no_listen,
2797 	.shutdown =	sock_no_shutdown,
2798 	.setsockopt =	netlink_setsockopt,
2799 	.getsockopt =	netlink_getsockopt,
2800 	.sendmsg =	netlink_sendmsg,
2801 	.recvmsg =	netlink_recvmsg,
2802 	.mmap =		sock_no_mmap,
2803 };
2804 
2805 static const struct net_proto_family netlink_family_ops = {
2806 	.family = PF_NETLINK,
2807 	.create = netlink_create,
2808 	.owner	= THIS_MODULE,	/* for consistency 8) */
2809 };
2810 
netlink_net_init(struct net * net)2811 static int __net_init netlink_net_init(struct net *net)
2812 {
2813 #ifdef CONFIG_PROC_FS
2814 	if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2815 			sizeof(struct nl_seq_iter)))
2816 		return -ENOMEM;
2817 #endif
2818 	return 0;
2819 }
2820 
netlink_net_exit(struct net * net)2821 static void __net_exit netlink_net_exit(struct net *net)
2822 {
2823 #ifdef CONFIG_PROC_FS
2824 	remove_proc_entry("netlink", net->proc_net);
2825 #endif
2826 }
2827 
netlink_add_usersock_entry(void)2828 static void __init netlink_add_usersock_entry(void)
2829 {
2830 	struct listeners *listeners;
2831 	int groups = 32;
2832 
2833 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2834 	if (!listeners)
2835 		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2836 
2837 	netlink_table_grab();
2838 
2839 	nl_table[NETLINK_USERSOCK].groups = groups;
2840 	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2841 	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2842 	nl_table[NETLINK_USERSOCK].registered = 1;
2843 	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2844 
2845 	netlink_table_ungrab();
2846 }
2847 
2848 static struct pernet_operations __net_initdata netlink_net_ops = {
2849 	.init = netlink_net_init,
2850 	.exit = netlink_net_exit,
2851 };
2852 
netlink_hash(const void * data,u32 len,u32 seed)2853 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2854 {
2855 	const struct netlink_sock *nlk = data;
2856 	struct netlink_compare_arg arg;
2857 
2858 	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2859 	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2860 }
2861 
2862 static const struct rhashtable_params netlink_rhashtable_params = {
2863 	.head_offset = offsetof(struct netlink_sock, node),
2864 	.key_len = netlink_compare_arg_len,
2865 	.obj_hashfn = netlink_hash,
2866 	.obj_cmpfn = netlink_compare,
2867 	.automatic_shrinking = true,
2868 };
2869 
2870 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2871 BTF_ID_LIST(btf_netlink_sock_id)
2872 BTF_ID(struct, netlink_sock)
2873 
2874 static const struct bpf_iter_seq_info netlink_seq_info = {
2875 	.seq_ops		= &netlink_seq_ops,
2876 	.init_seq_private	= bpf_iter_init_seq_net,
2877 	.fini_seq_private	= bpf_iter_fini_seq_net,
2878 	.seq_priv_size		= sizeof(struct nl_seq_iter),
2879 };
2880 
2881 static struct bpf_iter_reg netlink_reg_info = {
2882 	.target			= "netlink",
2883 	.ctx_arg_info_size	= 1,
2884 	.ctx_arg_info		= {
2885 		{ offsetof(struct bpf_iter__netlink, sk),
2886 		  PTR_TO_BTF_ID_OR_NULL },
2887 	},
2888 	.seq_info		= &netlink_seq_info,
2889 };
2890 
bpf_iter_register(void)2891 static int __init bpf_iter_register(void)
2892 {
2893 	netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2894 	return bpf_iter_reg_target(&netlink_reg_info);
2895 }
2896 #endif
2897 
netlink_proto_init(void)2898 static int __init netlink_proto_init(void)
2899 {
2900 	int i;
2901 	int err = proto_register(&netlink_proto, 0);
2902 
2903 	if (err != 0)
2904 		goto out;
2905 
2906 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2907 	err = bpf_iter_register();
2908 	if (err)
2909 		goto out;
2910 #endif
2911 
2912 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2913 
2914 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2915 	if (!nl_table)
2916 		goto panic;
2917 
2918 	for (i = 0; i < MAX_LINKS; i++) {
2919 		if (rhashtable_init(&nl_table[i].hash,
2920 				    &netlink_rhashtable_params) < 0)
2921 			goto panic;
2922 	}
2923 
2924 	netlink_add_usersock_entry();
2925 
2926 	sock_register(&netlink_family_ops);
2927 	register_pernet_subsys(&netlink_net_ops);
2928 	register_pernet_subsys(&netlink_tap_net_ops);
2929 	/* The netlink device handler may be needed early. */
2930 	rtnetlink_init();
2931 out:
2932 	return err;
2933 panic:
2934 	panic("netlink_init: Cannot allocate nl_table\n");
2935 }
2936 
2937 core_initcall(netlink_proto_init);
2938