xref: /linux/net/netlink/af_netlink.c (revision bc9ff192a6c940d9a26e21a0a82f2667067aaf5f)
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 	sk_mem_charge(sk, skb->truesize);
391 }
392 
netlink_sock_destruct(struct sock * sk)393 static void netlink_sock_destruct(struct sock *sk)
394 {
395 	skb_queue_purge(&sk->sk_receive_queue);
396 
397 	if (!sock_flag(sk, SOCK_DEAD)) {
398 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
399 		return;
400 	}
401 
402 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
403 	WARN_ON(refcount_read(&sk->sk_wmem_alloc));
404 	WARN_ON(nlk_sk(sk)->groups);
405 }
406 
407 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
408  * SMP. Look, when several writers sleep and reader wakes them up, all but one
409  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
410  * this, _but_ remember, it adds useless work on UP machines.
411  */
412 
netlink_table_grab(void)413 void netlink_table_grab(void)
414 	__acquires(nl_table_lock)
415 {
416 	might_sleep();
417 
418 	write_lock_irq(&nl_table_lock);
419 
420 	if (atomic_read(&nl_table_users)) {
421 		DECLARE_WAITQUEUE(wait, current);
422 
423 		add_wait_queue_exclusive(&nl_table_wait, &wait);
424 		for (;;) {
425 			set_current_state(TASK_UNINTERRUPTIBLE);
426 			if (atomic_read(&nl_table_users) == 0)
427 				break;
428 			write_unlock_irq(&nl_table_lock);
429 			schedule();
430 			write_lock_irq(&nl_table_lock);
431 		}
432 
433 		__set_current_state(TASK_RUNNING);
434 		remove_wait_queue(&nl_table_wait, &wait);
435 	}
436 }
437 
netlink_table_ungrab(void)438 void netlink_table_ungrab(void)
439 	__releases(nl_table_lock)
440 {
441 	write_unlock_irq(&nl_table_lock);
442 	wake_up(&nl_table_wait);
443 }
444 
445 static inline void
netlink_lock_table(void)446 netlink_lock_table(void)
447 {
448 	unsigned long flags;
449 
450 	/* read_lock() synchronizes us to netlink_table_grab */
451 
452 	read_lock_irqsave(&nl_table_lock, flags);
453 	atomic_inc(&nl_table_users);
454 	read_unlock_irqrestore(&nl_table_lock, flags);
455 }
456 
457 static inline void
netlink_unlock_table(void)458 netlink_unlock_table(void)
459 {
460 	if (atomic_dec_and_test(&nl_table_users))
461 		wake_up(&nl_table_wait);
462 }
463 
464 struct netlink_compare_arg
465 {
466 	possible_net_t pnet;
467 	u32 portid;
468 };
469 
470 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
471 #define netlink_compare_arg_len \
472 	(offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
473 
netlink_compare(struct rhashtable_compare_arg * arg,const void * ptr)474 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
475 				  const void *ptr)
476 {
477 	const struct netlink_compare_arg *x = arg->key;
478 	const struct netlink_sock *nlk = ptr;
479 
480 	return nlk->portid != x->portid ||
481 	       !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
482 }
483 
netlink_compare_arg_init(struct netlink_compare_arg * arg,struct net * net,u32 portid)484 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
485 				     struct net *net, u32 portid)
486 {
487 	memset(arg, 0, sizeof(*arg));
488 	write_pnet(&arg->pnet, net);
489 	arg->portid = portid;
490 }
491 
__netlink_lookup(struct netlink_table * table,u32 portid,struct net * net)492 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
493 				     struct net *net)
494 {
495 	struct netlink_compare_arg arg;
496 
497 	netlink_compare_arg_init(&arg, net, portid);
498 	return rhashtable_lookup_fast(&table->hash, &arg,
499 				      netlink_rhashtable_params);
500 }
501 
__netlink_insert(struct netlink_table * table,struct sock * sk)502 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
503 {
504 	struct netlink_compare_arg arg;
505 
506 	netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
507 	return rhashtable_lookup_insert_key(&table->hash, &arg,
508 					    &nlk_sk(sk)->node,
509 					    netlink_rhashtable_params);
510 }
511 
netlink_lookup(struct net * net,int protocol,u32 portid)512 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
513 {
514 	struct netlink_table *table = &nl_table[protocol];
515 	struct sock *sk;
516 
517 	rcu_read_lock();
518 	sk = __netlink_lookup(table, portid, net);
519 	if (sk)
520 		sock_hold(sk);
521 	rcu_read_unlock();
522 
523 	return sk;
524 }
525 
526 static const struct proto_ops netlink_ops;
527 
528 static void
netlink_update_listeners(struct sock * sk)529 netlink_update_listeners(struct sock *sk)
530 {
531 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
532 	unsigned long mask;
533 	unsigned int i;
534 	struct listeners *listeners;
535 
536 	listeners = nl_deref_protected(tbl->listeners);
537 	if (!listeners)
538 		return;
539 
540 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
541 		mask = 0;
542 		sk_for_each_bound(sk, &tbl->mc_list) {
543 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
544 				mask |= nlk_sk(sk)->groups[i];
545 		}
546 		listeners->masks[i] = mask;
547 	}
548 	/* this function is only called with the netlink table "grabbed", which
549 	 * makes sure updates are visible before bind or setsockopt return. */
550 }
551 
netlink_insert(struct sock * sk,u32 portid)552 static int netlink_insert(struct sock *sk, u32 portid)
553 {
554 	struct netlink_table *table = &nl_table[sk->sk_protocol];
555 	int err;
556 
557 	lock_sock(sk);
558 
559 	err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
560 	if (nlk_sk(sk)->bound)
561 		goto err;
562 
563 	/* portid can be read locklessly from netlink_getname(). */
564 	WRITE_ONCE(nlk_sk(sk)->portid, portid);
565 
566 	sock_hold(sk);
567 
568 	err = __netlink_insert(table, sk);
569 	if (err) {
570 		/* In case the hashtable backend returns with -EBUSY
571 		 * from here, it must not escape to the caller.
572 		 */
573 		if (unlikely(err == -EBUSY))
574 			err = -EOVERFLOW;
575 		if (err == -EEXIST)
576 			err = -EADDRINUSE;
577 		sock_put(sk);
578 		goto err;
579 	}
580 
581 	/* We need to ensure that the socket is hashed and visible. */
582 	smp_wmb();
583 	/* Paired with lockless reads from netlink_bind(),
584 	 * netlink_connect() and netlink_sendmsg().
585 	 */
586 	WRITE_ONCE(nlk_sk(sk)->bound, portid);
587 
588 err:
589 	release_sock(sk);
590 	return err;
591 }
592 
netlink_remove(struct sock * sk)593 static void netlink_remove(struct sock *sk)
594 {
595 	struct netlink_table *table;
596 
597 	table = &nl_table[sk->sk_protocol];
598 	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
599 				    netlink_rhashtable_params)) {
600 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
601 		__sock_put(sk);
602 	}
603 
604 	netlink_table_grab();
605 	if (nlk_sk(sk)->subscriptions) {
606 		__sk_del_bind_node(sk);
607 		netlink_update_listeners(sk);
608 	}
609 	if (sk->sk_protocol == NETLINK_GENERIC)
610 		atomic_inc(&genl_sk_destructing_cnt);
611 	netlink_table_ungrab();
612 }
613 
614 static struct proto netlink_proto = {
615 	.name	  = "NETLINK",
616 	.owner	  = THIS_MODULE,
617 	.obj_size = sizeof(struct netlink_sock),
618 };
619 
__netlink_create(struct net * net,struct socket * sock,int protocol,int kern)620 static int __netlink_create(struct net *net, struct socket *sock,
621 			    int protocol, int kern)
622 {
623 	struct sock *sk;
624 	struct netlink_sock *nlk;
625 
626 	sock->ops = &netlink_ops;
627 
628 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
629 	if (!sk)
630 		return -ENOMEM;
631 
632 	sock_init_data(sock, sk);
633 
634 	nlk = nlk_sk(sk);
635 	mutex_init(&nlk->nl_cb_mutex);
636 	lockdep_set_class_and_name(&nlk->nl_cb_mutex,
637 					   nlk_cb_mutex_keys + protocol,
638 					   nlk_cb_mutex_key_strings[protocol]);
639 	init_waitqueue_head(&nlk->wait);
640 
641 	sk->sk_destruct = netlink_sock_destruct;
642 	sk->sk_protocol = protocol;
643 	return 0;
644 }
645 
netlink_create(struct net * net,struct socket * sock,int protocol,int kern)646 static int netlink_create(struct net *net, struct socket *sock, int protocol,
647 			  int kern)
648 {
649 	struct module *module = NULL;
650 	struct netlink_sock *nlk;
651 	int (*bind)(struct net *net, int group);
652 	void (*unbind)(struct net *net, int group);
653 	void (*release)(struct sock *sock, unsigned long *groups);
654 	int err = 0;
655 
656 	sock->state = SS_UNCONNECTED;
657 
658 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
659 		return -ESOCKTNOSUPPORT;
660 
661 	if (protocol < 0 || protocol >= MAX_LINKS)
662 		return -EPROTONOSUPPORT;
663 	protocol = array_index_nospec(protocol, MAX_LINKS);
664 
665 	netlink_lock_table();
666 #ifdef CONFIG_MODULES
667 	if (!nl_table[protocol].registered) {
668 		netlink_unlock_table();
669 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
670 		netlink_lock_table();
671 	}
672 #endif
673 	if (nl_table[protocol].registered &&
674 	    try_module_get(nl_table[protocol].module))
675 		module = nl_table[protocol].module;
676 	else
677 		err = -EPROTONOSUPPORT;
678 	bind = nl_table[protocol].bind;
679 	unbind = nl_table[protocol].unbind;
680 	release = nl_table[protocol].release;
681 	netlink_unlock_table();
682 
683 	if (err < 0)
684 		goto out;
685 
686 	err = __netlink_create(net, sock, protocol, kern);
687 	if (err < 0)
688 		goto out_module;
689 
690 	sock_prot_inuse_add(net, &netlink_proto, 1);
691 
692 	nlk = nlk_sk(sock->sk);
693 	nlk->module = module;
694 	nlk->netlink_bind = bind;
695 	nlk->netlink_unbind = unbind;
696 	nlk->netlink_release = release;
697 out:
698 	return err;
699 
700 out_module:
701 	module_put(module);
702 	goto out;
703 }
704 
deferred_put_nlk_sk(struct rcu_head * head)705 static void deferred_put_nlk_sk(struct rcu_head *head)
706 {
707 	struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
708 	struct sock *sk = &nlk->sk;
709 
710 	kfree(nlk->groups);
711 	nlk->groups = NULL;
712 
713 	if (!refcount_dec_and_test(&sk->sk_refcnt))
714 		return;
715 
716 	sk_free(sk);
717 }
718 
netlink_release(struct socket * sock)719 static int netlink_release(struct socket *sock)
720 {
721 	struct sock *sk = sock->sk;
722 	struct netlink_sock *nlk;
723 
724 	if (!sk)
725 		return 0;
726 
727 	netlink_remove(sk);
728 	sock_orphan(sk);
729 	nlk = nlk_sk(sk);
730 
731 	/*
732 	 * OK. Socket is unlinked, any packets that arrive now
733 	 * will be purged.
734 	 */
735 	if (nlk->netlink_release)
736 		nlk->netlink_release(sk, nlk->groups);
737 
738 	/* must not acquire netlink_table_lock in any way again before unbind
739 	 * and notifying genetlink is done as otherwise it might deadlock
740 	 */
741 	if (nlk->netlink_unbind) {
742 		int i;
743 
744 		for (i = 0; i < nlk->ngroups; i++)
745 			if (test_bit(i, nlk->groups))
746 				nlk->netlink_unbind(sock_net(sk), i + 1);
747 	}
748 	if (sk->sk_protocol == NETLINK_GENERIC &&
749 	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
750 		wake_up(&genl_sk_destructing_waitq);
751 
752 	sock->sk = NULL;
753 	wake_up_interruptible_all(&nlk->wait);
754 
755 	skb_queue_purge(&sk->sk_write_queue);
756 
757 	if (nlk->portid && nlk->bound) {
758 		struct netlink_notify n = {
759 						.net = sock_net(sk),
760 						.protocol = sk->sk_protocol,
761 						.portid = nlk->portid,
762 					  };
763 		blocking_notifier_call_chain(&netlink_chain,
764 				NETLINK_URELEASE, &n);
765 	}
766 
767 	/* Terminate any outstanding dump */
768 	if (nlk->cb_running) {
769 		if (nlk->cb.done)
770 			nlk->cb.done(&nlk->cb);
771 		module_put(nlk->cb.module);
772 		kfree_skb(nlk->cb.skb);
773 		WRITE_ONCE(nlk->cb_running, false);
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 	DECLARE_WAITQUEUE(wait, current);
1215 	struct netlink_sock *nlk;
1216 	unsigned int rmem;
1217 
1218 	nlk = nlk_sk(sk);
1219 	rmem = atomic_add_return(skb->truesize, &sk->sk_rmem_alloc);
1220 
1221 	if ((rmem == skb->truesize || rmem < READ_ONCE(sk->sk_rcvbuf)) &&
1222 	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1223 		netlink_skb_set_owner_r(skb, sk);
1224 		return 0;
1225 	}
1226 
1227 	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1228 
1229 	if (!*timeo) {
1230 		if (!ssk || netlink_is_kernel(ssk))
1231 			netlink_overrun(sk);
1232 		sock_put(sk);
1233 		kfree_skb(skb);
1234 		return -EAGAIN;
1235 	}
1236 
1237 	__set_current_state(TASK_INTERRUPTIBLE);
1238 	add_wait_queue(&nlk->wait, &wait);
1239 	rmem = atomic_read(&sk->sk_rmem_alloc);
1240 
1241 	if (((rmem && rmem + skb->truesize > READ_ONCE(sk->sk_rcvbuf)) ||
1242 	     test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1243 	    !sock_flag(sk, SOCK_DEAD))
1244 		*timeo = schedule_timeout(*timeo);
1245 
1246 	__set_current_state(TASK_RUNNING);
1247 	remove_wait_queue(&nlk->wait, &wait);
1248 	sock_put(sk);
1249 
1250 	if (signal_pending(current)) {
1251 		kfree_skb(skb);
1252 		return sock_intr_errno(*timeo);
1253 	}
1254 
1255 	return 1;
1256 }
1257 
__netlink_sendskb(struct sock * sk,struct sk_buff * skb)1258 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1259 {
1260 	int len = skb->len;
1261 
1262 	netlink_deliver_tap(sock_net(sk), skb);
1263 
1264 	skb_queue_tail(&sk->sk_receive_queue, skb);
1265 	sk->sk_data_ready(sk);
1266 	return len;
1267 }
1268 
netlink_sendskb(struct sock * sk,struct sk_buff * skb)1269 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1270 {
1271 	int len = __netlink_sendskb(sk, skb);
1272 
1273 	sock_put(sk);
1274 	return len;
1275 }
1276 
netlink_detachskb(struct sock * sk,struct sk_buff * skb)1277 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1278 {
1279 	kfree_skb(skb);
1280 	sock_put(sk);
1281 }
1282 
netlink_trim(struct sk_buff * skb,gfp_t allocation)1283 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1284 {
1285 	int delta;
1286 
1287 	skb_assert_len(skb);
1288 	WARN_ON(skb->sk != NULL);
1289 	delta = skb->end - skb->tail;
1290 	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1291 		return skb;
1292 
1293 	if (skb_shared(skb)) {
1294 		struct sk_buff *nskb = skb_clone(skb, allocation);
1295 		if (!nskb)
1296 			return skb;
1297 		consume_skb(skb);
1298 		skb = nskb;
1299 	}
1300 
1301 	pskb_expand_head(skb, 0, -delta,
1302 			 (allocation & ~__GFP_DIRECT_RECLAIM) |
1303 			 __GFP_NOWARN | __GFP_NORETRY);
1304 	return skb;
1305 }
1306 
netlink_unicast_kernel(struct sock * sk,struct sk_buff * skb,struct sock * ssk)1307 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1308 				  struct sock *ssk)
1309 {
1310 	int ret;
1311 	struct netlink_sock *nlk = nlk_sk(sk);
1312 
1313 	ret = -ECONNREFUSED;
1314 	if (nlk->netlink_rcv != NULL) {
1315 		ret = skb->len;
1316 		atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1317 		netlink_skb_set_owner_r(skb, sk);
1318 		NETLINK_CB(skb).sk = ssk;
1319 		netlink_deliver_tap_kernel(sk, ssk, skb);
1320 		nlk->netlink_rcv(skb);
1321 		consume_skb(skb);
1322 	} else {
1323 		kfree_skb(skb);
1324 	}
1325 	sock_put(sk);
1326 	return ret;
1327 }
1328 
netlink_unicast(struct sock * ssk,struct sk_buff * skb,u32 portid,int nonblock)1329 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1330 		    u32 portid, int nonblock)
1331 {
1332 	struct sock *sk;
1333 	int err;
1334 	long timeo;
1335 
1336 	skb = netlink_trim(skb, gfp_any());
1337 
1338 	timeo = sock_sndtimeo(ssk, nonblock);
1339 retry:
1340 	sk = netlink_getsockbyportid(ssk, portid);
1341 	if (IS_ERR(sk)) {
1342 		kfree_skb(skb);
1343 		return PTR_ERR(sk);
1344 	}
1345 	if (netlink_is_kernel(sk))
1346 		return netlink_unicast_kernel(sk, skb, ssk);
1347 
1348 	if (sk_filter(sk, skb)) {
1349 		err = skb->len;
1350 		kfree_skb(skb);
1351 		sock_put(sk);
1352 		return err;
1353 	}
1354 
1355 	err = netlink_attachskb(sk, skb, &timeo, ssk);
1356 	if (err == 1)
1357 		goto retry;
1358 	if (err)
1359 		return err;
1360 
1361 	return netlink_sendskb(sk, skb);
1362 }
1363 EXPORT_SYMBOL(netlink_unicast);
1364 
netlink_has_listeners(struct sock * sk,unsigned int group)1365 int netlink_has_listeners(struct sock *sk, unsigned int group)
1366 {
1367 	int res = 0;
1368 	struct listeners *listeners;
1369 
1370 	BUG_ON(!netlink_is_kernel(sk));
1371 
1372 	rcu_read_lock();
1373 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1374 
1375 	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1376 		res = test_bit(group - 1, listeners->masks);
1377 
1378 	rcu_read_unlock();
1379 
1380 	return res;
1381 }
1382 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1383 
netlink_strict_get_check(struct sk_buff * skb)1384 bool netlink_strict_get_check(struct sk_buff *skb)
1385 {
1386 	return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1387 }
1388 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1389 
netlink_broadcast_deliver(struct sock * sk,struct sk_buff * skb)1390 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1391 {
1392 	struct netlink_sock *nlk = nlk_sk(sk);
1393 	unsigned int rmem, rcvbuf;
1394 
1395 	rmem = atomic_add_return(skb->truesize, &sk->sk_rmem_alloc);
1396 	rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1397 
1398 	if ((rmem != skb->truesize || rmem <= rcvbuf) &&
1399 	    !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1400 		netlink_skb_set_owner_r(skb, sk);
1401 		__netlink_sendskb(sk, skb);
1402 		return rmem > (rcvbuf >> 1);
1403 	}
1404 
1405 	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1406 	return -1;
1407 }
1408 
1409 struct netlink_broadcast_data {
1410 	struct sock *exclude_sk;
1411 	struct net *net;
1412 	u32 portid;
1413 	u32 group;
1414 	int failure;
1415 	int delivery_failure;
1416 	int congested;
1417 	int delivered;
1418 	gfp_t allocation;
1419 	struct sk_buff *skb, *skb2;
1420 	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1421 	void *tx_data;
1422 };
1423 
do_one_broadcast(struct sock * sk,struct netlink_broadcast_data * p)1424 static void do_one_broadcast(struct sock *sk,
1425 				    struct netlink_broadcast_data *p)
1426 {
1427 	struct netlink_sock *nlk = nlk_sk(sk);
1428 	int val;
1429 
1430 	if (p->exclude_sk == sk)
1431 		return;
1432 
1433 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1434 	    !test_bit(p->group - 1, nlk->groups))
1435 		return;
1436 
1437 	if (!net_eq(sock_net(sk), p->net)) {
1438 		if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1439 			return;
1440 
1441 		if (!peernet_has_id(sock_net(sk), p->net))
1442 			return;
1443 
1444 		if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1445 				     CAP_NET_BROADCAST))
1446 			return;
1447 	}
1448 
1449 	if (p->failure) {
1450 		netlink_overrun(sk);
1451 		return;
1452 	}
1453 
1454 	sock_hold(sk);
1455 	if (p->skb2 == NULL) {
1456 		if (skb_shared(p->skb)) {
1457 			p->skb2 = skb_clone(p->skb, p->allocation);
1458 		} else {
1459 			p->skb2 = skb_get(p->skb);
1460 			/*
1461 			 * skb ownership may have been set when
1462 			 * delivered to a previous socket.
1463 			 */
1464 			skb_orphan(p->skb2);
1465 		}
1466 	}
1467 	if (p->skb2 == NULL) {
1468 		netlink_overrun(sk);
1469 		/* Clone failed. Notify ALL listeners. */
1470 		p->failure = 1;
1471 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1472 			p->delivery_failure = 1;
1473 		goto out;
1474 	}
1475 
1476 	if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1477 		kfree_skb(p->skb2);
1478 		p->skb2 = NULL;
1479 		goto out;
1480 	}
1481 
1482 	if (sk_filter(sk, p->skb2)) {
1483 		kfree_skb(p->skb2);
1484 		p->skb2 = NULL;
1485 		goto out;
1486 	}
1487 	NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1488 	if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1489 		NETLINK_CB(p->skb2).nsid_is_set = true;
1490 	val = netlink_broadcast_deliver(sk, p->skb2);
1491 	if (val < 0) {
1492 		netlink_overrun(sk);
1493 		if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1494 			p->delivery_failure = 1;
1495 	} else {
1496 		p->congested |= val;
1497 		p->delivered = 1;
1498 		p->skb2 = NULL;
1499 	}
1500 out:
1501 	sock_put(sk);
1502 }
1503 
netlink_broadcast_filtered(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation,netlink_filter_fn filter,void * filter_data)1504 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1505 			       u32 portid,
1506 			       u32 group, gfp_t allocation,
1507 			       netlink_filter_fn filter,
1508 			       void *filter_data)
1509 {
1510 	struct net *net = sock_net(ssk);
1511 	struct netlink_broadcast_data info;
1512 	struct sock *sk;
1513 
1514 	skb = netlink_trim(skb, allocation);
1515 
1516 	info.exclude_sk = ssk;
1517 	info.net = net;
1518 	info.portid = portid;
1519 	info.group = group;
1520 	info.failure = 0;
1521 	info.delivery_failure = 0;
1522 	info.congested = 0;
1523 	info.delivered = 0;
1524 	info.allocation = allocation;
1525 	info.skb = skb;
1526 	info.skb2 = NULL;
1527 	info.tx_filter = filter;
1528 	info.tx_data = filter_data;
1529 
1530 	/* While we sleep in clone, do not allow to change socket list */
1531 
1532 	netlink_lock_table();
1533 
1534 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1535 		do_one_broadcast(sk, &info);
1536 
1537 	consume_skb(skb);
1538 
1539 	netlink_unlock_table();
1540 
1541 	if (info.delivery_failure) {
1542 		kfree_skb(info.skb2);
1543 		return -ENOBUFS;
1544 	}
1545 	consume_skb(info.skb2);
1546 
1547 	if (info.delivered) {
1548 		if (info.congested && gfpflags_allow_blocking(allocation))
1549 			yield();
1550 		return 0;
1551 	}
1552 	return -ESRCH;
1553 }
1554 EXPORT_SYMBOL(netlink_broadcast_filtered);
1555 
netlink_broadcast(struct sock * ssk,struct sk_buff * skb,u32 portid,u32 group,gfp_t allocation)1556 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1557 		      u32 group, gfp_t allocation)
1558 {
1559 	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1560 					  NULL, NULL);
1561 }
1562 EXPORT_SYMBOL(netlink_broadcast);
1563 
1564 struct netlink_set_err_data {
1565 	struct sock *exclude_sk;
1566 	u32 portid;
1567 	u32 group;
1568 	int code;
1569 };
1570 
do_one_set_err(struct sock * sk,struct netlink_set_err_data * p)1571 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1572 {
1573 	struct netlink_sock *nlk = nlk_sk(sk);
1574 	int ret = 0;
1575 
1576 	if (sk == p->exclude_sk)
1577 		goto out;
1578 
1579 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1580 		goto out;
1581 
1582 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1583 	    !test_bit(p->group - 1, nlk->groups))
1584 		goto out;
1585 
1586 	if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1587 		ret = 1;
1588 		goto out;
1589 	}
1590 
1591 	WRITE_ONCE(sk->sk_err, p->code);
1592 	sk_error_report(sk);
1593 out:
1594 	return ret;
1595 }
1596 
1597 /**
1598  * netlink_set_err - report error to broadcast listeners
1599  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1600  * @portid: the PORTID of a process that we want to skip (if any)
1601  * @group: the broadcast group that will notice the error
1602  * @code: error code, must be negative (as usual in kernelspace)
1603  *
1604  * This function returns the number of broadcast listeners that have set the
1605  * NETLINK_NO_ENOBUFS socket option.
1606  */
netlink_set_err(struct sock * ssk,u32 portid,u32 group,int code)1607 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1608 {
1609 	struct netlink_set_err_data info;
1610 	unsigned long flags;
1611 	struct sock *sk;
1612 	int ret = 0;
1613 
1614 	info.exclude_sk = ssk;
1615 	info.portid = portid;
1616 	info.group = group;
1617 	/* sk->sk_err wants a positive error value */
1618 	info.code = -code;
1619 
1620 	read_lock_irqsave(&nl_table_lock, flags);
1621 
1622 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1623 		ret += do_one_set_err(sk, &info);
1624 
1625 	read_unlock_irqrestore(&nl_table_lock, flags);
1626 	return ret;
1627 }
1628 EXPORT_SYMBOL(netlink_set_err);
1629 
1630 /* must be called with netlink table grabbed */
netlink_update_socket_mc(struct netlink_sock * nlk,unsigned int group,int is_new)1631 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1632 				     unsigned int group,
1633 				     int is_new)
1634 {
1635 	int old, new = !!is_new, subscriptions;
1636 
1637 	old = test_bit(group - 1, nlk->groups);
1638 	subscriptions = nlk->subscriptions - old + new;
1639 	__assign_bit(group - 1, nlk->groups, new);
1640 	netlink_update_subscriptions(&nlk->sk, subscriptions);
1641 	netlink_update_listeners(&nlk->sk);
1642 }
1643 
netlink_setsockopt(struct socket * sock,int level,int optname,sockptr_t optval,unsigned int optlen)1644 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1645 			      sockptr_t optval, unsigned int optlen)
1646 {
1647 	struct sock *sk = sock->sk;
1648 	struct netlink_sock *nlk = nlk_sk(sk);
1649 	unsigned int val = 0;
1650 	int nr = -1;
1651 
1652 	if (level != SOL_NETLINK)
1653 		return -ENOPROTOOPT;
1654 
1655 	if (optlen >= sizeof(int) &&
1656 	    copy_from_sockptr(&val, optval, sizeof(val)))
1657 		return -EFAULT;
1658 
1659 	switch (optname) {
1660 	case NETLINK_PKTINFO:
1661 		nr = NETLINK_F_RECV_PKTINFO;
1662 		break;
1663 	case NETLINK_ADD_MEMBERSHIP:
1664 	case NETLINK_DROP_MEMBERSHIP: {
1665 		int err;
1666 
1667 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1668 			return -EPERM;
1669 		err = netlink_realloc_groups(sk);
1670 		if (err)
1671 			return err;
1672 		if (!val || val - 1 >= nlk->ngroups)
1673 			return -EINVAL;
1674 		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1675 			err = nlk->netlink_bind(sock_net(sk), val);
1676 			if (err)
1677 				return err;
1678 		}
1679 		netlink_table_grab();
1680 		netlink_update_socket_mc(nlk, val,
1681 					 optname == NETLINK_ADD_MEMBERSHIP);
1682 		netlink_table_ungrab();
1683 		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1684 			nlk->netlink_unbind(sock_net(sk), val);
1685 
1686 		break;
1687 	}
1688 	case NETLINK_BROADCAST_ERROR:
1689 		nr = NETLINK_F_BROADCAST_SEND_ERROR;
1690 		break;
1691 	case NETLINK_NO_ENOBUFS:
1692 		assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1693 		if (val) {
1694 			clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1695 			wake_up_interruptible(&nlk->wait);
1696 		}
1697 		break;
1698 	case NETLINK_LISTEN_ALL_NSID:
1699 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1700 			return -EPERM;
1701 		nr = NETLINK_F_LISTEN_ALL_NSID;
1702 		break;
1703 	case NETLINK_CAP_ACK:
1704 		nr = NETLINK_F_CAP_ACK;
1705 		break;
1706 	case NETLINK_EXT_ACK:
1707 		nr = NETLINK_F_EXT_ACK;
1708 		break;
1709 	case NETLINK_GET_STRICT_CHK:
1710 		nr = NETLINK_F_STRICT_CHK;
1711 		break;
1712 	default:
1713 		return -ENOPROTOOPT;
1714 	}
1715 	if (nr >= 0)
1716 		assign_bit(nr, &nlk->flags, val);
1717 	return 0;
1718 }
1719 
netlink_getsockopt(struct socket * sock,int level,int optname,char __user * optval,int __user * optlen)1720 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1721 			      char __user *optval, int __user *optlen)
1722 {
1723 	struct sock *sk = sock->sk;
1724 	struct netlink_sock *nlk = nlk_sk(sk);
1725 	unsigned int flag;
1726 	int len, val;
1727 
1728 	if (level != SOL_NETLINK)
1729 		return -ENOPROTOOPT;
1730 
1731 	if (get_user(len, optlen))
1732 		return -EFAULT;
1733 	if (len < 0)
1734 		return -EINVAL;
1735 
1736 	switch (optname) {
1737 	case NETLINK_PKTINFO:
1738 		flag = NETLINK_F_RECV_PKTINFO;
1739 		break;
1740 	case NETLINK_BROADCAST_ERROR:
1741 		flag = NETLINK_F_BROADCAST_SEND_ERROR;
1742 		break;
1743 	case NETLINK_NO_ENOBUFS:
1744 		flag = NETLINK_F_RECV_NO_ENOBUFS;
1745 		break;
1746 	case NETLINK_LIST_MEMBERSHIPS: {
1747 		int pos, idx, shift, err = 0;
1748 
1749 		netlink_lock_table();
1750 		for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1751 			if (len - pos < sizeof(u32))
1752 				break;
1753 
1754 			idx = pos / sizeof(unsigned long);
1755 			shift = (pos % sizeof(unsigned long)) * 8;
1756 			if (put_user((u32)(nlk->groups[idx] >> shift),
1757 				     (u32 __user *)(optval + pos))) {
1758 				err = -EFAULT;
1759 				break;
1760 			}
1761 		}
1762 		if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1763 			err = -EFAULT;
1764 		netlink_unlock_table();
1765 		return err;
1766 	}
1767 	case NETLINK_LISTEN_ALL_NSID:
1768 		flag = NETLINK_F_LISTEN_ALL_NSID;
1769 		break;
1770 	case NETLINK_CAP_ACK:
1771 		flag = NETLINK_F_CAP_ACK;
1772 		break;
1773 	case NETLINK_EXT_ACK:
1774 		flag = NETLINK_F_EXT_ACK;
1775 		break;
1776 	case NETLINK_GET_STRICT_CHK:
1777 		flag = NETLINK_F_STRICT_CHK;
1778 		break;
1779 	default:
1780 		return -ENOPROTOOPT;
1781 	}
1782 
1783 	if (len < sizeof(int))
1784 		return -EINVAL;
1785 
1786 	len = sizeof(int);
1787 	val = test_bit(flag, &nlk->flags);
1788 
1789 	if (put_user(len, optlen) ||
1790 	    copy_to_user(optval, &val, len))
1791 		return -EFAULT;
1792 
1793 	return 0;
1794 }
1795 
netlink_cmsg_recv_pktinfo(struct msghdr * msg,struct sk_buff * skb)1796 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1797 {
1798 	struct nl_pktinfo info;
1799 
1800 	info.group = NETLINK_CB(skb).dst_group;
1801 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1802 }
1803 
netlink_cmsg_listen_all_nsid(struct sock * sk,struct msghdr * msg,struct sk_buff * skb)1804 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1805 					 struct sk_buff *skb)
1806 {
1807 	if (!NETLINK_CB(skb).nsid_is_set)
1808 		return;
1809 
1810 	put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1811 		 &NETLINK_CB(skb).nsid);
1812 }
1813 
netlink_sendmsg(struct socket * sock,struct msghdr * msg,size_t len)1814 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1815 {
1816 	struct sock *sk = sock->sk;
1817 	struct netlink_sock *nlk = nlk_sk(sk);
1818 	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1819 	u32 dst_portid;
1820 	u32 dst_group;
1821 	struct sk_buff *skb;
1822 	int err;
1823 	struct scm_cookie scm;
1824 	u32 netlink_skb_flags = 0;
1825 
1826 	if (msg->msg_flags & MSG_OOB)
1827 		return -EOPNOTSUPP;
1828 
1829 	if (len == 0) {
1830 		pr_warn_once("Zero length message leads to an empty skb\n");
1831 		return -ENODATA;
1832 	}
1833 
1834 	err = scm_send(sock, msg, &scm, true);
1835 	if (err < 0)
1836 		return err;
1837 
1838 	if (msg->msg_namelen) {
1839 		err = -EINVAL;
1840 		if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1841 			goto out;
1842 		if (addr->nl_family != AF_NETLINK)
1843 			goto out;
1844 		dst_portid = addr->nl_pid;
1845 		dst_group = ffs(addr->nl_groups);
1846 		err =  -EPERM;
1847 		if ((dst_group || dst_portid) &&
1848 		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1849 			goto out;
1850 		netlink_skb_flags |= NETLINK_SKB_DST;
1851 	} else {
1852 		/* Paired with WRITE_ONCE() in netlink_connect() */
1853 		dst_portid = READ_ONCE(nlk->dst_portid);
1854 		dst_group = READ_ONCE(nlk->dst_group);
1855 	}
1856 
1857 	/* Paired with WRITE_ONCE() in netlink_insert() */
1858 	if (!READ_ONCE(nlk->bound)) {
1859 		err = netlink_autobind(sock);
1860 		if (err)
1861 			goto out;
1862 	} else {
1863 		/* Ensure nlk is hashed and visible. */
1864 		smp_rmb();
1865 	}
1866 
1867 	err = -EMSGSIZE;
1868 	if (len > sk->sk_sndbuf - 32)
1869 		goto out;
1870 	err = -ENOBUFS;
1871 	skb = netlink_alloc_large_skb(len, dst_group);
1872 	if (skb == NULL)
1873 		goto out;
1874 
1875 	NETLINK_CB(skb).portid	= nlk->portid;
1876 	NETLINK_CB(skb).dst_group = dst_group;
1877 	NETLINK_CB(skb).creds	= scm.creds;
1878 	NETLINK_CB(skb).flags	= netlink_skb_flags;
1879 
1880 	err = -EFAULT;
1881 	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1882 		kfree_skb(skb);
1883 		goto out;
1884 	}
1885 
1886 	err = security_netlink_send(sk, skb);
1887 	if (err) {
1888 		kfree_skb(skb);
1889 		goto out;
1890 	}
1891 
1892 	if (dst_group) {
1893 		refcount_inc(&skb->users);
1894 		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1895 	}
1896 	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1897 
1898 out:
1899 	scm_destroy(&scm);
1900 	return err;
1901 }
1902 
netlink_recvmsg(struct socket * sock,struct msghdr * msg,size_t len,int flags)1903 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1904 			   int flags)
1905 {
1906 	struct scm_cookie scm;
1907 	struct sock *sk = sock->sk;
1908 	struct netlink_sock *nlk = nlk_sk(sk);
1909 	size_t copied, max_recvmsg_len;
1910 	struct sk_buff *skb, *data_skb;
1911 	int err, ret;
1912 
1913 	if (flags & MSG_OOB)
1914 		return -EOPNOTSUPP;
1915 
1916 	copied = 0;
1917 
1918 	skb = skb_recv_datagram(sk, flags, &err);
1919 	if (skb == NULL)
1920 		goto out;
1921 
1922 	data_skb = skb;
1923 
1924 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1925 	if (unlikely(skb_shinfo(skb)->frag_list)) {
1926 		/*
1927 		 * If this skb has a frag_list, then here that means that we
1928 		 * will have to use the frag_list skb's data for compat tasks
1929 		 * and the regular skb's data for normal (non-compat) tasks.
1930 		 *
1931 		 * If we need to send the compat skb, assign it to the
1932 		 * 'data_skb' variable so that it will be used below for data
1933 		 * copying. We keep 'skb' for everything else, including
1934 		 * freeing both later.
1935 		 */
1936 		if (flags & MSG_CMSG_COMPAT)
1937 			data_skb = skb_shinfo(skb)->frag_list;
1938 	}
1939 #endif
1940 
1941 	/* Record the max length of recvmsg() calls for future allocations */
1942 	max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1943 	max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1944 				SKB_WITH_OVERHEAD(32768));
1945 	WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1946 
1947 	copied = data_skb->len;
1948 	if (len < copied) {
1949 		msg->msg_flags |= MSG_TRUNC;
1950 		copied = len;
1951 	}
1952 
1953 	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1954 
1955 	if (msg->msg_name) {
1956 		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1957 		addr->nl_family = AF_NETLINK;
1958 		addr->nl_pad    = 0;
1959 		addr->nl_pid	= NETLINK_CB(skb).portid;
1960 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
1961 		msg->msg_namelen = sizeof(*addr);
1962 	}
1963 
1964 	if (nlk_test_bit(RECV_PKTINFO, sk))
1965 		netlink_cmsg_recv_pktinfo(msg, skb);
1966 	if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1967 		netlink_cmsg_listen_all_nsid(sk, msg, skb);
1968 
1969 	memset(&scm, 0, sizeof(scm));
1970 	scm.creds = *NETLINK_CREDS(skb);
1971 	if (flags & MSG_TRUNC)
1972 		copied = data_skb->len;
1973 
1974 	skb_free_datagram(sk, skb);
1975 
1976 	if (READ_ONCE(nlk->cb_running) &&
1977 	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1978 		ret = netlink_dump(sk, false);
1979 		if (ret) {
1980 			WRITE_ONCE(sk->sk_err, -ret);
1981 			sk_error_report(sk);
1982 		}
1983 	}
1984 
1985 	scm_recv(sock, msg, &scm, flags);
1986 out:
1987 	netlink_rcv_wake(sk);
1988 	return err ? : copied;
1989 }
1990 
netlink_data_ready(struct sock * sk)1991 static void netlink_data_ready(struct sock *sk)
1992 {
1993 	BUG();
1994 }
1995 
1996 /*
1997  *	We export these functions to other modules. They provide a
1998  *	complete set of kernel non-blocking support for message
1999  *	queueing.
2000  */
2001 
2002 struct sock *
__netlink_kernel_create(struct net * net,int unit,struct module * module,struct netlink_kernel_cfg * cfg)2003 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2004 			struct netlink_kernel_cfg *cfg)
2005 {
2006 	struct socket *sock;
2007 	struct sock *sk;
2008 	struct netlink_sock *nlk;
2009 	struct listeners *listeners = NULL;
2010 	unsigned int groups;
2011 
2012 	BUG_ON(!nl_table);
2013 
2014 	if (unit < 0 || unit >= MAX_LINKS)
2015 		return NULL;
2016 
2017 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2018 		return NULL;
2019 
2020 	if (__netlink_create(net, sock, unit, 1) < 0)
2021 		goto out_sock_release_nosk;
2022 
2023 	sk = sock->sk;
2024 
2025 	if (!cfg || cfg->groups < 32)
2026 		groups = 32;
2027 	else
2028 		groups = cfg->groups;
2029 
2030 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2031 	if (!listeners)
2032 		goto out_sock_release;
2033 
2034 	sk->sk_data_ready = netlink_data_ready;
2035 	if (cfg && cfg->input)
2036 		nlk_sk(sk)->netlink_rcv = cfg->input;
2037 
2038 	if (netlink_insert(sk, 0))
2039 		goto out_sock_release;
2040 
2041 	nlk = nlk_sk(sk);
2042 	set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2043 
2044 	netlink_table_grab();
2045 	if (!nl_table[unit].registered) {
2046 		nl_table[unit].groups = groups;
2047 		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2048 		nl_table[unit].module = module;
2049 		if (cfg) {
2050 			nl_table[unit].bind = cfg->bind;
2051 			nl_table[unit].unbind = cfg->unbind;
2052 			nl_table[unit].release = cfg->release;
2053 			nl_table[unit].flags = cfg->flags;
2054 		}
2055 		nl_table[unit].registered = 1;
2056 	} else {
2057 		kfree(listeners);
2058 		nl_table[unit].registered++;
2059 	}
2060 	netlink_table_ungrab();
2061 	return sk;
2062 
2063 out_sock_release:
2064 	kfree(listeners);
2065 	netlink_kernel_release(sk);
2066 	return NULL;
2067 
2068 out_sock_release_nosk:
2069 	sock_release(sock);
2070 	return NULL;
2071 }
2072 EXPORT_SYMBOL(__netlink_kernel_create);
2073 
2074 void
netlink_kernel_release(struct sock * sk)2075 netlink_kernel_release(struct sock *sk)
2076 {
2077 	if (sk == NULL || sk->sk_socket == NULL)
2078 		return;
2079 
2080 	sock_release(sk->sk_socket);
2081 }
2082 EXPORT_SYMBOL(netlink_kernel_release);
2083 
__netlink_change_ngroups(struct sock * sk,unsigned int groups)2084 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2085 {
2086 	struct listeners *new, *old;
2087 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2088 
2089 	if (groups < 32)
2090 		groups = 32;
2091 
2092 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2093 		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2094 		if (!new)
2095 			return -ENOMEM;
2096 		old = nl_deref_protected(tbl->listeners);
2097 		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2098 		rcu_assign_pointer(tbl->listeners, new);
2099 
2100 		kfree_rcu(old, rcu);
2101 	}
2102 	tbl->groups = groups;
2103 
2104 	return 0;
2105 }
2106 
2107 /**
2108  * netlink_change_ngroups - change number of multicast groups
2109  *
2110  * This changes the number of multicast groups that are available
2111  * on a certain netlink family. Note that it is not possible to
2112  * change the number of groups to below 32. Also note that it does
2113  * not implicitly call netlink_clear_multicast_users() when the
2114  * number of groups is reduced.
2115  *
2116  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2117  * @groups: The new number of groups.
2118  */
netlink_change_ngroups(struct sock * sk,unsigned int groups)2119 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2120 {
2121 	int err;
2122 
2123 	netlink_table_grab();
2124 	err = __netlink_change_ngroups(sk, groups);
2125 	netlink_table_ungrab();
2126 
2127 	return err;
2128 }
2129 
__netlink_clear_multicast_users(struct sock * ksk,unsigned int group)2130 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2131 {
2132 	struct sock *sk;
2133 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2134 	struct hlist_node *tmp;
2135 
2136 	sk_for_each_bound_safe(sk, tmp, &tbl->mc_list)
2137 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2138 }
2139 
2140 struct nlmsghdr *
__nlmsg_put(struct sk_buff * skb,u32 portid,u32 seq,int type,int len,int flags)2141 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2142 {
2143 	struct nlmsghdr *nlh;
2144 	int size = nlmsg_msg_size(len);
2145 
2146 	nlh = skb_put(skb, NLMSG_ALIGN(size));
2147 	nlh->nlmsg_type = type;
2148 	nlh->nlmsg_len = size;
2149 	nlh->nlmsg_flags = flags;
2150 	nlh->nlmsg_pid = portid;
2151 	nlh->nlmsg_seq = seq;
2152 	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2153 		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2154 	return nlh;
2155 }
2156 EXPORT_SYMBOL(__nlmsg_put);
2157 
2158 static size_t
netlink_ack_tlv_len(struct netlink_sock * nlk,int err,const struct netlink_ext_ack * extack)2159 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2160 		    const struct netlink_ext_ack *extack)
2161 {
2162 	size_t tlvlen;
2163 
2164 	if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2165 		return 0;
2166 
2167 	tlvlen = 0;
2168 	if (extack->_msg)
2169 		tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2170 	if (extack->cookie_len)
2171 		tlvlen += nla_total_size(extack->cookie_len);
2172 
2173 	/* Following attributes are only reported as error (not warning) */
2174 	if (!err)
2175 		return tlvlen;
2176 
2177 	if (extack->bad_attr)
2178 		tlvlen += nla_total_size(sizeof(u32));
2179 	if (extack->policy)
2180 		tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2181 	if (extack->miss_type)
2182 		tlvlen += nla_total_size(sizeof(u32));
2183 	if (extack->miss_nest)
2184 		tlvlen += nla_total_size(sizeof(u32));
2185 
2186 	return tlvlen;
2187 }
2188 
nlmsg_check_in_payload(const struct nlmsghdr * nlh,const void * addr)2189 static bool nlmsg_check_in_payload(const struct nlmsghdr *nlh, const void *addr)
2190 {
2191 	return !WARN_ON(addr < nlmsg_data(nlh) ||
2192 			addr - (const void *) nlh >= nlh->nlmsg_len);
2193 }
2194 
2195 static void
netlink_ack_tlv_fill(struct sk_buff * skb,const struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2196 netlink_ack_tlv_fill(struct sk_buff *skb, const struct nlmsghdr *nlh, int err,
2197 		     const struct netlink_ext_ack *extack)
2198 {
2199 	if (extack->_msg)
2200 		WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2201 	if (extack->cookie_len)
2202 		WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2203 				extack->cookie_len, extack->cookie));
2204 
2205 	if (!err)
2206 		return;
2207 
2208 	if (extack->bad_attr && nlmsg_check_in_payload(nlh, extack->bad_attr))
2209 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2210 				    (u8 *)extack->bad_attr - (const u8 *)nlh));
2211 	if (extack->policy)
2212 		netlink_policy_dump_write_attr(skb, extack->policy,
2213 					       NLMSGERR_ATTR_POLICY);
2214 	if (extack->miss_type)
2215 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2216 				    extack->miss_type));
2217 	if (extack->miss_nest && nlmsg_check_in_payload(nlh, extack->miss_nest))
2218 		WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2219 				    (u8 *)extack->miss_nest - (const u8 *)nlh));
2220 }
2221 
2222 /*
2223  * It looks a bit ugly.
2224  * It would be better to create kernel thread.
2225  */
2226 
netlink_dump_done(struct netlink_sock * nlk,struct sk_buff * skb,struct netlink_callback * cb,struct netlink_ext_ack * extack)2227 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2228 			     struct netlink_callback *cb,
2229 			     struct netlink_ext_ack *extack)
2230 {
2231 	struct nlmsghdr *nlh;
2232 	size_t extack_len;
2233 
2234 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2235 			       NLM_F_MULTI | cb->answer_flags);
2236 	if (WARN_ON(!nlh))
2237 		return -ENOBUFS;
2238 
2239 	nl_dump_check_consistent(cb, nlh);
2240 	memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2241 
2242 	extack_len = netlink_ack_tlv_len(nlk, nlk->dump_done_errno, extack);
2243 	if (extack_len) {
2244 		nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2245 		if (skb_tailroom(skb) >= extack_len) {
2246 			netlink_ack_tlv_fill(skb, cb->nlh,
2247 					     nlk->dump_done_errno, extack);
2248 			nlmsg_end(skb, nlh);
2249 		}
2250 	}
2251 
2252 	return 0;
2253 }
2254 
netlink_dump(struct sock * sk,bool lock_taken)2255 static int netlink_dump(struct sock *sk, bool lock_taken)
2256 {
2257 	struct netlink_sock *nlk = nlk_sk(sk);
2258 	struct netlink_ext_ack extack = {};
2259 	struct netlink_callback *cb;
2260 	struct sk_buff *skb = NULL;
2261 	size_t max_recvmsg_len;
2262 	struct module *module;
2263 	int err = -ENOBUFS;
2264 	int alloc_min_size;
2265 	unsigned int rmem;
2266 	int alloc_size;
2267 
2268 	if (!lock_taken)
2269 		mutex_lock(&nlk->nl_cb_mutex);
2270 	if (!nlk->cb_running) {
2271 		err = -EINVAL;
2272 		goto errout_skb;
2273 	}
2274 
2275 	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2276 	 * required, but it makes sense to _attempt_ a 32KiB allocation
2277 	 * to reduce number of system calls on dump operations, if user
2278 	 * ever provided a big enough buffer.
2279 	 */
2280 	cb = &nlk->cb;
2281 	alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2282 
2283 	max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2284 	if (alloc_min_size < max_recvmsg_len) {
2285 		alloc_size = max_recvmsg_len;
2286 		skb = alloc_skb(alloc_size,
2287 				(GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2288 				__GFP_NOWARN | __GFP_NORETRY);
2289 	}
2290 	if (!skb) {
2291 		alloc_size = alloc_min_size;
2292 		skb = alloc_skb(alloc_size, GFP_KERNEL);
2293 	}
2294 	if (!skb)
2295 		goto errout_skb;
2296 
2297 	rmem = atomic_add_return(skb->truesize, &sk->sk_rmem_alloc);
2298 	if (rmem >= READ_ONCE(sk->sk_rcvbuf)) {
2299 		atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
2300 		goto errout_skb;
2301 	}
2302 
2303 	/* Trim skb to allocated size. User is expected to provide buffer as
2304 	 * large as max(min_dump_alloc, 32KiB (max_recvmsg_len capped at
2305 	 * netlink_recvmsg())). dump will pack as many smaller messages as
2306 	 * could fit within the allocated skb. skb is typically allocated
2307 	 * with larger space than required (could be as much as near 2x the
2308 	 * requested size with align to next power of 2 approach). Allowing
2309 	 * dump to use the excess space makes it difficult for a user to have a
2310 	 * reasonable static buffer based on the expected largest dump of a
2311 	 * single netdev. The outcome is MSG_TRUNC error.
2312 	 */
2313 	skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2314 
2315 	/* Make sure malicious BPF programs can not read unitialized memory
2316 	 * from skb->head -> skb->data
2317 	 */
2318 	skb_reset_network_header(skb);
2319 	skb_reset_mac_header(skb);
2320 
2321 	netlink_skb_set_owner_r(skb, sk);
2322 
2323 	if (nlk->dump_done_errno > 0) {
2324 		cb->extack = &extack;
2325 
2326 		nlk->dump_done_errno = cb->dump(skb, cb);
2327 
2328 		/* EMSGSIZE plus something already in the skb means
2329 		 * that there's more to dump but current skb has filled up.
2330 		 * If the callback really wants to return EMSGSIZE to user space
2331 		 * it needs to do so again, on the next cb->dump() call,
2332 		 * without putting data in the skb.
2333 		 */
2334 		if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2335 			nlk->dump_done_errno = skb->len;
2336 
2337 		cb->extack = NULL;
2338 	}
2339 
2340 	if (nlk->dump_done_errno > 0 ||
2341 	    skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2342 		mutex_unlock(&nlk->nl_cb_mutex);
2343 
2344 		if (sk_filter(sk, skb))
2345 			kfree_skb(skb);
2346 		else
2347 			__netlink_sendskb(sk, skb);
2348 		return 0;
2349 	}
2350 
2351 	if (netlink_dump_done(nlk, skb, cb, &extack))
2352 		goto errout_skb;
2353 
2354 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2355 	/* frag_list skb's data is used for compat tasks
2356 	 * and the regular skb's data for normal (non-compat) tasks.
2357 	 * See netlink_recvmsg().
2358 	 */
2359 	if (unlikely(skb_shinfo(skb)->frag_list)) {
2360 		if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2361 			goto errout_skb;
2362 	}
2363 #endif
2364 
2365 	if (sk_filter(sk, skb))
2366 		kfree_skb(skb);
2367 	else
2368 		__netlink_sendskb(sk, skb);
2369 
2370 	if (cb->done)
2371 		cb->done(cb);
2372 
2373 	WRITE_ONCE(nlk->cb_running, false);
2374 	module = cb->module;
2375 	skb = cb->skb;
2376 	mutex_unlock(&nlk->nl_cb_mutex);
2377 	module_put(module);
2378 	consume_skb(skb);
2379 	return 0;
2380 
2381 errout_skb:
2382 	mutex_unlock(&nlk->nl_cb_mutex);
2383 	kfree_skb(skb);
2384 	return err;
2385 }
2386 
__netlink_dump_start(struct sock * ssk,struct sk_buff * skb,const struct nlmsghdr * nlh,struct netlink_dump_control * control)2387 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2388 			 const struct nlmsghdr *nlh,
2389 			 struct netlink_dump_control *control)
2390 {
2391 	struct netlink_callback *cb;
2392 	struct netlink_sock *nlk;
2393 	struct sock *sk;
2394 	int ret;
2395 
2396 	refcount_inc(&skb->users);
2397 
2398 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2399 	if (sk == NULL) {
2400 		ret = -ECONNREFUSED;
2401 		goto error_free;
2402 	}
2403 
2404 	nlk = nlk_sk(sk);
2405 	mutex_lock(&nlk->nl_cb_mutex);
2406 	/* A dump is in progress... */
2407 	if (nlk->cb_running) {
2408 		ret = -EBUSY;
2409 		goto error_unlock;
2410 	}
2411 	/* add reference of module which cb->dump belongs to */
2412 	if (!try_module_get(control->module)) {
2413 		ret = -EPROTONOSUPPORT;
2414 		goto error_unlock;
2415 	}
2416 
2417 	cb = &nlk->cb;
2418 	memset(cb, 0, sizeof(*cb));
2419 	cb->dump = control->dump;
2420 	cb->done = control->done;
2421 	cb->nlh = nlh;
2422 	cb->data = control->data;
2423 	cb->module = control->module;
2424 	cb->min_dump_alloc = control->min_dump_alloc;
2425 	cb->flags = control->flags;
2426 	cb->skb = skb;
2427 
2428 	cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2429 
2430 	if (control->start) {
2431 		cb->extack = control->extack;
2432 		ret = control->start(cb);
2433 		cb->extack = NULL;
2434 		if (ret)
2435 			goto error_put;
2436 	}
2437 
2438 	WRITE_ONCE(nlk->cb_running, true);
2439 	nlk->dump_done_errno = INT_MAX;
2440 
2441 	ret = netlink_dump(sk, true);
2442 
2443 	sock_put(sk);
2444 
2445 	if (ret)
2446 		return ret;
2447 
2448 	/* We successfully started a dump, by returning -EINTR we
2449 	 * signal not to send ACK even if it was requested.
2450 	 */
2451 	return -EINTR;
2452 
2453 error_put:
2454 	module_put(control->module);
2455 error_unlock:
2456 	sock_put(sk);
2457 	mutex_unlock(&nlk->nl_cb_mutex);
2458 error_free:
2459 	kfree_skb(skb);
2460 	return ret;
2461 }
2462 EXPORT_SYMBOL(__netlink_dump_start);
2463 
netlink_ack(struct sk_buff * in_skb,struct nlmsghdr * nlh,int err,const struct netlink_ext_ack * extack)2464 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2465 		 const struct netlink_ext_ack *extack)
2466 {
2467 	struct sk_buff *skb;
2468 	struct nlmsghdr *rep;
2469 	struct nlmsgerr *errmsg;
2470 	size_t payload = sizeof(*errmsg);
2471 	struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2472 	unsigned int flags = 0;
2473 	size_t tlvlen;
2474 
2475 	/* Error messages get the original request appened, unless the user
2476 	 * requests to cap the error message, and get extra error data if
2477 	 * requested.
2478 	 */
2479 	if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2480 		payload += nlmsg_len(nlh);
2481 	else
2482 		flags |= NLM_F_CAPPED;
2483 
2484 	tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2485 	if (tlvlen)
2486 		flags |= NLM_F_ACK_TLVS;
2487 
2488 	skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2489 	if (!skb)
2490 		goto err_skb;
2491 
2492 	rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2493 			NLMSG_ERROR, sizeof(*errmsg), flags);
2494 	if (!rep)
2495 		goto err_bad_put;
2496 	errmsg = nlmsg_data(rep);
2497 	errmsg->error = err;
2498 	errmsg->msg = *nlh;
2499 
2500 	if (!(flags & NLM_F_CAPPED)) {
2501 		if (!nlmsg_append(skb, nlmsg_len(nlh)))
2502 			goto err_bad_put;
2503 
2504 		memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2505 		       nlmsg_len(nlh));
2506 	}
2507 
2508 	if (tlvlen)
2509 		netlink_ack_tlv_fill(skb, nlh, err, extack);
2510 
2511 	nlmsg_end(skb, rep);
2512 
2513 	nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2514 
2515 	return;
2516 
2517 err_bad_put:
2518 	nlmsg_free(skb);
2519 err_skb:
2520 	WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2521 	sk_error_report(NETLINK_CB(in_skb).sk);
2522 }
2523 EXPORT_SYMBOL(netlink_ack);
2524 
netlink_rcv_skb(struct sk_buff * skb,int (* cb)(struct sk_buff *,struct nlmsghdr *,struct netlink_ext_ack *))2525 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2526 						   struct nlmsghdr *,
2527 						   struct netlink_ext_ack *))
2528 {
2529 	struct netlink_ext_ack extack;
2530 	struct nlmsghdr *nlh;
2531 	int err;
2532 
2533 	while (skb->len >= nlmsg_total_size(0)) {
2534 		int msglen;
2535 
2536 		memset(&extack, 0, sizeof(extack));
2537 		nlh = nlmsg_hdr(skb);
2538 		err = 0;
2539 
2540 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2541 			return 0;
2542 
2543 		/* Only requests are handled by the kernel */
2544 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2545 			goto ack;
2546 
2547 		/* Skip control messages */
2548 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2549 			goto ack;
2550 
2551 		err = cb(skb, nlh, &extack);
2552 		if (err == -EINTR)
2553 			goto skip;
2554 
2555 ack:
2556 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2557 			netlink_ack(skb, nlh, err, &extack);
2558 
2559 skip:
2560 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2561 		if (msglen > skb->len)
2562 			msglen = skb->len;
2563 		skb_pull(skb, msglen);
2564 	}
2565 
2566 	return 0;
2567 }
2568 EXPORT_SYMBOL(netlink_rcv_skb);
2569 
2570 /**
2571  * nlmsg_notify - send a notification netlink message
2572  * @sk: netlink socket to use
2573  * @skb: notification message
2574  * @portid: destination netlink portid for reports or 0
2575  * @group: destination multicast group or 0
2576  * @report: 1 to report back, 0 to disable
2577  * @flags: allocation flags
2578  */
nlmsg_notify(struct sock * sk,struct sk_buff * skb,u32 portid,unsigned int group,int report,gfp_t flags)2579 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2580 		 unsigned int group, int report, gfp_t flags)
2581 {
2582 	int err = 0;
2583 
2584 	if (group) {
2585 		int exclude_portid = 0;
2586 
2587 		if (report) {
2588 			refcount_inc(&skb->users);
2589 			exclude_portid = portid;
2590 		}
2591 
2592 		/* errors reported via destination sk->sk_err, but propagate
2593 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2594 		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2595 		if (err == -ESRCH)
2596 			err = 0;
2597 	}
2598 
2599 	if (report) {
2600 		int err2;
2601 
2602 		err2 = nlmsg_unicast(sk, skb, portid);
2603 		if (!err)
2604 			err = err2;
2605 	}
2606 
2607 	return err;
2608 }
2609 EXPORT_SYMBOL(nlmsg_notify);
2610 
2611 #ifdef CONFIG_PROC_FS
2612 struct nl_seq_iter {
2613 	struct seq_net_private p;
2614 	struct rhashtable_iter hti;
2615 	int link;
2616 };
2617 
netlink_walk_start(struct nl_seq_iter * iter)2618 static void netlink_walk_start(struct nl_seq_iter *iter)
2619 {
2620 	rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2621 	rhashtable_walk_start(&iter->hti);
2622 }
2623 
netlink_walk_stop(struct nl_seq_iter * iter)2624 static void netlink_walk_stop(struct nl_seq_iter *iter)
2625 {
2626 	rhashtable_walk_stop(&iter->hti);
2627 	rhashtable_walk_exit(&iter->hti);
2628 }
2629 
__netlink_seq_next(struct seq_file * seq)2630 static void *__netlink_seq_next(struct seq_file *seq)
2631 {
2632 	struct nl_seq_iter *iter = seq->private;
2633 	struct netlink_sock *nlk;
2634 
2635 	do {
2636 		for (;;) {
2637 			nlk = rhashtable_walk_next(&iter->hti);
2638 
2639 			if (IS_ERR(nlk)) {
2640 				if (PTR_ERR(nlk) == -EAGAIN)
2641 					continue;
2642 
2643 				return nlk;
2644 			}
2645 
2646 			if (nlk)
2647 				break;
2648 
2649 			netlink_walk_stop(iter);
2650 			if (++iter->link >= MAX_LINKS)
2651 				return NULL;
2652 
2653 			netlink_walk_start(iter);
2654 		}
2655 	} while (sock_net(&nlk->sk) != seq_file_net(seq));
2656 
2657 	return nlk;
2658 }
2659 
netlink_seq_start(struct seq_file * seq,loff_t * posp)2660 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2661 	__acquires(RCU)
2662 {
2663 	struct nl_seq_iter *iter = seq->private;
2664 	void *obj = SEQ_START_TOKEN;
2665 	loff_t pos;
2666 
2667 	iter->link = 0;
2668 
2669 	netlink_walk_start(iter);
2670 
2671 	for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2672 		obj = __netlink_seq_next(seq);
2673 
2674 	return obj;
2675 }
2676 
netlink_seq_next(struct seq_file * seq,void * v,loff_t * pos)2677 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2678 {
2679 	++*pos;
2680 	return __netlink_seq_next(seq);
2681 }
2682 
netlink_native_seq_stop(struct seq_file * seq,void * v)2683 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2684 {
2685 	struct nl_seq_iter *iter = seq->private;
2686 
2687 	if (iter->link >= MAX_LINKS)
2688 		return;
2689 
2690 	netlink_walk_stop(iter);
2691 }
2692 
2693 
netlink_native_seq_show(struct seq_file * seq,void * v)2694 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2695 {
2696 	if (v == SEQ_START_TOKEN) {
2697 		seq_puts(seq,
2698 			 "sk               Eth Pid        Groups   "
2699 			 "Rmem     Wmem     Dump  Locks    Drops    Inode\n");
2700 	} else {
2701 		struct sock *s = v;
2702 		struct netlink_sock *nlk = nlk_sk(s);
2703 
2704 		seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2705 			   s,
2706 			   s->sk_protocol,
2707 			   nlk->portid,
2708 			   nlk->groups ? (u32)nlk->groups[0] : 0,
2709 			   sk_rmem_alloc_get(s),
2710 			   sk_wmem_alloc_get(s),
2711 			   READ_ONCE(nlk->cb_running),
2712 			   refcount_read(&s->sk_refcnt),
2713 			   atomic_read(&s->sk_drops),
2714 			   sock_i_ino(s)
2715 			);
2716 
2717 	}
2718 	return 0;
2719 }
2720 
2721 #ifdef CONFIG_BPF_SYSCALL
2722 struct bpf_iter__netlink {
2723 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2724 	__bpf_md_ptr(struct netlink_sock *, sk);
2725 };
2726 
DEFINE_BPF_ITER_FUNC(netlink,struct bpf_iter_meta * meta,struct netlink_sock * sk)2727 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2728 
2729 static int netlink_prog_seq_show(struct bpf_prog *prog,
2730 				  struct bpf_iter_meta *meta,
2731 				  void *v)
2732 {
2733 	struct bpf_iter__netlink ctx;
2734 
2735 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2736 	ctx.meta = meta;
2737 	ctx.sk = nlk_sk((struct sock *)v);
2738 	return bpf_iter_run_prog(prog, &ctx);
2739 }
2740 
netlink_seq_show(struct seq_file * seq,void * v)2741 static int netlink_seq_show(struct seq_file *seq, void *v)
2742 {
2743 	struct bpf_iter_meta meta;
2744 	struct bpf_prog *prog;
2745 
2746 	meta.seq = seq;
2747 	prog = bpf_iter_get_info(&meta, false);
2748 	if (!prog)
2749 		return netlink_native_seq_show(seq, v);
2750 
2751 	if (v != SEQ_START_TOKEN)
2752 		return netlink_prog_seq_show(prog, &meta, v);
2753 
2754 	return 0;
2755 }
2756 
netlink_seq_stop(struct seq_file * seq,void * v)2757 static void netlink_seq_stop(struct seq_file *seq, void *v)
2758 {
2759 	struct bpf_iter_meta meta;
2760 	struct bpf_prog *prog;
2761 
2762 	if (!v) {
2763 		meta.seq = seq;
2764 		prog = bpf_iter_get_info(&meta, true);
2765 		if (prog)
2766 			(void)netlink_prog_seq_show(prog, &meta, v);
2767 	}
2768 
2769 	netlink_native_seq_stop(seq, v);
2770 }
2771 #else
netlink_seq_show(struct seq_file * seq,void * v)2772 static int netlink_seq_show(struct seq_file *seq, void *v)
2773 {
2774 	return netlink_native_seq_show(seq, v);
2775 }
2776 
netlink_seq_stop(struct seq_file * seq,void * v)2777 static void netlink_seq_stop(struct seq_file *seq, void *v)
2778 {
2779 	netlink_native_seq_stop(seq, v);
2780 }
2781 #endif
2782 
2783 static const struct seq_operations netlink_seq_ops = {
2784 	.start  = netlink_seq_start,
2785 	.next   = netlink_seq_next,
2786 	.stop   = netlink_seq_stop,
2787 	.show   = netlink_seq_show,
2788 };
2789 #endif
2790 
netlink_register_notifier(struct notifier_block * nb)2791 int netlink_register_notifier(struct notifier_block *nb)
2792 {
2793 	return blocking_notifier_chain_register(&netlink_chain, nb);
2794 }
2795 EXPORT_SYMBOL(netlink_register_notifier);
2796 
netlink_unregister_notifier(struct notifier_block * nb)2797 int netlink_unregister_notifier(struct notifier_block *nb)
2798 {
2799 	return blocking_notifier_chain_unregister(&netlink_chain, nb);
2800 }
2801 EXPORT_SYMBOL(netlink_unregister_notifier);
2802 
2803 static const struct proto_ops netlink_ops = {
2804 	.family =	PF_NETLINK,
2805 	.owner =	THIS_MODULE,
2806 	.release =	netlink_release,
2807 	.bind =		netlink_bind,
2808 	.connect =	netlink_connect,
2809 	.socketpair =	sock_no_socketpair,
2810 	.accept =	sock_no_accept,
2811 	.getname =	netlink_getname,
2812 	.poll =		datagram_poll,
2813 	.ioctl =	netlink_ioctl,
2814 	.listen =	sock_no_listen,
2815 	.shutdown =	sock_no_shutdown,
2816 	.setsockopt =	netlink_setsockopt,
2817 	.getsockopt =	netlink_getsockopt,
2818 	.sendmsg =	netlink_sendmsg,
2819 	.recvmsg =	netlink_recvmsg,
2820 	.mmap =		sock_no_mmap,
2821 };
2822 
2823 static const struct net_proto_family netlink_family_ops = {
2824 	.family = PF_NETLINK,
2825 	.create = netlink_create,
2826 	.owner	= THIS_MODULE,	/* for consistency 8) */
2827 };
2828 
netlink_net_init(struct net * net)2829 static int __net_init netlink_net_init(struct net *net)
2830 {
2831 #ifdef CONFIG_PROC_FS
2832 	if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2833 			sizeof(struct nl_seq_iter)))
2834 		return -ENOMEM;
2835 #endif
2836 	return 0;
2837 }
2838 
netlink_net_exit(struct net * net)2839 static void __net_exit netlink_net_exit(struct net *net)
2840 {
2841 #ifdef CONFIG_PROC_FS
2842 	remove_proc_entry("netlink", net->proc_net);
2843 #endif
2844 }
2845 
netlink_add_usersock_entry(void)2846 static void __init netlink_add_usersock_entry(void)
2847 {
2848 	struct listeners *listeners;
2849 	int groups = 32;
2850 
2851 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2852 	if (!listeners)
2853 		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2854 
2855 	netlink_table_grab();
2856 
2857 	nl_table[NETLINK_USERSOCK].groups = groups;
2858 	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2859 	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2860 	nl_table[NETLINK_USERSOCK].registered = 1;
2861 	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2862 
2863 	netlink_table_ungrab();
2864 }
2865 
2866 static struct pernet_operations __net_initdata netlink_net_ops = {
2867 	.init = netlink_net_init,
2868 	.exit = netlink_net_exit,
2869 };
2870 
netlink_hash(const void * data,u32 len,u32 seed)2871 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2872 {
2873 	const struct netlink_sock *nlk = data;
2874 	struct netlink_compare_arg arg;
2875 
2876 	netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2877 	return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2878 }
2879 
2880 static const struct rhashtable_params netlink_rhashtable_params = {
2881 	.head_offset = offsetof(struct netlink_sock, node),
2882 	.key_len = netlink_compare_arg_len,
2883 	.obj_hashfn = netlink_hash,
2884 	.obj_cmpfn = netlink_compare,
2885 	.automatic_shrinking = true,
2886 };
2887 
2888 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2889 BTF_ID_LIST(btf_netlink_sock_id)
2890 BTF_ID(struct, netlink_sock)
2891 
2892 static const struct bpf_iter_seq_info netlink_seq_info = {
2893 	.seq_ops		= &netlink_seq_ops,
2894 	.init_seq_private	= bpf_iter_init_seq_net,
2895 	.fini_seq_private	= bpf_iter_fini_seq_net,
2896 	.seq_priv_size		= sizeof(struct nl_seq_iter),
2897 };
2898 
2899 static struct bpf_iter_reg netlink_reg_info = {
2900 	.target			= "netlink",
2901 	.ctx_arg_info_size	= 1,
2902 	.ctx_arg_info		= {
2903 		{ offsetof(struct bpf_iter__netlink, sk),
2904 		  PTR_TO_BTF_ID_OR_NULL },
2905 	},
2906 	.seq_info		= &netlink_seq_info,
2907 };
2908 
bpf_iter_register(void)2909 static int __init bpf_iter_register(void)
2910 {
2911 	netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2912 	return bpf_iter_reg_target(&netlink_reg_info);
2913 }
2914 #endif
2915 
netlink_proto_init(void)2916 static int __init netlink_proto_init(void)
2917 {
2918 	int i;
2919 	int err = proto_register(&netlink_proto, 0);
2920 
2921 	if (err != 0)
2922 		goto out;
2923 
2924 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2925 	err = bpf_iter_register();
2926 	if (err)
2927 		goto out;
2928 #endif
2929 
2930 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2931 
2932 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2933 	if (!nl_table)
2934 		goto panic;
2935 
2936 	for (i = 0; i < MAX_LINKS; i++) {
2937 		if (rhashtable_init(&nl_table[i].hash,
2938 				    &netlink_rhashtable_params) < 0)
2939 			goto panic;
2940 	}
2941 
2942 	netlink_add_usersock_entry();
2943 
2944 	sock_register(&netlink_family_ops);
2945 	register_pernet_subsys(&netlink_net_ops);
2946 	register_pernet_subsys(&netlink_tap_net_ops);
2947 	/* The netlink device handler may be needed early. */
2948 	rtnetlink_init();
2949 out:
2950 	return err;
2951 panic:
2952 	panic("netlink_init: Cannot allocate nl_table\n");
2953 }
2954 
2955 core_initcall(netlink_proto_init);
2956