xref: /openbmc/linux/drivers/net/wireguard/device.c (revision bacf743e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
4  */
5 
6 #include "queueing.h"
7 #include "socket.h"
8 #include "timers.h"
9 #include "device.h"
10 #include "ratelimiter.h"
11 #include "peer.h"
12 #include "messages.h"
13 
14 #include <linux/module.h>
15 #include <linux/rtnetlink.h>
16 #include <linux/inet.h>
17 #include <linux/netdevice.h>
18 #include <linux/inetdevice.h>
19 #include <linux/if_arp.h>
20 #include <linux/icmp.h>
21 #include <linux/suspend.h>
22 #include <net/icmp.h>
23 #include <net/rtnetlink.h>
24 #include <net/ip_tunnels.h>
25 #include <net/addrconf.h>
26 
27 static LIST_HEAD(device_list);
28 
29 static int wg_open(struct net_device *dev)
30 {
31 	struct in_device *dev_v4 = __in_dev_get_rtnl(dev);
32 	struct inet6_dev *dev_v6 = __in6_dev_get(dev);
33 	struct wg_device *wg = netdev_priv(dev);
34 	struct wg_peer *peer;
35 	int ret;
36 
37 	if (dev_v4) {
38 		/* At some point we might put this check near the ip_rt_send_
39 		 * redirect call of ip_forward in net/ipv4/ip_forward.c, similar
40 		 * to the current secpath check.
41 		 */
42 		IN_DEV_CONF_SET(dev_v4, SEND_REDIRECTS, false);
43 		IPV4_DEVCONF_ALL(dev_net(dev), SEND_REDIRECTS) = false;
44 	}
45 	if (dev_v6)
46 		dev_v6->cnf.addr_gen_mode = IN6_ADDR_GEN_MODE_NONE;
47 
48 	mutex_lock(&wg->device_update_lock);
49 	ret = wg_socket_init(wg, wg->incoming_port);
50 	if (ret < 0)
51 		goto out;
52 	list_for_each_entry(peer, &wg->peer_list, peer_list) {
53 		wg_packet_send_staged_packets(peer);
54 		if (peer->persistent_keepalive_interval)
55 			wg_packet_send_keepalive(peer);
56 	}
57 out:
58 	mutex_unlock(&wg->device_update_lock);
59 	return ret;
60 }
61 
62 static int wg_pm_notification(struct notifier_block *nb, unsigned long action, void *data)
63 {
64 	struct wg_device *wg;
65 	struct wg_peer *peer;
66 
67 	/* If the machine is constantly suspending and resuming, as part of
68 	 * its normal operation rather than as a somewhat rare event, then we
69 	 * don't actually want to clear keys.
70 	 */
71 	if (IS_ENABLED(CONFIG_PM_AUTOSLEEP) || IS_ENABLED(CONFIG_ANDROID))
72 		return 0;
73 
74 	if (action != PM_HIBERNATION_PREPARE && action != PM_SUSPEND_PREPARE)
75 		return 0;
76 
77 	rtnl_lock();
78 	list_for_each_entry(wg, &device_list, device_list) {
79 		mutex_lock(&wg->device_update_lock);
80 		list_for_each_entry(peer, &wg->peer_list, peer_list) {
81 			del_timer(&peer->timer_zero_key_material);
82 			wg_noise_handshake_clear(&peer->handshake);
83 			wg_noise_keypairs_clear(&peer->keypairs);
84 		}
85 		mutex_unlock(&wg->device_update_lock);
86 	}
87 	rtnl_unlock();
88 	rcu_barrier();
89 	return 0;
90 }
91 
92 static struct notifier_block pm_notifier = { .notifier_call = wg_pm_notification };
93 
94 static int wg_vm_notification(struct notifier_block *nb, unsigned long action, void *data)
95 {
96 	struct wg_device *wg;
97 	struct wg_peer *peer;
98 
99 	rtnl_lock();
100 	list_for_each_entry(wg, &device_list, device_list) {
101 		mutex_lock(&wg->device_update_lock);
102 		list_for_each_entry(peer, &wg->peer_list, peer_list)
103 			wg_noise_expire_current_peer_keypairs(peer);
104 		mutex_unlock(&wg->device_update_lock);
105 	}
106 	rtnl_unlock();
107 	return 0;
108 }
109 
110 static struct notifier_block vm_notifier = { .notifier_call = wg_vm_notification };
111 
112 static int wg_stop(struct net_device *dev)
113 {
114 	struct wg_device *wg = netdev_priv(dev);
115 	struct wg_peer *peer;
116 	struct sk_buff *skb;
117 
118 	mutex_lock(&wg->device_update_lock);
119 	list_for_each_entry(peer, &wg->peer_list, peer_list) {
120 		wg_packet_purge_staged_packets(peer);
121 		wg_timers_stop(peer);
122 		wg_noise_handshake_clear(&peer->handshake);
123 		wg_noise_keypairs_clear(&peer->keypairs);
124 		wg_noise_reset_last_sent_handshake(&peer->last_sent_handshake);
125 	}
126 	mutex_unlock(&wg->device_update_lock);
127 	while ((skb = ptr_ring_consume(&wg->handshake_queue.ring)) != NULL)
128 		kfree_skb(skb);
129 	atomic_set(&wg->handshake_queue_len, 0);
130 	wg_socket_reinit(wg, NULL, NULL);
131 	return 0;
132 }
133 
134 static netdev_tx_t wg_xmit(struct sk_buff *skb, struct net_device *dev)
135 {
136 	struct wg_device *wg = netdev_priv(dev);
137 	struct sk_buff_head packets;
138 	struct wg_peer *peer;
139 	struct sk_buff *next;
140 	sa_family_t family;
141 	u32 mtu;
142 	int ret;
143 
144 	if (unlikely(!wg_check_packet_protocol(skb))) {
145 		ret = -EPROTONOSUPPORT;
146 		net_dbg_ratelimited("%s: Invalid IP packet\n", dev->name);
147 		goto err;
148 	}
149 
150 	peer = wg_allowedips_lookup_dst(&wg->peer_allowedips, skb);
151 	if (unlikely(!peer)) {
152 		ret = -ENOKEY;
153 		if (skb->protocol == htons(ETH_P_IP))
154 			net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI4\n",
155 					    dev->name, &ip_hdr(skb)->daddr);
156 		else if (skb->protocol == htons(ETH_P_IPV6))
157 			net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI6\n",
158 					    dev->name, &ipv6_hdr(skb)->daddr);
159 		goto err_icmp;
160 	}
161 
162 	family = READ_ONCE(peer->endpoint.addr.sa_family);
163 	if (unlikely(family != AF_INET && family != AF_INET6)) {
164 		ret = -EDESTADDRREQ;
165 		net_dbg_ratelimited("%s: No valid endpoint has been configured or discovered for peer %llu\n",
166 				    dev->name, peer->internal_id);
167 		goto err_peer;
168 	}
169 
170 	mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
171 
172 	__skb_queue_head_init(&packets);
173 	if (!skb_is_gso(skb)) {
174 		skb_mark_not_on_list(skb);
175 	} else {
176 		struct sk_buff *segs = skb_gso_segment(skb, 0);
177 
178 		if (IS_ERR(segs)) {
179 			ret = PTR_ERR(segs);
180 			goto err_peer;
181 		}
182 		dev_kfree_skb(skb);
183 		skb = segs;
184 	}
185 
186 	skb_list_walk_safe(skb, skb, next) {
187 		skb_mark_not_on_list(skb);
188 
189 		skb = skb_share_check(skb, GFP_ATOMIC);
190 		if (unlikely(!skb))
191 			continue;
192 
193 		/* We only need to keep the original dst around for icmp,
194 		 * so at this point we're in a position to drop it.
195 		 */
196 		skb_dst_drop(skb);
197 
198 		PACKET_CB(skb)->mtu = mtu;
199 
200 		__skb_queue_tail(&packets, skb);
201 	}
202 
203 	spin_lock_bh(&peer->staged_packet_queue.lock);
204 	/* If the queue is getting too big, we start removing the oldest packets
205 	 * until it's small again. We do this before adding the new packet, so
206 	 * we don't remove GSO segments that are in excess.
207 	 */
208 	while (skb_queue_len(&peer->staged_packet_queue) > MAX_STAGED_PACKETS) {
209 		dev_kfree_skb(__skb_dequeue(&peer->staged_packet_queue));
210 		++dev->stats.tx_dropped;
211 	}
212 	skb_queue_splice_tail(&packets, &peer->staged_packet_queue);
213 	spin_unlock_bh(&peer->staged_packet_queue.lock);
214 
215 	wg_packet_send_staged_packets(peer);
216 
217 	wg_peer_put(peer);
218 	return NETDEV_TX_OK;
219 
220 err_peer:
221 	wg_peer_put(peer);
222 err_icmp:
223 	if (skb->protocol == htons(ETH_P_IP))
224 		icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
225 	else if (skb->protocol == htons(ETH_P_IPV6))
226 		icmpv6_ndo_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
227 err:
228 	++dev->stats.tx_errors;
229 	kfree_skb(skb);
230 	return ret;
231 }
232 
233 static const struct net_device_ops netdev_ops = {
234 	.ndo_open		= wg_open,
235 	.ndo_stop		= wg_stop,
236 	.ndo_start_xmit		= wg_xmit,
237 	.ndo_get_stats64	= dev_get_tstats64
238 };
239 
240 static void wg_destruct(struct net_device *dev)
241 {
242 	struct wg_device *wg = netdev_priv(dev);
243 
244 	rtnl_lock();
245 	list_del(&wg->device_list);
246 	rtnl_unlock();
247 	mutex_lock(&wg->device_update_lock);
248 	rcu_assign_pointer(wg->creating_net, NULL);
249 	wg->incoming_port = 0;
250 	wg_socket_reinit(wg, NULL, NULL);
251 	/* The final references are cleared in the below calls to destroy_workqueue. */
252 	wg_peer_remove_all(wg);
253 	destroy_workqueue(wg->handshake_receive_wq);
254 	destroy_workqueue(wg->handshake_send_wq);
255 	destroy_workqueue(wg->packet_crypt_wq);
256 	wg_packet_queue_free(&wg->handshake_queue, true);
257 	wg_packet_queue_free(&wg->decrypt_queue, false);
258 	wg_packet_queue_free(&wg->encrypt_queue, false);
259 	rcu_barrier(); /* Wait for all the peers to be actually freed. */
260 	wg_ratelimiter_uninit();
261 	memzero_explicit(&wg->static_identity, sizeof(wg->static_identity));
262 	free_percpu(dev->tstats);
263 	kvfree(wg->index_hashtable);
264 	kvfree(wg->peer_hashtable);
265 	mutex_unlock(&wg->device_update_lock);
266 
267 	pr_debug("%s: Interface destroyed\n", dev->name);
268 	free_netdev(dev);
269 }
270 
271 static const struct device_type device_type = { .name = KBUILD_MODNAME };
272 
273 static void wg_setup(struct net_device *dev)
274 {
275 	struct wg_device *wg = netdev_priv(dev);
276 	enum { WG_NETDEV_FEATURES = NETIF_F_HW_CSUM | NETIF_F_RXCSUM |
277 				    NETIF_F_SG | NETIF_F_GSO |
278 				    NETIF_F_GSO_SOFTWARE | NETIF_F_HIGHDMA };
279 	const int overhead = MESSAGE_MINIMUM_LENGTH + sizeof(struct udphdr) +
280 			     max(sizeof(struct ipv6hdr), sizeof(struct iphdr));
281 
282 	dev->netdev_ops = &netdev_ops;
283 	dev->header_ops = &ip_tunnel_header_ops;
284 	dev->hard_header_len = 0;
285 	dev->addr_len = 0;
286 	dev->needed_headroom = DATA_PACKET_HEAD_ROOM;
287 	dev->needed_tailroom = noise_encrypted_len(MESSAGE_PADDING_MULTIPLE);
288 	dev->type = ARPHRD_NONE;
289 	dev->flags = IFF_POINTOPOINT | IFF_NOARP;
290 	dev->priv_flags |= IFF_NO_QUEUE;
291 	dev->features |= NETIF_F_LLTX;
292 	dev->features |= WG_NETDEV_FEATURES;
293 	dev->hw_features |= WG_NETDEV_FEATURES;
294 	dev->hw_enc_features |= WG_NETDEV_FEATURES;
295 	dev->mtu = ETH_DATA_LEN - overhead;
296 	dev->max_mtu = round_down(INT_MAX, MESSAGE_PADDING_MULTIPLE) - overhead;
297 
298 	SET_NETDEV_DEVTYPE(dev, &device_type);
299 
300 	/* We need to keep the dst around in case of icmp replies. */
301 	netif_keep_dst(dev);
302 
303 	memset(wg, 0, sizeof(*wg));
304 	wg->dev = dev;
305 }
306 
307 static int wg_newlink(struct net *src_net, struct net_device *dev,
308 		      struct nlattr *tb[], struct nlattr *data[],
309 		      struct netlink_ext_ack *extack)
310 {
311 	struct wg_device *wg = netdev_priv(dev);
312 	int ret = -ENOMEM;
313 
314 	rcu_assign_pointer(wg->creating_net, src_net);
315 	init_rwsem(&wg->static_identity.lock);
316 	mutex_init(&wg->socket_update_lock);
317 	mutex_init(&wg->device_update_lock);
318 	wg_allowedips_init(&wg->peer_allowedips);
319 	wg_cookie_checker_init(&wg->cookie_checker, wg);
320 	INIT_LIST_HEAD(&wg->peer_list);
321 	wg->device_update_gen = 1;
322 
323 	wg->peer_hashtable = wg_pubkey_hashtable_alloc();
324 	if (!wg->peer_hashtable)
325 		return ret;
326 
327 	wg->index_hashtable = wg_index_hashtable_alloc();
328 	if (!wg->index_hashtable)
329 		goto err_free_peer_hashtable;
330 
331 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
332 	if (!dev->tstats)
333 		goto err_free_index_hashtable;
334 
335 	wg->handshake_receive_wq = alloc_workqueue("wg-kex-%s",
336 			WQ_CPU_INTENSIVE | WQ_FREEZABLE, 0, dev->name);
337 	if (!wg->handshake_receive_wq)
338 		goto err_free_tstats;
339 
340 	wg->handshake_send_wq = alloc_workqueue("wg-kex-%s",
341 			WQ_UNBOUND | WQ_FREEZABLE, 0, dev->name);
342 	if (!wg->handshake_send_wq)
343 		goto err_destroy_handshake_receive;
344 
345 	wg->packet_crypt_wq = alloc_workqueue("wg-crypt-%s",
346 			WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 0, dev->name);
347 	if (!wg->packet_crypt_wq)
348 		goto err_destroy_handshake_send;
349 
350 	ret = wg_packet_queue_init(&wg->encrypt_queue, wg_packet_encrypt_worker,
351 				   MAX_QUEUED_PACKETS);
352 	if (ret < 0)
353 		goto err_destroy_packet_crypt;
354 
355 	ret = wg_packet_queue_init(&wg->decrypt_queue, wg_packet_decrypt_worker,
356 				   MAX_QUEUED_PACKETS);
357 	if (ret < 0)
358 		goto err_free_encrypt_queue;
359 
360 	ret = wg_packet_queue_init(&wg->handshake_queue, wg_packet_handshake_receive_worker,
361 				   MAX_QUEUED_INCOMING_HANDSHAKES);
362 	if (ret < 0)
363 		goto err_free_decrypt_queue;
364 
365 	ret = wg_ratelimiter_init();
366 	if (ret < 0)
367 		goto err_free_handshake_queue;
368 
369 	ret = register_netdevice(dev);
370 	if (ret < 0)
371 		goto err_uninit_ratelimiter;
372 
373 	list_add(&wg->device_list, &device_list);
374 
375 	/* We wait until the end to assign priv_destructor, so that
376 	 * register_netdevice doesn't call it for us if it fails.
377 	 */
378 	dev->priv_destructor = wg_destruct;
379 
380 	pr_debug("%s: Interface created\n", dev->name);
381 	return ret;
382 
383 err_uninit_ratelimiter:
384 	wg_ratelimiter_uninit();
385 err_free_handshake_queue:
386 	wg_packet_queue_free(&wg->handshake_queue, false);
387 err_free_decrypt_queue:
388 	wg_packet_queue_free(&wg->decrypt_queue, false);
389 err_free_encrypt_queue:
390 	wg_packet_queue_free(&wg->encrypt_queue, false);
391 err_destroy_packet_crypt:
392 	destroy_workqueue(wg->packet_crypt_wq);
393 err_destroy_handshake_send:
394 	destroy_workqueue(wg->handshake_send_wq);
395 err_destroy_handshake_receive:
396 	destroy_workqueue(wg->handshake_receive_wq);
397 err_free_tstats:
398 	free_percpu(dev->tstats);
399 err_free_index_hashtable:
400 	kvfree(wg->index_hashtable);
401 err_free_peer_hashtable:
402 	kvfree(wg->peer_hashtable);
403 	return ret;
404 }
405 
406 static struct rtnl_link_ops link_ops __read_mostly = {
407 	.kind			= KBUILD_MODNAME,
408 	.priv_size		= sizeof(struct wg_device),
409 	.setup			= wg_setup,
410 	.newlink		= wg_newlink,
411 };
412 
413 static void wg_netns_pre_exit(struct net *net)
414 {
415 	struct wg_device *wg;
416 	struct wg_peer *peer;
417 
418 	rtnl_lock();
419 	list_for_each_entry(wg, &device_list, device_list) {
420 		if (rcu_access_pointer(wg->creating_net) == net) {
421 			pr_debug("%s: Creating namespace exiting\n", wg->dev->name);
422 			netif_carrier_off(wg->dev);
423 			mutex_lock(&wg->device_update_lock);
424 			rcu_assign_pointer(wg->creating_net, NULL);
425 			wg_socket_reinit(wg, NULL, NULL);
426 			list_for_each_entry(peer, &wg->peer_list, peer_list)
427 				wg_socket_clear_peer_endpoint_src(peer);
428 			mutex_unlock(&wg->device_update_lock);
429 		}
430 	}
431 	rtnl_unlock();
432 }
433 
434 static struct pernet_operations pernet_ops = {
435 	.pre_exit = wg_netns_pre_exit
436 };
437 
438 int __init wg_device_init(void)
439 {
440 	int ret;
441 
442 	ret = register_pm_notifier(&pm_notifier);
443 	if (ret)
444 		return ret;
445 
446 	ret = register_random_vmfork_notifier(&vm_notifier);
447 	if (ret)
448 		goto error_pm;
449 
450 	ret = register_pernet_device(&pernet_ops);
451 	if (ret)
452 		goto error_vm;
453 
454 	ret = rtnl_link_register(&link_ops);
455 	if (ret)
456 		goto error_pernet;
457 
458 	return 0;
459 
460 error_pernet:
461 	unregister_pernet_device(&pernet_ops);
462 error_vm:
463 	unregister_random_vmfork_notifier(&vm_notifier);
464 error_pm:
465 	unregister_pm_notifier(&pm_notifier);
466 	return ret;
467 }
468 
469 void wg_device_uninit(void)
470 {
471 	rtnl_link_unregister(&link_ops);
472 	unregister_pernet_device(&pernet_ops);
473 	unregister_random_vmfork_notifier(&vm_notifier);
474 	unregister_pm_notifier(&pm_notifier);
475 	rcu_barrier();
476 }
477