xref: /openbmc/linux/drivers/net/wireguard/device.c (revision 6e10e219)
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 #ifdef CONFIG_PM_SLEEP
63 static int wg_pm_notification(struct notifier_block *nb, unsigned long action,
64 			      void *data)
65 {
66 	struct wg_device *wg;
67 	struct wg_peer *peer;
68 
69 	/* If the machine is constantly suspending and resuming, as part of
70 	 * its normal operation rather than as a somewhat rare event, then we
71 	 * don't actually want to clear keys.
72 	 */
73 	if (IS_ENABLED(CONFIG_PM_AUTOSLEEP) || IS_ENABLED(CONFIG_ANDROID))
74 		return 0;
75 
76 	if (action != PM_HIBERNATION_PREPARE && action != PM_SUSPEND_PREPARE)
77 		return 0;
78 
79 	rtnl_lock();
80 	list_for_each_entry(wg, &device_list, device_list) {
81 		mutex_lock(&wg->device_update_lock);
82 		list_for_each_entry(peer, &wg->peer_list, peer_list) {
83 			del_timer(&peer->timer_zero_key_material);
84 			wg_noise_handshake_clear(&peer->handshake);
85 			wg_noise_keypairs_clear(&peer->keypairs);
86 		}
87 		mutex_unlock(&wg->device_update_lock);
88 	}
89 	rtnl_unlock();
90 	rcu_barrier();
91 	return 0;
92 }
93 
94 static struct notifier_block pm_notifier = { .notifier_call = wg_pm_notification };
95 #endif
96 
97 static int wg_stop(struct net_device *dev)
98 {
99 	struct wg_device *wg = netdev_priv(dev);
100 	struct wg_peer *peer;
101 	struct sk_buff *skb;
102 
103 	mutex_lock(&wg->device_update_lock);
104 	list_for_each_entry(peer, &wg->peer_list, peer_list) {
105 		wg_packet_purge_staged_packets(peer);
106 		wg_timers_stop(peer);
107 		wg_noise_handshake_clear(&peer->handshake);
108 		wg_noise_keypairs_clear(&peer->keypairs);
109 		wg_noise_reset_last_sent_handshake(&peer->last_sent_handshake);
110 	}
111 	mutex_unlock(&wg->device_update_lock);
112 	while ((skb = ptr_ring_consume(&wg->handshake_queue.ring)) != NULL)
113 		kfree_skb(skb);
114 	atomic_set(&wg->handshake_queue_len, 0);
115 	wg_socket_reinit(wg, NULL, NULL);
116 	return 0;
117 }
118 
119 static netdev_tx_t wg_xmit(struct sk_buff *skb, struct net_device *dev)
120 {
121 	struct wg_device *wg = netdev_priv(dev);
122 	struct sk_buff_head packets;
123 	struct wg_peer *peer;
124 	struct sk_buff *next;
125 	sa_family_t family;
126 	u32 mtu;
127 	int ret;
128 
129 	if (unlikely(!wg_check_packet_protocol(skb))) {
130 		ret = -EPROTONOSUPPORT;
131 		net_dbg_ratelimited("%s: Invalid IP packet\n", dev->name);
132 		goto err;
133 	}
134 
135 	peer = wg_allowedips_lookup_dst(&wg->peer_allowedips, skb);
136 	if (unlikely(!peer)) {
137 		ret = -ENOKEY;
138 		if (skb->protocol == htons(ETH_P_IP))
139 			net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI4\n",
140 					    dev->name, &ip_hdr(skb)->daddr);
141 		else if (skb->protocol == htons(ETH_P_IPV6))
142 			net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI6\n",
143 					    dev->name, &ipv6_hdr(skb)->daddr);
144 		goto err_icmp;
145 	}
146 
147 	family = READ_ONCE(peer->endpoint.addr.sa_family);
148 	if (unlikely(family != AF_INET && family != AF_INET6)) {
149 		ret = -EDESTADDRREQ;
150 		net_dbg_ratelimited("%s: No valid endpoint has been configured or discovered for peer %llu\n",
151 				    dev->name, peer->internal_id);
152 		goto err_peer;
153 	}
154 
155 	mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
156 
157 	__skb_queue_head_init(&packets);
158 	if (!skb_is_gso(skb)) {
159 		skb_mark_not_on_list(skb);
160 	} else {
161 		struct sk_buff *segs = skb_gso_segment(skb, 0);
162 
163 		if (IS_ERR(segs)) {
164 			ret = PTR_ERR(segs);
165 			goto err_peer;
166 		}
167 		dev_kfree_skb(skb);
168 		skb = segs;
169 	}
170 
171 	skb_list_walk_safe(skb, skb, next) {
172 		skb_mark_not_on_list(skb);
173 
174 		skb = skb_share_check(skb, GFP_ATOMIC);
175 		if (unlikely(!skb))
176 			continue;
177 
178 		/* We only need to keep the original dst around for icmp,
179 		 * so at this point we're in a position to drop it.
180 		 */
181 		skb_dst_drop(skb);
182 
183 		PACKET_CB(skb)->mtu = mtu;
184 
185 		__skb_queue_tail(&packets, skb);
186 	}
187 
188 	spin_lock_bh(&peer->staged_packet_queue.lock);
189 	/* If the queue is getting too big, we start removing the oldest packets
190 	 * until it's small again. We do this before adding the new packet, so
191 	 * we don't remove GSO segments that are in excess.
192 	 */
193 	while (skb_queue_len(&peer->staged_packet_queue) > MAX_STAGED_PACKETS) {
194 		dev_kfree_skb(__skb_dequeue(&peer->staged_packet_queue));
195 		++dev->stats.tx_dropped;
196 	}
197 	skb_queue_splice_tail(&packets, &peer->staged_packet_queue);
198 	spin_unlock_bh(&peer->staged_packet_queue.lock);
199 
200 	wg_packet_send_staged_packets(peer);
201 
202 	wg_peer_put(peer);
203 	return NETDEV_TX_OK;
204 
205 err_peer:
206 	wg_peer_put(peer);
207 err_icmp:
208 	if (skb->protocol == htons(ETH_P_IP))
209 		icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
210 	else if (skb->protocol == htons(ETH_P_IPV6))
211 		icmpv6_ndo_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
212 err:
213 	++dev->stats.tx_errors;
214 	kfree_skb(skb);
215 	return ret;
216 }
217 
218 static const struct net_device_ops netdev_ops = {
219 	.ndo_open		= wg_open,
220 	.ndo_stop		= wg_stop,
221 	.ndo_start_xmit		= wg_xmit,
222 	.ndo_get_stats64	= dev_get_tstats64
223 };
224 
225 static void wg_destruct(struct net_device *dev)
226 {
227 	struct wg_device *wg = netdev_priv(dev);
228 
229 	rtnl_lock();
230 	list_del(&wg->device_list);
231 	rtnl_unlock();
232 	mutex_lock(&wg->device_update_lock);
233 	rcu_assign_pointer(wg->creating_net, NULL);
234 	wg->incoming_port = 0;
235 	wg_socket_reinit(wg, NULL, NULL);
236 	/* The final references are cleared in the below calls to destroy_workqueue. */
237 	wg_peer_remove_all(wg);
238 	destroy_workqueue(wg->handshake_receive_wq);
239 	destroy_workqueue(wg->handshake_send_wq);
240 	destroy_workqueue(wg->packet_crypt_wq);
241 	wg_packet_queue_free(&wg->handshake_queue, true);
242 	wg_packet_queue_free(&wg->decrypt_queue, false);
243 	wg_packet_queue_free(&wg->encrypt_queue, false);
244 	rcu_barrier(); /* Wait for all the peers to be actually freed. */
245 	wg_ratelimiter_uninit();
246 	memzero_explicit(&wg->static_identity, sizeof(wg->static_identity));
247 	free_percpu(dev->tstats);
248 	kvfree(wg->index_hashtable);
249 	kvfree(wg->peer_hashtable);
250 	mutex_unlock(&wg->device_update_lock);
251 
252 	pr_debug("%s: Interface destroyed\n", dev->name);
253 	free_netdev(dev);
254 }
255 
256 static const struct device_type device_type = { .name = KBUILD_MODNAME };
257 
258 static void wg_setup(struct net_device *dev)
259 {
260 	struct wg_device *wg = netdev_priv(dev);
261 	enum { WG_NETDEV_FEATURES = NETIF_F_HW_CSUM | NETIF_F_RXCSUM |
262 				    NETIF_F_SG | NETIF_F_GSO |
263 				    NETIF_F_GSO_SOFTWARE | NETIF_F_HIGHDMA };
264 	const int overhead = MESSAGE_MINIMUM_LENGTH + sizeof(struct udphdr) +
265 			     max(sizeof(struct ipv6hdr), sizeof(struct iphdr));
266 
267 	dev->netdev_ops = &netdev_ops;
268 	dev->header_ops = &ip_tunnel_header_ops;
269 	dev->hard_header_len = 0;
270 	dev->addr_len = 0;
271 	dev->needed_headroom = DATA_PACKET_HEAD_ROOM;
272 	dev->needed_tailroom = noise_encrypted_len(MESSAGE_PADDING_MULTIPLE);
273 	dev->type = ARPHRD_NONE;
274 	dev->flags = IFF_POINTOPOINT | IFF_NOARP;
275 	dev->priv_flags |= IFF_NO_QUEUE;
276 	dev->features |= NETIF_F_LLTX;
277 	dev->features |= WG_NETDEV_FEATURES;
278 	dev->hw_features |= WG_NETDEV_FEATURES;
279 	dev->hw_enc_features |= WG_NETDEV_FEATURES;
280 	dev->mtu = ETH_DATA_LEN - overhead;
281 	dev->max_mtu = round_down(INT_MAX, MESSAGE_PADDING_MULTIPLE) - overhead;
282 
283 	SET_NETDEV_DEVTYPE(dev, &device_type);
284 
285 	/* We need to keep the dst around in case of icmp replies. */
286 	netif_keep_dst(dev);
287 
288 	memset(wg, 0, sizeof(*wg));
289 	wg->dev = dev;
290 }
291 
292 static int wg_newlink(struct net *src_net, struct net_device *dev,
293 		      struct nlattr *tb[], struct nlattr *data[],
294 		      struct netlink_ext_ack *extack)
295 {
296 	struct wg_device *wg = netdev_priv(dev);
297 	int ret = -ENOMEM;
298 
299 	rcu_assign_pointer(wg->creating_net, src_net);
300 	init_rwsem(&wg->static_identity.lock);
301 	mutex_init(&wg->socket_update_lock);
302 	mutex_init(&wg->device_update_lock);
303 	wg_allowedips_init(&wg->peer_allowedips);
304 	wg_cookie_checker_init(&wg->cookie_checker, wg);
305 	INIT_LIST_HEAD(&wg->peer_list);
306 	wg->device_update_gen = 1;
307 
308 	wg->peer_hashtable = wg_pubkey_hashtable_alloc();
309 	if (!wg->peer_hashtable)
310 		return ret;
311 
312 	wg->index_hashtable = wg_index_hashtable_alloc();
313 	if (!wg->index_hashtable)
314 		goto err_free_peer_hashtable;
315 
316 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
317 	if (!dev->tstats)
318 		goto err_free_index_hashtable;
319 
320 	wg->handshake_receive_wq = alloc_workqueue("wg-kex-%s",
321 			WQ_CPU_INTENSIVE | WQ_FREEZABLE, 0, dev->name);
322 	if (!wg->handshake_receive_wq)
323 		goto err_free_tstats;
324 
325 	wg->handshake_send_wq = alloc_workqueue("wg-kex-%s",
326 			WQ_UNBOUND | WQ_FREEZABLE, 0, dev->name);
327 	if (!wg->handshake_send_wq)
328 		goto err_destroy_handshake_receive;
329 
330 	wg->packet_crypt_wq = alloc_workqueue("wg-crypt-%s",
331 			WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 0, dev->name);
332 	if (!wg->packet_crypt_wq)
333 		goto err_destroy_handshake_send;
334 
335 	ret = wg_packet_queue_init(&wg->encrypt_queue, wg_packet_encrypt_worker,
336 				   MAX_QUEUED_PACKETS);
337 	if (ret < 0)
338 		goto err_destroy_packet_crypt;
339 
340 	ret = wg_packet_queue_init(&wg->decrypt_queue, wg_packet_decrypt_worker,
341 				   MAX_QUEUED_PACKETS);
342 	if (ret < 0)
343 		goto err_free_encrypt_queue;
344 
345 	ret = wg_packet_queue_init(&wg->handshake_queue, wg_packet_handshake_receive_worker,
346 				   MAX_QUEUED_INCOMING_HANDSHAKES);
347 	if (ret < 0)
348 		goto err_free_decrypt_queue;
349 
350 	ret = wg_ratelimiter_init();
351 	if (ret < 0)
352 		goto err_free_handshake_queue;
353 
354 	ret = register_netdevice(dev);
355 	if (ret < 0)
356 		goto err_uninit_ratelimiter;
357 
358 	list_add(&wg->device_list, &device_list);
359 
360 	/* We wait until the end to assign priv_destructor, so that
361 	 * register_netdevice doesn't call it for us if it fails.
362 	 */
363 	dev->priv_destructor = wg_destruct;
364 
365 	pr_debug("%s: Interface created\n", dev->name);
366 	return ret;
367 
368 err_uninit_ratelimiter:
369 	wg_ratelimiter_uninit();
370 err_free_handshake_queue:
371 	wg_packet_queue_free(&wg->handshake_queue, false);
372 err_free_decrypt_queue:
373 	wg_packet_queue_free(&wg->decrypt_queue, false);
374 err_free_encrypt_queue:
375 	wg_packet_queue_free(&wg->encrypt_queue, false);
376 err_destroy_packet_crypt:
377 	destroy_workqueue(wg->packet_crypt_wq);
378 err_destroy_handshake_send:
379 	destroy_workqueue(wg->handshake_send_wq);
380 err_destroy_handshake_receive:
381 	destroy_workqueue(wg->handshake_receive_wq);
382 err_free_tstats:
383 	free_percpu(dev->tstats);
384 err_free_index_hashtable:
385 	kvfree(wg->index_hashtable);
386 err_free_peer_hashtable:
387 	kvfree(wg->peer_hashtable);
388 	return ret;
389 }
390 
391 static struct rtnl_link_ops link_ops __read_mostly = {
392 	.kind			= KBUILD_MODNAME,
393 	.priv_size		= sizeof(struct wg_device),
394 	.setup			= wg_setup,
395 	.newlink		= wg_newlink,
396 };
397 
398 static void wg_netns_pre_exit(struct net *net)
399 {
400 	struct wg_device *wg;
401 	struct wg_peer *peer;
402 
403 	rtnl_lock();
404 	list_for_each_entry(wg, &device_list, device_list) {
405 		if (rcu_access_pointer(wg->creating_net) == net) {
406 			pr_debug("%s: Creating namespace exiting\n", wg->dev->name);
407 			netif_carrier_off(wg->dev);
408 			mutex_lock(&wg->device_update_lock);
409 			rcu_assign_pointer(wg->creating_net, NULL);
410 			wg_socket_reinit(wg, NULL, NULL);
411 			list_for_each_entry(peer, &wg->peer_list, peer_list)
412 				wg_socket_clear_peer_endpoint_src(peer);
413 			mutex_unlock(&wg->device_update_lock);
414 		}
415 	}
416 	rtnl_unlock();
417 }
418 
419 static struct pernet_operations pernet_ops = {
420 	.pre_exit = wg_netns_pre_exit
421 };
422 
423 int __init wg_device_init(void)
424 {
425 	int ret;
426 
427 #ifdef CONFIG_PM_SLEEP
428 	ret = register_pm_notifier(&pm_notifier);
429 	if (ret)
430 		return ret;
431 #endif
432 
433 	ret = register_pernet_device(&pernet_ops);
434 	if (ret)
435 		goto error_pm;
436 
437 	ret = rtnl_link_register(&link_ops);
438 	if (ret)
439 		goto error_pernet;
440 
441 	return 0;
442 
443 error_pernet:
444 	unregister_pernet_device(&pernet_ops);
445 error_pm:
446 #ifdef CONFIG_PM_SLEEP
447 	unregister_pm_notifier(&pm_notifier);
448 #endif
449 	return ret;
450 }
451 
452 void wg_device_uninit(void)
453 {
454 	rtnl_link_unregister(&link_ops);
455 	unregister_pernet_device(&pernet_ops);
456 #ifdef CONFIG_PM_SLEEP
457 	unregister_pm_notifier(&pm_notifier);
458 #endif
459 	rcu_barrier();
460 }
461