xref: /openbmc/linux/drivers/net/wireguard/device.c (revision 86e281fc)
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/dst_metadata.h>
23 #include <net/gso.h>
24 #include <net/icmp.h>
25 #include <net/rtnetlink.h>
26 #include <net/ip_tunnels.h>
27 #include <net/addrconf.h>
28 
29 static LIST_HEAD(device_list);
30 
31 static int wg_open(struct net_device *dev)
32 {
33 	struct in_device *dev_v4 = __in_dev_get_rtnl(dev);
34 	struct inet6_dev *dev_v6 = __in6_dev_get(dev);
35 	struct wg_device *wg = netdev_priv(dev);
36 	struct wg_peer *peer;
37 	int ret;
38 
39 	if (dev_v4) {
40 		/* At some point we might put this check near the ip_rt_send_
41 		 * redirect call of ip_forward in net/ipv4/ip_forward.c, similar
42 		 * to the current secpath check.
43 		 */
44 		IN_DEV_CONF_SET(dev_v4, SEND_REDIRECTS, false);
45 		IPV4_DEVCONF_ALL(dev_net(dev), SEND_REDIRECTS) = false;
46 	}
47 	if (dev_v6)
48 		dev_v6->cnf.addr_gen_mode = IN6_ADDR_GEN_MODE_NONE;
49 
50 	mutex_lock(&wg->device_update_lock);
51 	ret = wg_socket_init(wg, wg->incoming_port);
52 	if (ret < 0)
53 		goto out;
54 	list_for_each_entry(peer, &wg->peer_list, peer_list) {
55 		wg_packet_send_staged_packets(peer);
56 		if (peer->persistent_keepalive_interval)
57 			wg_packet_send_keepalive(peer);
58 	}
59 out:
60 	mutex_unlock(&wg->device_update_lock);
61 	return ret;
62 }
63 
64 static int wg_pm_notification(struct notifier_block *nb, unsigned long action, 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) ||
74 	    IS_ENABLED(CONFIG_PM_USERSPACE_AUTOSLEEP))
75 		return 0;
76 
77 	if (action != PM_HIBERNATION_PREPARE && action != PM_SUSPEND_PREPARE)
78 		return 0;
79 
80 	rtnl_lock();
81 	list_for_each_entry(wg, &device_list, device_list) {
82 		mutex_lock(&wg->device_update_lock);
83 		list_for_each_entry(peer, &wg->peer_list, peer_list) {
84 			del_timer(&peer->timer_zero_key_material);
85 			wg_noise_handshake_clear(&peer->handshake);
86 			wg_noise_keypairs_clear(&peer->keypairs);
87 		}
88 		mutex_unlock(&wg->device_update_lock);
89 	}
90 	rtnl_unlock();
91 	rcu_barrier();
92 	return 0;
93 }
94 
95 static struct notifier_block pm_notifier = { .notifier_call = wg_pm_notification };
96 
97 static int wg_vm_notification(struct notifier_block *nb, unsigned long action, void *data)
98 {
99 	struct wg_device *wg;
100 	struct wg_peer *peer;
101 
102 	rtnl_lock();
103 	list_for_each_entry(wg, &device_list, device_list) {
104 		mutex_lock(&wg->device_update_lock);
105 		list_for_each_entry(peer, &wg->peer_list, peer_list)
106 			wg_noise_expire_current_peer_keypairs(peer);
107 		mutex_unlock(&wg->device_update_lock);
108 	}
109 	rtnl_unlock();
110 	return 0;
111 }
112 
113 static struct notifier_block vm_notifier = { .notifier_call = wg_vm_notification };
114 
115 static int wg_stop(struct net_device *dev)
116 {
117 	struct wg_device *wg = netdev_priv(dev);
118 	struct wg_peer *peer;
119 	struct sk_buff *skb;
120 
121 	mutex_lock(&wg->device_update_lock);
122 	list_for_each_entry(peer, &wg->peer_list, peer_list) {
123 		wg_packet_purge_staged_packets(peer);
124 		wg_timers_stop(peer);
125 		wg_noise_handshake_clear(&peer->handshake);
126 		wg_noise_keypairs_clear(&peer->keypairs);
127 		wg_noise_reset_last_sent_handshake(&peer->last_sent_handshake);
128 	}
129 	mutex_unlock(&wg->device_update_lock);
130 	while ((skb = ptr_ring_consume(&wg->handshake_queue.ring)) != NULL)
131 		kfree_skb(skb);
132 	atomic_set(&wg->handshake_queue_len, 0);
133 	wg_socket_reinit(wg, NULL, NULL);
134 	return 0;
135 }
136 
137 static netdev_tx_t wg_xmit(struct sk_buff *skb, struct net_device *dev)
138 {
139 	struct wg_device *wg = netdev_priv(dev);
140 	struct sk_buff_head packets;
141 	struct wg_peer *peer;
142 	struct sk_buff *next;
143 	sa_family_t family;
144 	u32 mtu;
145 	int ret;
146 
147 	if (unlikely(!wg_check_packet_protocol(skb))) {
148 		ret = -EPROTONOSUPPORT;
149 		net_dbg_ratelimited("%s: Invalid IP packet\n", dev->name);
150 		goto err;
151 	}
152 
153 	peer = wg_allowedips_lookup_dst(&wg->peer_allowedips, skb);
154 	if (unlikely(!peer)) {
155 		ret = -ENOKEY;
156 		if (skb->protocol == htons(ETH_P_IP))
157 			net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI4\n",
158 					    dev->name, &ip_hdr(skb)->daddr);
159 		else if (skb->protocol == htons(ETH_P_IPV6))
160 			net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI6\n",
161 					    dev->name, &ipv6_hdr(skb)->daddr);
162 		goto err_icmp;
163 	}
164 
165 	family = READ_ONCE(peer->endpoint.addr.sa_family);
166 	if (unlikely(family != AF_INET && family != AF_INET6)) {
167 		ret = -EDESTADDRREQ;
168 		net_dbg_ratelimited("%s: No valid endpoint has been configured or discovered for peer %llu\n",
169 				    dev->name, peer->internal_id);
170 		goto err_peer;
171 	}
172 
173 	mtu = skb_valid_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
174 
175 	__skb_queue_head_init(&packets);
176 	if (!skb_is_gso(skb)) {
177 		skb_mark_not_on_list(skb);
178 	} else {
179 		struct sk_buff *segs = skb_gso_segment(skb, 0);
180 
181 		if (IS_ERR(segs)) {
182 			ret = PTR_ERR(segs);
183 			goto err_peer;
184 		}
185 		dev_kfree_skb(skb);
186 		skb = segs;
187 	}
188 
189 	skb_list_walk_safe(skb, skb, next) {
190 		skb_mark_not_on_list(skb);
191 
192 		skb = skb_share_check(skb, GFP_ATOMIC);
193 		if (unlikely(!skb))
194 			continue;
195 
196 		/* We only need to keep the original dst around for icmp,
197 		 * so at this point we're in a position to drop it.
198 		 */
199 		skb_dst_drop(skb);
200 
201 		PACKET_CB(skb)->mtu = mtu;
202 
203 		__skb_queue_tail(&packets, skb);
204 	}
205 
206 	spin_lock_bh(&peer->staged_packet_queue.lock);
207 	/* If the queue is getting too big, we start removing the oldest packets
208 	 * until it's small again. We do this before adding the new packet, so
209 	 * we don't remove GSO segments that are in excess.
210 	 */
211 	while (skb_queue_len(&peer->staged_packet_queue) > MAX_STAGED_PACKETS) {
212 		dev_kfree_skb(__skb_dequeue(&peer->staged_packet_queue));
213 		++dev->stats.tx_dropped;
214 	}
215 	skb_queue_splice_tail(&packets, &peer->staged_packet_queue);
216 	spin_unlock_bh(&peer->staged_packet_queue.lock);
217 
218 	wg_packet_send_staged_packets(peer);
219 
220 	wg_peer_put(peer);
221 	return NETDEV_TX_OK;
222 
223 err_peer:
224 	wg_peer_put(peer);
225 err_icmp:
226 	if (skb->protocol == htons(ETH_P_IP))
227 		icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
228 	else if (skb->protocol == htons(ETH_P_IPV6))
229 		icmpv6_ndo_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
230 err:
231 	++dev->stats.tx_errors;
232 	kfree_skb(skb);
233 	return ret;
234 }
235 
236 static const struct net_device_ops netdev_ops = {
237 	.ndo_open		= wg_open,
238 	.ndo_stop		= wg_stop,
239 	.ndo_start_xmit		= wg_xmit,
240 	.ndo_get_stats64	= dev_get_tstats64
241 };
242 
243 static void wg_destruct(struct net_device *dev)
244 {
245 	struct wg_device *wg = netdev_priv(dev);
246 
247 	rtnl_lock();
248 	list_del(&wg->device_list);
249 	rtnl_unlock();
250 	mutex_lock(&wg->device_update_lock);
251 	rcu_assign_pointer(wg->creating_net, NULL);
252 	wg->incoming_port = 0;
253 	wg_socket_reinit(wg, NULL, NULL);
254 	/* The final references are cleared in the below calls to destroy_workqueue. */
255 	wg_peer_remove_all(wg);
256 	destroy_workqueue(wg->handshake_receive_wq);
257 	destroy_workqueue(wg->handshake_send_wq);
258 	destroy_workqueue(wg->packet_crypt_wq);
259 	wg_packet_queue_free(&wg->handshake_queue, true);
260 	wg_packet_queue_free(&wg->decrypt_queue, false);
261 	wg_packet_queue_free(&wg->encrypt_queue, false);
262 	rcu_barrier(); /* Wait for all the peers to be actually freed. */
263 	wg_ratelimiter_uninit();
264 	memzero_explicit(&wg->static_identity, sizeof(wg->static_identity));
265 	free_percpu(dev->tstats);
266 	kvfree(wg->index_hashtable);
267 	kvfree(wg->peer_hashtable);
268 	mutex_unlock(&wg->device_update_lock);
269 
270 	pr_debug("%s: Interface destroyed\n", dev->name);
271 	free_netdev(dev);
272 }
273 
274 static const struct device_type device_type = { .name = KBUILD_MODNAME };
275 
276 static void wg_setup(struct net_device *dev)
277 {
278 	struct wg_device *wg = netdev_priv(dev);
279 	enum { WG_NETDEV_FEATURES = NETIF_F_HW_CSUM | NETIF_F_RXCSUM |
280 				    NETIF_F_SG | NETIF_F_GSO |
281 				    NETIF_F_GSO_SOFTWARE | NETIF_F_HIGHDMA };
282 	const int overhead = MESSAGE_MINIMUM_LENGTH + sizeof(struct udphdr) +
283 			     max(sizeof(struct ipv6hdr), sizeof(struct iphdr));
284 
285 	dev->netdev_ops = &netdev_ops;
286 	dev->header_ops = &ip_tunnel_header_ops;
287 	dev->hard_header_len = 0;
288 	dev->addr_len = 0;
289 	dev->needed_headroom = DATA_PACKET_HEAD_ROOM;
290 	dev->needed_tailroom = noise_encrypted_len(MESSAGE_PADDING_MULTIPLE);
291 	dev->type = ARPHRD_NONE;
292 	dev->flags = IFF_POINTOPOINT | IFF_NOARP;
293 	dev->priv_flags |= IFF_NO_QUEUE;
294 	dev->features |= NETIF_F_LLTX;
295 	dev->features |= WG_NETDEV_FEATURES;
296 	dev->hw_features |= WG_NETDEV_FEATURES;
297 	dev->hw_enc_features |= WG_NETDEV_FEATURES;
298 	dev->mtu = ETH_DATA_LEN - overhead;
299 	dev->max_mtu = round_down(INT_MAX, MESSAGE_PADDING_MULTIPLE) - overhead;
300 
301 	SET_NETDEV_DEVTYPE(dev, &device_type);
302 
303 	/* We need to keep the dst around in case of icmp replies. */
304 	netif_keep_dst(dev);
305 
306 	memset(wg, 0, sizeof(*wg));
307 	wg->dev = dev;
308 }
309 
310 static int wg_newlink(struct net *src_net, struct net_device *dev,
311 		      struct nlattr *tb[], struct nlattr *data[],
312 		      struct netlink_ext_ack *extack)
313 {
314 	struct wg_device *wg = netdev_priv(dev);
315 	int ret = -ENOMEM;
316 
317 	rcu_assign_pointer(wg->creating_net, src_net);
318 	init_rwsem(&wg->static_identity.lock);
319 	mutex_init(&wg->socket_update_lock);
320 	mutex_init(&wg->device_update_lock);
321 	wg_allowedips_init(&wg->peer_allowedips);
322 	wg_cookie_checker_init(&wg->cookie_checker, wg);
323 	INIT_LIST_HEAD(&wg->peer_list);
324 	wg->device_update_gen = 1;
325 
326 	wg->peer_hashtable = wg_pubkey_hashtable_alloc();
327 	if (!wg->peer_hashtable)
328 		return ret;
329 
330 	wg->index_hashtable = wg_index_hashtable_alloc();
331 	if (!wg->index_hashtable)
332 		goto err_free_peer_hashtable;
333 
334 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
335 	if (!dev->tstats)
336 		goto err_free_index_hashtable;
337 
338 	wg->handshake_receive_wq = alloc_workqueue("wg-kex-%s",
339 			WQ_CPU_INTENSIVE | WQ_FREEZABLE, 0, dev->name);
340 	if (!wg->handshake_receive_wq)
341 		goto err_free_tstats;
342 
343 	wg->handshake_send_wq = alloc_workqueue("wg-kex-%s",
344 			WQ_UNBOUND | WQ_FREEZABLE, 0, dev->name);
345 	if (!wg->handshake_send_wq)
346 		goto err_destroy_handshake_receive;
347 
348 	wg->packet_crypt_wq = alloc_workqueue("wg-crypt-%s",
349 			WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 0, dev->name);
350 	if (!wg->packet_crypt_wq)
351 		goto err_destroy_handshake_send;
352 
353 	ret = wg_packet_queue_init(&wg->encrypt_queue, wg_packet_encrypt_worker,
354 				   MAX_QUEUED_PACKETS);
355 	if (ret < 0)
356 		goto err_destroy_packet_crypt;
357 
358 	ret = wg_packet_queue_init(&wg->decrypt_queue, wg_packet_decrypt_worker,
359 				   MAX_QUEUED_PACKETS);
360 	if (ret < 0)
361 		goto err_free_encrypt_queue;
362 
363 	ret = wg_packet_queue_init(&wg->handshake_queue, wg_packet_handshake_receive_worker,
364 				   MAX_QUEUED_INCOMING_HANDSHAKES);
365 	if (ret < 0)
366 		goto err_free_decrypt_queue;
367 
368 	ret = wg_ratelimiter_init();
369 	if (ret < 0)
370 		goto err_free_handshake_queue;
371 
372 	ret = register_netdevice(dev);
373 	if (ret < 0)
374 		goto err_uninit_ratelimiter;
375 
376 	list_add(&wg->device_list, &device_list);
377 
378 	/* We wait until the end to assign priv_destructor, so that
379 	 * register_netdevice doesn't call it for us if it fails.
380 	 */
381 	dev->priv_destructor = wg_destruct;
382 
383 	pr_debug("%s: Interface created\n", dev->name);
384 	return ret;
385 
386 err_uninit_ratelimiter:
387 	wg_ratelimiter_uninit();
388 err_free_handshake_queue:
389 	wg_packet_queue_free(&wg->handshake_queue, false);
390 err_free_decrypt_queue:
391 	wg_packet_queue_free(&wg->decrypt_queue, false);
392 err_free_encrypt_queue:
393 	wg_packet_queue_free(&wg->encrypt_queue, false);
394 err_destroy_packet_crypt:
395 	destroy_workqueue(wg->packet_crypt_wq);
396 err_destroy_handshake_send:
397 	destroy_workqueue(wg->handshake_send_wq);
398 err_destroy_handshake_receive:
399 	destroy_workqueue(wg->handshake_receive_wq);
400 err_free_tstats:
401 	free_percpu(dev->tstats);
402 err_free_index_hashtable:
403 	kvfree(wg->index_hashtable);
404 err_free_peer_hashtable:
405 	kvfree(wg->peer_hashtable);
406 	return ret;
407 }
408 
409 static struct rtnl_link_ops link_ops __read_mostly = {
410 	.kind			= KBUILD_MODNAME,
411 	.priv_size		= sizeof(struct wg_device),
412 	.setup			= wg_setup,
413 	.newlink		= wg_newlink,
414 };
415 
416 static void wg_netns_pre_exit(struct net *net)
417 {
418 	struct wg_device *wg;
419 	struct wg_peer *peer;
420 
421 	rtnl_lock();
422 	list_for_each_entry(wg, &device_list, device_list) {
423 		if (rcu_access_pointer(wg->creating_net) == net) {
424 			pr_debug("%s: Creating namespace exiting\n", wg->dev->name);
425 			netif_carrier_off(wg->dev);
426 			mutex_lock(&wg->device_update_lock);
427 			rcu_assign_pointer(wg->creating_net, NULL);
428 			wg_socket_reinit(wg, NULL, NULL);
429 			list_for_each_entry(peer, &wg->peer_list, peer_list)
430 				wg_socket_clear_peer_endpoint_src(peer);
431 			mutex_unlock(&wg->device_update_lock);
432 		}
433 	}
434 	rtnl_unlock();
435 }
436 
437 static struct pernet_operations pernet_ops = {
438 	.pre_exit = wg_netns_pre_exit
439 };
440 
441 int __init wg_device_init(void)
442 {
443 	int ret;
444 
445 	ret = register_pm_notifier(&pm_notifier);
446 	if (ret)
447 		return ret;
448 
449 	ret = register_random_vmfork_notifier(&vm_notifier);
450 	if (ret)
451 		goto error_pm;
452 
453 	ret = register_pernet_device(&pernet_ops);
454 	if (ret)
455 		goto error_vm;
456 
457 	ret = rtnl_link_register(&link_ops);
458 	if (ret)
459 		goto error_pernet;
460 
461 	return 0;
462 
463 error_pernet:
464 	unregister_pernet_device(&pernet_ops);
465 error_vm:
466 	unregister_random_vmfork_notifier(&vm_notifier);
467 error_pm:
468 	unregister_pm_notifier(&pm_notifier);
469 	return ret;
470 }
471 
472 void wg_device_uninit(void)
473 {
474 	rtnl_link_unregister(&link_ops);
475 	unregister_pernet_device(&pernet_ops);
476 	unregister_random_vmfork_notifier(&vm_notifier);
477 	unregister_pm_notifier(&pm_notifier);
478 	rcu_barrier();
479 }
480