xref: /openbmc/linux/drivers/net/wireguard/device.c (revision 1edd0337)
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/icmp.h>
24 #include <net/rtnetlink.h>
25 #include <net/ip_tunnels.h>
26 #include <net/addrconf.h>
27 
28 static LIST_HEAD(device_list);
29 
30 static int wg_open(struct net_device *dev)
31 {
32 	struct in_device *dev_v4 = __in_dev_get_rtnl(dev);
33 	struct inet6_dev *dev_v6 = __in6_dev_get(dev);
34 	struct wg_device *wg = netdev_priv(dev);
35 	struct wg_peer *peer;
36 	int ret;
37 
38 	if (dev_v4) {
39 		/* At some point we might put this check near the ip_rt_send_
40 		 * redirect call of ip_forward in net/ipv4/ip_forward.c, similar
41 		 * to the current secpath check.
42 		 */
43 		IN_DEV_CONF_SET(dev_v4, SEND_REDIRECTS, false);
44 		IPV4_DEVCONF_ALL(dev_net(dev), SEND_REDIRECTS) = false;
45 	}
46 	if (dev_v6)
47 		dev_v6->cnf.addr_gen_mode = IN6_ADDR_GEN_MODE_NONE;
48 
49 	mutex_lock(&wg->device_update_lock);
50 	ret = wg_socket_init(wg, wg->incoming_port);
51 	if (ret < 0)
52 		goto out;
53 	list_for_each_entry(peer, &wg->peer_list, peer_list) {
54 		wg_packet_send_staged_packets(peer);
55 		if (peer->persistent_keepalive_interval)
56 			wg_packet_send_keepalive(peer);
57 	}
58 out:
59 	mutex_unlock(&wg->device_update_lock);
60 	return ret;
61 }
62 
63 static int wg_pm_notification(struct notifier_block *nb, unsigned long action, void *data)
64 {
65 	struct wg_device *wg;
66 	struct wg_peer *peer;
67 
68 	/* If the machine is constantly suspending and resuming, as part of
69 	 * its normal operation rather than as a somewhat rare event, then we
70 	 * don't actually want to clear keys.
71 	 */
72 	if (IS_ENABLED(CONFIG_PM_AUTOSLEEP) ||
73 	    IS_ENABLED(CONFIG_PM_USERSPACE_AUTOSLEEP))
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 
96 static int wg_vm_notification(struct notifier_block *nb, unsigned long action, void *data)
97 {
98 	struct wg_device *wg;
99 	struct wg_peer *peer;
100 
101 	rtnl_lock();
102 	list_for_each_entry(wg, &device_list, device_list) {
103 		mutex_lock(&wg->device_update_lock);
104 		list_for_each_entry(peer, &wg->peer_list, peer_list)
105 			wg_noise_expire_current_peer_keypairs(peer);
106 		mutex_unlock(&wg->device_update_lock);
107 	}
108 	rtnl_unlock();
109 	return 0;
110 }
111 
112 static struct notifier_block vm_notifier = { .notifier_call = wg_vm_notification };
113 
114 static int wg_stop(struct net_device *dev)
115 {
116 	struct wg_device *wg = netdev_priv(dev);
117 	struct wg_peer *peer;
118 	struct sk_buff *skb;
119 
120 	mutex_lock(&wg->device_update_lock);
121 	list_for_each_entry(peer, &wg->peer_list, peer_list) {
122 		wg_packet_purge_staged_packets(peer);
123 		wg_timers_stop(peer);
124 		wg_noise_handshake_clear(&peer->handshake);
125 		wg_noise_keypairs_clear(&peer->keypairs);
126 		wg_noise_reset_last_sent_handshake(&peer->last_sent_handshake);
127 	}
128 	mutex_unlock(&wg->device_update_lock);
129 	while ((skb = ptr_ring_consume(&wg->handshake_queue.ring)) != NULL)
130 		kfree_skb(skb);
131 	atomic_set(&wg->handshake_queue_len, 0);
132 	wg_socket_reinit(wg, NULL, NULL);
133 	return 0;
134 }
135 
136 static netdev_tx_t wg_xmit(struct sk_buff *skb, struct net_device *dev)
137 {
138 	struct wg_device *wg = netdev_priv(dev);
139 	struct sk_buff_head packets;
140 	struct wg_peer *peer;
141 	struct sk_buff *next;
142 	sa_family_t family;
143 	u32 mtu;
144 	int ret;
145 
146 	if (unlikely(!wg_check_packet_protocol(skb))) {
147 		ret = -EPROTONOSUPPORT;
148 		net_dbg_ratelimited("%s: Invalid IP packet\n", dev->name);
149 		goto err;
150 	}
151 
152 	peer = wg_allowedips_lookup_dst(&wg->peer_allowedips, skb);
153 	if (unlikely(!peer)) {
154 		ret = -ENOKEY;
155 		if (skb->protocol == htons(ETH_P_IP))
156 			net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI4\n",
157 					    dev->name, &ip_hdr(skb)->daddr);
158 		else if (skb->protocol == htons(ETH_P_IPV6))
159 			net_dbg_ratelimited("%s: No peer has allowed IPs matching %pI6\n",
160 					    dev->name, &ipv6_hdr(skb)->daddr);
161 		goto err_icmp;
162 	}
163 
164 	family = READ_ONCE(peer->endpoint.addr.sa_family);
165 	if (unlikely(family != AF_INET && family != AF_INET6)) {
166 		ret = -EDESTADDRREQ;
167 		net_dbg_ratelimited("%s: No valid endpoint has been configured or discovered for peer %llu\n",
168 				    dev->name, peer->internal_id);
169 		goto err_peer;
170 	}
171 
172 	mtu = skb_valid_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
173 
174 	__skb_queue_head_init(&packets);
175 	if (!skb_is_gso(skb)) {
176 		skb_mark_not_on_list(skb);
177 	} else {
178 		struct sk_buff *segs = skb_gso_segment(skb, 0);
179 
180 		if (IS_ERR(segs)) {
181 			ret = PTR_ERR(segs);
182 			goto err_peer;
183 		}
184 		dev_kfree_skb(skb);
185 		skb = segs;
186 	}
187 
188 	skb_list_walk_safe(skb, skb, next) {
189 		skb_mark_not_on_list(skb);
190 
191 		skb = skb_share_check(skb, GFP_ATOMIC);
192 		if (unlikely(!skb))
193 			continue;
194 
195 		/* We only need to keep the original dst around for icmp,
196 		 * so at this point we're in a position to drop it.
197 		 */
198 		skb_dst_drop(skb);
199 
200 		PACKET_CB(skb)->mtu = mtu;
201 
202 		__skb_queue_tail(&packets, skb);
203 	}
204 
205 	spin_lock_bh(&peer->staged_packet_queue.lock);
206 	/* If the queue is getting too big, we start removing the oldest packets
207 	 * until it's small again. We do this before adding the new packet, so
208 	 * we don't remove GSO segments that are in excess.
209 	 */
210 	while (skb_queue_len(&peer->staged_packet_queue) > MAX_STAGED_PACKETS) {
211 		dev_kfree_skb(__skb_dequeue(&peer->staged_packet_queue));
212 		++dev->stats.tx_dropped;
213 	}
214 	skb_queue_splice_tail(&packets, &peer->staged_packet_queue);
215 	spin_unlock_bh(&peer->staged_packet_queue.lock);
216 
217 	wg_packet_send_staged_packets(peer);
218 
219 	wg_peer_put(peer);
220 	return NETDEV_TX_OK;
221 
222 err_peer:
223 	wg_peer_put(peer);
224 err_icmp:
225 	if (skb->protocol == htons(ETH_P_IP))
226 		icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
227 	else if (skb->protocol == htons(ETH_P_IPV6))
228 		icmpv6_ndo_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
229 err:
230 	++dev->stats.tx_errors;
231 	kfree_skb(skb);
232 	return ret;
233 }
234 
235 static const struct net_device_ops netdev_ops = {
236 	.ndo_open		= wg_open,
237 	.ndo_stop		= wg_stop,
238 	.ndo_start_xmit		= wg_xmit,
239 	.ndo_get_stats64	= dev_get_tstats64
240 };
241 
242 static void wg_destruct(struct net_device *dev)
243 {
244 	struct wg_device *wg = netdev_priv(dev);
245 
246 	rtnl_lock();
247 	list_del(&wg->device_list);
248 	rtnl_unlock();
249 	mutex_lock(&wg->device_update_lock);
250 	rcu_assign_pointer(wg->creating_net, NULL);
251 	wg->incoming_port = 0;
252 	wg_socket_reinit(wg, NULL, NULL);
253 	/* The final references are cleared in the below calls to destroy_workqueue. */
254 	wg_peer_remove_all(wg);
255 	destroy_workqueue(wg->handshake_receive_wq);
256 	destroy_workqueue(wg->handshake_send_wq);
257 	destroy_workqueue(wg->packet_crypt_wq);
258 	wg_packet_queue_free(&wg->handshake_queue, true);
259 	wg_packet_queue_free(&wg->decrypt_queue, false);
260 	wg_packet_queue_free(&wg->encrypt_queue, false);
261 	rcu_barrier(); /* Wait for all the peers to be actually freed. */
262 	wg_ratelimiter_uninit();
263 	memzero_explicit(&wg->static_identity, sizeof(wg->static_identity));
264 	free_percpu(dev->tstats);
265 	kvfree(wg->index_hashtable);
266 	kvfree(wg->peer_hashtable);
267 	mutex_unlock(&wg->device_update_lock);
268 
269 	pr_debug("%s: Interface destroyed\n", dev->name);
270 	free_netdev(dev);
271 }
272 
273 static const struct device_type device_type = { .name = KBUILD_MODNAME };
274 
275 static void wg_setup(struct net_device *dev)
276 {
277 	struct wg_device *wg = netdev_priv(dev);
278 	enum { WG_NETDEV_FEATURES = NETIF_F_HW_CSUM | NETIF_F_RXCSUM |
279 				    NETIF_F_SG | NETIF_F_GSO |
280 				    NETIF_F_GSO_SOFTWARE | NETIF_F_HIGHDMA };
281 	const int overhead = MESSAGE_MINIMUM_LENGTH + sizeof(struct udphdr) +
282 			     max(sizeof(struct ipv6hdr), sizeof(struct iphdr));
283 
284 	dev->netdev_ops = &netdev_ops;
285 	dev->header_ops = &ip_tunnel_header_ops;
286 	dev->hard_header_len = 0;
287 	dev->addr_len = 0;
288 	dev->needed_headroom = DATA_PACKET_HEAD_ROOM;
289 	dev->needed_tailroom = noise_encrypted_len(MESSAGE_PADDING_MULTIPLE);
290 	dev->type = ARPHRD_NONE;
291 	dev->flags = IFF_POINTOPOINT | IFF_NOARP;
292 	dev->priv_flags |= IFF_NO_QUEUE;
293 	dev->features |= NETIF_F_LLTX;
294 	dev->features |= WG_NETDEV_FEATURES;
295 	dev->hw_features |= WG_NETDEV_FEATURES;
296 	dev->hw_enc_features |= WG_NETDEV_FEATURES;
297 	dev->mtu = ETH_DATA_LEN - overhead;
298 	dev->max_mtu = round_down(INT_MAX, MESSAGE_PADDING_MULTIPLE) - overhead;
299 
300 	SET_NETDEV_DEVTYPE(dev, &device_type);
301 
302 	/* We need to keep the dst around in case of icmp replies. */
303 	netif_keep_dst(dev);
304 
305 	memset(wg, 0, sizeof(*wg));
306 	wg->dev = dev;
307 }
308 
309 static int wg_newlink(struct net *src_net, struct net_device *dev,
310 		      struct nlattr *tb[], struct nlattr *data[],
311 		      struct netlink_ext_ack *extack)
312 {
313 	struct wg_device *wg = netdev_priv(dev);
314 	int ret = -ENOMEM;
315 
316 	rcu_assign_pointer(wg->creating_net, src_net);
317 	init_rwsem(&wg->static_identity.lock);
318 	mutex_init(&wg->socket_update_lock);
319 	mutex_init(&wg->device_update_lock);
320 	wg_allowedips_init(&wg->peer_allowedips);
321 	wg_cookie_checker_init(&wg->cookie_checker, wg);
322 	INIT_LIST_HEAD(&wg->peer_list);
323 	wg->device_update_gen = 1;
324 
325 	wg->peer_hashtable = wg_pubkey_hashtable_alloc();
326 	if (!wg->peer_hashtable)
327 		return ret;
328 
329 	wg->index_hashtable = wg_index_hashtable_alloc();
330 	if (!wg->index_hashtable)
331 		goto err_free_peer_hashtable;
332 
333 	dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
334 	if (!dev->tstats)
335 		goto err_free_index_hashtable;
336 
337 	wg->handshake_receive_wq = alloc_workqueue("wg-kex-%s",
338 			WQ_CPU_INTENSIVE | WQ_FREEZABLE, 0, dev->name);
339 	if (!wg->handshake_receive_wq)
340 		goto err_free_tstats;
341 
342 	wg->handshake_send_wq = alloc_workqueue("wg-kex-%s",
343 			WQ_UNBOUND | WQ_FREEZABLE, 0, dev->name);
344 	if (!wg->handshake_send_wq)
345 		goto err_destroy_handshake_receive;
346 
347 	wg->packet_crypt_wq = alloc_workqueue("wg-crypt-%s",
348 			WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 0, dev->name);
349 	if (!wg->packet_crypt_wq)
350 		goto err_destroy_handshake_send;
351 
352 	ret = wg_packet_queue_init(&wg->encrypt_queue, wg_packet_encrypt_worker,
353 				   MAX_QUEUED_PACKETS);
354 	if (ret < 0)
355 		goto err_destroy_packet_crypt;
356 
357 	ret = wg_packet_queue_init(&wg->decrypt_queue, wg_packet_decrypt_worker,
358 				   MAX_QUEUED_PACKETS);
359 	if (ret < 0)
360 		goto err_free_encrypt_queue;
361 
362 	ret = wg_packet_queue_init(&wg->handshake_queue, wg_packet_handshake_receive_worker,
363 				   MAX_QUEUED_INCOMING_HANDSHAKES);
364 	if (ret < 0)
365 		goto err_free_decrypt_queue;
366 
367 	ret = wg_ratelimiter_init();
368 	if (ret < 0)
369 		goto err_free_handshake_queue;
370 
371 	ret = register_netdevice(dev);
372 	if (ret < 0)
373 		goto err_uninit_ratelimiter;
374 
375 	list_add(&wg->device_list, &device_list);
376 
377 	/* We wait until the end to assign priv_destructor, so that
378 	 * register_netdevice doesn't call it for us if it fails.
379 	 */
380 	dev->priv_destructor = wg_destruct;
381 
382 	pr_debug("%s: Interface created\n", dev->name);
383 	return ret;
384 
385 err_uninit_ratelimiter:
386 	wg_ratelimiter_uninit();
387 err_free_handshake_queue:
388 	wg_packet_queue_free(&wg->handshake_queue, false);
389 err_free_decrypt_queue:
390 	wg_packet_queue_free(&wg->decrypt_queue, false);
391 err_free_encrypt_queue:
392 	wg_packet_queue_free(&wg->encrypt_queue, false);
393 err_destroy_packet_crypt:
394 	destroy_workqueue(wg->packet_crypt_wq);
395 err_destroy_handshake_send:
396 	destroy_workqueue(wg->handshake_send_wq);
397 err_destroy_handshake_receive:
398 	destroy_workqueue(wg->handshake_receive_wq);
399 err_free_tstats:
400 	free_percpu(dev->tstats);
401 err_free_index_hashtable:
402 	kvfree(wg->index_hashtable);
403 err_free_peer_hashtable:
404 	kvfree(wg->peer_hashtable);
405 	return ret;
406 }
407 
408 static struct rtnl_link_ops link_ops __read_mostly = {
409 	.kind			= KBUILD_MODNAME,
410 	.priv_size		= sizeof(struct wg_device),
411 	.setup			= wg_setup,
412 	.newlink		= wg_newlink,
413 };
414 
415 static void wg_netns_pre_exit(struct net *net)
416 {
417 	struct wg_device *wg;
418 	struct wg_peer *peer;
419 
420 	rtnl_lock();
421 	list_for_each_entry(wg, &device_list, device_list) {
422 		if (rcu_access_pointer(wg->creating_net) == net) {
423 			pr_debug("%s: Creating namespace exiting\n", wg->dev->name);
424 			netif_carrier_off(wg->dev);
425 			mutex_lock(&wg->device_update_lock);
426 			rcu_assign_pointer(wg->creating_net, NULL);
427 			wg_socket_reinit(wg, NULL, NULL);
428 			list_for_each_entry(peer, &wg->peer_list, peer_list)
429 				wg_socket_clear_peer_endpoint_src(peer);
430 			mutex_unlock(&wg->device_update_lock);
431 		}
432 	}
433 	rtnl_unlock();
434 }
435 
436 static struct pernet_operations pernet_ops = {
437 	.pre_exit = wg_netns_pre_exit
438 };
439 
440 int __init wg_device_init(void)
441 {
442 	int ret;
443 
444 	ret = register_pm_notifier(&pm_notifier);
445 	if (ret)
446 		return ret;
447 
448 	ret = register_random_vmfork_notifier(&vm_notifier);
449 	if (ret)
450 		goto error_pm;
451 
452 	ret = register_pernet_device(&pernet_ops);
453 	if (ret)
454 		goto error_vm;
455 
456 	ret = rtnl_link_register(&link_ops);
457 	if (ret)
458 		goto error_pernet;
459 
460 	return 0;
461 
462 error_pernet:
463 	unregister_pernet_device(&pernet_ops);
464 error_vm:
465 	unregister_random_vmfork_notifier(&vm_notifier);
466 error_pm:
467 	unregister_pm_notifier(&pm_notifier);
468 	return ret;
469 }
470 
471 void wg_device_uninit(void)
472 {
473 	rtnl_link_unregister(&link_ops);
474 	unregister_pernet_device(&pernet_ops);
475 	unregister_random_vmfork_notifier(&vm_notifier);
476 	unregister_pm_notifier(&pm_notifier);
477 	rcu_barrier();
478 }
479