xref: /linux/drivers/net/wireguard/receive.c (revision eeb9f5c2dcec90009d7cf12e780e7f9631993fc5)
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 "device.h"
8 #include "peer.h"
9 #include "timers.h"
10 #include "messages.h"
11 #include "cookie.h"
12 #include "socket.h"
13 
14 #include <linux/ip.h>
15 #include <linux/ipv6.h>
16 #include <linux/udp.h>
17 #include <net/ip_tunnels.h>
18 
19 /* Must be called with bh disabled. */
20 static void update_rx_stats(struct wg_peer *peer, size_t len)
21 {
22 	dev_sw_netstats_rx_add(peer->device->dev, len);
23 	peer->rx_bytes += len;
24 }
25 
26 #define SKB_TYPE_LE32(skb) (((struct message_header *)(skb)->data)->type)
27 
28 static size_t validate_header_len(struct sk_buff *skb)
29 {
30 	if (unlikely(skb->len < sizeof(struct message_header)))
31 		return 0;
32 	if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_DATA) &&
33 	    skb->len >= MESSAGE_MINIMUM_LENGTH)
34 		return sizeof(struct message_data);
35 	if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION) &&
36 	    skb->len == sizeof(struct message_handshake_initiation))
37 		return sizeof(struct message_handshake_initiation);
38 	if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE) &&
39 	    skb->len == sizeof(struct message_handshake_response))
40 		return sizeof(struct message_handshake_response);
41 	if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE) &&
42 	    skb->len == sizeof(struct message_handshake_cookie))
43 		return sizeof(struct message_handshake_cookie);
44 	return 0;
45 }
46 
47 static int prepare_skb_header(struct sk_buff *skb, struct wg_device *wg)
48 {
49 	size_t data_offset, data_len, header_len;
50 	struct udphdr *udp;
51 
52 	if (unlikely(!wg_check_packet_protocol(skb) ||
53 		     skb_transport_header(skb) < skb->head ||
54 		     (skb_transport_header(skb) + sizeof(struct udphdr)) >
55 			     skb_tail_pointer(skb)))
56 		return -EINVAL; /* Bogus IP header */
57 	udp = udp_hdr(skb);
58 	data_offset = (u8 *)udp - skb->data;
59 	if (unlikely(data_offset > U16_MAX ||
60 		     data_offset + sizeof(struct udphdr) > skb->len))
61 		/* Packet has offset at impossible location or isn't big enough
62 		 * to have UDP fields.
63 		 */
64 		return -EINVAL;
65 	data_len = ntohs(udp->len);
66 	if (unlikely(data_len < sizeof(struct udphdr) ||
67 		     data_len > skb->len - data_offset))
68 		/* UDP packet is reporting too small of a size or lying about
69 		 * its size.
70 		 */
71 		return -EINVAL;
72 	data_len -= sizeof(struct udphdr);
73 	data_offset = (u8 *)udp + sizeof(struct udphdr) - skb->data;
74 	if (unlikely(!pskb_may_pull(skb,
75 				data_offset + sizeof(struct message_header)) ||
76 		     pskb_trim(skb, data_len + data_offset) < 0))
77 		return -EINVAL;
78 	skb_pull(skb, data_offset);
79 	if (unlikely(skb->len != data_len))
80 		/* Final len does not agree with calculated len */
81 		return -EINVAL;
82 	header_len = validate_header_len(skb);
83 	if (unlikely(!header_len))
84 		return -EINVAL;
85 	__skb_push(skb, data_offset);
86 	if (unlikely(!pskb_may_pull(skb, data_offset + header_len)))
87 		return -EINVAL;
88 	__skb_pull(skb, data_offset);
89 	return 0;
90 }
91 
92 static void wg_receive_handshake_packet(struct wg_device *wg,
93 					struct sk_buff *skb)
94 {
95 	enum cookie_mac_state mac_state;
96 	struct wg_peer *peer = NULL;
97 	/* This is global, so that our load calculation applies to the whole
98 	 * system. We don't care about races with it at all.
99 	 */
100 	static u64 last_under_load;
101 	bool packet_needs_cookie;
102 	bool under_load;
103 
104 	if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE)) {
105 		net_dbg_skb_ratelimited("%s: Receiving cookie response from %pISpfsc\n",
106 					wg->dev->name, skb);
107 		wg_cookie_message_consume(
108 			(struct message_handshake_cookie *)skb->data, wg);
109 		return;
110 	}
111 
112 	under_load = atomic_read(&wg->handshake_queue_len) >=
113 			MAX_QUEUED_INCOMING_HANDSHAKES / 8;
114 	if (under_load) {
115 		last_under_load = ktime_get_coarse_boottime_ns();
116 	} else if (last_under_load) {
117 		under_load = !wg_birthdate_has_expired(last_under_load, 1);
118 		if (!under_load)
119 			last_under_load = 0;
120 	}
121 	mac_state = wg_cookie_validate_packet(&wg->cookie_checker, skb,
122 					      under_load);
123 	if ((under_load && mac_state == VALID_MAC_WITH_COOKIE) ||
124 	    (!under_load && mac_state == VALID_MAC_BUT_NO_COOKIE)) {
125 		packet_needs_cookie = false;
126 	} else if (under_load && mac_state == VALID_MAC_BUT_NO_COOKIE) {
127 		packet_needs_cookie = true;
128 	} else {
129 		net_dbg_skb_ratelimited("%s: Invalid MAC of handshake, dropping packet from %pISpfsc\n",
130 					wg->dev->name, skb);
131 		return;
132 	}
133 
134 	switch (SKB_TYPE_LE32(skb)) {
135 	case cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION): {
136 		struct message_handshake_initiation *message =
137 			(struct message_handshake_initiation *)skb->data;
138 
139 		if (packet_needs_cookie) {
140 			wg_packet_send_handshake_cookie(wg, skb,
141 							message->sender_index);
142 			return;
143 		}
144 		peer = wg_noise_handshake_consume_initiation(message, wg);
145 		if (unlikely(!peer)) {
146 			net_dbg_skb_ratelimited("%s: Invalid handshake initiation from %pISpfsc\n",
147 						wg->dev->name, skb);
148 			return;
149 		}
150 		wg_socket_set_peer_endpoint_from_skb(peer, skb);
151 		net_dbg_ratelimited("%s: Receiving handshake initiation from peer %llu (%pISpfsc)\n",
152 				    wg->dev->name, peer->internal_id,
153 				    &peer->endpoint.addr);
154 		wg_packet_send_handshake_response(peer);
155 		break;
156 	}
157 	case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE): {
158 		struct message_handshake_response *message =
159 			(struct message_handshake_response *)skb->data;
160 
161 		if (packet_needs_cookie) {
162 			wg_packet_send_handshake_cookie(wg, skb,
163 							message->sender_index);
164 			return;
165 		}
166 		peer = wg_noise_handshake_consume_response(message, wg);
167 		if (unlikely(!peer)) {
168 			net_dbg_skb_ratelimited("%s: Invalid handshake response from %pISpfsc\n",
169 						wg->dev->name, skb);
170 			return;
171 		}
172 		wg_socket_set_peer_endpoint_from_skb(peer, skb);
173 		net_dbg_ratelimited("%s: Receiving handshake response from peer %llu (%pISpfsc)\n",
174 				    wg->dev->name, peer->internal_id,
175 				    &peer->endpoint.addr);
176 		if (wg_noise_handshake_begin_session(&peer->handshake,
177 						     &peer->keypairs)) {
178 			wg_timers_session_derived(peer);
179 			wg_timers_handshake_complete(peer);
180 			/* Calling this function will either send any existing
181 			 * packets in the queue and not send a keepalive, which
182 			 * is the best case, Or, if there's nothing in the
183 			 * queue, it will send a keepalive, in order to give
184 			 * immediate confirmation of the session.
185 			 */
186 			wg_packet_send_keepalive(peer);
187 		}
188 		break;
189 	}
190 	}
191 
192 	if (unlikely(!peer)) {
193 		WARN(1, "Somehow a wrong type of packet wound up in the handshake queue!\n");
194 		return;
195 	}
196 
197 	local_bh_disable();
198 	update_rx_stats(peer, skb->len);
199 	local_bh_enable();
200 
201 	wg_timers_any_authenticated_packet_received(peer);
202 	wg_timers_any_authenticated_packet_traversal(peer);
203 	wg_peer_put(peer);
204 }
205 
206 void wg_packet_handshake_receive_worker(struct work_struct *work)
207 {
208 	struct crypt_queue *queue = container_of(work, struct multicore_worker, work)->ptr;
209 	struct wg_device *wg = container_of(queue, struct wg_device, handshake_queue);
210 	struct sk_buff *skb;
211 
212 	while ((skb = ptr_ring_consume_bh(&queue->ring)) != NULL) {
213 		wg_receive_handshake_packet(wg, skb);
214 		dev_kfree_skb(skb);
215 		atomic_dec(&wg->handshake_queue_len);
216 		cond_resched();
217 	}
218 }
219 
220 static void keep_key_fresh(struct wg_peer *peer)
221 {
222 	struct noise_keypair *keypair;
223 	bool send;
224 
225 	if (peer->sent_lastminute_handshake)
226 		return;
227 
228 	rcu_read_lock_bh();
229 	keypair = rcu_dereference_bh(peer->keypairs.current_keypair);
230 	send = keypair && READ_ONCE(keypair->sending.is_valid) &&
231 	       keypair->i_am_the_initiator &&
232 	       wg_birthdate_has_expired(keypair->sending.birthdate,
233 			REJECT_AFTER_TIME - KEEPALIVE_TIMEOUT - REKEY_TIMEOUT);
234 	rcu_read_unlock_bh();
235 
236 	if (unlikely(send)) {
237 		peer->sent_lastminute_handshake = true;
238 		wg_packet_send_queued_handshake_initiation(peer, false);
239 	}
240 }
241 
242 static bool decrypt_packet(struct sk_buff *skb, struct noise_keypair *keypair)
243 {
244 	struct scatterlist sg[MAX_SKB_FRAGS + 8];
245 	struct sk_buff *trailer;
246 	unsigned int offset;
247 	int num_frags;
248 
249 	if (unlikely(!keypair))
250 		return false;
251 
252 	if (unlikely(!READ_ONCE(keypair->receiving.is_valid) ||
253 		  wg_birthdate_has_expired(keypair->receiving.birthdate, REJECT_AFTER_TIME) ||
254 		  keypair->receiving_counter.counter >= REJECT_AFTER_MESSAGES)) {
255 		WRITE_ONCE(keypair->receiving.is_valid, false);
256 		return false;
257 	}
258 
259 	PACKET_CB(skb)->nonce =
260 		le64_to_cpu(((struct message_data *)skb->data)->counter);
261 
262 	/* We ensure that the network header is part of the packet before we
263 	 * call skb_cow_data, so that there's no chance that data is removed
264 	 * from the skb, so that later we can extract the original endpoint.
265 	 */
266 	offset = skb->data - skb_network_header(skb);
267 	skb_push(skb, offset);
268 	num_frags = skb_cow_data(skb, 0, &trailer);
269 	offset += sizeof(struct message_data);
270 	skb_pull(skb, offset);
271 	if (unlikely(num_frags < 0 || num_frags > ARRAY_SIZE(sg)))
272 		return false;
273 
274 	sg_init_table(sg, num_frags);
275 	if (skb_to_sgvec(skb, sg, 0, skb->len) <= 0)
276 		return false;
277 
278 	if (!chacha20poly1305_decrypt_sg_inplace(sg, skb->len, NULL, 0,
279 					         PACKET_CB(skb)->nonce,
280 						 keypair->receiving.key))
281 		return false;
282 
283 	/* Another ugly situation of pushing and pulling the header so as to
284 	 * keep endpoint information intact.
285 	 */
286 	skb_push(skb, offset);
287 	if (pskb_trim(skb, skb->len - noise_encrypted_len(0)))
288 		return false;
289 	skb_pull(skb, offset);
290 
291 	return true;
292 }
293 
294 /* This is RFC6479, a replay detection bitmap algorithm that avoids bitshifts */
295 static bool counter_validate(struct noise_replay_counter *counter, u64 their_counter)
296 {
297 	unsigned long index, index_current, top, i;
298 	bool ret = false;
299 
300 	spin_lock_bh(&counter->lock);
301 
302 	if (unlikely(counter->counter >= REJECT_AFTER_MESSAGES + 1 ||
303 		     their_counter >= REJECT_AFTER_MESSAGES))
304 		goto out;
305 
306 	++their_counter;
307 
308 	if (unlikely((COUNTER_WINDOW_SIZE + their_counter) <
309 		     counter->counter))
310 		goto out;
311 
312 	index = their_counter >> ilog2(BITS_PER_LONG);
313 
314 	if (likely(their_counter > counter->counter)) {
315 		index_current = counter->counter >> ilog2(BITS_PER_LONG);
316 		top = min_t(unsigned long, index - index_current,
317 			    COUNTER_BITS_TOTAL / BITS_PER_LONG);
318 		for (i = 1; i <= top; ++i)
319 			counter->backtrack[(i + index_current) &
320 				((COUNTER_BITS_TOTAL / BITS_PER_LONG) - 1)] = 0;
321 		counter->counter = their_counter;
322 	}
323 
324 	index &= (COUNTER_BITS_TOTAL / BITS_PER_LONG) - 1;
325 	ret = !test_and_set_bit(their_counter & (BITS_PER_LONG - 1),
326 				&counter->backtrack[index]);
327 
328 out:
329 	spin_unlock_bh(&counter->lock);
330 	return ret;
331 }
332 
333 #include "selftest/counter.c"
334 
335 static void wg_packet_consume_data_done(struct wg_peer *peer,
336 					struct sk_buff *skb,
337 					struct endpoint *endpoint)
338 {
339 	struct net_device *dev = peer->device->dev;
340 	unsigned int len, len_before_trim;
341 	struct wg_peer *routed_peer;
342 
343 	wg_socket_set_peer_endpoint(peer, endpoint);
344 
345 	if (unlikely(wg_noise_received_with_keypair(&peer->keypairs,
346 						    PACKET_CB(skb)->keypair))) {
347 		wg_timers_handshake_complete(peer);
348 		wg_packet_send_staged_packets(peer);
349 	}
350 
351 	keep_key_fresh(peer);
352 
353 	wg_timers_any_authenticated_packet_received(peer);
354 	wg_timers_any_authenticated_packet_traversal(peer);
355 
356 	/* A packet with length 0 is a keepalive packet */
357 	if (unlikely(!skb->len)) {
358 		update_rx_stats(peer, message_data_len(0));
359 		net_dbg_ratelimited("%s: Receiving keepalive packet from peer %llu (%pISpfsc)\n",
360 				    dev->name, peer->internal_id,
361 				    &peer->endpoint.addr);
362 		goto packet_processed;
363 	}
364 
365 	wg_timers_data_received(peer);
366 
367 	if (unlikely(skb_network_header(skb) < skb->head))
368 		goto dishonest_packet_size;
369 	if (unlikely(!(pskb_network_may_pull(skb, sizeof(struct iphdr)) &&
370 		       (ip_hdr(skb)->version == 4 ||
371 			(ip_hdr(skb)->version == 6 &&
372 			 pskb_network_may_pull(skb, sizeof(struct ipv6hdr)))))))
373 		goto dishonest_packet_type;
374 
375 	skb->dev = dev;
376 	/* We've already verified the Poly1305 auth tag, which means this packet
377 	 * was not modified in transit. We can therefore tell the networking
378 	 * stack that all checksums of every layer of encapsulation have already
379 	 * been checked "by the hardware" and therefore is unnecessary to check
380 	 * again in software.
381 	 */
382 	skb->ip_summed = CHECKSUM_UNNECESSARY;
383 	skb->csum_level = ~0; /* All levels */
384 	skb->protocol = ip_tunnel_parse_protocol(skb);
385 	if (skb->protocol == htons(ETH_P_IP)) {
386 		len = ntohs(ip_hdr(skb)->tot_len);
387 		if (unlikely(len < sizeof(struct iphdr)))
388 			goto dishonest_packet_size;
389 		INET_ECN_decapsulate(skb, PACKET_CB(skb)->ds, ip_hdr(skb)->tos);
390 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
391 		len = ntohs(ipv6_hdr(skb)->payload_len) +
392 		      sizeof(struct ipv6hdr);
393 		INET_ECN_decapsulate(skb, PACKET_CB(skb)->ds, ipv6_get_dsfield(ipv6_hdr(skb)));
394 	} else {
395 		goto dishonest_packet_type;
396 	}
397 
398 	if (unlikely(len > skb->len))
399 		goto dishonest_packet_size;
400 	len_before_trim = skb->len;
401 	if (unlikely(pskb_trim(skb, len)))
402 		goto packet_processed;
403 
404 	routed_peer = wg_allowedips_lookup_src(&peer->device->peer_allowedips,
405 					       skb);
406 	wg_peer_put(routed_peer); /* We don't need the extra reference. */
407 
408 	if (unlikely(routed_peer != peer))
409 		goto dishonest_packet_peer;
410 
411 	napi_gro_receive(&peer->napi, skb);
412 	update_rx_stats(peer, message_data_len(len_before_trim));
413 	return;
414 
415 dishonest_packet_peer:
416 	net_dbg_skb_ratelimited("%s: Packet has unallowed src IP (%pISc) from peer %llu (%pISpfsc)\n",
417 				dev->name, skb, peer->internal_id,
418 				&peer->endpoint.addr);
419 	DEV_STATS_INC(dev, rx_errors);
420 	DEV_STATS_INC(dev, rx_frame_errors);
421 	goto packet_processed;
422 dishonest_packet_type:
423 	net_dbg_ratelimited("%s: Packet is neither ipv4 nor ipv6 from peer %llu (%pISpfsc)\n",
424 			    dev->name, peer->internal_id, &peer->endpoint.addr);
425 	DEV_STATS_INC(dev, rx_errors);
426 	DEV_STATS_INC(dev, rx_frame_errors);
427 	goto packet_processed;
428 dishonest_packet_size:
429 	net_dbg_ratelimited("%s: Packet has incorrect size from peer %llu (%pISpfsc)\n",
430 			    dev->name, peer->internal_id, &peer->endpoint.addr);
431 	DEV_STATS_INC(dev, rx_errors);
432 	DEV_STATS_INC(dev, rx_length_errors);
433 	goto packet_processed;
434 packet_processed:
435 	dev_kfree_skb(skb);
436 }
437 
438 int wg_packet_rx_poll(struct napi_struct *napi, int budget)
439 {
440 	struct wg_peer *peer = container_of(napi, struct wg_peer, napi);
441 	struct noise_keypair *keypair;
442 	struct endpoint endpoint;
443 	enum packet_state state;
444 	struct sk_buff *skb;
445 	int work_done = 0;
446 	bool free;
447 
448 	if (unlikely(budget <= 0))
449 		return 0;
450 
451 	while ((skb = wg_prev_queue_peek(&peer->rx_queue)) != NULL &&
452 	       (state = atomic_read_acquire(&PACKET_CB(skb)->state)) !=
453 		       PACKET_STATE_UNCRYPTED) {
454 		wg_prev_queue_drop_peeked(&peer->rx_queue);
455 		keypair = PACKET_CB(skb)->keypair;
456 		free = true;
457 
458 		if (unlikely(state != PACKET_STATE_CRYPTED))
459 			goto next;
460 
461 		if (unlikely(!counter_validate(&keypair->receiving_counter,
462 					       PACKET_CB(skb)->nonce))) {
463 			net_dbg_ratelimited("%s: Packet has invalid nonce %llu (max %llu)\n",
464 					    peer->device->dev->name,
465 					    PACKET_CB(skb)->nonce,
466 					    keypair->receiving_counter.counter);
467 			goto next;
468 		}
469 
470 		if (unlikely(wg_socket_endpoint_from_skb(&endpoint, skb)))
471 			goto next;
472 
473 		wg_reset_packet(skb, false);
474 		wg_packet_consume_data_done(peer, skb, &endpoint);
475 		free = false;
476 
477 next:
478 		wg_noise_keypair_put(keypair, false);
479 		wg_peer_put(peer);
480 		if (unlikely(free))
481 			dev_kfree_skb(skb);
482 
483 		if (++work_done >= budget)
484 			break;
485 	}
486 
487 	if (work_done < budget)
488 		napi_complete_done(napi, work_done);
489 
490 	return work_done;
491 }
492 
493 void wg_packet_decrypt_worker(struct work_struct *work)
494 {
495 	struct crypt_queue *queue = container_of(work, struct multicore_worker,
496 						 work)->ptr;
497 	struct sk_buff *skb;
498 
499 	while ((skb = ptr_ring_consume_bh(&queue->ring)) != NULL) {
500 		enum packet_state state =
501 			likely(decrypt_packet(skb, PACKET_CB(skb)->keypair)) ?
502 				PACKET_STATE_CRYPTED : PACKET_STATE_DEAD;
503 		wg_queue_enqueue_per_peer_rx(skb, state);
504 		if (need_resched())
505 			cond_resched();
506 	}
507 }
508 
509 static void wg_packet_consume_data(struct wg_device *wg, struct sk_buff *skb)
510 {
511 	__le32 idx = ((struct message_data *)skb->data)->key_idx;
512 	struct wg_peer *peer = NULL;
513 	int ret;
514 
515 	rcu_read_lock_bh();
516 	PACKET_CB(skb)->keypair =
517 		(struct noise_keypair *)wg_index_hashtable_lookup(
518 			wg->index_hashtable, INDEX_HASHTABLE_KEYPAIR, idx,
519 			&peer);
520 	if (unlikely(!wg_noise_keypair_get(PACKET_CB(skb)->keypair)))
521 		goto err_keypair;
522 
523 	if (unlikely(READ_ONCE(peer->is_dead)))
524 		goto err;
525 
526 	ret = wg_queue_enqueue_per_device_and_peer(&wg->decrypt_queue, &peer->rx_queue, skb,
527 						   wg->packet_crypt_wq);
528 	if (unlikely(ret == -EPIPE))
529 		wg_queue_enqueue_per_peer_rx(skb, PACKET_STATE_DEAD);
530 	if (likely(!ret || ret == -EPIPE)) {
531 		rcu_read_unlock_bh();
532 		return;
533 	}
534 err:
535 	wg_noise_keypair_put(PACKET_CB(skb)->keypair, false);
536 err_keypair:
537 	rcu_read_unlock_bh();
538 	wg_peer_put(peer);
539 	dev_kfree_skb(skb);
540 }
541 
542 void wg_packet_receive(struct wg_device *wg, struct sk_buff *skb)
543 {
544 	if (unlikely(prepare_skb_header(skb, wg) < 0))
545 		goto err;
546 	switch (SKB_TYPE_LE32(skb)) {
547 	case cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION):
548 	case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE):
549 	case cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE): {
550 		int cpu, ret = -EBUSY;
551 
552 		if (unlikely(!rng_is_initialized()))
553 			goto drop;
554 		if (atomic_read(&wg->handshake_queue_len) > MAX_QUEUED_INCOMING_HANDSHAKES / 2) {
555 			if (spin_trylock_bh(&wg->handshake_queue.ring.producer_lock)) {
556 				ret = __ptr_ring_produce(&wg->handshake_queue.ring, skb);
557 				spin_unlock_bh(&wg->handshake_queue.ring.producer_lock);
558 			}
559 		} else
560 			ret = ptr_ring_produce_bh(&wg->handshake_queue.ring, skb);
561 		if (ret) {
562 	drop:
563 			net_dbg_skb_ratelimited("%s: Dropping handshake packet from %pISpfsc\n",
564 						wg->dev->name, skb);
565 			goto err;
566 		}
567 		atomic_inc(&wg->handshake_queue_len);
568 		cpu = wg_cpumask_next_online(&wg->handshake_queue.last_cpu);
569 		/* Queues up a call to packet_process_queued_handshake_packets(skb): */
570 		queue_work_on(cpu, wg->handshake_receive_wq,
571 			      &per_cpu_ptr(wg->handshake_queue.worker, cpu)->work);
572 		break;
573 	}
574 	case cpu_to_le32(MESSAGE_DATA):
575 		PACKET_CB(skb)->ds = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
576 		wg_packet_consume_data(wg, skb);
577 		break;
578 	default:
579 		WARN(1, "Non-exhaustive parsing of packet header lead to unknown packet type!\n");
580 		goto err;
581 	}
582 	return;
583 
584 err:
585 	dev_kfree_skb(skb);
586 }
587