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. */
update_rx_stats(struct wg_peer * peer,size_t len)20 static void update_rx_stats(struct wg_peer *peer, size_t len)
21 {
22 struct pcpu_sw_netstats *tstats =
23 get_cpu_ptr(peer->device->dev->tstats);
24
25 u64_stats_update_begin(&tstats->syncp);
26 ++tstats->rx_packets;
27 tstats->rx_bytes += len;
28 peer->rx_bytes += len;
29 u64_stats_update_end(&tstats->syncp);
30 put_cpu_ptr(tstats);
31 }
32
33 #define SKB_TYPE_LE32(skb) (((struct message_header *)(skb)->data)->type)
34
validate_header_len(struct sk_buff * skb)35 static size_t validate_header_len(struct sk_buff *skb)
36 {
37 if (unlikely(skb->len < sizeof(struct message_header)))
38 return 0;
39 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_DATA) &&
40 skb->len >= MESSAGE_MINIMUM_LENGTH)
41 return sizeof(struct message_data);
42 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION) &&
43 skb->len == sizeof(struct message_handshake_initiation))
44 return sizeof(struct message_handshake_initiation);
45 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE) &&
46 skb->len == sizeof(struct message_handshake_response))
47 return sizeof(struct message_handshake_response);
48 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE) &&
49 skb->len == sizeof(struct message_handshake_cookie))
50 return sizeof(struct message_handshake_cookie);
51 return 0;
52 }
53
prepare_skb_header(struct sk_buff * skb,struct wg_device * wg)54 static int prepare_skb_header(struct sk_buff *skb, struct wg_device *wg)
55 {
56 size_t data_offset, data_len, header_len;
57 struct udphdr *udp;
58
59 if (unlikely(!wg_check_packet_protocol(skb) ||
60 skb_transport_header(skb) < skb->head ||
61 (skb_transport_header(skb) + sizeof(struct udphdr)) >
62 skb_tail_pointer(skb)))
63 return -EINVAL; /* Bogus IP header */
64 udp = udp_hdr(skb);
65 data_offset = (u8 *)udp - skb->data;
66 if (unlikely(data_offset > U16_MAX ||
67 data_offset + sizeof(struct udphdr) > skb->len))
68 /* Packet has offset at impossible location or isn't big enough
69 * to have UDP fields.
70 */
71 return -EINVAL;
72 data_len = ntohs(udp->len);
73 if (unlikely(data_len < sizeof(struct udphdr) ||
74 data_len > skb->len - data_offset))
75 /* UDP packet is reporting too small of a size or lying about
76 * its size.
77 */
78 return -EINVAL;
79 data_len -= sizeof(struct udphdr);
80 data_offset = (u8 *)udp + sizeof(struct udphdr) - skb->data;
81 if (unlikely(!pskb_may_pull(skb,
82 data_offset + sizeof(struct message_header)) ||
83 pskb_trim(skb, data_len + data_offset) < 0))
84 return -EINVAL;
85 skb_pull(skb, data_offset);
86 if (unlikely(skb->len != data_len))
87 /* Final len does not agree with calculated len */
88 return -EINVAL;
89 header_len = validate_header_len(skb);
90 if (unlikely(!header_len))
91 return -EINVAL;
92 __skb_push(skb, data_offset);
93 if (unlikely(!pskb_may_pull(skb, data_offset + header_len)))
94 return -EINVAL;
95 __skb_pull(skb, data_offset);
96 return 0;
97 }
98
wg_receive_handshake_packet(struct wg_device * wg,struct sk_buff * skb)99 static void wg_receive_handshake_packet(struct wg_device *wg,
100 struct sk_buff *skb)
101 {
102 enum cookie_mac_state mac_state;
103 struct wg_peer *peer = NULL;
104 /* This is global, so that our load calculation applies to the whole
105 * system. We don't care about races with it at all.
106 */
107 static u64 last_under_load;
108 bool packet_needs_cookie;
109 bool under_load;
110
111 if (SKB_TYPE_LE32(skb) == cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE)) {
112 net_dbg_skb_ratelimited("%s: Receiving cookie response from %pISpfsc\n",
113 wg->dev->name, skb);
114 wg_cookie_message_consume(
115 (struct message_handshake_cookie *)skb->data, wg);
116 return;
117 }
118
119 under_load = skb_queue_len(&wg->incoming_handshakes) >=
120 MAX_QUEUED_INCOMING_HANDSHAKES / 8;
121 if (under_load) {
122 last_under_load = ktime_get_coarse_boottime_ns();
123 } else if (last_under_load) {
124 under_load = !wg_birthdate_has_expired(last_under_load, 1);
125 if (!under_load)
126 last_under_load = 0;
127 }
128 mac_state = wg_cookie_validate_packet(&wg->cookie_checker, skb,
129 under_load);
130 if ((under_load && mac_state == VALID_MAC_WITH_COOKIE) ||
131 (!under_load && mac_state == VALID_MAC_BUT_NO_COOKIE)) {
132 packet_needs_cookie = false;
133 } else if (under_load && mac_state == VALID_MAC_BUT_NO_COOKIE) {
134 packet_needs_cookie = true;
135 } else {
136 net_dbg_skb_ratelimited("%s: Invalid MAC of handshake, dropping packet from %pISpfsc\n",
137 wg->dev->name, skb);
138 return;
139 }
140
141 switch (SKB_TYPE_LE32(skb)) {
142 case cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION): {
143 struct message_handshake_initiation *message =
144 (struct message_handshake_initiation *)skb->data;
145
146 if (packet_needs_cookie) {
147 wg_packet_send_handshake_cookie(wg, skb,
148 message->sender_index);
149 return;
150 }
151 peer = wg_noise_handshake_consume_initiation(message, wg);
152 if (unlikely(!peer)) {
153 net_dbg_skb_ratelimited("%s: Invalid handshake initiation from %pISpfsc\n",
154 wg->dev->name, skb);
155 return;
156 }
157 wg_socket_set_peer_endpoint_from_skb(peer, skb);
158 net_dbg_ratelimited("%s: Receiving handshake initiation from peer %llu (%pISpfsc)\n",
159 wg->dev->name, peer->internal_id,
160 &peer->endpoint.addr);
161 wg_packet_send_handshake_response(peer);
162 break;
163 }
164 case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE): {
165 struct message_handshake_response *message =
166 (struct message_handshake_response *)skb->data;
167
168 if (packet_needs_cookie) {
169 wg_packet_send_handshake_cookie(wg, skb,
170 message->sender_index);
171 return;
172 }
173 peer = wg_noise_handshake_consume_response(message, wg);
174 if (unlikely(!peer)) {
175 net_dbg_skb_ratelimited("%s: Invalid handshake response from %pISpfsc\n",
176 wg->dev->name, skb);
177 return;
178 }
179 wg_socket_set_peer_endpoint_from_skb(peer, skb);
180 net_dbg_ratelimited("%s: Receiving handshake response from peer %llu (%pISpfsc)\n",
181 wg->dev->name, peer->internal_id,
182 &peer->endpoint.addr);
183 if (wg_noise_handshake_begin_session(&peer->handshake,
184 &peer->keypairs)) {
185 wg_timers_session_derived(peer);
186 wg_timers_handshake_complete(peer);
187 /* Calling this function will either send any existing
188 * packets in the queue and not send a keepalive, which
189 * is the best case, Or, if there's nothing in the
190 * queue, it will send a keepalive, in order to give
191 * immediate confirmation of the session.
192 */
193 wg_packet_send_keepalive(peer);
194 }
195 break;
196 }
197 }
198
199 if (unlikely(!peer)) {
200 WARN(1, "Somehow a wrong type of packet wound up in the handshake queue!\n");
201 return;
202 }
203
204 local_bh_disable();
205 update_rx_stats(peer, skb->len);
206 local_bh_enable();
207
208 wg_timers_any_authenticated_packet_received(peer);
209 wg_timers_any_authenticated_packet_traversal(peer);
210 wg_peer_put(peer);
211 }
212
wg_packet_handshake_receive_worker(struct work_struct * work)213 void wg_packet_handshake_receive_worker(struct work_struct *work)
214 {
215 struct wg_device *wg = container_of(work, struct multicore_worker,
216 work)->ptr;
217 struct sk_buff *skb;
218
219 while ((skb = skb_dequeue(&wg->incoming_handshakes)) != NULL) {
220 wg_receive_handshake_packet(wg, skb);
221 dev_kfree_skb(skb);
222 cond_resched();
223 }
224 }
225
keep_key_fresh(struct wg_peer * peer)226 static void keep_key_fresh(struct wg_peer *peer)
227 {
228 struct noise_keypair *keypair;
229 bool send;
230
231 if (peer->sent_lastminute_handshake)
232 return;
233
234 rcu_read_lock_bh();
235 keypair = rcu_dereference_bh(peer->keypairs.current_keypair);
236 send = keypair && READ_ONCE(keypair->sending.is_valid) &&
237 keypair->i_am_the_initiator &&
238 wg_birthdate_has_expired(keypair->sending.birthdate,
239 REJECT_AFTER_TIME - KEEPALIVE_TIMEOUT - REKEY_TIMEOUT);
240 rcu_read_unlock_bh();
241
242 if (unlikely(send)) {
243 peer->sent_lastminute_handshake = true;
244 wg_packet_send_queued_handshake_initiation(peer, false);
245 }
246 }
247
decrypt_packet(struct sk_buff * skb,struct noise_keypair * keypair)248 static bool decrypt_packet(struct sk_buff *skb, struct noise_keypair *keypair)
249 {
250 struct scatterlist sg[MAX_SKB_FRAGS + 8];
251 struct sk_buff *trailer;
252 unsigned int offset;
253 int num_frags;
254
255 if (unlikely(!keypair))
256 return false;
257
258 if (unlikely(!READ_ONCE(keypair->receiving.is_valid) ||
259 wg_birthdate_has_expired(keypair->receiving.birthdate, REJECT_AFTER_TIME) ||
260 keypair->receiving_counter.counter >= REJECT_AFTER_MESSAGES)) {
261 WRITE_ONCE(keypair->receiving.is_valid, false);
262 return false;
263 }
264
265 PACKET_CB(skb)->nonce =
266 le64_to_cpu(((struct message_data *)skb->data)->counter);
267
268 /* We ensure that the network header is part of the packet before we
269 * call skb_cow_data, so that there's no chance that data is removed
270 * from the skb, so that later we can extract the original endpoint.
271 */
272 offset = skb->data - skb_network_header(skb);
273 skb_push(skb, offset);
274 num_frags = skb_cow_data(skb, 0, &trailer);
275 offset += sizeof(struct message_data);
276 skb_pull(skb, offset);
277 if (unlikely(num_frags < 0 || num_frags > ARRAY_SIZE(sg)))
278 return false;
279
280 sg_init_table(sg, num_frags);
281 if (skb_to_sgvec(skb, sg, 0, skb->len) <= 0)
282 return false;
283
284 if (!chacha20poly1305_decrypt_sg_inplace(sg, skb->len, NULL, 0,
285 PACKET_CB(skb)->nonce,
286 keypair->receiving.key))
287 return false;
288
289 /* Another ugly situation of pushing and pulling the header so as to
290 * keep endpoint information intact.
291 */
292 skb_push(skb, offset);
293 if (pskb_trim(skb, skb->len - noise_encrypted_len(0)))
294 return false;
295 skb_pull(skb, offset);
296
297 return true;
298 }
299
300 /* This is RFC6479, a replay detection bitmap algorithm that avoids bitshifts */
counter_validate(struct noise_replay_counter * counter,u64 their_counter)301 static bool counter_validate(struct noise_replay_counter *counter, u64 their_counter)
302 {
303 unsigned long index, index_current, top, i;
304 bool ret = false;
305
306 spin_lock_bh(&counter->lock);
307
308 if (unlikely(counter->counter >= REJECT_AFTER_MESSAGES + 1 ||
309 their_counter >= REJECT_AFTER_MESSAGES))
310 goto out;
311
312 ++their_counter;
313
314 if (unlikely((COUNTER_WINDOW_SIZE + their_counter) <
315 counter->counter))
316 goto out;
317
318 index = their_counter >> ilog2(BITS_PER_LONG);
319
320 if (likely(their_counter > counter->counter)) {
321 index_current = counter->counter >> ilog2(BITS_PER_LONG);
322 top = min_t(unsigned long, index - index_current,
323 COUNTER_BITS_TOTAL / BITS_PER_LONG);
324 for (i = 1; i <= top; ++i)
325 counter->backtrack[(i + index_current) &
326 ((COUNTER_BITS_TOTAL / BITS_PER_LONG) - 1)] = 0;
327 counter->counter = their_counter;
328 }
329
330 index &= (COUNTER_BITS_TOTAL / BITS_PER_LONG) - 1;
331 ret = !test_and_set_bit(their_counter & (BITS_PER_LONG - 1),
332 &counter->backtrack[index]);
333
334 out:
335 spin_unlock_bh(&counter->lock);
336 return ret;
337 }
338
339 #include "selftest/counter.c"
340
wg_packet_consume_data_done(struct wg_peer * peer,struct sk_buff * skb,struct endpoint * endpoint)341 static void wg_packet_consume_data_done(struct wg_peer *peer,
342 struct sk_buff *skb,
343 struct endpoint *endpoint)
344 {
345 struct net_device *dev = peer->device->dev;
346 unsigned int len, len_before_trim;
347 struct wg_peer *routed_peer;
348
349 wg_socket_set_peer_endpoint(peer, endpoint);
350
351 if (unlikely(wg_noise_received_with_keypair(&peer->keypairs,
352 PACKET_CB(skb)->keypair))) {
353 wg_timers_handshake_complete(peer);
354 wg_packet_send_staged_packets(peer);
355 }
356
357 keep_key_fresh(peer);
358
359 wg_timers_any_authenticated_packet_received(peer);
360 wg_timers_any_authenticated_packet_traversal(peer);
361
362 /* A packet with length 0 is a keepalive packet */
363 if (unlikely(!skb->len)) {
364 update_rx_stats(peer, message_data_len(0));
365 net_dbg_ratelimited("%s: Receiving keepalive packet from peer %llu (%pISpfsc)\n",
366 dev->name, peer->internal_id,
367 &peer->endpoint.addr);
368 goto packet_processed;
369 }
370
371 wg_timers_data_received(peer);
372
373 if (unlikely(skb_network_header(skb) < skb->head))
374 goto dishonest_packet_size;
375 if (unlikely(!(pskb_network_may_pull(skb, sizeof(struct iphdr)) &&
376 (ip_hdr(skb)->version == 4 ||
377 (ip_hdr(skb)->version == 6 &&
378 pskb_network_may_pull(skb, sizeof(struct ipv6hdr)))))))
379 goto dishonest_packet_type;
380
381 skb->dev = dev;
382 /* We've already verified the Poly1305 auth tag, which means this packet
383 * was not modified in transit. We can therefore tell the networking
384 * stack that all checksums of every layer of encapsulation have already
385 * been checked "by the hardware" and therefore is unnecessary to check
386 * again in software.
387 */
388 skb->ip_summed = CHECKSUM_UNNECESSARY;
389 skb->csum_level = ~0; /* All levels */
390 skb->protocol = ip_tunnel_parse_protocol(skb);
391 if (skb->protocol == htons(ETH_P_IP)) {
392 len = ntohs(ip_hdr(skb)->tot_len);
393 if (unlikely(len < sizeof(struct iphdr)))
394 goto dishonest_packet_size;
395 INET_ECN_decapsulate(skb, PACKET_CB(skb)->ds, ip_hdr(skb)->tos);
396 } else if (skb->protocol == htons(ETH_P_IPV6)) {
397 len = ntohs(ipv6_hdr(skb)->payload_len) +
398 sizeof(struct ipv6hdr);
399 INET_ECN_decapsulate(skb, PACKET_CB(skb)->ds, ipv6_get_dsfield(ipv6_hdr(skb)));
400 } else {
401 goto dishonest_packet_type;
402 }
403
404 if (unlikely(len > skb->len))
405 goto dishonest_packet_size;
406 len_before_trim = skb->len;
407 if (unlikely(pskb_trim(skb, len)))
408 goto packet_processed;
409
410 routed_peer = wg_allowedips_lookup_src(&peer->device->peer_allowedips,
411 skb);
412 wg_peer_put(routed_peer); /* We don't need the extra reference. */
413
414 if (unlikely(routed_peer != peer))
415 goto dishonest_packet_peer;
416
417 napi_gro_receive(&peer->napi, skb);
418 update_rx_stats(peer, message_data_len(len_before_trim));
419 return;
420
421 dishonest_packet_peer:
422 net_dbg_skb_ratelimited("%s: Packet has unallowed src IP (%pISc) from peer %llu (%pISpfsc)\n",
423 dev->name, skb, peer->internal_id,
424 &peer->endpoint.addr);
425 ++dev->stats.rx_errors;
426 ++dev->stats.rx_frame_errors;
427 goto packet_processed;
428 dishonest_packet_type:
429 net_dbg_ratelimited("%s: Packet is neither ipv4 nor ipv6 from peer %llu (%pISpfsc)\n",
430 dev->name, peer->internal_id, &peer->endpoint.addr);
431 ++dev->stats.rx_errors;
432 ++dev->stats.rx_frame_errors;
433 goto packet_processed;
434 dishonest_packet_size:
435 net_dbg_ratelimited("%s: Packet has incorrect size from peer %llu (%pISpfsc)\n",
436 dev->name, peer->internal_id, &peer->endpoint.addr);
437 ++dev->stats.rx_errors;
438 ++dev->stats.rx_length_errors;
439 goto packet_processed;
440 packet_processed:
441 dev_kfree_skb(skb);
442 }
443
wg_packet_rx_poll(struct napi_struct * napi,int budget)444 int wg_packet_rx_poll(struct napi_struct *napi, int budget)
445 {
446 struct wg_peer *peer = container_of(napi, struct wg_peer, napi);
447 struct noise_keypair *keypair;
448 struct endpoint endpoint;
449 enum packet_state state;
450 struct sk_buff *skb;
451 int work_done = 0;
452 bool free;
453
454 if (unlikely(budget <= 0))
455 return 0;
456
457 while ((skb = wg_prev_queue_peek(&peer->rx_queue)) != NULL &&
458 (state = atomic_read_acquire(&PACKET_CB(skb)->state)) !=
459 PACKET_STATE_UNCRYPTED) {
460 wg_prev_queue_drop_peeked(&peer->rx_queue);
461 keypair = PACKET_CB(skb)->keypair;
462 free = true;
463
464 if (unlikely(state != PACKET_STATE_CRYPTED))
465 goto next;
466
467 if (unlikely(!counter_validate(&keypair->receiving_counter,
468 PACKET_CB(skb)->nonce))) {
469 net_dbg_ratelimited("%s: Packet has invalid nonce %llu (max %llu)\n",
470 peer->device->dev->name,
471 PACKET_CB(skb)->nonce,
472 keypair->receiving_counter.counter);
473 goto next;
474 }
475
476 if (unlikely(wg_socket_endpoint_from_skb(&endpoint, skb)))
477 goto next;
478
479 wg_reset_packet(skb, false);
480 wg_packet_consume_data_done(peer, skb, &endpoint);
481 free = false;
482
483 next:
484 wg_noise_keypair_put(keypair, false);
485 wg_peer_put(peer);
486 if (unlikely(free))
487 dev_kfree_skb(skb);
488
489 if (++work_done >= budget)
490 break;
491 }
492
493 if (work_done < budget)
494 napi_complete_done(napi, work_done);
495
496 return work_done;
497 }
498
wg_packet_decrypt_worker(struct work_struct * work)499 void wg_packet_decrypt_worker(struct work_struct *work)
500 {
501 struct crypt_queue *queue = container_of(work, struct multicore_worker,
502 work)->ptr;
503 struct sk_buff *skb;
504
505 while ((skb = ptr_ring_consume_bh(&queue->ring)) != NULL) {
506 enum packet_state state =
507 likely(decrypt_packet(skb, PACKET_CB(skb)->keypair)) ?
508 PACKET_STATE_CRYPTED : PACKET_STATE_DEAD;
509 wg_queue_enqueue_per_peer_rx(skb, state);
510 if (need_resched())
511 cond_resched();
512 }
513 }
514
wg_packet_consume_data(struct wg_device * wg,struct sk_buff * skb)515 static void wg_packet_consume_data(struct wg_device *wg, struct sk_buff *skb)
516 {
517 __le32 idx = ((struct message_data *)skb->data)->key_idx;
518 struct wg_peer *peer = NULL;
519 int ret;
520
521 rcu_read_lock_bh();
522 PACKET_CB(skb)->keypair =
523 (struct noise_keypair *)wg_index_hashtable_lookup(
524 wg->index_hashtable, INDEX_HASHTABLE_KEYPAIR, idx,
525 &peer);
526 if (unlikely(!wg_noise_keypair_get(PACKET_CB(skb)->keypair)))
527 goto err_keypair;
528
529 if (unlikely(READ_ONCE(peer->is_dead)))
530 goto err;
531
532 ret = wg_queue_enqueue_per_device_and_peer(&wg->decrypt_queue, &peer->rx_queue, skb,
533 wg->packet_crypt_wq, &wg->decrypt_queue.last_cpu);
534 if (unlikely(ret == -EPIPE))
535 wg_queue_enqueue_per_peer_rx(skb, PACKET_STATE_DEAD);
536 if (likely(!ret || ret == -EPIPE)) {
537 rcu_read_unlock_bh();
538 return;
539 }
540 err:
541 wg_noise_keypair_put(PACKET_CB(skb)->keypair, false);
542 err_keypair:
543 rcu_read_unlock_bh();
544 wg_peer_put(peer);
545 dev_kfree_skb(skb);
546 }
547
wg_packet_receive(struct wg_device * wg,struct sk_buff * skb)548 void wg_packet_receive(struct wg_device *wg, struct sk_buff *skb)
549 {
550 if (unlikely(prepare_skb_header(skb, wg) < 0))
551 goto err;
552 switch (SKB_TYPE_LE32(skb)) {
553 case cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION):
554 case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE):
555 case cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE): {
556 int cpu;
557
558 if (skb_queue_len(&wg->incoming_handshakes) >
559 MAX_QUEUED_INCOMING_HANDSHAKES ||
560 unlikely(!rng_is_initialized())) {
561 net_dbg_skb_ratelimited("%s: Dropping handshake packet from %pISpfsc\n",
562 wg->dev->name, skb);
563 goto err;
564 }
565 skb_queue_tail(&wg->incoming_handshakes, skb);
566 /* Queues up a call to packet_process_queued_handshake_
567 * packets(skb):
568 */
569 cpu = wg_cpumask_next_online(&wg->incoming_handshake_cpu);
570 queue_work_on(cpu, wg->handshake_receive_wq,
571 &per_cpu_ptr(wg->incoming_handshakes_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