Merge branch 'linux-5.8' of git://github.com/skeggsb/linux into drm-fixes
[linux-2.6-microblaze.git] / drivers / net / wireguard / receive.c
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         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
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
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
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
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
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
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 */
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
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
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 crypt_queue *queue = &peer->rx_queue;
448         struct noise_keypair *keypair;
449         struct endpoint endpoint;
450         enum packet_state state;
451         struct sk_buff *skb;
452         int work_done = 0;
453         bool free;
454
455         if (unlikely(budget <= 0))
456                 return 0;
457
458         while ((skb = __ptr_ring_peek(&queue->ring)) != NULL &&
459                (state = atomic_read_acquire(&PACKET_CB(skb)->state)) !=
460                        PACKET_STATE_UNCRYPTED) {
461                 __ptr_ring_discard_one(&queue->ring);
462                 peer = PACKET_PEER(skb);
463                 keypair = PACKET_CB(skb)->keypair;
464                 free = true;
465
466                 if (unlikely(state != PACKET_STATE_CRYPTED))
467                         goto next;
468
469                 if (unlikely(!counter_validate(&keypair->receiving_counter,
470                                                PACKET_CB(skb)->nonce))) {
471                         net_dbg_ratelimited("%s: Packet has invalid nonce %llu (max %llu)\n",
472                                             peer->device->dev->name,
473                                             PACKET_CB(skb)->nonce,
474                                             keypair->receiving_counter.counter);
475                         goto next;
476                 }
477
478                 if (unlikely(wg_socket_endpoint_from_skb(&endpoint, skb)))
479                         goto next;
480
481                 wg_reset_packet(skb, false);
482                 wg_packet_consume_data_done(peer, skb, &endpoint);
483                 free = false;
484
485 next:
486                 wg_noise_keypair_put(keypair, false);
487                 wg_peer_put(peer);
488                 if (unlikely(free))
489                         dev_kfree_skb(skb);
490
491                 if (++work_done >= budget)
492                         break;
493         }
494
495         if (work_done < budget)
496                 napi_complete_done(napi, work_done);
497
498         return work_done;
499 }
500
501 void wg_packet_decrypt_worker(struct work_struct *work)
502 {
503         struct crypt_queue *queue = container_of(work, struct multicore_worker,
504                                                  work)->ptr;
505         struct sk_buff *skb;
506
507         while ((skb = ptr_ring_consume_bh(&queue->ring)) != NULL) {
508                 enum packet_state state =
509                         likely(decrypt_packet(skb, PACKET_CB(skb)->keypair)) ?
510                                 PACKET_STATE_CRYPTED : PACKET_STATE_DEAD;
511                 wg_queue_enqueue_per_peer_napi(skb, state);
512                 if (need_resched())
513                         cond_resched();
514         }
515 }
516
517 static void wg_packet_consume_data(struct wg_device *wg, struct sk_buff *skb)
518 {
519         __le32 idx = ((struct message_data *)skb->data)->key_idx;
520         struct wg_peer *peer = NULL;
521         int ret;
522
523         rcu_read_lock_bh();
524         PACKET_CB(skb)->keypair =
525                 (struct noise_keypair *)wg_index_hashtable_lookup(
526                         wg->index_hashtable, INDEX_HASHTABLE_KEYPAIR, idx,
527                         &peer);
528         if (unlikely(!wg_noise_keypair_get(PACKET_CB(skb)->keypair)))
529                 goto err_keypair;
530
531         if (unlikely(READ_ONCE(peer->is_dead)))
532                 goto err;
533
534         ret = wg_queue_enqueue_per_device_and_peer(&wg->decrypt_queue,
535                                                    &peer->rx_queue, skb,
536                                                    wg->packet_crypt_wq,
537                                                    &wg->decrypt_queue.last_cpu);
538         if (unlikely(ret == -EPIPE))
539                 wg_queue_enqueue_per_peer_napi(skb, PACKET_STATE_DEAD);
540         if (likely(!ret || ret == -EPIPE)) {
541                 rcu_read_unlock_bh();
542                 return;
543         }
544 err:
545         wg_noise_keypair_put(PACKET_CB(skb)->keypair, false);
546 err_keypair:
547         rcu_read_unlock_bh();
548         wg_peer_put(peer);
549         dev_kfree_skb(skb);
550 }
551
552 void wg_packet_receive(struct wg_device *wg, struct sk_buff *skb)
553 {
554         if (unlikely(prepare_skb_header(skb, wg) < 0))
555                 goto err;
556         switch (SKB_TYPE_LE32(skb)) {
557         case cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION):
558         case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE):
559         case cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE): {
560                 int cpu;
561
562                 if (skb_queue_len(&wg->incoming_handshakes) >
563                             MAX_QUEUED_INCOMING_HANDSHAKES ||
564                     unlikely(!rng_is_initialized())) {
565                         net_dbg_skb_ratelimited("%s: Dropping handshake packet from %pISpfsc\n",
566                                                 wg->dev->name, skb);
567                         goto err;
568                 }
569                 skb_queue_tail(&wg->incoming_handshakes, skb);
570                 /* Queues up a call to packet_process_queued_handshake_
571                  * packets(skb):
572                  */
573                 cpu = wg_cpumask_next_online(&wg->incoming_handshake_cpu);
574                 queue_work_on(cpu, wg->handshake_receive_wq,
575                         &per_cpu_ptr(wg->incoming_handshakes_worker, cpu)->work);
576                 break;
577         }
578         case cpu_to_le32(MESSAGE_DATA):
579                 PACKET_CB(skb)->ds = ip_tunnel_get_dsfield(ip_hdr(skb), skb);
580                 wg_packet_consume_data(wg, skb);
581                 break;
582         default:
583                 WARN(1, "Non-exhaustive parsing of packet header lead to unknown packet type!\n");
584                 goto err;
585         }
586         return;
587
588 err:
589         dev_kfree_skb(skb);
590 }