xref: /netbsd-src/sys/net/if_wg.c (revision 7d62b00eb9ad855ffcd7da46b41e23feb5476fac)
1 /*	$NetBSD: if_wg.c,v 1.74 2023/01/05 20:32:18 christos Exp $	*/
2 
3 /*
4  * Copyright (C) Ryota Ozaki <ozaki.ryota@gmail.com>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*
33  * This network interface aims to implement the WireGuard protocol.
34  * The implementation is based on the paper of WireGuard as of
35  * 2018-06-30 [1].  The paper is referred in the source code with label
36  * [W].  Also the specification of the Noise protocol framework as of
37  * 2018-07-11 [2] is referred with label [N].
38  *
39  * [1] https://www.wireguard.com/papers/wireguard.pdf
40  * [2] http://noiseprotocol.org/noise.pdf
41  */
42 
43 #include <sys/cdefs.h>
44 __KERNEL_RCSID(0, "$NetBSD: if_wg.c,v 1.74 2023/01/05 20:32:18 christos Exp $");
45 
46 #ifdef _KERNEL_OPT
47 #include "opt_altq_enabled.h"
48 #include "opt_inet.h"
49 #endif
50 
51 #include <sys/param.h>
52 #include <sys/types.h>
53 
54 #include <sys/atomic.h>
55 #include <sys/callout.h>
56 #include <sys/cprng.h>
57 #include <sys/cpu.h>
58 #include <sys/device.h>
59 #include <sys/domain.h>
60 #include <sys/errno.h>
61 #include <sys/intr.h>
62 #include <sys/ioctl.h>
63 #include <sys/kernel.h>
64 #include <sys/kmem.h>
65 #include <sys/mbuf.h>
66 #include <sys/module.h>
67 #include <sys/mutex.h>
68 #include <sys/once.h>
69 #include <sys/percpu.h>
70 #include <sys/pserialize.h>
71 #include <sys/psref.h>
72 #include <sys/queue.h>
73 #include <sys/rwlock.h>
74 #include <sys/socket.h>
75 #include <sys/socketvar.h>
76 #include <sys/sockio.h>
77 #include <sys/sysctl.h>
78 #include <sys/syslog.h>
79 #include <sys/systm.h>
80 #include <sys/thmap.h>
81 #include <sys/threadpool.h>
82 #include <sys/time.h>
83 #include <sys/timespec.h>
84 #include <sys/workqueue.h>
85 
86 #include <net/bpf.h>
87 #include <net/if.h>
88 #include <net/if_types.h>
89 #include <net/if_wg.h>
90 #include <net/pktqueue.h>
91 #include <net/route.h>
92 
93 #include <netinet/in.h>
94 #include <netinet/in_pcb.h>
95 #include <netinet/in_var.h>
96 #include <netinet/ip.h>
97 #include <netinet/ip_var.h>
98 #include <netinet/udp.h>
99 #include <netinet/udp_var.h>
100 
101 #ifdef INET6
102 #include <netinet/ip6.h>
103 #include <netinet6/in6_pcb.h>
104 #include <netinet6/in6_var.h>
105 #include <netinet6/ip6_var.h>
106 #include <netinet6/udp6_var.h>
107 #endif /* INET6 */
108 
109 #include <prop/proplib.h>
110 
111 #include <crypto/blake2/blake2s.h>
112 #include <crypto/sodium/crypto_aead_chacha20poly1305.h>
113 #include <crypto/sodium/crypto_aead_xchacha20poly1305.h>
114 #include <crypto/sodium/crypto_scalarmult.h>
115 
116 #include "ioconf.h"
117 
118 #ifdef WG_RUMPKERNEL
119 #include "wg_user.h"
120 #endif
121 
122 /*
123  * Data structures
124  * - struct wg_softc is an instance of wg interfaces
125  *   - It has a list of peers (struct wg_peer)
126  *   - It has a threadpool job that sends/receives handshake messages and
127  *     runs event handlers
128  *   - It has its own two routing tables: one is for IPv4 and the other IPv6
129  * - struct wg_peer is a representative of a peer
130  *   - It has a struct work to handle handshakes and timer tasks
131  *   - It has a pair of session instances (struct wg_session)
132  *   - It has a pair of endpoint instances (struct wg_sockaddr)
133  *     - Normally one endpoint is used and the second one is used only on
134  *       a peer migration (a change of peer's IP address)
135  *   - It has a list of IP addresses and sub networks called allowedips
136  *     (struct wg_allowedip)
137  *     - A packets sent over a session is allowed if its destination matches
138  *       any IP addresses or sub networks of the list
139  * - struct wg_session represents a session of a secure tunnel with a peer
140  *   - Two instances of sessions belong to a peer; a stable session and a
141  *     unstable session
142  *   - A handshake process of a session always starts with a unstable instance
143  *   - Once a session is established, its instance becomes stable and the
144  *     other becomes unstable instead
145  *   - Data messages are always sent via a stable session
146  *
147  * Locking notes:
148  * - Each wg has a mutex(9) wg_lock, and a rwlock(9) wg_rwlock
149  *   - Changes to the peer list are serialized by wg_lock
150  *   - The peer list may be read with pserialize(9) and psref(9)
151  *   - The rwlock (wg_rwlock) protects the routing tables (wg_rtable_ipv[46])
152  *     => XXX replace by pserialize when routing table is psz-safe
153  * - Each peer (struct wg_peer, wgp) has a mutex wgp_lock, which can be taken
154  *   only in thread context and serializes:
155  *   - the stable and unstable session pointers
156  *   - all unstable session state
157  * - Packet processing may be done in softint context:
158  *   - The stable session can be read under pserialize(9) or psref(9)
159  *     - The stable session is always ESTABLISHED
160  *     - On a session swap, we must wait for all readers to release a
161  *       reference to a stable session before changing wgs_state and
162  *       session states
163  * - Lock order: wg_lock -> wgp_lock
164  */
165 
166 
167 #define WGLOG(level, fmt, args...)					      \
168 	log(level, "%s: " fmt, __func__, ##args)
169 
170 /* Debug options */
171 #ifdef WG_DEBUG
172 /* Output debug logs */
173 #ifndef WG_DEBUG_LOG
174 #define WG_DEBUG_LOG
175 #endif
176 /* Output trace logs */
177 #ifndef WG_DEBUG_TRACE
178 #define WG_DEBUG_TRACE
179 #endif
180 /* Output hash values, etc. */
181 #ifndef WG_DEBUG_DUMP
182 #define WG_DEBUG_DUMP
183 #endif
184 /* Make some internal parameters configurable for testing and debugging */
185 #ifndef WG_DEBUG_PARAMS
186 #define WG_DEBUG_PARAMS
187 #endif
188 #endif
189 
190 #ifdef WG_DEBUG_TRACE
191 #define WG_TRACE(msg)							      \
192 	log(LOG_DEBUG, "%s:%d: %s\n", __func__, __LINE__, (msg))
193 #else
194 #define WG_TRACE(msg)	__nothing
195 #endif
196 
197 #ifdef WG_DEBUG_LOG
198 #define WG_DLOG(fmt, args...)	log(LOG_DEBUG, "%s: " fmt, __func__, ##args)
199 #else
200 #define WG_DLOG(fmt, args...)	__nothing
201 #endif
202 
203 #define WG_LOG_RATECHECK(wgprc, level, fmt, args...)	do {		\
204 	if (ppsratecheck(&(wgprc)->wgprc_lasttime,			\
205 	    &(wgprc)->wgprc_curpps, 1)) {				\
206 		log(level, fmt, ##args);				\
207 	}								\
208 } while (0)
209 
210 #ifdef WG_DEBUG_PARAMS
211 static bool wg_force_underload = false;
212 #endif
213 
214 #ifdef WG_DEBUG_DUMP
215 
216 static char *
217 gethexdump(const char *p, size_t n)
218 {
219 	char *buf;
220 	size_t i;
221 
222 	if (n > SIZE_MAX/3 - 1)
223 		return NULL;
224 	buf = kmem_alloc(3*n + 1, KM_NOSLEEP);
225 	if (buf == NULL)
226 		return NULL;
227 	for (i = 0; i < n; i++)
228 		snprintf(buf + 3*i, 3 + 1, " %02hhx", p[i]);
229 	return buf;
230 }
231 
232 static void
233 puthexdump(char *buf, const void *p, size_t n)
234 {
235 
236 	if (buf == NULL)
237 		return;
238 	kmem_free(buf, 3*n + 1);
239 }
240 
241 #ifdef WG_RUMPKERNEL
242 static void
243 wg_dump_buf(const char *func, const char *buf, const size_t size)
244 {
245 	char *hex = gethexdump(buf, size);
246 
247 	log(LOG_DEBUG, "%s: %s\n", func, hex ? hex : "(enomem)");
248 	puthexdump(hex, buf, size);
249 }
250 #endif
251 
252 static void
253 wg_dump_hash(const uint8_t *func, const uint8_t *name, const uint8_t *hash,
254     const size_t size)
255 {
256 	char *hex = gethexdump(hash, size);
257 
258 	log(LOG_DEBUG, "%s: %s: %s\n", func, name, hex ? hex : "(enomem)");
259 	puthexdump(hex, hash, size);
260 }
261 
262 #define WG_DUMP_HASH(name, hash) \
263 	wg_dump_hash(__func__, name, hash, WG_HASH_LEN)
264 #define WG_DUMP_HASH48(name, hash) \
265 	wg_dump_hash(__func__, name, hash, 48)
266 #define WG_DUMP_BUF(buf, size) \
267 	wg_dump_buf(__func__, buf, size)
268 #else
269 #define WG_DUMP_HASH(name, hash)	__nothing
270 #define WG_DUMP_HASH48(name, hash)	__nothing
271 #define WG_DUMP_BUF(buf, size)	__nothing
272 #endif /* WG_DEBUG_DUMP */
273 
274 /* chosen somewhat arbitrarily -- fits in signed 16 bits NUL-termintaed */
275 #define	WG_MAX_PROPLEN		32766
276 
277 #define WG_MTU			1420
278 #define WG_ALLOWEDIPS		16
279 
280 #define CURVE25519_KEY_LEN	32
281 #define TAI64N_LEN		sizeof(uint32_t) * 3
282 #define POLY1305_AUTHTAG_LEN	16
283 #define HMAC_BLOCK_LEN		64
284 
285 /* [N] 4.1: "DHLEN must be 32 or greater."  WireGuard chooses 32. */
286 /* [N] 4.3: Hash functions */
287 #define NOISE_DHLEN		32
288 /* [N] 4.3: "Must be 32 or 64."  WireGuard chooses 32. */
289 #define NOISE_HASHLEN		32
290 #define NOISE_BLOCKLEN		64
291 #define NOISE_HKDF_OUTPUT_LEN	NOISE_HASHLEN
292 /* [N] 5.1: "k" */
293 #define NOISE_CIPHER_KEY_LEN	32
294 /*
295  * [N] 9.2: "psk"
296  *          "... psk is a 32-byte secret value provided by the application."
297  */
298 #define NOISE_PRESHARED_KEY_LEN	32
299 
300 #define WG_STATIC_KEY_LEN	CURVE25519_KEY_LEN
301 #define WG_TIMESTAMP_LEN	TAI64N_LEN
302 
303 #define WG_PRESHARED_KEY_LEN	NOISE_PRESHARED_KEY_LEN
304 
305 #define WG_COOKIE_LEN		16
306 #define WG_MAC_LEN		16
307 #define WG_RANDVAL_LEN		24
308 
309 #define WG_EPHEMERAL_KEY_LEN	CURVE25519_KEY_LEN
310 /* [N] 5.2: "ck: A chaining key of HASHLEN bytes" */
311 #define WG_CHAINING_KEY_LEN	NOISE_HASHLEN
312 /* [N] 5.2: "h: A hash output of HASHLEN bytes" */
313 #define WG_HASH_LEN		NOISE_HASHLEN
314 #define WG_CIPHER_KEY_LEN	NOISE_CIPHER_KEY_LEN
315 #define WG_DH_OUTPUT_LEN	NOISE_DHLEN
316 #define WG_KDF_OUTPUT_LEN	NOISE_HKDF_OUTPUT_LEN
317 #define WG_AUTHTAG_LEN		POLY1305_AUTHTAG_LEN
318 #define WG_DATA_KEY_LEN		32
319 #define WG_SALT_LEN		24
320 
321 /*
322  * The protocol messages
323  */
324 struct wg_msg {
325 	uint32_t	wgm_type;
326 } __packed;
327 
328 /* [W] 5.4.2 First Message: Initiator to Responder */
329 struct wg_msg_init {
330 	uint32_t	wgmi_type;
331 	uint32_t	wgmi_sender;
332 	uint8_t		wgmi_ephemeral[WG_EPHEMERAL_KEY_LEN];
333 	uint8_t		wgmi_static[WG_STATIC_KEY_LEN + WG_AUTHTAG_LEN];
334 	uint8_t		wgmi_timestamp[WG_TIMESTAMP_LEN + WG_AUTHTAG_LEN];
335 	uint8_t		wgmi_mac1[WG_MAC_LEN];
336 	uint8_t		wgmi_mac2[WG_MAC_LEN];
337 } __packed;
338 
339 /* [W] 5.4.3 Second Message: Responder to Initiator */
340 struct wg_msg_resp {
341 	uint32_t	wgmr_type;
342 	uint32_t	wgmr_sender;
343 	uint32_t	wgmr_receiver;
344 	uint8_t		wgmr_ephemeral[WG_EPHEMERAL_KEY_LEN];
345 	uint8_t		wgmr_empty[0 + WG_AUTHTAG_LEN];
346 	uint8_t		wgmr_mac1[WG_MAC_LEN];
347 	uint8_t		wgmr_mac2[WG_MAC_LEN];
348 } __packed;
349 
350 /* [W] 5.4.6 Subsequent Messages: Transport Data Messages */
351 struct wg_msg_data {
352 	uint32_t	wgmd_type;
353 	uint32_t	wgmd_receiver;
354 	uint64_t	wgmd_counter;
355 	uint32_t	wgmd_packet[0];
356 } __packed;
357 
358 /* [W] 5.4.7 Under Load: Cookie Reply Message */
359 struct wg_msg_cookie {
360 	uint32_t	wgmc_type;
361 	uint32_t	wgmc_receiver;
362 	uint8_t		wgmc_salt[WG_SALT_LEN];
363 	uint8_t		wgmc_cookie[WG_COOKIE_LEN + WG_AUTHTAG_LEN];
364 } __packed;
365 
366 #define WG_MSG_TYPE_INIT		1
367 #define WG_MSG_TYPE_RESP		2
368 #define WG_MSG_TYPE_COOKIE		3
369 #define WG_MSG_TYPE_DATA		4
370 #define WG_MSG_TYPE_MAX			WG_MSG_TYPE_DATA
371 
372 /* Sliding windows */
373 
374 #define	SLIWIN_BITS	2048u
375 #define	SLIWIN_TYPE	uint32_t
376 #define	SLIWIN_BPW	NBBY*sizeof(SLIWIN_TYPE)
377 #define	SLIWIN_WORDS	howmany(SLIWIN_BITS, SLIWIN_BPW)
378 #define	SLIWIN_NPKT	(SLIWIN_BITS - NBBY*sizeof(SLIWIN_TYPE))
379 
380 struct sliwin {
381 	SLIWIN_TYPE	B[SLIWIN_WORDS];
382 	uint64_t	T;
383 };
384 
385 static void
386 sliwin_reset(struct sliwin *W)
387 {
388 
389 	memset(W, 0, sizeof(*W));
390 }
391 
392 static int
393 sliwin_check_fast(const volatile struct sliwin *W, uint64_t S)
394 {
395 
396 	/*
397 	 * If it's more than one window older than the highest sequence
398 	 * number we've seen, reject.
399 	 */
400 #ifdef __HAVE_ATOMIC64_LOADSTORE
401 	if (S + SLIWIN_NPKT < atomic_load_relaxed(&W->T))
402 		return EAUTH;
403 #endif
404 
405 	/*
406 	 * Otherwise, we need to take the lock to decide, so don't
407 	 * reject just yet.  Caller must serialize a call to
408 	 * sliwin_update in this case.
409 	 */
410 	return 0;
411 }
412 
413 static int
414 sliwin_update(struct sliwin *W, uint64_t S)
415 {
416 	unsigned word, bit;
417 
418 	/*
419 	 * If it's more than one window older than the highest sequence
420 	 * number we've seen, reject.
421 	 */
422 	if (S + SLIWIN_NPKT < W->T)
423 		return EAUTH;
424 
425 	/*
426 	 * If it's higher than the highest sequence number we've seen,
427 	 * advance the window.
428 	 */
429 	if (S > W->T) {
430 		uint64_t i = W->T / SLIWIN_BPW;
431 		uint64_t j = S / SLIWIN_BPW;
432 		unsigned k;
433 
434 		for (k = 0; k < MIN(j - i, SLIWIN_WORDS); k++)
435 			W->B[(i + k + 1) % SLIWIN_WORDS] = 0;
436 #ifdef __HAVE_ATOMIC64_LOADSTORE
437 		atomic_store_relaxed(&W->T, S);
438 #else
439 		W->T = S;
440 #endif
441 	}
442 
443 	/* Test and set the bit -- if already set, reject.  */
444 	word = (S / SLIWIN_BPW) % SLIWIN_WORDS;
445 	bit = S % SLIWIN_BPW;
446 	if (W->B[word] & (1UL << bit))
447 		return EAUTH;
448 	W->B[word] |= 1U << bit;
449 
450 	/* Accept!  */
451 	return 0;
452 }
453 
454 struct wg_session {
455 	struct wg_peer	*wgs_peer;
456 	struct psref_target
457 			wgs_psref;
458 
459 	int		wgs_state;
460 #define WGS_STATE_UNKNOWN	0
461 #define WGS_STATE_INIT_ACTIVE	1
462 #define WGS_STATE_INIT_PASSIVE	2
463 #define WGS_STATE_ESTABLISHED	3
464 #define WGS_STATE_DESTROYING	4
465 
466 	time_t		wgs_time_established;
467 	time_t		wgs_time_last_data_sent;
468 	bool		wgs_is_initiator;
469 
470 	uint32_t	wgs_local_index;
471 	uint32_t	wgs_remote_index;
472 #ifdef __HAVE_ATOMIC64_LOADSTORE
473 	volatile uint64_t
474 			wgs_send_counter;
475 #else
476 	kmutex_t	wgs_send_counter_lock;
477 	uint64_t	wgs_send_counter;
478 #endif
479 
480 	struct {
481 		kmutex_t	lock;
482 		struct sliwin	window;
483 	}		*wgs_recvwin;
484 
485 	uint8_t		wgs_handshake_hash[WG_HASH_LEN];
486 	uint8_t		wgs_chaining_key[WG_CHAINING_KEY_LEN];
487 	uint8_t		wgs_ephemeral_key_pub[WG_EPHEMERAL_KEY_LEN];
488 	uint8_t		wgs_ephemeral_key_priv[WG_EPHEMERAL_KEY_LEN];
489 	uint8_t		wgs_ephemeral_key_peer[WG_EPHEMERAL_KEY_LEN];
490 	uint8_t		wgs_tkey_send[WG_DATA_KEY_LEN];
491 	uint8_t		wgs_tkey_recv[WG_DATA_KEY_LEN];
492 };
493 
494 struct wg_sockaddr {
495 	union {
496 		struct sockaddr_storage _ss;
497 		struct sockaddr _sa;
498 		struct sockaddr_in _sin;
499 		struct sockaddr_in6 _sin6;
500 	};
501 	struct psref_target	wgsa_psref;
502 };
503 
504 #define wgsatoss(wgsa)		(&(wgsa)->_ss)
505 #define wgsatosa(wgsa)		(&(wgsa)->_sa)
506 #define wgsatosin(wgsa)		(&(wgsa)->_sin)
507 #define wgsatosin6(wgsa)	(&(wgsa)->_sin6)
508 
509 #define	wgsa_family(wgsa)	(wgsatosa(wgsa)->sa_family)
510 
511 struct wg_peer;
512 struct wg_allowedip {
513 	struct radix_node	wga_nodes[2];
514 	struct wg_sockaddr	_wga_sa_addr;
515 	struct wg_sockaddr	_wga_sa_mask;
516 #define wga_sa_addr		_wga_sa_addr._sa
517 #define wga_sa_mask		_wga_sa_mask._sa
518 
519 	int			wga_family;
520 	uint8_t			wga_cidr;
521 	union {
522 		struct in_addr _ip4;
523 		struct in6_addr _ip6;
524 	} wga_addr;
525 #define wga_addr4	wga_addr._ip4
526 #define wga_addr6	wga_addr._ip6
527 
528 	struct wg_peer		*wga_peer;
529 };
530 
531 typedef uint8_t wg_timestamp_t[WG_TIMESTAMP_LEN];
532 
533 struct wg_ppsratecheck {
534 	struct timeval		wgprc_lasttime;
535 	int			wgprc_curpps;
536 };
537 
538 struct wg_softc;
539 struct wg_peer {
540 	struct wg_softc		*wgp_sc;
541 	char			wgp_name[WG_PEER_NAME_MAXLEN + 1];
542 	struct pslist_entry	wgp_peerlist_entry;
543 	pserialize_t		wgp_psz;
544 	struct psref_target	wgp_psref;
545 	kmutex_t		*wgp_lock;
546 	kmutex_t		*wgp_intr_lock;
547 
548 	uint8_t	wgp_pubkey[WG_STATIC_KEY_LEN];
549 	struct wg_sockaddr	*wgp_endpoint;
550 	struct wg_sockaddr	*wgp_endpoint0;
551 	volatile unsigned	wgp_endpoint_changing;
552 	bool			wgp_endpoint_available;
553 
554 			/* The preshared key (optional) */
555 	uint8_t		wgp_psk[WG_PRESHARED_KEY_LEN];
556 
557 	struct wg_session	*wgp_session_stable;
558 	struct wg_session	*wgp_session_unstable;
559 
560 	/* first outgoing packet awaiting session initiation */
561 	struct mbuf		*wgp_pending;
562 
563 	/* timestamp in big-endian */
564 	wg_timestamp_t	wgp_timestamp_latest_init;
565 
566 	struct timespec		wgp_last_handshake_time;
567 
568 	callout_t		wgp_rekey_timer;
569 	callout_t		wgp_handshake_timeout_timer;
570 	callout_t		wgp_session_dtor_timer;
571 
572 	time_t			wgp_handshake_start_time;
573 
574 	int			wgp_n_allowedips;
575 	struct wg_allowedip	wgp_allowedips[WG_ALLOWEDIPS];
576 
577 	time_t			wgp_latest_cookie_time;
578 	uint8_t			wgp_latest_cookie[WG_COOKIE_LEN];
579 	uint8_t			wgp_last_sent_mac1[WG_MAC_LEN];
580 	bool			wgp_last_sent_mac1_valid;
581 	uint8_t			wgp_last_sent_cookie[WG_COOKIE_LEN];
582 	bool			wgp_last_sent_cookie_valid;
583 
584 	time_t			wgp_last_msg_received_time[WG_MSG_TYPE_MAX];
585 
586 	time_t			wgp_last_genrandval_time;
587 	uint32_t		wgp_randval;
588 
589 	struct wg_ppsratecheck	wgp_ppsratecheck;
590 
591 	struct work		wgp_work;
592 	unsigned int		wgp_tasks;
593 #define WGP_TASK_SEND_INIT_MESSAGE		__BIT(0)
594 #define WGP_TASK_RETRY_HANDSHAKE		__BIT(1)
595 #define WGP_TASK_ESTABLISH_SESSION		__BIT(2)
596 #define WGP_TASK_ENDPOINT_CHANGED		__BIT(3)
597 #define WGP_TASK_SEND_KEEPALIVE_MESSAGE		__BIT(4)
598 #define WGP_TASK_DESTROY_PREV_SESSION		__BIT(5)
599 };
600 
601 struct wg_ops;
602 
603 struct wg_softc {
604 	struct ifnet	wg_if;
605 	LIST_ENTRY(wg_softc) wg_list;
606 	kmutex_t	*wg_lock;
607 	kmutex_t	*wg_intr_lock;
608 	krwlock_t	*wg_rwlock;
609 
610 	uint8_t		wg_privkey[WG_STATIC_KEY_LEN];
611 	uint8_t		wg_pubkey[WG_STATIC_KEY_LEN];
612 
613 	int		wg_npeers;
614 	struct pslist_head	wg_peers;
615 	struct thmap	*wg_peers_bypubkey;
616 	struct thmap	*wg_peers_byname;
617 	struct thmap	*wg_sessions_byindex;
618 	uint16_t	wg_listen_port;
619 
620 	struct threadpool	*wg_threadpool;
621 
622 	struct threadpool_job	wg_job;
623 	int			wg_upcalls;
624 #define	WG_UPCALL_INET	__BIT(0)
625 #define	WG_UPCALL_INET6	__BIT(1)
626 
627 #ifdef INET
628 	struct socket		*wg_so4;
629 	struct radix_node_head	*wg_rtable_ipv4;
630 #endif
631 #ifdef INET6
632 	struct socket		*wg_so6;
633 	struct radix_node_head	*wg_rtable_ipv6;
634 #endif
635 
636 	struct wg_ppsratecheck	wg_ppsratecheck;
637 
638 	struct wg_ops		*wg_ops;
639 
640 #ifdef WG_RUMPKERNEL
641 	struct wg_user		*wg_user;
642 #endif
643 };
644 
645 /* [W] 6.1 Preliminaries */
646 #define WG_REKEY_AFTER_MESSAGES		(1ULL << 60)
647 #define WG_REJECT_AFTER_MESSAGES	(UINT64_MAX - (1 << 13))
648 #define WG_REKEY_AFTER_TIME		120
649 #define WG_REJECT_AFTER_TIME		180
650 #define WG_REKEY_ATTEMPT_TIME		 90
651 #define WG_REKEY_TIMEOUT		  5
652 #define WG_KEEPALIVE_TIMEOUT		 10
653 
654 #define WG_COOKIE_TIME			120
655 #define WG_RANDVAL_TIME			(2 * 60)
656 
657 static uint64_t wg_rekey_after_messages = WG_REKEY_AFTER_MESSAGES;
658 static uint64_t wg_reject_after_messages = WG_REJECT_AFTER_MESSAGES;
659 static unsigned wg_rekey_after_time = WG_REKEY_AFTER_TIME;
660 static unsigned wg_reject_after_time = WG_REJECT_AFTER_TIME;
661 static unsigned wg_rekey_attempt_time = WG_REKEY_ATTEMPT_TIME;
662 static unsigned wg_rekey_timeout = WG_REKEY_TIMEOUT;
663 static unsigned wg_keepalive_timeout = WG_KEEPALIVE_TIMEOUT;
664 
665 static struct mbuf *
666 		wg_get_mbuf(size_t, size_t);
667 
668 static int	wg_send_data_msg(struct wg_peer *, struct wg_session *,
669 		    struct mbuf *);
670 static int	wg_send_cookie_msg(struct wg_softc *, struct wg_peer *,
671 		    const uint32_t, const uint8_t [], const struct sockaddr *);
672 static int	wg_send_handshake_msg_resp(struct wg_softc *, struct wg_peer *,
673 		    struct wg_session *, const struct wg_msg_init *);
674 static void	wg_send_keepalive_msg(struct wg_peer *, struct wg_session *);
675 
676 static struct wg_peer *
677 		wg_pick_peer_by_sa(struct wg_softc *, const struct sockaddr *,
678 		    struct psref *);
679 static struct wg_peer *
680 		wg_lookup_peer_by_pubkey(struct wg_softc *,
681 		    const uint8_t [], struct psref *);
682 
683 static struct wg_session *
684 		wg_lookup_session_by_index(struct wg_softc *,
685 		    const uint32_t, struct psref *);
686 
687 static void	wg_update_endpoint_if_necessary(struct wg_peer *,
688 		    const struct sockaddr *);
689 
690 static void	wg_schedule_rekey_timer(struct wg_peer *);
691 static void	wg_schedule_session_dtor_timer(struct wg_peer *);
692 
693 static bool	wg_is_underload(struct wg_softc *, struct wg_peer *, int);
694 static void	wg_calculate_keys(struct wg_session *, const bool);
695 
696 static void	wg_clear_states(struct wg_session *);
697 
698 static void	wg_get_peer(struct wg_peer *, struct psref *);
699 static void	wg_put_peer(struct wg_peer *, struct psref *);
700 
701 static int	wg_send_so(struct wg_peer *, struct mbuf *);
702 static int	wg_send_udp(struct wg_peer *, struct mbuf *);
703 static int	wg_output(struct ifnet *, struct mbuf *,
704 			   const struct sockaddr *, const struct rtentry *);
705 static void	wg_input(struct ifnet *, struct mbuf *, const int);
706 static int	wg_ioctl(struct ifnet *, u_long, void *);
707 static int	wg_bind_port(struct wg_softc *, const uint16_t);
708 static int	wg_init(struct ifnet *);
709 #ifdef ALTQ
710 static void	wg_start(struct ifnet *);
711 #endif
712 static void	wg_stop(struct ifnet *, int);
713 
714 static void	wg_peer_work(struct work *, void *);
715 static void	wg_job(struct threadpool_job *);
716 static void	wgintr(void *);
717 static void	wg_purge_pending_packets(struct wg_peer *);
718 
719 static int	wg_clone_create(struct if_clone *, int);
720 static int	wg_clone_destroy(struct ifnet *);
721 
722 struct wg_ops {
723 	int (*send_hs_msg)(struct wg_peer *, struct mbuf *);
724 	int (*send_data_msg)(struct wg_peer *, struct mbuf *);
725 	void (*input)(struct ifnet *, struct mbuf *, const int);
726 	int (*bind_port)(struct wg_softc *, const uint16_t);
727 };
728 
729 struct wg_ops wg_ops_rumpkernel = {
730 	.send_hs_msg	= wg_send_so,
731 	.send_data_msg	= wg_send_udp,
732 	.input		= wg_input,
733 	.bind_port	= wg_bind_port,
734 };
735 
736 #ifdef WG_RUMPKERNEL
737 static bool	wg_user_mode(struct wg_softc *);
738 static int	wg_ioctl_linkstr(struct wg_softc *, struct ifdrv *);
739 
740 static int	wg_send_user(struct wg_peer *, struct mbuf *);
741 static void	wg_input_user(struct ifnet *, struct mbuf *, const int);
742 static int	wg_bind_port_user(struct wg_softc *, const uint16_t);
743 
744 struct wg_ops wg_ops_rumpuser = {
745 	.send_hs_msg	= wg_send_user,
746 	.send_data_msg	= wg_send_user,
747 	.input		= wg_input_user,
748 	.bind_port	= wg_bind_port_user,
749 };
750 #endif
751 
752 #define WG_PEER_READER_FOREACH(wgp, wg)					\
753 	PSLIST_READER_FOREACH((wgp), &(wg)->wg_peers, struct wg_peer,	\
754 	    wgp_peerlist_entry)
755 #define WG_PEER_WRITER_FOREACH(wgp, wg)					\
756 	PSLIST_WRITER_FOREACH((wgp), &(wg)->wg_peers, struct wg_peer,	\
757 	    wgp_peerlist_entry)
758 #define WG_PEER_WRITER_INSERT_HEAD(wgp, wg)				\
759 	PSLIST_WRITER_INSERT_HEAD(&(wg)->wg_peers, (wgp), wgp_peerlist_entry)
760 #define WG_PEER_WRITER_REMOVE(wgp)					\
761 	PSLIST_WRITER_REMOVE((wgp), wgp_peerlist_entry)
762 
763 struct wg_route {
764 	struct radix_node	wgr_nodes[2];
765 	struct wg_peer		*wgr_peer;
766 };
767 
768 static struct radix_node_head *
769 wg_rnh(struct wg_softc *wg, const int family)
770 {
771 
772 	switch (family) {
773 		case AF_INET:
774 			return wg->wg_rtable_ipv4;
775 #ifdef INET6
776 		case AF_INET6:
777 			return wg->wg_rtable_ipv6;
778 #endif
779 		default:
780 			return NULL;
781 	}
782 }
783 
784 
785 /*
786  * Global variables
787  */
788 static volatile unsigned wg_count __cacheline_aligned;
789 
790 struct psref_class *wg_psref_class __read_mostly;
791 
792 static struct if_clone wg_cloner =
793     IF_CLONE_INITIALIZER("wg", wg_clone_create, wg_clone_destroy);
794 
795 static struct pktqueue *wg_pktq __read_mostly;
796 static struct workqueue *wg_wq __read_mostly;
797 
798 void wgattach(int);
799 /* ARGSUSED */
800 void
801 wgattach(int count)
802 {
803 	/*
804 	 * Nothing to do here, initialization is handled by the
805 	 * module initialization code in wginit() below).
806 	 */
807 }
808 
809 static void
810 wginit(void)
811 {
812 
813 	wg_psref_class = psref_class_create("wg", IPL_SOFTNET);
814 
815 	if_clone_attach(&wg_cloner);
816 }
817 
818 /*
819  * XXX Kludge: This should just happen in wginit, but workqueue_create
820  * cannot be run until after CPUs have been detected, and wginit runs
821  * before configure.
822  */
823 static int
824 wginitqueues(void)
825 {
826 	int error __diagused;
827 
828 	wg_pktq = pktq_create(IFQ_MAXLEN, wgintr, NULL);
829 	KASSERT(wg_pktq != NULL);
830 
831 	error = workqueue_create(&wg_wq, "wgpeer", wg_peer_work, NULL,
832 	    PRI_NONE, IPL_SOFTNET, WQ_MPSAFE|WQ_PERCPU);
833 	KASSERT(error == 0);
834 
835 	return 0;
836 }
837 
838 static void
839 wg_guarantee_initialized(void)
840 {
841 	static ONCE_DECL(init);
842 	int error __diagused;
843 
844 	error = RUN_ONCE(&init, wginitqueues);
845 	KASSERT(error == 0);
846 }
847 
848 static int
849 wg_count_inc(void)
850 {
851 	unsigned o, n;
852 
853 	do {
854 		o = atomic_load_relaxed(&wg_count);
855 		if (o == UINT_MAX)
856 			return ENFILE;
857 		n = o + 1;
858 	} while (atomic_cas_uint(&wg_count, o, n) != o);
859 
860 	return 0;
861 }
862 
863 static void
864 wg_count_dec(void)
865 {
866 	unsigned c __diagused;
867 
868 	c = atomic_dec_uint_nv(&wg_count);
869 	KASSERT(c != UINT_MAX);
870 }
871 
872 static int
873 wgdetach(void)
874 {
875 
876 	/* Prevent new interface creation.  */
877 	if_clone_detach(&wg_cloner);
878 
879 	/* Check whether there are any existing interfaces.  */
880 	if (atomic_load_relaxed(&wg_count)) {
881 		/* Back out -- reattach the cloner.  */
882 		if_clone_attach(&wg_cloner);
883 		return EBUSY;
884 	}
885 
886 	/* No interfaces left.  Nuke it.  */
887 	workqueue_destroy(wg_wq);
888 	pktq_destroy(wg_pktq);
889 	psref_class_destroy(wg_psref_class);
890 
891 	return 0;
892 }
893 
894 static void
895 wg_init_key_and_hash(uint8_t ckey[WG_CHAINING_KEY_LEN],
896     uint8_t hash[WG_HASH_LEN])
897 {
898 	/* [W] 5.4: CONSTRUCTION */
899 	const char *signature = "Noise_IKpsk2_25519_ChaChaPoly_BLAKE2s";
900 	/* [W] 5.4: IDENTIFIER */
901 	const char *id = "WireGuard v1 zx2c4 Jason@zx2c4.com";
902 	struct blake2s state;
903 
904 	blake2s(ckey, WG_CHAINING_KEY_LEN, NULL, 0,
905 	    signature, strlen(signature));
906 
907 	CTASSERT(WG_HASH_LEN == WG_CHAINING_KEY_LEN);
908 	memcpy(hash, ckey, WG_CHAINING_KEY_LEN);
909 
910 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
911 	blake2s_update(&state, ckey, WG_CHAINING_KEY_LEN);
912 	blake2s_update(&state, id, strlen(id));
913 	blake2s_final(&state, hash);
914 
915 	WG_DUMP_HASH("ckey", ckey);
916 	WG_DUMP_HASH("hash", hash);
917 }
918 
919 static void
920 wg_algo_hash(uint8_t hash[WG_HASH_LEN], const uint8_t input[],
921     const size_t inputsize)
922 {
923 	struct blake2s state;
924 
925 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
926 	blake2s_update(&state, hash, WG_HASH_LEN);
927 	blake2s_update(&state, input, inputsize);
928 	blake2s_final(&state, hash);
929 }
930 
931 static void
932 wg_algo_mac(uint8_t out[], const size_t outsize,
933     const uint8_t key[], const size_t keylen,
934     const uint8_t input1[], const size_t input1len,
935     const uint8_t input2[], const size_t input2len)
936 {
937 	struct blake2s state;
938 
939 	blake2s_init(&state, outsize, key, keylen);
940 
941 	blake2s_update(&state, input1, input1len);
942 	if (input2 != NULL)
943 		blake2s_update(&state, input2, input2len);
944 	blake2s_final(&state, out);
945 }
946 
947 static void
948 wg_algo_mac_mac1(uint8_t out[], const size_t outsize,
949     const uint8_t input1[], const size_t input1len,
950     const uint8_t input2[], const size_t input2len)
951 {
952 	struct blake2s state;
953 	/* [W] 5.4: LABEL-MAC1 */
954 	const char *label = "mac1----";
955 	uint8_t key[WG_HASH_LEN];
956 
957 	blake2s_init(&state, sizeof(key), NULL, 0);
958 	blake2s_update(&state, label, strlen(label));
959 	blake2s_update(&state, input1, input1len);
960 	blake2s_final(&state, key);
961 
962 	blake2s_init(&state, outsize, key, sizeof(key));
963 	if (input2 != NULL)
964 		blake2s_update(&state, input2, input2len);
965 	blake2s_final(&state, out);
966 }
967 
968 static void
969 wg_algo_mac_cookie(uint8_t out[], const size_t outsize,
970     const uint8_t input1[], const size_t input1len)
971 {
972 	struct blake2s state;
973 	/* [W] 5.4: LABEL-COOKIE */
974 	const char *label = "cookie--";
975 
976 	blake2s_init(&state, outsize, NULL, 0);
977 	blake2s_update(&state, label, strlen(label));
978 	blake2s_update(&state, input1, input1len);
979 	blake2s_final(&state, out);
980 }
981 
982 static void
983 wg_algo_generate_keypair(uint8_t pubkey[WG_EPHEMERAL_KEY_LEN],
984     uint8_t privkey[WG_EPHEMERAL_KEY_LEN])
985 {
986 
987 	CTASSERT(WG_EPHEMERAL_KEY_LEN == crypto_scalarmult_curve25519_BYTES);
988 
989 	cprng_strong(kern_cprng, privkey, WG_EPHEMERAL_KEY_LEN, 0);
990 	crypto_scalarmult_base(pubkey, privkey);
991 }
992 
993 static void
994 wg_algo_dh(uint8_t out[WG_DH_OUTPUT_LEN],
995     const uint8_t privkey[WG_STATIC_KEY_LEN],
996     const uint8_t pubkey[WG_STATIC_KEY_LEN])
997 {
998 
999 	CTASSERT(WG_STATIC_KEY_LEN == crypto_scalarmult_curve25519_BYTES);
1000 
1001 	int ret __diagused = crypto_scalarmult(out, privkey, pubkey);
1002 	KASSERT(ret == 0);
1003 }
1004 
1005 static void
1006 wg_algo_hmac(uint8_t out[], const size_t outlen,
1007     const uint8_t key[], const size_t keylen,
1008     const uint8_t in[], const size_t inlen)
1009 {
1010 #define IPAD	0x36
1011 #define OPAD	0x5c
1012 	uint8_t hmackey[HMAC_BLOCK_LEN] = {0};
1013 	uint8_t ipad[HMAC_BLOCK_LEN];
1014 	uint8_t opad[HMAC_BLOCK_LEN];
1015 	size_t i;
1016 	struct blake2s state;
1017 
1018 	KASSERT(outlen == WG_HASH_LEN);
1019 	KASSERT(keylen <= HMAC_BLOCK_LEN);
1020 
1021 	memcpy(hmackey, key, keylen);
1022 
1023 	for (i = 0; i < sizeof(hmackey); i++) {
1024 		ipad[i] = hmackey[i] ^ IPAD;
1025 		opad[i] = hmackey[i] ^ OPAD;
1026 	}
1027 
1028 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
1029 	blake2s_update(&state, ipad, sizeof(ipad));
1030 	blake2s_update(&state, in, inlen);
1031 	blake2s_final(&state, out);
1032 
1033 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
1034 	blake2s_update(&state, opad, sizeof(opad));
1035 	blake2s_update(&state, out, WG_HASH_LEN);
1036 	blake2s_final(&state, out);
1037 #undef IPAD
1038 #undef OPAD
1039 }
1040 
1041 static void
1042 wg_algo_kdf(uint8_t out1[WG_KDF_OUTPUT_LEN], uint8_t out2[WG_KDF_OUTPUT_LEN],
1043     uint8_t out3[WG_KDF_OUTPUT_LEN], const uint8_t ckey[WG_CHAINING_KEY_LEN],
1044     const uint8_t input[], const size_t inputlen)
1045 {
1046 	uint8_t tmp1[WG_KDF_OUTPUT_LEN], tmp2[WG_KDF_OUTPUT_LEN + 1];
1047 	uint8_t one[1];
1048 
1049 	/*
1050 	 * [N] 4.3: "an input_key_material byte sequence with length
1051 	 * either zero bytes, 32 bytes, or DHLEN bytes."
1052 	 */
1053 	KASSERT(inputlen == 0 || inputlen == 32 || inputlen == NOISE_DHLEN);
1054 
1055 	WG_DUMP_HASH("ckey", ckey);
1056 	if (input != NULL)
1057 		WG_DUMP_HASH("input", input);
1058 	wg_algo_hmac(tmp1, sizeof(tmp1), ckey, WG_CHAINING_KEY_LEN,
1059 	    input, inputlen);
1060 	WG_DUMP_HASH("tmp1", tmp1);
1061 	one[0] = 1;
1062 	wg_algo_hmac(out1, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
1063 	    one, sizeof(one));
1064 	WG_DUMP_HASH("out1", out1);
1065 	if (out2 == NULL)
1066 		return;
1067 	memcpy(tmp2, out1, WG_KDF_OUTPUT_LEN);
1068 	tmp2[WG_KDF_OUTPUT_LEN] = 2;
1069 	wg_algo_hmac(out2, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
1070 	    tmp2, sizeof(tmp2));
1071 	WG_DUMP_HASH("out2", out2);
1072 	if (out3 == NULL)
1073 		return;
1074 	memcpy(tmp2, out2, WG_KDF_OUTPUT_LEN);
1075 	tmp2[WG_KDF_OUTPUT_LEN] = 3;
1076 	wg_algo_hmac(out3, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
1077 	    tmp2, sizeof(tmp2));
1078 	WG_DUMP_HASH("out3", out3);
1079 }
1080 
1081 static void __noinline
1082 wg_algo_dh_kdf(uint8_t ckey[WG_CHAINING_KEY_LEN],
1083     uint8_t cipher_key[WG_CIPHER_KEY_LEN],
1084     const uint8_t local_key[WG_STATIC_KEY_LEN],
1085     const uint8_t remote_key[WG_STATIC_KEY_LEN])
1086 {
1087 	uint8_t dhout[WG_DH_OUTPUT_LEN];
1088 
1089 	wg_algo_dh(dhout, local_key, remote_key);
1090 	wg_algo_kdf(ckey, cipher_key, NULL, ckey, dhout, sizeof(dhout));
1091 
1092 	WG_DUMP_HASH("dhout", dhout);
1093 	WG_DUMP_HASH("ckey", ckey);
1094 	if (cipher_key != NULL)
1095 		WG_DUMP_HASH("cipher_key", cipher_key);
1096 }
1097 
1098 static void
1099 wg_algo_aead_enc(uint8_t out[], size_t expected_outsize, const uint8_t key[],
1100     const uint64_t counter, const uint8_t plain[], const size_t plainsize,
1101     const uint8_t auth[], size_t authlen)
1102 {
1103 	uint8_t nonce[(32 + 64) / 8] = {0};
1104 	long long unsigned int outsize;
1105 	int error __diagused;
1106 
1107 	le64enc(&nonce[4], counter);
1108 
1109 	error = crypto_aead_chacha20poly1305_ietf_encrypt(out, &outsize, plain,
1110 	    plainsize, auth, authlen, NULL, nonce, key);
1111 	KASSERT(error == 0);
1112 	KASSERT(outsize == expected_outsize);
1113 }
1114 
1115 static int
1116 wg_algo_aead_dec(uint8_t out[], size_t expected_outsize, const uint8_t key[],
1117     const uint64_t counter, const uint8_t encrypted[],
1118     const size_t encryptedsize, const uint8_t auth[], size_t authlen)
1119 {
1120 	uint8_t nonce[(32 + 64) / 8] = {0};
1121 	long long unsigned int outsize;
1122 	int error;
1123 
1124 	le64enc(&nonce[4], counter);
1125 
1126 	error = crypto_aead_chacha20poly1305_ietf_decrypt(out, &outsize, NULL,
1127 	    encrypted, encryptedsize, auth, authlen, nonce, key);
1128 	if (error == 0)
1129 		KASSERT(outsize == expected_outsize);
1130 	return error;
1131 }
1132 
1133 static void
1134 wg_algo_xaead_enc(uint8_t out[], const size_t expected_outsize,
1135     const uint8_t key[], const uint8_t plain[], const size_t plainsize,
1136     const uint8_t auth[], size_t authlen,
1137     const uint8_t nonce[WG_SALT_LEN])
1138 {
1139 	long long unsigned int outsize;
1140 	int error __diagused;
1141 
1142 	CTASSERT(WG_SALT_LEN == crypto_aead_xchacha20poly1305_ietf_NPUBBYTES);
1143 	error = crypto_aead_xchacha20poly1305_ietf_encrypt(out, &outsize,
1144 	    plain, plainsize, auth, authlen, NULL, nonce, key);
1145 	KASSERT(error == 0);
1146 	KASSERT(outsize == expected_outsize);
1147 }
1148 
1149 static int
1150 wg_algo_xaead_dec(uint8_t out[], const size_t expected_outsize,
1151     const uint8_t key[], const uint8_t encrypted[], const size_t encryptedsize,
1152     const uint8_t auth[], size_t authlen,
1153     const uint8_t nonce[WG_SALT_LEN])
1154 {
1155 	long long unsigned int outsize;
1156 	int error;
1157 
1158 	error = crypto_aead_xchacha20poly1305_ietf_decrypt(out, &outsize, NULL,
1159 	    encrypted, encryptedsize, auth, authlen, nonce, key);
1160 	if (error == 0)
1161 		KASSERT(outsize == expected_outsize);
1162 	return error;
1163 }
1164 
1165 static void
1166 wg_algo_tai64n(wg_timestamp_t timestamp)
1167 {
1168 	struct timespec ts;
1169 
1170 	/* FIXME strict TAI64N (https://cr.yp.to/libtai/tai64.html) */
1171 	getnanotime(&ts);
1172 	/* TAI64 label in external TAI64 format */
1173 	be32enc(timestamp, 0x40000000U + (uint32_t)(ts.tv_sec >> 32));
1174 	/* second beginning from 1970 TAI */
1175 	be32enc(timestamp + 4, (uint32_t)(ts.tv_sec & 0xffffffffU));
1176 	/* nanosecond in big-endian format */
1177 	be32enc(timestamp + 8, (uint32_t)ts.tv_nsec);
1178 }
1179 
1180 /*
1181  * wg_get_stable_session(wgp, psref)
1182  *
1183  *	Get a passive reference to the current stable session, or
1184  *	return NULL if there is no current stable session.
1185  *
1186  *	The pointer is always there but the session is not necessarily
1187  *	ESTABLISHED; if it is not ESTABLISHED, return NULL.  However,
1188  *	the session may transition from ESTABLISHED to DESTROYING while
1189  *	holding the passive reference.
1190  */
1191 static struct wg_session *
1192 wg_get_stable_session(struct wg_peer *wgp, struct psref *psref)
1193 {
1194 	int s;
1195 	struct wg_session *wgs;
1196 
1197 	s = pserialize_read_enter();
1198 	wgs = atomic_load_consume(&wgp->wgp_session_stable);
1199 	if (__predict_false(wgs->wgs_state != WGS_STATE_ESTABLISHED))
1200 		wgs = NULL;
1201 	else
1202 		psref_acquire(psref, &wgs->wgs_psref, wg_psref_class);
1203 	pserialize_read_exit(s);
1204 
1205 	return wgs;
1206 }
1207 
1208 static void
1209 wg_put_session(struct wg_session *wgs, struct psref *psref)
1210 {
1211 
1212 	psref_release(psref, &wgs->wgs_psref, wg_psref_class);
1213 }
1214 
1215 static void
1216 wg_destroy_session(struct wg_softc *wg, struct wg_session *wgs)
1217 {
1218 	struct wg_peer *wgp = wgs->wgs_peer;
1219 	struct wg_session *wgs0 __diagused;
1220 	void *garbage;
1221 
1222 	KASSERT(mutex_owned(wgp->wgp_lock));
1223 	KASSERT(wgs->wgs_state != WGS_STATE_UNKNOWN);
1224 
1225 	/* Remove the session from the table.  */
1226 	wgs0 = thmap_del(wg->wg_sessions_byindex,
1227 	    &wgs->wgs_local_index, sizeof(wgs->wgs_local_index));
1228 	KASSERT(wgs0 == wgs);
1229 	garbage = thmap_stage_gc(wg->wg_sessions_byindex);
1230 
1231 	/* Wait for passive references to drain.  */
1232 	pserialize_perform(wgp->wgp_psz);
1233 	psref_target_destroy(&wgs->wgs_psref, wg_psref_class);
1234 
1235 	/* Free memory, zero state, and transition to UNKNOWN.  */
1236 	thmap_gc(wg->wg_sessions_byindex, garbage);
1237 	wg_clear_states(wgs);
1238 	wgs->wgs_state = WGS_STATE_UNKNOWN;
1239 }
1240 
1241 /*
1242  * wg_get_session_index(wg, wgs)
1243  *
1244  *	Choose a session index for wgs->wgs_local_index, and store it
1245  *	in wg's table of sessions by index.
1246  *
1247  *	wgs must be the unstable session of its peer, and must be
1248  *	transitioning out of the UNKNOWN state.
1249  */
1250 static void
1251 wg_get_session_index(struct wg_softc *wg, struct wg_session *wgs)
1252 {
1253 	struct wg_peer *wgp __diagused = wgs->wgs_peer;
1254 	struct wg_session *wgs0;
1255 	uint32_t index;
1256 
1257 	KASSERT(mutex_owned(wgp->wgp_lock));
1258 	KASSERT(wgs == wgp->wgp_session_unstable);
1259 	KASSERT(wgs->wgs_state == WGS_STATE_UNKNOWN);
1260 
1261 	do {
1262 		/* Pick a uniform random index.  */
1263 		index = cprng_strong32();
1264 
1265 		/* Try to take it.  */
1266 		wgs->wgs_local_index = index;
1267 		wgs0 = thmap_put(wg->wg_sessions_byindex,
1268 		    &wgs->wgs_local_index, sizeof wgs->wgs_local_index, wgs);
1269 
1270 		/* If someone else beat us, start over.  */
1271 	} while (__predict_false(wgs0 != wgs));
1272 }
1273 
1274 /*
1275  * wg_put_session_index(wg, wgs)
1276  *
1277  *	Remove wgs from the table of sessions by index, wait for any
1278  *	passive references to drain, and transition the session to the
1279  *	UNKNOWN state.
1280  *
1281  *	wgs must be the unstable session of its peer, and must not be
1282  *	UNKNOWN or ESTABLISHED.
1283  */
1284 static void
1285 wg_put_session_index(struct wg_softc *wg, struct wg_session *wgs)
1286 {
1287 	struct wg_peer *wgp __diagused = wgs->wgs_peer;
1288 
1289 	KASSERT(mutex_owned(wgp->wgp_lock));
1290 	KASSERT(wgs == wgp->wgp_session_unstable);
1291 	KASSERT(wgs->wgs_state != WGS_STATE_UNKNOWN);
1292 	KASSERT(wgs->wgs_state != WGS_STATE_ESTABLISHED);
1293 
1294 	wg_destroy_session(wg, wgs);
1295 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
1296 }
1297 
1298 /*
1299  * Handshake patterns
1300  *
1301  * [W] 5: "These messages use the "IK" pattern from Noise"
1302  * [N] 7.5. Interactive handshake patterns (fundamental)
1303  *     "The first character refers to the initiator’s static key:"
1304  *     "I = Static key for initiator Immediately transmitted to responder,
1305  *          despite reduced or absent identity hiding"
1306  *     "The second character refers to the responder’s static key:"
1307  *     "K = Static key for responder Known to initiator"
1308  *     "IK:
1309  *        <- s
1310  *        ...
1311  *        -> e, es, s, ss
1312  *        <- e, ee, se"
1313  * [N] 9.4. Pattern modifiers
1314  *     "IKpsk2:
1315  *        <- s
1316  *        ...
1317  *        -> e, es, s, ss
1318  *        <- e, ee, se, psk"
1319  */
1320 static void
1321 wg_fill_msg_init(struct wg_softc *wg, struct wg_peer *wgp,
1322     struct wg_session *wgs, struct wg_msg_init *wgmi)
1323 {
1324 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.2: Ci */
1325 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.2: Hi */
1326 	uint8_t cipher_key[WG_CIPHER_KEY_LEN];
1327 	uint8_t pubkey[WG_EPHEMERAL_KEY_LEN];
1328 	uint8_t privkey[WG_EPHEMERAL_KEY_LEN];
1329 
1330 	KASSERT(mutex_owned(wgp->wgp_lock));
1331 	KASSERT(wgs == wgp->wgp_session_unstable);
1332 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_ACTIVE);
1333 
1334 	wgmi->wgmi_type = htole32(WG_MSG_TYPE_INIT);
1335 	wgmi->wgmi_sender = wgs->wgs_local_index;
1336 
1337 	/* [W] 5.4.2: First Message: Initiator to Responder */
1338 
1339 	/* Ci := HASH(CONSTRUCTION) */
1340 	/* Hi := HASH(Ci || IDENTIFIER) */
1341 	wg_init_key_and_hash(ckey, hash);
1342 	/* Hi := HASH(Hi || Sr^pub) */
1343 	wg_algo_hash(hash, wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey));
1344 
1345 	WG_DUMP_HASH("hash", hash);
1346 
1347 	/* [N] 2.2: "e" */
1348 	/* Ei^priv, Ei^pub := DH-GENERATE() */
1349 	wg_algo_generate_keypair(pubkey, privkey);
1350 	/* Ci := KDF1(Ci, Ei^pub) */
1351 	wg_algo_kdf(ckey, NULL, NULL, ckey, pubkey, sizeof(pubkey));
1352 	/* msg.ephemeral := Ei^pub */
1353 	memcpy(wgmi->wgmi_ephemeral, pubkey, sizeof(wgmi->wgmi_ephemeral));
1354 	/* Hi := HASH(Hi || msg.ephemeral) */
1355 	wg_algo_hash(hash, pubkey, sizeof(pubkey));
1356 
1357 	WG_DUMP_HASH("ckey", ckey);
1358 	WG_DUMP_HASH("hash", hash);
1359 
1360 	/* [N] 2.2: "es" */
1361 	/* Ci, k := KDF2(Ci, DH(Ei^priv, Sr^pub)) */
1362 	wg_algo_dh_kdf(ckey, cipher_key, privkey, wgp->wgp_pubkey);
1363 
1364 	/* [N] 2.2: "s" */
1365 	/* msg.static := AEAD(k, 0, Si^pub, Hi) */
1366 	wg_algo_aead_enc(wgmi->wgmi_static, sizeof(wgmi->wgmi_static),
1367 	    cipher_key, 0, wg->wg_pubkey, sizeof(wg->wg_pubkey),
1368 	    hash, sizeof(hash));
1369 	/* Hi := HASH(Hi || msg.static) */
1370 	wg_algo_hash(hash, wgmi->wgmi_static, sizeof(wgmi->wgmi_static));
1371 
1372 	WG_DUMP_HASH48("wgmi_static", wgmi->wgmi_static);
1373 
1374 	/* [N] 2.2: "ss" */
1375 	/* Ci, k := KDF2(Ci, DH(Si^priv, Sr^pub)) */
1376 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgp->wgp_pubkey);
1377 
1378 	/* msg.timestamp := AEAD(k, TIMESTAMP(), Hi) */
1379 	wg_timestamp_t timestamp;
1380 	wg_algo_tai64n(timestamp);
1381 	wg_algo_aead_enc(wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp),
1382 	    cipher_key, 0, timestamp, sizeof(timestamp), hash, sizeof(hash));
1383 	/* Hi := HASH(Hi || msg.timestamp) */
1384 	wg_algo_hash(hash, wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp));
1385 
1386 	/* [W] 5.4.4 Cookie MACs */
1387 	wg_algo_mac_mac1(wgmi->wgmi_mac1, sizeof(wgmi->wgmi_mac1),
1388 	    wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey),
1389 	    (const uint8_t *)wgmi, offsetof(struct wg_msg_init, wgmi_mac1));
1390 	/* Need mac1 to decrypt a cookie from a cookie message */
1391 	memcpy(wgp->wgp_last_sent_mac1, wgmi->wgmi_mac1,
1392 	    sizeof(wgp->wgp_last_sent_mac1));
1393 	wgp->wgp_last_sent_mac1_valid = true;
1394 
1395 	if (wgp->wgp_latest_cookie_time == 0 ||
1396 	    (time_uptime - wgp->wgp_latest_cookie_time) >= WG_COOKIE_TIME)
1397 		memset(wgmi->wgmi_mac2, 0, sizeof(wgmi->wgmi_mac2));
1398 	else {
1399 		wg_algo_mac(wgmi->wgmi_mac2, sizeof(wgmi->wgmi_mac2),
1400 		    wgp->wgp_latest_cookie, WG_COOKIE_LEN,
1401 		    (const uint8_t *)wgmi,
1402 		    offsetof(struct wg_msg_init, wgmi_mac2),
1403 		    NULL, 0);
1404 	}
1405 
1406 	memcpy(wgs->wgs_ephemeral_key_pub, pubkey, sizeof(pubkey));
1407 	memcpy(wgs->wgs_ephemeral_key_priv, privkey, sizeof(privkey));
1408 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
1409 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
1410 	WG_DLOG("%s: sender=%x\n", __func__, wgs->wgs_local_index);
1411 }
1412 
1413 static void __noinline
1414 wg_handle_msg_init(struct wg_softc *wg, const struct wg_msg_init *wgmi,
1415     const struct sockaddr *src)
1416 {
1417 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.2: Ci */
1418 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.2: Hi */
1419 	uint8_t cipher_key[WG_CIPHER_KEY_LEN];
1420 	uint8_t peer_pubkey[WG_STATIC_KEY_LEN];
1421 	struct wg_peer *wgp;
1422 	struct wg_session *wgs;
1423 	int error, ret;
1424 	struct psref psref_peer;
1425 	uint8_t mac1[WG_MAC_LEN];
1426 
1427 	WG_TRACE("init msg received");
1428 
1429 	wg_algo_mac_mac1(mac1, sizeof(mac1),
1430 	    wg->wg_pubkey, sizeof(wg->wg_pubkey),
1431 	    (const uint8_t *)wgmi, offsetof(struct wg_msg_init, wgmi_mac1));
1432 
1433 	/*
1434 	 * [W] 5.3: Denial of Service Mitigation & Cookies
1435 	 * "the responder, ..., must always reject messages with an invalid
1436 	 *  msg.mac1"
1437 	 */
1438 	if (!consttime_memequal(mac1, wgmi->wgmi_mac1, sizeof(mac1))) {
1439 		WG_DLOG("mac1 is invalid\n");
1440 		return;
1441 	}
1442 
1443 	/*
1444 	 * [W] 5.4.2: First Message: Initiator to Responder
1445 	 * "When the responder receives this message, it does the same
1446 	 *  operations so that its final state variables are identical,
1447 	 *  replacing the operands of the DH function to produce equivalent
1448 	 *  values."
1449 	 *  Note that the following comments of operations are just copies of
1450 	 *  the initiator's ones.
1451 	 */
1452 
1453 	/* Ci := HASH(CONSTRUCTION) */
1454 	/* Hi := HASH(Ci || IDENTIFIER) */
1455 	wg_init_key_and_hash(ckey, hash);
1456 	/* Hi := HASH(Hi || Sr^pub) */
1457 	wg_algo_hash(hash, wg->wg_pubkey, sizeof(wg->wg_pubkey));
1458 
1459 	/* [N] 2.2: "e" */
1460 	/* Ci := KDF1(Ci, Ei^pub) */
1461 	wg_algo_kdf(ckey, NULL, NULL, ckey, wgmi->wgmi_ephemeral,
1462 	    sizeof(wgmi->wgmi_ephemeral));
1463 	/* Hi := HASH(Hi || msg.ephemeral) */
1464 	wg_algo_hash(hash, wgmi->wgmi_ephemeral, sizeof(wgmi->wgmi_ephemeral));
1465 
1466 	WG_DUMP_HASH("ckey", ckey);
1467 
1468 	/* [N] 2.2: "es" */
1469 	/* Ci, k := KDF2(Ci, DH(Ei^priv, Sr^pub)) */
1470 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgmi->wgmi_ephemeral);
1471 
1472 	WG_DUMP_HASH48("wgmi_static", wgmi->wgmi_static);
1473 
1474 	/* [N] 2.2: "s" */
1475 	/* msg.static := AEAD(k, 0, Si^pub, Hi) */
1476 	error = wg_algo_aead_dec(peer_pubkey, WG_STATIC_KEY_LEN, cipher_key, 0,
1477 	    wgmi->wgmi_static, sizeof(wgmi->wgmi_static), hash, sizeof(hash));
1478 	if (error != 0) {
1479 		WG_LOG_RATECHECK(&wg->wg_ppsratecheck, LOG_DEBUG,
1480 		    "wg_algo_aead_dec for secret key failed\n");
1481 		return;
1482 	}
1483 	/* Hi := HASH(Hi || msg.static) */
1484 	wg_algo_hash(hash, wgmi->wgmi_static, sizeof(wgmi->wgmi_static));
1485 
1486 	wgp = wg_lookup_peer_by_pubkey(wg, peer_pubkey, &psref_peer);
1487 	if (wgp == NULL) {
1488 		WG_DLOG("peer not found\n");
1489 		return;
1490 	}
1491 
1492 	/*
1493 	 * Lock the peer to serialize access to cookie state.
1494 	 *
1495 	 * XXX Can we safely avoid holding the lock across DH?  Take it
1496 	 * just to verify mac2 and then unlock/DH/lock?
1497 	 */
1498 	mutex_enter(wgp->wgp_lock);
1499 
1500 	if (__predict_false(wg_is_underload(wg, wgp, WG_MSG_TYPE_INIT))) {
1501 		WG_TRACE("under load");
1502 		/*
1503 		 * [W] 5.3: Denial of Service Mitigation & Cookies
1504 		 * "the responder, ..., and when under load may reject messages
1505 		 *  with an invalid msg.mac2.  If the responder receives a
1506 		 *  message with a valid msg.mac1 yet with an invalid msg.mac2,
1507 		 *  and is under load, it may respond with a cookie reply
1508 		 *  message"
1509 		 */
1510 		uint8_t zero[WG_MAC_LEN] = {0};
1511 		if (consttime_memequal(wgmi->wgmi_mac2, zero, sizeof(zero))) {
1512 			WG_TRACE("sending a cookie message: no cookie included");
1513 			(void)wg_send_cookie_msg(wg, wgp, wgmi->wgmi_sender,
1514 			    wgmi->wgmi_mac1, src);
1515 			goto out;
1516 		}
1517 		if (!wgp->wgp_last_sent_cookie_valid) {
1518 			WG_TRACE("sending a cookie message: no cookie sent ever");
1519 			(void)wg_send_cookie_msg(wg, wgp, wgmi->wgmi_sender,
1520 			    wgmi->wgmi_mac1, src);
1521 			goto out;
1522 		}
1523 		uint8_t mac2[WG_MAC_LEN];
1524 		wg_algo_mac(mac2, sizeof(mac2), wgp->wgp_last_sent_cookie,
1525 		    WG_COOKIE_LEN, (const uint8_t *)wgmi,
1526 		    offsetof(struct wg_msg_init, wgmi_mac2), NULL, 0);
1527 		if (!consttime_memequal(mac2, wgmi->wgmi_mac2, sizeof(mac2))) {
1528 			WG_DLOG("mac2 is invalid\n");
1529 			goto out;
1530 		}
1531 		WG_TRACE("under load, but continue to sending");
1532 	}
1533 
1534 	/* [N] 2.2: "ss" */
1535 	/* Ci, k := KDF2(Ci, DH(Si^priv, Sr^pub)) */
1536 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgp->wgp_pubkey);
1537 
1538 	/* msg.timestamp := AEAD(k, TIMESTAMP(), Hi) */
1539 	wg_timestamp_t timestamp;
1540 	error = wg_algo_aead_dec(timestamp, sizeof(timestamp), cipher_key, 0,
1541 	    wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp),
1542 	    hash, sizeof(hash));
1543 	if (error != 0) {
1544 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
1545 		    "wg_algo_aead_dec for timestamp failed\n");
1546 		goto out;
1547 	}
1548 	/* Hi := HASH(Hi || msg.timestamp) */
1549 	wg_algo_hash(hash, wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp));
1550 
1551 	/*
1552 	 * [W] 5.1 "The responder keeps track of the greatest timestamp
1553 	 *      received per peer and discards packets containing
1554 	 *      timestamps less than or equal to it."
1555 	 */
1556 	ret = memcmp(timestamp, wgp->wgp_timestamp_latest_init,
1557 	    sizeof(timestamp));
1558 	if (ret <= 0) {
1559 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
1560 		    "invalid init msg: timestamp is old\n");
1561 		goto out;
1562 	}
1563 	memcpy(wgp->wgp_timestamp_latest_init, timestamp, sizeof(timestamp));
1564 
1565 	/*
1566 	 * Message is good -- we're committing to handle it now, unless
1567 	 * we were already initiating a session.
1568 	 */
1569 	wgs = wgp->wgp_session_unstable;
1570 	switch (wgs->wgs_state) {
1571 	case WGS_STATE_UNKNOWN:		/* new session initiated by peer */
1572 		wg_get_session_index(wg, wgs);
1573 		break;
1574 	case WGS_STATE_INIT_ACTIVE:	/* we're already initiating, drop */
1575 		WG_TRACE("Session already initializing, ignoring the message");
1576 		goto out;
1577 	case WGS_STATE_INIT_PASSIVE:	/* peer is retrying, start over */
1578 		WG_TRACE("Session already initializing, destroying old states");
1579 		wg_clear_states(wgs);
1580 		/* keep session index */
1581 		break;
1582 	case WGS_STATE_ESTABLISHED:	/* can't happen */
1583 		panic("unstable session can't be established");
1584 		break;
1585 	case WGS_STATE_DESTROYING:	/* rekey initiated by peer */
1586 		WG_TRACE("Session destroying, but force to clear");
1587 		callout_stop(&wgp->wgp_session_dtor_timer);
1588 		wg_clear_states(wgs);
1589 		/* keep session index */
1590 		break;
1591 	default:
1592 		panic("invalid session state: %d", wgs->wgs_state);
1593 	}
1594 	wgs->wgs_state = WGS_STATE_INIT_PASSIVE;
1595 
1596 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
1597 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
1598 	memcpy(wgs->wgs_ephemeral_key_peer, wgmi->wgmi_ephemeral,
1599 	    sizeof(wgmi->wgmi_ephemeral));
1600 
1601 	wg_update_endpoint_if_necessary(wgp, src);
1602 
1603 	(void)wg_send_handshake_msg_resp(wg, wgp, wgs, wgmi);
1604 
1605 	wg_calculate_keys(wgs, false);
1606 	wg_clear_states(wgs);
1607 
1608 out:
1609 	mutex_exit(wgp->wgp_lock);
1610 	wg_put_peer(wgp, &psref_peer);
1611 }
1612 
1613 static struct socket *
1614 wg_get_so_by_af(struct wg_softc *wg, const int af)
1615 {
1616 
1617 	switch (af) {
1618 #ifdef INET
1619 	case AF_INET:
1620 		return wg->wg_so4;
1621 #endif
1622 #ifdef INET6
1623 	case AF_INET6:
1624 		return wg->wg_so6;
1625 #endif
1626 	default:
1627 		panic("wg: no such af: %d", af);
1628 	}
1629 }
1630 
1631 static struct socket *
1632 wg_get_so_by_peer(struct wg_peer *wgp, struct wg_sockaddr *wgsa)
1633 {
1634 
1635 	return wg_get_so_by_af(wgp->wgp_sc, wgsa_family(wgsa));
1636 }
1637 
1638 static struct wg_sockaddr *
1639 wg_get_endpoint_sa(struct wg_peer *wgp, struct psref *psref)
1640 {
1641 	struct wg_sockaddr *wgsa;
1642 	int s;
1643 
1644 	s = pserialize_read_enter();
1645 	wgsa = atomic_load_consume(&wgp->wgp_endpoint);
1646 	psref_acquire(psref, &wgsa->wgsa_psref, wg_psref_class);
1647 	pserialize_read_exit(s);
1648 
1649 	return wgsa;
1650 }
1651 
1652 static void
1653 wg_put_sa(struct wg_peer *wgp, struct wg_sockaddr *wgsa, struct psref *psref)
1654 {
1655 
1656 	psref_release(psref, &wgsa->wgsa_psref, wg_psref_class);
1657 }
1658 
1659 static int
1660 wg_send_so(struct wg_peer *wgp, struct mbuf *m)
1661 {
1662 	int error;
1663 	struct socket *so;
1664 	struct psref psref;
1665 	struct wg_sockaddr *wgsa;
1666 
1667 	wgsa = wg_get_endpoint_sa(wgp, &psref);
1668 	so = wg_get_so_by_peer(wgp, wgsa);
1669 	error = sosend(so, wgsatosa(wgsa), NULL, m, NULL, 0, curlwp);
1670 	wg_put_sa(wgp, wgsa, &psref);
1671 
1672 	return error;
1673 }
1674 
1675 static int
1676 wg_send_handshake_msg_init(struct wg_softc *wg, struct wg_peer *wgp)
1677 {
1678 	int error;
1679 	struct mbuf *m;
1680 	struct wg_msg_init *wgmi;
1681 	struct wg_session *wgs;
1682 
1683 	KASSERT(mutex_owned(wgp->wgp_lock));
1684 
1685 	wgs = wgp->wgp_session_unstable;
1686 	/* XXX pull dispatch out into wg_task_send_init_message */
1687 	switch (wgs->wgs_state) {
1688 	case WGS_STATE_UNKNOWN:		/* new session initiated by us */
1689 		wg_get_session_index(wg, wgs);
1690 		break;
1691 	case WGS_STATE_INIT_ACTIVE:	/* we're already initiating, stop */
1692 		WG_TRACE("Session already initializing, skip starting new one");
1693 		return EBUSY;
1694 	case WGS_STATE_INIT_PASSIVE:	/* peer was trying -- XXX what now? */
1695 		WG_TRACE("Session already initializing, destroying old states");
1696 		wg_clear_states(wgs);
1697 		/* keep session index */
1698 		break;
1699 	case WGS_STATE_ESTABLISHED:	/* can't happen */
1700 		panic("unstable session can't be established");
1701 		break;
1702 	case WGS_STATE_DESTROYING:	/* rekey initiated by us too early */
1703 		WG_TRACE("Session destroying");
1704 		/* XXX should wait? */
1705 		return EBUSY;
1706 	}
1707 	wgs->wgs_state = WGS_STATE_INIT_ACTIVE;
1708 
1709 	m = m_gethdr(M_WAIT, MT_DATA);
1710 	if (sizeof(*wgmi) > MHLEN) {
1711 		m_clget(m, M_WAIT);
1712 		CTASSERT(sizeof(*wgmi) <= MCLBYTES);
1713 	}
1714 	m->m_pkthdr.len = m->m_len = sizeof(*wgmi);
1715 	wgmi = mtod(m, struct wg_msg_init *);
1716 	wg_fill_msg_init(wg, wgp, wgs, wgmi);
1717 
1718 	error = wg->wg_ops->send_hs_msg(wgp, m);
1719 	if (error == 0) {
1720 		WG_TRACE("init msg sent");
1721 
1722 		if (wgp->wgp_handshake_start_time == 0)
1723 			wgp->wgp_handshake_start_time = time_uptime;
1724 		callout_schedule(&wgp->wgp_handshake_timeout_timer,
1725 		    MIN(wg_rekey_timeout, (unsigned)(INT_MAX / hz)) * hz);
1726 	} else {
1727 		wg_put_session_index(wg, wgs);
1728 		/* Initiation failed; toss packet waiting for it if any.  */
1729 		if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL)
1730 			m_freem(m);
1731 	}
1732 
1733 	return error;
1734 }
1735 
1736 static void
1737 wg_fill_msg_resp(struct wg_softc *wg, struct wg_peer *wgp,
1738     struct wg_session *wgs, struct wg_msg_resp *wgmr,
1739     const struct wg_msg_init *wgmi)
1740 {
1741 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.3: Cr */
1742 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.3: Hr */
1743 	uint8_t cipher_key[WG_KDF_OUTPUT_LEN];
1744 	uint8_t pubkey[WG_EPHEMERAL_KEY_LEN];
1745 	uint8_t privkey[WG_EPHEMERAL_KEY_LEN];
1746 
1747 	KASSERT(mutex_owned(wgp->wgp_lock));
1748 	KASSERT(wgs == wgp->wgp_session_unstable);
1749 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_PASSIVE);
1750 
1751 	memcpy(hash, wgs->wgs_handshake_hash, sizeof(hash));
1752 	memcpy(ckey, wgs->wgs_chaining_key, sizeof(ckey));
1753 
1754 	wgmr->wgmr_type = htole32(WG_MSG_TYPE_RESP);
1755 	wgmr->wgmr_sender = wgs->wgs_local_index;
1756 	wgmr->wgmr_receiver = wgmi->wgmi_sender;
1757 
1758 	/* [W] 5.4.3 Second Message: Responder to Initiator */
1759 
1760 	/* [N] 2.2: "e" */
1761 	/* Er^priv, Er^pub := DH-GENERATE() */
1762 	wg_algo_generate_keypair(pubkey, privkey);
1763 	/* Cr := KDF1(Cr, Er^pub) */
1764 	wg_algo_kdf(ckey, NULL, NULL, ckey, pubkey, sizeof(pubkey));
1765 	/* msg.ephemeral := Er^pub */
1766 	memcpy(wgmr->wgmr_ephemeral, pubkey, sizeof(wgmr->wgmr_ephemeral));
1767 	/* Hr := HASH(Hr || msg.ephemeral) */
1768 	wg_algo_hash(hash, pubkey, sizeof(pubkey));
1769 
1770 	WG_DUMP_HASH("ckey", ckey);
1771 	WG_DUMP_HASH("hash", hash);
1772 
1773 	/* [N] 2.2: "ee" */
1774 	/* Cr := KDF1(Cr, DH(Er^priv, Ei^pub)) */
1775 	wg_algo_dh_kdf(ckey, NULL, privkey, wgs->wgs_ephemeral_key_peer);
1776 
1777 	/* [N] 2.2: "se" */
1778 	/* Cr := KDF1(Cr, DH(Er^priv, Si^pub)) */
1779 	wg_algo_dh_kdf(ckey, NULL, privkey, wgp->wgp_pubkey);
1780 
1781 	/* [N] 9.2: "psk" */
1782     {
1783 	uint8_t kdfout[WG_KDF_OUTPUT_LEN];
1784 	/* Cr, r, k := KDF3(Cr, Q) */
1785 	wg_algo_kdf(ckey, kdfout, cipher_key, ckey, wgp->wgp_psk,
1786 	    sizeof(wgp->wgp_psk));
1787 	/* Hr := HASH(Hr || r) */
1788 	wg_algo_hash(hash, kdfout, sizeof(kdfout));
1789     }
1790 
1791 	/* msg.empty := AEAD(k, 0, e, Hr) */
1792 	wg_algo_aead_enc(wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty),
1793 	    cipher_key, 0, NULL, 0, hash, sizeof(hash));
1794 	/* Hr := HASH(Hr || msg.empty) */
1795 	wg_algo_hash(hash, wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty));
1796 
1797 	WG_DUMP_HASH("wgmr_empty", wgmr->wgmr_empty);
1798 
1799 	/* [W] 5.4.4: Cookie MACs */
1800 	/* msg.mac1 := MAC(HASH(LABEL-MAC1 || Sm'^pub), msg_a) */
1801 	wg_algo_mac_mac1(wgmr->wgmr_mac1, sizeof(wgmi->wgmi_mac1),
1802 	    wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey),
1803 	    (const uint8_t *)wgmr, offsetof(struct wg_msg_resp, wgmr_mac1));
1804 	/* Need mac1 to decrypt a cookie from a cookie message */
1805 	memcpy(wgp->wgp_last_sent_mac1, wgmr->wgmr_mac1,
1806 	    sizeof(wgp->wgp_last_sent_mac1));
1807 	wgp->wgp_last_sent_mac1_valid = true;
1808 
1809 	if (wgp->wgp_latest_cookie_time == 0 ||
1810 	    (time_uptime - wgp->wgp_latest_cookie_time) >= WG_COOKIE_TIME)
1811 		/* msg.mac2 := 0^16 */
1812 		memset(wgmr->wgmr_mac2, 0, sizeof(wgmr->wgmr_mac2));
1813 	else {
1814 		/* msg.mac2 := MAC(Lm, msg_b) */
1815 		wg_algo_mac(wgmr->wgmr_mac2, sizeof(wgmi->wgmi_mac2),
1816 		    wgp->wgp_latest_cookie, WG_COOKIE_LEN,
1817 		    (const uint8_t *)wgmr,
1818 		    offsetof(struct wg_msg_resp, wgmr_mac2),
1819 		    NULL, 0);
1820 	}
1821 
1822 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
1823 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
1824 	memcpy(wgs->wgs_ephemeral_key_pub, pubkey, sizeof(pubkey));
1825 	memcpy(wgs->wgs_ephemeral_key_priv, privkey, sizeof(privkey));
1826 	wgs->wgs_remote_index = wgmi->wgmi_sender;
1827 	WG_DLOG("sender=%x\n", wgs->wgs_local_index);
1828 	WG_DLOG("receiver=%x\n", wgs->wgs_remote_index);
1829 }
1830 
1831 static void
1832 wg_swap_sessions(struct wg_peer *wgp)
1833 {
1834 	struct wg_session *wgs, *wgs_prev;
1835 
1836 	KASSERT(mutex_owned(wgp->wgp_lock));
1837 
1838 	wgs = wgp->wgp_session_unstable;
1839 	KASSERT(wgs->wgs_state == WGS_STATE_ESTABLISHED);
1840 
1841 	wgs_prev = wgp->wgp_session_stable;
1842 	KASSERT(wgs_prev->wgs_state == WGS_STATE_ESTABLISHED ||
1843 	    wgs_prev->wgs_state == WGS_STATE_UNKNOWN);
1844 	atomic_store_release(&wgp->wgp_session_stable, wgs);
1845 	wgp->wgp_session_unstable = wgs_prev;
1846 }
1847 
1848 static void __noinline
1849 wg_handle_msg_resp(struct wg_softc *wg, const struct wg_msg_resp *wgmr,
1850     const struct sockaddr *src)
1851 {
1852 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.3: Cr */
1853 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.3: Kr */
1854 	uint8_t cipher_key[WG_KDF_OUTPUT_LEN];
1855 	struct wg_peer *wgp;
1856 	struct wg_session *wgs;
1857 	struct psref psref;
1858 	int error;
1859 	uint8_t mac1[WG_MAC_LEN];
1860 	struct wg_session *wgs_prev;
1861 	struct mbuf *m;
1862 
1863 	wg_algo_mac_mac1(mac1, sizeof(mac1),
1864 	    wg->wg_pubkey, sizeof(wg->wg_pubkey),
1865 	    (const uint8_t *)wgmr, offsetof(struct wg_msg_resp, wgmr_mac1));
1866 
1867 	/*
1868 	 * [W] 5.3: Denial of Service Mitigation & Cookies
1869 	 * "the responder, ..., must always reject messages with an invalid
1870 	 *  msg.mac1"
1871 	 */
1872 	if (!consttime_memequal(mac1, wgmr->wgmr_mac1, sizeof(mac1))) {
1873 		WG_DLOG("mac1 is invalid\n");
1874 		return;
1875 	}
1876 
1877 	WG_TRACE("resp msg received");
1878 	wgs = wg_lookup_session_by_index(wg, wgmr->wgmr_receiver, &psref);
1879 	if (wgs == NULL) {
1880 		WG_TRACE("No session found");
1881 		return;
1882 	}
1883 
1884 	wgp = wgs->wgs_peer;
1885 
1886 	mutex_enter(wgp->wgp_lock);
1887 
1888 	/* If we weren't waiting for a handshake response, drop it.  */
1889 	if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE) {
1890 		WG_TRACE("peer sent spurious handshake response, ignoring");
1891 		goto out;
1892 	}
1893 
1894 	if (__predict_false(wg_is_underload(wg, wgp, WG_MSG_TYPE_RESP))) {
1895 		WG_TRACE("under load");
1896 		/*
1897 		 * [W] 5.3: Denial of Service Mitigation & Cookies
1898 		 * "the responder, ..., and when under load may reject messages
1899 		 *  with an invalid msg.mac2.  If the responder receives a
1900 		 *  message with a valid msg.mac1 yet with an invalid msg.mac2,
1901 		 *  and is under load, it may respond with a cookie reply
1902 		 *  message"
1903 		 */
1904 		uint8_t zero[WG_MAC_LEN] = {0};
1905 		if (consttime_memequal(wgmr->wgmr_mac2, zero, sizeof(zero))) {
1906 			WG_TRACE("sending a cookie message: no cookie included");
1907 			(void)wg_send_cookie_msg(wg, wgp, wgmr->wgmr_sender,
1908 			    wgmr->wgmr_mac1, src);
1909 			goto out;
1910 		}
1911 		if (!wgp->wgp_last_sent_cookie_valid) {
1912 			WG_TRACE("sending a cookie message: no cookie sent ever");
1913 			(void)wg_send_cookie_msg(wg, wgp, wgmr->wgmr_sender,
1914 			    wgmr->wgmr_mac1, src);
1915 			goto out;
1916 		}
1917 		uint8_t mac2[WG_MAC_LEN];
1918 		wg_algo_mac(mac2, sizeof(mac2), wgp->wgp_last_sent_cookie,
1919 		    WG_COOKIE_LEN, (const uint8_t *)wgmr,
1920 		    offsetof(struct wg_msg_resp, wgmr_mac2), NULL, 0);
1921 		if (!consttime_memequal(mac2, wgmr->wgmr_mac2, sizeof(mac2))) {
1922 			WG_DLOG("mac2 is invalid\n");
1923 			goto out;
1924 		}
1925 		WG_TRACE("under load, but continue to sending");
1926 	}
1927 
1928 	memcpy(hash, wgs->wgs_handshake_hash, sizeof(hash));
1929 	memcpy(ckey, wgs->wgs_chaining_key, sizeof(ckey));
1930 
1931 	/*
1932 	 * [W] 5.4.3 Second Message: Responder to Initiator
1933 	 * "When the initiator receives this message, it does the same
1934 	 *  operations so that its final state variables are identical,
1935 	 *  replacing the operands of the DH function to produce equivalent
1936 	 *  values."
1937 	 *  Note that the following comments of operations are just copies of
1938 	 *  the initiator's ones.
1939 	 */
1940 
1941 	/* [N] 2.2: "e" */
1942 	/* Cr := KDF1(Cr, Er^pub) */
1943 	wg_algo_kdf(ckey, NULL, NULL, ckey, wgmr->wgmr_ephemeral,
1944 	    sizeof(wgmr->wgmr_ephemeral));
1945 	/* Hr := HASH(Hr || msg.ephemeral) */
1946 	wg_algo_hash(hash, wgmr->wgmr_ephemeral, sizeof(wgmr->wgmr_ephemeral));
1947 
1948 	WG_DUMP_HASH("ckey", ckey);
1949 	WG_DUMP_HASH("hash", hash);
1950 
1951 	/* [N] 2.2: "ee" */
1952 	/* Cr := KDF1(Cr, DH(Er^priv, Ei^pub)) */
1953 	wg_algo_dh_kdf(ckey, NULL, wgs->wgs_ephemeral_key_priv,
1954 	    wgmr->wgmr_ephemeral);
1955 
1956 	/* [N] 2.2: "se" */
1957 	/* Cr := KDF1(Cr, DH(Er^priv, Si^pub)) */
1958 	wg_algo_dh_kdf(ckey, NULL, wg->wg_privkey, wgmr->wgmr_ephemeral);
1959 
1960 	/* [N] 9.2: "psk" */
1961     {
1962 	uint8_t kdfout[WG_KDF_OUTPUT_LEN];
1963 	/* Cr, r, k := KDF3(Cr, Q) */
1964 	wg_algo_kdf(ckey, kdfout, cipher_key, ckey, wgp->wgp_psk,
1965 	    sizeof(wgp->wgp_psk));
1966 	/* Hr := HASH(Hr || r) */
1967 	wg_algo_hash(hash, kdfout, sizeof(kdfout));
1968     }
1969 
1970     {
1971 	uint8_t out[sizeof(wgmr->wgmr_empty)]; /* for safety */
1972 	/* msg.empty := AEAD(k, 0, e, Hr) */
1973 	error = wg_algo_aead_dec(out, 0, cipher_key, 0, wgmr->wgmr_empty,
1974 	    sizeof(wgmr->wgmr_empty), hash, sizeof(hash));
1975 	WG_DUMP_HASH("wgmr_empty", wgmr->wgmr_empty);
1976 	if (error != 0) {
1977 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
1978 		    "wg_algo_aead_dec for empty message failed\n");
1979 		goto out;
1980 	}
1981 	/* Hr := HASH(Hr || msg.empty) */
1982 	wg_algo_hash(hash, wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty));
1983     }
1984 
1985 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(wgs->wgs_handshake_hash));
1986 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(wgs->wgs_chaining_key));
1987 	wgs->wgs_remote_index = wgmr->wgmr_sender;
1988 	WG_DLOG("receiver=%x\n", wgs->wgs_remote_index);
1989 
1990 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_ACTIVE);
1991 	wgs->wgs_state = WGS_STATE_ESTABLISHED;
1992 	wgs->wgs_time_established = time_uptime;
1993 	wgs->wgs_time_last_data_sent = 0;
1994 	wgs->wgs_is_initiator = true;
1995 	wg_calculate_keys(wgs, true);
1996 	wg_clear_states(wgs);
1997 	WG_TRACE("WGS_STATE_ESTABLISHED");
1998 
1999 	callout_stop(&wgp->wgp_handshake_timeout_timer);
2000 
2001 	wg_swap_sessions(wgp);
2002 	KASSERT(wgs == wgp->wgp_session_stable);
2003 	wgs_prev = wgp->wgp_session_unstable;
2004 	getnanotime(&wgp->wgp_last_handshake_time);
2005 	wgp->wgp_handshake_start_time = 0;
2006 	wgp->wgp_last_sent_mac1_valid = false;
2007 	wgp->wgp_last_sent_cookie_valid = false;
2008 
2009 	wg_schedule_rekey_timer(wgp);
2010 
2011 	wg_update_endpoint_if_necessary(wgp, src);
2012 
2013 	/*
2014 	 * If we had a data packet queued up, send it; otherwise send a
2015 	 * keepalive message -- either way we have to send something
2016 	 * immediately or else the responder will never answer.
2017 	 */
2018 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL) {
2019 		kpreempt_disable();
2020 		const uint32_t h = curcpu()->ci_index; // pktq_rps_hash(m)
2021 		M_SETCTX(m, wgp);
2022 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
2023 			WGLOG(LOG_ERR, "pktq full, dropping\n");
2024 			m_freem(m);
2025 		}
2026 		kpreempt_enable();
2027 	} else {
2028 		wg_send_keepalive_msg(wgp, wgs);
2029 	}
2030 
2031 	if (wgs_prev->wgs_state == WGS_STATE_ESTABLISHED) {
2032 		/* Wait for wg_get_stable_session to drain.  */
2033 		pserialize_perform(wgp->wgp_psz);
2034 
2035 		/* Transition ESTABLISHED->DESTROYING.  */
2036 		wgs_prev->wgs_state = WGS_STATE_DESTROYING;
2037 
2038 		/* We can't destroy the old session immediately */
2039 		wg_schedule_session_dtor_timer(wgp);
2040 	} else {
2041 		KASSERTMSG(wgs_prev->wgs_state == WGS_STATE_UNKNOWN,
2042 		    "state=%d", wgs_prev->wgs_state);
2043 	}
2044 
2045 out:
2046 	mutex_exit(wgp->wgp_lock);
2047 	wg_put_session(wgs, &psref);
2048 }
2049 
2050 static int
2051 wg_send_handshake_msg_resp(struct wg_softc *wg, struct wg_peer *wgp,
2052     struct wg_session *wgs, const struct wg_msg_init *wgmi)
2053 {
2054 	int error;
2055 	struct mbuf *m;
2056 	struct wg_msg_resp *wgmr;
2057 
2058 	KASSERT(mutex_owned(wgp->wgp_lock));
2059 	KASSERT(wgs == wgp->wgp_session_unstable);
2060 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_PASSIVE);
2061 
2062 	m = m_gethdr(M_WAIT, MT_DATA);
2063 	if (sizeof(*wgmr) > MHLEN) {
2064 		m_clget(m, M_WAIT);
2065 		CTASSERT(sizeof(*wgmr) <= MCLBYTES);
2066 	}
2067 	m->m_pkthdr.len = m->m_len = sizeof(*wgmr);
2068 	wgmr = mtod(m, struct wg_msg_resp *);
2069 	wg_fill_msg_resp(wg, wgp, wgs, wgmr, wgmi);
2070 
2071 	error = wg->wg_ops->send_hs_msg(wgp, m);
2072 	if (error == 0)
2073 		WG_TRACE("resp msg sent");
2074 	return error;
2075 }
2076 
2077 static struct wg_peer *
2078 wg_lookup_peer_by_pubkey(struct wg_softc *wg,
2079     const uint8_t pubkey[WG_STATIC_KEY_LEN], struct psref *psref)
2080 {
2081 	struct wg_peer *wgp;
2082 
2083 	int s = pserialize_read_enter();
2084 	wgp = thmap_get(wg->wg_peers_bypubkey, pubkey, WG_STATIC_KEY_LEN);
2085 	if (wgp != NULL)
2086 		wg_get_peer(wgp, psref);
2087 	pserialize_read_exit(s);
2088 
2089 	return wgp;
2090 }
2091 
2092 static void
2093 wg_fill_msg_cookie(struct wg_softc *wg, struct wg_peer *wgp,
2094     struct wg_msg_cookie *wgmc, const uint32_t sender,
2095     const uint8_t mac1[WG_MAC_LEN], const struct sockaddr *src)
2096 {
2097 	uint8_t cookie[WG_COOKIE_LEN];
2098 	uint8_t key[WG_HASH_LEN];
2099 	uint8_t addr[sizeof(struct in6_addr)];
2100 	size_t addrlen;
2101 	uint16_t uh_sport; /* be */
2102 
2103 	KASSERT(mutex_owned(wgp->wgp_lock));
2104 
2105 	wgmc->wgmc_type = htole32(WG_MSG_TYPE_COOKIE);
2106 	wgmc->wgmc_receiver = sender;
2107 	cprng_fast(wgmc->wgmc_salt, sizeof(wgmc->wgmc_salt));
2108 
2109 	/*
2110 	 * [W] 5.4.7: Under Load: Cookie Reply Message
2111 	 * "The secret variable, Rm, changes every two minutes to a
2112 	 * random value"
2113 	 */
2114 	if ((time_uptime - wgp->wgp_last_genrandval_time) > WG_RANDVAL_TIME) {
2115 		wgp->wgp_randval = cprng_strong32();
2116 		wgp->wgp_last_genrandval_time = time_uptime;
2117 	}
2118 
2119 	switch (src->sa_family) {
2120 	case AF_INET: {
2121 		const struct sockaddr_in *sin = satocsin(src);
2122 		addrlen = sizeof(sin->sin_addr);
2123 		memcpy(addr, &sin->sin_addr, addrlen);
2124 		uh_sport = sin->sin_port;
2125 		break;
2126 	    }
2127 #ifdef INET6
2128 	case AF_INET6: {
2129 		const struct sockaddr_in6 *sin6 = satocsin6(src);
2130 		addrlen = sizeof(sin6->sin6_addr);
2131 		memcpy(addr, &sin6->sin6_addr, addrlen);
2132 		uh_sport = sin6->sin6_port;
2133 		break;
2134 	    }
2135 #endif
2136 	default:
2137 		panic("invalid af=%d", src->sa_family);
2138 	}
2139 
2140 	wg_algo_mac(cookie, sizeof(cookie),
2141 	    (const uint8_t *)&wgp->wgp_randval, sizeof(wgp->wgp_randval),
2142 	    addr, addrlen, (const uint8_t *)&uh_sport, sizeof(uh_sport));
2143 	wg_algo_mac_cookie(key, sizeof(key), wg->wg_pubkey,
2144 	    sizeof(wg->wg_pubkey));
2145 	wg_algo_xaead_enc(wgmc->wgmc_cookie, sizeof(wgmc->wgmc_cookie), key,
2146 	    cookie, sizeof(cookie), mac1, WG_MAC_LEN, wgmc->wgmc_salt);
2147 
2148 	/* Need to store to calculate mac2 */
2149 	memcpy(wgp->wgp_last_sent_cookie, cookie, sizeof(cookie));
2150 	wgp->wgp_last_sent_cookie_valid = true;
2151 }
2152 
2153 static int
2154 wg_send_cookie_msg(struct wg_softc *wg, struct wg_peer *wgp,
2155     const uint32_t sender, const uint8_t mac1[WG_MAC_LEN],
2156     const struct sockaddr *src)
2157 {
2158 	int error;
2159 	struct mbuf *m;
2160 	struct wg_msg_cookie *wgmc;
2161 
2162 	KASSERT(mutex_owned(wgp->wgp_lock));
2163 
2164 	m = m_gethdr(M_WAIT, MT_DATA);
2165 	if (sizeof(*wgmc) > MHLEN) {
2166 		m_clget(m, M_WAIT);
2167 		CTASSERT(sizeof(*wgmc) <= MCLBYTES);
2168 	}
2169 	m->m_pkthdr.len = m->m_len = sizeof(*wgmc);
2170 	wgmc = mtod(m, struct wg_msg_cookie *);
2171 	wg_fill_msg_cookie(wg, wgp, wgmc, sender, mac1, src);
2172 
2173 	error = wg->wg_ops->send_hs_msg(wgp, m);
2174 	if (error == 0)
2175 		WG_TRACE("cookie msg sent");
2176 	return error;
2177 }
2178 
2179 static bool
2180 wg_is_underload(struct wg_softc *wg, struct wg_peer *wgp, int msgtype)
2181 {
2182 #ifdef WG_DEBUG_PARAMS
2183 	if (wg_force_underload)
2184 		return true;
2185 #endif
2186 
2187 	/*
2188 	 * XXX we don't have a means of a load estimation.  The purpose of
2189 	 * the mechanism is a DoS mitigation, so we consider frequent handshake
2190 	 * messages as (a kind of) load; if a message of the same type comes
2191 	 * to a peer within 1 second, we consider we are under load.
2192 	 */
2193 	time_t last = wgp->wgp_last_msg_received_time[msgtype];
2194 	wgp->wgp_last_msg_received_time[msgtype] = time_uptime;
2195 	return (time_uptime - last) == 0;
2196 }
2197 
2198 static void
2199 wg_calculate_keys(struct wg_session *wgs, const bool initiator)
2200 {
2201 
2202 	KASSERT(mutex_owned(wgs->wgs_peer->wgp_lock));
2203 
2204 	/*
2205 	 * [W] 5.4.5: Ti^send = Tr^recv, Ti^recv = Tr^send := KDF2(Ci = Cr, e)
2206 	 */
2207 	if (initiator) {
2208 		wg_algo_kdf(wgs->wgs_tkey_send, wgs->wgs_tkey_recv, NULL,
2209 		    wgs->wgs_chaining_key, NULL, 0);
2210 	} else {
2211 		wg_algo_kdf(wgs->wgs_tkey_recv, wgs->wgs_tkey_send, NULL,
2212 		    wgs->wgs_chaining_key, NULL, 0);
2213 	}
2214 	WG_DUMP_HASH("wgs_tkey_send", wgs->wgs_tkey_send);
2215 	WG_DUMP_HASH("wgs_tkey_recv", wgs->wgs_tkey_recv);
2216 }
2217 
2218 static uint64_t
2219 wg_session_get_send_counter(struct wg_session *wgs)
2220 {
2221 #ifdef __HAVE_ATOMIC64_LOADSTORE
2222 	return atomic_load_relaxed(&wgs->wgs_send_counter);
2223 #else
2224 	uint64_t send_counter;
2225 
2226 	mutex_enter(&wgs->wgs_send_counter_lock);
2227 	send_counter = wgs->wgs_send_counter;
2228 	mutex_exit(&wgs->wgs_send_counter_lock);
2229 
2230 	return send_counter;
2231 #endif
2232 }
2233 
2234 static uint64_t
2235 wg_session_inc_send_counter(struct wg_session *wgs)
2236 {
2237 #ifdef __HAVE_ATOMIC64_LOADSTORE
2238 	return atomic_inc_64_nv(&wgs->wgs_send_counter) - 1;
2239 #else
2240 	uint64_t send_counter;
2241 
2242 	mutex_enter(&wgs->wgs_send_counter_lock);
2243 	send_counter = wgs->wgs_send_counter++;
2244 	mutex_exit(&wgs->wgs_send_counter_lock);
2245 
2246 	return send_counter;
2247 #endif
2248 }
2249 
2250 static void
2251 wg_clear_states(struct wg_session *wgs)
2252 {
2253 
2254 	KASSERT(mutex_owned(wgs->wgs_peer->wgp_lock));
2255 
2256 	wgs->wgs_send_counter = 0;
2257 	sliwin_reset(&wgs->wgs_recvwin->window);
2258 
2259 #define wgs_clear(v)	explicit_memset(wgs->wgs_##v, 0, sizeof(wgs->wgs_##v))
2260 	wgs_clear(handshake_hash);
2261 	wgs_clear(chaining_key);
2262 	wgs_clear(ephemeral_key_pub);
2263 	wgs_clear(ephemeral_key_priv);
2264 	wgs_clear(ephemeral_key_peer);
2265 #undef wgs_clear
2266 }
2267 
2268 static struct wg_session *
2269 wg_lookup_session_by_index(struct wg_softc *wg, const uint32_t index,
2270     struct psref *psref)
2271 {
2272 	struct wg_session *wgs;
2273 
2274 	int s = pserialize_read_enter();
2275 	wgs = thmap_get(wg->wg_sessions_byindex, &index, sizeof index);
2276 	if (wgs != NULL) {
2277 		KASSERT(atomic_load_relaxed(&wgs->wgs_state) !=
2278 		    WGS_STATE_UNKNOWN);
2279 		psref_acquire(psref, &wgs->wgs_psref, wg_psref_class);
2280 	}
2281 	pserialize_read_exit(s);
2282 
2283 	return wgs;
2284 }
2285 
2286 static void
2287 wg_schedule_rekey_timer(struct wg_peer *wgp)
2288 {
2289 	int timeout = MIN(wg_rekey_after_time, (unsigned)(INT_MAX / hz));
2290 
2291 	callout_schedule(&wgp->wgp_rekey_timer, timeout * hz);
2292 }
2293 
2294 static void
2295 wg_send_keepalive_msg(struct wg_peer *wgp, struct wg_session *wgs)
2296 {
2297 	struct mbuf *m;
2298 
2299 	/*
2300 	 * [W] 6.5 Passive Keepalive
2301 	 * "A keepalive message is simply a transport data message with
2302 	 *  a zero-length encapsulated encrypted inner-packet."
2303 	 */
2304 	m = m_gethdr(M_WAIT, MT_DATA);
2305 	wg_send_data_msg(wgp, wgs, m);
2306 }
2307 
2308 static bool
2309 wg_need_to_send_init_message(struct wg_session *wgs)
2310 {
2311 	/*
2312 	 * [W] 6.2 Transport Message Limits
2313 	 * "if a peer is the initiator of a current secure session,
2314 	 *  WireGuard will send a handshake initiation message to begin
2315 	 *  a new secure session ... if after receiving a transport data
2316 	 *  message, the current secure session is (REJECT-AFTER-TIME −
2317 	 *  KEEPALIVE-TIMEOUT − REKEY-TIMEOUT) seconds old and it has
2318 	 *  not yet acted upon this event."
2319 	 */
2320 	return wgs->wgs_is_initiator && wgs->wgs_time_last_data_sent == 0 &&
2321 	    (time_uptime - wgs->wgs_time_established) >=
2322 	    (wg_reject_after_time - wg_keepalive_timeout - wg_rekey_timeout);
2323 }
2324 
2325 static void
2326 wg_schedule_peer_task(struct wg_peer *wgp, unsigned int task)
2327 {
2328 
2329 	mutex_enter(wgp->wgp_intr_lock);
2330 	WG_DLOG("tasks=%d, task=%d\n", wgp->wgp_tasks, task);
2331 	if (wgp->wgp_tasks == 0)
2332 		/*
2333 		 * XXX If the current CPU is already loaded -- e.g., if
2334 		 * there's already a bunch of handshakes queued up --
2335 		 * consider tossing this over to another CPU to
2336 		 * distribute the load.
2337 		 */
2338 		workqueue_enqueue(wg_wq, &wgp->wgp_work, NULL);
2339 	wgp->wgp_tasks |= task;
2340 	mutex_exit(wgp->wgp_intr_lock);
2341 }
2342 
2343 static void
2344 wg_change_endpoint(struct wg_peer *wgp, const struct sockaddr *new)
2345 {
2346 	struct wg_sockaddr *wgsa_prev;
2347 
2348 	WG_TRACE("Changing endpoint");
2349 
2350 	memcpy(wgp->wgp_endpoint0, new, new->sa_len);
2351 	wgsa_prev = wgp->wgp_endpoint;
2352 	atomic_store_release(&wgp->wgp_endpoint, wgp->wgp_endpoint0);
2353 	wgp->wgp_endpoint0 = wgsa_prev;
2354 	atomic_store_release(&wgp->wgp_endpoint_available, true);
2355 
2356 	wg_schedule_peer_task(wgp, WGP_TASK_ENDPOINT_CHANGED);
2357 }
2358 
2359 static bool
2360 wg_validate_inner_packet(const char *packet, size_t decrypted_len, int *af)
2361 {
2362 	uint16_t packet_len;
2363 	const struct ip *ip;
2364 
2365 	if (__predict_false(decrypted_len < sizeof(struct ip)))
2366 		return false;
2367 
2368 	ip = (const struct ip *)packet;
2369 	if (ip->ip_v == 4)
2370 		*af = AF_INET;
2371 	else if (ip->ip_v == 6)
2372 		*af = AF_INET6;
2373 	else
2374 		return false;
2375 
2376 	WG_DLOG("af=%d\n", *af);
2377 
2378 	switch (*af) {
2379 #ifdef INET
2380 	case AF_INET:
2381 		packet_len = ntohs(ip->ip_len);
2382 		break;
2383 #endif
2384 #ifdef INET6
2385 	case AF_INET6: {
2386 		const struct ip6_hdr *ip6;
2387 
2388 		if (__predict_false(decrypted_len < sizeof(struct ip6_hdr)))
2389 			return false;
2390 
2391 		ip6 = (const struct ip6_hdr *)packet;
2392 		packet_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2393 		break;
2394 	}
2395 #endif
2396 	default:
2397 		return false;
2398 	}
2399 
2400 	WG_DLOG("packet_len=%u\n", packet_len);
2401 	if (packet_len > decrypted_len)
2402 		return false;
2403 
2404 	return true;
2405 }
2406 
2407 static bool
2408 wg_validate_route(struct wg_softc *wg, struct wg_peer *wgp_expected,
2409     int af, char *packet)
2410 {
2411 	struct sockaddr_storage ss;
2412 	struct sockaddr *sa;
2413 	struct psref psref;
2414 	struct wg_peer *wgp;
2415 	bool ok;
2416 
2417 	/*
2418 	 * II CRYPTOKEY ROUTING
2419 	 * "it will only accept it if its source IP resolves in the
2420 	 *  table to the public key used in the secure session for
2421 	 *  decrypting it."
2422 	 */
2423 
2424 	if (af == AF_INET) {
2425 		const struct ip *ip = (const struct ip *)packet;
2426 		struct sockaddr_in *sin = (struct sockaddr_in *)&ss;
2427 		sockaddr_in_init(sin, &ip->ip_src, 0);
2428 		sa = sintosa(sin);
2429 #ifdef INET6
2430 	} else {
2431 		const struct ip6_hdr *ip6 = (const struct ip6_hdr *)packet;
2432 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ss;
2433 		sockaddr_in6_init(sin6, &ip6->ip6_src, 0, 0, 0);
2434 		sa = sin6tosa(sin6);
2435 #endif
2436 	}
2437 
2438 	wgp = wg_pick_peer_by_sa(wg, sa, &psref);
2439 	ok = (wgp == wgp_expected);
2440 	if (wgp != NULL)
2441 		wg_put_peer(wgp, &psref);
2442 
2443 	return ok;
2444 }
2445 
2446 static void
2447 wg_session_dtor_timer(void *arg)
2448 {
2449 	struct wg_peer *wgp = arg;
2450 
2451 	WG_TRACE("enter");
2452 
2453 	wg_schedule_peer_task(wgp, WGP_TASK_DESTROY_PREV_SESSION);
2454 }
2455 
2456 static void
2457 wg_schedule_session_dtor_timer(struct wg_peer *wgp)
2458 {
2459 
2460 	/* 1 second grace period */
2461 	callout_schedule(&wgp->wgp_session_dtor_timer, hz);
2462 }
2463 
2464 static bool
2465 sockaddr_port_match(const struct sockaddr *sa1, const struct sockaddr *sa2)
2466 {
2467 	if (sa1->sa_family != sa2->sa_family)
2468 		return false;
2469 
2470 	switch (sa1->sa_family) {
2471 #ifdef INET
2472 	case AF_INET:
2473 		return satocsin(sa1)->sin_port == satocsin(sa2)->sin_port;
2474 #endif
2475 #ifdef INET6
2476 	case AF_INET6:
2477 		return satocsin6(sa1)->sin6_port == satocsin6(sa2)->sin6_port;
2478 #endif
2479 	default:
2480 		return false;
2481 	}
2482 }
2483 
2484 static void
2485 wg_update_endpoint_if_necessary(struct wg_peer *wgp,
2486     const struct sockaddr *src)
2487 {
2488 	struct wg_sockaddr *wgsa;
2489 	struct psref psref;
2490 
2491 	wgsa = wg_get_endpoint_sa(wgp, &psref);
2492 
2493 #ifdef WG_DEBUG_LOG
2494 	char oldaddr[128], newaddr[128];
2495 	sockaddr_format(wgsatosa(wgsa), oldaddr, sizeof(oldaddr));
2496 	sockaddr_format(src, newaddr, sizeof(newaddr));
2497 	WG_DLOG("old=%s, new=%s\n", oldaddr, newaddr);
2498 #endif
2499 
2500 	/*
2501 	 * III: "Since the packet has authenticated correctly, the source IP of
2502 	 * the outer UDP/IP packet is used to update the endpoint for peer..."
2503 	 */
2504 	if (__predict_false(sockaddr_cmp(src, wgsatosa(wgsa)) != 0 ||
2505 		!sockaddr_port_match(src, wgsatosa(wgsa)))) {
2506 		/* XXX We can't change the endpoint twice in a short period */
2507 		if (atomic_swap_uint(&wgp->wgp_endpoint_changing, 1) == 0) {
2508 			wg_change_endpoint(wgp, src);
2509 		}
2510 	}
2511 
2512 	wg_put_sa(wgp, wgsa, &psref);
2513 }
2514 
2515 static void __noinline
2516 wg_handle_msg_data(struct wg_softc *wg, struct mbuf *m,
2517     const struct sockaddr *src)
2518 {
2519 	struct wg_msg_data *wgmd;
2520 	char *encrypted_buf = NULL, *decrypted_buf;
2521 	size_t encrypted_len, decrypted_len;
2522 	struct wg_session *wgs;
2523 	struct wg_peer *wgp;
2524 	int state;
2525 	size_t mlen;
2526 	struct psref psref;
2527 	int error, af;
2528 	bool success, free_encrypted_buf = false, ok;
2529 	struct mbuf *n;
2530 
2531 	KASSERT(m->m_len >= sizeof(struct wg_msg_data));
2532 	wgmd = mtod(m, struct wg_msg_data *);
2533 
2534 	KASSERT(wgmd->wgmd_type == htole32(WG_MSG_TYPE_DATA));
2535 	WG_TRACE("data");
2536 
2537 	/* Find the putative session, or drop.  */
2538 	wgs = wg_lookup_session_by_index(wg, wgmd->wgmd_receiver, &psref);
2539 	if (wgs == NULL) {
2540 		WG_TRACE("No session found");
2541 		m_freem(m);
2542 		return;
2543 	}
2544 
2545 	/*
2546 	 * We are only ready to handle data when in INIT_PASSIVE,
2547 	 * ESTABLISHED, or DESTROYING.  All transitions out of that
2548 	 * state dissociate the session index and drain psrefs.
2549 	 */
2550 	state = atomic_load_relaxed(&wgs->wgs_state);
2551 	switch (state) {
2552 	case WGS_STATE_UNKNOWN:
2553 		panic("wg session %p in unknown state has session index %u",
2554 		    wgs, wgmd->wgmd_receiver);
2555 	case WGS_STATE_INIT_ACTIVE:
2556 		WG_TRACE("not yet ready for data");
2557 		goto out;
2558 	case WGS_STATE_INIT_PASSIVE:
2559 	case WGS_STATE_ESTABLISHED:
2560 	case WGS_STATE_DESTROYING:
2561 		break;
2562 	}
2563 
2564 	/*
2565 	 * Get the peer, for rate-limited logs (XXX MPSAFE, dtrace) and
2566 	 * to update the endpoint if authentication succeeds.
2567 	 */
2568 	wgp = wgs->wgs_peer;
2569 
2570 	/*
2571 	 * Reject outrageously wrong sequence numbers before doing any
2572 	 * crypto work or taking any locks.
2573 	 */
2574 	error = sliwin_check_fast(&wgs->wgs_recvwin->window,
2575 	    le64toh(wgmd->wgmd_counter));
2576 	if (error) {
2577 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
2578 		    "out-of-window packet: %"PRIu64"\n",
2579 		    le64toh(wgmd->wgmd_counter));
2580 		goto out;
2581 	}
2582 
2583 	/* Ensure the payload and authenticator are contiguous.  */
2584 	mlen = m_length(m);
2585 	encrypted_len = mlen - sizeof(*wgmd);
2586 	if (encrypted_len < WG_AUTHTAG_LEN) {
2587 		WG_DLOG("Short encrypted_len: %lu\n", encrypted_len);
2588 		goto out;
2589 	}
2590 	success = m_ensure_contig(&m, sizeof(*wgmd) + encrypted_len);
2591 	if (success) {
2592 		encrypted_buf = mtod(m, char *) + sizeof(*wgmd);
2593 	} else {
2594 		encrypted_buf = kmem_intr_alloc(encrypted_len, KM_NOSLEEP);
2595 		if (encrypted_buf == NULL) {
2596 			WG_DLOG("failed to allocate encrypted_buf\n");
2597 			goto out;
2598 		}
2599 		m_copydata(m, sizeof(*wgmd), encrypted_len, encrypted_buf);
2600 		free_encrypted_buf = true;
2601 	}
2602 	/* m_ensure_contig may change m regardless of its result */
2603 	KASSERT(m->m_len >= sizeof(*wgmd));
2604 	wgmd = mtod(m, struct wg_msg_data *);
2605 
2606 	/*
2607 	 * Get a buffer for the plaintext.  Add WG_AUTHTAG_LEN to avoid
2608 	 * a zero-length buffer (XXX).  Drop if plaintext is longer
2609 	 * than MCLBYTES (XXX).
2610 	 */
2611 	decrypted_len = encrypted_len - WG_AUTHTAG_LEN;
2612 	if (decrypted_len > MCLBYTES) {
2613 		/* FIXME handle larger data than MCLBYTES */
2614 		WG_DLOG("couldn't handle larger data than MCLBYTES\n");
2615 		goto out;
2616 	}
2617 	n = wg_get_mbuf(0, decrypted_len + WG_AUTHTAG_LEN);
2618 	if (n == NULL) {
2619 		WG_DLOG("wg_get_mbuf failed\n");
2620 		goto out;
2621 	}
2622 	decrypted_buf = mtod(n, char *);
2623 
2624 	/* Decrypt and verify the packet.  */
2625 	WG_DLOG("mlen=%lu, encrypted_len=%lu\n", mlen, encrypted_len);
2626 	error = wg_algo_aead_dec(decrypted_buf,
2627 	    encrypted_len - WG_AUTHTAG_LEN /* can be 0 */,
2628 	    wgs->wgs_tkey_recv, le64toh(wgmd->wgmd_counter), encrypted_buf,
2629 	    encrypted_len, NULL, 0);
2630 	if (error != 0) {
2631 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
2632 		    "failed to wg_algo_aead_dec\n");
2633 		m_freem(n);
2634 		goto out;
2635 	}
2636 	WG_DLOG("outsize=%u\n", (u_int)decrypted_len);
2637 
2638 	/* Packet is genuine.  Reject it if a replay or just too old.  */
2639 	mutex_enter(&wgs->wgs_recvwin->lock);
2640 	error = sliwin_update(&wgs->wgs_recvwin->window,
2641 	    le64toh(wgmd->wgmd_counter));
2642 	mutex_exit(&wgs->wgs_recvwin->lock);
2643 	if (error) {
2644 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
2645 		    "replay or out-of-window packet: %"PRIu64"\n",
2646 		    le64toh(wgmd->wgmd_counter));
2647 		m_freem(n);
2648 		goto out;
2649 	}
2650 
2651 	/* We're done with m now; free it and chuck the pointers.  */
2652 	m_freem(m);
2653 	m = NULL;
2654 	wgmd = NULL;
2655 
2656 	/*
2657 	 * Validate the encapsulated packet header and get the address
2658 	 * family, or drop.
2659 	 */
2660 	ok = wg_validate_inner_packet(decrypted_buf, decrypted_len, &af);
2661 	if (!ok) {
2662 		m_freem(n);
2663 		goto out;
2664 	}
2665 
2666 	/*
2667 	 * The packet is genuine.  Update the peer's endpoint if the
2668 	 * source address changed.
2669 	 *
2670 	 * XXX How to prevent DoS by replaying genuine packets from the
2671 	 * wrong source address?
2672 	 */
2673 	wg_update_endpoint_if_necessary(wgp, src);
2674 
2675 	/* Submit it into our network stack if routable.  */
2676 	ok = wg_validate_route(wg, wgp, af, decrypted_buf);
2677 	if (ok) {
2678 		wg->wg_ops->input(&wg->wg_if, n, af);
2679 	} else {
2680 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
2681 		    "invalid source address\n");
2682 		m_freem(n);
2683 		/*
2684 		 * The inner address is invalid however the session is valid
2685 		 * so continue the session processing below.
2686 		 */
2687 	}
2688 	n = NULL;
2689 
2690 	/* Update the state machine if necessary.  */
2691 	if (__predict_false(state == WGS_STATE_INIT_PASSIVE)) {
2692 		/*
2693 		 * We were waiting for the initiator to send their
2694 		 * first data transport message, and that has happened.
2695 		 * Schedule a task to establish this session.
2696 		 */
2697 		wg_schedule_peer_task(wgp, WGP_TASK_ESTABLISH_SESSION);
2698 	} else {
2699 		if (__predict_false(wg_need_to_send_init_message(wgs))) {
2700 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
2701 		}
2702 		/*
2703 		 * [W] 6.5 Passive Keepalive
2704 		 * "If a peer has received a validly-authenticated transport
2705 		 *  data message (section 5.4.6), but does not have any packets
2706 		 *  itself to send back for KEEPALIVE-TIMEOUT seconds, it sends
2707 		 *  a keepalive message."
2708 		 */
2709 		WG_DLOG("time_uptime=%ju wgs_time_last_data_sent=%ju\n",
2710 		    (uintmax_t)time_uptime,
2711 		    (uintmax_t)wgs->wgs_time_last_data_sent);
2712 		if ((time_uptime - wgs->wgs_time_last_data_sent) >=
2713 		    wg_keepalive_timeout) {
2714 			WG_TRACE("Schedule sending keepalive message");
2715 			/*
2716 			 * We can't send a keepalive message here to avoid
2717 			 * a deadlock;  we already hold the solock of a socket
2718 			 * that is used to send the message.
2719 			 */
2720 			wg_schedule_peer_task(wgp,
2721 			    WGP_TASK_SEND_KEEPALIVE_MESSAGE);
2722 		}
2723 	}
2724 out:
2725 	wg_put_session(wgs, &psref);
2726 	if (m != NULL)
2727 		m_freem(m);
2728 	if (free_encrypted_buf)
2729 		kmem_intr_free(encrypted_buf, encrypted_len);
2730 }
2731 
2732 static void __noinline
2733 wg_handle_msg_cookie(struct wg_softc *wg, const struct wg_msg_cookie *wgmc)
2734 {
2735 	struct wg_session *wgs;
2736 	struct wg_peer *wgp;
2737 	struct psref psref;
2738 	int error;
2739 	uint8_t key[WG_HASH_LEN];
2740 	uint8_t cookie[WG_COOKIE_LEN];
2741 
2742 	WG_TRACE("cookie msg received");
2743 
2744 	/* Find the putative session.  */
2745 	wgs = wg_lookup_session_by_index(wg, wgmc->wgmc_receiver, &psref);
2746 	if (wgs == NULL) {
2747 		WG_TRACE("No session found");
2748 		return;
2749 	}
2750 
2751 	/* Lock the peer so we can update the cookie state.  */
2752 	wgp = wgs->wgs_peer;
2753 	mutex_enter(wgp->wgp_lock);
2754 
2755 	if (!wgp->wgp_last_sent_mac1_valid) {
2756 		WG_TRACE("No valid mac1 sent (or expired)");
2757 		goto out;
2758 	}
2759 
2760 	/* Decrypt the cookie and store it for later handshake retry.  */
2761 	wg_algo_mac_cookie(key, sizeof(key), wgp->wgp_pubkey,
2762 	    sizeof(wgp->wgp_pubkey));
2763 	error = wg_algo_xaead_dec(cookie, sizeof(cookie), key,
2764 	    wgmc->wgmc_cookie, sizeof(wgmc->wgmc_cookie),
2765 	    wgp->wgp_last_sent_mac1, sizeof(wgp->wgp_last_sent_mac1),
2766 	    wgmc->wgmc_salt);
2767 	if (error != 0) {
2768 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
2769 		    "wg_algo_aead_dec for cookie failed: error=%d\n", error);
2770 		goto out;
2771 	}
2772 	/*
2773 	 * [W] 6.6: Interaction with Cookie Reply System
2774 	 * "it should simply store the decrypted cookie value from the cookie
2775 	 *  reply message, and wait for the expiration of the REKEY-TIMEOUT
2776 	 *  timer for retrying a handshake initiation message."
2777 	 */
2778 	wgp->wgp_latest_cookie_time = time_uptime;
2779 	memcpy(wgp->wgp_latest_cookie, cookie, sizeof(wgp->wgp_latest_cookie));
2780 out:
2781 	mutex_exit(wgp->wgp_lock);
2782 	wg_put_session(wgs, &psref);
2783 }
2784 
2785 static struct mbuf *
2786 wg_validate_msg_header(struct wg_softc *wg, struct mbuf *m)
2787 {
2788 	struct wg_msg wgm;
2789 	size_t mbuflen;
2790 	size_t msglen;
2791 
2792 	/*
2793 	 * Get the mbuf chain length.  It is already guaranteed, by
2794 	 * wg_overudp_cb, to be large enough for a struct wg_msg.
2795 	 */
2796 	mbuflen = m_length(m);
2797 	KASSERT(mbuflen >= sizeof(struct wg_msg));
2798 
2799 	/*
2800 	 * Copy the message header (32-bit message type) out -- we'll
2801 	 * worry about contiguity and alignment later.
2802 	 */
2803 	m_copydata(m, 0, sizeof(wgm), &wgm);
2804 	switch (le32toh(wgm.wgm_type)) {
2805 	case WG_MSG_TYPE_INIT:
2806 		msglen = sizeof(struct wg_msg_init);
2807 		break;
2808 	case WG_MSG_TYPE_RESP:
2809 		msglen = sizeof(struct wg_msg_resp);
2810 		break;
2811 	case WG_MSG_TYPE_COOKIE:
2812 		msglen = sizeof(struct wg_msg_cookie);
2813 		break;
2814 	case WG_MSG_TYPE_DATA:
2815 		msglen = sizeof(struct wg_msg_data);
2816 		break;
2817 	default:
2818 		WG_LOG_RATECHECK(&wg->wg_ppsratecheck, LOG_DEBUG,
2819 		    "Unexpected msg type: %u\n", le32toh(wgm.wgm_type));
2820 		goto error;
2821 	}
2822 
2823 	/* Verify the mbuf chain is long enough for this type of message.  */
2824 	if (__predict_false(mbuflen < msglen)) {
2825 		WG_DLOG("Invalid msg size: mbuflen=%lu type=%u\n", mbuflen,
2826 		    le32toh(wgm.wgm_type));
2827 		goto error;
2828 	}
2829 
2830 	/* Make the message header contiguous if necessary.  */
2831 	if (__predict_false(m->m_len < msglen)) {
2832 		m = m_pullup(m, msglen);
2833 		if (m == NULL)
2834 			return NULL;
2835 	}
2836 
2837 	return m;
2838 
2839 error:
2840 	m_freem(m);
2841 	return NULL;
2842 }
2843 
2844 static void
2845 wg_handle_packet(struct wg_softc *wg, struct mbuf *m,
2846     const struct sockaddr *src)
2847 {
2848 	struct wg_msg *wgm;
2849 
2850 	m = wg_validate_msg_header(wg, m);
2851 	if (__predict_false(m == NULL))
2852 		return;
2853 
2854 	KASSERT(m->m_len >= sizeof(struct wg_msg));
2855 	wgm = mtod(m, struct wg_msg *);
2856 	switch (le32toh(wgm->wgm_type)) {
2857 	case WG_MSG_TYPE_INIT:
2858 		wg_handle_msg_init(wg, (struct wg_msg_init *)wgm, src);
2859 		break;
2860 	case WG_MSG_TYPE_RESP:
2861 		wg_handle_msg_resp(wg, (struct wg_msg_resp *)wgm, src);
2862 		break;
2863 	case WG_MSG_TYPE_COOKIE:
2864 		wg_handle_msg_cookie(wg, (struct wg_msg_cookie *)wgm);
2865 		break;
2866 	case WG_MSG_TYPE_DATA:
2867 		wg_handle_msg_data(wg, m, src);
2868 		/* wg_handle_msg_data frees m for us */
2869 		return;
2870 	default:
2871 		panic("invalid message type: %d", le32toh(wgm->wgm_type));
2872 	}
2873 
2874 	m_freem(m);
2875 }
2876 
2877 static void
2878 wg_receive_packets(struct wg_softc *wg, const int af)
2879 {
2880 
2881 	for (;;) {
2882 		int error, flags;
2883 		struct socket *so;
2884 		struct mbuf *m = NULL;
2885 		struct uio dummy_uio;
2886 		struct mbuf *paddr = NULL;
2887 		struct sockaddr *src;
2888 
2889 		so = wg_get_so_by_af(wg, af);
2890 		flags = MSG_DONTWAIT;
2891 		dummy_uio.uio_resid = 1000000000;
2892 
2893 		error = so->so_receive(so, &paddr, &dummy_uio, &m, NULL,
2894 		    &flags);
2895 		if (error || m == NULL) {
2896 			//if (error == EWOULDBLOCK)
2897 			return;
2898 		}
2899 
2900 		KASSERT(paddr != NULL);
2901 		KASSERT(paddr->m_len >= sizeof(struct sockaddr));
2902 		src = mtod(paddr, struct sockaddr *);
2903 
2904 		wg_handle_packet(wg, m, src);
2905 	}
2906 }
2907 
2908 static void
2909 wg_get_peer(struct wg_peer *wgp, struct psref *psref)
2910 {
2911 
2912 	psref_acquire(psref, &wgp->wgp_psref, wg_psref_class);
2913 }
2914 
2915 static void
2916 wg_put_peer(struct wg_peer *wgp, struct psref *psref)
2917 {
2918 
2919 	psref_release(psref, &wgp->wgp_psref, wg_psref_class);
2920 }
2921 
2922 static void
2923 wg_task_send_init_message(struct wg_softc *wg, struct wg_peer *wgp)
2924 {
2925 	struct wg_session *wgs;
2926 
2927 	WG_TRACE("WGP_TASK_SEND_INIT_MESSAGE");
2928 
2929 	KASSERT(mutex_owned(wgp->wgp_lock));
2930 
2931 	if (!atomic_load_acquire(&wgp->wgp_endpoint_available)) {
2932 		WGLOG(LOG_DEBUG, "No endpoint available\n");
2933 		/* XXX should do something? */
2934 		return;
2935 	}
2936 
2937 	wgs = wgp->wgp_session_stable;
2938 	if (wgs->wgs_state == WGS_STATE_UNKNOWN) {
2939 		/* XXX What if the unstable session is already INIT_ACTIVE?  */
2940 		wg_send_handshake_msg_init(wg, wgp);
2941 	} else {
2942 		/* rekey */
2943 		wgs = wgp->wgp_session_unstable;
2944 		if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE)
2945 			wg_send_handshake_msg_init(wg, wgp);
2946 	}
2947 }
2948 
2949 static void
2950 wg_task_retry_handshake(struct wg_softc *wg, struct wg_peer *wgp)
2951 {
2952 	struct wg_session *wgs;
2953 
2954 	WG_TRACE("WGP_TASK_RETRY_HANDSHAKE");
2955 
2956 	KASSERT(mutex_owned(wgp->wgp_lock));
2957 	KASSERT(wgp->wgp_handshake_start_time != 0);
2958 
2959 	wgs = wgp->wgp_session_unstable;
2960 	if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE)
2961 		return;
2962 
2963 	/*
2964 	 * XXX no real need to assign a new index here, but we do need
2965 	 * to transition to UNKNOWN temporarily
2966 	 */
2967 	wg_put_session_index(wg, wgs);
2968 
2969 	/* [W] 6.4 Handshake Initiation Retransmission */
2970 	if ((time_uptime - wgp->wgp_handshake_start_time) >
2971 	    wg_rekey_attempt_time) {
2972 		/* Give up handshaking */
2973 		wgp->wgp_handshake_start_time = 0;
2974 		WG_TRACE("give up");
2975 
2976 		/*
2977 		 * If a new data packet comes, handshaking will be retried
2978 		 * and a new session would be established at that time,
2979 		 * however we don't want to send pending packets then.
2980 		 */
2981 		wg_purge_pending_packets(wgp);
2982 		return;
2983 	}
2984 
2985 	wg_task_send_init_message(wg, wgp);
2986 }
2987 
2988 static void
2989 wg_task_establish_session(struct wg_softc *wg, struct wg_peer *wgp)
2990 {
2991 	struct wg_session *wgs, *wgs_prev;
2992 	struct mbuf *m;
2993 
2994 	KASSERT(mutex_owned(wgp->wgp_lock));
2995 
2996 	wgs = wgp->wgp_session_unstable;
2997 	if (wgs->wgs_state != WGS_STATE_INIT_PASSIVE)
2998 		/* XXX Can this happen?  */
2999 		return;
3000 
3001 	wgs->wgs_state = WGS_STATE_ESTABLISHED;
3002 	wgs->wgs_time_established = time_uptime;
3003 	wgs->wgs_time_last_data_sent = 0;
3004 	wgs->wgs_is_initiator = false;
3005 	WG_TRACE("WGS_STATE_ESTABLISHED");
3006 
3007 	wg_swap_sessions(wgp);
3008 	KASSERT(wgs == wgp->wgp_session_stable);
3009 	wgs_prev = wgp->wgp_session_unstable;
3010 	getnanotime(&wgp->wgp_last_handshake_time);
3011 	wgp->wgp_handshake_start_time = 0;
3012 	wgp->wgp_last_sent_mac1_valid = false;
3013 	wgp->wgp_last_sent_cookie_valid = false;
3014 
3015 	/* If we had a data packet queued up, send it.  */
3016 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL) {
3017 		kpreempt_disable();
3018 		const uint32_t h = curcpu()->ci_index; // pktq_rps_hash(m)
3019 		M_SETCTX(m, wgp);
3020 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
3021 			WGLOG(LOG_ERR, "pktq full, dropping\n");
3022 			m_freem(m);
3023 		}
3024 		kpreempt_enable();
3025 	}
3026 
3027 	if (wgs_prev->wgs_state == WGS_STATE_ESTABLISHED) {
3028 		/* Wait for wg_get_stable_session to drain.  */
3029 		pserialize_perform(wgp->wgp_psz);
3030 
3031 		/* Transition ESTABLISHED->DESTROYING.  */
3032 		wgs_prev->wgs_state = WGS_STATE_DESTROYING;
3033 
3034 		/* We can't destroy the old session immediately */
3035 		wg_schedule_session_dtor_timer(wgp);
3036 	} else {
3037 		KASSERTMSG(wgs_prev->wgs_state == WGS_STATE_UNKNOWN,
3038 		    "state=%d", wgs_prev->wgs_state);
3039 		wg_clear_states(wgs_prev);
3040 		wgs_prev->wgs_state = WGS_STATE_UNKNOWN;
3041 	}
3042 }
3043 
3044 static void
3045 wg_task_endpoint_changed(struct wg_softc *wg, struct wg_peer *wgp)
3046 {
3047 
3048 	WG_TRACE("WGP_TASK_ENDPOINT_CHANGED");
3049 
3050 	KASSERT(mutex_owned(wgp->wgp_lock));
3051 
3052 	if (atomic_load_relaxed(&wgp->wgp_endpoint_changing)) {
3053 		pserialize_perform(wgp->wgp_psz);
3054 		mutex_exit(wgp->wgp_lock);
3055 		psref_target_destroy(&wgp->wgp_endpoint0->wgsa_psref,
3056 		    wg_psref_class);
3057 		psref_target_init(&wgp->wgp_endpoint0->wgsa_psref,
3058 		    wg_psref_class);
3059 		mutex_enter(wgp->wgp_lock);
3060 		atomic_store_release(&wgp->wgp_endpoint_changing, 0);
3061 	}
3062 }
3063 
3064 static void
3065 wg_task_send_keepalive_message(struct wg_softc *wg, struct wg_peer *wgp)
3066 {
3067 	struct wg_session *wgs;
3068 
3069 	WG_TRACE("WGP_TASK_SEND_KEEPALIVE_MESSAGE");
3070 
3071 	KASSERT(mutex_owned(wgp->wgp_lock));
3072 
3073 	wgs = wgp->wgp_session_stable;
3074 	if (wgs->wgs_state != WGS_STATE_ESTABLISHED)
3075 		return;
3076 
3077 	wg_send_keepalive_msg(wgp, wgs);
3078 }
3079 
3080 static void
3081 wg_task_destroy_prev_session(struct wg_softc *wg, struct wg_peer *wgp)
3082 {
3083 	struct wg_session *wgs;
3084 
3085 	WG_TRACE("WGP_TASK_DESTROY_PREV_SESSION");
3086 
3087 	KASSERT(mutex_owned(wgp->wgp_lock));
3088 
3089 	wgs = wgp->wgp_session_unstable;
3090 	if (wgs->wgs_state == WGS_STATE_DESTROYING) {
3091 		wg_put_session_index(wg, wgs);
3092 	}
3093 }
3094 
3095 static void
3096 wg_peer_work(struct work *wk, void *cookie)
3097 {
3098 	struct wg_peer *wgp = container_of(wk, struct wg_peer, wgp_work);
3099 	struct wg_softc *wg = wgp->wgp_sc;
3100 	unsigned int tasks;
3101 
3102 	mutex_enter(wgp->wgp_intr_lock);
3103 	while ((tasks = wgp->wgp_tasks) != 0) {
3104 		wgp->wgp_tasks = 0;
3105 		mutex_exit(wgp->wgp_intr_lock);
3106 
3107 		mutex_enter(wgp->wgp_lock);
3108 		if (ISSET(tasks, WGP_TASK_SEND_INIT_MESSAGE))
3109 			wg_task_send_init_message(wg, wgp);
3110 		if (ISSET(tasks, WGP_TASK_RETRY_HANDSHAKE))
3111 			wg_task_retry_handshake(wg, wgp);
3112 		if (ISSET(tasks, WGP_TASK_ESTABLISH_SESSION))
3113 			wg_task_establish_session(wg, wgp);
3114 		if (ISSET(tasks, WGP_TASK_ENDPOINT_CHANGED))
3115 			wg_task_endpoint_changed(wg, wgp);
3116 		if (ISSET(tasks, WGP_TASK_SEND_KEEPALIVE_MESSAGE))
3117 			wg_task_send_keepalive_message(wg, wgp);
3118 		if (ISSET(tasks, WGP_TASK_DESTROY_PREV_SESSION))
3119 			wg_task_destroy_prev_session(wg, wgp);
3120 		mutex_exit(wgp->wgp_lock);
3121 
3122 		mutex_enter(wgp->wgp_intr_lock);
3123 	}
3124 	mutex_exit(wgp->wgp_intr_lock);
3125 }
3126 
3127 static void
3128 wg_job(struct threadpool_job *job)
3129 {
3130 	struct wg_softc *wg = container_of(job, struct wg_softc, wg_job);
3131 	int bound, upcalls;
3132 
3133 	mutex_enter(wg->wg_intr_lock);
3134 	while ((upcalls = wg->wg_upcalls) != 0) {
3135 		wg->wg_upcalls = 0;
3136 		mutex_exit(wg->wg_intr_lock);
3137 		bound = curlwp_bind();
3138 		if (ISSET(upcalls, WG_UPCALL_INET))
3139 			wg_receive_packets(wg, AF_INET);
3140 		if (ISSET(upcalls, WG_UPCALL_INET6))
3141 			wg_receive_packets(wg, AF_INET6);
3142 		curlwp_bindx(bound);
3143 		mutex_enter(wg->wg_intr_lock);
3144 	}
3145 	threadpool_job_done(job);
3146 	mutex_exit(wg->wg_intr_lock);
3147 }
3148 
3149 static int
3150 wg_bind_port(struct wg_softc *wg, const uint16_t port)
3151 {
3152 	int error;
3153 	uint16_t old_port = wg->wg_listen_port;
3154 
3155 	if (port != 0 && old_port == port)
3156 		return 0;
3157 
3158 	struct sockaddr_in _sin, *sin = &_sin;
3159 	sin->sin_len = sizeof(*sin);
3160 	sin->sin_family = AF_INET;
3161 	sin->sin_addr.s_addr = INADDR_ANY;
3162 	sin->sin_port = htons(port);
3163 
3164 	error = sobind(wg->wg_so4, sintosa(sin), curlwp);
3165 	if (error != 0)
3166 		return error;
3167 
3168 #ifdef INET6
3169 	struct sockaddr_in6 _sin6, *sin6 = &_sin6;
3170 	sin6->sin6_len = sizeof(*sin6);
3171 	sin6->sin6_family = AF_INET6;
3172 	sin6->sin6_addr = in6addr_any;
3173 	sin6->sin6_port = htons(port);
3174 
3175 	error = sobind(wg->wg_so6, sin6tosa(sin6), curlwp);
3176 	if (error != 0)
3177 		return error;
3178 #endif
3179 
3180 	wg->wg_listen_port = port;
3181 
3182 	return 0;
3183 }
3184 
3185 static void
3186 wg_so_upcall(struct socket *so, void *cookie, int events, int waitflag)
3187 {
3188 	struct wg_softc *wg = cookie;
3189 	int reason;
3190 
3191 	reason = (so->so_proto->pr_domain->dom_family == AF_INET) ?
3192 	    WG_UPCALL_INET :
3193 	    WG_UPCALL_INET6;
3194 
3195 	mutex_enter(wg->wg_intr_lock);
3196 	wg->wg_upcalls |= reason;
3197 	threadpool_schedule_job(wg->wg_threadpool, &wg->wg_job);
3198 	mutex_exit(wg->wg_intr_lock);
3199 }
3200 
3201 static int
3202 wg_overudp_cb(struct mbuf **mp, int offset, struct socket *so,
3203     struct sockaddr *src, void *arg)
3204 {
3205 	struct wg_softc *wg = arg;
3206 	struct wg_msg wgm;
3207 	struct mbuf *m = *mp;
3208 
3209 	WG_TRACE("enter");
3210 
3211 	/* Verify the mbuf chain is long enough to have a wg msg header.  */
3212 	KASSERT(offset <= m_length(m));
3213 	if (__predict_false(m_length(m) - offset < sizeof(struct wg_msg))) {
3214 		/* drop on the floor */
3215 		m_freem(m);
3216 		return -1;
3217 	}
3218 
3219 	/*
3220 	 * Copy the message header (32-bit message type) out -- we'll
3221 	 * worry about contiguity and alignment later.
3222 	 */
3223 	m_copydata(m, offset, sizeof(struct wg_msg), &wgm);
3224 	WG_DLOG("type=%d\n", le32toh(wgm.wgm_type));
3225 
3226 	/*
3227 	 * Handle DATA packets promptly as they arrive.  Other packets
3228 	 * may require expensive public-key crypto and are not as
3229 	 * sensitive to latency, so defer them to the worker thread.
3230 	 */
3231 	switch (le32toh(wgm.wgm_type)) {
3232 	case WG_MSG_TYPE_DATA:
3233 		/* handle immediately */
3234 		m_adj(m, offset);
3235 		if (__predict_false(m->m_len < sizeof(struct wg_msg_data))) {
3236 			m = m_pullup(m, sizeof(struct wg_msg_data));
3237 			if (m == NULL)
3238 				return -1;
3239 		}
3240 		wg_handle_msg_data(wg, m, src);
3241 		*mp = NULL;
3242 		return 1;
3243 	case WG_MSG_TYPE_INIT:
3244 	case WG_MSG_TYPE_RESP:
3245 	case WG_MSG_TYPE_COOKIE:
3246 		/* pass through to so_receive in wg_receive_packets */
3247 		return 0;
3248 	default:
3249 		/* drop on the floor */
3250 		m_freem(m);
3251 		return -1;
3252 	}
3253 }
3254 
3255 static int
3256 wg_socreate(struct wg_softc *wg, int af, struct socket **sop)
3257 {
3258 	int error;
3259 	struct socket *so;
3260 
3261 	error = socreate(af, &so, SOCK_DGRAM, 0, curlwp, NULL);
3262 	if (error != 0)
3263 		return error;
3264 
3265 	solock(so);
3266 	so->so_upcallarg = wg;
3267 	so->so_upcall = wg_so_upcall;
3268 	so->so_rcv.sb_flags |= SB_UPCALL;
3269 	inpcb_register_overudp_cb(sotoinpcb(so), wg_overudp_cb, wg);
3270 	sounlock(so);
3271 
3272 	*sop = so;
3273 
3274 	return 0;
3275 }
3276 
3277 static bool
3278 wg_session_hit_limits(struct wg_session *wgs)
3279 {
3280 
3281 	/*
3282 	 * [W] 6.2: Transport Message Limits
3283 	 * "After REJECT-AFTER-MESSAGES transport data messages or after the
3284 	 *  current secure session is REJECT-AFTER-TIME seconds old, whichever
3285 	 *  comes first, WireGuard will refuse to send any more transport data
3286 	 *  messages using the current secure session, ..."
3287 	 */
3288 	KASSERT(wgs->wgs_time_established != 0);
3289 	if ((time_uptime - wgs->wgs_time_established) > wg_reject_after_time) {
3290 		WG_DLOG("The session hits REJECT_AFTER_TIME\n");
3291 		return true;
3292 	} else if (wg_session_get_send_counter(wgs) >
3293 	    wg_reject_after_messages) {
3294 		WG_DLOG("The session hits REJECT_AFTER_MESSAGES\n");
3295 		return true;
3296 	}
3297 
3298 	return false;
3299 }
3300 
3301 static void
3302 wgintr(void *cookie)
3303 {
3304 	struct wg_peer *wgp;
3305 	struct wg_session *wgs;
3306 	struct mbuf *m;
3307 	struct psref psref;
3308 
3309 	while ((m = pktq_dequeue(wg_pktq)) != NULL) {
3310 		wgp = M_GETCTX(m, struct wg_peer *);
3311 		if ((wgs = wg_get_stable_session(wgp, &psref)) == NULL) {
3312 			WG_TRACE("no stable session");
3313 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
3314 			goto next0;
3315 		}
3316 		if (__predict_false(wg_session_hit_limits(wgs))) {
3317 			WG_TRACE("stable session hit limits");
3318 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
3319 			goto next1;
3320 		}
3321 		wg_send_data_msg(wgp, wgs, m);
3322 		m = NULL;	/* consumed */
3323 next1:		wg_put_session(wgs, &psref);
3324 next0:		if (m)
3325 			m_freem(m);
3326 		/* XXX Yield to avoid userland starvation?  */
3327 	}
3328 }
3329 
3330 static void
3331 wg_rekey_timer(void *arg)
3332 {
3333 	struct wg_peer *wgp = arg;
3334 
3335 	wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
3336 }
3337 
3338 static void
3339 wg_purge_pending_packets(struct wg_peer *wgp)
3340 {
3341 	struct mbuf *m;
3342 
3343 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL)
3344 		m_freem(m);
3345 	pktq_barrier(wg_pktq);
3346 }
3347 
3348 static void
3349 wg_handshake_timeout_timer(void *arg)
3350 {
3351 	struct wg_peer *wgp = arg;
3352 
3353 	WG_TRACE("enter");
3354 
3355 	wg_schedule_peer_task(wgp, WGP_TASK_RETRY_HANDSHAKE);
3356 }
3357 
3358 static struct wg_peer *
3359 wg_alloc_peer(struct wg_softc *wg)
3360 {
3361 	struct wg_peer *wgp;
3362 
3363 	wgp = kmem_zalloc(sizeof(*wgp), KM_SLEEP);
3364 
3365 	wgp->wgp_sc = wg;
3366 	callout_init(&wgp->wgp_rekey_timer, CALLOUT_MPSAFE);
3367 	callout_setfunc(&wgp->wgp_rekey_timer, wg_rekey_timer, wgp);
3368 	callout_init(&wgp->wgp_handshake_timeout_timer, CALLOUT_MPSAFE);
3369 	callout_setfunc(&wgp->wgp_handshake_timeout_timer,
3370 	    wg_handshake_timeout_timer, wgp);
3371 	callout_init(&wgp->wgp_session_dtor_timer, CALLOUT_MPSAFE);
3372 	callout_setfunc(&wgp->wgp_session_dtor_timer,
3373 	    wg_session_dtor_timer, wgp);
3374 	PSLIST_ENTRY_INIT(wgp, wgp_peerlist_entry);
3375 	wgp->wgp_endpoint_changing = false;
3376 	wgp->wgp_endpoint_available = false;
3377 	wgp->wgp_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
3378 	wgp->wgp_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
3379 	wgp->wgp_psz = pserialize_create();
3380 	psref_target_init(&wgp->wgp_psref, wg_psref_class);
3381 
3382 	wgp->wgp_endpoint = kmem_zalloc(sizeof(*wgp->wgp_endpoint), KM_SLEEP);
3383 	wgp->wgp_endpoint0 = kmem_zalloc(sizeof(*wgp->wgp_endpoint0), KM_SLEEP);
3384 	psref_target_init(&wgp->wgp_endpoint->wgsa_psref, wg_psref_class);
3385 	psref_target_init(&wgp->wgp_endpoint0->wgsa_psref, wg_psref_class);
3386 
3387 	struct wg_session *wgs;
3388 	wgp->wgp_session_stable =
3389 	    kmem_zalloc(sizeof(*wgp->wgp_session_stable), KM_SLEEP);
3390 	wgp->wgp_session_unstable =
3391 	    kmem_zalloc(sizeof(*wgp->wgp_session_unstable), KM_SLEEP);
3392 	wgs = wgp->wgp_session_stable;
3393 	wgs->wgs_peer = wgp;
3394 	wgs->wgs_state = WGS_STATE_UNKNOWN;
3395 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
3396 #ifndef __HAVE_ATOMIC64_LOADSTORE
3397 	mutex_init(&wgs->wgs_send_counter_lock, MUTEX_DEFAULT, IPL_SOFTNET);
3398 #endif
3399 	wgs->wgs_recvwin = kmem_zalloc(sizeof(*wgs->wgs_recvwin), KM_SLEEP);
3400 	mutex_init(&wgs->wgs_recvwin->lock, MUTEX_DEFAULT, IPL_SOFTNET);
3401 
3402 	wgs = wgp->wgp_session_unstable;
3403 	wgs->wgs_peer = wgp;
3404 	wgs->wgs_state = WGS_STATE_UNKNOWN;
3405 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
3406 #ifndef __HAVE_ATOMIC64_LOADSTORE
3407 	mutex_init(&wgs->wgs_send_counter_lock, MUTEX_DEFAULT, IPL_SOFTNET);
3408 #endif
3409 	wgs->wgs_recvwin = kmem_zalloc(sizeof(*wgs->wgs_recvwin), KM_SLEEP);
3410 	mutex_init(&wgs->wgs_recvwin->lock, MUTEX_DEFAULT, IPL_SOFTNET);
3411 
3412 	return wgp;
3413 }
3414 
3415 static void
3416 wg_destroy_peer(struct wg_peer *wgp)
3417 {
3418 	struct wg_session *wgs;
3419 	struct wg_softc *wg = wgp->wgp_sc;
3420 
3421 	/* Prevent new packets from this peer on any source address.  */
3422 	rw_enter(wg->wg_rwlock, RW_WRITER);
3423 	for (int i = 0; i < wgp->wgp_n_allowedips; i++) {
3424 		struct wg_allowedip *wga = &wgp->wgp_allowedips[i];
3425 		struct radix_node_head *rnh = wg_rnh(wg, wga->wga_family);
3426 		struct radix_node *rn;
3427 
3428 		KASSERT(rnh != NULL);
3429 		rn = rnh->rnh_deladdr(&wga->wga_sa_addr,
3430 		    &wga->wga_sa_mask, rnh);
3431 		if (rn == NULL) {
3432 			char addrstr[128];
3433 			sockaddr_format(&wga->wga_sa_addr, addrstr,
3434 			    sizeof(addrstr));
3435 			WGLOG(LOG_WARNING, "Couldn't delete %s", addrstr);
3436 		}
3437 	}
3438 	rw_exit(wg->wg_rwlock);
3439 
3440 	/* Purge pending packets.  */
3441 	wg_purge_pending_packets(wgp);
3442 
3443 	/* Halt all packet processing and timeouts.  */
3444 	callout_halt(&wgp->wgp_rekey_timer, NULL);
3445 	callout_halt(&wgp->wgp_handshake_timeout_timer, NULL);
3446 	callout_halt(&wgp->wgp_session_dtor_timer, NULL);
3447 
3448 	/* Wait for any queued work to complete.  */
3449 	workqueue_wait(wg_wq, &wgp->wgp_work);
3450 
3451 	wgs = wgp->wgp_session_unstable;
3452 	if (wgs->wgs_state != WGS_STATE_UNKNOWN) {
3453 		mutex_enter(wgp->wgp_lock);
3454 		wg_destroy_session(wg, wgs);
3455 		mutex_exit(wgp->wgp_lock);
3456 	}
3457 	mutex_destroy(&wgs->wgs_recvwin->lock);
3458 	kmem_free(wgs->wgs_recvwin, sizeof(*wgs->wgs_recvwin));
3459 #ifndef __HAVE_ATOMIC64_LOADSTORE
3460 	mutex_destroy(&wgs->wgs_send_counter_lock);
3461 #endif
3462 	kmem_free(wgs, sizeof(*wgs));
3463 
3464 	wgs = wgp->wgp_session_stable;
3465 	if (wgs->wgs_state != WGS_STATE_UNKNOWN) {
3466 		mutex_enter(wgp->wgp_lock);
3467 		wg_destroy_session(wg, wgs);
3468 		mutex_exit(wgp->wgp_lock);
3469 	}
3470 	mutex_destroy(&wgs->wgs_recvwin->lock);
3471 	kmem_free(wgs->wgs_recvwin, sizeof(*wgs->wgs_recvwin));
3472 #ifndef __HAVE_ATOMIC64_LOADSTORE
3473 	mutex_destroy(&wgs->wgs_send_counter_lock);
3474 #endif
3475 	kmem_free(wgs, sizeof(*wgs));
3476 
3477 	psref_target_destroy(&wgp->wgp_endpoint->wgsa_psref, wg_psref_class);
3478 	psref_target_destroy(&wgp->wgp_endpoint0->wgsa_psref, wg_psref_class);
3479 	kmem_free(wgp->wgp_endpoint, sizeof(*wgp->wgp_endpoint));
3480 	kmem_free(wgp->wgp_endpoint0, sizeof(*wgp->wgp_endpoint0));
3481 
3482 	pserialize_destroy(wgp->wgp_psz);
3483 	mutex_obj_free(wgp->wgp_intr_lock);
3484 	mutex_obj_free(wgp->wgp_lock);
3485 
3486 	kmem_free(wgp, sizeof(*wgp));
3487 }
3488 
3489 static void
3490 wg_destroy_all_peers(struct wg_softc *wg)
3491 {
3492 	struct wg_peer *wgp, *wgp0 __diagused;
3493 	void *garbage_byname, *garbage_bypubkey;
3494 
3495 restart:
3496 	garbage_byname = garbage_bypubkey = NULL;
3497 	mutex_enter(wg->wg_lock);
3498 	WG_PEER_WRITER_FOREACH(wgp, wg) {
3499 		if (wgp->wgp_name[0]) {
3500 			wgp0 = thmap_del(wg->wg_peers_byname, wgp->wgp_name,
3501 			    strlen(wgp->wgp_name));
3502 			KASSERT(wgp0 == wgp);
3503 			garbage_byname = thmap_stage_gc(wg->wg_peers_byname);
3504 		}
3505 		wgp0 = thmap_del(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
3506 		    sizeof(wgp->wgp_pubkey));
3507 		KASSERT(wgp0 == wgp);
3508 		garbage_bypubkey = thmap_stage_gc(wg->wg_peers_bypubkey);
3509 		WG_PEER_WRITER_REMOVE(wgp);
3510 		wg->wg_npeers--;
3511 		mutex_enter(wgp->wgp_lock);
3512 		pserialize_perform(wgp->wgp_psz);
3513 		mutex_exit(wgp->wgp_lock);
3514 		PSLIST_ENTRY_DESTROY(wgp, wgp_peerlist_entry);
3515 		break;
3516 	}
3517 	mutex_exit(wg->wg_lock);
3518 
3519 	if (wgp == NULL)
3520 		return;
3521 
3522 	psref_target_destroy(&wgp->wgp_psref, wg_psref_class);
3523 
3524 	wg_destroy_peer(wgp);
3525 	thmap_gc(wg->wg_peers_byname, garbage_byname);
3526 	thmap_gc(wg->wg_peers_bypubkey, garbage_bypubkey);
3527 
3528 	goto restart;
3529 }
3530 
3531 static int
3532 wg_destroy_peer_name(struct wg_softc *wg, const char *name)
3533 {
3534 	struct wg_peer *wgp, *wgp0 __diagused;
3535 	void *garbage_byname, *garbage_bypubkey;
3536 
3537 	mutex_enter(wg->wg_lock);
3538 	wgp = thmap_del(wg->wg_peers_byname, name, strlen(name));
3539 	if (wgp != NULL) {
3540 		wgp0 = thmap_del(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
3541 		    sizeof(wgp->wgp_pubkey));
3542 		KASSERT(wgp0 == wgp);
3543 		garbage_byname = thmap_stage_gc(wg->wg_peers_byname);
3544 		garbage_bypubkey = thmap_stage_gc(wg->wg_peers_bypubkey);
3545 		WG_PEER_WRITER_REMOVE(wgp);
3546 		wg->wg_npeers--;
3547 		if (wg->wg_npeers == 0)
3548 			if_link_state_change(&wg->wg_if, LINK_STATE_DOWN);
3549 		mutex_enter(wgp->wgp_lock);
3550 		pserialize_perform(wgp->wgp_psz);
3551 		mutex_exit(wgp->wgp_lock);
3552 		PSLIST_ENTRY_DESTROY(wgp, wgp_peerlist_entry);
3553 	}
3554 	mutex_exit(wg->wg_lock);
3555 
3556 	if (wgp == NULL)
3557 		return ENOENT;
3558 
3559 	psref_target_destroy(&wgp->wgp_psref, wg_psref_class);
3560 
3561 	wg_destroy_peer(wgp);
3562 	thmap_gc(wg->wg_peers_byname, garbage_byname);
3563 	thmap_gc(wg->wg_peers_bypubkey, garbage_bypubkey);
3564 
3565 	return 0;
3566 }
3567 
3568 static int
3569 wg_if_attach(struct wg_softc *wg)
3570 {
3571 
3572 	wg->wg_if.if_addrlen = 0;
3573 	wg->wg_if.if_mtu = WG_MTU;
3574 	wg->wg_if.if_flags = IFF_MULTICAST;
3575 	wg->wg_if.if_extflags = IFEF_MPSAFE;
3576 	wg->wg_if.if_ioctl = wg_ioctl;
3577 	wg->wg_if.if_output = wg_output;
3578 	wg->wg_if.if_init = wg_init;
3579 #ifdef ALTQ
3580 	wg->wg_if.if_start = wg_start;
3581 #endif
3582 	wg->wg_if.if_stop = wg_stop;
3583 	wg->wg_if.if_type = IFT_OTHER;
3584 	wg->wg_if.if_dlt = DLT_NULL;
3585 	wg->wg_if.if_softc = wg;
3586 #ifdef ALTQ
3587 	IFQ_SET_READY(&wg->wg_if.if_snd);
3588 #endif
3589 	if_initialize(&wg->wg_if);
3590 
3591 	wg->wg_if.if_link_state = LINK_STATE_DOWN;
3592 	if_alloc_sadl(&wg->wg_if);
3593 	if_register(&wg->wg_if);
3594 
3595 	bpf_attach(&wg->wg_if, DLT_NULL, sizeof(uint32_t));
3596 
3597 	return 0;
3598 }
3599 
3600 static void
3601 wg_if_detach(struct wg_softc *wg)
3602 {
3603 	struct ifnet *ifp = &wg->wg_if;
3604 
3605 	bpf_detach(ifp);
3606 	if_detach(ifp);
3607 }
3608 
3609 static int
3610 wg_clone_create(struct if_clone *ifc, int unit)
3611 {
3612 	struct wg_softc *wg;
3613 	int error;
3614 
3615 	wg_guarantee_initialized();
3616 
3617 	error = wg_count_inc();
3618 	if (error)
3619 		return error;
3620 
3621 	wg = kmem_zalloc(sizeof(*wg), KM_SLEEP);
3622 
3623 	if_initname(&wg->wg_if, ifc->ifc_name, unit);
3624 
3625 	PSLIST_INIT(&wg->wg_peers);
3626 	wg->wg_peers_bypubkey = thmap_create(0, NULL, THMAP_NOCOPY);
3627 	wg->wg_peers_byname = thmap_create(0, NULL, THMAP_NOCOPY);
3628 	wg->wg_sessions_byindex = thmap_create(0, NULL, THMAP_NOCOPY);
3629 	wg->wg_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
3630 	wg->wg_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
3631 	wg->wg_rwlock = rw_obj_alloc();
3632 	threadpool_job_init(&wg->wg_job, wg_job, wg->wg_intr_lock,
3633 	    "%s", if_name(&wg->wg_if));
3634 	wg->wg_ops = &wg_ops_rumpkernel;
3635 
3636 	error = threadpool_get(&wg->wg_threadpool, PRI_NONE);
3637 	if (error)
3638 		goto fail0;
3639 
3640 #ifdef INET
3641 	error = wg_socreate(wg, AF_INET, &wg->wg_so4);
3642 	if (error)
3643 		goto fail1;
3644 	rn_inithead((void **)&wg->wg_rtable_ipv4,
3645 	    offsetof(struct sockaddr_in, sin_addr) * NBBY);
3646 #endif
3647 #ifdef INET6
3648 	error = wg_socreate(wg, AF_INET6, &wg->wg_so6);
3649 	if (error)
3650 		goto fail2;
3651 	rn_inithead((void **)&wg->wg_rtable_ipv6,
3652 	    offsetof(struct sockaddr_in6, sin6_addr) * NBBY);
3653 #endif
3654 
3655 	error = wg_if_attach(wg);
3656 	if (error)
3657 		goto fail3;
3658 
3659 	return 0;
3660 
3661 fail4: __unused
3662 	wg_if_detach(wg);
3663 fail3:	wg_destroy_all_peers(wg);
3664 #ifdef INET6
3665 	solock(wg->wg_so6);
3666 	wg->wg_so6->so_rcv.sb_flags &= ~SB_UPCALL;
3667 	sounlock(wg->wg_so6);
3668 #endif
3669 #ifdef INET
3670 	solock(wg->wg_so4);
3671 	wg->wg_so4->so_rcv.sb_flags &= ~SB_UPCALL;
3672 	sounlock(wg->wg_so4);
3673 #endif
3674 	mutex_enter(wg->wg_intr_lock);
3675 	threadpool_cancel_job(wg->wg_threadpool, &wg->wg_job);
3676 	mutex_exit(wg->wg_intr_lock);
3677 #ifdef INET6
3678 	if (wg->wg_rtable_ipv6 != NULL)
3679 		free(wg->wg_rtable_ipv6, M_RTABLE);
3680 	soclose(wg->wg_so6);
3681 fail2:
3682 #endif
3683 #ifdef INET
3684 	if (wg->wg_rtable_ipv4 != NULL)
3685 		free(wg->wg_rtable_ipv4, M_RTABLE);
3686 	soclose(wg->wg_so4);
3687 fail1:
3688 #endif
3689 	threadpool_put(wg->wg_threadpool, PRI_NONE);
3690 fail0:	threadpool_job_destroy(&wg->wg_job);
3691 	rw_obj_free(wg->wg_rwlock);
3692 	mutex_obj_free(wg->wg_intr_lock);
3693 	mutex_obj_free(wg->wg_lock);
3694 	thmap_destroy(wg->wg_sessions_byindex);
3695 	thmap_destroy(wg->wg_peers_byname);
3696 	thmap_destroy(wg->wg_peers_bypubkey);
3697 	PSLIST_DESTROY(&wg->wg_peers);
3698 	kmem_free(wg, sizeof(*wg));
3699 	wg_count_dec();
3700 	return error;
3701 }
3702 
3703 static int
3704 wg_clone_destroy(struct ifnet *ifp)
3705 {
3706 	struct wg_softc *wg = container_of(ifp, struct wg_softc, wg_if);
3707 
3708 #ifdef WG_RUMPKERNEL
3709 	if (wg_user_mode(wg)) {
3710 		rumpuser_wg_destroy(wg->wg_user);
3711 		wg->wg_user = NULL;
3712 	}
3713 #endif
3714 
3715 	wg_if_detach(wg);
3716 	wg_destroy_all_peers(wg);
3717 #ifdef INET6
3718 	solock(wg->wg_so6);
3719 	wg->wg_so6->so_rcv.sb_flags &= ~SB_UPCALL;
3720 	sounlock(wg->wg_so6);
3721 #endif
3722 #ifdef INET
3723 	solock(wg->wg_so4);
3724 	wg->wg_so4->so_rcv.sb_flags &= ~SB_UPCALL;
3725 	sounlock(wg->wg_so4);
3726 #endif
3727 	mutex_enter(wg->wg_intr_lock);
3728 	threadpool_cancel_job(wg->wg_threadpool, &wg->wg_job);
3729 	mutex_exit(wg->wg_intr_lock);
3730 #ifdef INET6
3731 	if (wg->wg_rtable_ipv6 != NULL)
3732 		free(wg->wg_rtable_ipv6, M_RTABLE);
3733 	soclose(wg->wg_so6);
3734 #endif
3735 #ifdef INET
3736 	if (wg->wg_rtable_ipv4 != NULL)
3737 		free(wg->wg_rtable_ipv4, M_RTABLE);
3738 	soclose(wg->wg_so4);
3739 #endif
3740 	threadpool_put(wg->wg_threadpool, PRI_NONE);
3741 	threadpool_job_destroy(&wg->wg_job);
3742 	rw_obj_free(wg->wg_rwlock);
3743 	mutex_obj_free(wg->wg_intr_lock);
3744 	mutex_obj_free(wg->wg_lock);
3745 	thmap_destroy(wg->wg_sessions_byindex);
3746 	thmap_destroy(wg->wg_peers_byname);
3747 	thmap_destroy(wg->wg_peers_bypubkey);
3748 	PSLIST_DESTROY(&wg->wg_peers);
3749 	kmem_free(wg, sizeof(*wg));
3750 	wg_count_dec();
3751 
3752 	return 0;
3753 }
3754 
3755 static struct wg_peer *
3756 wg_pick_peer_by_sa(struct wg_softc *wg, const struct sockaddr *sa,
3757     struct psref *psref)
3758 {
3759 	struct radix_node_head *rnh;
3760 	struct radix_node *rn;
3761 	struct wg_peer *wgp = NULL;
3762 	struct wg_allowedip *wga;
3763 
3764 #ifdef WG_DEBUG_LOG
3765 	char addrstr[128];
3766 	sockaddr_format(sa, addrstr, sizeof(addrstr));
3767 	WG_DLOG("sa=%s\n", addrstr);
3768 #endif
3769 
3770 	rw_enter(wg->wg_rwlock, RW_READER);
3771 
3772 	rnh = wg_rnh(wg, sa->sa_family);
3773 	if (rnh == NULL)
3774 		goto out;
3775 
3776 	rn = rnh->rnh_matchaddr(sa, rnh);
3777 	if (rn == NULL || (rn->rn_flags & RNF_ROOT) != 0)
3778 		goto out;
3779 
3780 	WG_TRACE("success");
3781 
3782 	wga = container_of(rn, struct wg_allowedip, wga_nodes[0]);
3783 	wgp = wga->wga_peer;
3784 	wg_get_peer(wgp, psref);
3785 
3786 out:
3787 	rw_exit(wg->wg_rwlock);
3788 	return wgp;
3789 }
3790 
3791 static void
3792 wg_fill_msg_data(struct wg_softc *wg, struct wg_peer *wgp,
3793     struct wg_session *wgs, struct wg_msg_data *wgmd)
3794 {
3795 
3796 	memset(wgmd, 0, sizeof(*wgmd));
3797 	wgmd->wgmd_type = htole32(WG_MSG_TYPE_DATA);
3798 	wgmd->wgmd_receiver = wgs->wgs_remote_index;
3799 	/* [W] 5.4.6: msg.counter := Nm^send */
3800 	/* [W] 5.4.6: Nm^send := Nm^send + 1 */
3801 	wgmd->wgmd_counter = htole64(wg_session_inc_send_counter(wgs));
3802 	WG_DLOG("counter=%"PRIu64"\n", le64toh(wgmd->wgmd_counter));
3803 }
3804 
3805 static int
3806 wg_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
3807     const struct rtentry *rt)
3808 {
3809 	struct wg_softc *wg = ifp->if_softc;
3810 	struct wg_peer *wgp = NULL;
3811 	struct wg_session *wgs = NULL;
3812 	struct psref wgp_psref, wgs_psref;
3813 	int bound;
3814 	int error;
3815 
3816 	bound = curlwp_bind();
3817 
3818 	/* TODO make the nest limit configurable via sysctl */
3819 	error = if_tunnel_check_nesting(ifp, m, 1);
3820 	if (error) {
3821 		WGLOG(LOG_ERR, "tunneling loop detected and packet dropped\n");
3822 		goto out0;
3823 	}
3824 
3825 #ifdef ALTQ
3826 	bool altq = atomic_load_relaxed(&ifp->if_snd.altq_flags)
3827 	    & ALTQF_ENABLED;
3828 	if (altq)
3829 		IFQ_CLASSIFY(&ifp->if_snd, m, dst->sa_family);
3830 #endif
3831 
3832 	bpf_mtap_af(ifp, dst->sa_family, m, BPF_D_OUT);
3833 
3834 	m->m_flags &= ~(M_BCAST|M_MCAST);
3835 
3836 	wgp = wg_pick_peer_by_sa(wg, dst, &wgp_psref);
3837 	if (wgp == NULL) {
3838 		WG_TRACE("peer not found");
3839 		error = EHOSTUNREACH;
3840 		goto out0;
3841 	}
3842 
3843 	/* Clear checksum-offload flags. */
3844 	m->m_pkthdr.csum_flags = 0;
3845 	m->m_pkthdr.csum_data = 0;
3846 
3847 	/* Check whether there's an established session.  */
3848 	wgs = wg_get_stable_session(wgp, &wgs_psref);
3849 	if (wgs == NULL) {
3850 		/*
3851 		 * No established session.  If we're the first to try
3852 		 * sending data, schedule a handshake and queue the
3853 		 * packet for when the handshake is done; otherwise
3854 		 * just drop the packet and let the ongoing handshake
3855 		 * attempt continue.  We could queue more data packets
3856 		 * but it's not clear that's worthwhile.
3857 		 */
3858 		if (atomic_cas_ptr(&wgp->wgp_pending, NULL, m) == NULL) {
3859 			m = NULL; /* consume */
3860 			WG_TRACE("queued first packet; init handshake");
3861 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
3862 		} else {
3863 			WG_TRACE("first packet already queued, dropping");
3864 		}
3865 		goto out1;
3866 	}
3867 
3868 	/* There's an established session.  Toss it in the queue.  */
3869 #ifdef ALTQ
3870 	if (altq) {
3871 		mutex_enter(ifp->if_snd.ifq_lock);
3872 		if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
3873 			M_SETCTX(m, wgp);
3874 			ALTQ_ENQUEUE(&ifp->if_snd, m, error);
3875 			m = NULL; /* consume */
3876 		}
3877 		mutex_exit(ifp->if_snd.ifq_lock);
3878 		if (m == NULL) {
3879 			wg_start(ifp);
3880 			goto out2;
3881 		}
3882 	}
3883 #endif
3884 	kpreempt_disable();
3885 	const uint32_t h = curcpu()->ci_index;	// pktq_rps_hash(m)
3886 	M_SETCTX(m, wgp);
3887 	if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
3888 		WGLOG(LOG_ERR, "pktq full, dropping\n");
3889 		error = ENOBUFS;
3890 		goto out3;
3891 	}
3892 	m = NULL;		/* consumed */
3893 	error = 0;
3894 out3:	kpreempt_enable();
3895 
3896 #ifdef ALTQ
3897 out2:
3898 #endif
3899 	wg_put_session(wgs, &wgs_psref);
3900 out1:	wg_put_peer(wgp, &wgp_psref);
3901 out0:	if (m)
3902 		m_freem(m);
3903 	curlwp_bindx(bound);
3904 	return error;
3905 }
3906 
3907 static int
3908 wg_send_udp(struct wg_peer *wgp, struct mbuf *m)
3909 {
3910 	struct psref psref;
3911 	struct wg_sockaddr *wgsa;
3912 	int error;
3913 	struct socket *so;
3914 
3915 	wgsa = wg_get_endpoint_sa(wgp, &psref);
3916 	so = wg_get_so_by_peer(wgp, wgsa);
3917 	solock(so);
3918 	if (wgsatosa(wgsa)->sa_family == AF_INET) {
3919 		error = udp_send(so, m, wgsatosa(wgsa), NULL, curlwp);
3920 	} else {
3921 #ifdef INET6
3922 		error = udp6_output(sotoinpcb(so), m, wgsatosin6(wgsa),
3923 		    NULL, curlwp);
3924 #else
3925 		m_freem(m);
3926 		error = EPFNOSUPPORT;
3927 #endif
3928 	}
3929 	sounlock(so);
3930 	wg_put_sa(wgp, wgsa, &psref);
3931 
3932 	return error;
3933 }
3934 
3935 /* Inspired by pppoe_get_mbuf */
3936 static struct mbuf *
3937 wg_get_mbuf(size_t leading_len, size_t len)
3938 {
3939 	struct mbuf *m;
3940 
3941 	KASSERT(leading_len <= MCLBYTES);
3942 	KASSERT(len <= MCLBYTES - leading_len);
3943 
3944 	m = m_gethdr(M_DONTWAIT, MT_DATA);
3945 	if (m == NULL)
3946 		return NULL;
3947 	if (len + leading_len > MHLEN) {
3948 		m_clget(m, M_DONTWAIT);
3949 		if ((m->m_flags & M_EXT) == 0) {
3950 			m_free(m);
3951 			return NULL;
3952 		}
3953 	}
3954 	m->m_data += leading_len;
3955 	m->m_pkthdr.len = m->m_len = len;
3956 
3957 	return m;
3958 }
3959 
3960 static int
3961 wg_send_data_msg(struct wg_peer *wgp, struct wg_session *wgs,
3962     struct mbuf *m)
3963 {
3964 	struct wg_softc *wg = wgp->wgp_sc;
3965 	int error;
3966 	size_t inner_len, padded_len, encrypted_len;
3967 	char *padded_buf = NULL;
3968 	size_t mlen;
3969 	struct wg_msg_data *wgmd;
3970 	bool free_padded_buf = false;
3971 	struct mbuf *n;
3972 	size_t leading_len = max_hdr + sizeof(struct udphdr);
3973 
3974 	mlen = m_length(m);
3975 	inner_len = mlen;
3976 	padded_len = roundup(mlen, 16);
3977 	encrypted_len = padded_len + WG_AUTHTAG_LEN;
3978 	WG_DLOG("inner=%lu, padded=%lu, encrypted_len=%lu\n",
3979 	    inner_len, padded_len, encrypted_len);
3980 	if (mlen != 0) {
3981 		bool success;
3982 		success = m_ensure_contig(&m, padded_len);
3983 		if (success) {
3984 			padded_buf = mtod(m, char *);
3985 		} else {
3986 			padded_buf = kmem_intr_alloc(padded_len, KM_NOSLEEP);
3987 			if (padded_buf == NULL) {
3988 				error = ENOBUFS;
3989 				goto end;
3990 			}
3991 			free_padded_buf = true;
3992 			m_copydata(m, 0, mlen, padded_buf);
3993 		}
3994 		memset(padded_buf + mlen, 0, padded_len - inner_len);
3995 	}
3996 
3997 	n = wg_get_mbuf(leading_len, sizeof(*wgmd) + encrypted_len);
3998 	if (n == NULL) {
3999 		error = ENOBUFS;
4000 		goto end;
4001 	}
4002 	KASSERT(n->m_len >= sizeof(*wgmd));
4003 	wgmd = mtod(n, struct wg_msg_data *);
4004 	wg_fill_msg_data(wg, wgp, wgs, wgmd);
4005 	/* [W] 5.4.6: AEAD(Tm^send, Nm^send, P, e) */
4006 	wg_algo_aead_enc((char *)wgmd + sizeof(*wgmd), encrypted_len,
4007 	    wgs->wgs_tkey_send, le64toh(wgmd->wgmd_counter),
4008 	    padded_buf, padded_len,
4009 	    NULL, 0);
4010 
4011 	error = wg->wg_ops->send_data_msg(wgp, n);
4012 	if (error == 0) {
4013 		struct ifnet *ifp = &wg->wg_if;
4014 		if_statadd(ifp, if_obytes, mlen);
4015 		if_statinc(ifp, if_opackets);
4016 		if (wgs->wgs_is_initiator &&
4017 		    wgs->wgs_time_last_data_sent == 0) {
4018 			/*
4019 			 * [W] 6.2 Transport Message Limits
4020 			 * "if a peer is the initiator of a current secure
4021 			 *  session, WireGuard will send a handshake initiation
4022 			 *  message to begin a new secure session if, after
4023 			 *  transmitting a transport data message, the current
4024 			 *  secure session is REKEY-AFTER-TIME seconds old,"
4025 			 */
4026 			wg_schedule_rekey_timer(wgp);
4027 		}
4028 		wgs->wgs_time_last_data_sent = time_uptime;
4029 		if (wg_session_get_send_counter(wgs) >=
4030 		    wg_rekey_after_messages) {
4031 			/*
4032 			 * [W] 6.2 Transport Message Limits
4033 			 * "WireGuard will try to create a new session, by
4034 			 *  sending a handshake initiation message (section
4035 			 *  5.4.2), after it has sent REKEY-AFTER-MESSAGES
4036 			 *  transport data messages..."
4037 			 */
4038 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
4039 		}
4040 	}
4041 end:
4042 	m_freem(m);
4043 	if (free_padded_buf)
4044 		kmem_intr_free(padded_buf, padded_len);
4045 	return error;
4046 }
4047 
4048 static void
4049 wg_input(struct ifnet *ifp, struct mbuf *m, const int af)
4050 {
4051 	pktqueue_t *pktq;
4052 	size_t pktlen;
4053 
4054 	KASSERT(af == AF_INET || af == AF_INET6);
4055 
4056 	WG_TRACE("");
4057 
4058 	m_set_rcvif(m, ifp);
4059 	pktlen = m->m_pkthdr.len;
4060 
4061 	bpf_mtap_af(ifp, af, m, BPF_D_IN);
4062 
4063 	switch (af) {
4064 	case AF_INET:
4065 		pktq = ip_pktq;
4066 		break;
4067 #ifdef INET6
4068 	case AF_INET6:
4069 		pktq = ip6_pktq;
4070 		break;
4071 #endif
4072 	default:
4073 		panic("invalid af=%d", af);
4074 	}
4075 
4076 	kpreempt_disable();
4077 	const u_int h = curcpu()->ci_index;
4078 	if (__predict_true(pktq_enqueue(pktq, m, h))) {
4079 		if_statadd(ifp, if_ibytes, pktlen);
4080 		if_statinc(ifp, if_ipackets);
4081 	} else {
4082 		m_freem(m);
4083 	}
4084 	kpreempt_enable();
4085 }
4086 
4087 static void
4088 wg_calc_pubkey(uint8_t pubkey[WG_STATIC_KEY_LEN],
4089     const uint8_t privkey[WG_STATIC_KEY_LEN])
4090 {
4091 
4092 	crypto_scalarmult_base(pubkey, privkey);
4093 }
4094 
4095 static int
4096 wg_rtable_add_route(struct wg_softc *wg, struct wg_allowedip *wga)
4097 {
4098 	struct radix_node_head *rnh;
4099 	struct radix_node *rn;
4100 	int error = 0;
4101 
4102 	rw_enter(wg->wg_rwlock, RW_WRITER);
4103 	rnh = wg_rnh(wg, wga->wga_family);
4104 	KASSERT(rnh != NULL);
4105 	rn = rnh->rnh_addaddr(&wga->wga_sa_addr, &wga->wga_sa_mask, rnh,
4106 	    wga->wga_nodes);
4107 	rw_exit(wg->wg_rwlock);
4108 
4109 	if (rn == NULL)
4110 		error = EEXIST;
4111 
4112 	return error;
4113 }
4114 
4115 static int
4116 wg_handle_prop_peer(struct wg_softc *wg, prop_dictionary_t peer,
4117     struct wg_peer **wgpp)
4118 {
4119 	int error = 0;
4120 	const void *pubkey;
4121 	size_t pubkey_len;
4122 	const void *psk;
4123 	size_t psk_len;
4124 	const char *name = NULL;
4125 
4126 	if (prop_dictionary_get_string(peer, "name", &name)) {
4127 		if (strlen(name) > WG_PEER_NAME_MAXLEN) {
4128 			error = EINVAL;
4129 			goto out;
4130 		}
4131 	}
4132 
4133 	if (!prop_dictionary_get_data(peer, "public_key",
4134 		&pubkey, &pubkey_len)) {
4135 		error = EINVAL;
4136 		goto out;
4137 	}
4138 #ifdef WG_DEBUG_DUMP
4139     {
4140 	char *hex = gethexdump(pubkey, pubkey_len);
4141 	log(LOG_DEBUG, "pubkey=%p, pubkey_len=%lu\n%s\n",
4142 	    pubkey, pubkey_len, hex);
4143 	puthexdump(hex, pubkey, pubkey_len);
4144     }
4145 #endif
4146 
4147 	struct wg_peer *wgp = wg_alloc_peer(wg);
4148 	memcpy(wgp->wgp_pubkey, pubkey, sizeof(wgp->wgp_pubkey));
4149 	if (name != NULL)
4150 		strncpy(wgp->wgp_name, name, sizeof(wgp->wgp_name));
4151 
4152 	if (prop_dictionary_get_data(peer, "preshared_key", &psk, &psk_len)) {
4153 		if (psk_len != sizeof(wgp->wgp_psk)) {
4154 			error = EINVAL;
4155 			goto out;
4156 		}
4157 		memcpy(wgp->wgp_psk, psk, sizeof(wgp->wgp_psk));
4158 	}
4159 
4160 	const void *addr;
4161 	size_t addr_len;
4162 	struct wg_sockaddr *wgsa = wgp->wgp_endpoint;
4163 
4164 	if (!prop_dictionary_get_data(peer, "endpoint", &addr, &addr_len))
4165 		goto skip_endpoint;
4166 	if (addr_len < sizeof(*wgsatosa(wgsa)) ||
4167 	    addr_len > sizeof(*wgsatoss(wgsa))) {
4168 		error = EINVAL;
4169 		goto out;
4170 	}
4171 	memcpy(wgsatoss(wgsa), addr, addr_len);
4172 	switch (wgsa_family(wgsa)) {
4173 	case AF_INET:
4174 #ifdef INET6
4175 	case AF_INET6:
4176 #endif
4177 		break;
4178 	default:
4179 		error = EPFNOSUPPORT;
4180 		goto out;
4181 	}
4182 	if (addr_len != sockaddr_getsize_by_family(wgsa_family(wgsa))) {
4183 		error = EINVAL;
4184 		goto out;
4185 	}
4186     {
4187 	char addrstr[128];
4188 	sockaddr_format(wgsatosa(wgsa), addrstr, sizeof(addrstr));
4189 	WG_DLOG("addr=%s\n", addrstr);
4190     }
4191 	wgp->wgp_endpoint_available = true;
4192 
4193 	prop_array_t allowedips;
4194 skip_endpoint:
4195 	allowedips = prop_dictionary_get(peer, "allowedips");
4196 	if (allowedips == NULL)
4197 		goto skip;
4198 
4199 	prop_object_iterator_t _it = prop_array_iterator(allowedips);
4200 	prop_dictionary_t prop_allowedip;
4201 	int j = 0;
4202 	while ((prop_allowedip = prop_object_iterator_next(_it)) != NULL) {
4203 		struct wg_allowedip *wga = &wgp->wgp_allowedips[j];
4204 
4205 		if (!prop_dictionary_get_int(prop_allowedip, "family",
4206 			&wga->wga_family))
4207 			continue;
4208 		if (!prop_dictionary_get_data(prop_allowedip, "ip",
4209 			&addr, &addr_len))
4210 			continue;
4211 		if (!prop_dictionary_get_uint8(prop_allowedip, "cidr",
4212 			&wga->wga_cidr))
4213 			continue;
4214 
4215 		switch (wga->wga_family) {
4216 		case AF_INET: {
4217 			struct sockaddr_in sin;
4218 			char addrstr[128];
4219 			struct in_addr mask;
4220 			struct sockaddr_in sin_mask;
4221 
4222 			if (addr_len != sizeof(struct in_addr))
4223 				return EINVAL;
4224 			memcpy(&wga->wga_addr4, addr, addr_len);
4225 
4226 			sockaddr_in_init(&sin, (const struct in_addr *)addr,
4227 			    0);
4228 			sockaddr_copy(&wga->wga_sa_addr,
4229 			    sizeof(sin), sintosa(&sin));
4230 
4231 			sockaddr_format(sintosa(&sin),
4232 			    addrstr, sizeof(addrstr));
4233 			WG_DLOG("addr=%s/%d\n", addrstr, wga->wga_cidr);
4234 
4235 			in_len2mask(&mask, wga->wga_cidr);
4236 			sockaddr_in_init(&sin_mask, &mask, 0);
4237 			sockaddr_copy(&wga->wga_sa_mask,
4238 			    sizeof(sin_mask), sintosa(&sin_mask));
4239 
4240 			break;
4241 		    }
4242 #ifdef INET6
4243 		case AF_INET6: {
4244 			struct sockaddr_in6 sin6;
4245 			char addrstr[128];
4246 			struct in6_addr mask;
4247 			struct sockaddr_in6 sin6_mask;
4248 
4249 			if (addr_len != sizeof(struct in6_addr))
4250 				return EINVAL;
4251 			memcpy(&wga->wga_addr6, addr, addr_len);
4252 
4253 			sockaddr_in6_init(&sin6, (const struct in6_addr *)addr,
4254 			    0, 0, 0);
4255 			sockaddr_copy(&wga->wga_sa_addr,
4256 			    sizeof(sin6), sin6tosa(&sin6));
4257 
4258 			sockaddr_format(sin6tosa(&sin6),
4259 			    addrstr, sizeof(addrstr));
4260 			WG_DLOG("addr=%s/%d\n", addrstr, wga->wga_cidr);
4261 
4262 			in6_prefixlen2mask(&mask, wga->wga_cidr);
4263 			sockaddr_in6_init(&sin6_mask, &mask, 0, 0, 0);
4264 			sockaddr_copy(&wga->wga_sa_mask,
4265 			    sizeof(sin6_mask), sin6tosa(&sin6_mask));
4266 
4267 			break;
4268 		    }
4269 #endif
4270 		default:
4271 			error = EINVAL;
4272 			goto out;
4273 		}
4274 		wga->wga_peer = wgp;
4275 
4276 		error = wg_rtable_add_route(wg, wga);
4277 		if (error != 0)
4278 			goto out;
4279 
4280 		j++;
4281 	}
4282 	wgp->wgp_n_allowedips = j;
4283 skip:
4284 	*wgpp = wgp;
4285 out:
4286 	return error;
4287 }
4288 
4289 static int
4290 wg_alloc_prop_buf(char **_buf, struct ifdrv *ifd)
4291 {
4292 	int error;
4293 	char *buf;
4294 
4295 	WG_DLOG("buf=%p, len=%lu\n", ifd->ifd_data, ifd->ifd_len);
4296 	if (ifd->ifd_len >= WG_MAX_PROPLEN)
4297 		return E2BIG;
4298 	buf = kmem_alloc(ifd->ifd_len + 1, KM_SLEEP);
4299 	error = copyin(ifd->ifd_data, buf, ifd->ifd_len);
4300 	if (error != 0)
4301 		return error;
4302 	buf[ifd->ifd_len] = '\0';
4303 #ifdef WG_DEBUG_DUMP
4304 	log(LOG_DEBUG, "%.*s\n",
4305 	    (int)MIN(INT_MAX, ifd->ifd_len),
4306 	    (const char *)buf);
4307 #endif
4308 	*_buf = buf;
4309 	return 0;
4310 }
4311 
4312 static int
4313 wg_ioctl_set_private_key(struct wg_softc *wg, struct ifdrv *ifd)
4314 {
4315 	int error;
4316 	prop_dictionary_t prop_dict;
4317 	char *buf = NULL;
4318 	const void *privkey;
4319 	size_t privkey_len;
4320 
4321 	error = wg_alloc_prop_buf(&buf, ifd);
4322 	if (error != 0)
4323 		return error;
4324 	error = EINVAL;
4325 	prop_dict = prop_dictionary_internalize(buf);
4326 	if (prop_dict == NULL)
4327 		goto out;
4328 	if (!prop_dictionary_get_data(prop_dict, "private_key",
4329 		&privkey, &privkey_len))
4330 		goto out;
4331 #ifdef WG_DEBUG_DUMP
4332     {
4333 	char *hex = gethexdump(privkey, privkey_len);
4334 	log(LOG_DEBUG, "privkey=%p, privkey_len=%lu\n%s\n",
4335 	    privkey, privkey_len, hex);
4336 	puthexdump(hex, privkey, privkey_len);
4337     }
4338 #endif
4339 	if (privkey_len != WG_STATIC_KEY_LEN)
4340 		goto out;
4341 	memcpy(wg->wg_privkey, privkey, WG_STATIC_KEY_LEN);
4342 	wg_calc_pubkey(wg->wg_pubkey, wg->wg_privkey);
4343 	error = 0;
4344 
4345 out:
4346 	kmem_free(buf, ifd->ifd_len + 1);
4347 	return error;
4348 }
4349 
4350 static int
4351 wg_ioctl_set_listen_port(struct wg_softc *wg, struct ifdrv *ifd)
4352 {
4353 	int error;
4354 	prop_dictionary_t prop_dict;
4355 	char *buf = NULL;
4356 	uint16_t port;
4357 
4358 	error = wg_alloc_prop_buf(&buf, ifd);
4359 	if (error != 0)
4360 		return error;
4361 	error = EINVAL;
4362 	prop_dict = prop_dictionary_internalize(buf);
4363 	if (prop_dict == NULL)
4364 		goto out;
4365 	if (!prop_dictionary_get_uint16(prop_dict, "listen_port", &port))
4366 		goto out;
4367 
4368 	error = wg->wg_ops->bind_port(wg, (uint16_t)port);
4369 
4370 out:
4371 	kmem_free(buf, ifd->ifd_len + 1);
4372 	return error;
4373 }
4374 
4375 static int
4376 wg_ioctl_add_peer(struct wg_softc *wg, struct ifdrv *ifd)
4377 {
4378 	int error;
4379 	prop_dictionary_t prop_dict;
4380 	char *buf = NULL;
4381 	struct wg_peer *wgp = NULL, *wgp0 __diagused;
4382 
4383 	error = wg_alloc_prop_buf(&buf, ifd);
4384 	if (error != 0)
4385 		return error;
4386 	error = EINVAL;
4387 	prop_dict = prop_dictionary_internalize(buf);
4388 	if (prop_dict == NULL)
4389 		goto out;
4390 
4391 	error = wg_handle_prop_peer(wg, prop_dict, &wgp);
4392 	if (error != 0)
4393 		goto out;
4394 
4395 	mutex_enter(wg->wg_lock);
4396 	if (thmap_get(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
4397 		sizeof(wgp->wgp_pubkey)) != NULL ||
4398 	    (wgp->wgp_name[0] &&
4399 		thmap_get(wg->wg_peers_byname, wgp->wgp_name,
4400 		    strlen(wgp->wgp_name)) != NULL)) {
4401 		mutex_exit(wg->wg_lock);
4402 		wg_destroy_peer(wgp);
4403 		error = EEXIST;
4404 		goto out;
4405 	}
4406 	wgp0 = thmap_put(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
4407 	    sizeof(wgp->wgp_pubkey), wgp);
4408 	KASSERT(wgp0 == wgp);
4409 	if (wgp->wgp_name[0]) {
4410 		wgp0 = thmap_put(wg->wg_peers_byname, wgp->wgp_name,
4411 		    strlen(wgp->wgp_name), wgp);
4412 		KASSERT(wgp0 == wgp);
4413 	}
4414 	WG_PEER_WRITER_INSERT_HEAD(wgp, wg);
4415 	wg->wg_npeers++;
4416 	mutex_exit(wg->wg_lock);
4417 
4418 	if_link_state_change(&wg->wg_if, LINK_STATE_UP);
4419 
4420 out:
4421 	kmem_free(buf, ifd->ifd_len + 1);
4422 	return error;
4423 }
4424 
4425 static int
4426 wg_ioctl_delete_peer(struct wg_softc *wg, struct ifdrv *ifd)
4427 {
4428 	int error;
4429 	prop_dictionary_t prop_dict;
4430 	char *buf = NULL;
4431 	const char *name;
4432 
4433 	error = wg_alloc_prop_buf(&buf, ifd);
4434 	if (error != 0)
4435 		return error;
4436 	error = EINVAL;
4437 	prop_dict = prop_dictionary_internalize(buf);
4438 	if (prop_dict == NULL)
4439 		goto out;
4440 
4441 	if (!prop_dictionary_get_string(prop_dict, "name", &name))
4442 		goto out;
4443 	if (strlen(name) > WG_PEER_NAME_MAXLEN)
4444 		goto out;
4445 
4446 	error = wg_destroy_peer_name(wg, name);
4447 out:
4448 	kmem_free(buf, ifd->ifd_len + 1);
4449 	return error;
4450 }
4451 
4452 static bool
4453 wg_is_authorized(struct wg_softc *wg, u_long cmd)
4454 {
4455 	int au = cmd == SIOCGDRVSPEC ?
4456 	    KAUTH_REQ_NETWORK_INTERFACE_WG_GETPRIV :
4457 	    KAUTH_REQ_NETWORK_INTERFACE_WG_SETPRIV;
4458 	return kauth_authorize_network(kauth_cred_get(),
4459 	    KAUTH_NETWORK_INTERFACE_WG, au, &wg->wg_if,
4460 	    (void *)cmd, NULL) == 0;
4461 }
4462 
4463 static int
4464 wg_ioctl_get(struct wg_softc *wg, struct ifdrv *ifd)
4465 {
4466 	int error = ENOMEM;
4467 	prop_dictionary_t prop_dict;
4468 	prop_array_t peers = NULL;
4469 	char *buf;
4470 	struct wg_peer *wgp;
4471 	int s, i;
4472 
4473 	prop_dict = prop_dictionary_create();
4474 	if (prop_dict == NULL)
4475 		goto error;
4476 
4477 	if (wg_is_authorized(wg, SIOCGDRVSPEC)) {
4478 		if (!prop_dictionary_set_data(prop_dict, "private_key",
4479 			wg->wg_privkey, WG_STATIC_KEY_LEN))
4480 			goto error;
4481 	}
4482 
4483 	if (wg->wg_listen_port != 0) {
4484 		if (!prop_dictionary_set_uint16(prop_dict, "listen_port",
4485 			wg->wg_listen_port))
4486 			goto error;
4487 	}
4488 
4489 	if (wg->wg_npeers == 0)
4490 		goto skip_peers;
4491 
4492 	peers = prop_array_create();
4493 	if (peers == NULL)
4494 		goto error;
4495 
4496 	s = pserialize_read_enter();
4497 	i = 0;
4498 	WG_PEER_READER_FOREACH(wgp, wg) {
4499 		struct wg_sockaddr *wgsa;
4500 		struct psref wgp_psref, wgsa_psref;
4501 		prop_dictionary_t prop_peer;
4502 
4503 		wg_get_peer(wgp, &wgp_psref);
4504 		pserialize_read_exit(s);
4505 
4506 		prop_peer = prop_dictionary_create();
4507 		if (prop_peer == NULL)
4508 			goto next;
4509 
4510 		if (strlen(wgp->wgp_name) > 0) {
4511 			if (!prop_dictionary_set_string(prop_peer, "name",
4512 				wgp->wgp_name))
4513 				goto next;
4514 		}
4515 
4516 		if (!prop_dictionary_set_data(prop_peer, "public_key",
4517 			wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey)))
4518 			goto next;
4519 
4520 		uint8_t psk_zero[WG_PRESHARED_KEY_LEN] = {0};
4521 		if (!consttime_memequal(wgp->wgp_psk, psk_zero,
4522 			sizeof(wgp->wgp_psk))) {
4523 			if (wg_is_authorized(wg, SIOCGDRVSPEC)) {
4524 				if (!prop_dictionary_set_data(prop_peer,
4525 					"preshared_key",
4526 					wgp->wgp_psk, sizeof(wgp->wgp_psk)))
4527 					goto next;
4528 			}
4529 		}
4530 
4531 		wgsa = wg_get_endpoint_sa(wgp, &wgsa_psref);
4532 		CTASSERT(AF_UNSPEC == 0);
4533 		if (wgsa_family(wgsa) != 0 /*AF_UNSPEC*/ &&
4534 		    !prop_dictionary_set_data(prop_peer, "endpoint",
4535 			wgsatoss(wgsa),
4536 			sockaddr_getsize_by_family(wgsa_family(wgsa)))) {
4537 			wg_put_sa(wgp, wgsa, &wgsa_psref);
4538 			goto next;
4539 		}
4540 		wg_put_sa(wgp, wgsa, &wgsa_psref);
4541 
4542 		const struct timespec *t = &wgp->wgp_last_handshake_time;
4543 
4544 		if (!prop_dictionary_set_uint64(prop_peer,
4545 			"last_handshake_time_sec", (uint64_t)t->tv_sec))
4546 			goto next;
4547 		if (!prop_dictionary_set_uint32(prop_peer,
4548 			"last_handshake_time_nsec", (uint32_t)t->tv_nsec))
4549 			goto next;
4550 
4551 		if (wgp->wgp_n_allowedips == 0)
4552 			goto skip_allowedips;
4553 
4554 		prop_array_t allowedips = prop_array_create();
4555 		if (allowedips == NULL)
4556 			goto next;
4557 		for (int j = 0; j < wgp->wgp_n_allowedips; j++) {
4558 			struct wg_allowedip *wga = &wgp->wgp_allowedips[j];
4559 			prop_dictionary_t prop_allowedip;
4560 
4561 			prop_allowedip = prop_dictionary_create();
4562 			if (prop_allowedip == NULL)
4563 				break;
4564 
4565 			if (!prop_dictionary_set_int(prop_allowedip, "family",
4566 				wga->wga_family))
4567 				goto _next;
4568 			if (!prop_dictionary_set_uint8(prop_allowedip, "cidr",
4569 				wga->wga_cidr))
4570 				goto _next;
4571 
4572 			switch (wga->wga_family) {
4573 			case AF_INET:
4574 				if (!prop_dictionary_set_data(prop_allowedip,
4575 					"ip", &wga->wga_addr4,
4576 					sizeof(wga->wga_addr4)))
4577 					goto _next;
4578 				break;
4579 #ifdef INET6
4580 			case AF_INET6:
4581 				if (!prop_dictionary_set_data(prop_allowedip,
4582 					"ip", &wga->wga_addr6,
4583 					sizeof(wga->wga_addr6)))
4584 					goto _next;
4585 				break;
4586 #endif
4587 			default:
4588 				break;
4589 			}
4590 			prop_array_set(allowedips, j, prop_allowedip);
4591 		_next:
4592 			prop_object_release(prop_allowedip);
4593 		}
4594 		prop_dictionary_set(prop_peer, "allowedips", allowedips);
4595 		prop_object_release(allowedips);
4596 
4597 	skip_allowedips:
4598 
4599 		prop_array_set(peers, i, prop_peer);
4600 	next:
4601 		if (prop_peer)
4602 			prop_object_release(prop_peer);
4603 		i++;
4604 
4605 		s = pserialize_read_enter();
4606 		wg_put_peer(wgp, &wgp_psref);
4607 	}
4608 	pserialize_read_exit(s);
4609 
4610 	prop_dictionary_set(prop_dict, "peers", peers);
4611 	prop_object_release(peers);
4612 	peers = NULL;
4613 
4614 skip_peers:
4615 	buf = prop_dictionary_externalize(prop_dict);
4616 	if (buf == NULL)
4617 		goto error;
4618 	if (ifd->ifd_len < (strlen(buf) + 1)) {
4619 		error = EINVAL;
4620 		goto error;
4621 	}
4622 	error = copyout(buf, ifd->ifd_data, strlen(buf) + 1);
4623 
4624 	free(buf, 0);
4625 error:
4626 	if (peers != NULL)
4627 		prop_object_release(peers);
4628 	if (prop_dict != NULL)
4629 		prop_object_release(prop_dict);
4630 
4631 	return error;
4632 }
4633 
4634 static int
4635 wg_ioctl(struct ifnet *ifp, u_long cmd, void *data)
4636 {
4637 	struct wg_softc *wg = ifp->if_softc;
4638 	struct ifreq *ifr = data;
4639 	struct ifaddr *ifa = data;
4640 	struct ifdrv *ifd = data;
4641 	int error = 0;
4642 
4643 	switch (cmd) {
4644 	case SIOCINITIFADDR:
4645 		if (ifa->ifa_addr->sa_family != AF_LINK &&
4646 		    (ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
4647 		    (IFF_UP | IFF_RUNNING)) {
4648 			ifp->if_flags |= IFF_UP;
4649 			error = if_init(ifp);
4650 		}
4651 		return error;
4652 	case SIOCADDMULTI:
4653 	case SIOCDELMULTI:
4654 		switch (ifr->ifr_addr.sa_family) {
4655 		case AF_INET:	/* IP supports Multicast */
4656 			break;
4657 #ifdef INET6
4658 		case AF_INET6:	/* IP6 supports Multicast */
4659 			break;
4660 #endif
4661 		default:  /* Other protocols doesn't support Multicast */
4662 			error = EAFNOSUPPORT;
4663 			break;
4664 		}
4665 		return error;
4666 	case SIOCSDRVSPEC:
4667 		if (!wg_is_authorized(wg, cmd)) {
4668 			return EPERM;
4669 		}
4670 		switch (ifd->ifd_cmd) {
4671 		case WG_IOCTL_SET_PRIVATE_KEY:
4672 			error = wg_ioctl_set_private_key(wg, ifd);
4673 			break;
4674 		case WG_IOCTL_SET_LISTEN_PORT:
4675 			error = wg_ioctl_set_listen_port(wg, ifd);
4676 			break;
4677 		case WG_IOCTL_ADD_PEER:
4678 			error = wg_ioctl_add_peer(wg, ifd);
4679 			break;
4680 		case WG_IOCTL_DELETE_PEER:
4681 			error = wg_ioctl_delete_peer(wg, ifd);
4682 			break;
4683 		default:
4684 			error = EINVAL;
4685 			break;
4686 		}
4687 		return error;
4688 	case SIOCGDRVSPEC:
4689 		return wg_ioctl_get(wg, ifd);
4690 	case SIOCSIFFLAGS:
4691 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
4692 			break;
4693 		switch (ifp->if_flags & (IFF_UP|IFF_RUNNING)) {
4694 		case IFF_RUNNING:
4695 			/*
4696 			 * If interface is marked down and it is running,
4697 			 * then stop and disable it.
4698 			 */
4699 			if_stop(ifp, 1);
4700 			break;
4701 		case IFF_UP:
4702 			/*
4703 			 * If interface is marked up and it is stopped, then
4704 			 * start it.
4705 			 */
4706 			error = if_init(ifp);
4707 			break;
4708 		default:
4709 			break;
4710 		}
4711 		return error;
4712 #ifdef WG_RUMPKERNEL
4713 	case SIOCSLINKSTR:
4714 		error = wg_ioctl_linkstr(wg, ifd);
4715 		if (error == 0)
4716 			wg->wg_ops = &wg_ops_rumpuser;
4717 		return error;
4718 #endif
4719 	default:
4720 		break;
4721 	}
4722 
4723 	error = ifioctl_common(ifp, cmd, data);
4724 
4725 #ifdef WG_RUMPKERNEL
4726 	if (!wg_user_mode(wg))
4727 		return error;
4728 
4729 	/* Do the same to the corresponding tun device on the host */
4730 	/*
4731 	 * XXX Actually the command has not been handled yet.  It
4732 	 *     will be handled via pr_ioctl form doifioctl later.
4733 	 */
4734 	switch (cmd) {
4735 	case SIOCAIFADDR:
4736 	case SIOCDIFADDR: {
4737 		struct in_aliasreq _ifra = *(const struct in_aliasreq *)data;
4738 		struct in_aliasreq *ifra = &_ifra;
4739 		KASSERT(error == ENOTTY);
4740 		strncpy(ifra->ifra_name, rumpuser_wg_get_tunname(wg->wg_user),
4741 		    IFNAMSIZ);
4742 		error = rumpuser_wg_ioctl(wg->wg_user, cmd, ifra, AF_INET);
4743 		if (error == 0)
4744 			error = ENOTTY;
4745 		break;
4746 	}
4747 #ifdef INET6
4748 	case SIOCAIFADDR_IN6:
4749 	case SIOCDIFADDR_IN6: {
4750 		struct in6_aliasreq _ifra = *(const struct in6_aliasreq *)data;
4751 		struct in6_aliasreq *ifra = &_ifra;
4752 		KASSERT(error == ENOTTY);
4753 		strncpy(ifra->ifra_name, rumpuser_wg_get_tunname(wg->wg_user),
4754 		    IFNAMSIZ);
4755 		error = rumpuser_wg_ioctl(wg->wg_user, cmd, ifra, AF_INET6);
4756 		if (error == 0)
4757 			error = ENOTTY;
4758 		break;
4759 	}
4760 #endif
4761 	}
4762 #endif /* WG_RUMPKERNEL */
4763 
4764 	return error;
4765 }
4766 
4767 static int
4768 wg_init(struct ifnet *ifp)
4769 {
4770 
4771 	ifp->if_flags |= IFF_RUNNING;
4772 
4773 	/* TODO flush pending packets. */
4774 	return 0;
4775 }
4776 
4777 #ifdef ALTQ
4778 static void
4779 wg_start(struct ifnet *ifp)
4780 {
4781 	struct mbuf *m;
4782 
4783 	for (;;) {
4784 		IFQ_DEQUEUE(&ifp->if_snd, m);
4785 		if (m == NULL)
4786 			break;
4787 
4788 		kpreempt_disable();
4789 		const uint32_t h = curcpu()->ci_index;	// pktq_rps_hash(m)
4790 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
4791 			WGLOG(LOG_ERR, "pktq full, dropping\n");
4792 			m_freem(m);
4793 		}
4794 		kpreempt_enable();
4795 	}
4796 }
4797 #endif
4798 
4799 static void
4800 wg_stop(struct ifnet *ifp, int disable)
4801 {
4802 
4803 	KASSERT((ifp->if_flags & IFF_RUNNING) != 0);
4804 	ifp->if_flags &= ~IFF_RUNNING;
4805 
4806 	/* Need to do something? */
4807 }
4808 
4809 #ifdef WG_DEBUG_PARAMS
4810 SYSCTL_SETUP(sysctl_net_wg_setup, "sysctl net.wg setup")
4811 {
4812 	const struct sysctlnode *node = NULL;
4813 
4814 	sysctl_createv(clog, 0, NULL, &node,
4815 	    CTLFLAG_PERMANENT,
4816 	    CTLTYPE_NODE, "wg",
4817 	    SYSCTL_DESCR("wg(4)"),
4818 	    NULL, 0, NULL, 0,
4819 	    CTL_NET, CTL_CREATE, CTL_EOL);
4820 	sysctl_createv(clog, 0, &node, NULL,
4821 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4822 	    CTLTYPE_QUAD, "rekey_after_messages",
4823 	    SYSCTL_DESCR("session liftime by messages"),
4824 	    NULL, 0, &wg_rekey_after_messages, 0, CTL_CREATE, CTL_EOL);
4825 	sysctl_createv(clog, 0, &node, NULL,
4826 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4827 	    CTLTYPE_INT, "rekey_after_time",
4828 	    SYSCTL_DESCR("session liftime"),
4829 	    NULL, 0, &wg_rekey_after_time, 0, CTL_CREATE, CTL_EOL);
4830 	sysctl_createv(clog, 0, &node, NULL,
4831 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4832 	    CTLTYPE_INT, "rekey_timeout",
4833 	    SYSCTL_DESCR("session handshake retry time"),
4834 	    NULL, 0, &wg_rekey_timeout, 0, CTL_CREATE, CTL_EOL);
4835 	sysctl_createv(clog, 0, &node, NULL,
4836 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4837 	    CTLTYPE_INT, "rekey_attempt_time",
4838 	    SYSCTL_DESCR("session handshake timeout"),
4839 	    NULL, 0, &wg_rekey_attempt_time, 0, CTL_CREATE, CTL_EOL);
4840 	sysctl_createv(clog, 0, &node, NULL,
4841 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4842 	    CTLTYPE_INT, "keepalive_timeout",
4843 	    SYSCTL_DESCR("keepalive timeout"),
4844 	    NULL, 0, &wg_keepalive_timeout, 0, CTL_CREATE, CTL_EOL);
4845 	sysctl_createv(clog, 0, &node, NULL,
4846 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
4847 	    CTLTYPE_BOOL, "force_underload",
4848 	    SYSCTL_DESCR("force to detemine under load"),
4849 	    NULL, 0, &wg_force_underload, 0, CTL_CREATE, CTL_EOL);
4850 }
4851 #endif
4852 
4853 #ifdef WG_RUMPKERNEL
4854 static bool
4855 wg_user_mode(struct wg_softc *wg)
4856 {
4857 
4858 	return wg->wg_user != NULL;
4859 }
4860 
4861 static int
4862 wg_ioctl_linkstr(struct wg_softc *wg, struct ifdrv *ifd)
4863 {
4864 	struct ifnet *ifp = &wg->wg_if;
4865 	int error;
4866 
4867 	if (ifp->if_flags & IFF_UP)
4868 		return EBUSY;
4869 
4870 	if (ifd->ifd_cmd == IFLINKSTR_UNSET) {
4871 		/* XXX do nothing */
4872 		return 0;
4873 	} else if (ifd->ifd_cmd != 0) {
4874 		return EINVAL;
4875 	} else if (wg->wg_user != NULL) {
4876 		return EBUSY;
4877 	}
4878 
4879 	/* Assume \0 included */
4880 	if (ifd->ifd_len > IFNAMSIZ) {
4881 		return E2BIG;
4882 	} else if (ifd->ifd_len < 1) {
4883 		return EINVAL;
4884 	}
4885 
4886 	char tun_name[IFNAMSIZ];
4887 	error = copyinstr(ifd->ifd_data, tun_name, ifd->ifd_len, NULL);
4888 	if (error != 0)
4889 		return error;
4890 
4891 	if (strncmp(tun_name, "tun", 3) != 0)
4892 		return EINVAL;
4893 
4894 	error = rumpuser_wg_create(tun_name, wg, &wg->wg_user);
4895 
4896 	return error;
4897 }
4898 
4899 static int
4900 wg_send_user(struct wg_peer *wgp, struct mbuf *m)
4901 {
4902 	int error;
4903 	struct psref psref;
4904 	struct wg_sockaddr *wgsa;
4905 	struct wg_softc *wg = wgp->wgp_sc;
4906 	struct iovec iov[1];
4907 
4908 	wgsa = wg_get_endpoint_sa(wgp, &psref);
4909 
4910 	iov[0].iov_base = mtod(m, void *);
4911 	iov[0].iov_len = m->m_len;
4912 
4913 	/* Send messages to a peer via an ordinary socket. */
4914 	error = rumpuser_wg_send_peer(wg->wg_user, wgsatosa(wgsa), iov, 1);
4915 
4916 	wg_put_sa(wgp, wgsa, &psref);
4917 
4918 	m_freem(m);
4919 
4920 	return error;
4921 }
4922 
4923 static void
4924 wg_input_user(struct ifnet *ifp, struct mbuf *m, const int af)
4925 {
4926 	struct wg_softc *wg = ifp->if_softc;
4927 	struct iovec iov[2];
4928 	struct sockaddr_storage ss;
4929 
4930 	KASSERT(af == AF_INET || af == AF_INET6);
4931 
4932 	WG_TRACE("");
4933 
4934 	if (af == AF_INET) {
4935 		struct sockaddr_in *sin = (struct sockaddr_in *)&ss;
4936 		struct ip *ip;
4937 
4938 		KASSERT(m->m_len >= sizeof(struct ip));
4939 		ip = mtod(m, struct ip *);
4940 		sockaddr_in_init(sin, &ip->ip_dst, 0);
4941 	} else {
4942 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ss;
4943 		struct ip6_hdr *ip6;
4944 
4945 		KASSERT(m->m_len >= sizeof(struct ip6_hdr));
4946 		ip6 = mtod(m, struct ip6_hdr *);
4947 		sockaddr_in6_init(sin6, &ip6->ip6_dst, 0, 0, 0);
4948 	}
4949 
4950 	iov[0].iov_base = &ss;
4951 	iov[0].iov_len = ss.ss_len;
4952 	iov[1].iov_base = mtod(m, void *);
4953 	iov[1].iov_len = m->m_len;
4954 
4955 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
4956 
4957 	/* Send decrypted packets to users via a tun. */
4958 	rumpuser_wg_send_user(wg->wg_user, iov, 2);
4959 
4960 	m_freem(m);
4961 }
4962 
4963 static int
4964 wg_bind_port_user(struct wg_softc *wg, const uint16_t port)
4965 {
4966 	int error;
4967 	uint16_t old_port = wg->wg_listen_port;
4968 
4969 	if (port != 0 && old_port == port)
4970 		return 0;
4971 
4972 	error = rumpuser_wg_sock_bind(wg->wg_user, port);
4973 	if (error == 0)
4974 		wg->wg_listen_port = port;
4975 	return error;
4976 }
4977 
4978 /*
4979  * Receive user packets.
4980  */
4981 void
4982 rumpkern_wg_recv_user(struct wg_softc *wg, struct iovec *iov, size_t iovlen)
4983 {
4984 	struct ifnet *ifp = &wg->wg_if;
4985 	struct mbuf *m;
4986 	const struct sockaddr *dst;
4987 
4988 	WG_TRACE("");
4989 
4990 	dst = iov[0].iov_base;
4991 
4992 	m = m_gethdr(M_DONTWAIT, MT_DATA);
4993 	if (m == NULL)
4994 		return;
4995 	m->m_len = m->m_pkthdr.len = 0;
4996 	m_copyback(m, 0, iov[1].iov_len, iov[1].iov_base);
4997 
4998 	WG_DLOG("iov_len=%lu\n", iov[1].iov_len);
4999 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
5000 
5001 	(void)wg_output(ifp, m, dst, NULL);
5002 }
5003 
5004 /*
5005  * Receive packets from a peer.
5006  */
5007 void
5008 rumpkern_wg_recv_peer(struct wg_softc *wg, struct iovec *iov, size_t iovlen)
5009 {
5010 	struct mbuf *m;
5011 	const struct sockaddr *src;
5012 
5013 	WG_TRACE("");
5014 
5015 	src = iov[0].iov_base;
5016 
5017 	m = m_gethdr(M_DONTWAIT, MT_DATA);
5018 	if (m == NULL)
5019 		return;
5020 	m->m_len = m->m_pkthdr.len = 0;
5021 	m_copyback(m, 0, iov[1].iov_len, iov[1].iov_base);
5022 
5023 	WG_DLOG("iov_len=%lu\n", iov[1].iov_len);
5024 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
5025 
5026 	wg_handle_packet(wg, m, src);
5027 }
5028 #endif /* WG_RUMPKERNEL */
5029 
5030 /*
5031  * Module infrastructure
5032  */
5033 #include "if_module.h"
5034 
5035 IF_MODULE(MODULE_CLASS_DRIVER, wg, "sodium,blake2s")
5036