xref: /freebsd/sys/dev/wg/if_wg.c (revision 3b68c491d37196bb76a95bce3c02f7c6d5ba22fd)
1 /* SPDX-License-Identifier: ISC
2  *
3  * Copyright (C) 2015-2021 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
4  * Copyright (C) 2019-2021 Matt Dunwoodie <ncon@noconroy.net>
5  * Copyright (c) 2019-2020 Rubicon Communications, LLC (Netgate)
6  * Copyright (c) 2021 Kyle Evans <kevans@FreeBSD.org>
7  * Copyright (c) 2022 The FreeBSD Foundation
8  */
9 
10 #include "opt_inet.h"
11 #include "opt_inet6.h"
12 
13 #include <sys/param.h>
14 #include <sys/systm.h>
15 #include <sys/counter.h>
16 #include <sys/gtaskqueue.h>
17 #include <sys/jail.h>
18 #include <sys/kernel.h>
19 #include <sys/lock.h>
20 #include <sys/mbuf.h>
21 #include <sys/module.h>
22 #include <sys/nv.h>
23 #include <sys/priv.h>
24 #include <sys/protosw.h>
25 #include <sys/rmlock.h>
26 #include <sys/rwlock.h>
27 #include <sys/smp.h>
28 #include <sys/socket.h>
29 #include <sys/socketvar.h>
30 #include <sys/sockio.h>
31 #include <sys/sysctl.h>
32 #include <sys/sx.h>
33 #include <machine/_inttypes.h>
34 #include <net/bpf.h>
35 #include <net/ethernet.h>
36 #include <net/if.h>
37 #include <net/if_clone.h>
38 #include <net/if_types.h>
39 #include <net/if_var.h>
40 #include <net/netisr.h>
41 #include <net/radix.h>
42 #include <netinet/in.h>
43 #include <netinet6/in6_var.h>
44 #include <netinet/ip.h>
45 #include <netinet/ip6.h>
46 #include <netinet/ip_icmp.h>
47 #include <netinet/icmp6.h>
48 #include <netinet/udp_var.h>
49 #include <netinet6/nd6.h>
50 
51 #include "wg_noise.h"
52 #include "wg_cookie.h"
53 #include "version.h"
54 #include "if_wg.h"
55 
56 #define DEFAULT_MTU		(ETHERMTU - 80)
57 #define MAX_MTU			(IF_MAXMTU - 80)
58 
59 #define MAX_STAGED_PKT		128
60 #define MAX_QUEUED_PKT		1024
61 #define MAX_QUEUED_PKT_MASK	(MAX_QUEUED_PKT - 1)
62 
63 #define MAX_QUEUED_HANDSHAKES	4096
64 
65 #define REKEY_TIMEOUT_JITTER	334 /* 1/3 sec, round for arc4random_uniform */
66 #define MAX_TIMER_HANDSHAKES	(90 / REKEY_TIMEOUT)
67 #define NEW_HANDSHAKE_TIMEOUT	(REKEY_TIMEOUT + KEEPALIVE_TIMEOUT)
68 #define UNDERLOAD_TIMEOUT	1
69 
70 #define DPRINTF(sc, ...) if (if_getflags(sc->sc_ifp) & IFF_DEBUG) if_printf(sc->sc_ifp, ##__VA_ARGS__)
71 
72 /* First byte indicating packet type on the wire */
73 #define WG_PKT_INITIATION htole32(1)
74 #define WG_PKT_RESPONSE htole32(2)
75 #define WG_PKT_COOKIE htole32(3)
76 #define WG_PKT_DATA htole32(4)
77 
78 #define WG_PKT_PADDING		16
79 #define WG_KEY_SIZE		32
80 
81 struct wg_pkt_initiation {
82 	uint32_t		t;
83 	uint32_t		s_idx;
84 	uint8_t			ue[NOISE_PUBLIC_KEY_LEN];
85 	uint8_t			es[NOISE_PUBLIC_KEY_LEN + NOISE_AUTHTAG_LEN];
86 	uint8_t			ets[NOISE_TIMESTAMP_LEN + NOISE_AUTHTAG_LEN];
87 	struct cookie_macs	m;
88 };
89 
90 struct wg_pkt_response {
91 	uint32_t		t;
92 	uint32_t		s_idx;
93 	uint32_t		r_idx;
94 	uint8_t			ue[NOISE_PUBLIC_KEY_LEN];
95 	uint8_t			en[0 + NOISE_AUTHTAG_LEN];
96 	struct cookie_macs	m;
97 };
98 
99 struct wg_pkt_cookie {
100 	uint32_t		t;
101 	uint32_t		r_idx;
102 	uint8_t			nonce[COOKIE_NONCE_SIZE];
103 	uint8_t			ec[COOKIE_ENCRYPTED_SIZE];
104 };
105 
106 struct wg_pkt_data {
107 	uint32_t		t;
108 	uint32_t		r_idx;
109 	uint64_t		nonce;
110 	uint8_t			buf[];
111 };
112 
113 struct wg_endpoint {
114 	union {
115 		struct sockaddr		r_sa;
116 		struct sockaddr_in	r_sin;
117 #ifdef INET6
118 		struct sockaddr_in6	r_sin6;
119 #endif
120 	} e_remote;
121 	union {
122 		struct in_addr		l_in;
123 #ifdef INET6
124 		struct in6_pktinfo	l_pktinfo6;
125 #define l_in6 l_pktinfo6.ipi6_addr
126 #endif
127 	} e_local;
128 };
129 
130 struct aip_addr {
131 	uint8_t		length;
132 	union {
133 		uint8_t		bytes[16];
134 		uint32_t	ip;
135 		uint32_t	ip6[4];
136 		struct in_addr	in;
137 		struct in6_addr	in6;
138 	};
139 };
140 
141 struct wg_aip {
142 	struct radix_node	 a_nodes[2];
143 	LIST_ENTRY(wg_aip)	 a_entry;
144 	struct aip_addr		 a_addr;
145 	struct aip_addr		 a_mask;
146 	struct wg_peer		*a_peer;
147 	sa_family_t		 a_af;
148 };
149 
150 struct wg_packet {
151 	STAILQ_ENTRY(wg_packet)	 p_serial;
152 	STAILQ_ENTRY(wg_packet)	 p_parallel;
153 	struct wg_endpoint	 p_endpoint;
154 	struct noise_keypair	*p_keypair;
155 	uint64_t		 p_nonce;
156 	struct mbuf		*p_mbuf;
157 	int			 p_mtu;
158 	sa_family_t		 p_af;
159 	enum wg_ring_state {
160 		WG_PACKET_UNCRYPTED,
161 		WG_PACKET_CRYPTED,
162 		WG_PACKET_DEAD,
163 	}			 p_state;
164 };
165 
166 STAILQ_HEAD(wg_packet_list, wg_packet);
167 
168 struct wg_queue {
169 	struct mtx		 q_mtx;
170 	struct wg_packet_list	 q_queue;
171 	size_t			 q_len;
172 };
173 
174 struct wg_peer {
175 	TAILQ_ENTRY(wg_peer)		 p_entry;
176 	uint64_t			 p_id;
177 	struct wg_softc			*p_sc;
178 
179 	struct noise_remote		*p_remote;
180 	struct cookie_maker		 p_cookie;
181 
182 	struct rwlock			 p_endpoint_lock;
183 	struct wg_endpoint		 p_endpoint;
184 
185 	struct wg_queue	 		 p_stage_queue;
186 	struct wg_queue	 		 p_encrypt_serial;
187 	struct wg_queue	 		 p_decrypt_serial;
188 
189 	bool				 p_enabled;
190 	bool				 p_need_another_keepalive;
191 	uint16_t			 p_persistent_keepalive_interval;
192 	struct callout			 p_new_handshake;
193 	struct callout			 p_send_keepalive;
194 	struct callout			 p_retry_handshake;
195 	struct callout			 p_zero_key_material;
196 	struct callout			 p_persistent_keepalive;
197 
198 	struct mtx			 p_handshake_mtx;
199 	struct timespec			 p_handshake_complete;	/* nanotime */
200 	int				 p_handshake_retries;
201 
202 	struct grouptask		 p_send;
203 	struct grouptask		 p_recv;
204 
205 	counter_u64_t			 p_tx_bytes;
206 	counter_u64_t			 p_rx_bytes;
207 
208 	LIST_HEAD(, wg_aip)		 p_aips;
209 	size_t				 p_aips_num;
210 };
211 
212 struct wg_socket {
213 	struct socket	*so_so4;
214 	struct socket	*so_so6;
215 	uint32_t	 so_user_cookie;
216 	int		 so_fibnum;
217 	in_port_t	 so_port;
218 };
219 
220 struct wg_softc {
221 	LIST_ENTRY(wg_softc)	 sc_entry;
222 	if_t			 sc_ifp;
223 	int			 sc_flags;
224 
225 	struct ucred		*sc_ucred;
226 	struct wg_socket	 sc_socket;
227 
228 	TAILQ_HEAD(,wg_peer)	 sc_peers;
229 	size_t			 sc_peers_num;
230 
231 	struct noise_local	*sc_local;
232 	struct cookie_checker	 sc_cookie;
233 
234 	struct radix_node_head	*sc_aip4;
235 	struct radix_node_head	*sc_aip6;
236 
237 	struct grouptask	 sc_handshake;
238 	struct wg_queue		 sc_handshake_queue;
239 
240 	struct grouptask	*sc_encrypt;
241 	struct grouptask	*sc_decrypt;
242 	struct wg_queue		 sc_encrypt_parallel;
243 	struct wg_queue		 sc_decrypt_parallel;
244 	u_int			 sc_encrypt_last_cpu;
245 	u_int			 sc_decrypt_last_cpu;
246 
247 	struct sx		 sc_lock;
248 };
249 
250 #define	WGF_DYING	0x0001
251 
252 #define MAX_LOOPS	8
253 #define MTAG_WGLOOP	0x77676c70 /* wglp */
254 
255 #define	GROUPTASK_DRAIN(gtask)			\
256 	gtaskqueue_drain((gtask)->gt_taskqueue, &(gtask)->gt_task)
257 
258 #define BPF_MTAP2_AF(ifp, m, af) do { \
259 		uint32_t __bpf_tap_af = (af); \
260 		BPF_MTAP2(ifp, &__bpf_tap_af, sizeof(__bpf_tap_af), m); \
261 	} while (0)
262 
263 static int clone_count;
264 static uma_zone_t wg_packet_zone;
265 static volatile unsigned long peer_counter = 0;
266 static const char wgname[] = "wg";
267 static unsigned wg_osd_jail_slot;
268 
269 static struct sx wg_sx;
270 SX_SYSINIT(wg_sx, &wg_sx, "wg_sx");
271 
272 static LIST_HEAD(, wg_softc) wg_list = LIST_HEAD_INITIALIZER(wg_list);
273 
274 static TASKQGROUP_DEFINE(wg_tqg, mp_ncpus, 1);
275 
276 MALLOC_DEFINE(M_WG, "WG", "wireguard");
277 
278 VNET_DEFINE_STATIC(struct if_clone *, wg_cloner);
279 
280 #define	V_wg_cloner	VNET(wg_cloner)
281 #define	WG_CAPS		IFCAP_LINKSTATE
282 
283 struct wg_timespec64 {
284 	uint64_t	tv_sec;
285 	uint64_t	tv_nsec;
286 };
287 
288 static int wg_socket_init(struct wg_softc *, in_port_t);
289 static int wg_socket_bind(struct socket **, struct socket **, in_port_t *);
290 static void wg_socket_set(struct wg_softc *, struct socket *, struct socket *);
291 static void wg_socket_uninit(struct wg_softc *);
292 static int wg_socket_set_sockopt(struct socket *, struct socket *, int, void *, size_t);
293 static int wg_socket_set_cookie(struct wg_softc *, uint32_t);
294 static int wg_socket_set_fibnum(struct wg_softc *, int);
295 static int wg_send(struct wg_softc *, struct wg_endpoint *, struct mbuf *);
296 static void wg_timers_enable(struct wg_peer *);
297 static void wg_timers_disable(struct wg_peer *);
298 static void wg_timers_set_persistent_keepalive(struct wg_peer *, uint16_t);
299 static void wg_timers_get_last_handshake(struct wg_peer *, struct wg_timespec64 *);
300 static void wg_timers_event_data_sent(struct wg_peer *);
301 static void wg_timers_event_data_received(struct wg_peer *);
302 static void wg_timers_event_any_authenticated_packet_sent(struct wg_peer *);
303 static void wg_timers_event_any_authenticated_packet_received(struct wg_peer *);
304 static void wg_timers_event_any_authenticated_packet_traversal(struct wg_peer *);
305 static void wg_timers_event_handshake_initiated(struct wg_peer *);
306 static void wg_timers_event_handshake_complete(struct wg_peer *);
307 static void wg_timers_event_session_derived(struct wg_peer *);
308 static void wg_timers_event_want_initiation(struct wg_peer *);
309 static void wg_timers_run_send_initiation(struct wg_peer *, bool);
310 static void wg_timers_run_retry_handshake(void *);
311 static void wg_timers_run_send_keepalive(void *);
312 static void wg_timers_run_new_handshake(void *);
313 static void wg_timers_run_zero_key_material(void *);
314 static void wg_timers_run_persistent_keepalive(void *);
315 static int wg_aip_add(struct wg_softc *, struct wg_peer *, sa_family_t, const void *, uint8_t);
316 static struct wg_peer *wg_aip_lookup(struct wg_softc *, sa_family_t, void *);
317 static void wg_aip_remove_all(struct wg_softc *, struct wg_peer *);
318 static struct wg_peer *wg_peer_alloc(struct wg_softc *, const uint8_t [WG_KEY_SIZE]);
319 static void wg_peer_free_deferred(struct noise_remote *);
320 static void wg_peer_destroy(struct wg_peer *);
321 static void wg_peer_destroy_all(struct wg_softc *);
322 static void wg_peer_send_buf(struct wg_peer *, uint8_t *, size_t);
323 static void wg_send_initiation(struct wg_peer *);
324 static void wg_send_response(struct wg_peer *);
325 static void wg_send_cookie(struct wg_softc *, struct cookie_macs *, uint32_t, struct wg_endpoint *);
326 static void wg_peer_set_endpoint(struct wg_peer *, struct wg_endpoint *);
327 static void wg_peer_clear_src(struct wg_peer *);
328 static void wg_peer_get_endpoint(struct wg_peer *, struct wg_endpoint *);
329 static void wg_send_buf(struct wg_softc *, struct wg_endpoint *, uint8_t *, size_t);
330 static void wg_send_keepalive(struct wg_peer *);
331 static void wg_handshake(struct wg_softc *, struct wg_packet *);
332 static void wg_encrypt(struct wg_softc *, struct wg_packet *);
333 static void wg_decrypt(struct wg_softc *, struct wg_packet *);
334 static void wg_softc_handshake_receive(struct wg_softc *);
335 static void wg_softc_decrypt(struct wg_softc *);
336 static void wg_softc_encrypt(struct wg_softc *);
337 static void wg_encrypt_dispatch(struct wg_softc *);
338 static void wg_decrypt_dispatch(struct wg_softc *);
339 static void wg_deliver_out(struct wg_peer *);
340 static void wg_deliver_in(struct wg_peer *);
341 static struct wg_packet *wg_packet_alloc(struct mbuf *);
342 static void wg_packet_free(struct wg_packet *);
343 static void wg_queue_init(struct wg_queue *, const char *);
344 static void wg_queue_deinit(struct wg_queue *);
345 static size_t wg_queue_len(struct wg_queue *);
346 static int wg_queue_enqueue_handshake(struct wg_queue *, struct wg_packet *);
347 static struct wg_packet *wg_queue_dequeue_handshake(struct wg_queue *);
348 static void wg_queue_push_staged(struct wg_queue *, struct wg_packet *);
349 static void wg_queue_enlist_staged(struct wg_queue *, struct wg_packet_list *);
350 static void wg_queue_delist_staged(struct wg_queue *, struct wg_packet_list *);
351 static void wg_queue_purge(struct wg_queue *);
352 static int wg_queue_both(struct wg_queue *, struct wg_queue *, struct wg_packet *);
353 static struct wg_packet *wg_queue_dequeue_serial(struct wg_queue *);
354 static struct wg_packet *wg_queue_dequeue_parallel(struct wg_queue *);
355 static bool wg_input(struct mbuf *, int, struct inpcb *, const struct sockaddr *, void *);
356 static void wg_peer_send_staged(struct wg_peer *);
357 static int wg_clone_create(struct if_clone *ifc, char *name, size_t len,
358 	struct ifc_data *ifd, if_t *ifpp);
359 static void wg_qflush(if_t);
360 static inline int determine_af_and_pullup(struct mbuf **m, sa_family_t *af);
361 static int wg_xmit(if_t, struct mbuf *, sa_family_t, uint32_t);
362 static int wg_transmit(if_t, struct mbuf *);
363 static int wg_output(if_t, struct mbuf *, const struct sockaddr *, struct route *);
364 static int wg_clone_destroy(struct if_clone *ifc, if_t ifp,
365 	uint32_t flags);
366 static bool wgc_privileged(struct wg_softc *);
367 static int wgc_get(struct wg_softc *, struct wg_data_io *);
368 static int wgc_set(struct wg_softc *, struct wg_data_io *);
369 static int wg_up(struct wg_softc *);
370 static void wg_down(struct wg_softc *);
371 static void wg_reassign(if_t, struct vnet *, char *unused);
372 static void wg_init(void *);
373 static int wg_ioctl(if_t, u_long, caddr_t);
374 static void vnet_wg_init(const void *);
375 static void vnet_wg_uninit(const void *);
376 static int wg_module_init(void);
377 static void wg_module_deinit(void);
378 
379 /* TODO Peer */
380 static struct wg_peer *
381 wg_peer_alloc(struct wg_softc *sc, const uint8_t pub_key[WG_KEY_SIZE])
382 {
383 	struct wg_peer *peer;
384 
385 	sx_assert(&sc->sc_lock, SX_XLOCKED);
386 
387 	peer = malloc(sizeof(*peer), M_WG, M_WAITOK | M_ZERO);
388 	peer->p_remote = noise_remote_alloc(sc->sc_local, peer, pub_key);
389 	peer->p_tx_bytes = counter_u64_alloc(M_WAITOK);
390 	peer->p_rx_bytes = counter_u64_alloc(M_WAITOK);
391 	peer->p_id = peer_counter++;
392 	peer->p_sc = sc;
393 
394 	cookie_maker_init(&peer->p_cookie, pub_key);
395 
396 	rw_init(&peer->p_endpoint_lock, "wg_peer_endpoint");
397 
398 	wg_queue_init(&peer->p_stage_queue, "stageq");
399 	wg_queue_init(&peer->p_encrypt_serial, "txq");
400 	wg_queue_init(&peer->p_decrypt_serial, "rxq");
401 
402 	peer->p_enabled = false;
403 	peer->p_need_another_keepalive = false;
404 	peer->p_persistent_keepalive_interval = 0;
405 	callout_init(&peer->p_new_handshake, true);
406 	callout_init(&peer->p_send_keepalive, true);
407 	callout_init(&peer->p_retry_handshake, true);
408 	callout_init(&peer->p_persistent_keepalive, true);
409 	callout_init(&peer->p_zero_key_material, true);
410 
411 	mtx_init(&peer->p_handshake_mtx, "peer handshake", NULL, MTX_DEF);
412 	bzero(&peer->p_handshake_complete, sizeof(peer->p_handshake_complete));
413 	peer->p_handshake_retries = 0;
414 
415 	GROUPTASK_INIT(&peer->p_send, 0, (gtask_fn_t *)wg_deliver_out, peer);
416 	taskqgroup_attach(qgroup_wg_tqg, &peer->p_send, peer, NULL, NULL, "wg send");
417 	GROUPTASK_INIT(&peer->p_recv, 0, (gtask_fn_t *)wg_deliver_in, peer);
418 	taskqgroup_attach(qgroup_wg_tqg, &peer->p_recv, peer, NULL, NULL, "wg recv");
419 
420 	LIST_INIT(&peer->p_aips);
421 	peer->p_aips_num = 0;
422 
423 	return (peer);
424 }
425 
426 static void
427 wg_peer_free_deferred(struct noise_remote *r)
428 {
429 	struct wg_peer *peer = noise_remote_arg(r);
430 
431 	/* While there are no references remaining, we may still have
432 	 * p_{send,recv} executing (think empty queue, but wg_deliver_{in,out}
433 	 * needs to check the queue. We should wait for them and then free. */
434 	GROUPTASK_DRAIN(&peer->p_recv);
435 	GROUPTASK_DRAIN(&peer->p_send);
436 	taskqgroup_detach(qgroup_wg_tqg, &peer->p_recv);
437 	taskqgroup_detach(qgroup_wg_tqg, &peer->p_send);
438 
439 	wg_queue_deinit(&peer->p_decrypt_serial);
440 	wg_queue_deinit(&peer->p_encrypt_serial);
441 	wg_queue_deinit(&peer->p_stage_queue);
442 
443 	counter_u64_free(peer->p_tx_bytes);
444 	counter_u64_free(peer->p_rx_bytes);
445 	rw_destroy(&peer->p_endpoint_lock);
446 	mtx_destroy(&peer->p_handshake_mtx);
447 
448 	cookie_maker_free(&peer->p_cookie);
449 
450 	free(peer, M_WG);
451 }
452 
453 static void
454 wg_peer_destroy(struct wg_peer *peer)
455 {
456 	struct wg_softc *sc = peer->p_sc;
457 	sx_assert(&sc->sc_lock, SX_XLOCKED);
458 
459 	/* Disable remote and timers. This will prevent any new handshakes
460 	 * occuring. */
461 	noise_remote_disable(peer->p_remote);
462 	wg_timers_disable(peer);
463 
464 	/* Now we can remove all allowed IPs so no more packets will be routed
465 	 * to the peer. */
466 	wg_aip_remove_all(sc, peer);
467 
468 	/* Remove peer from the interface, then free. Some references may still
469 	 * exist to p_remote, so noise_remote_free will wait until they're all
470 	 * put to call wg_peer_free_deferred. */
471 	sc->sc_peers_num--;
472 	TAILQ_REMOVE(&sc->sc_peers, peer, p_entry);
473 	DPRINTF(sc, "Peer %" PRIu64 " destroyed\n", peer->p_id);
474 	noise_remote_free(peer->p_remote, wg_peer_free_deferred);
475 }
476 
477 static void
478 wg_peer_destroy_all(struct wg_softc *sc)
479 {
480 	struct wg_peer *peer, *tpeer;
481 	TAILQ_FOREACH_SAFE(peer, &sc->sc_peers, p_entry, tpeer)
482 		wg_peer_destroy(peer);
483 }
484 
485 static void
486 wg_peer_set_endpoint(struct wg_peer *peer, struct wg_endpoint *e)
487 {
488 	MPASS(e->e_remote.r_sa.sa_family != 0);
489 	if (memcmp(e, &peer->p_endpoint, sizeof(*e)) == 0)
490 		return;
491 
492 	rw_wlock(&peer->p_endpoint_lock);
493 	peer->p_endpoint = *e;
494 	rw_wunlock(&peer->p_endpoint_lock);
495 }
496 
497 static void
498 wg_peer_clear_src(struct wg_peer *peer)
499 {
500 	rw_wlock(&peer->p_endpoint_lock);
501 	bzero(&peer->p_endpoint.e_local, sizeof(peer->p_endpoint.e_local));
502 	rw_wunlock(&peer->p_endpoint_lock);
503 }
504 
505 static void
506 wg_peer_get_endpoint(struct wg_peer *peer, struct wg_endpoint *e)
507 {
508 	rw_rlock(&peer->p_endpoint_lock);
509 	*e = peer->p_endpoint;
510 	rw_runlock(&peer->p_endpoint_lock);
511 }
512 
513 /* Allowed IP */
514 static int
515 wg_aip_add(struct wg_softc *sc, struct wg_peer *peer, sa_family_t af, const void *addr, uint8_t cidr)
516 {
517 	struct radix_node_head	*root;
518 	struct radix_node	*node;
519 	struct wg_aip		*aip;
520 	int			 ret = 0;
521 
522 	aip = malloc(sizeof(*aip), M_WG, M_WAITOK | M_ZERO);
523 	aip->a_peer = peer;
524 	aip->a_af = af;
525 
526 	switch (af) {
527 #ifdef INET
528 	case AF_INET:
529 		if (cidr > 32) cidr = 32;
530 		root = sc->sc_aip4;
531 		aip->a_addr.in = *(const struct in_addr *)addr;
532 		aip->a_mask.ip = htonl(~((1LL << (32 - cidr)) - 1) & 0xffffffff);
533 		aip->a_addr.ip &= aip->a_mask.ip;
534 		aip->a_addr.length = aip->a_mask.length = offsetof(struct aip_addr, in) + sizeof(struct in_addr);
535 		break;
536 #endif
537 #ifdef INET6
538 	case AF_INET6:
539 		if (cidr > 128) cidr = 128;
540 		root = sc->sc_aip6;
541 		aip->a_addr.in6 = *(const struct in6_addr *)addr;
542 		in6_prefixlen2mask(&aip->a_mask.in6, cidr);
543 		for (int i = 0; i < 4; i++)
544 			aip->a_addr.ip6[i] &= aip->a_mask.ip6[i];
545 		aip->a_addr.length = aip->a_mask.length = offsetof(struct aip_addr, in6) + sizeof(struct in6_addr);
546 		break;
547 #endif
548 	default:
549 		free(aip, M_WG);
550 		return (EAFNOSUPPORT);
551 	}
552 
553 	RADIX_NODE_HEAD_LOCK(root);
554 	node = root->rnh_addaddr(&aip->a_addr, &aip->a_mask, &root->rh, aip->a_nodes);
555 	if (node == aip->a_nodes) {
556 		LIST_INSERT_HEAD(&peer->p_aips, aip, a_entry);
557 		peer->p_aips_num++;
558 	} else if (!node)
559 		node = root->rnh_lookup(&aip->a_addr, &aip->a_mask, &root->rh);
560 	if (!node) {
561 		free(aip, M_WG);
562 		ret = ENOMEM;
563 	} else if (node != aip->a_nodes) {
564 		free(aip, M_WG);
565 		aip = (struct wg_aip *)node;
566 		if (aip->a_peer != peer) {
567 			LIST_REMOVE(aip, a_entry);
568 			aip->a_peer->p_aips_num--;
569 			aip->a_peer = peer;
570 			LIST_INSERT_HEAD(&peer->p_aips, aip, a_entry);
571 			aip->a_peer->p_aips_num++;
572 		}
573 	}
574 	RADIX_NODE_HEAD_UNLOCK(root);
575 	return (ret);
576 }
577 
578 static struct wg_peer *
579 wg_aip_lookup(struct wg_softc *sc, sa_family_t af, void *a)
580 {
581 	struct radix_node_head	*root;
582 	struct radix_node	*node;
583 	struct wg_peer		*peer;
584 	struct aip_addr		 addr;
585 	RADIX_NODE_HEAD_RLOCK_TRACKER;
586 
587 	switch (af) {
588 	case AF_INET:
589 		root = sc->sc_aip4;
590 		memcpy(&addr.in, a, sizeof(addr.in));
591 		addr.length = offsetof(struct aip_addr, in) + sizeof(struct in_addr);
592 		break;
593 	case AF_INET6:
594 		root = sc->sc_aip6;
595 		memcpy(&addr.in6, a, sizeof(addr.in6));
596 		addr.length = offsetof(struct aip_addr, in6) + sizeof(struct in6_addr);
597 		break;
598 	default:
599 		return NULL;
600 	}
601 
602 	RADIX_NODE_HEAD_RLOCK(root);
603 	node = root->rnh_matchaddr(&addr, &root->rh);
604 	if (node != NULL) {
605 		peer = ((struct wg_aip *)node)->a_peer;
606 		noise_remote_ref(peer->p_remote);
607 	} else {
608 		peer = NULL;
609 	}
610 	RADIX_NODE_HEAD_RUNLOCK(root);
611 
612 	return (peer);
613 }
614 
615 static void
616 wg_aip_remove_all(struct wg_softc *sc, struct wg_peer *peer)
617 {
618 	struct wg_aip		*aip, *taip;
619 
620 	RADIX_NODE_HEAD_LOCK(sc->sc_aip4);
621 	LIST_FOREACH_SAFE(aip, &peer->p_aips, a_entry, taip) {
622 		if (aip->a_af == AF_INET) {
623 			if (sc->sc_aip4->rnh_deladdr(&aip->a_addr, &aip->a_mask, &sc->sc_aip4->rh) == NULL)
624 				panic("failed to delete aip %p", aip);
625 			LIST_REMOVE(aip, a_entry);
626 			peer->p_aips_num--;
627 			free(aip, M_WG);
628 		}
629 	}
630 	RADIX_NODE_HEAD_UNLOCK(sc->sc_aip4);
631 
632 	RADIX_NODE_HEAD_LOCK(sc->sc_aip6);
633 	LIST_FOREACH_SAFE(aip, &peer->p_aips, a_entry, taip) {
634 		if (aip->a_af == AF_INET6) {
635 			if (sc->sc_aip6->rnh_deladdr(&aip->a_addr, &aip->a_mask, &sc->sc_aip6->rh) == NULL)
636 				panic("failed to delete aip %p", aip);
637 			LIST_REMOVE(aip, a_entry);
638 			peer->p_aips_num--;
639 			free(aip, M_WG);
640 		}
641 	}
642 	RADIX_NODE_HEAD_UNLOCK(sc->sc_aip6);
643 
644 	if (!LIST_EMPTY(&peer->p_aips) || peer->p_aips_num != 0)
645 		panic("wg_aip_remove_all could not delete all %p", peer);
646 }
647 
648 static int
649 wg_socket_init(struct wg_softc *sc, in_port_t port)
650 {
651 	struct ucred *cred = sc->sc_ucred;
652 	struct socket *so4 = NULL, *so6 = NULL;
653 	int rc;
654 
655 	sx_assert(&sc->sc_lock, SX_XLOCKED);
656 
657 	if (!cred)
658 		return (EBUSY);
659 
660 	/*
661 	 * For socket creation, we use the creds of the thread that created the
662 	 * tunnel rather than the current thread to maintain the semantics that
663 	 * WireGuard has on Linux with network namespaces -- that the sockets
664 	 * are created in their home vnet so that they can be configured and
665 	 * functionally attached to a foreign vnet as the jail's only interface
666 	 * to the network.
667 	 */
668 #ifdef INET
669 	rc = socreate(AF_INET, &so4, SOCK_DGRAM, IPPROTO_UDP, cred, curthread);
670 	if (rc)
671 		goto out;
672 
673 	rc = udp_set_kernel_tunneling(so4, wg_input, NULL, sc);
674 	/*
675 	 * udp_set_kernel_tunneling can only fail if there is already a tunneling function set.
676 	 * This should never happen with a new socket.
677 	 */
678 	MPASS(rc == 0);
679 #endif
680 
681 #ifdef INET6
682 	rc = socreate(AF_INET6, &so6, SOCK_DGRAM, IPPROTO_UDP, cred, curthread);
683 	if (rc)
684 		goto out;
685 	rc = udp_set_kernel_tunneling(so6, wg_input, NULL, sc);
686 	MPASS(rc == 0);
687 #endif
688 
689 	if (sc->sc_socket.so_user_cookie) {
690 		rc = wg_socket_set_sockopt(so4, so6, SO_USER_COOKIE, &sc->sc_socket.so_user_cookie, sizeof(sc->sc_socket.so_user_cookie));
691 		if (rc)
692 			goto out;
693 	}
694 	rc = wg_socket_set_sockopt(so4, so6, SO_SETFIB, &sc->sc_socket.so_fibnum, sizeof(sc->sc_socket.so_fibnum));
695 	if (rc)
696 		goto out;
697 
698 	rc = wg_socket_bind(&so4, &so6, &port);
699 	if (!rc) {
700 		sc->sc_socket.so_port = port;
701 		wg_socket_set(sc, so4, so6);
702 	}
703 out:
704 	if (rc) {
705 		if (so4 != NULL)
706 			soclose(so4);
707 		if (so6 != NULL)
708 			soclose(so6);
709 	}
710 	return (rc);
711 }
712 
713 static int wg_socket_set_sockopt(struct socket *so4, struct socket *so6, int name, void *val, size_t len)
714 {
715 	int ret4 = 0, ret6 = 0;
716 	struct sockopt sopt = {
717 		.sopt_dir = SOPT_SET,
718 		.sopt_level = SOL_SOCKET,
719 		.sopt_name = name,
720 		.sopt_val = val,
721 		.sopt_valsize = len
722 	};
723 
724 	if (so4)
725 		ret4 = sosetopt(so4, &sopt);
726 	if (so6)
727 		ret6 = sosetopt(so6, &sopt);
728 	return (ret4 ?: ret6);
729 }
730 
731 static int wg_socket_set_cookie(struct wg_softc *sc, uint32_t user_cookie)
732 {
733 	struct wg_socket *so = &sc->sc_socket;
734 	int ret;
735 
736 	sx_assert(&sc->sc_lock, SX_XLOCKED);
737 	ret = wg_socket_set_sockopt(so->so_so4, so->so_so6, SO_USER_COOKIE, &user_cookie, sizeof(user_cookie));
738 	if (!ret)
739 		so->so_user_cookie = user_cookie;
740 	return (ret);
741 }
742 
743 static int wg_socket_set_fibnum(struct wg_softc *sc, int fibnum)
744 {
745 	struct wg_socket *so = &sc->sc_socket;
746 	int ret;
747 
748 	sx_assert(&sc->sc_lock, SX_XLOCKED);
749 
750 	ret = wg_socket_set_sockopt(so->so_so4, so->so_so6, SO_SETFIB, &fibnum, sizeof(fibnum));
751 	if (!ret)
752 		so->so_fibnum = fibnum;
753 	return (ret);
754 }
755 
756 static void
757 wg_socket_uninit(struct wg_softc *sc)
758 {
759 	wg_socket_set(sc, NULL, NULL);
760 }
761 
762 static void
763 wg_socket_set(struct wg_softc *sc, struct socket *new_so4, struct socket *new_so6)
764 {
765 	struct wg_socket *so = &sc->sc_socket;
766 	struct socket *so4, *so6;
767 
768 	sx_assert(&sc->sc_lock, SX_XLOCKED);
769 
770 	so4 = atomic_load_ptr(&so->so_so4);
771 	so6 = atomic_load_ptr(&so->so_so6);
772 	atomic_store_ptr(&so->so_so4, new_so4);
773 	atomic_store_ptr(&so->so_so6, new_so6);
774 
775 	if (!so4 && !so6)
776 		return;
777 	NET_EPOCH_WAIT();
778 	if (so4)
779 		soclose(so4);
780 	if (so6)
781 		soclose(so6);
782 }
783 
784 static int
785 wg_socket_bind(struct socket **in_so4, struct socket **in_so6, in_port_t *requested_port)
786 {
787 	struct socket *so4 = *in_so4, *so6 = *in_so6;
788 	int ret4 = 0, ret6 = 0;
789 	in_port_t port = *requested_port;
790 	struct sockaddr_in sin = {
791 		.sin_len = sizeof(struct sockaddr_in),
792 		.sin_family = AF_INET,
793 		.sin_port = htons(port)
794 	};
795 	struct sockaddr_in6 sin6 = {
796 		.sin6_len = sizeof(struct sockaddr_in6),
797 		.sin6_family = AF_INET6,
798 		.sin6_port = htons(port)
799 	};
800 
801 	if (so4) {
802 		ret4 = sobind(so4, (struct sockaddr *)&sin, curthread);
803 		if (ret4 && ret4 != EADDRNOTAVAIL)
804 			return (ret4);
805 		if (!ret4 && !sin.sin_port) {
806 			struct sockaddr_in bound_sin =
807 			    { .sin_len = sizeof(bound_sin) };
808 			int ret;
809 
810 			ret = sosockaddr(so4, (struct sockaddr *)&bound_sin);
811 			if (ret)
812 				return (ret);
813 			port = ntohs(bound_sin.sin_port);
814 			sin6.sin6_port = bound_sin.sin_port;
815 		}
816 	}
817 
818 	if (so6) {
819 		ret6 = sobind(so6, (struct sockaddr *)&sin6, curthread);
820 		if (ret6 && ret6 != EADDRNOTAVAIL)
821 			return (ret6);
822 		if (!ret6 && !sin6.sin6_port) {
823 			struct sockaddr_in6 bound_sin6 =
824 			    { .sin6_len = sizeof(bound_sin6) };
825 			int ret;
826 
827 			ret = sosockaddr(so6, (struct sockaddr *)&bound_sin6);
828 			if (ret)
829 				return (ret);
830 			port = ntohs(bound_sin6.sin6_port);
831 		}
832 	}
833 
834 	if (ret4 && ret6)
835 		return (ret4);
836 	*requested_port = port;
837 	if (ret4 && !ret6 && so4) {
838 		soclose(so4);
839 		*in_so4 = NULL;
840 	} else if (ret6 && !ret4 && so6) {
841 		soclose(so6);
842 		*in_so6 = NULL;
843 	}
844 	return (0);
845 }
846 
847 static int
848 wg_send(struct wg_softc *sc, struct wg_endpoint *e, struct mbuf *m)
849 {
850 	struct epoch_tracker et;
851 	struct sockaddr *sa;
852 	struct wg_socket *so = &sc->sc_socket;
853 	struct socket *so4, *so6;
854 	struct mbuf *control = NULL;
855 	int ret = 0;
856 	size_t len = m->m_pkthdr.len;
857 
858 	/* Get local control address before locking */
859 	if (e->e_remote.r_sa.sa_family == AF_INET) {
860 		if (e->e_local.l_in.s_addr != INADDR_ANY)
861 			control = sbcreatecontrol((caddr_t)&e->e_local.l_in,
862 			    sizeof(struct in_addr), IP_SENDSRCADDR,
863 			    IPPROTO_IP, M_NOWAIT);
864 #ifdef INET6
865 	} else if (e->e_remote.r_sa.sa_family == AF_INET6) {
866 		if (!IN6_IS_ADDR_UNSPECIFIED(&e->e_local.l_in6))
867 			control = sbcreatecontrol((caddr_t)&e->e_local.l_pktinfo6,
868 			    sizeof(struct in6_pktinfo), IPV6_PKTINFO,
869 			    IPPROTO_IPV6, M_NOWAIT);
870 #endif
871 	} else {
872 		m_freem(m);
873 		return (EAFNOSUPPORT);
874 	}
875 
876 	/* Get remote address */
877 	sa = &e->e_remote.r_sa;
878 
879 	NET_EPOCH_ENTER(et);
880 	so4 = atomic_load_ptr(&so->so_so4);
881 	so6 = atomic_load_ptr(&so->so_so6);
882 	if (e->e_remote.r_sa.sa_family == AF_INET && so4 != NULL)
883 		ret = sosend(so4, sa, NULL, m, control, 0, curthread);
884 	else if (e->e_remote.r_sa.sa_family == AF_INET6 && so6 != NULL)
885 		ret = sosend(so6, sa, NULL, m, control, 0, curthread);
886 	else {
887 		ret = ENOTCONN;
888 		m_freem(control);
889 		m_freem(m);
890 	}
891 	NET_EPOCH_EXIT(et);
892 	if (ret == 0) {
893 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
894 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, len);
895 	}
896 	return (ret);
897 }
898 
899 static void
900 wg_send_buf(struct wg_softc *sc, struct wg_endpoint *e, uint8_t *buf, size_t len)
901 {
902 	struct mbuf	*m;
903 	int		 ret = 0;
904 	bool		 retried = false;
905 
906 retry:
907 	m = m_get2(len, M_NOWAIT, MT_DATA, M_PKTHDR);
908 	if (!m) {
909 		ret = ENOMEM;
910 		goto out;
911 	}
912 	m_copyback(m, 0, len, buf);
913 
914 	if (ret == 0) {
915 		ret = wg_send(sc, e, m);
916 		/* Retry if we couldn't bind to e->e_local */
917 		if (ret == EADDRNOTAVAIL && !retried) {
918 			bzero(&e->e_local, sizeof(e->e_local));
919 			retried = true;
920 			goto retry;
921 		}
922 	} else {
923 		ret = wg_send(sc, e, m);
924 	}
925 out:
926 	if (ret)
927 		DPRINTF(sc, "Unable to send packet: %d\n", ret);
928 }
929 
930 /* Timers */
931 static void
932 wg_timers_enable(struct wg_peer *peer)
933 {
934 	atomic_store_bool(&peer->p_enabled, true);
935 	wg_timers_run_persistent_keepalive(peer);
936 }
937 
938 static void
939 wg_timers_disable(struct wg_peer *peer)
940 {
941 	/* By setting p_enabled = false, then calling NET_EPOCH_WAIT, we can be
942 	 * sure no new handshakes are created after the wait. This is because
943 	 * all callout_resets (scheduling the callout) are guarded by
944 	 * p_enabled. We can be sure all sections that read p_enabled and then
945 	 * optionally call callout_reset are finished as they are surrounded by
946 	 * NET_EPOCH_{ENTER,EXIT}.
947 	 *
948 	 * However, as new callouts may be scheduled during NET_EPOCH_WAIT (but
949 	 * not after), we stop all callouts leaving no callouts active.
950 	 *
951 	 * We should also pull NET_EPOCH_WAIT out of the FOREACH(peer) loops, but the
952 	 * performance impact is acceptable for the time being. */
953 	atomic_store_bool(&peer->p_enabled, false);
954 	NET_EPOCH_WAIT();
955 	atomic_store_bool(&peer->p_need_another_keepalive, false);
956 
957 	callout_stop(&peer->p_new_handshake);
958 	callout_stop(&peer->p_send_keepalive);
959 	callout_stop(&peer->p_retry_handshake);
960 	callout_stop(&peer->p_persistent_keepalive);
961 	callout_stop(&peer->p_zero_key_material);
962 }
963 
964 static void
965 wg_timers_set_persistent_keepalive(struct wg_peer *peer, uint16_t interval)
966 {
967 	struct epoch_tracker et;
968 	if (interval != peer->p_persistent_keepalive_interval) {
969 		atomic_store_16(&peer->p_persistent_keepalive_interval, interval);
970 		NET_EPOCH_ENTER(et);
971 		if (atomic_load_bool(&peer->p_enabled))
972 			wg_timers_run_persistent_keepalive(peer);
973 		NET_EPOCH_EXIT(et);
974 	}
975 }
976 
977 static void
978 wg_timers_get_last_handshake(struct wg_peer *peer, struct wg_timespec64 *time)
979 {
980 	mtx_lock(&peer->p_handshake_mtx);
981 	time->tv_sec = peer->p_handshake_complete.tv_sec;
982 	time->tv_nsec = peer->p_handshake_complete.tv_nsec;
983 	mtx_unlock(&peer->p_handshake_mtx);
984 }
985 
986 static void
987 wg_timers_event_data_sent(struct wg_peer *peer)
988 {
989 	struct epoch_tracker et;
990 	NET_EPOCH_ENTER(et);
991 	if (atomic_load_bool(&peer->p_enabled) &&
992 	    !callout_pending(&peer->p_new_handshake))
993 		callout_reset(&peer->p_new_handshake, MSEC_2_TICKS(
994 		    NEW_HANDSHAKE_TIMEOUT * 1000 +
995 		    arc4random_uniform(REKEY_TIMEOUT_JITTER)),
996 		    wg_timers_run_new_handshake, peer);
997 	NET_EPOCH_EXIT(et);
998 }
999 
1000 static void
1001 wg_timers_event_data_received(struct wg_peer *peer)
1002 {
1003 	struct epoch_tracker et;
1004 	NET_EPOCH_ENTER(et);
1005 	if (atomic_load_bool(&peer->p_enabled)) {
1006 		if (!callout_pending(&peer->p_send_keepalive))
1007 			callout_reset(&peer->p_send_keepalive,
1008 			    MSEC_2_TICKS(KEEPALIVE_TIMEOUT * 1000),
1009 			    wg_timers_run_send_keepalive, peer);
1010 		else
1011 			atomic_store_bool(&peer->p_need_another_keepalive,
1012 			    true);
1013 	}
1014 	NET_EPOCH_EXIT(et);
1015 }
1016 
1017 static void
1018 wg_timers_event_any_authenticated_packet_sent(struct wg_peer *peer)
1019 {
1020 	callout_stop(&peer->p_send_keepalive);
1021 }
1022 
1023 static void
1024 wg_timers_event_any_authenticated_packet_received(struct wg_peer *peer)
1025 {
1026 	callout_stop(&peer->p_new_handshake);
1027 }
1028 
1029 static void
1030 wg_timers_event_any_authenticated_packet_traversal(struct wg_peer *peer)
1031 {
1032 	struct epoch_tracker et;
1033 	uint16_t interval;
1034 	NET_EPOCH_ENTER(et);
1035 	interval = atomic_load_16(&peer->p_persistent_keepalive_interval);
1036 	if (atomic_load_bool(&peer->p_enabled) && interval > 0)
1037 		callout_reset(&peer->p_persistent_keepalive,
1038 		     MSEC_2_TICKS(interval * 1000),
1039 		     wg_timers_run_persistent_keepalive, peer);
1040 	NET_EPOCH_EXIT(et);
1041 }
1042 
1043 static void
1044 wg_timers_event_handshake_initiated(struct wg_peer *peer)
1045 {
1046 	struct epoch_tracker et;
1047 	NET_EPOCH_ENTER(et);
1048 	if (atomic_load_bool(&peer->p_enabled))
1049 		callout_reset(&peer->p_retry_handshake, MSEC_2_TICKS(
1050 		    REKEY_TIMEOUT * 1000 +
1051 		    arc4random_uniform(REKEY_TIMEOUT_JITTER)),
1052 		    wg_timers_run_retry_handshake, peer);
1053 	NET_EPOCH_EXIT(et);
1054 }
1055 
1056 static void
1057 wg_timers_event_handshake_complete(struct wg_peer *peer)
1058 {
1059 	struct epoch_tracker et;
1060 	NET_EPOCH_ENTER(et);
1061 	if (atomic_load_bool(&peer->p_enabled)) {
1062 		mtx_lock(&peer->p_handshake_mtx);
1063 		callout_stop(&peer->p_retry_handshake);
1064 		peer->p_handshake_retries = 0;
1065 		getnanotime(&peer->p_handshake_complete);
1066 		mtx_unlock(&peer->p_handshake_mtx);
1067 		wg_timers_run_send_keepalive(peer);
1068 	}
1069 	NET_EPOCH_EXIT(et);
1070 }
1071 
1072 static void
1073 wg_timers_event_session_derived(struct wg_peer *peer)
1074 {
1075 	struct epoch_tracker et;
1076 	NET_EPOCH_ENTER(et);
1077 	if (atomic_load_bool(&peer->p_enabled))
1078 		callout_reset(&peer->p_zero_key_material,
1079 		    MSEC_2_TICKS(REJECT_AFTER_TIME * 3 * 1000),
1080 		    wg_timers_run_zero_key_material, peer);
1081 	NET_EPOCH_EXIT(et);
1082 }
1083 
1084 static void
1085 wg_timers_event_want_initiation(struct wg_peer *peer)
1086 {
1087 	struct epoch_tracker et;
1088 	NET_EPOCH_ENTER(et);
1089 	if (atomic_load_bool(&peer->p_enabled))
1090 		wg_timers_run_send_initiation(peer, false);
1091 	NET_EPOCH_EXIT(et);
1092 }
1093 
1094 static void
1095 wg_timers_run_send_initiation(struct wg_peer *peer, bool is_retry)
1096 {
1097 	if (!is_retry)
1098 		peer->p_handshake_retries = 0;
1099 	if (noise_remote_initiation_expired(peer->p_remote) == ETIMEDOUT)
1100 		wg_send_initiation(peer);
1101 }
1102 
1103 static void
1104 wg_timers_run_retry_handshake(void *_peer)
1105 {
1106 	struct epoch_tracker et;
1107 	struct wg_peer *peer = _peer;
1108 
1109 	mtx_lock(&peer->p_handshake_mtx);
1110 	if (peer->p_handshake_retries <= MAX_TIMER_HANDSHAKES) {
1111 		peer->p_handshake_retries++;
1112 		mtx_unlock(&peer->p_handshake_mtx);
1113 
1114 		DPRINTF(peer->p_sc, "Handshake for peer %" PRIu64 " did not complete "
1115 		    "after %d seconds, retrying (try %d)\n", peer->p_id,
1116 		    REKEY_TIMEOUT, peer->p_handshake_retries + 1);
1117 		wg_peer_clear_src(peer);
1118 		wg_timers_run_send_initiation(peer, true);
1119 	} else {
1120 		mtx_unlock(&peer->p_handshake_mtx);
1121 
1122 		DPRINTF(peer->p_sc, "Handshake for peer %" PRIu64 " did not complete "
1123 		    "after %d retries, giving up\n", peer->p_id,
1124 		    MAX_TIMER_HANDSHAKES + 2);
1125 
1126 		callout_stop(&peer->p_send_keepalive);
1127 		wg_queue_purge(&peer->p_stage_queue);
1128 		NET_EPOCH_ENTER(et);
1129 		if (atomic_load_bool(&peer->p_enabled) &&
1130 		    !callout_pending(&peer->p_zero_key_material))
1131 			callout_reset(&peer->p_zero_key_material,
1132 			    MSEC_2_TICKS(REJECT_AFTER_TIME * 3 * 1000),
1133 			    wg_timers_run_zero_key_material, peer);
1134 		NET_EPOCH_EXIT(et);
1135 	}
1136 }
1137 
1138 static void
1139 wg_timers_run_send_keepalive(void *_peer)
1140 {
1141 	struct epoch_tracker et;
1142 	struct wg_peer *peer = _peer;
1143 
1144 	wg_send_keepalive(peer);
1145 	NET_EPOCH_ENTER(et);
1146 	if (atomic_load_bool(&peer->p_enabled) &&
1147 	    atomic_load_bool(&peer->p_need_another_keepalive)) {
1148 		atomic_store_bool(&peer->p_need_another_keepalive, false);
1149 		callout_reset(&peer->p_send_keepalive,
1150 		    MSEC_2_TICKS(KEEPALIVE_TIMEOUT * 1000),
1151 		    wg_timers_run_send_keepalive, peer);
1152 	}
1153 	NET_EPOCH_EXIT(et);
1154 }
1155 
1156 static void
1157 wg_timers_run_new_handshake(void *_peer)
1158 {
1159 	struct wg_peer *peer = _peer;
1160 
1161 	DPRINTF(peer->p_sc, "Retrying handshake with peer %" PRIu64 " because we "
1162 	    "stopped hearing back after %d seconds\n",
1163 	    peer->p_id, NEW_HANDSHAKE_TIMEOUT);
1164 
1165 	wg_peer_clear_src(peer);
1166 	wg_timers_run_send_initiation(peer, false);
1167 }
1168 
1169 static void
1170 wg_timers_run_zero_key_material(void *_peer)
1171 {
1172 	struct wg_peer *peer = _peer;
1173 
1174 	DPRINTF(peer->p_sc, "Zeroing out keys for peer %" PRIu64 ", since we "
1175 	    "haven't received a new one in %d seconds\n",
1176 	    peer->p_id, REJECT_AFTER_TIME * 3);
1177 	noise_remote_keypairs_clear(peer->p_remote);
1178 }
1179 
1180 static void
1181 wg_timers_run_persistent_keepalive(void *_peer)
1182 {
1183 	struct wg_peer *peer = _peer;
1184 
1185 	if (atomic_load_16(&peer->p_persistent_keepalive_interval) > 0)
1186 		wg_send_keepalive(peer);
1187 }
1188 
1189 /* TODO Handshake */
1190 static void
1191 wg_peer_send_buf(struct wg_peer *peer, uint8_t *buf, size_t len)
1192 {
1193 	struct wg_endpoint endpoint;
1194 
1195 	counter_u64_add(peer->p_tx_bytes, len);
1196 	wg_timers_event_any_authenticated_packet_traversal(peer);
1197 	wg_timers_event_any_authenticated_packet_sent(peer);
1198 	wg_peer_get_endpoint(peer, &endpoint);
1199 	wg_send_buf(peer->p_sc, &endpoint, buf, len);
1200 }
1201 
1202 static void
1203 wg_send_initiation(struct wg_peer *peer)
1204 {
1205 	struct wg_pkt_initiation pkt;
1206 
1207 	if (noise_create_initiation(peer->p_remote, &pkt.s_idx, pkt.ue,
1208 	    pkt.es, pkt.ets) != 0)
1209 		return;
1210 
1211 	DPRINTF(peer->p_sc, "Sending handshake initiation to peer %" PRIu64 "\n", peer->p_id);
1212 
1213 	pkt.t = WG_PKT_INITIATION;
1214 	cookie_maker_mac(&peer->p_cookie, &pkt.m, &pkt,
1215 	    sizeof(pkt) - sizeof(pkt.m));
1216 	wg_peer_send_buf(peer, (uint8_t *)&pkt, sizeof(pkt));
1217 	wg_timers_event_handshake_initiated(peer);
1218 }
1219 
1220 static void
1221 wg_send_response(struct wg_peer *peer)
1222 {
1223 	struct wg_pkt_response pkt;
1224 
1225 	if (noise_create_response(peer->p_remote, &pkt.s_idx, &pkt.r_idx,
1226 	    pkt.ue, pkt.en) != 0)
1227 		return;
1228 
1229 	DPRINTF(peer->p_sc, "Sending handshake response to peer %" PRIu64 "\n", peer->p_id);
1230 
1231 	wg_timers_event_session_derived(peer);
1232 	pkt.t = WG_PKT_RESPONSE;
1233 	cookie_maker_mac(&peer->p_cookie, &pkt.m, &pkt,
1234 	     sizeof(pkt)-sizeof(pkt.m));
1235 	wg_peer_send_buf(peer, (uint8_t*)&pkt, sizeof(pkt));
1236 }
1237 
1238 static void
1239 wg_send_cookie(struct wg_softc *sc, struct cookie_macs *cm, uint32_t idx,
1240     struct wg_endpoint *e)
1241 {
1242 	struct wg_pkt_cookie	pkt;
1243 
1244 	DPRINTF(sc, "Sending cookie response for denied handshake message\n");
1245 
1246 	pkt.t = WG_PKT_COOKIE;
1247 	pkt.r_idx = idx;
1248 
1249 	cookie_checker_create_payload(&sc->sc_cookie, cm, pkt.nonce,
1250 	    pkt.ec, &e->e_remote.r_sa);
1251 	wg_send_buf(sc, e, (uint8_t *)&pkt, sizeof(pkt));
1252 }
1253 
1254 static void
1255 wg_send_keepalive(struct wg_peer *peer)
1256 {
1257 	struct wg_packet *pkt;
1258 	struct mbuf *m;
1259 
1260 	if (wg_queue_len(&peer->p_stage_queue) > 0)
1261 		goto send;
1262 	if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL)
1263 		return;
1264 	if ((pkt = wg_packet_alloc(m)) == NULL) {
1265 		m_freem(m);
1266 		return;
1267 	}
1268 	wg_queue_push_staged(&peer->p_stage_queue, pkt);
1269 	DPRINTF(peer->p_sc, "Sending keepalive packet to peer %" PRIu64 "\n", peer->p_id);
1270 send:
1271 	wg_peer_send_staged(peer);
1272 }
1273 
1274 static void
1275 wg_handshake(struct wg_softc *sc, struct wg_packet *pkt)
1276 {
1277 	struct wg_pkt_initiation	*init;
1278 	struct wg_pkt_response		*resp;
1279 	struct wg_pkt_cookie		*cook;
1280 	struct wg_endpoint		*e;
1281 	struct wg_peer			*peer;
1282 	struct mbuf			*m;
1283 	struct noise_remote		*remote = NULL;
1284 	int				 res;
1285 	bool				 underload = false;
1286 	static sbintime_t		 wg_last_underload; /* sbinuptime */
1287 
1288 	underload = wg_queue_len(&sc->sc_handshake_queue) >= MAX_QUEUED_HANDSHAKES / 8;
1289 	if (underload) {
1290 		wg_last_underload = getsbinuptime();
1291 	} else if (wg_last_underload) {
1292 		underload = wg_last_underload + UNDERLOAD_TIMEOUT * SBT_1S > getsbinuptime();
1293 		if (!underload)
1294 			wg_last_underload = 0;
1295 	}
1296 
1297 	m = pkt->p_mbuf;
1298 	e = &pkt->p_endpoint;
1299 
1300 	if ((pkt->p_mbuf = m = m_pullup(m, m->m_pkthdr.len)) == NULL)
1301 		goto error;
1302 
1303 	switch (*mtod(m, uint32_t *)) {
1304 	case WG_PKT_INITIATION:
1305 		init = mtod(m, struct wg_pkt_initiation *);
1306 
1307 		res = cookie_checker_validate_macs(&sc->sc_cookie, &init->m,
1308 				init, sizeof(*init) - sizeof(init->m),
1309 				underload, &e->e_remote.r_sa,
1310 				if_getvnet(sc->sc_ifp));
1311 
1312 		if (res == EINVAL) {
1313 			DPRINTF(sc, "Invalid initiation MAC\n");
1314 			goto error;
1315 		} else if (res == ECONNREFUSED) {
1316 			DPRINTF(sc, "Handshake ratelimited\n");
1317 			goto error;
1318 		} else if (res == EAGAIN) {
1319 			wg_send_cookie(sc, &init->m, init->s_idx, e);
1320 			goto error;
1321 		} else if (res != 0) {
1322 			panic("unexpected response: %d\n", res);
1323 		}
1324 
1325 		if (noise_consume_initiation(sc->sc_local, &remote,
1326 		    init->s_idx, init->ue, init->es, init->ets) != 0) {
1327 			DPRINTF(sc, "Invalid handshake initiation\n");
1328 			goto error;
1329 		}
1330 
1331 		peer = noise_remote_arg(remote);
1332 
1333 		DPRINTF(sc, "Receiving handshake initiation from peer %" PRIu64 "\n", peer->p_id);
1334 
1335 		wg_peer_set_endpoint(peer, e);
1336 		wg_send_response(peer);
1337 		break;
1338 	case WG_PKT_RESPONSE:
1339 		resp = mtod(m, struct wg_pkt_response *);
1340 
1341 		res = cookie_checker_validate_macs(&sc->sc_cookie, &resp->m,
1342 				resp, sizeof(*resp) - sizeof(resp->m),
1343 				underload, &e->e_remote.r_sa,
1344 				if_getvnet(sc->sc_ifp));
1345 
1346 		if (res == EINVAL) {
1347 			DPRINTF(sc, "Invalid response MAC\n");
1348 			goto error;
1349 		} else if (res == ECONNREFUSED) {
1350 			DPRINTF(sc, "Handshake ratelimited\n");
1351 			goto error;
1352 		} else if (res == EAGAIN) {
1353 			wg_send_cookie(sc, &resp->m, resp->s_idx, e);
1354 			goto error;
1355 		} else if (res != 0) {
1356 			panic("unexpected response: %d\n", res);
1357 		}
1358 
1359 		if (noise_consume_response(sc->sc_local, &remote,
1360 		    resp->s_idx, resp->r_idx, resp->ue, resp->en) != 0) {
1361 			DPRINTF(sc, "Invalid handshake response\n");
1362 			goto error;
1363 		}
1364 
1365 		peer = noise_remote_arg(remote);
1366 		DPRINTF(sc, "Receiving handshake response from peer %" PRIu64 "\n", peer->p_id);
1367 
1368 		wg_peer_set_endpoint(peer, e);
1369 		wg_timers_event_session_derived(peer);
1370 		wg_timers_event_handshake_complete(peer);
1371 		break;
1372 	case WG_PKT_COOKIE:
1373 		cook = mtod(m, struct wg_pkt_cookie *);
1374 
1375 		if ((remote = noise_remote_index(sc->sc_local, cook->r_idx)) == NULL) {
1376 			DPRINTF(sc, "Unknown cookie index\n");
1377 			goto error;
1378 		}
1379 
1380 		peer = noise_remote_arg(remote);
1381 
1382 		if (cookie_maker_consume_payload(&peer->p_cookie,
1383 		    cook->nonce, cook->ec) == 0) {
1384 			DPRINTF(sc, "Receiving cookie response\n");
1385 		} else {
1386 			DPRINTF(sc, "Could not decrypt cookie response\n");
1387 			goto error;
1388 		}
1389 
1390 		goto not_authenticated;
1391 	default:
1392 		panic("invalid packet in handshake queue");
1393 	}
1394 
1395 	wg_timers_event_any_authenticated_packet_received(peer);
1396 	wg_timers_event_any_authenticated_packet_traversal(peer);
1397 
1398 not_authenticated:
1399 	counter_u64_add(peer->p_rx_bytes, m->m_pkthdr.len);
1400 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
1401 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1402 error:
1403 	if (remote != NULL)
1404 		noise_remote_put(remote);
1405 	wg_packet_free(pkt);
1406 }
1407 
1408 static void
1409 wg_softc_handshake_receive(struct wg_softc *sc)
1410 {
1411 	struct wg_packet *pkt;
1412 	while ((pkt = wg_queue_dequeue_handshake(&sc->sc_handshake_queue)) != NULL)
1413 		wg_handshake(sc, pkt);
1414 }
1415 
1416 static void
1417 wg_mbuf_reset(struct mbuf *m)
1418 {
1419 
1420 	struct m_tag *t, *tmp;
1421 
1422 	/*
1423 	 * We want to reset the mbuf to a newly allocated state, containing
1424 	 * just the packet contents. Unfortunately FreeBSD doesn't seem to
1425 	 * offer this anywhere, so we have to make it up as we go. If we can
1426 	 * get this in kern/kern_mbuf.c, that would be best.
1427 	 *
1428 	 * Notice: this may break things unexpectedly but it is better to fail
1429 	 *         closed in the extreme case than leak informtion in every
1430 	 *         case.
1431 	 *
1432 	 * With that said, all this attempts to do is remove any extraneous
1433 	 * information that could be present.
1434 	 */
1435 
1436 	M_ASSERTPKTHDR(m);
1437 
1438 	m->m_flags &= ~(M_BCAST|M_MCAST|M_VLANTAG|M_PROMISC|M_PROTOFLAGS);
1439 
1440 	M_HASHTYPE_CLEAR(m);
1441 #ifdef NUMA
1442         m->m_pkthdr.numa_domain = M_NODOM;
1443 #endif
1444 	SLIST_FOREACH_SAFE(t, &m->m_pkthdr.tags, m_tag_link, tmp) {
1445 		if ((t->m_tag_id != 0 || t->m_tag_cookie != MTAG_WGLOOP) &&
1446 		    t->m_tag_id != PACKET_TAG_MACLABEL)
1447 			m_tag_delete(m, t);
1448 	}
1449 
1450 	KASSERT((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0,
1451 	    ("%s: mbuf %p has a send tag", __func__, m));
1452 
1453 	m->m_pkthdr.csum_flags = 0;
1454 	m->m_pkthdr.PH_per.sixtyfour[0] = 0;
1455 	m->m_pkthdr.PH_loc.sixtyfour[0] = 0;
1456 }
1457 
1458 static inline unsigned int
1459 calculate_padding(struct wg_packet *pkt)
1460 {
1461 	unsigned int padded_size, last_unit = pkt->p_mbuf->m_pkthdr.len;
1462 
1463 	/* Keepalive packets don't set p_mtu, but also have a length of zero. */
1464 	if (__predict_false(pkt->p_mtu == 0)) {
1465 		padded_size = (last_unit + (WG_PKT_PADDING - 1)) &
1466 		    ~(WG_PKT_PADDING - 1);
1467 		return (padded_size - last_unit);
1468 	}
1469 
1470 	if (__predict_false(last_unit > pkt->p_mtu))
1471 		last_unit %= pkt->p_mtu;
1472 
1473 	padded_size = (last_unit + (WG_PKT_PADDING - 1)) & ~(WG_PKT_PADDING - 1);
1474 	if (pkt->p_mtu < padded_size)
1475 		padded_size = pkt->p_mtu;
1476 	return (padded_size - last_unit);
1477 }
1478 
1479 static void
1480 wg_encrypt(struct wg_softc *sc, struct wg_packet *pkt)
1481 {
1482 	static const uint8_t	 padding[WG_PKT_PADDING] = { 0 };
1483 	struct wg_pkt_data	*data;
1484 	struct wg_peer		*peer;
1485 	struct noise_remote	*remote;
1486 	struct mbuf		*m;
1487 	uint32_t		 idx;
1488 	unsigned int		 padlen;
1489 	enum wg_ring_state	 state = WG_PACKET_DEAD;
1490 
1491 	remote = noise_keypair_remote(pkt->p_keypair);
1492 	peer = noise_remote_arg(remote);
1493 	m = pkt->p_mbuf;
1494 
1495 	/* Pad the packet */
1496 	padlen = calculate_padding(pkt);
1497 	if (padlen != 0 && !m_append(m, padlen, padding))
1498 		goto out;
1499 
1500 	/* Do encryption */
1501 	if (noise_keypair_encrypt(pkt->p_keypair, &idx, pkt->p_nonce, m) != 0)
1502 		goto out;
1503 
1504 	/* Put header into packet */
1505 	M_PREPEND(m, sizeof(struct wg_pkt_data), M_NOWAIT);
1506 	if (m == NULL)
1507 		goto out;
1508 	data = mtod(m, struct wg_pkt_data *);
1509 	data->t = WG_PKT_DATA;
1510 	data->r_idx = idx;
1511 	data->nonce = htole64(pkt->p_nonce);
1512 
1513 	wg_mbuf_reset(m);
1514 	state = WG_PACKET_CRYPTED;
1515 out:
1516 	pkt->p_mbuf = m;
1517 	atomic_store_rel_int(&pkt->p_state, state);
1518 	GROUPTASK_ENQUEUE(&peer->p_send);
1519 	noise_remote_put(remote);
1520 }
1521 
1522 static void
1523 wg_decrypt(struct wg_softc *sc, struct wg_packet *pkt)
1524 {
1525 	struct wg_peer		*peer, *allowed_peer;
1526 	struct noise_remote	*remote;
1527 	struct mbuf		*m;
1528 	int			 len;
1529 	enum wg_ring_state	 state = WG_PACKET_DEAD;
1530 
1531 	remote = noise_keypair_remote(pkt->p_keypair);
1532 	peer = noise_remote_arg(remote);
1533 	m = pkt->p_mbuf;
1534 
1535 	/* Read nonce and then adjust to remove the header. */
1536 	pkt->p_nonce = le64toh(mtod(m, struct wg_pkt_data *)->nonce);
1537 	m_adj(m, sizeof(struct wg_pkt_data));
1538 
1539 	if (noise_keypair_decrypt(pkt->p_keypair, pkt->p_nonce, m) != 0)
1540 		goto out;
1541 
1542 	/* A packet with length 0 is a keepalive packet */
1543 	if (__predict_false(m->m_pkthdr.len == 0)) {
1544 		DPRINTF(sc, "Receiving keepalive packet from peer "
1545 		    "%" PRIu64 "\n", peer->p_id);
1546 		state = WG_PACKET_CRYPTED;
1547 		goto out;
1548 	}
1549 
1550 	/*
1551 	 * We can let the network stack handle the intricate validation of the
1552 	 * IP header, we just worry about the sizeof and the version, so we can
1553 	 * read the source address in wg_aip_lookup.
1554 	 */
1555 
1556 	if (determine_af_and_pullup(&m, &pkt->p_af) == 0) {
1557 		if (pkt->p_af == AF_INET) {
1558 			struct ip *ip = mtod(m, struct ip *);
1559 			allowed_peer = wg_aip_lookup(sc, AF_INET, &ip->ip_src);
1560 			len = ntohs(ip->ip_len);
1561 			if (len >= sizeof(struct ip) && len < m->m_pkthdr.len)
1562 				m_adj(m, len - m->m_pkthdr.len);
1563 		} else if (pkt->p_af == AF_INET6) {
1564 			struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
1565 			allowed_peer = wg_aip_lookup(sc, AF_INET6, &ip6->ip6_src);
1566 			len = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr);
1567 			if (len < m->m_pkthdr.len)
1568 				m_adj(m, len - m->m_pkthdr.len);
1569 		} else
1570 			panic("determine_af_and_pullup returned unexpected value");
1571 	} else {
1572 		DPRINTF(sc, "Packet is neither ipv4 nor ipv6 from peer %" PRIu64 "\n", peer->p_id);
1573 		goto out;
1574 	}
1575 
1576 	/* We only want to compare the address, not dereference, so drop the ref. */
1577 	if (allowed_peer != NULL)
1578 		noise_remote_put(allowed_peer->p_remote);
1579 
1580 	if (__predict_false(peer != allowed_peer)) {
1581 		DPRINTF(sc, "Packet has unallowed src IP from peer %" PRIu64 "\n", peer->p_id);
1582 		goto out;
1583 	}
1584 
1585 	wg_mbuf_reset(m);
1586 	state = WG_PACKET_CRYPTED;
1587 out:
1588 	pkt->p_mbuf = m;
1589 	atomic_store_rel_int(&pkt->p_state, state);
1590 	GROUPTASK_ENQUEUE(&peer->p_recv);
1591 	noise_remote_put(remote);
1592 }
1593 
1594 static void
1595 wg_softc_decrypt(struct wg_softc *sc)
1596 {
1597 	struct wg_packet *pkt;
1598 
1599 	while ((pkt = wg_queue_dequeue_parallel(&sc->sc_decrypt_parallel)) != NULL)
1600 		wg_decrypt(sc, pkt);
1601 }
1602 
1603 static void
1604 wg_softc_encrypt(struct wg_softc *sc)
1605 {
1606 	struct wg_packet *pkt;
1607 
1608 	while ((pkt = wg_queue_dequeue_parallel(&sc->sc_encrypt_parallel)) != NULL)
1609 		wg_encrypt(sc, pkt);
1610 }
1611 
1612 static void
1613 wg_encrypt_dispatch(struct wg_softc *sc)
1614 {
1615 	/*
1616 	 * The update to encrypt_last_cpu is racey such that we may
1617 	 * reschedule the task for the same CPU multiple times, but
1618 	 * the race doesn't really matter.
1619 	 */
1620 	u_int cpu = (sc->sc_encrypt_last_cpu + 1) % mp_ncpus;
1621 	sc->sc_encrypt_last_cpu = cpu;
1622 	GROUPTASK_ENQUEUE(&sc->sc_encrypt[cpu]);
1623 }
1624 
1625 static void
1626 wg_decrypt_dispatch(struct wg_softc *sc)
1627 {
1628 	u_int cpu = (sc->sc_decrypt_last_cpu + 1) % mp_ncpus;
1629 	sc->sc_decrypt_last_cpu = cpu;
1630 	GROUPTASK_ENQUEUE(&sc->sc_decrypt[cpu]);
1631 }
1632 
1633 static void
1634 wg_deliver_out(struct wg_peer *peer)
1635 {
1636 	struct wg_endpoint	 endpoint;
1637 	struct wg_softc		*sc = peer->p_sc;
1638 	struct wg_packet	*pkt;
1639 	struct mbuf		*m;
1640 	int			 rc, len;
1641 
1642 	wg_peer_get_endpoint(peer, &endpoint);
1643 
1644 	while ((pkt = wg_queue_dequeue_serial(&peer->p_encrypt_serial)) != NULL) {
1645 		if (atomic_load_acq_int(&pkt->p_state) != WG_PACKET_CRYPTED)
1646 			goto error;
1647 
1648 		m = pkt->p_mbuf;
1649 		pkt->p_mbuf = NULL;
1650 
1651 		len = m->m_pkthdr.len;
1652 
1653 		wg_timers_event_any_authenticated_packet_traversal(peer);
1654 		wg_timers_event_any_authenticated_packet_sent(peer);
1655 		rc = wg_send(sc, &endpoint, m);
1656 		if (rc == 0) {
1657 			if (len > (sizeof(struct wg_pkt_data) + NOISE_AUTHTAG_LEN))
1658 				wg_timers_event_data_sent(peer);
1659 			counter_u64_add(peer->p_tx_bytes, len);
1660 		} else if (rc == EADDRNOTAVAIL) {
1661 			wg_peer_clear_src(peer);
1662 			wg_peer_get_endpoint(peer, &endpoint);
1663 			goto error;
1664 		} else {
1665 			goto error;
1666 		}
1667 		wg_packet_free(pkt);
1668 		if (noise_keep_key_fresh_send(peer->p_remote))
1669 			wg_timers_event_want_initiation(peer);
1670 		continue;
1671 error:
1672 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1673 		wg_packet_free(pkt);
1674 	}
1675 }
1676 
1677 #ifdef DEV_NETMAP
1678 /*
1679  * Hand a packet to the netmap RX ring, via netmap's
1680  * freebsd_generic_rx_handler().
1681  */
1682 static void
1683 wg_deliver_netmap(if_t ifp, struct mbuf *m, int af)
1684 {
1685 	struct ether_header *eh;
1686 
1687 	M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT);
1688 	if (__predict_false(m == NULL)) {
1689 		if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1690 		return;
1691 	}
1692 
1693 	eh = mtod(m, struct ether_header *);
1694 	eh->ether_type = af == AF_INET ?
1695 	    htons(ETHERTYPE_IP) : htons(ETHERTYPE_IPV6);
1696 	memcpy(eh->ether_shost, "\x02\x02\x02\x02\x02\x02", ETHER_ADDR_LEN);
1697 	memcpy(eh->ether_dhost, "\xff\xff\xff\xff\xff\xff", ETHER_ADDR_LEN);
1698 	if_input(ifp, m);
1699 }
1700 #endif
1701 
1702 static void
1703 wg_deliver_in(struct wg_peer *peer)
1704 {
1705 	struct wg_softc		*sc = peer->p_sc;
1706 	if_t			 ifp = sc->sc_ifp;
1707 	struct wg_packet	*pkt;
1708 	struct mbuf		*m;
1709 	struct epoch_tracker	 et;
1710 	int			 af;
1711 
1712 	while ((pkt = wg_queue_dequeue_serial(&peer->p_decrypt_serial)) != NULL) {
1713 		if (atomic_load_acq_int(&pkt->p_state) != WG_PACKET_CRYPTED)
1714 			goto error;
1715 
1716 		m = pkt->p_mbuf;
1717 		if (noise_keypair_nonce_check(pkt->p_keypair, pkt->p_nonce) != 0)
1718 			goto error;
1719 
1720 		if (noise_keypair_received_with(pkt->p_keypair) == ECONNRESET)
1721 			wg_timers_event_handshake_complete(peer);
1722 
1723 		wg_timers_event_any_authenticated_packet_received(peer);
1724 		wg_timers_event_any_authenticated_packet_traversal(peer);
1725 		wg_peer_set_endpoint(peer, &pkt->p_endpoint);
1726 
1727 		counter_u64_add(peer->p_rx_bytes, m->m_pkthdr.len +
1728 		    sizeof(struct wg_pkt_data) + NOISE_AUTHTAG_LEN);
1729 		if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
1730 		if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len +
1731 		    sizeof(struct wg_pkt_data) + NOISE_AUTHTAG_LEN);
1732 
1733 		if (m->m_pkthdr.len == 0)
1734 			goto done;
1735 
1736 		af = pkt->p_af;
1737 		MPASS(af == AF_INET || af == AF_INET6);
1738 		pkt->p_mbuf = NULL;
1739 
1740 		m->m_pkthdr.rcvif = ifp;
1741 
1742 		NET_EPOCH_ENTER(et);
1743 		BPF_MTAP2_AF(ifp, m, af);
1744 
1745 		CURVNET_SET(if_getvnet(ifp));
1746 		M_SETFIB(m, if_getfib(ifp));
1747 #ifdef DEV_NETMAP
1748 		if ((if_getcapenable(ifp) & IFCAP_NETMAP) != 0)
1749 			wg_deliver_netmap(ifp, m, af);
1750 		else
1751 #endif
1752 		if (af == AF_INET)
1753 			netisr_dispatch(NETISR_IP, m);
1754 		else if (af == AF_INET6)
1755 			netisr_dispatch(NETISR_IPV6, m);
1756 		CURVNET_RESTORE();
1757 		NET_EPOCH_EXIT(et);
1758 
1759 		wg_timers_event_data_received(peer);
1760 
1761 done:
1762 		if (noise_keep_key_fresh_recv(peer->p_remote))
1763 			wg_timers_event_want_initiation(peer);
1764 		wg_packet_free(pkt);
1765 		continue;
1766 error:
1767 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1768 		wg_packet_free(pkt);
1769 	}
1770 }
1771 
1772 static struct wg_packet *
1773 wg_packet_alloc(struct mbuf *m)
1774 {
1775 	struct wg_packet *pkt;
1776 
1777 	if ((pkt = uma_zalloc(wg_packet_zone, M_NOWAIT | M_ZERO)) == NULL)
1778 		return (NULL);
1779 	pkt->p_mbuf = m;
1780 	return (pkt);
1781 }
1782 
1783 static void
1784 wg_packet_free(struct wg_packet *pkt)
1785 {
1786 	if (pkt->p_keypair != NULL)
1787 		noise_keypair_put(pkt->p_keypair);
1788 	if (pkt->p_mbuf != NULL)
1789 		m_freem(pkt->p_mbuf);
1790 	uma_zfree(wg_packet_zone, pkt);
1791 }
1792 
1793 static void
1794 wg_queue_init(struct wg_queue *queue, const char *name)
1795 {
1796 	mtx_init(&queue->q_mtx, name, NULL, MTX_DEF);
1797 	STAILQ_INIT(&queue->q_queue);
1798 	queue->q_len = 0;
1799 }
1800 
1801 static void
1802 wg_queue_deinit(struct wg_queue *queue)
1803 {
1804 	wg_queue_purge(queue);
1805 	mtx_destroy(&queue->q_mtx);
1806 }
1807 
1808 static size_t
1809 wg_queue_len(struct wg_queue *queue)
1810 {
1811 	return (queue->q_len);
1812 }
1813 
1814 static int
1815 wg_queue_enqueue_handshake(struct wg_queue *hs, struct wg_packet *pkt)
1816 {
1817 	int ret = 0;
1818 	mtx_lock(&hs->q_mtx);
1819 	if (hs->q_len < MAX_QUEUED_HANDSHAKES) {
1820 		STAILQ_INSERT_TAIL(&hs->q_queue, pkt, p_parallel);
1821 		hs->q_len++;
1822 	} else {
1823 		ret = ENOBUFS;
1824 	}
1825 	mtx_unlock(&hs->q_mtx);
1826 	if (ret != 0)
1827 		wg_packet_free(pkt);
1828 	return (ret);
1829 }
1830 
1831 static struct wg_packet *
1832 wg_queue_dequeue_handshake(struct wg_queue *hs)
1833 {
1834 	struct wg_packet *pkt;
1835 	mtx_lock(&hs->q_mtx);
1836 	if ((pkt = STAILQ_FIRST(&hs->q_queue)) != NULL) {
1837 		STAILQ_REMOVE_HEAD(&hs->q_queue, p_parallel);
1838 		hs->q_len--;
1839 	}
1840 	mtx_unlock(&hs->q_mtx);
1841 	return (pkt);
1842 }
1843 
1844 static void
1845 wg_queue_push_staged(struct wg_queue *staged, struct wg_packet *pkt)
1846 {
1847 	struct wg_packet *old = NULL;
1848 
1849 	mtx_lock(&staged->q_mtx);
1850 	if (staged->q_len >= MAX_STAGED_PKT) {
1851 		old = STAILQ_FIRST(&staged->q_queue);
1852 		STAILQ_REMOVE_HEAD(&staged->q_queue, p_parallel);
1853 		staged->q_len--;
1854 	}
1855 	STAILQ_INSERT_TAIL(&staged->q_queue, pkt, p_parallel);
1856 	staged->q_len++;
1857 	mtx_unlock(&staged->q_mtx);
1858 
1859 	if (old != NULL)
1860 		wg_packet_free(old);
1861 }
1862 
1863 static void
1864 wg_queue_enlist_staged(struct wg_queue *staged, struct wg_packet_list *list)
1865 {
1866 	struct wg_packet *pkt, *tpkt;
1867 	STAILQ_FOREACH_SAFE(pkt, list, p_parallel, tpkt)
1868 		wg_queue_push_staged(staged, pkt);
1869 }
1870 
1871 static void
1872 wg_queue_delist_staged(struct wg_queue *staged, struct wg_packet_list *list)
1873 {
1874 	STAILQ_INIT(list);
1875 	mtx_lock(&staged->q_mtx);
1876 	STAILQ_CONCAT(list, &staged->q_queue);
1877 	staged->q_len = 0;
1878 	mtx_unlock(&staged->q_mtx);
1879 }
1880 
1881 static void
1882 wg_queue_purge(struct wg_queue *staged)
1883 {
1884 	struct wg_packet_list list;
1885 	struct wg_packet *pkt, *tpkt;
1886 	wg_queue_delist_staged(staged, &list);
1887 	STAILQ_FOREACH_SAFE(pkt, &list, p_parallel, tpkt)
1888 		wg_packet_free(pkt);
1889 }
1890 
1891 static int
1892 wg_queue_both(struct wg_queue *parallel, struct wg_queue *serial, struct wg_packet *pkt)
1893 {
1894 	pkt->p_state = WG_PACKET_UNCRYPTED;
1895 
1896 	mtx_lock(&serial->q_mtx);
1897 	if (serial->q_len < MAX_QUEUED_PKT) {
1898 		serial->q_len++;
1899 		STAILQ_INSERT_TAIL(&serial->q_queue, pkt, p_serial);
1900 	} else {
1901 		mtx_unlock(&serial->q_mtx);
1902 		wg_packet_free(pkt);
1903 		return (ENOBUFS);
1904 	}
1905 	mtx_unlock(&serial->q_mtx);
1906 
1907 	mtx_lock(&parallel->q_mtx);
1908 	if (parallel->q_len < MAX_QUEUED_PKT) {
1909 		parallel->q_len++;
1910 		STAILQ_INSERT_TAIL(&parallel->q_queue, pkt, p_parallel);
1911 	} else {
1912 		mtx_unlock(&parallel->q_mtx);
1913 		pkt->p_state = WG_PACKET_DEAD;
1914 		return (ENOBUFS);
1915 	}
1916 	mtx_unlock(&parallel->q_mtx);
1917 
1918 	return (0);
1919 }
1920 
1921 static struct wg_packet *
1922 wg_queue_dequeue_serial(struct wg_queue *serial)
1923 {
1924 	struct wg_packet *pkt = NULL;
1925 	mtx_lock(&serial->q_mtx);
1926 	if (serial->q_len > 0 && STAILQ_FIRST(&serial->q_queue)->p_state != WG_PACKET_UNCRYPTED) {
1927 		serial->q_len--;
1928 		pkt = STAILQ_FIRST(&serial->q_queue);
1929 		STAILQ_REMOVE_HEAD(&serial->q_queue, p_serial);
1930 	}
1931 	mtx_unlock(&serial->q_mtx);
1932 	return (pkt);
1933 }
1934 
1935 static struct wg_packet *
1936 wg_queue_dequeue_parallel(struct wg_queue *parallel)
1937 {
1938 	struct wg_packet *pkt = NULL;
1939 	mtx_lock(&parallel->q_mtx);
1940 	if (parallel->q_len > 0) {
1941 		parallel->q_len--;
1942 		pkt = STAILQ_FIRST(&parallel->q_queue);
1943 		STAILQ_REMOVE_HEAD(&parallel->q_queue, p_parallel);
1944 	}
1945 	mtx_unlock(&parallel->q_mtx);
1946 	return (pkt);
1947 }
1948 
1949 static bool
1950 wg_input(struct mbuf *m, int offset, struct inpcb *inpcb,
1951     const struct sockaddr *sa, void *_sc)
1952 {
1953 #ifdef INET
1954 	const struct sockaddr_in	*sin;
1955 #endif
1956 #ifdef INET6
1957 	const struct sockaddr_in6	*sin6;
1958 #endif
1959 	struct noise_remote		*remote;
1960 	struct wg_pkt_data		*data;
1961 	struct wg_packet		*pkt;
1962 	struct wg_peer			*peer;
1963 	struct wg_softc			*sc = _sc;
1964 	struct mbuf			*defragged;
1965 
1966 	defragged = m_defrag(m, M_NOWAIT);
1967 	if (defragged)
1968 		m = defragged;
1969 	m = m_unshare(m, M_NOWAIT);
1970 	if (!m) {
1971 		if_inc_counter(sc->sc_ifp, IFCOUNTER_IQDROPS, 1);
1972 		return true;
1973 	}
1974 
1975 	/* Caller provided us with `sa`, no need for this header. */
1976 	m_adj(m, offset + sizeof(struct udphdr));
1977 
1978 	/* Pullup enough to read packet type */
1979 	if ((m = m_pullup(m, sizeof(uint32_t))) == NULL) {
1980 		if_inc_counter(sc->sc_ifp, IFCOUNTER_IQDROPS, 1);
1981 		return true;
1982 	}
1983 
1984 	if ((pkt = wg_packet_alloc(m)) == NULL) {
1985 		if_inc_counter(sc->sc_ifp, IFCOUNTER_IQDROPS, 1);
1986 		m_freem(m);
1987 		return true;
1988 	}
1989 
1990 	/* Save send/recv address and port for later. */
1991 	switch (sa->sa_family) {
1992 #ifdef INET
1993 	case AF_INET:
1994 		sin = (const struct sockaddr_in *)sa;
1995 		pkt->p_endpoint.e_remote.r_sin = sin[0];
1996 		pkt->p_endpoint.e_local.l_in = sin[1].sin_addr;
1997 		break;
1998 #endif
1999 #ifdef INET6
2000 	case AF_INET6:
2001 		sin6 = (const struct sockaddr_in6 *)sa;
2002 		pkt->p_endpoint.e_remote.r_sin6 = sin6[0];
2003 		pkt->p_endpoint.e_local.l_in6 = sin6[1].sin6_addr;
2004 		break;
2005 #endif
2006 	default:
2007 		goto error;
2008 	}
2009 
2010 	if ((m->m_pkthdr.len == sizeof(struct wg_pkt_initiation) &&
2011 		*mtod(m, uint32_t *) == WG_PKT_INITIATION) ||
2012 	    (m->m_pkthdr.len == sizeof(struct wg_pkt_response) &&
2013 		*mtod(m, uint32_t *) == WG_PKT_RESPONSE) ||
2014 	    (m->m_pkthdr.len == sizeof(struct wg_pkt_cookie) &&
2015 		*mtod(m, uint32_t *) == WG_PKT_COOKIE)) {
2016 
2017 		if (wg_queue_enqueue_handshake(&sc->sc_handshake_queue, pkt) != 0) {
2018 			if_inc_counter(sc->sc_ifp, IFCOUNTER_IQDROPS, 1);
2019 			DPRINTF(sc, "Dropping handshake packet\n");
2020 		}
2021 		GROUPTASK_ENQUEUE(&sc->sc_handshake);
2022 	} else if (m->m_pkthdr.len >= sizeof(struct wg_pkt_data) +
2023 	    NOISE_AUTHTAG_LEN && *mtod(m, uint32_t *) == WG_PKT_DATA) {
2024 
2025 		/* Pullup whole header to read r_idx below. */
2026 		if ((pkt->p_mbuf = m_pullup(m, sizeof(struct wg_pkt_data))) == NULL)
2027 			goto error;
2028 
2029 		data = mtod(pkt->p_mbuf, struct wg_pkt_data *);
2030 		if ((pkt->p_keypair = noise_keypair_lookup(sc->sc_local, data->r_idx)) == NULL)
2031 			goto error;
2032 
2033 		remote = noise_keypair_remote(pkt->p_keypair);
2034 		peer = noise_remote_arg(remote);
2035 		if (wg_queue_both(&sc->sc_decrypt_parallel, &peer->p_decrypt_serial, pkt) != 0)
2036 			if_inc_counter(sc->sc_ifp, IFCOUNTER_IQDROPS, 1);
2037 		wg_decrypt_dispatch(sc);
2038 		noise_remote_put(remote);
2039 	} else {
2040 		goto error;
2041 	}
2042 	return true;
2043 error:
2044 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IERRORS, 1);
2045 	wg_packet_free(pkt);
2046 	return true;
2047 }
2048 
2049 static void
2050 wg_peer_send_staged(struct wg_peer *peer)
2051 {
2052 	struct wg_packet_list	 list;
2053 	struct noise_keypair	*keypair;
2054 	struct wg_packet	*pkt, *tpkt;
2055 	struct wg_softc		*sc = peer->p_sc;
2056 
2057 	wg_queue_delist_staged(&peer->p_stage_queue, &list);
2058 
2059 	if (STAILQ_EMPTY(&list))
2060 		return;
2061 
2062 	if ((keypair = noise_keypair_current(peer->p_remote)) == NULL)
2063 		goto error;
2064 
2065 	STAILQ_FOREACH(pkt, &list, p_parallel) {
2066 		if (noise_keypair_nonce_next(keypair, &pkt->p_nonce) != 0)
2067 			goto error_keypair;
2068 	}
2069 	STAILQ_FOREACH_SAFE(pkt, &list, p_parallel, tpkt) {
2070 		pkt->p_keypair = noise_keypair_ref(keypair);
2071 		if (wg_queue_both(&sc->sc_encrypt_parallel, &peer->p_encrypt_serial, pkt) != 0)
2072 			if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
2073 	}
2074 	wg_encrypt_dispatch(sc);
2075 	noise_keypair_put(keypair);
2076 	return;
2077 
2078 error_keypair:
2079 	noise_keypair_put(keypair);
2080 error:
2081 	wg_queue_enlist_staged(&peer->p_stage_queue, &list);
2082 	wg_timers_event_want_initiation(peer);
2083 }
2084 
2085 static inline void
2086 xmit_err(if_t ifp, struct mbuf *m, struct wg_packet *pkt, sa_family_t af)
2087 {
2088 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
2089 	switch (af) {
2090 #ifdef INET
2091 	case AF_INET:
2092 		icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0);
2093 		if (pkt)
2094 			pkt->p_mbuf = NULL;
2095 		m = NULL;
2096 		break;
2097 #endif
2098 #ifdef INET6
2099 	case AF_INET6:
2100 		icmp6_error(m, ICMP6_DST_UNREACH, 0, 0);
2101 		if (pkt)
2102 			pkt->p_mbuf = NULL;
2103 		m = NULL;
2104 		break;
2105 #endif
2106 	}
2107 	if (pkt)
2108 		wg_packet_free(pkt);
2109 	else if (m)
2110 		m_freem(m);
2111 }
2112 
2113 static int
2114 wg_xmit(if_t ifp, struct mbuf *m, sa_family_t af, uint32_t mtu)
2115 {
2116 	struct wg_packet	*pkt = NULL;
2117 	struct wg_softc		*sc = if_getsoftc(ifp);
2118 	struct wg_peer		*peer;
2119 	int			 rc = 0;
2120 	sa_family_t		 peer_af;
2121 
2122 	/* Work around lifetime issue in the ipv6 mld code. */
2123 	if (__predict_false((if_getflags(ifp) & IFF_DYING) || !sc)) {
2124 		rc = ENXIO;
2125 		goto err_xmit;
2126 	}
2127 
2128 	if ((pkt = wg_packet_alloc(m)) == NULL) {
2129 		rc = ENOBUFS;
2130 		goto err_xmit;
2131 	}
2132 	pkt->p_mtu = mtu;
2133 	pkt->p_af = af;
2134 
2135 	if (af == AF_INET) {
2136 		peer = wg_aip_lookup(sc, AF_INET, &mtod(m, struct ip *)->ip_dst);
2137 	} else if (af == AF_INET6) {
2138 		peer = wg_aip_lookup(sc, AF_INET6, &mtod(m, struct ip6_hdr *)->ip6_dst);
2139 	} else {
2140 		rc = EAFNOSUPPORT;
2141 		goto err_xmit;
2142 	}
2143 
2144 	BPF_MTAP2_AF(ifp, m, pkt->p_af);
2145 
2146 	if (__predict_false(peer == NULL)) {
2147 		rc = ENETUNREACH;
2148 		goto err_xmit;
2149 	}
2150 
2151 	if (__predict_false(if_tunnel_check_nesting(ifp, m, MTAG_WGLOOP, MAX_LOOPS))) {
2152 		DPRINTF(sc, "Packet looped");
2153 		rc = ELOOP;
2154 		goto err_peer;
2155 	}
2156 
2157 	peer_af = peer->p_endpoint.e_remote.r_sa.sa_family;
2158 	if (__predict_false(peer_af != AF_INET && peer_af != AF_INET6)) {
2159 		DPRINTF(sc, "No valid endpoint has been configured or "
2160 			    "discovered for peer %" PRIu64 "\n", peer->p_id);
2161 		rc = EHOSTUNREACH;
2162 		goto err_peer;
2163 	}
2164 
2165 	wg_queue_push_staged(&peer->p_stage_queue, pkt);
2166 	wg_peer_send_staged(peer);
2167 	noise_remote_put(peer->p_remote);
2168 	return (0);
2169 
2170 err_peer:
2171 	noise_remote_put(peer->p_remote);
2172 err_xmit:
2173 	xmit_err(ifp, m, pkt, af);
2174 	return (rc);
2175 }
2176 
2177 static inline int
2178 determine_af_and_pullup(struct mbuf **m, sa_family_t *af)
2179 {
2180 	u_char ipv;
2181 	if ((*m)->m_pkthdr.len >= sizeof(struct ip6_hdr))
2182 		*m = m_pullup(*m, sizeof(struct ip6_hdr));
2183 	else if ((*m)->m_pkthdr.len >= sizeof(struct ip))
2184 		*m = m_pullup(*m, sizeof(struct ip));
2185 	else
2186 		return (EAFNOSUPPORT);
2187 	if (*m == NULL)
2188 		return (ENOBUFS);
2189 	ipv = mtod(*m, struct ip *)->ip_v;
2190 	if (ipv == 4)
2191 		*af = AF_INET;
2192 	else if (ipv == 6 && (*m)->m_pkthdr.len >= sizeof(struct ip6_hdr))
2193 		*af = AF_INET6;
2194 	else
2195 		return (EAFNOSUPPORT);
2196 	return (0);
2197 }
2198 
2199 static int
2200 determine_ethertype_and_pullup(struct mbuf **m, int *etp)
2201 {
2202 	struct ether_header *eh;
2203 
2204 	*m = m_pullup(*m, sizeof(struct ether_header));
2205 	if (__predict_false(*m == NULL))
2206 		return (ENOBUFS);
2207 	eh = mtod(*m, struct ether_header *);
2208 	*etp = ntohs(eh->ether_type);
2209 	if (*etp != ETHERTYPE_IP && *etp != ETHERTYPE_IPV6)
2210 		return (EAFNOSUPPORT);
2211 	return (0);
2212 }
2213 
2214 /*
2215  * This should only be invoked by netmap, via nm_os_generic_xmit_frame(), to
2216  * transmit packets from the netmap TX ring.
2217  */
2218 static int
2219 wg_transmit(if_t ifp, struct mbuf *m)
2220 {
2221 	sa_family_t af;
2222 	int et, ret;
2223 	struct mbuf *defragged;
2224 
2225 	KASSERT((if_getcapenable(ifp) & IFCAP_NETMAP) != 0,
2226 	    ("%s: ifp %p is not in netmap mode", __func__, ifp));
2227 
2228 	defragged = m_defrag(m, M_NOWAIT);
2229 	if (defragged)
2230 		m = defragged;
2231 	m = m_unshare(m, M_NOWAIT);
2232 	if (!m) {
2233 		xmit_err(ifp, m, NULL, AF_UNSPEC);
2234 		return (ENOBUFS);
2235 	}
2236 
2237 	ret = determine_ethertype_and_pullup(&m, &et);
2238 	if (ret) {
2239 		xmit_err(ifp, m, NULL, AF_UNSPEC);
2240 		return (ret);
2241 	}
2242 	m_adj(m, sizeof(struct ether_header));
2243 
2244 	ret = determine_af_and_pullup(&m, &af);
2245 	if (ret) {
2246 		xmit_err(ifp, m, NULL, AF_UNSPEC);
2247 		return (ret);
2248 	}
2249 
2250 	/*
2251 	 * netmap only gets to see transient errors, since it handles errors by
2252 	 * refusing to advance the transmit ring and retrying later.
2253 	 */
2254 	ret = wg_xmit(ifp, m, af, if_getmtu(ifp));
2255 	if (ret == ENOBUFS)
2256 		return (ret);
2257 	return (0);
2258 }
2259 
2260 #ifdef DEV_NETMAP
2261 /*
2262  * This should only be invoked by netmap, via nm_os_send_up(), to process
2263  * packets from the host TX ring.
2264  */
2265 static void
2266 wg_if_input(if_t ifp, struct mbuf *m)
2267 {
2268 	int et;
2269 
2270 	KASSERT((if_getcapenable(ifp) & IFCAP_NETMAP) != 0,
2271 	    ("%s: ifp %p is not in netmap mode", __func__, ifp));
2272 
2273 	if (determine_ethertype_and_pullup(&m, &et) != 0) {
2274 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2275 		m_freem(m);
2276 		return;
2277 	}
2278 	CURVNET_SET(if_getvnet(ifp));
2279 	switch (et) {
2280 	case ETHERTYPE_IP:
2281 		m_adj(m, sizeof(struct ether_header));
2282 		netisr_dispatch(NETISR_IP, m);
2283 		break;
2284 	case ETHERTYPE_IPV6:
2285 		m_adj(m, sizeof(struct ether_header));
2286 		netisr_dispatch(NETISR_IPV6, m);
2287 		break;
2288 	default:
2289 		__assert_unreachable();
2290 	}
2291 	CURVNET_RESTORE();
2292 }
2293 
2294 /*
2295  * Deliver a packet to the host RX ring.  Because the interface is in netmap
2296  * mode, the if_transmit() call should pass the packet to netmap_transmit().
2297  */
2298 static int
2299 wg_xmit_netmap(if_t ifp, struct mbuf *m, int af)
2300 {
2301 	struct ether_header *eh;
2302 
2303 	if (__predict_false(if_tunnel_check_nesting(ifp, m, MTAG_WGLOOP,
2304 	    MAX_LOOPS))) {
2305 		printf("%s:%d\n", __func__, __LINE__);
2306 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
2307 		m_freem(m);
2308 		return (ELOOP);
2309 	}
2310 
2311 	M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT);
2312 	if (__predict_false(m == NULL)) {
2313 		if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2314 		return (ENOBUFS);
2315 	}
2316 
2317 	eh = mtod(m, struct ether_header *);
2318 	eh->ether_type = af == AF_INET ?
2319 	    htons(ETHERTYPE_IP) : htons(ETHERTYPE_IPV6);
2320 	memcpy(eh->ether_shost, "\x06\x06\x06\x06\x06\x06", ETHER_ADDR_LEN);
2321 	memcpy(eh->ether_dhost, "\xff\xff\xff\xff\xff\xff", ETHER_ADDR_LEN);
2322 	return (if_transmit(ifp, m));
2323 }
2324 #endif /* DEV_NETMAP */
2325 
2326 static int
2327 wg_output(if_t ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro)
2328 {
2329 	sa_family_t parsed_af;
2330 	uint32_t af, mtu;
2331 	int ret;
2332 	struct mbuf *defragged;
2333 
2334 	/* BPF writes need to be handled specially. */
2335 	if (dst->sa_family == AF_UNSPEC || dst->sa_family == pseudo_AF_HDRCMPLT)
2336 		memcpy(&af, dst->sa_data, sizeof(af));
2337 	else
2338 		af = dst->sa_family;
2339 	if (af == AF_UNSPEC) {
2340 		xmit_err(ifp, m, NULL, af);
2341 		return (EAFNOSUPPORT);
2342 	}
2343 
2344 #ifdef DEV_NETMAP
2345 	if ((if_getcapenable(ifp) & IFCAP_NETMAP) != 0)
2346 		return (wg_xmit_netmap(ifp, m, af));
2347 #endif
2348 
2349 	defragged = m_defrag(m, M_NOWAIT);
2350 	if (defragged)
2351 		m = defragged;
2352 	m = m_unshare(m, M_NOWAIT);
2353 	if (!m) {
2354 		xmit_err(ifp, m, NULL, AF_UNSPEC);
2355 		return (ENOBUFS);
2356 	}
2357 
2358 	ret = determine_af_and_pullup(&m, &parsed_af);
2359 	if (ret) {
2360 		xmit_err(ifp, m, NULL, AF_UNSPEC);
2361 		return (ret);
2362 	}
2363 	if (parsed_af != af) {
2364 		xmit_err(ifp, m, NULL, AF_UNSPEC);
2365 		return (EAFNOSUPPORT);
2366 	}
2367 	mtu = (ro != NULL && ro->ro_mtu > 0) ? ro->ro_mtu : if_getmtu(ifp);
2368 	return (wg_xmit(ifp, m, parsed_af, mtu));
2369 }
2370 
2371 static int
2372 wg_peer_add(struct wg_softc *sc, const nvlist_t *nvl)
2373 {
2374 	uint8_t			 public[WG_KEY_SIZE];
2375 	const void *pub_key, *preshared_key = NULL;
2376 	const struct sockaddr *endpoint;
2377 	int err;
2378 	size_t size;
2379 	struct noise_remote *remote;
2380 	struct wg_peer *peer = NULL;
2381 	bool need_insert = false;
2382 
2383 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2384 
2385 	if (!nvlist_exists_binary(nvl, "public-key")) {
2386 		return (EINVAL);
2387 	}
2388 	pub_key = nvlist_get_binary(nvl, "public-key", &size);
2389 	if (size != WG_KEY_SIZE) {
2390 		return (EINVAL);
2391 	}
2392 	if (noise_local_keys(sc->sc_local, public, NULL) == 0 &&
2393 	    bcmp(public, pub_key, WG_KEY_SIZE) == 0) {
2394 		return (0); // Silently ignored; not actually a failure.
2395 	}
2396 	if ((remote = noise_remote_lookup(sc->sc_local, pub_key)) != NULL)
2397 		peer = noise_remote_arg(remote);
2398 	if (nvlist_exists_bool(nvl, "remove") &&
2399 		nvlist_get_bool(nvl, "remove")) {
2400 		if (remote != NULL) {
2401 			wg_peer_destroy(peer);
2402 			noise_remote_put(remote);
2403 		}
2404 		return (0);
2405 	}
2406 	if (nvlist_exists_bool(nvl, "replace-allowedips") &&
2407 		nvlist_get_bool(nvl, "replace-allowedips") &&
2408 	    peer != NULL) {
2409 
2410 		wg_aip_remove_all(sc, peer);
2411 	}
2412 	if (peer == NULL) {
2413 		peer = wg_peer_alloc(sc, pub_key);
2414 		need_insert = true;
2415 	}
2416 	if (nvlist_exists_binary(nvl, "endpoint")) {
2417 		endpoint = nvlist_get_binary(nvl, "endpoint", &size);
2418 		if (size > sizeof(peer->p_endpoint.e_remote)) {
2419 			err = EINVAL;
2420 			goto out;
2421 		}
2422 		memcpy(&peer->p_endpoint.e_remote, endpoint, size);
2423 	}
2424 	if (nvlist_exists_binary(nvl, "preshared-key")) {
2425 		preshared_key = nvlist_get_binary(nvl, "preshared-key", &size);
2426 		if (size != WG_KEY_SIZE) {
2427 			err = EINVAL;
2428 			goto out;
2429 		}
2430 		noise_remote_set_psk(peer->p_remote, preshared_key);
2431 	}
2432 	if (nvlist_exists_number(nvl, "persistent-keepalive-interval")) {
2433 		uint64_t pki = nvlist_get_number(nvl, "persistent-keepalive-interval");
2434 		if (pki > UINT16_MAX) {
2435 			err = EINVAL;
2436 			goto out;
2437 		}
2438 		wg_timers_set_persistent_keepalive(peer, pki);
2439 	}
2440 	if (nvlist_exists_nvlist_array(nvl, "allowed-ips")) {
2441 		const void *addr;
2442 		uint64_t cidr;
2443 		const nvlist_t * const * aipl;
2444 		size_t allowedip_count;
2445 
2446 		aipl = nvlist_get_nvlist_array(nvl, "allowed-ips", &allowedip_count);
2447 		for (size_t idx = 0; idx < allowedip_count; idx++) {
2448 			if (!nvlist_exists_number(aipl[idx], "cidr"))
2449 				continue;
2450 			cidr = nvlist_get_number(aipl[idx], "cidr");
2451 			if (nvlist_exists_binary(aipl[idx], "ipv4")) {
2452 				addr = nvlist_get_binary(aipl[idx], "ipv4", &size);
2453 				if (addr == NULL || cidr > 32 || size != sizeof(struct in_addr)) {
2454 					err = EINVAL;
2455 					goto out;
2456 				}
2457 				if ((err = wg_aip_add(sc, peer, AF_INET, addr, cidr)) != 0)
2458 					goto out;
2459 			} else if (nvlist_exists_binary(aipl[idx], "ipv6")) {
2460 				addr = nvlist_get_binary(aipl[idx], "ipv6", &size);
2461 				if (addr == NULL || cidr > 128 || size != sizeof(struct in6_addr)) {
2462 					err = EINVAL;
2463 					goto out;
2464 				}
2465 				if ((err = wg_aip_add(sc, peer, AF_INET6, addr, cidr)) != 0)
2466 					goto out;
2467 			} else {
2468 				continue;
2469 			}
2470 		}
2471 	}
2472 	if (need_insert) {
2473 		if ((err = noise_remote_enable(peer->p_remote)) != 0)
2474 			goto out;
2475 		TAILQ_INSERT_TAIL(&sc->sc_peers, peer, p_entry);
2476 		sc->sc_peers_num++;
2477 		if (if_getlinkstate(sc->sc_ifp) == LINK_STATE_UP)
2478 			wg_timers_enable(peer);
2479 	}
2480 	if (remote != NULL)
2481 		noise_remote_put(remote);
2482 	return (0);
2483 out:
2484 	if (need_insert) /* If we fail, only destroy if it was new. */
2485 		wg_peer_destroy(peer);
2486 	if (remote != NULL)
2487 		noise_remote_put(remote);
2488 	return (err);
2489 }
2490 
2491 static int
2492 wgc_set(struct wg_softc *sc, struct wg_data_io *wgd)
2493 {
2494 	uint8_t public[WG_KEY_SIZE], private[WG_KEY_SIZE];
2495 	if_t ifp;
2496 	void *nvlpacked;
2497 	nvlist_t *nvl;
2498 	ssize_t size;
2499 	int err;
2500 
2501 	ifp = sc->sc_ifp;
2502 	if (wgd->wgd_size == 0 || wgd->wgd_data == NULL)
2503 		return (EFAULT);
2504 
2505 	/* Can nvlists be streamed in? It's not nice to impose arbitrary limits like that but
2506 	 * there needs to be _some_ limitation. */
2507 	if (wgd->wgd_size >= UINT32_MAX / 2)
2508 		return (E2BIG);
2509 
2510 	nvlpacked = malloc(wgd->wgd_size, M_TEMP, M_WAITOK | M_ZERO);
2511 
2512 	err = copyin(wgd->wgd_data, nvlpacked, wgd->wgd_size);
2513 	if (err)
2514 		goto out;
2515 	nvl = nvlist_unpack(nvlpacked, wgd->wgd_size, 0);
2516 	if (nvl == NULL) {
2517 		err = EBADMSG;
2518 		goto out;
2519 	}
2520 	sx_xlock(&sc->sc_lock);
2521 	if (nvlist_exists_bool(nvl, "replace-peers") &&
2522 		nvlist_get_bool(nvl, "replace-peers"))
2523 		wg_peer_destroy_all(sc);
2524 	if (nvlist_exists_number(nvl, "listen-port")) {
2525 		uint64_t new_port = nvlist_get_number(nvl, "listen-port");
2526 		if (new_port > UINT16_MAX) {
2527 			err = EINVAL;
2528 			goto out_locked;
2529 		}
2530 		if (new_port != sc->sc_socket.so_port) {
2531 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2532 				if ((err = wg_socket_init(sc, new_port)) != 0)
2533 					goto out_locked;
2534 			} else
2535 				sc->sc_socket.so_port = new_port;
2536 		}
2537 	}
2538 	if (nvlist_exists_binary(nvl, "private-key")) {
2539 		const void *key = nvlist_get_binary(nvl, "private-key", &size);
2540 		if (size != WG_KEY_SIZE) {
2541 			err = EINVAL;
2542 			goto out_locked;
2543 		}
2544 
2545 		if (noise_local_keys(sc->sc_local, NULL, private) != 0 ||
2546 		    timingsafe_bcmp(private, key, WG_KEY_SIZE) != 0) {
2547 			struct wg_peer *peer;
2548 
2549 			if (curve25519_generate_public(public, key)) {
2550 				/* Peer conflict: remove conflicting peer. */
2551 				struct noise_remote *remote;
2552 				if ((remote = noise_remote_lookup(sc->sc_local,
2553 				    public)) != NULL) {
2554 					peer = noise_remote_arg(remote);
2555 					wg_peer_destroy(peer);
2556 					noise_remote_put(remote);
2557 				}
2558 			}
2559 
2560 			/*
2561 			 * Set the private key and invalidate all existing
2562 			 * handshakes.
2563 			 */
2564 			/* Note: we might be removing the private key. */
2565 			noise_local_private(sc->sc_local, key);
2566 			if (noise_local_keys(sc->sc_local, NULL, NULL) == 0)
2567 				cookie_checker_update(&sc->sc_cookie, public);
2568 			else
2569 				cookie_checker_update(&sc->sc_cookie, NULL);
2570 		}
2571 	}
2572 	if (nvlist_exists_number(nvl, "user-cookie")) {
2573 		uint64_t user_cookie = nvlist_get_number(nvl, "user-cookie");
2574 		if (user_cookie > UINT32_MAX) {
2575 			err = EINVAL;
2576 			goto out_locked;
2577 		}
2578 		err = wg_socket_set_cookie(sc, user_cookie);
2579 		if (err)
2580 			goto out_locked;
2581 	}
2582 	if (nvlist_exists_nvlist_array(nvl, "peers")) {
2583 		size_t peercount;
2584 		const nvlist_t * const*nvl_peers;
2585 
2586 		nvl_peers = nvlist_get_nvlist_array(nvl, "peers", &peercount);
2587 		for (int i = 0; i < peercount; i++) {
2588 			err = wg_peer_add(sc, nvl_peers[i]);
2589 			if (err != 0)
2590 				goto out_locked;
2591 		}
2592 	}
2593 
2594 out_locked:
2595 	sx_xunlock(&sc->sc_lock);
2596 	nvlist_destroy(nvl);
2597 out:
2598 	zfree(nvlpacked, M_TEMP);
2599 	return (err);
2600 }
2601 
2602 static int
2603 wgc_get(struct wg_softc *sc, struct wg_data_io *wgd)
2604 {
2605 	uint8_t public_key[WG_KEY_SIZE] = { 0 };
2606 	uint8_t private_key[WG_KEY_SIZE] = { 0 };
2607 	uint8_t preshared_key[NOISE_SYMMETRIC_KEY_LEN] = { 0 };
2608 	nvlist_t *nvl, *nvl_peer, *nvl_aip, **nvl_peers, **nvl_aips;
2609 	size_t size, peer_count, aip_count, i, j;
2610 	struct wg_timespec64 ts64;
2611 	struct wg_peer *peer;
2612 	struct wg_aip *aip;
2613 	void *packed;
2614 	int err = 0;
2615 
2616 	nvl = nvlist_create(0);
2617 	if (!nvl)
2618 		return (ENOMEM);
2619 
2620 	sx_slock(&sc->sc_lock);
2621 
2622 	if (sc->sc_socket.so_port != 0)
2623 		nvlist_add_number(nvl, "listen-port", sc->sc_socket.so_port);
2624 	if (sc->sc_socket.so_user_cookie != 0)
2625 		nvlist_add_number(nvl, "user-cookie", sc->sc_socket.so_user_cookie);
2626 	if (noise_local_keys(sc->sc_local, public_key, private_key) == 0) {
2627 		nvlist_add_binary(nvl, "public-key", public_key, WG_KEY_SIZE);
2628 		if (wgc_privileged(sc))
2629 			nvlist_add_binary(nvl, "private-key", private_key, WG_KEY_SIZE);
2630 		explicit_bzero(private_key, sizeof(private_key));
2631 	}
2632 	peer_count = sc->sc_peers_num;
2633 	if (peer_count) {
2634 		nvl_peers = mallocarray(peer_count, sizeof(void *), M_NVLIST, M_WAITOK | M_ZERO);
2635 		i = 0;
2636 		TAILQ_FOREACH(peer, &sc->sc_peers, p_entry) {
2637 			if (i >= peer_count)
2638 				panic("peers changed from under us");
2639 
2640 			nvl_peers[i++] = nvl_peer = nvlist_create(0);
2641 			if (!nvl_peer) {
2642 				err = ENOMEM;
2643 				goto err_peer;
2644 			}
2645 
2646 			(void)noise_remote_keys(peer->p_remote, public_key, preshared_key);
2647 			nvlist_add_binary(nvl_peer, "public-key", public_key, sizeof(public_key));
2648 			if (wgc_privileged(sc))
2649 				nvlist_add_binary(nvl_peer, "preshared-key", preshared_key, sizeof(preshared_key));
2650 			explicit_bzero(preshared_key, sizeof(preshared_key));
2651 			if (peer->p_endpoint.e_remote.r_sa.sa_family == AF_INET)
2652 				nvlist_add_binary(nvl_peer, "endpoint", &peer->p_endpoint.e_remote, sizeof(struct sockaddr_in));
2653 			else if (peer->p_endpoint.e_remote.r_sa.sa_family == AF_INET6)
2654 				nvlist_add_binary(nvl_peer, "endpoint", &peer->p_endpoint.e_remote, sizeof(struct sockaddr_in6));
2655 			wg_timers_get_last_handshake(peer, &ts64);
2656 			nvlist_add_binary(nvl_peer, "last-handshake-time", &ts64, sizeof(ts64));
2657 			nvlist_add_number(nvl_peer, "persistent-keepalive-interval", peer->p_persistent_keepalive_interval);
2658 			nvlist_add_number(nvl_peer, "rx-bytes", counter_u64_fetch(peer->p_rx_bytes));
2659 			nvlist_add_number(nvl_peer, "tx-bytes", counter_u64_fetch(peer->p_tx_bytes));
2660 
2661 			aip_count = peer->p_aips_num;
2662 			if (aip_count) {
2663 				nvl_aips = mallocarray(aip_count, sizeof(void *), M_NVLIST, M_WAITOK | M_ZERO);
2664 				j = 0;
2665 				LIST_FOREACH(aip, &peer->p_aips, a_entry) {
2666 					if (j >= aip_count)
2667 						panic("aips changed from under us");
2668 
2669 					nvl_aips[j++] = nvl_aip = nvlist_create(0);
2670 					if (!nvl_aip) {
2671 						err = ENOMEM;
2672 						goto err_aip;
2673 					}
2674 					if (aip->a_af == AF_INET) {
2675 						nvlist_add_binary(nvl_aip, "ipv4", &aip->a_addr.in, sizeof(aip->a_addr.in));
2676 						nvlist_add_number(nvl_aip, "cidr", bitcount32(aip->a_mask.ip));
2677 					}
2678 #ifdef INET6
2679 					else if (aip->a_af == AF_INET6) {
2680 						nvlist_add_binary(nvl_aip, "ipv6", &aip->a_addr.in6, sizeof(aip->a_addr.in6));
2681 						nvlist_add_number(nvl_aip, "cidr", in6_mask2len(&aip->a_mask.in6, NULL));
2682 					}
2683 #endif
2684 				}
2685 				nvlist_add_nvlist_array(nvl_peer, "allowed-ips", (const nvlist_t *const *)nvl_aips, aip_count);
2686 			err_aip:
2687 				for (j = 0; j < aip_count; ++j)
2688 					nvlist_destroy(nvl_aips[j]);
2689 				free(nvl_aips, M_NVLIST);
2690 				if (err)
2691 					goto err_peer;
2692 			}
2693 		}
2694 		nvlist_add_nvlist_array(nvl, "peers", (const nvlist_t * const *)nvl_peers, peer_count);
2695 	err_peer:
2696 		for (i = 0; i < peer_count; ++i)
2697 			nvlist_destroy(nvl_peers[i]);
2698 		free(nvl_peers, M_NVLIST);
2699 		if (err) {
2700 			sx_sunlock(&sc->sc_lock);
2701 			goto err;
2702 		}
2703 	}
2704 	sx_sunlock(&sc->sc_lock);
2705 	packed = nvlist_pack(nvl, &size);
2706 	if (!packed) {
2707 		err = ENOMEM;
2708 		goto err;
2709 	}
2710 	if (!wgd->wgd_size) {
2711 		wgd->wgd_size = size;
2712 		goto out;
2713 	}
2714 	if (wgd->wgd_size < size) {
2715 		err = ENOSPC;
2716 		goto out;
2717 	}
2718 	err = copyout(packed, wgd->wgd_data, size);
2719 	wgd->wgd_size = size;
2720 
2721 out:
2722 	zfree(packed, M_NVLIST);
2723 err:
2724 	nvlist_destroy(nvl);
2725 	return (err);
2726 }
2727 
2728 static int
2729 wg_ioctl(if_t ifp, u_long cmd, caddr_t data)
2730 {
2731 	struct wg_data_io *wgd = (struct wg_data_io *)data;
2732 	struct ifreq *ifr = (struct ifreq *)data;
2733 	struct wg_softc *sc;
2734 	int ret = 0;
2735 
2736 	sx_slock(&wg_sx);
2737 	sc = if_getsoftc(ifp);
2738 	if (!sc) {
2739 		ret = ENXIO;
2740 		goto out;
2741 	}
2742 
2743 	switch (cmd) {
2744 	case SIOCSWG:
2745 		ret = priv_check(curthread, PRIV_NET_WG);
2746 		if (ret == 0)
2747 			ret = wgc_set(sc, wgd);
2748 		break;
2749 	case SIOCGWG:
2750 		ret = wgc_get(sc, wgd);
2751 		break;
2752 	/* Interface IOCTLs */
2753 	case SIOCSIFADDR:
2754 		/*
2755 		 * This differs from *BSD norms, but is more uniform with how
2756 		 * WireGuard behaves elsewhere.
2757 		 */
2758 		break;
2759 	case SIOCSIFFLAGS:
2760 		if (if_getflags(ifp) & IFF_UP)
2761 			ret = wg_up(sc);
2762 		else
2763 			wg_down(sc);
2764 		break;
2765 	case SIOCSIFMTU:
2766 		if (ifr->ifr_mtu <= 0 || ifr->ifr_mtu > MAX_MTU)
2767 			ret = EINVAL;
2768 		else
2769 			if_setmtu(ifp, ifr->ifr_mtu);
2770 		break;
2771 	case SIOCADDMULTI:
2772 	case SIOCDELMULTI:
2773 		break;
2774 	case SIOCGTUNFIB:
2775 		ifr->ifr_fib = sc->sc_socket.so_fibnum;
2776 		break;
2777 	case SIOCSTUNFIB:
2778 		ret = priv_check(curthread, PRIV_NET_WG);
2779 		if (ret)
2780 			break;
2781 		ret = priv_check(curthread, PRIV_NET_SETIFFIB);
2782 		if (ret)
2783 			break;
2784 		sx_xlock(&sc->sc_lock);
2785 		ret = wg_socket_set_fibnum(sc, ifr->ifr_fib);
2786 		sx_xunlock(&sc->sc_lock);
2787 		break;
2788 	default:
2789 		ret = ENOTTY;
2790 	}
2791 
2792 out:
2793 	sx_sunlock(&wg_sx);
2794 	return (ret);
2795 }
2796 
2797 static int
2798 wg_up(struct wg_softc *sc)
2799 {
2800 	if_t ifp = sc->sc_ifp;
2801 	struct wg_peer *peer;
2802 	int rc = EBUSY;
2803 
2804 	sx_xlock(&sc->sc_lock);
2805 	/* Jail's being removed, no more wg_up(). */
2806 	if ((sc->sc_flags & WGF_DYING) != 0)
2807 		goto out;
2808 
2809 	/* Silent success if we're already running. */
2810 	rc = 0;
2811 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
2812 		goto out;
2813 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
2814 
2815 	rc = wg_socket_init(sc, sc->sc_socket.so_port);
2816 	if (rc == 0) {
2817 		TAILQ_FOREACH(peer, &sc->sc_peers, p_entry)
2818 			wg_timers_enable(peer);
2819 		if_link_state_change(sc->sc_ifp, LINK_STATE_UP);
2820 	} else {
2821 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2822 		DPRINTF(sc, "Unable to initialize sockets: %d\n", rc);
2823 	}
2824 out:
2825 	sx_xunlock(&sc->sc_lock);
2826 	return (rc);
2827 }
2828 
2829 static void
2830 wg_down(struct wg_softc *sc)
2831 {
2832 	if_t ifp = sc->sc_ifp;
2833 	struct wg_peer *peer;
2834 
2835 	sx_xlock(&sc->sc_lock);
2836 	if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING)) {
2837 		sx_xunlock(&sc->sc_lock);
2838 		return;
2839 	}
2840 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2841 
2842 	TAILQ_FOREACH(peer, &sc->sc_peers, p_entry) {
2843 		wg_queue_purge(&peer->p_stage_queue);
2844 		wg_timers_disable(peer);
2845 	}
2846 
2847 	wg_queue_purge(&sc->sc_handshake_queue);
2848 
2849 	TAILQ_FOREACH(peer, &sc->sc_peers, p_entry) {
2850 		noise_remote_handshake_clear(peer->p_remote);
2851 		noise_remote_keypairs_clear(peer->p_remote);
2852 	}
2853 
2854 	if_link_state_change(sc->sc_ifp, LINK_STATE_DOWN);
2855 	wg_socket_uninit(sc);
2856 
2857 	sx_xunlock(&sc->sc_lock);
2858 }
2859 
2860 static int
2861 wg_clone_create(struct if_clone *ifc, char *name, size_t len,
2862     struct ifc_data *ifd, struct ifnet **ifpp)
2863 {
2864 	struct wg_softc *sc;
2865 	if_t ifp;
2866 
2867 	sc = malloc(sizeof(*sc), M_WG, M_WAITOK | M_ZERO);
2868 
2869 	sc->sc_local = noise_local_alloc(sc);
2870 
2871 	sc->sc_encrypt = mallocarray(sizeof(struct grouptask), mp_ncpus, M_WG, M_WAITOK | M_ZERO);
2872 
2873 	sc->sc_decrypt = mallocarray(sizeof(struct grouptask), mp_ncpus, M_WG, M_WAITOK | M_ZERO);
2874 
2875 	if (!rn_inithead((void **)&sc->sc_aip4, offsetof(struct aip_addr, in) * NBBY))
2876 		goto free_decrypt;
2877 
2878 	if (!rn_inithead((void **)&sc->sc_aip6, offsetof(struct aip_addr, in6) * NBBY))
2879 		goto free_aip4;
2880 
2881 	atomic_add_int(&clone_count, 1);
2882 	ifp = sc->sc_ifp = if_alloc(IFT_WIREGUARD);
2883 
2884 	sc->sc_ucred = crhold(curthread->td_ucred);
2885 	sc->sc_socket.so_fibnum = curthread->td_proc->p_fibnum;
2886 	sc->sc_socket.so_port = 0;
2887 
2888 	TAILQ_INIT(&sc->sc_peers);
2889 	sc->sc_peers_num = 0;
2890 
2891 	cookie_checker_init(&sc->sc_cookie);
2892 
2893 	RADIX_NODE_HEAD_LOCK_INIT(sc->sc_aip4);
2894 	RADIX_NODE_HEAD_LOCK_INIT(sc->sc_aip6);
2895 
2896 	GROUPTASK_INIT(&sc->sc_handshake, 0, (gtask_fn_t *)wg_softc_handshake_receive, sc);
2897 	taskqgroup_attach(qgroup_wg_tqg, &sc->sc_handshake, sc, NULL, NULL, "wg tx initiation");
2898 	wg_queue_init(&sc->sc_handshake_queue, "hsq");
2899 
2900 	for (int i = 0; i < mp_ncpus; i++) {
2901 		GROUPTASK_INIT(&sc->sc_encrypt[i], 0,
2902 		     (gtask_fn_t *)wg_softc_encrypt, sc);
2903 		taskqgroup_attach_cpu(qgroup_wg_tqg, &sc->sc_encrypt[i], sc, i, NULL, NULL, "wg encrypt");
2904 		GROUPTASK_INIT(&sc->sc_decrypt[i], 0,
2905 		    (gtask_fn_t *)wg_softc_decrypt, sc);
2906 		taskqgroup_attach_cpu(qgroup_wg_tqg, &sc->sc_decrypt[i], sc, i, NULL, NULL, "wg decrypt");
2907 	}
2908 
2909 	wg_queue_init(&sc->sc_encrypt_parallel, "encp");
2910 	wg_queue_init(&sc->sc_decrypt_parallel, "decp");
2911 
2912 	sx_init(&sc->sc_lock, "wg softc lock");
2913 
2914 	if_setsoftc(ifp, sc);
2915 	if_setcapabilities(ifp, WG_CAPS);
2916 	if_setcapenable(ifp, WG_CAPS);
2917 	if_initname(ifp, wgname, ifd->unit);
2918 
2919 	if_setmtu(ifp, DEFAULT_MTU);
2920 	if_setflags(ifp, IFF_NOARP | IFF_MULTICAST);
2921 	if_setinitfn(ifp, wg_init);
2922 	if_setreassignfn(ifp, wg_reassign);
2923 	if_setqflushfn(ifp, wg_qflush);
2924 	if_settransmitfn(ifp, wg_transmit);
2925 #ifdef DEV_NETMAP
2926 	if_setinputfn(ifp, wg_if_input);
2927 #endif
2928 	if_setoutputfn(ifp, wg_output);
2929 	if_setioctlfn(ifp, wg_ioctl);
2930 	if_attach(ifp);
2931 	bpfattach(ifp, DLT_NULL, sizeof(uint32_t));
2932 #ifdef INET6
2933 	ND_IFINFO(ifp)->flags &= ~ND6_IFF_AUTO_LINKLOCAL;
2934 	ND_IFINFO(ifp)->flags |= ND6_IFF_NO_DAD;
2935 #endif
2936 	sx_xlock(&wg_sx);
2937 	LIST_INSERT_HEAD(&wg_list, sc, sc_entry);
2938 	sx_xunlock(&wg_sx);
2939 	*ifpp = ifp;
2940 	return (0);
2941 free_aip4:
2942 	RADIX_NODE_HEAD_DESTROY(sc->sc_aip4);
2943 	free(sc->sc_aip4, M_RTABLE);
2944 free_decrypt:
2945 	free(sc->sc_decrypt, M_WG);
2946 	free(sc->sc_encrypt, M_WG);
2947 	noise_local_free(sc->sc_local, NULL);
2948 	free(sc, M_WG);
2949 	return (ENOMEM);
2950 }
2951 
2952 static void
2953 wg_clone_deferred_free(struct noise_local *l)
2954 {
2955 	struct wg_softc *sc = noise_local_arg(l);
2956 
2957 	free(sc, M_WG);
2958 	atomic_add_int(&clone_count, -1);
2959 }
2960 
2961 static int
2962 wg_clone_destroy(struct if_clone *ifc, if_t ifp, uint32_t flags)
2963 {
2964 	struct wg_softc *sc = if_getsoftc(ifp);
2965 	struct ucred *cred;
2966 
2967 	sx_xlock(&wg_sx);
2968 	if_setsoftc(ifp, NULL);
2969 	sx_xlock(&sc->sc_lock);
2970 	sc->sc_flags |= WGF_DYING;
2971 	cred = sc->sc_ucred;
2972 	sc->sc_ucred = NULL;
2973 	sx_xunlock(&sc->sc_lock);
2974 	LIST_REMOVE(sc, sc_entry);
2975 	sx_xunlock(&wg_sx);
2976 
2977 	if_link_state_change(sc->sc_ifp, LINK_STATE_DOWN);
2978 	CURVNET_SET(if_getvnet(sc->sc_ifp));
2979 	if_purgeaddrs(sc->sc_ifp);
2980 	CURVNET_RESTORE();
2981 
2982 	sx_xlock(&sc->sc_lock);
2983 	wg_socket_uninit(sc);
2984 	sx_xunlock(&sc->sc_lock);
2985 
2986 	/*
2987 	 * No guarantees that all traffic have passed until the epoch has
2988 	 * elapsed with the socket closed.
2989 	 */
2990 	NET_EPOCH_WAIT();
2991 
2992 	taskqgroup_drain_all(qgroup_wg_tqg);
2993 	sx_xlock(&sc->sc_lock);
2994 	wg_peer_destroy_all(sc);
2995 	NET_EPOCH_DRAIN_CALLBACKS();
2996 	sx_xunlock(&sc->sc_lock);
2997 	sx_destroy(&sc->sc_lock);
2998 	taskqgroup_detach(qgroup_wg_tqg, &sc->sc_handshake);
2999 	for (int i = 0; i < mp_ncpus; i++) {
3000 		taskqgroup_detach(qgroup_wg_tqg, &sc->sc_encrypt[i]);
3001 		taskqgroup_detach(qgroup_wg_tqg, &sc->sc_decrypt[i]);
3002 	}
3003 	free(sc->sc_encrypt, M_WG);
3004 	free(sc->sc_decrypt, M_WG);
3005 	wg_queue_deinit(&sc->sc_handshake_queue);
3006 	wg_queue_deinit(&sc->sc_encrypt_parallel);
3007 	wg_queue_deinit(&sc->sc_decrypt_parallel);
3008 
3009 	RADIX_NODE_HEAD_DESTROY(sc->sc_aip4);
3010 	RADIX_NODE_HEAD_DESTROY(sc->sc_aip6);
3011 	rn_detachhead((void **)&sc->sc_aip4);
3012 	rn_detachhead((void **)&sc->sc_aip6);
3013 
3014 	cookie_checker_free(&sc->sc_cookie);
3015 
3016 	if (cred != NULL)
3017 		crfree(cred);
3018 	bpfdetach(sc->sc_ifp);
3019 	if_detach(sc->sc_ifp);
3020 	if_free(sc->sc_ifp);
3021 
3022 	noise_local_free(sc->sc_local, wg_clone_deferred_free);
3023 
3024 	return (0);
3025 }
3026 
3027 static void
3028 wg_qflush(if_t ifp __unused)
3029 {
3030 }
3031 
3032 /*
3033  * Privileged information (private-key, preshared-key) are only exported for
3034  * root and jailed root by default.
3035  */
3036 static bool
3037 wgc_privileged(struct wg_softc *sc)
3038 {
3039 	struct thread *td;
3040 
3041 	td = curthread;
3042 	return (priv_check(td, PRIV_NET_WG) == 0);
3043 }
3044 
3045 static void
3046 wg_reassign(if_t ifp, struct vnet *new_vnet __unused,
3047     char *unused __unused)
3048 {
3049 	struct wg_softc *sc;
3050 
3051 	sc = if_getsoftc(ifp);
3052 	wg_down(sc);
3053 }
3054 
3055 static void
3056 wg_init(void *xsc)
3057 {
3058 	struct wg_softc *sc;
3059 
3060 	sc = xsc;
3061 	wg_up(sc);
3062 }
3063 
3064 static void
3065 vnet_wg_init(const void *unused __unused)
3066 {
3067 	struct if_clone_addreq req = {
3068 		.create_f = wg_clone_create,
3069 		.destroy_f = wg_clone_destroy,
3070 		.flags = IFC_F_AUTOUNIT,
3071 	};
3072 	V_wg_cloner = ifc_attach_cloner(wgname, &req);
3073 }
3074 VNET_SYSINIT(vnet_wg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
3075 	     vnet_wg_init, NULL);
3076 
3077 static void
3078 vnet_wg_uninit(const void *unused __unused)
3079 {
3080 	if (V_wg_cloner)
3081 		ifc_detach_cloner(V_wg_cloner);
3082 }
3083 VNET_SYSUNINIT(vnet_wg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
3084 	       vnet_wg_uninit, NULL);
3085 
3086 static int
3087 wg_prison_remove(void *obj, void *data __unused)
3088 {
3089 	const struct prison *pr = obj;
3090 	struct wg_softc *sc;
3091 
3092 	/*
3093 	 * Do a pass through all if_wg interfaces and release creds on any from
3094 	 * the jail that are supposed to be going away.  This will, in turn, let
3095 	 * the jail die so that we don't end up with Schrödinger's jail.
3096 	 */
3097 	sx_slock(&wg_sx);
3098 	LIST_FOREACH(sc, &wg_list, sc_entry) {
3099 		sx_xlock(&sc->sc_lock);
3100 		if (!(sc->sc_flags & WGF_DYING) && sc->sc_ucred && sc->sc_ucred->cr_prison == pr) {
3101 			struct ucred *cred = sc->sc_ucred;
3102 			DPRINTF(sc, "Creating jail exiting\n");
3103 			if_link_state_change(sc->sc_ifp, LINK_STATE_DOWN);
3104 			wg_socket_uninit(sc);
3105 			sc->sc_ucred = NULL;
3106 			crfree(cred);
3107 			sc->sc_flags |= WGF_DYING;
3108 		}
3109 		sx_xunlock(&sc->sc_lock);
3110 	}
3111 	sx_sunlock(&wg_sx);
3112 
3113 	return (0);
3114 }
3115 
3116 #ifdef SELFTESTS
3117 #include "selftest/allowedips.c"
3118 static bool wg_run_selftests(void)
3119 {
3120 	bool ret = true;
3121 	ret &= wg_allowedips_selftest();
3122 	ret &= noise_counter_selftest();
3123 	ret &= cookie_selftest();
3124 	return ret;
3125 }
3126 #else
3127 static inline bool wg_run_selftests(void) { return true; }
3128 #endif
3129 
3130 static int
3131 wg_module_init(void)
3132 {
3133 	int ret;
3134 	osd_method_t methods[PR_MAXMETHOD] = {
3135 		[PR_METHOD_REMOVE] = wg_prison_remove,
3136 	};
3137 
3138 	wg_packet_zone = uma_zcreate("wg packet", sizeof(struct wg_packet),
3139 	     NULL, NULL, NULL, NULL, 0, 0);
3140 
3141 	ret = crypto_init();
3142 	if (ret != 0)
3143 		return (ret);
3144 	ret = cookie_init();
3145 	if (ret != 0)
3146 		return (ret);
3147 
3148 	wg_osd_jail_slot = osd_jail_register(NULL, methods);
3149 
3150 	if (!wg_run_selftests())
3151 		return (ENOTRECOVERABLE);
3152 
3153 	return (0);
3154 }
3155 
3156 static void
3157 wg_module_deinit(void)
3158 {
3159 	VNET_ITERATOR_DECL(vnet_iter);
3160 	VNET_LIST_RLOCK();
3161 	VNET_FOREACH(vnet_iter) {
3162 		struct if_clone *clone = VNET_VNET(vnet_iter, wg_cloner);
3163 		if (clone) {
3164 			ifc_detach_cloner(clone);
3165 			VNET_VNET(vnet_iter, wg_cloner) = NULL;
3166 		}
3167 	}
3168 	VNET_LIST_RUNLOCK();
3169 	NET_EPOCH_WAIT();
3170 	MPASS(LIST_EMPTY(&wg_list));
3171 	if (wg_osd_jail_slot != 0)
3172 		osd_jail_deregister(wg_osd_jail_slot);
3173 	cookie_deinit();
3174 	crypto_deinit();
3175 	if (wg_packet_zone != NULL)
3176 		uma_zdestroy(wg_packet_zone);
3177 }
3178 
3179 static int
3180 wg_module_event_handler(module_t mod, int what, void *arg)
3181 {
3182 	switch (what) {
3183 		case MOD_LOAD:
3184 			return wg_module_init();
3185 		case MOD_UNLOAD:
3186 			wg_module_deinit();
3187 			break;
3188 		default:
3189 			return (EOPNOTSUPP);
3190 	}
3191 	return (0);
3192 }
3193 
3194 static moduledata_t wg_moduledata = {
3195 	"if_wg",
3196 	wg_module_event_handler,
3197 	NULL
3198 };
3199 
3200 DECLARE_MODULE(if_wg, wg_moduledata, SI_SUB_PSEUDO, SI_ORDER_ANY);
3201 MODULE_VERSION(if_wg, WIREGUARD_VERSION);
3202 MODULE_DEPEND(if_wg, crypto, 1, 1, 1);
3203