Home | History | Annotate | Line # | Download | only in net
if_wg.c revision 1.78
      1 /*	$NetBSD: if_wg.c,v 1.78 2024/03/10 04:21:47 riastradh Exp $	*/
      2 
      3 /*
      4  * Copyright (C) Ryota Ozaki <ozaki.ryota (at) gmail.com>
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. Neither the name of the project nor the names of its contributors
     16  *    may be used to endorse or promote products derived from this software
     17  *    without specific prior written permission.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
     20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
     23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     29  * SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * This network interface aims to implement the WireGuard protocol.
     34  * The implementation is based on the paper of WireGuard as of
     35  * 2018-06-30 [1].  The paper is referred in the source code with label
     36  * [W].  Also the specification of the Noise protocol framework as of
     37  * 2018-07-11 [2] is referred with label [N].
     38  *
     39  * [1] https://www.wireguard.com/papers/wireguard.pdf
     40  * [2] http://noiseprotocol.org/noise.pdf
     41  */
     42 
     43 #include <sys/cdefs.h>
     44 __KERNEL_RCSID(0, "$NetBSD: if_wg.c,v 1.78 2024/03/10 04:21:47 riastradh Exp $");
     45 
     46 #ifdef _KERNEL_OPT
     47 #include "opt_altq_enabled.h"
     48 #include "opt_inet.h"
     49 #endif
     50 
     51 #include <sys/param.h>
     52 #include <sys/types.h>
     53 
     54 #include <sys/atomic.h>
     55 #include <sys/callout.h>
     56 #include <sys/cprng.h>
     57 #include <sys/cpu.h>
     58 #include <sys/device.h>
     59 #include <sys/domain.h>
     60 #include <sys/errno.h>
     61 #include <sys/intr.h>
     62 #include <sys/ioctl.h>
     63 #include <sys/kernel.h>
     64 #include <sys/kmem.h>
     65 #include <sys/mbuf.h>
     66 #include <sys/module.h>
     67 #include <sys/mutex.h>
     68 #include <sys/once.h>
     69 #include <sys/percpu.h>
     70 #include <sys/pserialize.h>
     71 #include <sys/psref.h>
     72 #include <sys/queue.h>
     73 #include <sys/rwlock.h>
     74 #include <sys/socket.h>
     75 #include <sys/socketvar.h>
     76 #include <sys/sockio.h>
     77 #include <sys/sysctl.h>
     78 #include <sys/syslog.h>
     79 #include <sys/systm.h>
     80 #include <sys/thmap.h>
     81 #include <sys/threadpool.h>
     82 #include <sys/time.h>
     83 #include <sys/timespec.h>
     84 #include <sys/workqueue.h>
     85 
     86 #include <net/bpf.h>
     87 #include <net/if.h>
     88 #include <net/if_types.h>
     89 #include <net/if_wg.h>
     90 #include <net/pktqueue.h>
     91 #include <net/route.h>
     92 
     93 #include <netinet/in.h>
     94 #include <netinet/in_pcb.h>
     95 #include <netinet/in_var.h>
     96 #include <netinet/ip.h>
     97 #include <netinet/ip_var.h>
     98 #include <netinet/udp.h>
     99 #include <netinet/udp_var.h>
    100 
    101 #ifdef INET6
    102 #include <netinet/ip6.h>
    103 #include <netinet6/in6_pcb.h>
    104 #include <netinet6/in6_var.h>
    105 #include <netinet6/ip6_var.h>
    106 #include <netinet6/udp6_var.h>
    107 #endif /* INET6 */
    108 
    109 #include <prop/proplib.h>
    110 
    111 #include <crypto/blake2/blake2s.h>
    112 #include <crypto/sodium/crypto_aead_chacha20poly1305.h>
    113 #include <crypto/sodium/crypto_aead_xchacha20poly1305.h>
    114 #include <crypto/sodium/crypto_scalarmult.h>
    115 
    116 #include "ioconf.h"
    117 
    118 #ifdef WG_RUMPKERNEL
    119 #include "wg_user.h"
    120 #endif
    121 
    122 /*
    123  * Data structures
    124  * - struct wg_softc is an instance of wg interfaces
    125  *   - It has a list of peers (struct wg_peer)
    126  *   - It has a threadpool job that sends/receives handshake messages and
    127  *     runs event handlers
    128  *   - It has its own two routing tables: one is for IPv4 and the other IPv6
    129  * - struct wg_peer is a representative of a peer
    130  *   - It has a struct work to handle handshakes and timer tasks
    131  *   - It has a pair of session instances (struct wg_session)
    132  *   - It has a pair of endpoint instances (struct wg_sockaddr)
    133  *     - Normally one endpoint is used and the second one is used only on
    134  *       a peer migration (a change of peer's IP address)
    135  *   - It has a list of IP addresses and sub networks called allowedips
    136  *     (struct wg_allowedip)
    137  *     - A packets sent over a session is allowed if its destination matches
    138  *       any IP addresses or sub networks of the list
    139  * - struct wg_session represents a session of a secure tunnel with a peer
    140  *   - Two instances of sessions belong to a peer; a stable session and a
    141  *     unstable session
    142  *   - A handshake process of a session always starts with a unstable instance
    143  *   - Once a session is established, its instance becomes stable and the
    144  *     other becomes unstable instead
    145  *   - Data messages are always sent via a stable session
    146  *
    147  * Locking notes:
    148  * - Each wg has a mutex(9) wg_lock, and a rwlock(9) wg_rwlock
    149  *   - Changes to the peer list are serialized by wg_lock
    150  *   - The peer list may be read with pserialize(9) and psref(9)
    151  *   - The rwlock (wg_rwlock) protects the routing tables (wg_rtable_ipv[46])
    152  *     => XXX replace by pserialize when routing table is psz-safe
    153  * - Each peer (struct wg_peer, wgp) has a mutex wgp_lock, which can be taken
    154  *   only in thread context and serializes:
    155  *   - the stable and unstable session pointers
    156  *   - all unstable session state
    157  * - Packet processing may be done in softint context:
    158  *   - The stable session can be read under pserialize(9) or psref(9)
    159  *     - The stable session is always ESTABLISHED
    160  *     - On a session swap, we must wait for all readers to release a
    161  *       reference to a stable session before changing wgs_state and
    162  *       session states
    163  * - Lock order: wg_lock -> wgp_lock
    164  */
    165 
    166 
    167 #define WGLOG(level, fmt, args...)					      \
    168 	log(level, "%s: " fmt, __func__, ##args)
    169 
    170 /* Debug options */
    171 #ifdef WG_DEBUG
    172 /* Output debug logs */
    173 #ifndef WG_DEBUG_LOG
    174 #define WG_DEBUG_LOG
    175 #endif
    176 /* Output trace logs */
    177 #ifndef WG_DEBUG_TRACE
    178 #define WG_DEBUG_TRACE
    179 #endif
    180 /* Output hash values, etc. */
    181 #ifndef WG_DEBUG_DUMP
    182 #define WG_DEBUG_DUMP
    183 #endif
    184 /* Make some internal parameters configurable for testing and debugging */
    185 #ifndef WG_DEBUG_PARAMS
    186 #define WG_DEBUG_PARAMS
    187 #endif
    188 #endif
    189 
    190 #ifdef WG_DEBUG_TRACE
    191 #define WG_TRACE(msg)							      \
    192 	log(LOG_DEBUG, "%s:%d: %s\n", __func__, __LINE__, (msg))
    193 #else
    194 #define WG_TRACE(msg)	__nothing
    195 #endif
    196 
    197 #ifdef WG_DEBUG_LOG
    198 #define WG_DLOG(fmt, args...)	log(LOG_DEBUG, "%s: " fmt, __func__, ##args)
    199 #else
    200 #define WG_DLOG(fmt, args...)	__nothing
    201 #endif
    202 
    203 #define WG_LOG_RATECHECK(wgprc, level, fmt, args...)	do {		\
    204 	if (ppsratecheck(&(wgprc)->wgprc_lasttime,			\
    205 	    &(wgprc)->wgprc_curpps, 1)) {				\
    206 		log(level, fmt, ##args);				\
    207 	}								\
    208 } while (0)
    209 
    210 #ifdef WG_DEBUG_PARAMS
    211 static bool wg_force_underload = false;
    212 #endif
    213 
    214 #ifdef WG_DEBUG_DUMP
    215 
    216 static char *
    217 gethexdump(const char *p, size_t n)
    218 {
    219 	char *buf;
    220 	size_t i;
    221 
    222 	if (n > SIZE_MAX/3 - 1)
    223 		return NULL;
    224 	buf = kmem_alloc(3*n + 1, KM_NOSLEEP);
    225 	if (buf == NULL)
    226 		return NULL;
    227 	for (i = 0; i < n; i++)
    228 		snprintf(buf + 3*i, 3 + 1, " %02hhx", p[i]);
    229 	return buf;
    230 }
    231 
    232 static void
    233 puthexdump(char *buf, const void *p, size_t n)
    234 {
    235 
    236 	if (buf == NULL)
    237 		return;
    238 	kmem_free(buf, 3*n + 1);
    239 }
    240 
    241 #ifdef WG_RUMPKERNEL
    242 static void
    243 wg_dump_buf(const char *func, const char *buf, const size_t size)
    244 {
    245 	char *hex = gethexdump(buf, size);
    246 
    247 	log(LOG_DEBUG, "%s: %s\n", func, hex ? hex : "(enomem)");
    248 	puthexdump(hex, buf, size);
    249 }
    250 #endif
    251 
    252 static void
    253 wg_dump_hash(const uint8_t *func, const uint8_t *name, const uint8_t *hash,
    254     const size_t size)
    255 {
    256 	char *hex = gethexdump(hash, size);
    257 
    258 	log(LOG_DEBUG, "%s: %s: %s\n", func, name, hex ? hex : "(enomem)");
    259 	puthexdump(hex, hash, size);
    260 }
    261 
    262 #define WG_DUMP_HASH(name, hash) \
    263 	wg_dump_hash(__func__, name, hash, WG_HASH_LEN)
    264 #define WG_DUMP_HASH48(name, hash) \
    265 	wg_dump_hash(__func__, name, hash, 48)
    266 #define WG_DUMP_BUF(buf, size) \
    267 	wg_dump_buf(__func__, buf, size)
    268 #else
    269 #define WG_DUMP_HASH(name, hash)	__nothing
    270 #define WG_DUMP_HASH48(name, hash)	__nothing
    271 #define WG_DUMP_BUF(buf, size)	__nothing
    272 #endif /* WG_DEBUG_DUMP */
    273 
    274 /* chosen somewhat arbitrarily -- fits in signed 16 bits NUL-terminated */
    275 #define	WG_MAX_PROPLEN		32766
    276 
    277 #define WG_MTU			1420
    278 #define WG_ALLOWEDIPS		16
    279 
    280 #define CURVE25519_KEY_LEN	32
    281 #define TAI64N_LEN		sizeof(uint32_t) * 3
    282 #define POLY1305_AUTHTAG_LEN	16
    283 #define HMAC_BLOCK_LEN		64
    284 
    285 /* [N] 4.1: "DHLEN must be 32 or greater."  WireGuard chooses 32. */
    286 /* [N] 4.3: Hash functions */
    287 #define NOISE_DHLEN		32
    288 /* [N] 4.3: "Must be 32 or 64."  WireGuard chooses 32. */
    289 #define NOISE_HASHLEN		32
    290 #define NOISE_BLOCKLEN		64
    291 #define NOISE_HKDF_OUTPUT_LEN	NOISE_HASHLEN
    292 /* [N] 5.1: "k" */
    293 #define NOISE_CIPHER_KEY_LEN	32
    294 /*
    295  * [N] 9.2: "psk"
    296  *          "... psk is a 32-byte secret value provided by the application."
    297  */
    298 #define NOISE_PRESHARED_KEY_LEN	32
    299 
    300 #define WG_STATIC_KEY_LEN	CURVE25519_KEY_LEN
    301 #define WG_TIMESTAMP_LEN	TAI64N_LEN
    302 
    303 #define WG_PRESHARED_KEY_LEN	NOISE_PRESHARED_KEY_LEN
    304 
    305 #define WG_COOKIE_LEN		16
    306 #define WG_MAC_LEN		16
    307 #define WG_RANDVAL_LEN		24
    308 
    309 #define WG_EPHEMERAL_KEY_LEN	CURVE25519_KEY_LEN
    310 /* [N] 5.2: "ck: A chaining key of HASHLEN bytes" */
    311 #define WG_CHAINING_KEY_LEN	NOISE_HASHLEN
    312 /* [N] 5.2: "h: A hash output of HASHLEN bytes" */
    313 #define WG_HASH_LEN		NOISE_HASHLEN
    314 #define WG_CIPHER_KEY_LEN	NOISE_CIPHER_KEY_LEN
    315 #define WG_DH_OUTPUT_LEN	NOISE_DHLEN
    316 #define WG_KDF_OUTPUT_LEN	NOISE_HKDF_OUTPUT_LEN
    317 #define WG_AUTHTAG_LEN		POLY1305_AUTHTAG_LEN
    318 #define WG_DATA_KEY_LEN		32
    319 #define WG_SALT_LEN		24
    320 
    321 /*
    322  * The protocol messages
    323  */
    324 struct wg_msg {
    325 	uint32_t	wgm_type;
    326 } __packed;
    327 
    328 /* [W] 5.4.2 First Message: Initiator to Responder */
    329 struct wg_msg_init {
    330 	uint32_t	wgmi_type;
    331 	uint32_t	wgmi_sender;
    332 	uint8_t		wgmi_ephemeral[WG_EPHEMERAL_KEY_LEN];
    333 	uint8_t		wgmi_static[WG_STATIC_KEY_LEN + WG_AUTHTAG_LEN];
    334 	uint8_t		wgmi_timestamp[WG_TIMESTAMP_LEN + WG_AUTHTAG_LEN];
    335 	uint8_t		wgmi_mac1[WG_MAC_LEN];
    336 	uint8_t		wgmi_mac2[WG_MAC_LEN];
    337 } __packed;
    338 
    339 /* [W] 5.4.3 Second Message: Responder to Initiator */
    340 struct wg_msg_resp {
    341 	uint32_t	wgmr_type;
    342 	uint32_t	wgmr_sender;
    343 	uint32_t	wgmr_receiver;
    344 	uint8_t		wgmr_ephemeral[WG_EPHEMERAL_KEY_LEN];
    345 	uint8_t		wgmr_empty[0 + WG_AUTHTAG_LEN];
    346 	uint8_t		wgmr_mac1[WG_MAC_LEN];
    347 	uint8_t		wgmr_mac2[WG_MAC_LEN];
    348 } __packed;
    349 
    350 /* [W] 5.4.6 Subsequent Messages: Transport Data Messages */
    351 struct wg_msg_data {
    352 	uint32_t	wgmd_type;
    353 	uint32_t	wgmd_receiver;
    354 	uint64_t	wgmd_counter;
    355 	uint32_t	wgmd_packet[0];
    356 } __packed;
    357 
    358 /* [W] 5.4.7 Under Load: Cookie Reply Message */
    359 struct wg_msg_cookie {
    360 	uint32_t	wgmc_type;
    361 	uint32_t	wgmc_receiver;
    362 	uint8_t		wgmc_salt[WG_SALT_LEN];
    363 	uint8_t		wgmc_cookie[WG_COOKIE_LEN + WG_AUTHTAG_LEN];
    364 } __packed;
    365 
    366 #define WG_MSG_TYPE_INIT		1
    367 #define WG_MSG_TYPE_RESP		2
    368 #define WG_MSG_TYPE_COOKIE		3
    369 #define WG_MSG_TYPE_DATA		4
    370 #define WG_MSG_TYPE_MAX			WG_MSG_TYPE_DATA
    371 
    372 /* Sliding windows */
    373 
    374 #define	SLIWIN_BITS	2048u
    375 #define	SLIWIN_TYPE	uint32_t
    376 #define	SLIWIN_BPW	NBBY*sizeof(SLIWIN_TYPE)
    377 #define	SLIWIN_WORDS	howmany(SLIWIN_BITS, SLIWIN_BPW)
    378 #define	SLIWIN_NPKT	(SLIWIN_BITS - NBBY*sizeof(SLIWIN_TYPE))
    379 
    380 struct sliwin {
    381 	SLIWIN_TYPE	B[SLIWIN_WORDS];
    382 	uint64_t	T;
    383 };
    384 
    385 static void
    386 sliwin_reset(struct sliwin *W)
    387 {
    388 
    389 	memset(W, 0, sizeof(*W));
    390 }
    391 
    392 static int
    393 sliwin_check_fast(const volatile struct sliwin *W, uint64_t S)
    394 {
    395 
    396 	/*
    397 	 * If it's more than one window older than the highest sequence
    398 	 * number we've seen, reject.
    399 	 */
    400 #ifdef __HAVE_ATOMIC64_LOADSTORE
    401 	if (S + SLIWIN_NPKT < atomic_load_relaxed(&W->T))
    402 		return EAUTH;
    403 #endif
    404 
    405 	/*
    406 	 * Otherwise, we need to take the lock to decide, so don't
    407 	 * reject just yet.  Caller must serialize a call to
    408 	 * sliwin_update in this case.
    409 	 */
    410 	return 0;
    411 }
    412 
    413 static int
    414 sliwin_update(struct sliwin *W, uint64_t S)
    415 {
    416 	unsigned word, bit;
    417 
    418 	/*
    419 	 * If it's more than one window older than the highest sequence
    420 	 * number we've seen, reject.
    421 	 */
    422 	if (S + SLIWIN_NPKT < W->T)
    423 		return EAUTH;
    424 
    425 	/*
    426 	 * If it's higher than the highest sequence number we've seen,
    427 	 * advance the window.
    428 	 */
    429 	if (S > W->T) {
    430 		uint64_t i = W->T / SLIWIN_BPW;
    431 		uint64_t j = S / SLIWIN_BPW;
    432 		unsigned k;
    433 
    434 		for (k = 0; k < MIN(j - i, SLIWIN_WORDS); k++)
    435 			W->B[(i + k + 1) % SLIWIN_WORDS] = 0;
    436 #ifdef __HAVE_ATOMIC64_LOADSTORE
    437 		atomic_store_relaxed(&W->T, S);
    438 #else
    439 		W->T = S;
    440 #endif
    441 	}
    442 
    443 	/* Test and set the bit -- if already set, reject.  */
    444 	word = (S / SLIWIN_BPW) % SLIWIN_WORDS;
    445 	bit = S % SLIWIN_BPW;
    446 	if (W->B[word] & (1UL << bit))
    447 		return EAUTH;
    448 	W->B[word] |= 1U << bit;
    449 
    450 	/* Accept!  */
    451 	return 0;
    452 }
    453 
    454 struct wg_session {
    455 	struct wg_peer	*wgs_peer;
    456 	struct psref_target
    457 			wgs_psref;
    458 
    459 	int		wgs_state;
    460 #define WGS_STATE_UNKNOWN	0
    461 #define WGS_STATE_INIT_ACTIVE	1
    462 #define WGS_STATE_INIT_PASSIVE	2
    463 #define WGS_STATE_ESTABLISHED	3
    464 #define WGS_STATE_DESTROYING	4
    465 
    466 	time_t		wgs_time_established;
    467 	time_t		wgs_time_last_data_sent;
    468 	bool		wgs_is_initiator;
    469 
    470 	uint32_t	wgs_local_index;
    471 	uint32_t	wgs_remote_index;
    472 #ifdef __HAVE_ATOMIC64_LOADSTORE
    473 	volatile uint64_t
    474 			wgs_send_counter;
    475 #else
    476 	kmutex_t	wgs_send_counter_lock;
    477 	uint64_t	wgs_send_counter;
    478 #endif
    479 
    480 	struct {
    481 		kmutex_t	lock;
    482 		struct sliwin	window;
    483 	}		*wgs_recvwin;
    484 
    485 	uint8_t		wgs_handshake_hash[WG_HASH_LEN];
    486 	uint8_t		wgs_chaining_key[WG_CHAINING_KEY_LEN];
    487 	uint8_t		wgs_ephemeral_key_pub[WG_EPHEMERAL_KEY_LEN];
    488 	uint8_t		wgs_ephemeral_key_priv[WG_EPHEMERAL_KEY_LEN];
    489 	uint8_t		wgs_ephemeral_key_peer[WG_EPHEMERAL_KEY_LEN];
    490 	uint8_t		wgs_tkey_send[WG_DATA_KEY_LEN];
    491 	uint8_t		wgs_tkey_recv[WG_DATA_KEY_LEN];
    492 };
    493 
    494 struct wg_sockaddr {
    495 	union {
    496 		struct sockaddr_storage _ss;
    497 		struct sockaddr _sa;
    498 		struct sockaddr_in _sin;
    499 		struct sockaddr_in6 _sin6;
    500 	};
    501 	struct psref_target	wgsa_psref;
    502 };
    503 
    504 #define wgsatoss(wgsa)		(&(wgsa)->_ss)
    505 #define wgsatosa(wgsa)		(&(wgsa)->_sa)
    506 #define wgsatosin(wgsa)		(&(wgsa)->_sin)
    507 #define wgsatosin6(wgsa)	(&(wgsa)->_sin6)
    508 
    509 #define	wgsa_family(wgsa)	(wgsatosa(wgsa)->sa_family)
    510 
    511 struct wg_peer;
    512 struct wg_allowedip {
    513 	struct radix_node	wga_nodes[2];
    514 	struct wg_sockaddr	_wga_sa_addr;
    515 	struct wg_sockaddr	_wga_sa_mask;
    516 #define wga_sa_addr		_wga_sa_addr._sa
    517 #define wga_sa_mask		_wga_sa_mask._sa
    518 
    519 	int			wga_family;
    520 	uint8_t			wga_cidr;
    521 	union {
    522 		struct in_addr _ip4;
    523 		struct in6_addr _ip6;
    524 	} wga_addr;
    525 #define wga_addr4	wga_addr._ip4
    526 #define wga_addr6	wga_addr._ip6
    527 
    528 	struct wg_peer		*wga_peer;
    529 };
    530 
    531 typedef uint8_t wg_timestamp_t[WG_TIMESTAMP_LEN];
    532 
    533 struct wg_ppsratecheck {
    534 	struct timeval		wgprc_lasttime;
    535 	int			wgprc_curpps;
    536 };
    537 
    538 struct wg_softc;
    539 struct wg_peer {
    540 	struct wg_softc		*wgp_sc;
    541 	char			wgp_name[WG_PEER_NAME_MAXLEN + 1];
    542 	struct pslist_entry	wgp_peerlist_entry;
    543 	pserialize_t		wgp_psz;
    544 	struct psref_target	wgp_psref;
    545 	kmutex_t		*wgp_lock;
    546 	kmutex_t		*wgp_intr_lock;
    547 
    548 	uint8_t	wgp_pubkey[WG_STATIC_KEY_LEN];
    549 	struct wg_sockaddr	*wgp_endpoint;
    550 	struct wg_sockaddr	*wgp_endpoint0;
    551 	volatile unsigned	wgp_endpoint_changing;
    552 	bool			wgp_endpoint_available;
    553 
    554 			/* The preshared key (optional) */
    555 	uint8_t		wgp_psk[WG_PRESHARED_KEY_LEN];
    556 
    557 	struct wg_session	*wgp_session_stable;
    558 	struct wg_session	*wgp_session_unstable;
    559 
    560 	/* first outgoing packet awaiting session initiation */
    561 	struct mbuf		*wgp_pending;
    562 
    563 	/* timestamp in big-endian */
    564 	wg_timestamp_t	wgp_timestamp_latest_init;
    565 
    566 	struct timespec		wgp_last_handshake_time;
    567 
    568 	callout_t		wgp_rekey_timer;
    569 	callout_t		wgp_handshake_timeout_timer;
    570 	callout_t		wgp_session_dtor_timer;
    571 
    572 	time_t			wgp_handshake_start_time;
    573 
    574 	int			wgp_n_allowedips;
    575 	struct wg_allowedip	wgp_allowedips[WG_ALLOWEDIPS];
    576 
    577 	time_t			wgp_latest_cookie_time;
    578 	uint8_t			wgp_latest_cookie[WG_COOKIE_LEN];
    579 	uint8_t			wgp_last_sent_mac1[WG_MAC_LEN];
    580 	bool			wgp_last_sent_mac1_valid;
    581 	uint8_t			wgp_last_sent_cookie[WG_COOKIE_LEN];
    582 	bool			wgp_last_sent_cookie_valid;
    583 
    584 	time_t			wgp_last_msg_received_time[WG_MSG_TYPE_MAX];
    585 
    586 	time_t			wgp_last_genrandval_time;
    587 	uint32_t		wgp_randval;
    588 
    589 	struct wg_ppsratecheck	wgp_ppsratecheck;
    590 
    591 	struct work		wgp_work;
    592 	unsigned int		wgp_tasks;
    593 #define WGP_TASK_SEND_INIT_MESSAGE		__BIT(0)
    594 #define WGP_TASK_RETRY_HANDSHAKE		__BIT(1)
    595 #define WGP_TASK_ESTABLISH_SESSION		__BIT(2)
    596 #define WGP_TASK_ENDPOINT_CHANGED		__BIT(3)
    597 #define WGP_TASK_SEND_KEEPALIVE_MESSAGE		__BIT(4)
    598 #define WGP_TASK_DESTROY_PREV_SESSION		__BIT(5)
    599 };
    600 
    601 struct wg_ops;
    602 
    603 struct wg_softc {
    604 	struct ifnet	wg_if;
    605 	LIST_ENTRY(wg_softc) wg_list;
    606 	kmutex_t	*wg_lock;
    607 	kmutex_t	*wg_intr_lock;
    608 	krwlock_t	*wg_rwlock;
    609 
    610 	uint8_t		wg_privkey[WG_STATIC_KEY_LEN];
    611 	uint8_t		wg_pubkey[WG_STATIC_KEY_LEN];
    612 
    613 	int		wg_npeers;
    614 	struct pslist_head	wg_peers;
    615 	struct thmap	*wg_peers_bypubkey;
    616 	struct thmap	*wg_peers_byname;
    617 	struct thmap	*wg_sessions_byindex;
    618 	uint16_t	wg_listen_port;
    619 
    620 	struct threadpool	*wg_threadpool;
    621 
    622 	struct threadpool_job	wg_job;
    623 	int			wg_upcalls;
    624 #define	WG_UPCALL_INET	__BIT(0)
    625 #define	WG_UPCALL_INET6	__BIT(1)
    626 
    627 #ifdef INET
    628 	struct socket		*wg_so4;
    629 	struct radix_node_head	*wg_rtable_ipv4;
    630 #endif
    631 #ifdef INET6
    632 	struct socket		*wg_so6;
    633 	struct radix_node_head	*wg_rtable_ipv6;
    634 #endif
    635 
    636 	struct wg_ppsratecheck	wg_ppsratecheck;
    637 
    638 	struct wg_ops		*wg_ops;
    639 
    640 #ifdef WG_RUMPKERNEL
    641 	struct wg_user		*wg_user;
    642 #endif
    643 };
    644 
    645 /* [W] 6.1 Preliminaries */
    646 #define WG_REKEY_AFTER_MESSAGES		(1ULL << 60)
    647 #define WG_REJECT_AFTER_MESSAGES	(UINT64_MAX - (1 << 13))
    648 #define WG_REKEY_AFTER_TIME		120
    649 #define WG_REJECT_AFTER_TIME		180
    650 #define WG_REKEY_ATTEMPT_TIME		 90
    651 #define WG_REKEY_TIMEOUT		  5
    652 #define WG_KEEPALIVE_TIMEOUT		 10
    653 
    654 #define WG_COOKIE_TIME			120
    655 #define WG_RANDVAL_TIME			(2 * 60)
    656 
    657 static uint64_t wg_rekey_after_messages = WG_REKEY_AFTER_MESSAGES;
    658 static uint64_t wg_reject_after_messages = WG_REJECT_AFTER_MESSAGES;
    659 static unsigned wg_rekey_after_time = WG_REKEY_AFTER_TIME;
    660 static unsigned wg_reject_after_time = WG_REJECT_AFTER_TIME;
    661 static unsigned wg_rekey_attempt_time = WG_REKEY_ATTEMPT_TIME;
    662 static unsigned wg_rekey_timeout = WG_REKEY_TIMEOUT;
    663 static unsigned wg_keepalive_timeout = WG_KEEPALIVE_TIMEOUT;
    664 
    665 static struct mbuf *
    666 		wg_get_mbuf(size_t, size_t);
    667 
    668 static int	wg_send_data_msg(struct wg_peer *, struct wg_session *,
    669 		    struct mbuf *);
    670 static int	wg_send_cookie_msg(struct wg_softc *, struct wg_peer *,
    671 		    const uint32_t, const uint8_t [WG_MAC_LEN],
    672 		    const struct sockaddr *);
    673 static int	wg_send_handshake_msg_resp(struct wg_softc *, struct wg_peer *,
    674 		    struct wg_session *, const struct wg_msg_init *);
    675 static void	wg_send_keepalive_msg(struct wg_peer *, struct wg_session *);
    676 
    677 static struct wg_peer *
    678 		wg_pick_peer_by_sa(struct wg_softc *, const struct sockaddr *,
    679 		    struct psref *);
    680 static struct wg_peer *
    681 		wg_lookup_peer_by_pubkey(struct wg_softc *,
    682 		    const uint8_t [WG_STATIC_KEY_LEN], struct psref *);
    683 
    684 static struct wg_session *
    685 		wg_lookup_session_by_index(struct wg_softc *,
    686 		    const uint32_t, struct psref *);
    687 
    688 static void	wg_update_endpoint_if_necessary(struct wg_peer *,
    689 		    const struct sockaddr *);
    690 
    691 static void	wg_schedule_rekey_timer(struct wg_peer *);
    692 static void	wg_schedule_session_dtor_timer(struct wg_peer *);
    693 
    694 static bool	wg_is_underload(struct wg_softc *, struct wg_peer *, int);
    695 static void	wg_calculate_keys(struct wg_session *, const bool);
    696 
    697 static void	wg_clear_states(struct wg_session *);
    698 
    699 static void	wg_get_peer(struct wg_peer *, struct psref *);
    700 static void	wg_put_peer(struct wg_peer *, struct psref *);
    701 
    702 static int	wg_send_so(struct wg_peer *, struct mbuf *);
    703 static int	wg_send_udp(struct wg_peer *, struct mbuf *);
    704 static int	wg_output(struct ifnet *, struct mbuf *,
    705 			   const struct sockaddr *, const struct rtentry *);
    706 static void	wg_input(struct ifnet *, struct mbuf *, const int);
    707 static int	wg_ioctl(struct ifnet *, u_long, void *);
    708 static int	wg_bind_port(struct wg_softc *, const uint16_t);
    709 static int	wg_init(struct ifnet *);
    710 #ifdef ALTQ
    711 static void	wg_start(struct ifnet *);
    712 #endif
    713 static void	wg_stop(struct ifnet *, int);
    714 
    715 static void	wg_peer_work(struct work *, void *);
    716 static void	wg_job(struct threadpool_job *);
    717 static void	wgintr(void *);
    718 static void	wg_purge_pending_packets(struct wg_peer *);
    719 
    720 static int	wg_clone_create(struct if_clone *, int);
    721 static int	wg_clone_destroy(struct ifnet *);
    722 
    723 struct wg_ops {
    724 	int (*send_hs_msg)(struct wg_peer *, struct mbuf *);
    725 	int (*send_data_msg)(struct wg_peer *, struct mbuf *);
    726 	void (*input)(struct ifnet *, struct mbuf *, const int);
    727 	int (*bind_port)(struct wg_softc *, const uint16_t);
    728 };
    729 
    730 struct wg_ops wg_ops_rumpkernel = {
    731 	.send_hs_msg	= wg_send_so,
    732 	.send_data_msg	= wg_send_udp,
    733 	.input		= wg_input,
    734 	.bind_port	= wg_bind_port,
    735 };
    736 
    737 #ifdef WG_RUMPKERNEL
    738 static bool	wg_user_mode(struct wg_softc *);
    739 static int	wg_ioctl_linkstr(struct wg_softc *, struct ifdrv *);
    740 
    741 static int	wg_send_user(struct wg_peer *, struct mbuf *);
    742 static void	wg_input_user(struct ifnet *, struct mbuf *, const int);
    743 static int	wg_bind_port_user(struct wg_softc *, const uint16_t);
    744 
    745 struct wg_ops wg_ops_rumpuser = {
    746 	.send_hs_msg	= wg_send_user,
    747 	.send_data_msg	= wg_send_user,
    748 	.input		= wg_input_user,
    749 	.bind_port	= wg_bind_port_user,
    750 };
    751 #endif
    752 
    753 #define WG_PEER_READER_FOREACH(wgp, wg)					\
    754 	PSLIST_READER_FOREACH((wgp), &(wg)->wg_peers, struct wg_peer,	\
    755 	    wgp_peerlist_entry)
    756 #define WG_PEER_WRITER_FOREACH(wgp, wg)					\
    757 	PSLIST_WRITER_FOREACH((wgp), &(wg)->wg_peers, struct wg_peer,	\
    758 	    wgp_peerlist_entry)
    759 #define WG_PEER_WRITER_INSERT_HEAD(wgp, wg)				\
    760 	PSLIST_WRITER_INSERT_HEAD(&(wg)->wg_peers, (wgp), wgp_peerlist_entry)
    761 #define WG_PEER_WRITER_REMOVE(wgp)					\
    762 	PSLIST_WRITER_REMOVE((wgp), wgp_peerlist_entry)
    763 
    764 struct wg_route {
    765 	struct radix_node	wgr_nodes[2];
    766 	struct wg_peer		*wgr_peer;
    767 };
    768 
    769 static struct radix_node_head *
    770 wg_rnh(struct wg_softc *wg, const int family)
    771 {
    772 
    773 	switch (family) {
    774 		case AF_INET:
    775 			return wg->wg_rtable_ipv4;
    776 #ifdef INET6
    777 		case AF_INET6:
    778 			return wg->wg_rtable_ipv6;
    779 #endif
    780 		default:
    781 			return NULL;
    782 	}
    783 }
    784 
    785 
    786 /*
    787  * Global variables
    788  */
    789 static volatile unsigned wg_count __cacheline_aligned;
    790 
    791 struct psref_class *wg_psref_class __read_mostly;
    792 
    793 static struct if_clone wg_cloner =
    794     IF_CLONE_INITIALIZER("wg", wg_clone_create, wg_clone_destroy);
    795 
    796 static struct pktqueue *wg_pktq __read_mostly;
    797 static struct workqueue *wg_wq __read_mostly;
    798 
    799 void wgattach(int);
    800 /* ARGSUSED */
    801 void
    802 wgattach(int count)
    803 {
    804 	/*
    805 	 * Nothing to do here, initialization is handled by the
    806 	 * module initialization code in wginit() below).
    807 	 */
    808 }
    809 
    810 static void
    811 wginit(void)
    812 {
    813 
    814 	wg_psref_class = psref_class_create("wg", IPL_SOFTNET);
    815 
    816 	if_clone_attach(&wg_cloner);
    817 }
    818 
    819 /*
    820  * XXX Kludge: This should just happen in wginit, but workqueue_create
    821  * cannot be run until after CPUs have been detected, and wginit runs
    822  * before configure.
    823  */
    824 static int
    825 wginitqueues(void)
    826 {
    827 	int error __diagused;
    828 
    829 	wg_pktq = pktq_create(IFQ_MAXLEN, wgintr, NULL);
    830 	KASSERT(wg_pktq != NULL);
    831 
    832 	error = workqueue_create(&wg_wq, "wgpeer", wg_peer_work, NULL,
    833 	    PRI_NONE, IPL_SOFTNET, WQ_MPSAFE|WQ_PERCPU);
    834 	KASSERT(error == 0);
    835 
    836 	return 0;
    837 }
    838 
    839 static void
    840 wg_guarantee_initialized(void)
    841 {
    842 	static ONCE_DECL(init);
    843 	int error __diagused;
    844 
    845 	error = RUN_ONCE(&init, wginitqueues);
    846 	KASSERT(error == 0);
    847 }
    848 
    849 static int
    850 wg_count_inc(void)
    851 {
    852 	unsigned o, n;
    853 
    854 	do {
    855 		o = atomic_load_relaxed(&wg_count);
    856 		if (o == UINT_MAX)
    857 			return ENFILE;
    858 		n = o + 1;
    859 	} while (atomic_cas_uint(&wg_count, o, n) != o);
    860 
    861 	return 0;
    862 }
    863 
    864 static void
    865 wg_count_dec(void)
    866 {
    867 	unsigned c __diagused;
    868 
    869 	c = atomic_dec_uint_nv(&wg_count);
    870 	KASSERT(c != UINT_MAX);
    871 }
    872 
    873 static int
    874 wgdetach(void)
    875 {
    876 
    877 	/* Prevent new interface creation.  */
    878 	if_clone_detach(&wg_cloner);
    879 
    880 	/* Check whether there are any existing interfaces.  */
    881 	if (atomic_load_relaxed(&wg_count)) {
    882 		/* Back out -- reattach the cloner.  */
    883 		if_clone_attach(&wg_cloner);
    884 		return EBUSY;
    885 	}
    886 
    887 	/* No interfaces left.  Nuke it.  */
    888 	workqueue_destroy(wg_wq);
    889 	pktq_destroy(wg_pktq);
    890 	psref_class_destroy(wg_psref_class);
    891 
    892 	return 0;
    893 }
    894 
    895 static void
    896 wg_init_key_and_hash(uint8_t ckey[WG_CHAINING_KEY_LEN],
    897     uint8_t hash[WG_HASH_LEN])
    898 {
    899 	/* [W] 5.4: CONSTRUCTION */
    900 	const char *signature = "Noise_IKpsk2_25519_ChaChaPoly_BLAKE2s";
    901 	/* [W] 5.4: IDENTIFIER */
    902 	const char *id = "WireGuard v1 zx2c4 Jason (at) zx2c4.com";
    903 	struct blake2s state;
    904 
    905 	blake2s(ckey, WG_CHAINING_KEY_LEN, NULL, 0,
    906 	    signature, strlen(signature));
    907 
    908 	CTASSERT(WG_HASH_LEN == WG_CHAINING_KEY_LEN);
    909 	memcpy(hash, ckey, WG_CHAINING_KEY_LEN);
    910 
    911 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
    912 	blake2s_update(&state, ckey, WG_CHAINING_KEY_LEN);
    913 	blake2s_update(&state, id, strlen(id));
    914 	blake2s_final(&state, hash);
    915 
    916 	WG_DUMP_HASH("ckey", ckey);
    917 	WG_DUMP_HASH("hash", hash);
    918 }
    919 
    920 static void
    921 wg_algo_hash(uint8_t hash[WG_HASH_LEN], const uint8_t input[],
    922     const size_t inputsize)
    923 {
    924 	struct blake2s state;
    925 
    926 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
    927 	blake2s_update(&state, hash, WG_HASH_LEN);
    928 	blake2s_update(&state, input, inputsize);
    929 	blake2s_final(&state, hash);
    930 }
    931 
    932 static void
    933 wg_algo_mac(uint8_t out[], const size_t outsize,
    934     const uint8_t key[], const size_t keylen,
    935     const uint8_t input1[], const size_t input1len,
    936     const uint8_t input2[], const size_t input2len)
    937 {
    938 	struct blake2s state;
    939 
    940 	blake2s_init(&state, outsize, key, keylen);
    941 
    942 	blake2s_update(&state, input1, input1len);
    943 	if (input2 != NULL)
    944 		blake2s_update(&state, input2, input2len);
    945 	blake2s_final(&state, out);
    946 }
    947 
    948 static void
    949 wg_algo_mac_mac1(uint8_t out[], const size_t outsize,
    950     const uint8_t input1[], const size_t input1len,
    951     const uint8_t input2[], const size_t input2len)
    952 {
    953 	struct blake2s state;
    954 	/* [W] 5.4: LABEL-MAC1 */
    955 	const char *label = "mac1----";
    956 	uint8_t key[WG_HASH_LEN];
    957 
    958 	blake2s_init(&state, sizeof(key), NULL, 0);
    959 	blake2s_update(&state, label, strlen(label));
    960 	blake2s_update(&state, input1, input1len);
    961 	blake2s_final(&state, key);
    962 
    963 	blake2s_init(&state, outsize, key, sizeof(key));
    964 	if (input2 != NULL)
    965 		blake2s_update(&state, input2, input2len);
    966 	blake2s_final(&state, out);
    967 }
    968 
    969 static void
    970 wg_algo_mac_cookie(uint8_t out[], const size_t outsize,
    971     const uint8_t input1[], const size_t input1len)
    972 {
    973 	struct blake2s state;
    974 	/* [W] 5.4: LABEL-COOKIE */
    975 	const char *label = "cookie--";
    976 
    977 	blake2s_init(&state, outsize, NULL, 0);
    978 	blake2s_update(&state, label, strlen(label));
    979 	blake2s_update(&state, input1, input1len);
    980 	blake2s_final(&state, out);
    981 }
    982 
    983 static void
    984 wg_algo_generate_keypair(uint8_t pubkey[WG_EPHEMERAL_KEY_LEN],
    985     uint8_t privkey[WG_EPHEMERAL_KEY_LEN])
    986 {
    987 
    988 	CTASSERT(WG_EPHEMERAL_KEY_LEN == crypto_scalarmult_curve25519_BYTES);
    989 
    990 	cprng_strong(kern_cprng, privkey, WG_EPHEMERAL_KEY_LEN, 0);
    991 	crypto_scalarmult_base(pubkey, privkey);
    992 }
    993 
    994 static void
    995 wg_algo_dh(uint8_t out[WG_DH_OUTPUT_LEN],
    996     const uint8_t privkey[WG_STATIC_KEY_LEN],
    997     const uint8_t pubkey[WG_STATIC_KEY_LEN])
    998 {
    999 
   1000 	CTASSERT(WG_STATIC_KEY_LEN == crypto_scalarmult_curve25519_BYTES);
   1001 
   1002 	int ret __diagused = crypto_scalarmult(out, privkey, pubkey);
   1003 	KASSERT(ret == 0);
   1004 }
   1005 
   1006 static void
   1007 wg_algo_hmac(uint8_t out[], const size_t outlen,
   1008     const uint8_t key[], const size_t keylen,
   1009     const uint8_t in[], const size_t inlen)
   1010 {
   1011 #define IPAD	0x36
   1012 #define OPAD	0x5c
   1013 	uint8_t hmackey[HMAC_BLOCK_LEN] = {0};
   1014 	uint8_t ipad[HMAC_BLOCK_LEN];
   1015 	uint8_t opad[HMAC_BLOCK_LEN];
   1016 	size_t i;
   1017 	struct blake2s state;
   1018 
   1019 	KASSERT(outlen == WG_HASH_LEN);
   1020 	KASSERT(keylen <= HMAC_BLOCK_LEN);
   1021 
   1022 	memcpy(hmackey, key, keylen);
   1023 
   1024 	for (i = 0; i < sizeof(hmackey); i++) {
   1025 		ipad[i] = hmackey[i] ^ IPAD;
   1026 		opad[i] = hmackey[i] ^ OPAD;
   1027 	}
   1028 
   1029 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
   1030 	blake2s_update(&state, ipad, sizeof(ipad));
   1031 	blake2s_update(&state, in, inlen);
   1032 	blake2s_final(&state, out);
   1033 
   1034 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
   1035 	blake2s_update(&state, opad, sizeof(opad));
   1036 	blake2s_update(&state, out, WG_HASH_LEN);
   1037 	blake2s_final(&state, out);
   1038 #undef IPAD
   1039 #undef OPAD
   1040 }
   1041 
   1042 static void
   1043 wg_algo_kdf(uint8_t out1[WG_KDF_OUTPUT_LEN], uint8_t out2[WG_KDF_OUTPUT_LEN],
   1044     uint8_t out3[WG_KDF_OUTPUT_LEN], const uint8_t ckey[WG_CHAINING_KEY_LEN],
   1045     const uint8_t input[], const size_t inputlen)
   1046 {
   1047 	uint8_t tmp1[WG_KDF_OUTPUT_LEN], tmp2[WG_KDF_OUTPUT_LEN + 1];
   1048 	uint8_t one[1];
   1049 
   1050 	/*
   1051 	 * [N] 4.3: "an input_key_material byte sequence with length
   1052 	 * either zero bytes, 32 bytes, or DHLEN bytes."
   1053 	 */
   1054 	KASSERT(inputlen == 0 || inputlen == 32 || inputlen == NOISE_DHLEN);
   1055 
   1056 	WG_DUMP_HASH("ckey", ckey);
   1057 	if (input != NULL)
   1058 		WG_DUMP_HASH("input", input);
   1059 	wg_algo_hmac(tmp1, sizeof(tmp1), ckey, WG_CHAINING_KEY_LEN,
   1060 	    input, inputlen);
   1061 	WG_DUMP_HASH("tmp1", tmp1);
   1062 	one[0] = 1;
   1063 	wg_algo_hmac(out1, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
   1064 	    one, sizeof(one));
   1065 	WG_DUMP_HASH("out1", out1);
   1066 	if (out2 == NULL)
   1067 		return;
   1068 	memcpy(tmp2, out1, WG_KDF_OUTPUT_LEN);
   1069 	tmp2[WG_KDF_OUTPUT_LEN] = 2;
   1070 	wg_algo_hmac(out2, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
   1071 	    tmp2, sizeof(tmp2));
   1072 	WG_DUMP_HASH("out2", out2);
   1073 	if (out3 == NULL)
   1074 		return;
   1075 	memcpy(tmp2, out2, WG_KDF_OUTPUT_LEN);
   1076 	tmp2[WG_KDF_OUTPUT_LEN] = 3;
   1077 	wg_algo_hmac(out3, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
   1078 	    tmp2, sizeof(tmp2));
   1079 	WG_DUMP_HASH("out3", out3);
   1080 }
   1081 
   1082 static void __noinline
   1083 wg_algo_dh_kdf(uint8_t ckey[WG_CHAINING_KEY_LEN],
   1084     uint8_t cipher_key[WG_CIPHER_KEY_LEN],
   1085     const uint8_t local_key[WG_STATIC_KEY_LEN],
   1086     const uint8_t remote_key[WG_STATIC_KEY_LEN])
   1087 {
   1088 	uint8_t dhout[WG_DH_OUTPUT_LEN];
   1089 
   1090 	wg_algo_dh(dhout, local_key, remote_key);
   1091 	wg_algo_kdf(ckey, cipher_key, NULL, ckey, dhout, sizeof(dhout));
   1092 
   1093 	WG_DUMP_HASH("dhout", dhout);
   1094 	WG_DUMP_HASH("ckey", ckey);
   1095 	if (cipher_key != NULL)
   1096 		WG_DUMP_HASH("cipher_key", cipher_key);
   1097 }
   1098 
   1099 static void
   1100 wg_algo_aead_enc(uint8_t out[], size_t expected_outsize, const uint8_t key[],
   1101     const uint64_t counter, const uint8_t plain[], const size_t plainsize,
   1102     const uint8_t auth[], size_t authlen)
   1103 {
   1104 	uint8_t nonce[(32 + 64) / 8] = {0};
   1105 	long long unsigned int outsize;
   1106 	int error __diagused;
   1107 
   1108 	le64enc(&nonce[4], counter);
   1109 
   1110 	error = crypto_aead_chacha20poly1305_ietf_encrypt(out, &outsize, plain,
   1111 	    plainsize, auth, authlen, NULL, nonce, key);
   1112 	KASSERT(error == 0);
   1113 	KASSERT(outsize == expected_outsize);
   1114 }
   1115 
   1116 static int
   1117 wg_algo_aead_dec(uint8_t out[], size_t expected_outsize, const uint8_t key[],
   1118     const uint64_t counter, const uint8_t encrypted[],
   1119     const size_t encryptedsize, const uint8_t auth[], size_t authlen)
   1120 {
   1121 	uint8_t nonce[(32 + 64) / 8] = {0};
   1122 	long long unsigned int outsize;
   1123 	int error;
   1124 
   1125 	le64enc(&nonce[4], counter);
   1126 
   1127 	error = crypto_aead_chacha20poly1305_ietf_decrypt(out, &outsize, NULL,
   1128 	    encrypted, encryptedsize, auth, authlen, nonce, key);
   1129 	if (error == 0)
   1130 		KASSERT(outsize == expected_outsize);
   1131 	return error;
   1132 }
   1133 
   1134 static void
   1135 wg_algo_xaead_enc(uint8_t out[], const size_t expected_outsize,
   1136     const uint8_t key[], const uint8_t plain[], const size_t plainsize,
   1137     const uint8_t auth[], size_t authlen,
   1138     const uint8_t nonce[WG_SALT_LEN])
   1139 {
   1140 	long long unsigned int outsize;
   1141 	int error __diagused;
   1142 
   1143 	CTASSERT(WG_SALT_LEN == crypto_aead_xchacha20poly1305_ietf_NPUBBYTES);
   1144 	error = crypto_aead_xchacha20poly1305_ietf_encrypt(out, &outsize,
   1145 	    plain, plainsize, auth, authlen, NULL, nonce, key);
   1146 	KASSERT(error == 0);
   1147 	KASSERT(outsize == expected_outsize);
   1148 }
   1149 
   1150 static int
   1151 wg_algo_xaead_dec(uint8_t out[], const size_t expected_outsize,
   1152     const uint8_t key[], const uint8_t encrypted[], const size_t encryptedsize,
   1153     const uint8_t auth[], size_t authlen,
   1154     const uint8_t nonce[WG_SALT_LEN])
   1155 {
   1156 	long long unsigned int outsize;
   1157 	int error;
   1158 
   1159 	error = crypto_aead_xchacha20poly1305_ietf_decrypt(out, &outsize, NULL,
   1160 	    encrypted, encryptedsize, auth, authlen, nonce, key);
   1161 	if (error == 0)
   1162 		KASSERT(outsize == expected_outsize);
   1163 	return error;
   1164 }
   1165 
   1166 static void
   1167 wg_algo_tai64n(wg_timestamp_t timestamp)
   1168 {
   1169 	struct timespec ts;
   1170 
   1171 	/* FIXME strict TAI64N (https://cr.yp.to/libtai/tai64.html) */
   1172 	getnanotime(&ts);
   1173 	/* TAI64 label in external TAI64 format */
   1174 	be32enc(timestamp, 0x40000000U + (uint32_t)(ts.tv_sec >> 32));
   1175 	/* second beginning from 1970 TAI */
   1176 	be32enc(timestamp + 4, (uint32_t)(ts.tv_sec & 0xffffffffU));
   1177 	/* nanosecond in big-endian format */
   1178 	be32enc(timestamp + 8, (uint32_t)ts.tv_nsec);
   1179 }
   1180 
   1181 /*
   1182  * wg_get_stable_session(wgp, psref)
   1183  *
   1184  *	Get a passive reference to the current stable session, or
   1185  *	return NULL if there is no current stable session.
   1186  *
   1187  *	The pointer is always there but the session is not necessarily
   1188  *	ESTABLISHED; if it is not ESTABLISHED, return NULL.  However,
   1189  *	the session may transition from ESTABLISHED to DESTROYING while
   1190  *	holding the passive reference.
   1191  */
   1192 static struct wg_session *
   1193 wg_get_stable_session(struct wg_peer *wgp, struct psref *psref)
   1194 {
   1195 	int s;
   1196 	struct wg_session *wgs;
   1197 
   1198 	s = pserialize_read_enter();
   1199 	wgs = atomic_load_consume(&wgp->wgp_session_stable);
   1200 	if (__predict_false(wgs->wgs_state != WGS_STATE_ESTABLISHED))
   1201 		wgs = NULL;
   1202 	else
   1203 		psref_acquire(psref, &wgs->wgs_psref, wg_psref_class);
   1204 	pserialize_read_exit(s);
   1205 
   1206 	return wgs;
   1207 }
   1208 
   1209 static void
   1210 wg_put_session(struct wg_session *wgs, struct psref *psref)
   1211 {
   1212 
   1213 	psref_release(psref, &wgs->wgs_psref, wg_psref_class);
   1214 }
   1215 
   1216 static void
   1217 wg_destroy_session(struct wg_softc *wg, struct wg_session *wgs)
   1218 {
   1219 	struct wg_peer *wgp = wgs->wgs_peer;
   1220 	struct wg_session *wgs0 __diagused;
   1221 	void *garbage;
   1222 
   1223 	KASSERT(mutex_owned(wgp->wgp_lock));
   1224 	KASSERT(wgs->wgs_state != WGS_STATE_UNKNOWN);
   1225 
   1226 	/* Remove the session from the table.  */
   1227 	wgs0 = thmap_del(wg->wg_sessions_byindex,
   1228 	    &wgs->wgs_local_index, sizeof(wgs->wgs_local_index));
   1229 	KASSERT(wgs0 == wgs);
   1230 	garbage = thmap_stage_gc(wg->wg_sessions_byindex);
   1231 
   1232 	/* Wait for passive references to drain.  */
   1233 	pserialize_perform(wgp->wgp_psz);
   1234 	psref_target_destroy(&wgs->wgs_psref, wg_psref_class);
   1235 
   1236 	/* Free memory, zero state, and transition to UNKNOWN.  */
   1237 	thmap_gc(wg->wg_sessions_byindex, garbage);
   1238 	wg_clear_states(wgs);
   1239 	wgs->wgs_state = WGS_STATE_UNKNOWN;
   1240 }
   1241 
   1242 /*
   1243  * wg_get_session_index(wg, wgs)
   1244  *
   1245  *	Choose a session index for wgs->wgs_local_index, and store it
   1246  *	in wg's table of sessions by index.
   1247  *
   1248  *	wgs must be the unstable session of its peer, and must be
   1249  *	transitioning out of the UNKNOWN state.
   1250  */
   1251 static void
   1252 wg_get_session_index(struct wg_softc *wg, struct wg_session *wgs)
   1253 {
   1254 	struct wg_peer *wgp __diagused = wgs->wgs_peer;
   1255 	struct wg_session *wgs0;
   1256 	uint32_t index;
   1257 
   1258 	KASSERT(mutex_owned(wgp->wgp_lock));
   1259 	KASSERT(wgs == wgp->wgp_session_unstable);
   1260 	KASSERT(wgs->wgs_state == WGS_STATE_UNKNOWN);
   1261 
   1262 	do {
   1263 		/* Pick a uniform random index.  */
   1264 		index = cprng_strong32();
   1265 
   1266 		/* Try to take it.  */
   1267 		wgs->wgs_local_index = index;
   1268 		wgs0 = thmap_put(wg->wg_sessions_byindex,
   1269 		    &wgs->wgs_local_index, sizeof wgs->wgs_local_index, wgs);
   1270 
   1271 		/* If someone else beat us, start over.  */
   1272 	} while (__predict_false(wgs0 != wgs));
   1273 }
   1274 
   1275 /*
   1276  * wg_put_session_index(wg, wgs)
   1277  *
   1278  *	Remove wgs from the table of sessions by index, wait for any
   1279  *	passive references to drain, and transition the session to the
   1280  *	UNKNOWN state.
   1281  *
   1282  *	wgs must be the unstable session of its peer, and must not be
   1283  *	UNKNOWN or ESTABLISHED.
   1284  */
   1285 static void
   1286 wg_put_session_index(struct wg_softc *wg, struct wg_session *wgs)
   1287 {
   1288 	struct wg_peer *wgp __diagused = wgs->wgs_peer;
   1289 
   1290 	KASSERT(mutex_owned(wgp->wgp_lock));
   1291 	KASSERT(wgs == wgp->wgp_session_unstable);
   1292 	KASSERT(wgs->wgs_state != WGS_STATE_UNKNOWN);
   1293 	KASSERT(wgs->wgs_state != WGS_STATE_ESTABLISHED);
   1294 
   1295 	wg_destroy_session(wg, wgs);
   1296 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
   1297 }
   1298 
   1299 /*
   1300  * Handshake patterns
   1301  *
   1302  * [W] 5: "These messages use the "IK" pattern from Noise"
   1303  * [N] 7.5. Interactive handshake patterns (fundamental)
   1304  *     "The first character refers to the initiators static key:"
   1305  *     "I = Static key for initiator Immediately transmitted to responder,
   1306  *          despite reduced or absent identity hiding"
   1307  *     "The second character refers to the responders static key:"
   1308  *     "K = Static key for responder Known to initiator"
   1309  *     "IK:
   1310  *        <- s
   1311  *        ...
   1312  *        -> e, es, s, ss
   1313  *        <- e, ee, se"
   1314  * [N] 9.4. Pattern modifiers
   1315  *     "IKpsk2:
   1316  *        <- s
   1317  *        ...
   1318  *        -> e, es, s, ss
   1319  *        <- e, ee, se, psk"
   1320  */
   1321 static void
   1322 wg_fill_msg_init(struct wg_softc *wg, struct wg_peer *wgp,
   1323     struct wg_session *wgs, struct wg_msg_init *wgmi)
   1324 {
   1325 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.2: Ci */
   1326 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.2: Hi */
   1327 	uint8_t cipher_key[WG_CIPHER_KEY_LEN];
   1328 	uint8_t pubkey[WG_EPHEMERAL_KEY_LEN];
   1329 	uint8_t privkey[WG_EPHEMERAL_KEY_LEN];
   1330 
   1331 	KASSERT(mutex_owned(wgp->wgp_lock));
   1332 	KASSERT(wgs == wgp->wgp_session_unstable);
   1333 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_ACTIVE);
   1334 
   1335 	wgmi->wgmi_type = htole32(WG_MSG_TYPE_INIT);
   1336 	wgmi->wgmi_sender = wgs->wgs_local_index;
   1337 
   1338 	/* [W] 5.4.2: First Message: Initiator to Responder */
   1339 
   1340 	/* Ci := HASH(CONSTRUCTION) */
   1341 	/* Hi := HASH(Ci || IDENTIFIER) */
   1342 	wg_init_key_and_hash(ckey, hash);
   1343 	/* Hi := HASH(Hi || Sr^pub) */
   1344 	wg_algo_hash(hash, wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey));
   1345 
   1346 	WG_DUMP_HASH("hash", hash);
   1347 
   1348 	/* [N] 2.2: "e" */
   1349 	/* Ei^priv, Ei^pub := DH-GENERATE() */
   1350 	wg_algo_generate_keypair(pubkey, privkey);
   1351 	/* Ci := KDF1(Ci, Ei^pub) */
   1352 	wg_algo_kdf(ckey, NULL, NULL, ckey, pubkey, sizeof(pubkey));
   1353 	/* msg.ephemeral := Ei^pub */
   1354 	memcpy(wgmi->wgmi_ephemeral, pubkey, sizeof(wgmi->wgmi_ephemeral));
   1355 	/* Hi := HASH(Hi || msg.ephemeral) */
   1356 	wg_algo_hash(hash, pubkey, sizeof(pubkey));
   1357 
   1358 	WG_DUMP_HASH("ckey", ckey);
   1359 	WG_DUMP_HASH("hash", hash);
   1360 
   1361 	/* [N] 2.2: "es" */
   1362 	/* Ci, k := KDF2(Ci, DH(Ei^priv, Sr^pub)) */
   1363 	wg_algo_dh_kdf(ckey, cipher_key, privkey, wgp->wgp_pubkey);
   1364 
   1365 	/* [N] 2.2: "s" */
   1366 	/* msg.static := AEAD(k, 0, Si^pub, Hi) */
   1367 	wg_algo_aead_enc(wgmi->wgmi_static, sizeof(wgmi->wgmi_static),
   1368 	    cipher_key, 0, wg->wg_pubkey, sizeof(wg->wg_pubkey),
   1369 	    hash, sizeof(hash));
   1370 	/* Hi := HASH(Hi || msg.static) */
   1371 	wg_algo_hash(hash, wgmi->wgmi_static, sizeof(wgmi->wgmi_static));
   1372 
   1373 	WG_DUMP_HASH48("wgmi_static", wgmi->wgmi_static);
   1374 
   1375 	/* [N] 2.2: "ss" */
   1376 	/* Ci, k := KDF2(Ci, DH(Si^priv, Sr^pub)) */
   1377 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgp->wgp_pubkey);
   1378 
   1379 	/* msg.timestamp := AEAD(k, TIMESTAMP(), Hi) */
   1380 	wg_timestamp_t timestamp;
   1381 	wg_algo_tai64n(timestamp);
   1382 	wg_algo_aead_enc(wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp),
   1383 	    cipher_key, 0, timestamp, sizeof(timestamp), hash, sizeof(hash));
   1384 	/* Hi := HASH(Hi || msg.timestamp) */
   1385 	wg_algo_hash(hash, wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp));
   1386 
   1387 	/* [W] 5.4.4 Cookie MACs */
   1388 	wg_algo_mac_mac1(wgmi->wgmi_mac1, sizeof(wgmi->wgmi_mac1),
   1389 	    wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey),
   1390 	    (const uint8_t *)wgmi, offsetof(struct wg_msg_init, wgmi_mac1));
   1391 	/* Need mac1 to decrypt a cookie from a cookie message */
   1392 	memcpy(wgp->wgp_last_sent_mac1, wgmi->wgmi_mac1,
   1393 	    sizeof(wgp->wgp_last_sent_mac1));
   1394 	wgp->wgp_last_sent_mac1_valid = true;
   1395 
   1396 	if (wgp->wgp_latest_cookie_time == 0 ||
   1397 	    (time_uptime - wgp->wgp_latest_cookie_time) >= WG_COOKIE_TIME)
   1398 		memset(wgmi->wgmi_mac2, 0, sizeof(wgmi->wgmi_mac2));
   1399 	else {
   1400 		wg_algo_mac(wgmi->wgmi_mac2, sizeof(wgmi->wgmi_mac2),
   1401 		    wgp->wgp_latest_cookie, WG_COOKIE_LEN,
   1402 		    (const uint8_t *)wgmi,
   1403 		    offsetof(struct wg_msg_init, wgmi_mac2),
   1404 		    NULL, 0);
   1405 	}
   1406 
   1407 	memcpy(wgs->wgs_ephemeral_key_pub, pubkey, sizeof(pubkey));
   1408 	memcpy(wgs->wgs_ephemeral_key_priv, privkey, sizeof(privkey));
   1409 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
   1410 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
   1411 	WG_DLOG("%s: sender=%x\n", __func__, wgs->wgs_local_index);
   1412 }
   1413 
   1414 static void __noinline
   1415 wg_handle_msg_init(struct wg_softc *wg, const struct wg_msg_init *wgmi,
   1416     const struct sockaddr *src)
   1417 {
   1418 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.2: Ci */
   1419 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.2: Hi */
   1420 	uint8_t cipher_key[WG_CIPHER_KEY_LEN];
   1421 	uint8_t peer_pubkey[WG_STATIC_KEY_LEN];
   1422 	struct wg_peer *wgp;
   1423 	struct wg_session *wgs;
   1424 	int error, ret;
   1425 	struct psref psref_peer;
   1426 	uint8_t mac1[WG_MAC_LEN];
   1427 
   1428 	WG_TRACE("init msg received");
   1429 
   1430 	wg_algo_mac_mac1(mac1, sizeof(mac1),
   1431 	    wg->wg_pubkey, sizeof(wg->wg_pubkey),
   1432 	    (const uint8_t *)wgmi, offsetof(struct wg_msg_init, wgmi_mac1));
   1433 
   1434 	/*
   1435 	 * [W] 5.3: Denial of Service Mitigation & Cookies
   1436 	 * "the responder, ..., must always reject messages with an invalid
   1437 	 *  msg.mac1"
   1438 	 */
   1439 	if (!consttime_memequal(mac1, wgmi->wgmi_mac1, sizeof(mac1))) {
   1440 		WG_DLOG("mac1 is invalid\n");
   1441 		return;
   1442 	}
   1443 
   1444 	/*
   1445 	 * [W] 5.4.2: First Message: Initiator to Responder
   1446 	 * "When the responder receives this message, it does the same
   1447 	 *  operations so that its final state variables are identical,
   1448 	 *  replacing the operands of the DH function to produce equivalent
   1449 	 *  values."
   1450 	 *  Note that the following comments of operations are just copies of
   1451 	 *  the initiator's ones.
   1452 	 */
   1453 
   1454 	/* Ci := HASH(CONSTRUCTION) */
   1455 	/* Hi := HASH(Ci || IDENTIFIER) */
   1456 	wg_init_key_and_hash(ckey, hash);
   1457 	/* Hi := HASH(Hi || Sr^pub) */
   1458 	wg_algo_hash(hash, wg->wg_pubkey, sizeof(wg->wg_pubkey));
   1459 
   1460 	/* [N] 2.2: "e" */
   1461 	/* Ci := KDF1(Ci, Ei^pub) */
   1462 	wg_algo_kdf(ckey, NULL, NULL, ckey, wgmi->wgmi_ephemeral,
   1463 	    sizeof(wgmi->wgmi_ephemeral));
   1464 	/* Hi := HASH(Hi || msg.ephemeral) */
   1465 	wg_algo_hash(hash, wgmi->wgmi_ephemeral, sizeof(wgmi->wgmi_ephemeral));
   1466 
   1467 	WG_DUMP_HASH("ckey", ckey);
   1468 
   1469 	/* [N] 2.2: "es" */
   1470 	/* Ci, k := KDF2(Ci, DH(Ei^priv, Sr^pub)) */
   1471 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgmi->wgmi_ephemeral);
   1472 
   1473 	WG_DUMP_HASH48("wgmi_static", wgmi->wgmi_static);
   1474 
   1475 	/* [N] 2.2: "s" */
   1476 	/* msg.static := AEAD(k, 0, Si^pub, Hi) */
   1477 	error = wg_algo_aead_dec(peer_pubkey, WG_STATIC_KEY_LEN, cipher_key, 0,
   1478 	    wgmi->wgmi_static, sizeof(wgmi->wgmi_static), hash, sizeof(hash));
   1479 	if (error != 0) {
   1480 		WG_LOG_RATECHECK(&wg->wg_ppsratecheck, LOG_DEBUG,
   1481 		    "%s: wg_algo_aead_dec for secret key failed\n",
   1482 		    if_name(&wg->wg_if));
   1483 		return;
   1484 	}
   1485 	/* Hi := HASH(Hi || msg.static) */
   1486 	wg_algo_hash(hash, wgmi->wgmi_static, sizeof(wgmi->wgmi_static));
   1487 
   1488 	wgp = wg_lookup_peer_by_pubkey(wg, peer_pubkey, &psref_peer);
   1489 	if (wgp == NULL) {
   1490 		WG_DLOG("peer not found\n");
   1491 		return;
   1492 	}
   1493 
   1494 	/*
   1495 	 * Lock the peer to serialize access to cookie state.
   1496 	 *
   1497 	 * XXX Can we safely avoid holding the lock across DH?  Take it
   1498 	 * just to verify mac2 and then unlock/DH/lock?
   1499 	 */
   1500 	mutex_enter(wgp->wgp_lock);
   1501 
   1502 	if (__predict_false(wg_is_underload(wg, wgp, WG_MSG_TYPE_INIT))) {
   1503 		WG_TRACE("under load");
   1504 		/*
   1505 		 * [W] 5.3: Denial of Service Mitigation & Cookies
   1506 		 * "the responder, ..., and when under load may reject messages
   1507 		 *  with an invalid msg.mac2.  If the responder receives a
   1508 		 *  message with a valid msg.mac1 yet with an invalid msg.mac2,
   1509 		 *  and is under load, it may respond with a cookie reply
   1510 		 *  message"
   1511 		 */
   1512 		uint8_t zero[WG_MAC_LEN] = {0};
   1513 		if (consttime_memequal(wgmi->wgmi_mac2, zero, sizeof(zero))) {
   1514 			WG_TRACE("sending a cookie message: no cookie included");
   1515 			(void)wg_send_cookie_msg(wg, wgp, wgmi->wgmi_sender,
   1516 			    wgmi->wgmi_mac1, src);
   1517 			goto out;
   1518 		}
   1519 		if (!wgp->wgp_last_sent_cookie_valid) {
   1520 			WG_TRACE("sending a cookie message: no cookie sent ever");
   1521 			(void)wg_send_cookie_msg(wg, wgp, wgmi->wgmi_sender,
   1522 			    wgmi->wgmi_mac1, src);
   1523 			goto out;
   1524 		}
   1525 		uint8_t mac2[WG_MAC_LEN];
   1526 		wg_algo_mac(mac2, sizeof(mac2), wgp->wgp_last_sent_cookie,
   1527 		    WG_COOKIE_LEN, (const uint8_t *)wgmi,
   1528 		    offsetof(struct wg_msg_init, wgmi_mac2), NULL, 0);
   1529 		if (!consttime_memequal(mac2, wgmi->wgmi_mac2, sizeof(mac2))) {
   1530 			WG_DLOG("mac2 is invalid\n");
   1531 			goto out;
   1532 		}
   1533 		WG_TRACE("under load, but continue to sending");
   1534 	}
   1535 
   1536 	/* [N] 2.2: "ss" */
   1537 	/* Ci, k := KDF2(Ci, DH(Si^priv, Sr^pub)) */
   1538 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgp->wgp_pubkey);
   1539 
   1540 	/* msg.timestamp := AEAD(k, TIMESTAMP(), Hi) */
   1541 	wg_timestamp_t timestamp;
   1542 	error = wg_algo_aead_dec(timestamp, sizeof(timestamp), cipher_key, 0,
   1543 	    wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp),
   1544 	    hash, sizeof(hash));
   1545 	if (error != 0) {
   1546 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   1547 		    "%s: peer %s: wg_algo_aead_dec for timestamp failed\n",
   1548 		    if_name(&wg->wg_if), wgp->wgp_name);
   1549 		goto out;
   1550 	}
   1551 	/* Hi := HASH(Hi || msg.timestamp) */
   1552 	wg_algo_hash(hash, wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp));
   1553 
   1554 	/*
   1555 	 * [W] 5.1 "The responder keeps track of the greatest timestamp
   1556 	 *      received per peer and discards packets containing
   1557 	 *      timestamps less than or equal to it."
   1558 	 */
   1559 	ret = memcmp(timestamp, wgp->wgp_timestamp_latest_init,
   1560 	    sizeof(timestamp));
   1561 	if (ret <= 0) {
   1562 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   1563 		    "%s: peer %s: invalid init msg: timestamp is old\n",
   1564 		    if_name(&wg->wg_if), wgp->wgp_name);
   1565 		goto out;
   1566 	}
   1567 	memcpy(wgp->wgp_timestamp_latest_init, timestamp, sizeof(timestamp));
   1568 
   1569 	/*
   1570 	 * Message is good -- we're committing to handle it now, unless
   1571 	 * we were already initiating a session.
   1572 	 */
   1573 	wgs = wgp->wgp_session_unstable;
   1574 	switch (wgs->wgs_state) {
   1575 	case WGS_STATE_UNKNOWN:		/* new session initiated by peer */
   1576 		wg_get_session_index(wg, wgs);
   1577 		break;
   1578 	case WGS_STATE_INIT_ACTIVE:	/* we're already initiating, drop */
   1579 		WG_TRACE("Session already initializing, ignoring the message");
   1580 		goto out;
   1581 	case WGS_STATE_INIT_PASSIVE:	/* peer is retrying, start over */
   1582 		WG_TRACE("Session already initializing, destroying old states");
   1583 		wg_clear_states(wgs);
   1584 		/* keep session index */
   1585 		break;
   1586 	case WGS_STATE_ESTABLISHED:	/* can't happen */
   1587 		panic("unstable session can't be established");
   1588 		break;
   1589 	case WGS_STATE_DESTROYING:	/* rekey initiated by peer */
   1590 		WG_TRACE("Session destroying, but force to clear");
   1591 		callout_stop(&wgp->wgp_session_dtor_timer);
   1592 		wg_clear_states(wgs);
   1593 		/* keep session index */
   1594 		break;
   1595 	default:
   1596 		panic("invalid session state: %d", wgs->wgs_state);
   1597 	}
   1598 	wgs->wgs_state = WGS_STATE_INIT_PASSIVE;
   1599 
   1600 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
   1601 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
   1602 	memcpy(wgs->wgs_ephemeral_key_peer, wgmi->wgmi_ephemeral,
   1603 	    sizeof(wgmi->wgmi_ephemeral));
   1604 
   1605 	wg_update_endpoint_if_necessary(wgp, src);
   1606 
   1607 	(void)wg_send_handshake_msg_resp(wg, wgp, wgs, wgmi);
   1608 
   1609 	wg_calculate_keys(wgs, false);
   1610 	wg_clear_states(wgs);
   1611 
   1612 out:
   1613 	mutex_exit(wgp->wgp_lock);
   1614 	wg_put_peer(wgp, &psref_peer);
   1615 }
   1616 
   1617 static struct socket *
   1618 wg_get_so_by_af(struct wg_softc *wg, const int af)
   1619 {
   1620 
   1621 	switch (af) {
   1622 #ifdef INET
   1623 	case AF_INET:
   1624 		return wg->wg_so4;
   1625 #endif
   1626 #ifdef INET6
   1627 	case AF_INET6:
   1628 		return wg->wg_so6;
   1629 #endif
   1630 	default:
   1631 		panic("wg: no such af: %d", af);
   1632 	}
   1633 }
   1634 
   1635 static struct socket *
   1636 wg_get_so_by_peer(struct wg_peer *wgp, struct wg_sockaddr *wgsa)
   1637 {
   1638 
   1639 	return wg_get_so_by_af(wgp->wgp_sc, wgsa_family(wgsa));
   1640 }
   1641 
   1642 static struct wg_sockaddr *
   1643 wg_get_endpoint_sa(struct wg_peer *wgp, struct psref *psref)
   1644 {
   1645 	struct wg_sockaddr *wgsa;
   1646 	int s;
   1647 
   1648 	s = pserialize_read_enter();
   1649 	wgsa = atomic_load_consume(&wgp->wgp_endpoint);
   1650 	psref_acquire(psref, &wgsa->wgsa_psref, wg_psref_class);
   1651 	pserialize_read_exit(s);
   1652 
   1653 	return wgsa;
   1654 }
   1655 
   1656 static void
   1657 wg_put_sa(struct wg_peer *wgp, struct wg_sockaddr *wgsa, struct psref *psref)
   1658 {
   1659 
   1660 	psref_release(psref, &wgsa->wgsa_psref, wg_psref_class);
   1661 }
   1662 
   1663 static int
   1664 wg_send_so(struct wg_peer *wgp, struct mbuf *m)
   1665 {
   1666 	int error;
   1667 	struct socket *so;
   1668 	struct psref psref;
   1669 	struct wg_sockaddr *wgsa;
   1670 
   1671 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   1672 	so = wg_get_so_by_peer(wgp, wgsa);
   1673 	error = sosend(so, wgsatosa(wgsa), NULL, m, NULL, 0, curlwp);
   1674 	wg_put_sa(wgp, wgsa, &psref);
   1675 
   1676 	return error;
   1677 }
   1678 
   1679 static int
   1680 wg_send_handshake_msg_init(struct wg_softc *wg, struct wg_peer *wgp)
   1681 {
   1682 	int error;
   1683 	struct mbuf *m;
   1684 	struct wg_msg_init *wgmi;
   1685 	struct wg_session *wgs;
   1686 
   1687 	KASSERT(mutex_owned(wgp->wgp_lock));
   1688 
   1689 	wgs = wgp->wgp_session_unstable;
   1690 	/* XXX pull dispatch out into wg_task_send_init_message */
   1691 	switch (wgs->wgs_state) {
   1692 	case WGS_STATE_UNKNOWN:		/* new session initiated by us */
   1693 		wg_get_session_index(wg, wgs);
   1694 		break;
   1695 	case WGS_STATE_INIT_ACTIVE:	/* we're already initiating, stop */
   1696 		WG_TRACE("Session already initializing, skip starting new one");
   1697 		return EBUSY;
   1698 	case WGS_STATE_INIT_PASSIVE:	/* peer was trying -- XXX what now? */
   1699 		WG_TRACE("Session already initializing, destroying old states");
   1700 		wg_clear_states(wgs);
   1701 		/* keep session index */
   1702 		break;
   1703 	case WGS_STATE_ESTABLISHED:	/* can't happen */
   1704 		panic("unstable session can't be established");
   1705 		break;
   1706 	case WGS_STATE_DESTROYING:	/* rekey initiated by us too early */
   1707 		WG_TRACE("Session destroying");
   1708 		/* XXX should wait? */
   1709 		return EBUSY;
   1710 	}
   1711 	wgs->wgs_state = WGS_STATE_INIT_ACTIVE;
   1712 
   1713 	m = m_gethdr(M_WAIT, MT_DATA);
   1714 	if (sizeof(*wgmi) > MHLEN) {
   1715 		m_clget(m, M_WAIT);
   1716 		CTASSERT(sizeof(*wgmi) <= MCLBYTES);
   1717 	}
   1718 	m->m_pkthdr.len = m->m_len = sizeof(*wgmi);
   1719 	wgmi = mtod(m, struct wg_msg_init *);
   1720 	wg_fill_msg_init(wg, wgp, wgs, wgmi);
   1721 
   1722 	error = wg->wg_ops->send_hs_msg(wgp, m);
   1723 	if (error == 0) {
   1724 		WG_TRACE("init msg sent");
   1725 
   1726 		if (wgp->wgp_handshake_start_time == 0)
   1727 			wgp->wgp_handshake_start_time = time_uptime;
   1728 		callout_schedule(&wgp->wgp_handshake_timeout_timer,
   1729 		    MIN(wg_rekey_timeout, (unsigned)(INT_MAX / hz)) * hz);
   1730 	} else {
   1731 		wg_put_session_index(wg, wgs);
   1732 		/* Initiation failed; toss packet waiting for it if any.  */
   1733 		if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL)
   1734 			m_freem(m);
   1735 	}
   1736 
   1737 	return error;
   1738 }
   1739 
   1740 static void
   1741 wg_fill_msg_resp(struct wg_softc *wg, struct wg_peer *wgp,
   1742     struct wg_session *wgs, struct wg_msg_resp *wgmr,
   1743     const struct wg_msg_init *wgmi)
   1744 {
   1745 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.3: Cr */
   1746 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.3: Hr */
   1747 	uint8_t cipher_key[WG_KDF_OUTPUT_LEN];
   1748 	uint8_t pubkey[WG_EPHEMERAL_KEY_LEN];
   1749 	uint8_t privkey[WG_EPHEMERAL_KEY_LEN];
   1750 
   1751 	KASSERT(mutex_owned(wgp->wgp_lock));
   1752 	KASSERT(wgs == wgp->wgp_session_unstable);
   1753 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_PASSIVE);
   1754 
   1755 	memcpy(hash, wgs->wgs_handshake_hash, sizeof(hash));
   1756 	memcpy(ckey, wgs->wgs_chaining_key, sizeof(ckey));
   1757 
   1758 	wgmr->wgmr_type = htole32(WG_MSG_TYPE_RESP);
   1759 	wgmr->wgmr_sender = wgs->wgs_local_index;
   1760 	wgmr->wgmr_receiver = wgmi->wgmi_sender;
   1761 
   1762 	/* [W] 5.4.3 Second Message: Responder to Initiator */
   1763 
   1764 	/* [N] 2.2: "e" */
   1765 	/* Er^priv, Er^pub := DH-GENERATE() */
   1766 	wg_algo_generate_keypair(pubkey, privkey);
   1767 	/* Cr := KDF1(Cr, Er^pub) */
   1768 	wg_algo_kdf(ckey, NULL, NULL, ckey, pubkey, sizeof(pubkey));
   1769 	/* msg.ephemeral := Er^pub */
   1770 	memcpy(wgmr->wgmr_ephemeral, pubkey, sizeof(wgmr->wgmr_ephemeral));
   1771 	/* Hr := HASH(Hr || msg.ephemeral) */
   1772 	wg_algo_hash(hash, pubkey, sizeof(pubkey));
   1773 
   1774 	WG_DUMP_HASH("ckey", ckey);
   1775 	WG_DUMP_HASH("hash", hash);
   1776 
   1777 	/* [N] 2.2: "ee" */
   1778 	/* Cr := KDF1(Cr, DH(Er^priv, Ei^pub)) */
   1779 	wg_algo_dh_kdf(ckey, NULL, privkey, wgs->wgs_ephemeral_key_peer);
   1780 
   1781 	/* [N] 2.2: "se" */
   1782 	/* Cr := KDF1(Cr, DH(Er^priv, Si^pub)) */
   1783 	wg_algo_dh_kdf(ckey, NULL, privkey, wgp->wgp_pubkey);
   1784 
   1785 	/* [N] 9.2: "psk" */
   1786     {
   1787 	uint8_t kdfout[WG_KDF_OUTPUT_LEN];
   1788 	/* Cr, r, k := KDF3(Cr, Q) */
   1789 	wg_algo_kdf(ckey, kdfout, cipher_key, ckey, wgp->wgp_psk,
   1790 	    sizeof(wgp->wgp_psk));
   1791 	/* Hr := HASH(Hr || r) */
   1792 	wg_algo_hash(hash, kdfout, sizeof(kdfout));
   1793     }
   1794 
   1795 	/* msg.empty := AEAD(k, 0, e, Hr) */
   1796 	wg_algo_aead_enc(wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty),
   1797 	    cipher_key, 0, NULL, 0, hash, sizeof(hash));
   1798 	/* Hr := HASH(Hr || msg.empty) */
   1799 	wg_algo_hash(hash, wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty));
   1800 
   1801 	WG_DUMP_HASH("wgmr_empty", wgmr->wgmr_empty);
   1802 
   1803 	/* [W] 5.4.4: Cookie MACs */
   1804 	/* msg.mac1 := MAC(HASH(LABEL-MAC1 || Sm'^pub), msg_a) */
   1805 	wg_algo_mac_mac1(wgmr->wgmr_mac1, sizeof(wgmi->wgmi_mac1),
   1806 	    wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey),
   1807 	    (const uint8_t *)wgmr, offsetof(struct wg_msg_resp, wgmr_mac1));
   1808 	/* Need mac1 to decrypt a cookie from a cookie message */
   1809 	memcpy(wgp->wgp_last_sent_mac1, wgmr->wgmr_mac1,
   1810 	    sizeof(wgp->wgp_last_sent_mac1));
   1811 	wgp->wgp_last_sent_mac1_valid = true;
   1812 
   1813 	if (wgp->wgp_latest_cookie_time == 0 ||
   1814 	    (time_uptime - wgp->wgp_latest_cookie_time) >= WG_COOKIE_TIME)
   1815 		/* msg.mac2 := 0^16 */
   1816 		memset(wgmr->wgmr_mac2, 0, sizeof(wgmr->wgmr_mac2));
   1817 	else {
   1818 		/* msg.mac2 := MAC(Lm, msg_b) */
   1819 		wg_algo_mac(wgmr->wgmr_mac2, sizeof(wgmi->wgmi_mac2),
   1820 		    wgp->wgp_latest_cookie, WG_COOKIE_LEN,
   1821 		    (const uint8_t *)wgmr,
   1822 		    offsetof(struct wg_msg_resp, wgmr_mac2),
   1823 		    NULL, 0);
   1824 	}
   1825 
   1826 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
   1827 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
   1828 	memcpy(wgs->wgs_ephemeral_key_pub, pubkey, sizeof(pubkey));
   1829 	memcpy(wgs->wgs_ephemeral_key_priv, privkey, sizeof(privkey));
   1830 	wgs->wgs_remote_index = wgmi->wgmi_sender;
   1831 	WG_DLOG("sender=%x\n", wgs->wgs_local_index);
   1832 	WG_DLOG("receiver=%x\n", wgs->wgs_remote_index);
   1833 }
   1834 
   1835 static void
   1836 wg_swap_sessions(struct wg_peer *wgp)
   1837 {
   1838 	struct wg_session *wgs, *wgs_prev;
   1839 
   1840 	KASSERT(mutex_owned(wgp->wgp_lock));
   1841 
   1842 	wgs = wgp->wgp_session_unstable;
   1843 	KASSERT(wgs->wgs_state == WGS_STATE_ESTABLISHED);
   1844 
   1845 	wgs_prev = wgp->wgp_session_stable;
   1846 	KASSERT(wgs_prev->wgs_state == WGS_STATE_ESTABLISHED ||
   1847 	    wgs_prev->wgs_state == WGS_STATE_UNKNOWN);
   1848 	atomic_store_release(&wgp->wgp_session_stable, wgs);
   1849 	wgp->wgp_session_unstable = wgs_prev;
   1850 }
   1851 
   1852 static void __noinline
   1853 wg_handle_msg_resp(struct wg_softc *wg, const struct wg_msg_resp *wgmr,
   1854     const struct sockaddr *src)
   1855 {
   1856 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.3: Cr */
   1857 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.3: Kr */
   1858 	uint8_t cipher_key[WG_KDF_OUTPUT_LEN];
   1859 	struct wg_peer *wgp;
   1860 	struct wg_session *wgs;
   1861 	struct psref psref;
   1862 	int error;
   1863 	uint8_t mac1[WG_MAC_LEN];
   1864 	struct wg_session *wgs_prev;
   1865 	struct mbuf *m;
   1866 
   1867 	wg_algo_mac_mac1(mac1, sizeof(mac1),
   1868 	    wg->wg_pubkey, sizeof(wg->wg_pubkey),
   1869 	    (const uint8_t *)wgmr, offsetof(struct wg_msg_resp, wgmr_mac1));
   1870 
   1871 	/*
   1872 	 * [W] 5.3: Denial of Service Mitigation & Cookies
   1873 	 * "the responder, ..., must always reject messages with an invalid
   1874 	 *  msg.mac1"
   1875 	 */
   1876 	if (!consttime_memequal(mac1, wgmr->wgmr_mac1, sizeof(mac1))) {
   1877 		WG_DLOG("mac1 is invalid\n");
   1878 		return;
   1879 	}
   1880 
   1881 	WG_TRACE("resp msg received");
   1882 	wgs = wg_lookup_session_by_index(wg, wgmr->wgmr_receiver, &psref);
   1883 	if (wgs == NULL) {
   1884 		WG_TRACE("No session found");
   1885 		return;
   1886 	}
   1887 
   1888 	wgp = wgs->wgs_peer;
   1889 
   1890 	mutex_enter(wgp->wgp_lock);
   1891 
   1892 	/* If we weren't waiting for a handshake response, drop it.  */
   1893 	if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE) {
   1894 		WG_TRACE("peer sent spurious handshake response, ignoring");
   1895 		goto out;
   1896 	}
   1897 
   1898 	if (__predict_false(wg_is_underload(wg, wgp, WG_MSG_TYPE_RESP))) {
   1899 		WG_TRACE("under load");
   1900 		/*
   1901 		 * [W] 5.3: Denial of Service Mitigation & Cookies
   1902 		 * "the responder, ..., and when under load may reject messages
   1903 		 *  with an invalid msg.mac2.  If the responder receives a
   1904 		 *  message with a valid msg.mac1 yet with an invalid msg.mac2,
   1905 		 *  and is under load, it may respond with a cookie reply
   1906 		 *  message"
   1907 		 */
   1908 		uint8_t zero[WG_MAC_LEN] = {0};
   1909 		if (consttime_memequal(wgmr->wgmr_mac2, zero, sizeof(zero))) {
   1910 			WG_TRACE("sending a cookie message: no cookie included");
   1911 			(void)wg_send_cookie_msg(wg, wgp, wgmr->wgmr_sender,
   1912 			    wgmr->wgmr_mac1, src);
   1913 			goto out;
   1914 		}
   1915 		if (!wgp->wgp_last_sent_cookie_valid) {
   1916 			WG_TRACE("sending a cookie message: no cookie sent ever");
   1917 			(void)wg_send_cookie_msg(wg, wgp, wgmr->wgmr_sender,
   1918 			    wgmr->wgmr_mac1, src);
   1919 			goto out;
   1920 		}
   1921 		uint8_t mac2[WG_MAC_LEN];
   1922 		wg_algo_mac(mac2, sizeof(mac2), wgp->wgp_last_sent_cookie,
   1923 		    WG_COOKIE_LEN, (const uint8_t *)wgmr,
   1924 		    offsetof(struct wg_msg_resp, wgmr_mac2), NULL, 0);
   1925 		if (!consttime_memequal(mac2, wgmr->wgmr_mac2, sizeof(mac2))) {
   1926 			WG_DLOG("mac2 is invalid\n");
   1927 			goto out;
   1928 		}
   1929 		WG_TRACE("under load, but continue to sending");
   1930 	}
   1931 
   1932 	memcpy(hash, wgs->wgs_handshake_hash, sizeof(hash));
   1933 	memcpy(ckey, wgs->wgs_chaining_key, sizeof(ckey));
   1934 
   1935 	/*
   1936 	 * [W] 5.4.3 Second Message: Responder to Initiator
   1937 	 * "When the initiator receives this message, it does the same
   1938 	 *  operations so that its final state variables are identical,
   1939 	 *  replacing the operands of the DH function to produce equivalent
   1940 	 *  values."
   1941 	 *  Note that the following comments of operations are just copies of
   1942 	 *  the initiator's ones.
   1943 	 */
   1944 
   1945 	/* [N] 2.2: "e" */
   1946 	/* Cr := KDF1(Cr, Er^pub) */
   1947 	wg_algo_kdf(ckey, NULL, NULL, ckey, wgmr->wgmr_ephemeral,
   1948 	    sizeof(wgmr->wgmr_ephemeral));
   1949 	/* Hr := HASH(Hr || msg.ephemeral) */
   1950 	wg_algo_hash(hash, wgmr->wgmr_ephemeral, sizeof(wgmr->wgmr_ephemeral));
   1951 
   1952 	WG_DUMP_HASH("ckey", ckey);
   1953 	WG_DUMP_HASH("hash", hash);
   1954 
   1955 	/* [N] 2.2: "ee" */
   1956 	/* Cr := KDF1(Cr, DH(Er^priv, Ei^pub)) */
   1957 	wg_algo_dh_kdf(ckey, NULL, wgs->wgs_ephemeral_key_priv,
   1958 	    wgmr->wgmr_ephemeral);
   1959 
   1960 	/* [N] 2.2: "se" */
   1961 	/* Cr := KDF1(Cr, DH(Er^priv, Si^pub)) */
   1962 	wg_algo_dh_kdf(ckey, NULL, wg->wg_privkey, wgmr->wgmr_ephemeral);
   1963 
   1964 	/* [N] 9.2: "psk" */
   1965     {
   1966 	uint8_t kdfout[WG_KDF_OUTPUT_LEN];
   1967 	/* Cr, r, k := KDF3(Cr, Q) */
   1968 	wg_algo_kdf(ckey, kdfout, cipher_key, ckey, wgp->wgp_psk,
   1969 	    sizeof(wgp->wgp_psk));
   1970 	/* Hr := HASH(Hr || r) */
   1971 	wg_algo_hash(hash, kdfout, sizeof(kdfout));
   1972     }
   1973 
   1974     {
   1975 	uint8_t out[sizeof(wgmr->wgmr_empty)]; /* for safety */
   1976 	/* msg.empty := AEAD(k, 0, e, Hr) */
   1977 	error = wg_algo_aead_dec(out, 0, cipher_key, 0, wgmr->wgmr_empty,
   1978 	    sizeof(wgmr->wgmr_empty), hash, sizeof(hash));
   1979 	WG_DUMP_HASH("wgmr_empty", wgmr->wgmr_empty);
   1980 	if (error != 0) {
   1981 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   1982 		    "%s: peer %s: wg_algo_aead_dec for empty message failed\n",
   1983 		    if_name(&wg->wg_if), wgp->wgp_name);
   1984 		goto out;
   1985 	}
   1986 	/* Hr := HASH(Hr || msg.empty) */
   1987 	wg_algo_hash(hash, wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty));
   1988     }
   1989 
   1990 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(wgs->wgs_handshake_hash));
   1991 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(wgs->wgs_chaining_key));
   1992 	wgs->wgs_remote_index = wgmr->wgmr_sender;
   1993 	WG_DLOG("receiver=%x\n", wgs->wgs_remote_index);
   1994 
   1995 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_ACTIVE);
   1996 	wgs->wgs_state = WGS_STATE_ESTABLISHED;
   1997 	wgs->wgs_time_established = time_uptime;
   1998 	wgs->wgs_time_last_data_sent = 0;
   1999 	wgs->wgs_is_initiator = true;
   2000 	wg_calculate_keys(wgs, true);
   2001 	wg_clear_states(wgs);
   2002 	WG_TRACE("WGS_STATE_ESTABLISHED");
   2003 
   2004 	callout_stop(&wgp->wgp_handshake_timeout_timer);
   2005 
   2006 	wg_swap_sessions(wgp);
   2007 	KASSERT(wgs == wgp->wgp_session_stable);
   2008 	wgs_prev = wgp->wgp_session_unstable;
   2009 	getnanotime(&wgp->wgp_last_handshake_time);
   2010 	wgp->wgp_handshake_start_time = 0;
   2011 	wgp->wgp_last_sent_mac1_valid = false;
   2012 	wgp->wgp_last_sent_cookie_valid = false;
   2013 
   2014 	wg_schedule_rekey_timer(wgp);
   2015 
   2016 	wg_update_endpoint_if_necessary(wgp, src);
   2017 
   2018 	/*
   2019 	 * If we had a data packet queued up, send it; otherwise send a
   2020 	 * keepalive message -- either way we have to send something
   2021 	 * immediately or else the responder will never answer.
   2022 	 */
   2023 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL) {
   2024 		kpreempt_disable();
   2025 		const uint32_t h = curcpu()->ci_index; // pktq_rps_hash(m)
   2026 		M_SETCTX(m, wgp);
   2027 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   2028 			WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   2029 			    if_name(&wg->wg_if));
   2030 			m_freem(m);
   2031 		}
   2032 		kpreempt_enable();
   2033 	} else {
   2034 		wg_send_keepalive_msg(wgp, wgs);
   2035 	}
   2036 
   2037 	if (wgs_prev->wgs_state == WGS_STATE_ESTABLISHED) {
   2038 		/* Wait for wg_get_stable_session to drain.  */
   2039 		pserialize_perform(wgp->wgp_psz);
   2040 
   2041 		/* Transition ESTABLISHED->DESTROYING.  */
   2042 		wgs_prev->wgs_state = WGS_STATE_DESTROYING;
   2043 
   2044 		/* We can't destroy the old session immediately */
   2045 		wg_schedule_session_dtor_timer(wgp);
   2046 	} else {
   2047 		KASSERTMSG(wgs_prev->wgs_state == WGS_STATE_UNKNOWN,
   2048 		    "state=%d", wgs_prev->wgs_state);
   2049 	}
   2050 
   2051 out:
   2052 	mutex_exit(wgp->wgp_lock);
   2053 	wg_put_session(wgs, &psref);
   2054 }
   2055 
   2056 static int
   2057 wg_send_handshake_msg_resp(struct wg_softc *wg, struct wg_peer *wgp,
   2058     struct wg_session *wgs, const struct wg_msg_init *wgmi)
   2059 {
   2060 	int error;
   2061 	struct mbuf *m;
   2062 	struct wg_msg_resp *wgmr;
   2063 
   2064 	KASSERT(mutex_owned(wgp->wgp_lock));
   2065 	KASSERT(wgs == wgp->wgp_session_unstable);
   2066 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_PASSIVE);
   2067 
   2068 	m = m_gethdr(M_WAIT, MT_DATA);
   2069 	if (sizeof(*wgmr) > MHLEN) {
   2070 		m_clget(m, M_WAIT);
   2071 		CTASSERT(sizeof(*wgmr) <= MCLBYTES);
   2072 	}
   2073 	m->m_pkthdr.len = m->m_len = sizeof(*wgmr);
   2074 	wgmr = mtod(m, struct wg_msg_resp *);
   2075 	wg_fill_msg_resp(wg, wgp, wgs, wgmr, wgmi);
   2076 
   2077 	error = wg->wg_ops->send_hs_msg(wgp, m);
   2078 	if (error == 0)
   2079 		WG_TRACE("resp msg sent");
   2080 	return error;
   2081 }
   2082 
   2083 static struct wg_peer *
   2084 wg_lookup_peer_by_pubkey(struct wg_softc *wg,
   2085     const uint8_t pubkey[WG_STATIC_KEY_LEN], struct psref *psref)
   2086 {
   2087 	struct wg_peer *wgp;
   2088 
   2089 	int s = pserialize_read_enter();
   2090 	wgp = thmap_get(wg->wg_peers_bypubkey, pubkey, WG_STATIC_KEY_LEN);
   2091 	if (wgp != NULL)
   2092 		wg_get_peer(wgp, psref);
   2093 	pserialize_read_exit(s);
   2094 
   2095 	return wgp;
   2096 }
   2097 
   2098 static void
   2099 wg_fill_msg_cookie(struct wg_softc *wg, struct wg_peer *wgp,
   2100     struct wg_msg_cookie *wgmc, const uint32_t sender,
   2101     const uint8_t mac1[WG_MAC_LEN], const struct sockaddr *src)
   2102 {
   2103 	uint8_t cookie[WG_COOKIE_LEN];
   2104 	uint8_t key[WG_HASH_LEN];
   2105 	uint8_t addr[sizeof(struct in6_addr)];
   2106 	size_t addrlen;
   2107 	uint16_t uh_sport; /* be */
   2108 
   2109 	KASSERT(mutex_owned(wgp->wgp_lock));
   2110 
   2111 	wgmc->wgmc_type = htole32(WG_MSG_TYPE_COOKIE);
   2112 	wgmc->wgmc_receiver = sender;
   2113 	cprng_fast(wgmc->wgmc_salt, sizeof(wgmc->wgmc_salt));
   2114 
   2115 	/*
   2116 	 * [W] 5.4.7: Under Load: Cookie Reply Message
   2117 	 * "The secret variable, Rm, changes every two minutes to a
   2118 	 * random value"
   2119 	 */
   2120 	if ((time_uptime - wgp->wgp_last_genrandval_time) > WG_RANDVAL_TIME) {
   2121 		wgp->wgp_randval = cprng_strong32();
   2122 		wgp->wgp_last_genrandval_time = time_uptime;
   2123 	}
   2124 
   2125 	switch (src->sa_family) {
   2126 	case AF_INET: {
   2127 		const struct sockaddr_in *sin = satocsin(src);
   2128 		addrlen = sizeof(sin->sin_addr);
   2129 		memcpy(addr, &sin->sin_addr, addrlen);
   2130 		uh_sport = sin->sin_port;
   2131 		break;
   2132 	    }
   2133 #ifdef INET6
   2134 	case AF_INET6: {
   2135 		const struct sockaddr_in6 *sin6 = satocsin6(src);
   2136 		addrlen = sizeof(sin6->sin6_addr);
   2137 		memcpy(addr, &sin6->sin6_addr, addrlen);
   2138 		uh_sport = sin6->sin6_port;
   2139 		break;
   2140 	    }
   2141 #endif
   2142 	default:
   2143 		panic("invalid af=%d", src->sa_family);
   2144 	}
   2145 
   2146 	wg_algo_mac(cookie, sizeof(cookie),
   2147 	    (const uint8_t *)&wgp->wgp_randval, sizeof(wgp->wgp_randval),
   2148 	    addr, addrlen, (const uint8_t *)&uh_sport, sizeof(uh_sport));
   2149 	wg_algo_mac_cookie(key, sizeof(key), wg->wg_pubkey,
   2150 	    sizeof(wg->wg_pubkey));
   2151 	wg_algo_xaead_enc(wgmc->wgmc_cookie, sizeof(wgmc->wgmc_cookie), key,
   2152 	    cookie, sizeof(cookie), mac1, WG_MAC_LEN, wgmc->wgmc_salt);
   2153 
   2154 	/* Need to store to calculate mac2 */
   2155 	memcpy(wgp->wgp_last_sent_cookie, cookie, sizeof(cookie));
   2156 	wgp->wgp_last_sent_cookie_valid = true;
   2157 }
   2158 
   2159 static int
   2160 wg_send_cookie_msg(struct wg_softc *wg, struct wg_peer *wgp,
   2161     const uint32_t sender, const uint8_t mac1[WG_MAC_LEN],
   2162     const struct sockaddr *src)
   2163 {
   2164 	int error;
   2165 	struct mbuf *m;
   2166 	struct wg_msg_cookie *wgmc;
   2167 
   2168 	KASSERT(mutex_owned(wgp->wgp_lock));
   2169 
   2170 	m = m_gethdr(M_WAIT, MT_DATA);
   2171 	if (sizeof(*wgmc) > MHLEN) {
   2172 		m_clget(m, M_WAIT);
   2173 		CTASSERT(sizeof(*wgmc) <= MCLBYTES);
   2174 	}
   2175 	m->m_pkthdr.len = m->m_len = sizeof(*wgmc);
   2176 	wgmc = mtod(m, struct wg_msg_cookie *);
   2177 	wg_fill_msg_cookie(wg, wgp, wgmc, sender, mac1, src);
   2178 
   2179 	error = wg->wg_ops->send_hs_msg(wgp, m);
   2180 	if (error == 0)
   2181 		WG_TRACE("cookie msg sent");
   2182 	return error;
   2183 }
   2184 
   2185 static bool
   2186 wg_is_underload(struct wg_softc *wg, struct wg_peer *wgp, int msgtype)
   2187 {
   2188 #ifdef WG_DEBUG_PARAMS
   2189 	if (wg_force_underload)
   2190 		return true;
   2191 #endif
   2192 
   2193 	/*
   2194 	 * XXX we don't have a means of a load estimation.  The purpose of
   2195 	 * the mechanism is a DoS mitigation, so we consider frequent handshake
   2196 	 * messages as (a kind of) load; if a message of the same type comes
   2197 	 * to a peer within 1 second, we consider we are under load.
   2198 	 */
   2199 	time_t last = wgp->wgp_last_msg_received_time[msgtype];
   2200 	wgp->wgp_last_msg_received_time[msgtype] = time_uptime;
   2201 	return (time_uptime - last) == 0;
   2202 }
   2203 
   2204 static void
   2205 wg_calculate_keys(struct wg_session *wgs, const bool initiator)
   2206 {
   2207 
   2208 	KASSERT(mutex_owned(wgs->wgs_peer->wgp_lock));
   2209 
   2210 	/*
   2211 	 * [W] 5.4.5: Ti^send = Tr^recv, Ti^recv = Tr^send := KDF2(Ci = Cr, e)
   2212 	 */
   2213 	if (initiator) {
   2214 		wg_algo_kdf(wgs->wgs_tkey_send, wgs->wgs_tkey_recv, NULL,
   2215 		    wgs->wgs_chaining_key, NULL, 0);
   2216 	} else {
   2217 		wg_algo_kdf(wgs->wgs_tkey_recv, wgs->wgs_tkey_send, NULL,
   2218 		    wgs->wgs_chaining_key, NULL, 0);
   2219 	}
   2220 	WG_DUMP_HASH("wgs_tkey_send", wgs->wgs_tkey_send);
   2221 	WG_DUMP_HASH("wgs_tkey_recv", wgs->wgs_tkey_recv);
   2222 }
   2223 
   2224 static uint64_t
   2225 wg_session_get_send_counter(struct wg_session *wgs)
   2226 {
   2227 #ifdef __HAVE_ATOMIC64_LOADSTORE
   2228 	return atomic_load_relaxed(&wgs->wgs_send_counter);
   2229 #else
   2230 	uint64_t send_counter;
   2231 
   2232 	mutex_enter(&wgs->wgs_send_counter_lock);
   2233 	send_counter = wgs->wgs_send_counter;
   2234 	mutex_exit(&wgs->wgs_send_counter_lock);
   2235 
   2236 	return send_counter;
   2237 #endif
   2238 }
   2239 
   2240 static uint64_t
   2241 wg_session_inc_send_counter(struct wg_session *wgs)
   2242 {
   2243 #ifdef __HAVE_ATOMIC64_LOADSTORE
   2244 	return atomic_inc_64_nv(&wgs->wgs_send_counter) - 1;
   2245 #else
   2246 	uint64_t send_counter;
   2247 
   2248 	mutex_enter(&wgs->wgs_send_counter_lock);
   2249 	send_counter = wgs->wgs_send_counter++;
   2250 	mutex_exit(&wgs->wgs_send_counter_lock);
   2251 
   2252 	return send_counter;
   2253 #endif
   2254 }
   2255 
   2256 static void
   2257 wg_clear_states(struct wg_session *wgs)
   2258 {
   2259 
   2260 	KASSERT(mutex_owned(wgs->wgs_peer->wgp_lock));
   2261 
   2262 	wgs->wgs_send_counter = 0;
   2263 	sliwin_reset(&wgs->wgs_recvwin->window);
   2264 
   2265 #define wgs_clear(v)	explicit_memset(wgs->wgs_##v, 0, sizeof(wgs->wgs_##v))
   2266 	wgs_clear(handshake_hash);
   2267 	wgs_clear(chaining_key);
   2268 	wgs_clear(ephemeral_key_pub);
   2269 	wgs_clear(ephemeral_key_priv);
   2270 	wgs_clear(ephemeral_key_peer);
   2271 #undef wgs_clear
   2272 }
   2273 
   2274 static struct wg_session *
   2275 wg_lookup_session_by_index(struct wg_softc *wg, const uint32_t index,
   2276     struct psref *psref)
   2277 {
   2278 	struct wg_session *wgs;
   2279 
   2280 	int s = pserialize_read_enter();
   2281 	wgs = thmap_get(wg->wg_sessions_byindex, &index, sizeof index);
   2282 	if (wgs != NULL) {
   2283 		KASSERT(atomic_load_relaxed(&wgs->wgs_state) !=
   2284 		    WGS_STATE_UNKNOWN);
   2285 		psref_acquire(psref, &wgs->wgs_psref, wg_psref_class);
   2286 	}
   2287 	pserialize_read_exit(s);
   2288 
   2289 	return wgs;
   2290 }
   2291 
   2292 static void
   2293 wg_schedule_rekey_timer(struct wg_peer *wgp)
   2294 {
   2295 	int timeout = MIN(wg_rekey_after_time, (unsigned)(INT_MAX / hz));
   2296 
   2297 	callout_schedule(&wgp->wgp_rekey_timer, timeout * hz);
   2298 }
   2299 
   2300 static void
   2301 wg_send_keepalive_msg(struct wg_peer *wgp, struct wg_session *wgs)
   2302 {
   2303 	struct mbuf *m;
   2304 
   2305 	/*
   2306 	 * [W] 6.5 Passive Keepalive
   2307 	 * "A keepalive message is simply a transport data message with
   2308 	 *  a zero-length encapsulated encrypted inner-packet."
   2309 	 */
   2310 	m = m_gethdr(M_WAIT, MT_DATA);
   2311 	wg_send_data_msg(wgp, wgs, m);
   2312 }
   2313 
   2314 static bool
   2315 wg_need_to_send_init_message(struct wg_session *wgs)
   2316 {
   2317 	/*
   2318 	 * [W] 6.2 Transport Message Limits
   2319 	 * "if a peer is the initiator of a current secure session,
   2320 	 *  WireGuard will send a handshake initiation message to begin
   2321 	 *  a new secure session ... if after receiving a transport data
   2322 	 *  message, the current secure session is (REJECT-AFTER-TIME 
   2323 	 *  KEEPALIVE-TIMEOUT  REKEY-TIMEOUT) seconds old and it has
   2324 	 *  not yet acted upon this event."
   2325 	 */
   2326 	return wgs->wgs_is_initiator && wgs->wgs_time_last_data_sent == 0 &&
   2327 	    (time_uptime - wgs->wgs_time_established) >=
   2328 	    (wg_reject_after_time - wg_keepalive_timeout - wg_rekey_timeout);
   2329 }
   2330 
   2331 static void
   2332 wg_schedule_peer_task(struct wg_peer *wgp, unsigned int task)
   2333 {
   2334 
   2335 	mutex_enter(wgp->wgp_intr_lock);
   2336 	WG_DLOG("tasks=%d, task=%d\n", wgp->wgp_tasks, task);
   2337 	if (wgp->wgp_tasks == 0)
   2338 		/*
   2339 		 * XXX If the current CPU is already loaded -- e.g., if
   2340 		 * there's already a bunch of handshakes queued up --
   2341 		 * consider tossing this over to another CPU to
   2342 		 * distribute the load.
   2343 		 */
   2344 		workqueue_enqueue(wg_wq, &wgp->wgp_work, NULL);
   2345 	wgp->wgp_tasks |= task;
   2346 	mutex_exit(wgp->wgp_intr_lock);
   2347 }
   2348 
   2349 static void
   2350 wg_change_endpoint(struct wg_peer *wgp, const struct sockaddr *new)
   2351 {
   2352 	struct wg_sockaddr *wgsa_prev;
   2353 
   2354 	WG_TRACE("Changing endpoint");
   2355 
   2356 	memcpy(wgp->wgp_endpoint0, new, new->sa_len);
   2357 	wgsa_prev = wgp->wgp_endpoint;
   2358 	atomic_store_release(&wgp->wgp_endpoint, wgp->wgp_endpoint0);
   2359 	wgp->wgp_endpoint0 = wgsa_prev;
   2360 	atomic_store_release(&wgp->wgp_endpoint_available, true);
   2361 
   2362 	wg_schedule_peer_task(wgp, WGP_TASK_ENDPOINT_CHANGED);
   2363 }
   2364 
   2365 static bool
   2366 wg_validate_inner_packet(const char *packet, size_t decrypted_len, int *af)
   2367 {
   2368 	uint16_t packet_len;
   2369 	const struct ip *ip;
   2370 
   2371 	if (__predict_false(decrypted_len < sizeof(struct ip)))
   2372 		return false;
   2373 
   2374 	ip = (const struct ip *)packet;
   2375 	if (ip->ip_v == 4)
   2376 		*af = AF_INET;
   2377 	else if (ip->ip_v == 6)
   2378 		*af = AF_INET6;
   2379 	else
   2380 		return false;
   2381 
   2382 	WG_DLOG("af=%d\n", *af);
   2383 
   2384 	switch (*af) {
   2385 #ifdef INET
   2386 	case AF_INET:
   2387 		packet_len = ntohs(ip->ip_len);
   2388 		break;
   2389 #endif
   2390 #ifdef INET6
   2391 	case AF_INET6: {
   2392 		const struct ip6_hdr *ip6;
   2393 
   2394 		if (__predict_false(decrypted_len < sizeof(struct ip6_hdr)))
   2395 			return false;
   2396 
   2397 		ip6 = (const struct ip6_hdr *)packet;
   2398 		packet_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
   2399 		break;
   2400 	}
   2401 #endif
   2402 	default:
   2403 		return false;
   2404 	}
   2405 
   2406 	WG_DLOG("packet_len=%u\n", packet_len);
   2407 	if (packet_len > decrypted_len)
   2408 		return false;
   2409 
   2410 	return true;
   2411 }
   2412 
   2413 static bool
   2414 wg_validate_route(struct wg_softc *wg, struct wg_peer *wgp_expected,
   2415     int af, char *packet)
   2416 {
   2417 	struct sockaddr_storage ss;
   2418 	struct sockaddr *sa;
   2419 	struct psref psref;
   2420 	struct wg_peer *wgp;
   2421 	bool ok;
   2422 
   2423 	/*
   2424 	 * II CRYPTOKEY ROUTING
   2425 	 * "it will only accept it if its source IP resolves in the
   2426 	 *  table to the public key used in the secure session for
   2427 	 *  decrypting it."
   2428 	 */
   2429 
   2430 	if (af == AF_INET) {
   2431 		const struct ip *ip = (const struct ip *)packet;
   2432 		struct sockaddr_in *sin = (struct sockaddr_in *)&ss;
   2433 		sockaddr_in_init(sin, &ip->ip_src, 0);
   2434 		sa = sintosa(sin);
   2435 #ifdef INET6
   2436 	} else {
   2437 		const struct ip6_hdr *ip6 = (const struct ip6_hdr *)packet;
   2438 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ss;
   2439 		sockaddr_in6_init(sin6, &ip6->ip6_src, 0, 0, 0);
   2440 		sa = sin6tosa(sin6);
   2441 #endif
   2442 	}
   2443 
   2444 	wgp = wg_pick_peer_by_sa(wg, sa, &psref);
   2445 	ok = (wgp == wgp_expected);
   2446 	if (wgp != NULL)
   2447 		wg_put_peer(wgp, &psref);
   2448 
   2449 	return ok;
   2450 }
   2451 
   2452 static void
   2453 wg_session_dtor_timer(void *arg)
   2454 {
   2455 	struct wg_peer *wgp = arg;
   2456 
   2457 	WG_TRACE("enter");
   2458 
   2459 	wg_schedule_peer_task(wgp, WGP_TASK_DESTROY_PREV_SESSION);
   2460 }
   2461 
   2462 static void
   2463 wg_schedule_session_dtor_timer(struct wg_peer *wgp)
   2464 {
   2465 
   2466 	/* 1 second grace period */
   2467 	callout_schedule(&wgp->wgp_session_dtor_timer, hz);
   2468 }
   2469 
   2470 static bool
   2471 sockaddr_port_match(const struct sockaddr *sa1, const struct sockaddr *sa2)
   2472 {
   2473 	if (sa1->sa_family != sa2->sa_family)
   2474 		return false;
   2475 
   2476 	switch (sa1->sa_family) {
   2477 #ifdef INET
   2478 	case AF_INET:
   2479 		return satocsin(sa1)->sin_port == satocsin(sa2)->sin_port;
   2480 #endif
   2481 #ifdef INET6
   2482 	case AF_INET6:
   2483 		return satocsin6(sa1)->sin6_port == satocsin6(sa2)->sin6_port;
   2484 #endif
   2485 	default:
   2486 		return false;
   2487 	}
   2488 }
   2489 
   2490 static void
   2491 wg_update_endpoint_if_necessary(struct wg_peer *wgp,
   2492     const struct sockaddr *src)
   2493 {
   2494 	struct wg_sockaddr *wgsa;
   2495 	struct psref psref;
   2496 
   2497 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   2498 
   2499 #ifdef WG_DEBUG_LOG
   2500 	char oldaddr[128], newaddr[128];
   2501 	sockaddr_format(wgsatosa(wgsa), oldaddr, sizeof(oldaddr));
   2502 	sockaddr_format(src, newaddr, sizeof(newaddr));
   2503 	WG_DLOG("old=%s, new=%s\n", oldaddr, newaddr);
   2504 #endif
   2505 
   2506 	/*
   2507 	 * III: "Since the packet has authenticated correctly, the source IP of
   2508 	 * the outer UDP/IP packet is used to update the endpoint for peer..."
   2509 	 */
   2510 	if (__predict_false(sockaddr_cmp(src, wgsatosa(wgsa)) != 0 ||
   2511 		!sockaddr_port_match(src, wgsatosa(wgsa)))) {
   2512 		/* XXX We can't change the endpoint twice in a short period */
   2513 		if (atomic_swap_uint(&wgp->wgp_endpoint_changing, 1) == 0) {
   2514 			wg_change_endpoint(wgp, src);
   2515 		}
   2516 	}
   2517 
   2518 	wg_put_sa(wgp, wgsa, &psref);
   2519 }
   2520 
   2521 static void __noinline
   2522 wg_handle_msg_data(struct wg_softc *wg, struct mbuf *m,
   2523     const struct sockaddr *src)
   2524 {
   2525 	struct wg_msg_data *wgmd;
   2526 	char *encrypted_buf = NULL, *decrypted_buf;
   2527 	size_t encrypted_len, decrypted_len;
   2528 	struct wg_session *wgs;
   2529 	struct wg_peer *wgp;
   2530 	int state;
   2531 	size_t mlen;
   2532 	struct psref psref;
   2533 	int error, af;
   2534 	bool success, free_encrypted_buf = false, ok;
   2535 	struct mbuf *n;
   2536 
   2537 	KASSERT(m->m_len >= sizeof(struct wg_msg_data));
   2538 	wgmd = mtod(m, struct wg_msg_data *);
   2539 
   2540 	KASSERT(wgmd->wgmd_type == htole32(WG_MSG_TYPE_DATA));
   2541 	WG_TRACE("data");
   2542 
   2543 	/* Find the putative session, or drop.  */
   2544 	wgs = wg_lookup_session_by_index(wg, wgmd->wgmd_receiver, &psref);
   2545 	if (wgs == NULL) {
   2546 		WG_TRACE("No session found");
   2547 		m_freem(m);
   2548 		return;
   2549 	}
   2550 
   2551 	/*
   2552 	 * We are only ready to handle data when in INIT_PASSIVE,
   2553 	 * ESTABLISHED, or DESTROYING.  All transitions out of that
   2554 	 * state dissociate the session index and drain psrefs.
   2555 	 */
   2556 	state = atomic_load_relaxed(&wgs->wgs_state);
   2557 	switch (state) {
   2558 	case WGS_STATE_UNKNOWN:
   2559 		panic("wg session %p in unknown state has session index %u",
   2560 		    wgs, wgmd->wgmd_receiver);
   2561 	case WGS_STATE_INIT_ACTIVE:
   2562 		WG_TRACE("not yet ready for data");
   2563 		goto out;
   2564 	case WGS_STATE_INIT_PASSIVE:
   2565 	case WGS_STATE_ESTABLISHED:
   2566 	case WGS_STATE_DESTROYING:
   2567 		break;
   2568 	}
   2569 
   2570 	/*
   2571 	 * Get the peer, for rate-limited logs (XXX MPSAFE, dtrace) and
   2572 	 * to update the endpoint if authentication succeeds.
   2573 	 */
   2574 	wgp = wgs->wgs_peer;
   2575 
   2576 	/*
   2577 	 * Reject outrageously wrong sequence numbers before doing any
   2578 	 * crypto work or taking any locks.
   2579 	 */
   2580 	error = sliwin_check_fast(&wgs->wgs_recvwin->window,
   2581 	    le64toh(wgmd->wgmd_counter));
   2582 	if (error) {
   2583 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2584 		    "%s: peer %s: out-of-window packet: %"PRIu64"\n",
   2585 		    if_name(&wg->wg_if), wgp->wgp_name,
   2586 		    le64toh(wgmd->wgmd_counter));
   2587 		goto out;
   2588 	}
   2589 
   2590 	/* Ensure the payload and authenticator are contiguous.  */
   2591 	mlen = m_length(m);
   2592 	encrypted_len = mlen - sizeof(*wgmd);
   2593 	if (encrypted_len < WG_AUTHTAG_LEN) {
   2594 		WG_DLOG("Short encrypted_len: %lu\n", encrypted_len);
   2595 		goto out;
   2596 	}
   2597 	success = m_ensure_contig(&m, sizeof(*wgmd) + encrypted_len);
   2598 	if (success) {
   2599 		encrypted_buf = mtod(m, char *) + sizeof(*wgmd);
   2600 	} else {
   2601 		encrypted_buf = kmem_intr_alloc(encrypted_len, KM_NOSLEEP);
   2602 		if (encrypted_buf == NULL) {
   2603 			WG_DLOG("failed to allocate encrypted_buf\n");
   2604 			goto out;
   2605 		}
   2606 		m_copydata(m, sizeof(*wgmd), encrypted_len, encrypted_buf);
   2607 		free_encrypted_buf = true;
   2608 	}
   2609 	/* m_ensure_contig may change m regardless of its result */
   2610 	KASSERT(m->m_len >= sizeof(*wgmd));
   2611 	wgmd = mtod(m, struct wg_msg_data *);
   2612 
   2613 	/*
   2614 	 * Get a buffer for the plaintext.  Add WG_AUTHTAG_LEN to avoid
   2615 	 * a zero-length buffer (XXX).  Drop if plaintext is longer
   2616 	 * than MCLBYTES (XXX).
   2617 	 */
   2618 	decrypted_len = encrypted_len - WG_AUTHTAG_LEN;
   2619 	if (decrypted_len > MCLBYTES) {
   2620 		/* FIXME handle larger data than MCLBYTES */
   2621 		WG_DLOG("couldn't handle larger data than MCLBYTES\n");
   2622 		goto out;
   2623 	}
   2624 	n = wg_get_mbuf(0, decrypted_len + WG_AUTHTAG_LEN);
   2625 	if (n == NULL) {
   2626 		WG_DLOG("wg_get_mbuf failed\n");
   2627 		goto out;
   2628 	}
   2629 	decrypted_buf = mtod(n, char *);
   2630 
   2631 	/* Decrypt and verify the packet.  */
   2632 	WG_DLOG("mlen=%lu, encrypted_len=%lu\n", mlen, encrypted_len);
   2633 	error = wg_algo_aead_dec(decrypted_buf,
   2634 	    encrypted_len - WG_AUTHTAG_LEN /* can be 0 */,
   2635 	    wgs->wgs_tkey_recv, le64toh(wgmd->wgmd_counter), encrypted_buf,
   2636 	    encrypted_len, NULL, 0);
   2637 	if (error != 0) {
   2638 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2639 		    "%s: peer %s: failed to wg_algo_aead_dec\n",
   2640 		    if_name(&wg->wg_if), wgp->wgp_name);
   2641 		m_freem(n);
   2642 		goto out;
   2643 	}
   2644 	WG_DLOG("outsize=%u\n", (u_int)decrypted_len);
   2645 
   2646 	/* Packet is genuine.  Reject it if a replay or just too old.  */
   2647 	mutex_enter(&wgs->wgs_recvwin->lock);
   2648 	error = sliwin_update(&wgs->wgs_recvwin->window,
   2649 	    le64toh(wgmd->wgmd_counter));
   2650 	mutex_exit(&wgs->wgs_recvwin->lock);
   2651 	if (error) {
   2652 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2653 		    "%s: peer %s: replay or out-of-window packet: %"PRIu64"\n",
   2654 		    if_name(&wg->wg_if), wgp->wgp_name,
   2655 		    le64toh(wgmd->wgmd_counter));
   2656 		m_freem(n);
   2657 		goto out;
   2658 	}
   2659 
   2660 	/* We're done with m now; free it and chuck the pointers.  */
   2661 	m_freem(m);
   2662 	m = NULL;
   2663 	wgmd = NULL;
   2664 
   2665 	/*
   2666 	 * Validate the encapsulated packet header and get the address
   2667 	 * family, or drop.
   2668 	 */
   2669 	ok = wg_validate_inner_packet(decrypted_buf, decrypted_len, &af);
   2670 	if (!ok) {
   2671 		m_freem(n);
   2672 		goto out;
   2673 	}
   2674 
   2675 	/*
   2676 	 * The packet is genuine.  Update the peer's endpoint if the
   2677 	 * source address changed.
   2678 	 *
   2679 	 * XXX How to prevent DoS by replaying genuine packets from the
   2680 	 * wrong source address?
   2681 	 */
   2682 	wg_update_endpoint_if_necessary(wgp, src);
   2683 
   2684 	/* Submit it into our network stack if routable.  */
   2685 	ok = wg_validate_route(wg, wgp, af, decrypted_buf);
   2686 	if (ok) {
   2687 		wg->wg_ops->input(&wg->wg_if, n, af);
   2688 	} else {
   2689 		char addrstr[INET6_ADDRSTRLEN];
   2690 		memset(addrstr, 0, sizeof(addrstr));
   2691 		if (af == AF_INET) {
   2692 			const struct ip *ip = (const struct ip *)decrypted_buf;
   2693 			IN_PRINT(addrstr, &ip->ip_src);
   2694 #ifdef INET6
   2695 		} else if (af == AF_INET6) {
   2696 			const struct ip6_hdr *ip6 =
   2697 			    (const struct ip6_hdr *)decrypted_buf;
   2698 			IN6_PRINT(addrstr, &ip6->ip6_src);
   2699 #endif
   2700 		}
   2701 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2702 		    "%s: peer %s: invalid source address (%s)\n",
   2703 		    if_name(&wg->wg_if), wgp->wgp_name, addrstr);
   2704 		m_freem(n);
   2705 		/*
   2706 		 * The inner address is invalid however the session is valid
   2707 		 * so continue the session processing below.
   2708 		 */
   2709 	}
   2710 	n = NULL;
   2711 
   2712 	/* Update the state machine if necessary.  */
   2713 	if (__predict_false(state == WGS_STATE_INIT_PASSIVE)) {
   2714 		/*
   2715 		 * We were waiting for the initiator to send their
   2716 		 * first data transport message, and that has happened.
   2717 		 * Schedule a task to establish this session.
   2718 		 */
   2719 		wg_schedule_peer_task(wgp, WGP_TASK_ESTABLISH_SESSION);
   2720 	} else {
   2721 		if (__predict_false(wg_need_to_send_init_message(wgs))) {
   2722 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   2723 		}
   2724 		/*
   2725 		 * [W] 6.5 Passive Keepalive
   2726 		 * "If a peer has received a validly-authenticated transport
   2727 		 *  data message (section 5.4.6), but does not have any packets
   2728 		 *  itself to send back for KEEPALIVE-TIMEOUT seconds, it sends
   2729 		 *  a keepalive message."
   2730 		 */
   2731 		WG_DLOG("time_uptime=%ju wgs_time_last_data_sent=%ju\n",
   2732 		    (uintmax_t)time_uptime,
   2733 		    (uintmax_t)wgs->wgs_time_last_data_sent);
   2734 		if ((time_uptime - wgs->wgs_time_last_data_sent) >=
   2735 		    wg_keepalive_timeout) {
   2736 			WG_TRACE("Schedule sending keepalive message");
   2737 			/*
   2738 			 * We can't send a keepalive message here to avoid
   2739 			 * a deadlock;  we already hold the solock of a socket
   2740 			 * that is used to send the message.
   2741 			 */
   2742 			wg_schedule_peer_task(wgp,
   2743 			    WGP_TASK_SEND_KEEPALIVE_MESSAGE);
   2744 		}
   2745 	}
   2746 out:
   2747 	wg_put_session(wgs, &psref);
   2748 	if (m != NULL)
   2749 		m_freem(m);
   2750 	if (free_encrypted_buf)
   2751 		kmem_intr_free(encrypted_buf, encrypted_len);
   2752 }
   2753 
   2754 static void __noinline
   2755 wg_handle_msg_cookie(struct wg_softc *wg, const struct wg_msg_cookie *wgmc)
   2756 {
   2757 	struct wg_session *wgs;
   2758 	struct wg_peer *wgp;
   2759 	struct psref psref;
   2760 	int error;
   2761 	uint8_t key[WG_HASH_LEN];
   2762 	uint8_t cookie[WG_COOKIE_LEN];
   2763 
   2764 	WG_TRACE("cookie msg received");
   2765 
   2766 	/* Find the putative session.  */
   2767 	wgs = wg_lookup_session_by_index(wg, wgmc->wgmc_receiver, &psref);
   2768 	if (wgs == NULL) {
   2769 		WG_TRACE("No session found");
   2770 		return;
   2771 	}
   2772 
   2773 	/* Lock the peer so we can update the cookie state.  */
   2774 	wgp = wgs->wgs_peer;
   2775 	mutex_enter(wgp->wgp_lock);
   2776 
   2777 	if (!wgp->wgp_last_sent_mac1_valid) {
   2778 		WG_TRACE("No valid mac1 sent (or expired)");
   2779 		goto out;
   2780 	}
   2781 
   2782 	/* Decrypt the cookie and store it for later handshake retry.  */
   2783 	wg_algo_mac_cookie(key, sizeof(key), wgp->wgp_pubkey,
   2784 	    sizeof(wgp->wgp_pubkey));
   2785 	error = wg_algo_xaead_dec(cookie, sizeof(cookie), key,
   2786 	    wgmc->wgmc_cookie, sizeof(wgmc->wgmc_cookie),
   2787 	    wgp->wgp_last_sent_mac1, sizeof(wgp->wgp_last_sent_mac1),
   2788 	    wgmc->wgmc_salt);
   2789 	if (error != 0) {
   2790 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2791 		    "%s: peer %s: wg_algo_aead_dec for cookie failed: "
   2792 		    "error=%d\n", if_name(&wg->wg_if), wgp->wgp_name, error);
   2793 		goto out;
   2794 	}
   2795 	/*
   2796 	 * [W] 6.6: Interaction with Cookie Reply System
   2797 	 * "it should simply store the decrypted cookie value from the cookie
   2798 	 *  reply message, and wait for the expiration of the REKEY-TIMEOUT
   2799 	 *  timer for retrying a handshake initiation message."
   2800 	 */
   2801 	wgp->wgp_latest_cookie_time = time_uptime;
   2802 	memcpy(wgp->wgp_latest_cookie, cookie, sizeof(wgp->wgp_latest_cookie));
   2803 out:
   2804 	mutex_exit(wgp->wgp_lock);
   2805 	wg_put_session(wgs, &psref);
   2806 }
   2807 
   2808 static struct mbuf *
   2809 wg_validate_msg_header(struct wg_softc *wg, struct mbuf *m)
   2810 {
   2811 	struct wg_msg wgm;
   2812 	size_t mbuflen;
   2813 	size_t msglen;
   2814 
   2815 	/*
   2816 	 * Get the mbuf chain length.  It is already guaranteed, by
   2817 	 * wg_overudp_cb, to be large enough for a struct wg_msg.
   2818 	 */
   2819 	mbuflen = m_length(m);
   2820 	KASSERT(mbuflen >= sizeof(struct wg_msg));
   2821 
   2822 	/*
   2823 	 * Copy the message header (32-bit message type) out -- we'll
   2824 	 * worry about contiguity and alignment later.
   2825 	 */
   2826 	m_copydata(m, 0, sizeof(wgm), &wgm);
   2827 	switch (le32toh(wgm.wgm_type)) {
   2828 	case WG_MSG_TYPE_INIT:
   2829 		msglen = sizeof(struct wg_msg_init);
   2830 		break;
   2831 	case WG_MSG_TYPE_RESP:
   2832 		msglen = sizeof(struct wg_msg_resp);
   2833 		break;
   2834 	case WG_MSG_TYPE_COOKIE:
   2835 		msglen = sizeof(struct wg_msg_cookie);
   2836 		break;
   2837 	case WG_MSG_TYPE_DATA:
   2838 		msglen = sizeof(struct wg_msg_data);
   2839 		break;
   2840 	default:
   2841 		WG_LOG_RATECHECK(&wg->wg_ppsratecheck, LOG_DEBUG,
   2842 		    "%s: Unexpected msg type: %u\n", if_name(&wg->wg_if),
   2843 		    le32toh(wgm.wgm_type));
   2844 		goto error;
   2845 	}
   2846 
   2847 	/* Verify the mbuf chain is long enough for this type of message.  */
   2848 	if (__predict_false(mbuflen < msglen)) {
   2849 		WG_DLOG("Invalid msg size: mbuflen=%lu type=%u\n", mbuflen,
   2850 		    le32toh(wgm.wgm_type));
   2851 		goto error;
   2852 	}
   2853 
   2854 	/* Make the message header contiguous if necessary.  */
   2855 	if (__predict_false(m->m_len < msglen)) {
   2856 		m = m_pullup(m, msglen);
   2857 		if (m == NULL)
   2858 			return NULL;
   2859 	}
   2860 
   2861 	return m;
   2862 
   2863 error:
   2864 	m_freem(m);
   2865 	return NULL;
   2866 }
   2867 
   2868 static void
   2869 wg_handle_packet(struct wg_softc *wg, struct mbuf *m,
   2870     const struct sockaddr *src)
   2871 {
   2872 	struct wg_msg *wgm;
   2873 
   2874 	KASSERT(curlwp->l_pflag & LP_BOUND);
   2875 
   2876 	m = wg_validate_msg_header(wg, m);
   2877 	if (__predict_false(m == NULL))
   2878 		return;
   2879 
   2880 	KASSERT(m->m_len >= sizeof(struct wg_msg));
   2881 	wgm = mtod(m, struct wg_msg *);
   2882 	switch (le32toh(wgm->wgm_type)) {
   2883 	case WG_MSG_TYPE_INIT:
   2884 		wg_handle_msg_init(wg, (struct wg_msg_init *)wgm, src);
   2885 		break;
   2886 	case WG_MSG_TYPE_RESP:
   2887 		wg_handle_msg_resp(wg, (struct wg_msg_resp *)wgm, src);
   2888 		break;
   2889 	case WG_MSG_TYPE_COOKIE:
   2890 		wg_handle_msg_cookie(wg, (struct wg_msg_cookie *)wgm);
   2891 		break;
   2892 	case WG_MSG_TYPE_DATA:
   2893 		wg_handle_msg_data(wg, m, src);
   2894 		/* wg_handle_msg_data frees m for us */
   2895 		return;
   2896 	default:
   2897 		panic("invalid message type: %d", le32toh(wgm->wgm_type));
   2898 	}
   2899 
   2900 	m_freem(m);
   2901 }
   2902 
   2903 static void
   2904 wg_receive_packets(struct wg_softc *wg, const int af)
   2905 {
   2906 
   2907 	for (;;) {
   2908 		int error, flags;
   2909 		struct socket *so;
   2910 		struct mbuf *m = NULL;
   2911 		struct uio dummy_uio;
   2912 		struct mbuf *paddr = NULL;
   2913 		struct sockaddr *src;
   2914 
   2915 		so = wg_get_so_by_af(wg, af);
   2916 		flags = MSG_DONTWAIT;
   2917 		dummy_uio.uio_resid = 1000000000;
   2918 
   2919 		error = so->so_receive(so, &paddr, &dummy_uio, &m, NULL,
   2920 		    &flags);
   2921 		if (error || m == NULL) {
   2922 			//if (error == EWOULDBLOCK)
   2923 			return;
   2924 		}
   2925 
   2926 		KASSERT(paddr != NULL);
   2927 		KASSERT(paddr->m_len >= sizeof(struct sockaddr));
   2928 		src = mtod(paddr, struct sockaddr *);
   2929 
   2930 		wg_handle_packet(wg, m, src);
   2931 	}
   2932 }
   2933 
   2934 static void
   2935 wg_get_peer(struct wg_peer *wgp, struct psref *psref)
   2936 {
   2937 
   2938 	psref_acquire(psref, &wgp->wgp_psref, wg_psref_class);
   2939 }
   2940 
   2941 static void
   2942 wg_put_peer(struct wg_peer *wgp, struct psref *psref)
   2943 {
   2944 
   2945 	psref_release(psref, &wgp->wgp_psref, wg_psref_class);
   2946 }
   2947 
   2948 static void
   2949 wg_task_send_init_message(struct wg_softc *wg, struct wg_peer *wgp)
   2950 {
   2951 	struct wg_session *wgs;
   2952 
   2953 	WG_TRACE("WGP_TASK_SEND_INIT_MESSAGE");
   2954 
   2955 	KASSERT(mutex_owned(wgp->wgp_lock));
   2956 
   2957 	if (!atomic_load_acquire(&wgp->wgp_endpoint_available)) {
   2958 		WGLOG(LOG_DEBUG, "%s: No endpoint available\n",
   2959 		    if_name(&wg->wg_if));
   2960 		/* XXX should do something? */
   2961 		return;
   2962 	}
   2963 
   2964 	wgs = wgp->wgp_session_stable;
   2965 	if (wgs->wgs_state == WGS_STATE_UNKNOWN) {
   2966 		/* XXX What if the unstable session is already INIT_ACTIVE?  */
   2967 		wg_send_handshake_msg_init(wg, wgp);
   2968 	} else {
   2969 		/* rekey */
   2970 		wgs = wgp->wgp_session_unstable;
   2971 		if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE)
   2972 			wg_send_handshake_msg_init(wg, wgp);
   2973 	}
   2974 }
   2975 
   2976 static void
   2977 wg_task_retry_handshake(struct wg_softc *wg, struct wg_peer *wgp)
   2978 {
   2979 	struct wg_session *wgs;
   2980 
   2981 	WG_TRACE("WGP_TASK_RETRY_HANDSHAKE");
   2982 
   2983 	KASSERT(mutex_owned(wgp->wgp_lock));
   2984 	KASSERT(wgp->wgp_handshake_start_time != 0);
   2985 
   2986 	wgs = wgp->wgp_session_unstable;
   2987 	if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE)
   2988 		return;
   2989 
   2990 	/*
   2991 	 * XXX no real need to assign a new index here, but we do need
   2992 	 * to transition to UNKNOWN temporarily
   2993 	 */
   2994 	wg_put_session_index(wg, wgs);
   2995 
   2996 	/* [W] 6.4 Handshake Initiation Retransmission */
   2997 	if ((time_uptime - wgp->wgp_handshake_start_time) >
   2998 	    wg_rekey_attempt_time) {
   2999 		/* Give up handshaking */
   3000 		wgp->wgp_handshake_start_time = 0;
   3001 		WG_TRACE("give up");
   3002 
   3003 		/*
   3004 		 * If a new data packet comes, handshaking will be retried
   3005 		 * and a new session would be established at that time,
   3006 		 * however we don't want to send pending packets then.
   3007 		 */
   3008 		wg_purge_pending_packets(wgp);
   3009 		return;
   3010 	}
   3011 
   3012 	wg_task_send_init_message(wg, wgp);
   3013 }
   3014 
   3015 static void
   3016 wg_task_establish_session(struct wg_softc *wg, struct wg_peer *wgp)
   3017 {
   3018 	struct wg_session *wgs, *wgs_prev;
   3019 	struct mbuf *m;
   3020 
   3021 	KASSERT(mutex_owned(wgp->wgp_lock));
   3022 
   3023 	wgs = wgp->wgp_session_unstable;
   3024 	if (wgs->wgs_state != WGS_STATE_INIT_PASSIVE)
   3025 		/* XXX Can this happen?  */
   3026 		return;
   3027 
   3028 	wgs->wgs_state = WGS_STATE_ESTABLISHED;
   3029 	wgs->wgs_time_established = time_uptime;
   3030 	wgs->wgs_time_last_data_sent = 0;
   3031 	wgs->wgs_is_initiator = false;
   3032 	WG_TRACE("WGS_STATE_ESTABLISHED");
   3033 
   3034 	wg_swap_sessions(wgp);
   3035 	KASSERT(wgs == wgp->wgp_session_stable);
   3036 	wgs_prev = wgp->wgp_session_unstable;
   3037 	getnanotime(&wgp->wgp_last_handshake_time);
   3038 	wgp->wgp_handshake_start_time = 0;
   3039 	wgp->wgp_last_sent_mac1_valid = false;
   3040 	wgp->wgp_last_sent_cookie_valid = false;
   3041 
   3042 	/* If we had a data packet queued up, send it.  */
   3043 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL) {
   3044 		kpreempt_disable();
   3045 		const uint32_t h = curcpu()->ci_index; // pktq_rps_hash(m)
   3046 		M_SETCTX(m, wgp);
   3047 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   3048 			WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   3049 			    if_name(&wg->wg_if));
   3050 			m_freem(m);
   3051 		}
   3052 		kpreempt_enable();
   3053 	}
   3054 
   3055 	if (wgs_prev->wgs_state == WGS_STATE_ESTABLISHED) {
   3056 		/* Wait for wg_get_stable_session to drain.  */
   3057 		pserialize_perform(wgp->wgp_psz);
   3058 
   3059 		/* Transition ESTABLISHED->DESTROYING.  */
   3060 		wgs_prev->wgs_state = WGS_STATE_DESTROYING;
   3061 
   3062 		/* We can't destroy the old session immediately */
   3063 		wg_schedule_session_dtor_timer(wgp);
   3064 	} else {
   3065 		KASSERTMSG(wgs_prev->wgs_state == WGS_STATE_UNKNOWN,
   3066 		    "state=%d", wgs_prev->wgs_state);
   3067 		wg_clear_states(wgs_prev);
   3068 		wgs_prev->wgs_state = WGS_STATE_UNKNOWN;
   3069 	}
   3070 }
   3071 
   3072 static void
   3073 wg_task_endpoint_changed(struct wg_softc *wg, struct wg_peer *wgp)
   3074 {
   3075 
   3076 	WG_TRACE("WGP_TASK_ENDPOINT_CHANGED");
   3077 
   3078 	KASSERT(mutex_owned(wgp->wgp_lock));
   3079 
   3080 	if (atomic_load_relaxed(&wgp->wgp_endpoint_changing)) {
   3081 		pserialize_perform(wgp->wgp_psz);
   3082 		mutex_exit(wgp->wgp_lock);
   3083 		psref_target_destroy(&wgp->wgp_endpoint0->wgsa_psref,
   3084 		    wg_psref_class);
   3085 		psref_target_init(&wgp->wgp_endpoint0->wgsa_psref,
   3086 		    wg_psref_class);
   3087 		mutex_enter(wgp->wgp_lock);
   3088 		atomic_store_release(&wgp->wgp_endpoint_changing, 0);
   3089 	}
   3090 }
   3091 
   3092 static void
   3093 wg_task_send_keepalive_message(struct wg_softc *wg, struct wg_peer *wgp)
   3094 {
   3095 	struct wg_session *wgs;
   3096 
   3097 	WG_TRACE("WGP_TASK_SEND_KEEPALIVE_MESSAGE");
   3098 
   3099 	KASSERT(mutex_owned(wgp->wgp_lock));
   3100 
   3101 	wgs = wgp->wgp_session_stable;
   3102 	if (wgs->wgs_state != WGS_STATE_ESTABLISHED)
   3103 		return;
   3104 
   3105 	wg_send_keepalive_msg(wgp, wgs);
   3106 }
   3107 
   3108 static void
   3109 wg_task_destroy_prev_session(struct wg_softc *wg, struct wg_peer *wgp)
   3110 {
   3111 	struct wg_session *wgs;
   3112 
   3113 	WG_TRACE("WGP_TASK_DESTROY_PREV_SESSION");
   3114 
   3115 	KASSERT(mutex_owned(wgp->wgp_lock));
   3116 
   3117 	wgs = wgp->wgp_session_unstable;
   3118 	if (wgs->wgs_state == WGS_STATE_DESTROYING) {
   3119 		wg_put_session_index(wg, wgs);
   3120 	}
   3121 }
   3122 
   3123 static void
   3124 wg_peer_work(struct work *wk, void *cookie)
   3125 {
   3126 	struct wg_peer *wgp = container_of(wk, struct wg_peer, wgp_work);
   3127 	struct wg_softc *wg = wgp->wgp_sc;
   3128 	unsigned int tasks;
   3129 
   3130 	mutex_enter(wgp->wgp_intr_lock);
   3131 	while ((tasks = wgp->wgp_tasks) != 0) {
   3132 		wgp->wgp_tasks = 0;
   3133 		mutex_exit(wgp->wgp_intr_lock);
   3134 
   3135 		mutex_enter(wgp->wgp_lock);
   3136 		if (ISSET(tasks, WGP_TASK_SEND_INIT_MESSAGE))
   3137 			wg_task_send_init_message(wg, wgp);
   3138 		if (ISSET(tasks, WGP_TASK_RETRY_HANDSHAKE))
   3139 			wg_task_retry_handshake(wg, wgp);
   3140 		if (ISSET(tasks, WGP_TASK_ESTABLISH_SESSION))
   3141 			wg_task_establish_session(wg, wgp);
   3142 		if (ISSET(tasks, WGP_TASK_ENDPOINT_CHANGED))
   3143 			wg_task_endpoint_changed(wg, wgp);
   3144 		if (ISSET(tasks, WGP_TASK_SEND_KEEPALIVE_MESSAGE))
   3145 			wg_task_send_keepalive_message(wg, wgp);
   3146 		if (ISSET(tasks, WGP_TASK_DESTROY_PREV_SESSION))
   3147 			wg_task_destroy_prev_session(wg, wgp);
   3148 		mutex_exit(wgp->wgp_lock);
   3149 
   3150 		mutex_enter(wgp->wgp_intr_lock);
   3151 	}
   3152 	mutex_exit(wgp->wgp_intr_lock);
   3153 }
   3154 
   3155 static void
   3156 wg_job(struct threadpool_job *job)
   3157 {
   3158 	struct wg_softc *wg = container_of(job, struct wg_softc, wg_job);
   3159 	int bound, upcalls;
   3160 
   3161 	mutex_enter(wg->wg_intr_lock);
   3162 	while ((upcalls = wg->wg_upcalls) != 0) {
   3163 		wg->wg_upcalls = 0;
   3164 		mutex_exit(wg->wg_intr_lock);
   3165 		bound = curlwp_bind();
   3166 		if (ISSET(upcalls, WG_UPCALL_INET))
   3167 			wg_receive_packets(wg, AF_INET);
   3168 		if (ISSET(upcalls, WG_UPCALL_INET6))
   3169 			wg_receive_packets(wg, AF_INET6);
   3170 		curlwp_bindx(bound);
   3171 		mutex_enter(wg->wg_intr_lock);
   3172 	}
   3173 	threadpool_job_done(job);
   3174 	mutex_exit(wg->wg_intr_lock);
   3175 }
   3176 
   3177 static int
   3178 wg_bind_port(struct wg_softc *wg, const uint16_t port)
   3179 {
   3180 	int error;
   3181 	uint16_t old_port = wg->wg_listen_port;
   3182 
   3183 	if (port != 0 && old_port == port)
   3184 		return 0;
   3185 
   3186 	struct sockaddr_in _sin, *sin = &_sin;
   3187 	sin->sin_len = sizeof(*sin);
   3188 	sin->sin_family = AF_INET;
   3189 	sin->sin_addr.s_addr = INADDR_ANY;
   3190 	sin->sin_port = htons(port);
   3191 
   3192 	error = sobind(wg->wg_so4, sintosa(sin), curlwp);
   3193 	if (error != 0)
   3194 		return error;
   3195 
   3196 #ifdef INET6
   3197 	struct sockaddr_in6 _sin6, *sin6 = &_sin6;
   3198 	sin6->sin6_len = sizeof(*sin6);
   3199 	sin6->sin6_family = AF_INET6;
   3200 	sin6->sin6_addr = in6addr_any;
   3201 	sin6->sin6_port = htons(port);
   3202 
   3203 	error = sobind(wg->wg_so6, sin6tosa(sin6), curlwp);
   3204 	if (error != 0)
   3205 		return error;
   3206 #endif
   3207 
   3208 	wg->wg_listen_port = port;
   3209 
   3210 	return 0;
   3211 }
   3212 
   3213 static void
   3214 wg_so_upcall(struct socket *so, void *cookie, int events, int waitflag)
   3215 {
   3216 	struct wg_softc *wg = cookie;
   3217 	int reason;
   3218 
   3219 	reason = (so->so_proto->pr_domain->dom_family == AF_INET) ?
   3220 	    WG_UPCALL_INET :
   3221 	    WG_UPCALL_INET6;
   3222 
   3223 	mutex_enter(wg->wg_intr_lock);
   3224 	wg->wg_upcalls |= reason;
   3225 	threadpool_schedule_job(wg->wg_threadpool, &wg->wg_job);
   3226 	mutex_exit(wg->wg_intr_lock);
   3227 }
   3228 
   3229 static int
   3230 wg_overudp_cb(struct mbuf **mp, int offset, struct socket *so,
   3231     struct sockaddr *src, void *arg)
   3232 {
   3233 	struct wg_softc *wg = arg;
   3234 	struct wg_msg wgm;
   3235 	struct mbuf *m = *mp;
   3236 
   3237 	WG_TRACE("enter");
   3238 
   3239 	/* Verify the mbuf chain is long enough to have a wg msg header.  */
   3240 	KASSERT(offset <= m_length(m));
   3241 	if (__predict_false(m_length(m) - offset < sizeof(struct wg_msg))) {
   3242 		/* drop on the floor */
   3243 		m_freem(m);
   3244 		return -1;
   3245 	}
   3246 
   3247 	/*
   3248 	 * Copy the message header (32-bit message type) out -- we'll
   3249 	 * worry about contiguity and alignment later.
   3250 	 */
   3251 	m_copydata(m, offset, sizeof(struct wg_msg), &wgm);
   3252 	WG_DLOG("type=%d\n", le32toh(wgm.wgm_type));
   3253 
   3254 	/*
   3255 	 * Handle DATA packets promptly as they arrive.  Other packets
   3256 	 * may require expensive public-key crypto and are not as
   3257 	 * sensitive to latency, so defer them to the worker thread.
   3258 	 */
   3259 	switch (le32toh(wgm.wgm_type)) {
   3260 	case WG_MSG_TYPE_DATA:
   3261 		/* handle immediately */
   3262 		m_adj(m, offset);
   3263 		if (__predict_false(m->m_len < sizeof(struct wg_msg_data))) {
   3264 			m = m_pullup(m, sizeof(struct wg_msg_data));
   3265 			if (m == NULL)
   3266 				return -1;
   3267 		}
   3268 		wg_handle_msg_data(wg, m, src);
   3269 		*mp = NULL;
   3270 		return 1;
   3271 	case WG_MSG_TYPE_INIT:
   3272 	case WG_MSG_TYPE_RESP:
   3273 	case WG_MSG_TYPE_COOKIE:
   3274 		/* pass through to so_receive in wg_receive_packets */
   3275 		return 0;
   3276 	default:
   3277 		/* drop on the floor */
   3278 		m_freem(m);
   3279 		return -1;
   3280 	}
   3281 }
   3282 
   3283 static int
   3284 wg_socreate(struct wg_softc *wg, int af, struct socket **sop)
   3285 {
   3286 	int error;
   3287 	struct socket *so;
   3288 
   3289 	error = socreate(af, &so, SOCK_DGRAM, 0, curlwp, NULL);
   3290 	if (error != 0)
   3291 		return error;
   3292 
   3293 	solock(so);
   3294 	so->so_upcallarg = wg;
   3295 	so->so_upcall = wg_so_upcall;
   3296 	so->so_rcv.sb_flags |= SB_UPCALL;
   3297 	inpcb_register_overudp_cb(sotoinpcb(so), wg_overudp_cb, wg);
   3298 	sounlock(so);
   3299 
   3300 	*sop = so;
   3301 
   3302 	return 0;
   3303 }
   3304 
   3305 static bool
   3306 wg_session_hit_limits(struct wg_session *wgs)
   3307 {
   3308 
   3309 	/*
   3310 	 * [W] 6.2: Transport Message Limits
   3311 	 * "After REJECT-AFTER-MESSAGES transport data messages or after the
   3312 	 *  current secure session is REJECT-AFTER-TIME seconds old, whichever
   3313 	 *  comes first, WireGuard will refuse to send any more transport data
   3314 	 *  messages using the current secure session, ..."
   3315 	 */
   3316 	KASSERT(wgs->wgs_time_established != 0);
   3317 	if ((time_uptime - wgs->wgs_time_established) > wg_reject_after_time) {
   3318 		WG_DLOG("The session hits REJECT_AFTER_TIME\n");
   3319 		return true;
   3320 	} else if (wg_session_get_send_counter(wgs) >
   3321 	    wg_reject_after_messages) {
   3322 		WG_DLOG("The session hits REJECT_AFTER_MESSAGES\n");
   3323 		return true;
   3324 	}
   3325 
   3326 	return false;
   3327 }
   3328 
   3329 static void
   3330 wgintr(void *cookie)
   3331 {
   3332 	struct wg_peer *wgp;
   3333 	struct wg_session *wgs;
   3334 	struct mbuf *m;
   3335 	struct psref psref;
   3336 
   3337 	while ((m = pktq_dequeue(wg_pktq)) != NULL) {
   3338 		wgp = M_GETCTX(m, struct wg_peer *);
   3339 		if ((wgs = wg_get_stable_session(wgp, &psref)) == NULL) {
   3340 			WG_TRACE("no stable session");
   3341 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3342 			goto next0;
   3343 		}
   3344 		if (__predict_false(wg_session_hit_limits(wgs))) {
   3345 			WG_TRACE("stable session hit limits");
   3346 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3347 			goto next1;
   3348 		}
   3349 		wg_send_data_msg(wgp, wgs, m);
   3350 		m = NULL;	/* consumed */
   3351 next1:		wg_put_session(wgs, &psref);
   3352 next0:		if (m)
   3353 			m_freem(m);
   3354 		/* XXX Yield to avoid userland starvation?  */
   3355 	}
   3356 }
   3357 
   3358 static void
   3359 wg_rekey_timer(void *arg)
   3360 {
   3361 	struct wg_peer *wgp = arg;
   3362 
   3363 	wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3364 }
   3365 
   3366 static void
   3367 wg_purge_pending_packets(struct wg_peer *wgp)
   3368 {
   3369 	struct mbuf *m;
   3370 
   3371 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL)
   3372 		m_freem(m);
   3373 	pktq_barrier(wg_pktq);
   3374 }
   3375 
   3376 static void
   3377 wg_handshake_timeout_timer(void *arg)
   3378 {
   3379 	struct wg_peer *wgp = arg;
   3380 
   3381 	WG_TRACE("enter");
   3382 
   3383 	wg_schedule_peer_task(wgp, WGP_TASK_RETRY_HANDSHAKE);
   3384 }
   3385 
   3386 static struct wg_peer *
   3387 wg_alloc_peer(struct wg_softc *wg)
   3388 {
   3389 	struct wg_peer *wgp;
   3390 
   3391 	wgp = kmem_zalloc(sizeof(*wgp), KM_SLEEP);
   3392 
   3393 	wgp->wgp_sc = wg;
   3394 	callout_init(&wgp->wgp_rekey_timer, CALLOUT_MPSAFE);
   3395 	callout_setfunc(&wgp->wgp_rekey_timer, wg_rekey_timer, wgp);
   3396 	callout_init(&wgp->wgp_handshake_timeout_timer, CALLOUT_MPSAFE);
   3397 	callout_setfunc(&wgp->wgp_handshake_timeout_timer,
   3398 	    wg_handshake_timeout_timer, wgp);
   3399 	callout_init(&wgp->wgp_session_dtor_timer, CALLOUT_MPSAFE);
   3400 	callout_setfunc(&wgp->wgp_session_dtor_timer,
   3401 	    wg_session_dtor_timer, wgp);
   3402 	PSLIST_ENTRY_INIT(wgp, wgp_peerlist_entry);
   3403 	wgp->wgp_endpoint_changing = false;
   3404 	wgp->wgp_endpoint_available = false;
   3405 	wgp->wgp_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
   3406 	wgp->wgp_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
   3407 	wgp->wgp_psz = pserialize_create();
   3408 	psref_target_init(&wgp->wgp_psref, wg_psref_class);
   3409 
   3410 	wgp->wgp_endpoint = kmem_zalloc(sizeof(*wgp->wgp_endpoint), KM_SLEEP);
   3411 	wgp->wgp_endpoint0 = kmem_zalloc(sizeof(*wgp->wgp_endpoint0), KM_SLEEP);
   3412 	psref_target_init(&wgp->wgp_endpoint->wgsa_psref, wg_psref_class);
   3413 	psref_target_init(&wgp->wgp_endpoint0->wgsa_psref, wg_psref_class);
   3414 
   3415 	struct wg_session *wgs;
   3416 	wgp->wgp_session_stable =
   3417 	    kmem_zalloc(sizeof(*wgp->wgp_session_stable), KM_SLEEP);
   3418 	wgp->wgp_session_unstable =
   3419 	    kmem_zalloc(sizeof(*wgp->wgp_session_unstable), KM_SLEEP);
   3420 	wgs = wgp->wgp_session_stable;
   3421 	wgs->wgs_peer = wgp;
   3422 	wgs->wgs_state = WGS_STATE_UNKNOWN;
   3423 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
   3424 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3425 	mutex_init(&wgs->wgs_send_counter_lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3426 #endif
   3427 	wgs->wgs_recvwin = kmem_zalloc(sizeof(*wgs->wgs_recvwin), KM_SLEEP);
   3428 	mutex_init(&wgs->wgs_recvwin->lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3429 
   3430 	wgs = wgp->wgp_session_unstable;
   3431 	wgs->wgs_peer = wgp;
   3432 	wgs->wgs_state = WGS_STATE_UNKNOWN;
   3433 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
   3434 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3435 	mutex_init(&wgs->wgs_send_counter_lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3436 #endif
   3437 	wgs->wgs_recvwin = kmem_zalloc(sizeof(*wgs->wgs_recvwin), KM_SLEEP);
   3438 	mutex_init(&wgs->wgs_recvwin->lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3439 
   3440 	return wgp;
   3441 }
   3442 
   3443 static void
   3444 wg_destroy_peer(struct wg_peer *wgp)
   3445 {
   3446 	struct wg_session *wgs;
   3447 	struct wg_softc *wg = wgp->wgp_sc;
   3448 
   3449 	/* Prevent new packets from this peer on any source address.  */
   3450 	rw_enter(wg->wg_rwlock, RW_WRITER);
   3451 	for (int i = 0; i < wgp->wgp_n_allowedips; i++) {
   3452 		struct wg_allowedip *wga = &wgp->wgp_allowedips[i];
   3453 		struct radix_node_head *rnh = wg_rnh(wg, wga->wga_family);
   3454 		struct radix_node *rn;
   3455 
   3456 		KASSERT(rnh != NULL);
   3457 		rn = rnh->rnh_deladdr(&wga->wga_sa_addr,
   3458 		    &wga->wga_sa_mask, rnh);
   3459 		if (rn == NULL) {
   3460 			char addrstr[128];
   3461 			sockaddr_format(&wga->wga_sa_addr, addrstr,
   3462 			    sizeof(addrstr));
   3463 			WGLOG(LOG_WARNING, "%s: Couldn't delete %s",
   3464 			    if_name(&wg->wg_if), addrstr);
   3465 		}
   3466 	}
   3467 	rw_exit(wg->wg_rwlock);
   3468 
   3469 	/* Purge pending packets.  */
   3470 	wg_purge_pending_packets(wgp);
   3471 
   3472 	/* Halt all packet processing and timeouts.  */
   3473 	callout_halt(&wgp->wgp_rekey_timer, NULL);
   3474 	callout_halt(&wgp->wgp_handshake_timeout_timer, NULL);
   3475 	callout_halt(&wgp->wgp_session_dtor_timer, NULL);
   3476 
   3477 	/* Wait for any queued work to complete.  */
   3478 	workqueue_wait(wg_wq, &wgp->wgp_work);
   3479 
   3480 	wgs = wgp->wgp_session_unstable;
   3481 	if (wgs->wgs_state != WGS_STATE_UNKNOWN) {
   3482 		mutex_enter(wgp->wgp_lock);
   3483 		wg_destroy_session(wg, wgs);
   3484 		mutex_exit(wgp->wgp_lock);
   3485 	}
   3486 	mutex_destroy(&wgs->wgs_recvwin->lock);
   3487 	kmem_free(wgs->wgs_recvwin, sizeof(*wgs->wgs_recvwin));
   3488 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3489 	mutex_destroy(&wgs->wgs_send_counter_lock);
   3490 #endif
   3491 	kmem_free(wgs, sizeof(*wgs));
   3492 
   3493 	wgs = wgp->wgp_session_stable;
   3494 	if (wgs->wgs_state != WGS_STATE_UNKNOWN) {
   3495 		mutex_enter(wgp->wgp_lock);
   3496 		wg_destroy_session(wg, wgs);
   3497 		mutex_exit(wgp->wgp_lock);
   3498 	}
   3499 	mutex_destroy(&wgs->wgs_recvwin->lock);
   3500 	kmem_free(wgs->wgs_recvwin, sizeof(*wgs->wgs_recvwin));
   3501 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3502 	mutex_destroy(&wgs->wgs_send_counter_lock);
   3503 #endif
   3504 	kmem_free(wgs, sizeof(*wgs));
   3505 
   3506 	psref_target_destroy(&wgp->wgp_endpoint->wgsa_psref, wg_psref_class);
   3507 	psref_target_destroy(&wgp->wgp_endpoint0->wgsa_psref, wg_psref_class);
   3508 	kmem_free(wgp->wgp_endpoint, sizeof(*wgp->wgp_endpoint));
   3509 	kmem_free(wgp->wgp_endpoint0, sizeof(*wgp->wgp_endpoint0));
   3510 
   3511 	pserialize_destroy(wgp->wgp_psz);
   3512 	mutex_obj_free(wgp->wgp_intr_lock);
   3513 	mutex_obj_free(wgp->wgp_lock);
   3514 
   3515 	kmem_free(wgp, sizeof(*wgp));
   3516 }
   3517 
   3518 static void
   3519 wg_destroy_all_peers(struct wg_softc *wg)
   3520 {
   3521 	struct wg_peer *wgp, *wgp0 __diagused;
   3522 	void *garbage_byname, *garbage_bypubkey;
   3523 
   3524 restart:
   3525 	garbage_byname = garbage_bypubkey = NULL;
   3526 	mutex_enter(wg->wg_lock);
   3527 	WG_PEER_WRITER_FOREACH(wgp, wg) {
   3528 		if (wgp->wgp_name[0]) {
   3529 			wgp0 = thmap_del(wg->wg_peers_byname, wgp->wgp_name,
   3530 			    strlen(wgp->wgp_name));
   3531 			KASSERT(wgp0 == wgp);
   3532 			garbage_byname = thmap_stage_gc(wg->wg_peers_byname);
   3533 		}
   3534 		wgp0 = thmap_del(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   3535 		    sizeof(wgp->wgp_pubkey));
   3536 		KASSERT(wgp0 == wgp);
   3537 		garbage_bypubkey = thmap_stage_gc(wg->wg_peers_bypubkey);
   3538 		WG_PEER_WRITER_REMOVE(wgp);
   3539 		wg->wg_npeers--;
   3540 		mutex_enter(wgp->wgp_lock);
   3541 		pserialize_perform(wgp->wgp_psz);
   3542 		mutex_exit(wgp->wgp_lock);
   3543 		PSLIST_ENTRY_DESTROY(wgp, wgp_peerlist_entry);
   3544 		break;
   3545 	}
   3546 	mutex_exit(wg->wg_lock);
   3547 
   3548 	if (wgp == NULL)
   3549 		return;
   3550 
   3551 	psref_target_destroy(&wgp->wgp_psref, wg_psref_class);
   3552 
   3553 	wg_destroy_peer(wgp);
   3554 	thmap_gc(wg->wg_peers_byname, garbage_byname);
   3555 	thmap_gc(wg->wg_peers_bypubkey, garbage_bypubkey);
   3556 
   3557 	goto restart;
   3558 }
   3559 
   3560 static int
   3561 wg_destroy_peer_name(struct wg_softc *wg, const char *name)
   3562 {
   3563 	struct wg_peer *wgp, *wgp0 __diagused;
   3564 	void *garbage_byname, *garbage_bypubkey;
   3565 
   3566 	mutex_enter(wg->wg_lock);
   3567 	wgp = thmap_del(wg->wg_peers_byname, name, strlen(name));
   3568 	if (wgp != NULL) {
   3569 		wgp0 = thmap_del(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   3570 		    sizeof(wgp->wgp_pubkey));
   3571 		KASSERT(wgp0 == wgp);
   3572 		garbage_byname = thmap_stage_gc(wg->wg_peers_byname);
   3573 		garbage_bypubkey = thmap_stage_gc(wg->wg_peers_bypubkey);
   3574 		WG_PEER_WRITER_REMOVE(wgp);
   3575 		wg->wg_npeers--;
   3576 		if (wg->wg_npeers == 0)
   3577 			if_link_state_change(&wg->wg_if, LINK_STATE_DOWN);
   3578 		mutex_enter(wgp->wgp_lock);
   3579 		pserialize_perform(wgp->wgp_psz);
   3580 		mutex_exit(wgp->wgp_lock);
   3581 		PSLIST_ENTRY_DESTROY(wgp, wgp_peerlist_entry);
   3582 	}
   3583 	mutex_exit(wg->wg_lock);
   3584 
   3585 	if (wgp == NULL)
   3586 		return ENOENT;
   3587 
   3588 	psref_target_destroy(&wgp->wgp_psref, wg_psref_class);
   3589 
   3590 	wg_destroy_peer(wgp);
   3591 	thmap_gc(wg->wg_peers_byname, garbage_byname);
   3592 	thmap_gc(wg->wg_peers_bypubkey, garbage_bypubkey);
   3593 
   3594 	return 0;
   3595 }
   3596 
   3597 static int
   3598 wg_if_attach(struct wg_softc *wg)
   3599 {
   3600 
   3601 	wg->wg_if.if_addrlen = 0;
   3602 	wg->wg_if.if_mtu = WG_MTU;
   3603 	wg->wg_if.if_flags = IFF_MULTICAST;
   3604 	wg->wg_if.if_extflags = IFEF_MPSAFE;
   3605 	wg->wg_if.if_ioctl = wg_ioctl;
   3606 	wg->wg_if.if_output = wg_output;
   3607 	wg->wg_if.if_init = wg_init;
   3608 #ifdef ALTQ
   3609 	wg->wg_if.if_start = wg_start;
   3610 #endif
   3611 	wg->wg_if.if_stop = wg_stop;
   3612 	wg->wg_if.if_type = IFT_OTHER;
   3613 	wg->wg_if.if_dlt = DLT_NULL;
   3614 	wg->wg_if.if_softc = wg;
   3615 #ifdef ALTQ
   3616 	IFQ_SET_READY(&wg->wg_if.if_snd);
   3617 #endif
   3618 	if_initialize(&wg->wg_if);
   3619 
   3620 	wg->wg_if.if_link_state = LINK_STATE_DOWN;
   3621 	if_alloc_sadl(&wg->wg_if);
   3622 	if_register(&wg->wg_if);
   3623 
   3624 	bpf_attach(&wg->wg_if, DLT_NULL, sizeof(uint32_t));
   3625 
   3626 	return 0;
   3627 }
   3628 
   3629 static void
   3630 wg_if_detach(struct wg_softc *wg)
   3631 {
   3632 	struct ifnet *ifp = &wg->wg_if;
   3633 
   3634 	bpf_detach(ifp);
   3635 	if_detach(ifp);
   3636 }
   3637 
   3638 static int
   3639 wg_clone_create(struct if_clone *ifc, int unit)
   3640 {
   3641 	struct wg_softc *wg;
   3642 	int error;
   3643 
   3644 	wg_guarantee_initialized();
   3645 
   3646 	error = wg_count_inc();
   3647 	if (error)
   3648 		return error;
   3649 
   3650 	wg = kmem_zalloc(sizeof(*wg), KM_SLEEP);
   3651 
   3652 	if_initname(&wg->wg_if, ifc->ifc_name, unit);
   3653 
   3654 	PSLIST_INIT(&wg->wg_peers);
   3655 	wg->wg_peers_bypubkey = thmap_create(0, NULL, THMAP_NOCOPY);
   3656 	wg->wg_peers_byname = thmap_create(0, NULL, THMAP_NOCOPY);
   3657 	wg->wg_sessions_byindex = thmap_create(0, NULL, THMAP_NOCOPY);
   3658 	wg->wg_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
   3659 	wg->wg_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
   3660 	wg->wg_rwlock = rw_obj_alloc();
   3661 	threadpool_job_init(&wg->wg_job, wg_job, wg->wg_intr_lock,
   3662 	    "%s", if_name(&wg->wg_if));
   3663 	wg->wg_ops = &wg_ops_rumpkernel;
   3664 
   3665 	error = threadpool_get(&wg->wg_threadpool, PRI_NONE);
   3666 	if (error)
   3667 		goto fail0;
   3668 
   3669 #ifdef INET
   3670 	error = wg_socreate(wg, AF_INET, &wg->wg_so4);
   3671 	if (error)
   3672 		goto fail1;
   3673 	rn_inithead((void **)&wg->wg_rtable_ipv4,
   3674 	    offsetof(struct sockaddr_in, sin_addr) * NBBY);
   3675 #endif
   3676 #ifdef INET6
   3677 	error = wg_socreate(wg, AF_INET6, &wg->wg_so6);
   3678 	if (error)
   3679 		goto fail2;
   3680 	rn_inithead((void **)&wg->wg_rtable_ipv6,
   3681 	    offsetof(struct sockaddr_in6, sin6_addr) * NBBY);
   3682 #endif
   3683 
   3684 	error = wg_if_attach(wg);
   3685 	if (error)
   3686 		goto fail3;
   3687 
   3688 	return 0;
   3689 
   3690 fail4: __unused
   3691 	wg_if_detach(wg);
   3692 fail3:	wg_destroy_all_peers(wg);
   3693 #ifdef INET6
   3694 	solock(wg->wg_so6);
   3695 	wg->wg_so6->so_rcv.sb_flags &= ~SB_UPCALL;
   3696 	sounlock(wg->wg_so6);
   3697 #endif
   3698 #ifdef INET
   3699 	solock(wg->wg_so4);
   3700 	wg->wg_so4->so_rcv.sb_flags &= ~SB_UPCALL;
   3701 	sounlock(wg->wg_so4);
   3702 #endif
   3703 	mutex_enter(wg->wg_intr_lock);
   3704 	threadpool_cancel_job(wg->wg_threadpool, &wg->wg_job);
   3705 	mutex_exit(wg->wg_intr_lock);
   3706 #ifdef INET6
   3707 	if (wg->wg_rtable_ipv6 != NULL)
   3708 		free(wg->wg_rtable_ipv6, M_RTABLE);
   3709 	soclose(wg->wg_so6);
   3710 fail2:
   3711 #endif
   3712 #ifdef INET
   3713 	if (wg->wg_rtable_ipv4 != NULL)
   3714 		free(wg->wg_rtable_ipv4, M_RTABLE);
   3715 	soclose(wg->wg_so4);
   3716 fail1:
   3717 #endif
   3718 	threadpool_put(wg->wg_threadpool, PRI_NONE);
   3719 fail0:	threadpool_job_destroy(&wg->wg_job);
   3720 	rw_obj_free(wg->wg_rwlock);
   3721 	mutex_obj_free(wg->wg_intr_lock);
   3722 	mutex_obj_free(wg->wg_lock);
   3723 	thmap_destroy(wg->wg_sessions_byindex);
   3724 	thmap_destroy(wg->wg_peers_byname);
   3725 	thmap_destroy(wg->wg_peers_bypubkey);
   3726 	PSLIST_DESTROY(&wg->wg_peers);
   3727 	kmem_free(wg, sizeof(*wg));
   3728 	wg_count_dec();
   3729 	return error;
   3730 }
   3731 
   3732 static int
   3733 wg_clone_destroy(struct ifnet *ifp)
   3734 {
   3735 	struct wg_softc *wg = container_of(ifp, struct wg_softc, wg_if);
   3736 
   3737 #ifdef WG_RUMPKERNEL
   3738 	if (wg_user_mode(wg)) {
   3739 		rumpuser_wg_destroy(wg->wg_user);
   3740 		wg->wg_user = NULL;
   3741 	}
   3742 #endif
   3743 
   3744 	wg_if_detach(wg);
   3745 	wg_destroy_all_peers(wg);
   3746 #ifdef INET6
   3747 	solock(wg->wg_so6);
   3748 	wg->wg_so6->so_rcv.sb_flags &= ~SB_UPCALL;
   3749 	sounlock(wg->wg_so6);
   3750 #endif
   3751 #ifdef INET
   3752 	solock(wg->wg_so4);
   3753 	wg->wg_so4->so_rcv.sb_flags &= ~SB_UPCALL;
   3754 	sounlock(wg->wg_so4);
   3755 #endif
   3756 	mutex_enter(wg->wg_intr_lock);
   3757 	threadpool_cancel_job(wg->wg_threadpool, &wg->wg_job);
   3758 	mutex_exit(wg->wg_intr_lock);
   3759 #ifdef INET6
   3760 	if (wg->wg_rtable_ipv6 != NULL)
   3761 		free(wg->wg_rtable_ipv6, M_RTABLE);
   3762 	soclose(wg->wg_so6);
   3763 #endif
   3764 #ifdef INET
   3765 	if (wg->wg_rtable_ipv4 != NULL)
   3766 		free(wg->wg_rtable_ipv4, M_RTABLE);
   3767 	soclose(wg->wg_so4);
   3768 #endif
   3769 	threadpool_put(wg->wg_threadpool, PRI_NONE);
   3770 	threadpool_job_destroy(&wg->wg_job);
   3771 	rw_obj_free(wg->wg_rwlock);
   3772 	mutex_obj_free(wg->wg_intr_lock);
   3773 	mutex_obj_free(wg->wg_lock);
   3774 	thmap_destroy(wg->wg_sessions_byindex);
   3775 	thmap_destroy(wg->wg_peers_byname);
   3776 	thmap_destroy(wg->wg_peers_bypubkey);
   3777 	PSLIST_DESTROY(&wg->wg_peers);
   3778 	kmem_free(wg, sizeof(*wg));
   3779 	wg_count_dec();
   3780 
   3781 	return 0;
   3782 }
   3783 
   3784 static struct wg_peer *
   3785 wg_pick_peer_by_sa(struct wg_softc *wg, const struct sockaddr *sa,
   3786     struct psref *psref)
   3787 {
   3788 	struct radix_node_head *rnh;
   3789 	struct radix_node *rn;
   3790 	struct wg_peer *wgp = NULL;
   3791 	struct wg_allowedip *wga;
   3792 
   3793 #ifdef WG_DEBUG_LOG
   3794 	char addrstr[128];
   3795 	sockaddr_format(sa, addrstr, sizeof(addrstr));
   3796 	WG_DLOG("sa=%s\n", addrstr);
   3797 #endif
   3798 
   3799 	rw_enter(wg->wg_rwlock, RW_READER);
   3800 
   3801 	rnh = wg_rnh(wg, sa->sa_family);
   3802 	if (rnh == NULL)
   3803 		goto out;
   3804 
   3805 	rn = rnh->rnh_matchaddr(sa, rnh);
   3806 	if (rn == NULL || (rn->rn_flags & RNF_ROOT) != 0)
   3807 		goto out;
   3808 
   3809 	WG_TRACE("success");
   3810 
   3811 	wga = container_of(rn, struct wg_allowedip, wga_nodes[0]);
   3812 	wgp = wga->wga_peer;
   3813 	wg_get_peer(wgp, psref);
   3814 
   3815 out:
   3816 	rw_exit(wg->wg_rwlock);
   3817 	return wgp;
   3818 }
   3819 
   3820 static void
   3821 wg_fill_msg_data(struct wg_softc *wg, struct wg_peer *wgp,
   3822     struct wg_session *wgs, struct wg_msg_data *wgmd)
   3823 {
   3824 
   3825 	memset(wgmd, 0, sizeof(*wgmd));
   3826 	wgmd->wgmd_type = htole32(WG_MSG_TYPE_DATA);
   3827 	wgmd->wgmd_receiver = wgs->wgs_remote_index;
   3828 	/* [W] 5.4.6: msg.counter := Nm^send */
   3829 	/* [W] 5.4.6: Nm^send := Nm^send + 1 */
   3830 	wgmd->wgmd_counter = htole64(wg_session_inc_send_counter(wgs));
   3831 	WG_DLOG("counter=%"PRIu64"\n", le64toh(wgmd->wgmd_counter));
   3832 }
   3833 
   3834 static int
   3835 wg_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
   3836     const struct rtentry *rt)
   3837 {
   3838 	struct wg_softc *wg = ifp->if_softc;
   3839 	struct wg_peer *wgp = NULL;
   3840 	struct wg_session *wgs = NULL;
   3841 	struct psref wgp_psref, wgs_psref;
   3842 	int bound;
   3843 	int error;
   3844 
   3845 	bound = curlwp_bind();
   3846 
   3847 	/* TODO make the nest limit configurable via sysctl */
   3848 	error = if_tunnel_check_nesting(ifp, m, 1);
   3849 	if (error) {
   3850 		WGLOG(LOG_ERR,
   3851 		    "%s: tunneling loop detected and packet dropped\n",
   3852 		    if_name(&wg->wg_if));
   3853 		goto out0;
   3854 	}
   3855 
   3856 #ifdef ALTQ
   3857 	bool altq = atomic_load_relaxed(&ifp->if_snd.altq_flags)
   3858 	    & ALTQF_ENABLED;
   3859 	if (altq)
   3860 		IFQ_CLASSIFY(&ifp->if_snd, m, dst->sa_family);
   3861 #endif
   3862 
   3863 	bpf_mtap_af(ifp, dst->sa_family, m, BPF_D_OUT);
   3864 
   3865 	m->m_flags &= ~(M_BCAST|M_MCAST);
   3866 
   3867 	wgp = wg_pick_peer_by_sa(wg, dst, &wgp_psref);
   3868 	if (wgp == NULL) {
   3869 		WG_TRACE("peer not found");
   3870 		error = EHOSTUNREACH;
   3871 		goto out0;
   3872 	}
   3873 
   3874 	/* Clear checksum-offload flags. */
   3875 	m->m_pkthdr.csum_flags = 0;
   3876 	m->m_pkthdr.csum_data = 0;
   3877 
   3878 	/* Check whether there's an established session.  */
   3879 	wgs = wg_get_stable_session(wgp, &wgs_psref);
   3880 	if (wgs == NULL) {
   3881 		/*
   3882 		 * No established session.  If we're the first to try
   3883 		 * sending data, schedule a handshake and queue the
   3884 		 * packet for when the handshake is done; otherwise
   3885 		 * just drop the packet and let the ongoing handshake
   3886 		 * attempt continue.  We could queue more data packets
   3887 		 * but it's not clear that's worthwhile.
   3888 		 */
   3889 		if (atomic_cas_ptr(&wgp->wgp_pending, NULL, m) == NULL) {
   3890 			m = NULL; /* consume */
   3891 			WG_TRACE("queued first packet; init handshake");
   3892 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3893 		} else {
   3894 			WG_TRACE("first packet already queued, dropping");
   3895 		}
   3896 		goto out1;
   3897 	}
   3898 
   3899 	/* There's an established session.  Toss it in the queue.  */
   3900 #ifdef ALTQ
   3901 	if (altq) {
   3902 		mutex_enter(ifp->if_snd.ifq_lock);
   3903 		if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
   3904 			M_SETCTX(m, wgp);
   3905 			ALTQ_ENQUEUE(&ifp->if_snd, m, error);
   3906 			m = NULL; /* consume */
   3907 		}
   3908 		mutex_exit(ifp->if_snd.ifq_lock);
   3909 		if (m == NULL) {
   3910 			wg_start(ifp);
   3911 			goto out2;
   3912 		}
   3913 	}
   3914 #endif
   3915 	kpreempt_disable();
   3916 	const uint32_t h = curcpu()->ci_index;	// pktq_rps_hash(m)
   3917 	M_SETCTX(m, wgp);
   3918 	if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   3919 		WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   3920 		    if_name(&wg->wg_if));
   3921 		error = ENOBUFS;
   3922 		goto out3;
   3923 	}
   3924 	m = NULL;		/* consumed */
   3925 	error = 0;
   3926 out3:	kpreempt_enable();
   3927 
   3928 #ifdef ALTQ
   3929 out2:
   3930 #endif
   3931 	wg_put_session(wgs, &wgs_psref);
   3932 out1:	wg_put_peer(wgp, &wgp_psref);
   3933 out0:	if (m)
   3934 		m_freem(m);
   3935 	curlwp_bindx(bound);
   3936 	return error;
   3937 }
   3938 
   3939 static int
   3940 wg_send_udp(struct wg_peer *wgp, struct mbuf *m)
   3941 {
   3942 	struct psref psref;
   3943 	struct wg_sockaddr *wgsa;
   3944 	int error;
   3945 	struct socket *so;
   3946 
   3947 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   3948 	so = wg_get_so_by_peer(wgp, wgsa);
   3949 	solock(so);
   3950 	if (wgsatosa(wgsa)->sa_family == AF_INET) {
   3951 		error = udp_send(so, m, wgsatosa(wgsa), NULL, curlwp);
   3952 	} else {
   3953 #ifdef INET6
   3954 		error = udp6_output(sotoinpcb(so), m, wgsatosin6(wgsa),
   3955 		    NULL, curlwp);
   3956 #else
   3957 		m_freem(m);
   3958 		error = EPFNOSUPPORT;
   3959 #endif
   3960 	}
   3961 	sounlock(so);
   3962 	wg_put_sa(wgp, wgsa, &psref);
   3963 
   3964 	return error;
   3965 }
   3966 
   3967 /* Inspired by pppoe_get_mbuf */
   3968 static struct mbuf *
   3969 wg_get_mbuf(size_t leading_len, size_t len)
   3970 {
   3971 	struct mbuf *m;
   3972 
   3973 	KASSERT(leading_len <= MCLBYTES);
   3974 	KASSERT(len <= MCLBYTES - leading_len);
   3975 
   3976 	m = m_gethdr(M_DONTWAIT, MT_DATA);
   3977 	if (m == NULL)
   3978 		return NULL;
   3979 	if (len + leading_len > MHLEN) {
   3980 		m_clget(m, M_DONTWAIT);
   3981 		if ((m->m_flags & M_EXT) == 0) {
   3982 			m_free(m);
   3983 			return NULL;
   3984 		}
   3985 	}
   3986 	m->m_data += leading_len;
   3987 	m->m_pkthdr.len = m->m_len = len;
   3988 
   3989 	return m;
   3990 }
   3991 
   3992 static int
   3993 wg_send_data_msg(struct wg_peer *wgp, struct wg_session *wgs,
   3994     struct mbuf *m)
   3995 {
   3996 	struct wg_softc *wg = wgp->wgp_sc;
   3997 	int error;
   3998 	size_t inner_len, padded_len, encrypted_len;
   3999 	char *padded_buf = NULL;
   4000 	size_t mlen;
   4001 	struct wg_msg_data *wgmd;
   4002 	bool free_padded_buf = false;
   4003 	struct mbuf *n;
   4004 	size_t leading_len = max_hdr + sizeof(struct udphdr);
   4005 
   4006 	mlen = m_length(m);
   4007 	inner_len = mlen;
   4008 	padded_len = roundup(mlen, 16);
   4009 	encrypted_len = padded_len + WG_AUTHTAG_LEN;
   4010 	WG_DLOG("inner=%lu, padded=%lu, encrypted_len=%lu\n",
   4011 	    inner_len, padded_len, encrypted_len);
   4012 	if (mlen != 0) {
   4013 		bool success;
   4014 		success = m_ensure_contig(&m, padded_len);
   4015 		if (success) {
   4016 			padded_buf = mtod(m, char *);
   4017 		} else {
   4018 			padded_buf = kmem_intr_alloc(padded_len, KM_NOSLEEP);
   4019 			if (padded_buf == NULL) {
   4020 				error = ENOBUFS;
   4021 				goto end;
   4022 			}
   4023 			free_padded_buf = true;
   4024 			m_copydata(m, 0, mlen, padded_buf);
   4025 		}
   4026 		memset(padded_buf + mlen, 0, padded_len - inner_len);
   4027 	}
   4028 
   4029 	n = wg_get_mbuf(leading_len, sizeof(*wgmd) + encrypted_len);
   4030 	if (n == NULL) {
   4031 		error = ENOBUFS;
   4032 		goto end;
   4033 	}
   4034 	KASSERT(n->m_len >= sizeof(*wgmd));
   4035 	wgmd = mtod(n, struct wg_msg_data *);
   4036 	wg_fill_msg_data(wg, wgp, wgs, wgmd);
   4037 	/* [W] 5.4.6: AEAD(Tm^send, Nm^send, P, e) */
   4038 	wg_algo_aead_enc((char *)wgmd + sizeof(*wgmd), encrypted_len,
   4039 	    wgs->wgs_tkey_send, le64toh(wgmd->wgmd_counter),
   4040 	    padded_buf, padded_len,
   4041 	    NULL, 0);
   4042 
   4043 	error = wg->wg_ops->send_data_msg(wgp, n);
   4044 	if (error == 0) {
   4045 		struct ifnet *ifp = &wg->wg_if;
   4046 		if_statadd(ifp, if_obytes, mlen);
   4047 		if_statinc(ifp, if_opackets);
   4048 		if (wgs->wgs_is_initiator &&
   4049 		    wgs->wgs_time_last_data_sent == 0) {
   4050 			/*
   4051 			 * [W] 6.2 Transport Message Limits
   4052 			 * "if a peer is the initiator of a current secure
   4053 			 *  session, WireGuard will send a handshake initiation
   4054 			 *  message to begin a new secure session if, after
   4055 			 *  transmitting a transport data message, the current
   4056 			 *  secure session is REKEY-AFTER-TIME seconds old,"
   4057 			 */
   4058 			wg_schedule_rekey_timer(wgp);
   4059 		}
   4060 		wgs->wgs_time_last_data_sent = time_uptime;
   4061 		if (wg_session_get_send_counter(wgs) >=
   4062 		    wg_rekey_after_messages) {
   4063 			/*
   4064 			 * [W] 6.2 Transport Message Limits
   4065 			 * "WireGuard will try to create a new session, by
   4066 			 *  sending a handshake initiation message (section
   4067 			 *  5.4.2), after it has sent REKEY-AFTER-MESSAGES
   4068 			 *  transport data messages..."
   4069 			 */
   4070 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   4071 		}
   4072 	}
   4073 end:
   4074 	m_freem(m);
   4075 	if (free_padded_buf)
   4076 		kmem_intr_free(padded_buf, padded_len);
   4077 	return error;
   4078 }
   4079 
   4080 static void
   4081 wg_input(struct ifnet *ifp, struct mbuf *m, const int af)
   4082 {
   4083 	pktqueue_t *pktq;
   4084 	size_t pktlen;
   4085 
   4086 	KASSERT(af == AF_INET || af == AF_INET6);
   4087 
   4088 	WG_TRACE("");
   4089 
   4090 	m_set_rcvif(m, ifp);
   4091 	pktlen = m->m_pkthdr.len;
   4092 
   4093 	bpf_mtap_af(ifp, af, m, BPF_D_IN);
   4094 
   4095 	switch (af) {
   4096 	case AF_INET:
   4097 		pktq = ip_pktq;
   4098 		break;
   4099 #ifdef INET6
   4100 	case AF_INET6:
   4101 		pktq = ip6_pktq;
   4102 		break;
   4103 #endif
   4104 	default:
   4105 		panic("invalid af=%d", af);
   4106 	}
   4107 
   4108 	kpreempt_disable();
   4109 	const u_int h = curcpu()->ci_index;
   4110 	if (__predict_true(pktq_enqueue(pktq, m, h))) {
   4111 		if_statadd(ifp, if_ibytes, pktlen);
   4112 		if_statinc(ifp, if_ipackets);
   4113 	} else {
   4114 		m_freem(m);
   4115 	}
   4116 	kpreempt_enable();
   4117 }
   4118 
   4119 static void
   4120 wg_calc_pubkey(uint8_t pubkey[WG_STATIC_KEY_LEN],
   4121     const uint8_t privkey[WG_STATIC_KEY_LEN])
   4122 {
   4123 
   4124 	crypto_scalarmult_base(pubkey, privkey);
   4125 }
   4126 
   4127 static int
   4128 wg_rtable_add_route(struct wg_softc *wg, struct wg_allowedip *wga)
   4129 {
   4130 	struct radix_node_head *rnh;
   4131 	struct radix_node *rn;
   4132 	int error = 0;
   4133 
   4134 	rw_enter(wg->wg_rwlock, RW_WRITER);
   4135 	rnh = wg_rnh(wg, wga->wga_family);
   4136 	KASSERT(rnh != NULL);
   4137 	rn = rnh->rnh_addaddr(&wga->wga_sa_addr, &wga->wga_sa_mask, rnh,
   4138 	    wga->wga_nodes);
   4139 	rw_exit(wg->wg_rwlock);
   4140 
   4141 	if (rn == NULL)
   4142 		error = EEXIST;
   4143 
   4144 	return error;
   4145 }
   4146 
   4147 static int
   4148 wg_handle_prop_peer(struct wg_softc *wg, prop_dictionary_t peer,
   4149     struct wg_peer **wgpp)
   4150 {
   4151 	int error = 0;
   4152 	const void *pubkey;
   4153 	size_t pubkey_len;
   4154 	const void *psk;
   4155 	size_t psk_len;
   4156 	const char *name = NULL;
   4157 
   4158 	if (prop_dictionary_get_string(peer, "name", &name)) {
   4159 		if (strlen(name) > WG_PEER_NAME_MAXLEN) {
   4160 			error = EINVAL;
   4161 			goto out;
   4162 		}
   4163 	}
   4164 
   4165 	if (!prop_dictionary_get_data(peer, "public_key",
   4166 		&pubkey, &pubkey_len)) {
   4167 		error = EINVAL;
   4168 		goto out;
   4169 	}
   4170 #ifdef WG_DEBUG_DUMP
   4171     {
   4172 	char *hex = gethexdump(pubkey, pubkey_len);
   4173 	log(LOG_DEBUG, "pubkey=%p, pubkey_len=%lu\n%s\n",
   4174 	    pubkey, pubkey_len, hex);
   4175 	puthexdump(hex, pubkey, pubkey_len);
   4176     }
   4177 #endif
   4178 
   4179 	struct wg_peer *wgp = wg_alloc_peer(wg);
   4180 	memcpy(wgp->wgp_pubkey, pubkey, sizeof(wgp->wgp_pubkey));
   4181 	if (name != NULL)
   4182 		strncpy(wgp->wgp_name, name, sizeof(wgp->wgp_name));
   4183 
   4184 	if (prop_dictionary_get_data(peer, "preshared_key", &psk, &psk_len)) {
   4185 		if (psk_len != sizeof(wgp->wgp_psk)) {
   4186 			error = EINVAL;
   4187 			goto out;
   4188 		}
   4189 		memcpy(wgp->wgp_psk, psk, sizeof(wgp->wgp_psk));
   4190 	}
   4191 
   4192 	const void *addr;
   4193 	size_t addr_len;
   4194 	struct wg_sockaddr *wgsa = wgp->wgp_endpoint;
   4195 
   4196 	if (!prop_dictionary_get_data(peer, "endpoint", &addr, &addr_len))
   4197 		goto skip_endpoint;
   4198 	if (addr_len < sizeof(*wgsatosa(wgsa)) ||
   4199 	    addr_len > sizeof(*wgsatoss(wgsa))) {
   4200 		error = EINVAL;
   4201 		goto out;
   4202 	}
   4203 	memcpy(wgsatoss(wgsa), addr, addr_len);
   4204 	switch (wgsa_family(wgsa)) {
   4205 	case AF_INET:
   4206 #ifdef INET6
   4207 	case AF_INET6:
   4208 #endif
   4209 		break;
   4210 	default:
   4211 		error = EPFNOSUPPORT;
   4212 		goto out;
   4213 	}
   4214 	if (addr_len != sockaddr_getsize_by_family(wgsa_family(wgsa))) {
   4215 		error = EINVAL;
   4216 		goto out;
   4217 	}
   4218     {
   4219 	char addrstr[128];
   4220 	sockaddr_format(wgsatosa(wgsa), addrstr, sizeof(addrstr));
   4221 	WG_DLOG("addr=%s\n", addrstr);
   4222     }
   4223 	wgp->wgp_endpoint_available = true;
   4224 
   4225 	prop_array_t allowedips;
   4226 skip_endpoint:
   4227 	allowedips = prop_dictionary_get(peer, "allowedips");
   4228 	if (allowedips == NULL)
   4229 		goto skip;
   4230 
   4231 	prop_object_iterator_t _it = prop_array_iterator(allowedips);
   4232 	prop_dictionary_t prop_allowedip;
   4233 	int j = 0;
   4234 	while ((prop_allowedip = prop_object_iterator_next(_it)) != NULL) {
   4235 		struct wg_allowedip *wga = &wgp->wgp_allowedips[j];
   4236 
   4237 		if (!prop_dictionary_get_int(prop_allowedip, "family",
   4238 			&wga->wga_family))
   4239 			continue;
   4240 		if (!prop_dictionary_get_data(prop_allowedip, "ip",
   4241 			&addr, &addr_len))
   4242 			continue;
   4243 		if (!prop_dictionary_get_uint8(prop_allowedip, "cidr",
   4244 			&wga->wga_cidr))
   4245 			continue;
   4246 
   4247 		switch (wga->wga_family) {
   4248 		case AF_INET: {
   4249 			struct sockaddr_in sin;
   4250 			char addrstr[128];
   4251 			struct in_addr mask;
   4252 			struct sockaddr_in sin_mask;
   4253 
   4254 			if (addr_len != sizeof(struct in_addr))
   4255 				return EINVAL;
   4256 			memcpy(&wga->wga_addr4, addr, addr_len);
   4257 
   4258 			sockaddr_in_init(&sin, (const struct in_addr *)addr,
   4259 			    0);
   4260 			sockaddr_copy(&wga->wga_sa_addr,
   4261 			    sizeof(sin), sintosa(&sin));
   4262 
   4263 			sockaddr_format(sintosa(&sin),
   4264 			    addrstr, sizeof(addrstr));
   4265 			WG_DLOG("addr=%s/%d\n", addrstr, wga->wga_cidr);
   4266 
   4267 			in_len2mask(&mask, wga->wga_cidr);
   4268 			sockaddr_in_init(&sin_mask, &mask, 0);
   4269 			sockaddr_copy(&wga->wga_sa_mask,
   4270 			    sizeof(sin_mask), sintosa(&sin_mask));
   4271 
   4272 			break;
   4273 		    }
   4274 #ifdef INET6
   4275 		case AF_INET6: {
   4276 			struct sockaddr_in6 sin6;
   4277 			char addrstr[128];
   4278 			struct in6_addr mask;
   4279 			struct sockaddr_in6 sin6_mask;
   4280 
   4281 			if (addr_len != sizeof(struct in6_addr))
   4282 				return EINVAL;
   4283 			memcpy(&wga->wga_addr6, addr, addr_len);
   4284 
   4285 			sockaddr_in6_init(&sin6, (const struct in6_addr *)addr,
   4286 			    0, 0, 0);
   4287 			sockaddr_copy(&wga->wga_sa_addr,
   4288 			    sizeof(sin6), sin6tosa(&sin6));
   4289 
   4290 			sockaddr_format(sin6tosa(&sin6),
   4291 			    addrstr, sizeof(addrstr));
   4292 			WG_DLOG("addr=%s/%d\n", addrstr, wga->wga_cidr);
   4293 
   4294 			in6_prefixlen2mask(&mask, wga->wga_cidr);
   4295 			sockaddr_in6_init(&sin6_mask, &mask, 0, 0, 0);
   4296 			sockaddr_copy(&wga->wga_sa_mask,
   4297 			    sizeof(sin6_mask), sin6tosa(&sin6_mask));
   4298 
   4299 			break;
   4300 		    }
   4301 #endif
   4302 		default:
   4303 			error = EINVAL;
   4304 			goto out;
   4305 		}
   4306 		wga->wga_peer = wgp;
   4307 
   4308 		error = wg_rtable_add_route(wg, wga);
   4309 		if (error != 0)
   4310 			goto out;
   4311 
   4312 		j++;
   4313 	}
   4314 	wgp->wgp_n_allowedips = j;
   4315 skip:
   4316 	*wgpp = wgp;
   4317 out:
   4318 	return error;
   4319 }
   4320 
   4321 static int
   4322 wg_alloc_prop_buf(char **_buf, struct ifdrv *ifd)
   4323 {
   4324 	int error;
   4325 	char *buf;
   4326 
   4327 	WG_DLOG("buf=%p, len=%lu\n", ifd->ifd_data, ifd->ifd_len);
   4328 	if (ifd->ifd_len >= WG_MAX_PROPLEN)
   4329 		return E2BIG;
   4330 	buf = kmem_alloc(ifd->ifd_len + 1, KM_SLEEP);
   4331 	error = copyin(ifd->ifd_data, buf, ifd->ifd_len);
   4332 	if (error != 0)
   4333 		return error;
   4334 	buf[ifd->ifd_len] = '\0';
   4335 #ifdef WG_DEBUG_DUMP
   4336 	log(LOG_DEBUG, "%.*s\n",
   4337 	    (int)MIN(INT_MAX, ifd->ifd_len),
   4338 	    (const char *)buf);
   4339 #endif
   4340 	*_buf = buf;
   4341 	return 0;
   4342 }
   4343 
   4344 static int
   4345 wg_ioctl_set_private_key(struct wg_softc *wg, struct ifdrv *ifd)
   4346 {
   4347 	int error;
   4348 	prop_dictionary_t prop_dict;
   4349 	char *buf = NULL;
   4350 	const void *privkey;
   4351 	size_t privkey_len;
   4352 
   4353 	error = wg_alloc_prop_buf(&buf, ifd);
   4354 	if (error != 0)
   4355 		return error;
   4356 	error = EINVAL;
   4357 	prop_dict = prop_dictionary_internalize(buf);
   4358 	if (prop_dict == NULL)
   4359 		goto out;
   4360 	if (!prop_dictionary_get_data(prop_dict, "private_key",
   4361 		&privkey, &privkey_len))
   4362 		goto out;
   4363 #ifdef WG_DEBUG_DUMP
   4364     {
   4365 	char *hex = gethexdump(privkey, privkey_len);
   4366 	log(LOG_DEBUG, "privkey=%p, privkey_len=%lu\n%s\n",
   4367 	    privkey, privkey_len, hex);
   4368 	puthexdump(hex, privkey, privkey_len);
   4369     }
   4370 #endif
   4371 	if (privkey_len != WG_STATIC_KEY_LEN)
   4372 		goto out;
   4373 	memcpy(wg->wg_privkey, privkey, WG_STATIC_KEY_LEN);
   4374 	wg_calc_pubkey(wg->wg_pubkey, wg->wg_privkey);
   4375 	error = 0;
   4376 
   4377 out:
   4378 	kmem_free(buf, ifd->ifd_len + 1);
   4379 	return error;
   4380 }
   4381 
   4382 static int
   4383 wg_ioctl_set_listen_port(struct wg_softc *wg, struct ifdrv *ifd)
   4384 {
   4385 	int error;
   4386 	prop_dictionary_t prop_dict;
   4387 	char *buf = NULL;
   4388 	uint16_t port;
   4389 
   4390 	error = wg_alloc_prop_buf(&buf, ifd);
   4391 	if (error != 0)
   4392 		return error;
   4393 	error = EINVAL;
   4394 	prop_dict = prop_dictionary_internalize(buf);
   4395 	if (prop_dict == NULL)
   4396 		goto out;
   4397 	if (!prop_dictionary_get_uint16(prop_dict, "listen_port", &port))
   4398 		goto out;
   4399 
   4400 	error = wg->wg_ops->bind_port(wg, (uint16_t)port);
   4401 
   4402 out:
   4403 	kmem_free(buf, ifd->ifd_len + 1);
   4404 	return error;
   4405 }
   4406 
   4407 static int
   4408 wg_ioctl_add_peer(struct wg_softc *wg, struct ifdrv *ifd)
   4409 {
   4410 	int error;
   4411 	prop_dictionary_t prop_dict;
   4412 	char *buf = NULL;
   4413 	struct wg_peer *wgp = NULL, *wgp0 __diagused;
   4414 
   4415 	error = wg_alloc_prop_buf(&buf, ifd);
   4416 	if (error != 0)
   4417 		return error;
   4418 	error = EINVAL;
   4419 	prop_dict = prop_dictionary_internalize(buf);
   4420 	if (prop_dict == NULL)
   4421 		goto out;
   4422 
   4423 	error = wg_handle_prop_peer(wg, prop_dict, &wgp);
   4424 	if (error != 0)
   4425 		goto out;
   4426 
   4427 	mutex_enter(wg->wg_lock);
   4428 	if (thmap_get(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   4429 		sizeof(wgp->wgp_pubkey)) != NULL ||
   4430 	    (wgp->wgp_name[0] &&
   4431 		thmap_get(wg->wg_peers_byname, wgp->wgp_name,
   4432 		    strlen(wgp->wgp_name)) != NULL)) {
   4433 		mutex_exit(wg->wg_lock);
   4434 		wg_destroy_peer(wgp);
   4435 		error = EEXIST;
   4436 		goto out;
   4437 	}
   4438 	wgp0 = thmap_put(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   4439 	    sizeof(wgp->wgp_pubkey), wgp);
   4440 	KASSERT(wgp0 == wgp);
   4441 	if (wgp->wgp_name[0]) {
   4442 		wgp0 = thmap_put(wg->wg_peers_byname, wgp->wgp_name,
   4443 		    strlen(wgp->wgp_name), wgp);
   4444 		KASSERT(wgp0 == wgp);
   4445 	}
   4446 	WG_PEER_WRITER_INSERT_HEAD(wgp, wg);
   4447 	wg->wg_npeers++;
   4448 	mutex_exit(wg->wg_lock);
   4449 
   4450 	if_link_state_change(&wg->wg_if, LINK_STATE_UP);
   4451 
   4452 out:
   4453 	kmem_free(buf, ifd->ifd_len + 1);
   4454 	return error;
   4455 }
   4456 
   4457 static int
   4458 wg_ioctl_delete_peer(struct wg_softc *wg, struct ifdrv *ifd)
   4459 {
   4460 	int error;
   4461 	prop_dictionary_t prop_dict;
   4462 	char *buf = NULL;
   4463 	const char *name;
   4464 
   4465 	error = wg_alloc_prop_buf(&buf, ifd);
   4466 	if (error != 0)
   4467 		return error;
   4468 	error = EINVAL;
   4469 	prop_dict = prop_dictionary_internalize(buf);
   4470 	if (prop_dict == NULL)
   4471 		goto out;
   4472 
   4473 	if (!prop_dictionary_get_string(prop_dict, "name", &name))
   4474 		goto out;
   4475 	if (strlen(name) > WG_PEER_NAME_MAXLEN)
   4476 		goto out;
   4477 
   4478 	error = wg_destroy_peer_name(wg, name);
   4479 out:
   4480 	kmem_free(buf, ifd->ifd_len + 1);
   4481 	return error;
   4482 }
   4483 
   4484 static bool
   4485 wg_is_authorized(struct wg_softc *wg, u_long cmd)
   4486 {
   4487 	int au = cmd == SIOCGDRVSPEC ?
   4488 	    KAUTH_REQ_NETWORK_INTERFACE_WG_GETPRIV :
   4489 	    KAUTH_REQ_NETWORK_INTERFACE_WG_SETPRIV;
   4490 	return kauth_authorize_network(kauth_cred_get(),
   4491 	    KAUTH_NETWORK_INTERFACE_WG, au, &wg->wg_if,
   4492 	    (void *)cmd, NULL) == 0;
   4493 }
   4494 
   4495 static int
   4496 wg_ioctl_get(struct wg_softc *wg, struct ifdrv *ifd)
   4497 {
   4498 	int error = ENOMEM;
   4499 	prop_dictionary_t prop_dict;
   4500 	prop_array_t peers = NULL;
   4501 	char *buf;
   4502 	struct wg_peer *wgp;
   4503 	int s, i;
   4504 
   4505 	prop_dict = prop_dictionary_create();
   4506 	if (prop_dict == NULL)
   4507 		goto error;
   4508 
   4509 	if (wg_is_authorized(wg, SIOCGDRVSPEC)) {
   4510 		if (!prop_dictionary_set_data(prop_dict, "private_key",
   4511 			wg->wg_privkey, WG_STATIC_KEY_LEN))
   4512 			goto error;
   4513 	}
   4514 
   4515 	if (wg->wg_listen_port != 0) {
   4516 		if (!prop_dictionary_set_uint16(prop_dict, "listen_port",
   4517 			wg->wg_listen_port))
   4518 			goto error;
   4519 	}
   4520 
   4521 	if (wg->wg_npeers == 0)
   4522 		goto skip_peers;
   4523 
   4524 	peers = prop_array_create();
   4525 	if (peers == NULL)
   4526 		goto error;
   4527 
   4528 	s = pserialize_read_enter();
   4529 	i = 0;
   4530 	WG_PEER_READER_FOREACH(wgp, wg) {
   4531 		struct wg_sockaddr *wgsa;
   4532 		struct psref wgp_psref, wgsa_psref;
   4533 		prop_dictionary_t prop_peer;
   4534 
   4535 		wg_get_peer(wgp, &wgp_psref);
   4536 		pserialize_read_exit(s);
   4537 
   4538 		prop_peer = prop_dictionary_create();
   4539 		if (prop_peer == NULL)
   4540 			goto next;
   4541 
   4542 		if (strlen(wgp->wgp_name) > 0) {
   4543 			if (!prop_dictionary_set_string(prop_peer, "name",
   4544 				wgp->wgp_name))
   4545 				goto next;
   4546 		}
   4547 
   4548 		if (!prop_dictionary_set_data(prop_peer, "public_key",
   4549 			wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey)))
   4550 			goto next;
   4551 
   4552 		uint8_t psk_zero[WG_PRESHARED_KEY_LEN] = {0};
   4553 		if (!consttime_memequal(wgp->wgp_psk, psk_zero,
   4554 			sizeof(wgp->wgp_psk))) {
   4555 			if (wg_is_authorized(wg, SIOCGDRVSPEC)) {
   4556 				if (!prop_dictionary_set_data(prop_peer,
   4557 					"preshared_key",
   4558 					wgp->wgp_psk, sizeof(wgp->wgp_psk)))
   4559 					goto next;
   4560 			}
   4561 		}
   4562 
   4563 		wgsa = wg_get_endpoint_sa(wgp, &wgsa_psref);
   4564 		CTASSERT(AF_UNSPEC == 0);
   4565 		if (wgsa_family(wgsa) != 0 /*AF_UNSPEC*/ &&
   4566 		    !prop_dictionary_set_data(prop_peer, "endpoint",
   4567 			wgsatoss(wgsa),
   4568 			sockaddr_getsize_by_family(wgsa_family(wgsa)))) {
   4569 			wg_put_sa(wgp, wgsa, &wgsa_psref);
   4570 			goto next;
   4571 		}
   4572 		wg_put_sa(wgp, wgsa, &wgsa_psref);
   4573 
   4574 		const struct timespec *t = &wgp->wgp_last_handshake_time;
   4575 
   4576 		if (!prop_dictionary_set_uint64(prop_peer,
   4577 			"last_handshake_time_sec", (uint64_t)t->tv_sec))
   4578 			goto next;
   4579 		if (!prop_dictionary_set_uint32(prop_peer,
   4580 			"last_handshake_time_nsec", (uint32_t)t->tv_nsec))
   4581 			goto next;
   4582 
   4583 		if (wgp->wgp_n_allowedips == 0)
   4584 			goto skip_allowedips;
   4585 
   4586 		prop_array_t allowedips = prop_array_create();
   4587 		if (allowedips == NULL)
   4588 			goto next;
   4589 		for (int j = 0; j < wgp->wgp_n_allowedips; j++) {
   4590 			struct wg_allowedip *wga = &wgp->wgp_allowedips[j];
   4591 			prop_dictionary_t prop_allowedip;
   4592 
   4593 			prop_allowedip = prop_dictionary_create();
   4594 			if (prop_allowedip == NULL)
   4595 				break;
   4596 
   4597 			if (!prop_dictionary_set_int(prop_allowedip, "family",
   4598 				wga->wga_family))
   4599 				goto _next;
   4600 			if (!prop_dictionary_set_uint8(prop_allowedip, "cidr",
   4601 				wga->wga_cidr))
   4602 				goto _next;
   4603 
   4604 			switch (wga->wga_family) {
   4605 			case AF_INET:
   4606 				if (!prop_dictionary_set_data(prop_allowedip,
   4607 					"ip", &wga->wga_addr4,
   4608 					sizeof(wga->wga_addr4)))
   4609 					goto _next;
   4610 				break;
   4611 #ifdef INET6
   4612 			case AF_INET6:
   4613 				if (!prop_dictionary_set_data(prop_allowedip,
   4614 					"ip", &wga->wga_addr6,
   4615 					sizeof(wga->wga_addr6)))
   4616 					goto _next;
   4617 				break;
   4618 #endif
   4619 			default:
   4620 				break;
   4621 			}
   4622 			prop_array_set(allowedips, j, prop_allowedip);
   4623 		_next:
   4624 			prop_object_release(prop_allowedip);
   4625 		}
   4626 		prop_dictionary_set(prop_peer, "allowedips", allowedips);
   4627 		prop_object_release(allowedips);
   4628 
   4629 	skip_allowedips:
   4630 
   4631 		prop_array_set(peers, i, prop_peer);
   4632 	next:
   4633 		if (prop_peer)
   4634 			prop_object_release(prop_peer);
   4635 		i++;
   4636 
   4637 		s = pserialize_read_enter();
   4638 		wg_put_peer(wgp, &wgp_psref);
   4639 	}
   4640 	pserialize_read_exit(s);
   4641 
   4642 	prop_dictionary_set(prop_dict, "peers", peers);
   4643 	prop_object_release(peers);
   4644 	peers = NULL;
   4645 
   4646 skip_peers:
   4647 	buf = prop_dictionary_externalize(prop_dict);
   4648 	if (buf == NULL)
   4649 		goto error;
   4650 	if (ifd->ifd_len < (strlen(buf) + 1)) {
   4651 		error = EINVAL;
   4652 		goto error;
   4653 	}
   4654 	error = copyout(buf, ifd->ifd_data, strlen(buf) + 1);
   4655 
   4656 	free(buf, 0);
   4657 error:
   4658 	if (peers != NULL)
   4659 		prop_object_release(peers);
   4660 	if (prop_dict != NULL)
   4661 		prop_object_release(prop_dict);
   4662 
   4663 	return error;
   4664 }
   4665 
   4666 static int
   4667 wg_ioctl(struct ifnet *ifp, u_long cmd, void *data)
   4668 {
   4669 	struct wg_softc *wg = ifp->if_softc;
   4670 	struct ifreq *ifr = data;
   4671 	struct ifaddr *ifa = data;
   4672 	struct ifdrv *ifd = data;
   4673 	int error = 0;
   4674 
   4675 	switch (cmd) {
   4676 	case SIOCINITIFADDR:
   4677 		if (ifa->ifa_addr->sa_family != AF_LINK &&
   4678 		    (ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
   4679 		    (IFF_UP | IFF_RUNNING)) {
   4680 			ifp->if_flags |= IFF_UP;
   4681 			error = if_init(ifp);
   4682 		}
   4683 		return error;
   4684 	case SIOCADDMULTI:
   4685 	case SIOCDELMULTI:
   4686 		switch (ifr->ifr_addr.sa_family) {
   4687 		case AF_INET:	/* IP supports Multicast */
   4688 			break;
   4689 #ifdef INET6
   4690 		case AF_INET6:	/* IP6 supports Multicast */
   4691 			break;
   4692 #endif
   4693 		default:  /* Other protocols doesn't support Multicast */
   4694 			error = EAFNOSUPPORT;
   4695 			break;
   4696 		}
   4697 		return error;
   4698 	case SIOCSDRVSPEC:
   4699 		if (!wg_is_authorized(wg, cmd)) {
   4700 			return EPERM;
   4701 		}
   4702 		switch (ifd->ifd_cmd) {
   4703 		case WG_IOCTL_SET_PRIVATE_KEY:
   4704 			error = wg_ioctl_set_private_key(wg, ifd);
   4705 			break;
   4706 		case WG_IOCTL_SET_LISTEN_PORT:
   4707 			error = wg_ioctl_set_listen_port(wg, ifd);
   4708 			break;
   4709 		case WG_IOCTL_ADD_PEER:
   4710 			error = wg_ioctl_add_peer(wg, ifd);
   4711 			break;
   4712 		case WG_IOCTL_DELETE_PEER:
   4713 			error = wg_ioctl_delete_peer(wg, ifd);
   4714 			break;
   4715 		default:
   4716 			error = EINVAL;
   4717 			break;
   4718 		}
   4719 		return error;
   4720 	case SIOCGDRVSPEC:
   4721 		return wg_ioctl_get(wg, ifd);
   4722 	case SIOCSIFFLAGS:
   4723 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
   4724 			break;
   4725 		switch (ifp->if_flags & (IFF_UP|IFF_RUNNING)) {
   4726 		case IFF_RUNNING:
   4727 			/*
   4728 			 * If interface is marked down and it is running,
   4729 			 * then stop and disable it.
   4730 			 */
   4731 			if_stop(ifp, 1);
   4732 			break;
   4733 		case IFF_UP:
   4734 			/*
   4735 			 * If interface is marked up and it is stopped, then
   4736 			 * start it.
   4737 			 */
   4738 			error = if_init(ifp);
   4739 			break;
   4740 		default:
   4741 			break;
   4742 		}
   4743 		return error;
   4744 #ifdef WG_RUMPKERNEL
   4745 	case SIOCSLINKSTR:
   4746 		error = wg_ioctl_linkstr(wg, ifd);
   4747 		if (error == 0)
   4748 			wg->wg_ops = &wg_ops_rumpuser;
   4749 		return error;
   4750 #endif
   4751 	default:
   4752 		break;
   4753 	}
   4754 
   4755 	error = ifioctl_common(ifp, cmd, data);
   4756 
   4757 #ifdef WG_RUMPKERNEL
   4758 	if (!wg_user_mode(wg))
   4759 		return error;
   4760 
   4761 	/* Do the same to the corresponding tun device on the host */
   4762 	/*
   4763 	 * XXX Actually the command has not been handled yet.  It
   4764 	 *     will be handled via pr_ioctl form doifioctl later.
   4765 	 */
   4766 	switch (cmd) {
   4767 	case SIOCAIFADDR:
   4768 	case SIOCDIFADDR: {
   4769 		struct in_aliasreq _ifra = *(const struct in_aliasreq *)data;
   4770 		struct in_aliasreq *ifra = &_ifra;
   4771 		KASSERT(error == ENOTTY);
   4772 		strncpy(ifra->ifra_name, rumpuser_wg_get_tunname(wg->wg_user),
   4773 		    IFNAMSIZ);
   4774 		error = rumpuser_wg_ioctl(wg->wg_user, cmd, ifra, AF_INET);
   4775 		if (error == 0)
   4776 			error = ENOTTY;
   4777 		break;
   4778 	}
   4779 #ifdef INET6
   4780 	case SIOCAIFADDR_IN6:
   4781 	case SIOCDIFADDR_IN6: {
   4782 		struct in6_aliasreq _ifra = *(const struct in6_aliasreq *)data;
   4783 		struct in6_aliasreq *ifra = &_ifra;
   4784 		KASSERT(error == ENOTTY);
   4785 		strncpy(ifra->ifra_name, rumpuser_wg_get_tunname(wg->wg_user),
   4786 		    IFNAMSIZ);
   4787 		error = rumpuser_wg_ioctl(wg->wg_user, cmd, ifra, AF_INET6);
   4788 		if (error == 0)
   4789 			error = ENOTTY;
   4790 		break;
   4791 	}
   4792 #endif
   4793 	}
   4794 #endif /* WG_RUMPKERNEL */
   4795 
   4796 	return error;
   4797 }
   4798 
   4799 static int
   4800 wg_init(struct ifnet *ifp)
   4801 {
   4802 
   4803 	ifp->if_flags |= IFF_RUNNING;
   4804 
   4805 	/* TODO flush pending packets. */
   4806 	return 0;
   4807 }
   4808 
   4809 #ifdef ALTQ
   4810 static void
   4811 wg_start(struct ifnet *ifp)
   4812 {
   4813 	struct mbuf *m;
   4814 
   4815 	for (;;) {
   4816 		IFQ_DEQUEUE(&ifp->if_snd, m);
   4817 		if (m == NULL)
   4818 			break;
   4819 
   4820 		kpreempt_disable();
   4821 		const uint32_t h = curcpu()->ci_index;	// pktq_rps_hash(m)
   4822 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   4823 			WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   4824 			    if_name(ifp));
   4825 			m_freem(m);
   4826 		}
   4827 		kpreempt_enable();
   4828 	}
   4829 }
   4830 #endif
   4831 
   4832 static void
   4833 wg_stop(struct ifnet *ifp, int disable)
   4834 {
   4835 
   4836 	KASSERT((ifp->if_flags & IFF_RUNNING) != 0);
   4837 	ifp->if_flags &= ~IFF_RUNNING;
   4838 
   4839 	/* Need to do something? */
   4840 }
   4841 
   4842 #ifdef WG_DEBUG_PARAMS
   4843 SYSCTL_SETUP(sysctl_net_wg_setup, "sysctl net.wg setup")
   4844 {
   4845 	const struct sysctlnode *node = NULL;
   4846 
   4847 	sysctl_createv(clog, 0, NULL, &node,
   4848 	    CTLFLAG_PERMANENT,
   4849 	    CTLTYPE_NODE, "wg",
   4850 	    SYSCTL_DESCR("wg(4)"),
   4851 	    NULL, 0, NULL, 0,
   4852 	    CTL_NET, CTL_CREATE, CTL_EOL);
   4853 	sysctl_createv(clog, 0, &node, NULL,
   4854 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4855 	    CTLTYPE_QUAD, "rekey_after_messages",
   4856 	    SYSCTL_DESCR("session liftime by messages"),
   4857 	    NULL, 0, &wg_rekey_after_messages, 0, CTL_CREATE, CTL_EOL);
   4858 	sysctl_createv(clog, 0, &node, NULL,
   4859 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4860 	    CTLTYPE_INT, "rekey_after_time",
   4861 	    SYSCTL_DESCR("session liftime"),
   4862 	    NULL, 0, &wg_rekey_after_time, 0, CTL_CREATE, CTL_EOL);
   4863 	sysctl_createv(clog, 0, &node, NULL,
   4864 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4865 	    CTLTYPE_INT, "rekey_timeout",
   4866 	    SYSCTL_DESCR("session handshake retry time"),
   4867 	    NULL, 0, &wg_rekey_timeout, 0, CTL_CREATE, CTL_EOL);
   4868 	sysctl_createv(clog, 0, &node, NULL,
   4869 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4870 	    CTLTYPE_INT, "rekey_attempt_time",
   4871 	    SYSCTL_DESCR("session handshake timeout"),
   4872 	    NULL, 0, &wg_rekey_attempt_time, 0, CTL_CREATE, CTL_EOL);
   4873 	sysctl_createv(clog, 0, &node, NULL,
   4874 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4875 	    CTLTYPE_INT, "keepalive_timeout",
   4876 	    SYSCTL_DESCR("keepalive timeout"),
   4877 	    NULL, 0, &wg_keepalive_timeout, 0, CTL_CREATE, CTL_EOL);
   4878 	sysctl_createv(clog, 0, &node, NULL,
   4879 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4880 	    CTLTYPE_BOOL, "force_underload",
   4881 	    SYSCTL_DESCR("force to detemine under load"),
   4882 	    NULL, 0, &wg_force_underload, 0, CTL_CREATE, CTL_EOL);
   4883 }
   4884 #endif
   4885 
   4886 #ifdef WG_RUMPKERNEL
   4887 static bool
   4888 wg_user_mode(struct wg_softc *wg)
   4889 {
   4890 
   4891 	return wg->wg_user != NULL;
   4892 }
   4893 
   4894 static int
   4895 wg_ioctl_linkstr(struct wg_softc *wg, struct ifdrv *ifd)
   4896 {
   4897 	struct ifnet *ifp = &wg->wg_if;
   4898 	int error;
   4899 
   4900 	if (ifp->if_flags & IFF_UP)
   4901 		return EBUSY;
   4902 
   4903 	if (ifd->ifd_cmd == IFLINKSTR_UNSET) {
   4904 		/* XXX do nothing */
   4905 		return 0;
   4906 	} else if (ifd->ifd_cmd != 0) {
   4907 		return EINVAL;
   4908 	} else if (wg->wg_user != NULL) {
   4909 		return EBUSY;
   4910 	}
   4911 
   4912 	/* Assume \0 included */
   4913 	if (ifd->ifd_len > IFNAMSIZ) {
   4914 		return E2BIG;
   4915 	} else if (ifd->ifd_len < 1) {
   4916 		return EINVAL;
   4917 	}
   4918 
   4919 	char tun_name[IFNAMSIZ];
   4920 	error = copyinstr(ifd->ifd_data, tun_name, ifd->ifd_len, NULL);
   4921 	if (error != 0)
   4922 		return error;
   4923 
   4924 	if (strncmp(tun_name, "tun", 3) != 0)
   4925 		return EINVAL;
   4926 
   4927 	error = rumpuser_wg_create(tun_name, wg, &wg->wg_user);
   4928 
   4929 	return error;
   4930 }
   4931 
   4932 static int
   4933 wg_send_user(struct wg_peer *wgp, struct mbuf *m)
   4934 {
   4935 	int error;
   4936 	struct psref psref;
   4937 	struct wg_sockaddr *wgsa;
   4938 	struct wg_softc *wg = wgp->wgp_sc;
   4939 	struct iovec iov[1];
   4940 
   4941 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   4942 
   4943 	iov[0].iov_base = mtod(m, void *);
   4944 	iov[0].iov_len = m->m_len;
   4945 
   4946 	/* Send messages to a peer via an ordinary socket. */
   4947 	error = rumpuser_wg_send_peer(wg->wg_user, wgsatosa(wgsa), iov, 1);
   4948 
   4949 	wg_put_sa(wgp, wgsa, &psref);
   4950 
   4951 	m_freem(m);
   4952 
   4953 	return error;
   4954 }
   4955 
   4956 static void
   4957 wg_input_user(struct ifnet *ifp, struct mbuf *m, const int af)
   4958 {
   4959 	struct wg_softc *wg = ifp->if_softc;
   4960 	struct iovec iov[2];
   4961 	struct sockaddr_storage ss;
   4962 
   4963 	KASSERT(af == AF_INET || af == AF_INET6);
   4964 
   4965 	WG_TRACE("");
   4966 
   4967 	if (af == AF_INET) {
   4968 		struct sockaddr_in *sin = (struct sockaddr_in *)&ss;
   4969 		struct ip *ip;
   4970 
   4971 		KASSERT(m->m_len >= sizeof(struct ip));
   4972 		ip = mtod(m, struct ip *);
   4973 		sockaddr_in_init(sin, &ip->ip_dst, 0);
   4974 	} else {
   4975 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ss;
   4976 		struct ip6_hdr *ip6;
   4977 
   4978 		KASSERT(m->m_len >= sizeof(struct ip6_hdr));
   4979 		ip6 = mtod(m, struct ip6_hdr *);
   4980 		sockaddr_in6_init(sin6, &ip6->ip6_dst, 0, 0, 0);
   4981 	}
   4982 
   4983 	iov[0].iov_base = &ss;
   4984 	iov[0].iov_len = ss.ss_len;
   4985 	iov[1].iov_base = mtod(m, void *);
   4986 	iov[1].iov_len = m->m_len;
   4987 
   4988 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
   4989 
   4990 	/* Send decrypted packets to users via a tun. */
   4991 	rumpuser_wg_send_user(wg->wg_user, iov, 2);
   4992 
   4993 	m_freem(m);
   4994 }
   4995 
   4996 static int
   4997 wg_bind_port_user(struct wg_softc *wg, const uint16_t port)
   4998 {
   4999 	int error;
   5000 	uint16_t old_port = wg->wg_listen_port;
   5001 
   5002 	if (port != 0 && old_port == port)
   5003 		return 0;
   5004 
   5005 	error = rumpuser_wg_sock_bind(wg->wg_user, port);
   5006 	if (error == 0)
   5007 		wg->wg_listen_port = port;
   5008 	return error;
   5009 }
   5010 
   5011 /*
   5012  * Receive user packets.
   5013  */
   5014 void
   5015 rumpkern_wg_recv_user(struct wg_softc *wg, struct iovec *iov, size_t iovlen)
   5016 {
   5017 	struct ifnet *ifp = &wg->wg_if;
   5018 	struct mbuf *m;
   5019 	const struct sockaddr *dst;
   5020 
   5021 	WG_TRACE("");
   5022 
   5023 	dst = iov[0].iov_base;
   5024 
   5025 	m = m_gethdr(M_DONTWAIT, MT_DATA);
   5026 	if (m == NULL)
   5027 		return;
   5028 	m->m_len = m->m_pkthdr.len = 0;
   5029 	m_copyback(m, 0, iov[1].iov_len, iov[1].iov_base);
   5030 
   5031 	WG_DLOG("iov_len=%lu\n", iov[1].iov_len);
   5032 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
   5033 
   5034 	(void)wg_output(ifp, m, dst, NULL);
   5035 }
   5036 
   5037 /*
   5038  * Receive packets from a peer.
   5039  */
   5040 void
   5041 rumpkern_wg_recv_peer(struct wg_softc *wg, struct iovec *iov, size_t iovlen)
   5042 {
   5043 	struct mbuf *m;
   5044 	const struct sockaddr *src;
   5045 	int bound;
   5046 
   5047 	WG_TRACE("");
   5048 
   5049 	src = iov[0].iov_base;
   5050 
   5051 	m = m_gethdr(M_DONTWAIT, MT_DATA);
   5052 	if (m == NULL)
   5053 		return;
   5054 	m->m_len = m->m_pkthdr.len = 0;
   5055 	m_copyback(m, 0, iov[1].iov_len, iov[1].iov_base);
   5056 
   5057 	WG_DLOG("iov_len=%lu\n", iov[1].iov_len);
   5058 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
   5059 
   5060 	bound = curlwp_bind();
   5061 	wg_handle_packet(wg, m, src);
   5062 	curlwp_bindx(bound);
   5063 }
   5064 #endif /* WG_RUMPKERNEL */
   5065 
   5066 /*
   5067  * Module infrastructure
   5068  */
   5069 #include "if_module.h"
   5070 
   5071 IF_MODULE(MODULE_CLASS_DRIVER, wg, "sodium,blake2s")
   5072