Home | History | Annotate | Line # | Download | only in net
if_wg.c revision 1.92
      1 /*	$NetBSD: if_wg.c,v 1.92 2024/07/26 18:34:35 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 1.92 2024/07/26 18:34:35 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 <lib/libkern/libkern.h>
     87 
     88 #include <net/bpf.h>
     89 #include <net/if.h>
     90 #include <net/if_types.h>
     91 #include <net/if_wg.h>
     92 #include <net/pktqueue.h>
     93 #include <net/route.h>
     94 
     95 #include <netinet/in.h>
     96 #include <netinet/in_pcb.h>
     97 #include <netinet/in_var.h>
     98 #include <netinet/ip.h>
     99 #include <netinet/ip_var.h>
    100 #include <netinet/udp.h>
    101 #include <netinet/udp_var.h>
    102 
    103 #ifdef INET6
    104 #include <netinet/ip6.h>
    105 #include <netinet6/in6_pcb.h>
    106 #include <netinet6/in6_var.h>
    107 #include <netinet6/ip6_var.h>
    108 #include <netinet6/udp6_var.h>
    109 #endif /* INET6 */
    110 
    111 #include <prop/proplib.h>
    112 
    113 #include <crypto/blake2/blake2s.h>
    114 #include <crypto/sodium/crypto_aead_chacha20poly1305.h>
    115 #include <crypto/sodium/crypto_aead_xchacha20poly1305.h>
    116 #include <crypto/sodium/crypto_scalarmult.h>
    117 
    118 #include "ioconf.h"
    119 
    120 #ifdef WG_RUMPKERNEL
    121 #include "wg_user.h"
    122 #endif
    123 
    124 /*
    125  * Data structures
    126  * - struct wg_softc is an instance of wg interfaces
    127  *   - It has a list of peers (struct wg_peer)
    128  *   - It has a threadpool job that sends/receives handshake messages and
    129  *     runs event handlers
    130  *   - It has its own two routing tables: one is for IPv4 and the other IPv6
    131  * - struct wg_peer is a representative of a peer
    132  *   - It has a struct work to handle handshakes and timer tasks
    133  *   - It has a pair of session instances (struct wg_session)
    134  *   - It has a pair of endpoint instances (struct wg_sockaddr)
    135  *     - Normally one endpoint is used and the second one is used only on
    136  *       a peer migration (a change of peer's IP address)
    137  *   - It has a list of IP addresses and sub networks called allowedips
    138  *     (struct wg_allowedip)
    139  *     - A packets sent over a session is allowed if its destination matches
    140  *       any IP addresses or sub networks of the list
    141  * - struct wg_session represents a session of a secure tunnel with a peer
    142  *   - Two instances of sessions belong to a peer; a stable session and a
    143  *     unstable session
    144  *   - A handshake process of a session always starts with a unstable instance
    145  *   - Once a session is established, its instance becomes stable and the
    146  *     other becomes unstable instead
    147  *   - Data messages are always sent via a stable session
    148  *
    149  * Locking notes:
    150  * - Each wg has a mutex(9) wg_lock, and a rwlock(9) wg_rwlock
    151  *   - Changes to the peer list are serialized by wg_lock
    152  *   - The peer list may be read with pserialize(9) and psref(9)
    153  *   - The rwlock (wg_rwlock) protects the routing tables (wg_rtable_ipv[46])
    154  *     => XXX replace by pserialize when routing table is psz-safe
    155  * - Each peer (struct wg_peer, wgp) has a mutex wgp_lock, which can be taken
    156  *   only in thread context and serializes:
    157  *   - the stable and unstable session pointers
    158  *   - all unstable session state
    159  * - Packet processing may be done in softint context:
    160  *   - The stable session can be read under pserialize(9) or psref(9)
    161  *     - The stable session is always ESTABLISHED
    162  *     - On a session swap, we must wait for all readers to release a
    163  *       reference to a stable session before changing wgs_state and
    164  *       session states
    165  * - Lock order: wg_lock -> wgp_lock
    166  */
    167 
    168 
    169 #define WGLOG(level, fmt, args...)					      \
    170 	log(level, "%s: " fmt, __func__, ##args)
    171 
    172 #define WG_DEBUG
    173 
    174 /* Debug options */
    175 #ifdef WG_DEBUG
    176 /* Output debug logs */
    177 #ifndef WG_DEBUG_LOG
    178 #define WG_DEBUG_LOG
    179 #endif
    180 /* Output trace logs */
    181 #ifndef WG_DEBUG_TRACE
    182 #define WG_DEBUG_TRACE
    183 #endif
    184 /* Output hash values, etc. */
    185 #ifndef WG_DEBUG_DUMP
    186 #define WG_DEBUG_DUMP
    187 #endif
    188 /* debug packets */
    189 #ifndef WG_DEBUG_PACKET
    190 #define WG_DEBUG_PACKET
    191 #endif
    192 /* Make some internal parameters configurable for testing and debugging */
    193 #ifndef WG_DEBUG_PARAMS
    194 #define WG_DEBUG_PARAMS
    195 #endif
    196 #endif /* WG_DEBUG */
    197 
    198 #ifndef WG_DEBUG
    199 # if defined(WG_DEBUG_LOG) || defined(WG_DEBUG_TRACE) ||		    \
    200 	defined(WG_DEBUG_DUMP) || defined(WG_DEBUG_PARAMS) ||		    \
    201 	defined(WG_DEBUG_PACKET)
    202 #   define WG_DEBUG
    203 # endif
    204 #endif
    205 
    206 #ifdef WG_DEBUG
    207 int wg_debug;
    208 #define WG_DEBUG_FLAGS_LOG	1
    209 #define WG_DEBUG_FLAGS_TRACE	2
    210 #define WG_DEBUG_FLAGS_DUMP	4
    211 #define WG_DEBUG_FLAGS_PACKET	8
    212 #endif
    213 
    214 
    215 #ifdef WG_DEBUG_TRACE
    216 #define WG_TRACE(msg)	 do {						\
    217 	if (wg_debug & WG_DEBUG_FLAGS_TRACE)				\
    218 	    log(LOG_DEBUG, "%s:%d: %s\n", __func__, __LINE__, (msg));	\
    219 } while (0)
    220 #else
    221 #define WG_TRACE(msg)	__nothing
    222 #endif
    223 
    224 #ifdef WG_DEBUG_LOG
    225 #define WG_DLOG(fmt, args...)	 do {					\
    226 	if (wg_debug & WG_DEBUG_FLAGS_LOG)				\
    227 	    log(LOG_DEBUG, "%s: " fmt, __func__, ##args);		\
    228 } while (0)
    229 #else
    230 #define WG_DLOG(fmt, args...)	__nothing
    231 #endif
    232 
    233 #define WG_LOG_RATECHECK(wgprc, level, fmt, args...)	do {		\
    234 	if (ppsratecheck(&(wgprc)->wgprc_lasttime,			\
    235 	    &(wgprc)->wgprc_curpps, 1)) {				\
    236 		log(level, fmt, ##args);				\
    237 	}								\
    238 } while (0)
    239 
    240 #ifdef WG_DEBUG_PARAMS
    241 static bool wg_force_underload = false;
    242 #endif
    243 
    244 #ifdef WG_DEBUG_DUMP
    245 
    246 static char enomem[10] = "[enomem]";
    247 
    248 static char *
    249 gethexdump(const void *vp, size_t n)
    250 {
    251 	char *buf;
    252 	const uint8_t *p = vp;
    253 	size_t i;
    254 
    255 	if (n > (SIZE_MAX - 1) / 3)
    256 		return enomem;
    257 	buf = kmem_alloc(3 * n + 1, KM_NOSLEEP);
    258 	if (buf == NULL)
    259 		return enomem;
    260 	for (i = 0; i < n; i++)
    261 		snprintf(buf + 3 * i, 3 + 1, " %02hhx", p[i]);
    262 	return buf;
    263 }
    264 
    265 static void
    266 puthexdump(char *buf, const void *p, size_t n)
    267 {
    268 
    269 	if (buf == NULL || buf == enomem)
    270 		return;
    271 	kmem_free(buf, 3*n + 1);
    272 }
    273 
    274 #ifdef WG_RUMPKERNEL
    275 static void
    276 wg_dump_buf(const char *func, const char *buf, const size_t size)
    277 {
    278 	if ((wg_debug & WG_DEBUG_FLAGS_DUMP) == 0)
    279 		return;
    280 
    281 	char *hex = gethexdump(buf, size);
    282 
    283 	log(LOG_DEBUG, "%s: %s\n", func, hex);
    284 	puthexdump(hex, buf, size);
    285 }
    286 #endif
    287 
    288 static void
    289 wg_dump_hash(const uint8_t *func, const uint8_t *name, const uint8_t *hash,
    290     const size_t size)
    291 {
    292 	if ((wg_debug & WG_DEBUG_FLAGS_DUMP) == 0)
    293 		return;
    294 
    295 	char *hex = gethexdump(hash, size);
    296 
    297 	log(LOG_DEBUG, "%s: %s: %s\n", func, name, hex);
    298 	puthexdump(hex, hash, size);
    299 }
    300 
    301 #define WG_DUMP_HASH(name, hash) \
    302 	wg_dump_hash(__func__, name, hash, WG_HASH_LEN)
    303 #define WG_DUMP_HASH48(name, hash) \
    304 	wg_dump_hash(__func__, name, hash, 48)
    305 #define WG_DUMP_BUF(buf, size) \
    306 	wg_dump_buf(__func__, buf, size)
    307 #else
    308 #define WG_DUMP_HASH(name, hash)	__nothing
    309 #define WG_DUMP_HASH48(name, hash)	__nothing
    310 #define WG_DUMP_BUF(buf, size)	__nothing
    311 #endif /* WG_DEBUG_DUMP */
    312 
    313 /* chosen somewhat arbitrarily -- fits in signed 16 bits NUL-terminated */
    314 #define	WG_MAX_PROPLEN		32766
    315 
    316 #define WG_MTU			1420
    317 #define WG_ALLOWEDIPS		16
    318 
    319 #define CURVE25519_KEY_LEN	32
    320 #define TAI64N_LEN		sizeof(uint32_t) * 3
    321 #define POLY1305_AUTHTAG_LEN	16
    322 #define HMAC_BLOCK_LEN		64
    323 
    324 /* [N] 4.1: "DHLEN must be 32 or greater."  WireGuard chooses 32. */
    325 /* [N] 4.3: Hash functions */
    326 #define NOISE_DHLEN		32
    327 /* [N] 4.3: "Must be 32 or 64."  WireGuard chooses 32. */
    328 #define NOISE_HASHLEN		32
    329 #define NOISE_BLOCKLEN		64
    330 #define NOISE_HKDF_OUTPUT_LEN	NOISE_HASHLEN
    331 /* [N] 5.1: "k" */
    332 #define NOISE_CIPHER_KEY_LEN	32
    333 /*
    334  * [N] 9.2: "psk"
    335  *          "... psk is a 32-byte secret value provided by the application."
    336  */
    337 #define NOISE_PRESHARED_KEY_LEN	32
    338 
    339 #define WG_STATIC_KEY_LEN	CURVE25519_KEY_LEN
    340 #define WG_TIMESTAMP_LEN	TAI64N_LEN
    341 
    342 #define WG_PRESHARED_KEY_LEN	NOISE_PRESHARED_KEY_LEN
    343 
    344 #define WG_COOKIE_LEN		16
    345 #define WG_MAC_LEN		16
    346 #define WG_RANDVAL_LEN		24
    347 
    348 #define WG_EPHEMERAL_KEY_LEN	CURVE25519_KEY_LEN
    349 /* [N] 5.2: "ck: A chaining key of HASHLEN bytes" */
    350 #define WG_CHAINING_KEY_LEN	NOISE_HASHLEN
    351 /* [N] 5.2: "h: A hash output of HASHLEN bytes" */
    352 #define WG_HASH_LEN		NOISE_HASHLEN
    353 #define WG_CIPHER_KEY_LEN	NOISE_CIPHER_KEY_LEN
    354 #define WG_DH_OUTPUT_LEN	NOISE_DHLEN
    355 #define WG_KDF_OUTPUT_LEN	NOISE_HKDF_OUTPUT_LEN
    356 #define WG_AUTHTAG_LEN		POLY1305_AUTHTAG_LEN
    357 #define WG_DATA_KEY_LEN		32
    358 #define WG_SALT_LEN		24
    359 
    360 /*
    361  * The protocol messages
    362  */
    363 struct wg_msg {
    364 	uint32_t	wgm_type;
    365 } __packed;
    366 
    367 /* [W] 5.4.2 First Message: Initiator to Responder */
    368 struct wg_msg_init {
    369 	uint32_t	wgmi_type;
    370 	uint32_t	wgmi_sender;
    371 	uint8_t		wgmi_ephemeral[WG_EPHEMERAL_KEY_LEN];
    372 	uint8_t		wgmi_static[WG_STATIC_KEY_LEN + WG_AUTHTAG_LEN];
    373 	uint8_t		wgmi_timestamp[WG_TIMESTAMP_LEN + WG_AUTHTAG_LEN];
    374 	uint8_t		wgmi_mac1[WG_MAC_LEN];
    375 	uint8_t		wgmi_mac2[WG_MAC_LEN];
    376 } __packed;
    377 
    378 /* [W] 5.4.3 Second Message: Responder to Initiator */
    379 struct wg_msg_resp {
    380 	uint32_t	wgmr_type;
    381 	uint32_t	wgmr_sender;
    382 	uint32_t	wgmr_receiver;
    383 	uint8_t		wgmr_ephemeral[WG_EPHEMERAL_KEY_LEN];
    384 	uint8_t		wgmr_empty[0 + WG_AUTHTAG_LEN];
    385 	uint8_t		wgmr_mac1[WG_MAC_LEN];
    386 	uint8_t		wgmr_mac2[WG_MAC_LEN];
    387 } __packed;
    388 
    389 /* [W] 5.4.6 Subsequent Messages: Transport Data Messages */
    390 struct wg_msg_data {
    391 	uint32_t	wgmd_type;
    392 	uint32_t	wgmd_receiver;
    393 	uint64_t	wgmd_counter;
    394 	uint32_t	wgmd_packet[0];
    395 } __packed;
    396 
    397 /* [W] 5.4.7 Under Load: Cookie Reply Message */
    398 struct wg_msg_cookie {
    399 	uint32_t	wgmc_type;
    400 	uint32_t	wgmc_receiver;
    401 	uint8_t		wgmc_salt[WG_SALT_LEN];
    402 	uint8_t		wgmc_cookie[WG_COOKIE_LEN + WG_AUTHTAG_LEN];
    403 } __packed;
    404 
    405 #define WG_MSG_TYPE_INIT		1
    406 #define WG_MSG_TYPE_RESP		2
    407 #define WG_MSG_TYPE_COOKIE		3
    408 #define WG_MSG_TYPE_DATA		4
    409 #define WG_MSG_TYPE_MAX			WG_MSG_TYPE_DATA
    410 
    411 /* Sliding windows */
    412 
    413 #define	SLIWIN_BITS	2048u
    414 #define	SLIWIN_TYPE	uint32_t
    415 #define	SLIWIN_BPW	NBBY*sizeof(SLIWIN_TYPE)
    416 #define	SLIWIN_WORDS	howmany(SLIWIN_BITS, SLIWIN_BPW)
    417 #define	SLIWIN_NPKT	(SLIWIN_BITS - NBBY*sizeof(SLIWIN_TYPE))
    418 
    419 struct sliwin {
    420 	SLIWIN_TYPE	B[SLIWIN_WORDS];
    421 	uint64_t	T;
    422 };
    423 
    424 static void
    425 sliwin_reset(struct sliwin *W)
    426 {
    427 
    428 	memset(W, 0, sizeof(*W));
    429 }
    430 
    431 static int
    432 sliwin_check_fast(const volatile struct sliwin *W, uint64_t S)
    433 {
    434 
    435 	/*
    436 	 * If it's more than one window older than the highest sequence
    437 	 * number we've seen, reject.
    438 	 */
    439 #ifdef __HAVE_ATOMIC64_LOADSTORE
    440 	if (S + SLIWIN_NPKT < atomic_load_relaxed(&W->T))
    441 		return EAUTH;
    442 #endif
    443 
    444 	/*
    445 	 * Otherwise, we need to take the lock to decide, so don't
    446 	 * reject just yet.  Caller must serialize a call to
    447 	 * sliwin_update in this case.
    448 	 */
    449 	return 0;
    450 }
    451 
    452 static int
    453 sliwin_update(struct sliwin *W, uint64_t S)
    454 {
    455 	unsigned word, bit;
    456 
    457 	/*
    458 	 * If it's more than one window older than the highest sequence
    459 	 * number we've seen, reject.
    460 	 */
    461 	if (S + SLIWIN_NPKT < W->T)
    462 		return EAUTH;
    463 
    464 	/*
    465 	 * If it's higher than the highest sequence number we've seen,
    466 	 * advance the window.
    467 	 */
    468 	if (S > W->T) {
    469 		uint64_t i = W->T / SLIWIN_BPW;
    470 		uint64_t j = S / SLIWIN_BPW;
    471 		unsigned k;
    472 
    473 		for (k = 0; k < MIN(j - i, SLIWIN_WORDS); k++)
    474 			W->B[(i + k + 1) % SLIWIN_WORDS] = 0;
    475 #ifdef __HAVE_ATOMIC64_LOADSTORE
    476 		atomic_store_relaxed(&W->T, S);
    477 #else
    478 		W->T = S;
    479 #endif
    480 	}
    481 
    482 	/* Test and set the bit -- if already set, reject.  */
    483 	word = (S / SLIWIN_BPW) % SLIWIN_WORDS;
    484 	bit = S % SLIWIN_BPW;
    485 	if (W->B[word] & (1UL << bit))
    486 		return EAUTH;
    487 	W->B[word] |= 1U << bit;
    488 
    489 	/* Accept!  */
    490 	return 0;
    491 }
    492 
    493 struct wg_session {
    494 	struct wg_peer	*wgs_peer;
    495 	struct psref_target
    496 			wgs_psref;
    497 
    498 	int		wgs_state;
    499 #define WGS_STATE_UNKNOWN	0
    500 #define WGS_STATE_INIT_ACTIVE	1
    501 #define WGS_STATE_INIT_PASSIVE	2
    502 #define WGS_STATE_ESTABLISHED	3
    503 #define WGS_STATE_DESTROYING	4
    504 
    505 	time_t		wgs_time_established;
    506 	time_t		wgs_time_last_data_sent;
    507 	bool		wgs_is_initiator;
    508 
    509 	uint32_t	wgs_local_index;
    510 	uint32_t	wgs_remote_index;
    511 #ifdef __HAVE_ATOMIC64_LOADSTORE
    512 	volatile uint64_t
    513 			wgs_send_counter;
    514 #else
    515 	kmutex_t	wgs_send_counter_lock;
    516 	uint64_t	wgs_send_counter;
    517 #endif
    518 
    519 	struct {
    520 		kmutex_t	lock;
    521 		struct sliwin	window;
    522 	}		*wgs_recvwin;
    523 
    524 	uint8_t		wgs_handshake_hash[WG_HASH_LEN];
    525 	uint8_t		wgs_chaining_key[WG_CHAINING_KEY_LEN];
    526 	uint8_t		wgs_ephemeral_key_pub[WG_EPHEMERAL_KEY_LEN];
    527 	uint8_t		wgs_ephemeral_key_priv[WG_EPHEMERAL_KEY_LEN];
    528 	uint8_t		wgs_ephemeral_key_peer[WG_EPHEMERAL_KEY_LEN];
    529 	uint8_t		wgs_tkey_send[WG_DATA_KEY_LEN];
    530 	uint8_t		wgs_tkey_recv[WG_DATA_KEY_LEN];
    531 };
    532 
    533 struct wg_sockaddr {
    534 	union {
    535 		struct sockaddr_storage _ss;
    536 		struct sockaddr _sa;
    537 		struct sockaddr_in _sin;
    538 		struct sockaddr_in6 _sin6;
    539 	};
    540 	struct psref_target	wgsa_psref;
    541 };
    542 
    543 #define wgsatoss(wgsa)		(&(wgsa)->_ss)
    544 #define wgsatosa(wgsa)		(&(wgsa)->_sa)
    545 #define wgsatosin(wgsa)		(&(wgsa)->_sin)
    546 #define wgsatosin6(wgsa)	(&(wgsa)->_sin6)
    547 
    548 #define	wgsa_family(wgsa)	(wgsatosa(wgsa)->sa_family)
    549 
    550 struct wg_peer;
    551 struct wg_allowedip {
    552 	struct radix_node	wga_nodes[2];
    553 	struct wg_sockaddr	_wga_sa_addr;
    554 	struct wg_sockaddr	_wga_sa_mask;
    555 #define wga_sa_addr		_wga_sa_addr._sa
    556 #define wga_sa_mask		_wga_sa_mask._sa
    557 
    558 	int			wga_family;
    559 	uint8_t			wga_cidr;
    560 	union {
    561 		struct in_addr _ip4;
    562 		struct in6_addr _ip6;
    563 	} wga_addr;
    564 #define wga_addr4	wga_addr._ip4
    565 #define wga_addr6	wga_addr._ip6
    566 
    567 	struct wg_peer		*wga_peer;
    568 };
    569 
    570 typedef uint8_t wg_timestamp_t[WG_TIMESTAMP_LEN];
    571 
    572 struct wg_ppsratecheck {
    573 	struct timeval		wgprc_lasttime;
    574 	int			wgprc_curpps;
    575 };
    576 
    577 struct wg_softc;
    578 struct wg_peer {
    579 	struct wg_softc		*wgp_sc;
    580 	char			wgp_name[WG_PEER_NAME_MAXLEN + 1];
    581 	struct pslist_entry	wgp_peerlist_entry;
    582 	pserialize_t		wgp_psz;
    583 	struct psref_target	wgp_psref;
    584 	kmutex_t		*wgp_lock;
    585 	kmutex_t		*wgp_intr_lock;
    586 
    587 	uint8_t	wgp_pubkey[WG_STATIC_KEY_LEN];
    588 	struct wg_sockaddr	*wgp_endpoint;
    589 	struct wg_sockaddr	*wgp_endpoint0;
    590 	volatile unsigned	wgp_endpoint_changing;
    591 	bool			wgp_endpoint_available;
    592 
    593 			/* The preshared key (optional) */
    594 	uint8_t		wgp_psk[WG_PRESHARED_KEY_LEN];
    595 
    596 	struct wg_session	*wgp_session_stable;
    597 	struct wg_session	*wgp_session_unstable;
    598 
    599 	/* first outgoing packet awaiting session initiation */
    600 	struct mbuf		*wgp_pending;
    601 
    602 	/* timestamp in big-endian */
    603 	wg_timestamp_t	wgp_timestamp_latest_init;
    604 
    605 	struct timespec		wgp_last_handshake_time;
    606 
    607 	callout_t		wgp_rekey_timer;
    608 	callout_t		wgp_handshake_timeout_timer;
    609 	callout_t		wgp_session_dtor_timer;
    610 
    611 	time_t			wgp_handshake_start_time;
    612 
    613 	int			wgp_n_allowedips;
    614 	struct wg_allowedip	wgp_allowedips[WG_ALLOWEDIPS];
    615 
    616 	time_t			wgp_latest_cookie_time;
    617 	uint8_t			wgp_latest_cookie[WG_COOKIE_LEN];
    618 	uint8_t			wgp_last_sent_mac1[WG_MAC_LEN];
    619 	bool			wgp_last_sent_mac1_valid;
    620 	uint8_t			wgp_last_sent_cookie[WG_COOKIE_LEN];
    621 	bool			wgp_last_sent_cookie_valid;
    622 
    623 	time_t			wgp_last_msg_received_time[WG_MSG_TYPE_MAX];
    624 
    625 	time_t			wgp_last_genrandval_time;
    626 	uint32_t		wgp_randval;
    627 
    628 	struct wg_ppsratecheck	wgp_ppsratecheck;
    629 
    630 	struct work		wgp_work;
    631 	unsigned int		wgp_tasks;
    632 #define WGP_TASK_SEND_INIT_MESSAGE		__BIT(0)
    633 #define WGP_TASK_RETRY_HANDSHAKE		__BIT(1)
    634 #define WGP_TASK_ESTABLISH_SESSION		__BIT(2)
    635 #define WGP_TASK_ENDPOINT_CHANGED		__BIT(3)
    636 #define WGP_TASK_SEND_KEEPALIVE_MESSAGE		__BIT(4)
    637 #define WGP_TASK_DESTROY_PREV_SESSION		__BIT(5)
    638 };
    639 
    640 struct wg_ops;
    641 
    642 struct wg_softc {
    643 	struct ifnet	wg_if;
    644 	LIST_ENTRY(wg_softc) wg_list;
    645 	kmutex_t	*wg_lock;
    646 	kmutex_t	*wg_intr_lock;
    647 	krwlock_t	*wg_rwlock;
    648 
    649 	uint8_t		wg_privkey[WG_STATIC_KEY_LEN];
    650 	uint8_t		wg_pubkey[WG_STATIC_KEY_LEN];
    651 
    652 	int		wg_npeers;
    653 	struct pslist_head	wg_peers;
    654 	struct thmap	*wg_peers_bypubkey;
    655 	struct thmap	*wg_peers_byname;
    656 	struct thmap	*wg_sessions_byindex;
    657 	uint16_t	wg_listen_port;
    658 
    659 	struct threadpool	*wg_threadpool;
    660 
    661 	struct threadpool_job	wg_job;
    662 	int			wg_upcalls;
    663 #define	WG_UPCALL_INET	__BIT(0)
    664 #define	WG_UPCALL_INET6	__BIT(1)
    665 
    666 #ifdef INET
    667 	struct socket		*wg_so4;
    668 	struct radix_node_head	*wg_rtable_ipv4;
    669 #endif
    670 #ifdef INET6
    671 	struct socket		*wg_so6;
    672 	struct radix_node_head	*wg_rtable_ipv6;
    673 #endif
    674 
    675 	struct wg_ppsratecheck	wg_ppsratecheck;
    676 
    677 	struct wg_ops		*wg_ops;
    678 
    679 #ifdef WG_RUMPKERNEL
    680 	struct wg_user		*wg_user;
    681 #endif
    682 };
    683 
    684 /* [W] 6.1 Preliminaries */
    685 #define WG_REKEY_AFTER_MESSAGES		(1ULL << 60)
    686 #define WG_REJECT_AFTER_MESSAGES	(UINT64_MAX - (1 << 13))
    687 #define WG_REKEY_AFTER_TIME		120
    688 #define WG_REJECT_AFTER_TIME		180
    689 #define WG_REKEY_ATTEMPT_TIME		 90
    690 #define WG_REKEY_TIMEOUT		  5
    691 #define WG_KEEPALIVE_TIMEOUT		 10
    692 
    693 #define WG_COOKIE_TIME			120
    694 #define WG_RANDVAL_TIME			(2 * 60)
    695 
    696 static uint64_t wg_rekey_after_messages = WG_REKEY_AFTER_MESSAGES;
    697 static uint64_t wg_reject_after_messages = WG_REJECT_AFTER_MESSAGES;
    698 static unsigned wg_rekey_after_time = WG_REKEY_AFTER_TIME;
    699 static unsigned wg_reject_after_time = WG_REJECT_AFTER_TIME;
    700 static unsigned wg_rekey_attempt_time = WG_REKEY_ATTEMPT_TIME;
    701 static unsigned wg_rekey_timeout = WG_REKEY_TIMEOUT;
    702 static unsigned wg_keepalive_timeout = WG_KEEPALIVE_TIMEOUT;
    703 
    704 static struct mbuf *
    705 		wg_get_mbuf(size_t, size_t);
    706 
    707 static int	wg_send_data_msg(struct wg_peer *, struct wg_session *,
    708 		    struct mbuf *);
    709 static int	wg_send_cookie_msg(struct wg_softc *, struct wg_peer *,
    710 		    const uint32_t, const uint8_t [WG_MAC_LEN],
    711 		    const struct sockaddr *);
    712 static int	wg_send_handshake_msg_resp(struct wg_softc *, struct wg_peer *,
    713 		    struct wg_session *, const struct wg_msg_init *);
    714 static void	wg_send_keepalive_msg(struct wg_peer *, struct wg_session *);
    715 
    716 static struct wg_peer *
    717 		wg_pick_peer_by_sa(struct wg_softc *, const struct sockaddr *,
    718 		    struct psref *);
    719 static struct wg_peer *
    720 		wg_lookup_peer_by_pubkey(struct wg_softc *,
    721 		    const uint8_t [WG_STATIC_KEY_LEN], struct psref *);
    722 
    723 static struct wg_session *
    724 		wg_lookup_session_by_index(struct wg_softc *,
    725 		    const uint32_t, struct psref *);
    726 
    727 static void	wg_update_endpoint_if_necessary(struct wg_peer *,
    728 		    const struct sockaddr *);
    729 
    730 static void	wg_schedule_rekey_timer(struct wg_peer *);
    731 static void	wg_schedule_session_dtor_timer(struct wg_peer *);
    732 
    733 static bool	wg_is_underload(struct wg_softc *, struct wg_peer *, int);
    734 static void	wg_calculate_keys(struct wg_session *, const bool);
    735 
    736 static void	wg_clear_states(struct wg_session *);
    737 
    738 static void	wg_get_peer(struct wg_peer *, struct psref *);
    739 static void	wg_put_peer(struct wg_peer *, struct psref *);
    740 
    741 static int	wg_send_so(struct wg_peer *, struct mbuf *);
    742 static int	wg_send_udp(struct wg_peer *, struct mbuf *);
    743 static int	wg_output(struct ifnet *, struct mbuf *,
    744 			   const struct sockaddr *, const struct rtentry *);
    745 static void	wg_input(struct ifnet *, struct mbuf *, const int);
    746 static int	wg_ioctl(struct ifnet *, u_long, void *);
    747 static int	wg_bind_port(struct wg_softc *, const uint16_t);
    748 static int	wg_init(struct ifnet *);
    749 #ifdef ALTQ
    750 static void	wg_start(struct ifnet *);
    751 #endif
    752 static void	wg_stop(struct ifnet *, int);
    753 
    754 static void	wg_peer_work(struct work *, void *);
    755 static void	wg_job(struct threadpool_job *);
    756 static void	wgintr(void *);
    757 static void	wg_purge_pending_packets(struct wg_peer *);
    758 
    759 static int	wg_clone_create(struct if_clone *, int);
    760 static int	wg_clone_destroy(struct ifnet *);
    761 
    762 struct wg_ops {
    763 	int (*send_hs_msg)(struct wg_peer *, struct mbuf *);
    764 	int (*send_data_msg)(struct wg_peer *, struct mbuf *);
    765 	void (*input)(struct ifnet *, struct mbuf *, const int);
    766 	int (*bind_port)(struct wg_softc *, const uint16_t);
    767 };
    768 
    769 struct wg_ops wg_ops_rumpkernel = {
    770 	.send_hs_msg	= wg_send_so,
    771 	.send_data_msg	= wg_send_udp,
    772 	.input		= wg_input,
    773 	.bind_port	= wg_bind_port,
    774 };
    775 
    776 #ifdef WG_RUMPKERNEL
    777 static bool	wg_user_mode(struct wg_softc *);
    778 static int	wg_ioctl_linkstr(struct wg_softc *, struct ifdrv *);
    779 
    780 static int	wg_send_user(struct wg_peer *, struct mbuf *);
    781 static void	wg_input_user(struct ifnet *, struct mbuf *, const int);
    782 static int	wg_bind_port_user(struct wg_softc *, const uint16_t);
    783 
    784 struct wg_ops wg_ops_rumpuser = {
    785 	.send_hs_msg	= wg_send_user,
    786 	.send_data_msg	= wg_send_user,
    787 	.input		= wg_input_user,
    788 	.bind_port	= wg_bind_port_user,
    789 };
    790 #endif
    791 
    792 #define WG_PEER_READER_FOREACH(wgp, wg)					\
    793 	PSLIST_READER_FOREACH((wgp), &(wg)->wg_peers, struct wg_peer,	\
    794 	    wgp_peerlist_entry)
    795 #define WG_PEER_WRITER_FOREACH(wgp, wg)					\
    796 	PSLIST_WRITER_FOREACH((wgp), &(wg)->wg_peers, struct wg_peer,	\
    797 	    wgp_peerlist_entry)
    798 #define WG_PEER_WRITER_INSERT_HEAD(wgp, wg)				\
    799 	PSLIST_WRITER_INSERT_HEAD(&(wg)->wg_peers, (wgp), wgp_peerlist_entry)
    800 #define WG_PEER_WRITER_REMOVE(wgp)					\
    801 	PSLIST_WRITER_REMOVE((wgp), wgp_peerlist_entry)
    802 
    803 struct wg_route {
    804 	struct radix_node	wgr_nodes[2];
    805 	struct wg_peer		*wgr_peer;
    806 };
    807 
    808 static struct radix_node_head *
    809 wg_rnh(struct wg_softc *wg, const int family)
    810 {
    811 
    812 	switch (family) {
    813 		case AF_INET:
    814 			return wg->wg_rtable_ipv4;
    815 #ifdef INET6
    816 		case AF_INET6:
    817 			return wg->wg_rtable_ipv6;
    818 #endif
    819 		default:
    820 			return NULL;
    821 	}
    822 }
    823 
    824 
    825 /*
    826  * Global variables
    827  */
    828 static volatile unsigned wg_count __cacheline_aligned;
    829 
    830 struct psref_class *wg_psref_class __read_mostly;
    831 
    832 static struct if_clone wg_cloner =
    833     IF_CLONE_INITIALIZER("wg", wg_clone_create, wg_clone_destroy);
    834 
    835 static struct pktqueue *wg_pktq __read_mostly;
    836 static struct workqueue *wg_wq __read_mostly;
    837 
    838 void wgattach(int);
    839 /* ARGSUSED */
    840 void
    841 wgattach(int count)
    842 {
    843 	/*
    844 	 * Nothing to do here, initialization is handled by the
    845 	 * module initialization code in wginit() below).
    846 	 */
    847 }
    848 
    849 static void
    850 wginit(void)
    851 {
    852 
    853 	wg_psref_class = psref_class_create("wg", IPL_SOFTNET);
    854 
    855 	if_clone_attach(&wg_cloner);
    856 }
    857 
    858 /*
    859  * XXX Kludge: This should just happen in wginit, but workqueue_create
    860  * cannot be run until after CPUs have been detected, and wginit runs
    861  * before configure.
    862  */
    863 static int
    864 wginitqueues(void)
    865 {
    866 	int error __diagused;
    867 
    868 	wg_pktq = pktq_create(IFQ_MAXLEN, wgintr, NULL);
    869 	KASSERT(wg_pktq != NULL);
    870 
    871 	error = workqueue_create(&wg_wq, "wgpeer", wg_peer_work, NULL,
    872 	    PRI_NONE, IPL_SOFTNET, WQ_MPSAFE|WQ_PERCPU);
    873 	KASSERT(error == 0);
    874 
    875 	return 0;
    876 }
    877 
    878 static void
    879 wg_guarantee_initialized(void)
    880 {
    881 	static ONCE_DECL(init);
    882 	int error __diagused;
    883 
    884 	error = RUN_ONCE(&init, wginitqueues);
    885 	KASSERT(error == 0);
    886 }
    887 
    888 static int
    889 wg_count_inc(void)
    890 {
    891 	unsigned o, n;
    892 
    893 	do {
    894 		o = atomic_load_relaxed(&wg_count);
    895 		if (o == UINT_MAX)
    896 			return ENFILE;
    897 		n = o + 1;
    898 	} while (atomic_cas_uint(&wg_count, o, n) != o);
    899 
    900 	return 0;
    901 }
    902 
    903 static void
    904 wg_count_dec(void)
    905 {
    906 	unsigned c __diagused;
    907 
    908 	c = atomic_dec_uint_nv(&wg_count);
    909 	KASSERT(c != UINT_MAX);
    910 }
    911 
    912 static int
    913 wgdetach(void)
    914 {
    915 
    916 	/* Prevent new interface creation.  */
    917 	if_clone_detach(&wg_cloner);
    918 
    919 	/* Check whether there are any existing interfaces.  */
    920 	if (atomic_load_relaxed(&wg_count)) {
    921 		/* Back out -- reattach the cloner.  */
    922 		if_clone_attach(&wg_cloner);
    923 		return EBUSY;
    924 	}
    925 
    926 	/* No interfaces left.  Nuke it.  */
    927 	if (wg_wq)
    928 		workqueue_destroy(wg_wq);
    929 	if (wg_pktq)
    930 		pktq_destroy(wg_pktq);
    931 	psref_class_destroy(wg_psref_class);
    932 
    933 	return 0;
    934 }
    935 
    936 static void
    937 wg_init_key_and_hash(uint8_t ckey[WG_CHAINING_KEY_LEN],
    938     uint8_t hash[WG_HASH_LEN])
    939 {
    940 	/* [W] 5.4: CONSTRUCTION */
    941 	const char *signature = "Noise_IKpsk2_25519_ChaChaPoly_BLAKE2s";
    942 	/* [W] 5.4: IDENTIFIER */
    943 	const char *id = "WireGuard v1 zx2c4 Jason (at) zx2c4.com";
    944 	struct blake2s state;
    945 
    946 	blake2s(ckey, WG_CHAINING_KEY_LEN, NULL, 0,
    947 	    signature, strlen(signature));
    948 
    949 	CTASSERT(WG_HASH_LEN == WG_CHAINING_KEY_LEN);
    950 	memcpy(hash, ckey, WG_CHAINING_KEY_LEN);
    951 
    952 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
    953 	blake2s_update(&state, ckey, WG_CHAINING_KEY_LEN);
    954 	blake2s_update(&state, id, strlen(id));
    955 	blake2s_final(&state, hash);
    956 
    957 	WG_DUMP_HASH("ckey", ckey);
    958 	WG_DUMP_HASH("hash", hash);
    959 }
    960 
    961 static void
    962 wg_algo_hash(uint8_t hash[WG_HASH_LEN], const uint8_t input[],
    963     const size_t inputsize)
    964 {
    965 	struct blake2s state;
    966 
    967 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
    968 	blake2s_update(&state, hash, WG_HASH_LEN);
    969 	blake2s_update(&state, input, inputsize);
    970 	blake2s_final(&state, hash);
    971 }
    972 
    973 static void
    974 wg_algo_mac(uint8_t out[], const size_t outsize,
    975     const uint8_t key[], const size_t keylen,
    976     const uint8_t input1[], const size_t input1len,
    977     const uint8_t input2[], const size_t input2len)
    978 {
    979 	struct blake2s state;
    980 
    981 	blake2s_init(&state, outsize, key, keylen);
    982 
    983 	blake2s_update(&state, input1, input1len);
    984 	if (input2 != NULL)
    985 		blake2s_update(&state, input2, input2len);
    986 	blake2s_final(&state, out);
    987 }
    988 
    989 static void
    990 wg_algo_mac_mac1(uint8_t out[], const size_t outsize,
    991     const uint8_t input1[], const size_t input1len,
    992     const uint8_t input2[], const size_t input2len)
    993 {
    994 	struct blake2s state;
    995 	/* [W] 5.4: LABEL-MAC1 */
    996 	const char *label = "mac1----";
    997 	uint8_t key[WG_HASH_LEN];
    998 
    999 	blake2s_init(&state, sizeof(key), NULL, 0);
   1000 	blake2s_update(&state, label, strlen(label));
   1001 	blake2s_update(&state, input1, input1len);
   1002 	blake2s_final(&state, key);
   1003 
   1004 	blake2s_init(&state, outsize, key, sizeof(key));
   1005 	if (input2 != NULL)
   1006 		blake2s_update(&state, input2, input2len);
   1007 	blake2s_final(&state, out);
   1008 }
   1009 
   1010 static void
   1011 wg_algo_mac_cookie(uint8_t out[], const size_t outsize,
   1012     const uint8_t input1[], const size_t input1len)
   1013 {
   1014 	struct blake2s state;
   1015 	/* [W] 5.4: LABEL-COOKIE */
   1016 	const char *label = "cookie--";
   1017 
   1018 	blake2s_init(&state, outsize, NULL, 0);
   1019 	blake2s_update(&state, label, strlen(label));
   1020 	blake2s_update(&state, input1, input1len);
   1021 	blake2s_final(&state, out);
   1022 }
   1023 
   1024 static void
   1025 wg_algo_generate_keypair(uint8_t pubkey[WG_EPHEMERAL_KEY_LEN],
   1026     uint8_t privkey[WG_EPHEMERAL_KEY_LEN])
   1027 {
   1028 
   1029 	CTASSERT(WG_EPHEMERAL_KEY_LEN == crypto_scalarmult_curve25519_BYTES);
   1030 
   1031 	cprng_strong(kern_cprng, privkey, WG_EPHEMERAL_KEY_LEN, 0);
   1032 	crypto_scalarmult_base(pubkey, privkey);
   1033 }
   1034 
   1035 static void
   1036 wg_algo_dh(uint8_t out[WG_DH_OUTPUT_LEN],
   1037     const uint8_t privkey[WG_STATIC_KEY_LEN],
   1038     const uint8_t pubkey[WG_STATIC_KEY_LEN])
   1039 {
   1040 
   1041 	CTASSERT(WG_STATIC_KEY_LEN == crypto_scalarmult_curve25519_BYTES);
   1042 
   1043 	int ret __diagused = crypto_scalarmult(out, privkey, pubkey);
   1044 	KASSERT(ret == 0);
   1045 }
   1046 
   1047 static void
   1048 wg_algo_hmac(uint8_t out[], const size_t outlen,
   1049     const uint8_t key[], const size_t keylen,
   1050     const uint8_t in[], const size_t inlen)
   1051 {
   1052 #define IPAD	0x36
   1053 #define OPAD	0x5c
   1054 	uint8_t hmackey[HMAC_BLOCK_LEN] = {0};
   1055 	uint8_t ipad[HMAC_BLOCK_LEN];
   1056 	uint8_t opad[HMAC_BLOCK_LEN];
   1057 	size_t i;
   1058 	struct blake2s state;
   1059 
   1060 	KASSERT(outlen == WG_HASH_LEN);
   1061 	KASSERT(keylen <= HMAC_BLOCK_LEN);
   1062 
   1063 	memcpy(hmackey, key, keylen);
   1064 
   1065 	for (i = 0; i < sizeof(hmackey); i++) {
   1066 		ipad[i] = hmackey[i] ^ IPAD;
   1067 		opad[i] = hmackey[i] ^ OPAD;
   1068 	}
   1069 
   1070 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
   1071 	blake2s_update(&state, ipad, sizeof(ipad));
   1072 	blake2s_update(&state, in, inlen);
   1073 	blake2s_final(&state, out);
   1074 
   1075 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
   1076 	blake2s_update(&state, opad, sizeof(opad));
   1077 	blake2s_update(&state, out, WG_HASH_LEN);
   1078 	blake2s_final(&state, out);
   1079 #undef IPAD
   1080 #undef OPAD
   1081 }
   1082 
   1083 static void
   1084 wg_algo_kdf(uint8_t out1[WG_KDF_OUTPUT_LEN], uint8_t out2[WG_KDF_OUTPUT_LEN],
   1085     uint8_t out3[WG_KDF_OUTPUT_LEN], const uint8_t ckey[WG_CHAINING_KEY_LEN],
   1086     const uint8_t input[], const size_t inputlen)
   1087 {
   1088 	uint8_t tmp1[WG_KDF_OUTPUT_LEN], tmp2[WG_KDF_OUTPUT_LEN + 1];
   1089 	uint8_t one[1];
   1090 
   1091 	/*
   1092 	 * [N] 4.3: "an input_key_material byte sequence with length
   1093 	 * either zero bytes, 32 bytes, or DHLEN bytes."
   1094 	 */
   1095 	KASSERT(inputlen == 0 || inputlen == 32 || inputlen == NOISE_DHLEN);
   1096 
   1097 	WG_DUMP_HASH("ckey", ckey);
   1098 	if (input != NULL)
   1099 		WG_DUMP_HASH("input", input);
   1100 	wg_algo_hmac(tmp1, sizeof(tmp1), ckey, WG_CHAINING_KEY_LEN,
   1101 	    input, inputlen);
   1102 	WG_DUMP_HASH("tmp1", tmp1);
   1103 	one[0] = 1;
   1104 	wg_algo_hmac(out1, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
   1105 	    one, sizeof(one));
   1106 	WG_DUMP_HASH("out1", out1);
   1107 	if (out2 == NULL)
   1108 		return;
   1109 	memcpy(tmp2, out1, WG_KDF_OUTPUT_LEN);
   1110 	tmp2[WG_KDF_OUTPUT_LEN] = 2;
   1111 	wg_algo_hmac(out2, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
   1112 	    tmp2, sizeof(tmp2));
   1113 	WG_DUMP_HASH("out2", out2);
   1114 	if (out3 == NULL)
   1115 		return;
   1116 	memcpy(tmp2, out2, WG_KDF_OUTPUT_LEN);
   1117 	tmp2[WG_KDF_OUTPUT_LEN] = 3;
   1118 	wg_algo_hmac(out3, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
   1119 	    tmp2, sizeof(tmp2));
   1120 	WG_DUMP_HASH("out3", out3);
   1121 }
   1122 
   1123 static void __noinline
   1124 wg_algo_dh_kdf(uint8_t ckey[WG_CHAINING_KEY_LEN],
   1125     uint8_t cipher_key[WG_CIPHER_KEY_LEN],
   1126     const uint8_t local_key[WG_STATIC_KEY_LEN],
   1127     const uint8_t remote_key[WG_STATIC_KEY_LEN])
   1128 {
   1129 	uint8_t dhout[WG_DH_OUTPUT_LEN];
   1130 
   1131 	wg_algo_dh(dhout, local_key, remote_key);
   1132 	wg_algo_kdf(ckey, cipher_key, NULL, ckey, dhout, sizeof(dhout));
   1133 
   1134 	WG_DUMP_HASH("dhout", dhout);
   1135 	WG_DUMP_HASH("ckey", ckey);
   1136 	if (cipher_key != NULL)
   1137 		WG_DUMP_HASH("cipher_key", cipher_key);
   1138 }
   1139 
   1140 static void
   1141 wg_algo_aead_enc(uint8_t out[], size_t expected_outsize, const uint8_t key[],
   1142     const uint64_t counter, const uint8_t plain[], const size_t plainsize,
   1143     const uint8_t auth[], size_t authlen)
   1144 {
   1145 	uint8_t nonce[(32 + 64) / 8] = {0};
   1146 	long long unsigned int outsize;
   1147 	int error __diagused;
   1148 
   1149 	le64enc(&nonce[4], counter);
   1150 
   1151 	error = crypto_aead_chacha20poly1305_ietf_encrypt(out, &outsize, plain,
   1152 	    plainsize, auth, authlen, NULL, nonce, key);
   1153 	KASSERT(error == 0);
   1154 	KASSERT(outsize == expected_outsize);
   1155 }
   1156 
   1157 static int
   1158 wg_algo_aead_dec(uint8_t out[], size_t expected_outsize, const uint8_t key[],
   1159     const uint64_t counter, const uint8_t encrypted[],
   1160     const size_t encryptedsize, const uint8_t auth[], size_t authlen)
   1161 {
   1162 	uint8_t nonce[(32 + 64) / 8] = {0};
   1163 	long long unsigned int outsize;
   1164 	int error;
   1165 
   1166 	le64enc(&nonce[4], counter);
   1167 
   1168 	error = crypto_aead_chacha20poly1305_ietf_decrypt(out, &outsize, NULL,
   1169 	    encrypted, encryptedsize, auth, authlen, nonce, key);
   1170 	if (error == 0)
   1171 		KASSERT(outsize == expected_outsize);
   1172 	return error;
   1173 }
   1174 
   1175 static void
   1176 wg_algo_xaead_enc(uint8_t out[], const size_t expected_outsize,
   1177     const uint8_t key[], const uint8_t plain[], const size_t plainsize,
   1178     const uint8_t auth[], size_t authlen,
   1179     const uint8_t nonce[WG_SALT_LEN])
   1180 {
   1181 	long long unsigned int outsize;
   1182 	int error __diagused;
   1183 
   1184 	CTASSERT(WG_SALT_LEN == crypto_aead_xchacha20poly1305_ietf_NPUBBYTES);
   1185 	error = crypto_aead_xchacha20poly1305_ietf_encrypt(out, &outsize,
   1186 	    plain, plainsize, auth, authlen, NULL, nonce, key);
   1187 	KASSERT(error == 0);
   1188 	KASSERT(outsize == expected_outsize);
   1189 }
   1190 
   1191 static int
   1192 wg_algo_xaead_dec(uint8_t out[], const size_t expected_outsize,
   1193     const uint8_t key[], const uint8_t encrypted[], const size_t encryptedsize,
   1194     const uint8_t auth[], size_t authlen,
   1195     const uint8_t nonce[WG_SALT_LEN])
   1196 {
   1197 	long long unsigned int outsize;
   1198 	int error;
   1199 
   1200 	error = crypto_aead_xchacha20poly1305_ietf_decrypt(out, &outsize, NULL,
   1201 	    encrypted, encryptedsize, auth, authlen, nonce, key);
   1202 	if (error == 0)
   1203 		KASSERT(outsize == expected_outsize);
   1204 	return error;
   1205 }
   1206 
   1207 static void
   1208 wg_algo_tai64n(wg_timestamp_t timestamp)
   1209 {
   1210 	struct timespec ts;
   1211 
   1212 	/* FIXME strict TAI64N (https://cr.yp.to/libtai/tai64.html) */
   1213 	getnanotime(&ts);
   1214 	/* TAI64 label in external TAI64 format */
   1215 	be32enc(timestamp, 0x40000000U + (uint32_t)(ts.tv_sec >> 32));
   1216 	/* second beginning from 1970 TAI */
   1217 	be32enc(timestamp + 4, (uint32_t)(ts.tv_sec & 0xffffffffU));
   1218 	/* nanosecond in big-endian format */
   1219 	be32enc(timestamp + 8, (uint32_t)ts.tv_nsec);
   1220 }
   1221 
   1222 /*
   1223  * wg_get_stable_session(wgp, psref)
   1224  *
   1225  *	Get a passive reference to the current stable session, or
   1226  *	return NULL if there is no current stable session.
   1227  *
   1228  *	The pointer is always there but the session is not necessarily
   1229  *	ESTABLISHED; if it is not ESTABLISHED, return NULL.  However,
   1230  *	the session may transition from ESTABLISHED to DESTROYING while
   1231  *	holding the passive reference.
   1232  */
   1233 static struct wg_session *
   1234 wg_get_stable_session(struct wg_peer *wgp, struct psref *psref)
   1235 {
   1236 	int s;
   1237 	struct wg_session *wgs;
   1238 
   1239 	s = pserialize_read_enter();
   1240 	wgs = atomic_load_consume(&wgp->wgp_session_stable);
   1241 	if (__predict_false(wgs->wgs_state != WGS_STATE_ESTABLISHED))
   1242 		wgs = NULL;
   1243 	else
   1244 		psref_acquire(psref, &wgs->wgs_psref, wg_psref_class);
   1245 	pserialize_read_exit(s);
   1246 
   1247 	return wgs;
   1248 }
   1249 
   1250 static void
   1251 wg_put_session(struct wg_session *wgs, struct psref *psref)
   1252 {
   1253 
   1254 	psref_release(psref, &wgs->wgs_psref, wg_psref_class);
   1255 }
   1256 
   1257 static void
   1258 wg_destroy_session(struct wg_softc *wg, struct wg_session *wgs)
   1259 {
   1260 	struct wg_peer *wgp = wgs->wgs_peer;
   1261 	struct wg_session *wgs0 __diagused;
   1262 	void *garbage;
   1263 
   1264 	KASSERT(mutex_owned(wgp->wgp_lock));
   1265 	KASSERT(wgs->wgs_state != WGS_STATE_UNKNOWN);
   1266 
   1267 	/* Remove the session from the table.  */
   1268 	wgs0 = thmap_del(wg->wg_sessions_byindex,
   1269 	    &wgs->wgs_local_index, sizeof(wgs->wgs_local_index));
   1270 	KASSERT(wgs0 == wgs);
   1271 	garbage = thmap_stage_gc(wg->wg_sessions_byindex);
   1272 
   1273 	/* Wait for passive references to drain.  */
   1274 	pserialize_perform(wgp->wgp_psz);
   1275 	psref_target_destroy(&wgs->wgs_psref, wg_psref_class);
   1276 
   1277 	/* Free memory, zero state, and transition to UNKNOWN.  */
   1278 	thmap_gc(wg->wg_sessions_byindex, garbage);
   1279 	wg_clear_states(wgs);
   1280 	wgs->wgs_state = WGS_STATE_UNKNOWN;
   1281 }
   1282 
   1283 /*
   1284  * wg_get_session_index(wg, wgs)
   1285  *
   1286  *	Choose a session index for wgs->wgs_local_index, and store it
   1287  *	in wg's table of sessions by index.
   1288  *
   1289  *	wgs must be the unstable session of its peer, and must be
   1290  *	transitioning out of the UNKNOWN state.
   1291  */
   1292 static void
   1293 wg_get_session_index(struct wg_softc *wg, struct wg_session *wgs)
   1294 {
   1295 	struct wg_peer *wgp __diagused = wgs->wgs_peer;
   1296 	struct wg_session *wgs0;
   1297 	uint32_t index;
   1298 
   1299 	KASSERT(mutex_owned(wgp->wgp_lock));
   1300 	KASSERT(wgs == wgp->wgp_session_unstable);
   1301 	KASSERT(wgs->wgs_state == WGS_STATE_UNKNOWN);
   1302 
   1303 	do {
   1304 		/* Pick a uniform random index.  */
   1305 		index = cprng_strong32();
   1306 
   1307 		/* Try to take it.  */
   1308 		wgs->wgs_local_index = index;
   1309 		wgs0 = thmap_put(wg->wg_sessions_byindex,
   1310 		    &wgs->wgs_local_index, sizeof wgs->wgs_local_index, wgs);
   1311 
   1312 		/* If someone else beat us, start over.  */
   1313 	} while (__predict_false(wgs0 != wgs));
   1314 }
   1315 
   1316 /*
   1317  * wg_put_session_index(wg, wgs)
   1318  *
   1319  *	Remove wgs from the table of sessions by index, wait for any
   1320  *	passive references to drain, and transition the session to the
   1321  *	UNKNOWN state.
   1322  *
   1323  *	wgs must be the unstable session of its peer, and must not be
   1324  *	UNKNOWN or ESTABLISHED.
   1325  */
   1326 static void
   1327 wg_put_session_index(struct wg_softc *wg, struct wg_session *wgs)
   1328 {
   1329 	struct wg_peer *wgp __diagused = wgs->wgs_peer;
   1330 
   1331 	KASSERT(mutex_owned(wgp->wgp_lock));
   1332 	KASSERT(wgs == wgp->wgp_session_unstable);
   1333 	KASSERT(wgs->wgs_state != WGS_STATE_UNKNOWN);
   1334 	KASSERT(wgs->wgs_state != WGS_STATE_ESTABLISHED);
   1335 
   1336 	wg_destroy_session(wg, wgs);
   1337 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
   1338 }
   1339 
   1340 /*
   1341  * Handshake patterns
   1342  *
   1343  * [W] 5: "These messages use the "IK" pattern from Noise"
   1344  * [N] 7.5. Interactive handshake patterns (fundamental)
   1345  *     "The first character refers to the initiators static key:"
   1346  *     "I = Static key for initiator Immediately transmitted to responder,
   1347  *          despite reduced or absent identity hiding"
   1348  *     "The second character refers to the responders static key:"
   1349  *     "K = Static key for responder Known to initiator"
   1350  *     "IK:
   1351  *        <- s
   1352  *        ...
   1353  *        -> e, es, s, ss
   1354  *        <- e, ee, se"
   1355  * [N] 9.4. Pattern modifiers
   1356  *     "IKpsk2:
   1357  *        <- s
   1358  *        ...
   1359  *        -> e, es, s, ss
   1360  *        <- e, ee, se, psk"
   1361  */
   1362 static void
   1363 wg_fill_msg_init(struct wg_softc *wg, struct wg_peer *wgp,
   1364     struct wg_session *wgs, struct wg_msg_init *wgmi)
   1365 {
   1366 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.2: Ci */
   1367 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.2: Hi */
   1368 	uint8_t cipher_key[WG_CIPHER_KEY_LEN];
   1369 	uint8_t pubkey[WG_EPHEMERAL_KEY_LEN];
   1370 	uint8_t privkey[WG_EPHEMERAL_KEY_LEN];
   1371 
   1372 	KASSERT(mutex_owned(wgp->wgp_lock));
   1373 	KASSERT(wgs == wgp->wgp_session_unstable);
   1374 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_ACTIVE);
   1375 
   1376 	wgmi->wgmi_type = htole32(WG_MSG_TYPE_INIT);
   1377 	wgmi->wgmi_sender = wgs->wgs_local_index;
   1378 
   1379 	/* [W] 5.4.2: First Message: Initiator to Responder */
   1380 
   1381 	/* Ci := HASH(CONSTRUCTION) */
   1382 	/* Hi := HASH(Ci || IDENTIFIER) */
   1383 	wg_init_key_and_hash(ckey, hash);
   1384 	/* Hi := HASH(Hi || Sr^pub) */
   1385 	wg_algo_hash(hash, wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey));
   1386 
   1387 	WG_DUMP_HASH("hash", hash);
   1388 
   1389 	/* [N] 2.2: "e" */
   1390 	/* Ei^priv, Ei^pub := DH-GENERATE() */
   1391 	wg_algo_generate_keypair(pubkey, privkey);
   1392 	/* Ci := KDF1(Ci, Ei^pub) */
   1393 	wg_algo_kdf(ckey, NULL, NULL, ckey, pubkey, sizeof(pubkey));
   1394 	/* msg.ephemeral := Ei^pub */
   1395 	memcpy(wgmi->wgmi_ephemeral, pubkey, sizeof(wgmi->wgmi_ephemeral));
   1396 	/* Hi := HASH(Hi || msg.ephemeral) */
   1397 	wg_algo_hash(hash, pubkey, sizeof(pubkey));
   1398 
   1399 	WG_DUMP_HASH("ckey", ckey);
   1400 	WG_DUMP_HASH("hash", hash);
   1401 
   1402 	/* [N] 2.2: "es" */
   1403 	/* Ci, k := KDF2(Ci, DH(Ei^priv, Sr^pub)) */
   1404 	wg_algo_dh_kdf(ckey, cipher_key, privkey, wgp->wgp_pubkey);
   1405 
   1406 	/* [N] 2.2: "s" */
   1407 	/* msg.static := AEAD(k, 0, Si^pub, Hi) */
   1408 	wg_algo_aead_enc(wgmi->wgmi_static, sizeof(wgmi->wgmi_static),
   1409 	    cipher_key, 0, wg->wg_pubkey, sizeof(wg->wg_pubkey),
   1410 	    hash, sizeof(hash));
   1411 	/* Hi := HASH(Hi || msg.static) */
   1412 	wg_algo_hash(hash, wgmi->wgmi_static, sizeof(wgmi->wgmi_static));
   1413 
   1414 	WG_DUMP_HASH48("wgmi_static", wgmi->wgmi_static);
   1415 
   1416 	/* [N] 2.2: "ss" */
   1417 	/* Ci, k := KDF2(Ci, DH(Si^priv, Sr^pub)) */
   1418 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgp->wgp_pubkey);
   1419 
   1420 	/* msg.timestamp := AEAD(k, TIMESTAMP(), Hi) */
   1421 	wg_timestamp_t timestamp;
   1422 	wg_algo_tai64n(timestamp);
   1423 	wg_algo_aead_enc(wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp),
   1424 	    cipher_key, 0, timestamp, sizeof(timestamp), hash, sizeof(hash));
   1425 	/* Hi := HASH(Hi || msg.timestamp) */
   1426 	wg_algo_hash(hash, wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp));
   1427 
   1428 	/* [W] 5.4.4 Cookie MACs */
   1429 	wg_algo_mac_mac1(wgmi->wgmi_mac1, sizeof(wgmi->wgmi_mac1),
   1430 	    wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey),
   1431 	    (const uint8_t *)wgmi, offsetof(struct wg_msg_init, wgmi_mac1));
   1432 	/* Need mac1 to decrypt a cookie from a cookie message */
   1433 	memcpy(wgp->wgp_last_sent_mac1, wgmi->wgmi_mac1,
   1434 	    sizeof(wgp->wgp_last_sent_mac1));
   1435 	wgp->wgp_last_sent_mac1_valid = true;
   1436 
   1437 	if (wgp->wgp_latest_cookie_time == 0 ||
   1438 	    (time_uptime - wgp->wgp_latest_cookie_time) >= WG_COOKIE_TIME)
   1439 		memset(wgmi->wgmi_mac2, 0, sizeof(wgmi->wgmi_mac2));
   1440 	else {
   1441 		wg_algo_mac(wgmi->wgmi_mac2, sizeof(wgmi->wgmi_mac2),
   1442 		    wgp->wgp_latest_cookie, WG_COOKIE_LEN,
   1443 		    (const uint8_t *)wgmi,
   1444 		    offsetof(struct wg_msg_init, wgmi_mac2),
   1445 		    NULL, 0);
   1446 	}
   1447 
   1448 	memcpy(wgs->wgs_ephemeral_key_pub, pubkey, sizeof(pubkey));
   1449 	memcpy(wgs->wgs_ephemeral_key_priv, privkey, sizeof(privkey));
   1450 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
   1451 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
   1452 	WG_DLOG("%s: sender=%x\n", __func__, wgs->wgs_local_index);
   1453 }
   1454 
   1455 static void __noinline
   1456 wg_handle_msg_init(struct wg_softc *wg, const struct wg_msg_init *wgmi,
   1457     const struct sockaddr *src)
   1458 {
   1459 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.2: Ci */
   1460 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.2: Hi */
   1461 	uint8_t cipher_key[WG_CIPHER_KEY_LEN];
   1462 	uint8_t peer_pubkey[WG_STATIC_KEY_LEN];
   1463 	struct wg_peer *wgp;
   1464 	struct wg_session *wgs;
   1465 	int error, ret;
   1466 	struct psref psref_peer;
   1467 	uint8_t mac1[WG_MAC_LEN];
   1468 
   1469 	WG_TRACE("init msg received");
   1470 
   1471 	wg_algo_mac_mac1(mac1, sizeof(mac1),
   1472 	    wg->wg_pubkey, sizeof(wg->wg_pubkey),
   1473 	    (const uint8_t *)wgmi, offsetof(struct wg_msg_init, wgmi_mac1));
   1474 
   1475 	/*
   1476 	 * [W] 5.3: Denial of Service Mitigation & Cookies
   1477 	 * "the responder, ..., must always reject messages with an invalid
   1478 	 *  msg.mac1"
   1479 	 */
   1480 	if (!consttime_memequal(mac1, wgmi->wgmi_mac1, sizeof(mac1))) {
   1481 		WG_DLOG("mac1 is invalid\n");
   1482 		return;
   1483 	}
   1484 
   1485 	/*
   1486 	 * [W] 5.4.2: First Message: Initiator to Responder
   1487 	 * "When the responder receives this message, it does the same
   1488 	 *  operations so that its final state variables are identical,
   1489 	 *  replacing the operands of the DH function to produce equivalent
   1490 	 *  values."
   1491 	 *  Note that the following comments of operations are just copies of
   1492 	 *  the initiator's ones.
   1493 	 */
   1494 
   1495 	/* Ci := HASH(CONSTRUCTION) */
   1496 	/* Hi := HASH(Ci || IDENTIFIER) */
   1497 	wg_init_key_and_hash(ckey, hash);
   1498 	/* Hi := HASH(Hi || Sr^pub) */
   1499 	wg_algo_hash(hash, wg->wg_pubkey, sizeof(wg->wg_pubkey));
   1500 
   1501 	/* [N] 2.2: "e" */
   1502 	/* Ci := KDF1(Ci, Ei^pub) */
   1503 	wg_algo_kdf(ckey, NULL, NULL, ckey, wgmi->wgmi_ephemeral,
   1504 	    sizeof(wgmi->wgmi_ephemeral));
   1505 	/* Hi := HASH(Hi || msg.ephemeral) */
   1506 	wg_algo_hash(hash, wgmi->wgmi_ephemeral, sizeof(wgmi->wgmi_ephemeral));
   1507 
   1508 	WG_DUMP_HASH("ckey", ckey);
   1509 
   1510 	/* [N] 2.2: "es" */
   1511 	/* Ci, k := KDF2(Ci, DH(Ei^priv, Sr^pub)) */
   1512 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgmi->wgmi_ephemeral);
   1513 
   1514 	WG_DUMP_HASH48("wgmi_static", wgmi->wgmi_static);
   1515 
   1516 	/* [N] 2.2: "s" */
   1517 	/* msg.static := AEAD(k, 0, Si^pub, Hi) */
   1518 	error = wg_algo_aead_dec(peer_pubkey, WG_STATIC_KEY_LEN, cipher_key, 0,
   1519 	    wgmi->wgmi_static, sizeof(wgmi->wgmi_static), hash, sizeof(hash));
   1520 	if (error != 0) {
   1521 		WG_LOG_RATECHECK(&wg->wg_ppsratecheck, LOG_DEBUG,
   1522 		    "%s: wg_algo_aead_dec for secret key failed\n",
   1523 		    if_name(&wg->wg_if));
   1524 		return;
   1525 	}
   1526 	/* Hi := HASH(Hi || msg.static) */
   1527 	wg_algo_hash(hash, wgmi->wgmi_static, sizeof(wgmi->wgmi_static));
   1528 
   1529 	wgp = wg_lookup_peer_by_pubkey(wg, peer_pubkey, &psref_peer);
   1530 	if (wgp == NULL) {
   1531 		WG_DLOG("peer not found\n");
   1532 		return;
   1533 	}
   1534 
   1535 	/*
   1536 	 * Lock the peer to serialize access to cookie state.
   1537 	 *
   1538 	 * XXX Can we safely avoid holding the lock across DH?  Take it
   1539 	 * just to verify mac2 and then unlock/DH/lock?
   1540 	 */
   1541 	mutex_enter(wgp->wgp_lock);
   1542 
   1543 	if (__predict_false(wg_is_underload(wg, wgp, WG_MSG_TYPE_INIT))) {
   1544 		WG_TRACE("under load");
   1545 		/*
   1546 		 * [W] 5.3: Denial of Service Mitigation & Cookies
   1547 		 * "the responder, ..., and when under load may reject messages
   1548 		 *  with an invalid msg.mac2.  If the responder receives a
   1549 		 *  message with a valid msg.mac1 yet with an invalid msg.mac2,
   1550 		 *  and is under load, it may respond with a cookie reply
   1551 		 *  message"
   1552 		 */
   1553 		uint8_t zero[WG_MAC_LEN] = {0};
   1554 		if (consttime_memequal(wgmi->wgmi_mac2, zero, sizeof(zero))) {
   1555 			WG_TRACE("sending a cookie message: no cookie included");
   1556 			(void)wg_send_cookie_msg(wg, wgp, wgmi->wgmi_sender,
   1557 			    wgmi->wgmi_mac1, src);
   1558 			goto out;
   1559 		}
   1560 		if (!wgp->wgp_last_sent_cookie_valid) {
   1561 			WG_TRACE("sending a cookie message: no cookie sent ever");
   1562 			(void)wg_send_cookie_msg(wg, wgp, wgmi->wgmi_sender,
   1563 			    wgmi->wgmi_mac1, src);
   1564 			goto out;
   1565 		}
   1566 		uint8_t mac2[WG_MAC_LEN];
   1567 		wg_algo_mac(mac2, sizeof(mac2), wgp->wgp_last_sent_cookie,
   1568 		    WG_COOKIE_LEN, (const uint8_t *)wgmi,
   1569 		    offsetof(struct wg_msg_init, wgmi_mac2), NULL, 0);
   1570 		if (!consttime_memequal(mac2, wgmi->wgmi_mac2, sizeof(mac2))) {
   1571 			WG_DLOG("mac2 is invalid\n");
   1572 			goto out;
   1573 		}
   1574 		WG_TRACE("under load, but continue to sending");
   1575 	}
   1576 
   1577 	/* [N] 2.2: "ss" */
   1578 	/* Ci, k := KDF2(Ci, DH(Si^priv, Sr^pub)) */
   1579 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgp->wgp_pubkey);
   1580 
   1581 	/* msg.timestamp := AEAD(k, TIMESTAMP(), Hi) */
   1582 	wg_timestamp_t timestamp;
   1583 	error = wg_algo_aead_dec(timestamp, sizeof(timestamp), cipher_key, 0,
   1584 	    wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp),
   1585 	    hash, sizeof(hash));
   1586 	if (error != 0) {
   1587 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   1588 		    "%s: peer %s: wg_algo_aead_dec for timestamp failed\n",
   1589 		    if_name(&wg->wg_if), wgp->wgp_name);
   1590 		goto out;
   1591 	}
   1592 	/* Hi := HASH(Hi || msg.timestamp) */
   1593 	wg_algo_hash(hash, wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp));
   1594 
   1595 	/*
   1596 	 * [W] 5.1 "The responder keeps track of the greatest timestamp
   1597 	 *      received per peer and discards packets containing
   1598 	 *      timestamps less than or equal to it."
   1599 	 */
   1600 	ret = memcmp(timestamp, wgp->wgp_timestamp_latest_init,
   1601 	    sizeof(timestamp));
   1602 	if (ret <= 0) {
   1603 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   1604 		    "%s: peer %s: invalid init msg: timestamp is old\n",
   1605 		    if_name(&wg->wg_if), wgp->wgp_name);
   1606 		goto out;
   1607 	}
   1608 	memcpy(wgp->wgp_timestamp_latest_init, timestamp, sizeof(timestamp));
   1609 
   1610 	/*
   1611 	 * Message is good -- we're committing to handle it now, unless
   1612 	 * we were already initiating a session.
   1613 	 */
   1614 	wgs = wgp->wgp_session_unstable;
   1615 	switch (wgs->wgs_state) {
   1616 	case WGS_STATE_UNKNOWN:		/* new session initiated by peer */
   1617 		wg_get_session_index(wg, wgs);
   1618 		break;
   1619 	case WGS_STATE_INIT_ACTIVE:	/* we're already initiating, drop */
   1620 		WG_TRACE("Session already initializing, ignoring the message");
   1621 		goto out;
   1622 	case WGS_STATE_INIT_PASSIVE:	/* peer is retrying, start over */
   1623 		WG_TRACE("Session already initializing, destroying old states");
   1624 		wg_clear_states(wgs);
   1625 		/* keep session index */
   1626 		break;
   1627 	case WGS_STATE_ESTABLISHED:	/* can't happen */
   1628 		panic("unstable session can't be established");
   1629 		break;
   1630 	case WGS_STATE_DESTROYING:	/* rekey initiated by peer */
   1631 		WG_TRACE("Session destroying, but force to clear");
   1632 		callout_stop(&wgp->wgp_session_dtor_timer);
   1633 		wg_clear_states(wgs);
   1634 		/* keep session index */
   1635 		break;
   1636 	default:
   1637 		panic("invalid session state: %d", wgs->wgs_state);
   1638 	}
   1639 	wgs->wgs_state = WGS_STATE_INIT_PASSIVE;
   1640 
   1641 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
   1642 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
   1643 	memcpy(wgs->wgs_ephemeral_key_peer, wgmi->wgmi_ephemeral,
   1644 	    sizeof(wgmi->wgmi_ephemeral));
   1645 
   1646 	wg_update_endpoint_if_necessary(wgp, src);
   1647 
   1648 	(void)wg_send_handshake_msg_resp(wg, wgp, wgs, wgmi);
   1649 
   1650 	wg_calculate_keys(wgs, false);
   1651 	wg_clear_states(wgs);
   1652 
   1653 out:
   1654 	mutex_exit(wgp->wgp_lock);
   1655 	wg_put_peer(wgp, &psref_peer);
   1656 }
   1657 
   1658 static struct socket *
   1659 wg_get_so_by_af(struct wg_softc *wg, const int af)
   1660 {
   1661 
   1662 	switch (af) {
   1663 #ifdef INET
   1664 	case AF_INET:
   1665 		return wg->wg_so4;
   1666 #endif
   1667 #ifdef INET6
   1668 	case AF_INET6:
   1669 		return wg->wg_so6;
   1670 #endif
   1671 	default:
   1672 		panic("wg: no such af: %d", af);
   1673 	}
   1674 }
   1675 
   1676 static struct socket *
   1677 wg_get_so_by_peer(struct wg_peer *wgp, struct wg_sockaddr *wgsa)
   1678 {
   1679 
   1680 	return wg_get_so_by_af(wgp->wgp_sc, wgsa_family(wgsa));
   1681 }
   1682 
   1683 static struct wg_sockaddr *
   1684 wg_get_endpoint_sa(struct wg_peer *wgp, struct psref *psref)
   1685 {
   1686 	struct wg_sockaddr *wgsa;
   1687 	int s;
   1688 
   1689 	s = pserialize_read_enter();
   1690 	wgsa = atomic_load_consume(&wgp->wgp_endpoint);
   1691 	psref_acquire(psref, &wgsa->wgsa_psref, wg_psref_class);
   1692 	pserialize_read_exit(s);
   1693 
   1694 	return wgsa;
   1695 }
   1696 
   1697 static void
   1698 wg_put_sa(struct wg_peer *wgp, struct wg_sockaddr *wgsa, struct psref *psref)
   1699 {
   1700 
   1701 	psref_release(psref, &wgsa->wgsa_psref, wg_psref_class);
   1702 }
   1703 
   1704 static int
   1705 wg_send_so(struct wg_peer *wgp, struct mbuf *m)
   1706 {
   1707 	int error;
   1708 	struct socket *so;
   1709 	struct psref psref;
   1710 	struct wg_sockaddr *wgsa;
   1711 
   1712 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   1713 	so = wg_get_so_by_peer(wgp, wgsa);
   1714 	error = sosend(so, wgsatosa(wgsa), NULL, m, NULL, 0, curlwp);
   1715 	wg_put_sa(wgp, wgsa, &psref);
   1716 
   1717 	return error;
   1718 }
   1719 
   1720 static int
   1721 wg_send_handshake_msg_init(struct wg_softc *wg, struct wg_peer *wgp)
   1722 {
   1723 	int error;
   1724 	struct mbuf *m;
   1725 	struct wg_msg_init *wgmi;
   1726 	struct wg_session *wgs;
   1727 
   1728 	KASSERT(mutex_owned(wgp->wgp_lock));
   1729 
   1730 	wgs = wgp->wgp_session_unstable;
   1731 	/* XXX pull dispatch out into wg_task_send_init_message */
   1732 	switch (wgs->wgs_state) {
   1733 	case WGS_STATE_UNKNOWN:		/* new session initiated by us */
   1734 		wg_get_session_index(wg, wgs);
   1735 		break;
   1736 	case WGS_STATE_INIT_ACTIVE:	/* we're already initiating, stop */
   1737 		WG_TRACE("Session already initializing, skip starting new one");
   1738 		return EBUSY;
   1739 	case WGS_STATE_INIT_PASSIVE:	/* peer was trying -- XXX what now? */
   1740 		WG_TRACE("Session already initializing, destroying old states");
   1741 		wg_clear_states(wgs);
   1742 		/* keep session index */
   1743 		break;
   1744 	case WGS_STATE_ESTABLISHED:	/* can't happen */
   1745 		panic("unstable session can't be established");
   1746 		break;
   1747 	case WGS_STATE_DESTROYING:	/* rekey initiated by us too early */
   1748 		WG_TRACE("Session destroying");
   1749 		/* XXX should wait? */
   1750 		return EBUSY;
   1751 	}
   1752 	wgs->wgs_state = WGS_STATE_INIT_ACTIVE;
   1753 
   1754 	m = m_gethdr(M_WAIT, MT_DATA);
   1755 	if (sizeof(*wgmi) > MHLEN) {
   1756 		m_clget(m, M_WAIT);
   1757 		CTASSERT(sizeof(*wgmi) <= MCLBYTES);
   1758 	}
   1759 	m->m_pkthdr.len = m->m_len = sizeof(*wgmi);
   1760 	wgmi = mtod(m, struct wg_msg_init *);
   1761 	wg_fill_msg_init(wg, wgp, wgs, wgmi);
   1762 
   1763 	error = wg->wg_ops->send_hs_msg(wgp, m);
   1764 	if (error == 0) {
   1765 		WG_TRACE("init msg sent");
   1766 
   1767 		if (wgp->wgp_handshake_start_time == 0)
   1768 			wgp->wgp_handshake_start_time = time_uptime;
   1769 		callout_schedule(&wgp->wgp_handshake_timeout_timer,
   1770 		    MIN(wg_rekey_timeout, (unsigned)(INT_MAX / hz)) * hz);
   1771 	} else {
   1772 		wg_put_session_index(wg, wgs);
   1773 		/* Initiation failed; toss packet waiting for it if any.  */
   1774 		m = atomic_swap_ptr(&wgp->wgp_pending, NULL);
   1775 		m_freem(m);
   1776 	}
   1777 
   1778 	return error;
   1779 }
   1780 
   1781 static void
   1782 wg_fill_msg_resp(struct wg_softc *wg, struct wg_peer *wgp,
   1783     struct wg_session *wgs, struct wg_msg_resp *wgmr,
   1784     const struct wg_msg_init *wgmi)
   1785 {
   1786 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.3: Cr */
   1787 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.3: Hr */
   1788 	uint8_t cipher_key[WG_KDF_OUTPUT_LEN];
   1789 	uint8_t pubkey[WG_EPHEMERAL_KEY_LEN];
   1790 	uint8_t privkey[WG_EPHEMERAL_KEY_LEN];
   1791 
   1792 	KASSERT(mutex_owned(wgp->wgp_lock));
   1793 	KASSERT(wgs == wgp->wgp_session_unstable);
   1794 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_PASSIVE);
   1795 
   1796 	memcpy(hash, wgs->wgs_handshake_hash, sizeof(hash));
   1797 	memcpy(ckey, wgs->wgs_chaining_key, sizeof(ckey));
   1798 
   1799 	wgmr->wgmr_type = htole32(WG_MSG_TYPE_RESP);
   1800 	wgmr->wgmr_sender = wgs->wgs_local_index;
   1801 	wgmr->wgmr_receiver = wgmi->wgmi_sender;
   1802 
   1803 	/* [W] 5.4.3 Second Message: Responder to Initiator */
   1804 
   1805 	/* [N] 2.2: "e" */
   1806 	/* Er^priv, Er^pub := DH-GENERATE() */
   1807 	wg_algo_generate_keypair(pubkey, privkey);
   1808 	/* Cr := KDF1(Cr, Er^pub) */
   1809 	wg_algo_kdf(ckey, NULL, NULL, ckey, pubkey, sizeof(pubkey));
   1810 	/* msg.ephemeral := Er^pub */
   1811 	memcpy(wgmr->wgmr_ephemeral, pubkey, sizeof(wgmr->wgmr_ephemeral));
   1812 	/* Hr := HASH(Hr || msg.ephemeral) */
   1813 	wg_algo_hash(hash, pubkey, sizeof(pubkey));
   1814 
   1815 	WG_DUMP_HASH("ckey", ckey);
   1816 	WG_DUMP_HASH("hash", hash);
   1817 
   1818 	/* [N] 2.2: "ee" */
   1819 	/* Cr := KDF1(Cr, DH(Er^priv, Ei^pub)) */
   1820 	wg_algo_dh_kdf(ckey, NULL, privkey, wgs->wgs_ephemeral_key_peer);
   1821 
   1822 	/* [N] 2.2: "se" */
   1823 	/* Cr := KDF1(Cr, DH(Er^priv, Si^pub)) */
   1824 	wg_algo_dh_kdf(ckey, NULL, privkey, wgp->wgp_pubkey);
   1825 
   1826 	/* [N] 9.2: "psk" */
   1827     {
   1828 	uint8_t kdfout[WG_KDF_OUTPUT_LEN];
   1829 	/* Cr, r, k := KDF3(Cr, Q) */
   1830 	wg_algo_kdf(ckey, kdfout, cipher_key, ckey, wgp->wgp_psk,
   1831 	    sizeof(wgp->wgp_psk));
   1832 	/* Hr := HASH(Hr || r) */
   1833 	wg_algo_hash(hash, kdfout, sizeof(kdfout));
   1834     }
   1835 
   1836 	/* msg.empty := AEAD(k, 0, e, Hr) */
   1837 	wg_algo_aead_enc(wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty),
   1838 	    cipher_key, 0, NULL, 0, hash, sizeof(hash));
   1839 	/* Hr := HASH(Hr || msg.empty) */
   1840 	wg_algo_hash(hash, wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty));
   1841 
   1842 	WG_DUMP_HASH("wgmr_empty", wgmr->wgmr_empty);
   1843 
   1844 	/* [W] 5.4.4: Cookie MACs */
   1845 	/* msg.mac1 := MAC(HASH(LABEL-MAC1 || Sm'^pub), msg_a) */
   1846 	wg_algo_mac_mac1(wgmr->wgmr_mac1, sizeof(wgmi->wgmi_mac1),
   1847 	    wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey),
   1848 	    (const uint8_t *)wgmr, offsetof(struct wg_msg_resp, wgmr_mac1));
   1849 	/* Need mac1 to decrypt a cookie from a cookie message */
   1850 	memcpy(wgp->wgp_last_sent_mac1, wgmr->wgmr_mac1,
   1851 	    sizeof(wgp->wgp_last_sent_mac1));
   1852 	wgp->wgp_last_sent_mac1_valid = true;
   1853 
   1854 	if (wgp->wgp_latest_cookie_time == 0 ||
   1855 	    (time_uptime - wgp->wgp_latest_cookie_time) >= WG_COOKIE_TIME)
   1856 		/* msg.mac2 := 0^16 */
   1857 		memset(wgmr->wgmr_mac2, 0, sizeof(wgmr->wgmr_mac2));
   1858 	else {
   1859 		/* msg.mac2 := MAC(Lm, msg_b) */
   1860 		wg_algo_mac(wgmr->wgmr_mac2, sizeof(wgmi->wgmi_mac2),
   1861 		    wgp->wgp_latest_cookie, WG_COOKIE_LEN,
   1862 		    (const uint8_t *)wgmr,
   1863 		    offsetof(struct wg_msg_resp, wgmr_mac2),
   1864 		    NULL, 0);
   1865 	}
   1866 
   1867 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
   1868 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
   1869 	memcpy(wgs->wgs_ephemeral_key_pub, pubkey, sizeof(pubkey));
   1870 	memcpy(wgs->wgs_ephemeral_key_priv, privkey, sizeof(privkey));
   1871 	wgs->wgs_remote_index = wgmi->wgmi_sender;
   1872 	WG_DLOG("sender=%x\n", wgs->wgs_local_index);
   1873 	WG_DLOG("receiver=%x\n", wgs->wgs_remote_index);
   1874 }
   1875 
   1876 static void
   1877 wg_swap_sessions(struct wg_peer *wgp)
   1878 {
   1879 	struct wg_session *wgs, *wgs_prev;
   1880 
   1881 	KASSERT(mutex_owned(wgp->wgp_lock));
   1882 
   1883 	wgs = wgp->wgp_session_unstable;
   1884 	KASSERT(wgs->wgs_state == WGS_STATE_ESTABLISHED);
   1885 
   1886 	wgs_prev = wgp->wgp_session_stable;
   1887 	KASSERT(wgs_prev->wgs_state == WGS_STATE_ESTABLISHED ||
   1888 	    wgs_prev->wgs_state == WGS_STATE_UNKNOWN);
   1889 	atomic_store_release(&wgp->wgp_session_stable, wgs);
   1890 	wgp->wgp_session_unstable = wgs_prev;
   1891 }
   1892 
   1893 static void __noinline
   1894 wg_handle_msg_resp(struct wg_softc *wg, const struct wg_msg_resp *wgmr,
   1895     const struct sockaddr *src)
   1896 {
   1897 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.3: Cr */
   1898 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.3: Kr */
   1899 	uint8_t cipher_key[WG_KDF_OUTPUT_LEN];
   1900 	struct wg_peer *wgp;
   1901 	struct wg_session *wgs;
   1902 	struct psref psref;
   1903 	int error;
   1904 	uint8_t mac1[WG_MAC_LEN];
   1905 	struct wg_session *wgs_prev;
   1906 	struct mbuf *m;
   1907 
   1908 	wg_algo_mac_mac1(mac1, sizeof(mac1),
   1909 	    wg->wg_pubkey, sizeof(wg->wg_pubkey),
   1910 	    (const uint8_t *)wgmr, offsetof(struct wg_msg_resp, wgmr_mac1));
   1911 
   1912 	/*
   1913 	 * [W] 5.3: Denial of Service Mitigation & Cookies
   1914 	 * "the responder, ..., must always reject messages with an invalid
   1915 	 *  msg.mac1"
   1916 	 */
   1917 	if (!consttime_memequal(mac1, wgmr->wgmr_mac1, sizeof(mac1))) {
   1918 		WG_DLOG("mac1 is invalid\n");
   1919 		return;
   1920 	}
   1921 
   1922 	WG_TRACE("resp msg received");
   1923 	wgs = wg_lookup_session_by_index(wg, wgmr->wgmr_receiver, &psref);
   1924 	if (wgs == NULL) {
   1925 		WG_TRACE("No session found");
   1926 		return;
   1927 	}
   1928 
   1929 	wgp = wgs->wgs_peer;
   1930 
   1931 	mutex_enter(wgp->wgp_lock);
   1932 
   1933 	/* If we weren't waiting for a handshake response, drop it.  */
   1934 	if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE) {
   1935 		WG_TRACE("peer sent spurious handshake response, ignoring");
   1936 		goto out;
   1937 	}
   1938 
   1939 	if (__predict_false(wg_is_underload(wg, wgp, WG_MSG_TYPE_RESP))) {
   1940 		WG_TRACE("under load");
   1941 		/*
   1942 		 * [W] 5.3: Denial of Service Mitigation & Cookies
   1943 		 * "the responder, ..., and when under load may reject messages
   1944 		 *  with an invalid msg.mac2.  If the responder receives a
   1945 		 *  message with a valid msg.mac1 yet with an invalid msg.mac2,
   1946 		 *  and is under load, it may respond with a cookie reply
   1947 		 *  message"
   1948 		 */
   1949 		uint8_t zero[WG_MAC_LEN] = {0};
   1950 		if (consttime_memequal(wgmr->wgmr_mac2, zero, sizeof(zero))) {
   1951 			WG_TRACE("sending a cookie message: no cookie included");
   1952 			(void)wg_send_cookie_msg(wg, wgp, wgmr->wgmr_sender,
   1953 			    wgmr->wgmr_mac1, src);
   1954 			goto out;
   1955 		}
   1956 		if (!wgp->wgp_last_sent_cookie_valid) {
   1957 			WG_TRACE("sending a cookie message: no cookie sent ever");
   1958 			(void)wg_send_cookie_msg(wg, wgp, wgmr->wgmr_sender,
   1959 			    wgmr->wgmr_mac1, src);
   1960 			goto out;
   1961 		}
   1962 		uint8_t mac2[WG_MAC_LEN];
   1963 		wg_algo_mac(mac2, sizeof(mac2), wgp->wgp_last_sent_cookie,
   1964 		    WG_COOKIE_LEN, (const uint8_t *)wgmr,
   1965 		    offsetof(struct wg_msg_resp, wgmr_mac2), NULL, 0);
   1966 		if (!consttime_memequal(mac2, wgmr->wgmr_mac2, sizeof(mac2))) {
   1967 			WG_DLOG("mac2 is invalid\n");
   1968 			goto out;
   1969 		}
   1970 		WG_TRACE("under load, but continue to sending");
   1971 	}
   1972 
   1973 	memcpy(hash, wgs->wgs_handshake_hash, sizeof(hash));
   1974 	memcpy(ckey, wgs->wgs_chaining_key, sizeof(ckey));
   1975 
   1976 	/*
   1977 	 * [W] 5.4.3 Second Message: Responder to Initiator
   1978 	 * "When the initiator receives this message, it does the same
   1979 	 *  operations so that its final state variables are identical,
   1980 	 *  replacing the operands of the DH function to produce equivalent
   1981 	 *  values."
   1982 	 *  Note that the following comments of operations are just copies of
   1983 	 *  the initiator's ones.
   1984 	 */
   1985 
   1986 	/* [N] 2.2: "e" */
   1987 	/* Cr := KDF1(Cr, Er^pub) */
   1988 	wg_algo_kdf(ckey, NULL, NULL, ckey, wgmr->wgmr_ephemeral,
   1989 	    sizeof(wgmr->wgmr_ephemeral));
   1990 	/* Hr := HASH(Hr || msg.ephemeral) */
   1991 	wg_algo_hash(hash, wgmr->wgmr_ephemeral, sizeof(wgmr->wgmr_ephemeral));
   1992 
   1993 	WG_DUMP_HASH("ckey", ckey);
   1994 	WG_DUMP_HASH("hash", hash);
   1995 
   1996 	/* [N] 2.2: "ee" */
   1997 	/* Cr := KDF1(Cr, DH(Er^priv, Ei^pub)) */
   1998 	wg_algo_dh_kdf(ckey, NULL, wgs->wgs_ephemeral_key_priv,
   1999 	    wgmr->wgmr_ephemeral);
   2000 
   2001 	/* [N] 2.2: "se" */
   2002 	/* Cr := KDF1(Cr, DH(Er^priv, Si^pub)) */
   2003 	wg_algo_dh_kdf(ckey, NULL, wg->wg_privkey, wgmr->wgmr_ephemeral);
   2004 
   2005 	/* [N] 9.2: "psk" */
   2006     {
   2007 	uint8_t kdfout[WG_KDF_OUTPUT_LEN];
   2008 	/* Cr, r, k := KDF3(Cr, Q) */
   2009 	wg_algo_kdf(ckey, kdfout, cipher_key, ckey, wgp->wgp_psk,
   2010 	    sizeof(wgp->wgp_psk));
   2011 	/* Hr := HASH(Hr || r) */
   2012 	wg_algo_hash(hash, kdfout, sizeof(kdfout));
   2013     }
   2014 
   2015     {
   2016 	uint8_t out[sizeof(wgmr->wgmr_empty)]; /* for safety */
   2017 	/* msg.empty := AEAD(k, 0, e, Hr) */
   2018 	error = wg_algo_aead_dec(out, 0, cipher_key, 0, wgmr->wgmr_empty,
   2019 	    sizeof(wgmr->wgmr_empty), hash, sizeof(hash));
   2020 	WG_DUMP_HASH("wgmr_empty", wgmr->wgmr_empty);
   2021 	if (error != 0) {
   2022 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2023 		    "%s: peer %s: wg_algo_aead_dec for empty message failed\n",
   2024 		    if_name(&wg->wg_if), wgp->wgp_name);
   2025 		goto out;
   2026 	}
   2027 	/* Hr := HASH(Hr || msg.empty) */
   2028 	wg_algo_hash(hash, wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty));
   2029     }
   2030 
   2031 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(wgs->wgs_handshake_hash));
   2032 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(wgs->wgs_chaining_key));
   2033 	wgs->wgs_remote_index = wgmr->wgmr_sender;
   2034 	WG_DLOG("receiver=%x\n", wgs->wgs_remote_index);
   2035 
   2036 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_ACTIVE);
   2037 	wgs->wgs_state = WGS_STATE_ESTABLISHED;
   2038 	wgs->wgs_time_established = time_uptime;
   2039 	wgs->wgs_time_last_data_sent = 0;
   2040 	wgs->wgs_is_initiator = true;
   2041 	wg_calculate_keys(wgs, true);
   2042 	wg_clear_states(wgs);
   2043 	WG_TRACE("WGS_STATE_ESTABLISHED");
   2044 
   2045 	callout_stop(&wgp->wgp_handshake_timeout_timer);
   2046 
   2047 	wg_swap_sessions(wgp);
   2048 	KASSERT(wgs == wgp->wgp_session_stable);
   2049 	wgs_prev = wgp->wgp_session_unstable;
   2050 	getnanotime(&wgp->wgp_last_handshake_time);
   2051 	wgp->wgp_handshake_start_time = 0;
   2052 	wgp->wgp_last_sent_mac1_valid = false;
   2053 	wgp->wgp_last_sent_cookie_valid = false;
   2054 
   2055 	wg_schedule_rekey_timer(wgp);
   2056 
   2057 	wg_update_endpoint_if_necessary(wgp, src);
   2058 
   2059 	/*
   2060 	 * If we had a data packet queued up, send it; otherwise send a
   2061 	 * keepalive message -- either way we have to send something
   2062 	 * immediately or else the responder will never answer.
   2063 	 */
   2064 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL) {
   2065 		kpreempt_disable();
   2066 		const uint32_t h = curcpu()->ci_index; // pktq_rps_hash(m)
   2067 		M_SETCTX(m, wgp);
   2068 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   2069 			WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   2070 			    if_name(&wg->wg_if));
   2071 			m_freem(m);
   2072 		}
   2073 		kpreempt_enable();
   2074 	} else {
   2075 		wg_send_keepalive_msg(wgp, wgs);
   2076 	}
   2077 
   2078 	if (wgs_prev->wgs_state == WGS_STATE_ESTABLISHED) {
   2079 		/* Wait for wg_get_stable_session to drain.  */
   2080 		pserialize_perform(wgp->wgp_psz);
   2081 
   2082 		/* Transition ESTABLISHED->DESTROYING.  */
   2083 		wgs_prev->wgs_state = WGS_STATE_DESTROYING;
   2084 
   2085 		/* We can't destroy the old session immediately */
   2086 		wg_schedule_session_dtor_timer(wgp);
   2087 	} else {
   2088 		KASSERTMSG(wgs_prev->wgs_state == WGS_STATE_UNKNOWN,
   2089 		    "state=%d", wgs_prev->wgs_state);
   2090 	}
   2091 
   2092 out:
   2093 	mutex_exit(wgp->wgp_lock);
   2094 	wg_put_session(wgs, &psref);
   2095 }
   2096 
   2097 static int
   2098 wg_send_handshake_msg_resp(struct wg_softc *wg, struct wg_peer *wgp,
   2099     struct wg_session *wgs, const struct wg_msg_init *wgmi)
   2100 {
   2101 	int error;
   2102 	struct mbuf *m;
   2103 	struct wg_msg_resp *wgmr;
   2104 
   2105 	KASSERT(mutex_owned(wgp->wgp_lock));
   2106 	KASSERT(wgs == wgp->wgp_session_unstable);
   2107 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_PASSIVE);
   2108 
   2109 	m = m_gethdr(M_WAIT, MT_DATA);
   2110 	if (sizeof(*wgmr) > MHLEN) {
   2111 		m_clget(m, M_WAIT);
   2112 		CTASSERT(sizeof(*wgmr) <= MCLBYTES);
   2113 	}
   2114 	m->m_pkthdr.len = m->m_len = sizeof(*wgmr);
   2115 	wgmr = mtod(m, struct wg_msg_resp *);
   2116 	wg_fill_msg_resp(wg, wgp, wgs, wgmr, wgmi);
   2117 
   2118 	error = wg->wg_ops->send_hs_msg(wgp, m);
   2119 	if (error == 0)
   2120 		WG_TRACE("resp msg sent");
   2121 	return error;
   2122 }
   2123 
   2124 static struct wg_peer *
   2125 wg_lookup_peer_by_pubkey(struct wg_softc *wg,
   2126     const uint8_t pubkey[WG_STATIC_KEY_LEN], struct psref *psref)
   2127 {
   2128 	struct wg_peer *wgp;
   2129 
   2130 	int s = pserialize_read_enter();
   2131 	wgp = thmap_get(wg->wg_peers_bypubkey, pubkey, WG_STATIC_KEY_LEN);
   2132 	if (wgp != NULL)
   2133 		wg_get_peer(wgp, psref);
   2134 	pserialize_read_exit(s);
   2135 
   2136 	return wgp;
   2137 }
   2138 
   2139 static void
   2140 wg_fill_msg_cookie(struct wg_softc *wg, struct wg_peer *wgp,
   2141     struct wg_msg_cookie *wgmc, const uint32_t sender,
   2142     const uint8_t mac1[WG_MAC_LEN], const struct sockaddr *src)
   2143 {
   2144 	uint8_t cookie[WG_COOKIE_LEN];
   2145 	uint8_t key[WG_HASH_LEN];
   2146 	uint8_t addr[sizeof(struct in6_addr)];
   2147 	size_t addrlen;
   2148 	uint16_t uh_sport; /* be */
   2149 
   2150 	KASSERT(mutex_owned(wgp->wgp_lock));
   2151 
   2152 	wgmc->wgmc_type = htole32(WG_MSG_TYPE_COOKIE);
   2153 	wgmc->wgmc_receiver = sender;
   2154 	cprng_fast(wgmc->wgmc_salt, sizeof(wgmc->wgmc_salt));
   2155 
   2156 	/*
   2157 	 * [W] 5.4.7: Under Load: Cookie Reply Message
   2158 	 * "The secret variable, Rm, changes every two minutes to a
   2159 	 * random value"
   2160 	 */
   2161 	if ((time_uptime - wgp->wgp_last_genrandval_time) > WG_RANDVAL_TIME) {
   2162 		wgp->wgp_randval = cprng_strong32();
   2163 		wgp->wgp_last_genrandval_time = time_uptime;
   2164 	}
   2165 
   2166 	switch (src->sa_family) {
   2167 	case AF_INET: {
   2168 		const struct sockaddr_in *sin = satocsin(src);
   2169 		addrlen = sizeof(sin->sin_addr);
   2170 		memcpy(addr, &sin->sin_addr, addrlen);
   2171 		uh_sport = sin->sin_port;
   2172 		break;
   2173 	    }
   2174 #ifdef INET6
   2175 	case AF_INET6: {
   2176 		const struct sockaddr_in6 *sin6 = satocsin6(src);
   2177 		addrlen = sizeof(sin6->sin6_addr);
   2178 		memcpy(addr, &sin6->sin6_addr, addrlen);
   2179 		uh_sport = sin6->sin6_port;
   2180 		break;
   2181 	    }
   2182 #endif
   2183 	default:
   2184 		panic("invalid af=%d", src->sa_family);
   2185 	}
   2186 
   2187 	wg_algo_mac(cookie, sizeof(cookie),
   2188 	    (const uint8_t *)&wgp->wgp_randval, sizeof(wgp->wgp_randval),
   2189 	    addr, addrlen, (const uint8_t *)&uh_sport, sizeof(uh_sport));
   2190 	wg_algo_mac_cookie(key, sizeof(key), wg->wg_pubkey,
   2191 	    sizeof(wg->wg_pubkey));
   2192 	wg_algo_xaead_enc(wgmc->wgmc_cookie, sizeof(wgmc->wgmc_cookie), key,
   2193 	    cookie, sizeof(cookie), mac1, WG_MAC_LEN, wgmc->wgmc_salt);
   2194 
   2195 	/* Need to store to calculate mac2 */
   2196 	memcpy(wgp->wgp_last_sent_cookie, cookie, sizeof(cookie));
   2197 	wgp->wgp_last_sent_cookie_valid = true;
   2198 }
   2199 
   2200 static int
   2201 wg_send_cookie_msg(struct wg_softc *wg, struct wg_peer *wgp,
   2202     const uint32_t sender, const uint8_t mac1[WG_MAC_LEN],
   2203     const struct sockaddr *src)
   2204 {
   2205 	int error;
   2206 	struct mbuf *m;
   2207 	struct wg_msg_cookie *wgmc;
   2208 
   2209 	KASSERT(mutex_owned(wgp->wgp_lock));
   2210 
   2211 	m = m_gethdr(M_WAIT, MT_DATA);
   2212 	if (sizeof(*wgmc) > MHLEN) {
   2213 		m_clget(m, M_WAIT);
   2214 		CTASSERT(sizeof(*wgmc) <= MCLBYTES);
   2215 	}
   2216 	m->m_pkthdr.len = m->m_len = sizeof(*wgmc);
   2217 	wgmc = mtod(m, struct wg_msg_cookie *);
   2218 	wg_fill_msg_cookie(wg, wgp, wgmc, sender, mac1, src);
   2219 
   2220 	error = wg->wg_ops->send_hs_msg(wgp, m);
   2221 	if (error == 0)
   2222 		WG_TRACE("cookie msg sent");
   2223 	return error;
   2224 }
   2225 
   2226 static bool
   2227 wg_is_underload(struct wg_softc *wg, struct wg_peer *wgp, int msgtype)
   2228 {
   2229 #ifdef WG_DEBUG_PARAMS
   2230 	if (wg_force_underload)
   2231 		return true;
   2232 #endif
   2233 
   2234 	/*
   2235 	 * XXX we don't have a means of a load estimation.  The purpose of
   2236 	 * the mechanism is a DoS mitigation, so we consider frequent handshake
   2237 	 * messages as (a kind of) load; if a message of the same type comes
   2238 	 * to a peer within 1 second, we consider we are under load.
   2239 	 */
   2240 	time_t last = wgp->wgp_last_msg_received_time[msgtype];
   2241 	wgp->wgp_last_msg_received_time[msgtype] = time_uptime;
   2242 	return (time_uptime - last) == 0;
   2243 }
   2244 
   2245 static void
   2246 wg_calculate_keys(struct wg_session *wgs, const bool initiator)
   2247 {
   2248 
   2249 	KASSERT(mutex_owned(wgs->wgs_peer->wgp_lock));
   2250 
   2251 	/*
   2252 	 * [W] 5.4.5: Ti^send = Tr^recv, Ti^recv = Tr^send := KDF2(Ci = Cr, e)
   2253 	 */
   2254 	if (initiator) {
   2255 		wg_algo_kdf(wgs->wgs_tkey_send, wgs->wgs_tkey_recv, NULL,
   2256 		    wgs->wgs_chaining_key, NULL, 0);
   2257 	} else {
   2258 		wg_algo_kdf(wgs->wgs_tkey_recv, wgs->wgs_tkey_send, NULL,
   2259 		    wgs->wgs_chaining_key, NULL, 0);
   2260 	}
   2261 	WG_DUMP_HASH("wgs_tkey_send", wgs->wgs_tkey_send);
   2262 	WG_DUMP_HASH("wgs_tkey_recv", wgs->wgs_tkey_recv);
   2263 }
   2264 
   2265 static uint64_t
   2266 wg_session_get_send_counter(struct wg_session *wgs)
   2267 {
   2268 #ifdef __HAVE_ATOMIC64_LOADSTORE
   2269 	return atomic_load_relaxed(&wgs->wgs_send_counter);
   2270 #else
   2271 	uint64_t send_counter;
   2272 
   2273 	mutex_enter(&wgs->wgs_send_counter_lock);
   2274 	send_counter = wgs->wgs_send_counter;
   2275 	mutex_exit(&wgs->wgs_send_counter_lock);
   2276 
   2277 	return send_counter;
   2278 #endif
   2279 }
   2280 
   2281 static uint64_t
   2282 wg_session_inc_send_counter(struct wg_session *wgs)
   2283 {
   2284 #ifdef __HAVE_ATOMIC64_LOADSTORE
   2285 	return atomic_inc_64_nv(&wgs->wgs_send_counter) - 1;
   2286 #else
   2287 	uint64_t send_counter;
   2288 
   2289 	mutex_enter(&wgs->wgs_send_counter_lock);
   2290 	send_counter = wgs->wgs_send_counter++;
   2291 	mutex_exit(&wgs->wgs_send_counter_lock);
   2292 
   2293 	return send_counter;
   2294 #endif
   2295 }
   2296 
   2297 static void
   2298 wg_clear_states(struct wg_session *wgs)
   2299 {
   2300 
   2301 	KASSERT(mutex_owned(wgs->wgs_peer->wgp_lock));
   2302 
   2303 	wgs->wgs_send_counter = 0;
   2304 	sliwin_reset(&wgs->wgs_recvwin->window);
   2305 
   2306 #define wgs_clear(v)	explicit_memset(wgs->wgs_##v, 0, sizeof(wgs->wgs_##v))
   2307 	wgs_clear(handshake_hash);
   2308 	wgs_clear(chaining_key);
   2309 	wgs_clear(ephemeral_key_pub);
   2310 	wgs_clear(ephemeral_key_priv);
   2311 	wgs_clear(ephemeral_key_peer);
   2312 #undef wgs_clear
   2313 }
   2314 
   2315 static struct wg_session *
   2316 wg_lookup_session_by_index(struct wg_softc *wg, const uint32_t index,
   2317     struct psref *psref)
   2318 {
   2319 	struct wg_session *wgs;
   2320 
   2321 	int s = pserialize_read_enter();
   2322 	wgs = thmap_get(wg->wg_sessions_byindex, &index, sizeof index);
   2323 	if (wgs != NULL) {
   2324 		KASSERT(atomic_load_relaxed(&wgs->wgs_state) !=
   2325 		    WGS_STATE_UNKNOWN);
   2326 		psref_acquire(psref, &wgs->wgs_psref, wg_psref_class);
   2327 	}
   2328 	pserialize_read_exit(s);
   2329 
   2330 	return wgs;
   2331 }
   2332 
   2333 static void
   2334 wg_schedule_rekey_timer(struct wg_peer *wgp)
   2335 {
   2336 	int timeout = MIN(wg_rekey_after_time, (unsigned)(INT_MAX / hz));
   2337 
   2338 	callout_schedule(&wgp->wgp_rekey_timer, timeout * hz);
   2339 }
   2340 
   2341 static void
   2342 wg_send_keepalive_msg(struct wg_peer *wgp, struct wg_session *wgs)
   2343 {
   2344 	struct mbuf *m;
   2345 
   2346 	/*
   2347 	 * [W] 6.5 Passive Keepalive
   2348 	 * "A keepalive message is simply a transport data message with
   2349 	 *  a zero-length encapsulated encrypted inner-packet."
   2350 	 */
   2351 	WG_TRACE("");
   2352 	m = m_gethdr(M_WAIT, MT_DATA);
   2353 	wg_send_data_msg(wgp, wgs, m);
   2354 }
   2355 
   2356 static bool
   2357 wg_need_to_send_init_message(struct wg_session *wgs)
   2358 {
   2359 	/*
   2360 	 * [W] 6.2 Transport Message Limits
   2361 	 * "if a peer is the initiator of a current secure session,
   2362 	 *  WireGuard will send a handshake initiation message to begin
   2363 	 *  a new secure session ... if after receiving a transport data
   2364 	 *  message, the current secure session is (REJECT-AFTER-TIME 
   2365 	 *  KEEPALIVE-TIMEOUT  REKEY-TIMEOUT) seconds old and it has
   2366 	 *  not yet acted upon this event."
   2367 	 */
   2368 	return wgs->wgs_is_initiator && wgs->wgs_time_last_data_sent == 0 &&
   2369 	    (time_uptime - wgs->wgs_time_established) >=
   2370 	    (wg_reject_after_time - wg_keepalive_timeout - wg_rekey_timeout);
   2371 }
   2372 
   2373 static void
   2374 wg_schedule_peer_task(struct wg_peer *wgp, unsigned int task)
   2375 {
   2376 
   2377 	mutex_enter(wgp->wgp_intr_lock);
   2378 	WG_DLOG("tasks=%d, task=%d\n", wgp->wgp_tasks, task);
   2379 	if (wgp->wgp_tasks == 0)
   2380 		/*
   2381 		 * XXX If the current CPU is already loaded -- e.g., if
   2382 		 * there's already a bunch of handshakes queued up --
   2383 		 * consider tossing this over to another CPU to
   2384 		 * distribute the load.
   2385 		 */
   2386 		workqueue_enqueue(wg_wq, &wgp->wgp_work, NULL);
   2387 	wgp->wgp_tasks |= task;
   2388 	mutex_exit(wgp->wgp_intr_lock);
   2389 }
   2390 
   2391 static void
   2392 wg_change_endpoint(struct wg_peer *wgp, const struct sockaddr *new)
   2393 {
   2394 	struct wg_sockaddr *wgsa_prev;
   2395 
   2396 	WG_TRACE("Changing endpoint");
   2397 
   2398 	memcpy(wgp->wgp_endpoint0, new, new->sa_len);
   2399 	wgsa_prev = wgp->wgp_endpoint;
   2400 	atomic_store_release(&wgp->wgp_endpoint, wgp->wgp_endpoint0);
   2401 	wgp->wgp_endpoint0 = wgsa_prev;
   2402 	atomic_store_release(&wgp->wgp_endpoint_available, true);
   2403 
   2404 	wg_schedule_peer_task(wgp, WGP_TASK_ENDPOINT_CHANGED);
   2405 }
   2406 
   2407 static bool
   2408 wg_validate_inner_packet(const char *packet, size_t decrypted_len, int *af)
   2409 {
   2410 	uint16_t packet_len;
   2411 	const struct ip *ip;
   2412 
   2413 	if (__predict_false(decrypted_len < sizeof(*ip))) {
   2414 		WG_DLOG("decrypted_len=%zu < %zu\n", decrypted_len,
   2415 		    sizeof(*ip));
   2416 		return false;
   2417 	}
   2418 
   2419 	ip = (const struct ip *)packet;
   2420 	if (ip->ip_v == 4)
   2421 		*af = AF_INET;
   2422 	else if (ip->ip_v == 6)
   2423 		*af = AF_INET6;
   2424 	else {
   2425 		WG_DLOG("ip_v=%d\n", ip->ip_v);
   2426 		return false;
   2427 	}
   2428 
   2429 	WG_DLOG("af=%d\n", *af);
   2430 
   2431 	switch (*af) {
   2432 #ifdef INET
   2433 	case AF_INET:
   2434 		packet_len = ntohs(ip->ip_len);
   2435 		break;
   2436 #endif
   2437 #ifdef INET6
   2438 	case AF_INET6: {
   2439 		const struct ip6_hdr *ip6;
   2440 
   2441 		if (__predict_false(decrypted_len < sizeof(*ip6))) {
   2442 			WG_DLOG("decrypted_len=%zu < %zu\n", decrypted_len,
   2443 			    sizeof(*ip6));
   2444 			return false;
   2445 		}
   2446 
   2447 		ip6 = (const struct ip6_hdr *)packet;
   2448 		packet_len = sizeof(*ip6) + ntohs(ip6->ip6_plen);
   2449 		break;
   2450 	}
   2451 #endif
   2452 	default:
   2453 		return false;
   2454 	}
   2455 
   2456 	if (packet_len > decrypted_len) {
   2457 		WG_DLOG("packet_len %u > decrypted_len %zu\n", packet_len,
   2458 		    decrypted_len);
   2459 		return false;
   2460 	}
   2461 
   2462 	return true;
   2463 }
   2464 
   2465 static bool
   2466 wg_validate_route(struct wg_softc *wg, struct wg_peer *wgp_expected,
   2467     int af, char *packet)
   2468 {
   2469 	struct sockaddr_storage ss;
   2470 	struct sockaddr *sa;
   2471 	struct psref psref;
   2472 	struct wg_peer *wgp;
   2473 	bool ok;
   2474 
   2475 	/*
   2476 	 * II CRYPTOKEY ROUTING
   2477 	 * "it will only accept it if its source IP resolves in the
   2478 	 *  table to the public key used in the secure session for
   2479 	 *  decrypting it."
   2480 	 */
   2481 
   2482 	if (af == AF_INET) {
   2483 		const struct ip *ip = (const struct ip *)packet;
   2484 		struct sockaddr_in *sin = (struct sockaddr_in *)&ss;
   2485 		sockaddr_in_init(sin, &ip->ip_src, 0);
   2486 		sa = sintosa(sin);
   2487 #ifdef INET6
   2488 	} else {
   2489 		const struct ip6_hdr *ip6 = (const struct ip6_hdr *)packet;
   2490 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ss;
   2491 		sockaddr_in6_init(sin6, &ip6->ip6_src, 0, 0, 0);
   2492 		sa = sin6tosa(sin6);
   2493 #endif
   2494 	}
   2495 
   2496 	wgp = wg_pick_peer_by_sa(wg, sa, &psref);
   2497 	ok = (wgp == wgp_expected);
   2498 	if (wgp != NULL)
   2499 		wg_put_peer(wgp, &psref);
   2500 
   2501 	return ok;
   2502 }
   2503 
   2504 static void
   2505 wg_session_dtor_timer(void *arg)
   2506 {
   2507 	struct wg_peer *wgp = arg;
   2508 
   2509 	WG_TRACE("enter");
   2510 
   2511 	wg_schedule_peer_task(wgp, WGP_TASK_DESTROY_PREV_SESSION);
   2512 }
   2513 
   2514 static void
   2515 wg_schedule_session_dtor_timer(struct wg_peer *wgp)
   2516 {
   2517 
   2518 	/* 1 second grace period */
   2519 	callout_schedule(&wgp->wgp_session_dtor_timer, hz);
   2520 }
   2521 
   2522 static bool
   2523 sockaddr_port_match(const struct sockaddr *sa1, const struct sockaddr *sa2)
   2524 {
   2525 	if (sa1->sa_family != sa2->sa_family)
   2526 		return false;
   2527 
   2528 	switch (sa1->sa_family) {
   2529 #ifdef INET
   2530 	case AF_INET:
   2531 		return satocsin(sa1)->sin_port == satocsin(sa2)->sin_port;
   2532 #endif
   2533 #ifdef INET6
   2534 	case AF_INET6:
   2535 		return satocsin6(sa1)->sin6_port == satocsin6(sa2)->sin6_port;
   2536 #endif
   2537 	default:
   2538 		return false;
   2539 	}
   2540 }
   2541 
   2542 static void
   2543 wg_update_endpoint_if_necessary(struct wg_peer *wgp,
   2544     const struct sockaddr *src)
   2545 {
   2546 	struct wg_sockaddr *wgsa;
   2547 	struct psref psref;
   2548 
   2549 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   2550 
   2551 #ifdef WG_DEBUG_LOG
   2552 	char oldaddr[128], newaddr[128];
   2553 	sockaddr_format(wgsatosa(wgsa), oldaddr, sizeof(oldaddr));
   2554 	sockaddr_format(src, newaddr, sizeof(newaddr));
   2555 	WG_DLOG("old=%s, new=%s\n", oldaddr, newaddr);
   2556 #endif
   2557 
   2558 	/*
   2559 	 * III: "Since the packet has authenticated correctly, the source IP of
   2560 	 * the outer UDP/IP packet is used to update the endpoint for peer..."
   2561 	 */
   2562 	if (__predict_false(sockaddr_cmp(src, wgsatosa(wgsa)) != 0 ||
   2563 		!sockaddr_port_match(src, wgsatosa(wgsa)))) {
   2564 		/* XXX We can't change the endpoint twice in a short period */
   2565 		if (atomic_swap_uint(&wgp->wgp_endpoint_changing, 1) == 0) {
   2566 			wg_change_endpoint(wgp, src);
   2567 		}
   2568 	}
   2569 
   2570 	wg_put_sa(wgp, wgsa, &psref);
   2571 }
   2572 
   2573 static void __noinline
   2574 wg_handle_msg_data(struct wg_softc *wg, struct mbuf *m,
   2575     const struct sockaddr *src)
   2576 {
   2577 	struct wg_msg_data *wgmd;
   2578 	char *encrypted_buf = NULL, *decrypted_buf;
   2579 	size_t encrypted_len, decrypted_len;
   2580 	struct wg_session *wgs;
   2581 	struct wg_peer *wgp;
   2582 	int state;
   2583 	size_t mlen;
   2584 	struct psref psref;
   2585 	int error, af;
   2586 	bool success, free_encrypted_buf = false, ok;
   2587 	struct mbuf *n;
   2588 
   2589 	KASSERT(m->m_len >= sizeof(struct wg_msg_data));
   2590 	wgmd = mtod(m, struct wg_msg_data *);
   2591 
   2592 	KASSERT(wgmd->wgmd_type == htole32(WG_MSG_TYPE_DATA));
   2593 	WG_TRACE("data");
   2594 
   2595 	/* Find the putative session, or drop.  */
   2596 	wgs = wg_lookup_session_by_index(wg, wgmd->wgmd_receiver, &psref);
   2597 	if (wgs == NULL) {
   2598 		WG_TRACE("No session found");
   2599 		m_freem(m);
   2600 		return;
   2601 	}
   2602 
   2603 	/*
   2604 	 * We are only ready to handle data when in INIT_PASSIVE,
   2605 	 * ESTABLISHED, or DESTROYING.  All transitions out of that
   2606 	 * state dissociate the session index and drain psrefs.
   2607 	 */
   2608 	state = atomic_load_relaxed(&wgs->wgs_state);
   2609 	switch (state) {
   2610 	case WGS_STATE_UNKNOWN:
   2611 		panic("wg session %p in unknown state has session index %u",
   2612 		    wgs, wgmd->wgmd_receiver);
   2613 	case WGS_STATE_INIT_ACTIVE:
   2614 		WG_TRACE("not yet ready for data");
   2615 		goto out;
   2616 	case WGS_STATE_INIT_PASSIVE:
   2617 	case WGS_STATE_ESTABLISHED:
   2618 	case WGS_STATE_DESTROYING:
   2619 		break;
   2620 	}
   2621 
   2622 	/*
   2623 	 * Get the peer, for rate-limited logs (XXX MPSAFE, dtrace) and
   2624 	 * to update the endpoint if authentication succeeds.
   2625 	 */
   2626 	wgp = wgs->wgs_peer;
   2627 
   2628 	/*
   2629 	 * Reject outrageously wrong sequence numbers before doing any
   2630 	 * crypto work or taking any locks.
   2631 	 */
   2632 	error = sliwin_check_fast(&wgs->wgs_recvwin->window,
   2633 	    le64toh(wgmd->wgmd_counter));
   2634 	if (error) {
   2635 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2636 		    "%s: peer %s: out-of-window packet: %"PRIu64"\n",
   2637 		    if_name(&wg->wg_if), wgp->wgp_name,
   2638 		    le64toh(wgmd->wgmd_counter));
   2639 		goto out;
   2640 	}
   2641 
   2642 	/* Ensure the payload and authenticator are contiguous.  */
   2643 	mlen = m_length(m);
   2644 	encrypted_len = mlen - sizeof(*wgmd);
   2645 	if (encrypted_len < WG_AUTHTAG_LEN) {
   2646 		WG_DLOG("Short encrypted_len: %zu\n", encrypted_len);
   2647 		goto out;
   2648 	}
   2649 	success = m_ensure_contig(&m, sizeof(*wgmd) + encrypted_len);
   2650 	if (success) {
   2651 		encrypted_buf = mtod(m, char *) + sizeof(*wgmd);
   2652 	} else {
   2653 		encrypted_buf = kmem_intr_alloc(encrypted_len, KM_NOSLEEP);
   2654 		if (encrypted_buf == NULL) {
   2655 			WG_DLOG("failed to allocate encrypted_buf\n");
   2656 			goto out;
   2657 		}
   2658 		m_copydata(m, sizeof(*wgmd), encrypted_len, encrypted_buf);
   2659 		free_encrypted_buf = true;
   2660 	}
   2661 	/* m_ensure_contig may change m regardless of its result */
   2662 	KASSERT(m->m_len >= sizeof(*wgmd));
   2663 	wgmd = mtod(m, struct wg_msg_data *);
   2664 
   2665 #ifdef WG_DEBUG_PACKET
   2666 	if (wg_debug & WG_DEBUG_FLAGS_PACKET) {
   2667 		hexdump(printf, "incoming packet", encrypted_buf,
   2668 		    encrypted_len);
   2669 	}
   2670 #endif
   2671 	/*
   2672 	 * Get a buffer for the plaintext.  Add WG_AUTHTAG_LEN to avoid
   2673 	 * a zero-length buffer (XXX).  Drop if plaintext is longer
   2674 	 * than MCLBYTES (XXX).
   2675 	 */
   2676 	decrypted_len = encrypted_len - WG_AUTHTAG_LEN;
   2677 	if (decrypted_len > MCLBYTES) {
   2678 		/* FIXME handle larger data than MCLBYTES */
   2679 		WG_DLOG("couldn't handle larger data than MCLBYTES\n");
   2680 		goto out;
   2681 	}
   2682 	n = wg_get_mbuf(0, decrypted_len + WG_AUTHTAG_LEN);
   2683 	if (n == NULL) {
   2684 		WG_DLOG("wg_get_mbuf failed\n");
   2685 		goto out;
   2686 	}
   2687 	decrypted_buf = mtod(n, char *);
   2688 
   2689 	/* Decrypt and verify the packet.  */
   2690 	WG_DLOG("mlen=%zu, encrypted_len=%zu\n", mlen, encrypted_len);
   2691 	error = wg_algo_aead_dec(decrypted_buf,
   2692 	    encrypted_len - WG_AUTHTAG_LEN /* can be 0 */,
   2693 	    wgs->wgs_tkey_recv, le64toh(wgmd->wgmd_counter), encrypted_buf,
   2694 	    encrypted_len, NULL, 0);
   2695 	if (error != 0) {
   2696 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2697 		    "%s: peer %s: failed to wg_algo_aead_dec\n",
   2698 		    if_name(&wg->wg_if), wgp->wgp_name);
   2699 		m_freem(n);
   2700 		goto out;
   2701 	}
   2702 	WG_DLOG("outsize=%u\n", (u_int)decrypted_len);
   2703 
   2704 	/* Packet is genuine.  Reject it if a replay or just too old.  */
   2705 	mutex_enter(&wgs->wgs_recvwin->lock);
   2706 	error = sliwin_update(&wgs->wgs_recvwin->window,
   2707 	    le64toh(wgmd->wgmd_counter));
   2708 	mutex_exit(&wgs->wgs_recvwin->lock);
   2709 	if (error) {
   2710 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2711 		    "%s: peer %s: replay or out-of-window packet: %"PRIu64"\n",
   2712 		    if_name(&wg->wg_if), wgp->wgp_name,
   2713 		    le64toh(wgmd->wgmd_counter));
   2714 		m_freem(n);
   2715 		goto out;
   2716 	}
   2717 
   2718 #ifdef WG_DEBUG_PACKET
   2719 	if (wg_debug & WG_DEBUG_FLAGS_PACKET) {
   2720 		hexdump(printf, "tkey_recv", wgs->wgs_tkey_recv,
   2721 		    sizeof(wgs->wgs_tkey_recv));
   2722 		hexdump(printf, "wgmd", wgmd, sizeof(*wgmd));
   2723 		hexdump(printf, "decrypted_buf", decrypted_buf,
   2724 		    decrypted_len);
   2725 	}
   2726 #endif
   2727 	/* We're done with m now; free it and chuck the pointers.  */
   2728 	m_freem(m);
   2729 	m = NULL;
   2730 	wgmd = NULL;
   2731 
   2732 	/*
   2733 	 * Validate the encapsulated packet header and get the address
   2734 	 * family, or drop.
   2735 	 */
   2736 	ok = wg_validate_inner_packet(decrypted_buf, decrypted_len, &af);
   2737 	if (!ok) {
   2738 		m_freem(n);
   2739 		goto out;
   2740 	}
   2741 
   2742 	/*
   2743 	 * The packet is genuine.  Update the peer's endpoint if the
   2744 	 * source address changed.
   2745 	 *
   2746 	 * XXX How to prevent DoS by replaying genuine packets from the
   2747 	 * wrong source address?
   2748 	 */
   2749 	wg_update_endpoint_if_necessary(wgp, src);
   2750 
   2751 	/* Submit it into our network stack if routable.  */
   2752 	ok = wg_validate_route(wg, wgp, af, decrypted_buf);
   2753 	if (ok) {
   2754 		wg->wg_ops->input(&wg->wg_if, n, af);
   2755 	} else {
   2756 		char addrstr[INET6_ADDRSTRLEN];
   2757 		memset(addrstr, 0, sizeof(addrstr));
   2758 		if (af == AF_INET) {
   2759 			const struct ip *ip = (const struct ip *)decrypted_buf;
   2760 			IN_PRINT(addrstr, &ip->ip_src);
   2761 #ifdef INET6
   2762 		} else if (af == AF_INET6) {
   2763 			const struct ip6_hdr *ip6 =
   2764 			    (const struct ip6_hdr *)decrypted_buf;
   2765 			IN6_PRINT(addrstr, &ip6->ip6_src);
   2766 #endif
   2767 		}
   2768 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2769 		    "%s: peer %s: invalid source address (%s)\n",
   2770 		    if_name(&wg->wg_if), wgp->wgp_name, addrstr);
   2771 		m_freem(n);
   2772 		/*
   2773 		 * The inner address is invalid however the session is valid
   2774 		 * so continue the session processing below.
   2775 		 */
   2776 	}
   2777 	n = NULL;
   2778 
   2779 	/* Update the state machine if necessary.  */
   2780 	if (__predict_false(state == WGS_STATE_INIT_PASSIVE)) {
   2781 		/*
   2782 		 * We were waiting for the initiator to send their
   2783 		 * first data transport message, and that has happened.
   2784 		 * Schedule a task to establish this session.
   2785 		 */
   2786 		wg_schedule_peer_task(wgp, WGP_TASK_ESTABLISH_SESSION);
   2787 	} else {
   2788 		if (__predict_false(wg_need_to_send_init_message(wgs))) {
   2789 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   2790 		}
   2791 		/*
   2792 		 * [W] 6.5 Passive Keepalive
   2793 		 * "If a peer has received a validly-authenticated transport
   2794 		 *  data message (section 5.4.6), but does not have any packets
   2795 		 *  itself to send back for KEEPALIVE-TIMEOUT seconds, it sends
   2796 		 *  a keepalive message."
   2797 		 */
   2798 		WG_DLOG("time_uptime=%ju wgs_time_last_data_sent=%ju\n",
   2799 		    (uintmax_t)time_uptime,
   2800 		    (uintmax_t)wgs->wgs_time_last_data_sent);
   2801 		if ((time_uptime - wgs->wgs_time_last_data_sent) >=
   2802 		    wg_keepalive_timeout) {
   2803 			WG_TRACE("Schedule sending keepalive message");
   2804 			/*
   2805 			 * We can't send a keepalive message here to avoid
   2806 			 * a deadlock;  we already hold the solock of a socket
   2807 			 * that is used to send the message.
   2808 			 */
   2809 			wg_schedule_peer_task(wgp,
   2810 			    WGP_TASK_SEND_KEEPALIVE_MESSAGE);
   2811 		}
   2812 	}
   2813 out:
   2814 	wg_put_session(wgs, &psref);
   2815 	m_freem(m);
   2816 	if (free_encrypted_buf)
   2817 		kmem_intr_free(encrypted_buf, encrypted_len);
   2818 }
   2819 
   2820 static void __noinline
   2821 wg_handle_msg_cookie(struct wg_softc *wg, const struct wg_msg_cookie *wgmc)
   2822 {
   2823 	struct wg_session *wgs;
   2824 	struct wg_peer *wgp;
   2825 	struct psref psref;
   2826 	int error;
   2827 	uint8_t key[WG_HASH_LEN];
   2828 	uint8_t cookie[WG_COOKIE_LEN];
   2829 
   2830 	WG_TRACE("cookie msg received");
   2831 
   2832 	/* Find the putative session.  */
   2833 	wgs = wg_lookup_session_by_index(wg, wgmc->wgmc_receiver, &psref);
   2834 	if (wgs == NULL) {
   2835 		WG_TRACE("No session found");
   2836 		return;
   2837 	}
   2838 
   2839 	/* Lock the peer so we can update the cookie state.  */
   2840 	wgp = wgs->wgs_peer;
   2841 	mutex_enter(wgp->wgp_lock);
   2842 
   2843 	if (!wgp->wgp_last_sent_mac1_valid) {
   2844 		WG_TRACE("No valid mac1 sent (or expired)");
   2845 		goto out;
   2846 	}
   2847 
   2848 	/* Decrypt the cookie and store it for later handshake retry.  */
   2849 	wg_algo_mac_cookie(key, sizeof(key), wgp->wgp_pubkey,
   2850 	    sizeof(wgp->wgp_pubkey));
   2851 	error = wg_algo_xaead_dec(cookie, sizeof(cookie), key,
   2852 	    wgmc->wgmc_cookie, sizeof(wgmc->wgmc_cookie),
   2853 	    wgp->wgp_last_sent_mac1, sizeof(wgp->wgp_last_sent_mac1),
   2854 	    wgmc->wgmc_salt);
   2855 	if (error != 0) {
   2856 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2857 		    "%s: peer %s: wg_algo_aead_dec for cookie failed: "
   2858 		    "error=%d\n", if_name(&wg->wg_if), wgp->wgp_name, error);
   2859 		goto out;
   2860 	}
   2861 	/*
   2862 	 * [W] 6.6: Interaction with Cookie Reply System
   2863 	 * "it should simply store the decrypted cookie value from the cookie
   2864 	 *  reply message, and wait for the expiration of the REKEY-TIMEOUT
   2865 	 *  timer for retrying a handshake initiation message."
   2866 	 */
   2867 	wgp->wgp_latest_cookie_time = time_uptime;
   2868 	memcpy(wgp->wgp_latest_cookie, cookie, sizeof(wgp->wgp_latest_cookie));
   2869 out:
   2870 	mutex_exit(wgp->wgp_lock);
   2871 	wg_put_session(wgs, &psref);
   2872 }
   2873 
   2874 static struct mbuf *
   2875 wg_validate_msg_header(struct wg_softc *wg, struct mbuf *m)
   2876 {
   2877 	struct wg_msg wgm;
   2878 	size_t mbuflen;
   2879 	size_t msglen;
   2880 
   2881 	/*
   2882 	 * Get the mbuf chain length.  It is already guaranteed, by
   2883 	 * wg_overudp_cb, to be large enough for a struct wg_msg.
   2884 	 */
   2885 	mbuflen = m_length(m);
   2886 	KASSERT(mbuflen >= sizeof(struct wg_msg));
   2887 
   2888 	/*
   2889 	 * Copy the message header (32-bit message type) out -- we'll
   2890 	 * worry about contiguity and alignment later.
   2891 	 */
   2892 	m_copydata(m, 0, sizeof(wgm), &wgm);
   2893 	switch (le32toh(wgm.wgm_type)) {
   2894 	case WG_MSG_TYPE_INIT:
   2895 		msglen = sizeof(struct wg_msg_init);
   2896 		break;
   2897 	case WG_MSG_TYPE_RESP:
   2898 		msglen = sizeof(struct wg_msg_resp);
   2899 		break;
   2900 	case WG_MSG_TYPE_COOKIE:
   2901 		msglen = sizeof(struct wg_msg_cookie);
   2902 		break;
   2903 	case WG_MSG_TYPE_DATA:
   2904 		msglen = sizeof(struct wg_msg_data);
   2905 		break;
   2906 	default:
   2907 		WG_LOG_RATECHECK(&wg->wg_ppsratecheck, LOG_DEBUG,
   2908 		    "%s: Unexpected msg type: %u\n", if_name(&wg->wg_if),
   2909 		    le32toh(wgm.wgm_type));
   2910 		goto error;
   2911 	}
   2912 
   2913 	/* Verify the mbuf chain is long enough for this type of message.  */
   2914 	if (__predict_false(mbuflen < msglen)) {
   2915 		WG_DLOG("Invalid msg size: mbuflen=%zu type=%u\n", mbuflen,
   2916 		    le32toh(wgm.wgm_type));
   2917 		goto error;
   2918 	}
   2919 
   2920 	/* Make the message header contiguous if necessary.  */
   2921 	if (__predict_false(m->m_len < msglen)) {
   2922 		m = m_pullup(m, msglen);
   2923 		if (m == NULL)
   2924 			return NULL;
   2925 	}
   2926 
   2927 	return m;
   2928 
   2929 error:
   2930 	m_freem(m);
   2931 	return NULL;
   2932 }
   2933 
   2934 static void
   2935 wg_handle_packet(struct wg_softc *wg, struct mbuf *m,
   2936     const struct sockaddr *src)
   2937 {
   2938 	struct wg_msg *wgm;
   2939 
   2940 	KASSERT(curlwp->l_pflag & LP_BOUND);
   2941 
   2942 	m = wg_validate_msg_header(wg, m);
   2943 	if (__predict_false(m == NULL))
   2944 		return;
   2945 
   2946 	KASSERT(m->m_len >= sizeof(struct wg_msg));
   2947 	wgm = mtod(m, struct wg_msg *);
   2948 	switch (le32toh(wgm->wgm_type)) {
   2949 	case WG_MSG_TYPE_INIT:
   2950 		wg_handle_msg_init(wg, (struct wg_msg_init *)wgm, src);
   2951 		break;
   2952 	case WG_MSG_TYPE_RESP:
   2953 		wg_handle_msg_resp(wg, (struct wg_msg_resp *)wgm, src);
   2954 		break;
   2955 	case WG_MSG_TYPE_COOKIE:
   2956 		wg_handle_msg_cookie(wg, (struct wg_msg_cookie *)wgm);
   2957 		break;
   2958 	case WG_MSG_TYPE_DATA:
   2959 		wg_handle_msg_data(wg, m, src);
   2960 		/* wg_handle_msg_data frees m for us */
   2961 		return;
   2962 	default:
   2963 		panic("invalid message type: %d", le32toh(wgm->wgm_type));
   2964 	}
   2965 
   2966 	m_freem(m);
   2967 }
   2968 
   2969 static void
   2970 wg_receive_packets(struct wg_softc *wg, const int af)
   2971 {
   2972 
   2973 	for (;;) {
   2974 		int error, flags;
   2975 		struct socket *so;
   2976 		struct mbuf *m = NULL;
   2977 		struct uio dummy_uio;
   2978 		struct mbuf *paddr = NULL;
   2979 		struct sockaddr *src;
   2980 
   2981 		so = wg_get_so_by_af(wg, af);
   2982 		flags = MSG_DONTWAIT;
   2983 		dummy_uio.uio_resid = 1000000000;
   2984 
   2985 		error = so->so_receive(so, &paddr, &dummy_uio, &m, NULL,
   2986 		    &flags);
   2987 		if (error || m == NULL) {
   2988 			//if (error == EWOULDBLOCK)
   2989 			return;
   2990 		}
   2991 
   2992 		KASSERT(paddr != NULL);
   2993 		KASSERT(paddr->m_len >= sizeof(struct sockaddr));
   2994 		src = mtod(paddr, struct sockaddr *);
   2995 
   2996 		wg_handle_packet(wg, m, src);
   2997 	}
   2998 }
   2999 
   3000 static void
   3001 wg_get_peer(struct wg_peer *wgp, struct psref *psref)
   3002 {
   3003 
   3004 	psref_acquire(psref, &wgp->wgp_psref, wg_psref_class);
   3005 }
   3006 
   3007 static void
   3008 wg_put_peer(struct wg_peer *wgp, struct psref *psref)
   3009 {
   3010 
   3011 	psref_release(psref, &wgp->wgp_psref, wg_psref_class);
   3012 }
   3013 
   3014 static void
   3015 wg_task_send_init_message(struct wg_softc *wg, struct wg_peer *wgp)
   3016 {
   3017 	struct wg_session *wgs;
   3018 
   3019 	WG_TRACE("WGP_TASK_SEND_INIT_MESSAGE");
   3020 
   3021 	KASSERT(mutex_owned(wgp->wgp_lock));
   3022 
   3023 	if (!atomic_load_acquire(&wgp->wgp_endpoint_available)) {
   3024 		WGLOG(LOG_DEBUG, "%s: No endpoint available\n",
   3025 		    if_name(&wg->wg_if));
   3026 		/* XXX should do something? */
   3027 		return;
   3028 	}
   3029 
   3030 	wgs = wgp->wgp_session_stable;
   3031 	if (wgs->wgs_state == WGS_STATE_UNKNOWN) {
   3032 		/* XXX What if the unstable session is already INIT_ACTIVE?  */
   3033 		wg_send_handshake_msg_init(wg, wgp);
   3034 	} else {
   3035 		/* rekey */
   3036 		wgs = wgp->wgp_session_unstable;
   3037 		if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE)
   3038 			wg_send_handshake_msg_init(wg, wgp);
   3039 	}
   3040 }
   3041 
   3042 static void
   3043 wg_task_retry_handshake(struct wg_softc *wg, struct wg_peer *wgp)
   3044 {
   3045 	struct wg_session *wgs;
   3046 
   3047 	WG_TRACE("WGP_TASK_RETRY_HANDSHAKE");
   3048 
   3049 	KASSERT(mutex_owned(wgp->wgp_lock));
   3050 	KASSERT(wgp->wgp_handshake_start_time != 0);
   3051 
   3052 	wgs = wgp->wgp_session_unstable;
   3053 	if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE)
   3054 		return;
   3055 
   3056 	/*
   3057 	 * XXX no real need to assign a new index here, but we do need
   3058 	 * to transition to UNKNOWN temporarily
   3059 	 */
   3060 	wg_put_session_index(wg, wgs);
   3061 
   3062 	/* [W] 6.4 Handshake Initiation Retransmission */
   3063 	if ((time_uptime - wgp->wgp_handshake_start_time) >
   3064 	    wg_rekey_attempt_time) {
   3065 		/* Give up handshaking */
   3066 		wgp->wgp_handshake_start_time = 0;
   3067 		WG_TRACE("give up");
   3068 
   3069 		/*
   3070 		 * If a new data packet comes, handshaking will be retried
   3071 		 * and a new session would be established at that time,
   3072 		 * however we don't want to send pending packets then.
   3073 		 */
   3074 		wg_purge_pending_packets(wgp);
   3075 		return;
   3076 	}
   3077 
   3078 	wg_task_send_init_message(wg, wgp);
   3079 }
   3080 
   3081 static void
   3082 wg_task_establish_session(struct wg_softc *wg, struct wg_peer *wgp)
   3083 {
   3084 	struct wg_session *wgs, *wgs_prev;
   3085 	struct mbuf *m;
   3086 
   3087 	KASSERT(mutex_owned(wgp->wgp_lock));
   3088 
   3089 	wgs = wgp->wgp_session_unstable;
   3090 	if (wgs->wgs_state != WGS_STATE_INIT_PASSIVE)
   3091 		/* XXX Can this happen?  */
   3092 		return;
   3093 
   3094 	wgs->wgs_state = WGS_STATE_ESTABLISHED;
   3095 	wgs->wgs_time_established = time_uptime;
   3096 	wgs->wgs_time_last_data_sent = 0;
   3097 	wgs->wgs_is_initiator = false;
   3098 	WG_TRACE("WGS_STATE_ESTABLISHED");
   3099 
   3100 	wg_swap_sessions(wgp);
   3101 	KASSERT(wgs == wgp->wgp_session_stable);
   3102 	wgs_prev = wgp->wgp_session_unstable;
   3103 	getnanotime(&wgp->wgp_last_handshake_time);
   3104 	wgp->wgp_handshake_start_time = 0;
   3105 	wgp->wgp_last_sent_mac1_valid = false;
   3106 	wgp->wgp_last_sent_cookie_valid = false;
   3107 
   3108 	/* If we had a data packet queued up, send it.  */
   3109 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL) {
   3110 		kpreempt_disable();
   3111 		const uint32_t h = curcpu()->ci_index; // pktq_rps_hash(m)
   3112 		M_SETCTX(m, wgp);
   3113 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   3114 			WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   3115 			    if_name(&wg->wg_if));
   3116 			m_freem(m);
   3117 		}
   3118 		kpreempt_enable();
   3119 	}
   3120 
   3121 	if (wgs_prev->wgs_state == WGS_STATE_ESTABLISHED) {
   3122 		/* Wait for wg_get_stable_session to drain.  */
   3123 		pserialize_perform(wgp->wgp_psz);
   3124 
   3125 		/* Transition ESTABLISHED->DESTROYING.  */
   3126 		wgs_prev->wgs_state = WGS_STATE_DESTROYING;
   3127 
   3128 		/* We can't destroy the old session immediately */
   3129 		wg_schedule_session_dtor_timer(wgp);
   3130 	} else {
   3131 		KASSERTMSG(wgs_prev->wgs_state == WGS_STATE_UNKNOWN,
   3132 		    "state=%d", wgs_prev->wgs_state);
   3133 		wg_clear_states(wgs_prev);
   3134 		wgs_prev->wgs_state = WGS_STATE_UNKNOWN;
   3135 	}
   3136 }
   3137 
   3138 static void
   3139 wg_task_endpoint_changed(struct wg_softc *wg, struct wg_peer *wgp)
   3140 {
   3141 
   3142 	WG_TRACE("WGP_TASK_ENDPOINT_CHANGED");
   3143 
   3144 	KASSERT(mutex_owned(wgp->wgp_lock));
   3145 
   3146 	if (atomic_load_relaxed(&wgp->wgp_endpoint_changing)) {
   3147 		pserialize_perform(wgp->wgp_psz);
   3148 		mutex_exit(wgp->wgp_lock);
   3149 		psref_target_destroy(&wgp->wgp_endpoint0->wgsa_psref,
   3150 		    wg_psref_class);
   3151 		psref_target_init(&wgp->wgp_endpoint0->wgsa_psref,
   3152 		    wg_psref_class);
   3153 		mutex_enter(wgp->wgp_lock);
   3154 		atomic_store_release(&wgp->wgp_endpoint_changing, 0);
   3155 	}
   3156 }
   3157 
   3158 static void
   3159 wg_task_send_keepalive_message(struct wg_softc *wg, struct wg_peer *wgp)
   3160 {
   3161 	struct wg_session *wgs;
   3162 
   3163 	WG_TRACE("WGP_TASK_SEND_KEEPALIVE_MESSAGE");
   3164 
   3165 	KASSERT(mutex_owned(wgp->wgp_lock));
   3166 
   3167 	wgs = wgp->wgp_session_stable;
   3168 	if (wgs->wgs_state != WGS_STATE_ESTABLISHED)
   3169 		return;
   3170 
   3171 	wg_send_keepalive_msg(wgp, wgs);
   3172 }
   3173 
   3174 static void
   3175 wg_task_destroy_prev_session(struct wg_softc *wg, struct wg_peer *wgp)
   3176 {
   3177 	struct wg_session *wgs;
   3178 
   3179 	WG_TRACE("WGP_TASK_DESTROY_PREV_SESSION");
   3180 
   3181 	KASSERT(mutex_owned(wgp->wgp_lock));
   3182 
   3183 	wgs = wgp->wgp_session_unstable;
   3184 	if (wgs->wgs_state == WGS_STATE_DESTROYING) {
   3185 		wg_put_session_index(wg, wgs);
   3186 	}
   3187 }
   3188 
   3189 static void
   3190 wg_peer_work(struct work *wk, void *cookie)
   3191 {
   3192 	struct wg_peer *wgp = container_of(wk, struct wg_peer, wgp_work);
   3193 	struct wg_softc *wg = wgp->wgp_sc;
   3194 	unsigned int tasks;
   3195 
   3196 	mutex_enter(wgp->wgp_intr_lock);
   3197 	while ((tasks = wgp->wgp_tasks) != 0) {
   3198 		wgp->wgp_tasks = 0;
   3199 		mutex_exit(wgp->wgp_intr_lock);
   3200 
   3201 		mutex_enter(wgp->wgp_lock);
   3202 		if (ISSET(tasks, WGP_TASK_SEND_INIT_MESSAGE))
   3203 			wg_task_send_init_message(wg, wgp);
   3204 		if (ISSET(tasks, WGP_TASK_RETRY_HANDSHAKE))
   3205 			wg_task_retry_handshake(wg, wgp);
   3206 		if (ISSET(tasks, WGP_TASK_ESTABLISH_SESSION))
   3207 			wg_task_establish_session(wg, wgp);
   3208 		if (ISSET(tasks, WGP_TASK_ENDPOINT_CHANGED))
   3209 			wg_task_endpoint_changed(wg, wgp);
   3210 		if (ISSET(tasks, WGP_TASK_SEND_KEEPALIVE_MESSAGE))
   3211 			wg_task_send_keepalive_message(wg, wgp);
   3212 		if (ISSET(tasks, WGP_TASK_DESTROY_PREV_SESSION))
   3213 			wg_task_destroy_prev_session(wg, wgp);
   3214 		mutex_exit(wgp->wgp_lock);
   3215 
   3216 		mutex_enter(wgp->wgp_intr_lock);
   3217 	}
   3218 	mutex_exit(wgp->wgp_intr_lock);
   3219 }
   3220 
   3221 static void
   3222 wg_job(struct threadpool_job *job)
   3223 {
   3224 	struct wg_softc *wg = container_of(job, struct wg_softc, wg_job);
   3225 	int bound, upcalls;
   3226 
   3227 	mutex_enter(wg->wg_intr_lock);
   3228 	while ((upcalls = wg->wg_upcalls) != 0) {
   3229 		wg->wg_upcalls = 0;
   3230 		mutex_exit(wg->wg_intr_lock);
   3231 		bound = curlwp_bind();
   3232 		if (ISSET(upcalls, WG_UPCALL_INET))
   3233 			wg_receive_packets(wg, AF_INET);
   3234 		if (ISSET(upcalls, WG_UPCALL_INET6))
   3235 			wg_receive_packets(wg, AF_INET6);
   3236 		curlwp_bindx(bound);
   3237 		mutex_enter(wg->wg_intr_lock);
   3238 	}
   3239 	threadpool_job_done(job);
   3240 	mutex_exit(wg->wg_intr_lock);
   3241 }
   3242 
   3243 static int
   3244 wg_bind_port(struct wg_softc *wg, const uint16_t port)
   3245 {
   3246 	int error;
   3247 	uint16_t old_port = wg->wg_listen_port;
   3248 
   3249 	if (port != 0 && old_port == port)
   3250 		return 0;
   3251 
   3252 	struct sockaddr_in _sin, *sin = &_sin;
   3253 	sin->sin_len = sizeof(*sin);
   3254 	sin->sin_family = AF_INET;
   3255 	sin->sin_addr.s_addr = INADDR_ANY;
   3256 	sin->sin_port = htons(port);
   3257 
   3258 	error = sobind(wg->wg_so4, sintosa(sin), curlwp);
   3259 	if (error != 0)
   3260 		return error;
   3261 
   3262 #ifdef INET6
   3263 	struct sockaddr_in6 _sin6, *sin6 = &_sin6;
   3264 	sin6->sin6_len = sizeof(*sin6);
   3265 	sin6->sin6_family = AF_INET6;
   3266 	sin6->sin6_addr = in6addr_any;
   3267 	sin6->sin6_port = htons(port);
   3268 
   3269 	error = sobind(wg->wg_so6, sin6tosa(sin6), curlwp);
   3270 	if (error != 0)
   3271 		return error;
   3272 #endif
   3273 
   3274 	wg->wg_listen_port = port;
   3275 
   3276 	return 0;
   3277 }
   3278 
   3279 static void
   3280 wg_so_upcall(struct socket *so, void *cookie, int events, int waitflag)
   3281 {
   3282 	struct wg_softc *wg = cookie;
   3283 	int reason;
   3284 
   3285 	reason = (so->so_proto->pr_domain->dom_family == AF_INET) ?
   3286 	    WG_UPCALL_INET :
   3287 	    WG_UPCALL_INET6;
   3288 
   3289 	mutex_enter(wg->wg_intr_lock);
   3290 	wg->wg_upcalls |= reason;
   3291 	threadpool_schedule_job(wg->wg_threadpool, &wg->wg_job);
   3292 	mutex_exit(wg->wg_intr_lock);
   3293 }
   3294 
   3295 static int
   3296 wg_overudp_cb(struct mbuf **mp, int offset, struct socket *so,
   3297     struct sockaddr *src, void *arg)
   3298 {
   3299 	struct wg_softc *wg = arg;
   3300 	struct wg_msg wgm;
   3301 	struct mbuf *m = *mp;
   3302 
   3303 	WG_TRACE("enter");
   3304 
   3305 	/* Verify the mbuf chain is long enough to have a wg msg header.  */
   3306 	KASSERT(offset <= m_length(m));
   3307 	if (__predict_false(m_length(m) - offset < sizeof(struct wg_msg))) {
   3308 		/* drop on the floor */
   3309 		m_freem(m);
   3310 		return -1;
   3311 	}
   3312 
   3313 	/*
   3314 	 * Copy the message header (32-bit message type) out -- we'll
   3315 	 * worry about contiguity and alignment later.
   3316 	 */
   3317 	m_copydata(m, offset, sizeof(struct wg_msg), &wgm);
   3318 	WG_DLOG("type=%d\n", le32toh(wgm.wgm_type));
   3319 
   3320 	/*
   3321 	 * Handle DATA packets promptly as they arrive.  Other packets
   3322 	 * may require expensive public-key crypto and are not as
   3323 	 * sensitive to latency, so defer them to the worker thread.
   3324 	 */
   3325 	switch (le32toh(wgm.wgm_type)) {
   3326 	case WG_MSG_TYPE_DATA:
   3327 		/* handle immediately */
   3328 		m_adj(m, offset);
   3329 		if (__predict_false(m->m_len < sizeof(struct wg_msg_data))) {
   3330 			m = m_pullup(m, sizeof(struct wg_msg_data));
   3331 			if (m == NULL)
   3332 				return -1;
   3333 		}
   3334 		wg_handle_msg_data(wg, m, src);
   3335 		*mp = NULL;
   3336 		return 1;
   3337 	case WG_MSG_TYPE_INIT:
   3338 	case WG_MSG_TYPE_RESP:
   3339 	case WG_MSG_TYPE_COOKIE:
   3340 		/* pass through to so_receive in wg_receive_packets */
   3341 		return 0;
   3342 	default:
   3343 		/* drop on the floor */
   3344 		m_freem(m);
   3345 		return -1;
   3346 	}
   3347 }
   3348 
   3349 static int
   3350 wg_socreate(struct wg_softc *wg, int af, struct socket **sop)
   3351 {
   3352 	int error;
   3353 	struct socket *so;
   3354 
   3355 	error = socreate(af, &so, SOCK_DGRAM, 0, curlwp, NULL);
   3356 	if (error != 0)
   3357 		return error;
   3358 
   3359 	solock(so);
   3360 	so->so_upcallarg = wg;
   3361 	so->so_upcall = wg_so_upcall;
   3362 	so->so_rcv.sb_flags |= SB_UPCALL;
   3363 	inpcb_register_overudp_cb(sotoinpcb(so), wg_overudp_cb, wg);
   3364 	sounlock(so);
   3365 
   3366 	*sop = so;
   3367 
   3368 	return 0;
   3369 }
   3370 
   3371 static bool
   3372 wg_session_hit_limits(struct wg_session *wgs)
   3373 {
   3374 
   3375 	/*
   3376 	 * [W] 6.2: Transport Message Limits
   3377 	 * "After REJECT-AFTER-MESSAGES transport data messages or after the
   3378 	 *  current secure session is REJECT-AFTER-TIME seconds old, whichever
   3379 	 *  comes first, WireGuard will refuse to send any more transport data
   3380 	 *  messages using the current secure session, ..."
   3381 	 */
   3382 	KASSERT(wgs->wgs_time_established != 0);
   3383 	if ((time_uptime - wgs->wgs_time_established) > wg_reject_after_time) {
   3384 		WG_DLOG("The session hits REJECT_AFTER_TIME\n");
   3385 		return true;
   3386 	} else if (wg_session_get_send_counter(wgs) >
   3387 	    wg_reject_after_messages) {
   3388 		WG_DLOG("The session hits REJECT_AFTER_MESSAGES\n");
   3389 		return true;
   3390 	}
   3391 
   3392 	return false;
   3393 }
   3394 
   3395 static void
   3396 wgintr(void *cookie)
   3397 {
   3398 	struct wg_peer *wgp;
   3399 	struct wg_session *wgs;
   3400 	struct mbuf *m;
   3401 	struct psref psref;
   3402 
   3403 	while ((m = pktq_dequeue(wg_pktq)) != NULL) {
   3404 		wgp = M_GETCTX(m, struct wg_peer *);
   3405 		if ((wgs = wg_get_stable_session(wgp, &psref)) == NULL) {
   3406 			WG_TRACE("no stable session");
   3407 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3408 			goto next0;
   3409 		}
   3410 		if (__predict_false(wg_session_hit_limits(wgs))) {
   3411 			WG_TRACE("stable session hit limits");
   3412 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3413 			goto next1;
   3414 		}
   3415 		wg_send_data_msg(wgp, wgs, m);
   3416 		m = NULL;	/* consumed */
   3417 next1:		wg_put_session(wgs, &psref);
   3418 next0:		m_freem(m);
   3419 		/* XXX Yield to avoid userland starvation?  */
   3420 	}
   3421 }
   3422 
   3423 static void
   3424 wg_rekey_timer(void *arg)
   3425 {
   3426 	struct wg_peer *wgp = arg;
   3427 
   3428 	wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3429 }
   3430 
   3431 static void
   3432 wg_purge_pending_packets(struct wg_peer *wgp)
   3433 {
   3434 	struct mbuf *m;
   3435 
   3436 	m = atomic_swap_ptr(&wgp->wgp_pending, NULL);
   3437 	m_freem(m);
   3438 	pktq_barrier(wg_pktq);
   3439 }
   3440 
   3441 static void
   3442 wg_handshake_timeout_timer(void *arg)
   3443 {
   3444 	struct wg_peer *wgp = arg;
   3445 
   3446 	WG_TRACE("enter");
   3447 
   3448 	wg_schedule_peer_task(wgp, WGP_TASK_RETRY_HANDSHAKE);
   3449 }
   3450 
   3451 static struct wg_peer *
   3452 wg_alloc_peer(struct wg_softc *wg)
   3453 {
   3454 	struct wg_peer *wgp;
   3455 
   3456 	wgp = kmem_zalloc(sizeof(*wgp), KM_SLEEP);
   3457 
   3458 	wgp->wgp_sc = wg;
   3459 	callout_init(&wgp->wgp_rekey_timer, CALLOUT_MPSAFE);
   3460 	callout_setfunc(&wgp->wgp_rekey_timer, wg_rekey_timer, wgp);
   3461 	callout_init(&wgp->wgp_handshake_timeout_timer, CALLOUT_MPSAFE);
   3462 	callout_setfunc(&wgp->wgp_handshake_timeout_timer,
   3463 	    wg_handshake_timeout_timer, wgp);
   3464 	callout_init(&wgp->wgp_session_dtor_timer, CALLOUT_MPSAFE);
   3465 	callout_setfunc(&wgp->wgp_session_dtor_timer,
   3466 	    wg_session_dtor_timer, wgp);
   3467 	PSLIST_ENTRY_INIT(wgp, wgp_peerlist_entry);
   3468 	wgp->wgp_endpoint_changing = false;
   3469 	wgp->wgp_endpoint_available = false;
   3470 	wgp->wgp_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
   3471 	wgp->wgp_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
   3472 	wgp->wgp_psz = pserialize_create();
   3473 	psref_target_init(&wgp->wgp_psref, wg_psref_class);
   3474 
   3475 	wgp->wgp_endpoint = kmem_zalloc(sizeof(*wgp->wgp_endpoint), KM_SLEEP);
   3476 	wgp->wgp_endpoint0 = kmem_zalloc(sizeof(*wgp->wgp_endpoint0), KM_SLEEP);
   3477 	psref_target_init(&wgp->wgp_endpoint->wgsa_psref, wg_psref_class);
   3478 	psref_target_init(&wgp->wgp_endpoint0->wgsa_psref, wg_psref_class);
   3479 
   3480 	struct wg_session *wgs;
   3481 	wgp->wgp_session_stable =
   3482 	    kmem_zalloc(sizeof(*wgp->wgp_session_stable), KM_SLEEP);
   3483 	wgp->wgp_session_unstable =
   3484 	    kmem_zalloc(sizeof(*wgp->wgp_session_unstable), KM_SLEEP);
   3485 	wgs = wgp->wgp_session_stable;
   3486 	wgs->wgs_peer = wgp;
   3487 	wgs->wgs_state = WGS_STATE_UNKNOWN;
   3488 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
   3489 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3490 	mutex_init(&wgs->wgs_send_counter_lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3491 #endif
   3492 	wgs->wgs_recvwin = kmem_zalloc(sizeof(*wgs->wgs_recvwin), KM_SLEEP);
   3493 	mutex_init(&wgs->wgs_recvwin->lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3494 
   3495 	wgs = wgp->wgp_session_unstable;
   3496 	wgs->wgs_peer = wgp;
   3497 	wgs->wgs_state = WGS_STATE_UNKNOWN;
   3498 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
   3499 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3500 	mutex_init(&wgs->wgs_send_counter_lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3501 #endif
   3502 	wgs->wgs_recvwin = kmem_zalloc(sizeof(*wgs->wgs_recvwin), KM_SLEEP);
   3503 	mutex_init(&wgs->wgs_recvwin->lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3504 
   3505 	return wgp;
   3506 }
   3507 
   3508 static void
   3509 wg_destroy_peer(struct wg_peer *wgp)
   3510 {
   3511 	struct wg_session *wgs;
   3512 	struct wg_softc *wg = wgp->wgp_sc;
   3513 
   3514 	/* Prevent new packets from this peer on any source address.  */
   3515 	rw_enter(wg->wg_rwlock, RW_WRITER);
   3516 	for (int i = 0; i < wgp->wgp_n_allowedips; i++) {
   3517 		struct wg_allowedip *wga = &wgp->wgp_allowedips[i];
   3518 		struct radix_node_head *rnh = wg_rnh(wg, wga->wga_family);
   3519 		struct radix_node *rn;
   3520 
   3521 		KASSERT(rnh != NULL);
   3522 		rn = rnh->rnh_deladdr(&wga->wga_sa_addr,
   3523 		    &wga->wga_sa_mask, rnh);
   3524 		if (rn == NULL) {
   3525 			char addrstr[128];
   3526 			sockaddr_format(&wga->wga_sa_addr, addrstr,
   3527 			    sizeof(addrstr));
   3528 			WGLOG(LOG_WARNING, "%s: Couldn't delete %s",
   3529 			    if_name(&wg->wg_if), addrstr);
   3530 		}
   3531 	}
   3532 	rw_exit(wg->wg_rwlock);
   3533 
   3534 	/* Purge pending packets.  */
   3535 	wg_purge_pending_packets(wgp);
   3536 
   3537 	/* Halt all packet processing and timeouts.  */
   3538 	callout_halt(&wgp->wgp_rekey_timer, NULL);
   3539 	callout_halt(&wgp->wgp_handshake_timeout_timer, NULL);
   3540 	callout_halt(&wgp->wgp_session_dtor_timer, NULL);
   3541 
   3542 	/* Wait for any queued work to complete.  */
   3543 	workqueue_wait(wg_wq, &wgp->wgp_work);
   3544 
   3545 	wgs = wgp->wgp_session_unstable;
   3546 	if (wgs->wgs_state != WGS_STATE_UNKNOWN) {
   3547 		mutex_enter(wgp->wgp_lock);
   3548 		wg_destroy_session(wg, wgs);
   3549 		mutex_exit(wgp->wgp_lock);
   3550 	}
   3551 	mutex_destroy(&wgs->wgs_recvwin->lock);
   3552 	kmem_free(wgs->wgs_recvwin, sizeof(*wgs->wgs_recvwin));
   3553 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3554 	mutex_destroy(&wgs->wgs_send_counter_lock);
   3555 #endif
   3556 	kmem_free(wgs, sizeof(*wgs));
   3557 
   3558 	wgs = wgp->wgp_session_stable;
   3559 	if (wgs->wgs_state != WGS_STATE_UNKNOWN) {
   3560 		mutex_enter(wgp->wgp_lock);
   3561 		wg_destroy_session(wg, wgs);
   3562 		mutex_exit(wgp->wgp_lock);
   3563 	}
   3564 	mutex_destroy(&wgs->wgs_recvwin->lock);
   3565 	kmem_free(wgs->wgs_recvwin, sizeof(*wgs->wgs_recvwin));
   3566 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3567 	mutex_destroy(&wgs->wgs_send_counter_lock);
   3568 #endif
   3569 	kmem_free(wgs, sizeof(*wgs));
   3570 
   3571 	psref_target_destroy(&wgp->wgp_endpoint->wgsa_psref, wg_psref_class);
   3572 	psref_target_destroy(&wgp->wgp_endpoint0->wgsa_psref, wg_psref_class);
   3573 	kmem_free(wgp->wgp_endpoint, sizeof(*wgp->wgp_endpoint));
   3574 	kmem_free(wgp->wgp_endpoint0, sizeof(*wgp->wgp_endpoint0));
   3575 
   3576 	pserialize_destroy(wgp->wgp_psz);
   3577 	mutex_obj_free(wgp->wgp_intr_lock);
   3578 	mutex_obj_free(wgp->wgp_lock);
   3579 
   3580 	kmem_free(wgp, sizeof(*wgp));
   3581 }
   3582 
   3583 static void
   3584 wg_destroy_all_peers(struct wg_softc *wg)
   3585 {
   3586 	struct wg_peer *wgp, *wgp0 __diagused;
   3587 	void *garbage_byname, *garbage_bypubkey;
   3588 
   3589 restart:
   3590 	garbage_byname = garbage_bypubkey = NULL;
   3591 	mutex_enter(wg->wg_lock);
   3592 	WG_PEER_WRITER_FOREACH(wgp, wg) {
   3593 		if (wgp->wgp_name[0]) {
   3594 			wgp0 = thmap_del(wg->wg_peers_byname, wgp->wgp_name,
   3595 			    strlen(wgp->wgp_name));
   3596 			KASSERT(wgp0 == wgp);
   3597 			garbage_byname = thmap_stage_gc(wg->wg_peers_byname);
   3598 		}
   3599 		wgp0 = thmap_del(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   3600 		    sizeof(wgp->wgp_pubkey));
   3601 		KASSERT(wgp0 == wgp);
   3602 		garbage_bypubkey = thmap_stage_gc(wg->wg_peers_bypubkey);
   3603 		WG_PEER_WRITER_REMOVE(wgp);
   3604 		wg->wg_npeers--;
   3605 		mutex_enter(wgp->wgp_lock);
   3606 		pserialize_perform(wgp->wgp_psz);
   3607 		mutex_exit(wgp->wgp_lock);
   3608 		PSLIST_ENTRY_DESTROY(wgp, wgp_peerlist_entry);
   3609 		break;
   3610 	}
   3611 	mutex_exit(wg->wg_lock);
   3612 
   3613 	if (wgp == NULL)
   3614 		return;
   3615 
   3616 	psref_target_destroy(&wgp->wgp_psref, wg_psref_class);
   3617 
   3618 	wg_destroy_peer(wgp);
   3619 	thmap_gc(wg->wg_peers_byname, garbage_byname);
   3620 	thmap_gc(wg->wg_peers_bypubkey, garbage_bypubkey);
   3621 
   3622 	goto restart;
   3623 }
   3624 
   3625 static int
   3626 wg_destroy_peer_name(struct wg_softc *wg, const char *name)
   3627 {
   3628 	struct wg_peer *wgp, *wgp0 __diagused;
   3629 	void *garbage_byname, *garbage_bypubkey;
   3630 
   3631 	mutex_enter(wg->wg_lock);
   3632 	wgp = thmap_del(wg->wg_peers_byname, name, strlen(name));
   3633 	if (wgp != NULL) {
   3634 		wgp0 = thmap_del(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   3635 		    sizeof(wgp->wgp_pubkey));
   3636 		KASSERT(wgp0 == wgp);
   3637 		garbage_byname = thmap_stage_gc(wg->wg_peers_byname);
   3638 		garbage_bypubkey = thmap_stage_gc(wg->wg_peers_bypubkey);
   3639 		WG_PEER_WRITER_REMOVE(wgp);
   3640 		wg->wg_npeers--;
   3641 		if (wg->wg_npeers == 0)
   3642 			if_link_state_change(&wg->wg_if, LINK_STATE_DOWN);
   3643 		mutex_enter(wgp->wgp_lock);
   3644 		pserialize_perform(wgp->wgp_psz);
   3645 		mutex_exit(wgp->wgp_lock);
   3646 		PSLIST_ENTRY_DESTROY(wgp, wgp_peerlist_entry);
   3647 	}
   3648 	mutex_exit(wg->wg_lock);
   3649 
   3650 	if (wgp == NULL)
   3651 		return ENOENT;
   3652 
   3653 	psref_target_destroy(&wgp->wgp_psref, wg_psref_class);
   3654 
   3655 	wg_destroy_peer(wgp);
   3656 	thmap_gc(wg->wg_peers_byname, garbage_byname);
   3657 	thmap_gc(wg->wg_peers_bypubkey, garbage_bypubkey);
   3658 
   3659 	return 0;
   3660 }
   3661 
   3662 static int
   3663 wg_if_attach(struct wg_softc *wg)
   3664 {
   3665 
   3666 	wg->wg_if.if_addrlen = 0;
   3667 	wg->wg_if.if_mtu = WG_MTU;
   3668 	wg->wg_if.if_flags = IFF_MULTICAST;
   3669 	wg->wg_if.if_extflags = IFEF_MPSAFE;
   3670 	wg->wg_if.if_ioctl = wg_ioctl;
   3671 	wg->wg_if.if_output = wg_output;
   3672 	wg->wg_if.if_init = wg_init;
   3673 #ifdef ALTQ
   3674 	wg->wg_if.if_start = wg_start;
   3675 #endif
   3676 	wg->wg_if.if_stop = wg_stop;
   3677 	wg->wg_if.if_type = IFT_OTHER;
   3678 	wg->wg_if.if_dlt = DLT_NULL;
   3679 	wg->wg_if.if_softc = wg;
   3680 #ifdef ALTQ
   3681 	IFQ_SET_READY(&wg->wg_if.if_snd);
   3682 #endif
   3683 	if_initialize(&wg->wg_if);
   3684 
   3685 	wg->wg_if.if_link_state = LINK_STATE_DOWN;
   3686 	if_alloc_sadl(&wg->wg_if);
   3687 	if_register(&wg->wg_if);
   3688 
   3689 	bpf_attach(&wg->wg_if, DLT_NULL, sizeof(uint32_t));
   3690 
   3691 	return 0;
   3692 }
   3693 
   3694 static void
   3695 wg_if_detach(struct wg_softc *wg)
   3696 {
   3697 	struct ifnet *ifp = &wg->wg_if;
   3698 
   3699 	bpf_detach(ifp);
   3700 	if_detach(ifp);
   3701 }
   3702 
   3703 static int
   3704 wg_clone_create(struct if_clone *ifc, int unit)
   3705 {
   3706 	struct wg_softc *wg;
   3707 	int error;
   3708 
   3709 	wg_guarantee_initialized();
   3710 
   3711 	error = wg_count_inc();
   3712 	if (error)
   3713 		return error;
   3714 
   3715 	wg = kmem_zalloc(sizeof(*wg), KM_SLEEP);
   3716 
   3717 	if_initname(&wg->wg_if, ifc->ifc_name, unit);
   3718 
   3719 	PSLIST_INIT(&wg->wg_peers);
   3720 	wg->wg_peers_bypubkey = thmap_create(0, NULL, THMAP_NOCOPY);
   3721 	wg->wg_peers_byname = thmap_create(0, NULL, THMAP_NOCOPY);
   3722 	wg->wg_sessions_byindex = thmap_create(0, NULL, THMAP_NOCOPY);
   3723 	wg->wg_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
   3724 	wg->wg_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
   3725 	wg->wg_rwlock = rw_obj_alloc();
   3726 	threadpool_job_init(&wg->wg_job, wg_job, wg->wg_intr_lock,
   3727 	    "%s", if_name(&wg->wg_if));
   3728 	wg->wg_ops = &wg_ops_rumpkernel;
   3729 
   3730 	error = threadpool_get(&wg->wg_threadpool, PRI_NONE);
   3731 	if (error)
   3732 		goto fail0;
   3733 
   3734 #ifdef INET
   3735 	error = wg_socreate(wg, AF_INET, &wg->wg_so4);
   3736 	if (error)
   3737 		goto fail1;
   3738 	rn_inithead((void **)&wg->wg_rtable_ipv4,
   3739 	    offsetof(struct sockaddr_in, sin_addr) * NBBY);
   3740 #endif
   3741 #ifdef INET6
   3742 	error = wg_socreate(wg, AF_INET6, &wg->wg_so6);
   3743 	if (error)
   3744 		goto fail2;
   3745 	rn_inithead((void **)&wg->wg_rtable_ipv6,
   3746 	    offsetof(struct sockaddr_in6, sin6_addr) * NBBY);
   3747 #endif
   3748 
   3749 	error = wg_if_attach(wg);
   3750 	if (error)
   3751 		goto fail3;
   3752 
   3753 	return 0;
   3754 
   3755 fail4: __unused
   3756 	wg_if_detach(wg);
   3757 fail3:	wg_destroy_all_peers(wg);
   3758 #ifdef INET6
   3759 	solock(wg->wg_so6);
   3760 	wg->wg_so6->so_rcv.sb_flags &= ~SB_UPCALL;
   3761 	sounlock(wg->wg_so6);
   3762 #endif
   3763 #ifdef INET
   3764 	solock(wg->wg_so4);
   3765 	wg->wg_so4->so_rcv.sb_flags &= ~SB_UPCALL;
   3766 	sounlock(wg->wg_so4);
   3767 #endif
   3768 	mutex_enter(wg->wg_intr_lock);
   3769 	threadpool_cancel_job(wg->wg_threadpool, &wg->wg_job);
   3770 	mutex_exit(wg->wg_intr_lock);
   3771 #ifdef INET6
   3772 	if (wg->wg_rtable_ipv6 != NULL)
   3773 		free(wg->wg_rtable_ipv6, M_RTABLE);
   3774 	soclose(wg->wg_so6);
   3775 fail2:
   3776 #endif
   3777 #ifdef INET
   3778 	if (wg->wg_rtable_ipv4 != NULL)
   3779 		free(wg->wg_rtable_ipv4, M_RTABLE);
   3780 	soclose(wg->wg_so4);
   3781 fail1:
   3782 #endif
   3783 	threadpool_put(wg->wg_threadpool, PRI_NONE);
   3784 fail0:	threadpool_job_destroy(&wg->wg_job);
   3785 	rw_obj_free(wg->wg_rwlock);
   3786 	mutex_obj_free(wg->wg_intr_lock);
   3787 	mutex_obj_free(wg->wg_lock);
   3788 	thmap_destroy(wg->wg_sessions_byindex);
   3789 	thmap_destroy(wg->wg_peers_byname);
   3790 	thmap_destroy(wg->wg_peers_bypubkey);
   3791 	PSLIST_DESTROY(&wg->wg_peers);
   3792 	kmem_free(wg, sizeof(*wg));
   3793 	wg_count_dec();
   3794 	return error;
   3795 }
   3796 
   3797 static int
   3798 wg_clone_destroy(struct ifnet *ifp)
   3799 {
   3800 	struct wg_softc *wg = container_of(ifp, struct wg_softc, wg_if);
   3801 
   3802 #ifdef WG_RUMPKERNEL
   3803 	if (wg_user_mode(wg)) {
   3804 		rumpuser_wg_destroy(wg->wg_user);
   3805 		wg->wg_user = NULL;
   3806 	}
   3807 #endif
   3808 
   3809 	wg_if_detach(wg);
   3810 	wg_destroy_all_peers(wg);
   3811 #ifdef INET6
   3812 	solock(wg->wg_so6);
   3813 	wg->wg_so6->so_rcv.sb_flags &= ~SB_UPCALL;
   3814 	sounlock(wg->wg_so6);
   3815 #endif
   3816 #ifdef INET
   3817 	solock(wg->wg_so4);
   3818 	wg->wg_so4->so_rcv.sb_flags &= ~SB_UPCALL;
   3819 	sounlock(wg->wg_so4);
   3820 #endif
   3821 	mutex_enter(wg->wg_intr_lock);
   3822 	threadpool_cancel_job(wg->wg_threadpool, &wg->wg_job);
   3823 	mutex_exit(wg->wg_intr_lock);
   3824 #ifdef INET6
   3825 	if (wg->wg_rtable_ipv6 != NULL)
   3826 		free(wg->wg_rtable_ipv6, M_RTABLE);
   3827 	soclose(wg->wg_so6);
   3828 #endif
   3829 #ifdef INET
   3830 	if (wg->wg_rtable_ipv4 != NULL)
   3831 		free(wg->wg_rtable_ipv4, M_RTABLE);
   3832 	soclose(wg->wg_so4);
   3833 #endif
   3834 	threadpool_put(wg->wg_threadpool, PRI_NONE);
   3835 	threadpool_job_destroy(&wg->wg_job);
   3836 	rw_obj_free(wg->wg_rwlock);
   3837 	mutex_obj_free(wg->wg_intr_lock);
   3838 	mutex_obj_free(wg->wg_lock);
   3839 	thmap_destroy(wg->wg_sessions_byindex);
   3840 	thmap_destroy(wg->wg_peers_byname);
   3841 	thmap_destroy(wg->wg_peers_bypubkey);
   3842 	PSLIST_DESTROY(&wg->wg_peers);
   3843 	kmem_free(wg, sizeof(*wg));
   3844 	wg_count_dec();
   3845 
   3846 	return 0;
   3847 }
   3848 
   3849 static struct wg_peer *
   3850 wg_pick_peer_by_sa(struct wg_softc *wg, const struct sockaddr *sa,
   3851     struct psref *psref)
   3852 {
   3853 	struct radix_node_head *rnh;
   3854 	struct radix_node *rn;
   3855 	struct wg_peer *wgp = NULL;
   3856 	struct wg_allowedip *wga;
   3857 
   3858 #ifdef WG_DEBUG_LOG
   3859 	char addrstr[128];
   3860 	sockaddr_format(sa, addrstr, sizeof(addrstr));
   3861 	WG_DLOG("sa=%s\n", addrstr);
   3862 #endif
   3863 
   3864 	rw_enter(wg->wg_rwlock, RW_READER);
   3865 
   3866 	rnh = wg_rnh(wg, sa->sa_family);
   3867 	if (rnh == NULL)
   3868 		goto out;
   3869 
   3870 	rn = rnh->rnh_matchaddr(sa, rnh);
   3871 	if (rn == NULL || (rn->rn_flags & RNF_ROOT) != 0)
   3872 		goto out;
   3873 
   3874 	WG_TRACE("success");
   3875 
   3876 	wga = container_of(rn, struct wg_allowedip, wga_nodes[0]);
   3877 	wgp = wga->wga_peer;
   3878 	wg_get_peer(wgp, psref);
   3879 
   3880 out:
   3881 	rw_exit(wg->wg_rwlock);
   3882 	return wgp;
   3883 }
   3884 
   3885 static void
   3886 wg_fill_msg_data(struct wg_softc *wg, struct wg_peer *wgp,
   3887     struct wg_session *wgs, struct wg_msg_data *wgmd)
   3888 {
   3889 
   3890 	memset(wgmd, 0, sizeof(*wgmd));
   3891 	wgmd->wgmd_type = htole32(WG_MSG_TYPE_DATA);
   3892 	wgmd->wgmd_receiver = wgs->wgs_remote_index;
   3893 	/* [W] 5.4.6: msg.counter := Nm^send */
   3894 	/* [W] 5.4.6: Nm^send := Nm^send + 1 */
   3895 	wgmd->wgmd_counter = htole64(wg_session_inc_send_counter(wgs));
   3896 	WG_DLOG("counter=%"PRIu64"\n", le64toh(wgmd->wgmd_counter));
   3897 }
   3898 
   3899 static int
   3900 wg_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
   3901     const struct rtentry *rt)
   3902 {
   3903 	struct wg_softc *wg = ifp->if_softc;
   3904 	struct wg_peer *wgp = NULL;
   3905 	struct wg_session *wgs = NULL;
   3906 	struct psref wgp_psref, wgs_psref;
   3907 	int bound;
   3908 	int error;
   3909 
   3910 	bound = curlwp_bind();
   3911 
   3912 	/* TODO make the nest limit configurable via sysctl */
   3913 	error = if_tunnel_check_nesting(ifp, m, 1);
   3914 	if (error) {
   3915 		WGLOG(LOG_ERR,
   3916 		    "%s: tunneling loop detected and packet dropped\n",
   3917 		    if_name(&wg->wg_if));
   3918 		goto out0;
   3919 	}
   3920 
   3921 #ifdef ALTQ
   3922 	bool altq = atomic_load_relaxed(&ifp->if_snd.altq_flags)
   3923 	    & ALTQF_ENABLED;
   3924 	if (altq)
   3925 		IFQ_CLASSIFY(&ifp->if_snd, m, dst->sa_family);
   3926 #endif
   3927 
   3928 	bpf_mtap_af(ifp, dst->sa_family, m, BPF_D_OUT);
   3929 
   3930 	m->m_flags &= ~(M_BCAST|M_MCAST);
   3931 
   3932 	wgp = wg_pick_peer_by_sa(wg, dst, &wgp_psref);
   3933 	if (wgp == NULL) {
   3934 		WG_TRACE("peer not found");
   3935 		error = EHOSTUNREACH;
   3936 		goto out0;
   3937 	}
   3938 
   3939 	/* Clear checksum-offload flags. */
   3940 	m->m_pkthdr.csum_flags = 0;
   3941 	m->m_pkthdr.csum_data = 0;
   3942 
   3943 	/* Check whether there's an established session.  */
   3944 	wgs = wg_get_stable_session(wgp, &wgs_psref);
   3945 	if (wgs == NULL) {
   3946 		/*
   3947 		 * No established session.  If we're the first to try
   3948 		 * sending data, schedule a handshake and queue the
   3949 		 * packet for when the handshake is done; otherwise
   3950 		 * just drop the packet and let the ongoing handshake
   3951 		 * attempt continue.  We could queue more data packets
   3952 		 * but it's not clear that's worthwhile.
   3953 		 */
   3954 		if (atomic_cas_ptr(&wgp->wgp_pending, NULL, m) == NULL) {
   3955 			m = NULL; /* consume */
   3956 			WG_TRACE("queued first packet; init handshake");
   3957 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3958 		} else {
   3959 			WG_TRACE("first packet already queued, dropping");
   3960 		}
   3961 		goto out1;
   3962 	}
   3963 
   3964 	/* There's an established session.  Toss it in the queue.  */
   3965 #ifdef ALTQ
   3966 	if (altq) {
   3967 		mutex_enter(ifp->if_snd.ifq_lock);
   3968 		if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
   3969 			M_SETCTX(m, wgp);
   3970 			ALTQ_ENQUEUE(&ifp->if_snd, m, error);
   3971 			m = NULL; /* consume */
   3972 		}
   3973 		mutex_exit(ifp->if_snd.ifq_lock);
   3974 		if (m == NULL) {
   3975 			wg_start(ifp);
   3976 			goto out2;
   3977 		}
   3978 	}
   3979 #endif
   3980 	kpreempt_disable();
   3981 	const uint32_t h = curcpu()->ci_index;	// pktq_rps_hash(m)
   3982 	M_SETCTX(m, wgp);
   3983 	if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   3984 		WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   3985 		    if_name(&wg->wg_if));
   3986 		error = ENOBUFS;
   3987 		goto out3;
   3988 	}
   3989 	m = NULL;		/* consumed */
   3990 	error = 0;
   3991 out3:	kpreempt_enable();
   3992 
   3993 #ifdef ALTQ
   3994 out2:
   3995 #endif
   3996 	wg_put_session(wgs, &wgs_psref);
   3997 out1:	wg_put_peer(wgp, &wgp_psref);
   3998 out0:	m_freem(m);
   3999 	curlwp_bindx(bound);
   4000 	return error;
   4001 }
   4002 
   4003 static int
   4004 wg_send_udp(struct wg_peer *wgp, struct mbuf *m)
   4005 {
   4006 	struct psref psref;
   4007 	struct wg_sockaddr *wgsa;
   4008 	int error;
   4009 	struct socket *so;
   4010 
   4011 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   4012 	so = wg_get_so_by_peer(wgp, wgsa);
   4013 	solock(so);
   4014 	if (wgsatosa(wgsa)->sa_family == AF_INET) {
   4015 		error = udp_send(so, m, wgsatosa(wgsa), NULL, curlwp);
   4016 	} else {
   4017 #ifdef INET6
   4018 		error = udp6_output(sotoinpcb(so), m, wgsatosin6(wgsa),
   4019 		    NULL, curlwp);
   4020 #else
   4021 		m_freem(m);
   4022 		error = EPFNOSUPPORT;
   4023 #endif
   4024 	}
   4025 	sounlock(so);
   4026 	wg_put_sa(wgp, wgsa, &psref);
   4027 
   4028 	return error;
   4029 }
   4030 
   4031 /* Inspired by pppoe_get_mbuf */
   4032 static struct mbuf *
   4033 wg_get_mbuf(size_t leading_len, size_t len)
   4034 {
   4035 	struct mbuf *m;
   4036 
   4037 	KASSERT(leading_len <= MCLBYTES);
   4038 	KASSERT(len <= MCLBYTES - leading_len);
   4039 
   4040 	m = m_gethdr(M_DONTWAIT, MT_DATA);
   4041 	if (m == NULL)
   4042 		return NULL;
   4043 	if (len + leading_len > MHLEN) {
   4044 		m_clget(m, M_DONTWAIT);
   4045 		if ((m->m_flags & M_EXT) == 0) {
   4046 			m_free(m);
   4047 			return NULL;
   4048 		}
   4049 	}
   4050 	m->m_data += leading_len;
   4051 	m->m_pkthdr.len = m->m_len = len;
   4052 
   4053 	return m;
   4054 }
   4055 
   4056 static int
   4057 wg_send_data_msg(struct wg_peer *wgp, struct wg_session *wgs,
   4058     struct mbuf *m)
   4059 {
   4060 	struct wg_softc *wg = wgp->wgp_sc;
   4061 	int error;
   4062 	size_t inner_len, padded_len, encrypted_len;
   4063 	char *padded_buf = NULL;
   4064 	size_t mlen;
   4065 	struct wg_msg_data *wgmd;
   4066 	bool free_padded_buf = false;
   4067 	struct mbuf *n;
   4068 	size_t leading_len = max_hdr + sizeof(struct udphdr);
   4069 
   4070 	mlen = m_length(m);
   4071 	inner_len = mlen;
   4072 	padded_len = roundup(mlen, 16);
   4073 	encrypted_len = padded_len + WG_AUTHTAG_LEN;
   4074 	WG_DLOG("inner=%zu, padded=%zu, encrypted_len=%zu\n",
   4075 	    inner_len, padded_len, encrypted_len);
   4076 	if (mlen != 0) {
   4077 		bool success;
   4078 		success = m_ensure_contig(&m, padded_len);
   4079 		if (success) {
   4080 			padded_buf = mtod(m, char *);
   4081 		} else {
   4082 			padded_buf = kmem_intr_alloc(padded_len, KM_NOSLEEP);
   4083 			if (padded_buf == NULL) {
   4084 				error = ENOBUFS;
   4085 				goto end;
   4086 			}
   4087 			free_padded_buf = true;
   4088 			m_copydata(m, 0, mlen, padded_buf);
   4089 		}
   4090 		memset(padded_buf + mlen, 0, padded_len - inner_len);
   4091 	}
   4092 
   4093 	n = wg_get_mbuf(leading_len, sizeof(*wgmd) + encrypted_len);
   4094 	if (n == NULL) {
   4095 		error = ENOBUFS;
   4096 		goto end;
   4097 	}
   4098 	KASSERT(n->m_len >= sizeof(*wgmd));
   4099 	wgmd = mtod(n, struct wg_msg_data *);
   4100 	wg_fill_msg_data(wg, wgp, wgs, wgmd);
   4101 #ifdef WG_DEBUG_PACKET
   4102 	if (wg_debug & WG_DEBUG_FLAGS_PACKET) {
   4103 		hexdump(printf, "padded_buf", padded_buf,
   4104 		    padded_len);
   4105 	}
   4106 #endif
   4107 	/* [W] 5.4.6: AEAD(Tm^send, Nm^send, P, e) */
   4108 	wg_algo_aead_enc((char *)wgmd + sizeof(*wgmd), encrypted_len,
   4109 	    wgs->wgs_tkey_send, le64toh(wgmd->wgmd_counter),
   4110 	    padded_buf, padded_len,
   4111 	    NULL, 0);
   4112 #ifdef WG_DEBUG_PACKET
   4113 	if (wg_debug & WG_DEBUG_FLAGS_PACKET) {
   4114 		hexdump(printf, "tkey_send", wgs->wgs_tkey_send,
   4115 		    sizeof(wgs->wgs_tkey_send));
   4116 		hexdump(printf, "wgmd", wgmd, sizeof(*wgmd));
   4117 		hexdump(printf, "outgoing packet",
   4118 		    (char *)wgmd + sizeof(*wgmd), encrypted_len);
   4119 		size_t decrypted_len = encrypted_len - WG_AUTHTAG_LEN;
   4120 		char *decrypted_buf = kmem_intr_alloc((decrypted_len +
   4121 			WG_AUTHTAG_LEN/*XXX*/), KM_NOSLEEP);
   4122 		if (decrypted_buf != NULL) {
   4123 			error = wg_algo_aead_dec(
   4124 			    1 + decrypted_buf /* force misalignment */,
   4125 			    encrypted_len - WG_AUTHTAG_LEN /* XXX */,
   4126 			    wgs->wgs_tkey_send, le64toh(wgmd->wgmd_counter),
   4127 			    (char *)wgmd + sizeof(*wgmd), encrypted_len,
   4128 			    NULL, 0);
   4129 			if (error) {
   4130 				WG_DLOG("wg_algo_aead_dec failed: %d\n",
   4131 				    error);
   4132 			}
   4133 			if (!consttime_memequal(1 + decrypted_buf,
   4134 				(char *)wgmd + sizeof(*wgmd),
   4135 				decrypted_len)) {
   4136 				WG_DLOG("wg_algo_aead_dec returned garbage\n");
   4137 			}
   4138 			kmem_intr_free(decrypted_buf, (decrypted_len +
   4139 				WG_AUTHTAG_LEN/*XXX*/));
   4140 		}
   4141 	}
   4142 #endif
   4143 
   4144 	error = wg->wg_ops->send_data_msg(wgp, n);
   4145 	if (error == 0) {
   4146 		struct ifnet *ifp = &wg->wg_if;
   4147 		if_statadd(ifp, if_obytes, mlen);
   4148 		if_statinc(ifp, if_opackets);
   4149 		if (wgs->wgs_is_initiator &&
   4150 		    wgs->wgs_time_last_data_sent == 0) {
   4151 			/*
   4152 			 * [W] 6.2 Transport Message Limits
   4153 			 * "if a peer is the initiator of a current secure
   4154 			 *  session, WireGuard will send a handshake initiation
   4155 			 *  message to begin a new secure session if, after
   4156 			 *  transmitting a transport data message, the current
   4157 			 *  secure session is REKEY-AFTER-TIME seconds old,"
   4158 			 */
   4159 			wg_schedule_rekey_timer(wgp);
   4160 		}
   4161 		wgs->wgs_time_last_data_sent = time_uptime;
   4162 		if (wg_session_get_send_counter(wgs) >=
   4163 		    wg_rekey_after_messages) {
   4164 			/*
   4165 			 * [W] 6.2 Transport Message Limits
   4166 			 * "WireGuard will try to create a new session, by
   4167 			 *  sending a handshake initiation message (section
   4168 			 *  5.4.2), after it has sent REKEY-AFTER-MESSAGES
   4169 			 *  transport data messages..."
   4170 			 */
   4171 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   4172 		}
   4173 	}
   4174 end:
   4175 	m_freem(m);
   4176 	if (free_padded_buf)
   4177 		kmem_intr_free(padded_buf, padded_len);
   4178 	return error;
   4179 }
   4180 
   4181 static void
   4182 wg_input(struct ifnet *ifp, struct mbuf *m, const int af)
   4183 {
   4184 	pktqueue_t *pktq;
   4185 	size_t pktlen;
   4186 
   4187 	KASSERT(af == AF_INET || af == AF_INET6);
   4188 
   4189 	WG_TRACE("");
   4190 
   4191 	m_set_rcvif(m, ifp);
   4192 	pktlen = m->m_pkthdr.len;
   4193 
   4194 	bpf_mtap_af(ifp, af, m, BPF_D_IN);
   4195 
   4196 	switch (af) {
   4197 	case AF_INET:
   4198 		pktq = ip_pktq;
   4199 		break;
   4200 #ifdef INET6
   4201 	case AF_INET6:
   4202 		pktq = ip6_pktq;
   4203 		break;
   4204 #endif
   4205 	default:
   4206 		panic("invalid af=%d", af);
   4207 	}
   4208 
   4209 	kpreempt_disable();
   4210 	const u_int h = curcpu()->ci_index;
   4211 	if (__predict_true(pktq_enqueue(pktq, m, h))) {
   4212 		if_statadd(ifp, if_ibytes, pktlen);
   4213 		if_statinc(ifp, if_ipackets);
   4214 	} else {
   4215 		m_freem(m);
   4216 	}
   4217 	kpreempt_enable();
   4218 }
   4219 
   4220 static void
   4221 wg_calc_pubkey(uint8_t pubkey[WG_STATIC_KEY_LEN],
   4222     const uint8_t privkey[WG_STATIC_KEY_LEN])
   4223 {
   4224 
   4225 	crypto_scalarmult_base(pubkey, privkey);
   4226 }
   4227 
   4228 static int
   4229 wg_rtable_add_route(struct wg_softc *wg, struct wg_allowedip *wga)
   4230 {
   4231 	struct radix_node_head *rnh;
   4232 	struct radix_node *rn;
   4233 	int error = 0;
   4234 
   4235 	rw_enter(wg->wg_rwlock, RW_WRITER);
   4236 	rnh = wg_rnh(wg, wga->wga_family);
   4237 	KASSERT(rnh != NULL);
   4238 	rn = rnh->rnh_addaddr(&wga->wga_sa_addr, &wga->wga_sa_mask, rnh,
   4239 	    wga->wga_nodes);
   4240 	rw_exit(wg->wg_rwlock);
   4241 
   4242 	if (rn == NULL)
   4243 		error = EEXIST;
   4244 
   4245 	return error;
   4246 }
   4247 
   4248 static int
   4249 wg_handle_prop_peer(struct wg_softc *wg, prop_dictionary_t peer,
   4250     struct wg_peer **wgpp)
   4251 {
   4252 	int error = 0;
   4253 	const void *pubkey;
   4254 	size_t pubkey_len;
   4255 	const void *psk;
   4256 	size_t psk_len;
   4257 	const char *name = NULL;
   4258 
   4259 	if (prop_dictionary_get_string(peer, "name", &name)) {
   4260 		if (strlen(name) > WG_PEER_NAME_MAXLEN) {
   4261 			error = EINVAL;
   4262 			goto out;
   4263 		}
   4264 	}
   4265 
   4266 	if (!prop_dictionary_get_data(peer, "public_key",
   4267 		&pubkey, &pubkey_len)) {
   4268 		error = EINVAL;
   4269 		goto out;
   4270 	}
   4271 #ifdef WG_DEBUG_DUMP
   4272         if (wg_debug & WG_DEBUG_FLAGS_DUMP) {
   4273 		char *hex = gethexdump(pubkey, pubkey_len);
   4274 		log(LOG_DEBUG, "pubkey=%p, pubkey_len=%zu\n%s\n",
   4275 		    pubkey, pubkey_len, hex);
   4276 		puthexdump(hex, pubkey, pubkey_len);
   4277 	}
   4278 #endif
   4279 
   4280 	struct wg_peer *wgp = wg_alloc_peer(wg);
   4281 	memcpy(wgp->wgp_pubkey, pubkey, sizeof(wgp->wgp_pubkey));
   4282 	if (name != NULL)
   4283 		strncpy(wgp->wgp_name, name, sizeof(wgp->wgp_name));
   4284 
   4285 	if (prop_dictionary_get_data(peer, "preshared_key", &psk, &psk_len)) {
   4286 		if (psk_len != sizeof(wgp->wgp_psk)) {
   4287 			error = EINVAL;
   4288 			goto out;
   4289 		}
   4290 		memcpy(wgp->wgp_psk, psk, sizeof(wgp->wgp_psk));
   4291 	}
   4292 
   4293 	const void *addr;
   4294 	size_t addr_len;
   4295 	struct wg_sockaddr *wgsa = wgp->wgp_endpoint;
   4296 
   4297 	if (!prop_dictionary_get_data(peer, "endpoint", &addr, &addr_len))
   4298 		goto skip_endpoint;
   4299 	if (addr_len < sizeof(*wgsatosa(wgsa)) ||
   4300 	    addr_len > sizeof(*wgsatoss(wgsa))) {
   4301 		error = EINVAL;
   4302 		goto out;
   4303 	}
   4304 	memcpy(wgsatoss(wgsa), addr, addr_len);
   4305 	switch (wgsa_family(wgsa)) {
   4306 	case AF_INET:
   4307 #ifdef INET6
   4308 	case AF_INET6:
   4309 #endif
   4310 		break;
   4311 	default:
   4312 		error = EPFNOSUPPORT;
   4313 		goto out;
   4314 	}
   4315 	if (addr_len != sockaddr_getsize_by_family(wgsa_family(wgsa))) {
   4316 		error = EINVAL;
   4317 		goto out;
   4318 	}
   4319     {
   4320 	char addrstr[128];
   4321 	sockaddr_format(wgsatosa(wgsa), addrstr, sizeof(addrstr));
   4322 	WG_DLOG("addr=%s\n", addrstr);
   4323     }
   4324 	wgp->wgp_endpoint_available = true;
   4325 
   4326 	prop_array_t allowedips;
   4327 skip_endpoint:
   4328 	allowedips = prop_dictionary_get(peer, "allowedips");
   4329 	if (allowedips == NULL)
   4330 		goto skip;
   4331 
   4332 	prop_object_iterator_t _it = prop_array_iterator(allowedips);
   4333 	prop_dictionary_t prop_allowedip;
   4334 	int j = 0;
   4335 	while ((prop_allowedip = prop_object_iterator_next(_it)) != NULL) {
   4336 		struct wg_allowedip *wga = &wgp->wgp_allowedips[j];
   4337 
   4338 		if (!prop_dictionary_get_int(prop_allowedip, "family",
   4339 			&wga->wga_family))
   4340 			continue;
   4341 		if (!prop_dictionary_get_data(prop_allowedip, "ip",
   4342 			&addr, &addr_len))
   4343 			continue;
   4344 		if (!prop_dictionary_get_uint8(prop_allowedip, "cidr",
   4345 			&wga->wga_cidr))
   4346 			continue;
   4347 
   4348 		switch (wga->wga_family) {
   4349 		case AF_INET: {
   4350 			struct sockaddr_in sin;
   4351 			char addrstr[128];
   4352 			struct in_addr mask;
   4353 			struct sockaddr_in sin_mask;
   4354 
   4355 			if (addr_len != sizeof(struct in_addr))
   4356 				return EINVAL;
   4357 			memcpy(&wga->wga_addr4, addr, addr_len);
   4358 
   4359 			sockaddr_in_init(&sin, (const struct in_addr *)addr,
   4360 			    0);
   4361 			sockaddr_copy(&wga->wga_sa_addr,
   4362 			    sizeof(sin), sintosa(&sin));
   4363 
   4364 			sockaddr_format(sintosa(&sin),
   4365 			    addrstr, sizeof(addrstr));
   4366 			WG_DLOG("addr=%s/%d\n", addrstr, wga->wga_cidr);
   4367 
   4368 			in_len2mask(&mask, wga->wga_cidr);
   4369 			sockaddr_in_init(&sin_mask, &mask, 0);
   4370 			sockaddr_copy(&wga->wga_sa_mask,
   4371 			    sizeof(sin_mask), sintosa(&sin_mask));
   4372 
   4373 			break;
   4374 		    }
   4375 #ifdef INET6
   4376 		case AF_INET6: {
   4377 			struct sockaddr_in6 sin6;
   4378 			char addrstr[128];
   4379 			struct in6_addr mask;
   4380 			struct sockaddr_in6 sin6_mask;
   4381 
   4382 			if (addr_len != sizeof(struct in6_addr))
   4383 				return EINVAL;
   4384 			memcpy(&wga->wga_addr6, addr, addr_len);
   4385 
   4386 			sockaddr_in6_init(&sin6, (const struct in6_addr *)addr,
   4387 			    0, 0, 0);
   4388 			sockaddr_copy(&wga->wga_sa_addr,
   4389 			    sizeof(sin6), sin6tosa(&sin6));
   4390 
   4391 			sockaddr_format(sin6tosa(&sin6),
   4392 			    addrstr, sizeof(addrstr));
   4393 			WG_DLOG("addr=%s/%d\n", addrstr, wga->wga_cidr);
   4394 
   4395 			in6_prefixlen2mask(&mask, wga->wga_cidr);
   4396 			sockaddr_in6_init(&sin6_mask, &mask, 0, 0, 0);
   4397 			sockaddr_copy(&wga->wga_sa_mask,
   4398 			    sizeof(sin6_mask), sin6tosa(&sin6_mask));
   4399 
   4400 			break;
   4401 		    }
   4402 #endif
   4403 		default:
   4404 			error = EINVAL;
   4405 			goto out;
   4406 		}
   4407 		wga->wga_peer = wgp;
   4408 
   4409 		error = wg_rtable_add_route(wg, wga);
   4410 		if (error != 0)
   4411 			goto out;
   4412 
   4413 		j++;
   4414 	}
   4415 	wgp->wgp_n_allowedips = j;
   4416 skip:
   4417 	*wgpp = wgp;
   4418 out:
   4419 	return error;
   4420 }
   4421 
   4422 static int
   4423 wg_alloc_prop_buf(char **_buf, struct ifdrv *ifd)
   4424 {
   4425 	int error;
   4426 	char *buf;
   4427 
   4428 	WG_DLOG("buf=%p, len=%zu\n", ifd->ifd_data, ifd->ifd_len);
   4429 	if (ifd->ifd_len >= WG_MAX_PROPLEN)
   4430 		return E2BIG;
   4431 	buf = kmem_alloc(ifd->ifd_len + 1, KM_SLEEP);
   4432 	error = copyin(ifd->ifd_data, buf, ifd->ifd_len);
   4433 	if (error != 0)
   4434 		return error;
   4435 	buf[ifd->ifd_len] = '\0';
   4436 #ifdef WG_DEBUG_DUMP
   4437 	if (wg_debug & WG_DEBUG_FLAGS_DUMP) {
   4438 		log(LOG_DEBUG, "%.*s\n", (int)MIN(INT_MAX, ifd->ifd_len),
   4439 		    (const char *)buf);
   4440 	}
   4441 #endif
   4442 	*_buf = buf;
   4443 	return 0;
   4444 }
   4445 
   4446 static int
   4447 wg_ioctl_set_private_key(struct wg_softc *wg, struct ifdrv *ifd)
   4448 {
   4449 	int error;
   4450 	prop_dictionary_t prop_dict;
   4451 	char *buf = NULL;
   4452 	const void *privkey;
   4453 	size_t privkey_len;
   4454 
   4455 	error = wg_alloc_prop_buf(&buf, ifd);
   4456 	if (error != 0)
   4457 		return error;
   4458 	error = EINVAL;
   4459 	prop_dict = prop_dictionary_internalize(buf);
   4460 	if (prop_dict == NULL)
   4461 		goto out;
   4462 	if (!prop_dictionary_get_data(prop_dict, "private_key",
   4463 		&privkey, &privkey_len))
   4464 		goto out;
   4465 #ifdef WG_DEBUG_DUMP
   4466 	if (wg_debug & WG_DEBUG_FLAGS_DUMP) {
   4467 		char *hex = gethexdump(privkey, privkey_len);
   4468 		log(LOG_DEBUG, "privkey=%p, privkey_len=%zu\n%s\n",
   4469 		    privkey, privkey_len, hex);
   4470 		puthexdump(hex, privkey, privkey_len);
   4471 	}
   4472 #endif
   4473 	if (privkey_len != WG_STATIC_KEY_LEN)
   4474 		goto out;
   4475 	memcpy(wg->wg_privkey, privkey, WG_STATIC_KEY_LEN);
   4476 	wg_calc_pubkey(wg->wg_pubkey, wg->wg_privkey);
   4477 	error = 0;
   4478 
   4479 out:
   4480 	kmem_free(buf, ifd->ifd_len + 1);
   4481 	return error;
   4482 }
   4483 
   4484 static int
   4485 wg_ioctl_set_listen_port(struct wg_softc *wg, struct ifdrv *ifd)
   4486 {
   4487 	int error;
   4488 	prop_dictionary_t prop_dict;
   4489 	char *buf = NULL;
   4490 	uint16_t port;
   4491 
   4492 	error = wg_alloc_prop_buf(&buf, ifd);
   4493 	if (error != 0)
   4494 		return error;
   4495 	error = EINVAL;
   4496 	prop_dict = prop_dictionary_internalize(buf);
   4497 	if (prop_dict == NULL)
   4498 		goto out;
   4499 	if (!prop_dictionary_get_uint16(prop_dict, "listen_port", &port))
   4500 		goto out;
   4501 
   4502 	error = wg->wg_ops->bind_port(wg, (uint16_t)port);
   4503 
   4504 out:
   4505 	kmem_free(buf, ifd->ifd_len + 1);
   4506 	return error;
   4507 }
   4508 
   4509 static int
   4510 wg_ioctl_add_peer(struct wg_softc *wg, struct ifdrv *ifd)
   4511 {
   4512 	int error;
   4513 	prop_dictionary_t prop_dict;
   4514 	char *buf = NULL;
   4515 	struct wg_peer *wgp = NULL, *wgp0 __diagused;
   4516 
   4517 	error = wg_alloc_prop_buf(&buf, ifd);
   4518 	if (error != 0)
   4519 		return error;
   4520 	error = EINVAL;
   4521 	prop_dict = prop_dictionary_internalize(buf);
   4522 	if (prop_dict == NULL)
   4523 		goto out;
   4524 
   4525 	error = wg_handle_prop_peer(wg, prop_dict, &wgp);
   4526 	if (error != 0)
   4527 		goto out;
   4528 
   4529 	mutex_enter(wg->wg_lock);
   4530 	if (thmap_get(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   4531 		sizeof(wgp->wgp_pubkey)) != NULL ||
   4532 	    (wgp->wgp_name[0] &&
   4533 		thmap_get(wg->wg_peers_byname, wgp->wgp_name,
   4534 		    strlen(wgp->wgp_name)) != NULL)) {
   4535 		mutex_exit(wg->wg_lock);
   4536 		wg_destroy_peer(wgp);
   4537 		error = EEXIST;
   4538 		goto out;
   4539 	}
   4540 	wgp0 = thmap_put(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   4541 	    sizeof(wgp->wgp_pubkey), wgp);
   4542 	KASSERT(wgp0 == wgp);
   4543 	if (wgp->wgp_name[0]) {
   4544 		wgp0 = thmap_put(wg->wg_peers_byname, wgp->wgp_name,
   4545 		    strlen(wgp->wgp_name), wgp);
   4546 		KASSERT(wgp0 == wgp);
   4547 	}
   4548 	WG_PEER_WRITER_INSERT_HEAD(wgp, wg);
   4549 	wg->wg_npeers++;
   4550 	mutex_exit(wg->wg_lock);
   4551 
   4552 	if_link_state_change(&wg->wg_if, LINK_STATE_UP);
   4553 
   4554 out:
   4555 	kmem_free(buf, ifd->ifd_len + 1);
   4556 	return error;
   4557 }
   4558 
   4559 static int
   4560 wg_ioctl_delete_peer(struct wg_softc *wg, struct ifdrv *ifd)
   4561 {
   4562 	int error;
   4563 	prop_dictionary_t prop_dict;
   4564 	char *buf = NULL;
   4565 	const char *name;
   4566 
   4567 	error = wg_alloc_prop_buf(&buf, ifd);
   4568 	if (error != 0)
   4569 		return error;
   4570 	error = EINVAL;
   4571 	prop_dict = prop_dictionary_internalize(buf);
   4572 	if (prop_dict == NULL)
   4573 		goto out;
   4574 
   4575 	if (!prop_dictionary_get_string(prop_dict, "name", &name))
   4576 		goto out;
   4577 	if (strlen(name) > WG_PEER_NAME_MAXLEN)
   4578 		goto out;
   4579 
   4580 	error = wg_destroy_peer_name(wg, name);
   4581 out:
   4582 	kmem_free(buf, ifd->ifd_len + 1);
   4583 	return error;
   4584 }
   4585 
   4586 static bool
   4587 wg_is_authorized(struct wg_softc *wg, u_long cmd)
   4588 {
   4589 	int au = cmd == SIOCGDRVSPEC ?
   4590 	    KAUTH_REQ_NETWORK_INTERFACE_WG_GETPRIV :
   4591 	    KAUTH_REQ_NETWORK_INTERFACE_WG_SETPRIV;
   4592 	return kauth_authorize_network(kauth_cred_get(),
   4593 	    KAUTH_NETWORK_INTERFACE_WG, au, &wg->wg_if,
   4594 	    (void *)cmd, NULL) == 0;
   4595 }
   4596 
   4597 static int
   4598 wg_ioctl_get(struct wg_softc *wg, struct ifdrv *ifd)
   4599 {
   4600 	int error = ENOMEM;
   4601 	prop_dictionary_t prop_dict;
   4602 	prop_array_t peers = NULL;
   4603 	char *buf;
   4604 	struct wg_peer *wgp;
   4605 	int s, i;
   4606 
   4607 	prop_dict = prop_dictionary_create();
   4608 	if (prop_dict == NULL)
   4609 		goto error;
   4610 
   4611 	if (wg_is_authorized(wg, SIOCGDRVSPEC)) {
   4612 		if (!prop_dictionary_set_data(prop_dict, "private_key",
   4613 			wg->wg_privkey, WG_STATIC_KEY_LEN))
   4614 			goto error;
   4615 	}
   4616 
   4617 	if (wg->wg_listen_port != 0) {
   4618 		if (!prop_dictionary_set_uint16(prop_dict, "listen_port",
   4619 			wg->wg_listen_port))
   4620 			goto error;
   4621 	}
   4622 
   4623 	if (wg->wg_npeers == 0)
   4624 		goto skip_peers;
   4625 
   4626 	peers = prop_array_create();
   4627 	if (peers == NULL)
   4628 		goto error;
   4629 
   4630 	s = pserialize_read_enter();
   4631 	i = 0;
   4632 	WG_PEER_READER_FOREACH(wgp, wg) {
   4633 		struct wg_sockaddr *wgsa;
   4634 		struct psref wgp_psref, wgsa_psref;
   4635 		prop_dictionary_t prop_peer;
   4636 
   4637 		wg_get_peer(wgp, &wgp_psref);
   4638 		pserialize_read_exit(s);
   4639 
   4640 		prop_peer = prop_dictionary_create();
   4641 		if (prop_peer == NULL)
   4642 			goto next;
   4643 
   4644 		if (strlen(wgp->wgp_name) > 0) {
   4645 			if (!prop_dictionary_set_string(prop_peer, "name",
   4646 				wgp->wgp_name))
   4647 				goto next;
   4648 		}
   4649 
   4650 		if (!prop_dictionary_set_data(prop_peer, "public_key",
   4651 			wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey)))
   4652 			goto next;
   4653 
   4654 		uint8_t psk_zero[WG_PRESHARED_KEY_LEN] = {0};
   4655 		if (!consttime_memequal(wgp->wgp_psk, psk_zero,
   4656 			sizeof(wgp->wgp_psk))) {
   4657 			if (wg_is_authorized(wg, SIOCGDRVSPEC)) {
   4658 				if (!prop_dictionary_set_data(prop_peer,
   4659 					"preshared_key",
   4660 					wgp->wgp_psk, sizeof(wgp->wgp_psk)))
   4661 					goto next;
   4662 			}
   4663 		}
   4664 
   4665 		wgsa = wg_get_endpoint_sa(wgp, &wgsa_psref);
   4666 		CTASSERT(AF_UNSPEC == 0);
   4667 		if (wgsa_family(wgsa) != 0 /*AF_UNSPEC*/ &&
   4668 		    !prop_dictionary_set_data(prop_peer, "endpoint",
   4669 			wgsatoss(wgsa),
   4670 			sockaddr_getsize_by_family(wgsa_family(wgsa)))) {
   4671 			wg_put_sa(wgp, wgsa, &wgsa_psref);
   4672 			goto next;
   4673 		}
   4674 		wg_put_sa(wgp, wgsa, &wgsa_psref);
   4675 
   4676 		const struct timespec *t = &wgp->wgp_last_handshake_time;
   4677 
   4678 		if (!prop_dictionary_set_uint64(prop_peer,
   4679 			"last_handshake_time_sec", (uint64_t)t->tv_sec))
   4680 			goto next;
   4681 		if (!prop_dictionary_set_uint32(prop_peer,
   4682 			"last_handshake_time_nsec", (uint32_t)t->tv_nsec))
   4683 			goto next;
   4684 
   4685 		if (wgp->wgp_n_allowedips == 0)
   4686 			goto skip_allowedips;
   4687 
   4688 		prop_array_t allowedips = prop_array_create();
   4689 		if (allowedips == NULL)
   4690 			goto next;
   4691 		for (int j = 0; j < wgp->wgp_n_allowedips; j++) {
   4692 			struct wg_allowedip *wga = &wgp->wgp_allowedips[j];
   4693 			prop_dictionary_t prop_allowedip;
   4694 
   4695 			prop_allowedip = prop_dictionary_create();
   4696 			if (prop_allowedip == NULL)
   4697 				break;
   4698 
   4699 			if (!prop_dictionary_set_int(prop_allowedip, "family",
   4700 				wga->wga_family))
   4701 				goto _next;
   4702 			if (!prop_dictionary_set_uint8(prop_allowedip, "cidr",
   4703 				wga->wga_cidr))
   4704 				goto _next;
   4705 
   4706 			switch (wga->wga_family) {
   4707 			case AF_INET:
   4708 				if (!prop_dictionary_set_data(prop_allowedip,
   4709 					"ip", &wga->wga_addr4,
   4710 					sizeof(wga->wga_addr4)))
   4711 					goto _next;
   4712 				break;
   4713 #ifdef INET6
   4714 			case AF_INET6:
   4715 				if (!prop_dictionary_set_data(prop_allowedip,
   4716 					"ip", &wga->wga_addr6,
   4717 					sizeof(wga->wga_addr6)))
   4718 					goto _next;
   4719 				break;
   4720 #endif
   4721 			default:
   4722 				break;
   4723 			}
   4724 			prop_array_set(allowedips, j, prop_allowedip);
   4725 		_next:
   4726 			prop_object_release(prop_allowedip);
   4727 		}
   4728 		prop_dictionary_set(prop_peer, "allowedips", allowedips);
   4729 		prop_object_release(allowedips);
   4730 
   4731 	skip_allowedips:
   4732 
   4733 		prop_array_set(peers, i, prop_peer);
   4734 	next:
   4735 		if (prop_peer)
   4736 			prop_object_release(prop_peer);
   4737 		i++;
   4738 
   4739 		s = pserialize_read_enter();
   4740 		wg_put_peer(wgp, &wgp_psref);
   4741 	}
   4742 	pserialize_read_exit(s);
   4743 
   4744 	prop_dictionary_set(prop_dict, "peers", peers);
   4745 	prop_object_release(peers);
   4746 	peers = NULL;
   4747 
   4748 skip_peers:
   4749 	buf = prop_dictionary_externalize(prop_dict);
   4750 	if (buf == NULL)
   4751 		goto error;
   4752 	if (ifd->ifd_len < (strlen(buf) + 1)) {
   4753 		error = EINVAL;
   4754 		goto error;
   4755 	}
   4756 	error = copyout(buf, ifd->ifd_data, strlen(buf) + 1);
   4757 
   4758 	free(buf, 0);
   4759 error:
   4760 	if (peers != NULL)
   4761 		prop_object_release(peers);
   4762 	if (prop_dict != NULL)
   4763 		prop_object_release(prop_dict);
   4764 
   4765 	return error;
   4766 }
   4767 
   4768 static int
   4769 wg_ioctl(struct ifnet *ifp, u_long cmd, void *data)
   4770 {
   4771 	struct wg_softc *wg = ifp->if_softc;
   4772 	struct ifreq *ifr = data;
   4773 	struct ifaddr *ifa = data;
   4774 	struct ifdrv *ifd = data;
   4775 	int error = 0;
   4776 
   4777 	switch (cmd) {
   4778 	case SIOCINITIFADDR:
   4779 		if (ifa->ifa_addr->sa_family != AF_LINK &&
   4780 		    (ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
   4781 		    (IFF_UP | IFF_RUNNING)) {
   4782 			ifp->if_flags |= IFF_UP;
   4783 			error = if_init(ifp);
   4784 		}
   4785 		return error;
   4786 	case SIOCADDMULTI:
   4787 	case SIOCDELMULTI:
   4788 		switch (ifr->ifr_addr.sa_family) {
   4789 		case AF_INET:	/* IP supports Multicast */
   4790 			break;
   4791 #ifdef INET6
   4792 		case AF_INET6:	/* IP6 supports Multicast */
   4793 			break;
   4794 #endif
   4795 		default:  /* Other protocols doesn't support Multicast */
   4796 			error = EAFNOSUPPORT;
   4797 			break;
   4798 		}
   4799 		return error;
   4800 	case SIOCSDRVSPEC:
   4801 		if (!wg_is_authorized(wg, cmd)) {
   4802 			return EPERM;
   4803 		}
   4804 		switch (ifd->ifd_cmd) {
   4805 		case WG_IOCTL_SET_PRIVATE_KEY:
   4806 			error = wg_ioctl_set_private_key(wg, ifd);
   4807 			break;
   4808 		case WG_IOCTL_SET_LISTEN_PORT:
   4809 			error = wg_ioctl_set_listen_port(wg, ifd);
   4810 			break;
   4811 		case WG_IOCTL_ADD_PEER:
   4812 			error = wg_ioctl_add_peer(wg, ifd);
   4813 			break;
   4814 		case WG_IOCTL_DELETE_PEER:
   4815 			error = wg_ioctl_delete_peer(wg, ifd);
   4816 			break;
   4817 		default:
   4818 			error = EINVAL;
   4819 			break;
   4820 		}
   4821 		return error;
   4822 	case SIOCGDRVSPEC:
   4823 		return wg_ioctl_get(wg, ifd);
   4824 	case SIOCSIFFLAGS:
   4825 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
   4826 			break;
   4827 		switch (ifp->if_flags & (IFF_UP|IFF_RUNNING)) {
   4828 		case IFF_RUNNING:
   4829 			/*
   4830 			 * If interface is marked down and it is running,
   4831 			 * then stop and disable it.
   4832 			 */
   4833 			if_stop(ifp, 1);
   4834 			break;
   4835 		case IFF_UP:
   4836 			/*
   4837 			 * If interface is marked up and it is stopped, then
   4838 			 * start it.
   4839 			 */
   4840 			error = if_init(ifp);
   4841 			break;
   4842 		default:
   4843 			break;
   4844 		}
   4845 		return error;
   4846 #ifdef WG_RUMPKERNEL
   4847 	case SIOCSLINKSTR:
   4848 		error = wg_ioctl_linkstr(wg, ifd);
   4849 		if (error == 0)
   4850 			wg->wg_ops = &wg_ops_rumpuser;
   4851 		return error;
   4852 #endif
   4853 	default:
   4854 		break;
   4855 	}
   4856 
   4857 	error = ifioctl_common(ifp, cmd, data);
   4858 
   4859 #ifdef WG_RUMPKERNEL
   4860 	if (!wg_user_mode(wg))
   4861 		return error;
   4862 
   4863 	/* Do the same to the corresponding tun device on the host */
   4864 	/*
   4865 	 * XXX Actually the command has not been handled yet.  It
   4866 	 *     will be handled via pr_ioctl form doifioctl later.
   4867 	 */
   4868 	switch (cmd) {
   4869 	case SIOCAIFADDR:
   4870 	case SIOCDIFADDR: {
   4871 		struct in_aliasreq _ifra = *(const struct in_aliasreq *)data;
   4872 		struct in_aliasreq *ifra = &_ifra;
   4873 		KASSERT(error == ENOTTY);
   4874 		strncpy(ifra->ifra_name, rumpuser_wg_get_tunname(wg->wg_user),
   4875 		    IFNAMSIZ);
   4876 		error = rumpuser_wg_ioctl(wg->wg_user, cmd, ifra, AF_INET);
   4877 		if (error == 0)
   4878 			error = ENOTTY;
   4879 		break;
   4880 	}
   4881 #ifdef INET6
   4882 	case SIOCAIFADDR_IN6:
   4883 	case SIOCDIFADDR_IN6: {
   4884 		struct in6_aliasreq _ifra = *(const struct in6_aliasreq *)data;
   4885 		struct in6_aliasreq *ifra = &_ifra;
   4886 		KASSERT(error == ENOTTY);
   4887 		strncpy(ifra->ifra_name, rumpuser_wg_get_tunname(wg->wg_user),
   4888 		    IFNAMSIZ);
   4889 		error = rumpuser_wg_ioctl(wg->wg_user, cmd, ifra, AF_INET6);
   4890 		if (error == 0)
   4891 			error = ENOTTY;
   4892 		break;
   4893 	}
   4894 #endif
   4895 	}
   4896 #endif /* WG_RUMPKERNEL */
   4897 
   4898 	return error;
   4899 }
   4900 
   4901 static int
   4902 wg_init(struct ifnet *ifp)
   4903 {
   4904 
   4905 	ifp->if_flags |= IFF_RUNNING;
   4906 
   4907 	/* TODO flush pending packets. */
   4908 	return 0;
   4909 }
   4910 
   4911 #ifdef ALTQ
   4912 static void
   4913 wg_start(struct ifnet *ifp)
   4914 {
   4915 	struct mbuf *m;
   4916 
   4917 	for (;;) {
   4918 		IFQ_DEQUEUE(&ifp->if_snd, m);
   4919 		if (m == NULL)
   4920 			break;
   4921 
   4922 		kpreempt_disable();
   4923 		const uint32_t h = curcpu()->ci_index;	// pktq_rps_hash(m)
   4924 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   4925 			WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   4926 			    if_name(ifp));
   4927 			m_freem(m);
   4928 		}
   4929 		kpreempt_enable();
   4930 	}
   4931 }
   4932 #endif
   4933 
   4934 static void
   4935 wg_stop(struct ifnet *ifp, int disable)
   4936 {
   4937 
   4938 	KASSERT((ifp->if_flags & IFF_RUNNING) != 0);
   4939 	ifp->if_flags &= ~IFF_RUNNING;
   4940 
   4941 	/* Need to do something? */
   4942 }
   4943 
   4944 #ifdef WG_DEBUG_PARAMS
   4945 SYSCTL_SETUP(sysctl_net_wg_setup, "sysctl net.wg setup")
   4946 {
   4947 	const struct sysctlnode *node = NULL;
   4948 
   4949 	sysctl_createv(clog, 0, NULL, &node,
   4950 	    CTLFLAG_PERMANENT,
   4951 	    CTLTYPE_NODE, "wg",
   4952 	    SYSCTL_DESCR("wg(4)"),
   4953 	    NULL, 0, NULL, 0,
   4954 	    CTL_NET, CTL_CREATE, CTL_EOL);
   4955 	sysctl_createv(clog, 0, &node, NULL,
   4956 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4957 	    CTLTYPE_QUAD, "rekey_after_messages",
   4958 	    SYSCTL_DESCR("session liftime by messages"),
   4959 	    NULL, 0, &wg_rekey_after_messages, 0, CTL_CREATE, CTL_EOL);
   4960 	sysctl_createv(clog, 0, &node, NULL,
   4961 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4962 	    CTLTYPE_INT, "rekey_after_time",
   4963 	    SYSCTL_DESCR("session liftime"),
   4964 	    NULL, 0, &wg_rekey_after_time, 0, CTL_CREATE, CTL_EOL);
   4965 	sysctl_createv(clog, 0, &node, NULL,
   4966 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4967 	    CTLTYPE_INT, "rekey_timeout",
   4968 	    SYSCTL_DESCR("session handshake retry time"),
   4969 	    NULL, 0, &wg_rekey_timeout, 0, CTL_CREATE, CTL_EOL);
   4970 	sysctl_createv(clog, 0, &node, NULL,
   4971 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4972 	    CTLTYPE_INT, "rekey_attempt_time",
   4973 	    SYSCTL_DESCR("session handshake timeout"),
   4974 	    NULL, 0, &wg_rekey_attempt_time, 0, CTL_CREATE, CTL_EOL);
   4975 	sysctl_createv(clog, 0, &node, NULL,
   4976 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4977 	    CTLTYPE_INT, "keepalive_timeout",
   4978 	    SYSCTL_DESCR("keepalive timeout"),
   4979 	    NULL, 0, &wg_keepalive_timeout, 0, CTL_CREATE, CTL_EOL);
   4980 	sysctl_createv(clog, 0, &node, NULL,
   4981 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4982 	    CTLTYPE_BOOL, "force_underload",
   4983 	    SYSCTL_DESCR("force to detemine under load"),
   4984 	    NULL, 0, &wg_force_underload, 0, CTL_CREATE, CTL_EOL);
   4985 	sysctl_createv(clog, 0, &node, NULL,
   4986 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4987 	    CTLTYPE_INT, "debug",
   4988 	    SYSCTL_DESCR("set debug flags 1=log 2=trace 4=dump 8=packet"),
   4989 	    NULL, 0, &wg_debug, 0, CTL_CREATE, CTL_EOL);
   4990 }
   4991 #endif
   4992 
   4993 #ifdef WG_RUMPKERNEL
   4994 static bool
   4995 wg_user_mode(struct wg_softc *wg)
   4996 {
   4997 
   4998 	return wg->wg_user != NULL;
   4999 }
   5000 
   5001 static int
   5002 wg_ioctl_linkstr(struct wg_softc *wg, struct ifdrv *ifd)
   5003 {
   5004 	struct ifnet *ifp = &wg->wg_if;
   5005 	int error;
   5006 
   5007 	if (ifp->if_flags & IFF_UP)
   5008 		return EBUSY;
   5009 
   5010 	if (ifd->ifd_cmd == IFLINKSTR_UNSET) {
   5011 		/* XXX do nothing */
   5012 		return 0;
   5013 	} else if (ifd->ifd_cmd != 0) {
   5014 		return EINVAL;
   5015 	} else if (wg->wg_user != NULL) {
   5016 		return EBUSY;
   5017 	}
   5018 
   5019 	/* Assume \0 included */
   5020 	if (ifd->ifd_len > IFNAMSIZ) {
   5021 		return E2BIG;
   5022 	} else if (ifd->ifd_len < 1) {
   5023 		return EINVAL;
   5024 	}
   5025 
   5026 	char tun_name[IFNAMSIZ];
   5027 	error = copyinstr(ifd->ifd_data, tun_name, ifd->ifd_len, NULL);
   5028 	if (error != 0)
   5029 		return error;
   5030 
   5031 	if (strncmp(tun_name, "tun", 3) != 0)
   5032 		return EINVAL;
   5033 
   5034 	error = rumpuser_wg_create(tun_name, wg, &wg->wg_user);
   5035 
   5036 	return error;
   5037 }
   5038 
   5039 static int
   5040 wg_send_user(struct wg_peer *wgp, struct mbuf *m)
   5041 {
   5042 	int error;
   5043 	struct psref psref;
   5044 	struct wg_sockaddr *wgsa;
   5045 	struct wg_softc *wg = wgp->wgp_sc;
   5046 	struct iovec iov[1];
   5047 
   5048 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   5049 
   5050 	iov[0].iov_base = mtod(m, void *);
   5051 	iov[0].iov_len = m->m_len;
   5052 
   5053 	/* Send messages to a peer via an ordinary socket. */
   5054 	error = rumpuser_wg_send_peer(wg->wg_user, wgsatosa(wgsa), iov, 1);
   5055 
   5056 	wg_put_sa(wgp, wgsa, &psref);
   5057 
   5058 	m_freem(m);
   5059 
   5060 	return error;
   5061 }
   5062 
   5063 static void
   5064 wg_input_user(struct ifnet *ifp, struct mbuf *m, const int af)
   5065 {
   5066 	struct wg_softc *wg = ifp->if_softc;
   5067 	struct iovec iov[2];
   5068 	struct sockaddr_storage ss;
   5069 
   5070 	KASSERT(af == AF_INET || af == AF_INET6);
   5071 
   5072 	WG_TRACE("");
   5073 
   5074 	if (af == AF_INET) {
   5075 		struct sockaddr_in *sin = (struct sockaddr_in *)&ss;
   5076 		struct ip *ip;
   5077 
   5078 		KASSERT(m->m_len >= sizeof(struct ip));
   5079 		ip = mtod(m, struct ip *);
   5080 		sockaddr_in_init(sin, &ip->ip_dst, 0);
   5081 	} else {
   5082 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ss;
   5083 		struct ip6_hdr *ip6;
   5084 
   5085 		KASSERT(m->m_len >= sizeof(struct ip6_hdr));
   5086 		ip6 = mtod(m, struct ip6_hdr *);
   5087 		sockaddr_in6_init(sin6, &ip6->ip6_dst, 0, 0, 0);
   5088 	}
   5089 
   5090 	iov[0].iov_base = &ss;
   5091 	iov[0].iov_len = ss.ss_len;
   5092 	iov[1].iov_base = mtod(m, void *);
   5093 	iov[1].iov_len = m->m_len;
   5094 
   5095 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
   5096 
   5097 	/* Send decrypted packets to users via a tun. */
   5098 	rumpuser_wg_send_user(wg->wg_user, iov, 2);
   5099 
   5100 	m_freem(m);
   5101 }
   5102 
   5103 static int
   5104 wg_bind_port_user(struct wg_softc *wg, const uint16_t port)
   5105 {
   5106 	int error;
   5107 	uint16_t old_port = wg->wg_listen_port;
   5108 
   5109 	if (port != 0 && old_port == port)
   5110 		return 0;
   5111 
   5112 	error = rumpuser_wg_sock_bind(wg->wg_user, port);
   5113 	if (error == 0)
   5114 		wg->wg_listen_port = port;
   5115 	return error;
   5116 }
   5117 
   5118 /*
   5119  * Receive user packets.
   5120  */
   5121 void
   5122 rumpkern_wg_recv_user(struct wg_softc *wg, struct iovec *iov, size_t iovlen)
   5123 {
   5124 	struct ifnet *ifp = &wg->wg_if;
   5125 	struct mbuf *m;
   5126 	const struct sockaddr *dst;
   5127 
   5128 	WG_TRACE("");
   5129 
   5130 	dst = iov[0].iov_base;
   5131 
   5132 	m = m_gethdr(M_DONTWAIT, MT_DATA);
   5133 	if (m == NULL)
   5134 		return;
   5135 	m->m_len = m->m_pkthdr.len = 0;
   5136 	m_copyback(m, 0, iov[1].iov_len, iov[1].iov_base);
   5137 
   5138 	WG_DLOG("iov_len=%zu\n", iov[1].iov_len);
   5139 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
   5140 
   5141 	(void)wg_output(ifp, m, dst, NULL);
   5142 }
   5143 
   5144 /*
   5145  * Receive packets from a peer.
   5146  */
   5147 void
   5148 rumpkern_wg_recv_peer(struct wg_softc *wg, struct iovec *iov, size_t iovlen)
   5149 {
   5150 	struct mbuf *m;
   5151 	const struct sockaddr *src;
   5152 	int bound;
   5153 
   5154 	WG_TRACE("");
   5155 
   5156 	src = iov[0].iov_base;
   5157 
   5158 	m = m_gethdr(M_DONTWAIT, MT_DATA);
   5159 	if (m == NULL)
   5160 		return;
   5161 	m->m_len = m->m_pkthdr.len = 0;
   5162 	m_copyback(m, 0, iov[1].iov_len, iov[1].iov_base);
   5163 
   5164 	WG_DLOG("iov_len=%zu\n", iov[1].iov_len);
   5165 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
   5166 
   5167 	bound = curlwp_bind();
   5168 	wg_handle_packet(wg, m, src);
   5169 	curlwp_bindx(bound);
   5170 }
   5171 #endif /* WG_RUMPKERNEL */
   5172 
   5173 /*
   5174  * Module infrastructure
   5175  */
   5176 #include "if_module.h"
   5177 
   5178 IF_MODULE(MODULE_CLASS_DRIVER, wg, "sodium,blake2s")
   5179