Home | History | Annotate | Line # | Download | only in net
if_wg.c revision 1.85
      1 /*	$NetBSD: if_wg.c,v 1.85 2024/07/25 00:07:33 christos 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 1.85 2024/07/25 00:07:33 christos Exp $");
     45 
     46 #ifdef _KERNEL_OPT
     47 #include "opt_altq_enabled.h"
     48 #include "opt_inet.h"
     49 #endif
     50 
     51 #include <sys/param.h>
     52 #include <sys/types.h>
     53 
     54 #include <sys/atomic.h>
     55 #include <sys/callout.h>
     56 #include <sys/cprng.h>
     57 #include <sys/cpu.h>
     58 #include <sys/device.h>
     59 #include <sys/domain.h>
     60 #include <sys/errno.h>
     61 #include <sys/intr.h>
     62 #include <sys/ioctl.h>
     63 #include <sys/kernel.h>
     64 #include <sys/kmem.h>
     65 #include <sys/mbuf.h>
     66 #include <sys/module.h>
     67 #include <sys/mutex.h>
     68 #include <sys/once.h>
     69 #include <sys/percpu.h>
     70 #include <sys/pserialize.h>
     71 #include <sys/psref.h>
     72 #include <sys/queue.h>
     73 #include <sys/rwlock.h>
     74 #include <sys/socket.h>
     75 #include <sys/socketvar.h>
     76 #include <sys/sockio.h>
     77 #include <sys/sysctl.h>
     78 #include <sys/syslog.h>
     79 #include <sys/systm.h>
     80 #include <sys/thmap.h>
     81 #include <sys/threadpool.h>
     82 #include <sys/time.h>
     83 #include <sys/timespec.h>
     84 #include <sys/workqueue.h>
     85 
     86 #include <net/bpf.h>
     87 #include <net/if.h>
     88 #include <net/if_types.h>
     89 #include <net/if_wg.h>
     90 #include <net/pktqueue.h>
     91 #include <net/route.h>
     92 
     93 #include <netinet/in.h>
     94 #include <netinet/in_pcb.h>
     95 #include <netinet/in_var.h>
     96 #include <netinet/ip.h>
     97 #include <netinet/ip_var.h>
     98 #include <netinet/udp.h>
     99 #include <netinet/udp_var.h>
    100 
    101 #ifdef INET6
    102 #include <netinet/ip6.h>
    103 #include <netinet6/in6_pcb.h>
    104 #include <netinet6/in6_var.h>
    105 #include <netinet6/ip6_var.h>
    106 #include <netinet6/udp6_var.h>
    107 #endif /* INET6 */
    108 
    109 #include <prop/proplib.h>
    110 
    111 #include <crypto/blake2/blake2s.h>
    112 #include <crypto/sodium/crypto_aead_chacha20poly1305.h>
    113 #include <crypto/sodium/crypto_aead_xchacha20poly1305.h>
    114 #include <crypto/sodium/crypto_scalarmult.h>
    115 
    116 #include "ioconf.h"
    117 
    118 #ifdef WG_RUMPKERNEL
    119 #include "wg_user.h"
    120 #endif
    121 
    122 /*
    123  * Data structures
    124  * - struct wg_softc is an instance of wg interfaces
    125  *   - It has a list of peers (struct wg_peer)
    126  *   - It has a threadpool job that sends/receives handshake messages and
    127  *     runs event handlers
    128  *   - It has its own two routing tables: one is for IPv4 and the other IPv6
    129  * - struct wg_peer is a representative of a peer
    130  *   - It has a struct work to handle handshakes and timer tasks
    131  *   - It has a pair of session instances (struct wg_session)
    132  *   - It has a pair of endpoint instances (struct wg_sockaddr)
    133  *     - Normally one endpoint is used and the second one is used only on
    134  *       a peer migration (a change of peer's IP address)
    135  *   - It has a list of IP addresses and sub networks called allowedips
    136  *     (struct wg_allowedip)
    137  *     - A packets sent over a session is allowed if its destination matches
    138  *       any IP addresses or sub networks of the list
    139  * - struct wg_session represents a session of a secure tunnel with a peer
    140  *   - Two instances of sessions belong to a peer; a stable session and a
    141  *     unstable session
    142  *   - A handshake process of a session always starts with a unstable instance
    143  *   - Once a session is established, its instance becomes stable and the
    144  *     other becomes unstable instead
    145  *   - Data messages are always sent via a stable session
    146  *
    147  * Locking notes:
    148  * - Each wg has a mutex(9) wg_lock, and a rwlock(9) wg_rwlock
    149  *   - Changes to the peer list are serialized by wg_lock
    150  *   - The peer list may be read with pserialize(9) and psref(9)
    151  *   - The rwlock (wg_rwlock) protects the routing tables (wg_rtable_ipv[46])
    152  *     => XXX replace by pserialize when routing table is psz-safe
    153  * - Each peer (struct wg_peer, wgp) has a mutex wgp_lock, which can be taken
    154  *   only in thread context and serializes:
    155  *   - the stable and unstable session pointers
    156  *   - all unstable session state
    157  * - Packet processing may be done in softint context:
    158  *   - The stable session can be read under pserialize(9) or psref(9)
    159  *     - The stable session is always ESTABLISHED
    160  *     - On a session swap, we must wait for all readers to release a
    161  *       reference to a stable session before changing wgs_state and
    162  *       session states
    163  * - Lock order: wg_lock -> wgp_lock
    164  */
    165 
    166 
    167 #define WGLOG(level, fmt, args...)					      \
    168 	log(level, "%s: " fmt, __func__, ##args)
    169 
    170 #define WG_DEBUG
    171 
    172 /* Debug options */
    173 #ifdef WG_DEBUG
    174 /* Output debug logs */
    175 #ifndef WG_DEBUG_LOG
    176 #define WG_DEBUG_LOG
    177 #endif
    178 /* Output trace logs */
    179 #ifndef WG_DEBUG_TRACE
    180 #define WG_DEBUG_TRACE
    181 #endif
    182 /* Output hash values, etc. */
    183 #ifndef WG_DEBUG_DUMP
    184 #define WG_DEBUG_DUMP
    185 #endif
    186 /* debug packets */
    187 #ifndef WG_DEBUG_PACKET
    188 #define WG_DEBUG_PACKET
    189 #endif
    190 /* Make some internal parameters configurable for testing and debugging */
    191 #ifndef WG_DEBUG_PARAMS
    192 #define WG_DEBUG_PARAMS
    193 #endif
    194 #endif /* WG_DEBUG */
    195 
    196 #ifndef WG_DEBUG
    197 # if defined(WG_DEBUG_LOG) || defined(WG_DEBUG_TRACE) ||		    \
    198 	defined(WG_DEBUG_DUMP) || defined(WG_DEBUG_PARAMS)
    199 #   define WG_DEBUG
    200 # endif
    201 #endif
    202 
    203 #ifdef WG_DEBUG
    204 int wg_debug;
    205 #define WG_DEBUG_FLAGS_LOG	1
    206 #define WG_DEBUG_FLAGS_TRACE	2
    207 #define WG_DEBUG_FLAGS_DUMP	4
    208 #define WG_DEBUG_FLAGS_PACKET	8
    209 #endif
    210 
    211 
    212 #ifdef WG_DEBUG_TRACE
    213 #define WG_TRACE(msg)	 do {						\
    214 	if (wg_debug & WG_DEBUG_FLAGS_TRACE)				\
    215 	    log(LOG_DEBUG, "%s:%d: %s\n", __func__, __LINE__, (msg));	\
    216 } while (0)
    217 #else
    218 #define WG_TRACE(msg)	__nothing
    219 #endif
    220 
    221 #ifdef WG_DEBUG_LOG
    222 #define WG_DLOG(fmt, args...)	 do {					\
    223 	if (wg_debug & WG_DEBUG_FLAGS_LOG)				\
    224 	    log(LOG_DEBUG, "%s: " fmt, __func__, ##args);		\
    225 } while (0)
    226 #else
    227 #define WG_DLOG(fmt, args...)	__nothing
    228 #endif
    229 
    230 #define WG_LOG_RATECHECK(wgprc, level, fmt, args...)	do {		\
    231 	if (ppsratecheck(&(wgprc)->wgprc_lasttime,			\
    232 	    &(wgprc)->wgprc_curpps, 1)) {				\
    233 		log(level, fmt, ##args);				\
    234 	}								\
    235 } while (0)
    236 
    237 #ifdef WG_DEBUG_PARAMS
    238 static bool wg_force_underload = false;
    239 #endif
    240 
    241 #ifdef WG_DEBUG_DUMP
    242 
    243 static char *
    244 gethexdump(const void *vp, size_t n)
    245 {
    246 	char *buf;
    247 	const uint8_t *p = vp;
    248 	size_t i;
    249 
    250 	if (n > (SIZE_MAX - 1) / 3)
    251 		return NULL;
    252 	buf = kmem_alloc(3 * n + 1, KM_NOSLEEP);
    253 	if (buf == NULL)
    254 		return NULL;
    255 	for (i = 0; i < n; i++)
    256 		snprintf(buf + 3 * i, 3 + 1, " %02hhx", p[i]);
    257 	return buf;
    258 }
    259 
    260 static void
    261 puthexdump(char *buf, const void *p, size_t n)
    262 {
    263 
    264 	if (buf == NULL)
    265 		return;
    266 	kmem_free(buf, 3*n + 1);
    267 }
    268 
    269 #ifdef WG_RUMPKERNEL
    270 static void
    271 wg_dump_buf(const char *func, const char *buf, const size_t size)
    272 {
    273 	if ((wg_debug & WG_DEBUG_FLAGS_DUMP) == 0)
    274 		return;
    275 
    276 	char *hex = gethexdump(buf, size);
    277 
    278 	log(LOG_DEBUG, "%s: %s\n", func, hex ? hex : "(enomem)");
    279 	puthexdump(hex, buf, size);
    280 }
    281 #endif
    282 
    283 static void
    284 wg_dump_hash(const uint8_t *func, const uint8_t *name, const uint8_t *hash,
    285     const size_t size)
    286 {
    287 	if ((wg_debug & WG_DEBUG_FLAGS_DUMP) == 0)
    288 		return;
    289 
    290 	char *hex = gethexdump(hash, size);
    291 
    292 	log(LOG_DEBUG, "%s: %s: %s\n", func, name, hex ? hex : "(enomem)");
    293 	puthexdump(hex, hash, size);
    294 }
    295 
    296 #define WG_DUMP_HASH(name, hash) \
    297 	wg_dump_hash(__func__, name, hash, WG_HASH_LEN)
    298 #define WG_DUMP_HASH48(name, hash) \
    299 	wg_dump_hash(__func__, name, hash, 48)
    300 #define WG_DUMP_BUF(buf, size) \
    301 	wg_dump_buf(__func__, buf, size)
    302 #else
    303 #define WG_DUMP_HASH(name, hash)	__nothing
    304 #define WG_DUMP_HASH48(name, hash)	__nothing
    305 #define WG_DUMP_BUF(buf, size)	__nothing
    306 #endif /* WG_DEBUG_DUMP */
    307 
    308 /* chosen somewhat arbitrarily -- fits in signed 16 bits NUL-terminated */
    309 #define	WG_MAX_PROPLEN		32766
    310 
    311 #define WG_MTU			1420
    312 #define WG_ALLOWEDIPS		16
    313 
    314 #define CURVE25519_KEY_LEN	32
    315 #define TAI64N_LEN		sizeof(uint32_t) * 3
    316 #define POLY1305_AUTHTAG_LEN	16
    317 #define HMAC_BLOCK_LEN		64
    318 
    319 /* [N] 4.1: "DHLEN must be 32 or greater."  WireGuard chooses 32. */
    320 /* [N] 4.3: Hash functions */
    321 #define NOISE_DHLEN		32
    322 /* [N] 4.3: "Must be 32 or 64."  WireGuard chooses 32. */
    323 #define NOISE_HASHLEN		32
    324 #define NOISE_BLOCKLEN		64
    325 #define NOISE_HKDF_OUTPUT_LEN	NOISE_HASHLEN
    326 /* [N] 5.1: "k" */
    327 #define NOISE_CIPHER_KEY_LEN	32
    328 /*
    329  * [N] 9.2: "psk"
    330  *          "... psk is a 32-byte secret value provided by the application."
    331  */
    332 #define NOISE_PRESHARED_KEY_LEN	32
    333 
    334 #define WG_STATIC_KEY_LEN	CURVE25519_KEY_LEN
    335 #define WG_TIMESTAMP_LEN	TAI64N_LEN
    336 
    337 #define WG_PRESHARED_KEY_LEN	NOISE_PRESHARED_KEY_LEN
    338 
    339 #define WG_COOKIE_LEN		16
    340 #define WG_MAC_LEN		16
    341 #define WG_RANDVAL_LEN		24
    342 
    343 #define WG_EPHEMERAL_KEY_LEN	CURVE25519_KEY_LEN
    344 /* [N] 5.2: "ck: A chaining key of HASHLEN bytes" */
    345 #define WG_CHAINING_KEY_LEN	NOISE_HASHLEN
    346 /* [N] 5.2: "h: A hash output of HASHLEN bytes" */
    347 #define WG_HASH_LEN		NOISE_HASHLEN
    348 #define WG_CIPHER_KEY_LEN	NOISE_CIPHER_KEY_LEN
    349 #define WG_DH_OUTPUT_LEN	NOISE_DHLEN
    350 #define WG_KDF_OUTPUT_LEN	NOISE_HKDF_OUTPUT_LEN
    351 #define WG_AUTHTAG_LEN		POLY1305_AUTHTAG_LEN
    352 #define WG_DATA_KEY_LEN		32
    353 #define WG_SALT_LEN		24
    354 
    355 /*
    356  * The protocol messages
    357  */
    358 struct wg_msg {
    359 	uint32_t	wgm_type;
    360 } __packed;
    361 
    362 /* [W] 5.4.2 First Message: Initiator to Responder */
    363 struct wg_msg_init {
    364 	uint32_t	wgmi_type;
    365 	uint32_t	wgmi_sender;
    366 	uint8_t		wgmi_ephemeral[WG_EPHEMERAL_KEY_LEN];
    367 	uint8_t		wgmi_static[WG_STATIC_KEY_LEN + WG_AUTHTAG_LEN];
    368 	uint8_t		wgmi_timestamp[WG_TIMESTAMP_LEN + WG_AUTHTAG_LEN];
    369 	uint8_t		wgmi_mac1[WG_MAC_LEN];
    370 	uint8_t		wgmi_mac2[WG_MAC_LEN];
    371 } __packed;
    372 
    373 /* [W] 5.4.3 Second Message: Responder to Initiator */
    374 struct wg_msg_resp {
    375 	uint32_t	wgmr_type;
    376 	uint32_t	wgmr_sender;
    377 	uint32_t	wgmr_receiver;
    378 	uint8_t		wgmr_ephemeral[WG_EPHEMERAL_KEY_LEN];
    379 	uint8_t		wgmr_empty[0 + WG_AUTHTAG_LEN];
    380 	uint8_t		wgmr_mac1[WG_MAC_LEN];
    381 	uint8_t		wgmr_mac2[WG_MAC_LEN];
    382 } __packed;
    383 
    384 /* [W] 5.4.6 Subsequent Messages: Transport Data Messages */
    385 struct wg_msg_data {
    386 	uint32_t	wgmd_type;
    387 	uint32_t	wgmd_receiver;
    388 	uint64_t	wgmd_counter;
    389 	uint32_t	wgmd_packet[0];
    390 } __packed;
    391 
    392 /* [W] 5.4.7 Under Load: Cookie Reply Message */
    393 struct wg_msg_cookie {
    394 	uint32_t	wgmc_type;
    395 	uint32_t	wgmc_receiver;
    396 	uint8_t		wgmc_salt[WG_SALT_LEN];
    397 	uint8_t		wgmc_cookie[WG_COOKIE_LEN + WG_AUTHTAG_LEN];
    398 } __packed;
    399 
    400 #define WG_MSG_TYPE_INIT		1
    401 #define WG_MSG_TYPE_RESP		2
    402 #define WG_MSG_TYPE_COOKIE		3
    403 #define WG_MSG_TYPE_DATA		4
    404 #define WG_MSG_TYPE_MAX			WG_MSG_TYPE_DATA
    405 
    406 /* Sliding windows */
    407 
    408 #define	SLIWIN_BITS	2048u
    409 #define	SLIWIN_TYPE	uint32_t
    410 #define	SLIWIN_BPW	NBBY*sizeof(SLIWIN_TYPE)
    411 #define	SLIWIN_WORDS	howmany(SLIWIN_BITS, SLIWIN_BPW)
    412 #define	SLIWIN_NPKT	(SLIWIN_BITS - NBBY*sizeof(SLIWIN_TYPE))
    413 
    414 struct sliwin {
    415 	SLIWIN_TYPE	B[SLIWIN_WORDS];
    416 	uint64_t	T;
    417 };
    418 
    419 static void
    420 sliwin_reset(struct sliwin *W)
    421 {
    422 
    423 	memset(W, 0, sizeof(*W));
    424 }
    425 
    426 static int
    427 sliwin_check_fast(const volatile struct sliwin *W, uint64_t S)
    428 {
    429 
    430 	/*
    431 	 * If it's more than one window older than the highest sequence
    432 	 * number we've seen, reject.
    433 	 */
    434 #ifdef __HAVE_ATOMIC64_LOADSTORE
    435 	if (S + SLIWIN_NPKT < atomic_load_relaxed(&W->T))
    436 		return EAUTH;
    437 #endif
    438 
    439 	/*
    440 	 * Otherwise, we need to take the lock to decide, so don't
    441 	 * reject just yet.  Caller must serialize a call to
    442 	 * sliwin_update in this case.
    443 	 */
    444 	return 0;
    445 }
    446 
    447 static int
    448 sliwin_update(struct sliwin *W, uint64_t S)
    449 {
    450 	unsigned word, bit;
    451 
    452 	/*
    453 	 * If it's more than one window older than the highest sequence
    454 	 * number we've seen, reject.
    455 	 */
    456 	if (S + SLIWIN_NPKT < W->T)
    457 		return EAUTH;
    458 
    459 	/*
    460 	 * If it's higher than the highest sequence number we've seen,
    461 	 * advance the window.
    462 	 */
    463 	if (S > W->T) {
    464 		uint64_t i = W->T / SLIWIN_BPW;
    465 		uint64_t j = S / SLIWIN_BPW;
    466 		unsigned k;
    467 
    468 		for (k = 0; k < MIN(j - i, SLIWIN_WORDS); k++)
    469 			W->B[(i + k + 1) % SLIWIN_WORDS] = 0;
    470 #ifdef __HAVE_ATOMIC64_LOADSTORE
    471 		atomic_store_relaxed(&W->T, S);
    472 #else
    473 		W->T = S;
    474 #endif
    475 	}
    476 
    477 	/* Test and set the bit -- if already set, reject.  */
    478 	word = (S / SLIWIN_BPW) % SLIWIN_WORDS;
    479 	bit = S % SLIWIN_BPW;
    480 	if (W->B[word] & (1UL << bit))
    481 		return EAUTH;
    482 	W->B[word] |= 1U << bit;
    483 
    484 	/* Accept!  */
    485 	return 0;
    486 }
    487 
    488 struct wg_session {
    489 	struct wg_peer	*wgs_peer;
    490 	struct psref_target
    491 			wgs_psref;
    492 
    493 	int		wgs_state;
    494 #define WGS_STATE_UNKNOWN	0
    495 #define WGS_STATE_INIT_ACTIVE	1
    496 #define WGS_STATE_INIT_PASSIVE	2
    497 #define WGS_STATE_ESTABLISHED	3
    498 #define WGS_STATE_DESTROYING	4
    499 
    500 	time_t		wgs_time_established;
    501 	time_t		wgs_time_last_data_sent;
    502 	bool		wgs_is_initiator;
    503 
    504 	uint32_t	wgs_local_index;
    505 	uint32_t	wgs_remote_index;
    506 #ifdef __HAVE_ATOMIC64_LOADSTORE
    507 	volatile uint64_t
    508 			wgs_send_counter;
    509 #else
    510 	kmutex_t	wgs_send_counter_lock;
    511 	uint64_t	wgs_send_counter;
    512 #endif
    513 
    514 	struct {
    515 		kmutex_t	lock;
    516 		struct sliwin	window;
    517 	}		*wgs_recvwin;
    518 
    519 	uint8_t		wgs_handshake_hash[WG_HASH_LEN];
    520 	uint8_t		wgs_chaining_key[WG_CHAINING_KEY_LEN];
    521 	uint8_t		wgs_ephemeral_key_pub[WG_EPHEMERAL_KEY_LEN];
    522 	uint8_t		wgs_ephemeral_key_priv[WG_EPHEMERAL_KEY_LEN];
    523 	uint8_t		wgs_ephemeral_key_peer[WG_EPHEMERAL_KEY_LEN];
    524 	uint8_t		wgs_tkey_send[WG_DATA_KEY_LEN];
    525 	uint8_t		wgs_tkey_recv[WG_DATA_KEY_LEN];
    526 };
    527 
    528 struct wg_sockaddr {
    529 	union {
    530 		struct sockaddr_storage _ss;
    531 		struct sockaddr _sa;
    532 		struct sockaddr_in _sin;
    533 		struct sockaddr_in6 _sin6;
    534 	};
    535 	struct psref_target	wgsa_psref;
    536 };
    537 
    538 #define wgsatoss(wgsa)		(&(wgsa)->_ss)
    539 #define wgsatosa(wgsa)		(&(wgsa)->_sa)
    540 #define wgsatosin(wgsa)		(&(wgsa)->_sin)
    541 #define wgsatosin6(wgsa)	(&(wgsa)->_sin6)
    542 
    543 #define	wgsa_family(wgsa)	(wgsatosa(wgsa)->sa_family)
    544 
    545 struct wg_peer;
    546 struct wg_allowedip {
    547 	struct radix_node	wga_nodes[2];
    548 	struct wg_sockaddr	_wga_sa_addr;
    549 	struct wg_sockaddr	_wga_sa_mask;
    550 #define wga_sa_addr		_wga_sa_addr._sa
    551 #define wga_sa_mask		_wga_sa_mask._sa
    552 
    553 	int			wga_family;
    554 	uint8_t			wga_cidr;
    555 	union {
    556 		struct in_addr _ip4;
    557 		struct in6_addr _ip6;
    558 	} wga_addr;
    559 #define wga_addr4	wga_addr._ip4
    560 #define wga_addr6	wga_addr._ip6
    561 
    562 	struct wg_peer		*wga_peer;
    563 };
    564 
    565 typedef uint8_t wg_timestamp_t[WG_TIMESTAMP_LEN];
    566 
    567 struct wg_ppsratecheck {
    568 	struct timeval		wgprc_lasttime;
    569 	int			wgprc_curpps;
    570 };
    571 
    572 struct wg_softc;
    573 struct wg_peer {
    574 	struct wg_softc		*wgp_sc;
    575 	char			wgp_name[WG_PEER_NAME_MAXLEN + 1];
    576 	struct pslist_entry	wgp_peerlist_entry;
    577 	pserialize_t		wgp_psz;
    578 	struct psref_target	wgp_psref;
    579 	kmutex_t		*wgp_lock;
    580 	kmutex_t		*wgp_intr_lock;
    581 
    582 	uint8_t	wgp_pubkey[WG_STATIC_KEY_LEN];
    583 	struct wg_sockaddr	*wgp_endpoint;
    584 	struct wg_sockaddr	*wgp_endpoint0;
    585 	volatile unsigned	wgp_endpoint_changing;
    586 	bool			wgp_endpoint_available;
    587 
    588 			/* The preshared key (optional) */
    589 	uint8_t		wgp_psk[WG_PRESHARED_KEY_LEN];
    590 
    591 	struct wg_session	*wgp_session_stable;
    592 	struct wg_session	*wgp_session_unstable;
    593 
    594 	/* first outgoing packet awaiting session initiation */
    595 	struct mbuf		*wgp_pending;
    596 
    597 	/* timestamp in big-endian */
    598 	wg_timestamp_t	wgp_timestamp_latest_init;
    599 
    600 	struct timespec		wgp_last_handshake_time;
    601 
    602 	callout_t		wgp_rekey_timer;
    603 	callout_t		wgp_handshake_timeout_timer;
    604 	callout_t		wgp_session_dtor_timer;
    605 
    606 	time_t			wgp_handshake_start_time;
    607 
    608 	int			wgp_n_allowedips;
    609 	struct wg_allowedip	wgp_allowedips[WG_ALLOWEDIPS];
    610 
    611 	time_t			wgp_latest_cookie_time;
    612 	uint8_t			wgp_latest_cookie[WG_COOKIE_LEN];
    613 	uint8_t			wgp_last_sent_mac1[WG_MAC_LEN];
    614 	bool			wgp_last_sent_mac1_valid;
    615 	uint8_t			wgp_last_sent_cookie[WG_COOKIE_LEN];
    616 	bool			wgp_last_sent_cookie_valid;
    617 
    618 	time_t			wgp_last_msg_received_time[WG_MSG_TYPE_MAX];
    619 
    620 	time_t			wgp_last_genrandval_time;
    621 	uint32_t		wgp_randval;
    622 
    623 	struct wg_ppsratecheck	wgp_ppsratecheck;
    624 
    625 	struct work		wgp_work;
    626 	unsigned int		wgp_tasks;
    627 #define WGP_TASK_SEND_INIT_MESSAGE		__BIT(0)
    628 #define WGP_TASK_RETRY_HANDSHAKE		__BIT(1)
    629 #define WGP_TASK_ESTABLISH_SESSION		__BIT(2)
    630 #define WGP_TASK_ENDPOINT_CHANGED		__BIT(3)
    631 #define WGP_TASK_SEND_KEEPALIVE_MESSAGE		__BIT(4)
    632 #define WGP_TASK_DESTROY_PREV_SESSION		__BIT(5)
    633 };
    634 
    635 struct wg_ops;
    636 
    637 struct wg_softc {
    638 	struct ifnet	wg_if;
    639 	LIST_ENTRY(wg_softc) wg_list;
    640 	kmutex_t	*wg_lock;
    641 	kmutex_t	*wg_intr_lock;
    642 	krwlock_t	*wg_rwlock;
    643 
    644 	uint8_t		wg_privkey[WG_STATIC_KEY_LEN];
    645 	uint8_t		wg_pubkey[WG_STATIC_KEY_LEN];
    646 
    647 	int		wg_npeers;
    648 	struct pslist_head	wg_peers;
    649 	struct thmap	*wg_peers_bypubkey;
    650 	struct thmap	*wg_peers_byname;
    651 	struct thmap	*wg_sessions_byindex;
    652 	uint16_t	wg_listen_port;
    653 
    654 	struct threadpool	*wg_threadpool;
    655 
    656 	struct threadpool_job	wg_job;
    657 	int			wg_upcalls;
    658 #define	WG_UPCALL_INET	__BIT(0)
    659 #define	WG_UPCALL_INET6	__BIT(1)
    660 
    661 #ifdef INET
    662 	struct socket		*wg_so4;
    663 	struct radix_node_head	*wg_rtable_ipv4;
    664 #endif
    665 #ifdef INET6
    666 	struct socket		*wg_so6;
    667 	struct radix_node_head	*wg_rtable_ipv6;
    668 #endif
    669 
    670 	struct wg_ppsratecheck	wg_ppsratecheck;
    671 
    672 	struct wg_ops		*wg_ops;
    673 
    674 #ifdef WG_RUMPKERNEL
    675 	struct wg_user		*wg_user;
    676 #endif
    677 };
    678 
    679 /* [W] 6.1 Preliminaries */
    680 #define WG_REKEY_AFTER_MESSAGES		(1ULL << 60)
    681 #define WG_REJECT_AFTER_MESSAGES	(UINT64_MAX - (1 << 13))
    682 #define WG_REKEY_AFTER_TIME		120
    683 #define WG_REJECT_AFTER_TIME		180
    684 #define WG_REKEY_ATTEMPT_TIME		 90
    685 #define WG_REKEY_TIMEOUT		  5
    686 #define WG_KEEPALIVE_TIMEOUT		 10
    687 
    688 #define WG_COOKIE_TIME			120
    689 #define WG_RANDVAL_TIME			(2 * 60)
    690 
    691 static uint64_t wg_rekey_after_messages = WG_REKEY_AFTER_MESSAGES;
    692 static uint64_t wg_reject_after_messages = WG_REJECT_AFTER_MESSAGES;
    693 static unsigned wg_rekey_after_time = WG_REKEY_AFTER_TIME;
    694 static unsigned wg_reject_after_time = WG_REJECT_AFTER_TIME;
    695 static unsigned wg_rekey_attempt_time = WG_REKEY_ATTEMPT_TIME;
    696 static unsigned wg_rekey_timeout = WG_REKEY_TIMEOUT;
    697 static unsigned wg_keepalive_timeout = WG_KEEPALIVE_TIMEOUT;
    698 
    699 static struct mbuf *
    700 		wg_get_mbuf(size_t, size_t);
    701 
    702 static int	wg_send_data_msg(struct wg_peer *, struct wg_session *,
    703 		    struct mbuf *);
    704 static int	wg_send_cookie_msg(struct wg_softc *, struct wg_peer *,
    705 		    const uint32_t, const uint8_t [WG_MAC_LEN],
    706 		    const struct sockaddr *);
    707 static int	wg_send_handshake_msg_resp(struct wg_softc *, struct wg_peer *,
    708 		    struct wg_session *, const struct wg_msg_init *);
    709 static void	wg_send_keepalive_msg(struct wg_peer *, struct wg_session *);
    710 
    711 static struct wg_peer *
    712 		wg_pick_peer_by_sa(struct wg_softc *, const struct sockaddr *,
    713 		    struct psref *);
    714 static struct wg_peer *
    715 		wg_lookup_peer_by_pubkey(struct wg_softc *,
    716 		    const uint8_t [WG_STATIC_KEY_LEN], struct psref *);
    717 
    718 static struct wg_session *
    719 		wg_lookup_session_by_index(struct wg_softc *,
    720 		    const uint32_t, struct psref *);
    721 
    722 static void	wg_update_endpoint_if_necessary(struct wg_peer *,
    723 		    const struct sockaddr *);
    724 
    725 static void	wg_schedule_rekey_timer(struct wg_peer *);
    726 static void	wg_schedule_session_dtor_timer(struct wg_peer *);
    727 
    728 static bool	wg_is_underload(struct wg_softc *, struct wg_peer *, int);
    729 static void	wg_calculate_keys(struct wg_session *, const bool);
    730 
    731 static void	wg_clear_states(struct wg_session *);
    732 
    733 static void	wg_get_peer(struct wg_peer *, struct psref *);
    734 static void	wg_put_peer(struct wg_peer *, struct psref *);
    735 
    736 static int	wg_send_so(struct wg_peer *, struct mbuf *);
    737 static int	wg_send_udp(struct wg_peer *, struct mbuf *);
    738 static int	wg_output(struct ifnet *, struct mbuf *,
    739 			   const struct sockaddr *, const struct rtentry *);
    740 static void	wg_input(struct ifnet *, struct mbuf *, const int);
    741 static int	wg_ioctl(struct ifnet *, u_long, void *);
    742 static int	wg_bind_port(struct wg_softc *, const uint16_t);
    743 static int	wg_init(struct ifnet *);
    744 #ifdef ALTQ
    745 static void	wg_start(struct ifnet *);
    746 #endif
    747 static void	wg_stop(struct ifnet *, int);
    748 
    749 static void	wg_peer_work(struct work *, void *);
    750 static void	wg_job(struct threadpool_job *);
    751 static void	wgintr(void *);
    752 static void	wg_purge_pending_packets(struct wg_peer *);
    753 
    754 static int	wg_clone_create(struct if_clone *, int);
    755 static int	wg_clone_destroy(struct ifnet *);
    756 
    757 struct wg_ops {
    758 	int (*send_hs_msg)(struct wg_peer *, struct mbuf *);
    759 	int (*send_data_msg)(struct wg_peer *, struct mbuf *);
    760 	void (*input)(struct ifnet *, struct mbuf *, const int);
    761 	int (*bind_port)(struct wg_softc *, const uint16_t);
    762 };
    763 
    764 struct wg_ops wg_ops_rumpkernel = {
    765 	.send_hs_msg	= wg_send_so,
    766 	.send_data_msg	= wg_send_udp,
    767 	.input		= wg_input,
    768 	.bind_port	= wg_bind_port,
    769 };
    770 
    771 #ifdef WG_RUMPKERNEL
    772 static bool	wg_user_mode(struct wg_softc *);
    773 static int	wg_ioctl_linkstr(struct wg_softc *, struct ifdrv *);
    774 
    775 static int	wg_send_user(struct wg_peer *, struct mbuf *);
    776 static void	wg_input_user(struct ifnet *, struct mbuf *, const int);
    777 static int	wg_bind_port_user(struct wg_softc *, const uint16_t);
    778 
    779 struct wg_ops wg_ops_rumpuser = {
    780 	.send_hs_msg	= wg_send_user,
    781 	.send_data_msg	= wg_send_user,
    782 	.input		= wg_input_user,
    783 	.bind_port	= wg_bind_port_user,
    784 };
    785 #endif
    786 
    787 #define WG_PEER_READER_FOREACH(wgp, wg)					\
    788 	PSLIST_READER_FOREACH((wgp), &(wg)->wg_peers, struct wg_peer,	\
    789 	    wgp_peerlist_entry)
    790 #define WG_PEER_WRITER_FOREACH(wgp, wg)					\
    791 	PSLIST_WRITER_FOREACH((wgp), &(wg)->wg_peers, struct wg_peer,	\
    792 	    wgp_peerlist_entry)
    793 #define WG_PEER_WRITER_INSERT_HEAD(wgp, wg)				\
    794 	PSLIST_WRITER_INSERT_HEAD(&(wg)->wg_peers, (wgp), wgp_peerlist_entry)
    795 #define WG_PEER_WRITER_REMOVE(wgp)					\
    796 	PSLIST_WRITER_REMOVE((wgp), wgp_peerlist_entry)
    797 
    798 struct wg_route {
    799 	struct radix_node	wgr_nodes[2];
    800 	struct wg_peer		*wgr_peer;
    801 };
    802 
    803 static struct radix_node_head *
    804 wg_rnh(struct wg_softc *wg, const int family)
    805 {
    806 
    807 	switch (family) {
    808 		case AF_INET:
    809 			return wg->wg_rtable_ipv4;
    810 #ifdef INET6
    811 		case AF_INET6:
    812 			return wg->wg_rtable_ipv6;
    813 #endif
    814 		default:
    815 			return NULL;
    816 	}
    817 }
    818 
    819 
    820 /*
    821  * Global variables
    822  */
    823 static volatile unsigned wg_count __cacheline_aligned;
    824 
    825 struct psref_class *wg_psref_class __read_mostly;
    826 
    827 static struct if_clone wg_cloner =
    828     IF_CLONE_INITIALIZER("wg", wg_clone_create, wg_clone_destroy);
    829 
    830 static struct pktqueue *wg_pktq __read_mostly;
    831 static struct workqueue *wg_wq __read_mostly;
    832 
    833 void wgattach(int);
    834 /* ARGSUSED */
    835 void
    836 wgattach(int count)
    837 {
    838 	/*
    839 	 * Nothing to do here, initialization is handled by the
    840 	 * module initialization code in wginit() below).
    841 	 */
    842 }
    843 
    844 static void
    845 wginit(void)
    846 {
    847 
    848 	wg_psref_class = psref_class_create("wg", IPL_SOFTNET);
    849 
    850 	if_clone_attach(&wg_cloner);
    851 }
    852 
    853 /*
    854  * XXX Kludge: This should just happen in wginit, but workqueue_create
    855  * cannot be run until after CPUs have been detected, and wginit runs
    856  * before configure.
    857  */
    858 static int
    859 wginitqueues(void)
    860 {
    861 	int error __diagused;
    862 
    863 	wg_pktq = pktq_create(IFQ_MAXLEN, wgintr, NULL);
    864 	KASSERT(wg_pktq != NULL);
    865 
    866 	error = workqueue_create(&wg_wq, "wgpeer", wg_peer_work, NULL,
    867 	    PRI_NONE, IPL_SOFTNET, WQ_MPSAFE|WQ_PERCPU);
    868 	KASSERT(error == 0);
    869 
    870 	return 0;
    871 }
    872 
    873 static void
    874 wg_guarantee_initialized(void)
    875 {
    876 	static ONCE_DECL(init);
    877 	int error __diagused;
    878 
    879 	error = RUN_ONCE(&init, wginitqueues);
    880 	KASSERT(error == 0);
    881 }
    882 
    883 static int
    884 wg_count_inc(void)
    885 {
    886 	unsigned o, n;
    887 
    888 	do {
    889 		o = atomic_load_relaxed(&wg_count);
    890 		if (o == UINT_MAX)
    891 			return ENFILE;
    892 		n = o + 1;
    893 	} while (atomic_cas_uint(&wg_count, o, n) != o);
    894 
    895 	return 0;
    896 }
    897 
    898 static void
    899 wg_count_dec(void)
    900 {
    901 	unsigned c __diagused;
    902 
    903 	c = atomic_dec_uint_nv(&wg_count);
    904 	KASSERT(c != UINT_MAX);
    905 }
    906 
    907 static int
    908 wgdetach(void)
    909 {
    910 
    911 	/* Prevent new interface creation.  */
    912 	if_clone_detach(&wg_cloner);
    913 
    914 	/* Check whether there are any existing interfaces.  */
    915 	if (atomic_load_relaxed(&wg_count)) {
    916 		/* Back out -- reattach the cloner.  */
    917 		if_clone_attach(&wg_cloner);
    918 		return EBUSY;
    919 	}
    920 
    921 	/* No interfaces left.  Nuke it.  */
    922 	workqueue_destroy(wg_wq);
    923 	pktq_destroy(wg_pktq);
    924 	psref_class_destroy(wg_psref_class);
    925 
    926 	return 0;
    927 }
    928 
    929 static void
    930 wg_init_key_and_hash(uint8_t ckey[WG_CHAINING_KEY_LEN],
    931     uint8_t hash[WG_HASH_LEN])
    932 {
    933 	/* [W] 5.4: CONSTRUCTION */
    934 	const char *signature = "Noise_IKpsk2_25519_ChaChaPoly_BLAKE2s";
    935 	/* [W] 5.4: IDENTIFIER */
    936 	const char *id = "WireGuard v1 zx2c4 Jason (at) zx2c4.com";
    937 	struct blake2s state;
    938 
    939 	blake2s(ckey, WG_CHAINING_KEY_LEN, NULL, 0,
    940 	    signature, strlen(signature));
    941 
    942 	CTASSERT(WG_HASH_LEN == WG_CHAINING_KEY_LEN);
    943 	memcpy(hash, ckey, WG_CHAINING_KEY_LEN);
    944 
    945 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
    946 	blake2s_update(&state, ckey, WG_CHAINING_KEY_LEN);
    947 	blake2s_update(&state, id, strlen(id));
    948 	blake2s_final(&state, hash);
    949 
    950 	WG_DUMP_HASH("ckey", ckey);
    951 	WG_DUMP_HASH("hash", hash);
    952 }
    953 
    954 static void
    955 wg_algo_hash(uint8_t hash[WG_HASH_LEN], const uint8_t input[],
    956     const size_t inputsize)
    957 {
    958 	struct blake2s state;
    959 
    960 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
    961 	blake2s_update(&state, hash, WG_HASH_LEN);
    962 	blake2s_update(&state, input, inputsize);
    963 	blake2s_final(&state, hash);
    964 }
    965 
    966 static void
    967 wg_algo_mac(uint8_t out[], const size_t outsize,
    968     const uint8_t key[], const size_t keylen,
    969     const uint8_t input1[], const size_t input1len,
    970     const uint8_t input2[], const size_t input2len)
    971 {
    972 	struct blake2s state;
    973 
    974 	blake2s_init(&state, outsize, key, keylen);
    975 
    976 	blake2s_update(&state, input1, input1len);
    977 	if (input2 != NULL)
    978 		blake2s_update(&state, input2, input2len);
    979 	blake2s_final(&state, out);
    980 }
    981 
    982 static void
    983 wg_algo_mac_mac1(uint8_t out[], const size_t outsize,
    984     const uint8_t input1[], const size_t input1len,
    985     const uint8_t input2[], const size_t input2len)
    986 {
    987 	struct blake2s state;
    988 	/* [W] 5.4: LABEL-MAC1 */
    989 	const char *label = "mac1----";
    990 	uint8_t key[WG_HASH_LEN];
    991 
    992 	blake2s_init(&state, sizeof(key), NULL, 0);
    993 	blake2s_update(&state, label, strlen(label));
    994 	blake2s_update(&state, input1, input1len);
    995 	blake2s_final(&state, key);
    996 
    997 	blake2s_init(&state, outsize, key, sizeof(key));
    998 	if (input2 != NULL)
    999 		blake2s_update(&state, input2, input2len);
   1000 	blake2s_final(&state, out);
   1001 }
   1002 
   1003 static void
   1004 wg_algo_mac_cookie(uint8_t out[], const size_t outsize,
   1005     const uint8_t input1[], const size_t input1len)
   1006 {
   1007 	struct blake2s state;
   1008 	/* [W] 5.4: LABEL-COOKIE */
   1009 	const char *label = "cookie--";
   1010 
   1011 	blake2s_init(&state, outsize, NULL, 0);
   1012 	blake2s_update(&state, label, strlen(label));
   1013 	blake2s_update(&state, input1, input1len);
   1014 	blake2s_final(&state, out);
   1015 }
   1016 
   1017 static void
   1018 wg_algo_generate_keypair(uint8_t pubkey[WG_EPHEMERAL_KEY_LEN],
   1019     uint8_t privkey[WG_EPHEMERAL_KEY_LEN])
   1020 {
   1021 
   1022 	CTASSERT(WG_EPHEMERAL_KEY_LEN == crypto_scalarmult_curve25519_BYTES);
   1023 
   1024 	cprng_strong(kern_cprng, privkey, WG_EPHEMERAL_KEY_LEN, 0);
   1025 	crypto_scalarmult_base(pubkey, privkey);
   1026 }
   1027 
   1028 static void
   1029 wg_algo_dh(uint8_t out[WG_DH_OUTPUT_LEN],
   1030     const uint8_t privkey[WG_STATIC_KEY_LEN],
   1031     const uint8_t pubkey[WG_STATIC_KEY_LEN])
   1032 {
   1033 
   1034 	CTASSERT(WG_STATIC_KEY_LEN == crypto_scalarmult_curve25519_BYTES);
   1035 
   1036 	int ret __diagused = crypto_scalarmult(out, privkey, pubkey);
   1037 	KASSERT(ret == 0);
   1038 }
   1039 
   1040 static void
   1041 wg_algo_hmac(uint8_t out[], const size_t outlen,
   1042     const uint8_t key[], const size_t keylen,
   1043     const uint8_t in[], const size_t inlen)
   1044 {
   1045 #define IPAD	0x36
   1046 #define OPAD	0x5c
   1047 	uint8_t hmackey[HMAC_BLOCK_LEN] = {0};
   1048 	uint8_t ipad[HMAC_BLOCK_LEN];
   1049 	uint8_t opad[HMAC_BLOCK_LEN];
   1050 	size_t i;
   1051 	struct blake2s state;
   1052 
   1053 	KASSERT(outlen == WG_HASH_LEN);
   1054 	KASSERT(keylen <= HMAC_BLOCK_LEN);
   1055 
   1056 	memcpy(hmackey, key, keylen);
   1057 
   1058 	for (i = 0; i < sizeof(hmackey); i++) {
   1059 		ipad[i] = hmackey[i] ^ IPAD;
   1060 		opad[i] = hmackey[i] ^ OPAD;
   1061 	}
   1062 
   1063 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
   1064 	blake2s_update(&state, ipad, sizeof(ipad));
   1065 	blake2s_update(&state, in, inlen);
   1066 	blake2s_final(&state, out);
   1067 
   1068 	blake2s_init(&state, WG_HASH_LEN, NULL, 0);
   1069 	blake2s_update(&state, opad, sizeof(opad));
   1070 	blake2s_update(&state, out, WG_HASH_LEN);
   1071 	blake2s_final(&state, out);
   1072 #undef IPAD
   1073 #undef OPAD
   1074 }
   1075 
   1076 static void
   1077 wg_algo_kdf(uint8_t out1[WG_KDF_OUTPUT_LEN], uint8_t out2[WG_KDF_OUTPUT_LEN],
   1078     uint8_t out3[WG_KDF_OUTPUT_LEN], const uint8_t ckey[WG_CHAINING_KEY_LEN],
   1079     const uint8_t input[], const size_t inputlen)
   1080 {
   1081 	uint8_t tmp1[WG_KDF_OUTPUT_LEN], tmp2[WG_KDF_OUTPUT_LEN + 1];
   1082 	uint8_t one[1];
   1083 
   1084 	/*
   1085 	 * [N] 4.3: "an input_key_material byte sequence with length
   1086 	 * either zero bytes, 32 bytes, or DHLEN bytes."
   1087 	 */
   1088 	KASSERT(inputlen == 0 || inputlen == 32 || inputlen == NOISE_DHLEN);
   1089 
   1090 	WG_DUMP_HASH("ckey", ckey);
   1091 	if (input != NULL)
   1092 		WG_DUMP_HASH("input", input);
   1093 	wg_algo_hmac(tmp1, sizeof(tmp1), ckey, WG_CHAINING_KEY_LEN,
   1094 	    input, inputlen);
   1095 	WG_DUMP_HASH("tmp1", tmp1);
   1096 	one[0] = 1;
   1097 	wg_algo_hmac(out1, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
   1098 	    one, sizeof(one));
   1099 	WG_DUMP_HASH("out1", out1);
   1100 	if (out2 == NULL)
   1101 		return;
   1102 	memcpy(tmp2, out1, WG_KDF_OUTPUT_LEN);
   1103 	tmp2[WG_KDF_OUTPUT_LEN] = 2;
   1104 	wg_algo_hmac(out2, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
   1105 	    tmp2, sizeof(tmp2));
   1106 	WG_DUMP_HASH("out2", out2);
   1107 	if (out3 == NULL)
   1108 		return;
   1109 	memcpy(tmp2, out2, WG_KDF_OUTPUT_LEN);
   1110 	tmp2[WG_KDF_OUTPUT_LEN] = 3;
   1111 	wg_algo_hmac(out3, WG_KDF_OUTPUT_LEN, tmp1, sizeof(tmp1),
   1112 	    tmp2, sizeof(tmp2));
   1113 	WG_DUMP_HASH("out3", out3);
   1114 }
   1115 
   1116 static void __noinline
   1117 wg_algo_dh_kdf(uint8_t ckey[WG_CHAINING_KEY_LEN],
   1118     uint8_t cipher_key[WG_CIPHER_KEY_LEN],
   1119     const uint8_t local_key[WG_STATIC_KEY_LEN],
   1120     const uint8_t remote_key[WG_STATIC_KEY_LEN])
   1121 {
   1122 	uint8_t dhout[WG_DH_OUTPUT_LEN];
   1123 
   1124 	wg_algo_dh(dhout, local_key, remote_key);
   1125 	wg_algo_kdf(ckey, cipher_key, NULL, ckey, dhout, sizeof(dhout));
   1126 
   1127 	WG_DUMP_HASH("dhout", dhout);
   1128 	WG_DUMP_HASH("ckey", ckey);
   1129 	if (cipher_key != NULL)
   1130 		WG_DUMP_HASH("cipher_key", cipher_key);
   1131 }
   1132 
   1133 static void
   1134 wg_algo_aead_enc(uint8_t out[], size_t expected_outsize, const uint8_t key[],
   1135     const uint64_t counter, const uint8_t plain[], const size_t plainsize,
   1136     const uint8_t auth[], size_t authlen)
   1137 {
   1138 	uint8_t nonce[(32 + 64) / 8] = {0};
   1139 	long long unsigned int outsize;
   1140 	int error __diagused;
   1141 
   1142 	le64enc(&nonce[4], counter);
   1143 
   1144 	error = crypto_aead_chacha20poly1305_ietf_encrypt(out, &outsize, plain,
   1145 	    plainsize, auth, authlen, NULL, nonce, key);
   1146 	KASSERT(error == 0);
   1147 	KASSERT(outsize == expected_outsize);
   1148 }
   1149 
   1150 static int
   1151 wg_algo_aead_dec(uint8_t out[], size_t expected_outsize, const uint8_t key[],
   1152     const uint64_t counter, const uint8_t encrypted[],
   1153     const size_t encryptedsize, const uint8_t auth[], size_t authlen)
   1154 {
   1155 	uint8_t nonce[(32 + 64) / 8] = {0};
   1156 	long long unsigned int outsize;
   1157 	int error;
   1158 
   1159 	le64enc(&nonce[4], counter);
   1160 
   1161 	error = crypto_aead_chacha20poly1305_ietf_decrypt(out, &outsize, NULL,
   1162 	    encrypted, encryptedsize, auth, authlen, nonce, key);
   1163 	if (error == 0)
   1164 		KASSERT(outsize == expected_outsize);
   1165 	return error;
   1166 }
   1167 
   1168 static void
   1169 wg_algo_xaead_enc(uint8_t out[], const size_t expected_outsize,
   1170     const uint8_t key[], const uint8_t plain[], const size_t plainsize,
   1171     const uint8_t auth[], size_t authlen,
   1172     const uint8_t nonce[WG_SALT_LEN])
   1173 {
   1174 	long long unsigned int outsize;
   1175 	int error __diagused;
   1176 
   1177 	CTASSERT(WG_SALT_LEN == crypto_aead_xchacha20poly1305_ietf_NPUBBYTES);
   1178 	error = crypto_aead_xchacha20poly1305_ietf_encrypt(out, &outsize,
   1179 	    plain, plainsize, auth, authlen, NULL, nonce, key);
   1180 	KASSERT(error == 0);
   1181 	KASSERT(outsize == expected_outsize);
   1182 }
   1183 
   1184 static int
   1185 wg_algo_xaead_dec(uint8_t out[], const size_t expected_outsize,
   1186     const uint8_t key[], const uint8_t encrypted[], const size_t encryptedsize,
   1187     const uint8_t auth[], size_t authlen,
   1188     const uint8_t nonce[WG_SALT_LEN])
   1189 {
   1190 	long long unsigned int outsize;
   1191 	int error;
   1192 
   1193 	error = crypto_aead_xchacha20poly1305_ietf_decrypt(out, &outsize, NULL,
   1194 	    encrypted, encryptedsize, auth, authlen, nonce, key);
   1195 	if (error == 0)
   1196 		KASSERT(outsize == expected_outsize);
   1197 	return error;
   1198 }
   1199 
   1200 static void
   1201 wg_algo_tai64n(wg_timestamp_t timestamp)
   1202 {
   1203 	struct timespec ts;
   1204 
   1205 	/* FIXME strict TAI64N (https://cr.yp.to/libtai/tai64.html) */
   1206 	getnanotime(&ts);
   1207 	/* TAI64 label in external TAI64 format */
   1208 	be32enc(timestamp, 0x40000000U + (uint32_t)(ts.tv_sec >> 32));
   1209 	/* second beginning from 1970 TAI */
   1210 	be32enc(timestamp + 4, (uint32_t)(ts.tv_sec & 0xffffffffU));
   1211 	/* nanosecond in big-endian format */
   1212 	be32enc(timestamp + 8, (uint32_t)ts.tv_nsec);
   1213 }
   1214 
   1215 /*
   1216  * wg_get_stable_session(wgp, psref)
   1217  *
   1218  *	Get a passive reference to the current stable session, or
   1219  *	return NULL if there is no current stable session.
   1220  *
   1221  *	The pointer is always there but the session is not necessarily
   1222  *	ESTABLISHED; if it is not ESTABLISHED, return NULL.  However,
   1223  *	the session may transition from ESTABLISHED to DESTROYING while
   1224  *	holding the passive reference.
   1225  */
   1226 static struct wg_session *
   1227 wg_get_stable_session(struct wg_peer *wgp, struct psref *psref)
   1228 {
   1229 	int s;
   1230 	struct wg_session *wgs;
   1231 
   1232 	s = pserialize_read_enter();
   1233 	wgs = atomic_load_consume(&wgp->wgp_session_stable);
   1234 	if (__predict_false(wgs->wgs_state != WGS_STATE_ESTABLISHED))
   1235 		wgs = NULL;
   1236 	else
   1237 		psref_acquire(psref, &wgs->wgs_psref, wg_psref_class);
   1238 	pserialize_read_exit(s);
   1239 
   1240 	return wgs;
   1241 }
   1242 
   1243 static void
   1244 wg_put_session(struct wg_session *wgs, struct psref *psref)
   1245 {
   1246 
   1247 	psref_release(psref, &wgs->wgs_psref, wg_psref_class);
   1248 }
   1249 
   1250 static void
   1251 wg_destroy_session(struct wg_softc *wg, struct wg_session *wgs)
   1252 {
   1253 	struct wg_peer *wgp = wgs->wgs_peer;
   1254 	struct wg_session *wgs0 __diagused;
   1255 	void *garbage;
   1256 
   1257 	KASSERT(mutex_owned(wgp->wgp_lock));
   1258 	KASSERT(wgs->wgs_state != WGS_STATE_UNKNOWN);
   1259 
   1260 	/* Remove the session from the table.  */
   1261 	wgs0 = thmap_del(wg->wg_sessions_byindex,
   1262 	    &wgs->wgs_local_index, sizeof(wgs->wgs_local_index));
   1263 	KASSERT(wgs0 == wgs);
   1264 	garbage = thmap_stage_gc(wg->wg_sessions_byindex);
   1265 
   1266 	/* Wait for passive references to drain.  */
   1267 	pserialize_perform(wgp->wgp_psz);
   1268 	psref_target_destroy(&wgs->wgs_psref, wg_psref_class);
   1269 
   1270 	/* Free memory, zero state, and transition to UNKNOWN.  */
   1271 	thmap_gc(wg->wg_sessions_byindex, garbage);
   1272 	wg_clear_states(wgs);
   1273 	wgs->wgs_state = WGS_STATE_UNKNOWN;
   1274 }
   1275 
   1276 /*
   1277  * wg_get_session_index(wg, wgs)
   1278  *
   1279  *	Choose a session index for wgs->wgs_local_index, and store it
   1280  *	in wg's table of sessions by index.
   1281  *
   1282  *	wgs must be the unstable session of its peer, and must be
   1283  *	transitioning out of the UNKNOWN state.
   1284  */
   1285 static void
   1286 wg_get_session_index(struct wg_softc *wg, struct wg_session *wgs)
   1287 {
   1288 	struct wg_peer *wgp __diagused = wgs->wgs_peer;
   1289 	struct wg_session *wgs0;
   1290 	uint32_t index;
   1291 
   1292 	KASSERT(mutex_owned(wgp->wgp_lock));
   1293 	KASSERT(wgs == wgp->wgp_session_unstable);
   1294 	KASSERT(wgs->wgs_state == WGS_STATE_UNKNOWN);
   1295 
   1296 	do {
   1297 		/* Pick a uniform random index.  */
   1298 		index = cprng_strong32();
   1299 
   1300 		/* Try to take it.  */
   1301 		wgs->wgs_local_index = index;
   1302 		wgs0 = thmap_put(wg->wg_sessions_byindex,
   1303 		    &wgs->wgs_local_index, sizeof wgs->wgs_local_index, wgs);
   1304 
   1305 		/* If someone else beat us, start over.  */
   1306 	} while (__predict_false(wgs0 != wgs));
   1307 }
   1308 
   1309 /*
   1310  * wg_put_session_index(wg, wgs)
   1311  *
   1312  *	Remove wgs from the table of sessions by index, wait for any
   1313  *	passive references to drain, and transition the session to the
   1314  *	UNKNOWN state.
   1315  *
   1316  *	wgs must be the unstable session of its peer, and must not be
   1317  *	UNKNOWN or ESTABLISHED.
   1318  */
   1319 static void
   1320 wg_put_session_index(struct wg_softc *wg, struct wg_session *wgs)
   1321 {
   1322 	struct wg_peer *wgp __diagused = wgs->wgs_peer;
   1323 
   1324 	KASSERT(mutex_owned(wgp->wgp_lock));
   1325 	KASSERT(wgs == wgp->wgp_session_unstable);
   1326 	KASSERT(wgs->wgs_state != WGS_STATE_UNKNOWN);
   1327 	KASSERT(wgs->wgs_state != WGS_STATE_ESTABLISHED);
   1328 
   1329 	wg_destroy_session(wg, wgs);
   1330 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
   1331 }
   1332 
   1333 /*
   1334  * Handshake patterns
   1335  *
   1336  * [W] 5: "These messages use the "IK" pattern from Noise"
   1337  * [N] 7.5. Interactive handshake patterns (fundamental)
   1338  *     "The first character refers to the initiators static key:"
   1339  *     "I = Static key for initiator Immediately transmitted to responder,
   1340  *          despite reduced or absent identity hiding"
   1341  *     "The second character refers to the responders static key:"
   1342  *     "K = Static key for responder Known to initiator"
   1343  *     "IK:
   1344  *        <- s
   1345  *        ...
   1346  *        -> e, es, s, ss
   1347  *        <- e, ee, se"
   1348  * [N] 9.4. Pattern modifiers
   1349  *     "IKpsk2:
   1350  *        <- s
   1351  *        ...
   1352  *        -> e, es, s, ss
   1353  *        <- e, ee, se, psk"
   1354  */
   1355 static void
   1356 wg_fill_msg_init(struct wg_softc *wg, struct wg_peer *wgp,
   1357     struct wg_session *wgs, struct wg_msg_init *wgmi)
   1358 {
   1359 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.2: Ci */
   1360 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.2: Hi */
   1361 	uint8_t cipher_key[WG_CIPHER_KEY_LEN];
   1362 	uint8_t pubkey[WG_EPHEMERAL_KEY_LEN];
   1363 	uint8_t privkey[WG_EPHEMERAL_KEY_LEN];
   1364 
   1365 	KASSERT(mutex_owned(wgp->wgp_lock));
   1366 	KASSERT(wgs == wgp->wgp_session_unstable);
   1367 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_ACTIVE);
   1368 
   1369 	wgmi->wgmi_type = htole32(WG_MSG_TYPE_INIT);
   1370 	wgmi->wgmi_sender = wgs->wgs_local_index;
   1371 
   1372 	/* [W] 5.4.2: First Message: Initiator to Responder */
   1373 
   1374 	/* Ci := HASH(CONSTRUCTION) */
   1375 	/* Hi := HASH(Ci || IDENTIFIER) */
   1376 	wg_init_key_and_hash(ckey, hash);
   1377 	/* Hi := HASH(Hi || Sr^pub) */
   1378 	wg_algo_hash(hash, wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey));
   1379 
   1380 	WG_DUMP_HASH("hash", hash);
   1381 
   1382 	/* [N] 2.2: "e" */
   1383 	/* Ei^priv, Ei^pub := DH-GENERATE() */
   1384 	wg_algo_generate_keypair(pubkey, privkey);
   1385 	/* Ci := KDF1(Ci, Ei^pub) */
   1386 	wg_algo_kdf(ckey, NULL, NULL, ckey, pubkey, sizeof(pubkey));
   1387 	/* msg.ephemeral := Ei^pub */
   1388 	memcpy(wgmi->wgmi_ephemeral, pubkey, sizeof(wgmi->wgmi_ephemeral));
   1389 	/* Hi := HASH(Hi || msg.ephemeral) */
   1390 	wg_algo_hash(hash, pubkey, sizeof(pubkey));
   1391 
   1392 	WG_DUMP_HASH("ckey", ckey);
   1393 	WG_DUMP_HASH("hash", hash);
   1394 
   1395 	/* [N] 2.2: "es" */
   1396 	/* Ci, k := KDF2(Ci, DH(Ei^priv, Sr^pub)) */
   1397 	wg_algo_dh_kdf(ckey, cipher_key, privkey, wgp->wgp_pubkey);
   1398 
   1399 	/* [N] 2.2: "s" */
   1400 	/* msg.static := AEAD(k, 0, Si^pub, Hi) */
   1401 	wg_algo_aead_enc(wgmi->wgmi_static, sizeof(wgmi->wgmi_static),
   1402 	    cipher_key, 0, wg->wg_pubkey, sizeof(wg->wg_pubkey),
   1403 	    hash, sizeof(hash));
   1404 	/* Hi := HASH(Hi || msg.static) */
   1405 	wg_algo_hash(hash, wgmi->wgmi_static, sizeof(wgmi->wgmi_static));
   1406 
   1407 	WG_DUMP_HASH48("wgmi_static", wgmi->wgmi_static);
   1408 
   1409 	/* [N] 2.2: "ss" */
   1410 	/* Ci, k := KDF2(Ci, DH(Si^priv, Sr^pub)) */
   1411 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgp->wgp_pubkey);
   1412 
   1413 	/* msg.timestamp := AEAD(k, TIMESTAMP(), Hi) */
   1414 	wg_timestamp_t timestamp;
   1415 	wg_algo_tai64n(timestamp);
   1416 	wg_algo_aead_enc(wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp),
   1417 	    cipher_key, 0, timestamp, sizeof(timestamp), hash, sizeof(hash));
   1418 	/* Hi := HASH(Hi || msg.timestamp) */
   1419 	wg_algo_hash(hash, wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp));
   1420 
   1421 	/* [W] 5.4.4 Cookie MACs */
   1422 	wg_algo_mac_mac1(wgmi->wgmi_mac1, sizeof(wgmi->wgmi_mac1),
   1423 	    wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey),
   1424 	    (const uint8_t *)wgmi, offsetof(struct wg_msg_init, wgmi_mac1));
   1425 	/* Need mac1 to decrypt a cookie from a cookie message */
   1426 	memcpy(wgp->wgp_last_sent_mac1, wgmi->wgmi_mac1,
   1427 	    sizeof(wgp->wgp_last_sent_mac1));
   1428 	wgp->wgp_last_sent_mac1_valid = true;
   1429 
   1430 	if (wgp->wgp_latest_cookie_time == 0 ||
   1431 	    (time_uptime - wgp->wgp_latest_cookie_time) >= WG_COOKIE_TIME)
   1432 		memset(wgmi->wgmi_mac2, 0, sizeof(wgmi->wgmi_mac2));
   1433 	else {
   1434 		wg_algo_mac(wgmi->wgmi_mac2, sizeof(wgmi->wgmi_mac2),
   1435 		    wgp->wgp_latest_cookie, WG_COOKIE_LEN,
   1436 		    (const uint8_t *)wgmi,
   1437 		    offsetof(struct wg_msg_init, wgmi_mac2),
   1438 		    NULL, 0);
   1439 	}
   1440 
   1441 	memcpy(wgs->wgs_ephemeral_key_pub, pubkey, sizeof(pubkey));
   1442 	memcpy(wgs->wgs_ephemeral_key_priv, privkey, sizeof(privkey));
   1443 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
   1444 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
   1445 	WG_DLOG("%s: sender=%x\n", __func__, wgs->wgs_local_index);
   1446 }
   1447 
   1448 static void __noinline
   1449 wg_handle_msg_init(struct wg_softc *wg, const struct wg_msg_init *wgmi,
   1450     const struct sockaddr *src)
   1451 {
   1452 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.2: Ci */
   1453 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.2: Hi */
   1454 	uint8_t cipher_key[WG_CIPHER_KEY_LEN];
   1455 	uint8_t peer_pubkey[WG_STATIC_KEY_LEN];
   1456 	struct wg_peer *wgp;
   1457 	struct wg_session *wgs;
   1458 	int error, ret;
   1459 	struct psref psref_peer;
   1460 	uint8_t mac1[WG_MAC_LEN];
   1461 
   1462 	WG_TRACE("init msg received");
   1463 
   1464 	wg_algo_mac_mac1(mac1, sizeof(mac1),
   1465 	    wg->wg_pubkey, sizeof(wg->wg_pubkey),
   1466 	    (const uint8_t *)wgmi, offsetof(struct wg_msg_init, wgmi_mac1));
   1467 
   1468 	/*
   1469 	 * [W] 5.3: Denial of Service Mitigation & Cookies
   1470 	 * "the responder, ..., must always reject messages with an invalid
   1471 	 *  msg.mac1"
   1472 	 */
   1473 	if (!consttime_memequal(mac1, wgmi->wgmi_mac1, sizeof(mac1))) {
   1474 		WG_DLOG("mac1 is invalid\n");
   1475 		return;
   1476 	}
   1477 
   1478 	/*
   1479 	 * [W] 5.4.2: First Message: Initiator to Responder
   1480 	 * "When the responder receives this message, it does the same
   1481 	 *  operations so that its final state variables are identical,
   1482 	 *  replacing the operands of the DH function to produce equivalent
   1483 	 *  values."
   1484 	 *  Note that the following comments of operations are just copies of
   1485 	 *  the initiator's ones.
   1486 	 */
   1487 
   1488 	/* Ci := HASH(CONSTRUCTION) */
   1489 	/* Hi := HASH(Ci || IDENTIFIER) */
   1490 	wg_init_key_and_hash(ckey, hash);
   1491 	/* Hi := HASH(Hi || Sr^pub) */
   1492 	wg_algo_hash(hash, wg->wg_pubkey, sizeof(wg->wg_pubkey));
   1493 
   1494 	/* [N] 2.2: "e" */
   1495 	/* Ci := KDF1(Ci, Ei^pub) */
   1496 	wg_algo_kdf(ckey, NULL, NULL, ckey, wgmi->wgmi_ephemeral,
   1497 	    sizeof(wgmi->wgmi_ephemeral));
   1498 	/* Hi := HASH(Hi || msg.ephemeral) */
   1499 	wg_algo_hash(hash, wgmi->wgmi_ephemeral, sizeof(wgmi->wgmi_ephemeral));
   1500 
   1501 	WG_DUMP_HASH("ckey", ckey);
   1502 
   1503 	/* [N] 2.2: "es" */
   1504 	/* Ci, k := KDF2(Ci, DH(Ei^priv, Sr^pub)) */
   1505 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgmi->wgmi_ephemeral);
   1506 
   1507 	WG_DUMP_HASH48("wgmi_static", wgmi->wgmi_static);
   1508 
   1509 	/* [N] 2.2: "s" */
   1510 	/* msg.static := AEAD(k, 0, Si^pub, Hi) */
   1511 	error = wg_algo_aead_dec(peer_pubkey, WG_STATIC_KEY_LEN, cipher_key, 0,
   1512 	    wgmi->wgmi_static, sizeof(wgmi->wgmi_static), hash, sizeof(hash));
   1513 	if (error != 0) {
   1514 		WG_LOG_RATECHECK(&wg->wg_ppsratecheck, LOG_DEBUG,
   1515 		    "%s: wg_algo_aead_dec for secret key failed\n",
   1516 		    if_name(&wg->wg_if));
   1517 		return;
   1518 	}
   1519 	/* Hi := HASH(Hi || msg.static) */
   1520 	wg_algo_hash(hash, wgmi->wgmi_static, sizeof(wgmi->wgmi_static));
   1521 
   1522 	wgp = wg_lookup_peer_by_pubkey(wg, peer_pubkey, &psref_peer);
   1523 	if (wgp == NULL) {
   1524 		WG_DLOG("peer not found\n");
   1525 		return;
   1526 	}
   1527 
   1528 	/*
   1529 	 * Lock the peer to serialize access to cookie state.
   1530 	 *
   1531 	 * XXX Can we safely avoid holding the lock across DH?  Take it
   1532 	 * just to verify mac2 and then unlock/DH/lock?
   1533 	 */
   1534 	mutex_enter(wgp->wgp_lock);
   1535 
   1536 	if (__predict_false(wg_is_underload(wg, wgp, WG_MSG_TYPE_INIT))) {
   1537 		WG_TRACE("under load");
   1538 		/*
   1539 		 * [W] 5.3: Denial of Service Mitigation & Cookies
   1540 		 * "the responder, ..., and when under load may reject messages
   1541 		 *  with an invalid msg.mac2.  If the responder receives a
   1542 		 *  message with a valid msg.mac1 yet with an invalid msg.mac2,
   1543 		 *  and is under load, it may respond with a cookie reply
   1544 		 *  message"
   1545 		 */
   1546 		uint8_t zero[WG_MAC_LEN] = {0};
   1547 		if (consttime_memequal(wgmi->wgmi_mac2, zero, sizeof(zero))) {
   1548 			WG_TRACE("sending a cookie message: no cookie included");
   1549 			(void)wg_send_cookie_msg(wg, wgp, wgmi->wgmi_sender,
   1550 			    wgmi->wgmi_mac1, src);
   1551 			goto out;
   1552 		}
   1553 		if (!wgp->wgp_last_sent_cookie_valid) {
   1554 			WG_TRACE("sending a cookie message: no cookie sent ever");
   1555 			(void)wg_send_cookie_msg(wg, wgp, wgmi->wgmi_sender,
   1556 			    wgmi->wgmi_mac1, src);
   1557 			goto out;
   1558 		}
   1559 		uint8_t mac2[WG_MAC_LEN];
   1560 		wg_algo_mac(mac2, sizeof(mac2), wgp->wgp_last_sent_cookie,
   1561 		    WG_COOKIE_LEN, (const uint8_t *)wgmi,
   1562 		    offsetof(struct wg_msg_init, wgmi_mac2), NULL, 0);
   1563 		if (!consttime_memequal(mac2, wgmi->wgmi_mac2, sizeof(mac2))) {
   1564 			WG_DLOG("mac2 is invalid\n");
   1565 			goto out;
   1566 		}
   1567 		WG_TRACE("under load, but continue to sending");
   1568 	}
   1569 
   1570 	/* [N] 2.2: "ss" */
   1571 	/* Ci, k := KDF2(Ci, DH(Si^priv, Sr^pub)) */
   1572 	wg_algo_dh_kdf(ckey, cipher_key, wg->wg_privkey, wgp->wgp_pubkey);
   1573 
   1574 	/* msg.timestamp := AEAD(k, TIMESTAMP(), Hi) */
   1575 	wg_timestamp_t timestamp;
   1576 	error = wg_algo_aead_dec(timestamp, sizeof(timestamp), cipher_key, 0,
   1577 	    wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp),
   1578 	    hash, sizeof(hash));
   1579 	if (error != 0) {
   1580 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   1581 		    "%s: peer %s: wg_algo_aead_dec for timestamp failed\n",
   1582 		    if_name(&wg->wg_if), wgp->wgp_name);
   1583 		goto out;
   1584 	}
   1585 	/* Hi := HASH(Hi || msg.timestamp) */
   1586 	wg_algo_hash(hash, wgmi->wgmi_timestamp, sizeof(wgmi->wgmi_timestamp));
   1587 
   1588 	/*
   1589 	 * [W] 5.1 "The responder keeps track of the greatest timestamp
   1590 	 *      received per peer and discards packets containing
   1591 	 *      timestamps less than or equal to it."
   1592 	 */
   1593 	ret = memcmp(timestamp, wgp->wgp_timestamp_latest_init,
   1594 	    sizeof(timestamp));
   1595 	if (ret <= 0) {
   1596 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   1597 		    "%s: peer %s: invalid init msg: timestamp is old\n",
   1598 		    if_name(&wg->wg_if), wgp->wgp_name);
   1599 		goto out;
   1600 	}
   1601 	memcpy(wgp->wgp_timestamp_latest_init, timestamp, sizeof(timestamp));
   1602 
   1603 	/*
   1604 	 * Message is good -- we're committing to handle it now, unless
   1605 	 * we were already initiating a session.
   1606 	 */
   1607 	wgs = wgp->wgp_session_unstable;
   1608 	switch (wgs->wgs_state) {
   1609 	case WGS_STATE_UNKNOWN:		/* new session initiated by peer */
   1610 		wg_get_session_index(wg, wgs);
   1611 		break;
   1612 	case WGS_STATE_INIT_ACTIVE:	/* we're already initiating, drop */
   1613 		WG_TRACE("Session already initializing, ignoring the message");
   1614 		goto out;
   1615 	case WGS_STATE_INIT_PASSIVE:	/* peer is retrying, start over */
   1616 		WG_TRACE("Session already initializing, destroying old states");
   1617 		wg_clear_states(wgs);
   1618 		/* keep session index */
   1619 		break;
   1620 	case WGS_STATE_ESTABLISHED:	/* can't happen */
   1621 		panic("unstable session can't be established");
   1622 		break;
   1623 	case WGS_STATE_DESTROYING:	/* rekey initiated by peer */
   1624 		WG_TRACE("Session destroying, but force to clear");
   1625 		callout_stop(&wgp->wgp_session_dtor_timer);
   1626 		wg_clear_states(wgs);
   1627 		/* keep session index */
   1628 		break;
   1629 	default:
   1630 		panic("invalid session state: %d", wgs->wgs_state);
   1631 	}
   1632 	wgs->wgs_state = WGS_STATE_INIT_PASSIVE;
   1633 
   1634 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
   1635 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
   1636 	memcpy(wgs->wgs_ephemeral_key_peer, wgmi->wgmi_ephemeral,
   1637 	    sizeof(wgmi->wgmi_ephemeral));
   1638 
   1639 	wg_update_endpoint_if_necessary(wgp, src);
   1640 
   1641 	(void)wg_send_handshake_msg_resp(wg, wgp, wgs, wgmi);
   1642 
   1643 	wg_calculate_keys(wgs, false);
   1644 	wg_clear_states(wgs);
   1645 
   1646 out:
   1647 	mutex_exit(wgp->wgp_lock);
   1648 	wg_put_peer(wgp, &psref_peer);
   1649 }
   1650 
   1651 static struct socket *
   1652 wg_get_so_by_af(struct wg_softc *wg, const int af)
   1653 {
   1654 
   1655 	switch (af) {
   1656 #ifdef INET
   1657 	case AF_INET:
   1658 		return wg->wg_so4;
   1659 #endif
   1660 #ifdef INET6
   1661 	case AF_INET6:
   1662 		return wg->wg_so6;
   1663 #endif
   1664 	default:
   1665 		panic("wg: no such af: %d", af);
   1666 	}
   1667 }
   1668 
   1669 static struct socket *
   1670 wg_get_so_by_peer(struct wg_peer *wgp, struct wg_sockaddr *wgsa)
   1671 {
   1672 
   1673 	return wg_get_so_by_af(wgp->wgp_sc, wgsa_family(wgsa));
   1674 }
   1675 
   1676 static struct wg_sockaddr *
   1677 wg_get_endpoint_sa(struct wg_peer *wgp, struct psref *psref)
   1678 {
   1679 	struct wg_sockaddr *wgsa;
   1680 	int s;
   1681 
   1682 	s = pserialize_read_enter();
   1683 	wgsa = atomic_load_consume(&wgp->wgp_endpoint);
   1684 	psref_acquire(psref, &wgsa->wgsa_psref, wg_psref_class);
   1685 	pserialize_read_exit(s);
   1686 
   1687 	return wgsa;
   1688 }
   1689 
   1690 static void
   1691 wg_put_sa(struct wg_peer *wgp, struct wg_sockaddr *wgsa, struct psref *psref)
   1692 {
   1693 
   1694 	psref_release(psref, &wgsa->wgsa_psref, wg_psref_class);
   1695 }
   1696 
   1697 static int
   1698 wg_send_so(struct wg_peer *wgp, struct mbuf *m)
   1699 {
   1700 	int error;
   1701 	struct socket *so;
   1702 	struct psref psref;
   1703 	struct wg_sockaddr *wgsa;
   1704 
   1705 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   1706 	so = wg_get_so_by_peer(wgp, wgsa);
   1707 	error = sosend(so, wgsatosa(wgsa), NULL, m, NULL, 0, curlwp);
   1708 	wg_put_sa(wgp, wgsa, &psref);
   1709 
   1710 	return error;
   1711 }
   1712 
   1713 static int
   1714 wg_send_handshake_msg_init(struct wg_softc *wg, struct wg_peer *wgp)
   1715 {
   1716 	int error;
   1717 	struct mbuf *m;
   1718 	struct wg_msg_init *wgmi;
   1719 	struct wg_session *wgs;
   1720 
   1721 	KASSERT(mutex_owned(wgp->wgp_lock));
   1722 
   1723 	wgs = wgp->wgp_session_unstable;
   1724 	/* XXX pull dispatch out into wg_task_send_init_message */
   1725 	switch (wgs->wgs_state) {
   1726 	case WGS_STATE_UNKNOWN:		/* new session initiated by us */
   1727 		wg_get_session_index(wg, wgs);
   1728 		break;
   1729 	case WGS_STATE_INIT_ACTIVE:	/* we're already initiating, stop */
   1730 		WG_TRACE("Session already initializing, skip starting new one");
   1731 		return EBUSY;
   1732 	case WGS_STATE_INIT_PASSIVE:	/* peer was trying -- XXX what now? */
   1733 		WG_TRACE("Session already initializing, destroying old states");
   1734 		wg_clear_states(wgs);
   1735 		/* keep session index */
   1736 		break;
   1737 	case WGS_STATE_ESTABLISHED:	/* can't happen */
   1738 		panic("unstable session can't be established");
   1739 		break;
   1740 	case WGS_STATE_DESTROYING:	/* rekey initiated by us too early */
   1741 		WG_TRACE("Session destroying");
   1742 		/* XXX should wait? */
   1743 		return EBUSY;
   1744 	}
   1745 	wgs->wgs_state = WGS_STATE_INIT_ACTIVE;
   1746 
   1747 	m = m_gethdr(M_WAIT, MT_DATA);
   1748 	if (sizeof(*wgmi) > MHLEN) {
   1749 		m_clget(m, M_WAIT);
   1750 		CTASSERT(sizeof(*wgmi) <= MCLBYTES);
   1751 	}
   1752 	m->m_pkthdr.len = m->m_len = sizeof(*wgmi);
   1753 	wgmi = mtod(m, struct wg_msg_init *);
   1754 	wg_fill_msg_init(wg, wgp, wgs, wgmi);
   1755 
   1756 	error = wg->wg_ops->send_hs_msg(wgp, m);
   1757 	if (error == 0) {
   1758 		WG_TRACE("init msg sent");
   1759 
   1760 		if (wgp->wgp_handshake_start_time == 0)
   1761 			wgp->wgp_handshake_start_time = time_uptime;
   1762 		callout_schedule(&wgp->wgp_handshake_timeout_timer,
   1763 		    MIN(wg_rekey_timeout, (unsigned)(INT_MAX / hz)) * hz);
   1764 	} else {
   1765 		wg_put_session_index(wg, wgs);
   1766 		/* Initiation failed; toss packet waiting for it if any.  */
   1767 		m = atomic_swap_ptr(&wgp->wgp_pending, NULL);
   1768 		m_freem(m);
   1769 	}
   1770 
   1771 	return error;
   1772 }
   1773 
   1774 static void
   1775 wg_fill_msg_resp(struct wg_softc *wg, struct wg_peer *wgp,
   1776     struct wg_session *wgs, struct wg_msg_resp *wgmr,
   1777     const struct wg_msg_init *wgmi)
   1778 {
   1779 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.3: Cr */
   1780 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.3: Hr */
   1781 	uint8_t cipher_key[WG_KDF_OUTPUT_LEN];
   1782 	uint8_t pubkey[WG_EPHEMERAL_KEY_LEN];
   1783 	uint8_t privkey[WG_EPHEMERAL_KEY_LEN];
   1784 
   1785 	KASSERT(mutex_owned(wgp->wgp_lock));
   1786 	KASSERT(wgs == wgp->wgp_session_unstable);
   1787 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_PASSIVE);
   1788 
   1789 	memcpy(hash, wgs->wgs_handshake_hash, sizeof(hash));
   1790 	memcpy(ckey, wgs->wgs_chaining_key, sizeof(ckey));
   1791 
   1792 	wgmr->wgmr_type = htole32(WG_MSG_TYPE_RESP);
   1793 	wgmr->wgmr_sender = wgs->wgs_local_index;
   1794 	wgmr->wgmr_receiver = wgmi->wgmi_sender;
   1795 
   1796 	/* [W] 5.4.3 Second Message: Responder to Initiator */
   1797 
   1798 	/* [N] 2.2: "e" */
   1799 	/* Er^priv, Er^pub := DH-GENERATE() */
   1800 	wg_algo_generate_keypair(pubkey, privkey);
   1801 	/* Cr := KDF1(Cr, Er^pub) */
   1802 	wg_algo_kdf(ckey, NULL, NULL, ckey, pubkey, sizeof(pubkey));
   1803 	/* msg.ephemeral := Er^pub */
   1804 	memcpy(wgmr->wgmr_ephemeral, pubkey, sizeof(wgmr->wgmr_ephemeral));
   1805 	/* Hr := HASH(Hr || msg.ephemeral) */
   1806 	wg_algo_hash(hash, pubkey, sizeof(pubkey));
   1807 
   1808 	WG_DUMP_HASH("ckey", ckey);
   1809 	WG_DUMP_HASH("hash", hash);
   1810 
   1811 	/* [N] 2.2: "ee" */
   1812 	/* Cr := KDF1(Cr, DH(Er^priv, Ei^pub)) */
   1813 	wg_algo_dh_kdf(ckey, NULL, privkey, wgs->wgs_ephemeral_key_peer);
   1814 
   1815 	/* [N] 2.2: "se" */
   1816 	/* Cr := KDF1(Cr, DH(Er^priv, Si^pub)) */
   1817 	wg_algo_dh_kdf(ckey, NULL, privkey, wgp->wgp_pubkey);
   1818 
   1819 	/* [N] 9.2: "psk" */
   1820     {
   1821 	uint8_t kdfout[WG_KDF_OUTPUT_LEN];
   1822 	/* Cr, r, k := KDF3(Cr, Q) */
   1823 	wg_algo_kdf(ckey, kdfout, cipher_key, ckey, wgp->wgp_psk,
   1824 	    sizeof(wgp->wgp_psk));
   1825 	/* Hr := HASH(Hr || r) */
   1826 	wg_algo_hash(hash, kdfout, sizeof(kdfout));
   1827     }
   1828 
   1829 	/* msg.empty := AEAD(k, 0, e, Hr) */
   1830 	wg_algo_aead_enc(wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty),
   1831 	    cipher_key, 0, NULL, 0, hash, sizeof(hash));
   1832 	/* Hr := HASH(Hr || msg.empty) */
   1833 	wg_algo_hash(hash, wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty));
   1834 
   1835 	WG_DUMP_HASH("wgmr_empty", wgmr->wgmr_empty);
   1836 
   1837 	/* [W] 5.4.4: Cookie MACs */
   1838 	/* msg.mac1 := MAC(HASH(LABEL-MAC1 || Sm'^pub), msg_a) */
   1839 	wg_algo_mac_mac1(wgmr->wgmr_mac1, sizeof(wgmi->wgmi_mac1),
   1840 	    wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey),
   1841 	    (const uint8_t *)wgmr, offsetof(struct wg_msg_resp, wgmr_mac1));
   1842 	/* Need mac1 to decrypt a cookie from a cookie message */
   1843 	memcpy(wgp->wgp_last_sent_mac1, wgmr->wgmr_mac1,
   1844 	    sizeof(wgp->wgp_last_sent_mac1));
   1845 	wgp->wgp_last_sent_mac1_valid = true;
   1846 
   1847 	if (wgp->wgp_latest_cookie_time == 0 ||
   1848 	    (time_uptime - wgp->wgp_latest_cookie_time) >= WG_COOKIE_TIME)
   1849 		/* msg.mac2 := 0^16 */
   1850 		memset(wgmr->wgmr_mac2, 0, sizeof(wgmr->wgmr_mac2));
   1851 	else {
   1852 		/* msg.mac2 := MAC(Lm, msg_b) */
   1853 		wg_algo_mac(wgmr->wgmr_mac2, sizeof(wgmi->wgmi_mac2),
   1854 		    wgp->wgp_latest_cookie, WG_COOKIE_LEN,
   1855 		    (const uint8_t *)wgmr,
   1856 		    offsetof(struct wg_msg_resp, wgmr_mac2),
   1857 		    NULL, 0);
   1858 	}
   1859 
   1860 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(hash));
   1861 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(ckey));
   1862 	memcpy(wgs->wgs_ephemeral_key_pub, pubkey, sizeof(pubkey));
   1863 	memcpy(wgs->wgs_ephemeral_key_priv, privkey, sizeof(privkey));
   1864 	wgs->wgs_remote_index = wgmi->wgmi_sender;
   1865 	WG_DLOG("sender=%x\n", wgs->wgs_local_index);
   1866 	WG_DLOG("receiver=%x\n", wgs->wgs_remote_index);
   1867 }
   1868 
   1869 static void
   1870 wg_swap_sessions(struct wg_peer *wgp)
   1871 {
   1872 	struct wg_session *wgs, *wgs_prev;
   1873 
   1874 	KASSERT(mutex_owned(wgp->wgp_lock));
   1875 
   1876 	wgs = wgp->wgp_session_unstable;
   1877 	KASSERT(wgs->wgs_state == WGS_STATE_ESTABLISHED);
   1878 
   1879 	wgs_prev = wgp->wgp_session_stable;
   1880 	KASSERT(wgs_prev->wgs_state == WGS_STATE_ESTABLISHED ||
   1881 	    wgs_prev->wgs_state == WGS_STATE_UNKNOWN);
   1882 	atomic_store_release(&wgp->wgp_session_stable, wgs);
   1883 	wgp->wgp_session_unstable = wgs_prev;
   1884 }
   1885 
   1886 static void __noinline
   1887 wg_handle_msg_resp(struct wg_softc *wg, const struct wg_msg_resp *wgmr,
   1888     const struct sockaddr *src)
   1889 {
   1890 	uint8_t ckey[WG_CHAINING_KEY_LEN]; /* [W] 5.4.3: Cr */
   1891 	uint8_t hash[WG_HASH_LEN]; /* [W] 5.4.3: Kr */
   1892 	uint8_t cipher_key[WG_KDF_OUTPUT_LEN];
   1893 	struct wg_peer *wgp;
   1894 	struct wg_session *wgs;
   1895 	struct psref psref;
   1896 	int error;
   1897 	uint8_t mac1[WG_MAC_LEN];
   1898 	struct wg_session *wgs_prev;
   1899 	struct mbuf *m;
   1900 
   1901 	wg_algo_mac_mac1(mac1, sizeof(mac1),
   1902 	    wg->wg_pubkey, sizeof(wg->wg_pubkey),
   1903 	    (const uint8_t *)wgmr, offsetof(struct wg_msg_resp, wgmr_mac1));
   1904 
   1905 	/*
   1906 	 * [W] 5.3: Denial of Service Mitigation & Cookies
   1907 	 * "the responder, ..., must always reject messages with an invalid
   1908 	 *  msg.mac1"
   1909 	 */
   1910 	if (!consttime_memequal(mac1, wgmr->wgmr_mac1, sizeof(mac1))) {
   1911 		WG_DLOG("mac1 is invalid\n");
   1912 		return;
   1913 	}
   1914 
   1915 	WG_TRACE("resp msg received");
   1916 	wgs = wg_lookup_session_by_index(wg, wgmr->wgmr_receiver, &psref);
   1917 	if (wgs == NULL) {
   1918 		WG_TRACE("No session found");
   1919 		return;
   1920 	}
   1921 
   1922 	wgp = wgs->wgs_peer;
   1923 
   1924 	mutex_enter(wgp->wgp_lock);
   1925 
   1926 	/* If we weren't waiting for a handshake response, drop it.  */
   1927 	if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE) {
   1928 		WG_TRACE("peer sent spurious handshake response, ignoring");
   1929 		goto out;
   1930 	}
   1931 
   1932 	if (__predict_false(wg_is_underload(wg, wgp, WG_MSG_TYPE_RESP))) {
   1933 		WG_TRACE("under load");
   1934 		/*
   1935 		 * [W] 5.3: Denial of Service Mitigation & Cookies
   1936 		 * "the responder, ..., and when under load may reject messages
   1937 		 *  with an invalid msg.mac2.  If the responder receives a
   1938 		 *  message with a valid msg.mac1 yet with an invalid msg.mac2,
   1939 		 *  and is under load, it may respond with a cookie reply
   1940 		 *  message"
   1941 		 */
   1942 		uint8_t zero[WG_MAC_LEN] = {0};
   1943 		if (consttime_memequal(wgmr->wgmr_mac2, zero, sizeof(zero))) {
   1944 			WG_TRACE("sending a cookie message: no cookie included");
   1945 			(void)wg_send_cookie_msg(wg, wgp, wgmr->wgmr_sender,
   1946 			    wgmr->wgmr_mac1, src);
   1947 			goto out;
   1948 		}
   1949 		if (!wgp->wgp_last_sent_cookie_valid) {
   1950 			WG_TRACE("sending a cookie message: no cookie sent ever");
   1951 			(void)wg_send_cookie_msg(wg, wgp, wgmr->wgmr_sender,
   1952 			    wgmr->wgmr_mac1, src);
   1953 			goto out;
   1954 		}
   1955 		uint8_t mac2[WG_MAC_LEN];
   1956 		wg_algo_mac(mac2, sizeof(mac2), wgp->wgp_last_sent_cookie,
   1957 		    WG_COOKIE_LEN, (const uint8_t *)wgmr,
   1958 		    offsetof(struct wg_msg_resp, wgmr_mac2), NULL, 0);
   1959 		if (!consttime_memequal(mac2, wgmr->wgmr_mac2, sizeof(mac2))) {
   1960 			WG_DLOG("mac2 is invalid\n");
   1961 			goto out;
   1962 		}
   1963 		WG_TRACE("under load, but continue to sending");
   1964 	}
   1965 
   1966 	memcpy(hash, wgs->wgs_handshake_hash, sizeof(hash));
   1967 	memcpy(ckey, wgs->wgs_chaining_key, sizeof(ckey));
   1968 
   1969 	/*
   1970 	 * [W] 5.4.3 Second Message: Responder to Initiator
   1971 	 * "When the initiator receives this message, it does the same
   1972 	 *  operations so that its final state variables are identical,
   1973 	 *  replacing the operands of the DH function to produce equivalent
   1974 	 *  values."
   1975 	 *  Note that the following comments of operations are just copies of
   1976 	 *  the initiator's ones.
   1977 	 */
   1978 
   1979 	/* [N] 2.2: "e" */
   1980 	/* Cr := KDF1(Cr, Er^pub) */
   1981 	wg_algo_kdf(ckey, NULL, NULL, ckey, wgmr->wgmr_ephemeral,
   1982 	    sizeof(wgmr->wgmr_ephemeral));
   1983 	/* Hr := HASH(Hr || msg.ephemeral) */
   1984 	wg_algo_hash(hash, wgmr->wgmr_ephemeral, sizeof(wgmr->wgmr_ephemeral));
   1985 
   1986 	WG_DUMP_HASH("ckey", ckey);
   1987 	WG_DUMP_HASH("hash", hash);
   1988 
   1989 	/* [N] 2.2: "ee" */
   1990 	/* Cr := KDF1(Cr, DH(Er^priv, Ei^pub)) */
   1991 	wg_algo_dh_kdf(ckey, NULL, wgs->wgs_ephemeral_key_priv,
   1992 	    wgmr->wgmr_ephemeral);
   1993 
   1994 	/* [N] 2.2: "se" */
   1995 	/* Cr := KDF1(Cr, DH(Er^priv, Si^pub)) */
   1996 	wg_algo_dh_kdf(ckey, NULL, wg->wg_privkey, wgmr->wgmr_ephemeral);
   1997 
   1998 	/* [N] 9.2: "psk" */
   1999     {
   2000 	uint8_t kdfout[WG_KDF_OUTPUT_LEN];
   2001 	/* Cr, r, k := KDF3(Cr, Q) */
   2002 	wg_algo_kdf(ckey, kdfout, cipher_key, ckey, wgp->wgp_psk,
   2003 	    sizeof(wgp->wgp_psk));
   2004 	/* Hr := HASH(Hr || r) */
   2005 	wg_algo_hash(hash, kdfout, sizeof(kdfout));
   2006     }
   2007 
   2008     {
   2009 	uint8_t out[sizeof(wgmr->wgmr_empty)]; /* for safety */
   2010 	/* msg.empty := AEAD(k, 0, e, Hr) */
   2011 	error = wg_algo_aead_dec(out, 0, cipher_key, 0, wgmr->wgmr_empty,
   2012 	    sizeof(wgmr->wgmr_empty), hash, sizeof(hash));
   2013 	WG_DUMP_HASH("wgmr_empty", wgmr->wgmr_empty);
   2014 	if (error != 0) {
   2015 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2016 		    "%s: peer %s: wg_algo_aead_dec for empty message failed\n",
   2017 		    if_name(&wg->wg_if), wgp->wgp_name);
   2018 		goto out;
   2019 	}
   2020 	/* Hr := HASH(Hr || msg.empty) */
   2021 	wg_algo_hash(hash, wgmr->wgmr_empty, sizeof(wgmr->wgmr_empty));
   2022     }
   2023 
   2024 	memcpy(wgs->wgs_handshake_hash, hash, sizeof(wgs->wgs_handshake_hash));
   2025 	memcpy(wgs->wgs_chaining_key, ckey, sizeof(wgs->wgs_chaining_key));
   2026 	wgs->wgs_remote_index = wgmr->wgmr_sender;
   2027 	WG_DLOG("receiver=%x\n", wgs->wgs_remote_index);
   2028 
   2029 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_ACTIVE);
   2030 	wgs->wgs_state = WGS_STATE_ESTABLISHED;
   2031 	wgs->wgs_time_established = time_uptime;
   2032 	wgs->wgs_time_last_data_sent = 0;
   2033 	wgs->wgs_is_initiator = true;
   2034 	wg_calculate_keys(wgs, true);
   2035 	wg_clear_states(wgs);
   2036 	WG_TRACE("WGS_STATE_ESTABLISHED");
   2037 
   2038 	callout_stop(&wgp->wgp_handshake_timeout_timer);
   2039 
   2040 	wg_swap_sessions(wgp);
   2041 	KASSERT(wgs == wgp->wgp_session_stable);
   2042 	wgs_prev = wgp->wgp_session_unstable;
   2043 	getnanotime(&wgp->wgp_last_handshake_time);
   2044 	wgp->wgp_handshake_start_time = 0;
   2045 	wgp->wgp_last_sent_mac1_valid = false;
   2046 	wgp->wgp_last_sent_cookie_valid = false;
   2047 
   2048 	wg_schedule_rekey_timer(wgp);
   2049 
   2050 	wg_update_endpoint_if_necessary(wgp, src);
   2051 
   2052 	/*
   2053 	 * If we had a data packet queued up, send it; otherwise send a
   2054 	 * keepalive message -- either way we have to send something
   2055 	 * immediately or else the responder will never answer.
   2056 	 */
   2057 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL) {
   2058 		kpreempt_disable();
   2059 		const uint32_t h = curcpu()->ci_index; // pktq_rps_hash(m)
   2060 		M_SETCTX(m, wgp);
   2061 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   2062 			WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   2063 			    if_name(&wg->wg_if));
   2064 			m_freem(m);
   2065 		}
   2066 		kpreempt_enable();
   2067 	} else {
   2068 		wg_send_keepalive_msg(wgp, wgs);
   2069 	}
   2070 
   2071 	if (wgs_prev->wgs_state == WGS_STATE_ESTABLISHED) {
   2072 		/* Wait for wg_get_stable_session to drain.  */
   2073 		pserialize_perform(wgp->wgp_psz);
   2074 
   2075 		/* Transition ESTABLISHED->DESTROYING.  */
   2076 		wgs_prev->wgs_state = WGS_STATE_DESTROYING;
   2077 
   2078 		/* We can't destroy the old session immediately */
   2079 		wg_schedule_session_dtor_timer(wgp);
   2080 	} else {
   2081 		KASSERTMSG(wgs_prev->wgs_state == WGS_STATE_UNKNOWN,
   2082 		    "state=%d", wgs_prev->wgs_state);
   2083 	}
   2084 
   2085 out:
   2086 	mutex_exit(wgp->wgp_lock);
   2087 	wg_put_session(wgs, &psref);
   2088 }
   2089 
   2090 static int
   2091 wg_send_handshake_msg_resp(struct wg_softc *wg, struct wg_peer *wgp,
   2092     struct wg_session *wgs, const struct wg_msg_init *wgmi)
   2093 {
   2094 	int error;
   2095 	struct mbuf *m;
   2096 	struct wg_msg_resp *wgmr;
   2097 
   2098 	KASSERT(mutex_owned(wgp->wgp_lock));
   2099 	KASSERT(wgs == wgp->wgp_session_unstable);
   2100 	KASSERT(wgs->wgs_state == WGS_STATE_INIT_PASSIVE);
   2101 
   2102 	m = m_gethdr(M_WAIT, MT_DATA);
   2103 	if (sizeof(*wgmr) > MHLEN) {
   2104 		m_clget(m, M_WAIT);
   2105 		CTASSERT(sizeof(*wgmr) <= MCLBYTES);
   2106 	}
   2107 	m->m_pkthdr.len = m->m_len = sizeof(*wgmr);
   2108 	wgmr = mtod(m, struct wg_msg_resp *);
   2109 	wg_fill_msg_resp(wg, wgp, wgs, wgmr, wgmi);
   2110 
   2111 	error = wg->wg_ops->send_hs_msg(wgp, m);
   2112 	if (error == 0)
   2113 		WG_TRACE("resp msg sent");
   2114 	return error;
   2115 }
   2116 
   2117 static struct wg_peer *
   2118 wg_lookup_peer_by_pubkey(struct wg_softc *wg,
   2119     const uint8_t pubkey[WG_STATIC_KEY_LEN], struct psref *psref)
   2120 {
   2121 	struct wg_peer *wgp;
   2122 
   2123 	int s = pserialize_read_enter();
   2124 	wgp = thmap_get(wg->wg_peers_bypubkey, pubkey, WG_STATIC_KEY_LEN);
   2125 	if (wgp != NULL)
   2126 		wg_get_peer(wgp, psref);
   2127 	pserialize_read_exit(s);
   2128 
   2129 	return wgp;
   2130 }
   2131 
   2132 static void
   2133 wg_fill_msg_cookie(struct wg_softc *wg, struct wg_peer *wgp,
   2134     struct wg_msg_cookie *wgmc, const uint32_t sender,
   2135     const uint8_t mac1[WG_MAC_LEN], const struct sockaddr *src)
   2136 {
   2137 	uint8_t cookie[WG_COOKIE_LEN];
   2138 	uint8_t key[WG_HASH_LEN];
   2139 	uint8_t addr[sizeof(struct in6_addr)];
   2140 	size_t addrlen;
   2141 	uint16_t uh_sport; /* be */
   2142 
   2143 	KASSERT(mutex_owned(wgp->wgp_lock));
   2144 
   2145 	wgmc->wgmc_type = htole32(WG_MSG_TYPE_COOKIE);
   2146 	wgmc->wgmc_receiver = sender;
   2147 	cprng_fast(wgmc->wgmc_salt, sizeof(wgmc->wgmc_salt));
   2148 
   2149 	/*
   2150 	 * [W] 5.4.7: Under Load: Cookie Reply Message
   2151 	 * "The secret variable, Rm, changes every two minutes to a
   2152 	 * random value"
   2153 	 */
   2154 	if ((time_uptime - wgp->wgp_last_genrandval_time) > WG_RANDVAL_TIME) {
   2155 		wgp->wgp_randval = cprng_strong32();
   2156 		wgp->wgp_last_genrandval_time = time_uptime;
   2157 	}
   2158 
   2159 	switch (src->sa_family) {
   2160 	case AF_INET: {
   2161 		const struct sockaddr_in *sin = satocsin(src);
   2162 		addrlen = sizeof(sin->sin_addr);
   2163 		memcpy(addr, &sin->sin_addr, addrlen);
   2164 		uh_sport = sin->sin_port;
   2165 		break;
   2166 	    }
   2167 #ifdef INET6
   2168 	case AF_INET6: {
   2169 		const struct sockaddr_in6 *sin6 = satocsin6(src);
   2170 		addrlen = sizeof(sin6->sin6_addr);
   2171 		memcpy(addr, &sin6->sin6_addr, addrlen);
   2172 		uh_sport = sin6->sin6_port;
   2173 		break;
   2174 	    }
   2175 #endif
   2176 	default:
   2177 		panic("invalid af=%d", src->sa_family);
   2178 	}
   2179 
   2180 	wg_algo_mac(cookie, sizeof(cookie),
   2181 	    (const uint8_t *)&wgp->wgp_randval, sizeof(wgp->wgp_randval),
   2182 	    addr, addrlen, (const uint8_t *)&uh_sport, sizeof(uh_sport));
   2183 	wg_algo_mac_cookie(key, sizeof(key), wg->wg_pubkey,
   2184 	    sizeof(wg->wg_pubkey));
   2185 	wg_algo_xaead_enc(wgmc->wgmc_cookie, sizeof(wgmc->wgmc_cookie), key,
   2186 	    cookie, sizeof(cookie), mac1, WG_MAC_LEN, wgmc->wgmc_salt);
   2187 
   2188 	/* Need to store to calculate mac2 */
   2189 	memcpy(wgp->wgp_last_sent_cookie, cookie, sizeof(cookie));
   2190 	wgp->wgp_last_sent_cookie_valid = true;
   2191 }
   2192 
   2193 static int
   2194 wg_send_cookie_msg(struct wg_softc *wg, struct wg_peer *wgp,
   2195     const uint32_t sender, const uint8_t mac1[WG_MAC_LEN],
   2196     const struct sockaddr *src)
   2197 {
   2198 	int error;
   2199 	struct mbuf *m;
   2200 	struct wg_msg_cookie *wgmc;
   2201 
   2202 	KASSERT(mutex_owned(wgp->wgp_lock));
   2203 
   2204 	m = m_gethdr(M_WAIT, MT_DATA);
   2205 	if (sizeof(*wgmc) > MHLEN) {
   2206 		m_clget(m, M_WAIT);
   2207 		CTASSERT(sizeof(*wgmc) <= MCLBYTES);
   2208 	}
   2209 	m->m_pkthdr.len = m->m_len = sizeof(*wgmc);
   2210 	wgmc = mtod(m, struct wg_msg_cookie *);
   2211 	wg_fill_msg_cookie(wg, wgp, wgmc, sender, mac1, src);
   2212 
   2213 	error = wg->wg_ops->send_hs_msg(wgp, m);
   2214 	if (error == 0)
   2215 		WG_TRACE("cookie msg sent");
   2216 	return error;
   2217 }
   2218 
   2219 static bool
   2220 wg_is_underload(struct wg_softc *wg, struct wg_peer *wgp, int msgtype)
   2221 {
   2222 #ifdef WG_DEBUG_PARAMS
   2223 	if (wg_force_underload)
   2224 		return true;
   2225 #endif
   2226 
   2227 	/*
   2228 	 * XXX we don't have a means of a load estimation.  The purpose of
   2229 	 * the mechanism is a DoS mitigation, so we consider frequent handshake
   2230 	 * messages as (a kind of) load; if a message of the same type comes
   2231 	 * to a peer within 1 second, we consider we are under load.
   2232 	 */
   2233 	time_t last = wgp->wgp_last_msg_received_time[msgtype];
   2234 	wgp->wgp_last_msg_received_time[msgtype] = time_uptime;
   2235 	return (time_uptime - last) == 0;
   2236 }
   2237 
   2238 static void
   2239 wg_calculate_keys(struct wg_session *wgs, const bool initiator)
   2240 {
   2241 
   2242 	KASSERT(mutex_owned(wgs->wgs_peer->wgp_lock));
   2243 
   2244 	/*
   2245 	 * [W] 5.4.5: Ti^send = Tr^recv, Ti^recv = Tr^send := KDF2(Ci = Cr, e)
   2246 	 */
   2247 	if (initiator) {
   2248 		wg_algo_kdf(wgs->wgs_tkey_send, wgs->wgs_tkey_recv, NULL,
   2249 		    wgs->wgs_chaining_key, NULL, 0);
   2250 	} else {
   2251 		wg_algo_kdf(wgs->wgs_tkey_recv, wgs->wgs_tkey_send, NULL,
   2252 		    wgs->wgs_chaining_key, NULL, 0);
   2253 	}
   2254 	WG_DUMP_HASH("wgs_tkey_send", wgs->wgs_tkey_send);
   2255 	WG_DUMP_HASH("wgs_tkey_recv", wgs->wgs_tkey_recv);
   2256 }
   2257 
   2258 static uint64_t
   2259 wg_session_get_send_counter(struct wg_session *wgs)
   2260 {
   2261 #ifdef __HAVE_ATOMIC64_LOADSTORE
   2262 	return atomic_load_relaxed(&wgs->wgs_send_counter);
   2263 #else
   2264 	uint64_t send_counter;
   2265 
   2266 	mutex_enter(&wgs->wgs_send_counter_lock);
   2267 	send_counter = wgs->wgs_send_counter;
   2268 	mutex_exit(&wgs->wgs_send_counter_lock);
   2269 
   2270 	return send_counter;
   2271 #endif
   2272 }
   2273 
   2274 static uint64_t
   2275 wg_session_inc_send_counter(struct wg_session *wgs)
   2276 {
   2277 #ifdef __HAVE_ATOMIC64_LOADSTORE
   2278 	return atomic_inc_64_nv(&wgs->wgs_send_counter) - 1;
   2279 #else
   2280 	uint64_t send_counter;
   2281 
   2282 	mutex_enter(&wgs->wgs_send_counter_lock);
   2283 	send_counter = wgs->wgs_send_counter++;
   2284 	mutex_exit(&wgs->wgs_send_counter_lock);
   2285 
   2286 	return send_counter;
   2287 #endif
   2288 }
   2289 
   2290 static void
   2291 wg_clear_states(struct wg_session *wgs)
   2292 {
   2293 
   2294 	KASSERT(mutex_owned(wgs->wgs_peer->wgp_lock));
   2295 
   2296 	wgs->wgs_send_counter = 0;
   2297 	sliwin_reset(&wgs->wgs_recvwin->window);
   2298 
   2299 #define wgs_clear(v)	explicit_memset(wgs->wgs_##v, 0, sizeof(wgs->wgs_##v))
   2300 	wgs_clear(handshake_hash);
   2301 	wgs_clear(chaining_key);
   2302 	wgs_clear(ephemeral_key_pub);
   2303 	wgs_clear(ephemeral_key_priv);
   2304 	wgs_clear(ephemeral_key_peer);
   2305 #undef wgs_clear
   2306 }
   2307 
   2308 static struct wg_session *
   2309 wg_lookup_session_by_index(struct wg_softc *wg, const uint32_t index,
   2310     struct psref *psref)
   2311 {
   2312 	struct wg_session *wgs;
   2313 
   2314 	int s = pserialize_read_enter();
   2315 	wgs = thmap_get(wg->wg_sessions_byindex, &index, sizeof index);
   2316 	if (wgs != NULL) {
   2317 		KASSERT(atomic_load_relaxed(&wgs->wgs_state) !=
   2318 		    WGS_STATE_UNKNOWN);
   2319 		psref_acquire(psref, &wgs->wgs_psref, wg_psref_class);
   2320 	}
   2321 	pserialize_read_exit(s);
   2322 
   2323 	return wgs;
   2324 }
   2325 
   2326 static void
   2327 wg_schedule_rekey_timer(struct wg_peer *wgp)
   2328 {
   2329 	int timeout = MIN(wg_rekey_after_time, (unsigned)(INT_MAX / hz));
   2330 
   2331 	callout_schedule(&wgp->wgp_rekey_timer, timeout * hz);
   2332 }
   2333 
   2334 static void
   2335 wg_send_keepalive_msg(struct wg_peer *wgp, struct wg_session *wgs)
   2336 {
   2337 	struct mbuf *m;
   2338 
   2339 	/*
   2340 	 * [W] 6.5 Passive Keepalive
   2341 	 * "A keepalive message is simply a transport data message with
   2342 	 *  a zero-length encapsulated encrypted inner-packet."
   2343 	 */
   2344 	WG_TRACE("");
   2345 	m = m_gethdr(M_WAIT, MT_DATA);
   2346 	wg_send_data_msg(wgp, wgs, m);
   2347 }
   2348 
   2349 static bool
   2350 wg_need_to_send_init_message(struct wg_session *wgs)
   2351 {
   2352 	/*
   2353 	 * [W] 6.2 Transport Message Limits
   2354 	 * "if a peer is the initiator of a current secure session,
   2355 	 *  WireGuard will send a handshake initiation message to begin
   2356 	 *  a new secure session ... if after receiving a transport data
   2357 	 *  message, the current secure session is (REJECT-AFTER-TIME 
   2358 	 *  KEEPALIVE-TIMEOUT  REKEY-TIMEOUT) seconds old and it has
   2359 	 *  not yet acted upon this event."
   2360 	 */
   2361 	return wgs->wgs_is_initiator && wgs->wgs_time_last_data_sent == 0 &&
   2362 	    (time_uptime - wgs->wgs_time_established) >=
   2363 	    (wg_reject_after_time - wg_keepalive_timeout - wg_rekey_timeout);
   2364 }
   2365 
   2366 static void
   2367 wg_schedule_peer_task(struct wg_peer *wgp, unsigned int task)
   2368 {
   2369 
   2370 	mutex_enter(wgp->wgp_intr_lock);
   2371 	WG_DLOG("tasks=%d, task=%d\n", wgp->wgp_tasks, task);
   2372 	if (wgp->wgp_tasks == 0)
   2373 		/*
   2374 		 * XXX If the current CPU is already loaded -- e.g., if
   2375 		 * there's already a bunch of handshakes queued up --
   2376 		 * consider tossing this over to another CPU to
   2377 		 * distribute the load.
   2378 		 */
   2379 		workqueue_enqueue(wg_wq, &wgp->wgp_work, NULL);
   2380 	wgp->wgp_tasks |= task;
   2381 	mutex_exit(wgp->wgp_intr_lock);
   2382 }
   2383 
   2384 static void
   2385 wg_change_endpoint(struct wg_peer *wgp, const struct sockaddr *new)
   2386 {
   2387 	struct wg_sockaddr *wgsa_prev;
   2388 
   2389 	WG_TRACE("Changing endpoint");
   2390 
   2391 	memcpy(wgp->wgp_endpoint0, new, new->sa_len);
   2392 	wgsa_prev = wgp->wgp_endpoint;
   2393 	atomic_store_release(&wgp->wgp_endpoint, wgp->wgp_endpoint0);
   2394 	wgp->wgp_endpoint0 = wgsa_prev;
   2395 	atomic_store_release(&wgp->wgp_endpoint_available, true);
   2396 
   2397 	wg_schedule_peer_task(wgp, WGP_TASK_ENDPOINT_CHANGED);
   2398 }
   2399 
   2400 static bool
   2401 wg_validate_inner_packet(const char *packet, size_t decrypted_len, int *af)
   2402 {
   2403 	uint16_t packet_len;
   2404 	const struct ip *ip;
   2405 
   2406 	if (__predict_false(decrypted_len < sizeof(*ip))) {
   2407 		WG_DLOG("decrypted_len=%zu < %zu\n", decrypted_len,
   2408 		    sizeof(*ip));
   2409 		return false;
   2410 	}
   2411 
   2412 	ip = (const struct ip *)packet;
   2413 	if (ip->ip_v == 4)
   2414 		*af = AF_INET;
   2415 	else if (ip->ip_v == 6)
   2416 		*af = AF_INET6;
   2417 	else {
   2418 		WG_DLOG("ip_v=%d\n", ip->ip_v);
   2419 		return false;
   2420 	}
   2421 
   2422 	WG_DLOG("af=%d\n", *af);
   2423 
   2424 	switch (*af) {
   2425 #ifdef INET
   2426 	case AF_INET:
   2427 		packet_len = ntohs(ip->ip_len);
   2428 		break;
   2429 #endif
   2430 #ifdef INET6
   2431 	case AF_INET6: {
   2432 		const struct ip6_hdr *ip6;
   2433 
   2434 		if (__predict_false(decrypted_len < sizeof(*ip6))) {
   2435 			WG_DLOG("decrypted_len=%zu < %zu\n", decrypted_len,
   2436 			    sizeof(*ip6));
   2437 			return false;
   2438 		}
   2439 
   2440 		ip6 = (const struct ip6_hdr *)packet;
   2441 		packet_len = sizeof(*ip6) + ntohs(ip6->ip6_plen);
   2442 		break;
   2443 	}
   2444 #endif
   2445 	default:
   2446 		return false;
   2447 	}
   2448 
   2449 	if (packet_len > decrypted_len) {
   2450 		WG_DLOG("packet_len %u > decrypted_len %zu\n", packet_len,
   2451 		    decrypted_len);
   2452 		return false;
   2453 	}
   2454 
   2455 	return true;
   2456 }
   2457 
   2458 static bool
   2459 wg_validate_route(struct wg_softc *wg, struct wg_peer *wgp_expected,
   2460     int af, char *packet)
   2461 {
   2462 	struct sockaddr_storage ss;
   2463 	struct sockaddr *sa;
   2464 	struct psref psref;
   2465 	struct wg_peer *wgp;
   2466 	bool ok;
   2467 
   2468 	/*
   2469 	 * II CRYPTOKEY ROUTING
   2470 	 * "it will only accept it if its source IP resolves in the
   2471 	 *  table to the public key used in the secure session for
   2472 	 *  decrypting it."
   2473 	 */
   2474 
   2475 	if (af == AF_INET) {
   2476 		const struct ip *ip = (const struct ip *)packet;
   2477 		struct sockaddr_in *sin = (struct sockaddr_in *)&ss;
   2478 		sockaddr_in_init(sin, &ip->ip_src, 0);
   2479 		sa = sintosa(sin);
   2480 #ifdef INET6
   2481 	} else {
   2482 		const struct ip6_hdr *ip6 = (const struct ip6_hdr *)packet;
   2483 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ss;
   2484 		sockaddr_in6_init(sin6, &ip6->ip6_src, 0, 0, 0);
   2485 		sa = sin6tosa(sin6);
   2486 #endif
   2487 	}
   2488 
   2489 	wgp = wg_pick_peer_by_sa(wg, sa, &psref);
   2490 	ok = (wgp == wgp_expected);
   2491 	if (wgp != NULL)
   2492 		wg_put_peer(wgp, &psref);
   2493 
   2494 	return ok;
   2495 }
   2496 
   2497 static void
   2498 wg_session_dtor_timer(void *arg)
   2499 {
   2500 	struct wg_peer *wgp = arg;
   2501 
   2502 	WG_TRACE("enter");
   2503 
   2504 	wg_schedule_peer_task(wgp, WGP_TASK_DESTROY_PREV_SESSION);
   2505 }
   2506 
   2507 static void
   2508 wg_schedule_session_dtor_timer(struct wg_peer *wgp)
   2509 {
   2510 
   2511 	/* 1 second grace period */
   2512 	callout_schedule(&wgp->wgp_session_dtor_timer, hz);
   2513 }
   2514 
   2515 static bool
   2516 sockaddr_port_match(const struct sockaddr *sa1, const struct sockaddr *sa2)
   2517 {
   2518 	if (sa1->sa_family != sa2->sa_family)
   2519 		return false;
   2520 
   2521 	switch (sa1->sa_family) {
   2522 #ifdef INET
   2523 	case AF_INET:
   2524 		return satocsin(sa1)->sin_port == satocsin(sa2)->sin_port;
   2525 #endif
   2526 #ifdef INET6
   2527 	case AF_INET6:
   2528 		return satocsin6(sa1)->sin6_port == satocsin6(sa2)->sin6_port;
   2529 #endif
   2530 	default:
   2531 		return false;
   2532 	}
   2533 }
   2534 
   2535 static void
   2536 wg_update_endpoint_if_necessary(struct wg_peer *wgp,
   2537     const struct sockaddr *src)
   2538 {
   2539 	struct wg_sockaddr *wgsa;
   2540 	struct psref psref;
   2541 
   2542 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   2543 
   2544 #ifdef WG_DEBUG_LOG
   2545 	char oldaddr[128], newaddr[128];
   2546 	sockaddr_format(wgsatosa(wgsa), oldaddr, sizeof(oldaddr));
   2547 	sockaddr_format(src, newaddr, sizeof(newaddr));
   2548 	WG_DLOG("old=%s, new=%s\n", oldaddr, newaddr);
   2549 #endif
   2550 
   2551 	/*
   2552 	 * III: "Since the packet has authenticated correctly, the source IP of
   2553 	 * the outer UDP/IP packet is used to update the endpoint for peer..."
   2554 	 */
   2555 	if (__predict_false(sockaddr_cmp(src, wgsatosa(wgsa)) != 0 ||
   2556 		!sockaddr_port_match(src, wgsatosa(wgsa)))) {
   2557 		/* XXX We can't change the endpoint twice in a short period */
   2558 		if (atomic_swap_uint(&wgp->wgp_endpoint_changing, 1) == 0) {
   2559 			wg_change_endpoint(wgp, src);
   2560 		}
   2561 	}
   2562 
   2563 	wg_put_sa(wgp, wgsa, &psref);
   2564 }
   2565 
   2566 static void __noinline
   2567 wg_handle_msg_data(struct wg_softc *wg, struct mbuf *m,
   2568     const struct sockaddr *src)
   2569 {
   2570 	struct wg_msg_data *wgmd;
   2571 	char *encrypted_buf = NULL, *decrypted_buf;
   2572 	size_t encrypted_len, decrypted_len;
   2573 	struct wg_session *wgs;
   2574 	struct wg_peer *wgp;
   2575 	int state;
   2576 	size_t mlen;
   2577 	struct psref psref;
   2578 	int error, af;
   2579 	bool success, free_encrypted_buf = false, ok;
   2580 	struct mbuf *n;
   2581 
   2582 	KASSERT(m->m_len >= sizeof(struct wg_msg_data));
   2583 	wgmd = mtod(m, struct wg_msg_data *);
   2584 
   2585 	KASSERT(wgmd->wgmd_type == htole32(WG_MSG_TYPE_DATA));
   2586 	WG_TRACE("data");
   2587 
   2588 	/* Find the putative session, or drop.  */
   2589 	wgs = wg_lookup_session_by_index(wg, wgmd->wgmd_receiver, &psref);
   2590 	if (wgs == NULL) {
   2591 		WG_TRACE("No session found");
   2592 		m_freem(m);
   2593 		return;
   2594 	}
   2595 
   2596 	/*
   2597 	 * We are only ready to handle data when in INIT_PASSIVE,
   2598 	 * ESTABLISHED, or DESTROYING.  All transitions out of that
   2599 	 * state dissociate the session index and drain psrefs.
   2600 	 */
   2601 	state = atomic_load_relaxed(&wgs->wgs_state);
   2602 	switch (state) {
   2603 	case WGS_STATE_UNKNOWN:
   2604 		panic("wg session %p in unknown state has session index %u",
   2605 		    wgs, wgmd->wgmd_receiver);
   2606 	case WGS_STATE_INIT_ACTIVE:
   2607 		WG_TRACE("not yet ready for data");
   2608 		goto out;
   2609 	case WGS_STATE_INIT_PASSIVE:
   2610 	case WGS_STATE_ESTABLISHED:
   2611 	case WGS_STATE_DESTROYING:
   2612 		break;
   2613 	}
   2614 
   2615 	/*
   2616 	 * Get the peer, for rate-limited logs (XXX MPSAFE, dtrace) and
   2617 	 * to update the endpoint if authentication succeeds.
   2618 	 */
   2619 	wgp = wgs->wgs_peer;
   2620 
   2621 	/*
   2622 	 * Reject outrageously wrong sequence numbers before doing any
   2623 	 * crypto work or taking any locks.
   2624 	 */
   2625 	error = sliwin_check_fast(&wgs->wgs_recvwin->window,
   2626 	    le64toh(wgmd->wgmd_counter));
   2627 	if (error) {
   2628 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2629 		    "%s: peer %s: out-of-window packet: %"PRIu64"\n",
   2630 		    if_name(&wg->wg_if), wgp->wgp_name,
   2631 		    le64toh(wgmd->wgmd_counter));
   2632 		goto out;
   2633 	}
   2634 
   2635 	/* Ensure the payload and authenticator are contiguous.  */
   2636 	mlen = m_length(m);
   2637 	encrypted_len = mlen - sizeof(*wgmd);
   2638 	if (encrypted_len < WG_AUTHTAG_LEN) {
   2639 		WG_DLOG("Short encrypted_len: %lu\n", encrypted_len);
   2640 		goto out;
   2641 	}
   2642 	success = m_ensure_contig(&m, sizeof(*wgmd) + encrypted_len);
   2643 	if (success) {
   2644 		encrypted_buf = mtod(m, char *) + sizeof(*wgmd);
   2645 	} else {
   2646 		encrypted_buf = kmem_intr_alloc(encrypted_len, KM_NOSLEEP);
   2647 		if (encrypted_buf == NULL) {
   2648 			WG_DLOG("failed to allocate encrypted_buf\n");
   2649 			goto out;
   2650 		}
   2651 		m_copydata(m, sizeof(*wgmd), encrypted_len, encrypted_buf);
   2652 		free_encrypted_buf = true;
   2653 	}
   2654 	/* m_ensure_contig may change m regardless of its result */
   2655 	KASSERT(m->m_len >= sizeof(*wgmd));
   2656 	wgmd = mtod(m, struct wg_msg_data *);
   2657 
   2658 	/*
   2659 	 * Get a buffer for the plaintext.  Add WG_AUTHTAG_LEN to avoid
   2660 	 * a zero-length buffer (XXX).  Drop if plaintext is longer
   2661 	 * than MCLBYTES (XXX).
   2662 	 */
   2663 	decrypted_len = encrypted_len - WG_AUTHTAG_LEN;
   2664 	if (decrypted_len > MCLBYTES) {
   2665 		/* FIXME handle larger data than MCLBYTES */
   2666 		WG_DLOG("couldn't handle larger data than MCLBYTES\n");
   2667 		goto out;
   2668 	}
   2669 	n = wg_get_mbuf(0, decrypted_len + WG_AUTHTAG_LEN);
   2670 	if (n == NULL) {
   2671 		WG_DLOG("wg_get_mbuf failed\n");
   2672 		goto out;
   2673 	}
   2674 	decrypted_buf = mtod(n, char *);
   2675 
   2676 	/* Decrypt and verify the packet.  */
   2677 	WG_DLOG("mlen=%lu, encrypted_len=%lu\n", mlen, encrypted_len);
   2678 	error = wg_algo_aead_dec(decrypted_buf,
   2679 	    encrypted_len - WG_AUTHTAG_LEN /* can be 0 */,
   2680 	    wgs->wgs_tkey_recv, le64toh(wgmd->wgmd_counter), encrypted_buf,
   2681 	    encrypted_len, NULL, 0);
   2682 	if (error != 0) {
   2683 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2684 		    "%s: peer %s: failed to wg_algo_aead_dec\n",
   2685 		    if_name(&wg->wg_if), wgp->wgp_name);
   2686 		m_freem(n);
   2687 		goto out;
   2688 	}
   2689 	WG_DLOG("outsize=%u\n", (u_int)decrypted_len);
   2690 
   2691 	/* Packet is genuine.  Reject it if a replay or just too old.  */
   2692 	mutex_enter(&wgs->wgs_recvwin->lock);
   2693 	error = sliwin_update(&wgs->wgs_recvwin->window,
   2694 	    le64toh(wgmd->wgmd_counter));
   2695 	mutex_exit(&wgs->wgs_recvwin->lock);
   2696 	if (error) {
   2697 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2698 		    "%s: peer %s: replay or out-of-window packet: %"PRIu64"\n",
   2699 		    if_name(&wg->wg_if), wgp->wgp_name,
   2700 		    le64toh(wgmd->wgmd_counter));
   2701 		m_freem(n);
   2702 		goto out;
   2703 	}
   2704 
   2705 #ifdef WG_DEBUG_PACKET
   2706 	if (wg_debug & WG_DEBUG_FLAGS_PACKET) {
   2707 		char *hex = gethexdump(wgmd, sizeof(*wgmd));
   2708 		if (hex != NULL) {
   2709 			log(LOG_DEBUG, "wgmd=%p, sizeof(*wgmd)=%zu\n%s\n",
   2710 			    wgmd, sizeof(*wgmd), hex);
   2711 			puthexdump(hex, wgmd, sizeof(*wgmd));
   2712 		}
   2713 		hex = gethexdump(decrypted_buf, decrypted_len);
   2714 		if (hex != NULL) {
   2715 			log(LOG_DEBUG, "decrypted_buf=%p, "
   2716 			    "decrypted_len=%zu\n%s\n",
   2717 			decrypted_buf, decrypted_len, hex);
   2718 			puthexdump(hex, decrypted_buf, decrypted_len);
   2719 		}
   2720 	}
   2721 #endif
   2722 	/* We're done with m now; free it and chuck the pointers.  */
   2723 	m_freem(m);
   2724 	m = NULL;
   2725 	wgmd = NULL;
   2726 
   2727 	/*
   2728 	 * Validate the encapsulated packet header and get the address
   2729 	 * family, or drop.
   2730 	 */
   2731 	ok = wg_validate_inner_packet(decrypted_buf, decrypted_len, &af);
   2732 	if (!ok) {
   2733 		m_freem(n);
   2734 		goto out;
   2735 	}
   2736 
   2737 	/*
   2738 	 * The packet is genuine.  Update the peer's endpoint if the
   2739 	 * source address changed.
   2740 	 *
   2741 	 * XXX How to prevent DoS by replaying genuine packets from the
   2742 	 * wrong source address?
   2743 	 */
   2744 	wg_update_endpoint_if_necessary(wgp, src);
   2745 
   2746 	/* Submit it into our network stack if routable.  */
   2747 	ok = wg_validate_route(wg, wgp, af, decrypted_buf);
   2748 	if (ok) {
   2749 		wg->wg_ops->input(&wg->wg_if, n, af);
   2750 	} else {
   2751 		char addrstr[INET6_ADDRSTRLEN];
   2752 		memset(addrstr, 0, sizeof(addrstr));
   2753 		if (af == AF_INET) {
   2754 			const struct ip *ip = (const struct ip *)decrypted_buf;
   2755 			IN_PRINT(addrstr, &ip->ip_src);
   2756 #ifdef INET6
   2757 		} else if (af == AF_INET6) {
   2758 			const struct ip6_hdr *ip6 =
   2759 			    (const struct ip6_hdr *)decrypted_buf;
   2760 			IN6_PRINT(addrstr, &ip6->ip6_src);
   2761 #endif
   2762 		}
   2763 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2764 		    "%s: peer %s: invalid source address (%s)\n",
   2765 		    if_name(&wg->wg_if), wgp->wgp_name, addrstr);
   2766 		m_freem(n);
   2767 		/*
   2768 		 * The inner address is invalid however the session is valid
   2769 		 * so continue the session processing below.
   2770 		 */
   2771 	}
   2772 	n = NULL;
   2773 
   2774 	/* Update the state machine if necessary.  */
   2775 	if (__predict_false(state == WGS_STATE_INIT_PASSIVE)) {
   2776 		/*
   2777 		 * We were waiting for the initiator to send their
   2778 		 * first data transport message, and that has happened.
   2779 		 * Schedule a task to establish this session.
   2780 		 */
   2781 		wg_schedule_peer_task(wgp, WGP_TASK_ESTABLISH_SESSION);
   2782 	} else {
   2783 		if (__predict_false(wg_need_to_send_init_message(wgs))) {
   2784 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   2785 		}
   2786 		/*
   2787 		 * [W] 6.5 Passive Keepalive
   2788 		 * "If a peer has received a validly-authenticated transport
   2789 		 *  data message (section 5.4.6), but does not have any packets
   2790 		 *  itself to send back for KEEPALIVE-TIMEOUT seconds, it sends
   2791 		 *  a keepalive message."
   2792 		 */
   2793 		WG_DLOG("time_uptime=%ju wgs_time_last_data_sent=%ju\n",
   2794 		    (uintmax_t)time_uptime,
   2795 		    (uintmax_t)wgs->wgs_time_last_data_sent);
   2796 		if ((time_uptime - wgs->wgs_time_last_data_sent) >=
   2797 		    wg_keepalive_timeout) {
   2798 			WG_TRACE("Schedule sending keepalive message");
   2799 			/*
   2800 			 * We can't send a keepalive message here to avoid
   2801 			 * a deadlock;  we already hold the solock of a socket
   2802 			 * that is used to send the message.
   2803 			 */
   2804 			wg_schedule_peer_task(wgp,
   2805 			    WGP_TASK_SEND_KEEPALIVE_MESSAGE);
   2806 		}
   2807 	}
   2808 out:
   2809 	wg_put_session(wgs, &psref);
   2810 	m_freem(m);
   2811 	if (free_encrypted_buf)
   2812 		kmem_intr_free(encrypted_buf, encrypted_len);
   2813 }
   2814 
   2815 static void __noinline
   2816 wg_handle_msg_cookie(struct wg_softc *wg, const struct wg_msg_cookie *wgmc)
   2817 {
   2818 	struct wg_session *wgs;
   2819 	struct wg_peer *wgp;
   2820 	struct psref psref;
   2821 	int error;
   2822 	uint8_t key[WG_HASH_LEN];
   2823 	uint8_t cookie[WG_COOKIE_LEN];
   2824 
   2825 	WG_TRACE("cookie msg received");
   2826 
   2827 	/* Find the putative session.  */
   2828 	wgs = wg_lookup_session_by_index(wg, wgmc->wgmc_receiver, &psref);
   2829 	if (wgs == NULL) {
   2830 		WG_TRACE("No session found");
   2831 		return;
   2832 	}
   2833 
   2834 	/* Lock the peer so we can update the cookie state.  */
   2835 	wgp = wgs->wgs_peer;
   2836 	mutex_enter(wgp->wgp_lock);
   2837 
   2838 	if (!wgp->wgp_last_sent_mac1_valid) {
   2839 		WG_TRACE("No valid mac1 sent (or expired)");
   2840 		goto out;
   2841 	}
   2842 
   2843 	/* Decrypt the cookie and store it for later handshake retry.  */
   2844 	wg_algo_mac_cookie(key, sizeof(key), wgp->wgp_pubkey,
   2845 	    sizeof(wgp->wgp_pubkey));
   2846 	error = wg_algo_xaead_dec(cookie, sizeof(cookie), key,
   2847 	    wgmc->wgmc_cookie, sizeof(wgmc->wgmc_cookie),
   2848 	    wgp->wgp_last_sent_mac1, sizeof(wgp->wgp_last_sent_mac1),
   2849 	    wgmc->wgmc_salt);
   2850 	if (error != 0) {
   2851 		WG_LOG_RATECHECK(&wgp->wgp_ppsratecheck, LOG_DEBUG,
   2852 		    "%s: peer %s: wg_algo_aead_dec for cookie failed: "
   2853 		    "error=%d\n", if_name(&wg->wg_if), wgp->wgp_name, error);
   2854 		goto out;
   2855 	}
   2856 	/*
   2857 	 * [W] 6.6: Interaction with Cookie Reply System
   2858 	 * "it should simply store the decrypted cookie value from the cookie
   2859 	 *  reply message, and wait for the expiration of the REKEY-TIMEOUT
   2860 	 *  timer for retrying a handshake initiation message."
   2861 	 */
   2862 	wgp->wgp_latest_cookie_time = time_uptime;
   2863 	memcpy(wgp->wgp_latest_cookie, cookie, sizeof(wgp->wgp_latest_cookie));
   2864 out:
   2865 	mutex_exit(wgp->wgp_lock);
   2866 	wg_put_session(wgs, &psref);
   2867 }
   2868 
   2869 static struct mbuf *
   2870 wg_validate_msg_header(struct wg_softc *wg, struct mbuf *m)
   2871 {
   2872 	struct wg_msg wgm;
   2873 	size_t mbuflen;
   2874 	size_t msglen;
   2875 
   2876 	/*
   2877 	 * Get the mbuf chain length.  It is already guaranteed, by
   2878 	 * wg_overudp_cb, to be large enough for a struct wg_msg.
   2879 	 */
   2880 	mbuflen = m_length(m);
   2881 	KASSERT(mbuflen >= sizeof(struct wg_msg));
   2882 
   2883 	/*
   2884 	 * Copy the message header (32-bit message type) out -- we'll
   2885 	 * worry about contiguity and alignment later.
   2886 	 */
   2887 	m_copydata(m, 0, sizeof(wgm), &wgm);
   2888 	switch (le32toh(wgm.wgm_type)) {
   2889 	case WG_MSG_TYPE_INIT:
   2890 		msglen = sizeof(struct wg_msg_init);
   2891 		break;
   2892 	case WG_MSG_TYPE_RESP:
   2893 		msglen = sizeof(struct wg_msg_resp);
   2894 		break;
   2895 	case WG_MSG_TYPE_COOKIE:
   2896 		msglen = sizeof(struct wg_msg_cookie);
   2897 		break;
   2898 	case WG_MSG_TYPE_DATA:
   2899 		msglen = sizeof(struct wg_msg_data);
   2900 		break;
   2901 	default:
   2902 		WG_LOG_RATECHECK(&wg->wg_ppsratecheck, LOG_DEBUG,
   2903 		    "%s: Unexpected msg type: %u\n", if_name(&wg->wg_if),
   2904 		    le32toh(wgm.wgm_type));
   2905 		goto error;
   2906 	}
   2907 
   2908 	/* Verify the mbuf chain is long enough for this type of message.  */
   2909 	if (__predict_false(mbuflen < msglen)) {
   2910 		WG_DLOG("Invalid msg size: mbuflen=%lu type=%u\n", mbuflen,
   2911 		    le32toh(wgm.wgm_type));
   2912 		goto error;
   2913 	}
   2914 
   2915 	/* Make the message header contiguous if necessary.  */
   2916 	if (__predict_false(m->m_len < msglen)) {
   2917 		m = m_pullup(m, msglen);
   2918 		if (m == NULL)
   2919 			return NULL;
   2920 	}
   2921 
   2922 	return m;
   2923 
   2924 error:
   2925 	m_freem(m);
   2926 	return NULL;
   2927 }
   2928 
   2929 static void
   2930 wg_handle_packet(struct wg_softc *wg, struct mbuf *m,
   2931     const struct sockaddr *src)
   2932 {
   2933 	struct wg_msg *wgm;
   2934 
   2935 	KASSERT(curlwp->l_pflag & LP_BOUND);
   2936 
   2937 	m = wg_validate_msg_header(wg, m);
   2938 	if (__predict_false(m == NULL))
   2939 		return;
   2940 
   2941 	KASSERT(m->m_len >= sizeof(struct wg_msg));
   2942 	wgm = mtod(m, struct wg_msg *);
   2943 	switch (le32toh(wgm->wgm_type)) {
   2944 	case WG_MSG_TYPE_INIT:
   2945 		wg_handle_msg_init(wg, (struct wg_msg_init *)wgm, src);
   2946 		break;
   2947 	case WG_MSG_TYPE_RESP:
   2948 		wg_handle_msg_resp(wg, (struct wg_msg_resp *)wgm, src);
   2949 		break;
   2950 	case WG_MSG_TYPE_COOKIE:
   2951 		wg_handle_msg_cookie(wg, (struct wg_msg_cookie *)wgm);
   2952 		break;
   2953 	case WG_MSG_TYPE_DATA:
   2954 		wg_handle_msg_data(wg, m, src);
   2955 		/* wg_handle_msg_data frees m for us */
   2956 		return;
   2957 	default:
   2958 		panic("invalid message type: %d", le32toh(wgm->wgm_type));
   2959 	}
   2960 
   2961 	m_freem(m);
   2962 }
   2963 
   2964 static void
   2965 wg_receive_packets(struct wg_softc *wg, const int af)
   2966 {
   2967 
   2968 	for (;;) {
   2969 		int error, flags;
   2970 		struct socket *so;
   2971 		struct mbuf *m = NULL;
   2972 		struct uio dummy_uio;
   2973 		struct mbuf *paddr = NULL;
   2974 		struct sockaddr *src;
   2975 
   2976 		so = wg_get_so_by_af(wg, af);
   2977 		flags = MSG_DONTWAIT;
   2978 		dummy_uio.uio_resid = 1000000000;
   2979 
   2980 		error = so->so_receive(so, &paddr, &dummy_uio, &m, NULL,
   2981 		    &flags);
   2982 		if (error || m == NULL) {
   2983 			//if (error == EWOULDBLOCK)
   2984 			return;
   2985 		}
   2986 
   2987 		KASSERT(paddr != NULL);
   2988 		KASSERT(paddr->m_len >= sizeof(struct sockaddr));
   2989 		src = mtod(paddr, struct sockaddr *);
   2990 
   2991 		wg_handle_packet(wg, m, src);
   2992 	}
   2993 }
   2994 
   2995 static void
   2996 wg_get_peer(struct wg_peer *wgp, struct psref *psref)
   2997 {
   2998 
   2999 	psref_acquire(psref, &wgp->wgp_psref, wg_psref_class);
   3000 }
   3001 
   3002 static void
   3003 wg_put_peer(struct wg_peer *wgp, struct psref *psref)
   3004 {
   3005 
   3006 	psref_release(psref, &wgp->wgp_psref, wg_psref_class);
   3007 }
   3008 
   3009 static void
   3010 wg_task_send_init_message(struct wg_softc *wg, struct wg_peer *wgp)
   3011 {
   3012 	struct wg_session *wgs;
   3013 
   3014 	WG_TRACE("WGP_TASK_SEND_INIT_MESSAGE");
   3015 
   3016 	KASSERT(mutex_owned(wgp->wgp_lock));
   3017 
   3018 	if (!atomic_load_acquire(&wgp->wgp_endpoint_available)) {
   3019 		WGLOG(LOG_DEBUG, "%s: No endpoint available\n",
   3020 		    if_name(&wg->wg_if));
   3021 		/* XXX should do something? */
   3022 		return;
   3023 	}
   3024 
   3025 	wgs = wgp->wgp_session_stable;
   3026 	if (wgs->wgs_state == WGS_STATE_UNKNOWN) {
   3027 		/* XXX What if the unstable session is already INIT_ACTIVE?  */
   3028 		wg_send_handshake_msg_init(wg, wgp);
   3029 	} else {
   3030 		/* rekey */
   3031 		wgs = wgp->wgp_session_unstable;
   3032 		if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE)
   3033 			wg_send_handshake_msg_init(wg, wgp);
   3034 	}
   3035 }
   3036 
   3037 static void
   3038 wg_task_retry_handshake(struct wg_softc *wg, struct wg_peer *wgp)
   3039 {
   3040 	struct wg_session *wgs;
   3041 
   3042 	WG_TRACE("WGP_TASK_RETRY_HANDSHAKE");
   3043 
   3044 	KASSERT(mutex_owned(wgp->wgp_lock));
   3045 	KASSERT(wgp->wgp_handshake_start_time != 0);
   3046 
   3047 	wgs = wgp->wgp_session_unstable;
   3048 	if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE)
   3049 		return;
   3050 
   3051 	/*
   3052 	 * XXX no real need to assign a new index here, but we do need
   3053 	 * to transition to UNKNOWN temporarily
   3054 	 */
   3055 	wg_put_session_index(wg, wgs);
   3056 
   3057 	/* [W] 6.4 Handshake Initiation Retransmission */
   3058 	if ((time_uptime - wgp->wgp_handshake_start_time) >
   3059 	    wg_rekey_attempt_time) {
   3060 		/* Give up handshaking */
   3061 		wgp->wgp_handshake_start_time = 0;
   3062 		WG_TRACE("give up");
   3063 
   3064 		/*
   3065 		 * If a new data packet comes, handshaking will be retried
   3066 		 * and a new session would be established at that time,
   3067 		 * however we don't want to send pending packets then.
   3068 		 */
   3069 		wg_purge_pending_packets(wgp);
   3070 		return;
   3071 	}
   3072 
   3073 	wg_task_send_init_message(wg, wgp);
   3074 }
   3075 
   3076 static void
   3077 wg_task_establish_session(struct wg_softc *wg, struct wg_peer *wgp)
   3078 {
   3079 	struct wg_session *wgs, *wgs_prev;
   3080 	struct mbuf *m;
   3081 
   3082 	KASSERT(mutex_owned(wgp->wgp_lock));
   3083 
   3084 	wgs = wgp->wgp_session_unstable;
   3085 	if (wgs->wgs_state != WGS_STATE_INIT_PASSIVE)
   3086 		/* XXX Can this happen?  */
   3087 		return;
   3088 
   3089 	wgs->wgs_state = WGS_STATE_ESTABLISHED;
   3090 	wgs->wgs_time_established = time_uptime;
   3091 	wgs->wgs_time_last_data_sent = 0;
   3092 	wgs->wgs_is_initiator = false;
   3093 	WG_TRACE("WGS_STATE_ESTABLISHED");
   3094 
   3095 	wg_swap_sessions(wgp);
   3096 	KASSERT(wgs == wgp->wgp_session_stable);
   3097 	wgs_prev = wgp->wgp_session_unstable;
   3098 	getnanotime(&wgp->wgp_last_handshake_time);
   3099 	wgp->wgp_handshake_start_time = 0;
   3100 	wgp->wgp_last_sent_mac1_valid = false;
   3101 	wgp->wgp_last_sent_cookie_valid = false;
   3102 
   3103 	/* If we had a data packet queued up, send it.  */
   3104 	if ((m = atomic_swap_ptr(&wgp->wgp_pending, NULL)) != NULL) {
   3105 		kpreempt_disable();
   3106 		const uint32_t h = curcpu()->ci_index; // pktq_rps_hash(m)
   3107 		M_SETCTX(m, wgp);
   3108 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   3109 			WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   3110 			    if_name(&wg->wg_if));
   3111 			m_freem(m);
   3112 		}
   3113 		kpreempt_enable();
   3114 	}
   3115 
   3116 	if (wgs_prev->wgs_state == WGS_STATE_ESTABLISHED) {
   3117 		/* Wait for wg_get_stable_session to drain.  */
   3118 		pserialize_perform(wgp->wgp_psz);
   3119 
   3120 		/* Transition ESTABLISHED->DESTROYING.  */
   3121 		wgs_prev->wgs_state = WGS_STATE_DESTROYING;
   3122 
   3123 		/* We can't destroy the old session immediately */
   3124 		wg_schedule_session_dtor_timer(wgp);
   3125 	} else {
   3126 		KASSERTMSG(wgs_prev->wgs_state == WGS_STATE_UNKNOWN,
   3127 		    "state=%d", wgs_prev->wgs_state);
   3128 		wg_clear_states(wgs_prev);
   3129 		wgs_prev->wgs_state = WGS_STATE_UNKNOWN;
   3130 	}
   3131 }
   3132 
   3133 static void
   3134 wg_task_endpoint_changed(struct wg_softc *wg, struct wg_peer *wgp)
   3135 {
   3136 
   3137 	WG_TRACE("WGP_TASK_ENDPOINT_CHANGED");
   3138 
   3139 	KASSERT(mutex_owned(wgp->wgp_lock));
   3140 
   3141 	if (atomic_load_relaxed(&wgp->wgp_endpoint_changing)) {
   3142 		pserialize_perform(wgp->wgp_psz);
   3143 		mutex_exit(wgp->wgp_lock);
   3144 		psref_target_destroy(&wgp->wgp_endpoint0->wgsa_psref,
   3145 		    wg_psref_class);
   3146 		psref_target_init(&wgp->wgp_endpoint0->wgsa_psref,
   3147 		    wg_psref_class);
   3148 		mutex_enter(wgp->wgp_lock);
   3149 		atomic_store_release(&wgp->wgp_endpoint_changing, 0);
   3150 	}
   3151 }
   3152 
   3153 static void
   3154 wg_task_send_keepalive_message(struct wg_softc *wg, struct wg_peer *wgp)
   3155 {
   3156 	struct wg_session *wgs;
   3157 
   3158 	WG_TRACE("WGP_TASK_SEND_KEEPALIVE_MESSAGE");
   3159 
   3160 	KASSERT(mutex_owned(wgp->wgp_lock));
   3161 
   3162 	wgs = wgp->wgp_session_stable;
   3163 	if (wgs->wgs_state != WGS_STATE_ESTABLISHED)
   3164 		return;
   3165 
   3166 	wg_send_keepalive_msg(wgp, wgs);
   3167 }
   3168 
   3169 static void
   3170 wg_task_destroy_prev_session(struct wg_softc *wg, struct wg_peer *wgp)
   3171 {
   3172 	struct wg_session *wgs;
   3173 
   3174 	WG_TRACE("WGP_TASK_DESTROY_PREV_SESSION");
   3175 
   3176 	KASSERT(mutex_owned(wgp->wgp_lock));
   3177 
   3178 	wgs = wgp->wgp_session_unstable;
   3179 	if (wgs->wgs_state == WGS_STATE_DESTROYING) {
   3180 		wg_put_session_index(wg, wgs);
   3181 	}
   3182 }
   3183 
   3184 static void
   3185 wg_peer_work(struct work *wk, void *cookie)
   3186 {
   3187 	struct wg_peer *wgp = container_of(wk, struct wg_peer, wgp_work);
   3188 	struct wg_softc *wg = wgp->wgp_sc;
   3189 	unsigned int tasks;
   3190 
   3191 	mutex_enter(wgp->wgp_intr_lock);
   3192 	while ((tasks = wgp->wgp_tasks) != 0) {
   3193 		wgp->wgp_tasks = 0;
   3194 		mutex_exit(wgp->wgp_intr_lock);
   3195 
   3196 		mutex_enter(wgp->wgp_lock);
   3197 		if (ISSET(tasks, WGP_TASK_SEND_INIT_MESSAGE))
   3198 			wg_task_send_init_message(wg, wgp);
   3199 		if (ISSET(tasks, WGP_TASK_RETRY_HANDSHAKE))
   3200 			wg_task_retry_handshake(wg, wgp);
   3201 		if (ISSET(tasks, WGP_TASK_ESTABLISH_SESSION))
   3202 			wg_task_establish_session(wg, wgp);
   3203 		if (ISSET(tasks, WGP_TASK_ENDPOINT_CHANGED))
   3204 			wg_task_endpoint_changed(wg, wgp);
   3205 		if (ISSET(tasks, WGP_TASK_SEND_KEEPALIVE_MESSAGE))
   3206 			wg_task_send_keepalive_message(wg, wgp);
   3207 		if (ISSET(tasks, WGP_TASK_DESTROY_PREV_SESSION))
   3208 			wg_task_destroy_prev_session(wg, wgp);
   3209 		mutex_exit(wgp->wgp_lock);
   3210 
   3211 		mutex_enter(wgp->wgp_intr_lock);
   3212 	}
   3213 	mutex_exit(wgp->wgp_intr_lock);
   3214 }
   3215 
   3216 static void
   3217 wg_job(struct threadpool_job *job)
   3218 {
   3219 	struct wg_softc *wg = container_of(job, struct wg_softc, wg_job);
   3220 	int bound, upcalls;
   3221 
   3222 	mutex_enter(wg->wg_intr_lock);
   3223 	while ((upcalls = wg->wg_upcalls) != 0) {
   3224 		wg->wg_upcalls = 0;
   3225 		mutex_exit(wg->wg_intr_lock);
   3226 		bound = curlwp_bind();
   3227 		if (ISSET(upcalls, WG_UPCALL_INET))
   3228 			wg_receive_packets(wg, AF_INET);
   3229 		if (ISSET(upcalls, WG_UPCALL_INET6))
   3230 			wg_receive_packets(wg, AF_INET6);
   3231 		curlwp_bindx(bound);
   3232 		mutex_enter(wg->wg_intr_lock);
   3233 	}
   3234 	threadpool_job_done(job);
   3235 	mutex_exit(wg->wg_intr_lock);
   3236 }
   3237 
   3238 static int
   3239 wg_bind_port(struct wg_softc *wg, const uint16_t port)
   3240 {
   3241 	int error;
   3242 	uint16_t old_port = wg->wg_listen_port;
   3243 
   3244 	if (port != 0 && old_port == port)
   3245 		return 0;
   3246 
   3247 	struct sockaddr_in _sin, *sin = &_sin;
   3248 	sin->sin_len = sizeof(*sin);
   3249 	sin->sin_family = AF_INET;
   3250 	sin->sin_addr.s_addr = INADDR_ANY;
   3251 	sin->sin_port = htons(port);
   3252 
   3253 	error = sobind(wg->wg_so4, sintosa(sin), curlwp);
   3254 	if (error != 0)
   3255 		return error;
   3256 
   3257 #ifdef INET6
   3258 	struct sockaddr_in6 _sin6, *sin6 = &_sin6;
   3259 	sin6->sin6_len = sizeof(*sin6);
   3260 	sin6->sin6_family = AF_INET6;
   3261 	sin6->sin6_addr = in6addr_any;
   3262 	sin6->sin6_port = htons(port);
   3263 
   3264 	error = sobind(wg->wg_so6, sin6tosa(sin6), curlwp);
   3265 	if (error != 0)
   3266 		return error;
   3267 #endif
   3268 
   3269 	wg->wg_listen_port = port;
   3270 
   3271 	return 0;
   3272 }
   3273 
   3274 static void
   3275 wg_so_upcall(struct socket *so, void *cookie, int events, int waitflag)
   3276 {
   3277 	struct wg_softc *wg = cookie;
   3278 	int reason;
   3279 
   3280 	reason = (so->so_proto->pr_domain->dom_family == AF_INET) ?
   3281 	    WG_UPCALL_INET :
   3282 	    WG_UPCALL_INET6;
   3283 
   3284 	mutex_enter(wg->wg_intr_lock);
   3285 	wg->wg_upcalls |= reason;
   3286 	threadpool_schedule_job(wg->wg_threadpool, &wg->wg_job);
   3287 	mutex_exit(wg->wg_intr_lock);
   3288 }
   3289 
   3290 static int
   3291 wg_overudp_cb(struct mbuf **mp, int offset, struct socket *so,
   3292     struct sockaddr *src, void *arg)
   3293 {
   3294 	struct wg_softc *wg = arg;
   3295 	struct wg_msg wgm;
   3296 	struct mbuf *m = *mp;
   3297 
   3298 	WG_TRACE("enter");
   3299 
   3300 	/* Verify the mbuf chain is long enough to have a wg msg header.  */
   3301 	KASSERT(offset <= m_length(m));
   3302 	if (__predict_false(m_length(m) - offset < sizeof(struct wg_msg))) {
   3303 		/* drop on the floor */
   3304 		m_freem(m);
   3305 		return -1;
   3306 	}
   3307 
   3308 	/*
   3309 	 * Copy the message header (32-bit message type) out -- we'll
   3310 	 * worry about contiguity and alignment later.
   3311 	 */
   3312 	m_copydata(m, offset, sizeof(struct wg_msg), &wgm);
   3313 	WG_DLOG("type=%d\n", le32toh(wgm.wgm_type));
   3314 
   3315 	/*
   3316 	 * Handle DATA packets promptly as they arrive.  Other packets
   3317 	 * may require expensive public-key crypto and are not as
   3318 	 * sensitive to latency, so defer them to the worker thread.
   3319 	 */
   3320 	switch (le32toh(wgm.wgm_type)) {
   3321 	case WG_MSG_TYPE_DATA:
   3322 		/* handle immediately */
   3323 		m_adj(m, offset);
   3324 		if (__predict_false(m->m_len < sizeof(struct wg_msg_data))) {
   3325 			m = m_pullup(m, sizeof(struct wg_msg_data));
   3326 			if (m == NULL)
   3327 				return -1;
   3328 		}
   3329 		wg_handle_msg_data(wg, m, src);
   3330 		*mp = NULL;
   3331 		return 1;
   3332 	case WG_MSG_TYPE_INIT:
   3333 	case WG_MSG_TYPE_RESP:
   3334 	case WG_MSG_TYPE_COOKIE:
   3335 		/* pass through to so_receive in wg_receive_packets */
   3336 		return 0;
   3337 	default:
   3338 		/* drop on the floor */
   3339 		m_freem(m);
   3340 		return -1;
   3341 	}
   3342 }
   3343 
   3344 static int
   3345 wg_socreate(struct wg_softc *wg, int af, struct socket **sop)
   3346 {
   3347 	int error;
   3348 	struct socket *so;
   3349 
   3350 	error = socreate(af, &so, SOCK_DGRAM, 0, curlwp, NULL);
   3351 	if (error != 0)
   3352 		return error;
   3353 
   3354 	solock(so);
   3355 	so->so_upcallarg = wg;
   3356 	so->so_upcall = wg_so_upcall;
   3357 	so->so_rcv.sb_flags |= SB_UPCALL;
   3358 	inpcb_register_overudp_cb(sotoinpcb(so), wg_overudp_cb, wg);
   3359 	sounlock(so);
   3360 
   3361 	*sop = so;
   3362 
   3363 	return 0;
   3364 }
   3365 
   3366 static bool
   3367 wg_session_hit_limits(struct wg_session *wgs)
   3368 {
   3369 
   3370 	/*
   3371 	 * [W] 6.2: Transport Message Limits
   3372 	 * "After REJECT-AFTER-MESSAGES transport data messages or after the
   3373 	 *  current secure session is REJECT-AFTER-TIME seconds old, whichever
   3374 	 *  comes first, WireGuard will refuse to send any more transport data
   3375 	 *  messages using the current secure session, ..."
   3376 	 */
   3377 	KASSERT(wgs->wgs_time_established != 0);
   3378 	if ((time_uptime - wgs->wgs_time_established) > wg_reject_after_time) {
   3379 		WG_DLOG("The session hits REJECT_AFTER_TIME\n");
   3380 		return true;
   3381 	} else if (wg_session_get_send_counter(wgs) >
   3382 	    wg_reject_after_messages) {
   3383 		WG_DLOG("The session hits REJECT_AFTER_MESSAGES\n");
   3384 		return true;
   3385 	}
   3386 
   3387 	return false;
   3388 }
   3389 
   3390 static void
   3391 wgintr(void *cookie)
   3392 {
   3393 	struct wg_peer *wgp;
   3394 	struct wg_session *wgs;
   3395 	struct mbuf *m;
   3396 	struct psref psref;
   3397 
   3398 	while ((m = pktq_dequeue(wg_pktq)) != NULL) {
   3399 		wgp = M_GETCTX(m, struct wg_peer *);
   3400 		if ((wgs = wg_get_stable_session(wgp, &psref)) == NULL) {
   3401 			WG_TRACE("no stable session");
   3402 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3403 			goto next0;
   3404 		}
   3405 		if (__predict_false(wg_session_hit_limits(wgs))) {
   3406 			WG_TRACE("stable session hit limits");
   3407 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3408 			goto next1;
   3409 		}
   3410 		wg_send_data_msg(wgp, wgs, m);
   3411 		m = NULL;	/* consumed */
   3412 next1:		wg_put_session(wgs, &psref);
   3413 next0:		m_freem(m);
   3414 		/* XXX Yield to avoid userland starvation?  */
   3415 	}
   3416 }
   3417 
   3418 static void
   3419 wg_rekey_timer(void *arg)
   3420 {
   3421 	struct wg_peer *wgp = arg;
   3422 
   3423 	wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3424 }
   3425 
   3426 static void
   3427 wg_purge_pending_packets(struct wg_peer *wgp)
   3428 {
   3429 	struct mbuf *m;
   3430 
   3431 	m = atomic_swap_ptr(&wgp->wgp_pending, NULL);
   3432 	m_freem(m);
   3433 	pktq_barrier(wg_pktq);
   3434 }
   3435 
   3436 static void
   3437 wg_handshake_timeout_timer(void *arg)
   3438 {
   3439 	struct wg_peer *wgp = arg;
   3440 
   3441 	WG_TRACE("enter");
   3442 
   3443 	wg_schedule_peer_task(wgp, WGP_TASK_RETRY_HANDSHAKE);
   3444 }
   3445 
   3446 static struct wg_peer *
   3447 wg_alloc_peer(struct wg_softc *wg)
   3448 {
   3449 	struct wg_peer *wgp;
   3450 
   3451 	wgp = kmem_zalloc(sizeof(*wgp), KM_SLEEP);
   3452 
   3453 	wgp->wgp_sc = wg;
   3454 	callout_init(&wgp->wgp_rekey_timer, CALLOUT_MPSAFE);
   3455 	callout_setfunc(&wgp->wgp_rekey_timer, wg_rekey_timer, wgp);
   3456 	callout_init(&wgp->wgp_handshake_timeout_timer, CALLOUT_MPSAFE);
   3457 	callout_setfunc(&wgp->wgp_handshake_timeout_timer,
   3458 	    wg_handshake_timeout_timer, wgp);
   3459 	callout_init(&wgp->wgp_session_dtor_timer, CALLOUT_MPSAFE);
   3460 	callout_setfunc(&wgp->wgp_session_dtor_timer,
   3461 	    wg_session_dtor_timer, wgp);
   3462 	PSLIST_ENTRY_INIT(wgp, wgp_peerlist_entry);
   3463 	wgp->wgp_endpoint_changing = false;
   3464 	wgp->wgp_endpoint_available = false;
   3465 	wgp->wgp_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
   3466 	wgp->wgp_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
   3467 	wgp->wgp_psz = pserialize_create();
   3468 	psref_target_init(&wgp->wgp_psref, wg_psref_class);
   3469 
   3470 	wgp->wgp_endpoint = kmem_zalloc(sizeof(*wgp->wgp_endpoint), KM_SLEEP);
   3471 	wgp->wgp_endpoint0 = kmem_zalloc(sizeof(*wgp->wgp_endpoint0), KM_SLEEP);
   3472 	psref_target_init(&wgp->wgp_endpoint->wgsa_psref, wg_psref_class);
   3473 	psref_target_init(&wgp->wgp_endpoint0->wgsa_psref, wg_psref_class);
   3474 
   3475 	struct wg_session *wgs;
   3476 	wgp->wgp_session_stable =
   3477 	    kmem_zalloc(sizeof(*wgp->wgp_session_stable), KM_SLEEP);
   3478 	wgp->wgp_session_unstable =
   3479 	    kmem_zalloc(sizeof(*wgp->wgp_session_unstable), KM_SLEEP);
   3480 	wgs = wgp->wgp_session_stable;
   3481 	wgs->wgs_peer = wgp;
   3482 	wgs->wgs_state = WGS_STATE_UNKNOWN;
   3483 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
   3484 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3485 	mutex_init(&wgs->wgs_send_counter_lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3486 #endif
   3487 	wgs->wgs_recvwin = kmem_zalloc(sizeof(*wgs->wgs_recvwin), KM_SLEEP);
   3488 	mutex_init(&wgs->wgs_recvwin->lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3489 
   3490 	wgs = wgp->wgp_session_unstable;
   3491 	wgs->wgs_peer = wgp;
   3492 	wgs->wgs_state = WGS_STATE_UNKNOWN;
   3493 	psref_target_init(&wgs->wgs_psref, wg_psref_class);
   3494 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3495 	mutex_init(&wgs->wgs_send_counter_lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3496 #endif
   3497 	wgs->wgs_recvwin = kmem_zalloc(sizeof(*wgs->wgs_recvwin), KM_SLEEP);
   3498 	mutex_init(&wgs->wgs_recvwin->lock, MUTEX_DEFAULT, IPL_SOFTNET);
   3499 
   3500 	return wgp;
   3501 }
   3502 
   3503 static void
   3504 wg_destroy_peer(struct wg_peer *wgp)
   3505 {
   3506 	struct wg_session *wgs;
   3507 	struct wg_softc *wg = wgp->wgp_sc;
   3508 
   3509 	/* Prevent new packets from this peer on any source address.  */
   3510 	rw_enter(wg->wg_rwlock, RW_WRITER);
   3511 	for (int i = 0; i < wgp->wgp_n_allowedips; i++) {
   3512 		struct wg_allowedip *wga = &wgp->wgp_allowedips[i];
   3513 		struct radix_node_head *rnh = wg_rnh(wg, wga->wga_family);
   3514 		struct radix_node *rn;
   3515 
   3516 		KASSERT(rnh != NULL);
   3517 		rn = rnh->rnh_deladdr(&wga->wga_sa_addr,
   3518 		    &wga->wga_sa_mask, rnh);
   3519 		if (rn == NULL) {
   3520 			char addrstr[128];
   3521 			sockaddr_format(&wga->wga_sa_addr, addrstr,
   3522 			    sizeof(addrstr));
   3523 			WGLOG(LOG_WARNING, "%s: Couldn't delete %s",
   3524 			    if_name(&wg->wg_if), addrstr);
   3525 		}
   3526 	}
   3527 	rw_exit(wg->wg_rwlock);
   3528 
   3529 	/* Purge pending packets.  */
   3530 	wg_purge_pending_packets(wgp);
   3531 
   3532 	/* Halt all packet processing and timeouts.  */
   3533 	callout_halt(&wgp->wgp_rekey_timer, NULL);
   3534 	callout_halt(&wgp->wgp_handshake_timeout_timer, NULL);
   3535 	callout_halt(&wgp->wgp_session_dtor_timer, NULL);
   3536 
   3537 	/* Wait for any queued work to complete.  */
   3538 	workqueue_wait(wg_wq, &wgp->wgp_work);
   3539 
   3540 	wgs = wgp->wgp_session_unstable;
   3541 	if (wgs->wgs_state != WGS_STATE_UNKNOWN) {
   3542 		mutex_enter(wgp->wgp_lock);
   3543 		wg_destroy_session(wg, wgs);
   3544 		mutex_exit(wgp->wgp_lock);
   3545 	}
   3546 	mutex_destroy(&wgs->wgs_recvwin->lock);
   3547 	kmem_free(wgs->wgs_recvwin, sizeof(*wgs->wgs_recvwin));
   3548 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3549 	mutex_destroy(&wgs->wgs_send_counter_lock);
   3550 #endif
   3551 	kmem_free(wgs, sizeof(*wgs));
   3552 
   3553 	wgs = wgp->wgp_session_stable;
   3554 	if (wgs->wgs_state != WGS_STATE_UNKNOWN) {
   3555 		mutex_enter(wgp->wgp_lock);
   3556 		wg_destroy_session(wg, wgs);
   3557 		mutex_exit(wgp->wgp_lock);
   3558 	}
   3559 	mutex_destroy(&wgs->wgs_recvwin->lock);
   3560 	kmem_free(wgs->wgs_recvwin, sizeof(*wgs->wgs_recvwin));
   3561 #ifndef __HAVE_ATOMIC64_LOADSTORE
   3562 	mutex_destroy(&wgs->wgs_send_counter_lock);
   3563 #endif
   3564 	kmem_free(wgs, sizeof(*wgs));
   3565 
   3566 	psref_target_destroy(&wgp->wgp_endpoint->wgsa_psref, wg_psref_class);
   3567 	psref_target_destroy(&wgp->wgp_endpoint0->wgsa_psref, wg_psref_class);
   3568 	kmem_free(wgp->wgp_endpoint, sizeof(*wgp->wgp_endpoint));
   3569 	kmem_free(wgp->wgp_endpoint0, sizeof(*wgp->wgp_endpoint0));
   3570 
   3571 	pserialize_destroy(wgp->wgp_psz);
   3572 	mutex_obj_free(wgp->wgp_intr_lock);
   3573 	mutex_obj_free(wgp->wgp_lock);
   3574 
   3575 	kmem_free(wgp, sizeof(*wgp));
   3576 }
   3577 
   3578 static void
   3579 wg_destroy_all_peers(struct wg_softc *wg)
   3580 {
   3581 	struct wg_peer *wgp, *wgp0 __diagused;
   3582 	void *garbage_byname, *garbage_bypubkey;
   3583 
   3584 restart:
   3585 	garbage_byname = garbage_bypubkey = NULL;
   3586 	mutex_enter(wg->wg_lock);
   3587 	WG_PEER_WRITER_FOREACH(wgp, wg) {
   3588 		if (wgp->wgp_name[0]) {
   3589 			wgp0 = thmap_del(wg->wg_peers_byname, wgp->wgp_name,
   3590 			    strlen(wgp->wgp_name));
   3591 			KASSERT(wgp0 == wgp);
   3592 			garbage_byname = thmap_stage_gc(wg->wg_peers_byname);
   3593 		}
   3594 		wgp0 = thmap_del(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   3595 		    sizeof(wgp->wgp_pubkey));
   3596 		KASSERT(wgp0 == wgp);
   3597 		garbage_bypubkey = thmap_stage_gc(wg->wg_peers_bypubkey);
   3598 		WG_PEER_WRITER_REMOVE(wgp);
   3599 		wg->wg_npeers--;
   3600 		mutex_enter(wgp->wgp_lock);
   3601 		pserialize_perform(wgp->wgp_psz);
   3602 		mutex_exit(wgp->wgp_lock);
   3603 		PSLIST_ENTRY_DESTROY(wgp, wgp_peerlist_entry);
   3604 		break;
   3605 	}
   3606 	mutex_exit(wg->wg_lock);
   3607 
   3608 	if (wgp == NULL)
   3609 		return;
   3610 
   3611 	psref_target_destroy(&wgp->wgp_psref, wg_psref_class);
   3612 
   3613 	wg_destroy_peer(wgp);
   3614 	thmap_gc(wg->wg_peers_byname, garbage_byname);
   3615 	thmap_gc(wg->wg_peers_bypubkey, garbage_bypubkey);
   3616 
   3617 	goto restart;
   3618 }
   3619 
   3620 static int
   3621 wg_destroy_peer_name(struct wg_softc *wg, const char *name)
   3622 {
   3623 	struct wg_peer *wgp, *wgp0 __diagused;
   3624 	void *garbage_byname, *garbage_bypubkey;
   3625 
   3626 	mutex_enter(wg->wg_lock);
   3627 	wgp = thmap_del(wg->wg_peers_byname, name, strlen(name));
   3628 	if (wgp != NULL) {
   3629 		wgp0 = thmap_del(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   3630 		    sizeof(wgp->wgp_pubkey));
   3631 		KASSERT(wgp0 == wgp);
   3632 		garbage_byname = thmap_stage_gc(wg->wg_peers_byname);
   3633 		garbage_bypubkey = thmap_stage_gc(wg->wg_peers_bypubkey);
   3634 		WG_PEER_WRITER_REMOVE(wgp);
   3635 		wg->wg_npeers--;
   3636 		if (wg->wg_npeers == 0)
   3637 			if_link_state_change(&wg->wg_if, LINK_STATE_DOWN);
   3638 		mutex_enter(wgp->wgp_lock);
   3639 		pserialize_perform(wgp->wgp_psz);
   3640 		mutex_exit(wgp->wgp_lock);
   3641 		PSLIST_ENTRY_DESTROY(wgp, wgp_peerlist_entry);
   3642 	}
   3643 	mutex_exit(wg->wg_lock);
   3644 
   3645 	if (wgp == NULL)
   3646 		return ENOENT;
   3647 
   3648 	psref_target_destroy(&wgp->wgp_psref, wg_psref_class);
   3649 
   3650 	wg_destroy_peer(wgp);
   3651 	thmap_gc(wg->wg_peers_byname, garbage_byname);
   3652 	thmap_gc(wg->wg_peers_bypubkey, garbage_bypubkey);
   3653 
   3654 	return 0;
   3655 }
   3656 
   3657 static int
   3658 wg_if_attach(struct wg_softc *wg)
   3659 {
   3660 
   3661 	wg->wg_if.if_addrlen = 0;
   3662 	wg->wg_if.if_mtu = WG_MTU;
   3663 	wg->wg_if.if_flags = IFF_MULTICAST;
   3664 	wg->wg_if.if_extflags = IFEF_MPSAFE;
   3665 	wg->wg_if.if_ioctl = wg_ioctl;
   3666 	wg->wg_if.if_output = wg_output;
   3667 	wg->wg_if.if_init = wg_init;
   3668 #ifdef ALTQ
   3669 	wg->wg_if.if_start = wg_start;
   3670 #endif
   3671 	wg->wg_if.if_stop = wg_stop;
   3672 	wg->wg_if.if_type = IFT_OTHER;
   3673 	wg->wg_if.if_dlt = DLT_NULL;
   3674 	wg->wg_if.if_softc = wg;
   3675 #ifdef ALTQ
   3676 	IFQ_SET_READY(&wg->wg_if.if_snd);
   3677 #endif
   3678 	if_initialize(&wg->wg_if);
   3679 
   3680 	wg->wg_if.if_link_state = LINK_STATE_DOWN;
   3681 	if_alloc_sadl(&wg->wg_if);
   3682 	if_register(&wg->wg_if);
   3683 
   3684 	bpf_attach(&wg->wg_if, DLT_NULL, sizeof(uint32_t));
   3685 
   3686 	return 0;
   3687 }
   3688 
   3689 static void
   3690 wg_if_detach(struct wg_softc *wg)
   3691 {
   3692 	struct ifnet *ifp = &wg->wg_if;
   3693 
   3694 	bpf_detach(ifp);
   3695 	if_detach(ifp);
   3696 }
   3697 
   3698 static int
   3699 wg_clone_create(struct if_clone *ifc, int unit)
   3700 {
   3701 	struct wg_softc *wg;
   3702 	int error;
   3703 
   3704 	wg_guarantee_initialized();
   3705 
   3706 	error = wg_count_inc();
   3707 	if (error)
   3708 		return error;
   3709 
   3710 	wg = kmem_zalloc(sizeof(*wg), KM_SLEEP);
   3711 
   3712 	if_initname(&wg->wg_if, ifc->ifc_name, unit);
   3713 
   3714 	PSLIST_INIT(&wg->wg_peers);
   3715 	wg->wg_peers_bypubkey = thmap_create(0, NULL, THMAP_NOCOPY);
   3716 	wg->wg_peers_byname = thmap_create(0, NULL, THMAP_NOCOPY);
   3717 	wg->wg_sessions_byindex = thmap_create(0, NULL, THMAP_NOCOPY);
   3718 	wg->wg_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
   3719 	wg->wg_intr_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET);
   3720 	wg->wg_rwlock = rw_obj_alloc();
   3721 	threadpool_job_init(&wg->wg_job, wg_job, wg->wg_intr_lock,
   3722 	    "%s", if_name(&wg->wg_if));
   3723 	wg->wg_ops = &wg_ops_rumpkernel;
   3724 
   3725 	error = threadpool_get(&wg->wg_threadpool, PRI_NONE);
   3726 	if (error)
   3727 		goto fail0;
   3728 
   3729 #ifdef INET
   3730 	error = wg_socreate(wg, AF_INET, &wg->wg_so4);
   3731 	if (error)
   3732 		goto fail1;
   3733 	rn_inithead((void **)&wg->wg_rtable_ipv4,
   3734 	    offsetof(struct sockaddr_in, sin_addr) * NBBY);
   3735 #endif
   3736 #ifdef INET6
   3737 	error = wg_socreate(wg, AF_INET6, &wg->wg_so6);
   3738 	if (error)
   3739 		goto fail2;
   3740 	rn_inithead((void **)&wg->wg_rtable_ipv6,
   3741 	    offsetof(struct sockaddr_in6, sin6_addr) * NBBY);
   3742 #endif
   3743 
   3744 	error = wg_if_attach(wg);
   3745 	if (error)
   3746 		goto fail3;
   3747 
   3748 	return 0;
   3749 
   3750 fail4: __unused
   3751 	wg_if_detach(wg);
   3752 fail3:	wg_destroy_all_peers(wg);
   3753 #ifdef INET6
   3754 	solock(wg->wg_so6);
   3755 	wg->wg_so6->so_rcv.sb_flags &= ~SB_UPCALL;
   3756 	sounlock(wg->wg_so6);
   3757 #endif
   3758 #ifdef INET
   3759 	solock(wg->wg_so4);
   3760 	wg->wg_so4->so_rcv.sb_flags &= ~SB_UPCALL;
   3761 	sounlock(wg->wg_so4);
   3762 #endif
   3763 	mutex_enter(wg->wg_intr_lock);
   3764 	threadpool_cancel_job(wg->wg_threadpool, &wg->wg_job);
   3765 	mutex_exit(wg->wg_intr_lock);
   3766 #ifdef INET6
   3767 	if (wg->wg_rtable_ipv6 != NULL)
   3768 		free(wg->wg_rtable_ipv6, M_RTABLE);
   3769 	soclose(wg->wg_so6);
   3770 fail2:
   3771 #endif
   3772 #ifdef INET
   3773 	if (wg->wg_rtable_ipv4 != NULL)
   3774 		free(wg->wg_rtable_ipv4, M_RTABLE);
   3775 	soclose(wg->wg_so4);
   3776 fail1:
   3777 #endif
   3778 	threadpool_put(wg->wg_threadpool, PRI_NONE);
   3779 fail0:	threadpool_job_destroy(&wg->wg_job);
   3780 	rw_obj_free(wg->wg_rwlock);
   3781 	mutex_obj_free(wg->wg_intr_lock);
   3782 	mutex_obj_free(wg->wg_lock);
   3783 	thmap_destroy(wg->wg_sessions_byindex);
   3784 	thmap_destroy(wg->wg_peers_byname);
   3785 	thmap_destroy(wg->wg_peers_bypubkey);
   3786 	PSLIST_DESTROY(&wg->wg_peers);
   3787 	kmem_free(wg, sizeof(*wg));
   3788 	wg_count_dec();
   3789 	return error;
   3790 }
   3791 
   3792 static int
   3793 wg_clone_destroy(struct ifnet *ifp)
   3794 {
   3795 	struct wg_softc *wg = container_of(ifp, struct wg_softc, wg_if);
   3796 
   3797 #ifdef WG_RUMPKERNEL
   3798 	if (wg_user_mode(wg)) {
   3799 		rumpuser_wg_destroy(wg->wg_user);
   3800 		wg->wg_user = NULL;
   3801 	}
   3802 #endif
   3803 
   3804 	wg_if_detach(wg);
   3805 	wg_destroy_all_peers(wg);
   3806 #ifdef INET6
   3807 	solock(wg->wg_so6);
   3808 	wg->wg_so6->so_rcv.sb_flags &= ~SB_UPCALL;
   3809 	sounlock(wg->wg_so6);
   3810 #endif
   3811 #ifdef INET
   3812 	solock(wg->wg_so4);
   3813 	wg->wg_so4->so_rcv.sb_flags &= ~SB_UPCALL;
   3814 	sounlock(wg->wg_so4);
   3815 #endif
   3816 	mutex_enter(wg->wg_intr_lock);
   3817 	threadpool_cancel_job(wg->wg_threadpool, &wg->wg_job);
   3818 	mutex_exit(wg->wg_intr_lock);
   3819 #ifdef INET6
   3820 	if (wg->wg_rtable_ipv6 != NULL)
   3821 		free(wg->wg_rtable_ipv6, M_RTABLE);
   3822 	soclose(wg->wg_so6);
   3823 #endif
   3824 #ifdef INET
   3825 	if (wg->wg_rtable_ipv4 != NULL)
   3826 		free(wg->wg_rtable_ipv4, M_RTABLE);
   3827 	soclose(wg->wg_so4);
   3828 #endif
   3829 	threadpool_put(wg->wg_threadpool, PRI_NONE);
   3830 	threadpool_job_destroy(&wg->wg_job);
   3831 	rw_obj_free(wg->wg_rwlock);
   3832 	mutex_obj_free(wg->wg_intr_lock);
   3833 	mutex_obj_free(wg->wg_lock);
   3834 	thmap_destroy(wg->wg_sessions_byindex);
   3835 	thmap_destroy(wg->wg_peers_byname);
   3836 	thmap_destroy(wg->wg_peers_bypubkey);
   3837 	PSLIST_DESTROY(&wg->wg_peers);
   3838 	kmem_free(wg, sizeof(*wg));
   3839 	wg_count_dec();
   3840 
   3841 	return 0;
   3842 }
   3843 
   3844 static struct wg_peer *
   3845 wg_pick_peer_by_sa(struct wg_softc *wg, const struct sockaddr *sa,
   3846     struct psref *psref)
   3847 {
   3848 	struct radix_node_head *rnh;
   3849 	struct radix_node *rn;
   3850 	struct wg_peer *wgp = NULL;
   3851 	struct wg_allowedip *wga;
   3852 
   3853 #ifdef WG_DEBUG_LOG
   3854 	char addrstr[128];
   3855 	sockaddr_format(sa, addrstr, sizeof(addrstr));
   3856 	WG_DLOG("sa=%s\n", addrstr);
   3857 #endif
   3858 
   3859 	rw_enter(wg->wg_rwlock, RW_READER);
   3860 
   3861 	rnh = wg_rnh(wg, sa->sa_family);
   3862 	if (rnh == NULL)
   3863 		goto out;
   3864 
   3865 	rn = rnh->rnh_matchaddr(sa, rnh);
   3866 	if (rn == NULL || (rn->rn_flags & RNF_ROOT) != 0)
   3867 		goto out;
   3868 
   3869 	WG_TRACE("success");
   3870 
   3871 	wga = container_of(rn, struct wg_allowedip, wga_nodes[0]);
   3872 	wgp = wga->wga_peer;
   3873 	wg_get_peer(wgp, psref);
   3874 
   3875 out:
   3876 	rw_exit(wg->wg_rwlock);
   3877 	return wgp;
   3878 }
   3879 
   3880 static void
   3881 wg_fill_msg_data(struct wg_softc *wg, struct wg_peer *wgp,
   3882     struct wg_session *wgs, struct wg_msg_data *wgmd)
   3883 {
   3884 
   3885 	memset(wgmd, 0, sizeof(*wgmd));
   3886 	wgmd->wgmd_type = htole32(WG_MSG_TYPE_DATA);
   3887 	wgmd->wgmd_receiver = wgs->wgs_remote_index;
   3888 	/* [W] 5.4.6: msg.counter := Nm^send */
   3889 	/* [W] 5.4.6: Nm^send := Nm^send + 1 */
   3890 	wgmd->wgmd_counter = htole64(wg_session_inc_send_counter(wgs));
   3891 	WG_DLOG("counter=%"PRIu64"\n", le64toh(wgmd->wgmd_counter));
   3892 }
   3893 
   3894 static int
   3895 wg_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
   3896     const struct rtentry *rt)
   3897 {
   3898 	struct wg_softc *wg = ifp->if_softc;
   3899 	struct wg_peer *wgp = NULL;
   3900 	struct wg_session *wgs = NULL;
   3901 	struct psref wgp_psref, wgs_psref;
   3902 	int bound;
   3903 	int error;
   3904 
   3905 	bound = curlwp_bind();
   3906 
   3907 	/* TODO make the nest limit configurable via sysctl */
   3908 	error = if_tunnel_check_nesting(ifp, m, 1);
   3909 	if (error) {
   3910 		WGLOG(LOG_ERR,
   3911 		    "%s: tunneling loop detected and packet dropped\n",
   3912 		    if_name(&wg->wg_if));
   3913 		goto out0;
   3914 	}
   3915 
   3916 #ifdef ALTQ
   3917 	bool altq = atomic_load_relaxed(&ifp->if_snd.altq_flags)
   3918 	    & ALTQF_ENABLED;
   3919 	if (altq)
   3920 		IFQ_CLASSIFY(&ifp->if_snd, m, dst->sa_family);
   3921 #endif
   3922 
   3923 	bpf_mtap_af(ifp, dst->sa_family, m, BPF_D_OUT);
   3924 
   3925 	m->m_flags &= ~(M_BCAST|M_MCAST);
   3926 
   3927 	wgp = wg_pick_peer_by_sa(wg, dst, &wgp_psref);
   3928 	if (wgp == NULL) {
   3929 		WG_TRACE("peer not found");
   3930 		error = EHOSTUNREACH;
   3931 		goto out0;
   3932 	}
   3933 
   3934 	/* Clear checksum-offload flags. */
   3935 	m->m_pkthdr.csum_flags = 0;
   3936 	m->m_pkthdr.csum_data = 0;
   3937 
   3938 	/* Check whether there's an established session.  */
   3939 	wgs = wg_get_stable_session(wgp, &wgs_psref);
   3940 	if (wgs == NULL) {
   3941 		/*
   3942 		 * No established session.  If we're the first to try
   3943 		 * sending data, schedule a handshake and queue the
   3944 		 * packet for when the handshake is done; otherwise
   3945 		 * just drop the packet and let the ongoing handshake
   3946 		 * attempt continue.  We could queue more data packets
   3947 		 * but it's not clear that's worthwhile.
   3948 		 */
   3949 		if (atomic_cas_ptr(&wgp->wgp_pending, NULL, m) == NULL) {
   3950 			m = NULL; /* consume */
   3951 			WG_TRACE("queued first packet; init handshake");
   3952 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   3953 		} else {
   3954 			WG_TRACE("first packet already queued, dropping");
   3955 		}
   3956 		goto out1;
   3957 	}
   3958 
   3959 	/* There's an established session.  Toss it in the queue.  */
   3960 #ifdef ALTQ
   3961 	if (altq) {
   3962 		mutex_enter(ifp->if_snd.ifq_lock);
   3963 		if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
   3964 			M_SETCTX(m, wgp);
   3965 			ALTQ_ENQUEUE(&ifp->if_snd, m, error);
   3966 			m = NULL; /* consume */
   3967 		}
   3968 		mutex_exit(ifp->if_snd.ifq_lock);
   3969 		if (m == NULL) {
   3970 			wg_start(ifp);
   3971 			goto out2;
   3972 		}
   3973 	}
   3974 #endif
   3975 	kpreempt_disable();
   3976 	const uint32_t h = curcpu()->ci_index;	// pktq_rps_hash(m)
   3977 	M_SETCTX(m, wgp);
   3978 	if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   3979 		WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   3980 		    if_name(&wg->wg_if));
   3981 		error = ENOBUFS;
   3982 		goto out3;
   3983 	}
   3984 	m = NULL;		/* consumed */
   3985 	error = 0;
   3986 out3:	kpreempt_enable();
   3987 
   3988 #ifdef ALTQ
   3989 out2:
   3990 #endif
   3991 	wg_put_session(wgs, &wgs_psref);
   3992 out1:	wg_put_peer(wgp, &wgp_psref);
   3993 out0:	m_freem(m);
   3994 	curlwp_bindx(bound);
   3995 	return error;
   3996 }
   3997 
   3998 static int
   3999 wg_send_udp(struct wg_peer *wgp, struct mbuf *m)
   4000 {
   4001 	struct psref psref;
   4002 	struct wg_sockaddr *wgsa;
   4003 	int error;
   4004 	struct socket *so;
   4005 
   4006 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   4007 	so = wg_get_so_by_peer(wgp, wgsa);
   4008 	solock(so);
   4009 	if (wgsatosa(wgsa)->sa_family == AF_INET) {
   4010 		error = udp_send(so, m, wgsatosa(wgsa), NULL, curlwp);
   4011 	} else {
   4012 #ifdef INET6
   4013 		error = udp6_output(sotoinpcb(so), m, wgsatosin6(wgsa),
   4014 		    NULL, curlwp);
   4015 #else
   4016 		m_freem(m);
   4017 		error = EPFNOSUPPORT;
   4018 #endif
   4019 	}
   4020 	sounlock(so);
   4021 	wg_put_sa(wgp, wgsa, &psref);
   4022 
   4023 	return error;
   4024 }
   4025 
   4026 /* Inspired by pppoe_get_mbuf */
   4027 static struct mbuf *
   4028 wg_get_mbuf(size_t leading_len, size_t len)
   4029 {
   4030 	struct mbuf *m;
   4031 
   4032 	KASSERT(leading_len <= MCLBYTES);
   4033 	KASSERT(len <= MCLBYTES - leading_len);
   4034 
   4035 	m = m_gethdr(M_DONTWAIT, MT_DATA);
   4036 	if (m == NULL)
   4037 		return NULL;
   4038 	if (len + leading_len > MHLEN) {
   4039 		m_clget(m, M_DONTWAIT);
   4040 		if ((m->m_flags & M_EXT) == 0) {
   4041 			m_free(m);
   4042 			return NULL;
   4043 		}
   4044 	}
   4045 	m->m_data += leading_len;
   4046 	m->m_pkthdr.len = m->m_len = len;
   4047 
   4048 	return m;
   4049 }
   4050 
   4051 static int
   4052 wg_send_data_msg(struct wg_peer *wgp, struct wg_session *wgs,
   4053     struct mbuf *m)
   4054 {
   4055 	struct wg_softc *wg = wgp->wgp_sc;
   4056 	int error;
   4057 	size_t inner_len, padded_len, encrypted_len;
   4058 	char *padded_buf = NULL;
   4059 	size_t mlen;
   4060 	struct wg_msg_data *wgmd;
   4061 	bool free_padded_buf = false;
   4062 	struct mbuf *n;
   4063 	size_t leading_len = max_hdr + sizeof(struct udphdr);
   4064 
   4065 	mlen = m_length(m);
   4066 	inner_len = mlen;
   4067 	padded_len = roundup(mlen, 16);
   4068 	encrypted_len = padded_len + WG_AUTHTAG_LEN;
   4069 	WG_DLOG("inner=%lu, padded=%lu, encrypted_len=%lu\n",
   4070 	    inner_len, padded_len, encrypted_len);
   4071 	if (mlen != 0) {
   4072 		bool success;
   4073 		success = m_ensure_contig(&m, padded_len);
   4074 		if (success) {
   4075 			padded_buf = mtod(m, char *);
   4076 		} else {
   4077 			padded_buf = kmem_intr_alloc(padded_len, KM_NOSLEEP);
   4078 			if (padded_buf == NULL) {
   4079 				error = ENOBUFS;
   4080 				goto end;
   4081 			}
   4082 			free_padded_buf = true;
   4083 			m_copydata(m, 0, mlen, padded_buf);
   4084 		}
   4085 		memset(padded_buf + mlen, 0, padded_len - inner_len);
   4086 	}
   4087 
   4088 	n = wg_get_mbuf(leading_len, sizeof(*wgmd) + encrypted_len);
   4089 	if (n == NULL) {
   4090 		error = ENOBUFS;
   4091 		goto end;
   4092 	}
   4093 	KASSERT(n->m_len >= sizeof(*wgmd));
   4094 	wgmd = mtod(n, struct wg_msg_data *);
   4095 	wg_fill_msg_data(wg, wgp, wgs, wgmd);
   4096 #ifdef WG_DEBUG_PACKET
   4097 	if (wg_debug & WG_DEBUG_FLAGS_PACKET) {
   4098 		char *hex = gethexdump(wgmd, sizeof(*wgmd));
   4099 		if (hex != NULL) {
   4100 			log(LOG_DEBUG, "wgmd=%p, sizeof(*wgmd)=%zu\n%s\n",
   4101 			    wgmd, sizeof(*wgmd), hex);
   4102 			puthexdump(hex, wgmd, sizeof(*wgmd));
   4103 		}
   4104 		hex = gethexdump(padded_buf, padded_len);
   4105 		if (hex != NULL) {
   4106 			log(LOG_DEBUG, "padded_buf=%p, padded_len=%zu\n%s\n",
   4107 			    padded_buf, padded_len, hex);
   4108 			puthexdump(hex, padded_buf, padded_len);
   4109 		}
   4110 	}
   4111 #endif
   4112 	/* [W] 5.4.6: AEAD(Tm^send, Nm^send, P, e) */
   4113 	wg_algo_aead_enc((char *)wgmd + sizeof(*wgmd), encrypted_len,
   4114 	    wgs->wgs_tkey_send, le64toh(wgmd->wgmd_counter),
   4115 	    padded_buf, padded_len,
   4116 	    NULL, 0);
   4117 
   4118 	error = wg->wg_ops->send_data_msg(wgp, n);
   4119 	if (error == 0) {
   4120 		struct ifnet *ifp = &wg->wg_if;
   4121 		if_statadd(ifp, if_obytes, mlen);
   4122 		if_statinc(ifp, if_opackets);
   4123 		if (wgs->wgs_is_initiator &&
   4124 		    wgs->wgs_time_last_data_sent == 0) {
   4125 			/*
   4126 			 * [W] 6.2 Transport Message Limits
   4127 			 * "if a peer is the initiator of a current secure
   4128 			 *  session, WireGuard will send a handshake initiation
   4129 			 *  message to begin a new secure session if, after
   4130 			 *  transmitting a transport data message, the current
   4131 			 *  secure session is REKEY-AFTER-TIME seconds old,"
   4132 			 */
   4133 			wg_schedule_rekey_timer(wgp);
   4134 		}
   4135 		wgs->wgs_time_last_data_sent = time_uptime;
   4136 		if (wg_session_get_send_counter(wgs) >=
   4137 		    wg_rekey_after_messages) {
   4138 			/*
   4139 			 * [W] 6.2 Transport Message Limits
   4140 			 * "WireGuard will try to create a new session, by
   4141 			 *  sending a handshake initiation message (section
   4142 			 *  5.4.2), after it has sent REKEY-AFTER-MESSAGES
   4143 			 *  transport data messages..."
   4144 			 */
   4145 			wg_schedule_peer_task(wgp, WGP_TASK_SEND_INIT_MESSAGE);
   4146 		}
   4147 	}
   4148 end:
   4149 	m_freem(m);
   4150 	if (free_padded_buf)
   4151 		kmem_intr_free(padded_buf, padded_len);
   4152 	return error;
   4153 }
   4154 
   4155 static void
   4156 wg_input(struct ifnet *ifp, struct mbuf *m, const int af)
   4157 {
   4158 	pktqueue_t *pktq;
   4159 	size_t pktlen;
   4160 
   4161 	KASSERT(af == AF_INET || af == AF_INET6);
   4162 
   4163 	WG_TRACE("");
   4164 
   4165 	m_set_rcvif(m, ifp);
   4166 	pktlen = m->m_pkthdr.len;
   4167 
   4168 	bpf_mtap_af(ifp, af, m, BPF_D_IN);
   4169 
   4170 	switch (af) {
   4171 	case AF_INET:
   4172 		pktq = ip_pktq;
   4173 		break;
   4174 #ifdef INET6
   4175 	case AF_INET6:
   4176 		pktq = ip6_pktq;
   4177 		break;
   4178 #endif
   4179 	default:
   4180 		panic("invalid af=%d", af);
   4181 	}
   4182 
   4183 	kpreempt_disable();
   4184 	const u_int h = curcpu()->ci_index;
   4185 	if (__predict_true(pktq_enqueue(pktq, m, h))) {
   4186 		if_statadd(ifp, if_ibytes, pktlen);
   4187 		if_statinc(ifp, if_ipackets);
   4188 	} else {
   4189 		m_freem(m);
   4190 	}
   4191 	kpreempt_enable();
   4192 }
   4193 
   4194 static void
   4195 wg_calc_pubkey(uint8_t pubkey[WG_STATIC_KEY_LEN],
   4196     const uint8_t privkey[WG_STATIC_KEY_LEN])
   4197 {
   4198 
   4199 	crypto_scalarmult_base(pubkey, privkey);
   4200 }
   4201 
   4202 static int
   4203 wg_rtable_add_route(struct wg_softc *wg, struct wg_allowedip *wga)
   4204 {
   4205 	struct radix_node_head *rnh;
   4206 	struct radix_node *rn;
   4207 	int error = 0;
   4208 
   4209 	rw_enter(wg->wg_rwlock, RW_WRITER);
   4210 	rnh = wg_rnh(wg, wga->wga_family);
   4211 	KASSERT(rnh != NULL);
   4212 	rn = rnh->rnh_addaddr(&wga->wga_sa_addr, &wga->wga_sa_mask, rnh,
   4213 	    wga->wga_nodes);
   4214 	rw_exit(wg->wg_rwlock);
   4215 
   4216 	if (rn == NULL)
   4217 		error = EEXIST;
   4218 
   4219 	return error;
   4220 }
   4221 
   4222 static int
   4223 wg_handle_prop_peer(struct wg_softc *wg, prop_dictionary_t peer,
   4224     struct wg_peer **wgpp)
   4225 {
   4226 	int error = 0;
   4227 	const void *pubkey;
   4228 	size_t pubkey_len;
   4229 	const void *psk;
   4230 	size_t psk_len;
   4231 	const char *name = NULL;
   4232 
   4233 	if (prop_dictionary_get_string(peer, "name", &name)) {
   4234 		if (strlen(name) > WG_PEER_NAME_MAXLEN) {
   4235 			error = EINVAL;
   4236 			goto out;
   4237 		}
   4238 	}
   4239 
   4240 	if (!prop_dictionary_get_data(peer, "public_key",
   4241 		&pubkey, &pubkey_len)) {
   4242 		error = EINVAL;
   4243 		goto out;
   4244 	}
   4245 #ifdef WG_DEBUG_DUMP
   4246         if (wg_debug & WG_DEBUG_FLAGS_DUMP) {
   4247 		char *hex = gethexdump(pubkey, pubkey_len);
   4248 		log(LOG_DEBUG, "pubkey=%p, pubkey_len=%lu\n%s\n",
   4249 		    pubkey, pubkey_len, hex);
   4250 		puthexdump(hex, pubkey, pubkey_len);
   4251 	}
   4252 #endif
   4253 
   4254 	struct wg_peer *wgp = wg_alloc_peer(wg);
   4255 	memcpy(wgp->wgp_pubkey, pubkey, sizeof(wgp->wgp_pubkey));
   4256 	if (name != NULL)
   4257 		strncpy(wgp->wgp_name, name, sizeof(wgp->wgp_name));
   4258 
   4259 	if (prop_dictionary_get_data(peer, "preshared_key", &psk, &psk_len)) {
   4260 		if (psk_len != sizeof(wgp->wgp_psk)) {
   4261 			error = EINVAL;
   4262 			goto out;
   4263 		}
   4264 		memcpy(wgp->wgp_psk, psk, sizeof(wgp->wgp_psk));
   4265 	}
   4266 
   4267 	const void *addr;
   4268 	size_t addr_len;
   4269 	struct wg_sockaddr *wgsa = wgp->wgp_endpoint;
   4270 
   4271 	if (!prop_dictionary_get_data(peer, "endpoint", &addr, &addr_len))
   4272 		goto skip_endpoint;
   4273 	if (addr_len < sizeof(*wgsatosa(wgsa)) ||
   4274 	    addr_len > sizeof(*wgsatoss(wgsa))) {
   4275 		error = EINVAL;
   4276 		goto out;
   4277 	}
   4278 	memcpy(wgsatoss(wgsa), addr, addr_len);
   4279 	switch (wgsa_family(wgsa)) {
   4280 	case AF_INET:
   4281 #ifdef INET6
   4282 	case AF_INET6:
   4283 #endif
   4284 		break;
   4285 	default:
   4286 		error = EPFNOSUPPORT;
   4287 		goto out;
   4288 	}
   4289 	if (addr_len != sockaddr_getsize_by_family(wgsa_family(wgsa))) {
   4290 		error = EINVAL;
   4291 		goto out;
   4292 	}
   4293     {
   4294 	char addrstr[128];
   4295 	sockaddr_format(wgsatosa(wgsa), addrstr, sizeof(addrstr));
   4296 	WG_DLOG("addr=%s\n", addrstr);
   4297     }
   4298 	wgp->wgp_endpoint_available = true;
   4299 
   4300 	prop_array_t allowedips;
   4301 skip_endpoint:
   4302 	allowedips = prop_dictionary_get(peer, "allowedips");
   4303 	if (allowedips == NULL)
   4304 		goto skip;
   4305 
   4306 	prop_object_iterator_t _it = prop_array_iterator(allowedips);
   4307 	prop_dictionary_t prop_allowedip;
   4308 	int j = 0;
   4309 	while ((prop_allowedip = prop_object_iterator_next(_it)) != NULL) {
   4310 		struct wg_allowedip *wga = &wgp->wgp_allowedips[j];
   4311 
   4312 		if (!prop_dictionary_get_int(prop_allowedip, "family",
   4313 			&wga->wga_family))
   4314 			continue;
   4315 		if (!prop_dictionary_get_data(prop_allowedip, "ip",
   4316 			&addr, &addr_len))
   4317 			continue;
   4318 		if (!prop_dictionary_get_uint8(prop_allowedip, "cidr",
   4319 			&wga->wga_cidr))
   4320 			continue;
   4321 
   4322 		switch (wga->wga_family) {
   4323 		case AF_INET: {
   4324 			struct sockaddr_in sin;
   4325 			char addrstr[128];
   4326 			struct in_addr mask;
   4327 			struct sockaddr_in sin_mask;
   4328 
   4329 			if (addr_len != sizeof(struct in_addr))
   4330 				return EINVAL;
   4331 			memcpy(&wga->wga_addr4, addr, addr_len);
   4332 
   4333 			sockaddr_in_init(&sin, (const struct in_addr *)addr,
   4334 			    0);
   4335 			sockaddr_copy(&wga->wga_sa_addr,
   4336 			    sizeof(sin), sintosa(&sin));
   4337 
   4338 			sockaddr_format(sintosa(&sin),
   4339 			    addrstr, sizeof(addrstr));
   4340 			WG_DLOG("addr=%s/%d\n", addrstr, wga->wga_cidr);
   4341 
   4342 			in_len2mask(&mask, wga->wga_cidr);
   4343 			sockaddr_in_init(&sin_mask, &mask, 0);
   4344 			sockaddr_copy(&wga->wga_sa_mask,
   4345 			    sizeof(sin_mask), sintosa(&sin_mask));
   4346 
   4347 			break;
   4348 		    }
   4349 #ifdef INET6
   4350 		case AF_INET6: {
   4351 			struct sockaddr_in6 sin6;
   4352 			char addrstr[128];
   4353 			struct in6_addr mask;
   4354 			struct sockaddr_in6 sin6_mask;
   4355 
   4356 			if (addr_len != sizeof(struct in6_addr))
   4357 				return EINVAL;
   4358 			memcpy(&wga->wga_addr6, addr, addr_len);
   4359 
   4360 			sockaddr_in6_init(&sin6, (const struct in6_addr *)addr,
   4361 			    0, 0, 0);
   4362 			sockaddr_copy(&wga->wga_sa_addr,
   4363 			    sizeof(sin6), sin6tosa(&sin6));
   4364 
   4365 			sockaddr_format(sin6tosa(&sin6),
   4366 			    addrstr, sizeof(addrstr));
   4367 			WG_DLOG("addr=%s/%d\n", addrstr, wga->wga_cidr);
   4368 
   4369 			in6_prefixlen2mask(&mask, wga->wga_cidr);
   4370 			sockaddr_in6_init(&sin6_mask, &mask, 0, 0, 0);
   4371 			sockaddr_copy(&wga->wga_sa_mask,
   4372 			    sizeof(sin6_mask), sin6tosa(&sin6_mask));
   4373 
   4374 			break;
   4375 		    }
   4376 #endif
   4377 		default:
   4378 			error = EINVAL;
   4379 			goto out;
   4380 		}
   4381 		wga->wga_peer = wgp;
   4382 
   4383 		error = wg_rtable_add_route(wg, wga);
   4384 		if (error != 0)
   4385 			goto out;
   4386 
   4387 		j++;
   4388 	}
   4389 	wgp->wgp_n_allowedips = j;
   4390 skip:
   4391 	*wgpp = wgp;
   4392 out:
   4393 	return error;
   4394 }
   4395 
   4396 static int
   4397 wg_alloc_prop_buf(char **_buf, struct ifdrv *ifd)
   4398 {
   4399 	int error;
   4400 	char *buf;
   4401 
   4402 	WG_DLOG("buf=%p, len=%lu\n", ifd->ifd_data, ifd->ifd_len);
   4403 	if (ifd->ifd_len >= WG_MAX_PROPLEN)
   4404 		return E2BIG;
   4405 	buf = kmem_alloc(ifd->ifd_len + 1, KM_SLEEP);
   4406 	error = copyin(ifd->ifd_data, buf, ifd->ifd_len);
   4407 	if (error != 0)
   4408 		return error;
   4409 	buf[ifd->ifd_len] = '\0';
   4410 #ifdef WG_DEBUG_DUMP
   4411 	if (wg_debug & WG_DEBUG_FLAGS_DUMP) {
   4412 		log(LOG_DEBUG, "%.*s\n", (int)MIN(INT_MAX, ifd->ifd_len),
   4413 		    (const char *)buf);
   4414 	}
   4415 #endif
   4416 	*_buf = buf;
   4417 	return 0;
   4418 }
   4419 
   4420 static int
   4421 wg_ioctl_set_private_key(struct wg_softc *wg, struct ifdrv *ifd)
   4422 {
   4423 	int error;
   4424 	prop_dictionary_t prop_dict;
   4425 	char *buf = NULL;
   4426 	const void *privkey;
   4427 	size_t privkey_len;
   4428 
   4429 	error = wg_alloc_prop_buf(&buf, ifd);
   4430 	if (error != 0)
   4431 		return error;
   4432 	error = EINVAL;
   4433 	prop_dict = prop_dictionary_internalize(buf);
   4434 	if (prop_dict == NULL)
   4435 		goto out;
   4436 	if (!prop_dictionary_get_data(prop_dict, "private_key",
   4437 		&privkey, &privkey_len))
   4438 		goto out;
   4439 #ifdef WG_DEBUG_DUMP
   4440 	if (wg_debug & WG_DEBUG_FLAGS_DUMP) {
   4441 		char *hex = gethexdump(privkey, privkey_len);
   4442 		log(LOG_DEBUG, "privkey=%p, privkey_len=%lu\n%s\n",
   4443 		    privkey, privkey_len, hex);
   4444 		puthexdump(hex, privkey, privkey_len);
   4445 	}
   4446 #endif
   4447 	if (privkey_len != WG_STATIC_KEY_LEN)
   4448 		goto out;
   4449 	memcpy(wg->wg_privkey, privkey, WG_STATIC_KEY_LEN);
   4450 	wg_calc_pubkey(wg->wg_pubkey, wg->wg_privkey);
   4451 	error = 0;
   4452 
   4453 out:
   4454 	kmem_free(buf, ifd->ifd_len + 1);
   4455 	return error;
   4456 }
   4457 
   4458 static int
   4459 wg_ioctl_set_listen_port(struct wg_softc *wg, struct ifdrv *ifd)
   4460 {
   4461 	int error;
   4462 	prop_dictionary_t prop_dict;
   4463 	char *buf = NULL;
   4464 	uint16_t port;
   4465 
   4466 	error = wg_alloc_prop_buf(&buf, ifd);
   4467 	if (error != 0)
   4468 		return error;
   4469 	error = EINVAL;
   4470 	prop_dict = prop_dictionary_internalize(buf);
   4471 	if (prop_dict == NULL)
   4472 		goto out;
   4473 	if (!prop_dictionary_get_uint16(prop_dict, "listen_port", &port))
   4474 		goto out;
   4475 
   4476 	error = wg->wg_ops->bind_port(wg, (uint16_t)port);
   4477 
   4478 out:
   4479 	kmem_free(buf, ifd->ifd_len + 1);
   4480 	return error;
   4481 }
   4482 
   4483 static int
   4484 wg_ioctl_add_peer(struct wg_softc *wg, struct ifdrv *ifd)
   4485 {
   4486 	int error;
   4487 	prop_dictionary_t prop_dict;
   4488 	char *buf = NULL;
   4489 	struct wg_peer *wgp = NULL, *wgp0 __diagused;
   4490 
   4491 	error = wg_alloc_prop_buf(&buf, ifd);
   4492 	if (error != 0)
   4493 		return error;
   4494 	error = EINVAL;
   4495 	prop_dict = prop_dictionary_internalize(buf);
   4496 	if (prop_dict == NULL)
   4497 		goto out;
   4498 
   4499 	error = wg_handle_prop_peer(wg, prop_dict, &wgp);
   4500 	if (error != 0)
   4501 		goto out;
   4502 
   4503 	mutex_enter(wg->wg_lock);
   4504 	if (thmap_get(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   4505 		sizeof(wgp->wgp_pubkey)) != NULL ||
   4506 	    (wgp->wgp_name[0] &&
   4507 		thmap_get(wg->wg_peers_byname, wgp->wgp_name,
   4508 		    strlen(wgp->wgp_name)) != NULL)) {
   4509 		mutex_exit(wg->wg_lock);
   4510 		wg_destroy_peer(wgp);
   4511 		error = EEXIST;
   4512 		goto out;
   4513 	}
   4514 	wgp0 = thmap_put(wg->wg_peers_bypubkey, wgp->wgp_pubkey,
   4515 	    sizeof(wgp->wgp_pubkey), wgp);
   4516 	KASSERT(wgp0 == wgp);
   4517 	if (wgp->wgp_name[0]) {
   4518 		wgp0 = thmap_put(wg->wg_peers_byname, wgp->wgp_name,
   4519 		    strlen(wgp->wgp_name), wgp);
   4520 		KASSERT(wgp0 == wgp);
   4521 	}
   4522 	WG_PEER_WRITER_INSERT_HEAD(wgp, wg);
   4523 	wg->wg_npeers++;
   4524 	mutex_exit(wg->wg_lock);
   4525 
   4526 	if_link_state_change(&wg->wg_if, LINK_STATE_UP);
   4527 
   4528 out:
   4529 	kmem_free(buf, ifd->ifd_len + 1);
   4530 	return error;
   4531 }
   4532 
   4533 static int
   4534 wg_ioctl_delete_peer(struct wg_softc *wg, struct ifdrv *ifd)
   4535 {
   4536 	int error;
   4537 	prop_dictionary_t prop_dict;
   4538 	char *buf = NULL;
   4539 	const char *name;
   4540 
   4541 	error = wg_alloc_prop_buf(&buf, ifd);
   4542 	if (error != 0)
   4543 		return error;
   4544 	error = EINVAL;
   4545 	prop_dict = prop_dictionary_internalize(buf);
   4546 	if (prop_dict == NULL)
   4547 		goto out;
   4548 
   4549 	if (!prop_dictionary_get_string(prop_dict, "name", &name))
   4550 		goto out;
   4551 	if (strlen(name) > WG_PEER_NAME_MAXLEN)
   4552 		goto out;
   4553 
   4554 	error = wg_destroy_peer_name(wg, name);
   4555 out:
   4556 	kmem_free(buf, ifd->ifd_len + 1);
   4557 	return error;
   4558 }
   4559 
   4560 static bool
   4561 wg_is_authorized(struct wg_softc *wg, u_long cmd)
   4562 {
   4563 	int au = cmd == SIOCGDRVSPEC ?
   4564 	    KAUTH_REQ_NETWORK_INTERFACE_WG_GETPRIV :
   4565 	    KAUTH_REQ_NETWORK_INTERFACE_WG_SETPRIV;
   4566 	return kauth_authorize_network(kauth_cred_get(),
   4567 	    KAUTH_NETWORK_INTERFACE_WG, au, &wg->wg_if,
   4568 	    (void *)cmd, NULL) == 0;
   4569 }
   4570 
   4571 static int
   4572 wg_ioctl_get(struct wg_softc *wg, struct ifdrv *ifd)
   4573 {
   4574 	int error = ENOMEM;
   4575 	prop_dictionary_t prop_dict;
   4576 	prop_array_t peers = NULL;
   4577 	char *buf;
   4578 	struct wg_peer *wgp;
   4579 	int s, i;
   4580 
   4581 	prop_dict = prop_dictionary_create();
   4582 	if (prop_dict == NULL)
   4583 		goto error;
   4584 
   4585 	if (wg_is_authorized(wg, SIOCGDRVSPEC)) {
   4586 		if (!prop_dictionary_set_data(prop_dict, "private_key",
   4587 			wg->wg_privkey, WG_STATIC_KEY_LEN))
   4588 			goto error;
   4589 	}
   4590 
   4591 	if (wg->wg_listen_port != 0) {
   4592 		if (!prop_dictionary_set_uint16(prop_dict, "listen_port",
   4593 			wg->wg_listen_port))
   4594 			goto error;
   4595 	}
   4596 
   4597 	if (wg->wg_npeers == 0)
   4598 		goto skip_peers;
   4599 
   4600 	peers = prop_array_create();
   4601 	if (peers == NULL)
   4602 		goto error;
   4603 
   4604 	s = pserialize_read_enter();
   4605 	i = 0;
   4606 	WG_PEER_READER_FOREACH(wgp, wg) {
   4607 		struct wg_sockaddr *wgsa;
   4608 		struct psref wgp_psref, wgsa_psref;
   4609 		prop_dictionary_t prop_peer;
   4610 
   4611 		wg_get_peer(wgp, &wgp_psref);
   4612 		pserialize_read_exit(s);
   4613 
   4614 		prop_peer = prop_dictionary_create();
   4615 		if (prop_peer == NULL)
   4616 			goto next;
   4617 
   4618 		if (strlen(wgp->wgp_name) > 0) {
   4619 			if (!prop_dictionary_set_string(prop_peer, "name",
   4620 				wgp->wgp_name))
   4621 				goto next;
   4622 		}
   4623 
   4624 		if (!prop_dictionary_set_data(prop_peer, "public_key",
   4625 			wgp->wgp_pubkey, sizeof(wgp->wgp_pubkey)))
   4626 			goto next;
   4627 
   4628 		uint8_t psk_zero[WG_PRESHARED_KEY_LEN] = {0};
   4629 		if (!consttime_memequal(wgp->wgp_psk, psk_zero,
   4630 			sizeof(wgp->wgp_psk))) {
   4631 			if (wg_is_authorized(wg, SIOCGDRVSPEC)) {
   4632 				if (!prop_dictionary_set_data(prop_peer,
   4633 					"preshared_key",
   4634 					wgp->wgp_psk, sizeof(wgp->wgp_psk)))
   4635 					goto next;
   4636 			}
   4637 		}
   4638 
   4639 		wgsa = wg_get_endpoint_sa(wgp, &wgsa_psref);
   4640 		CTASSERT(AF_UNSPEC == 0);
   4641 		if (wgsa_family(wgsa) != 0 /*AF_UNSPEC*/ &&
   4642 		    !prop_dictionary_set_data(prop_peer, "endpoint",
   4643 			wgsatoss(wgsa),
   4644 			sockaddr_getsize_by_family(wgsa_family(wgsa)))) {
   4645 			wg_put_sa(wgp, wgsa, &wgsa_psref);
   4646 			goto next;
   4647 		}
   4648 		wg_put_sa(wgp, wgsa, &wgsa_psref);
   4649 
   4650 		const struct timespec *t = &wgp->wgp_last_handshake_time;
   4651 
   4652 		if (!prop_dictionary_set_uint64(prop_peer,
   4653 			"last_handshake_time_sec", (uint64_t)t->tv_sec))
   4654 			goto next;
   4655 		if (!prop_dictionary_set_uint32(prop_peer,
   4656 			"last_handshake_time_nsec", (uint32_t)t->tv_nsec))
   4657 			goto next;
   4658 
   4659 		if (wgp->wgp_n_allowedips == 0)
   4660 			goto skip_allowedips;
   4661 
   4662 		prop_array_t allowedips = prop_array_create();
   4663 		if (allowedips == NULL)
   4664 			goto next;
   4665 		for (int j = 0; j < wgp->wgp_n_allowedips; j++) {
   4666 			struct wg_allowedip *wga = &wgp->wgp_allowedips[j];
   4667 			prop_dictionary_t prop_allowedip;
   4668 
   4669 			prop_allowedip = prop_dictionary_create();
   4670 			if (prop_allowedip == NULL)
   4671 				break;
   4672 
   4673 			if (!prop_dictionary_set_int(prop_allowedip, "family",
   4674 				wga->wga_family))
   4675 				goto _next;
   4676 			if (!prop_dictionary_set_uint8(prop_allowedip, "cidr",
   4677 				wga->wga_cidr))
   4678 				goto _next;
   4679 
   4680 			switch (wga->wga_family) {
   4681 			case AF_INET:
   4682 				if (!prop_dictionary_set_data(prop_allowedip,
   4683 					"ip", &wga->wga_addr4,
   4684 					sizeof(wga->wga_addr4)))
   4685 					goto _next;
   4686 				break;
   4687 #ifdef INET6
   4688 			case AF_INET6:
   4689 				if (!prop_dictionary_set_data(prop_allowedip,
   4690 					"ip", &wga->wga_addr6,
   4691 					sizeof(wga->wga_addr6)))
   4692 					goto _next;
   4693 				break;
   4694 #endif
   4695 			default:
   4696 				break;
   4697 			}
   4698 			prop_array_set(allowedips, j, prop_allowedip);
   4699 		_next:
   4700 			prop_object_release(prop_allowedip);
   4701 		}
   4702 		prop_dictionary_set(prop_peer, "allowedips", allowedips);
   4703 		prop_object_release(allowedips);
   4704 
   4705 	skip_allowedips:
   4706 
   4707 		prop_array_set(peers, i, prop_peer);
   4708 	next:
   4709 		if (prop_peer)
   4710 			prop_object_release(prop_peer);
   4711 		i++;
   4712 
   4713 		s = pserialize_read_enter();
   4714 		wg_put_peer(wgp, &wgp_psref);
   4715 	}
   4716 	pserialize_read_exit(s);
   4717 
   4718 	prop_dictionary_set(prop_dict, "peers", peers);
   4719 	prop_object_release(peers);
   4720 	peers = NULL;
   4721 
   4722 skip_peers:
   4723 	buf = prop_dictionary_externalize(prop_dict);
   4724 	if (buf == NULL)
   4725 		goto error;
   4726 	if (ifd->ifd_len < (strlen(buf) + 1)) {
   4727 		error = EINVAL;
   4728 		goto error;
   4729 	}
   4730 	error = copyout(buf, ifd->ifd_data, strlen(buf) + 1);
   4731 
   4732 	free(buf, 0);
   4733 error:
   4734 	if (peers != NULL)
   4735 		prop_object_release(peers);
   4736 	if (prop_dict != NULL)
   4737 		prop_object_release(prop_dict);
   4738 
   4739 	return error;
   4740 }
   4741 
   4742 static int
   4743 wg_ioctl(struct ifnet *ifp, u_long cmd, void *data)
   4744 {
   4745 	struct wg_softc *wg = ifp->if_softc;
   4746 	struct ifreq *ifr = data;
   4747 	struct ifaddr *ifa = data;
   4748 	struct ifdrv *ifd = data;
   4749 	int error = 0;
   4750 
   4751 	switch (cmd) {
   4752 	case SIOCINITIFADDR:
   4753 		if (ifa->ifa_addr->sa_family != AF_LINK &&
   4754 		    (ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
   4755 		    (IFF_UP | IFF_RUNNING)) {
   4756 			ifp->if_flags |= IFF_UP;
   4757 			error = if_init(ifp);
   4758 		}
   4759 		return error;
   4760 	case SIOCADDMULTI:
   4761 	case SIOCDELMULTI:
   4762 		switch (ifr->ifr_addr.sa_family) {
   4763 		case AF_INET:	/* IP supports Multicast */
   4764 			break;
   4765 #ifdef INET6
   4766 		case AF_INET6:	/* IP6 supports Multicast */
   4767 			break;
   4768 #endif
   4769 		default:  /* Other protocols doesn't support Multicast */
   4770 			error = EAFNOSUPPORT;
   4771 			break;
   4772 		}
   4773 		return error;
   4774 	case SIOCSDRVSPEC:
   4775 		if (!wg_is_authorized(wg, cmd)) {
   4776 			return EPERM;
   4777 		}
   4778 		switch (ifd->ifd_cmd) {
   4779 		case WG_IOCTL_SET_PRIVATE_KEY:
   4780 			error = wg_ioctl_set_private_key(wg, ifd);
   4781 			break;
   4782 		case WG_IOCTL_SET_LISTEN_PORT:
   4783 			error = wg_ioctl_set_listen_port(wg, ifd);
   4784 			break;
   4785 		case WG_IOCTL_ADD_PEER:
   4786 			error = wg_ioctl_add_peer(wg, ifd);
   4787 			break;
   4788 		case WG_IOCTL_DELETE_PEER:
   4789 			error = wg_ioctl_delete_peer(wg, ifd);
   4790 			break;
   4791 		default:
   4792 			error = EINVAL;
   4793 			break;
   4794 		}
   4795 		return error;
   4796 	case SIOCGDRVSPEC:
   4797 		return wg_ioctl_get(wg, ifd);
   4798 	case SIOCSIFFLAGS:
   4799 		if ((error = ifioctl_common(ifp, cmd, data)) != 0)
   4800 			break;
   4801 		switch (ifp->if_flags & (IFF_UP|IFF_RUNNING)) {
   4802 		case IFF_RUNNING:
   4803 			/*
   4804 			 * If interface is marked down and it is running,
   4805 			 * then stop and disable it.
   4806 			 */
   4807 			if_stop(ifp, 1);
   4808 			break;
   4809 		case IFF_UP:
   4810 			/*
   4811 			 * If interface is marked up and it is stopped, then
   4812 			 * start it.
   4813 			 */
   4814 			error = if_init(ifp);
   4815 			break;
   4816 		default:
   4817 			break;
   4818 		}
   4819 		return error;
   4820 #ifdef WG_RUMPKERNEL
   4821 	case SIOCSLINKSTR:
   4822 		error = wg_ioctl_linkstr(wg, ifd);
   4823 		if (error == 0)
   4824 			wg->wg_ops = &wg_ops_rumpuser;
   4825 		return error;
   4826 #endif
   4827 	default:
   4828 		break;
   4829 	}
   4830 
   4831 	error = ifioctl_common(ifp, cmd, data);
   4832 
   4833 #ifdef WG_RUMPKERNEL
   4834 	if (!wg_user_mode(wg))
   4835 		return error;
   4836 
   4837 	/* Do the same to the corresponding tun device on the host */
   4838 	/*
   4839 	 * XXX Actually the command has not been handled yet.  It
   4840 	 *     will be handled via pr_ioctl form doifioctl later.
   4841 	 */
   4842 	switch (cmd) {
   4843 	case SIOCAIFADDR:
   4844 	case SIOCDIFADDR: {
   4845 		struct in_aliasreq _ifra = *(const struct in_aliasreq *)data;
   4846 		struct in_aliasreq *ifra = &_ifra;
   4847 		KASSERT(error == ENOTTY);
   4848 		strncpy(ifra->ifra_name, rumpuser_wg_get_tunname(wg->wg_user),
   4849 		    IFNAMSIZ);
   4850 		error = rumpuser_wg_ioctl(wg->wg_user, cmd, ifra, AF_INET);
   4851 		if (error == 0)
   4852 			error = ENOTTY;
   4853 		break;
   4854 	}
   4855 #ifdef INET6
   4856 	case SIOCAIFADDR_IN6:
   4857 	case SIOCDIFADDR_IN6: {
   4858 		struct in6_aliasreq _ifra = *(const struct in6_aliasreq *)data;
   4859 		struct in6_aliasreq *ifra = &_ifra;
   4860 		KASSERT(error == ENOTTY);
   4861 		strncpy(ifra->ifra_name, rumpuser_wg_get_tunname(wg->wg_user),
   4862 		    IFNAMSIZ);
   4863 		error = rumpuser_wg_ioctl(wg->wg_user, cmd, ifra, AF_INET6);
   4864 		if (error == 0)
   4865 			error = ENOTTY;
   4866 		break;
   4867 	}
   4868 #endif
   4869 	}
   4870 #endif /* WG_RUMPKERNEL */
   4871 
   4872 	return error;
   4873 }
   4874 
   4875 static int
   4876 wg_init(struct ifnet *ifp)
   4877 {
   4878 
   4879 	ifp->if_flags |= IFF_RUNNING;
   4880 
   4881 	/* TODO flush pending packets. */
   4882 	return 0;
   4883 }
   4884 
   4885 #ifdef ALTQ
   4886 static void
   4887 wg_start(struct ifnet *ifp)
   4888 {
   4889 	struct mbuf *m;
   4890 
   4891 	for (;;) {
   4892 		IFQ_DEQUEUE(&ifp->if_snd, m);
   4893 		if (m == NULL)
   4894 			break;
   4895 
   4896 		kpreempt_disable();
   4897 		const uint32_t h = curcpu()->ci_index;	// pktq_rps_hash(m)
   4898 		if (__predict_false(!pktq_enqueue(wg_pktq, m, h))) {
   4899 			WGLOG(LOG_ERR, "%s: pktq full, dropping\n",
   4900 			    if_name(ifp));
   4901 			m_freem(m);
   4902 		}
   4903 		kpreempt_enable();
   4904 	}
   4905 }
   4906 #endif
   4907 
   4908 static void
   4909 wg_stop(struct ifnet *ifp, int disable)
   4910 {
   4911 
   4912 	KASSERT((ifp->if_flags & IFF_RUNNING) != 0);
   4913 	ifp->if_flags &= ~IFF_RUNNING;
   4914 
   4915 	/* Need to do something? */
   4916 }
   4917 
   4918 #ifdef WG_DEBUG_PARAMS
   4919 SYSCTL_SETUP(sysctl_net_wg_setup, "sysctl net.wg setup")
   4920 {
   4921 	const struct sysctlnode *node = NULL;
   4922 
   4923 	sysctl_createv(clog, 0, NULL, &node,
   4924 	    CTLFLAG_PERMANENT,
   4925 	    CTLTYPE_NODE, "wg",
   4926 	    SYSCTL_DESCR("wg(4)"),
   4927 	    NULL, 0, NULL, 0,
   4928 	    CTL_NET, CTL_CREATE, CTL_EOL);
   4929 	sysctl_createv(clog, 0, &node, NULL,
   4930 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4931 	    CTLTYPE_QUAD, "rekey_after_messages",
   4932 	    SYSCTL_DESCR("session liftime by messages"),
   4933 	    NULL, 0, &wg_rekey_after_messages, 0, CTL_CREATE, CTL_EOL);
   4934 	sysctl_createv(clog, 0, &node, NULL,
   4935 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4936 	    CTLTYPE_INT, "rekey_after_time",
   4937 	    SYSCTL_DESCR("session liftime"),
   4938 	    NULL, 0, &wg_rekey_after_time, 0, CTL_CREATE, CTL_EOL);
   4939 	sysctl_createv(clog, 0, &node, NULL,
   4940 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4941 	    CTLTYPE_INT, "rekey_timeout",
   4942 	    SYSCTL_DESCR("session handshake retry time"),
   4943 	    NULL, 0, &wg_rekey_timeout, 0, CTL_CREATE, CTL_EOL);
   4944 	sysctl_createv(clog, 0, &node, NULL,
   4945 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4946 	    CTLTYPE_INT, "rekey_attempt_time",
   4947 	    SYSCTL_DESCR("session handshake timeout"),
   4948 	    NULL, 0, &wg_rekey_attempt_time, 0, CTL_CREATE, CTL_EOL);
   4949 	sysctl_createv(clog, 0, &node, NULL,
   4950 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4951 	    CTLTYPE_INT, "keepalive_timeout",
   4952 	    SYSCTL_DESCR("keepalive timeout"),
   4953 	    NULL, 0, &wg_keepalive_timeout, 0, CTL_CREATE, CTL_EOL);
   4954 	sysctl_createv(clog, 0, &node, NULL,
   4955 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4956 	    CTLTYPE_BOOL, "force_underload",
   4957 	    SYSCTL_DESCR("force to detemine under load"),
   4958 	    NULL, 0, &wg_force_underload, 0, CTL_CREATE, CTL_EOL);
   4959 	sysctl_createv(clog, 0, &node, NULL,
   4960 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   4961 	    CTLTYPE_INT, "debug",
   4962 	    SYSCTL_DESCR("set debug flags 1=log 2=trace 4=dump 8=packet"),
   4963 	    NULL, 0, &wg_debug, 0, CTL_CREATE, CTL_EOL);
   4964 }
   4965 #endif
   4966 
   4967 #ifdef WG_RUMPKERNEL
   4968 static bool
   4969 wg_user_mode(struct wg_softc *wg)
   4970 {
   4971 
   4972 	return wg->wg_user != NULL;
   4973 }
   4974 
   4975 static int
   4976 wg_ioctl_linkstr(struct wg_softc *wg, struct ifdrv *ifd)
   4977 {
   4978 	struct ifnet *ifp = &wg->wg_if;
   4979 	int error;
   4980 
   4981 	if (ifp->if_flags & IFF_UP)
   4982 		return EBUSY;
   4983 
   4984 	if (ifd->ifd_cmd == IFLINKSTR_UNSET) {
   4985 		/* XXX do nothing */
   4986 		return 0;
   4987 	} else if (ifd->ifd_cmd != 0) {
   4988 		return EINVAL;
   4989 	} else if (wg->wg_user != NULL) {
   4990 		return EBUSY;
   4991 	}
   4992 
   4993 	/* Assume \0 included */
   4994 	if (ifd->ifd_len > IFNAMSIZ) {
   4995 		return E2BIG;
   4996 	} else if (ifd->ifd_len < 1) {
   4997 		return EINVAL;
   4998 	}
   4999 
   5000 	char tun_name[IFNAMSIZ];
   5001 	error = copyinstr(ifd->ifd_data, tun_name, ifd->ifd_len, NULL);
   5002 	if (error != 0)
   5003 		return error;
   5004 
   5005 	if (strncmp(tun_name, "tun", 3) != 0)
   5006 		return EINVAL;
   5007 
   5008 	error = rumpuser_wg_create(tun_name, wg, &wg->wg_user);
   5009 
   5010 	return error;
   5011 }
   5012 
   5013 static int
   5014 wg_send_user(struct wg_peer *wgp, struct mbuf *m)
   5015 {
   5016 	int error;
   5017 	struct psref psref;
   5018 	struct wg_sockaddr *wgsa;
   5019 	struct wg_softc *wg = wgp->wgp_sc;
   5020 	struct iovec iov[1];
   5021 
   5022 	wgsa = wg_get_endpoint_sa(wgp, &psref);
   5023 
   5024 	iov[0].iov_base = mtod(m, void *);
   5025 	iov[0].iov_len = m->m_len;
   5026 
   5027 	/* Send messages to a peer via an ordinary socket. */
   5028 	error = rumpuser_wg_send_peer(wg->wg_user, wgsatosa(wgsa), iov, 1);
   5029 
   5030 	wg_put_sa(wgp, wgsa, &psref);
   5031 
   5032 	m_freem(m);
   5033 
   5034 	return error;
   5035 }
   5036 
   5037 static void
   5038 wg_input_user(struct ifnet *ifp, struct mbuf *m, const int af)
   5039 {
   5040 	struct wg_softc *wg = ifp->if_softc;
   5041 	struct iovec iov[2];
   5042 	struct sockaddr_storage ss;
   5043 
   5044 	KASSERT(af == AF_INET || af == AF_INET6);
   5045 
   5046 	WG_TRACE("");
   5047 
   5048 	if (af == AF_INET) {
   5049 		struct sockaddr_in *sin = (struct sockaddr_in *)&ss;
   5050 		struct ip *ip;
   5051 
   5052 		KASSERT(m->m_len >= sizeof(struct ip));
   5053 		ip = mtod(m, struct ip *);
   5054 		sockaddr_in_init(sin, &ip->ip_dst, 0);
   5055 	} else {
   5056 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ss;
   5057 		struct ip6_hdr *ip6;
   5058 
   5059 		KASSERT(m->m_len >= sizeof(struct ip6_hdr));
   5060 		ip6 = mtod(m, struct ip6_hdr *);
   5061 		sockaddr_in6_init(sin6, &ip6->ip6_dst, 0, 0, 0);
   5062 	}
   5063 
   5064 	iov[0].iov_base = &ss;
   5065 	iov[0].iov_len = ss.ss_len;
   5066 	iov[1].iov_base = mtod(m, void *);
   5067 	iov[1].iov_len = m->m_len;
   5068 
   5069 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
   5070 
   5071 	/* Send decrypted packets to users via a tun. */
   5072 	rumpuser_wg_send_user(wg->wg_user, iov, 2);
   5073 
   5074 	m_freem(m);
   5075 }
   5076 
   5077 static int
   5078 wg_bind_port_user(struct wg_softc *wg, const uint16_t port)
   5079 {
   5080 	int error;
   5081 	uint16_t old_port = wg->wg_listen_port;
   5082 
   5083 	if (port != 0 && old_port == port)
   5084 		return 0;
   5085 
   5086 	error = rumpuser_wg_sock_bind(wg->wg_user, port);
   5087 	if (error == 0)
   5088 		wg->wg_listen_port = port;
   5089 	return error;
   5090 }
   5091 
   5092 /*
   5093  * Receive user packets.
   5094  */
   5095 void
   5096 rumpkern_wg_recv_user(struct wg_softc *wg, struct iovec *iov, size_t iovlen)
   5097 {
   5098 	struct ifnet *ifp = &wg->wg_if;
   5099 	struct mbuf *m;
   5100 	const struct sockaddr *dst;
   5101 
   5102 	WG_TRACE("");
   5103 
   5104 	dst = iov[0].iov_base;
   5105 
   5106 	m = m_gethdr(M_DONTWAIT, MT_DATA);
   5107 	if (m == NULL)
   5108 		return;
   5109 	m->m_len = m->m_pkthdr.len = 0;
   5110 	m_copyback(m, 0, iov[1].iov_len, iov[1].iov_base);
   5111 
   5112 	WG_DLOG("iov_len=%lu\n", iov[1].iov_len);
   5113 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
   5114 
   5115 	(void)wg_output(ifp, m, dst, NULL);
   5116 }
   5117 
   5118 /*
   5119  * Receive packets from a peer.
   5120  */
   5121 void
   5122 rumpkern_wg_recv_peer(struct wg_softc *wg, struct iovec *iov, size_t iovlen)
   5123 {
   5124 	struct mbuf *m;
   5125 	const struct sockaddr *src;
   5126 	int bound;
   5127 
   5128 	WG_TRACE("");
   5129 
   5130 	src = 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=%lu\n", iov[1].iov_len);
   5139 	WG_DUMP_BUF(iov[1].iov_base, iov[1].iov_len);
   5140 
   5141 	bound = curlwp_bind();
   5142 	wg_handle_packet(wg, m, src);
   5143 	curlwp_bindx(bound);
   5144 }
   5145 #endif /* WG_RUMPKERNEL */
   5146 
   5147 /*
   5148  * Module infrastructure
   5149  */
   5150 #include "if_module.h"
   5151 
   5152 IF_MODULE(MODULE_CLASS_DRIVER, wg, "sodium,blake2s")
   5153