xref: /freebsd/sbin/hastd/primary.c (revision f5f47d5068fb97df18eb114a66ae8ef51a0b3c8c)
1 /*-
2  * Copyright (c) 2009 The FreeBSD Foundation
3  * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net>
4  * All rights reserved.
5  *
6  * This software was developed by Pawel Jakub Dawidek under sponsorship from
7  * the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/types.h>
35 #include <sys/time.h>
36 #include <sys/bio.h>
37 #include <sys/disk.h>
38 #include <sys/refcount.h>
39 #include <sys/stat.h>
40 
41 #include <geom/gate/g_gate.h>
42 
43 #include <err.h>
44 #include <errno.h>
45 #include <fcntl.h>
46 #include <libgeom.h>
47 #include <pthread.h>
48 #include <signal.h>
49 #include <stdint.h>
50 #include <stdio.h>
51 #include <string.h>
52 #include <sysexits.h>
53 #include <unistd.h>
54 
55 #include <activemap.h>
56 #include <nv.h>
57 #include <rangelock.h>
58 
59 #include "control.h"
60 #include "event.h"
61 #include "hast.h"
62 #include "hast_proto.h"
63 #include "hastd.h"
64 #include "hooks.h"
65 #include "metadata.h"
66 #include "proto.h"
67 #include "pjdlog.h"
68 #include "subr.h"
69 #include "synch.h"
70 
71 /* The is only one remote component for now. */
72 #define	ISREMOTE(no)	((no) == 1)
73 
74 struct hio {
75 	/*
76 	 * Number of components we are still waiting for.
77 	 * When this field goes to 0, we can send the request back to the
78 	 * kernel. Each component has to decrease this counter by one
79 	 * even on failure.
80 	 */
81 	unsigned int		 hio_countdown;
82 	/*
83 	 * Each component has a place to store its own error.
84 	 * Once the request is handled by all components we can decide if the
85 	 * request overall is successful or not.
86 	 */
87 	int			*hio_errors;
88 	/*
89 	 * Structure used to communicate with GEOM Gate class.
90 	 */
91 	struct g_gate_ctl_io	 hio_ggio;
92 	/*
93 	 * Request was already confirmed to GEOM Gate.
94 	 */
95 	bool			 hio_done;
96 	/*
97 	 * Remember replication from the time the request was initiated,
98 	 * so we won't get confused when replication changes on reload.
99 	 */
100 	int			 hio_replication;
101 	TAILQ_ENTRY(hio)	*hio_next;
102 };
103 #define	hio_free_next	hio_next[0]
104 #define	hio_done_next	hio_next[0]
105 
106 /*
107  * Free list holds unused structures. When free list is empty, we have to wait
108  * until some in-progress requests are freed.
109  */
110 static TAILQ_HEAD(, hio) hio_free_list;
111 static pthread_mutex_t hio_free_list_lock;
112 static pthread_cond_t hio_free_list_cond;
113 /*
114  * There is one send list for every component. One requests is placed on all
115  * send lists - each component gets the same request, but each component is
116  * responsible for managing his own send list.
117  */
118 static TAILQ_HEAD(, hio) *hio_send_list;
119 static pthread_mutex_t *hio_send_list_lock;
120 static pthread_cond_t *hio_send_list_cond;
121 /*
122  * There is one recv list for every component, although local components don't
123  * use recv lists as local requests are done synchronously.
124  */
125 static TAILQ_HEAD(, hio) *hio_recv_list;
126 static pthread_mutex_t *hio_recv_list_lock;
127 static pthread_cond_t *hio_recv_list_cond;
128 /*
129  * Request is placed on done list by the slowest component (the one that
130  * decreased hio_countdown from 1 to 0).
131  */
132 static TAILQ_HEAD(, hio) hio_done_list;
133 static pthread_mutex_t hio_done_list_lock;
134 static pthread_cond_t hio_done_list_cond;
135 /*
136  * Structure below are for interaction with sync thread.
137  */
138 static bool sync_inprogress;
139 static pthread_mutex_t sync_lock;
140 static pthread_cond_t sync_cond;
141 /*
142  * The lock below allows to synchornize access to remote connections.
143  */
144 static pthread_rwlock_t *hio_remote_lock;
145 
146 /*
147  * Lock to synchronize metadata updates. Also synchronize access to
148  * hr_primary_localcnt and hr_primary_remotecnt fields.
149  */
150 static pthread_mutex_t metadata_lock;
151 
152 /*
153  * Maximum number of outstanding I/O requests.
154  */
155 #define	HAST_HIO_MAX	256
156 /*
157  * Number of components. At this point there are only two components: local
158  * and remote, but in the future it might be possible to use multiple local
159  * and remote components.
160  */
161 #define	HAST_NCOMPONENTS	2
162 
163 #define	ISCONNECTED(res, no)	\
164 	((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL)
165 
166 #define	QUEUE_INSERT1(hio, name, ncomp)	do {				\
167 	bool _wakeup;							\
168 									\
169 	mtx_lock(&hio_##name##_list_lock[(ncomp)]);			\
170 	_wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]);		\
171 	TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio),		\
172 	    hio_next[(ncomp)]);						\
173 	mtx_unlock(&hio_##name##_list_lock[ncomp]);			\
174 	if (_wakeup)							\
175 		cv_signal(&hio_##name##_list_cond[(ncomp)]);		\
176 } while (0)
177 #define	QUEUE_INSERT2(hio, name)	do {				\
178 	bool _wakeup;							\
179 									\
180 	mtx_lock(&hio_##name##_list_lock);				\
181 	_wakeup = TAILQ_EMPTY(&hio_##name##_list);			\
182 	TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\
183 	mtx_unlock(&hio_##name##_list_lock);				\
184 	if (_wakeup)							\
185 		cv_signal(&hio_##name##_list_cond);			\
186 } while (0)
187 #define	QUEUE_TAKE1(hio, name, ncomp, timeout)	do {			\
188 	bool _last;							\
189 									\
190 	mtx_lock(&hio_##name##_list_lock[(ncomp)]);			\
191 	_last = false;							\
192 	while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL && !_last) { \
193 		cv_timedwait(&hio_##name##_list_cond[(ncomp)],		\
194 		    &hio_##name##_list_lock[(ncomp)], (timeout));	\
195 		if ((timeout) != 0)					\
196 			_last = true;					\
197 	}								\
198 	if (hio != NULL) {						\
199 		TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio),	\
200 		    hio_next[(ncomp)]);					\
201 	}								\
202 	mtx_unlock(&hio_##name##_list_lock[(ncomp)]);			\
203 } while (0)
204 #define	QUEUE_TAKE2(hio, name)	do {					\
205 	mtx_lock(&hio_##name##_list_lock);				\
206 	while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) {	\
207 		cv_wait(&hio_##name##_list_cond,			\
208 		    &hio_##name##_list_lock);				\
209 	}								\
210 	TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next);	\
211 	mtx_unlock(&hio_##name##_list_lock);				\
212 } while (0)
213 
214 #define	SYNCREQ(hio)		do {					\
215 	(hio)->hio_ggio.gctl_unit = -1;					\
216 	(hio)->hio_ggio.gctl_seq = 1;					\
217 } while (0)
218 #define	ISSYNCREQ(hio)		((hio)->hio_ggio.gctl_unit == -1)
219 #define	SYNCREQDONE(hio)	do { (hio)->hio_ggio.gctl_unit = -2; } while (0)
220 #define	ISSYNCREQDONE(hio)	((hio)->hio_ggio.gctl_unit == -2)
221 
222 static struct hast_resource *gres;
223 
224 static pthread_mutex_t range_lock;
225 static struct rangelocks *range_regular;
226 static bool range_regular_wait;
227 static pthread_cond_t range_regular_cond;
228 static struct rangelocks *range_sync;
229 static bool range_sync_wait;
230 static pthread_cond_t range_sync_cond;
231 static bool fullystarted;
232 
233 static void *ggate_recv_thread(void *arg);
234 static void *local_send_thread(void *arg);
235 static void *remote_send_thread(void *arg);
236 static void *remote_recv_thread(void *arg);
237 static void *ggate_send_thread(void *arg);
238 static void *sync_thread(void *arg);
239 static void *guard_thread(void *arg);
240 
241 static void
242 cleanup(struct hast_resource *res)
243 {
244 	int rerrno;
245 
246 	/* Remember errno. */
247 	rerrno = errno;
248 
249 	/* Destroy ggate provider if we created one. */
250 	if (res->hr_ggateunit >= 0) {
251 		struct g_gate_ctl_destroy ggiod;
252 
253 		bzero(&ggiod, sizeof(ggiod));
254 		ggiod.gctl_version = G_GATE_VERSION;
255 		ggiod.gctl_unit = res->hr_ggateunit;
256 		ggiod.gctl_force = 1;
257 		if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) == -1) {
258 			pjdlog_errno(LOG_WARNING,
259 			    "Unable to destroy hast/%s device",
260 			    res->hr_provname);
261 		}
262 		res->hr_ggateunit = -1;
263 	}
264 
265 	/* Restore errno. */
266 	errno = rerrno;
267 }
268 
269 static __dead2 void
270 primary_exit(int exitcode, const char *fmt, ...)
271 {
272 	va_list ap;
273 
274 	PJDLOG_ASSERT(exitcode != EX_OK);
275 	va_start(ap, fmt);
276 	pjdlogv_errno(LOG_ERR, fmt, ap);
277 	va_end(ap);
278 	cleanup(gres);
279 	exit(exitcode);
280 }
281 
282 static __dead2 void
283 primary_exitx(int exitcode, const char *fmt, ...)
284 {
285 	va_list ap;
286 
287 	va_start(ap, fmt);
288 	pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap);
289 	va_end(ap);
290 	cleanup(gres);
291 	exit(exitcode);
292 }
293 
294 static int
295 hast_activemap_flush(struct hast_resource *res)
296 {
297 	const unsigned char *buf;
298 	size_t size;
299 
300 	buf = activemap_bitmap(res->hr_amp, &size);
301 	PJDLOG_ASSERT(buf != NULL);
302 	PJDLOG_ASSERT((size % res->hr_local_sectorsize) == 0);
303 	if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) !=
304 	    (ssize_t)size) {
305 		pjdlog_errno(LOG_ERR, "Unable to flush activemap to disk");
306 		return (-1);
307 	}
308 	if (res->hr_metaflush == 1 && g_flush(res->hr_localfd) == -1) {
309 		if (errno == EOPNOTSUPP) {
310 			pjdlog_warning("The %s provider doesn't support flushing write cache. Disabling it.",
311 			    res->hr_localpath);
312 			res->hr_metaflush = 0;
313 		} else {
314 			pjdlog_errno(LOG_ERR,
315 			    "Unable to flush disk cache on activemap update");
316 			return (-1);
317 		}
318 	}
319 	return (0);
320 }
321 
322 static bool
323 real_remote(const struct hast_resource *res)
324 {
325 
326 	return (strcmp(res->hr_remoteaddr, "none") != 0);
327 }
328 
329 static void
330 init_environment(struct hast_resource *res __unused)
331 {
332 	struct hio *hio;
333 	unsigned int ii, ncomps;
334 
335 	/*
336 	 * In the future it might be per-resource value.
337 	 */
338 	ncomps = HAST_NCOMPONENTS;
339 
340 	/*
341 	 * Allocate memory needed by lists.
342 	 */
343 	hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps);
344 	if (hio_send_list == NULL) {
345 		primary_exitx(EX_TEMPFAIL,
346 		    "Unable to allocate %zu bytes of memory for send lists.",
347 		    sizeof(hio_send_list[0]) * ncomps);
348 	}
349 	hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps);
350 	if (hio_send_list_lock == NULL) {
351 		primary_exitx(EX_TEMPFAIL,
352 		    "Unable to allocate %zu bytes of memory for send list locks.",
353 		    sizeof(hio_send_list_lock[0]) * ncomps);
354 	}
355 	hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps);
356 	if (hio_send_list_cond == NULL) {
357 		primary_exitx(EX_TEMPFAIL,
358 		    "Unable to allocate %zu bytes of memory for send list condition variables.",
359 		    sizeof(hio_send_list_cond[0]) * ncomps);
360 	}
361 	hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps);
362 	if (hio_recv_list == NULL) {
363 		primary_exitx(EX_TEMPFAIL,
364 		    "Unable to allocate %zu bytes of memory for recv lists.",
365 		    sizeof(hio_recv_list[0]) * ncomps);
366 	}
367 	hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps);
368 	if (hio_recv_list_lock == NULL) {
369 		primary_exitx(EX_TEMPFAIL,
370 		    "Unable to allocate %zu bytes of memory for recv list locks.",
371 		    sizeof(hio_recv_list_lock[0]) * ncomps);
372 	}
373 	hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps);
374 	if (hio_recv_list_cond == NULL) {
375 		primary_exitx(EX_TEMPFAIL,
376 		    "Unable to allocate %zu bytes of memory for recv list condition variables.",
377 		    sizeof(hio_recv_list_cond[0]) * ncomps);
378 	}
379 	hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps);
380 	if (hio_remote_lock == NULL) {
381 		primary_exitx(EX_TEMPFAIL,
382 		    "Unable to allocate %zu bytes of memory for remote connections locks.",
383 		    sizeof(hio_remote_lock[0]) * ncomps);
384 	}
385 
386 	/*
387 	 * Initialize lists, their locks and theirs condition variables.
388 	 */
389 	TAILQ_INIT(&hio_free_list);
390 	mtx_init(&hio_free_list_lock);
391 	cv_init(&hio_free_list_cond);
392 	for (ii = 0; ii < HAST_NCOMPONENTS; ii++) {
393 		TAILQ_INIT(&hio_send_list[ii]);
394 		mtx_init(&hio_send_list_lock[ii]);
395 		cv_init(&hio_send_list_cond[ii]);
396 		TAILQ_INIT(&hio_recv_list[ii]);
397 		mtx_init(&hio_recv_list_lock[ii]);
398 		cv_init(&hio_recv_list_cond[ii]);
399 		rw_init(&hio_remote_lock[ii]);
400 	}
401 	TAILQ_INIT(&hio_done_list);
402 	mtx_init(&hio_done_list_lock);
403 	cv_init(&hio_done_list_cond);
404 	mtx_init(&metadata_lock);
405 
406 	/*
407 	 * Allocate requests pool and initialize requests.
408 	 */
409 	for (ii = 0; ii < HAST_HIO_MAX; ii++) {
410 		hio = malloc(sizeof(*hio));
411 		if (hio == NULL) {
412 			primary_exitx(EX_TEMPFAIL,
413 			    "Unable to allocate %zu bytes of memory for hio request.",
414 			    sizeof(*hio));
415 		}
416 		hio->hio_countdown = 0;
417 		hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps);
418 		if (hio->hio_errors == NULL) {
419 			primary_exitx(EX_TEMPFAIL,
420 			    "Unable allocate %zu bytes of memory for hio errors.",
421 			    sizeof(hio->hio_errors[0]) * ncomps);
422 		}
423 		hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps);
424 		if (hio->hio_next == NULL) {
425 			primary_exitx(EX_TEMPFAIL,
426 			    "Unable allocate %zu bytes of memory for hio_next field.",
427 			    sizeof(hio->hio_next[0]) * ncomps);
428 		}
429 		hio->hio_ggio.gctl_version = G_GATE_VERSION;
430 		hio->hio_ggio.gctl_data = malloc(MAXPHYS);
431 		if (hio->hio_ggio.gctl_data == NULL) {
432 			primary_exitx(EX_TEMPFAIL,
433 			    "Unable to allocate %zu bytes of memory for gctl_data.",
434 			    MAXPHYS);
435 		}
436 		hio->hio_ggio.gctl_length = MAXPHYS;
437 		hio->hio_ggio.gctl_error = 0;
438 		TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next);
439 	}
440 }
441 
442 static bool
443 init_resuid(struct hast_resource *res)
444 {
445 
446 	mtx_lock(&metadata_lock);
447 	if (res->hr_resuid != 0) {
448 		mtx_unlock(&metadata_lock);
449 		return (false);
450 	} else {
451 		/* Initialize unique resource identifier. */
452 		arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid));
453 		mtx_unlock(&metadata_lock);
454 		if (metadata_write(res) == -1)
455 			exit(EX_NOINPUT);
456 		return (true);
457 	}
458 }
459 
460 static void
461 init_local(struct hast_resource *res)
462 {
463 	unsigned char *buf;
464 	size_t mapsize;
465 
466 	if (metadata_read(res, true) == -1)
467 		exit(EX_NOINPUT);
468 	mtx_init(&res->hr_amp_lock);
469 	if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize,
470 	    res->hr_local_sectorsize, res->hr_keepdirty) == -1) {
471 		primary_exit(EX_TEMPFAIL, "Unable to create activemap");
472 	}
473 	mtx_init(&range_lock);
474 	cv_init(&range_regular_cond);
475 	if (rangelock_init(&range_regular) == -1)
476 		primary_exit(EX_TEMPFAIL, "Unable to create regular range lock");
477 	cv_init(&range_sync_cond);
478 	if (rangelock_init(&range_sync) == -1)
479 		primary_exit(EX_TEMPFAIL, "Unable to create sync range lock");
480 	mapsize = activemap_ondisk_size(res->hr_amp);
481 	buf = calloc(1, mapsize);
482 	if (buf == NULL) {
483 		primary_exitx(EX_TEMPFAIL,
484 		    "Unable to allocate buffer for activemap.");
485 	}
486 	if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) !=
487 	    (ssize_t)mapsize) {
488 		primary_exit(EX_NOINPUT, "Unable to read activemap");
489 	}
490 	activemap_copyin(res->hr_amp, buf, mapsize);
491 	free(buf);
492 	if (res->hr_resuid != 0)
493 		return;
494 	/*
495 	 * We're using provider for the first time. Initialize local and remote
496 	 * counters. We don't initialize resuid here, as we want to do it just
497 	 * in time. The reason for this is that we want to inform secondary
498 	 * that there were no writes yet, so there is no need to synchronize
499 	 * anything.
500 	 */
501 	res->hr_primary_localcnt = 0;
502 	res->hr_primary_remotecnt = 0;
503 	if (metadata_write(res) == -1)
504 		exit(EX_NOINPUT);
505 }
506 
507 static int
508 primary_connect(struct hast_resource *res, struct proto_conn **connp)
509 {
510 	struct proto_conn *conn;
511 	int16_t val;
512 
513 	val = 1;
514 	if (proto_send(res->hr_conn, &val, sizeof(val)) == -1) {
515 		primary_exit(EX_TEMPFAIL,
516 		    "Unable to send connection request to parent");
517 	}
518 	if (proto_recv(res->hr_conn, &val, sizeof(val)) == -1) {
519 		primary_exit(EX_TEMPFAIL,
520 		    "Unable to receive reply to connection request from parent");
521 	}
522 	if (val != 0) {
523 		errno = val;
524 		pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
525 		    res->hr_remoteaddr);
526 		return (-1);
527 	}
528 	if (proto_connection_recv(res->hr_conn, true, &conn) == -1) {
529 		primary_exit(EX_TEMPFAIL,
530 		    "Unable to receive connection from parent");
531 	}
532 	if (proto_connect_wait(conn, res->hr_timeout) == -1) {
533 		pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
534 		    res->hr_remoteaddr);
535 		proto_close(conn);
536 		return (-1);
537 	}
538 	/* Error in setting timeout is not critical, but why should it fail? */
539 	if (proto_timeout(conn, res->hr_timeout) == -1)
540 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
541 
542 	*connp = conn;
543 
544 	return (0);
545 }
546 
547 /*
548  * Function instructs GEOM_GATE to handle reads directly from within the kernel.
549  */
550 static void
551 enable_direct_reads(struct hast_resource *res)
552 {
553 	struct g_gate_ctl_modify ggiomodify;
554 
555 	bzero(&ggiomodify, sizeof(ggiomodify));
556 	ggiomodify.gctl_version = G_GATE_VERSION;
557 	ggiomodify.gctl_unit = res->hr_ggateunit;
558 	ggiomodify.gctl_modify = GG_MODIFY_READPROV | GG_MODIFY_READOFFSET;
559 	strlcpy(ggiomodify.gctl_readprov, res->hr_localpath,
560 	    sizeof(ggiomodify.gctl_readprov));
561 	ggiomodify.gctl_readoffset = res->hr_localoff;
562 	if (ioctl(res->hr_ggatefd, G_GATE_CMD_MODIFY, &ggiomodify) == 0)
563 		pjdlog_debug(1, "Direct reads enabled.");
564 	else
565 		pjdlog_errno(LOG_WARNING, "Failed to enable direct reads");
566 }
567 
568 static int
569 init_remote(struct hast_resource *res, struct proto_conn **inp,
570     struct proto_conn **outp)
571 {
572 	struct proto_conn *in, *out;
573 	struct nv *nvout, *nvin;
574 	const unsigned char *token;
575 	unsigned char *map;
576 	const char *errmsg;
577 	int32_t extentsize;
578 	int64_t datasize;
579 	uint32_t mapsize;
580 	size_t size;
581 	int error;
582 
583 	PJDLOG_ASSERT((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL));
584 	PJDLOG_ASSERT(real_remote(res));
585 
586 	in = out = NULL;
587 	errmsg = NULL;
588 
589 	if (primary_connect(res, &out) == -1)
590 		return (ECONNREFUSED);
591 
592 	error = ECONNABORTED;
593 
594 	/*
595 	 * First handshake step.
596 	 * Setup outgoing connection with remote node.
597 	 */
598 	nvout = nv_alloc();
599 	nv_add_string(nvout, res->hr_name, "resource");
600 	if (nv_error(nvout) != 0) {
601 		pjdlog_common(LOG_WARNING, 0, nv_error(nvout),
602 		    "Unable to allocate header for connection with %s",
603 		    res->hr_remoteaddr);
604 		nv_free(nvout);
605 		goto close;
606 	}
607 	if (hast_proto_send(res, out, nvout, NULL, 0) == -1) {
608 		pjdlog_errno(LOG_WARNING,
609 		    "Unable to send handshake header to %s",
610 		    res->hr_remoteaddr);
611 		nv_free(nvout);
612 		goto close;
613 	}
614 	nv_free(nvout);
615 	if (hast_proto_recv_hdr(out, &nvin) == -1) {
616 		pjdlog_errno(LOG_WARNING,
617 		    "Unable to receive handshake header from %s",
618 		    res->hr_remoteaddr);
619 		goto close;
620 	}
621 	errmsg = nv_get_string(nvin, "errmsg");
622 	if (errmsg != NULL) {
623 		pjdlog_warning("%s", errmsg);
624 		if (nv_exists(nvin, "wait"))
625 			error = EBUSY;
626 		nv_free(nvin);
627 		goto close;
628 	}
629 	token = nv_get_uint8_array(nvin, &size, "token");
630 	if (token == NULL) {
631 		pjdlog_warning("Handshake header from %s has no 'token' field.",
632 		    res->hr_remoteaddr);
633 		nv_free(nvin);
634 		goto close;
635 	}
636 	if (size != sizeof(res->hr_token)) {
637 		pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).",
638 		    res->hr_remoteaddr, size, sizeof(res->hr_token));
639 		nv_free(nvin);
640 		goto close;
641 	}
642 	bcopy(token, res->hr_token, sizeof(res->hr_token));
643 	nv_free(nvin);
644 
645 	/*
646 	 * Second handshake step.
647 	 * Setup incoming connection with remote node.
648 	 */
649 	if (primary_connect(res, &in) == -1)
650 		goto close;
651 
652 	nvout = nv_alloc();
653 	nv_add_string(nvout, res->hr_name, "resource");
654 	nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token),
655 	    "token");
656 	if (res->hr_resuid == 0) {
657 		/*
658 		 * The resuid field was not yet initialized.
659 		 * Because we do synchronization inside init_resuid(), it is
660 		 * possible that someone already initialized it, the function
661 		 * will return false then, but if we successfully initialized
662 		 * it, we will get true. True means that there were no writes
663 		 * to this resource yet and we want to inform secondary that
664 		 * synchronization is not needed by sending "virgin" argument.
665 		 */
666 		if (init_resuid(res))
667 			nv_add_int8(nvout, 1, "virgin");
668 	}
669 	nv_add_uint64(nvout, res->hr_resuid, "resuid");
670 	nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt");
671 	nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt");
672 	if (nv_error(nvout) != 0) {
673 		pjdlog_common(LOG_WARNING, 0, nv_error(nvout),
674 		    "Unable to allocate header for connection with %s",
675 		    res->hr_remoteaddr);
676 		nv_free(nvout);
677 		goto close;
678 	}
679 	if (hast_proto_send(res, in, nvout, NULL, 0) == -1) {
680 		pjdlog_errno(LOG_WARNING,
681 		    "Unable to send handshake header to %s",
682 		    res->hr_remoteaddr);
683 		nv_free(nvout);
684 		goto close;
685 	}
686 	nv_free(nvout);
687 	if (hast_proto_recv_hdr(out, &nvin) == -1) {
688 		pjdlog_errno(LOG_WARNING,
689 		    "Unable to receive handshake header from %s",
690 		    res->hr_remoteaddr);
691 		goto close;
692 	}
693 	errmsg = nv_get_string(nvin, "errmsg");
694 	if (errmsg != NULL) {
695 		pjdlog_warning("%s", errmsg);
696 		nv_free(nvin);
697 		goto close;
698 	}
699 	datasize = nv_get_int64(nvin, "datasize");
700 	if (datasize != res->hr_datasize) {
701 		pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).",
702 		    (intmax_t)res->hr_datasize, (intmax_t)datasize);
703 		nv_free(nvin);
704 		goto close;
705 	}
706 	extentsize = nv_get_int32(nvin, "extentsize");
707 	if (extentsize != res->hr_extentsize) {
708 		pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).",
709 		    (ssize_t)res->hr_extentsize, (ssize_t)extentsize);
710 		nv_free(nvin);
711 		goto close;
712 	}
713 	res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt");
714 	res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt");
715 	res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc");
716 	if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY)
717 		enable_direct_reads(res);
718 	if (nv_exists(nvin, "virgin")) {
719 		/*
720 		 * Secondary was reinitialized, bump localcnt if it is 0 as
721 		 * only we have the data.
722 		 */
723 		PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_PRIMARY);
724 		PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
725 
726 		if (res->hr_primary_localcnt == 0) {
727 			PJDLOG_ASSERT(res->hr_secondary_remotecnt == 0);
728 
729 			mtx_lock(&metadata_lock);
730 			res->hr_primary_localcnt++;
731 			pjdlog_debug(1, "Increasing localcnt to %ju.",
732 			    (uintmax_t)res->hr_primary_localcnt);
733 			(void)metadata_write(res);
734 			mtx_unlock(&metadata_lock);
735 		}
736 	}
737 	map = NULL;
738 	mapsize = nv_get_uint32(nvin, "mapsize");
739 	if (mapsize > 0) {
740 		map = malloc(mapsize);
741 		if (map == NULL) {
742 			pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).",
743 			    (uintmax_t)mapsize);
744 			nv_free(nvin);
745 			goto close;
746 		}
747 		/*
748 		 * Remote node have some dirty extents on its own, lets
749 		 * download its activemap.
750 		 */
751 		if (hast_proto_recv_data(res, out, nvin, map,
752 		    mapsize) == -1) {
753 			pjdlog_errno(LOG_ERR,
754 			    "Unable to receive remote activemap");
755 			nv_free(nvin);
756 			free(map);
757 			goto close;
758 		}
759 		/*
760 		 * Merge local and remote bitmaps.
761 		 */
762 		activemap_merge(res->hr_amp, map, mapsize);
763 		free(map);
764 		/*
765 		 * Now that we merged bitmaps from both nodes, flush it to the
766 		 * disk before we start to synchronize.
767 		 */
768 		(void)hast_activemap_flush(res);
769 	}
770 	nv_free(nvin);
771 #ifdef notyet
772 	/* Setup directions. */
773 	if (proto_send(out, NULL, 0) == -1)
774 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
775 	if (proto_recv(in, NULL, 0) == -1)
776 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
777 #endif
778 	pjdlog_info("Connected to %s.", res->hr_remoteaddr);
779 	if (inp != NULL && outp != NULL) {
780 		*inp = in;
781 		*outp = out;
782 	} else {
783 		res->hr_remotein = in;
784 		res->hr_remoteout = out;
785 	}
786 	event_send(res, EVENT_CONNECT);
787 	return (0);
788 close:
789 	if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0)
790 		event_send(res, EVENT_SPLITBRAIN);
791 	proto_close(out);
792 	if (in != NULL)
793 		proto_close(in);
794 	return (error);
795 }
796 
797 static void
798 sync_start(void)
799 {
800 
801 	mtx_lock(&sync_lock);
802 	sync_inprogress = true;
803 	mtx_unlock(&sync_lock);
804 	cv_signal(&sync_cond);
805 }
806 
807 static void
808 sync_stop(void)
809 {
810 
811 	mtx_lock(&sync_lock);
812 	if (sync_inprogress)
813 		sync_inprogress = false;
814 	mtx_unlock(&sync_lock);
815 }
816 
817 static void
818 init_ggate(struct hast_resource *res)
819 {
820 	struct g_gate_ctl_create ggiocreate;
821 	struct g_gate_ctl_cancel ggiocancel;
822 
823 	/*
824 	 * We communicate with ggate via /dev/ggctl. Open it.
825 	 */
826 	res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR);
827 	if (res->hr_ggatefd == -1)
828 		primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME);
829 	/*
830 	 * Create provider before trying to connect, as connection failure
831 	 * is not critical, but may take some time.
832 	 */
833 	bzero(&ggiocreate, sizeof(ggiocreate));
834 	ggiocreate.gctl_version = G_GATE_VERSION;
835 	ggiocreate.gctl_mediasize = res->hr_datasize;
836 	ggiocreate.gctl_sectorsize = res->hr_local_sectorsize;
837 	ggiocreate.gctl_flags = 0;
838 	ggiocreate.gctl_maxcount = 0;
839 	ggiocreate.gctl_timeout = 0;
840 	ggiocreate.gctl_unit = G_GATE_NAME_GIVEN;
841 	snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s",
842 	    res->hr_provname);
843 	if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) {
844 		pjdlog_info("Device hast/%s created.", res->hr_provname);
845 		res->hr_ggateunit = ggiocreate.gctl_unit;
846 		return;
847 	}
848 	if (errno != EEXIST) {
849 		primary_exit(EX_OSERR, "Unable to create hast/%s device",
850 		    res->hr_provname);
851 	}
852 	pjdlog_debug(1,
853 	    "Device hast/%s already exists, we will try to take it over.",
854 	    res->hr_provname);
855 	/*
856 	 * If we received EEXIST, we assume that the process who created the
857 	 * provider died and didn't clean up. In that case we will start from
858 	 * where he left of.
859 	 */
860 	bzero(&ggiocancel, sizeof(ggiocancel));
861 	ggiocancel.gctl_version = G_GATE_VERSION;
862 	ggiocancel.gctl_unit = G_GATE_NAME_GIVEN;
863 	snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s",
864 	    res->hr_provname);
865 	if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) {
866 		pjdlog_info("Device hast/%s recovered.", res->hr_provname);
867 		res->hr_ggateunit = ggiocancel.gctl_unit;
868 		return;
869 	}
870 	primary_exit(EX_OSERR, "Unable to take over hast/%s device",
871 	    res->hr_provname);
872 }
873 
874 void
875 hastd_primary(struct hast_resource *res)
876 {
877 	pthread_t td;
878 	pid_t pid;
879 	int error, mode, debuglevel;
880 
881 	/*
882 	 * Create communication channel for sending control commands from
883 	 * parent to child.
884 	 */
885 	if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) {
886 		/* TODO: There's no need for this to be fatal error. */
887 		KEEP_ERRNO((void)pidfile_remove(pfh));
888 		pjdlog_exit(EX_OSERR,
889 		    "Unable to create control sockets between parent and child");
890 	}
891 	/*
892 	 * Create communication channel for sending events from child to parent.
893 	 */
894 	if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) {
895 		/* TODO: There's no need for this to be fatal error. */
896 		KEEP_ERRNO((void)pidfile_remove(pfh));
897 		pjdlog_exit(EX_OSERR,
898 		    "Unable to create event sockets between child and parent");
899 	}
900 	/*
901 	 * Create communication channel for sending connection requests from
902 	 * child to parent.
903 	 */
904 	if (proto_client(NULL, "socketpair://", &res->hr_conn) == -1) {
905 		/* TODO: There's no need for this to be fatal error. */
906 		KEEP_ERRNO((void)pidfile_remove(pfh));
907 		pjdlog_exit(EX_OSERR,
908 		    "Unable to create connection sockets between child and parent");
909 	}
910 
911 	pid = fork();
912 	if (pid == -1) {
913 		/* TODO: There's no need for this to be fatal error. */
914 		KEEP_ERRNO((void)pidfile_remove(pfh));
915 		pjdlog_exit(EX_TEMPFAIL, "Unable to fork");
916 	}
917 
918 	if (pid > 0) {
919 		/* This is parent. */
920 		/* Declare that we are receiver. */
921 		proto_recv(res->hr_event, NULL, 0);
922 		proto_recv(res->hr_conn, NULL, 0);
923 		/* Declare that we are sender. */
924 		proto_send(res->hr_ctrl, NULL, 0);
925 		res->hr_workerpid = pid;
926 		return;
927 	}
928 
929 	gres = res;
930 	mode = pjdlog_mode_get();
931 	debuglevel = pjdlog_debug_get();
932 
933 	/* Declare that we are sender. */
934 	proto_send(res->hr_event, NULL, 0);
935 	proto_send(res->hr_conn, NULL, 0);
936 	/* Declare that we are receiver. */
937 	proto_recv(res->hr_ctrl, NULL, 0);
938 	descriptors_cleanup(res);
939 
940 	descriptors_assert(res, mode);
941 
942 	pjdlog_init(mode);
943 	pjdlog_debug_set(debuglevel);
944 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
945 	setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
946 
947 	init_local(res);
948 	init_ggate(res);
949 	init_environment(res);
950 
951 	if (drop_privs(res) != 0) {
952 		cleanup(res);
953 		exit(EX_CONFIG);
954 	}
955 	pjdlog_info("Privileges successfully dropped.");
956 
957 	/*
958 	 * Create the guard thread first, so we can handle signals from the
959 	 * very beginning.
960 	 */
961 	error = pthread_create(&td, NULL, guard_thread, res);
962 	PJDLOG_ASSERT(error == 0);
963 	/*
964 	 * Create the control thread before sending any event to the parent,
965 	 * as we can deadlock when parent sends control request to worker,
966 	 * but worker has no control thread started yet, so parent waits.
967 	 * In the meantime worker sends an event to the parent, but parent
968 	 * is unable to handle the event, because it waits for control
969 	 * request response.
970 	 */
971 	error = pthread_create(&td, NULL, ctrl_thread, res);
972 	PJDLOG_ASSERT(error == 0);
973 	if (real_remote(res)) {
974 		error = init_remote(res, NULL, NULL);
975 		if (error == 0) {
976 			sync_start();
977 		} else if (error == EBUSY) {
978 			time_t start = time(NULL);
979 
980 			pjdlog_warning("Waiting for remote node to become %s for %ds.",
981 			    role2str(HAST_ROLE_SECONDARY),
982 			    res->hr_timeout);
983 			for (;;) {
984 				sleep(1);
985 				error = init_remote(res, NULL, NULL);
986 				if (error != EBUSY)
987 					break;
988 				if (time(NULL) > start + res->hr_timeout)
989 					break;
990 			}
991 			if (error == EBUSY) {
992 				pjdlog_warning("Remote node is still %s, starting anyway.",
993 				    role2str(HAST_ROLE_PRIMARY));
994 			}
995 		}
996 	}
997 	error = pthread_create(&td, NULL, ggate_recv_thread, res);
998 	PJDLOG_ASSERT(error == 0);
999 	error = pthread_create(&td, NULL, local_send_thread, res);
1000 	PJDLOG_ASSERT(error == 0);
1001 	error = pthread_create(&td, NULL, remote_send_thread, res);
1002 	PJDLOG_ASSERT(error == 0);
1003 	error = pthread_create(&td, NULL, remote_recv_thread, res);
1004 	PJDLOG_ASSERT(error == 0);
1005 	error = pthread_create(&td, NULL, ggate_send_thread, res);
1006 	PJDLOG_ASSERT(error == 0);
1007 	fullystarted = true;
1008 	(void)sync_thread(res);
1009 }
1010 
1011 static void
1012 reqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...)
1013 {
1014 	char msg[1024];
1015 	va_list ap;
1016 
1017 	va_start(ap, fmt);
1018 	(void)vsnprintf(msg, sizeof(msg), fmt, ap);
1019 	va_end(ap);
1020 	switch (ggio->gctl_cmd) {
1021 	case BIO_READ:
1022 		(void)snprlcat(msg, sizeof(msg), "READ(%ju, %ju).",
1023 		    (uintmax_t)ggio->gctl_offset,
1024 		    (uintmax_t)ggio->gctl_length);
1025 		break;
1026 	case BIO_DELETE:
1027 		(void)snprlcat(msg, sizeof(msg), "DELETE(%ju, %ju).",
1028 		    (uintmax_t)ggio->gctl_offset,
1029 		    (uintmax_t)ggio->gctl_length);
1030 		break;
1031 	case BIO_FLUSH:
1032 		(void)snprlcat(msg, sizeof(msg), "FLUSH.");
1033 		break;
1034 	case BIO_WRITE:
1035 		(void)snprlcat(msg, sizeof(msg), "WRITE(%ju, %ju).",
1036 		    (uintmax_t)ggio->gctl_offset,
1037 		    (uintmax_t)ggio->gctl_length);
1038 		break;
1039 	default:
1040 		(void)snprlcat(msg, sizeof(msg), "UNKNOWN(%u).",
1041 		    (unsigned int)ggio->gctl_cmd);
1042 		break;
1043 	}
1044 	pjdlog_common(loglevel, debuglevel, -1, "%s", msg);
1045 }
1046 
1047 static void
1048 remote_close(struct hast_resource *res, int ncomp)
1049 {
1050 
1051 	rw_wlock(&hio_remote_lock[ncomp]);
1052 	/*
1053 	 * Check for a race between dropping rlock and acquiring wlock -
1054 	 * another thread can close connection in-between.
1055 	 */
1056 	if (!ISCONNECTED(res, ncomp)) {
1057 		PJDLOG_ASSERT(res->hr_remotein == NULL);
1058 		PJDLOG_ASSERT(res->hr_remoteout == NULL);
1059 		rw_unlock(&hio_remote_lock[ncomp]);
1060 		return;
1061 	}
1062 
1063 	PJDLOG_ASSERT(res->hr_remotein != NULL);
1064 	PJDLOG_ASSERT(res->hr_remoteout != NULL);
1065 
1066 	pjdlog_debug(2, "Closing incoming connection to %s.",
1067 	    res->hr_remoteaddr);
1068 	proto_close(res->hr_remotein);
1069 	res->hr_remotein = NULL;
1070 	pjdlog_debug(2, "Closing outgoing connection to %s.",
1071 	    res->hr_remoteaddr);
1072 	proto_close(res->hr_remoteout);
1073 	res->hr_remoteout = NULL;
1074 
1075 	rw_unlock(&hio_remote_lock[ncomp]);
1076 
1077 	pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr);
1078 
1079 	/*
1080 	 * Stop synchronization if in-progress.
1081 	 */
1082 	sync_stop();
1083 
1084 	event_send(res, EVENT_DISCONNECT);
1085 }
1086 
1087 /*
1088  * Acknowledge write completion to the kernel, but don't update activemap yet.
1089  */
1090 static void
1091 write_complete(struct hast_resource *res, struct hio *hio)
1092 {
1093 	struct g_gate_ctl_io *ggio;
1094 	unsigned int ncomp;
1095 
1096 	PJDLOG_ASSERT(!hio->hio_done);
1097 
1098 	ggio = &hio->hio_ggio;
1099 	PJDLOG_ASSERT(ggio->gctl_cmd == BIO_WRITE);
1100 
1101 	/*
1102 	 * Bump local count if this is first write after
1103 	 * connection failure with remote node.
1104 	 */
1105 	ncomp = 1;
1106 	rw_rlock(&hio_remote_lock[ncomp]);
1107 	if (!ISCONNECTED(res, ncomp)) {
1108 		mtx_lock(&metadata_lock);
1109 		if (res->hr_primary_localcnt == res->hr_secondary_remotecnt) {
1110 			res->hr_primary_localcnt++;
1111 			pjdlog_debug(1, "Increasing localcnt to %ju.",
1112 			    (uintmax_t)res->hr_primary_localcnt);
1113 			(void)metadata_write(res);
1114 		}
1115 		mtx_unlock(&metadata_lock);
1116 	}
1117 	rw_unlock(&hio_remote_lock[ncomp]);
1118 	if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) == -1)
1119 		primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed");
1120 	hio->hio_done = true;
1121 }
1122 
1123 /*
1124  * Thread receives ggate I/O requests from the kernel and passes them to
1125  * appropriate threads:
1126  * WRITE - always goes to both local_send and remote_send threads
1127  * READ (when the block is up-to-date on local component) -
1128  *	only local_send thread
1129  * READ (when the block isn't up-to-date on local component) -
1130  *	only remote_send thread
1131  * DELETE - always goes to both local_send and remote_send threads
1132  * FLUSH - always goes to both local_send and remote_send threads
1133  */
1134 static void *
1135 ggate_recv_thread(void *arg)
1136 {
1137 	struct hast_resource *res = arg;
1138 	struct g_gate_ctl_io *ggio;
1139 	struct hio *hio;
1140 	unsigned int ii, ncomp, ncomps;
1141 	int error;
1142 
1143 	for (;;) {
1144 		pjdlog_debug(2, "ggate_recv: Taking free request.");
1145 		QUEUE_TAKE2(hio, free);
1146 		pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio);
1147 		ggio = &hio->hio_ggio;
1148 		ggio->gctl_unit = res->hr_ggateunit;
1149 		ggio->gctl_length = MAXPHYS;
1150 		ggio->gctl_error = 0;
1151 		hio->hio_done = false;
1152 		hio->hio_replication = res->hr_replication;
1153 		pjdlog_debug(2,
1154 		    "ggate_recv: (%p) Waiting for request from the kernel.",
1155 		    hio);
1156 		if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) == -1) {
1157 			if (sigexit_received)
1158 				pthread_exit(NULL);
1159 			primary_exit(EX_OSERR, "G_GATE_CMD_START failed");
1160 		}
1161 		error = ggio->gctl_error;
1162 		switch (error) {
1163 		case 0:
1164 			break;
1165 		case ECANCELED:
1166 			/* Exit gracefully. */
1167 			if (!sigexit_received) {
1168 				pjdlog_debug(2,
1169 				    "ggate_recv: (%p) Received cancel from the kernel.",
1170 				    hio);
1171 				pjdlog_info("Received cancel from the kernel, exiting.");
1172 			}
1173 			pthread_exit(NULL);
1174 		case ENOMEM:
1175 			/*
1176 			 * Buffer too small? Impossible, we allocate MAXPHYS
1177 			 * bytes - request can't be bigger than that.
1178 			 */
1179 			/* FALLTHROUGH */
1180 		case ENXIO:
1181 		default:
1182 			primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.",
1183 			    strerror(error));
1184 		}
1185 
1186 		ncomp = 0;
1187 		ncomps = HAST_NCOMPONENTS;
1188 
1189 		for (ii = 0; ii < ncomps; ii++)
1190 			hio->hio_errors[ii] = EINVAL;
1191 		reqlog(LOG_DEBUG, 2, ggio,
1192 		    "ggate_recv: (%p) Request received from the kernel: ",
1193 		    hio);
1194 
1195 		/*
1196 		 * Inform all components about new write request.
1197 		 * For read request prefer local component unless the given
1198 		 * range is out-of-date, then use remote component.
1199 		 */
1200 		switch (ggio->gctl_cmd) {
1201 		case BIO_READ:
1202 			res->hr_stat_read++;
1203 			ncomps = 1;
1204 			mtx_lock(&metadata_lock);
1205 			if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF ||
1206 			    res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
1207 				/*
1208 				 * This range is up-to-date on local component,
1209 				 * so handle request locally.
1210 				 */
1211 				 /* Local component is 0 for now. */
1212 				ncomp = 0;
1213 			} else /* if (res->hr_syncsrc ==
1214 			    HAST_SYNCSRC_SECONDARY) */ {
1215 				PJDLOG_ASSERT(res->hr_syncsrc ==
1216 				    HAST_SYNCSRC_SECONDARY);
1217 				/*
1218 				 * This range is out-of-date on local component,
1219 				 * so send request to the remote node.
1220 				 */
1221 				 /* Remote component is 1 for now. */
1222 				ncomp = 1;
1223 			}
1224 			mtx_unlock(&metadata_lock);
1225 			break;
1226 		case BIO_WRITE:
1227 			res->hr_stat_write++;
1228 			if (res->hr_resuid == 0 &&
1229 			    res->hr_primary_localcnt == 0) {
1230 				/* This is first write. */
1231 				res->hr_primary_localcnt = 1;
1232 			}
1233 			for (;;) {
1234 				mtx_lock(&range_lock);
1235 				if (rangelock_islocked(range_sync,
1236 				    ggio->gctl_offset, ggio->gctl_length)) {
1237 					pjdlog_debug(2,
1238 					    "regular: Range offset=%jd length=%zu locked.",
1239 					    (intmax_t)ggio->gctl_offset,
1240 					    (size_t)ggio->gctl_length);
1241 					range_regular_wait = true;
1242 					cv_wait(&range_regular_cond, &range_lock);
1243 					range_regular_wait = false;
1244 					mtx_unlock(&range_lock);
1245 					continue;
1246 				}
1247 				if (rangelock_add(range_regular,
1248 				    ggio->gctl_offset, ggio->gctl_length) == -1) {
1249 					mtx_unlock(&range_lock);
1250 					pjdlog_debug(2,
1251 					    "regular: Range offset=%jd length=%zu is already locked, waiting.",
1252 					    (intmax_t)ggio->gctl_offset,
1253 					    (size_t)ggio->gctl_length);
1254 					sleep(1);
1255 					continue;
1256 				}
1257 				mtx_unlock(&range_lock);
1258 				break;
1259 			}
1260 			mtx_lock(&res->hr_amp_lock);
1261 			if (activemap_write_start(res->hr_amp,
1262 			    ggio->gctl_offset, ggio->gctl_length)) {
1263 				res->hr_stat_activemap_update++;
1264 				(void)hast_activemap_flush(res);
1265 			}
1266 			mtx_unlock(&res->hr_amp_lock);
1267 			break;
1268 		case BIO_DELETE:
1269 			res->hr_stat_delete++;
1270 			break;
1271 		case BIO_FLUSH:
1272 			res->hr_stat_flush++;
1273 			break;
1274 		}
1275 		pjdlog_debug(2,
1276 		    "ggate_recv: (%p) Moving request to the send queues.", hio);
1277 		refcount_init(&hio->hio_countdown, ncomps);
1278 		for (ii = ncomp; ii < ncomp + ncomps; ii++)
1279 			QUEUE_INSERT1(hio, send, ii);
1280 	}
1281 	/* NOTREACHED */
1282 	return (NULL);
1283 }
1284 
1285 /*
1286  * Thread reads from or writes to local component.
1287  * If local read fails, it redirects it to remote_send thread.
1288  */
1289 static void *
1290 local_send_thread(void *arg)
1291 {
1292 	struct hast_resource *res = arg;
1293 	struct g_gate_ctl_io *ggio;
1294 	struct hio *hio;
1295 	unsigned int ncomp, rncomp;
1296 	ssize_t ret;
1297 
1298 	/* Local component is 0 for now. */
1299 	ncomp = 0;
1300 	/* Remote component is 1 for now. */
1301 	rncomp = 1;
1302 
1303 	for (;;) {
1304 		pjdlog_debug(2, "local_send: Taking request.");
1305 		QUEUE_TAKE1(hio, send, ncomp, 0);
1306 		pjdlog_debug(2, "local_send: (%p) Got request.", hio);
1307 		ggio = &hio->hio_ggio;
1308 		switch (ggio->gctl_cmd) {
1309 		case BIO_READ:
1310 			ret = pread(res->hr_localfd, ggio->gctl_data,
1311 			    ggio->gctl_length,
1312 			    ggio->gctl_offset + res->hr_localoff);
1313 			if (ret == ggio->gctl_length)
1314 				hio->hio_errors[ncomp] = 0;
1315 			else if (!ISSYNCREQ(hio)) {
1316 				/*
1317 				 * If READ failed, try to read from remote node.
1318 				 */
1319 				if (ret == -1) {
1320 					reqlog(LOG_WARNING, 0, ggio,
1321 					    "Local request failed (%s), trying remote node. ",
1322 					    strerror(errno));
1323 				} else if (ret != ggio->gctl_length) {
1324 					reqlog(LOG_WARNING, 0, ggio,
1325 					    "Local request failed (%zd != %jd), trying remote node. ",
1326 					    ret, (intmax_t)ggio->gctl_length);
1327 				}
1328 				QUEUE_INSERT1(hio, send, rncomp);
1329 				continue;
1330 			}
1331 			break;
1332 		case BIO_WRITE:
1333 			ret = pwrite(res->hr_localfd, ggio->gctl_data,
1334 			    ggio->gctl_length,
1335 			    ggio->gctl_offset + res->hr_localoff);
1336 			if (ret == -1) {
1337 				hio->hio_errors[ncomp] = errno;
1338 				reqlog(LOG_WARNING, 0, ggio,
1339 				    "Local request failed (%s): ",
1340 				    strerror(errno));
1341 			} else if (ret != ggio->gctl_length) {
1342 				hio->hio_errors[ncomp] = EIO;
1343 				reqlog(LOG_WARNING, 0, ggio,
1344 				    "Local request failed (%zd != %jd): ",
1345 				    ret, (intmax_t)ggio->gctl_length);
1346 			} else {
1347 				hio->hio_errors[ncomp] = 0;
1348 				if (hio->hio_replication ==
1349 				    HAST_REPLICATION_ASYNC &&
1350 				    !ISSYNCREQ(hio)) {
1351 					ggio->gctl_error = 0;
1352 					write_complete(res, hio);
1353 				}
1354 			}
1355 			break;
1356 		case BIO_DELETE:
1357 			ret = g_delete(res->hr_localfd,
1358 			    ggio->gctl_offset + res->hr_localoff,
1359 			    ggio->gctl_length);
1360 			if (ret == -1) {
1361 				hio->hio_errors[ncomp] = errno;
1362 				reqlog(LOG_WARNING, 0, ggio,
1363 				    "Local request failed (%s): ",
1364 				    strerror(errno));
1365 			} else {
1366 				hio->hio_errors[ncomp] = 0;
1367 			}
1368 			break;
1369 		case BIO_FLUSH:
1370 			if (!res->hr_localflush) {
1371 				ret = -1;
1372 				errno = EOPNOTSUPP;
1373 				break;
1374 			}
1375 			ret = g_flush(res->hr_localfd);
1376 			if (ret == -1) {
1377 				if (errno == EOPNOTSUPP)
1378 					res->hr_localflush = false;
1379 				hio->hio_errors[ncomp] = errno;
1380 				reqlog(LOG_WARNING, 0, ggio,
1381 				    "Local request failed (%s): ",
1382 				    strerror(errno));
1383 			} else {
1384 				hio->hio_errors[ncomp] = 0;
1385 			}
1386 			break;
1387 		}
1388 		if (!refcount_release(&hio->hio_countdown))
1389 			continue;
1390 		if (ISSYNCREQ(hio)) {
1391 			mtx_lock(&sync_lock);
1392 			SYNCREQDONE(hio);
1393 			mtx_unlock(&sync_lock);
1394 			cv_signal(&sync_cond);
1395 		} else {
1396 			pjdlog_debug(2,
1397 			    "local_send: (%p) Moving request to the done queue.",
1398 			    hio);
1399 			QUEUE_INSERT2(hio, done);
1400 		}
1401 	}
1402 	/* NOTREACHED */
1403 	return (NULL);
1404 }
1405 
1406 static void
1407 keepalive_send(struct hast_resource *res, unsigned int ncomp)
1408 {
1409 	struct nv *nv;
1410 
1411 	rw_rlock(&hio_remote_lock[ncomp]);
1412 
1413 	if (!ISCONNECTED(res, ncomp)) {
1414 		rw_unlock(&hio_remote_lock[ncomp]);
1415 		return;
1416 	}
1417 
1418 	PJDLOG_ASSERT(res->hr_remotein != NULL);
1419 	PJDLOG_ASSERT(res->hr_remoteout != NULL);
1420 
1421 	nv = nv_alloc();
1422 	nv_add_uint8(nv, HIO_KEEPALIVE, "cmd");
1423 	if (nv_error(nv) != 0) {
1424 		rw_unlock(&hio_remote_lock[ncomp]);
1425 		nv_free(nv);
1426 		pjdlog_debug(1,
1427 		    "keepalive_send: Unable to prepare header to send.");
1428 		return;
1429 	}
1430 	if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) == -1) {
1431 		rw_unlock(&hio_remote_lock[ncomp]);
1432 		pjdlog_common(LOG_DEBUG, 1, errno,
1433 		    "keepalive_send: Unable to send request");
1434 		nv_free(nv);
1435 		remote_close(res, ncomp);
1436 		return;
1437 	}
1438 
1439 	rw_unlock(&hio_remote_lock[ncomp]);
1440 	nv_free(nv);
1441 	pjdlog_debug(2, "keepalive_send: Request sent.");
1442 }
1443 
1444 /*
1445  * Thread sends request to secondary node.
1446  */
1447 static void *
1448 remote_send_thread(void *arg)
1449 {
1450 	struct hast_resource *res = arg;
1451 	struct g_gate_ctl_io *ggio;
1452 	time_t lastcheck, now;
1453 	struct hio *hio;
1454 	struct nv *nv;
1455 	unsigned int ncomp;
1456 	bool wakeup;
1457 	uint64_t offset, length;
1458 	uint8_t cmd;
1459 	void *data;
1460 
1461 	/* Remote component is 1 for now. */
1462 	ncomp = 1;
1463 	lastcheck = time(NULL);
1464 
1465 	for (;;) {
1466 		pjdlog_debug(2, "remote_send: Taking request.");
1467 		QUEUE_TAKE1(hio, send, ncomp, HAST_KEEPALIVE);
1468 		if (hio == NULL) {
1469 			now = time(NULL);
1470 			if (lastcheck + HAST_KEEPALIVE <= now) {
1471 				keepalive_send(res, ncomp);
1472 				lastcheck = now;
1473 			}
1474 			continue;
1475 		}
1476 		pjdlog_debug(2, "remote_send: (%p) Got request.", hio);
1477 		ggio = &hio->hio_ggio;
1478 		switch (ggio->gctl_cmd) {
1479 		case BIO_READ:
1480 			cmd = HIO_READ;
1481 			data = NULL;
1482 			offset = ggio->gctl_offset;
1483 			length = ggio->gctl_length;
1484 			break;
1485 		case BIO_WRITE:
1486 			cmd = HIO_WRITE;
1487 			data = ggio->gctl_data;
1488 			offset = ggio->gctl_offset;
1489 			length = ggio->gctl_length;
1490 			break;
1491 		case BIO_DELETE:
1492 			cmd = HIO_DELETE;
1493 			data = NULL;
1494 			offset = ggio->gctl_offset;
1495 			length = ggio->gctl_length;
1496 			break;
1497 		case BIO_FLUSH:
1498 			cmd = HIO_FLUSH;
1499 			data = NULL;
1500 			offset = 0;
1501 			length = 0;
1502 			break;
1503 		default:
1504 			PJDLOG_ABORT("invalid condition");
1505 		}
1506 		nv = nv_alloc();
1507 		nv_add_uint8(nv, cmd, "cmd");
1508 		nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq");
1509 		nv_add_uint64(nv, offset, "offset");
1510 		nv_add_uint64(nv, length, "length");
1511 		if (nv_error(nv) != 0) {
1512 			hio->hio_errors[ncomp] = nv_error(nv);
1513 			pjdlog_debug(2,
1514 			    "remote_send: (%p) Unable to prepare header to send.",
1515 			    hio);
1516 			reqlog(LOG_ERR, 0, ggio,
1517 			    "Unable to prepare header to send (%s): ",
1518 			    strerror(nv_error(nv)));
1519 			/* Move failed request immediately to the done queue. */
1520 			goto done_queue;
1521 		}
1522 		/*
1523 		 * Protect connection from disappearing.
1524 		 */
1525 		rw_rlock(&hio_remote_lock[ncomp]);
1526 		if (!ISCONNECTED(res, ncomp)) {
1527 			rw_unlock(&hio_remote_lock[ncomp]);
1528 			hio->hio_errors[ncomp] = ENOTCONN;
1529 			goto done_queue;
1530 		}
1531 		/*
1532 		 * Move the request to recv queue before sending it, because
1533 		 * in different order we can get reply before we move request
1534 		 * to recv queue.
1535 		 */
1536 		pjdlog_debug(2,
1537 		    "remote_send: (%p) Moving request to the recv queue.",
1538 		    hio);
1539 		mtx_lock(&hio_recv_list_lock[ncomp]);
1540 		wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]);
1541 		TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]);
1542 		mtx_unlock(&hio_recv_list_lock[ncomp]);
1543 		if (hast_proto_send(res, res->hr_remoteout, nv, data,
1544 		    data != NULL ? length : 0) == -1) {
1545 			hio->hio_errors[ncomp] = errno;
1546 			rw_unlock(&hio_remote_lock[ncomp]);
1547 			pjdlog_debug(2,
1548 			    "remote_send: (%p) Unable to send request.", hio);
1549 			reqlog(LOG_ERR, 0, ggio,
1550 			    "Unable to send request (%s): ",
1551 			    strerror(hio->hio_errors[ncomp]));
1552 			remote_close(res, ncomp);
1553 			/*
1554 			 * Take request back from the receive queue and move
1555 			 * it immediately to the done queue.
1556 			 */
1557 			mtx_lock(&hio_recv_list_lock[ncomp]);
1558 			TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1559 			    hio_next[ncomp]);
1560 			mtx_unlock(&hio_recv_list_lock[ncomp]);
1561 			goto done_queue;
1562 		}
1563 		rw_unlock(&hio_remote_lock[ncomp]);
1564 		nv_free(nv);
1565 		if (wakeup)
1566 			cv_signal(&hio_recv_list_cond[ncomp]);
1567 		continue;
1568 done_queue:
1569 		nv_free(nv);
1570 		if (ISSYNCREQ(hio)) {
1571 			if (!refcount_release(&hio->hio_countdown))
1572 				continue;
1573 			mtx_lock(&sync_lock);
1574 			SYNCREQDONE(hio);
1575 			mtx_unlock(&sync_lock);
1576 			cv_signal(&sync_cond);
1577 			continue;
1578 		}
1579 		if (ggio->gctl_cmd == BIO_WRITE) {
1580 			mtx_lock(&res->hr_amp_lock);
1581 			if (activemap_need_sync(res->hr_amp, ggio->gctl_offset,
1582 			    ggio->gctl_length)) {
1583 				(void)hast_activemap_flush(res);
1584 			}
1585 			mtx_unlock(&res->hr_amp_lock);
1586 		}
1587 		if (!refcount_release(&hio->hio_countdown))
1588 			continue;
1589 		pjdlog_debug(2,
1590 		    "remote_send: (%p) Moving request to the done queue.",
1591 		    hio);
1592 		QUEUE_INSERT2(hio, done);
1593 	}
1594 	/* NOTREACHED */
1595 	return (NULL);
1596 }
1597 
1598 /*
1599  * Thread receives answer from secondary node and passes it to ggate_send
1600  * thread.
1601  */
1602 static void *
1603 remote_recv_thread(void *arg)
1604 {
1605 	struct hast_resource *res = arg;
1606 	struct g_gate_ctl_io *ggio;
1607 	struct hio *hio;
1608 	struct nv *nv;
1609 	unsigned int ncomp;
1610 	uint64_t seq;
1611 	int error;
1612 
1613 	/* Remote component is 1 for now. */
1614 	ncomp = 1;
1615 
1616 	for (;;) {
1617 		/* Wait until there is anything to receive. */
1618 		mtx_lock(&hio_recv_list_lock[ncomp]);
1619 		while (TAILQ_EMPTY(&hio_recv_list[ncomp])) {
1620 			pjdlog_debug(2, "remote_recv: No requests, waiting.");
1621 			cv_wait(&hio_recv_list_cond[ncomp],
1622 			    &hio_recv_list_lock[ncomp]);
1623 		}
1624 		mtx_unlock(&hio_recv_list_lock[ncomp]);
1625 
1626 		rw_rlock(&hio_remote_lock[ncomp]);
1627 		if (!ISCONNECTED(res, ncomp)) {
1628 			rw_unlock(&hio_remote_lock[ncomp]);
1629 			/*
1630 			 * Connection is dead, so move all pending requests to
1631 			 * the done queue (one-by-one).
1632 			 */
1633 			mtx_lock(&hio_recv_list_lock[ncomp]);
1634 			hio = TAILQ_FIRST(&hio_recv_list[ncomp]);
1635 			PJDLOG_ASSERT(hio != NULL);
1636 			TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1637 			    hio_next[ncomp]);
1638 			mtx_unlock(&hio_recv_list_lock[ncomp]);
1639 			goto done_queue;
1640 		}
1641 		if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) {
1642 			pjdlog_errno(LOG_ERR,
1643 			    "Unable to receive reply header");
1644 			rw_unlock(&hio_remote_lock[ncomp]);
1645 			remote_close(res, ncomp);
1646 			continue;
1647 		}
1648 		rw_unlock(&hio_remote_lock[ncomp]);
1649 		seq = nv_get_uint64(nv, "seq");
1650 		if (seq == 0) {
1651 			pjdlog_error("Header contains no 'seq' field.");
1652 			nv_free(nv);
1653 			continue;
1654 		}
1655 		mtx_lock(&hio_recv_list_lock[ncomp]);
1656 		TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) {
1657 			if (hio->hio_ggio.gctl_seq == seq) {
1658 				TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1659 				    hio_next[ncomp]);
1660 				break;
1661 			}
1662 		}
1663 		mtx_unlock(&hio_recv_list_lock[ncomp]);
1664 		if (hio == NULL) {
1665 			pjdlog_error("Found no request matching received 'seq' field (%ju).",
1666 			    (uintmax_t)seq);
1667 			nv_free(nv);
1668 			continue;
1669 		}
1670 		ggio = &hio->hio_ggio;
1671 		error = nv_get_int16(nv, "error");
1672 		if (error != 0) {
1673 			/* Request failed on remote side. */
1674 			hio->hio_errors[ncomp] = error;
1675 			reqlog(LOG_WARNING, 0, ggio,
1676 			    "Remote request failed (%s): ", strerror(error));
1677 			nv_free(nv);
1678 			goto done_queue;
1679 		}
1680 		switch (ggio->gctl_cmd) {
1681 		case BIO_READ:
1682 			rw_rlock(&hio_remote_lock[ncomp]);
1683 			if (!ISCONNECTED(res, ncomp)) {
1684 				rw_unlock(&hio_remote_lock[ncomp]);
1685 				nv_free(nv);
1686 				goto done_queue;
1687 			}
1688 			if (hast_proto_recv_data(res, res->hr_remotein, nv,
1689 			    ggio->gctl_data, ggio->gctl_length) == -1) {
1690 				hio->hio_errors[ncomp] = errno;
1691 				pjdlog_errno(LOG_ERR,
1692 				    "Unable to receive reply data");
1693 				rw_unlock(&hio_remote_lock[ncomp]);
1694 				nv_free(nv);
1695 				remote_close(res, ncomp);
1696 				goto done_queue;
1697 			}
1698 			rw_unlock(&hio_remote_lock[ncomp]);
1699 			break;
1700 		case BIO_WRITE:
1701 		case BIO_DELETE:
1702 		case BIO_FLUSH:
1703 			break;
1704 		default:
1705 			PJDLOG_ABORT("invalid condition");
1706 		}
1707 		hio->hio_errors[ncomp] = 0;
1708 		nv_free(nv);
1709 done_queue:
1710 		if (!refcount_release(&hio->hio_countdown))
1711 			continue;
1712 		if (ISSYNCREQ(hio)) {
1713 			mtx_lock(&sync_lock);
1714 			SYNCREQDONE(hio);
1715 			mtx_unlock(&sync_lock);
1716 			cv_signal(&sync_cond);
1717 		} else {
1718 			pjdlog_debug(2,
1719 			    "remote_recv: (%p) Moving request to the done queue.",
1720 			    hio);
1721 			QUEUE_INSERT2(hio, done);
1722 		}
1723 	}
1724 	/* NOTREACHED */
1725 	return (NULL);
1726 }
1727 
1728 /*
1729  * Thread sends answer to the kernel.
1730  */
1731 static void *
1732 ggate_send_thread(void *arg)
1733 {
1734 	struct hast_resource *res = arg;
1735 	struct g_gate_ctl_io *ggio;
1736 	struct hio *hio;
1737 	unsigned int ii, ncomps;
1738 
1739 	ncomps = HAST_NCOMPONENTS;
1740 
1741 	for (;;) {
1742 		pjdlog_debug(2, "ggate_send: Taking request.");
1743 		QUEUE_TAKE2(hio, done);
1744 		pjdlog_debug(2, "ggate_send: (%p) Got request.", hio);
1745 		ggio = &hio->hio_ggio;
1746 		for (ii = 0; ii < ncomps; ii++) {
1747 			if (hio->hio_errors[ii] == 0) {
1748 				/*
1749 				 * One successful request is enough to declare
1750 				 * success.
1751 				 */
1752 				ggio->gctl_error = 0;
1753 				break;
1754 			}
1755 		}
1756 		if (ii == ncomps) {
1757 			/*
1758 			 * None of the requests were successful.
1759 			 * Use the error from local component except the
1760 			 * case when we did only remote request.
1761 			 */
1762 			if (ggio->gctl_cmd == BIO_READ &&
1763 			    res->hr_syncsrc == HAST_SYNCSRC_SECONDARY)
1764 				ggio->gctl_error = hio->hio_errors[1];
1765 			else
1766 				ggio->gctl_error = hio->hio_errors[0];
1767 		}
1768 		if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) {
1769 			mtx_lock(&res->hr_amp_lock);
1770 			if (activemap_write_complete(res->hr_amp,
1771 			    ggio->gctl_offset, ggio->gctl_length)) {
1772 				res->hr_stat_activemap_update++;
1773 				(void)hast_activemap_flush(res);
1774 			}
1775 			mtx_unlock(&res->hr_amp_lock);
1776 		}
1777 		if (ggio->gctl_cmd == BIO_WRITE) {
1778 			/*
1779 			 * Unlock range we locked.
1780 			 */
1781 			mtx_lock(&range_lock);
1782 			rangelock_del(range_regular, ggio->gctl_offset,
1783 			    ggio->gctl_length);
1784 			if (range_sync_wait)
1785 				cv_signal(&range_sync_cond);
1786 			mtx_unlock(&range_lock);
1787 			if (!hio->hio_done)
1788 				write_complete(res, hio);
1789 		} else {
1790 			if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) == -1) {
1791 				primary_exit(EX_OSERR,
1792 				    "G_GATE_CMD_DONE failed");
1793 			}
1794 		}
1795 		pjdlog_debug(2,
1796 		    "ggate_send: (%p) Moving request to the free queue.", hio);
1797 		QUEUE_INSERT2(hio, free);
1798 	}
1799 	/* NOTREACHED */
1800 	return (NULL);
1801 }
1802 
1803 /*
1804  * Thread synchronize local and remote components.
1805  */
1806 static void *
1807 sync_thread(void *arg __unused)
1808 {
1809 	struct hast_resource *res = arg;
1810 	struct hio *hio;
1811 	struct g_gate_ctl_io *ggio;
1812 	struct timeval tstart, tend, tdiff;
1813 	unsigned int ii, ncomp, ncomps;
1814 	off_t offset, length, synced;
1815 	bool dorewind, directreads;
1816 	int syncext;
1817 
1818 	ncomps = HAST_NCOMPONENTS;
1819 	dorewind = true;
1820 	synced = 0;
1821 	offset = -1;
1822 	directreads = false;
1823 
1824 	for (;;) {
1825 		mtx_lock(&sync_lock);
1826 		if (offset >= 0 && !sync_inprogress) {
1827 			gettimeofday(&tend, NULL);
1828 			timersub(&tend, &tstart, &tdiff);
1829 			pjdlog_info("Synchronization interrupted after %#.0T. "
1830 			    "%NB synchronized so far.", &tdiff,
1831 			    (intmax_t)synced);
1832 			event_send(res, EVENT_SYNCINTR);
1833 		}
1834 		while (!sync_inprogress) {
1835 			dorewind = true;
1836 			synced = 0;
1837 			cv_wait(&sync_cond, &sync_lock);
1838 		}
1839 		mtx_unlock(&sync_lock);
1840 		/*
1841 		 * Obtain offset at which we should synchronize.
1842 		 * Rewind synchronization if needed.
1843 		 */
1844 		mtx_lock(&res->hr_amp_lock);
1845 		if (dorewind)
1846 			activemap_sync_rewind(res->hr_amp);
1847 		offset = activemap_sync_offset(res->hr_amp, &length, &syncext);
1848 		if (syncext != -1) {
1849 			/*
1850 			 * We synchronized entire syncext extent, we can mark
1851 			 * it as clean now.
1852 			 */
1853 			if (activemap_extent_complete(res->hr_amp, syncext))
1854 				(void)hast_activemap_flush(res);
1855 		}
1856 		mtx_unlock(&res->hr_amp_lock);
1857 		if (dorewind) {
1858 			dorewind = false;
1859 			if (offset == -1)
1860 				pjdlog_info("Nodes are in sync.");
1861 			else {
1862 				pjdlog_info("Synchronization started. %NB to go.",
1863 				    (intmax_t)(res->hr_extentsize *
1864 				    activemap_ndirty(res->hr_amp)));
1865 				event_send(res, EVENT_SYNCSTART);
1866 				gettimeofday(&tstart, NULL);
1867 			}
1868 		}
1869 		if (offset == -1) {
1870 			sync_stop();
1871 			pjdlog_debug(1, "Nothing to synchronize.");
1872 			/*
1873 			 * Synchronization complete, make both localcnt and
1874 			 * remotecnt equal.
1875 			 */
1876 			ncomp = 1;
1877 			rw_rlock(&hio_remote_lock[ncomp]);
1878 			if (ISCONNECTED(res, ncomp)) {
1879 				if (synced > 0) {
1880 					int64_t bps;
1881 
1882 					gettimeofday(&tend, NULL);
1883 					timersub(&tend, &tstart, &tdiff);
1884 					bps = (int64_t)((double)synced /
1885 					    ((double)tdiff.tv_sec +
1886 					    (double)tdiff.tv_usec / 1000000));
1887 					pjdlog_info("Synchronization complete. "
1888 					    "%NB synchronized in %#.0lT (%NB/sec).",
1889 					    (intmax_t)synced, &tdiff,
1890 					    (intmax_t)bps);
1891 					event_send(res, EVENT_SYNCDONE);
1892 				}
1893 				mtx_lock(&metadata_lock);
1894 				if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY)
1895 					directreads = true;
1896 				res->hr_syncsrc = HAST_SYNCSRC_UNDEF;
1897 				res->hr_primary_localcnt =
1898 				    res->hr_secondary_remotecnt;
1899 				res->hr_primary_remotecnt =
1900 				    res->hr_secondary_localcnt;
1901 				pjdlog_debug(1,
1902 				    "Setting localcnt to %ju and remotecnt to %ju.",
1903 				    (uintmax_t)res->hr_primary_localcnt,
1904 				    (uintmax_t)res->hr_primary_remotecnt);
1905 				(void)metadata_write(res);
1906 				mtx_unlock(&metadata_lock);
1907 			}
1908 			rw_unlock(&hio_remote_lock[ncomp]);
1909 			if (directreads) {
1910 				directreads = false;
1911 				enable_direct_reads(res);
1912 			}
1913 			continue;
1914 		}
1915 		pjdlog_debug(2, "sync: Taking free request.");
1916 		QUEUE_TAKE2(hio, free);
1917 		pjdlog_debug(2, "sync: (%p) Got free request.", hio);
1918 		/*
1919 		 * Lock the range we are going to synchronize. We don't want
1920 		 * race where someone writes between our read and write.
1921 		 */
1922 		for (;;) {
1923 			mtx_lock(&range_lock);
1924 			if (rangelock_islocked(range_regular, offset, length)) {
1925 				pjdlog_debug(2,
1926 				    "sync: Range offset=%jd length=%jd locked.",
1927 				    (intmax_t)offset, (intmax_t)length);
1928 				range_sync_wait = true;
1929 				cv_wait(&range_sync_cond, &range_lock);
1930 				range_sync_wait = false;
1931 				mtx_unlock(&range_lock);
1932 				continue;
1933 			}
1934 			if (rangelock_add(range_sync, offset, length) == -1) {
1935 				mtx_unlock(&range_lock);
1936 				pjdlog_debug(2,
1937 				    "sync: Range offset=%jd length=%jd is already locked, waiting.",
1938 				    (intmax_t)offset, (intmax_t)length);
1939 				sleep(1);
1940 				continue;
1941 			}
1942 			mtx_unlock(&range_lock);
1943 			break;
1944 		}
1945 		/*
1946 		 * First read the data from synchronization source.
1947 		 */
1948 		SYNCREQ(hio);
1949 		ggio = &hio->hio_ggio;
1950 		ggio->gctl_cmd = BIO_READ;
1951 		ggio->gctl_offset = offset;
1952 		ggio->gctl_length = length;
1953 		ggio->gctl_error = 0;
1954 		hio->hio_done = false;
1955 		hio->hio_replication = res->hr_replication;
1956 		for (ii = 0; ii < ncomps; ii++)
1957 			hio->hio_errors[ii] = EINVAL;
1958 		reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ",
1959 		    hio);
1960 		pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
1961 		    hio);
1962 		mtx_lock(&metadata_lock);
1963 		if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
1964 			/*
1965 			 * This range is up-to-date on local component,
1966 			 * so handle request locally.
1967 			 */
1968 			 /* Local component is 0 for now. */
1969 			ncomp = 0;
1970 		} else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ {
1971 			PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY);
1972 			/*
1973 			 * This range is out-of-date on local component,
1974 			 * so send request to the remote node.
1975 			 */
1976 			 /* Remote component is 1 for now. */
1977 			ncomp = 1;
1978 		}
1979 		mtx_unlock(&metadata_lock);
1980 		refcount_init(&hio->hio_countdown, 1);
1981 		QUEUE_INSERT1(hio, send, ncomp);
1982 
1983 		/*
1984 		 * Let's wait for READ to finish.
1985 		 */
1986 		mtx_lock(&sync_lock);
1987 		while (!ISSYNCREQDONE(hio))
1988 			cv_wait(&sync_cond, &sync_lock);
1989 		mtx_unlock(&sync_lock);
1990 
1991 		if (hio->hio_errors[ncomp] != 0) {
1992 			pjdlog_error("Unable to read synchronization data: %s.",
1993 			    strerror(hio->hio_errors[ncomp]));
1994 			goto free_queue;
1995 		}
1996 
1997 		/*
1998 		 * We read the data from synchronization source, now write it
1999 		 * to synchronization target.
2000 		 */
2001 		SYNCREQ(hio);
2002 		ggio->gctl_cmd = BIO_WRITE;
2003 		for (ii = 0; ii < ncomps; ii++)
2004 			hio->hio_errors[ii] = EINVAL;
2005 		reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ",
2006 		    hio);
2007 		pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2008 		    hio);
2009 		mtx_lock(&metadata_lock);
2010 		if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
2011 			/*
2012 			 * This range is up-to-date on local component,
2013 			 * so we update remote component.
2014 			 */
2015 			 /* Remote component is 1 for now. */
2016 			ncomp = 1;
2017 		} else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ {
2018 			PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY);
2019 			/*
2020 			 * This range is out-of-date on local component,
2021 			 * so we update it.
2022 			 */
2023 			 /* Local component is 0 for now. */
2024 			ncomp = 0;
2025 		}
2026 		mtx_unlock(&metadata_lock);
2027 
2028 		pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2029 		    hio);
2030 		refcount_init(&hio->hio_countdown, 1);
2031 		QUEUE_INSERT1(hio, send, ncomp);
2032 
2033 		/*
2034 		 * Let's wait for WRITE to finish.
2035 		 */
2036 		mtx_lock(&sync_lock);
2037 		while (!ISSYNCREQDONE(hio))
2038 			cv_wait(&sync_cond, &sync_lock);
2039 		mtx_unlock(&sync_lock);
2040 
2041 		if (hio->hio_errors[ncomp] != 0) {
2042 			pjdlog_error("Unable to write synchronization data: %s.",
2043 			    strerror(hio->hio_errors[ncomp]));
2044 			goto free_queue;
2045 		}
2046 
2047 		synced += length;
2048 free_queue:
2049 		mtx_lock(&range_lock);
2050 		rangelock_del(range_sync, offset, length);
2051 		if (range_regular_wait)
2052 			cv_signal(&range_regular_cond);
2053 		mtx_unlock(&range_lock);
2054 		pjdlog_debug(2, "sync: (%p) Moving request to the free queue.",
2055 		    hio);
2056 		QUEUE_INSERT2(hio, free);
2057 	}
2058 	/* NOTREACHED */
2059 	return (NULL);
2060 }
2061 
2062 void
2063 primary_config_reload(struct hast_resource *res, struct nv *nv)
2064 {
2065 	unsigned int ii, ncomps;
2066 	int modified, vint;
2067 	const char *vstr;
2068 
2069 	pjdlog_info("Reloading configuration...");
2070 
2071 	PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
2072 	PJDLOG_ASSERT(gres == res);
2073 	nv_assert(nv, "remoteaddr");
2074 	nv_assert(nv, "sourceaddr");
2075 	nv_assert(nv, "replication");
2076 	nv_assert(nv, "checksum");
2077 	nv_assert(nv, "compression");
2078 	nv_assert(nv, "timeout");
2079 	nv_assert(nv, "exec");
2080 	nv_assert(nv, "metaflush");
2081 
2082 	ncomps = HAST_NCOMPONENTS;
2083 
2084 #define MODIFIED_REMOTEADDR	0x01
2085 #define MODIFIED_SOURCEADDR	0x02
2086 #define MODIFIED_REPLICATION	0x04
2087 #define MODIFIED_CHECKSUM	0x08
2088 #define MODIFIED_COMPRESSION	0x10
2089 #define MODIFIED_TIMEOUT	0x20
2090 #define MODIFIED_EXEC		0x40
2091 #define MODIFIED_METAFLUSH	0x80
2092 	modified = 0;
2093 
2094 	vstr = nv_get_string(nv, "remoteaddr");
2095 	if (strcmp(gres->hr_remoteaddr, vstr) != 0) {
2096 		/*
2097 		 * Don't copy res->hr_remoteaddr to gres just yet.
2098 		 * We want remote_close() to log disconnect from the old
2099 		 * addresses, not from the new ones.
2100 		 */
2101 		modified |= MODIFIED_REMOTEADDR;
2102 	}
2103 	vstr = nv_get_string(nv, "sourceaddr");
2104 	if (strcmp(gres->hr_sourceaddr, vstr) != 0) {
2105 		strlcpy(gres->hr_sourceaddr, vstr, sizeof(gres->hr_sourceaddr));
2106 		modified |= MODIFIED_SOURCEADDR;
2107 	}
2108 	vint = nv_get_int32(nv, "replication");
2109 	if (gres->hr_replication != vint) {
2110 		gres->hr_replication = vint;
2111 		modified |= MODIFIED_REPLICATION;
2112 	}
2113 	vint = nv_get_int32(nv, "checksum");
2114 	if (gres->hr_checksum != vint) {
2115 		gres->hr_checksum = vint;
2116 		modified |= MODIFIED_CHECKSUM;
2117 	}
2118 	vint = nv_get_int32(nv, "compression");
2119 	if (gres->hr_compression != vint) {
2120 		gres->hr_compression = vint;
2121 		modified |= MODIFIED_COMPRESSION;
2122 	}
2123 	vint = nv_get_int32(nv, "timeout");
2124 	if (gres->hr_timeout != vint) {
2125 		gres->hr_timeout = vint;
2126 		modified |= MODIFIED_TIMEOUT;
2127 	}
2128 	vstr = nv_get_string(nv, "exec");
2129 	if (strcmp(gres->hr_exec, vstr) != 0) {
2130 		strlcpy(gres->hr_exec, vstr, sizeof(gres->hr_exec));
2131 		modified |= MODIFIED_EXEC;
2132 	}
2133 	vint = nv_get_int32(nv, "metaflush");
2134 	if (gres->hr_metaflush != vint) {
2135 		gres->hr_metaflush = vint;
2136 		modified |= MODIFIED_METAFLUSH;
2137 	}
2138 
2139 	/*
2140 	 * Change timeout for connected sockets.
2141 	 * Don't bother if we need to reconnect.
2142 	 */
2143 	if ((modified & MODIFIED_TIMEOUT) != 0 &&
2144 	    (modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR)) == 0) {
2145 		for (ii = 0; ii < ncomps; ii++) {
2146 			if (!ISREMOTE(ii))
2147 				continue;
2148 			rw_rlock(&hio_remote_lock[ii]);
2149 			if (!ISCONNECTED(gres, ii)) {
2150 				rw_unlock(&hio_remote_lock[ii]);
2151 				continue;
2152 			}
2153 			rw_unlock(&hio_remote_lock[ii]);
2154 			if (proto_timeout(gres->hr_remotein,
2155 			    gres->hr_timeout) == -1) {
2156 				pjdlog_errno(LOG_WARNING,
2157 				    "Unable to set connection timeout");
2158 			}
2159 			if (proto_timeout(gres->hr_remoteout,
2160 			    gres->hr_timeout) == -1) {
2161 				pjdlog_errno(LOG_WARNING,
2162 				    "Unable to set connection timeout");
2163 			}
2164 		}
2165 	}
2166 	if ((modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR)) != 0) {
2167 		for (ii = 0; ii < ncomps; ii++) {
2168 			if (!ISREMOTE(ii))
2169 				continue;
2170 			remote_close(gres, ii);
2171 		}
2172 		if (modified & MODIFIED_REMOTEADDR) {
2173 			vstr = nv_get_string(nv, "remoteaddr");
2174 			strlcpy(gres->hr_remoteaddr, vstr,
2175 			    sizeof(gres->hr_remoteaddr));
2176 		}
2177 	}
2178 #undef	MODIFIED_REMOTEADDR
2179 #undef	MODIFIED_SOURCEADDR
2180 #undef	MODIFIED_REPLICATION
2181 #undef	MODIFIED_CHECKSUM
2182 #undef	MODIFIED_COMPRESSION
2183 #undef	MODIFIED_TIMEOUT
2184 #undef	MODIFIED_EXEC
2185 #undef	MODIFIED_METAFLUSH
2186 
2187 	pjdlog_info("Configuration reloaded successfully.");
2188 }
2189 
2190 static void
2191 guard_one(struct hast_resource *res, unsigned int ncomp)
2192 {
2193 	struct proto_conn *in, *out;
2194 
2195 	if (!ISREMOTE(ncomp))
2196 		return;
2197 
2198 	rw_rlock(&hio_remote_lock[ncomp]);
2199 
2200 	if (!real_remote(res)) {
2201 		rw_unlock(&hio_remote_lock[ncomp]);
2202 		return;
2203 	}
2204 
2205 	if (ISCONNECTED(res, ncomp)) {
2206 		PJDLOG_ASSERT(res->hr_remotein != NULL);
2207 		PJDLOG_ASSERT(res->hr_remoteout != NULL);
2208 		rw_unlock(&hio_remote_lock[ncomp]);
2209 		pjdlog_debug(2, "remote_guard: Connection to %s is ok.",
2210 		    res->hr_remoteaddr);
2211 		return;
2212 	}
2213 
2214 	PJDLOG_ASSERT(res->hr_remotein == NULL);
2215 	PJDLOG_ASSERT(res->hr_remoteout == NULL);
2216 	/*
2217 	 * Upgrade the lock. It doesn't have to be atomic as no other thread
2218 	 * can change connection status from disconnected to connected.
2219 	 */
2220 	rw_unlock(&hio_remote_lock[ncomp]);
2221 	pjdlog_debug(2, "remote_guard: Reconnecting to %s.",
2222 	    res->hr_remoteaddr);
2223 	in = out = NULL;
2224 	if (init_remote(res, &in, &out) == 0) {
2225 		rw_wlock(&hio_remote_lock[ncomp]);
2226 		PJDLOG_ASSERT(res->hr_remotein == NULL);
2227 		PJDLOG_ASSERT(res->hr_remoteout == NULL);
2228 		PJDLOG_ASSERT(in != NULL && out != NULL);
2229 		res->hr_remotein = in;
2230 		res->hr_remoteout = out;
2231 		rw_unlock(&hio_remote_lock[ncomp]);
2232 		pjdlog_info("Successfully reconnected to %s.",
2233 		    res->hr_remoteaddr);
2234 		sync_start();
2235 	} else {
2236 		/* Both connections should be NULL. */
2237 		PJDLOG_ASSERT(res->hr_remotein == NULL);
2238 		PJDLOG_ASSERT(res->hr_remoteout == NULL);
2239 		PJDLOG_ASSERT(in == NULL && out == NULL);
2240 		pjdlog_debug(2, "remote_guard: Reconnect to %s failed.",
2241 		    res->hr_remoteaddr);
2242 	}
2243 }
2244 
2245 /*
2246  * Thread guards remote connections and reconnects when needed, handles
2247  * signals, etc.
2248  */
2249 static void *
2250 guard_thread(void *arg)
2251 {
2252 	struct hast_resource *res = arg;
2253 	unsigned int ii, ncomps;
2254 	struct timespec timeout;
2255 	time_t lastcheck, now;
2256 	sigset_t mask;
2257 	int signo;
2258 
2259 	ncomps = HAST_NCOMPONENTS;
2260 	lastcheck = time(NULL);
2261 
2262 	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
2263 	PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
2264 	PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
2265 
2266 	timeout.tv_sec = HAST_KEEPALIVE;
2267 	timeout.tv_nsec = 0;
2268 	signo = -1;
2269 
2270 	for (;;) {
2271 		switch (signo) {
2272 		case SIGINT:
2273 		case SIGTERM:
2274 			sigexit_received = true;
2275 			primary_exitx(EX_OK,
2276 			    "Termination signal received, exiting.");
2277 			break;
2278 		default:
2279 			break;
2280 		}
2281 
2282 		/*
2283 		 * Don't check connections until we fully started,
2284 		 * as we may still be looping, waiting for remote node
2285 		 * to switch from primary to secondary.
2286 		 */
2287 		if (fullystarted) {
2288 			pjdlog_debug(2, "remote_guard: Checking connections.");
2289 			now = time(NULL);
2290 			if (lastcheck + HAST_KEEPALIVE <= now) {
2291 				for (ii = 0; ii < ncomps; ii++)
2292 					guard_one(res, ii);
2293 				lastcheck = now;
2294 			}
2295 		}
2296 		signo = sigtimedwait(&mask, NULL, &timeout);
2297 	}
2298 	/* NOTREACHED */
2299 	return (NULL);
2300 }
2301