xref: /freebsd/sbin/hastd/primary.c (revision ce3adf4362fcca6a43e500b2531f0038adbfbd21)
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/stat.h>
39 
40 #include <geom/gate/g_gate.h>
41 
42 #include <err.h>
43 #include <errno.h>
44 #include <fcntl.h>
45 #include <libgeom.h>
46 #include <pthread.h>
47 #include <signal.h>
48 #include <stdint.h>
49 #include <stdio.h>
50 #include <string.h>
51 #include <sysexits.h>
52 #include <unistd.h>
53 
54 #include <activemap.h>
55 #include <nv.h>
56 #include <rangelock.h>
57 
58 #include "control.h"
59 #include "event.h"
60 #include "hast.h"
61 #include "hast_proto.h"
62 #include "hastd.h"
63 #include "hooks.h"
64 #include "metadata.h"
65 #include "proto.h"
66 #include "pjdlog.h"
67 #include "refcnt.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 	refcnt_t		 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_broadcast(&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_broadcast(&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 /* Expects res->hr_amp locked, returns unlocked. */
295 static int
296 hast_activemap_flush(struct hast_resource *res)
297 {
298 	const unsigned char *buf;
299 	size_t size;
300 	int ret;
301 
302 	mtx_lock(&res->hr_amp_diskmap_lock);
303 	buf = activemap_bitmap(res->hr_amp, &size);
304 	mtx_unlock(&res->hr_amp_lock);
305 	PJDLOG_ASSERT(buf != NULL);
306 	PJDLOG_ASSERT((size % res->hr_local_sectorsize) == 0);
307 	ret = 0;
308 	if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) !=
309 	    (ssize_t)size) {
310 		pjdlog_errno(LOG_ERR, "Unable to flush activemap to disk");
311 		res->hr_stat_activemap_write_error++;
312 		ret = -1;
313 	}
314 	if (ret == 0 && res->hr_metaflush == 1 &&
315 	    g_flush(res->hr_localfd) == -1) {
316 		if (errno == EOPNOTSUPP) {
317 			pjdlog_warning("The %s provider doesn't support flushing write cache. Disabling it.",
318 			    res->hr_localpath);
319 			res->hr_metaflush = 0;
320 		} else {
321 			pjdlog_errno(LOG_ERR,
322 			    "Unable to flush disk cache on activemap update");
323 			res->hr_stat_activemap_flush_error++;
324 			ret = -1;
325 		}
326 	}
327 	mtx_unlock(&res->hr_amp_diskmap_lock);
328 	return (ret);
329 }
330 
331 static bool
332 real_remote(const struct hast_resource *res)
333 {
334 
335 	return (strcmp(res->hr_remoteaddr, "none") != 0);
336 }
337 
338 static void
339 init_environment(struct hast_resource *res __unused)
340 {
341 	struct hio *hio;
342 	unsigned int ii, ncomps;
343 
344 	/*
345 	 * In the future it might be per-resource value.
346 	 */
347 	ncomps = HAST_NCOMPONENTS;
348 
349 	/*
350 	 * Allocate memory needed by lists.
351 	 */
352 	hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps);
353 	if (hio_send_list == NULL) {
354 		primary_exitx(EX_TEMPFAIL,
355 		    "Unable to allocate %zu bytes of memory for send lists.",
356 		    sizeof(hio_send_list[0]) * ncomps);
357 	}
358 	hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps);
359 	if (hio_send_list_lock == NULL) {
360 		primary_exitx(EX_TEMPFAIL,
361 		    "Unable to allocate %zu bytes of memory for send list locks.",
362 		    sizeof(hio_send_list_lock[0]) * ncomps);
363 	}
364 	hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps);
365 	if (hio_send_list_cond == NULL) {
366 		primary_exitx(EX_TEMPFAIL,
367 		    "Unable to allocate %zu bytes of memory for send list condition variables.",
368 		    sizeof(hio_send_list_cond[0]) * ncomps);
369 	}
370 	hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps);
371 	if (hio_recv_list == NULL) {
372 		primary_exitx(EX_TEMPFAIL,
373 		    "Unable to allocate %zu bytes of memory for recv lists.",
374 		    sizeof(hio_recv_list[0]) * ncomps);
375 	}
376 	hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps);
377 	if (hio_recv_list_lock == NULL) {
378 		primary_exitx(EX_TEMPFAIL,
379 		    "Unable to allocate %zu bytes of memory for recv list locks.",
380 		    sizeof(hio_recv_list_lock[0]) * ncomps);
381 	}
382 	hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps);
383 	if (hio_recv_list_cond == NULL) {
384 		primary_exitx(EX_TEMPFAIL,
385 		    "Unable to allocate %zu bytes of memory for recv list condition variables.",
386 		    sizeof(hio_recv_list_cond[0]) * ncomps);
387 	}
388 	hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps);
389 	if (hio_remote_lock == NULL) {
390 		primary_exitx(EX_TEMPFAIL,
391 		    "Unable to allocate %zu bytes of memory for remote connections locks.",
392 		    sizeof(hio_remote_lock[0]) * ncomps);
393 	}
394 
395 	/*
396 	 * Initialize lists, their locks and theirs condition variables.
397 	 */
398 	TAILQ_INIT(&hio_free_list);
399 	mtx_init(&hio_free_list_lock);
400 	cv_init(&hio_free_list_cond);
401 	for (ii = 0; ii < HAST_NCOMPONENTS; ii++) {
402 		TAILQ_INIT(&hio_send_list[ii]);
403 		mtx_init(&hio_send_list_lock[ii]);
404 		cv_init(&hio_send_list_cond[ii]);
405 		TAILQ_INIT(&hio_recv_list[ii]);
406 		mtx_init(&hio_recv_list_lock[ii]);
407 		cv_init(&hio_recv_list_cond[ii]);
408 		rw_init(&hio_remote_lock[ii]);
409 	}
410 	TAILQ_INIT(&hio_done_list);
411 	mtx_init(&hio_done_list_lock);
412 	cv_init(&hio_done_list_cond);
413 	mtx_init(&metadata_lock);
414 
415 	/*
416 	 * Allocate requests pool and initialize requests.
417 	 */
418 	for (ii = 0; ii < HAST_HIO_MAX; ii++) {
419 		hio = malloc(sizeof(*hio));
420 		if (hio == NULL) {
421 			primary_exitx(EX_TEMPFAIL,
422 			    "Unable to allocate %zu bytes of memory for hio request.",
423 			    sizeof(*hio));
424 		}
425 		refcnt_init(&hio->hio_countdown, 0);
426 		hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps);
427 		if (hio->hio_errors == NULL) {
428 			primary_exitx(EX_TEMPFAIL,
429 			    "Unable allocate %zu bytes of memory for hio errors.",
430 			    sizeof(hio->hio_errors[0]) * ncomps);
431 		}
432 		hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps);
433 		if (hio->hio_next == NULL) {
434 			primary_exitx(EX_TEMPFAIL,
435 			    "Unable allocate %zu bytes of memory for hio_next field.",
436 			    sizeof(hio->hio_next[0]) * ncomps);
437 		}
438 		hio->hio_ggio.gctl_version = G_GATE_VERSION;
439 		hio->hio_ggio.gctl_data = malloc(MAXPHYS);
440 		if (hio->hio_ggio.gctl_data == NULL) {
441 			primary_exitx(EX_TEMPFAIL,
442 			    "Unable to allocate %zu bytes of memory for gctl_data.",
443 			    MAXPHYS);
444 		}
445 		hio->hio_ggio.gctl_length = MAXPHYS;
446 		hio->hio_ggio.gctl_error = 0;
447 		TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next);
448 	}
449 }
450 
451 static bool
452 init_resuid(struct hast_resource *res)
453 {
454 
455 	mtx_lock(&metadata_lock);
456 	if (res->hr_resuid != 0) {
457 		mtx_unlock(&metadata_lock);
458 		return (false);
459 	} else {
460 		/* Initialize unique resource identifier. */
461 		arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid));
462 		mtx_unlock(&metadata_lock);
463 		if (metadata_write(res) == -1)
464 			exit(EX_NOINPUT);
465 		return (true);
466 	}
467 }
468 
469 static void
470 init_local(struct hast_resource *res)
471 {
472 	unsigned char *buf;
473 	size_t mapsize;
474 
475 	if (metadata_read(res, true) == -1)
476 		exit(EX_NOINPUT);
477 	mtx_init(&res->hr_amp_lock);
478 	if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize,
479 	    res->hr_local_sectorsize, res->hr_keepdirty) == -1) {
480 		primary_exit(EX_TEMPFAIL, "Unable to create activemap");
481 	}
482 	mtx_init(&range_lock);
483 	cv_init(&range_regular_cond);
484 	if (rangelock_init(&range_regular) == -1)
485 		primary_exit(EX_TEMPFAIL, "Unable to create regular range lock");
486 	cv_init(&range_sync_cond);
487 	if (rangelock_init(&range_sync) == -1)
488 		primary_exit(EX_TEMPFAIL, "Unable to create sync range lock");
489 	mapsize = activemap_ondisk_size(res->hr_amp);
490 	buf = calloc(1, mapsize);
491 	if (buf == NULL) {
492 		primary_exitx(EX_TEMPFAIL,
493 		    "Unable to allocate buffer for activemap.");
494 	}
495 	if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) !=
496 	    (ssize_t)mapsize) {
497 		primary_exit(EX_NOINPUT, "Unable to read activemap");
498 	}
499 	activemap_copyin(res->hr_amp, buf, mapsize);
500 	free(buf);
501 	if (res->hr_resuid != 0)
502 		return;
503 	/*
504 	 * We're using provider for the first time. Initialize local and remote
505 	 * counters. We don't initialize resuid here, as we want to do it just
506 	 * in time. The reason for this is that we want to inform secondary
507 	 * that there were no writes yet, so there is no need to synchronize
508 	 * anything.
509 	 */
510 	res->hr_primary_localcnt = 0;
511 	res->hr_primary_remotecnt = 0;
512 	if (metadata_write(res) == -1)
513 		exit(EX_NOINPUT);
514 }
515 
516 static int
517 primary_connect(struct hast_resource *res, struct proto_conn **connp)
518 {
519 	struct proto_conn *conn;
520 	int16_t val;
521 
522 	val = 1;
523 	if (proto_send(res->hr_conn, &val, sizeof(val)) == -1) {
524 		primary_exit(EX_TEMPFAIL,
525 		    "Unable to send connection request to parent");
526 	}
527 	if (proto_recv(res->hr_conn, &val, sizeof(val)) == -1) {
528 		primary_exit(EX_TEMPFAIL,
529 		    "Unable to receive reply to connection request from parent");
530 	}
531 	if (val != 0) {
532 		errno = val;
533 		pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
534 		    res->hr_remoteaddr);
535 		return (-1);
536 	}
537 	if (proto_connection_recv(res->hr_conn, true, &conn) == -1) {
538 		primary_exit(EX_TEMPFAIL,
539 		    "Unable to receive connection from parent");
540 	}
541 	if (proto_connect_wait(conn, res->hr_timeout) == -1) {
542 		pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
543 		    res->hr_remoteaddr);
544 		proto_close(conn);
545 		return (-1);
546 	}
547 	/* Error in setting timeout is not critical, but why should it fail? */
548 	if (proto_timeout(conn, res->hr_timeout) == -1)
549 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
550 
551 	*connp = conn;
552 
553 	return (0);
554 }
555 
556 /*
557  * Function instructs GEOM_GATE to handle reads directly from within the kernel.
558  */
559 static void
560 enable_direct_reads(struct hast_resource *res)
561 {
562 	struct g_gate_ctl_modify ggiomodify;
563 
564 	bzero(&ggiomodify, sizeof(ggiomodify));
565 	ggiomodify.gctl_version = G_GATE_VERSION;
566 	ggiomodify.gctl_unit = res->hr_ggateunit;
567 	ggiomodify.gctl_modify = GG_MODIFY_READPROV | GG_MODIFY_READOFFSET;
568 	strlcpy(ggiomodify.gctl_readprov, res->hr_localpath,
569 	    sizeof(ggiomodify.gctl_readprov));
570 	ggiomodify.gctl_readoffset = res->hr_localoff;
571 	if (ioctl(res->hr_ggatefd, G_GATE_CMD_MODIFY, &ggiomodify) == 0)
572 		pjdlog_debug(1, "Direct reads enabled.");
573 	else
574 		pjdlog_errno(LOG_WARNING, "Failed to enable direct reads");
575 }
576 
577 static int
578 init_remote(struct hast_resource *res, struct proto_conn **inp,
579     struct proto_conn **outp)
580 {
581 	struct proto_conn *in, *out;
582 	struct nv *nvout, *nvin;
583 	const unsigned char *token;
584 	unsigned char *map;
585 	const char *errmsg;
586 	int32_t extentsize;
587 	int64_t datasize;
588 	uint32_t mapsize;
589 	uint8_t version;
590 	size_t size;
591 	int error;
592 
593 	PJDLOG_ASSERT((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL));
594 	PJDLOG_ASSERT(real_remote(res));
595 
596 	in = out = NULL;
597 	errmsg = NULL;
598 
599 	if (primary_connect(res, &out) == -1)
600 		return (ECONNREFUSED);
601 
602 	error = ECONNABORTED;
603 
604 	/*
605 	 * First handshake step.
606 	 * Setup outgoing connection with remote node.
607 	 */
608 	nvout = nv_alloc();
609 	nv_add_string(nvout, res->hr_name, "resource");
610 	nv_add_uint8(nvout, HAST_PROTO_VERSION, "version");
611 	if (nv_error(nvout) != 0) {
612 		pjdlog_common(LOG_WARNING, 0, nv_error(nvout),
613 		    "Unable to allocate header for connection with %s",
614 		    res->hr_remoteaddr);
615 		nv_free(nvout);
616 		goto close;
617 	}
618 	if (hast_proto_send(res, out, nvout, NULL, 0) == -1) {
619 		pjdlog_errno(LOG_WARNING,
620 		    "Unable to send handshake header to %s",
621 		    res->hr_remoteaddr);
622 		nv_free(nvout);
623 		goto close;
624 	}
625 	nv_free(nvout);
626 	if (hast_proto_recv_hdr(out, &nvin) == -1) {
627 		pjdlog_errno(LOG_WARNING,
628 		    "Unable to receive handshake header from %s",
629 		    res->hr_remoteaddr);
630 		goto close;
631 	}
632 	errmsg = nv_get_string(nvin, "errmsg");
633 	if (errmsg != NULL) {
634 		pjdlog_warning("%s", errmsg);
635 		if (nv_exists(nvin, "wait"))
636 			error = EBUSY;
637 		nv_free(nvin);
638 		goto close;
639 	}
640 	version = nv_get_uint8(nvin, "version");
641 	if (version == 0) {
642 		/*
643 		 * If no version is sent, it means this is protocol version 1.
644 		 */
645 		version = 1;
646 	}
647 	if (version > HAST_PROTO_VERSION) {
648 		pjdlog_warning("Invalid version received (%hhu).", version);
649 		nv_free(nvin);
650 		goto close;
651 	}
652 	res->hr_version = version;
653 	pjdlog_debug(1, "Negotiated protocol version %d.", res->hr_version);
654 	token = nv_get_uint8_array(nvin, &size, "token");
655 	if (token == NULL) {
656 		pjdlog_warning("Handshake header from %s has no 'token' field.",
657 		    res->hr_remoteaddr);
658 		nv_free(nvin);
659 		goto close;
660 	}
661 	if (size != sizeof(res->hr_token)) {
662 		pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).",
663 		    res->hr_remoteaddr, size, sizeof(res->hr_token));
664 		nv_free(nvin);
665 		goto close;
666 	}
667 	bcopy(token, res->hr_token, sizeof(res->hr_token));
668 	nv_free(nvin);
669 
670 	/*
671 	 * Second handshake step.
672 	 * Setup incoming connection with remote node.
673 	 */
674 	if (primary_connect(res, &in) == -1)
675 		goto close;
676 
677 	nvout = nv_alloc();
678 	nv_add_string(nvout, res->hr_name, "resource");
679 	nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token),
680 	    "token");
681 	if (res->hr_resuid == 0) {
682 		/*
683 		 * The resuid field was not yet initialized.
684 		 * Because we do synchronization inside init_resuid(), it is
685 		 * possible that someone already initialized it, the function
686 		 * will return false then, but if we successfully initialized
687 		 * it, we will get true. True means that there were no writes
688 		 * to this resource yet and we want to inform secondary that
689 		 * synchronization is not needed by sending "virgin" argument.
690 		 */
691 		if (init_resuid(res))
692 			nv_add_int8(nvout, 1, "virgin");
693 	}
694 	nv_add_uint64(nvout, res->hr_resuid, "resuid");
695 	nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt");
696 	nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt");
697 	if (nv_error(nvout) != 0) {
698 		pjdlog_common(LOG_WARNING, 0, nv_error(nvout),
699 		    "Unable to allocate header for connection with %s",
700 		    res->hr_remoteaddr);
701 		nv_free(nvout);
702 		goto close;
703 	}
704 	if (hast_proto_send(res, in, nvout, NULL, 0) == -1) {
705 		pjdlog_errno(LOG_WARNING,
706 		    "Unable to send handshake header to %s",
707 		    res->hr_remoteaddr);
708 		nv_free(nvout);
709 		goto close;
710 	}
711 	nv_free(nvout);
712 	if (hast_proto_recv_hdr(out, &nvin) == -1) {
713 		pjdlog_errno(LOG_WARNING,
714 		    "Unable to receive handshake header from %s",
715 		    res->hr_remoteaddr);
716 		goto close;
717 	}
718 	errmsg = nv_get_string(nvin, "errmsg");
719 	if (errmsg != NULL) {
720 		pjdlog_warning("%s", errmsg);
721 		nv_free(nvin);
722 		goto close;
723 	}
724 	datasize = nv_get_int64(nvin, "datasize");
725 	if (datasize != res->hr_datasize) {
726 		pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).",
727 		    (intmax_t)res->hr_datasize, (intmax_t)datasize);
728 		nv_free(nvin);
729 		goto close;
730 	}
731 	extentsize = nv_get_int32(nvin, "extentsize");
732 	if (extentsize != res->hr_extentsize) {
733 		pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).",
734 		    (ssize_t)res->hr_extentsize, (ssize_t)extentsize);
735 		nv_free(nvin);
736 		goto close;
737 	}
738 	res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt");
739 	res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt");
740 	res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc");
741 	if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY)
742 		enable_direct_reads(res);
743 	if (nv_exists(nvin, "virgin")) {
744 		/*
745 		 * Secondary was reinitialized, bump localcnt if it is 0 as
746 		 * only we have the data.
747 		 */
748 		PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_PRIMARY);
749 		PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
750 
751 		if (res->hr_primary_localcnt == 0) {
752 			PJDLOG_ASSERT(res->hr_secondary_remotecnt == 0);
753 
754 			mtx_lock(&metadata_lock);
755 			res->hr_primary_localcnt++;
756 			pjdlog_debug(1, "Increasing localcnt to %ju.",
757 			    (uintmax_t)res->hr_primary_localcnt);
758 			(void)metadata_write(res);
759 			mtx_unlock(&metadata_lock);
760 		}
761 	}
762 	map = NULL;
763 	mapsize = nv_get_uint32(nvin, "mapsize");
764 	if (mapsize > 0) {
765 		map = malloc(mapsize);
766 		if (map == NULL) {
767 			pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).",
768 			    (uintmax_t)mapsize);
769 			nv_free(nvin);
770 			goto close;
771 		}
772 		/*
773 		 * Remote node have some dirty extents on its own, lets
774 		 * download its activemap.
775 		 */
776 		if (hast_proto_recv_data(res, out, nvin, map,
777 		    mapsize) == -1) {
778 			pjdlog_errno(LOG_ERR,
779 			    "Unable to receive remote activemap");
780 			nv_free(nvin);
781 			free(map);
782 			goto close;
783 		}
784 		/*
785 		 * Merge local and remote bitmaps.
786 		 */
787 		activemap_merge(res->hr_amp, map, mapsize);
788 		free(map);
789 		/*
790 		 * Now that we merged bitmaps from both nodes, flush it to the
791 		 * disk before we start to synchronize.
792 		 */
793 		mtx_lock(&res->hr_amp_lock);
794 		(void)hast_activemap_flush(res);
795 	}
796 	nv_free(nvin);
797 #ifdef notyet
798 	/* Setup directions. */
799 	if (proto_send(out, NULL, 0) == -1)
800 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
801 	if (proto_recv(in, NULL, 0) == -1)
802 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
803 #endif
804 	pjdlog_info("Connected to %s.", res->hr_remoteaddr);
805 	if (res->hr_original_replication == HAST_REPLICATION_MEMSYNC &&
806 	    res->hr_version < 2) {
807 		pjdlog_warning("The 'memsync' replication mode is not supported by the remote node, falling back to 'fullsync' mode.");
808 		res->hr_replication = HAST_REPLICATION_FULLSYNC;
809 	} else if (res->hr_replication != res->hr_original_replication) {
810 		/*
811 		 * This is in case hastd disconnected and was upgraded.
812 		 */
813 		res->hr_replication = res->hr_original_replication;
814 	}
815 	if (inp != NULL && outp != NULL) {
816 		*inp = in;
817 		*outp = out;
818 	} else {
819 		res->hr_remotein = in;
820 		res->hr_remoteout = out;
821 	}
822 	event_send(res, EVENT_CONNECT);
823 	return (0);
824 close:
825 	if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0)
826 		event_send(res, EVENT_SPLITBRAIN);
827 	proto_close(out);
828 	if (in != NULL)
829 		proto_close(in);
830 	return (error);
831 }
832 
833 static void
834 sync_start(void)
835 {
836 
837 	mtx_lock(&sync_lock);
838 	sync_inprogress = true;
839 	mtx_unlock(&sync_lock);
840 	cv_signal(&sync_cond);
841 }
842 
843 static void
844 sync_stop(void)
845 {
846 
847 	mtx_lock(&sync_lock);
848 	if (sync_inprogress)
849 		sync_inprogress = false;
850 	mtx_unlock(&sync_lock);
851 }
852 
853 static void
854 init_ggate(struct hast_resource *res)
855 {
856 	struct g_gate_ctl_create ggiocreate;
857 	struct g_gate_ctl_cancel ggiocancel;
858 
859 	/*
860 	 * We communicate with ggate via /dev/ggctl. Open it.
861 	 */
862 	res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR);
863 	if (res->hr_ggatefd == -1)
864 		primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME);
865 	/*
866 	 * Create provider before trying to connect, as connection failure
867 	 * is not critical, but may take some time.
868 	 */
869 	bzero(&ggiocreate, sizeof(ggiocreate));
870 	ggiocreate.gctl_version = G_GATE_VERSION;
871 	ggiocreate.gctl_mediasize = res->hr_datasize;
872 	ggiocreate.gctl_sectorsize = res->hr_local_sectorsize;
873 	ggiocreate.gctl_flags = 0;
874 	ggiocreate.gctl_maxcount = 0;
875 	ggiocreate.gctl_timeout = 0;
876 	ggiocreate.gctl_unit = G_GATE_NAME_GIVEN;
877 	snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s",
878 	    res->hr_provname);
879 	if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) {
880 		pjdlog_info("Device hast/%s created.", res->hr_provname);
881 		res->hr_ggateunit = ggiocreate.gctl_unit;
882 		return;
883 	}
884 	if (errno != EEXIST) {
885 		primary_exit(EX_OSERR, "Unable to create hast/%s device",
886 		    res->hr_provname);
887 	}
888 	pjdlog_debug(1,
889 	    "Device hast/%s already exists, we will try to take it over.",
890 	    res->hr_provname);
891 	/*
892 	 * If we received EEXIST, we assume that the process who created the
893 	 * provider died and didn't clean up. In that case we will start from
894 	 * where he left of.
895 	 */
896 	bzero(&ggiocancel, sizeof(ggiocancel));
897 	ggiocancel.gctl_version = G_GATE_VERSION;
898 	ggiocancel.gctl_unit = G_GATE_NAME_GIVEN;
899 	snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s",
900 	    res->hr_provname);
901 	if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) {
902 		pjdlog_info("Device hast/%s recovered.", res->hr_provname);
903 		res->hr_ggateunit = ggiocancel.gctl_unit;
904 		return;
905 	}
906 	primary_exit(EX_OSERR, "Unable to take over hast/%s device",
907 	    res->hr_provname);
908 }
909 
910 void
911 hastd_primary(struct hast_resource *res)
912 {
913 	pthread_t td;
914 	pid_t pid;
915 	int error, mode, debuglevel;
916 
917 	/*
918 	 * Create communication channel for sending control commands from
919 	 * parent to child.
920 	 */
921 	if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) {
922 		/* TODO: There's no need for this to be fatal error. */
923 		KEEP_ERRNO((void)pidfile_remove(pfh));
924 		pjdlog_exit(EX_OSERR,
925 		    "Unable to create control sockets between parent and child");
926 	}
927 	/*
928 	 * Create communication channel for sending events from child to parent.
929 	 */
930 	if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) {
931 		/* TODO: There's no need for this to be fatal error. */
932 		KEEP_ERRNO((void)pidfile_remove(pfh));
933 		pjdlog_exit(EX_OSERR,
934 		    "Unable to create event sockets between child and parent");
935 	}
936 	/*
937 	 * Create communication channel for sending connection requests from
938 	 * child to parent.
939 	 */
940 	if (proto_client(NULL, "socketpair://", &res->hr_conn) == -1) {
941 		/* TODO: There's no need for this to be fatal error. */
942 		KEEP_ERRNO((void)pidfile_remove(pfh));
943 		pjdlog_exit(EX_OSERR,
944 		    "Unable to create connection sockets between child and parent");
945 	}
946 
947 	pid = fork();
948 	if (pid == -1) {
949 		/* TODO: There's no need for this to be fatal error. */
950 		KEEP_ERRNO((void)pidfile_remove(pfh));
951 		pjdlog_exit(EX_TEMPFAIL, "Unable to fork");
952 	}
953 
954 	if (pid > 0) {
955 		/* This is parent. */
956 		/* Declare that we are receiver. */
957 		proto_recv(res->hr_event, NULL, 0);
958 		proto_recv(res->hr_conn, NULL, 0);
959 		/* Declare that we are sender. */
960 		proto_send(res->hr_ctrl, NULL, 0);
961 		res->hr_workerpid = pid;
962 		return;
963 	}
964 
965 	gres = res;
966 	mode = pjdlog_mode_get();
967 	debuglevel = pjdlog_debug_get();
968 
969 	/* Declare that we are sender. */
970 	proto_send(res->hr_event, NULL, 0);
971 	proto_send(res->hr_conn, NULL, 0);
972 	/* Declare that we are receiver. */
973 	proto_recv(res->hr_ctrl, NULL, 0);
974 	descriptors_cleanup(res);
975 
976 	descriptors_assert(res, mode);
977 
978 	pjdlog_init(mode);
979 	pjdlog_debug_set(debuglevel);
980 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
981 	setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
982 
983 	init_local(res);
984 	init_ggate(res);
985 	init_environment(res);
986 
987 	if (drop_privs(res) != 0) {
988 		cleanup(res);
989 		exit(EX_CONFIG);
990 	}
991 	pjdlog_info("Privileges successfully dropped.");
992 
993 	/*
994 	 * Create the guard thread first, so we can handle signals from the
995 	 * very beginning.
996 	 */
997 	error = pthread_create(&td, NULL, guard_thread, res);
998 	PJDLOG_ASSERT(error == 0);
999 	/*
1000 	 * Create the control thread before sending any event to the parent,
1001 	 * as we can deadlock when parent sends control request to worker,
1002 	 * but worker has no control thread started yet, so parent waits.
1003 	 * In the meantime worker sends an event to the parent, but parent
1004 	 * is unable to handle the event, because it waits for control
1005 	 * request response.
1006 	 */
1007 	error = pthread_create(&td, NULL, ctrl_thread, res);
1008 	PJDLOG_ASSERT(error == 0);
1009 	if (real_remote(res)) {
1010 		error = init_remote(res, NULL, NULL);
1011 		if (error == 0) {
1012 			sync_start();
1013 		} else if (error == EBUSY) {
1014 			time_t start = time(NULL);
1015 
1016 			pjdlog_warning("Waiting for remote node to become %s for %ds.",
1017 			    role2str(HAST_ROLE_SECONDARY),
1018 			    res->hr_timeout);
1019 			for (;;) {
1020 				sleep(1);
1021 				error = init_remote(res, NULL, NULL);
1022 				if (error != EBUSY)
1023 					break;
1024 				if (time(NULL) > start + res->hr_timeout)
1025 					break;
1026 			}
1027 			if (error == EBUSY) {
1028 				pjdlog_warning("Remote node is still %s, starting anyway.",
1029 				    role2str(HAST_ROLE_PRIMARY));
1030 			}
1031 		}
1032 	}
1033 	error = pthread_create(&td, NULL, ggate_recv_thread, res);
1034 	PJDLOG_ASSERT(error == 0);
1035 	error = pthread_create(&td, NULL, local_send_thread, res);
1036 	PJDLOG_ASSERT(error == 0);
1037 	error = pthread_create(&td, NULL, remote_send_thread, res);
1038 	PJDLOG_ASSERT(error == 0);
1039 	error = pthread_create(&td, NULL, remote_recv_thread, res);
1040 	PJDLOG_ASSERT(error == 0);
1041 	error = pthread_create(&td, NULL, ggate_send_thread, res);
1042 	PJDLOG_ASSERT(error == 0);
1043 	fullystarted = true;
1044 	(void)sync_thread(res);
1045 }
1046 
1047 static void
1048 reqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio,
1049     const char *fmt, ...)
1050 {
1051 	char msg[1024];
1052 	va_list ap;
1053 
1054 	va_start(ap, fmt);
1055 	(void)vsnprintf(msg, sizeof(msg), fmt, ap);
1056 	va_end(ap);
1057 	switch (ggio->gctl_cmd) {
1058 	case BIO_READ:
1059 		(void)snprlcat(msg, sizeof(msg), "READ(%ju, %ju).",
1060 		    (uintmax_t)ggio->gctl_offset, (uintmax_t)ggio->gctl_length);
1061 		break;
1062 	case BIO_DELETE:
1063 		(void)snprlcat(msg, sizeof(msg), "DELETE(%ju, %ju).",
1064 		    (uintmax_t)ggio->gctl_offset, (uintmax_t)ggio->gctl_length);
1065 		break;
1066 	case BIO_FLUSH:
1067 		(void)snprlcat(msg, sizeof(msg), "FLUSH.");
1068 		break;
1069 	case BIO_WRITE:
1070 		(void)snprlcat(msg, sizeof(msg), "WRITE(%ju, %ju).",
1071 		    (uintmax_t)ggio->gctl_offset, (uintmax_t)ggio->gctl_length);
1072 		break;
1073 	default:
1074 		(void)snprlcat(msg, sizeof(msg), "UNKNOWN(%u).",
1075 		    (unsigned int)ggio->gctl_cmd);
1076 		break;
1077 	}
1078 	pjdlog_common(loglevel, debuglevel, -1, "%s", msg);
1079 }
1080 
1081 static void
1082 remote_close(struct hast_resource *res, int ncomp)
1083 {
1084 
1085 	rw_wlock(&hio_remote_lock[ncomp]);
1086 	/*
1087 	 * Check for a race between dropping rlock and acquiring wlock -
1088 	 * another thread can close connection in-between.
1089 	 */
1090 	if (!ISCONNECTED(res, ncomp)) {
1091 		PJDLOG_ASSERT(res->hr_remotein == NULL);
1092 		PJDLOG_ASSERT(res->hr_remoteout == NULL);
1093 		rw_unlock(&hio_remote_lock[ncomp]);
1094 		return;
1095 	}
1096 
1097 	PJDLOG_ASSERT(res->hr_remotein != NULL);
1098 	PJDLOG_ASSERT(res->hr_remoteout != NULL);
1099 
1100 	pjdlog_debug(2, "Closing incoming connection to %s.",
1101 	    res->hr_remoteaddr);
1102 	proto_close(res->hr_remotein);
1103 	res->hr_remotein = NULL;
1104 	pjdlog_debug(2, "Closing outgoing connection to %s.",
1105 	    res->hr_remoteaddr);
1106 	proto_close(res->hr_remoteout);
1107 	res->hr_remoteout = NULL;
1108 
1109 	rw_unlock(&hio_remote_lock[ncomp]);
1110 
1111 	pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr);
1112 
1113 	/*
1114 	 * Stop synchronization if in-progress.
1115 	 */
1116 	sync_stop();
1117 
1118 	event_send(res, EVENT_DISCONNECT);
1119 }
1120 
1121 /*
1122  * Acknowledge write completion to the kernel, but don't update activemap yet.
1123  */
1124 static void
1125 write_complete(struct hast_resource *res, struct hio *hio)
1126 {
1127 	struct g_gate_ctl_io *ggio;
1128 	unsigned int ncomp;
1129 
1130 	PJDLOG_ASSERT(!hio->hio_done);
1131 
1132 	ggio = &hio->hio_ggio;
1133 	PJDLOG_ASSERT(ggio->gctl_cmd == BIO_WRITE);
1134 
1135 	/*
1136 	 * Bump local count if this is first write after
1137 	 * connection failure with remote node.
1138 	 */
1139 	ncomp = 1;
1140 	rw_rlock(&hio_remote_lock[ncomp]);
1141 	if (!ISCONNECTED(res, ncomp)) {
1142 		mtx_lock(&metadata_lock);
1143 		if (res->hr_primary_localcnt == res->hr_secondary_remotecnt) {
1144 			res->hr_primary_localcnt++;
1145 			pjdlog_debug(1, "Increasing localcnt to %ju.",
1146 			    (uintmax_t)res->hr_primary_localcnt);
1147 			(void)metadata_write(res);
1148 		}
1149 		mtx_unlock(&metadata_lock);
1150 	}
1151 	rw_unlock(&hio_remote_lock[ncomp]);
1152 	if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) == -1)
1153 		primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed");
1154 	hio->hio_done = true;
1155 }
1156 
1157 /*
1158  * Thread receives ggate I/O requests from the kernel and passes them to
1159  * appropriate threads:
1160  * WRITE - always goes to both local_send and remote_send threads
1161  * READ (when the block is up-to-date on local component) -
1162  *	only local_send thread
1163  * READ (when the block isn't up-to-date on local component) -
1164  *	only remote_send thread
1165  * DELETE - always goes to both local_send and remote_send threads
1166  * FLUSH - always goes to both local_send and remote_send threads
1167  */
1168 static void *
1169 ggate_recv_thread(void *arg)
1170 {
1171 	struct hast_resource *res = arg;
1172 	struct g_gate_ctl_io *ggio;
1173 	struct hio *hio;
1174 	unsigned int ii, ncomp, ncomps;
1175 	int error;
1176 
1177 	for (;;) {
1178 		pjdlog_debug(2, "ggate_recv: Taking free request.");
1179 		QUEUE_TAKE2(hio, free);
1180 		pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio);
1181 		ggio = &hio->hio_ggio;
1182 		ggio->gctl_unit = res->hr_ggateunit;
1183 		ggio->gctl_length = MAXPHYS;
1184 		ggio->gctl_error = 0;
1185 		hio->hio_done = false;
1186 		hio->hio_replication = res->hr_replication;
1187 		pjdlog_debug(2,
1188 		    "ggate_recv: (%p) Waiting for request from the kernel.",
1189 		    hio);
1190 		if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) == -1) {
1191 			if (sigexit_received)
1192 				pthread_exit(NULL);
1193 			primary_exit(EX_OSERR, "G_GATE_CMD_START failed");
1194 		}
1195 		error = ggio->gctl_error;
1196 		switch (error) {
1197 		case 0:
1198 			break;
1199 		case ECANCELED:
1200 			/* Exit gracefully. */
1201 			if (!sigexit_received) {
1202 				pjdlog_debug(2,
1203 				    "ggate_recv: (%p) Received cancel from the kernel.",
1204 				    hio);
1205 				pjdlog_info("Received cancel from the kernel, exiting.");
1206 			}
1207 			pthread_exit(NULL);
1208 		case ENOMEM:
1209 			/*
1210 			 * Buffer too small? Impossible, we allocate MAXPHYS
1211 			 * bytes - request can't be bigger than that.
1212 			 */
1213 			/* FALLTHROUGH */
1214 		case ENXIO:
1215 		default:
1216 			primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.",
1217 			    strerror(error));
1218 		}
1219 
1220 		ncomp = 0;
1221 		ncomps = HAST_NCOMPONENTS;
1222 
1223 		for (ii = 0; ii < ncomps; ii++)
1224 			hio->hio_errors[ii] = EINVAL;
1225 		reqlog(LOG_DEBUG, 2, ggio,
1226 		    "ggate_recv: (%p) Request received from the kernel: ",
1227 		    hio);
1228 
1229 		/*
1230 		 * Inform all components about new write request.
1231 		 * For read request prefer local component unless the given
1232 		 * range is out-of-date, then use remote component.
1233 		 */
1234 		switch (ggio->gctl_cmd) {
1235 		case BIO_READ:
1236 			res->hr_stat_read++;
1237 			ncomps = 1;
1238 			mtx_lock(&metadata_lock);
1239 			if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF ||
1240 			    res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
1241 				/*
1242 				 * This range is up-to-date on local component,
1243 				 * so handle request locally.
1244 				 */
1245 				 /* Local component is 0 for now. */
1246 				ncomp = 0;
1247 			} else /* if (res->hr_syncsrc ==
1248 			    HAST_SYNCSRC_SECONDARY) */ {
1249 				PJDLOG_ASSERT(res->hr_syncsrc ==
1250 				    HAST_SYNCSRC_SECONDARY);
1251 				/*
1252 				 * This range is out-of-date on local component,
1253 				 * so send request to the remote node.
1254 				 */
1255 				 /* Remote component is 1 for now. */
1256 				ncomp = 1;
1257 			}
1258 			mtx_unlock(&metadata_lock);
1259 			break;
1260 		case BIO_WRITE:
1261 			res->hr_stat_write++;
1262 			if (res->hr_resuid == 0 &&
1263 			    res->hr_primary_localcnt == 0) {
1264 				/* This is first write. */
1265 				res->hr_primary_localcnt = 1;
1266 			}
1267 			for (;;) {
1268 				mtx_lock(&range_lock);
1269 				if (rangelock_islocked(range_sync,
1270 				    ggio->gctl_offset, ggio->gctl_length)) {
1271 					pjdlog_debug(2,
1272 					    "regular: Range offset=%jd length=%zu locked.",
1273 					    (intmax_t)ggio->gctl_offset,
1274 					    (size_t)ggio->gctl_length);
1275 					range_regular_wait = true;
1276 					cv_wait(&range_regular_cond, &range_lock);
1277 					range_regular_wait = false;
1278 					mtx_unlock(&range_lock);
1279 					continue;
1280 				}
1281 				if (rangelock_add(range_regular,
1282 				    ggio->gctl_offset, ggio->gctl_length) == -1) {
1283 					mtx_unlock(&range_lock);
1284 					pjdlog_debug(2,
1285 					    "regular: Range offset=%jd length=%zu is already locked, waiting.",
1286 					    (intmax_t)ggio->gctl_offset,
1287 					    (size_t)ggio->gctl_length);
1288 					sleep(1);
1289 					continue;
1290 				}
1291 				mtx_unlock(&range_lock);
1292 				break;
1293 			}
1294 			mtx_lock(&res->hr_amp_lock);
1295 			if (activemap_write_start(res->hr_amp,
1296 			    ggio->gctl_offset, ggio->gctl_length)) {
1297 				res->hr_stat_activemap_update++;
1298 				(void)hast_activemap_flush(res);
1299 			} else {
1300 				mtx_unlock(&res->hr_amp_lock);
1301 			}
1302 			break;
1303 		case BIO_DELETE:
1304 			res->hr_stat_delete++;
1305 			break;
1306 		case BIO_FLUSH:
1307 			res->hr_stat_flush++;
1308 			break;
1309 		}
1310 		pjdlog_debug(2,
1311 		    "ggate_recv: (%p) Moving request to the send queues.", hio);
1312 		if (hio->hio_replication == HAST_REPLICATION_MEMSYNC &&
1313 		    ggio->gctl_cmd == BIO_WRITE) {
1314 			/* Each remote request needs two responses in memsync. */
1315 			refcnt_init(&hio->hio_countdown, ncomps + 1);
1316 		} else {
1317 			refcnt_init(&hio->hio_countdown, ncomps);
1318 		}
1319 		for (ii = ncomp; ii < ncomps; ii++)
1320 			QUEUE_INSERT1(hio, send, ii);
1321 	}
1322 	/* NOTREACHED */
1323 	return (NULL);
1324 }
1325 
1326 /*
1327  * Thread reads from or writes to local component.
1328  * If local read fails, it redirects it to remote_send thread.
1329  */
1330 static void *
1331 local_send_thread(void *arg)
1332 {
1333 	struct hast_resource *res = arg;
1334 	struct g_gate_ctl_io *ggio;
1335 	struct hio *hio;
1336 	unsigned int ncomp, rncomp;
1337 	ssize_t ret;
1338 
1339 	/* Local component is 0 for now. */
1340 	ncomp = 0;
1341 	/* Remote component is 1 for now. */
1342 	rncomp = 1;
1343 
1344 	for (;;) {
1345 		pjdlog_debug(2, "local_send: Taking request.");
1346 		QUEUE_TAKE1(hio, send, ncomp, 0);
1347 		pjdlog_debug(2, "local_send: (%p) Got request.", hio);
1348 		ggio = &hio->hio_ggio;
1349 		switch (ggio->gctl_cmd) {
1350 		case BIO_READ:
1351 			ret = pread(res->hr_localfd, ggio->gctl_data,
1352 			    ggio->gctl_length,
1353 			    ggio->gctl_offset + res->hr_localoff);
1354 			if (ret == ggio->gctl_length)
1355 				hio->hio_errors[ncomp] = 0;
1356 			else if (!ISSYNCREQ(hio)) {
1357 				/*
1358 				 * If READ failed, try to read from remote node.
1359 				 */
1360 				if (ret == -1) {
1361 					reqlog(LOG_WARNING, 0, ggio,
1362 					    "Local request failed (%s), trying remote node. ",
1363 					    strerror(errno));
1364 				} else if (ret != ggio->gctl_length) {
1365 					reqlog(LOG_WARNING, 0, ggio,
1366 					    "Local request failed (%zd != %jd), trying remote node. ",
1367 					    ret, (intmax_t)ggio->gctl_length);
1368 				}
1369 				QUEUE_INSERT1(hio, send, rncomp);
1370 				continue;
1371 			}
1372 			break;
1373 		case BIO_WRITE:
1374 			ret = pwrite(res->hr_localfd, ggio->gctl_data,
1375 			    ggio->gctl_length,
1376 			    ggio->gctl_offset + res->hr_localoff);
1377 			if (ret == -1) {
1378 				hio->hio_errors[ncomp] = errno;
1379 				reqlog(LOG_WARNING, 0, ggio,
1380 				    "Local request failed (%s): ",
1381 				    strerror(errno));
1382 			} else if (ret != ggio->gctl_length) {
1383 				hio->hio_errors[ncomp] = EIO;
1384 				reqlog(LOG_WARNING, 0, ggio,
1385 				    "Local request failed (%zd != %jd): ",
1386 				    ret, (intmax_t)ggio->gctl_length);
1387 			} else {
1388 				hio->hio_errors[ncomp] = 0;
1389 				if (hio->hio_replication ==
1390 				    HAST_REPLICATION_ASYNC) {
1391 					ggio->gctl_error = 0;
1392 					write_complete(res, hio);
1393 				}
1394 			}
1395 			break;
1396 		case BIO_DELETE:
1397 			ret = g_delete(res->hr_localfd,
1398 			    ggio->gctl_offset + res->hr_localoff,
1399 			    ggio->gctl_length);
1400 			if (ret == -1) {
1401 				hio->hio_errors[ncomp] = errno;
1402 				reqlog(LOG_WARNING, 0, ggio,
1403 				    "Local request failed (%s): ",
1404 				    strerror(errno));
1405 			} else {
1406 				hio->hio_errors[ncomp] = 0;
1407 			}
1408 			break;
1409 		case BIO_FLUSH:
1410 			if (!res->hr_localflush) {
1411 				ret = -1;
1412 				errno = EOPNOTSUPP;
1413 				break;
1414 			}
1415 			ret = g_flush(res->hr_localfd);
1416 			if (ret == -1) {
1417 				if (errno == EOPNOTSUPP)
1418 					res->hr_localflush = false;
1419 				hio->hio_errors[ncomp] = errno;
1420 				reqlog(LOG_WARNING, 0, ggio,
1421 				    "Local request failed (%s): ",
1422 				    strerror(errno));
1423 			} else {
1424 				hio->hio_errors[ncomp] = 0;
1425 			}
1426 			break;
1427 		}
1428 
1429 		if (hio->hio_replication != HAST_REPLICATION_MEMSYNC ||
1430 		    ggio->gctl_cmd != BIO_WRITE || ISSYNCREQ(hio)) {
1431 			if (refcnt_release(&hio->hio_countdown) > 0)
1432 				continue;
1433 		} else {
1434 			/*
1435 			 * Depending on hio_countdown value, requests finished
1436 			 * in the following order:
1437 			 * 0: remote memsync, remote final, local write
1438 			 * 1: remote memsync, local write, (remote final)
1439 			 * 2: local write, (remote memsync), (remote final)
1440 			 */
1441 			switch (refcnt_release(&hio->hio_countdown)) {
1442 			case 0:
1443 				/*
1444 				 * Local write finished as last.
1445 				 */
1446 				break;
1447 			case 1:
1448 				/*
1449 				 * Local write finished after remote memsync
1450 				 * reply arrvied. We can complete the write now.
1451 				 */
1452 				if (hio->hio_errors[0] == 0)
1453 					write_complete(res, hio);
1454 				continue;
1455 			case 2:
1456 				/*
1457 				 * Local write finished as first.
1458 				 */
1459 				continue;
1460 			default:
1461 				PJDLOG_ABORT("Invalid hio_countdown.");
1462 			}
1463 		}
1464 		if (ISSYNCREQ(hio)) {
1465 			mtx_lock(&sync_lock);
1466 			SYNCREQDONE(hio);
1467 			mtx_unlock(&sync_lock);
1468 			cv_signal(&sync_cond);
1469 		} else {
1470 			pjdlog_debug(2,
1471 			    "local_send: (%p) Moving request to the done queue.",
1472 			    hio);
1473 			QUEUE_INSERT2(hio, done);
1474 		}
1475 	}
1476 	/* NOTREACHED */
1477 	return (NULL);
1478 }
1479 
1480 static void
1481 keepalive_send(struct hast_resource *res, unsigned int ncomp)
1482 {
1483 	struct nv *nv;
1484 
1485 	rw_rlock(&hio_remote_lock[ncomp]);
1486 
1487 	if (!ISCONNECTED(res, ncomp)) {
1488 		rw_unlock(&hio_remote_lock[ncomp]);
1489 		return;
1490 	}
1491 
1492 	PJDLOG_ASSERT(res->hr_remotein != NULL);
1493 	PJDLOG_ASSERT(res->hr_remoteout != NULL);
1494 
1495 	nv = nv_alloc();
1496 	nv_add_uint8(nv, HIO_KEEPALIVE, "cmd");
1497 	if (nv_error(nv) != 0) {
1498 		rw_unlock(&hio_remote_lock[ncomp]);
1499 		nv_free(nv);
1500 		pjdlog_debug(1,
1501 		    "keepalive_send: Unable to prepare header to send.");
1502 		return;
1503 	}
1504 	if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) == -1) {
1505 		rw_unlock(&hio_remote_lock[ncomp]);
1506 		pjdlog_common(LOG_DEBUG, 1, errno,
1507 		    "keepalive_send: Unable to send request");
1508 		nv_free(nv);
1509 		remote_close(res, ncomp);
1510 		return;
1511 	}
1512 
1513 	rw_unlock(&hio_remote_lock[ncomp]);
1514 	nv_free(nv);
1515 	pjdlog_debug(2, "keepalive_send: Request sent.");
1516 }
1517 
1518 /*
1519  * Thread sends request to secondary node.
1520  */
1521 static void *
1522 remote_send_thread(void *arg)
1523 {
1524 	struct hast_resource *res = arg;
1525 	struct g_gate_ctl_io *ggio;
1526 	time_t lastcheck, now;
1527 	struct hio *hio;
1528 	struct nv *nv;
1529 	unsigned int ncomp;
1530 	bool wakeup;
1531 	uint64_t offset, length;
1532 	uint8_t cmd;
1533 	void *data;
1534 
1535 	/* Remote component is 1 for now. */
1536 	ncomp = 1;
1537 	lastcheck = time(NULL);
1538 
1539 	for (;;) {
1540 		pjdlog_debug(2, "remote_send: Taking request.");
1541 		QUEUE_TAKE1(hio, send, ncomp, HAST_KEEPALIVE);
1542 		if (hio == NULL) {
1543 			now = time(NULL);
1544 			if (lastcheck + HAST_KEEPALIVE <= now) {
1545 				keepalive_send(res, ncomp);
1546 				lastcheck = now;
1547 			}
1548 			continue;
1549 		}
1550 		pjdlog_debug(2, "remote_send: (%p) Got request.", hio);
1551 		ggio = &hio->hio_ggio;
1552 		switch (ggio->gctl_cmd) {
1553 		case BIO_READ:
1554 			cmd = HIO_READ;
1555 			data = NULL;
1556 			offset = ggio->gctl_offset;
1557 			length = ggio->gctl_length;
1558 			break;
1559 		case BIO_WRITE:
1560 			cmd = HIO_WRITE;
1561 			data = ggio->gctl_data;
1562 			offset = ggio->gctl_offset;
1563 			length = ggio->gctl_length;
1564 			break;
1565 		case BIO_DELETE:
1566 			cmd = HIO_DELETE;
1567 			data = NULL;
1568 			offset = ggio->gctl_offset;
1569 			length = ggio->gctl_length;
1570 			break;
1571 		case BIO_FLUSH:
1572 			cmd = HIO_FLUSH;
1573 			data = NULL;
1574 			offset = 0;
1575 			length = 0;
1576 			break;
1577 		default:
1578 			PJDLOG_ABORT("invalid condition");
1579 		}
1580 		nv = nv_alloc();
1581 		nv_add_uint8(nv, cmd, "cmd");
1582 		nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq");
1583 		nv_add_uint64(nv, offset, "offset");
1584 		nv_add_uint64(nv, length, "length");
1585 		if (hio->hio_replication == HAST_REPLICATION_MEMSYNC &&
1586 		    ggio->gctl_cmd == BIO_WRITE && !ISSYNCREQ(hio)) {
1587 			nv_add_uint8(nv, 1, "memsync");
1588 		}
1589 		if (nv_error(nv) != 0) {
1590 			hio->hio_errors[ncomp] = nv_error(nv);
1591 			pjdlog_debug(2,
1592 			    "remote_send: (%p) Unable to prepare header to send.",
1593 			    hio);
1594 			reqlog(LOG_ERR, 0, ggio,
1595 			    "Unable to prepare header to send (%s): ",
1596 			    strerror(nv_error(nv)));
1597 			/* Move failed request immediately to the done queue. */
1598 			goto done_queue;
1599 		}
1600 		/*
1601 		 * Protect connection from disappearing.
1602 		 */
1603 		rw_rlock(&hio_remote_lock[ncomp]);
1604 		if (!ISCONNECTED(res, ncomp)) {
1605 			rw_unlock(&hio_remote_lock[ncomp]);
1606 			hio->hio_errors[ncomp] = ENOTCONN;
1607 			goto done_queue;
1608 		}
1609 		/*
1610 		 * Move the request to recv queue before sending it, because
1611 		 * in different order we can get reply before we move request
1612 		 * to recv queue.
1613 		 */
1614 		pjdlog_debug(2,
1615 		    "remote_send: (%p) Moving request to the recv queue.",
1616 		    hio);
1617 		mtx_lock(&hio_recv_list_lock[ncomp]);
1618 		wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]);
1619 		TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]);
1620 		mtx_unlock(&hio_recv_list_lock[ncomp]);
1621 		if (hast_proto_send(res, res->hr_remoteout, nv, data,
1622 		    data != NULL ? length : 0) == -1) {
1623 			hio->hio_errors[ncomp] = errno;
1624 			rw_unlock(&hio_remote_lock[ncomp]);
1625 			pjdlog_debug(2,
1626 			    "remote_send: (%p) Unable to send request.", hio);
1627 			reqlog(LOG_ERR, 0, ggio,
1628 			    "Unable to send request (%s): ",
1629 			    strerror(hio->hio_errors[ncomp]));
1630 			remote_close(res, ncomp);
1631 			/*
1632 			 * Take request back from the receive queue and move
1633 			 * it immediately to the done queue.
1634 			 */
1635 			mtx_lock(&hio_recv_list_lock[ncomp]);
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 		rw_unlock(&hio_remote_lock[ncomp]);
1642 		nv_free(nv);
1643 		if (wakeup)
1644 			cv_signal(&hio_recv_list_cond[ncomp]);
1645 		continue;
1646 done_queue:
1647 		nv_free(nv);
1648 		if (ISSYNCREQ(hio)) {
1649 			if (refcnt_release(&hio->hio_countdown) > 0)
1650 				continue;
1651 			mtx_lock(&sync_lock);
1652 			SYNCREQDONE(hio);
1653 			mtx_unlock(&sync_lock);
1654 			cv_signal(&sync_cond);
1655 			continue;
1656 		}
1657 		if (ggio->gctl_cmd == BIO_WRITE) {
1658 			mtx_lock(&res->hr_amp_lock);
1659 			if (activemap_need_sync(res->hr_amp, ggio->gctl_offset,
1660 			    ggio->gctl_length)) {
1661 				(void)hast_activemap_flush(res);
1662 			} else {
1663 				mtx_unlock(&res->hr_amp_lock);
1664 			}
1665 			if (hio->hio_replication == HAST_REPLICATION_MEMSYNC)
1666 				(void)refcnt_release(&hio->hio_countdown);
1667 		}
1668 		if (refcnt_release(&hio->hio_countdown) > 0)
1669 			continue;
1670 		pjdlog_debug(2,
1671 		    "remote_send: (%p) Moving request to the done queue.",
1672 		    hio);
1673 		QUEUE_INSERT2(hio, done);
1674 	}
1675 	/* NOTREACHED */
1676 	return (NULL);
1677 }
1678 
1679 /*
1680  * Thread receives answer from secondary node and passes it to ggate_send
1681  * thread.
1682  */
1683 static void *
1684 remote_recv_thread(void *arg)
1685 {
1686 	struct hast_resource *res = arg;
1687 	struct g_gate_ctl_io *ggio;
1688 	struct hio *hio;
1689 	struct nv *nv;
1690 	unsigned int ncomp;
1691 	uint64_t seq;
1692 	bool memsyncack;
1693 	int error;
1694 
1695 	/* Remote component is 1 for now. */
1696 	ncomp = 1;
1697 
1698 	for (;;) {
1699 		/* Wait until there is anything to receive. */
1700 		mtx_lock(&hio_recv_list_lock[ncomp]);
1701 		while (TAILQ_EMPTY(&hio_recv_list[ncomp])) {
1702 			pjdlog_debug(2, "remote_recv: No requests, waiting.");
1703 			cv_wait(&hio_recv_list_cond[ncomp],
1704 			    &hio_recv_list_lock[ncomp]);
1705 		}
1706 		mtx_unlock(&hio_recv_list_lock[ncomp]);
1707 
1708 		memsyncack = false;
1709 
1710 		rw_rlock(&hio_remote_lock[ncomp]);
1711 		if (!ISCONNECTED(res, ncomp)) {
1712 			rw_unlock(&hio_remote_lock[ncomp]);
1713 			/*
1714 			 * Connection is dead, so move all pending requests to
1715 			 * the done queue (one-by-one).
1716 			 */
1717 			mtx_lock(&hio_recv_list_lock[ncomp]);
1718 			hio = TAILQ_FIRST(&hio_recv_list[ncomp]);
1719 			PJDLOG_ASSERT(hio != NULL);
1720 			TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1721 			    hio_next[ncomp]);
1722 			mtx_unlock(&hio_recv_list_lock[ncomp]);
1723 			goto done_queue;
1724 		}
1725 		if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) {
1726 			pjdlog_errno(LOG_ERR,
1727 			    "Unable to receive reply header");
1728 			rw_unlock(&hio_remote_lock[ncomp]);
1729 			remote_close(res, ncomp);
1730 			continue;
1731 		}
1732 		rw_unlock(&hio_remote_lock[ncomp]);
1733 		seq = nv_get_uint64(nv, "seq");
1734 		if (seq == 0) {
1735 			pjdlog_error("Header contains no 'seq' field.");
1736 			nv_free(nv);
1737 			continue;
1738 		}
1739 		memsyncack = nv_exists(nv, "received");
1740 		mtx_lock(&hio_recv_list_lock[ncomp]);
1741 		TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) {
1742 			if (hio->hio_ggio.gctl_seq == seq) {
1743 				TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1744 				    hio_next[ncomp]);
1745 				break;
1746 			}
1747 		}
1748 		mtx_unlock(&hio_recv_list_lock[ncomp]);
1749 		if (hio == NULL) {
1750 			pjdlog_error("Found no request matching received 'seq' field (%ju).",
1751 			    (uintmax_t)seq);
1752 			nv_free(nv);
1753 			continue;
1754 		}
1755 		ggio = &hio->hio_ggio;
1756 		error = nv_get_int16(nv, "error");
1757 		if (error != 0) {
1758 			/* Request failed on remote side. */
1759 			hio->hio_errors[ncomp] = error;
1760 			reqlog(LOG_WARNING, 0, ggio,
1761 			    "Remote request failed (%s): ", strerror(error));
1762 			nv_free(nv);
1763 			goto done_queue;
1764 		}
1765 		switch (ggio->gctl_cmd) {
1766 		case BIO_READ:
1767 			rw_rlock(&hio_remote_lock[ncomp]);
1768 			if (!ISCONNECTED(res, ncomp)) {
1769 				rw_unlock(&hio_remote_lock[ncomp]);
1770 				nv_free(nv);
1771 				goto done_queue;
1772 			}
1773 			if (hast_proto_recv_data(res, res->hr_remotein, nv,
1774 			    ggio->gctl_data, ggio->gctl_length) == -1) {
1775 				hio->hio_errors[ncomp] = errno;
1776 				pjdlog_errno(LOG_ERR,
1777 				    "Unable to receive reply data");
1778 				rw_unlock(&hio_remote_lock[ncomp]);
1779 				nv_free(nv);
1780 				remote_close(res, ncomp);
1781 				goto done_queue;
1782 			}
1783 			rw_unlock(&hio_remote_lock[ncomp]);
1784 			break;
1785 		case BIO_WRITE:
1786 		case BIO_DELETE:
1787 		case BIO_FLUSH:
1788 			break;
1789 		default:
1790 			PJDLOG_ABORT("invalid condition");
1791 		}
1792 		hio->hio_errors[ncomp] = 0;
1793 		nv_free(nv);
1794 done_queue:
1795 		if (hio->hio_replication != HAST_REPLICATION_MEMSYNC ||
1796 		    hio->hio_ggio.gctl_cmd != BIO_WRITE || ISSYNCREQ(hio)) {
1797 			if (refcnt_release(&hio->hio_countdown) > 0)
1798 				continue;
1799 		} else {
1800 			/*
1801 			 * Depending on hio_countdown value, requests finished
1802 			 * in the following order:
1803 			 *
1804 			 * 0: local write, remote memsync, remote final
1805 			 * or
1806 			 * 0: remote memsync, local write, remote final
1807 			 *
1808 			 * 1: local write, remote memsync, (remote final)
1809 			 * or
1810 			 * 1: remote memsync, remote final, (local write)
1811 			 *
1812 			 * 2: remote memsync, (local write), (remote final)
1813 			 * or
1814 			 * 2: remote memsync, (remote final), (local write)
1815 			 */
1816 			switch (refcnt_release(&hio->hio_countdown)) {
1817 			case 0:
1818 				/*
1819 				 * Remote final reply arrived.
1820 				 */
1821 				PJDLOG_ASSERT(!memsyncack);
1822 				break;
1823 			case 1:
1824 				if (memsyncack) {
1825 					/*
1826 					 * Local request already finished, so we
1827 					 * can complete the write.
1828 					 */
1829 					if (hio->hio_errors[0] == 0)
1830 						write_complete(res, hio);
1831 					/*
1832 					 * We still need to wait for final
1833 					 * remote reply.
1834 					 */
1835 					pjdlog_debug(2,
1836 					    "remote_recv: (%p) Moving request back to the recv queue.",
1837 					    hio);
1838 					mtx_lock(&hio_recv_list_lock[ncomp]);
1839 					TAILQ_INSERT_TAIL(&hio_recv_list[ncomp],
1840 					    hio, hio_next[ncomp]);
1841 					mtx_unlock(&hio_recv_list_lock[ncomp]);
1842 				} else {
1843 					/*
1844 					 * Remote final reply arrived before
1845 					 * local write finished.
1846 					 * Nothing to do in such case.
1847 					 */
1848 				}
1849 				continue;
1850 			case 2:
1851 				/*
1852 				 * We received remote memsync reply even before
1853 				 * local write finished.
1854 				 */
1855 				PJDLOG_ASSERT(memsyncack);
1856 
1857 				pjdlog_debug(2,
1858 				    "remote_recv: (%p) Moving request back to the recv queue.",
1859 				    hio);
1860 				mtx_lock(&hio_recv_list_lock[ncomp]);
1861 				TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio,
1862 				    hio_next[ncomp]);
1863 				mtx_unlock(&hio_recv_list_lock[ncomp]);
1864 				continue;
1865 			default:
1866 				PJDLOG_ABORT("Invalid hio_countdown.");
1867 			}
1868 		}
1869 		if (ISSYNCREQ(hio)) {
1870 			mtx_lock(&sync_lock);
1871 			SYNCREQDONE(hio);
1872 			mtx_unlock(&sync_lock);
1873 			cv_signal(&sync_cond);
1874 		} else {
1875 			pjdlog_debug(2,
1876 			    "remote_recv: (%p) Moving request to the done queue.",
1877 			    hio);
1878 			QUEUE_INSERT2(hio, done);
1879 		}
1880 	}
1881 	/* NOTREACHED */
1882 	return (NULL);
1883 }
1884 
1885 /*
1886  * Thread sends answer to the kernel.
1887  */
1888 static void *
1889 ggate_send_thread(void *arg)
1890 {
1891 	struct hast_resource *res = arg;
1892 	struct g_gate_ctl_io *ggio;
1893 	struct hio *hio;
1894 	unsigned int ii, ncomps;
1895 
1896 	ncomps = HAST_NCOMPONENTS;
1897 
1898 	for (;;) {
1899 		pjdlog_debug(2, "ggate_send: Taking request.");
1900 		QUEUE_TAKE2(hio, done);
1901 		pjdlog_debug(2, "ggate_send: (%p) Got request.", hio);
1902 		ggio = &hio->hio_ggio;
1903 		for (ii = 0; ii < ncomps; ii++) {
1904 			if (hio->hio_errors[ii] == 0) {
1905 				/*
1906 				 * One successful request is enough to declare
1907 				 * success.
1908 				 */
1909 				ggio->gctl_error = 0;
1910 				break;
1911 			}
1912 		}
1913 		if (ii == ncomps) {
1914 			/*
1915 			 * None of the requests were successful.
1916 			 * Use the error from local component except the
1917 			 * case when we did only remote request.
1918 			 */
1919 			if (ggio->gctl_cmd == BIO_READ &&
1920 			    res->hr_syncsrc == HAST_SYNCSRC_SECONDARY)
1921 				ggio->gctl_error = hio->hio_errors[1];
1922 			else
1923 				ggio->gctl_error = hio->hio_errors[0];
1924 		}
1925 		if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) {
1926 			mtx_lock(&res->hr_amp_lock);
1927 			if (activemap_write_complete(res->hr_amp,
1928 			    ggio->gctl_offset, ggio->gctl_length)) {
1929 				res->hr_stat_activemap_update++;
1930 				(void)hast_activemap_flush(res);
1931 			} else {
1932 				mtx_unlock(&res->hr_amp_lock);
1933 			}
1934 		}
1935 		if (ggio->gctl_cmd == BIO_WRITE) {
1936 			/*
1937 			 * Unlock range we locked.
1938 			 */
1939 			mtx_lock(&range_lock);
1940 			rangelock_del(range_regular, ggio->gctl_offset,
1941 			    ggio->gctl_length);
1942 			if (range_sync_wait)
1943 				cv_signal(&range_sync_cond);
1944 			mtx_unlock(&range_lock);
1945 			if (!hio->hio_done)
1946 				write_complete(res, hio);
1947 		} else {
1948 			if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) == -1) {
1949 				primary_exit(EX_OSERR,
1950 				    "G_GATE_CMD_DONE failed");
1951 			}
1952 		}
1953 		if (hio->hio_errors[0]) {
1954 			switch (ggio->gctl_cmd) {
1955 			case BIO_READ:
1956 				res->hr_stat_read_error++;
1957 				break;
1958 			case BIO_WRITE:
1959 				res->hr_stat_write_error++;
1960 				break;
1961 			case BIO_DELETE:
1962 				res->hr_stat_delete_error++;
1963 				break;
1964 			case BIO_FLUSH:
1965 				res->hr_stat_flush_error++;
1966 				break;
1967 			}
1968 		}
1969 		pjdlog_debug(2,
1970 		    "ggate_send: (%p) Moving request to the free queue.", hio);
1971 		QUEUE_INSERT2(hio, free);
1972 	}
1973 	/* NOTREACHED */
1974 	return (NULL);
1975 }
1976 
1977 /*
1978  * Thread synchronize local and remote components.
1979  */
1980 static void *
1981 sync_thread(void *arg __unused)
1982 {
1983 	struct hast_resource *res = arg;
1984 	struct hio *hio;
1985 	struct g_gate_ctl_io *ggio;
1986 	struct timeval tstart, tend, tdiff;
1987 	unsigned int ii, ncomp, ncomps;
1988 	off_t offset, length, synced;
1989 	bool dorewind, directreads;
1990 	int syncext;
1991 
1992 	ncomps = HAST_NCOMPONENTS;
1993 	dorewind = true;
1994 	synced = 0;
1995 	offset = -1;
1996 	directreads = false;
1997 
1998 	for (;;) {
1999 		mtx_lock(&sync_lock);
2000 		if (offset >= 0 && !sync_inprogress) {
2001 			gettimeofday(&tend, NULL);
2002 			timersub(&tend, &tstart, &tdiff);
2003 			pjdlog_info("Synchronization interrupted after %#.0T. "
2004 			    "%NB synchronized so far.", &tdiff,
2005 			    (intmax_t)synced);
2006 			event_send(res, EVENT_SYNCINTR);
2007 		}
2008 		while (!sync_inprogress) {
2009 			dorewind = true;
2010 			synced = 0;
2011 			cv_wait(&sync_cond, &sync_lock);
2012 		}
2013 		mtx_unlock(&sync_lock);
2014 		/*
2015 		 * Obtain offset at which we should synchronize.
2016 		 * Rewind synchronization if needed.
2017 		 */
2018 		mtx_lock(&res->hr_amp_lock);
2019 		if (dorewind)
2020 			activemap_sync_rewind(res->hr_amp);
2021 		offset = activemap_sync_offset(res->hr_amp, &length, &syncext);
2022 		if (syncext != -1) {
2023 			/*
2024 			 * We synchronized entire syncext extent, we can mark
2025 			 * it as clean now.
2026 			 */
2027 			if (activemap_extent_complete(res->hr_amp, syncext))
2028 				(void)hast_activemap_flush(res);
2029 			else
2030 				mtx_unlock(&res->hr_amp_lock);
2031 		} else {
2032 			mtx_unlock(&res->hr_amp_lock);
2033 		}
2034 		if (dorewind) {
2035 			dorewind = false;
2036 			if (offset == -1)
2037 				pjdlog_info("Nodes are in sync.");
2038 			else {
2039 				pjdlog_info("Synchronization started. %NB to go.",
2040 				    (intmax_t)(res->hr_extentsize *
2041 				    activemap_ndirty(res->hr_amp)));
2042 				event_send(res, EVENT_SYNCSTART);
2043 				gettimeofday(&tstart, NULL);
2044 			}
2045 		}
2046 		if (offset == -1) {
2047 			sync_stop();
2048 			pjdlog_debug(1, "Nothing to synchronize.");
2049 			/*
2050 			 * Synchronization complete, make both localcnt and
2051 			 * remotecnt equal.
2052 			 */
2053 			ncomp = 1;
2054 			rw_rlock(&hio_remote_lock[ncomp]);
2055 			if (ISCONNECTED(res, ncomp)) {
2056 				if (synced > 0) {
2057 					int64_t bps;
2058 
2059 					gettimeofday(&tend, NULL);
2060 					timersub(&tend, &tstart, &tdiff);
2061 					bps = (int64_t)((double)synced /
2062 					    ((double)tdiff.tv_sec +
2063 					    (double)tdiff.tv_usec / 1000000));
2064 					pjdlog_info("Synchronization complete. "
2065 					    "%NB synchronized in %#.0lT (%NB/sec).",
2066 					    (intmax_t)synced, &tdiff,
2067 					    (intmax_t)bps);
2068 					event_send(res, EVENT_SYNCDONE);
2069 				}
2070 				mtx_lock(&metadata_lock);
2071 				if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY)
2072 					directreads = true;
2073 				res->hr_syncsrc = HAST_SYNCSRC_UNDEF;
2074 				res->hr_primary_localcnt =
2075 				    res->hr_secondary_remotecnt;
2076 				res->hr_primary_remotecnt =
2077 				    res->hr_secondary_localcnt;
2078 				pjdlog_debug(1,
2079 				    "Setting localcnt to %ju and remotecnt to %ju.",
2080 				    (uintmax_t)res->hr_primary_localcnt,
2081 				    (uintmax_t)res->hr_primary_remotecnt);
2082 				(void)metadata_write(res);
2083 				mtx_unlock(&metadata_lock);
2084 			}
2085 			rw_unlock(&hio_remote_lock[ncomp]);
2086 			if (directreads) {
2087 				directreads = false;
2088 				enable_direct_reads(res);
2089 			}
2090 			continue;
2091 		}
2092 		pjdlog_debug(2, "sync: Taking free request.");
2093 		QUEUE_TAKE2(hio, free);
2094 		pjdlog_debug(2, "sync: (%p) Got free request.", hio);
2095 		/*
2096 		 * Lock the range we are going to synchronize. We don't want
2097 		 * race where someone writes between our read and write.
2098 		 */
2099 		for (;;) {
2100 			mtx_lock(&range_lock);
2101 			if (rangelock_islocked(range_regular, offset, length)) {
2102 				pjdlog_debug(2,
2103 				    "sync: Range offset=%jd length=%jd locked.",
2104 				    (intmax_t)offset, (intmax_t)length);
2105 				range_sync_wait = true;
2106 				cv_wait(&range_sync_cond, &range_lock);
2107 				range_sync_wait = false;
2108 				mtx_unlock(&range_lock);
2109 				continue;
2110 			}
2111 			if (rangelock_add(range_sync, offset, length) == -1) {
2112 				mtx_unlock(&range_lock);
2113 				pjdlog_debug(2,
2114 				    "sync: Range offset=%jd length=%jd is already locked, waiting.",
2115 				    (intmax_t)offset, (intmax_t)length);
2116 				sleep(1);
2117 				continue;
2118 			}
2119 			mtx_unlock(&range_lock);
2120 			break;
2121 		}
2122 		/*
2123 		 * First read the data from synchronization source.
2124 		 */
2125 		SYNCREQ(hio);
2126 		ggio = &hio->hio_ggio;
2127 		ggio->gctl_cmd = BIO_READ;
2128 		ggio->gctl_offset = offset;
2129 		ggio->gctl_length = length;
2130 		ggio->gctl_error = 0;
2131 		hio->hio_done = false;
2132 		hio->hio_replication = res->hr_replication;
2133 		for (ii = 0; ii < ncomps; ii++)
2134 			hio->hio_errors[ii] = EINVAL;
2135 		reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ",
2136 		    hio);
2137 		pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2138 		    hio);
2139 		mtx_lock(&metadata_lock);
2140 		if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
2141 			/*
2142 			 * This range is up-to-date on local component,
2143 			 * so handle request locally.
2144 			 */
2145 			 /* Local component is 0 for now. */
2146 			ncomp = 0;
2147 		} else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ {
2148 			PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY);
2149 			/*
2150 			 * This range is out-of-date on local component,
2151 			 * so send request to the remote node.
2152 			 */
2153 			 /* Remote component is 1 for now. */
2154 			ncomp = 1;
2155 		}
2156 		mtx_unlock(&metadata_lock);
2157 		refcnt_init(&hio->hio_countdown, 1);
2158 		QUEUE_INSERT1(hio, send, ncomp);
2159 
2160 		/*
2161 		 * Let's wait for READ to finish.
2162 		 */
2163 		mtx_lock(&sync_lock);
2164 		while (!ISSYNCREQDONE(hio))
2165 			cv_wait(&sync_cond, &sync_lock);
2166 		mtx_unlock(&sync_lock);
2167 
2168 		if (hio->hio_errors[ncomp] != 0) {
2169 			pjdlog_error("Unable to read synchronization data: %s.",
2170 			    strerror(hio->hio_errors[ncomp]));
2171 			goto free_queue;
2172 		}
2173 
2174 		/*
2175 		 * We read the data from synchronization source, now write it
2176 		 * to synchronization target.
2177 		 */
2178 		SYNCREQ(hio);
2179 		ggio->gctl_cmd = BIO_WRITE;
2180 		for (ii = 0; ii < ncomps; ii++)
2181 			hio->hio_errors[ii] = EINVAL;
2182 		reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ",
2183 		    hio);
2184 		pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2185 		    hio);
2186 		mtx_lock(&metadata_lock);
2187 		if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
2188 			/*
2189 			 * This range is up-to-date on local component,
2190 			 * so we update remote component.
2191 			 */
2192 			 /* Remote component is 1 for now. */
2193 			ncomp = 1;
2194 		} else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ {
2195 			PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY);
2196 			/*
2197 			 * This range is out-of-date on local component,
2198 			 * so we update it.
2199 			 */
2200 			 /* Local component is 0 for now. */
2201 			ncomp = 0;
2202 		}
2203 		mtx_unlock(&metadata_lock);
2204 
2205 		pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2206 		    hio);
2207 		refcnt_init(&hio->hio_countdown, 1);
2208 		QUEUE_INSERT1(hio, send, ncomp);
2209 
2210 		/*
2211 		 * Let's wait for WRITE to finish.
2212 		 */
2213 		mtx_lock(&sync_lock);
2214 		while (!ISSYNCREQDONE(hio))
2215 			cv_wait(&sync_cond, &sync_lock);
2216 		mtx_unlock(&sync_lock);
2217 
2218 		if (hio->hio_errors[ncomp] != 0) {
2219 			pjdlog_error("Unable to write synchronization data: %s.",
2220 			    strerror(hio->hio_errors[ncomp]));
2221 			goto free_queue;
2222 		}
2223 
2224 		synced += length;
2225 free_queue:
2226 		mtx_lock(&range_lock);
2227 		rangelock_del(range_sync, offset, length);
2228 		if (range_regular_wait)
2229 			cv_signal(&range_regular_cond);
2230 		mtx_unlock(&range_lock);
2231 		pjdlog_debug(2, "sync: (%p) Moving request to the free queue.",
2232 		    hio);
2233 		QUEUE_INSERT2(hio, free);
2234 	}
2235 	/* NOTREACHED */
2236 	return (NULL);
2237 }
2238 
2239 void
2240 primary_config_reload(struct hast_resource *res, struct nv *nv)
2241 {
2242 	unsigned int ii, ncomps;
2243 	int modified, vint;
2244 	const char *vstr;
2245 
2246 	pjdlog_info("Reloading configuration...");
2247 
2248 	PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
2249 	PJDLOG_ASSERT(gres == res);
2250 	nv_assert(nv, "remoteaddr");
2251 	nv_assert(nv, "sourceaddr");
2252 	nv_assert(nv, "replication");
2253 	nv_assert(nv, "checksum");
2254 	nv_assert(nv, "compression");
2255 	nv_assert(nv, "timeout");
2256 	nv_assert(nv, "exec");
2257 	nv_assert(nv, "metaflush");
2258 
2259 	ncomps = HAST_NCOMPONENTS;
2260 
2261 #define MODIFIED_REMOTEADDR	0x01
2262 #define MODIFIED_SOURCEADDR	0x02
2263 #define MODIFIED_REPLICATION	0x04
2264 #define MODIFIED_CHECKSUM	0x08
2265 #define MODIFIED_COMPRESSION	0x10
2266 #define MODIFIED_TIMEOUT	0x20
2267 #define MODIFIED_EXEC		0x40
2268 #define MODIFIED_METAFLUSH	0x80
2269 	modified = 0;
2270 
2271 	vstr = nv_get_string(nv, "remoteaddr");
2272 	if (strcmp(gres->hr_remoteaddr, vstr) != 0) {
2273 		/*
2274 		 * Don't copy res->hr_remoteaddr to gres just yet.
2275 		 * We want remote_close() to log disconnect from the old
2276 		 * addresses, not from the new ones.
2277 		 */
2278 		modified |= MODIFIED_REMOTEADDR;
2279 	}
2280 	vstr = nv_get_string(nv, "sourceaddr");
2281 	if (strcmp(gres->hr_sourceaddr, vstr) != 0) {
2282 		strlcpy(gres->hr_sourceaddr, vstr, sizeof(gres->hr_sourceaddr));
2283 		modified |= MODIFIED_SOURCEADDR;
2284 	}
2285 	vint = nv_get_int32(nv, "replication");
2286 	if (gres->hr_replication != vint) {
2287 		gres->hr_replication = vint;
2288 		modified |= MODIFIED_REPLICATION;
2289 	}
2290 	vint = nv_get_int32(nv, "checksum");
2291 	if (gres->hr_checksum != vint) {
2292 		gres->hr_checksum = vint;
2293 		modified |= MODIFIED_CHECKSUM;
2294 	}
2295 	vint = nv_get_int32(nv, "compression");
2296 	if (gres->hr_compression != vint) {
2297 		gres->hr_compression = vint;
2298 		modified |= MODIFIED_COMPRESSION;
2299 	}
2300 	vint = nv_get_int32(nv, "timeout");
2301 	if (gres->hr_timeout != vint) {
2302 		gres->hr_timeout = vint;
2303 		modified |= MODIFIED_TIMEOUT;
2304 	}
2305 	vstr = nv_get_string(nv, "exec");
2306 	if (strcmp(gres->hr_exec, vstr) != 0) {
2307 		strlcpy(gres->hr_exec, vstr, sizeof(gres->hr_exec));
2308 		modified |= MODIFIED_EXEC;
2309 	}
2310 	vint = nv_get_int32(nv, "metaflush");
2311 	if (gres->hr_metaflush != vint) {
2312 		gres->hr_metaflush = vint;
2313 		modified |= MODIFIED_METAFLUSH;
2314 	}
2315 
2316 	/*
2317 	 * Change timeout for connected sockets.
2318 	 * Don't bother if we need to reconnect.
2319 	 */
2320 	if ((modified & MODIFIED_TIMEOUT) != 0 &&
2321 	    (modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR)) == 0) {
2322 		for (ii = 0; ii < ncomps; ii++) {
2323 			if (!ISREMOTE(ii))
2324 				continue;
2325 			rw_rlock(&hio_remote_lock[ii]);
2326 			if (!ISCONNECTED(gres, ii)) {
2327 				rw_unlock(&hio_remote_lock[ii]);
2328 				continue;
2329 			}
2330 			rw_unlock(&hio_remote_lock[ii]);
2331 			if (proto_timeout(gres->hr_remotein,
2332 			    gres->hr_timeout) == -1) {
2333 				pjdlog_errno(LOG_WARNING,
2334 				    "Unable to set connection timeout");
2335 			}
2336 			if (proto_timeout(gres->hr_remoteout,
2337 			    gres->hr_timeout) == -1) {
2338 				pjdlog_errno(LOG_WARNING,
2339 				    "Unable to set connection timeout");
2340 			}
2341 		}
2342 	}
2343 	if ((modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR)) != 0) {
2344 		for (ii = 0; ii < ncomps; ii++) {
2345 			if (!ISREMOTE(ii))
2346 				continue;
2347 			remote_close(gres, ii);
2348 		}
2349 		if (modified & MODIFIED_REMOTEADDR) {
2350 			vstr = nv_get_string(nv, "remoteaddr");
2351 			strlcpy(gres->hr_remoteaddr, vstr,
2352 			    sizeof(gres->hr_remoteaddr));
2353 		}
2354 	}
2355 #undef	MODIFIED_REMOTEADDR
2356 #undef	MODIFIED_SOURCEADDR
2357 #undef	MODIFIED_REPLICATION
2358 #undef	MODIFIED_CHECKSUM
2359 #undef	MODIFIED_COMPRESSION
2360 #undef	MODIFIED_TIMEOUT
2361 #undef	MODIFIED_EXEC
2362 #undef	MODIFIED_METAFLUSH
2363 
2364 	pjdlog_info("Configuration reloaded successfully.");
2365 }
2366 
2367 static void
2368 guard_one(struct hast_resource *res, unsigned int ncomp)
2369 {
2370 	struct proto_conn *in, *out;
2371 
2372 	if (!ISREMOTE(ncomp))
2373 		return;
2374 
2375 	rw_rlock(&hio_remote_lock[ncomp]);
2376 
2377 	if (!real_remote(res)) {
2378 		rw_unlock(&hio_remote_lock[ncomp]);
2379 		return;
2380 	}
2381 
2382 	if (ISCONNECTED(res, ncomp)) {
2383 		PJDLOG_ASSERT(res->hr_remotein != NULL);
2384 		PJDLOG_ASSERT(res->hr_remoteout != NULL);
2385 		rw_unlock(&hio_remote_lock[ncomp]);
2386 		pjdlog_debug(2, "remote_guard: Connection to %s is ok.",
2387 		    res->hr_remoteaddr);
2388 		return;
2389 	}
2390 
2391 	PJDLOG_ASSERT(res->hr_remotein == NULL);
2392 	PJDLOG_ASSERT(res->hr_remoteout == NULL);
2393 	/*
2394 	 * Upgrade the lock. It doesn't have to be atomic as no other thread
2395 	 * can change connection status from disconnected to connected.
2396 	 */
2397 	rw_unlock(&hio_remote_lock[ncomp]);
2398 	pjdlog_debug(2, "remote_guard: Reconnecting to %s.",
2399 	    res->hr_remoteaddr);
2400 	in = out = NULL;
2401 	if (init_remote(res, &in, &out) == 0) {
2402 		rw_wlock(&hio_remote_lock[ncomp]);
2403 		PJDLOG_ASSERT(res->hr_remotein == NULL);
2404 		PJDLOG_ASSERT(res->hr_remoteout == NULL);
2405 		PJDLOG_ASSERT(in != NULL && out != NULL);
2406 		res->hr_remotein = in;
2407 		res->hr_remoteout = out;
2408 		rw_unlock(&hio_remote_lock[ncomp]);
2409 		pjdlog_info("Successfully reconnected to %s.",
2410 		    res->hr_remoteaddr);
2411 		sync_start();
2412 	} else {
2413 		/* Both connections should be NULL. */
2414 		PJDLOG_ASSERT(res->hr_remotein == NULL);
2415 		PJDLOG_ASSERT(res->hr_remoteout == NULL);
2416 		PJDLOG_ASSERT(in == NULL && out == NULL);
2417 		pjdlog_debug(2, "remote_guard: Reconnect to %s failed.",
2418 		    res->hr_remoteaddr);
2419 	}
2420 }
2421 
2422 /*
2423  * Thread guards remote connections and reconnects when needed, handles
2424  * signals, etc.
2425  */
2426 static void *
2427 guard_thread(void *arg)
2428 {
2429 	struct hast_resource *res = arg;
2430 	unsigned int ii, ncomps;
2431 	struct timespec timeout;
2432 	time_t lastcheck, now;
2433 	sigset_t mask;
2434 	int signo;
2435 
2436 	ncomps = HAST_NCOMPONENTS;
2437 	lastcheck = time(NULL);
2438 
2439 	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
2440 	PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
2441 	PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
2442 
2443 	timeout.tv_sec = HAST_KEEPALIVE;
2444 	timeout.tv_nsec = 0;
2445 	signo = -1;
2446 
2447 	for (;;) {
2448 		switch (signo) {
2449 		case SIGINT:
2450 		case SIGTERM:
2451 			sigexit_received = true;
2452 			primary_exitx(EX_OK,
2453 			    "Termination signal received, exiting.");
2454 			break;
2455 		default:
2456 			break;
2457 		}
2458 
2459 		/*
2460 		 * Don't check connections until we fully started,
2461 		 * as we may still be looping, waiting for remote node
2462 		 * to switch from primary to secondary.
2463 		 */
2464 		if (fullystarted) {
2465 			pjdlog_debug(2, "remote_guard: Checking connections.");
2466 			now = time(NULL);
2467 			if (lastcheck + HAST_KEEPALIVE <= now) {
2468 				for (ii = 0; ii < ncomps; ii++)
2469 					guard_one(res, ii);
2470 				lastcheck = now;
2471 			}
2472 		}
2473 		signo = sigtimedwait(&mask, NULL, &timeout);
2474 	}
2475 	/* NOTREACHED */
2476 	return (NULL);
2477 }
2478