xref: /freebsd/sbin/hastd/secondary.c (revision a0409676120c1e558d0ade943019934e0f15118d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2009-2010 The FreeBSD Foundation
5  * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
6  * All rights reserved.
7  *
8  * This software was developed by Pawel Jakub Dawidek under sponsorship from
9  * the FreeBSD Foundation.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include <sys/param.h>
37 #include <sys/time.h>
38 #include <sys/bio.h>
39 #include <sys/disk.h>
40 #include <sys/stat.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 <pjdlog.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 "subr.h"
67 #include "synch.h"
68 
69 struct hio {
70 	uint64_t	 hio_seq;
71 	int		 hio_error;
72 	void		*hio_data;
73 	uint8_t		 hio_cmd;
74 	uint64_t	 hio_offset;
75 	uint64_t	 hio_length;
76 	bool		 hio_memsync;
77 	TAILQ_ENTRY(hio) hio_next;
78 };
79 
80 static struct hast_resource *gres;
81 
82 /*
83  * Free list holds unused structures. When free list is empty, we have to wait
84  * until some in-progress requests are freed.
85  */
86 static TAILQ_HEAD(, hio) hio_free_list;
87 static size_t hio_free_list_size;
88 static pthread_mutex_t hio_free_list_lock;
89 static pthread_cond_t hio_free_list_cond;
90 /*
91  * Disk thread (the one that does I/O requests) takes requests from this list.
92  */
93 static TAILQ_HEAD(, hio) hio_disk_list;
94 static size_t hio_disk_list_size;
95 static pthread_mutex_t hio_disk_list_lock;
96 static pthread_cond_t hio_disk_list_cond;
97 /*
98  * Thread that sends requests back to primary takes requests from this list.
99  */
100 static TAILQ_HEAD(, hio) hio_send_list;
101 static size_t hio_send_list_size;
102 static pthread_mutex_t hio_send_list_lock;
103 static pthread_cond_t hio_send_list_cond;
104 
105 /*
106  * Maximum number of outstanding I/O requests.
107  */
108 #define	HAST_HIO_MAX	256
109 
110 static void *recv_thread(void *arg);
111 static void *disk_thread(void *arg);
112 static void *send_thread(void *arg);
113 
114 #define	QUEUE_INSERT(name, hio)	do {					\
115 	mtx_lock(&hio_##name##_list_lock);				\
116 	if (TAILQ_EMPTY(&hio_##name##_list))				\
117 		cv_broadcast(&hio_##name##_list_cond);			\
118 	TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next);		\
119 	hio_##name##_list_size++;					\
120 	mtx_unlock(&hio_##name##_list_lock);				\
121 } while (0)
122 #define	QUEUE_TAKE(name, hio)	do {					\
123 	mtx_lock(&hio_##name##_list_lock);				\
124 	while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) {	\
125 		cv_wait(&hio_##name##_list_cond,			\
126 		    &hio_##name##_list_lock);				\
127 	}								\
128 	PJDLOG_ASSERT(hio_##name##_list_size != 0);			\
129 	hio_##name##_list_size--;					\
130 	TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next);		\
131 	mtx_unlock(&hio_##name##_list_lock);				\
132 } while (0)
133 
134 static void
135 output_status_aux(struct nv *nvout)
136 {
137 
138 	nv_add_uint64(nvout, (uint64_t)hio_free_list_size, "idle_queue_size");
139 	nv_add_uint64(nvout, (uint64_t)hio_disk_list_size, "local_queue_size");
140 	nv_add_uint64(nvout, (uint64_t)hio_send_list_size, "send_queue_size");
141 }
142 
143 static void
144 hio_clear(struct hio *hio)
145 {
146 
147 	hio->hio_seq = 0;
148 	hio->hio_error = 0;
149 	hio->hio_cmd = HIO_UNDEF;
150 	hio->hio_offset = 0;
151 	hio->hio_length = 0;
152 	hio->hio_memsync = false;
153 }
154 
155 static void
156 hio_copy(const struct hio *srchio, struct hio *dsthio)
157 {
158 
159 	/*
160 	 * We don't copy hio_error, hio_data and hio_next fields.
161 	 */
162 
163 	dsthio->hio_seq = srchio->hio_seq;
164 	dsthio->hio_cmd = srchio->hio_cmd;
165 	dsthio->hio_offset = srchio->hio_offset;
166 	dsthio->hio_length = srchio->hio_length;
167 	dsthio->hio_memsync = srchio->hio_memsync;
168 }
169 
170 static void
171 init_environment(void)
172 {
173 	struct hio *hio;
174 	unsigned int ii;
175 
176 	/*
177 	 * Initialize lists, their locks and theirs condition variables.
178 	 */
179 	TAILQ_INIT(&hio_free_list);
180 	mtx_init(&hio_free_list_lock);
181 	cv_init(&hio_free_list_cond);
182 	TAILQ_INIT(&hio_disk_list);
183 	mtx_init(&hio_disk_list_lock);
184 	cv_init(&hio_disk_list_cond);
185 	TAILQ_INIT(&hio_send_list);
186 	mtx_init(&hio_send_list_lock);
187 	cv_init(&hio_send_list_cond);
188 
189 	/*
190 	 * Allocate requests pool and initialize requests.
191 	 */
192 	for (ii = 0; ii < HAST_HIO_MAX; ii++) {
193 		hio = malloc(sizeof(*hio));
194 		if (hio == NULL) {
195 			pjdlog_exitx(EX_TEMPFAIL,
196 			    "Unable to allocate memory (%zu bytes) for hio request.",
197 			    sizeof(*hio));
198 		}
199 		hio->hio_data = malloc(MAXPHYS);
200 		if (hio->hio_data == NULL) {
201 			pjdlog_exitx(EX_TEMPFAIL,
202 			    "Unable to allocate memory (%zu bytes) for gctl_data.",
203 			    (size_t)MAXPHYS);
204 		}
205 		hio_clear(hio);
206 		TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);
207 		hio_free_list_size++;
208 	}
209 }
210 
211 static void
212 init_local(struct hast_resource *res)
213 {
214 
215 	if (metadata_read(res, true) == -1)
216 		exit(EX_NOINPUT);
217 }
218 
219 static void
220 init_remote(struct hast_resource *res, struct nv *nvin)
221 {
222 	uint64_t resuid;
223 	struct nv *nvout;
224 	unsigned char *map;
225 	size_t mapsize;
226 
227 #ifdef notyet
228 	/* Setup direction. */
229 	if (proto_send(res->hr_remoteout, NULL, 0) == -1)
230 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
231 #endif
232 
233 	nvout = nv_alloc();
234 	nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize");
235 	nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize");
236 	resuid = nv_get_uint64(nvin, "resuid");
237 	res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt");
238 	res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt");
239 	nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt");
240 	nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt");
241 	mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
242 	    METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize);
243 	map = malloc(mapsize);
244 	if (map == NULL) {
245 		pjdlog_exitx(EX_TEMPFAIL,
246 		    "Unable to allocate memory (%zu bytes) for activemap.",
247 		    mapsize);
248 	}
249 	/*
250 	 * When we work as primary and secondary is missing we will increase
251 	 * localcnt in our metadata. When secondary is connected and synced
252 	 * we make localcnt be equal to remotecnt, which means nodes are more
253 	 * or less in sync.
254 	 * Split-brain condition is when both nodes are not able to communicate
255 	 * and are both configured as primary nodes. In turn, they can both
256 	 * make incompatible changes to the data and we have to detect that.
257 	 * Under split-brain condition we will increase our localcnt on first
258 	 * write and remote node will increase its localcnt on first write.
259 	 * When we connect we can see that primary's localcnt is greater than
260 	 * our remotecnt (primary was modified while we weren't watching) and
261 	 * our localcnt is greater than primary's remotecnt (we were modified
262 	 * while primary wasn't watching).
263 	 * There are many possible combinations which are all gathered below.
264 	 * Don't pay too much attention to exact numbers, the more important
265 	 * is to compare them. We compare secondary's local with primary's
266 	 * remote and secondary's remote with primary's local.
267 	 * Note that every case where primary's localcnt is smaller than
268 	 * secondary's remotecnt and where secondary's localcnt is smaller than
269 	 * primary's remotecnt should be impossible in practise. We will perform
270 	 * full synchronization then. Those cases are marked with an asterisk.
271 	 * Regular synchronization means that only extents marked as dirty are
272 	 * synchronized (regular synchronization).
273 	 *
274 	 * SECONDARY METADATA PRIMARY METADATA
275 	 * local=3 remote=3   local=2 remote=2*  ?! Full sync from secondary.
276 	 * local=3 remote=3   local=2 remote=3*  ?! Full sync from primary.
277 	 * local=3 remote=3   local=2 remote=4*  ?! Full sync from primary.
278 	 * local=3 remote=3   local=3 remote=2   Primary is out-of-date,
279 	 *                                       regular sync from secondary.
280 	 * local=3 remote=3   local=3 remote=3   Regular sync just in case.
281 	 * local=3 remote=3   local=3 remote=4*  ?! Full sync from primary.
282 	 * local=3 remote=3   local=4 remote=2   Split-brain condition.
283 	 * local=3 remote=3   local=4 remote=3   Secondary out-of-date,
284 	 *                                       regular sync from primary.
285 	 * local=3 remote=3   local=4 remote=4*  ?! Full sync from primary.
286 	 */
287 	if (res->hr_resuid == 0) {
288 		/*
289 		 * Provider is used for the first time. If primary node done no
290 		 * writes yet as well (we will find "virgin" argument) then
291 		 * there is no need to synchronize anything. If primary node
292 		 * done any writes already we have to synchronize everything.
293 		 */
294 		PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
295 		res->hr_resuid = resuid;
296 		if (metadata_write(res) == -1)
297 			exit(EX_NOINPUT);
298 		if (nv_exists(nvin, "virgin")) {
299 			free(map);
300 			map = NULL;
301 			mapsize = 0;
302 		} else {
303 			memset(map, 0xff, mapsize);
304 		}
305 		nv_add_int8(nvout, 1, "virgin");
306 		nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
307 	} else if (res->hr_resuid != resuid) {
308 		char errmsg[256];
309 
310 		free(map);
311 		(void)snprintf(errmsg, sizeof(errmsg),
312 		    "Resource unique ID mismatch (primary=%ju, secondary=%ju).",
313 		    (uintmax_t)resuid, (uintmax_t)res->hr_resuid);
314 		pjdlog_error("%s", errmsg);
315 		nv_add_string(nvout, errmsg, "errmsg");
316 		if (hast_proto_send(res, res->hr_remotein, nvout,
317 		    NULL, 0) == -1) {
318 			pjdlog_exit(EX_TEMPFAIL,
319 			    "Unable to send response to %s",
320 			    res->hr_remoteaddr);
321 		}
322 		nv_free(nvout);
323 		exit(EX_CONFIG);
324 	} else if (
325 	    /* Is primary out-of-date? */
326 	    (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
327 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
328 	    /* Are the nodes more or less in sync? */
329 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
330 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
331 	    /* Is secondary out-of-date? */
332 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
333 	     res->hr_secondary_remotecnt < res->hr_primary_localcnt)) {
334 		/*
335 		 * Nodes are more or less in sync or one of the nodes is
336 		 * out-of-date.
337 		 * It doesn't matter at this point which one, we just have to
338 		 * send out local bitmap to the remote node.
339 		 */
340 		if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) !=
341 		    (ssize_t)mapsize) {
342 			pjdlog_exit(LOG_ERR, "Unable to read activemap");
343 		}
344 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
345 		     res->hr_secondary_remotecnt == res->hr_primary_localcnt) {
346 			/* Primary is out-of-date, sync from secondary. */
347 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
348 		} else {
349 			/*
350 			 * Secondary is out-of-date or counts match.
351 			 * Sync from primary.
352 			 */
353 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
354 		}
355 	} else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
356 	     res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
357 		/*
358 		 * Not good, we have split-brain condition.
359 		 */
360 		free(map);
361 		pjdlog_error("Split-brain detected, exiting.");
362 		nv_add_string(nvout, "Split-brain condition!", "errmsg");
363 		if (hast_proto_send(res, res->hr_remotein, nvout,
364 		    NULL, 0) == -1) {
365 			pjdlog_exit(EX_TEMPFAIL,
366 			    "Unable to send response to %s",
367 			    res->hr_remoteaddr);
368 		}
369 		nv_free(nvout);
370 		/* Exit on split-brain. */
371 		event_send(res, EVENT_SPLITBRAIN);
372 		exit(EX_CONFIG);
373 	} else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
374 	    res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ {
375 		/*
376 		 * This should never happen in practise, but we will perform
377 		 * full synchronization.
378 		 */
379 		PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
380 		    res->hr_primary_localcnt < res->hr_secondary_remotecnt);
381 		mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
382 		    METADATA_SIZE, res->hr_extentsize,
383 		    res->hr_local_sectorsize);
384 		memset(map, 0xff, mapsize);
385 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) {
386 			/* In this one of five cases sync from secondary. */
387 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
388 		} else {
389 			/* For the rest four cases sync from primary. */
390 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
391 		}
392 		pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).",
393 		    (uintmax_t)res->hr_primary_localcnt,
394 		    (uintmax_t)res->hr_primary_remotecnt,
395 		    (uintmax_t)res->hr_secondary_localcnt,
396 		    (uintmax_t)res->hr_secondary_remotecnt);
397 	}
398 	nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize");
399 	if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) {
400 		pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s",
401 		    res->hr_remoteaddr);
402 	}
403 	if (map != NULL)
404 		free(map);
405 	nv_free(nvout);
406 #ifdef notyet
407 	/* Setup direction. */
408 	if (proto_recv(res->hr_remotein, NULL, 0) == -1)
409 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
410 #endif
411 }
412 
413 void
414 hastd_secondary(struct hast_resource *res, struct nv *nvin)
415 {
416 	sigset_t mask;
417 	pthread_t td;
418 	pid_t pid;
419 	int error, mode, debuglevel;
420 
421 	/*
422 	 * Create communication channel between parent and child.
423 	 */
424 	if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) {
425 		KEEP_ERRNO((void)pidfile_remove(pfh));
426 		pjdlog_exit(EX_OSERR,
427 		    "Unable to create control sockets between parent and child");
428 	}
429 	/*
430 	 * Create communication channel between child and parent.
431 	 */
432 	if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) {
433 		KEEP_ERRNO((void)pidfile_remove(pfh));
434 		pjdlog_exit(EX_OSERR,
435 		    "Unable to create event sockets between child and parent");
436 	}
437 
438 	pid = fork();
439 	if (pid == -1) {
440 		KEEP_ERRNO((void)pidfile_remove(pfh));
441 		pjdlog_exit(EX_OSERR, "Unable to fork");
442 	}
443 
444 	if (pid > 0) {
445 		/* This is parent. */
446 		proto_close(res->hr_remotein);
447 		res->hr_remotein = NULL;
448 		proto_close(res->hr_remoteout);
449 		res->hr_remoteout = NULL;
450 		/* Declare that we are receiver. */
451 		proto_recv(res->hr_event, NULL, 0);
452 		/* Declare that we are sender. */
453 		proto_send(res->hr_ctrl, NULL, 0);
454 		res->hr_workerpid = pid;
455 		return;
456 	}
457 
458 	gres = res;
459 	res->output_status_aux = output_status_aux;
460 	mode = pjdlog_mode_get();
461 	debuglevel = pjdlog_debug_get();
462 
463 	/* Declare that we are sender. */
464 	proto_send(res->hr_event, NULL, 0);
465 	/* Declare that we are receiver. */
466 	proto_recv(res->hr_ctrl, NULL, 0);
467 	descriptors_cleanup(res);
468 
469 	descriptors_assert(res, mode);
470 
471 	pjdlog_init(mode);
472 	pjdlog_debug_set(debuglevel);
473 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
474 	setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
475 
476 	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
477 	PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
478 
479 	/* Error in setting timeout is not critical, but why should it fail? */
480 	if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1)
481 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
482 	if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1)
483 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
484 
485 	init_local(res);
486 	init_environment();
487 
488 	if (drop_privs(res) != 0)
489 		exit(EX_CONFIG);
490 	pjdlog_info("Privileges successfully dropped.");
491 
492 	/*
493 	 * Create the control thread before sending any event to the parent,
494 	 * as we can deadlock when parent sends control request to worker,
495 	 * but worker has no control thread started yet, so parent waits.
496 	 * In the meantime worker sends an event to the parent, but parent
497 	 * is unable to handle the event, because it waits for control
498 	 * request response.
499 	 */
500 	error = pthread_create(&td, NULL, ctrl_thread, res);
501 	PJDLOG_ASSERT(error == 0);
502 
503 	init_remote(res, nvin);
504 	event_send(res, EVENT_CONNECT);
505 
506 	error = pthread_create(&td, NULL, recv_thread, res);
507 	PJDLOG_ASSERT(error == 0);
508 	error = pthread_create(&td, NULL, disk_thread, res);
509 	PJDLOG_ASSERT(error == 0);
510 	(void)send_thread(res);
511 }
512 
513 static void
514 reqlog(int loglevel, int debuglevel, int error, struct hio *hio,
515     const char *fmt, ...)
516 {
517 	char msg[1024];
518 	va_list ap;
519 	int len;
520 
521 	va_start(ap, fmt);
522 	len = vsnprintf(msg, sizeof(msg), fmt, ap);
523 	va_end(ap);
524 	if ((size_t)len < sizeof(msg)) {
525 		switch (hio->hio_cmd) {
526 		case HIO_READ:
527 			(void)snprintf(msg + len, sizeof(msg) - len,
528 			    "READ(%ju, %ju).", (uintmax_t)hio->hio_offset,
529 			    (uintmax_t)hio->hio_length);
530 			break;
531 		case HIO_DELETE:
532 			(void)snprintf(msg + len, sizeof(msg) - len,
533 			    "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset,
534 			    (uintmax_t)hio->hio_length);
535 			break;
536 		case HIO_FLUSH:
537 			(void)snprintf(msg + len, sizeof(msg) - len, "FLUSH.");
538 			break;
539 		case HIO_WRITE:
540 			(void)snprintf(msg + len, sizeof(msg) - len,
541 			    "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset,
542 			    (uintmax_t)hio->hio_length);
543 			break;
544 		case HIO_KEEPALIVE:
545 			(void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE.");
546 			break;
547 		default:
548 			(void)snprintf(msg + len, sizeof(msg) - len,
549 			    "UNKNOWN(%u).", (unsigned int)hio->hio_cmd);
550 			break;
551 		}
552 	}
553 	pjdlog_common(loglevel, debuglevel, error, "%s", msg);
554 }
555 
556 static int
557 requnpack(struct hast_resource *res, struct hio *hio, struct nv *nv)
558 {
559 
560 	hio->hio_cmd = nv_get_uint8(nv, "cmd");
561 	if (hio->hio_cmd == 0) {
562 		pjdlog_error("Header contains no 'cmd' field.");
563 		hio->hio_error = EINVAL;
564 		goto end;
565 	}
566 	if (hio->hio_cmd != HIO_KEEPALIVE) {
567 		hio->hio_seq = nv_get_uint64(nv, "seq");
568 		if (hio->hio_seq == 0) {
569 			pjdlog_error("Header contains no 'seq' field.");
570 			hio->hio_error = EINVAL;
571 			goto end;
572 		}
573 	}
574 	switch (hio->hio_cmd) {
575 	case HIO_FLUSH:
576 	case HIO_KEEPALIVE:
577 		break;
578 	case HIO_WRITE:
579 		hio->hio_memsync = nv_exists(nv, "memsync");
580 		/* FALLTHROUGH */
581 	case HIO_READ:
582 	case HIO_DELETE:
583 		hio->hio_offset = nv_get_uint64(nv, "offset");
584 		if (nv_error(nv) != 0) {
585 			pjdlog_error("Header is missing 'offset' field.");
586 			hio->hio_error = EINVAL;
587 			goto end;
588 		}
589 		hio->hio_length = nv_get_uint64(nv, "length");
590 		if (nv_error(nv) != 0) {
591 			pjdlog_error("Header is missing 'length' field.");
592 			hio->hio_error = EINVAL;
593 			goto end;
594 		}
595 		if (hio->hio_length == 0) {
596 			pjdlog_error("Data length is zero.");
597 			hio->hio_error = EINVAL;
598 			goto end;
599 		}
600 		if (hio->hio_cmd != HIO_DELETE && hio->hio_length > MAXPHYS) {
601 			pjdlog_error("Data length is too large (%ju > %ju).",
602 			    (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);
603 			hio->hio_error = EINVAL;
604 			goto end;
605 		}
606 		if ((hio->hio_offset % res->hr_local_sectorsize) != 0) {
607 			pjdlog_error("Offset %ju is not multiple of sector size.",
608 			    (uintmax_t)hio->hio_offset);
609 			hio->hio_error = EINVAL;
610 			goto end;
611 		}
612 		if ((hio->hio_length % res->hr_local_sectorsize) != 0) {
613 			pjdlog_error("Length %ju is not multiple of sector size.",
614 			    (uintmax_t)hio->hio_length);
615 			hio->hio_error = EINVAL;
616 			goto end;
617 		}
618 		if (hio->hio_offset + hio->hio_length >
619 		    (uint64_t)res->hr_datasize) {
620 			pjdlog_error("Data offset is too large (%ju > %ju).",
621 			    (uintmax_t)(hio->hio_offset + hio->hio_length),
622 			    (uintmax_t)res->hr_datasize);
623 			hio->hio_error = EINVAL;
624 			goto end;
625 		}
626 		break;
627 	default:
628 		pjdlog_error("Header contains invalid 'cmd' (%hhu).",
629 		    hio->hio_cmd);
630 		hio->hio_error = EINVAL;
631 		goto end;
632 	}
633 	hio->hio_error = 0;
634 end:
635 	return (hio->hio_error);
636 }
637 
638 static __dead2 void
639 secondary_exit(int exitcode, const char *fmt, ...)
640 {
641 	va_list ap;
642 
643 	PJDLOG_ASSERT(exitcode != EX_OK);
644 	va_start(ap, fmt);
645 	pjdlogv_errno(LOG_ERR, fmt, ap);
646 	va_end(ap);
647 	event_send(gres, EVENT_DISCONNECT);
648 	exit(exitcode);
649 }
650 
651 /*
652  * Thread receives requests from the primary node.
653  */
654 static void *
655 recv_thread(void *arg)
656 {
657 	struct hast_resource *res = arg;
658 	struct hio *hio, *mshio;
659 	struct nv *nv;
660 
661 	for (;;) {
662 		pjdlog_debug(2, "recv: Taking free request.");
663 		QUEUE_TAKE(free, hio);
664 		pjdlog_debug(2, "recv: (%p) Got request.", hio);
665 		if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) {
666 			secondary_exit(EX_TEMPFAIL,
667 			    "Unable to receive request header");
668 		}
669 		if (requnpack(res, hio, nv) != 0) {
670 			nv_free(nv);
671 			pjdlog_debug(2,
672 			    "recv: (%p) Moving request to the send queue.",
673 			    hio);
674 			QUEUE_INSERT(send, hio);
675 			continue;
676 		}
677 		switch (hio->hio_cmd) {
678 		case HIO_READ:
679 			res->hr_stat_read++;
680 			break;
681 		case HIO_WRITE:
682 			res->hr_stat_write++;
683 			break;
684 		case HIO_DELETE:
685 			res->hr_stat_delete++;
686 			break;
687 		case HIO_FLUSH:
688 			res->hr_stat_flush++;
689 			break;
690 		case HIO_KEEPALIVE:
691 			break;
692 		default:
693 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
694 			    hio->hio_cmd);
695 		}
696 		reqlog(LOG_DEBUG, 2, -1, hio,
697 		    "recv: (%p) Got request header: ", hio);
698 		if (hio->hio_cmd == HIO_KEEPALIVE) {
699 			nv_free(nv);
700 			pjdlog_debug(2,
701 			    "recv: (%p) Moving request to the free queue.",
702 			    hio);
703 			hio_clear(hio);
704 			QUEUE_INSERT(free, hio);
705 			continue;
706 		} else if (hio->hio_cmd == HIO_WRITE) {
707 			if (hast_proto_recv_data(res, res->hr_remotein, nv,
708 			    hio->hio_data, MAXPHYS) == -1) {
709 				secondary_exit(EX_TEMPFAIL,
710 				    "Unable to receive request data");
711 			}
712 			if (hio->hio_memsync) {
713 				/*
714 				 * For memsync requests we expect two replies.
715 				 * Clone the hio so we can handle both of them.
716 				 */
717 				pjdlog_debug(2, "recv: Taking free request.");
718 				QUEUE_TAKE(free, mshio);
719 				pjdlog_debug(2, "recv: (%p) Got request.",
720 				    mshio);
721 				hio_copy(hio, mshio);
722 				mshio->hio_error = 0;
723 				/*
724 				 * We want to keep 'memsync' tag only on the
725 				 * request going onto send queue (mshio).
726 				 */
727 				hio->hio_memsync = false;
728 				pjdlog_debug(2,
729 				    "recv: (%p) Moving memsync request to the send queue.",
730 				    mshio);
731 				QUEUE_INSERT(send, mshio);
732 			}
733 		}
734 		nv_free(nv);
735 		pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.",
736 		    hio);
737 		QUEUE_INSERT(disk, hio);
738 	}
739 	/* NOTREACHED */
740 	return (NULL);
741 }
742 
743 /*
744  * Thread reads from or writes to local component and also handles DELETE and
745  * FLUSH requests.
746  */
747 static void *
748 disk_thread(void *arg)
749 {
750 	struct hast_resource *res = arg;
751 	struct hio *hio;
752 	ssize_t ret;
753 	bool clear_activemap, logerror;
754 
755 	clear_activemap = true;
756 
757 	for (;;) {
758 		pjdlog_debug(2, "disk: Taking request.");
759 		QUEUE_TAKE(disk, hio);
760 		while (clear_activemap) {
761 			unsigned char *map;
762 			size_t mapsize;
763 
764 			/*
765 			 * When first request is received, it means that primary
766 			 * already received our activemap, merged it and stored
767 			 * locally. We can now safely clear our activemap.
768 			 */
769 			mapsize =
770 			    activemap_calc_ondisk_size(res->hr_local_mediasize -
771 			    METADATA_SIZE, res->hr_extentsize,
772 			    res->hr_local_sectorsize);
773 			map = calloc(1, mapsize);
774 			if (map == NULL) {
775 				pjdlog_warning("Unable to allocate memory to clear local activemap.");
776 				break;
777 			}
778 			if (pwrite(res->hr_localfd, map, mapsize,
779 			    METADATA_SIZE) != (ssize_t)mapsize) {
780 				pjdlog_errno(LOG_WARNING,
781 				    "Unable to store cleared activemap");
782 				free(map);
783 				res->hr_stat_activemap_write_error++;
784 				break;
785 			}
786 			free(map);
787 			clear_activemap = false;
788 			pjdlog_debug(1, "Local activemap cleared.");
789 			break;
790 		}
791 		reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio);
792 		logerror = true;
793 		/* Handle the actual request. */
794 		switch (hio->hio_cmd) {
795 		case HIO_READ:
796 			ret = pread(res->hr_localfd, hio->hio_data,
797 			    hio->hio_length,
798 			    hio->hio_offset + res->hr_localoff);
799 			if (ret == -1)
800 				hio->hio_error = errno;
801 			else if (ret != (int64_t)hio->hio_length)
802 				hio->hio_error = EIO;
803 			else
804 				hio->hio_error = 0;
805 			break;
806 		case HIO_WRITE:
807 			ret = pwrite(res->hr_localfd, hio->hio_data,
808 			    hio->hio_length,
809 			    hio->hio_offset + res->hr_localoff);
810 			if (ret == -1)
811 				hio->hio_error = errno;
812 			else if (ret != (int64_t)hio->hio_length)
813 				hio->hio_error = EIO;
814 			else
815 				hio->hio_error = 0;
816 			break;
817 		case HIO_DELETE:
818 			ret = g_delete(res->hr_localfd,
819 			    hio->hio_offset + res->hr_localoff,
820 			    hio->hio_length);
821 			if (ret == -1)
822 				hio->hio_error = errno;
823 			else
824 				hio->hio_error = 0;
825 			break;
826 		case HIO_FLUSH:
827 			if (!res->hr_localflush) {
828 				ret = -1;
829 				hio->hio_error = EOPNOTSUPP;
830 				logerror = false;
831 				break;
832 			}
833 			ret = g_flush(res->hr_localfd);
834 			if (ret == -1) {
835 				if (errno == EOPNOTSUPP)
836 					res->hr_localflush = false;
837 				hio->hio_error = errno;
838 			} else {
839 				hio->hio_error = 0;
840 			}
841 			break;
842 		default:
843 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
844 			    hio->hio_cmd);
845 		}
846 		if (logerror && hio->hio_error != 0) {
847 			reqlog(LOG_ERR, 0, hio->hio_error, hio,
848 			    "Request failed: ");
849 		}
850 		pjdlog_debug(2, "disk: (%p) Moving request to the send queue.",
851 		    hio);
852 		QUEUE_INSERT(send, hio);
853 	}
854 	/* NOTREACHED */
855 	return (NULL);
856 }
857 
858 /*
859  * Thread sends requests back to primary node.
860  */
861 static void *
862 send_thread(void *arg)
863 {
864 	struct hast_resource *res = arg;
865 	struct nv *nvout;
866 	struct hio *hio;
867 	void *data;
868 	size_t length;
869 
870 	for (;;) {
871 		pjdlog_debug(2, "send: Taking request.");
872 		QUEUE_TAKE(send, hio);
873 		reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio);
874 		nvout = nv_alloc();
875 		/* Copy sequence number. */
876 		nv_add_uint64(nvout, hio->hio_seq, "seq");
877 		if (hio->hio_memsync) {
878 			PJDLOG_ASSERT(hio->hio_cmd == HIO_WRITE);
879 			nv_add_int8(nvout, 1, "received");
880 		}
881 		switch (hio->hio_cmd) {
882 		case HIO_READ:
883 			if (hio->hio_error == 0) {
884 				data = hio->hio_data;
885 				length = hio->hio_length;
886 				break;
887 			}
888 			/*
889 			 * We send no data in case of an error.
890 			 */
891 			/* FALLTHROUGH */
892 		case HIO_DELETE:
893 		case HIO_FLUSH:
894 		case HIO_WRITE:
895 			data = NULL;
896 			length = 0;
897 			break;
898 		default:
899 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
900 			    hio->hio_cmd);
901 		}
902 		if (hio->hio_error != 0) {
903 			switch (hio->hio_cmd) {
904 			case HIO_READ:
905 				res->hr_stat_read_error++;
906 				break;
907 			case HIO_WRITE:
908 				res->hr_stat_write_error++;
909 				break;
910 			case HIO_DELETE:
911 				res->hr_stat_delete_error++;
912 				break;
913 			case HIO_FLUSH:
914 				res->hr_stat_flush_error++;
915 				break;
916 			}
917 			nv_add_int16(nvout, hio->hio_error, "error");
918 		}
919 		if (hast_proto_send(res, res->hr_remoteout, nvout, data,
920 		    length) == -1) {
921 			secondary_exit(EX_TEMPFAIL, "Unable to send reply");
922 		}
923 		nv_free(nvout);
924 		pjdlog_debug(2, "send: (%p) Moving request to the free queue.",
925 		    hio);
926 		hio_clear(hio);
927 		QUEUE_INSERT(free, hio);
928 	}
929 	/* NOTREACHED */
930 	return (NULL);
931 }
932