xref: /freebsd/sbin/hastd/secondary.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include <sys/param.h>
35 #include <sys/time.h>
36 #include <sys/bio.h>
37 #include <sys/disk.h>
38 #include <sys/stat.h>
39 
40 #include <err.h>
41 #include <errno.h>
42 #include <fcntl.h>
43 #include <libgeom.h>
44 #include <pthread.h>
45 #include <signal.h>
46 #include <stdint.h>
47 #include <stdio.h>
48 #include <string.h>
49 #include <sysexits.h>
50 #include <unistd.h>
51 
52 #include <activemap.h>
53 #include <nv.h>
54 #include <pjdlog.h>
55 
56 #include "control.h"
57 #include "event.h"
58 #include "hast.h"
59 #include "hast_proto.h"
60 #include "hastd.h"
61 #include "hooks.h"
62 #include "metadata.h"
63 #include "proto.h"
64 #include "subr.h"
65 #include "synch.h"
66 
67 struct hio {
68 	uint64_t	 hio_seq;
69 	int		 hio_error;
70 	void		*hio_data;
71 	uint8_t		 hio_cmd;
72 	uint64_t	 hio_offset;
73 	uint64_t	 hio_length;
74 	bool		 hio_memsync;
75 	TAILQ_ENTRY(hio) hio_next;
76 };
77 
78 static struct hast_resource *gres;
79 
80 /*
81  * Free list holds unused structures. When free list is empty, we have to wait
82  * until some in-progress requests are freed.
83  */
84 static TAILQ_HEAD(, hio) hio_free_list;
85 static size_t hio_free_list_size;
86 static pthread_mutex_t hio_free_list_lock;
87 static pthread_cond_t hio_free_list_cond;
88 /*
89  * Disk thread (the one that does I/O requests) takes requests from this list.
90  */
91 static TAILQ_HEAD(, hio) hio_disk_list;
92 static size_t hio_disk_list_size;
93 static pthread_mutex_t hio_disk_list_lock;
94 static pthread_cond_t hio_disk_list_cond;
95 /*
96  * Thread that sends requests back to primary takes requests from this list.
97  */
98 static TAILQ_HEAD(, hio) hio_send_list;
99 static size_t hio_send_list_size;
100 static pthread_mutex_t hio_send_list_lock;
101 static pthread_cond_t hio_send_list_cond;
102 
103 /*
104  * Maximum number of outstanding I/O requests.
105  */
106 #define	HAST_HIO_MAX	256
107 
108 static void *recv_thread(void *arg);
109 static void *disk_thread(void *arg);
110 static void *send_thread(void *arg);
111 
112 #define	QUEUE_INSERT(name, hio)	do {					\
113 	mtx_lock(&hio_##name##_list_lock);				\
114 	if (TAILQ_EMPTY(&hio_##name##_list))				\
115 		cv_broadcast(&hio_##name##_list_cond);			\
116 	TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next);		\
117 	hio_##name##_list_size++;					\
118 	mtx_unlock(&hio_##name##_list_lock);				\
119 } while (0)
120 #define	QUEUE_TAKE(name, hio)	do {					\
121 	mtx_lock(&hio_##name##_list_lock);				\
122 	while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) {	\
123 		cv_wait(&hio_##name##_list_cond,			\
124 		    &hio_##name##_list_lock);				\
125 	}								\
126 	PJDLOG_ASSERT(hio_##name##_list_size != 0);			\
127 	hio_##name##_list_size--;					\
128 	TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next);		\
129 	mtx_unlock(&hio_##name##_list_lock);				\
130 } while (0)
131 
132 static void
133 output_status_aux(struct nv *nvout)
134 {
135 
136 	nv_add_uint64(nvout, (uint64_t)hio_free_list_size, "idle_queue_size");
137 	nv_add_uint64(nvout, (uint64_t)hio_disk_list_size, "local_queue_size");
138 	nv_add_uint64(nvout, (uint64_t)hio_send_list_size, "send_queue_size");
139 }
140 
141 static void
142 hio_clear(struct hio *hio)
143 {
144 
145 	hio->hio_seq = 0;
146 	hio->hio_error = 0;
147 	hio->hio_cmd = HIO_UNDEF;
148 	hio->hio_offset = 0;
149 	hio->hio_length = 0;
150 	hio->hio_memsync = false;
151 }
152 
153 static void
154 hio_copy(const struct hio *srchio, struct hio *dsthio)
155 {
156 
157 	/*
158 	 * We don't copy hio_error, hio_data and hio_next fields.
159 	 */
160 
161 	dsthio->hio_seq = srchio->hio_seq;
162 	dsthio->hio_cmd = srchio->hio_cmd;
163 	dsthio->hio_offset = srchio->hio_offset;
164 	dsthio->hio_length = srchio->hio_length;
165 	dsthio->hio_memsync = srchio->hio_memsync;
166 }
167 
168 static void
169 init_environment(void)
170 {
171 	struct hio *hio;
172 	unsigned int ii;
173 
174 	/*
175 	 * Initialize lists, their locks and theirs condition variables.
176 	 */
177 	TAILQ_INIT(&hio_free_list);
178 	mtx_init(&hio_free_list_lock);
179 	cv_init(&hio_free_list_cond);
180 	TAILQ_INIT(&hio_disk_list);
181 	mtx_init(&hio_disk_list_lock);
182 	cv_init(&hio_disk_list_cond);
183 	TAILQ_INIT(&hio_send_list);
184 	mtx_init(&hio_send_list_lock);
185 	cv_init(&hio_send_list_cond);
186 
187 	/*
188 	 * Allocate requests pool and initialize requests.
189 	 */
190 	for (ii = 0; ii < HAST_HIO_MAX; ii++) {
191 		hio = malloc(sizeof(*hio));
192 		if (hio == NULL) {
193 			pjdlog_exitx(EX_TEMPFAIL,
194 			    "Unable to allocate memory (%zu bytes) for hio request.",
195 			    sizeof(*hio));
196 		}
197 		hio->hio_data = malloc(MAXPHYS);
198 		if (hio->hio_data == NULL) {
199 			pjdlog_exitx(EX_TEMPFAIL,
200 			    "Unable to allocate memory (%zu bytes) for gctl_data.",
201 			    (size_t)MAXPHYS);
202 		}
203 		hio_clear(hio);
204 		TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);
205 		hio_free_list_size++;
206 	}
207 }
208 
209 static void
210 init_local(struct hast_resource *res)
211 {
212 
213 	if (metadata_read(res, true) == -1)
214 		exit(EX_NOINPUT);
215 }
216 
217 static void
218 init_remote(struct hast_resource *res, struct nv *nvin)
219 {
220 	uint64_t resuid;
221 	struct nv *nvout;
222 	unsigned char *map;
223 	size_t mapsize;
224 
225 #ifdef notyet
226 	/* Setup direction. */
227 	if (proto_send(res->hr_remoteout, NULL, 0) == -1)
228 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
229 #endif
230 
231 	nvout = nv_alloc();
232 	nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize");
233 	nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize");
234 	resuid = nv_get_uint64(nvin, "resuid");
235 	res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt");
236 	res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt");
237 	nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt");
238 	nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt");
239 	mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
240 	    METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize);
241 	map = malloc(mapsize);
242 	if (map == NULL) {
243 		pjdlog_exitx(EX_TEMPFAIL,
244 		    "Unable to allocate memory (%zu bytes) for activemap.",
245 		    mapsize);
246 	}
247 	/*
248 	 * When we work as primary and secondary is missing we will increase
249 	 * localcnt in our metadata. When secondary is connected and synced
250 	 * we make localcnt be equal to remotecnt, which means nodes are more
251 	 * or less in sync.
252 	 * Split-brain condition is when both nodes are not able to communicate
253 	 * and are both configured as primary nodes. In turn, they can both
254 	 * make incompatible changes to the data and we have to detect that.
255 	 * Under split-brain condition we will increase our localcnt on first
256 	 * write and remote node will increase its localcnt on first write.
257 	 * When we connect we can see that primary's localcnt is greater than
258 	 * our remotecnt (primary was modified while we weren't watching) and
259 	 * our localcnt is greater than primary's remotecnt (we were modified
260 	 * while primary wasn't watching).
261 	 * There are many possible combinations which are all gathered below.
262 	 * Don't pay too much attention to exact numbers, the more important
263 	 * is to compare them. We compare secondary's local with primary's
264 	 * remote and secondary's remote with primary's local.
265 	 * Note that every case where primary's localcnt is smaller than
266 	 * secondary's remotecnt and where secondary's localcnt is smaller than
267 	 * primary's remotecnt should be impossible in practise. We will perform
268 	 * full synchronization then. Those cases are marked with an asterisk.
269 	 * Regular synchronization means that only extents marked as dirty are
270 	 * synchronized (regular synchronization).
271 	 *
272 	 * SECONDARY METADATA PRIMARY METADATA
273 	 * local=3 remote=3   local=2 remote=2*  ?! Full sync from secondary.
274 	 * local=3 remote=3   local=2 remote=3*  ?! Full sync from primary.
275 	 * local=3 remote=3   local=2 remote=4*  ?! Full sync from primary.
276 	 * local=3 remote=3   local=3 remote=2   Primary is out-of-date,
277 	 *                                       regular sync from secondary.
278 	 * local=3 remote=3   local=3 remote=3   Regular sync just in case.
279 	 * local=3 remote=3   local=3 remote=4*  ?! Full sync from primary.
280 	 * local=3 remote=3   local=4 remote=2   Split-brain condition.
281 	 * local=3 remote=3   local=4 remote=3   Secondary out-of-date,
282 	 *                                       regular sync from primary.
283 	 * local=3 remote=3   local=4 remote=4*  ?! Full sync from primary.
284 	 */
285 	if (res->hr_resuid == 0) {
286 		/*
287 		 * Provider is used for the first time. If primary node done no
288 		 * writes yet as well (we will find "virgin" argument) then
289 		 * there is no need to synchronize anything. If primary node
290 		 * done any writes already we have to synchronize everything.
291 		 */
292 		PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
293 		res->hr_resuid = resuid;
294 		if (metadata_write(res) == -1)
295 			exit(EX_NOINPUT);
296 		if (nv_exists(nvin, "virgin")) {
297 			free(map);
298 			map = NULL;
299 			mapsize = 0;
300 		} else {
301 			memset(map, 0xff, mapsize);
302 		}
303 		nv_add_int8(nvout, 1, "virgin");
304 		nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
305 	} else if (res->hr_resuid != resuid) {
306 		char errmsg[256];
307 
308 		free(map);
309 		(void)snprintf(errmsg, sizeof(errmsg),
310 		    "Resource unique ID mismatch (primary=%ju, secondary=%ju).",
311 		    (uintmax_t)resuid, (uintmax_t)res->hr_resuid);
312 		pjdlog_error("%s", errmsg);
313 		nv_add_string(nvout, errmsg, "errmsg");
314 		if (hast_proto_send(res, res->hr_remotein, nvout,
315 		    NULL, 0) == -1) {
316 			pjdlog_exit(EX_TEMPFAIL,
317 			    "Unable to send response to %s",
318 			    res->hr_remoteaddr);
319 		}
320 		nv_free(nvout);
321 		exit(EX_CONFIG);
322 	} else if (
323 	    /* Is primary out-of-date? */
324 	    (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
325 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
326 	    /* Are the nodes more or less in sync? */
327 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
328 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
329 	    /* Is secondary out-of-date? */
330 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
331 	     res->hr_secondary_remotecnt < res->hr_primary_localcnt)) {
332 		/*
333 		 * Nodes are more or less in sync or one of the nodes is
334 		 * out-of-date.
335 		 * It doesn't matter at this point which one, we just have to
336 		 * send out local bitmap to the remote node.
337 		 */
338 		if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) !=
339 		    (ssize_t)mapsize) {
340 			pjdlog_exit(LOG_ERR, "Unable to read activemap");
341 		}
342 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
343 		     res->hr_secondary_remotecnt == res->hr_primary_localcnt) {
344 			/* Primary is out-of-date, sync from secondary. */
345 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
346 		} else {
347 			/*
348 			 * Secondary is out-of-date or counts match.
349 			 * Sync from primary.
350 			 */
351 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
352 		}
353 	} else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
354 	     res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
355 		/*
356 		 * Not good, we have split-brain condition.
357 		 */
358 		free(map);
359 		pjdlog_error("Split-brain detected, exiting.");
360 		nv_add_string(nvout, "Split-brain condition!", "errmsg");
361 		if (hast_proto_send(res, res->hr_remotein, nvout,
362 		    NULL, 0) == -1) {
363 			pjdlog_exit(EX_TEMPFAIL,
364 			    "Unable to send response to %s",
365 			    res->hr_remoteaddr);
366 		}
367 		nv_free(nvout);
368 		/* Exit on split-brain. */
369 		event_send(res, EVENT_SPLITBRAIN);
370 		exit(EX_CONFIG);
371 	} else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
372 	    res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ {
373 		/*
374 		 * This should never happen in practise, but we will perform
375 		 * full synchronization.
376 		 */
377 		PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
378 		    res->hr_primary_localcnt < res->hr_secondary_remotecnt);
379 		mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
380 		    METADATA_SIZE, res->hr_extentsize,
381 		    res->hr_local_sectorsize);
382 		memset(map, 0xff, mapsize);
383 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) {
384 			/* In this one of five cases sync from secondary. */
385 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
386 		} else {
387 			/* For the rest four cases sync from primary. */
388 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
389 		}
390 		pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).",
391 		    (uintmax_t)res->hr_primary_localcnt,
392 		    (uintmax_t)res->hr_primary_remotecnt,
393 		    (uintmax_t)res->hr_secondary_localcnt,
394 		    (uintmax_t)res->hr_secondary_remotecnt);
395 	}
396 	nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize");
397 	if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) {
398 		pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s",
399 		    res->hr_remoteaddr);
400 	}
401 	if (map != NULL)
402 		free(map);
403 	nv_free(nvout);
404 #ifdef notyet
405 	/* Setup direction. */
406 	if (proto_recv(res->hr_remotein, NULL, 0) == -1)
407 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
408 #endif
409 }
410 
411 void
412 hastd_secondary(struct hast_resource *res, struct nv *nvin)
413 {
414 	sigset_t mask;
415 	pthread_t td;
416 	pid_t pid;
417 	int error, mode, debuglevel;
418 
419 	/*
420 	 * Create communication channel between parent and child.
421 	 */
422 	if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) {
423 		KEEP_ERRNO((void)pidfile_remove(pfh));
424 		pjdlog_exit(EX_OSERR,
425 		    "Unable to create control sockets between parent and child");
426 	}
427 	/*
428 	 * Create communication channel between child and parent.
429 	 */
430 	if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) {
431 		KEEP_ERRNO((void)pidfile_remove(pfh));
432 		pjdlog_exit(EX_OSERR,
433 		    "Unable to create event sockets between child and parent");
434 	}
435 
436 	pid = fork();
437 	if (pid == -1) {
438 		KEEP_ERRNO((void)pidfile_remove(pfh));
439 		pjdlog_exit(EX_OSERR, "Unable to fork");
440 	}
441 
442 	if (pid > 0) {
443 		/* This is parent. */
444 		proto_close(res->hr_remotein);
445 		res->hr_remotein = NULL;
446 		proto_close(res->hr_remoteout);
447 		res->hr_remoteout = NULL;
448 		/* Declare that we are receiver. */
449 		proto_recv(res->hr_event, NULL, 0);
450 		/* Declare that we are sender. */
451 		proto_send(res->hr_ctrl, NULL, 0);
452 		res->hr_workerpid = pid;
453 		return;
454 	}
455 
456 	gres = res;
457 	res->output_status_aux = output_status_aux;
458 	mode = pjdlog_mode_get();
459 	debuglevel = pjdlog_debug_get();
460 
461 	/* Declare that we are sender. */
462 	proto_send(res->hr_event, NULL, 0);
463 	/* Declare that we are receiver. */
464 	proto_recv(res->hr_ctrl, NULL, 0);
465 	descriptors_cleanup(res);
466 
467 	descriptors_assert(res, mode);
468 
469 	pjdlog_init(mode);
470 	pjdlog_debug_set(debuglevel);
471 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
472 	setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
473 
474 	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
475 	PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
476 
477 	/* Error in setting timeout is not critical, but why should it fail? */
478 	if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1)
479 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
480 	if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1)
481 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
482 
483 	init_local(res);
484 	init_environment();
485 
486 	if (drop_privs(res) != 0)
487 		exit(EX_CONFIG);
488 	pjdlog_info("Privileges successfully dropped.");
489 
490 	/*
491 	 * Create the control thread before sending any event to the parent,
492 	 * as we can deadlock when parent sends control request to worker,
493 	 * but worker has no control thread started yet, so parent waits.
494 	 * In the meantime worker sends an event to the parent, but parent
495 	 * is unable to handle the event, because it waits for control
496 	 * request response.
497 	 */
498 	error = pthread_create(&td, NULL, ctrl_thread, res);
499 	PJDLOG_ASSERT(error == 0);
500 
501 	init_remote(res, nvin);
502 	event_send(res, EVENT_CONNECT);
503 
504 	error = pthread_create(&td, NULL, recv_thread, res);
505 	PJDLOG_ASSERT(error == 0);
506 	error = pthread_create(&td, NULL, disk_thread, res);
507 	PJDLOG_ASSERT(error == 0);
508 	(void)send_thread(res);
509 }
510 
511 static void
512 reqlog(int loglevel, int debuglevel, int error, struct hio *hio,
513     const char *fmt, ...)
514 {
515 	char msg[1024];
516 	va_list ap;
517 	int len;
518 
519 	va_start(ap, fmt);
520 	len = vsnprintf(msg, sizeof(msg), fmt, ap);
521 	va_end(ap);
522 	if ((size_t)len < sizeof(msg)) {
523 		switch (hio->hio_cmd) {
524 		case HIO_READ:
525 			(void)snprintf(msg + len, sizeof(msg) - len,
526 			    "READ(%ju, %ju).", (uintmax_t)hio->hio_offset,
527 			    (uintmax_t)hio->hio_length);
528 			break;
529 		case HIO_DELETE:
530 			(void)snprintf(msg + len, sizeof(msg) - len,
531 			    "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset,
532 			    (uintmax_t)hio->hio_length);
533 			break;
534 		case HIO_FLUSH:
535 			(void)snprintf(msg + len, sizeof(msg) - len, "FLUSH.");
536 			break;
537 		case HIO_WRITE:
538 			(void)snprintf(msg + len, sizeof(msg) - len,
539 			    "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset,
540 			    (uintmax_t)hio->hio_length);
541 			break;
542 		case HIO_KEEPALIVE:
543 			(void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE.");
544 			break;
545 		default:
546 			(void)snprintf(msg + len, sizeof(msg) - len,
547 			    "UNKNOWN(%u).", (unsigned int)hio->hio_cmd);
548 			break;
549 		}
550 	}
551 	pjdlog_common(loglevel, debuglevel, error, "%s", msg);
552 }
553 
554 static int
555 requnpack(struct hast_resource *res, struct hio *hio, struct nv *nv)
556 {
557 
558 	hio->hio_cmd = nv_get_uint8(nv, "cmd");
559 	if (hio->hio_cmd == 0) {
560 		pjdlog_error("Header contains no 'cmd' field.");
561 		hio->hio_error = EINVAL;
562 		goto end;
563 	}
564 	if (hio->hio_cmd != HIO_KEEPALIVE) {
565 		hio->hio_seq = nv_get_uint64(nv, "seq");
566 		if (hio->hio_seq == 0) {
567 			pjdlog_error("Header contains no 'seq' field.");
568 			hio->hio_error = EINVAL;
569 			goto end;
570 		}
571 	}
572 	switch (hio->hio_cmd) {
573 	case HIO_FLUSH:
574 	case HIO_KEEPALIVE:
575 		break;
576 	case HIO_WRITE:
577 		hio->hio_memsync = nv_exists(nv, "memsync");
578 		/* FALLTHROUGH */
579 	case HIO_READ:
580 	case HIO_DELETE:
581 		hio->hio_offset = nv_get_uint64(nv, "offset");
582 		if (nv_error(nv) != 0) {
583 			pjdlog_error("Header is missing 'offset' field.");
584 			hio->hio_error = EINVAL;
585 			goto end;
586 		}
587 		hio->hio_length = nv_get_uint64(nv, "length");
588 		if (nv_error(nv) != 0) {
589 			pjdlog_error("Header is missing 'length' field.");
590 			hio->hio_error = EINVAL;
591 			goto end;
592 		}
593 		if (hio->hio_length == 0) {
594 			pjdlog_error("Data length is zero.");
595 			hio->hio_error = EINVAL;
596 			goto end;
597 		}
598 		if (hio->hio_cmd != HIO_DELETE && hio->hio_length > MAXPHYS) {
599 			pjdlog_error("Data length is too large (%ju > %ju).",
600 			    (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);
601 			hio->hio_error = EINVAL;
602 			goto end;
603 		}
604 		if ((hio->hio_offset % res->hr_local_sectorsize) != 0) {
605 			pjdlog_error("Offset %ju is not multiple of sector size.",
606 			    (uintmax_t)hio->hio_offset);
607 			hio->hio_error = EINVAL;
608 			goto end;
609 		}
610 		if ((hio->hio_length % res->hr_local_sectorsize) != 0) {
611 			pjdlog_error("Length %ju is not multiple of sector size.",
612 			    (uintmax_t)hio->hio_length);
613 			hio->hio_error = EINVAL;
614 			goto end;
615 		}
616 		if (hio->hio_offset + hio->hio_length >
617 		    (uint64_t)res->hr_datasize) {
618 			pjdlog_error("Data offset is too large (%ju > %ju).",
619 			    (uintmax_t)(hio->hio_offset + hio->hio_length),
620 			    (uintmax_t)res->hr_datasize);
621 			hio->hio_error = EINVAL;
622 			goto end;
623 		}
624 		break;
625 	default:
626 		pjdlog_error("Header contains invalid 'cmd' (%hhu).",
627 		    hio->hio_cmd);
628 		hio->hio_error = EINVAL;
629 		goto end;
630 	}
631 	hio->hio_error = 0;
632 end:
633 	return (hio->hio_error);
634 }
635 
636 static __dead2 void
637 secondary_exit(int exitcode, const char *fmt, ...)
638 {
639 	va_list ap;
640 
641 	PJDLOG_ASSERT(exitcode != EX_OK);
642 	va_start(ap, fmt);
643 	pjdlogv_errno(LOG_ERR, fmt, ap);
644 	va_end(ap);
645 	event_send(gres, EVENT_DISCONNECT);
646 	exit(exitcode);
647 }
648 
649 /*
650  * Thread receives requests from the primary node.
651  */
652 static void *
653 recv_thread(void *arg)
654 {
655 	struct hast_resource *res = arg;
656 	struct hio *hio, *mshio;
657 	struct nv *nv;
658 
659 	for (;;) {
660 		pjdlog_debug(2, "recv: Taking free request.");
661 		QUEUE_TAKE(free, hio);
662 		pjdlog_debug(2, "recv: (%p) Got request.", hio);
663 		if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) {
664 			secondary_exit(EX_TEMPFAIL,
665 			    "Unable to receive request header");
666 		}
667 		if (requnpack(res, hio, nv) != 0) {
668 			nv_free(nv);
669 			pjdlog_debug(2,
670 			    "recv: (%p) Moving request to the send queue.",
671 			    hio);
672 			QUEUE_INSERT(send, hio);
673 			continue;
674 		}
675 		switch (hio->hio_cmd) {
676 		case HIO_READ:
677 			res->hr_stat_read++;
678 			break;
679 		case HIO_WRITE:
680 			res->hr_stat_write++;
681 			break;
682 		case HIO_DELETE:
683 			res->hr_stat_delete++;
684 			break;
685 		case HIO_FLUSH:
686 			res->hr_stat_flush++;
687 			break;
688 		case HIO_KEEPALIVE:
689 			break;
690 		default:
691 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
692 			    hio->hio_cmd);
693 		}
694 		reqlog(LOG_DEBUG, 2, -1, hio,
695 		    "recv: (%p) Got request header: ", hio);
696 		if (hio->hio_cmd == HIO_KEEPALIVE) {
697 			nv_free(nv);
698 			pjdlog_debug(2,
699 			    "recv: (%p) Moving request to the free queue.",
700 			    hio);
701 			hio_clear(hio);
702 			QUEUE_INSERT(free, hio);
703 			continue;
704 		} else if (hio->hio_cmd == HIO_WRITE) {
705 			if (hast_proto_recv_data(res, res->hr_remotein, nv,
706 			    hio->hio_data, MAXPHYS) == -1) {
707 				secondary_exit(EX_TEMPFAIL,
708 				    "Unable to receive request data");
709 			}
710 			if (hio->hio_memsync) {
711 				/*
712 				 * For memsync requests we expect two replies.
713 				 * Clone the hio so we can handle both of them.
714 				 */
715 				pjdlog_debug(2, "recv: Taking free request.");
716 				QUEUE_TAKE(free, mshio);
717 				pjdlog_debug(2, "recv: (%p) Got request.",
718 				    mshio);
719 				hio_copy(hio, mshio);
720 				mshio->hio_error = 0;
721 				/*
722 				 * We want to keep 'memsync' tag only on the
723 				 * request going onto send queue (mshio).
724 				 */
725 				hio->hio_memsync = false;
726 				pjdlog_debug(2,
727 				    "recv: (%p) Moving memsync request to the send queue.",
728 				    mshio);
729 				QUEUE_INSERT(send, mshio);
730 			}
731 		}
732 		nv_free(nv);
733 		pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.",
734 		    hio);
735 		QUEUE_INSERT(disk, hio);
736 	}
737 	/* NOTREACHED */
738 	return (NULL);
739 }
740 
741 /*
742  * Thread reads from or writes to local component and also handles DELETE and
743  * FLUSH requests.
744  */
745 static void *
746 disk_thread(void *arg)
747 {
748 	struct hast_resource *res = arg;
749 	struct hio *hio;
750 	ssize_t ret;
751 	bool clear_activemap, logerror;
752 
753 	clear_activemap = true;
754 
755 	for (;;) {
756 		pjdlog_debug(2, "disk: Taking request.");
757 		QUEUE_TAKE(disk, hio);
758 		while (clear_activemap) {
759 			unsigned char *map;
760 			size_t mapsize;
761 
762 			/*
763 			 * When first request is received, it means that primary
764 			 * already received our activemap, merged it and stored
765 			 * locally. We can now safely clear our activemap.
766 			 */
767 			mapsize =
768 			    activemap_calc_ondisk_size(res->hr_local_mediasize -
769 			    METADATA_SIZE, res->hr_extentsize,
770 			    res->hr_local_sectorsize);
771 			map = calloc(1, mapsize);
772 			if (map == NULL) {
773 				pjdlog_warning("Unable to allocate memory to clear local activemap.");
774 				break;
775 			}
776 			if (pwrite(res->hr_localfd, map, mapsize,
777 			    METADATA_SIZE) != (ssize_t)mapsize) {
778 				pjdlog_errno(LOG_WARNING,
779 				    "Unable to store cleared activemap");
780 				free(map);
781 				res->hr_stat_activemap_write_error++;
782 				break;
783 			}
784 			free(map);
785 			clear_activemap = false;
786 			pjdlog_debug(1, "Local activemap cleared.");
787 			break;
788 		}
789 		reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio);
790 		logerror = true;
791 		/* Handle the actual request. */
792 		switch (hio->hio_cmd) {
793 		case HIO_READ:
794 			ret = pread(res->hr_localfd, hio->hio_data,
795 			    hio->hio_length,
796 			    hio->hio_offset + res->hr_localoff);
797 			if (ret == -1)
798 				hio->hio_error = errno;
799 			else if (ret != (int64_t)hio->hio_length)
800 				hio->hio_error = EIO;
801 			else
802 				hio->hio_error = 0;
803 			break;
804 		case HIO_WRITE:
805 			ret = pwrite(res->hr_localfd, hio->hio_data,
806 			    hio->hio_length,
807 			    hio->hio_offset + res->hr_localoff);
808 			if (ret == -1)
809 				hio->hio_error = errno;
810 			else if (ret != (int64_t)hio->hio_length)
811 				hio->hio_error = EIO;
812 			else
813 				hio->hio_error = 0;
814 			break;
815 		case HIO_DELETE:
816 			ret = g_delete(res->hr_localfd,
817 			    hio->hio_offset + res->hr_localoff,
818 			    hio->hio_length);
819 			if (ret == -1)
820 				hio->hio_error = errno;
821 			else
822 				hio->hio_error = 0;
823 			break;
824 		case HIO_FLUSH:
825 			if (!res->hr_localflush) {
826 				ret = -1;
827 				hio->hio_error = EOPNOTSUPP;
828 				logerror = false;
829 				break;
830 			}
831 			ret = g_flush(res->hr_localfd);
832 			if (ret == -1) {
833 				if (errno == EOPNOTSUPP)
834 					res->hr_localflush = false;
835 				hio->hio_error = errno;
836 			} else {
837 				hio->hio_error = 0;
838 			}
839 			break;
840 		default:
841 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
842 			    hio->hio_cmd);
843 		}
844 		if (logerror && hio->hio_error != 0) {
845 			reqlog(LOG_ERR, 0, hio->hio_error, hio,
846 			    "Request failed: ");
847 		}
848 		pjdlog_debug(2, "disk: (%p) Moving request to the send queue.",
849 		    hio);
850 		QUEUE_INSERT(send, hio);
851 	}
852 	/* NOTREACHED */
853 	return (NULL);
854 }
855 
856 /*
857  * Thread sends requests back to primary node.
858  */
859 static void *
860 send_thread(void *arg)
861 {
862 	struct hast_resource *res = arg;
863 	struct nv *nvout;
864 	struct hio *hio;
865 	void *data;
866 	size_t length;
867 
868 	for (;;) {
869 		pjdlog_debug(2, "send: Taking request.");
870 		QUEUE_TAKE(send, hio);
871 		reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio);
872 		nvout = nv_alloc();
873 		/* Copy sequence number. */
874 		nv_add_uint64(nvout, hio->hio_seq, "seq");
875 		if (hio->hio_memsync) {
876 			PJDLOG_ASSERT(hio->hio_cmd == HIO_WRITE);
877 			nv_add_int8(nvout, 1, "received");
878 		}
879 		switch (hio->hio_cmd) {
880 		case HIO_READ:
881 			if (hio->hio_error == 0) {
882 				data = hio->hio_data;
883 				length = hio->hio_length;
884 				break;
885 			}
886 			/*
887 			 * We send no data in case of an error.
888 			 */
889 			/* FALLTHROUGH */
890 		case HIO_DELETE:
891 		case HIO_FLUSH:
892 		case HIO_WRITE:
893 			data = NULL;
894 			length = 0;
895 			break;
896 		default:
897 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
898 			    hio->hio_cmd);
899 		}
900 		if (hio->hio_error != 0) {
901 			switch (hio->hio_cmd) {
902 			case HIO_READ:
903 				res->hr_stat_read_error++;
904 				break;
905 			case HIO_WRITE:
906 				res->hr_stat_write_error++;
907 				break;
908 			case HIO_DELETE:
909 				res->hr_stat_delete_error++;
910 				break;
911 			case HIO_FLUSH:
912 				res->hr_stat_flush_error++;
913 				break;
914 			}
915 			nv_add_int16(nvout, hio->hio_error, "error");
916 		}
917 		if (hast_proto_send(res, res->hr_remoteout, nvout, data,
918 		    length) == -1) {
919 			secondary_exit(EX_TEMPFAIL, "Unable to send reply");
920 		}
921 		nv_free(nvout);
922 		pjdlog_debug(2, "send: (%p) Moving request to the free queue.",
923 		    hio);
924 		hio_clear(hio);
925 		QUEUE_INSERT(free, hio);
926 	}
927 	/* NOTREACHED */
928 	return (NULL);
929 }
930