xref: /freebsd/sbin/hastd/secondary.c (revision 10b59a9b4add0320d52c15ce057dd697261e7dfc)
1 /*-
2  * Copyright (c) 2009-2010 The FreeBSD Foundation
3  * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
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/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 	TAILQ_ENTRY(hio) hio_next;
75 };
76 
77 static struct hast_resource *gres;
78 
79 /*
80  * Free list holds unused structures. When free list is empty, we have to wait
81  * until some in-progress requests are freed.
82  */
83 static TAILQ_HEAD(, hio) hio_free_list;
84 static pthread_mutex_t hio_free_list_lock;
85 static pthread_cond_t hio_free_list_cond;
86 /*
87  * Disk thread (the one that do I/O requests) takes requests from this list.
88  */
89 static TAILQ_HEAD(, hio) hio_disk_list;
90 static pthread_mutex_t hio_disk_list_lock;
91 static pthread_cond_t hio_disk_list_cond;
92 /*
93  * There is one recv list for every component, although local components don't
94  * use recv lists as local requests are done synchronously.
95  */
96 static TAILQ_HEAD(, hio) hio_send_list;
97 static pthread_mutex_t hio_send_list_lock;
98 static pthread_cond_t hio_send_list_cond;
99 
100 /*
101  * Maximum number of outstanding I/O requests.
102  */
103 #define	HAST_HIO_MAX	256
104 
105 static void *recv_thread(void *arg);
106 static void *disk_thread(void *arg);
107 static void *send_thread(void *arg);
108 
109 #define	QUEUE_INSERT(name, hio)	do {					\
110 	bool _wakeup;							\
111 									\
112 	mtx_lock(&hio_##name##_list_lock);				\
113 	_wakeup = TAILQ_EMPTY(&hio_##name##_list);			\
114 	TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next);		\
115 	mtx_unlock(&hio_##name##_list_lock);				\
116 	if (_wakeup)							\
117 		cv_signal(&hio_##name##_list_cond);			\
118 } while (0)
119 #define	QUEUE_TAKE(name, hio)	do {					\
120 	mtx_lock(&hio_##name##_list_lock);				\
121 	while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) {	\
122 		cv_wait(&hio_##name##_list_cond,			\
123 		    &hio_##name##_list_lock);				\
124 	}								\
125 	TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next);		\
126 	mtx_unlock(&hio_##name##_list_lock);				\
127 } while (0)
128 
129 static void
130 hio_clear(struct hio *hio)
131 {
132 
133 	hio->hio_seq = 0;
134 	hio->hio_error = 0;
135 	hio->hio_cmd = HIO_UNDEF;
136 	hio->hio_offset = 0;
137 	hio->hio_length = 0;
138 }
139 
140 static void
141 init_environment(void)
142 {
143 	struct hio *hio;
144 	unsigned int ii;
145 
146 	/*
147 	 * Initialize lists, their locks and theirs condition variables.
148 	 */
149 	TAILQ_INIT(&hio_free_list);
150 	mtx_init(&hio_free_list_lock);
151 	cv_init(&hio_free_list_cond);
152 	TAILQ_INIT(&hio_disk_list);
153 	mtx_init(&hio_disk_list_lock);
154 	cv_init(&hio_disk_list_cond);
155 	TAILQ_INIT(&hio_send_list);
156 	mtx_init(&hio_send_list_lock);
157 	cv_init(&hio_send_list_cond);
158 
159 	/*
160 	 * Allocate requests pool and initialize requests.
161 	 */
162 	for (ii = 0; ii < HAST_HIO_MAX; ii++) {
163 		hio = malloc(sizeof(*hio));
164 		if (hio == NULL) {
165 			pjdlog_exitx(EX_TEMPFAIL,
166 			    "Unable to allocate memory (%zu bytes) for hio request.",
167 			    sizeof(*hio));
168 		}
169 		hio->hio_data = malloc(MAXPHYS);
170 		if (hio->hio_data == NULL) {
171 			pjdlog_exitx(EX_TEMPFAIL,
172 			    "Unable to allocate memory (%zu bytes) for gctl_data.",
173 			    (size_t)MAXPHYS);
174 		}
175 		hio_clear(hio);
176 		TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);
177 	}
178 }
179 
180 static void
181 init_local(struct hast_resource *res)
182 {
183 
184 	if (metadata_read(res, true) < 0)
185 		exit(EX_NOINPUT);
186 }
187 
188 static void
189 init_remote(struct hast_resource *res, struct nv *nvin)
190 {
191 	uint64_t resuid;
192 	struct nv *nvout;
193 	unsigned char *map;
194 	size_t mapsize;
195 
196 #ifdef notyet
197 	/* Setup direction. */
198 	if (proto_send(res->hr_remoteout, NULL, 0) == -1)
199 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
200 #endif
201 
202 	map = NULL;
203 	mapsize = 0;
204 	nvout = nv_alloc();
205 	nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize");
206 	nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize");
207 	resuid = nv_get_uint64(nvin, "resuid");
208 	res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt");
209 	res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt");
210 	nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt");
211 	nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt");
212 	mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
213 	    METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize);
214 	map = malloc(mapsize);
215 	if (map == NULL) {
216 		pjdlog_exitx(EX_TEMPFAIL,
217 		    "Unable to allocate memory (%zu bytes) for activemap.",
218 		    mapsize);
219 	}
220 	/*
221 	 * When we work as primary and secondary is missing we will increase
222 	 * localcnt in our metadata. When secondary is connected and synced
223 	 * we make localcnt be equal to remotecnt, which means nodes are more
224 	 * or less in sync.
225 	 * Split-brain condition is when both nodes are not able to communicate
226 	 * and are both configured as primary nodes. In turn, they can both
227 	 * make incompatible changes to the data and we have to detect that.
228 	 * Under split-brain condition we will increase our localcnt on first
229 	 * write and remote node will increase its localcnt on first write.
230 	 * When we connect we can see that primary's localcnt is greater than
231 	 * our remotecnt (primary was modified while we weren't watching) and
232 	 * our localcnt is greater than primary's remotecnt (we were modified
233 	 * while primary wasn't watching).
234 	 * There are many possible combinations which are all gathered below.
235 	 * Don't pay too much attention to exact numbers, the more important
236 	 * is to compare them. We compare secondary's local with primary's
237 	 * remote and secondary's remote with primary's local.
238 	 * Note that every case where primary's localcnt is smaller than
239 	 * secondary's remotecnt and where secondary's localcnt is smaller than
240 	 * primary's remotecnt should be impossible in practise. We will perform
241 	 * full synchronization then. Those cases are marked with an asterisk.
242 	 * Regular synchronization means that only extents marked as dirty are
243 	 * synchronized (regular synchronization).
244 	 *
245 	 * SECONDARY METADATA PRIMARY METADATA
246 	 * local=3 remote=3   local=2 remote=2*  ?! Full sync from secondary.
247 	 * local=3 remote=3   local=2 remote=3*  ?! Full sync from primary.
248 	 * local=3 remote=3   local=2 remote=4*  ?! Full sync from primary.
249 	 * local=3 remote=3   local=3 remote=2   Primary is out-of-date,
250 	 *                                       regular sync from secondary.
251 	 * local=3 remote=3   local=3 remote=3   Regular sync just in case.
252 	 * local=3 remote=3   local=3 remote=4*  ?! Full sync from primary.
253 	 * local=3 remote=3   local=4 remote=2   Split-brain condition.
254 	 * local=3 remote=3   local=4 remote=3   Secondary out-of-date,
255 	 *                                       regular sync from primary.
256 	 * local=3 remote=3   local=4 remote=4*  ?! Full sync from primary.
257 	 */
258 	if (res->hr_resuid == 0) {
259 		/*
260 		 * Provider is used for the first time. If primary node done no
261 		 * writes yet as well (we will find "virgin" argument) then
262 		 * there is no need to synchronize anything. If primary node
263 		 * done any writes already we have to synchronize everything.
264 		 */
265 		PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
266 		res->hr_resuid = resuid;
267 		if (metadata_write(res) < 0)
268 			exit(EX_NOINPUT);
269 		if (nv_exists(nvin, "virgin")) {
270 			free(map);
271 			map = NULL;
272 			mapsize = 0;
273 		} else {
274 			memset(map, 0xff, mapsize);
275 		}
276 		nv_add_int8(nvout, 1, "virgin");
277 		nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
278 	} else if (res->hr_resuid != resuid) {
279 		char errmsg[256];
280 
281 		free(map);
282 		(void)snprintf(errmsg, sizeof(errmsg),
283 		    "Resource unique ID mismatch (primary=%ju, secondary=%ju).",
284 		    (uintmax_t)resuid, (uintmax_t)res->hr_resuid);
285 		pjdlog_error("%s", errmsg);
286 		nv_add_string(nvout, errmsg, "errmsg");
287 		if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) {
288 			pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s",
289 			    res->hr_remoteaddr);
290 		}
291 		nv_free(nvout);
292 		exit(EX_CONFIG);
293 	} else if (
294 	    /* Is primary out-of-date? */
295 	    (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
296 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
297 	    /* Are the nodes more or less in sync? */
298 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
299 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
300 	    /* Is secondary out-of-date? */
301 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
302 	     res->hr_secondary_remotecnt < res->hr_primary_localcnt)) {
303 		/*
304 		 * Nodes are more or less in sync or one of the nodes is
305 		 * out-of-date.
306 		 * It doesn't matter at this point which one, we just have to
307 		 * send out local bitmap to the remote node.
308 		 */
309 		if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) !=
310 		    (ssize_t)mapsize) {
311 			pjdlog_exit(LOG_ERR, "Unable to read activemap");
312 		}
313 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
314 		     res->hr_secondary_remotecnt == res->hr_primary_localcnt) {
315 			/* Primary is out-of-date, sync from secondary. */
316 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
317 		} else {
318 			/*
319 			 * Secondary is out-of-date or counts match.
320 			 * Sync from primary.
321 			 */
322 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
323 		}
324 	} else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
325 	     res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
326 		/*
327 		 * Not good, we have split-brain condition.
328 		 */
329 		free(map);
330 		pjdlog_error("Split-brain detected, exiting.");
331 		nv_add_string(nvout, "Split-brain condition!", "errmsg");
332 		if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) {
333 			pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s",
334 			    res->hr_remoteaddr);
335 		}
336 		nv_free(nvout);
337 		/* Exit on split-brain. */
338 		event_send(res, EVENT_SPLITBRAIN);
339 		exit(EX_CONFIG);
340 	} else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
341 	    res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ {
342 		/*
343 		 * This should never happen in practise, but we will perform
344 		 * full synchronization.
345 		 */
346 		PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
347 		    res->hr_primary_localcnt < res->hr_secondary_remotecnt);
348 		mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
349 		    METADATA_SIZE, res->hr_extentsize,
350 		    res->hr_local_sectorsize);
351 		memset(map, 0xff, mapsize);
352 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) {
353 			/* In this one of five cases sync from secondary. */
354 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
355 		} else {
356 			/* For the rest four cases sync from primary. */
357 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
358 		}
359 		pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).",
360 		    (uintmax_t)res->hr_primary_localcnt,
361 		    (uintmax_t)res->hr_primary_remotecnt,
362 		    (uintmax_t)res->hr_secondary_localcnt,
363 		    (uintmax_t)res->hr_secondary_remotecnt);
364 	}
365 	nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize");
366 	if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) {
367 		pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s",
368 		    res->hr_remoteaddr);
369 	}
370 	if (map != NULL)
371 		free(map);
372 	nv_free(nvout);
373 #ifdef notyet
374 	/* Setup direction. */
375 	if (proto_recv(res->hr_remotein, NULL, 0) == -1)
376 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
377 #endif
378 }
379 
380 void
381 hastd_secondary(struct hast_resource *res, struct nv *nvin)
382 {
383 	sigset_t mask;
384 	pthread_t td;
385 	pid_t pid;
386 	int error, mode, debuglevel;
387 
388 	/*
389 	 * Create communication channel between parent and child.
390 	 */
391 	if (proto_client(NULL, "socketpair://", &res->hr_ctrl) < 0) {
392 		KEEP_ERRNO((void)pidfile_remove(pfh));
393 		pjdlog_exit(EX_OSERR,
394 		    "Unable to create control sockets between parent and child");
395 	}
396 	/*
397 	 * Create communication channel between child and parent.
398 	 */
399 	if (proto_client(NULL, "socketpair://", &res->hr_event) < 0) {
400 		KEEP_ERRNO((void)pidfile_remove(pfh));
401 		pjdlog_exit(EX_OSERR,
402 		    "Unable to create event sockets between child and parent");
403 	}
404 
405 	pid = fork();
406 	if (pid < 0) {
407 		KEEP_ERRNO((void)pidfile_remove(pfh));
408 		pjdlog_exit(EX_OSERR, "Unable to fork");
409 	}
410 
411 	if (pid > 0) {
412 		/* This is parent. */
413 		proto_close(res->hr_remotein);
414 		res->hr_remotein = NULL;
415 		proto_close(res->hr_remoteout);
416 		res->hr_remoteout = NULL;
417 		/* Declare that we are receiver. */
418 		proto_recv(res->hr_event, NULL, 0);
419 		/* Declare that we are sender. */
420 		proto_send(res->hr_ctrl, NULL, 0);
421 		res->hr_workerpid = pid;
422 		return;
423 	}
424 
425 	gres = res;
426 	mode = pjdlog_mode_get();
427 	debuglevel = pjdlog_debug_get();
428 
429 	/* Declare that we are sender. */
430 	proto_send(res->hr_event, NULL, 0);
431 	/* Declare that we are receiver. */
432 	proto_recv(res->hr_ctrl, NULL, 0);
433 	descriptors_cleanup(res);
434 
435 	descriptors_assert(res, mode);
436 
437 	pjdlog_init(mode);
438 	pjdlog_debug_set(debuglevel);
439 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
440 	setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
441 
442 	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
443 	PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
444 
445 	/* Error in setting timeout is not critical, but why should it fail? */
446 	if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0)
447 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
448 	if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0)
449 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
450 
451 	init_local(res);
452 	init_environment();
453 
454 	if (drop_privs(res) != 0)
455 		exit(EX_CONFIG);
456 	pjdlog_info("Privileges successfully dropped.");
457 
458 	/*
459 	 * Create the control thread before sending any event to the parent,
460 	 * as we can deadlock when parent sends control request to worker,
461 	 * but worker has no control thread started yet, so parent waits.
462 	 * In the meantime worker sends an event to the parent, but parent
463 	 * is unable to handle the event, because it waits for control
464 	 * request response.
465 	 */
466 	error = pthread_create(&td, NULL, ctrl_thread, res);
467 	PJDLOG_ASSERT(error == 0);
468 
469 	init_remote(res, nvin);
470 	event_send(res, EVENT_CONNECT);
471 
472 	error = pthread_create(&td, NULL, recv_thread, res);
473 	PJDLOG_ASSERT(error == 0);
474 	error = pthread_create(&td, NULL, disk_thread, res);
475 	PJDLOG_ASSERT(error == 0);
476 	(void)send_thread(res);
477 }
478 
479 static void
480 reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...)
481 {
482 	char msg[1024];
483 	va_list ap;
484 	int len;
485 
486 	va_start(ap, fmt);
487 	len = vsnprintf(msg, sizeof(msg), fmt, ap);
488 	va_end(ap);
489 	if ((size_t)len < sizeof(msg)) {
490 		switch (hio->hio_cmd) {
491 		case HIO_READ:
492 			(void)snprintf(msg + len, sizeof(msg) - len,
493 			    "READ(%ju, %ju).", (uintmax_t)hio->hio_offset,
494 			    (uintmax_t)hio->hio_length);
495 			break;
496 		case HIO_DELETE:
497 			(void)snprintf(msg + len, sizeof(msg) - len,
498 			    "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset,
499 			    (uintmax_t)hio->hio_length);
500 			break;
501 		case HIO_FLUSH:
502 			(void)snprintf(msg + len, sizeof(msg) - len, "FLUSH.");
503 			break;
504 		case HIO_WRITE:
505 			(void)snprintf(msg + len, sizeof(msg) - len,
506 			    "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset,
507 			    (uintmax_t)hio->hio_length);
508 			break;
509 		case HIO_KEEPALIVE:
510 			(void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE.");
511 			break;
512 		default:
513 			(void)snprintf(msg + len, sizeof(msg) - len,
514 			    "UNKNOWN(%u).", (unsigned int)hio->hio_cmd);
515 			break;
516 		}
517 	}
518 	pjdlog_common(loglevel, debuglevel, error, "%s", msg);
519 }
520 
521 static int
522 requnpack(struct hast_resource *res, struct hio *hio, struct nv *nv)
523 {
524 
525 	hio->hio_cmd = nv_get_uint8(nv, "cmd");
526 	if (hio->hio_cmd == 0) {
527 		pjdlog_error("Header contains no 'cmd' field.");
528 		hio->hio_error = EINVAL;
529 		goto end;
530 	}
531 	if (hio->hio_cmd != HIO_KEEPALIVE) {
532 		hio->hio_seq = nv_get_uint64(nv, "seq");
533 		if (hio->hio_seq == 0) {
534 			pjdlog_error("Header contains no 'seq' field.");
535 			hio->hio_error = EINVAL;
536 			goto end;
537 		}
538 	}
539 	switch (hio->hio_cmd) {
540 	case HIO_FLUSH:
541 	case HIO_KEEPALIVE:
542 		break;
543 	case HIO_READ:
544 	case HIO_WRITE:
545 	case HIO_DELETE:
546 		hio->hio_offset = nv_get_uint64(nv, "offset");
547 		if (nv_error(nv) != 0) {
548 			pjdlog_error("Header is missing 'offset' field.");
549 			hio->hio_error = EINVAL;
550 			goto end;
551 		}
552 		hio->hio_length = nv_get_uint64(nv, "length");
553 		if (nv_error(nv) != 0) {
554 			pjdlog_error("Header is missing 'length' field.");
555 			hio->hio_error = EINVAL;
556 			goto end;
557 		}
558 		if (hio->hio_length == 0) {
559 			pjdlog_error("Data length is zero.");
560 			hio->hio_error = EINVAL;
561 			goto end;
562 		}
563 		if (hio->hio_length > MAXPHYS) {
564 			pjdlog_error("Data length is too large (%ju > %ju).",
565 			    (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);
566 			hio->hio_error = EINVAL;
567 			goto end;
568 		}
569 		if ((hio->hio_offset % res->hr_local_sectorsize) != 0) {
570 			pjdlog_error("Offset %ju is not multiple of sector size.",
571 			    (uintmax_t)hio->hio_offset);
572 			hio->hio_error = EINVAL;
573 			goto end;
574 		}
575 		if ((hio->hio_length % res->hr_local_sectorsize) != 0) {
576 			pjdlog_error("Length %ju is not multiple of sector size.",
577 			    (uintmax_t)hio->hio_length);
578 			hio->hio_error = EINVAL;
579 			goto end;
580 		}
581 		if (hio->hio_offset + hio->hio_length >
582 		    (uint64_t)res->hr_datasize) {
583 			pjdlog_error("Data offset is too large (%ju > %ju).",
584 			    (uintmax_t)(hio->hio_offset + hio->hio_length),
585 			    (uintmax_t)res->hr_datasize);
586 			hio->hio_error = EINVAL;
587 			goto end;
588 		}
589 		break;
590 	default:
591 		pjdlog_error("Header contains invalid 'cmd' (%hhu).",
592 		    hio->hio_cmd);
593 		hio->hio_error = EINVAL;
594 		goto end;
595 	}
596 	hio->hio_error = 0;
597 end:
598 	return (hio->hio_error);
599 }
600 
601 static __dead2 void
602 secondary_exit(int exitcode, const char *fmt, ...)
603 {
604 	va_list ap;
605 
606 	PJDLOG_ASSERT(exitcode != EX_OK);
607 	va_start(ap, fmt);
608 	pjdlogv_errno(LOG_ERR, fmt, ap);
609 	va_end(ap);
610 	event_send(gres, EVENT_DISCONNECT);
611 	exit(exitcode);
612 }
613 
614 /*
615  * Thread receives requests from the primary node.
616  */
617 static void *
618 recv_thread(void *arg)
619 {
620 	struct hast_resource *res = arg;
621 	struct hio *hio;
622 	struct nv *nv;
623 
624 	for (;;) {
625 		pjdlog_debug(2, "recv: Taking free request.");
626 		QUEUE_TAKE(free, hio);
627 		pjdlog_debug(2, "recv: (%p) Got request.", hio);
628 		if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) {
629 			secondary_exit(EX_TEMPFAIL,
630 			    "Unable to receive request header");
631 		}
632 		if (requnpack(res, hio, nv) != 0) {
633 			nv_free(nv);
634 			pjdlog_debug(2,
635 			    "recv: (%p) Moving request to the send queue.",
636 			    hio);
637 			QUEUE_INSERT(send, hio);
638 			continue;
639 		}
640 		switch (hio->hio_cmd) {
641 		case HIO_READ:
642 			res->hr_stat_read++;
643 			break;
644 		case HIO_WRITE:
645 			res->hr_stat_write++;
646 			break;
647 		case HIO_DELETE:
648 			res->hr_stat_delete++;
649 			break;
650 		case HIO_FLUSH:
651 			res->hr_stat_flush++;
652 			break;
653 		case HIO_KEEPALIVE:
654 			break;
655 		default:
656 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
657 			    hio->hio_cmd);
658 		}
659 		reqlog(LOG_DEBUG, 2, -1, hio,
660 		    "recv: (%p) Got request header: ", hio);
661 		if (hio->hio_cmd == HIO_KEEPALIVE) {
662 			nv_free(nv);
663 			pjdlog_debug(2,
664 			    "recv: (%p) Moving request to the free queue.",
665 			    hio);
666 			hio_clear(hio);
667 			QUEUE_INSERT(free, hio);
668 			continue;
669 		} else if (hio->hio_cmd == HIO_WRITE) {
670 			if (hast_proto_recv_data(res, res->hr_remotein, nv,
671 			    hio->hio_data, MAXPHYS) < 0) {
672 				secondary_exit(EX_TEMPFAIL,
673 				    "Unable to receive request data");
674 			}
675 		}
676 		nv_free(nv);
677 		pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.",
678 		    hio);
679 		QUEUE_INSERT(disk, hio);
680 	}
681 	/* NOTREACHED */
682 	return (NULL);
683 }
684 
685 /*
686  * Thread reads from or writes to local component and also handles DELETE and
687  * FLUSH requests.
688  */
689 static void *
690 disk_thread(void *arg)
691 {
692 	struct hast_resource *res = arg;
693 	struct hio *hio;
694 	ssize_t ret;
695 	bool clear_activemap, logerror;
696 
697 	clear_activemap = true;
698 
699 	for (;;) {
700 		pjdlog_debug(2, "disk: Taking request.");
701 		QUEUE_TAKE(disk, hio);
702 		while (clear_activemap) {
703 			unsigned char *map;
704 			size_t mapsize;
705 
706 			/*
707 			 * When first request is received, it means that primary
708 			 * already received our activemap, merged it and stored
709 			 * locally. We can now safely clear our activemap.
710 			 */
711 			mapsize =
712 			    activemap_calc_ondisk_size(res->hr_local_mediasize -
713 			    METADATA_SIZE, res->hr_extentsize,
714 			    res->hr_local_sectorsize);
715 			map = calloc(1, mapsize);
716 			if (map == NULL) {
717 				pjdlog_warning("Unable to allocate memory to clear local activemap.");
718 				break;
719 			}
720 			if (pwrite(res->hr_localfd, map, mapsize,
721 			    METADATA_SIZE) != (ssize_t)mapsize) {
722 				pjdlog_errno(LOG_WARNING,
723 				    "Unable to store cleared activemap");
724 				free(map);
725 				break;
726 			}
727 			free(map);
728 			clear_activemap = false;
729 			pjdlog_debug(1, "Local activemap cleared.");
730 			break;
731 		}
732 		reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio);
733 		logerror = true;
734 		/* Handle the actual request. */
735 		switch (hio->hio_cmd) {
736 		case HIO_READ:
737 			ret = pread(res->hr_localfd, hio->hio_data,
738 			    hio->hio_length,
739 			    hio->hio_offset + res->hr_localoff);
740 			if (ret < 0)
741 				hio->hio_error = errno;
742 			else if (ret != (int64_t)hio->hio_length)
743 				hio->hio_error = EIO;
744 			else
745 				hio->hio_error = 0;
746 			break;
747 		case HIO_WRITE:
748 			ret = pwrite(res->hr_localfd, hio->hio_data,
749 			    hio->hio_length,
750 			    hio->hio_offset + res->hr_localoff);
751 			if (ret < 0)
752 				hio->hio_error = errno;
753 			else if (ret != (int64_t)hio->hio_length)
754 				hio->hio_error = EIO;
755 			else
756 				hio->hio_error = 0;
757 			break;
758 		case HIO_DELETE:
759 			ret = g_delete(res->hr_localfd,
760 			    hio->hio_offset + res->hr_localoff,
761 			    hio->hio_length);
762 			if (ret < 0)
763 				hio->hio_error = errno;
764 			else
765 				hio->hio_error = 0;
766 			break;
767 		case HIO_FLUSH:
768 			if (!res->hr_localflush) {
769 				ret = -1;
770 				hio->hio_error = EOPNOTSUPP;
771 				logerror = false;
772 				break;
773 			}
774 			ret = g_flush(res->hr_localfd);
775 			if (ret < 0) {
776 				if (errno == EOPNOTSUPP)
777 					res->hr_localflush = false;
778 				hio->hio_error = errno;
779 			} else {
780 				hio->hio_error = 0;
781 			}
782 			break;
783 		default:
784 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
785 			    hio->hio_cmd);
786 		}
787 		if (logerror && hio->hio_error != 0) {
788 			reqlog(LOG_ERR, 0, hio->hio_error, hio,
789 			    "Request failed: ");
790 		}
791 		pjdlog_debug(2, "disk: (%p) Moving request to the send queue.",
792 		    hio);
793 		QUEUE_INSERT(send, hio);
794 	}
795 	/* NOTREACHED */
796 	return (NULL);
797 }
798 
799 /*
800  * Thread sends requests back to primary node.
801  */
802 static void *
803 send_thread(void *arg)
804 {
805 	struct hast_resource *res = arg;
806 	struct nv *nvout;
807 	struct hio *hio;
808 	void *data;
809 	size_t length;
810 
811 	for (;;) {
812 		pjdlog_debug(2, "send: Taking request.");
813 		QUEUE_TAKE(send, hio);
814 		reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio);
815 		nvout = nv_alloc();
816 		/* Copy sequence number. */
817 		nv_add_uint64(nvout, hio->hio_seq, "seq");
818 		switch (hio->hio_cmd) {
819 		case HIO_READ:
820 			if (hio->hio_error == 0) {
821 				data = hio->hio_data;
822 				length = hio->hio_length;
823 				break;
824 			}
825 			/*
826 			 * We send no data in case of an error.
827 			 */
828 			/* FALLTHROUGH */
829 		case HIO_DELETE:
830 		case HIO_FLUSH:
831 		case HIO_WRITE:
832 			data = NULL;
833 			length = 0;
834 			break;
835 		default:
836 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
837 			    hio->hio_cmd);
838 		}
839 		if (hio->hio_error != 0)
840 			nv_add_int16(nvout, hio->hio_error, "error");
841 		if (hast_proto_send(res, res->hr_remoteout, nvout, data,
842 		    length) < 0) {
843 			secondary_exit(EX_TEMPFAIL, "Unable to send reply.");
844 		}
845 		nv_free(nvout);
846 		pjdlog_debug(2, "send: (%p) Moving request to the free queue.",
847 		    hio);
848 		hio_clear(hio);
849 		QUEUE_INSERT(free, hio);
850 	}
851 	/* NOTREACHED */
852 	return (NULL);
853 }
854