xref: /freebsd/sbin/hastd/secondary.c (revision 3f0164abf32b9b761e0a2cb4bdca3a8b84f156d4)
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 	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 pthread_mutex_t hio_free_list_lock;
86 static pthread_cond_t hio_free_list_cond;
87 /*
88  * Disk thread (the one that do I/O requests) takes requests from this list.
89  */
90 static TAILQ_HEAD(, hio) hio_disk_list;
91 static pthread_mutex_t hio_disk_list_lock;
92 static pthread_cond_t hio_disk_list_cond;
93 /*
94  * There is one recv list for every component, although local components don't
95  * use recv lists as local requests are done synchronously.
96  */
97 static TAILQ_HEAD(, hio) hio_send_list;
98 static pthread_mutex_t hio_send_list_lock;
99 static pthread_cond_t hio_send_list_cond;
100 
101 /*
102  * Maximum number of outstanding I/O requests.
103  */
104 #define	HAST_HIO_MAX	256
105 
106 static void *recv_thread(void *arg);
107 static void *disk_thread(void *arg);
108 static void *send_thread(void *arg);
109 
110 #define	QUEUE_INSERT(name, hio)	do {					\
111 	bool _wakeup;							\
112 									\
113 	mtx_lock(&hio_##name##_list_lock);				\
114 	_wakeup = TAILQ_EMPTY(&hio_##name##_list);			\
115 	TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next);		\
116 	mtx_unlock(&hio_##name##_list_lock);				\
117 	if (_wakeup)							\
118 		cv_signal(&hio_##name##_list_cond);			\
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 	TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next);		\
127 	mtx_unlock(&hio_##name##_list_lock);				\
128 } while (0)
129 
130 static void
131 hio_clear(struct hio *hio)
132 {
133 
134 	hio->hio_seq = 0;
135 	hio->hio_error = 0;
136 	hio->hio_cmd = HIO_UNDEF;
137 	hio->hio_offset = 0;
138 	hio->hio_length = 0;
139 	hio->hio_memsync = false;
140 }
141 
142 static void
143 hio_copy(const struct hio *srchio, struct hio *dsthio)
144 {
145 
146 	/*
147 	 * We don't copy hio_error, hio_data and hio_next fields.
148 	 */
149 
150 	dsthio->hio_seq = srchio->hio_seq;
151 	dsthio->hio_cmd = srchio->hio_cmd;
152 	dsthio->hio_offset = srchio->hio_offset;
153 	dsthio->hio_length = srchio->hio_length;
154 	dsthio->hio_memsync = srchio->hio_memsync;
155 }
156 
157 static void
158 init_environment(void)
159 {
160 	struct hio *hio;
161 	unsigned int ii;
162 
163 	/*
164 	 * Initialize lists, their locks and theirs condition variables.
165 	 */
166 	TAILQ_INIT(&hio_free_list);
167 	mtx_init(&hio_free_list_lock);
168 	cv_init(&hio_free_list_cond);
169 	TAILQ_INIT(&hio_disk_list);
170 	mtx_init(&hio_disk_list_lock);
171 	cv_init(&hio_disk_list_cond);
172 	TAILQ_INIT(&hio_send_list);
173 	mtx_init(&hio_send_list_lock);
174 	cv_init(&hio_send_list_cond);
175 
176 	/*
177 	 * Allocate requests pool and initialize requests.
178 	 */
179 	for (ii = 0; ii < HAST_HIO_MAX; ii++) {
180 		hio = malloc(sizeof(*hio));
181 		if (hio == NULL) {
182 			pjdlog_exitx(EX_TEMPFAIL,
183 			    "Unable to allocate memory (%zu bytes) for hio request.",
184 			    sizeof(*hio));
185 		}
186 		hio->hio_data = malloc(MAXPHYS);
187 		if (hio->hio_data == NULL) {
188 			pjdlog_exitx(EX_TEMPFAIL,
189 			    "Unable to allocate memory (%zu bytes) for gctl_data.",
190 			    (size_t)MAXPHYS);
191 		}
192 		hio_clear(hio);
193 		TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);
194 	}
195 }
196 
197 static void
198 init_local(struct hast_resource *res)
199 {
200 
201 	if (metadata_read(res, true) == -1)
202 		exit(EX_NOINPUT);
203 }
204 
205 static void
206 init_remote(struct hast_resource *res, struct nv *nvin)
207 {
208 	uint64_t resuid;
209 	struct nv *nvout;
210 	unsigned char *map;
211 	size_t mapsize;
212 
213 #ifdef notyet
214 	/* Setup direction. */
215 	if (proto_send(res->hr_remoteout, NULL, 0) == -1)
216 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
217 #endif
218 
219 	nvout = nv_alloc();
220 	nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize");
221 	nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize");
222 	resuid = nv_get_uint64(nvin, "resuid");
223 	res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt");
224 	res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt");
225 	nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt");
226 	nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt");
227 	mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
228 	    METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize);
229 	map = malloc(mapsize);
230 	if (map == NULL) {
231 		pjdlog_exitx(EX_TEMPFAIL,
232 		    "Unable to allocate memory (%zu bytes) for activemap.",
233 		    mapsize);
234 	}
235 	/*
236 	 * When we work as primary and secondary is missing we will increase
237 	 * localcnt in our metadata. When secondary is connected and synced
238 	 * we make localcnt be equal to remotecnt, which means nodes are more
239 	 * or less in sync.
240 	 * Split-brain condition is when both nodes are not able to communicate
241 	 * and are both configured as primary nodes. In turn, they can both
242 	 * make incompatible changes to the data and we have to detect that.
243 	 * Under split-brain condition we will increase our localcnt on first
244 	 * write and remote node will increase its localcnt on first write.
245 	 * When we connect we can see that primary's localcnt is greater than
246 	 * our remotecnt (primary was modified while we weren't watching) and
247 	 * our localcnt is greater than primary's remotecnt (we were modified
248 	 * while primary wasn't watching).
249 	 * There are many possible combinations which are all gathered below.
250 	 * Don't pay too much attention to exact numbers, the more important
251 	 * is to compare them. We compare secondary's local with primary's
252 	 * remote and secondary's remote with primary's local.
253 	 * Note that every case where primary's localcnt is smaller than
254 	 * secondary's remotecnt and where secondary's localcnt is smaller than
255 	 * primary's remotecnt should be impossible in practise. We will perform
256 	 * full synchronization then. Those cases are marked with an asterisk.
257 	 * Regular synchronization means that only extents marked as dirty are
258 	 * synchronized (regular synchronization).
259 	 *
260 	 * SECONDARY METADATA PRIMARY METADATA
261 	 * local=3 remote=3   local=2 remote=2*  ?! Full sync from secondary.
262 	 * local=3 remote=3   local=2 remote=3*  ?! Full sync from primary.
263 	 * local=3 remote=3   local=2 remote=4*  ?! Full sync from primary.
264 	 * local=3 remote=3   local=3 remote=2   Primary is out-of-date,
265 	 *                                       regular sync from secondary.
266 	 * local=3 remote=3   local=3 remote=3   Regular sync just in case.
267 	 * local=3 remote=3   local=3 remote=4*  ?! Full sync from primary.
268 	 * local=3 remote=3   local=4 remote=2   Split-brain condition.
269 	 * local=3 remote=3   local=4 remote=3   Secondary out-of-date,
270 	 *                                       regular sync from primary.
271 	 * local=3 remote=3   local=4 remote=4*  ?! Full sync from primary.
272 	 */
273 	if (res->hr_resuid == 0) {
274 		/*
275 		 * Provider is used for the first time. If primary node done no
276 		 * writes yet as well (we will find "virgin" argument) then
277 		 * there is no need to synchronize anything. If primary node
278 		 * done any writes already we have to synchronize everything.
279 		 */
280 		PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
281 		res->hr_resuid = resuid;
282 		if (metadata_write(res) == -1)
283 			exit(EX_NOINPUT);
284 		if (nv_exists(nvin, "virgin")) {
285 			free(map);
286 			map = NULL;
287 			mapsize = 0;
288 		} else {
289 			memset(map, 0xff, mapsize);
290 		}
291 		nv_add_int8(nvout, 1, "virgin");
292 		nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
293 	} else if (res->hr_resuid != resuid) {
294 		char errmsg[256];
295 
296 		free(map);
297 		(void)snprintf(errmsg, sizeof(errmsg),
298 		    "Resource unique ID mismatch (primary=%ju, secondary=%ju).",
299 		    (uintmax_t)resuid, (uintmax_t)res->hr_resuid);
300 		pjdlog_error("%s", errmsg);
301 		nv_add_string(nvout, errmsg, "errmsg");
302 		if (hast_proto_send(res, res->hr_remotein, nvout,
303 		    NULL, 0) == -1) {
304 			pjdlog_exit(EX_TEMPFAIL,
305 			    "Unable to send response to %s",
306 			    res->hr_remoteaddr);
307 		}
308 		nv_free(nvout);
309 		exit(EX_CONFIG);
310 	} else if (
311 	    /* Is primary out-of-date? */
312 	    (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
313 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
314 	    /* Are the nodes more or less in sync? */
315 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
316 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
317 	    /* Is secondary out-of-date? */
318 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
319 	     res->hr_secondary_remotecnt < res->hr_primary_localcnt)) {
320 		/*
321 		 * Nodes are more or less in sync or one of the nodes is
322 		 * out-of-date.
323 		 * It doesn't matter at this point which one, we just have to
324 		 * send out local bitmap to the remote node.
325 		 */
326 		if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) !=
327 		    (ssize_t)mapsize) {
328 			pjdlog_exit(LOG_ERR, "Unable to read activemap");
329 		}
330 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
331 		     res->hr_secondary_remotecnt == res->hr_primary_localcnt) {
332 			/* Primary is out-of-date, sync from secondary. */
333 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
334 		} else {
335 			/*
336 			 * Secondary is out-of-date or counts match.
337 			 * Sync from primary.
338 			 */
339 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
340 		}
341 	} else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
342 	     res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
343 		/*
344 		 * Not good, we have split-brain condition.
345 		 */
346 		free(map);
347 		pjdlog_error("Split-brain detected, exiting.");
348 		nv_add_string(nvout, "Split-brain condition!", "errmsg");
349 		if (hast_proto_send(res, res->hr_remotein, nvout,
350 		    NULL, 0) == -1) {
351 			pjdlog_exit(EX_TEMPFAIL,
352 			    "Unable to send response to %s",
353 			    res->hr_remoteaddr);
354 		}
355 		nv_free(nvout);
356 		/* Exit on split-brain. */
357 		event_send(res, EVENT_SPLITBRAIN);
358 		exit(EX_CONFIG);
359 	} else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
360 	    res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ {
361 		/*
362 		 * This should never happen in practise, but we will perform
363 		 * full synchronization.
364 		 */
365 		PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
366 		    res->hr_primary_localcnt < res->hr_secondary_remotecnt);
367 		mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
368 		    METADATA_SIZE, res->hr_extentsize,
369 		    res->hr_local_sectorsize);
370 		memset(map, 0xff, mapsize);
371 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) {
372 			/* In this one of five cases sync from secondary. */
373 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
374 		} else {
375 			/* For the rest four cases sync from primary. */
376 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
377 		}
378 		pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).",
379 		    (uintmax_t)res->hr_primary_localcnt,
380 		    (uintmax_t)res->hr_primary_remotecnt,
381 		    (uintmax_t)res->hr_secondary_localcnt,
382 		    (uintmax_t)res->hr_secondary_remotecnt);
383 	}
384 	nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize");
385 	if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) {
386 		pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s",
387 		    res->hr_remoteaddr);
388 	}
389 	if (map != NULL)
390 		free(map);
391 	nv_free(nvout);
392 #ifdef notyet
393 	/* Setup direction. */
394 	if (proto_recv(res->hr_remotein, NULL, 0) == -1)
395 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
396 #endif
397 }
398 
399 void
400 hastd_secondary(struct hast_resource *res, struct nv *nvin)
401 {
402 	sigset_t mask;
403 	pthread_t td;
404 	pid_t pid;
405 	int error, mode, debuglevel;
406 
407 	/*
408 	 * Create communication channel between parent and child.
409 	 */
410 	if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) {
411 		KEEP_ERRNO((void)pidfile_remove(pfh));
412 		pjdlog_exit(EX_OSERR,
413 		    "Unable to create control sockets between parent and child");
414 	}
415 	/*
416 	 * Create communication channel between child and parent.
417 	 */
418 	if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) {
419 		KEEP_ERRNO((void)pidfile_remove(pfh));
420 		pjdlog_exit(EX_OSERR,
421 		    "Unable to create event sockets between child and parent");
422 	}
423 
424 	pid = fork();
425 	if (pid == -1) {
426 		KEEP_ERRNO((void)pidfile_remove(pfh));
427 		pjdlog_exit(EX_OSERR, "Unable to fork");
428 	}
429 
430 	if (pid > 0) {
431 		/* This is parent. */
432 		proto_close(res->hr_remotein);
433 		res->hr_remotein = NULL;
434 		proto_close(res->hr_remoteout);
435 		res->hr_remoteout = NULL;
436 		/* Declare that we are receiver. */
437 		proto_recv(res->hr_event, NULL, 0);
438 		/* Declare that we are sender. */
439 		proto_send(res->hr_ctrl, NULL, 0);
440 		res->hr_workerpid = pid;
441 		return;
442 	}
443 
444 	gres = res;
445 	mode = pjdlog_mode_get();
446 	debuglevel = pjdlog_debug_get();
447 
448 	/* Declare that we are sender. */
449 	proto_send(res->hr_event, NULL, 0);
450 	/* Declare that we are receiver. */
451 	proto_recv(res->hr_ctrl, NULL, 0);
452 	descriptors_cleanup(res);
453 
454 	descriptors_assert(res, mode);
455 
456 	pjdlog_init(mode);
457 	pjdlog_debug_set(debuglevel);
458 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
459 	setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
460 
461 	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
462 	PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
463 
464 	/* Error in setting timeout is not critical, but why should it fail? */
465 	if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1)
466 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
467 	if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1)
468 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
469 
470 	init_local(res);
471 	init_environment();
472 
473 	if (drop_privs(res) != 0)
474 		exit(EX_CONFIG);
475 	pjdlog_info("Privileges successfully dropped.");
476 
477 	/*
478 	 * Create the control thread before sending any event to the parent,
479 	 * as we can deadlock when parent sends control request to worker,
480 	 * but worker has no control thread started yet, so parent waits.
481 	 * In the meantime worker sends an event to the parent, but parent
482 	 * is unable to handle the event, because it waits for control
483 	 * request response.
484 	 */
485 	error = pthread_create(&td, NULL, ctrl_thread, res);
486 	PJDLOG_ASSERT(error == 0);
487 
488 	init_remote(res, nvin);
489 	event_send(res, EVENT_CONNECT);
490 
491 	error = pthread_create(&td, NULL, recv_thread, res);
492 	PJDLOG_ASSERT(error == 0);
493 	error = pthread_create(&td, NULL, disk_thread, res);
494 	PJDLOG_ASSERT(error == 0);
495 	(void)send_thread(res);
496 }
497 
498 static void
499 reqlog(int loglevel, int debuglevel, int error, struct hio *hio,
500     const char *fmt, ...)
501 {
502 	char msg[1024];
503 	va_list ap;
504 	int len;
505 
506 	va_start(ap, fmt);
507 	len = vsnprintf(msg, sizeof(msg), fmt, ap);
508 	va_end(ap);
509 	if ((size_t)len < sizeof(msg)) {
510 		switch (hio->hio_cmd) {
511 		case HIO_READ:
512 			(void)snprintf(msg + len, sizeof(msg) - len,
513 			    "READ(%ju, %ju).", (uintmax_t)hio->hio_offset,
514 			    (uintmax_t)hio->hio_length);
515 			break;
516 		case HIO_DELETE:
517 			(void)snprintf(msg + len, sizeof(msg) - len,
518 			    "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset,
519 			    (uintmax_t)hio->hio_length);
520 			break;
521 		case HIO_FLUSH:
522 			(void)snprintf(msg + len, sizeof(msg) - len, "FLUSH.");
523 			break;
524 		case HIO_WRITE:
525 			(void)snprintf(msg + len, sizeof(msg) - len,
526 			    "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset,
527 			    (uintmax_t)hio->hio_length);
528 			break;
529 		case HIO_KEEPALIVE:
530 			(void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE.");
531 			break;
532 		default:
533 			(void)snprintf(msg + len, sizeof(msg) - len,
534 			    "UNKNOWN(%u).", (unsigned int)hio->hio_cmd);
535 			break;
536 		}
537 	}
538 	pjdlog_common(loglevel, debuglevel, error, "%s", msg);
539 }
540 
541 static int
542 requnpack(struct hast_resource *res, struct hio *hio, struct nv *nv)
543 {
544 
545 	hio->hio_cmd = nv_get_uint8(nv, "cmd");
546 	if (hio->hio_cmd == 0) {
547 		pjdlog_error("Header contains no 'cmd' field.");
548 		hio->hio_error = EINVAL;
549 		goto end;
550 	}
551 	if (hio->hio_cmd != HIO_KEEPALIVE) {
552 		hio->hio_seq = nv_get_uint64(nv, "seq");
553 		if (hio->hio_seq == 0) {
554 			pjdlog_error("Header contains no 'seq' field.");
555 			hio->hio_error = EINVAL;
556 			goto end;
557 		}
558 	}
559 	switch (hio->hio_cmd) {
560 	case HIO_FLUSH:
561 	case HIO_KEEPALIVE:
562 		break;
563 	case HIO_WRITE:
564 		hio->hio_memsync = nv_exists(nv, "memsync");
565 		/* FALLTHROUGH */
566 	case HIO_READ:
567 	case HIO_DELETE:
568 		hio->hio_offset = nv_get_uint64(nv, "offset");
569 		if (nv_error(nv) != 0) {
570 			pjdlog_error("Header is missing 'offset' field.");
571 			hio->hio_error = EINVAL;
572 			goto end;
573 		}
574 		hio->hio_length = nv_get_uint64(nv, "length");
575 		if (nv_error(nv) != 0) {
576 			pjdlog_error("Header is missing 'length' field.");
577 			hio->hio_error = EINVAL;
578 			goto end;
579 		}
580 		if (hio->hio_length == 0) {
581 			pjdlog_error("Data length is zero.");
582 			hio->hio_error = EINVAL;
583 			goto end;
584 		}
585 		if (hio->hio_cmd != HIO_DELETE && hio->hio_length > MAXPHYS) {
586 			pjdlog_error("Data length is too large (%ju > %ju).",
587 			    (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);
588 			hio->hio_error = EINVAL;
589 			goto end;
590 		}
591 		if ((hio->hio_offset % res->hr_local_sectorsize) != 0) {
592 			pjdlog_error("Offset %ju is not multiple of sector size.",
593 			    (uintmax_t)hio->hio_offset);
594 			hio->hio_error = EINVAL;
595 			goto end;
596 		}
597 		if ((hio->hio_length % res->hr_local_sectorsize) != 0) {
598 			pjdlog_error("Length %ju is not multiple of sector size.",
599 			    (uintmax_t)hio->hio_length);
600 			hio->hio_error = EINVAL;
601 			goto end;
602 		}
603 		if (hio->hio_offset + hio->hio_length >
604 		    (uint64_t)res->hr_datasize) {
605 			pjdlog_error("Data offset is too large (%ju > %ju).",
606 			    (uintmax_t)(hio->hio_offset + hio->hio_length),
607 			    (uintmax_t)res->hr_datasize);
608 			hio->hio_error = EINVAL;
609 			goto end;
610 		}
611 		break;
612 	default:
613 		pjdlog_error("Header contains invalid 'cmd' (%hhu).",
614 		    hio->hio_cmd);
615 		hio->hio_error = EINVAL;
616 		goto end;
617 	}
618 	hio->hio_error = 0;
619 end:
620 	return (hio->hio_error);
621 }
622 
623 static __dead2 void
624 secondary_exit(int exitcode, const char *fmt, ...)
625 {
626 	va_list ap;
627 
628 	PJDLOG_ASSERT(exitcode != EX_OK);
629 	va_start(ap, fmt);
630 	pjdlogv_errno(LOG_ERR, fmt, ap);
631 	va_end(ap);
632 	event_send(gres, EVENT_DISCONNECT);
633 	exit(exitcode);
634 }
635 
636 /*
637  * Thread receives requests from the primary node.
638  */
639 static void *
640 recv_thread(void *arg)
641 {
642 	struct hast_resource *res = arg;
643 	struct hio *hio, *mshio;
644 	struct nv *nv;
645 
646 	for (;;) {
647 		pjdlog_debug(2, "recv: Taking free request.");
648 		QUEUE_TAKE(free, hio);
649 		pjdlog_debug(2, "recv: (%p) Got request.", hio);
650 		if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) {
651 			secondary_exit(EX_TEMPFAIL,
652 			    "Unable to receive request header");
653 		}
654 		if (requnpack(res, hio, nv) != 0) {
655 			nv_free(nv);
656 			pjdlog_debug(2,
657 			    "recv: (%p) Moving request to the send queue.",
658 			    hio);
659 			QUEUE_INSERT(send, hio);
660 			continue;
661 		}
662 		switch (hio->hio_cmd) {
663 		case HIO_READ:
664 			res->hr_stat_read++;
665 			break;
666 		case HIO_WRITE:
667 			res->hr_stat_write++;
668 			break;
669 		case HIO_DELETE:
670 			res->hr_stat_delete++;
671 			break;
672 		case HIO_FLUSH:
673 			res->hr_stat_flush++;
674 			break;
675 		case HIO_KEEPALIVE:
676 			break;
677 		default:
678 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
679 			    hio->hio_cmd);
680 		}
681 		reqlog(LOG_DEBUG, 2, -1, hio,
682 		    "recv: (%p) Got request header: ", hio);
683 		if (hio->hio_cmd == HIO_KEEPALIVE) {
684 			nv_free(nv);
685 			pjdlog_debug(2,
686 			    "recv: (%p) Moving request to the free queue.",
687 			    hio);
688 			hio_clear(hio);
689 			QUEUE_INSERT(free, hio);
690 			continue;
691 		} else if (hio->hio_cmd == HIO_WRITE) {
692 			if (hast_proto_recv_data(res, res->hr_remotein, nv,
693 			    hio->hio_data, MAXPHYS) == -1) {
694 				secondary_exit(EX_TEMPFAIL,
695 				    "Unable to receive request data");
696 			}
697 			if (hio->hio_memsync) {
698 				/*
699 				 * For memsync requests we expect two replies.
700 				 * Clone the hio so we can handle both of them.
701 				 */
702 				pjdlog_debug(2, "recv: Taking free request.");
703 				QUEUE_TAKE(free, mshio);
704 				pjdlog_debug(2, "recv: (%p) Got request.",
705 				    mshio);
706 				hio_copy(hio, mshio);
707 				mshio->hio_error = 0;
708 				/*
709 				 * We want to keep 'memsync' tag only on the
710 				 * request going onto send queue (mshio).
711 				 */
712 				hio->hio_memsync = false;
713 				pjdlog_debug(2,
714 				    "recv: (%p) Moving memsync request to the send queue.",
715 				    mshio);
716 				QUEUE_INSERT(send, mshio);
717 			}
718 		}
719 		nv_free(nv);
720 		pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.",
721 		    hio);
722 		QUEUE_INSERT(disk, hio);
723 	}
724 	/* NOTREACHED */
725 	return (NULL);
726 }
727 
728 /*
729  * Thread reads from or writes to local component and also handles DELETE and
730  * FLUSH requests.
731  */
732 static void *
733 disk_thread(void *arg)
734 {
735 	struct hast_resource *res = arg;
736 	struct hio *hio;
737 	ssize_t ret;
738 	bool clear_activemap, logerror;
739 
740 	clear_activemap = true;
741 
742 	for (;;) {
743 		pjdlog_debug(2, "disk: Taking request.");
744 		QUEUE_TAKE(disk, hio);
745 		while (clear_activemap) {
746 			unsigned char *map;
747 			size_t mapsize;
748 
749 			/*
750 			 * When first request is received, it means that primary
751 			 * already received our activemap, merged it and stored
752 			 * locally. We can now safely clear our activemap.
753 			 */
754 			mapsize =
755 			    activemap_calc_ondisk_size(res->hr_local_mediasize -
756 			    METADATA_SIZE, res->hr_extentsize,
757 			    res->hr_local_sectorsize);
758 			map = calloc(1, mapsize);
759 			if (map == NULL) {
760 				pjdlog_warning("Unable to allocate memory to clear local activemap.");
761 				break;
762 			}
763 			if (pwrite(res->hr_localfd, map, mapsize,
764 			    METADATA_SIZE) != (ssize_t)mapsize) {
765 				pjdlog_errno(LOG_WARNING,
766 				    "Unable to store cleared activemap");
767 				free(map);
768 				res->hr_stat_activemap_write_error++;
769 				break;
770 			}
771 			free(map);
772 			clear_activemap = false;
773 			pjdlog_debug(1, "Local activemap cleared.");
774 			break;
775 		}
776 		reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio);
777 		logerror = true;
778 		/* Handle the actual request. */
779 		switch (hio->hio_cmd) {
780 		case HIO_READ:
781 			ret = pread(res->hr_localfd, hio->hio_data,
782 			    hio->hio_length,
783 			    hio->hio_offset + res->hr_localoff);
784 			if (ret == -1)
785 				hio->hio_error = errno;
786 			else if (ret != (int64_t)hio->hio_length)
787 				hio->hio_error = EIO;
788 			else
789 				hio->hio_error = 0;
790 			break;
791 		case HIO_WRITE:
792 			ret = pwrite(res->hr_localfd, hio->hio_data,
793 			    hio->hio_length,
794 			    hio->hio_offset + res->hr_localoff);
795 			if (ret == -1)
796 				hio->hio_error = errno;
797 			else if (ret != (int64_t)hio->hio_length)
798 				hio->hio_error = EIO;
799 			else
800 				hio->hio_error = 0;
801 			break;
802 		case HIO_DELETE:
803 			ret = g_delete(res->hr_localfd,
804 			    hio->hio_offset + res->hr_localoff,
805 			    hio->hio_length);
806 			if (ret == -1)
807 				hio->hio_error = errno;
808 			else
809 				hio->hio_error = 0;
810 			break;
811 		case HIO_FLUSH:
812 			if (!res->hr_localflush) {
813 				ret = -1;
814 				hio->hio_error = EOPNOTSUPP;
815 				logerror = false;
816 				break;
817 			}
818 			ret = g_flush(res->hr_localfd);
819 			if (ret == -1) {
820 				if (errno == EOPNOTSUPP)
821 					res->hr_localflush = false;
822 				hio->hio_error = errno;
823 			} else {
824 				hio->hio_error = 0;
825 			}
826 			break;
827 		default:
828 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
829 			    hio->hio_cmd);
830 		}
831 		if (logerror && hio->hio_error != 0) {
832 			reqlog(LOG_ERR, 0, hio->hio_error, hio,
833 			    "Request failed: ");
834 		}
835 		pjdlog_debug(2, "disk: (%p) Moving request to the send queue.",
836 		    hio);
837 		QUEUE_INSERT(send, hio);
838 	}
839 	/* NOTREACHED */
840 	return (NULL);
841 }
842 
843 /*
844  * Thread sends requests back to primary node.
845  */
846 static void *
847 send_thread(void *arg)
848 {
849 	struct hast_resource *res = arg;
850 	struct nv *nvout;
851 	struct hio *hio;
852 	void *data;
853 	size_t length;
854 
855 	for (;;) {
856 		pjdlog_debug(2, "send: Taking request.");
857 		QUEUE_TAKE(send, hio);
858 		reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio);
859 		nvout = nv_alloc();
860 		/* Copy sequence number. */
861 		nv_add_uint64(nvout, hio->hio_seq, "seq");
862 		if (hio->hio_memsync) {
863 			PJDLOG_ASSERT(hio->hio_cmd == HIO_WRITE);
864 			nv_add_int8(nvout, 1, "received");
865 		}
866 		switch (hio->hio_cmd) {
867 		case HIO_READ:
868 			if (hio->hio_error == 0) {
869 				data = hio->hio_data;
870 				length = hio->hio_length;
871 				break;
872 			}
873 			/*
874 			 * We send no data in case of an error.
875 			 */
876 			/* FALLTHROUGH */
877 		case HIO_DELETE:
878 		case HIO_FLUSH:
879 		case HIO_WRITE:
880 			data = NULL;
881 			length = 0;
882 			break;
883 		default:
884 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
885 			    hio->hio_cmd);
886 		}
887 		if (hio->hio_error != 0) {
888 			switch (hio->hio_cmd) {
889 			case HIO_READ:
890 				res->hr_stat_read_error++;
891 				break;
892 			case HIO_WRITE:
893 				res->hr_stat_write_error++;
894 				break;
895 			case HIO_DELETE:
896 				res->hr_stat_delete_error++;
897 				break;
898 			case HIO_FLUSH:
899 				res->hr_stat_flush_error++;
900 				break;
901 			}
902 			nv_add_int16(nvout, hio->hio_error, "error");
903 		}
904 		if (hast_proto_send(res, res->hr_remoteout, nvout, data,
905 		    length) == -1) {
906 			secondary_exit(EX_TEMPFAIL, "Unable to send reply");
907 		}
908 		nv_free(nvout);
909 		pjdlog_debug(2, "send: (%p) Moving request to the free queue.",
910 		    hio);
911 		hio_clear(hio);
912 		QUEUE_INSERT(free, hio);
913 	}
914 	/* NOTREACHED */
915 	return (NULL);
916 }
917