xref: /freebsd/sbin/hastd/secondary.c (revision f5463265955b829775bbb32e1fd0bc11dafc36ce)
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/param.h>
34 #include <sys/time.h>
35 #include <sys/bio.h>
36 #include <sys/disk.h>
37 #include <sys/stat.h>
38 
39 #include <err.h>
40 #include <errno.h>
41 #include <fcntl.h>
42 #include <libgeom.h>
43 #include <pthread.h>
44 #include <signal.h>
45 #include <stdint.h>
46 #include <stdio.h>
47 #include <string.h>
48 #include <sysexits.h>
49 #include <unistd.h>
50 
51 #include <activemap.h>
52 #include <nv.h>
53 #include <pjdlog.h>
54 
55 #include "control.h"
56 #include "event.h"
57 #include "hast.h"
58 #include "hast_proto.h"
59 #include "hastd.h"
60 #include "hooks.h"
61 #include "metadata.h"
62 #include "proto.h"
63 #include "subr.h"
64 #include "synch.h"
65 
66 struct hio {
67 	uint64_t	 hio_seq;
68 	int		 hio_error;
69 	void		*hio_data;
70 	uint8_t		 hio_cmd;
71 	uint64_t	 hio_offset;
72 	uint64_t	 hio_length;
73 	bool		 hio_memsync;
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 size_t hio_free_list_size;
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 does I/O requests) takes requests from this list.
89  */
90 static TAILQ_HEAD(, hio) hio_disk_list;
91 static size_t hio_disk_list_size;
92 static pthread_mutex_t hio_disk_list_lock;
93 static pthread_cond_t hio_disk_list_cond;
94 /*
95  * Thread that sends requests back to primary takes requests from this list.
96  */
97 static TAILQ_HEAD(, hio) hio_send_list;
98 static size_t hio_send_list_size;
99 static pthread_mutex_t hio_send_list_lock;
100 static pthread_cond_t hio_send_list_cond;
101 
102 /*
103  * Maximum number of outstanding I/O requests.
104  */
105 #define	HAST_HIO_MAX	256
106 
107 static void *recv_thread(void *arg);
108 static void *disk_thread(void *arg);
109 static void *send_thread(void *arg);
110 
111 #define	QUEUE_INSERT(name, hio)	do {					\
112 	mtx_lock(&hio_##name##_list_lock);				\
113 	if (TAILQ_EMPTY(&hio_##name##_list))				\
114 		cv_broadcast(&hio_##name##_list_cond);			\
115 	TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next);		\
116 	hio_##name##_list_size++;					\
117 	mtx_unlock(&hio_##name##_list_lock);				\
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 	PJDLOG_ASSERT(hio_##name##_list_size != 0);			\
126 	hio_##name##_list_size--;					\
127 	TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next);		\
128 	mtx_unlock(&hio_##name##_list_lock);				\
129 } while (0)
130 
131 static void
132 output_status_aux(struct nv *nvout)
133 {
134 
135 	nv_add_uint64(nvout, (uint64_t)hio_free_list_size, "idle_queue_size");
136 	nv_add_uint64(nvout, (uint64_t)hio_disk_list_size, "local_queue_size");
137 	nv_add_uint64(nvout, (uint64_t)hio_send_list_size, "send_queue_size");
138 }
139 
140 static void
141 hio_clear(struct hio *hio)
142 {
143 
144 	hio->hio_seq = 0;
145 	hio->hio_error = 0;
146 	hio->hio_cmd = HIO_UNDEF;
147 	hio->hio_offset = 0;
148 	hio->hio_length = 0;
149 	hio->hio_memsync = false;
150 }
151 
152 static void
153 hio_copy(const struct hio *srchio, struct hio *dsthio)
154 {
155 
156 	/*
157 	 * We don't copy hio_error, hio_data and hio_next fields.
158 	 */
159 
160 	dsthio->hio_seq = srchio->hio_seq;
161 	dsthio->hio_cmd = srchio->hio_cmd;
162 	dsthio->hio_offset = srchio->hio_offset;
163 	dsthio->hio_length = srchio->hio_length;
164 	dsthio->hio_memsync = srchio->hio_memsync;
165 }
166 
167 static void
168 init_environment(void)
169 {
170 	struct hio *hio;
171 	unsigned int ii;
172 
173 	/*
174 	 * Initialize lists, their locks and theirs condition variables.
175 	 */
176 	TAILQ_INIT(&hio_free_list);
177 	mtx_init(&hio_free_list_lock);
178 	cv_init(&hio_free_list_cond);
179 	TAILQ_INIT(&hio_disk_list);
180 	mtx_init(&hio_disk_list_lock);
181 	cv_init(&hio_disk_list_cond);
182 	TAILQ_INIT(&hio_send_list);
183 	mtx_init(&hio_send_list_lock);
184 	cv_init(&hio_send_list_cond);
185 
186 	/*
187 	 * Allocate requests pool and initialize requests.
188 	 */
189 	for (ii = 0; ii < HAST_HIO_MAX; ii++) {
190 		hio = malloc(sizeof(*hio));
191 		if (hio == NULL) {
192 			pjdlog_exitx(EX_TEMPFAIL,
193 			    "Unable to allocate memory (%zu bytes) for hio request.",
194 			    sizeof(*hio));
195 		}
196 		hio->hio_data = malloc(MAXPHYS);
197 		if (hio->hio_data == NULL) {
198 			pjdlog_exitx(EX_TEMPFAIL,
199 			    "Unable to allocate memory (%zu bytes) for gctl_data.",
200 			    (size_t)MAXPHYS);
201 		}
202 		hio_clear(hio);
203 		TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);
204 		hio_free_list_size++;
205 	}
206 }
207 
208 static void
209 init_local(struct hast_resource *res)
210 {
211 
212 	if (metadata_read(res, true) == -1)
213 		exit(EX_NOINPUT);
214 }
215 
216 static void
217 init_remote(struct hast_resource *res, struct nv *nvin)
218 {
219 	uint64_t resuid;
220 	struct nv *nvout;
221 	unsigned char *map;
222 	size_t mapsize;
223 
224 #ifdef notyet
225 	/* Setup direction. */
226 	if (proto_send(res->hr_remoteout, NULL, 0) == -1)
227 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
228 #endif
229 
230 	nvout = nv_alloc();
231 	nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize");
232 	nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize");
233 	resuid = nv_get_uint64(nvin, "resuid");
234 	res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt");
235 	res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt");
236 	nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt");
237 	nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt");
238 	mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
239 	    METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize);
240 	map = malloc(mapsize);
241 	if (map == NULL) {
242 		pjdlog_exitx(EX_TEMPFAIL,
243 		    "Unable to allocate memory (%zu bytes) for activemap.",
244 		    mapsize);
245 	}
246 	/*
247 	 * When we work as primary and secondary is missing we will increase
248 	 * localcnt in our metadata. When secondary is connected and synced
249 	 * we make localcnt be equal to remotecnt, which means nodes are more
250 	 * or less in sync.
251 	 * Split-brain condition is when both nodes are not able to communicate
252 	 * and are both configured as primary nodes. In turn, they can both
253 	 * make incompatible changes to the data and we have to detect that.
254 	 * Under split-brain condition we will increase our localcnt on first
255 	 * write and remote node will increase its localcnt on first write.
256 	 * When we connect we can see that primary's localcnt is greater than
257 	 * our remotecnt (primary was modified while we weren't watching) and
258 	 * our localcnt is greater than primary's remotecnt (we were modified
259 	 * while primary wasn't watching).
260 	 * There are many possible combinations which are all gathered below.
261 	 * Don't pay too much attention to exact numbers, the more important
262 	 * is to compare them. We compare secondary's local with primary's
263 	 * remote and secondary's remote with primary's local.
264 	 * Note that every case where primary's localcnt is smaller than
265 	 * secondary's remotecnt and where secondary's localcnt is smaller than
266 	 * primary's remotecnt should be impossible in practise. We will perform
267 	 * full synchronization then. Those cases are marked with an asterisk.
268 	 * Regular synchronization means that only extents marked as dirty are
269 	 * synchronized (regular synchronization).
270 	 *
271 	 * SECONDARY METADATA PRIMARY METADATA
272 	 * local=3 remote=3   local=2 remote=2*  ?! Full sync from secondary.
273 	 * local=3 remote=3   local=2 remote=3*  ?! Full sync from primary.
274 	 * local=3 remote=3   local=2 remote=4*  ?! Full sync from primary.
275 	 * local=3 remote=3   local=3 remote=2   Primary is out-of-date,
276 	 *                                       regular sync from secondary.
277 	 * local=3 remote=3   local=3 remote=3   Regular sync just in case.
278 	 * local=3 remote=3   local=3 remote=4*  ?! Full sync from primary.
279 	 * local=3 remote=3   local=4 remote=2   Split-brain condition.
280 	 * local=3 remote=3   local=4 remote=3   Secondary out-of-date,
281 	 *                                       regular sync from primary.
282 	 * local=3 remote=3   local=4 remote=4*  ?! Full sync from primary.
283 	 */
284 	if (res->hr_resuid == 0) {
285 		/*
286 		 * Provider is used for the first time. If primary node done no
287 		 * writes yet as well (we will find "virgin" argument) then
288 		 * there is no need to synchronize anything. If primary node
289 		 * done any writes already we have to synchronize everything.
290 		 */
291 		PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
292 		res->hr_resuid = resuid;
293 		if (metadata_write(res) == -1)
294 			exit(EX_NOINPUT);
295 		if (nv_exists(nvin, "virgin")) {
296 			free(map);
297 			map = NULL;
298 			mapsize = 0;
299 		} else {
300 			memset(map, 0xff, mapsize);
301 		}
302 		nv_add_int8(nvout, 1, "virgin");
303 		nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
304 	} else if (res->hr_resuid != resuid) {
305 		char errmsg[256];
306 
307 		free(map);
308 		(void)snprintf(errmsg, sizeof(errmsg),
309 		    "Resource unique ID mismatch (primary=%ju, secondary=%ju).",
310 		    (uintmax_t)resuid, (uintmax_t)res->hr_resuid);
311 		pjdlog_error("%s", errmsg);
312 		nv_add_string(nvout, errmsg, "errmsg");
313 		if (hast_proto_send(res, res->hr_remotein, nvout,
314 		    NULL, 0) == -1) {
315 			pjdlog_exit(EX_TEMPFAIL,
316 			    "Unable to send response to %s",
317 			    res->hr_remoteaddr);
318 		}
319 		nv_free(nvout);
320 		exit(EX_CONFIG);
321 	} else if (
322 	    /* Is primary out-of-date? */
323 	    (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
324 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
325 	    /* Are the nodes more or less in sync? */
326 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
327 	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
328 	    /* Is secondary out-of-date? */
329 	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
330 	     res->hr_secondary_remotecnt < res->hr_primary_localcnt)) {
331 		/*
332 		 * Nodes are more or less in sync or one of the nodes is
333 		 * out-of-date.
334 		 * It doesn't matter at this point which one, we just have to
335 		 * send out local bitmap to the remote node.
336 		 */
337 		if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) !=
338 		    (ssize_t)mapsize) {
339 			pjdlog_exit(LOG_ERR, "Unable to read activemap");
340 		}
341 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
342 		     res->hr_secondary_remotecnt == res->hr_primary_localcnt) {
343 			/* Primary is out-of-date, sync from secondary. */
344 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
345 		} else {
346 			/*
347 			 * Secondary is out-of-date or counts match.
348 			 * Sync from primary.
349 			 */
350 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
351 		}
352 	} else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
353 	     res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
354 		/*
355 		 * Not good, we have split-brain condition.
356 		 */
357 		free(map);
358 		pjdlog_error("Split-brain detected, exiting.");
359 		nv_add_string(nvout, "Split-brain condition!", "errmsg");
360 		if (hast_proto_send(res, res->hr_remotein, nvout,
361 		    NULL, 0) == -1) {
362 			pjdlog_exit(EX_TEMPFAIL,
363 			    "Unable to send response to %s",
364 			    res->hr_remoteaddr);
365 		}
366 		nv_free(nvout);
367 		/* Exit on split-brain. */
368 		event_send(res, EVENT_SPLITBRAIN);
369 		exit(EX_CONFIG);
370 	} else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
371 	    res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ {
372 		/*
373 		 * This should never happen in practise, but we will perform
374 		 * full synchronization.
375 		 */
376 		PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
377 		    res->hr_primary_localcnt < res->hr_secondary_remotecnt);
378 		mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
379 		    METADATA_SIZE, res->hr_extentsize,
380 		    res->hr_local_sectorsize);
381 		memset(map, 0xff, mapsize);
382 		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) {
383 			/* In this one of five cases sync from secondary. */
384 			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
385 		} else {
386 			/* For the rest four cases sync from primary. */
387 			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
388 		}
389 		pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).",
390 		    (uintmax_t)res->hr_primary_localcnt,
391 		    (uintmax_t)res->hr_primary_remotecnt,
392 		    (uintmax_t)res->hr_secondary_localcnt,
393 		    (uintmax_t)res->hr_secondary_remotecnt);
394 	}
395 	nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize");
396 	if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) {
397 		pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s",
398 		    res->hr_remoteaddr);
399 	}
400 	if (map != NULL)
401 		free(map);
402 	nv_free(nvout);
403 #ifdef notyet
404 	/* Setup direction. */
405 	if (proto_recv(res->hr_remotein, NULL, 0) == -1)
406 		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
407 #endif
408 }
409 
410 void
411 hastd_secondary(struct hast_resource *res, struct nv *nvin)
412 {
413 	sigset_t mask;
414 	pthread_t td;
415 	pid_t pid;
416 	int error, mode, debuglevel;
417 
418 	/*
419 	 * Create communication channel between parent and child.
420 	 */
421 	if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) {
422 		KEEP_ERRNO((void)pidfile_remove(pfh));
423 		pjdlog_exit(EX_OSERR,
424 		    "Unable to create control sockets between parent and child");
425 	}
426 	/*
427 	 * Create communication channel between child and parent.
428 	 */
429 	if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) {
430 		KEEP_ERRNO((void)pidfile_remove(pfh));
431 		pjdlog_exit(EX_OSERR,
432 		    "Unable to create event sockets between child and parent");
433 	}
434 
435 	pid = fork();
436 	if (pid == -1) {
437 		KEEP_ERRNO((void)pidfile_remove(pfh));
438 		pjdlog_exit(EX_OSERR, "Unable to fork");
439 	}
440 
441 	if (pid > 0) {
442 		/* This is parent. */
443 		proto_close(res->hr_remotein);
444 		res->hr_remotein = NULL;
445 		proto_close(res->hr_remoteout);
446 		res->hr_remoteout = NULL;
447 		/* Declare that we are receiver. */
448 		proto_recv(res->hr_event, NULL, 0);
449 		/* Declare that we are sender. */
450 		proto_send(res->hr_ctrl, NULL, 0);
451 		res->hr_workerpid = pid;
452 		return;
453 	}
454 
455 	gres = res;
456 	res->output_status_aux = output_status_aux;
457 	mode = pjdlog_mode_get();
458 	debuglevel = pjdlog_debug_get();
459 
460 	/* Declare that we are sender. */
461 	proto_send(res->hr_event, NULL, 0);
462 	/* Declare that we are receiver. */
463 	proto_recv(res->hr_ctrl, NULL, 0);
464 	descriptors_cleanup(res);
465 
466 	descriptors_assert(res, mode);
467 
468 	pjdlog_init(mode);
469 	pjdlog_debug_set(debuglevel);
470 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
471 	setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
472 
473 	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
474 	PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
475 
476 	/* Error in setting timeout is not critical, but why should it fail? */
477 	if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1)
478 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
479 	if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1)
480 		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
481 
482 	init_local(res);
483 	init_environment();
484 
485 	if (drop_privs(res) != 0)
486 		exit(EX_CONFIG);
487 	pjdlog_info("Privileges successfully dropped.");
488 
489 	/*
490 	 * Create the control thread before sending any event to the parent,
491 	 * as we can deadlock when parent sends control request to worker,
492 	 * but worker has no control thread started yet, so parent waits.
493 	 * In the meantime worker sends an event to the parent, but parent
494 	 * is unable to handle the event, because it waits for control
495 	 * request response.
496 	 */
497 	error = pthread_create(&td, NULL, ctrl_thread, res);
498 	PJDLOG_ASSERT(error == 0);
499 
500 	init_remote(res, nvin);
501 	event_send(res, EVENT_CONNECT);
502 
503 	error = pthread_create(&td, NULL, recv_thread, res);
504 	PJDLOG_ASSERT(error == 0);
505 	error = pthread_create(&td, NULL, disk_thread, res);
506 	PJDLOG_ASSERT(error == 0);
507 	(void)send_thread(res);
508 }
509 
510 static void
511 reqlog(int loglevel, int debuglevel, int error, struct hio *hio,
512     const char *fmt, ...)
513 {
514 	char msg[1024];
515 	va_list ap;
516 	int len;
517 
518 	va_start(ap, fmt);
519 	len = vsnprintf(msg, sizeof(msg), fmt, ap);
520 	va_end(ap);
521 	if ((size_t)len < sizeof(msg)) {
522 		switch (hio->hio_cmd) {
523 		case HIO_READ:
524 			(void)snprintf(msg + len, sizeof(msg) - len,
525 			    "READ(%ju, %ju).", (uintmax_t)hio->hio_offset,
526 			    (uintmax_t)hio->hio_length);
527 			break;
528 		case HIO_DELETE:
529 			(void)snprintf(msg + len, sizeof(msg) - len,
530 			    "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset,
531 			    (uintmax_t)hio->hio_length);
532 			break;
533 		case HIO_FLUSH:
534 			(void)snprintf(msg + len, sizeof(msg) - len, "FLUSH.");
535 			break;
536 		case HIO_WRITE:
537 			(void)snprintf(msg + len, sizeof(msg) - len,
538 			    "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset,
539 			    (uintmax_t)hio->hio_length);
540 			break;
541 		case HIO_KEEPALIVE:
542 			(void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE.");
543 			break;
544 		default:
545 			(void)snprintf(msg + len, sizeof(msg) - len,
546 			    "UNKNOWN(%u).", (unsigned int)hio->hio_cmd);
547 			break;
548 		}
549 	}
550 	pjdlog_common(loglevel, debuglevel, error, "%s", msg);
551 }
552 
553 static int
554 requnpack(struct hast_resource *res, struct hio *hio, struct nv *nv)
555 {
556 
557 	hio->hio_cmd = nv_get_uint8(nv, "cmd");
558 	if (hio->hio_cmd == 0) {
559 		pjdlog_error("Header contains no 'cmd' field.");
560 		hio->hio_error = EINVAL;
561 		goto end;
562 	}
563 	if (hio->hio_cmd != HIO_KEEPALIVE) {
564 		hio->hio_seq = nv_get_uint64(nv, "seq");
565 		if (hio->hio_seq == 0) {
566 			pjdlog_error("Header contains no 'seq' field.");
567 			hio->hio_error = EINVAL;
568 			goto end;
569 		}
570 	}
571 	switch (hio->hio_cmd) {
572 	case HIO_FLUSH:
573 	case HIO_KEEPALIVE:
574 		break;
575 	case HIO_WRITE:
576 		hio->hio_memsync = nv_exists(nv, "memsync");
577 		/* FALLTHROUGH */
578 	case HIO_READ:
579 	case HIO_DELETE:
580 		hio->hio_offset = nv_get_uint64(nv, "offset");
581 		if (nv_error(nv) != 0) {
582 			pjdlog_error("Header is missing 'offset' field.");
583 			hio->hio_error = EINVAL;
584 			goto end;
585 		}
586 		hio->hio_length = nv_get_uint64(nv, "length");
587 		if (nv_error(nv) != 0) {
588 			pjdlog_error("Header is missing 'length' field.");
589 			hio->hio_error = EINVAL;
590 			goto end;
591 		}
592 		if (hio->hio_length == 0) {
593 			pjdlog_error("Data length is zero.");
594 			hio->hio_error = EINVAL;
595 			goto end;
596 		}
597 		if (hio->hio_cmd != HIO_DELETE && hio->hio_length > MAXPHYS) {
598 			pjdlog_error("Data length is too large (%ju > %ju).",
599 			    (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);
600 			hio->hio_error = EINVAL;
601 			goto end;
602 		}
603 		if ((hio->hio_offset % res->hr_local_sectorsize) != 0) {
604 			pjdlog_error("Offset %ju is not multiple of sector size.",
605 			    (uintmax_t)hio->hio_offset);
606 			hio->hio_error = EINVAL;
607 			goto end;
608 		}
609 		if ((hio->hio_length % res->hr_local_sectorsize) != 0) {
610 			pjdlog_error("Length %ju is not multiple of sector size.",
611 			    (uintmax_t)hio->hio_length);
612 			hio->hio_error = EINVAL;
613 			goto end;
614 		}
615 		if (hio->hio_offset + hio->hio_length >
616 		    (uint64_t)res->hr_datasize) {
617 			pjdlog_error("Data offset is too large (%ju > %ju).",
618 			    (uintmax_t)(hio->hio_offset + hio->hio_length),
619 			    (uintmax_t)res->hr_datasize);
620 			hio->hio_error = EINVAL;
621 			goto end;
622 		}
623 		break;
624 	default:
625 		pjdlog_error("Header contains invalid 'cmd' (%hhu).",
626 		    hio->hio_cmd);
627 		hio->hio_error = EINVAL;
628 		goto end;
629 	}
630 	hio->hio_error = 0;
631 end:
632 	return (hio->hio_error);
633 }
634 
635 static __dead2 void
636 secondary_exit(int exitcode, const char *fmt, ...)
637 {
638 	va_list ap;
639 
640 	PJDLOG_ASSERT(exitcode != EX_OK);
641 	va_start(ap, fmt);
642 	pjdlogv_errno(LOG_ERR, fmt, ap);
643 	va_end(ap);
644 	event_send(gres, EVENT_DISCONNECT);
645 	exit(exitcode);
646 }
647 
648 /*
649  * Thread receives requests from the primary node.
650  */
651 static void *
652 recv_thread(void *arg)
653 {
654 	struct hast_resource *res = arg;
655 	struct hio *hio, *mshio;
656 	struct nv *nv;
657 
658 	for (;;) {
659 		pjdlog_debug(2, "recv: Taking free request.");
660 		QUEUE_TAKE(free, hio);
661 		pjdlog_debug(2, "recv: (%p) Got request.", hio);
662 		if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) {
663 			secondary_exit(EX_TEMPFAIL,
664 			    "Unable to receive request header");
665 		}
666 		if (requnpack(res, hio, nv) != 0) {
667 			nv_free(nv);
668 			pjdlog_debug(2,
669 			    "recv: (%p) Moving request to the send queue.",
670 			    hio);
671 			QUEUE_INSERT(send, hio);
672 			continue;
673 		}
674 		switch (hio->hio_cmd) {
675 		case HIO_READ:
676 			res->hr_stat_read++;
677 			break;
678 		case HIO_WRITE:
679 			res->hr_stat_write++;
680 			break;
681 		case HIO_DELETE:
682 			res->hr_stat_delete++;
683 			break;
684 		case HIO_FLUSH:
685 			res->hr_stat_flush++;
686 			break;
687 		case HIO_KEEPALIVE:
688 			break;
689 		default:
690 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
691 			    hio->hio_cmd);
692 		}
693 		reqlog(LOG_DEBUG, 2, -1, hio,
694 		    "recv: (%p) Got request header: ", hio);
695 		if (hio->hio_cmd == HIO_KEEPALIVE) {
696 			nv_free(nv);
697 			pjdlog_debug(2,
698 			    "recv: (%p) Moving request to the free queue.",
699 			    hio);
700 			hio_clear(hio);
701 			QUEUE_INSERT(free, hio);
702 			continue;
703 		} else if (hio->hio_cmd == HIO_WRITE) {
704 			if (hast_proto_recv_data(res, res->hr_remotein, nv,
705 			    hio->hio_data, MAXPHYS) == -1) {
706 				secondary_exit(EX_TEMPFAIL,
707 				    "Unable to receive request data");
708 			}
709 			if (hio->hio_memsync) {
710 				/*
711 				 * For memsync requests we expect two replies.
712 				 * Clone the hio so we can handle both of them.
713 				 */
714 				pjdlog_debug(2, "recv: Taking free request.");
715 				QUEUE_TAKE(free, mshio);
716 				pjdlog_debug(2, "recv: (%p) Got request.",
717 				    mshio);
718 				hio_copy(hio, mshio);
719 				mshio->hio_error = 0;
720 				/*
721 				 * We want to keep 'memsync' tag only on the
722 				 * request going onto send queue (mshio).
723 				 */
724 				hio->hio_memsync = false;
725 				pjdlog_debug(2,
726 				    "recv: (%p) Moving memsync request to the send queue.",
727 				    mshio);
728 				QUEUE_INSERT(send, mshio);
729 			}
730 		}
731 		nv_free(nv);
732 		pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.",
733 		    hio);
734 		QUEUE_INSERT(disk, hio);
735 	}
736 	/* NOTREACHED */
737 	return (NULL);
738 }
739 
740 /*
741  * Thread reads from or writes to local component and also handles DELETE and
742  * FLUSH requests.
743  */
744 static void *
745 disk_thread(void *arg)
746 {
747 	struct hast_resource *res = arg;
748 	struct hio *hio;
749 	ssize_t ret;
750 	bool clear_activemap, logerror;
751 
752 	clear_activemap = true;
753 
754 	for (;;) {
755 		pjdlog_debug(2, "disk: Taking request.");
756 		QUEUE_TAKE(disk, hio);
757 		while (clear_activemap) {
758 			unsigned char *map;
759 			size_t mapsize;
760 
761 			/*
762 			 * When first request is received, it means that primary
763 			 * already received our activemap, merged it and stored
764 			 * locally. We can now safely clear our activemap.
765 			 */
766 			mapsize =
767 			    activemap_calc_ondisk_size(res->hr_local_mediasize -
768 			    METADATA_SIZE, res->hr_extentsize,
769 			    res->hr_local_sectorsize);
770 			map = calloc(1, mapsize);
771 			if (map == NULL) {
772 				pjdlog_warning("Unable to allocate memory to clear local activemap.");
773 				break;
774 			}
775 			if (pwrite(res->hr_localfd, map, mapsize,
776 			    METADATA_SIZE) != (ssize_t)mapsize) {
777 				pjdlog_errno(LOG_WARNING,
778 				    "Unable to store cleared activemap");
779 				free(map);
780 				res->hr_stat_activemap_write_error++;
781 				break;
782 			}
783 			free(map);
784 			clear_activemap = false;
785 			pjdlog_debug(1, "Local activemap cleared.");
786 			break;
787 		}
788 		reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio);
789 		logerror = true;
790 		/* Handle the actual request. */
791 		switch (hio->hio_cmd) {
792 		case HIO_READ:
793 			ret = pread(res->hr_localfd, hio->hio_data,
794 			    hio->hio_length,
795 			    hio->hio_offset + res->hr_localoff);
796 			if (ret == -1)
797 				hio->hio_error = errno;
798 			else if (ret != (int64_t)hio->hio_length)
799 				hio->hio_error = EIO;
800 			else
801 				hio->hio_error = 0;
802 			break;
803 		case HIO_WRITE:
804 			ret = pwrite(res->hr_localfd, hio->hio_data,
805 			    hio->hio_length,
806 			    hio->hio_offset + res->hr_localoff);
807 			if (ret == -1)
808 				hio->hio_error = errno;
809 			else if (ret != (int64_t)hio->hio_length)
810 				hio->hio_error = EIO;
811 			else
812 				hio->hio_error = 0;
813 			break;
814 		case HIO_DELETE:
815 			ret = g_delete(res->hr_localfd,
816 			    hio->hio_offset + res->hr_localoff,
817 			    hio->hio_length);
818 			if (ret == -1)
819 				hio->hio_error = errno;
820 			else
821 				hio->hio_error = 0;
822 			break;
823 		case HIO_FLUSH:
824 			if (!res->hr_localflush) {
825 				ret = -1;
826 				hio->hio_error = EOPNOTSUPP;
827 				logerror = false;
828 				break;
829 			}
830 			ret = g_flush(res->hr_localfd);
831 			if (ret == -1) {
832 				if (errno == EOPNOTSUPP)
833 					res->hr_localflush = false;
834 				hio->hio_error = errno;
835 			} else {
836 				hio->hio_error = 0;
837 			}
838 			break;
839 		default:
840 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
841 			    hio->hio_cmd);
842 		}
843 		if (logerror && hio->hio_error != 0) {
844 			reqlog(LOG_ERR, 0, hio->hio_error, hio,
845 			    "Request failed: ");
846 		}
847 		pjdlog_debug(2, "disk: (%p) Moving request to the send queue.",
848 		    hio);
849 		QUEUE_INSERT(send, hio);
850 	}
851 	/* NOTREACHED */
852 	return (NULL);
853 }
854 
855 /*
856  * Thread sends requests back to primary node.
857  */
858 static void *
859 send_thread(void *arg)
860 {
861 	struct hast_resource *res = arg;
862 	struct nv *nvout;
863 	struct hio *hio;
864 	void *data;
865 	size_t length;
866 
867 	for (;;) {
868 		pjdlog_debug(2, "send: Taking request.");
869 		QUEUE_TAKE(send, hio);
870 		reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio);
871 		nvout = nv_alloc();
872 		/* Copy sequence number. */
873 		nv_add_uint64(nvout, hio->hio_seq, "seq");
874 		if (hio->hio_memsync) {
875 			PJDLOG_ASSERT(hio->hio_cmd == HIO_WRITE);
876 			nv_add_int8(nvout, 1, "received");
877 		}
878 		switch (hio->hio_cmd) {
879 		case HIO_READ:
880 			if (hio->hio_error == 0) {
881 				data = hio->hio_data;
882 				length = hio->hio_length;
883 				break;
884 			}
885 			/*
886 			 * We send no data in case of an error.
887 			 */
888 			/* FALLTHROUGH */
889 		case HIO_DELETE:
890 		case HIO_FLUSH:
891 		case HIO_WRITE:
892 			data = NULL;
893 			length = 0;
894 			break;
895 		default:
896 			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
897 			    hio->hio_cmd);
898 		}
899 		if (hio->hio_error != 0) {
900 			switch (hio->hio_cmd) {
901 			case HIO_READ:
902 				res->hr_stat_read_error++;
903 				break;
904 			case HIO_WRITE:
905 				res->hr_stat_write_error++;
906 				break;
907 			case HIO_DELETE:
908 				res->hr_stat_delete_error++;
909 				break;
910 			case HIO_FLUSH:
911 				res->hr_stat_flush_error++;
912 				break;
913 			}
914 			nv_add_int16(nvout, hio->hio_error, "error");
915 		}
916 		if (hast_proto_send(res, res->hr_remoteout, nvout, data,
917 		    length) == -1) {
918 			secondary_exit(EX_TEMPFAIL, "Unable to send reply");
919 		}
920 		nv_free(nvout);
921 		pjdlog_debug(2, "send: (%p) Moving request to the free queue.",
922 		    hio);
923 		hio_clear(hio);
924 		QUEUE_INSERT(free, hio);
925 	}
926 	/* NOTREACHED */
927 	return (NULL);
928 }
929