secondary.c revision 229509
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: stable/9/sbin/hastd/secondary.c 229509 2012-01-04 17:22:10Z trociny $");
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
67struct 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
77static 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 */
83static TAILQ_HEAD(, hio) hio_free_list;
84static pthread_mutex_t hio_free_list_lock;
85static pthread_cond_t hio_free_list_cond;
86/*
87 * Disk thread (the one that do I/O requests) takes requests from this list.
88 */
89static TAILQ_HEAD(, hio) hio_disk_list;
90static pthread_mutex_t hio_disk_list_lock;
91static 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 */
96static TAILQ_HEAD(, hio) hio_send_list;
97static pthread_mutex_t hio_send_list_lock;
98static 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
105static void *recv_thread(void *arg);
106static void *disk_thread(void *arg);
107static 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
129static void
130hio_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
140static void
141init_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
180static void
181init_local(struct hast_resource *res)
182{
183
184	if (metadata_read(res, true) < 0)
185		exit(EX_NOINPUT);
186}
187
188static void
189init_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	nvout = nv_alloc();
203	nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize");
204	nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize");
205	resuid = nv_get_uint64(nvin, "resuid");
206	res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt");
207	res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt");
208	nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt");
209	nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt");
210	mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
211	    METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize);
212	map = malloc(mapsize);
213	if (map == NULL) {
214		pjdlog_exitx(EX_TEMPFAIL,
215		    "Unable to allocate memory (%zu bytes) for activemap.",
216		    mapsize);
217	}
218	/*
219	 * When we work as primary and secondary is missing we will increase
220	 * localcnt in our metadata. When secondary is connected and synced
221	 * we make localcnt be equal to remotecnt, which means nodes are more
222	 * or less in sync.
223	 * Split-brain condition is when both nodes are not able to communicate
224	 * and are both configured as primary nodes. In turn, they can both
225	 * make incompatible changes to the data and we have to detect that.
226	 * Under split-brain condition we will increase our localcnt on first
227	 * write and remote node will increase its localcnt on first write.
228	 * When we connect we can see that primary's localcnt is greater than
229	 * our remotecnt (primary was modified while we weren't watching) and
230	 * our localcnt is greater than primary's remotecnt (we were modified
231	 * while primary wasn't watching).
232	 * There are many possible combinations which are all gathered below.
233	 * Don't pay too much attention to exact numbers, the more important
234	 * is to compare them. We compare secondary's local with primary's
235	 * remote and secondary's remote with primary's local.
236	 * Note that every case where primary's localcnt is smaller than
237	 * secondary's remotecnt and where secondary's localcnt is smaller than
238	 * primary's remotecnt should be impossible in practise. We will perform
239	 * full synchronization then. Those cases are marked with an asterisk.
240	 * Regular synchronization means that only extents marked as dirty are
241	 * synchronized (regular synchronization).
242	 *
243	 * SECONDARY METADATA PRIMARY METADATA
244	 * local=3 remote=3   local=2 remote=2*  ?! Full sync from secondary.
245	 * local=3 remote=3   local=2 remote=3*  ?! Full sync from primary.
246	 * local=3 remote=3   local=2 remote=4*  ?! Full sync from primary.
247	 * local=3 remote=3   local=3 remote=2   Primary is out-of-date,
248	 *                                       regular sync from secondary.
249	 * local=3 remote=3   local=3 remote=3   Regular sync just in case.
250	 * local=3 remote=3   local=3 remote=4*  ?! Full sync from primary.
251	 * local=3 remote=3   local=4 remote=2   Split-brain condition.
252	 * local=3 remote=3   local=4 remote=3   Secondary out-of-date,
253	 *                                       regular sync from primary.
254	 * local=3 remote=3   local=4 remote=4*  ?! Full sync from primary.
255	 */
256	if (res->hr_resuid == 0) {
257		/*
258		 * Provider is used for the first time. If primary node done no
259		 * writes yet as well (we will find "virgin" argument) then
260		 * there is no need to synchronize anything. If primary node
261		 * done any writes already we have to synchronize everything.
262		 */
263		PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
264		res->hr_resuid = resuid;
265		if (metadata_write(res) < 0)
266			exit(EX_NOINPUT);
267		if (nv_exists(nvin, "virgin")) {
268			free(map);
269			map = NULL;
270			mapsize = 0;
271		} else {
272			memset(map, 0xff, mapsize);
273		}
274		nv_add_int8(nvout, 1, "virgin");
275		nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
276	} else if (res->hr_resuid != resuid) {
277		char errmsg[256];
278
279		free(map);
280		(void)snprintf(errmsg, sizeof(errmsg),
281		    "Resource unique ID mismatch (primary=%ju, secondary=%ju).",
282		    (uintmax_t)resuid, (uintmax_t)res->hr_resuid);
283		pjdlog_error("%s", errmsg);
284		nv_add_string(nvout, errmsg, "errmsg");
285		if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) {
286			pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s",
287			    res->hr_remoteaddr);
288		}
289		nv_free(nvout);
290		exit(EX_CONFIG);
291	} else if (
292	    /* Is primary out-of-date? */
293	    (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
294	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
295	    /* Are the nodes more or less in sync? */
296	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
297	     res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
298	    /* Is secondary out-of-date? */
299	    (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
300	     res->hr_secondary_remotecnt < res->hr_primary_localcnt)) {
301		/*
302		 * Nodes are more or less in sync or one of the nodes is
303		 * out-of-date.
304		 * It doesn't matter at this point which one, we just have to
305		 * send out local bitmap to the remote node.
306		 */
307		if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) !=
308		    (ssize_t)mapsize) {
309			pjdlog_exit(LOG_ERR, "Unable to read activemap");
310		}
311		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
312		     res->hr_secondary_remotecnt == res->hr_primary_localcnt) {
313			/* Primary is out-of-date, sync from secondary. */
314			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
315		} else {
316			/*
317			 * Secondary is out-of-date or counts match.
318			 * Sync from primary.
319			 */
320			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
321		}
322	} else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
323	     res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
324		/*
325		 * Not good, we have split-brain condition.
326		 */
327		free(map);
328		pjdlog_error("Split-brain detected, exiting.");
329		nv_add_string(nvout, "Split-brain condition!", "errmsg");
330		if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) {
331			pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s",
332			    res->hr_remoteaddr);
333		}
334		nv_free(nvout);
335		/* Exit on split-brain. */
336		event_send(res, EVENT_SPLITBRAIN);
337		exit(EX_CONFIG);
338	} else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
339	    res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ {
340		/*
341		 * This should never happen in practise, but we will perform
342		 * full synchronization.
343		 */
344		PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
345		    res->hr_primary_localcnt < res->hr_secondary_remotecnt);
346		mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
347		    METADATA_SIZE, res->hr_extentsize,
348		    res->hr_local_sectorsize);
349		memset(map, 0xff, mapsize);
350		if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) {
351			/* In this one of five cases sync from secondary. */
352			nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
353		} else {
354			/* For the rest four cases sync from primary. */
355			nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
356		}
357		pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).",
358		    (uintmax_t)res->hr_primary_localcnt,
359		    (uintmax_t)res->hr_primary_remotecnt,
360		    (uintmax_t)res->hr_secondary_localcnt,
361		    (uintmax_t)res->hr_secondary_remotecnt);
362	}
363	nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize");
364	if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) {
365		pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s",
366		    res->hr_remoteaddr);
367	}
368	if (map != NULL)
369		free(map);
370	nv_free(nvout);
371#ifdef notyet
372	/* Setup direction. */
373	if (proto_recv(res->hr_remotein, NULL, 0) == -1)
374		pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
375#endif
376}
377
378void
379hastd_secondary(struct hast_resource *res, struct nv *nvin)
380{
381	sigset_t mask;
382	pthread_t td;
383	pid_t pid;
384	int error, mode, debuglevel;
385
386	/*
387	 * Create communication channel between parent and child.
388	 */
389	if (proto_client(NULL, "socketpair://", &res->hr_ctrl) < 0) {
390		KEEP_ERRNO((void)pidfile_remove(pfh));
391		pjdlog_exit(EX_OSERR,
392		    "Unable to create control sockets between parent and child");
393	}
394	/*
395	 * Create communication channel between child and parent.
396	 */
397	if (proto_client(NULL, "socketpair://", &res->hr_event) < 0) {
398		KEEP_ERRNO((void)pidfile_remove(pfh));
399		pjdlog_exit(EX_OSERR,
400		    "Unable to create event sockets between child and parent");
401	}
402
403	pid = fork();
404	if (pid < 0) {
405		KEEP_ERRNO((void)pidfile_remove(pfh));
406		pjdlog_exit(EX_OSERR, "Unable to fork");
407	}
408
409	if (pid > 0) {
410		/* This is parent. */
411		proto_close(res->hr_remotein);
412		res->hr_remotein = NULL;
413		proto_close(res->hr_remoteout);
414		res->hr_remoteout = NULL;
415		/* Declare that we are receiver. */
416		proto_recv(res->hr_event, NULL, 0);
417		/* Declare that we are sender. */
418		proto_send(res->hr_ctrl, NULL, 0);
419		res->hr_workerpid = pid;
420		return;
421	}
422
423	gres = res;
424	mode = pjdlog_mode_get();
425	debuglevel = pjdlog_debug_get();
426
427	/* Declare that we are sender. */
428	proto_send(res->hr_event, NULL, 0);
429	/* Declare that we are receiver. */
430	proto_recv(res->hr_ctrl, NULL, 0);
431	descriptors_cleanup(res);
432
433	descriptors_assert(res, mode);
434
435	pjdlog_init(mode);
436	pjdlog_debug_set(debuglevel);
437	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
438	setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
439
440	PJDLOG_VERIFY(sigemptyset(&mask) == 0);
441	PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
442
443	/* Error in setting timeout is not critical, but why should it fail? */
444	if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0)
445		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
446	if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0)
447		pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
448
449	init_local(res);
450	init_environment();
451
452	if (drop_privs(res) != 0)
453		exit(EX_CONFIG);
454	pjdlog_info("Privileges successfully dropped.");
455
456	/*
457	 * Create the control thread before sending any event to the parent,
458	 * as we can deadlock when parent sends control request to worker,
459	 * but worker has no control thread started yet, so parent waits.
460	 * In the meantime worker sends an event to the parent, but parent
461	 * is unable to handle the event, because it waits for control
462	 * request response.
463	 */
464	error = pthread_create(&td, NULL, ctrl_thread, res);
465	PJDLOG_ASSERT(error == 0);
466
467	init_remote(res, nvin);
468	event_send(res, EVENT_CONNECT);
469
470	error = pthread_create(&td, NULL, recv_thread, res);
471	PJDLOG_ASSERT(error == 0);
472	error = pthread_create(&td, NULL, disk_thread, res);
473	PJDLOG_ASSERT(error == 0);
474	(void)send_thread(res);
475}
476
477static void
478reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...)
479{
480	char msg[1024];
481	va_list ap;
482	int len;
483
484	va_start(ap, fmt);
485	len = vsnprintf(msg, sizeof(msg), fmt, ap);
486	va_end(ap);
487	if ((size_t)len < sizeof(msg)) {
488		switch (hio->hio_cmd) {
489		case HIO_READ:
490			(void)snprintf(msg + len, sizeof(msg) - len,
491			    "READ(%ju, %ju).", (uintmax_t)hio->hio_offset,
492			    (uintmax_t)hio->hio_length);
493			break;
494		case HIO_DELETE:
495			(void)snprintf(msg + len, sizeof(msg) - len,
496			    "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset,
497			    (uintmax_t)hio->hio_length);
498			break;
499		case HIO_FLUSH:
500			(void)snprintf(msg + len, sizeof(msg) - len, "FLUSH.");
501			break;
502		case HIO_WRITE:
503			(void)snprintf(msg + len, sizeof(msg) - len,
504			    "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset,
505			    (uintmax_t)hio->hio_length);
506			break;
507		case HIO_KEEPALIVE:
508			(void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE.");
509			break;
510		default:
511			(void)snprintf(msg + len, sizeof(msg) - len,
512			    "UNKNOWN(%u).", (unsigned int)hio->hio_cmd);
513			break;
514		}
515	}
516	pjdlog_common(loglevel, debuglevel, error, "%s", msg);
517}
518
519static int
520requnpack(struct hast_resource *res, struct hio *hio, struct nv *nv)
521{
522
523	hio->hio_cmd = nv_get_uint8(nv, "cmd");
524	if (hio->hio_cmd == 0) {
525		pjdlog_error("Header contains no 'cmd' field.");
526		hio->hio_error = EINVAL;
527		goto end;
528	}
529	if (hio->hio_cmd != HIO_KEEPALIVE) {
530		hio->hio_seq = nv_get_uint64(nv, "seq");
531		if (hio->hio_seq == 0) {
532			pjdlog_error("Header contains no 'seq' field.");
533			hio->hio_error = EINVAL;
534			goto end;
535		}
536	}
537	switch (hio->hio_cmd) {
538	case HIO_FLUSH:
539	case HIO_KEEPALIVE:
540		break;
541	case HIO_READ:
542	case HIO_WRITE:
543	case HIO_DELETE:
544		hio->hio_offset = nv_get_uint64(nv, "offset");
545		if (nv_error(nv) != 0) {
546			pjdlog_error("Header is missing 'offset' field.");
547			hio->hio_error = EINVAL;
548			goto end;
549		}
550		hio->hio_length = nv_get_uint64(nv, "length");
551		if (nv_error(nv) != 0) {
552			pjdlog_error("Header is missing 'length' field.");
553			hio->hio_error = EINVAL;
554			goto end;
555		}
556		if (hio->hio_length == 0) {
557			pjdlog_error("Data length is zero.");
558			hio->hio_error = EINVAL;
559			goto end;
560		}
561		if (hio->hio_length > MAXPHYS) {
562			pjdlog_error("Data length is too large (%ju > %ju).",
563			    (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);
564			hio->hio_error = EINVAL;
565			goto end;
566		}
567		if ((hio->hio_offset % res->hr_local_sectorsize) != 0) {
568			pjdlog_error("Offset %ju is not multiple of sector size.",
569			    (uintmax_t)hio->hio_offset);
570			hio->hio_error = EINVAL;
571			goto end;
572		}
573		if ((hio->hio_length % res->hr_local_sectorsize) != 0) {
574			pjdlog_error("Length %ju is not multiple of sector size.",
575			    (uintmax_t)hio->hio_length);
576			hio->hio_error = EINVAL;
577			goto end;
578		}
579		if (hio->hio_offset + hio->hio_length >
580		    (uint64_t)res->hr_datasize) {
581			pjdlog_error("Data offset is too large (%ju > %ju).",
582			    (uintmax_t)(hio->hio_offset + hio->hio_length),
583			    (uintmax_t)res->hr_datasize);
584			hio->hio_error = EINVAL;
585			goto end;
586		}
587		break;
588	default:
589		pjdlog_error("Header contains invalid 'cmd' (%hhu).",
590		    hio->hio_cmd);
591		hio->hio_error = EINVAL;
592		goto end;
593	}
594	hio->hio_error = 0;
595end:
596	return (hio->hio_error);
597}
598
599static __dead2 void
600secondary_exit(int exitcode, const char *fmt, ...)
601{
602	va_list ap;
603
604	PJDLOG_ASSERT(exitcode != EX_OK);
605	va_start(ap, fmt);
606	pjdlogv_errno(LOG_ERR, fmt, ap);
607	va_end(ap);
608	event_send(gres, EVENT_DISCONNECT);
609	exit(exitcode);
610}
611
612/*
613 * Thread receives requests from the primary node.
614 */
615static void *
616recv_thread(void *arg)
617{
618	struct hast_resource *res = arg;
619	struct hio *hio;
620	struct nv *nv;
621
622	for (;;) {
623		pjdlog_debug(2, "recv: Taking free request.");
624		QUEUE_TAKE(free, hio);
625		pjdlog_debug(2, "recv: (%p) Got request.", hio);
626		if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) {
627			secondary_exit(EX_TEMPFAIL,
628			    "Unable to receive request header");
629		}
630		if (requnpack(res, hio, nv) != 0) {
631			nv_free(nv);
632			pjdlog_debug(2,
633			    "recv: (%p) Moving request to the send queue.",
634			    hio);
635			QUEUE_INSERT(send, hio);
636			continue;
637		}
638		switch (hio->hio_cmd) {
639		case HIO_READ:
640			res->hr_stat_read++;
641			break;
642		case HIO_WRITE:
643			res->hr_stat_write++;
644			break;
645		case HIO_DELETE:
646			res->hr_stat_delete++;
647			break;
648		case HIO_FLUSH:
649			res->hr_stat_flush++;
650			break;
651		case HIO_KEEPALIVE:
652			break;
653		default:
654			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
655			    hio->hio_cmd);
656		}
657		reqlog(LOG_DEBUG, 2, -1, hio,
658		    "recv: (%p) Got request header: ", hio);
659		if (hio->hio_cmd == HIO_KEEPALIVE) {
660			nv_free(nv);
661			pjdlog_debug(2,
662			    "recv: (%p) Moving request to the free queue.",
663			    hio);
664			hio_clear(hio);
665			QUEUE_INSERT(free, hio);
666			continue;
667		} else if (hio->hio_cmd == HIO_WRITE) {
668			if (hast_proto_recv_data(res, res->hr_remotein, nv,
669			    hio->hio_data, MAXPHYS) < 0) {
670				secondary_exit(EX_TEMPFAIL,
671				    "Unable to receive request data");
672			}
673		}
674		nv_free(nv);
675		pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.",
676		    hio);
677		QUEUE_INSERT(disk, hio);
678	}
679	/* NOTREACHED */
680	return (NULL);
681}
682
683/*
684 * Thread reads from or writes to local component and also handles DELETE and
685 * FLUSH requests.
686 */
687static void *
688disk_thread(void *arg)
689{
690	struct hast_resource *res = arg;
691	struct hio *hio;
692	ssize_t ret;
693	bool clear_activemap, logerror;
694
695	clear_activemap = true;
696
697	for (;;) {
698		pjdlog_debug(2, "disk: Taking request.");
699		QUEUE_TAKE(disk, hio);
700		while (clear_activemap) {
701			unsigned char *map;
702			size_t mapsize;
703
704			/*
705			 * When first request is received, it means that primary
706			 * already received our activemap, merged it and stored
707			 * locally. We can now safely clear our activemap.
708			 */
709			mapsize =
710			    activemap_calc_ondisk_size(res->hr_local_mediasize -
711			    METADATA_SIZE, res->hr_extentsize,
712			    res->hr_local_sectorsize);
713			map = calloc(1, mapsize);
714			if (map == NULL) {
715				pjdlog_warning("Unable to allocate memory to clear local activemap.");
716				break;
717			}
718			if (pwrite(res->hr_localfd, map, mapsize,
719			    METADATA_SIZE) != (ssize_t)mapsize) {
720				pjdlog_errno(LOG_WARNING,
721				    "Unable to store cleared activemap");
722				free(map);
723				break;
724			}
725			free(map);
726			clear_activemap = false;
727			pjdlog_debug(1, "Local activemap cleared.");
728			break;
729		}
730		reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio);
731		logerror = true;
732		/* Handle the actual request. */
733		switch (hio->hio_cmd) {
734		case HIO_READ:
735			ret = pread(res->hr_localfd, hio->hio_data,
736			    hio->hio_length,
737			    hio->hio_offset + res->hr_localoff);
738			if (ret < 0)
739				hio->hio_error = errno;
740			else if (ret != (int64_t)hio->hio_length)
741				hio->hio_error = EIO;
742			else
743				hio->hio_error = 0;
744			break;
745		case HIO_WRITE:
746			ret = pwrite(res->hr_localfd, hio->hio_data,
747			    hio->hio_length,
748			    hio->hio_offset + res->hr_localoff);
749			if (ret < 0)
750				hio->hio_error = errno;
751			else if (ret != (int64_t)hio->hio_length)
752				hio->hio_error = EIO;
753			else
754				hio->hio_error = 0;
755			break;
756		case HIO_DELETE:
757			ret = g_delete(res->hr_localfd,
758			    hio->hio_offset + res->hr_localoff,
759			    hio->hio_length);
760			if (ret < 0)
761				hio->hio_error = errno;
762			else
763				hio->hio_error = 0;
764			break;
765		case HIO_FLUSH:
766			if (!res->hr_localflush) {
767				ret = -1;
768				hio->hio_error = EOPNOTSUPP;
769				logerror = false;
770				break;
771			}
772			ret = g_flush(res->hr_localfd);
773			if (ret < 0) {
774				if (errno == EOPNOTSUPP)
775					res->hr_localflush = false;
776				hio->hio_error = errno;
777			} else {
778				hio->hio_error = 0;
779			}
780			break;
781		default:
782			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
783			    hio->hio_cmd);
784		}
785		if (logerror && hio->hio_error != 0) {
786			reqlog(LOG_ERR, 0, hio->hio_error, hio,
787			    "Request failed: ");
788		}
789		pjdlog_debug(2, "disk: (%p) Moving request to the send queue.",
790		    hio);
791		QUEUE_INSERT(send, hio);
792	}
793	/* NOTREACHED */
794	return (NULL);
795}
796
797/*
798 * Thread sends requests back to primary node.
799 */
800static void *
801send_thread(void *arg)
802{
803	struct hast_resource *res = arg;
804	struct nv *nvout;
805	struct hio *hio;
806	void *data;
807	size_t length;
808
809	for (;;) {
810		pjdlog_debug(2, "send: Taking request.");
811		QUEUE_TAKE(send, hio);
812		reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio);
813		nvout = nv_alloc();
814		/* Copy sequence number. */
815		nv_add_uint64(nvout, hio->hio_seq, "seq");
816		switch (hio->hio_cmd) {
817		case HIO_READ:
818			if (hio->hio_error == 0) {
819				data = hio->hio_data;
820				length = hio->hio_length;
821				break;
822			}
823			/*
824			 * We send no data in case of an error.
825			 */
826			/* FALLTHROUGH */
827		case HIO_DELETE:
828		case HIO_FLUSH:
829		case HIO_WRITE:
830			data = NULL;
831			length = 0;
832			break;
833		default:
834			PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
835			    hio->hio_cmd);
836		}
837		if (hio->hio_error != 0)
838			nv_add_int16(nvout, hio->hio_error, "error");
839		if (hast_proto_send(res, res->hr_remoteout, nvout, data,
840		    length) < 0) {
841			secondary_exit(EX_TEMPFAIL, "Unable to send reply.");
842		}
843		nv_free(nvout);
844		pjdlog_debug(2, "send: (%p) Moving request to the free queue.",
845		    hio);
846		hio_clear(hio);
847		QUEUE_INSERT(free, hio);
848	}
849	/* NOTREACHED */
850	return (NULL);
851}
852