2 * Copyright (c) 2009-2010 The FreeBSD Foundation
3 * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
6 * This software was developed by Pawel Jakub Dawidek under sponsorship from
7 * the FreeBSD Foundation.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
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.
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
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
52 #include <activemap.h>
59 #include "hast_proto.h"
75 TAILQ_ENTRY(hio) hio_next;
78 static struct hast_resource *gres;
81 * Free list holds unused structures. When free list is empty, we have to wait
82 * until some in-progress requests are freed.
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;
88 * Disk thread (the one that does I/O requests) takes requests from this list.
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;
94 * Thread that sends requests back to primary takes requests from this list.
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;
101 * Maximum number of outstanding I/O requests.
103 #define HAST_HIO_MAX 256
105 static void *recv_thread(void *arg);
106 static void *disk_thread(void *arg);
107 static void *send_thread(void *arg);
109 #define QUEUE_INSERT(name, hio) do { \
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); \
117 cv_broadcast(&hio_##name##_list_cond); \
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); \
125 TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \
126 mtx_unlock(&hio_##name##_list_lock); \
130 hio_clear(struct hio *hio)
135 hio->hio_cmd = HIO_UNDEF;
138 hio->hio_memsync = false;
142 hio_copy(const struct hio *srchio, struct hio *dsthio)
146 * We don't copy hio_error, hio_data and hio_next fields.
149 dsthio->hio_seq = srchio->hio_seq;
150 dsthio->hio_cmd = srchio->hio_cmd;
151 dsthio->hio_offset = srchio->hio_offset;
152 dsthio->hio_length = srchio->hio_length;
153 dsthio->hio_memsync = srchio->hio_memsync;
157 init_environment(void)
163 * Initialize lists, their locks and theirs condition variables.
165 TAILQ_INIT(&hio_free_list);
166 mtx_init(&hio_free_list_lock);
167 cv_init(&hio_free_list_cond);
168 TAILQ_INIT(&hio_disk_list);
169 mtx_init(&hio_disk_list_lock);
170 cv_init(&hio_disk_list_cond);
171 TAILQ_INIT(&hio_send_list);
172 mtx_init(&hio_send_list_lock);
173 cv_init(&hio_send_list_cond);
176 * Allocate requests pool and initialize requests.
178 for (ii = 0; ii < HAST_HIO_MAX; ii++) {
179 hio = malloc(sizeof(*hio));
181 pjdlog_exitx(EX_TEMPFAIL,
182 "Unable to allocate memory (%zu bytes) for hio request.",
185 hio->hio_data = malloc(MAXPHYS);
186 if (hio->hio_data == NULL) {
187 pjdlog_exitx(EX_TEMPFAIL,
188 "Unable to allocate memory (%zu bytes) for gctl_data.",
192 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next);
197 init_local(struct hast_resource *res)
200 if (metadata_read(res, true) == -1)
205 init_remote(struct hast_resource *res, struct nv *nvin)
213 /* Setup direction. */
214 if (proto_send(res->hr_remoteout, NULL, 0) == -1)
215 pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
219 nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize");
220 nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize");
221 resuid = nv_get_uint64(nvin, "resuid");
222 res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt");
223 res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt");
224 nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt");
225 nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt");
226 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
227 METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize);
228 map = malloc(mapsize);
230 pjdlog_exitx(EX_TEMPFAIL,
231 "Unable to allocate memory (%zu bytes) for activemap.",
235 * When we work as primary and secondary is missing we will increase
236 * localcnt in our metadata. When secondary is connected and synced
237 * we make localcnt be equal to remotecnt, which means nodes are more
239 * Split-brain condition is when both nodes are not able to communicate
240 * and are both configured as primary nodes. In turn, they can both
241 * make incompatible changes to the data and we have to detect that.
242 * Under split-brain condition we will increase our localcnt on first
243 * write and remote node will increase its localcnt on first write.
244 * When we connect we can see that primary's localcnt is greater than
245 * our remotecnt (primary was modified while we weren't watching) and
246 * our localcnt is greater than primary's remotecnt (we were modified
247 * while primary wasn't watching).
248 * There are many possible combinations which are all gathered below.
249 * Don't pay too much attention to exact numbers, the more important
250 * is to compare them. We compare secondary's local with primary's
251 * remote and secondary's remote with primary's local.
252 * Note that every case where primary's localcnt is smaller than
253 * secondary's remotecnt and where secondary's localcnt is smaller than
254 * primary's remotecnt should be impossible in practise. We will perform
255 * full synchronization then. Those cases are marked with an asterisk.
256 * Regular synchronization means that only extents marked as dirty are
257 * synchronized (regular synchronization).
259 * SECONDARY METADATA PRIMARY METADATA
260 * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary.
261 * local=3 remote=3 local=2 remote=3* ?! Full sync from primary.
262 * local=3 remote=3 local=2 remote=4* ?! Full sync from primary.
263 * local=3 remote=3 local=3 remote=2 Primary is out-of-date,
264 * regular sync from secondary.
265 * local=3 remote=3 local=3 remote=3 Regular sync just in case.
266 * local=3 remote=3 local=3 remote=4* ?! Full sync from primary.
267 * local=3 remote=3 local=4 remote=2 Split-brain condition.
268 * local=3 remote=3 local=4 remote=3 Secondary out-of-date,
269 * regular sync from primary.
270 * local=3 remote=3 local=4 remote=4* ?! Full sync from primary.
272 if (res->hr_resuid == 0) {
274 * Provider is used for the first time. If primary node done no
275 * writes yet as well (we will find "virgin" argument) then
276 * there is no need to synchronize anything. If primary node
277 * done any writes already we have to synchronize everything.
279 PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
280 res->hr_resuid = resuid;
281 if (metadata_write(res) == -1)
283 if (nv_exists(nvin, "virgin")) {
288 memset(map, 0xff, mapsize);
290 nv_add_int8(nvout, 1, "virgin");
291 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
292 } else if (res->hr_resuid != resuid) {
296 (void)snprintf(errmsg, sizeof(errmsg),
297 "Resource unique ID mismatch (primary=%ju, secondary=%ju).",
298 (uintmax_t)resuid, (uintmax_t)res->hr_resuid);
299 pjdlog_error("%s", errmsg);
300 nv_add_string(nvout, errmsg, "errmsg");
301 if (hast_proto_send(res, res->hr_remotein, nvout,
303 pjdlog_exit(EX_TEMPFAIL,
304 "Unable to send response to %s",
310 /* Is primary out-of-date? */
311 (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
312 res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
313 /* Are the nodes more or less in sync? */
314 (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
315 res->hr_secondary_remotecnt == res->hr_primary_localcnt) ||
316 /* Is secondary out-of-date? */
317 (res->hr_secondary_localcnt == res->hr_primary_remotecnt &&
318 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) {
320 * Nodes are more or less in sync or one of the nodes is
322 * It doesn't matter at this point which one, we just have to
323 * send out local bitmap to the remote node.
325 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) !=
327 pjdlog_exit(LOG_ERR, "Unable to read activemap");
329 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
330 res->hr_secondary_remotecnt == res->hr_primary_localcnt) {
331 /* Primary is out-of-date, sync from secondary. */
332 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
335 * Secondary is out-of-date or counts match.
338 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
340 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt &&
341 res->hr_primary_localcnt > res->hr_secondary_remotecnt) {
343 * Not good, we have split-brain condition.
346 pjdlog_error("Split-brain detected, exiting.");
347 nv_add_string(nvout, "Split-brain condition!", "errmsg");
348 if (hast_proto_send(res, res->hr_remotein, nvout,
350 pjdlog_exit(EX_TEMPFAIL,
351 "Unable to send response to %s",
355 /* Exit on split-brain. */
356 event_send(res, EVENT_SPLITBRAIN);
358 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
359 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ {
361 * This should never happen in practise, but we will perform
362 * full synchronization.
364 PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt ||
365 res->hr_primary_localcnt < res->hr_secondary_remotecnt);
366 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize -
367 METADATA_SIZE, res->hr_extentsize,
368 res->hr_local_sectorsize);
369 memset(map, 0xff, mapsize);
370 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) {
371 /* In this one of five cases sync from secondary. */
372 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc");
374 /* For the rest four cases sync from primary. */
375 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc");
377 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).",
378 (uintmax_t)res->hr_primary_localcnt,
379 (uintmax_t)res->hr_primary_remotecnt,
380 (uintmax_t)res->hr_secondary_localcnt,
381 (uintmax_t)res->hr_secondary_remotecnt);
383 nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize");
384 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) {
385 pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s",
392 /* Setup direction. */
393 if (proto_recv(res->hr_remotein, NULL, 0) == -1)
394 pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
399 hastd_secondary(struct hast_resource *res, struct nv *nvin)
404 int error, mode, debuglevel;
407 * Create communication channel between parent and child.
409 if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) {
410 KEEP_ERRNO((void)pidfile_remove(pfh));
411 pjdlog_exit(EX_OSERR,
412 "Unable to create control sockets between parent and child");
415 * Create communication channel between child and parent.
417 if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) {
418 KEEP_ERRNO((void)pidfile_remove(pfh));
419 pjdlog_exit(EX_OSERR,
420 "Unable to create event sockets between child and parent");
425 KEEP_ERRNO((void)pidfile_remove(pfh));
426 pjdlog_exit(EX_OSERR, "Unable to fork");
430 /* This is parent. */
431 proto_close(res->hr_remotein);
432 res->hr_remotein = NULL;
433 proto_close(res->hr_remoteout);
434 res->hr_remoteout = NULL;
435 /* Declare that we are receiver. */
436 proto_recv(res->hr_event, NULL, 0);
437 /* Declare that we are sender. */
438 proto_send(res->hr_ctrl, NULL, 0);
439 res->hr_workerpid = pid;
444 mode = pjdlog_mode_get();
445 debuglevel = pjdlog_debug_get();
447 /* Declare that we are sender. */
448 proto_send(res->hr_event, NULL, 0);
449 /* Declare that we are receiver. */
450 proto_recv(res->hr_ctrl, NULL, 0);
451 descriptors_cleanup(res);
453 descriptors_assert(res, mode);
456 pjdlog_debug_set(debuglevel);
457 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
458 setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
460 PJDLOG_VERIFY(sigemptyset(&mask) == 0);
461 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
463 /* Error in setting timeout is not critical, but why should it fail? */
464 if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1)
465 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
466 if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1)
467 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
472 if (drop_privs(res) != 0)
474 pjdlog_info("Privileges successfully dropped.");
477 * Create the control thread before sending any event to the parent,
478 * as we can deadlock when parent sends control request to worker,
479 * but worker has no control thread started yet, so parent waits.
480 * In the meantime worker sends an event to the parent, but parent
481 * is unable to handle the event, because it waits for control
484 error = pthread_create(&td, NULL, ctrl_thread, res);
485 PJDLOG_ASSERT(error == 0);
487 init_remote(res, nvin);
488 event_send(res, EVENT_CONNECT);
490 error = pthread_create(&td, NULL, recv_thread, res);
491 PJDLOG_ASSERT(error == 0);
492 error = pthread_create(&td, NULL, disk_thread, res);
493 PJDLOG_ASSERT(error == 0);
494 (void)send_thread(res);
498 reqlog(int loglevel, int debuglevel, int error, struct hio *hio,
499 const char *fmt, ...)
506 len = vsnprintf(msg, sizeof(msg), fmt, ap);
508 if ((size_t)len < sizeof(msg)) {
509 switch (hio->hio_cmd) {
511 (void)snprintf(msg + len, sizeof(msg) - len,
512 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset,
513 (uintmax_t)hio->hio_length);
516 (void)snprintf(msg + len, sizeof(msg) - len,
517 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset,
518 (uintmax_t)hio->hio_length);
521 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH.");
524 (void)snprintf(msg + len, sizeof(msg) - len,
525 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset,
526 (uintmax_t)hio->hio_length);
529 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE.");
532 (void)snprintf(msg + len, sizeof(msg) - len,
533 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd);
537 pjdlog_common(loglevel, debuglevel, error, "%s", msg);
541 requnpack(struct hast_resource *res, struct hio *hio, struct nv *nv)
544 hio->hio_cmd = nv_get_uint8(nv, "cmd");
545 if (hio->hio_cmd == 0) {
546 pjdlog_error("Header contains no 'cmd' field.");
547 hio->hio_error = EINVAL;
550 if (hio->hio_cmd != HIO_KEEPALIVE) {
551 hio->hio_seq = nv_get_uint64(nv, "seq");
552 if (hio->hio_seq == 0) {
553 pjdlog_error("Header contains no 'seq' field.");
554 hio->hio_error = EINVAL;
558 switch (hio->hio_cmd) {
563 hio->hio_memsync = nv_exists(nv, "memsync");
567 hio->hio_offset = nv_get_uint64(nv, "offset");
568 if (nv_error(nv) != 0) {
569 pjdlog_error("Header is missing 'offset' field.");
570 hio->hio_error = EINVAL;
573 hio->hio_length = nv_get_uint64(nv, "length");
574 if (nv_error(nv) != 0) {
575 pjdlog_error("Header is missing 'length' field.");
576 hio->hio_error = EINVAL;
579 if (hio->hio_length == 0) {
580 pjdlog_error("Data length is zero.");
581 hio->hio_error = EINVAL;
584 if (hio->hio_cmd != HIO_DELETE && hio->hio_length > MAXPHYS) {
585 pjdlog_error("Data length is too large (%ju > %ju).",
586 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS);
587 hio->hio_error = EINVAL;
590 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) {
591 pjdlog_error("Offset %ju is not multiple of sector size.",
592 (uintmax_t)hio->hio_offset);
593 hio->hio_error = EINVAL;
596 if ((hio->hio_length % res->hr_local_sectorsize) != 0) {
597 pjdlog_error("Length %ju is not multiple of sector size.",
598 (uintmax_t)hio->hio_length);
599 hio->hio_error = EINVAL;
602 if (hio->hio_offset + hio->hio_length >
603 (uint64_t)res->hr_datasize) {
604 pjdlog_error("Data offset is too large (%ju > %ju).",
605 (uintmax_t)(hio->hio_offset + hio->hio_length),
606 (uintmax_t)res->hr_datasize);
607 hio->hio_error = EINVAL;
612 pjdlog_error("Header contains invalid 'cmd' (%hhu).",
614 hio->hio_error = EINVAL;
619 return (hio->hio_error);
623 secondary_exit(int exitcode, const char *fmt, ...)
627 PJDLOG_ASSERT(exitcode != EX_OK);
629 pjdlogv_errno(LOG_ERR, fmt, ap);
631 event_send(gres, EVENT_DISCONNECT);
636 * Thread receives requests from the primary node.
639 recv_thread(void *arg)
641 struct hast_resource *res = arg;
642 struct hio *hio, *mshio;
646 pjdlog_debug(2, "recv: Taking free request.");
647 QUEUE_TAKE(free, hio);
648 pjdlog_debug(2, "recv: (%p) Got request.", hio);
649 if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) {
650 secondary_exit(EX_TEMPFAIL,
651 "Unable to receive request header");
653 if (requnpack(res, hio, nv) != 0) {
656 "recv: (%p) Moving request to the send queue.",
658 QUEUE_INSERT(send, hio);
661 switch (hio->hio_cmd) {
666 res->hr_stat_write++;
669 res->hr_stat_delete++;
672 res->hr_stat_flush++;
677 PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
680 reqlog(LOG_DEBUG, 2, -1, hio,
681 "recv: (%p) Got request header: ", hio);
682 if (hio->hio_cmd == HIO_KEEPALIVE) {
685 "recv: (%p) Moving request to the free queue.",
688 QUEUE_INSERT(free, hio);
690 } else if (hio->hio_cmd == HIO_WRITE) {
691 if (hast_proto_recv_data(res, res->hr_remotein, nv,
692 hio->hio_data, MAXPHYS) == -1) {
693 secondary_exit(EX_TEMPFAIL,
694 "Unable to receive request data");
696 if (hio->hio_memsync) {
698 * For memsync requests we expect two replies.
699 * Clone the hio so we can handle both of them.
701 pjdlog_debug(2, "recv: Taking free request.");
702 QUEUE_TAKE(free, mshio);
703 pjdlog_debug(2, "recv: (%p) Got request.",
705 hio_copy(hio, mshio);
706 mshio->hio_error = 0;
708 * We want to keep 'memsync' tag only on the
709 * request going onto send queue (mshio).
711 hio->hio_memsync = false;
713 "recv: (%p) Moving memsync request to the send queue.",
715 QUEUE_INSERT(send, mshio);
719 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.",
721 QUEUE_INSERT(disk, hio);
728 * Thread reads from or writes to local component and also handles DELETE and
732 disk_thread(void *arg)
734 struct hast_resource *res = arg;
737 bool clear_activemap, logerror;
739 clear_activemap = true;
742 pjdlog_debug(2, "disk: Taking request.");
743 QUEUE_TAKE(disk, hio);
744 while (clear_activemap) {
749 * When first request is received, it means that primary
750 * already received our activemap, merged it and stored
751 * locally. We can now safely clear our activemap.
754 activemap_calc_ondisk_size(res->hr_local_mediasize -
755 METADATA_SIZE, res->hr_extentsize,
756 res->hr_local_sectorsize);
757 map = calloc(1, mapsize);
759 pjdlog_warning("Unable to allocate memory to clear local activemap.");
762 if (pwrite(res->hr_localfd, map, mapsize,
763 METADATA_SIZE) != (ssize_t)mapsize) {
764 pjdlog_errno(LOG_WARNING,
765 "Unable to store cleared activemap");
767 res->hr_stat_activemap_write_error++;
771 clear_activemap = false;
772 pjdlog_debug(1, "Local activemap cleared.");
775 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio);
777 /* Handle the actual request. */
778 switch (hio->hio_cmd) {
780 ret = pread(res->hr_localfd, hio->hio_data,
782 hio->hio_offset + res->hr_localoff);
784 hio->hio_error = errno;
785 else if (ret != (int64_t)hio->hio_length)
786 hio->hio_error = EIO;
791 ret = pwrite(res->hr_localfd, hio->hio_data,
793 hio->hio_offset + res->hr_localoff);
795 hio->hio_error = errno;
796 else if (ret != (int64_t)hio->hio_length)
797 hio->hio_error = EIO;
802 ret = g_delete(res->hr_localfd,
803 hio->hio_offset + res->hr_localoff,
806 hio->hio_error = errno;
811 if (!res->hr_localflush) {
813 hio->hio_error = EOPNOTSUPP;
817 ret = g_flush(res->hr_localfd);
819 if (errno == EOPNOTSUPP)
820 res->hr_localflush = false;
821 hio->hio_error = errno;
827 PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
830 if (logerror && hio->hio_error != 0) {
831 reqlog(LOG_ERR, 0, hio->hio_error, hio,
834 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.",
836 QUEUE_INSERT(send, hio);
843 * Thread sends requests back to primary node.
846 send_thread(void *arg)
848 struct hast_resource *res = arg;
855 pjdlog_debug(2, "send: Taking request.");
856 QUEUE_TAKE(send, hio);
857 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio);
859 /* Copy sequence number. */
860 nv_add_uint64(nvout, hio->hio_seq, "seq");
861 if (hio->hio_memsync) {
862 PJDLOG_ASSERT(hio->hio_cmd == HIO_WRITE);
863 nv_add_int8(nvout, 1, "received");
865 switch (hio->hio_cmd) {
867 if (hio->hio_error == 0) {
868 data = hio->hio_data;
869 length = hio->hio_length;
873 * We send no data in case of an error.
883 PJDLOG_ABORT("Unexpected command (cmd=%hhu).",
886 if (hio->hio_error != 0) {
887 switch (hio->hio_cmd) {
889 res->hr_stat_read_error++;
892 res->hr_stat_write_error++;
895 res->hr_stat_delete_error++;
898 res->hr_stat_flush_error++;
901 nv_add_int16(nvout, hio->hio_error, "error");
903 if (hast_proto_send(res, res->hr_remoteout, nvout, data,
905 secondary_exit(EX_TEMPFAIL, "Unable to send reply");
908 pjdlog_debug(2, "send: (%p) Moving request to the free queue.",
911 QUEUE_INSERT(free, hio);