secondary.c revision 210879
1/*- 2 * Copyright (c) 2009-2010 The FreeBSD Foundation 3 * All rights reserved. 4 * 5 * This software was developed by Pawel Jakub Dawidek under sponsorship from 6 * the FreeBSD Foundation. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30#include <sys/cdefs.h> 31__FBSDID("$FreeBSD: head/sbin/hastd/secondary.c 210879 2010-08-05 18:56:24Z pjd $"); 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 <assert.h> 40#include <err.h> 41#include <errno.h> 42#include <fcntl.h> 43#include <libgeom.h> 44#include <pthread.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 "hast.h" 57#include "hast_proto.h" 58#include "hastd.h" 59#include "metadata.h" 60#include "proto.h" 61#include "subr.h" 62#include "synch.h" 63 64struct hio { 65 uint64_t hio_seq; 66 int hio_error; 67 struct nv *hio_nv; 68 void *hio_data; 69 uint8_t hio_cmd; 70 uint64_t hio_offset; 71 uint64_t hio_length; 72 TAILQ_ENTRY(hio) hio_next; 73}; 74 75/* 76 * Free list holds unused structures. When free list is empty, we have to wait 77 * until some in-progress requests are freed. 78 */ 79static TAILQ_HEAD(, hio) hio_free_list; 80static pthread_mutex_t hio_free_list_lock; 81static pthread_cond_t hio_free_list_cond; 82/* 83 * Disk thread (the one that do I/O requests) takes requests from this list. 84 */ 85static TAILQ_HEAD(, hio) hio_disk_list; 86static pthread_mutex_t hio_disk_list_lock; 87static pthread_cond_t hio_disk_list_cond; 88/* 89 * There is one recv list for every component, although local components don't 90 * use recv lists as local requests are done synchronously. 91 */ 92static TAILQ_HEAD(, hio) hio_send_list; 93static pthread_mutex_t hio_send_list_lock; 94static pthread_cond_t hio_send_list_cond; 95 96/* 97 * Maximum number of outstanding I/O requests. 98 */ 99#define HAST_HIO_MAX 256 100 101static void *recv_thread(void *arg); 102static void *disk_thread(void *arg); 103static void *send_thread(void *arg); 104 105static void 106init_environment(void) 107{ 108 struct hio *hio; 109 unsigned int ii; 110 111 /* 112 * Initialize lists, their locks and theirs condition variables. 113 */ 114 TAILQ_INIT(&hio_free_list); 115 mtx_init(&hio_free_list_lock); 116 cv_init(&hio_free_list_cond); 117 TAILQ_INIT(&hio_disk_list); 118 mtx_init(&hio_disk_list_lock); 119 cv_init(&hio_disk_list_cond); 120 TAILQ_INIT(&hio_send_list); 121 mtx_init(&hio_send_list_lock); 122 cv_init(&hio_send_list_cond); 123 124 /* 125 * Allocate requests pool and initialize requests. 126 */ 127 for (ii = 0; ii < HAST_HIO_MAX; ii++) { 128 hio = malloc(sizeof(*hio)); 129 if (hio == NULL) { 130 pjdlog_exitx(EX_TEMPFAIL, 131 "Unable to allocate memory (%zu bytes) for hio request.", 132 sizeof(*hio)); 133 } 134 hio->hio_error = 0; 135 hio->hio_data = malloc(MAXPHYS); 136 if (hio->hio_data == NULL) { 137 pjdlog_exitx(EX_TEMPFAIL, 138 "Unable to allocate memory (%zu bytes) for gctl_data.", 139 (size_t)MAXPHYS); 140 } 141 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 142 } 143} 144 145static void 146init_local(struct hast_resource *res) 147{ 148 149 if (metadata_read(res, true) < 0) 150 exit(EX_NOINPUT); 151} 152 153static void 154init_remote(struct hast_resource *res, struct nv *nvin) 155{ 156 uint64_t resuid; 157 struct nv *nvout; 158 unsigned char *map; 159 size_t mapsize; 160 161 map = NULL; 162 mapsize = 0; 163 nvout = nv_alloc(); 164 nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 165 nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 166 resuid = nv_get_uint64(nvin, "resuid"); 167 res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 168 res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 169 nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 170 nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 171 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 172 METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 173 map = malloc(mapsize); 174 if (map == NULL) { 175 pjdlog_exitx(EX_TEMPFAIL, 176 "Unable to allocate memory (%zu bytes) for activemap.", 177 mapsize); 178 } 179 nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 180 /* 181 * When we work as primary and secondary is missing we will increase 182 * localcnt in our metadata. When secondary is connected and synced 183 * we make localcnt be equal to remotecnt, which means nodes are more 184 * or less in sync. 185 * Split-brain condition is when both nodes are not able to communicate 186 * and are both configured as primary nodes. In turn, they can both 187 * make incompatible changes to the data and we have to detect that. 188 * Under split-brain condition we will increase our localcnt on first 189 * write and remote node will increase its localcnt on first write. 190 * When we connect we can see that primary's localcnt is greater than 191 * our remotecnt (primary was modified while we weren't watching) and 192 * our localcnt is greater than primary's remotecnt (we were modified 193 * while primary wasn't watching). 194 * There are many possible combinations which are all gathered below. 195 * Don't pay too much attention to exact numbers, the more important 196 * is to compare them. We compare secondary's local with primary's 197 * remote and secondary's remote with primary's local. 198 * Note that every case where primary's localcnt is smaller than 199 * secondary's remotecnt and where secondary's localcnt is smaller than 200 * primary's remotecnt should be impossible in practise. We will perform 201 * full synchronization then. Those cases are marked with an asterisk. 202 * Regular synchronization means that only extents marked as dirty are 203 * synchronized (regular synchronization). 204 * 205 * SECONDARY METADATA PRIMARY METADATA 206 * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 207 * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 208 * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 209 * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 210 * regular sync from secondary. 211 * local=3 remote=3 local=3 remote=3 Regular sync just in case. 212 * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 213 * local=3 remote=3 local=4 remote=2 Split-brain condition. 214 * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 215 * regular sync from primary. 216 * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 217 */ 218 if (res->hr_resuid == 0) { 219 /* 220 * Provider is used for the first time. Initialize everything. 221 */ 222 assert(res->hr_secondary_localcnt == 0); 223 res->hr_resuid = resuid; 224 if (metadata_write(res) < 0) 225 exit(EX_NOINPUT); 226 memset(map, 0xff, mapsize); 227 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 228 } else if ( 229 /* Is primary is out-of-date? */ 230 (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 231 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 232 /* Node are more or less in sync? */ 233 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 234 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 235 /* Is secondary is out-of-date? */ 236 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 237 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 238 /* 239 * Nodes are more or less in sync or one of the nodes is 240 * out-of-date. 241 * It doesn't matter at this point which one, we just have to 242 * send out local bitmap to the remote node. 243 */ 244 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 245 (ssize_t)mapsize) { 246 pjdlog_exit(LOG_ERR, "Unable to read activemap"); 247 } 248 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 249 res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 250 /* Primary is out-of-date, sync from secondary. */ 251 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 252 } else { 253 /* 254 * Secondary is out-of-date or counts match. 255 * Sync from primary. 256 */ 257 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 258 } 259 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 260 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 261 /* 262 * Not good, we have split-brain condition. 263 */ 264 pjdlog_error("Split-brain detected, exiting."); 265 nv_add_string(nvout, "Split-brain condition!", "errmsg"); 266 free(map); 267 map = NULL; 268 mapsize = 0; 269 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 270 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 271 /* 272 * This should never happen in practise, but we will perform 273 * full synchronization. 274 */ 275 assert(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 276 res->hr_primary_localcnt < res->hr_secondary_remotecnt); 277 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 278 METADATA_SIZE, res->hr_extentsize, 279 res->hr_local_sectorsize); 280 memset(map, 0xff, mapsize); 281 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 282 /* In this one of five cases sync from secondary. */ 283 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 284 } else { 285 /* For the rest four cases sync from primary. */ 286 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 287 } 288 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 289 (uintmax_t)res->hr_primary_localcnt, 290 (uintmax_t)res->hr_primary_remotecnt, 291 (uintmax_t)res->hr_secondary_localcnt, 292 (uintmax_t)res->hr_secondary_remotecnt); 293 } 294 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 295 pjdlog_errno(LOG_WARNING, "Unable to send activemap to %s", 296 res->hr_remoteaddr); 297 nv_free(nvout); 298 exit(EX_TEMPFAIL); 299 } 300 nv_free(nvout); 301 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 302 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 303 /* Exit on split-brain. */ 304 exit(EX_CONFIG); 305 } 306} 307 308void 309hastd_secondary(struct hast_resource *res, struct nv *nvin) 310{ 311 pthread_t td; 312 pid_t pid; 313 int error; 314 315 /* 316 * Create communication channel between parent and child. 317 */ 318 if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 319 KEEP_ERRNO((void)pidfile_remove(pfh)); 320 pjdlog_exit(EX_OSERR, 321 "Unable to create control sockets between parent and child"); 322 } 323 324 pid = fork(); 325 if (pid < 0) { 326 KEEP_ERRNO((void)pidfile_remove(pfh)); 327 pjdlog_exit(EX_OSERR, "Unable to fork"); 328 } 329 330 if (pid > 0) { 331 /* This is parent. */ 332 proto_close(res->hr_remotein); 333 res->hr_remotein = NULL; 334 proto_close(res->hr_remoteout); 335 res->hr_remoteout = NULL; 336 res->hr_workerpid = pid; 337 return; 338 } 339 (void)pidfile_close(pfh); 340 341 setproctitle("%s (secondary)", res->hr_name); 342 343 /* Error in setting timeout is not critical, but why should it fail? */ 344 if (proto_timeout(res->hr_remotein, 0) < 0) 345 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 346 if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 347 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 348 349 init_local(res); 350 init_remote(res, nvin); 351 init_environment(); 352 353 error = pthread_create(&td, NULL, recv_thread, res); 354 assert(error == 0); 355 error = pthread_create(&td, NULL, disk_thread, res); 356 assert(error == 0); 357 error = pthread_create(&td, NULL, send_thread, res); 358 assert(error == 0); 359 (void)ctrl_thread(res); 360} 361 362static void 363reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 364{ 365 char msg[1024]; 366 va_list ap; 367 int len; 368 369 va_start(ap, fmt); 370 len = vsnprintf(msg, sizeof(msg), fmt, ap); 371 va_end(ap); 372 if ((size_t)len < sizeof(msg)) { 373 switch (hio->hio_cmd) { 374 case HIO_READ: 375 (void)snprintf(msg + len, sizeof(msg) - len, 376 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 377 (uintmax_t)hio->hio_length); 378 break; 379 case HIO_DELETE: 380 (void)snprintf(msg + len, sizeof(msg) - len, 381 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 382 (uintmax_t)hio->hio_length); 383 break; 384 case HIO_FLUSH: 385 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 386 break; 387 case HIO_WRITE: 388 (void)snprintf(msg + len, sizeof(msg) - len, 389 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 390 (uintmax_t)hio->hio_length); 391 break; 392 default: 393 (void)snprintf(msg + len, sizeof(msg) - len, 394 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 395 break; 396 } 397 } 398 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 399} 400 401static int 402requnpack(struct hast_resource *res, struct hio *hio) 403{ 404 405 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 406 if (hio->hio_cmd == 0) { 407 pjdlog_error("Header contains no 'cmd' field."); 408 hio->hio_error = EINVAL; 409 goto end; 410 } 411 switch (hio->hio_cmd) { 412 case HIO_READ: 413 case HIO_WRITE: 414 case HIO_DELETE: 415 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 416 if (nv_error(hio->hio_nv) != 0) { 417 pjdlog_error("Header is missing 'offset' field."); 418 hio->hio_error = EINVAL; 419 goto end; 420 } 421 hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 422 if (nv_error(hio->hio_nv) != 0) { 423 pjdlog_error("Header is missing 'length' field."); 424 hio->hio_error = EINVAL; 425 goto end; 426 } 427 if (hio->hio_length == 0) { 428 pjdlog_error("Data length is zero."); 429 hio->hio_error = EINVAL; 430 goto end; 431 } 432 if (hio->hio_length > MAXPHYS) { 433 pjdlog_error("Data length is too large (%ju > %ju).", 434 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 435 hio->hio_error = EINVAL; 436 goto end; 437 } 438 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 439 pjdlog_error("Offset %ju is not multiple of sector size.", 440 (uintmax_t)hio->hio_offset); 441 hio->hio_error = EINVAL; 442 goto end; 443 } 444 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 445 pjdlog_error("Length %ju is not multiple of sector size.", 446 (uintmax_t)hio->hio_length); 447 hio->hio_error = EINVAL; 448 goto end; 449 } 450 if (hio->hio_offset + hio->hio_length > 451 (uint64_t)res->hr_datasize) { 452 pjdlog_error("Data offset is too large (%ju > %ju).", 453 (uintmax_t)(hio->hio_offset + hio->hio_length), 454 (uintmax_t)res->hr_datasize); 455 hio->hio_error = EINVAL; 456 goto end; 457 } 458 break; 459 default: 460 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 461 hio->hio_cmd); 462 hio->hio_error = EINVAL; 463 goto end; 464 } 465 hio->hio_error = 0; 466end: 467 return (hio->hio_error); 468} 469 470/* 471 * Thread receives requests from the primary node. 472 */ 473static void * 474recv_thread(void *arg) 475{ 476 struct hast_resource *res = arg; 477 struct hio *hio; 478 bool wakeup; 479 480 for (;;) { 481 pjdlog_debug(2, "recv: Taking free request."); 482 mtx_lock(&hio_free_list_lock); 483 while ((hio = TAILQ_FIRST(&hio_free_list)) == NULL) { 484 pjdlog_debug(2, "recv: No free requests, waiting."); 485 cv_wait(&hio_free_list_cond, &hio_free_list_lock); 486 } 487 TAILQ_REMOVE(&hio_free_list, hio, hio_next); 488 mtx_unlock(&hio_free_list_lock); 489 pjdlog_debug(2, "recv: (%p) Got request.", hio); 490 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 491 pjdlog_exit(EX_TEMPFAIL, 492 "Unable to receive request header"); 493 } 494 if (requnpack(res, hio) != 0) 495 goto send_queue; 496 reqlog(LOG_DEBUG, 2, -1, hio, 497 "recv: (%p) Got request header: ", hio); 498 if (hio->hio_cmd == HIO_WRITE) { 499 if (hast_proto_recv_data(res, res->hr_remotein, 500 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 501 pjdlog_exit(EX_TEMPFAIL, 502 "Unable to receive reply data"); 503 } 504 } 505 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 506 hio); 507 mtx_lock(&hio_disk_list_lock); 508 wakeup = TAILQ_EMPTY(&hio_disk_list); 509 TAILQ_INSERT_TAIL(&hio_disk_list, hio, hio_next); 510 mtx_unlock(&hio_disk_list_lock); 511 if (wakeup) 512 cv_signal(&hio_disk_list_cond); 513 continue; 514send_queue: 515 pjdlog_debug(2, "recv: (%p) Moving request to the send queue.", 516 hio); 517 mtx_lock(&hio_send_list_lock); 518 wakeup = TAILQ_EMPTY(&hio_send_list); 519 TAILQ_INSERT_TAIL(&hio_send_list, hio, hio_next); 520 mtx_unlock(&hio_send_list_lock); 521 if (wakeup) 522 cv_signal(&hio_send_list_cond); 523 } 524 /* NOTREACHED */ 525 return (NULL); 526} 527 528/* 529 * Thread reads from or writes to local component and also handles DELETE and 530 * FLUSH requests. 531 */ 532static void * 533disk_thread(void *arg) 534{ 535 struct hast_resource *res = arg; 536 struct hio *hio; 537 ssize_t ret; 538 bool clear_activemap, wakeup; 539 540 clear_activemap = true; 541 542 for (;;) { 543 pjdlog_debug(2, "disk: Taking request."); 544 mtx_lock(&hio_disk_list_lock); 545 while ((hio = TAILQ_FIRST(&hio_disk_list)) == NULL) { 546 pjdlog_debug(2, "disk: No requests, waiting."); 547 cv_wait(&hio_disk_list_cond, &hio_disk_list_lock); 548 } 549 TAILQ_REMOVE(&hio_disk_list, hio, hio_next); 550 mtx_unlock(&hio_disk_list_lock); 551 while (clear_activemap) { 552 unsigned char *map; 553 size_t mapsize; 554 555 /* 556 * When first request is received, it means that primary 557 * already received our activemap, merged it and stored 558 * locally. We can now safely clear our activemap. 559 */ 560 mapsize = 561 activemap_calc_ondisk_size(res->hr_local_mediasize - 562 METADATA_SIZE, res->hr_extentsize, 563 res->hr_local_sectorsize); 564 map = calloc(1, mapsize); 565 if (map == NULL) { 566 pjdlog_warning("Unable to allocate memory to clear local activemap."); 567 break; 568 } 569 if (pwrite(res->hr_localfd, map, mapsize, 570 METADATA_SIZE) != (ssize_t)mapsize) { 571 pjdlog_errno(LOG_WARNING, 572 "Unable to store cleared activemap"); 573 free(map); 574 break; 575 } 576 free(map); 577 clear_activemap = false; 578 pjdlog_debug(1, "Local activemap cleared."); 579 } 580 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 581 /* Handle the actual request. */ 582 switch (hio->hio_cmd) { 583 case HIO_READ: 584 ret = pread(res->hr_localfd, hio->hio_data, 585 hio->hio_length, 586 hio->hio_offset + res->hr_localoff); 587 if (ret < 0) 588 hio->hio_error = errno; 589 else if (ret != (int64_t)hio->hio_length) 590 hio->hio_error = EIO; 591 else 592 hio->hio_error = 0; 593 break; 594 case HIO_WRITE: 595 ret = pwrite(res->hr_localfd, hio->hio_data, 596 hio->hio_length, 597 hio->hio_offset + res->hr_localoff); 598 if (ret < 0) 599 hio->hio_error = errno; 600 else if (ret != (int64_t)hio->hio_length) 601 hio->hio_error = EIO; 602 else 603 hio->hio_error = 0; 604 break; 605 case HIO_DELETE: 606 ret = g_delete(res->hr_localfd, 607 hio->hio_offset + res->hr_localoff, 608 hio->hio_length); 609 if (ret < 0) 610 hio->hio_error = errno; 611 else 612 hio->hio_error = 0; 613 break; 614 case HIO_FLUSH: 615 ret = g_flush(res->hr_localfd); 616 if (ret < 0) 617 hio->hio_error = errno; 618 else 619 hio->hio_error = 0; 620 break; 621 } 622 if (hio->hio_error != 0) { 623 reqlog(LOG_ERR, 0, hio->hio_error, hio, 624 "Request failed: "); 625 } 626 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 627 hio); 628 mtx_lock(&hio_send_list_lock); 629 wakeup = TAILQ_EMPTY(&hio_send_list); 630 TAILQ_INSERT_TAIL(&hio_send_list, hio, hio_next); 631 mtx_unlock(&hio_send_list_lock); 632 if (wakeup) 633 cv_signal(&hio_send_list_cond); 634 } 635 /* NOTREACHED */ 636 return (NULL); 637} 638 639/* 640 * Thread sends requests back to primary node. 641 */ 642static void * 643send_thread(void *arg) 644{ 645 struct hast_resource *res = arg; 646 struct nv *nvout; 647 struct hio *hio; 648 void *data; 649 size_t length; 650 bool wakeup; 651 652 for (;;) { 653 pjdlog_debug(2, "send: Taking request."); 654 mtx_lock(&hio_send_list_lock); 655 while ((hio = TAILQ_FIRST(&hio_send_list)) == NULL) { 656 pjdlog_debug(2, "send: No requests, waiting."); 657 cv_wait(&hio_send_list_cond, &hio_send_list_lock); 658 } 659 TAILQ_REMOVE(&hio_send_list, hio, hio_next); 660 mtx_unlock(&hio_send_list_lock); 661 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 662 nvout = nv_alloc(); 663 /* Copy sequence number. */ 664 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 665 switch (hio->hio_cmd) { 666 case HIO_READ: 667 if (hio->hio_error == 0) { 668 data = hio->hio_data; 669 length = hio->hio_length; 670 break; 671 } 672 /* 673 * We send no data in case of an error. 674 */ 675 /* FALLTHROUGH */ 676 case HIO_DELETE: 677 case HIO_FLUSH: 678 case HIO_WRITE: 679 data = NULL; 680 length = 0; 681 break; 682 default: 683 abort(); 684 break; 685 } 686 if (hio->hio_error != 0) 687 nv_add_int16(nvout, hio->hio_error, "error"); 688 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 689 length) < 0) { 690 pjdlog_exit(EX_TEMPFAIL, "Unable to send reply."); 691 } 692 nv_free(nvout); 693 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 694 hio); 695 nv_free(hio->hio_nv); 696 hio->hio_error = 0; 697 mtx_lock(&hio_free_list_lock); 698 wakeup = TAILQ_EMPTY(&hio_free_list); 699 TAILQ_INSERT_TAIL(&hio_free_list, hio, hio_next); 700 mtx_unlock(&hio_free_list_lock); 701 if (wakeup) 702 cv_signal(&hio_free_list_cond); 703 } 704 /* NOTREACHED */ 705 return (NULL); 706} 707