secondary.c revision 211984
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: head/sbin/hastd/secondary.c 211984 2010-08-30 00:31:30Z pjd $"); 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 <assert.h> 41#include <err.h> 42#include <errno.h> 43#include <fcntl.h> 44#include <libgeom.h> 45#include <pthread.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 "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 66struct hio { 67 uint64_t hio_seq; 68 int hio_error; 69 struct nv *hio_nv; 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 130init_environment(void) 131{ 132 struct hio *hio; 133 unsigned int ii; 134 135 /* 136 * Initialize lists, their locks and theirs condition variables. 137 */ 138 TAILQ_INIT(&hio_free_list); 139 mtx_init(&hio_free_list_lock); 140 cv_init(&hio_free_list_cond); 141 TAILQ_INIT(&hio_disk_list); 142 mtx_init(&hio_disk_list_lock); 143 cv_init(&hio_disk_list_cond); 144 TAILQ_INIT(&hio_send_list); 145 mtx_init(&hio_send_list_lock); 146 cv_init(&hio_send_list_cond); 147 148 /* 149 * Allocate requests pool and initialize requests. 150 */ 151 for (ii = 0; ii < HAST_HIO_MAX; ii++) { 152 hio = malloc(sizeof(*hio)); 153 if (hio == NULL) { 154 pjdlog_exitx(EX_TEMPFAIL, 155 "Unable to allocate memory (%zu bytes) for hio request.", 156 sizeof(*hio)); 157 } 158 hio->hio_error = 0; 159 hio->hio_data = malloc(MAXPHYS); 160 if (hio->hio_data == NULL) { 161 pjdlog_exitx(EX_TEMPFAIL, 162 "Unable to allocate memory (%zu bytes) for gctl_data.", 163 (size_t)MAXPHYS); 164 } 165 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 166 } 167} 168 169static void 170init_local(struct hast_resource *res) 171{ 172 173 if (metadata_read(res, true) < 0) 174 exit(EX_NOINPUT); 175} 176 177static void 178init_remote(struct hast_resource *res, struct nv *nvin) 179{ 180 uint64_t resuid; 181 struct nv *nvout; 182 unsigned char *map; 183 size_t mapsize; 184 185 map = NULL; 186 mapsize = 0; 187 nvout = nv_alloc(); 188 nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 189 nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 190 resuid = nv_get_uint64(nvin, "resuid"); 191 res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 192 res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 193 nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 194 nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 195 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 196 METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 197 map = malloc(mapsize); 198 if (map == NULL) { 199 pjdlog_exitx(EX_TEMPFAIL, 200 "Unable to allocate memory (%zu bytes) for activemap.", 201 mapsize); 202 } 203 nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 204 /* 205 * When we work as primary and secondary is missing we will increase 206 * localcnt in our metadata. When secondary is connected and synced 207 * we make localcnt be equal to remotecnt, which means nodes are more 208 * or less in sync. 209 * Split-brain condition is when both nodes are not able to communicate 210 * and are both configured as primary nodes. In turn, they can both 211 * make incompatible changes to the data and we have to detect that. 212 * Under split-brain condition we will increase our localcnt on first 213 * write and remote node will increase its localcnt on first write. 214 * When we connect we can see that primary's localcnt is greater than 215 * our remotecnt (primary was modified while we weren't watching) and 216 * our localcnt is greater than primary's remotecnt (we were modified 217 * while primary wasn't watching). 218 * There are many possible combinations which are all gathered below. 219 * Don't pay too much attention to exact numbers, the more important 220 * is to compare them. We compare secondary's local with primary's 221 * remote and secondary's remote with primary's local. 222 * Note that every case where primary's localcnt is smaller than 223 * secondary's remotecnt and where secondary's localcnt is smaller than 224 * primary's remotecnt should be impossible in practise. We will perform 225 * full synchronization then. Those cases are marked with an asterisk. 226 * Regular synchronization means that only extents marked as dirty are 227 * synchronized (regular synchronization). 228 * 229 * SECONDARY METADATA PRIMARY METADATA 230 * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 231 * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 232 * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 233 * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 234 * regular sync from secondary. 235 * local=3 remote=3 local=3 remote=3 Regular sync just in case. 236 * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 237 * local=3 remote=3 local=4 remote=2 Split-brain condition. 238 * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 239 * regular sync from primary. 240 * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 241 */ 242 if (res->hr_resuid == 0) { 243 /* 244 * Provider is used for the first time. Initialize everything. 245 */ 246 assert(res->hr_secondary_localcnt == 0); 247 res->hr_resuid = resuid; 248 if (metadata_write(res) < 0) 249 exit(EX_NOINPUT); 250 memset(map, 0xff, mapsize); 251 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 252 } else if ( 253 /* Is primary is out-of-date? */ 254 (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 255 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 256 /* Node are more or less in sync? */ 257 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 258 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 259 /* Is secondary is out-of-date? */ 260 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 261 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 262 /* 263 * Nodes are more or less in sync or one of the nodes is 264 * out-of-date. 265 * It doesn't matter at this point which one, we just have to 266 * send out local bitmap to the remote node. 267 */ 268 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 269 (ssize_t)mapsize) { 270 pjdlog_exit(LOG_ERR, "Unable to read activemap"); 271 } 272 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 273 res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 274 /* Primary is out-of-date, sync from secondary. */ 275 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 276 } else { 277 /* 278 * Secondary is out-of-date or counts match. 279 * Sync from primary. 280 */ 281 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 282 } 283 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 284 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 285 /* 286 * Not good, we have split-brain condition. 287 */ 288 pjdlog_error("Split-brain detected, exiting."); 289 nv_add_string(nvout, "Split-brain condition!", "errmsg"); 290 free(map); 291 map = NULL; 292 mapsize = 0; 293 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 294 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 295 /* 296 * This should never happen in practise, but we will perform 297 * full synchronization. 298 */ 299 assert(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 300 res->hr_primary_localcnt < res->hr_secondary_remotecnt); 301 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 302 METADATA_SIZE, res->hr_extentsize, 303 res->hr_local_sectorsize); 304 memset(map, 0xff, mapsize); 305 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 306 /* In this one of five cases sync from secondary. */ 307 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 308 } else { 309 /* For the rest four cases sync from primary. */ 310 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 311 } 312 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 313 (uintmax_t)res->hr_primary_localcnt, 314 (uintmax_t)res->hr_primary_remotecnt, 315 (uintmax_t)res->hr_secondary_localcnt, 316 (uintmax_t)res->hr_secondary_remotecnt); 317 } 318 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 319 pjdlog_errno(LOG_WARNING, "Unable to send activemap to %s", 320 res->hr_remoteaddr); 321 nv_free(nvout); 322 exit(EX_TEMPFAIL); 323 } 324 nv_free(nvout); 325 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 326 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 327 /* Exit on split-brain. */ 328 hook_exec(res->hr_exec, "split-brain", res->hr_name, NULL); 329 exit(EX_CONFIG); 330 } 331} 332 333void 334hastd_secondary(struct hast_resource *res, struct nv *nvin) 335{ 336 pthread_t td; 337 pid_t pid; 338 int error; 339 340 /* 341 * Create communication channel between parent and child. 342 */ 343 if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 344 KEEP_ERRNO((void)pidfile_remove(pfh)); 345 pjdlog_exit(EX_OSERR, 346 "Unable to create control sockets between parent and child"); 347 } 348 349 pid = fork(); 350 if (pid < 0) { 351 KEEP_ERRNO((void)pidfile_remove(pfh)); 352 pjdlog_exit(EX_OSERR, "Unable to fork"); 353 } 354 355 if (pid > 0) { 356 /* This is parent. */ 357 proto_close(res->hr_remotein); 358 res->hr_remotein = NULL; 359 proto_close(res->hr_remoteout); 360 res->hr_remoteout = NULL; 361 res->hr_workerpid = pid; 362 return; 363 } 364 365 gres = res; 366 367 (void)pidfile_close(pfh); 368 hook_fini(); 369 370 setproctitle("%s (secondary)", res->hr_name); 371 372 signal(SIGHUP, SIG_DFL); 373 signal(SIGCHLD, SIG_DFL); 374 375 /* Error in setting timeout is not critical, but why should it fail? */ 376 if (proto_timeout(res->hr_remotein, 0) < 0) 377 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 378 if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 379 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 380 381 hook_init(); 382 init_local(res); 383 init_remote(res, nvin); 384 init_environment(); 385 hook_exec(res->hr_exec, "connect", res->hr_name, NULL); 386 387 error = pthread_create(&td, NULL, recv_thread, res); 388 assert(error == 0); 389 error = pthread_create(&td, NULL, disk_thread, res); 390 assert(error == 0); 391 error = pthread_create(&td, NULL, send_thread, res); 392 assert(error == 0); 393 (void)ctrl_thread(res); 394} 395 396static void 397reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 398{ 399 char msg[1024]; 400 va_list ap; 401 int len; 402 403 va_start(ap, fmt); 404 len = vsnprintf(msg, sizeof(msg), fmt, ap); 405 va_end(ap); 406 if ((size_t)len < sizeof(msg)) { 407 switch (hio->hio_cmd) { 408 case HIO_READ: 409 (void)snprintf(msg + len, sizeof(msg) - len, 410 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 411 (uintmax_t)hio->hio_length); 412 break; 413 case HIO_DELETE: 414 (void)snprintf(msg + len, sizeof(msg) - len, 415 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 416 (uintmax_t)hio->hio_length); 417 break; 418 case HIO_FLUSH: 419 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 420 break; 421 case HIO_WRITE: 422 (void)snprintf(msg + len, sizeof(msg) - len, 423 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 424 (uintmax_t)hio->hio_length); 425 break; 426 case HIO_KEEPALIVE: 427 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 428 break; 429 default: 430 (void)snprintf(msg + len, sizeof(msg) - len, 431 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 432 break; 433 } 434 } 435 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 436} 437 438static int 439requnpack(struct hast_resource *res, struct hio *hio) 440{ 441 442 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 443 if (hio->hio_cmd == 0) { 444 pjdlog_error("Header contains no 'cmd' field."); 445 hio->hio_error = EINVAL; 446 goto end; 447 } 448 switch (hio->hio_cmd) { 449 case HIO_KEEPALIVE: 450 break; 451 case HIO_READ: 452 case HIO_WRITE: 453 case HIO_DELETE: 454 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 455 if (nv_error(hio->hio_nv) != 0) { 456 pjdlog_error("Header is missing 'offset' field."); 457 hio->hio_error = EINVAL; 458 goto end; 459 } 460 hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 461 if (nv_error(hio->hio_nv) != 0) { 462 pjdlog_error("Header is missing 'length' field."); 463 hio->hio_error = EINVAL; 464 goto end; 465 } 466 if (hio->hio_length == 0) { 467 pjdlog_error("Data length is zero."); 468 hio->hio_error = EINVAL; 469 goto end; 470 } 471 if (hio->hio_length > MAXPHYS) { 472 pjdlog_error("Data length is too large (%ju > %ju).", 473 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 474 hio->hio_error = EINVAL; 475 goto end; 476 } 477 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 478 pjdlog_error("Offset %ju is not multiple of sector size.", 479 (uintmax_t)hio->hio_offset); 480 hio->hio_error = EINVAL; 481 goto end; 482 } 483 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 484 pjdlog_error("Length %ju is not multiple of sector size.", 485 (uintmax_t)hio->hio_length); 486 hio->hio_error = EINVAL; 487 goto end; 488 } 489 if (hio->hio_offset + hio->hio_length > 490 (uint64_t)res->hr_datasize) { 491 pjdlog_error("Data offset is too large (%ju > %ju).", 492 (uintmax_t)(hio->hio_offset + hio->hio_length), 493 (uintmax_t)res->hr_datasize); 494 hio->hio_error = EINVAL; 495 goto end; 496 } 497 break; 498 default: 499 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 500 hio->hio_cmd); 501 hio->hio_error = EINVAL; 502 goto end; 503 } 504 hio->hio_error = 0; 505end: 506 return (hio->hio_error); 507} 508 509static void 510secondary_exit(int exitcode, const char *fmt, ...) 511{ 512 va_list ap; 513 514 assert(exitcode != EX_OK); 515 va_start(ap, fmt); 516 pjdlogv_errno(LOG_ERR, fmt, ap); 517 va_end(ap); 518 hook_exec(gres->hr_exec, "disconnect", gres->hr_name, NULL); 519 exit(exitcode); 520} 521 522/* 523 * Thread receives requests from the primary node. 524 */ 525static void * 526recv_thread(void *arg) 527{ 528 struct hast_resource *res = arg; 529 struct hio *hio; 530 531 for (;;) { 532 pjdlog_debug(2, "recv: Taking free request."); 533 QUEUE_TAKE(free, hio); 534 pjdlog_debug(2, "recv: (%p) Got request.", hio); 535 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 536 secondary_exit(EX_TEMPFAIL, 537 "Unable to receive request header"); 538 } 539 if (requnpack(res, hio) != 0) { 540 pjdlog_debug(2, 541 "recv: (%p) Moving request to the send queue.", 542 hio); 543 QUEUE_INSERT(send, hio); 544 continue; 545 } 546 reqlog(LOG_DEBUG, 2, -1, hio, 547 "recv: (%p) Got request header: ", hio); 548 if (hio->hio_cmd == HIO_KEEPALIVE) { 549 pjdlog_debug(2, 550 "recv: (%p) Moving request to the free queue.", 551 hio); 552 nv_free(hio->hio_nv); 553 QUEUE_INSERT(free, hio); 554 continue; 555 } else if (hio->hio_cmd == HIO_WRITE) { 556 if (hast_proto_recv_data(res, res->hr_remotein, 557 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 558 secondary_exit(EX_TEMPFAIL, 559 "Unable to receive reply data"); 560 } 561 } 562 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 563 hio); 564 QUEUE_INSERT(disk, hio); 565 } 566 /* NOTREACHED */ 567 return (NULL); 568} 569 570/* 571 * Thread reads from or writes to local component and also handles DELETE and 572 * FLUSH requests. 573 */ 574static void * 575disk_thread(void *arg) 576{ 577 struct hast_resource *res = arg; 578 struct hio *hio; 579 ssize_t ret; 580 bool clear_activemap; 581 582 clear_activemap = true; 583 584 for (;;) { 585 pjdlog_debug(2, "disk: Taking request."); 586 QUEUE_TAKE(disk, hio); 587 while (clear_activemap) { 588 unsigned char *map; 589 size_t mapsize; 590 591 /* 592 * When first request is received, it means that primary 593 * already received our activemap, merged it and stored 594 * locally. We can now safely clear our activemap. 595 */ 596 mapsize = 597 activemap_calc_ondisk_size(res->hr_local_mediasize - 598 METADATA_SIZE, res->hr_extentsize, 599 res->hr_local_sectorsize); 600 map = calloc(1, mapsize); 601 if (map == NULL) { 602 pjdlog_warning("Unable to allocate memory to clear local activemap."); 603 break; 604 } 605 if (pwrite(res->hr_localfd, map, mapsize, 606 METADATA_SIZE) != (ssize_t)mapsize) { 607 pjdlog_errno(LOG_WARNING, 608 "Unable to store cleared activemap"); 609 free(map); 610 break; 611 } 612 free(map); 613 clear_activemap = false; 614 pjdlog_debug(1, "Local activemap cleared."); 615 } 616 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 617 /* Handle the actual request. */ 618 switch (hio->hio_cmd) { 619 case HIO_READ: 620 ret = pread(res->hr_localfd, hio->hio_data, 621 hio->hio_length, 622 hio->hio_offset + res->hr_localoff); 623 if (ret < 0) 624 hio->hio_error = errno; 625 else if (ret != (int64_t)hio->hio_length) 626 hio->hio_error = EIO; 627 else 628 hio->hio_error = 0; 629 break; 630 case HIO_WRITE: 631 ret = pwrite(res->hr_localfd, hio->hio_data, 632 hio->hio_length, 633 hio->hio_offset + res->hr_localoff); 634 if (ret < 0) 635 hio->hio_error = errno; 636 else if (ret != (int64_t)hio->hio_length) 637 hio->hio_error = EIO; 638 else 639 hio->hio_error = 0; 640 break; 641 case HIO_DELETE: 642 ret = g_delete(res->hr_localfd, 643 hio->hio_offset + res->hr_localoff, 644 hio->hio_length); 645 if (ret < 0) 646 hio->hio_error = errno; 647 else 648 hio->hio_error = 0; 649 break; 650 case HIO_FLUSH: 651 ret = g_flush(res->hr_localfd); 652 if (ret < 0) 653 hio->hio_error = errno; 654 else 655 hio->hio_error = 0; 656 break; 657 } 658 if (hio->hio_error != 0) { 659 reqlog(LOG_ERR, 0, hio->hio_error, hio, 660 "Request failed: "); 661 } 662 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 663 hio); 664 QUEUE_INSERT(send, hio); 665 } 666 /* NOTREACHED */ 667 return (NULL); 668} 669 670/* 671 * Thread sends requests back to primary node. 672 */ 673static void * 674send_thread(void *arg) 675{ 676 struct hast_resource *res = arg; 677 struct nv *nvout; 678 struct hio *hio; 679 void *data; 680 size_t length; 681 682 for (;;) { 683 pjdlog_debug(2, "send: Taking request."); 684 QUEUE_TAKE(send, hio); 685 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 686 nvout = nv_alloc(); 687 /* Copy sequence number. */ 688 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 689 switch (hio->hio_cmd) { 690 case HIO_READ: 691 if (hio->hio_error == 0) { 692 data = hio->hio_data; 693 length = hio->hio_length; 694 break; 695 } 696 /* 697 * We send no data in case of an error. 698 */ 699 /* FALLTHROUGH */ 700 case HIO_DELETE: 701 case HIO_FLUSH: 702 case HIO_WRITE: 703 data = NULL; 704 length = 0; 705 break; 706 default: 707 abort(); 708 break; 709 } 710 if (hio->hio_error != 0) 711 nv_add_int16(nvout, hio->hio_error, "error"); 712 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 713 length) < 0) { 714 secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 715 } 716 nv_free(nvout); 717 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 718 hio); 719 nv_free(hio->hio_nv); 720 hio->hio_error = 0; 721 QUEUE_INSERT(free, hio); 722 } 723 /* NOTREACHED */ 724 return (NULL); 725} 726