secondary.c revision 212038
1155324Simp/*- 2155324Simp * Copyright (c) 2009-2010 The FreeBSD Foundation 3155324Simp * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 4155324Simp * All rights reserved. 5155324Simp * 6155324Simp * 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 212038 2010-08-30 23:26:10Z 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 "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 struct nv *hio_nv; 71 void *hio_data; 72 uint8_t hio_cmd; 73 uint64_t hio_offset; 74 uint64_t hio_length; 75 TAILQ_ENTRY(hio) hio_next; 76}; 77 78static struct hast_resource *gres; 79 80/* 81 * Free list holds unused structures. When free list is empty, we have to wait 82 * until some in-progress requests are freed. 83 */ 84static TAILQ_HEAD(, hio) hio_free_list; 85static pthread_mutex_t hio_free_list_lock; 86static pthread_cond_t hio_free_list_cond; 87/* 88 * Disk thread (the one that do I/O requests) takes requests from this list. 89 */ 90static TAILQ_HEAD(, hio) hio_disk_list; 91static pthread_mutex_t hio_disk_list_lock; 92static pthread_cond_t hio_disk_list_cond; 93/* 94 * There is one recv list for every component, although local components don't 95 * use recv lists as local requests are done synchronously. 96 */ 97static TAILQ_HEAD(, hio) hio_send_list; 98static pthread_mutex_t hio_send_list_lock; 99static pthread_cond_t hio_send_list_cond; 100 101/* 102 * Maximum number of outstanding I/O requests. 103 */ 104#define HAST_HIO_MAX 256 105 106static void *recv_thread(void *arg); 107static void *disk_thread(void *arg); 108static void *send_thread(void *arg); 109 110#define QUEUE_INSERT(name, hio) do { \ 111 bool _wakeup; \ 112 \ 113 mtx_lock(&hio_##name##_list_lock); \ 114 _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 115 TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \ 116 mtx_unlock(&hio_##name##_list_lock); \ 117 if (_wakeup) \ 118 cv_signal(&hio_##name##_list_cond); \ 119} while (0) 120#define QUEUE_TAKE(name, hio) do { \ 121 mtx_lock(&hio_##name##_list_lock); \ 122 while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 123 cv_wait(&hio_##name##_list_cond, \ 124 &hio_##name##_list_lock); \ 125 } \ 126 TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \ 127 mtx_unlock(&hio_##name##_list_lock); \ 128} while (0) 129 130static void 131init_environment(void) 132{ 133 struct hio *hio; 134 unsigned int ii; 135 136 /* 137 * Initialize lists, their locks and theirs condition variables. 138 */ 139 TAILQ_INIT(&hio_free_list); 140 mtx_init(&hio_free_list_lock); 141 cv_init(&hio_free_list_cond); 142 TAILQ_INIT(&hio_disk_list); 143 mtx_init(&hio_disk_list_lock); 144 cv_init(&hio_disk_list_cond); 145 TAILQ_INIT(&hio_send_list); 146 mtx_init(&hio_send_list_lock); 147 cv_init(&hio_send_list_cond); 148 149 /* 150 * Allocate requests pool and initialize requests. 151 */ 152 for (ii = 0; ii < HAST_HIO_MAX; ii++) { 153 hio = malloc(sizeof(*hio)); 154 if (hio == NULL) { 155 pjdlog_exitx(EX_TEMPFAIL, 156 "Unable to allocate memory (%zu bytes) for hio request.", 157 sizeof(*hio)); 158 } 159 hio->hio_error = 0; 160 hio->hio_data = malloc(MAXPHYS); 161 if (hio->hio_data == NULL) { 162 pjdlog_exitx(EX_TEMPFAIL, 163 "Unable to allocate memory (%zu bytes) for gctl_data.", 164 (size_t)MAXPHYS); 165 } 166 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 167 } 168} 169 170static void 171init_local(struct hast_resource *res) 172{ 173 174 if (metadata_read(res, true) < 0) 175 exit(EX_NOINPUT); 176} 177 178static void 179init_remote(struct hast_resource *res, struct nv *nvin) 180{ 181 uint64_t resuid; 182 struct nv *nvout; 183 unsigned char *map; 184 size_t mapsize; 185 186 map = NULL; 187 mapsize = 0; 188 nvout = nv_alloc(); 189 nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 190 nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 191 resuid = nv_get_uint64(nvin, "resuid"); 192 res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 193 res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 194 nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 195 nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 196 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 197 METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 198 map = malloc(mapsize); 199 if (map == NULL) { 200 pjdlog_exitx(EX_TEMPFAIL, 201 "Unable to allocate memory (%zu bytes) for activemap.", 202 mapsize); 203 } 204 nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 205 /* 206 * When we work as primary and secondary is missing we will increase 207 * localcnt in our metadata. When secondary is connected and synced 208 * we make localcnt be equal to remotecnt, which means nodes are more 209 * or less in sync. 210 * Split-brain condition is when both nodes are not able to communicate 211 * and are both configured as primary nodes. In turn, they can both 212 * make incompatible changes to the data and we have to detect that. 213 * Under split-brain condition we will increase our localcnt on first 214 * write and remote node will increase its localcnt on first write. 215 * When we connect we can see that primary's localcnt is greater than 216 * our remotecnt (primary was modified while we weren't watching) and 217 * our localcnt is greater than primary's remotecnt (we were modified 218 * while primary wasn't watching). 219 * There are many possible combinations which are all gathered below. 220 * Don't pay too much attention to exact numbers, the more important 221 * is to compare them. We compare secondary's local with primary's 222 * remote and secondary's remote with primary's local. 223 * Note that every case where primary's localcnt is smaller than 224 * secondary's remotecnt and where secondary's localcnt is smaller than 225 * primary's remotecnt should be impossible in practise. We will perform 226 * full synchronization then. Those cases are marked with an asterisk. 227 * Regular synchronization means that only extents marked as dirty are 228 * synchronized (regular synchronization). 229 * 230 * SECONDARY METADATA PRIMARY METADATA 231 * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 232 * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 233 * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 234 * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 235 * regular sync from secondary. 236 * local=3 remote=3 local=3 remote=3 Regular sync just in case. 237 * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 238 * local=3 remote=3 local=4 remote=2 Split-brain condition. 239 * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 240 * regular sync from primary. 241 * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 242 */ 243 if (res->hr_resuid == 0) { 244 /* 245 * Provider is used for the first time. Initialize everything. 246 */ 247 assert(res->hr_secondary_localcnt == 0); 248 res->hr_resuid = resuid; 249 if (metadata_write(res) < 0) 250 exit(EX_NOINPUT); 251 memset(map, 0xff, mapsize); 252 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 253 } else if ( 254 /* Is primary is out-of-date? */ 255 (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 256 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 257 /* Node are more or less in sync? */ 258 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 259 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 260 /* Is secondary is out-of-date? */ 261 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 262 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 263 /* 264 * Nodes are more or less in sync or one of the nodes is 265 * out-of-date. 266 * It doesn't matter at this point which one, we just have to 267 * send out local bitmap to the remote node. 268 */ 269 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 270 (ssize_t)mapsize) { 271 pjdlog_exit(LOG_ERR, "Unable to read activemap"); 272 } 273 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 274 res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 275 /* Primary is out-of-date, sync from secondary. */ 276 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 277 } else { 278 /* 279 * Secondary is out-of-date or counts match. 280 * Sync from primary. 281 */ 282 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 283 } 284 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 285 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 286 /* 287 * Not good, we have split-brain condition. 288 */ 289 pjdlog_error("Split-brain detected, exiting."); 290 nv_add_string(nvout, "Split-brain condition!", "errmsg"); 291 free(map); 292 map = NULL; 293 mapsize = 0; 294 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 295 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 296 /* 297 * This should never happen in practise, but we will perform 298 * full synchronization. 299 */ 300 assert(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 301 res->hr_primary_localcnt < res->hr_secondary_remotecnt); 302 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 303 METADATA_SIZE, res->hr_extentsize, 304 res->hr_local_sectorsize); 305 memset(map, 0xff, mapsize); 306 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 307 /* In this one of five cases sync from secondary. */ 308 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 309 } else { 310 /* For the rest four cases sync from primary. */ 311 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 312 } 313 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 314 (uintmax_t)res->hr_primary_localcnt, 315 (uintmax_t)res->hr_primary_remotecnt, 316 (uintmax_t)res->hr_secondary_localcnt, 317 (uintmax_t)res->hr_secondary_remotecnt); 318 } 319 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 320 pjdlog_errno(LOG_WARNING, "Unable to send activemap to %s", 321 res->hr_remoteaddr); 322 nv_free(nvout); 323 exit(EX_TEMPFAIL); 324 } 325 nv_free(nvout); 326 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 327 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 328 /* Exit on split-brain. */ 329 event_send(res, EVENT_SPLITBRAIN); 330 exit(EX_CONFIG); 331 } 332} 333 334void 335hastd_secondary(struct hast_resource *res, struct nv *nvin) 336{ 337 pthread_t td; 338 pid_t pid; 339 int error; 340 341 /* 342 * Create communication channel between parent and child. 343 */ 344 if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 345 KEEP_ERRNO((void)pidfile_remove(pfh)); 346 pjdlog_exit(EX_OSERR, 347 "Unable to create control sockets between parent and child"); 348 } 349 /* 350 * Create communication channel between child and parent. 351 */ 352 if (proto_client("socketpair://", &res->hr_event) < 0) { 353 KEEP_ERRNO((void)pidfile_remove(pfh)); 354 pjdlog_exit(EX_OSERR, 355 "Unable to create event sockets between child and parent"); 356 } 357 358 pid = fork(); 359 if (pid < 0) { 360 KEEP_ERRNO((void)pidfile_remove(pfh)); 361 pjdlog_exit(EX_OSERR, "Unable to fork"); 362 } 363 364 if (pid > 0) { 365 /* This is parent. */ 366 proto_close(res->hr_remotein); 367 res->hr_remotein = NULL; 368 proto_close(res->hr_remoteout); 369 res->hr_remoteout = NULL; 370 /* Declare that we are receiver. */ 371 proto_recv(res->hr_event, NULL, 0); 372 res->hr_workerpid = pid; 373 return; 374 } 375 376 gres = res; 377 378 (void)pidfile_close(pfh); 379 hook_fini(); 380 381 setproctitle("%s (secondary)", res->hr_name); 382 383 signal(SIGHUP, SIG_DFL); 384 signal(SIGCHLD, SIG_DFL); 385 386 /* Declare that we are sender. */ 387 proto_send(res->hr_event, NULL, 0); 388 389 /* Error in setting timeout is not critical, but why should it fail? */ 390 if (proto_timeout(res->hr_remotein, 0) < 0) 391 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 392 if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 393 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 394 395 init_local(res); 396 init_remote(res, nvin); 397 init_environment(); 398 event_send(res, EVENT_CONNECT); 399 400 error = pthread_create(&td, NULL, recv_thread, res); 401 assert(error == 0); 402 error = pthread_create(&td, NULL, disk_thread, res); 403 assert(error == 0); 404 error = pthread_create(&td, NULL, send_thread, res); 405 assert(error == 0); 406 (void)ctrl_thread(res); 407} 408 409static void 410reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 411{ 412 char msg[1024]; 413 va_list ap; 414 int len; 415 416 va_start(ap, fmt); 417 len = vsnprintf(msg, sizeof(msg), fmt, ap); 418 va_end(ap); 419 if ((size_t)len < sizeof(msg)) { 420 switch (hio->hio_cmd) { 421 case HIO_READ: 422 (void)snprintf(msg + len, sizeof(msg) - len, 423 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 424 (uintmax_t)hio->hio_length); 425 break; 426 case HIO_DELETE: 427 (void)snprintf(msg + len, sizeof(msg) - len, 428 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 429 (uintmax_t)hio->hio_length); 430 break; 431 case HIO_FLUSH: 432 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 433 break; 434 case HIO_WRITE: 435 (void)snprintf(msg + len, sizeof(msg) - len, 436 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 437 (uintmax_t)hio->hio_length); 438 break; 439 case HIO_KEEPALIVE: 440 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 441 break; 442 default: 443 (void)snprintf(msg + len, sizeof(msg) - len, 444 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 445 break; 446 } 447 } 448 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 449} 450 451static int 452requnpack(struct hast_resource *res, struct hio *hio) 453{ 454 455 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 456 if (hio->hio_cmd == 0) { 457 pjdlog_error("Header contains no 'cmd' field."); 458 hio->hio_error = EINVAL; 459 goto end; 460 } 461 switch (hio->hio_cmd) { 462 case HIO_KEEPALIVE: 463 break; 464 case HIO_READ: 465 case HIO_WRITE: 466 case HIO_DELETE: 467 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 468 if (nv_error(hio->hio_nv) != 0) { 469 pjdlog_error("Header is missing 'offset' field."); 470 hio->hio_error = EINVAL; 471 goto end; 472 } 473 hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 474 if (nv_error(hio->hio_nv) != 0) { 475 pjdlog_error("Header is missing 'length' field."); 476 hio->hio_error = EINVAL; 477 goto end; 478 } 479 if (hio->hio_length == 0) { 480 pjdlog_error("Data length is zero."); 481 hio->hio_error = EINVAL; 482 goto end; 483 } 484 if (hio->hio_length > MAXPHYS) { 485 pjdlog_error("Data length is too large (%ju > %ju).", 486 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 487 hio->hio_error = EINVAL; 488 goto end; 489 } 490 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 491 pjdlog_error("Offset %ju is not multiple of sector size.", 492 (uintmax_t)hio->hio_offset); 493 hio->hio_error = EINVAL; 494 goto end; 495 } 496 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 497 pjdlog_error("Length %ju is not multiple of sector size.", 498 (uintmax_t)hio->hio_length); 499 hio->hio_error = EINVAL; 500 goto end; 501 } 502 if (hio->hio_offset + hio->hio_length > 503 (uint64_t)res->hr_datasize) { 504 pjdlog_error("Data offset is too large (%ju > %ju).", 505 (uintmax_t)(hio->hio_offset + hio->hio_length), 506 (uintmax_t)res->hr_datasize); 507 hio->hio_error = EINVAL; 508 goto end; 509 } 510 break; 511 default: 512 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 513 hio->hio_cmd); 514 hio->hio_error = EINVAL; 515 goto end; 516 } 517 hio->hio_error = 0; 518end: 519 return (hio->hio_error); 520} 521 522static void 523secondary_exit(int exitcode, const char *fmt, ...) 524{ 525 va_list ap; 526 527 assert(exitcode != EX_OK); 528 va_start(ap, fmt); 529 pjdlogv_errno(LOG_ERR, fmt, ap); 530 va_end(ap); 531 event_send(gres, EVENT_DISCONNECT); 532 exit(exitcode); 533} 534 535/* 536 * Thread receives requests from the primary node. 537 */ 538static void * 539recv_thread(void *arg) 540{ 541 struct hast_resource *res = arg; 542 struct hio *hio; 543 544 for (;;) { 545 pjdlog_debug(2, "recv: Taking free request."); 546 QUEUE_TAKE(free, hio); 547 pjdlog_debug(2, "recv: (%p) Got request.", hio); 548 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 549 secondary_exit(EX_TEMPFAIL, 550 "Unable to receive request header"); 551 } 552 if (requnpack(res, hio) != 0) { 553 pjdlog_debug(2, 554 "recv: (%p) Moving request to the send queue.", 555 hio); 556 QUEUE_INSERT(send, hio); 557 continue; 558 } 559 reqlog(LOG_DEBUG, 2, -1, hio, 560 "recv: (%p) Got request header: ", hio); 561 if (hio->hio_cmd == HIO_KEEPALIVE) { 562 pjdlog_debug(2, 563 "recv: (%p) Moving request to the free queue.", 564 hio); 565 nv_free(hio->hio_nv); 566 QUEUE_INSERT(free, hio); 567 continue; 568 } else if (hio->hio_cmd == HIO_WRITE) { 569 if (hast_proto_recv_data(res, res->hr_remotein, 570 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 571 secondary_exit(EX_TEMPFAIL, 572 "Unable to receive reply data"); 573 } 574 } 575 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 576 hio); 577 QUEUE_INSERT(disk, hio); 578 } 579 /* NOTREACHED */ 580 return (NULL); 581} 582 583/* 584 * Thread reads from or writes to local component and also handles DELETE and 585 * FLUSH requests. 586 */ 587static void * 588disk_thread(void *arg) 589{ 590 struct hast_resource *res = arg; 591 struct hio *hio; 592 ssize_t ret; 593 bool clear_activemap; 594 595 clear_activemap = true; 596 597 for (;;) { 598 pjdlog_debug(2, "disk: Taking request."); 599 QUEUE_TAKE(disk, hio); 600 while (clear_activemap) { 601 unsigned char *map; 602 size_t mapsize; 603 604 /* 605 * When first request is received, it means that primary 606 * already received our activemap, merged it and stored 607 * locally. We can now safely clear our activemap. 608 */ 609 mapsize = 610 activemap_calc_ondisk_size(res->hr_local_mediasize - 611 METADATA_SIZE, res->hr_extentsize, 612 res->hr_local_sectorsize); 613 map = calloc(1, mapsize); 614 if (map == NULL) { 615 pjdlog_warning("Unable to allocate memory to clear local activemap."); 616 break; 617 } 618 if (pwrite(res->hr_localfd, map, mapsize, 619 METADATA_SIZE) != (ssize_t)mapsize) { 620 pjdlog_errno(LOG_WARNING, 621 "Unable to store cleared activemap"); 622 free(map); 623 break; 624 } 625 free(map); 626 clear_activemap = false; 627 pjdlog_debug(1, "Local activemap cleared."); 628 } 629 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 630 /* Handle the actual request. */ 631 switch (hio->hio_cmd) { 632 case HIO_READ: 633 ret = pread(res->hr_localfd, hio->hio_data, 634 hio->hio_length, 635 hio->hio_offset + res->hr_localoff); 636 if (ret < 0) 637 hio->hio_error = errno; 638 else if (ret != (int64_t)hio->hio_length) 639 hio->hio_error = EIO; 640 else 641 hio->hio_error = 0; 642 break; 643 case HIO_WRITE: 644 ret = pwrite(res->hr_localfd, hio->hio_data, 645 hio->hio_length, 646 hio->hio_offset + res->hr_localoff); 647 if (ret < 0) 648 hio->hio_error = errno; 649 else if (ret != (int64_t)hio->hio_length) 650 hio->hio_error = EIO; 651 else 652 hio->hio_error = 0; 653 break; 654 case HIO_DELETE: 655 ret = g_delete(res->hr_localfd, 656 hio->hio_offset + res->hr_localoff, 657 hio->hio_length); 658 if (ret < 0) 659 hio->hio_error = errno; 660 else 661 hio->hio_error = 0; 662 break; 663 case HIO_FLUSH: 664 ret = g_flush(res->hr_localfd); 665 if (ret < 0) 666 hio->hio_error = errno; 667 else 668 hio->hio_error = 0; 669 break; 670 } 671 if (hio->hio_error != 0) { 672 reqlog(LOG_ERR, 0, hio->hio_error, hio, 673 "Request failed: "); 674 } 675 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 676 hio); 677 QUEUE_INSERT(send, hio); 678 } 679 /* NOTREACHED */ 680 return (NULL); 681} 682 683/* 684 * Thread sends requests back to primary node. 685 */ 686static void * 687send_thread(void *arg) 688{ 689 struct hast_resource *res = arg; 690 struct nv *nvout; 691 struct hio *hio; 692 void *data; 693 size_t length; 694 695 for (;;) { 696 pjdlog_debug(2, "send: Taking request."); 697 QUEUE_TAKE(send, hio); 698 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 699 nvout = nv_alloc(); 700 /* Copy sequence number. */ 701 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 702 switch (hio->hio_cmd) { 703 case HIO_READ: 704 if (hio->hio_error == 0) { 705 data = hio->hio_data; 706 length = hio->hio_length; 707 break; 708 } 709 /* 710 * We send no data in case of an error. 711 */ 712 /* FALLTHROUGH */ 713 case HIO_DELETE: 714 case HIO_FLUSH: 715 case HIO_WRITE: 716 data = NULL; 717 length = 0; 718 break; 719 default: 720 abort(); 721 break; 722 } 723 if (hio->hio_error != 0) 724 nv_add_int16(nvout, hio->hio_error, "error"); 725 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 726 length) < 0) { 727 secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 728 } 729 nv_free(nvout); 730 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 731 hio); 732 nv_free(hio->hio_nv); 733 hio->hio_error = 0; 734 QUEUE_INSERT(free, hio); 735 } 736 /* NOTREACHED */ 737 return (NULL); 738} 739