secondary.c revision 218138
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 218138 2011-01-31 18:32:17Z 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 <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 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. If primary node done no 246 * writes yet as well (we will find "virgin" argument) then 247 * there is no need to synchronize anything. If primary node 248 * done any writes already we have to synchronize everything. 249 */ 250 PJDLOG_ASSERT(res->hr_secondary_localcnt == 0); 251 res->hr_resuid = resuid; 252 if (metadata_write(res) < 0) 253 exit(EX_NOINPUT); 254 if (nv_exists(nvin, "virgin")) { 255 free(map); 256 map = NULL; 257 mapsize = 0; 258 } else { 259 memset(map, 0xff, mapsize); 260 } 261 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 262 } else if ( 263 /* Is primary is out-of-date? */ 264 (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 265 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 266 /* Node are more or less in sync? */ 267 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 268 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 269 /* Is secondary is out-of-date? */ 270 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 271 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 272 /* 273 * Nodes are more or less in sync or one of the nodes is 274 * out-of-date. 275 * It doesn't matter at this point which one, we just have to 276 * send out local bitmap to the remote node. 277 */ 278 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 279 (ssize_t)mapsize) { 280 pjdlog_exit(LOG_ERR, "Unable to read activemap"); 281 } 282 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 283 res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 284 /* Primary is out-of-date, sync from secondary. */ 285 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 286 } else { 287 /* 288 * Secondary is out-of-date or counts match. 289 * Sync from primary. 290 */ 291 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 292 } 293 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 294 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 295 /* 296 * Not good, we have split-brain condition. 297 */ 298 pjdlog_error("Split-brain detected, exiting."); 299 nv_add_string(nvout, "Split-brain condition!", "errmsg"); 300 free(map); 301 map = NULL; 302 mapsize = 0; 303 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 304 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 305 /* 306 * This should never happen in practise, but we will perform 307 * full synchronization. 308 */ 309 PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 310 res->hr_primary_localcnt < res->hr_secondary_remotecnt); 311 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 312 METADATA_SIZE, res->hr_extentsize, 313 res->hr_local_sectorsize); 314 memset(map, 0xff, mapsize); 315 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 316 /* In this one of five cases sync from secondary. */ 317 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 318 } else { 319 /* For the rest four cases sync from primary. */ 320 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 321 } 322 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 323 (uintmax_t)res->hr_primary_localcnt, 324 (uintmax_t)res->hr_primary_remotecnt, 325 (uintmax_t)res->hr_secondary_localcnt, 326 (uintmax_t)res->hr_secondary_remotecnt); 327 } 328 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 329 pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 330 res->hr_remoteaddr); 331 } 332 if (map != NULL) 333 free(map); 334 nv_free(nvout); 335 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 336 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 337 /* Exit on split-brain. */ 338 event_send(res, EVENT_SPLITBRAIN); 339 exit(EX_CONFIG); 340 } 341} 342 343void 344hastd_secondary(struct hast_resource *res, struct nv *nvin) 345{ 346 sigset_t mask; 347 pthread_t td; 348 pid_t pid; 349 int error, mode; 350 351 /* 352 * Create communication channel between parent and child. 353 */ 354 if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 355 KEEP_ERRNO((void)pidfile_remove(pfh)); 356 pjdlog_exit(EX_OSERR, 357 "Unable to create control sockets between parent and child"); 358 } 359 /* 360 * Create communication channel between child and parent. 361 */ 362 if (proto_client("socketpair://", &res->hr_event) < 0) { 363 KEEP_ERRNO((void)pidfile_remove(pfh)); 364 pjdlog_exit(EX_OSERR, 365 "Unable to create event sockets between child and parent"); 366 } 367 368 pid = fork(); 369 if (pid < 0) { 370 KEEP_ERRNO((void)pidfile_remove(pfh)); 371 pjdlog_exit(EX_OSERR, "Unable to fork"); 372 } 373 374 if (pid > 0) { 375 /* This is parent. */ 376 proto_close(res->hr_remotein); 377 res->hr_remotein = NULL; 378 proto_close(res->hr_remoteout); 379 res->hr_remoteout = NULL; 380 /* Declare that we are receiver. */ 381 proto_recv(res->hr_event, NULL, 0); 382 /* Declare that we are sender. */ 383 proto_send(res->hr_ctrl, NULL, 0); 384 res->hr_workerpid = pid; 385 return; 386 } 387 388 gres = res; 389 mode = pjdlog_mode_get(); 390 391 /* Declare that we are sender. */ 392 proto_send(res->hr_event, NULL, 0); 393 /* Declare that we are receiver. */ 394 proto_recv(res->hr_ctrl, NULL, 0); 395 descriptors_cleanup(res); 396 397 descriptors_assert(res, mode); 398 399 pjdlog_init(mode); 400 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 401 setproctitle("%s (secondary)", res->hr_name); 402 403 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 404 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 405 406 /* Error in setting timeout is not critical, but why should it fail? */ 407 if (proto_timeout(res->hr_remotein, 0) < 0) 408 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 409 if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 410 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 411 412 init_local(res); 413 init_environment(); 414 415 if (drop_privs() != 0) 416 exit(EX_CONFIG); 417 418 /* 419 * Create the control thread before sending any event to the parent, 420 * as we can deadlock when parent sends control request to worker, 421 * but worker has no control thread started yet, so parent waits. 422 * In the meantime worker sends an event to the parent, but parent 423 * is unable to handle the event, because it waits for control 424 * request response. 425 */ 426 error = pthread_create(&td, NULL, ctrl_thread, res); 427 PJDLOG_ASSERT(error == 0); 428 429 init_remote(res, nvin); 430 event_send(res, EVENT_CONNECT); 431 432 error = pthread_create(&td, NULL, recv_thread, res); 433 PJDLOG_ASSERT(error == 0); 434 error = pthread_create(&td, NULL, disk_thread, res); 435 PJDLOG_ASSERT(error == 0); 436 (void)send_thread(res); 437} 438 439static void 440reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 441{ 442 char msg[1024]; 443 va_list ap; 444 int len; 445 446 va_start(ap, fmt); 447 len = vsnprintf(msg, sizeof(msg), fmt, ap); 448 va_end(ap); 449 if ((size_t)len < sizeof(msg)) { 450 switch (hio->hio_cmd) { 451 case HIO_READ: 452 (void)snprintf(msg + len, sizeof(msg) - len, 453 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 454 (uintmax_t)hio->hio_length); 455 break; 456 case HIO_DELETE: 457 (void)snprintf(msg + len, sizeof(msg) - len, 458 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 459 (uintmax_t)hio->hio_length); 460 break; 461 case HIO_FLUSH: 462 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 463 break; 464 case HIO_WRITE: 465 (void)snprintf(msg + len, sizeof(msg) - len, 466 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 467 (uintmax_t)hio->hio_length); 468 break; 469 case HIO_KEEPALIVE: 470 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 471 break; 472 default: 473 (void)snprintf(msg + len, sizeof(msg) - len, 474 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 475 break; 476 } 477 } 478 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 479} 480 481static int 482requnpack(struct hast_resource *res, struct hio *hio) 483{ 484 485 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 486 if (hio->hio_cmd == 0) { 487 pjdlog_error("Header contains no 'cmd' field."); 488 hio->hio_error = EINVAL; 489 goto end; 490 } 491 switch (hio->hio_cmd) { 492 case HIO_KEEPALIVE: 493 break; 494 case HIO_READ: 495 case HIO_WRITE: 496 case HIO_DELETE: 497 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 498 if (nv_error(hio->hio_nv) != 0) { 499 pjdlog_error("Header is missing 'offset' field."); 500 hio->hio_error = EINVAL; 501 goto end; 502 } 503 hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 504 if (nv_error(hio->hio_nv) != 0) { 505 pjdlog_error("Header is missing 'length' field."); 506 hio->hio_error = EINVAL; 507 goto end; 508 } 509 if (hio->hio_length == 0) { 510 pjdlog_error("Data length is zero."); 511 hio->hio_error = EINVAL; 512 goto end; 513 } 514 if (hio->hio_length > MAXPHYS) { 515 pjdlog_error("Data length is too large (%ju > %ju).", 516 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 517 hio->hio_error = EINVAL; 518 goto end; 519 } 520 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 521 pjdlog_error("Offset %ju is not multiple of sector size.", 522 (uintmax_t)hio->hio_offset); 523 hio->hio_error = EINVAL; 524 goto end; 525 } 526 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 527 pjdlog_error("Length %ju is not multiple of sector size.", 528 (uintmax_t)hio->hio_length); 529 hio->hio_error = EINVAL; 530 goto end; 531 } 532 if (hio->hio_offset + hio->hio_length > 533 (uint64_t)res->hr_datasize) { 534 pjdlog_error("Data offset is too large (%ju > %ju).", 535 (uintmax_t)(hio->hio_offset + hio->hio_length), 536 (uintmax_t)res->hr_datasize); 537 hio->hio_error = EINVAL; 538 goto end; 539 } 540 break; 541 default: 542 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 543 hio->hio_cmd); 544 hio->hio_error = EINVAL; 545 goto end; 546 } 547 hio->hio_error = 0; 548end: 549 return (hio->hio_error); 550} 551 552static __dead2 void 553secondary_exit(int exitcode, const char *fmt, ...) 554{ 555 va_list ap; 556 557 PJDLOG_ASSERT(exitcode != EX_OK); 558 va_start(ap, fmt); 559 pjdlogv_errno(LOG_ERR, fmt, ap); 560 va_end(ap); 561 event_send(gres, EVENT_DISCONNECT); 562 exit(exitcode); 563} 564 565/* 566 * Thread receives requests from the primary node. 567 */ 568static void * 569recv_thread(void *arg) 570{ 571 struct hast_resource *res = arg; 572 struct hio *hio; 573 574 for (;;) { 575 pjdlog_debug(2, "recv: Taking free request."); 576 QUEUE_TAKE(free, hio); 577 pjdlog_debug(2, "recv: (%p) Got request.", hio); 578 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 579 secondary_exit(EX_TEMPFAIL, 580 "Unable to receive request header"); 581 } 582 if (requnpack(res, hio) != 0) { 583 pjdlog_debug(2, 584 "recv: (%p) Moving request to the send queue.", 585 hio); 586 QUEUE_INSERT(send, hio); 587 continue; 588 } 589 reqlog(LOG_DEBUG, 2, -1, hio, 590 "recv: (%p) Got request header: ", hio); 591 if (hio->hio_cmd == HIO_KEEPALIVE) { 592 pjdlog_debug(2, 593 "recv: (%p) Moving request to the free queue.", 594 hio); 595 nv_free(hio->hio_nv); 596 QUEUE_INSERT(free, hio); 597 continue; 598 } else if (hio->hio_cmd == HIO_WRITE) { 599 if (hast_proto_recv_data(res, res->hr_remotein, 600 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 601 secondary_exit(EX_TEMPFAIL, 602 "Unable to receive request data"); 603 } 604 } 605 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 606 hio); 607 QUEUE_INSERT(disk, hio); 608 } 609 /* NOTREACHED */ 610 return (NULL); 611} 612 613/* 614 * Thread reads from or writes to local component and also handles DELETE and 615 * FLUSH requests. 616 */ 617static void * 618disk_thread(void *arg) 619{ 620 struct hast_resource *res = arg; 621 struct hio *hio; 622 ssize_t ret; 623 bool clear_activemap; 624 625 clear_activemap = true; 626 627 for (;;) { 628 pjdlog_debug(2, "disk: Taking request."); 629 QUEUE_TAKE(disk, hio); 630 while (clear_activemap) { 631 unsigned char *map; 632 size_t mapsize; 633 634 /* 635 * When first request is received, it means that primary 636 * already received our activemap, merged it and stored 637 * locally. We can now safely clear our activemap. 638 */ 639 mapsize = 640 activemap_calc_ondisk_size(res->hr_local_mediasize - 641 METADATA_SIZE, res->hr_extentsize, 642 res->hr_local_sectorsize); 643 map = calloc(1, mapsize); 644 if (map == NULL) { 645 pjdlog_warning("Unable to allocate memory to clear local activemap."); 646 break; 647 } 648 if (pwrite(res->hr_localfd, map, mapsize, 649 METADATA_SIZE) != (ssize_t)mapsize) { 650 pjdlog_errno(LOG_WARNING, 651 "Unable to store cleared activemap"); 652 free(map); 653 break; 654 } 655 free(map); 656 clear_activemap = false; 657 pjdlog_debug(1, "Local activemap cleared."); 658 } 659 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 660 /* Handle the actual request. */ 661 switch (hio->hio_cmd) { 662 case HIO_READ: 663 ret = pread(res->hr_localfd, hio->hio_data, 664 hio->hio_length, 665 hio->hio_offset + res->hr_localoff); 666 if (ret < 0) 667 hio->hio_error = errno; 668 else if (ret != (int64_t)hio->hio_length) 669 hio->hio_error = EIO; 670 else 671 hio->hio_error = 0; 672 break; 673 case HIO_WRITE: 674 ret = pwrite(res->hr_localfd, hio->hio_data, 675 hio->hio_length, 676 hio->hio_offset + res->hr_localoff); 677 if (ret < 0) 678 hio->hio_error = errno; 679 else if (ret != (int64_t)hio->hio_length) 680 hio->hio_error = EIO; 681 else 682 hio->hio_error = 0; 683 break; 684 case HIO_DELETE: 685 ret = g_delete(res->hr_localfd, 686 hio->hio_offset + res->hr_localoff, 687 hio->hio_length); 688 if (ret < 0) 689 hio->hio_error = errno; 690 else 691 hio->hio_error = 0; 692 break; 693 case HIO_FLUSH: 694 ret = g_flush(res->hr_localfd); 695 if (ret < 0) 696 hio->hio_error = errno; 697 else 698 hio->hio_error = 0; 699 break; 700 } 701 if (hio->hio_error != 0) { 702 reqlog(LOG_ERR, 0, hio->hio_error, hio, 703 "Request failed: "); 704 } 705 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 706 hio); 707 QUEUE_INSERT(send, hio); 708 } 709 /* NOTREACHED */ 710 return (NULL); 711} 712 713/* 714 * Thread sends requests back to primary node. 715 */ 716static void * 717send_thread(void *arg) 718{ 719 struct hast_resource *res = arg; 720 struct nv *nvout; 721 struct hio *hio; 722 void *data; 723 size_t length; 724 725 for (;;) { 726 pjdlog_debug(2, "send: Taking request."); 727 QUEUE_TAKE(send, hio); 728 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 729 nvout = nv_alloc(); 730 /* Copy sequence number. */ 731 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 732 switch (hio->hio_cmd) { 733 case HIO_READ: 734 if (hio->hio_error == 0) { 735 data = hio->hio_data; 736 length = hio->hio_length; 737 break; 738 } 739 /* 740 * We send no data in case of an error. 741 */ 742 /* FALLTHROUGH */ 743 case HIO_DELETE: 744 case HIO_FLUSH: 745 case HIO_WRITE: 746 data = NULL; 747 length = 0; 748 break; 749 default: 750 abort(); 751 break; 752 } 753 if (hio->hio_error != 0) 754 nv_add_int16(nvout, hio->hio_error, "error"); 755 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 756 length) < 0) { 757 secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 758 } 759 nv_free(nvout); 760 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 761 hio); 762 nv_free(hio->hio_nv); 763 hio->hio_error = 0; 764 QUEUE_INSERT(free, hio); 765 } 766 /* NOTREACHED */ 767 return (NULL); 768} 769