secondary.c revision 219831
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 219831 2011-03-21 14:51:16Z 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 (res->hr_resuid != resuid) { 263 char errmsg[256]; 264 265 (void)snprintf(errmsg, sizeof(errmsg), 266 "Resource unique ID mismatch (primary=%ju, secondary=%ju).", 267 (uintmax_t)resuid, (uintmax_t)res->hr_resuid); 268 pjdlog_error("%s", errmsg); 269 nv_add_string(nvout, errmsg, "errmsg"); 270 if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) { 271 pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s", 272 res->hr_remoteaddr); 273 } 274 nv_free(nvout); 275 exit(EX_CONFIG); 276 } else if ( 277 /* Is primary is out-of-date? */ 278 (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 279 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 280 /* Node are more or less in sync? */ 281 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 282 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 283 /* Is secondary is out-of-date? */ 284 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 285 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 286 /* 287 * Nodes are more or less in sync or one of the nodes is 288 * out-of-date. 289 * It doesn't matter at this point which one, we just have to 290 * send out local bitmap to the remote node. 291 */ 292 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 293 (ssize_t)mapsize) { 294 pjdlog_exit(LOG_ERR, "Unable to read activemap"); 295 } 296 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 297 res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 298 /* Primary is out-of-date, sync from secondary. */ 299 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 300 } else { 301 /* 302 * Secondary is out-of-date or counts match. 303 * Sync from primary. 304 */ 305 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 306 } 307 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 308 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 309 /* 310 * Not good, we have split-brain condition. 311 */ 312 pjdlog_error("Split-brain detected, exiting."); 313 nv_add_string(nvout, "Split-brain condition!", "errmsg"); 314 free(map); 315 map = NULL; 316 mapsize = 0; 317 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 318 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 319 /* 320 * This should never happen in practise, but we will perform 321 * full synchronization. 322 */ 323 PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 324 res->hr_primary_localcnt < res->hr_secondary_remotecnt); 325 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 326 METADATA_SIZE, res->hr_extentsize, 327 res->hr_local_sectorsize); 328 memset(map, 0xff, mapsize); 329 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 330 /* In this one of five cases sync from secondary. */ 331 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 332 } else { 333 /* For the rest four cases sync from primary. */ 334 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 335 } 336 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 337 (uintmax_t)res->hr_primary_localcnt, 338 (uintmax_t)res->hr_primary_remotecnt, 339 (uintmax_t)res->hr_secondary_localcnt, 340 (uintmax_t)res->hr_secondary_remotecnt); 341 } 342 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 343 pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 344 res->hr_remoteaddr); 345 } 346 if (map != NULL) 347 free(map); 348 nv_free(nvout); 349 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 350 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 351 /* Exit on split-brain. */ 352 event_send(res, EVENT_SPLITBRAIN); 353 exit(EX_CONFIG); 354 } 355} 356 357void 358hastd_secondary(struct hast_resource *res, struct nv *nvin) 359{ 360 sigset_t mask; 361 pthread_t td; 362 pid_t pid; 363 int error, mode, debuglevel; 364 365 /* 366 * Create communication channel between parent and child. 367 */ 368 if (proto_client(NULL, "socketpair://", &res->hr_ctrl) < 0) { 369 KEEP_ERRNO((void)pidfile_remove(pfh)); 370 pjdlog_exit(EX_OSERR, 371 "Unable to create control sockets between parent and child"); 372 } 373 /* 374 * Create communication channel between child and parent. 375 */ 376 if (proto_client(NULL, "socketpair://", &res->hr_event) < 0) { 377 KEEP_ERRNO((void)pidfile_remove(pfh)); 378 pjdlog_exit(EX_OSERR, 379 "Unable to create event sockets between child and parent"); 380 } 381 /* 382 * Create communication channel for sending connection requests from 383 * parent to child. 384 */ 385 if (proto_client(NULL, "socketpair://", &res->hr_conn) < 0) { 386 /* TODO: There's no need for this to be fatal error. */ 387 KEEP_ERRNO((void)pidfile_remove(pfh)); 388 pjdlog_exit(EX_OSERR, 389 "Unable to create connection sockets between parent and child"); 390 } 391 392 pid = fork(); 393 if (pid < 0) { 394 KEEP_ERRNO((void)pidfile_remove(pfh)); 395 pjdlog_exit(EX_OSERR, "Unable to fork"); 396 } 397 398 if (pid > 0) { 399 /* This is parent. */ 400 proto_close(res->hr_remotein); 401 res->hr_remotein = NULL; 402 proto_close(res->hr_remoteout); 403 res->hr_remoteout = NULL; 404 /* Declare that we are receiver. */ 405 proto_recv(res->hr_event, NULL, 0); 406 /* Declare that we are sender. */ 407 proto_send(res->hr_ctrl, NULL, 0); 408 proto_send(res->hr_conn, NULL, 0); 409 res->hr_workerpid = pid; 410 return; 411 } 412 413 gres = res; 414 mode = pjdlog_mode_get(); 415 debuglevel = pjdlog_debug_get(); 416 417 /* Declare that we are sender. */ 418 proto_send(res->hr_event, NULL, 0); 419 /* Declare that we are receiver. */ 420 proto_recv(res->hr_ctrl, NULL, 0); 421 proto_recv(res->hr_conn, NULL, 0); 422 descriptors_cleanup(res); 423 424 descriptors_assert(res, mode); 425 426 pjdlog_init(mode); 427 pjdlog_debug_set(debuglevel); 428 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 429 setproctitle("%s (secondary)", res->hr_name); 430 431 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 432 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 433 434 /* Error in setting timeout is not critical, but why should it fail? */ 435 if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0) 436 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 437 if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 438 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 439 440 init_local(res); 441 init_environment(); 442 443 if (drop_privs() != 0) 444 exit(EX_CONFIG); 445 pjdlog_info("Privileges successfully dropped."); 446 447 /* 448 * Create the control thread before sending any event to the parent, 449 * as we can deadlock when parent sends control request to worker, 450 * but worker has no control thread started yet, so parent waits. 451 * In the meantime worker sends an event to the parent, but parent 452 * is unable to handle the event, because it waits for control 453 * request response. 454 */ 455 error = pthread_create(&td, NULL, ctrl_thread, res); 456 PJDLOG_ASSERT(error == 0); 457 458 init_remote(res, nvin); 459 event_send(res, EVENT_CONNECT); 460 461 error = pthread_create(&td, NULL, recv_thread, res); 462 PJDLOG_ASSERT(error == 0); 463 error = pthread_create(&td, NULL, disk_thread, res); 464 PJDLOG_ASSERT(error == 0); 465 (void)send_thread(res); 466} 467 468static void 469reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 470{ 471 char msg[1024]; 472 va_list ap; 473 int len; 474 475 va_start(ap, fmt); 476 len = vsnprintf(msg, sizeof(msg), fmt, ap); 477 va_end(ap); 478 if ((size_t)len < sizeof(msg)) { 479 switch (hio->hio_cmd) { 480 case HIO_READ: 481 (void)snprintf(msg + len, sizeof(msg) - len, 482 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 483 (uintmax_t)hio->hio_length); 484 break; 485 case HIO_DELETE: 486 (void)snprintf(msg + len, sizeof(msg) - len, 487 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 488 (uintmax_t)hio->hio_length); 489 break; 490 case HIO_FLUSH: 491 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 492 break; 493 case HIO_WRITE: 494 (void)snprintf(msg + len, sizeof(msg) - len, 495 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 496 (uintmax_t)hio->hio_length); 497 break; 498 case HIO_KEEPALIVE: 499 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 500 break; 501 default: 502 (void)snprintf(msg + len, sizeof(msg) - len, 503 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 504 break; 505 } 506 } 507 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 508} 509 510static int 511requnpack(struct hast_resource *res, struct hio *hio) 512{ 513 514 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 515 if (hio->hio_cmd == 0) { 516 pjdlog_error("Header contains no 'cmd' field."); 517 hio->hio_error = EINVAL; 518 goto end; 519 } 520 switch (hio->hio_cmd) { 521 case HIO_KEEPALIVE: 522 break; 523 case HIO_READ: 524 case HIO_WRITE: 525 case HIO_DELETE: 526 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 527 if (nv_error(hio->hio_nv) != 0) { 528 pjdlog_error("Header is missing 'offset' field."); 529 hio->hio_error = EINVAL; 530 goto end; 531 } 532 hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 533 if (nv_error(hio->hio_nv) != 0) { 534 pjdlog_error("Header is missing 'length' field."); 535 hio->hio_error = EINVAL; 536 goto end; 537 } 538 if (hio->hio_length == 0) { 539 pjdlog_error("Data length is zero."); 540 hio->hio_error = EINVAL; 541 goto end; 542 } 543 if (hio->hio_length > MAXPHYS) { 544 pjdlog_error("Data length is too large (%ju > %ju).", 545 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 546 hio->hio_error = EINVAL; 547 goto end; 548 } 549 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 550 pjdlog_error("Offset %ju is not multiple of sector size.", 551 (uintmax_t)hio->hio_offset); 552 hio->hio_error = EINVAL; 553 goto end; 554 } 555 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 556 pjdlog_error("Length %ju is not multiple of sector size.", 557 (uintmax_t)hio->hio_length); 558 hio->hio_error = EINVAL; 559 goto end; 560 } 561 if (hio->hio_offset + hio->hio_length > 562 (uint64_t)res->hr_datasize) { 563 pjdlog_error("Data offset is too large (%ju > %ju).", 564 (uintmax_t)(hio->hio_offset + hio->hio_length), 565 (uintmax_t)res->hr_datasize); 566 hio->hio_error = EINVAL; 567 goto end; 568 } 569 break; 570 default: 571 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 572 hio->hio_cmd); 573 hio->hio_error = EINVAL; 574 goto end; 575 } 576 hio->hio_error = 0; 577end: 578 return (hio->hio_error); 579} 580 581static __dead2 void 582secondary_exit(int exitcode, const char *fmt, ...) 583{ 584 va_list ap; 585 586 PJDLOG_ASSERT(exitcode != EX_OK); 587 va_start(ap, fmt); 588 pjdlogv_errno(LOG_ERR, fmt, ap); 589 va_end(ap); 590 event_send(gres, EVENT_DISCONNECT); 591 exit(exitcode); 592} 593 594/* 595 * Thread receives requests from the primary node. 596 */ 597static void * 598recv_thread(void *arg) 599{ 600 struct hast_resource *res = arg; 601 struct hio *hio; 602 603 for (;;) { 604 pjdlog_debug(2, "recv: Taking free request."); 605 QUEUE_TAKE(free, hio); 606 pjdlog_debug(2, "recv: (%p) Got request.", hio); 607 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 608 secondary_exit(EX_TEMPFAIL, 609 "Unable to receive request header"); 610 } 611 if (requnpack(res, hio) != 0) { 612 pjdlog_debug(2, 613 "recv: (%p) Moving request to the send queue.", 614 hio); 615 QUEUE_INSERT(send, hio); 616 continue; 617 } 618 reqlog(LOG_DEBUG, 2, -1, hio, 619 "recv: (%p) Got request header: ", hio); 620 if (hio->hio_cmd == HIO_KEEPALIVE) { 621 pjdlog_debug(2, 622 "recv: (%p) Moving request to the free queue.", 623 hio); 624 nv_free(hio->hio_nv); 625 QUEUE_INSERT(free, hio); 626 continue; 627 } else if (hio->hio_cmd == HIO_WRITE) { 628 if (hast_proto_recv_data(res, res->hr_remotein, 629 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 630 secondary_exit(EX_TEMPFAIL, 631 "Unable to receive request data"); 632 } 633 } 634 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 635 hio); 636 QUEUE_INSERT(disk, hio); 637 } 638 /* NOTREACHED */ 639 return (NULL); 640} 641 642/* 643 * Thread reads from or writes to local component and also handles DELETE and 644 * FLUSH requests. 645 */ 646static void * 647disk_thread(void *arg) 648{ 649 struct hast_resource *res = arg; 650 struct hio *hio; 651 ssize_t ret; 652 bool clear_activemap; 653 654 clear_activemap = true; 655 656 for (;;) { 657 pjdlog_debug(2, "disk: Taking request."); 658 QUEUE_TAKE(disk, hio); 659 while (clear_activemap) { 660 unsigned char *map; 661 size_t mapsize; 662 663 /* 664 * When first request is received, it means that primary 665 * already received our activemap, merged it and stored 666 * locally. We can now safely clear our activemap. 667 */ 668 mapsize = 669 activemap_calc_ondisk_size(res->hr_local_mediasize - 670 METADATA_SIZE, res->hr_extentsize, 671 res->hr_local_sectorsize); 672 map = calloc(1, mapsize); 673 if (map == NULL) { 674 pjdlog_warning("Unable to allocate memory to clear local activemap."); 675 break; 676 } 677 if (pwrite(res->hr_localfd, map, mapsize, 678 METADATA_SIZE) != (ssize_t)mapsize) { 679 pjdlog_errno(LOG_WARNING, 680 "Unable to store cleared activemap"); 681 free(map); 682 break; 683 } 684 free(map); 685 clear_activemap = false; 686 pjdlog_debug(1, "Local activemap cleared."); 687 } 688 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 689 /* Handle the actual request. */ 690 switch (hio->hio_cmd) { 691 case HIO_READ: 692 ret = pread(res->hr_localfd, hio->hio_data, 693 hio->hio_length, 694 hio->hio_offset + res->hr_localoff); 695 if (ret < 0) 696 hio->hio_error = errno; 697 else if (ret != (int64_t)hio->hio_length) 698 hio->hio_error = EIO; 699 else 700 hio->hio_error = 0; 701 break; 702 case HIO_WRITE: 703 ret = pwrite(res->hr_localfd, hio->hio_data, 704 hio->hio_length, 705 hio->hio_offset + res->hr_localoff); 706 if (ret < 0) 707 hio->hio_error = errno; 708 else if (ret != (int64_t)hio->hio_length) 709 hio->hio_error = EIO; 710 else 711 hio->hio_error = 0; 712 break; 713 case HIO_DELETE: 714 ret = g_delete(res->hr_localfd, 715 hio->hio_offset + res->hr_localoff, 716 hio->hio_length); 717 if (ret < 0) 718 hio->hio_error = errno; 719 else 720 hio->hio_error = 0; 721 break; 722 case HIO_FLUSH: 723 ret = g_flush(res->hr_localfd); 724 if (ret < 0) 725 hio->hio_error = errno; 726 else 727 hio->hio_error = 0; 728 break; 729 } 730 if (hio->hio_error != 0) { 731 reqlog(LOG_ERR, 0, hio->hio_error, hio, 732 "Request failed: "); 733 } 734 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 735 hio); 736 QUEUE_INSERT(send, hio); 737 } 738 /* NOTREACHED */ 739 return (NULL); 740} 741 742/* 743 * Thread sends requests back to primary node. 744 */ 745static void * 746send_thread(void *arg) 747{ 748 struct hast_resource *res = arg; 749 struct nv *nvout; 750 struct hio *hio; 751 void *data; 752 size_t length; 753 754 for (;;) { 755 pjdlog_debug(2, "send: Taking request."); 756 QUEUE_TAKE(send, hio); 757 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 758 nvout = nv_alloc(); 759 /* Copy sequence number. */ 760 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 761 switch (hio->hio_cmd) { 762 case HIO_READ: 763 if (hio->hio_error == 0) { 764 data = hio->hio_data; 765 length = hio->hio_length; 766 break; 767 } 768 /* 769 * We send no data in case of an error. 770 */ 771 /* FALLTHROUGH */ 772 case HIO_DELETE: 773 case HIO_FLUSH: 774 case HIO_WRITE: 775 data = NULL; 776 length = 0; 777 break; 778 default: 779 abort(); 780 break; 781 } 782 if (hio->hio_error != 0) 783 nv_add_int16(nvout, hio->hio_error, "error"); 784 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 785 length) < 0) { 786 secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 787 } 788 nv_free(nvout); 789 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 790 hio); 791 nv_free(hio->hio_nv); 792 hio->hio_error = 0; 793 QUEUE_INSERT(free, hio); 794 } 795 /* NOTREACHED */ 796 return (NULL); 797} 798