secondary.c revision 219900
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 219900 2011-03-23 11:09:04Z 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 /* Nodes 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 pid = fork(); 383 if (pid < 0) { 384 KEEP_ERRNO((void)pidfile_remove(pfh)); 385 pjdlog_exit(EX_OSERR, "Unable to fork"); 386 } 387 388 if (pid > 0) { 389 /* This is parent. */ 390 proto_close(res->hr_remotein); 391 res->hr_remotein = NULL; 392 proto_close(res->hr_remoteout); 393 res->hr_remoteout = NULL; 394 /* Declare that we are receiver. */ 395 proto_recv(res->hr_event, NULL, 0); 396 /* Declare that we are sender. */ 397 proto_send(res->hr_ctrl, NULL, 0); 398 res->hr_workerpid = pid; 399 return; 400 } 401 402 gres = res; 403 mode = pjdlog_mode_get(); 404 debuglevel = pjdlog_debug_get(); 405 406 /* Declare that we are sender. */ 407 proto_send(res->hr_event, NULL, 0); 408 /* Declare that we are receiver. */ 409 proto_recv(res->hr_ctrl, NULL, 0); 410 descriptors_cleanup(res); 411 412 descriptors_assert(res, mode); 413 414 pjdlog_init(mode); 415 pjdlog_debug_set(debuglevel); 416 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 417 setproctitle("%s (secondary)", res->hr_name); 418 419 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 420 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 421 422 /* Error in setting timeout is not critical, but why should it fail? */ 423 if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0) 424 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 425 if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 426 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 427 428 init_local(res); 429 init_environment(); 430 431 if (drop_privs(true) != 0) 432 exit(EX_CONFIG); 433 pjdlog_info("Privileges successfully dropped."); 434 435 /* 436 * Create the control thread before sending any event to the parent, 437 * as we can deadlock when parent sends control request to worker, 438 * but worker has no control thread started yet, so parent waits. 439 * In the meantime worker sends an event to the parent, but parent 440 * is unable to handle the event, because it waits for control 441 * request response. 442 */ 443 error = pthread_create(&td, NULL, ctrl_thread, res); 444 PJDLOG_ASSERT(error == 0); 445 446 init_remote(res, nvin); 447 event_send(res, EVENT_CONNECT); 448 449 error = pthread_create(&td, NULL, recv_thread, res); 450 PJDLOG_ASSERT(error == 0); 451 error = pthread_create(&td, NULL, disk_thread, res); 452 PJDLOG_ASSERT(error == 0); 453 (void)send_thread(res); 454} 455 456static void 457reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 458{ 459 char msg[1024]; 460 va_list ap; 461 int len; 462 463 va_start(ap, fmt); 464 len = vsnprintf(msg, sizeof(msg), fmt, ap); 465 va_end(ap); 466 if ((size_t)len < sizeof(msg)) { 467 switch (hio->hio_cmd) { 468 case HIO_READ: 469 (void)snprintf(msg + len, sizeof(msg) - len, 470 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 471 (uintmax_t)hio->hio_length); 472 break; 473 case HIO_DELETE: 474 (void)snprintf(msg + len, sizeof(msg) - len, 475 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 476 (uintmax_t)hio->hio_length); 477 break; 478 case HIO_FLUSH: 479 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 480 break; 481 case HIO_WRITE: 482 (void)snprintf(msg + len, sizeof(msg) - len, 483 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 484 (uintmax_t)hio->hio_length); 485 break; 486 case HIO_KEEPALIVE: 487 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 488 break; 489 default: 490 (void)snprintf(msg + len, sizeof(msg) - len, 491 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 492 break; 493 } 494 } 495 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 496} 497 498static int 499requnpack(struct hast_resource *res, struct hio *hio) 500{ 501 502 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 503 if (hio->hio_cmd == 0) { 504 pjdlog_error("Header contains no 'cmd' field."); 505 hio->hio_error = EINVAL; 506 goto end; 507 } 508 switch (hio->hio_cmd) { 509 case HIO_KEEPALIVE: 510 break; 511 case HIO_READ: 512 case HIO_WRITE: 513 case HIO_DELETE: 514 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 515 if (nv_error(hio->hio_nv) != 0) { 516 pjdlog_error("Header is missing 'offset' field."); 517 hio->hio_error = EINVAL; 518 goto end; 519 } 520 hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 521 if (nv_error(hio->hio_nv) != 0) { 522 pjdlog_error("Header is missing 'length' field."); 523 hio->hio_error = EINVAL; 524 goto end; 525 } 526 if (hio->hio_length == 0) { 527 pjdlog_error("Data length is zero."); 528 hio->hio_error = EINVAL; 529 goto end; 530 } 531 if (hio->hio_length > MAXPHYS) { 532 pjdlog_error("Data length is too large (%ju > %ju).", 533 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 534 hio->hio_error = EINVAL; 535 goto end; 536 } 537 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 538 pjdlog_error("Offset %ju is not multiple of sector size.", 539 (uintmax_t)hio->hio_offset); 540 hio->hio_error = EINVAL; 541 goto end; 542 } 543 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 544 pjdlog_error("Length %ju is not multiple of sector size.", 545 (uintmax_t)hio->hio_length); 546 hio->hio_error = EINVAL; 547 goto end; 548 } 549 if (hio->hio_offset + hio->hio_length > 550 (uint64_t)res->hr_datasize) { 551 pjdlog_error("Data offset is too large (%ju > %ju).", 552 (uintmax_t)(hio->hio_offset + hio->hio_length), 553 (uintmax_t)res->hr_datasize); 554 hio->hio_error = EINVAL; 555 goto end; 556 } 557 break; 558 default: 559 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 560 hio->hio_cmd); 561 hio->hio_error = EINVAL; 562 goto end; 563 } 564 hio->hio_error = 0; 565end: 566 return (hio->hio_error); 567} 568 569static __dead2 void 570secondary_exit(int exitcode, const char *fmt, ...) 571{ 572 va_list ap; 573 574 PJDLOG_ASSERT(exitcode != EX_OK); 575 va_start(ap, fmt); 576 pjdlogv_errno(LOG_ERR, fmt, ap); 577 va_end(ap); 578 event_send(gres, EVENT_DISCONNECT); 579 exit(exitcode); 580} 581 582/* 583 * Thread receives requests from the primary node. 584 */ 585static void * 586recv_thread(void *arg) 587{ 588 struct hast_resource *res = arg; 589 struct hio *hio; 590 591 for (;;) { 592 pjdlog_debug(2, "recv: Taking free request."); 593 QUEUE_TAKE(free, hio); 594 pjdlog_debug(2, "recv: (%p) Got request.", hio); 595 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 596 secondary_exit(EX_TEMPFAIL, 597 "Unable to receive request header"); 598 } 599 if (requnpack(res, hio) != 0) { 600 pjdlog_debug(2, 601 "recv: (%p) Moving request to the send queue.", 602 hio); 603 QUEUE_INSERT(send, hio); 604 continue; 605 } 606 reqlog(LOG_DEBUG, 2, -1, hio, 607 "recv: (%p) Got request header: ", hio); 608 if (hio->hio_cmd == HIO_KEEPALIVE) { 609 pjdlog_debug(2, 610 "recv: (%p) Moving request to the free queue.", 611 hio); 612 nv_free(hio->hio_nv); 613 QUEUE_INSERT(free, hio); 614 continue; 615 } else if (hio->hio_cmd == HIO_WRITE) { 616 if (hast_proto_recv_data(res, res->hr_remotein, 617 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 618 secondary_exit(EX_TEMPFAIL, 619 "Unable to receive request data"); 620 } 621 } 622 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 623 hio); 624 QUEUE_INSERT(disk, hio); 625 } 626 /* NOTREACHED */ 627 return (NULL); 628} 629 630/* 631 * Thread reads from or writes to local component and also handles DELETE and 632 * FLUSH requests. 633 */ 634static void * 635disk_thread(void *arg) 636{ 637 struct hast_resource *res = arg; 638 struct hio *hio; 639 ssize_t ret; 640 bool clear_activemap; 641 642 clear_activemap = true; 643 644 for (;;) { 645 pjdlog_debug(2, "disk: Taking request."); 646 QUEUE_TAKE(disk, hio); 647 while (clear_activemap) { 648 unsigned char *map; 649 size_t mapsize; 650 651 /* 652 * When first request is received, it means that primary 653 * already received our activemap, merged it and stored 654 * locally. We can now safely clear our activemap. 655 */ 656 mapsize = 657 activemap_calc_ondisk_size(res->hr_local_mediasize - 658 METADATA_SIZE, res->hr_extentsize, 659 res->hr_local_sectorsize); 660 map = calloc(1, mapsize); 661 if (map == NULL) { 662 pjdlog_warning("Unable to allocate memory to clear local activemap."); 663 break; 664 } 665 if (pwrite(res->hr_localfd, map, mapsize, 666 METADATA_SIZE) != (ssize_t)mapsize) { 667 pjdlog_errno(LOG_WARNING, 668 "Unable to store cleared activemap"); 669 free(map); 670 break; 671 } 672 free(map); 673 clear_activemap = false; 674 pjdlog_debug(1, "Local activemap cleared."); 675 } 676 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 677 /* Handle the actual request. */ 678 switch (hio->hio_cmd) { 679 case HIO_READ: 680 ret = pread(res->hr_localfd, hio->hio_data, 681 hio->hio_length, 682 hio->hio_offset + res->hr_localoff); 683 if (ret < 0) 684 hio->hio_error = errno; 685 else if (ret != (int64_t)hio->hio_length) 686 hio->hio_error = EIO; 687 else 688 hio->hio_error = 0; 689 break; 690 case HIO_WRITE: 691 ret = pwrite(res->hr_localfd, hio->hio_data, 692 hio->hio_length, 693 hio->hio_offset + res->hr_localoff); 694 if (ret < 0) 695 hio->hio_error = errno; 696 else if (ret != (int64_t)hio->hio_length) 697 hio->hio_error = EIO; 698 else 699 hio->hio_error = 0; 700 break; 701 case HIO_DELETE: 702 ret = g_delete(res->hr_localfd, 703 hio->hio_offset + res->hr_localoff, 704 hio->hio_length); 705 if (ret < 0) 706 hio->hio_error = errno; 707 else 708 hio->hio_error = 0; 709 break; 710 case HIO_FLUSH: 711 ret = g_flush(res->hr_localfd); 712 if (ret < 0) 713 hio->hio_error = errno; 714 else 715 hio->hio_error = 0; 716 break; 717 } 718 if (hio->hio_error != 0) { 719 reqlog(LOG_ERR, 0, hio->hio_error, hio, 720 "Request failed: "); 721 } 722 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 723 hio); 724 QUEUE_INSERT(send, hio); 725 } 726 /* NOTREACHED */ 727 return (NULL); 728} 729 730/* 731 * Thread sends requests back to primary node. 732 */ 733static void * 734send_thread(void *arg) 735{ 736 struct hast_resource *res = arg; 737 struct nv *nvout; 738 struct hio *hio; 739 void *data; 740 size_t length; 741 742 for (;;) { 743 pjdlog_debug(2, "send: Taking request."); 744 QUEUE_TAKE(send, hio); 745 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 746 nvout = nv_alloc(); 747 /* Copy sequence number. */ 748 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 749 switch (hio->hio_cmd) { 750 case HIO_READ: 751 if (hio->hio_error == 0) { 752 data = hio->hio_data; 753 length = hio->hio_length; 754 break; 755 } 756 /* 757 * We send no data in case of an error. 758 */ 759 /* FALLTHROUGH */ 760 case HIO_DELETE: 761 case HIO_FLUSH: 762 case HIO_WRITE: 763 data = NULL; 764 length = 0; 765 break; 766 default: 767 abort(); 768 break; 769 } 770 if (hio->hio_error != 0) 771 nv_add_int16(nvout, hio->hio_error, "error"); 772 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 773 length) < 0) { 774 secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 775 } 776 nv_free(nvout); 777 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 778 hio); 779 nv_free(hio->hio_nv); 780 hio->hio_error = 0; 781 QUEUE_INSERT(free, hio); 782 } 783 /* NOTREACHED */ 784 return (NULL); 785} 786