secondary.c revision 207371
1/*- 2 * Copyright (c) 2009-2010 The FreeBSD Foundation 3 * All rights reserved. 4 * 5 * This software was developed by Pawel Jakub Dawidek under sponsorship from 6 * the FreeBSD Foundation. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30#include <sys/cdefs.h> 31__FBSDID("$FreeBSD: head/sbin/hastd/secondary.c 207371 2010-04-29 15:36:32Z pjd $"); 32 33#include <sys/param.h> 34#include <sys/time.h> 35#include <sys/bio.h> 36#include <sys/disk.h> 37#include <sys/stat.h> 38 39#include <assert.h> 40#include <err.h> 41#include <errno.h> 42#include <fcntl.h> 43#include <libgeom.h> 44#include <pthread.h> 45#include <stdint.h> 46#include <stdio.h> 47#include <string.h> 48#include <sysexits.h> 49#include <unistd.h> 50 51#include <activemap.h> 52#include <nv.h> 53#include <pjdlog.h> 54 55#include "control.h" 56#include "hast.h" 57#include "hast_proto.h" 58#include "hastd.h" 59#include "metadata.h" 60#include "proto.h" 61#include "subr.h" 62#include "synch.h" 63 64struct hio { 65 uint64_t hio_seq; 66 int hio_error; 67 struct nv *hio_nv; 68 void *hio_data; 69 uint8_t hio_cmd; 70 uint64_t hio_offset; 71 uint64_t hio_length; 72 TAILQ_ENTRY(hio) hio_next; 73}; 74 75/* 76 * Free list holds unused structures. When free list is empty, we have to wait 77 * until some in-progress requests are freed. 78 */ 79static TAILQ_HEAD(, hio) hio_free_list; 80static pthread_mutex_t hio_free_list_lock; 81static pthread_cond_t hio_free_list_cond; 82/* 83 * Disk thread (the one that do I/O requests) takes requests from this list. 84 */ 85static TAILQ_HEAD(, hio) hio_disk_list; 86static pthread_mutex_t hio_disk_list_lock; 87static pthread_cond_t hio_disk_list_cond; 88/* 89 * There is one recv list for every component, although local components don't 90 * use recv lists as local requests are done synchronously. 91 */ 92static TAILQ_HEAD(, hio) hio_send_list; 93static pthread_mutex_t hio_send_list_lock; 94static pthread_cond_t hio_send_list_cond; 95 96/* 97 * Maximum number of outstanding I/O requests. 98 */ 99#define HAST_HIO_MAX 256 100 101static void *recv_thread(void *arg); 102static void *disk_thread(void *arg); 103static void *send_thread(void *arg); 104 105static void 106init_environment(void) 107{ 108 struct hio *hio; 109 unsigned int ii; 110 111 /* 112 * Initialize lists, their locks and theirs condition variables. 113 */ 114 TAILQ_INIT(&hio_free_list); 115 mtx_init(&hio_free_list_lock); 116 cv_init(&hio_free_list_cond); 117 TAILQ_INIT(&hio_disk_list); 118 mtx_init(&hio_disk_list_lock); 119 cv_init(&hio_disk_list_cond); 120 TAILQ_INIT(&hio_send_list); 121 mtx_init(&hio_send_list_lock); 122 cv_init(&hio_send_list_cond); 123 124 /* 125 * Allocate requests pool and initialize requests. 126 */ 127 for (ii = 0; ii < HAST_HIO_MAX; ii++) { 128 hio = malloc(sizeof(*hio)); 129 if (hio == NULL) { 130 errx(EX_TEMPFAIL, "cannot allocate %zu bytes of memory " 131 "for hio request", sizeof(*hio)); 132 } 133 hio->hio_error = 0; 134 hio->hio_data = malloc(MAXPHYS); 135 if (hio->hio_data == NULL) { 136 errx(EX_TEMPFAIL, "cannot allocate %zu bytes of memory " 137 "for gctl_data", (size_t)MAXPHYS); 138 } 139 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 140 } 141} 142 143static void 144init_local(struct hast_resource *res) 145{ 146 147 if (metadata_read(res, true) < 0) 148 exit(EX_NOINPUT); 149} 150 151static void 152init_remote(struct hast_resource *res, struct nv *nvin) 153{ 154 uint64_t resuid; 155 struct nv *nvout; 156 unsigned char *map; 157 size_t mapsize; 158 159 map = NULL; 160 mapsize = 0; 161 nvout = nv_alloc(); 162 nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 163 nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 164 resuid = nv_get_uint64(nvin, "resuid"); 165 res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 166 res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 167 nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 168 nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 169 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 170 METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 171 map = malloc(mapsize); 172 if (map == NULL) { 173 pjdlog_exitx(EX_TEMPFAIL, 174 "Unable to allocate memory (%zu bytes) for activemap.", 175 mapsize); 176 } 177 nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 178 /* 179 * When we work as primary and secondary is missing we will increase 180 * localcnt in our metadata. When secondary is connected and synced 181 * we make localcnt be equal to remotecnt, which means nodes are more 182 * or less in sync. 183 * Split-brain condition is when both nodes are not able to communicate 184 * and are both configured as primary nodes. In turn, they can both 185 * make incompatible changes to the data and we have to detect that. 186 * Under split-brain condition we will increase our localcnt on first 187 * write and remote node will increase its localcnt on first write. 188 * When we connect we can see that primary's localcnt is greater than 189 * our remotecnt (primary was modified while we weren't watching) and 190 * our localcnt is greater than primary's remotecnt (we were modified 191 * while primary wasn't watching). 192 * There are many possible combinations which are all gathered below. 193 * Don't pay too much attention to exact numbers, the more important 194 * is to compare them. We compare secondary's local with primary's 195 * remote and secondary's remote with primary's local. 196 * Note that every case where primary's localcnt is smaller than 197 * secondary's remotecnt and where secondary's localcnt is smaller than 198 * primary's remotecnt should be impossible in practise. We will perform 199 * full synchronization then. Those cases are marked with an asterisk. 200 * Regular synchronization means that only extents marked as dirty are 201 * synchronized (regular synchronization). 202 * 203 * SECONDARY METADATA PRIMARY METADATA 204 * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 205 * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 206 * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 207 * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 208 * regular sync from secondary. 209 * local=3 remote=3 local=3 remote=3 Regular sync just in case. 210 * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 211 * local=3 remote=3 local=4 remote=2 Split-brain condition. 212 * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 213 * regular sync from primary. 214 * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 215 */ 216 if (res->hr_resuid == 0) { 217 /* 218 * Provider is used for the first time. Initialize everything. 219 */ 220 assert(res->hr_secondary_localcnt == 0); 221 res->hr_resuid = resuid; 222 if (metadata_write(res) < 0) 223 exit(EX_NOINPUT); 224 memset(map, 0xff, mapsize); 225 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 226 } else if ( 227 /* Is primary is out-of-date? */ 228 (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 229 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 230 /* Node are more or less in sync? */ 231 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 232 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 233 /* Is secondary is out-of-date? */ 234 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 235 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 236 /* 237 * Nodes are more or less in sync or one of the nodes is 238 * out-of-date. 239 * It doesn't matter at this point which one, we just have to 240 * send out local bitmap to the remote node. 241 */ 242 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 243 (ssize_t)mapsize) { 244 pjdlog_exit(LOG_ERR, "Unable to read activemap"); 245 } 246 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 247 res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 248 /* Primary is out-of-date, sync from secondary. */ 249 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 250 } else { 251 /* 252 * Secondary is out-of-date or counts match. 253 * Sync from primary. 254 */ 255 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 256 } 257 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 258 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 259 /* 260 * Not good, we have split-brain condition. 261 */ 262 pjdlog_error("Split-brain detected, exiting."); 263 nv_add_string(nvout, "Split-brain condition!", "errmsg"); 264 free(map); 265 map = NULL; 266 mapsize = 0; 267 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 268 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 269 /* 270 * This should never happen in practise, but we will perform 271 * full synchronization. 272 */ 273 assert(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 274 res->hr_primary_localcnt < res->hr_secondary_remotecnt); 275 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 276 METADATA_SIZE, res->hr_extentsize, 277 res->hr_local_sectorsize); 278 memset(map, 0xff, mapsize); 279 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 280 /* In this one of five cases sync from secondary. */ 281 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 282 } else { 283 /* For the rest four cases sync from primary. */ 284 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 285 } 286 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 287 (uintmax_t)res->hr_primary_localcnt, 288 (uintmax_t)res->hr_primary_remotecnt, 289 (uintmax_t)res->hr_secondary_localcnt, 290 (uintmax_t)res->hr_secondary_remotecnt); 291 } 292 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 293 pjdlog_errno(LOG_WARNING, "Unable to send activemap to %s", 294 res->hr_remoteaddr); 295 nv_free(nvout); 296 exit(EX_TEMPFAIL); 297 } 298 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 299 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 300 /* Exit on split-brain. */ 301 exit(EX_CONFIG); 302 } 303} 304 305void 306hastd_secondary(struct hast_resource *res, struct nv *nvin) 307{ 308 pthread_t td; 309 pid_t pid; 310 int error; 311 312 /* 313 * Create communication channel between parent and child. 314 */ 315 if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 316 KEEP_ERRNO((void)pidfile_remove(pfh)); 317 pjdlog_exit(EX_OSERR, 318 "Unable to create control sockets between parent and child"); 319 } 320 321 pid = fork(); 322 if (pid < 0) { 323 KEEP_ERRNO((void)pidfile_remove(pfh)); 324 pjdlog_exit(EX_OSERR, "Unable to fork"); 325 } 326 327 if (pid > 0) { 328 /* This is parent. */ 329 proto_close(res->hr_remotein); 330 res->hr_remotein = NULL; 331 proto_close(res->hr_remoteout); 332 res->hr_remoteout = NULL; 333 res->hr_workerpid = pid; 334 return; 335 } 336 (void)pidfile_close(pfh); 337 338 setproctitle("%s (secondary)", res->hr_name); 339 340 /* Error in setting timeout is not critical, but why should it fail? */ 341 if (proto_timeout(res->hr_remotein, 0) < 0) 342 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 343 if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 344 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 345 346 init_local(res); 347 init_remote(res, nvin); 348 init_environment(); 349 350 error = pthread_create(&td, NULL, recv_thread, res); 351 assert(error == 0); 352 error = pthread_create(&td, NULL, disk_thread, res); 353 assert(error == 0); 354 error = pthread_create(&td, NULL, send_thread, res); 355 assert(error == 0); 356 (void)ctrl_thread(res); 357} 358 359static void 360reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 361{ 362 char msg[1024]; 363 va_list ap; 364 int len; 365 366 va_start(ap, fmt); 367 len = vsnprintf(msg, sizeof(msg), fmt, ap); 368 va_end(ap); 369 if ((size_t)len < sizeof(msg)) { 370 switch (hio->hio_cmd) { 371 case HIO_READ: 372 (void)snprintf(msg + len, sizeof(msg) - len, 373 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 374 (uintmax_t)hio->hio_length); 375 break; 376 case HIO_DELETE: 377 (void)snprintf(msg + len, sizeof(msg) - len, 378 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 379 (uintmax_t)hio->hio_length); 380 break; 381 case HIO_FLUSH: 382 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 383 break; 384 case HIO_WRITE: 385 (void)snprintf(msg + len, sizeof(msg) - len, 386 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 387 (uintmax_t)hio->hio_length); 388 break; 389 default: 390 (void)snprintf(msg + len, sizeof(msg) - len, 391 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 392 break; 393 } 394 } 395 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 396} 397 398static int 399requnpack(struct hast_resource *res, struct hio *hio) 400{ 401 402 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 403 if (hio->hio_cmd == 0) { 404 pjdlog_error("Header contains no 'cmd' field."); 405 hio->hio_error = EINVAL; 406 goto end; 407 } 408 switch (hio->hio_cmd) { 409 case HIO_READ: 410 case HIO_WRITE: 411 case HIO_DELETE: 412 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 413 if (nv_error(hio->hio_nv) != 0) { 414 pjdlog_error("Header is missing 'offset' field."); 415 hio->hio_error = EINVAL; 416 goto end; 417 } 418 hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 419 if (nv_error(hio->hio_nv) != 0) { 420 pjdlog_error("Header is missing 'length' field."); 421 hio->hio_error = EINVAL; 422 goto end; 423 } 424 if (hio->hio_length == 0) { 425 pjdlog_error("Data length is zero."); 426 hio->hio_error = EINVAL; 427 goto end; 428 } 429 if (hio->hio_length > MAXPHYS) { 430 pjdlog_error("Data length is too large (%ju > %ju).", 431 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 432 hio->hio_error = EINVAL; 433 goto end; 434 } 435 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 436 pjdlog_error("Offset %ju is not multiple of sector size.", 437 (uintmax_t)hio->hio_offset); 438 hio->hio_error = EINVAL; 439 goto end; 440 } 441 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 442 pjdlog_error("Length %ju is not multiple of sector size.", 443 (uintmax_t)hio->hio_length); 444 hio->hio_error = EINVAL; 445 goto end; 446 } 447 if (hio->hio_offset + hio->hio_length > 448 (uint64_t)res->hr_datasize) { 449 pjdlog_error("Data offset is too large (%ju > %ju).", 450 (uintmax_t)(hio->hio_offset + hio->hio_length), 451 (uintmax_t)res->hr_datasize); 452 hio->hio_error = EINVAL; 453 goto end; 454 } 455 break; 456 default: 457 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 458 hio->hio_cmd); 459 hio->hio_error = EINVAL; 460 goto end; 461 } 462 hio->hio_error = 0; 463end: 464 return (hio->hio_error); 465} 466 467/* 468 * Thread receives requests from the primary node. 469 */ 470static void * 471recv_thread(void *arg) 472{ 473 struct hast_resource *res = arg; 474 struct hio *hio; 475 bool wakeup; 476 477 for (;;) { 478 pjdlog_debug(2, "recv: Taking free request."); 479 mtx_lock(&hio_free_list_lock); 480 while ((hio = TAILQ_FIRST(&hio_free_list)) == NULL) { 481 pjdlog_debug(2, "recv: No free requests, waiting."); 482 cv_wait(&hio_free_list_cond, &hio_free_list_lock); 483 } 484 TAILQ_REMOVE(&hio_free_list, hio, hio_next); 485 mtx_unlock(&hio_free_list_lock); 486 pjdlog_debug(2, "recv: (%p) Got request.", hio); 487 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 488 pjdlog_exit(EX_TEMPFAIL, 489 "Unable to receive request header"); 490 } 491 if (requnpack(res, hio) != 0) 492 goto send_queue; 493 reqlog(LOG_DEBUG, 2, -1, hio, 494 "recv: (%p) Got request header: ", hio); 495 if (hio->hio_cmd == HIO_WRITE) { 496 if (hast_proto_recv_data(res, res->hr_remotein, 497 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 498 pjdlog_exit(EX_TEMPFAIL, 499 "Unable to receive reply data"); 500 } 501 } 502 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 503 hio); 504 mtx_lock(&hio_disk_list_lock); 505 wakeup = TAILQ_EMPTY(&hio_disk_list); 506 TAILQ_INSERT_TAIL(&hio_disk_list, hio, hio_next); 507 mtx_unlock(&hio_disk_list_lock); 508 if (wakeup) 509 cv_signal(&hio_disk_list_cond); 510 continue; 511send_queue: 512 pjdlog_debug(2, "recv: (%p) Moving request to the send queue.", 513 hio); 514 mtx_lock(&hio_send_list_lock); 515 wakeup = TAILQ_EMPTY(&hio_send_list); 516 TAILQ_INSERT_TAIL(&hio_send_list, hio, hio_next); 517 mtx_unlock(&hio_send_list_lock); 518 if (wakeup) 519 cv_signal(&hio_send_list_cond); 520 } 521 /* NOTREACHED */ 522 return (NULL); 523} 524 525/* 526 * Thread reads from or writes to local component and also handles DELETE and 527 * FLUSH requests. 528 */ 529static void * 530disk_thread(void *arg) 531{ 532 struct hast_resource *res = arg; 533 struct hio *hio; 534 ssize_t ret; 535 bool clear_activemap, wakeup; 536 537 clear_activemap = true; 538 539 for (;;) { 540 pjdlog_debug(2, "disk: Taking request."); 541 mtx_lock(&hio_disk_list_lock); 542 while ((hio = TAILQ_FIRST(&hio_disk_list)) == NULL) { 543 pjdlog_debug(2, "disk: No requests, waiting."); 544 cv_wait(&hio_disk_list_cond, &hio_disk_list_lock); 545 } 546 TAILQ_REMOVE(&hio_disk_list, hio, hio_next); 547 mtx_unlock(&hio_disk_list_lock); 548 while (clear_activemap) { 549 unsigned char *map; 550 size_t mapsize; 551 552 /* 553 * When first request is received, it means that primary 554 * already received our activemap, merged it and stored 555 * locally. We can now safely clear our activemap. 556 */ 557 mapsize = 558 activemap_calc_ondisk_size(res->hr_local_mediasize - 559 METADATA_SIZE, res->hr_extentsize, 560 res->hr_local_sectorsize); 561 map = calloc(1, mapsize); 562 if (map == NULL) { 563 pjdlog_warning("Unable to allocate memory to clear local activemap."); 564 break; 565 } 566 if (pwrite(res->hr_localfd, map, mapsize, 567 METADATA_SIZE) != (ssize_t)mapsize) { 568 pjdlog_errno(LOG_WARNING, 569 "Unable to store cleared activemap"); 570 free(map); 571 break; 572 } 573 free(map); 574 clear_activemap = false; 575 pjdlog_debug(1, "Local activemap cleared."); 576 } 577 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 578 /* Handle the actual request. */ 579 switch (hio->hio_cmd) { 580 case HIO_READ: 581 ret = pread(res->hr_localfd, hio->hio_data, 582 hio->hio_length, 583 hio->hio_offset + res->hr_localoff); 584 if (ret < 0) 585 hio->hio_error = errno; 586 else if (ret != (int64_t)hio->hio_length) 587 hio->hio_error = EIO; 588 else 589 hio->hio_error = 0; 590 break; 591 case HIO_WRITE: 592 ret = pwrite(res->hr_localfd, hio->hio_data, 593 hio->hio_length, 594 hio->hio_offset + res->hr_localoff); 595 if (ret < 0) 596 hio->hio_error = errno; 597 else if (ret != (int64_t)hio->hio_length) 598 hio->hio_error = EIO; 599 else 600 hio->hio_error = 0; 601 break; 602 case HIO_DELETE: 603 ret = g_delete(res->hr_localfd, 604 hio->hio_offset + res->hr_localoff, 605 hio->hio_length); 606 if (ret < 0) 607 hio->hio_error = errno; 608 else 609 hio->hio_error = 0; 610 break; 611 case HIO_FLUSH: 612 ret = g_flush(res->hr_localfd); 613 if (ret < 0) 614 hio->hio_error = errno; 615 else 616 hio->hio_error = 0; 617 break; 618 } 619 if (hio->hio_error != 0) { 620 reqlog(LOG_ERR, 0, hio->hio_error, hio, 621 "Request failed: "); 622 } 623 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 624 hio); 625 mtx_lock(&hio_send_list_lock); 626 wakeup = TAILQ_EMPTY(&hio_send_list); 627 TAILQ_INSERT_TAIL(&hio_send_list, hio, hio_next); 628 mtx_unlock(&hio_send_list_lock); 629 if (wakeup) 630 cv_signal(&hio_send_list_cond); 631 } 632 /* NOTREACHED */ 633 return (NULL); 634} 635 636/* 637 * Thread sends requests back to primary node. 638 */ 639static void * 640send_thread(void *arg) 641{ 642 struct hast_resource *res = arg; 643 struct nv *nvout; 644 struct hio *hio; 645 void *data; 646 size_t length; 647 bool wakeup; 648 649 for (;;) { 650 pjdlog_debug(2, "send: Taking request."); 651 mtx_lock(&hio_send_list_lock); 652 while ((hio = TAILQ_FIRST(&hio_send_list)) == NULL) { 653 pjdlog_debug(2, "send: No requests, waiting."); 654 cv_wait(&hio_send_list_cond, &hio_send_list_lock); 655 } 656 TAILQ_REMOVE(&hio_send_list, hio, hio_next); 657 mtx_unlock(&hio_send_list_lock); 658 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 659 nvout = nv_alloc(); 660 /* Copy sequence number. */ 661 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 662 switch (hio->hio_cmd) { 663 case HIO_READ: 664 if (hio->hio_error == 0) { 665 data = hio->hio_data; 666 length = hio->hio_length; 667 break; 668 } 669 /* 670 * We send no data in case of an error. 671 */ 672 /* FALLTHROUGH */ 673 case HIO_DELETE: 674 case HIO_FLUSH: 675 case HIO_WRITE: 676 data = NULL; 677 length = 0; 678 break; 679 default: 680 abort(); 681 break; 682 } 683 if (hio->hio_error != 0) 684 nv_add_int16(nvout, hio->hio_error, "error"); 685 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 686 length) < 0) { 687 pjdlog_exit(EX_TEMPFAIL, "Unable to send reply."); 688 } 689 nv_free(nvout); 690 pjdlog_debug(2, "disk: (%p) Moving request to the free queue.", 691 hio); 692 nv_free(hio->hio_nv); 693 hio->hio_error = 0; 694 mtx_lock(&hio_free_list_lock); 695 wakeup = TAILQ_EMPTY(&hio_free_list); 696 TAILQ_INSERT_TAIL(&hio_free_list, hio, hio_next); 697 mtx_unlock(&hio_free_list_lock); 698 if (wakeup) 699 cv_signal(&hio_free_list_cond); 700 } 701 /* NOTREACHED */ 702 return (NULL); 703} 704