1/*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2009-2010 The FreeBSD Foundation 5 * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 6 * All rights reserved. 7 * 8 * This software was developed by Pawel Jakub Dawidek under sponsorship from 9 * the FreeBSD Foundation. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 */ 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 <err.h> 40#include <errno.h> 41#include <fcntl.h> 42#include <libgeom.h> 43#include <pthread.h> 44#include <signal.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 "event.h" 57#include "hast.h" 58#include "hast_proto.h" 59#include "hastd.h" 60#include "hooks.h" 61#include "metadata.h" 62#include "proto.h" 63#include "subr.h" 64#include "synch.h" 65 66struct hio { 67 uint64_t hio_seq; 68 int hio_error; 69 void *hio_data; 70 uint8_t hio_cmd; 71 uint64_t hio_offset; 72 uint64_t hio_length; 73 bool hio_memsync; 74 TAILQ_ENTRY(hio) hio_next; 75}; 76 77static struct hast_resource *gres; 78 79/* 80 * Free list holds unused structures. When free list is empty, we have to wait 81 * until some in-progress requests are freed. 82 */ 83static TAILQ_HEAD(, hio) hio_free_list; 84static size_t hio_free_list_size; 85static pthread_mutex_t hio_free_list_lock; 86static pthread_cond_t hio_free_list_cond; 87/* 88 * Disk thread (the one that does I/O requests) takes requests from this list. 89 */ 90static TAILQ_HEAD(, hio) hio_disk_list; 91static size_t hio_disk_list_size; 92static pthread_mutex_t hio_disk_list_lock; 93static pthread_cond_t hio_disk_list_cond; 94/* 95 * Thread that sends requests back to primary takes requests from this list. 96 */ 97static TAILQ_HEAD(, hio) hio_send_list; 98static size_t hio_send_list_size; 99static pthread_mutex_t hio_send_list_lock; 100static pthread_cond_t hio_send_list_cond; 101 102/* 103 * Maximum number of outstanding I/O requests. 104 */ 105#define HAST_HIO_MAX 256 106 107static void *recv_thread(void *arg); 108static void *disk_thread(void *arg); 109static void *send_thread(void *arg); 110 111#define QUEUE_INSERT(name, hio) do { \ 112 mtx_lock(&hio_##name##_list_lock); \ 113 if (TAILQ_EMPTY(&hio_##name##_list)) \ 114 cv_broadcast(&hio_##name##_list_cond); \ 115 TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \ 116 hio_##name##_list_size++; \ 117 mtx_unlock(&hio_##name##_list_lock); \ 118} while (0) 119#define QUEUE_TAKE(name, hio) do { \ 120 mtx_lock(&hio_##name##_list_lock); \ 121 while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 122 cv_wait(&hio_##name##_list_cond, \ 123 &hio_##name##_list_lock); \ 124 } \ 125 PJDLOG_ASSERT(hio_##name##_list_size != 0); \ 126 hio_##name##_list_size--; \ 127 TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \ 128 mtx_unlock(&hio_##name##_list_lock); \ 129} while (0) 130 131static void 132output_status_aux(struct nv *nvout) 133{ 134 135 nv_add_uint64(nvout, (uint64_t)hio_free_list_size, "idle_queue_size"); 136 nv_add_uint64(nvout, (uint64_t)hio_disk_list_size, "local_queue_size"); 137 nv_add_uint64(nvout, (uint64_t)hio_send_list_size, "send_queue_size"); 138} 139 140static void 141hio_clear(struct hio *hio) 142{ 143 144 hio->hio_seq = 0; 145 hio->hio_error = 0; 146 hio->hio_cmd = HIO_UNDEF; 147 hio->hio_offset = 0; 148 hio->hio_length = 0; 149 hio->hio_memsync = false; 150} 151 152static void 153hio_copy(const struct hio *srchio, struct hio *dsthio) 154{ 155 156 /* 157 * We don't copy hio_error, hio_data and hio_next fields. 158 */ 159 160 dsthio->hio_seq = srchio->hio_seq; 161 dsthio->hio_cmd = srchio->hio_cmd; 162 dsthio->hio_offset = srchio->hio_offset; 163 dsthio->hio_length = srchio->hio_length; 164 dsthio->hio_memsync = srchio->hio_memsync; 165} 166 167static void 168init_environment(void) 169{ 170 struct hio *hio; 171 unsigned int ii; 172 173 /* 174 * Initialize lists, their locks and theirs condition variables. 175 */ 176 TAILQ_INIT(&hio_free_list); 177 mtx_init(&hio_free_list_lock); 178 cv_init(&hio_free_list_cond); 179 TAILQ_INIT(&hio_disk_list); 180 mtx_init(&hio_disk_list_lock); 181 cv_init(&hio_disk_list_cond); 182 TAILQ_INIT(&hio_send_list); 183 mtx_init(&hio_send_list_lock); 184 cv_init(&hio_send_list_cond); 185 186 /* 187 * Allocate requests pool and initialize requests. 188 */ 189 for (ii = 0; ii < HAST_HIO_MAX; ii++) { 190 hio = malloc(sizeof(*hio)); 191 if (hio == NULL) { 192 pjdlog_exitx(EX_TEMPFAIL, 193 "Unable to allocate memory (%zu bytes) for hio request.", 194 sizeof(*hio)); 195 } 196 hio->hio_data = malloc(MAXPHYS); 197 if (hio->hio_data == NULL) { 198 pjdlog_exitx(EX_TEMPFAIL, 199 "Unable to allocate memory (%zu bytes) for gctl_data.", 200 (size_t)MAXPHYS); 201 } 202 hio_clear(hio); 203 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 204 hio_free_list_size++; 205 } 206} 207 208static void 209init_local(struct hast_resource *res) 210{ 211 212 if (metadata_read(res, true) == -1) 213 exit(EX_NOINPUT); 214} 215 216static void 217init_remote(struct hast_resource *res, struct nv *nvin) 218{ 219 uint64_t resuid; 220 struct nv *nvout; 221 unsigned char *map; 222 size_t mapsize; 223 224#ifdef notyet 225 /* Setup direction. */ 226 if (proto_send(res->hr_remoteout, NULL, 0) == -1) 227 pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 228#endif 229 230 nvout = nv_alloc(); 231 nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 232 nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 233 resuid = nv_get_uint64(nvin, "resuid"); 234 res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 235 res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 236 nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 237 nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 238 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 239 METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 240 map = malloc(mapsize); 241 if (map == NULL) { 242 pjdlog_exitx(EX_TEMPFAIL, 243 "Unable to allocate memory (%zu bytes) for activemap.", 244 mapsize); 245 } 246 /* 247 * When we work as primary and secondary is missing we will increase 248 * localcnt in our metadata. When secondary is connected and synced 249 * we make localcnt be equal to remotecnt, which means nodes are more 250 * or less in sync. 251 * Split-brain condition is when both nodes are not able to communicate 252 * and are both configured as primary nodes. In turn, they can both 253 * make incompatible changes to the data and we have to detect that. 254 * Under split-brain condition we will increase our localcnt on first 255 * write and remote node will increase its localcnt on first write. 256 * When we connect we can see that primary's localcnt is greater than 257 * our remotecnt (primary was modified while we weren't watching) and 258 * our localcnt is greater than primary's remotecnt (we were modified 259 * while primary wasn't watching). 260 * There are many possible combinations which are all gathered below. 261 * Don't pay too much attention to exact numbers, the more important 262 * is to compare them. We compare secondary's local with primary's 263 * remote and secondary's remote with primary's local. 264 * Note that every case where primary's localcnt is smaller than 265 * secondary's remotecnt and where secondary's localcnt is smaller than 266 * primary's remotecnt should be impossible in practise. We will perform 267 * full synchronization then. Those cases are marked with an asterisk. 268 * Regular synchronization means that only extents marked as dirty are 269 * synchronized (regular synchronization). 270 * 271 * SECONDARY METADATA PRIMARY METADATA 272 * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 273 * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 274 * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 275 * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 276 * regular sync from secondary. 277 * local=3 remote=3 local=3 remote=3 Regular sync just in case. 278 * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 279 * local=3 remote=3 local=4 remote=2 Split-brain condition. 280 * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 281 * regular sync from primary. 282 * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 283 */ 284 if (res->hr_resuid == 0) { 285 /* 286 * Provider is used for the first time. If primary node done no 287 * writes yet as well (we will find "virgin" argument) then 288 * there is no need to synchronize anything. If primary node 289 * done any writes already we have to synchronize everything. 290 */ 291 PJDLOG_ASSERT(res->hr_secondary_localcnt == 0); 292 res->hr_resuid = resuid; 293 if (metadata_write(res) == -1) 294 exit(EX_NOINPUT); 295 if (nv_exists(nvin, "virgin")) { 296 free(map); 297 map = NULL; 298 mapsize = 0; 299 } else { 300 memset(map, 0xff, mapsize); 301 } 302 nv_add_int8(nvout, 1, "virgin"); 303 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 304 } else if (res->hr_resuid != resuid) { 305 char errmsg[256]; 306 307 free(map); 308 (void)snprintf(errmsg, sizeof(errmsg), 309 "Resource unique ID mismatch (primary=%ju, secondary=%ju).", 310 (uintmax_t)resuid, (uintmax_t)res->hr_resuid); 311 pjdlog_error("%s", errmsg); 312 nv_add_string(nvout, errmsg, "errmsg"); 313 if (hast_proto_send(res, res->hr_remotein, nvout, 314 NULL, 0) == -1) { 315 pjdlog_exit(EX_TEMPFAIL, 316 "Unable to send response to %s", 317 res->hr_remoteaddr); 318 } 319 nv_free(nvout); 320 exit(EX_CONFIG); 321 } else if ( 322 /* Is primary out-of-date? */ 323 (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 324 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 325 /* Are the nodes more or less in sync? */ 326 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 327 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 328 /* Is secondary out-of-date? */ 329 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 330 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 331 /* 332 * Nodes are more or less in sync or one of the nodes is 333 * out-of-date. 334 * It doesn't matter at this point which one, we just have to 335 * send out local bitmap to the remote node. 336 */ 337 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 338 (ssize_t)mapsize) { 339 pjdlog_exit(LOG_ERR, "Unable to read activemap"); 340 } 341 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 342 res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 343 /* Primary is out-of-date, sync from secondary. */ 344 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 345 } else { 346 /* 347 * Secondary is out-of-date or counts match. 348 * Sync from primary. 349 */ 350 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 351 } 352 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 353 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 354 /* 355 * Not good, we have split-brain condition. 356 */ 357 free(map); 358 pjdlog_error("Split-brain detected, exiting."); 359 nv_add_string(nvout, "Split-brain condition!", "errmsg"); 360 if (hast_proto_send(res, res->hr_remotein, nvout, 361 NULL, 0) == -1) { 362 pjdlog_exit(EX_TEMPFAIL, 363 "Unable to send response to %s", 364 res->hr_remoteaddr); 365 } 366 nv_free(nvout); 367 /* Exit on split-brain. */ 368 event_send(res, EVENT_SPLITBRAIN); 369 exit(EX_CONFIG); 370 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 371 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 372 /* 373 * This should never happen in practise, but we will perform 374 * full synchronization. 375 */ 376 PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 377 res->hr_primary_localcnt < res->hr_secondary_remotecnt); 378 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 379 METADATA_SIZE, res->hr_extentsize, 380 res->hr_local_sectorsize); 381 memset(map, 0xff, mapsize); 382 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 383 /* In this one of five cases sync from secondary. */ 384 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 385 } else { 386 /* For the rest four cases sync from primary. */ 387 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 388 } 389 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 390 (uintmax_t)res->hr_primary_localcnt, 391 (uintmax_t)res->hr_primary_remotecnt, 392 (uintmax_t)res->hr_secondary_localcnt, 393 (uintmax_t)res->hr_secondary_remotecnt); 394 } 395 nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 396 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) { 397 pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 398 res->hr_remoteaddr); 399 } 400 if (map != NULL) 401 free(map); 402 nv_free(nvout); 403#ifdef notyet 404 /* Setup direction. */ 405 if (proto_recv(res->hr_remotein, NULL, 0) == -1) 406 pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 407#endif 408} 409 410void 411hastd_secondary(struct hast_resource *res, struct nv *nvin) 412{ 413 sigset_t mask; 414 pthread_t td; 415 pid_t pid; 416 int error, mode, debuglevel; 417 418 /* 419 * Create communication channel between parent and child. 420 */ 421 if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) { 422 KEEP_ERRNO((void)pidfile_remove(pfh)); 423 pjdlog_exit(EX_OSERR, 424 "Unable to create control sockets between parent and child"); 425 } 426 /* 427 * Create communication channel between child and parent. 428 */ 429 if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) { 430 KEEP_ERRNO((void)pidfile_remove(pfh)); 431 pjdlog_exit(EX_OSERR, 432 "Unable to create event sockets between child and parent"); 433 } 434 435 pid = fork(); 436 if (pid == -1) { 437 KEEP_ERRNO((void)pidfile_remove(pfh)); 438 pjdlog_exit(EX_OSERR, "Unable to fork"); 439 } 440 441 if (pid > 0) { 442 /* This is parent. */ 443 proto_close(res->hr_remotein); 444 res->hr_remotein = NULL; 445 proto_close(res->hr_remoteout); 446 res->hr_remoteout = NULL; 447 /* Declare that we are receiver. */ 448 proto_recv(res->hr_event, NULL, 0); 449 /* Declare that we are sender. */ 450 proto_send(res->hr_ctrl, NULL, 0); 451 res->hr_workerpid = pid; 452 return; 453 } 454 455 gres = res; 456 res->output_status_aux = output_status_aux; 457 mode = pjdlog_mode_get(); 458 debuglevel = pjdlog_debug_get(); 459 460 /* Declare that we are sender. */ 461 proto_send(res->hr_event, NULL, 0); 462 /* Declare that we are receiver. */ 463 proto_recv(res->hr_ctrl, NULL, 0); 464 descriptors_cleanup(res); 465 466 descriptors_assert(res, mode); 467 468 pjdlog_init(mode); 469 pjdlog_debug_set(debuglevel); 470 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 471 setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role)); 472 473 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 474 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 475 476 /* Error in setting timeout is not critical, but why should it fail? */ 477 if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1) 478 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 479 if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1) 480 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 481 482 init_local(res); 483 init_environment(); 484 485 if (drop_privs(res) != 0) 486 exit(EX_CONFIG); 487 pjdlog_info("Privileges successfully dropped."); 488 489 /* 490 * Create the control thread before sending any event to the parent, 491 * as we can deadlock when parent sends control request to worker, 492 * but worker has no control thread started yet, so parent waits. 493 * In the meantime worker sends an event to the parent, but parent 494 * is unable to handle the event, because it waits for control 495 * request response. 496 */ 497 error = pthread_create(&td, NULL, ctrl_thread, res); 498 PJDLOG_ASSERT(error == 0); 499 500 init_remote(res, nvin); 501 event_send(res, EVENT_CONNECT); 502 503 error = pthread_create(&td, NULL, recv_thread, res); 504 PJDLOG_ASSERT(error == 0); 505 error = pthread_create(&td, NULL, disk_thread, res); 506 PJDLOG_ASSERT(error == 0); 507 (void)send_thread(res); 508} 509 510static void 511reqlog(int loglevel, int debuglevel, int error, struct hio *hio, 512 const char *fmt, ...) 513{ 514 char msg[1024]; 515 va_list ap; 516 int len; 517 518 va_start(ap, fmt); 519 len = vsnprintf(msg, sizeof(msg), fmt, ap); 520 va_end(ap); 521 if ((size_t)len < sizeof(msg)) { 522 switch (hio->hio_cmd) { 523 case HIO_READ: 524 (void)snprintf(msg + len, sizeof(msg) - len, 525 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 526 (uintmax_t)hio->hio_length); 527 break; 528 case HIO_DELETE: 529 (void)snprintf(msg + len, sizeof(msg) - len, 530 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 531 (uintmax_t)hio->hio_length); 532 break; 533 case HIO_FLUSH: 534 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 535 break; 536 case HIO_WRITE: 537 (void)snprintf(msg + len, sizeof(msg) - len, 538 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 539 (uintmax_t)hio->hio_length); 540 break; 541 case HIO_KEEPALIVE: 542 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 543 break; 544 default: 545 (void)snprintf(msg + len, sizeof(msg) - len, 546 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 547 break; 548 } 549 } 550 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 551} 552 553static int 554requnpack(struct hast_resource *res, struct hio *hio, struct nv *nv) 555{ 556 557 hio->hio_cmd = nv_get_uint8(nv, "cmd"); 558 if (hio->hio_cmd == 0) { 559 pjdlog_error("Header contains no 'cmd' field."); 560 hio->hio_error = EINVAL; 561 goto end; 562 } 563 if (hio->hio_cmd != HIO_KEEPALIVE) { 564 hio->hio_seq = nv_get_uint64(nv, "seq"); 565 if (hio->hio_seq == 0) { 566 pjdlog_error("Header contains no 'seq' field."); 567 hio->hio_error = EINVAL; 568 goto end; 569 } 570 } 571 switch (hio->hio_cmd) { 572 case HIO_FLUSH: 573 case HIO_KEEPALIVE: 574 break; 575 case HIO_WRITE: 576 hio->hio_memsync = nv_exists(nv, "memsync"); 577 /* FALLTHROUGH */ 578 case HIO_READ: 579 case HIO_DELETE: 580 hio->hio_offset = nv_get_uint64(nv, "offset"); 581 if (nv_error(nv) != 0) { 582 pjdlog_error("Header is missing 'offset' field."); 583 hio->hio_error = EINVAL; 584 goto end; 585 } 586 hio->hio_length = nv_get_uint64(nv, "length"); 587 if (nv_error(nv) != 0) { 588 pjdlog_error("Header is missing 'length' field."); 589 hio->hio_error = EINVAL; 590 goto end; 591 } 592 if (hio->hio_length == 0) { 593 pjdlog_error("Data length is zero."); 594 hio->hio_error = EINVAL; 595 goto end; 596 } 597 if (hio->hio_cmd != HIO_DELETE && hio->hio_length > MAXPHYS) { 598 pjdlog_error("Data length is too large (%ju > %ju).", 599 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 600 hio->hio_error = EINVAL; 601 goto end; 602 } 603 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 604 pjdlog_error("Offset %ju is not multiple of sector size.", 605 (uintmax_t)hio->hio_offset); 606 hio->hio_error = EINVAL; 607 goto end; 608 } 609 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 610 pjdlog_error("Length %ju is not multiple of sector size.", 611 (uintmax_t)hio->hio_length); 612 hio->hio_error = EINVAL; 613 goto end; 614 } 615 if (hio->hio_offset + hio->hio_length > 616 (uint64_t)res->hr_datasize) { 617 pjdlog_error("Data offset is too large (%ju > %ju).", 618 (uintmax_t)(hio->hio_offset + hio->hio_length), 619 (uintmax_t)res->hr_datasize); 620 hio->hio_error = EINVAL; 621 goto end; 622 } 623 break; 624 default: 625 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 626 hio->hio_cmd); 627 hio->hio_error = EINVAL; 628 goto end; 629 } 630 hio->hio_error = 0; 631end: 632 return (hio->hio_error); 633} 634 635static __dead2 void 636secondary_exit(int exitcode, const char *fmt, ...) 637{ 638 va_list ap; 639 640 PJDLOG_ASSERT(exitcode != EX_OK); 641 va_start(ap, fmt); 642 pjdlogv_errno(LOG_ERR, fmt, ap); 643 va_end(ap); 644 event_send(gres, EVENT_DISCONNECT); 645 exit(exitcode); 646} 647 648/* 649 * Thread receives requests from the primary node. 650 */ 651static void * 652recv_thread(void *arg) 653{ 654 struct hast_resource *res = arg; 655 struct hio *hio, *mshio; 656 struct nv *nv; 657 658 for (;;) { 659 pjdlog_debug(2, "recv: Taking free request."); 660 QUEUE_TAKE(free, hio); 661 pjdlog_debug(2, "recv: (%p) Got request.", hio); 662 if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) { 663 secondary_exit(EX_TEMPFAIL, 664 "Unable to receive request header"); 665 } 666 if (requnpack(res, hio, nv) != 0) { 667 nv_free(nv); 668 pjdlog_debug(2, 669 "recv: (%p) Moving request to the send queue.", 670 hio); 671 QUEUE_INSERT(send, hio); 672 continue; 673 } 674 switch (hio->hio_cmd) { 675 case HIO_READ: 676 res->hr_stat_read++; 677 break; 678 case HIO_WRITE: 679 res->hr_stat_write++; 680 break; 681 case HIO_DELETE: 682 res->hr_stat_delete++; 683 break; 684 case HIO_FLUSH: 685 res->hr_stat_flush++; 686 break; 687 case HIO_KEEPALIVE: 688 break; 689 default: 690 PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 691 hio->hio_cmd); 692 } 693 reqlog(LOG_DEBUG, 2, -1, hio, 694 "recv: (%p) Got request header: ", hio); 695 if (hio->hio_cmd == HIO_KEEPALIVE) { 696 nv_free(nv); 697 pjdlog_debug(2, 698 "recv: (%p) Moving request to the free queue.", 699 hio); 700 hio_clear(hio); 701 QUEUE_INSERT(free, hio); 702 continue; 703 } else if (hio->hio_cmd == HIO_WRITE) { 704 if (hast_proto_recv_data(res, res->hr_remotein, nv, 705 hio->hio_data, MAXPHYS) == -1) { 706 secondary_exit(EX_TEMPFAIL, 707 "Unable to receive request data"); 708 } 709 if (hio->hio_memsync) { 710 /* 711 * For memsync requests we expect two replies. 712 * Clone the hio so we can handle both of them. 713 */ 714 pjdlog_debug(2, "recv: Taking free request."); 715 QUEUE_TAKE(free, mshio); 716 pjdlog_debug(2, "recv: (%p) Got request.", 717 mshio); 718 hio_copy(hio, mshio); 719 mshio->hio_error = 0; 720 /* 721 * We want to keep 'memsync' tag only on the 722 * request going onto send queue (mshio). 723 */ 724 hio->hio_memsync = false; 725 pjdlog_debug(2, 726 "recv: (%p) Moving memsync request to the send queue.", 727 mshio); 728 QUEUE_INSERT(send, mshio); 729 } 730 } 731 nv_free(nv); 732 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 733 hio); 734 QUEUE_INSERT(disk, hio); 735 } 736 /* NOTREACHED */ 737 return (NULL); 738} 739 740/* 741 * Thread reads from or writes to local component and also handles DELETE and 742 * FLUSH requests. 743 */ 744static void * 745disk_thread(void *arg) 746{ 747 struct hast_resource *res = arg; 748 struct hio *hio; 749 ssize_t ret; 750 bool clear_activemap, logerror; 751 752 clear_activemap = true; 753 754 for (;;) { 755 pjdlog_debug(2, "disk: Taking request."); 756 QUEUE_TAKE(disk, hio); 757 while (clear_activemap) { 758 unsigned char *map; 759 size_t mapsize; 760 761 /* 762 * When first request is received, it means that primary 763 * already received our activemap, merged it and stored 764 * locally. We can now safely clear our activemap. 765 */ 766 mapsize = 767 activemap_calc_ondisk_size(res->hr_local_mediasize - 768 METADATA_SIZE, res->hr_extentsize, 769 res->hr_local_sectorsize); 770 map = calloc(1, mapsize); 771 if (map == NULL) { 772 pjdlog_warning("Unable to allocate memory to clear local activemap."); 773 break; 774 } 775 if (pwrite(res->hr_localfd, map, mapsize, 776 METADATA_SIZE) != (ssize_t)mapsize) { 777 pjdlog_errno(LOG_WARNING, 778 "Unable to store cleared activemap"); 779 free(map); 780 res->hr_stat_activemap_write_error++; 781 break; 782 } 783 free(map); 784 clear_activemap = false; 785 pjdlog_debug(1, "Local activemap cleared."); 786 break; 787 } 788 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 789 logerror = true; 790 /* Handle the actual request. */ 791 switch (hio->hio_cmd) { 792 case HIO_READ: 793 ret = pread(res->hr_localfd, hio->hio_data, 794 hio->hio_length, 795 hio->hio_offset + res->hr_localoff); 796 if (ret == -1) 797 hio->hio_error = errno; 798 else if (ret != (int64_t)hio->hio_length) 799 hio->hio_error = EIO; 800 else 801 hio->hio_error = 0; 802 break; 803 case HIO_WRITE: 804 ret = pwrite(res->hr_localfd, hio->hio_data, 805 hio->hio_length, 806 hio->hio_offset + res->hr_localoff); 807 if (ret == -1) 808 hio->hio_error = errno; 809 else if (ret != (int64_t)hio->hio_length) 810 hio->hio_error = EIO; 811 else 812 hio->hio_error = 0; 813 break; 814 case HIO_DELETE: 815 ret = g_delete(res->hr_localfd, 816 hio->hio_offset + res->hr_localoff, 817 hio->hio_length); 818 if (ret == -1) 819 hio->hio_error = errno; 820 else 821 hio->hio_error = 0; 822 break; 823 case HIO_FLUSH: 824 if (!res->hr_localflush) { 825 ret = -1; 826 hio->hio_error = EOPNOTSUPP; 827 logerror = false; 828 break; 829 } 830 ret = g_flush(res->hr_localfd); 831 if (ret == -1) { 832 if (errno == EOPNOTSUPP) 833 res->hr_localflush = false; 834 hio->hio_error = errno; 835 } else { 836 hio->hio_error = 0; 837 } 838 break; 839 default: 840 PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 841 hio->hio_cmd); 842 } 843 if (logerror && hio->hio_error != 0) { 844 reqlog(LOG_ERR, 0, hio->hio_error, hio, 845 "Request failed: "); 846 } 847 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 848 hio); 849 QUEUE_INSERT(send, hio); 850 } 851 /* NOTREACHED */ 852 return (NULL); 853} 854 855/* 856 * Thread sends requests back to primary node. 857 */ 858static void * 859send_thread(void *arg) 860{ 861 struct hast_resource *res = arg; 862 struct nv *nvout; 863 struct hio *hio; 864 void *data; 865 size_t length; 866 867 for (;;) { 868 pjdlog_debug(2, "send: Taking request."); 869 QUEUE_TAKE(send, hio); 870 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 871 nvout = nv_alloc(); 872 /* Copy sequence number. */ 873 nv_add_uint64(nvout, hio->hio_seq, "seq"); 874 if (hio->hio_memsync) { 875 PJDLOG_ASSERT(hio->hio_cmd == HIO_WRITE); 876 nv_add_int8(nvout, 1, "received"); 877 } 878 switch (hio->hio_cmd) { 879 case HIO_READ: 880 if (hio->hio_error == 0) { 881 data = hio->hio_data; 882 length = hio->hio_length; 883 break; 884 } 885 /* 886 * We send no data in case of an error. 887 */ 888 /* FALLTHROUGH */ 889 case HIO_DELETE: 890 case HIO_FLUSH: 891 case HIO_WRITE: 892 data = NULL; 893 length = 0; 894 break; 895 default: 896 PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 897 hio->hio_cmd); 898 } 899 if (hio->hio_error != 0) { 900 switch (hio->hio_cmd) { 901 case HIO_READ: 902 res->hr_stat_read_error++; 903 break; 904 case HIO_WRITE: 905 res->hr_stat_write_error++; 906 break; 907 case HIO_DELETE: 908 res->hr_stat_delete_error++; 909 break; 910 case HIO_FLUSH: 911 res->hr_stat_flush_error++; 912 break; 913 } 914 nv_add_int16(nvout, hio->hio_error, "error"); 915 } 916 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 917 length) == -1) { 918 secondary_exit(EX_TEMPFAIL, "Unable to send reply"); 919 } 920 nv_free(nvout); 921 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 922 hio); 923 hio_clear(hio); 924 QUEUE_INSERT(free, hio); 925 } 926 /* NOTREACHED */ 927 return (NULL); 928} 929