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/cdefs.h> 34 __FBSDID("$FreeBSD$"); 35 36 #include <sys/param.h> 37 #include <sys/time.h> 38 #include <sys/bio.h> 39 #include <sys/disk.h> 40 #include <sys/stat.h> 41 42 #include <err.h> 43 #include <errno.h> 44 #include <fcntl.h> 45 #include <libgeom.h> 46 #include <pthread.h> 47 #include <signal.h> 48 #include <stdint.h> 49 #include <stdio.h> 50 #include <string.h> 51 #include <sysexits.h> 52 #include <unistd.h> 53 54 #include <activemap.h> 55 #include <nv.h> 56 #include <pjdlog.h> 57 58 #include "control.h" 59 #include "event.h" 60 #include "hast.h" 61 #include "hast_proto.h" 62 #include "hastd.h" 63 #include "hooks.h" 64 #include "metadata.h" 65 #include "proto.h" 66 #include "subr.h" 67 #include "synch.h" 68 69 struct hio { 70 uint64_t hio_seq; 71 int hio_error; 72 void *hio_data; 73 uint8_t hio_cmd; 74 uint64_t hio_offset; 75 uint64_t hio_length; 76 bool hio_memsync; 77 TAILQ_ENTRY(hio) hio_next; 78 }; 79 80 static struct hast_resource *gres; 81 82 /* 83 * Free list holds unused structures. When free list is empty, we have to wait 84 * until some in-progress requests are freed. 85 */ 86 static TAILQ_HEAD(, hio) hio_free_list; 87 static size_t hio_free_list_size; 88 static pthread_mutex_t hio_free_list_lock; 89 static pthread_cond_t hio_free_list_cond; 90 /* 91 * Disk thread (the one that does I/O requests) takes requests from this list. 92 */ 93 static TAILQ_HEAD(, hio) hio_disk_list; 94 static size_t hio_disk_list_size; 95 static pthread_mutex_t hio_disk_list_lock; 96 static pthread_cond_t hio_disk_list_cond; 97 /* 98 * Thread that sends requests back to primary takes requests from this list. 99 */ 100 static TAILQ_HEAD(, hio) hio_send_list; 101 static size_t hio_send_list_size; 102 static pthread_mutex_t hio_send_list_lock; 103 static pthread_cond_t hio_send_list_cond; 104 105 /* 106 * Maximum number of outstanding I/O requests. 107 */ 108 #define HAST_HIO_MAX 256 109 110 static void *recv_thread(void *arg); 111 static void *disk_thread(void *arg); 112 static void *send_thread(void *arg); 113 114 #define QUEUE_INSERT(name, hio) do { \ 115 mtx_lock(&hio_##name##_list_lock); \ 116 if (TAILQ_EMPTY(&hio_##name##_list)) \ 117 cv_broadcast(&hio_##name##_list_cond); \ 118 TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \ 119 hio_##name##_list_size++; \ 120 mtx_unlock(&hio_##name##_list_lock); \ 121 } while (0) 122 #define QUEUE_TAKE(name, hio) do { \ 123 mtx_lock(&hio_##name##_list_lock); \ 124 while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 125 cv_wait(&hio_##name##_list_cond, \ 126 &hio_##name##_list_lock); \ 127 } \ 128 PJDLOG_ASSERT(hio_##name##_list_size != 0); \ 129 hio_##name##_list_size--; \ 130 TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \ 131 mtx_unlock(&hio_##name##_list_lock); \ 132 } while (0) 133 134 static void 135 output_status_aux(struct nv *nvout) 136 { 137 138 nv_add_uint64(nvout, (uint64_t)hio_free_list_size, "idle_queue_size"); 139 nv_add_uint64(nvout, (uint64_t)hio_disk_list_size, "local_queue_size"); 140 nv_add_uint64(nvout, (uint64_t)hio_send_list_size, "send_queue_size"); 141 } 142 143 static void 144 hio_clear(struct hio *hio) 145 { 146 147 hio->hio_seq = 0; 148 hio->hio_error = 0; 149 hio->hio_cmd = HIO_UNDEF; 150 hio->hio_offset = 0; 151 hio->hio_length = 0; 152 hio->hio_memsync = false; 153 } 154 155 static void 156 hio_copy(const struct hio *srchio, struct hio *dsthio) 157 { 158 159 /* 160 * We don't copy hio_error, hio_data and hio_next fields. 161 */ 162 163 dsthio->hio_seq = srchio->hio_seq; 164 dsthio->hio_cmd = srchio->hio_cmd; 165 dsthio->hio_offset = srchio->hio_offset; 166 dsthio->hio_length = srchio->hio_length; 167 dsthio->hio_memsync = srchio->hio_memsync; 168 } 169 170 static void 171 init_environment(void) 172 { 173 struct hio *hio; 174 unsigned int ii; 175 176 /* 177 * Initialize lists, their locks and theirs condition variables. 178 */ 179 TAILQ_INIT(&hio_free_list); 180 mtx_init(&hio_free_list_lock); 181 cv_init(&hio_free_list_cond); 182 TAILQ_INIT(&hio_disk_list); 183 mtx_init(&hio_disk_list_lock); 184 cv_init(&hio_disk_list_cond); 185 TAILQ_INIT(&hio_send_list); 186 mtx_init(&hio_send_list_lock); 187 cv_init(&hio_send_list_cond); 188 189 /* 190 * Allocate requests pool and initialize requests. 191 */ 192 for (ii = 0; ii < HAST_HIO_MAX; ii++) { 193 hio = malloc(sizeof(*hio)); 194 if (hio == NULL) { 195 pjdlog_exitx(EX_TEMPFAIL, 196 "Unable to allocate memory (%zu bytes) for hio request.", 197 sizeof(*hio)); 198 } 199 hio->hio_data = malloc(MAXPHYS); 200 if (hio->hio_data == NULL) { 201 pjdlog_exitx(EX_TEMPFAIL, 202 "Unable to allocate memory (%zu bytes) for gctl_data.", 203 (size_t)MAXPHYS); 204 } 205 hio_clear(hio); 206 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 207 hio_free_list_size++; 208 } 209 } 210 211 static void 212 init_local(struct hast_resource *res) 213 { 214 215 if (metadata_read(res, true) == -1) 216 exit(EX_NOINPUT); 217 } 218 219 static void 220 init_remote(struct hast_resource *res, struct nv *nvin) 221 { 222 uint64_t resuid; 223 struct nv *nvout; 224 unsigned char *map; 225 size_t mapsize; 226 227 #ifdef notyet 228 /* Setup direction. */ 229 if (proto_send(res->hr_remoteout, NULL, 0) == -1) 230 pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 231 #endif 232 233 nvout = nv_alloc(); 234 nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 235 nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 236 resuid = nv_get_uint64(nvin, "resuid"); 237 res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 238 res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 239 nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 240 nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 241 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 242 METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 243 map = malloc(mapsize); 244 if (map == NULL) { 245 pjdlog_exitx(EX_TEMPFAIL, 246 "Unable to allocate memory (%zu bytes) for activemap.", 247 mapsize); 248 } 249 /* 250 * When we work as primary and secondary is missing we will increase 251 * localcnt in our metadata. When secondary is connected and synced 252 * we make localcnt be equal to remotecnt, which means nodes are more 253 * or less in sync. 254 * Split-brain condition is when both nodes are not able to communicate 255 * and are both configured as primary nodes. In turn, they can both 256 * make incompatible changes to the data and we have to detect that. 257 * Under split-brain condition we will increase our localcnt on first 258 * write and remote node will increase its localcnt on first write. 259 * When we connect we can see that primary's localcnt is greater than 260 * our remotecnt (primary was modified while we weren't watching) and 261 * our localcnt is greater than primary's remotecnt (we were modified 262 * while primary wasn't watching). 263 * There are many possible combinations which are all gathered below. 264 * Don't pay too much attention to exact numbers, the more important 265 * is to compare them. We compare secondary's local with primary's 266 * remote and secondary's remote with primary's local. 267 * Note that every case where primary's localcnt is smaller than 268 * secondary's remotecnt and where secondary's localcnt is smaller than 269 * primary's remotecnt should be impossible in practise. We will perform 270 * full synchronization then. Those cases are marked with an asterisk. 271 * Regular synchronization means that only extents marked as dirty are 272 * synchronized (regular synchronization). 273 * 274 * SECONDARY METADATA PRIMARY METADATA 275 * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 276 * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 277 * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 278 * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 279 * regular sync from secondary. 280 * local=3 remote=3 local=3 remote=3 Regular sync just in case. 281 * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 282 * local=3 remote=3 local=4 remote=2 Split-brain condition. 283 * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 284 * regular sync from primary. 285 * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 286 */ 287 if (res->hr_resuid == 0) { 288 /* 289 * Provider is used for the first time. If primary node done no 290 * writes yet as well (we will find "virgin" argument) then 291 * there is no need to synchronize anything. If primary node 292 * done any writes already we have to synchronize everything. 293 */ 294 PJDLOG_ASSERT(res->hr_secondary_localcnt == 0); 295 res->hr_resuid = resuid; 296 if (metadata_write(res) == -1) 297 exit(EX_NOINPUT); 298 if (nv_exists(nvin, "virgin")) { 299 free(map); 300 map = NULL; 301 mapsize = 0; 302 } else { 303 memset(map, 0xff, mapsize); 304 } 305 nv_add_int8(nvout, 1, "virgin"); 306 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 307 } else if (res->hr_resuid != resuid) { 308 char errmsg[256]; 309 310 free(map); 311 (void)snprintf(errmsg, sizeof(errmsg), 312 "Resource unique ID mismatch (primary=%ju, secondary=%ju).", 313 (uintmax_t)resuid, (uintmax_t)res->hr_resuid); 314 pjdlog_error("%s", errmsg); 315 nv_add_string(nvout, errmsg, "errmsg"); 316 if (hast_proto_send(res, res->hr_remotein, nvout, 317 NULL, 0) == -1) { 318 pjdlog_exit(EX_TEMPFAIL, 319 "Unable to send response to %s", 320 res->hr_remoteaddr); 321 } 322 nv_free(nvout); 323 exit(EX_CONFIG); 324 } else if ( 325 /* Is primary out-of-date? */ 326 (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 327 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 328 /* Are the nodes more or less in sync? */ 329 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 330 res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 331 /* Is secondary out-of-date? */ 332 (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 333 res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 334 /* 335 * Nodes are more or less in sync or one of the nodes is 336 * out-of-date. 337 * It doesn't matter at this point which one, we just have to 338 * send out local bitmap to the remote node. 339 */ 340 if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 341 (ssize_t)mapsize) { 342 pjdlog_exit(LOG_ERR, "Unable to read activemap"); 343 } 344 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 345 res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 346 /* Primary is out-of-date, sync from secondary. */ 347 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 348 } else { 349 /* 350 * Secondary is out-of-date or counts match. 351 * Sync from primary. 352 */ 353 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 354 } 355 } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 356 res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 357 /* 358 * Not good, we have split-brain condition. 359 */ 360 free(map); 361 pjdlog_error("Split-brain detected, exiting."); 362 nv_add_string(nvout, "Split-brain condition!", "errmsg"); 363 if (hast_proto_send(res, res->hr_remotein, nvout, 364 NULL, 0) == -1) { 365 pjdlog_exit(EX_TEMPFAIL, 366 "Unable to send response to %s", 367 res->hr_remoteaddr); 368 } 369 nv_free(nvout); 370 /* Exit on split-brain. */ 371 event_send(res, EVENT_SPLITBRAIN); 372 exit(EX_CONFIG); 373 } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 374 res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 375 /* 376 * This should never happen in practise, but we will perform 377 * full synchronization. 378 */ 379 PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 380 res->hr_primary_localcnt < res->hr_secondary_remotecnt); 381 mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 382 METADATA_SIZE, res->hr_extentsize, 383 res->hr_local_sectorsize); 384 memset(map, 0xff, mapsize); 385 if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 386 /* In this one of five cases sync from secondary. */ 387 nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 388 } else { 389 /* For the rest four cases sync from primary. */ 390 nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 391 } 392 pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 393 (uintmax_t)res->hr_primary_localcnt, 394 (uintmax_t)res->hr_primary_remotecnt, 395 (uintmax_t)res->hr_secondary_localcnt, 396 (uintmax_t)res->hr_secondary_remotecnt); 397 } 398 nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 399 if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) { 400 pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 401 res->hr_remoteaddr); 402 } 403 if (map != NULL) 404 free(map); 405 nv_free(nvout); 406 #ifdef notyet 407 /* Setup direction. */ 408 if (proto_recv(res->hr_remotein, NULL, 0) == -1) 409 pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 410 #endif 411 } 412 413 void 414 hastd_secondary(struct hast_resource *res, struct nv *nvin) 415 { 416 sigset_t mask; 417 pthread_t td; 418 pid_t pid; 419 int error, mode, debuglevel; 420 421 /* 422 * Create communication channel between parent and child. 423 */ 424 if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) { 425 KEEP_ERRNO((void)pidfile_remove(pfh)); 426 pjdlog_exit(EX_OSERR, 427 "Unable to create control sockets between parent and child"); 428 } 429 /* 430 * Create communication channel between child and parent. 431 */ 432 if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) { 433 KEEP_ERRNO((void)pidfile_remove(pfh)); 434 pjdlog_exit(EX_OSERR, 435 "Unable to create event sockets between child and parent"); 436 } 437 438 pid = fork(); 439 if (pid == -1) { 440 KEEP_ERRNO((void)pidfile_remove(pfh)); 441 pjdlog_exit(EX_OSERR, "Unable to fork"); 442 } 443 444 if (pid > 0) { 445 /* This is parent. */ 446 proto_close(res->hr_remotein); 447 res->hr_remotein = NULL; 448 proto_close(res->hr_remoteout); 449 res->hr_remoteout = NULL; 450 /* Declare that we are receiver. */ 451 proto_recv(res->hr_event, NULL, 0); 452 /* Declare that we are sender. */ 453 proto_send(res->hr_ctrl, NULL, 0); 454 res->hr_workerpid = pid; 455 return; 456 } 457 458 gres = res; 459 res->output_status_aux = output_status_aux; 460 mode = pjdlog_mode_get(); 461 debuglevel = pjdlog_debug_get(); 462 463 /* Declare that we are sender. */ 464 proto_send(res->hr_event, NULL, 0); 465 /* Declare that we are receiver. */ 466 proto_recv(res->hr_ctrl, NULL, 0); 467 descriptors_cleanup(res); 468 469 descriptors_assert(res, mode); 470 471 pjdlog_init(mode); 472 pjdlog_debug_set(debuglevel); 473 pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 474 setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role)); 475 476 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 477 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 478 479 /* Error in setting timeout is not critical, but why should it fail? */ 480 if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1) 481 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 482 if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1) 483 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 484 485 init_local(res); 486 init_environment(); 487 488 if (drop_privs(res) != 0) 489 exit(EX_CONFIG); 490 pjdlog_info("Privileges successfully dropped."); 491 492 /* 493 * Create the control thread before sending any event to the parent, 494 * as we can deadlock when parent sends control request to worker, 495 * but worker has no control thread started yet, so parent waits. 496 * In the meantime worker sends an event to the parent, but parent 497 * is unable to handle the event, because it waits for control 498 * request response. 499 */ 500 error = pthread_create(&td, NULL, ctrl_thread, res); 501 PJDLOG_ASSERT(error == 0); 502 503 init_remote(res, nvin); 504 event_send(res, EVENT_CONNECT); 505 506 error = pthread_create(&td, NULL, recv_thread, res); 507 PJDLOG_ASSERT(error == 0); 508 error = pthread_create(&td, NULL, disk_thread, res); 509 PJDLOG_ASSERT(error == 0); 510 (void)send_thread(res); 511 } 512 513 static void 514 reqlog(int loglevel, int debuglevel, int error, struct hio *hio, 515 const char *fmt, ...) 516 { 517 char msg[1024]; 518 va_list ap; 519 int len; 520 521 va_start(ap, fmt); 522 len = vsnprintf(msg, sizeof(msg), fmt, ap); 523 va_end(ap); 524 if ((size_t)len < sizeof(msg)) { 525 switch (hio->hio_cmd) { 526 case HIO_READ: 527 (void)snprintf(msg + len, sizeof(msg) - len, 528 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 529 (uintmax_t)hio->hio_length); 530 break; 531 case HIO_DELETE: 532 (void)snprintf(msg + len, sizeof(msg) - len, 533 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 534 (uintmax_t)hio->hio_length); 535 break; 536 case HIO_FLUSH: 537 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 538 break; 539 case HIO_WRITE: 540 (void)snprintf(msg + len, sizeof(msg) - len, 541 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 542 (uintmax_t)hio->hio_length); 543 break; 544 case HIO_KEEPALIVE: 545 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 546 break; 547 default: 548 (void)snprintf(msg + len, sizeof(msg) - len, 549 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 550 break; 551 } 552 } 553 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 554 } 555 556 static int 557 requnpack(struct hast_resource *res, struct hio *hio, struct nv *nv) 558 { 559 560 hio->hio_cmd = nv_get_uint8(nv, "cmd"); 561 if (hio->hio_cmd == 0) { 562 pjdlog_error("Header contains no 'cmd' field."); 563 hio->hio_error = EINVAL; 564 goto end; 565 } 566 if (hio->hio_cmd != HIO_KEEPALIVE) { 567 hio->hio_seq = nv_get_uint64(nv, "seq"); 568 if (hio->hio_seq == 0) { 569 pjdlog_error("Header contains no 'seq' field."); 570 hio->hio_error = EINVAL; 571 goto end; 572 } 573 } 574 switch (hio->hio_cmd) { 575 case HIO_FLUSH: 576 case HIO_KEEPALIVE: 577 break; 578 case HIO_WRITE: 579 hio->hio_memsync = nv_exists(nv, "memsync"); 580 /* FALLTHROUGH */ 581 case HIO_READ: 582 case HIO_DELETE: 583 hio->hio_offset = nv_get_uint64(nv, "offset"); 584 if (nv_error(nv) != 0) { 585 pjdlog_error("Header is missing 'offset' field."); 586 hio->hio_error = EINVAL; 587 goto end; 588 } 589 hio->hio_length = nv_get_uint64(nv, "length"); 590 if (nv_error(nv) != 0) { 591 pjdlog_error("Header is missing 'length' field."); 592 hio->hio_error = EINVAL; 593 goto end; 594 } 595 if (hio->hio_length == 0) { 596 pjdlog_error("Data length is zero."); 597 hio->hio_error = EINVAL; 598 goto end; 599 } 600 if (hio->hio_cmd != HIO_DELETE && hio->hio_length > MAXPHYS) { 601 pjdlog_error("Data length is too large (%ju > %ju).", 602 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 603 hio->hio_error = EINVAL; 604 goto end; 605 } 606 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 607 pjdlog_error("Offset %ju is not multiple of sector size.", 608 (uintmax_t)hio->hio_offset); 609 hio->hio_error = EINVAL; 610 goto end; 611 } 612 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 613 pjdlog_error("Length %ju is not multiple of sector size.", 614 (uintmax_t)hio->hio_length); 615 hio->hio_error = EINVAL; 616 goto end; 617 } 618 if (hio->hio_offset + hio->hio_length > 619 (uint64_t)res->hr_datasize) { 620 pjdlog_error("Data offset is too large (%ju > %ju).", 621 (uintmax_t)(hio->hio_offset + hio->hio_length), 622 (uintmax_t)res->hr_datasize); 623 hio->hio_error = EINVAL; 624 goto end; 625 } 626 break; 627 default: 628 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 629 hio->hio_cmd); 630 hio->hio_error = EINVAL; 631 goto end; 632 } 633 hio->hio_error = 0; 634 end: 635 return (hio->hio_error); 636 } 637 638 static __dead2 void 639 secondary_exit(int exitcode, const char *fmt, ...) 640 { 641 va_list ap; 642 643 PJDLOG_ASSERT(exitcode != EX_OK); 644 va_start(ap, fmt); 645 pjdlogv_errno(LOG_ERR, fmt, ap); 646 va_end(ap); 647 event_send(gres, EVENT_DISCONNECT); 648 exit(exitcode); 649 } 650 651 /* 652 * Thread receives requests from the primary node. 653 */ 654 static void * 655 recv_thread(void *arg) 656 { 657 struct hast_resource *res = arg; 658 struct hio *hio, *mshio; 659 struct nv *nv; 660 661 for (;;) { 662 pjdlog_debug(2, "recv: Taking free request."); 663 QUEUE_TAKE(free, hio); 664 pjdlog_debug(2, "recv: (%p) Got request.", hio); 665 if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) { 666 secondary_exit(EX_TEMPFAIL, 667 "Unable to receive request header"); 668 } 669 if (requnpack(res, hio, nv) != 0) { 670 nv_free(nv); 671 pjdlog_debug(2, 672 "recv: (%p) Moving request to the send queue.", 673 hio); 674 QUEUE_INSERT(send, hio); 675 continue; 676 } 677 switch (hio->hio_cmd) { 678 case HIO_READ: 679 res->hr_stat_read++; 680 break; 681 case HIO_WRITE: 682 res->hr_stat_write++; 683 break; 684 case HIO_DELETE: 685 res->hr_stat_delete++; 686 break; 687 case HIO_FLUSH: 688 res->hr_stat_flush++; 689 break; 690 case HIO_KEEPALIVE: 691 break; 692 default: 693 PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 694 hio->hio_cmd); 695 } 696 reqlog(LOG_DEBUG, 2, -1, hio, 697 "recv: (%p) Got request header: ", hio); 698 if (hio->hio_cmd == HIO_KEEPALIVE) { 699 nv_free(nv); 700 pjdlog_debug(2, 701 "recv: (%p) Moving request to the free queue.", 702 hio); 703 hio_clear(hio); 704 QUEUE_INSERT(free, hio); 705 continue; 706 } else if (hio->hio_cmd == HIO_WRITE) { 707 if (hast_proto_recv_data(res, res->hr_remotein, nv, 708 hio->hio_data, MAXPHYS) == -1) { 709 secondary_exit(EX_TEMPFAIL, 710 "Unable to receive request data"); 711 } 712 if (hio->hio_memsync) { 713 /* 714 * For memsync requests we expect two replies. 715 * Clone the hio so we can handle both of them. 716 */ 717 pjdlog_debug(2, "recv: Taking free request."); 718 QUEUE_TAKE(free, mshio); 719 pjdlog_debug(2, "recv: (%p) Got request.", 720 mshio); 721 hio_copy(hio, mshio); 722 mshio->hio_error = 0; 723 /* 724 * We want to keep 'memsync' tag only on the 725 * request going onto send queue (mshio). 726 */ 727 hio->hio_memsync = false; 728 pjdlog_debug(2, 729 "recv: (%p) Moving memsync request to the send queue.", 730 mshio); 731 QUEUE_INSERT(send, mshio); 732 } 733 } 734 nv_free(nv); 735 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 736 hio); 737 QUEUE_INSERT(disk, hio); 738 } 739 /* NOTREACHED */ 740 return (NULL); 741 } 742 743 /* 744 * Thread reads from or writes to local component and also handles DELETE and 745 * FLUSH requests. 746 */ 747 static void * 748 disk_thread(void *arg) 749 { 750 struct hast_resource *res = arg; 751 struct hio *hio; 752 ssize_t ret; 753 bool clear_activemap, logerror; 754 755 clear_activemap = true; 756 757 for (;;) { 758 pjdlog_debug(2, "disk: Taking request."); 759 QUEUE_TAKE(disk, hio); 760 while (clear_activemap) { 761 unsigned char *map; 762 size_t mapsize; 763 764 /* 765 * When first request is received, it means that primary 766 * already received our activemap, merged it and stored 767 * locally. We can now safely clear our activemap. 768 */ 769 mapsize = 770 activemap_calc_ondisk_size(res->hr_local_mediasize - 771 METADATA_SIZE, res->hr_extentsize, 772 res->hr_local_sectorsize); 773 map = calloc(1, mapsize); 774 if (map == NULL) { 775 pjdlog_warning("Unable to allocate memory to clear local activemap."); 776 break; 777 } 778 if (pwrite(res->hr_localfd, map, mapsize, 779 METADATA_SIZE) != (ssize_t)mapsize) { 780 pjdlog_errno(LOG_WARNING, 781 "Unable to store cleared activemap"); 782 free(map); 783 res->hr_stat_activemap_write_error++; 784 break; 785 } 786 free(map); 787 clear_activemap = false; 788 pjdlog_debug(1, "Local activemap cleared."); 789 break; 790 } 791 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 792 logerror = true; 793 /* Handle the actual request. */ 794 switch (hio->hio_cmd) { 795 case HIO_READ: 796 ret = pread(res->hr_localfd, hio->hio_data, 797 hio->hio_length, 798 hio->hio_offset + res->hr_localoff); 799 if (ret == -1) 800 hio->hio_error = errno; 801 else if (ret != (int64_t)hio->hio_length) 802 hio->hio_error = EIO; 803 else 804 hio->hio_error = 0; 805 break; 806 case HIO_WRITE: 807 ret = pwrite(res->hr_localfd, hio->hio_data, 808 hio->hio_length, 809 hio->hio_offset + res->hr_localoff); 810 if (ret == -1) 811 hio->hio_error = errno; 812 else if (ret != (int64_t)hio->hio_length) 813 hio->hio_error = EIO; 814 else 815 hio->hio_error = 0; 816 break; 817 case HIO_DELETE: 818 ret = g_delete(res->hr_localfd, 819 hio->hio_offset + res->hr_localoff, 820 hio->hio_length); 821 if (ret == -1) 822 hio->hio_error = errno; 823 else 824 hio->hio_error = 0; 825 break; 826 case HIO_FLUSH: 827 if (!res->hr_localflush) { 828 ret = -1; 829 hio->hio_error = EOPNOTSUPP; 830 logerror = false; 831 break; 832 } 833 ret = g_flush(res->hr_localfd); 834 if (ret == -1) { 835 if (errno == EOPNOTSUPP) 836 res->hr_localflush = false; 837 hio->hio_error = errno; 838 } else { 839 hio->hio_error = 0; 840 } 841 break; 842 default: 843 PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 844 hio->hio_cmd); 845 } 846 if (logerror && hio->hio_error != 0) { 847 reqlog(LOG_ERR, 0, hio->hio_error, hio, 848 "Request failed: "); 849 } 850 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 851 hio); 852 QUEUE_INSERT(send, hio); 853 } 854 /* NOTREACHED */ 855 return (NULL); 856 } 857 858 /* 859 * Thread sends requests back to primary node. 860 */ 861 static void * 862 send_thread(void *arg) 863 { 864 struct hast_resource *res = arg; 865 struct nv *nvout; 866 struct hio *hio; 867 void *data; 868 size_t length; 869 870 for (;;) { 871 pjdlog_debug(2, "send: Taking request."); 872 QUEUE_TAKE(send, hio); 873 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 874 nvout = nv_alloc(); 875 /* Copy sequence number. */ 876 nv_add_uint64(nvout, hio->hio_seq, "seq"); 877 if (hio->hio_memsync) { 878 PJDLOG_ASSERT(hio->hio_cmd == HIO_WRITE); 879 nv_add_int8(nvout, 1, "received"); 880 } 881 switch (hio->hio_cmd) { 882 case HIO_READ: 883 if (hio->hio_error == 0) { 884 data = hio->hio_data; 885 length = hio->hio_length; 886 break; 887 } 888 /* 889 * We send no data in case of an error. 890 */ 891 /* FALLTHROUGH */ 892 case HIO_DELETE: 893 case HIO_FLUSH: 894 case HIO_WRITE: 895 data = NULL; 896 length = 0; 897 break; 898 default: 899 PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 900 hio->hio_cmd); 901 } 902 if (hio->hio_error != 0) { 903 switch (hio->hio_cmd) { 904 case HIO_READ: 905 res->hr_stat_read_error++; 906 break; 907 case HIO_WRITE: 908 res->hr_stat_write_error++; 909 break; 910 case HIO_DELETE: 911 res->hr_stat_delete_error++; 912 break; 913 case HIO_FLUSH: 914 res->hr_stat_flush_error++; 915 break; 916 } 917 nv_add_int16(nvout, hio->hio_error, "error"); 918 } 919 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 920 length) == -1) { 921 secondary_exit(EX_TEMPFAIL, "Unable to send reply"); 922 } 923 nv_free(nvout); 924 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 925 hio); 926 hio_clear(hio); 927 QUEUE_INSERT(free, hio); 928 } 929 /* NOTREACHED */ 930 return (NULL); 931 } 932