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