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