primary.c revision 212034
1/*- 2 * Copyright (c) 2009 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: head/sbin/hastd/primary.c 212034 2010-08-30 22:28:04Z pjd $"); 33 34#include <sys/types.h> 35#include <sys/time.h> 36#include <sys/bio.h> 37#include <sys/disk.h> 38#include <sys/refcount.h> 39#include <sys/stat.h> 40 41#include <geom/gate/g_gate.h> 42 43#include <assert.h> 44#include <err.h> 45#include <errno.h> 46#include <fcntl.h> 47#include <libgeom.h> 48#include <pthread.h> 49#include <signal.h> 50#include <stdint.h> 51#include <stdio.h> 52#include <string.h> 53#include <sysexits.h> 54#include <unistd.h> 55 56#include <activemap.h> 57#include <nv.h> 58#include <rangelock.h> 59 60#include "control.h" 61#include "hast.h" 62#include "hast_proto.h" 63#include "hastd.h" 64#include "hooks.h" 65#include "metadata.h" 66#include "proto.h" 67#include "pjdlog.h" 68#include "subr.h" 69#include "synch.h" 70 71/* The is only one remote component for now. */ 72#define ISREMOTE(no) ((no) == 1) 73 74struct hio { 75 /* 76 * Number of components we are still waiting for. 77 * When this field goes to 0, we can send the request back to the 78 * kernel. Each component has to decrease this counter by one 79 * even on failure. 80 */ 81 unsigned int hio_countdown; 82 /* 83 * Each component has a place to store its own error. 84 * Once the request is handled by all components we can decide if the 85 * request overall is successful or not. 86 */ 87 int *hio_errors; 88 /* 89 * Structure used to comunicate with GEOM Gate class. 90 */ 91 struct g_gate_ctl_io hio_ggio; 92 TAILQ_ENTRY(hio) *hio_next; 93}; 94#define hio_free_next hio_next[0] 95#define hio_done_next hio_next[0] 96 97/* 98 * Free list holds unused structures. When free list is empty, we have to wait 99 * until some in-progress requests are freed. 100 */ 101static TAILQ_HEAD(, hio) hio_free_list; 102static pthread_mutex_t hio_free_list_lock; 103static pthread_cond_t hio_free_list_cond; 104/* 105 * There is one send list for every component. One requests is placed on all 106 * send lists - each component gets the same request, but each component is 107 * responsible for managing his own send list. 108 */ 109static TAILQ_HEAD(, hio) *hio_send_list; 110static pthread_mutex_t *hio_send_list_lock; 111static pthread_cond_t *hio_send_list_cond; 112/* 113 * There is one recv list for every component, although local components don't 114 * use recv lists as local requests are done synchronously. 115 */ 116static TAILQ_HEAD(, hio) *hio_recv_list; 117static pthread_mutex_t *hio_recv_list_lock; 118static pthread_cond_t *hio_recv_list_cond; 119/* 120 * Request is placed on done list by the slowest component (the one that 121 * decreased hio_countdown from 1 to 0). 122 */ 123static TAILQ_HEAD(, hio) hio_done_list; 124static pthread_mutex_t hio_done_list_lock; 125static pthread_cond_t hio_done_list_cond; 126/* 127 * Structure below are for interaction with sync thread. 128 */ 129static bool sync_inprogress; 130static pthread_mutex_t sync_lock; 131static pthread_cond_t sync_cond; 132/* 133 * The lock below allows to synchornize access to remote connections. 134 */ 135static pthread_rwlock_t *hio_remote_lock; 136 137/* 138 * Lock to synchronize metadata updates. Also synchronize access to 139 * hr_primary_localcnt and hr_primary_remotecnt fields. 140 */ 141static pthread_mutex_t metadata_lock; 142 143/* 144 * Maximum number of outstanding I/O requests. 145 */ 146#define HAST_HIO_MAX 256 147/* 148 * Number of components. At this point there are only two components: local 149 * and remote, but in the future it might be possible to use multiple local 150 * and remote components. 151 */ 152#define HAST_NCOMPONENTS 2 153/* 154 * Number of seconds to sleep between reconnect retries or keepalive packets. 155 */ 156#define RETRY_SLEEP 10 157 158#define ISCONNECTED(res, no) \ 159 ((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL) 160 161#define QUEUE_INSERT1(hio, name, ncomp) do { \ 162 bool _wakeup; \ 163 \ 164 mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 165 _wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]); \ 166 TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio), \ 167 hio_next[(ncomp)]); \ 168 mtx_unlock(&hio_##name##_list_lock[ncomp]); \ 169 if (_wakeup) \ 170 cv_signal(&hio_##name##_list_cond[(ncomp)]); \ 171} while (0) 172#define QUEUE_INSERT2(hio, name) do { \ 173 bool _wakeup; \ 174 \ 175 mtx_lock(&hio_##name##_list_lock); \ 176 _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 177 TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\ 178 mtx_unlock(&hio_##name##_list_lock); \ 179 if (_wakeup) \ 180 cv_signal(&hio_##name##_list_cond); \ 181} while (0) 182#define QUEUE_TAKE1(hio, name, ncomp) do { \ 183 mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 184 while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL) { \ 185 cv_wait(&hio_##name##_list_cond[(ncomp)], \ 186 &hio_##name##_list_lock[(ncomp)]); \ 187 } \ 188 TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio), \ 189 hio_next[(ncomp)]); \ 190 mtx_unlock(&hio_##name##_list_lock[(ncomp)]); \ 191} while (0) 192#define QUEUE_TAKE2(hio, name) do { \ 193 mtx_lock(&hio_##name##_list_lock); \ 194 while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 195 cv_wait(&hio_##name##_list_cond, \ 196 &hio_##name##_list_lock); \ 197 } \ 198 TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next); \ 199 mtx_unlock(&hio_##name##_list_lock); \ 200} while (0) 201 202#define SYNCREQ(hio) do { \ 203 (hio)->hio_ggio.gctl_unit = -1; \ 204 (hio)->hio_ggio.gctl_seq = 1; \ 205} while (0) 206#define ISSYNCREQ(hio) ((hio)->hio_ggio.gctl_unit == -1) 207#define SYNCREQDONE(hio) do { (hio)->hio_ggio.gctl_unit = -2; } while (0) 208#define ISSYNCREQDONE(hio) ((hio)->hio_ggio.gctl_unit == -2) 209 210static struct hast_resource *gres; 211 212static pthread_mutex_t range_lock; 213static struct rangelocks *range_regular; 214static bool range_regular_wait; 215static pthread_cond_t range_regular_cond; 216static struct rangelocks *range_sync; 217static bool range_sync_wait; 218static pthread_cond_t range_sync_cond; 219 220static void *ggate_recv_thread(void *arg); 221static void *local_send_thread(void *arg); 222static void *remote_send_thread(void *arg); 223static void *remote_recv_thread(void *arg); 224static void *ggate_send_thread(void *arg); 225static void *sync_thread(void *arg); 226static void *guard_thread(void *arg); 227 228static void 229dummy_sighandler(int sig __unused) 230{ 231 /* Nothing to do. */ 232} 233 234static void 235cleanup(struct hast_resource *res) 236{ 237 int rerrno; 238 239 /* Remember errno. */ 240 rerrno = errno; 241 242 /* 243 * Close descriptor to /dev/hast/<name> 244 * to work-around race in the kernel. 245 */ 246 close(res->hr_localfd); 247 248 /* Destroy ggate provider if we created one. */ 249 if (res->hr_ggateunit >= 0) { 250 struct g_gate_ctl_destroy ggiod; 251 252 ggiod.gctl_version = G_GATE_VERSION; 253 ggiod.gctl_unit = res->hr_ggateunit; 254 ggiod.gctl_force = 1; 255 if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 256 pjdlog_warning("Unable to destroy hast/%s device", 257 res->hr_provname); 258 } 259 res->hr_ggateunit = -1; 260 } 261 262 /* Restore errno. */ 263 errno = rerrno; 264} 265 266static void 267primary_exit(int exitcode, const char *fmt, ...) 268{ 269 va_list ap; 270 271 assert(exitcode != EX_OK); 272 va_start(ap, fmt); 273 pjdlogv_errno(LOG_ERR, fmt, ap); 274 va_end(ap); 275 cleanup(gres); 276 exit(exitcode); 277} 278 279static void 280primary_exitx(int exitcode, const char *fmt, ...) 281{ 282 va_list ap; 283 284 va_start(ap, fmt); 285 pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 286 va_end(ap); 287 cleanup(gres); 288 exit(exitcode); 289} 290 291static int 292hast_activemap_flush(struct hast_resource *res) 293{ 294 const unsigned char *buf; 295 size_t size; 296 297 buf = activemap_bitmap(res->hr_amp, &size); 298 assert(buf != NULL); 299 assert((size % res->hr_local_sectorsize) == 0); 300 if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 301 (ssize_t)size) { 302 KEEP_ERRNO(pjdlog_errno(LOG_ERR, 303 "Unable to flush activemap to disk")); 304 return (-1); 305 } 306 return (0); 307} 308 309static bool 310real_remote(const struct hast_resource *res) 311{ 312 313 return (strcmp(res->hr_remoteaddr, "none") != 0); 314} 315 316static void 317init_environment(struct hast_resource *res __unused) 318{ 319 struct hio *hio; 320 unsigned int ii, ncomps; 321 sigset_t mask; 322 323 /* 324 * In the future it might be per-resource value. 325 */ 326 ncomps = HAST_NCOMPONENTS; 327 328 /* 329 * Allocate memory needed by lists. 330 */ 331 hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 332 if (hio_send_list == NULL) { 333 primary_exitx(EX_TEMPFAIL, 334 "Unable to allocate %zu bytes of memory for send lists.", 335 sizeof(hio_send_list[0]) * ncomps); 336 } 337 hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 338 if (hio_send_list_lock == NULL) { 339 primary_exitx(EX_TEMPFAIL, 340 "Unable to allocate %zu bytes of memory for send list locks.", 341 sizeof(hio_send_list_lock[0]) * ncomps); 342 } 343 hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 344 if (hio_send_list_cond == NULL) { 345 primary_exitx(EX_TEMPFAIL, 346 "Unable to allocate %zu bytes of memory for send list condition variables.", 347 sizeof(hio_send_list_cond[0]) * ncomps); 348 } 349 hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 350 if (hio_recv_list == NULL) { 351 primary_exitx(EX_TEMPFAIL, 352 "Unable to allocate %zu bytes of memory for recv lists.", 353 sizeof(hio_recv_list[0]) * ncomps); 354 } 355 hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 356 if (hio_recv_list_lock == NULL) { 357 primary_exitx(EX_TEMPFAIL, 358 "Unable to allocate %zu bytes of memory for recv list locks.", 359 sizeof(hio_recv_list_lock[0]) * ncomps); 360 } 361 hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 362 if (hio_recv_list_cond == NULL) { 363 primary_exitx(EX_TEMPFAIL, 364 "Unable to allocate %zu bytes of memory for recv list condition variables.", 365 sizeof(hio_recv_list_cond[0]) * ncomps); 366 } 367 hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 368 if (hio_remote_lock == NULL) { 369 primary_exitx(EX_TEMPFAIL, 370 "Unable to allocate %zu bytes of memory for remote connections locks.", 371 sizeof(hio_remote_lock[0]) * ncomps); 372 } 373 374 /* 375 * Initialize lists, their locks and theirs condition variables. 376 */ 377 TAILQ_INIT(&hio_free_list); 378 mtx_init(&hio_free_list_lock); 379 cv_init(&hio_free_list_cond); 380 for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 381 TAILQ_INIT(&hio_send_list[ii]); 382 mtx_init(&hio_send_list_lock[ii]); 383 cv_init(&hio_send_list_cond[ii]); 384 TAILQ_INIT(&hio_recv_list[ii]); 385 mtx_init(&hio_recv_list_lock[ii]); 386 cv_init(&hio_recv_list_cond[ii]); 387 rw_init(&hio_remote_lock[ii]); 388 } 389 TAILQ_INIT(&hio_done_list); 390 mtx_init(&hio_done_list_lock); 391 cv_init(&hio_done_list_cond); 392 mtx_init(&metadata_lock); 393 394 /* 395 * Allocate requests pool and initialize requests. 396 */ 397 for (ii = 0; ii < HAST_HIO_MAX; ii++) { 398 hio = malloc(sizeof(*hio)); 399 if (hio == NULL) { 400 primary_exitx(EX_TEMPFAIL, 401 "Unable to allocate %zu bytes of memory for hio request.", 402 sizeof(*hio)); 403 } 404 hio->hio_countdown = 0; 405 hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 406 if (hio->hio_errors == NULL) { 407 primary_exitx(EX_TEMPFAIL, 408 "Unable allocate %zu bytes of memory for hio errors.", 409 sizeof(hio->hio_errors[0]) * ncomps); 410 } 411 hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 412 if (hio->hio_next == NULL) { 413 primary_exitx(EX_TEMPFAIL, 414 "Unable allocate %zu bytes of memory for hio_next field.", 415 sizeof(hio->hio_next[0]) * ncomps); 416 } 417 hio->hio_ggio.gctl_version = G_GATE_VERSION; 418 hio->hio_ggio.gctl_data = malloc(MAXPHYS); 419 if (hio->hio_ggio.gctl_data == NULL) { 420 primary_exitx(EX_TEMPFAIL, 421 "Unable to allocate %zu bytes of memory for gctl_data.", 422 MAXPHYS); 423 } 424 hio->hio_ggio.gctl_length = MAXPHYS; 425 hio->hio_ggio.gctl_error = 0; 426 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 427 } 428 429 /* 430 * Turn on signals handling. 431 */ 432 /* Because SIGCHLD is ignored by default, setup dummy handler for it. */ 433 PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 434 PJDLOG_VERIFY(sigfillset(&mask) == 0); 435 PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 436} 437 438static void 439init_local(struct hast_resource *res) 440{ 441 unsigned char *buf; 442 size_t mapsize; 443 444 if (metadata_read(res, true) < 0) 445 exit(EX_NOINPUT); 446 mtx_init(&res->hr_amp_lock); 447 if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 448 res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 449 primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 450 } 451 mtx_init(&range_lock); 452 cv_init(&range_regular_cond); 453 if (rangelock_init(&range_regular) < 0) 454 primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 455 cv_init(&range_sync_cond); 456 if (rangelock_init(&range_sync) < 0) 457 primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 458 mapsize = activemap_ondisk_size(res->hr_amp); 459 buf = calloc(1, mapsize); 460 if (buf == NULL) { 461 primary_exitx(EX_TEMPFAIL, 462 "Unable to allocate buffer for activemap."); 463 } 464 if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 465 (ssize_t)mapsize) { 466 primary_exit(EX_NOINPUT, "Unable to read activemap"); 467 } 468 activemap_copyin(res->hr_amp, buf, mapsize); 469 free(buf); 470 if (res->hr_resuid != 0) 471 return; 472 /* 473 * We're using provider for the first time, so we have to generate 474 * resource unique identifier and initialize local and remote counts. 475 */ 476 arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 477 res->hr_primary_localcnt = 1; 478 res->hr_primary_remotecnt = 0; 479 if (metadata_write(res) < 0) 480 exit(EX_NOINPUT); 481} 482 483static bool 484init_remote(struct hast_resource *res, struct proto_conn **inp, 485 struct proto_conn **outp) 486{ 487 struct proto_conn *in, *out; 488 struct nv *nvout, *nvin; 489 const unsigned char *token; 490 unsigned char *map; 491 const char *errmsg; 492 int32_t extentsize; 493 int64_t datasize; 494 uint32_t mapsize; 495 size_t size; 496 497 assert((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 498 assert(real_remote(res)); 499 500 in = out = NULL; 501 errmsg = NULL; 502 503 /* Prepare outgoing connection with remote node. */ 504 if (proto_client(res->hr_remoteaddr, &out) < 0) { 505 primary_exit(EX_TEMPFAIL, "Unable to create connection to %s", 506 res->hr_remoteaddr); 507 } 508 /* Try to connect, but accept failure. */ 509 if (proto_connect(out) < 0) { 510 pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 511 res->hr_remoteaddr); 512 goto close; 513 } 514 /* Error in setting timeout is not critical, but why should it fail? */ 515 if (proto_timeout(out, res->hr_timeout) < 0) 516 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 517 /* 518 * First handshake step. 519 * Setup outgoing connection with remote node. 520 */ 521 nvout = nv_alloc(); 522 nv_add_string(nvout, res->hr_name, "resource"); 523 if (nv_error(nvout) != 0) { 524 pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 525 "Unable to allocate header for connection with %s", 526 res->hr_remoteaddr); 527 nv_free(nvout); 528 goto close; 529 } 530 if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 531 pjdlog_errno(LOG_WARNING, 532 "Unable to send handshake header to %s", 533 res->hr_remoteaddr); 534 nv_free(nvout); 535 goto close; 536 } 537 nv_free(nvout); 538 if (hast_proto_recv_hdr(out, &nvin) < 0) { 539 pjdlog_errno(LOG_WARNING, 540 "Unable to receive handshake header from %s", 541 res->hr_remoteaddr); 542 goto close; 543 } 544 errmsg = nv_get_string(nvin, "errmsg"); 545 if (errmsg != NULL) { 546 pjdlog_warning("%s", errmsg); 547 nv_free(nvin); 548 goto close; 549 } 550 token = nv_get_uint8_array(nvin, &size, "token"); 551 if (token == NULL) { 552 pjdlog_warning("Handshake header from %s has no 'token' field.", 553 res->hr_remoteaddr); 554 nv_free(nvin); 555 goto close; 556 } 557 if (size != sizeof(res->hr_token)) { 558 pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 559 res->hr_remoteaddr, size, sizeof(res->hr_token)); 560 nv_free(nvin); 561 goto close; 562 } 563 bcopy(token, res->hr_token, sizeof(res->hr_token)); 564 nv_free(nvin); 565 566 /* 567 * Second handshake step. 568 * Setup incoming connection with remote node. 569 */ 570 if (proto_client(res->hr_remoteaddr, &in) < 0) { 571 pjdlog_errno(LOG_WARNING, "Unable to create connection to %s", 572 res->hr_remoteaddr); 573 } 574 /* Try to connect, but accept failure. */ 575 if (proto_connect(in) < 0) { 576 pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 577 res->hr_remoteaddr); 578 goto close; 579 } 580 /* Error in setting timeout is not critical, but why should it fail? */ 581 if (proto_timeout(in, res->hr_timeout) < 0) 582 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 583 nvout = nv_alloc(); 584 nv_add_string(nvout, res->hr_name, "resource"); 585 nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 586 "token"); 587 nv_add_uint64(nvout, res->hr_resuid, "resuid"); 588 nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 589 nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 590 if (nv_error(nvout) != 0) { 591 pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 592 "Unable to allocate header for connection with %s", 593 res->hr_remoteaddr); 594 nv_free(nvout); 595 goto close; 596 } 597 if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 598 pjdlog_errno(LOG_WARNING, 599 "Unable to send handshake header to %s", 600 res->hr_remoteaddr); 601 nv_free(nvout); 602 goto close; 603 } 604 nv_free(nvout); 605 if (hast_proto_recv_hdr(out, &nvin) < 0) { 606 pjdlog_errno(LOG_WARNING, 607 "Unable to receive handshake header from %s", 608 res->hr_remoteaddr); 609 goto close; 610 } 611 errmsg = nv_get_string(nvin, "errmsg"); 612 if (errmsg != NULL) { 613 pjdlog_warning("%s", errmsg); 614 nv_free(nvin); 615 goto close; 616 } 617 datasize = nv_get_int64(nvin, "datasize"); 618 if (datasize != res->hr_datasize) { 619 pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 620 (intmax_t)res->hr_datasize, (intmax_t)datasize); 621 nv_free(nvin); 622 goto close; 623 } 624 extentsize = nv_get_int32(nvin, "extentsize"); 625 if (extentsize != res->hr_extentsize) { 626 pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 627 (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 628 nv_free(nvin); 629 goto close; 630 } 631 res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 632 res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 633 res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 634 map = NULL; 635 mapsize = nv_get_uint32(nvin, "mapsize"); 636 if (mapsize > 0) { 637 map = malloc(mapsize); 638 if (map == NULL) { 639 pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 640 (uintmax_t)mapsize); 641 nv_free(nvin); 642 goto close; 643 } 644 /* 645 * Remote node have some dirty extents on its own, lets 646 * download its activemap. 647 */ 648 if (hast_proto_recv_data(res, out, nvin, map, 649 mapsize) < 0) { 650 pjdlog_errno(LOG_ERR, 651 "Unable to receive remote activemap"); 652 nv_free(nvin); 653 free(map); 654 goto close; 655 } 656 /* 657 * Merge local and remote bitmaps. 658 */ 659 activemap_merge(res->hr_amp, map, mapsize); 660 free(map); 661 /* 662 * Now that we merged bitmaps from both nodes, flush it to the 663 * disk before we start to synchronize. 664 */ 665 (void)hast_activemap_flush(res); 666 } 667 pjdlog_info("Connected to %s.", res->hr_remoteaddr); 668 if (inp != NULL && outp != NULL) { 669 *inp = in; 670 *outp = out; 671 } else { 672 res->hr_remotein = in; 673 res->hr_remoteout = out; 674 } 675 hook_exec(res->hr_exec, "connect", res->hr_name, NULL); 676 return (true); 677close: 678 if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0) 679 hook_exec(res->hr_exec, "split-brain", res->hr_name, NULL); 680 proto_close(out); 681 if (in != NULL) 682 proto_close(in); 683 return (false); 684} 685 686static void 687sync_start(void) 688{ 689 690 mtx_lock(&sync_lock); 691 sync_inprogress = true; 692 mtx_unlock(&sync_lock); 693 cv_signal(&sync_cond); 694} 695 696static void 697sync_stop(void) 698{ 699 700 mtx_lock(&sync_lock); 701 if (sync_inprogress) 702 sync_inprogress = false; 703 mtx_unlock(&sync_lock); 704} 705 706static void 707init_ggate(struct hast_resource *res) 708{ 709 struct g_gate_ctl_create ggiocreate; 710 struct g_gate_ctl_cancel ggiocancel; 711 712 /* 713 * We communicate with ggate via /dev/ggctl. Open it. 714 */ 715 res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 716 if (res->hr_ggatefd < 0) 717 primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 718 /* 719 * Create provider before trying to connect, as connection failure 720 * is not critical, but may take some time. 721 */ 722 ggiocreate.gctl_version = G_GATE_VERSION; 723 ggiocreate.gctl_mediasize = res->hr_datasize; 724 ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 725 ggiocreate.gctl_flags = 0; 726 ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 727 ggiocreate.gctl_timeout = 0; 728 ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 729 snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 730 res->hr_provname); 731 bzero(ggiocreate.gctl_info, sizeof(ggiocreate.gctl_info)); 732 if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 733 pjdlog_info("Device hast/%s created.", res->hr_provname); 734 res->hr_ggateunit = ggiocreate.gctl_unit; 735 return; 736 } 737 if (errno != EEXIST) { 738 primary_exit(EX_OSERR, "Unable to create hast/%s device", 739 res->hr_provname); 740 } 741 pjdlog_debug(1, 742 "Device hast/%s already exists, we will try to take it over.", 743 res->hr_provname); 744 /* 745 * If we received EEXIST, we assume that the process who created the 746 * provider died and didn't clean up. In that case we will start from 747 * where he left of. 748 */ 749 ggiocancel.gctl_version = G_GATE_VERSION; 750 ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 751 snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 752 res->hr_provname); 753 if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 754 pjdlog_info("Device hast/%s recovered.", res->hr_provname); 755 res->hr_ggateunit = ggiocancel.gctl_unit; 756 return; 757 } 758 primary_exit(EX_OSERR, "Unable to take over hast/%s device", 759 res->hr_provname); 760} 761 762void 763hastd_primary(struct hast_resource *res) 764{ 765 pthread_t td; 766 pid_t pid; 767 int error; 768 769 /* 770 * Create communication channel between parent and child. 771 */ 772 if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 773 KEEP_ERRNO((void)pidfile_remove(pfh)); 774 pjdlog_exit(EX_OSERR, 775 "Unable to create control sockets between parent and child"); 776 } 777 778 pid = fork(); 779 if (pid < 0) { 780 KEEP_ERRNO((void)pidfile_remove(pfh)); 781 pjdlog_exit(EX_TEMPFAIL, "Unable to fork"); 782 } 783 784 if (pid > 0) { 785 /* This is parent. */ 786 res->hr_workerpid = pid; 787 return; 788 } 789 790 gres = res; 791 792 (void)pidfile_close(pfh); 793 hook_fini(); 794 795 setproctitle("%s (primary)", res->hr_name); 796 797 signal(SIGHUP, SIG_DFL); 798 signal(SIGCHLD, SIG_DFL); 799 800 hook_init(); 801 init_local(res); 802 if (real_remote(res) && init_remote(res, NULL, NULL)) 803 sync_start(); 804 init_ggate(res); 805 init_environment(res); 806 error = pthread_create(&td, NULL, ggate_recv_thread, res); 807 assert(error == 0); 808 error = pthread_create(&td, NULL, local_send_thread, res); 809 assert(error == 0); 810 error = pthread_create(&td, NULL, remote_send_thread, res); 811 assert(error == 0); 812 error = pthread_create(&td, NULL, remote_recv_thread, res); 813 assert(error == 0); 814 error = pthread_create(&td, NULL, ggate_send_thread, res); 815 assert(error == 0); 816 error = pthread_create(&td, NULL, sync_thread, res); 817 assert(error == 0); 818 error = pthread_create(&td, NULL, ctrl_thread, res); 819 assert(error == 0); 820 (void)guard_thread(res); 821} 822 823static void 824reqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 825{ 826 char msg[1024]; 827 va_list ap; 828 int len; 829 830 va_start(ap, fmt); 831 len = vsnprintf(msg, sizeof(msg), fmt, ap); 832 va_end(ap); 833 if ((size_t)len < sizeof(msg)) { 834 switch (ggio->gctl_cmd) { 835 case BIO_READ: 836 (void)snprintf(msg + len, sizeof(msg) - len, 837 "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 838 (uintmax_t)ggio->gctl_length); 839 break; 840 case BIO_DELETE: 841 (void)snprintf(msg + len, sizeof(msg) - len, 842 "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 843 (uintmax_t)ggio->gctl_length); 844 break; 845 case BIO_FLUSH: 846 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 847 break; 848 case BIO_WRITE: 849 (void)snprintf(msg + len, sizeof(msg) - len, 850 "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 851 (uintmax_t)ggio->gctl_length); 852 break; 853 default: 854 (void)snprintf(msg + len, sizeof(msg) - len, 855 "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 856 break; 857 } 858 } 859 pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 860} 861 862static void 863remote_close(struct hast_resource *res, int ncomp) 864{ 865 866 rw_wlock(&hio_remote_lock[ncomp]); 867 /* 868 * A race is possible between dropping rlock and acquiring wlock - 869 * another thread can close connection in-between. 870 */ 871 if (!ISCONNECTED(res, ncomp)) { 872 assert(res->hr_remotein == NULL); 873 assert(res->hr_remoteout == NULL); 874 rw_unlock(&hio_remote_lock[ncomp]); 875 return; 876 } 877 878 assert(res->hr_remotein != NULL); 879 assert(res->hr_remoteout != NULL); 880 881 pjdlog_debug(2, "Closing incoming connection to %s.", 882 res->hr_remoteaddr); 883 proto_close(res->hr_remotein); 884 res->hr_remotein = NULL; 885 pjdlog_debug(2, "Closing outgoing connection to %s.", 886 res->hr_remoteaddr); 887 proto_close(res->hr_remoteout); 888 res->hr_remoteout = NULL; 889 890 rw_unlock(&hio_remote_lock[ncomp]); 891 892 pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 893 894 /* 895 * Stop synchronization if in-progress. 896 */ 897 sync_stop(); 898 899 hook_exec(res->hr_exec, "disconnect", res->hr_name, NULL); 900} 901 902/* 903 * Thread receives ggate I/O requests from the kernel and passes them to 904 * appropriate threads: 905 * WRITE - always goes to both local_send and remote_send threads 906 * READ (when the block is up-to-date on local component) - 907 * only local_send thread 908 * READ (when the block isn't up-to-date on local component) - 909 * only remote_send thread 910 * DELETE - always goes to both local_send and remote_send threads 911 * FLUSH - always goes to both local_send and remote_send threads 912 */ 913static void * 914ggate_recv_thread(void *arg) 915{ 916 struct hast_resource *res = arg; 917 struct g_gate_ctl_io *ggio; 918 struct hio *hio; 919 unsigned int ii, ncomp, ncomps; 920 int error; 921 922 ncomps = HAST_NCOMPONENTS; 923 924 for (;;) { 925 pjdlog_debug(2, "ggate_recv: Taking free request."); 926 QUEUE_TAKE2(hio, free); 927 pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 928 ggio = &hio->hio_ggio; 929 ggio->gctl_unit = res->hr_ggateunit; 930 ggio->gctl_length = MAXPHYS; 931 ggio->gctl_error = 0; 932 pjdlog_debug(2, 933 "ggate_recv: (%p) Waiting for request from the kernel.", 934 hio); 935 if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 936 if (sigexit_received) 937 pthread_exit(NULL); 938 primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 939 } 940 error = ggio->gctl_error; 941 switch (error) { 942 case 0: 943 break; 944 case ECANCELED: 945 /* Exit gracefully. */ 946 if (!sigexit_received) { 947 pjdlog_debug(2, 948 "ggate_recv: (%p) Received cancel from the kernel.", 949 hio); 950 pjdlog_info("Received cancel from the kernel, exiting."); 951 } 952 pthread_exit(NULL); 953 case ENOMEM: 954 /* 955 * Buffer too small? Impossible, we allocate MAXPHYS 956 * bytes - request can't be bigger than that. 957 */ 958 /* FALLTHROUGH */ 959 case ENXIO: 960 default: 961 primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 962 strerror(error)); 963 } 964 for (ii = 0; ii < ncomps; ii++) 965 hio->hio_errors[ii] = EINVAL; 966 reqlog(LOG_DEBUG, 2, ggio, 967 "ggate_recv: (%p) Request received from the kernel: ", 968 hio); 969 /* 970 * Inform all components about new write request. 971 * For read request prefer local component unless the given 972 * range is out-of-date, then use remote component. 973 */ 974 switch (ggio->gctl_cmd) { 975 case BIO_READ: 976 pjdlog_debug(2, 977 "ggate_recv: (%p) Moving request to the send queue.", 978 hio); 979 refcount_init(&hio->hio_countdown, 1); 980 mtx_lock(&metadata_lock); 981 if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 982 res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 983 /* 984 * This range is up-to-date on local component, 985 * so handle request locally. 986 */ 987 /* Local component is 0 for now. */ 988 ncomp = 0; 989 } else /* if (res->hr_syncsrc == 990 HAST_SYNCSRC_SECONDARY) */ { 991 assert(res->hr_syncsrc == 992 HAST_SYNCSRC_SECONDARY); 993 /* 994 * This range is out-of-date on local component, 995 * so send request to the remote node. 996 */ 997 /* Remote component is 1 for now. */ 998 ncomp = 1; 999 } 1000 mtx_unlock(&metadata_lock); 1001 QUEUE_INSERT1(hio, send, ncomp); 1002 break; 1003 case BIO_WRITE: 1004 for (;;) { 1005 mtx_lock(&range_lock); 1006 if (rangelock_islocked(range_sync, 1007 ggio->gctl_offset, ggio->gctl_length)) { 1008 pjdlog_debug(2, 1009 "regular: Range offset=%jd length=%zu locked.", 1010 (intmax_t)ggio->gctl_offset, 1011 (size_t)ggio->gctl_length); 1012 range_regular_wait = true; 1013 cv_wait(&range_regular_cond, &range_lock); 1014 range_regular_wait = false; 1015 mtx_unlock(&range_lock); 1016 continue; 1017 } 1018 if (rangelock_add(range_regular, 1019 ggio->gctl_offset, ggio->gctl_length) < 0) { 1020 mtx_unlock(&range_lock); 1021 pjdlog_debug(2, 1022 "regular: Range offset=%jd length=%zu is already locked, waiting.", 1023 (intmax_t)ggio->gctl_offset, 1024 (size_t)ggio->gctl_length); 1025 sleep(1); 1026 continue; 1027 } 1028 mtx_unlock(&range_lock); 1029 break; 1030 } 1031 mtx_lock(&res->hr_amp_lock); 1032 if (activemap_write_start(res->hr_amp, 1033 ggio->gctl_offset, ggio->gctl_length)) { 1034 (void)hast_activemap_flush(res); 1035 } 1036 mtx_unlock(&res->hr_amp_lock); 1037 /* FALLTHROUGH */ 1038 case BIO_DELETE: 1039 case BIO_FLUSH: 1040 pjdlog_debug(2, 1041 "ggate_recv: (%p) Moving request to the send queues.", 1042 hio); 1043 refcount_init(&hio->hio_countdown, ncomps); 1044 for (ii = 0; ii < ncomps; ii++) 1045 QUEUE_INSERT1(hio, send, ii); 1046 break; 1047 } 1048 } 1049 /* NOTREACHED */ 1050 return (NULL); 1051} 1052 1053/* 1054 * Thread reads from or writes to local component. 1055 * If local read fails, it redirects it to remote_send thread. 1056 */ 1057static void * 1058local_send_thread(void *arg) 1059{ 1060 struct hast_resource *res = arg; 1061 struct g_gate_ctl_io *ggio; 1062 struct hio *hio; 1063 unsigned int ncomp, rncomp; 1064 ssize_t ret; 1065 1066 /* Local component is 0 for now. */ 1067 ncomp = 0; 1068 /* Remote component is 1 for now. */ 1069 rncomp = 1; 1070 1071 for (;;) { 1072 pjdlog_debug(2, "local_send: Taking request."); 1073 QUEUE_TAKE1(hio, send, ncomp); 1074 pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1075 ggio = &hio->hio_ggio; 1076 switch (ggio->gctl_cmd) { 1077 case BIO_READ: 1078 ret = pread(res->hr_localfd, ggio->gctl_data, 1079 ggio->gctl_length, 1080 ggio->gctl_offset + res->hr_localoff); 1081 if (ret == ggio->gctl_length) 1082 hio->hio_errors[ncomp] = 0; 1083 else { 1084 /* 1085 * If READ failed, try to read from remote node. 1086 */ 1087 QUEUE_INSERT1(hio, send, rncomp); 1088 continue; 1089 } 1090 break; 1091 case BIO_WRITE: 1092 ret = pwrite(res->hr_localfd, ggio->gctl_data, 1093 ggio->gctl_length, 1094 ggio->gctl_offset + res->hr_localoff); 1095 if (ret < 0) 1096 hio->hio_errors[ncomp] = errno; 1097 else if (ret != ggio->gctl_length) 1098 hio->hio_errors[ncomp] = EIO; 1099 else 1100 hio->hio_errors[ncomp] = 0; 1101 break; 1102 case BIO_DELETE: 1103 ret = g_delete(res->hr_localfd, 1104 ggio->gctl_offset + res->hr_localoff, 1105 ggio->gctl_length); 1106 if (ret < 0) 1107 hio->hio_errors[ncomp] = errno; 1108 else 1109 hio->hio_errors[ncomp] = 0; 1110 break; 1111 case BIO_FLUSH: 1112 ret = g_flush(res->hr_localfd); 1113 if (ret < 0) 1114 hio->hio_errors[ncomp] = errno; 1115 else 1116 hio->hio_errors[ncomp] = 0; 1117 break; 1118 } 1119 if (refcount_release(&hio->hio_countdown)) { 1120 if (ISSYNCREQ(hio)) { 1121 mtx_lock(&sync_lock); 1122 SYNCREQDONE(hio); 1123 mtx_unlock(&sync_lock); 1124 cv_signal(&sync_cond); 1125 } else { 1126 pjdlog_debug(2, 1127 "local_send: (%p) Moving request to the done queue.", 1128 hio); 1129 QUEUE_INSERT2(hio, done); 1130 } 1131 } 1132 } 1133 /* NOTREACHED */ 1134 return (NULL); 1135} 1136 1137/* 1138 * Thread sends request to secondary node. 1139 */ 1140static void * 1141remote_send_thread(void *arg) 1142{ 1143 struct hast_resource *res = arg; 1144 struct g_gate_ctl_io *ggio; 1145 struct hio *hio; 1146 struct nv *nv; 1147 unsigned int ncomp; 1148 bool wakeup; 1149 uint64_t offset, length; 1150 uint8_t cmd; 1151 void *data; 1152 1153 /* Remote component is 1 for now. */ 1154 ncomp = 1; 1155 1156 for (;;) { 1157 pjdlog_debug(2, "remote_send: Taking request."); 1158 QUEUE_TAKE1(hio, send, ncomp); 1159 pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1160 ggio = &hio->hio_ggio; 1161 switch (ggio->gctl_cmd) { 1162 case BIO_READ: 1163 cmd = HIO_READ; 1164 data = NULL; 1165 offset = ggio->gctl_offset; 1166 length = ggio->gctl_length; 1167 break; 1168 case BIO_WRITE: 1169 cmd = HIO_WRITE; 1170 data = ggio->gctl_data; 1171 offset = ggio->gctl_offset; 1172 length = ggio->gctl_length; 1173 break; 1174 case BIO_DELETE: 1175 cmd = HIO_DELETE; 1176 data = NULL; 1177 offset = ggio->gctl_offset; 1178 length = ggio->gctl_length; 1179 break; 1180 case BIO_FLUSH: 1181 cmd = HIO_FLUSH; 1182 data = NULL; 1183 offset = 0; 1184 length = 0; 1185 break; 1186 default: 1187 assert(!"invalid condition"); 1188 abort(); 1189 } 1190 nv = nv_alloc(); 1191 nv_add_uint8(nv, cmd, "cmd"); 1192 nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1193 nv_add_uint64(nv, offset, "offset"); 1194 nv_add_uint64(nv, length, "length"); 1195 if (nv_error(nv) != 0) { 1196 hio->hio_errors[ncomp] = nv_error(nv); 1197 pjdlog_debug(2, 1198 "remote_send: (%p) Unable to prepare header to send.", 1199 hio); 1200 reqlog(LOG_ERR, 0, ggio, 1201 "Unable to prepare header to send (%s): ", 1202 strerror(nv_error(nv))); 1203 /* Move failed request immediately to the done queue. */ 1204 goto done_queue; 1205 } 1206 pjdlog_debug(2, 1207 "remote_send: (%p) Moving request to the recv queue.", 1208 hio); 1209 /* 1210 * Protect connection from disappearing. 1211 */ 1212 rw_rlock(&hio_remote_lock[ncomp]); 1213 if (!ISCONNECTED(res, ncomp)) { 1214 rw_unlock(&hio_remote_lock[ncomp]); 1215 hio->hio_errors[ncomp] = ENOTCONN; 1216 goto done_queue; 1217 } 1218 /* 1219 * Move the request to recv queue before sending it, because 1220 * in different order we can get reply before we move request 1221 * to recv queue. 1222 */ 1223 mtx_lock(&hio_recv_list_lock[ncomp]); 1224 wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1225 TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1226 mtx_unlock(&hio_recv_list_lock[ncomp]); 1227 if (hast_proto_send(res, res->hr_remoteout, nv, data, 1228 data != NULL ? length : 0) < 0) { 1229 hio->hio_errors[ncomp] = errno; 1230 rw_unlock(&hio_remote_lock[ncomp]); 1231 pjdlog_debug(2, 1232 "remote_send: (%p) Unable to send request.", hio); 1233 reqlog(LOG_ERR, 0, ggio, 1234 "Unable to send request (%s): ", 1235 strerror(hio->hio_errors[ncomp])); 1236 remote_close(res, ncomp); 1237 /* 1238 * Take request back from the receive queue and move 1239 * it immediately to the done queue. 1240 */ 1241 mtx_lock(&hio_recv_list_lock[ncomp]); 1242 TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1243 mtx_unlock(&hio_recv_list_lock[ncomp]); 1244 goto done_queue; 1245 } 1246 rw_unlock(&hio_remote_lock[ncomp]); 1247 nv_free(nv); 1248 if (wakeup) 1249 cv_signal(&hio_recv_list_cond[ncomp]); 1250 continue; 1251done_queue: 1252 nv_free(nv); 1253 if (ISSYNCREQ(hio)) { 1254 if (!refcount_release(&hio->hio_countdown)) 1255 continue; 1256 mtx_lock(&sync_lock); 1257 SYNCREQDONE(hio); 1258 mtx_unlock(&sync_lock); 1259 cv_signal(&sync_cond); 1260 continue; 1261 } 1262 if (ggio->gctl_cmd == BIO_WRITE) { 1263 mtx_lock(&res->hr_amp_lock); 1264 if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1265 ggio->gctl_length)) { 1266 (void)hast_activemap_flush(res); 1267 } 1268 mtx_unlock(&res->hr_amp_lock); 1269 } 1270 if (!refcount_release(&hio->hio_countdown)) 1271 continue; 1272 pjdlog_debug(2, 1273 "remote_send: (%p) Moving request to the done queue.", 1274 hio); 1275 QUEUE_INSERT2(hio, done); 1276 } 1277 /* NOTREACHED */ 1278 return (NULL); 1279} 1280 1281/* 1282 * Thread receives answer from secondary node and passes it to ggate_send 1283 * thread. 1284 */ 1285static void * 1286remote_recv_thread(void *arg) 1287{ 1288 struct hast_resource *res = arg; 1289 struct g_gate_ctl_io *ggio; 1290 struct hio *hio; 1291 struct nv *nv; 1292 unsigned int ncomp; 1293 uint64_t seq; 1294 int error; 1295 1296 /* Remote component is 1 for now. */ 1297 ncomp = 1; 1298 1299 for (;;) { 1300 /* Wait until there is anything to receive. */ 1301 mtx_lock(&hio_recv_list_lock[ncomp]); 1302 while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1303 pjdlog_debug(2, "remote_recv: No requests, waiting."); 1304 cv_wait(&hio_recv_list_cond[ncomp], 1305 &hio_recv_list_lock[ncomp]); 1306 } 1307 mtx_unlock(&hio_recv_list_lock[ncomp]); 1308 rw_rlock(&hio_remote_lock[ncomp]); 1309 if (!ISCONNECTED(res, ncomp)) { 1310 rw_unlock(&hio_remote_lock[ncomp]); 1311 /* 1312 * Connection is dead, so move all pending requests to 1313 * the done queue (one-by-one). 1314 */ 1315 mtx_lock(&hio_recv_list_lock[ncomp]); 1316 hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1317 assert(hio != NULL); 1318 TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1319 hio_next[ncomp]); 1320 mtx_unlock(&hio_recv_list_lock[ncomp]); 1321 goto done_queue; 1322 } 1323 if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1324 pjdlog_errno(LOG_ERR, 1325 "Unable to receive reply header"); 1326 rw_unlock(&hio_remote_lock[ncomp]); 1327 remote_close(res, ncomp); 1328 continue; 1329 } 1330 rw_unlock(&hio_remote_lock[ncomp]); 1331 seq = nv_get_uint64(nv, "seq"); 1332 if (seq == 0) { 1333 pjdlog_error("Header contains no 'seq' field."); 1334 nv_free(nv); 1335 continue; 1336 } 1337 mtx_lock(&hio_recv_list_lock[ncomp]); 1338 TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1339 if (hio->hio_ggio.gctl_seq == seq) { 1340 TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1341 hio_next[ncomp]); 1342 break; 1343 } 1344 } 1345 mtx_unlock(&hio_recv_list_lock[ncomp]); 1346 if (hio == NULL) { 1347 pjdlog_error("Found no request matching received 'seq' field (%ju).", 1348 (uintmax_t)seq); 1349 nv_free(nv); 1350 continue; 1351 } 1352 error = nv_get_int16(nv, "error"); 1353 if (error != 0) { 1354 /* Request failed on remote side. */ 1355 hio->hio_errors[ncomp] = 0; 1356 nv_free(nv); 1357 goto done_queue; 1358 } 1359 ggio = &hio->hio_ggio; 1360 switch (ggio->gctl_cmd) { 1361 case BIO_READ: 1362 rw_rlock(&hio_remote_lock[ncomp]); 1363 if (!ISCONNECTED(res, ncomp)) { 1364 rw_unlock(&hio_remote_lock[ncomp]); 1365 nv_free(nv); 1366 goto done_queue; 1367 } 1368 if (hast_proto_recv_data(res, res->hr_remotein, nv, 1369 ggio->gctl_data, ggio->gctl_length) < 0) { 1370 hio->hio_errors[ncomp] = errno; 1371 pjdlog_errno(LOG_ERR, 1372 "Unable to receive reply data"); 1373 rw_unlock(&hio_remote_lock[ncomp]); 1374 nv_free(nv); 1375 remote_close(res, ncomp); 1376 goto done_queue; 1377 } 1378 rw_unlock(&hio_remote_lock[ncomp]); 1379 break; 1380 case BIO_WRITE: 1381 case BIO_DELETE: 1382 case BIO_FLUSH: 1383 break; 1384 default: 1385 assert(!"invalid condition"); 1386 abort(); 1387 } 1388 hio->hio_errors[ncomp] = 0; 1389 nv_free(nv); 1390done_queue: 1391 if (refcount_release(&hio->hio_countdown)) { 1392 if (ISSYNCREQ(hio)) { 1393 mtx_lock(&sync_lock); 1394 SYNCREQDONE(hio); 1395 mtx_unlock(&sync_lock); 1396 cv_signal(&sync_cond); 1397 } else { 1398 pjdlog_debug(2, 1399 "remote_recv: (%p) Moving request to the done queue.", 1400 hio); 1401 QUEUE_INSERT2(hio, done); 1402 } 1403 } 1404 } 1405 /* NOTREACHED */ 1406 return (NULL); 1407} 1408 1409/* 1410 * Thread sends answer to the kernel. 1411 */ 1412static void * 1413ggate_send_thread(void *arg) 1414{ 1415 struct hast_resource *res = arg; 1416 struct g_gate_ctl_io *ggio; 1417 struct hio *hio; 1418 unsigned int ii, ncomp, ncomps; 1419 1420 ncomps = HAST_NCOMPONENTS; 1421 1422 for (;;) { 1423 pjdlog_debug(2, "ggate_send: Taking request."); 1424 QUEUE_TAKE2(hio, done); 1425 pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1426 ggio = &hio->hio_ggio; 1427 for (ii = 0; ii < ncomps; ii++) { 1428 if (hio->hio_errors[ii] == 0) { 1429 /* 1430 * One successful request is enough to declare 1431 * success. 1432 */ 1433 ggio->gctl_error = 0; 1434 break; 1435 } 1436 } 1437 if (ii == ncomps) { 1438 /* 1439 * None of the requests were successful. 1440 * Use first error. 1441 */ 1442 ggio->gctl_error = hio->hio_errors[0]; 1443 } 1444 if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1445 mtx_lock(&res->hr_amp_lock); 1446 activemap_write_complete(res->hr_amp, 1447 ggio->gctl_offset, ggio->gctl_length); 1448 mtx_unlock(&res->hr_amp_lock); 1449 } 1450 if (ggio->gctl_cmd == BIO_WRITE) { 1451 /* 1452 * Unlock range we locked. 1453 */ 1454 mtx_lock(&range_lock); 1455 rangelock_del(range_regular, ggio->gctl_offset, 1456 ggio->gctl_length); 1457 if (range_sync_wait) 1458 cv_signal(&range_sync_cond); 1459 mtx_unlock(&range_lock); 1460 /* 1461 * Bump local count if this is first write after 1462 * connection failure with remote node. 1463 */ 1464 ncomp = 1; 1465 rw_rlock(&hio_remote_lock[ncomp]); 1466 if (!ISCONNECTED(res, ncomp)) { 1467 mtx_lock(&metadata_lock); 1468 if (res->hr_primary_localcnt == 1469 res->hr_secondary_remotecnt) { 1470 res->hr_primary_localcnt++; 1471 pjdlog_debug(1, 1472 "Increasing localcnt to %ju.", 1473 (uintmax_t)res->hr_primary_localcnt); 1474 (void)metadata_write(res); 1475 } 1476 mtx_unlock(&metadata_lock); 1477 } 1478 rw_unlock(&hio_remote_lock[ncomp]); 1479 } 1480 if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1481 primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1482 pjdlog_debug(2, 1483 "ggate_send: (%p) Moving request to the free queue.", hio); 1484 QUEUE_INSERT2(hio, free); 1485 } 1486 /* NOTREACHED */ 1487 return (NULL); 1488} 1489 1490/* 1491 * Thread synchronize local and remote components. 1492 */ 1493static void * 1494sync_thread(void *arg __unused) 1495{ 1496 struct hast_resource *res = arg; 1497 struct hio *hio; 1498 struct g_gate_ctl_io *ggio; 1499 unsigned int ii, ncomp, ncomps; 1500 off_t offset, length, synced; 1501 bool dorewind; 1502 int syncext; 1503 1504 ncomps = HAST_NCOMPONENTS; 1505 dorewind = true; 1506 synced = 0; 1507 offset = -1; 1508 1509 for (;;) { 1510 mtx_lock(&sync_lock); 1511 if (offset >= 0 && !sync_inprogress) { 1512 pjdlog_info("Synchronization interrupted. " 1513 "%jd bytes synchronized so far.", 1514 (intmax_t)synced); 1515 hook_exec(res->hr_exec, "syncintr", res->hr_name, NULL); 1516 } 1517 while (!sync_inprogress) { 1518 dorewind = true; 1519 synced = 0; 1520 cv_wait(&sync_cond, &sync_lock); 1521 } 1522 mtx_unlock(&sync_lock); 1523 /* 1524 * Obtain offset at which we should synchronize. 1525 * Rewind synchronization if needed. 1526 */ 1527 mtx_lock(&res->hr_amp_lock); 1528 if (dorewind) 1529 activemap_sync_rewind(res->hr_amp); 1530 offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1531 if (syncext != -1) { 1532 /* 1533 * We synchronized entire syncext extent, we can mark 1534 * it as clean now. 1535 */ 1536 if (activemap_extent_complete(res->hr_amp, syncext)) 1537 (void)hast_activemap_flush(res); 1538 } 1539 mtx_unlock(&res->hr_amp_lock); 1540 if (dorewind) { 1541 dorewind = false; 1542 if (offset < 0) 1543 pjdlog_info("Nodes are in sync."); 1544 else { 1545 pjdlog_info("Synchronization started. %ju bytes to go.", 1546 (uintmax_t)(res->hr_extentsize * 1547 activemap_ndirty(res->hr_amp))); 1548 hook_exec(res->hr_exec, "syncstart", 1549 res->hr_name, NULL); 1550 } 1551 } 1552 if (offset < 0) { 1553 sync_stop(); 1554 pjdlog_debug(1, "Nothing to synchronize."); 1555 /* 1556 * Synchronization complete, make both localcnt and 1557 * remotecnt equal. 1558 */ 1559 ncomp = 1; 1560 rw_rlock(&hio_remote_lock[ncomp]); 1561 if (ISCONNECTED(res, ncomp)) { 1562 if (synced > 0) { 1563 pjdlog_info("Synchronization complete. " 1564 "%jd bytes synchronized.", 1565 (intmax_t)synced); 1566 hook_exec(res->hr_exec, "syncdone", 1567 res->hr_name, NULL); 1568 } 1569 mtx_lock(&metadata_lock); 1570 res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1571 res->hr_primary_localcnt = 1572 res->hr_secondary_localcnt; 1573 res->hr_primary_remotecnt = 1574 res->hr_secondary_remotecnt; 1575 pjdlog_debug(1, 1576 "Setting localcnt to %ju and remotecnt to %ju.", 1577 (uintmax_t)res->hr_primary_localcnt, 1578 (uintmax_t)res->hr_secondary_localcnt); 1579 (void)metadata_write(res); 1580 mtx_unlock(&metadata_lock); 1581 } 1582 rw_unlock(&hio_remote_lock[ncomp]); 1583 continue; 1584 } 1585 pjdlog_debug(2, "sync: Taking free request."); 1586 QUEUE_TAKE2(hio, free); 1587 pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1588 /* 1589 * Lock the range we are going to synchronize. We don't want 1590 * race where someone writes between our read and write. 1591 */ 1592 for (;;) { 1593 mtx_lock(&range_lock); 1594 if (rangelock_islocked(range_regular, offset, length)) { 1595 pjdlog_debug(2, 1596 "sync: Range offset=%jd length=%jd locked.", 1597 (intmax_t)offset, (intmax_t)length); 1598 range_sync_wait = true; 1599 cv_wait(&range_sync_cond, &range_lock); 1600 range_sync_wait = false; 1601 mtx_unlock(&range_lock); 1602 continue; 1603 } 1604 if (rangelock_add(range_sync, offset, length) < 0) { 1605 mtx_unlock(&range_lock); 1606 pjdlog_debug(2, 1607 "sync: Range offset=%jd length=%jd is already locked, waiting.", 1608 (intmax_t)offset, (intmax_t)length); 1609 sleep(1); 1610 continue; 1611 } 1612 mtx_unlock(&range_lock); 1613 break; 1614 } 1615 /* 1616 * First read the data from synchronization source. 1617 */ 1618 SYNCREQ(hio); 1619 ggio = &hio->hio_ggio; 1620 ggio->gctl_cmd = BIO_READ; 1621 ggio->gctl_offset = offset; 1622 ggio->gctl_length = length; 1623 ggio->gctl_error = 0; 1624 for (ii = 0; ii < ncomps; ii++) 1625 hio->hio_errors[ii] = EINVAL; 1626 reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1627 hio); 1628 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1629 hio); 1630 mtx_lock(&metadata_lock); 1631 if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1632 /* 1633 * This range is up-to-date on local component, 1634 * so handle request locally. 1635 */ 1636 /* Local component is 0 for now. */ 1637 ncomp = 0; 1638 } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1639 assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1640 /* 1641 * This range is out-of-date on local component, 1642 * so send request to the remote node. 1643 */ 1644 /* Remote component is 1 for now. */ 1645 ncomp = 1; 1646 } 1647 mtx_unlock(&metadata_lock); 1648 refcount_init(&hio->hio_countdown, 1); 1649 QUEUE_INSERT1(hio, send, ncomp); 1650 1651 /* 1652 * Let's wait for READ to finish. 1653 */ 1654 mtx_lock(&sync_lock); 1655 while (!ISSYNCREQDONE(hio)) 1656 cv_wait(&sync_cond, &sync_lock); 1657 mtx_unlock(&sync_lock); 1658 1659 if (hio->hio_errors[ncomp] != 0) { 1660 pjdlog_error("Unable to read synchronization data: %s.", 1661 strerror(hio->hio_errors[ncomp])); 1662 goto free_queue; 1663 } 1664 1665 /* 1666 * We read the data from synchronization source, now write it 1667 * to synchronization target. 1668 */ 1669 SYNCREQ(hio); 1670 ggio->gctl_cmd = BIO_WRITE; 1671 for (ii = 0; ii < ncomps; ii++) 1672 hio->hio_errors[ii] = EINVAL; 1673 reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1674 hio); 1675 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1676 hio); 1677 mtx_lock(&metadata_lock); 1678 if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1679 /* 1680 * This range is up-to-date on local component, 1681 * so we update remote component. 1682 */ 1683 /* Remote component is 1 for now. */ 1684 ncomp = 1; 1685 } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1686 assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1687 /* 1688 * This range is out-of-date on local component, 1689 * so we update it. 1690 */ 1691 /* Local component is 0 for now. */ 1692 ncomp = 0; 1693 } 1694 mtx_unlock(&metadata_lock); 1695 1696 pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1697 hio); 1698 refcount_init(&hio->hio_countdown, 1); 1699 QUEUE_INSERT1(hio, send, ncomp); 1700 1701 /* 1702 * Let's wait for WRITE to finish. 1703 */ 1704 mtx_lock(&sync_lock); 1705 while (!ISSYNCREQDONE(hio)) 1706 cv_wait(&sync_cond, &sync_lock); 1707 mtx_unlock(&sync_lock); 1708 1709 if (hio->hio_errors[ncomp] != 0) { 1710 pjdlog_error("Unable to write synchronization data: %s.", 1711 strerror(hio->hio_errors[ncomp])); 1712 goto free_queue; 1713 } 1714 1715 synced += length; 1716free_queue: 1717 mtx_lock(&range_lock); 1718 rangelock_del(range_sync, offset, length); 1719 if (range_regular_wait) 1720 cv_signal(&range_regular_cond); 1721 mtx_unlock(&range_lock); 1722 pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1723 hio); 1724 QUEUE_INSERT2(hio, free); 1725 } 1726 /* NOTREACHED */ 1727 return (NULL); 1728} 1729 1730static void 1731config_reload(void) 1732{ 1733 struct hastd_config *newcfg; 1734 struct hast_resource *res; 1735 unsigned int ii, ncomps; 1736 int modified; 1737 1738 pjdlog_info("Reloading configuration..."); 1739 1740 ncomps = HAST_NCOMPONENTS; 1741 1742 newcfg = yy_config_parse(cfgpath, false); 1743 if (newcfg == NULL) 1744 goto failed; 1745 1746 TAILQ_FOREACH(res, &newcfg->hc_resources, hr_next) { 1747 if (strcmp(res->hr_name, gres->hr_name) == 0) 1748 break; 1749 } 1750 /* 1751 * If resource was removed from the configuration file, resource 1752 * name, provider name or path to local component was modified we 1753 * shouldn't be here. This means that someone modified configuration 1754 * file and send SIGHUP to us instead of main hastd process. 1755 * Log advice and ignore the signal. 1756 */ 1757 if (res == NULL || strcmp(gres->hr_name, res->hr_name) != 0 || 1758 strcmp(gres->hr_provname, res->hr_provname) != 0 || 1759 strcmp(gres->hr_localpath, res->hr_localpath) != 0) { 1760 pjdlog_warning("To reload configuration send SIGHUP to the main hastd process (pid %u).", 1761 (unsigned int)getppid()); 1762 goto failed; 1763 } 1764 1765#define MODIFIED_REMOTEADDR 0x1 1766#define MODIFIED_REPLICATION 0x2 1767#define MODIFIED_TIMEOUT 0x4 1768#define MODIFIED_EXEC 0x8 1769 modified = 0; 1770 if (strcmp(gres->hr_remoteaddr, res->hr_remoteaddr) != 0) { 1771 /* 1772 * Don't copy res->hr_remoteaddr to gres just yet. 1773 * We want remote_close() to log disconnect from the old 1774 * addresses, not from the new ones. 1775 */ 1776 modified |= MODIFIED_REMOTEADDR; 1777 } 1778 if (gres->hr_replication != res->hr_replication) { 1779 gres->hr_replication = res->hr_replication; 1780 modified |= MODIFIED_REPLICATION; 1781 } 1782 if (gres->hr_timeout != res->hr_timeout) { 1783 gres->hr_timeout = res->hr_timeout; 1784 modified |= MODIFIED_TIMEOUT; 1785 } 1786 if (strcmp(gres->hr_exec, res->hr_exec) != 0) { 1787 strlcpy(gres->hr_exec, res->hr_exec, sizeof(gres->hr_exec)); 1788 modified |= MODIFIED_EXEC; 1789 } 1790 /* 1791 * If only timeout was modified we only need to change it without 1792 * reconnecting. 1793 */ 1794 if (modified == MODIFIED_TIMEOUT) { 1795 for (ii = 0; ii < ncomps; ii++) { 1796 if (!ISREMOTE(ii)) 1797 continue; 1798 rw_rlock(&hio_remote_lock[ii]); 1799 if (!ISCONNECTED(gres, ii)) { 1800 rw_unlock(&hio_remote_lock[ii]); 1801 continue; 1802 } 1803 rw_unlock(&hio_remote_lock[ii]); 1804 if (proto_timeout(gres->hr_remotein, 1805 gres->hr_timeout) < 0) { 1806 pjdlog_errno(LOG_WARNING, 1807 "Unable to set connection timeout"); 1808 } 1809 if (proto_timeout(gres->hr_remoteout, 1810 gres->hr_timeout) < 0) { 1811 pjdlog_errno(LOG_WARNING, 1812 "Unable to set connection timeout"); 1813 } 1814 } 1815 } else if ((modified & 1816 (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1817 for (ii = 0; ii < ncomps; ii++) { 1818 if (!ISREMOTE(ii)) 1819 continue; 1820 remote_close(gres, ii); 1821 } 1822 if (modified & MODIFIED_REMOTEADDR) { 1823 strlcpy(gres->hr_remoteaddr, res->hr_remoteaddr, 1824 sizeof(gres->hr_remoteaddr)); 1825 } 1826 } 1827#undef MODIFIED_REMOTEADDR 1828#undef MODIFIED_REPLICATION 1829#undef MODIFIED_TIMEOUT 1830#undef MODIFIED_EXEC 1831 1832 pjdlog_info("Configuration reloaded successfully."); 1833 return; 1834failed: 1835 if (newcfg != NULL) { 1836 if (newcfg->hc_controlconn != NULL) 1837 proto_close(newcfg->hc_controlconn); 1838 if (newcfg->hc_listenconn != NULL) 1839 proto_close(newcfg->hc_listenconn); 1840 yy_config_free(newcfg); 1841 } 1842 pjdlog_warning("Configuration not reloaded."); 1843} 1844 1845static void 1846keepalive_send(struct hast_resource *res, unsigned int ncomp) 1847{ 1848 struct nv *nv; 1849 1850 nv = nv_alloc(); 1851 nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1852 if (nv_error(nv) != 0) { 1853 nv_free(nv); 1854 pjdlog_debug(1, 1855 "keepalive_send: Unable to prepare header to send."); 1856 return; 1857 } 1858 if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1859 pjdlog_common(LOG_DEBUG, 1, errno, 1860 "keepalive_send: Unable to send request"); 1861 nv_free(nv); 1862 rw_unlock(&hio_remote_lock[ncomp]); 1863 remote_close(res, ncomp); 1864 rw_rlock(&hio_remote_lock[ncomp]); 1865 return; 1866 } 1867 nv_free(nv); 1868 pjdlog_debug(2, "keepalive_send: Request sent."); 1869} 1870 1871static void 1872guard_one(struct hast_resource *res, unsigned int ncomp) 1873{ 1874 struct proto_conn *in, *out; 1875 1876 if (!ISREMOTE(ncomp)) 1877 return; 1878 1879 rw_rlock(&hio_remote_lock[ncomp]); 1880 1881 if (!real_remote(res)) { 1882 rw_unlock(&hio_remote_lock[ncomp]); 1883 return; 1884 } 1885 1886 if (ISCONNECTED(res, ncomp)) { 1887 assert(res->hr_remotein != NULL); 1888 assert(res->hr_remoteout != NULL); 1889 keepalive_send(res, ncomp); 1890 } 1891 1892 if (ISCONNECTED(res, ncomp)) { 1893 assert(res->hr_remotein != NULL); 1894 assert(res->hr_remoteout != NULL); 1895 rw_unlock(&hio_remote_lock[ncomp]); 1896 pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 1897 res->hr_remoteaddr); 1898 return; 1899 } 1900 1901 assert(res->hr_remotein == NULL); 1902 assert(res->hr_remoteout == NULL); 1903 /* 1904 * Upgrade the lock. It doesn't have to be atomic as no other thread 1905 * can change connection status from disconnected to connected. 1906 */ 1907 rw_unlock(&hio_remote_lock[ncomp]); 1908 pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 1909 res->hr_remoteaddr); 1910 in = out = NULL; 1911 if (init_remote(res, &in, &out)) { 1912 rw_wlock(&hio_remote_lock[ncomp]); 1913 assert(res->hr_remotein == NULL); 1914 assert(res->hr_remoteout == NULL); 1915 assert(in != NULL && out != NULL); 1916 res->hr_remotein = in; 1917 res->hr_remoteout = out; 1918 rw_unlock(&hio_remote_lock[ncomp]); 1919 pjdlog_info("Successfully reconnected to %s.", 1920 res->hr_remoteaddr); 1921 sync_start(); 1922 } else { 1923 /* Both connections should be NULL. */ 1924 assert(res->hr_remotein == NULL); 1925 assert(res->hr_remoteout == NULL); 1926 assert(in == NULL && out == NULL); 1927 pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 1928 res->hr_remoteaddr); 1929 } 1930} 1931 1932/* 1933 * Thread guards remote connections and reconnects when needed, handles 1934 * signals, etc. 1935 */ 1936static void * 1937guard_thread(void *arg) 1938{ 1939 struct hast_resource *res = arg; 1940 unsigned int ii, ncomps; 1941 struct timespec timeout; 1942 time_t lastcheck, now; 1943 sigset_t mask; 1944 int signo; 1945 1946 ncomps = HAST_NCOMPONENTS; 1947 lastcheck = time(NULL); 1948 1949 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1950 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1951 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1952 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1953 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1954 1955 timeout.tv_nsec = 0; 1956 signo = -1; 1957 1958 for (;;) { 1959 switch (signo) { 1960 case SIGHUP: 1961 config_reload(); 1962 break; 1963 case SIGINT: 1964 case SIGTERM: 1965 sigexit_received = true; 1966 primary_exitx(EX_OK, 1967 "Termination signal received, exiting."); 1968 break; 1969 default: 1970 break; 1971 } 1972 hook_check(signo == SIGCHLD); 1973 1974 pjdlog_debug(2, "remote_guard: Checking connections."); 1975 now = time(NULL); 1976 if (lastcheck + RETRY_SLEEP <= now) { 1977 for (ii = 0; ii < ncomps; ii++) 1978 guard_one(res, ii); 1979 lastcheck = now; 1980 } 1981 timeout.tv_sec = RETRY_SLEEP; 1982 signo = sigtimedwait(&mask, NULL, &timeout); 1983 } 1984 /* NOTREACHED */ 1985 return (NULL); 1986} 1987