primary.c revision 211982
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 211982 2010-08-30 00:06:05Z 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; 501 502 /* Prepare outgoing connection with remote node. */ 503 if (proto_client(res->hr_remoteaddr, &out) < 0) { 504 primary_exit(EX_TEMPFAIL, "Unable to create connection to %s", 505 res->hr_remoteaddr); 506 } 507 /* Try to connect, but accept failure. */ 508 if (proto_connect(out) < 0) { 509 pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 510 res->hr_remoteaddr); 511 goto close; 512 } 513 /* Error in setting timeout is not critical, but why should it fail? */ 514 if (proto_timeout(out, res->hr_timeout) < 0) 515 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 516 /* 517 * First handshake step. 518 * Setup outgoing connection with remote node. 519 */ 520 nvout = nv_alloc(); 521 nv_add_string(nvout, res->hr_name, "resource"); 522 if (nv_error(nvout) != 0) { 523 pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 524 "Unable to allocate header for connection with %s", 525 res->hr_remoteaddr); 526 nv_free(nvout); 527 goto close; 528 } 529 if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 530 pjdlog_errno(LOG_WARNING, 531 "Unable to send handshake header to %s", 532 res->hr_remoteaddr); 533 nv_free(nvout); 534 goto close; 535 } 536 nv_free(nvout); 537 if (hast_proto_recv_hdr(out, &nvin) < 0) { 538 pjdlog_errno(LOG_WARNING, 539 "Unable to receive handshake header from %s", 540 res->hr_remoteaddr); 541 goto close; 542 } 543 errmsg = nv_get_string(nvin, "errmsg"); 544 if (errmsg != NULL) { 545 pjdlog_warning("%s", errmsg); 546 nv_free(nvin); 547 goto close; 548 } 549 token = nv_get_uint8_array(nvin, &size, "token"); 550 if (token == NULL) { 551 pjdlog_warning("Handshake header from %s has no 'token' field.", 552 res->hr_remoteaddr); 553 nv_free(nvin); 554 goto close; 555 } 556 if (size != sizeof(res->hr_token)) { 557 pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 558 res->hr_remoteaddr, size, sizeof(res->hr_token)); 559 nv_free(nvin); 560 goto close; 561 } 562 bcopy(token, res->hr_token, sizeof(res->hr_token)); 563 nv_free(nvin); 564 565 /* 566 * Second handshake step. 567 * Setup incoming connection with remote node. 568 */ 569 if (proto_client(res->hr_remoteaddr, &in) < 0) { 570 pjdlog_errno(LOG_WARNING, "Unable to create connection to %s", 571 res->hr_remoteaddr); 572 } 573 /* Try to connect, but accept failure. */ 574 if (proto_connect(in) < 0) { 575 pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 576 res->hr_remoteaddr); 577 goto close; 578 } 579 /* Error in setting timeout is not critical, but why should it fail? */ 580 if (proto_timeout(in, res->hr_timeout) < 0) 581 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 582 nvout = nv_alloc(); 583 nv_add_string(nvout, res->hr_name, "resource"); 584 nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 585 "token"); 586 nv_add_uint64(nvout, res->hr_resuid, "resuid"); 587 nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 588 nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 589 if (nv_error(nvout) != 0) { 590 pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 591 "Unable to allocate header for connection with %s", 592 res->hr_remoteaddr); 593 nv_free(nvout); 594 goto close; 595 } 596 if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 597 pjdlog_errno(LOG_WARNING, 598 "Unable to send handshake header to %s", 599 res->hr_remoteaddr); 600 nv_free(nvout); 601 goto close; 602 } 603 nv_free(nvout); 604 if (hast_proto_recv_hdr(out, &nvin) < 0) { 605 pjdlog_errno(LOG_WARNING, 606 "Unable to receive handshake header from %s", 607 res->hr_remoteaddr); 608 goto close; 609 } 610 errmsg = nv_get_string(nvin, "errmsg"); 611 if (errmsg != NULL) { 612 pjdlog_warning("%s", errmsg); 613 nv_free(nvin); 614 goto close; 615 } 616 datasize = nv_get_int64(nvin, "datasize"); 617 if (datasize != res->hr_datasize) { 618 pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 619 (intmax_t)res->hr_datasize, (intmax_t)datasize); 620 nv_free(nvin); 621 goto close; 622 } 623 extentsize = nv_get_int32(nvin, "extentsize"); 624 if (extentsize != res->hr_extentsize) { 625 pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 626 (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 627 nv_free(nvin); 628 goto close; 629 } 630 res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 631 res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 632 res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 633 map = NULL; 634 mapsize = nv_get_uint32(nvin, "mapsize"); 635 if (mapsize > 0) { 636 map = malloc(mapsize); 637 if (map == NULL) { 638 pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 639 (uintmax_t)mapsize); 640 nv_free(nvin); 641 goto close; 642 } 643 /* 644 * Remote node have some dirty extents on its own, lets 645 * download its activemap. 646 */ 647 if (hast_proto_recv_data(res, out, nvin, map, 648 mapsize) < 0) { 649 pjdlog_errno(LOG_ERR, 650 "Unable to receive remote activemap"); 651 nv_free(nvin); 652 free(map); 653 goto close; 654 } 655 /* 656 * Merge local and remote bitmaps. 657 */ 658 activemap_merge(res->hr_amp, map, mapsize); 659 free(map); 660 /* 661 * Now that we merged bitmaps from both nodes, flush it to the 662 * disk before we start to synchronize. 663 */ 664 (void)hast_activemap_flush(res); 665 } 666 pjdlog_info("Connected to %s.", res->hr_remoteaddr); 667 if (inp != NULL && outp != NULL) { 668 *inp = in; 669 *outp = out; 670 } else { 671 res->hr_remotein = in; 672 res->hr_remoteout = out; 673 } 674 return (true); 675close: 676 proto_close(out); 677 if (in != NULL) 678 proto_close(in); 679 return (false); 680} 681 682static void 683sync_start(void) 684{ 685 686 mtx_lock(&sync_lock); 687 sync_inprogress = true; 688 mtx_unlock(&sync_lock); 689 cv_signal(&sync_cond); 690} 691 692static void 693sync_stop(void) 694{ 695 696 mtx_lock(&sync_lock); 697 if (sync_inprogress) 698 sync_inprogress = false; 699 mtx_unlock(&sync_lock); 700} 701 702static void 703init_ggate(struct hast_resource *res) 704{ 705 struct g_gate_ctl_create ggiocreate; 706 struct g_gate_ctl_cancel ggiocancel; 707 708 /* 709 * We communicate with ggate via /dev/ggctl. Open it. 710 */ 711 res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 712 if (res->hr_ggatefd < 0) 713 primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 714 /* 715 * Create provider before trying to connect, as connection failure 716 * is not critical, but may take some time. 717 */ 718 ggiocreate.gctl_version = G_GATE_VERSION; 719 ggiocreate.gctl_mediasize = res->hr_datasize; 720 ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 721 ggiocreate.gctl_flags = 0; 722 ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 723 ggiocreate.gctl_timeout = 0; 724 ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 725 snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 726 res->hr_provname); 727 bzero(ggiocreate.gctl_info, sizeof(ggiocreate.gctl_info)); 728 if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 729 pjdlog_info("Device hast/%s created.", res->hr_provname); 730 res->hr_ggateunit = ggiocreate.gctl_unit; 731 return; 732 } 733 if (errno != EEXIST) { 734 primary_exit(EX_OSERR, "Unable to create hast/%s device", 735 res->hr_provname); 736 } 737 pjdlog_debug(1, 738 "Device hast/%s already exists, we will try to take it over.", 739 res->hr_provname); 740 /* 741 * If we received EEXIST, we assume that the process who created the 742 * provider died and didn't clean up. In that case we will start from 743 * where he left of. 744 */ 745 ggiocancel.gctl_version = G_GATE_VERSION; 746 ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 747 snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 748 res->hr_provname); 749 if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 750 pjdlog_info("Device hast/%s recovered.", res->hr_provname); 751 res->hr_ggateunit = ggiocancel.gctl_unit; 752 return; 753 } 754 primary_exit(EX_OSERR, "Unable to take over hast/%s device", 755 res->hr_provname); 756} 757 758void 759hastd_primary(struct hast_resource *res) 760{ 761 pthread_t td; 762 pid_t pid; 763 int error; 764 765 gres = res; 766 767 /* 768 * Create communication channel between parent and child. 769 */ 770 if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 771 KEEP_ERRNO((void)pidfile_remove(pfh)); 772 primary_exit(EX_OSERR, 773 "Unable to create control sockets between parent and child"); 774 } 775 776 pid = fork(); 777 if (pid < 0) { 778 KEEP_ERRNO((void)pidfile_remove(pfh)); 779 primary_exit(EX_TEMPFAIL, "Unable to fork"); 780 } 781 782 if (pid > 0) { 783 /* This is parent. */ 784 res->hr_workerpid = pid; 785 return; 786 } 787 788 (void)pidfile_close(pfh); 789 hook_fini(); 790 791 setproctitle("%s (primary)", res->hr_name); 792 793 signal(SIGHUP, SIG_DFL); 794 signal(SIGCHLD, SIG_DFL); 795 796 hook_init(); 797 init_local(res); 798 if (real_remote(res) && init_remote(res, NULL, NULL)) 799 sync_start(); 800 init_ggate(res); 801 init_environment(res); 802 error = pthread_create(&td, NULL, ggate_recv_thread, res); 803 assert(error == 0); 804 error = pthread_create(&td, NULL, local_send_thread, res); 805 assert(error == 0); 806 error = pthread_create(&td, NULL, remote_send_thread, res); 807 assert(error == 0); 808 error = pthread_create(&td, NULL, remote_recv_thread, res); 809 assert(error == 0); 810 error = pthread_create(&td, NULL, ggate_send_thread, res); 811 assert(error == 0); 812 error = pthread_create(&td, NULL, sync_thread, res); 813 assert(error == 0); 814 error = pthread_create(&td, NULL, ctrl_thread, res); 815 assert(error == 0); 816 (void)guard_thread(res); 817} 818 819static void 820reqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 821{ 822 char msg[1024]; 823 va_list ap; 824 int len; 825 826 va_start(ap, fmt); 827 len = vsnprintf(msg, sizeof(msg), fmt, ap); 828 va_end(ap); 829 if ((size_t)len < sizeof(msg)) { 830 switch (ggio->gctl_cmd) { 831 case BIO_READ: 832 (void)snprintf(msg + len, sizeof(msg) - len, 833 "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 834 (uintmax_t)ggio->gctl_length); 835 break; 836 case BIO_DELETE: 837 (void)snprintf(msg + len, sizeof(msg) - len, 838 "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 839 (uintmax_t)ggio->gctl_length); 840 break; 841 case BIO_FLUSH: 842 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 843 break; 844 case BIO_WRITE: 845 (void)snprintf(msg + len, sizeof(msg) - len, 846 "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 847 (uintmax_t)ggio->gctl_length); 848 break; 849 default: 850 (void)snprintf(msg + len, sizeof(msg) - len, 851 "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 852 break; 853 } 854 } 855 pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 856} 857 858static void 859remote_close(struct hast_resource *res, int ncomp) 860{ 861 862 rw_wlock(&hio_remote_lock[ncomp]); 863 /* 864 * A race is possible between dropping rlock and acquiring wlock - 865 * another thread can close connection in-between. 866 */ 867 if (!ISCONNECTED(res, ncomp)) { 868 assert(res->hr_remotein == NULL); 869 assert(res->hr_remoteout == NULL); 870 rw_unlock(&hio_remote_lock[ncomp]); 871 return; 872 } 873 874 assert(res->hr_remotein != NULL); 875 assert(res->hr_remoteout != NULL); 876 877 pjdlog_debug(2, "Closing incoming connection to %s.", 878 res->hr_remoteaddr); 879 proto_close(res->hr_remotein); 880 res->hr_remotein = NULL; 881 pjdlog_debug(2, "Closing outgoing connection to %s.", 882 res->hr_remoteaddr); 883 proto_close(res->hr_remoteout); 884 res->hr_remoteout = NULL; 885 886 rw_unlock(&hio_remote_lock[ncomp]); 887 888 pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 889 890 /* 891 * Stop synchronization if in-progress. 892 */ 893 sync_stop(); 894} 895 896/* 897 * Thread receives ggate I/O requests from the kernel and passes them to 898 * appropriate threads: 899 * WRITE - always goes to both local_send and remote_send threads 900 * READ (when the block is up-to-date on local component) - 901 * only local_send thread 902 * READ (when the block isn't up-to-date on local component) - 903 * only remote_send thread 904 * DELETE - always goes to both local_send and remote_send threads 905 * FLUSH - always goes to both local_send and remote_send threads 906 */ 907static void * 908ggate_recv_thread(void *arg) 909{ 910 struct hast_resource *res = arg; 911 struct g_gate_ctl_io *ggio; 912 struct hio *hio; 913 unsigned int ii, ncomp, ncomps; 914 int error; 915 916 ncomps = HAST_NCOMPONENTS; 917 918 for (;;) { 919 pjdlog_debug(2, "ggate_recv: Taking free request."); 920 QUEUE_TAKE2(hio, free); 921 pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 922 ggio = &hio->hio_ggio; 923 ggio->gctl_unit = res->hr_ggateunit; 924 ggio->gctl_length = MAXPHYS; 925 ggio->gctl_error = 0; 926 pjdlog_debug(2, 927 "ggate_recv: (%p) Waiting for request from the kernel.", 928 hio); 929 if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 930 if (sigexit_received) 931 pthread_exit(NULL); 932 primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 933 } 934 error = ggio->gctl_error; 935 switch (error) { 936 case 0: 937 break; 938 case ECANCELED: 939 /* Exit gracefully. */ 940 if (!sigexit_received) { 941 pjdlog_debug(2, 942 "ggate_recv: (%p) Received cancel from the kernel.", 943 hio); 944 pjdlog_info("Received cancel from the kernel, exiting."); 945 } 946 pthread_exit(NULL); 947 case ENOMEM: 948 /* 949 * Buffer too small? Impossible, we allocate MAXPHYS 950 * bytes - request can't be bigger than that. 951 */ 952 /* FALLTHROUGH */ 953 case ENXIO: 954 default: 955 primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 956 strerror(error)); 957 } 958 for (ii = 0; ii < ncomps; ii++) 959 hio->hio_errors[ii] = EINVAL; 960 reqlog(LOG_DEBUG, 2, ggio, 961 "ggate_recv: (%p) Request received from the kernel: ", 962 hio); 963 /* 964 * Inform all components about new write request. 965 * For read request prefer local component unless the given 966 * range is out-of-date, then use remote component. 967 */ 968 switch (ggio->gctl_cmd) { 969 case BIO_READ: 970 pjdlog_debug(2, 971 "ggate_recv: (%p) Moving request to the send queue.", 972 hio); 973 refcount_init(&hio->hio_countdown, 1); 974 mtx_lock(&metadata_lock); 975 if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 976 res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 977 /* 978 * This range is up-to-date on local component, 979 * so handle request locally. 980 */ 981 /* Local component is 0 for now. */ 982 ncomp = 0; 983 } else /* if (res->hr_syncsrc == 984 HAST_SYNCSRC_SECONDARY) */ { 985 assert(res->hr_syncsrc == 986 HAST_SYNCSRC_SECONDARY); 987 /* 988 * This range is out-of-date on local component, 989 * so send request to the remote node. 990 */ 991 /* Remote component is 1 for now. */ 992 ncomp = 1; 993 } 994 mtx_unlock(&metadata_lock); 995 QUEUE_INSERT1(hio, send, ncomp); 996 break; 997 case BIO_WRITE: 998 for (;;) { 999 mtx_lock(&range_lock); 1000 if (rangelock_islocked(range_sync, 1001 ggio->gctl_offset, ggio->gctl_length)) { 1002 pjdlog_debug(2, 1003 "regular: Range offset=%jd length=%zu locked.", 1004 (intmax_t)ggio->gctl_offset, 1005 (size_t)ggio->gctl_length); 1006 range_regular_wait = true; 1007 cv_wait(&range_regular_cond, &range_lock); 1008 range_regular_wait = false; 1009 mtx_unlock(&range_lock); 1010 continue; 1011 } 1012 if (rangelock_add(range_regular, 1013 ggio->gctl_offset, ggio->gctl_length) < 0) { 1014 mtx_unlock(&range_lock); 1015 pjdlog_debug(2, 1016 "regular: Range offset=%jd length=%zu is already locked, waiting.", 1017 (intmax_t)ggio->gctl_offset, 1018 (size_t)ggio->gctl_length); 1019 sleep(1); 1020 continue; 1021 } 1022 mtx_unlock(&range_lock); 1023 break; 1024 } 1025 mtx_lock(&res->hr_amp_lock); 1026 if (activemap_write_start(res->hr_amp, 1027 ggio->gctl_offset, ggio->gctl_length)) { 1028 (void)hast_activemap_flush(res); 1029 } 1030 mtx_unlock(&res->hr_amp_lock); 1031 /* FALLTHROUGH */ 1032 case BIO_DELETE: 1033 case BIO_FLUSH: 1034 pjdlog_debug(2, 1035 "ggate_recv: (%p) Moving request to the send queues.", 1036 hio); 1037 refcount_init(&hio->hio_countdown, ncomps); 1038 for (ii = 0; ii < ncomps; ii++) 1039 QUEUE_INSERT1(hio, send, ii); 1040 break; 1041 } 1042 } 1043 /* NOTREACHED */ 1044 return (NULL); 1045} 1046 1047/* 1048 * Thread reads from or writes to local component. 1049 * If local read fails, it redirects it to remote_send thread. 1050 */ 1051static void * 1052local_send_thread(void *arg) 1053{ 1054 struct hast_resource *res = arg; 1055 struct g_gate_ctl_io *ggio; 1056 struct hio *hio; 1057 unsigned int ncomp, rncomp; 1058 ssize_t ret; 1059 1060 /* Local component is 0 for now. */ 1061 ncomp = 0; 1062 /* Remote component is 1 for now. */ 1063 rncomp = 1; 1064 1065 for (;;) { 1066 pjdlog_debug(2, "local_send: Taking request."); 1067 QUEUE_TAKE1(hio, send, ncomp); 1068 pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1069 ggio = &hio->hio_ggio; 1070 switch (ggio->gctl_cmd) { 1071 case BIO_READ: 1072 ret = pread(res->hr_localfd, ggio->gctl_data, 1073 ggio->gctl_length, 1074 ggio->gctl_offset + res->hr_localoff); 1075 if (ret == ggio->gctl_length) 1076 hio->hio_errors[ncomp] = 0; 1077 else { 1078 /* 1079 * If READ failed, try to read from remote node. 1080 */ 1081 QUEUE_INSERT1(hio, send, rncomp); 1082 continue; 1083 } 1084 break; 1085 case BIO_WRITE: 1086 ret = pwrite(res->hr_localfd, ggio->gctl_data, 1087 ggio->gctl_length, 1088 ggio->gctl_offset + res->hr_localoff); 1089 if (ret < 0) 1090 hio->hio_errors[ncomp] = errno; 1091 else if (ret != ggio->gctl_length) 1092 hio->hio_errors[ncomp] = EIO; 1093 else 1094 hio->hio_errors[ncomp] = 0; 1095 break; 1096 case BIO_DELETE: 1097 ret = g_delete(res->hr_localfd, 1098 ggio->gctl_offset + res->hr_localoff, 1099 ggio->gctl_length); 1100 if (ret < 0) 1101 hio->hio_errors[ncomp] = errno; 1102 else 1103 hio->hio_errors[ncomp] = 0; 1104 break; 1105 case BIO_FLUSH: 1106 ret = g_flush(res->hr_localfd); 1107 if (ret < 0) 1108 hio->hio_errors[ncomp] = errno; 1109 else 1110 hio->hio_errors[ncomp] = 0; 1111 break; 1112 } 1113 if (refcount_release(&hio->hio_countdown)) { 1114 if (ISSYNCREQ(hio)) { 1115 mtx_lock(&sync_lock); 1116 SYNCREQDONE(hio); 1117 mtx_unlock(&sync_lock); 1118 cv_signal(&sync_cond); 1119 } else { 1120 pjdlog_debug(2, 1121 "local_send: (%p) Moving request to the done queue.", 1122 hio); 1123 QUEUE_INSERT2(hio, done); 1124 } 1125 } 1126 } 1127 /* NOTREACHED */ 1128 return (NULL); 1129} 1130 1131/* 1132 * Thread sends request to secondary node. 1133 */ 1134static void * 1135remote_send_thread(void *arg) 1136{ 1137 struct hast_resource *res = arg; 1138 struct g_gate_ctl_io *ggio; 1139 struct hio *hio; 1140 struct nv *nv; 1141 unsigned int ncomp; 1142 bool wakeup; 1143 uint64_t offset, length; 1144 uint8_t cmd; 1145 void *data; 1146 1147 /* Remote component is 1 for now. */ 1148 ncomp = 1; 1149 1150 for (;;) { 1151 pjdlog_debug(2, "remote_send: Taking request."); 1152 QUEUE_TAKE1(hio, send, ncomp); 1153 pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1154 ggio = &hio->hio_ggio; 1155 switch (ggio->gctl_cmd) { 1156 case BIO_READ: 1157 cmd = HIO_READ; 1158 data = NULL; 1159 offset = ggio->gctl_offset; 1160 length = ggio->gctl_length; 1161 break; 1162 case BIO_WRITE: 1163 cmd = HIO_WRITE; 1164 data = ggio->gctl_data; 1165 offset = ggio->gctl_offset; 1166 length = ggio->gctl_length; 1167 break; 1168 case BIO_DELETE: 1169 cmd = HIO_DELETE; 1170 data = NULL; 1171 offset = ggio->gctl_offset; 1172 length = ggio->gctl_length; 1173 break; 1174 case BIO_FLUSH: 1175 cmd = HIO_FLUSH; 1176 data = NULL; 1177 offset = 0; 1178 length = 0; 1179 break; 1180 default: 1181 assert(!"invalid condition"); 1182 abort(); 1183 } 1184 nv = nv_alloc(); 1185 nv_add_uint8(nv, cmd, "cmd"); 1186 nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1187 nv_add_uint64(nv, offset, "offset"); 1188 nv_add_uint64(nv, length, "length"); 1189 if (nv_error(nv) != 0) { 1190 hio->hio_errors[ncomp] = nv_error(nv); 1191 pjdlog_debug(2, 1192 "remote_send: (%p) Unable to prepare header to send.", 1193 hio); 1194 reqlog(LOG_ERR, 0, ggio, 1195 "Unable to prepare header to send (%s): ", 1196 strerror(nv_error(nv))); 1197 /* Move failed request immediately to the done queue. */ 1198 goto done_queue; 1199 } 1200 pjdlog_debug(2, 1201 "remote_send: (%p) Moving request to the recv queue.", 1202 hio); 1203 /* 1204 * Protect connection from disappearing. 1205 */ 1206 rw_rlock(&hio_remote_lock[ncomp]); 1207 if (!ISCONNECTED(res, ncomp)) { 1208 rw_unlock(&hio_remote_lock[ncomp]); 1209 hio->hio_errors[ncomp] = ENOTCONN; 1210 goto done_queue; 1211 } 1212 /* 1213 * Move the request to recv queue before sending it, because 1214 * in different order we can get reply before we move request 1215 * to recv queue. 1216 */ 1217 mtx_lock(&hio_recv_list_lock[ncomp]); 1218 wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1219 TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1220 mtx_unlock(&hio_recv_list_lock[ncomp]); 1221 if (hast_proto_send(res, res->hr_remoteout, nv, data, 1222 data != NULL ? length : 0) < 0) { 1223 hio->hio_errors[ncomp] = errno; 1224 rw_unlock(&hio_remote_lock[ncomp]); 1225 pjdlog_debug(2, 1226 "remote_send: (%p) Unable to send request.", hio); 1227 reqlog(LOG_ERR, 0, ggio, 1228 "Unable to send request (%s): ", 1229 strerror(hio->hio_errors[ncomp])); 1230 remote_close(res, ncomp); 1231 /* 1232 * Take request back from the receive queue and move 1233 * it immediately to the done queue. 1234 */ 1235 mtx_lock(&hio_recv_list_lock[ncomp]); 1236 TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1237 mtx_unlock(&hio_recv_list_lock[ncomp]); 1238 goto done_queue; 1239 } 1240 rw_unlock(&hio_remote_lock[ncomp]); 1241 nv_free(nv); 1242 if (wakeup) 1243 cv_signal(&hio_recv_list_cond[ncomp]); 1244 continue; 1245done_queue: 1246 nv_free(nv); 1247 if (ISSYNCREQ(hio)) { 1248 if (!refcount_release(&hio->hio_countdown)) 1249 continue; 1250 mtx_lock(&sync_lock); 1251 SYNCREQDONE(hio); 1252 mtx_unlock(&sync_lock); 1253 cv_signal(&sync_cond); 1254 continue; 1255 } 1256 if (ggio->gctl_cmd == BIO_WRITE) { 1257 mtx_lock(&res->hr_amp_lock); 1258 if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1259 ggio->gctl_length)) { 1260 (void)hast_activemap_flush(res); 1261 } 1262 mtx_unlock(&res->hr_amp_lock); 1263 } 1264 if (!refcount_release(&hio->hio_countdown)) 1265 continue; 1266 pjdlog_debug(2, 1267 "remote_send: (%p) Moving request to the done queue.", 1268 hio); 1269 QUEUE_INSERT2(hio, done); 1270 } 1271 /* NOTREACHED */ 1272 return (NULL); 1273} 1274 1275/* 1276 * Thread receives answer from secondary node and passes it to ggate_send 1277 * thread. 1278 */ 1279static void * 1280remote_recv_thread(void *arg) 1281{ 1282 struct hast_resource *res = arg; 1283 struct g_gate_ctl_io *ggio; 1284 struct hio *hio; 1285 struct nv *nv; 1286 unsigned int ncomp; 1287 uint64_t seq; 1288 int error; 1289 1290 /* Remote component is 1 for now. */ 1291 ncomp = 1; 1292 1293 for (;;) { 1294 /* Wait until there is anything to receive. */ 1295 mtx_lock(&hio_recv_list_lock[ncomp]); 1296 while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1297 pjdlog_debug(2, "remote_recv: No requests, waiting."); 1298 cv_wait(&hio_recv_list_cond[ncomp], 1299 &hio_recv_list_lock[ncomp]); 1300 } 1301 mtx_unlock(&hio_recv_list_lock[ncomp]); 1302 rw_rlock(&hio_remote_lock[ncomp]); 1303 if (!ISCONNECTED(res, ncomp)) { 1304 rw_unlock(&hio_remote_lock[ncomp]); 1305 /* 1306 * Connection is dead, so move all pending requests to 1307 * the done queue (one-by-one). 1308 */ 1309 mtx_lock(&hio_recv_list_lock[ncomp]); 1310 hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1311 assert(hio != NULL); 1312 TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1313 hio_next[ncomp]); 1314 mtx_unlock(&hio_recv_list_lock[ncomp]); 1315 goto done_queue; 1316 } 1317 if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1318 pjdlog_errno(LOG_ERR, 1319 "Unable to receive reply header"); 1320 rw_unlock(&hio_remote_lock[ncomp]); 1321 remote_close(res, ncomp); 1322 continue; 1323 } 1324 rw_unlock(&hio_remote_lock[ncomp]); 1325 seq = nv_get_uint64(nv, "seq"); 1326 if (seq == 0) { 1327 pjdlog_error("Header contains no 'seq' field."); 1328 nv_free(nv); 1329 continue; 1330 } 1331 mtx_lock(&hio_recv_list_lock[ncomp]); 1332 TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1333 if (hio->hio_ggio.gctl_seq == seq) { 1334 TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1335 hio_next[ncomp]); 1336 break; 1337 } 1338 } 1339 mtx_unlock(&hio_recv_list_lock[ncomp]); 1340 if (hio == NULL) { 1341 pjdlog_error("Found no request matching received 'seq' field (%ju).", 1342 (uintmax_t)seq); 1343 nv_free(nv); 1344 continue; 1345 } 1346 error = nv_get_int16(nv, "error"); 1347 if (error != 0) { 1348 /* Request failed on remote side. */ 1349 hio->hio_errors[ncomp] = 0; 1350 nv_free(nv); 1351 goto done_queue; 1352 } 1353 ggio = &hio->hio_ggio; 1354 switch (ggio->gctl_cmd) { 1355 case BIO_READ: 1356 rw_rlock(&hio_remote_lock[ncomp]); 1357 if (!ISCONNECTED(res, ncomp)) { 1358 rw_unlock(&hio_remote_lock[ncomp]); 1359 nv_free(nv); 1360 goto done_queue; 1361 } 1362 if (hast_proto_recv_data(res, res->hr_remotein, nv, 1363 ggio->gctl_data, ggio->gctl_length) < 0) { 1364 hio->hio_errors[ncomp] = errno; 1365 pjdlog_errno(LOG_ERR, 1366 "Unable to receive reply data"); 1367 rw_unlock(&hio_remote_lock[ncomp]); 1368 nv_free(nv); 1369 remote_close(res, ncomp); 1370 goto done_queue; 1371 } 1372 rw_unlock(&hio_remote_lock[ncomp]); 1373 break; 1374 case BIO_WRITE: 1375 case BIO_DELETE: 1376 case BIO_FLUSH: 1377 break; 1378 default: 1379 assert(!"invalid condition"); 1380 abort(); 1381 } 1382 hio->hio_errors[ncomp] = 0; 1383 nv_free(nv); 1384done_queue: 1385 if (refcount_release(&hio->hio_countdown)) { 1386 if (ISSYNCREQ(hio)) { 1387 mtx_lock(&sync_lock); 1388 SYNCREQDONE(hio); 1389 mtx_unlock(&sync_lock); 1390 cv_signal(&sync_cond); 1391 } else { 1392 pjdlog_debug(2, 1393 "remote_recv: (%p) Moving request to the done queue.", 1394 hio); 1395 QUEUE_INSERT2(hio, done); 1396 } 1397 } 1398 } 1399 /* NOTREACHED */ 1400 return (NULL); 1401} 1402 1403/* 1404 * Thread sends answer to the kernel. 1405 */ 1406static void * 1407ggate_send_thread(void *arg) 1408{ 1409 struct hast_resource *res = arg; 1410 struct g_gate_ctl_io *ggio; 1411 struct hio *hio; 1412 unsigned int ii, ncomp, ncomps; 1413 1414 ncomps = HAST_NCOMPONENTS; 1415 1416 for (;;) { 1417 pjdlog_debug(2, "ggate_send: Taking request."); 1418 QUEUE_TAKE2(hio, done); 1419 pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1420 ggio = &hio->hio_ggio; 1421 for (ii = 0; ii < ncomps; ii++) { 1422 if (hio->hio_errors[ii] == 0) { 1423 /* 1424 * One successful request is enough to declare 1425 * success. 1426 */ 1427 ggio->gctl_error = 0; 1428 break; 1429 } 1430 } 1431 if (ii == ncomps) { 1432 /* 1433 * None of the requests were successful. 1434 * Use first error. 1435 */ 1436 ggio->gctl_error = hio->hio_errors[0]; 1437 } 1438 if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1439 mtx_lock(&res->hr_amp_lock); 1440 activemap_write_complete(res->hr_amp, 1441 ggio->gctl_offset, ggio->gctl_length); 1442 mtx_unlock(&res->hr_amp_lock); 1443 } 1444 if (ggio->gctl_cmd == BIO_WRITE) { 1445 /* 1446 * Unlock range we locked. 1447 */ 1448 mtx_lock(&range_lock); 1449 rangelock_del(range_regular, ggio->gctl_offset, 1450 ggio->gctl_length); 1451 if (range_sync_wait) 1452 cv_signal(&range_sync_cond); 1453 mtx_unlock(&range_lock); 1454 /* 1455 * Bump local count if this is first write after 1456 * connection failure with remote node. 1457 */ 1458 ncomp = 1; 1459 rw_rlock(&hio_remote_lock[ncomp]); 1460 if (!ISCONNECTED(res, ncomp)) { 1461 mtx_lock(&metadata_lock); 1462 if (res->hr_primary_localcnt == 1463 res->hr_secondary_remotecnt) { 1464 res->hr_primary_localcnt++; 1465 pjdlog_debug(1, 1466 "Increasing localcnt to %ju.", 1467 (uintmax_t)res->hr_primary_localcnt); 1468 (void)metadata_write(res); 1469 } 1470 mtx_unlock(&metadata_lock); 1471 } 1472 rw_unlock(&hio_remote_lock[ncomp]); 1473 } 1474 if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1475 primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1476 pjdlog_debug(2, 1477 "ggate_send: (%p) Moving request to the free queue.", hio); 1478 QUEUE_INSERT2(hio, free); 1479 } 1480 /* NOTREACHED */ 1481 return (NULL); 1482} 1483 1484/* 1485 * Thread synchronize local and remote components. 1486 */ 1487static void * 1488sync_thread(void *arg __unused) 1489{ 1490 struct hast_resource *res = arg; 1491 struct hio *hio; 1492 struct g_gate_ctl_io *ggio; 1493 unsigned int ii, ncomp, ncomps; 1494 off_t offset, length, synced; 1495 bool dorewind; 1496 int syncext; 1497 1498 ncomps = HAST_NCOMPONENTS; 1499 dorewind = true; 1500 synced = 0; 1501 offset = -1; 1502 1503 for (;;) { 1504 mtx_lock(&sync_lock); 1505 if (offset >= 0 && !sync_inprogress) { 1506 pjdlog_info("Synchronization interrupted. " 1507 "%jd bytes synchronized so far.", 1508 (intmax_t)synced); 1509 hook_exec(res->hr_exec, "syncintr", res->hr_name, NULL); 1510 } 1511 while (!sync_inprogress) { 1512 dorewind = true; 1513 synced = 0; 1514 cv_wait(&sync_cond, &sync_lock); 1515 } 1516 mtx_unlock(&sync_lock); 1517 /* 1518 * Obtain offset at which we should synchronize. 1519 * Rewind synchronization if needed. 1520 */ 1521 mtx_lock(&res->hr_amp_lock); 1522 if (dorewind) 1523 activemap_sync_rewind(res->hr_amp); 1524 offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1525 if (syncext != -1) { 1526 /* 1527 * We synchronized entire syncext extent, we can mark 1528 * it as clean now. 1529 */ 1530 if (activemap_extent_complete(res->hr_amp, syncext)) 1531 (void)hast_activemap_flush(res); 1532 } 1533 mtx_unlock(&res->hr_amp_lock); 1534 if (dorewind) { 1535 dorewind = false; 1536 if (offset < 0) 1537 pjdlog_info("Nodes are in sync."); 1538 else { 1539 pjdlog_info("Synchronization started. %ju bytes to go.", 1540 (uintmax_t)(res->hr_extentsize * 1541 activemap_ndirty(res->hr_amp))); 1542 hook_exec(res->hr_exec, "syncstart", 1543 res->hr_name, NULL); 1544 } 1545 } 1546 if (offset < 0) { 1547 sync_stop(); 1548 pjdlog_debug(1, "Nothing to synchronize."); 1549 /* 1550 * Synchronization complete, make both localcnt and 1551 * remotecnt equal. 1552 */ 1553 ncomp = 1; 1554 rw_rlock(&hio_remote_lock[ncomp]); 1555 if (ISCONNECTED(res, ncomp)) { 1556 if (synced > 0) { 1557 pjdlog_info("Synchronization complete. " 1558 "%jd bytes synchronized.", 1559 (intmax_t)synced); 1560 hook_exec(res->hr_exec, "syncdone", 1561 res->hr_name, NULL); 1562 } 1563 mtx_lock(&metadata_lock); 1564 res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1565 res->hr_primary_localcnt = 1566 res->hr_secondary_localcnt; 1567 res->hr_primary_remotecnt = 1568 res->hr_secondary_remotecnt; 1569 pjdlog_debug(1, 1570 "Setting localcnt to %ju and remotecnt to %ju.", 1571 (uintmax_t)res->hr_primary_localcnt, 1572 (uintmax_t)res->hr_secondary_localcnt); 1573 (void)metadata_write(res); 1574 mtx_unlock(&metadata_lock); 1575 } 1576 rw_unlock(&hio_remote_lock[ncomp]); 1577 continue; 1578 } 1579 pjdlog_debug(2, "sync: Taking free request."); 1580 QUEUE_TAKE2(hio, free); 1581 pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1582 /* 1583 * Lock the range we are going to synchronize. We don't want 1584 * race where someone writes between our read and write. 1585 */ 1586 for (;;) { 1587 mtx_lock(&range_lock); 1588 if (rangelock_islocked(range_regular, offset, length)) { 1589 pjdlog_debug(2, 1590 "sync: Range offset=%jd length=%jd locked.", 1591 (intmax_t)offset, (intmax_t)length); 1592 range_sync_wait = true; 1593 cv_wait(&range_sync_cond, &range_lock); 1594 range_sync_wait = false; 1595 mtx_unlock(&range_lock); 1596 continue; 1597 } 1598 if (rangelock_add(range_sync, offset, length) < 0) { 1599 mtx_unlock(&range_lock); 1600 pjdlog_debug(2, 1601 "sync: Range offset=%jd length=%jd is already locked, waiting.", 1602 (intmax_t)offset, (intmax_t)length); 1603 sleep(1); 1604 continue; 1605 } 1606 mtx_unlock(&range_lock); 1607 break; 1608 } 1609 /* 1610 * First read the data from synchronization source. 1611 */ 1612 SYNCREQ(hio); 1613 ggio = &hio->hio_ggio; 1614 ggio->gctl_cmd = BIO_READ; 1615 ggio->gctl_offset = offset; 1616 ggio->gctl_length = length; 1617 ggio->gctl_error = 0; 1618 for (ii = 0; ii < ncomps; ii++) 1619 hio->hio_errors[ii] = EINVAL; 1620 reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1621 hio); 1622 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1623 hio); 1624 mtx_lock(&metadata_lock); 1625 if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1626 /* 1627 * This range is up-to-date on local component, 1628 * so handle request locally. 1629 */ 1630 /* Local component is 0 for now. */ 1631 ncomp = 0; 1632 } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1633 assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1634 /* 1635 * This range is out-of-date on local component, 1636 * so send request to the remote node. 1637 */ 1638 /* Remote component is 1 for now. */ 1639 ncomp = 1; 1640 } 1641 mtx_unlock(&metadata_lock); 1642 refcount_init(&hio->hio_countdown, 1); 1643 QUEUE_INSERT1(hio, send, ncomp); 1644 1645 /* 1646 * Let's wait for READ to finish. 1647 */ 1648 mtx_lock(&sync_lock); 1649 while (!ISSYNCREQDONE(hio)) 1650 cv_wait(&sync_cond, &sync_lock); 1651 mtx_unlock(&sync_lock); 1652 1653 if (hio->hio_errors[ncomp] != 0) { 1654 pjdlog_error("Unable to read synchronization data: %s.", 1655 strerror(hio->hio_errors[ncomp])); 1656 goto free_queue; 1657 } 1658 1659 /* 1660 * We read the data from synchronization source, now write it 1661 * to synchronization target. 1662 */ 1663 SYNCREQ(hio); 1664 ggio->gctl_cmd = BIO_WRITE; 1665 for (ii = 0; ii < ncomps; ii++) 1666 hio->hio_errors[ii] = EINVAL; 1667 reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1668 hio); 1669 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1670 hio); 1671 mtx_lock(&metadata_lock); 1672 if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1673 /* 1674 * This range is up-to-date on local component, 1675 * so we update remote component. 1676 */ 1677 /* Remote component is 1 for now. */ 1678 ncomp = 1; 1679 } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1680 assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1681 /* 1682 * This range is out-of-date on local component, 1683 * so we update it. 1684 */ 1685 /* Local component is 0 for now. */ 1686 ncomp = 0; 1687 } 1688 mtx_unlock(&metadata_lock); 1689 1690 pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1691 hio); 1692 refcount_init(&hio->hio_countdown, 1); 1693 QUEUE_INSERT1(hio, send, ncomp); 1694 1695 /* 1696 * Let's wait for WRITE to finish. 1697 */ 1698 mtx_lock(&sync_lock); 1699 while (!ISSYNCREQDONE(hio)) 1700 cv_wait(&sync_cond, &sync_lock); 1701 mtx_unlock(&sync_lock); 1702 1703 if (hio->hio_errors[ncomp] != 0) { 1704 pjdlog_error("Unable to write synchronization data: %s.", 1705 strerror(hio->hio_errors[ncomp])); 1706 goto free_queue; 1707 } 1708 1709 synced += length; 1710free_queue: 1711 mtx_lock(&range_lock); 1712 rangelock_del(range_sync, offset, length); 1713 if (range_regular_wait) 1714 cv_signal(&range_regular_cond); 1715 mtx_unlock(&range_lock); 1716 pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1717 hio); 1718 QUEUE_INSERT2(hio, free); 1719 } 1720 /* NOTREACHED */ 1721 return (NULL); 1722} 1723 1724static void 1725config_reload(void) 1726{ 1727 struct hastd_config *newcfg; 1728 struct hast_resource *res; 1729 unsigned int ii, ncomps; 1730 int modified; 1731 1732 pjdlog_info("Reloading configuration..."); 1733 1734 ncomps = HAST_NCOMPONENTS; 1735 1736 newcfg = yy_config_parse(cfgpath, false); 1737 if (newcfg == NULL) 1738 goto failed; 1739 1740 TAILQ_FOREACH(res, &newcfg->hc_resources, hr_next) { 1741 if (strcmp(res->hr_name, gres->hr_name) == 0) 1742 break; 1743 } 1744 /* 1745 * If resource was removed from the configuration file, resource 1746 * name, provider name or path to local component was modified we 1747 * shouldn't be here. This means that someone modified configuration 1748 * file and send SIGHUP to us instead of main hastd process. 1749 * Log advice and ignore the signal. 1750 */ 1751 if (res == NULL || strcmp(gres->hr_name, res->hr_name) != 0 || 1752 strcmp(gres->hr_provname, res->hr_provname) != 0 || 1753 strcmp(gres->hr_localpath, res->hr_localpath) != 0) { 1754 pjdlog_warning("To reload configuration send SIGHUP to the main hastd process (pid %u).", 1755 (unsigned int)getppid()); 1756 goto failed; 1757 } 1758 1759#define MODIFIED_REMOTEADDR 0x1 1760#define MODIFIED_REPLICATION 0x2 1761#define MODIFIED_TIMEOUT 0x4 1762#define MODIFIED_EXEC 0x8 1763 modified = 0; 1764 if (strcmp(gres->hr_remoteaddr, res->hr_remoteaddr) != 0) { 1765 /* 1766 * Don't copy res->hr_remoteaddr to gres just yet. 1767 * We want remote_close() to log disconnect from the old 1768 * addresses, not from the new ones. 1769 */ 1770 modified |= MODIFIED_REMOTEADDR; 1771 } 1772 if (gres->hr_replication != res->hr_replication) { 1773 gres->hr_replication = res->hr_replication; 1774 modified |= MODIFIED_REPLICATION; 1775 } 1776 if (gres->hr_timeout != res->hr_timeout) { 1777 gres->hr_timeout = res->hr_timeout; 1778 modified |= MODIFIED_TIMEOUT; 1779 } 1780 if (strcmp(gres->hr_exec, res->hr_exec) != 0) { 1781 strlcpy(gres->hr_exec, res->hr_exec, sizeof(gres->hr_exec)); 1782 modified |= MODIFIED_EXEC; 1783 } 1784 /* 1785 * If only timeout was modified we only need to change it without 1786 * reconnecting. 1787 */ 1788 if (modified == MODIFIED_TIMEOUT) { 1789 for (ii = 0; ii < ncomps; ii++) { 1790 if (!ISREMOTE(ii)) 1791 continue; 1792 rw_rlock(&hio_remote_lock[ii]); 1793 if (!ISCONNECTED(gres, ii)) { 1794 rw_unlock(&hio_remote_lock[ii]); 1795 continue; 1796 } 1797 rw_unlock(&hio_remote_lock[ii]); 1798 if (proto_timeout(gres->hr_remotein, 1799 gres->hr_timeout) < 0) { 1800 pjdlog_errno(LOG_WARNING, 1801 "Unable to set connection timeout"); 1802 } 1803 if (proto_timeout(gres->hr_remoteout, 1804 gres->hr_timeout) < 0) { 1805 pjdlog_errno(LOG_WARNING, 1806 "Unable to set connection timeout"); 1807 } 1808 } 1809 } else if ((modified & 1810 (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1811 for (ii = 0; ii < ncomps; ii++) { 1812 if (!ISREMOTE(ii)) 1813 continue; 1814 remote_close(gres, ii); 1815 } 1816 if (modified & MODIFIED_REMOTEADDR) { 1817 strlcpy(gres->hr_remoteaddr, res->hr_remoteaddr, 1818 sizeof(gres->hr_remoteaddr)); 1819 } 1820 } 1821#undef MODIFIED_REMOTEADDR 1822#undef MODIFIED_REPLICATION 1823#undef MODIFIED_TIMEOUT 1824#undef MODIFIED_EXEC 1825 1826 pjdlog_info("Configuration reloaded successfully."); 1827 return; 1828failed: 1829 if (newcfg != NULL) { 1830 if (newcfg->hc_controlconn != NULL) 1831 proto_close(newcfg->hc_controlconn); 1832 if (newcfg->hc_listenconn != NULL) 1833 proto_close(newcfg->hc_listenconn); 1834 yy_config_free(newcfg); 1835 } 1836 pjdlog_warning("Configuration not reloaded."); 1837} 1838 1839static void 1840keepalive_send(struct hast_resource *res, unsigned int ncomp) 1841{ 1842 struct nv *nv; 1843 1844 nv = nv_alloc(); 1845 nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1846 if (nv_error(nv) != 0) { 1847 nv_free(nv); 1848 pjdlog_debug(1, 1849 "keepalive_send: Unable to prepare header to send."); 1850 return; 1851 } 1852 if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1853 pjdlog_common(LOG_DEBUG, 1, errno, 1854 "keepalive_send: Unable to send request"); 1855 nv_free(nv); 1856 rw_unlock(&hio_remote_lock[ncomp]); 1857 remote_close(res, ncomp); 1858 rw_rlock(&hio_remote_lock[ncomp]); 1859 return; 1860 } 1861 nv_free(nv); 1862 pjdlog_debug(2, "keepalive_send: Request sent."); 1863} 1864 1865static void 1866guard_one(struct hast_resource *res, unsigned int ncomp) 1867{ 1868 struct proto_conn *in, *out; 1869 1870 if (!ISREMOTE(ncomp)) 1871 return; 1872 1873 rw_rlock(&hio_remote_lock[ncomp]); 1874 1875 if (!real_remote(res)) { 1876 rw_unlock(&hio_remote_lock[ncomp]); 1877 return; 1878 } 1879 1880 if (ISCONNECTED(res, ncomp)) { 1881 assert(res->hr_remotein != NULL); 1882 assert(res->hr_remoteout != NULL); 1883 keepalive_send(res, ncomp); 1884 } 1885 1886 if (ISCONNECTED(res, ncomp)) { 1887 assert(res->hr_remotein != NULL); 1888 assert(res->hr_remoteout != NULL); 1889 rw_unlock(&hio_remote_lock[ncomp]); 1890 pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 1891 res->hr_remoteaddr); 1892 return; 1893 } 1894 1895 assert(res->hr_remotein == NULL); 1896 assert(res->hr_remoteout == NULL); 1897 /* 1898 * Upgrade the lock. It doesn't have to be atomic as no other thread 1899 * can change connection status from disconnected to connected. 1900 */ 1901 rw_unlock(&hio_remote_lock[ncomp]); 1902 pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 1903 res->hr_remoteaddr); 1904 in = out = NULL; 1905 if (init_remote(res, &in, &out)) { 1906 rw_wlock(&hio_remote_lock[ncomp]); 1907 assert(res->hr_remotein == NULL); 1908 assert(res->hr_remoteout == NULL); 1909 assert(in != NULL && out != NULL); 1910 res->hr_remotein = in; 1911 res->hr_remoteout = out; 1912 rw_unlock(&hio_remote_lock[ncomp]); 1913 pjdlog_info("Successfully reconnected to %s.", 1914 res->hr_remoteaddr); 1915 sync_start(); 1916 } else { 1917 /* Both connections should be NULL. */ 1918 assert(res->hr_remotein == NULL); 1919 assert(res->hr_remoteout == NULL); 1920 assert(in == NULL && out == NULL); 1921 pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 1922 res->hr_remoteaddr); 1923 } 1924} 1925 1926/* 1927 * Thread guards remote connections and reconnects when needed, handles 1928 * signals, etc. 1929 */ 1930static void * 1931guard_thread(void *arg) 1932{ 1933 struct hast_resource *res = arg; 1934 unsigned int ii, ncomps; 1935 struct timespec timeout; 1936 time_t lastcheck, now; 1937 sigset_t mask; 1938 int signo; 1939 1940 ncomps = HAST_NCOMPONENTS; 1941 lastcheck = time(NULL); 1942 1943 PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1944 PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1945 PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1946 PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1947 PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1948 1949 timeout.tv_nsec = 0; 1950 signo = -1; 1951 1952 for (;;) { 1953 switch (signo) { 1954 case SIGHUP: 1955 config_reload(); 1956 break; 1957 case SIGINT: 1958 case SIGTERM: 1959 sigexit_received = true; 1960 primary_exitx(EX_OK, 1961 "Termination signal received, exiting."); 1962 break; 1963 default: 1964 break; 1965 } 1966 hook_check(signo == SIGCHLD); 1967 1968 pjdlog_debug(2, "remote_guard: Checking connections."); 1969 now = time(NULL); 1970 if (lastcheck + RETRY_SLEEP <= now) { 1971 for (ii = 0; ii < ncomps; ii++) 1972 guard_one(res, ii); 1973 lastcheck = now; 1974 } 1975 timeout.tv_sec = RETRY_SLEEP; 1976 signo = sigtimedwait(&mask, NULL, &timeout); 1977 } 1978 /* NOTREACHED */ 1979 return (NULL); 1980} 1981