primary.c revision 211880
1204076Spjd/*- 2204076Spjd * Copyright (c) 2009 The FreeBSD Foundation 3210886Spjd * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 4204076Spjd * All rights reserved. 5204076Spjd * 6204076Spjd * This software was developed by Pawel Jakub Dawidek under sponsorship from 7204076Spjd * the FreeBSD Foundation. 8204076Spjd * 9204076Spjd * Redistribution and use in source and binary forms, with or without 10204076Spjd * modification, are permitted provided that the following conditions 11204076Spjd * are met: 12204076Spjd * 1. Redistributions of source code must retain the above copyright 13204076Spjd * notice, this list of conditions and the following disclaimer. 14204076Spjd * 2. Redistributions in binary form must reproduce the above copyright 15204076Spjd * notice, this list of conditions and the following disclaimer in the 16204076Spjd * documentation and/or other materials provided with the distribution. 17204076Spjd * 18204076Spjd * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 19204076Spjd * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20204076Spjd * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21204076Spjd * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 22204076Spjd * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23204076Spjd * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24204076Spjd * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25204076Spjd * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26204076Spjd * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27204076Spjd * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28204076Spjd * SUCH DAMAGE. 29204076Spjd */ 30204076Spjd 31204076Spjd#include <sys/cdefs.h> 32204076Spjd__FBSDID("$FreeBSD: head/sbin/hastd/primary.c 211880 2010-08-27 14:10:25Z pjd $"); 33204076Spjd 34204076Spjd#include <sys/types.h> 35204076Spjd#include <sys/time.h> 36204076Spjd#include <sys/bio.h> 37204076Spjd#include <sys/disk.h> 38204076Spjd#include <sys/refcount.h> 39204076Spjd#include <sys/stat.h> 40204076Spjd 41204076Spjd#include <geom/gate/g_gate.h> 42204076Spjd 43204076Spjd#include <assert.h> 44204076Spjd#include <err.h> 45204076Spjd#include <errno.h> 46204076Spjd#include <fcntl.h> 47204076Spjd#include <libgeom.h> 48204076Spjd#include <pthread.h> 49204076Spjd#include <stdint.h> 50204076Spjd#include <stdio.h> 51204076Spjd#include <string.h> 52204076Spjd#include <sysexits.h> 53204076Spjd#include <unistd.h> 54204076Spjd 55204076Spjd#include <activemap.h> 56204076Spjd#include <nv.h> 57204076Spjd#include <rangelock.h> 58204076Spjd 59204076Spjd#include "control.h" 60204076Spjd#include "hast.h" 61204076Spjd#include "hast_proto.h" 62204076Spjd#include "hastd.h" 63204076Spjd#include "metadata.h" 64204076Spjd#include "proto.h" 65204076Spjd#include "pjdlog.h" 66204076Spjd#include "subr.h" 67204076Spjd#include "synch.h" 68204076Spjd 69210886Spjd/* The is only one remote component for now. */ 70210886Spjd#define ISREMOTE(no) ((no) == 1) 71210886Spjd 72204076Spjdstruct hio { 73204076Spjd /* 74204076Spjd * Number of components we are still waiting for. 75204076Spjd * When this field goes to 0, we can send the request back to the 76204076Spjd * kernel. Each component has to decrease this counter by one 77204076Spjd * even on failure. 78204076Spjd */ 79204076Spjd unsigned int hio_countdown; 80204076Spjd /* 81204076Spjd * Each component has a place to store its own error. 82204076Spjd * Once the request is handled by all components we can decide if the 83204076Spjd * request overall is successful or not. 84204076Spjd */ 85204076Spjd int *hio_errors; 86204076Spjd /* 87204076Spjd * Structure used to comunicate with GEOM Gate class. 88204076Spjd */ 89204076Spjd struct g_gate_ctl_io hio_ggio; 90204076Spjd TAILQ_ENTRY(hio) *hio_next; 91204076Spjd}; 92204076Spjd#define hio_free_next hio_next[0] 93204076Spjd#define hio_done_next hio_next[0] 94204076Spjd 95204076Spjd/* 96204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 97204076Spjd * until some in-progress requests are freed. 98204076Spjd */ 99204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 100204076Spjdstatic pthread_mutex_t hio_free_list_lock; 101204076Spjdstatic pthread_cond_t hio_free_list_cond; 102204076Spjd/* 103204076Spjd * There is one send list for every component. One requests is placed on all 104204076Spjd * send lists - each component gets the same request, but each component is 105204076Spjd * responsible for managing his own send list. 106204076Spjd */ 107204076Spjdstatic TAILQ_HEAD(, hio) *hio_send_list; 108204076Spjdstatic pthread_mutex_t *hio_send_list_lock; 109204076Spjdstatic pthread_cond_t *hio_send_list_cond; 110204076Spjd/* 111204076Spjd * There is one recv list for every component, although local components don't 112204076Spjd * use recv lists as local requests are done synchronously. 113204076Spjd */ 114204076Spjdstatic TAILQ_HEAD(, hio) *hio_recv_list; 115204076Spjdstatic pthread_mutex_t *hio_recv_list_lock; 116204076Spjdstatic pthread_cond_t *hio_recv_list_cond; 117204076Spjd/* 118204076Spjd * Request is placed on done list by the slowest component (the one that 119204076Spjd * decreased hio_countdown from 1 to 0). 120204076Spjd */ 121204076Spjdstatic TAILQ_HEAD(, hio) hio_done_list; 122204076Spjdstatic pthread_mutex_t hio_done_list_lock; 123204076Spjdstatic pthread_cond_t hio_done_list_cond; 124204076Spjd/* 125204076Spjd * Structure below are for interaction with sync thread. 126204076Spjd */ 127204076Spjdstatic bool sync_inprogress; 128204076Spjdstatic pthread_mutex_t sync_lock; 129204076Spjdstatic pthread_cond_t sync_cond; 130204076Spjd/* 131204076Spjd * The lock below allows to synchornize access to remote connections. 132204076Spjd */ 133204076Spjdstatic pthread_rwlock_t *hio_remote_lock; 134204076Spjdstatic pthread_mutex_t hio_guard_lock; 135204076Spjdstatic pthread_cond_t hio_guard_cond; 136204076Spjd 137204076Spjd/* 138204076Spjd * Lock to synchronize metadata updates. Also synchronize access to 139204076Spjd * hr_primary_localcnt and hr_primary_remotecnt fields. 140204076Spjd */ 141204076Spjdstatic pthread_mutex_t metadata_lock; 142204076Spjd 143204076Spjd/* 144204076Spjd * Maximum number of outstanding I/O requests. 145204076Spjd */ 146204076Spjd#define HAST_HIO_MAX 256 147204076Spjd/* 148204076Spjd * Number of components. At this point there are only two components: local 149204076Spjd * and remote, but in the future it might be possible to use multiple local 150204076Spjd * and remote components. 151204076Spjd */ 152204076Spjd#define HAST_NCOMPONENTS 2 153204076Spjd/* 154204076Spjd * Number of seconds to sleep before next reconnect try. 155204076Spjd */ 156204076Spjd#define RECONNECT_SLEEP 5 157204076Spjd 158204076Spjd#define ISCONNECTED(res, no) \ 159204076Spjd ((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL) 160204076Spjd 161204076Spjd#define QUEUE_INSERT1(hio, name, ncomp) do { \ 162204076Spjd bool _wakeup; \ 163204076Spjd \ 164204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 165204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]); \ 166204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio), \ 167204076Spjd hio_next[(ncomp)]); \ 168204076Spjd mtx_unlock(&hio_##name##_list_lock[ncomp]); \ 169204076Spjd if (_wakeup) \ 170204076Spjd cv_signal(&hio_##name##_list_cond[(ncomp)]); \ 171204076Spjd} while (0) 172204076Spjd#define QUEUE_INSERT2(hio, name) do { \ 173204076Spjd bool _wakeup; \ 174204076Spjd \ 175204076Spjd mtx_lock(&hio_##name##_list_lock); \ 176204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 177204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\ 178204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 179204076Spjd if (_wakeup) \ 180204076Spjd cv_signal(&hio_##name##_list_cond); \ 181204076Spjd} while (0) 182204076Spjd#define QUEUE_TAKE1(hio, name, ncomp) do { \ 183204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 184204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL) { \ 185204076Spjd cv_wait(&hio_##name##_list_cond[(ncomp)], \ 186204076Spjd &hio_##name##_list_lock[(ncomp)]); \ 187204076Spjd } \ 188204076Spjd TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio), \ 189204076Spjd hio_next[(ncomp)]); \ 190204076Spjd mtx_unlock(&hio_##name##_list_lock[(ncomp)]); \ 191204076Spjd} while (0) 192204076Spjd#define QUEUE_TAKE2(hio, name) do { \ 193204076Spjd mtx_lock(&hio_##name##_list_lock); \ 194204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 195204076Spjd cv_wait(&hio_##name##_list_cond, \ 196204076Spjd &hio_##name##_list_lock); \ 197204076Spjd } \ 198204076Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next); \ 199204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 200204076Spjd} while (0) 201204076Spjd 202209183Spjd#define SYNCREQ(hio) do { \ 203209183Spjd (hio)->hio_ggio.gctl_unit = -1; \ 204209183Spjd (hio)->hio_ggio.gctl_seq = 1; \ 205209183Spjd} while (0) 206204076Spjd#define ISSYNCREQ(hio) ((hio)->hio_ggio.gctl_unit == -1) 207204076Spjd#define SYNCREQDONE(hio) do { (hio)->hio_ggio.gctl_unit = -2; } while (0) 208204076Spjd#define ISSYNCREQDONE(hio) ((hio)->hio_ggio.gctl_unit == -2) 209204076Spjd 210204076Spjdstatic struct hast_resource *gres; 211204076Spjd 212204076Spjdstatic pthread_mutex_t range_lock; 213204076Spjdstatic struct rangelocks *range_regular; 214204076Spjdstatic bool range_regular_wait; 215204076Spjdstatic pthread_cond_t range_regular_cond; 216204076Spjdstatic struct rangelocks *range_sync; 217204076Spjdstatic bool range_sync_wait; 218204076Spjdstatic pthread_cond_t range_sync_cond; 219204076Spjd 220204076Spjdstatic void *ggate_recv_thread(void *arg); 221204076Spjdstatic void *local_send_thread(void *arg); 222204076Spjdstatic void *remote_send_thread(void *arg); 223204076Spjdstatic void *remote_recv_thread(void *arg); 224204076Spjdstatic void *ggate_send_thread(void *arg); 225204076Spjdstatic void *sync_thread(void *arg); 226204076Spjdstatic void *guard_thread(void *arg); 227204076Spjd 228204076Spjdstatic void sighandler(int sig); 229204076Spjd 230204076Spjdstatic void 231204076Spjdcleanup(struct hast_resource *res) 232204076Spjd{ 233204076Spjd int rerrno; 234204076Spjd 235204076Spjd /* Remember errno. */ 236204076Spjd rerrno = errno; 237204076Spjd 238204076Spjd /* 239204076Spjd * Close descriptor to /dev/hast/<name> 240204076Spjd * to work-around race in the kernel. 241204076Spjd */ 242204076Spjd close(res->hr_localfd); 243204076Spjd 244204076Spjd /* Destroy ggate provider if we created one. */ 245204076Spjd if (res->hr_ggateunit >= 0) { 246204076Spjd struct g_gate_ctl_destroy ggiod; 247204076Spjd 248204076Spjd ggiod.gctl_version = G_GATE_VERSION; 249204076Spjd ggiod.gctl_unit = res->hr_ggateunit; 250204076Spjd ggiod.gctl_force = 1; 251204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 252204076Spjd pjdlog_warning("Unable to destroy hast/%s device", 253204076Spjd res->hr_provname); 254204076Spjd } 255204076Spjd res->hr_ggateunit = -1; 256204076Spjd } 257204076Spjd 258204076Spjd /* Restore errno. */ 259204076Spjd errno = rerrno; 260204076Spjd} 261204076Spjd 262204076Spjdstatic void 263204076Spjdprimary_exit(int exitcode, const char *fmt, ...) 264204076Spjd{ 265204076Spjd va_list ap; 266204076Spjd 267204076Spjd assert(exitcode != EX_OK); 268204076Spjd va_start(ap, fmt); 269204076Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 270204076Spjd va_end(ap); 271204076Spjd cleanup(gres); 272204076Spjd exit(exitcode); 273204076Spjd} 274204076Spjd 275204076Spjdstatic void 276204076Spjdprimary_exitx(int exitcode, const char *fmt, ...) 277204076Spjd{ 278204076Spjd va_list ap; 279204076Spjd 280204076Spjd va_start(ap, fmt); 281204076Spjd pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 282204076Spjd va_end(ap); 283204076Spjd cleanup(gres); 284204076Spjd exit(exitcode); 285204076Spjd} 286204076Spjd 287204076Spjdstatic int 288204076Spjdhast_activemap_flush(struct hast_resource *res) 289204076Spjd{ 290204076Spjd const unsigned char *buf; 291204076Spjd size_t size; 292204076Spjd 293204076Spjd buf = activemap_bitmap(res->hr_amp, &size); 294204076Spjd assert(buf != NULL); 295204076Spjd assert((size % res->hr_local_sectorsize) == 0); 296204076Spjd if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 297204076Spjd (ssize_t)size) { 298204076Spjd KEEP_ERRNO(pjdlog_errno(LOG_ERR, 299204076Spjd "Unable to flush activemap to disk")); 300204076Spjd return (-1); 301204076Spjd } 302204076Spjd return (0); 303204076Spjd} 304204076Spjd 305210881Spjdstatic bool 306210881Spjdreal_remote(const struct hast_resource *res) 307210881Spjd{ 308210881Spjd 309210881Spjd return (strcmp(res->hr_remoteaddr, "none") != 0); 310210881Spjd} 311210881Spjd 312204076Spjdstatic void 313204076Spjdinit_environment(struct hast_resource *res __unused) 314204076Spjd{ 315204076Spjd struct hio *hio; 316204076Spjd unsigned int ii, ncomps; 317204076Spjd 318204076Spjd /* 319204076Spjd * In the future it might be per-resource value. 320204076Spjd */ 321204076Spjd ncomps = HAST_NCOMPONENTS; 322204076Spjd 323204076Spjd /* 324204076Spjd * Allocate memory needed by lists. 325204076Spjd */ 326204076Spjd hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 327204076Spjd if (hio_send_list == NULL) { 328204076Spjd primary_exitx(EX_TEMPFAIL, 329204076Spjd "Unable to allocate %zu bytes of memory for send lists.", 330204076Spjd sizeof(hio_send_list[0]) * ncomps); 331204076Spjd } 332204076Spjd hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 333204076Spjd if (hio_send_list_lock == NULL) { 334204076Spjd primary_exitx(EX_TEMPFAIL, 335204076Spjd "Unable to allocate %zu bytes of memory for send list locks.", 336204076Spjd sizeof(hio_send_list_lock[0]) * ncomps); 337204076Spjd } 338204076Spjd hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 339204076Spjd if (hio_send_list_cond == NULL) { 340204076Spjd primary_exitx(EX_TEMPFAIL, 341204076Spjd "Unable to allocate %zu bytes of memory for send list condition variables.", 342204076Spjd sizeof(hio_send_list_cond[0]) * ncomps); 343204076Spjd } 344204076Spjd hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 345204076Spjd if (hio_recv_list == NULL) { 346204076Spjd primary_exitx(EX_TEMPFAIL, 347204076Spjd "Unable to allocate %zu bytes of memory for recv lists.", 348204076Spjd sizeof(hio_recv_list[0]) * ncomps); 349204076Spjd } 350204076Spjd hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 351204076Spjd if (hio_recv_list_lock == NULL) { 352204076Spjd primary_exitx(EX_TEMPFAIL, 353204076Spjd "Unable to allocate %zu bytes of memory for recv list locks.", 354204076Spjd sizeof(hio_recv_list_lock[0]) * ncomps); 355204076Spjd } 356204076Spjd hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 357204076Spjd if (hio_recv_list_cond == NULL) { 358204076Spjd primary_exitx(EX_TEMPFAIL, 359204076Spjd "Unable to allocate %zu bytes of memory for recv list condition variables.", 360204076Spjd sizeof(hio_recv_list_cond[0]) * ncomps); 361204076Spjd } 362204076Spjd hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 363204076Spjd if (hio_remote_lock == NULL) { 364204076Spjd primary_exitx(EX_TEMPFAIL, 365204076Spjd "Unable to allocate %zu bytes of memory for remote connections locks.", 366204076Spjd sizeof(hio_remote_lock[0]) * ncomps); 367204076Spjd } 368204076Spjd 369204076Spjd /* 370204076Spjd * Initialize lists, their locks and theirs condition variables. 371204076Spjd */ 372204076Spjd TAILQ_INIT(&hio_free_list); 373204076Spjd mtx_init(&hio_free_list_lock); 374204076Spjd cv_init(&hio_free_list_cond); 375204076Spjd for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 376204076Spjd TAILQ_INIT(&hio_send_list[ii]); 377204076Spjd mtx_init(&hio_send_list_lock[ii]); 378204076Spjd cv_init(&hio_send_list_cond[ii]); 379204076Spjd TAILQ_INIT(&hio_recv_list[ii]); 380204076Spjd mtx_init(&hio_recv_list_lock[ii]); 381204076Spjd cv_init(&hio_recv_list_cond[ii]); 382204076Spjd rw_init(&hio_remote_lock[ii]); 383204076Spjd } 384204076Spjd TAILQ_INIT(&hio_done_list); 385204076Spjd mtx_init(&hio_done_list_lock); 386204076Spjd cv_init(&hio_done_list_cond); 387204076Spjd mtx_init(&hio_guard_lock); 388204076Spjd cv_init(&hio_guard_cond); 389204076Spjd mtx_init(&metadata_lock); 390204076Spjd 391204076Spjd /* 392204076Spjd * Allocate requests pool and initialize requests. 393204076Spjd */ 394204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 395204076Spjd hio = malloc(sizeof(*hio)); 396204076Spjd if (hio == NULL) { 397204076Spjd primary_exitx(EX_TEMPFAIL, 398204076Spjd "Unable to allocate %zu bytes of memory for hio request.", 399204076Spjd sizeof(*hio)); 400204076Spjd } 401204076Spjd hio->hio_countdown = 0; 402204076Spjd hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 403204076Spjd if (hio->hio_errors == NULL) { 404204076Spjd primary_exitx(EX_TEMPFAIL, 405204076Spjd "Unable allocate %zu bytes of memory for hio errors.", 406204076Spjd sizeof(hio->hio_errors[0]) * ncomps); 407204076Spjd } 408204076Spjd hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 409204076Spjd if (hio->hio_next == NULL) { 410204076Spjd primary_exitx(EX_TEMPFAIL, 411204076Spjd "Unable allocate %zu bytes of memory for hio_next field.", 412204076Spjd sizeof(hio->hio_next[0]) * ncomps); 413204076Spjd } 414204076Spjd hio->hio_ggio.gctl_version = G_GATE_VERSION; 415204076Spjd hio->hio_ggio.gctl_data = malloc(MAXPHYS); 416204076Spjd if (hio->hio_ggio.gctl_data == NULL) { 417204076Spjd primary_exitx(EX_TEMPFAIL, 418204076Spjd "Unable to allocate %zu bytes of memory for gctl_data.", 419204076Spjd MAXPHYS); 420204076Spjd } 421204076Spjd hio->hio_ggio.gctl_length = MAXPHYS; 422204076Spjd hio->hio_ggio.gctl_error = 0; 423204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 424204076Spjd } 425204076Spjd 426204076Spjd /* 427204076Spjd * Turn on signals handling. 428204076Spjd */ 429204076Spjd signal(SIGINT, sighandler); 430204076Spjd signal(SIGTERM, sighandler); 431210886Spjd signal(SIGHUP, sighandler); 432204076Spjd} 433204076Spjd 434204076Spjdstatic void 435204076Spjdinit_local(struct hast_resource *res) 436204076Spjd{ 437204076Spjd unsigned char *buf; 438204076Spjd size_t mapsize; 439204076Spjd 440204076Spjd if (metadata_read(res, true) < 0) 441204076Spjd exit(EX_NOINPUT); 442204076Spjd mtx_init(&res->hr_amp_lock); 443204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 444204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 445204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 446204076Spjd } 447204076Spjd mtx_init(&range_lock); 448204076Spjd cv_init(&range_regular_cond); 449204076Spjd if (rangelock_init(&range_regular) < 0) 450204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 451204076Spjd cv_init(&range_sync_cond); 452204076Spjd if (rangelock_init(&range_sync) < 0) 453204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 454204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 455204076Spjd buf = calloc(1, mapsize); 456204076Spjd if (buf == NULL) { 457204076Spjd primary_exitx(EX_TEMPFAIL, 458204076Spjd "Unable to allocate buffer for activemap."); 459204076Spjd } 460204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 461204076Spjd (ssize_t)mapsize) { 462204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 463204076Spjd } 464204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 465209181Spjd free(buf); 466204076Spjd if (res->hr_resuid != 0) 467204076Spjd return; 468204076Spjd /* 469204076Spjd * We're using provider for the first time, so we have to generate 470204076Spjd * resource unique identifier and initialize local and remote counts. 471204076Spjd */ 472204076Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 473204076Spjd res->hr_primary_localcnt = 1; 474204076Spjd res->hr_primary_remotecnt = 0; 475204076Spjd if (metadata_write(res) < 0) 476204076Spjd exit(EX_NOINPUT); 477204076Spjd} 478204076Spjd 479205738Spjdstatic bool 480205738Spjdinit_remote(struct hast_resource *res, struct proto_conn **inp, 481205738Spjd struct proto_conn **outp) 482204076Spjd{ 483205738Spjd struct proto_conn *in, *out; 484204076Spjd struct nv *nvout, *nvin; 485204076Spjd const unsigned char *token; 486204076Spjd unsigned char *map; 487204076Spjd const char *errmsg; 488204076Spjd int32_t extentsize; 489204076Spjd int64_t datasize; 490204076Spjd uint32_t mapsize; 491204076Spjd size_t size; 492204076Spjd 493205738Spjd assert((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 494210881Spjd assert(real_remote(res)); 495205738Spjd 496205738Spjd in = out = NULL; 497205738Spjd 498204076Spjd /* Prepare outgoing connection with remote node. */ 499205738Spjd if (proto_client(res->hr_remoteaddr, &out) < 0) { 500207347Spjd primary_exit(EX_TEMPFAIL, "Unable to create connection to %s", 501204076Spjd res->hr_remoteaddr); 502204076Spjd } 503204076Spjd /* Try to connect, but accept failure. */ 504205738Spjd if (proto_connect(out) < 0) { 505204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 506204076Spjd res->hr_remoteaddr); 507204076Spjd goto close; 508204076Spjd } 509207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 510207371Spjd if (proto_timeout(out, res->hr_timeout) < 0) 511207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 512204076Spjd /* 513204076Spjd * First handshake step. 514204076Spjd * Setup outgoing connection with remote node. 515204076Spjd */ 516204076Spjd nvout = nv_alloc(); 517204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 518204076Spjd if (nv_error(nvout) != 0) { 519204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 520204076Spjd "Unable to allocate header for connection with %s", 521204076Spjd res->hr_remoteaddr); 522204076Spjd nv_free(nvout); 523204076Spjd goto close; 524204076Spjd } 525205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 526204076Spjd pjdlog_errno(LOG_WARNING, 527204076Spjd "Unable to send handshake header to %s", 528204076Spjd res->hr_remoteaddr); 529204076Spjd nv_free(nvout); 530204076Spjd goto close; 531204076Spjd } 532204076Spjd nv_free(nvout); 533205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 534204076Spjd pjdlog_errno(LOG_WARNING, 535204076Spjd "Unable to receive handshake header from %s", 536204076Spjd res->hr_remoteaddr); 537204076Spjd goto close; 538204076Spjd } 539204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 540204076Spjd if (errmsg != NULL) { 541204076Spjd pjdlog_warning("%s", errmsg); 542204076Spjd nv_free(nvin); 543204076Spjd goto close; 544204076Spjd } 545204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 546204076Spjd if (token == NULL) { 547204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 548204076Spjd res->hr_remoteaddr); 549204076Spjd nv_free(nvin); 550204076Spjd goto close; 551204076Spjd } 552204076Spjd if (size != sizeof(res->hr_token)) { 553204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 554204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 555204076Spjd nv_free(nvin); 556204076Spjd goto close; 557204076Spjd } 558204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 559204076Spjd nv_free(nvin); 560204076Spjd 561204076Spjd /* 562204076Spjd * Second handshake step. 563204076Spjd * Setup incoming connection with remote node. 564204076Spjd */ 565205738Spjd if (proto_client(res->hr_remoteaddr, &in) < 0) { 566204076Spjd pjdlog_errno(LOG_WARNING, "Unable to create connection to %s", 567204076Spjd res->hr_remoteaddr); 568204076Spjd } 569204076Spjd /* Try to connect, but accept failure. */ 570205738Spjd if (proto_connect(in) < 0) { 571204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 572204076Spjd res->hr_remoteaddr); 573204076Spjd goto close; 574204076Spjd } 575207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 576207371Spjd if (proto_timeout(in, res->hr_timeout) < 0) 577207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 578204076Spjd nvout = nv_alloc(); 579204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 580204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 581204076Spjd "token"); 582204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 583204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 584204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 585204076Spjd if (nv_error(nvout) != 0) { 586204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 587204076Spjd "Unable to allocate header for connection with %s", 588204076Spjd res->hr_remoteaddr); 589204076Spjd nv_free(nvout); 590204076Spjd goto close; 591204076Spjd } 592205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 593204076Spjd pjdlog_errno(LOG_WARNING, 594204076Spjd "Unable to send handshake header to %s", 595204076Spjd res->hr_remoteaddr); 596204076Spjd nv_free(nvout); 597204076Spjd goto close; 598204076Spjd } 599204076Spjd nv_free(nvout); 600205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 601204076Spjd pjdlog_errno(LOG_WARNING, 602204076Spjd "Unable to receive handshake header from %s", 603204076Spjd res->hr_remoteaddr); 604204076Spjd goto close; 605204076Spjd } 606204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 607204076Spjd if (errmsg != NULL) { 608204076Spjd pjdlog_warning("%s", errmsg); 609204076Spjd nv_free(nvin); 610204076Spjd goto close; 611204076Spjd } 612204076Spjd datasize = nv_get_int64(nvin, "datasize"); 613204076Spjd if (datasize != res->hr_datasize) { 614204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 615204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 616204076Spjd nv_free(nvin); 617204076Spjd goto close; 618204076Spjd } 619204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 620204076Spjd if (extentsize != res->hr_extentsize) { 621204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 622204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 623204076Spjd nv_free(nvin); 624204076Spjd goto close; 625204076Spjd } 626204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 627204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 628204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 629204076Spjd map = NULL; 630204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 631204076Spjd if (mapsize > 0) { 632204076Spjd map = malloc(mapsize); 633204076Spjd if (map == NULL) { 634204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 635204076Spjd (uintmax_t)mapsize); 636204076Spjd nv_free(nvin); 637204076Spjd goto close; 638204076Spjd } 639204076Spjd /* 640204076Spjd * Remote node have some dirty extents on its own, lets 641204076Spjd * download its activemap. 642204076Spjd */ 643205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 644204076Spjd mapsize) < 0) { 645204076Spjd pjdlog_errno(LOG_ERR, 646204076Spjd "Unable to receive remote activemap"); 647204076Spjd nv_free(nvin); 648204076Spjd free(map); 649204076Spjd goto close; 650204076Spjd } 651204076Spjd /* 652204076Spjd * Merge local and remote bitmaps. 653204076Spjd */ 654204076Spjd activemap_merge(res->hr_amp, map, mapsize); 655204076Spjd free(map); 656204076Spjd /* 657204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 658204076Spjd * disk before we start to synchronize. 659204076Spjd */ 660204076Spjd (void)hast_activemap_flush(res); 661204076Spjd } 662204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 663205738Spjd if (inp != NULL && outp != NULL) { 664205738Spjd *inp = in; 665205738Spjd *outp = out; 666205738Spjd } else { 667205738Spjd res->hr_remotein = in; 668205738Spjd res->hr_remoteout = out; 669205738Spjd } 670205738Spjd return (true); 671205738Spjdclose: 672205738Spjd proto_close(out); 673205738Spjd if (in != NULL) 674205738Spjd proto_close(in); 675205738Spjd return (false); 676205738Spjd} 677205738Spjd 678205738Spjdstatic void 679205738Spjdsync_start(void) 680205738Spjd{ 681205738Spjd 682204076Spjd mtx_lock(&sync_lock); 683204076Spjd sync_inprogress = true; 684204076Spjd mtx_unlock(&sync_lock); 685204076Spjd cv_signal(&sync_cond); 686204076Spjd} 687204076Spjd 688204076Spjdstatic void 689211878Spjdsync_stop(void) 690211878Spjd{ 691211878Spjd 692211878Spjd mtx_lock(&sync_lock); 693211878Spjd if (sync_inprogress) 694211878Spjd sync_inprogress = false; 695211878Spjd mtx_unlock(&sync_lock); 696211878Spjd} 697211878Spjd 698211878Spjdstatic void 699204076Spjdinit_ggate(struct hast_resource *res) 700204076Spjd{ 701204076Spjd struct g_gate_ctl_create ggiocreate; 702204076Spjd struct g_gate_ctl_cancel ggiocancel; 703204076Spjd 704204076Spjd /* 705204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 706204076Spjd */ 707204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 708204076Spjd if (res->hr_ggatefd < 0) 709204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 710204076Spjd /* 711204076Spjd * Create provider before trying to connect, as connection failure 712204076Spjd * is not critical, but may take some time. 713204076Spjd */ 714204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 715204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 716204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 717204076Spjd ggiocreate.gctl_flags = 0; 718206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 719204076Spjd ggiocreate.gctl_timeout = 0; 720204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 721204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 722204076Spjd res->hr_provname); 723204076Spjd bzero(ggiocreate.gctl_info, sizeof(ggiocreate.gctl_info)); 724204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 725204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 726204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 727204076Spjd return; 728204076Spjd } 729204076Spjd if (errno != EEXIST) { 730204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 731204076Spjd res->hr_provname); 732204076Spjd } 733204076Spjd pjdlog_debug(1, 734204076Spjd "Device hast/%s already exists, we will try to take it over.", 735204076Spjd res->hr_provname); 736204076Spjd /* 737204076Spjd * If we received EEXIST, we assume that the process who created the 738204076Spjd * provider died and didn't clean up. In that case we will start from 739204076Spjd * where he left of. 740204076Spjd */ 741204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 742204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 743204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 744204076Spjd res->hr_provname); 745204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 746204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 747204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 748204076Spjd return; 749204076Spjd } 750204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 751204076Spjd res->hr_provname); 752204076Spjd} 753204076Spjd 754204076Spjdvoid 755204076Spjdhastd_primary(struct hast_resource *res) 756204076Spjd{ 757204076Spjd pthread_t td; 758204076Spjd pid_t pid; 759204076Spjd int error; 760204076Spjd 761204076Spjd gres = res; 762204076Spjd 763204076Spjd /* 764204076Spjd * Create communication channel between parent and child. 765204076Spjd */ 766204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 767204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 768204076Spjd primary_exit(EX_OSERR, 769204076Spjd "Unable to create control sockets between parent and child"); 770204076Spjd } 771204076Spjd 772204076Spjd pid = fork(); 773204076Spjd if (pid < 0) { 774204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 775207347Spjd primary_exit(EX_TEMPFAIL, "Unable to fork"); 776204076Spjd } 777204076Spjd 778204076Spjd if (pid > 0) { 779204076Spjd /* This is parent. */ 780204076Spjd res->hr_workerpid = pid; 781204076Spjd return; 782204076Spjd } 783204076Spjd (void)pidfile_close(pfh); 784204076Spjd 785204076Spjd setproctitle("%s (primary)", res->hr_name); 786204076Spjd 787210880Spjd signal(SIGHUP, SIG_DFL); 788210880Spjd signal(SIGCHLD, SIG_DFL); 789210880Spjd 790204076Spjd init_local(res); 791210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 792205738Spjd sync_start(); 793204076Spjd init_ggate(res); 794204076Spjd init_environment(res); 795204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 796204076Spjd assert(error == 0); 797204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 798204076Spjd assert(error == 0); 799204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 800204076Spjd assert(error == 0); 801204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 802204076Spjd assert(error == 0); 803204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 804204076Spjd assert(error == 0); 805204076Spjd error = pthread_create(&td, NULL, sync_thread, res); 806204076Spjd assert(error == 0); 807204076Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 808204076Spjd assert(error == 0); 809204076Spjd (void)guard_thread(res); 810204076Spjd} 811204076Spjd 812204076Spjdstatic void 813204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 814204076Spjd{ 815204076Spjd char msg[1024]; 816204076Spjd va_list ap; 817204076Spjd int len; 818204076Spjd 819204076Spjd va_start(ap, fmt); 820204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 821204076Spjd va_end(ap); 822204076Spjd if ((size_t)len < sizeof(msg)) { 823204076Spjd switch (ggio->gctl_cmd) { 824204076Spjd case BIO_READ: 825204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 826204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 827204076Spjd (uintmax_t)ggio->gctl_length); 828204076Spjd break; 829204076Spjd case BIO_DELETE: 830204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 831204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 832204076Spjd (uintmax_t)ggio->gctl_length); 833204076Spjd break; 834204076Spjd case BIO_FLUSH: 835204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 836204076Spjd break; 837204076Spjd case BIO_WRITE: 838204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 839204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 840204076Spjd (uintmax_t)ggio->gctl_length); 841204076Spjd break; 842204076Spjd default: 843204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 844204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 845204076Spjd break; 846204076Spjd } 847204076Spjd } 848204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 849204076Spjd} 850204076Spjd 851204076Spjdstatic void 852204076Spjdremote_close(struct hast_resource *res, int ncomp) 853204076Spjd{ 854204076Spjd 855204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 856204076Spjd /* 857204076Spjd * A race is possible between dropping rlock and acquiring wlock - 858204076Spjd * another thread can close connection in-between. 859204076Spjd */ 860204076Spjd if (!ISCONNECTED(res, ncomp)) { 861204076Spjd assert(res->hr_remotein == NULL); 862204076Spjd assert(res->hr_remoteout == NULL); 863204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 864204076Spjd return; 865204076Spjd } 866204076Spjd 867204076Spjd assert(res->hr_remotein != NULL); 868204076Spjd assert(res->hr_remoteout != NULL); 869204076Spjd 870204076Spjd pjdlog_debug(2, "Closing old incoming connection to %s.", 871204076Spjd res->hr_remoteaddr); 872204076Spjd proto_close(res->hr_remotein); 873204076Spjd res->hr_remotein = NULL; 874204076Spjd pjdlog_debug(2, "Closing old outgoing connection to %s.", 875204076Spjd res->hr_remoteaddr); 876204076Spjd proto_close(res->hr_remoteout); 877204076Spjd res->hr_remoteout = NULL; 878204076Spjd 879204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 880204076Spjd 881204076Spjd /* 882204076Spjd * Stop synchronization if in-progress. 883204076Spjd */ 884211878Spjd sync_stop(); 885204076Spjd 886204076Spjd /* 887204076Spjd * Wake up guard thread, so it can immediately start reconnect. 888204076Spjd */ 889204076Spjd mtx_lock(&hio_guard_lock); 890204076Spjd cv_signal(&hio_guard_cond); 891204076Spjd mtx_unlock(&hio_guard_lock); 892204076Spjd} 893204076Spjd 894204076Spjd/* 895204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 896204076Spjd * appropriate threads: 897204076Spjd * WRITE - always goes to both local_send and remote_send threads 898204076Spjd * READ (when the block is up-to-date on local component) - 899204076Spjd * only local_send thread 900204076Spjd * READ (when the block isn't up-to-date on local component) - 901204076Spjd * only remote_send thread 902204076Spjd * DELETE - always goes to both local_send and remote_send threads 903204076Spjd * FLUSH - always goes to both local_send and remote_send threads 904204076Spjd */ 905204076Spjdstatic void * 906204076Spjdggate_recv_thread(void *arg) 907204076Spjd{ 908204076Spjd struct hast_resource *res = arg; 909204076Spjd struct g_gate_ctl_io *ggio; 910204076Spjd struct hio *hio; 911204076Spjd unsigned int ii, ncomp, ncomps; 912204076Spjd int error; 913204076Spjd 914204076Spjd ncomps = HAST_NCOMPONENTS; 915204076Spjd 916204076Spjd for (;;) { 917204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 918204076Spjd QUEUE_TAKE2(hio, free); 919204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 920204076Spjd ggio = &hio->hio_ggio; 921204076Spjd ggio->gctl_unit = res->hr_ggateunit; 922204076Spjd ggio->gctl_length = MAXPHYS; 923204076Spjd ggio->gctl_error = 0; 924204076Spjd pjdlog_debug(2, 925204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 926204076Spjd hio); 927204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 928204076Spjd if (sigexit_received) 929204076Spjd pthread_exit(NULL); 930204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 931204076Spjd } 932204076Spjd error = ggio->gctl_error; 933204076Spjd switch (error) { 934204076Spjd case 0: 935204076Spjd break; 936204076Spjd case ECANCELED: 937204076Spjd /* Exit gracefully. */ 938204076Spjd if (!sigexit_received) { 939204076Spjd pjdlog_debug(2, 940204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 941204076Spjd hio); 942204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 943204076Spjd } 944204076Spjd pthread_exit(NULL); 945204076Spjd case ENOMEM: 946204076Spjd /* 947204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 948204076Spjd * bytes - request can't be bigger than that. 949204076Spjd */ 950204076Spjd /* FALLTHROUGH */ 951204076Spjd case ENXIO: 952204076Spjd default: 953204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 954204076Spjd strerror(error)); 955204076Spjd } 956204076Spjd for (ii = 0; ii < ncomps; ii++) 957204076Spjd hio->hio_errors[ii] = EINVAL; 958204076Spjd reqlog(LOG_DEBUG, 2, ggio, 959204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 960204076Spjd hio); 961204076Spjd /* 962204076Spjd * Inform all components about new write request. 963204076Spjd * For read request prefer local component unless the given 964204076Spjd * range is out-of-date, then use remote component. 965204076Spjd */ 966204076Spjd switch (ggio->gctl_cmd) { 967204076Spjd case BIO_READ: 968204076Spjd pjdlog_debug(2, 969204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 970204076Spjd hio); 971204076Spjd refcount_init(&hio->hio_countdown, 1); 972204076Spjd mtx_lock(&metadata_lock); 973204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 974204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 975204076Spjd /* 976204076Spjd * This range is up-to-date on local component, 977204076Spjd * so handle request locally. 978204076Spjd */ 979204076Spjd /* Local component is 0 for now. */ 980204076Spjd ncomp = 0; 981204076Spjd } else /* if (res->hr_syncsrc == 982204076Spjd HAST_SYNCSRC_SECONDARY) */ { 983204076Spjd assert(res->hr_syncsrc == 984204076Spjd HAST_SYNCSRC_SECONDARY); 985204076Spjd /* 986204076Spjd * This range is out-of-date on local component, 987204076Spjd * so send request to the remote node. 988204076Spjd */ 989204076Spjd /* Remote component is 1 for now. */ 990204076Spjd ncomp = 1; 991204076Spjd } 992204076Spjd mtx_unlock(&metadata_lock); 993204076Spjd QUEUE_INSERT1(hio, send, ncomp); 994204076Spjd break; 995204076Spjd case BIO_WRITE: 996204076Spjd for (;;) { 997204076Spjd mtx_lock(&range_lock); 998204076Spjd if (rangelock_islocked(range_sync, 999204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1000204076Spjd pjdlog_debug(2, 1001204076Spjd "regular: Range offset=%jd length=%zu locked.", 1002204076Spjd (intmax_t)ggio->gctl_offset, 1003204076Spjd (size_t)ggio->gctl_length); 1004204076Spjd range_regular_wait = true; 1005204076Spjd cv_wait(&range_regular_cond, &range_lock); 1006204076Spjd range_regular_wait = false; 1007204076Spjd mtx_unlock(&range_lock); 1008204076Spjd continue; 1009204076Spjd } 1010204076Spjd if (rangelock_add(range_regular, 1011204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1012204076Spjd mtx_unlock(&range_lock); 1013204076Spjd pjdlog_debug(2, 1014204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1015204076Spjd (intmax_t)ggio->gctl_offset, 1016204076Spjd (size_t)ggio->gctl_length); 1017204076Spjd sleep(1); 1018204076Spjd continue; 1019204076Spjd } 1020204076Spjd mtx_unlock(&range_lock); 1021204076Spjd break; 1022204076Spjd } 1023204076Spjd mtx_lock(&res->hr_amp_lock); 1024204076Spjd if (activemap_write_start(res->hr_amp, 1025204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1026204076Spjd (void)hast_activemap_flush(res); 1027204076Spjd } 1028204076Spjd mtx_unlock(&res->hr_amp_lock); 1029204076Spjd /* FALLTHROUGH */ 1030204076Spjd case BIO_DELETE: 1031204076Spjd case BIO_FLUSH: 1032204076Spjd pjdlog_debug(2, 1033204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1034204076Spjd hio); 1035204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1036204076Spjd for (ii = 0; ii < ncomps; ii++) 1037204076Spjd QUEUE_INSERT1(hio, send, ii); 1038204076Spjd break; 1039204076Spjd } 1040204076Spjd } 1041204076Spjd /* NOTREACHED */ 1042204076Spjd return (NULL); 1043204076Spjd} 1044204076Spjd 1045204076Spjd/* 1046204076Spjd * Thread reads from or writes to local component. 1047204076Spjd * If local read fails, it redirects it to remote_send thread. 1048204076Spjd */ 1049204076Spjdstatic void * 1050204076Spjdlocal_send_thread(void *arg) 1051204076Spjd{ 1052204076Spjd struct hast_resource *res = arg; 1053204076Spjd struct g_gate_ctl_io *ggio; 1054204076Spjd struct hio *hio; 1055204076Spjd unsigned int ncomp, rncomp; 1056204076Spjd ssize_t ret; 1057204076Spjd 1058204076Spjd /* Local component is 0 for now. */ 1059204076Spjd ncomp = 0; 1060204076Spjd /* Remote component is 1 for now. */ 1061204076Spjd rncomp = 1; 1062204076Spjd 1063204076Spjd for (;;) { 1064204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1065204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1066204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1067204076Spjd ggio = &hio->hio_ggio; 1068204076Spjd switch (ggio->gctl_cmd) { 1069204076Spjd case BIO_READ: 1070204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1071204076Spjd ggio->gctl_length, 1072204076Spjd ggio->gctl_offset + res->hr_localoff); 1073204076Spjd if (ret == ggio->gctl_length) 1074204076Spjd hio->hio_errors[ncomp] = 0; 1075204076Spjd else { 1076204076Spjd /* 1077204076Spjd * If READ failed, try to read from remote node. 1078204076Spjd */ 1079204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1080204076Spjd continue; 1081204076Spjd } 1082204076Spjd break; 1083204076Spjd case BIO_WRITE: 1084204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1085204076Spjd ggio->gctl_length, 1086204076Spjd ggio->gctl_offset + res->hr_localoff); 1087204076Spjd if (ret < 0) 1088204076Spjd hio->hio_errors[ncomp] = errno; 1089204076Spjd else if (ret != ggio->gctl_length) 1090204076Spjd hio->hio_errors[ncomp] = EIO; 1091204076Spjd else 1092204076Spjd hio->hio_errors[ncomp] = 0; 1093204076Spjd break; 1094204076Spjd case BIO_DELETE: 1095204076Spjd ret = g_delete(res->hr_localfd, 1096204076Spjd ggio->gctl_offset + res->hr_localoff, 1097204076Spjd ggio->gctl_length); 1098204076Spjd if (ret < 0) 1099204076Spjd hio->hio_errors[ncomp] = errno; 1100204076Spjd else 1101204076Spjd hio->hio_errors[ncomp] = 0; 1102204076Spjd break; 1103204076Spjd case BIO_FLUSH: 1104204076Spjd ret = g_flush(res->hr_localfd); 1105204076Spjd if (ret < 0) 1106204076Spjd hio->hio_errors[ncomp] = errno; 1107204076Spjd else 1108204076Spjd hio->hio_errors[ncomp] = 0; 1109204076Spjd break; 1110204076Spjd } 1111204076Spjd if (refcount_release(&hio->hio_countdown)) { 1112204076Spjd if (ISSYNCREQ(hio)) { 1113204076Spjd mtx_lock(&sync_lock); 1114204076Spjd SYNCREQDONE(hio); 1115204076Spjd mtx_unlock(&sync_lock); 1116204076Spjd cv_signal(&sync_cond); 1117204076Spjd } else { 1118204076Spjd pjdlog_debug(2, 1119204076Spjd "local_send: (%p) Moving request to the done queue.", 1120204076Spjd hio); 1121204076Spjd QUEUE_INSERT2(hio, done); 1122204076Spjd } 1123204076Spjd } 1124204076Spjd } 1125204076Spjd /* NOTREACHED */ 1126204076Spjd return (NULL); 1127204076Spjd} 1128204076Spjd 1129204076Spjd/* 1130204076Spjd * Thread sends request to secondary node. 1131204076Spjd */ 1132204076Spjdstatic void * 1133204076Spjdremote_send_thread(void *arg) 1134204076Spjd{ 1135204076Spjd struct hast_resource *res = arg; 1136204076Spjd struct g_gate_ctl_io *ggio; 1137204076Spjd struct hio *hio; 1138204076Spjd struct nv *nv; 1139204076Spjd unsigned int ncomp; 1140204076Spjd bool wakeup; 1141204076Spjd uint64_t offset, length; 1142204076Spjd uint8_t cmd; 1143204076Spjd void *data; 1144204076Spjd 1145204076Spjd /* Remote component is 1 for now. */ 1146204076Spjd ncomp = 1; 1147204076Spjd 1148204076Spjd for (;;) { 1149204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1150204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1151204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1152204076Spjd ggio = &hio->hio_ggio; 1153204076Spjd switch (ggio->gctl_cmd) { 1154204076Spjd case BIO_READ: 1155204076Spjd cmd = HIO_READ; 1156204076Spjd data = NULL; 1157204076Spjd offset = ggio->gctl_offset; 1158204076Spjd length = ggio->gctl_length; 1159204076Spjd break; 1160204076Spjd case BIO_WRITE: 1161204076Spjd cmd = HIO_WRITE; 1162204076Spjd data = ggio->gctl_data; 1163204076Spjd offset = ggio->gctl_offset; 1164204076Spjd length = ggio->gctl_length; 1165204076Spjd break; 1166204076Spjd case BIO_DELETE: 1167204076Spjd cmd = HIO_DELETE; 1168204076Spjd data = NULL; 1169204076Spjd offset = ggio->gctl_offset; 1170204076Spjd length = ggio->gctl_length; 1171204076Spjd break; 1172204076Spjd case BIO_FLUSH: 1173204076Spjd cmd = HIO_FLUSH; 1174204076Spjd data = NULL; 1175204076Spjd offset = 0; 1176204076Spjd length = 0; 1177204076Spjd break; 1178204076Spjd default: 1179204076Spjd assert(!"invalid condition"); 1180204076Spjd abort(); 1181204076Spjd } 1182204076Spjd nv = nv_alloc(); 1183204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1184204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1185204076Spjd nv_add_uint64(nv, offset, "offset"); 1186204076Spjd nv_add_uint64(nv, length, "length"); 1187204076Spjd if (nv_error(nv) != 0) { 1188204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1189204076Spjd pjdlog_debug(2, 1190204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1191204076Spjd hio); 1192204076Spjd reqlog(LOG_ERR, 0, ggio, 1193204076Spjd "Unable to prepare header to send (%s): ", 1194204076Spjd strerror(nv_error(nv))); 1195204076Spjd /* Move failed request immediately to the done queue. */ 1196204076Spjd goto done_queue; 1197204076Spjd } 1198204076Spjd pjdlog_debug(2, 1199204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1200204076Spjd hio); 1201204076Spjd /* 1202204076Spjd * Protect connection from disappearing. 1203204076Spjd */ 1204204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1205204076Spjd if (!ISCONNECTED(res, ncomp)) { 1206204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1207204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1208204076Spjd goto done_queue; 1209204076Spjd } 1210204076Spjd /* 1211204076Spjd * Move the request to recv queue before sending it, because 1212204076Spjd * in different order we can get reply before we move request 1213204076Spjd * to recv queue. 1214204076Spjd */ 1215204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1216204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1217204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1218204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1219204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1220204076Spjd data != NULL ? length : 0) < 0) { 1221204076Spjd hio->hio_errors[ncomp] = errno; 1222204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1223204076Spjd remote_close(res, ncomp); 1224204076Spjd pjdlog_debug(2, 1225204076Spjd "remote_send: (%p) Unable to send request.", hio); 1226204076Spjd reqlog(LOG_ERR, 0, ggio, 1227204076Spjd "Unable to send request (%s): ", 1228204076Spjd strerror(hio->hio_errors[ncomp])); 1229204076Spjd /* 1230204076Spjd * Take request back from the receive queue and move 1231204076Spjd * it immediately to the done queue. 1232204076Spjd */ 1233204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1234204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1235204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1236204076Spjd goto done_queue; 1237204076Spjd } 1238204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1239204076Spjd nv_free(nv); 1240204076Spjd if (wakeup) 1241204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1242204076Spjd continue; 1243204076Spjddone_queue: 1244204076Spjd nv_free(nv); 1245204076Spjd if (ISSYNCREQ(hio)) { 1246204076Spjd if (!refcount_release(&hio->hio_countdown)) 1247204076Spjd continue; 1248204076Spjd mtx_lock(&sync_lock); 1249204076Spjd SYNCREQDONE(hio); 1250204076Spjd mtx_unlock(&sync_lock); 1251204076Spjd cv_signal(&sync_cond); 1252204076Spjd continue; 1253204076Spjd } 1254204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1255204076Spjd mtx_lock(&res->hr_amp_lock); 1256204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1257204076Spjd ggio->gctl_length)) { 1258204076Spjd (void)hast_activemap_flush(res); 1259204076Spjd } 1260204076Spjd mtx_unlock(&res->hr_amp_lock); 1261204076Spjd } 1262204076Spjd if (!refcount_release(&hio->hio_countdown)) 1263204076Spjd continue; 1264204076Spjd pjdlog_debug(2, 1265204076Spjd "remote_send: (%p) Moving request to the done queue.", 1266204076Spjd hio); 1267204076Spjd QUEUE_INSERT2(hio, done); 1268204076Spjd } 1269204076Spjd /* NOTREACHED */ 1270204076Spjd return (NULL); 1271204076Spjd} 1272204076Spjd 1273204076Spjd/* 1274204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1275204076Spjd * thread. 1276204076Spjd */ 1277204076Spjdstatic void * 1278204076Spjdremote_recv_thread(void *arg) 1279204076Spjd{ 1280204076Spjd struct hast_resource *res = arg; 1281204076Spjd struct g_gate_ctl_io *ggio; 1282204076Spjd struct hio *hio; 1283204076Spjd struct nv *nv; 1284204076Spjd unsigned int ncomp; 1285204076Spjd uint64_t seq; 1286204076Spjd int error; 1287204076Spjd 1288204076Spjd /* Remote component is 1 for now. */ 1289204076Spjd ncomp = 1; 1290204076Spjd 1291204076Spjd for (;;) { 1292204076Spjd /* Wait until there is anything to receive. */ 1293204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1294204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1295204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1296204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1297204076Spjd &hio_recv_list_lock[ncomp]); 1298204076Spjd } 1299204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1300204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1301204076Spjd if (!ISCONNECTED(res, ncomp)) { 1302204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1303204076Spjd /* 1304204076Spjd * Connection is dead, so move all pending requests to 1305204076Spjd * the done queue (one-by-one). 1306204076Spjd */ 1307204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1308204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1309204076Spjd assert(hio != NULL); 1310204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1311204076Spjd hio_next[ncomp]); 1312204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1313204076Spjd goto done_queue; 1314204076Spjd } 1315204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1316204076Spjd pjdlog_errno(LOG_ERR, 1317204076Spjd "Unable to receive reply header"); 1318204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1319204076Spjd remote_close(res, ncomp); 1320204076Spjd continue; 1321204076Spjd } 1322204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1323204076Spjd seq = nv_get_uint64(nv, "seq"); 1324204076Spjd if (seq == 0) { 1325204076Spjd pjdlog_error("Header contains no 'seq' field."); 1326204076Spjd nv_free(nv); 1327204076Spjd continue; 1328204076Spjd } 1329204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1330204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1331204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1332204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1333204076Spjd hio_next[ncomp]); 1334204076Spjd break; 1335204076Spjd } 1336204076Spjd } 1337204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1338204076Spjd if (hio == NULL) { 1339204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1340204076Spjd (uintmax_t)seq); 1341204076Spjd nv_free(nv); 1342204076Spjd continue; 1343204076Spjd } 1344204076Spjd error = nv_get_int16(nv, "error"); 1345204076Spjd if (error != 0) { 1346204076Spjd /* Request failed on remote side. */ 1347204076Spjd hio->hio_errors[ncomp] = 0; 1348204076Spjd nv_free(nv); 1349204076Spjd goto done_queue; 1350204076Spjd } 1351204076Spjd ggio = &hio->hio_ggio; 1352204076Spjd switch (ggio->gctl_cmd) { 1353204076Spjd case BIO_READ: 1354204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1355204076Spjd if (!ISCONNECTED(res, ncomp)) { 1356204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1357204076Spjd nv_free(nv); 1358204076Spjd goto done_queue; 1359204076Spjd } 1360204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1361204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1362204076Spjd hio->hio_errors[ncomp] = errno; 1363204076Spjd pjdlog_errno(LOG_ERR, 1364204076Spjd "Unable to receive reply data"); 1365204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1366204076Spjd nv_free(nv); 1367204076Spjd remote_close(res, ncomp); 1368204076Spjd goto done_queue; 1369204076Spjd } 1370204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1371204076Spjd break; 1372204076Spjd case BIO_WRITE: 1373204076Spjd case BIO_DELETE: 1374204076Spjd case BIO_FLUSH: 1375204076Spjd break; 1376204076Spjd default: 1377204076Spjd assert(!"invalid condition"); 1378204076Spjd abort(); 1379204076Spjd } 1380204076Spjd hio->hio_errors[ncomp] = 0; 1381204076Spjd nv_free(nv); 1382204076Spjddone_queue: 1383204076Spjd if (refcount_release(&hio->hio_countdown)) { 1384204076Spjd if (ISSYNCREQ(hio)) { 1385204076Spjd mtx_lock(&sync_lock); 1386204076Spjd SYNCREQDONE(hio); 1387204076Spjd mtx_unlock(&sync_lock); 1388204076Spjd cv_signal(&sync_cond); 1389204076Spjd } else { 1390204076Spjd pjdlog_debug(2, 1391204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1392204076Spjd hio); 1393204076Spjd QUEUE_INSERT2(hio, done); 1394204076Spjd } 1395204076Spjd } 1396204076Spjd } 1397204076Spjd /* NOTREACHED */ 1398204076Spjd return (NULL); 1399204076Spjd} 1400204076Spjd 1401204076Spjd/* 1402204076Spjd * Thread sends answer to the kernel. 1403204076Spjd */ 1404204076Spjdstatic void * 1405204076Spjdggate_send_thread(void *arg) 1406204076Spjd{ 1407204076Spjd struct hast_resource *res = arg; 1408204076Spjd struct g_gate_ctl_io *ggio; 1409204076Spjd struct hio *hio; 1410204076Spjd unsigned int ii, ncomp, ncomps; 1411204076Spjd 1412204076Spjd ncomps = HAST_NCOMPONENTS; 1413204076Spjd 1414204076Spjd for (;;) { 1415204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1416204076Spjd QUEUE_TAKE2(hio, done); 1417204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1418204076Spjd ggio = &hio->hio_ggio; 1419204076Spjd for (ii = 0; ii < ncomps; ii++) { 1420204076Spjd if (hio->hio_errors[ii] == 0) { 1421204076Spjd /* 1422204076Spjd * One successful request is enough to declare 1423204076Spjd * success. 1424204076Spjd */ 1425204076Spjd ggio->gctl_error = 0; 1426204076Spjd break; 1427204076Spjd } 1428204076Spjd } 1429204076Spjd if (ii == ncomps) { 1430204076Spjd /* 1431204076Spjd * None of the requests were successful. 1432204076Spjd * Use first error. 1433204076Spjd */ 1434204076Spjd ggio->gctl_error = hio->hio_errors[0]; 1435204076Spjd } 1436204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1437204076Spjd mtx_lock(&res->hr_amp_lock); 1438204076Spjd activemap_write_complete(res->hr_amp, 1439204076Spjd ggio->gctl_offset, ggio->gctl_length); 1440204076Spjd mtx_unlock(&res->hr_amp_lock); 1441204076Spjd } 1442204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1443204076Spjd /* 1444204076Spjd * Unlock range we locked. 1445204076Spjd */ 1446204076Spjd mtx_lock(&range_lock); 1447204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1448204076Spjd ggio->gctl_length); 1449204076Spjd if (range_sync_wait) 1450204076Spjd cv_signal(&range_sync_cond); 1451204076Spjd mtx_unlock(&range_lock); 1452204076Spjd /* 1453204076Spjd * Bump local count if this is first write after 1454204076Spjd * connection failure with remote node. 1455204076Spjd */ 1456204076Spjd ncomp = 1; 1457204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1458204076Spjd if (!ISCONNECTED(res, ncomp)) { 1459204076Spjd mtx_lock(&metadata_lock); 1460204076Spjd if (res->hr_primary_localcnt == 1461204076Spjd res->hr_secondary_remotecnt) { 1462204076Spjd res->hr_primary_localcnt++; 1463204076Spjd pjdlog_debug(1, 1464204076Spjd "Increasing localcnt to %ju.", 1465204076Spjd (uintmax_t)res->hr_primary_localcnt); 1466204076Spjd (void)metadata_write(res); 1467204076Spjd } 1468204076Spjd mtx_unlock(&metadata_lock); 1469204076Spjd } 1470204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1471204076Spjd } 1472204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1473204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1474204076Spjd pjdlog_debug(2, 1475204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1476204076Spjd QUEUE_INSERT2(hio, free); 1477204076Spjd } 1478204076Spjd /* NOTREACHED */ 1479204076Spjd return (NULL); 1480204076Spjd} 1481204076Spjd 1482204076Spjd/* 1483204076Spjd * Thread synchronize local and remote components. 1484204076Spjd */ 1485204076Spjdstatic void * 1486204076Spjdsync_thread(void *arg __unused) 1487204076Spjd{ 1488204076Spjd struct hast_resource *res = arg; 1489204076Spjd struct hio *hio; 1490204076Spjd struct g_gate_ctl_io *ggio; 1491204076Spjd unsigned int ii, ncomp, ncomps; 1492204076Spjd off_t offset, length, synced; 1493204076Spjd bool dorewind; 1494204076Spjd int syncext; 1495204076Spjd 1496204076Spjd ncomps = HAST_NCOMPONENTS; 1497204076Spjd dorewind = true; 1498211879Spjd synced = -1; 1499204076Spjd 1500204076Spjd for (;;) { 1501204076Spjd mtx_lock(&sync_lock); 1502211879Spjd if (synced == -1) 1503211879Spjd synced = 0; 1504211879Spjd else if (!sync_inprogress) { 1505211879Spjd pjdlog_info("Synchronization interrupted. " 1506211879Spjd "%jd bytes synchronized so far.", 1507211879Spjd (intmax_t)synced); 1508211879Spjd } 1509204076Spjd while (!sync_inprogress) { 1510204076Spjd dorewind = true; 1511204076Spjd synced = 0; 1512204076Spjd cv_wait(&sync_cond, &sync_lock); 1513204076Spjd } 1514204076Spjd mtx_unlock(&sync_lock); 1515204076Spjd /* 1516204076Spjd * Obtain offset at which we should synchronize. 1517204076Spjd * Rewind synchronization if needed. 1518204076Spjd */ 1519204076Spjd mtx_lock(&res->hr_amp_lock); 1520204076Spjd if (dorewind) 1521204076Spjd activemap_sync_rewind(res->hr_amp); 1522204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1523204076Spjd if (syncext != -1) { 1524204076Spjd /* 1525204076Spjd * We synchronized entire syncext extent, we can mark 1526204076Spjd * it as clean now. 1527204076Spjd */ 1528204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1529204076Spjd (void)hast_activemap_flush(res); 1530204076Spjd } 1531204076Spjd mtx_unlock(&res->hr_amp_lock); 1532204076Spjd if (dorewind) { 1533204076Spjd dorewind = false; 1534204076Spjd if (offset < 0) 1535204076Spjd pjdlog_info("Nodes are in sync."); 1536204076Spjd else { 1537204076Spjd pjdlog_info("Synchronization started. %ju bytes to go.", 1538204076Spjd (uintmax_t)(res->hr_extentsize * 1539204076Spjd activemap_ndirty(res->hr_amp))); 1540204076Spjd } 1541204076Spjd } 1542204076Spjd if (offset < 0) { 1543211878Spjd sync_stop(); 1544204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1545204076Spjd /* 1546204076Spjd * Synchronization complete, make both localcnt and 1547204076Spjd * remotecnt equal. 1548204076Spjd */ 1549204076Spjd ncomp = 1; 1550204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1551204076Spjd if (ISCONNECTED(res, ncomp)) { 1552204076Spjd if (synced > 0) { 1553204076Spjd pjdlog_info("Synchronization complete. " 1554204076Spjd "%jd bytes synchronized.", 1555204076Spjd (intmax_t)synced); 1556204076Spjd } 1557204076Spjd mtx_lock(&metadata_lock); 1558204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1559204076Spjd res->hr_primary_localcnt = 1560204076Spjd res->hr_secondary_localcnt; 1561204076Spjd res->hr_primary_remotecnt = 1562204076Spjd res->hr_secondary_remotecnt; 1563204076Spjd pjdlog_debug(1, 1564204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1565204076Spjd (uintmax_t)res->hr_primary_localcnt, 1566204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1567204076Spjd (void)metadata_write(res); 1568204076Spjd mtx_unlock(&metadata_lock); 1569204076Spjd } 1570204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1571204076Spjd continue; 1572204076Spjd } 1573204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1574204076Spjd QUEUE_TAKE2(hio, free); 1575204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1576204076Spjd /* 1577204076Spjd * Lock the range we are going to synchronize. We don't want 1578204076Spjd * race where someone writes between our read and write. 1579204076Spjd */ 1580204076Spjd for (;;) { 1581204076Spjd mtx_lock(&range_lock); 1582204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1583204076Spjd pjdlog_debug(2, 1584204076Spjd "sync: Range offset=%jd length=%jd locked.", 1585204076Spjd (intmax_t)offset, (intmax_t)length); 1586204076Spjd range_sync_wait = true; 1587204076Spjd cv_wait(&range_sync_cond, &range_lock); 1588204076Spjd range_sync_wait = false; 1589204076Spjd mtx_unlock(&range_lock); 1590204076Spjd continue; 1591204076Spjd } 1592204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1593204076Spjd mtx_unlock(&range_lock); 1594204076Spjd pjdlog_debug(2, 1595204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1596204076Spjd (intmax_t)offset, (intmax_t)length); 1597204076Spjd sleep(1); 1598204076Spjd continue; 1599204076Spjd } 1600204076Spjd mtx_unlock(&range_lock); 1601204076Spjd break; 1602204076Spjd } 1603204076Spjd /* 1604204076Spjd * First read the data from synchronization source. 1605204076Spjd */ 1606204076Spjd SYNCREQ(hio); 1607204076Spjd ggio = &hio->hio_ggio; 1608204076Spjd ggio->gctl_cmd = BIO_READ; 1609204076Spjd ggio->gctl_offset = offset; 1610204076Spjd ggio->gctl_length = length; 1611204076Spjd ggio->gctl_error = 0; 1612204076Spjd for (ii = 0; ii < ncomps; ii++) 1613204076Spjd hio->hio_errors[ii] = EINVAL; 1614204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1615204076Spjd hio); 1616204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1617204076Spjd hio); 1618204076Spjd mtx_lock(&metadata_lock); 1619204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1620204076Spjd /* 1621204076Spjd * This range is up-to-date on local component, 1622204076Spjd * so handle request locally. 1623204076Spjd */ 1624204076Spjd /* Local component is 0 for now. */ 1625204076Spjd ncomp = 0; 1626204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1627204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1628204076Spjd /* 1629204076Spjd * This range is out-of-date on local component, 1630204076Spjd * so send request to the remote node. 1631204076Spjd */ 1632204076Spjd /* Remote component is 1 for now. */ 1633204076Spjd ncomp = 1; 1634204076Spjd } 1635204076Spjd mtx_unlock(&metadata_lock); 1636204076Spjd refcount_init(&hio->hio_countdown, 1); 1637204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1638204076Spjd 1639204076Spjd /* 1640204076Spjd * Let's wait for READ to finish. 1641204076Spjd */ 1642204076Spjd mtx_lock(&sync_lock); 1643204076Spjd while (!ISSYNCREQDONE(hio)) 1644204076Spjd cv_wait(&sync_cond, &sync_lock); 1645204076Spjd mtx_unlock(&sync_lock); 1646204076Spjd 1647204076Spjd if (hio->hio_errors[ncomp] != 0) { 1648204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1649204076Spjd strerror(hio->hio_errors[ncomp])); 1650204076Spjd goto free_queue; 1651204076Spjd } 1652204076Spjd 1653204076Spjd /* 1654204076Spjd * We read the data from synchronization source, now write it 1655204076Spjd * to synchronization target. 1656204076Spjd */ 1657204076Spjd SYNCREQ(hio); 1658204076Spjd ggio->gctl_cmd = BIO_WRITE; 1659204076Spjd for (ii = 0; ii < ncomps; ii++) 1660204076Spjd hio->hio_errors[ii] = EINVAL; 1661204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1662204076Spjd hio); 1663204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1664204076Spjd hio); 1665204076Spjd mtx_lock(&metadata_lock); 1666204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1667204076Spjd /* 1668204076Spjd * This range is up-to-date on local component, 1669204076Spjd * so we update remote component. 1670204076Spjd */ 1671204076Spjd /* Remote component is 1 for now. */ 1672204076Spjd ncomp = 1; 1673204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1674204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1675204076Spjd /* 1676204076Spjd * This range is out-of-date on local component, 1677204076Spjd * so we update it. 1678204076Spjd */ 1679204076Spjd /* Local component is 0 for now. */ 1680204076Spjd ncomp = 0; 1681204076Spjd } 1682204076Spjd mtx_unlock(&metadata_lock); 1683204076Spjd 1684204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1685204076Spjd hio); 1686204076Spjd refcount_init(&hio->hio_countdown, 1); 1687204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1688204076Spjd 1689204076Spjd /* 1690204076Spjd * Let's wait for WRITE to finish. 1691204076Spjd */ 1692204076Spjd mtx_lock(&sync_lock); 1693204076Spjd while (!ISSYNCREQDONE(hio)) 1694204076Spjd cv_wait(&sync_cond, &sync_lock); 1695204076Spjd mtx_unlock(&sync_lock); 1696204076Spjd 1697204076Spjd if (hio->hio_errors[ncomp] != 0) { 1698204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1699204076Spjd strerror(hio->hio_errors[ncomp])); 1700204076Spjd goto free_queue; 1701204076Spjd } 1702211880Spjd 1703211880Spjd synced += length; 1704204076Spjdfree_queue: 1705204076Spjd mtx_lock(&range_lock); 1706204076Spjd rangelock_del(range_sync, offset, length); 1707204076Spjd if (range_regular_wait) 1708204076Spjd cv_signal(&range_regular_cond); 1709204076Spjd mtx_unlock(&range_lock); 1710204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1711204076Spjd hio); 1712204076Spjd QUEUE_INSERT2(hio, free); 1713204076Spjd } 1714204076Spjd /* NOTREACHED */ 1715204076Spjd return (NULL); 1716204076Spjd} 1717204076Spjd 1718204076Spjdstatic void 1719204076Spjdsighandler(int sig) 1720204076Spjd{ 1721204076Spjd bool unlock; 1722204076Spjd 1723204076Spjd switch (sig) { 1724204076Spjd case SIGINT: 1725204076Spjd case SIGTERM: 1726204076Spjd sigexit_received = true; 1727204076Spjd break; 1728210886Spjd case SIGHUP: 1729210886Spjd sighup_received = true; 1730210886Spjd break; 1731204076Spjd default: 1732204076Spjd assert(!"invalid condition"); 1733204076Spjd } 1734204076Spjd /* 1735204076Spjd * XXX: Racy, but if we cannot obtain hio_guard_lock here, we don't 1736204076Spjd * want to risk deadlock. 1737204076Spjd */ 1738204076Spjd unlock = mtx_trylock(&hio_guard_lock); 1739204076Spjd cv_signal(&hio_guard_cond); 1740204076Spjd if (unlock) 1741204076Spjd mtx_unlock(&hio_guard_lock); 1742204076Spjd} 1743204076Spjd 1744210886Spjdstatic void 1745210886Spjdconfig_reload(void) 1746210886Spjd{ 1747210886Spjd struct hastd_config *newcfg; 1748210886Spjd struct hast_resource *res; 1749210886Spjd unsigned int ii, ncomps; 1750210886Spjd int modified; 1751210886Spjd 1752210886Spjd pjdlog_info("Reloading configuration..."); 1753210886Spjd 1754210886Spjd ncomps = HAST_NCOMPONENTS; 1755210886Spjd 1756210886Spjd newcfg = yy_config_parse(cfgpath, false); 1757210886Spjd if (newcfg == NULL) 1758210886Spjd goto failed; 1759210886Spjd 1760210886Spjd TAILQ_FOREACH(res, &newcfg->hc_resources, hr_next) { 1761210886Spjd if (strcmp(res->hr_name, gres->hr_name) == 0) 1762210886Spjd break; 1763210886Spjd } 1764210886Spjd /* 1765210886Spjd * If resource was removed from the configuration file, resource 1766210886Spjd * name, provider name or path to local component was modified we 1767210886Spjd * shouldn't be here. This means that someone modified configuration 1768210886Spjd * file and send SIGHUP to us instead of main hastd process. 1769210886Spjd * Log advice and ignore the signal. 1770210886Spjd */ 1771210886Spjd if (res == NULL || strcmp(gres->hr_name, res->hr_name) != 0 || 1772210886Spjd strcmp(gres->hr_provname, res->hr_provname) != 0 || 1773210886Spjd strcmp(gres->hr_localpath, res->hr_localpath) != 0) { 1774210886Spjd pjdlog_warning("To reload configuration send SIGHUP to the main hastd process (pid %u).", 1775210886Spjd (unsigned int)getppid()); 1776210886Spjd goto failed; 1777210886Spjd } 1778210886Spjd 1779210886Spjd#define MODIFIED_REMOTEADDR 0x1 1780210886Spjd#define MODIFIED_REPLICATION 0x2 1781210886Spjd#define MODIFIED_TIMEOUT 0x4 1782210886Spjd modified = 0; 1783210886Spjd if (strcmp(gres->hr_remoteaddr, res->hr_remoteaddr) != 0) { 1784210886Spjd /* 1785210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1786210886Spjd * We want remote_close() to log disconnect from the old 1787210886Spjd * addresses, not from the new ones. 1788210886Spjd */ 1789210886Spjd modified |= MODIFIED_REMOTEADDR; 1790210886Spjd } 1791210886Spjd if (gres->hr_replication != res->hr_replication) { 1792210886Spjd gres->hr_replication = res->hr_replication; 1793210886Spjd modified |= MODIFIED_REPLICATION; 1794210886Spjd } 1795210886Spjd if (gres->hr_timeout != res->hr_timeout) { 1796210886Spjd gres->hr_timeout = res->hr_timeout; 1797210886Spjd modified |= MODIFIED_TIMEOUT; 1798210886Spjd } 1799210886Spjd /* 1800210886Spjd * If only timeout was modified we only need to change it without 1801210886Spjd * reconnecting. 1802210886Spjd */ 1803210886Spjd if (modified == MODIFIED_TIMEOUT) { 1804210886Spjd for (ii = 0; ii < ncomps; ii++) { 1805210886Spjd if (!ISREMOTE(ii)) 1806210886Spjd continue; 1807210886Spjd rw_rlock(&hio_remote_lock[ii]); 1808210886Spjd if (!ISCONNECTED(gres, ii)) { 1809210886Spjd rw_unlock(&hio_remote_lock[ii]); 1810210886Spjd continue; 1811210886Spjd } 1812210886Spjd rw_unlock(&hio_remote_lock[ii]); 1813210886Spjd if (proto_timeout(gres->hr_remotein, 1814210886Spjd gres->hr_timeout) < 0) { 1815210886Spjd pjdlog_errno(LOG_WARNING, 1816210886Spjd "Unable to set connection timeout"); 1817210886Spjd } 1818210886Spjd if (proto_timeout(gres->hr_remoteout, 1819210886Spjd gres->hr_timeout) < 0) { 1820210886Spjd pjdlog_errno(LOG_WARNING, 1821210886Spjd "Unable to set connection timeout"); 1822210886Spjd } 1823210886Spjd } 1824210886Spjd } else { 1825210886Spjd for (ii = 0; ii < ncomps; ii++) { 1826210886Spjd if (!ISREMOTE(ii)) 1827210886Spjd continue; 1828210886Spjd remote_close(gres, ii); 1829210886Spjd } 1830210886Spjd if (modified & MODIFIED_REMOTEADDR) { 1831210886Spjd strlcpy(gres->hr_remoteaddr, res->hr_remoteaddr, 1832210886Spjd sizeof(gres->hr_remoteaddr)); 1833210886Spjd } 1834210886Spjd } 1835210886Spjd#undef MODIFIED_REMOTEADDR 1836210886Spjd#undef MODIFIED_REPLICATION 1837210886Spjd#undef MODIFIED_TIMEOUT 1838210886Spjd 1839210886Spjd pjdlog_info("Configuration reloaded successfully."); 1840210886Spjd return; 1841210886Spjdfailed: 1842210886Spjd if (newcfg != NULL) { 1843210886Spjd if (newcfg->hc_controlconn != NULL) 1844210886Spjd proto_close(newcfg->hc_controlconn); 1845210886Spjd if (newcfg->hc_listenconn != NULL) 1846210886Spjd proto_close(newcfg->hc_listenconn); 1847210886Spjd yy_config_free(newcfg); 1848210886Spjd } 1849210886Spjd pjdlog_warning("Configuration not reloaded."); 1850210886Spjd} 1851210886Spjd 1852204076Spjd/* 1853204076Spjd * Thread guards remote connections and reconnects when needed, handles 1854204076Spjd * signals, etc. 1855204076Spjd */ 1856204076Spjdstatic void * 1857204076Spjdguard_thread(void *arg) 1858204076Spjd{ 1859204076Spjd struct hast_resource *res = arg; 1860205738Spjd struct proto_conn *in, *out; 1861204076Spjd unsigned int ii, ncomps; 1862204076Spjd int timeout; 1863204076Spjd 1864204076Spjd ncomps = HAST_NCOMPONENTS; 1865204076Spjd 1866204076Spjd for (;;) { 1867204076Spjd if (sigexit_received) { 1868204076Spjd primary_exitx(EX_OK, 1869204076Spjd "Termination signal received, exiting."); 1870204076Spjd } 1871210886Spjd if (sighup_received) { 1872210886Spjd sighup_received = false; 1873210886Spjd config_reload(); 1874210886Spjd } 1875204076Spjd /* 1876204076Spjd * If all the connection will be fine, we will sleep until 1877204076Spjd * someone wakes us up. 1878204076Spjd * If any of the connections will be broken and we won't be 1879204076Spjd * able to connect, we will sleep only for RECONNECT_SLEEP 1880204076Spjd * seconds so we can retry soon. 1881204076Spjd */ 1882204076Spjd timeout = 0; 1883204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 1884204076Spjd mtx_lock(&hio_guard_lock); 1885204076Spjd for (ii = 0; ii < ncomps; ii++) { 1886204076Spjd if (!ISREMOTE(ii)) 1887204076Spjd continue; 1888204076Spjd rw_rlock(&hio_remote_lock[ii]); 1889204076Spjd if (ISCONNECTED(res, ii)) { 1890204076Spjd assert(res->hr_remotein != NULL); 1891204076Spjd assert(res->hr_remoteout != NULL); 1892204076Spjd rw_unlock(&hio_remote_lock[ii]); 1893204076Spjd pjdlog_debug(2, 1894204076Spjd "remote_guard: Connection to %s is ok.", 1895204076Spjd res->hr_remoteaddr); 1896210881Spjd } else if (real_remote(res)) { 1897204076Spjd assert(res->hr_remotein == NULL); 1898204076Spjd assert(res->hr_remoteout == NULL); 1899204076Spjd /* 1900204076Spjd * Upgrade the lock. It doesn't have to be 1901204076Spjd * atomic as no other thread can change 1902204076Spjd * connection status from disconnected to 1903204076Spjd * connected. 1904204076Spjd */ 1905204076Spjd rw_unlock(&hio_remote_lock[ii]); 1906204076Spjd pjdlog_debug(2, 1907204076Spjd "remote_guard: Reconnecting to %s.", 1908204076Spjd res->hr_remoteaddr); 1909205738Spjd in = out = NULL; 1910205738Spjd if (init_remote(res, &in, &out)) { 1911205738Spjd rw_wlock(&hio_remote_lock[ii]); 1912205738Spjd assert(res->hr_remotein == NULL); 1913205738Spjd assert(res->hr_remoteout == NULL); 1914205738Spjd assert(in != NULL && out != NULL); 1915205738Spjd res->hr_remotein = in; 1916205738Spjd res->hr_remoteout = out; 1917205738Spjd rw_unlock(&hio_remote_lock[ii]); 1918204076Spjd pjdlog_info("Successfully reconnected to %s.", 1919204076Spjd res->hr_remoteaddr); 1920205738Spjd sync_start(); 1921204076Spjd } else { 1922204076Spjd /* Both connections should be NULL. */ 1923204076Spjd assert(res->hr_remotein == NULL); 1924204076Spjd assert(res->hr_remoteout == NULL); 1925205738Spjd assert(in == NULL && out == NULL); 1926204076Spjd pjdlog_debug(2, 1927204076Spjd "remote_guard: Reconnect to %s failed.", 1928204076Spjd res->hr_remoteaddr); 1929204076Spjd timeout = RECONNECT_SLEEP; 1930204076Spjd } 1931210881Spjd } else { 1932210881Spjd rw_unlock(&hio_remote_lock[ii]); 1933204076Spjd } 1934204076Spjd } 1935204076Spjd (void)cv_timedwait(&hio_guard_cond, &hio_guard_lock, timeout); 1936204076Spjd mtx_unlock(&hio_guard_lock); 1937204076Spjd } 1938204076Spjd /* NOTREACHED */ 1939204076Spjd return (NULL); 1940204076Spjd} 1941