primary.c revision 211897
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 211897 2010-08-27 21:20:32Z 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" 63211886Spjd#include "hooks.h" 64204076Spjd#include "metadata.h" 65204076Spjd#include "proto.h" 66204076Spjd#include "pjdlog.h" 67204076Spjd#include "subr.h" 68204076Spjd#include "synch.h" 69204076Spjd 70210886Spjd/* The is only one remote component for now. */ 71210886Spjd#define ISREMOTE(no) ((no) == 1) 72210886Spjd 73204076Spjdstruct hio { 74204076Spjd /* 75204076Spjd * Number of components we are still waiting for. 76204076Spjd * When this field goes to 0, we can send the request back to the 77204076Spjd * kernel. Each component has to decrease this counter by one 78204076Spjd * even on failure. 79204076Spjd */ 80204076Spjd unsigned int hio_countdown; 81204076Spjd /* 82204076Spjd * Each component has a place to store its own error. 83204076Spjd * Once the request is handled by all components we can decide if the 84204076Spjd * request overall is successful or not. 85204076Spjd */ 86204076Spjd int *hio_errors; 87204076Spjd /* 88204076Spjd * Structure used to comunicate with GEOM Gate class. 89204076Spjd */ 90204076Spjd struct g_gate_ctl_io hio_ggio; 91204076Spjd TAILQ_ENTRY(hio) *hio_next; 92204076Spjd}; 93204076Spjd#define hio_free_next hio_next[0] 94204076Spjd#define hio_done_next hio_next[0] 95204076Spjd 96204076Spjd/* 97204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 98204076Spjd * until some in-progress requests are freed. 99204076Spjd */ 100204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 101204076Spjdstatic pthread_mutex_t hio_free_list_lock; 102204076Spjdstatic pthread_cond_t hio_free_list_cond; 103204076Spjd/* 104204076Spjd * There is one send list for every component. One requests is placed on all 105204076Spjd * send lists - each component gets the same request, but each component is 106204076Spjd * responsible for managing his own send list. 107204076Spjd */ 108204076Spjdstatic TAILQ_HEAD(, hio) *hio_send_list; 109204076Spjdstatic pthread_mutex_t *hio_send_list_lock; 110204076Spjdstatic pthread_cond_t *hio_send_list_cond; 111204076Spjd/* 112204076Spjd * There is one recv list for every component, although local components don't 113204076Spjd * use recv lists as local requests are done synchronously. 114204076Spjd */ 115204076Spjdstatic TAILQ_HEAD(, hio) *hio_recv_list; 116204076Spjdstatic pthread_mutex_t *hio_recv_list_lock; 117204076Spjdstatic pthread_cond_t *hio_recv_list_cond; 118204076Spjd/* 119204076Spjd * Request is placed on done list by the slowest component (the one that 120204076Spjd * decreased hio_countdown from 1 to 0). 121204076Spjd */ 122204076Spjdstatic TAILQ_HEAD(, hio) hio_done_list; 123204076Spjdstatic pthread_mutex_t hio_done_list_lock; 124204076Spjdstatic pthread_cond_t hio_done_list_cond; 125204076Spjd/* 126204076Spjd * Structure below are for interaction with sync thread. 127204076Spjd */ 128204076Spjdstatic bool sync_inprogress; 129204076Spjdstatic pthread_mutex_t sync_lock; 130204076Spjdstatic pthread_cond_t sync_cond; 131204076Spjd/* 132204076Spjd * The lock below allows to synchornize access to remote connections. 133204076Spjd */ 134204076Spjdstatic pthread_rwlock_t *hio_remote_lock; 135204076Spjdstatic pthread_mutex_t hio_guard_lock; 136204076Spjdstatic pthread_cond_t hio_guard_cond; 137204076Spjd 138204076Spjd/* 139204076Spjd * Lock to synchronize metadata updates. Also synchronize access to 140204076Spjd * hr_primary_localcnt and hr_primary_remotecnt fields. 141204076Spjd */ 142204076Spjdstatic pthread_mutex_t metadata_lock; 143204076Spjd 144204076Spjd/* 145204076Spjd * Maximum number of outstanding I/O requests. 146204076Spjd */ 147204076Spjd#define HAST_HIO_MAX 256 148204076Spjd/* 149204076Spjd * Number of components. At this point there are only two components: local 150204076Spjd * and remote, but in the future it might be possible to use multiple local 151204076Spjd * and remote components. 152204076Spjd */ 153204076Spjd#define HAST_NCOMPONENTS 2 154204076Spjd/* 155211882Spjd * Number of seconds to sleep between keepalive packets. 156204076Spjd */ 157211882Spjd#define KEEPALIVE_SLEEP 10 158211882Spjd/* 159211882Spjd * Number of seconds to sleep between reconnect retries. 160211882Spjd */ 161204076Spjd#define RECONNECT_SLEEP 5 162204076Spjd 163204076Spjd#define ISCONNECTED(res, no) \ 164204076Spjd ((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL) 165204076Spjd 166204076Spjd#define QUEUE_INSERT1(hio, name, ncomp) do { \ 167204076Spjd bool _wakeup; \ 168204076Spjd \ 169204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 170204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]); \ 171204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio), \ 172204076Spjd hio_next[(ncomp)]); \ 173204076Spjd mtx_unlock(&hio_##name##_list_lock[ncomp]); \ 174204076Spjd if (_wakeup) \ 175204076Spjd cv_signal(&hio_##name##_list_cond[(ncomp)]); \ 176204076Spjd} while (0) 177204076Spjd#define QUEUE_INSERT2(hio, name) do { \ 178204076Spjd bool _wakeup; \ 179204076Spjd \ 180204076Spjd mtx_lock(&hio_##name##_list_lock); \ 181204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 182204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\ 183204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 184204076Spjd if (_wakeup) \ 185204076Spjd cv_signal(&hio_##name##_list_cond); \ 186204076Spjd} while (0) 187204076Spjd#define QUEUE_TAKE1(hio, name, ncomp) do { \ 188204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 189204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL) { \ 190204076Spjd cv_wait(&hio_##name##_list_cond[(ncomp)], \ 191204076Spjd &hio_##name##_list_lock[(ncomp)]); \ 192204076Spjd } \ 193204076Spjd TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio), \ 194204076Spjd hio_next[(ncomp)]); \ 195204076Spjd mtx_unlock(&hio_##name##_list_lock[(ncomp)]); \ 196204076Spjd} while (0) 197204076Spjd#define QUEUE_TAKE2(hio, name) do { \ 198204076Spjd mtx_lock(&hio_##name##_list_lock); \ 199204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 200204076Spjd cv_wait(&hio_##name##_list_cond, \ 201204076Spjd &hio_##name##_list_lock); \ 202204076Spjd } \ 203204076Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next); \ 204204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 205204076Spjd} while (0) 206204076Spjd 207209183Spjd#define SYNCREQ(hio) do { \ 208209183Spjd (hio)->hio_ggio.gctl_unit = -1; \ 209209183Spjd (hio)->hio_ggio.gctl_seq = 1; \ 210209183Spjd} while (0) 211204076Spjd#define ISSYNCREQ(hio) ((hio)->hio_ggio.gctl_unit == -1) 212204076Spjd#define SYNCREQDONE(hio) do { (hio)->hio_ggio.gctl_unit = -2; } while (0) 213204076Spjd#define ISSYNCREQDONE(hio) ((hio)->hio_ggio.gctl_unit == -2) 214204076Spjd 215204076Spjdstatic struct hast_resource *gres; 216204076Spjd 217204076Spjdstatic pthread_mutex_t range_lock; 218204076Spjdstatic struct rangelocks *range_regular; 219204076Spjdstatic bool range_regular_wait; 220204076Spjdstatic pthread_cond_t range_regular_cond; 221204076Spjdstatic struct rangelocks *range_sync; 222204076Spjdstatic bool range_sync_wait; 223204076Spjdstatic pthread_cond_t range_sync_cond; 224204076Spjd 225204076Spjdstatic void *ggate_recv_thread(void *arg); 226204076Spjdstatic void *local_send_thread(void *arg); 227204076Spjdstatic void *remote_send_thread(void *arg); 228204076Spjdstatic void *remote_recv_thread(void *arg); 229204076Spjdstatic void *ggate_send_thread(void *arg); 230204076Spjdstatic void *sync_thread(void *arg); 231204076Spjdstatic void *guard_thread(void *arg); 232204076Spjd 233204076Spjdstatic void sighandler(int sig); 234204076Spjd 235204076Spjdstatic void 236204076Spjdcleanup(struct hast_resource *res) 237204076Spjd{ 238204076Spjd int rerrno; 239204076Spjd 240204076Spjd /* Remember errno. */ 241204076Spjd rerrno = errno; 242204076Spjd 243204076Spjd /* 244204076Spjd * Close descriptor to /dev/hast/<name> 245204076Spjd * to work-around race in the kernel. 246204076Spjd */ 247204076Spjd close(res->hr_localfd); 248204076Spjd 249204076Spjd /* Destroy ggate provider if we created one. */ 250204076Spjd if (res->hr_ggateunit >= 0) { 251204076Spjd struct g_gate_ctl_destroy ggiod; 252204076Spjd 253204076Spjd ggiod.gctl_version = G_GATE_VERSION; 254204076Spjd ggiod.gctl_unit = res->hr_ggateunit; 255204076Spjd ggiod.gctl_force = 1; 256204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 257204076Spjd pjdlog_warning("Unable to destroy hast/%s device", 258204076Spjd res->hr_provname); 259204076Spjd } 260204076Spjd res->hr_ggateunit = -1; 261204076Spjd } 262204076Spjd 263204076Spjd /* Restore errno. */ 264204076Spjd errno = rerrno; 265204076Spjd} 266204076Spjd 267204076Spjdstatic void 268204076Spjdprimary_exit(int exitcode, const char *fmt, ...) 269204076Spjd{ 270204076Spjd va_list ap; 271204076Spjd 272204076Spjd assert(exitcode != EX_OK); 273204076Spjd va_start(ap, fmt); 274204076Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 275204076Spjd va_end(ap); 276204076Spjd cleanup(gres); 277204076Spjd exit(exitcode); 278204076Spjd} 279204076Spjd 280204076Spjdstatic void 281204076Spjdprimary_exitx(int exitcode, const char *fmt, ...) 282204076Spjd{ 283204076Spjd va_list ap; 284204076Spjd 285204076Spjd va_start(ap, fmt); 286204076Spjd pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 287204076Spjd va_end(ap); 288204076Spjd cleanup(gres); 289204076Spjd exit(exitcode); 290204076Spjd} 291204076Spjd 292204076Spjdstatic int 293204076Spjdhast_activemap_flush(struct hast_resource *res) 294204076Spjd{ 295204076Spjd const unsigned char *buf; 296204076Spjd size_t size; 297204076Spjd 298204076Spjd buf = activemap_bitmap(res->hr_amp, &size); 299204076Spjd assert(buf != NULL); 300204076Spjd assert((size % res->hr_local_sectorsize) == 0); 301204076Spjd if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 302204076Spjd (ssize_t)size) { 303204076Spjd KEEP_ERRNO(pjdlog_errno(LOG_ERR, 304204076Spjd "Unable to flush activemap to disk")); 305204076Spjd return (-1); 306204076Spjd } 307204076Spjd return (0); 308204076Spjd} 309204076Spjd 310210881Spjdstatic bool 311210881Spjdreal_remote(const struct hast_resource *res) 312210881Spjd{ 313210881Spjd 314210881Spjd return (strcmp(res->hr_remoteaddr, "none") != 0); 315210881Spjd} 316210881Spjd 317204076Spjdstatic void 318204076Spjdinit_environment(struct hast_resource *res __unused) 319204076Spjd{ 320204076Spjd struct hio *hio; 321204076Spjd unsigned int ii, ncomps; 322204076Spjd 323204076Spjd /* 324204076Spjd * In the future it might be per-resource value. 325204076Spjd */ 326204076Spjd ncomps = HAST_NCOMPONENTS; 327204076Spjd 328204076Spjd /* 329204076Spjd * Allocate memory needed by lists. 330204076Spjd */ 331204076Spjd hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 332204076Spjd if (hio_send_list == NULL) { 333204076Spjd primary_exitx(EX_TEMPFAIL, 334204076Spjd "Unable to allocate %zu bytes of memory for send lists.", 335204076Spjd sizeof(hio_send_list[0]) * ncomps); 336204076Spjd } 337204076Spjd hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 338204076Spjd if (hio_send_list_lock == NULL) { 339204076Spjd primary_exitx(EX_TEMPFAIL, 340204076Spjd "Unable to allocate %zu bytes of memory for send list locks.", 341204076Spjd sizeof(hio_send_list_lock[0]) * ncomps); 342204076Spjd } 343204076Spjd hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 344204076Spjd if (hio_send_list_cond == NULL) { 345204076Spjd primary_exitx(EX_TEMPFAIL, 346204076Spjd "Unable to allocate %zu bytes of memory for send list condition variables.", 347204076Spjd sizeof(hio_send_list_cond[0]) * ncomps); 348204076Spjd } 349204076Spjd hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 350204076Spjd if (hio_recv_list == NULL) { 351204076Spjd primary_exitx(EX_TEMPFAIL, 352204076Spjd "Unable to allocate %zu bytes of memory for recv lists.", 353204076Spjd sizeof(hio_recv_list[0]) * ncomps); 354204076Spjd } 355204076Spjd hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 356204076Spjd if (hio_recv_list_lock == NULL) { 357204076Spjd primary_exitx(EX_TEMPFAIL, 358204076Spjd "Unable to allocate %zu bytes of memory for recv list locks.", 359204076Spjd sizeof(hio_recv_list_lock[0]) * ncomps); 360204076Spjd } 361204076Spjd hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 362204076Spjd if (hio_recv_list_cond == NULL) { 363204076Spjd primary_exitx(EX_TEMPFAIL, 364204076Spjd "Unable to allocate %zu bytes of memory for recv list condition variables.", 365204076Spjd sizeof(hio_recv_list_cond[0]) * ncomps); 366204076Spjd } 367204076Spjd hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 368204076Spjd if (hio_remote_lock == NULL) { 369204076Spjd primary_exitx(EX_TEMPFAIL, 370204076Spjd "Unable to allocate %zu bytes of memory for remote connections locks.", 371204076Spjd sizeof(hio_remote_lock[0]) * ncomps); 372204076Spjd } 373204076Spjd 374204076Spjd /* 375204076Spjd * Initialize lists, their locks and theirs condition variables. 376204076Spjd */ 377204076Spjd TAILQ_INIT(&hio_free_list); 378204076Spjd mtx_init(&hio_free_list_lock); 379204076Spjd cv_init(&hio_free_list_cond); 380204076Spjd for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 381204076Spjd TAILQ_INIT(&hio_send_list[ii]); 382204076Spjd mtx_init(&hio_send_list_lock[ii]); 383204076Spjd cv_init(&hio_send_list_cond[ii]); 384204076Spjd TAILQ_INIT(&hio_recv_list[ii]); 385204076Spjd mtx_init(&hio_recv_list_lock[ii]); 386204076Spjd cv_init(&hio_recv_list_cond[ii]); 387204076Spjd rw_init(&hio_remote_lock[ii]); 388204076Spjd } 389204076Spjd TAILQ_INIT(&hio_done_list); 390204076Spjd mtx_init(&hio_done_list_lock); 391204076Spjd cv_init(&hio_done_list_cond); 392204076Spjd mtx_init(&hio_guard_lock); 393204076Spjd cv_init(&hio_guard_cond); 394204076Spjd mtx_init(&metadata_lock); 395204076Spjd 396204076Spjd /* 397204076Spjd * Allocate requests pool and initialize requests. 398204076Spjd */ 399204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 400204076Spjd hio = malloc(sizeof(*hio)); 401204076Spjd if (hio == NULL) { 402204076Spjd primary_exitx(EX_TEMPFAIL, 403204076Spjd "Unable to allocate %zu bytes of memory for hio request.", 404204076Spjd sizeof(*hio)); 405204076Spjd } 406204076Spjd hio->hio_countdown = 0; 407204076Spjd hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 408204076Spjd if (hio->hio_errors == NULL) { 409204076Spjd primary_exitx(EX_TEMPFAIL, 410204076Spjd "Unable allocate %zu bytes of memory for hio errors.", 411204076Spjd sizeof(hio->hio_errors[0]) * ncomps); 412204076Spjd } 413204076Spjd hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 414204076Spjd if (hio->hio_next == NULL) { 415204076Spjd primary_exitx(EX_TEMPFAIL, 416204076Spjd "Unable allocate %zu bytes of memory for hio_next field.", 417204076Spjd sizeof(hio->hio_next[0]) * ncomps); 418204076Spjd } 419204076Spjd hio->hio_ggio.gctl_version = G_GATE_VERSION; 420204076Spjd hio->hio_ggio.gctl_data = malloc(MAXPHYS); 421204076Spjd if (hio->hio_ggio.gctl_data == NULL) { 422204076Spjd primary_exitx(EX_TEMPFAIL, 423204076Spjd "Unable to allocate %zu bytes of memory for gctl_data.", 424204076Spjd MAXPHYS); 425204076Spjd } 426204076Spjd hio->hio_ggio.gctl_length = MAXPHYS; 427204076Spjd hio->hio_ggio.gctl_error = 0; 428204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 429204076Spjd } 430204076Spjd 431204076Spjd /* 432204076Spjd * Turn on signals handling. 433204076Spjd */ 434204076Spjd signal(SIGINT, sighandler); 435204076Spjd signal(SIGTERM, sighandler); 436210886Spjd signal(SIGHUP, sighandler); 437211886Spjd signal(SIGCHLD, sighandler); 438204076Spjd} 439204076Spjd 440204076Spjdstatic void 441204076Spjdinit_local(struct hast_resource *res) 442204076Spjd{ 443204076Spjd unsigned char *buf; 444204076Spjd size_t mapsize; 445204076Spjd 446204076Spjd if (metadata_read(res, true) < 0) 447204076Spjd exit(EX_NOINPUT); 448204076Spjd mtx_init(&res->hr_amp_lock); 449204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 450204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 451204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 452204076Spjd } 453204076Spjd mtx_init(&range_lock); 454204076Spjd cv_init(&range_regular_cond); 455204076Spjd if (rangelock_init(&range_regular) < 0) 456204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 457204076Spjd cv_init(&range_sync_cond); 458204076Spjd if (rangelock_init(&range_sync) < 0) 459204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 460204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 461204076Spjd buf = calloc(1, mapsize); 462204076Spjd if (buf == NULL) { 463204076Spjd primary_exitx(EX_TEMPFAIL, 464204076Spjd "Unable to allocate buffer for activemap."); 465204076Spjd } 466204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 467204076Spjd (ssize_t)mapsize) { 468204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 469204076Spjd } 470204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 471209181Spjd free(buf); 472204076Spjd if (res->hr_resuid != 0) 473204076Spjd return; 474204076Spjd /* 475204076Spjd * We're using provider for the first time, so we have to generate 476204076Spjd * resource unique identifier and initialize local and remote counts. 477204076Spjd */ 478204076Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 479204076Spjd res->hr_primary_localcnt = 1; 480204076Spjd res->hr_primary_remotecnt = 0; 481204076Spjd if (metadata_write(res) < 0) 482204076Spjd exit(EX_NOINPUT); 483204076Spjd} 484204076Spjd 485205738Spjdstatic bool 486205738Spjdinit_remote(struct hast_resource *res, struct proto_conn **inp, 487205738Spjd struct proto_conn **outp) 488204076Spjd{ 489205738Spjd struct proto_conn *in, *out; 490204076Spjd struct nv *nvout, *nvin; 491204076Spjd const unsigned char *token; 492204076Spjd unsigned char *map; 493204076Spjd const char *errmsg; 494204076Spjd int32_t extentsize; 495204076Spjd int64_t datasize; 496204076Spjd uint32_t mapsize; 497204076Spjd size_t size; 498204076Spjd 499205738Spjd assert((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 500210881Spjd assert(real_remote(res)); 501205738Spjd 502205738Spjd in = out = NULL; 503205738Spjd 504204076Spjd /* Prepare outgoing connection with remote node. */ 505205738Spjd if (proto_client(res->hr_remoteaddr, &out) < 0) { 506207347Spjd primary_exit(EX_TEMPFAIL, "Unable to create connection to %s", 507204076Spjd res->hr_remoteaddr); 508204076Spjd } 509204076Spjd /* Try to connect, but accept failure. */ 510205738Spjd if (proto_connect(out) < 0) { 511204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 512204076Spjd res->hr_remoteaddr); 513204076Spjd goto close; 514204076Spjd } 515207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 516207371Spjd if (proto_timeout(out, res->hr_timeout) < 0) 517207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 518204076Spjd /* 519204076Spjd * First handshake step. 520204076Spjd * Setup outgoing connection with remote node. 521204076Spjd */ 522204076Spjd nvout = nv_alloc(); 523204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 524204076Spjd if (nv_error(nvout) != 0) { 525204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 526204076Spjd "Unable to allocate header for connection with %s", 527204076Spjd res->hr_remoteaddr); 528204076Spjd nv_free(nvout); 529204076Spjd goto close; 530204076Spjd } 531205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 532204076Spjd pjdlog_errno(LOG_WARNING, 533204076Spjd "Unable to send handshake header to %s", 534204076Spjd res->hr_remoteaddr); 535204076Spjd nv_free(nvout); 536204076Spjd goto close; 537204076Spjd } 538204076Spjd nv_free(nvout); 539205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 540204076Spjd pjdlog_errno(LOG_WARNING, 541204076Spjd "Unable to receive handshake header from %s", 542204076Spjd res->hr_remoteaddr); 543204076Spjd goto close; 544204076Spjd } 545204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 546204076Spjd if (errmsg != NULL) { 547204076Spjd pjdlog_warning("%s", errmsg); 548204076Spjd nv_free(nvin); 549204076Spjd goto close; 550204076Spjd } 551204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 552204076Spjd if (token == NULL) { 553204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 554204076Spjd res->hr_remoteaddr); 555204076Spjd nv_free(nvin); 556204076Spjd goto close; 557204076Spjd } 558204076Spjd if (size != sizeof(res->hr_token)) { 559204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 560204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 561204076Spjd nv_free(nvin); 562204076Spjd goto close; 563204076Spjd } 564204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 565204076Spjd nv_free(nvin); 566204076Spjd 567204076Spjd /* 568204076Spjd * Second handshake step. 569204076Spjd * Setup incoming connection with remote node. 570204076Spjd */ 571205738Spjd if (proto_client(res->hr_remoteaddr, &in) < 0) { 572204076Spjd pjdlog_errno(LOG_WARNING, "Unable to create connection to %s", 573204076Spjd res->hr_remoteaddr); 574204076Spjd } 575204076Spjd /* Try to connect, but accept failure. */ 576205738Spjd if (proto_connect(in) < 0) { 577204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 578204076Spjd res->hr_remoteaddr); 579204076Spjd goto close; 580204076Spjd } 581207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 582207371Spjd if (proto_timeout(in, res->hr_timeout) < 0) 583207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 584204076Spjd nvout = nv_alloc(); 585204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 586204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 587204076Spjd "token"); 588204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 589204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 590204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 591204076Spjd if (nv_error(nvout) != 0) { 592204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 593204076Spjd "Unable to allocate header for connection with %s", 594204076Spjd res->hr_remoteaddr); 595204076Spjd nv_free(nvout); 596204076Spjd goto close; 597204076Spjd } 598205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 599204076Spjd pjdlog_errno(LOG_WARNING, 600204076Spjd "Unable to send handshake header to %s", 601204076Spjd res->hr_remoteaddr); 602204076Spjd nv_free(nvout); 603204076Spjd goto close; 604204076Spjd } 605204076Spjd nv_free(nvout); 606205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 607204076Spjd pjdlog_errno(LOG_WARNING, 608204076Spjd "Unable to receive handshake header from %s", 609204076Spjd res->hr_remoteaddr); 610204076Spjd goto close; 611204076Spjd } 612204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 613204076Spjd if (errmsg != NULL) { 614204076Spjd pjdlog_warning("%s", errmsg); 615204076Spjd nv_free(nvin); 616204076Spjd goto close; 617204076Spjd } 618204076Spjd datasize = nv_get_int64(nvin, "datasize"); 619204076Spjd if (datasize != res->hr_datasize) { 620204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 621204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 622204076Spjd nv_free(nvin); 623204076Spjd goto close; 624204076Spjd } 625204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 626204076Spjd if (extentsize != res->hr_extentsize) { 627204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 628204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 629204076Spjd nv_free(nvin); 630204076Spjd goto close; 631204076Spjd } 632204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 633204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 634204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 635204076Spjd map = NULL; 636204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 637204076Spjd if (mapsize > 0) { 638204076Spjd map = malloc(mapsize); 639204076Spjd if (map == NULL) { 640204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 641204076Spjd (uintmax_t)mapsize); 642204076Spjd nv_free(nvin); 643204076Spjd goto close; 644204076Spjd } 645204076Spjd /* 646204076Spjd * Remote node have some dirty extents on its own, lets 647204076Spjd * download its activemap. 648204076Spjd */ 649205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 650204076Spjd mapsize) < 0) { 651204076Spjd pjdlog_errno(LOG_ERR, 652204076Spjd "Unable to receive remote activemap"); 653204076Spjd nv_free(nvin); 654204076Spjd free(map); 655204076Spjd goto close; 656204076Spjd } 657204076Spjd /* 658204076Spjd * Merge local and remote bitmaps. 659204076Spjd */ 660204076Spjd activemap_merge(res->hr_amp, map, mapsize); 661204076Spjd free(map); 662204076Spjd /* 663204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 664204076Spjd * disk before we start to synchronize. 665204076Spjd */ 666204076Spjd (void)hast_activemap_flush(res); 667204076Spjd } 668204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 669205738Spjd if (inp != NULL && outp != NULL) { 670205738Spjd *inp = in; 671205738Spjd *outp = out; 672205738Spjd } else { 673205738Spjd res->hr_remotein = in; 674205738Spjd res->hr_remoteout = out; 675205738Spjd } 676205738Spjd return (true); 677205738Spjdclose: 678205738Spjd proto_close(out); 679205738Spjd if (in != NULL) 680205738Spjd proto_close(in); 681205738Spjd return (false); 682205738Spjd} 683205738Spjd 684205738Spjdstatic void 685205738Spjdsync_start(void) 686205738Spjd{ 687205738Spjd 688204076Spjd mtx_lock(&sync_lock); 689204076Spjd sync_inprogress = true; 690204076Spjd mtx_unlock(&sync_lock); 691204076Spjd cv_signal(&sync_cond); 692204076Spjd} 693204076Spjd 694204076Spjdstatic void 695211878Spjdsync_stop(void) 696211878Spjd{ 697211878Spjd 698211878Spjd mtx_lock(&sync_lock); 699211878Spjd if (sync_inprogress) 700211878Spjd sync_inprogress = false; 701211878Spjd mtx_unlock(&sync_lock); 702211878Spjd} 703211878Spjd 704211878Spjdstatic void 705204076Spjdinit_ggate(struct hast_resource *res) 706204076Spjd{ 707204076Spjd struct g_gate_ctl_create ggiocreate; 708204076Spjd struct g_gate_ctl_cancel ggiocancel; 709204076Spjd 710204076Spjd /* 711204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 712204076Spjd */ 713204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 714204076Spjd if (res->hr_ggatefd < 0) 715204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 716204076Spjd /* 717204076Spjd * Create provider before trying to connect, as connection failure 718204076Spjd * is not critical, but may take some time. 719204076Spjd */ 720204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 721204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 722204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 723204076Spjd ggiocreate.gctl_flags = 0; 724206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 725204076Spjd ggiocreate.gctl_timeout = 0; 726204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 727204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 728204076Spjd res->hr_provname); 729204076Spjd bzero(ggiocreate.gctl_info, sizeof(ggiocreate.gctl_info)); 730204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 731204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 732204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 733204076Spjd return; 734204076Spjd } 735204076Spjd if (errno != EEXIST) { 736204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 737204076Spjd res->hr_provname); 738204076Spjd } 739204076Spjd pjdlog_debug(1, 740204076Spjd "Device hast/%s already exists, we will try to take it over.", 741204076Spjd res->hr_provname); 742204076Spjd /* 743204076Spjd * If we received EEXIST, we assume that the process who created the 744204076Spjd * provider died and didn't clean up. In that case we will start from 745204076Spjd * where he left of. 746204076Spjd */ 747204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 748204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 749204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 750204076Spjd res->hr_provname); 751204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 752204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 753204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 754204076Spjd return; 755204076Spjd } 756204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 757204076Spjd res->hr_provname); 758204076Spjd} 759204076Spjd 760204076Spjdvoid 761204076Spjdhastd_primary(struct hast_resource *res) 762204076Spjd{ 763204076Spjd pthread_t td; 764204076Spjd pid_t pid; 765204076Spjd int error; 766204076Spjd 767204076Spjd gres = res; 768204076Spjd 769204076Spjd /* 770204076Spjd * Create communication channel between parent and child. 771204076Spjd */ 772204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 773204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 774204076Spjd primary_exit(EX_OSERR, 775204076Spjd "Unable to create control sockets between parent and child"); 776204076Spjd } 777204076Spjd 778204076Spjd pid = fork(); 779204076Spjd if (pid < 0) { 780204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 781207347Spjd primary_exit(EX_TEMPFAIL, "Unable to fork"); 782204076Spjd } 783204076Spjd 784204076Spjd if (pid > 0) { 785204076Spjd /* This is parent. */ 786204076Spjd res->hr_workerpid = pid; 787204076Spjd return; 788204076Spjd } 789204076Spjd (void)pidfile_close(pfh); 790204076Spjd 791204076Spjd setproctitle("%s (primary)", res->hr_name); 792204076Spjd 793210880Spjd signal(SIGHUP, SIG_DFL); 794210880Spjd signal(SIGCHLD, SIG_DFL); 795210880Spjd 796211886Spjd hook_init(); 797204076Spjd init_local(res); 798210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 799205738Spjd sync_start(); 800204076Spjd init_ggate(res); 801204076Spjd init_environment(res); 802204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 803204076Spjd assert(error == 0); 804204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 805204076Spjd assert(error == 0); 806204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 807204076Spjd assert(error == 0); 808204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 809204076Spjd assert(error == 0); 810204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 811204076Spjd assert(error == 0); 812204076Spjd error = pthread_create(&td, NULL, sync_thread, res); 813204076Spjd assert(error == 0); 814204076Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 815204076Spjd assert(error == 0); 816204076Spjd (void)guard_thread(res); 817204076Spjd} 818204076Spjd 819204076Spjdstatic void 820204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 821204076Spjd{ 822204076Spjd char msg[1024]; 823204076Spjd va_list ap; 824204076Spjd int len; 825204076Spjd 826204076Spjd va_start(ap, fmt); 827204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 828204076Spjd va_end(ap); 829204076Spjd if ((size_t)len < sizeof(msg)) { 830204076Spjd switch (ggio->gctl_cmd) { 831204076Spjd case BIO_READ: 832204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 833204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 834204076Spjd (uintmax_t)ggio->gctl_length); 835204076Spjd break; 836204076Spjd case BIO_DELETE: 837204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 838204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 839204076Spjd (uintmax_t)ggio->gctl_length); 840204076Spjd break; 841204076Spjd case BIO_FLUSH: 842204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 843204076Spjd break; 844204076Spjd case BIO_WRITE: 845204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 846204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 847204076Spjd (uintmax_t)ggio->gctl_length); 848204076Spjd break; 849204076Spjd default: 850204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 851204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 852204076Spjd break; 853204076Spjd } 854204076Spjd } 855204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 856204076Spjd} 857204076Spjd 858204076Spjdstatic void 859204076Spjdremote_close(struct hast_resource *res, int ncomp) 860204076Spjd{ 861204076Spjd 862204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 863204076Spjd /* 864204076Spjd * A race is possible between dropping rlock and acquiring wlock - 865204076Spjd * another thread can close connection in-between. 866204076Spjd */ 867204076Spjd if (!ISCONNECTED(res, ncomp)) { 868204076Spjd assert(res->hr_remotein == NULL); 869204076Spjd assert(res->hr_remoteout == NULL); 870204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 871204076Spjd return; 872204076Spjd } 873204076Spjd 874204076Spjd assert(res->hr_remotein != NULL); 875204076Spjd assert(res->hr_remoteout != NULL); 876204076Spjd 877211881Spjd pjdlog_debug(2, "Closing incoming connection to %s.", 878204076Spjd res->hr_remoteaddr); 879204076Spjd proto_close(res->hr_remotein); 880204076Spjd res->hr_remotein = NULL; 881211881Spjd pjdlog_debug(2, "Closing outgoing connection to %s.", 882204076Spjd res->hr_remoteaddr); 883204076Spjd proto_close(res->hr_remoteout); 884204076Spjd res->hr_remoteout = NULL; 885204076Spjd 886204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 887204076Spjd 888211881Spjd pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 889211881Spjd 890204076Spjd /* 891204076Spjd * Stop synchronization if in-progress. 892204076Spjd */ 893211878Spjd sync_stop(); 894204076Spjd 895204076Spjd /* 896211882Spjd * Wake up guard thread (if we are not called from within guard thread), 897211882Spjd * so it can immediately start reconnect. 898204076Spjd */ 899211882Spjd if (!mtx_owned(&hio_guard_lock)) { 900211882Spjd mtx_lock(&hio_guard_lock); 901211882Spjd cv_signal(&hio_guard_cond); 902211882Spjd mtx_unlock(&hio_guard_lock); 903211882Spjd } 904204076Spjd} 905204076Spjd 906204076Spjd/* 907204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 908204076Spjd * appropriate threads: 909204076Spjd * WRITE - always goes to both local_send and remote_send threads 910204076Spjd * READ (when the block is up-to-date on local component) - 911204076Spjd * only local_send thread 912204076Spjd * READ (when the block isn't up-to-date on local component) - 913204076Spjd * only remote_send thread 914204076Spjd * DELETE - always goes to both local_send and remote_send threads 915204076Spjd * FLUSH - always goes to both local_send and remote_send threads 916204076Spjd */ 917204076Spjdstatic void * 918204076Spjdggate_recv_thread(void *arg) 919204076Spjd{ 920204076Spjd struct hast_resource *res = arg; 921204076Spjd struct g_gate_ctl_io *ggio; 922204076Spjd struct hio *hio; 923204076Spjd unsigned int ii, ncomp, ncomps; 924204076Spjd int error; 925204076Spjd 926204076Spjd ncomps = HAST_NCOMPONENTS; 927204076Spjd 928204076Spjd for (;;) { 929204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 930204076Spjd QUEUE_TAKE2(hio, free); 931204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 932204076Spjd ggio = &hio->hio_ggio; 933204076Spjd ggio->gctl_unit = res->hr_ggateunit; 934204076Spjd ggio->gctl_length = MAXPHYS; 935204076Spjd ggio->gctl_error = 0; 936204076Spjd pjdlog_debug(2, 937204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 938204076Spjd hio); 939204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 940204076Spjd if (sigexit_received) 941204076Spjd pthread_exit(NULL); 942204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 943204076Spjd } 944204076Spjd error = ggio->gctl_error; 945204076Spjd switch (error) { 946204076Spjd case 0: 947204076Spjd break; 948204076Spjd case ECANCELED: 949204076Spjd /* Exit gracefully. */ 950204076Spjd if (!sigexit_received) { 951204076Spjd pjdlog_debug(2, 952204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 953204076Spjd hio); 954204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 955204076Spjd } 956204076Spjd pthread_exit(NULL); 957204076Spjd case ENOMEM: 958204076Spjd /* 959204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 960204076Spjd * bytes - request can't be bigger than that. 961204076Spjd */ 962204076Spjd /* FALLTHROUGH */ 963204076Spjd case ENXIO: 964204076Spjd default: 965204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 966204076Spjd strerror(error)); 967204076Spjd } 968204076Spjd for (ii = 0; ii < ncomps; ii++) 969204076Spjd hio->hio_errors[ii] = EINVAL; 970204076Spjd reqlog(LOG_DEBUG, 2, ggio, 971204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 972204076Spjd hio); 973204076Spjd /* 974204076Spjd * Inform all components about new write request. 975204076Spjd * For read request prefer local component unless the given 976204076Spjd * range is out-of-date, then use remote component. 977204076Spjd */ 978204076Spjd switch (ggio->gctl_cmd) { 979204076Spjd case BIO_READ: 980204076Spjd pjdlog_debug(2, 981204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 982204076Spjd hio); 983204076Spjd refcount_init(&hio->hio_countdown, 1); 984204076Spjd mtx_lock(&metadata_lock); 985204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 986204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 987204076Spjd /* 988204076Spjd * This range is up-to-date on local component, 989204076Spjd * so handle request locally. 990204076Spjd */ 991204076Spjd /* Local component is 0 for now. */ 992204076Spjd ncomp = 0; 993204076Spjd } else /* if (res->hr_syncsrc == 994204076Spjd HAST_SYNCSRC_SECONDARY) */ { 995204076Spjd assert(res->hr_syncsrc == 996204076Spjd HAST_SYNCSRC_SECONDARY); 997204076Spjd /* 998204076Spjd * This range is out-of-date on local component, 999204076Spjd * so send request to the remote node. 1000204076Spjd */ 1001204076Spjd /* Remote component is 1 for now. */ 1002204076Spjd ncomp = 1; 1003204076Spjd } 1004204076Spjd mtx_unlock(&metadata_lock); 1005204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1006204076Spjd break; 1007204076Spjd case BIO_WRITE: 1008204076Spjd for (;;) { 1009204076Spjd mtx_lock(&range_lock); 1010204076Spjd if (rangelock_islocked(range_sync, 1011204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1012204076Spjd pjdlog_debug(2, 1013204076Spjd "regular: Range offset=%jd length=%zu locked.", 1014204076Spjd (intmax_t)ggio->gctl_offset, 1015204076Spjd (size_t)ggio->gctl_length); 1016204076Spjd range_regular_wait = true; 1017204076Spjd cv_wait(&range_regular_cond, &range_lock); 1018204076Spjd range_regular_wait = false; 1019204076Spjd mtx_unlock(&range_lock); 1020204076Spjd continue; 1021204076Spjd } 1022204076Spjd if (rangelock_add(range_regular, 1023204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1024204076Spjd mtx_unlock(&range_lock); 1025204076Spjd pjdlog_debug(2, 1026204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1027204076Spjd (intmax_t)ggio->gctl_offset, 1028204076Spjd (size_t)ggio->gctl_length); 1029204076Spjd sleep(1); 1030204076Spjd continue; 1031204076Spjd } 1032204076Spjd mtx_unlock(&range_lock); 1033204076Spjd break; 1034204076Spjd } 1035204076Spjd mtx_lock(&res->hr_amp_lock); 1036204076Spjd if (activemap_write_start(res->hr_amp, 1037204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1038204076Spjd (void)hast_activemap_flush(res); 1039204076Spjd } 1040204076Spjd mtx_unlock(&res->hr_amp_lock); 1041204076Spjd /* FALLTHROUGH */ 1042204076Spjd case BIO_DELETE: 1043204076Spjd case BIO_FLUSH: 1044204076Spjd pjdlog_debug(2, 1045204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1046204076Spjd hio); 1047204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1048204076Spjd for (ii = 0; ii < ncomps; ii++) 1049204076Spjd QUEUE_INSERT1(hio, send, ii); 1050204076Spjd break; 1051204076Spjd } 1052204076Spjd } 1053204076Spjd /* NOTREACHED */ 1054204076Spjd return (NULL); 1055204076Spjd} 1056204076Spjd 1057204076Spjd/* 1058204076Spjd * Thread reads from or writes to local component. 1059204076Spjd * If local read fails, it redirects it to remote_send thread. 1060204076Spjd */ 1061204076Spjdstatic void * 1062204076Spjdlocal_send_thread(void *arg) 1063204076Spjd{ 1064204076Spjd struct hast_resource *res = arg; 1065204076Spjd struct g_gate_ctl_io *ggio; 1066204076Spjd struct hio *hio; 1067204076Spjd unsigned int ncomp, rncomp; 1068204076Spjd ssize_t ret; 1069204076Spjd 1070204076Spjd /* Local component is 0 for now. */ 1071204076Spjd ncomp = 0; 1072204076Spjd /* Remote component is 1 for now. */ 1073204076Spjd rncomp = 1; 1074204076Spjd 1075204076Spjd for (;;) { 1076204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1077204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1078204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1079204076Spjd ggio = &hio->hio_ggio; 1080204076Spjd switch (ggio->gctl_cmd) { 1081204076Spjd case BIO_READ: 1082204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1083204076Spjd ggio->gctl_length, 1084204076Spjd ggio->gctl_offset + res->hr_localoff); 1085204076Spjd if (ret == ggio->gctl_length) 1086204076Spjd hio->hio_errors[ncomp] = 0; 1087204076Spjd else { 1088204076Spjd /* 1089204076Spjd * If READ failed, try to read from remote node. 1090204076Spjd */ 1091204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1092204076Spjd continue; 1093204076Spjd } 1094204076Spjd break; 1095204076Spjd case BIO_WRITE: 1096204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1097204076Spjd ggio->gctl_length, 1098204076Spjd ggio->gctl_offset + res->hr_localoff); 1099204076Spjd if (ret < 0) 1100204076Spjd hio->hio_errors[ncomp] = errno; 1101204076Spjd else if (ret != ggio->gctl_length) 1102204076Spjd hio->hio_errors[ncomp] = EIO; 1103204076Spjd else 1104204076Spjd hio->hio_errors[ncomp] = 0; 1105204076Spjd break; 1106204076Spjd case BIO_DELETE: 1107204076Spjd ret = g_delete(res->hr_localfd, 1108204076Spjd ggio->gctl_offset + res->hr_localoff, 1109204076Spjd ggio->gctl_length); 1110204076Spjd if (ret < 0) 1111204076Spjd hio->hio_errors[ncomp] = errno; 1112204076Spjd else 1113204076Spjd hio->hio_errors[ncomp] = 0; 1114204076Spjd break; 1115204076Spjd case BIO_FLUSH: 1116204076Spjd ret = g_flush(res->hr_localfd); 1117204076Spjd if (ret < 0) 1118204076Spjd hio->hio_errors[ncomp] = errno; 1119204076Spjd else 1120204076Spjd hio->hio_errors[ncomp] = 0; 1121204076Spjd break; 1122204076Spjd } 1123204076Spjd if (refcount_release(&hio->hio_countdown)) { 1124204076Spjd if (ISSYNCREQ(hio)) { 1125204076Spjd mtx_lock(&sync_lock); 1126204076Spjd SYNCREQDONE(hio); 1127204076Spjd mtx_unlock(&sync_lock); 1128204076Spjd cv_signal(&sync_cond); 1129204076Spjd } else { 1130204076Spjd pjdlog_debug(2, 1131204076Spjd "local_send: (%p) Moving request to the done queue.", 1132204076Spjd hio); 1133204076Spjd QUEUE_INSERT2(hio, done); 1134204076Spjd } 1135204076Spjd } 1136204076Spjd } 1137204076Spjd /* NOTREACHED */ 1138204076Spjd return (NULL); 1139204076Spjd} 1140204076Spjd 1141204076Spjd/* 1142204076Spjd * Thread sends request to secondary node. 1143204076Spjd */ 1144204076Spjdstatic void * 1145204076Spjdremote_send_thread(void *arg) 1146204076Spjd{ 1147204076Spjd struct hast_resource *res = arg; 1148204076Spjd struct g_gate_ctl_io *ggio; 1149204076Spjd struct hio *hio; 1150204076Spjd struct nv *nv; 1151204076Spjd unsigned int ncomp; 1152204076Spjd bool wakeup; 1153204076Spjd uint64_t offset, length; 1154204076Spjd uint8_t cmd; 1155204076Spjd void *data; 1156204076Spjd 1157204076Spjd /* Remote component is 1 for now. */ 1158204076Spjd ncomp = 1; 1159204076Spjd 1160204076Spjd for (;;) { 1161204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1162204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1163204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1164204076Spjd ggio = &hio->hio_ggio; 1165204076Spjd switch (ggio->gctl_cmd) { 1166204076Spjd case BIO_READ: 1167204076Spjd cmd = HIO_READ; 1168204076Spjd data = NULL; 1169204076Spjd offset = ggio->gctl_offset; 1170204076Spjd length = ggio->gctl_length; 1171204076Spjd break; 1172204076Spjd case BIO_WRITE: 1173204076Spjd cmd = HIO_WRITE; 1174204076Spjd data = ggio->gctl_data; 1175204076Spjd offset = ggio->gctl_offset; 1176204076Spjd length = ggio->gctl_length; 1177204076Spjd break; 1178204076Spjd case BIO_DELETE: 1179204076Spjd cmd = HIO_DELETE; 1180204076Spjd data = NULL; 1181204076Spjd offset = ggio->gctl_offset; 1182204076Spjd length = ggio->gctl_length; 1183204076Spjd break; 1184204076Spjd case BIO_FLUSH: 1185204076Spjd cmd = HIO_FLUSH; 1186204076Spjd data = NULL; 1187204076Spjd offset = 0; 1188204076Spjd length = 0; 1189204076Spjd break; 1190204076Spjd default: 1191204076Spjd assert(!"invalid condition"); 1192204076Spjd abort(); 1193204076Spjd } 1194204076Spjd nv = nv_alloc(); 1195204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1196204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1197204076Spjd nv_add_uint64(nv, offset, "offset"); 1198204076Spjd nv_add_uint64(nv, length, "length"); 1199204076Spjd if (nv_error(nv) != 0) { 1200204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1201204076Spjd pjdlog_debug(2, 1202204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1203204076Spjd hio); 1204204076Spjd reqlog(LOG_ERR, 0, ggio, 1205204076Spjd "Unable to prepare header to send (%s): ", 1206204076Spjd strerror(nv_error(nv))); 1207204076Spjd /* Move failed request immediately to the done queue. */ 1208204076Spjd goto done_queue; 1209204076Spjd } 1210204076Spjd pjdlog_debug(2, 1211204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1212204076Spjd hio); 1213204076Spjd /* 1214204076Spjd * Protect connection from disappearing. 1215204076Spjd */ 1216204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1217204076Spjd if (!ISCONNECTED(res, ncomp)) { 1218204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1219204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1220204076Spjd goto done_queue; 1221204076Spjd } 1222204076Spjd /* 1223204076Spjd * Move the request to recv queue before sending it, because 1224204076Spjd * in different order we can get reply before we move request 1225204076Spjd * to recv queue. 1226204076Spjd */ 1227204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1228204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1229204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1230204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1231204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1232204076Spjd data != NULL ? length : 0) < 0) { 1233204076Spjd hio->hio_errors[ncomp] = errno; 1234204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1235204076Spjd remote_close(res, ncomp); 1236204076Spjd pjdlog_debug(2, 1237204076Spjd "remote_send: (%p) Unable to send request.", hio); 1238204076Spjd reqlog(LOG_ERR, 0, ggio, 1239204076Spjd "Unable to send request (%s): ", 1240204076Spjd strerror(hio->hio_errors[ncomp])); 1241204076Spjd /* 1242204076Spjd * Take request back from the receive queue and move 1243204076Spjd * it immediately to the done queue. 1244204076Spjd */ 1245204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1246204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1247204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1248204076Spjd goto done_queue; 1249204076Spjd } 1250204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1251204076Spjd nv_free(nv); 1252204076Spjd if (wakeup) 1253204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1254204076Spjd continue; 1255204076Spjddone_queue: 1256204076Spjd nv_free(nv); 1257204076Spjd if (ISSYNCREQ(hio)) { 1258204076Spjd if (!refcount_release(&hio->hio_countdown)) 1259204076Spjd continue; 1260204076Spjd mtx_lock(&sync_lock); 1261204076Spjd SYNCREQDONE(hio); 1262204076Spjd mtx_unlock(&sync_lock); 1263204076Spjd cv_signal(&sync_cond); 1264204076Spjd continue; 1265204076Spjd } 1266204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1267204076Spjd mtx_lock(&res->hr_amp_lock); 1268204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1269204076Spjd ggio->gctl_length)) { 1270204076Spjd (void)hast_activemap_flush(res); 1271204076Spjd } 1272204076Spjd mtx_unlock(&res->hr_amp_lock); 1273204076Spjd } 1274204076Spjd if (!refcount_release(&hio->hio_countdown)) 1275204076Spjd continue; 1276204076Spjd pjdlog_debug(2, 1277204076Spjd "remote_send: (%p) Moving request to the done queue.", 1278204076Spjd hio); 1279204076Spjd QUEUE_INSERT2(hio, done); 1280204076Spjd } 1281204076Spjd /* NOTREACHED */ 1282204076Spjd return (NULL); 1283204076Spjd} 1284204076Spjd 1285204076Spjd/* 1286204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1287204076Spjd * thread. 1288204076Spjd */ 1289204076Spjdstatic void * 1290204076Spjdremote_recv_thread(void *arg) 1291204076Spjd{ 1292204076Spjd struct hast_resource *res = arg; 1293204076Spjd struct g_gate_ctl_io *ggio; 1294204076Spjd struct hio *hio; 1295204076Spjd struct nv *nv; 1296204076Spjd unsigned int ncomp; 1297204076Spjd uint64_t seq; 1298204076Spjd int error; 1299204076Spjd 1300204076Spjd /* Remote component is 1 for now. */ 1301204076Spjd ncomp = 1; 1302204076Spjd 1303204076Spjd for (;;) { 1304204076Spjd /* Wait until there is anything to receive. */ 1305204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1306204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1307204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1308204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1309204076Spjd &hio_recv_list_lock[ncomp]); 1310204076Spjd } 1311204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1312204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1313204076Spjd if (!ISCONNECTED(res, ncomp)) { 1314204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1315204076Spjd /* 1316204076Spjd * Connection is dead, so move all pending requests to 1317204076Spjd * the done queue (one-by-one). 1318204076Spjd */ 1319204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1320204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1321204076Spjd assert(hio != NULL); 1322204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1323204076Spjd hio_next[ncomp]); 1324204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1325204076Spjd goto done_queue; 1326204076Spjd } 1327204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1328204076Spjd pjdlog_errno(LOG_ERR, 1329204076Spjd "Unable to receive reply header"); 1330204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1331204076Spjd remote_close(res, ncomp); 1332204076Spjd continue; 1333204076Spjd } 1334204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1335204076Spjd seq = nv_get_uint64(nv, "seq"); 1336204076Spjd if (seq == 0) { 1337204076Spjd pjdlog_error("Header contains no 'seq' field."); 1338204076Spjd nv_free(nv); 1339204076Spjd continue; 1340204076Spjd } 1341204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1342204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1343204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1344204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1345204076Spjd hio_next[ncomp]); 1346204076Spjd break; 1347204076Spjd } 1348204076Spjd } 1349204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1350204076Spjd if (hio == NULL) { 1351204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1352204076Spjd (uintmax_t)seq); 1353204076Spjd nv_free(nv); 1354204076Spjd continue; 1355204076Spjd } 1356204076Spjd error = nv_get_int16(nv, "error"); 1357204076Spjd if (error != 0) { 1358204076Spjd /* Request failed on remote side. */ 1359204076Spjd hio->hio_errors[ncomp] = 0; 1360204076Spjd nv_free(nv); 1361204076Spjd goto done_queue; 1362204076Spjd } 1363204076Spjd ggio = &hio->hio_ggio; 1364204076Spjd switch (ggio->gctl_cmd) { 1365204076Spjd case BIO_READ: 1366204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1367204076Spjd if (!ISCONNECTED(res, ncomp)) { 1368204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1369204076Spjd nv_free(nv); 1370204076Spjd goto done_queue; 1371204076Spjd } 1372204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1373204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1374204076Spjd hio->hio_errors[ncomp] = errno; 1375204076Spjd pjdlog_errno(LOG_ERR, 1376204076Spjd "Unable to receive reply data"); 1377204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1378204076Spjd nv_free(nv); 1379204076Spjd remote_close(res, ncomp); 1380204076Spjd goto done_queue; 1381204076Spjd } 1382204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1383204076Spjd break; 1384204076Spjd case BIO_WRITE: 1385204076Spjd case BIO_DELETE: 1386204076Spjd case BIO_FLUSH: 1387204076Spjd break; 1388204076Spjd default: 1389204076Spjd assert(!"invalid condition"); 1390204076Spjd abort(); 1391204076Spjd } 1392204076Spjd hio->hio_errors[ncomp] = 0; 1393204076Spjd nv_free(nv); 1394204076Spjddone_queue: 1395204076Spjd if (refcount_release(&hio->hio_countdown)) { 1396204076Spjd if (ISSYNCREQ(hio)) { 1397204076Spjd mtx_lock(&sync_lock); 1398204076Spjd SYNCREQDONE(hio); 1399204076Spjd mtx_unlock(&sync_lock); 1400204076Spjd cv_signal(&sync_cond); 1401204076Spjd } else { 1402204076Spjd pjdlog_debug(2, 1403204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1404204076Spjd hio); 1405204076Spjd QUEUE_INSERT2(hio, done); 1406204076Spjd } 1407204076Spjd } 1408204076Spjd } 1409204076Spjd /* NOTREACHED */ 1410204076Spjd return (NULL); 1411204076Spjd} 1412204076Spjd 1413204076Spjd/* 1414204076Spjd * Thread sends answer to the kernel. 1415204076Spjd */ 1416204076Spjdstatic void * 1417204076Spjdggate_send_thread(void *arg) 1418204076Spjd{ 1419204076Spjd struct hast_resource *res = arg; 1420204076Spjd struct g_gate_ctl_io *ggio; 1421204076Spjd struct hio *hio; 1422204076Spjd unsigned int ii, ncomp, ncomps; 1423204076Spjd 1424204076Spjd ncomps = HAST_NCOMPONENTS; 1425204076Spjd 1426204076Spjd for (;;) { 1427204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1428204076Spjd QUEUE_TAKE2(hio, done); 1429204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1430204076Spjd ggio = &hio->hio_ggio; 1431204076Spjd for (ii = 0; ii < ncomps; ii++) { 1432204076Spjd if (hio->hio_errors[ii] == 0) { 1433204076Spjd /* 1434204076Spjd * One successful request is enough to declare 1435204076Spjd * success. 1436204076Spjd */ 1437204076Spjd ggio->gctl_error = 0; 1438204076Spjd break; 1439204076Spjd } 1440204076Spjd } 1441204076Spjd if (ii == ncomps) { 1442204076Spjd /* 1443204076Spjd * None of the requests were successful. 1444204076Spjd * Use first error. 1445204076Spjd */ 1446204076Spjd ggio->gctl_error = hio->hio_errors[0]; 1447204076Spjd } 1448204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1449204076Spjd mtx_lock(&res->hr_amp_lock); 1450204076Spjd activemap_write_complete(res->hr_amp, 1451204076Spjd ggio->gctl_offset, ggio->gctl_length); 1452204076Spjd mtx_unlock(&res->hr_amp_lock); 1453204076Spjd } 1454204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1455204076Spjd /* 1456204076Spjd * Unlock range we locked. 1457204076Spjd */ 1458204076Spjd mtx_lock(&range_lock); 1459204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1460204076Spjd ggio->gctl_length); 1461204076Spjd if (range_sync_wait) 1462204076Spjd cv_signal(&range_sync_cond); 1463204076Spjd mtx_unlock(&range_lock); 1464204076Spjd /* 1465204076Spjd * Bump local count if this is first write after 1466204076Spjd * connection failure with remote node. 1467204076Spjd */ 1468204076Spjd ncomp = 1; 1469204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1470204076Spjd if (!ISCONNECTED(res, ncomp)) { 1471204076Spjd mtx_lock(&metadata_lock); 1472204076Spjd if (res->hr_primary_localcnt == 1473204076Spjd res->hr_secondary_remotecnt) { 1474204076Spjd res->hr_primary_localcnt++; 1475204076Spjd pjdlog_debug(1, 1476204076Spjd "Increasing localcnt to %ju.", 1477204076Spjd (uintmax_t)res->hr_primary_localcnt); 1478204076Spjd (void)metadata_write(res); 1479204076Spjd } 1480204076Spjd mtx_unlock(&metadata_lock); 1481204076Spjd } 1482204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1483204076Spjd } 1484204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1485204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1486204076Spjd pjdlog_debug(2, 1487204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1488204076Spjd QUEUE_INSERT2(hio, free); 1489204076Spjd } 1490204076Spjd /* NOTREACHED */ 1491204076Spjd return (NULL); 1492204076Spjd} 1493204076Spjd 1494204076Spjd/* 1495204076Spjd * Thread synchronize local and remote components. 1496204076Spjd */ 1497204076Spjdstatic void * 1498204076Spjdsync_thread(void *arg __unused) 1499204076Spjd{ 1500204076Spjd struct hast_resource *res = arg; 1501204076Spjd struct hio *hio; 1502204076Spjd struct g_gate_ctl_io *ggio; 1503204076Spjd unsigned int ii, ncomp, ncomps; 1504204076Spjd off_t offset, length, synced; 1505204076Spjd bool dorewind; 1506204076Spjd int syncext; 1507204076Spjd 1508204076Spjd ncomps = HAST_NCOMPONENTS; 1509204076Spjd dorewind = true; 1510211897Spjd synced = 0; 1511211897Spjd offset = -1; 1512204076Spjd 1513204076Spjd for (;;) { 1514204076Spjd mtx_lock(&sync_lock); 1515211897Spjd if (offset >= 0 && !sync_inprogress) { 1516211879Spjd pjdlog_info("Synchronization interrupted. " 1517211879Spjd "%jd bytes synchronized so far.", 1518211879Spjd (intmax_t)synced); 1519211897Spjd hook_exec(res->hr_exec, "syncintr", res->hr_name, NULL); 1520211879Spjd } 1521204076Spjd while (!sync_inprogress) { 1522204076Spjd dorewind = true; 1523204076Spjd synced = 0; 1524204076Spjd cv_wait(&sync_cond, &sync_lock); 1525204076Spjd } 1526204076Spjd mtx_unlock(&sync_lock); 1527204076Spjd /* 1528204076Spjd * Obtain offset at which we should synchronize. 1529204076Spjd * Rewind synchronization if needed. 1530204076Spjd */ 1531204076Spjd mtx_lock(&res->hr_amp_lock); 1532204076Spjd if (dorewind) 1533204076Spjd activemap_sync_rewind(res->hr_amp); 1534204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1535204076Spjd if (syncext != -1) { 1536204076Spjd /* 1537204076Spjd * We synchronized entire syncext extent, we can mark 1538204076Spjd * it as clean now. 1539204076Spjd */ 1540204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1541204076Spjd (void)hast_activemap_flush(res); 1542204076Spjd } 1543204076Spjd mtx_unlock(&res->hr_amp_lock); 1544204076Spjd if (dorewind) { 1545204076Spjd dorewind = false; 1546204076Spjd if (offset < 0) 1547204076Spjd pjdlog_info("Nodes are in sync."); 1548204076Spjd else { 1549204076Spjd pjdlog_info("Synchronization started. %ju bytes to go.", 1550204076Spjd (uintmax_t)(res->hr_extentsize * 1551204076Spjd activemap_ndirty(res->hr_amp))); 1552211895Spjd hook_exec(res->hr_exec, "syncstart", 1553211895Spjd res->hr_name, NULL); 1554204076Spjd } 1555204076Spjd } 1556204076Spjd if (offset < 0) { 1557211878Spjd sync_stop(); 1558204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1559204076Spjd /* 1560204076Spjd * Synchronization complete, make both localcnt and 1561204076Spjd * remotecnt equal. 1562204076Spjd */ 1563204076Spjd ncomp = 1; 1564204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1565204076Spjd if (ISCONNECTED(res, ncomp)) { 1566204076Spjd if (synced > 0) { 1567204076Spjd pjdlog_info("Synchronization complete. " 1568204076Spjd "%jd bytes synchronized.", 1569204076Spjd (intmax_t)synced); 1570211895Spjd hook_exec(res->hr_exec, "syncdone", 1571211895Spjd res->hr_name, NULL); 1572204076Spjd } 1573204076Spjd mtx_lock(&metadata_lock); 1574204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1575204076Spjd res->hr_primary_localcnt = 1576204076Spjd res->hr_secondary_localcnt; 1577204076Spjd res->hr_primary_remotecnt = 1578204076Spjd res->hr_secondary_remotecnt; 1579204076Spjd pjdlog_debug(1, 1580204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1581204076Spjd (uintmax_t)res->hr_primary_localcnt, 1582204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1583204076Spjd (void)metadata_write(res); 1584204076Spjd mtx_unlock(&metadata_lock); 1585204076Spjd } 1586204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1587204076Spjd continue; 1588204076Spjd } 1589204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1590204076Spjd QUEUE_TAKE2(hio, free); 1591204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1592204076Spjd /* 1593204076Spjd * Lock the range we are going to synchronize. We don't want 1594204076Spjd * race where someone writes between our read and write. 1595204076Spjd */ 1596204076Spjd for (;;) { 1597204076Spjd mtx_lock(&range_lock); 1598204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1599204076Spjd pjdlog_debug(2, 1600204076Spjd "sync: Range offset=%jd length=%jd locked.", 1601204076Spjd (intmax_t)offset, (intmax_t)length); 1602204076Spjd range_sync_wait = true; 1603204076Spjd cv_wait(&range_sync_cond, &range_lock); 1604204076Spjd range_sync_wait = false; 1605204076Spjd mtx_unlock(&range_lock); 1606204076Spjd continue; 1607204076Spjd } 1608204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1609204076Spjd mtx_unlock(&range_lock); 1610204076Spjd pjdlog_debug(2, 1611204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1612204076Spjd (intmax_t)offset, (intmax_t)length); 1613204076Spjd sleep(1); 1614204076Spjd continue; 1615204076Spjd } 1616204076Spjd mtx_unlock(&range_lock); 1617204076Spjd break; 1618204076Spjd } 1619204076Spjd /* 1620204076Spjd * First read the data from synchronization source. 1621204076Spjd */ 1622204076Spjd SYNCREQ(hio); 1623204076Spjd ggio = &hio->hio_ggio; 1624204076Spjd ggio->gctl_cmd = BIO_READ; 1625204076Spjd ggio->gctl_offset = offset; 1626204076Spjd ggio->gctl_length = length; 1627204076Spjd ggio->gctl_error = 0; 1628204076Spjd for (ii = 0; ii < ncomps; ii++) 1629204076Spjd hio->hio_errors[ii] = EINVAL; 1630204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1631204076Spjd hio); 1632204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1633204076Spjd hio); 1634204076Spjd mtx_lock(&metadata_lock); 1635204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1636204076Spjd /* 1637204076Spjd * This range is up-to-date on local component, 1638204076Spjd * so handle request locally. 1639204076Spjd */ 1640204076Spjd /* Local component is 0 for now. */ 1641204076Spjd ncomp = 0; 1642204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1643204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1644204076Spjd /* 1645204076Spjd * This range is out-of-date on local component, 1646204076Spjd * so send request to the remote node. 1647204076Spjd */ 1648204076Spjd /* Remote component is 1 for now. */ 1649204076Spjd ncomp = 1; 1650204076Spjd } 1651204076Spjd mtx_unlock(&metadata_lock); 1652204076Spjd refcount_init(&hio->hio_countdown, 1); 1653204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1654204076Spjd 1655204076Spjd /* 1656204076Spjd * Let's wait for READ to finish. 1657204076Spjd */ 1658204076Spjd mtx_lock(&sync_lock); 1659204076Spjd while (!ISSYNCREQDONE(hio)) 1660204076Spjd cv_wait(&sync_cond, &sync_lock); 1661204076Spjd mtx_unlock(&sync_lock); 1662204076Spjd 1663204076Spjd if (hio->hio_errors[ncomp] != 0) { 1664204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1665204076Spjd strerror(hio->hio_errors[ncomp])); 1666204076Spjd goto free_queue; 1667204076Spjd } 1668204076Spjd 1669204076Spjd /* 1670204076Spjd * We read the data from synchronization source, now write it 1671204076Spjd * to synchronization target. 1672204076Spjd */ 1673204076Spjd SYNCREQ(hio); 1674204076Spjd ggio->gctl_cmd = BIO_WRITE; 1675204076Spjd for (ii = 0; ii < ncomps; ii++) 1676204076Spjd hio->hio_errors[ii] = EINVAL; 1677204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1678204076Spjd hio); 1679204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1680204076Spjd hio); 1681204076Spjd mtx_lock(&metadata_lock); 1682204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1683204076Spjd /* 1684204076Spjd * This range is up-to-date on local component, 1685204076Spjd * so we update remote component. 1686204076Spjd */ 1687204076Spjd /* Remote component is 1 for now. */ 1688204076Spjd ncomp = 1; 1689204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1690204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1691204076Spjd /* 1692204076Spjd * This range is out-of-date on local component, 1693204076Spjd * so we update it. 1694204076Spjd */ 1695204076Spjd /* Local component is 0 for now. */ 1696204076Spjd ncomp = 0; 1697204076Spjd } 1698204076Spjd mtx_unlock(&metadata_lock); 1699204076Spjd 1700204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1701204076Spjd hio); 1702204076Spjd refcount_init(&hio->hio_countdown, 1); 1703204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1704204076Spjd 1705204076Spjd /* 1706204076Spjd * Let's wait for WRITE to finish. 1707204076Spjd */ 1708204076Spjd mtx_lock(&sync_lock); 1709204076Spjd while (!ISSYNCREQDONE(hio)) 1710204076Spjd cv_wait(&sync_cond, &sync_lock); 1711204076Spjd mtx_unlock(&sync_lock); 1712204076Spjd 1713204076Spjd if (hio->hio_errors[ncomp] != 0) { 1714204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1715204076Spjd strerror(hio->hio_errors[ncomp])); 1716204076Spjd goto free_queue; 1717204076Spjd } 1718211880Spjd 1719211880Spjd synced += length; 1720204076Spjdfree_queue: 1721204076Spjd mtx_lock(&range_lock); 1722204076Spjd rangelock_del(range_sync, offset, length); 1723204076Spjd if (range_regular_wait) 1724204076Spjd cv_signal(&range_regular_cond); 1725204076Spjd mtx_unlock(&range_lock); 1726204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1727204076Spjd hio); 1728204076Spjd QUEUE_INSERT2(hio, free); 1729204076Spjd } 1730204076Spjd /* NOTREACHED */ 1731204076Spjd return (NULL); 1732204076Spjd} 1733204076Spjd 1734204076Spjdstatic void 1735204076Spjdsighandler(int sig) 1736204076Spjd{ 1737204076Spjd bool unlock; 1738204076Spjd 1739204076Spjd switch (sig) { 1740204076Spjd case SIGINT: 1741204076Spjd case SIGTERM: 1742204076Spjd sigexit_received = true; 1743204076Spjd break; 1744210886Spjd case SIGHUP: 1745210886Spjd sighup_received = true; 1746210886Spjd break; 1747211886Spjd case SIGCHLD: 1748211886Spjd sigchld_received = true; 1749211886Spjd break; 1750204076Spjd default: 1751204076Spjd assert(!"invalid condition"); 1752204076Spjd } 1753204076Spjd /* 1754211882Spjd * Racy, but if we cannot obtain hio_guard_lock here, we don't 1755204076Spjd * want to risk deadlock. 1756204076Spjd */ 1757204076Spjd unlock = mtx_trylock(&hio_guard_lock); 1758204076Spjd cv_signal(&hio_guard_cond); 1759204076Spjd if (unlock) 1760204076Spjd mtx_unlock(&hio_guard_lock); 1761204076Spjd} 1762204076Spjd 1763210886Spjdstatic void 1764210886Spjdconfig_reload(void) 1765210886Spjd{ 1766210886Spjd struct hastd_config *newcfg; 1767210886Spjd struct hast_resource *res; 1768210886Spjd unsigned int ii, ncomps; 1769210886Spjd int modified; 1770210886Spjd 1771210886Spjd pjdlog_info("Reloading configuration..."); 1772210886Spjd 1773210886Spjd ncomps = HAST_NCOMPONENTS; 1774210886Spjd 1775210886Spjd newcfg = yy_config_parse(cfgpath, false); 1776210886Spjd if (newcfg == NULL) 1777210886Spjd goto failed; 1778210886Spjd 1779210886Spjd TAILQ_FOREACH(res, &newcfg->hc_resources, hr_next) { 1780210886Spjd if (strcmp(res->hr_name, gres->hr_name) == 0) 1781210886Spjd break; 1782210886Spjd } 1783210886Spjd /* 1784210886Spjd * If resource was removed from the configuration file, resource 1785210886Spjd * name, provider name or path to local component was modified we 1786210886Spjd * shouldn't be here. This means that someone modified configuration 1787210886Spjd * file and send SIGHUP to us instead of main hastd process. 1788210886Spjd * Log advice and ignore the signal. 1789210886Spjd */ 1790210886Spjd if (res == NULL || strcmp(gres->hr_name, res->hr_name) != 0 || 1791210886Spjd strcmp(gres->hr_provname, res->hr_provname) != 0 || 1792210886Spjd strcmp(gres->hr_localpath, res->hr_localpath) != 0) { 1793210886Spjd pjdlog_warning("To reload configuration send SIGHUP to the main hastd process (pid %u).", 1794210886Spjd (unsigned int)getppid()); 1795210886Spjd goto failed; 1796210886Spjd } 1797210886Spjd 1798210886Spjd#define MODIFIED_REMOTEADDR 0x1 1799210886Spjd#define MODIFIED_REPLICATION 0x2 1800210886Spjd#define MODIFIED_TIMEOUT 0x4 1801211886Spjd#define MODIFIED_EXEC 0x8 1802210886Spjd modified = 0; 1803210886Spjd if (strcmp(gres->hr_remoteaddr, res->hr_remoteaddr) != 0) { 1804210886Spjd /* 1805210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1806210886Spjd * We want remote_close() to log disconnect from the old 1807210886Spjd * addresses, not from the new ones. 1808210886Spjd */ 1809210886Spjd modified |= MODIFIED_REMOTEADDR; 1810210886Spjd } 1811210886Spjd if (gres->hr_replication != res->hr_replication) { 1812210886Spjd gres->hr_replication = res->hr_replication; 1813210886Spjd modified |= MODIFIED_REPLICATION; 1814210886Spjd } 1815210886Spjd if (gres->hr_timeout != res->hr_timeout) { 1816210886Spjd gres->hr_timeout = res->hr_timeout; 1817210886Spjd modified |= MODIFIED_TIMEOUT; 1818210886Spjd } 1819211886Spjd if (strcmp(gres->hr_exec, res->hr_exec) != 0) { 1820211886Spjd strlcpy(gres->hr_exec, res->hr_exec, sizeof(gres->hr_exec)); 1821211886Spjd modified |= MODIFIED_EXEC; 1822211886Spjd } 1823210886Spjd /* 1824210886Spjd * If only timeout was modified we only need to change it without 1825210886Spjd * reconnecting. 1826210886Spjd */ 1827210886Spjd if (modified == MODIFIED_TIMEOUT) { 1828210886Spjd for (ii = 0; ii < ncomps; ii++) { 1829210886Spjd if (!ISREMOTE(ii)) 1830210886Spjd continue; 1831210886Spjd rw_rlock(&hio_remote_lock[ii]); 1832210886Spjd if (!ISCONNECTED(gres, ii)) { 1833210886Spjd rw_unlock(&hio_remote_lock[ii]); 1834210886Spjd continue; 1835210886Spjd } 1836210886Spjd rw_unlock(&hio_remote_lock[ii]); 1837210886Spjd if (proto_timeout(gres->hr_remotein, 1838210886Spjd gres->hr_timeout) < 0) { 1839210886Spjd pjdlog_errno(LOG_WARNING, 1840210886Spjd "Unable to set connection timeout"); 1841210886Spjd } 1842210886Spjd if (proto_timeout(gres->hr_remoteout, 1843210886Spjd gres->hr_timeout) < 0) { 1844210886Spjd pjdlog_errno(LOG_WARNING, 1845210886Spjd "Unable to set connection timeout"); 1846210886Spjd } 1847210886Spjd } 1848211886Spjd } else if ((modified & 1849211886Spjd (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1850210886Spjd for (ii = 0; ii < ncomps; ii++) { 1851210886Spjd if (!ISREMOTE(ii)) 1852210886Spjd continue; 1853210886Spjd remote_close(gres, ii); 1854210886Spjd } 1855210886Spjd if (modified & MODIFIED_REMOTEADDR) { 1856210886Spjd strlcpy(gres->hr_remoteaddr, res->hr_remoteaddr, 1857210886Spjd sizeof(gres->hr_remoteaddr)); 1858210886Spjd } 1859210886Spjd } 1860210886Spjd#undef MODIFIED_REMOTEADDR 1861210886Spjd#undef MODIFIED_REPLICATION 1862210886Spjd#undef MODIFIED_TIMEOUT 1863211886Spjd#undef MODIFIED_EXEC 1864210886Spjd 1865210886Spjd pjdlog_info("Configuration reloaded successfully."); 1866210886Spjd return; 1867210886Spjdfailed: 1868210886Spjd if (newcfg != NULL) { 1869210886Spjd if (newcfg->hc_controlconn != NULL) 1870210886Spjd proto_close(newcfg->hc_controlconn); 1871210886Spjd if (newcfg->hc_listenconn != NULL) 1872210886Spjd proto_close(newcfg->hc_listenconn); 1873210886Spjd yy_config_free(newcfg); 1874210886Spjd } 1875210886Spjd pjdlog_warning("Configuration not reloaded."); 1876210886Spjd} 1877210886Spjd 1878211882Spjdstatic void 1879211882Spjdkeepalive_send(struct hast_resource *res, unsigned int ncomp) 1880211882Spjd{ 1881211882Spjd struct nv *nv; 1882211882Spjd 1883211882Spjd nv = nv_alloc(); 1884211882Spjd nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1885211882Spjd if (nv_error(nv) != 0) { 1886211882Spjd nv_free(nv); 1887211882Spjd pjdlog_debug(1, 1888211882Spjd "keepalive_send: Unable to prepare header to send."); 1889211882Spjd return; 1890211882Spjd } 1891211882Spjd if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1892211882Spjd pjdlog_common(LOG_DEBUG, 1, errno, 1893211882Spjd "keepalive_send: Unable to send request"); 1894211882Spjd nv_free(nv); 1895211882Spjd rw_unlock(&hio_remote_lock[ncomp]); 1896211882Spjd remote_close(res, ncomp); 1897211882Spjd rw_rlock(&hio_remote_lock[ncomp]); 1898211882Spjd return; 1899211882Spjd } 1900211882Spjd nv_free(nv); 1901211882Spjd pjdlog_debug(2, "keepalive_send: Request sent."); 1902211882Spjd} 1903211882Spjd 1904204076Spjd/* 1905204076Spjd * Thread guards remote connections and reconnects when needed, handles 1906204076Spjd * signals, etc. 1907204076Spjd */ 1908204076Spjdstatic void * 1909204076Spjdguard_thread(void *arg) 1910204076Spjd{ 1911204076Spjd struct hast_resource *res = arg; 1912205738Spjd struct proto_conn *in, *out; 1913204076Spjd unsigned int ii, ncomps; 1914204076Spjd int timeout; 1915204076Spjd 1916204076Spjd ncomps = HAST_NCOMPONENTS; 1917204076Spjd 1918204076Spjd for (;;) { 1919204076Spjd if (sigexit_received) { 1920204076Spjd primary_exitx(EX_OK, 1921204076Spjd "Termination signal received, exiting."); 1922204076Spjd } 1923210886Spjd if (sighup_received) { 1924210886Spjd sighup_received = false; 1925210886Spjd config_reload(); 1926210886Spjd } 1927211886Spjd hook_check(sigchld_received); 1928211886Spjd if (sigchld_received) 1929211886Spjd sigchld_received = false; 1930211882Spjd 1931211882Spjd timeout = KEEPALIVE_SLEEP; 1932204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 1933204076Spjd mtx_lock(&hio_guard_lock); 1934204076Spjd for (ii = 0; ii < ncomps; ii++) { 1935204076Spjd if (!ISREMOTE(ii)) 1936204076Spjd continue; 1937204076Spjd rw_rlock(&hio_remote_lock[ii]); 1938204076Spjd if (ISCONNECTED(res, ii)) { 1939204076Spjd assert(res->hr_remotein != NULL); 1940204076Spjd assert(res->hr_remoteout != NULL); 1941211882Spjd keepalive_send(res, ii); 1942211882Spjd } 1943211882Spjd if (ISCONNECTED(res, ii)) { 1944211882Spjd assert(res->hr_remotein != NULL); 1945211882Spjd assert(res->hr_remoteout != NULL); 1946204076Spjd rw_unlock(&hio_remote_lock[ii]); 1947204076Spjd pjdlog_debug(2, 1948204076Spjd "remote_guard: Connection to %s is ok.", 1949204076Spjd res->hr_remoteaddr); 1950210881Spjd } else if (real_remote(res)) { 1951204076Spjd assert(res->hr_remotein == NULL); 1952204076Spjd assert(res->hr_remoteout == NULL); 1953204076Spjd /* 1954204076Spjd * Upgrade the lock. It doesn't have to be 1955204076Spjd * atomic as no other thread can change 1956204076Spjd * connection status from disconnected to 1957204076Spjd * connected. 1958204076Spjd */ 1959204076Spjd rw_unlock(&hio_remote_lock[ii]); 1960204076Spjd pjdlog_debug(2, 1961204076Spjd "remote_guard: Reconnecting to %s.", 1962204076Spjd res->hr_remoteaddr); 1963205738Spjd in = out = NULL; 1964205738Spjd if (init_remote(res, &in, &out)) { 1965205738Spjd rw_wlock(&hio_remote_lock[ii]); 1966205738Spjd assert(res->hr_remotein == NULL); 1967205738Spjd assert(res->hr_remoteout == NULL); 1968205738Spjd assert(in != NULL && out != NULL); 1969205738Spjd res->hr_remotein = in; 1970205738Spjd res->hr_remoteout = out; 1971205738Spjd rw_unlock(&hio_remote_lock[ii]); 1972204076Spjd pjdlog_info("Successfully reconnected to %s.", 1973204076Spjd res->hr_remoteaddr); 1974205738Spjd sync_start(); 1975204076Spjd } else { 1976204076Spjd /* Both connections should be NULL. */ 1977204076Spjd assert(res->hr_remotein == NULL); 1978204076Spjd assert(res->hr_remoteout == NULL); 1979205738Spjd assert(in == NULL && out == NULL); 1980204076Spjd pjdlog_debug(2, 1981204076Spjd "remote_guard: Reconnect to %s failed.", 1982204076Spjd res->hr_remoteaddr); 1983204076Spjd timeout = RECONNECT_SLEEP; 1984204076Spjd } 1985210881Spjd } else { 1986210881Spjd rw_unlock(&hio_remote_lock[ii]); 1987204076Spjd } 1988204076Spjd } 1989211896Spjd /* Sleep only if a signal wasn't delivered in the meantime. */ 1990211896Spjd if (!sigexit_received && !sighup_received && !sigchld_received) 1991211896Spjd cv_timedwait(&hio_guard_cond, &hio_guard_lock, timeout); 1992204076Spjd mtx_unlock(&hio_guard_lock); 1993204076Spjd } 1994204076Spjd /* NOTREACHED */ 1995204076Spjd return (NULL); 1996204076Spjd} 1997