primary.c revision 214284
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 214284 2010-10-24 17:28:25Z 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> 49211982Spjd#include <signal.h> 50204076Spjd#include <stdint.h> 51204076Spjd#include <stdio.h> 52204076Spjd#include <string.h> 53204076Spjd#include <sysexits.h> 54204076Spjd#include <unistd.h> 55204076Spjd 56204076Spjd#include <activemap.h> 57204076Spjd#include <nv.h> 58204076Spjd#include <rangelock.h> 59204076Spjd 60204076Spjd#include "control.h" 61212038Spjd#include "event.h" 62204076Spjd#include "hast.h" 63204076Spjd#include "hast_proto.h" 64204076Spjd#include "hastd.h" 65211886Spjd#include "hooks.h" 66204076Spjd#include "metadata.h" 67204076Spjd#include "proto.h" 68204076Spjd#include "pjdlog.h" 69204076Spjd#include "subr.h" 70204076Spjd#include "synch.h" 71204076Spjd 72210886Spjd/* The is only one remote component for now. */ 73210886Spjd#define ISREMOTE(no) ((no) == 1) 74210886Spjd 75204076Spjdstruct hio { 76204076Spjd /* 77204076Spjd * Number of components we are still waiting for. 78204076Spjd * When this field goes to 0, we can send the request back to the 79204076Spjd * kernel. Each component has to decrease this counter by one 80204076Spjd * even on failure. 81204076Spjd */ 82204076Spjd unsigned int hio_countdown; 83204076Spjd /* 84204076Spjd * Each component has a place to store its own error. 85204076Spjd * Once the request is handled by all components we can decide if the 86204076Spjd * request overall is successful or not. 87204076Spjd */ 88204076Spjd int *hio_errors; 89204076Spjd /* 90204076Spjd * Structure used to comunicate with GEOM Gate class. 91204076Spjd */ 92204076Spjd struct g_gate_ctl_io hio_ggio; 93204076Spjd TAILQ_ENTRY(hio) *hio_next; 94204076Spjd}; 95204076Spjd#define hio_free_next hio_next[0] 96204076Spjd#define hio_done_next hio_next[0] 97204076Spjd 98204076Spjd/* 99204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 100204076Spjd * until some in-progress requests are freed. 101204076Spjd */ 102204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 103204076Spjdstatic pthread_mutex_t hio_free_list_lock; 104204076Spjdstatic pthread_cond_t hio_free_list_cond; 105204076Spjd/* 106204076Spjd * There is one send list for every component. One requests is placed on all 107204076Spjd * send lists - each component gets the same request, but each component is 108204076Spjd * responsible for managing his own send list. 109204076Spjd */ 110204076Spjdstatic TAILQ_HEAD(, hio) *hio_send_list; 111204076Spjdstatic pthread_mutex_t *hio_send_list_lock; 112204076Spjdstatic pthread_cond_t *hio_send_list_cond; 113204076Spjd/* 114204076Spjd * There is one recv list for every component, although local components don't 115204076Spjd * use recv lists as local requests are done synchronously. 116204076Spjd */ 117204076Spjdstatic TAILQ_HEAD(, hio) *hio_recv_list; 118204076Spjdstatic pthread_mutex_t *hio_recv_list_lock; 119204076Spjdstatic pthread_cond_t *hio_recv_list_cond; 120204076Spjd/* 121204076Spjd * Request is placed on done list by the slowest component (the one that 122204076Spjd * decreased hio_countdown from 1 to 0). 123204076Spjd */ 124204076Spjdstatic TAILQ_HEAD(, hio) hio_done_list; 125204076Spjdstatic pthread_mutex_t hio_done_list_lock; 126204076Spjdstatic pthread_cond_t hio_done_list_cond; 127204076Spjd/* 128204076Spjd * Structure below are for interaction with sync thread. 129204076Spjd */ 130204076Spjdstatic bool sync_inprogress; 131204076Spjdstatic pthread_mutex_t sync_lock; 132204076Spjdstatic pthread_cond_t sync_cond; 133204076Spjd/* 134204076Spjd * The lock below allows to synchornize access to remote connections. 135204076Spjd */ 136204076Spjdstatic pthread_rwlock_t *hio_remote_lock; 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/* 155211982Spjd * Number of seconds to sleep between reconnect retries or keepalive packets. 156204076Spjd */ 157211982Spjd#define RETRY_SLEEP 10 158204076Spjd 159204076Spjd#define ISCONNECTED(res, no) \ 160204076Spjd ((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL) 161204076Spjd 162204076Spjd#define QUEUE_INSERT1(hio, name, ncomp) do { \ 163204076Spjd bool _wakeup; \ 164204076Spjd \ 165204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 166204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]); \ 167204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio), \ 168204076Spjd hio_next[(ncomp)]); \ 169204076Spjd mtx_unlock(&hio_##name##_list_lock[ncomp]); \ 170204076Spjd if (_wakeup) \ 171204076Spjd cv_signal(&hio_##name##_list_cond[(ncomp)]); \ 172204076Spjd} while (0) 173204076Spjd#define QUEUE_INSERT2(hio, name) do { \ 174204076Spjd bool _wakeup; \ 175204076Spjd \ 176204076Spjd mtx_lock(&hio_##name##_list_lock); \ 177204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 178204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\ 179204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 180204076Spjd if (_wakeup) \ 181204076Spjd cv_signal(&hio_##name##_list_cond); \ 182204076Spjd} while (0) 183204076Spjd#define QUEUE_TAKE1(hio, name, ncomp) do { \ 184204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 185204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL) { \ 186204076Spjd cv_wait(&hio_##name##_list_cond[(ncomp)], \ 187204076Spjd &hio_##name##_list_lock[(ncomp)]); \ 188204076Spjd } \ 189204076Spjd TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio), \ 190204076Spjd hio_next[(ncomp)]); \ 191204076Spjd mtx_unlock(&hio_##name##_list_lock[(ncomp)]); \ 192204076Spjd} while (0) 193204076Spjd#define QUEUE_TAKE2(hio, name) do { \ 194204076Spjd mtx_lock(&hio_##name##_list_lock); \ 195204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 196204076Spjd cv_wait(&hio_##name##_list_cond, \ 197204076Spjd &hio_##name##_list_lock); \ 198204076Spjd } \ 199204076Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next); \ 200204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 201204076Spjd} while (0) 202204076Spjd 203209183Spjd#define SYNCREQ(hio) do { \ 204209183Spjd (hio)->hio_ggio.gctl_unit = -1; \ 205209183Spjd (hio)->hio_ggio.gctl_seq = 1; \ 206209183Spjd} while (0) 207204076Spjd#define ISSYNCREQ(hio) ((hio)->hio_ggio.gctl_unit == -1) 208204076Spjd#define SYNCREQDONE(hio) do { (hio)->hio_ggio.gctl_unit = -2; } while (0) 209204076Spjd#define ISSYNCREQDONE(hio) ((hio)->hio_ggio.gctl_unit == -2) 210204076Spjd 211204076Spjdstatic struct hast_resource *gres; 212204076Spjd 213204076Spjdstatic pthread_mutex_t range_lock; 214204076Spjdstatic struct rangelocks *range_regular; 215204076Spjdstatic bool range_regular_wait; 216204076Spjdstatic pthread_cond_t range_regular_cond; 217204076Spjdstatic struct rangelocks *range_sync; 218204076Spjdstatic bool range_sync_wait; 219204076Spjdstatic pthread_cond_t range_sync_cond; 220204076Spjd 221204076Spjdstatic void *ggate_recv_thread(void *arg); 222204076Spjdstatic void *local_send_thread(void *arg); 223204076Spjdstatic void *remote_send_thread(void *arg); 224204076Spjdstatic void *remote_recv_thread(void *arg); 225204076Spjdstatic void *ggate_send_thread(void *arg); 226204076Spjdstatic void *sync_thread(void *arg); 227204076Spjdstatic void *guard_thread(void *arg); 228204076Spjd 229211982Spjdstatic void 230204076Spjdcleanup(struct hast_resource *res) 231204076Spjd{ 232204076Spjd int rerrno; 233204076Spjd 234204076Spjd /* Remember errno. */ 235204076Spjd rerrno = errno; 236204076Spjd 237204076Spjd /* Destroy ggate provider if we created one. */ 238204076Spjd if (res->hr_ggateunit >= 0) { 239204076Spjd struct g_gate_ctl_destroy ggiod; 240204076Spjd 241213533Spjd bzero(&ggiod, sizeof(ggiod)); 242204076Spjd ggiod.gctl_version = G_GATE_VERSION; 243204076Spjd ggiod.gctl_unit = res->hr_ggateunit; 244204076Spjd ggiod.gctl_force = 1; 245204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 246213531Spjd pjdlog_errno(LOG_WARNING, 247213531Spjd "Unable to destroy hast/%s device", 248204076Spjd res->hr_provname); 249204076Spjd } 250204076Spjd res->hr_ggateunit = -1; 251204076Spjd } 252204076Spjd 253204076Spjd /* Restore errno. */ 254204076Spjd errno = rerrno; 255204076Spjd} 256204076Spjd 257212899Spjdstatic __dead2 void 258204076Spjdprimary_exit(int exitcode, const char *fmt, ...) 259204076Spjd{ 260204076Spjd va_list ap; 261204076Spjd 262204076Spjd assert(exitcode != EX_OK); 263204076Spjd va_start(ap, fmt); 264204076Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 265204076Spjd va_end(ap); 266204076Spjd cleanup(gres); 267204076Spjd exit(exitcode); 268204076Spjd} 269204076Spjd 270212899Spjdstatic __dead2 void 271204076Spjdprimary_exitx(int exitcode, const char *fmt, ...) 272204076Spjd{ 273204076Spjd va_list ap; 274204076Spjd 275204076Spjd va_start(ap, fmt); 276204076Spjd pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 277204076Spjd va_end(ap); 278204076Spjd cleanup(gres); 279204076Spjd exit(exitcode); 280204076Spjd} 281204076Spjd 282204076Spjdstatic int 283204076Spjdhast_activemap_flush(struct hast_resource *res) 284204076Spjd{ 285204076Spjd const unsigned char *buf; 286204076Spjd size_t size; 287204076Spjd 288204076Spjd buf = activemap_bitmap(res->hr_amp, &size); 289204076Spjd assert(buf != NULL); 290204076Spjd assert((size % res->hr_local_sectorsize) == 0); 291204076Spjd if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 292204076Spjd (ssize_t)size) { 293204076Spjd KEEP_ERRNO(pjdlog_errno(LOG_ERR, 294204076Spjd "Unable to flush activemap to disk")); 295204076Spjd return (-1); 296204076Spjd } 297204076Spjd return (0); 298204076Spjd} 299204076Spjd 300210881Spjdstatic bool 301210881Spjdreal_remote(const struct hast_resource *res) 302210881Spjd{ 303210881Spjd 304210881Spjd return (strcmp(res->hr_remoteaddr, "none") != 0); 305210881Spjd} 306210881Spjd 307204076Spjdstatic void 308204076Spjdinit_environment(struct hast_resource *res __unused) 309204076Spjd{ 310204076Spjd struct hio *hio; 311204076Spjd unsigned int ii, ncomps; 312204076Spjd 313204076Spjd /* 314204076Spjd * In the future it might be per-resource value. 315204076Spjd */ 316204076Spjd ncomps = HAST_NCOMPONENTS; 317204076Spjd 318204076Spjd /* 319204076Spjd * Allocate memory needed by lists. 320204076Spjd */ 321204076Spjd hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 322204076Spjd if (hio_send_list == NULL) { 323204076Spjd primary_exitx(EX_TEMPFAIL, 324204076Spjd "Unable to allocate %zu bytes of memory for send lists.", 325204076Spjd sizeof(hio_send_list[0]) * ncomps); 326204076Spjd } 327204076Spjd hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 328204076Spjd if (hio_send_list_lock == NULL) { 329204076Spjd primary_exitx(EX_TEMPFAIL, 330204076Spjd "Unable to allocate %zu bytes of memory for send list locks.", 331204076Spjd sizeof(hio_send_list_lock[0]) * ncomps); 332204076Spjd } 333204076Spjd hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 334204076Spjd if (hio_send_list_cond == NULL) { 335204076Spjd primary_exitx(EX_TEMPFAIL, 336204076Spjd "Unable to allocate %zu bytes of memory for send list condition variables.", 337204076Spjd sizeof(hio_send_list_cond[0]) * ncomps); 338204076Spjd } 339204076Spjd hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 340204076Spjd if (hio_recv_list == NULL) { 341204076Spjd primary_exitx(EX_TEMPFAIL, 342204076Spjd "Unable to allocate %zu bytes of memory for recv lists.", 343204076Spjd sizeof(hio_recv_list[0]) * ncomps); 344204076Spjd } 345204076Spjd hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 346204076Spjd if (hio_recv_list_lock == NULL) { 347204076Spjd primary_exitx(EX_TEMPFAIL, 348204076Spjd "Unable to allocate %zu bytes of memory for recv list locks.", 349204076Spjd sizeof(hio_recv_list_lock[0]) * ncomps); 350204076Spjd } 351204076Spjd hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 352204076Spjd if (hio_recv_list_cond == NULL) { 353204076Spjd primary_exitx(EX_TEMPFAIL, 354204076Spjd "Unable to allocate %zu bytes of memory for recv list condition variables.", 355204076Spjd sizeof(hio_recv_list_cond[0]) * ncomps); 356204076Spjd } 357204076Spjd hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 358204076Spjd if (hio_remote_lock == NULL) { 359204076Spjd primary_exitx(EX_TEMPFAIL, 360204076Spjd "Unable to allocate %zu bytes of memory for remote connections locks.", 361204076Spjd sizeof(hio_remote_lock[0]) * ncomps); 362204076Spjd } 363204076Spjd 364204076Spjd /* 365204076Spjd * Initialize lists, their locks and theirs condition variables. 366204076Spjd */ 367204076Spjd TAILQ_INIT(&hio_free_list); 368204076Spjd mtx_init(&hio_free_list_lock); 369204076Spjd cv_init(&hio_free_list_cond); 370204076Spjd for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 371204076Spjd TAILQ_INIT(&hio_send_list[ii]); 372204076Spjd mtx_init(&hio_send_list_lock[ii]); 373204076Spjd cv_init(&hio_send_list_cond[ii]); 374204076Spjd TAILQ_INIT(&hio_recv_list[ii]); 375204076Spjd mtx_init(&hio_recv_list_lock[ii]); 376204076Spjd cv_init(&hio_recv_list_cond[ii]); 377204076Spjd rw_init(&hio_remote_lock[ii]); 378204076Spjd } 379204076Spjd TAILQ_INIT(&hio_done_list); 380204076Spjd mtx_init(&hio_done_list_lock); 381204076Spjd cv_init(&hio_done_list_cond); 382204076Spjd mtx_init(&metadata_lock); 383204076Spjd 384204076Spjd /* 385204076Spjd * Allocate requests pool and initialize requests. 386204076Spjd */ 387204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 388204076Spjd hio = malloc(sizeof(*hio)); 389204076Spjd if (hio == NULL) { 390204076Spjd primary_exitx(EX_TEMPFAIL, 391204076Spjd "Unable to allocate %zu bytes of memory for hio request.", 392204076Spjd sizeof(*hio)); 393204076Spjd } 394204076Spjd hio->hio_countdown = 0; 395204076Spjd hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 396204076Spjd if (hio->hio_errors == NULL) { 397204076Spjd primary_exitx(EX_TEMPFAIL, 398204076Spjd "Unable allocate %zu bytes of memory for hio errors.", 399204076Spjd sizeof(hio->hio_errors[0]) * ncomps); 400204076Spjd } 401204076Spjd hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 402204076Spjd if (hio->hio_next == NULL) { 403204076Spjd primary_exitx(EX_TEMPFAIL, 404204076Spjd "Unable allocate %zu bytes of memory for hio_next field.", 405204076Spjd sizeof(hio->hio_next[0]) * ncomps); 406204076Spjd } 407204076Spjd hio->hio_ggio.gctl_version = G_GATE_VERSION; 408204076Spjd hio->hio_ggio.gctl_data = malloc(MAXPHYS); 409204076Spjd if (hio->hio_ggio.gctl_data == NULL) { 410204076Spjd primary_exitx(EX_TEMPFAIL, 411204076Spjd "Unable to allocate %zu bytes of memory for gctl_data.", 412204076Spjd MAXPHYS); 413204076Spjd } 414204076Spjd hio->hio_ggio.gctl_length = MAXPHYS; 415204076Spjd hio->hio_ggio.gctl_error = 0; 416204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 417204076Spjd } 418204076Spjd} 419204076Spjd 420214284Spjdstatic bool 421214284Spjdinit_resuid(struct hast_resource *res) 422214284Spjd{ 423214284Spjd 424214284Spjd mtx_lock(&metadata_lock); 425214284Spjd if (res->hr_resuid != 0) { 426214284Spjd mtx_unlock(&metadata_lock); 427214284Spjd return (false); 428214284Spjd } else { 429214284Spjd /* Initialize unique resource identifier. */ 430214284Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 431214284Spjd mtx_unlock(&metadata_lock); 432214284Spjd if (metadata_write(res) < 0) 433214284Spjd exit(EX_NOINPUT); 434214284Spjd return (true); 435214284Spjd } 436214284Spjd} 437214284Spjd 438204076Spjdstatic void 439204076Spjdinit_local(struct hast_resource *res) 440204076Spjd{ 441204076Spjd unsigned char *buf; 442204076Spjd size_t mapsize; 443204076Spjd 444204076Spjd if (metadata_read(res, true) < 0) 445204076Spjd exit(EX_NOINPUT); 446204076Spjd mtx_init(&res->hr_amp_lock); 447204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 448204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 449204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 450204076Spjd } 451204076Spjd mtx_init(&range_lock); 452204076Spjd cv_init(&range_regular_cond); 453204076Spjd if (rangelock_init(&range_regular) < 0) 454204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 455204076Spjd cv_init(&range_sync_cond); 456204076Spjd if (rangelock_init(&range_sync) < 0) 457204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 458204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 459204076Spjd buf = calloc(1, mapsize); 460204076Spjd if (buf == NULL) { 461204076Spjd primary_exitx(EX_TEMPFAIL, 462204076Spjd "Unable to allocate buffer for activemap."); 463204076Spjd } 464204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 465204076Spjd (ssize_t)mapsize) { 466204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 467204076Spjd } 468204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 469209181Spjd free(buf); 470204076Spjd if (res->hr_resuid != 0) 471204076Spjd return; 472204076Spjd /* 473214284Spjd * We're using provider for the first time. Initialize local and remote 474214284Spjd * counters. We don't initialize resuid here, as we want to do it just 475214284Spjd * in time. The reason for this is that we want to inform secondary 476214284Spjd * that there were no writes yet, so there is no need to synchronize 477214284Spjd * anything. 478204076Spjd */ 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; 503211983Spjd errmsg = NULL; 504205738Spjd 505204076Spjd /* Prepare outgoing connection with remote node. */ 506205738Spjd if (proto_client(res->hr_remoteaddr, &out) < 0) { 507207347Spjd primary_exit(EX_TEMPFAIL, "Unable to create connection to %s", 508204076Spjd res->hr_remoteaddr); 509204076Spjd } 510204076Spjd /* Try to connect, but accept failure. */ 511205738Spjd if (proto_connect(out) < 0) { 512204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 513204076Spjd res->hr_remoteaddr); 514204076Spjd goto close; 515204076Spjd } 516207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 517207371Spjd if (proto_timeout(out, res->hr_timeout) < 0) 518207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 519204076Spjd /* 520204076Spjd * First handshake step. 521204076Spjd * Setup outgoing connection with remote node. 522204076Spjd */ 523204076Spjd nvout = nv_alloc(); 524204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 525204076Spjd if (nv_error(nvout) != 0) { 526204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 527204076Spjd "Unable to allocate header for connection with %s", 528204076Spjd res->hr_remoteaddr); 529204076Spjd nv_free(nvout); 530204076Spjd goto close; 531204076Spjd } 532205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 533204076Spjd pjdlog_errno(LOG_WARNING, 534204076Spjd "Unable to send handshake header to %s", 535204076Spjd res->hr_remoteaddr); 536204076Spjd nv_free(nvout); 537204076Spjd goto close; 538204076Spjd } 539204076Spjd nv_free(nvout); 540205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 541204076Spjd pjdlog_errno(LOG_WARNING, 542204076Spjd "Unable to receive handshake header from %s", 543204076Spjd res->hr_remoteaddr); 544204076Spjd goto close; 545204076Spjd } 546204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 547204076Spjd if (errmsg != NULL) { 548204076Spjd pjdlog_warning("%s", errmsg); 549204076Spjd nv_free(nvin); 550204076Spjd goto close; 551204076Spjd } 552204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 553204076Spjd if (token == NULL) { 554204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 555204076Spjd res->hr_remoteaddr); 556204076Spjd nv_free(nvin); 557204076Spjd goto close; 558204076Spjd } 559204076Spjd if (size != sizeof(res->hr_token)) { 560204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 561204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 562204076Spjd nv_free(nvin); 563204076Spjd goto close; 564204076Spjd } 565204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 566204076Spjd nv_free(nvin); 567204076Spjd 568204076Spjd /* 569204076Spjd * Second handshake step. 570204076Spjd * Setup incoming connection with remote node. 571204076Spjd */ 572205738Spjd if (proto_client(res->hr_remoteaddr, &in) < 0) { 573204076Spjd pjdlog_errno(LOG_WARNING, "Unable to create connection to %s", 574204076Spjd res->hr_remoteaddr); 575204076Spjd } 576204076Spjd /* Try to connect, but accept failure. */ 577205738Spjd if (proto_connect(in) < 0) { 578204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 579204076Spjd res->hr_remoteaddr); 580204076Spjd goto close; 581204076Spjd } 582207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 583207371Spjd if (proto_timeout(in, res->hr_timeout) < 0) 584207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 585204076Spjd nvout = nv_alloc(); 586204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 587204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 588204076Spjd "token"); 589214284Spjd if (res->hr_resuid == 0) { 590214284Spjd /* 591214284Spjd * The resuid field was not yet initialized. 592214284Spjd * Because we do synchronization inside init_resuid(), it is 593214284Spjd * possible that someone already initialized it, the function 594214284Spjd * will return false then, but if we successfully initialized 595214284Spjd * it, we will get true. True means that there were no writes 596214284Spjd * to this resource yet and we want to inform secondary that 597214284Spjd * synchronization is not needed by sending "virgin" argument. 598214284Spjd */ 599214284Spjd if (init_resuid(res)) 600214284Spjd nv_add_int8(nvout, 1, "virgin"); 601214284Spjd } 602204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 603204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 604204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 605204076Spjd if (nv_error(nvout) != 0) { 606204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 607204076Spjd "Unable to allocate header for connection with %s", 608204076Spjd res->hr_remoteaddr); 609204076Spjd nv_free(nvout); 610204076Spjd goto close; 611204076Spjd } 612205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 613204076Spjd pjdlog_errno(LOG_WARNING, 614204076Spjd "Unable to send handshake header to %s", 615204076Spjd res->hr_remoteaddr); 616204076Spjd nv_free(nvout); 617204076Spjd goto close; 618204076Spjd } 619204076Spjd nv_free(nvout); 620205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 621204076Spjd pjdlog_errno(LOG_WARNING, 622204076Spjd "Unable to receive handshake header from %s", 623204076Spjd res->hr_remoteaddr); 624204076Spjd goto close; 625204076Spjd } 626204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 627204076Spjd if (errmsg != NULL) { 628204076Spjd pjdlog_warning("%s", errmsg); 629204076Spjd nv_free(nvin); 630204076Spjd goto close; 631204076Spjd } 632204076Spjd datasize = nv_get_int64(nvin, "datasize"); 633204076Spjd if (datasize != res->hr_datasize) { 634204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 635204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 636204076Spjd nv_free(nvin); 637204076Spjd goto close; 638204076Spjd } 639204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 640204076Spjd if (extentsize != res->hr_extentsize) { 641204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 642204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 643204076Spjd nv_free(nvin); 644204076Spjd goto close; 645204076Spjd } 646204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 647204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 648204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 649204076Spjd map = NULL; 650204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 651204076Spjd if (mapsize > 0) { 652204076Spjd map = malloc(mapsize); 653204076Spjd if (map == NULL) { 654204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 655204076Spjd (uintmax_t)mapsize); 656204076Spjd nv_free(nvin); 657204076Spjd goto close; 658204076Spjd } 659204076Spjd /* 660204076Spjd * Remote node have some dirty extents on its own, lets 661204076Spjd * download its activemap. 662204076Spjd */ 663205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 664204076Spjd mapsize) < 0) { 665204076Spjd pjdlog_errno(LOG_ERR, 666204076Spjd "Unable to receive remote activemap"); 667204076Spjd nv_free(nvin); 668204076Spjd free(map); 669204076Spjd goto close; 670204076Spjd } 671204076Spjd /* 672204076Spjd * Merge local and remote bitmaps. 673204076Spjd */ 674204076Spjd activemap_merge(res->hr_amp, map, mapsize); 675204076Spjd free(map); 676204076Spjd /* 677204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 678204076Spjd * disk before we start to synchronize. 679204076Spjd */ 680204076Spjd (void)hast_activemap_flush(res); 681204076Spjd } 682214274Spjd nv_free(nvin); 683204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 684205738Spjd if (inp != NULL && outp != NULL) { 685205738Spjd *inp = in; 686205738Spjd *outp = out; 687205738Spjd } else { 688205738Spjd res->hr_remotein = in; 689205738Spjd res->hr_remoteout = out; 690205738Spjd } 691212038Spjd event_send(res, EVENT_CONNECT); 692205738Spjd return (true); 693205738Spjdclose: 694211983Spjd if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0) 695212038Spjd event_send(res, EVENT_SPLITBRAIN); 696205738Spjd proto_close(out); 697205738Spjd if (in != NULL) 698205738Spjd proto_close(in); 699205738Spjd return (false); 700205738Spjd} 701205738Spjd 702205738Spjdstatic void 703205738Spjdsync_start(void) 704205738Spjd{ 705205738Spjd 706204076Spjd mtx_lock(&sync_lock); 707204076Spjd sync_inprogress = true; 708204076Spjd mtx_unlock(&sync_lock); 709204076Spjd cv_signal(&sync_cond); 710204076Spjd} 711204076Spjd 712204076Spjdstatic void 713211878Spjdsync_stop(void) 714211878Spjd{ 715211878Spjd 716211878Spjd mtx_lock(&sync_lock); 717211878Spjd if (sync_inprogress) 718211878Spjd sync_inprogress = false; 719211878Spjd mtx_unlock(&sync_lock); 720211878Spjd} 721211878Spjd 722211878Spjdstatic void 723204076Spjdinit_ggate(struct hast_resource *res) 724204076Spjd{ 725204076Spjd struct g_gate_ctl_create ggiocreate; 726204076Spjd struct g_gate_ctl_cancel ggiocancel; 727204076Spjd 728204076Spjd /* 729204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 730204076Spjd */ 731204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 732204076Spjd if (res->hr_ggatefd < 0) 733204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 734204076Spjd /* 735204076Spjd * Create provider before trying to connect, as connection failure 736204076Spjd * is not critical, but may take some time. 737204076Spjd */ 738213533Spjd bzero(&ggiocreate, sizeof(ggiocreate)); 739204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 740204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 741204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 742204076Spjd ggiocreate.gctl_flags = 0; 743206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 744204076Spjd ggiocreate.gctl_timeout = 0; 745204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 746204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 747204076Spjd res->hr_provname); 748204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 749204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 750204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 751204076Spjd return; 752204076Spjd } 753204076Spjd if (errno != EEXIST) { 754204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 755204076Spjd res->hr_provname); 756204076Spjd } 757204076Spjd pjdlog_debug(1, 758204076Spjd "Device hast/%s already exists, we will try to take it over.", 759204076Spjd res->hr_provname); 760204076Spjd /* 761204076Spjd * If we received EEXIST, we assume that the process who created the 762204076Spjd * provider died and didn't clean up. In that case we will start from 763204076Spjd * where he left of. 764204076Spjd */ 765213533Spjd bzero(&ggiocancel, sizeof(ggiocancel)); 766204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 767204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 768204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 769204076Spjd res->hr_provname); 770204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 771204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 772204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 773204076Spjd return; 774204076Spjd } 775204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 776204076Spjd res->hr_provname); 777204076Spjd} 778204076Spjd 779204076Spjdvoid 780204076Spjdhastd_primary(struct hast_resource *res) 781204076Spjd{ 782204076Spjd pthread_t td; 783204076Spjd pid_t pid; 784204076Spjd int error; 785204076Spjd 786204076Spjd /* 787204076Spjd * Create communication channel between parent and child. 788204076Spjd */ 789204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 790204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 791212034Spjd pjdlog_exit(EX_OSERR, 792204076Spjd "Unable to create control sockets between parent and child"); 793204076Spjd } 794212038Spjd /* 795212038Spjd * Create communication channel between child and parent. 796212038Spjd */ 797212038Spjd if (proto_client("socketpair://", &res->hr_event) < 0) { 798212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 799212038Spjd pjdlog_exit(EX_OSERR, 800212038Spjd "Unable to create event sockets between child and parent"); 801212038Spjd } 802204076Spjd 803204076Spjd pid = fork(); 804204076Spjd if (pid < 0) { 805204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 806212034Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to fork"); 807204076Spjd } 808204076Spjd 809204076Spjd if (pid > 0) { 810204076Spjd /* This is parent. */ 811212038Spjd /* Declare that we are receiver. */ 812212038Spjd proto_recv(res->hr_event, NULL, 0); 813204076Spjd res->hr_workerpid = pid; 814204076Spjd return; 815204076Spjd } 816211977Spjd 817211984Spjd gres = res; 818211984Spjd 819204076Spjd (void)pidfile_close(pfh); 820211977Spjd hook_fini(); 821204076Spjd 822204076Spjd setproctitle("%s (primary)", res->hr_name); 823204076Spjd 824212038Spjd /* Declare that we are sender. */ 825212038Spjd proto_send(res->hr_event, NULL, 0); 826212038Spjd 827204076Spjd init_local(res); 828213007Spjd init_ggate(res); 829213007Spjd init_environment(res); 830213007Spjd /* 831213530Spjd * Create the guard thread first, so we can handle signals from the 832213530Spjd * very begining. 833213530Spjd */ 834213530Spjd error = pthread_create(&td, NULL, guard_thread, res); 835213530Spjd assert(error == 0); 836213530Spjd /* 837213007Spjd * Create the control thread before sending any event to the parent, 838213007Spjd * as we can deadlock when parent sends control request to worker, 839213007Spjd * but worker has no control thread started yet, so parent waits. 840213007Spjd * In the meantime worker sends an event to the parent, but parent 841213007Spjd * is unable to handle the event, because it waits for control 842213007Spjd * request response. 843213007Spjd */ 844213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 845213007Spjd assert(error == 0); 846210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 847205738Spjd sync_start(); 848204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 849204076Spjd assert(error == 0); 850204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 851204076Spjd assert(error == 0); 852204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 853204076Spjd assert(error == 0); 854204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 855204076Spjd assert(error == 0); 856204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 857204076Spjd assert(error == 0); 858213530Spjd (void)sync_thread(res); 859204076Spjd} 860204076Spjd 861204076Spjdstatic void 862204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 863204076Spjd{ 864204076Spjd char msg[1024]; 865204076Spjd va_list ap; 866204076Spjd int len; 867204076Spjd 868204076Spjd va_start(ap, fmt); 869204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 870204076Spjd va_end(ap); 871204076Spjd if ((size_t)len < sizeof(msg)) { 872204076Spjd switch (ggio->gctl_cmd) { 873204076Spjd case BIO_READ: 874204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 875204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 876204076Spjd (uintmax_t)ggio->gctl_length); 877204076Spjd break; 878204076Spjd case BIO_DELETE: 879204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 880204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 881204076Spjd (uintmax_t)ggio->gctl_length); 882204076Spjd break; 883204076Spjd case BIO_FLUSH: 884204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 885204076Spjd break; 886204076Spjd case BIO_WRITE: 887204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 888204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 889204076Spjd (uintmax_t)ggio->gctl_length); 890204076Spjd break; 891204076Spjd default: 892204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 893204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 894204076Spjd break; 895204076Spjd } 896204076Spjd } 897204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 898204076Spjd} 899204076Spjd 900204076Spjdstatic void 901204076Spjdremote_close(struct hast_resource *res, int ncomp) 902204076Spjd{ 903204076Spjd 904204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 905204076Spjd /* 906204076Spjd * A race is possible between dropping rlock and acquiring wlock - 907204076Spjd * another thread can close connection in-between. 908204076Spjd */ 909204076Spjd if (!ISCONNECTED(res, ncomp)) { 910204076Spjd assert(res->hr_remotein == NULL); 911204076Spjd assert(res->hr_remoteout == NULL); 912204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 913204076Spjd return; 914204076Spjd } 915204076Spjd 916204076Spjd assert(res->hr_remotein != NULL); 917204076Spjd assert(res->hr_remoteout != NULL); 918204076Spjd 919211881Spjd pjdlog_debug(2, "Closing incoming connection to %s.", 920204076Spjd res->hr_remoteaddr); 921204076Spjd proto_close(res->hr_remotein); 922204076Spjd res->hr_remotein = NULL; 923211881Spjd pjdlog_debug(2, "Closing outgoing connection to %s.", 924204076Spjd res->hr_remoteaddr); 925204076Spjd proto_close(res->hr_remoteout); 926204076Spjd res->hr_remoteout = NULL; 927204076Spjd 928204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 929204076Spjd 930211881Spjd pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 931211881Spjd 932204076Spjd /* 933204076Spjd * Stop synchronization if in-progress. 934204076Spjd */ 935211878Spjd sync_stop(); 936211984Spjd 937212038Spjd event_send(res, EVENT_DISCONNECT); 938204076Spjd} 939204076Spjd 940204076Spjd/* 941204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 942204076Spjd * appropriate threads: 943204076Spjd * WRITE - always goes to both local_send and remote_send threads 944204076Spjd * READ (when the block is up-to-date on local component) - 945204076Spjd * only local_send thread 946204076Spjd * READ (when the block isn't up-to-date on local component) - 947204076Spjd * only remote_send thread 948204076Spjd * DELETE - always goes to both local_send and remote_send threads 949204076Spjd * FLUSH - always goes to both local_send and remote_send threads 950204076Spjd */ 951204076Spjdstatic void * 952204076Spjdggate_recv_thread(void *arg) 953204076Spjd{ 954204076Spjd struct hast_resource *res = arg; 955204076Spjd struct g_gate_ctl_io *ggio; 956204076Spjd struct hio *hio; 957204076Spjd unsigned int ii, ncomp, ncomps; 958204076Spjd int error; 959204076Spjd 960204076Spjd ncomps = HAST_NCOMPONENTS; 961204076Spjd 962204076Spjd for (;;) { 963204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 964204076Spjd QUEUE_TAKE2(hio, free); 965204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 966204076Spjd ggio = &hio->hio_ggio; 967204076Spjd ggio->gctl_unit = res->hr_ggateunit; 968204076Spjd ggio->gctl_length = MAXPHYS; 969204076Spjd ggio->gctl_error = 0; 970204076Spjd pjdlog_debug(2, 971204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 972204076Spjd hio); 973204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 974204076Spjd if (sigexit_received) 975204076Spjd pthread_exit(NULL); 976204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 977204076Spjd } 978204076Spjd error = ggio->gctl_error; 979204076Spjd switch (error) { 980204076Spjd case 0: 981204076Spjd break; 982204076Spjd case ECANCELED: 983204076Spjd /* Exit gracefully. */ 984204076Spjd if (!sigexit_received) { 985204076Spjd pjdlog_debug(2, 986204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 987204076Spjd hio); 988204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 989204076Spjd } 990204076Spjd pthread_exit(NULL); 991204076Spjd case ENOMEM: 992204076Spjd /* 993204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 994204076Spjd * bytes - request can't be bigger than that. 995204076Spjd */ 996204076Spjd /* FALLTHROUGH */ 997204076Spjd case ENXIO: 998204076Spjd default: 999204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 1000204076Spjd strerror(error)); 1001204076Spjd } 1002204076Spjd for (ii = 0; ii < ncomps; ii++) 1003204076Spjd hio->hio_errors[ii] = EINVAL; 1004204076Spjd reqlog(LOG_DEBUG, 2, ggio, 1005204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 1006204076Spjd hio); 1007204076Spjd /* 1008204076Spjd * Inform all components about new write request. 1009204076Spjd * For read request prefer local component unless the given 1010204076Spjd * range is out-of-date, then use remote component. 1011204076Spjd */ 1012204076Spjd switch (ggio->gctl_cmd) { 1013204076Spjd case BIO_READ: 1014204076Spjd pjdlog_debug(2, 1015204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 1016204076Spjd hio); 1017204076Spjd refcount_init(&hio->hio_countdown, 1); 1018204076Spjd mtx_lock(&metadata_lock); 1019204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 1020204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1021204076Spjd /* 1022204076Spjd * This range is up-to-date on local component, 1023204076Spjd * so handle request locally. 1024204076Spjd */ 1025204076Spjd /* Local component is 0 for now. */ 1026204076Spjd ncomp = 0; 1027204076Spjd } else /* if (res->hr_syncsrc == 1028204076Spjd HAST_SYNCSRC_SECONDARY) */ { 1029204076Spjd assert(res->hr_syncsrc == 1030204076Spjd HAST_SYNCSRC_SECONDARY); 1031204076Spjd /* 1032204076Spjd * This range is out-of-date on local component, 1033204076Spjd * so send request to the remote node. 1034204076Spjd */ 1035204076Spjd /* Remote component is 1 for now. */ 1036204076Spjd ncomp = 1; 1037204076Spjd } 1038204076Spjd mtx_unlock(&metadata_lock); 1039204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1040204076Spjd break; 1041204076Spjd case BIO_WRITE: 1042214284Spjd if (res->hr_resuid == 0) { 1043214284Spjd /* This is first write, initialize resuid. */ 1044214284Spjd (void)init_resuid(res); 1045214284Spjd } 1046204076Spjd for (;;) { 1047204076Spjd mtx_lock(&range_lock); 1048204076Spjd if (rangelock_islocked(range_sync, 1049204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1050204076Spjd pjdlog_debug(2, 1051204076Spjd "regular: Range offset=%jd length=%zu locked.", 1052204076Spjd (intmax_t)ggio->gctl_offset, 1053204076Spjd (size_t)ggio->gctl_length); 1054204076Spjd range_regular_wait = true; 1055204076Spjd cv_wait(&range_regular_cond, &range_lock); 1056204076Spjd range_regular_wait = false; 1057204076Spjd mtx_unlock(&range_lock); 1058204076Spjd continue; 1059204076Spjd } 1060204076Spjd if (rangelock_add(range_regular, 1061204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1062204076Spjd mtx_unlock(&range_lock); 1063204076Spjd pjdlog_debug(2, 1064204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1065204076Spjd (intmax_t)ggio->gctl_offset, 1066204076Spjd (size_t)ggio->gctl_length); 1067204076Spjd sleep(1); 1068204076Spjd continue; 1069204076Spjd } 1070204076Spjd mtx_unlock(&range_lock); 1071204076Spjd break; 1072204076Spjd } 1073204076Spjd mtx_lock(&res->hr_amp_lock); 1074204076Spjd if (activemap_write_start(res->hr_amp, 1075204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1076204076Spjd (void)hast_activemap_flush(res); 1077204076Spjd } 1078204076Spjd mtx_unlock(&res->hr_amp_lock); 1079204076Spjd /* FALLTHROUGH */ 1080204076Spjd case BIO_DELETE: 1081204076Spjd case BIO_FLUSH: 1082204076Spjd pjdlog_debug(2, 1083204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1084204076Spjd hio); 1085204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1086204076Spjd for (ii = 0; ii < ncomps; ii++) 1087204076Spjd QUEUE_INSERT1(hio, send, ii); 1088204076Spjd break; 1089204076Spjd } 1090204076Spjd } 1091204076Spjd /* NOTREACHED */ 1092204076Spjd return (NULL); 1093204076Spjd} 1094204076Spjd 1095204076Spjd/* 1096204076Spjd * Thread reads from or writes to local component. 1097204076Spjd * If local read fails, it redirects it to remote_send thread. 1098204076Spjd */ 1099204076Spjdstatic void * 1100204076Spjdlocal_send_thread(void *arg) 1101204076Spjd{ 1102204076Spjd struct hast_resource *res = arg; 1103204076Spjd struct g_gate_ctl_io *ggio; 1104204076Spjd struct hio *hio; 1105204076Spjd unsigned int ncomp, rncomp; 1106204076Spjd ssize_t ret; 1107204076Spjd 1108204076Spjd /* Local component is 0 for now. */ 1109204076Spjd ncomp = 0; 1110204076Spjd /* Remote component is 1 for now. */ 1111204076Spjd rncomp = 1; 1112204076Spjd 1113204076Spjd for (;;) { 1114204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1115204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1116204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1117204076Spjd ggio = &hio->hio_ggio; 1118204076Spjd switch (ggio->gctl_cmd) { 1119204076Spjd case BIO_READ: 1120204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1121204076Spjd ggio->gctl_length, 1122204076Spjd ggio->gctl_offset + res->hr_localoff); 1123204076Spjd if (ret == ggio->gctl_length) 1124204076Spjd hio->hio_errors[ncomp] = 0; 1125204076Spjd else { 1126204076Spjd /* 1127204076Spjd * If READ failed, try to read from remote node. 1128204076Spjd */ 1129204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1130204076Spjd continue; 1131204076Spjd } 1132204076Spjd break; 1133204076Spjd case BIO_WRITE: 1134204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1135204076Spjd ggio->gctl_length, 1136204076Spjd ggio->gctl_offset + res->hr_localoff); 1137204076Spjd if (ret < 0) 1138204076Spjd hio->hio_errors[ncomp] = errno; 1139204076Spjd else if (ret != ggio->gctl_length) 1140204076Spjd hio->hio_errors[ncomp] = EIO; 1141204076Spjd else 1142204076Spjd hio->hio_errors[ncomp] = 0; 1143204076Spjd break; 1144204076Spjd case BIO_DELETE: 1145204076Spjd ret = g_delete(res->hr_localfd, 1146204076Spjd ggio->gctl_offset + res->hr_localoff, 1147204076Spjd ggio->gctl_length); 1148204076Spjd if (ret < 0) 1149204076Spjd hio->hio_errors[ncomp] = errno; 1150204076Spjd else 1151204076Spjd hio->hio_errors[ncomp] = 0; 1152204076Spjd break; 1153204076Spjd case BIO_FLUSH: 1154204076Spjd ret = g_flush(res->hr_localfd); 1155204076Spjd if (ret < 0) 1156204076Spjd hio->hio_errors[ncomp] = errno; 1157204076Spjd else 1158204076Spjd hio->hio_errors[ncomp] = 0; 1159204076Spjd break; 1160204076Spjd } 1161204076Spjd if (refcount_release(&hio->hio_countdown)) { 1162204076Spjd if (ISSYNCREQ(hio)) { 1163204076Spjd mtx_lock(&sync_lock); 1164204076Spjd SYNCREQDONE(hio); 1165204076Spjd mtx_unlock(&sync_lock); 1166204076Spjd cv_signal(&sync_cond); 1167204076Spjd } else { 1168204076Spjd pjdlog_debug(2, 1169204076Spjd "local_send: (%p) Moving request to the done queue.", 1170204076Spjd hio); 1171204076Spjd QUEUE_INSERT2(hio, done); 1172204076Spjd } 1173204076Spjd } 1174204076Spjd } 1175204076Spjd /* NOTREACHED */ 1176204076Spjd return (NULL); 1177204076Spjd} 1178204076Spjd 1179204076Spjd/* 1180204076Spjd * Thread sends request to secondary node. 1181204076Spjd */ 1182204076Spjdstatic void * 1183204076Spjdremote_send_thread(void *arg) 1184204076Spjd{ 1185204076Spjd struct hast_resource *res = arg; 1186204076Spjd struct g_gate_ctl_io *ggio; 1187204076Spjd struct hio *hio; 1188204076Spjd struct nv *nv; 1189204076Spjd unsigned int ncomp; 1190204076Spjd bool wakeup; 1191204076Spjd uint64_t offset, length; 1192204076Spjd uint8_t cmd; 1193204076Spjd void *data; 1194204076Spjd 1195204076Spjd /* Remote component is 1 for now. */ 1196204076Spjd ncomp = 1; 1197204076Spjd 1198204076Spjd for (;;) { 1199204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1200204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1201204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1202204076Spjd ggio = &hio->hio_ggio; 1203204076Spjd switch (ggio->gctl_cmd) { 1204204076Spjd case BIO_READ: 1205204076Spjd cmd = HIO_READ; 1206204076Spjd data = NULL; 1207204076Spjd offset = ggio->gctl_offset; 1208204076Spjd length = ggio->gctl_length; 1209204076Spjd break; 1210204076Spjd case BIO_WRITE: 1211204076Spjd cmd = HIO_WRITE; 1212204076Spjd data = ggio->gctl_data; 1213204076Spjd offset = ggio->gctl_offset; 1214204076Spjd length = ggio->gctl_length; 1215204076Spjd break; 1216204076Spjd case BIO_DELETE: 1217204076Spjd cmd = HIO_DELETE; 1218204076Spjd data = NULL; 1219204076Spjd offset = ggio->gctl_offset; 1220204076Spjd length = ggio->gctl_length; 1221204076Spjd break; 1222204076Spjd case BIO_FLUSH: 1223204076Spjd cmd = HIO_FLUSH; 1224204076Spjd data = NULL; 1225204076Spjd offset = 0; 1226204076Spjd length = 0; 1227204076Spjd break; 1228204076Spjd default: 1229204076Spjd assert(!"invalid condition"); 1230204076Spjd abort(); 1231204076Spjd } 1232204076Spjd nv = nv_alloc(); 1233204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1234204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1235204076Spjd nv_add_uint64(nv, offset, "offset"); 1236204076Spjd nv_add_uint64(nv, length, "length"); 1237204076Spjd if (nv_error(nv) != 0) { 1238204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1239204076Spjd pjdlog_debug(2, 1240204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1241204076Spjd hio); 1242204076Spjd reqlog(LOG_ERR, 0, ggio, 1243204076Spjd "Unable to prepare header to send (%s): ", 1244204076Spjd strerror(nv_error(nv))); 1245204076Spjd /* Move failed request immediately to the done queue. */ 1246204076Spjd goto done_queue; 1247204076Spjd } 1248204076Spjd pjdlog_debug(2, 1249204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1250204076Spjd hio); 1251204076Spjd /* 1252204076Spjd * Protect connection from disappearing. 1253204076Spjd */ 1254204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1255204076Spjd if (!ISCONNECTED(res, ncomp)) { 1256204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1257204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1258204076Spjd goto done_queue; 1259204076Spjd } 1260204076Spjd /* 1261204076Spjd * Move the request to recv queue before sending it, because 1262204076Spjd * in different order we can get reply before we move request 1263204076Spjd * to recv queue. 1264204076Spjd */ 1265204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1266204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1267204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1268204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1269204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1270204076Spjd data != NULL ? length : 0) < 0) { 1271204076Spjd hio->hio_errors[ncomp] = errno; 1272204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1273204076Spjd pjdlog_debug(2, 1274204076Spjd "remote_send: (%p) Unable to send request.", hio); 1275204076Spjd reqlog(LOG_ERR, 0, ggio, 1276204076Spjd "Unable to send request (%s): ", 1277204076Spjd strerror(hio->hio_errors[ncomp])); 1278211979Spjd remote_close(res, ncomp); 1279204076Spjd /* 1280204076Spjd * Take request back from the receive queue and move 1281204076Spjd * it immediately to the done queue. 1282204076Spjd */ 1283204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1284204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1285204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1286204076Spjd goto done_queue; 1287204076Spjd } 1288204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1289204076Spjd nv_free(nv); 1290204076Spjd if (wakeup) 1291204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1292204076Spjd continue; 1293204076Spjddone_queue: 1294204076Spjd nv_free(nv); 1295204076Spjd if (ISSYNCREQ(hio)) { 1296204076Spjd if (!refcount_release(&hio->hio_countdown)) 1297204076Spjd continue; 1298204076Spjd mtx_lock(&sync_lock); 1299204076Spjd SYNCREQDONE(hio); 1300204076Spjd mtx_unlock(&sync_lock); 1301204076Spjd cv_signal(&sync_cond); 1302204076Spjd continue; 1303204076Spjd } 1304204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1305204076Spjd mtx_lock(&res->hr_amp_lock); 1306204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1307204076Spjd ggio->gctl_length)) { 1308204076Spjd (void)hast_activemap_flush(res); 1309204076Spjd } 1310204076Spjd mtx_unlock(&res->hr_amp_lock); 1311204076Spjd } 1312204076Spjd if (!refcount_release(&hio->hio_countdown)) 1313204076Spjd continue; 1314204076Spjd pjdlog_debug(2, 1315204076Spjd "remote_send: (%p) Moving request to the done queue.", 1316204076Spjd hio); 1317204076Spjd QUEUE_INSERT2(hio, done); 1318204076Spjd } 1319204076Spjd /* NOTREACHED */ 1320204076Spjd return (NULL); 1321204076Spjd} 1322204076Spjd 1323204076Spjd/* 1324204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1325204076Spjd * thread. 1326204076Spjd */ 1327204076Spjdstatic void * 1328204076Spjdremote_recv_thread(void *arg) 1329204076Spjd{ 1330204076Spjd struct hast_resource *res = arg; 1331204076Spjd struct g_gate_ctl_io *ggio; 1332204076Spjd struct hio *hio; 1333204076Spjd struct nv *nv; 1334204076Spjd unsigned int ncomp; 1335204076Spjd uint64_t seq; 1336204076Spjd int error; 1337204076Spjd 1338204076Spjd /* Remote component is 1 for now. */ 1339204076Spjd ncomp = 1; 1340204076Spjd 1341204076Spjd for (;;) { 1342204076Spjd /* Wait until there is anything to receive. */ 1343204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1344204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1345204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1346204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1347204076Spjd &hio_recv_list_lock[ncomp]); 1348204076Spjd } 1349204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1350204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1351204076Spjd if (!ISCONNECTED(res, ncomp)) { 1352204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1353204076Spjd /* 1354204076Spjd * Connection is dead, so move all pending requests to 1355204076Spjd * the done queue (one-by-one). 1356204076Spjd */ 1357204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1358204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1359204076Spjd assert(hio != NULL); 1360204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1361204076Spjd hio_next[ncomp]); 1362204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1363204076Spjd goto done_queue; 1364204076Spjd } 1365204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1366204076Spjd pjdlog_errno(LOG_ERR, 1367204076Spjd "Unable to receive reply header"); 1368204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1369204076Spjd remote_close(res, ncomp); 1370204076Spjd continue; 1371204076Spjd } 1372204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1373204076Spjd seq = nv_get_uint64(nv, "seq"); 1374204076Spjd if (seq == 0) { 1375204076Spjd pjdlog_error("Header contains no 'seq' field."); 1376204076Spjd nv_free(nv); 1377204076Spjd continue; 1378204076Spjd } 1379204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1380204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1381204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1382204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1383204076Spjd hio_next[ncomp]); 1384204076Spjd break; 1385204076Spjd } 1386204076Spjd } 1387204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1388204076Spjd if (hio == NULL) { 1389204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1390204076Spjd (uintmax_t)seq); 1391204076Spjd nv_free(nv); 1392204076Spjd continue; 1393204076Spjd } 1394204076Spjd error = nv_get_int16(nv, "error"); 1395204076Spjd if (error != 0) { 1396204076Spjd /* Request failed on remote side. */ 1397204076Spjd hio->hio_errors[ncomp] = 0; 1398204076Spjd nv_free(nv); 1399204076Spjd goto done_queue; 1400204076Spjd } 1401204076Spjd ggio = &hio->hio_ggio; 1402204076Spjd switch (ggio->gctl_cmd) { 1403204076Spjd case BIO_READ: 1404204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1405204076Spjd if (!ISCONNECTED(res, ncomp)) { 1406204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1407204076Spjd nv_free(nv); 1408204076Spjd goto done_queue; 1409204076Spjd } 1410204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1411204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1412204076Spjd hio->hio_errors[ncomp] = errno; 1413204076Spjd pjdlog_errno(LOG_ERR, 1414204076Spjd "Unable to receive reply data"); 1415204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1416204076Spjd nv_free(nv); 1417204076Spjd remote_close(res, ncomp); 1418204076Spjd goto done_queue; 1419204076Spjd } 1420204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1421204076Spjd break; 1422204076Spjd case BIO_WRITE: 1423204076Spjd case BIO_DELETE: 1424204076Spjd case BIO_FLUSH: 1425204076Spjd break; 1426204076Spjd default: 1427204076Spjd assert(!"invalid condition"); 1428204076Spjd abort(); 1429204076Spjd } 1430204076Spjd hio->hio_errors[ncomp] = 0; 1431204076Spjd nv_free(nv); 1432204076Spjddone_queue: 1433204076Spjd if (refcount_release(&hio->hio_countdown)) { 1434204076Spjd if (ISSYNCREQ(hio)) { 1435204076Spjd mtx_lock(&sync_lock); 1436204076Spjd SYNCREQDONE(hio); 1437204076Spjd mtx_unlock(&sync_lock); 1438204076Spjd cv_signal(&sync_cond); 1439204076Spjd } else { 1440204076Spjd pjdlog_debug(2, 1441204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1442204076Spjd hio); 1443204076Spjd QUEUE_INSERT2(hio, done); 1444204076Spjd } 1445204076Spjd } 1446204076Spjd } 1447204076Spjd /* NOTREACHED */ 1448204076Spjd return (NULL); 1449204076Spjd} 1450204076Spjd 1451204076Spjd/* 1452204076Spjd * Thread sends answer to the kernel. 1453204076Spjd */ 1454204076Spjdstatic void * 1455204076Spjdggate_send_thread(void *arg) 1456204076Spjd{ 1457204076Spjd struct hast_resource *res = arg; 1458204076Spjd struct g_gate_ctl_io *ggio; 1459204076Spjd struct hio *hio; 1460204076Spjd unsigned int ii, ncomp, ncomps; 1461204076Spjd 1462204076Spjd ncomps = HAST_NCOMPONENTS; 1463204076Spjd 1464204076Spjd for (;;) { 1465204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1466204076Spjd QUEUE_TAKE2(hio, done); 1467204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1468204076Spjd ggio = &hio->hio_ggio; 1469204076Spjd for (ii = 0; ii < ncomps; ii++) { 1470204076Spjd if (hio->hio_errors[ii] == 0) { 1471204076Spjd /* 1472204076Spjd * One successful request is enough to declare 1473204076Spjd * success. 1474204076Spjd */ 1475204076Spjd ggio->gctl_error = 0; 1476204076Spjd break; 1477204076Spjd } 1478204076Spjd } 1479204076Spjd if (ii == ncomps) { 1480204076Spjd /* 1481204076Spjd * None of the requests were successful. 1482204076Spjd * Use first error. 1483204076Spjd */ 1484204076Spjd ggio->gctl_error = hio->hio_errors[0]; 1485204076Spjd } 1486204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1487204076Spjd mtx_lock(&res->hr_amp_lock); 1488204076Spjd activemap_write_complete(res->hr_amp, 1489204076Spjd ggio->gctl_offset, ggio->gctl_length); 1490204076Spjd mtx_unlock(&res->hr_amp_lock); 1491204076Spjd } 1492204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1493204076Spjd /* 1494204076Spjd * Unlock range we locked. 1495204076Spjd */ 1496204076Spjd mtx_lock(&range_lock); 1497204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1498204076Spjd ggio->gctl_length); 1499204076Spjd if (range_sync_wait) 1500204076Spjd cv_signal(&range_sync_cond); 1501204076Spjd mtx_unlock(&range_lock); 1502204076Spjd /* 1503204076Spjd * Bump local count if this is first write after 1504204076Spjd * connection failure with remote node. 1505204076Spjd */ 1506204076Spjd ncomp = 1; 1507204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1508204076Spjd if (!ISCONNECTED(res, ncomp)) { 1509204076Spjd mtx_lock(&metadata_lock); 1510204076Spjd if (res->hr_primary_localcnt == 1511204076Spjd res->hr_secondary_remotecnt) { 1512204076Spjd res->hr_primary_localcnt++; 1513204076Spjd pjdlog_debug(1, 1514204076Spjd "Increasing localcnt to %ju.", 1515204076Spjd (uintmax_t)res->hr_primary_localcnt); 1516204076Spjd (void)metadata_write(res); 1517204076Spjd } 1518204076Spjd mtx_unlock(&metadata_lock); 1519204076Spjd } 1520204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1521204076Spjd } 1522204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1523204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1524204076Spjd pjdlog_debug(2, 1525204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1526204076Spjd QUEUE_INSERT2(hio, free); 1527204076Spjd } 1528204076Spjd /* NOTREACHED */ 1529204076Spjd return (NULL); 1530204076Spjd} 1531204076Spjd 1532204076Spjd/* 1533204076Spjd * Thread synchronize local and remote components. 1534204076Spjd */ 1535204076Spjdstatic void * 1536204076Spjdsync_thread(void *arg __unused) 1537204076Spjd{ 1538204076Spjd struct hast_resource *res = arg; 1539204076Spjd struct hio *hio; 1540204076Spjd struct g_gate_ctl_io *ggio; 1541204076Spjd unsigned int ii, ncomp, ncomps; 1542204076Spjd off_t offset, length, synced; 1543204076Spjd bool dorewind; 1544204076Spjd int syncext; 1545204076Spjd 1546204076Spjd ncomps = HAST_NCOMPONENTS; 1547204076Spjd dorewind = true; 1548211897Spjd synced = 0; 1549211897Spjd offset = -1; 1550204076Spjd 1551204076Spjd for (;;) { 1552204076Spjd mtx_lock(&sync_lock); 1553211897Spjd if (offset >= 0 && !sync_inprogress) { 1554211879Spjd pjdlog_info("Synchronization interrupted. " 1555211879Spjd "%jd bytes synchronized so far.", 1556211879Spjd (intmax_t)synced); 1557212038Spjd event_send(res, EVENT_SYNCINTR); 1558211879Spjd } 1559204076Spjd while (!sync_inprogress) { 1560204076Spjd dorewind = true; 1561204076Spjd synced = 0; 1562204076Spjd cv_wait(&sync_cond, &sync_lock); 1563204076Spjd } 1564204076Spjd mtx_unlock(&sync_lock); 1565204076Spjd /* 1566204076Spjd * Obtain offset at which we should synchronize. 1567204076Spjd * Rewind synchronization if needed. 1568204076Spjd */ 1569204076Spjd mtx_lock(&res->hr_amp_lock); 1570204076Spjd if (dorewind) 1571204076Spjd activemap_sync_rewind(res->hr_amp); 1572204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1573204076Spjd if (syncext != -1) { 1574204076Spjd /* 1575204076Spjd * We synchronized entire syncext extent, we can mark 1576204076Spjd * it as clean now. 1577204076Spjd */ 1578204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1579204076Spjd (void)hast_activemap_flush(res); 1580204076Spjd } 1581204076Spjd mtx_unlock(&res->hr_amp_lock); 1582204076Spjd if (dorewind) { 1583204076Spjd dorewind = false; 1584204076Spjd if (offset < 0) 1585204076Spjd pjdlog_info("Nodes are in sync."); 1586204076Spjd else { 1587204076Spjd pjdlog_info("Synchronization started. %ju bytes to go.", 1588204076Spjd (uintmax_t)(res->hr_extentsize * 1589204076Spjd activemap_ndirty(res->hr_amp))); 1590212038Spjd event_send(res, EVENT_SYNCSTART); 1591204076Spjd } 1592204076Spjd } 1593204076Spjd if (offset < 0) { 1594211878Spjd sync_stop(); 1595204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1596204076Spjd /* 1597204076Spjd * Synchronization complete, make both localcnt and 1598204076Spjd * remotecnt equal. 1599204076Spjd */ 1600204076Spjd ncomp = 1; 1601204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1602204076Spjd if (ISCONNECTED(res, ncomp)) { 1603204076Spjd if (synced > 0) { 1604204076Spjd pjdlog_info("Synchronization complete. " 1605204076Spjd "%jd bytes synchronized.", 1606204076Spjd (intmax_t)synced); 1607212038Spjd event_send(res, EVENT_SYNCDONE); 1608204076Spjd } 1609204076Spjd mtx_lock(&metadata_lock); 1610204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1611204076Spjd res->hr_primary_localcnt = 1612204076Spjd res->hr_secondary_localcnt; 1613204076Spjd res->hr_primary_remotecnt = 1614204076Spjd res->hr_secondary_remotecnt; 1615204076Spjd pjdlog_debug(1, 1616204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1617204076Spjd (uintmax_t)res->hr_primary_localcnt, 1618204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1619204076Spjd (void)metadata_write(res); 1620204076Spjd mtx_unlock(&metadata_lock); 1621204076Spjd } 1622204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1623204076Spjd continue; 1624204076Spjd } 1625204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1626204076Spjd QUEUE_TAKE2(hio, free); 1627204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1628204076Spjd /* 1629204076Spjd * Lock the range we are going to synchronize. We don't want 1630204076Spjd * race where someone writes between our read and write. 1631204076Spjd */ 1632204076Spjd for (;;) { 1633204076Spjd mtx_lock(&range_lock); 1634204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1635204076Spjd pjdlog_debug(2, 1636204076Spjd "sync: Range offset=%jd length=%jd locked.", 1637204076Spjd (intmax_t)offset, (intmax_t)length); 1638204076Spjd range_sync_wait = true; 1639204076Spjd cv_wait(&range_sync_cond, &range_lock); 1640204076Spjd range_sync_wait = false; 1641204076Spjd mtx_unlock(&range_lock); 1642204076Spjd continue; 1643204076Spjd } 1644204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1645204076Spjd mtx_unlock(&range_lock); 1646204076Spjd pjdlog_debug(2, 1647204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1648204076Spjd (intmax_t)offset, (intmax_t)length); 1649204076Spjd sleep(1); 1650204076Spjd continue; 1651204076Spjd } 1652204076Spjd mtx_unlock(&range_lock); 1653204076Spjd break; 1654204076Spjd } 1655204076Spjd /* 1656204076Spjd * First read the data from synchronization source. 1657204076Spjd */ 1658204076Spjd SYNCREQ(hio); 1659204076Spjd ggio = &hio->hio_ggio; 1660204076Spjd ggio->gctl_cmd = BIO_READ; 1661204076Spjd ggio->gctl_offset = offset; 1662204076Spjd ggio->gctl_length = length; 1663204076Spjd ggio->gctl_error = 0; 1664204076Spjd for (ii = 0; ii < ncomps; ii++) 1665204076Spjd hio->hio_errors[ii] = EINVAL; 1666204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1667204076Spjd hio); 1668204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1669204076Spjd hio); 1670204076Spjd mtx_lock(&metadata_lock); 1671204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1672204076Spjd /* 1673204076Spjd * This range is up-to-date on local component, 1674204076Spjd * so handle request locally. 1675204076Spjd */ 1676204076Spjd /* Local component is 0 for now. */ 1677204076Spjd ncomp = 0; 1678204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1679204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1680204076Spjd /* 1681204076Spjd * This range is out-of-date on local component, 1682204076Spjd * so send request to the remote node. 1683204076Spjd */ 1684204076Spjd /* Remote component is 1 for now. */ 1685204076Spjd ncomp = 1; 1686204076Spjd } 1687204076Spjd mtx_unlock(&metadata_lock); 1688204076Spjd refcount_init(&hio->hio_countdown, 1); 1689204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1690204076Spjd 1691204076Spjd /* 1692204076Spjd * Let's wait for READ to finish. 1693204076Spjd */ 1694204076Spjd mtx_lock(&sync_lock); 1695204076Spjd while (!ISSYNCREQDONE(hio)) 1696204076Spjd cv_wait(&sync_cond, &sync_lock); 1697204076Spjd mtx_unlock(&sync_lock); 1698204076Spjd 1699204076Spjd if (hio->hio_errors[ncomp] != 0) { 1700204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1701204076Spjd strerror(hio->hio_errors[ncomp])); 1702204076Spjd goto free_queue; 1703204076Spjd } 1704204076Spjd 1705204076Spjd /* 1706204076Spjd * We read the data from synchronization source, now write it 1707204076Spjd * to synchronization target. 1708204076Spjd */ 1709204076Spjd SYNCREQ(hio); 1710204076Spjd ggio->gctl_cmd = BIO_WRITE; 1711204076Spjd for (ii = 0; ii < ncomps; ii++) 1712204076Spjd hio->hio_errors[ii] = EINVAL; 1713204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1714204076Spjd hio); 1715204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1716204076Spjd hio); 1717204076Spjd mtx_lock(&metadata_lock); 1718204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1719204076Spjd /* 1720204076Spjd * This range is up-to-date on local component, 1721204076Spjd * so we update remote component. 1722204076Spjd */ 1723204076Spjd /* Remote component is 1 for now. */ 1724204076Spjd ncomp = 1; 1725204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1726204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1727204076Spjd /* 1728204076Spjd * This range is out-of-date on local component, 1729204076Spjd * so we update it. 1730204076Spjd */ 1731204076Spjd /* Local component is 0 for now. */ 1732204076Spjd ncomp = 0; 1733204076Spjd } 1734204076Spjd mtx_unlock(&metadata_lock); 1735204076Spjd 1736204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1737204076Spjd hio); 1738204076Spjd refcount_init(&hio->hio_countdown, 1); 1739204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1740204076Spjd 1741204076Spjd /* 1742204076Spjd * Let's wait for WRITE to finish. 1743204076Spjd */ 1744204076Spjd mtx_lock(&sync_lock); 1745204076Spjd while (!ISSYNCREQDONE(hio)) 1746204076Spjd cv_wait(&sync_cond, &sync_lock); 1747204076Spjd mtx_unlock(&sync_lock); 1748204076Spjd 1749204076Spjd if (hio->hio_errors[ncomp] != 0) { 1750204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1751204076Spjd strerror(hio->hio_errors[ncomp])); 1752204076Spjd goto free_queue; 1753204076Spjd } 1754211880Spjd 1755211880Spjd synced += length; 1756204076Spjdfree_queue: 1757204076Spjd mtx_lock(&range_lock); 1758204076Spjd rangelock_del(range_sync, offset, length); 1759204076Spjd if (range_regular_wait) 1760204076Spjd cv_signal(&range_regular_cond); 1761204076Spjd mtx_unlock(&range_lock); 1762204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1763204076Spjd hio); 1764204076Spjd QUEUE_INSERT2(hio, free); 1765204076Spjd } 1766204076Spjd /* NOTREACHED */ 1767204076Spjd return (NULL); 1768204076Spjd} 1769204076Spjd 1770204076Spjdstatic void 1771210886Spjdconfig_reload(void) 1772210886Spjd{ 1773210886Spjd struct hastd_config *newcfg; 1774210886Spjd struct hast_resource *res; 1775210886Spjd unsigned int ii, ncomps; 1776210886Spjd int modified; 1777210886Spjd 1778210886Spjd pjdlog_info("Reloading configuration..."); 1779210886Spjd 1780210886Spjd ncomps = HAST_NCOMPONENTS; 1781210886Spjd 1782210886Spjd newcfg = yy_config_parse(cfgpath, false); 1783210886Spjd if (newcfg == NULL) 1784210886Spjd goto failed; 1785210886Spjd 1786210886Spjd TAILQ_FOREACH(res, &newcfg->hc_resources, hr_next) { 1787210886Spjd if (strcmp(res->hr_name, gres->hr_name) == 0) 1788210886Spjd break; 1789210886Spjd } 1790210886Spjd /* 1791210886Spjd * If resource was removed from the configuration file, resource 1792210886Spjd * name, provider name or path to local component was modified we 1793210886Spjd * shouldn't be here. This means that someone modified configuration 1794210886Spjd * file and send SIGHUP to us instead of main hastd process. 1795210886Spjd * Log advice and ignore the signal. 1796210886Spjd */ 1797210886Spjd if (res == NULL || strcmp(gres->hr_name, res->hr_name) != 0 || 1798210886Spjd strcmp(gres->hr_provname, res->hr_provname) != 0 || 1799210886Spjd strcmp(gres->hr_localpath, res->hr_localpath) != 0) { 1800210886Spjd pjdlog_warning("To reload configuration send SIGHUP to the main hastd process (pid %u).", 1801210886Spjd (unsigned int)getppid()); 1802210886Spjd goto failed; 1803210886Spjd } 1804210886Spjd 1805210886Spjd#define MODIFIED_REMOTEADDR 0x1 1806210886Spjd#define MODIFIED_REPLICATION 0x2 1807210886Spjd#define MODIFIED_TIMEOUT 0x4 1808211886Spjd#define MODIFIED_EXEC 0x8 1809210886Spjd modified = 0; 1810210886Spjd if (strcmp(gres->hr_remoteaddr, res->hr_remoteaddr) != 0) { 1811210886Spjd /* 1812210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1813210886Spjd * We want remote_close() to log disconnect from the old 1814210886Spjd * addresses, not from the new ones. 1815210886Spjd */ 1816210886Spjd modified |= MODIFIED_REMOTEADDR; 1817210886Spjd } 1818210886Spjd if (gres->hr_replication != res->hr_replication) { 1819210886Spjd gres->hr_replication = res->hr_replication; 1820210886Spjd modified |= MODIFIED_REPLICATION; 1821210886Spjd } 1822210886Spjd if (gres->hr_timeout != res->hr_timeout) { 1823210886Spjd gres->hr_timeout = res->hr_timeout; 1824210886Spjd modified |= MODIFIED_TIMEOUT; 1825210886Spjd } 1826211886Spjd if (strcmp(gres->hr_exec, res->hr_exec) != 0) { 1827211886Spjd strlcpy(gres->hr_exec, res->hr_exec, sizeof(gres->hr_exec)); 1828211886Spjd modified |= MODIFIED_EXEC; 1829211886Spjd } 1830210886Spjd /* 1831210886Spjd * If only timeout was modified we only need to change it without 1832210886Spjd * reconnecting. 1833210886Spjd */ 1834210886Spjd if (modified == MODIFIED_TIMEOUT) { 1835210886Spjd for (ii = 0; ii < ncomps; ii++) { 1836210886Spjd if (!ISREMOTE(ii)) 1837210886Spjd continue; 1838210886Spjd rw_rlock(&hio_remote_lock[ii]); 1839210886Spjd if (!ISCONNECTED(gres, ii)) { 1840210886Spjd rw_unlock(&hio_remote_lock[ii]); 1841210886Spjd continue; 1842210886Spjd } 1843210886Spjd rw_unlock(&hio_remote_lock[ii]); 1844210886Spjd if (proto_timeout(gres->hr_remotein, 1845210886Spjd gres->hr_timeout) < 0) { 1846210886Spjd pjdlog_errno(LOG_WARNING, 1847210886Spjd "Unable to set connection timeout"); 1848210886Spjd } 1849210886Spjd if (proto_timeout(gres->hr_remoteout, 1850210886Spjd gres->hr_timeout) < 0) { 1851210886Spjd pjdlog_errno(LOG_WARNING, 1852210886Spjd "Unable to set connection timeout"); 1853210886Spjd } 1854210886Spjd } 1855211886Spjd } else if ((modified & 1856211886Spjd (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1857210886Spjd for (ii = 0; ii < ncomps; ii++) { 1858210886Spjd if (!ISREMOTE(ii)) 1859210886Spjd continue; 1860210886Spjd remote_close(gres, ii); 1861210886Spjd } 1862210886Spjd if (modified & MODIFIED_REMOTEADDR) { 1863210886Spjd strlcpy(gres->hr_remoteaddr, res->hr_remoteaddr, 1864210886Spjd sizeof(gres->hr_remoteaddr)); 1865210886Spjd } 1866210886Spjd } 1867210886Spjd#undef MODIFIED_REMOTEADDR 1868210886Spjd#undef MODIFIED_REPLICATION 1869210886Spjd#undef MODIFIED_TIMEOUT 1870211886Spjd#undef MODIFIED_EXEC 1871210886Spjd 1872210886Spjd pjdlog_info("Configuration reloaded successfully."); 1873210886Spjd return; 1874210886Spjdfailed: 1875210886Spjd if (newcfg != NULL) { 1876210886Spjd if (newcfg->hc_controlconn != NULL) 1877210886Spjd proto_close(newcfg->hc_controlconn); 1878210886Spjd if (newcfg->hc_listenconn != NULL) 1879210886Spjd proto_close(newcfg->hc_listenconn); 1880210886Spjd yy_config_free(newcfg); 1881210886Spjd } 1882210886Spjd pjdlog_warning("Configuration not reloaded."); 1883210886Spjd} 1884210886Spjd 1885211882Spjdstatic void 1886211882Spjdkeepalive_send(struct hast_resource *res, unsigned int ncomp) 1887211882Spjd{ 1888211882Spjd struct nv *nv; 1889211882Spjd 1890211882Spjd nv = nv_alloc(); 1891211882Spjd nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1892211882Spjd if (nv_error(nv) != 0) { 1893211882Spjd nv_free(nv); 1894211882Spjd pjdlog_debug(1, 1895211882Spjd "keepalive_send: Unable to prepare header to send."); 1896211882Spjd return; 1897211882Spjd } 1898211882Spjd if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1899211882Spjd pjdlog_common(LOG_DEBUG, 1, errno, 1900211882Spjd "keepalive_send: Unable to send request"); 1901211882Spjd nv_free(nv); 1902211882Spjd rw_unlock(&hio_remote_lock[ncomp]); 1903211882Spjd remote_close(res, ncomp); 1904211882Spjd rw_rlock(&hio_remote_lock[ncomp]); 1905211882Spjd return; 1906211882Spjd } 1907211882Spjd nv_free(nv); 1908211882Spjd pjdlog_debug(2, "keepalive_send: Request sent."); 1909211882Spjd} 1910211882Spjd 1911211981Spjdstatic void 1912211981Spjdguard_one(struct hast_resource *res, unsigned int ncomp) 1913211981Spjd{ 1914211981Spjd struct proto_conn *in, *out; 1915211981Spjd 1916211981Spjd if (!ISREMOTE(ncomp)) 1917211981Spjd return; 1918211981Spjd 1919211981Spjd rw_rlock(&hio_remote_lock[ncomp]); 1920211981Spjd 1921211981Spjd if (!real_remote(res)) { 1922211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1923211981Spjd return; 1924211981Spjd } 1925211981Spjd 1926211981Spjd if (ISCONNECTED(res, ncomp)) { 1927211981Spjd assert(res->hr_remotein != NULL); 1928211981Spjd assert(res->hr_remoteout != NULL); 1929211981Spjd keepalive_send(res, ncomp); 1930211981Spjd } 1931211981Spjd 1932211981Spjd if (ISCONNECTED(res, ncomp)) { 1933211981Spjd assert(res->hr_remotein != NULL); 1934211981Spjd assert(res->hr_remoteout != NULL); 1935211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1936211981Spjd pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 1937211981Spjd res->hr_remoteaddr); 1938211981Spjd return; 1939211981Spjd } 1940211981Spjd 1941211981Spjd assert(res->hr_remotein == NULL); 1942211981Spjd assert(res->hr_remoteout == NULL); 1943211981Spjd /* 1944211981Spjd * Upgrade the lock. It doesn't have to be atomic as no other thread 1945211981Spjd * can change connection status from disconnected to connected. 1946211981Spjd */ 1947211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1948211981Spjd pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 1949211981Spjd res->hr_remoteaddr); 1950211981Spjd in = out = NULL; 1951211981Spjd if (init_remote(res, &in, &out)) { 1952211981Spjd rw_wlock(&hio_remote_lock[ncomp]); 1953211981Spjd assert(res->hr_remotein == NULL); 1954211981Spjd assert(res->hr_remoteout == NULL); 1955211981Spjd assert(in != NULL && out != NULL); 1956211981Spjd res->hr_remotein = in; 1957211981Spjd res->hr_remoteout = out; 1958211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1959211981Spjd pjdlog_info("Successfully reconnected to %s.", 1960211981Spjd res->hr_remoteaddr); 1961211981Spjd sync_start(); 1962211981Spjd } else { 1963211981Spjd /* Both connections should be NULL. */ 1964211981Spjd assert(res->hr_remotein == NULL); 1965211981Spjd assert(res->hr_remoteout == NULL); 1966211981Spjd assert(in == NULL && out == NULL); 1967211981Spjd pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 1968211981Spjd res->hr_remoteaddr); 1969211981Spjd } 1970211981Spjd} 1971211981Spjd 1972204076Spjd/* 1973204076Spjd * Thread guards remote connections and reconnects when needed, handles 1974204076Spjd * signals, etc. 1975204076Spjd */ 1976204076Spjdstatic void * 1977204076Spjdguard_thread(void *arg) 1978204076Spjd{ 1979204076Spjd struct hast_resource *res = arg; 1980204076Spjd unsigned int ii, ncomps; 1981211982Spjd struct timespec timeout; 1982211981Spjd time_t lastcheck, now; 1983211982Spjd sigset_t mask; 1984211982Spjd int signo; 1985204076Spjd 1986204076Spjd ncomps = HAST_NCOMPONENTS; 1987211981Spjd lastcheck = time(NULL); 1988204076Spjd 1989211982Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1990211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1991211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1992211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1993211982Spjd 1994211982Spjd timeout.tv_nsec = 0; 1995211982Spjd signo = -1; 1996211982Spjd 1997204076Spjd for (;;) { 1998211982Spjd switch (signo) { 1999211982Spjd case SIGHUP: 2000211982Spjd config_reload(); 2001211982Spjd break; 2002211982Spjd case SIGINT: 2003211982Spjd case SIGTERM: 2004211982Spjd sigexit_received = true; 2005204076Spjd primary_exitx(EX_OK, 2006204076Spjd "Termination signal received, exiting."); 2007211982Spjd break; 2008211982Spjd default: 2009211982Spjd break; 2010204076Spjd } 2011211882Spjd 2012204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 2013211981Spjd now = time(NULL); 2014211982Spjd if (lastcheck + RETRY_SLEEP <= now) { 2015211982Spjd for (ii = 0; ii < ncomps; ii++) 2016211981Spjd guard_one(res, ii); 2017211981Spjd lastcheck = now; 2018204076Spjd } 2019211982Spjd timeout.tv_sec = RETRY_SLEEP; 2020211982Spjd signo = sigtimedwait(&mask, NULL, &timeout); 2021204076Spjd } 2022204076Spjd /* NOTREACHED */ 2023204076Spjd return (NULL); 2024204076Spjd} 2025