primary.c revision 218214
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 218214 2011-02-03 10:37:44Z 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 <err.h> 44204076Spjd#include <errno.h> 45204076Spjd#include <fcntl.h> 46204076Spjd#include <libgeom.h> 47204076Spjd#include <pthread.h> 48211982Spjd#include <signal.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" 60212038Spjd#include "event.h" 61204076Spjd#include "hast.h" 62204076Spjd#include "hast_proto.h" 63204076Spjd#include "hastd.h" 64211886Spjd#include "hooks.h" 65204076Spjd#include "metadata.h" 66204076Spjd#include "proto.h" 67204076Spjd#include "pjdlog.h" 68204076Spjd#include "subr.h" 69204076Spjd#include "synch.h" 70204076Spjd 71210886Spjd/* The is only one remote component for now. */ 72210886Spjd#define ISREMOTE(no) ((no) == 1) 73210886Spjd 74204076Spjdstruct hio { 75204076Spjd /* 76204076Spjd * Number of components we are still waiting for. 77204076Spjd * When this field goes to 0, we can send the request back to the 78204076Spjd * kernel. Each component has to decrease this counter by one 79204076Spjd * even on failure. 80204076Spjd */ 81204076Spjd unsigned int hio_countdown; 82204076Spjd /* 83204076Spjd * Each component has a place to store its own error. 84204076Spjd * Once the request is handled by all components we can decide if the 85204076Spjd * request overall is successful or not. 86204076Spjd */ 87204076Spjd int *hio_errors; 88204076Spjd /* 89204076Spjd * Structure used to comunicate with GEOM Gate class. 90204076Spjd */ 91204076Spjd struct g_gate_ctl_io hio_ggio; 92204076Spjd TAILQ_ENTRY(hio) *hio_next; 93204076Spjd}; 94204076Spjd#define hio_free_next hio_next[0] 95204076Spjd#define hio_done_next hio_next[0] 96204076Spjd 97204076Spjd/* 98204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 99204076Spjd * until some in-progress requests are freed. 100204076Spjd */ 101204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 102204076Spjdstatic pthread_mutex_t hio_free_list_lock; 103204076Spjdstatic pthread_cond_t hio_free_list_cond; 104204076Spjd/* 105204076Spjd * There is one send list for every component. One requests is placed on all 106204076Spjd * send lists - each component gets the same request, but each component is 107204076Spjd * responsible for managing his own send list. 108204076Spjd */ 109204076Spjdstatic TAILQ_HEAD(, hio) *hio_send_list; 110204076Spjdstatic pthread_mutex_t *hio_send_list_lock; 111204076Spjdstatic pthread_cond_t *hio_send_list_cond; 112204076Spjd/* 113204076Spjd * There is one recv list for every component, although local components don't 114204076Spjd * use recv lists as local requests are done synchronously. 115204076Spjd */ 116204076Spjdstatic TAILQ_HEAD(, hio) *hio_recv_list; 117204076Spjdstatic pthread_mutex_t *hio_recv_list_lock; 118204076Spjdstatic pthread_cond_t *hio_recv_list_cond; 119204076Spjd/* 120204076Spjd * Request is placed on done list by the slowest component (the one that 121204076Spjd * decreased hio_countdown from 1 to 0). 122204076Spjd */ 123204076Spjdstatic TAILQ_HEAD(, hio) hio_done_list; 124204076Spjdstatic pthread_mutex_t hio_done_list_lock; 125204076Spjdstatic pthread_cond_t hio_done_list_cond; 126204076Spjd/* 127204076Spjd * Structure below are for interaction with sync thread. 128204076Spjd */ 129204076Spjdstatic bool sync_inprogress; 130204076Spjdstatic pthread_mutex_t sync_lock; 131204076Spjdstatic pthread_cond_t sync_cond; 132204076Spjd/* 133204076Spjd * The lock below allows to synchornize access to remote connections. 134204076Spjd */ 135204076Spjdstatic pthread_rwlock_t *hio_remote_lock; 136204076Spjd 137204076Spjd/* 138204076Spjd * Lock to synchronize metadata updates. Also synchronize access to 139204076Spjd * hr_primary_localcnt and hr_primary_remotecnt fields. 140204076Spjd */ 141204076Spjdstatic pthread_mutex_t metadata_lock; 142204076Spjd 143204076Spjd/* 144204076Spjd * Maximum number of outstanding I/O requests. 145204076Spjd */ 146204076Spjd#define HAST_HIO_MAX 256 147204076Spjd/* 148204076Spjd * Number of components. At this point there are only two components: local 149204076Spjd * and remote, but in the future it might be possible to use multiple local 150204076Spjd * and remote components. 151204076Spjd */ 152204076Spjd#define HAST_NCOMPONENTS 2 153204076Spjd/* 154211982Spjd * Number of seconds to sleep between reconnect retries or keepalive packets. 155204076Spjd */ 156211982Spjd#define RETRY_SLEEP 10 157204076Spjd 158204076Spjd#define ISCONNECTED(res, no) \ 159204076Spjd ((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL) 160204076Spjd 161204076Spjd#define QUEUE_INSERT1(hio, name, ncomp) do { \ 162204076Spjd bool _wakeup; \ 163204076Spjd \ 164204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 165204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]); \ 166204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio), \ 167204076Spjd hio_next[(ncomp)]); \ 168204076Spjd mtx_unlock(&hio_##name##_list_lock[ncomp]); \ 169204076Spjd if (_wakeup) \ 170204076Spjd cv_signal(&hio_##name##_list_cond[(ncomp)]); \ 171204076Spjd} while (0) 172204076Spjd#define QUEUE_INSERT2(hio, name) do { \ 173204076Spjd bool _wakeup; \ 174204076Spjd \ 175204076Spjd mtx_lock(&hio_##name##_list_lock); \ 176204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 177204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\ 178204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 179204076Spjd if (_wakeup) \ 180204076Spjd cv_signal(&hio_##name##_list_cond); \ 181204076Spjd} while (0) 182214692Spjd#define QUEUE_TAKE1(hio, name, ncomp, timeout) do { \ 183214692Spjd bool _last; \ 184214692Spjd \ 185204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 186214692Spjd _last = false; \ 187214692Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL && !_last) { \ 188214692Spjd cv_timedwait(&hio_##name##_list_cond[(ncomp)], \ 189214692Spjd &hio_##name##_list_lock[(ncomp)], (timeout)); \ 190214692Spjd if ((timeout) != 0) \ 191214692Spjd _last = true; \ 192204076Spjd } \ 193214692Spjd if (hio != NULL) { \ 194214692Spjd TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio), \ 195214692Spjd hio_next[(ncomp)]); \ 196214692Spjd } \ 197204076Spjd mtx_unlock(&hio_##name##_list_lock[(ncomp)]); \ 198204076Spjd} while (0) 199204076Spjd#define QUEUE_TAKE2(hio, name) do { \ 200204076Spjd mtx_lock(&hio_##name##_list_lock); \ 201204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 202204076Spjd cv_wait(&hio_##name##_list_cond, \ 203204076Spjd &hio_##name##_list_lock); \ 204204076Spjd } \ 205204076Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next); \ 206204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 207204076Spjd} while (0) 208204076Spjd 209209183Spjd#define SYNCREQ(hio) do { \ 210209183Spjd (hio)->hio_ggio.gctl_unit = -1; \ 211209183Spjd (hio)->hio_ggio.gctl_seq = 1; \ 212209183Spjd} while (0) 213204076Spjd#define ISSYNCREQ(hio) ((hio)->hio_ggio.gctl_unit == -1) 214204076Spjd#define SYNCREQDONE(hio) do { (hio)->hio_ggio.gctl_unit = -2; } while (0) 215204076Spjd#define ISSYNCREQDONE(hio) ((hio)->hio_ggio.gctl_unit == -2) 216204076Spjd 217204076Spjdstatic struct hast_resource *gres; 218204076Spjd 219204076Spjdstatic pthread_mutex_t range_lock; 220204076Spjdstatic struct rangelocks *range_regular; 221204076Spjdstatic bool range_regular_wait; 222204076Spjdstatic pthread_cond_t range_regular_cond; 223204076Spjdstatic struct rangelocks *range_sync; 224204076Spjdstatic bool range_sync_wait; 225204076Spjdstatic pthread_cond_t range_sync_cond; 226204076Spjd 227204076Spjdstatic void *ggate_recv_thread(void *arg); 228204076Spjdstatic void *local_send_thread(void *arg); 229204076Spjdstatic void *remote_send_thread(void *arg); 230204076Spjdstatic void *remote_recv_thread(void *arg); 231204076Spjdstatic void *ggate_send_thread(void *arg); 232204076Spjdstatic void *sync_thread(void *arg); 233204076Spjdstatic void *guard_thread(void *arg); 234204076Spjd 235211982Spjdstatic void 236204076Spjdcleanup(struct hast_resource *res) 237204076Spjd{ 238204076Spjd int rerrno; 239204076Spjd 240204076Spjd /* Remember errno. */ 241204076Spjd rerrno = errno; 242204076Spjd 243204076Spjd /* Destroy ggate provider if we created one. */ 244204076Spjd if (res->hr_ggateunit >= 0) { 245204076Spjd struct g_gate_ctl_destroy ggiod; 246204076Spjd 247213533Spjd bzero(&ggiod, sizeof(ggiod)); 248204076Spjd ggiod.gctl_version = G_GATE_VERSION; 249204076Spjd ggiod.gctl_unit = res->hr_ggateunit; 250204076Spjd ggiod.gctl_force = 1; 251204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 252213531Spjd pjdlog_errno(LOG_WARNING, 253213531Spjd "Unable to destroy hast/%s device", 254204076Spjd res->hr_provname); 255204076Spjd } 256204076Spjd res->hr_ggateunit = -1; 257204076Spjd } 258204076Spjd 259204076Spjd /* Restore errno. */ 260204076Spjd errno = rerrno; 261204076Spjd} 262204076Spjd 263212899Spjdstatic __dead2 void 264204076Spjdprimary_exit(int exitcode, const char *fmt, ...) 265204076Spjd{ 266204076Spjd va_list ap; 267204076Spjd 268218138Spjd PJDLOG_ASSERT(exitcode != EX_OK); 269204076Spjd va_start(ap, fmt); 270204076Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 271204076Spjd va_end(ap); 272204076Spjd cleanup(gres); 273204076Spjd exit(exitcode); 274204076Spjd} 275204076Spjd 276212899Spjdstatic __dead2 void 277204076Spjdprimary_exitx(int exitcode, const char *fmt, ...) 278204076Spjd{ 279204076Spjd va_list ap; 280204076Spjd 281204076Spjd va_start(ap, fmt); 282204076Spjd pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 283204076Spjd va_end(ap); 284204076Spjd cleanup(gres); 285204076Spjd exit(exitcode); 286204076Spjd} 287204076Spjd 288204076Spjdstatic int 289204076Spjdhast_activemap_flush(struct hast_resource *res) 290204076Spjd{ 291204076Spjd const unsigned char *buf; 292204076Spjd size_t size; 293204076Spjd 294204076Spjd buf = activemap_bitmap(res->hr_amp, &size); 295218138Spjd PJDLOG_ASSERT(buf != NULL); 296218138Spjd PJDLOG_ASSERT((size % res->hr_local_sectorsize) == 0); 297204076Spjd if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 298204076Spjd (ssize_t)size) { 299204076Spjd KEEP_ERRNO(pjdlog_errno(LOG_ERR, 300204076Spjd "Unable to flush activemap to disk")); 301204076Spjd return (-1); 302204076Spjd } 303204076Spjd return (0); 304204076Spjd} 305204076Spjd 306210881Spjdstatic bool 307210881Spjdreal_remote(const struct hast_resource *res) 308210881Spjd{ 309210881Spjd 310210881Spjd return (strcmp(res->hr_remoteaddr, "none") != 0); 311210881Spjd} 312210881Spjd 313204076Spjdstatic void 314204076Spjdinit_environment(struct hast_resource *res __unused) 315204076Spjd{ 316204076Spjd struct hio *hio; 317204076Spjd unsigned int ii, ncomps; 318204076Spjd 319204076Spjd /* 320204076Spjd * In the future it might be per-resource value. 321204076Spjd */ 322204076Spjd ncomps = HAST_NCOMPONENTS; 323204076Spjd 324204076Spjd /* 325204076Spjd * Allocate memory needed by lists. 326204076Spjd */ 327204076Spjd hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 328204076Spjd if (hio_send_list == NULL) { 329204076Spjd primary_exitx(EX_TEMPFAIL, 330204076Spjd "Unable to allocate %zu bytes of memory for send lists.", 331204076Spjd sizeof(hio_send_list[0]) * ncomps); 332204076Spjd } 333204076Spjd hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 334204076Spjd if (hio_send_list_lock == NULL) { 335204076Spjd primary_exitx(EX_TEMPFAIL, 336204076Spjd "Unable to allocate %zu bytes of memory for send list locks.", 337204076Spjd sizeof(hio_send_list_lock[0]) * ncomps); 338204076Spjd } 339204076Spjd hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 340204076Spjd if (hio_send_list_cond == NULL) { 341204076Spjd primary_exitx(EX_TEMPFAIL, 342204076Spjd "Unable to allocate %zu bytes of memory for send list condition variables.", 343204076Spjd sizeof(hio_send_list_cond[0]) * ncomps); 344204076Spjd } 345204076Spjd hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 346204076Spjd if (hio_recv_list == NULL) { 347204076Spjd primary_exitx(EX_TEMPFAIL, 348204076Spjd "Unable to allocate %zu bytes of memory for recv lists.", 349204076Spjd sizeof(hio_recv_list[0]) * ncomps); 350204076Spjd } 351204076Spjd hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 352204076Spjd if (hio_recv_list_lock == NULL) { 353204076Spjd primary_exitx(EX_TEMPFAIL, 354204076Spjd "Unable to allocate %zu bytes of memory for recv list locks.", 355204076Spjd sizeof(hio_recv_list_lock[0]) * ncomps); 356204076Spjd } 357204076Spjd hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 358204076Spjd if (hio_recv_list_cond == NULL) { 359204076Spjd primary_exitx(EX_TEMPFAIL, 360204076Spjd "Unable to allocate %zu bytes of memory for recv list condition variables.", 361204076Spjd sizeof(hio_recv_list_cond[0]) * ncomps); 362204076Spjd } 363204076Spjd hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 364204076Spjd if (hio_remote_lock == NULL) { 365204076Spjd primary_exitx(EX_TEMPFAIL, 366204076Spjd "Unable to allocate %zu bytes of memory for remote connections locks.", 367204076Spjd sizeof(hio_remote_lock[0]) * ncomps); 368204076Spjd } 369204076Spjd 370204076Spjd /* 371204076Spjd * Initialize lists, their locks and theirs condition variables. 372204076Spjd */ 373204076Spjd TAILQ_INIT(&hio_free_list); 374204076Spjd mtx_init(&hio_free_list_lock); 375204076Spjd cv_init(&hio_free_list_cond); 376204076Spjd for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 377204076Spjd TAILQ_INIT(&hio_send_list[ii]); 378204076Spjd mtx_init(&hio_send_list_lock[ii]); 379204076Spjd cv_init(&hio_send_list_cond[ii]); 380204076Spjd TAILQ_INIT(&hio_recv_list[ii]); 381204076Spjd mtx_init(&hio_recv_list_lock[ii]); 382204076Spjd cv_init(&hio_recv_list_cond[ii]); 383204076Spjd rw_init(&hio_remote_lock[ii]); 384204076Spjd } 385204076Spjd TAILQ_INIT(&hio_done_list); 386204076Spjd mtx_init(&hio_done_list_lock); 387204076Spjd cv_init(&hio_done_list_cond); 388204076Spjd mtx_init(&metadata_lock); 389204076Spjd 390204076Spjd /* 391204076Spjd * Allocate requests pool and initialize requests. 392204076Spjd */ 393204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 394204076Spjd hio = malloc(sizeof(*hio)); 395204076Spjd if (hio == NULL) { 396204076Spjd primary_exitx(EX_TEMPFAIL, 397204076Spjd "Unable to allocate %zu bytes of memory for hio request.", 398204076Spjd sizeof(*hio)); 399204076Spjd } 400204076Spjd hio->hio_countdown = 0; 401204076Spjd hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 402204076Spjd if (hio->hio_errors == NULL) { 403204076Spjd primary_exitx(EX_TEMPFAIL, 404204076Spjd "Unable allocate %zu bytes of memory for hio errors.", 405204076Spjd sizeof(hio->hio_errors[0]) * ncomps); 406204076Spjd } 407204076Spjd hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 408204076Spjd if (hio->hio_next == NULL) { 409204076Spjd primary_exitx(EX_TEMPFAIL, 410204076Spjd "Unable allocate %zu bytes of memory for hio_next field.", 411204076Spjd sizeof(hio->hio_next[0]) * ncomps); 412204076Spjd } 413204076Spjd hio->hio_ggio.gctl_version = G_GATE_VERSION; 414204076Spjd hio->hio_ggio.gctl_data = malloc(MAXPHYS); 415204076Spjd if (hio->hio_ggio.gctl_data == NULL) { 416204076Spjd primary_exitx(EX_TEMPFAIL, 417204076Spjd "Unable to allocate %zu bytes of memory for gctl_data.", 418204076Spjd MAXPHYS); 419204076Spjd } 420204076Spjd hio->hio_ggio.gctl_length = MAXPHYS; 421204076Spjd hio->hio_ggio.gctl_error = 0; 422204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 423204076Spjd } 424204076Spjd} 425204076Spjd 426214284Spjdstatic bool 427214284Spjdinit_resuid(struct hast_resource *res) 428214284Spjd{ 429214284Spjd 430214284Spjd mtx_lock(&metadata_lock); 431214284Spjd if (res->hr_resuid != 0) { 432214284Spjd mtx_unlock(&metadata_lock); 433214284Spjd return (false); 434214284Spjd } else { 435214284Spjd /* Initialize unique resource identifier. */ 436214284Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 437214284Spjd mtx_unlock(&metadata_lock); 438214284Spjd if (metadata_write(res) < 0) 439214284Spjd exit(EX_NOINPUT); 440214284Spjd return (true); 441214284Spjd } 442214284Spjd} 443214284Spjd 444204076Spjdstatic void 445204076Spjdinit_local(struct hast_resource *res) 446204076Spjd{ 447204076Spjd unsigned char *buf; 448204076Spjd size_t mapsize; 449204076Spjd 450204076Spjd if (metadata_read(res, true) < 0) 451204076Spjd exit(EX_NOINPUT); 452204076Spjd mtx_init(&res->hr_amp_lock); 453204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 454204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 455204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 456204076Spjd } 457204076Spjd mtx_init(&range_lock); 458204076Spjd cv_init(&range_regular_cond); 459204076Spjd if (rangelock_init(&range_regular) < 0) 460204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 461204076Spjd cv_init(&range_sync_cond); 462204076Spjd if (rangelock_init(&range_sync) < 0) 463204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 464204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 465204076Spjd buf = calloc(1, mapsize); 466204076Spjd if (buf == NULL) { 467204076Spjd primary_exitx(EX_TEMPFAIL, 468204076Spjd "Unable to allocate buffer for activemap."); 469204076Spjd } 470204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 471204076Spjd (ssize_t)mapsize) { 472204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 473204076Spjd } 474204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 475209181Spjd free(buf); 476204076Spjd if (res->hr_resuid != 0) 477204076Spjd return; 478204076Spjd /* 479214284Spjd * We're using provider for the first time. Initialize local and remote 480214284Spjd * counters. We don't initialize resuid here, as we want to do it just 481214284Spjd * in time. The reason for this is that we want to inform secondary 482214284Spjd * that there were no writes yet, so there is no need to synchronize 483214284Spjd * anything. 484204076Spjd */ 485204076Spjd res->hr_primary_localcnt = 1; 486204076Spjd res->hr_primary_remotecnt = 0; 487204076Spjd if (metadata_write(res) < 0) 488204076Spjd exit(EX_NOINPUT); 489204076Spjd} 490204076Spjd 491205738Spjdstatic bool 492205738Spjdinit_remote(struct hast_resource *res, struct proto_conn **inp, 493205738Spjd struct proto_conn **outp) 494204076Spjd{ 495205738Spjd struct proto_conn *in, *out; 496204076Spjd struct nv *nvout, *nvin; 497204076Spjd const unsigned char *token; 498204076Spjd unsigned char *map; 499204076Spjd const char *errmsg; 500204076Spjd int32_t extentsize; 501204076Spjd int64_t datasize; 502204076Spjd uint32_t mapsize; 503204076Spjd size_t size; 504204076Spjd 505218138Spjd PJDLOG_ASSERT((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 506218138Spjd PJDLOG_ASSERT(real_remote(res)); 507205738Spjd 508205738Spjd in = out = NULL; 509211983Spjd errmsg = NULL; 510205738Spjd 511204076Spjd /* Prepare outgoing connection with remote node. */ 512205738Spjd if (proto_client(res->hr_remoteaddr, &out) < 0) { 513215331Spjd primary_exit(EX_TEMPFAIL, 514215331Spjd "Unable to create outgoing connection to %s", 515204076Spjd res->hr_remoteaddr); 516204076Spjd } 517204076Spjd /* Try to connect, but accept failure. */ 518218192Spjd if (proto_connect(out, HAST_TIMEOUT) < 0) { 519204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 520204076Spjd res->hr_remoteaddr); 521204076Spjd goto close; 522204076Spjd } 523207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 524207371Spjd if (proto_timeout(out, res->hr_timeout) < 0) 525207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 526204076Spjd /* 527204076Spjd * First handshake step. 528204076Spjd * Setup outgoing connection with remote node. 529204076Spjd */ 530204076Spjd nvout = nv_alloc(); 531204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 532204076Spjd if (nv_error(nvout) != 0) { 533204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 534204076Spjd "Unable to allocate header for connection with %s", 535204076Spjd res->hr_remoteaddr); 536204076Spjd nv_free(nvout); 537204076Spjd goto close; 538204076Spjd } 539205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 540204076Spjd pjdlog_errno(LOG_WARNING, 541204076Spjd "Unable to send handshake header to %s", 542204076Spjd res->hr_remoteaddr); 543204076Spjd nv_free(nvout); 544204076Spjd goto close; 545204076Spjd } 546204076Spjd nv_free(nvout); 547205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 548204076Spjd pjdlog_errno(LOG_WARNING, 549204076Spjd "Unable to receive handshake header from %s", 550204076Spjd res->hr_remoteaddr); 551204076Spjd goto close; 552204076Spjd } 553204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 554204076Spjd if (errmsg != NULL) { 555204076Spjd pjdlog_warning("%s", errmsg); 556204076Spjd nv_free(nvin); 557204076Spjd goto close; 558204076Spjd } 559204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 560204076Spjd if (token == NULL) { 561204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 562204076Spjd res->hr_remoteaddr); 563204076Spjd nv_free(nvin); 564204076Spjd goto close; 565204076Spjd } 566204076Spjd if (size != sizeof(res->hr_token)) { 567204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 568204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 569204076Spjd nv_free(nvin); 570204076Spjd goto close; 571204076Spjd } 572204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 573204076Spjd nv_free(nvin); 574204076Spjd 575204076Spjd /* 576204076Spjd * Second handshake step. 577204076Spjd * Setup incoming connection with remote node. 578204076Spjd */ 579205738Spjd if (proto_client(res->hr_remoteaddr, &in) < 0) { 580215331Spjd primary_exit(EX_TEMPFAIL, 581215331Spjd "Unable to create incoming connection to %s", 582204076Spjd res->hr_remoteaddr); 583204076Spjd } 584204076Spjd /* Try to connect, but accept failure. */ 585218192Spjd if (proto_connect(in, HAST_TIMEOUT) < 0) { 586204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 587204076Spjd res->hr_remoteaddr); 588204076Spjd goto close; 589204076Spjd } 590207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 591207371Spjd if (proto_timeout(in, res->hr_timeout) < 0) 592207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 593204076Spjd nvout = nv_alloc(); 594204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 595204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 596204076Spjd "token"); 597214284Spjd if (res->hr_resuid == 0) { 598214284Spjd /* 599214284Spjd * The resuid field was not yet initialized. 600214284Spjd * Because we do synchronization inside init_resuid(), it is 601214284Spjd * possible that someone already initialized it, the function 602214284Spjd * will return false then, but if we successfully initialized 603214284Spjd * it, we will get true. True means that there were no writes 604214284Spjd * to this resource yet and we want to inform secondary that 605214284Spjd * synchronization is not needed by sending "virgin" argument. 606214284Spjd */ 607214284Spjd if (init_resuid(res)) 608214284Spjd nv_add_int8(nvout, 1, "virgin"); 609214284Spjd } 610204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 611204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 612204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 613204076Spjd if (nv_error(nvout) != 0) { 614204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 615204076Spjd "Unable to allocate header for connection with %s", 616204076Spjd res->hr_remoteaddr); 617204076Spjd nv_free(nvout); 618204076Spjd goto close; 619204076Spjd } 620205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 621204076Spjd pjdlog_errno(LOG_WARNING, 622204076Spjd "Unable to send handshake header to %s", 623204076Spjd res->hr_remoteaddr); 624204076Spjd nv_free(nvout); 625204076Spjd goto close; 626204076Spjd } 627204076Spjd nv_free(nvout); 628205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 629204076Spjd pjdlog_errno(LOG_WARNING, 630204076Spjd "Unable to receive handshake header from %s", 631204076Spjd res->hr_remoteaddr); 632204076Spjd goto close; 633204076Spjd } 634204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 635204076Spjd if (errmsg != NULL) { 636204076Spjd pjdlog_warning("%s", errmsg); 637204076Spjd nv_free(nvin); 638204076Spjd goto close; 639204076Spjd } 640204076Spjd datasize = nv_get_int64(nvin, "datasize"); 641204076Spjd if (datasize != res->hr_datasize) { 642204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 643204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 644204076Spjd nv_free(nvin); 645204076Spjd goto close; 646204076Spjd } 647204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 648204076Spjd if (extentsize != res->hr_extentsize) { 649204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 650204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 651204076Spjd nv_free(nvin); 652204076Spjd goto close; 653204076Spjd } 654204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 655204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 656204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 657204076Spjd map = NULL; 658204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 659204076Spjd if (mapsize > 0) { 660204076Spjd map = malloc(mapsize); 661204076Spjd if (map == NULL) { 662204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 663204076Spjd (uintmax_t)mapsize); 664204076Spjd nv_free(nvin); 665204076Spjd goto close; 666204076Spjd } 667204076Spjd /* 668204076Spjd * Remote node have some dirty extents on its own, lets 669204076Spjd * download its activemap. 670204076Spjd */ 671205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 672204076Spjd mapsize) < 0) { 673204076Spjd pjdlog_errno(LOG_ERR, 674204076Spjd "Unable to receive remote activemap"); 675204076Spjd nv_free(nvin); 676204076Spjd free(map); 677204076Spjd goto close; 678204076Spjd } 679204076Spjd /* 680204076Spjd * Merge local and remote bitmaps. 681204076Spjd */ 682204076Spjd activemap_merge(res->hr_amp, map, mapsize); 683204076Spjd free(map); 684204076Spjd /* 685204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 686204076Spjd * disk before we start to synchronize. 687204076Spjd */ 688204076Spjd (void)hast_activemap_flush(res); 689204076Spjd } 690214274Spjd nv_free(nvin); 691204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 692205738Spjd if (inp != NULL && outp != NULL) { 693205738Spjd *inp = in; 694205738Spjd *outp = out; 695205738Spjd } else { 696205738Spjd res->hr_remotein = in; 697205738Spjd res->hr_remoteout = out; 698205738Spjd } 699212038Spjd event_send(res, EVENT_CONNECT); 700205738Spjd return (true); 701205738Spjdclose: 702211983Spjd if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0) 703212038Spjd event_send(res, EVENT_SPLITBRAIN); 704205738Spjd proto_close(out); 705205738Spjd if (in != NULL) 706205738Spjd proto_close(in); 707205738Spjd return (false); 708205738Spjd} 709205738Spjd 710205738Spjdstatic void 711205738Spjdsync_start(void) 712205738Spjd{ 713205738Spjd 714204076Spjd mtx_lock(&sync_lock); 715204076Spjd sync_inprogress = true; 716204076Spjd mtx_unlock(&sync_lock); 717204076Spjd cv_signal(&sync_cond); 718204076Spjd} 719204076Spjd 720204076Spjdstatic void 721211878Spjdsync_stop(void) 722211878Spjd{ 723211878Spjd 724211878Spjd mtx_lock(&sync_lock); 725211878Spjd if (sync_inprogress) 726211878Spjd sync_inprogress = false; 727211878Spjd mtx_unlock(&sync_lock); 728211878Spjd} 729211878Spjd 730211878Spjdstatic void 731204076Spjdinit_ggate(struct hast_resource *res) 732204076Spjd{ 733204076Spjd struct g_gate_ctl_create ggiocreate; 734204076Spjd struct g_gate_ctl_cancel ggiocancel; 735204076Spjd 736204076Spjd /* 737204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 738204076Spjd */ 739204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 740204076Spjd if (res->hr_ggatefd < 0) 741204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 742204076Spjd /* 743204076Spjd * Create provider before trying to connect, as connection failure 744204076Spjd * is not critical, but may take some time. 745204076Spjd */ 746213533Spjd bzero(&ggiocreate, sizeof(ggiocreate)); 747204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 748204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 749204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 750204076Spjd ggiocreate.gctl_flags = 0; 751206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 752204076Spjd ggiocreate.gctl_timeout = 0; 753204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 754204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 755204076Spjd res->hr_provname); 756204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 757204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 758204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 759204076Spjd return; 760204076Spjd } 761204076Spjd if (errno != EEXIST) { 762204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 763204076Spjd res->hr_provname); 764204076Spjd } 765204076Spjd pjdlog_debug(1, 766204076Spjd "Device hast/%s already exists, we will try to take it over.", 767204076Spjd res->hr_provname); 768204076Spjd /* 769204076Spjd * If we received EEXIST, we assume that the process who created the 770204076Spjd * provider died and didn't clean up. In that case we will start from 771204076Spjd * where he left of. 772204076Spjd */ 773213533Spjd bzero(&ggiocancel, sizeof(ggiocancel)); 774204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 775204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 776204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 777204076Spjd res->hr_provname); 778204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 779204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 780204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 781204076Spjd return; 782204076Spjd } 783204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 784204076Spjd res->hr_provname); 785204076Spjd} 786204076Spjd 787204076Spjdvoid 788204076Spjdhastd_primary(struct hast_resource *res) 789204076Spjd{ 790204076Spjd pthread_t td; 791204076Spjd pid_t pid; 792218043Spjd int error, mode; 793204076Spjd 794204076Spjd /* 795204076Spjd * Create communication channel between parent and child. 796204076Spjd */ 797204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 798218042Spjd /* TODO: There's no need for this to be fatal error. */ 799204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 800212034Spjd pjdlog_exit(EX_OSERR, 801204076Spjd "Unable to create control sockets between parent and child"); 802204076Spjd } 803212038Spjd /* 804212038Spjd * Create communication channel between child and parent. 805212038Spjd */ 806212038Spjd if (proto_client("socketpair://", &res->hr_event) < 0) { 807218042Spjd /* TODO: There's no need for this to be fatal error. */ 808212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 809212038Spjd pjdlog_exit(EX_OSERR, 810212038Spjd "Unable to create event sockets between child and parent"); 811212038Spjd } 812204076Spjd 813204076Spjd pid = fork(); 814204076Spjd if (pid < 0) { 815218042Spjd /* TODO: There's no need for this to be fatal error. */ 816204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 817212034Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to fork"); 818204076Spjd } 819204076Spjd 820204076Spjd if (pid > 0) { 821204076Spjd /* This is parent. */ 822212038Spjd /* Declare that we are receiver. */ 823212038Spjd proto_recv(res->hr_event, NULL, 0); 824218043Spjd /* Declare that we are sender. */ 825218043Spjd proto_send(res->hr_ctrl, NULL, 0); 826204076Spjd res->hr_workerpid = pid; 827204076Spjd return; 828204076Spjd } 829211977Spjd 830211984Spjd gres = res; 831218043Spjd mode = pjdlog_mode_get(); 832211984Spjd 833218043Spjd /* Declare that we are sender. */ 834218043Spjd proto_send(res->hr_event, NULL, 0); 835218043Spjd /* Declare that we are receiver. */ 836218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 837218043Spjd descriptors_cleanup(res); 838204076Spjd 839218045Spjd descriptors_assert(res, mode); 840218045Spjd 841218043Spjd pjdlog_init(mode); 842218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 843204076Spjd setproctitle("%s (primary)", res->hr_name); 844204076Spjd 845204076Spjd init_local(res); 846213007Spjd init_ggate(res); 847213007Spjd init_environment(res); 848217784Spjd 849218049Spjd if (drop_privs() != 0) { 850218049Spjd cleanup(res); 851218049Spjd exit(EX_CONFIG); 852218049Spjd } 853218214Spjd pjdlog_info("Privileges successfully dropped."); 854218049Spjd 855213007Spjd /* 856213530Spjd * Create the guard thread first, so we can handle signals from the 857213530Spjd * very begining. 858213530Spjd */ 859213530Spjd error = pthread_create(&td, NULL, guard_thread, res); 860218138Spjd PJDLOG_ASSERT(error == 0); 861213530Spjd /* 862213007Spjd * Create the control thread before sending any event to the parent, 863213007Spjd * as we can deadlock when parent sends control request to worker, 864213007Spjd * but worker has no control thread started yet, so parent waits. 865213007Spjd * In the meantime worker sends an event to the parent, but parent 866213007Spjd * is unable to handle the event, because it waits for control 867213007Spjd * request response. 868213007Spjd */ 869213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 870218138Spjd PJDLOG_ASSERT(error == 0); 871210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 872205738Spjd sync_start(); 873204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 874218138Spjd PJDLOG_ASSERT(error == 0); 875204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 876218138Spjd PJDLOG_ASSERT(error == 0); 877204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 878218138Spjd PJDLOG_ASSERT(error == 0); 879204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 880218138Spjd PJDLOG_ASSERT(error == 0); 881204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 882218138Spjd PJDLOG_ASSERT(error == 0); 883213530Spjd (void)sync_thread(res); 884204076Spjd} 885204076Spjd 886204076Spjdstatic void 887204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 888204076Spjd{ 889204076Spjd char msg[1024]; 890204076Spjd va_list ap; 891204076Spjd int len; 892204076Spjd 893204076Spjd va_start(ap, fmt); 894204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 895204076Spjd va_end(ap); 896204076Spjd if ((size_t)len < sizeof(msg)) { 897204076Spjd switch (ggio->gctl_cmd) { 898204076Spjd case BIO_READ: 899204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 900204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 901204076Spjd (uintmax_t)ggio->gctl_length); 902204076Spjd break; 903204076Spjd case BIO_DELETE: 904204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 905204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 906204076Spjd (uintmax_t)ggio->gctl_length); 907204076Spjd break; 908204076Spjd case BIO_FLUSH: 909204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 910204076Spjd break; 911204076Spjd case BIO_WRITE: 912204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 913204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 914204076Spjd (uintmax_t)ggio->gctl_length); 915204076Spjd break; 916204076Spjd default: 917204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 918204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 919204076Spjd break; 920204076Spjd } 921204076Spjd } 922204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 923204076Spjd} 924204076Spjd 925204076Spjdstatic void 926204076Spjdremote_close(struct hast_resource *res, int ncomp) 927204076Spjd{ 928204076Spjd 929204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 930204076Spjd /* 931204076Spjd * A race is possible between dropping rlock and acquiring wlock - 932204076Spjd * another thread can close connection in-between. 933204076Spjd */ 934204076Spjd if (!ISCONNECTED(res, ncomp)) { 935218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 936218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 937204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 938204076Spjd return; 939204076Spjd } 940204076Spjd 941218138Spjd PJDLOG_ASSERT(res->hr_remotein != NULL); 942218138Spjd PJDLOG_ASSERT(res->hr_remoteout != NULL); 943204076Spjd 944211881Spjd pjdlog_debug(2, "Closing incoming connection to %s.", 945204076Spjd res->hr_remoteaddr); 946204076Spjd proto_close(res->hr_remotein); 947204076Spjd res->hr_remotein = NULL; 948211881Spjd pjdlog_debug(2, "Closing outgoing connection to %s.", 949204076Spjd res->hr_remoteaddr); 950204076Spjd proto_close(res->hr_remoteout); 951204076Spjd res->hr_remoteout = NULL; 952204076Spjd 953204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 954204076Spjd 955211881Spjd pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 956211881Spjd 957204076Spjd /* 958204076Spjd * Stop synchronization if in-progress. 959204076Spjd */ 960211878Spjd sync_stop(); 961211984Spjd 962212038Spjd event_send(res, EVENT_DISCONNECT); 963204076Spjd} 964204076Spjd 965204076Spjd/* 966204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 967204076Spjd * appropriate threads: 968204076Spjd * WRITE - always goes to both local_send and remote_send threads 969204076Spjd * READ (when the block is up-to-date on local component) - 970204076Spjd * only local_send thread 971204076Spjd * READ (when the block isn't up-to-date on local component) - 972204076Spjd * only remote_send thread 973204076Spjd * DELETE - always goes to both local_send and remote_send threads 974204076Spjd * FLUSH - always goes to both local_send and remote_send threads 975204076Spjd */ 976204076Spjdstatic void * 977204076Spjdggate_recv_thread(void *arg) 978204076Spjd{ 979204076Spjd struct hast_resource *res = arg; 980204076Spjd struct g_gate_ctl_io *ggio; 981204076Spjd struct hio *hio; 982204076Spjd unsigned int ii, ncomp, ncomps; 983204076Spjd int error; 984204076Spjd 985204076Spjd ncomps = HAST_NCOMPONENTS; 986204076Spjd 987204076Spjd for (;;) { 988204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 989204076Spjd QUEUE_TAKE2(hio, free); 990204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 991204076Spjd ggio = &hio->hio_ggio; 992204076Spjd ggio->gctl_unit = res->hr_ggateunit; 993204076Spjd ggio->gctl_length = MAXPHYS; 994204076Spjd ggio->gctl_error = 0; 995204076Spjd pjdlog_debug(2, 996204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 997204076Spjd hio); 998204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 999204076Spjd if (sigexit_received) 1000204076Spjd pthread_exit(NULL); 1001204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 1002204076Spjd } 1003204076Spjd error = ggio->gctl_error; 1004204076Spjd switch (error) { 1005204076Spjd case 0: 1006204076Spjd break; 1007204076Spjd case ECANCELED: 1008204076Spjd /* Exit gracefully. */ 1009204076Spjd if (!sigexit_received) { 1010204076Spjd pjdlog_debug(2, 1011204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 1012204076Spjd hio); 1013204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 1014204076Spjd } 1015204076Spjd pthread_exit(NULL); 1016204076Spjd case ENOMEM: 1017204076Spjd /* 1018204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 1019204076Spjd * bytes - request can't be bigger than that. 1020204076Spjd */ 1021204076Spjd /* FALLTHROUGH */ 1022204076Spjd case ENXIO: 1023204076Spjd default: 1024204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 1025204076Spjd strerror(error)); 1026204076Spjd } 1027204076Spjd for (ii = 0; ii < ncomps; ii++) 1028204076Spjd hio->hio_errors[ii] = EINVAL; 1029204076Spjd reqlog(LOG_DEBUG, 2, ggio, 1030204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 1031204076Spjd hio); 1032204076Spjd /* 1033204076Spjd * Inform all components about new write request. 1034204076Spjd * For read request prefer local component unless the given 1035204076Spjd * range is out-of-date, then use remote component. 1036204076Spjd */ 1037204076Spjd switch (ggio->gctl_cmd) { 1038204076Spjd case BIO_READ: 1039204076Spjd pjdlog_debug(2, 1040204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 1041204076Spjd hio); 1042204076Spjd refcount_init(&hio->hio_countdown, 1); 1043204076Spjd mtx_lock(&metadata_lock); 1044204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 1045204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1046204076Spjd /* 1047204076Spjd * This range is up-to-date on local component, 1048204076Spjd * so handle request locally. 1049204076Spjd */ 1050204076Spjd /* Local component is 0 for now. */ 1051204076Spjd ncomp = 0; 1052204076Spjd } else /* if (res->hr_syncsrc == 1053204076Spjd HAST_SYNCSRC_SECONDARY) */ { 1054218138Spjd PJDLOG_ASSERT(res->hr_syncsrc == 1055204076Spjd HAST_SYNCSRC_SECONDARY); 1056204076Spjd /* 1057204076Spjd * This range is out-of-date on local component, 1058204076Spjd * so send request to the remote node. 1059204076Spjd */ 1060204076Spjd /* Remote component is 1 for now. */ 1061204076Spjd ncomp = 1; 1062204076Spjd } 1063204076Spjd mtx_unlock(&metadata_lock); 1064204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1065204076Spjd break; 1066204076Spjd case BIO_WRITE: 1067214284Spjd if (res->hr_resuid == 0) { 1068214284Spjd /* This is first write, initialize resuid. */ 1069214284Spjd (void)init_resuid(res); 1070214284Spjd } 1071204076Spjd for (;;) { 1072204076Spjd mtx_lock(&range_lock); 1073204076Spjd if (rangelock_islocked(range_sync, 1074204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1075204076Spjd pjdlog_debug(2, 1076204076Spjd "regular: Range offset=%jd length=%zu locked.", 1077204076Spjd (intmax_t)ggio->gctl_offset, 1078204076Spjd (size_t)ggio->gctl_length); 1079204076Spjd range_regular_wait = true; 1080204076Spjd cv_wait(&range_regular_cond, &range_lock); 1081204076Spjd range_regular_wait = false; 1082204076Spjd mtx_unlock(&range_lock); 1083204076Spjd continue; 1084204076Spjd } 1085204076Spjd if (rangelock_add(range_regular, 1086204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1087204076Spjd mtx_unlock(&range_lock); 1088204076Spjd pjdlog_debug(2, 1089204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1090204076Spjd (intmax_t)ggio->gctl_offset, 1091204076Spjd (size_t)ggio->gctl_length); 1092204076Spjd sleep(1); 1093204076Spjd continue; 1094204076Spjd } 1095204076Spjd mtx_unlock(&range_lock); 1096204076Spjd break; 1097204076Spjd } 1098204076Spjd mtx_lock(&res->hr_amp_lock); 1099204076Spjd if (activemap_write_start(res->hr_amp, 1100204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1101204076Spjd (void)hast_activemap_flush(res); 1102204076Spjd } 1103204076Spjd mtx_unlock(&res->hr_amp_lock); 1104204076Spjd /* FALLTHROUGH */ 1105204076Spjd case BIO_DELETE: 1106204076Spjd case BIO_FLUSH: 1107204076Spjd pjdlog_debug(2, 1108204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1109204076Spjd hio); 1110204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1111204076Spjd for (ii = 0; ii < ncomps; ii++) 1112204076Spjd QUEUE_INSERT1(hio, send, ii); 1113204076Spjd break; 1114204076Spjd } 1115204076Spjd } 1116204076Spjd /* NOTREACHED */ 1117204076Spjd return (NULL); 1118204076Spjd} 1119204076Spjd 1120204076Spjd/* 1121204076Spjd * Thread reads from or writes to local component. 1122204076Spjd * If local read fails, it redirects it to remote_send thread. 1123204076Spjd */ 1124204076Spjdstatic void * 1125204076Spjdlocal_send_thread(void *arg) 1126204076Spjd{ 1127204076Spjd struct hast_resource *res = arg; 1128204076Spjd struct g_gate_ctl_io *ggio; 1129204076Spjd struct hio *hio; 1130204076Spjd unsigned int ncomp, rncomp; 1131204076Spjd ssize_t ret; 1132204076Spjd 1133204076Spjd /* Local component is 0 for now. */ 1134204076Spjd ncomp = 0; 1135204076Spjd /* Remote component is 1 for now. */ 1136204076Spjd rncomp = 1; 1137204076Spjd 1138204076Spjd for (;;) { 1139204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1140214692Spjd QUEUE_TAKE1(hio, send, ncomp, 0); 1141204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1142204076Spjd ggio = &hio->hio_ggio; 1143204076Spjd switch (ggio->gctl_cmd) { 1144204076Spjd case BIO_READ: 1145204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1146204076Spjd ggio->gctl_length, 1147204076Spjd ggio->gctl_offset + res->hr_localoff); 1148204076Spjd if (ret == ggio->gctl_length) 1149204076Spjd hio->hio_errors[ncomp] = 0; 1150204076Spjd else { 1151204076Spjd /* 1152204076Spjd * If READ failed, try to read from remote node. 1153204076Spjd */ 1154216479Spjd if (ret < 0) { 1155216479Spjd reqlog(LOG_WARNING, 0, ggio, 1156216479Spjd "Local request failed (%s), trying remote node. ", 1157216479Spjd strerror(errno)); 1158216479Spjd } else if (ret != ggio->gctl_length) { 1159216479Spjd reqlog(LOG_WARNING, 0, ggio, 1160216479Spjd "Local request failed (%zd != %jd), trying remote node. ", 1161216494Spjd ret, (intmax_t)ggio->gctl_length); 1162216479Spjd } 1163204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1164204076Spjd continue; 1165204076Spjd } 1166204076Spjd break; 1167204076Spjd case BIO_WRITE: 1168204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1169204076Spjd ggio->gctl_length, 1170204076Spjd ggio->gctl_offset + res->hr_localoff); 1171216479Spjd if (ret < 0) { 1172204076Spjd hio->hio_errors[ncomp] = errno; 1173216479Spjd reqlog(LOG_WARNING, 0, ggio, 1174216479Spjd "Local request failed (%s): ", 1175216479Spjd strerror(errno)); 1176216479Spjd } else if (ret != ggio->gctl_length) { 1177204076Spjd hio->hio_errors[ncomp] = EIO; 1178216479Spjd reqlog(LOG_WARNING, 0, ggio, 1179216479Spjd "Local request failed (%zd != %jd): ", 1180216494Spjd ret, (intmax_t)ggio->gctl_length); 1181216479Spjd } else { 1182204076Spjd hio->hio_errors[ncomp] = 0; 1183216479Spjd } 1184204076Spjd break; 1185204076Spjd case BIO_DELETE: 1186204076Spjd ret = g_delete(res->hr_localfd, 1187204076Spjd ggio->gctl_offset + res->hr_localoff, 1188204076Spjd ggio->gctl_length); 1189216479Spjd if (ret < 0) { 1190204076Spjd hio->hio_errors[ncomp] = errno; 1191216479Spjd reqlog(LOG_WARNING, 0, ggio, 1192216479Spjd "Local request failed (%s): ", 1193216479Spjd strerror(errno)); 1194216479Spjd } else { 1195204076Spjd hio->hio_errors[ncomp] = 0; 1196216479Spjd } 1197204076Spjd break; 1198204076Spjd case BIO_FLUSH: 1199204076Spjd ret = g_flush(res->hr_localfd); 1200216479Spjd if (ret < 0) { 1201204076Spjd hio->hio_errors[ncomp] = errno; 1202216479Spjd reqlog(LOG_WARNING, 0, ggio, 1203216479Spjd "Local request failed (%s): ", 1204216479Spjd strerror(errno)); 1205216479Spjd } else { 1206204076Spjd hio->hio_errors[ncomp] = 0; 1207216479Spjd } 1208204076Spjd break; 1209204076Spjd } 1210204076Spjd if (refcount_release(&hio->hio_countdown)) { 1211204076Spjd if (ISSYNCREQ(hio)) { 1212204076Spjd mtx_lock(&sync_lock); 1213204076Spjd SYNCREQDONE(hio); 1214204076Spjd mtx_unlock(&sync_lock); 1215204076Spjd cv_signal(&sync_cond); 1216204076Spjd } else { 1217204076Spjd pjdlog_debug(2, 1218204076Spjd "local_send: (%p) Moving request to the done queue.", 1219204076Spjd hio); 1220204076Spjd QUEUE_INSERT2(hio, done); 1221204076Spjd } 1222204076Spjd } 1223204076Spjd } 1224204076Spjd /* NOTREACHED */ 1225204076Spjd return (NULL); 1226204076Spjd} 1227204076Spjd 1228214692Spjdstatic void 1229214692Spjdkeepalive_send(struct hast_resource *res, unsigned int ncomp) 1230214692Spjd{ 1231214692Spjd struct nv *nv; 1232214692Spjd 1233214692Spjd if (!ISCONNECTED(res, ncomp)) 1234214692Spjd return; 1235214692Spjd 1236218138Spjd PJDLOG_ASSERT(res->hr_remotein != NULL); 1237218138Spjd PJDLOG_ASSERT(res->hr_remoteout != NULL); 1238214692Spjd 1239214692Spjd nv = nv_alloc(); 1240214692Spjd nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1241214692Spjd if (nv_error(nv) != 0) { 1242214692Spjd nv_free(nv); 1243214692Spjd pjdlog_debug(1, 1244214692Spjd "keepalive_send: Unable to prepare header to send."); 1245214692Spjd return; 1246214692Spjd } 1247214692Spjd if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1248214692Spjd pjdlog_common(LOG_DEBUG, 1, errno, 1249214692Spjd "keepalive_send: Unable to send request"); 1250214692Spjd nv_free(nv); 1251214692Spjd rw_unlock(&hio_remote_lock[ncomp]); 1252214692Spjd remote_close(res, ncomp); 1253214692Spjd rw_rlock(&hio_remote_lock[ncomp]); 1254214692Spjd return; 1255214692Spjd } 1256214692Spjd nv_free(nv); 1257214692Spjd pjdlog_debug(2, "keepalive_send: Request sent."); 1258214692Spjd} 1259214692Spjd 1260204076Spjd/* 1261204076Spjd * Thread sends request to secondary node. 1262204076Spjd */ 1263204076Spjdstatic void * 1264204076Spjdremote_send_thread(void *arg) 1265204076Spjd{ 1266204076Spjd struct hast_resource *res = arg; 1267204076Spjd struct g_gate_ctl_io *ggio; 1268214692Spjd time_t lastcheck, now; 1269204076Spjd struct hio *hio; 1270204076Spjd struct nv *nv; 1271204076Spjd unsigned int ncomp; 1272204076Spjd bool wakeup; 1273204076Spjd uint64_t offset, length; 1274204076Spjd uint8_t cmd; 1275204076Spjd void *data; 1276204076Spjd 1277204076Spjd /* Remote component is 1 for now. */ 1278204076Spjd ncomp = 1; 1279214692Spjd lastcheck = time(NULL); 1280204076Spjd 1281204076Spjd for (;;) { 1282204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1283214692Spjd QUEUE_TAKE1(hio, send, ncomp, RETRY_SLEEP); 1284214692Spjd if (hio == NULL) { 1285214692Spjd now = time(NULL); 1286214692Spjd if (lastcheck + RETRY_SLEEP <= now) { 1287214692Spjd keepalive_send(res, ncomp); 1288214692Spjd lastcheck = now; 1289214692Spjd } 1290214692Spjd continue; 1291214692Spjd } 1292204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1293204076Spjd ggio = &hio->hio_ggio; 1294204076Spjd switch (ggio->gctl_cmd) { 1295204076Spjd case BIO_READ: 1296204076Spjd cmd = HIO_READ; 1297204076Spjd data = NULL; 1298204076Spjd offset = ggio->gctl_offset; 1299204076Spjd length = ggio->gctl_length; 1300204076Spjd break; 1301204076Spjd case BIO_WRITE: 1302204076Spjd cmd = HIO_WRITE; 1303204076Spjd data = ggio->gctl_data; 1304204076Spjd offset = ggio->gctl_offset; 1305204076Spjd length = ggio->gctl_length; 1306204076Spjd break; 1307204076Spjd case BIO_DELETE: 1308204076Spjd cmd = HIO_DELETE; 1309204076Spjd data = NULL; 1310204076Spjd offset = ggio->gctl_offset; 1311204076Spjd length = ggio->gctl_length; 1312204076Spjd break; 1313204076Spjd case BIO_FLUSH: 1314204076Spjd cmd = HIO_FLUSH; 1315204076Spjd data = NULL; 1316204076Spjd offset = 0; 1317204076Spjd length = 0; 1318204076Spjd break; 1319204076Spjd default: 1320218138Spjd PJDLOG_ASSERT(!"invalid condition"); 1321204076Spjd abort(); 1322204076Spjd } 1323204076Spjd nv = nv_alloc(); 1324204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1325204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1326204076Spjd nv_add_uint64(nv, offset, "offset"); 1327204076Spjd nv_add_uint64(nv, length, "length"); 1328204076Spjd if (nv_error(nv) != 0) { 1329204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1330204076Spjd pjdlog_debug(2, 1331204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1332204076Spjd hio); 1333204076Spjd reqlog(LOG_ERR, 0, ggio, 1334204076Spjd "Unable to prepare header to send (%s): ", 1335204076Spjd strerror(nv_error(nv))); 1336204076Spjd /* Move failed request immediately to the done queue. */ 1337204076Spjd goto done_queue; 1338204076Spjd } 1339204076Spjd pjdlog_debug(2, 1340204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1341204076Spjd hio); 1342204076Spjd /* 1343204076Spjd * Protect connection from disappearing. 1344204076Spjd */ 1345204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1346204076Spjd if (!ISCONNECTED(res, ncomp)) { 1347204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1348204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1349204076Spjd goto done_queue; 1350204076Spjd } 1351204076Spjd /* 1352204076Spjd * Move the request to recv queue before sending it, because 1353204076Spjd * in different order we can get reply before we move request 1354204076Spjd * to recv queue. 1355204076Spjd */ 1356204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1357204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1358204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1359204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1360204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1361204076Spjd data != NULL ? length : 0) < 0) { 1362204076Spjd hio->hio_errors[ncomp] = errno; 1363204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1364204076Spjd pjdlog_debug(2, 1365204076Spjd "remote_send: (%p) Unable to send request.", hio); 1366204076Spjd reqlog(LOG_ERR, 0, ggio, 1367204076Spjd "Unable to send request (%s): ", 1368204076Spjd strerror(hio->hio_errors[ncomp])); 1369211979Spjd remote_close(res, ncomp); 1370204076Spjd /* 1371204076Spjd * Take request back from the receive queue and move 1372204076Spjd * it immediately to the done queue. 1373204076Spjd */ 1374204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1375204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1376204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1377204076Spjd goto done_queue; 1378204076Spjd } 1379204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1380204076Spjd nv_free(nv); 1381204076Spjd if (wakeup) 1382204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1383204076Spjd continue; 1384204076Spjddone_queue: 1385204076Spjd nv_free(nv); 1386204076Spjd if (ISSYNCREQ(hio)) { 1387204076Spjd if (!refcount_release(&hio->hio_countdown)) 1388204076Spjd continue; 1389204076Spjd mtx_lock(&sync_lock); 1390204076Spjd SYNCREQDONE(hio); 1391204076Spjd mtx_unlock(&sync_lock); 1392204076Spjd cv_signal(&sync_cond); 1393204076Spjd continue; 1394204076Spjd } 1395204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1396204076Spjd mtx_lock(&res->hr_amp_lock); 1397204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1398204076Spjd ggio->gctl_length)) { 1399204076Spjd (void)hast_activemap_flush(res); 1400204076Spjd } 1401204076Spjd mtx_unlock(&res->hr_amp_lock); 1402204076Spjd } 1403204076Spjd if (!refcount_release(&hio->hio_countdown)) 1404204076Spjd continue; 1405204076Spjd pjdlog_debug(2, 1406204076Spjd "remote_send: (%p) Moving request to the done queue.", 1407204076Spjd hio); 1408204076Spjd QUEUE_INSERT2(hio, done); 1409204076Spjd } 1410204076Spjd /* NOTREACHED */ 1411204076Spjd return (NULL); 1412204076Spjd} 1413204076Spjd 1414204076Spjd/* 1415204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1416204076Spjd * thread. 1417204076Spjd */ 1418204076Spjdstatic void * 1419204076Spjdremote_recv_thread(void *arg) 1420204076Spjd{ 1421204076Spjd struct hast_resource *res = arg; 1422204076Spjd struct g_gate_ctl_io *ggio; 1423204076Spjd struct hio *hio; 1424204076Spjd struct nv *nv; 1425204076Spjd unsigned int ncomp; 1426204076Spjd uint64_t seq; 1427204076Spjd int error; 1428204076Spjd 1429204076Spjd /* Remote component is 1 for now. */ 1430204076Spjd ncomp = 1; 1431204076Spjd 1432204076Spjd for (;;) { 1433204076Spjd /* Wait until there is anything to receive. */ 1434204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1435204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1436204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1437204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1438204076Spjd &hio_recv_list_lock[ncomp]); 1439204076Spjd } 1440204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1441204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1442204076Spjd if (!ISCONNECTED(res, ncomp)) { 1443204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1444204076Spjd /* 1445204076Spjd * Connection is dead, so move all pending requests to 1446204076Spjd * the done queue (one-by-one). 1447204076Spjd */ 1448204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1449204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1450218138Spjd PJDLOG_ASSERT(hio != NULL); 1451204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1452204076Spjd hio_next[ncomp]); 1453204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1454204076Spjd goto done_queue; 1455204076Spjd } 1456204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1457204076Spjd pjdlog_errno(LOG_ERR, 1458204076Spjd "Unable to receive reply header"); 1459204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1460204076Spjd remote_close(res, ncomp); 1461204076Spjd continue; 1462204076Spjd } 1463204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1464204076Spjd seq = nv_get_uint64(nv, "seq"); 1465204076Spjd if (seq == 0) { 1466204076Spjd pjdlog_error("Header contains no 'seq' field."); 1467204076Spjd nv_free(nv); 1468204076Spjd continue; 1469204076Spjd } 1470204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1471204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1472204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1473204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1474204076Spjd hio_next[ncomp]); 1475204076Spjd break; 1476204076Spjd } 1477204076Spjd } 1478204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1479204076Spjd if (hio == NULL) { 1480204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1481204076Spjd (uintmax_t)seq); 1482204076Spjd nv_free(nv); 1483204076Spjd continue; 1484204076Spjd } 1485204076Spjd error = nv_get_int16(nv, "error"); 1486204076Spjd if (error != 0) { 1487204076Spjd /* Request failed on remote side. */ 1488216478Spjd hio->hio_errors[ncomp] = error; 1489216479Spjd reqlog(LOG_WARNING, 0, &hio->hio_ggio, 1490216479Spjd "Remote request failed (%s): ", strerror(error)); 1491204076Spjd nv_free(nv); 1492204076Spjd goto done_queue; 1493204076Spjd } 1494204076Spjd ggio = &hio->hio_ggio; 1495204076Spjd switch (ggio->gctl_cmd) { 1496204076Spjd case BIO_READ: 1497204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1498204076Spjd if (!ISCONNECTED(res, ncomp)) { 1499204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1500204076Spjd nv_free(nv); 1501204076Spjd goto done_queue; 1502204076Spjd } 1503204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1504204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1505204076Spjd hio->hio_errors[ncomp] = errno; 1506204076Spjd pjdlog_errno(LOG_ERR, 1507204076Spjd "Unable to receive reply data"); 1508204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1509204076Spjd nv_free(nv); 1510204076Spjd remote_close(res, ncomp); 1511204076Spjd goto done_queue; 1512204076Spjd } 1513204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1514204076Spjd break; 1515204076Spjd case BIO_WRITE: 1516204076Spjd case BIO_DELETE: 1517204076Spjd case BIO_FLUSH: 1518204076Spjd break; 1519204076Spjd default: 1520218138Spjd PJDLOG_ASSERT(!"invalid condition"); 1521204076Spjd abort(); 1522204076Spjd } 1523204076Spjd hio->hio_errors[ncomp] = 0; 1524204076Spjd nv_free(nv); 1525204076Spjddone_queue: 1526204076Spjd if (refcount_release(&hio->hio_countdown)) { 1527204076Spjd if (ISSYNCREQ(hio)) { 1528204076Spjd mtx_lock(&sync_lock); 1529204076Spjd SYNCREQDONE(hio); 1530204076Spjd mtx_unlock(&sync_lock); 1531204076Spjd cv_signal(&sync_cond); 1532204076Spjd } else { 1533204076Spjd pjdlog_debug(2, 1534204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1535204076Spjd hio); 1536204076Spjd QUEUE_INSERT2(hio, done); 1537204076Spjd } 1538204076Spjd } 1539204076Spjd } 1540204076Spjd /* NOTREACHED */ 1541204076Spjd return (NULL); 1542204076Spjd} 1543204076Spjd 1544204076Spjd/* 1545204076Spjd * Thread sends answer to the kernel. 1546204076Spjd */ 1547204076Spjdstatic void * 1548204076Spjdggate_send_thread(void *arg) 1549204076Spjd{ 1550204076Spjd struct hast_resource *res = arg; 1551204076Spjd struct g_gate_ctl_io *ggio; 1552204076Spjd struct hio *hio; 1553204076Spjd unsigned int ii, ncomp, ncomps; 1554204076Spjd 1555204076Spjd ncomps = HAST_NCOMPONENTS; 1556204076Spjd 1557204076Spjd for (;;) { 1558204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1559204076Spjd QUEUE_TAKE2(hio, done); 1560204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1561204076Spjd ggio = &hio->hio_ggio; 1562204076Spjd for (ii = 0; ii < ncomps; ii++) { 1563204076Spjd if (hio->hio_errors[ii] == 0) { 1564204076Spjd /* 1565204076Spjd * One successful request is enough to declare 1566204076Spjd * success. 1567204076Spjd */ 1568204076Spjd ggio->gctl_error = 0; 1569204076Spjd break; 1570204076Spjd } 1571204076Spjd } 1572204076Spjd if (ii == ncomps) { 1573204076Spjd /* 1574204076Spjd * None of the requests were successful. 1575204076Spjd * Use first error. 1576204076Spjd */ 1577204076Spjd ggio->gctl_error = hio->hio_errors[0]; 1578204076Spjd } 1579204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1580204076Spjd mtx_lock(&res->hr_amp_lock); 1581204076Spjd activemap_write_complete(res->hr_amp, 1582204076Spjd ggio->gctl_offset, ggio->gctl_length); 1583204076Spjd mtx_unlock(&res->hr_amp_lock); 1584204076Spjd } 1585204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1586204076Spjd /* 1587204076Spjd * Unlock range we locked. 1588204076Spjd */ 1589204076Spjd mtx_lock(&range_lock); 1590204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1591204076Spjd ggio->gctl_length); 1592204076Spjd if (range_sync_wait) 1593204076Spjd cv_signal(&range_sync_cond); 1594204076Spjd mtx_unlock(&range_lock); 1595204076Spjd /* 1596204076Spjd * Bump local count if this is first write after 1597204076Spjd * connection failure with remote node. 1598204076Spjd */ 1599204076Spjd ncomp = 1; 1600204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1601204076Spjd if (!ISCONNECTED(res, ncomp)) { 1602204076Spjd mtx_lock(&metadata_lock); 1603204076Spjd if (res->hr_primary_localcnt == 1604204076Spjd res->hr_secondary_remotecnt) { 1605204076Spjd res->hr_primary_localcnt++; 1606204076Spjd pjdlog_debug(1, 1607204076Spjd "Increasing localcnt to %ju.", 1608204076Spjd (uintmax_t)res->hr_primary_localcnt); 1609204076Spjd (void)metadata_write(res); 1610204076Spjd } 1611204076Spjd mtx_unlock(&metadata_lock); 1612204076Spjd } 1613204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1614204076Spjd } 1615204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1616204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1617204076Spjd pjdlog_debug(2, 1618204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1619204076Spjd QUEUE_INSERT2(hio, free); 1620204076Spjd } 1621204076Spjd /* NOTREACHED */ 1622204076Spjd return (NULL); 1623204076Spjd} 1624204076Spjd 1625204076Spjd/* 1626204076Spjd * Thread synchronize local and remote components. 1627204076Spjd */ 1628204076Spjdstatic void * 1629204076Spjdsync_thread(void *arg __unused) 1630204076Spjd{ 1631204076Spjd struct hast_resource *res = arg; 1632204076Spjd struct hio *hio; 1633204076Spjd struct g_gate_ctl_io *ggio; 1634204076Spjd unsigned int ii, ncomp, ncomps; 1635204076Spjd off_t offset, length, synced; 1636204076Spjd bool dorewind; 1637204076Spjd int syncext; 1638204076Spjd 1639204076Spjd ncomps = HAST_NCOMPONENTS; 1640204076Spjd dorewind = true; 1641211897Spjd synced = 0; 1642211897Spjd offset = -1; 1643204076Spjd 1644204076Spjd for (;;) { 1645204076Spjd mtx_lock(&sync_lock); 1646211897Spjd if (offset >= 0 && !sync_inprogress) { 1647211879Spjd pjdlog_info("Synchronization interrupted. " 1648211879Spjd "%jd bytes synchronized so far.", 1649211879Spjd (intmax_t)synced); 1650212038Spjd event_send(res, EVENT_SYNCINTR); 1651211879Spjd } 1652204076Spjd while (!sync_inprogress) { 1653204076Spjd dorewind = true; 1654204076Spjd synced = 0; 1655204076Spjd cv_wait(&sync_cond, &sync_lock); 1656204076Spjd } 1657204076Spjd mtx_unlock(&sync_lock); 1658204076Spjd /* 1659204076Spjd * Obtain offset at which we should synchronize. 1660204076Spjd * Rewind synchronization if needed. 1661204076Spjd */ 1662204076Spjd mtx_lock(&res->hr_amp_lock); 1663204076Spjd if (dorewind) 1664204076Spjd activemap_sync_rewind(res->hr_amp); 1665204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1666204076Spjd if (syncext != -1) { 1667204076Spjd /* 1668204076Spjd * We synchronized entire syncext extent, we can mark 1669204076Spjd * it as clean now. 1670204076Spjd */ 1671204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1672204076Spjd (void)hast_activemap_flush(res); 1673204076Spjd } 1674204076Spjd mtx_unlock(&res->hr_amp_lock); 1675204076Spjd if (dorewind) { 1676204076Spjd dorewind = false; 1677204076Spjd if (offset < 0) 1678204076Spjd pjdlog_info("Nodes are in sync."); 1679204076Spjd else { 1680204076Spjd pjdlog_info("Synchronization started. %ju bytes to go.", 1681204076Spjd (uintmax_t)(res->hr_extentsize * 1682204076Spjd activemap_ndirty(res->hr_amp))); 1683212038Spjd event_send(res, EVENT_SYNCSTART); 1684204076Spjd } 1685204076Spjd } 1686204076Spjd if (offset < 0) { 1687211878Spjd sync_stop(); 1688204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1689204076Spjd /* 1690204076Spjd * Synchronization complete, make both localcnt and 1691204076Spjd * remotecnt equal. 1692204076Spjd */ 1693204076Spjd ncomp = 1; 1694204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1695204076Spjd if (ISCONNECTED(res, ncomp)) { 1696204076Spjd if (synced > 0) { 1697204076Spjd pjdlog_info("Synchronization complete. " 1698204076Spjd "%jd bytes synchronized.", 1699204076Spjd (intmax_t)synced); 1700212038Spjd event_send(res, EVENT_SYNCDONE); 1701204076Spjd } 1702204076Spjd mtx_lock(&metadata_lock); 1703204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1704204076Spjd res->hr_primary_localcnt = 1705204076Spjd res->hr_secondary_localcnt; 1706204076Spjd res->hr_primary_remotecnt = 1707204076Spjd res->hr_secondary_remotecnt; 1708204076Spjd pjdlog_debug(1, 1709204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1710204076Spjd (uintmax_t)res->hr_primary_localcnt, 1711204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1712204076Spjd (void)metadata_write(res); 1713204076Spjd mtx_unlock(&metadata_lock); 1714204076Spjd } 1715204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1716204076Spjd continue; 1717204076Spjd } 1718204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1719204076Spjd QUEUE_TAKE2(hio, free); 1720204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1721204076Spjd /* 1722204076Spjd * Lock the range we are going to synchronize. We don't want 1723204076Spjd * race where someone writes between our read and write. 1724204076Spjd */ 1725204076Spjd for (;;) { 1726204076Spjd mtx_lock(&range_lock); 1727204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1728204076Spjd pjdlog_debug(2, 1729204076Spjd "sync: Range offset=%jd length=%jd locked.", 1730204076Spjd (intmax_t)offset, (intmax_t)length); 1731204076Spjd range_sync_wait = true; 1732204076Spjd cv_wait(&range_sync_cond, &range_lock); 1733204076Spjd range_sync_wait = false; 1734204076Spjd mtx_unlock(&range_lock); 1735204076Spjd continue; 1736204076Spjd } 1737204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1738204076Spjd mtx_unlock(&range_lock); 1739204076Spjd pjdlog_debug(2, 1740204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1741204076Spjd (intmax_t)offset, (intmax_t)length); 1742204076Spjd sleep(1); 1743204076Spjd continue; 1744204076Spjd } 1745204076Spjd mtx_unlock(&range_lock); 1746204076Spjd break; 1747204076Spjd } 1748204076Spjd /* 1749204076Spjd * First read the data from synchronization source. 1750204076Spjd */ 1751204076Spjd SYNCREQ(hio); 1752204076Spjd ggio = &hio->hio_ggio; 1753204076Spjd ggio->gctl_cmd = BIO_READ; 1754204076Spjd ggio->gctl_offset = offset; 1755204076Spjd ggio->gctl_length = length; 1756204076Spjd ggio->gctl_error = 0; 1757204076Spjd for (ii = 0; ii < ncomps; ii++) 1758204076Spjd hio->hio_errors[ii] = EINVAL; 1759204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1760204076Spjd hio); 1761204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1762204076Spjd hio); 1763204076Spjd mtx_lock(&metadata_lock); 1764204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1765204076Spjd /* 1766204076Spjd * This range is up-to-date on local component, 1767204076Spjd * so handle request locally. 1768204076Spjd */ 1769204076Spjd /* Local component is 0 for now. */ 1770204076Spjd ncomp = 0; 1771204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1772218138Spjd PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1773204076Spjd /* 1774204076Spjd * This range is out-of-date on local component, 1775204076Spjd * so send request to the remote node. 1776204076Spjd */ 1777204076Spjd /* Remote component is 1 for now. */ 1778204076Spjd ncomp = 1; 1779204076Spjd } 1780204076Spjd mtx_unlock(&metadata_lock); 1781204076Spjd refcount_init(&hio->hio_countdown, 1); 1782204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1783204076Spjd 1784204076Spjd /* 1785204076Spjd * Let's wait for READ to finish. 1786204076Spjd */ 1787204076Spjd mtx_lock(&sync_lock); 1788204076Spjd while (!ISSYNCREQDONE(hio)) 1789204076Spjd cv_wait(&sync_cond, &sync_lock); 1790204076Spjd mtx_unlock(&sync_lock); 1791204076Spjd 1792204076Spjd if (hio->hio_errors[ncomp] != 0) { 1793204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1794204076Spjd strerror(hio->hio_errors[ncomp])); 1795204076Spjd goto free_queue; 1796204076Spjd } 1797204076Spjd 1798204076Spjd /* 1799204076Spjd * We read the data from synchronization source, now write it 1800204076Spjd * to synchronization target. 1801204076Spjd */ 1802204076Spjd SYNCREQ(hio); 1803204076Spjd ggio->gctl_cmd = BIO_WRITE; 1804204076Spjd for (ii = 0; ii < ncomps; ii++) 1805204076Spjd hio->hio_errors[ii] = EINVAL; 1806204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1807204076Spjd hio); 1808204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1809204076Spjd hio); 1810204076Spjd mtx_lock(&metadata_lock); 1811204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1812204076Spjd /* 1813204076Spjd * This range is up-to-date on local component, 1814204076Spjd * so we update remote component. 1815204076Spjd */ 1816204076Spjd /* Remote component is 1 for now. */ 1817204076Spjd ncomp = 1; 1818204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1819218138Spjd PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1820204076Spjd /* 1821204076Spjd * This range is out-of-date on local component, 1822204076Spjd * so we update it. 1823204076Spjd */ 1824204076Spjd /* Local component is 0 for now. */ 1825204076Spjd ncomp = 0; 1826204076Spjd } 1827204076Spjd mtx_unlock(&metadata_lock); 1828204076Spjd 1829204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1830204076Spjd hio); 1831204076Spjd refcount_init(&hio->hio_countdown, 1); 1832204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1833204076Spjd 1834204076Spjd /* 1835204076Spjd * Let's wait for WRITE to finish. 1836204076Spjd */ 1837204076Spjd mtx_lock(&sync_lock); 1838204076Spjd while (!ISSYNCREQDONE(hio)) 1839204076Spjd cv_wait(&sync_cond, &sync_lock); 1840204076Spjd mtx_unlock(&sync_lock); 1841204076Spjd 1842204076Spjd if (hio->hio_errors[ncomp] != 0) { 1843204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1844204076Spjd strerror(hio->hio_errors[ncomp])); 1845204076Spjd goto free_queue; 1846204076Spjd } 1847211880Spjd 1848211880Spjd synced += length; 1849204076Spjdfree_queue: 1850204076Spjd mtx_lock(&range_lock); 1851204076Spjd rangelock_del(range_sync, offset, length); 1852204076Spjd if (range_regular_wait) 1853204076Spjd cv_signal(&range_regular_cond); 1854204076Spjd mtx_unlock(&range_lock); 1855204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1856204076Spjd hio); 1857204076Spjd QUEUE_INSERT2(hio, free); 1858204076Spjd } 1859204076Spjd /* NOTREACHED */ 1860204076Spjd return (NULL); 1861204076Spjd} 1862204076Spjd 1863217784Spjdvoid 1864217784Spjdprimary_config_reload(struct hast_resource *res, struct nv *nv) 1865210886Spjd{ 1866210886Spjd unsigned int ii, ncomps; 1867217784Spjd int modified, vint; 1868217784Spjd const char *vstr; 1869210886Spjd 1870210886Spjd pjdlog_info("Reloading configuration..."); 1871210886Spjd 1872218138Spjd PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 1873218138Spjd PJDLOG_ASSERT(gres == res); 1874217784Spjd nv_assert(nv, "remoteaddr"); 1875217784Spjd nv_assert(nv, "replication"); 1876217784Spjd nv_assert(nv, "timeout"); 1877217784Spjd nv_assert(nv, "exec"); 1878217784Spjd 1879210886Spjd ncomps = HAST_NCOMPONENTS; 1880210886Spjd 1881210886Spjd#define MODIFIED_REMOTEADDR 0x1 1882210886Spjd#define MODIFIED_REPLICATION 0x2 1883210886Spjd#define MODIFIED_TIMEOUT 0x4 1884211886Spjd#define MODIFIED_EXEC 0x8 1885210886Spjd modified = 0; 1886217784Spjd 1887217784Spjd vstr = nv_get_string(nv, "remoteaddr"); 1888217784Spjd if (strcmp(gres->hr_remoteaddr, vstr) != 0) { 1889210886Spjd /* 1890210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1891210886Spjd * We want remote_close() to log disconnect from the old 1892210886Spjd * addresses, not from the new ones. 1893210886Spjd */ 1894210886Spjd modified |= MODIFIED_REMOTEADDR; 1895210886Spjd } 1896217784Spjd vint = nv_get_int32(nv, "replication"); 1897217784Spjd if (gres->hr_replication != vint) { 1898217784Spjd gres->hr_replication = vint; 1899210886Spjd modified |= MODIFIED_REPLICATION; 1900210886Spjd } 1901217784Spjd vint = nv_get_int32(nv, "timeout"); 1902217784Spjd if (gres->hr_timeout != vint) { 1903217784Spjd gres->hr_timeout = vint; 1904210886Spjd modified |= MODIFIED_TIMEOUT; 1905210886Spjd } 1906217784Spjd vstr = nv_get_string(nv, "exec"); 1907217784Spjd if (strcmp(gres->hr_exec, vstr) != 0) { 1908217784Spjd strlcpy(gres->hr_exec, vstr, sizeof(gres->hr_exec)); 1909211886Spjd modified |= MODIFIED_EXEC; 1910211886Spjd } 1911217784Spjd 1912210886Spjd /* 1913210886Spjd * If only timeout was modified we only need to change it without 1914210886Spjd * reconnecting. 1915210886Spjd */ 1916210886Spjd if (modified == MODIFIED_TIMEOUT) { 1917210886Spjd for (ii = 0; ii < ncomps; ii++) { 1918210886Spjd if (!ISREMOTE(ii)) 1919210886Spjd continue; 1920210886Spjd rw_rlock(&hio_remote_lock[ii]); 1921210886Spjd if (!ISCONNECTED(gres, ii)) { 1922210886Spjd rw_unlock(&hio_remote_lock[ii]); 1923210886Spjd continue; 1924210886Spjd } 1925210886Spjd rw_unlock(&hio_remote_lock[ii]); 1926210886Spjd if (proto_timeout(gres->hr_remotein, 1927210886Spjd gres->hr_timeout) < 0) { 1928210886Spjd pjdlog_errno(LOG_WARNING, 1929210886Spjd "Unable to set connection timeout"); 1930210886Spjd } 1931210886Spjd if (proto_timeout(gres->hr_remoteout, 1932210886Spjd gres->hr_timeout) < 0) { 1933210886Spjd pjdlog_errno(LOG_WARNING, 1934210886Spjd "Unable to set connection timeout"); 1935210886Spjd } 1936210886Spjd } 1937211886Spjd } else if ((modified & 1938211886Spjd (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1939210886Spjd for (ii = 0; ii < ncomps; ii++) { 1940210886Spjd if (!ISREMOTE(ii)) 1941210886Spjd continue; 1942210886Spjd remote_close(gres, ii); 1943210886Spjd } 1944210886Spjd if (modified & MODIFIED_REMOTEADDR) { 1945217784Spjd vstr = nv_get_string(nv, "remoteaddr"); 1946217784Spjd strlcpy(gres->hr_remoteaddr, vstr, 1947210886Spjd sizeof(gres->hr_remoteaddr)); 1948210886Spjd } 1949210886Spjd } 1950210886Spjd#undef MODIFIED_REMOTEADDR 1951210886Spjd#undef MODIFIED_REPLICATION 1952210886Spjd#undef MODIFIED_TIMEOUT 1953211886Spjd#undef MODIFIED_EXEC 1954210886Spjd 1955210886Spjd pjdlog_info("Configuration reloaded successfully."); 1956210886Spjd} 1957210886Spjd 1958211882Spjdstatic void 1959211981Spjdguard_one(struct hast_resource *res, unsigned int ncomp) 1960211981Spjd{ 1961211981Spjd struct proto_conn *in, *out; 1962211981Spjd 1963211981Spjd if (!ISREMOTE(ncomp)) 1964211981Spjd return; 1965211981Spjd 1966211981Spjd rw_rlock(&hio_remote_lock[ncomp]); 1967211981Spjd 1968211981Spjd if (!real_remote(res)) { 1969211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1970211981Spjd return; 1971211981Spjd } 1972211981Spjd 1973211981Spjd if (ISCONNECTED(res, ncomp)) { 1974218138Spjd PJDLOG_ASSERT(res->hr_remotein != NULL); 1975218138Spjd PJDLOG_ASSERT(res->hr_remoteout != NULL); 1976211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1977211981Spjd pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 1978211981Spjd res->hr_remoteaddr); 1979211981Spjd return; 1980211981Spjd } 1981211981Spjd 1982218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 1983218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 1984211981Spjd /* 1985211981Spjd * Upgrade the lock. It doesn't have to be atomic as no other thread 1986211981Spjd * can change connection status from disconnected to connected. 1987211981Spjd */ 1988211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1989211981Spjd pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 1990211981Spjd res->hr_remoteaddr); 1991211981Spjd in = out = NULL; 1992211981Spjd if (init_remote(res, &in, &out)) { 1993211981Spjd rw_wlock(&hio_remote_lock[ncomp]); 1994218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 1995218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 1996218138Spjd PJDLOG_ASSERT(in != NULL && out != NULL); 1997211981Spjd res->hr_remotein = in; 1998211981Spjd res->hr_remoteout = out; 1999211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 2000211981Spjd pjdlog_info("Successfully reconnected to %s.", 2001211981Spjd res->hr_remoteaddr); 2002211981Spjd sync_start(); 2003211981Spjd } else { 2004211981Spjd /* Both connections should be NULL. */ 2005218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 2006218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 2007218138Spjd PJDLOG_ASSERT(in == NULL && out == NULL); 2008211981Spjd pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 2009211981Spjd res->hr_remoteaddr); 2010211981Spjd } 2011211981Spjd} 2012211981Spjd 2013204076Spjd/* 2014204076Spjd * Thread guards remote connections and reconnects when needed, handles 2015204076Spjd * signals, etc. 2016204076Spjd */ 2017204076Spjdstatic void * 2018204076Spjdguard_thread(void *arg) 2019204076Spjd{ 2020204076Spjd struct hast_resource *res = arg; 2021204076Spjd unsigned int ii, ncomps; 2022211982Spjd struct timespec timeout; 2023211981Spjd time_t lastcheck, now; 2024211982Spjd sigset_t mask; 2025211982Spjd int signo; 2026204076Spjd 2027204076Spjd ncomps = HAST_NCOMPONENTS; 2028211981Spjd lastcheck = time(NULL); 2029204076Spjd 2030211982Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 2031211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 2032211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 2033211982Spjd 2034215332Spjd timeout.tv_sec = RETRY_SLEEP; 2035211982Spjd timeout.tv_nsec = 0; 2036211982Spjd signo = -1; 2037211982Spjd 2038204076Spjd for (;;) { 2039211982Spjd switch (signo) { 2040211982Spjd case SIGINT: 2041211982Spjd case SIGTERM: 2042211982Spjd sigexit_received = true; 2043204076Spjd primary_exitx(EX_OK, 2044204076Spjd "Termination signal received, exiting."); 2045211982Spjd break; 2046211982Spjd default: 2047211982Spjd break; 2048204076Spjd } 2049211882Spjd 2050204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 2051211981Spjd now = time(NULL); 2052211982Spjd if (lastcheck + RETRY_SLEEP <= now) { 2053211982Spjd for (ii = 0; ii < ncomps; ii++) 2054211981Spjd guard_one(res, ii); 2055211981Spjd lastcheck = now; 2056204076Spjd } 2057211982Spjd signo = sigtimedwait(&mask, NULL, &timeout); 2058204076Spjd } 2059204076Spjd /* NOTREACHED */ 2060204076Spjd return (NULL); 2061204076Spjd} 2062