primary.c revision 218043
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 218043 2011-01-28 21:52:37Z 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) 183214692Spjd#define QUEUE_TAKE1(hio, name, ncomp, timeout) do { \ 184214692Spjd bool _last; \ 185214692Spjd \ 186204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 187214692Spjd _last = false; \ 188214692Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL && !_last) { \ 189214692Spjd cv_timedwait(&hio_##name##_list_cond[(ncomp)], \ 190214692Spjd &hio_##name##_list_lock[(ncomp)], (timeout)); \ 191214692Spjd if ((timeout) != 0) \ 192214692Spjd _last = true; \ 193204076Spjd } \ 194214692Spjd if (hio != NULL) { \ 195214692Spjd TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio), \ 196214692Spjd hio_next[(ncomp)]); \ 197214692Spjd } \ 198204076Spjd mtx_unlock(&hio_##name##_list_lock[(ncomp)]); \ 199204076Spjd} while (0) 200204076Spjd#define QUEUE_TAKE2(hio, name) do { \ 201204076Spjd mtx_lock(&hio_##name##_list_lock); \ 202204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 203204076Spjd cv_wait(&hio_##name##_list_cond, \ 204204076Spjd &hio_##name##_list_lock); \ 205204076Spjd } \ 206204076Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next); \ 207204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 208204076Spjd} while (0) 209204076Spjd 210209183Spjd#define SYNCREQ(hio) do { \ 211209183Spjd (hio)->hio_ggio.gctl_unit = -1; \ 212209183Spjd (hio)->hio_ggio.gctl_seq = 1; \ 213209183Spjd} while (0) 214204076Spjd#define ISSYNCREQ(hio) ((hio)->hio_ggio.gctl_unit == -1) 215204076Spjd#define SYNCREQDONE(hio) do { (hio)->hio_ggio.gctl_unit = -2; } while (0) 216204076Spjd#define ISSYNCREQDONE(hio) ((hio)->hio_ggio.gctl_unit == -2) 217204076Spjd 218204076Spjdstatic struct hast_resource *gres; 219204076Spjd 220204076Spjdstatic pthread_mutex_t range_lock; 221204076Spjdstatic struct rangelocks *range_regular; 222204076Spjdstatic bool range_regular_wait; 223204076Spjdstatic pthread_cond_t range_regular_cond; 224204076Spjdstatic struct rangelocks *range_sync; 225204076Spjdstatic bool range_sync_wait; 226204076Spjdstatic pthread_cond_t range_sync_cond; 227204076Spjd 228204076Spjdstatic void *ggate_recv_thread(void *arg); 229204076Spjdstatic void *local_send_thread(void *arg); 230204076Spjdstatic void *remote_send_thread(void *arg); 231204076Spjdstatic void *remote_recv_thread(void *arg); 232204076Spjdstatic void *ggate_send_thread(void *arg); 233204076Spjdstatic void *sync_thread(void *arg); 234204076Spjdstatic void *guard_thread(void *arg); 235204076Spjd 236211982Spjdstatic void 237204076Spjdcleanup(struct hast_resource *res) 238204076Spjd{ 239204076Spjd int rerrno; 240204076Spjd 241204076Spjd /* Remember errno. */ 242204076Spjd rerrno = errno; 243204076Spjd 244204076Spjd /* Destroy ggate provider if we created one. */ 245204076Spjd if (res->hr_ggateunit >= 0) { 246204076Spjd struct g_gate_ctl_destroy ggiod; 247204076Spjd 248213533Spjd bzero(&ggiod, sizeof(ggiod)); 249204076Spjd ggiod.gctl_version = G_GATE_VERSION; 250204076Spjd ggiod.gctl_unit = res->hr_ggateunit; 251204076Spjd ggiod.gctl_force = 1; 252204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 253213531Spjd pjdlog_errno(LOG_WARNING, 254213531Spjd "Unable to destroy hast/%s device", 255204076Spjd res->hr_provname); 256204076Spjd } 257204076Spjd res->hr_ggateunit = -1; 258204076Spjd } 259204076Spjd 260204076Spjd /* Restore errno. */ 261204076Spjd errno = rerrno; 262204076Spjd} 263204076Spjd 264212899Spjdstatic __dead2 void 265204076Spjdprimary_exit(int exitcode, const char *fmt, ...) 266204076Spjd{ 267204076Spjd va_list ap; 268204076Spjd 269204076Spjd assert(exitcode != EX_OK); 270204076Spjd va_start(ap, fmt); 271204076Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 272204076Spjd va_end(ap); 273204076Spjd cleanup(gres); 274204076Spjd exit(exitcode); 275204076Spjd} 276204076Spjd 277212899Spjdstatic __dead2 void 278204076Spjdprimary_exitx(int exitcode, const char *fmt, ...) 279204076Spjd{ 280204076Spjd va_list ap; 281204076Spjd 282204076Spjd va_start(ap, fmt); 283204076Spjd pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 284204076Spjd va_end(ap); 285204076Spjd cleanup(gres); 286204076Spjd exit(exitcode); 287204076Spjd} 288204076Spjd 289204076Spjdstatic int 290204076Spjdhast_activemap_flush(struct hast_resource *res) 291204076Spjd{ 292204076Spjd const unsigned char *buf; 293204076Spjd size_t size; 294204076Spjd 295204076Spjd buf = activemap_bitmap(res->hr_amp, &size); 296204076Spjd assert(buf != NULL); 297204076Spjd assert((size % res->hr_local_sectorsize) == 0); 298204076Spjd if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 299204076Spjd (ssize_t)size) { 300204076Spjd KEEP_ERRNO(pjdlog_errno(LOG_ERR, 301204076Spjd "Unable to flush activemap to disk")); 302204076Spjd return (-1); 303204076Spjd } 304204076Spjd return (0); 305204076Spjd} 306204076Spjd 307210881Spjdstatic bool 308210881Spjdreal_remote(const struct hast_resource *res) 309210881Spjd{ 310210881Spjd 311210881Spjd return (strcmp(res->hr_remoteaddr, "none") != 0); 312210881Spjd} 313210881Spjd 314204076Spjdstatic void 315204076Spjdinit_environment(struct hast_resource *res __unused) 316204076Spjd{ 317204076Spjd struct hio *hio; 318204076Spjd unsigned int ii, ncomps; 319204076Spjd 320204076Spjd /* 321204076Spjd * In the future it might be per-resource value. 322204076Spjd */ 323204076Spjd ncomps = HAST_NCOMPONENTS; 324204076Spjd 325204076Spjd /* 326204076Spjd * Allocate memory needed by lists. 327204076Spjd */ 328204076Spjd hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 329204076Spjd if (hio_send_list == NULL) { 330204076Spjd primary_exitx(EX_TEMPFAIL, 331204076Spjd "Unable to allocate %zu bytes of memory for send lists.", 332204076Spjd sizeof(hio_send_list[0]) * ncomps); 333204076Spjd } 334204076Spjd hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 335204076Spjd if (hio_send_list_lock == NULL) { 336204076Spjd primary_exitx(EX_TEMPFAIL, 337204076Spjd "Unable to allocate %zu bytes of memory for send list locks.", 338204076Spjd sizeof(hio_send_list_lock[0]) * ncomps); 339204076Spjd } 340204076Spjd hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 341204076Spjd if (hio_send_list_cond == NULL) { 342204076Spjd primary_exitx(EX_TEMPFAIL, 343204076Spjd "Unable to allocate %zu bytes of memory for send list condition variables.", 344204076Spjd sizeof(hio_send_list_cond[0]) * ncomps); 345204076Spjd } 346204076Spjd hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 347204076Spjd if (hio_recv_list == NULL) { 348204076Spjd primary_exitx(EX_TEMPFAIL, 349204076Spjd "Unable to allocate %zu bytes of memory for recv lists.", 350204076Spjd sizeof(hio_recv_list[0]) * ncomps); 351204076Spjd } 352204076Spjd hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 353204076Spjd if (hio_recv_list_lock == NULL) { 354204076Spjd primary_exitx(EX_TEMPFAIL, 355204076Spjd "Unable to allocate %zu bytes of memory for recv list locks.", 356204076Spjd sizeof(hio_recv_list_lock[0]) * ncomps); 357204076Spjd } 358204076Spjd hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 359204076Spjd if (hio_recv_list_cond == NULL) { 360204076Spjd primary_exitx(EX_TEMPFAIL, 361204076Spjd "Unable to allocate %zu bytes of memory for recv list condition variables.", 362204076Spjd sizeof(hio_recv_list_cond[0]) * ncomps); 363204076Spjd } 364204076Spjd hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 365204076Spjd if (hio_remote_lock == NULL) { 366204076Spjd primary_exitx(EX_TEMPFAIL, 367204076Spjd "Unable to allocate %zu bytes of memory for remote connections locks.", 368204076Spjd sizeof(hio_remote_lock[0]) * ncomps); 369204076Spjd } 370204076Spjd 371204076Spjd /* 372204076Spjd * Initialize lists, their locks and theirs condition variables. 373204076Spjd */ 374204076Spjd TAILQ_INIT(&hio_free_list); 375204076Spjd mtx_init(&hio_free_list_lock); 376204076Spjd cv_init(&hio_free_list_cond); 377204076Spjd for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 378204076Spjd TAILQ_INIT(&hio_send_list[ii]); 379204076Spjd mtx_init(&hio_send_list_lock[ii]); 380204076Spjd cv_init(&hio_send_list_cond[ii]); 381204076Spjd TAILQ_INIT(&hio_recv_list[ii]); 382204076Spjd mtx_init(&hio_recv_list_lock[ii]); 383204076Spjd cv_init(&hio_recv_list_cond[ii]); 384204076Spjd rw_init(&hio_remote_lock[ii]); 385204076Spjd } 386204076Spjd TAILQ_INIT(&hio_done_list); 387204076Spjd mtx_init(&hio_done_list_lock); 388204076Spjd cv_init(&hio_done_list_cond); 389204076Spjd mtx_init(&metadata_lock); 390204076Spjd 391204076Spjd /* 392204076Spjd * Allocate requests pool and initialize requests. 393204076Spjd */ 394204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 395204076Spjd hio = malloc(sizeof(*hio)); 396204076Spjd if (hio == NULL) { 397204076Spjd primary_exitx(EX_TEMPFAIL, 398204076Spjd "Unable to allocate %zu bytes of memory for hio request.", 399204076Spjd sizeof(*hio)); 400204076Spjd } 401204076Spjd hio->hio_countdown = 0; 402204076Spjd hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 403204076Spjd if (hio->hio_errors == NULL) { 404204076Spjd primary_exitx(EX_TEMPFAIL, 405204076Spjd "Unable allocate %zu bytes of memory for hio errors.", 406204076Spjd sizeof(hio->hio_errors[0]) * ncomps); 407204076Spjd } 408204076Spjd hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 409204076Spjd if (hio->hio_next == NULL) { 410204076Spjd primary_exitx(EX_TEMPFAIL, 411204076Spjd "Unable allocate %zu bytes of memory for hio_next field.", 412204076Spjd sizeof(hio->hio_next[0]) * ncomps); 413204076Spjd } 414204076Spjd hio->hio_ggio.gctl_version = G_GATE_VERSION; 415204076Spjd hio->hio_ggio.gctl_data = malloc(MAXPHYS); 416204076Spjd if (hio->hio_ggio.gctl_data == NULL) { 417204076Spjd primary_exitx(EX_TEMPFAIL, 418204076Spjd "Unable to allocate %zu bytes of memory for gctl_data.", 419204076Spjd MAXPHYS); 420204076Spjd } 421204076Spjd hio->hio_ggio.gctl_length = MAXPHYS; 422204076Spjd hio->hio_ggio.gctl_error = 0; 423204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 424204076Spjd } 425204076Spjd} 426204076Spjd 427214284Spjdstatic bool 428214284Spjdinit_resuid(struct hast_resource *res) 429214284Spjd{ 430214284Spjd 431214284Spjd mtx_lock(&metadata_lock); 432214284Spjd if (res->hr_resuid != 0) { 433214284Spjd mtx_unlock(&metadata_lock); 434214284Spjd return (false); 435214284Spjd } else { 436214284Spjd /* Initialize unique resource identifier. */ 437214284Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 438214284Spjd mtx_unlock(&metadata_lock); 439214284Spjd if (metadata_write(res) < 0) 440214284Spjd exit(EX_NOINPUT); 441214284Spjd return (true); 442214284Spjd } 443214284Spjd} 444214284Spjd 445204076Spjdstatic void 446204076Spjdinit_local(struct hast_resource *res) 447204076Spjd{ 448204076Spjd unsigned char *buf; 449204076Spjd size_t mapsize; 450204076Spjd 451204076Spjd if (metadata_read(res, true) < 0) 452204076Spjd exit(EX_NOINPUT); 453204076Spjd mtx_init(&res->hr_amp_lock); 454204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 455204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 456204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 457204076Spjd } 458204076Spjd mtx_init(&range_lock); 459204076Spjd cv_init(&range_regular_cond); 460204076Spjd if (rangelock_init(&range_regular) < 0) 461204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 462204076Spjd cv_init(&range_sync_cond); 463204076Spjd if (rangelock_init(&range_sync) < 0) 464204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 465204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 466204076Spjd buf = calloc(1, mapsize); 467204076Spjd if (buf == NULL) { 468204076Spjd primary_exitx(EX_TEMPFAIL, 469204076Spjd "Unable to allocate buffer for activemap."); 470204076Spjd } 471204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 472204076Spjd (ssize_t)mapsize) { 473204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 474204076Spjd } 475204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 476209181Spjd free(buf); 477204076Spjd if (res->hr_resuid != 0) 478204076Spjd return; 479204076Spjd /* 480214284Spjd * We're using provider for the first time. Initialize local and remote 481214284Spjd * counters. We don't initialize resuid here, as we want to do it just 482214284Spjd * in time. The reason for this is that we want to inform secondary 483214284Spjd * that there were no writes yet, so there is no need to synchronize 484214284Spjd * anything. 485204076Spjd */ 486204076Spjd res->hr_primary_localcnt = 1; 487204076Spjd res->hr_primary_remotecnt = 0; 488204076Spjd if (metadata_write(res) < 0) 489204076Spjd exit(EX_NOINPUT); 490204076Spjd} 491204076Spjd 492205738Spjdstatic bool 493205738Spjdinit_remote(struct hast_resource *res, struct proto_conn **inp, 494205738Spjd struct proto_conn **outp) 495204076Spjd{ 496205738Spjd struct proto_conn *in, *out; 497204076Spjd struct nv *nvout, *nvin; 498204076Spjd const unsigned char *token; 499204076Spjd unsigned char *map; 500204076Spjd const char *errmsg; 501204076Spjd int32_t extentsize; 502204076Spjd int64_t datasize; 503204076Spjd uint32_t mapsize; 504204076Spjd size_t size; 505204076Spjd 506205738Spjd assert((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 507210881Spjd assert(real_remote(res)); 508205738Spjd 509205738Spjd in = out = NULL; 510211983Spjd errmsg = NULL; 511205738Spjd 512204076Spjd /* Prepare outgoing connection with remote node. */ 513205738Spjd if (proto_client(res->hr_remoteaddr, &out) < 0) { 514215331Spjd primary_exit(EX_TEMPFAIL, 515215331Spjd "Unable to create outgoing connection to %s", 516204076Spjd res->hr_remoteaddr); 517204076Spjd } 518204076Spjd /* Try to connect, but accept failure. */ 519205738Spjd if (proto_connect(out) < 0) { 520204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 521204076Spjd res->hr_remoteaddr); 522204076Spjd goto close; 523204076Spjd } 524207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 525207371Spjd if (proto_timeout(out, res->hr_timeout) < 0) 526207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 527204076Spjd /* 528204076Spjd * First handshake step. 529204076Spjd * Setup outgoing connection with remote node. 530204076Spjd */ 531204076Spjd nvout = nv_alloc(); 532204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 533204076Spjd if (nv_error(nvout) != 0) { 534204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 535204076Spjd "Unable to allocate header for connection with %s", 536204076Spjd res->hr_remoteaddr); 537204076Spjd nv_free(nvout); 538204076Spjd goto close; 539204076Spjd } 540205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 541204076Spjd pjdlog_errno(LOG_WARNING, 542204076Spjd "Unable to send handshake header to %s", 543204076Spjd res->hr_remoteaddr); 544204076Spjd nv_free(nvout); 545204076Spjd goto close; 546204076Spjd } 547204076Spjd nv_free(nvout); 548205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 549204076Spjd pjdlog_errno(LOG_WARNING, 550204076Spjd "Unable to receive handshake header from %s", 551204076Spjd res->hr_remoteaddr); 552204076Spjd goto close; 553204076Spjd } 554204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 555204076Spjd if (errmsg != NULL) { 556204076Spjd pjdlog_warning("%s", errmsg); 557204076Spjd nv_free(nvin); 558204076Spjd goto close; 559204076Spjd } 560204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 561204076Spjd if (token == NULL) { 562204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 563204076Spjd res->hr_remoteaddr); 564204076Spjd nv_free(nvin); 565204076Spjd goto close; 566204076Spjd } 567204076Spjd if (size != sizeof(res->hr_token)) { 568204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 569204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 570204076Spjd nv_free(nvin); 571204076Spjd goto close; 572204076Spjd } 573204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 574204076Spjd nv_free(nvin); 575204076Spjd 576204076Spjd /* 577204076Spjd * Second handshake step. 578204076Spjd * Setup incoming connection with remote node. 579204076Spjd */ 580205738Spjd if (proto_client(res->hr_remoteaddr, &in) < 0) { 581215331Spjd primary_exit(EX_TEMPFAIL, 582215331Spjd "Unable to create incoming connection to %s", 583204076Spjd res->hr_remoteaddr); 584204076Spjd } 585204076Spjd /* Try to connect, but accept failure. */ 586205738Spjd if (proto_connect(in) < 0) { 587204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 588204076Spjd res->hr_remoteaddr); 589204076Spjd goto close; 590204076Spjd } 591207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 592207371Spjd if (proto_timeout(in, res->hr_timeout) < 0) 593207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 594204076Spjd nvout = nv_alloc(); 595204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 596204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 597204076Spjd "token"); 598214284Spjd if (res->hr_resuid == 0) { 599214284Spjd /* 600214284Spjd * The resuid field was not yet initialized. 601214284Spjd * Because we do synchronization inside init_resuid(), it is 602214284Spjd * possible that someone already initialized it, the function 603214284Spjd * will return false then, but if we successfully initialized 604214284Spjd * it, we will get true. True means that there were no writes 605214284Spjd * to this resource yet and we want to inform secondary that 606214284Spjd * synchronization is not needed by sending "virgin" argument. 607214284Spjd */ 608214284Spjd if (init_resuid(res)) 609214284Spjd nv_add_int8(nvout, 1, "virgin"); 610214284Spjd } 611204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 612204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 613204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 614204076Spjd if (nv_error(nvout) != 0) { 615204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 616204076Spjd "Unable to allocate header for connection with %s", 617204076Spjd res->hr_remoteaddr); 618204076Spjd nv_free(nvout); 619204076Spjd goto close; 620204076Spjd } 621205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 622204076Spjd pjdlog_errno(LOG_WARNING, 623204076Spjd "Unable to send handshake header to %s", 624204076Spjd res->hr_remoteaddr); 625204076Spjd nv_free(nvout); 626204076Spjd goto close; 627204076Spjd } 628204076Spjd nv_free(nvout); 629205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 630204076Spjd pjdlog_errno(LOG_WARNING, 631204076Spjd "Unable to receive handshake header from %s", 632204076Spjd res->hr_remoteaddr); 633204076Spjd goto close; 634204076Spjd } 635204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 636204076Spjd if (errmsg != NULL) { 637204076Spjd pjdlog_warning("%s", errmsg); 638204076Spjd nv_free(nvin); 639204076Spjd goto close; 640204076Spjd } 641204076Spjd datasize = nv_get_int64(nvin, "datasize"); 642204076Spjd if (datasize != res->hr_datasize) { 643204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 644204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 645204076Spjd nv_free(nvin); 646204076Spjd goto close; 647204076Spjd } 648204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 649204076Spjd if (extentsize != res->hr_extentsize) { 650204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 651204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 652204076Spjd nv_free(nvin); 653204076Spjd goto close; 654204076Spjd } 655204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 656204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 657204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 658204076Spjd map = NULL; 659204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 660204076Spjd if (mapsize > 0) { 661204076Spjd map = malloc(mapsize); 662204076Spjd if (map == NULL) { 663204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 664204076Spjd (uintmax_t)mapsize); 665204076Spjd nv_free(nvin); 666204076Spjd goto close; 667204076Spjd } 668204076Spjd /* 669204076Spjd * Remote node have some dirty extents on its own, lets 670204076Spjd * download its activemap. 671204076Spjd */ 672205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 673204076Spjd mapsize) < 0) { 674204076Spjd pjdlog_errno(LOG_ERR, 675204076Spjd "Unable to receive remote activemap"); 676204076Spjd nv_free(nvin); 677204076Spjd free(map); 678204076Spjd goto close; 679204076Spjd } 680204076Spjd /* 681204076Spjd * Merge local and remote bitmaps. 682204076Spjd */ 683204076Spjd activemap_merge(res->hr_amp, map, mapsize); 684204076Spjd free(map); 685204076Spjd /* 686204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 687204076Spjd * disk before we start to synchronize. 688204076Spjd */ 689204076Spjd (void)hast_activemap_flush(res); 690204076Spjd } 691214274Spjd nv_free(nvin); 692204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 693205738Spjd if (inp != NULL && outp != NULL) { 694205738Spjd *inp = in; 695205738Spjd *outp = out; 696205738Spjd } else { 697205738Spjd res->hr_remotein = in; 698205738Spjd res->hr_remoteout = out; 699205738Spjd } 700212038Spjd event_send(res, EVENT_CONNECT); 701205738Spjd return (true); 702205738Spjdclose: 703211983Spjd if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0) 704212038Spjd event_send(res, EVENT_SPLITBRAIN); 705205738Spjd proto_close(out); 706205738Spjd if (in != NULL) 707205738Spjd proto_close(in); 708205738Spjd return (false); 709205738Spjd} 710205738Spjd 711205738Spjdstatic void 712205738Spjdsync_start(void) 713205738Spjd{ 714205738Spjd 715204076Spjd mtx_lock(&sync_lock); 716204076Spjd sync_inprogress = true; 717204076Spjd mtx_unlock(&sync_lock); 718204076Spjd cv_signal(&sync_cond); 719204076Spjd} 720204076Spjd 721204076Spjdstatic void 722211878Spjdsync_stop(void) 723211878Spjd{ 724211878Spjd 725211878Spjd mtx_lock(&sync_lock); 726211878Spjd if (sync_inprogress) 727211878Spjd sync_inprogress = false; 728211878Spjd mtx_unlock(&sync_lock); 729211878Spjd} 730211878Spjd 731211878Spjdstatic void 732204076Spjdinit_ggate(struct hast_resource *res) 733204076Spjd{ 734204076Spjd struct g_gate_ctl_create ggiocreate; 735204076Spjd struct g_gate_ctl_cancel ggiocancel; 736204076Spjd 737204076Spjd /* 738204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 739204076Spjd */ 740204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 741204076Spjd if (res->hr_ggatefd < 0) 742204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 743204076Spjd /* 744204076Spjd * Create provider before trying to connect, as connection failure 745204076Spjd * is not critical, but may take some time. 746204076Spjd */ 747213533Spjd bzero(&ggiocreate, sizeof(ggiocreate)); 748204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 749204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 750204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 751204076Spjd ggiocreate.gctl_flags = 0; 752206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 753204076Spjd ggiocreate.gctl_timeout = 0; 754204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 755204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 756204076Spjd res->hr_provname); 757204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 758204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 759204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 760204076Spjd return; 761204076Spjd } 762204076Spjd if (errno != EEXIST) { 763204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 764204076Spjd res->hr_provname); 765204076Spjd } 766204076Spjd pjdlog_debug(1, 767204076Spjd "Device hast/%s already exists, we will try to take it over.", 768204076Spjd res->hr_provname); 769204076Spjd /* 770204076Spjd * If we received EEXIST, we assume that the process who created the 771204076Spjd * provider died and didn't clean up. In that case we will start from 772204076Spjd * where he left of. 773204076Spjd */ 774213533Spjd bzero(&ggiocancel, sizeof(ggiocancel)); 775204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 776204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 777204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 778204076Spjd res->hr_provname); 779204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 780204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 781204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 782204076Spjd return; 783204076Spjd } 784204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 785204076Spjd res->hr_provname); 786204076Spjd} 787204076Spjd 788204076Spjdvoid 789204076Spjdhastd_primary(struct hast_resource *res) 790204076Spjd{ 791204076Spjd pthread_t td; 792204076Spjd pid_t pid; 793218043Spjd int error, mode; 794204076Spjd 795204076Spjd /* 796204076Spjd * Create communication channel between parent and child. 797204076Spjd */ 798204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 799218042Spjd /* TODO: There's no need for this to be fatal error. */ 800204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 801212034Spjd pjdlog_exit(EX_OSERR, 802204076Spjd "Unable to create control sockets between parent and child"); 803204076Spjd } 804212038Spjd /* 805212038Spjd * Create communication channel between child and parent. 806212038Spjd */ 807212038Spjd if (proto_client("socketpair://", &res->hr_event) < 0) { 808218042Spjd /* TODO: There's no need for this to be fatal error. */ 809212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 810212038Spjd pjdlog_exit(EX_OSERR, 811212038Spjd "Unable to create event sockets between child and parent"); 812212038Spjd } 813204076Spjd 814204076Spjd pid = fork(); 815204076Spjd if (pid < 0) { 816218042Spjd /* TODO: There's no need for this to be fatal error. */ 817204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 818212034Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to fork"); 819204076Spjd } 820204076Spjd 821204076Spjd if (pid > 0) { 822204076Spjd /* This is parent. */ 823212038Spjd /* Declare that we are receiver. */ 824212038Spjd proto_recv(res->hr_event, NULL, 0); 825218043Spjd /* Declare that we are sender. */ 826218043Spjd proto_send(res->hr_ctrl, NULL, 0); 827204076Spjd res->hr_workerpid = pid; 828204076Spjd return; 829204076Spjd } 830211977Spjd 831211984Spjd gres = res; 832218043Spjd mode = pjdlog_mode_get(); 833211984Spjd 834218043Spjd /* Declare that we are sender. */ 835218043Spjd proto_send(res->hr_event, NULL, 0); 836218043Spjd /* Declare that we are receiver. */ 837218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 838218043Spjd descriptors_cleanup(res); 839204076Spjd 840218043Spjd pjdlog_init(mode); 841218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 842204076Spjd setproctitle("%s (primary)", res->hr_name); 843204076Spjd 844204076Spjd init_local(res); 845213007Spjd init_ggate(res); 846213007Spjd init_environment(res); 847217784Spjd 848213007Spjd /* 849213530Spjd * Create the guard thread first, so we can handle signals from the 850213530Spjd * very begining. 851213530Spjd */ 852213530Spjd error = pthread_create(&td, NULL, guard_thread, res); 853213530Spjd assert(error == 0); 854213530Spjd /* 855213007Spjd * Create the control thread before sending any event to the parent, 856213007Spjd * as we can deadlock when parent sends control request to worker, 857213007Spjd * but worker has no control thread started yet, so parent waits. 858213007Spjd * In the meantime worker sends an event to the parent, but parent 859213007Spjd * is unable to handle the event, because it waits for control 860213007Spjd * request response. 861213007Spjd */ 862213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 863213007Spjd assert(error == 0); 864210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 865205738Spjd sync_start(); 866204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 867204076Spjd assert(error == 0); 868204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 869204076Spjd assert(error == 0); 870204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 871204076Spjd assert(error == 0); 872204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 873204076Spjd assert(error == 0); 874204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 875204076Spjd assert(error == 0); 876213530Spjd (void)sync_thread(res); 877204076Spjd} 878204076Spjd 879204076Spjdstatic void 880204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 881204076Spjd{ 882204076Spjd char msg[1024]; 883204076Spjd va_list ap; 884204076Spjd int len; 885204076Spjd 886204076Spjd va_start(ap, fmt); 887204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 888204076Spjd va_end(ap); 889204076Spjd if ((size_t)len < sizeof(msg)) { 890204076Spjd switch (ggio->gctl_cmd) { 891204076Spjd case BIO_READ: 892204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 893204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 894204076Spjd (uintmax_t)ggio->gctl_length); 895204076Spjd break; 896204076Spjd case BIO_DELETE: 897204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 898204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 899204076Spjd (uintmax_t)ggio->gctl_length); 900204076Spjd break; 901204076Spjd case BIO_FLUSH: 902204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 903204076Spjd break; 904204076Spjd case BIO_WRITE: 905204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 906204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 907204076Spjd (uintmax_t)ggio->gctl_length); 908204076Spjd break; 909204076Spjd default: 910204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 911204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 912204076Spjd break; 913204076Spjd } 914204076Spjd } 915204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 916204076Spjd} 917204076Spjd 918204076Spjdstatic void 919204076Spjdremote_close(struct hast_resource *res, int ncomp) 920204076Spjd{ 921204076Spjd 922204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 923204076Spjd /* 924204076Spjd * A race is possible between dropping rlock and acquiring wlock - 925204076Spjd * another thread can close connection in-between. 926204076Spjd */ 927204076Spjd if (!ISCONNECTED(res, ncomp)) { 928204076Spjd assert(res->hr_remotein == NULL); 929204076Spjd assert(res->hr_remoteout == NULL); 930204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 931204076Spjd return; 932204076Spjd } 933204076Spjd 934204076Spjd assert(res->hr_remotein != NULL); 935204076Spjd assert(res->hr_remoteout != NULL); 936204076Spjd 937211881Spjd pjdlog_debug(2, "Closing incoming connection to %s.", 938204076Spjd res->hr_remoteaddr); 939204076Spjd proto_close(res->hr_remotein); 940204076Spjd res->hr_remotein = NULL; 941211881Spjd pjdlog_debug(2, "Closing outgoing connection to %s.", 942204076Spjd res->hr_remoteaddr); 943204076Spjd proto_close(res->hr_remoteout); 944204076Spjd res->hr_remoteout = NULL; 945204076Spjd 946204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 947204076Spjd 948211881Spjd pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 949211881Spjd 950204076Spjd /* 951204076Spjd * Stop synchronization if in-progress. 952204076Spjd */ 953211878Spjd sync_stop(); 954211984Spjd 955212038Spjd event_send(res, EVENT_DISCONNECT); 956204076Spjd} 957204076Spjd 958204076Spjd/* 959204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 960204076Spjd * appropriate threads: 961204076Spjd * WRITE - always goes to both local_send and remote_send threads 962204076Spjd * READ (when the block is up-to-date on local component) - 963204076Spjd * only local_send thread 964204076Spjd * READ (when the block isn't up-to-date on local component) - 965204076Spjd * only remote_send thread 966204076Spjd * DELETE - always goes to both local_send and remote_send threads 967204076Spjd * FLUSH - always goes to both local_send and remote_send threads 968204076Spjd */ 969204076Spjdstatic void * 970204076Spjdggate_recv_thread(void *arg) 971204076Spjd{ 972204076Spjd struct hast_resource *res = arg; 973204076Spjd struct g_gate_ctl_io *ggio; 974204076Spjd struct hio *hio; 975204076Spjd unsigned int ii, ncomp, ncomps; 976204076Spjd int error; 977204076Spjd 978204076Spjd ncomps = HAST_NCOMPONENTS; 979204076Spjd 980204076Spjd for (;;) { 981204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 982204076Spjd QUEUE_TAKE2(hio, free); 983204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 984204076Spjd ggio = &hio->hio_ggio; 985204076Spjd ggio->gctl_unit = res->hr_ggateunit; 986204076Spjd ggio->gctl_length = MAXPHYS; 987204076Spjd ggio->gctl_error = 0; 988204076Spjd pjdlog_debug(2, 989204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 990204076Spjd hio); 991204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 992204076Spjd if (sigexit_received) 993204076Spjd pthread_exit(NULL); 994204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 995204076Spjd } 996204076Spjd error = ggio->gctl_error; 997204076Spjd switch (error) { 998204076Spjd case 0: 999204076Spjd break; 1000204076Spjd case ECANCELED: 1001204076Spjd /* Exit gracefully. */ 1002204076Spjd if (!sigexit_received) { 1003204076Spjd pjdlog_debug(2, 1004204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 1005204076Spjd hio); 1006204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 1007204076Spjd } 1008204076Spjd pthread_exit(NULL); 1009204076Spjd case ENOMEM: 1010204076Spjd /* 1011204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 1012204076Spjd * bytes - request can't be bigger than that. 1013204076Spjd */ 1014204076Spjd /* FALLTHROUGH */ 1015204076Spjd case ENXIO: 1016204076Spjd default: 1017204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 1018204076Spjd strerror(error)); 1019204076Spjd } 1020204076Spjd for (ii = 0; ii < ncomps; ii++) 1021204076Spjd hio->hio_errors[ii] = EINVAL; 1022204076Spjd reqlog(LOG_DEBUG, 2, ggio, 1023204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 1024204076Spjd hio); 1025204076Spjd /* 1026204076Spjd * Inform all components about new write request. 1027204076Spjd * For read request prefer local component unless the given 1028204076Spjd * range is out-of-date, then use remote component. 1029204076Spjd */ 1030204076Spjd switch (ggio->gctl_cmd) { 1031204076Spjd case BIO_READ: 1032204076Spjd pjdlog_debug(2, 1033204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 1034204076Spjd hio); 1035204076Spjd refcount_init(&hio->hio_countdown, 1); 1036204076Spjd mtx_lock(&metadata_lock); 1037204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 1038204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1039204076Spjd /* 1040204076Spjd * This range is up-to-date on local component, 1041204076Spjd * so handle request locally. 1042204076Spjd */ 1043204076Spjd /* Local component is 0 for now. */ 1044204076Spjd ncomp = 0; 1045204076Spjd } else /* if (res->hr_syncsrc == 1046204076Spjd HAST_SYNCSRC_SECONDARY) */ { 1047204076Spjd assert(res->hr_syncsrc == 1048204076Spjd HAST_SYNCSRC_SECONDARY); 1049204076Spjd /* 1050204076Spjd * This range is out-of-date on local component, 1051204076Spjd * so send request to the remote node. 1052204076Spjd */ 1053204076Spjd /* Remote component is 1 for now. */ 1054204076Spjd ncomp = 1; 1055204076Spjd } 1056204076Spjd mtx_unlock(&metadata_lock); 1057204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1058204076Spjd break; 1059204076Spjd case BIO_WRITE: 1060214284Spjd if (res->hr_resuid == 0) { 1061214284Spjd /* This is first write, initialize resuid. */ 1062214284Spjd (void)init_resuid(res); 1063214284Spjd } 1064204076Spjd for (;;) { 1065204076Spjd mtx_lock(&range_lock); 1066204076Spjd if (rangelock_islocked(range_sync, 1067204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1068204076Spjd pjdlog_debug(2, 1069204076Spjd "regular: Range offset=%jd length=%zu locked.", 1070204076Spjd (intmax_t)ggio->gctl_offset, 1071204076Spjd (size_t)ggio->gctl_length); 1072204076Spjd range_regular_wait = true; 1073204076Spjd cv_wait(&range_regular_cond, &range_lock); 1074204076Spjd range_regular_wait = false; 1075204076Spjd mtx_unlock(&range_lock); 1076204076Spjd continue; 1077204076Spjd } 1078204076Spjd if (rangelock_add(range_regular, 1079204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1080204076Spjd mtx_unlock(&range_lock); 1081204076Spjd pjdlog_debug(2, 1082204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1083204076Spjd (intmax_t)ggio->gctl_offset, 1084204076Spjd (size_t)ggio->gctl_length); 1085204076Spjd sleep(1); 1086204076Spjd continue; 1087204076Spjd } 1088204076Spjd mtx_unlock(&range_lock); 1089204076Spjd break; 1090204076Spjd } 1091204076Spjd mtx_lock(&res->hr_amp_lock); 1092204076Spjd if (activemap_write_start(res->hr_amp, 1093204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1094204076Spjd (void)hast_activemap_flush(res); 1095204076Spjd } 1096204076Spjd mtx_unlock(&res->hr_amp_lock); 1097204076Spjd /* FALLTHROUGH */ 1098204076Spjd case BIO_DELETE: 1099204076Spjd case BIO_FLUSH: 1100204076Spjd pjdlog_debug(2, 1101204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1102204076Spjd hio); 1103204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1104204076Spjd for (ii = 0; ii < ncomps; ii++) 1105204076Spjd QUEUE_INSERT1(hio, send, ii); 1106204076Spjd break; 1107204076Spjd } 1108204076Spjd } 1109204076Spjd /* NOTREACHED */ 1110204076Spjd return (NULL); 1111204076Spjd} 1112204076Spjd 1113204076Spjd/* 1114204076Spjd * Thread reads from or writes to local component. 1115204076Spjd * If local read fails, it redirects it to remote_send thread. 1116204076Spjd */ 1117204076Spjdstatic void * 1118204076Spjdlocal_send_thread(void *arg) 1119204076Spjd{ 1120204076Spjd struct hast_resource *res = arg; 1121204076Spjd struct g_gate_ctl_io *ggio; 1122204076Spjd struct hio *hio; 1123204076Spjd unsigned int ncomp, rncomp; 1124204076Spjd ssize_t ret; 1125204076Spjd 1126204076Spjd /* Local component is 0 for now. */ 1127204076Spjd ncomp = 0; 1128204076Spjd /* Remote component is 1 for now. */ 1129204076Spjd rncomp = 1; 1130204076Spjd 1131204076Spjd for (;;) { 1132204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1133214692Spjd QUEUE_TAKE1(hio, send, ncomp, 0); 1134204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1135204076Spjd ggio = &hio->hio_ggio; 1136204076Spjd switch (ggio->gctl_cmd) { 1137204076Spjd case BIO_READ: 1138204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1139204076Spjd ggio->gctl_length, 1140204076Spjd ggio->gctl_offset + res->hr_localoff); 1141204076Spjd if (ret == ggio->gctl_length) 1142204076Spjd hio->hio_errors[ncomp] = 0; 1143204076Spjd else { 1144204076Spjd /* 1145204076Spjd * If READ failed, try to read from remote node. 1146204076Spjd */ 1147216479Spjd if (ret < 0) { 1148216479Spjd reqlog(LOG_WARNING, 0, ggio, 1149216479Spjd "Local request failed (%s), trying remote node. ", 1150216479Spjd strerror(errno)); 1151216479Spjd } else if (ret != ggio->gctl_length) { 1152216479Spjd reqlog(LOG_WARNING, 0, ggio, 1153216479Spjd "Local request failed (%zd != %jd), trying remote node. ", 1154216494Spjd ret, (intmax_t)ggio->gctl_length); 1155216479Spjd } 1156204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1157204076Spjd continue; 1158204076Spjd } 1159204076Spjd break; 1160204076Spjd case BIO_WRITE: 1161204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1162204076Spjd ggio->gctl_length, 1163204076Spjd ggio->gctl_offset + res->hr_localoff); 1164216479Spjd if (ret < 0) { 1165204076Spjd hio->hio_errors[ncomp] = errno; 1166216479Spjd reqlog(LOG_WARNING, 0, ggio, 1167216479Spjd "Local request failed (%s): ", 1168216479Spjd strerror(errno)); 1169216479Spjd } else if (ret != ggio->gctl_length) { 1170204076Spjd hio->hio_errors[ncomp] = EIO; 1171216479Spjd reqlog(LOG_WARNING, 0, ggio, 1172216479Spjd "Local request failed (%zd != %jd): ", 1173216494Spjd ret, (intmax_t)ggio->gctl_length); 1174216479Spjd } else { 1175204076Spjd hio->hio_errors[ncomp] = 0; 1176216479Spjd } 1177204076Spjd break; 1178204076Spjd case BIO_DELETE: 1179204076Spjd ret = g_delete(res->hr_localfd, 1180204076Spjd ggio->gctl_offset + res->hr_localoff, 1181204076Spjd ggio->gctl_length); 1182216479Spjd if (ret < 0) { 1183204076Spjd hio->hio_errors[ncomp] = errno; 1184216479Spjd reqlog(LOG_WARNING, 0, ggio, 1185216479Spjd "Local request failed (%s): ", 1186216479Spjd strerror(errno)); 1187216479Spjd } else { 1188204076Spjd hio->hio_errors[ncomp] = 0; 1189216479Spjd } 1190204076Spjd break; 1191204076Spjd case BIO_FLUSH: 1192204076Spjd ret = g_flush(res->hr_localfd); 1193216479Spjd if (ret < 0) { 1194204076Spjd hio->hio_errors[ncomp] = errno; 1195216479Spjd reqlog(LOG_WARNING, 0, ggio, 1196216479Spjd "Local request failed (%s): ", 1197216479Spjd strerror(errno)); 1198216479Spjd } else { 1199204076Spjd hio->hio_errors[ncomp] = 0; 1200216479Spjd } 1201204076Spjd break; 1202204076Spjd } 1203204076Spjd if (refcount_release(&hio->hio_countdown)) { 1204204076Spjd if (ISSYNCREQ(hio)) { 1205204076Spjd mtx_lock(&sync_lock); 1206204076Spjd SYNCREQDONE(hio); 1207204076Spjd mtx_unlock(&sync_lock); 1208204076Spjd cv_signal(&sync_cond); 1209204076Spjd } else { 1210204076Spjd pjdlog_debug(2, 1211204076Spjd "local_send: (%p) Moving request to the done queue.", 1212204076Spjd hio); 1213204076Spjd QUEUE_INSERT2(hio, done); 1214204076Spjd } 1215204076Spjd } 1216204076Spjd } 1217204076Spjd /* NOTREACHED */ 1218204076Spjd return (NULL); 1219204076Spjd} 1220204076Spjd 1221214692Spjdstatic void 1222214692Spjdkeepalive_send(struct hast_resource *res, unsigned int ncomp) 1223214692Spjd{ 1224214692Spjd struct nv *nv; 1225214692Spjd 1226214692Spjd if (!ISCONNECTED(res, ncomp)) 1227214692Spjd return; 1228214692Spjd 1229214692Spjd assert(res->hr_remotein != NULL); 1230214692Spjd assert(res->hr_remoteout != NULL); 1231214692Spjd 1232214692Spjd nv = nv_alloc(); 1233214692Spjd nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1234214692Spjd if (nv_error(nv) != 0) { 1235214692Spjd nv_free(nv); 1236214692Spjd pjdlog_debug(1, 1237214692Spjd "keepalive_send: Unable to prepare header to send."); 1238214692Spjd return; 1239214692Spjd } 1240214692Spjd if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1241214692Spjd pjdlog_common(LOG_DEBUG, 1, errno, 1242214692Spjd "keepalive_send: Unable to send request"); 1243214692Spjd nv_free(nv); 1244214692Spjd rw_unlock(&hio_remote_lock[ncomp]); 1245214692Spjd remote_close(res, ncomp); 1246214692Spjd rw_rlock(&hio_remote_lock[ncomp]); 1247214692Spjd return; 1248214692Spjd } 1249214692Spjd nv_free(nv); 1250214692Spjd pjdlog_debug(2, "keepalive_send: Request sent."); 1251214692Spjd} 1252214692Spjd 1253204076Spjd/* 1254204076Spjd * Thread sends request to secondary node. 1255204076Spjd */ 1256204076Spjdstatic void * 1257204076Spjdremote_send_thread(void *arg) 1258204076Spjd{ 1259204076Spjd struct hast_resource *res = arg; 1260204076Spjd struct g_gate_ctl_io *ggio; 1261214692Spjd time_t lastcheck, now; 1262204076Spjd struct hio *hio; 1263204076Spjd struct nv *nv; 1264204076Spjd unsigned int ncomp; 1265204076Spjd bool wakeup; 1266204076Spjd uint64_t offset, length; 1267204076Spjd uint8_t cmd; 1268204076Spjd void *data; 1269204076Spjd 1270204076Spjd /* Remote component is 1 for now. */ 1271204076Spjd ncomp = 1; 1272214692Spjd lastcheck = time(NULL); 1273204076Spjd 1274204076Spjd for (;;) { 1275204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1276214692Spjd QUEUE_TAKE1(hio, send, ncomp, RETRY_SLEEP); 1277214692Spjd if (hio == NULL) { 1278214692Spjd now = time(NULL); 1279214692Spjd if (lastcheck + RETRY_SLEEP <= now) { 1280214692Spjd keepalive_send(res, ncomp); 1281214692Spjd lastcheck = now; 1282214692Spjd } 1283214692Spjd continue; 1284214692Spjd } 1285204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1286204076Spjd ggio = &hio->hio_ggio; 1287204076Spjd switch (ggio->gctl_cmd) { 1288204076Spjd case BIO_READ: 1289204076Spjd cmd = HIO_READ; 1290204076Spjd data = NULL; 1291204076Spjd offset = ggio->gctl_offset; 1292204076Spjd length = ggio->gctl_length; 1293204076Spjd break; 1294204076Spjd case BIO_WRITE: 1295204076Spjd cmd = HIO_WRITE; 1296204076Spjd data = ggio->gctl_data; 1297204076Spjd offset = ggio->gctl_offset; 1298204076Spjd length = ggio->gctl_length; 1299204076Spjd break; 1300204076Spjd case BIO_DELETE: 1301204076Spjd cmd = HIO_DELETE; 1302204076Spjd data = NULL; 1303204076Spjd offset = ggio->gctl_offset; 1304204076Spjd length = ggio->gctl_length; 1305204076Spjd break; 1306204076Spjd case BIO_FLUSH: 1307204076Spjd cmd = HIO_FLUSH; 1308204076Spjd data = NULL; 1309204076Spjd offset = 0; 1310204076Spjd length = 0; 1311204076Spjd break; 1312204076Spjd default: 1313204076Spjd assert(!"invalid condition"); 1314204076Spjd abort(); 1315204076Spjd } 1316204076Spjd nv = nv_alloc(); 1317204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1318204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1319204076Spjd nv_add_uint64(nv, offset, "offset"); 1320204076Spjd nv_add_uint64(nv, length, "length"); 1321204076Spjd if (nv_error(nv) != 0) { 1322204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1323204076Spjd pjdlog_debug(2, 1324204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1325204076Spjd hio); 1326204076Spjd reqlog(LOG_ERR, 0, ggio, 1327204076Spjd "Unable to prepare header to send (%s): ", 1328204076Spjd strerror(nv_error(nv))); 1329204076Spjd /* Move failed request immediately to the done queue. */ 1330204076Spjd goto done_queue; 1331204076Spjd } 1332204076Spjd pjdlog_debug(2, 1333204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1334204076Spjd hio); 1335204076Spjd /* 1336204076Spjd * Protect connection from disappearing. 1337204076Spjd */ 1338204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1339204076Spjd if (!ISCONNECTED(res, ncomp)) { 1340204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1341204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1342204076Spjd goto done_queue; 1343204076Spjd } 1344204076Spjd /* 1345204076Spjd * Move the request to recv queue before sending it, because 1346204076Spjd * in different order we can get reply before we move request 1347204076Spjd * to recv queue. 1348204076Spjd */ 1349204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1350204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1351204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1352204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1353204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1354204076Spjd data != NULL ? length : 0) < 0) { 1355204076Spjd hio->hio_errors[ncomp] = errno; 1356204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1357204076Spjd pjdlog_debug(2, 1358204076Spjd "remote_send: (%p) Unable to send request.", hio); 1359204076Spjd reqlog(LOG_ERR, 0, ggio, 1360204076Spjd "Unable to send request (%s): ", 1361204076Spjd strerror(hio->hio_errors[ncomp])); 1362211979Spjd remote_close(res, ncomp); 1363204076Spjd /* 1364204076Spjd * Take request back from the receive queue and move 1365204076Spjd * it immediately to the done queue. 1366204076Spjd */ 1367204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1368204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1369204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1370204076Spjd goto done_queue; 1371204076Spjd } 1372204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1373204076Spjd nv_free(nv); 1374204076Spjd if (wakeup) 1375204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1376204076Spjd continue; 1377204076Spjddone_queue: 1378204076Spjd nv_free(nv); 1379204076Spjd if (ISSYNCREQ(hio)) { 1380204076Spjd if (!refcount_release(&hio->hio_countdown)) 1381204076Spjd continue; 1382204076Spjd mtx_lock(&sync_lock); 1383204076Spjd SYNCREQDONE(hio); 1384204076Spjd mtx_unlock(&sync_lock); 1385204076Spjd cv_signal(&sync_cond); 1386204076Spjd continue; 1387204076Spjd } 1388204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1389204076Spjd mtx_lock(&res->hr_amp_lock); 1390204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1391204076Spjd ggio->gctl_length)) { 1392204076Spjd (void)hast_activemap_flush(res); 1393204076Spjd } 1394204076Spjd mtx_unlock(&res->hr_amp_lock); 1395204076Spjd } 1396204076Spjd if (!refcount_release(&hio->hio_countdown)) 1397204076Spjd continue; 1398204076Spjd pjdlog_debug(2, 1399204076Spjd "remote_send: (%p) Moving request to the done queue.", 1400204076Spjd hio); 1401204076Spjd QUEUE_INSERT2(hio, done); 1402204076Spjd } 1403204076Spjd /* NOTREACHED */ 1404204076Spjd return (NULL); 1405204076Spjd} 1406204076Spjd 1407204076Spjd/* 1408204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1409204076Spjd * thread. 1410204076Spjd */ 1411204076Spjdstatic void * 1412204076Spjdremote_recv_thread(void *arg) 1413204076Spjd{ 1414204076Spjd struct hast_resource *res = arg; 1415204076Spjd struct g_gate_ctl_io *ggio; 1416204076Spjd struct hio *hio; 1417204076Spjd struct nv *nv; 1418204076Spjd unsigned int ncomp; 1419204076Spjd uint64_t seq; 1420204076Spjd int error; 1421204076Spjd 1422204076Spjd /* Remote component is 1 for now. */ 1423204076Spjd ncomp = 1; 1424204076Spjd 1425204076Spjd for (;;) { 1426204076Spjd /* Wait until there is anything to receive. */ 1427204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1428204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1429204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1430204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1431204076Spjd &hio_recv_list_lock[ncomp]); 1432204076Spjd } 1433204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1434204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1435204076Spjd if (!ISCONNECTED(res, ncomp)) { 1436204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1437204076Spjd /* 1438204076Spjd * Connection is dead, so move all pending requests to 1439204076Spjd * the done queue (one-by-one). 1440204076Spjd */ 1441204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1442204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1443204076Spjd assert(hio != NULL); 1444204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1445204076Spjd hio_next[ncomp]); 1446204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1447204076Spjd goto done_queue; 1448204076Spjd } 1449204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1450204076Spjd pjdlog_errno(LOG_ERR, 1451204076Spjd "Unable to receive reply header"); 1452204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1453204076Spjd remote_close(res, ncomp); 1454204076Spjd continue; 1455204076Spjd } 1456204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1457204076Spjd seq = nv_get_uint64(nv, "seq"); 1458204076Spjd if (seq == 0) { 1459204076Spjd pjdlog_error("Header contains no 'seq' field."); 1460204076Spjd nv_free(nv); 1461204076Spjd continue; 1462204076Spjd } 1463204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1464204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1465204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1466204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1467204076Spjd hio_next[ncomp]); 1468204076Spjd break; 1469204076Spjd } 1470204076Spjd } 1471204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1472204076Spjd if (hio == NULL) { 1473204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1474204076Spjd (uintmax_t)seq); 1475204076Spjd nv_free(nv); 1476204076Spjd continue; 1477204076Spjd } 1478204076Spjd error = nv_get_int16(nv, "error"); 1479204076Spjd if (error != 0) { 1480204076Spjd /* Request failed on remote side. */ 1481216478Spjd hio->hio_errors[ncomp] = error; 1482216479Spjd reqlog(LOG_WARNING, 0, &hio->hio_ggio, 1483216479Spjd "Remote request failed (%s): ", strerror(error)); 1484204076Spjd nv_free(nv); 1485204076Spjd goto done_queue; 1486204076Spjd } 1487204076Spjd ggio = &hio->hio_ggio; 1488204076Spjd switch (ggio->gctl_cmd) { 1489204076Spjd case BIO_READ: 1490204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1491204076Spjd if (!ISCONNECTED(res, ncomp)) { 1492204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1493204076Spjd nv_free(nv); 1494204076Spjd goto done_queue; 1495204076Spjd } 1496204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1497204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1498204076Spjd hio->hio_errors[ncomp] = errno; 1499204076Spjd pjdlog_errno(LOG_ERR, 1500204076Spjd "Unable to receive reply data"); 1501204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1502204076Spjd nv_free(nv); 1503204076Spjd remote_close(res, ncomp); 1504204076Spjd goto done_queue; 1505204076Spjd } 1506204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1507204076Spjd break; 1508204076Spjd case BIO_WRITE: 1509204076Spjd case BIO_DELETE: 1510204076Spjd case BIO_FLUSH: 1511204076Spjd break; 1512204076Spjd default: 1513204076Spjd assert(!"invalid condition"); 1514204076Spjd abort(); 1515204076Spjd } 1516204076Spjd hio->hio_errors[ncomp] = 0; 1517204076Spjd nv_free(nv); 1518204076Spjddone_queue: 1519204076Spjd if (refcount_release(&hio->hio_countdown)) { 1520204076Spjd if (ISSYNCREQ(hio)) { 1521204076Spjd mtx_lock(&sync_lock); 1522204076Spjd SYNCREQDONE(hio); 1523204076Spjd mtx_unlock(&sync_lock); 1524204076Spjd cv_signal(&sync_cond); 1525204076Spjd } else { 1526204076Spjd pjdlog_debug(2, 1527204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1528204076Spjd hio); 1529204076Spjd QUEUE_INSERT2(hio, done); 1530204076Spjd } 1531204076Spjd } 1532204076Spjd } 1533204076Spjd /* NOTREACHED */ 1534204076Spjd return (NULL); 1535204076Spjd} 1536204076Spjd 1537204076Spjd/* 1538204076Spjd * Thread sends answer to the kernel. 1539204076Spjd */ 1540204076Spjdstatic void * 1541204076Spjdggate_send_thread(void *arg) 1542204076Spjd{ 1543204076Spjd struct hast_resource *res = arg; 1544204076Spjd struct g_gate_ctl_io *ggio; 1545204076Spjd struct hio *hio; 1546204076Spjd unsigned int ii, ncomp, ncomps; 1547204076Spjd 1548204076Spjd ncomps = HAST_NCOMPONENTS; 1549204076Spjd 1550204076Spjd for (;;) { 1551204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1552204076Spjd QUEUE_TAKE2(hio, done); 1553204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1554204076Spjd ggio = &hio->hio_ggio; 1555204076Spjd for (ii = 0; ii < ncomps; ii++) { 1556204076Spjd if (hio->hio_errors[ii] == 0) { 1557204076Spjd /* 1558204076Spjd * One successful request is enough to declare 1559204076Spjd * success. 1560204076Spjd */ 1561204076Spjd ggio->gctl_error = 0; 1562204076Spjd break; 1563204076Spjd } 1564204076Spjd } 1565204076Spjd if (ii == ncomps) { 1566204076Spjd /* 1567204076Spjd * None of the requests were successful. 1568204076Spjd * Use first error. 1569204076Spjd */ 1570204076Spjd ggio->gctl_error = hio->hio_errors[0]; 1571204076Spjd } 1572204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1573204076Spjd mtx_lock(&res->hr_amp_lock); 1574204076Spjd activemap_write_complete(res->hr_amp, 1575204076Spjd ggio->gctl_offset, ggio->gctl_length); 1576204076Spjd mtx_unlock(&res->hr_amp_lock); 1577204076Spjd } 1578204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1579204076Spjd /* 1580204076Spjd * Unlock range we locked. 1581204076Spjd */ 1582204076Spjd mtx_lock(&range_lock); 1583204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1584204076Spjd ggio->gctl_length); 1585204076Spjd if (range_sync_wait) 1586204076Spjd cv_signal(&range_sync_cond); 1587204076Spjd mtx_unlock(&range_lock); 1588204076Spjd /* 1589204076Spjd * Bump local count if this is first write after 1590204076Spjd * connection failure with remote node. 1591204076Spjd */ 1592204076Spjd ncomp = 1; 1593204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1594204076Spjd if (!ISCONNECTED(res, ncomp)) { 1595204076Spjd mtx_lock(&metadata_lock); 1596204076Spjd if (res->hr_primary_localcnt == 1597204076Spjd res->hr_secondary_remotecnt) { 1598204076Spjd res->hr_primary_localcnt++; 1599204076Spjd pjdlog_debug(1, 1600204076Spjd "Increasing localcnt to %ju.", 1601204076Spjd (uintmax_t)res->hr_primary_localcnt); 1602204076Spjd (void)metadata_write(res); 1603204076Spjd } 1604204076Spjd mtx_unlock(&metadata_lock); 1605204076Spjd } 1606204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1607204076Spjd } 1608204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1609204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1610204076Spjd pjdlog_debug(2, 1611204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1612204076Spjd QUEUE_INSERT2(hio, free); 1613204076Spjd } 1614204076Spjd /* NOTREACHED */ 1615204076Spjd return (NULL); 1616204076Spjd} 1617204076Spjd 1618204076Spjd/* 1619204076Spjd * Thread synchronize local and remote components. 1620204076Spjd */ 1621204076Spjdstatic void * 1622204076Spjdsync_thread(void *arg __unused) 1623204076Spjd{ 1624204076Spjd struct hast_resource *res = arg; 1625204076Spjd struct hio *hio; 1626204076Spjd struct g_gate_ctl_io *ggio; 1627204076Spjd unsigned int ii, ncomp, ncomps; 1628204076Spjd off_t offset, length, synced; 1629204076Spjd bool dorewind; 1630204076Spjd int syncext; 1631204076Spjd 1632204076Spjd ncomps = HAST_NCOMPONENTS; 1633204076Spjd dorewind = true; 1634211897Spjd synced = 0; 1635211897Spjd offset = -1; 1636204076Spjd 1637204076Spjd for (;;) { 1638204076Spjd mtx_lock(&sync_lock); 1639211897Spjd if (offset >= 0 && !sync_inprogress) { 1640211879Spjd pjdlog_info("Synchronization interrupted. " 1641211879Spjd "%jd bytes synchronized so far.", 1642211879Spjd (intmax_t)synced); 1643212038Spjd event_send(res, EVENT_SYNCINTR); 1644211879Spjd } 1645204076Spjd while (!sync_inprogress) { 1646204076Spjd dorewind = true; 1647204076Spjd synced = 0; 1648204076Spjd cv_wait(&sync_cond, &sync_lock); 1649204076Spjd } 1650204076Spjd mtx_unlock(&sync_lock); 1651204076Spjd /* 1652204076Spjd * Obtain offset at which we should synchronize. 1653204076Spjd * Rewind synchronization if needed. 1654204076Spjd */ 1655204076Spjd mtx_lock(&res->hr_amp_lock); 1656204076Spjd if (dorewind) 1657204076Spjd activemap_sync_rewind(res->hr_amp); 1658204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1659204076Spjd if (syncext != -1) { 1660204076Spjd /* 1661204076Spjd * We synchronized entire syncext extent, we can mark 1662204076Spjd * it as clean now. 1663204076Spjd */ 1664204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1665204076Spjd (void)hast_activemap_flush(res); 1666204076Spjd } 1667204076Spjd mtx_unlock(&res->hr_amp_lock); 1668204076Spjd if (dorewind) { 1669204076Spjd dorewind = false; 1670204076Spjd if (offset < 0) 1671204076Spjd pjdlog_info("Nodes are in sync."); 1672204076Spjd else { 1673204076Spjd pjdlog_info("Synchronization started. %ju bytes to go.", 1674204076Spjd (uintmax_t)(res->hr_extentsize * 1675204076Spjd activemap_ndirty(res->hr_amp))); 1676212038Spjd event_send(res, EVENT_SYNCSTART); 1677204076Spjd } 1678204076Spjd } 1679204076Spjd if (offset < 0) { 1680211878Spjd sync_stop(); 1681204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1682204076Spjd /* 1683204076Spjd * Synchronization complete, make both localcnt and 1684204076Spjd * remotecnt equal. 1685204076Spjd */ 1686204076Spjd ncomp = 1; 1687204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1688204076Spjd if (ISCONNECTED(res, ncomp)) { 1689204076Spjd if (synced > 0) { 1690204076Spjd pjdlog_info("Synchronization complete. " 1691204076Spjd "%jd bytes synchronized.", 1692204076Spjd (intmax_t)synced); 1693212038Spjd event_send(res, EVENT_SYNCDONE); 1694204076Spjd } 1695204076Spjd mtx_lock(&metadata_lock); 1696204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1697204076Spjd res->hr_primary_localcnt = 1698204076Spjd res->hr_secondary_localcnt; 1699204076Spjd res->hr_primary_remotecnt = 1700204076Spjd res->hr_secondary_remotecnt; 1701204076Spjd pjdlog_debug(1, 1702204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1703204076Spjd (uintmax_t)res->hr_primary_localcnt, 1704204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1705204076Spjd (void)metadata_write(res); 1706204076Spjd mtx_unlock(&metadata_lock); 1707204076Spjd } 1708204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1709204076Spjd continue; 1710204076Spjd } 1711204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1712204076Spjd QUEUE_TAKE2(hio, free); 1713204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1714204076Spjd /* 1715204076Spjd * Lock the range we are going to synchronize. We don't want 1716204076Spjd * race where someone writes between our read and write. 1717204076Spjd */ 1718204076Spjd for (;;) { 1719204076Spjd mtx_lock(&range_lock); 1720204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1721204076Spjd pjdlog_debug(2, 1722204076Spjd "sync: Range offset=%jd length=%jd locked.", 1723204076Spjd (intmax_t)offset, (intmax_t)length); 1724204076Spjd range_sync_wait = true; 1725204076Spjd cv_wait(&range_sync_cond, &range_lock); 1726204076Spjd range_sync_wait = false; 1727204076Spjd mtx_unlock(&range_lock); 1728204076Spjd continue; 1729204076Spjd } 1730204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1731204076Spjd mtx_unlock(&range_lock); 1732204076Spjd pjdlog_debug(2, 1733204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1734204076Spjd (intmax_t)offset, (intmax_t)length); 1735204076Spjd sleep(1); 1736204076Spjd continue; 1737204076Spjd } 1738204076Spjd mtx_unlock(&range_lock); 1739204076Spjd break; 1740204076Spjd } 1741204076Spjd /* 1742204076Spjd * First read the data from synchronization source. 1743204076Spjd */ 1744204076Spjd SYNCREQ(hio); 1745204076Spjd ggio = &hio->hio_ggio; 1746204076Spjd ggio->gctl_cmd = BIO_READ; 1747204076Spjd ggio->gctl_offset = offset; 1748204076Spjd ggio->gctl_length = length; 1749204076Spjd ggio->gctl_error = 0; 1750204076Spjd for (ii = 0; ii < ncomps; ii++) 1751204076Spjd hio->hio_errors[ii] = EINVAL; 1752204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1753204076Spjd hio); 1754204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1755204076Spjd hio); 1756204076Spjd mtx_lock(&metadata_lock); 1757204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1758204076Spjd /* 1759204076Spjd * This range is up-to-date on local component, 1760204076Spjd * so handle request locally. 1761204076Spjd */ 1762204076Spjd /* Local component is 0 for now. */ 1763204076Spjd ncomp = 0; 1764204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1765204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1766204076Spjd /* 1767204076Spjd * This range is out-of-date on local component, 1768204076Spjd * so send request to the remote node. 1769204076Spjd */ 1770204076Spjd /* Remote component is 1 for now. */ 1771204076Spjd ncomp = 1; 1772204076Spjd } 1773204076Spjd mtx_unlock(&metadata_lock); 1774204076Spjd refcount_init(&hio->hio_countdown, 1); 1775204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1776204076Spjd 1777204076Spjd /* 1778204076Spjd * Let's wait for READ to finish. 1779204076Spjd */ 1780204076Spjd mtx_lock(&sync_lock); 1781204076Spjd while (!ISSYNCREQDONE(hio)) 1782204076Spjd cv_wait(&sync_cond, &sync_lock); 1783204076Spjd mtx_unlock(&sync_lock); 1784204076Spjd 1785204076Spjd if (hio->hio_errors[ncomp] != 0) { 1786204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1787204076Spjd strerror(hio->hio_errors[ncomp])); 1788204076Spjd goto free_queue; 1789204076Spjd } 1790204076Spjd 1791204076Spjd /* 1792204076Spjd * We read the data from synchronization source, now write it 1793204076Spjd * to synchronization target. 1794204076Spjd */ 1795204076Spjd SYNCREQ(hio); 1796204076Spjd ggio->gctl_cmd = BIO_WRITE; 1797204076Spjd for (ii = 0; ii < ncomps; ii++) 1798204076Spjd hio->hio_errors[ii] = EINVAL; 1799204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1800204076Spjd hio); 1801204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1802204076Spjd hio); 1803204076Spjd mtx_lock(&metadata_lock); 1804204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1805204076Spjd /* 1806204076Spjd * This range is up-to-date on local component, 1807204076Spjd * so we update remote component. 1808204076Spjd */ 1809204076Spjd /* Remote component is 1 for now. */ 1810204076Spjd ncomp = 1; 1811204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1812204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1813204076Spjd /* 1814204076Spjd * This range is out-of-date on local component, 1815204076Spjd * so we update it. 1816204076Spjd */ 1817204076Spjd /* Local component is 0 for now. */ 1818204076Spjd ncomp = 0; 1819204076Spjd } 1820204076Spjd mtx_unlock(&metadata_lock); 1821204076Spjd 1822204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1823204076Spjd hio); 1824204076Spjd refcount_init(&hio->hio_countdown, 1); 1825204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1826204076Spjd 1827204076Spjd /* 1828204076Spjd * Let's wait for WRITE to finish. 1829204076Spjd */ 1830204076Spjd mtx_lock(&sync_lock); 1831204076Spjd while (!ISSYNCREQDONE(hio)) 1832204076Spjd cv_wait(&sync_cond, &sync_lock); 1833204076Spjd mtx_unlock(&sync_lock); 1834204076Spjd 1835204076Spjd if (hio->hio_errors[ncomp] != 0) { 1836204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1837204076Spjd strerror(hio->hio_errors[ncomp])); 1838204076Spjd goto free_queue; 1839204076Spjd } 1840211880Spjd 1841211880Spjd synced += length; 1842204076Spjdfree_queue: 1843204076Spjd mtx_lock(&range_lock); 1844204076Spjd rangelock_del(range_sync, offset, length); 1845204076Spjd if (range_regular_wait) 1846204076Spjd cv_signal(&range_regular_cond); 1847204076Spjd mtx_unlock(&range_lock); 1848204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1849204076Spjd hio); 1850204076Spjd QUEUE_INSERT2(hio, free); 1851204076Spjd } 1852204076Spjd /* NOTREACHED */ 1853204076Spjd return (NULL); 1854204076Spjd} 1855204076Spjd 1856217784Spjdvoid 1857217784Spjdprimary_config_reload(struct hast_resource *res, struct nv *nv) 1858210886Spjd{ 1859210886Spjd unsigned int ii, ncomps; 1860217784Spjd int modified, vint; 1861217784Spjd const char *vstr; 1862210886Spjd 1863210886Spjd pjdlog_info("Reloading configuration..."); 1864210886Spjd 1865217784Spjd assert(res->hr_role == HAST_ROLE_PRIMARY); 1866217784Spjd assert(gres == res); 1867217784Spjd nv_assert(nv, "remoteaddr"); 1868217784Spjd nv_assert(nv, "replication"); 1869217784Spjd nv_assert(nv, "timeout"); 1870217784Spjd nv_assert(nv, "exec"); 1871217784Spjd 1872210886Spjd ncomps = HAST_NCOMPONENTS; 1873210886Spjd 1874210886Spjd#define MODIFIED_REMOTEADDR 0x1 1875210886Spjd#define MODIFIED_REPLICATION 0x2 1876210886Spjd#define MODIFIED_TIMEOUT 0x4 1877211886Spjd#define MODIFIED_EXEC 0x8 1878210886Spjd modified = 0; 1879217784Spjd 1880217784Spjd vstr = nv_get_string(nv, "remoteaddr"); 1881217784Spjd if (strcmp(gres->hr_remoteaddr, vstr) != 0) { 1882210886Spjd /* 1883210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1884210886Spjd * We want remote_close() to log disconnect from the old 1885210886Spjd * addresses, not from the new ones. 1886210886Spjd */ 1887210886Spjd modified |= MODIFIED_REMOTEADDR; 1888210886Spjd } 1889217784Spjd vint = nv_get_int32(nv, "replication"); 1890217784Spjd if (gres->hr_replication != vint) { 1891217784Spjd gres->hr_replication = vint; 1892210886Spjd modified |= MODIFIED_REPLICATION; 1893210886Spjd } 1894217784Spjd vint = nv_get_int32(nv, "timeout"); 1895217784Spjd if (gres->hr_timeout != vint) { 1896217784Spjd gres->hr_timeout = vint; 1897210886Spjd modified |= MODIFIED_TIMEOUT; 1898210886Spjd } 1899217784Spjd vstr = nv_get_string(nv, "exec"); 1900217784Spjd if (strcmp(gres->hr_exec, vstr) != 0) { 1901217784Spjd strlcpy(gres->hr_exec, vstr, sizeof(gres->hr_exec)); 1902211886Spjd modified |= MODIFIED_EXEC; 1903211886Spjd } 1904217784Spjd 1905210886Spjd /* 1906210886Spjd * If only timeout was modified we only need to change it without 1907210886Spjd * reconnecting. 1908210886Spjd */ 1909210886Spjd if (modified == MODIFIED_TIMEOUT) { 1910210886Spjd for (ii = 0; ii < ncomps; ii++) { 1911210886Spjd if (!ISREMOTE(ii)) 1912210886Spjd continue; 1913210886Spjd rw_rlock(&hio_remote_lock[ii]); 1914210886Spjd if (!ISCONNECTED(gres, ii)) { 1915210886Spjd rw_unlock(&hio_remote_lock[ii]); 1916210886Spjd continue; 1917210886Spjd } 1918210886Spjd rw_unlock(&hio_remote_lock[ii]); 1919210886Spjd if (proto_timeout(gres->hr_remotein, 1920210886Spjd gres->hr_timeout) < 0) { 1921210886Spjd pjdlog_errno(LOG_WARNING, 1922210886Spjd "Unable to set connection timeout"); 1923210886Spjd } 1924210886Spjd if (proto_timeout(gres->hr_remoteout, 1925210886Spjd gres->hr_timeout) < 0) { 1926210886Spjd pjdlog_errno(LOG_WARNING, 1927210886Spjd "Unable to set connection timeout"); 1928210886Spjd } 1929210886Spjd } 1930211886Spjd } else if ((modified & 1931211886Spjd (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1932210886Spjd for (ii = 0; ii < ncomps; ii++) { 1933210886Spjd if (!ISREMOTE(ii)) 1934210886Spjd continue; 1935210886Spjd remote_close(gres, ii); 1936210886Spjd } 1937210886Spjd if (modified & MODIFIED_REMOTEADDR) { 1938217784Spjd vstr = nv_get_string(nv, "remoteaddr"); 1939217784Spjd strlcpy(gres->hr_remoteaddr, vstr, 1940210886Spjd sizeof(gres->hr_remoteaddr)); 1941210886Spjd } 1942210886Spjd } 1943210886Spjd#undef MODIFIED_REMOTEADDR 1944210886Spjd#undef MODIFIED_REPLICATION 1945210886Spjd#undef MODIFIED_TIMEOUT 1946211886Spjd#undef MODIFIED_EXEC 1947210886Spjd 1948210886Spjd pjdlog_info("Configuration reloaded successfully."); 1949210886Spjd} 1950210886Spjd 1951211882Spjdstatic void 1952211981Spjdguard_one(struct hast_resource *res, unsigned int ncomp) 1953211981Spjd{ 1954211981Spjd struct proto_conn *in, *out; 1955211981Spjd 1956211981Spjd if (!ISREMOTE(ncomp)) 1957211981Spjd return; 1958211981Spjd 1959211981Spjd rw_rlock(&hio_remote_lock[ncomp]); 1960211981Spjd 1961211981Spjd if (!real_remote(res)) { 1962211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1963211981Spjd return; 1964211981Spjd } 1965211981Spjd 1966211981Spjd if (ISCONNECTED(res, ncomp)) { 1967211981Spjd assert(res->hr_remotein != NULL); 1968211981Spjd assert(res->hr_remoteout != NULL); 1969211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1970211981Spjd pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 1971211981Spjd res->hr_remoteaddr); 1972211981Spjd return; 1973211981Spjd } 1974211981Spjd 1975211981Spjd assert(res->hr_remotein == NULL); 1976211981Spjd assert(res->hr_remoteout == NULL); 1977211981Spjd /* 1978211981Spjd * Upgrade the lock. It doesn't have to be atomic as no other thread 1979211981Spjd * can change connection status from disconnected to connected. 1980211981Spjd */ 1981211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1982211981Spjd pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 1983211981Spjd res->hr_remoteaddr); 1984211981Spjd in = out = NULL; 1985211981Spjd if (init_remote(res, &in, &out)) { 1986211981Spjd rw_wlock(&hio_remote_lock[ncomp]); 1987211981Spjd assert(res->hr_remotein == NULL); 1988211981Spjd assert(res->hr_remoteout == NULL); 1989211981Spjd assert(in != NULL && out != NULL); 1990211981Spjd res->hr_remotein = in; 1991211981Spjd res->hr_remoteout = out; 1992211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1993211981Spjd pjdlog_info("Successfully reconnected to %s.", 1994211981Spjd res->hr_remoteaddr); 1995211981Spjd sync_start(); 1996211981Spjd } else { 1997211981Spjd /* Both connections should be NULL. */ 1998211981Spjd assert(res->hr_remotein == NULL); 1999211981Spjd assert(res->hr_remoteout == NULL); 2000211981Spjd assert(in == NULL && out == NULL); 2001211981Spjd pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 2002211981Spjd res->hr_remoteaddr); 2003211981Spjd } 2004211981Spjd} 2005211981Spjd 2006204076Spjd/* 2007204076Spjd * Thread guards remote connections and reconnects when needed, handles 2008204076Spjd * signals, etc. 2009204076Spjd */ 2010204076Spjdstatic void * 2011204076Spjdguard_thread(void *arg) 2012204076Spjd{ 2013204076Spjd struct hast_resource *res = arg; 2014204076Spjd unsigned int ii, ncomps; 2015211982Spjd struct timespec timeout; 2016211981Spjd time_t lastcheck, now; 2017211982Spjd sigset_t mask; 2018211982Spjd int signo; 2019204076Spjd 2020204076Spjd ncomps = HAST_NCOMPONENTS; 2021211981Spjd lastcheck = time(NULL); 2022204076Spjd 2023211982Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 2024211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 2025211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 2026211982Spjd 2027215332Spjd timeout.tv_sec = RETRY_SLEEP; 2028211982Spjd timeout.tv_nsec = 0; 2029211982Spjd signo = -1; 2030211982Spjd 2031204076Spjd for (;;) { 2032211982Spjd switch (signo) { 2033211982Spjd case SIGINT: 2034211982Spjd case SIGTERM: 2035211982Spjd sigexit_received = true; 2036204076Spjd primary_exitx(EX_OK, 2037204076Spjd "Termination signal received, exiting."); 2038211982Spjd break; 2039211982Spjd default: 2040211982Spjd break; 2041204076Spjd } 2042211882Spjd 2043204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 2044211981Spjd now = time(NULL); 2045211982Spjd if (lastcheck + RETRY_SLEEP <= now) { 2046211982Spjd for (ii = 0; ii < ncomps; ii++) 2047211981Spjd guard_one(res, ii); 2048211981Spjd lastcheck = now; 2049204076Spjd } 2050211982Spjd signo = sigtimedwait(&mask, NULL, &timeout); 2051204076Spjd } 2052204076Spjd /* NOTREACHED */ 2053204076Spjd return (NULL); 2054204076Spjd} 2055