primary.c revision 213531
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 213531 2010-10-07 18:20:16Z pjd $"); 33204076Spjd 34204076Spjd#include <sys/types.h> 35204076Spjd#include <sys/time.h> 36204076Spjd#include <sys/bio.h> 37204076Spjd#include <sys/disk.h> 38204076Spjd#include <sys/refcount.h> 39204076Spjd#include <sys/stat.h> 40204076Spjd 41204076Spjd#include <geom/gate/g_gate.h> 42204076Spjd 43204076Spjd#include <assert.h> 44204076Spjd#include <err.h> 45204076Spjd#include <errno.h> 46204076Spjd#include <fcntl.h> 47204076Spjd#include <libgeom.h> 48204076Spjd#include <pthread.h> 49211982Spjd#include <signal.h> 50204076Spjd#include <stdint.h> 51204076Spjd#include <stdio.h> 52204076Spjd#include <string.h> 53204076Spjd#include <sysexits.h> 54204076Spjd#include <unistd.h> 55204076Spjd 56204076Spjd#include <activemap.h> 57204076Spjd#include <nv.h> 58204076Spjd#include <rangelock.h> 59204076Spjd 60204076Spjd#include "control.h" 61212038Spjd#include "event.h" 62204076Spjd#include "hast.h" 63204076Spjd#include "hast_proto.h" 64204076Spjd#include "hastd.h" 65211886Spjd#include "hooks.h" 66204076Spjd#include "metadata.h" 67204076Spjd#include "proto.h" 68204076Spjd#include "pjdlog.h" 69204076Spjd#include "subr.h" 70204076Spjd#include "synch.h" 71204076Spjd 72210886Spjd/* The is only one remote component for now. */ 73210886Spjd#define ISREMOTE(no) ((no) == 1) 74210886Spjd 75204076Spjdstruct hio { 76204076Spjd /* 77204076Spjd * Number of components we are still waiting for. 78204076Spjd * When this field goes to 0, we can send the request back to the 79204076Spjd * kernel. Each component has to decrease this counter by one 80204076Spjd * even on failure. 81204076Spjd */ 82204076Spjd unsigned int hio_countdown; 83204076Spjd /* 84204076Spjd * Each component has a place to store its own error. 85204076Spjd * Once the request is handled by all components we can decide if the 86204076Spjd * request overall is successful or not. 87204076Spjd */ 88204076Spjd int *hio_errors; 89204076Spjd /* 90204076Spjd * Structure used to comunicate with GEOM Gate class. 91204076Spjd */ 92204076Spjd struct g_gate_ctl_io hio_ggio; 93204076Spjd TAILQ_ENTRY(hio) *hio_next; 94204076Spjd}; 95204076Spjd#define hio_free_next hio_next[0] 96204076Spjd#define hio_done_next hio_next[0] 97204076Spjd 98204076Spjd/* 99204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 100204076Spjd * until some in-progress requests are freed. 101204076Spjd */ 102204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 103204076Spjdstatic pthread_mutex_t hio_free_list_lock; 104204076Spjdstatic pthread_cond_t hio_free_list_cond; 105204076Spjd/* 106204076Spjd * There is one send list for every component. One requests is placed on all 107204076Spjd * send lists - each component gets the same request, but each component is 108204076Spjd * responsible for managing his own send list. 109204076Spjd */ 110204076Spjdstatic TAILQ_HEAD(, hio) *hio_send_list; 111204076Spjdstatic pthread_mutex_t *hio_send_list_lock; 112204076Spjdstatic pthread_cond_t *hio_send_list_cond; 113204076Spjd/* 114204076Spjd * There is one recv list for every component, although local components don't 115204076Spjd * use recv lists as local requests are done synchronously. 116204076Spjd */ 117204076Spjdstatic TAILQ_HEAD(, hio) *hio_recv_list; 118204076Spjdstatic pthread_mutex_t *hio_recv_list_lock; 119204076Spjdstatic pthread_cond_t *hio_recv_list_cond; 120204076Spjd/* 121204076Spjd * Request is placed on done list by the slowest component (the one that 122204076Spjd * decreased hio_countdown from 1 to 0). 123204076Spjd */ 124204076Spjdstatic TAILQ_HEAD(, hio) hio_done_list; 125204076Spjdstatic pthread_mutex_t hio_done_list_lock; 126204076Spjdstatic pthread_cond_t hio_done_list_cond; 127204076Spjd/* 128204076Spjd * Structure below are for interaction with sync thread. 129204076Spjd */ 130204076Spjdstatic bool sync_inprogress; 131204076Spjdstatic pthread_mutex_t sync_lock; 132204076Spjdstatic pthread_cond_t sync_cond; 133204076Spjd/* 134204076Spjd * The lock below allows to synchornize access to remote connections. 135204076Spjd */ 136204076Spjdstatic pthread_rwlock_t *hio_remote_lock; 137204076Spjd 138204076Spjd/* 139204076Spjd * Lock to synchronize metadata updates. Also synchronize access to 140204076Spjd * hr_primary_localcnt and hr_primary_remotecnt fields. 141204076Spjd */ 142204076Spjdstatic pthread_mutex_t metadata_lock; 143204076Spjd 144204076Spjd/* 145204076Spjd * Maximum number of outstanding I/O requests. 146204076Spjd */ 147204076Spjd#define HAST_HIO_MAX 256 148204076Spjd/* 149204076Spjd * Number of components. At this point there are only two components: local 150204076Spjd * and remote, but in the future it might be possible to use multiple local 151204076Spjd * and remote components. 152204076Spjd */ 153204076Spjd#define HAST_NCOMPONENTS 2 154204076Spjd/* 155211982Spjd * Number of seconds to sleep between reconnect retries or keepalive packets. 156204076Spjd */ 157211982Spjd#define RETRY_SLEEP 10 158204076Spjd 159204076Spjd#define ISCONNECTED(res, no) \ 160204076Spjd ((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL) 161204076Spjd 162204076Spjd#define QUEUE_INSERT1(hio, name, ncomp) do { \ 163204076Spjd bool _wakeup; \ 164204076Spjd \ 165204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 166204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]); \ 167204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio), \ 168204076Spjd hio_next[(ncomp)]); \ 169204076Spjd mtx_unlock(&hio_##name##_list_lock[ncomp]); \ 170204076Spjd if (_wakeup) \ 171204076Spjd cv_signal(&hio_##name##_list_cond[(ncomp)]); \ 172204076Spjd} while (0) 173204076Spjd#define QUEUE_INSERT2(hio, name) do { \ 174204076Spjd bool _wakeup; \ 175204076Spjd \ 176204076Spjd mtx_lock(&hio_##name##_list_lock); \ 177204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 178204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\ 179204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 180204076Spjd if (_wakeup) \ 181204076Spjd cv_signal(&hio_##name##_list_cond); \ 182204076Spjd} while (0) 183204076Spjd#define QUEUE_TAKE1(hio, name, ncomp) do { \ 184204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 185204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL) { \ 186204076Spjd cv_wait(&hio_##name##_list_cond[(ncomp)], \ 187204076Spjd &hio_##name##_list_lock[(ncomp)]); \ 188204076Spjd } \ 189204076Spjd TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio), \ 190204076Spjd hio_next[(ncomp)]); \ 191204076Spjd mtx_unlock(&hio_##name##_list_lock[(ncomp)]); \ 192204076Spjd} while (0) 193204076Spjd#define QUEUE_TAKE2(hio, name) do { \ 194204076Spjd mtx_lock(&hio_##name##_list_lock); \ 195204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 196204076Spjd cv_wait(&hio_##name##_list_cond, \ 197204076Spjd &hio_##name##_list_lock); \ 198204076Spjd } \ 199204076Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next); \ 200204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 201204076Spjd} while (0) 202204076Spjd 203209183Spjd#define SYNCREQ(hio) do { \ 204209183Spjd (hio)->hio_ggio.gctl_unit = -1; \ 205209183Spjd (hio)->hio_ggio.gctl_seq = 1; \ 206209183Spjd} while (0) 207204076Spjd#define ISSYNCREQ(hio) ((hio)->hio_ggio.gctl_unit == -1) 208204076Spjd#define SYNCREQDONE(hio) do { (hio)->hio_ggio.gctl_unit = -2; } while (0) 209204076Spjd#define ISSYNCREQDONE(hio) ((hio)->hio_ggio.gctl_unit == -2) 210204076Spjd 211204076Spjdstatic struct hast_resource *gres; 212204076Spjd 213204076Spjdstatic pthread_mutex_t range_lock; 214204076Spjdstatic struct rangelocks *range_regular; 215204076Spjdstatic bool range_regular_wait; 216204076Spjdstatic pthread_cond_t range_regular_cond; 217204076Spjdstatic struct rangelocks *range_sync; 218204076Spjdstatic bool range_sync_wait; 219204076Spjdstatic pthread_cond_t range_sync_cond; 220204076Spjd 221204076Spjdstatic void *ggate_recv_thread(void *arg); 222204076Spjdstatic void *local_send_thread(void *arg); 223204076Spjdstatic void *remote_send_thread(void *arg); 224204076Spjdstatic void *remote_recv_thread(void *arg); 225204076Spjdstatic void *ggate_send_thread(void *arg); 226204076Spjdstatic void *sync_thread(void *arg); 227204076Spjdstatic void *guard_thread(void *arg); 228204076Spjd 229211982Spjdstatic void 230204076Spjdcleanup(struct hast_resource *res) 231204076Spjd{ 232204076Spjd int rerrno; 233204076Spjd 234204076Spjd /* Remember errno. */ 235204076Spjd rerrno = errno; 236204076Spjd 237204076Spjd /* Destroy ggate provider if we created one. */ 238204076Spjd if (res->hr_ggateunit >= 0) { 239204076Spjd struct g_gate_ctl_destroy ggiod; 240204076Spjd 241204076Spjd ggiod.gctl_version = G_GATE_VERSION; 242204076Spjd ggiod.gctl_unit = res->hr_ggateunit; 243204076Spjd ggiod.gctl_force = 1; 244204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 245213531Spjd pjdlog_errno(LOG_WARNING, 246213531Spjd "Unable to destroy hast/%s device", 247204076Spjd res->hr_provname); 248204076Spjd } 249204076Spjd res->hr_ggateunit = -1; 250204076Spjd } 251204076Spjd 252204076Spjd /* Restore errno. */ 253204076Spjd errno = rerrno; 254204076Spjd} 255204076Spjd 256212899Spjdstatic __dead2 void 257204076Spjdprimary_exit(int exitcode, const char *fmt, ...) 258204076Spjd{ 259204076Spjd va_list ap; 260204076Spjd 261204076Spjd assert(exitcode != EX_OK); 262204076Spjd va_start(ap, fmt); 263204076Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 264204076Spjd va_end(ap); 265204076Spjd cleanup(gres); 266204076Spjd exit(exitcode); 267204076Spjd} 268204076Spjd 269212899Spjdstatic __dead2 void 270204076Spjdprimary_exitx(int exitcode, const char *fmt, ...) 271204076Spjd{ 272204076Spjd va_list ap; 273204076Spjd 274204076Spjd va_start(ap, fmt); 275204076Spjd pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 276204076Spjd va_end(ap); 277204076Spjd cleanup(gres); 278204076Spjd exit(exitcode); 279204076Spjd} 280204076Spjd 281204076Spjdstatic int 282204076Spjdhast_activemap_flush(struct hast_resource *res) 283204076Spjd{ 284204076Spjd const unsigned char *buf; 285204076Spjd size_t size; 286204076Spjd 287204076Spjd buf = activemap_bitmap(res->hr_amp, &size); 288204076Spjd assert(buf != NULL); 289204076Spjd assert((size % res->hr_local_sectorsize) == 0); 290204076Spjd if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 291204076Spjd (ssize_t)size) { 292204076Spjd KEEP_ERRNO(pjdlog_errno(LOG_ERR, 293204076Spjd "Unable to flush activemap to disk")); 294204076Spjd return (-1); 295204076Spjd } 296204076Spjd return (0); 297204076Spjd} 298204076Spjd 299210881Spjdstatic bool 300210881Spjdreal_remote(const struct hast_resource *res) 301210881Spjd{ 302210881Spjd 303210881Spjd return (strcmp(res->hr_remoteaddr, "none") != 0); 304210881Spjd} 305210881Spjd 306204076Spjdstatic void 307204076Spjdinit_environment(struct hast_resource *res __unused) 308204076Spjd{ 309204076Spjd struct hio *hio; 310204076Spjd unsigned int ii, ncomps; 311204076Spjd 312204076Spjd /* 313204076Spjd * In the future it might be per-resource value. 314204076Spjd */ 315204076Spjd ncomps = HAST_NCOMPONENTS; 316204076Spjd 317204076Spjd /* 318204076Spjd * Allocate memory needed by lists. 319204076Spjd */ 320204076Spjd hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 321204076Spjd if (hio_send_list == NULL) { 322204076Spjd primary_exitx(EX_TEMPFAIL, 323204076Spjd "Unable to allocate %zu bytes of memory for send lists.", 324204076Spjd sizeof(hio_send_list[0]) * ncomps); 325204076Spjd } 326204076Spjd hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 327204076Spjd if (hio_send_list_lock == NULL) { 328204076Spjd primary_exitx(EX_TEMPFAIL, 329204076Spjd "Unable to allocate %zu bytes of memory for send list locks.", 330204076Spjd sizeof(hio_send_list_lock[0]) * ncomps); 331204076Spjd } 332204076Spjd hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 333204076Spjd if (hio_send_list_cond == NULL) { 334204076Spjd primary_exitx(EX_TEMPFAIL, 335204076Spjd "Unable to allocate %zu bytes of memory for send list condition variables.", 336204076Spjd sizeof(hio_send_list_cond[0]) * ncomps); 337204076Spjd } 338204076Spjd hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 339204076Spjd if (hio_recv_list == NULL) { 340204076Spjd primary_exitx(EX_TEMPFAIL, 341204076Spjd "Unable to allocate %zu bytes of memory for recv lists.", 342204076Spjd sizeof(hio_recv_list[0]) * ncomps); 343204076Spjd } 344204076Spjd hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 345204076Spjd if (hio_recv_list_lock == NULL) { 346204076Spjd primary_exitx(EX_TEMPFAIL, 347204076Spjd "Unable to allocate %zu bytes of memory for recv list locks.", 348204076Spjd sizeof(hio_recv_list_lock[0]) * ncomps); 349204076Spjd } 350204076Spjd hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 351204076Spjd if (hio_recv_list_cond == NULL) { 352204076Spjd primary_exitx(EX_TEMPFAIL, 353204076Spjd "Unable to allocate %zu bytes of memory for recv list condition variables.", 354204076Spjd sizeof(hio_recv_list_cond[0]) * ncomps); 355204076Spjd } 356204076Spjd hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 357204076Spjd if (hio_remote_lock == NULL) { 358204076Spjd primary_exitx(EX_TEMPFAIL, 359204076Spjd "Unable to allocate %zu bytes of memory for remote connections locks.", 360204076Spjd sizeof(hio_remote_lock[0]) * ncomps); 361204076Spjd } 362204076Spjd 363204076Spjd /* 364204076Spjd * Initialize lists, their locks and theirs condition variables. 365204076Spjd */ 366204076Spjd TAILQ_INIT(&hio_free_list); 367204076Spjd mtx_init(&hio_free_list_lock); 368204076Spjd cv_init(&hio_free_list_cond); 369204076Spjd for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 370204076Spjd TAILQ_INIT(&hio_send_list[ii]); 371204076Spjd mtx_init(&hio_send_list_lock[ii]); 372204076Spjd cv_init(&hio_send_list_cond[ii]); 373204076Spjd TAILQ_INIT(&hio_recv_list[ii]); 374204076Spjd mtx_init(&hio_recv_list_lock[ii]); 375204076Spjd cv_init(&hio_recv_list_cond[ii]); 376204076Spjd rw_init(&hio_remote_lock[ii]); 377204076Spjd } 378204076Spjd TAILQ_INIT(&hio_done_list); 379204076Spjd mtx_init(&hio_done_list_lock); 380204076Spjd cv_init(&hio_done_list_cond); 381204076Spjd mtx_init(&metadata_lock); 382204076Spjd 383204076Spjd /* 384204076Spjd * Allocate requests pool and initialize requests. 385204076Spjd */ 386204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 387204076Spjd hio = malloc(sizeof(*hio)); 388204076Spjd if (hio == NULL) { 389204076Spjd primary_exitx(EX_TEMPFAIL, 390204076Spjd "Unable to allocate %zu bytes of memory for hio request.", 391204076Spjd sizeof(*hio)); 392204076Spjd } 393204076Spjd hio->hio_countdown = 0; 394204076Spjd hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 395204076Spjd if (hio->hio_errors == NULL) { 396204076Spjd primary_exitx(EX_TEMPFAIL, 397204076Spjd "Unable allocate %zu bytes of memory for hio errors.", 398204076Spjd sizeof(hio->hio_errors[0]) * ncomps); 399204076Spjd } 400204076Spjd hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 401204076Spjd if (hio->hio_next == NULL) { 402204076Spjd primary_exitx(EX_TEMPFAIL, 403204076Spjd "Unable allocate %zu bytes of memory for hio_next field.", 404204076Spjd sizeof(hio->hio_next[0]) * ncomps); 405204076Spjd } 406204076Spjd hio->hio_ggio.gctl_version = G_GATE_VERSION; 407204076Spjd hio->hio_ggio.gctl_data = malloc(MAXPHYS); 408204076Spjd if (hio->hio_ggio.gctl_data == NULL) { 409204076Spjd primary_exitx(EX_TEMPFAIL, 410204076Spjd "Unable to allocate %zu bytes of memory for gctl_data.", 411204076Spjd MAXPHYS); 412204076Spjd } 413204076Spjd hio->hio_ggio.gctl_length = MAXPHYS; 414204076Spjd hio->hio_ggio.gctl_error = 0; 415204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 416204076Spjd } 417204076Spjd} 418204076Spjd 419204076Spjdstatic void 420204076Spjdinit_local(struct hast_resource *res) 421204076Spjd{ 422204076Spjd unsigned char *buf; 423204076Spjd size_t mapsize; 424204076Spjd 425204076Spjd if (metadata_read(res, true) < 0) 426204076Spjd exit(EX_NOINPUT); 427204076Spjd mtx_init(&res->hr_amp_lock); 428204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 429204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 430204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 431204076Spjd } 432204076Spjd mtx_init(&range_lock); 433204076Spjd cv_init(&range_regular_cond); 434204076Spjd if (rangelock_init(&range_regular) < 0) 435204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 436204076Spjd cv_init(&range_sync_cond); 437204076Spjd if (rangelock_init(&range_sync) < 0) 438204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 439204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 440204076Spjd buf = calloc(1, mapsize); 441204076Spjd if (buf == NULL) { 442204076Spjd primary_exitx(EX_TEMPFAIL, 443204076Spjd "Unable to allocate buffer for activemap."); 444204076Spjd } 445204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 446204076Spjd (ssize_t)mapsize) { 447204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 448204076Spjd } 449204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 450209181Spjd free(buf); 451204076Spjd if (res->hr_resuid != 0) 452204076Spjd return; 453204076Spjd /* 454204076Spjd * We're using provider for the first time, so we have to generate 455204076Spjd * resource unique identifier and initialize local and remote counts. 456204076Spjd */ 457204076Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 458204076Spjd res->hr_primary_localcnt = 1; 459204076Spjd res->hr_primary_remotecnt = 0; 460204076Spjd if (metadata_write(res) < 0) 461204076Spjd exit(EX_NOINPUT); 462204076Spjd} 463204076Spjd 464205738Spjdstatic bool 465205738Spjdinit_remote(struct hast_resource *res, struct proto_conn **inp, 466205738Spjd struct proto_conn **outp) 467204076Spjd{ 468205738Spjd struct proto_conn *in, *out; 469204076Spjd struct nv *nvout, *nvin; 470204076Spjd const unsigned char *token; 471204076Spjd unsigned char *map; 472204076Spjd const char *errmsg; 473204076Spjd int32_t extentsize; 474204076Spjd int64_t datasize; 475204076Spjd uint32_t mapsize; 476204076Spjd size_t size; 477204076Spjd 478205738Spjd assert((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 479210881Spjd assert(real_remote(res)); 480205738Spjd 481205738Spjd in = out = NULL; 482211983Spjd errmsg = NULL; 483205738Spjd 484204076Spjd /* Prepare outgoing connection with remote node. */ 485205738Spjd if (proto_client(res->hr_remoteaddr, &out) < 0) { 486207347Spjd primary_exit(EX_TEMPFAIL, "Unable to create connection to %s", 487204076Spjd res->hr_remoteaddr); 488204076Spjd } 489204076Spjd /* Try to connect, but accept failure. */ 490205738Spjd if (proto_connect(out) < 0) { 491204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 492204076Spjd res->hr_remoteaddr); 493204076Spjd goto close; 494204076Spjd } 495207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 496207371Spjd if (proto_timeout(out, res->hr_timeout) < 0) 497207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 498204076Spjd /* 499204076Spjd * First handshake step. 500204076Spjd * Setup outgoing connection with remote node. 501204076Spjd */ 502204076Spjd nvout = nv_alloc(); 503204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 504204076Spjd if (nv_error(nvout) != 0) { 505204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 506204076Spjd "Unable to allocate header for connection with %s", 507204076Spjd res->hr_remoteaddr); 508204076Spjd nv_free(nvout); 509204076Spjd goto close; 510204076Spjd } 511205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 512204076Spjd pjdlog_errno(LOG_WARNING, 513204076Spjd "Unable to send handshake header to %s", 514204076Spjd res->hr_remoteaddr); 515204076Spjd nv_free(nvout); 516204076Spjd goto close; 517204076Spjd } 518204076Spjd nv_free(nvout); 519205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 520204076Spjd pjdlog_errno(LOG_WARNING, 521204076Spjd "Unable to receive handshake header from %s", 522204076Spjd res->hr_remoteaddr); 523204076Spjd goto close; 524204076Spjd } 525204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 526204076Spjd if (errmsg != NULL) { 527204076Spjd pjdlog_warning("%s", errmsg); 528204076Spjd nv_free(nvin); 529204076Spjd goto close; 530204076Spjd } 531204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 532204076Spjd if (token == NULL) { 533204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 534204076Spjd res->hr_remoteaddr); 535204076Spjd nv_free(nvin); 536204076Spjd goto close; 537204076Spjd } 538204076Spjd if (size != sizeof(res->hr_token)) { 539204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 540204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 541204076Spjd nv_free(nvin); 542204076Spjd goto close; 543204076Spjd } 544204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 545204076Spjd nv_free(nvin); 546204076Spjd 547204076Spjd /* 548204076Spjd * Second handshake step. 549204076Spjd * Setup incoming connection with remote node. 550204076Spjd */ 551205738Spjd if (proto_client(res->hr_remoteaddr, &in) < 0) { 552204076Spjd pjdlog_errno(LOG_WARNING, "Unable to create connection to %s", 553204076Spjd res->hr_remoteaddr); 554204076Spjd } 555204076Spjd /* Try to connect, but accept failure. */ 556205738Spjd if (proto_connect(in) < 0) { 557204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 558204076Spjd res->hr_remoteaddr); 559204076Spjd goto close; 560204076Spjd } 561207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 562207371Spjd if (proto_timeout(in, res->hr_timeout) < 0) 563207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 564204076Spjd nvout = nv_alloc(); 565204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 566204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 567204076Spjd "token"); 568204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 569204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 570204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 571204076Spjd if (nv_error(nvout) != 0) { 572204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 573204076Spjd "Unable to allocate header for connection with %s", 574204076Spjd res->hr_remoteaddr); 575204076Spjd nv_free(nvout); 576204076Spjd goto close; 577204076Spjd } 578205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 579204076Spjd pjdlog_errno(LOG_WARNING, 580204076Spjd "Unable to send handshake header to %s", 581204076Spjd res->hr_remoteaddr); 582204076Spjd nv_free(nvout); 583204076Spjd goto close; 584204076Spjd } 585204076Spjd nv_free(nvout); 586205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 587204076Spjd pjdlog_errno(LOG_WARNING, 588204076Spjd "Unable to receive handshake header from %s", 589204076Spjd res->hr_remoteaddr); 590204076Spjd goto close; 591204076Spjd } 592204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 593204076Spjd if (errmsg != NULL) { 594204076Spjd pjdlog_warning("%s", errmsg); 595204076Spjd nv_free(nvin); 596204076Spjd goto close; 597204076Spjd } 598204076Spjd datasize = nv_get_int64(nvin, "datasize"); 599204076Spjd if (datasize != res->hr_datasize) { 600204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 601204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 602204076Spjd nv_free(nvin); 603204076Spjd goto close; 604204076Spjd } 605204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 606204076Spjd if (extentsize != res->hr_extentsize) { 607204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 608204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 609204076Spjd nv_free(nvin); 610204076Spjd goto close; 611204076Spjd } 612204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 613204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 614204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 615204076Spjd map = NULL; 616204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 617204076Spjd if (mapsize > 0) { 618204076Spjd map = malloc(mapsize); 619204076Spjd if (map == NULL) { 620204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 621204076Spjd (uintmax_t)mapsize); 622204076Spjd nv_free(nvin); 623204076Spjd goto close; 624204076Spjd } 625204076Spjd /* 626204076Spjd * Remote node have some dirty extents on its own, lets 627204076Spjd * download its activemap. 628204076Spjd */ 629205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 630204076Spjd mapsize) < 0) { 631204076Spjd pjdlog_errno(LOG_ERR, 632204076Spjd "Unable to receive remote activemap"); 633204076Spjd nv_free(nvin); 634204076Spjd free(map); 635204076Spjd goto close; 636204076Spjd } 637204076Spjd /* 638204076Spjd * Merge local and remote bitmaps. 639204076Spjd */ 640204076Spjd activemap_merge(res->hr_amp, map, mapsize); 641204076Spjd free(map); 642204076Spjd /* 643204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 644204076Spjd * disk before we start to synchronize. 645204076Spjd */ 646204076Spjd (void)hast_activemap_flush(res); 647204076Spjd } 648204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 649205738Spjd if (inp != NULL && outp != NULL) { 650205738Spjd *inp = in; 651205738Spjd *outp = out; 652205738Spjd } else { 653205738Spjd res->hr_remotein = in; 654205738Spjd res->hr_remoteout = out; 655205738Spjd } 656212038Spjd event_send(res, EVENT_CONNECT); 657205738Spjd return (true); 658205738Spjdclose: 659211983Spjd if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0) 660212038Spjd event_send(res, EVENT_SPLITBRAIN); 661205738Spjd proto_close(out); 662205738Spjd if (in != NULL) 663205738Spjd proto_close(in); 664205738Spjd return (false); 665205738Spjd} 666205738Spjd 667205738Spjdstatic void 668205738Spjdsync_start(void) 669205738Spjd{ 670205738Spjd 671204076Spjd mtx_lock(&sync_lock); 672204076Spjd sync_inprogress = true; 673204076Spjd mtx_unlock(&sync_lock); 674204076Spjd cv_signal(&sync_cond); 675204076Spjd} 676204076Spjd 677204076Spjdstatic void 678211878Spjdsync_stop(void) 679211878Spjd{ 680211878Spjd 681211878Spjd mtx_lock(&sync_lock); 682211878Spjd if (sync_inprogress) 683211878Spjd sync_inprogress = false; 684211878Spjd mtx_unlock(&sync_lock); 685211878Spjd} 686211878Spjd 687211878Spjdstatic void 688204076Spjdinit_ggate(struct hast_resource *res) 689204076Spjd{ 690204076Spjd struct g_gate_ctl_create ggiocreate; 691204076Spjd struct g_gate_ctl_cancel ggiocancel; 692204076Spjd 693204076Spjd /* 694204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 695204076Spjd */ 696204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 697204076Spjd if (res->hr_ggatefd < 0) 698204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 699204076Spjd /* 700204076Spjd * Create provider before trying to connect, as connection failure 701204076Spjd * is not critical, but may take some time. 702204076Spjd */ 703204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 704204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 705204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 706204076Spjd ggiocreate.gctl_flags = 0; 707206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 708204076Spjd ggiocreate.gctl_timeout = 0; 709204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 710204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 711204076Spjd res->hr_provname); 712204076Spjd bzero(ggiocreate.gctl_info, sizeof(ggiocreate.gctl_info)); 713204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 714204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 715204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 716204076Spjd return; 717204076Spjd } 718204076Spjd if (errno != EEXIST) { 719204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 720204076Spjd res->hr_provname); 721204076Spjd } 722204076Spjd pjdlog_debug(1, 723204076Spjd "Device hast/%s already exists, we will try to take it over.", 724204076Spjd res->hr_provname); 725204076Spjd /* 726204076Spjd * If we received EEXIST, we assume that the process who created the 727204076Spjd * provider died and didn't clean up. In that case we will start from 728204076Spjd * where he left of. 729204076Spjd */ 730204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 731204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 732204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 733204076Spjd res->hr_provname); 734204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 735204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 736204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 737204076Spjd return; 738204076Spjd } 739204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 740204076Spjd res->hr_provname); 741204076Spjd} 742204076Spjd 743204076Spjdvoid 744204076Spjdhastd_primary(struct hast_resource *res) 745204076Spjd{ 746204076Spjd pthread_t td; 747204076Spjd pid_t pid; 748204076Spjd int error; 749204076Spjd 750204076Spjd /* 751204076Spjd * Create communication channel between parent and child. 752204076Spjd */ 753204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 754204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 755212034Spjd pjdlog_exit(EX_OSERR, 756204076Spjd "Unable to create control sockets between parent and child"); 757204076Spjd } 758212038Spjd /* 759212038Spjd * Create communication channel between child and parent. 760212038Spjd */ 761212038Spjd if (proto_client("socketpair://", &res->hr_event) < 0) { 762212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 763212038Spjd pjdlog_exit(EX_OSERR, 764212038Spjd "Unable to create event sockets between child and parent"); 765212038Spjd } 766204076Spjd 767204076Spjd pid = fork(); 768204076Spjd if (pid < 0) { 769204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 770212034Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to fork"); 771204076Spjd } 772204076Spjd 773204076Spjd if (pid > 0) { 774204076Spjd /* This is parent. */ 775212038Spjd /* Declare that we are receiver. */ 776212038Spjd proto_recv(res->hr_event, NULL, 0); 777204076Spjd res->hr_workerpid = pid; 778204076Spjd return; 779204076Spjd } 780211977Spjd 781211984Spjd gres = res; 782211984Spjd 783204076Spjd (void)pidfile_close(pfh); 784211977Spjd hook_fini(); 785204076Spjd 786204076Spjd setproctitle("%s (primary)", res->hr_name); 787204076Spjd 788212038Spjd /* Declare that we are sender. */ 789212038Spjd proto_send(res->hr_event, NULL, 0); 790212038Spjd 791204076Spjd init_local(res); 792213007Spjd init_ggate(res); 793213007Spjd init_environment(res); 794213007Spjd /* 795213530Spjd * Create the guard thread first, so we can handle signals from the 796213530Spjd * very begining. 797213530Spjd */ 798213530Spjd error = pthread_create(&td, NULL, guard_thread, res); 799213530Spjd assert(error == 0); 800213530Spjd /* 801213007Spjd * Create the control thread before sending any event to the parent, 802213007Spjd * as we can deadlock when parent sends control request to worker, 803213007Spjd * but worker has no control thread started yet, so parent waits. 804213007Spjd * In the meantime worker sends an event to the parent, but parent 805213007Spjd * is unable to handle the event, because it waits for control 806213007Spjd * request response. 807213007Spjd */ 808213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 809213007Spjd assert(error == 0); 810210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 811205738Spjd sync_start(); 812204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 813204076Spjd assert(error == 0); 814204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 815204076Spjd assert(error == 0); 816204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 817204076Spjd assert(error == 0); 818204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 819204076Spjd assert(error == 0); 820204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 821204076Spjd assert(error == 0); 822213530Spjd (void)sync_thread(res); 823204076Spjd} 824204076Spjd 825204076Spjdstatic void 826204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 827204076Spjd{ 828204076Spjd char msg[1024]; 829204076Spjd va_list ap; 830204076Spjd int len; 831204076Spjd 832204076Spjd va_start(ap, fmt); 833204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 834204076Spjd va_end(ap); 835204076Spjd if ((size_t)len < sizeof(msg)) { 836204076Spjd switch (ggio->gctl_cmd) { 837204076Spjd case BIO_READ: 838204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 839204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 840204076Spjd (uintmax_t)ggio->gctl_length); 841204076Spjd break; 842204076Spjd case BIO_DELETE: 843204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 844204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 845204076Spjd (uintmax_t)ggio->gctl_length); 846204076Spjd break; 847204076Spjd case BIO_FLUSH: 848204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 849204076Spjd break; 850204076Spjd case BIO_WRITE: 851204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 852204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 853204076Spjd (uintmax_t)ggio->gctl_length); 854204076Spjd break; 855204076Spjd default: 856204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 857204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 858204076Spjd break; 859204076Spjd } 860204076Spjd } 861204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 862204076Spjd} 863204076Spjd 864204076Spjdstatic void 865204076Spjdremote_close(struct hast_resource *res, int ncomp) 866204076Spjd{ 867204076Spjd 868204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 869204076Spjd /* 870204076Spjd * A race is possible between dropping rlock and acquiring wlock - 871204076Spjd * another thread can close connection in-between. 872204076Spjd */ 873204076Spjd if (!ISCONNECTED(res, ncomp)) { 874204076Spjd assert(res->hr_remotein == NULL); 875204076Spjd assert(res->hr_remoteout == NULL); 876204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 877204076Spjd return; 878204076Spjd } 879204076Spjd 880204076Spjd assert(res->hr_remotein != NULL); 881204076Spjd assert(res->hr_remoteout != NULL); 882204076Spjd 883211881Spjd pjdlog_debug(2, "Closing incoming connection to %s.", 884204076Spjd res->hr_remoteaddr); 885204076Spjd proto_close(res->hr_remotein); 886204076Spjd res->hr_remotein = NULL; 887211881Spjd pjdlog_debug(2, "Closing outgoing connection to %s.", 888204076Spjd res->hr_remoteaddr); 889204076Spjd proto_close(res->hr_remoteout); 890204076Spjd res->hr_remoteout = NULL; 891204076Spjd 892204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 893204076Spjd 894211881Spjd pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 895211881Spjd 896204076Spjd /* 897204076Spjd * Stop synchronization if in-progress. 898204076Spjd */ 899211878Spjd sync_stop(); 900211984Spjd 901212038Spjd event_send(res, EVENT_DISCONNECT); 902204076Spjd} 903204076Spjd 904204076Spjd/* 905204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 906204076Spjd * appropriate threads: 907204076Spjd * WRITE - always goes to both local_send and remote_send threads 908204076Spjd * READ (when the block is up-to-date on local component) - 909204076Spjd * only local_send thread 910204076Spjd * READ (when the block isn't up-to-date on local component) - 911204076Spjd * only remote_send thread 912204076Spjd * DELETE - always goes to both local_send and remote_send threads 913204076Spjd * FLUSH - always goes to both local_send and remote_send threads 914204076Spjd */ 915204076Spjdstatic void * 916204076Spjdggate_recv_thread(void *arg) 917204076Spjd{ 918204076Spjd struct hast_resource *res = arg; 919204076Spjd struct g_gate_ctl_io *ggio; 920204076Spjd struct hio *hio; 921204076Spjd unsigned int ii, ncomp, ncomps; 922204076Spjd int error; 923204076Spjd 924204076Spjd ncomps = HAST_NCOMPONENTS; 925204076Spjd 926204076Spjd for (;;) { 927204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 928204076Spjd QUEUE_TAKE2(hio, free); 929204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 930204076Spjd ggio = &hio->hio_ggio; 931204076Spjd ggio->gctl_unit = res->hr_ggateunit; 932204076Spjd ggio->gctl_length = MAXPHYS; 933204076Spjd ggio->gctl_error = 0; 934204076Spjd pjdlog_debug(2, 935204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 936204076Spjd hio); 937204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 938204076Spjd if (sigexit_received) 939204076Spjd pthread_exit(NULL); 940204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 941204076Spjd } 942204076Spjd error = ggio->gctl_error; 943204076Spjd switch (error) { 944204076Spjd case 0: 945204076Spjd break; 946204076Spjd case ECANCELED: 947204076Spjd /* Exit gracefully. */ 948204076Spjd if (!sigexit_received) { 949204076Spjd pjdlog_debug(2, 950204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 951204076Spjd hio); 952204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 953204076Spjd } 954204076Spjd pthread_exit(NULL); 955204076Spjd case ENOMEM: 956204076Spjd /* 957204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 958204076Spjd * bytes - request can't be bigger than that. 959204076Spjd */ 960204076Spjd /* FALLTHROUGH */ 961204076Spjd case ENXIO: 962204076Spjd default: 963204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 964204076Spjd strerror(error)); 965204076Spjd } 966204076Spjd for (ii = 0; ii < ncomps; ii++) 967204076Spjd hio->hio_errors[ii] = EINVAL; 968204076Spjd reqlog(LOG_DEBUG, 2, ggio, 969204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 970204076Spjd hio); 971204076Spjd /* 972204076Spjd * Inform all components about new write request. 973204076Spjd * For read request prefer local component unless the given 974204076Spjd * range is out-of-date, then use remote component. 975204076Spjd */ 976204076Spjd switch (ggio->gctl_cmd) { 977204076Spjd case BIO_READ: 978204076Spjd pjdlog_debug(2, 979204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 980204076Spjd hio); 981204076Spjd refcount_init(&hio->hio_countdown, 1); 982204076Spjd mtx_lock(&metadata_lock); 983204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 984204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 985204076Spjd /* 986204076Spjd * This range is up-to-date on local component, 987204076Spjd * so handle request locally. 988204076Spjd */ 989204076Spjd /* Local component is 0 for now. */ 990204076Spjd ncomp = 0; 991204076Spjd } else /* if (res->hr_syncsrc == 992204076Spjd HAST_SYNCSRC_SECONDARY) */ { 993204076Spjd assert(res->hr_syncsrc == 994204076Spjd HAST_SYNCSRC_SECONDARY); 995204076Spjd /* 996204076Spjd * This range is out-of-date on local component, 997204076Spjd * so send request to the remote node. 998204076Spjd */ 999204076Spjd /* Remote component is 1 for now. */ 1000204076Spjd ncomp = 1; 1001204076Spjd } 1002204076Spjd mtx_unlock(&metadata_lock); 1003204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1004204076Spjd break; 1005204076Spjd case BIO_WRITE: 1006204076Spjd for (;;) { 1007204076Spjd mtx_lock(&range_lock); 1008204076Spjd if (rangelock_islocked(range_sync, 1009204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1010204076Spjd pjdlog_debug(2, 1011204076Spjd "regular: Range offset=%jd length=%zu locked.", 1012204076Spjd (intmax_t)ggio->gctl_offset, 1013204076Spjd (size_t)ggio->gctl_length); 1014204076Spjd range_regular_wait = true; 1015204076Spjd cv_wait(&range_regular_cond, &range_lock); 1016204076Spjd range_regular_wait = false; 1017204076Spjd mtx_unlock(&range_lock); 1018204076Spjd continue; 1019204076Spjd } 1020204076Spjd if (rangelock_add(range_regular, 1021204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1022204076Spjd mtx_unlock(&range_lock); 1023204076Spjd pjdlog_debug(2, 1024204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1025204076Spjd (intmax_t)ggio->gctl_offset, 1026204076Spjd (size_t)ggio->gctl_length); 1027204076Spjd sleep(1); 1028204076Spjd continue; 1029204076Spjd } 1030204076Spjd mtx_unlock(&range_lock); 1031204076Spjd break; 1032204076Spjd } 1033204076Spjd mtx_lock(&res->hr_amp_lock); 1034204076Spjd if (activemap_write_start(res->hr_amp, 1035204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1036204076Spjd (void)hast_activemap_flush(res); 1037204076Spjd } 1038204076Spjd mtx_unlock(&res->hr_amp_lock); 1039204076Spjd /* FALLTHROUGH */ 1040204076Spjd case BIO_DELETE: 1041204076Spjd case BIO_FLUSH: 1042204076Spjd pjdlog_debug(2, 1043204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1044204076Spjd hio); 1045204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1046204076Spjd for (ii = 0; ii < ncomps; ii++) 1047204076Spjd QUEUE_INSERT1(hio, send, ii); 1048204076Spjd break; 1049204076Spjd } 1050204076Spjd } 1051204076Spjd /* NOTREACHED */ 1052204076Spjd return (NULL); 1053204076Spjd} 1054204076Spjd 1055204076Spjd/* 1056204076Spjd * Thread reads from or writes to local component. 1057204076Spjd * If local read fails, it redirects it to remote_send thread. 1058204076Spjd */ 1059204076Spjdstatic void * 1060204076Spjdlocal_send_thread(void *arg) 1061204076Spjd{ 1062204076Spjd struct hast_resource *res = arg; 1063204076Spjd struct g_gate_ctl_io *ggio; 1064204076Spjd struct hio *hio; 1065204076Spjd unsigned int ncomp, rncomp; 1066204076Spjd ssize_t ret; 1067204076Spjd 1068204076Spjd /* Local component is 0 for now. */ 1069204076Spjd ncomp = 0; 1070204076Spjd /* Remote component is 1 for now. */ 1071204076Spjd rncomp = 1; 1072204076Spjd 1073204076Spjd for (;;) { 1074204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1075204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1076204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1077204076Spjd ggio = &hio->hio_ggio; 1078204076Spjd switch (ggio->gctl_cmd) { 1079204076Spjd case BIO_READ: 1080204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1081204076Spjd ggio->gctl_length, 1082204076Spjd ggio->gctl_offset + res->hr_localoff); 1083204076Spjd if (ret == ggio->gctl_length) 1084204076Spjd hio->hio_errors[ncomp] = 0; 1085204076Spjd else { 1086204076Spjd /* 1087204076Spjd * If READ failed, try to read from remote node. 1088204076Spjd */ 1089204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1090204076Spjd continue; 1091204076Spjd } 1092204076Spjd break; 1093204076Spjd case BIO_WRITE: 1094204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1095204076Spjd ggio->gctl_length, 1096204076Spjd ggio->gctl_offset + res->hr_localoff); 1097204076Spjd if (ret < 0) 1098204076Spjd hio->hio_errors[ncomp] = errno; 1099204076Spjd else if (ret != ggio->gctl_length) 1100204076Spjd hio->hio_errors[ncomp] = EIO; 1101204076Spjd else 1102204076Spjd hio->hio_errors[ncomp] = 0; 1103204076Spjd break; 1104204076Spjd case BIO_DELETE: 1105204076Spjd ret = g_delete(res->hr_localfd, 1106204076Spjd ggio->gctl_offset + res->hr_localoff, 1107204076Spjd ggio->gctl_length); 1108204076Spjd if (ret < 0) 1109204076Spjd hio->hio_errors[ncomp] = errno; 1110204076Spjd else 1111204076Spjd hio->hio_errors[ncomp] = 0; 1112204076Spjd break; 1113204076Spjd case BIO_FLUSH: 1114204076Spjd ret = g_flush(res->hr_localfd); 1115204076Spjd if (ret < 0) 1116204076Spjd hio->hio_errors[ncomp] = errno; 1117204076Spjd else 1118204076Spjd hio->hio_errors[ncomp] = 0; 1119204076Spjd break; 1120204076Spjd } 1121204076Spjd if (refcount_release(&hio->hio_countdown)) { 1122204076Spjd if (ISSYNCREQ(hio)) { 1123204076Spjd mtx_lock(&sync_lock); 1124204076Spjd SYNCREQDONE(hio); 1125204076Spjd mtx_unlock(&sync_lock); 1126204076Spjd cv_signal(&sync_cond); 1127204076Spjd } else { 1128204076Spjd pjdlog_debug(2, 1129204076Spjd "local_send: (%p) Moving request to the done queue.", 1130204076Spjd hio); 1131204076Spjd QUEUE_INSERT2(hio, done); 1132204076Spjd } 1133204076Spjd } 1134204076Spjd } 1135204076Spjd /* NOTREACHED */ 1136204076Spjd return (NULL); 1137204076Spjd} 1138204076Spjd 1139204076Spjd/* 1140204076Spjd * Thread sends request to secondary node. 1141204076Spjd */ 1142204076Spjdstatic void * 1143204076Spjdremote_send_thread(void *arg) 1144204076Spjd{ 1145204076Spjd struct hast_resource *res = arg; 1146204076Spjd struct g_gate_ctl_io *ggio; 1147204076Spjd struct hio *hio; 1148204076Spjd struct nv *nv; 1149204076Spjd unsigned int ncomp; 1150204076Spjd bool wakeup; 1151204076Spjd uint64_t offset, length; 1152204076Spjd uint8_t cmd; 1153204076Spjd void *data; 1154204076Spjd 1155204076Spjd /* Remote component is 1 for now. */ 1156204076Spjd ncomp = 1; 1157204076Spjd 1158204076Spjd for (;;) { 1159204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1160204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1161204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1162204076Spjd ggio = &hio->hio_ggio; 1163204076Spjd switch (ggio->gctl_cmd) { 1164204076Spjd case BIO_READ: 1165204076Spjd cmd = HIO_READ; 1166204076Spjd data = NULL; 1167204076Spjd offset = ggio->gctl_offset; 1168204076Spjd length = ggio->gctl_length; 1169204076Spjd break; 1170204076Spjd case BIO_WRITE: 1171204076Spjd cmd = HIO_WRITE; 1172204076Spjd data = ggio->gctl_data; 1173204076Spjd offset = ggio->gctl_offset; 1174204076Spjd length = ggio->gctl_length; 1175204076Spjd break; 1176204076Spjd case BIO_DELETE: 1177204076Spjd cmd = HIO_DELETE; 1178204076Spjd data = NULL; 1179204076Spjd offset = ggio->gctl_offset; 1180204076Spjd length = ggio->gctl_length; 1181204076Spjd break; 1182204076Spjd case BIO_FLUSH: 1183204076Spjd cmd = HIO_FLUSH; 1184204076Spjd data = NULL; 1185204076Spjd offset = 0; 1186204076Spjd length = 0; 1187204076Spjd break; 1188204076Spjd default: 1189204076Spjd assert(!"invalid condition"); 1190204076Spjd abort(); 1191204076Spjd } 1192204076Spjd nv = nv_alloc(); 1193204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1194204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1195204076Spjd nv_add_uint64(nv, offset, "offset"); 1196204076Spjd nv_add_uint64(nv, length, "length"); 1197204076Spjd if (nv_error(nv) != 0) { 1198204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1199204076Spjd pjdlog_debug(2, 1200204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1201204076Spjd hio); 1202204076Spjd reqlog(LOG_ERR, 0, ggio, 1203204076Spjd "Unable to prepare header to send (%s): ", 1204204076Spjd strerror(nv_error(nv))); 1205204076Spjd /* Move failed request immediately to the done queue. */ 1206204076Spjd goto done_queue; 1207204076Spjd } 1208204076Spjd pjdlog_debug(2, 1209204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1210204076Spjd hio); 1211204076Spjd /* 1212204076Spjd * Protect connection from disappearing. 1213204076Spjd */ 1214204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1215204076Spjd if (!ISCONNECTED(res, ncomp)) { 1216204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1217204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1218204076Spjd goto done_queue; 1219204076Spjd } 1220204076Spjd /* 1221204076Spjd * Move the request to recv queue before sending it, because 1222204076Spjd * in different order we can get reply before we move request 1223204076Spjd * to recv queue. 1224204076Spjd */ 1225204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1226204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1227204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1228204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1229204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1230204076Spjd data != NULL ? length : 0) < 0) { 1231204076Spjd hio->hio_errors[ncomp] = errno; 1232204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1233204076Spjd pjdlog_debug(2, 1234204076Spjd "remote_send: (%p) Unable to send request.", hio); 1235204076Spjd reqlog(LOG_ERR, 0, ggio, 1236204076Spjd "Unable to send request (%s): ", 1237204076Spjd strerror(hio->hio_errors[ncomp])); 1238211979Spjd remote_close(res, ncomp); 1239204076Spjd /* 1240204076Spjd * Take request back from the receive queue and move 1241204076Spjd * it immediately to the done queue. 1242204076Spjd */ 1243204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1244204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1245204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1246204076Spjd goto done_queue; 1247204076Spjd } 1248204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1249204076Spjd nv_free(nv); 1250204076Spjd if (wakeup) 1251204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1252204076Spjd continue; 1253204076Spjddone_queue: 1254204076Spjd nv_free(nv); 1255204076Spjd if (ISSYNCREQ(hio)) { 1256204076Spjd if (!refcount_release(&hio->hio_countdown)) 1257204076Spjd continue; 1258204076Spjd mtx_lock(&sync_lock); 1259204076Spjd SYNCREQDONE(hio); 1260204076Spjd mtx_unlock(&sync_lock); 1261204076Spjd cv_signal(&sync_cond); 1262204076Spjd continue; 1263204076Spjd } 1264204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1265204076Spjd mtx_lock(&res->hr_amp_lock); 1266204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1267204076Spjd ggio->gctl_length)) { 1268204076Spjd (void)hast_activemap_flush(res); 1269204076Spjd } 1270204076Spjd mtx_unlock(&res->hr_amp_lock); 1271204076Spjd } 1272204076Spjd if (!refcount_release(&hio->hio_countdown)) 1273204076Spjd continue; 1274204076Spjd pjdlog_debug(2, 1275204076Spjd "remote_send: (%p) Moving request to the done queue.", 1276204076Spjd hio); 1277204076Spjd QUEUE_INSERT2(hio, done); 1278204076Spjd } 1279204076Spjd /* NOTREACHED */ 1280204076Spjd return (NULL); 1281204076Spjd} 1282204076Spjd 1283204076Spjd/* 1284204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1285204076Spjd * thread. 1286204076Spjd */ 1287204076Spjdstatic void * 1288204076Spjdremote_recv_thread(void *arg) 1289204076Spjd{ 1290204076Spjd struct hast_resource *res = arg; 1291204076Spjd struct g_gate_ctl_io *ggio; 1292204076Spjd struct hio *hio; 1293204076Spjd struct nv *nv; 1294204076Spjd unsigned int ncomp; 1295204076Spjd uint64_t seq; 1296204076Spjd int error; 1297204076Spjd 1298204076Spjd /* Remote component is 1 for now. */ 1299204076Spjd ncomp = 1; 1300204076Spjd 1301204076Spjd for (;;) { 1302204076Spjd /* Wait until there is anything to receive. */ 1303204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1304204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1305204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1306204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1307204076Spjd &hio_recv_list_lock[ncomp]); 1308204076Spjd } 1309204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1310204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1311204076Spjd if (!ISCONNECTED(res, ncomp)) { 1312204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1313204076Spjd /* 1314204076Spjd * Connection is dead, so move all pending requests to 1315204076Spjd * the done queue (one-by-one). 1316204076Spjd */ 1317204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1318204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1319204076Spjd assert(hio != NULL); 1320204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1321204076Spjd hio_next[ncomp]); 1322204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1323204076Spjd goto done_queue; 1324204076Spjd } 1325204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1326204076Spjd pjdlog_errno(LOG_ERR, 1327204076Spjd "Unable to receive reply header"); 1328204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1329204076Spjd remote_close(res, ncomp); 1330204076Spjd continue; 1331204076Spjd } 1332204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1333204076Spjd seq = nv_get_uint64(nv, "seq"); 1334204076Spjd if (seq == 0) { 1335204076Spjd pjdlog_error("Header contains no 'seq' field."); 1336204076Spjd nv_free(nv); 1337204076Spjd continue; 1338204076Spjd } 1339204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1340204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1341204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1342204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1343204076Spjd hio_next[ncomp]); 1344204076Spjd break; 1345204076Spjd } 1346204076Spjd } 1347204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1348204076Spjd if (hio == NULL) { 1349204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1350204076Spjd (uintmax_t)seq); 1351204076Spjd nv_free(nv); 1352204076Spjd continue; 1353204076Spjd } 1354204076Spjd error = nv_get_int16(nv, "error"); 1355204076Spjd if (error != 0) { 1356204076Spjd /* Request failed on remote side. */ 1357204076Spjd hio->hio_errors[ncomp] = 0; 1358204076Spjd nv_free(nv); 1359204076Spjd goto done_queue; 1360204076Spjd } 1361204076Spjd ggio = &hio->hio_ggio; 1362204076Spjd switch (ggio->gctl_cmd) { 1363204076Spjd case BIO_READ: 1364204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1365204076Spjd if (!ISCONNECTED(res, ncomp)) { 1366204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1367204076Spjd nv_free(nv); 1368204076Spjd goto done_queue; 1369204076Spjd } 1370204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1371204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1372204076Spjd hio->hio_errors[ncomp] = errno; 1373204076Spjd pjdlog_errno(LOG_ERR, 1374204076Spjd "Unable to receive reply data"); 1375204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1376204076Spjd nv_free(nv); 1377204076Spjd remote_close(res, ncomp); 1378204076Spjd goto done_queue; 1379204076Spjd } 1380204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1381204076Spjd break; 1382204076Spjd case BIO_WRITE: 1383204076Spjd case BIO_DELETE: 1384204076Spjd case BIO_FLUSH: 1385204076Spjd break; 1386204076Spjd default: 1387204076Spjd assert(!"invalid condition"); 1388204076Spjd abort(); 1389204076Spjd } 1390204076Spjd hio->hio_errors[ncomp] = 0; 1391204076Spjd nv_free(nv); 1392204076Spjddone_queue: 1393204076Spjd if (refcount_release(&hio->hio_countdown)) { 1394204076Spjd if (ISSYNCREQ(hio)) { 1395204076Spjd mtx_lock(&sync_lock); 1396204076Spjd SYNCREQDONE(hio); 1397204076Spjd mtx_unlock(&sync_lock); 1398204076Spjd cv_signal(&sync_cond); 1399204076Spjd } else { 1400204076Spjd pjdlog_debug(2, 1401204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1402204076Spjd hio); 1403204076Spjd QUEUE_INSERT2(hio, done); 1404204076Spjd } 1405204076Spjd } 1406204076Spjd } 1407204076Spjd /* NOTREACHED */ 1408204076Spjd return (NULL); 1409204076Spjd} 1410204076Spjd 1411204076Spjd/* 1412204076Spjd * Thread sends answer to the kernel. 1413204076Spjd */ 1414204076Spjdstatic void * 1415204076Spjdggate_send_thread(void *arg) 1416204076Spjd{ 1417204076Spjd struct hast_resource *res = arg; 1418204076Spjd struct g_gate_ctl_io *ggio; 1419204076Spjd struct hio *hio; 1420204076Spjd unsigned int ii, ncomp, ncomps; 1421204076Spjd 1422204076Spjd ncomps = HAST_NCOMPONENTS; 1423204076Spjd 1424204076Spjd for (;;) { 1425204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1426204076Spjd QUEUE_TAKE2(hio, done); 1427204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1428204076Spjd ggio = &hio->hio_ggio; 1429204076Spjd for (ii = 0; ii < ncomps; ii++) { 1430204076Spjd if (hio->hio_errors[ii] == 0) { 1431204076Spjd /* 1432204076Spjd * One successful request is enough to declare 1433204076Spjd * success. 1434204076Spjd */ 1435204076Spjd ggio->gctl_error = 0; 1436204076Spjd break; 1437204076Spjd } 1438204076Spjd } 1439204076Spjd if (ii == ncomps) { 1440204076Spjd /* 1441204076Spjd * None of the requests were successful. 1442204076Spjd * Use first error. 1443204076Spjd */ 1444204076Spjd ggio->gctl_error = hio->hio_errors[0]; 1445204076Spjd } 1446204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1447204076Spjd mtx_lock(&res->hr_amp_lock); 1448204076Spjd activemap_write_complete(res->hr_amp, 1449204076Spjd ggio->gctl_offset, ggio->gctl_length); 1450204076Spjd mtx_unlock(&res->hr_amp_lock); 1451204076Spjd } 1452204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1453204076Spjd /* 1454204076Spjd * Unlock range we locked. 1455204076Spjd */ 1456204076Spjd mtx_lock(&range_lock); 1457204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1458204076Spjd ggio->gctl_length); 1459204076Spjd if (range_sync_wait) 1460204076Spjd cv_signal(&range_sync_cond); 1461204076Spjd mtx_unlock(&range_lock); 1462204076Spjd /* 1463204076Spjd * Bump local count if this is first write after 1464204076Spjd * connection failure with remote node. 1465204076Spjd */ 1466204076Spjd ncomp = 1; 1467204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1468204076Spjd if (!ISCONNECTED(res, ncomp)) { 1469204076Spjd mtx_lock(&metadata_lock); 1470204076Spjd if (res->hr_primary_localcnt == 1471204076Spjd res->hr_secondary_remotecnt) { 1472204076Spjd res->hr_primary_localcnt++; 1473204076Spjd pjdlog_debug(1, 1474204076Spjd "Increasing localcnt to %ju.", 1475204076Spjd (uintmax_t)res->hr_primary_localcnt); 1476204076Spjd (void)metadata_write(res); 1477204076Spjd } 1478204076Spjd mtx_unlock(&metadata_lock); 1479204076Spjd } 1480204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1481204076Spjd } 1482204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1483204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1484204076Spjd pjdlog_debug(2, 1485204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1486204076Spjd QUEUE_INSERT2(hio, free); 1487204076Spjd } 1488204076Spjd /* NOTREACHED */ 1489204076Spjd return (NULL); 1490204076Spjd} 1491204076Spjd 1492204076Spjd/* 1493204076Spjd * Thread synchronize local and remote components. 1494204076Spjd */ 1495204076Spjdstatic void * 1496204076Spjdsync_thread(void *arg __unused) 1497204076Spjd{ 1498204076Spjd struct hast_resource *res = arg; 1499204076Spjd struct hio *hio; 1500204076Spjd struct g_gate_ctl_io *ggio; 1501204076Spjd unsigned int ii, ncomp, ncomps; 1502204076Spjd off_t offset, length, synced; 1503204076Spjd bool dorewind; 1504204076Spjd int syncext; 1505204076Spjd 1506204076Spjd ncomps = HAST_NCOMPONENTS; 1507204076Spjd dorewind = true; 1508211897Spjd synced = 0; 1509211897Spjd offset = -1; 1510204076Spjd 1511204076Spjd for (;;) { 1512204076Spjd mtx_lock(&sync_lock); 1513211897Spjd if (offset >= 0 && !sync_inprogress) { 1514211879Spjd pjdlog_info("Synchronization interrupted. " 1515211879Spjd "%jd bytes synchronized so far.", 1516211879Spjd (intmax_t)synced); 1517212038Spjd event_send(res, EVENT_SYNCINTR); 1518211879Spjd } 1519204076Spjd while (!sync_inprogress) { 1520204076Spjd dorewind = true; 1521204076Spjd synced = 0; 1522204076Spjd cv_wait(&sync_cond, &sync_lock); 1523204076Spjd } 1524204076Spjd mtx_unlock(&sync_lock); 1525204076Spjd /* 1526204076Spjd * Obtain offset at which we should synchronize. 1527204076Spjd * Rewind synchronization if needed. 1528204076Spjd */ 1529204076Spjd mtx_lock(&res->hr_amp_lock); 1530204076Spjd if (dorewind) 1531204076Spjd activemap_sync_rewind(res->hr_amp); 1532204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1533204076Spjd if (syncext != -1) { 1534204076Spjd /* 1535204076Spjd * We synchronized entire syncext extent, we can mark 1536204076Spjd * it as clean now. 1537204076Spjd */ 1538204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1539204076Spjd (void)hast_activemap_flush(res); 1540204076Spjd } 1541204076Spjd mtx_unlock(&res->hr_amp_lock); 1542204076Spjd if (dorewind) { 1543204076Spjd dorewind = false; 1544204076Spjd if (offset < 0) 1545204076Spjd pjdlog_info("Nodes are in sync."); 1546204076Spjd else { 1547204076Spjd pjdlog_info("Synchronization started. %ju bytes to go.", 1548204076Spjd (uintmax_t)(res->hr_extentsize * 1549204076Spjd activemap_ndirty(res->hr_amp))); 1550212038Spjd event_send(res, EVENT_SYNCSTART); 1551204076Spjd } 1552204076Spjd } 1553204076Spjd if (offset < 0) { 1554211878Spjd sync_stop(); 1555204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1556204076Spjd /* 1557204076Spjd * Synchronization complete, make both localcnt and 1558204076Spjd * remotecnt equal. 1559204076Spjd */ 1560204076Spjd ncomp = 1; 1561204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1562204076Spjd if (ISCONNECTED(res, ncomp)) { 1563204076Spjd if (synced > 0) { 1564204076Spjd pjdlog_info("Synchronization complete. " 1565204076Spjd "%jd bytes synchronized.", 1566204076Spjd (intmax_t)synced); 1567212038Spjd event_send(res, EVENT_SYNCDONE); 1568204076Spjd } 1569204076Spjd mtx_lock(&metadata_lock); 1570204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1571204076Spjd res->hr_primary_localcnt = 1572204076Spjd res->hr_secondary_localcnt; 1573204076Spjd res->hr_primary_remotecnt = 1574204076Spjd res->hr_secondary_remotecnt; 1575204076Spjd pjdlog_debug(1, 1576204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1577204076Spjd (uintmax_t)res->hr_primary_localcnt, 1578204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1579204076Spjd (void)metadata_write(res); 1580204076Spjd mtx_unlock(&metadata_lock); 1581204076Spjd } 1582204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1583204076Spjd continue; 1584204076Spjd } 1585204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1586204076Spjd QUEUE_TAKE2(hio, free); 1587204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1588204076Spjd /* 1589204076Spjd * Lock the range we are going to synchronize. We don't want 1590204076Spjd * race where someone writes between our read and write. 1591204076Spjd */ 1592204076Spjd for (;;) { 1593204076Spjd mtx_lock(&range_lock); 1594204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1595204076Spjd pjdlog_debug(2, 1596204076Spjd "sync: Range offset=%jd length=%jd locked.", 1597204076Spjd (intmax_t)offset, (intmax_t)length); 1598204076Spjd range_sync_wait = true; 1599204076Spjd cv_wait(&range_sync_cond, &range_lock); 1600204076Spjd range_sync_wait = false; 1601204076Spjd mtx_unlock(&range_lock); 1602204076Spjd continue; 1603204076Spjd } 1604204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1605204076Spjd mtx_unlock(&range_lock); 1606204076Spjd pjdlog_debug(2, 1607204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1608204076Spjd (intmax_t)offset, (intmax_t)length); 1609204076Spjd sleep(1); 1610204076Spjd continue; 1611204076Spjd } 1612204076Spjd mtx_unlock(&range_lock); 1613204076Spjd break; 1614204076Spjd } 1615204076Spjd /* 1616204076Spjd * First read the data from synchronization source. 1617204076Spjd */ 1618204076Spjd SYNCREQ(hio); 1619204076Spjd ggio = &hio->hio_ggio; 1620204076Spjd ggio->gctl_cmd = BIO_READ; 1621204076Spjd ggio->gctl_offset = offset; 1622204076Spjd ggio->gctl_length = length; 1623204076Spjd ggio->gctl_error = 0; 1624204076Spjd for (ii = 0; ii < ncomps; ii++) 1625204076Spjd hio->hio_errors[ii] = EINVAL; 1626204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1627204076Spjd hio); 1628204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1629204076Spjd hio); 1630204076Spjd mtx_lock(&metadata_lock); 1631204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1632204076Spjd /* 1633204076Spjd * This range is up-to-date on local component, 1634204076Spjd * so handle request locally. 1635204076Spjd */ 1636204076Spjd /* Local component is 0 for now. */ 1637204076Spjd ncomp = 0; 1638204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1639204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1640204076Spjd /* 1641204076Spjd * This range is out-of-date on local component, 1642204076Spjd * so send request to the remote node. 1643204076Spjd */ 1644204076Spjd /* Remote component is 1 for now. */ 1645204076Spjd ncomp = 1; 1646204076Spjd } 1647204076Spjd mtx_unlock(&metadata_lock); 1648204076Spjd refcount_init(&hio->hio_countdown, 1); 1649204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1650204076Spjd 1651204076Spjd /* 1652204076Spjd * Let's wait for READ to finish. 1653204076Spjd */ 1654204076Spjd mtx_lock(&sync_lock); 1655204076Spjd while (!ISSYNCREQDONE(hio)) 1656204076Spjd cv_wait(&sync_cond, &sync_lock); 1657204076Spjd mtx_unlock(&sync_lock); 1658204076Spjd 1659204076Spjd if (hio->hio_errors[ncomp] != 0) { 1660204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1661204076Spjd strerror(hio->hio_errors[ncomp])); 1662204076Spjd goto free_queue; 1663204076Spjd } 1664204076Spjd 1665204076Spjd /* 1666204076Spjd * We read the data from synchronization source, now write it 1667204076Spjd * to synchronization target. 1668204076Spjd */ 1669204076Spjd SYNCREQ(hio); 1670204076Spjd ggio->gctl_cmd = BIO_WRITE; 1671204076Spjd for (ii = 0; ii < ncomps; ii++) 1672204076Spjd hio->hio_errors[ii] = EINVAL; 1673204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1674204076Spjd hio); 1675204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1676204076Spjd hio); 1677204076Spjd mtx_lock(&metadata_lock); 1678204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1679204076Spjd /* 1680204076Spjd * This range is up-to-date on local component, 1681204076Spjd * so we update remote component. 1682204076Spjd */ 1683204076Spjd /* Remote component is 1 for now. */ 1684204076Spjd ncomp = 1; 1685204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1686204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1687204076Spjd /* 1688204076Spjd * This range is out-of-date on local component, 1689204076Spjd * so we update it. 1690204076Spjd */ 1691204076Spjd /* Local component is 0 for now. */ 1692204076Spjd ncomp = 0; 1693204076Spjd } 1694204076Spjd mtx_unlock(&metadata_lock); 1695204076Spjd 1696204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1697204076Spjd hio); 1698204076Spjd refcount_init(&hio->hio_countdown, 1); 1699204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1700204076Spjd 1701204076Spjd /* 1702204076Spjd * Let's wait for WRITE to finish. 1703204076Spjd */ 1704204076Spjd mtx_lock(&sync_lock); 1705204076Spjd while (!ISSYNCREQDONE(hio)) 1706204076Spjd cv_wait(&sync_cond, &sync_lock); 1707204076Spjd mtx_unlock(&sync_lock); 1708204076Spjd 1709204076Spjd if (hio->hio_errors[ncomp] != 0) { 1710204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1711204076Spjd strerror(hio->hio_errors[ncomp])); 1712204076Spjd goto free_queue; 1713204076Spjd } 1714211880Spjd 1715211880Spjd synced += length; 1716204076Spjdfree_queue: 1717204076Spjd mtx_lock(&range_lock); 1718204076Spjd rangelock_del(range_sync, offset, length); 1719204076Spjd if (range_regular_wait) 1720204076Spjd cv_signal(&range_regular_cond); 1721204076Spjd mtx_unlock(&range_lock); 1722204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1723204076Spjd hio); 1724204076Spjd QUEUE_INSERT2(hio, free); 1725204076Spjd } 1726204076Spjd /* NOTREACHED */ 1727204076Spjd return (NULL); 1728204076Spjd} 1729204076Spjd 1730204076Spjdstatic void 1731210886Spjdconfig_reload(void) 1732210886Spjd{ 1733210886Spjd struct hastd_config *newcfg; 1734210886Spjd struct hast_resource *res; 1735210886Spjd unsigned int ii, ncomps; 1736210886Spjd int modified; 1737210886Spjd 1738210886Spjd pjdlog_info("Reloading configuration..."); 1739210886Spjd 1740210886Spjd ncomps = HAST_NCOMPONENTS; 1741210886Spjd 1742210886Spjd newcfg = yy_config_parse(cfgpath, false); 1743210886Spjd if (newcfg == NULL) 1744210886Spjd goto failed; 1745210886Spjd 1746210886Spjd TAILQ_FOREACH(res, &newcfg->hc_resources, hr_next) { 1747210886Spjd if (strcmp(res->hr_name, gres->hr_name) == 0) 1748210886Spjd break; 1749210886Spjd } 1750210886Spjd /* 1751210886Spjd * If resource was removed from the configuration file, resource 1752210886Spjd * name, provider name or path to local component was modified we 1753210886Spjd * shouldn't be here. This means that someone modified configuration 1754210886Spjd * file and send SIGHUP to us instead of main hastd process. 1755210886Spjd * Log advice and ignore the signal. 1756210886Spjd */ 1757210886Spjd if (res == NULL || strcmp(gres->hr_name, res->hr_name) != 0 || 1758210886Spjd strcmp(gres->hr_provname, res->hr_provname) != 0 || 1759210886Spjd strcmp(gres->hr_localpath, res->hr_localpath) != 0) { 1760210886Spjd pjdlog_warning("To reload configuration send SIGHUP to the main hastd process (pid %u).", 1761210886Spjd (unsigned int)getppid()); 1762210886Spjd goto failed; 1763210886Spjd } 1764210886Spjd 1765210886Spjd#define MODIFIED_REMOTEADDR 0x1 1766210886Spjd#define MODIFIED_REPLICATION 0x2 1767210886Spjd#define MODIFIED_TIMEOUT 0x4 1768211886Spjd#define MODIFIED_EXEC 0x8 1769210886Spjd modified = 0; 1770210886Spjd if (strcmp(gres->hr_remoteaddr, res->hr_remoteaddr) != 0) { 1771210886Spjd /* 1772210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1773210886Spjd * We want remote_close() to log disconnect from the old 1774210886Spjd * addresses, not from the new ones. 1775210886Spjd */ 1776210886Spjd modified |= MODIFIED_REMOTEADDR; 1777210886Spjd } 1778210886Spjd if (gres->hr_replication != res->hr_replication) { 1779210886Spjd gres->hr_replication = res->hr_replication; 1780210886Spjd modified |= MODIFIED_REPLICATION; 1781210886Spjd } 1782210886Spjd if (gres->hr_timeout != res->hr_timeout) { 1783210886Spjd gres->hr_timeout = res->hr_timeout; 1784210886Spjd modified |= MODIFIED_TIMEOUT; 1785210886Spjd } 1786211886Spjd if (strcmp(gres->hr_exec, res->hr_exec) != 0) { 1787211886Spjd strlcpy(gres->hr_exec, res->hr_exec, sizeof(gres->hr_exec)); 1788211886Spjd modified |= MODIFIED_EXEC; 1789211886Spjd } 1790210886Spjd /* 1791210886Spjd * If only timeout was modified we only need to change it without 1792210886Spjd * reconnecting. 1793210886Spjd */ 1794210886Spjd if (modified == MODIFIED_TIMEOUT) { 1795210886Spjd for (ii = 0; ii < ncomps; ii++) { 1796210886Spjd if (!ISREMOTE(ii)) 1797210886Spjd continue; 1798210886Spjd rw_rlock(&hio_remote_lock[ii]); 1799210886Spjd if (!ISCONNECTED(gres, ii)) { 1800210886Spjd rw_unlock(&hio_remote_lock[ii]); 1801210886Spjd continue; 1802210886Spjd } 1803210886Spjd rw_unlock(&hio_remote_lock[ii]); 1804210886Spjd if (proto_timeout(gres->hr_remotein, 1805210886Spjd gres->hr_timeout) < 0) { 1806210886Spjd pjdlog_errno(LOG_WARNING, 1807210886Spjd "Unable to set connection timeout"); 1808210886Spjd } 1809210886Spjd if (proto_timeout(gres->hr_remoteout, 1810210886Spjd gres->hr_timeout) < 0) { 1811210886Spjd pjdlog_errno(LOG_WARNING, 1812210886Spjd "Unable to set connection timeout"); 1813210886Spjd } 1814210886Spjd } 1815211886Spjd } else if ((modified & 1816211886Spjd (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1817210886Spjd for (ii = 0; ii < ncomps; ii++) { 1818210886Spjd if (!ISREMOTE(ii)) 1819210886Spjd continue; 1820210886Spjd remote_close(gres, ii); 1821210886Spjd } 1822210886Spjd if (modified & MODIFIED_REMOTEADDR) { 1823210886Spjd strlcpy(gres->hr_remoteaddr, res->hr_remoteaddr, 1824210886Spjd sizeof(gres->hr_remoteaddr)); 1825210886Spjd } 1826210886Spjd } 1827210886Spjd#undef MODIFIED_REMOTEADDR 1828210886Spjd#undef MODIFIED_REPLICATION 1829210886Spjd#undef MODIFIED_TIMEOUT 1830211886Spjd#undef MODIFIED_EXEC 1831210886Spjd 1832210886Spjd pjdlog_info("Configuration reloaded successfully."); 1833210886Spjd return; 1834210886Spjdfailed: 1835210886Spjd if (newcfg != NULL) { 1836210886Spjd if (newcfg->hc_controlconn != NULL) 1837210886Spjd proto_close(newcfg->hc_controlconn); 1838210886Spjd if (newcfg->hc_listenconn != NULL) 1839210886Spjd proto_close(newcfg->hc_listenconn); 1840210886Spjd yy_config_free(newcfg); 1841210886Spjd } 1842210886Spjd pjdlog_warning("Configuration not reloaded."); 1843210886Spjd} 1844210886Spjd 1845211882Spjdstatic void 1846211882Spjdkeepalive_send(struct hast_resource *res, unsigned int ncomp) 1847211882Spjd{ 1848211882Spjd struct nv *nv; 1849211882Spjd 1850211882Spjd nv = nv_alloc(); 1851211882Spjd nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1852211882Spjd if (nv_error(nv) != 0) { 1853211882Spjd nv_free(nv); 1854211882Spjd pjdlog_debug(1, 1855211882Spjd "keepalive_send: Unable to prepare header to send."); 1856211882Spjd return; 1857211882Spjd } 1858211882Spjd if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1859211882Spjd pjdlog_common(LOG_DEBUG, 1, errno, 1860211882Spjd "keepalive_send: Unable to send request"); 1861211882Spjd nv_free(nv); 1862211882Spjd rw_unlock(&hio_remote_lock[ncomp]); 1863211882Spjd remote_close(res, ncomp); 1864211882Spjd rw_rlock(&hio_remote_lock[ncomp]); 1865211882Spjd return; 1866211882Spjd } 1867211882Spjd nv_free(nv); 1868211882Spjd pjdlog_debug(2, "keepalive_send: Request sent."); 1869211882Spjd} 1870211882Spjd 1871211981Spjdstatic void 1872211981Spjdguard_one(struct hast_resource *res, unsigned int ncomp) 1873211981Spjd{ 1874211981Spjd struct proto_conn *in, *out; 1875211981Spjd 1876211981Spjd if (!ISREMOTE(ncomp)) 1877211981Spjd return; 1878211981Spjd 1879211981Spjd rw_rlock(&hio_remote_lock[ncomp]); 1880211981Spjd 1881211981Spjd if (!real_remote(res)) { 1882211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1883211981Spjd return; 1884211981Spjd } 1885211981Spjd 1886211981Spjd if (ISCONNECTED(res, ncomp)) { 1887211981Spjd assert(res->hr_remotein != NULL); 1888211981Spjd assert(res->hr_remoteout != NULL); 1889211981Spjd keepalive_send(res, ncomp); 1890211981Spjd } 1891211981Spjd 1892211981Spjd if (ISCONNECTED(res, ncomp)) { 1893211981Spjd assert(res->hr_remotein != NULL); 1894211981Spjd assert(res->hr_remoteout != NULL); 1895211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1896211981Spjd pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 1897211981Spjd res->hr_remoteaddr); 1898211981Spjd return; 1899211981Spjd } 1900211981Spjd 1901211981Spjd assert(res->hr_remotein == NULL); 1902211981Spjd assert(res->hr_remoteout == NULL); 1903211981Spjd /* 1904211981Spjd * Upgrade the lock. It doesn't have to be atomic as no other thread 1905211981Spjd * can change connection status from disconnected to connected. 1906211981Spjd */ 1907211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1908211981Spjd pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 1909211981Spjd res->hr_remoteaddr); 1910211981Spjd in = out = NULL; 1911211981Spjd if (init_remote(res, &in, &out)) { 1912211981Spjd rw_wlock(&hio_remote_lock[ncomp]); 1913211981Spjd assert(res->hr_remotein == NULL); 1914211981Spjd assert(res->hr_remoteout == NULL); 1915211981Spjd assert(in != NULL && out != NULL); 1916211981Spjd res->hr_remotein = in; 1917211981Spjd res->hr_remoteout = out; 1918211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1919211981Spjd pjdlog_info("Successfully reconnected to %s.", 1920211981Spjd res->hr_remoteaddr); 1921211981Spjd sync_start(); 1922211981Spjd } else { 1923211981Spjd /* Both connections should be NULL. */ 1924211981Spjd assert(res->hr_remotein == NULL); 1925211981Spjd assert(res->hr_remoteout == NULL); 1926211981Spjd assert(in == NULL && out == NULL); 1927211981Spjd pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 1928211981Spjd res->hr_remoteaddr); 1929211981Spjd } 1930211981Spjd} 1931211981Spjd 1932204076Spjd/* 1933204076Spjd * Thread guards remote connections and reconnects when needed, handles 1934204076Spjd * signals, etc. 1935204076Spjd */ 1936204076Spjdstatic void * 1937204076Spjdguard_thread(void *arg) 1938204076Spjd{ 1939204076Spjd struct hast_resource *res = arg; 1940204076Spjd unsigned int ii, ncomps; 1941211982Spjd struct timespec timeout; 1942211981Spjd time_t lastcheck, now; 1943211982Spjd sigset_t mask; 1944211982Spjd int signo; 1945204076Spjd 1946204076Spjd ncomps = HAST_NCOMPONENTS; 1947211981Spjd lastcheck = time(NULL); 1948204076Spjd 1949211982Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1950211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1951211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1952211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1953211982Spjd 1954211982Spjd timeout.tv_nsec = 0; 1955211982Spjd signo = -1; 1956211982Spjd 1957204076Spjd for (;;) { 1958211982Spjd switch (signo) { 1959211982Spjd case SIGHUP: 1960211982Spjd config_reload(); 1961211982Spjd break; 1962211982Spjd case SIGINT: 1963211982Spjd case SIGTERM: 1964211982Spjd sigexit_received = true; 1965204076Spjd primary_exitx(EX_OK, 1966204076Spjd "Termination signal received, exiting."); 1967211982Spjd break; 1968211982Spjd default: 1969211982Spjd break; 1970204076Spjd } 1971211882Spjd 1972204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 1973211981Spjd now = time(NULL); 1974211982Spjd if (lastcheck + RETRY_SLEEP <= now) { 1975211982Spjd for (ii = 0; ii < ncomps; ii++) 1976211981Spjd guard_one(res, ii); 1977211981Spjd lastcheck = now; 1978204076Spjd } 1979211982Spjd timeout.tv_sec = RETRY_SLEEP; 1980211982Spjd signo = sigtimedwait(&mask, NULL, &timeout); 1981204076Spjd } 1982204076Spjd /* NOTREACHED */ 1983204076Spjd return (NULL); 1984204076Spjd} 1985