primary.c revision 219879
1204076Spjd/*- 2204076Spjd * Copyright (c) 2009 The FreeBSD Foundation 3219351Spjd * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net> 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 219879 2011-03-22 19:49:27Z trociny $"); 33204076Spjd 34204076Spjd#include <sys/types.h> 35204076Spjd#include <sys/time.h> 36204076Spjd#include <sys/bio.h> 37204076Spjd#include <sys/disk.h> 38204076Spjd#include <sys/refcount.h> 39204076Spjd#include <sys/stat.h> 40204076Spjd 41204076Spjd#include <geom/gate/g_gate.h> 42204076Spjd 43204076Spjd#include <err.h> 44204076Spjd#include <errno.h> 45204076Spjd#include <fcntl.h> 46204076Spjd#include <libgeom.h> 47204076Spjd#include <pthread.h> 48211982Spjd#include <signal.h> 49204076Spjd#include <stdint.h> 50204076Spjd#include <stdio.h> 51204076Spjd#include <string.h> 52204076Spjd#include <sysexits.h> 53204076Spjd#include <unistd.h> 54204076Spjd 55204076Spjd#include <activemap.h> 56204076Spjd#include <nv.h> 57204076Spjd#include <rangelock.h> 58204076Spjd 59204076Spjd#include "control.h" 60212038Spjd#include "event.h" 61204076Spjd#include "hast.h" 62204076Spjd#include "hast_proto.h" 63204076Spjd#include "hastd.h" 64211886Spjd#include "hooks.h" 65204076Spjd#include "metadata.h" 66204076Spjd#include "proto.h" 67204076Spjd#include "pjdlog.h" 68204076Spjd#include "subr.h" 69204076Spjd#include "synch.h" 70204076Spjd 71210886Spjd/* The is only one remote component for now. */ 72210886Spjd#define ISREMOTE(no) ((no) == 1) 73210886Spjd 74204076Spjdstruct hio { 75204076Spjd /* 76204076Spjd * Number of components we are still waiting for. 77204076Spjd * When this field goes to 0, we can send the request back to the 78204076Spjd * kernel. Each component has to decrease this counter by one 79204076Spjd * even on failure. 80204076Spjd */ 81204076Spjd unsigned int hio_countdown; 82204076Spjd /* 83204076Spjd * Each component has a place to store its own error. 84204076Spjd * Once the request is handled by all components we can decide if the 85204076Spjd * request overall is successful or not. 86204076Spjd */ 87204076Spjd int *hio_errors; 88204076Spjd /* 89219818Spjd * Structure used to communicate with GEOM Gate class. 90204076Spjd */ 91204076Spjd struct g_gate_ctl_io hio_ggio; 92204076Spjd TAILQ_ENTRY(hio) *hio_next; 93204076Spjd}; 94204076Spjd#define hio_free_next hio_next[0] 95204076Spjd#define hio_done_next hio_next[0] 96204076Spjd 97204076Spjd/* 98204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 99204076Spjd * until some in-progress requests are freed. 100204076Spjd */ 101204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 102204076Spjdstatic pthread_mutex_t hio_free_list_lock; 103204076Spjdstatic pthread_cond_t hio_free_list_cond; 104204076Spjd/* 105204076Spjd * There is one send list for every component. One requests is placed on all 106204076Spjd * send lists - each component gets the same request, but each component is 107204076Spjd * responsible for managing his own send list. 108204076Spjd */ 109204076Spjdstatic TAILQ_HEAD(, hio) *hio_send_list; 110204076Spjdstatic pthread_mutex_t *hio_send_list_lock; 111204076Spjdstatic pthread_cond_t *hio_send_list_cond; 112204076Spjd/* 113204076Spjd * There is one recv list for every component, although local components don't 114204076Spjd * use recv lists as local requests are done synchronously. 115204076Spjd */ 116204076Spjdstatic TAILQ_HEAD(, hio) *hio_recv_list; 117204076Spjdstatic pthread_mutex_t *hio_recv_list_lock; 118204076Spjdstatic pthread_cond_t *hio_recv_list_cond; 119204076Spjd/* 120204076Spjd * Request is placed on done list by the slowest component (the one that 121204076Spjd * decreased hio_countdown from 1 to 0). 122204076Spjd */ 123204076Spjdstatic TAILQ_HEAD(, hio) hio_done_list; 124204076Spjdstatic pthread_mutex_t hio_done_list_lock; 125204076Spjdstatic pthread_cond_t hio_done_list_cond; 126204076Spjd/* 127204076Spjd * Structure below are for interaction with sync thread. 128204076Spjd */ 129204076Spjdstatic bool sync_inprogress; 130204076Spjdstatic pthread_mutex_t sync_lock; 131204076Spjdstatic pthread_cond_t sync_cond; 132204076Spjd/* 133204076Spjd * The lock below allows to synchornize access to remote connections. 134204076Spjd */ 135204076Spjdstatic pthread_rwlock_t *hio_remote_lock; 136204076Spjd 137204076Spjd/* 138204076Spjd * Lock to synchronize metadata updates. Also synchronize access to 139204076Spjd * hr_primary_localcnt and hr_primary_remotecnt fields. 140204076Spjd */ 141204076Spjdstatic pthread_mutex_t metadata_lock; 142204076Spjd 143204076Spjd/* 144204076Spjd * Maximum number of outstanding I/O requests. 145204076Spjd */ 146204076Spjd#define HAST_HIO_MAX 256 147204076Spjd/* 148204076Spjd * Number of components. At this point there are only two components: local 149204076Spjd * and remote, but in the future it might be possible to use multiple local 150204076Spjd * and remote components. 151204076Spjd */ 152204076Spjd#define HAST_NCOMPONENTS 2 153204076Spjd 154204076Spjd#define ISCONNECTED(res, no) \ 155204076Spjd ((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL) 156204076Spjd 157204076Spjd#define QUEUE_INSERT1(hio, name, ncomp) do { \ 158204076Spjd bool _wakeup; \ 159204076Spjd \ 160204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 161204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]); \ 162204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio), \ 163204076Spjd hio_next[(ncomp)]); \ 164204076Spjd mtx_unlock(&hio_##name##_list_lock[ncomp]); \ 165204076Spjd if (_wakeup) \ 166204076Spjd cv_signal(&hio_##name##_list_cond[(ncomp)]); \ 167204076Spjd} while (0) 168204076Spjd#define QUEUE_INSERT2(hio, name) do { \ 169204076Spjd bool _wakeup; \ 170204076Spjd \ 171204076Spjd mtx_lock(&hio_##name##_list_lock); \ 172204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 173204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\ 174204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 175204076Spjd if (_wakeup) \ 176204076Spjd cv_signal(&hio_##name##_list_cond); \ 177204076Spjd} while (0) 178214692Spjd#define QUEUE_TAKE1(hio, name, ncomp, timeout) do { \ 179214692Spjd bool _last; \ 180214692Spjd \ 181204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 182214692Spjd _last = false; \ 183214692Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL && !_last) { \ 184214692Spjd cv_timedwait(&hio_##name##_list_cond[(ncomp)], \ 185214692Spjd &hio_##name##_list_lock[(ncomp)], (timeout)); \ 186219864Spjd if ((timeout) != 0) \ 187214692Spjd _last = true; \ 188204076Spjd } \ 189214692Spjd if (hio != NULL) { \ 190214692Spjd TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio), \ 191214692Spjd hio_next[(ncomp)]); \ 192214692Spjd } \ 193204076Spjd mtx_unlock(&hio_##name##_list_lock[(ncomp)]); \ 194204076Spjd} while (0) 195204076Spjd#define QUEUE_TAKE2(hio, name) do { \ 196204076Spjd mtx_lock(&hio_##name##_list_lock); \ 197204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 198204076Spjd cv_wait(&hio_##name##_list_cond, \ 199204076Spjd &hio_##name##_list_lock); \ 200204076Spjd } \ 201204076Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next); \ 202204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 203204076Spjd} while (0) 204204076Spjd 205209183Spjd#define SYNCREQ(hio) do { \ 206209183Spjd (hio)->hio_ggio.gctl_unit = -1; \ 207209183Spjd (hio)->hio_ggio.gctl_seq = 1; \ 208209183Spjd} while (0) 209204076Spjd#define ISSYNCREQ(hio) ((hio)->hio_ggio.gctl_unit == -1) 210204076Spjd#define SYNCREQDONE(hio) do { (hio)->hio_ggio.gctl_unit = -2; } while (0) 211204076Spjd#define ISSYNCREQDONE(hio) ((hio)->hio_ggio.gctl_unit == -2) 212204076Spjd 213204076Spjdstatic struct hast_resource *gres; 214204076Spjd 215204076Spjdstatic pthread_mutex_t range_lock; 216204076Spjdstatic struct rangelocks *range_regular; 217204076Spjdstatic bool range_regular_wait; 218204076Spjdstatic pthread_cond_t range_regular_cond; 219204076Spjdstatic struct rangelocks *range_sync; 220204076Spjdstatic bool range_sync_wait; 221204076Spjdstatic pthread_cond_t range_sync_cond; 222204076Spjd 223204076Spjdstatic void *ggate_recv_thread(void *arg); 224204076Spjdstatic void *local_send_thread(void *arg); 225204076Spjdstatic void *remote_send_thread(void *arg); 226204076Spjdstatic void *remote_recv_thread(void *arg); 227204076Spjdstatic void *ggate_send_thread(void *arg); 228204076Spjdstatic void *sync_thread(void *arg); 229204076Spjdstatic void *guard_thread(void *arg); 230204076Spjd 231211982Spjdstatic void 232204076Spjdcleanup(struct hast_resource *res) 233204076Spjd{ 234204076Spjd int rerrno; 235204076Spjd 236204076Spjd /* Remember errno. */ 237204076Spjd rerrno = errno; 238204076Spjd 239204076Spjd /* Destroy ggate provider if we created one. */ 240204076Spjd if (res->hr_ggateunit >= 0) { 241204076Spjd struct g_gate_ctl_destroy ggiod; 242204076Spjd 243213533Spjd bzero(&ggiod, sizeof(ggiod)); 244204076Spjd ggiod.gctl_version = G_GATE_VERSION; 245204076Spjd ggiod.gctl_unit = res->hr_ggateunit; 246204076Spjd ggiod.gctl_force = 1; 247204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 248213531Spjd pjdlog_errno(LOG_WARNING, 249213531Spjd "Unable to destroy hast/%s device", 250204076Spjd res->hr_provname); 251204076Spjd } 252204076Spjd res->hr_ggateunit = -1; 253204076Spjd } 254204076Spjd 255204076Spjd /* Restore errno. */ 256204076Spjd errno = rerrno; 257204076Spjd} 258204076Spjd 259212899Spjdstatic __dead2 void 260204076Spjdprimary_exit(int exitcode, const char *fmt, ...) 261204076Spjd{ 262204076Spjd va_list ap; 263204076Spjd 264218138Spjd PJDLOG_ASSERT(exitcode != EX_OK); 265204076Spjd va_start(ap, fmt); 266204076Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 267204076Spjd va_end(ap); 268204076Spjd cleanup(gres); 269204076Spjd exit(exitcode); 270204076Spjd} 271204076Spjd 272212899Spjdstatic __dead2 void 273204076Spjdprimary_exitx(int exitcode, const char *fmt, ...) 274204076Spjd{ 275204076Spjd va_list ap; 276204076Spjd 277204076Spjd va_start(ap, fmt); 278204076Spjd pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 279204076Spjd va_end(ap); 280204076Spjd cleanup(gres); 281204076Spjd exit(exitcode); 282204076Spjd} 283204076Spjd 284204076Spjdstatic int 285204076Spjdhast_activemap_flush(struct hast_resource *res) 286204076Spjd{ 287204076Spjd const unsigned char *buf; 288204076Spjd size_t size; 289204076Spjd 290204076Spjd buf = activemap_bitmap(res->hr_amp, &size); 291218138Spjd PJDLOG_ASSERT(buf != NULL); 292218138Spjd PJDLOG_ASSERT((size % res->hr_local_sectorsize) == 0); 293204076Spjd if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 294204076Spjd (ssize_t)size) { 295204076Spjd KEEP_ERRNO(pjdlog_errno(LOG_ERR, 296204076Spjd "Unable to flush activemap to disk")); 297204076Spjd return (-1); 298204076Spjd } 299204076Spjd return (0); 300204076Spjd} 301204076Spjd 302210881Spjdstatic bool 303210881Spjdreal_remote(const struct hast_resource *res) 304210881Spjd{ 305210881Spjd 306210881Spjd return (strcmp(res->hr_remoteaddr, "none") != 0); 307210881Spjd} 308210881Spjd 309204076Spjdstatic void 310204076Spjdinit_environment(struct hast_resource *res __unused) 311204076Spjd{ 312204076Spjd struct hio *hio; 313204076Spjd unsigned int ii, ncomps; 314204076Spjd 315204076Spjd /* 316204076Spjd * In the future it might be per-resource value. 317204076Spjd */ 318204076Spjd ncomps = HAST_NCOMPONENTS; 319204076Spjd 320204076Spjd /* 321204076Spjd * Allocate memory needed by lists. 322204076Spjd */ 323204076Spjd hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 324204076Spjd if (hio_send_list == NULL) { 325204076Spjd primary_exitx(EX_TEMPFAIL, 326204076Spjd "Unable to allocate %zu bytes of memory for send lists.", 327204076Spjd sizeof(hio_send_list[0]) * ncomps); 328204076Spjd } 329204076Spjd hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 330204076Spjd if (hio_send_list_lock == NULL) { 331204076Spjd primary_exitx(EX_TEMPFAIL, 332204076Spjd "Unable to allocate %zu bytes of memory for send list locks.", 333204076Spjd sizeof(hio_send_list_lock[0]) * ncomps); 334204076Spjd } 335204076Spjd hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 336204076Spjd if (hio_send_list_cond == NULL) { 337204076Spjd primary_exitx(EX_TEMPFAIL, 338204076Spjd "Unable to allocate %zu bytes of memory for send list condition variables.", 339204076Spjd sizeof(hio_send_list_cond[0]) * ncomps); 340204076Spjd } 341204076Spjd hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 342204076Spjd if (hio_recv_list == NULL) { 343204076Spjd primary_exitx(EX_TEMPFAIL, 344204076Spjd "Unable to allocate %zu bytes of memory for recv lists.", 345204076Spjd sizeof(hio_recv_list[0]) * ncomps); 346204076Spjd } 347204076Spjd hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 348204076Spjd if (hio_recv_list_lock == NULL) { 349204076Spjd primary_exitx(EX_TEMPFAIL, 350204076Spjd "Unable to allocate %zu bytes of memory for recv list locks.", 351204076Spjd sizeof(hio_recv_list_lock[0]) * ncomps); 352204076Spjd } 353204076Spjd hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 354204076Spjd if (hio_recv_list_cond == NULL) { 355204076Spjd primary_exitx(EX_TEMPFAIL, 356204076Spjd "Unable to allocate %zu bytes of memory for recv list condition variables.", 357204076Spjd sizeof(hio_recv_list_cond[0]) * ncomps); 358204076Spjd } 359204076Spjd hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 360204076Spjd if (hio_remote_lock == NULL) { 361204076Spjd primary_exitx(EX_TEMPFAIL, 362204076Spjd "Unable to allocate %zu bytes of memory for remote connections locks.", 363204076Spjd sizeof(hio_remote_lock[0]) * ncomps); 364204076Spjd } 365204076Spjd 366204076Spjd /* 367204076Spjd * Initialize lists, their locks and theirs condition variables. 368204076Spjd */ 369204076Spjd TAILQ_INIT(&hio_free_list); 370204076Spjd mtx_init(&hio_free_list_lock); 371204076Spjd cv_init(&hio_free_list_cond); 372204076Spjd for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 373204076Spjd TAILQ_INIT(&hio_send_list[ii]); 374204076Spjd mtx_init(&hio_send_list_lock[ii]); 375204076Spjd cv_init(&hio_send_list_cond[ii]); 376204076Spjd TAILQ_INIT(&hio_recv_list[ii]); 377204076Spjd mtx_init(&hio_recv_list_lock[ii]); 378204076Spjd cv_init(&hio_recv_list_cond[ii]); 379204076Spjd rw_init(&hio_remote_lock[ii]); 380204076Spjd } 381204076Spjd TAILQ_INIT(&hio_done_list); 382204076Spjd mtx_init(&hio_done_list_lock); 383204076Spjd cv_init(&hio_done_list_cond); 384204076Spjd mtx_init(&metadata_lock); 385204076Spjd 386204076Spjd /* 387204076Spjd * Allocate requests pool and initialize requests. 388204076Spjd */ 389204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 390204076Spjd hio = malloc(sizeof(*hio)); 391204076Spjd if (hio == NULL) { 392204076Spjd primary_exitx(EX_TEMPFAIL, 393204076Spjd "Unable to allocate %zu bytes of memory for hio request.", 394204076Spjd sizeof(*hio)); 395204076Spjd } 396204076Spjd hio->hio_countdown = 0; 397204076Spjd hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 398204076Spjd if (hio->hio_errors == NULL) { 399204076Spjd primary_exitx(EX_TEMPFAIL, 400204076Spjd "Unable allocate %zu bytes of memory for hio errors.", 401204076Spjd sizeof(hio->hio_errors[0]) * ncomps); 402204076Spjd } 403204076Spjd hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 404204076Spjd if (hio->hio_next == NULL) { 405204076Spjd primary_exitx(EX_TEMPFAIL, 406204076Spjd "Unable allocate %zu bytes of memory for hio_next field.", 407204076Spjd sizeof(hio->hio_next[0]) * ncomps); 408204076Spjd } 409204076Spjd hio->hio_ggio.gctl_version = G_GATE_VERSION; 410204076Spjd hio->hio_ggio.gctl_data = malloc(MAXPHYS); 411204076Spjd if (hio->hio_ggio.gctl_data == NULL) { 412204076Spjd primary_exitx(EX_TEMPFAIL, 413204076Spjd "Unable to allocate %zu bytes of memory for gctl_data.", 414204076Spjd MAXPHYS); 415204076Spjd } 416204076Spjd hio->hio_ggio.gctl_length = MAXPHYS; 417204076Spjd hio->hio_ggio.gctl_error = 0; 418204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 419204076Spjd } 420204076Spjd} 421204076Spjd 422214284Spjdstatic bool 423214284Spjdinit_resuid(struct hast_resource *res) 424214284Spjd{ 425214284Spjd 426214284Spjd mtx_lock(&metadata_lock); 427214284Spjd if (res->hr_resuid != 0) { 428214284Spjd mtx_unlock(&metadata_lock); 429214284Spjd return (false); 430214284Spjd } else { 431214284Spjd /* Initialize unique resource identifier. */ 432214284Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 433214284Spjd mtx_unlock(&metadata_lock); 434214284Spjd if (metadata_write(res) < 0) 435214284Spjd exit(EX_NOINPUT); 436214284Spjd return (true); 437214284Spjd } 438214284Spjd} 439214284Spjd 440204076Spjdstatic void 441204076Spjdinit_local(struct hast_resource *res) 442204076Spjd{ 443204076Spjd unsigned char *buf; 444204076Spjd size_t mapsize; 445204076Spjd 446204076Spjd if (metadata_read(res, true) < 0) 447204076Spjd exit(EX_NOINPUT); 448204076Spjd mtx_init(&res->hr_amp_lock); 449204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 450204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 451204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 452204076Spjd } 453204076Spjd mtx_init(&range_lock); 454204076Spjd cv_init(&range_regular_cond); 455204076Spjd if (rangelock_init(&range_regular) < 0) 456204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 457204076Spjd cv_init(&range_sync_cond); 458204076Spjd if (rangelock_init(&range_sync) < 0) 459204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 460204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 461204076Spjd buf = calloc(1, mapsize); 462204076Spjd if (buf == NULL) { 463204076Spjd primary_exitx(EX_TEMPFAIL, 464204076Spjd "Unable to allocate buffer for activemap."); 465204076Spjd } 466204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 467204076Spjd (ssize_t)mapsize) { 468204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 469204076Spjd } 470204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 471209181Spjd free(buf); 472204076Spjd if (res->hr_resuid != 0) 473204076Spjd return; 474204076Spjd /* 475214284Spjd * We're using provider for the first time. Initialize local and remote 476214284Spjd * counters. We don't initialize resuid here, as we want to do it just 477214284Spjd * in time. The reason for this is that we want to inform secondary 478214284Spjd * that there were no writes yet, so there is no need to synchronize 479214284Spjd * anything. 480204076Spjd */ 481219844Spjd res->hr_primary_localcnt = 0; 482204076Spjd res->hr_primary_remotecnt = 0; 483204076Spjd if (metadata_write(res) < 0) 484204076Spjd exit(EX_NOINPUT); 485204076Spjd} 486204076Spjd 487218218Spjdstatic int 488218218Spjdprimary_connect(struct hast_resource *res, struct proto_conn **connp) 489218218Spjd{ 490218218Spjd struct proto_conn *conn; 491218218Spjd int16_t val; 492218218Spjd 493218218Spjd val = 1; 494218218Spjd if (proto_send(res->hr_conn, &val, sizeof(val)) < 0) { 495218218Spjd primary_exit(EX_TEMPFAIL, 496218218Spjd "Unable to send connection request to parent"); 497218218Spjd } 498218218Spjd if (proto_recv(res->hr_conn, &val, sizeof(val)) < 0) { 499218218Spjd primary_exit(EX_TEMPFAIL, 500218218Spjd "Unable to receive reply to connection request from parent"); 501218218Spjd } 502218218Spjd if (val != 0) { 503218218Spjd errno = val; 504218218Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 505218218Spjd res->hr_remoteaddr); 506218218Spjd return (-1); 507218218Spjd } 508218218Spjd if (proto_connection_recv(res->hr_conn, true, &conn) < 0) { 509218218Spjd primary_exit(EX_TEMPFAIL, 510218218Spjd "Unable to receive connection from parent"); 511218218Spjd } 512218218Spjd if (proto_connect_wait(conn, HAST_TIMEOUT) < 0) { 513218218Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 514218218Spjd res->hr_remoteaddr); 515218218Spjd proto_close(conn); 516218218Spjd return (-1); 517218218Spjd } 518218218Spjd /* Error in setting timeout is not critical, but why should it fail? */ 519218218Spjd if (proto_timeout(conn, res->hr_timeout) < 0) 520218218Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 521218218Spjd 522218218Spjd *connp = conn; 523218218Spjd 524218218Spjd return (0); 525218218Spjd} 526218218Spjd 527205738Spjdstatic bool 528205738Spjdinit_remote(struct hast_resource *res, struct proto_conn **inp, 529205738Spjd struct proto_conn **outp) 530204076Spjd{ 531205738Spjd struct proto_conn *in, *out; 532204076Spjd struct nv *nvout, *nvin; 533204076Spjd const unsigned char *token; 534204076Spjd unsigned char *map; 535204076Spjd const char *errmsg; 536204076Spjd int32_t extentsize; 537204076Spjd int64_t datasize; 538204076Spjd uint32_t mapsize; 539204076Spjd size_t size; 540204076Spjd 541218138Spjd PJDLOG_ASSERT((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 542218138Spjd PJDLOG_ASSERT(real_remote(res)); 543205738Spjd 544205738Spjd in = out = NULL; 545211983Spjd errmsg = NULL; 546205738Spjd 547218218Spjd if (primary_connect(res, &out) == -1) 548218218Spjd return (false); 549218218Spjd 550204076Spjd /* 551204076Spjd * First handshake step. 552204076Spjd * Setup outgoing connection with remote node. 553204076Spjd */ 554204076Spjd nvout = nv_alloc(); 555204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 556204076Spjd if (nv_error(nvout) != 0) { 557204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 558204076Spjd "Unable to allocate header for connection with %s", 559204076Spjd res->hr_remoteaddr); 560204076Spjd nv_free(nvout); 561204076Spjd goto close; 562204076Spjd } 563205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 564204076Spjd pjdlog_errno(LOG_WARNING, 565204076Spjd "Unable to send handshake header to %s", 566204076Spjd res->hr_remoteaddr); 567204076Spjd nv_free(nvout); 568204076Spjd goto close; 569204076Spjd } 570204076Spjd nv_free(nvout); 571205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 572204076Spjd pjdlog_errno(LOG_WARNING, 573204076Spjd "Unable to receive handshake header from %s", 574204076Spjd res->hr_remoteaddr); 575204076Spjd goto close; 576204076Spjd } 577204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 578204076Spjd if (errmsg != NULL) { 579204076Spjd pjdlog_warning("%s", errmsg); 580204076Spjd nv_free(nvin); 581204076Spjd goto close; 582204076Spjd } 583204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 584204076Spjd if (token == NULL) { 585204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 586204076Spjd res->hr_remoteaddr); 587204076Spjd nv_free(nvin); 588204076Spjd goto close; 589204076Spjd } 590204076Spjd if (size != sizeof(res->hr_token)) { 591204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 592204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 593204076Spjd nv_free(nvin); 594204076Spjd goto close; 595204076Spjd } 596204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 597204076Spjd nv_free(nvin); 598204076Spjd 599204076Spjd /* 600204076Spjd * Second handshake step. 601204076Spjd * Setup incoming connection with remote node. 602204076Spjd */ 603218218Spjd if (primary_connect(res, &in) == -1) 604204076Spjd goto close; 605218218Spjd 606204076Spjd nvout = nv_alloc(); 607204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 608204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 609204076Spjd "token"); 610214284Spjd if (res->hr_resuid == 0) { 611214284Spjd /* 612214284Spjd * The resuid field was not yet initialized. 613214284Spjd * Because we do synchronization inside init_resuid(), it is 614214284Spjd * possible that someone already initialized it, the function 615214284Spjd * will return false then, but if we successfully initialized 616214284Spjd * it, we will get true. True means that there were no writes 617214284Spjd * to this resource yet and we want to inform secondary that 618214284Spjd * synchronization is not needed by sending "virgin" argument. 619214284Spjd */ 620214284Spjd if (init_resuid(res)) 621214284Spjd nv_add_int8(nvout, 1, "virgin"); 622214284Spjd } 623204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 624204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 625204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 626204076Spjd if (nv_error(nvout) != 0) { 627204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 628204076Spjd "Unable to allocate header for connection with %s", 629204076Spjd res->hr_remoteaddr); 630204076Spjd nv_free(nvout); 631204076Spjd goto close; 632204076Spjd } 633205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 634204076Spjd pjdlog_errno(LOG_WARNING, 635204076Spjd "Unable to send handshake header to %s", 636204076Spjd res->hr_remoteaddr); 637204076Spjd nv_free(nvout); 638204076Spjd goto close; 639204076Spjd } 640204076Spjd nv_free(nvout); 641205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 642204076Spjd pjdlog_errno(LOG_WARNING, 643204076Spjd "Unable to receive handshake header from %s", 644204076Spjd res->hr_remoteaddr); 645204076Spjd goto close; 646204076Spjd } 647204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 648204076Spjd if (errmsg != NULL) { 649204076Spjd pjdlog_warning("%s", errmsg); 650204076Spjd nv_free(nvin); 651204076Spjd goto close; 652204076Spjd } 653204076Spjd datasize = nv_get_int64(nvin, "datasize"); 654204076Spjd if (datasize != res->hr_datasize) { 655204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 656204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 657204076Spjd nv_free(nvin); 658204076Spjd goto close; 659204076Spjd } 660204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 661204076Spjd if (extentsize != res->hr_extentsize) { 662204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 663204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 664204076Spjd nv_free(nvin); 665204076Spjd goto close; 666204076Spjd } 667204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 668204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 669204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 670204076Spjd map = NULL; 671204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 672204076Spjd if (mapsize > 0) { 673204076Spjd map = malloc(mapsize); 674204076Spjd if (map == NULL) { 675204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 676204076Spjd (uintmax_t)mapsize); 677204076Spjd nv_free(nvin); 678204076Spjd goto close; 679204076Spjd } 680204076Spjd /* 681204076Spjd * Remote node have some dirty extents on its own, lets 682204076Spjd * download its activemap. 683204076Spjd */ 684205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 685204076Spjd mapsize) < 0) { 686204076Spjd pjdlog_errno(LOG_ERR, 687204076Spjd "Unable to receive remote activemap"); 688204076Spjd nv_free(nvin); 689204076Spjd free(map); 690204076Spjd goto close; 691204076Spjd } 692204076Spjd /* 693204076Spjd * Merge local and remote bitmaps. 694204076Spjd */ 695204076Spjd activemap_merge(res->hr_amp, map, mapsize); 696204076Spjd free(map); 697204076Spjd /* 698204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 699204076Spjd * disk before we start to synchronize. 700204076Spjd */ 701204076Spjd (void)hast_activemap_flush(res); 702204076Spjd } 703214274Spjd nv_free(nvin); 704204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 705205738Spjd if (inp != NULL && outp != NULL) { 706205738Spjd *inp = in; 707205738Spjd *outp = out; 708205738Spjd } else { 709205738Spjd res->hr_remotein = in; 710205738Spjd res->hr_remoteout = out; 711205738Spjd } 712212038Spjd event_send(res, EVENT_CONNECT); 713205738Spjd return (true); 714205738Spjdclose: 715211983Spjd if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0) 716212038Spjd event_send(res, EVENT_SPLITBRAIN); 717205738Spjd proto_close(out); 718205738Spjd if (in != NULL) 719205738Spjd proto_close(in); 720205738Spjd return (false); 721205738Spjd} 722205738Spjd 723205738Spjdstatic void 724205738Spjdsync_start(void) 725205738Spjd{ 726205738Spjd 727204076Spjd mtx_lock(&sync_lock); 728204076Spjd sync_inprogress = true; 729204076Spjd mtx_unlock(&sync_lock); 730204076Spjd cv_signal(&sync_cond); 731204076Spjd} 732204076Spjd 733204076Spjdstatic void 734211878Spjdsync_stop(void) 735211878Spjd{ 736211878Spjd 737211878Spjd mtx_lock(&sync_lock); 738211878Spjd if (sync_inprogress) 739211878Spjd sync_inprogress = false; 740211878Spjd mtx_unlock(&sync_lock); 741211878Spjd} 742211878Spjd 743211878Spjdstatic void 744204076Spjdinit_ggate(struct hast_resource *res) 745204076Spjd{ 746204076Spjd struct g_gate_ctl_create ggiocreate; 747204076Spjd struct g_gate_ctl_cancel ggiocancel; 748204076Spjd 749204076Spjd /* 750204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 751204076Spjd */ 752204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 753204076Spjd if (res->hr_ggatefd < 0) 754204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 755204076Spjd /* 756204076Spjd * Create provider before trying to connect, as connection failure 757204076Spjd * is not critical, but may take some time. 758204076Spjd */ 759213533Spjd bzero(&ggiocreate, sizeof(ggiocreate)); 760204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 761204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 762204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 763204076Spjd ggiocreate.gctl_flags = 0; 764206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 765204076Spjd ggiocreate.gctl_timeout = 0; 766204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 767204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 768204076Spjd res->hr_provname); 769204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 770204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 771204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 772204076Spjd return; 773204076Spjd } 774204076Spjd if (errno != EEXIST) { 775204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 776204076Spjd res->hr_provname); 777204076Spjd } 778204076Spjd pjdlog_debug(1, 779204076Spjd "Device hast/%s already exists, we will try to take it over.", 780204076Spjd res->hr_provname); 781204076Spjd /* 782204076Spjd * If we received EEXIST, we assume that the process who created the 783204076Spjd * provider died and didn't clean up. In that case we will start from 784204076Spjd * where he left of. 785204076Spjd */ 786213533Spjd bzero(&ggiocancel, sizeof(ggiocancel)); 787204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 788204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 789204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 790204076Spjd res->hr_provname); 791204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 792204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 793204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 794204076Spjd return; 795204076Spjd } 796204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 797204076Spjd res->hr_provname); 798204076Spjd} 799204076Spjd 800204076Spjdvoid 801204076Spjdhastd_primary(struct hast_resource *res) 802204076Spjd{ 803204076Spjd pthread_t td; 804204076Spjd pid_t pid; 805219482Strociny int error, mode, debuglevel; 806204076Spjd 807204076Spjd /* 808218218Spjd * Create communication channel for sending control commands from 809218218Spjd * parent to child. 810204076Spjd */ 811219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_ctrl) < 0) { 812218042Spjd /* TODO: There's no need for this to be fatal error. */ 813204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 814212034Spjd pjdlog_exit(EX_OSERR, 815204076Spjd "Unable to create control sockets between parent and child"); 816204076Spjd } 817212038Spjd /* 818218218Spjd * Create communication channel for sending events from child to parent. 819212038Spjd */ 820219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_event) < 0) { 821218042Spjd /* TODO: There's no need for this to be fatal error. */ 822212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 823212038Spjd pjdlog_exit(EX_OSERR, 824212038Spjd "Unable to create event sockets between child and parent"); 825212038Spjd } 826218218Spjd /* 827218218Spjd * Create communication channel for sending connection requests from 828218218Spjd * child to parent. 829218218Spjd */ 830219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_conn) < 0) { 831218218Spjd /* TODO: There's no need for this to be fatal error. */ 832218218Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 833218218Spjd pjdlog_exit(EX_OSERR, 834218218Spjd "Unable to create connection sockets between child and parent"); 835218218Spjd } 836204076Spjd 837204076Spjd pid = fork(); 838204076Spjd if (pid < 0) { 839218042Spjd /* TODO: There's no need for this to be fatal error. */ 840204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 841212034Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to fork"); 842204076Spjd } 843204076Spjd 844204076Spjd if (pid > 0) { 845204076Spjd /* This is parent. */ 846212038Spjd /* Declare that we are receiver. */ 847212038Spjd proto_recv(res->hr_event, NULL, 0); 848218218Spjd proto_recv(res->hr_conn, NULL, 0); 849218043Spjd /* Declare that we are sender. */ 850218043Spjd proto_send(res->hr_ctrl, NULL, 0); 851204076Spjd res->hr_workerpid = pid; 852204076Spjd return; 853204076Spjd } 854211977Spjd 855211984Spjd gres = res; 856218043Spjd mode = pjdlog_mode_get(); 857219482Strociny debuglevel = pjdlog_debug_get(); 858211984Spjd 859218043Spjd /* Declare that we are sender. */ 860218043Spjd proto_send(res->hr_event, NULL, 0); 861218218Spjd proto_send(res->hr_conn, NULL, 0); 862218043Spjd /* Declare that we are receiver. */ 863218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 864218043Spjd descriptors_cleanup(res); 865204076Spjd 866218045Spjd descriptors_assert(res, mode); 867218045Spjd 868218043Spjd pjdlog_init(mode); 869219482Strociny pjdlog_debug_set(debuglevel); 870218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 871204076Spjd setproctitle("%s (primary)", res->hr_name); 872204076Spjd 873204076Spjd init_local(res); 874213007Spjd init_ggate(res); 875213007Spjd init_environment(res); 876217784Spjd 877219847Spjd if (drop_privs(true) != 0) { 878218049Spjd cleanup(res); 879218049Spjd exit(EX_CONFIG); 880218049Spjd } 881218214Spjd pjdlog_info("Privileges successfully dropped."); 882218049Spjd 883213007Spjd /* 884213530Spjd * Create the guard thread first, so we can handle signals from the 885213530Spjd * very begining. 886213530Spjd */ 887213530Spjd error = pthread_create(&td, NULL, guard_thread, res); 888218138Spjd PJDLOG_ASSERT(error == 0); 889213530Spjd /* 890213007Spjd * Create the control thread before sending any event to the parent, 891213007Spjd * as we can deadlock when parent sends control request to worker, 892213007Spjd * but worker has no control thread started yet, so parent waits. 893213007Spjd * In the meantime worker sends an event to the parent, but parent 894213007Spjd * is unable to handle the event, because it waits for control 895213007Spjd * request response. 896213007Spjd */ 897213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 898218138Spjd PJDLOG_ASSERT(error == 0); 899210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 900205738Spjd sync_start(); 901204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 902218138Spjd PJDLOG_ASSERT(error == 0); 903204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 904218138Spjd PJDLOG_ASSERT(error == 0); 905204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 906218138Spjd PJDLOG_ASSERT(error == 0); 907204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 908218138Spjd PJDLOG_ASSERT(error == 0); 909204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 910218138Spjd PJDLOG_ASSERT(error == 0); 911213530Spjd (void)sync_thread(res); 912204076Spjd} 913204076Spjd 914204076Spjdstatic void 915204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 916204076Spjd{ 917204076Spjd char msg[1024]; 918204076Spjd va_list ap; 919204076Spjd int len; 920204076Spjd 921204076Spjd va_start(ap, fmt); 922204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 923204076Spjd va_end(ap); 924204076Spjd if ((size_t)len < sizeof(msg)) { 925204076Spjd switch (ggio->gctl_cmd) { 926204076Spjd case BIO_READ: 927204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 928204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 929204076Spjd (uintmax_t)ggio->gctl_length); 930204076Spjd break; 931204076Spjd case BIO_DELETE: 932204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 933204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 934204076Spjd (uintmax_t)ggio->gctl_length); 935204076Spjd break; 936204076Spjd case BIO_FLUSH: 937204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 938204076Spjd break; 939204076Spjd case BIO_WRITE: 940204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 941204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 942204076Spjd (uintmax_t)ggio->gctl_length); 943204076Spjd break; 944204076Spjd default: 945204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 946204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 947204076Spjd break; 948204076Spjd } 949204076Spjd } 950204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 951204076Spjd} 952204076Spjd 953204076Spjdstatic void 954204076Spjdremote_close(struct hast_resource *res, int ncomp) 955204076Spjd{ 956204076Spjd 957204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 958204076Spjd /* 959204076Spjd * A race is possible between dropping rlock and acquiring wlock - 960204076Spjd * another thread can close connection in-between. 961204076Spjd */ 962204076Spjd if (!ISCONNECTED(res, ncomp)) { 963218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 964218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 965204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 966204076Spjd return; 967204076Spjd } 968204076Spjd 969218138Spjd PJDLOG_ASSERT(res->hr_remotein != NULL); 970218138Spjd PJDLOG_ASSERT(res->hr_remoteout != NULL); 971204076Spjd 972211881Spjd pjdlog_debug(2, "Closing incoming connection to %s.", 973204076Spjd res->hr_remoteaddr); 974204076Spjd proto_close(res->hr_remotein); 975204076Spjd res->hr_remotein = NULL; 976211881Spjd pjdlog_debug(2, "Closing outgoing connection to %s.", 977204076Spjd res->hr_remoteaddr); 978204076Spjd proto_close(res->hr_remoteout); 979204076Spjd res->hr_remoteout = NULL; 980204076Spjd 981204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 982204076Spjd 983211881Spjd pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 984211881Spjd 985204076Spjd /* 986204076Spjd * Stop synchronization if in-progress. 987204076Spjd */ 988211878Spjd sync_stop(); 989211984Spjd 990212038Spjd event_send(res, EVENT_DISCONNECT); 991204076Spjd} 992204076Spjd 993204076Spjd/* 994204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 995204076Spjd * appropriate threads: 996204076Spjd * WRITE - always goes to both local_send and remote_send threads 997204076Spjd * READ (when the block is up-to-date on local component) - 998204076Spjd * only local_send thread 999204076Spjd * READ (when the block isn't up-to-date on local component) - 1000204076Spjd * only remote_send thread 1001204076Spjd * DELETE - always goes to both local_send and remote_send threads 1002204076Spjd * FLUSH - always goes to both local_send and remote_send threads 1003204076Spjd */ 1004204076Spjdstatic void * 1005204076Spjdggate_recv_thread(void *arg) 1006204076Spjd{ 1007204076Spjd struct hast_resource *res = arg; 1008204076Spjd struct g_gate_ctl_io *ggio; 1009204076Spjd struct hio *hio; 1010204076Spjd unsigned int ii, ncomp, ncomps; 1011204076Spjd int error; 1012204076Spjd 1013204076Spjd ncomps = HAST_NCOMPONENTS; 1014204076Spjd 1015204076Spjd for (;;) { 1016204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 1017204076Spjd QUEUE_TAKE2(hio, free); 1018204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 1019204076Spjd ggio = &hio->hio_ggio; 1020204076Spjd ggio->gctl_unit = res->hr_ggateunit; 1021204076Spjd ggio->gctl_length = MAXPHYS; 1022204076Spjd ggio->gctl_error = 0; 1023204076Spjd pjdlog_debug(2, 1024204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 1025204076Spjd hio); 1026204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 1027204076Spjd if (sigexit_received) 1028204076Spjd pthread_exit(NULL); 1029204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 1030204076Spjd } 1031204076Spjd error = ggio->gctl_error; 1032204076Spjd switch (error) { 1033204076Spjd case 0: 1034204076Spjd break; 1035204076Spjd case ECANCELED: 1036204076Spjd /* Exit gracefully. */ 1037204076Spjd if (!sigexit_received) { 1038204076Spjd pjdlog_debug(2, 1039204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 1040204076Spjd hio); 1041204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 1042204076Spjd } 1043204076Spjd pthread_exit(NULL); 1044204076Spjd case ENOMEM: 1045204076Spjd /* 1046204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 1047204076Spjd * bytes - request can't be bigger than that. 1048204076Spjd */ 1049204076Spjd /* FALLTHROUGH */ 1050204076Spjd case ENXIO: 1051204076Spjd default: 1052204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 1053204076Spjd strerror(error)); 1054204076Spjd } 1055204076Spjd for (ii = 0; ii < ncomps; ii++) 1056204076Spjd hio->hio_errors[ii] = EINVAL; 1057204076Spjd reqlog(LOG_DEBUG, 2, ggio, 1058204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 1059204076Spjd hio); 1060204076Spjd /* 1061204076Spjd * Inform all components about new write request. 1062204076Spjd * For read request prefer local component unless the given 1063204076Spjd * range is out-of-date, then use remote component. 1064204076Spjd */ 1065204076Spjd switch (ggio->gctl_cmd) { 1066204076Spjd case BIO_READ: 1067204076Spjd pjdlog_debug(2, 1068204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 1069204076Spjd hio); 1070204076Spjd refcount_init(&hio->hio_countdown, 1); 1071204076Spjd mtx_lock(&metadata_lock); 1072204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 1073204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1074204076Spjd /* 1075204076Spjd * This range is up-to-date on local component, 1076204076Spjd * so handle request locally. 1077204076Spjd */ 1078204076Spjd /* Local component is 0 for now. */ 1079204076Spjd ncomp = 0; 1080204076Spjd } else /* if (res->hr_syncsrc == 1081204076Spjd HAST_SYNCSRC_SECONDARY) */ { 1082218138Spjd PJDLOG_ASSERT(res->hr_syncsrc == 1083204076Spjd HAST_SYNCSRC_SECONDARY); 1084204076Spjd /* 1085204076Spjd * This range is out-of-date on local component, 1086204076Spjd * so send request to the remote node. 1087204076Spjd */ 1088204076Spjd /* Remote component is 1 for now. */ 1089204076Spjd ncomp = 1; 1090204076Spjd } 1091204076Spjd mtx_unlock(&metadata_lock); 1092204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1093204076Spjd break; 1094204076Spjd case BIO_WRITE: 1095214284Spjd if (res->hr_resuid == 0) { 1096219844Spjd /* 1097219844Spjd * This is first write, initialize localcnt and 1098219844Spjd * resuid. 1099219844Spjd */ 1100219844Spjd res->hr_primary_localcnt = 1; 1101214284Spjd (void)init_resuid(res); 1102214284Spjd } 1103204076Spjd for (;;) { 1104204076Spjd mtx_lock(&range_lock); 1105204076Spjd if (rangelock_islocked(range_sync, 1106204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1107204076Spjd pjdlog_debug(2, 1108204076Spjd "regular: Range offset=%jd length=%zu locked.", 1109204076Spjd (intmax_t)ggio->gctl_offset, 1110204076Spjd (size_t)ggio->gctl_length); 1111204076Spjd range_regular_wait = true; 1112204076Spjd cv_wait(&range_regular_cond, &range_lock); 1113204076Spjd range_regular_wait = false; 1114204076Spjd mtx_unlock(&range_lock); 1115204076Spjd continue; 1116204076Spjd } 1117204076Spjd if (rangelock_add(range_regular, 1118204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1119204076Spjd mtx_unlock(&range_lock); 1120204076Spjd pjdlog_debug(2, 1121204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1122204076Spjd (intmax_t)ggio->gctl_offset, 1123204076Spjd (size_t)ggio->gctl_length); 1124204076Spjd sleep(1); 1125204076Spjd continue; 1126204076Spjd } 1127204076Spjd mtx_unlock(&range_lock); 1128204076Spjd break; 1129204076Spjd } 1130204076Spjd mtx_lock(&res->hr_amp_lock); 1131204076Spjd if (activemap_write_start(res->hr_amp, 1132204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1133204076Spjd (void)hast_activemap_flush(res); 1134204076Spjd } 1135204076Spjd mtx_unlock(&res->hr_amp_lock); 1136204076Spjd /* FALLTHROUGH */ 1137204076Spjd case BIO_DELETE: 1138204076Spjd case BIO_FLUSH: 1139204076Spjd pjdlog_debug(2, 1140204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1141204076Spjd hio); 1142204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1143204076Spjd for (ii = 0; ii < ncomps; ii++) 1144204076Spjd QUEUE_INSERT1(hio, send, ii); 1145204076Spjd break; 1146204076Spjd } 1147204076Spjd } 1148204076Spjd /* NOTREACHED */ 1149204076Spjd return (NULL); 1150204076Spjd} 1151204076Spjd 1152204076Spjd/* 1153204076Spjd * Thread reads from or writes to local component. 1154204076Spjd * If local read fails, it redirects it to remote_send thread. 1155204076Spjd */ 1156204076Spjdstatic void * 1157204076Spjdlocal_send_thread(void *arg) 1158204076Spjd{ 1159204076Spjd struct hast_resource *res = arg; 1160204076Spjd struct g_gate_ctl_io *ggio; 1161204076Spjd struct hio *hio; 1162204076Spjd unsigned int ncomp, rncomp; 1163204076Spjd ssize_t ret; 1164204076Spjd 1165204076Spjd /* Local component is 0 for now. */ 1166204076Spjd ncomp = 0; 1167204076Spjd /* Remote component is 1 for now. */ 1168204076Spjd rncomp = 1; 1169204076Spjd 1170204076Spjd for (;;) { 1171204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1172214692Spjd QUEUE_TAKE1(hio, send, ncomp, 0); 1173204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1174204076Spjd ggio = &hio->hio_ggio; 1175204076Spjd switch (ggio->gctl_cmd) { 1176204076Spjd case BIO_READ: 1177204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1178204076Spjd ggio->gctl_length, 1179204076Spjd ggio->gctl_offset + res->hr_localoff); 1180204076Spjd if (ret == ggio->gctl_length) 1181204076Spjd hio->hio_errors[ncomp] = 0; 1182204076Spjd else { 1183204076Spjd /* 1184204076Spjd * If READ failed, try to read from remote node. 1185204076Spjd */ 1186216479Spjd if (ret < 0) { 1187216479Spjd reqlog(LOG_WARNING, 0, ggio, 1188216479Spjd "Local request failed (%s), trying remote node. ", 1189216479Spjd strerror(errno)); 1190216479Spjd } else if (ret != ggio->gctl_length) { 1191216479Spjd reqlog(LOG_WARNING, 0, ggio, 1192216479Spjd "Local request failed (%zd != %jd), trying remote node. ", 1193216494Spjd ret, (intmax_t)ggio->gctl_length); 1194216479Spjd } 1195204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1196204076Spjd continue; 1197204076Spjd } 1198204076Spjd break; 1199204076Spjd case BIO_WRITE: 1200204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1201204076Spjd ggio->gctl_length, 1202204076Spjd ggio->gctl_offset + res->hr_localoff); 1203216479Spjd if (ret < 0) { 1204204076Spjd hio->hio_errors[ncomp] = errno; 1205216479Spjd reqlog(LOG_WARNING, 0, ggio, 1206216479Spjd "Local request failed (%s): ", 1207216479Spjd strerror(errno)); 1208216479Spjd } else if (ret != ggio->gctl_length) { 1209204076Spjd hio->hio_errors[ncomp] = EIO; 1210216479Spjd reqlog(LOG_WARNING, 0, ggio, 1211216479Spjd "Local request failed (%zd != %jd): ", 1212216494Spjd ret, (intmax_t)ggio->gctl_length); 1213216479Spjd } else { 1214204076Spjd hio->hio_errors[ncomp] = 0; 1215216479Spjd } 1216204076Spjd break; 1217204076Spjd case BIO_DELETE: 1218204076Spjd ret = g_delete(res->hr_localfd, 1219204076Spjd ggio->gctl_offset + res->hr_localoff, 1220204076Spjd ggio->gctl_length); 1221216479Spjd if (ret < 0) { 1222204076Spjd hio->hio_errors[ncomp] = errno; 1223216479Spjd reqlog(LOG_WARNING, 0, ggio, 1224216479Spjd "Local request failed (%s): ", 1225216479Spjd strerror(errno)); 1226216479Spjd } else { 1227204076Spjd hio->hio_errors[ncomp] = 0; 1228216479Spjd } 1229204076Spjd break; 1230204076Spjd case BIO_FLUSH: 1231204076Spjd ret = g_flush(res->hr_localfd); 1232216479Spjd if (ret < 0) { 1233204076Spjd hio->hio_errors[ncomp] = errno; 1234216479Spjd reqlog(LOG_WARNING, 0, ggio, 1235216479Spjd "Local request failed (%s): ", 1236216479Spjd strerror(errno)); 1237216479Spjd } else { 1238204076Spjd hio->hio_errors[ncomp] = 0; 1239216479Spjd } 1240204076Spjd break; 1241204076Spjd } 1242204076Spjd if (refcount_release(&hio->hio_countdown)) { 1243204076Spjd if (ISSYNCREQ(hio)) { 1244204076Spjd mtx_lock(&sync_lock); 1245204076Spjd SYNCREQDONE(hio); 1246204076Spjd mtx_unlock(&sync_lock); 1247204076Spjd cv_signal(&sync_cond); 1248204076Spjd } else { 1249204076Spjd pjdlog_debug(2, 1250204076Spjd "local_send: (%p) Moving request to the done queue.", 1251204076Spjd hio); 1252204076Spjd QUEUE_INSERT2(hio, done); 1253204076Spjd } 1254204076Spjd } 1255204076Spjd } 1256204076Spjd /* NOTREACHED */ 1257204076Spjd return (NULL); 1258204076Spjd} 1259204076Spjd 1260214692Spjdstatic void 1261214692Spjdkeepalive_send(struct hast_resource *res, unsigned int ncomp) 1262214692Spjd{ 1263214692Spjd struct nv *nv; 1264214692Spjd 1265218217Spjd rw_rlock(&hio_remote_lock[ncomp]); 1266218217Spjd 1267218217Spjd if (!ISCONNECTED(res, ncomp)) { 1268218217Spjd rw_unlock(&hio_remote_lock[ncomp]); 1269214692Spjd return; 1270218217Spjd } 1271219864Spjd 1272218138Spjd PJDLOG_ASSERT(res->hr_remotein != NULL); 1273218138Spjd PJDLOG_ASSERT(res->hr_remoteout != NULL); 1274214692Spjd 1275214692Spjd nv = nv_alloc(); 1276214692Spjd nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1277214692Spjd if (nv_error(nv) != 0) { 1278218217Spjd rw_unlock(&hio_remote_lock[ncomp]); 1279214692Spjd nv_free(nv); 1280214692Spjd pjdlog_debug(1, 1281214692Spjd "keepalive_send: Unable to prepare header to send."); 1282214692Spjd return; 1283214692Spjd } 1284214692Spjd if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1285218217Spjd rw_unlock(&hio_remote_lock[ncomp]); 1286214692Spjd pjdlog_common(LOG_DEBUG, 1, errno, 1287214692Spjd "keepalive_send: Unable to send request"); 1288214692Spjd nv_free(nv); 1289214692Spjd remote_close(res, ncomp); 1290214692Spjd return; 1291214692Spjd } 1292218217Spjd 1293218217Spjd rw_unlock(&hio_remote_lock[ncomp]); 1294214692Spjd nv_free(nv); 1295214692Spjd pjdlog_debug(2, "keepalive_send: Request sent."); 1296214692Spjd} 1297214692Spjd 1298204076Spjd/* 1299204076Spjd * Thread sends request to secondary node. 1300204076Spjd */ 1301204076Spjdstatic void * 1302204076Spjdremote_send_thread(void *arg) 1303204076Spjd{ 1304204076Spjd struct hast_resource *res = arg; 1305204076Spjd struct g_gate_ctl_io *ggio; 1306214692Spjd time_t lastcheck, now; 1307204076Spjd struct hio *hio; 1308204076Spjd struct nv *nv; 1309204076Spjd unsigned int ncomp; 1310204076Spjd bool wakeup; 1311204076Spjd uint64_t offset, length; 1312204076Spjd uint8_t cmd; 1313204076Spjd void *data; 1314204076Spjd 1315204076Spjd /* Remote component is 1 for now. */ 1316204076Spjd ncomp = 1; 1317219864Spjd lastcheck = time(NULL); 1318204076Spjd 1319204076Spjd for (;;) { 1320204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1321219721Strociny QUEUE_TAKE1(hio, send, ncomp, HAST_KEEPALIVE); 1322214692Spjd if (hio == NULL) { 1323214692Spjd now = time(NULL); 1324219721Strociny if (lastcheck + HAST_KEEPALIVE <= now) { 1325214692Spjd keepalive_send(res, ncomp); 1326214692Spjd lastcheck = now; 1327214692Spjd } 1328214692Spjd continue; 1329214692Spjd } 1330204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1331204076Spjd ggio = &hio->hio_ggio; 1332204076Spjd switch (ggio->gctl_cmd) { 1333204076Spjd case BIO_READ: 1334204076Spjd cmd = HIO_READ; 1335204076Spjd data = NULL; 1336204076Spjd offset = ggio->gctl_offset; 1337204076Spjd length = ggio->gctl_length; 1338204076Spjd break; 1339204076Spjd case BIO_WRITE: 1340204076Spjd cmd = HIO_WRITE; 1341204076Spjd data = ggio->gctl_data; 1342204076Spjd offset = ggio->gctl_offset; 1343204076Spjd length = ggio->gctl_length; 1344204076Spjd break; 1345204076Spjd case BIO_DELETE: 1346204076Spjd cmd = HIO_DELETE; 1347204076Spjd data = NULL; 1348204076Spjd offset = ggio->gctl_offset; 1349204076Spjd length = ggio->gctl_length; 1350204076Spjd break; 1351204076Spjd case BIO_FLUSH: 1352204076Spjd cmd = HIO_FLUSH; 1353204076Spjd data = NULL; 1354204076Spjd offset = 0; 1355204076Spjd length = 0; 1356204076Spjd break; 1357204076Spjd default: 1358218138Spjd PJDLOG_ASSERT(!"invalid condition"); 1359204076Spjd abort(); 1360204076Spjd } 1361204076Spjd nv = nv_alloc(); 1362204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1363204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1364204076Spjd nv_add_uint64(nv, offset, "offset"); 1365204076Spjd nv_add_uint64(nv, length, "length"); 1366204076Spjd if (nv_error(nv) != 0) { 1367204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1368204076Spjd pjdlog_debug(2, 1369204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1370204076Spjd hio); 1371204076Spjd reqlog(LOG_ERR, 0, ggio, 1372204076Spjd "Unable to prepare header to send (%s): ", 1373204076Spjd strerror(nv_error(nv))); 1374204076Spjd /* Move failed request immediately to the done queue. */ 1375204076Spjd goto done_queue; 1376204076Spjd } 1377204076Spjd pjdlog_debug(2, 1378204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1379204076Spjd hio); 1380204076Spjd /* 1381204076Spjd * Protect connection from disappearing. 1382204076Spjd */ 1383204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1384204076Spjd if (!ISCONNECTED(res, ncomp)) { 1385204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1386204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1387204076Spjd goto done_queue; 1388204076Spjd } 1389204076Spjd /* 1390204076Spjd * Move the request to recv queue before sending it, because 1391204076Spjd * in different order we can get reply before we move request 1392204076Spjd * to recv queue. 1393204076Spjd */ 1394204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1395204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1396204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1397204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1398204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1399204076Spjd data != NULL ? length : 0) < 0) { 1400204076Spjd hio->hio_errors[ncomp] = errno; 1401204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1402204076Spjd pjdlog_debug(2, 1403204076Spjd "remote_send: (%p) Unable to send request.", hio); 1404204076Spjd reqlog(LOG_ERR, 0, ggio, 1405204076Spjd "Unable to send request (%s): ", 1406204076Spjd strerror(hio->hio_errors[ncomp])); 1407211979Spjd remote_close(res, ncomp); 1408204076Spjd /* 1409204076Spjd * Take request back from the receive queue and move 1410204076Spjd * it immediately to the done queue. 1411204076Spjd */ 1412204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1413204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1414204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1415204076Spjd goto done_queue; 1416204076Spjd } 1417204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1418204076Spjd nv_free(nv); 1419204076Spjd if (wakeup) 1420204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1421204076Spjd continue; 1422204076Spjddone_queue: 1423204076Spjd nv_free(nv); 1424204076Spjd if (ISSYNCREQ(hio)) { 1425204076Spjd if (!refcount_release(&hio->hio_countdown)) 1426204076Spjd continue; 1427204076Spjd mtx_lock(&sync_lock); 1428204076Spjd SYNCREQDONE(hio); 1429204076Spjd mtx_unlock(&sync_lock); 1430204076Spjd cv_signal(&sync_cond); 1431204076Spjd continue; 1432204076Spjd } 1433204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1434204076Spjd mtx_lock(&res->hr_amp_lock); 1435204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1436204076Spjd ggio->gctl_length)) { 1437204076Spjd (void)hast_activemap_flush(res); 1438204076Spjd } 1439204076Spjd mtx_unlock(&res->hr_amp_lock); 1440204076Spjd } 1441204076Spjd if (!refcount_release(&hio->hio_countdown)) 1442204076Spjd continue; 1443204076Spjd pjdlog_debug(2, 1444204076Spjd "remote_send: (%p) Moving request to the done queue.", 1445204076Spjd hio); 1446204076Spjd QUEUE_INSERT2(hio, done); 1447204076Spjd } 1448204076Spjd /* NOTREACHED */ 1449204076Spjd return (NULL); 1450204076Spjd} 1451204076Spjd 1452204076Spjd/* 1453204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1454204076Spjd * thread. 1455204076Spjd */ 1456204076Spjdstatic void * 1457204076Spjdremote_recv_thread(void *arg) 1458204076Spjd{ 1459204076Spjd struct hast_resource *res = arg; 1460204076Spjd struct g_gate_ctl_io *ggio; 1461204076Spjd struct hio *hio; 1462204076Spjd struct nv *nv; 1463204076Spjd unsigned int ncomp; 1464204076Spjd uint64_t seq; 1465204076Spjd int error; 1466204076Spjd 1467204076Spjd /* Remote component is 1 for now. */ 1468204076Spjd ncomp = 1; 1469204076Spjd 1470204076Spjd for (;;) { 1471204076Spjd /* Wait until there is anything to receive. */ 1472204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1473204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1474204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1475204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1476204076Spjd &hio_recv_list_lock[ncomp]); 1477204076Spjd } 1478204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1479204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1480204076Spjd if (!ISCONNECTED(res, ncomp)) { 1481204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1482204076Spjd /* 1483204076Spjd * Connection is dead, so move all pending requests to 1484204076Spjd * the done queue (one-by-one). 1485204076Spjd */ 1486204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1487204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1488218138Spjd PJDLOG_ASSERT(hio != NULL); 1489204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1490204076Spjd hio_next[ncomp]); 1491204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1492204076Spjd goto done_queue; 1493204076Spjd } 1494204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1495204076Spjd pjdlog_errno(LOG_ERR, 1496204076Spjd "Unable to receive reply header"); 1497204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1498204076Spjd remote_close(res, ncomp); 1499204076Spjd continue; 1500204076Spjd } 1501204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1502204076Spjd seq = nv_get_uint64(nv, "seq"); 1503204076Spjd if (seq == 0) { 1504204076Spjd pjdlog_error("Header contains no 'seq' field."); 1505204076Spjd nv_free(nv); 1506204076Spjd continue; 1507204076Spjd } 1508204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1509204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1510204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1511204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1512204076Spjd hio_next[ncomp]); 1513204076Spjd break; 1514204076Spjd } 1515204076Spjd } 1516204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1517204076Spjd if (hio == NULL) { 1518204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1519204076Spjd (uintmax_t)seq); 1520204076Spjd nv_free(nv); 1521204076Spjd continue; 1522204076Spjd } 1523204076Spjd error = nv_get_int16(nv, "error"); 1524204076Spjd if (error != 0) { 1525204076Spjd /* Request failed on remote side. */ 1526216478Spjd hio->hio_errors[ncomp] = error; 1527216479Spjd reqlog(LOG_WARNING, 0, &hio->hio_ggio, 1528216479Spjd "Remote request failed (%s): ", strerror(error)); 1529204076Spjd nv_free(nv); 1530204076Spjd goto done_queue; 1531204076Spjd } 1532204076Spjd ggio = &hio->hio_ggio; 1533204076Spjd switch (ggio->gctl_cmd) { 1534204076Spjd case BIO_READ: 1535204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1536204076Spjd if (!ISCONNECTED(res, ncomp)) { 1537204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1538204076Spjd nv_free(nv); 1539204076Spjd goto done_queue; 1540204076Spjd } 1541204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1542204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1543204076Spjd hio->hio_errors[ncomp] = errno; 1544204076Spjd pjdlog_errno(LOG_ERR, 1545204076Spjd "Unable to receive reply data"); 1546204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1547204076Spjd nv_free(nv); 1548204076Spjd remote_close(res, ncomp); 1549204076Spjd goto done_queue; 1550204076Spjd } 1551204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1552204076Spjd break; 1553204076Spjd case BIO_WRITE: 1554204076Spjd case BIO_DELETE: 1555204076Spjd case BIO_FLUSH: 1556204076Spjd break; 1557204076Spjd default: 1558218138Spjd PJDLOG_ASSERT(!"invalid condition"); 1559204076Spjd abort(); 1560204076Spjd } 1561204076Spjd hio->hio_errors[ncomp] = 0; 1562204076Spjd nv_free(nv); 1563204076Spjddone_queue: 1564204076Spjd if (refcount_release(&hio->hio_countdown)) { 1565204076Spjd if (ISSYNCREQ(hio)) { 1566204076Spjd mtx_lock(&sync_lock); 1567204076Spjd SYNCREQDONE(hio); 1568204076Spjd mtx_unlock(&sync_lock); 1569204076Spjd cv_signal(&sync_cond); 1570204076Spjd } else { 1571204076Spjd pjdlog_debug(2, 1572204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1573204076Spjd hio); 1574204076Spjd QUEUE_INSERT2(hio, done); 1575204076Spjd } 1576204076Spjd } 1577204076Spjd } 1578204076Spjd /* NOTREACHED */ 1579204076Spjd return (NULL); 1580204076Spjd} 1581204076Spjd 1582204076Spjd/* 1583204076Spjd * Thread sends answer to the kernel. 1584204076Spjd */ 1585204076Spjdstatic void * 1586204076Spjdggate_send_thread(void *arg) 1587204076Spjd{ 1588204076Spjd struct hast_resource *res = arg; 1589204076Spjd struct g_gate_ctl_io *ggio; 1590204076Spjd struct hio *hio; 1591204076Spjd unsigned int ii, ncomp, ncomps; 1592204076Spjd 1593204076Spjd ncomps = HAST_NCOMPONENTS; 1594204076Spjd 1595204076Spjd for (;;) { 1596204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1597204076Spjd QUEUE_TAKE2(hio, done); 1598204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1599204076Spjd ggio = &hio->hio_ggio; 1600204076Spjd for (ii = 0; ii < ncomps; ii++) { 1601204076Spjd if (hio->hio_errors[ii] == 0) { 1602204076Spjd /* 1603204076Spjd * One successful request is enough to declare 1604204076Spjd * success. 1605204076Spjd */ 1606204076Spjd ggio->gctl_error = 0; 1607204076Spjd break; 1608204076Spjd } 1609204076Spjd } 1610204076Spjd if (ii == ncomps) { 1611204076Spjd /* 1612204076Spjd * None of the requests were successful. 1613219879Strociny * Use the error from local component except the 1614219879Strociny * case when we did only remote request. 1615204076Spjd */ 1616219879Strociny if (ggio->gctl_cmd == BIO_READ && 1617219879Strociny res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) 1618219879Strociny ggio->gctl_error = hio->hio_errors[1]; 1619219879Strociny else 1620219879Strociny ggio->gctl_error = hio->hio_errors[0]; 1621204076Spjd } 1622204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1623204076Spjd mtx_lock(&res->hr_amp_lock); 1624204076Spjd activemap_write_complete(res->hr_amp, 1625204076Spjd ggio->gctl_offset, ggio->gctl_length); 1626204076Spjd mtx_unlock(&res->hr_amp_lock); 1627204076Spjd } 1628204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1629204076Spjd /* 1630204076Spjd * Unlock range we locked. 1631204076Spjd */ 1632204076Spjd mtx_lock(&range_lock); 1633204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1634204076Spjd ggio->gctl_length); 1635204076Spjd if (range_sync_wait) 1636204076Spjd cv_signal(&range_sync_cond); 1637204076Spjd mtx_unlock(&range_lock); 1638204076Spjd /* 1639204076Spjd * Bump local count if this is first write after 1640204076Spjd * connection failure with remote node. 1641204076Spjd */ 1642204076Spjd ncomp = 1; 1643204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1644204076Spjd if (!ISCONNECTED(res, ncomp)) { 1645204076Spjd mtx_lock(&metadata_lock); 1646204076Spjd if (res->hr_primary_localcnt == 1647204076Spjd res->hr_secondary_remotecnt) { 1648204076Spjd res->hr_primary_localcnt++; 1649204076Spjd pjdlog_debug(1, 1650204076Spjd "Increasing localcnt to %ju.", 1651204076Spjd (uintmax_t)res->hr_primary_localcnt); 1652204076Spjd (void)metadata_write(res); 1653204076Spjd } 1654204076Spjd mtx_unlock(&metadata_lock); 1655204076Spjd } 1656204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1657204076Spjd } 1658204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1659204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1660204076Spjd pjdlog_debug(2, 1661204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1662204076Spjd QUEUE_INSERT2(hio, free); 1663204076Spjd } 1664204076Spjd /* NOTREACHED */ 1665204076Spjd return (NULL); 1666204076Spjd} 1667204076Spjd 1668204076Spjd/* 1669204076Spjd * Thread synchronize local and remote components. 1670204076Spjd */ 1671204076Spjdstatic void * 1672204076Spjdsync_thread(void *arg __unused) 1673204076Spjd{ 1674204076Spjd struct hast_resource *res = arg; 1675204076Spjd struct hio *hio; 1676204076Spjd struct g_gate_ctl_io *ggio; 1677219372Spjd struct timeval tstart, tend, tdiff; 1678204076Spjd unsigned int ii, ncomp, ncomps; 1679204076Spjd off_t offset, length, synced; 1680204076Spjd bool dorewind; 1681204076Spjd int syncext; 1682204076Spjd 1683204076Spjd ncomps = HAST_NCOMPONENTS; 1684204076Spjd dorewind = true; 1685211897Spjd synced = 0; 1686211897Spjd offset = -1; 1687204076Spjd 1688204076Spjd for (;;) { 1689204076Spjd mtx_lock(&sync_lock); 1690211897Spjd if (offset >= 0 && !sync_inprogress) { 1691219372Spjd gettimeofday(&tend, NULL); 1692219372Spjd timersub(&tend, &tstart, &tdiff); 1693219372Spjd pjdlog_info("Synchronization interrupted after %#.0T. " 1694219372Spjd "%NB synchronized so far.", &tdiff, 1695211879Spjd (intmax_t)synced); 1696212038Spjd event_send(res, EVENT_SYNCINTR); 1697211879Spjd } 1698204076Spjd while (!sync_inprogress) { 1699204076Spjd dorewind = true; 1700204076Spjd synced = 0; 1701204076Spjd cv_wait(&sync_cond, &sync_lock); 1702204076Spjd } 1703204076Spjd mtx_unlock(&sync_lock); 1704204076Spjd /* 1705204076Spjd * Obtain offset at which we should synchronize. 1706204076Spjd * Rewind synchronization if needed. 1707204076Spjd */ 1708204076Spjd mtx_lock(&res->hr_amp_lock); 1709204076Spjd if (dorewind) 1710204076Spjd activemap_sync_rewind(res->hr_amp); 1711204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1712204076Spjd if (syncext != -1) { 1713204076Spjd /* 1714204076Spjd * We synchronized entire syncext extent, we can mark 1715204076Spjd * it as clean now. 1716204076Spjd */ 1717204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1718204076Spjd (void)hast_activemap_flush(res); 1719204076Spjd } 1720204076Spjd mtx_unlock(&res->hr_amp_lock); 1721204076Spjd if (dorewind) { 1722204076Spjd dorewind = false; 1723204076Spjd if (offset < 0) 1724204076Spjd pjdlog_info("Nodes are in sync."); 1725204076Spjd else { 1726219372Spjd pjdlog_info("Synchronization started. %NB to go.", 1727219372Spjd (intmax_t)(res->hr_extentsize * 1728204076Spjd activemap_ndirty(res->hr_amp))); 1729212038Spjd event_send(res, EVENT_SYNCSTART); 1730219372Spjd gettimeofday(&tstart, NULL); 1731204076Spjd } 1732204076Spjd } 1733204076Spjd if (offset < 0) { 1734211878Spjd sync_stop(); 1735204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1736204076Spjd /* 1737204076Spjd * Synchronization complete, make both localcnt and 1738204076Spjd * remotecnt equal. 1739204076Spjd */ 1740204076Spjd ncomp = 1; 1741204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1742204076Spjd if (ISCONNECTED(res, ncomp)) { 1743204076Spjd if (synced > 0) { 1744219372Spjd int64_t bps; 1745219372Spjd 1746219372Spjd gettimeofday(&tend, NULL); 1747219372Spjd timersub(&tend, &tstart, &tdiff); 1748219372Spjd bps = (int64_t)((double)synced / 1749219372Spjd ((double)tdiff.tv_sec + 1750219372Spjd (double)tdiff.tv_usec / 1000000)); 1751204076Spjd pjdlog_info("Synchronization complete. " 1752219372Spjd "%NB synchronized in %#.0lT (%NB/sec).", 1753219372Spjd (intmax_t)synced, &tdiff, 1754219372Spjd (intmax_t)bps); 1755212038Spjd event_send(res, EVENT_SYNCDONE); 1756204076Spjd } 1757204076Spjd mtx_lock(&metadata_lock); 1758204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1759204076Spjd res->hr_primary_localcnt = 1760204076Spjd res->hr_secondary_localcnt; 1761204076Spjd res->hr_primary_remotecnt = 1762204076Spjd res->hr_secondary_remotecnt; 1763204076Spjd pjdlog_debug(1, 1764204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1765204076Spjd (uintmax_t)res->hr_primary_localcnt, 1766204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1767204076Spjd (void)metadata_write(res); 1768204076Spjd mtx_unlock(&metadata_lock); 1769204076Spjd } 1770204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1771204076Spjd continue; 1772204076Spjd } 1773204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1774204076Spjd QUEUE_TAKE2(hio, free); 1775204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1776204076Spjd /* 1777204076Spjd * Lock the range we are going to synchronize. We don't want 1778204076Spjd * race where someone writes between our read and write. 1779204076Spjd */ 1780204076Spjd for (;;) { 1781204076Spjd mtx_lock(&range_lock); 1782204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1783204076Spjd pjdlog_debug(2, 1784204076Spjd "sync: Range offset=%jd length=%jd locked.", 1785204076Spjd (intmax_t)offset, (intmax_t)length); 1786204076Spjd range_sync_wait = true; 1787204076Spjd cv_wait(&range_sync_cond, &range_lock); 1788204076Spjd range_sync_wait = false; 1789204076Spjd mtx_unlock(&range_lock); 1790204076Spjd continue; 1791204076Spjd } 1792204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1793204076Spjd mtx_unlock(&range_lock); 1794204076Spjd pjdlog_debug(2, 1795204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1796204076Spjd (intmax_t)offset, (intmax_t)length); 1797204076Spjd sleep(1); 1798204076Spjd continue; 1799204076Spjd } 1800204076Spjd mtx_unlock(&range_lock); 1801204076Spjd break; 1802204076Spjd } 1803204076Spjd /* 1804204076Spjd * First read the data from synchronization source. 1805204076Spjd */ 1806204076Spjd SYNCREQ(hio); 1807204076Spjd ggio = &hio->hio_ggio; 1808204076Spjd ggio->gctl_cmd = BIO_READ; 1809204076Spjd ggio->gctl_offset = offset; 1810204076Spjd ggio->gctl_length = length; 1811204076Spjd ggio->gctl_error = 0; 1812204076Spjd for (ii = 0; ii < ncomps; ii++) 1813204076Spjd hio->hio_errors[ii] = EINVAL; 1814204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1815204076Spjd hio); 1816204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1817204076Spjd hio); 1818204076Spjd mtx_lock(&metadata_lock); 1819204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1820204076Spjd /* 1821204076Spjd * This range is up-to-date on local component, 1822204076Spjd * so handle request locally. 1823204076Spjd */ 1824204076Spjd /* Local component is 0 for now. */ 1825204076Spjd ncomp = 0; 1826204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1827218138Spjd PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1828204076Spjd /* 1829204076Spjd * This range is out-of-date on local component, 1830204076Spjd * so send request to the remote node. 1831204076Spjd */ 1832204076Spjd /* Remote component is 1 for now. */ 1833204076Spjd ncomp = 1; 1834204076Spjd } 1835204076Spjd mtx_unlock(&metadata_lock); 1836204076Spjd refcount_init(&hio->hio_countdown, 1); 1837204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1838204076Spjd 1839204076Spjd /* 1840204076Spjd * Let's wait for READ to finish. 1841204076Spjd */ 1842204076Spjd mtx_lock(&sync_lock); 1843204076Spjd while (!ISSYNCREQDONE(hio)) 1844204076Spjd cv_wait(&sync_cond, &sync_lock); 1845204076Spjd mtx_unlock(&sync_lock); 1846204076Spjd 1847204076Spjd if (hio->hio_errors[ncomp] != 0) { 1848204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1849204076Spjd strerror(hio->hio_errors[ncomp])); 1850204076Spjd goto free_queue; 1851204076Spjd } 1852204076Spjd 1853204076Spjd /* 1854204076Spjd * We read the data from synchronization source, now write it 1855204076Spjd * to synchronization target. 1856204076Spjd */ 1857204076Spjd SYNCREQ(hio); 1858204076Spjd ggio->gctl_cmd = BIO_WRITE; 1859204076Spjd for (ii = 0; ii < ncomps; ii++) 1860204076Spjd hio->hio_errors[ii] = EINVAL; 1861204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1862204076Spjd hio); 1863204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1864204076Spjd hio); 1865204076Spjd mtx_lock(&metadata_lock); 1866204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1867204076Spjd /* 1868204076Spjd * This range is up-to-date on local component, 1869204076Spjd * so we update remote component. 1870204076Spjd */ 1871204076Spjd /* Remote component is 1 for now. */ 1872204076Spjd ncomp = 1; 1873204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1874218138Spjd PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1875204076Spjd /* 1876204076Spjd * This range is out-of-date on local component, 1877204076Spjd * so we update it. 1878204076Spjd */ 1879204076Spjd /* Local component is 0 for now. */ 1880204076Spjd ncomp = 0; 1881204076Spjd } 1882204076Spjd mtx_unlock(&metadata_lock); 1883204076Spjd 1884204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1885204076Spjd hio); 1886204076Spjd refcount_init(&hio->hio_countdown, 1); 1887204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1888204076Spjd 1889204076Spjd /* 1890204076Spjd * Let's wait for WRITE to finish. 1891204076Spjd */ 1892204076Spjd mtx_lock(&sync_lock); 1893204076Spjd while (!ISSYNCREQDONE(hio)) 1894204076Spjd cv_wait(&sync_cond, &sync_lock); 1895204076Spjd mtx_unlock(&sync_lock); 1896204076Spjd 1897204076Spjd if (hio->hio_errors[ncomp] != 0) { 1898204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1899204076Spjd strerror(hio->hio_errors[ncomp])); 1900204076Spjd goto free_queue; 1901204076Spjd } 1902211880Spjd 1903211880Spjd synced += length; 1904204076Spjdfree_queue: 1905204076Spjd mtx_lock(&range_lock); 1906204076Spjd rangelock_del(range_sync, offset, length); 1907204076Spjd if (range_regular_wait) 1908204076Spjd cv_signal(&range_regular_cond); 1909204076Spjd mtx_unlock(&range_lock); 1910204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1911204076Spjd hio); 1912204076Spjd QUEUE_INSERT2(hio, free); 1913204076Spjd } 1914204076Spjd /* NOTREACHED */ 1915204076Spjd return (NULL); 1916204076Spjd} 1917204076Spjd 1918217784Spjdvoid 1919217784Spjdprimary_config_reload(struct hast_resource *res, struct nv *nv) 1920210886Spjd{ 1921210886Spjd unsigned int ii, ncomps; 1922217784Spjd int modified, vint; 1923217784Spjd const char *vstr; 1924210886Spjd 1925210886Spjd pjdlog_info("Reloading configuration..."); 1926210886Spjd 1927218138Spjd PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY); 1928218138Spjd PJDLOG_ASSERT(gres == res); 1929217784Spjd nv_assert(nv, "remoteaddr"); 1930219818Spjd nv_assert(nv, "sourceaddr"); 1931217784Spjd nv_assert(nv, "replication"); 1932219351Spjd nv_assert(nv, "checksum"); 1933219354Spjd nv_assert(nv, "compression"); 1934217784Spjd nv_assert(nv, "timeout"); 1935217784Spjd nv_assert(nv, "exec"); 1936217784Spjd 1937210886Spjd ncomps = HAST_NCOMPONENTS; 1938210886Spjd 1939219351Spjd#define MODIFIED_REMOTEADDR 0x01 1940219818Spjd#define MODIFIED_SOURCEADDR 0x02 1941219818Spjd#define MODIFIED_REPLICATION 0x04 1942219818Spjd#define MODIFIED_CHECKSUM 0x08 1943219818Spjd#define MODIFIED_COMPRESSION 0x10 1944219818Spjd#define MODIFIED_TIMEOUT 0x20 1945219818Spjd#define MODIFIED_EXEC 0x40 1946210886Spjd modified = 0; 1947217784Spjd 1948217784Spjd vstr = nv_get_string(nv, "remoteaddr"); 1949217784Spjd if (strcmp(gres->hr_remoteaddr, vstr) != 0) { 1950210886Spjd /* 1951210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1952210886Spjd * We want remote_close() to log disconnect from the old 1953210886Spjd * addresses, not from the new ones. 1954210886Spjd */ 1955210886Spjd modified |= MODIFIED_REMOTEADDR; 1956210886Spjd } 1957219818Spjd vstr = nv_get_string(nv, "sourceaddr"); 1958219818Spjd if (strcmp(gres->hr_sourceaddr, vstr) != 0) { 1959219818Spjd strlcpy(gres->hr_sourceaddr, vstr, sizeof(gres->hr_sourceaddr)); 1960219818Spjd modified |= MODIFIED_SOURCEADDR; 1961219818Spjd } 1962217784Spjd vint = nv_get_int32(nv, "replication"); 1963217784Spjd if (gres->hr_replication != vint) { 1964217784Spjd gres->hr_replication = vint; 1965210886Spjd modified |= MODIFIED_REPLICATION; 1966210886Spjd } 1967219351Spjd vint = nv_get_int32(nv, "checksum"); 1968219351Spjd if (gres->hr_checksum != vint) { 1969219351Spjd gres->hr_checksum = vint; 1970219351Spjd modified |= MODIFIED_CHECKSUM; 1971219351Spjd } 1972219354Spjd vint = nv_get_int32(nv, "compression"); 1973219354Spjd if (gres->hr_compression != vint) { 1974219354Spjd gres->hr_compression = vint; 1975219354Spjd modified |= MODIFIED_COMPRESSION; 1976219354Spjd } 1977217784Spjd vint = nv_get_int32(nv, "timeout"); 1978217784Spjd if (gres->hr_timeout != vint) { 1979217784Spjd gres->hr_timeout = vint; 1980210886Spjd modified |= MODIFIED_TIMEOUT; 1981210886Spjd } 1982217784Spjd vstr = nv_get_string(nv, "exec"); 1983217784Spjd if (strcmp(gres->hr_exec, vstr) != 0) { 1984217784Spjd strlcpy(gres->hr_exec, vstr, sizeof(gres->hr_exec)); 1985211886Spjd modified |= MODIFIED_EXEC; 1986211886Spjd } 1987217784Spjd 1988210886Spjd /* 1989219351Spjd * Change timeout for connected sockets. 1990219351Spjd * Don't bother if we need to reconnect. 1991210886Spjd */ 1992219351Spjd if ((modified & MODIFIED_TIMEOUT) != 0 && 1993219818Spjd (modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR | 1994219818Spjd MODIFIED_REPLICATION)) == 0) { 1995210886Spjd for (ii = 0; ii < ncomps; ii++) { 1996210886Spjd if (!ISREMOTE(ii)) 1997210886Spjd continue; 1998210886Spjd rw_rlock(&hio_remote_lock[ii]); 1999210886Spjd if (!ISCONNECTED(gres, ii)) { 2000210886Spjd rw_unlock(&hio_remote_lock[ii]); 2001210886Spjd continue; 2002210886Spjd } 2003210886Spjd rw_unlock(&hio_remote_lock[ii]); 2004210886Spjd if (proto_timeout(gres->hr_remotein, 2005210886Spjd gres->hr_timeout) < 0) { 2006210886Spjd pjdlog_errno(LOG_WARNING, 2007210886Spjd "Unable to set connection timeout"); 2008210886Spjd } 2009210886Spjd if (proto_timeout(gres->hr_remoteout, 2010210886Spjd gres->hr_timeout) < 0) { 2011210886Spjd pjdlog_errno(LOG_WARNING, 2012210886Spjd "Unable to set connection timeout"); 2013210886Spjd } 2014210886Spjd } 2015219351Spjd } 2016219818Spjd if ((modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR | 2017219818Spjd MODIFIED_REPLICATION)) != 0) { 2018210886Spjd for (ii = 0; ii < ncomps; ii++) { 2019210886Spjd if (!ISREMOTE(ii)) 2020210886Spjd continue; 2021210886Spjd remote_close(gres, ii); 2022210886Spjd } 2023210886Spjd if (modified & MODIFIED_REMOTEADDR) { 2024217784Spjd vstr = nv_get_string(nv, "remoteaddr"); 2025217784Spjd strlcpy(gres->hr_remoteaddr, vstr, 2026210886Spjd sizeof(gres->hr_remoteaddr)); 2027210886Spjd } 2028210886Spjd } 2029210886Spjd#undef MODIFIED_REMOTEADDR 2030219818Spjd#undef MODIFIED_SOURCEADDR 2031210886Spjd#undef MODIFIED_REPLICATION 2032219351Spjd#undef MODIFIED_CHECKSUM 2033219354Spjd#undef MODIFIED_COMPRESSION 2034210886Spjd#undef MODIFIED_TIMEOUT 2035211886Spjd#undef MODIFIED_EXEC 2036210886Spjd 2037210886Spjd pjdlog_info("Configuration reloaded successfully."); 2038210886Spjd} 2039210886Spjd 2040211882Spjdstatic void 2041211981Spjdguard_one(struct hast_resource *res, unsigned int ncomp) 2042211981Spjd{ 2043211981Spjd struct proto_conn *in, *out; 2044211981Spjd 2045211981Spjd if (!ISREMOTE(ncomp)) 2046211981Spjd return; 2047211981Spjd 2048211981Spjd rw_rlock(&hio_remote_lock[ncomp]); 2049211981Spjd 2050211981Spjd if (!real_remote(res)) { 2051211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 2052211981Spjd return; 2053211981Spjd } 2054211981Spjd 2055211981Spjd if (ISCONNECTED(res, ncomp)) { 2056218138Spjd PJDLOG_ASSERT(res->hr_remotein != NULL); 2057218138Spjd PJDLOG_ASSERT(res->hr_remoteout != NULL); 2058211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 2059211981Spjd pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 2060211981Spjd res->hr_remoteaddr); 2061211981Spjd return; 2062211981Spjd } 2063211981Spjd 2064218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 2065218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 2066211981Spjd /* 2067211981Spjd * Upgrade the lock. It doesn't have to be atomic as no other thread 2068211981Spjd * can change connection status from disconnected to connected. 2069211981Spjd */ 2070211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 2071211981Spjd pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 2072211981Spjd res->hr_remoteaddr); 2073211981Spjd in = out = NULL; 2074211981Spjd if (init_remote(res, &in, &out)) { 2075211981Spjd rw_wlock(&hio_remote_lock[ncomp]); 2076218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 2077218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 2078218138Spjd PJDLOG_ASSERT(in != NULL && out != NULL); 2079211981Spjd res->hr_remotein = in; 2080211981Spjd res->hr_remoteout = out; 2081211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 2082211981Spjd pjdlog_info("Successfully reconnected to %s.", 2083211981Spjd res->hr_remoteaddr); 2084211981Spjd sync_start(); 2085211981Spjd } else { 2086211981Spjd /* Both connections should be NULL. */ 2087218138Spjd PJDLOG_ASSERT(res->hr_remotein == NULL); 2088218138Spjd PJDLOG_ASSERT(res->hr_remoteout == NULL); 2089218138Spjd PJDLOG_ASSERT(in == NULL && out == NULL); 2090211981Spjd pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 2091211981Spjd res->hr_remoteaddr); 2092211981Spjd } 2093211981Spjd} 2094211981Spjd 2095204076Spjd/* 2096204076Spjd * Thread guards remote connections and reconnects when needed, handles 2097204076Spjd * signals, etc. 2098204076Spjd */ 2099204076Spjdstatic void * 2100204076Spjdguard_thread(void *arg) 2101204076Spjd{ 2102204076Spjd struct hast_resource *res = arg; 2103204076Spjd unsigned int ii, ncomps; 2104211982Spjd struct timespec timeout; 2105211981Spjd time_t lastcheck, now; 2106211982Spjd sigset_t mask; 2107211982Spjd int signo; 2108204076Spjd 2109204076Spjd ncomps = HAST_NCOMPONENTS; 2110211981Spjd lastcheck = time(NULL); 2111204076Spjd 2112211982Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 2113211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 2114211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 2115211982Spjd 2116219721Strociny timeout.tv_sec = HAST_KEEPALIVE; 2117211982Spjd timeout.tv_nsec = 0; 2118211982Spjd signo = -1; 2119211982Spjd 2120204076Spjd for (;;) { 2121211982Spjd switch (signo) { 2122211982Spjd case SIGINT: 2123211982Spjd case SIGTERM: 2124211982Spjd sigexit_received = true; 2125204076Spjd primary_exitx(EX_OK, 2126204076Spjd "Termination signal received, exiting."); 2127211982Spjd break; 2128211982Spjd default: 2129211982Spjd break; 2130204076Spjd } 2131211882Spjd 2132204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 2133211981Spjd now = time(NULL); 2134219721Strociny if (lastcheck + HAST_KEEPALIVE <= now) { 2135211982Spjd for (ii = 0; ii < ncomps; ii++) 2136211981Spjd guard_one(res, ii); 2137211981Spjd lastcheck = now; 2138204076Spjd } 2139211982Spjd signo = sigtimedwait(&mask, NULL, &timeout); 2140204076Spjd } 2141204076Spjd /* NOTREACHED */ 2142204076Spjd return (NULL); 2143204076Spjd} 2144