primary.c revision 212038
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 212038 2010-08-30 23:26:10Z 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 /* 238204076Spjd * Close descriptor to /dev/hast/<name> 239204076Spjd * to work-around race in the kernel. 240204076Spjd */ 241204076Spjd close(res->hr_localfd); 242204076Spjd 243204076Spjd /* Destroy ggate provider if we created one. */ 244204076Spjd if (res->hr_ggateunit >= 0) { 245204076Spjd struct g_gate_ctl_destroy ggiod; 246204076Spjd 247204076Spjd ggiod.gctl_version = G_GATE_VERSION; 248204076Spjd ggiod.gctl_unit = res->hr_ggateunit; 249204076Spjd ggiod.gctl_force = 1; 250204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 251204076Spjd pjdlog_warning("Unable to destroy hast/%s device", 252204076Spjd res->hr_provname); 253204076Spjd } 254204076Spjd res->hr_ggateunit = -1; 255204076Spjd } 256204076Spjd 257204076Spjd /* Restore errno. */ 258204076Spjd errno = rerrno; 259204076Spjd} 260204076Spjd 261204076Spjdstatic void 262204076Spjdprimary_exit(int exitcode, const char *fmt, ...) 263204076Spjd{ 264204076Spjd va_list ap; 265204076Spjd 266204076Spjd assert(exitcode != EX_OK); 267204076Spjd va_start(ap, fmt); 268204076Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 269204076Spjd va_end(ap); 270204076Spjd cleanup(gres); 271204076Spjd exit(exitcode); 272204076Spjd} 273204076Spjd 274204076Spjdstatic void 275204076Spjdprimary_exitx(int exitcode, const char *fmt, ...) 276204076Spjd{ 277204076Spjd va_list ap; 278204076Spjd 279204076Spjd va_start(ap, fmt); 280204076Spjd pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 281204076Spjd va_end(ap); 282204076Spjd cleanup(gres); 283204076Spjd exit(exitcode); 284204076Spjd} 285204076Spjd 286204076Spjdstatic int 287204076Spjdhast_activemap_flush(struct hast_resource *res) 288204076Spjd{ 289204076Spjd const unsigned char *buf; 290204076Spjd size_t size; 291204076Spjd 292204076Spjd buf = activemap_bitmap(res->hr_amp, &size); 293204076Spjd assert(buf != NULL); 294204076Spjd assert((size % res->hr_local_sectorsize) == 0); 295204076Spjd if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 296204076Spjd (ssize_t)size) { 297204076Spjd KEEP_ERRNO(pjdlog_errno(LOG_ERR, 298204076Spjd "Unable to flush activemap to disk")); 299204076Spjd return (-1); 300204076Spjd } 301204076Spjd return (0); 302204076Spjd} 303204076Spjd 304210881Spjdstatic bool 305210881Spjdreal_remote(const struct hast_resource *res) 306210881Spjd{ 307210881Spjd 308210881Spjd return (strcmp(res->hr_remoteaddr, "none") != 0); 309210881Spjd} 310210881Spjd 311204076Spjdstatic void 312204076Spjdinit_environment(struct hast_resource *res __unused) 313204076Spjd{ 314204076Spjd struct hio *hio; 315204076Spjd unsigned int ii, ncomps; 316211982Spjd sigset_t mask; 317204076Spjd 318204076Spjd /* 319204076Spjd * In the future it might be per-resource value. 320204076Spjd */ 321204076Spjd ncomps = HAST_NCOMPONENTS; 322204076Spjd 323204076Spjd /* 324204076Spjd * Allocate memory needed by lists. 325204076Spjd */ 326204076Spjd hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 327204076Spjd if (hio_send_list == NULL) { 328204076Spjd primary_exitx(EX_TEMPFAIL, 329204076Spjd "Unable to allocate %zu bytes of memory for send lists.", 330204076Spjd sizeof(hio_send_list[0]) * ncomps); 331204076Spjd } 332204076Spjd hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 333204076Spjd if (hio_send_list_lock == NULL) { 334204076Spjd primary_exitx(EX_TEMPFAIL, 335204076Spjd "Unable to allocate %zu bytes of memory for send list locks.", 336204076Spjd sizeof(hio_send_list_lock[0]) * ncomps); 337204076Spjd } 338204076Spjd hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 339204076Spjd if (hio_send_list_cond == NULL) { 340204076Spjd primary_exitx(EX_TEMPFAIL, 341204076Spjd "Unable to allocate %zu bytes of memory for send list condition variables.", 342204076Spjd sizeof(hio_send_list_cond[0]) * ncomps); 343204076Spjd } 344204076Spjd hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 345204076Spjd if (hio_recv_list == NULL) { 346204076Spjd primary_exitx(EX_TEMPFAIL, 347204076Spjd "Unable to allocate %zu bytes of memory for recv lists.", 348204076Spjd sizeof(hio_recv_list[0]) * ncomps); 349204076Spjd } 350204076Spjd hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 351204076Spjd if (hio_recv_list_lock == NULL) { 352204076Spjd primary_exitx(EX_TEMPFAIL, 353204076Spjd "Unable to allocate %zu bytes of memory for recv list locks.", 354204076Spjd sizeof(hio_recv_list_lock[0]) * ncomps); 355204076Spjd } 356204076Spjd hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 357204076Spjd if (hio_recv_list_cond == NULL) { 358204076Spjd primary_exitx(EX_TEMPFAIL, 359204076Spjd "Unable to allocate %zu bytes of memory for recv list condition variables.", 360204076Spjd sizeof(hio_recv_list_cond[0]) * ncomps); 361204076Spjd } 362204076Spjd hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 363204076Spjd if (hio_remote_lock == NULL) { 364204076Spjd primary_exitx(EX_TEMPFAIL, 365204076Spjd "Unable to allocate %zu bytes of memory for remote connections locks.", 366204076Spjd sizeof(hio_remote_lock[0]) * ncomps); 367204076Spjd } 368204076Spjd 369204076Spjd /* 370204076Spjd * Initialize lists, their locks and theirs condition variables. 371204076Spjd */ 372204076Spjd TAILQ_INIT(&hio_free_list); 373204076Spjd mtx_init(&hio_free_list_lock); 374204076Spjd cv_init(&hio_free_list_cond); 375204076Spjd for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 376204076Spjd TAILQ_INIT(&hio_send_list[ii]); 377204076Spjd mtx_init(&hio_send_list_lock[ii]); 378204076Spjd cv_init(&hio_send_list_cond[ii]); 379204076Spjd TAILQ_INIT(&hio_recv_list[ii]); 380204076Spjd mtx_init(&hio_recv_list_lock[ii]); 381204076Spjd cv_init(&hio_recv_list_cond[ii]); 382204076Spjd rw_init(&hio_remote_lock[ii]); 383204076Spjd } 384204076Spjd TAILQ_INIT(&hio_done_list); 385204076Spjd mtx_init(&hio_done_list_lock); 386204076Spjd cv_init(&hio_done_list_cond); 387204076Spjd mtx_init(&metadata_lock); 388204076Spjd 389204076Spjd /* 390204076Spjd * Allocate requests pool and initialize requests. 391204076Spjd */ 392204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 393204076Spjd hio = malloc(sizeof(*hio)); 394204076Spjd if (hio == NULL) { 395204076Spjd primary_exitx(EX_TEMPFAIL, 396204076Spjd "Unable to allocate %zu bytes of memory for hio request.", 397204076Spjd sizeof(*hio)); 398204076Spjd } 399204076Spjd hio->hio_countdown = 0; 400204076Spjd hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 401204076Spjd if (hio->hio_errors == NULL) { 402204076Spjd primary_exitx(EX_TEMPFAIL, 403204076Spjd "Unable allocate %zu bytes of memory for hio errors.", 404204076Spjd sizeof(hio->hio_errors[0]) * ncomps); 405204076Spjd } 406204076Spjd hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 407204076Spjd if (hio->hio_next == NULL) { 408204076Spjd primary_exitx(EX_TEMPFAIL, 409204076Spjd "Unable allocate %zu bytes of memory for hio_next field.", 410204076Spjd sizeof(hio->hio_next[0]) * ncomps); 411204076Spjd } 412204076Spjd hio->hio_ggio.gctl_version = G_GATE_VERSION; 413204076Spjd hio->hio_ggio.gctl_data = malloc(MAXPHYS); 414204076Spjd if (hio->hio_ggio.gctl_data == NULL) { 415204076Spjd primary_exitx(EX_TEMPFAIL, 416204076Spjd "Unable to allocate %zu bytes of memory for gctl_data.", 417204076Spjd MAXPHYS); 418204076Spjd } 419204076Spjd hio->hio_ggio.gctl_length = MAXPHYS; 420204076Spjd hio->hio_ggio.gctl_error = 0; 421204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 422204076Spjd } 423204076Spjd 424204076Spjd /* 425204076Spjd * Turn on signals handling. 426204076Spjd */ 427211982Spjd PJDLOG_VERIFY(sigfillset(&mask) == 0); 428211982Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 429204076Spjd} 430204076Spjd 431204076Spjdstatic void 432204076Spjdinit_local(struct hast_resource *res) 433204076Spjd{ 434204076Spjd unsigned char *buf; 435204076Spjd size_t mapsize; 436204076Spjd 437204076Spjd if (metadata_read(res, true) < 0) 438204076Spjd exit(EX_NOINPUT); 439204076Spjd mtx_init(&res->hr_amp_lock); 440204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 441204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 442204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 443204076Spjd } 444204076Spjd mtx_init(&range_lock); 445204076Spjd cv_init(&range_regular_cond); 446204076Spjd if (rangelock_init(&range_regular) < 0) 447204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 448204076Spjd cv_init(&range_sync_cond); 449204076Spjd if (rangelock_init(&range_sync) < 0) 450204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 451204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 452204076Spjd buf = calloc(1, mapsize); 453204076Spjd if (buf == NULL) { 454204076Spjd primary_exitx(EX_TEMPFAIL, 455204076Spjd "Unable to allocate buffer for activemap."); 456204076Spjd } 457204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 458204076Spjd (ssize_t)mapsize) { 459204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 460204076Spjd } 461204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 462209181Spjd free(buf); 463204076Spjd if (res->hr_resuid != 0) 464204076Spjd return; 465204076Spjd /* 466204076Spjd * We're using provider for the first time, so we have to generate 467204076Spjd * resource unique identifier and initialize local and remote counts. 468204076Spjd */ 469204076Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 470204076Spjd res->hr_primary_localcnt = 1; 471204076Spjd res->hr_primary_remotecnt = 0; 472204076Spjd if (metadata_write(res) < 0) 473204076Spjd exit(EX_NOINPUT); 474204076Spjd} 475204076Spjd 476205738Spjdstatic bool 477205738Spjdinit_remote(struct hast_resource *res, struct proto_conn **inp, 478205738Spjd struct proto_conn **outp) 479204076Spjd{ 480205738Spjd struct proto_conn *in, *out; 481204076Spjd struct nv *nvout, *nvin; 482204076Spjd const unsigned char *token; 483204076Spjd unsigned char *map; 484204076Spjd const char *errmsg; 485204076Spjd int32_t extentsize; 486204076Spjd int64_t datasize; 487204076Spjd uint32_t mapsize; 488204076Spjd size_t size; 489204076Spjd 490205738Spjd assert((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 491210881Spjd assert(real_remote(res)); 492205738Spjd 493205738Spjd in = out = NULL; 494211983Spjd errmsg = NULL; 495205738Spjd 496204076Spjd /* Prepare outgoing connection with remote node. */ 497205738Spjd if (proto_client(res->hr_remoteaddr, &out) < 0) { 498207347Spjd primary_exit(EX_TEMPFAIL, "Unable to create connection to %s", 499204076Spjd res->hr_remoteaddr); 500204076Spjd } 501204076Spjd /* Try to connect, but accept failure. */ 502205738Spjd if (proto_connect(out) < 0) { 503204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 504204076Spjd res->hr_remoteaddr); 505204076Spjd goto close; 506204076Spjd } 507207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 508207371Spjd if (proto_timeout(out, res->hr_timeout) < 0) 509207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 510204076Spjd /* 511204076Spjd * First handshake step. 512204076Spjd * Setup outgoing connection with remote node. 513204076Spjd */ 514204076Spjd nvout = nv_alloc(); 515204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 516204076Spjd if (nv_error(nvout) != 0) { 517204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 518204076Spjd "Unable to allocate header for connection with %s", 519204076Spjd res->hr_remoteaddr); 520204076Spjd nv_free(nvout); 521204076Spjd goto close; 522204076Spjd } 523205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 524204076Spjd pjdlog_errno(LOG_WARNING, 525204076Spjd "Unable to send handshake header to %s", 526204076Spjd res->hr_remoteaddr); 527204076Spjd nv_free(nvout); 528204076Spjd goto close; 529204076Spjd } 530204076Spjd nv_free(nvout); 531205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 532204076Spjd pjdlog_errno(LOG_WARNING, 533204076Spjd "Unable to receive handshake header from %s", 534204076Spjd res->hr_remoteaddr); 535204076Spjd goto close; 536204076Spjd } 537204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 538204076Spjd if (errmsg != NULL) { 539204076Spjd pjdlog_warning("%s", errmsg); 540204076Spjd nv_free(nvin); 541204076Spjd goto close; 542204076Spjd } 543204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 544204076Spjd if (token == NULL) { 545204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 546204076Spjd res->hr_remoteaddr); 547204076Spjd nv_free(nvin); 548204076Spjd goto close; 549204076Spjd } 550204076Spjd if (size != sizeof(res->hr_token)) { 551204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 552204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 553204076Spjd nv_free(nvin); 554204076Spjd goto close; 555204076Spjd } 556204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 557204076Spjd nv_free(nvin); 558204076Spjd 559204076Spjd /* 560204076Spjd * Second handshake step. 561204076Spjd * Setup incoming connection with remote node. 562204076Spjd */ 563205738Spjd if (proto_client(res->hr_remoteaddr, &in) < 0) { 564204076Spjd pjdlog_errno(LOG_WARNING, "Unable to create connection to %s", 565204076Spjd res->hr_remoteaddr); 566204076Spjd } 567204076Spjd /* Try to connect, but accept failure. */ 568205738Spjd if (proto_connect(in) < 0) { 569204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 570204076Spjd res->hr_remoteaddr); 571204076Spjd goto close; 572204076Spjd } 573207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 574207371Spjd if (proto_timeout(in, res->hr_timeout) < 0) 575207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 576204076Spjd nvout = nv_alloc(); 577204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 578204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 579204076Spjd "token"); 580204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 581204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 582204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 583204076Spjd if (nv_error(nvout) != 0) { 584204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 585204076Spjd "Unable to allocate header for connection with %s", 586204076Spjd res->hr_remoteaddr); 587204076Spjd nv_free(nvout); 588204076Spjd goto close; 589204076Spjd } 590205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 591204076Spjd pjdlog_errno(LOG_WARNING, 592204076Spjd "Unable to send handshake header to %s", 593204076Spjd res->hr_remoteaddr); 594204076Spjd nv_free(nvout); 595204076Spjd goto close; 596204076Spjd } 597204076Spjd nv_free(nvout); 598205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 599204076Spjd pjdlog_errno(LOG_WARNING, 600204076Spjd "Unable to receive handshake header from %s", 601204076Spjd res->hr_remoteaddr); 602204076Spjd goto close; 603204076Spjd } 604204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 605204076Spjd if (errmsg != NULL) { 606204076Spjd pjdlog_warning("%s", errmsg); 607204076Spjd nv_free(nvin); 608204076Spjd goto close; 609204076Spjd } 610204076Spjd datasize = nv_get_int64(nvin, "datasize"); 611204076Spjd if (datasize != res->hr_datasize) { 612204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 613204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 614204076Spjd nv_free(nvin); 615204076Spjd goto close; 616204076Spjd } 617204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 618204076Spjd if (extentsize != res->hr_extentsize) { 619204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 620204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 621204076Spjd nv_free(nvin); 622204076Spjd goto close; 623204076Spjd } 624204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 625204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 626204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 627204076Spjd map = NULL; 628204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 629204076Spjd if (mapsize > 0) { 630204076Spjd map = malloc(mapsize); 631204076Spjd if (map == NULL) { 632204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 633204076Spjd (uintmax_t)mapsize); 634204076Spjd nv_free(nvin); 635204076Spjd goto close; 636204076Spjd } 637204076Spjd /* 638204076Spjd * Remote node have some dirty extents on its own, lets 639204076Spjd * download its activemap. 640204076Spjd */ 641205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 642204076Spjd mapsize) < 0) { 643204076Spjd pjdlog_errno(LOG_ERR, 644204076Spjd "Unable to receive remote activemap"); 645204076Spjd nv_free(nvin); 646204076Spjd free(map); 647204076Spjd goto close; 648204076Spjd } 649204076Spjd /* 650204076Spjd * Merge local and remote bitmaps. 651204076Spjd */ 652204076Spjd activemap_merge(res->hr_amp, map, mapsize); 653204076Spjd free(map); 654204076Spjd /* 655204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 656204076Spjd * disk before we start to synchronize. 657204076Spjd */ 658204076Spjd (void)hast_activemap_flush(res); 659204076Spjd } 660204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 661205738Spjd if (inp != NULL && outp != NULL) { 662205738Spjd *inp = in; 663205738Spjd *outp = out; 664205738Spjd } else { 665205738Spjd res->hr_remotein = in; 666205738Spjd res->hr_remoteout = out; 667205738Spjd } 668212038Spjd event_send(res, EVENT_CONNECT); 669205738Spjd return (true); 670205738Spjdclose: 671211983Spjd if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0) 672212038Spjd event_send(res, EVENT_SPLITBRAIN); 673205738Spjd proto_close(out); 674205738Spjd if (in != NULL) 675205738Spjd proto_close(in); 676205738Spjd return (false); 677205738Spjd} 678205738Spjd 679205738Spjdstatic void 680205738Spjdsync_start(void) 681205738Spjd{ 682205738Spjd 683204076Spjd mtx_lock(&sync_lock); 684204076Spjd sync_inprogress = true; 685204076Spjd mtx_unlock(&sync_lock); 686204076Spjd cv_signal(&sync_cond); 687204076Spjd} 688204076Spjd 689204076Spjdstatic void 690211878Spjdsync_stop(void) 691211878Spjd{ 692211878Spjd 693211878Spjd mtx_lock(&sync_lock); 694211878Spjd if (sync_inprogress) 695211878Spjd sync_inprogress = false; 696211878Spjd mtx_unlock(&sync_lock); 697211878Spjd} 698211878Spjd 699211878Spjdstatic void 700204076Spjdinit_ggate(struct hast_resource *res) 701204076Spjd{ 702204076Spjd struct g_gate_ctl_create ggiocreate; 703204076Spjd struct g_gate_ctl_cancel ggiocancel; 704204076Spjd 705204076Spjd /* 706204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 707204076Spjd */ 708204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 709204076Spjd if (res->hr_ggatefd < 0) 710204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 711204076Spjd /* 712204076Spjd * Create provider before trying to connect, as connection failure 713204076Spjd * is not critical, but may take some time. 714204076Spjd */ 715204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 716204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 717204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 718204076Spjd ggiocreate.gctl_flags = 0; 719206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 720204076Spjd ggiocreate.gctl_timeout = 0; 721204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 722204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 723204076Spjd res->hr_provname); 724204076Spjd bzero(ggiocreate.gctl_info, sizeof(ggiocreate.gctl_info)); 725204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 726204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 727204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 728204076Spjd return; 729204076Spjd } 730204076Spjd if (errno != EEXIST) { 731204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 732204076Spjd res->hr_provname); 733204076Spjd } 734204076Spjd pjdlog_debug(1, 735204076Spjd "Device hast/%s already exists, we will try to take it over.", 736204076Spjd res->hr_provname); 737204076Spjd /* 738204076Spjd * If we received EEXIST, we assume that the process who created the 739204076Spjd * provider died and didn't clean up. In that case we will start from 740204076Spjd * where he left of. 741204076Spjd */ 742204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 743204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 744204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 745204076Spjd res->hr_provname); 746204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 747204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 748204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 749204076Spjd return; 750204076Spjd } 751204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 752204076Spjd res->hr_provname); 753204076Spjd} 754204076Spjd 755204076Spjdvoid 756204076Spjdhastd_primary(struct hast_resource *res) 757204076Spjd{ 758204076Spjd pthread_t td; 759204076Spjd pid_t pid; 760204076Spjd int error; 761204076Spjd 762204076Spjd /* 763204076Spjd * Create communication channel between parent and child. 764204076Spjd */ 765204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 766204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 767212034Spjd pjdlog_exit(EX_OSERR, 768204076Spjd "Unable to create control sockets between parent and child"); 769204076Spjd } 770212038Spjd /* 771212038Spjd * Create communication channel between child and parent. 772212038Spjd */ 773212038Spjd if (proto_client("socketpair://", &res->hr_event) < 0) { 774212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 775212038Spjd pjdlog_exit(EX_OSERR, 776212038Spjd "Unable to create event sockets between child and parent"); 777212038Spjd } 778204076Spjd 779204076Spjd pid = fork(); 780204076Spjd if (pid < 0) { 781204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 782212034Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to fork"); 783204076Spjd } 784204076Spjd 785204076Spjd if (pid > 0) { 786204076Spjd /* This is parent. */ 787212038Spjd /* Declare that we are receiver. */ 788212038Spjd proto_recv(res->hr_event, NULL, 0); 789204076Spjd res->hr_workerpid = pid; 790204076Spjd return; 791204076Spjd } 792211977Spjd 793211984Spjd gres = res; 794211984Spjd 795204076Spjd (void)pidfile_close(pfh); 796211977Spjd hook_fini(); 797204076Spjd 798204076Spjd setproctitle("%s (primary)", res->hr_name); 799204076Spjd 800210880Spjd signal(SIGHUP, SIG_DFL); 801210880Spjd signal(SIGCHLD, SIG_DFL); 802210880Spjd 803212038Spjd /* Declare that we are sender. */ 804212038Spjd proto_send(res->hr_event, NULL, 0); 805212038Spjd 806204076Spjd init_local(res); 807210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 808205738Spjd sync_start(); 809204076Spjd init_ggate(res); 810204076Spjd init_environment(res); 811204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 812204076Spjd assert(error == 0); 813204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 814204076Spjd assert(error == 0); 815204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 816204076Spjd assert(error == 0); 817204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 818204076Spjd assert(error == 0); 819204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 820204076Spjd assert(error == 0); 821204076Spjd error = pthread_create(&td, NULL, sync_thread, res); 822204076Spjd assert(error == 0); 823204076Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 824204076Spjd assert(error == 0); 825204076Spjd (void)guard_thread(res); 826204076Spjd} 827204076Spjd 828204076Spjdstatic void 829204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 830204076Spjd{ 831204076Spjd char msg[1024]; 832204076Spjd va_list ap; 833204076Spjd int len; 834204076Spjd 835204076Spjd va_start(ap, fmt); 836204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 837204076Spjd va_end(ap); 838204076Spjd if ((size_t)len < sizeof(msg)) { 839204076Spjd switch (ggio->gctl_cmd) { 840204076Spjd case BIO_READ: 841204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 842204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 843204076Spjd (uintmax_t)ggio->gctl_length); 844204076Spjd break; 845204076Spjd case BIO_DELETE: 846204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 847204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 848204076Spjd (uintmax_t)ggio->gctl_length); 849204076Spjd break; 850204076Spjd case BIO_FLUSH: 851204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 852204076Spjd break; 853204076Spjd case BIO_WRITE: 854204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 855204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 856204076Spjd (uintmax_t)ggio->gctl_length); 857204076Spjd break; 858204076Spjd default: 859204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 860204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 861204076Spjd break; 862204076Spjd } 863204076Spjd } 864204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 865204076Spjd} 866204076Spjd 867204076Spjdstatic void 868204076Spjdremote_close(struct hast_resource *res, int ncomp) 869204076Spjd{ 870204076Spjd 871204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 872204076Spjd /* 873204076Spjd * A race is possible between dropping rlock and acquiring wlock - 874204076Spjd * another thread can close connection in-between. 875204076Spjd */ 876204076Spjd if (!ISCONNECTED(res, ncomp)) { 877204076Spjd assert(res->hr_remotein == NULL); 878204076Spjd assert(res->hr_remoteout == NULL); 879204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 880204076Spjd return; 881204076Spjd } 882204076Spjd 883204076Spjd assert(res->hr_remotein != NULL); 884204076Spjd assert(res->hr_remoteout != NULL); 885204076Spjd 886211881Spjd pjdlog_debug(2, "Closing incoming connection to %s.", 887204076Spjd res->hr_remoteaddr); 888204076Spjd proto_close(res->hr_remotein); 889204076Spjd res->hr_remotein = NULL; 890211881Spjd pjdlog_debug(2, "Closing outgoing connection to %s.", 891204076Spjd res->hr_remoteaddr); 892204076Spjd proto_close(res->hr_remoteout); 893204076Spjd res->hr_remoteout = NULL; 894204076Spjd 895204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 896204076Spjd 897211881Spjd pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 898211881Spjd 899204076Spjd /* 900204076Spjd * Stop synchronization if in-progress. 901204076Spjd */ 902211878Spjd sync_stop(); 903211984Spjd 904212038Spjd event_send(res, EVENT_DISCONNECT); 905204076Spjd} 906204076Spjd 907204076Spjd/* 908204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 909204076Spjd * appropriate threads: 910204076Spjd * WRITE - always goes to both local_send and remote_send threads 911204076Spjd * READ (when the block is up-to-date on local component) - 912204076Spjd * only local_send thread 913204076Spjd * READ (when the block isn't up-to-date on local component) - 914204076Spjd * only remote_send thread 915204076Spjd * DELETE - always goes to both local_send and remote_send threads 916204076Spjd * FLUSH - always goes to both local_send and remote_send threads 917204076Spjd */ 918204076Spjdstatic void * 919204076Spjdggate_recv_thread(void *arg) 920204076Spjd{ 921204076Spjd struct hast_resource *res = arg; 922204076Spjd struct g_gate_ctl_io *ggio; 923204076Spjd struct hio *hio; 924204076Spjd unsigned int ii, ncomp, ncomps; 925204076Spjd int error; 926204076Spjd 927204076Spjd ncomps = HAST_NCOMPONENTS; 928204076Spjd 929204076Spjd for (;;) { 930204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 931204076Spjd QUEUE_TAKE2(hio, free); 932204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 933204076Spjd ggio = &hio->hio_ggio; 934204076Spjd ggio->gctl_unit = res->hr_ggateunit; 935204076Spjd ggio->gctl_length = MAXPHYS; 936204076Spjd ggio->gctl_error = 0; 937204076Spjd pjdlog_debug(2, 938204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 939204076Spjd hio); 940204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 941204076Spjd if (sigexit_received) 942204076Spjd pthread_exit(NULL); 943204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 944204076Spjd } 945204076Spjd error = ggio->gctl_error; 946204076Spjd switch (error) { 947204076Spjd case 0: 948204076Spjd break; 949204076Spjd case ECANCELED: 950204076Spjd /* Exit gracefully. */ 951204076Spjd if (!sigexit_received) { 952204076Spjd pjdlog_debug(2, 953204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 954204076Spjd hio); 955204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 956204076Spjd } 957204076Spjd pthread_exit(NULL); 958204076Spjd case ENOMEM: 959204076Spjd /* 960204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 961204076Spjd * bytes - request can't be bigger than that. 962204076Spjd */ 963204076Spjd /* FALLTHROUGH */ 964204076Spjd case ENXIO: 965204076Spjd default: 966204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 967204076Spjd strerror(error)); 968204076Spjd } 969204076Spjd for (ii = 0; ii < ncomps; ii++) 970204076Spjd hio->hio_errors[ii] = EINVAL; 971204076Spjd reqlog(LOG_DEBUG, 2, ggio, 972204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 973204076Spjd hio); 974204076Spjd /* 975204076Spjd * Inform all components about new write request. 976204076Spjd * For read request prefer local component unless the given 977204076Spjd * range is out-of-date, then use remote component. 978204076Spjd */ 979204076Spjd switch (ggio->gctl_cmd) { 980204076Spjd case BIO_READ: 981204076Spjd pjdlog_debug(2, 982204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 983204076Spjd hio); 984204076Spjd refcount_init(&hio->hio_countdown, 1); 985204076Spjd mtx_lock(&metadata_lock); 986204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 987204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 988204076Spjd /* 989204076Spjd * This range is up-to-date on local component, 990204076Spjd * so handle request locally. 991204076Spjd */ 992204076Spjd /* Local component is 0 for now. */ 993204076Spjd ncomp = 0; 994204076Spjd } else /* if (res->hr_syncsrc == 995204076Spjd HAST_SYNCSRC_SECONDARY) */ { 996204076Spjd assert(res->hr_syncsrc == 997204076Spjd HAST_SYNCSRC_SECONDARY); 998204076Spjd /* 999204076Spjd * This range is out-of-date on local component, 1000204076Spjd * so send request to the remote node. 1001204076Spjd */ 1002204076Spjd /* Remote component is 1 for now. */ 1003204076Spjd ncomp = 1; 1004204076Spjd } 1005204076Spjd mtx_unlock(&metadata_lock); 1006204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1007204076Spjd break; 1008204076Spjd case BIO_WRITE: 1009204076Spjd for (;;) { 1010204076Spjd mtx_lock(&range_lock); 1011204076Spjd if (rangelock_islocked(range_sync, 1012204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1013204076Spjd pjdlog_debug(2, 1014204076Spjd "regular: Range offset=%jd length=%zu locked.", 1015204076Spjd (intmax_t)ggio->gctl_offset, 1016204076Spjd (size_t)ggio->gctl_length); 1017204076Spjd range_regular_wait = true; 1018204076Spjd cv_wait(&range_regular_cond, &range_lock); 1019204076Spjd range_regular_wait = false; 1020204076Spjd mtx_unlock(&range_lock); 1021204076Spjd continue; 1022204076Spjd } 1023204076Spjd if (rangelock_add(range_regular, 1024204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1025204076Spjd mtx_unlock(&range_lock); 1026204076Spjd pjdlog_debug(2, 1027204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1028204076Spjd (intmax_t)ggio->gctl_offset, 1029204076Spjd (size_t)ggio->gctl_length); 1030204076Spjd sleep(1); 1031204076Spjd continue; 1032204076Spjd } 1033204076Spjd mtx_unlock(&range_lock); 1034204076Spjd break; 1035204076Spjd } 1036204076Spjd mtx_lock(&res->hr_amp_lock); 1037204076Spjd if (activemap_write_start(res->hr_amp, 1038204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1039204076Spjd (void)hast_activemap_flush(res); 1040204076Spjd } 1041204076Spjd mtx_unlock(&res->hr_amp_lock); 1042204076Spjd /* FALLTHROUGH */ 1043204076Spjd case BIO_DELETE: 1044204076Spjd case BIO_FLUSH: 1045204076Spjd pjdlog_debug(2, 1046204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1047204076Spjd hio); 1048204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1049204076Spjd for (ii = 0; ii < ncomps; ii++) 1050204076Spjd QUEUE_INSERT1(hio, send, ii); 1051204076Spjd break; 1052204076Spjd } 1053204076Spjd } 1054204076Spjd /* NOTREACHED */ 1055204076Spjd return (NULL); 1056204076Spjd} 1057204076Spjd 1058204076Spjd/* 1059204076Spjd * Thread reads from or writes to local component. 1060204076Spjd * If local read fails, it redirects it to remote_send thread. 1061204076Spjd */ 1062204076Spjdstatic void * 1063204076Spjdlocal_send_thread(void *arg) 1064204076Spjd{ 1065204076Spjd struct hast_resource *res = arg; 1066204076Spjd struct g_gate_ctl_io *ggio; 1067204076Spjd struct hio *hio; 1068204076Spjd unsigned int ncomp, rncomp; 1069204076Spjd ssize_t ret; 1070204076Spjd 1071204076Spjd /* Local component is 0 for now. */ 1072204076Spjd ncomp = 0; 1073204076Spjd /* Remote component is 1 for now. */ 1074204076Spjd rncomp = 1; 1075204076Spjd 1076204076Spjd for (;;) { 1077204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1078204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1079204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1080204076Spjd ggio = &hio->hio_ggio; 1081204076Spjd switch (ggio->gctl_cmd) { 1082204076Spjd case BIO_READ: 1083204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1084204076Spjd ggio->gctl_length, 1085204076Spjd ggio->gctl_offset + res->hr_localoff); 1086204076Spjd if (ret == ggio->gctl_length) 1087204076Spjd hio->hio_errors[ncomp] = 0; 1088204076Spjd else { 1089204076Spjd /* 1090204076Spjd * If READ failed, try to read from remote node. 1091204076Spjd */ 1092204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1093204076Spjd continue; 1094204076Spjd } 1095204076Spjd break; 1096204076Spjd case BIO_WRITE: 1097204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1098204076Spjd ggio->gctl_length, 1099204076Spjd ggio->gctl_offset + res->hr_localoff); 1100204076Spjd if (ret < 0) 1101204076Spjd hio->hio_errors[ncomp] = errno; 1102204076Spjd else if (ret != ggio->gctl_length) 1103204076Spjd hio->hio_errors[ncomp] = EIO; 1104204076Spjd else 1105204076Spjd hio->hio_errors[ncomp] = 0; 1106204076Spjd break; 1107204076Spjd case BIO_DELETE: 1108204076Spjd ret = g_delete(res->hr_localfd, 1109204076Spjd ggio->gctl_offset + res->hr_localoff, 1110204076Spjd ggio->gctl_length); 1111204076Spjd if (ret < 0) 1112204076Spjd hio->hio_errors[ncomp] = errno; 1113204076Spjd else 1114204076Spjd hio->hio_errors[ncomp] = 0; 1115204076Spjd break; 1116204076Spjd case BIO_FLUSH: 1117204076Spjd ret = g_flush(res->hr_localfd); 1118204076Spjd if (ret < 0) 1119204076Spjd hio->hio_errors[ncomp] = errno; 1120204076Spjd else 1121204076Spjd hio->hio_errors[ncomp] = 0; 1122204076Spjd break; 1123204076Spjd } 1124204076Spjd if (refcount_release(&hio->hio_countdown)) { 1125204076Spjd if (ISSYNCREQ(hio)) { 1126204076Spjd mtx_lock(&sync_lock); 1127204076Spjd SYNCREQDONE(hio); 1128204076Spjd mtx_unlock(&sync_lock); 1129204076Spjd cv_signal(&sync_cond); 1130204076Spjd } else { 1131204076Spjd pjdlog_debug(2, 1132204076Spjd "local_send: (%p) Moving request to the done queue.", 1133204076Spjd hio); 1134204076Spjd QUEUE_INSERT2(hio, done); 1135204076Spjd } 1136204076Spjd } 1137204076Spjd } 1138204076Spjd /* NOTREACHED */ 1139204076Spjd return (NULL); 1140204076Spjd} 1141204076Spjd 1142204076Spjd/* 1143204076Spjd * Thread sends request to secondary node. 1144204076Spjd */ 1145204076Spjdstatic void * 1146204076Spjdremote_send_thread(void *arg) 1147204076Spjd{ 1148204076Spjd struct hast_resource *res = arg; 1149204076Spjd struct g_gate_ctl_io *ggio; 1150204076Spjd struct hio *hio; 1151204076Spjd struct nv *nv; 1152204076Spjd unsigned int ncomp; 1153204076Spjd bool wakeup; 1154204076Spjd uint64_t offset, length; 1155204076Spjd uint8_t cmd; 1156204076Spjd void *data; 1157204076Spjd 1158204076Spjd /* Remote component is 1 for now. */ 1159204076Spjd ncomp = 1; 1160204076Spjd 1161204076Spjd for (;;) { 1162204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1163204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1164204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1165204076Spjd ggio = &hio->hio_ggio; 1166204076Spjd switch (ggio->gctl_cmd) { 1167204076Spjd case BIO_READ: 1168204076Spjd cmd = HIO_READ; 1169204076Spjd data = NULL; 1170204076Spjd offset = ggio->gctl_offset; 1171204076Spjd length = ggio->gctl_length; 1172204076Spjd break; 1173204076Spjd case BIO_WRITE: 1174204076Spjd cmd = HIO_WRITE; 1175204076Spjd data = ggio->gctl_data; 1176204076Spjd offset = ggio->gctl_offset; 1177204076Spjd length = ggio->gctl_length; 1178204076Spjd break; 1179204076Spjd case BIO_DELETE: 1180204076Spjd cmd = HIO_DELETE; 1181204076Spjd data = NULL; 1182204076Spjd offset = ggio->gctl_offset; 1183204076Spjd length = ggio->gctl_length; 1184204076Spjd break; 1185204076Spjd case BIO_FLUSH: 1186204076Spjd cmd = HIO_FLUSH; 1187204076Spjd data = NULL; 1188204076Spjd offset = 0; 1189204076Spjd length = 0; 1190204076Spjd break; 1191204076Spjd default: 1192204076Spjd assert(!"invalid condition"); 1193204076Spjd abort(); 1194204076Spjd } 1195204076Spjd nv = nv_alloc(); 1196204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1197204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1198204076Spjd nv_add_uint64(nv, offset, "offset"); 1199204076Spjd nv_add_uint64(nv, length, "length"); 1200204076Spjd if (nv_error(nv) != 0) { 1201204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1202204076Spjd pjdlog_debug(2, 1203204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1204204076Spjd hio); 1205204076Spjd reqlog(LOG_ERR, 0, ggio, 1206204076Spjd "Unable to prepare header to send (%s): ", 1207204076Spjd strerror(nv_error(nv))); 1208204076Spjd /* Move failed request immediately to the done queue. */ 1209204076Spjd goto done_queue; 1210204076Spjd } 1211204076Spjd pjdlog_debug(2, 1212204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1213204076Spjd hio); 1214204076Spjd /* 1215204076Spjd * Protect connection from disappearing. 1216204076Spjd */ 1217204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1218204076Spjd if (!ISCONNECTED(res, ncomp)) { 1219204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1220204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1221204076Spjd goto done_queue; 1222204076Spjd } 1223204076Spjd /* 1224204076Spjd * Move the request to recv queue before sending it, because 1225204076Spjd * in different order we can get reply before we move request 1226204076Spjd * to recv queue. 1227204076Spjd */ 1228204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1229204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1230204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1231204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1232204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1233204076Spjd data != NULL ? length : 0) < 0) { 1234204076Spjd hio->hio_errors[ncomp] = errno; 1235204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1236204076Spjd pjdlog_debug(2, 1237204076Spjd "remote_send: (%p) Unable to send request.", hio); 1238204076Spjd reqlog(LOG_ERR, 0, ggio, 1239204076Spjd "Unable to send request (%s): ", 1240204076Spjd strerror(hio->hio_errors[ncomp])); 1241211979Spjd remote_close(res, ncomp); 1242204076Spjd /* 1243204076Spjd * Take request back from the receive queue and move 1244204076Spjd * it immediately to the done queue. 1245204076Spjd */ 1246204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1247204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1248204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1249204076Spjd goto done_queue; 1250204076Spjd } 1251204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1252204076Spjd nv_free(nv); 1253204076Spjd if (wakeup) 1254204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1255204076Spjd continue; 1256204076Spjddone_queue: 1257204076Spjd nv_free(nv); 1258204076Spjd if (ISSYNCREQ(hio)) { 1259204076Spjd if (!refcount_release(&hio->hio_countdown)) 1260204076Spjd continue; 1261204076Spjd mtx_lock(&sync_lock); 1262204076Spjd SYNCREQDONE(hio); 1263204076Spjd mtx_unlock(&sync_lock); 1264204076Spjd cv_signal(&sync_cond); 1265204076Spjd continue; 1266204076Spjd } 1267204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1268204076Spjd mtx_lock(&res->hr_amp_lock); 1269204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1270204076Spjd ggio->gctl_length)) { 1271204076Spjd (void)hast_activemap_flush(res); 1272204076Spjd } 1273204076Spjd mtx_unlock(&res->hr_amp_lock); 1274204076Spjd } 1275204076Spjd if (!refcount_release(&hio->hio_countdown)) 1276204076Spjd continue; 1277204076Spjd pjdlog_debug(2, 1278204076Spjd "remote_send: (%p) Moving request to the done queue.", 1279204076Spjd hio); 1280204076Spjd QUEUE_INSERT2(hio, done); 1281204076Spjd } 1282204076Spjd /* NOTREACHED */ 1283204076Spjd return (NULL); 1284204076Spjd} 1285204076Spjd 1286204076Spjd/* 1287204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1288204076Spjd * thread. 1289204076Spjd */ 1290204076Spjdstatic void * 1291204076Spjdremote_recv_thread(void *arg) 1292204076Spjd{ 1293204076Spjd struct hast_resource *res = arg; 1294204076Spjd struct g_gate_ctl_io *ggio; 1295204076Spjd struct hio *hio; 1296204076Spjd struct nv *nv; 1297204076Spjd unsigned int ncomp; 1298204076Spjd uint64_t seq; 1299204076Spjd int error; 1300204076Spjd 1301204076Spjd /* Remote component is 1 for now. */ 1302204076Spjd ncomp = 1; 1303204076Spjd 1304204076Spjd for (;;) { 1305204076Spjd /* Wait until there is anything to receive. */ 1306204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1307204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1308204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1309204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1310204076Spjd &hio_recv_list_lock[ncomp]); 1311204076Spjd } 1312204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1313204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1314204076Spjd if (!ISCONNECTED(res, ncomp)) { 1315204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1316204076Spjd /* 1317204076Spjd * Connection is dead, so move all pending requests to 1318204076Spjd * the done queue (one-by-one). 1319204076Spjd */ 1320204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1321204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1322204076Spjd assert(hio != NULL); 1323204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1324204076Spjd hio_next[ncomp]); 1325204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1326204076Spjd goto done_queue; 1327204076Spjd } 1328204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1329204076Spjd pjdlog_errno(LOG_ERR, 1330204076Spjd "Unable to receive reply header"); 1331204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1332204076Spjd remote_close(res, ncomp); 1333204076Spjd continue; 1334204076Spjd } 1335204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1336204076Spjd seq = nv_get_uint64(nv, "seq"); 1337204076Spjd if (seq == 0) { 1338204076Spjd pjdlog_error("Header contains no 'seq' field."); 1339204076Spjd nv_free(nv); 1340204076Spjd continue; 1341204076Spjd } 1342204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1343204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1344204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1345204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1346204076Spjd hio_next[ncomp]); 1347204076Spjd break; 1348204076Spjd } 1349204076Spjd } 1350204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1351204076Spjd if (hio == NULL) { 1352204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1353204076Spjd (uintmax_t)seq); 1354204076Spjd nv_free(nv); 1355204076Spjd continue; 1356204076Spjd } 1357204076Spjd error = nv_get_int16(nv, "error"); 1358204076Spjd if (error != 0) { 1359204076Spjd /* Request failed on remote side. */ 1360204076Spjd hio->hio_errors[ncomp] = 0; 1361204076Spjd nv_free(nv); 1362204076Spjd goto done_queue; 1363204076Spjd } 1364204076Spjd ggio = &hio->hio_ggio; 1365204076Spjd switch (ggio->gctl_cmd) { 1366204076Spjd case BIO_READ: 1367204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1368204076Spjd if (!ISCONNECTED(res, ncomp)) { 1369204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1370204076Spjd nv_free(nv); 1371204076Spjd goto done_queue; 1372204076Spjd } 1373204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1374204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1375204076Spjd hio->hio_errors[ncomp] = errno; 1376204076Spjd pjdlog_errno(LOG_ERR, 1377204076Spjd "Unable to receive reply data"); 1378204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1379204076Spjd nv_free(nv); 1380204076Spjd remote_close(res, ncomp); 1381204076Spjd goto done_queue; 1382204076Spjd } 1383204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1384204076Spjd break; 1385204076Spjd case BIO_WRITE: 1386204076Spjd case BIO_DELETE: 1387204076Spjd case BIO_FLUSH: 1388204076Spjd break; 1389204076Spjd default: 1390204076Spjd assert(!"invalid condition"); 1391204076Spjd abort(); 1392204076Spjd } 1393204076Spjd hio->hio_errors[ncomp] = 0; 1394204076Spjd nv_free(nv); 1395204076Spjddone_queue: 1396204076Spjd if (refcount_release(&hio->hio_countdown)) { 1397204076Spjd if (ISSYNCREQ(hio)) { 1398204076Spjd mtx_lock(&sync_lock); 1399204076Spjd SYNCREQDONE(hio); 1400204076Spjd mtx_unlock(&sync_lock); 1401204076Spjd cv_signal(&sync_cond); 1402204076Spjd } else { 1403204076Spjd pjdlog_debug(2, 1404204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1405204076Spjd hio); 1406204076Spjd QUEUE_INSERT2(hio, done); 1407204076Spjd } 1408204076Spjd } 1409204076Spjd } 1410204076Spjd /* NOTREACHED */ 1411204076Spjd return (NULL); 1412204076Spjd} 1413204076Spjd 1414204076Spjd/* 1415204076Spjd * Thread sends answer to the kernel. 1416204076Spjd */ 1417204076Spjdstatic void * 1418204076Spjdggate_send_thread(void *arg) 1419204076Spjd{ 1420204076Spjd struct hast_resource *res = arg; 1421204076Spjd struct g_gate_ctl_io *ggio; 1422204076Spjd struct hio *hio; 1423204076Spjd unsigned int ii, ncomp, ncomps; 1424204076Spjd 1425204076Spjd ncomps = HAST_NCOMPONENTS; 1426204076Spjd 1427204076Spjd for (;;) { 1428204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1429204076Spjd QUEUE_TAKE2(hio, done); 1430204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1431204076Spjd ggio = &hio->hio_ggio; 1432204076Spjd for (ii = 0; ii < ncomps; ii++) { 1433204076Spjd if (hio->hio_errors[ii] == 0) { 1434204076Spjd /* 1435204076Spjd * One successful request is enough to declare 1436204076Spjd * success. 1437204076Spjd */ 1438204076Spjd ggio->gctl_error = 0; 1439204076Spjd break; 1440204076Spjd } 1441204076Spjd } 1442204076Spjd if (ii == ncomps) { 1443204076Spjd /* 1444204076Spjd * None of the requests were successful. 1445204076Spjd * Use first error. 1446204076Spjd */ 1447204076Spjd ggio->gctl_error = hio->hio_errors[0]; 1448204076Spjd } 1449204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1450204076Spjd mtx_lock(&res->hr_amp_lock); 1451204076Spjd activemap_write_complete(res->hr_amp, 1452204076Spjd ggio->gctl_offset, ggio->gctl_length); 1453204076Spjd mtx_unlock(&res->hr_amp_lock); 1454204076Spjd } 1455204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1456204076Spjd /* 1457204076Spjd * Unlock range we locked. 1458204076Spjd */ 1459204076Spjd mtx_lock(&range_lock); 1460204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1461204076Spjd ggio->gctl_length); 1462204076Spjd if (range_sync_wait) 1463204076Spjd cv_signal(&range_sync_cond); 1464204076Spjd mtx_unlock(&range_lock); 1465204076Spjd /* 1466204076Spjd * Bump local count if this is first write after 1467204076Spjd * connection failure with remote node. 1468204076Spjd */ 1469204076Spjd ncomp = 1; 1470204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1471204076Spjd if (!ISCONNECTED(res, ncomp)) { 1472204076Spjd mtx_lock(&metadata_lock); 1473204076Spjd if (res->hr_primary_localcnt == 1474204076Spjd res->hr_secondary_remotecnt) { 1475204076Spjd res->hr_primary_localcnt++; 1476204076Spjd pjdlog_debug(1, 1477204076Spjd "Increasing localcnt to %ju.", 1478204076Spjd (uintmax_t)res->hr_primary_localcnt); 1479204076Spjd (void)metadata_write(res); 1480204076Spjd } 1481204076Spjd mtx_unlock(&metadata_lock); 1482204076Spjd } 1483204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1484204076Spjd } 1485204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1486204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1487204076Spjd pjdlog_debug(2, 1488204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1489204076Spjd QUEUE_INSERT2(hio, free); 1490204076Spjd } 1491204076Spjd /* NOTREACHED */ 1492204076Spjd return (NULL); 1493204076Spjd} 1494204076Spjd 1495204076Spjd/* 1496204076Spjd * Thread synchronize local and remote components. 1497204076Spjd */ 1498204076Spjdstatic void * 1499204076Spjdsync_thread(void *arg __unused) 1500204076Spjd{ 1501204076Spjd struct hast_resource *res = arg; 1502204076Spjd struct hio *hio; 1503204076Spjd struct g_gate_ctl_io *ggio; 1504204076Spjd unsigned int ii, ncomp, ncomps; 1505204076Spjd off_t offset, length, synced; 1506204076Spjd bool dorewind; 1507204076Spjd int syncext; 1508204076Spjd 1509204076Spjd ncomps = HAST_NCOMPONENTS; 1510204076Spjd dorewind = true; 1511211897Spjd synced = 0; 1512211897Spjd offset = -1; 1513204076Spjd 1514204076Spjd for (;;) { 1515204076Spjd mtx_lock(&sync_lock); 1516211897Spjd if (offset >= 0 && !sync_inprogress) { 1517211879Spjd pjdlog_info("Synchronization interrupted. " 1518211879Spjd "%jd bytes synchronized so far.", 1519211879Spjd (intmax_t)synced); 1520212038Spjd event_send(res, EVENT_SYNCINTR); 1521211879Spjd } 1522204076Spjd while (!sync_inprogress) { 1523204076Spjd dorewind = true; 1524204076Spjd synced = 0; 1525204076Spjd cv_wait(&sync_cond, &sync_lock); 1526204076Spjd } 1527204076Spjd mtx_unlock(&sync_lock); 1528204076Spjd /* 1529204076Spjd * Obtain offset at which we should synchronize. 1530204076Spjd * Rewind synchronization if needed. 1531204076Spjd */ 1532204076Spjd mtx_lock(&res->hr_amp_lock); 1533204076Spjd if (dorewind) 1534204076Spjd activemap_sync_rewind(res->hr_amp); 1535204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1536204076Spjd if (syncext != -1) { 1537204076Spjd /* 1538204076Spjd * We synchronized entire syncext extent, we can mark 1539204076Spjd * it as clean now. 1540204076Spjd */ 1541204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1542204076Spjd (void)hast_activemap_flush(res); 1543204076Spjd } 1544204076Spjd mtx_unlock(&res->hr_amp_lock); 1545204076Spjd if (dorewind) { 1546204076Spjd dorewind = false; 1547204076Spjd if (offset < 0) 1548204076Spjd pjdlog_info("Nodes are in sync."); 1549204076Spjd else { 1550204076Spjd pjdlog_info("Synchronization started. %ju bytes to go.", 1551204076Spjd (uintmax_t)(res->hr_extentsize * 1552204076Spjd activemap_ndirty(res->hr_amp))); 1553212038Spjd event_send(res, EVENT_SYNCSTART); 1554204076Spjd } 1555204076Spjd } 1556204076Spjd if (offset < 0) { 1557211878Spjd sync_stop(); 1558204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1559204076Spjd /* 1560204076Spjd * Synchronization complete, make both localcnt and 1561204076Spjd * remotecnt equal. 1562204076Spjd */ 1563204076Spjd ncomp = 1; 1564204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1565204076Spjd if (ISCONNECTED(res, ncomp)) { 1566204076Spjd if (synced > 0) { 1567204076Spjd pjdlog_info("Synchronization complete. " 1568204076Spjd "%jd bytes synchronized.", 1569204076Spjd (intmax_t)synced); 1570212038Spjd event_send(res, EVENT_SYNCDONE); 1571204076Spjd } 1572204076Spjd mtx_lock(&metadata_lock); 1573204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1574204076Spjd res->hr_primary_localcnt = 1575204076Spjd res->hr_secondary_localcnt; 1576204076Spjd res->hr_primary_remotecnt = 1577204076Spjd res->hr_secondary_remotecnt; 1578204076Spjd pjdlog_debug(1, 1579204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1580204076Spjd (uintmax_t)res->hr_primary_localcnt, 1581204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1582204076Spjd (void)metadata_write(res); 1583204076Spjd mtx_unlock(&metadata_lock); 1584204076Spjd } 1585204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1586204076Spjd continue; 1587204076Spjd } 1588204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1589204076Spjd QUEUE_TAKE2(hio, free); 1590204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1591204076Spjd /* 1592204076Spjd * Lock the range we are going to synchronize. We don't want 1593204076Spjd * race where someone writes between our read and write. 1594204076Spjd */ 1595204076Spjd for (;;) { 1596204076Spjd mtx_lock(&range_lock); 1597204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1598204076Spjd pjdlog_debug(2, 1599204076Spjd "sync: Range offset=%jd length=%jd locked.", 1600204076Spjd (intmax_t)offset, (intmax_t)length); 1601204076Spjd range_sync_wait = true; 1602204076Spjd cv_wait(&range_sync_cond, &range_lock); 1603204076Spjd range_sync_wait = false; 1604204076Spjd mtx_unlock(&range_lock); 1605204076Spjd continue; 1606204076Spjd } 1607204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1608204076Spjd mtx_unlock(&range_lock); 1609204076Spjd pjdlog_debug(2, 1610204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1611204076Spjd (intmax_t)offset, (intmax_t)length); 1612204076Spjd sleep(1); 1613204076Spjd continue; 1614204076Spjd } 1615204076Spjd mtx_unlock(&range_lock); 1616204076Spjd break; 1617204076Spjd } 1618204076Spjd /* 1619204076Spjd * First read the data from synchronization source. 1620204076Spjd */ 1621204076Spjd SYNCREQ(hio); 1622204076Spjd ggio = &hio->hio_ggio; 1623204076Spjd ggio->gctl_cmd = BIO_READ; 1624204076Spjd ggio->gctl_offset = offset; 1625204076Spjd ggio->gctl_length = length; 1626204076Spjd ggio->gctl_error = 0; 1627204076Spjd for (ii = 0; ii < ncomps; ii++) 1628204076Spjd hio->hio_errors[ii] = EINVAL; 1629204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1630204076Spjd hio); 1631204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1632204076Spjd hio); 1633204076Spjd mtx_lock(&metadata_lock); 1634204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1635204076Spjd /* 1636204076Spjd * This range is up-to-date on local component, 1637204076Spjd * so handle request locally. 1638204076Spjd */ 1639204076Spjd /* Local component is 0 for now. */ 1640204076Spjd ncomp = 0; 1641204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1642204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1643204076Spjd /* 1644204076Spjd * This range is out-of-date on local component, 1645204076Spjd * so send request to the remote node. 1646204076Spjd */ 1647204076Spjd /* Remote component is 1 for now. */ 1648204076Spjd ncomp = 1; 1649204076Spjd } 1650204076Spjd mtx_unlock(&metadata_lock); 1651204076Spjd refcount_init(&hio->hio_countdown, 1); 1652204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1653204076Spjd 1654204076Spjd /* 1655204076Spjd * Let's wait for READ to finish. 1656204076Spjd */ 1657204076Spjd mtx_lock(&sync_lock); 1658204076Spjd while (!ISSYNCREQDONE(hio)) 1659204076Spjd cv_wait(&sync_cond, &sync_lock); 1660204076Spjd mtx_unlock(&sync_lock); 1661204076Spjd 1662204076Spjd if (hio->hio_errors[ncomp] != 0) { 1663204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1664204076Spjd strerror(hio->hio_errors[ncomp])); 1665204076Spjd goto free_queue; 1666204076Spjd } 1667204076Spjd 1668204076Spjd /* 1669204076Spjd * We read the data from synchronization source, now write it 1670204076Spjd * to synchronization target. 1671204076Spjd */ 1672204076Spjd SYNCREQ(hio); 1673204076Spjd ggio->gctl_cmd = BIO_WRITE; 1674204076Spjd for (ii = 0; ii < ncomps; ii++) 1675204076Spjd hio->hio_errors[ii] = EINVAL; 1676204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1677204076Spjd hio); 1678204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1679204076Spjd hio); 1680204076Spjd mtx_lock(&metadata_lock); 1681204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1682204076Spjd /* 1683204076Spjd * This range is up-to-date on local component, 1684204076Spjd * so we update remote component. 1685204076Spjd */ 1686204076Spjd /* Remote component is 1 for now. */ 1687204076Spjd ncomp = 1; 1688204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1689204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1690204076Spjd /* 1691204076Spjd * This range is out-of-date on local component, 1692204076Spjd * so we update it. 1693204076Spjd */ 1694204076Spjd /* Local component is 0 for now. */ 1695204076Spjd ncomp = 0; 1696204076Spjd } 1697204076Spjd mtx_unlock(&metadata_lock); 1698204076Spjd 1699204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1700204076Spjd hio); 1701204076Spjd refcount_init(&hio->hio_countdown, 1); 1702204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1703204076Spjd 1704204076Spjd /* 1705204076Spjd * Let's wait for WRITE to finish. 1706204076Spjd */ 1707204076Spjd mtx_lock(&sync_lock); 1708204076Spjd while (!ISSYNCREQDONE(hio)) 1709204076Spjd cv_wait(&sync_cond, &sync_lock); 1710204076Spjd mtx_unlock(&sync_lock); 1711204076Spjd 1712204076Spjd if (hio->hio_errors[ncomp] != 0) { 1713204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1714204076Spjd strerror(hio->hio_errors[ncomp])); 1715204076Spjd goto free_queue; 1716204076Spjd } 1717211880Spjd 1718211880Spjd synced += length; 1719204076Spjdfree_queue: 1720204076Spjd mtx_lock(&range_lock); 1721204076Spjd rangelock_del(range_sync, offset, length); 1722204076Spjd if (range_regular_wait) 1723204076Spjd cv_signal(&range_regular_cond); 1724204076Spjd mtx_unlock(&range_lock); 1725204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1726204076Spjd hio); 1727204076Spjd QUEUE_INSERT2(hio, free); 1728204076Spjd } 1729204076Spjd /* NOTREACHED */ 1730204076Spjd return (NULL); 1731204076Spjd} 1732204076Spjd 1733204076Spjdstatic void 1734210886Spjdconfig_reload(void) 1735210886Spjd{ 1736210886Spjd struct hastd_config *newcfg; 1737210886Spjd struct hast_resource *res; 1738210886Spjd unsigned int ii, ncomps; 1739210886Spjd int modified; 1740210886Spjd 1741210886Spjd pjdlog_info("Reloading configuration..."); 1742210886Spjd 1743210886Spjd ncomps = HAST_NCOMPONENTS; 1744210886Spjd 1745210886Spjd newcfg = yy_config_parse(cfgpath, false); 1746210886Spjd if (newcfg == NULL) 1747210886Spjd goto failed; 1748210886Spjd 1749210886Spjd TAILQ_FOREACH(res, &newcfg->hc_resources, hr_next) { 1750210886Spjd if (strcmp(res->hr_name, gres->hr_name) == 0) 1751210886Spjd break; 1752210886Spjd } 1753210886Spjd /* 1754210886Spjd * If resource was removed from the configuration file, resource 1755210886Spjd * name, provider name or path to local component was modified we 1756210886Spjd * shouldn't be here. This means that someone modified configuration 1757210886Spjd * file and send SIGHUP to us instead of main hastd process. 1758210886Spjd * Log advice and ignore the signal. 1759210886Spjd */ 1760210886Spjd if (res == NULL || strcmp(gres->hr_name, res->hr_name) != 0 || 1761210886Spjd strcmp(gres->hr_provname, res->hr_provname) != 0 || 1762210886Spjd strcmp(gres->hr_localpath, res->hr_localpath) != 0) { 1763210886Spjd pjdlog_warning("To reload configuration send SIGHUP to the main hastd process (pid %u).", 1764210886Spjd (unsigned int)getppid()); 1765210886Spjd goto failed; 1766210886Spjd } 1767210886Spjd 1768210886Spjd#define MODIFIED_REMOTEADDR 0x1 1769210886Spjd#define MODIFIED_REPLICATION 0x2 1770210886Spjd#define MODIFIED_TIMEOUT 0x4 1771211886Spjd#define MODIFIED_EXEC 0x8 1772210886Spjd modified = 0; 1773210886Spjd if (strcmp(gres->hr_remoteaddr, res->hr_remoteaddr) != 0) { 1774210886Spjd /* 1775210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1776210886Spjd * We want remote_close() to log disconnect from the old 1777210886Spjd * addresses, not from the new ones. 1778210886Spjd */ 1779210886Spjd modified |= MODIFIED_REMOTEADDR; 1780210886Spjd } 1781210886Spjd if (gres->hr_replication != res->hr_replication) { 1782210886Spjd gres->hr_replication = res->hr_replication; 1783210886Spjd modified |= MODIFIED_REPLICATION; 1784210886Spjd } 1785210886Spjd if (gres->hr_timeout != res->hr_timeout) { 1786210886Spjd gres->hr_timeout = res->hr_timeout; 1787210886Spjd modified |= MODIFIED_TIMEOUT; 1788210886Spjd } 1789211886Spjd if (strcmp(gres->hr_exec, res->hr_exec) != 0) { 1790211886Spjd strlcpy(gres->hr_exec, res->hr_exec, sizeof(gres->hr_exec)); 1791211886Spjd modified |= MODIFIED_EXEC; 1792211886Spjd } 1793210886Spjd /* 1794210886Spjd * If only timeout was modified we only need to change it without 1795210886Spjd * reconnecting. 1796210886Spjd */ 1797210886Spjd if (modified == MODIFIED_TIMEOUT) { 1798210886Spjd for (ii = 0; ii < ncomps; ii++) { 1799210886Spjd if (!ISREMOTE(ii)) 1800210886Spjd continue; 1801210886Spjd rw_rlock(&hio_remote_lock[ii]); 1802210886Spjd if (!ISCONNECTED(gres, ii)) { 1803210886Spjd rw_unlock(&hio_remote_lock[ii]); 1804210886Spjd continue; 1805210886Spjd } 1806210886Spjd rw_unlock(&hio_remote_lock[ii]); 1807210886Spjd if (proto_timeout(gres->hr_remotein, 1808210886Spjd gres->hr_timeout) < 0) { 1809210886Spjd pjdlog_errno(LOG_WARNING, 1810210886Spjd "Unable to set connection timeout"); 1811210886Spjd } 1812210886Spjd if (proto_timeout(gres->hr_remoteout, 1813210886Spjd gres->hr_timeout) < 0) { 1814210886Spjd pjdlog_errno(LOG_WARNING, 1815210886Spjd "Unable to set connection timeout"); 1816210886Spjd } 1817210886Spjd } 1818211886Spjd } else if ((modified & 1819211886Spjd (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1820210886Spjd for (ii = 0; ii < ncomps; ii++) { 1821210886Spjd if (!ISREMOTE(ii)) 1822210886Spjd continue; 1823210886Spjd remote_close(gres, ii); 1824210886Spjd } 1825210886Spjd if (modified & MODIFIED_REMOTEADDR) { 1826210886Spjd strlcpy(gres->hr_remoteaddr, res->hr_remoteaddr, 1827210886Spjd sizeof(gres->hr_remoteaddr)); 1828210886Spjd } 1829210886Spjd } 1830210886Spjd#undef MODIFIED_REMOTEADDR 1831210886Spjd#undef MODIFIED_REPLICATION 1832210886Spjd#undef MODIFIED_TIMEOUT 1833211886Spjd#undef MODIFIED_EXEC 1834210886Spjd 1835210886Spjd pjdlog_info("Configuration reloaded successfully."); 1836210886Spjd return; 1837210886Spjdfailed: 1838210886Spjd if (newcfg != NULL) { 1839210886Spjd if (newcfg->hc_controlconn != NULL) 1840210886Spjd proto_close(newcfg->hc_controlconn); 1841210886Spjd if (newcfg->hc_listenconn != NULL) 1842210886Spjd proto_close(newcfg->hc_listenconn); 1843210886Spjd yy_config_free(newcfg); 1844210886Spjd } 1845210886Spjd pjdlog_warning("Configuration not reloaded."); 1846210886Spjd} 1847210886Spjd 1848211882Spjdstatic void 1849211882Spjdkeepalive_send(struct hast_resource *res, unsigned int ncomp) 1850211882Spjd{ 1851211882Spjd struct nv *nv; 1852211882Spjd 1853211882Spjd nv = nv_alloc(); 1854211882Spjd nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1855211882Spjd if (nv_error(nv) != 0) { 1856211882Spjd nv_free(nv); 1857211882Spjd pjdlog_debug(1, 1858211882Spjd "keepalive_send: Unable to prepare header to send."); 1859211882Spjd return; 1860211882Spjd } 1861211882Spjd if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1862211882Spjd pjdlog_common(LOG_DEBUG, 1, errno, 1863211882Spjd "keepalive_send: Unable to send request"); 1864211882Spjd nv_free(nv); 1865211882Spjd rw_unlock(&hio_remote_lock[ncomp]); 1866211882Spjd remote_close(res, ncomp); 1867211882Spjd rw_rlock(&hio_remote_lock[ncomp]); 1868211882Spjd return; 1869211882Spjd } 1870211882Spjd nv_free(nv); 1871211882Spjd pjdlog_debug(2, "keepalive_send: Request sent."); 1872211882Spjd} 1873211882Spjd 1874211981Spjdstatic void 1875211981Spjdguard_one(struct hast_resource *res, unsigned int ncomp) 1876211981Spjd{ 1877211981Spjd struct proto_conn *in, *out; 1878211981Spjd 1879211981Spjd if (!ISREMOTE(ncomp)) 1880211981Spjd return; 1881211981Spjd 1882211981Spjd rw_rlock(&hio_remote_lock[ncomp]); 1883211981Spjd 1884211981Spjd if (!real_remote(res)) { 1885211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1886211981Spjd return; 1887211981Spjd } 1888211981Spjd 1889211981Spjd if (ISCONNECTED(res, ncomp)) { 1890211981Spjd assert(res->hr_remotein != NULL); 1891211981Spjd assert(res->hr_remoteout != NULL); 1892211981Spjd keepalive_send(res, ncomp); 1893211981Spjd } 1894211981Spjd 1895211981Spjd if (ISCONNECTED(res, ncomp)) { 1896211981Spjd assert(res->hr_remotein != NULL); 1897211981Spjd assert(res->hr_remoteout != NULL); 1898211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1899211981Spjd pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 1900211981Spjd res->hr_remoteaddr); 1901211981Spjd return; 1902211981Spjd } 1903211981Spjd 1904211981Spjd assert(res->hr_remotein == NULL); 1905211981Spjd assert(res->hr_remoteout == NULL); 1906211981Spjd /* 1907211981Spjd * Upgrade the lock. It doesn't have to be atomic as no other thread 1908211981Spjd * can change connection status from disconnected to connected. 1909211981Spjd */ 1910211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1911211981Spjd pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 1912211981Spjd res->hr_remoteaddr); 1913211981Spjd in = out = NULL; 1914211981Spjd if (init_remote(res, &in, &out)) { 1915211981Spjd rw_wlock(&hio_remote_lock[ncomp]); 1916211981Spjd assert(res->hr_remotein == NULL); 1917211981Spjd assert(res->hr_remoteout == NULL); 1918211981Spjd assert(in != NULL && out != NULL); 1919211981Spjd res->hr_remotein = in; 1920211981Spjd res->hr_remoteout = out; 1921211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1922211981Spjd pjdlog_info("Successfully reconnected to %s.", 1923211981Spjd res->hr_remoteaddr); 1924211981Spjd sync_start(); 1925211981Spjd } else { 1926211981Spjd /* Both connections should be NULL. */ 1927211981Spjd assert(res->hr_remotein == NULL); 1928211981Spjd assert(res->hr_remoteout == NULL); 1929211981Spjd assert(in == NULL && out == NULL); 1930211981Spjd pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 1931211981Spjd res->hr_remoteaddr); 1932211981Spjd } 1933211981Spjd} 1934211981Spjd 1935204076Spjd/* 1936204076Spjd * Thread guards remote connections and reconnects when needed, handles 1937204076Spjd * signals, etc. 1938204076Spjd */ 1939204076Spjdstatic void * 1940204076Spjdguard_thread(void *arg) 1941204076Spjd{ 1942204076Spjd struct hast_resource *res = arg; 1943204076Spjd unsigned int ii, ncomps; 1944211982Spjd struct timespec timeout; 1945211981Spjd time_t lastcheck, now; 1946211982Spjd sigset_t mask; 1947211982Spjd int signo; 1948204076Spjd 1949204076Spjd ncomps = HAST_NCOMPONENTS; 1950211981Spjd lastcheck = time(NULL); 1951204076Spjd 1952211982Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1953211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1954211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1955211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1956211982Spjd 1957211982Spjd timeout.tv_nsec = 0; 1958211982Spjd signo = -1; 1959211982Spjd 1960204076Spjd for (;;) { 1961211982Spjd switch (signo) { 1962211982Spjd case SIGHUP: 1963211982Spjd config_reload(); 1964211982Spjd break; 1965211982Spjd case SIGINT: 1966211982Spjd case SIGTERM: 1967211982Spjd sigexit_received = true; 1968204076Spjd primary_exitx(EX_OK, 1969204076Spjd "Termination signal received, exiting."); 1970211982Spjd break; 1971211982Spjd default: 1972211982Spjd break; 1973204076Spjd } 1974211882Spjd 1975204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 1976211981Spjd now = time(NULL); 1977211982Spjd if (lastcheck + RETRY_SLEEP <= now) { 1978211982Spjd for (ii = 0; ii < ncomps; ii++) 1979211981Spjd guard_one(res, ii); 1980211981Spjd lastcheck = now; 1981204076Spjd } 1982211982Spjd timeout.tv_sec = RETRY_SLEEP; 1983211982Spjd signo = sigtimedwait(&mask, NULL, &timeout); 1984204076Spjd } 1985204076Spjd /* NOTREACHED */ 1986204076Spjd return (NULL); 1987204076Spjd} 1988