primary.c revision 213007
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 213007 2010-09-22 19:03:11Z 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 261212899Spjdstatic __dead2 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 274212899Spjdstatic __dead2 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 */ 427212046Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 428212046Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 429212046Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 430212046Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 431211982Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 432204076Spjd} 433204076Spjd 434204076Spjdstatic void 435204076Spjdinit_local(struct hast_resource *res) 436204076Spjd{ 437204076Spjd unsigned char *buf; 438204076Spjd size_t mapsize; 439204076Spjd 440204076Spjd if (metadata_read(res, true) < 0) 441204076Spjd exit(EX_NOINPUT); 442204076Spjd mtx_init(&res->hr_amp_lock); 443204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 444204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 445204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 446204076Spjd } 447204076Spjd mtx_init(&range_lock); 448204076Spjd cv_init(&range_regular_cond); 449204076Spjd if (rangelock_init(&range_regular) < 0) 450204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 451204076Spjd cv_init(&range_sync_cond); 452204076Spjd if (rangelock_init(&range_sync) < 0) 453204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 454204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 455204076Spjd buf = calloc(1, mapsize); 456204076Spjd if (buf == NULL) { 457204076Spjd primary_exitx(EX_TEMPFAIL, 458204076Spjd "Unable to allocate buffer for activemap."); 459204076Spjd } 460204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 461204076Spjd (ssize_t)mapsize) { 462204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 463204076Spjd } 464204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 465209181Spjd free(buf); 466204076Spjd if (res->hr_resuid != 0) 467204076Spjd return; 468204076Spjd /* 469204076Spjd * We're using provider for the first time, so we have to generate 470204076Spjd * resource unique identifier and initialize local and remote counts. 471204076Spjd */ 472204076Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 473204076Spjd res->hr_primary_localcnt = 1; 474204076Spjd res->hr_primary_remotecnt = 0; 475204076Spjd if (metadata_write(res) < 0) 476204076Spjd exit(EX_NOINPUT); 477204076Spjd} 478204076Spjd 479205738Spjdstatic bool 480205738Spjdinit_remote(struct hast_resource *res, struct proto_conn **inp, 481205738Spjd struct proto_conn **outp) 482204076Spjd{ 483205738Spjd struct proto_conn *in, *out; 484204076Spjd struct nv *nvout, *nvin; 485204076Spjd const unsigned char *token; 486204076Spjd unsigned char *map; 487204076Spjd const char *errmsg; 488204076Spjd int32_t extentsize; 489204076Spjd int64_t datasize; 490204076Spjd uint32_t mapsize; 491204076Spjd size_t size; 492204076Spjd 493205738Spjd assert((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 494210881Spjd assert(real_remote(res)); 495205738Spjd 496205738Spjd in = out = NULL; 497211983Spjd errmsg = NULL; 498205738Spjd 499204076Spjd /* Prepare outgoing connection with remote node. */ 500205738Spjd if (proto_client(res->hr_remoteaddr, &out) < 0) { 501207347Spjd primary_exit(EX_TEMPFAIL, "Unable to create connection to %s", 502204076Spjd res->hr_remoteaddr); 503204076Spjd } 504204076Spjd /* Try to connect, but accept failure. */ 505205738Spjd if (proto_connect(out) < 0) { 506204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 507204076Spjd res->hr_remoteaddr); 508204076Spjd goto close; 509204076Spjd } 510207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 511207371Spjd if (proto_timeout(out, res->hr_timeout) < 0) 512207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 513204076Spjd /* 514204076Spjd * First handshake step. 515204076Spjd * Setup outgoing connection with remote node. 516204076Spjd */ 517204076Spjd nvout = nv_alloc(); 518204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 519204076Spjd if (nv_error(nvout) != 0) { 520204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 521204076Spjd "Unable to allocate header for connection with %s", 522204076Spjd res->hr_remoteaddr); 523204076Spjd nv_free(nvout); 524204076Spjd goto close; 525204076Spjd } 526205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 527204076Spjd pjdlog_errno(LOG_WARNING, 528204076Spjd "Unable to send handshake header to %s", 529204076Spjd res->hr_remoteaddr); 530204076Spjd nv_free(nvout); 531204076Spjd goto close; 532204076Spjd } 533204076Spjd nv_free(nvout); 534205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 535204076Spjd pjdlog_errno(LOG_WARNING, 536204076Spjd "Unable to receive handshake header from %s", 537204076Spjd res->hr_remoteaddr); 538204076Spjd goto close; 539204076Spjd } 540204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 541204076Spjd if (errmsg != NULL) { 542204076Spjd pjdlog_warning("%s", errmsg); 543204076Spjd nv_free(nvin); 544204076Spjd goto close; 545204076Spjd } 546204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 547204076Spjd if (token == NULL) { 548204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 549204076Spjd res->hr_remoteaddr); 550204076Spjd nv_free(nvin); 551204076Spjd goto close; 552204076Spjd } 553204076Spjd if (size != sizeof(res->hr_token)) { 554204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 555204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 556204076Spjd nv_free(nvin); 557204076Spjd goto close; 558204076Spjd } 559204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 560204076Spjd nv_free(nvin); 561204076Spjd 562204076Spjd /* 563204076Spjd * Second handshake step. 564204076Spjd * Setup incoming connection with remote node. 565204076Spjd */ 566205738Spjd if (proto_client(res->hr_remoteaddr, &in) < 0) { 567204076Spjd pjdlog_errno(LOG_WARNING, "Unable to create connection to %s", 568204076Spjd res->hr_remoteaddr); 569204076Spjd } 570204076Spjd /* Try to connect, but accept failure. */ 571205738Spjd if (proto_connect(in) < 0) { 572204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 573204076Spjd res->hr_remoteaddr); 574204076Spjd goto close; 575204076Spjd } 576207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 577207371Spjd if (proto_timeout(in, res->hr_timeout) < 0) 578207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 579204076Spjd nvout = nv_alloc(); 580204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 581204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 582204076Spjd "token"); 583204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 584204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 585204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 586204076Spjd if (nv_error(nvout) != 0) { 587204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 588204076Spjd "Unable to allocate header for connection with %s", 589204076Spjd res->hr_remoteaddr); 590204076Spjd nv_free(nvout); 591204076Spjd goto close; 592204076Spjd } 593205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 594204076Spjd pjdlog_errno(LOG_WARNING, 595204076Spjd "Unable to send handshake header to %s", 596204076Spjd res->hr_remoteaddr); 597204076Spjd nv_free(nvout); 598204076Spjd goto close; 599204076Spjd } 600204076Spjd nv_free(nvout); 601205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 602204076Spjd pjdlog_errno(LOG_WARNING, 603204076Spjd "Unable to receive handshake header from %s", 604204076Spjd res->hr_remoteaddr); 605204076Spjd goto close; 606204076Spjd } 607204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 608204076Spjd if (errmsg != NULL) { 609204076Spjd pjdlog_warning("%s", errmsg); 610204076Spjd nv_free(nvin); 611204076Spjd goto close; 612204076Spjd } 613204076Spjd datasize = nv_get_int64(nvin, "datasize"); 614204076Spjd if (datasize != res->hr_datasize) { 615204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 616204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 617204076Spjd nv_free(nvin); 618204076Spjd goto close; 619204076Spjd } 620204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 621204076Spjd if (extentsize != res->hr_extentsize) { 622204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 623204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 624204076Spjd nv_free(nvin); 625204076Spjd goto close; 626204076Spjd } 627204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 628204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 629204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 630204076Spjd map = NULL; 631204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 632204076Spjd if (mapsize > 0) { 633204076Spjd map = malloc(mapsize); 634204076Spjd if (map == NULL) { 635204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 636204076Spjd (uintmax_t)mapsize); 637204076Spjd nv_free(nvin); 638204076Spjd goto close; 639204076Spjd } 640204076Spjd /* 641204076Spjd * Remote node have some dirty extents on its own, lets 642204076Spjd * download its activemap. 643204076Spjd */ 644205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 645204076Spjd mapsize) < 0) { 646204076Spjd pjdlog_errno(LOG_ERR, 647204076Spjd "Unable to receive remote activemap"); 648204076Spjd nv_free(nvin); 649204076Spjd free(map); 650204076Spjd goto close; 651204076Spjd } 652204076Spjd /* 653204076Spjd * Merge local and remote bitmaps. 654204076Spjd */ 655204076Spjd activemap_merge(res->hr_amp, map, mapsize); 656204076Spjd free(map); 657204076Spjd /* 658204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 659204076Spjd * disk before we start to synchronize. 660204076Spjd */ 661204076Spjd (void)hast_activemap_flush(res); 662204076Spjd } 663204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 664205738Spjd if (inp != NULL && outp != NULL) { 665205738Spjd *inp = in; 666205738Spjd *outp = out; 667205738Spjd } else { 668205738Spjd res->hr_remotein = in; 669205738Spjd res->hr_remoteout = out; 670205738Spjd } 671212038Spjd event_send(res, EVENT_CONNECT); 672205738Spjd return (true); 673205738Spjdclose: 674211983Spjd if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0) 675212038Spjd event_send(res, EVENT_SPLITBRAIN); 676205738Spjd proto_close(out); 677205738Spjd if (in != NULL) 678205738Spjd proto_close(in); 679205738Spjd return (false); 680205738Spjd} 681205738Spjd 682205738Spjdstatic void 683205738Spjdsync_start(void) 684205738Spjd{ 685205738Spjd 686204076Spjd mtx_lock(&sync_lock); 687204076Spjd sync_inprogress = true; 688204076Spjd mtx_unlock(&sync_lock); 689204076Spjd cv_signal(&sync_cond); 690204076Spjd} 691204076Spjd 692204076Spjdstatic void 693211878Spjdsync_stop(void) 694211878Spjd{ 695211878Spjd 696211878Spjd mtx_lock(&sync_lock); 697211878Spjd if (sync_inprogress) 698211878Spjd sync_inprogress = false; 699211878Spjd mtx_unlock(&sync_lock); 700211878Spjd} 701211878Spjd 702211878Spjdstatic void 703204076Spjdinit_ggate(struct hast_resource *res) 704204076Spjd{ 705204076Spjd struct g_gate_ctl_create ggiocreate; 706204076Spjd struct g_gate_ctl_cancel ggiocancel; 707204076Spjd 708204076Spjd /* 709204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 710204076Spjd */ 711204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 712204076Spjd if (res->hr_ggatefd < 0) 713204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 714204076Spjd /* 715204076Spjd * Create provider before trying to connect, as connection failure 716204076Spjd * is not critical, but may take some time. 717204076Spjd */ 718204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 719204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 720204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 721204076Spjd ggiocreate.gctl_flags = 0; 722206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 723204076Spjd ggiocreate.gctl_timeout = 0; 724204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 725204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 726204076Spjd res->hr_provname); 727204076Spjd bzero(ggiocreate.gctl_info, sizeof(ggiocreate.gctl_info)); 728204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 729204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 730204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 731204076Spjd return; 732204076Spjd } 733204076Spjd if (errno != EEXIST) { 734204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 735204076Spjd res->hr_provname); 736204076Spjd } 737204076Spjd pjdlog_debug(1, 738204076Spjd "Device hast/%s already exists, we will try to take it over.", 739204076Spjd res->hr_provname); 740204076Spjd /* 741204076Spjd * If we received EEXIST, we assume that the process who created the 742204076Spjd * provider died and didn't clean up. In that case we will start from 743204076Spjd * where he left of. 744204076Spjd */ 745204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 746204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 747204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 748204076Spjd res->hr_provname); 749204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 750204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 751204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 752204076Spjd return; 753204076Spjd } 754204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 755204076Spjd res->hr_provname); 756204076Spjd} 757204076Spjd 758204076Spjdvoid 759204076Spjdhastd_primary(struct hast_resource *res) 760204076Spjd{ 761204076Spjd pthread_t td; 762204076Spjd pid_t pid; 763204076Spjd int error; 764204076Spjd 765204076Spjd /* 766204076Spjd * Create communication channel between parent and child. 767204076Spjd */ 768204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 769204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 770212034Spjd pjdlog_exit(EX_OSERR, 771204076Spjd "Unable to create control sockets between parent and child"); 772204076Spjd } 773212038Spjd /* 774212038Spjd * Create communication channel between child and parent. 775212038Spjd */ 776212038Spjd if (proto_client("socketpair://", &res->hr_event) < 0) { 777212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 778212038Spjd pjdlog_exit(EX_OSERR, 779212038Spjd "Unable to create event sockets between child and parent"); 780212038Spjd } 781204076Spjd 782204076Spjd pid = fork(); 783204076Spjd if (pid < 0) { 784204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 785212034Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to fork"); 786204076Spjd } 787204076Spjd 788204076Spjd if (pid > 0) { 789204076Spjd /* This is parent. */ 790212038Spjd /* Declare that we are receiver. */ 791212038Spjd proto_recv(res->hr_event, NULL, 0); 792204076Spjd res->hr_workerpid = pid; 793204076Spjd return; 794204076Spjd } 795211977Spjd 796211984Spjd gres = res; 797211984Spjd 798204076Spjd (void)pidfile_close(pfh); 799211977Spjd hook_fini(); 800204076Spjd 801204076Spjd setproctitle("%s (primary)", res->hr_name); 802204076Spjd 803210880Spjd signal(SIGHUP, SIG_DFL); 804210880Spjd signal(SIGCHLD, SIG_DFL); 805210880Spjd 806212038Spjd /* Declare that we are sender. */ 807212038Spjd proto_send(res->hr_event, NULL, 0); 808212038Spjd 809204076Spjd init_local(res); 810213007Spjd init_ggate(res); 811213007Spjd init_environment(res); 812213007Spjd /* 813213007Spjd * Create the control thread before sending any event to the parent, 814213007Spjd * as we can deadlock when parent sends control request to worker, 815213007Spjd * but worker has no control thread started yet, so parent waits. 816213007Spjd * In the meantime worker sends an event to the parent, but parent 817213007Spjd * is unable to handle the event, because it waits for control 818213007Spjd * request response. 819213007Spjd */ 820213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 821213007Spjd assert(error == 0); 822210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 823205738Spjd sync_start(); 824204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 825204076Spjd assert(error == 0); 826204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 827204076Spjd assert(error == 0); 828204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 829204076Spjd assert(error == 0); 830204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 831204076Spjd assert(error == 0); 832204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 833204076Spjd assert(error == 0); 834204076Spjd error = pthread_create(&td, NULL, sync_thread, res); 835204076Spjd assert(error == 0); 836204076Spjd (void)guard_thread(res); 837204076Spjd} 838204076Spjd 839204076Spjdstatic void 840204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 841204076Spjd{ 842204076Spjd char msg[1024]; 843204076Spjd va_list ap; 844204076Spjd int len; 845204076Spjd 846204076Spjd va_start(ap, fmt); 847204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 848204076Spjd va_end(ap); 849204076Spjd if ((size_t)len < sizeof(msg)) { 850204076Spjd switch (ggio->gctl_cmd) { 851204076Spjd case BIO_READ: 852204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 853204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 854204076Spjd (uintmax_t)ggio->gctl_length); 855204076Spjd break; 856204076Spjd case BIO_DELETE: 857204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 858204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 859204076Spjd (uintmax_t)ggio->gctl_length); 860204076Spjd break; 861204076Spjd case BIO_FLUSH: 862204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 863204076Spjd break; 864204076Spjd case BIO_WRITE: 865204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 866204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 867204076Spjd (uintmax_t)ggio->gctl_length); 868204076Spjd break; 869204076Spjd default: 870204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 871204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 872204076Spjd break; 873204076Spjd } 874204076Spjd } 875204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 876204076Spjd} 877204076Spjd 878204076Spjdstatic void 879204076Spjdremote_close(struct hast_resource *res, int ncomp) 880204076Spjd{ 881204076Spjd 882204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 883204076Spjd /* 884204076Spjd * A race is possible between dropping rlock and acquiring wlock - 885204076Spjd * another thread can close connection in-between. 886204076Spjd */ 887204076Spjd if (!ISCONNECTED(res, ncomp)) { 888204076Spjd assert(res->hr_remotein == NULL); 889204076Spjd assert(res->hr_remoteout == NULL); 890204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 891204076Spjd return; 892204076Spjd } 893204076Spjd 894204076Spjd assert(res->hr_remotein != NULL); 895204076Spjd assert(res->hr_remoteout != NULL); 896204076Spjd 897211881Spjd pjdlog_debug(2, "Closing incoming connection to %s.", 898204076Spjd res->hr_remoteaddr); 899204076Spjd proto_close(res->hr_remotein); 900204076Spjd res->hr_remotein = NULL; 901211881Spjd pjdlog_debug(2, "Closing outgoing connection to %s.", 902204076Spjd res->hr_remoteaddr); 903204076Spjd proto_close(res->hr_remoteout); 904204076Spjd res->hr_remoteout = NULL; 905204076Spjd 906204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 907204076Spjd 908211881Spjd pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 909211881Spjd 910204076Spjd /* 911204076Spjd * Stop synchronization if in-progress. 912204076Spjd */ 913211878Spjd sync_stop(); 914211984Spjd 915212038Spjd event_send(res, EVENT_DISCONNECT); 916204076Spjd} 917204076Spjd 918204076Spjd/* 919204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 920204076Spjd * appropriate threads: 921204076Spjd * WRITE - always goes to both local_send and remote_send threads 922204076Spjd * READ (when the block is up-to-date on local component) - 923204076Spjd * only local_send thread 924204076Spjd * READ (when the block isn't up-to-date on local component) - 925204076Spjd * only remote_send thread 926204076Spjd * DELETE - always goes to both local_send and remote_send threads 927204076Spjd * FLUSH - always goes to both local_send and remote_send threads 928204076Spjd */ 929204076Spjdstatic void * 930204076Spjdggate_recv_thread(void *arg) 931204076Spjd{ 932204076Spjd struct hast_resource *res = arg; 933204076Spjd struct g_gate_ctl_io *ggio; 934204076Spjd struct hio *hio; 935204076Spjd unsigned int ii, ncomp, ncomps; 936204076Spjd int error; 937204076Spjd 938204076Spjd ncomps = HAST_NCOMPONENTS; 939204076Spjd 940204076Spjd for (;;) { 941204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 942204076Spjd QUEUE_TAKE2(hio, free); 943204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 944204076Spjd ggio = &hio->hio_ggio; 945204076Spjd ggio->gctl_unit = res->hr_ggateunit; 946204076Spjd ggio->gctl_length = MAXPHYS; 947204076Spjd ggio->gctl_error = 0; 948204076Spjd pjdlog_debug(2, 949204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 950204076Spjd hio); 951204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 952204076Spjd if (sigexit_received) 953204076Spjd pthread_exit(NULL); 954204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 955204076Spjd } 956204076Spjd error = ggio->gctl_error; 957204076Spjd switch (error) { 958204076Spjd case 0: 959204076Spjd break; 960204076Spjd case ECANCELED: 961204076Spjd /* Exit gracefully. */ 962204076Spjd if (!sigexit_received) { 963204076Spjd pjdlog_debug(2, 964204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 965204076Spjd hio); 966204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 967204076Spjd } 968204076Spjd pthread_exit(NULL); 969204076Spjd case ENOMEM: 970204076Spjd /* 971204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 972204076Spjd * bytes - request can't be bigger than that. 973204076Spjd */ 974204076Spjd /* FALLTHROUGH */ 975204076Spjd case ENXIO: 976204076Spjd default: 977204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 978204076Spjd strerror(error)); 979204076Spjd } 980204076Spjd for (ii = 0; ii < ncomps; ii++) 981204076Spjd hio->hio_errors[ii] = EINVAL; 982204076Spjd reqlog(LOG_DEBUG, 2, ggio, 983204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 984204076Spjd hio); 985204076Spjd /* 986204076Spjd * Inform all components about new write request. 987204076Spjd * For read request prefer local component unless the given 988204076Spjd * range is out-of-date, then use remote component. 989204076Spjd */ 990204076Spjd switch (ggio->gctl_cmd) { 991204076Spjd case BIO_READ: 992204076Spjd pjdlog_debug(2, 993204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 994204076Spjd hio); 995204076Spjd refcount_init(&hio->hio_countdown, 1); 996204076Spjd mtx_lock(&metadata_lock); 997204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 998204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 999204076Spjd /* 1000204076Spjd * This range is up-to-date on local component, 1001204076Spjd * so handle request locally. 1002204076Spjd */ 1003204076Spjd /* Local component is 0 for now. */ 1004204076Spjd ncomp = 0; 1005204076Spjd } else /* if (res->hr_syncsrc == 1006204076Spjd HAST_SYNCSRC_SECONDARY) */ { 1007204076Spjd assert(res->hr_syncsrc == 1008204076Spjd HAST_SYNCSRC_SECONDARY); 1009204076Spjd /* 1010204076Spjd * This range is out-of-date on local component, 1011204076Spjd * so send request to the remote node. 1012204076Spjd */ 1013204076Spjd /* Remote component is 1 for now. */ 1014204076Spjd ncomp = 1; 1015204076Spjd } 1016204076Spjd mtx_unlock(&metadata_lock); 1017204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1018204076Spjd break; 1019204076Spjd case BIO_WRITE: 1020204076Spjd for (;;) { 1021204076Spjd mtx_lock(&range_lock); 1022204076Spjd if (rangelock_islocked(range_sync, 1023204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1024204076Spjd pjdlog_debug(2, 1025204076Spjd "regular: Range offset=%jd length=%zu locked.", 1026204076Spjd (intmax_t)ggio->gctl_offset, 1027204076Spjd (size_t)ggio->gctl_length); 1028204076Spjd range_regular_wait = true; 1029204076Spjd cv_wait(&range_regular_cond, &range_lock); 1030204076Spjd range_regular_wait = false; 1031204076Spjd mtx_unlock(&range_lock); 1032204076Spjd continue; 1033204076Spjd } 1034204076Spjd if (rangelock_add(range_regular, 1035204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1036204076Spjd mtx_unlock(&range_lock); 1037204076Spjd pjdlog_debug(2, 1038204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1039204076Spjd (intmax_t)ggio->gctl_offset, 1040204076Spjd (size_t)ggio->gctl_length); 1041204076Spjd sleep(1); 1042204076Spjd continue; 1043204076Spjd } 1044204076Spjd mtx_unlock(&range_lock); 1045204076Spjd break; 1046204076Spjd } 1047204076Spjd mtx_lock(&res->hr_amp_lock); 1048204076Spjd if (activemap_write_start(res->hr_amp, 1049204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1050204076Spjd (void)hast_activemap_flush(res); 1051204076Spjd } 1052204076Spjd mtx_unlock(&res->hr_amp_lock); 1053204076Spjd /* FALLTHROUGH */ 1054204076Spjd case BIO_DELETE: 1055204076Spjd case BIO_FLUSH: 1056204076Spjd pjdlog_debug(2, 1057204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1058204076Spjd hio); 1059204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1060204076Spjd for (ii = 0; ii < ncomps; ii++) 1061204076Spjd QUEUE_INSERT1(hio, send, ii); 1062204076Spjd break; 1063204076Spjd } 1064204076Spjd } 1065204076Spjd /* NOTREACHED */ 1066204076Spjd return (NULL); 1067204076Spjd} 1068204076Spjd 1069204076Spjd/* 1070204076Spjd * Thread reads from or writes to local component. 1071204076Spjd * If local read fails, it redirects it to remote_send thread. 1072204076Spjd */ 1073204076Spjdstatic void * 1074204076Spjdlocal_send_thread(void *arg) 1075204076Spjd{ 1076204076Spjd struct hast_resource *res = arg; 1077204076Spjd struct g_gate_ctl_io *ggio; 1078204076Spjd struct hio *hio; 1079204076Spjd unsigned int ncomp, rncomp; 1080204076Spjd ssize_t ret; 1081204076Spjd 1082204076Spjd /* Local component is 0 for now. */ 1083204076Spjd ncomp = 0; 1084204076Spjd /* Remote component is 1 for now. */ 1085204076Spjd rncomp = 1; 1086204076Spjd 1087204076Spjd for (;;) { 1088204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1089204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1090204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1091204076Spjd ggio = &hio->hio_ggio; 1092204076Spjd switch (ggio->gctl_cmd) { 1093204076Spjd case BIO_READ: 1094204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1095204076Spjd ggio->gctl_length, 1096204076Spjd ggio->gctl_offset + res->hr_localoff); 1097204076Spjd if (ret == ggio->gctl_length) 1098204076Spjd hio->hio_errors[ncomp] = 0; 1099204076Spjd else { 1100204076Spjd /* 1101204076Spjd * If READ failed, try to read from remote node. 1102204076Spjd */ 1103204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1104204076Spjd continue; 1105204076Spjd } 1106204076Spjd break; 1107204076Spjd case BIO_WRITE: 1108204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1109204076Spjd ggio->gctl_length, 1110204076Spjd ggio->gctl_offset + res->hr_localoff); 1111204076Spjd if (ret < 0) 1112204076Spjd hio->hio_errors[ncomp] = errno; 1113204076Spjd else if (ret != ggio->gctl_length) 1114204076Spjd hio->hio_errors[ncomp] = EIO; 1115204076Spjd else 1116204076Spjd hio->hio_errors[ncomp] = 0; 1117204076Spjd break; 1118204076Spjd case BIO_DELETE: 1119204076Spjd ret = g_delete(res->hr_localfd, 1120204076Spjd ggio->gctl_offset + res->hr_localoff, 1121204076Spjd ggio->gctl_length); 1122204076Spjd if (ret < 0) 1123204076Spjd hio->hio_errors[ncomp] = errno; 1124204076Spjd else 1125204076Spjd hio->hio_errors[ncomp] = 0; 1126204076Spjd break; 1127204076Spjd case BIO_FLUSH: 1128204076Spjd ret = g_flush(res->hr_localfd); 1129204076Spjd if (ret < 0) 1130204076Spjd hio->hio_errors[ncomp] = errno; 1131204076Spjd else 1132204076Spjd hio->hio_errors[ncomp] = 0; 1133204076Spjd break; 1134204076Spjd } 1135204076Spjd if (refcount_release(&hio->hio_countdown)) { 1136204076Spjd if (ISSYNCREQ(hio)) { 1137204076Spjd mtx_lock(&sync_lock); 1138204076Spjd SYNCREQDONE(hio); 1139204076Spjd mtx_unlock(&sync_lock); 1140204076Spjd cv_signal(&sync_cond); 1141204076Spjd } else { 1142204076Spjd pjdlog_debug(2, 1143204076Spjd "local_send: (%p) Moving request to the done queue.", 1144204076Spjd hio); 1145204076Spjd QUEUE_INSERT2(hio, done); 1146204076Spjd } 1147204076Spjd } 1148204076Spjd } 1149204076Spjd /* NOTREACHED */ 1150204076Spjd return (NULL); 1151204076Spjd} 1152204076Spjd 1153204076Spjd/* 1154204076Spjd * Thread sends request to secondary node. 1155204076Spjd */ 1156204076Spjdstatic void * 1157204076Spjdremote_send_thread(void *arg) 1158204076Spjd{ 1159204076Spjd struct hast_resource *res = arg; 1160204076Spjd struct g_gate_ctl_io *ggio; 1161204076Spjd struct hio *hio; 1162204076Spjd struct nv *nv; 1163204076Spjd unsigned int ncomp; 1164204076Spjd bool wakeup; 1165204076Spjd uint64_t offset, length; 1166204076Spjd uint8_t cmd; 1167204076Spjd void *data; 1168204076Spjd 1169204076Spjd /* Remote component is 1 for now. */ 1170204076Spjd ncomp = 1; 1171204076Spjd 1172204076Spjd for (;;) { 1173204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1174204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1175204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1176204076Spjd ggio = &hio->hio_ggio; 1177204076Spjd switch (ggio->gctl_cmd) { 1178204076Spjd case BIO_READ: 1179204076Spjd cmd = HIO_READ; 1180204076Spjd data = NULL; 1181204076Spjd offset = ggio->gctl_offset; 1182204076Spjd length = ggio->gctl_length; 1183204076Spjd break; 1184204076Spjd case BIO_WRITE: 1185204076Spjd cmd = HIO_WRITE; 1186204076Spjd data = ggio->gctl_data; 1187204076Spjd offset = ggio->gctl_offset; 1188204076Spjd length = ggio->gctl_length; 1189204076Spjd break; 1190204076Spjd case BIO_DELETE: 1191204076Spjd cmd = HIO_DELETE; 1192204076Spjd data = NULL; 1193204076Spjd offset = ggio->gctl_offset; 1194204076Spjd length = ggio->gctl_length; 1195204076Spjd break; 1196204076Spjd case BIO_FLUSH: 1197204076Spjd cmd = HIO_FLUSH; 1198204076Spjd data = NULL; 1199204076Spjd offset = 0; 1200204076Spjd length = 0; 1201204076Spjd break; 1202204076Spjd default: 1203204076Spjd assert(!"invalid condition"); 1204204076Spjd abort(); 1205204076Spjd } 1206204076Spjd nv = nv_alloc(); 1207204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1208204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1209204076Spjd nv_add_uint64(nv, offset, "offset"); 1210204076Spjd nv_add_uint64(nv, length, "length"); 1211204076Spjd if (nv_error(nv) != 0) { 1212204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1213204076Spjd pjdlog_debug(2, 1214204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1215204076Spjd hio); 1216204076Spjd reqlog(LOG_ERR, 0, ggio, 1217204076Spjd "Unable to prepare header to send (%s): ", 1218204076Spjd strerror(nv_error(nv))); 1219204076Spjd /* Move failed request immediately to the done queue. */ 1220204076Spjd goto done_queue; 1221204076Spjd } 1222204076Spjd pjdlog_debug(2, 1223204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1224204076Spjd hio); 1225204076Spjd /* 1226204076Spjd * Protect connection from disappearing. 1227204076Spjd */ 1228204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1229204076Spjd if (!ISCONNECTED(res, ncomp)) { 1230204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1231204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1232204076Spjd goto done_queue; 1233204076Spjd } 1234204076Spjd /* 1235204076Spjd * Move the request to recv queue before sending it, because 1236204076Spjd * in different order we can get reply before we move request 1237204076Spjd * to recv queue. 1238204076Spjd */ 1239204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1240204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1241204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1242204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1243204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1244204076Spjd data != NULL ? length : 0) < 0) { 1245204076Spjd hio->hio_errors[ncomp] = errno; 1246204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1247204076Spjd pjdlog_debug(2, 1248204076Spjd "remote_send: (%p) Unable to send request.", hio); 1249204076Spjd reqlog(LOG_ERR, 0, ggio, 1250204076Spjd "Unable to send request (%s): ", 1251204076Spjd strerror(hio->hio_errors[ncomp])); 1252211979Spjd remote_close(res, ncomp); 1253204076Spjd /* 1254204076Spjd * Take request back from the receive queue and move 1255204076Spjd * it immediately to the done queue. 1256204076Spjd */ 1257204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1258204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1259204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1260204076Spjd goto done_queue; 1261204076Spjd } 1262204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1263204076Spjd nv_free(nv); 1264204076Spjd if (wakeup) 1265204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1266204076Spjd continue; 1267204076Spjddone_queue: 1268204076Spjd nv_free(nv); 1269204076Spjd if (ISSYNCREQ(hio)) { 1270204076Spjd if (!refcount_release(&hio->hio_countdown)) 1271204076Spjd continue; 1272204076Spjd mtx_lock(&sync_lock); 1273204076Spjd SYNCREQDONE(hio); 1274204076Spjd mtx_unlock(&sync_lock); 1275204076Spjd cv_signal(&sync_cond); 1276204076Spjd continue; 1277204076Spjd } 1278204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1279204076Spjd mtx_lock(&res->hr_amp_lock); 1280204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1281204076Spjd ggio->gctl_length)) { 1282204076Spjd (void)hast_activemap_flush(res); 1283204076Spjd } 1284204076Spjd mtx_unlock(&res->hr_amp_lock); 1285204076Spjd } 1286204076Spjd if (!refcount_release(&hio->hio_countdown)) 1287204076Spjd continue; 1288204076Spjd pjdlog_debug(2, 1289204076Spjd "remote_send: (%p) Moving request to the done queue.", 1290204076Spjd hio); 1291204076Spjd QUEUE_INSERT2(hio, done); 1292204076Spjd } 1293204076Spjd /* NOTREACHED */ 1294204076Spjd return (NULL); 1295204076Spjd} 1296204076Spjd 1297204076Spjd/* 1298204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1299204076Spjd * thread. 1300204076Spjd */ 1301204076Spjdstatic void * 1302204076Spjdremote_recv_thread(void *arg) 1303204076Spjd{ 1304204076Spjd struct hast_resource *res = arg; 1305204076Spjd struct g_gate_ctl_io *ggio; 1306204076Spjd struct hio *hio; 1307204076Spjd struct nv *nv; 1308204076Spjd unsigned int ncomp; 1309204076Spjd uint64_t seq; 1310204076Spjd int error; 1311204076Spjd 1312204076Spjd /* Remote component is 1 for now. */ 1313204076Spjd ncomp = 1; 1314204076Spjd 1315204076Spjd for (;;) { 1316204076Spjd /* Wait until there is anything to receive. */ 1317204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1318204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1319204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1320204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1321204076Spjd &hio_recv_list_lock[ncomp]); 1322204076Spjd } 1323204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1324204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1325204076Spjd if (!ISCONNECTED(res, ncomp)) { 1326204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1327204076Spjd /* 1328204076Spjd * Connection is dead, so move all pending requests to 1329204076Spjd * the done queue (one-by-one). 1330204076Spjd */ 1331204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1332204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1333204076Spjd assert(hio != NULL); 1334204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1335204076Spjd hio_next[ncomp]); 1336204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1337204076Spjd goto done_queue; 1338204076Spjd } 1339204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1340204076Spjd pjdlog_errno(LOG_ERR, 1341204076Spjd "Unable to receive reply header"); 1342204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1343204076Spjd remote_close(res, ncomp); 1344204076Spjd continue; 1345204076Spjd } 1346204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1347204076Spjd seq = nv_get_uint64(nv, "seq"); 1348204076Spjd if (seq == 0) { 1349204076Spjd pjdlog_error("Header contains no 'seq' field."); 1350204076Spjd nv_free(nv); 1351204076Spjd continue; 1352204076Spjd } 1353204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1354204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1355204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1356204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1357204076Spjd hio_next[ncomp]); 1358204076Spjd break; 1359204076Spjd } 1360204076Spjd } 1361204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1362204076Spjd if (hio == NULL) { 1363204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1364204076Spjd (uintmax_t)seq); 1365204076Spjd nv_free(nv); 1366204076Spjd continue; 1367204076Spjd } 1368204076Spjd error = nv_get_int16(nv, "error"); 1369204076Spjd if (error != 0) { 1370204076Spjd /* Request failed on remote side. */ 1371204076Spjd hio->hio_errors[ncomp] = 0; 1372204076Spjd nv_free(nv); 1373204076Spjd goto done_queue; 1374204076Spjd } 1375204076Spjd ggio = &hio->hio_ggio; 1376204076Spjd switch (ggio->gctl_cmd) { 1377204076Spjd case BIO_READ: 1378204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1379204076Spjd if (!ISCONNECTED(res, ncomp)) { 1380204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1381204076Spjd nv_free(nv); 1382204076Spjd goto done_queue; 1383204076Spjd } 1384204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1385204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1386204076Spjd hio->hio_errors[ncomp] = errno; 1387204076Spjd pjdlog_errno(LOG_ERR, 1388204076Spjd "Unable to receive reply data"); 1389204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1390204076Spjd nv_free(nv); 1391204076Spjd remote_close(res, ncomp); 1392204076Spjd goto done_queue; 1393204076Spjd } 1394204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1395204076Spjd break; 1396204076Spjd case BIO_WRITE: 1397204076Spjd case BIO_DELETE: 1398204076Spjd case BIO_FLUSH: 1399204076Spjd break; 1400204076Spjd default: 1401204076Spjd assert(!"invalid condition"); 1402204076Spjd abort(); 1403204076Spjd } 1404204076Spjd hio->hio_errors[ncomp] = 0; 1405204076Spjd nv_free(nv); 1406204076Spjddone_queue: 1407204076Spjd if (refcount_release(&hio->hio_countdown)) { 1408204076Spjd if (ISSYNCREQ(hio)) { 1409204076Spjd mtx_lock(&sync_lock); 1410204076Spjd SYNCREQDONE(hio); 1411204076Spjd mtx_unlock(&sync_lock); 1412204076Spjd cv_signal(&sync_cond); 1413204076Spjd } else { 1414204076Spjd pjdlog_debug(2, 1415204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1416204076Spjd hio); 1417204076Spjd QUEUE_INSERT2(hio, done); 1418204076Spjd } 1419204076Spjd } 1420204076Spjd } 1421204076Spjd /* NOTREACHED */ 1422204076Spjd return (NULL); 1423204076Spjd} 1424204076Spjd 1425204076Spjd/* 1426204076Spjd * Thread sends answer to the kernel. 1427204076Spjd */ 1428204076Spjdstatic void * 1429204076Spjdggate_send_thread(void *arg) 1430204076Spjd{ 1431204076Spjd struct hast_resource *res = arg; 1432204076Spjd struct g_gate_ctl_io *ggio; 1433204076Spjd struct hio *hio; 1434204076Spjd unsigned int ii, ncomp, ncomps; 1435204076Spjd 1436204076Spjd ncomps = HAST_NCOMPONENTS; 1437204076Spjd 1438204076Spjd for (;;) { 1439204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1440204076Spjd QUEUE_TAKE2(hio, done); 1441204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1442204076Spjd ggio = &hio->hio_ggio; 1443204076Spjd for (ii = 0; ii < ncomps; ii++) { 1444204076Spjd if (hio->hio_errors[ii] == 0) { 1445204076Spjd /* 1446204076Spjd * One successful request is enough to declare 1447204076Spjd * success. 1448204076Spjd */ 1449204076Spjd ggio->gctl_error = 0; 1450204076Spjd break; 1451204076Spjd } 1452204076Spjd } 1453204076Spjd if (ii == ncomps) { 1454204076Spjd /* 1455204076Spjd * None of the requests were successful. 1456204076Spjd * Use first error. 1457204076Spjd */ 1458204076Spjd ggio->gctl_error = hio->hio_errors[0]; 1459204076Spjd } 1460204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1461204076Spjd mtx_lock(&res->hr_amp_lock); 1462204076Spjd activemap_write_complete(res->hr_amp, 1463204076Spjd ggio->gctl_offset, ggio->gctl_length); 1464204076Spjd mtx_unlock(&res->hr_amp_lock); 1465204076Spjd } 1466204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1467204076Spjd /* 1468204076Spjd * Unlock range we locked. 1469204076Spjd */ 1470204076Spjd mtx_lock(&range_lock); 1471204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1472204076Spjd ggio->gctl_length); 1473204076Spjd if (range_sync_wait) 1474204076Spjd cv_signal(&range_sync_cond); 1475204076Spjd mtx_unlock(&range_lock); 1476204076Spjd /* 1477204076Spjd * Bump local count if this is first write after 1478204076Spjd * connection failure with remote node. 1479204076Spjd */ 1480204076Spjd ncomp = 1; 1481204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1482204076Spjd if (!ISCONNECTED(res, ncomp)) { 1483204076Spjd mtx_lock(&metadata_lock); 1484204076Spjd if (res->hr_primary_localcnt == 1485204076Spjd res->hr_secondary_remotecnt) { 1486204076Spjd res->hr_primary_localcnt++; 1487204076Spjd pjdlog_debug(1, 1488204076Spjd "Increasing localcnt to %ju.", 1489204076Spjd (uintmax_t)res->hr_primary_localcnt); 1490204076Spjd (void)metadata_write(res); 1491204076Spjd } 1492204076Spjd mtx_unlock(&metadata_lock); 1493204076Spjd } 1494204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1495204076Spjd } 1496204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1497204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1498204076Spjd pjdlog_debug(2, 1499204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1500204076Spjd QUEUE_INSERT2(hio, free); 1501204076Spjd } 1502204076Spjd /* NOTREACHED */ 1503204076Spjd return (NULL); 1504204076Spjd} 1505204076Spjd 1506204076Spjd/* 1507204076Spjd * Thread synchronize local and remote components. 1508204076Spjd */ 1509204076Spjdstatic void * 1510204076Spjdsync_thread(void *arg __unused) 1511204076Spjd{ 1512204076Spjd struct hast_resource *res = arg; 1513204076Spjd struct hio *hio; 1514204076Spjd struct g_gate_ctl_io *ggio; 1515204076Spjd unsigned int ii, ncomp, ncomps; 1516204076Spjd off_t offset, length, synced; 1517204076Spjd bool dorewind; 1518204076Spjd int syncext; 1519204076Spjd 1520204076Spjd ncomps = HAST_NCOMPONENTS; 1521204076Spjd dorewind = true; 1522211897Spjd synced = 0; 1523211897Spjd offset = -1; 1524204076Spjd 1525204076Spjd for (;;) { 1526204076Spjd mtx_lock(&sync_lock); 1527211897Spjd if (offset >= 0 && !sync_inprogress) { 1528211879Spjd pjdlog_info("Synchronization interrupted. " 1529211879Spjd "%jd bytes synchronized so far.", 1530211879Spjd (intmax_t)synced); 1531212038Spjd event_send(res, EVENT_SYNCINTR); 1532211879Spjd } 1533204076Spjd while (!sync_inprogress) { 1534204076Spjd dorewind = true; 1535204076Spjd synced = 0; 1536204076Spjd cv_wait(&sync_cond, &sync_lock); 1537204076Spjd } 1538204076Spjd mtx_unlock(&sync_lock); 1539204076Spjd /* 1540204076Spjd * Obtain offset at which we should synchronize. 1541204076Spjd * Rewind synchronization if needed. 1542204076Spjd */ 1543204076Spjd mtx_lock(&res->hr_amp_lock); 1544204076Spjd if (dorewind) 1545204076Spjd activemap_sync_rewind(res->hr_amp); 1546204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1547204076Spjd if (syncext != -1) { 1548204076Spjd /* 1549204076Spjd * We synchronized entire syncext extent, we can mark 1550204076Spjd * it as clean now. 1551204076Spjd */ 1552204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1553204076Spjd (void)hast_activemap_flush(res); 1554204076Spjd } 1555204076Spjd mtx_unlock(&res->hr_amp_lock); 1556204076Spjd if (dorewind) { 1557204076Spjd dorewind = false; 1558204076Spjd if (offset < 0) 1559204076Spjd pjdlog_info("Nodes are in sync."); 1560204076Spjd else { 1561204076Spjd pjdlog_info("Synchronization started. %ju bytes to go.", 1562204076Spjd (uintmax_t)(res->hr_extentsize * 1563204076Spjd activemap_ndirty(res->hr_amp))); 1564212038Spjd event_send(res, EVENT_SYNCSTART); 1565204076Spjd } 1566204076Spjd } 1567204076Spjd if (offset < 0) { 1568211878Spjd sync_stop(); 1569204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1570204076Spjd /* 1571204076Spjd * Synchronization complete, make both localcnt and 1572204076Spjd * remotecnt equal. 1573204076Spjd */ 1574204076Spjd ncomp = 1; 1575204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1576204076Spjd if (ISCONNECTED(res, ncomp)) { 1577204076Spjd if (synced > 0) { 1578204076Spjd pjdlog_info("Synchronization complete. " 1579204076Spjd "%jd bytes synchronized.", 1580204076Spjd (intmax_t)synced); 1581212038Spjd event_send(res, EVENT_SYNCDONE); 1582204076Spjd } 1583204076Spjd mtx_lock(&metadata_lock); 1584204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1585204076Spjd res->hr_primary_localcnt = 1586204076Spjd res->hr_secondary_localcnt; 1587204076Spjd res->hr_primary_remotecnt = 1588204076Spjd res->hr_secondary_remotecnt; 1589204076Spjd pjdlog_debug(1, 1590204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1591204076Spjd (uintmax_t)res->hr_primary_localcnt, 1592204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1593204076Spjd (void)metadata_write(res); 1594204076Spjd mtx_unlock(&metadata_lock); 1595204076Spjd } 1596204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1597204076Spjd continue; 1598204076Spjd } 1599204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1600204076Spjd QUEUE_TAKE2(hio, free); 1601204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1602204076Spjd /* 1603204076Spjd * Lock the range we are going to synchronize. We don't want 1604204076Spjd * race where someone writes between our read and write. 1605204076Spjd */ 1606204076Spjd for (;;) { 1607204076Spjd mtx_lock(&range_lock); 1608204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1609204076Spjd pjdlog_debug(2, 1610204076Spjd "sync: Range offset=%jd length=%jd locked.", 1611204076Spjd (intmax_t)offset, (intmax_t)length); 1612204076Spjd range_sync_wait = true; 1613204076Spjd cv_wait(&range_sync_cond, &range_lock); 1614204076Spjd range_sync_wait = false; 1615204076Spjd mtx_unlock(&range_lock); 1616204076Spjd continue; 1617204076Spjd } 1618204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1619204076Spjd mtx_unlock(&range_lock); 1620204076Spjd pjdlog_debug(2, 1621204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1622204076Spjd (intmax_t)offset, (intmax_t)length); 1623204076Spjd sleep(1); 1624204076Spjd continue; 1625204076Spjd } 1626204076Spjd mtx_unlock(&range_lock); 1627204076Spjd break; 1628204076Spjd } 1629204076Spjd /* 1630204076Spjd * First read the data from synchronization source. 1631204076Spjd */ 1632204076Spjd SYNCREQ(hio); 1633204076Spjd ggio = &hio->hio_ggio; 1634204076Spjd ggio->gctl_cmd = BIO_READ; 1635204076Spjd ggio->gctl_offset = offset; 1636204076Spjd ggio->gctl_length = length; 1637204076Spjd ggio->gctl_error = 0; 1638204076Spjd for (ii = 0; ii < ncomps; ii++) 1639204076Spjd hio->hio_errors[ii] = EINVAL; 1640204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1641204076Spjd hio); 1642204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1643204076Spjd hio); 1644204076Spjd mtx_lock(&metadata_lock); 1645204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1646204076Spjd /* 1647204076Spjd * This range is up-to-date on local component, 1648204076Spjd * so handle request locally. 1649204076Spjd */ 1650204076Spjd /* Local component is 0 for now. */ 1651204076Spjd ncomp = 0; 1652204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1653204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1654204076Spjd /* 1655204076Spjd * This range is out-of-date on local component, 1656204076Spjd * so send request to the remote node. 1657204076Spjd */ 1658204076Spjd /* Remote component is 1 for now. */ 1659204076Spjd ncomp = 1; 1660204076Spjd } 1661204076Spjd mtx_unlock(&metadata_lock); 1662204076Spjd refcount_init(&hio->hio_countdown, 1); 1663204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1664204076Spjd 1665204076Spjd /* 1666204076Spjd * Let's wait for READ to finish. 1667204076Spjd */ 1668204076Spjd mtx_lock(&sync_lock); 1669204076Spjd while (!ISSYNCREQDONE(hio)) 1670204076Spjd cv_wait(&sync_cond, &sync_lock); 1671204076Spjd mtx_unlock(&sync_lock); 1672204076Spjd 1673204076Spjd if (hio->hio_errors[ncomp] != 0) { 1674204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1675204076Spjd strerror(hio->hio_errors[ncomp])); 1676204076Spjd goto free_queue; 1677204076Spjd } 1678204076Spjd 1679204076Spjd /* 1680204076Spjd * We read the data from synchronization source, now write it 1681204076Spjd * to synchronization target. 1682204076Spjd */ 1683204076Spjd SYNCREQ(hio); 1684204076Spjd ggio->gctl_cmd = BIO_WRITE; 1685204076Spjd for (ii = 0; ii < ncomps; ii++) 1686204076Spjd hio->hio_errors[ii] = EINVAL; 1687204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1688204076Spjd hio); 1689204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1690204076Spjd hio); 1691204076Spjd mtx_lock(&metadata_lock); 1692204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1693204076Spjd /* 1694204076Spjd * This range is up-to-date on local component, 1695204076Spjd * so we update remote component. 1696204076Spjd */ 1697204076Spjd /* Remote component is 1 for now. */ 1698204076Spjd ncomp = 1; 1699204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1700204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1701204076Spjd /* 1702204076Spjd * This range is out-of-date on local component, 1703204076Spjd * so we update it. 1704204076Spjd */ 1705204076Spjd /* Local component is 0 for now. */ 1706204076Spjd ncomp = 0; 1707204076Spjd } 1708204076Spjd mtx_unlock(&metadata_lock); 1709204076Spjd 1710204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1711204076Spjd hio); 1712204076Spjd refcount_init(&hio->hio_countdown, 1); 1713204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1714204076Spjd 1715204076Spjd /* 1716204076Spjd * Let's wait for WRITE to finish. 1717204076Spjd */ 1718204076Spjd mtx_lock(&sync_lock); 1719204076Spjd while (!ISSYNCREQDONE(hio)) 1720204076Spjd cv_wait(&sync_cond, &sync_lock); 1721204076Spjd mtx_unlock(&sync_lock); 1722204076Spjd 1723204076Spjd if (hio->hio_errors[ncomp] != 0) { 1724204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1725204076Spjd strerror(hio->hio_errors[ncomp])); 1726204076Spjd goto free_queue; 1727204076Spjd } 1728211880Spjd 1729211880Spjd synced += length; 1730204076Spjdfree_queue: 1731204076Spjd mtx_lock(&range_lock); 1732204076Spjd rangelock_del(range_sync, offset, length); 1733204076Spjd if (range_regular_wait) 1734204076Spjd cv_signal(&range_regular_cond); 1735204076Spjd mtx_unlock(&range_lock); 1736204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1737204076Spjd hio); 1738204076Spjd QUEUE_INSERT2(hio, free); 1739204076Spjd } 1740204076Spjd /* NOTREACHED */ 1741204076Spjd return (NULL); 1742204076Spjd} 1743204076Spjd 1744204076Spjdstatic void 1745210886Spjdconfig_reload(void) 1746210886Spjd{ 1747210886Spjd struct hastd_config *newcfg; 1748210886Spjd struct hast_resource *res; 1749210886Spjd unsigned int ii, ncomps; 1750210886Spjd int modified; 1751210886Spjd 1752210886Spjd pjdlog_info("Reloading configuration..."); 1753210886Spjd 1754210886Spjd ncomps = HAST_NCOMPONENTS; 1755210886Spjd 1756210886Spjd newcfg = yy_config_parse(cfgpath, false); 1757210886Spjd if (newcfg == NULL) 1758210886Spjd goto failed; 1759210886Spjd 1760210886Spjd TAILQ_FOREACH(res, &newcfg->hc_resources, hr_next) { 1761210886Spjd if (strcmp(res->hr_name, gres->hr_name) == 0) 1762210886Spjd break; 1763210886Spjd } 1764210886Spjd /* 1765210886Spjd * If resource was removed from the configuration file, resource 1766210886Spjd * name, provider name or path to local component was modified we 1767210886Spjd * shouldn't be here. This means that someone modified configuration 1768210886Spjd * file and send SIGHUP to us instead of main hastd process. 1769210886Spjd * Log advice and ignore the signal. 1770210886Spjd */ 1771210886Spjd if (res == NULL || strcmp(gres->hr_name, res->hr_name) != 0 || 1772210886Spjd strcmp(gres->hr_provname, res->hr_provname) != 0 || 1773210886Spjd strcmp(gres->hr_localpath, res->hr_localpath) != 0) { 1774210886Spjd pjdlog_warning("To reload configuration send SIGHUP to the main hastd process (pid %u).", 1775210886Spjd (unsigned int)getppid()); 1776210886Spjd goto failed; 1777210886Spjd } 1778210886Spjd 1779210886Spjd#define MODIFIED_REMOTEADDR 0x1 1780210886Spjd#define MODIFIED_REPLICATION 0x2 1781210886Spjd#define MODIFIED_TIMEOUT 0x4 1782211886Spjd#define MODIFIED_EXEC 0x8 1783210886Spjd modified = 0; 1784210886Spjd if (strcmp(gres->hr_remoteaddr, res->hr_remoteaddr) != 0) { 1785210886Spjd /* 1786210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1787210886Spjd * We want remote_close() to log disconnect from the old 1788210886Spjd * addresses, not from the new ones. 1789210886Spjd */ 1790210886Spjd modified |= MODIFIED_REMOTEADDR; 1791210886Spjd } 1792210886Spjd if (gres->hr_replication != res->hr_replication) { 1793210886Spjd gres->hr_replication = res->hr_replication; 1794210886Spjd modified |= MODIFIED_REPLICATION; 1795210886Spjd } 1796210886Spjd if (gres->hr_timeout != res->hr_timeout) { 1797210886Spjd gres->hr_timeout = res->hr_timeout; 1798210886Spjd modified |= MODIFIED_TIMEOUT; 1799210886Spjd } 1800211886Spjd if (strcmp(gres->hr_exec, res->hr_exec) != 0) { 1801211886Spjd strlcpy(gres->hr_exec, res->hr_exec, sizeof(gres->hr_exec)); 1802211886Spjd modified |= MODIFIED_EXEC; 1803211886Spjd } 1804210886Spjd /* 1805210886Spjd * If only timeout was modified we only need to change it without 1806210886Spjd * reconnecting. 1807210886Spjd */ 1808210886Spjd if (modified == MODIFIED_TIMEOUT) { 1809210886Spjd for (ii = 0; ii < ncomps; ii++) { 1810210886Spjd if (!ISREMOTE(ii)) 1811210886Spjd continue; 1812210886Spjd rw_rlock(&hio_remote_lock[ii]); 1813210886Spjd if (!ISCONNECTED(gres, ii)) { 1814210886Spjd rw_unlock(&hio_remote_lock[ii]); 1815210886Spjd continue; 1816210886Spjd } 1817210886Spjd rw_unlock(&hio_remote_lock[ii]); 1818210886Spjd if (proto_timeout(gres->hr_remotein, 1819210886Spjd gres->hr_timeout) < 0) { 1820210886Spjd pjdlog_errno(LOG_WARNING, 1821210886Spjd "Unable to set connection timeout"); 1822210886Spjd } 1823210886Spjd if (proto_timeout(gres->hr_remoteout, 1824210886Spjd gres->hr_timeout) < 0) { 1825210886Spjd pjdlog_errno(LOG_WARNING, 1826210886Spjd "Unable to set connection timeout"); 1827210886Spjd } 1828210886Spjd } 1829211886Spjd } else if ((modified & 1830211886Spjd (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1831210886Spjd for (ii = 0; ii < ncomps; ii++) { 1832210886Spjd if (!ISREMOTE(ii)) 1833210886Spjd continue; 1834210886Spjd remote_close(gres, ii); 1835210886Spjd } 1836210886Spjd if (modified & MODIFIED_REMOTEADDR) { 1837210886Spjd strlcpy(gres->hr_remoteaddr, res->hr_remoteaddr, 1838210886Spjd sizeof(gres->hr_remoteaddr)); 1839210886Spjd } 1840210886Spjd } 1841210886Spjd#undef MODIFIED_REMOTEADDR 1842210886Spjd#undef MODIFIED_REPLICATION 1843210886Spjd#undef MODIFIED_TIMEOUT 1844211886Spjd#undef MODIFIED_EXEC 1845210886Spjd 1846210886Spjd pjdlog_info("Configuration reloaded successfully."); 1847210886Spjd return; 1848210886Spjdfailed: 1849210886Spjd if (newcfg != NULL) { 1850210886Spjd if (newcfg->hc_controlconn != NULL) 1851210886Spjd proto_close(newcfg->hc_controlconn); 1852210886Spjd if (newcfg->hc_listenconn != NULL) 1853210886Spjd proto_close(newcfg->hc_listenconn); 1854210886Spjd yy_config_free(newcfg); 1855210886Spjd } 1856210886Spjd pjdlog_warning("Configuration not reloaded."); 1857210886Spjd} 1858210886Spjd 1859211882Spjdstatic void 1860211882Spjdkeepalive_send(struct hast_resource *res, unsigned int ncomp) 1861211882Spjd{ 1862211882Spjd struct nv *nv; 1863211882Spjd 1864211882Spjd nv = nv_alloc(); 1865211882Spjd nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1866211882Spjd if (nv_error(nv) != 0) { 1867211882Spjd nv_free(nv); 1868211882Spjd pjdlog_debug(1, 1869211882Spjd "keepalive_send: Unable to prepare header to send."); 1870211882Spjd return; 1871211882Spjd } 1872211882Spjd if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1873211882Spjd pjdlog_common(LOG_DEBUG, 1, errno, 1874211882Spjd "keepalive_send: Unable to send request"); 1875211882Spjd nv_free(nv); 1876211882Spjd rw_unlock(&hio_remote_lock[ncomp]); 1877211882Spjd remote_close(res, ncomp); 1878211882Spjd rw_rlock(&hio_remote_lock[ncomp]); 1879211882Spjd return; 1880211882Spjd } 1881211882Spjd nv_free(nv); 1882211882Spjd pjdlog_debug(2, "keepalive_send: Request sent."); 1883211882Spjd} 1884211882Spjd 1885211981Spjdstatic void 1886211981Spjdguard_one(struct hast_resource *res, unsigned int ncomp) 1887211981Spjd{ 1888211981Spjd struct proto_conn *in, *out; 1889211981Spjd 1890211981Spjd if (!ISREMOTE(ncomp)) 1891211981Spjd return; 1892211981Spjd 1893211981Spjd rw_rlock(&hio_remote_lock[ncomp]); 1894211981Spjd 1895211981Spjd if (!real_remote(res)) { 1896211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1897211981Spjd return; 1898211981Spjd } 1899211981Spjd 1900211981Spjd if (ISCONNECTED(res, ncomp)) { 1901211981Spjd assert(res->hr_remotein != NULL); 1902211981Spjd assert(res->hr_remoteout != NULL); 1903211981Spjd keepalive_send(res, ncomp); 1904211981Spjd } 1905211981Spjd 1906211981Spjd if (ISCONNECTED(res, ncomp)) { 1907211981Spjd assert(res->hr_remotein != NULL); 1908211981Spjd assert(res->hr_remoteout != NULL); 1909211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1910211981Spjd pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 1911211981Spjd res->hr_remoteaddr); 1912211981Spjd return; 1913211981Spjd } 1914211981Spjd 1915211981Spjd assert(res->hr_remotein == NULL); 1916211981Spjd assert(res->hr_remoteout == NULL); 1917211981Spjd /* 1918211981Spjd * Upgrade the lock. It doesn't have to be atomic as no other thread 1919211981Spjd * can change connection status from disconnected to connected. 1920211981Spjd */ 1921211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1922211981Spjd pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 1923211981Spjd res->hr_remoteaddr); 1924211981Spjd in = out = NULL; 1925211981Spjd if (init_remote(res, &in, &out)) { 1926211981Spjd rw_wlock(&hio_remote_lock[ncomp]); 1927211981Spjd assert(res->hr_remotein == NULL); 1928211981Spjd assert(res->hr_remoteout == NULL); 1929211981Spjd assert(in != NULL && out != NULL); 1930211981Spjd res->hr_remotein = in; 1931211981Spjd res->hr_remoteout = out; 1932211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1933211981Spjd pjdlog_info("Successfully reconnected to %s.", 1934211981Spjd res->hr_remoteaddr); 1935211981Spjd sync_start(); 1936211981Spjd } else { 1937211981Spjd /* Both connections should be NULL. */ 1938211981Spjd assert(res->hr_remotein == NULL); 1939211981Spjd assert(res->hr_remoteout == NULL); 1940211981Spjd assert(in == NULL && out == NULL); 1941211981Spjd pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 1942211981Spjd res->hr_remoteaddr); 1943211981Spjd } 1944211981Spjd} 1945211981Spjd 1946204076Spjd/* 1947204076Spjd * Thread guards remote connections and reconnects when needed, handles 1948204076Spjd * signals, etc. 1949204076Spjd */ 1950204076Spjdstatic void * 1951204076Spjdguard_thread(void *arg) 1952204076Spjd{ 1953204076Spjd struct hast_resource *res = arg; 1954204076Spjd unsigned int ii, ncomps; 1955211982Spjd struct timespec timeout; 1956211981Spjd time_t lastcheck, now; 1957211982Spjd sigset_t mask; 1958211982Spjd int signo; 1959204076Spjd 1960204076Spjd ncomps = HAST_NCOMPONENTS; 1961211981Spjd lastcheck = time(NULL); 1962204076Spjd 1963211982Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1964211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1965211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1966211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1967211982Spjd 1968211982Spjd timeout.tv_nsec = 0; 1969211982Spjd signo = -1; 1970211982Spjd 1971204076Spjd for (;;) { 1972211982Spjd switch (signo) { 1973211982Spjd case SIGHUP: 1974211982Spjd config_reload(); 1975211982Spjd break; 1976211982Spjd case SIGINT: 1977211982Spjd case SIGTERM: 1978211982Spjd sigexit_received = true; 1979204076Spjd primary_exitx(EX_OK, 1980204076Spjd "Termination signal received, exiting."); 1981211982Spjd break; 1982211982Spjd default: 1983211982Spjd break; 1984204076Spjd } 1985211882Spjd 1986204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 1987211981Spjd now = time(NULL); 1988211982Spjd if (lastcheck + RETRY_SLEEP <= now) { 1989211982Spjd for (ii = 0; ii < ncomps; ii++) 1990211981Spjd guard_one(res, ii); 1991211981Spjd lastcheck = now; 1992204076Spjd } 1993211982Spjd timeout.tv_sec = RETRY_SLEEP; 1994211982Spjd signo = sigtimedwait(&mask, NULL, &timeout); 1995204076Spjd } 1996204076Spjd /* NOTREACHED */ 1997204076Spjd return (NULL); 1998204076Spjd} 1999