primary.c revision 212034
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 212034 2010-08-30 22:28:04Z 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" 61204076Spjd#include "hast.h" 62204076Spjd#include "hast_proto.h" 63204076Spjd#include "hastd.h" 64211886Spjd#include "hooks.h" 65204076Spjd#include "metadata.h" 66204076Spjd#include "proto.h" 67204076Spjd#include "pjdlog.h" 68204076Spjd#include "subr.h" 69204076Spjd#include "synch.h" 70204076Spjd 71210886Spjd/* The is only one remote component for now. */ 72210886Spjd#define ISREMOTE(no) ((no) == 1) 73210886Spjd 74204076Spjdstruct hio { 75204076Spjd /* 76204076Spjd * Number of components we are still waiting for. 77204076Spjd * When this field goes to 0, we can send the request back to the 78204076Spjd * kernel. Each component has to decrease this counter by one 79204076Spjd * even on failure. 80204076Spjd */ 81204076Spjd unsigned int hio_countdown; 82204076Spjd /* 83204076Spjd * Each component has a place to store its own error. 84204076Spjd * Once the request is handled by all components we can decide if the 85204076Spjd * request overall is successful or not. 86204076Spjd */ 87204076Spjd int *hio_errors; 88204076Spjd /* 89204076Spjd * Structure used to comunicate with GEOM Gate class. 90204076Spjd */ 91204076Spjd struct g_gate_ctl_io hio_ggio; 92204076Spjd TAILQ_ENTRY(hio) *hio_next; 93204076Spjd}; 94204076Spjd#define hio_free_next hio_next[0] 95204076Spjd#define hio_done_next hio_next[0] 96204076Spjd 97204076Spjd/* 98204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 99204076Spjd * until some in-progress requests are freed. 100204076Spjd */ 101204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 102204076Spjdstatic pthread_mutex_t hio_free_list_lock; 103204076Spjdstatic pthread_cond_t hio_free_list_cond; 104204076Spjd/* 105204076Spjd * There is one send list for every component. One requests is placed on all 106204076Spjd * send lists - each component gets the same request, but each component is 107204076Spjd * responsible for managing his own send list. 108204076Spjd */ 109204076Spjdstatic TAILQ_HEAD(, hio) *hio_send_list; 110204076Spjdstatic pthread_mutex_t *hio_send_list_lock; 111204076Spjdstatic pthread_cond_t *hio_send_list_cond; 112204076Spjd/* 113204076Spjd * There is one recv list for every component, although local components don't 114204076Spjd * use recv lists as local requests are done synchronously. 115204076Spjd */ 116204076Spjdstatic TAILQ_HEAD(, hio) *hio_recv_list; 117204076Spjdstatic pthread_mutex_t *hio_recv_list_lock; 118204076Spjdstatic pthread_cond_t *hio_recv_list_cond; 119204076Spjd/* 120204076Spjd * Request is placed on done list by the slowest component (the one that 121204076Spjd * decreased hio_countdown from 1 to 0). 122204076Spjd */ 123204076Spjdstatic TAILQ_HEAD(, hio) hio_done_list; 124204076Spjdstatic pthread_mutex_t hio_done_list_lock; 125204076Spjdstatic pthread_cond_t hio_done_list_cond; 126204076Spjd/* 127204076Spjd * Structure below are for interaction with sync thread. 128204076Spjd */ 129204076Spjdstatic bool sync_inprogress; 130204076Spjdstatic pthread_mutex_t sync_lock; 131204076Spjdstatic pthread_cond_t sync_cond; 132204076Spjd/* 133204076Spjd * The lock below allows to synchornize access to remote connections. 134204076Spjd */ 135204076Spjdstatic pthread_rwlock_t *hio_remote_lock; 136204076Spjd 137204076Spjd/* 138204076Spjd * Lock to synchronize metadata updates. Also synchronize access to 139204076Spjd * hr_primary_localcnt and hr_primary_remotecnt fields. 140204076Spjd */ 141204076Spjdstatic pthread_mutex_t metadata_lock; 142204076Spjd 143204076Spjd/* 144204076Spjd * Maximum number of outstanding I/O requests. 145204076Spjd */ 146204076Spjd#define HAST_HIO_MAX 256 147204076Spjd/* 148204076Spjd * Number of components. At this point there are only two components: local 149204076Spjd * and remote, but in the future it might be possible to use multiple local 150204076Spjd * and remote components. 151204076Spjd */ 152204076Spjd#define HAST_NCOMPONENTS 2 153204076Spjd/* 154211982Spjd * Number of seconds to sleep between reconnect retries or keepalive packets. 155204076Spjd */ 156211982Spjd#define RETRY_SLEEP 10 157204076Spjd 158204076Spjd#define ISCONNECTED(res, no) \ 159204076Spjd ((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL) 160204076Spjd 161204076Spjd#define QUEUE_INSERT1(hio, name, ncomp) do { \ 162204076Spjd bool _wakeup; \ 163204076Spjd \ 164204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 165204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]); \ 166204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio), \ 167204076Spjd hio_next[(ncomp)]); \ 168204076Spjd mtx_unlock(&hio_##name##_list_lock[ncomp]); \ 169204076Spjd if (_wakeup) \ 170204076Spjd cv_signal(&hio_##name##_list_cond[(ncomp)]); \ 171204076Spjd} while (0) 172204076Spjd#define QUEUE_INSERT2(hio, name) do { \ 173204076Spjd bool _wakeup; \ 174204076Spjd \ 175204076Spjd mtx_lock(&hio_##name##_list_lock); \ 176204076Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 177204076Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\ 178204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 179204076Spjd if (_wakeup) \ 180204076Spjd cv_signal(&hio_##name##_list_cond); \ 181204076Spjd} while (0) 182204076Spjd#define QUEUE_TAKE1(hio, name, ncomp) do { \ 183204076Spjd mtx_lock(&hio_##name##_list_lock[(ncomp)]); \ 184204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL) { \ 185204076Spjd cv_wait(&hio_##name##_list_cond[(ncomp)], \ 186204076Spjd &hio_##name##_list_lock[(ncomp)]); \ 187204076Spjd } \ 188204076Spjd TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio), \ 189204076Spjd hio_next[(ncomp)]); \ 190204076Spjd mtx_unlock(&hio_##name##_list_lock[(ncomp)]); \ 191204076Spjd} while (0) 192204076Spjd#define QUEUE_TAKE2(hio, name) do { \ 193204076Spjd mtx_lock(&hio_##name##_list_lock); \ 194204076Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 195204076Spjd cv_wait(&hio_##name##_list_cond, \ 196204076Spjd &hio_##name##_list_lock); \ 197204076Spjd } \ 198204076Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next); \ 199204076Spjd mtx_unlock(&hio_##name##_list_lock); \ 200204076Spjd} while (0) 201204076Spjd 202209183Spjd#define SYNCREQ(hio) do { \ 203209183Spjd (hio)->hio_ggio.gctl_unit = -1; \ 204209183Spjd (hio)->hio_ggio.gctl_seq = 1; \ 205209183Spjd} while (0) 206204076Spjd#define ISSYNCREQ(hio) ((hio)->hio_ggio.gctl_unit == -1) 207204076Spjd#define SYNCREQDONE(hio) do { (hio)->hio_ggio.gctl_unit = -2; } while (0) 208204076Spjd#define ISSYNCREQDONE(hio) ((hio)->hio_ggio.gctl_unit == -2) 209204076Spjd 210204076Spjdstatic struct hast_resource *gres; 211204076Spjd 212204076Spjdstatic pthread_mutex_t range_lock; 213204076Spjdstatic struct rangelocks *range_regular; 214204076Spjdstatic bool range_regular_wait; 215204076Spjdstatic pthread_cond_t range_regular_cond; 216204076Spjdstatic struct rangelocks *range_sync; 217204076Spjdstatic bool range_sync_wait; 218204076Spjdstatic pthread_cond_t range_sync_cond; 219204076Spjd 220204076Spjdstatic void *ggate_recv_thread(void *arg); 221204076Spjdstatic void *local_send_thread(void *arg); 222204076Spjdstatic void *remote_send_thread(void *arg); 223204076Spjdstatic void *remote_recv_thread(void *arg); 224204076Spjdstatic void *ggate_send_thread(void *arg); 225204076Spjdstatic void *sync_thread(void *arg); 226204076Spjdstatic void *guard_thread(void *arg); 227204076Spjd 228211982Spjdstatic void 229211982Spjddummy_sighandler(int sig __unused) 230211982Spjd{ 231211982Spjd /* Nothing to do. */ 232211982Spjd} 233204076Spjd 234204076Spjdstatic void 235204076Spjdcleanup(struct hast_resource *res) 236204076Spjd{ 237204076Spjd int rerrno; 238204076Spjd 239204076Spjd /* Remember errno. */ 240204076Spjd rerrno = errno; 241204076Spjd 242204076Spjd /* 243204076Spjd * Close descriptor to /dev/hast/<name> 244204076Spjd * to work-around race in the kernel. 245204076Spjd */ 246204076Spjd close(res->hr_localfd); 247204076Spjd 248204076Spjd /* Destroy ggate provider if we created one. */ 249204076Spjd if (res->hr_ggateunit >= 0) { 250204076Spjd struct g_gate_ctl_destroy ggiod; 251204076Spjd 252204076Spjd ggiod.gctl_version = G_GATE_VERSION; 253204076Spjd ggiod.gctl_unit = res->hr_ggateunit; 254204076Spjd ggiod.gctl_force = 1; 255204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) < 0) { 256204076Spjd pjdlog_warning("Unable to destroy hast/%s device", 257204076Spjd res->hr_provname); 258204076Spjd } 259204076Spjd res->hr_ggateunit = -1; 260204076Spjd } 261204076Spjd 262204076Spjd /* Restore errno. */ 263204076Spjd errno = rerrno; 264204076Spjd} 265204076Spjd 266204076Spjdstatic void 267204076Spjdprimary_exit(int exitcode, const char *fmt, ...) 268204076Spjd{ 269204076Spjd va_list ap; 270204076Spjd 271204076Spjd assert(exitcode != EX_OK); 272204076Spjd va_start(ap, fmt); 273204076Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 274204076Spjd va_end(ap); 275204076Spjd cleanup(gres); 276204076Spjd exit(exitcode); 277204076Spjd} 278204076Spjd 279204076Spjdstatic void 280204076Spjdprimary_exitx(int exitcode, const char *fmt, ...) 281204076Spjd{ 282204076Spjd va_list ap; 283204076Spjd 284204076Spjd va_start(ap, fmt); 285204076Spjd pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap); 286204076Spjd va_end(ap); 287204076Spjd cleanup(gres); 288204076Spjd exit(exitcode); 289204076Spjd} 290204076Spjd 291204076Spjdstatic int 292204076Spjdhast_activemap_flush(struct hast_resource *res) 293204076Spjd{ 294204076Spjd const unsigned char *buf; 295204076Spjd size_t size; 296204076Spjd 297204076Spjd buf = activemap_bitmap(res->hr_amp, &size); 298204076Spjd assert(buf != NULL); 299204076Spjd assert((size % res->hr_local_sectorsize) == 0); 300204076Spjd if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) != 301204076Spjd (ssize_t)size) { 302204076Spjd KEEP_ERRNO(pjdlog_errno(LOG_ERR, 303204076Spjd "Unable to flush activemap to disk")); 304204076Spjd return (-1); 305204076Spjd } 306204076Spjd return (0); 307204076Spjd} 308204076Spjd 309210881Spjdstatic bool 310210881Spjdreal_remote(const struct hast_resource *res) 311210881Spjd{ 312210881Spjd 313210881Spjd return (strcmp(res->hr_remoteaddr, "none") != 0); 314210881Spjd} 315210881Spjd 316204076Spjdstatic void 317204076Spjdinit_environment(struct hast_resource *res __unused) 318204076Spjd{ 319204076Spjd struct hio *hio; 320204076Spjd unsigned int ii, ncomps; 321211982Spjd sigset_t mask; 322204076Spjd 323204076Spjd /* 324204076Spjd * In the future it might be per-resource value. 325204076Spjd */ 326204076Spjd ncomps = HAST_NCOMPONENTS; 327204076Spjd 328204076Spjd /* 329204076Spjd * Allocate memory needed by lists. 330204076Spjd */ 331204076Spjd hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps); 332204076Spjd if (hio_send_list == NULL) { 333204076Spjd primary_exitx(EX_TEMPFAIL, 334204076Spjd "Unable to allocate %zu bytes of memory for send lists.", 335204076Spjd sizeof(hio_send_list[0]) * ncomps); 336204076Spjd } 337204076Spjd hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps); 338204076Spjd if (hio_send_list_lock == NULL) { 339204076Spjd primary_exitx(EX_TEMPFAIL, 340204076Spjd "Unable to allocate %zu bytes of memory for send list locks.", 341204076Spjd sizeof(hio_send_list_lock[0]) * ncomps); 342204076Spjd } 343204076Spjd hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps); 344204076Spjd if (hio_send_list_cond == NULL) { 345204076Spjd primary_exitx(EX_TEMPFAIL, 346204076Spjd "Unable to allocate %zu bytes of memory for send list condition variables.", 347204076Spjd sizeof(hio_send_list_cond[0]) * ncomps); 348204076Spjd } 349204076Spjd hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps); 350204076Spjd if (hio_recv_list == NULL) { 351204076Spjd primary_exitx(EX_TEMPFAIL, 352204076Spjd "Unable to allocate %zu bytes of memory for recv lists.", 353204076Spjd sizeof(hio_recv_list[0]) * ncomps); 354204076Spjd } 355204076Spjd hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps); 356204076Spjd if (hio_recv_list_lock == NULL) { 357204076Spjd primary_exitx(EX_TEMPFAIL, 358204076Spjd "Unable to allocate %zu bytes of memory for recv list locks.", 359204076Spjd sizeof(hio_recv_list_lock[0]) * ncomps); 360204076Spjd } 361204076Spjd hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps); 362204076Spjd if (hio_recv_list_cond == NULL) { 363204076Spjd primary_exitx(EX_TEMPFAIL, 364204076Spjd "Unable to allocate %zu bytes of memory for recv list condition variables.", 365204076Spjd sizeof(hio_recv_list_cond[0]) * ncomps); 366204076Spjd } 367204076Spjd hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps); 368204076Spjd if (hio_remote_lock == NULL) { 369204076Spjd primary_exitx(EX_TEMPFAIL, 370204076Spjd "Unable to allocate %zu bytes of memory for remote connections locks.", 371204076Spjd sizeof(hio_remote_lock[0]) * ncomps); 372204076Spjd } 373204076Spjd 374204076Spjd /* 375204076Spjd * Initialize lists, their locks and theirs condition variables. 376204076Spjd */ 377204076Spjd TAILQ_INIT(&hio_free_list); 378204076Spjd mtx_init(&hio_free_list_lock); 379204076Spjd cv_init(&hio_free_list_cond); 380204076Spjd for (ii = 0; ii < HAST_NCOMPONENTS; ii++) { 381204076Spjd TAILQ_INIT(&hio_send_list[ii]); 382204076Spjd mtx_init(&hio_send_list_lock[ii]); 383204076Spjd cv_init(&hio_send_list_cond[ii]); 384204076Spjd TAILQ_INIT(&hio_recv_list[ii]); 385204076Spjd mtx_init(&hio_recv_list_lock[ii]); 386204076Spjd cv_init(&hio_recv_list_cond[ii]); 387204076Spjd rw_init(&hio_remote_lock[ii]); 388204076Spjd } 389204076Spjd TAILQ_INIT(&hio_done_list); 390204076Spjd mtx_init(&hio_done_list_lock); 391204076Spjd cv_init(&hio_done_list_cond); 392204076Spjd mtx_init(&metadata_lock); 393204076Spjd 394204076Spjd /* 395204076Spjd * Allocate requests pool and initialize requests. 396204076Spjd */ 397204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 398204076Spjd hio = malloc(sizeof(*hio)); 399204076Spjd if (hio == NULL) { 400204076Spjd primary_exitx(EX_TEMPFAIL, 401204076Spjd "Unable to allocate %zu bytes of memory for hio request.", 402204076Spjd sizeof(*hio)); 403204076Spjd } 404204076Spjd hio->hio_countdown = 0; 405204076Spjd hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps); 406204076Spjd if (hio->hio_errors == NULL) { 407204076Spjd primary_exitx(EX_TEMPFAIL, 408204076Spjd "Unable allocate %zu bytes of memory for hio errors.", 409204076Spjd sizeof(hio->hio_errors[0]) * ncomps); 410204076Spjd } 411204076Spjd hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps); 412204076Spjd if (hio->hio_next == NULL) { 413204076Spjd primary_exitx(EX_TEMPFAIL, 414204076Spjd "Unable allocate %zu bytes of memory for hio_next field.", 415204076Spjd sizeof(hio->hio_next[0]) * ncomps); 416204076Spjd } 417204076Spjd hio->hio_ggio.gctl_version = G_GATE_VERSION; 418204076Spjd hio->hio_ggio.gctl_data = malloc(MAXPHYS); 419204076Spjd if (hio->hio_ggio.gctl_data == NULL) { 420204076Spjd primary_exitx(EX_TEMPFAIL, 421204076Spjd "Unable to allocate %zu bytes of memory for gctl_data.", 422204076Spjd MAXPHYS); 423204076Spjd } 424204076Spjd hio->hio_ggio.gctl_length = MAXPHYS; 425204076Spjd hio->hio_ggio.gctl_error = 0; 426204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next); 427204076Spjd } 428204076Spjd 429204076Spjd /* 430204076Spjd * Turn on signals handling. 431204076Spjd */ 432211982Spjd /* Because SIGCHLD is ignored by default, setup dummy handler for it. */ 433211982Spjd PJDLOG_VERIFY(signal(SIGCHLD, dummy_sighandler) != SIG_ERR); 434211982Spjd PJDLOG_VERIFY(sigfillset(&mask) == 0); 435211982Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 436204076Spjd} 437204076Spjd 438204076Spjdstatic void 439204076Spjdinit_local(struct hast_resource *res) 440204076Spjd{ 441204076Spjd unsigned char *buf; 442204076Spjd size_t mapsize; 443204076Spjd 444204076Spjd if (metadata_read(res, true) < 0) 445204076Spjd exit(EX_NOINPUT); 446204076Spjd mtx_init(&res->hr_amp_lock); 447204076Spjd if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize, 448204076Spjd res->hr_local_sectorsize, res->hr_keepdirty) < 0) { 449204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create activemap"); 450204076Spjd } 451204076Spjd mtx_init(&range_lock); 452204076Spjd cv_init(&range_regular_cond); 453204076Spjd if (rangelock_init(&range_regular) < 0) 454204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create regular range lock"); 455204076Spjd cv_init(&range_sync_cond); 456204076Spjd if (rangelock_init(&range_sync) < 0) 457204076Spjd primary_exit(EX_TEMPFAIL, "Unable to create sync range lock"); 458204076Spjd mapsize = activemap_ondisk_size(res->hr_amp); 459204076Spjd buf = calloc(1, mapsize); 460204076Spjd if (buf == NULL) { 461204076Spjd primary_exitx(EX_TEMPFAIL, 462204076Spjd "Unable to allocate buffer for activemap."); 463204076Spjd } 464204076Spjd if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) != 465204076Spjd (ssize_t)mapsize) { 466204076Spjd primary_exit(EX_NOINPUT, "Unable to read activemap"); 467204076Spjd } 468204076Spjd activemap_copyin(res->hr_amp, buf, mapsize); 469209181Spjd free(buf); 470204076Spjd if (res->hr_resuid != 0) 471204076Spjd return; 472204076Spjd /* 473204076Spjd * We're using provider for the first time, so we have to generate 474204076Spjd * resource unique identifier and initialize local and remote counts. 475204076Spjd */ 476204076Spjd arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid)); 477204076Spjd res->hr_primary_localcnt = 1; 478204076Spjd res->hr_primary_remotecnt = 0; 479204076Spjd if (metadata_write(res) < 0) 480204076Spjd exit(EX_NOINPUT); 481204076Spjd} 482204076Spjd 483205738Spjdstatic bool 484205738Spjdinit_remote(struct hast_resource *res, struct proto_conn **inp, 485205738Spjd struct proto_conn **outp) 486204076Spjd{ 487205738Spjd struct proto_conn *in, *out; 488204076Spjd struct nv *nvout, *nvin; 489204076Spjd const unsigned char *token; 490204076Spjd unsigned char *map; 491204076Spjd const char *errmsg; 492204076Spjd int32_t extentsize; 493204076Spjd int64_t datasize; 494204076Spjd uint32_t mapsize; 495204076Spjd size_t size; 496204076Spjd 497205738Spjd assert((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL)); 498210881Spjd assert(real_remote(res)); 499205738Spjd 500205738Spjd in = out = NULL; 501211983Spjd errmsg = NULL; 502205738Spjd 503204076Spjd /* Prepare outgoing connection with remote node. */ 504205738Spjd if (proto_client(res->hr_remoteaddr, &out) < 0) { 505207347Spjd primary_exit(EX_TEMPFAIL, "Unable to create connection to %s", 506204076Spjd res->hr_remoteaddr); 507204076Spjd } 508204076Spjd /* Try to connect, but accept failure. */ 509205738Spjd if (proto_connect(out) < 0) { 510204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 511204076Spjd res->hr_remoteaddr); 512204076Spjd goto close; 513204076Spjd } 514207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 515207371Spjd if (proto_timeout(out, res->hr_timeout) < 0) 516207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 517204076Spjd /* 518204076Spjd * First handshake step. 519204076Spjd * Setup outgoing connection with remote node. 520204076Spjd */ 521204076Spjd nvout = nv_alloc(); 522204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 523204076Spjd if (nv_error(nvout) != 0) { 524204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 525204076Spjd "Unable to allocate header for connection with %s", 526204076Spjd res->hr_remoteaddr); 527204076Spjd nv_free(nvout); 528204076Spjd goto close; 529204076Spjd } 530205738Spjd if (hast_proto_send(res, out, nvout, NULL, 0) < 0) { 531204076Spjd pjdlog_errno(LOG_WARNING, 532204076Spjd "Unable to send handshake header to %s", 533204076Spjd res->hr_remoteaddr); 534204076Spjd nv_free(nvout); 535204076Spjd goto close; 536204076Spjd } 537204076Spjd nv_free(nvout); 538205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 539204076Spjd pjdlog_errno(LOG_WARNING, 540204076Spjd "Unable to receive handshake header from %s", 541204076Spjd res->hr_remoteaddr); 542204076Spjd goto close; 543204076Spjd } 544204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 545204076Spjd if (errmsg != NULL) { 546204076Spjd pjdlog_warning("%s", errmsg); 547204076Spjd nv_free(nvin); 548204076Spjd goto close; 549204076Spjd } 550204076Spjd token = nv_get_uint8_array(nvin, &size, "token"); 551204076Spjd if (token == NULL) { 552204076Spjd pjdlog_warning("Handshake header from %s has no 'token' field.", 553204076Spjd res->hr_remoteaddr); 554204076Spjd nv_free(nvin); 555204076Spjd goto close; 556204076Spjd } 557204076Spjd if (size != sizeof(res->hr_token)) { 558204076Spjd pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).", 559204076Spjd res->hr_remoteaddr, size, sizeof(res->hr_token)); 560204076Spjd nv_free(nvin); 561204076Spjd goto close; 562204076Spjd } 563204076Spjd bcopy(token, res->hr_token, sizeof(res->hr_token)); 564204076Spjd nv_free(nvin); 565204076Spjd 566204076Spjd /* 567204076Spjd * Second handshake step. 568204076Spjd * Setup incoming connection with remote node. 569204076Spjd */ 570205738Spjd if (proto_client(res->hr_remoteaddr, &in) < 0) { 571204076Spjd pjdlog_errno(LOG_WARNING, "Unable to create connection to %s", 572204076Spjd res->hr_remoteaddr); 573204076Spjd } 574204076Spjd /* Try to connect, but accept failure. */ 575205738Spjd if (proto_connect(in) < 0) { 576204076Spjd pjdlog_errno(LOG_WARNING, "Unable to connect to %s", 577204076Spjd res->hr_remoteaddr); 578204076Spjd goto close; 579204076Spjd } 580207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 581207371Spjd if (proto_timeout(in, res->hr_timeout) < 0) 582207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 583204076Spjd nvout = nv_alloc(); 584204076Spjd nv_add_string(nvout, res->hr_name, "resource"); 585204076Spjd nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token), 586204076Spjd "token"); 587204076Spjd nv_add_uint64(nvout, res->hr_resuid, "resuid"); 588204076Spjd nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt"); 589204076Spjd nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt"); 590204076Spjd if (nv_error(nvout) != 0) { 591204076Spjd pjdlog_common(LOG_WARNING, 0, nv_error(nvout), 592204076Spjd "Unable to allocate header for connection with %s", 593204076Spjd res->hr_remoteaddr); 594204076Spjd nv_free(nvout); 595204076Spjd goto close; 596204076Spjd } 597205738Spjd if (hast_proto_send(res, in, nvout, NULL, 0) < 0) { 598204076Spjd pjdlog_errno(LOG_WARNING, 599204076Spjd "Unable to send handshake header to %s", 600204076Spjd res->hr_remoteaddr); 601204076Spjd nv_free(nvout); 602204076Spjd goto close; 603204076Spjd } 604204076Spjd nv_free(nvout); 605205738Spjd if (hast_proto_recv_hdr(out, &nvin) < 0) { 606204076Spjd pjdlog_errno(LOG_WARNING, 607204076Spjd "Unable to receive handshake header from %s", 608204076Spjd res->hr_remoteaddr); 609204076Spjd goto close; 610204076Spjd } 611204076Spjd errmsg = nv_get_string(nvin, "errmsg"); 612204076Spjd if (errmsg != NULL) { 613204076Spjd pjdlog_warning("%s", errmsg); 614204076Spjd nv_free(nvin); 615204076Spjd goto close; 616204076Spjd } 617204076Spjd datasize = nv_get_int64(nvin, "datasize"); 618204076Spjd if (datasize != res->hr_datasize) { 619204076Spjd pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).", 620204076Spjd (intmax_t)res->hr_datasize, (intmax_t)datasize); 621204076Spjd nv_free(nvin); 622204076Spjd goto close; 623204076Spjd } 624204076Spjd extentsize = nv_get_int32(nvin, "extentsize"); 625204076Spjd if (extentsize != res->hr_extentsize) { 626204076Spjd pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).", 627204076Spjd (ssize_t)res->hr_extentsize, (ssize_t)extentsize); 628204076Spjd nv_free(nvin); 629204076Spjd goto close; 630204076Spjd } 631204076Spjd res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt"); 632204076Spjd res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 633204076Spjd res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc"); 634204076Spjd map = NULL; 635204076Spjd mapsize = nv_get_uint32(nvin, "mapsize"); 636204076Spjd if (mapsize > 0) { 637204076Spjd map = malloc(mapsize); 638204076Spjd if (map == NULL) { 639204076Spjd pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).", 640204076Spjd (uintmax_t)mapsize); 641204076Spjd nv_free(nvin); 642204076Spjd goto close; 643204076Spjd } 644204076Spjd /* 645204076Spjd * Remote node have some dirty extents on its own, lets 646204076Spjd * download its activemap. 647204076Spjd */ 648205738Spjd if (hast_proto_recv_data(res, out, nvin, map, 649204076Spjd mapsize) < 0) { 650204076Spjd pjdlog_errno(LOG_ERR, 651204076Spjd "Unable to receive remote activemap"); 652204076Spjd nv_free(nvin); 653204076Spjd free(map); 654204076Spjd goto close; 655204076Spjd } 656204076Spjd /* 657204076Spjd * Merge local and remote bitmaps. 658204076Spjd */ 659204076Spjd activemap_merge(res->hr_amp, map, mapsize); 660204076Spjd free(map); 661204076Spjd /* 662204076Spjd * Now that we merged bitmaps from both nodes, flush it to the 663204076Spjd * disk before we start to synchronize. 664204076Spjd */ 665204076Spjd (void)hast_activemap_flush(res); 666204076Spjd } 667204076Spjd pjdlog_info("Connected to %s.", res->hr_remoteaddr); 668205738Spjd if (inp != NULL && outp != NULL) { 669205738Spjd *inp = in; 670205738Spjd *outp = out; 671205738Spjd } else { 672205738Spjd res->hr_remotein = in; 673205738Spjd res->hr_remoteout = out; 674205738Spjd } 675211984Spjd hook_exec(res->hr_exec, "connect", res->hr_name, NULL); 676205738Spjd return (true); 677205738Spjdclose: 678211983Spjd if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0) 679211983Spjd hook_exec(res->hr_exec, "split-brain", res->hr_name, NULL); 680205738Spjd proto_close(out); 681205738Spjd if (in != NULL) 682205738Spjd proto_close(in); 683205738Spjd return (false); 684205738Spjd} 685205738Spjd 686205738Spjdstatic void 687205738Spjdsync_start(void) 688205738Spjd{ 689205738Spjd 690204076Spjd mtx_lock(&sync_lock); 691204076Spjd sync_inprogress = true; 692204076Spjd mtx_unlock(&sync_lock); 693204076Spjd cv_signal(&sync_cond); 694204076Spjd} 695204076Spjd 696204076Spjdstatic void 697211878Spjdsync_stop(void) 698211878Spjd{ 699211878Spjd 700211878Spjd mtx_lock(&sync_lock); 701211878Spjd if (sync_inprogress) 702211878Spjd sync_inprogress = false; 703211878Spjd mtx_unlock(&sync_lock); 704211878Spjd} 705211878Spjd 706211878Spjdstatic void 707204076Spjdinit_ggate(struct hast_resource *res) 708204076Spjd{ 709204076Spjd struct g_gate_ctl_create ggiocreate; 710204076Spjd struct g_gate_ctl_cancel ggiocancel; 711204076Spjd 712204076Spjd /* 713204076Spjd * We communicate with ggate via /dev/ggctl. Open it. 714204076Spjd */ 715204076Spjd res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR); 716204076Spjd if (res->hr_ggatefd < 0) 717204076Spjd primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME); 718204076Spjd /* 719204076Spjd * Create provider before trying to connect, as connection failure 720204076Spjd * is not critical, but may take some time. 721204076Spjd */ 722204076Spjd ggiocreate.gctl_version = G_GATE_VERSION; 723204076Spjd ggiocreate.gctl_mediasize = res->hr_datasize; 724204076Spjd ggiocreate.gctl_sectorsize = res->hr_local_sectorsize; 725204076Spjd ggiocreate.gctl_flags = 0; 726206669Spjd ggiocreate.gctl_maxcount = G_GATE_MAX_QUEUE_SIZE; 727204076Spjd ggiocreate.gctl_timeout = 0; 728204076Spjd ggiocreate.gctl_unit = G_GATE_NAME_GIVEN; 729204076Spjd snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s", 730204076Spjd res->hr_provname); 731204076Spjd bzero(ggiocreate.gctl_info, sizeof(ggiocreate.gctl_info)); 732204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) { 733204076Spjd pjdlog_info("Device hast/%s created.", res->hr_provname); 734204076Spjd res->hr_ggateunit = ggiocreate.gctl_unit; 735204076Spjd return; 736204076Spjd } 737204076Spjd if (errno != EEXIST) { 738204076Spjd primary_exit(EX_OSERR, "Unable to create hast/%s device", 739204076Spjd res->hr_provname); 740204076Spjd } 741204076Spjd pjdlog_debug(1, 742204076Spjd "Device hast/%s already exists, we will try to take it over.", 743204076Spjd res->hr_provname); 744204076Spjd /* 745204076Spjd * If we received EEXIST, we assume that the process who created the 746204076Spjd * provider died and didn't clean up. In that case we will start from 747204076Spjd * where he left of. 748204076Spjd */ 749204076Spjd ggiocancel.gctl_version = G_GATE_VERSION; 750204076Spjd ggiocancel.gctl_unit = G_GATE_NAME_GIVEN; 751204076Spjd snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s", 752204076Spjd res->hr_provname); 753204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) { 754204076Spjd pjdlog_info("Device hast/%s recovered.", res->hr_provname); 755204076Spjd res->hr_ggateunit = ggiocancel.gctl_unit; 756204076Spjd return; 757204076Spjd } 758204076Spjd primary_exit(EX_OSERR, "Unable to take over hast/%s device", 759204076Spjd res->hr_provname); 760204076Spjd} 761204076Spjd 762204076Spjdvoid 763204076Spjdhastd_primary(struct hast_resource *res) 764204076Spjd{ 765204076Spjd pthread_t td; 766204076Spjd pid_t pid; 767204076Spjd int error; 768204076Spjd 769204076Spjd /* 770204076Spjd * Create communication channel between parent and child. 771204076Spjd */ 772204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 773204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 774212034Spjd pjdlog_exit(EX_OSERR, 775204076Spjd "Unable to create control sockets between parent and child"); 776204076Spjd } 777204076Spjd 778204076Spjd pid = fork(); 779204076Spjd if (pid < 0) { 780204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 781212034Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to fork"); 782204076Spjd } 783204076Spjd 784204076Spjd if (pid > 0) { 785204076Spjd /* This is parent. */ 786204076Spjd res->hr_workerpid = pid; 787204076Spjd return; 788204076Spjd } 789211977Spjd 790211984Spjd gres = res; 791211984Spjd 792204076Spjd (void)pidfile_close(pfh); 793211977Spjd hook_fini(); 794204076Spjd 795204076Spjd setproctitle("%s (primary)", res->hr_name); 796204076Spjd 797210880Spjd signal(SIGHUP, SIG_DFL); 798210880Spjd signal(SIGCHLD, SIG_DFL); 799210880Spjd 800211886Spjd hook_init(); 801204076Spjd init_local(res); 802210881Spjd if (real_remote(res) && init_remote(res, NULL, NULL)) 803205738Spjd sync_start(); 804204076Spjd init_ggate(res); 805204076Spjd init_environment(res); 806204076Spjd error = pthread_create(&td, NULL, ggate_recv_thread, res); 807204076Spjd assert(error == 0); 808204076Spjd error = pthread_create(&td, NULL, local_send_thread, res); 809204076Spjd assert(error == 0); 810204076Spjd error = pthread_create(&td, NULL, remote_send_thread, res); 811204076Spjd assert(error == 0); 812204076Spjd error = pthread_create(&td, NULL, remote_recv_thread, res); 813204076Spjd assert(error == 0); 814204076Spjd error = pthread_create(&td, NULL, ggate_send_thread, res); 815204076Spjd assert(error == 0); 816204076Spjd error = pthread_create(&td, NULL, sync_thread, res); 817204076Spjd assert(error == 0); 818204076Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 819204076Spjd assert(error == 0); 820204076Spjd (void)guard_thread(res); 821204076Spjd} 822204076Spjd 823204076Spjdstatic void 824204076Spjdreqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio, const char *fmt, ...) 825204076Spjd{ 826204076Spjd char msg[1024]; 827204076Spjd va_list ap; 828204076Spjd int len; 829204076Spjd 830204076Spjd va_start(ap, fmt); 831204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 832204076Spjd va_end(ap); 833204076Spjd if ((size_t)len < sizeof(msg)) { 834204076Spjd switch (ggio->gctl_cmd) { 835204076Spjd case BIO_READ: 836204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 837204076Spjd "READ(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 838204076Spjd (uintmax_t)ggio->gctl_length); 839204076Spjd break; 840204076Spjd case BIO_DELETE: 841204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 842204076Spjd "DELETE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 843204076Spjd (uintmax_t)ggio->gctl_length); 844204076Spjd break; 845204076Spjd case BIO_FLUSH: 846204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 847204076Spjd break; 848204076Spjd case BIO_WRITE: 849204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 850204076Spjd "WRITE(%ju, %ju).", (uintmax_t)ggio->gctl_offset, 851204076Spjd (uintmax_t)ggio->gctl_length); 852204076Spjd break; 853204076Spjd default: 854204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 855204076Spjd "UNKNOWN(%u).", (unsigned int)ggio->gctl_cmd); 856204076Spjd break; 857204076Spjd } 858204076Spjd } 859204076Spjd pjdlog_common(loglevel, debuglevel, -1, "%s", msg); 860204076Spjd} 861204076Spjd 862204076Spjdstatic void 863204076Spjdremote_close(struct hast_resource *res, int ncomp) 864204076Spjd{ 865204076Spjd 866204076Spjd rw_wlock(&hio_remote_lock[ncomp]); 867204076Spjd /* 868204076Spjd * A race is possible between dropping rlock and acquiring wlock - 869204076Spjd * another thread can close connection in-between. 870204076Spjd */ 871204076Spjd if (!ISCONNECTED(res, ncomp)) { 872204076Spjd assert(res->hr_remotein == NULL); 873204076Spjd assert(res->hr_remoteout == NULL); 874204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 875204076Spjd return; 876204076Spjd } 877204076Spjd 878204076Spjd assert(res->hr_remotein != NULL); 879204076Spjd assert(res->hr_remoteout != NULL); 880204076Spjd 881211881Spjd pjdlog_debug(2, "Closing incoming connection to %s.", 882204076Spjd res->hr_remoteaddr); 883204076Spjd proto_close(res->hr_remotein); 884204076Spjd res->hr_remotein = NULL; 885211881Spjd pjdlog_debug(2, "Closing outgoing connection to %s.", 886204076Spjd res->hr_remoteaddr); 887204076Spjd proto_close(res->hr_remoteout); 888204076Spjd res->hr_remoteout = NULL; 889204076Spjd 890204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 891204076Spjd 892211881Spjd pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr); 893211881Spjd 894204076Spjd /* 895204076Spjd * Stop synchronization if in-progress. 896204076Spjd */ 897211878Spjd sync_stop(); 898211984Spjd 899211984Spjd hook_exec(res->hr_exec, "disconnect", res->hr_name, NULL); 900204076Spjd} 901204076Spjd 902204076Spjd/* 903204076Spjd * Thread receives ggate I/O requests from the kernel and passes them to 904204076Spjd * appropriate threads: 905204076Spjd * WRITE - always goes to both local_send and remote_send threads 906204076Spjd * READ (when the block is up-to-date on local component) - 907204076Spjd * only local_send thread 908204076Spjd * READ (when the block isn't up-to-date on local component) - 909204076Spjd * only remote_send thread 910204076Spjd * DELETE - always goes to both local_send and remote_send threads 911204076Spjd * FLUSH - always goes to both local_send and remote_send threads 912204076Spjd */ 913204076Spjdstatic void * 914204076Spjdggate_recv_thread(void *arg) 915204076Spjd{ 916204076Spjd struct hast_resource *res = arg; 917204076Spjd struct g_gate_ctl_io *ggio; 918204076Spjd struct hio *hio; 919204076Spjd unsigned int ii, ncomp, ncomps; 920204076Spjd int error; 921204076Spjd 922204076Spjd ncomps = HAST_NCOMPONENTS; 923204076Spjd 924204076Spjd for (;;) { 925204076Spjd pjdlog_debug(2, "ggate_recv: Taking free request."); 926204076Spjd QUEUE_TAKE2(hio, free); 927204076Spjd pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio); 928204076Spjd ggio = &hio->hio_ggio; 929204076Spjd ggio->gctl_unit = res->hr_ggateunit; 930204076Spjd ggio->gctl_length = MAXPHYS; 931204076Spjd ggio->gctl_error = 0; 932204076Spjd pjdlog_debug(2, 933204076Spjd "ggate_recv: (%p) Waiting for request from the kernel.", 934204076Spjd hio); 935204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) < 0) { 936204076Spjd if (sigexit_received) 937204076Spjd pthread_exit(NULL); 938204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_START failed"); 939204076Spjd } 940204076Spjd error = ggio->gctl_error; 941204076Spjd switch (error) { 942204076Spjd case 0: 943204076Spjd break; 944204076Spjd case ECANCELED: 945204076Spjd /* Exit gracefully. */ 946204076Spjd if (!sigexit_received) { 947204076Spjd pjdlog_debug(2, 948204076Spjd "ggate_recv: (%p) Received cancel from the kernel.", 949204076Spjd hio); 950204076Spjd pjdlog_info("Received cancel from the kernel, exiting."); 951204076Spjd } 952204076Spjd pthread_exit(NULL); 953204076Spjd case ENOMEM: 954204076Spjd /* 955204076Spjd * Buffer too small? Impossible, we allocate MAXPHYS 956204076Spjd * bytes - request can't be bigger than that. 957204076Spjd */ 958204076Spjd /* FALLTHROUGH */ 959204076Spjd case ENXIO: 960204076Spjd default: 961204076Spjd primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.", 962204076Spjd strerror(error)); 963204076Spjd } 964204076Spjd for (ii = 0; ii < ncomps; ii++) 965204076Spjd hio->hio_errors[ii] = EINVAL; 966204076Spjd reqlog(LOG_DEBUG, 2, ggio, 967204076Spjd "ggate_recv: (%p) Request received from the kernel: ", 968204076Spjd hio); 969204076Spjd /* 970204076Spjd * Inform all components about new write request. 971204076Spjd * For read request prefer local component unless the given 972204076Spjd * range is out-of-date, then use remote component. 973204076Spjd */ 974204076Spjd switch (ggio->gctl_cmd) { 975204076Spjd case BIO_READ: 976204076Spjd pjdlog_debug(2, 977204076Spjd "ggate_recv: (%p) Moving request to the send queue.", 978204076Spjd hio); 979204076Spjd refcount_init(&hio->hio_countdown, 1); 980204076Spjd mtx_lock(&metadata_lock); 981204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF || 982204076Spjd res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 983204076Spjd /* 984204076Spjd * This range is up-to-date on local component, 985204076Spjd * so handle request locally. 986204076Spjd */ 987204076Spjd /* Local component is 0 for now. */ 988204076Spjd ncomp = 0; 989204076Spjd } else /* if (res->hr_syncsrc == 990204076Spjd HAST_SYNCSRC_SECONDARY) */ { 991204076Spjd assert(res->hr_syncsrc == 992204076Spjd HAST_SYNCSRC_SECONDARY); 993204076Spjd /* 994204076Spjd * This range is out-of-date on local component, 995204076Spjd * so send request to the remote node. 996204076Spjd */ 997204076Spjd /* Remote component is 1 for now. */ 998204076Spjd ncomp = 1; 999204076Spjd } 1000204076Spjd mtx_unlock(&metadata_lock); 1001204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1002204076Spjd break; 1003204076Spjd case BIO_WRITE: 1004204076Spjd for (;;) { 1005204076Spjd mtx_lock(&range_lock); 1006204076Spjd if (rangelock_islocked(range_sync, 1007204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1008204076Spjd pjdlog_debug(2, 1009204076Spjd "regular: Range offset=%jd length=%zu locked.", 1010204076Spjd (intmax_t)ggio->gctl_offset, 1011204076Spjd (size_t)ggio->gctl_length); 1012204076Spjd range_regular_wait = true; 1013204076Spjd cv_wait(&range_regular_cond, &range_lock); 1014204076Spjd range_regular_wait = false; 1015204076Spjd mtx_unlock(&range_lock); 1016204076Spjd continue; 1017204076Spjd } 1018204076Spjd if (rangelock_add(range_regular, 1019204076Spjd ggio->gctl_offset, ggio->gctl_length) < 0) { 1020204076Spjd mtx_unlock(&range_lock); 1021204076Spjd pjdlog_debug(2, 1022204076Spjd "regular: Range offset=%jd length=%zu is already locked, waiting.", 1023204076Spjd (intmax_t)ggio->gctl_offset, 1024204076Spjd (size_t)ggio->gctl_length); 1025204076Spjd sleep(1); 1026204076Spjd continue; 1027204076Spjd } 1028204076Spjd mtx_unlock(&range_lock); 1029204076Spjd break; 1030204076Spjd } 1031204076Spjd mtx_lock(&res->hr_amp_lock); 1032204076Spjd if (activemap_write_start(res->hr_amp, 1033204076Spjd ggio->gctl_offset, ggio->gctl_length)) { 1034204076Spjd (void)hast_activemap_flush(res); 1035204076Spjd } 1036204076Spjd mtx_unlock(&res->hr_amp_lock); 1037204076Spjd /* FALLTHROUGH */ 1038204076Spjd case BIO_DELETE: 1039204076Spjd case BIO_FLUSH: 1040204076Spjd pjdlog_debug(2, 1041204076Spjd "ggate_recv: (%p) Moving request to the send queues.", 1042204076Spjd hio); 1043204076Spjd refcount_init(&hio->hio_countdown, ncomps); 1044204076Spjd for (ii = 0; ii < ncomps; ii++) 1045204076Spjd QUEUE_INSERT1(hio, send, ii); 1046204076Spjd break; 1047204076Spjd } 1048204076Spjd } 1049204076Spjd /* NOTREACHED */ 1050204076Spjd return (NULL); 1051204076Spjd} 1052204076Spjd 1053204076Spjd/* 1054204076Spjd * Thread reads from or writes to local component. 1055204076Spjd * If local read fails, it redirects it to remote_send thread. 1056204076Spjd */ 1057204076Spjdstatic void * 1058204076Spjdlocal_send_thread(void *arg) 1059204076Spjd{ 1060204076Spjd struct hast_resource *res = arg; 1061204076Spjd struct g_gate_ctl_io *ggio; 1062204076Spjd struct hio *hio; 1063204076Spjd unsigned int ncomp, rncomp; 1064204076Spjd ssize_t ret; 1065204076Spjd 1066204076Spjd /* Local component is 0 for now. */ 1067204076Spjd ncomp = 0; 1068204076Spjd /* Remote component is 1 for now. */ 1069204076Spjd rncomp = 1; 1070204076Spjd 1071204076Spjd for (;;) { 1072204076Spjd pjdlog_debug(2, "local_send: Taking request."); 1073204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1074204076Spjd pjdlog_debug(2, "local_send: (%p) Got request.", hio); 1075204076Spjd ggio = &hio->hio_ggio; 1076204076Spjd switch (ggio->gctl_cmd) { 1077204076Spjd case BIO_READ: 1078204076Spjd ret = pread(res->hr_localfd, ggio->gctl_data, 1079204076Spjd ggio->gctl_length, 1080204076Spjd ggio->gctl_offset + res->hr_localoff); 1081204076Spjd if (ret == ggio->gctl_length) 1082204076Spjd hio->hio_errors[ncomp] = 0; 1083204076Spjd else { 1084204076Spjd /* 1085204076Spjd * If READ failed, try to read from remote node. 1086204076Spjd */ 1087204076Spjd QUEUE_INSERT1(hio, send, rncomp); 1088204076Spjd continue; 1089204076Spjd } 1090204076Spjd break; 1091204076Spjd case BIO_WRITE: 1092204076Spjd ret = pwrite(res->hr_localfd, ggio->gctl_data, 1093204076Spjd ggio->gctl_length, 1094204076Spjd ggio->gctl_offset + res->hr_localoff); 1095204076Spjd if (ret < 0) 1096204076Spjd hio->hio_errors[ncomp] = errno; 1097204076Spjd else if (ret != ggio->gctl_length) 1098204076Spjd hio->hio_errors[ncomp] = EIO; 1099204076Spjd else 1100204076Spjd hio->hio_errors[ncomp] = 0; 1101204076Spjd break; 1102204076Spjd case BIO_DELETE: 1103204076Spjd ret = g_delete(res->hr_localfd, 1104204076Spjd ggio->gctl_offset + res->hr_localoff, 1105204076Spjd ggio->gctl_length); 1106204076Spjd if (ret < 0) 1107204076Spjd hio->hio_errors[ncomp] = errno; 1108204076Spjd else 1109204076Spjd hio->hio_errors[ncomp] = 0; 1110204076Spjd break; 1111204076Spjd case BIO_FLUSH: 1112204076Spjd ret = g_flush(res->hr_localfd); 1113204076Spjd if (ret < 0) 1114204076Spjd hio->hio_errors[ncomp] = errno; 1115204076Spjd else 1116204076Spjd hio->hio_errors[ncomp] = 0; 1117204076Spjd break; 1118204076Spjd } 1119204076Spjd if (refcount_release(&hio->hio_countdown)) { 1120204076Spjd if (ISSYNCREQ(hio)) { 1121204076Spjd mtx_lock(&sync_lock); 1122204076Spjd SYNCREQDONE(hio); 1123204076Spjd mtx_unlock(&sync_lock); 1124204076Spjd cv_signal(&sync_cond); 1125204076Spjd } else { 1126204076Spjd pjdlog_debug(2, 1127204076Spjd "local_send: (%p) Moving request to the done queue.", 1128204076Spjd hio); 1129204076Spjd QUEUE_INSERT2(hio, done); 1130204076Spjd } 1131204076Spjd } 1132204076Spjd } 1133204076Spjd /* NOTREACHED */ 1134204076Spjd return (NULL); 1135204076Spjd} 1136204076Spjd 1137204076Spjd/* 1138204076Spjd * Thread sends request to secondary node. 1139204076Spjd */ 1140204076Spjdstatic void * 1141204076Spjdremote_send_thread(void *arg) 1142204076Spjd{ 1143204076Spjd struct hast_resource *res = arg; 1144204076Spjd struct g_gate_ctl_io *ggio; 1145204076Spjd struct hio *hio; 1146204076Spjd struct nv *nv; 1147204076Spjd unsigned int ncomp; 1148204076Spjd bool wakeup; 1149204076Spjd uint64_t offset, length; 1150204076Spjd uint8_t cmd; 1151204076Spjd void *data; 1152204076Spjd 1153204076Spjd /* Remote component is 1 for now. */ 1154204076Spjd ncomp = 1; 1155204076Spjd 1156204076Spjd for (;;) { 1157204076Spjd pjdlog_debug(2, "remote_send: Taking request."); 1158204076Spjd QUEUE_TAKE1(hio, send, ncomp); 1159204076Spjd pjdlog_debug(2, "remote_send: (%p) Got request.", hio); 1160204076Spjd ggio = &hio->hio_ggio; 1161204076Spjd switch (ggio->gctl_cmd) { 1162204076Spjd case BIO_READ: 1163204076Spjd cmd = HIO_READ; 1164204076Spjd data = NULL; 1165204076Spjd offset = ggio->gctl_offset; 1166204076Spjd length = ggio->gctl_length; 1167204076Spjd break; 1168204076Spjd case BIO_WRITE: 1169204076Spjd cmd = HIO_WRITE; 1170204076Spjd data = ggio->gctl_data; 1171204076Spjd offset = ggio->gctl_offset; 1172204076Spjd length = ggio->gctl_length; 1173204076Spjd break; 1174204076Spjd case BIO_DELETE: 1175204076Spjd cmd = HIO_DELETE; 1176204076Spjd data = NULL; 1177204076Spjd offset = ggio->gctl_offset; 1178204076Spjd length = ggio->gctl_length; 1179204076Spjd break; 1180204076Spjd case BIO_FLUSH: 1181204076Spjd cmd = HIO_FLUSH; 1182204076Spjd data = NULL; 1183204076Spjd offset = 0; 1184204076Spjd length = 0; 1185204076Spjd break; 1186204076Spjd default: 1187204076Spjd assert(!"invalid condition"); 1188204076Spjd abort(); 1189204076Spjd } 1190204076Spjd nv = nv_alloc(); 1191204076Spjd nv_add_uint8(nv, cmd, "cmd"); 1192204076Spjd nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq"); 1193204076Spjd nv_add_uint64(nv, offset, "offset"); 1194204076Spjd nv_add_uint64(nv, length, "length"); 1195204076Spjd if (nv_error(nv) != 0) { 1196204076Spjd hio->hio_errors[ncomp] = nv_error(nv); 1197204076Spjd pjdlog_debug(2, 1198204076Spjd "remote_send: (%p) Unable to prepare header to send.", 1199204076Spjd hio); 1200204076Spjd reqlog(LOG_ERR, 0, ggio, 1201204076Spjd "Unable to prepare header to send (%s): ", 1202204076Spjd strerror(nv_error(nv))); 1203204076Spjd /* Move failed request immediately to the done queue. */ 1204204076Spjd goto done_queue; 1205204076Spjd } 1206204076Spjd pjdlog_debug(2, 1207204076Spjd "remote_send: (%p) Moving request to the recv queue.", 1208204076Spjd hio); 1209204076Spjd /* 1210204076Spjd * Protect connection from disappearing. 1211204076Spjd */ 1212204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1213204076Spjd if (!ISCONNECTED(res, ncomp)) { 1214204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1215204076Spjd hio->hio_errors[ncomp] = ENOTCONN; 1216204076Spjd goto done_queue; 1217204076Spjd } 1218204076Spjd /* 1219204076Spjd * Move the request to recv queue before sending it, because 1220204076Spjd * in different order we can get reply before we move request 1221204076Spjd * to recv queue. 1222204076Spjd */ 1223204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1224204076Spjd wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]); 1225204076Spjd TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1226204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1227204076Spjd if (hast_proto_send(res, res->hr_remoteout, nv, data, 1228204076Spjd data != NULL ? length : 0) < 0) { 1229204076Spjd hio->hio_errors[ncomp] = errno; 1230204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1231204076Spjd pjdlog_debug(2, 1232204076Spjd "remote_send: (%p) Unable to send request.", hio); 1233204076Spjd reqlog(LOG_ERR, 0, ggio, 1234204076Spjd "Unable to send request (%s): ", 1235204076Spjd strerror(hio->hio_errors[ncomp])); 1236211979Spjd remote_close(res, ncomp); 1237204076Spjd /* 1238204076Spjd * Take request back from the receive queue and move 1239204076Spjd * it immediately to the done queue. 1240204076Spjd */ 1241204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1242204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, hio_next[ncomp]); 1243204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1244204076Spjd goto done_queue; 1245204076Spjd } 1246204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1247204076Spjd nv_free(nv); 1248204076Spjd if (wakeup) 1249204076Spjd cv_signal(&hio_recv_list_cond[ncomp]); 1250204076Spjd continue; 1251204076Spjddone_queue: 1252204076Spjd nv_free(nv); 1253204076Spjd if (ISSYNCREQ(hio)) { 1254204076Spjd if (!refcount_release(&hio->hio_countdown)) 1255204076Spjd continue; 1256204076Spjd mtx_lock(&sync_lock); 1257204076Spjd SYNCREQDONE(hio); 1258204076Spjd mtx_unlock(&sync_lock); 1259204076Spjd cv_signal(&sync_cond); 1260204076Spjd continue; 1261204076Spjd } 1262204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1263204076Spjd mtx_lock(&res->hr_amp_lock); 1264204076Spjd if (activemap_need_sync(res->hr_amp, ggio->gctl_offset, 1265204076Spjd ggio->gctl_length)) { 1266204076Spjd (void)hast_activemap_flush(res); 1267204076Spjd } 1268204076Spjd mtx_unlock(&res->hr_amp_lock); 1269204076Spjd } 1270204076Spjd if (!refcount_release(&hio->hio_countdown)) 1271204076Spjd continue; 1272204076Spjd pjdlog_debug(2, 1273204076Spjd "remote_send: (%p) Moving request to the done queue.", 1274204076Spjd hio); 1275204076Spjd QUEUE_INSERT2(hio, done); 1276204076Spjd } 1277204076Spjd /* NOTREACHED */ 1278204076Spjd return (NULL); 1279204076Spjd} 1280204076Spjd 1281204076Spjd/* 1282204076Spjd * Thread receives answer from secondary node and passes it to ggate_send 1283204076Spjd * thread. 1284204076Spjd */ 1285204076Spjdstatic void * 1286204076Spjdremote_recv_thread(void *arg) 1287204076Spjd{ 1288204076Spjd struct hast_resource *res = arg; 1289204076Spjd struct g_gate_ctl_io *ggio; 1290204076Spjd struct hio *hio; 1291204076Spjd struct nv *nv; 1292204076Spjd unsigned int ncomp; 1293204076Spjd uint64_t seq; 1294204076Spjd int error; 1295204076Spjd 1296204076Spjd /* Remote component is 1 for now. */ 1297204076Spjd ncomp = 1; 1298204076Spjd 1299204076Spjd for (;;) { 1300204076Spjd /* Wait until there is anything to receive. */ 1301204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1302204076Spjd while (TAILQ_EMPTY(&hio_recv_list[ncomp])) { 1303204076Spjd pjdlog_debug(2, "remote_recv: No requests, waiting."); 1304204076Spjd cv_wait(&hio_recv_list_cond[ncomp], 1305204076Spjd &hio_recv_list_lock[ncomp]); 1306204076Spjd } 1307204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1308204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1309204076Spjd if (!ISCONNECTED(res, ncomp)) { 1310204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1311204076Spjd /* 1312204076Spjd * Connection is dead, so move all pending requests to 1313204076Spjd * the done queue (one-by-one). 1314204076Spjd */ 1315204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1316204076Spjd hio = TAILQ_FIRST(&hio_recv_list[ncomp]); 1317204076Spjd assert(hio != NULL); 1318204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1319204076Spjd hio_next[ncomp]); 1320204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1321204076Spjd goto done_queue; 1322204076Spjd } 1323204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 1324204076Spjd pjdlog_errno(LOG_ERR, 1325204076Spjd "Unable to receive reply header"); 1326204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1327204076Spjd remote_close(res, ncomp); 1328204076Spjd continue; 1329204076Spjd } 1330204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1331204076Spjd seq = nv_get_uint64(nv, "seq"); 1332204076Spjd if (seq == 0) { 1333204076Spjd pjdlog_error("Header contains no 'seq' field."); 1334204076Spjd nv_free(nv); 1335204076Spjd continue; 1336204076Spjd } 1337204076Spjd mtx_lock(&hio_recv_list_lock[ncomp]); 1338204076Spjd TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) { 1339204076Spjd if (hio->hio_ggio.gctl_seq == seq) { 1340204076Spjd TAILQ_REMOVE(&hio_recv_list[ncomp], hio, 1341204076Spjd hio_next[ncomp]); 1342204076Spjd break; 1343204076Spjd } 1344204076Spjd } 1345204076Spjd mtx_unlock(&hio_recv_list_lock[ncomp]); 1346204076Spjd if (hio == NULL) { 1347204076Spjd pjdlog_error("Found no request matching received 'seq' field (%ju).", 1348204076Spjd (uintmax_t)seq); 1349204076Spjd nv_free(nv); 1350204076Spjd continue; 1351204076Spjd } 1352204076Spjd error = nv_get_int16(nv, "error"); 1353204076Spjd if (error != 0) { 1354204076Spjd /* Request failed on remote side. */ 1355204076Spjd hio->hio_errors[ncomp] = 0; 1356204076Spjd nv_free(nv); 1357204076Spjd goto done_queue; 1358204076Spjd } 1359204076Spjd ggio = &hio->hio_ggio; 1360204076Spjd switch (ggio->gctl_cmd) { 1361204076Spjd case BIO_READ: 1362204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1363204076Spjd if (!ISCONNECTED(res, ncomp)) { 1364204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1365204076Spjd nv_free(nv); 1366204076Spjd goto done_queue; 1367204076Spjd } 1368204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 1369204076Spjd ggio->gctl_data, ggio->gctl_length) < 0) { 1370204076Spjd hio->hio_errors[ncomp] = errno; 1371204076Spjd pjdlog_errno(LOG_ERR, 1372204076Spjd "Unable to receive reply data"); 1373204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1374204076Spjd nv_free(nv); 1375204076Spjd remote_close(res, ncomp); 1376204076Spjd goto done_queue; 1377204076Spjd } 1378204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1379204076Spjd break; 1380204076Spjd case BIO_WRITE: 1381204076Spjd case BIO_DELETE: 1382204076Spjd case BIO_FLUSH: 1383204076Spjd break; 1384204076Spjd default: 1385204076Spjd assert(!"invalid condition"); 1386204076Spjd abort(); 1387204076Spjd } 1388204076Spjd hio->hio_errors[ncomp] = 0; 1389204076Spjd nv_free(nv); 1390204076Spjddone_queue: 1391204076Spjd if (refcount_release(&hio->hio_countdown)) { 1392204076Spjd if (ISSYNCREQ(hio)) { 1393204076Spjd mtx_lock(&sync_lock); 1394204076Spjd SYNCREQDONE(hio); 1395204076Spjd mtx_unlock(&sync_lock); 1396204076Spjd cv_signal(&sync_cond); 1397204076Spjd } else { 1398204076Spjd pjdlog_debug(2, 1399204076Spjd "remote_recv: (%p) Moving request to the done queue.", 1400204076Spjd hio); 1401204076Spjd QUEUE_INSERT2(hio, done); 1402204076Spjd } 1403204076Spjd } 1404204076Spjd } 1405204076Spjd /* NOTREACHED */ 1406204076Spjd return (NULL); 1407204076Spjd} 1408204076Spjd 1409204076Spjd/* 1410204076Spjd * Thread sends answer to the kernel. 1411204076Spjd */ 1412204076Spjdstatic void * 1413204076Spjdggate_send_thread(void *arg) 1414204076Spjd{ 1415204076Spjd struct hast_resource *res = arg; 1416204076Spjd struct g_gate_ctl_io *ggio; 1417204076Spjd struct hio *hio; 1418204076Spjd unsigned int ii, ncomp, ncomps; 1419204076Spjd 1420204076Spjd ncomps = HAST_NCOMPONENTS; 1421204076Spjd 1422204076Spjd for (;;) { 1423204076Spjd pjdlog_debug(2, "ggate_send: Taking request."); 1424204076Spjd QUEUE_TAKE2(hio, done); 1425204076Spjd pjdlog_debug(2, "ggate_send: (%p) Got request.", hio); 1426204076Spjd ggio = &hio->hio_ggio; 1427204076Spjd for (ii = 0; ii < ncomps; ii++) { 1428204076Spjd if (hio->hio_errors[ii] == 0) { 1429204076Spjd /* 1430204076Spjd * One successful request is enough to declare 1431204076Spjd * success. 1432204076Spjd */ 1433204076Spjd ggio->gctl_error = 0; 1434204076Spjd break; 1435204076Spjd } 1436204076Spjd } 1437204076Spjd if (ii == ncomps) { 1438204076Spjd /* 1439204076Spjd * None of the requests were successful. 1440204076Spjd * Use first error. 1441204076Spjd */ 1442204076Spjd ggio->gctl_error = hio->hio_errors[0]; 1443204076Spjd } 1444204076Spjd if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) { 1445204076Spjd mtx_lock(&res->hr_amp_lock); 1446204076Spjd activemap_write_complete(res->hr_amp, 1447204076Spjd ggio->gctl_offset, ggio->gctl_length); 1448204076Spjd mtx_unlock(&res->hr_amp_lock); 1449204076Spjd } 1450204076Spjd if (ggio->gctl_cmd == BIO_WRITE) { 1451204076Spjd /* 1452204076Spjd * Unlock range we locked. 1453204076Spjd */ 1454204076Spjd mtx_lock(&range_lock); 1455204076Spjd rangelock_del(range_regular, ggio->gctl_offset, 1456204076Spjd ggio->gctl_length); 1457204076Spjd if (range_sync_wait) 1458204076Spjd cv_signal(&range_sync_cond); 1459204076Spjd mtx_unlock(&range_lock); 1460204076Spjd /* 1461204076Spjd * Bump local count if this is first write after 1462204076Spjd * connection failure with remote node. 1463204076Spjd */ 1464204076Spjd ncomp = 1; 1465204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1466204076Spjd if (!ISCONNECTED(res, ncomp)) { 1467204076Spjd mtx_lock(&metadata_lock); 1468204076Spjd if (res->hr_primary_localcnt == 1469204076Spjd res->hr_secondary_remotecnt) { 1470204076Spjd res->hr_primary_localcnt++; 1471204076Spjd pjdlog_debug(1, 1472204076Spjd "Increasing localcnt to %ju.", 1473204076Spjd (uintmax_t)res->hr_primary_localcnt); 1474204076Spjd (void)metadata_write(res); 1475204076Spjd } 1476204076Spjd mtx_unlock(&metadata_lock); 1477204076Spjd } 1478204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1479204076Spjd } 1480204076Spjd if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) < 0) 1481204076Spjd primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed"); 1482204076Spjd pjdlog_debug(2, 1483204076Spjd "ggate_send: (%p) Moving request to the free queue.", hio); 1484204076Spjd QUEUE_INSERT2(hio, free); 1485204076Spjd } 1486204076Spjd /* NOTREACHED */ 1487204076Spjd return (NULL); 1488204076Spjd} 1489204076Spjd 1490204076Spjd/* 1491204076Spjd * Thread synchronize local and remote components. 1492204076Spjd */ 1493204076Spjdstatic void * 1494204076Spjdsync_thread(void *arg __unused) 1495204076Spjd{ 1496204076Spjd struct hast_resource *res = arg; 1497204076Spjd struct hio *hio; 1498204076Spjd struct g_gate_ctl_io *ggio; 1499204076Spjd unsigned int ii, ncomp, ncomps; 1500204076Spjd off_t offset, length, synced; 1501204076Spjd bool dorewind; 1502204076Spjd int syncext; 1503204076Spjd 1504204076Spjd ncomps = HAST_NCOMPONENTS; 1505204076Spjd dorewind = true; 1506211897Spjd synced = 0; 1507211897Spjd offset = -1; 1508204076Spjd 1509204076Spjd for (;;) { 1510204076Spjd mtx_lock(&sync_lock); 1511211897Spjd if (offset >= 0 && !sync_inprogress) { 1512211879Spjd pjdlog_info("Synchronization interrupted. " 1513211879Spjd "%jd bytes synchronized so far.", 1514211879Spjd (intmax_t)synced); 1515211897Spjd hook_exec(res->hr_exec, "syncintr", res->hr_name, NULL); 1516211879Spjd } 1517204076Spjd while (!sync_inprogress) { 1518204076Spjd dorewind = true; 1519204076Spjd synced = 0; 1520204076Spjd cv_wait(&sync_cond, &sync_lock); 1521204076Spjd } 1522204076Spjd mtx_unlock(&sync_lock); 1523204076Spjd /* 1524204076Spjd * Obtain offset at which we should synchronize. 1525204076Spjd * Rewind synchronization if needed. 1526204076Spjd */ 1527204076Spjd mtx_lock(&res->hr_amp_lock); 1528204076Spjd if (dorewind) 1529204076Spjd activemap_sync_rewind(res->hr_amp); 1530204076Spjd offset = activemap_sync_offset(res->hr_amp, &length, &syncext); 1531204076Spjd if (syncext != -1) { 1532204076Spjd /* 1533204076Spjd * We synchronized entire syncext extent, we can mark 1534204076Spjd * it as clean now. 1535204076Spjd */ 1536204076Spjd if (activemap_extent_complete(res->hr_amp, syncext)) 1537204076Spjd (void)hast_activemap_flush(res); 1538204076Spjd } 1539204076Spjd mtx_unlock(&res->hr_amp_lock); 1540204076Spjd if (dorewind) { 1541204076Spjd dorewind = false; 1542204076Spjd if (offset < 0) 1543204076Spjd pjdlog_info("Nodes are in sync."); 1544204076Spjd else { 1545204076Spjd pjdlog_info("Synchronization started. %ju bytes to go.", 1546204076Spjd (uintmax_t)(res->hr_extentsize * 1547204076Spjd activemap_ndirty(res->hr_amp))); 1548211895Spjd hook_exec(res->hr_exec, "syncstart", 1549211895Spjd res->hr_name, NULL); 1550204076Spjd } 1551204076Spjd } 1552204076Spjd if (offset < 0) { 1553211878Spjd sync_stop(); 1554204076Spjd pjdlog_debug(1, "Nothing to synchronize."); 1555204076Spjd /* 1556204076Spjd * Synchronization complete, make both localcnt and 1557204076Spjd * remotecnt equal. 1558204076Spjd */ 1559204076Spjd ncomp = 1; 1560204076Spjd rw_rlock(&hio_remote_lock[ncomp]); 1561204076Spjd if (ISCONNECTED(res, ncomp)) { 1562204076Spjd if (synced > 0) { 1563204076Spjd pjdlog_info("Synchronization complete. " 1564204076Spjd "%jd bytes synchronized.", 1565204076Spjd (intmax_t)synced); 1566211895Spjd hook_exec(res->hr_exec, "syncdone", 1567211895Spjd res->hr_name, NULL); 1568204076Spjd } 1569204076Spjd mtx_lock(&metadata_lock); 1570204076Spjd res->hr_syncsrc = HAST_SYNCSRC_UNDEF; 1571204076Spjd res->hr_primary_localcnt = 1572204076Spjd res->hr_secondary_localcnt; 1573204076Spjd res->hr_primary_remotecnt = 1574204076Spjd res->hr_secondary_remotecnt; 1575204076Spjd pjdlog_debug(1, 1576204076Spjd "Setting localcnt to %ju and remotecnt to %ju.", 1577204076Spjd (uintmax_t)res->hr_primary_localcnt, 1578204076Spjd (uintmax_t)res->hr_secondary_localcnt); 1579204076Spjd (void)metadata_write(res); 1580204076Spjd mtx_unlock(&metadata_lock); 1581204076Spjd } 1582204076Spjd rw_unlock(&hio_remote_lock[ncomp]); 1583204076Spjd continue; 1584204076Spjd } 1585204076Spjd pjdlog_debug(2, "sync: Taking free request."); 1586204076Spjd QUEUE_TAKE2(hio, free); 1587204076Spjd pjdlog_debug(2, "sync: (%p) Got free request.", hio); 1588204076Spjd /* 1589204076Spjd * Lock the range we are going to synchronize. We don't want 1590204076Spjd * race where someone writes between our read and write. 1591204076Spjd */ 1592204076Spjd for (;;) { 1593204076Spjd mtx_lock(&range_lock); 1594204076Spjd if (rangelock_islocked(range_regular, offset, length)) { 1595204076Spjd pjdlog_debug(2, 1596204076Spjd "sync: Range offset=%jd length=%jd locked.", 1597204076Spjd (intmax_t)offset, (intmax_t)length); 1598204076Spjd range_sync_wait = true; 1599204076Spjd cv_wait(&range_sync_cond, &range_lock); 1600204076Spjd range_sync_wait = false; 1601204076Spjd mtx_unlock(&range_lock); 1602204076Spjd continue; 1603204076Spjd } 1604204076Spjd if (rangelock_add(range_sync, offset, length) < 0) { 1605204076Spjd mtx_unlock(&range_lock); 1606204076Spjd pjdlog_debug(2, 1607204076Spjd "sync: Range offset=%jd length=%jd is already locked, waiting.", 1608204076Spjd (intmax_t)offset, (intmax_t)length); 1609204076Spjd sleep(1); 1610204076Spjd continue; 1611204076Spjd } 1612204076Spjd mtx_unlock(&range_lock); 1613204076Spjd break; 1614204076Spjd } 1615204076Spjd /* 1616204076Spjd * First read the data from synchronization source. 1617204076Spjd */ 1618204076Spjd SYNCREQ(hio); 1619204076Spjd ggio = &hio->hio_ggio; 1620204076Spjd ggio->gctl_cmd = BIO_READ; 1621204076Spjd ggio->gctl_offset = offset; 1622204076Spjd ggio->gctl_length = length; 1623204076Spjd ggio->gctl_error = 0; 1624204076Spjd for (ii = 0; ii < ncomps; ii++) 1625204076Spjd hio->hio_errors[ii] = EINVAL; 1626204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1627204076Spjd hio); 1628204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1629204076Spjd hio); 1630204076Spjd mtx_lock(&metadata_lock); 1631204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1632204076Spjd /* 1633204076Spjd * This range is up-to-date on local component, 1634204076Spjd * so handle request locally. 1635204076Spjd */ 1636204076Spjd /* Local component is 0 for now. */ 1637204076Spjd ncomp = 0; 1638204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1639204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1640204076Spjd /* 1641204076Spjd * This range is out-of-date on local component, 1642204076Spjd * so send request to the remote node. 1643204076Spjd */ 1644204076Spjd /* Remote component is 1 for now. */ 1645204076Spjd ncomp = 1; 1646204076Spjd } 1647204076Spjd mtx_unlock(&metadata_lock); 1648204076Spjd refcount_init(&hio->hio_countdown, 1); 1649204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1650204076Spjd 1651204076Spjd /* 1652204076Spjd * Let's wait for READ to finish. 1653204076Spjd */ 1654204076Spjd mtx_lock(&sync_lock); 1655204076Spjd while (!ISSYNCREQDONE(hio)) 1656204076Spjd cv_wait(&sync_cond, &sync_lock); 1657204076Spjd mtx_unlock(&sync_lock); 1658204076Spjd 1659204076Spjd if (hio->hio_errors[ncomp] != 0) { 1660204076Spjd pjdlog_error("Unable to read synchronization data: %s.", 1661204076Spjd strerror(hio->hio_errors[ncomp])); 1662204076Spjd goto free_queue; 1663204076Spjd } 1664204076Spjd 1665204076Spjd /* 1666204076Spjd * We read the data from synchronization source, now write it 1667204076Spjd * to synchronization target. 1668204076Spjd */ 1669204076Spjd SYNCREQ(hio); 1670204076Spjd ggio->gctl_cmd = BIO_WRITE; 1671204076Spjd for (ii = 0; ii < ncomps; ii++) 1672204076Spjd hio->hio_errors[ii] = EINVAL; 1673204076Spjd reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ", 1674204076Spjd hio); 1675204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queue.", 1676204076Spjd hio); 1677204076Spjd mtx_lock(&metadata_lock); 1678204076Spjd if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) { 1679204076Spjd /* 1680204076Spjd * This range is up-to-date on local component, 1681204076Spjd * so we update remote component. 1682204076Spjd */ 1683204076Spjd /* Remote component is 1 for now. */ 1684204076Spjd ncomp = 1; 1685204076Spjd } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ { 1686204076Spjd assert(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY); 1687204076Spjd /* 1688204076Spjd * This range is out-of-date on local component, 1689204076Spjd * so we update it. 1690204076Spjd */ 1691204076Spjd /* Local component is 0 for now. */ 1692204076Spjd ncomp = 0; 1693204076Spjd } 1694204076Spjd mtx_unlock(&metadata_lock); 1695204076Spjd 1696204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the send queues.", 1697204076Spjd hio); 1698204076Spjd refcount_init(&hio->hio_countdown, 1); 1699204076Spjd QUEUE_INSERT1(hio, send, ncomp); 1700204076Spjd 1701204076Spjd /* 1702204076Spjd * Let's wait for WRITE to finish. 1703204076Spjd */ 1704204076Spjd mtx_lock(&sync_lock); 1705204076Spjd while (!ISSYNCREQDONE(hio)) 1706204076Spjd cv_wait(&sync_cond, &sync_lock); 1707204076Spjd mtx_unlock(&sync_lock); 1708204076Spjd 1709204076Spjd if (hio->hio_errors[ncomp] != 0) { 1710204076Spjd pjdlog_error("Unable to write synchronization data: %s.", 1711204076Spjd strerror(hio->hio_errors[ncomp])); 1712204076Spjd goto free_queue; 1713204076Spjd } 1714211880Spjd 1715211880Spjd synced += length; 1716204076Spjdfree_queue: 1717204076Spjd mtx_lock(&range_lock); 1718204076Spjd rangelock_del(range_sync, offset, length); 1719204076Spjd if (range_regular_wait) 1720204076Spjd cv_signal(&range_regular_cond); 1721204076Spjd mtx_unlock(&range_lock); 1722204076Spjd pjdlog_debug(2, "sync: (%p) Moving request to the free queue.", 1723204076Spjd hio); 1724204076Spjd QUEUE_INSERT2(hio, free); 1725204076Spjd } 1726204076Spjd /* NOTREACHED */ 1727204076Spjd return (NULL); 1728204076Spjd} 1729204076Spjd 1730204076Spjdstatic void 1731210886Spjdconfig_reload(void) 1732210886Spjd{ 1733210886Spjd struct hastd_config *newcfg; 1734210886Spjd struct hast_resource *res; 1735210886Spjd unsigned int ii, ncomps; 1736210886Spjd int modified; 1737210886Spjd 1738210886Spjd pjdlog_info("Reloading configuration..."); 1739210886Spjd 1740210886Spjd ncomps = HAST_NCOMPONENTS; 1741210886Spjd 1742210886Spjd newcfg = yy_config_parse(cfgpath, false); 1743210886Spjd if (newcfg == NULL) 1744210886Spjd goto failed; 1745210886Spjd 1746210886Spjd TAILQ_FOREACH(res, &newcfg->hc_resources, hr_next) { 1747210886Spjd if (strcmp(res->hr_name, gres->hr_name) == 0) 1748210886Spjd break; 1749210886Spjd } 1750210886Spjd /* 1751210886Spjd * If resource was removed from the configuration file, resource 1752210886Spjd * name, provider name or path to local component was modified we 1753210886Spjd * shouldn't be here. This means that someone modified configuration 1754210886Spjd * file and send SIGHUP to us instead of main hastd process. 1755210886Spjd * Log advice and ignore the signal. 1756210886Spjd */ 1757210886Spjd if (res == NULL || strcmp(gres->hr_name, res->hr_name) != 0 || 1758210886Spjd strcmp(gres->hr_provname, res->hr_provname) != 0 || 1759210886Spjd strcmp(gres->hr_localpath, res->hr_localpath) != 0) { 1760210886Spjd pjdlog_warning("To reload configuration send SIGHUP to the main hastd process (pid %u).", 1761210886Spjd (unsigned int)getppid()); 1762210886Spjd goto failed; 1763210886Spjd } 1764210886Spjd 1765210886Spjd#define MODIFIED_REMOTEADDR 0x1 1766210886Spjd#define MODIFIED_REPLICATION 0x2 1767210886Spjd#define MODIFIED_TIMEOUT 0x4 1768211886Spjd#define MODIFIED_EXEC 0x8 1769210886Spjd modified = 0; 1770210886Spjd if (strcmp(gres->hr_remoteaddr, res->hr_remoteaddr) != 0) { 1771210886Spjd /* 1772210886Spjd * Don't copy res->hr_remoteaddr to gres just yet. 1773210886Spjd * We want remote_close() to log disconnect from the old 1774210886Spjd * addresses, not from the new ones. 1775210886Spjd */ 1776210886Spjd modified |= MODIFIED_REMOTEADDR; 1777210886Spjd } 1778210886Spjd if (gres->hr_replication != res->hr_replication) { 1779210886Spjd gres->hr_replication = res->hr_replication; 1780210886Spjd modified |= MODIFIED_REPLICATION; 1781210886Spjd } 1782210886Spjd if (gres->hr_timeout != res->hr_timeout) { 1783210886Spjd gres->hr_timeout = res->hr_timeout; 1784210886Spjd modified |= MODIFIED_TIMEOUT; 1785210886Spjd } 1786211886Spjd if (strcmp(gres->hr_exec, res->hr_exec) != 0) { 1787211886Spjd strlcpy(gres->hr_exec, res->hr_exec, sizeof(gres->hr_exec)); 1788211886Spjd modified |= MODIFIED_EXEC; 1789211886Spjd } 1790210886Spjd /* 1791210886Spjd * If only timeout was modified we only need to change it without 1792210886Spjd * reconnecting. 1793210886Spjd */ 1794210886Spjd if (modified == MODIFIED_TIMEOUT) { 1795210886Spjd for (ii = 0; ii < ncomps; ii++) { 1796210886Spjd if (!ISREMOTE(ii)) 1797210886Spjd continue; 1798210886Spjd rw_rlock(&hio_remote_lock[ii]); 1799210886Spjd if (!ISCONNECTED(gres, ii)) { 1800210886Spjd rw_unlock(&hio_remote_lock[ii]); 1801210886Spjd continue; 1802210886Spjd } 1803210886Spjd rw_unlock(&hio_remote_lock[ii]); 1804210886Spjd if (proto_timeout(gres->hr_remotein, 1805210886Spjd gres->hr_timeout) < 0) { 1806210886Spjd pjdlog_errno(LOG_WARNING, 1807210886Spjd "Unable to set connection timeout"); 1808210886Spjd } 1809210886Spjd if (proto_timeout(gres->hr_remoteout, 1810210886Spjd gres->hr_timeout) < 0) { 1811210886Spjd pjdlog_errno(LOG_WARNING, 1812210886Spjd "Unable to set connection timeout"); 1813210886Spjd } 1814210886Spjd } 1815211886Spjd } else if ((modified & 1816211886Spjd (MODIFIED_REMOTEADDR | MODIFIED_REPLICATION)) != 0) { 1817210886Spjd for (ii = 0; ii < ncomps; ii++) { 1818210886Spjd if (!ISREMOTE(ii)) 1819210886Spjd continue; 1820210886Spjd remote_close(gres, ii); 1821210886Spjd } 1822210886Spjd if (modified & MODIFIED_REMOTEADDR) { 1823210886Spjd strlcpy(gres->hr_remoteaddr, res->hr_remoteaddr, 1824210886Spjd sizeof(gres->hr_remoteaddr)); 1825210886Spjd } 1826210886Spjd } 1827210886Spjd#undef MODIFIED_REMOTEADDR 1828210886Spjd#undef MODIFIED_REPLICATION 1829210886Spjd#undef MODIFIED_TIMEOUT 1830211886Spjd#undef MODIFIED_EXEC 1831210886Spjd 1832210886Spjd pjdlog_info("Configuration reloaded successfully."); 1833210886Spjd return; 1834210886Spjdfailed: 1835210886Spjd if (newcfg != NULL) { 1836210886Spjd if (newcfg->hc_controlconn != NULL) 1837210886Spjd proto_close(newcfg->hc_controlconn); 1838210886Spjd if (newcfg->hc_listenconn != NULL) 1839210886Spjd proto_close(newcfg->hc_listenconn); 1840210886Spjd yy_config_free(newcfg); 1841210886Spjd } 1842210886Spjd pjdlog_warning("Configuration not reloaded."); 1843210886Spjd} 1844210886Spjd 1845211882Spjdstatic void 1846211882Spjdkeepalive_send(struct hast_resource *res, unsigned int ncomp) 1847211882Spjd{ 1848211882Spjd struct nv *nv; 1849211882Spjd 1850211882Spjd nv = nv_alloc(); 1851211882Spjd nv_add_uint8(nv, HIO_KEEPALIVE, "cmd"); 1852211882Spjd if (nv_error(nv) != 0) { 1853211882Spjd nv_free(nv); 1854211882Spjd pjdlog_debug(1, 1855211882Spjd "keepalive_send: Unable to prepare header to send."); 1856211882Spjd return; 1857211882Spjd } 1858211882Spjd if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) < 0) { 1859211882Spjd pjdlog_common(LOG_DEBUG, 1, errno, 1860211882Spjd "keepalive_send: Unable to send request"); 1861211882Spjd nv_free(nv); 1862211882Spjd rw_unlock(&hio_remote_lock[ncomp]); 1863211882Spjd remote_close(res, ncomp); 1864211882Spjd rw_rlock(&hio_remote_lock[ncomp]); 1865211882Spjd return; 1866211882Spjd } 1867211882Spjd nv_free(nv); 1868211882Spjd pjdlog_debug(2, "keepalive_send: Request sent."); 1869211882Spjd} 1870211882Spjd 1871211981Spjdstatic void 1872211981Spjdguard_one(struct hast_resource *res, unsigned int ncomp) 1873211981Spjd{ 1874211981Spjd struct proto_conn *in, *out; 1875211981Spjd 1876211981Spjd if (!ISREMOTE(ncomp)) 1877211981Spjd return; 1878211981Spjd 1879211981Spjd rw_rlock(&hio_remote_lock[ncomp]); 1880211981Spjd 1881211981Spjd if (!real_remote(res)) { 1882211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1883211981Spjd return; 1884211981Spjd } 1885211981Spjd 1886211981Spjd if (ISCONNECTED(res, ncomp)) { 1887211981Spjd assert(res->hr_remotein != NULL); 1888211981Spjd assert(res->hr_remoteout != NULL); 1889211981Spjd keepalive_send(res, ncomp); 1890211981Spjd } 1891211981Spjd 1892211981Spjd if (ISCONNECTED(res, ncomp)) { 1893211981Spjd assert(res->hr_remotein != NULL); 1894211981Spjd assert(res->hr_remoteout != NULL); 1895211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1896211981Spjd pjdlog_debug(2, "remote_guard: Connection to %s is ok.", 1897211981Spjd res->hr_remoteaddr); 1898211981Spjd return; 1899211981Spjd } 1900211981Spjd 1901211981Spjd assert(res->hr_remotein == NULL); 1902211981Spjd assert(res->hr_remoteout == NULL); 1903211981Spjd /* 1904211981Spjd * Upgrade the lock. It doesn't have to be atomic as no other thread 1905211981Spjd * can change connection status from disconnected to connected. 1906211981Spjd */ 1907211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1908211981Spjd pjdlog_debug(2, "remote_guard: Reconnecting to %s.", 1909211981Spjd res->hr_remoteaddr); 1910211981Spjd in = out = NULL; 1911211981Spjd if (init_remote(res, &in, &out)) { 1912211981Spjd rw_wlock(&hio_remote_lock[ncomp]); 1913211981Spjd assert(res->hr_remotein == NULL); 1914211981Spjd assert(res->hr_remoteout == NULL); 1915211981Spjd assert(in != NULL && out != NULL); 1916211981Spjd res->hr_remotein = in; 1917211981Spjd res->hr_remoteout = out; 1918211981Spjd rw_unlock(&hio_remote_lock[ncomp]); 1919211981Spjd pjdlog_info("Successfully reconnected to %s.", 1920211981Spjd res->hr_remoteaddr); 1921211981Spjd sync_start(); 1922211981Spjd } else { 1923211981Spjd /* Both connections should be NULL. */ 1924211981Spjd assert(res->hr_remotein == NULL); 1925211981Spjd assert(res->hr_remoteout == NULL); 1926211981Spjd assert(in == NULL && out == NULL); 1927211981Spjd pjdlog_debug(2, "remote_guard: Reconnect to %s failed.", 1928211981Spjd res->hr_remoteaddr); 1929211981Spjd } 1930211981Spjd} 1931211981Spjd 1932204076Spjd/* 1933204076Spjd * Thread guards remote connections and reconnects when needed, handles 1934204076Spjd * signals, etc. 1935204076Spjd */ 1936204076Spjdstatic void * 1937204076Spjdguard_thread(void *arg) 1938204076Spjd{ 1939204076Spjd struct hast_resource *res = arg; 1940204076Spjd unsigned int ii, ncomps; 1941211982Spjd struct timespec timeout; 1942211981Spjd time_t lastcheck, now; 1943211982Spjd sigset_t mask; 1944211982Spjd int signo; 1945204076Spjd 1946204076Spjd ncomps = HAST_NCOMPONENTS; 1947211981Spjd lastcheck = time(NULL); 1948204076Spjd 1949211982Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 1950211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGHUP) == 0); 1951211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0); 1952211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0); 1953211982Spjd PJDLOG_VERIFY(sigaddset(&mask, SIGCHLD) == 0); 1954211982Spjd 1955211982Spjd timeout.tv_nsec = 0; 1956211982Spjd signo = -1; 1957211982Spjd 1958204076Spjd for (;;) { 1959211982Spjd switch (signo) { 1960211982Spjd case SIGHUP: 1961211982Spjd config_reload(); 1962211982Spjd break; 1963211982Spjd case SIGINT: 1964211982Spjd case SIGTERM: 1965211982Spjd sigexit_received = true; 1966204076Spjd primary_exitx(EX_OK, 1967204076Spjd "Termination signal received, exiting."); 1968211982Spjd break; 1969211982Spjd default: 1970211982Spjd break; 1971204076Spjd } 1972211982Spjd hook_check(signo == SIGCHLD); 1973211882Spjd 1974204076Spjd pjdlog_debug(2, "remote_guard: Checking connections."); 1975211981Spjd now = time(NULL); 1976211982Spjd if (lastcheck + RETRY_SLEEP <= now) { 1977211982Spjd for (ii = 0; ii < ncomps; ii++) 1978211981Spjd guard_one(res, ii); 1979211981Spjd lastcheck = now; 1980204076Spjd } 1981211982Spjd timeout.tv_sec = RETRY_SLEEP; 1982211982Spjd signo = sigtimedwait(&mask, NULL, &timeout); 1983204076Spjd } 1984204076Spjd /* NOTREACHED */ 1985204076Spjd return (NULL); 1986204076Spjd} 1987