secondary.c revision 211984
1204076Spjd/*- 2204076Spjd * Copyright (c) 2009-2010 The FreeBSD Foundation 3211877Spjd * 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/secondary.c 211984 2010-08-30 00:31:30Z pjd $"); 33204076Spjd 34204076Spjd#include <sys/param.h> 35204076Spjd#include <sys/time.h> 36204076Spjd#include <sys/bio.h> 37204076Spjd#include <sys/disk.h> 38204076Spjd#include <sys/stat.h> 39204076Spjd 40204076Spjd#include <assert.h> 41204076Spjd#include <err.h> 42204076Spjd#include <errno.h> 43204076Spjd#include <fcntl.h> 44204076Spjd#include <libgeom.h> 45204076Spjd#include <pthread.h> 46204076Spjd#include <stdint.h> 47204076Spjd#include <stdio.h> 48204076Spjd#include <string.h> 49204076Spjd#include <sysexits.h> 50204076Spjd#include <unistd.h> 51204076Spjd 52204076Spjd#include <activemap.h> 53204076Spjd#include <nv.h> 54204076Spjd#include <pjdlog.h> 55204076Spjd 56204076Spjd#include "control.h" 57204076Spjd#include "hast.h" 58204076Spjd#include "hast_proto.h" 59204076Spjd#include "hastd.h" 60211977Spjd#include "hooks.h" 61204076Spjd#include "metadata.h" 62204076Spjd#include "proto.h" 63204076Spjd#include "subr.h" 64204076Spjd#include "synch.h" 65204076Spjd 66204076Spjdstruct hio { 67204076Spjd uint64_t hio_seq; 68204076Spjd int hio_error; 69204076Spjd struct nv *hio_nv; 70204076Spjd void *hio_data; 71204076Spjd uint8_t hio_cmd; 72204076Spjd uint64_t hio_offset; 73204076Spjd uint64_t hio_length; 74204076Spjd TAILQ_ENTRY(hio) hio_next; 75204076Spjd}; 76204076Spjd 77211984Spjdstatic struct hast_resource *gres; 78211984Spjd 79204076Spjd/* 80204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 81204076Spjd * until some in-progress requests are freed. 82204076Spjd */ 83204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 84204076Spjdstatic pthread_mutex_t hio_free_list_lock; 85204076Spjdstatic pthread_cond_t hio_free_list_cond; 86204076Spjd/* 87204076Spjd * Disk thread (the one that do I/O requests) takes requests from this list. 88204076Spjd */ 89204076Spjdstatic TAILQ_HEAD(, hio) hio_disk_list; 90204076Spjdstatic pthread_mutex_t hio_disk_list_lock; 91204076Spjdstatic pthread_cond_t hio_disk_list_cond; 92204076Spjd/* 93204076Spjd * There is one recv list for every component, although local components don't 94204076Spjd * use recv lists as local requests are done synchronously. 95204076Spjd */ 96204076Spjdstatic TAILQ_HEAD(, hio) hio_send_list; 97204076Spjdstatic pthread_mutex_t hio_send_list_lock; 98204076Spjdstatic pthread_cond_t hio_send_list_cond; 99204076Spjd 100204076Spjd/* 101204076Spjd * Maximum number of outstanding I/O requests. 102204076Spjd */ 103204076Spjd#define HAST_HIO_MAX 256 104204076Spjd 105204076Spjdstatic void *recv_thread(void *arg); 106204076Spjdstatic void *disk_thread(void *arg); 107204076Spjdstatic void *send_thread(void *arg); 108204076Spjd 109211877Spjd#define QUEUE_INSERT(name, hio) do { \ 110211877Spjd bool _wakeup; \ 111211877Spjd \ 112211877Spjd mtx_lock(&hio_##name##_list_lock); \ 113211877Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 114211877Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \ 115211877Spjd mtx_unlock(&hio_##name##_list_lock); \ 116211877Spjd if (_wakeup) \ 117211877Spjd cv_signal(&hio_##name##_list_cond); \ 118211877Spjd} while (0) 119211877Spjd#define QUEUE_TAKE(name, hio) do { \ 120211877Spjd mtx_lock(&hio_##name##_list_lock); \ 121211877Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 122211877Spjd cv_wait(&hio_##name##_list_cond, \ 123211877Spjd &hio_##name##_list_lock); \ 124211877Spjd } \ 125211877Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \ 126211877Spjd mtx_unlock(&hio_##name##_list_lock); \ 127211877Spjd} while (0) 128211877Spjd 129204076Spjdstatic void 130204076Spjdinit_environment(void) 131204076Spjd{ 132204076Spjd struct hio *hio; 133204076Spjd unsigned int ii; 134204076Spjd 135204076Spjd /* 136204076Spjd * Initialize lists, their locks and theirs condition variables. 137204076Spjd */ 138204076Spjd TAILQ_INIT(&hio_free_list); 139204076Spjd mtx_init(&hio_free_list_lock); 140204076Spjd cv_init(&hio_free_list_cond); 141204076Spjd TAILQ_INIT(&hio_disk_list); 142204076Spjd mtx_init(&hio_disk_list_lock); 143204076Spjd cv_init(&hio_disk_list_cond); 144204076Spjd TAILQ_INIT(&hio_send_list); 145204076Spjd mtx_init(&hio_send_list_lock); 146204076Spjd cv_init(&hio_send_list_cond); 147204076Spjd 148204076Spjd /* 149204076Spjd * Allocate requests pool and initialize requests. 150204076Spjd */ 151204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 152204076Spjd hio = malloc(sizeof(*hio)); 153204076Spjd if (hio == NULL) { 154210879Spjd pjdlog_exitx(EX_TEMPFAIL, 155210879Spjd "Unable to allocate memory (%zu bytes) for hio request.", 156210879Spjd sizeof(*hio)); 157204076Spjd } 158204076Spjd hio->hio_error = 0; 159204076Spjd hio->hio_data = malloc(MAXPHYS); 160204076Spjd if (hio->hio_data == NULL) { 161210879Spjd pjdlog_exitx(EX_TEMPFAIL, 162210879Spjd "Unable to allocate memory (%zu bytes) for gctl_data.", 163210879Spjd (size_t)MAXPHYS); 164204076Spjd } 165204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 166204076Spjd } 167204076Spjd} 168204076Spjd 169204076Spjdstatic void 170204076Spjdinit_local(struct hast_resource *res) 171204076Spjd{ 172204076Spjd 173204076Spjd if (metadata_read(res, true) < 0) 174204076Spjd exit(EX_NOINPUT); 175204076Spjd} 176204076Spjd 177204076Spjdstatic void 178204076Spjdinit_remote(struct hast_resource *res, struct nv *nvin) 179204076Spjd{ 180204076Spjd uint64_t resuid; 181204076Spjd struct nv *nvout; 182204076Spjd unsigned char *map; 183204076Spjd size_t mapsize; 184204076Spjd 185204076Spjd map = NULL; 186204076Spjd mapsize = 0; 187204076Spjd nvout = nv_alloc(); 188204076Spjd nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 189204076Spjd nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 190204076Spjd resuid = nv_get_uint64(nvin, "resuid"); 191204076Spjd res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 192204076Spjd res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 193204076Spjd nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 194204076Spjd nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 195204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 196204076Spjd METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 197204076Spjd map = malloc(mapsize); 198204076Spjd if (map == NULL) { 199204076Spjd pjdlog_exitx(EX_TEMPFAIL, 200204076Spjd "Unable to allocate memory (%zu bytes) for activemap.", 201204076Spjd mapsize); 202204076Spjd } 203204076Spjd nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 204204076Spjd /* 205204076Spjd * When we work as primary and secondary is missing we will increase 206204076Spjd * localcnt in our metadata. When secondary is connected and synced 207204076Spjd * we make localcnt be equal to remotecnt, which means nodes are more 208204076Spjd * or less in sync. 209204076Spjd * Split-brain condition is when both nodes are not able to communicate 210204076Spjd * and are both configured as primary nodes. In turn, they can both 211204076Spjd * make incompatible changes to the data and we have to detect that. 212204076Spjd * Under split-brain condition we will increase our localcnt on first 213204076Spjd * write and remote node will increase its localcnt on first write. 214204076Spjd * When we connect we can see that primary's localcnt is greater than 215204076Spjd * our remotecnt (primary was modified while we weren't watching) and 216204076Spjd * our localcnt is greater than primary's remotecnt (we were modified 217204076Spjd * while primary wasn't watching). 218204076Spjd * There are many possible combinations which are all gathered below. 219204076Spjd * Don't pay too much attention to exact numbers, the more important 220204076Spjd * is to compare them. We compare secondary's local with primary's 221204076Spjd * remote and secondary's remote with primary's local. 222204076Spjd * Note that every case where primary's localcnt is smaller than 223204076Spjd * secondary's remotecnt and where secondary's localcnt is smaller than 224204076Spjd * primary's remotecnt should be impossible in practise. We will perform 225204076Spjd * full synchronization then. Those cases are marked with an asterisk. 226204076Spjd * Regular synchronization means that only extents marked as dirty are 227204076Spjd * synchronized (regular synchronization). 228204076Spjd * 229204076Spjd * SECONDARY METADATA PRIMARY METADATA 230204076Spjd * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 231204076Spjd * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 232204076Spjd * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 233204076Spjd * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 234204076Spjd * regular sync from secondary. 235204076Spjd * local=3 remote=3 local=3 remote=3 Regular sync just in case. 236204076Spjd * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 237204076Spjd * local=3 remote=3 local=4 remote=2 Split-brain condition. 238204076Spjd * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 239204076Spjd * regular sync from primary. 240204076Spjd * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 241204076Spjd */ 242204076Spjd if (res->hr_resuid == 0) { 243204076Spjd /* 244204076Spjd * Provider is used for the first time. Initialize everything. 245204076Spjd */ 246204076Spjd assert(res->hr_secondary_localcnt == 0); 247204076Spjd res->hr_resuid = resuid; 248204076Spjd if (metadata_write(res) < 0) 249204076Spjd exit(EX_NOINPUT); 250204076Spjd memset(map, 0xff, mapsize); 251204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 252204076Spjd } else if ( 253204076Spjd /* Is primary is out-of-date? */ 254204076Spjd (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 255204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 256204076Spjd /* Node are more or less in sync? */ 257204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 258204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 259204076Spjd /* Is secondary is out-of-date? */ 260204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 261204076Spjd res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 262204076Spjd /* 263204076Spjd * Nodes are more or less in sync or one of the nodes is 264204076Spjd * out-of-date. 265204076Spjd * It doesn't matter at this point which one, we just have to 266204076Spjd * send out local bitmap to the remote node. 267204076Spjd */ 268204076Spjd if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 269204076Spjd (ssize_t)mapsize) { 270204076Spjd pjdlog_exit(LOG_ERR, "Unable to read activemap"); 271204076Spjd } 272204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 273204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 274204076Spjd /* Primary is out-of-date, sync from secondary. */ 275204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 276204076Spjd } else { 277204076Spjd /* 278204076Spjd * Secondary is out-of-date or counts match. 279204076Spjd * Sync from primary. 280204076Spjd */ 281204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 282204076Spjd } 283204076Spjd } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 284204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 285204076Spjd /* 286204076Spjd * Not good, we have split-brain condition. 287204076Spjd */ 288204076Spjd pjdlog_error("Split-brain detected, exiting."); 289204076Spjd nv_add_string(nvout, "Split-brain condition!", "errmsg"); 290204076Spjd free(map); 291204076Spjd map = NULL; 292204076Spjd mapsize = 0; 293204076Spjd } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 294204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 295204076Spjd /* 296204076Spjd * This should never happen in practise, but we will perform 297204076Spjd * full synchronization. 298204076Spjd */ 299204076Spjd assert(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 300204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt); 301204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 302204076Spjd METADATA_SIZE, res->hr_extentsize, 303204076Spjd res->hr_local_sectorsize); 304204076Spjd memset(map, 0xff, mapsize); 305204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 306204076Spjd /* In this one of five cases sync from secondary. */ 307204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 308204076Spjd } else { 309204076Spjd /* For the rest four cases sync from primary. */ 310204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 311204076Spjd } 312204076Spjd pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 313204076Spjd (uintmax_t)res->hr_primary_localcnt, 314204076Spjd (uintmax_t)res->hr_primary_remotecnt, 315204076Spjd (uintmax_t)res->hr_secondary_localcnt, 316204076Spjd (uintmax_t)res->hr_secondary_remotecnt); 317204076Spjd } 318204076Spjd if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 319204076Spjd pjdlog_errno(LOG_WARNING, "Unable to send activemap to %s", 320204076Spjd res->hr_remoteaddr); 321204076Spjd nv_free(nvout); 322204076Spjd exit(EX_TEMPFAIL); 323204076Spjd } 324209182Spjd nv_free(nvout); 325204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 326204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 327204076Spjd /* Exit on split-brain. */ 328211983Spjd hook_exec(res->hr_exec, "split-brain", res->hr_name, NULL); 329204076Spjd exit(EX_CONFIG); 330204076Spjd } 331204076Spjd} 332204076Spjd 333204076Spjdvoid 334204076Spjdhastd_secondary(struct hast_resource *res, struct nv *nvin) 335204076Spjd{ 336204076Spjd pthread_t td; 337204076Spjd pid_t pid; 338204076Spjd int error; 339204076Spjd 340204076Spjd /* 341204076Spjd * Create communication channel between parent and child. 342204076Spjd */ 343204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 344204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 345204076Spjd pjdlog_exit(EX_OSERR, 346204076Spjd "Unable to create control sockets between parent and child"); 347204076Spjd } 348204076Spjd 349204076Spjd pid = fork(); 350204076Spjd if (pid < 0) { 351204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 352204076Spjd pjdlog_exit(EX_OSERR, "Unable to fork"); 353204076Spjd } 354204076Spjd 355204076Spjd if (pid > 0) { 356204076Spjd /* This is parent. */ 357204076Spjd proto_close(res->hr_remotein); 358204076Spjd res->hr_remotein = NULL; 359204076Spjd proto_close(res->hr_remoteout); 360204076Spjd res->hr_remoteout = NULL; 361204076Spjd res->hr_workerpid = pid; 362204076Spjd return; 363204076Spjd } 364211977Spjd 365211984Spjd gres = res; 366211984Spjd 367204076Spjd (void)pidfile_close(pfh); 368211977Spjd hook_fini(); 369204076Spjd 370204076Spjd setproctitle("%s (secondary)", res->hr_name); 371204076Spjd 372210880Spjd signal(SIGHUP, SIG_DFL); 373210880Spjd signal(SIGCHLD, SIG_DFL); 374210880Spjd 375207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 376207371Spjd if (proto_timeout(res->hr_remotein, 0) < 0) 377207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 378207371Spjd if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 379207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 380207371Spjd 381211983Spjd hook_init(); 382204076Spjd init_local(res); 383204076Spjd init_remote(res, nvin); 384204076Spjd init_environment(); 385211984Spjd hook_exec(res->hr_exec, "connect", res->hr_name, NULL); 386204076Spjd 387204076Spjd error = pthread_create(&td, NULL, recv_thread, res); 388204076Spjd assert(error == 0); 389204076Spjd error = pthread_create(&td, NULL, disk_thread, res); 390204076Spjd assert(error == 0); 391204076Spjd error = pthread_create(&td, NULL, send_thread, res); 392204076Spjd assert(error == 0); 393204076Spjd (void)ctrl_thread(res); 394204076Spjd} 395204076Spjd 396204076Spjdstatic void 397204076Spjdreqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 398204076Spjd{ 399204076Spjd char msg[1024]; 400204076Spjd va_list ap; 401204076Spjd int len; 402204076Spjd 403204076Spjd va_start(ap, fmt); 404204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 405204076Spjd va_end(ap); 406204076Spjd if ((size_t)len < sizeof(msg)) { 407204076Spjd switch (hio->hio_cmd) { 408204076Spjd case HIO_READ: 409204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 410204076Spjd "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 411204076Spjd (uintmax_t)hio->hio_length); 412204076Spjd break; 413204076Spjd case HIO_DELETE: 414204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 415204076Spjd "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 416204076Spjd (uintmax_t)hio->hio_length); 417204076Spjd break; 418204076Spjd case HIO_FLUSH: 419204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 420204076Spjd break; 421204076Spjd case HIO_WRITE: 422204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 423204076Spjd "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 424204076Spjd (uintmax_t)hio->hio_length); 425204076Spjd break; 426211882Spjd case HIO_KEEPALIVE: 427211882Spjd (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 428211882Spjd break; 429204076Spjd default: 430204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 431204076Spjd "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 432204076Spjd break; 433204076Spjd } 434204076Spjd } 435204076Spjd pjdlog_common(loglevel, debuglevel, error, "%s", msg); 436204076Spjd} 437204076Spjd 438204076Spjdstatic int 439204076Spjdrequnpack(struct hast_resource *res, struct hio *hio) 440204076Spjd{ 441204076Spjd 442204076Spjd hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 443204076Spjd if (hio->hio_cmd == 0) { 444204076Spjd pjdlog_error("Header contains no 'cmd' field."); 445204076Spjd hio->hio_error = EINVAL; 446204076Spjd goto end; 447204076Spjd } 448204076Spjd switch (hio->hio_cmd) { 449211882Spjd case HIO_KEEPALIVE: 450211882Spjd break; 451204076Spjd case HIO_READ: 452204076Spjd case HIO_WRITE: 453204076Spjd case HIO_DELETE: 454204076Spjd hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 455204076Spjd if (nv_error(hio->hio_nv) != 0) { 456204076Spjd pjdlog_error("Header is missing 'offset' field."); 457204076Spjd hio->hio_error = EINVAL; 458204076Spjd goto end; 459204076Spjd } 460204076Spjd hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 461204076Spjd if (nv_error(hio->hio_nv) != 0) { 462204076Spjd pjdlog_error("Header is missing 'length' field."); 463204076Spjd hio->hio_error = EINVAL; 464204076Spjd goto end; 465204076Spjd } 466204076Spjd if (hio->hio_length == 0) { 467204076Spjd pjdlog_error("Data length is zero."); 468204076Spjd hio->hio_error = EINVAL; 469204076Spjd goto end; 470204076Spjd } 471204076Spjd if (hio->hio_length > MAXPHYS) { 472204076Spjd pjdlog_error("Data length is too large (%ju > %ju).", 473204076Spjd (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 474204076Spjd hio->hio_error = EINVAL; 475204076Spjd goto end; 476204076Spjd } 477204076Spjd if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 478204076Spjd pjdlog_error("Offset %ju is not multiple of sector size.", 479204076Spjd (uintmax_t)hio->hio_offset); 480204076Spjd hio->hio_error = EINVAL; 481204076Spjd goto end; 482204076Spjd } 483204076Spjd if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 484204076Spjd pjdlog_error("Length %ju is not multiple of sector size.", 485204076Spjd (uintmax_t)hio->hio_length); 486204076Spjd hio->hio_error = EINVAL; 487204076Spjd goto end; 488204076Spjd } 489204076Spjd if (hio->hio_offset + hio->hio_length > 490204076Spjd (uint64_t)res->hr_datasize) { 491204076Spjd pjdlog_error("Data offset is too large (%ju > %ju).", 492204076Spjd (uintmax_t)(hio->hio_offset + hio->hio_length), 493204076Spjd (uintmax_t)res->hr_datasize); 494204076Spjd hio->hio_error = EINVAL; 495204076Spjd goto end; 496204076Spjd } 497204076Spjd break; 498204076Spjd default: 499204076Spjd pjdlog_error("Header contains invalid 'cmd' (%hhu).", 500204076Spjd hio->hio_cmd); 501204076Spjd hio->hio_error = EINVAL; 502204076Spjd goto end; 503204076Spjd } 504204076Spjd hio->hio_error = 0; 505204076Spjdend: 506204076Spjd return (hio->hio_error); 507204076Spjd} 508204076Spjd 509211984Spjdstatic void 510211984Spjdsecondary_exit(int exitcode, const char *fmt, ...) 511211984Spjd{ 512211984Spjd va_list ap; 513211984Spjd 514211984Spjd assert(exitcode != EX_OK); 515211984Spjd va_start(ap, fmt); 516211984Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 517211984Spjd va_end(ap); 518211984Spjd hook_exec(gres->hr_exec, "disconnect", gres->hr_name, NULL); 519211984Spjd exit(exitcode); 520211984Spjd} 521211984Spjd 522204076Spjd/* 523204076Spjd * Thread receives requests from the primary node. 524204076Spjd */ 525204076Spjdstatic void * 526204076Spjdrecv_thread(void *arg) 527204076Spjd{ 528204076Spjd struct hast_resource *res = arg; 529204076Spjd struct hio *hio; 530204076Spjd 531204076Spjd for (;;) { 532204076Spjd pjdlog_debug(2, "recv: Taking free request."); 533211877Spjd QUEUE_TAKE(free, hio); 534204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", hio); 535204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 536211984Spjd secondary_exit(EX_TEMPFAIL, 537204076Spjd "Unable to receive request header"); 538204076Spjd } 539211877Spjd if (requnpack(res, hio) != 0) { 540211877Spjd pjdlog_debug(2, 541211877Spjd "recv: (%p) Moving request to the send queue.", 542211877Spjd hio); 543211877Spjd QUEUE_INSERT(send, hio); 544211877Spjd continue; 545211877Spjd } 546204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, 547204076Spjd "recv: (%p) Got request header: ", hio); 548211882Spjd if (hio->hio_cmd == HIO_KEEPALIVE) { 549211882Spjd pjdlog_debug(2, 550211882Spjd "recv: (%p) Moving request to the free queue.", 551211882Spjd hio); 552211882Spjd nv_free(hio->hio_nv); 553211882Spjd QUEUE_INSERT(free, hio); 554211882Spjd continue; 555211882Spjd } else if (hio->hio_cmd == HIO_WRITE) { 556204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, 557204076Spjd hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 558211984Spjd secondary_exit(EX_TEMPFAIL, 559204076Spjd "Unable to receive reply data"); 560204076Spjd } 561204076Spjd } 562204076Spjd pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 563204076Spjd hio); 564211877Spjd QUEUE_INSERT(disk, hio); 565204076Spjd } 566204076Spjd /* NOTREACHED */ 567204076Spjd return (NULL); 568204076Spjd} 569204076Spjd 570204076Spjd/* 571204076Spjd * Thread reads from or writes to local component and also handles DELETE and 572204076Spjd * FLUSH requests. 573204076Spjd */ 574204076Spjdstatic void * 575204076Spjddisk_thread(void *arg) 576204076Spjd{ 577204076Spjd struct hast_resource *res = arg; 578204076Spjd struct hio *hio; 579204076Spjd ssize_t ret; 580211877Spjd bool clear_activemap; 581204076Spjd 582204076Spjd clear_activemap = true; 583204076Spjd 584204076Spjd for (;;) { 585204076Spjd pjdlog_debug(2, "disk: Taking request."); 586211877Spjd QUEUE_TAKE(disk, hio); 587204076Spjd while (clear_activemap) { 588204076Spjd unsigned char *map; 589204076Spjd size_t mapsize; 590204076Spjd 591204076Spjd /* 592204076Spjd * When first request is received, it means that primary 593204076Spjd * already received our activemap, merged it and stored 594204076Spjd * locally. We can now safely clear our activemap. 595204076Spjd */ 596204076Spjd mapsize = 597204076Spjd activemap_calc_ondisk_size(res->hr_local_mediasize - 598204076Spjd METADATA_SIZE, res->hr_extentsize, 599204076Spjd res->hr_local_sectorsize); 600204076Spjd map = calloc(1, mapsize); 601204076Spjd if (map == NULL) { 602204076Spjd pjdlog_warning("Unable to allocate memory to clear local activemap."); 603204076Spjd break; 604204076Spjd } 605204076Spjd if (pwrite(res->hr_localfd, map, mapsize, 606204076Spjd METADATA_SIZE) != (ssize_t)mapsize) { 607204076Spjd pjdlog_errno(LOG_WARNING, 608204076Spjd "Unable to store cleared activemap"); 609204076Spjd free(map); 610204076Spjd break; 611204076Spjd } 612204076Spjd free(map); 613204076Spjd clear_activemap = false; 614204076Spjd pjdlog_debug(1, "Local activemap cleared."); 615204076Spjd } 616204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 617204076Spjd /* Handle the actual request. */ 618204076Spjd switch (hio->hio_cmd) { 619204076Spjd case HIO_READ: 620204076Spjd ret = pread(res->hr_localfd, hio->hio_data, 621204076Spjd hio->hio_length, 622204076Spjd hio->hio_offset + res->hr_localoff); 623204076Spjd if (ret < 0) 624204076Spjd hio->hio_error = errno; 625204076Spjd else if (ret != (int64_t)hio->hio_length) 626204076Spjd hio->hio_error = EIO; 627204076Spjd else 628204076Spjd hio->hio_error = 0; 629204076Spjd break; 630204076Spjd case HIO_WRITE: 631204076Spjd ret = pwrite(res->hr_localfd, hio->hio_data, 632204076Spjd hio->hio_length, 633204076Spjd hio->hio_offset + res->hr_localoff); 634204076Spjd if (ret < 0) 635204076Spjd hio->hio_error = errno; 636204076Spjd else if (ret != (int64_t)hio->hio_length) 637204076Spjd hio->hio_error = EIO; 638204076Spjd else 639204076Spjd hio->hio_error = 0; 640204076Spjd break; 641204076Spjd case HIO_DELETE: 642204076Spjd ret = g_delete(res->hr_localfd, 643204076Spjd hio->hio_offset + res->hr_localoff, 644204076Spjd hio->hio_length); 645204076Spjd if (ret < 0) 646204076Spjd hio->hio_error = errno; 647204076Spjd else 648204076Spjd hio->hio_error = 0; 649204076Spjd break; 650204076Spjd case HIO_FLUSH: 651204076Spjd ret = g_flush(res->hr_localfd); 652204076Spjd if (ret < 0) 653204076Spjd hio->hio_error = errno; 654204076Spjd else 655204076Spjd hio->hio_error = 0; 656204076Spjd break; 657204076Spjd } 658204076Spjd if (hio->hio_error != 0) { 659204076Spjd reqlog(LOG_ERR, 0, hio->hio_error, hio, 660204076Spjd "Request failed: "); 661204076Spjd } 662204076Spjd pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 663204076Spjd hio); 664211877Spjd QUEUE_INSERT(send, hio); 665204076Spjd } 666204076Spjd /* NOTREACHED */ 667204076Spjd return (NULL); 668204076Spjd} 669204076Spjd 670204076Spjd/* 671204076Spjd * Thread sends requests back to primary node. 672204076Spjd */ 673204076Spjdstatic void * 674204076Spjdsend_thread(void *arg) 675204076Spjd{ 676204076Spjd struct hast_resource *res = arg; 677204076Spjd struct nv *nvout; 678204076Spjd struct hio *hio; 679204076Spjd void *data; 680204076Spjd size_t length; 681204076Spjd 682204076Spjd for (;;) { 683204076Spjd pjdlog_debug(2, "send: Taking request."); 684211877Spjd QUEUE_TAKE(send, hio); 685204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 686204076Spjd nvout = nv_alloc(); 687204076Spjd /* Copy sequence number. */ 688204076Spjd nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 689204076Spjd switch (hio->hio_cmd) { 690204076Spjd case HIO_READ: 691204076Spjd if (hio->hio_error == 0) { 692204076Spjd data = hio->hio_data; 693204076Spjd length = hio->hio_length; 694204076Spjd break; 695204076Spjd } 696204076Spjd /* 697204076Spjd * We send no data in case of an error. 698204076Spjd */ 699204076Spjd /* FALLTHROUGH */ 700204076Spjd case HIO_DELETE: 701204076Spjd case HIO_FLUSH: 702204076Spjd case HIO_WRITE: 703204076Spjd data = NULL; 704204076Spjd length = 0; 705204076Spjd break; 706204076Spjd default: 707204076Spjd abort(); 708204076Spjd break; 709204076Spjd } 710204076Spjd if (hio->hio_error != 0) 711204076Spjd nv_add_int16(nvout, hio->hio_error, "error"); 712204076Spjd if (hast_proto_send(res, res->hr_remoteout, nvout, data, 713204076Spjd length) < 0) { 714211984Spjd secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 715204076Spjd } 716204076Spjd nv_free(nvout); 717209185Spjd pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 718204076Spjd hio); 719204076Spjd nv_free(hio->hio_nv); 720204076Spjd hio->hio_error = 0; 721211877Spjd QUEUE_INSERT(free, hio); 722204076Spjd } 723204076Spjd /* NOTREACHED */ 724204076Spjd return (NULL); 725204076Spjd} 726