secondary.c revision 219830
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 219830 2011-03-21 14:50:12Z 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 <err.h> 41204076Spjd#include <errno.h> 42204076Spjd#include <fcntl.h> 43204076Spjd#include <libgeom.h> 44204076Spjd#include <pthread.h> 45213009Spjd#include <signal.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" 57212038Spjd#include "event.h" 58204076Spjd#include "hast.h" 59204076Spjd#include "hast_proto.h" 60204076Spjd#include "hastd.h" 61211977Spjd#include "hooks.h" 62204076Spjd#include "metadata.h" 63204076Spjd#include "proto.h" 64204076Spjd#include "subr.h" 65204076Spjd#include "synch.h" 66204076Spjd 67204076Spjdstruct hio { 68204076Spjd uint64_t hio_seq; 69204076Spjd int hio_error; 70204076Spjd struct nv *hio_nv; 71204076Spjd void *hio_data; 72204076Spjd uint8_t hio_cmd; 73204076Spjd uint64_t hio_offset; 74204076Spjd uint64_t hio_length; 75204076Spjd TAILQ_ENTRY(hio) hio_next; 76204076Spjd}; 77204076Spjd 78211984Spjdstatic struct hast_resource *gres; 79211984Spjd 80204076Spjd/* 81204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 82204076Spjd * until some in-progress requests are freed. 83204076Spjd */ 84204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 85204076Spjdstatic pthread_mutex_t hio_free_list_lock; 86204076Spjdstatic pthread_cond_t hio_free_list_cond; 87204076Spjd/* 88204076Spjd * Disk thread (the one that do I/O requests) takes requests from this list. 89204076Spjd */ 90204076Spjdstatic TAILQ_HEAD(, hio) hio_disk_list; 91204076Spjdstatic pthread_mutex_t hio_disk_list_lock; 92204076Spjdstatic pthread_cond_t hio_disk_list_cond; 93204076Spjd/* 94204076Spjd * There is one recv list for every component, although local components don't 95204076Spjd * use recv lists as local requests are done synchronously. 96204076Spjd */ 97204076Spjdstatic TAILQ_HEAD(, hio) hio_send_list; 98204076Spjdstatic pthread_mutex_t hio_send_list_lock; 99204076Spjdstatic pthread_cond_t hio_send_list_cond; 100204076Spjd 101204076Spjd/* 102204076Spjd * Maximum number of outstanding I/O requests. 103204076Spjd */ 104204076Spjd#define HAST_HIO_MAX 256 105204076Spjd 106204076Spjdstatic void *recv_thread(void *arg); 107204076Spjdstatic void *disk_thread(void *arg); 108204076Spjdstatic void *send_thread(void *arg); 109204076Spjd 110211877Spjd#define QUEUE_INSERT(name, hio) do { \ 111211877Spjd bool _wakeup; \ 112211877Spjd \ 113211877Spjd mtx_lock(&hio_##name##_list_lock); \ 114211877Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 115211877Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \ 116211877Spjd mtx_unlock(&hio_##name##_list_lock); \ 117211877Spjd if (_wakeup) \ 118211877Spjd cv_signal(&hio_##name##_list_cond); \ 119211877Spjd} while (0) 120211877Spjd#define QUEUE_TAKE(name, hio) do { \ 121211877Spjd mtx_lock(&hio_##name##_list_lock); \ 122211877Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 123211877Spjd cv_wait(&hio_##name##_list_cond, \ 124211877Spjd &hio_##name##_list_lock); \ 125211877Spjd } \ 126211877Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \ 127211877Spjd mtx_unlock(&hio_##name##_list_lock); \ 128211877Spjd} while (0) 129211877Spjd 130204076Spjdstatic void 131204076Spjdinit_environment(void) 132204076Spjd{ 133204076Spjd struct hio *hio; 134204076Spjd unsigned int ii; 135204076Spjd 136204076Spjd /* 137204076Spjd * Initialize lists, their locks and theirs condition variables. 138204076Spjd */ 139204076Spjd TAILQ_INIT(&hio_free_list); 140204076Spjd mtx_init(&hio_free_list_lock); 141204076Spjd cv_init(&hio_free_list_cond); 142204076Spjd TAILQ_INIT(&hio_disk_list); 143204076Spjd mtx_init(&hio_disk_list_lock); 144204076Spjd cv_init(&hio_disk_list_cond); 145204076Spjd TAILQ_INIT(&hio_send_list); 146204076Spjd mtx_init(&hio_send_list_lock); 147204076Spjd cv_init(&hio_send_list_cond); 148204076Spjd 149204076Spjd /* 150204076Spjd * Allocate requests pool and initialize requests. 151204076Spjd */ 152204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 153204076Spjd hio = malloc(sizeof(*hio)); 154204076Spjd if (hio == NULL) { 155210879Spjd pjdlog_exitx(EX_TEMPFAIL, 156210879Spjd "Unable to allocate memory (%zu bytes) for hio request.", 157210879Spjd sizeof(*hio)); 158204076Spjd } 159204076Spjd hio->hio_error = 0; 160204076Spjd hio->hio_data = malloc(MAXPHYS); 161204076Spjd if (hio->hio_data == NULL) { 162210879Spjd pjdlog_exitx(EX_TEMPFAIL, 163210879Spjd "Unable to allocate memory (%zu bytes) for gctl_data.", 164210879Spjd (size_t)MAXPHYS); 165204076Spjd } 166204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 167204076Spjd } 168204076Spjd} 169204076Spjd 170204076Spjdstatic void 171204076Spjdinit_local(struct hast_resource *res) 172204076Spjd{ 173204076Spjd 174204076Spjd if (metadata_read(res, true) < 0) 175204076Spjd exit(EX_NOINPUT); 176204076Spjd} 177204076Spjd 178204076Spjdstatic void 179204076Spjdinit_remote(struct hast_resource *res, struct nv *nvin) 180204076Spjd{ 181204076Spjd uint64_t resuid; 182204076Spjd struct nv *nvout; 183204076Spjd unsigned char *map; 184204076Spjd size_t mapsize; 185204076Spjd 186204076Spjd map = NULL; 187204076Spjd mapsize = 0; 188204076Spjd nvout = nv_alloc(); 189204076Spjd nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 190204076Spjd nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 191204076Spjd resuid = nv_get_uint64(nvin, "resuid"); 192204076Spjd res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 193204076Spjd res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 194204076Spjd nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 195204076Spjd nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 196204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 197204076Spjd METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 198204076Spjd map = malloc(mapsize); 199204076Spjd if (map == NULL) { 200204076Spjd pjdlog_exitx(EX_TEMPFAIL, 201204076Spjd "Unable to allocate memory (%zu bytes) for activemap.", 202204076Spjd mapsize); 203204076Spjd } 204204076Spjd nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 205204076Spjd /* 206204076Spjd * When we work as primary and secondary is missing we will increase 207204076Spjd * localcnt in our metadata. When secondary is connected and synced 208204076Spjd * we make localcnt be equal to remotecnt, which means nodes are more 209204076Spjd * or less in sync. 210204076Spjd * Split-brain condition is when both nodes are not able to communicate 211204076Spjd * and are both configured as primary nodes. In turn, they can both 212204076Spjd * make incompatible changes to the data and we have to detect that. 213204076Spjd * Under split-brain condition we will increase our localcnt on first 214204076Spjd * write and remote node will increase its localcnt on first write. 215204076Spjd * When we connect we can see that primary's localcnt is greater than 216204076Spjd * our remotecnt (primary was modified while we weren't watching) and 217204076Spjd * our localcnt is greater than primary's remotecnt (we were modified 218204076Spjd * while primary wasn't watching). 219204076Spjd * There are many possible combinations which are all gathered below. 220204076Spjd * Don't pay too much attention to exact numbers, the more important 221204076Spjd * is to compare them. We compare secondary's local with primary's 222204076Spjd * remote and secondary's remote with primary's local. 223204076Spjd * Note that every case where primary's localcnt is smaller than 224204076Spjd * secondary's remotecnt and where secondary's localcnt is smaller than 225204076Spjd * primary's remotecnt should be impossible in practise. We will perform 226204076Spjd * full synchronization then. Those cases are marked with an asterisk. 227204076Spjd * Regular synchronization means that only extents marked as dirty are 228204076Spjd * synchronized (regular synchronization). 229204076Spjd * 230204076Spjd * SECONDARY METADATA PRIMARY METADATA 231204076Spjd * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 232204076Spjd * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 233204076Spjd * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 234204076Spjd * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 235204076Spjd * regular sync from secondary. 236204076Spjd * local=3 remote=3 local=3 remote=3 Regular sync just in case. 237204076Spjd * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 238204076Spjd * local=3 remote=3 local=4 remote=2 Split-brain condition. 239204076Spjd * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 240204076Spjd * regular sync from primary. 241204076Spjd * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 242204076Spjd */ 243204076Spjd if (res->hr_resuid == 0) { 244204076Spjd /* 245214284Spjd * Provider is used for the first time. If primary node done no 246214284Spjd * writes yet as well (we will find "virgin" argument) then 247214284Spjd * there is no need to synchronize anything. If primary node 248214284Spjd * done any writes already we have to synchronize everything. 249204076Spjd */ 250218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt == 0); 251204076Spjd res->hr_resuid = resuid; 252204076Spjd if (metadata_write(res) < 0) 253204076Spjd exit(EX_NOINPUT); 254214284Spjd if (nv_exists(nvin, "virgin")) { 255214284Spjd free(map); 256214284Spjd map = NULL; 257214284Spjd mapsize = 0; 258214284Spjd } else { 259214284Spjd memset(map, 0xff, mapsize); 260214284Spjd } 261204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 262219830Spjd } else if (res->hr_resuid != resuid) { 263219830Spjd char errmsg[256]; 264219830Spjd 265219830Spjd (void)snprintf(errmsg, sizeof(errmsg), 266219830Spjd "Resource unique ID mismatch (primary=%ju, secondary=%ju).", 267219830Spjd (uintmax_t)resuid, (uintmax_t)res->hr_resuid); 268219830Spjd pjdlog_error("%s", errmsg); 269219830Spjd nv_add_string(nvout, errmsg, "errmsg"); 270219830Spjd if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) { 271219830Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s", 272219830Spjd res->hr_remoteaddr); 273219830Spjd } 274219830Spjd exit(EX_CONFIG); 275204076Spjd } else if ( 276204076Spjd /* Is primary is out-of-date? */ 277204076Spjd (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 278204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 279204076Spjd /* Node are more or less in sync? */ 280204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 281204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 282204076Spjd /* Is secondary is out-of-date? */ 283204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 284204076Spjd res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 285204076Spjd /* 286204076Spjd * Nodes are more or less in sync or one of the nodes is 287204076Spjd * out-of-date. 288204076Spjd * It doesn't matter at this point which one, we just have to 289204076Spjd * send out local bitmap to the remote node. 290204076Spjd */ 291204076Spjd if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 292204076Spjd (ssize_t)mapsize) { 293204076Spjd pjdlog_exit(LOG_ERR, "Unable to read activemap"); 294204076Spjd } 295204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 296204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 297204076Spjd /* Primary is out-of-date, sync from secondary. */ 298204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 299204076Spjd } else { 300204076Spjd /* 301204076Spjd * Secondary is out-of-date or counts match. 302204076Spjd * Sync from primary. 303204076Spjd */ 304204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 305204076Spjd } 306204076Spjd } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 307204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 308204076Spjd /* 309204076Spjd * Not good, we have split-brain condition. 310204076Spjd */ 311204076Spjd pjdlog_error("Split-brain detected, exiting."); 312204076Spjd nv_add_string(nvout, "Split-brain condition!", "errmsg"); 313204076Spjd free(map); 314204076Spjd map = NULL; 315204076Spjd mapsize = 0; 316204076Spjd } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 317204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 318204076Spjd /* 319204076Spjd * This should never happen in practise, but we will perform 320204076Spjd * full synchronization. 321204076Spjd */ 322218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 323204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt); 324204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 325204076Spjd METADATA_SIZE, res->hr_extentsize, 326204076Spjd res->hr_local_sectorsize); 327204076Spjd memset(map, 0xff, mapsize); 328204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 329204076Spjd /* In this one of five cases sync from secondary. */ 330204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 331204076Spjd } else { 332204076Spjd /* For the rest four cases sync from primary. */ 333204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 334204076Spjd } 335204076Spjd pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 336204076Spjd (uintmax_t)res->hr_primary_localcnt, 337204076Spjd (uintmax_t)res->hr_primary_remotecnt, 338204076Spjd (uintmax_t)res->hr_secondary_localcnt, 339204076Spjd (uintmax_t)res->hr_secondary_remotecnt); 340204076Spjd } 341204076Spjd if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 342214276Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 343204076Spjd res->hr_remoteaddr); 344204076Spjd } 345214275Spjd if (map != NULL) 346214275Spjd free(map); 347209182Spjd nv_free(nvout); 348204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 349204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 350204076Spjd /* Exit on split-brain. */ 351212038Spjd event_send(res, EVENT_SPLITBRAIN); 352204076Spjd exit(EX_CONFIG); 353204076Spjd } 354204076Spjd} 355204076Spjd 356204076Spjdvoid 357204076Spjdhastd_secondary(struct hast_resource *res, struct nv *nvin) 358204076Spjd{ 359213009Spjd sigset_t mask; 360204076Spjd pthread_t td; 361204076Spjd pid_t pid; 362219482Strociny int error, mode, debuglevel; 363204076Spjd 364204076Spjd /* 365204076Spjd * Create communication channel between parent and child. 366204076Spjd */ 367219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_ctrl) < 0) { 368204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 369204076Spjd pjdlog_exit(EX_OSERR, 370204076Spjd "Unable to create control sockets between parent and child"); 371204076Spjd } 372212038Spjd /* 373212038Spjd * Create communication channel between child and parent. 374212038Spjd */ 375219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_event) < 0) { 376212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 377212038Spjd pjdlog_exit(EX_OSERR, 378212038Spjd "Unable to create event sockets between child and parent"); 379212038Spjd } 380218218Spjd /* 381218218Spjd * Create communication channel for sending connection requests from 382218218Spjd * parent to child. 383218218Spjd */ 384219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_conn) < 0) { 385218218Spjd /* TODO: There's no need for this to be fatal error. */ 386218218Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 387218218Spjd pjdlog_exit(EX_OSERR, 388218218Spjd "Unable to create connection sockets between parent and child"); 389218218Spjd } 390204076Spjd 391204076Spjd pid = fork(); 392204076Spjd if (pid < 0) { 393204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 394204076Spjd pjdlog_exit(EX_OSERR, "Unable to fork"); 395204076Spjd } 396204076Spjd 397204076Spjd if (pid > 0) { 398204076Spjd /* This is parent. */ 399204076Spjd proto_close(res->hr_remotein); 400204076Spjd res->hr_remotein = NULL; 401204076Spjd proto_close(res->hr_remoteout); 402204076Spjd res->hr_remoteout = NULL; 403212038Spjd /* Declare that we are receiver. */ 404212038Spjd proto_recv(res->hr_event, NULL, 0); 405218043Spjd /* Declare that we are sender. */ 406218043Spjd proto_send(res->hr_ctrl, NULL, 0); 407218218Spjd proto_send(res->hr_conn, NULL, 0); 408204076Spjd res->hr_workerpid = pid; 409204076Spjd return; 410204076Spjd } 411211977Spjd 412211984Spjd gres = res; 413218043Spjd mode = pjdlog_mode_get(); 414219482Strociny debuglevel = pjdlog_debug_get(); 415211984Spjd 416218043Spjd /* Declare that we are sender. */ 417218043Spjd proto_send(res->hr_event, NULL, 0); 418218043Spjd /* Declare that we are receiver. */ 419218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 420218218Spjd proto_recv(res->hr_conn, NULL, 0); 421218043Spjd descriptors_cleanup(res); 422204076Spjd 423218045Spjd descriptors_assert(res, mode); 424218045Spjd 425218043Spjd pjdlog_init(mode); 426219482Strociny pjdlog_debug_set(debuglevel); 427218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 428204076Spjd setproctitle("%s (secondary)", res->hr_name); 429204076Spjd 430213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 431213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 432210880Spjd 433207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 434219721Strociny if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0) 435207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 436207371Spjd if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 437207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 438207371Spjd 439204076Spjd init_local(res); 440213007Spjd init_environment(); 441213007Spjd 442218049Spjd if (drop_privs() != 0) 443218049Spjd exit(EX_CONFIG); 444218214Spjd pjdlog_info("Privileges successfully dropped."); 445218049Spjd 446213007Spjd /* 447213007Spjd * Create the control thread before sending any event to the parent, 448213007Spjd * as we can deadlock when parent sends control request to worker, 449213007Spjd * but worker has no control thread started yet, so parent waits. 450213007Spjd * In the meantime worker sends an event to the parent, but parent 451213007Spjd * is unable to handle the event, because it waits for control 452213007Spjd * request response. 453213007Spjd */ 454213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 455218138Spjd PJDLOG_ASSERT(error == 0); 456213007Spjd 457204076Spjd init_remote(res, nvin); 458212038Spjd event_send(res, EVENT_CONNECT); 459204076Spjd 460204076Spjd error = pthread_create(&td, NULL, recv_thread, res); 461218138Spjd PJDLOG_ASSERT(error == 0); 462204076Spjd error = pthread_create(&td, NULL, disk_thread, res); 463218138Spjd PJDLOG_ASSERT(error == 0); 464213007Spjd (void)send_thread(res); 465204076Spjd} 466204076Spjd 467204076Spjdstatic void 468204076Spjdreqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 469204076Spjd{ 470204076Spjd char msg[1024]; 471204076Spjd va_list ap; 472204076Spjd int len; 473204076Spjd 474204076Spjd va_start(ap, fmt); 475204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 476204076Spjd va_end(ap); 477204076Spjd if ((size_t)len < sizeof(msg)) { 478204076Spjd switch (hio->hio_cmd) { 479204076Spjd case HIO_READ: 480204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 481204076Spjd "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 482204076Spjd (uintmax_t)hio->hio_length); 483204076Spjd break; 484204076Spjd case HIO_DELETE: 485204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 486204076Spjd "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 487204076Spjd (uintmax_t)hio->hio_length); 488204076Spjd break; 489204076Spjd case HIO_FLUSH: 490204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 491204076Spjd break; 492204076Spjd case HIO_WRITE: 493204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 494204076Spjd "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 495204076Spjd (uintmax_t)hio->hio_length); 496204076Spjd break; 497211882Spjd case HIO_KEEPALIVE: 498211882Spjd (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 499211882Spjd break; 500204076Spjd default: 501204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 502204076Spjd "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 503204076Spjd break; 504204076Spjd } 505204076Spjd } 506204076Spjd pjdlog_common(loglevel, debuglevel, error, "%s", msg); 507204076Spjd} 508204076Spjd 509204076Spjdstatic int 510204076Spjdrequnpack(struct hast_resource *res, struct hio *hio) 511204076Spjd{ 512204076Spjd 513204076Spjd hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 514204076Spjd if (hio->hio_cmd == 0) { 515204076Spjd pjdlog_error("Header contains no 'cmd' field."); 516204076Spjd hio->hio_error = EINVAL; 517204076Spjd goto end; 518204076Spjd } 519204076Spjd switch (hio->hio_cmd) { 520211882Spjd case HIO_KEEPALIVE: 521211882Spjd break; 522204076Spjd case HIO_READ: 523204076Spjd case HIO_WRITE: 524204076Spjd case HIO_DELETE: 525204076Spjd hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 526204076Spjd if (nv_error(hio->hio_nv) != 0) { 527204076Spjd pjdlog_error("Header is missing 'offset' field."); 528204076Spjd hio->hio_error = EINVAL; 529204076Spjd goto end; 530204076Spjd } 531204076Spjd hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 532204076Spjd if (nv_error(hio->hio_nv) != 0) { 533204076Spjd pjdlog_error("Header is missing 'length' field."); 534204076Spjd hio->hio_error = EINVAL; 535204076Spjd goto end; 536204076Spjd } 537204076Spjd if (hio->hio_length == 0) { 538204076Spjd pjdlog_error("Data length is zero."); 539204076Spjd hio->hio_error = EINVAL; 540204076Spjd goto end; 541204076Spjd } 542204076Spjd if (hio->hio_length > MAXPHYS) { 543204076Spjd pjdlog_error("Data length is too large (%ju > %ju).", 544204076Spjd (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 545204076Spjd hio->hio_error = EINVAL; 546204076Spjd goto end; 547204076Spjd } 548204076Spjd if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 549204076Spjd pjdlog_error("Offset %ju is not multiple of sector size.", 550204076Spjd (uintmax_t)hio->hio_offset); 551204076Spjd hio->hio_error = EINVAL; 552204076Spjd goto end; 553204076Spjd } 554204076Spjd if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 555204076Spjd pjdlog_error("Length %ju is not multiple of sector size.", 556204076Spjd (uintmax_t)hio->hio_length); 557204076Spjd hio->hio_error = EINVAL; 558204076Spjd goto end; 559204076Spjd } 560204076Spjd if (hio->hio_offset + hio->hio_length > 561204076Spjd (uint64_t)res->hr_datasize) { 562204076Spjd pjdlog_error("Data offset is too large (%ju > %ju).", 563204076Spjd (uintmax_t)(hio->hio_offset + hio->hio_length), 564204076Spjd (uintmax_t)res->hr_datasize); 565204076Spjd hio->hio_error = EINVAL; 566204076Spjd goto end; 567204076Spjd } 568204076Spjd break; 569204076Spjd default: 570204076Spjd pjdlog_error("Header contains invalid 'cmd' (%hhu).", 571204076Spjd hio->hio_cmd); 572204076Spjd hio->hio_error = EINVAL; 573204076Spjd goto end; 574204076Spjd } 575204076Spjd hio->hio_error = 0; 576204076Spjdend: 577204076Spjd return (hio->hio_error); 578204076Spjd} 579204076Spjd 580212899Spjdstatic __dead2 void 581211984Spjdsecondary_exit(int exitcode, const char *fmt, ...) 582211984Spjd{ 583211984Spjd va_list ap; 584211984Spjd 585218138Spjd PJDLOG_ASSERT(exitcode != EX_OK); 586211984Spjd va_start(ap, fmt); 587211984Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 588211984Spjd va_end(ap); 589212038Spjd event_send(gres, EVENT_DISCONNECT); 590211984Spjd exit(exitcode); 591211984Spjd} 592211984Spjd 593204076Spjd/* 594204076Spjd * Thread receives requests from the primary node. 595204076Spjd */ 596204076Spjdstatic void * 597204076Spjdrecv_thread(void *arg) 598204076Spjd{ 599204076Spjd struct hast_resource *res = arg; 600204076Spjd struct hio *hio; 601204076Spjd 602204076Spjd for (;;) { 603204076Spjd pjdlog_debug(2, "recv: Taking free request."); 604211877Spjd QUEUE_TAKE(free, hio); 605204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", hio); 606204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 607211984Spjd secondary_exit(EX_TEMPFAIL, 608204076Spjd "Unable to receive request header"); 609204076Spjd } 610211877Spjd if (requnpack(res, hio) != 0) { 611211877Spjd pjdlog_debug(2, 612211877Spjd "recv: (%p) Moving request to the send queue.", 613211877Spjd hio); 614211877Spjd QUEUE_INSERT(send, hio); 615211877Spjd continue; 616211877Spjd } 617204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, 618204076Spjd "recv: (%p) Got request header: ", hio); 619211882Spjd if (hio->hio_cmd == HIO_KEEPALIVE) { 620211882Spjd pjdlog_debug(2, 621211882Spjd "recv: (%p) Moving request to the free queue.", 622211882Spjd hio); 623211882Spjd nv_free(hio->hio_nv); 624211882Spjd QUEUE_INSERT(free, hio); 625211882Spjd continue; 626211882Spjd } else if (hio->hio_cmd == HIO_WRITE) { 627204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, 628204076Spjd hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 629211984Spjd secondary_exit(EX_TEMPFAIL, 630212051Spjd "Unable to receive request data"); 631204076Spjd } 632204076Spjd } 633204076Spjd pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 634204076Spjd hio); 635211877Spjd QUEUE_INSERT(disk, hio); 636204076Spjd } 637204076Spjd /* NOTREACHED */ 638204076Spjd return (NULL); 639204076Spjd} 640204076Spjd 641204076Spjd/* 642204076Spjd * Thread reads from or writes to local component and also handles DELETE and 643204076Spjd * FLUSH requests. 644204076Spjd */ 645204076Spjdstatic void * 646204076Spjddisk_thread(void *arg) 647204076Spjd{ 648204076Spjd struct hast_resource *res = arg; 649204076Spjd struct hio *hio; 650204076Spjd ssize_t ret; 651211877Spjd bool clear_activemap; 652204076Spjd 653204076Spjd clear_activemap = true; 654204076Spjd 655204076Spjd for (;;) { 656204076Spjd pjdlog_debug(2, "disk: Taking request."); 657211877Spjd QUEUE_TAKE(disk, hio); 658204076Spjd while (clear_activemap) { 659204076Spjd unsigned char *map; 660204076Spjd size_t mapsize; 661204076Spjd 662204076Spjd /* 663204076Spjd * When first request is received, it means that primary 664204076Spjd * already received our activemap, merged it and stored 665204076Spjd * locally. We can now safely clear our activemap. 666204076Spjd */ 667204076Spjd mapsize = 668204076Spjd activemap_calc_ondisk_size(res->hr_local_mediasize - 669204076Spjd METADATA_SIZE, res->hr_extentsize, 670204076Spjd res->hr_local_sectorsize); 671204076Spjd map = calloc(1, mapsize); 672204076Spjd if (map == NULL) { 673204076Spjd pjdlog_warning("Unable to allocate memory to clear local activemap."); 674204076Spjd break; 675204076Spjd } 676204076Spjd if (pwrite(res->hr_localfd, map, mapsize, 677204076Spjd METADATA_SIZE) != (ssize_t)mapsize) { 678204076Spjd pjdlog_errno(LOG_WARNING, 679204076Spjd "Unable to store cleared activemap"); 680204076Spjd free(map); 681204076Spjd break; 682204076Spjd } 683204076Spjd free(map); 684204076Spjd clear_activemap = false; 685204076Spjd pjdlog_debug(1, "Local activemap cleared."); 686204076Spjd } 687204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 688204076Spjd /* Handle the actual request. */ 689204076Spjd switch (hio->hio_cmd) { 690204076Spjd case HIO_READ: 691204076Spjd ret = pread(res->hr_localfd, hio->hio_data, 692204076Spjd hio->hio_length, 693204076Spjd hio->hio_offset + res->hr_localoff); 694204076Spjd if (ret < 0) 695204076Spjd hio->hio_error = errno; 696204076Spjd else if (ret != (int64_t)hio->hio_length) 697204076Spjd hio->hio_error = EIO; 698204076Spjd else 699204076Spjd hio->hio_error = 0; 700204076Spjd break; 701204076Spjd case HIO_WRITE: 702204076Spjd ret = pwrite(res->hr_localfd, hio->hio_data, 703204076Spjd hio->hio_length, 704204076Spjd hio->hio_offset + res->hr_localoff); 705204076Spjd if (ret < 0) 706204076Spjd hio->hio_error = errno; 707204076Spjd else if (ret != (int64_t)hio->hio_length) 708204076Spjd hio->hio_error = EIO; 709204076Spjd else 710204076Spjd hio->hio_error = 0; 711204076Spjd break; 712204076Spjd case HIO_DELETE: 713204076Spjd ret = g_delete(res->hr_localfd, 714204076Spjd hio->hio_offset + res->hr_localoff, 715204076Spjd hio->hio_length); 716204076Spjd if (ret < 0) 717204076Spjd hio->hio_error = errno; 718204076Spjd else 719204076Spjd hio->hio_error = 0; 720204076Spjd break; 721204076Spjd case HIO_FLUSH: 722204076Spjd ret = g_flush(res->hr_localfd); 723204076Spjd if (ret < 0) 724204076Spjd hio->hio_error = errno; 725204076Spjd else 726204076Spjd hio->hio_error = 0; 727204076Spjd break; 728204076Spjd } 729204076Spjd if (hio->hio_error != 0) { 730204076Spjd reqlog(LOG_ERR, 0, hio->hio_error, hio, 731204076Spjd "Request failed: "); 732204076Spjd } 733204076Spjd pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 734204076Spjd hio); 735211877Spjd QUEUE_INSERT(send, hio); 736204076Spjd } 737204076Spjd /* NOTREACHED */ 738204076Spjd return (NULL); 739204076Spjd} 740204076Spjd 741204076Spjd/* 742204076Spjd * Thread sends requests back to primary node. 743204076Spjd */ 744204076Spjdstatic void * 745204076Spjdsend_thread(void *arg) 746204076Spjd{ 747204076Spjd struct hast_resource *res = arg; 748204076Spjd struct nv *nvout; 749204076Spjd struct hio *hio; 750204076Spjd void *data; 751204076Spjd size_t length; 752204076Spjd 753204076Spjd for (;;) { 754204076Spjd pjdlog_debug(2, "send: Taking request."); 755211877Spjd QUEUE_TAKE(send, hio); 756204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 757204076Spjd nvout = nv_alloc(); 758204076Spjd /* Copy sequence number. */ 759204076Spjd nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 760204076Spjd switch (hio->hio_cmd) { 761204076Spjd case HIO_READ: 762204076Spjd if (hio->hio_error == 0) { 763204076Spjd data = hio->hio_data; 764204076Spjd length = hio->hio_length; 765204076Spjd break; 766204076Spjd } 767204076Spjd /* 768204076Spjd * We send no data in case of an error. 769204076Spjd */ 770204076Spjd /* FALLTHROUGH */ 771204076Spjd case HIO_DELETE: 772204076Spjd case HIO_FLUSH: 773204076Spjd case HIO_WRITE: 774204076Spjd data = NULL; 775204076Spjd length = 0; 776204076Spjd break; 777204076Spjd default: 778204076Spjd abort(); 779204076Spjd break; 780204076Spjd } 781204076Spjd if (hio->hio_error != 0) 782204076Spjd nv_add_int16(nvout, hio->hio_error, "error"); 783204076Spjd if (hast_proto_send(res, res->hr_remoteout, nvout, data, 784204076Spjd length) < 0) { 785211984Spjd secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 786204076Spjd } 787204076Spjd nv_free(nvout); 788209185Spjd pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 789204076Spjd hio); 790204076Spjd nv_free(hio->hio_nv); 791204076Spjd hio->hio_error = 0; 792211877Spjd QUEUE_INSERT(free, hio); 793204076Spjd } 794204076Spjd /* NOTREACHED */ 795204076Spjd return (NULL); 796204076Spjd} 797