secondary.c revision 219482
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 219482 2011-03-11 12:12:35Z trociny $"); 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"); 262204076Spjd } else if ( 263204076Spjd /* Is primary is out-of-date? */ 264204076Spjd (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 265204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 266204076Spjd /* Node are more or less in sync? */ 267204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 268204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 269204076Spjd /* Is secondary is out-of-date? */ 270204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 271204076Spjd res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 272204076Spjd /* 273204076Spjd * Nodes are more or less in sync or one of the nodes is 274204076Spjd * out-of-date. 275204076Spjd * It doesn't matter at this point which one, we just have to 276204076Spjd * send out local bitmap to the remote node. 277204076Spjd */ 278204076Spjd if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 279204076Spjd (ssize_t)mapsize) { 280204076Spjd pjdlog_exit(LOG_ERR, "Unable to read activemap"); 281204076Spjd } 282204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 283204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 284204076Spjd /* Primary is out-of-date, sync from secondary. */ 285204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 286204076Spjd } else { 287204076Spjd /* 288204076Spjd * Secondary is out-of-date or counts match. 289204076Spjd * Sync from primary. 290204076Spjd */ 291204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 292204076Spjd } 293204076Spjd } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 294204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 295204076Spjd /* 296204076Spjd * Not good, we have split-brain condition. 297204076Spjd */ 298204076Spjd pjdlog_error("Split-brain detected, exiting."); 299204076Spjd nv_add_string(nvout, "Split-brain condition!", "errmsg"); 300204076Spjd free(map); 301204076Spjd map = NULL; 302204076Spjd mapsize = 0; 303204076Spjd } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 304204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 305204076Spjd /* 306204076Spjd * This should never happen in practise, but we will perform 307204076Spjd * full synchronization. 308204076Spjd */ 309218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 310204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt); 311204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 312204076Spjd METADATA_SIZE, res->hr_extentsize, 313204076Spjd res->hr_local_sectorsize); 314204076Spjd memset(map, 0xff, mapsize); 315204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 316204076Spjd /* In this one of five cases sync from secondary. */ 317204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 318204076Spjd } else { 319204076Spjd /* For the rest four cases sync from primary. */ 320204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 321204076Spjd } 322204076Spjd pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 323204076Spjd (uintmax_t)res->hr_primary_localcnt, 324204076Spjd (uintmax_t)res->hr_primary_remotecnt, 325204076Spjd (uintmax_t)res->hr_secondary_localcnt, 326204076Spjd (uintmax_t)res->hr_secondary_remotecnt); 327204076Spjd } 328204076Spjd if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 329214276Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 330204076Spjd res->hr_remoteaddr); 331204076Spjd } 332214275Spjd if (map != NULL) 333214275Spjd free(map); 334209182Spjd nv_free(nvout); 335204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 336204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 337204076Spjd /* Exit on split-brain. */ 338212038Spjd event_send(res, EVENT_SPLITBRAIN); 339204076Spjd exit(EX_CONFIG); 340204076Spjd } 341204076Spjd} 342204076Spjd 343204076Spjdvoid 344204076Spjdhastd_secondary(struct hast_resource *res, struct nv *nvin) 345204076Spjd{ 346213009Spjd sigset_t mask; 347204076Spjd pthread_t td; 348204076Spjd pid_t pid; 349219482Strociny int error, mode, debuglevel; 350204076Spjd 351204076Spjd /* 352204076Spjd * Create communication channel between parent and child. 353204076Spjd */ 354204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 355204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 356204076Spjd pjdlog_exit(EX_OSERR, 357204076Spjd "Unable to create control sockets between parent and child"); 358204076Spjd } 359212038Spjd /* 360212038Spjd * Create communication channel between child and parent. 361212038Spjd */ 362212038Spjd if (proto_client("socketpair://", &res->hr_event) < 0) { 363212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 364212038Spjd pjdlog_exit(EX_OSERR, 365212038Spjd "Unable to create event sockets between child and parent"); 366212038Spjd } 367218218Spjd /* 368218218Spjd * Create communication channel for sending connection requests from 369218218Spjd * parent to child. 370218218Spjd */ 371218218Spjd if (proto_client("socketpair://", &res->hr_conn) < 0) { 372218218Spjd /* TODO: There's no need for this to be fatal error. */ 373218218Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 374218218Spjd pjdlog_exit(EX_OSERR, 375218218Spjd "Unable to create connection sockets between parent and child"); 376218218Spjd } 377204076Spjd 378204076Spjd pid = fork(); 379204076Spjd if (pid < 0) { 380204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 381204076Spjd pjdlog_exit(EX_OSERR, "Unable to fork"); 382204076Spjd } 383204076Spjd 384204076Spjd if (pid > 0) { 385204076Spjd /* This is parent. */ 386204076Spjd proto_close(res->hr_remotein); 387204076Spjd res->hr_remotein = NULL; 388204076Spjd proto_close(res->hr_remoteout); 389204076Spjd res->hr_remoteout = NULL; 390212038Spjd /* Declare that we are receiver. */ 391212038Spjd proto_recv(res->hr_event, NULL, 0); 392218043Spjd /* Declare that we are sender. */ 393218043Spjd proto_send(res->hr_ctrl, NULL, 0); 394218218Spjd proto_send(res->hr_conn, NULL, 0); 395204076Spjd res->hr_workerpid = pid; 396204076Spjd return; 397204076Spjd } 398211977Spjd 399211984Spjd gres = res; 400218043Spjd mode = pjdlog_mode_get(); 401219482Strociny debuglevel = pjdlog_debug_get(); 402211984Spjd 403218043Spjd /* Declare that we are sender. */ 404218043Spjd proto_send(res->hr_event, NULL, 0); 405218043Spjd /* Declare that we are receiver. */ 406218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 407218218Spjd proto_recv(res->hr_conn, NULL, 0); 408218043Spjd descriptors_cleanup(res); 409204076Spjd 410218045Spjd descriptors_assert(res, mode); 411218045Spjd 412218043Spjd pjdlog_init(mode); 413219482Strociny pjdlog_debug_set(debuglevel); 414218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 415204076Spjd setproctitle("%s (secondary)", res->hr_name); 416204076Spjd 417213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 418213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 419210880Spjd 420207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 421207371Spjd if (proto_timeout(res->hr_remotein, 0) < 0) 422207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 423207371Spjd if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 424207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 425207371Spjd 426204076Spjd init_local(res); 427213007Spjd init_environment(); 428213007Spjd 429218049Spjd if (drop_privs() != 0) 430218049Spjd exit(EX_CONFIG); 431218214Spjd pjdlog_info("Privileges successfully dropped."); 432218049Spjd 433213007Spjd /* 434213007Spjd * Create the control thread before sending any event to the parent, 435213007Spjd * as we can deadlock when parent sends control request to worker, 436213007Spjd * but worker has no control thread started yet, so parent waits. 437213007Spjd * In the meantime worker sends an event to the parent, but parent 438213007Spjd * is unable to handle the event, because it waits for control 439213007Spjd * request response. 440213007Spjd */ 441213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 442218138Spjd PJDLOG_ASSERT(error == 0); 443213007Spjd 444204076Spjd init_remote(res, nvin); 445212038Spjd event_send(res, EVENT_CONNECT); 446204076Spjd 447204076Spjd error = pthread_create(&td, NULL, recv_thread, res); 448218138Spjd PJDLOG_ASSERT(error == 0); 449204076Spjd error = pthread_create(&td, NULL, disk_thread, res); 450218138Spjd PJDLOG_ASSERT(error == 0); 451213007Spjd (void)send_thread(res); 452204076Spjd} 453204076Spjd 454204076Spjdstatic void 455204076Spjdreqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 456204076Spjd{ 457204076Spjd char msg[1024]; 458204076Spjd va_list ap; 459204076Spjd int len; 460204076Spjd 461204076Spjd va_start(ap, fmt); 462204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 463204076Spjd va_end(ap); 464204076Spjd if ((size_t)len < sizeof(msg)) { 465204076Spjd switch (hio->hio_cmd) { 466204076Spjd case HIO_READ: 467204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 468204076Spjd "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 469204076Spjd (uintmax_t)hio->hio_length); 470204076Spjd break; 471204076Spjd case HIO_DELETE: 472204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 473204076Spjd "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 474204076Spjd (uintmax_t)hio->hio_length); 475204076Spjd break; 476204076Spjd case HIO_FLUSH: 477204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 478204076Spjd break; 479204076Spjd case HIO_WRITE: 480204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 481204076Spjd "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 482204076Spjd (uintmax_t)hio->hio_length); 483204076Spjd break; 484211882Spjd case HIO_KEEPALIVE: 485211882Spjd (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 486211882Spjd break; 487204076Spjd default: 488204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 489204076Spjd "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 490204076Spjd break; 491204076Spjd } 492204076Spjd } 493204076Spjd pjdlog_common(loglevel, debuglevel, error, "%s", msg); 494204076Spjd} 495204076Spjd 496204076Spjdstatic int 497204076Spjdrequnpack(struct hast_resource *res, struct hio *hio) 498204076Spjd{ 499204076Spjd 500204076Spjd hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 501204076Spjd if (hio->hio_cmd == 0) { 502204076Spjd pjdlog_error("Header contains no 'cmd' field."); 503204076Spjd hio->hio_error = EINVAL; 504204076Spjd goto end; 505204076Spjd } 506204076Spjd switch (hio->hio_cmd) { 507211882Spjd case HIO_KEEPALIVE: 508211882Spjd break; 509204076Spjd case HIO_READ: 510204076Spjd case HIO_WRITE: 511204076Spjd case HIO_DELETE: 512204076Spjd hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 513204076Spjd if (nv_error(hio->hio_nv) != 0) { 514204076Spjd pjdlog_error("Header is missing 'offset' field."); 515204076Spjd hio->hio_error = EINVAL; 516204076Spjd goto end; 517204076Spjd } 518204076Spjd hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 519204076Spjd if (nv_error(hio->hio_nv) != 0) { 520204076Spjd pjdlog_error("Header is missing 'length' field."); 521204076Spjd hio->hio_error = EINVAL; 522204076Spjd goto end; 523204076Spjd } 524204076Spjd if (hio->hio_length == 0) { 525204076Spjd pjdlog_error("Data length is zero."); 526204076Spjd hio->hio_error = EINVAL; 527204076Spjd goto end; 528204076Spjd } 529204076Spjd if (hio->hio_length > MAXPHYS) { 530204076Spjd pjdlog_error("Data length is too large (%ju > %ju).", 531204076Spjd (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 532204076Spjd hio->hio_error = EINVAL; 533204076Spjd goto end; 534204076Spjd } 535204076Spjd if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 536204076Spjd pjdlog_error("Offset %ju is not multiple of sector size.", 537204076Spjd (uintmax_t)hio->hio_offset); 538204076Spjd hio->hio_error = EINVAL; 539204076Spjd goto end; 540204076Spjd } 541204076Spjd if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 542204076Spjd pjdlog_error("Length %ju is not multiple of sector size.", 543204076Spjd (uintmax_t)hio->hio_length); 544204076Spjd hio->hio_error = EINVAL; 545204076Spjd goto end; 546204076Spjd } 547204076Spjd if (hio->hio_offset + hio->hio_length > 548204076Spjd (uint64_t)res->hr_datasize) { 549204076Spjd pjdlog_error("Data offset is too large (%ju > %ju).", 550204076Spjd (uintmax_t)(hio->hio_offset + hio->hio_length), 551204076Spjd (uintmax_t)res->hr_datasize); 552204076Spjd hio->hio_error = EINVAL; 553204076Spjd goto end; 554204076Spjd } 555204076Spjd break; 556204076Spjd default: 557204076Spjd pjdlog_error("Header contains invalid 'cmd' (%hhu).", 558204076Spjd hio->hio_cmd); 559204076Spjd hio->hio_error = EINVAL; 560204076Spjd goto end; 561204076Spjd } 562204076Spjd hio->hio_error = 0; 563204076Spjdend: 564204076Spjd return (hio->hio_error); 565204076Spjd} 566204076Spjd 567212899Spjdstatic __dead2 void 568211984Spjdsecondary_exit(int exitcode, const char *fmt, ...) 569211984Spjd{ 570211984Spjd va_list ap; 571211984Spjd 572218138Spjd PJDLOG_ASSERT(exitcode != EX_OK); 573211984Spjd va_start(ap, fmt); 574211984Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 575211984Spjd va_end(ap); 576212038Spjd event_send(gres, EVENT_DISCONNECT); 577211984Spjd exit(exitcode); 578211984Spjd} 579211984Spjd 580204076Spjd/* 581204076Spjd * Thread receives requests from the primary node. 582204076Spjd */ 583204076Spjdstatic void * 584204076Spjdrecv_thread(void *arg) 585204076Spjd{ 586204076Spjd struct hast_resource *res = arg; 587204076Spjd struct hio *hio; 588204076Spjd 589204076Spjd for (;;) { 590204076Spjd pjdlog_debug(2, "recv: Taking free request."); 591211877Spjd QUEUE_TAKE(free, hio); 592204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", hio); 593204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 594211984Spjd secondary_exit(EX_TEMPFAIL, 595204076Spjd "Unable to receive request header"); 596204076Spjd } 597211877Spjd if (requnpack(res, hio) != 0) { 598211877Spjd pjdlog_debug(2, 599211877Spjd "recv: (%p) Moving request to the send queue.", 600211877Spjd hio); 601211877Spjd QUEUE_INSERT(send, hio); 602211877Spjd continue; 603211877Spjd } 604204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, 605204076Spjd "recv: (%p) Got request header: ", hio); 606211882Spjd if (hio->hio_cmd == HIO_KEEPALIVE) { 607211882Spjd pjdlog_debug(2, 608211882Spjd "recv: (%p) Moving request to the free queue.", 609211882Spjd hio); 610211882Spjd nv_free(hio->hio_nv); 611211882Spjd QUEUE_INSERT(free, hio); 612211882Spjd continue; 613211882Spjd } else if (hio->hio_cmd == HIO_WRITE) { 614204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, 615204076Spjd hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 616211984Spjd secondary_exit(EX_TEMPFAIL, 617212051Spjd "Unable to receive request data"); 618204076Spjd } 619204076Spjd } 620204076Spjd pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 621204076Spjd hio); 622211877Spjd QUEUE_INSERT(disk, hio); 623204076Spjd } 624204076Spjd /* NOTREACHED */ 625204076Spjd return (NULL); 626204076Spjd} 627204076Spjd 628204076Spjd/* 629204076Spjd * Thread reads from or writes to local component and also handles DELETE and 630204076Spjd * FLUSH requests. 631204076Spjd */ 632204076Spjdstatic void * 633204076Spjddisk_thread(void *arg) 634204076Spjd{ 635204076Spjd struct hast_resource *res = arg; 636204076Spjd struct hio *hio; 637204076Spjd ssize_t ret; 638211877Spjd bool clear_activemap; 639204076Spjd 640204076Spjd clear_activemap = true; 641204076Spjd 642204076Spjd for (;;) { 643204076Spjd pjdlog_debug(2, "disk: Taking request."); 644211877Spjd QUEUE_TAKE(disk, hio); 645204076Spjd while (clear_activemap) { 646204076Spjd unsigned char *map; 647204076Spjd size_t mapsize; 648204076Spjd 649204076Spjd /* 650204076Spjd * When first request is received, it means that primary 651204076Spjd * already received our activemap, merged it and stored 652204076Spjd * locally. We can now safely clear our activemap. 653204076Spjd */ 654204076Spjd mapsize = 655204076Spjd activemap_calc_ondisk_size(res->hr_local_mediasize - 656204076Spjd METADATA_SIZE, res->hr_extentsize, 657204076Spjd res->hr_local_sectorsize); 658204076Spjd map = calloc(1, mapsize); 659204076Spjd if (map == NULL) { 660204076Spjd pjdlog_warning("Unable to allocate memory to clear local activemap."); 661204076Spjd break; 662204076Spjd } 663204076Spjd if (pwrite(res->hr_localfd, map, mapsize, 664204076Spjd METADATA_SIZE) != (ssize_t)mapsize) { 665204076Spjd pjdlog_errno(LOG_WARNING, 666204076Spjd "Unable to store cleared activemap"); 667204076Spjd free(map); 668204076Spjd break; 669204076Spjd } 670204076Spjd free(map); 671204076Spjd clear_activemap = false; 672204076Spjd pjdlog_debug(1, "Local activemap cleared."); 673204076Spjd } 674204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 675204076Spjd /* Handle the actual request. */ 676204076Spjd switch (hio->hio_cmd) { 677204076Spjd case HIO_READ: 678204076Spjd ret = pread(res->hr_localfd, hio->hio_data, 679204076Spjd hio->hio_length, 680204076Spjd hio->hio_offset + res->hr_localoff); 681204076Spjd if (ret < 0) 682204076Spjd hio->hio_error = errno; 683204076Spjd else if (ret != (int64_t)hio->hio_length) 684204076Spjd hio->hio_error = EIO; 685204076Spjd else 686204076Spjd hio->hio_error = 0; 687204076Spjd break; 688204076Spjd case HIO_WRITE: 689204076Spjd ret = pwrite(res->hr_localfd, hio->hio_data, 690204076Spjd hio->hio_length, 691204076Spjd hio->hio_offset + res->hr_localoff); 692204076Spjd if (ret < 0) 693204076Spjd hio->hio_error = errno; 694204076Spjd else if (ret != (int64_t)hio->hio_length) 695204076Spjd hio->hio_error = EIO; 696204076Spjd else 697204076Spjd hio->hio_error = 0; 698204076Spjd break; 699204076Spjd case HIO_DELETE: 700204076Spjd ret = g_delete(res->hr_localfd, 701204076Spjd hio->hio_offset + res->hr_localoff, 702204076Spjd hio->hio_length); 703204076Spjd if (ret < 0) 704204076Spjd hio->hio_error = errno; 705204076Spjd else 706204076Spjd hio->hio_error = 0; 707204076Spjd break; 708204076Spjd case HIO_FLUSH: 709204076Spjd ret = g_flush(res->hr_localfd); 710204076Spjd if (ret < 0) 711204076Spjd hio->hio_error = errno; 712204076Spjd else 713204076Spjd hio->hio_error = 0; 714204076Spjd break; 715204076Spjd } 716204076Spjd if (hio->hio_error != 0) { 717204076Spjd reqlog(LOG_ERR, 0, hio->hio_error, hio, 718204076Spjd "Request failed: "); 719204076Spjd } 720204076Spjd pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 721204076Spjd hio); 722211877Spjd QUEUE_INSERT(send, hio); 723204076Spjd } 724204076Spjd /* NOTREACHED */ 725204076Spjd return (NULL); 726204076Spjd} 727204076Spjd 728204076Spjd/* 729204076Spjd * Thread sends requests back to primary node. 730204076Spjd */ 731204076Spjdstatic void * 732204076Spjdsend_thread(void *arg) 733204076Spjd{ 734204076Spjd struct hast_resource *res = arg; 735204076Spjd struct nv *nvout; 736204076Spjd struct hio *hio; 737204076Spjd void *data; 738204076Spjd size_t length; 739204076Spjd 740204076Spjd for (;;) { 741204076Spjd pjdlog_debug(2, "send: Taking request."); 742211877Spjd QUEUE_TAKE(send, hio); 743204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 744204076Spjd nvout = nv_alloc(); 745204076Spjd /* Copy sequence number. */ 746204076Spjd nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 747204076Spjd switch (hio->hio_cmd) { 748204076Spjd case HIO_READ: 749204076Spjd if (hio->hio_error == 0) { 750204076Spjd data = hio->hio_data; 751204076Spjd length = hio->hio_length; 752204076Spjd break; 753204076Spjd } 754204076Spjd /* 755204076Spjd * We send no data in case of an error. 756204076Spjd */ 757204076Spjd /* FALLTHROUGH */ 758204076Spjd case HIO_DELETE: 759204076Spjd case HIO_FLUSH: 760204076Spjd case HIO_WRITE: 761204076Spjd data = NULL; 762204076Spjd length = 0; 763204076Spjd break; 764204076Spjd default: 765204076Spjd abort(); 766204076Spjd break; 767204076Spjd } 768204076Spjd if (hio->hio_error != 0) 769204076Spjd nv_add_int16(nvout, hio->hio_error, "error"); 770204076Spjd if (hast_proto_send(res, res->hr_remoteout, nvout, data, 771204076Spjd length) < 0) { 772211984Spjd secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 773204076Spjd } 774204076Spjd nv_free(nvout); 775209185Spjd pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 776204076Spjd hio); 777204076Spjd nv_free(hio->hio_nv); 778204076Spjd hio->hio_error = 0; 779211877Spjd QUEUE_INSERT(free, hio); 780204076Spjd } 781204076Spjd /* NOTREACHED */ 782204076Spjd return (NULL); 783204076Spjd} 784