secondary.c revision 218043
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 218043 2011-01-28 21:52:37Z 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> 46213009Spjd#include <signal.h> 47204076Spjd#include <stdint.h> 48204076Spjd#include <stdio.h> 49204076Spjd#include <string.h> 50204076Spjd#include <sysexits.h> 51204076Spjd#include <unistd.h> 52204076Spjd 53204076Spjd#include <activemap.h> 54204076Spjd#include <nv.h> 55204076Spjd#include <pjdlog.h> 56204076Spjd 57204076Spjd#include "control.h" 58212038Spjd#include "event.h" 59204076Spjd#include "hast.h" 60204076Spjd#include "hast_proto.h" 61204076Spjd#include "hastd.h" 62211977Spjd#include "hooks.h" 63204076Spjd#include "metadata.h" 64204076Spjd#include "proto.h" 65204076Spjd#include "subr.h" 66204076Spjd#include "synch.h" 67204076Spjd 68204076Spjdstruct hio { 69204076Spjd uint64_t hio_seq; 70204076Spjd int hio_error; 71204076Spjd struct nv *hio_nv; 72204076Spjd void *hio_data; 73204076Spjd uint8_t hio_cmd; 74204076Spjd uint64_t hio_offset; 75204076Spjd uint64_t hio_length; 76204076Spjd TAILQ_ENTRY(hio) hio_next; 77204076Spjd}; 78204076Spjd 79211984Spjdstatic struct hast_resource *gres; 80211984Spjd 81204076Spjd/* 82204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 83204076Spjd * until some in-progress requests are freed. 84204076Spjd */ 85204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 86204076Spjdstatic pthread_mutex_t hio_free_list_lock; 87204076Spjdstatic pthread_cond_t hio_free_list_cond; 88204076Spjd/* 89204076Spjd * Disk thread (the one that do I/O requests) takes requests from this list. 90204076Spjd */ 91204076Spjdstatic TAILQ_HEAD(, hio) hio_disk_list; 92204076Spjdstatic pthread_mutex_t hio_disk_list_lock; 93204076Spjdstatic pthread_cond_t hio_disk_list_cond; 94204076Spjd/* 95204076Spjd * There is one recv list for every component, although local components don't 96204076Spjd * use recv lists as local requests are done synchronously. 97204076Spjd */ 98204076Spjdstatic TAILQ_HEAD(, hio) hio_send_list; 99204076Spjdstatic pthread_mutex_t hio_send_list_lock; 100204076Spjdstatic pthread_cond_t hio_send_list_cond; 101204076Spjd 102204076Spjd/* 103204076Spjd * Maximum number of outstanding I/O requests. 104204076Spjd */ 105204076Spjd#define HAST_HIO_MAX 256 106204076Spjd 107204076Spjdstatic void *recv_thread(void *arg); 108204076Spjdstatic void *disk_thread(void *arg); 109204076Spjdstatic void *send_thread(void *arg); 110204076Spjd 111211877Spjd#define QUEUE_INSERT(name, hio) do { \ 112211877Spjd bool _wakeup; \ 113211877Spjd \ 114211877Spjd mtx_lock(&hio_##name##_list_lock); \ 115211877Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 116211877Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \ 117211877Spjd mtx_unlock(&hio_##name##_list_lock); \ 118211877Spjd if (_wakeup) \ 119211877Spjd cv_signal(&hio_##name##_list_cond); \ 120211877Spjd} while (0) 121211877Spjd#define QUEUE_TAKE(name, hio) do { \ 122211877Spjd mtx_lock(&hio_##name##_list_lock); \ 123211877Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 124211877Spjd cv_wait(&hio_##name##_list_cond, \ 125211877Spjd &hio_##name##_list_lock); \ 126211877Spjd } \ 127211877Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \ 128211877Spjd mtx_unlock(&hio_##name##_list_lock); \ 129211877Spjd} while (0) 130211877Spjd 131204076Spjdstatic void 132204076Spjdinit_environment(void) 133204076Spjd{ 134204076Spjd struct hio *hio; 135204076Spjd unsigned int ii; 136204076Spjd 137204076Spjd /* 138204076Spjd * Initialize lists, their locks and theirs condition variables. 139204076Spjd */ 140204076Spjd TAILQ_INIT(&hio_free_list); 141204076Spjd mtx_init(&hio_free_list_lock); 142204076Spjd cv_init(&hio_free_list_cond); 143204076Spjd TAILQ_INIT(&hio_disk_list); 144204076Spjd mtx_init(&hio_disk_list_lock); 145204076Spjd cv_init(&hio_disk_list_cond); 146204076Spjd TAILQ_INIT(&hio_send_list); 147204076Spjd mtx_init(&hio_send_list_lock); 148204076Spjd cv_init(&hio_send_list_cond); 149204076Spjd 150204076Spjd /* 151204076Spjd * Allocate requests pool and initialize requests. 152204076Spjd */ 153204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 154204076Spjd hio = malloc(sizeof(*hio)); 155204076Spjd if (hio == NULL) { 156210879Spjd pjdlog_exitx(EX_TEMPFAIL, 157210879Spjd "Unable to allocate memory (%zu bytes) for hio request.", 158210879Spjd sizeof(*hio)); 159204076Spjd } 160204076Spjd hio->hio_error = 0; 161204076Spjd hio->hio_data = malloc(MAXPHYS); 162204076Spjd if (hio->hio_data == NULL) { 163210879Spjd pjdlog_exitx(EX_TEMPFAIL, 164210879Spjd "Unable to allocate memory (%zu bytes) for gctl_data.", 165210879Spjd (size_t)MAXPHYS); 166204076Spjd } 167204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 168204076Spjd } 169204076Spjd} 170204076Spjd 171204076Spjdstatic void 172204076Spjdinit_local(struct hast_resource *res) 173204076Spjd{ 174204076Spjd 175204076Spjd if (metadata_read(res, true) < 0) 176204076Spjd exit(EX_NOINPUT); 177204076Spjd} 178204076Spjd 179204076Spjdstatic void 180204076Spjdinit_remote(struct hast_resource *res, struct nv *nvin) 181204076Spjd{ 182204076Spjd uint64_t resuid; 183204076Spjd struct nv *nvout; 184204076Spjd unsigned char *map; 185204076Spjd size_t mapsize; 186204076Spjd 187204076Spjd map = NULL; 188204076Spjd mapsize = 0; 189204076Spjd nvout = nv_alloc(); 190204076Spjd nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 191204076Spjd nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 192204076Spjd resuid = nv_get_uint64(nvin, "resuid"); 193204076Spjd res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 194204076Spjd res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 195204076Spjd nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 196204076Spjd nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 197204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 198204076Spjd METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 199204076Spjd map = malloc(mapsize); 200204076Spjd if (map == NULL) { 201204076Spjd pjdlog_exitx(EX_TEMPFAIL, 202204076Spjd "Unable to allocate memory (%zu bytes) for activemap.", 203204076Spjd mapsize); 204204076Spjd } 205204076Spjd nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 206204076Spjd /* 207204076Spjd * When we work as primary and secondary is missing we will increase 208204076Spjd * localcnt in our metadata. When secondary is connected and synced 209204076Spjd * we make localcnt be equal to remotecnt, which means nodes are more 210204076Spjd * or less in sync. 211204076Spjd * Split-brain condition is when both nodes are not able to communicate 212204076Spjd * and are both configured as primary nodes. In turn, they can both 213204076Spjd * make incompatible changes to the data and we have to detect that. 214204076Spjd * Under split-brain condition we will increase our localcnt on first 215204076Spjd * write and remote node will increase its localcnt on first write. 216204076Spjd * When we connect we can see that primary's localcnt is greater than 217204076Spjd * our remotecnt (primary was modified while we weren't watching) and 218204076Spjd * our localcnt is greater than primary's remotecnt (we were modified 219204076Spjd * while primary wasn't watching). 220204076Spjd * There are many possible combinations which are all gathered below. 221204076Spjd * Don't pay too much attention to exact numbers, the more important 222204076Spjd * is to compare them. We compare secondary's local with primary's 223204076Spjd * remote and secondary's remote with primary's local. 224204076Spjd * Note that every case where primary's localcnt is smaller than 225204076Spjd * secondary's remotecnt and where secondary's localcnt is smaller than 226204076Spjd * primary's remotecnt should be impossible in practise. We will perform 227204076Spjd * full synchronization then. Those cases are marked with an asterisk. 228204076Spjd * Regular synchronization means that only extents marked as dirty are 229204076Spjd * synchronized (regular synchronization). 230204076Spjd * 231204076Spjd * SECONDARY METADATA PRIMARY METADATA 232204076Spjd * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 233204076Spjd * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 234204076Spjd * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 235204076Spjd * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 236204076Spjd * regular sync from secondary. 237204076Spjd * local=3 remote=3 local=3 remote=3 Regular sync just in case. 238204076Spjd * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 239204076Spjd * local=3 remote=3 local=4 remote=2 Split-brain condition. 240204076Spjd * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 241204076Spjd * regular sync from primary. 242204076Spjd * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 243204076Spjd */ 244204076Spjd if (res->hr_resuid == 0) { 245204076Spjd /* 246214284Spjd * Provider is used for the first time. If primary node done no 247214284Spjd * writes yet as well (we will find "virgin" argument) then 248214284Spjd * there is no need to synchronize anything. If primary node 249214284Spjd * done any writes already we have to synchronize everything. 250204076Spjd */ 251204076Spjd assert(res->hr_secondary_localcnt == 0); 252204076Spjd res->hr_resuid = resuid; 253204076Spjd if (metadata_write(res) < 0) 254204076Spjd exit(EX_NOINPUT); 255214284Spjd if (nv_exists(nvin, "virgin")) { 256214284Spjd free(map); 257214284Spjd map = NULL; 258214284Spjd mapsize = 0; 259214284Spjd } else { 260214284Spjd memset(map, 0xff, mapsize); 261214284Spjd } 262204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 263204076Spjd } else if ( 264204076Spjd /* Is primary is out-of-date? */ 265204076Spjd (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 266204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 267204076Spjd /* Node are more or less in sync? */ 268204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 269204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 270204076Spjd /* Is secondary is out-of-date? */ 271204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 272204076Spjd res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 273204076Spjd /* 274204076Spjd * Nodes are more or less in sync or one of the nodes is 275204076Spjd * out-of-date. 276204076Spjd * It doesn't matter at this point which one, we just have to 277204076Spjd * send out local bitmap to the remote node. 278204076Spjd */ 279204076Spjd if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 280204076Spjd (ssize_t)mapsize) { 281204076Spjd pjdlog_exit(LOG_ERR, "Unable to read activemap"); 282204076Spjd } 283204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 284204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 285204076Spjd /* Primary is out-of-date, sync from secondary. */ 286204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 287204076Spjd } else { 288204076Spjd /* 289204076Spjd * Secondary is out-of-date or counts match. 290204076Spjd * Sync from primary. 291204076Spjd */ 292204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 293204076Spjd } 294204076Spjd } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 295204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 296204076Spjd /* 297204076Spjd * Not good, we have split-brain condition. 298204076Spjd */ 299204076Spjd pjdlog_error("Split-brain detected, exiting."); 300204076Spjd nv_add_string(nvout, "Split-brain condition!", "errmsg"); 301204076Spjd free(map); 302204076Spjd map = NULL; 303204076Spjd mapsize = 0; 304204076Spjd } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 305204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 306204076Spjd /* 307204076Spjd * This should never happen in practise, but we will perform 308204076Spjd * full synchronization. 309204076Spjd */ 310204076Spjd assert(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 311204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt); 312204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 313204076Spjd METADATA_SIZE, res->hr_extentsize, 314204076Spjd res->hr_local_sectorsize); 315204076Spjd memset(map, 0xff, mapsize); 316204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 317204076Spjd /* In this one of five cases sync from secondary. */ 318204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 319204076Spjd } else { 320204076Spjd /* For the rest four cases sync from primary. */ 321204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 322204076Spjd } 323204076Spjd pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 324204076Spjd (uintmax_t)res->hr_primary_localcnt, 325204076Spjd (uintmax_t)res->hr_primary_remotecnt, 326204076Spjd (uintmax_t)res->hr_secondary_localcnt, 327204076Spjd (uintmax_t)res->hr_secondary_remotecnt); 328204076Spjd } 329204076Spjd if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 330214276Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 331204076Spjd res->hr_remoteaddr); 332204076Spjd } 333214275Spjd if (map != NULL) 334214275Spjd free(map); 335209182Spjd nv_free(nvout); 336204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 337204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 338204076Spjd /* Exit on split-brain. */ 339212038Spjd event_send(res, EVENT_SPLITBRAIN); 340204076Spjd exit(EX_CONFIG); 341204076Spjd } 342204076Spjd} 343204076Spjd 344204076Spjdvoid 345204076Spjdhastd_secondary(struct hast_resource *res, struct nv *nvin) 346204076Spjd{ 347213009Spjd sigset_t mask; 348204076Spjd pthread_t td; 349204076Spjd pid_t pid; 350218043Spjd int error, mode; 351204076Spjd 352204076Spjd /* 353204076Spjd * Create communication channel between parent and child. 354204076Spjd */ 355204076Spjd if (proto_client("socketpair://", &res->hr_ctrl) < 0) { 356204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 357204076Spjd pjdlog_exit(EX_OSERR, 358204076Spjd "Unable to create control sockets between parent and child"); 359204076Spjd } 360212038Spjd /* 361212038Spjd * Create communication channel between child and parent. 362212038Spjd */ 363212038Spjd if (proto_client("socketpair://", &res->hr_event) < 0) { 364212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 365212038Spjd pjdlog_exit(EX_OSERR, 366212038Spjd "Unable to create event sockets between child and parent"); 367212038Spjd } 368204076Spjd 369204076Spjd pid = fork(); 370204076Spjd if (pid < 0) { 371204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 372204076Spjd pjdlog_exit(EX_OSERR, "Unable to fork"); 373204076Spjd } 374204076Spjd 375204076Spjd if (pid > 0) { 376204076Spjd /* This is parent. */ 377204076Spjd proto_close(res->hr_remotein); 378204076Spjd res->hr_remotein = NULL; 379204076Spjd proto_close(res->hr_remoteout); 380204076Spjd res->hr_remoteout = NULL; 381212038Spjd /* Declare that we are receiver. */ 382212038Spjd proto_recv(res->hr_event, NULL, 0); 383218043Spjd /* Declare that we are sender. */ 384218043Spjd proto_send(res->hr_ctrl, NULL, 0); 385204076Spjd res->hr_workerpid = pid; 386204076Spjd return; 387204076Spjd } 388211977Spjd 389211984Spjd gres = res; 390218043Spjd mode = pjdlog_mode_get(); 391211984Spjd 392218043Spjd /* Declare that we are sender. */ 393218043Spjd proto_send(res->hr_event, NULL, 0); 394218043Spjd /* Declare that we are receiver. */ 395218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 396218043Spjd descriptors_cleanup(res); 397204076Spjd 398218043Spjd pjdlog_init(mode); 399218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 400204076Spjd setproctitle("%s (secondary)", res->hr_name); 401204076Spjd 402213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 403213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 404210880Spjd 405207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 406207371Spjd if (proto_timeout(res->hr_remotein, 0) < 0) 407207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 408207371Spjd if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 409207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 410207371Spjd 411204076Spjd init_local(res); 412213007Spjd init_environment(); 413213007Spjd 414213007Spjd /* 415213007Spjd * Create the control thread before sending any event to the parent, 416213007Spjd * as we can deadlock when parent sends control request to worker, 417213007Spjd * but worker has no control thread started yet, so parent waits. 418213007Spjd * In the meantime worker sends an event to the parent, but parent 419213007Spjd * is unable to handle the event, because it waits for control 420213007Spjd * request response. 421213007Spjd */ 422213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 423213007Spjd assert(error == 0); 424213007Spjd 425204076Spjd init_remote(res, nvin); 426212038Spjd event_send(res, EVENT_CONNECT); 427204076Spjd 428204076Spjd error = pthread_create(&td, NULL, recv_thread, res); 429204076Spjd assert(error == 0); 430204076Spjd error = pthread_create(&td, NULL, disk_thread, res); 431204076Spjd assert(error == 0); 432213007Spjd (void)send_thread(res); 433204076Spjd} 434204076Spjd 435204076Spjdstatic void 436204076Spjdreqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 437204076Spjd{ 438204076Spjd char msg[1024]; 439204076Spjd va_list ap; 440204076Spjd int len; 441204076Spjd 442204076Spjd va_start(ap, fmt); 443204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 444204076Spjd va_end(ap); 445204076Spjd if ((size_t)len < sizeof(msg)) { 446204076Spjd switch (hio->hio_cmd) { 447204076Spjd case HIO_READ: 448204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 449204076Spjd "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 450204076Spjd (uintmax_t)hio->hio_length); 451204076Spjd break; 452204076Spjd case HIO_DELETE: 453204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 454204076Spjd "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 455204076Spjd (uintmax_t)hio->hio_length); 456204076Spjd break; 457204076Spjd case HIO_FLUSH: 458204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 459204076Spjd break; 460204076Spjd case HIO_WRITE: 461204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 462204076Spjd "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 463204076Spjd (uintmax_t)hio->hio_length); 464204076Spjd break; 465211882Spjd case HIO_KEEPALIVE: 466211882Spjd (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 467211882Spjd break; 468204076Spjd default: 469204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 470204076Spjd "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 471204076Spjd break; 472204076Spjd } 473204076Spjd } 474204076Spjd pjdlog_common(loglevel, debuglevel, error, "%s", msg); 475204076Spjd} 476204076Spjd 477204076Spjdstatic int 478204076Spjdrequnpack(struct hast_resource *res, struct hio *hio) 479204076Spjd{ 480204076Spjd 481204076Spjd hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 482204076Spjd if (hio->hio_cmd == 0) { 483204076Spjd pjdlog_error("Header contains no 'cmd' field."); 484204076Spjd hio->hio_error = EINVAL; 485204076Spjd goto end; 486204076Spjd } 487204076Spjd switch (hio->hio_cmd) { 488211882Spjd case HIO_KEEPALIVE: 489211882Spjd break; 490204076Spjd case HIO_READ: 491204076Spjd case HIO_WRITE: 492204076Spjd case HIO_DELETE: 493204076Spjd hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 494204076Spjd if (nv_error(hio->hio_nv) != 0) { 495204076Spjd pjdlog_error("Header is missing 'offset' field."); 496204076Spjd hio->hio_error = EINVAL; 497204076Spjd goto end; 498204076Spjd } 499204076Spjd hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 500204076Spjd if (nv_error(hio->hio_nv) != 0) { 501204076Spjd pjdlog_error("Header is missing 'length' field."); 502204076Spjd hio->hio_error = EINVAL; 503204076Spjd goto end; 504204076Spjd } 505204076Spjd if (hio->hio_length == 0) { 506204076Spjd pjdlog_error("Data length is zero."); 507204076Spjd hio->hio_error = EINVAL; 508204076Spjd goto end; 509204076Spjd } 510204076Spjd if (hio->hio_length > MAXPHYS) { 511204076Spjd pjdlog_error("Data length is too large (%ju > %ju).", 512204076Spjd (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 513204076Spjd hio->hio_error = EINVAL; 514204076Spjd goto end; 515204076Spjd } 516204076Spjd if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 517204076Spjd pjdlog_error("Offset %ju is not multiple of sector size.", 518204076Spjd (uintmax_t)hio->hio_offset); 519204076Spjd hio->hio_error = EINVAL; 520204076Spjd goto end; 521204076Spjd } 522204076Spjd if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 523204076Spjd pjdlog_error("Length %ju is not multiple of sector size.", 524204076Spjd (uintmax_t)hio->hio_length); 525204076Spjd hio->hio_error = EINVAL; 526204076Spjd goto end; 527204076Spjd } 528204076Spjd if (hio->hio_offset + hio->hio_length > 529204076Spjd (uint64_t)res->hr_datasize) { 530204076Spjd pjdlog_error("Data offset is too large (%ju > %ju).", 531204076Spjd (uintmax_t)(hio->hio_offset + hio->hio_length), 532204076Spjd (uintmax_t)res->hr_datasize); 533204076Spjd hio->hio_error = EINVAL; 534204076Spjd goto end; 535204076Spjd } 536204076Spjd break; 537204076Spjd default: 538204076Spjd pjdlog_error("Header contains invalid 'cmd' (%hhu).", 539204076Spjd hio->hio_cmd); 540204076Spjd hio->hio_error = EINVAL; 541204076Spjd goto end; 542204076Spjd } 543204076Spjd hio->hio_error = 0; 544204076Spjdend: 545204076Spjd return (hio->hio_error); 546204076Spjd} 547204076Spjd 548212899Spjdstatic __dead2 void 549211984Spjdsecondary_exit(int exitcode, const char *fmt, ...) 550211984Spjd{ 551211984Spjd va_list ap; 552211984Spjd 553211984Spjd assert(exitcode != EX_OK); 554211984Spjd va_start(ap, fmt); 555211984Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 556211984Spjd va_end(ap); 557212038Spjd event_send(gres, EVENT_DISCONNECT); 558211984Spjd exit(exitcode); 559211984Spjd} 560211984Spjd 561204076Spjd/* 562204076Spjd * Thread receives requests from the primary node. 563204076Spjd */ 564204076Spjdstatic void * 565204076Spjdrecv_thread(void *arg) 566204076Spjd{ 567204076Spjd struct hast_resource *res = arg; 568204076Spjd struct hio *hio; 569204076Spjd 570204076Spjd for (;;) { 571204076Spjd pjdlog_debug(2, "recv: Taking free request."); 572211877Spjd QUEUE_TAKE(free, hio); 573204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", hio); 574204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 575211984Spjd secondary_exit(EX_TEMPFAIL, 576204076Spjd "Unable to receive request header"); 577204076Spjd } 578211877Spjd if (requnpack(res, hio) != 0) { 579211877Spjd pjdlog_debug(2, 580211877Spjd "recv: (%p) Moving request to the send queue.", 581211877Spjd hio); 582211877Spjd QUEUE_INSERT(send, hio); 583211877Spjd continue; 584211877Spjd } 585204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, 586204076Spjd "recv: (%p) Got request header: ", hio); 587211882Spjd if (hio->hio_cmd == HIO_KEEPALIVE) { 588211882Spjd pjdlog_debug(2, 589211882Spjd "recv: (%p) Moving request to the free queue.", 590211882Spjd hio); 591211882Spjd nv_free(hio->hio_nv); 592211882Spjd QUEUE_INSERT(free, hio); 593211882Spjd continue; 594211882Spjd } else if (hio->hio_cmd == HIO_WRITE) { 595204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, 596204076Spjd hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 597211984Spjd secondary_exit(EX_TEMPFAIL, 598212051Spjd "Unable to receive request data"); 599204076Spjd } 600204076Spjd } 601204076Spjd pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 602204076Spjd hio); 603211877Spjd QUEUE_INSERT(disk, hio); 604204076Spjd } 605204076Spjd /* NOTREACHED */ 606204076Spjd return (NULL); 607204076Spjd} 608204076Spjd 609204076Spjd/* 610204076Spjd * Thread reads from or writes to local component and also handles DELETE and 611204076Spjd * FLUSH requests. 612204076Spjd */ 613204076Spjdstatic void * 614204076Spjddisk_thread(void *arg) 615204076Spjd{ 616204076Spjd struct hast_resource *res = arg; 617204076Spjd struct hio *hio; 618204076Spjd ssize_t ret; 619211877Spjd bool clear_activemap; 620204076Spjd 621204076Spjd clear_activemap = true; 622204076Spjd 623204076Spjd for (;;) { 624204076Spjd pjdlog_debug(2, "disk: Taking request."); 625211877Spjd QUEUE_TAKE(disk, hio); 626204076Spjd while (clear_activemap) { 627204076Spjd unsigned char *map; 628204076Spjd size_t mapsize; 629204076Spjd 630204076Spjd /* 631204076Spjd * When first request is received, it means that primary 632204076Spjd * already received our activemap, merged it and stored 633204076Spjd * locally. We can now safely clear our activemap. 634204076Spjd */ 635204076Spjd mapsize = 636204076Spjd activemap_calc_ondisk_size(res->hr_local_mediasize - 637204076Spjd METADATA_SIZE, res->hr_extentsize, 638204076Spjd res->hr_local_sectorsize); 639204076Spjd map = calloc(1, mapsize); 640204076Spjd if (map == NULL) { 641204076Spjd pjdlog_warning("Unable to allocate memory to clear local activemap."); 642204076Spjd break; 643204076Spjd } 644204076Spjd if (pwrite(res->hr_localfd, map, mapsize, 645204076Spjd METADATA_SIZE) != (ssize_t)mapsize) { 646204076Spjd pjdlog_errno(LOG_WARNING, 647204076Spjd "Unable to store cleared activemap"); 648204076Spjd free(map); 649204076Spjd break; 650204076Spjd } 651204076Spjd free(map); 652204076Spjd clear_activemap = false; 653204076Spjd pjdlog_debug(1, "Local activemap cleared."); 654204076Spjd } 655204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 656204076Spjd /* Handle the actual request. */ 657204076Spjd switch (hio->hio_cmd) { 658204076Spjd case HIO_READ: 659204076Spjd ret = pread(res->hr_localfd, hio->hio_data, 660204076Spjd hio->hio_length, 661204076Spjd hio->hio_offset + res->hr_localoff); 662204076Spjd if (ret < 0) 663204076Spjd hio->hio_error = errno; 664204076Spjd else if (ret != (int64_t)hio->hio_length) 665204076Spjd hio->hio_error = EIO; 666204076Spjd else 667204076Spjd hio->hio_error = 0; 668204076Spjd break; 669204076Spjd case HIO_WRITE: 670204076Spjd ret = pwrite(res->hr_localfd, hio->hio_data, 671204076Spjd hio->hio_length, 672204076Spjd hio->hio_offset + res->hr_localoff); 673204076Spjd if (ret < 0) 674204076Spjd hio->hio_error = errno; 675204076Spjd else if (ret != (int64_t)hio->hio_length) 676204076Spjd hio->hio_error = EIO; 677204076Spjd else 678204076Spjd hio->hio_error = 0; 679204076Spjd break; 680204076Spjd case HIO_DELETE: 681204076Spjd ret = g_delete(res->hr_localfd, 682204076Spjd hio->hio_offset + res->hr_localoff, 683204076Spjd hio->hio_length); 684204076Spjd if (ret < 0) 685204076Spjd hio->hio_error = errno; 686204076Spjd else 687204076Spjd hio->hio_error = 0; 688204076Spjd break; 689204076Spjd case HIO_FLUSH: 690204076Spjd ret = g_flush(res->hr_localfd); 691204076Spjd if (ret < 0) 692204076Spjd hio->hio_error = errno; 693204076Spjd else 694204076Spjd hio->hio_error = 0; 695204076Spjd break; 696204076Spjd } 697204076Spjd if (hio->hio_error != 0) { 698204076Spjd reqlog(LOG_ERR, 0, hio->hio_error, hio, 699204076Spjd "Request failed: "); 700204076Spjd } 701204076Spjd pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 702204076Spjd hio); 703211877Spjd QUEUE_INSERT(send, hio); 704204076Spjd } 705204076Spjd /* NOTREACHED */ 706204076Spjd return (NULL); 707204076Spjd} 708204076Spjd 709204076Spjd/* 710204076Spjd * Thread sends requests back to primary node. 711204076Spjd */ 712204076Spjdstatic void * 713204076Spjdsend_thread(void *arg) 714204076Spjd{ 715204076Spjd struct hast_resource *res = arg; 716204076Spjd struct nv *nvout; 717204076Spjd struct hio *hio; 718204076Spjd void *data; 719204076Spjd size_t length; 720204076Spjd 721204076Spjd for (;;) { 722204076Spjd pjdlog_debug(2, "send: Taking request."); 723211877Spjd QUEUE_TAKE(send, hio); 724204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 725204076Spjd nvout = nv_alloc(); 726204076Spjd /* Copy sequence number. */ 727204076Spjd nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 728204076Spjd switch (hio->hio_cmd) { 729204076Spjd case HIO_READ: 730204076Spjd if (hio->hio_error == 0) { 731204076Spjd data = hio->hio_data; 732204076Spjd length = hio->hio_length; 733204076Spjd break; 734204076Spjd } 735204076Spjd /* 736204076Spjd * We send no data in case of an error. 737204076Spjd */ 738204076Spjd /* FALLTHROUGH */ 739204076Spjd case HIO_DELETE: 740204076Spjd case HIO_FLUSH: 741204076Spjd case HIO_WRITE: 742204076Spjd data = NULL; 743204076Spjd length = 0; 744204076Spjd break; 745204076Spjd default: 746204076Spjd abort(); 747204076Spjd break; 748204076Spjd } 749204076Spjd if (hio->hio_error != 0) 750204076Spjd nv_add_int16(nvout, hio->hio_error, "error"); 751204076Spjd if (hast_proto_send(res, res->hr_remoteout, nvout, data, 752204076Spjd length) < 0) { 753211984Spjd secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 754204076Spjd } 755204076Spjd nv_free(nvout); 756209185Spjd pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 757204076Spjd hio); 758204076Spjd nv_free(hio->hio_nv); 759204076Spjd hio->hio_error = 0; 760211877Spjd QUEUE_INSERT(free, hio); 761204076Spjd } 762204076Spjd /* NOTREACHED */ 763204076Spjd return (NULL); 764204076Spjd} 765