secondary.c revision 220271
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 220271 2011-04-02 09:25:13Z 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 { 68219864Spjd uint64_t hio_seq; 69219864Spjd 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 186220271Spjd /* Setup direction. */ 187220271Spjd if (proto_send(res->hr_remoteout, NULL, 0) == -1) 188220271Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 189220271Spjd 190204076Spjd map = NULL; 191204076Spjd mapsize = 0; 192204076Spjd nvout = nv_alloc(); 193204076Spjd nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 194204076Spjd nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 195204076Spjd resuid = nv_get_uint64(nvin, "resuid"); 196204076Spjd res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 197204076Spjd res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 198204076Spjd nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 199204076Spjd nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 200204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 201204076Spjd METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 202204076Spjd map = malloc(mapsize); 203204076Spjd if (map == NULL) { 204204076Spjd pjdlog_exitx(EX_TEMPFAIL, 205204076Spjd "Unable to allocate memory (%zu bytes) for activemap.", 206204076Spjd mapsize); 207204076Spjd } 208204076Spjd /* 209204076Spjd * When we work as primary and secondary is missing we will increase 210204076Spjd * localcnt in our metadata. When secondary is connected and synced 211204076Spjd * we make localcnt be equal to remotecnt, which means nodes are more 212204076Spjd * or less in sync. 213204076Spjd * Split-brain condition is when both nodes are not able to communicate 214204076Spjd * and are both configured as primary nodes. In turn, they can both 215204076Spjd * make incompatible changes to the data and we have to detect that. 216204076Spjd * Under split-brain condition we will increase our localcnt on first 217204076Spjd * write and remote node will increase its localcnt on first write. 218204076Spjd * When we connect we can see that primary's localcnt is greater than 219204076Spjd * our remotecnt (primary was modified while we weren't watching) and 220204076Spjd * our localcnt is greater than primary's remotecnt (we were modified 221204076Spjd * while primary wasn't watching). 222204076Spjd * There are many possible combinations which are all gathered below. 223204076Spjd * Don't pay too much attention to exact numbers, the more important 224204076Spjd * is to compare them. We compare secondary's local with primary's 225204076Spjd * remote and secondary's remote with primary's local. 226204076Spjd * Note that every case where primary's localcnt is smaller than 227204076Spjd * secondary's remotecnt and where secondary's localcnt is smaller than 228204076Spjd * primary's remotecnt should be impossible in practise. We will perform 229204076Spjd * full synchronization then. Those cases are marked with an asterisk. 230204076Spjd * Regular synchronization means that only extents marked as dirty are 231204076Spjd * synchronized (regular synchronization). 232204076Spjd * 233204076Spjd * SECONDARY METADATA PRIMARY METADATA 234204076Spjd * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 235204076Spjd * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 236204076Spjd * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 237204076Spjd * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 238204076Spjd * regular sync from secondary. 239204076Spjd * local=3 remote=3 local=3 remote=3 Regular sync just in case. 240204076Spjd * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 241204076Spjd * local=3 remote=3 local=4 remote=2 Split-brain condition. 242204076Spjd * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 243204076Spjd * regular sync from primary. 244204076Spjd * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 245204076Spjd */ 246204076Spjd if (res->hr_resuid == 0) { 247204076Spjd /* 248214284Spjd * Provider is used for the first time. If primary node done no 249214284Spjd * writes yet as well (we will find "virgin" argument) then 250214284Spjd * there is no need to synchronize anything. If primary node 251214284Spjd * done any writes already we have to synchronize everything. 252204076Spjd */ 253218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt == 0); 254204076Spjd res->hr_resuid = resuid; 255204076Spjd if (metadata_write(res) < 0) 256204076Spjd exit(EX_NOINPUT); 257214284Spjd if (nv_exists(nvin, "virgin")) { 258214284Spjd free(map); 259214284Spjd map = NULL; 260214284Spjd mapsize = 0; 261214284Spjd } else { 262214284Spjd memset(map, 0xff, mapsize); 263214284Spjd } 264204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 265219830Spjd } else if (res->hr_resuid != resuid) { 266219830Spjd char errmsg[256]; 267219830Spjd 268219830Spjd (void)snprintf(errmsg, sizeof(errmsg), 269219830Spjd "Resource unique ID mismatch (primary=%ju, secondary=%ju).", 270219830Spjd (uintmax_t)resuid, (uintmax_t)res->hr_resuid); 271219830Spjd pjdlog_error("%s", errmsg); 272219830Spjd nv_add_string(nvout, errmsg, "errmsg"); 273219830Spjd if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) { 274219830Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s", 275219830Spjd res->hr_remoteaddr); 276219830Spjd } 277219831Spjd nv_free(nvout); 278219830Spjd exit(EX_CONFIG); 279204076Spjd } else if ( 280204076Spjd /* Is primary is out-of-date? */ 281204076Spjd (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 282204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 283219843Spjd /* Nodes are more or less in sync? */ 284204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 285204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 286204076Spjd /* Is secondary is out-of-date? */ 287204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 288204076Spjd res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 289204076Spjd /* 290204076Spjd * Nodes are more or less in sync or one of the nodes is 291204076Spjd * out-of-date. 292204076Spjd * It doesn't matter at this point which one, we just have to 293204076Spjd * send out local bitmap to the remote node. 294204076Spjd */ 295204076Spjd if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 296204076Spjd (ssize_t)mapsize) { 297204076Spjd pjdlog_exit(LOG_ERR, "Unable to read activemap"); 298204076Spjd } 299204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 300204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 301204076Spjd /* Primary is out-of-date, sync from secondary. */ 302204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 303204076Spjd } else { 304204076Spjd /* 305204076Spjd * Secondary is out-of-date or counts match. 306204076Spjd * Sync from primary. 307204076Spjd */ 308204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 309204076Spjd } 310204076Spjd } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 311204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 312204076Spjd /* 313204076Spjd * Not good, we have split-brain condition. 314204076Spjd */ 315204076Spjd pjdlog_error("Split-brain detected, exiting."); 316204076Spjd nv_add_string(nvout, "Split-brain condition!", "errmsg"); 317204076Spjd free(map); 318204076Spjd map = NULL; 319204076Spjd mapsize = 0; 320204076Spjd } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 321204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 322204076Spjd /* 323204076Spjd * This should never happen in practise, but we will perform 324204076Spjd * full synchronization. 325204076Spjd */ 326218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 327204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt); 328204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 329204076Spjd METADATA_SIZE, res->hr_extentsize, 330204076Spjd res->hr_local_sectorsize); 331204076Spjd memset(map, 0xff, mapsize); 332204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 333204076Spjd /* In this one of five cases sync from secondary. */ 334204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 335204076Spjd } else { 336204076Spjd /* For the rest four cases sync from primary. */ 337204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 338204076Spjd } 339204076Spjd pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 340204076Spjd (uintmax_t)res->hr_primary_localcnt, 341204076Spjd (uintmax_t)res->hr_primary_remotecnt, 342204076Spjd (uintmax_t)res->hr_secondary_localcnt, 343204076Spjd (uintmax_t)res->hr_secondary_remotecnt); 344204076Spjd } 345220007Spjd nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 346204076Spjd if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 347214276Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 348204076Spjd res->hr_remoteaddr); 349204076Spjd } 350214275Spjd if (map != NULL) 351214275Spjd free(map); 352209182Spjd nv_free(nvout); 353220271Spjd /* Setup direction. */ 354220271Spjd if (proto_recv(res->hr_remotein, NULL, 0) == -1) 355220271Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 356204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 357204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 358204076Spjd /* Exit on split-brain. */ 359212038Spjd event_send(res, EVENT_SPLITBRAIN); 360204076Spjd exit(EX_CONFIG); 361204076Spjd } 362204076Spjd} 363204076Spjd 364204076Spjdvoid 365204076Spjdhastd_secondary(struct hast_resource *res, struct nv *nvin) 366204076Spjd{ 367213009Spjd sigset_t mask; 368204076Spjd pthread_t td; 369204076Spjd pid_t pid; 370219482Strociny int error, mode, debuglevel; 371204076Spjd 372204076Spjd /* 373204076Spjd * Create communication channel between parent and child. 374204076Spjd */ 375219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_ctrl) < 0) { 376204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 377204076Spjd pjdlog_exit(EX_OSERR, 378204076Spjd "Unable to create control sockets between parent and child"); 379204076Spjd } 380212038Spjd /* 381212038Spjd * Create communication channel between child and parent. 382212038Spjd */ 383219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_event) < 0) { 384212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 385212038Spjd pjdlog_exit(EX_OSERR, 386212038Spjd "Unable to create event sockets between child and parent"); 387212038Spjd } 388204076Spjd 389204076Spjd pid = fork(); 390204076Spjd if (pid < 0) { 391204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 392204076Spjd pjdlog_exit(EX_OSERR, "Unable to fork"); 393204076Spjd } 394204076Spjd 395204076Spjd if (pid > 0) { 396204076Spjd /* This is parent. */ 397204076Spjd proto_close(res->hr_remotein); 398204076Spjd res->hr_remotein = NULL; 399204076Spjd proto_close(res->hr_remoteout); 400204076Spjd res->hr_remoteout = NULL; 401212038Spjd /* Declare that we are receiver. */ 402212038Spjd proto_recv(res->hr_event, NULL, 0); 403218043Spjd /* Declare that we are sender. */ 404218043Spjd proto_send(res->hr_ctrl, NULL, 0); 405204076Spjd res->hr_workerpid = pid; 406204076Spjd return; 407204076Spjd } 408211977Spjd 409211984Spjd gres = res; 410218043Spjd mode = pjdlog_mode_get(); 411219482Strociny debuglevel = pjdlog_debug_get(); 412211984Spjd 413218043Spjd /* Declare that we are sender. */ 414218043Spjd proto_send(res->hr_event, NULL, 0); 415218043Spjd /* Declare that we are receiver. */ 416218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 417218043Spjd descriptors_cleanup(res); 418204076Spjd 419218045Spjd descriptors_assert(res, mode); 420218045Spjd 421218043Spjd pjdlog_init(mode); 422219482Strociny pjdlog_debug_set(debuglevel); 423218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 424220005Spjd setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role)); 425204076Spjd 426213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 427213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 428210880Spjd 429207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 430219721Strociny if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0) 431207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 432207371Spjd if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 433207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 434207371Spjd 435204076Spjd init_local(res); 436213007Spjd init_environment(); 437213007Spjd 438219847Spjd if (drop_privs(true) != 0) 439218049Spjd exit(EX_CONFIG); 440218214Spjd pjdlog_info("Privileges successfully dropped."); 441218049Spjd 442213007Spjd /* 443213007Spjd * Create the control thread before sending any event to the parent, 444213007Spjd * as we can deadlock when parent sends control request to worker, 445213007Spjd * but worker has no control thread started yet, so parent waits. 446213007Spjd * In the meantime worker sends an event to the parent, but parent 447213007Spjd * is unable to handle the event, because it waits for control 448213007Spjd * request response. 449213007Spjd */ 450213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 451218138Spjd PJDLOG_ASSERT(error == 0); 452213007Spjd 453204076Spjd init_remote(res, nvin); 454212038Spjd event_send(res, EVENT_CONNECT); 455204076Spjd 456204076Spjd error = pthread_create(&td, NULL, recv_thread, res); 457218138Spjd PJDLOG_ASSERT(error == 0); 458204076Spjd error = pthread_create(&td, NULL, disk_thread, res); 459218138Spjd PJDLOG_ASSERT(error == 0); 460213007Spjd (void)send_thread(res); 461204076Spjd} 462204076Spjd 463204076Spjdstatic void 464204076Spjdreqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 465204076Spjd{ 466204076Spjd char msg[1024]; 467204076Spjd va_list ap; 468204076Spjd int len; 469204076Spjd 470204076Spjd va_start(ap, fmt); 471204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 472204076Spjd va_end(ap); 473204076Spjd if ((size_t)len < sizeof(msg)) { 474204076Spjd switch (hio->hio_cmd) { 475204076Spjd case HIO_READ: 476204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 477204076Spjd "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 478204076Spjd (uintmax_t)hio->hio_length); 479204076Spjd break; 480204076Spjd case HIO_DELETE: 481204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 482204076Spjd "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 483204076Spjd (uintmax_t)hio->hio_length); 484204076Spjd break; 485204076Spjd case HIO_FLUSH: 486204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 487204076Spjd break; 488204076Spjd case HIO_WRITE: 489204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 490204076Spjd "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 491204076Spjd (uintmax_t)hio->hio_length); 492204076Spjd break; 493211882Spjd case HIO_KEEPALIVE: 494211882Spjd (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 495211882Spjd break; 496204076Spjd default: 497204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 498204076Spjd "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 499204076Spjd break; 500204076Spjd } 501204076Spjd } 502204076Spjd pjdlog_common(loglevel, debuglevel, error, "%s", msg); 503204076Spjd} 504204076Spjd 505204076Spjdstatic int 506204076Spjdrequnpack(struct hast_resource *res, struct hio *hio) 507204076Spjd{ 508204076Spjd 509204076Spjd hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 510204076Spjd if (hio->hio_cmd == 0) { 511204076Spjd pjdlog_error("Header contains no 'cmd' field."); 512204076Spjd hio->hio_error = EINVAL; 513204076Spjd goto end; 514204076Spjd } 515204076Spjd switch (hio->hio_cmd) { 516211882Spjd case HIO_KEEPALIVE: 517211882Spjd break; 518204076Spjd case HIO_READ: 519204076Spjd case HIO_WRITE: 520204076Spjd case HIO_DELETE: 521204076Spjd hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 522204076Spjd if (nv_error(hio->hio_nv) != 0) { 523204076Spjd pjdlog_error("Header is missing 'offset' field."); 524204076Spjd hio->hio_error = EINVAL; 525204076Spjd goto end; 526204076Spjd } 527204076Spjd hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 528204076Spjd if (nv_error(hio->hio_nv) != 0) { 529204076Spjd pjdlog_error("Header is missing 'length' field."); 530204076Spjd hio->hio_error = EINVAL; 531204076Spjd goto end; 532204076Spjd } 533204076Spjd if (hio->hio_length == 0) { 534204076Spjd pjdlog_error("Data length is zero."); 535204076Spjd hio->hio_error = EINVAL; 536204076Spjd goto end; 537204076Spjd } 538204076Spjd if (hio->hio_length > MAXPHYS) { 539204076Spjd pjdlog_error("Data length is too large (%ju > %ju).", 540204076Spjd (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 541204076Spjd hio->hio_error = EINVAL; 542204076Spjd goto end; 543204076Spjd } 544204076Spjd if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 545204076Spjd pjdlog_error("Offset %ju is not multiple of sector size.", 546204076Spjd (uintmax_t)hio->hio_offset); 547204076Spjd hio->hio_error = EINVAL; 548204076Spjd goto end; 549204076Spjd } 550204076Spjd if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 551204076Spjd pjdlog_error("Length %ju is not multiple of sector size.", 552204076Spjd (uintmax_t)hio->hio_length); 553204076Spjd hio->hio_error = EINVAL; 554204076Spjd goto end; 555204076Spjd } 556204076Spjd if (hio->hio_offset + hio->hio_length > 557204076Spjd (uint64_t)res->hr_datasize) { 558204076Spjd pjdlog_error("Data offset is too large (%ju > %ju).", 559204076Spjd (uintmax_t)(hio->hio_offset + hio->hio_length), 560204076Spjd (uintmax_t)res->hr_datasize); 561204076Spjd hio->hio_error = EINVAL; 562204076Spjd goto end; 563204076Spjd } 564204076Spjd break; 565204076Spjd default: 566204076Spjd pjdlog_error("Header contains invalid 'cmd' (%hhu).", 567204076Spjd hio->hio_cmd); 568204076Spjd hio->hio_error = EINVAL; 569204076Spjd goto end; 570204076Spjd } 571204076Spjd hio->hio_error = 0; 572204076Spjdend: 573204076Spjd return (hio->hio_error); 574204076Spjd} 575204076Spjd 576212899Spjdstatic __dead2 void 577211984Spjdsecondary_exit(int exitcode, const char *fmt, ...) 578211984Spjd{ 579211984Spjd va_list ap; 580211984Spjd 581218138Spjd PJDLOG_ASSERT(exitcode != EX_OK); 582211984Spjd va_start(ap, fmt); 583211984Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 584211984Spjd va_end(ap); 585212038Spjd event_send(gres, EVENT_DISCONNECT); 586211984Spjd exit(exitcode); 587211984Spjd} 588211984Spjd 589204076Spjd/* 590204076Spjd * Thread receives requests from the primary node. 591204076Spjd */ 592204076Spjdstatic void * 593204076Spjdrecv_thread(void *arg) 594204076Spjd{ 595204076Spjd struct hast_resource *res = arg; 596204076Spjd struct hio *hio; 597204076Spjd 598204076Spjd for (;;) { 599204076Spjd pjdlog_debug(2, "recv: Taking free request."); 600211877Spjd QUEUE_TAKE(free, hio); 601204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", hio); 602204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 603211984Spjd secondary_exit(EX_TEMPFAIL, 604204076Spjd "Unable to receive request header"); 605204076Spjd } 606211877Spjd if (requnpack(res, hio) != 0) { 607211877Spjd pjdlog_debug(2, 608211877Spjd "recv: (%p) Moving request to the send queue.", 609211877Spjd hio); 610211877Spjd QUEUE_INSERT(send, hio); 611211877Spjd continue; 612211877Spjd } 613204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, 614204076Spjd "recv: (%p) Got request header: ", hio); 615211882Spjd if (hio->hio_cmd == HIO_KEEPALIVE) { 616211882Spjd pjdlog_debug(2, 617211882Spjd "recv: (%p) Moving request to the free queue.", 618211882Spjd hio); 619211882Spjd nv_free(hio->hio_nv); 620211882Spjd QUEUE_INSERT(free, hio); 621211882Spjd continue; 622211882Spjd } else if (hio->hio_cmd == HIO_WRITE) { 623204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, 624204076Spjd hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 625211984Spjd secondary_exit(EX_TEMPFAIL, 626212051Spjd "Unable to receive request data"); 627204076Spjd } 628204076Spjd } 629204076Spjd pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 630204076Spjd hio); 631211877Spjd QUEUE_INSERT(disk, hio); 632204076Spjd } 633204076Spjd /* NOTREACHED */ 634204076Spjd return (NULL); 635204076Spjd} 636204076Spjd 637204076Spjd/* 638204076Spjd * Thread reads from or writes to local component and also handles DELETE and 639204076Spjd * FLUSH requests. 640204076Spjd */ 641204076Spjdstatic void * 642204076Spjddisk_thread(void *arg) 643204076Spjd{ 644204076Spjd struct hast_resource *res = arg; 645204076Spjd struct hio *hio; 646204076Spjd ssize_t ret; 647211877Spjd bool clear_activemap; 648204076Spjd 649204076Spjd clear_activemap = true; 650204076Spjd 651204076Spjd for (;;) { 652204076Spjd pjdlog_debug(2, "disk: Taking request."); 653211877Spjd QUEUE_TAKE(disk, hio); 654204076Spjd while (clear_activemap) { 655204076Spjd unsigned char *map; 656204076Spjd size_t mapsize; 657204076Spjd 658204076Spjd /* 659204076Spjd * When first request is received, it means that primary 660204076Spjd * already received our activemap, merged it and stored 661204076Spjd * locally. We can now safely clear our activemap. 662204076Spjd */ 663204076Spjd mapsize = 664204076Spjd activemap_calc_ondisk_size(res->hr_local_mediasize - 665204076Spjd METADATA_SIZE, res->hr_extentsize, 666204076Spjd res->hr_local_sectorsize); 667204076Spjd map = calloc(1, mapsize); 668204076Spjd if (map == NULL) { 669204076Spjd pjdlog_warning("Unable to allocate memory to clear local activemap."); 670204076Spjd break; 671204076Spjd } 672204076Spjd if (pwrite(res->hr_localfd, map, mapsize, 673204076Spjd METADATA_SIZE) != (ssize_t)mapsize) { 674204076Spjd pjdlog_errno(LOG_WARNING, 675204076Spjd "Unable to store cleared activemap"); 676204076Spjd free(map); 677204076Spjd break; 678204076Spjd } 679204076Spjd free(map); 680204076Spjd clear_activemap = false; 681204076Spjd pjdlog_debug(1, "Local activemap cleared."); 682204076Spjd } 683204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 684204076Spjd /* Handle the actual request. */ 685204076Spjd switch (hio->hio_cmd) { 686204076Spjd case HIO_READ: 687204076Spjd ret = pread(res->hr_localfd, hio->hio_data, 688204076Spjd hio->hio_length, 689204076Spjd hio->hio_offset + res->hr_localoff); 690204076Spjd if (ret < 0) 691204076Spjd hio->hio_error = errno; 692204076Spjd else if (ret != (int64_t)hio->hio_length) 693204076Spjd hio->hio_error = EIO; 694204076Spjd else 695204076Spjd hio->hio_error = 0; 696204076Spjd break; 697204076Spjd case HIO_WRITE: 698204076Spjd ret = pwrite(res->hr_localfd, hio->hio_data, 699204076Spjd hio->hio_length, 700204076Spjd hio->hio_offset + res->hr_localoff); 701204076Spjd if (ret < 0) 702204076Spjd hio->hio_error = errno; 703204076Spjd else if (ret != (int64_t)hio->hio_length) 704204076Spjd hio->hio_error = EIO; 705204076Spjd else 706204076Spjd hio->hio_error = 0; 707204076Spjd break; 708204076Spjd case HIO_DELETE: 709204076Spjd ret = g_delete(res->hr_localfd, 710204076Spjd hio->hio_offset + res->hr_localoff, 711204076Spjd hio->hio_length); 712204076Spjd if (ret < 0) 713204076Spjd hio->hio_error = errno; 714204076Spjd else 715204076Spjd hio->hio_error = 0; 716204076Spjd break; 717204076Spjd case HIO_FLUSH: 718204076Spjd ret = g_flush(res->hr_localfd); 719204076Spjd if (ret < 0) 720204076Spjd hio->hio_error = errno; 721204076Spjd else 722204076Spjd hio->hio_error = 0; 723204076Spjd break; 724204076Spjd } 725204076Spjd if (hio->hio_error != 0) { 726204076Spjd reqlog(LOG_ERR, 0, hio->hio_error, hio, 727204076Spjd "Request failed: "); 728204076Spjd } 729204076Spjd pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 730204076Spjd hio); 731211877Spjd QUEUE_INSERT(send, hio); 732204076Spjd } 733204076Spjd /* NOTREACHED */ 734204076Spjd return (NULL); 735204076Spjd} 736204076Spjd 737204076Spjd/* 738204076Spjd * Thread sends requests back to primary node. 739204076Spjd */ 740204076Spjdstatic void * 741204076Spjdsend_thread(void *arg) 742204076Spjd{ 743204076Spjd struct hast_resource *res = arg; 744204076Spjd struct nv *nvout; 745204076Spjd struct hio *hio; 746204076Spjd void *data; 747204076Spjd size_t length; 748204076Spjd 749204076Spjd for (;;) { 750204076Spjd pjdlog_debug(2, "send: Taking request."); 751211877Spjd QUEUE_TAKE(send, hio); 752204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 753204076Spjd nvout = nv_alloc(); 754204076Spjd /* Copy sequence number. */ 755204076Spjd nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 756204076Spjd switch (hio->hio_cmd) { 757204076Spjd case HIO_READ: 758204076Spjd if (hio->hio_error == 0) { 759204076Spjd data = hio->hio_data; 760204076Spjd length = hio->hio_length; 761204076Spjd break; 762204076Spjd } 763204076Spjd /* 764204076Spjd * We send no data in case of an error. 765204076Spjd */ 766204076Spjd /* FALLTHROUGH */ 767204076Spjd case HIO_DELETE: 768204076Spjd case HIO_FLUSH: 769204076Spjd case HIO_WRITE: 770204076Spjd data = NULL; 771204076Spjd length = 0; 772204076Spjd break; 773204076Spjd default: 774204076Spjd abort(); 775204076Spjd break; 776204076Spjd } 777204076Spjd if (hio->hio_error != 0) 778204076Spjd nv_add_int16(nvout, hio->hio_error, "error"); 779204076Spjd if (hast_proto_send(res, res->hr_remoteout, nvout, data, 780204076Spjd length) < 0) { 781211984Spjd secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 782204076Spjd } 783204076Spjd nv_free(nvout); 784209185Spjd pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 785204076Spjd hio); 786204076Spjd nv_free(hio->hio_nv); 787204076Spjd hio->hio_error = 0; 788211877Spjd QUEUE_INSERT(free, hio); 789204076Spjd } 790204076Spjd /* NOTREACHED */ 791204076Spjd return (NULL); 792204076Spjd} 793