secondary.c revision 256027
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: stable/9/sbin/hastd/secondary.c 256027 2013-10-03 18:52:04Z 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 { 68219864Spjd uint64_t hio_seq; 69219864Spjd int hio_error; 70204076Spjd void *hio_data; 71204076Spjd uint8_t hio_cmd; 72204076Spjd uint64_t hio_offset; 73204076Spjd uint64_t hio_length; 74204076Spjd bool hio_memsync; 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 does 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 * Thread that sends requests back to primary takes requests from this list. 95204076Spjd */ 96204076Spjdstatic TAILQ_HEAD(, hio) hio_send_list; 97204076Spjdstatic pthread_mutex_t hio_send_list_lock; 98204076Spjdstatic pthread_cond_t hio_send_list_cond; 99204076Spjd 100204076Spjd/* 101204076Spjd * Maximum number of outstanding I/O requests. 102204076Spjd */ 103204076Spjd#define HAST_HIO_MAX 256 104204076Spjd 105204076Spjdstatic void *recv_thread(void *arg); 106204076Spjdstatic void *disk_thread(void *arg); 107204076Spjdstatic void *send_thread(void *arg); 108204076Spjd 109204076Spjd#define QUEUE_INSERT(name, hio) do { \ 110211877Spjd bool _wakeup; \ 111211877Spjd \ 112211877Spjd mtx_lock(&hio_##name##_list_lock); \ 113211877Spjd _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 114211877Spjd TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \ 115211877Spjd mtx_unlock(&hio_##name##_list_lock); \ 116211877Spjd if (_wakeup) \ 117211877Spjd cv_broadcast(&hio_##name##_list_cond); \ 118211877Spjd} while (0) 119211877Spjd#define QUEUE_TAKE(name, hio) do { \ 120211877Spjd mtx_lock(&hio_##name##_list_lock); \ 121211877Spjd while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 122211877Spjd cv_wait(&hio_##name##_list_cond, \ 123211877Spjd &hio_##name##_list_lock); \ 124211877Spjd } \ 125211877Spjd TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \ 126211877Spjd mtx_unlock(&hio_##name##_list_lock); \ 127211877Spjd} while (0) 128211877Spjd 129211877Spjdstatic void 130204076Spjdhio_clear(struct hio *hio) 131204076Spjd{ 132204076Spjd 133204076Spjd hio->hio_seq = 0; 134204076Spjd hio->hio_error = 0; 135204076Spjd hio->hio_cmd = HIO_UNDEF; 136204076Spjd hio->hio_offset = 0; 137204076Spjd hio->hio_length = 0; 138204076Spjd hio->hio_memsync = false; 139204076Spjd} 140204076Spjd 141204076Spjdstatic void 142204076Spjdhio_copy(const struct hio *srchio, struct hio *dsthio) 143204076Spjd{ 144204076Spjd 145204076Spjd /* 146204076Spjd * We don't copy hio_error, hio_data and hio_next fields. 147204076Spjd */ 148204076Spjd 149204076Spjd dsthio->hio_seq = srchio->hio_seq; 150204076Spjd dsthio->hio_cmd = srchio->hio_cmd; 151204076Spjd dsthio->hio_offset = srchio->hio_offset; 152204076Spjd dsthio->hio_length = srchio->hio_length; 153204076Spjd dsthio->hio_memsync = srchio->hio_memsync; 154204076Spjd} 155210879Spjd 156210879Spjdstatic void 157210879Spjdinit_environment(void) 158204076Spjd{ 159204076Spjd struct hio *hio; 160204076Spjd unsigned int ii; 161204076Spjd 162210879Spjd /* 163210879Spjd * Initialize lists, their locks and theirs condition variables. 164210879Spjd */ 165204076Spjd TAILQ_INIT(&hio_free_list); 166204076Spjd mtx_init(&hio_free_list_lock); 167204076Spjd cv_init(&hio_free_list_cond); 168204076Spjd TAILQ_INIT(&hio_disk_list); 169204076Spjd mtx_init(&hio_disk_list_lock); 170204076Spjd cv_init(&hio_disk_list_cond); 171204076Spjd TAILQ_INIT(&hio_send_list); 172204076Spjd mtx_init(&hio_send_list_lock); 173204076Spjd cv_init(&hio_send_list_cond); 174204076Spjd 175204076Spjd /* 176204076Spjd * Allocate requests pool and initialize requests. 177204076Spjd */ 178204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 179204076Spjd hio = malloc(sizeof(*hio)); 180204076Spjd if (hio == NULL) { 181204076Spjd pjdlog_exitx(EX_TEMPFAIL, 182204076Spjd "Unable to allocate memory (%zu bytes) for hio request.", 183204076Spjd sizeof(*hio)); 184204076Spjd } 185204076Spjd hio->hio_data = malloc(MAXPHYS); 186223181Strociny if (hio->hio_data == NULL) { 187220271Spjd pjdlog_exitx(EX_TEMPFAIL, 188220271Spjd "Unable to allocate memory (%zu bytes) for gctl_data.", 189220271Spjd (size_t)MAXPHYS); 190223181Strociny } 191220271Spjd hio_clear(hio); 192204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 193204076Spjd } 194204076Spjd} 195204076Spjd 196204076Spjdstatic void 197204076Spjdinit_local(struct hast_resource *res) 198204076Spjd{ 199204076Spjd 200204076Spjd if (metadata_read(res, true) == -1) 201204076Spjd exit(EX_NOINPUT); 202204076Spjd} 203204076Spjd 204204076Spjdstatic void 205204076Spjdinit_remote(struct hast_resource *res, struct nv *nvin) 206204076Spjd{ 207204076Spjd uint64_t resuid; 208204076Spjd struct nv *nvout; 209204076Spjd unsigned char *map; 210204076Spjd size_t mapsize; 211204076Spjd 212204076Spjd#ifdef notyet 213204076Spjd /* Setup direction. */ 214204076Spjd if (proto_send(res->hr_remoteout, NULL, 0) == -1) 215204076Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 216204076Spjd#endif 217204076Spjd 218204076Spjd nvout = nv_alloc(); 219204076Spjd nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 220204076Spjd nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 221204076Spjd resuid = nv_get_uint64(nvin, "resuid"); 222204076Spjd res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 223204076Spjd res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 224204076Spjd nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 225204076Spjd nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 226204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 227204076Spjd METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 228204076Spjd map = malloc(mapsize); 229204076Spjd if (map == NULL) { 230204076Spjd pjdlog_exitx(EX_TEMPFAIL, 231204076Spjd "Unable to allocate memory (%zu bytes) for activemap.", 232204076Spjd mapsize); 233204076Spjd } 234204076Spjd /* 235204076Spjd * When we work as primary and secondary is missing we will increase 236204076Spjd * localcnt in our metadata. When secondary is connected and synced 237204076Spjd * we make localcnt be equal to remotecnt, which means nodes are more 238204076Spjd * or less in sync. 239204076Spjd * Split-brain condition is when both nodes are not able to communicate 240204076Spjd * and are both configured as primary nodes. In turn, they can both 241204076Spjd * make incompatible changes to the data and we have to detect that. 242204076Spjd * Under split-brain condition we will increase our localcnt on first 243204076Spjd * write and remote node will increase its localcnt on first write. 244204076Spjd * When we connect we can see that primary's localcnt is greater than 245204076Spjd * our remotecnt (primary was modified while we weren't watching) and 246204076Spjd * our localcnt is greater than primary's remotecnt (we were modified 247204076Spjd * while primary wasn't watching). 248204076Spjd * There are many possible combinations which are all gathered below. 249204076Spjd * Don't pay too much attention to exact numbers, the more important 250214284Spjd * is to compare them. We compare secondary's local with primary's 251214284Spjd * remote and secondary's remote with primary's local. 252214284Spjd * Note that every case where primary's localcnt is smaller than 253214284Spjd * secondary's remotecnt and where secondary's localcnt is smaller than 254204076Spjd * primary's remotecnt should be impossible in practise. We will perform 255218138Spjd * full synchronization then. Those cases are marked with an asterisk. 256204076Spjd * Regular synchronization means that only extents marked as dirty are 257204076Spjd * synchronized (regular synchronization). 258204076Spjd * 259214284Spjd * SECONDARY METADATA PRIMARY METADATA 260214284Spjd * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 261214284Spjd * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 262214284Spjd * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 263214284Spjd * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 264214284Spjd * regular sync from secondary. 265214284Spjd * local=3 remote=3 local=3 remote=3 Regular sync just in case. 266220865Spjd * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 267204076Spjd * local=3 remote=3 local=4 remote=2 Split-brain condition. 268219830Spjd * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 269219830Spjd * regular sync from primary. 270219830Spjd * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 271219830Spjd */ 272219830Spjd if (res->hr_resuid == 0) { 273219830Spjd /* 274219830Spjd * Provider is used for the first time. If primary node done no 275219830Spjd * writes yet as well (we will find "virgin" argument) then 276219830Spjd * there is no need to synchronize anything. If primary node 277219830Spjd * done any writes already we have to synchronize everything. 278219830Spjd */ 279219830Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt == 0); 280219831Spjd res->hr_resuid = resuid; 281219830Spjd if (metadata_write(res) == -1) 282204076Spjd exit(EX_NOINPUT); 283204076Spjd if (nv_exists(nvin, "virgin")) { 284204076Spjd free(map); 285204076Spjd map = NULL; 286219843Spjd mapsize = 0; 287204076Spjd } else { 288204076Spjd memset(map, 0xff, mapsize); 289204076Spjd } 290204076Spjd nv_add_int8(nvout, 1, "virgin"); 291204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 292204076Spjd } else if (res->hr_resuid != resuid) { 293204076Spjd char errmsg[256]; 294204076Spjd 295204076Spjd free(map); 296204076Spjd (void)snprintf(errmsg, sizeof(errmsg), 297204076Spjd "Resource unique ID mismatch (primary=%ju, secondary=%ju).", 298204076Spjd (uintmax_t)resuid, (uintmax_t)res->hr_resuid); 299204076Spjd pjdlog_error("%s", errmsg); 300204076Spjd nv_add_string(nvout, errmsg, "errmsg"); 301204076Spjd if (hast_proto_send(res, res->hr_remotein, nvout, 302204076Spjd NULL, 0) == -1) { 303204076Spjd pjdlog_exit(EX_TEMPFAIL, 304204076Spjd "Unable to send response to %s", 305204076Spjd res->hr_remoteaddr); 306204076Spjd } 307204076Spjd nv_free(nvout); 308204076Spjd exit(EX_CONFIG); 309204076Spjd } else if ( 310204076Spjd /* Is primary out-of-date? */ 311204076Spjd (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 312204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 313204076Spjd /* Are the nodes more or less in sync? */ 314204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 315204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 316204076Spjd /* Is secondary out-of-date? */ 317204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 318204076Spjd res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 319204076Spjd /* 320204076Spjd * Nodes are more or less in sync or one of the nodes is 321204076Spjd * out-of-date. 322204076Spjd * It doesn't matter at this point which one, we just have to 323204076Spjd * send out local bitmap to the remote node. 324204076Spjd */ 325204076Spjd if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 326204076Spjd (ssize_t)mapsize) { 327204076Spjd pjdlog_exit(LOG_ERR, "Unable to read activemap"); 328204076Spjd } 329218138Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 330204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 331204076Spjd /* Primary is out-of-date, sync from secondary. */ 332204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 333204076Spjd } else { 334204076Spjd /* 335204076Spjd * Secondary is out-of-date or counts match. 336204076Spjd * Sync from primary. 337204076Spjd */ 338204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 339204076Spjd } 340204076Spjd } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 341204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 342204076Spjd /* 343204076Spjd * Not good, we have split-brain condition. 344204076Spjd */ 345204076Spjd free(map); 346204076Spjd pjdlog_error("Split-brain detected, exiting."); 347204076Spjd nv_add_string(nvout, "Split-brain condition!", "errmsg"); 348220007Spjd if (hast_proto_send(res, res->hr_remotein, nvout, 349204076Spjd NULL, 0) == -1) { 350214276Spjd pjdlog_exit(EX_TEMPFAIL, 351204076Spjd "Unable to send response to %s", 352204076Spjd res->hr_remoteaddr); 353214275Spjd } 354214275Spjd nv_free(nvout); 355209182Spjd /* Exit on split-brain. */ 356223181Strociny event_send(res, EVENT_SPLITBRAIN); 357220271Spjd exit(EX_CONFIG); 358220271Spjd } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 359220271Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 360223181Strociny /* 361204076Spjd * This should never happen in practise, but we will perform 362204076Spjd * full synchronization. 363204076Spjd */ 364212038Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 365204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt); 366204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 367204076Spjd METADATA_SIZE, res->hr_extentsize, 368204076Spjd res->hr_local_sectorsize); 369204076Spjd memset(map, 0xff, mapsize); 370204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 371204076Spjd /* In this one of five cases sync from secondary. */ 372213009Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 373204076Spjd } else { 374204076Spjd /* For the rest four cases sync from primary. */ 375219482Strociny nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 376204076Spjd } 377204076Spjd pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 378204076Spjd (uintmax_t)res->hr_primary_localcnt, 379204076Spjd (uintmax_t)res->hr_primary_remotecnt, 380219818Spjd (uintmax_t)res->hr_secondary_localcnt, 381204076Spjd (uintmax_t)res->hr_secondary_remotecnt); 382204076Spjd } 383204076Spjd nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 384204076Spjd if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) { 385212038Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 386212038Spjd res->hr_remoteaddr); 387212038Spjd } 388219818Spjd if (map != NULL) 389212038Spjd free(map); 390212038Spjd nv_free(nvout); 391212038Spjd#ifdef notyet 392212038Spjd /* Setup direction. */ 393204076Spjd if (proto_recv(res->hr_remotein, NULL, 0) == -1) 394204076Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 395204076Spjd#endif 396204076Spjd} 397204076Spjd 398204076Spjdvoid 399204076Spjdhastd_secondary(struct hast_resource *res, struct nv *nvin) 400204076Spjd{ 401204076Spjd sigset_t mask; 402204076Spjd pthread_t td; 403204076Spjd pid_t pid; 404204076Spjd int error, mode, debuglevel; 405204076Spjd 406212038Spjd /* 407212038Spjd * Create communication channel between parent and child. 408218043Spjd */ 409218043Spjd if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) { 410204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 411204076Spjd pjdlog_exit(EX_OSERR, 412204076Spjd "Unable to create control sockets between parent and child"); 413211977Spjd } 414211984Spjd /* 415218043Spjd * Create communication channel between child and parent. 416219482Strociny */ 417211984Spjd if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) { 418218043Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 419218043Spjd pjdlog_exit(EX_OSERR, 420218043Spjd "Unable to create event sockets between child and parent"); 421218043Spjd } 422218043Spjd 423204076Spjd pid = fork(); 424218045Spjd if (pid == -1) { 425218045Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 426218043Spjd pjdlog_exit(EX_OSERR, "Unable to fork"); 427219482Strociny } 428218043Spjd 429220005Spjd if (pid > 0) { 430204076Spjd /* This is parent. */ 431213009Spjd proto_close(res->hr_remotein); 432213009Spjd res->hr_remotein = NULL; 433210880Spjd proto_close(res->hr_remoteout); 434207371Spjd res->hr_remoteout = NULL; 435219721Strociny /* Declare that we are receiver. */ 436207371Spjd proto_recv(res->hr_event, NULL, 0); 437207371Spjd /* Declare that we are sender. */ 438207371Spjd proto_send(res->hr_ctrl, NULL, 0); 439207371Spjd res->hr_workerpid = pid; 440204076Spjd return; 441213007Spjd } 442213007Spjd 443221899Spjd gres = res; 444218049Spjd mode = pjdlog_mode_get(); 445218214Spjd debuglevel = pjdlog_debug_get(); 446218049Spjd 447213007Spjd /* Declare that we are sender. */ 448213007Spjd proto_send(res->hr_event, NULL, 0); 449213007Spjd /* Declare that we are receiver. */ 450213007Spjd proto_recv(res->hr_ctrl, NULL, 0); 451213007Spjd descriptors_cleanup(res); 452213007Spjd 453213007Spjd descriptors_assert(res, mode); 454213007Spjd 455213007Spjd pjdlog_init(mode); 456218138Spjd pjdlog_debug_set(debuglevel); 457213007Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 458204076Spjd setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role)); 459212038Spjd 460204076Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 461204076Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 462218138Spjd 463204076Spjd /* Error in setting timeout is not critical, but why should it fail? */ 464218138Spjd if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1) 465213007Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 466204076Spjd if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1) 467204076Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 468204076Spjd 469204076Spjd init_local(res); 470204076Spjd init_environment(); 471204076Spjd 472204076Spjd if (drop_privs(res) != 0) 473204076Spjd exit(EX_CONFIG); 474204076Spjd pjdlog_info("Privileges successfully dropped."); 475204076Spjd 476204076Spjd /* 477204076Spjd * Create the control thread before sending any event to the parent, 478204076Spjd * as we can deadlock when parent sends control request to worker, 479204076Spjd * but worker has no control thread started yet, so parent waits. 480204076Spjd * In the meantime worker sends an event to the parent, but parent 481204076Spjd * is unable to handle the event, because it waits for control 482204076Spjd * request response. 483204076Spjd */ 484204076Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 485204076Spjd PJDLOG_ASSERT(error == 0); 486204076Spjd 487204076Spjd init_remote(res, nvin); 488204076Spjd event_send(res, EVENT_CONNECT); 489204076Spjd 490204076Spjd error = pthread_create(&td, NULL, recv_thread, res); 491204076Spjd PJDLOG_ASSERT(error == 0); 492204076Spjd error = pthread_create(&td, NULL, disk_thread, res); 493204076Spjd PJDLOG_ASSERT(error == 0); 494204076Spjd (void)send_thread(res); 495204076Spjd} 496204076Spjd 497204076Spjdstatic void 498211882Spjdreqlog(int loglevel, int debuglevel, int error, struct hio *hio, 499211882Spjd const char *fmt, ...) 500211882Spjd{ 501204076Spjd char msg[1024]; 502204076Spjd va_list ap; 503204076Spjd int len; 504204076Spjd 505204076Spjd va_start(ap, fmt); 506204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 507204076Spjd va_end(ap); 508204076Spjd if ((size_t)len < sizeof(msg)) { 509204076Spjd switch (hio->hio_cmd) { 510204076Spjd case HIO_READ: 511204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 512204076Spjd "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 513204076Spjd (uintmax_t)hio->hio_length); 514204076Spjd break; 515204076Spjd case HIO_DELETE: 516204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 517204076Spjd "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 518204076Spjd (uintmax_t)hio->hio_length); 519204076Spjd break; 520204076Spjd case HIO_FLUSH: 521222164Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 522211882Spjd break; 523211882Spjd case HIO_WRITE: 524204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 525204076Spjd "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 526204076Spjd (uintmax_t)hio->hio_length); 527204076Spjd break; 528204076Spjd case HIO_KEEPALIVE: 529204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 530204076Spjd break; 531204076Spjd default: 532204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 533204076Spjd "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 534204076Spjd break; 535204076Spjd } 536204076Spjd } 537204076Spjd pjdlog_common(loglevel, debuglevel, error, "%s", msg); 538204076Spjd} 539204076Spjd 540204076Spjdstatic int 541204076Spjdrequnpack(struct hast_resource *res, struct hio *hio, struct nv *nv) 542204076Spjd{ 543204076Spjd 544204076Spjd hio->hio_cmd = nv_get_uint8(nv, "cmd"); 545204076Spjd if (hio->hio_cmd == 0) { 546204076Spjd pjdlog_error("Header contains no 'cmd' field."); 547204076Spjd hio->hio_error = EINVAL; 548204076Spjd goto end; 549204076Spjd } 550204076Spjd if (hio->hio_cmd != HIO_KEEPALIVE) { 551204076Spjd hio->hio_seq = nv_get_uint64(nv, "seq"); 552204076Spjd if (hio->hio_seq == 0) { 553204076Spjd pjdlog_error("Header contains no 'seq' field."); 554204076Spjd hio->hio_error = EINVAL; 555204076Spjd goto end; 556204076Spjd } 557204076Spjd } 558204076Spjd switch (hio->hio_cmd) { 559204076Spjd case HIO_FLUSH: 560204076Spjd case HIO_KEEPALIVE: 561204076Spjd break; 562204076Spjd case HIO_WRITE: 563204076Spjd hio->hio_memsync = nv_exists(nv, "memsync"); 564204076Spjd /* FALLTHROUGH */ 565204076Spjd case HIO_READ: 566204076Spjd case HIO_DELETE: 567204076Spjd hio->hio_offset = nv_get_uint64(nv, "offset"); 568204076Spjd if (nv_error(nv) != 0) { 569204076Spjd pjdlog_error("Header is missing 'offset' field."); 570204076Spjd hio->hio_error = EINVAL; 571204076Spjd goto end; 572204076Spjd } 573204076Spjd hio->hio_length = nv_get_uint64(nv, "length"); 574204076Spjd if (nv_error(nv) != 0) { 575204076Spjd pjdlog_error("Header is missing 'length' field."); 576204076Spjd hio->hio_error = EINVAL; 577204076Spjd goto end; 578204076Spjd } 579204076Spjd if (hio->hio_length == 0) { 580204076Spjd pjdlog_error("Data length is zero."); 581204076Spjd hio->hio_error = EINVAL; 582212899Spjd goto end; 583211984Spjd } 584211984Spjd if (hio->hio_cmd != HIO_DELETE && hio->hio_length > MAXPHYS) { 585211984Spjd pjdlog_error("Data length is too large (%ju > %ju).", 586211984Spjd (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 587218138Spjd hio->hio_error = EINVAL; 588211984Spjd goto end; 589211984Spjd } 590211984Spjd if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 591212038Spjd pjdlog_error("Offset %ju is not multiple of sector size.", 592211984Spjd (uintmax_t)hio->hio_offset); 593211984Spjd hio->hio_error = EINVAL; 594211984Spjd goto end; 595204076Spjd } 596204076Spjd if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 597204076Spjd pjdlog_error("Length %ju is not multiple of sector size.", 598204076Spjd (uintmax_t)hio->hio_length); 599204076Spjd hio->hio_error = EINVAL; 600204076Spjd goto end; 601204076Spjd } 602204076Spjd if (hio->hio_offset + hio->hio_length > 603204076Spjd (uint64_t)res->hr_datasize) { 604204076Spjd pjdlog_error("Data offset is too large (%ju > %ju).", 605204076Spjd (uintmax_t)(hio->hio_offset + hio->hio_length), 606211877Spjd (uintmax_t)res->hr_datasize); 607204076Spjd hio->hio_error = EINVAL; 608204076Spjd goto end; 609211984Spjd } 610204076Spjd break; 611204076Spjd default: 612211877Spjd pjdlog_error("Header contains invalid 'cmd' (%hhu).", 613211877Spjd hio->hio_cmd); 614211877Spjd hio->hio_error = EINVAL; 615211877Spjd goto end; 616211877Spjd } 617211877Spjd hio->hio_error = 0; 618211877Spjdend: 619222228Spjd return (hio->hio_error); 620222228Spjd} 621222228Spjd 622222228Spjdstatic __dead2 void 623222228Spjdsecondary_exit(int exitcode, const char *fmt, ...) 624222228Spjd{ 625222228Spjd va_list ap; 626222228Spjd 627222228Spjd PJDLOG_ASSERT(exitcode != EX_OK); 628222228Spjd va_start(ap, fmt); 629222228Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 630222228Spjd va_end(ap); 631222228Spjd event_send(gres, EVENT_DISCONNECT); 632222228Spjd exit(exitcode); 633204076Spjd} 634204076Spjd 635211882Spjd/* 636211882Spjd * Thread receives requests from the primary node. 637211882Spjd */ 638211882Spjdstatic void * 639211882Spjdrecv_thread(void *arg) 640211882Spjd{ 641211882Spjd struct hast_resource *res = arg; 642211882Spjd struct hio *hio, *mshio; 643204076Spjd struct nv *nv; 644204076Spjd 645211984Spjd for (;;) { 646212051Spjd pjdlog_debug(2, "recv: Taking free request."); 647204076Spjd QUEUE_TAKE(free, hio); 648204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", hio); 649204076Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) { 650204076Spjd secondary_exit(EX_TEMPFAIL, 651211877Spjd "Unable to receive request header"); 652204076Spjd } 653204076Spjd if (requnpack(res, hio, nv) != 0) { 654204076Spjd nv_free(nv); 655204076Spjd pjdlog_debug(2, 656204076Spjd "recv: (%p) Moving request to the send queue.", 657204076Spjd hio); 658204076Spjd QUEUE_INSERT(send, hio); 659204076Spjd continue; 660204076Spjd } 661204076Spjd switch (hio->hio_cmd) { 662204076Spjd case HIO_READ: 663204076Spjd res->hr_stat_read++; 664204076Spjd break; 665204076Spjd case HIO_WRITE: 666204076Spjd res->hr_stat_write++; 667211877Spjd break; 668204076Spjd case HIO_DELETE: 669204076Spjd res->hr_stat_delete++; 670204076Spjd break; 671204076Spjd case HIO_FLUSH: 672204076Spjd res->hr_stat_flush++; 673211877Spjd break; 674204076Spjd case HIO_KEEPALIVE: 675204076Spjd break; 676204076Spjd default: 677204076Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 678204076Spjd hio->hio_cmd); 679204076Spjd } 680204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, 681204076Spjd "recv: (%p) Got request header: ", hio); 682204076Spjd if (hio->hio_cmd == HIO_KEEPALIVE) { 683204076Spjd nv_free(nv); 684204076Spjd pjdlog_debug(2, 685204076Spjd "recv: (%p) Moving request to the free queue.", 686204076Spjd hio); 687204076Spjd hio_clear(hio); 688204076Spjd QUEUE_INSERT(free, hio); 689204076Spjd continue; 690204076Spjd } else if (hio->hio_cmd == HIO_WRITE) { 691204076Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 692204076Spjd hio->hio_data, MAXPHYS) == -1) { 693204076Spjd secondary_exit(EX_TEMPFAIL, 694204076Spjd "Unable to receive request data"); 695204076Spjd } 696204076Spjd if (hio->hio_memsync) { 697204076Spjd /* 698204076Spjd * For memsync requests we expect two replies. 699204076Spjd * Clone the hio so we can handle both of them. 700204076Spjd */ 701204076Spjd pjdlog_debug(2, "recv: Taking free request."); 702204076Spjd QUEUE_TAKE(free, mshio); 703204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", 704204076Spjd mshio); 705204076Spjd hio_copy(hio, mshio); 706204076Spjd mshio->hio_error = 0; 707204076Spjd /* 708204076Spjd * We want to keep 'memsync' tag only on the 709204076Spjd * request going onto send queue (mshio). 710204076Spjd */ 711204076Spjd hio->hio_memsync = false; 712204076Spjd pjdlog_debug(2, 713204076Spjd "recv: (%p) Moving memsync request to the send queue.", 714204076Spjd mshio); 715204076Spjd QUEUE_INSERT(send, mshio); 716204076Spjd } 717204076Spjd } 718204076Spjd nv_free(nv); 719204076Spjd pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 720204076Spjd hio); 721204076Spjd QUEUE_INSERT(disk, hio); 722204076Spjd } 723204076Spjd /* NOTREACHED */ 724204076Spjd return (NULL); 725204076Spjd} 726204076Spjd 727204076Spjd/* 728204076Spjd * Thread reads from or writes to local component and also handles DELETE and 729204076Spjd * FLUSH requests. 730204076Spjd */ 731204076Spjdstatic void * 732204076Spjddisk_thread(void *arg) 733204076Spjd{ 734204076Spjd struct hast_resource *res = arg; 735204076Spjd struct hio *hio; 736204076Spjd ssize_t ret; 737204076Spjd bool clear_activemap, logerror; 738204076Spjd 739204076Spjd clear_activemap = true; 740204076Spjd 741204076Spjd for (;;) { 742204076Spjd pjdlog_debug(2, "disk: Taking request."); 743204076Spjd QUEUE_TAKE(disk, hio); 744204076Spjd while (clear_activemap) { 745204076Spjd unsigned char *map; 746204076Spjd size_t mapsize; 747204076Spjd 748204076Spjd /* 749204076Spjd * When first request is received, it means that primary 750204076Spjd * already received our activemap, merged it and stored 751211877Spjd * locally. We can now safely clear our activemap. 752204076Spjd */ 753204076Spjd mapsize = 754204076Spjd activemap_calc_ondisk_size(res->hr_local_mediasize - 755204076Spjd METADATA_SIZE, res->hr_extentsize, 756204076Spjd res->hr_local_sectorsize); 757204076Spjd map = calloc(1, mapsize); 758204076Spjd if (map == NULL) { 759204076Spjd pjdlog_warning("Unable to allocate memory to clear local activemap."); 760204076Spjd break; 761204076Spjd } 762204076Spjd if (pwrite(res->hr_localfd, map, mapsize, 763204076Spjd METADATA_SIZE) != (ssize_t)mapsize) { 764204076Spjd pjdlog_errno(LOG_WARNING, 765204076Spjd "Unable to store cleared activemap"); 766204076Spjd free(map); 767204076Spjd res->hr_stat_activemap_write_error++; 768204076Spjd break; 769204076Spjd } 770204076Spjd free(map); 771211877Spjd clear_activemap = false; 772204076Spjd pjdlog_debug(1, "Local activemap cleared."); 773204076Spjd break; 774204076Spjd } 775204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 776204076Spjd logerror = true; 777204076Spjd /* Handle the actual request. */ 778204076Spjd switch (hio->hio_cmd) { 779204076Spjd case HIO_READ: 780204076Spjd ret = pread(res->hr_localfd, hio->hio_data, 781204076Spjd hio->hio_length, 782204076Spjd hio->hio_offset + res->hr_localoff); 783204076Spjd if (ret == -1) 784204076Spjd hio->hio_error = errno; 785204076Spjd else if (ret != (int64_t)hio->hio_length) 786204076Spjd hio->hio_error = EIO; 787204076Spjd else 788204076Spjd hio->hio_error = 0; 789204076Spjd break; 790204076Spjd case HIO_WRITE: 791204076Spjd ret = pwrite(res->hr_localfd, hio->hio_data, 792204076Spjd hio->hio_length, 793204076Spjd hio->hio_offset + res->hr_localoff); 794225782Spjd if (ret == -1) 795225782Spjd hio->hio_error = errno; 796204076Spjd else if (ret != (int64_t)hio->hio_length) 797204076Spjd hio->hio_error = EIO; 798204076Spjd else 799204076Spjd hio->hio_error = 0; 800204076Spjd break; 801211984Spjd case HIO_DELETE: 802204076Spjd ret = g_delete(res->hr_localfd, 803204076Spjd hio->hio_offset + res->hr_localoff, 804209185Spjd hio->hio_length); 805204076Spjd if (ret == -1) 806204076Spjd hio->hio_error = errno; 807204076Spjd else 808211877Spjd hio->hio_error = 0; 809204076Spjd break; 810204076Spjd case HIO_FLUSH: 811204076Spjd if (!res->hr_localflush) { 812204076Spjd ret = -1; 813 hio->hio_error = EOPNOTSUPP; 814 logerror = false; 815 break; 816 } 817 ret = g_flush(res->hr_localfd); 818 if (ret == -1) { 819 if (errno == EOPNOTSUPP) 820 res->hr_localflush = false; 821 hio->hio_error = errno; 822 } else { 823 hio->hio_error = 0; 824 } 825 break; 826 default: 827 PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 828 hio->hio_cmd); 829 } 830 if (logerror && hio->hio_error != 0) { 831 reqlog(LOG_ERR, 0, hio->hio_error, hio, 832 "Request failed: "); 833 } 834 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 835 hio); 836 QUEUE_INSERT(send, hio); 837 } 838 /* NOTREACHED */ 839 return (NULL); 840} 841 842/* 843 * Thread sends requests back to primary node. 844 */ 845static void * 846send_thread(void *arg) 847{ 848 struct hast_resource *res = arg; 849 struct nv *nvout; 850 struct hio *hio; 851 void *data; 852 size_t length; 853 854 for (;;) { 855 pjdlog_debug(2, "send: Taking request."); 856 QUEUE_TAKE(send, hio); 857 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 858 nvout = nv_alloc(); 859 /* Copy sequence number. */ 860 nv_add_uint64(nvout, hio->hio_seq, "seq"); 861 if (hio->hio_memsync) { 862 PJDLOG_ASSERT(hio->hio_cmd == HIO_WRITE); 863 nv_add_int8(nvout, 1, "received"); 864 } 865 switch (hio->hio_cmd) { 866 case HIO_READ: 867 if (hio->hio_error == 0) { 868 data = hio->hio_data; 869 length = hio->hio_length; 870 break; 871 } 872 /* 873 * We send no data in case of an error. 874 */ 875 /* FALLTHROUGH */ 876 case HIO_DELETE: 877 case HIO_FLUSH: 878 case HIO_WRITE: 879 data = NULL; 880 length = 0; 881 break; 882 default: 883 PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 884 hio->hio_cmd); 885 } 886 if (hio->hio_error != 0) { 887 switch (hio->hio_cmd) { 888 case HIO_READ: 889 res->hr_stat_read_error++; 890 break; 891 case HIO_WRITE: 892 res->hr_stat_write_error++; 893 break; 894 case HIO_DELETE: 895 res->hr_stat_delete_error++; 896 break; 897 case HIO_FLUSH: 898 res->hr_stat_flush_error++; 899 break; 900 } 901 nv_add_int16(nvout, hio->hio_error, "error"); 902 } 903 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 904 length) == -1) { 905 secondary_exit(EX_TEMPFAIL, "Unable to send reply"); 906 } 907 nv_free(nvout); 908 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 909 hio); 910 hio_clear(hio); 911 QUEUE_INSERT(free, hio); 912 } 913 /* NOTREACHED */ 914 return (NULL); 915} 916