secondary.c revision 230092
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 230092 2012-01-13 23:25:35Z 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 void *hio_data; 71204076Spjd uint8_t hio_cmd; 72204076Spjd uint64_t hio_offset; 73204076Spjd uint64_t hio_length; 74204076Spjd TAILQ_ENTRY(hio) hio_next; 75204076Spjd}; 76204076Spjd 77211984Spjdstatic struct hast_resource *gres; 78211984Spjd 79204076Spjd/* 80204076Spjd * Free list holds unused structures. When free list is empty, we have to wait 81204076Spjd * until some in-progress requests are freed. 82204076Spjd */ 83204076Spjdstatic TAILQ_HEAD(, hio) hio_free_list; 84204076Spjdstatic pthread_mutex_t hio_free_list_lock; 85204076Spjdstatic pthread_cond_t hio_free_list_cond; 86204076Spjd/* 87204076Spjd * Disk thread (the one that do I/O requests) takes requests from this list. 88204076Spjd */ 89204076Spjdstatic TAILQ_HEAD(, hio) hio_disk_list; 90204076Spjdstatic pthread_mutex_t hio_disk_list_lock; 91204076Spjdstatic pthread_cond_t hio_disk_list_cond; 92204076Spjd/* 93204076Spjd * There is one recv list for every component, although local components don't 94204076Spjd * use recv lists as local requests are done synchronously. 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 109211877Spjd#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_signal(&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) 128226861Spjd 129226854Spjdstatic void 130226854Spjdhio_clear(struct hio *hio) 131226854Spjd{ 132211877Spjd 133226854Spjd hio->hio_seq = 0; 134226854Spjd hio->hio_error = 0; 135226854Spjd hio->hio_cmd = HIO_UNDEF; 136226854Spjd hio->hio_offset = 0; 137226854Spjd hio->hio_length = 0; 138226854Spjd} 139226854Spjd 140204076Spjdstatic void 141204076Spjdinit_environment(void) 142204076Spjd{ 143204076Spjd struct hio *hio; 144204076Spjd unsigned int ii; 145204076Spjd 146204076Spjd /* 147204076Spjd * Initialize lists, their locks and theirs condition variables. 148204076Spjd */ 149204076Spjd TAILQ_INIT(&hio_free_list); 150204076Spjd mtx_init(&hio_free_list_lock); 151204076Spjd cv_init(&hio_free_list_cond); 152204076Spjd TAILQ_INIT(&hio_disk_list); 153204076Spjd mtx_init(&hio_disk_list_lock); 154204076Spjd cv_init(&hio_disk_list_cond); 155204076Spjd TAILQ_INIT(&hio_send_list); 156204076Spjd mtx_init(&hio_send_list_lock); 157204076Spjd cv_init(&hio_send_list_cond); 158204076Spjd 159204076Spjd /* 160204076Spjd * Allocate requests pool and initialize requests. 161204076Spjd */ 162204076Spjd for (ii = 0; ii < HAST_HIO_MAX; ii++) { 163204076Spjd hio = malloc(sizeof(*hio)); 164204076Spjd if (hio == NULL) { 165210879Spjd pjdlog_exitx(EX_TEMPFAIL, 166210879Spjd "Unable to allocate memory (%zu bytes) for hio request.", 167210879Spjd sizeof(*hio)); 168204076Spjd } 169204076Spjd hio->hio_data = malloc(MAXPHYS); 170204076Spjd if (hio->hio_data == NULL) { 171210879Spjd pjdlog_exitx(EX_TEMPFAIL, 172210879Spjd "Unable to allocate memory (%zu bytes) for gctl_data.", 173210879Spjd (size_t)MAXPHYS); 174204076Spjd } 175226854Spjd hio_clear(hio); 176204076Spjd TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 177204076Spjd } 178204076Spjd} 179204076Spjd 180204076Spjdstatic void 181204076Spjdinit_local(struct hast_resource *res) 182204076Spjd{ 183204076Spjd 184229945Spjd if (metadata_read(res, true) == -1) 185204076Spjd exit(EX_NOINPUT); 186204076Spjd} 187204076Spjd 188204076Spjdstatic void 189204076Spjdinit_remote(struct hast_resource *res, struct nv *nvin) 190204076Spjd{ 191204076Spjd uint64_t resuid; 192204076Spjd struct nv *nvout; 193204076Spjd unsigned char *map; 194204076Spjd size_t mapsize; 195204076Spjd 196223181Strociny#ifdef notyet 197220271Spjd /* Setup direction. */ 198220271Spjd if (proto_send(res->hr_remoteout, NULL, 0) == -1) 199220271Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 200223181Strociny#endif 201220271Spjd 202204076Spjd nvout = nv_alloc(); 203204076Spjd nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 204204076Spjd nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 205204076Spjd resuid = nv_get_uint64(nvin, "resuid"); 206204076Spjd res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 207204076Spjd res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 208204076Spjd nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 209204076Spjd nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 210204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 211204076Spjd METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 212204076Spjd map = malloc(mapsize); 213204076Spjd if (map == NULL) { 214204076Spjd pjdlog_exitx(EX_TEMPFAIL, 215204076Spjd "Unable to allocate memory (%zu bytes) for activemap.", 216204076Spjd mapsize); 217204076Spjd } 218204076Spjd /* 219204076Spjd * When we work as primary and secondary is missing we will increase 220204076Spjd * localcnt in our metadata. When secondary is connected and synced 221204076Spjd * we make localcnt be equal to remotecnt, which means nodes are more 222204076Spjd * or less in sync. 223204076Spjd * Split-brain condition is when both nodes are not able to communicate 224204076Spjd * and are both configured as primary nodes. In turn, they can both 225204076Spjd * make incompatible changes to the data and we have to detect that. 226204076Spjd * Under split-brain condition we will increase our localcnt on first 227204076Spjd * write and remote node will increase its localcnt on first write. 228204076Spjd * When we connect we can see that primary's localcnt is greater than 229204076Spjd * our remotecnt (primary was modified while we weren't watching) and 230204076Spjd * our localcnt is greater than primary's remotecnt (we were modified 231204076Spjd * while primary wasn't watching). 232204076Spjd * There are many possible combinations which are all gathered below. 233204076Spjd * Don't pay too much attention to exact numbers, the more important 234204076Spjd * is to compare them. We compare secondary's local with primary's 235204076Spjd * remote and secondary's remote with primary's local. 236204076Spjd * Note that every case where primary's localcnt is smaller than 237204076Spjd * secondary's remotecnt and where secondary's localcnt is smaller than 238204076Spjd * primary's remotecnt should be impossible in practise. We will perform 239204076Spjd * full synchronization then. Those cases are marked with an asterisk. 240204076Spjd * Regular synchronization means that only extents marked as dirty are 241204076Spjd * synchronized (regular synchronization). 242204076Spjd * 243204076Spjd * SECONDARY METADATA PRIMARY METADATA 244204076Spjd * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 245204076Spjd * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 246204076Spjd * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 247204076Spjd * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 248204076Spjd * regular sync from secondary. 249204076Spjd * local=3 remote=3 local=3 remote=3 Regular sync just in case. 250204076Spjd * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 251204076Spjd * local=3 remote=3 local=4 remote=2 Split-brain condition. 252204076Spjd * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 253204076Spjd * regular sync from primary. 254204076Spjd * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 255204076Spjd */ 256204076Spjd if (res->hr_resuid == 0) { 257204076Spjd /* 258214284Spjd * Provider is used for the first time. If primary node done no 259214284Spjd * writes yet as well (we will find "virgin" argument) then 260214284Spjd * there is no need to synchronize anything. If primary node 261214284Spjd * done any writes already we have to synchronize everything. 262204076Spjd */ 263218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt == 0); 264204076Spjd res->hr_resuid = resuid; 265229945Spjd if (metadata_write(res) == -1) 266204076Spjd exit(EX_NOINPUT); 267214284Spjd if (nv_exists(nvin, "virgin")) { 268214284Spjd free(map); 269214284Spjd map = NULL; 270214284Spjd mapsize = 0; 271214284Spjd } else { 272214284Spjd memset(map, 0xff, mapsize); 273214284Spjd } 274220865Spjd nv_add_int8(nvout, 1, "virgin"); 275204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 276219830Spjd } else if (res->hr_resuid != resuid) { 277219830Spjd char errmsg[256]; 278219830Spjd 279226854Spjd free(map); 280219830Spjd (void)snprintf(errmsg, sizeof(errmsg), 281219830Spjd "Resource unique ID mismatch (primary=%ju, secondary=%ju).", 282219830Spjd (uintmax_t)resuid, (uintmax_t)res->hr_resuid); 283219830Spjd pjdlog_error("%s", errmsg); 284219830Spjd nv_add_string(nvout, errmsg, "errmsg"); 285230092Spjd if (hast_proto_send(res, res->hr_remotein, nvout, 286230092Spjd NULL, 0) == -1) { 287230092Spjd pjdlog_exit(EX_TEMPFAIL, 288230092Spjd "Unable to send response to %s", 289219830Spjd res->hr_remoteaddr); 290219830Spjd } 291219831Spjd nv_free(nvout); 292219830Spjd exit(EX_CONFIG); 293204076Spjd } else if ( 294226842Spjd /* Is primary out-of-date? */ 295204076Spjd (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 296204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 297226842Spjd /* Are the nodes more or less in sync? */ 298204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 299204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 300226842Spjd /* Is secondary out-of-date? */ 301204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 302204076Spjd res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 303204076Spjd /* 304204076Spjd * Nodes are more or less in sync or one of the nodes is 305204076Spjd * out-of-date. 306204076Spjd * It doesn't matter at this point which one, we just have to 307204076Spjd * send out local bitmap to the remote node. 308204076Spjd */ 309204076Spjd if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 310204076Spjd (ssize_t)mapsize) { 311204076Spjd pjdlog_exit(LOG_ERR, "Unable to read activemap"); 312204076Spjd } 313204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 314204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 315204076Spjd /* Primary is out-of-date, sync from secondary. */ 316204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 317204076Spjd } else { 318204076Spjd /* 319204076Spjd * Secondary is out-of-date or counts match. 320204076Spjd * Sync from primary. 321204076Spjd */ 322204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 323204076Spjd } 324204076Spjd } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 325204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 326204076Spjd /* 327204076Spjd * Not good, we have split-brain condition. 328204076Spjd */ 329226854Spjd free(map); 330204076Spjd pjdlog_error("Split-brain detected, exiting."); 331204076Spjd nv_add_string(nvout, "Split-brain condition!", "errmsg"); 332230092Spjd if (hast_proto_send(res, res->hr_remotein, nvout, 333230092Spjd NULL, 0) == -1) { 334230092Spjd pjdlog_exit(EX_TEMPFAIL, 335230092Spjd "Unable to send response to %s", 336226854Spjd res->hr_remoteaddr); 337226854Spjd } 338226854Spjd nv_free(nvout); 339226854Spjd /* Exit on split-brain. */ 340226854Spjd event_send(res, EVENT_SPLITBRAIN); 341226854Spjd exit(EX_CONFIG); 342204076Spjd } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 343204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 344204076Spjd /* 345204076Spjd * This should never happen in practise, but we will perform 346204076Spjd * full synchronization. 347204076Spjd */ 348218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 349204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt); 350204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 351204076Spjd METADATA_SIZE, res->hr_extentsize, 352204076Spjd res->hr_local_sectorsize); 353204076Spjd memset(map, 0xff, mapsize); 354204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 355204076Spjd /* In this one of five cases sync from secondary. */ 356204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 357204076Spjd } else { 358204076Spjd /* For the rest four cases sync from primary. */ 359204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 360204076Spjd } 361204076Spjd pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 362204076Spjd (uintmax_t)res->hr_primary_localcnt, 363204076Spjd (uintmax_t)res->hr_primary_remotecnt, 364204076Spjd (uintmax_t)res->hr_secondary_localcnt, 365204076Spjd (uintmax_t)res->hr_secondary_remotecnt); 366204076Spjd } 367220007Spjd nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 368229945Spjd if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) { 369214276Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 370204076Spjd res->hr_remoteaddr); 371204076Spjd } 372214275Spjd if (map != NULL) 373214275Spjd free(map); 374209182Spjd nv_free(nvout); 375223181Strociny#ifdef notyet 376220271Spjd /* Setup direction. */ 377220271Spjd if (proto_recv(res->hr_remotein, NULL, 0) == -1) 378220271Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 379223181Strociny#endif 380204076Spjd} 381204076Spjd 382204076Spjdvoid 383204076Spjdhastd_secondary(struct hast_resource *res, struct nv *nvin) 384204076Spjd{ 385213009Spjd sigset_t mask; 386204076Spjd pthread_t td; 387204076Spjd pid_t pid; 388219482Strociny int error, mode, debuglevel; 389204076Spjd 390204076Spjd /* 391204076Spjd * Create communication channel between parent and child. 392204076Spjd */ 393229945Spjd if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) { 394204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 395204076Spjd pjdlog_exit(EX_OSERR, 396204076Spjd "Unable to create control sockets between parent and child"); 397204076Spjd } 398212038Spjd /* 399212038Spjd * Create communication channel between child and parent. 400212038Spjd */ 401229945Spjd if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) { 402212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 403212038Spjd pjdlog_exit(EX_OSERR, 404212038Spjd "Unable to create event sockets between child and parent"); 405212038Spjd } 406204076Spjd 407204076Spjd pid = fork(); 408229744Spjd if (pid == -1) { 409204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 410204076Spjd pjdlog_exit(EX_OSERR, "Unable to fork"); 411204076Spjd } 412204076Spjd 413204076Spjd if (pid > 0) { 414204076Spjd /* This is parent. */ 415204076Spjd proto_close(res->hr_remotein); 416204076Spjd res->hr_remotein = NULL; 417204076Spjd proto_close(res->hr_remoteout); 418204076Spjd res->hr_remoteout = NULL; 419212038Spjd /* Declare that we are receiver. */ 420212038Spjd proto_recv(res->hr_event, NULL, 0); 421218043Spjd /* Declare that we are sender. */ 422218043Spjd proto_send(res->hr_ctrl, NULL, 0); 423204076Spjd res->hr_workerpid = pid; 424204076Spjd return; 425204076Spjd } 426211977Spjd 427211984Spjd gres = res; 428218043Spjd mode = pjdlog_mode_get(); 429219482Strociny debuglevel = pjdlog_debug_get(); 430211984Spjd 431218043Spjd /* Declare that we are sender. */ 432218043Spjd proto_send(res->hr_event, NULL, 0); 433218043Spjd /* Declare that we are receiver. */ 434218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 435218043Spjd descriptors_cleanup(res); 436204076Spjd 437218045Spjd descriptors_assert(res, mode); 438218045Spjd 439218043Spjd pjdlog_init(mode); 440219482Strociny pjdlog_debug_set(debuglevel); 441218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 442220005Spjd setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role)); 443204076Spjd 444213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 445213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 446210880Spjd 447207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 448229945Spjd if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1) 449207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 450229945Spjd if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1) 451207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 452207371Spjd 453204076Spjd init_local(res); 454213007Spjd init_environment(); 455213007Spjd 456221899Spjd if (drop_privs(res) != 0) 457218049Spjd exit(EX_CONFIG); 458218214Spjd pjdlog_info("Privileges successfully dropped."); 459218049Spjd 460213007Spjd /* 461213007Spjd * Create the control thread before sending any event to the parent, 462213007Spjd * as we can deadlock when parent sends control request to worker, 463213007Spjd * but worker has no control thread started yet, so parent waits. 464213007Spjd * In the meantime worker sends an event to the parent, but parent 465213007Spjd * is unable to handle the event, because it waits for control 466213007Spjd * request response. 467213007Spjd */ 468213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 469218138Spjd PJDLOG_ASSERT(error == 0); 470213007Spjd 471204076Spjd init_remote(res, nvin); 472212038Spjd event_send(res, EVENT_CONNECT); 473204076Spjd 474204076Spjd error = pthread_create(&td, NULL, recv_thread, res); 475218138Spjd PJDLOG_ASSERT(error == 0); 476204076Spjd error = pthread_create(&td, NULL, disk_thread, res); 477218138Spjd PJDLOG_ASSERT(error == 0); 478213007Spjd (void)send_thread(res); 479204076Spjd} 480204076Spjd 481204076Spjdstatic void 482230092Spjdreqlog(int loglevel, int debuglevel, int error, struct hio *hio, 483230092Spjd const char *fmt, ...) 484204076Spjd{ 485204076Spjd char msg[1024]; 486204076Spjd va_list ap; 487204076Spjd int len; 488204076Spjd 489204076Spjd va_start(ap, fmt); 490204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 491204076Spjd va_end(ap); 492204076Spjd if ((size_t)len < sizeof(msg)) { 493204076Spjd switch (hio->hio_cmd) { 494204076Spjd case HIO_READ: 495204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 496204076Spjd "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 497204076Spjd (uintmax_t)hio->hio_length); 498204076Spjd break; 499204076Spjd case HIO_DELETE: 500204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 501204076Spjd "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 502204076Spjd (uintmax_t)hio->hio_length); 503204076Spjd break; 504204076Spjd case HIO_FLUSH: 505204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 506204076Spjd break; 507204076Spjd case HIO_WRITE: 508204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 509204076Spjd "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 510204076Spjd (uintmax_t)hio->hio_length); 511204076Spjd break; 512211882Spjd case HIO_KEEPALIVE: 513211882Spjd (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 514211882Spjd break; 515204076Spjd default: 516204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 517204076Spjd "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 518204076Spjd break; 519204076Spjd } 520204076Spjd } 521204076Spjd pjdlog_common(loglevel, debuglevel, error, "%s", msg); 522204076Spjd} 523204076Spjd 524204076Spjdstatic int 525226854Spjdrequnpack(struct hast_resource *res, struct hio *hio, struct nv *nv) 526204076Spjd{ 527204076Spjd 528226854Spjd hio->hio_cmd = nv_get_uint8(nv, "cmd"); 529204076Spjd if (hio->hio_cmd == 0) { 530204076Spjd pjdlog_error("Header contains no 'cmd' field."); 531204076Spjd hio->hio_error = EINVAL; 532204076Spjd goto end; 533204076Spjd } 534226854Spjd if (hio->hio_cmd != HIO_KEEPALIVE) { 535226854Spjd hio->hio_seq = nv_get_uint64(nv, "seq"); 536226854Spjd if (hio->hio_seq == 0) { 537226854Spjd pjdlog_error("Header contains no 'seq' field."); 538226854Spjd hio->hio_error = EINVAL; 539226854Spjd goto end; 540226854Spjd } 541226854Spjd } 542204076Spjd switch (hio->hio_cmd) { 543222164Spjd case HIO_FLUSH: 544211882Spjd case HIO_KEEPALIVE: 545211882Spjd break; 546204076Spjd case HIO_READ: 547204076Spjd case HIO_WRITE: 548204076Spjd case HIO_DELETE: 549226854Spjd hio->hio_offset = nv_get_uint64(nv, "offset"); 550226854Spjd if (nv_error(nv) != 0) { 551204076Spjd pjdlog_error("Header is missing 'offset' field."); 552204076Spjd hio->hio_error = EINVAL; 553204076Spjd goto end; 554204076Spjd } 555226854Spjd hio->hio_length = nv_get_uint64(nv, "length"); 556226854Spjd if (nv_error(nv) != 0) { 557204076Spjd pjdlog_error("Header is missing 'length' field."); 558204076Spjd hio->hio_error = EINVAL; 559204076Spjd goto end; 560204076Spjd } 561204076Spjd if (hio->hio_length == 0) { 562204076Spjd pjdlog_error("Data length is zero."); 563204076Spjd hio->hio_error = EINVAL; 564204076Spjd goto end; 565204076Spjd } 566204076Spjd if (hio->hio_length > MAXPHYS) { 567204076Spjd pjdlog_error("Data length is too large (%ju > %ju).", 568204076Spjd (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 569204076Spjd hio->hio_error = EINVAL; 570204076Spjd goto end; 571204076Spjd } 572204076Spjd if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 573204076Spjd pjdlog_error("Offset %ju is not multiple of sector size.", 574204076Spjd (uintmax_t)hio->hio_offset); 575204076Spjd hio->hio_error = EINVAL; 576204076Spjd goto end; 577204076Spjd } 578204076Spjd if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 579204076Spjd pjdlog_error("Length %ju is not multiple of sector size.", 580204076Spjd (uintmax_t)hio->hio_length); 581204076Spjd hio->hio_error = EINVAL; 582204076Spjd goto end; 583204076Spjd } 584204076Spjd if (hio->hio_offset + hio->hio_length > 585204076Spjd (uint64_t)res->hr_datasize) { 586204076Spjd pjdlog_error("Data offset is too large (%ju > %ju).", 587204076Spjd (uintmax_t)(hio->hio_offset + hio->hio_length), 588204076Spjd (uintmax_t)res->hr_datasize); 589204076Spjd hio->hio_error = EINVAL; 590204076Spjd goto end; 591204076Spjd } 592204076Spjd break; 593204076Spjd default: 594204076Spjd pjdlog_error("Header contains invalid 'cmd' (%hhu).", 595204076Spjd hio->hio_cmd); 596204076Spjd hio->hio_error = EINVAL; 597204076Spjd goto end; 598204076Spjd } 599204076Spjd hio->hio_error = 0; 600204076Spjdend: 601204076Spjd return (hio->hio_error); 602204076Spjd} 603204076Spjd 604212899Spjdstatic __dead2 void 605211984Spjdsecondary_exit(int exitcode, const char *fmt, ...) 606211984Spjd{ 607211984Spjd va_list ap; 608211984Spjd 609218138Spjd PJDLOG_ASSERT(exitcode != EX_OK); 610211984Spjd va_start(ap, fmt); 611211984Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 612211984Spjd va_end(ap); 613212038Spjd event_send(gres, EVENT_DISCONNECT); 614211984Spjd exit(exitcode); 615211984Spjd} 616211984Spjd 617204076Spjd/* 618204076Spjd * Thread receives requests from the primary node. 619204076Spjd */ 620204076Spjdstatic void * 621204076Spjdrecv_thread(void *arg) 622204076Spjd{ 623204076Spjd struct hast_resource *res = arg; 624204076Spjd struct hio *hio; 625226854Spjd struct nv *nv; 626204076Spjd 627204076Spjd for (;;) { 628204076Spjd pjdlog_debug(2, "recv: Taking free request."); 629211877Spjd QUEUE_TAKE(free, hio); 630204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", hio); 631229945Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) { 632211984Spjd secondary_exit(EX_TEMPFAIL, 633204076Spjd "Unable to receive request header"); 634204076Spjd } 635226854Spjd if (requnpack(res, hio, nv) != 0) { 636226854Spjd nv_free(nv); 637211877Spjd pjdlog_debug(2, 638211877Spjd "recv: (%p) Moving request to the send queue.", 639211877Spjd hio); 640211877Spjd QUEUE_INSERT(send, hio); 641211877Spjd continue; 642211877Spjd } 643222228Spjd switch (hio->hio_cmd) { 644222228Spjd case HIO_READ: 645222228Spjd res->hr_stat_read++; 646222228Spjd break; 647222228Spjd case HIO_WRITE: 648222228Spjd res->hr_stat_write++; 649222228Spjd break; 650222228Spjd case HIO_DELETE: 651222228Spjd res->hr_stat_delete++; 652222228Spjd break; 653222228Spjd case HIO_FLUSH: 654222228Spjd res->hr_stat_flush++; 655222228Spjd break; 656226854Spjd case HIO_KEEPALIVE: 657226854Spjd break; 658226854Spjd default: 659226854Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 660226854Spjd hio->hio_cmd); 661222228Spjd } 662204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, 663204076Spjd "recv: (%p) Got request header: ", hio); 664211882Spjd if (hio->hio_cmd == HIO_KEEPALIVE) { 665226854Spjd nv_free(nv); 666211882Spjd pjdlog_debug(2, 667211882Spjd "recv: (%p) Moving request to the free queue.", 668211882Spjd hio); 669226854Spjd hio_clear(hio); 670211882Spjd QUEUE_INSERT(free, hio); 671211882Spjd continue; 672211882Spjd } else if (hio->hio_cmd == HIO_WRITE) { 673226854Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 674229945Spjd hio->hio_data, MAXPHYS) == -1) { 675211984Spjd secondary_exit(EX_TEMPFAIL, 676212051Spjd "Unable to receive request data"); 677204076Spjd } 678204076Spjd } 679226854Spjd nv_free(nv); 680204076Spjd pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 681204076Spjd hio); 682211877Spjd QUEUE_INSERT(disk, hio); 683204076Spjd } 684204076Spjd /* NOTREACHED */ 685204076Spjd return (NULL); 686204076Spjd} 687204076Spjd 688204076Spjd/* 689204076Spjd * Thread reads from or writes to local component and also handles DELETE and 690204076Spjd * FLUSH requests. 691204076Spjd */ 692204076Spjdstatic void * 693204076Spjddisk_thread(void *arg) 694204076Spjd{ 695204076Spjd struct hast_resource *res = arg; 696204076Spjd struct hio *hio; 697204076Spjd ssize_t ret; 698225832Spjd bool clear_activemap, logerror; 699204076Spjd 700204076Spjd clear_activemap = true; 701204076Spjd 702204076Spjd for (;;) { 703204076Spjd pjdlog_debug(2, "disk: Taking request."); 704211877Spjd QUEUE_TAKE(disk, hio); 705204076Spjd while (clear_activemap) { 706204076Spjd unsigned char *map; 707204076Spjd size_t mapsize; 708204076Spjd 709204076Spjd /* 710204076Spjd * When first request is received, it means that primary 711204076Spjd * already received our activemap, merged it and stored 712204076Spjd * locally. We can now safely clear our activemap. 713204076Spjd */ 714204076Spjd mapsize = 715204076Spjd activemap_calc_ondisk_size(res->hr_local_mediasize - 716204076Spjd METADATA_SIZE, res->hr_extentsize, 717204076Spjd res->hr_local_sectorsize); 718204076Spjd map = calloc(1, mapsize); 719204076Spjd if (map == NULL) { 720204076Spjd pjdlog_warning("Unable to allocate memory to clear local activemap."); 721204076Spjd break; 722204076Spjd } 723204076Spjd if (pwrite(res->hr_localfd, map, mapsize, 724204076Spjd METADATA_SIZE) != (ssize_t)mapsize) { 725204076Spjd pjdlog_errno(LOG_WARNING, 726204076Spjd "Unable to store cleared activemap"); 727204076Spjd free(map); 728204076Spjd break; 729204076Spjd } 730204076Spjd free(map); 731204076Spjd clear_activemap = false; 732204076Spjd pjdlog_debug(1, "Local activemap cleared."); 733225831Spjd break; 734204076Spjd } 735204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 736225832Spjd logerror = true; 737204076Spjd /* Handle the actual request. */ 738204076Spjd switch (hio->hio_cmd) { 739204076Spjd case HIO_READ: 740204076Spjd ret = pread(res->hr_localfd, hio->hio_data, 741204076Spjd hio->hio_length, 742204076Spjd hio->hio_offset + res->hr_localoff); 743229945Spjd if (ret == -1) 744204076Spjd hio->hio_error = errno; 745204076Spjd else if (ret != (int64_t)hio->hio_length) 746204076Spjd hio->hio_error = EIO; 747204076Spjd else 748204076Spjd hio->hio_error = 0; 749204076Spjd break; 750204076Spjd case HIO_WRITE: 751204076Spjd ret = pwrite(res->hr_localfd, hio->hio_data, 752204076Spjd hio->hio_length, 753204076Spjd hio->hio_offset + res->hr_localoff); 754229945Spjd if (ret == -1) 755204076Spjd hio->hio_error = errno; 756204076Spjd else if (ret != (int64_t)hio->hio_length) 757204076Spjd hio->hio_error = EIO; 758204076Spjd else 759204076Spjd hio->hio_error = 0; 760204076Spjd break; 761204076Spjd case HIO_DELETE: 762204076Spjd ret = g_delete(res->hr_localfd, 763204076Spjd hio->hio_offset + res->hr_localoff, 764204076Spjd hio->hio_length); 765229945Spjd if (ret == -1) 766204076Spjd hio->hio_error = errno; 767204076Spjd else 768204076Spjd hio->hio_error = 0; 769204076Spjd break; 770204076Spjd case HIO_FLUSH: 771225832Spjd if (!res->hr_localflush) { 772225832Spjd ret = -1; 773225832Spjd hio->hio_error = EOPNOTSUPP; 774225832Spjd logerror = false; 775225832Spjd break; 776225832Spjd } 777204076Spjd ret = g_flush(res->hr_localfd); 778229945Spjd if (ret == -1) { 779225832Spjd if (errno == EOPNOTSUPP) 780225832Spjd res->hr_localflush = false; 781204076Spjd hio->hio_error = errno; 782225832Spjd } else { 783204076Spjd hio->hio_error = 0; 784225832Spjd } 785204076Spjd break; 786226854Spjd default: 787226854Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 788226854Spjd hio->hio_cmd); 789204076Spjd } 790225832Spjd if (logerror && hio->hio_error != 0) { 791204076Spjd reqlog(LOG_ERR, 0, hio->hio_error, hio, 792204076Spjd "Request failed: "); 793204076Spjd } 794204076Spjd pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 795204076Spjd hio); 796211877Spjd QUEUE_INSERT(send, hio); 797204076Spjd } 798204076Spjd /* NOTREACHED */ 799204076Spjd return (NULL); 800204076Spjd} 801204076Spjd 802204076Spjd/* 803204076Spjd * Thread sends requests back to primary node. 804204076Spjd */ 805204076Spjdstatic void * 806204076Spjdsend_thread(void *arg) 807204076Spjd{ 808204076Spjd struct hast_resource *res = arg; 809204076Spjd struct nv *nvout; 810204076Spjd struct hio *hio; 811204076Spjd void *data; 812204076Spjd size_t length; 813204076Spjd 814204076Spjd for (;;) { 815204076Spjd pjdlog_debug(2, "send: Taking request."); 816211877Spjd QUEUE_TAKE(send, hio); 817204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 818204076Spjd nvout = nv_alloc(); 819204076Spjd /* Copy sequence number. */ 820226854Spjd nv_add_uint64(nvout, hio->hio_seq, "seq"); 821204076Spjd switch (hio->hio_cmd) { 822204076Spjd case HIO_READ: 823204076Spjd if (hio->hio_error == 0) { 824204076Spjd data = hio->hio_data; 825204076Spjd length = hio->hio_length; 826204076Spjd break; 827204076Spjd } 828204076Spjd /* 829204076Spjd * We send no data in case of an error. 830204076Spjd */ 831204076Spjd /* FALLTHROUGH */ 832204076Spjd case HIO_DELETE: 833204076Spjd case HIO_FLUSH: 834204076Spjd case HIO_WRITE: 835204076Spjd data = NULL; 836204076Spjd length = 0; 837204076Spjd break; 838204076Spjd default: 839225782Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 840225782Spjd hio->hio_cmd); 841204076Spjd } 842204076Spjd if (hio->hio_error != 0) 843204076Spjd nv_add_int16(nvout, hio->hio_error, "error"); 844204076Spjd if (hast_proto_send(res, res->hr_remoteout, nvout, data, 845229945Spjd length) == -1) { 846230092Spjd secondary_exit(EX_TEMPFAIL, "Unable to send reply"); 847204076Spjd } 848204076Spjd nv_free(nvout); 849209185Spjd pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 850204076Spjd hio); 851226854Spjd hio_clear(hio); 852211877Spjd QUEUE_INSERT(free, hio); 853204076Spjd } 854204076Spjd /* NOTREACHED */ 855204076Spjd return (NULL); 856204076Spjd} 857