secondary.c revision 229945
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 229945 2012-01-10 22:39:07Z 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"); 285229945Spjd if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) == -1) { 286219830Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s", 287219830Spjd res->hr_remoteaddr); 288219830Spjd } 289219831Spjd nv_free(nvout); 290219830Spjd exit(EX_CONFIG); 291204076Spjd } else if ( 292226842Spjd /* Is primary out-of-date? */ 293204076Spjd (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 294204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 295226842Spjd /* Are the nodes more or less in sync? */ 296204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 297204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 298226842Spjd /* Is secondary out-of-date? */ 299204076Spjd (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 300204076Spjd res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 301204076Spjd /* 302204076Spjd * Nodes are more or less in sync or one of the nodes is 303204076Spjd * out-of-date. 304204076Spjd * It doesn't matter at this point which one, we just have to 305204076Spjd * send out local bitmap to the remote node. 306204076Spjd */ 307204076Spjd if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 308204076Spjd (ssize_t)mapsize) { 309204076Spjd pjdlog_exit(LOG_ERR, "Unable to read activemap"); 310204076Spjd } 311204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 312204076Spjd res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 313204076Spjd /* Primary is out-of-date, sync from secondary. */ 314204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 315204076Spjd } else { 316204076Spjd /* 317204076Spjd * Secondary is out-of-date or counts match. 318204076Spjd * Sync from primary. 319204076Spjd */ 320204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 321204076Spjd } 322204076Spjd } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 323204076Spjd res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 324204076Spjd /* 325204076Spjd * Not good, we have split-brain condition. 326204076Spjd */ 327226854Spjd free(map); 328204076Spjd pjdlog_error("Split-brain detected, exiting."); 329204076Spjd nv_add_string(nvout, "Split-brain condition!", "errmsg"); 330229945Spjd if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) == -1) { 331226854Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s", 332226854Spjd res->hr_remoteaddr); 333226854Spjd } 334226854Spjd nv_free(nvout); 335226854Spjd /* Exit on split-brain. */ 336226854Spjd event_send(res, EVENT_SPLITBRAIN); 337226854Spjd exit(EX_CONFIG); 338204076Spjd } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 339204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 340204076Spjd /* 341204076Spjd * This should never happen in practise, but we will perform 342204076Spjd * full synchronization. 343204076Spjd */ 344218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 345204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt); 346204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 347204076Spjd METADATA_SIZE, res->hr_extentsize, 348204076Spjd res->hr_local_sectorsize); 349204076Spjd memset(map, 0xff, mapsize); 350204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 351204076Spjd /* In this one of five cases sync from secondary. */ 352204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 353204076Spjd } else { 354204076Spjd /* For the rest four cases sync from primary. */ 355204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 356204076Spjd } 357204076Spjd pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 358204076Spjd (uintmax_t)res->hr_primary_localcnt, 359204076Spjd (uintmax_t)res->hr_primary_remotecnt, 360204076Spjd (uintmax_t)res->hr_secondary_localcnt, 361204076Spjd (uintmax_t)res->hr_secondary_remotecnt); 362204076Spjd } 363220007Spjd nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 364229945Spjd if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) == -1) { 365214276Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 366204076Spjd res->hr_remoteaddr); 367204076Spjd } 368214275Spjd if (map != NULL) 369214275Spjd free(map); 370209182Spjd nv_free(nvout); 371223181Strociny#ifdef notyet 372220271Spjd /* Setup direction. */ 373220271Spjd if (proto_recv(res->hr_remotein, NULL, 0) == -1) 374220271Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 375223181Strociny#endif 376204076Spjd} 377204076Spjd 378204076Spjdvoid 379204076Spjdhastd_secondary(struct hast_resource *res, struct nv *nvin) 380204076Spjd{ 381213009Spjd sigset_t mask; 382204076Spjd pthread_t td; 383204076Spjd pid_t pid; 384219482Strociny int error, mode, debuglevel; 385204076Spjd 386204076Spjd /* 387204076Spjd * Create communication channel between parent and child. 388204076Spjd */ 389229945Spjd if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) { 390204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 391204076Spjd pjdlog_exit(EX_OSERR, 392204076Spjd "Unable to create control sockets between parent and child"); 393204076Spjd } 394212038Spjd /* 395212038Spjd * Create communication channel between child and parent. 396212038Spjd */ 397229945Spjd if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) { 398212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 399212038Spjd pjdlog_exit(EX_OSERR, 400212038Spjd "Unable to create event sockets between child and parent"); 401212038Spjd } 402204076Spjd 403204076Spjd pid = fork(); 404229744Spjd if (pid == -1) { 405204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 406204076Spjd pjdlog_exit(EX_OSERR, "Unable to fork"); 407204076Spjd } 408204076Spjd 409204076Spjd if (pid > 0) { 410204076Spjd /* This is parent. */ 411204076Spjd proto_close(res->hr_remotein); 412204076Spjd res->hr_remotein = NULL; 413204076Spjd proto_close(res->hr_remoteout); 414204076Spjd res->hr_remoteout = NULL; 415212038Spjd /* Declare that we are receiver. */ 416212038Spjd proto_recv(res->hr_event, NULL, 0); 417218043Spjd /* Declare that we are sender. */ 418218043Spjd proto_send(res->hr_ctrl, NULL, 0); 419204076Spjd res->hr_workerpid = pid; 420204076Spjd return; 421204076Spjd } 422211977Spjd 423211984Spjd gres = res; 424218043Spjd mode = pjdlog_mode_get(); 425219482Strociny debuglevel = pjdlog_debug_get(); 426211984Spjd 427218043Spjd /* Declare that we are sender. */ 428218043Spjd proto_send(res->hr_event, NULL, 0); 429218043Spjd /* Declare that we are receiver. */ 430218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 431218043Spjd descriptors_cleanup(res); 432204076Spjd 433218045Spjd descriptors_assert(res, mode); 434218045Spjd 435218043Spjd pjdlog_init(mode); 436219482Strociny pjdlog_debug_set(debuglevel); 437218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 438220005Spjd setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role)); 439204076Spjd 440213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 441213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 442210880Spjd 443207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 444229945Spjd if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) == -1) 445207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 446229945Spjd if (proto_timeout(res->hr_remoteout, res->hr_timeout) == -1) 447207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 448207371Spjd 449204076Spjd init_local(res); 450213007Spjd init_environment(); 451213007Spjd 452221899Spjd if (drop_privs(res) != 0) 453218049Spjd exit(EX_CONFIG); 454218214Spjd pjdlog_info("Privileges successfully dropped."); 455218049Spjd 456213007Spjd /* 457213007Spjd * Create the control thread before sending any event to the parent, 458213007Spjd * as we can deadlock when parent sends control request to worker, 459213007Spjd * but worker has no control thread started yet, so parent waits. 460213007Spjd * In the meantime worker sends an event to the parent, but parent 461213007Spjd * is unable to handle the event, because it waits for control 462213007Spjd * request response. 463213007Spjd */ 464213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 465218138Spjd PJDLOG_ASSERT(error == 0); 466213007Spjd 467204076Spjd init_remote(res, nvin); 468212038Spjd event_send(res, EVENT_CONNECT); 469204076Spjd 470204076Spjd error = pthread_create(&td, NULL, recv_thread, res); 471218138Spjd PJDLOG_ASSERT(error == 0); 472204076Spjd error = pthread_create(&td, NULL, disk_thread, res); 473218138Spjd PJDLOG_ASSERT(error == 0); 474213007Spjd (void)send_thread(res); 475204076Spjd} 476204076Spjd 477204076Spjdstatic void 478204076Spjdreqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 479204076Spjd{ 480204076Spjd char msg[1024]; 481204076Spjd va_list ap; 482204076Spjd int len; 483204076Spjd 484204076Spjd va_start(ap, fmt); 485204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 486204076Spjd va_end(ap); 487204076Spjd if ((size_t)len < sizeof(msg)) { 488204076Spjd switch (hio->hio_cmd) { 489204076Spjd case HIO_READ: 490204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 491204076Spjd "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 492204076Spjd (uintmax_t)hio->hio_length); 493204076Spjd break; 494204076Spjd case HIO_DELETE: 495204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 496204076Spjd "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 497204076Spjd (uintmax_t)hio->hio_length); 498204076Spjd break; 499204076Spjd case HIO_FLUSH: 500204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 501204076Spjd break; 502204076Spjd case HIO_WRITE: 503204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 504204076Spjd "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 505204076Spjd (uintmax_t)hio->hio_length); 506204076Spjd break; 507211882Spjd case HIO_KEEPALIVE: 508211882Spjd (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 509211882Spjd break; 510204076Spjd default: 511204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 512204076Spjd "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 513204076Spjd break; 514204076Spjd } 515204076Spjd } 516204076Spjd pjdlog_common(loglevel, debuglevel, error, "%s", msg); 517204076Spjd} 518204076Spjd 519204076Spjdstatic int 520226854Spjdrequnpack(struct hast_resource *res, struct hio *hio, struct nv *nv) 521204076Spjd{ 522204076Spjd 523226854Spjd hio->hio_cmd = nv_get_uint8(nv, "cmd"); 524204076Spjd if (hio->hio_cmd == 0) { 525204076Spjd pjdlog_error("Header contains no 'cmd' field."); 526204076Spjd hio->hio_error = EINVAL; 527204076Spjd goto end; 528204076Spjd } 529226854Spjd if (hio->hio_cmd != HIO_KEEPALIVE) { 530226854Spjd hio->hio_seq = nv_get_uint64(nv, "seq"); 531226854Spjd if (hio->hio_seq == 0) { 532226854Spjd pjdlog_error("Header contains no 'seq' field."); 533226854Spjd hio->hio_error = EINVAL; 534226854Spjd goto end; 535226854Spjd } 536226854Spjd } 537204076Spjd switch (hio->hio_cmd) { 538222164Spjd case HIO_FLUSH: 539211882Spjd case HIO_KEEPALIVE: 540211882Spjd break; 541204076Spjd case HIO_READ: 542204076Spjd case HIO_WRITE: 543204076Spjd case HIO_DELETE: 544226854Spjd hio->hio_offset = nv_get_uint64(nv, "offset"); 545226854Spjd if (nv_error(nv) != 0) { 546204076Spjd pjdlog_error("Header is missing 'offset' field."); 547204076Spjd hio->hio_error = EINVAL; 548204076Spjd goto end; 549204076Spjd } 550226854Spjd hio->hio_length = nv_get_uint64(nv, "length"); 551226854Spjd if (nv_error(nv) != 0) { 552204076Spjd pjdlog_error("Header is missing 'length' field."); 553204076Spjd hio->hio_error = EINVAL; 554204076Spjd goto end; 555204076Spjd } 556204076Spjd if (hio->hio_length == 0) { 557204076Spjd pjdlog_error("Data length is zero."); 558204076Spjd hio->hio_error = EINVAL; 559204076Spjd goto end; 560204076Spjd } 561204076Spjd if (hio->hio_length > MAXPHYS) { 562204076Spjd pjdlog_error("Data length is too large (%ju > %ju).", 563204076Spjd (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 564204076Spjd hio->hio_error = EINVAL; 565204076Spjd goto end; 566204076Spjd } 567204076Spjd if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 568204076Spjd pjdlog_error("Offset %ju is not multiple of sector size.", 569204076Spjd (uintmax_t)hio->hio_offset); 570204076Spjd hio->hio_error = EINVAL; 571204076Spjd goto end; 572204076Spjd } 573204076Spjd if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 574204076Spjd pjdlog_error("Length %ju is not multiple of sector size.", 575204076Spjd (uintmax_t)hio->hio_length); 576204076Spjd hio->hio_error = EINVAL; 577204076Spjd goto end; 578204076Spjd } 579204076Spjd if (hio->hio_offset + hio->hio_length > 580204076Spjd (uint64_t)res->hr_datasize) { 581204076Spjd pjdlog_error("Data offset is too large (%ju > %ju).", 582204076Spjd (uintmax_t)(hio->hio_offset + hio->hio_length), 583204076Spjd (uintmax_t)res->hr_datasize); 584204076Spjd hio->hio_error = EINVAL; 585204076Spjd goto end; 586204076Spjd } 587204076Spjd break; 588204076Spjd default: 589204076Spjd pjdlog_error("Header contains invalid 'cmd' (%hhu).", 590204076Spjd hio->hio_cmd); 591204076Spjd hio->hio_error = EINVAL; 592204076Spjd goto end; 593204076Spjd } 594204076Spjd hio->hio_error = 0; 595204076Spjdend: 596204076Spjd return (hio->hio_error); 597204076Spjd} 598204076Spjd 599212899Spjdstatic __dead2 void 600211984Spjdsecondary_exit(int exitcode, const char *fmt, ...) 601211984Spjd{ 602211984Spjd va_list ap; 603211984Spjd 604218138Spjd PJDLOG_ASSERT(exitcode != EX_OK); 605211984Spjd va_start(ap, fmt); 606211984Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 607211984Spjd va_end(ap); 608212038Spjd event_send(gres, EVENT_DISCONNECT); 609211984Spjd exit(exitcode); 610211984Spjd} 611211984Spjd 612204076Spjd/* 613204076Spjd * Thread receives requests from the primary node. 614204076Spjd */ 615204076Spjdstatic void * 616204076Spjdrecv_thread(void *arg) 617204076Spjd{ 618204076Spjd struct hast_resource *res = arg; 619204076Spjd struct hio *hio; 620226854Spjd struct nv *nv; 621204076Spjd 622204076Spjd for (;;) { 623204076Spjd pjdlog_debug(2, "recv: Taking free request."); 624211877Spjd QUEUE_TAKE(free, hio); 625204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", hio); 626229945Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) { 627211984Spjd secondary_exit(EX_TEMPFAIL, 628204076Spjd "Unable to receive request header"); 629204076Spjd } 630226854Spjd if (requnpack(res, hio, nv) != 0) { 631226854Spjd nv_free(nv); 632211877Spjd pjdlog_debug(2, 633211877Spjd "recv: (%p) Moving request to the send queue.", 634211877Spjd hio); 635211877Spjd QUEUE_INSERT(send, hio); 636211877Spjd continue; 637211877Spjd } 638222228Spjd switch (hio->hio_cmd) { 639222228Spjd case HIO_READ: 640222228Spjd res->hr_stat_read++; 641222228Spjd break; 642222228Spjd case HIO_WRITE: 643222228Spjd res->hr_stat_write++; 644222228Spjd break; 645222228Spjd case HIO_DELETE: 646222228Spjd res->hr_stat_delete++; 647222228Spjd break; 648222228Spjd case HIO_FLUSH: 649222228Spjd res->hr_stat_flush++; 650222228Spjd break; 651226854Spjd case HIO_KEEPALIVE: 652226854Spjd break; 653226854Spjd default: 654226854Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 655226854Spjd hio->hio_cmd); 656222228Spjd } 657204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, 658204076Spjd "recv: (%p) Got request header: ", hio); 659211882Spjd if (hio->hio_cmd == HIO_KEEPALIVE) { 660226854Spjd nv_free(nv); 661211882Spjd pjdlog_debug(2, 662211882Spjd "recv: (%p) Moving request to the free queue.", 663211882Spjd hio); 664226854Spjd hio_clear(hio); 665211882Spjd QUEUE_INSERT(free, hio); 666211882Spjd continue; 667211882Spjd } else if (hio->hio_cmd == HIO_WRITE) { 668226854Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 669229945Spjd hio->hio_data, MAXPHYS) == -1) { 670211984Spjd secondary_exit(EX_TEMPFAIL, 671212051Spjd "Unable to receive request data"); 672204076Spjd } 673204076Spjd } 674226854Spjd nv_free(nv); 675204076Spjd pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 676204076Spjd hio); 677211877Spjd QUEUE_INSERT(disk, hio); 678204076Spjd } 679204076Spjd /* NOTREACHED */ 680204076Spjd return (NULL); 681204076Spjd} 682204076Spjd 683204076Spjd/* 684204076Spjd * Thread reads from or writes to local component and also handles DELETE and 685204076Spjd * FLUSH requests. 686204076Spjd */ 687204076Spjdstatic void * 688204076Spjddisk_thread(void *arg) 689204076Spjd{ 690204076Spjd struct hast_resource *res = arg; 691204076Spjd struct hio *hio; 692204076Spjd ssize_t ret; 693225832Spjd bool clear_activemap, logerror; 694204076Spjd 695204076Spjd clear_activemap = true; 696204076Spjd 697204076Spjd for (;;) { 698204076Spjd pjdlog_debug(2, "disk: Taking request."); 699211877Spjd QUEUE_TAKE(disk, hio); 700204076Spjd while (clear_activemap) { 701204076Spjd unsigned char *map; 702204076Spjd size_t mapsize; 703204076Spjd 704204076Spjd /* 705204076Spjd * When first request is received, it means that primary 706204076Spjd * already received our activemap, merged it and stored 707204076Spjd * locally. We can now safely clear our activemap. 708204076Spjd */ 709204076Spjd mapsize = 710204076Spjd activemap_calc_ondisk_size(res->hr_local_mediasize - 711204076Spjd METADATA_SIZE, res->hr_extentsize, 712204076Spjd res->hr_local_sectorsize); 713204076Spjd map = calloc(1, mapsize); 714204076Spjd if (map == NULL) { 715204076Spjd pjdlog_warning("Unable to allocate memory to clear local activemap."); 716204076Spjd break; 717204076Spjd } 718204076Spjd if (pwrite(res->hr_localfd, map, mapsize, 719204076Spjd METADATA_SIZE) != (ssize_t)mapsize) { 720204076Spjd pjdlog_errno(LOG_WARNING, 721204076Spjd "Unable to store cleared activemap"); 722204076Spjd free(map); 723204076Spjd break; 724204076Spjd } 725204076Spjd free(map); 726204076Spjd clear_activemap = false; 727204076Spjd pjdlog_debug(1, "Local activemap cleared."); 728225831Spjd break; 729204076Spjd } 730204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 731225832Spjd logerror = true; 732204076Spjd /* Handle the actual request. */ 733204076Spjd switch (hio->hio_cmd) { 734204076Spjd case HIO_READ: 735204076Spjd ret = pread(res->hr_localfd, hio->hio_data, 736204076Spjd hio->hio_length, 737204076Spjd hio->hio_offset + res->hr_localoff); 738229945Spjd if (ret == -1) 739204076Spjd hio->hio_error = errno; 740204076Spjd else if (ret != (int64_t)hio->hio_length) 741204076Spjd hio->hio_error = EIO; 742204076Spjd else 743204076Spjd hio->hio_error = 0; 744204076Spjd break; 745204076Spjd case HIO_WRITE: 746204076Spjd ret = pwrite(res->hr_localfd, hio->hio_data, 747204076Spjd hio->hio_length, 748204076Spjd hio->hio_offset + res->hr_localoff); 749229945Spjd if (ret == -1) 750204076Spjd hio->hio_error = errno; 751204076Spjd else if (ret != (int64_t)hio->hio_length) 752204076Spjd hio->hio_error = EIO; 753204076Spjd else 754204076Spjd hio->hio_error = 0; 755204076Spjd break; 756204076Spjd case HIO_DELETE: 757204076Spjd ret = g_delete(res->hr_localfd, 758204076Spjd hio->hio_offset + res->hr_localoff, 759204076Spjd hio->hio_length); 760229945Spjd if (ret == -1) 761204076Spjd hio->hio_error = errno; 762204076Spjd else 763204076Spjd hio->hio_error = 0; 764204076Spjd break; 765204076Spjd case HIO_FLUSH: 766225832Spjd if (!res->hr_localflush) { 767225832Spjd ret = -1; 768225832Spjd hio->hio_error = EOPNOTSUPP; 769225832Spjd logerror = false; 770225832Spjd break; 771225832Spjd } 772204076Spjd ret = g_flush(res->hr_localfd); 773229945Spjd if (ret == -1) { 774225832Spjd if (errno == EOPNOTSUPP) 775225832Spjd res->hr_localflush = false; 776204076Spjd hio->hio_error = errno; 777225832Spjd } else { 778204076Spjd hio->hio_error = 0; 779225832Spjd } 780204076Spjd break; 781226854Spjd default: 782226854Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 783226854Spjd hio->hio_cmd); 784204076Spjd } 785225832Spjd if (logerror && hio->hio_error != 0) { 786204076Spjd reqlog(LOG_ERR, 0, hio->hio_error, hio, 787204076Spjd "Request failed: "); 788204076Spjd } 789204076Spjd pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 790204076Spjd hio); 791211877Spjd QUEUE_INSERT(send, hio); 792204076Spjd } 793204076Spjd /* NOTREACHED */ 794204076Spjd return (NULL); 795204076Spjd} 796204076Spjd 797204076Spjd/* 798204076Spjd * Thread sends requests back to primary node. 799204076Spjd */ 800204076Spjdstatic void * 801204076Spjdsend_thread(void *arg) 802204076Spjd{ 803204076Spjd struct hast_resource *res = arg; 804204076Spjd struct nv *nvout; 805204076Spjd struct hio *hio; 806204076Spjd void *data; 807204076Spjd size_t length; 808204076Spjd 809204076Spjd for (;;) { 810204076Spjd pjdlog_debug(2, "send: Taking request."); 811211877Spjd QUEUE_TAKE(send, hio); 812204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 813204076Spjd nvout = nv_alloc(); 814204076Spjd /* Copy sequence number. */ 815226854Spjd nv_add_uint64(nvout, hio->hio_seq, "seq"); 816204076Spjd switch (hio->hio_cmd) { 817204076Spjd case HIO_READ: 818204076Spjd if (hio->hio_error == 0) { 819204076Spjd data = hio->hio_data; 820204076Spjd length = hio->hio_length; 821204076Spjd break; 822204076Spjd } 823204076Spjd /* 824204076Spjd * We send no data in case of an error. 825204076Spjd */ 826204076Spjd /* FALLTHROUGH */ 827204076Spjd case HIO_DELETE: 828204076Spjd case HIO_FLUSH: 829204076Spjd case HIO_WRITE: 830204076Spjd data = NULL; 831204076Spjd length = 0; 832204076Spjd break; 833204076Spjd default: 834225782Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 835225782Spjd hio->hio_cmd); 836204076Spjd } 837204076Spjd if (hio->hio_error != 0) 838204076Spjd nv_add_int16(nvout, hio->hio_error, "error"); 839204076Spjd if (hast_proto_send(res, res->hr_remoteout, nvout, data, 840229945Spjd length) == -1) { 841211984Spjd secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 842204076Spjd } 843204076Spjd nv_free(nvout); 844209185Spjd pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 845204076Spjd hio); 846226854Spjd hio_clear(hio); 847211877Spjd QUEUE_INSERT(free, hio); 848204076Spjd } 849204076Spjd /* NOTREACHED */ 850204076Spjd return (NULL); 851204076Spjd} 852