secondary.c revision 226861
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 226861 2011-10-27 20:36:35Z 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 184204076Spjd if (metadata_read(res, true) < 0) 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 map = NULL; 203204076Spjd mapsize = 0; 204204076Spjd nvout = nv_alloc(); 205204076Spjd nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 206204076Spjd nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 207204076Spjd resuid = nv_get_uint64(nvin, "resuid"); 208204076Spjd res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 209204076Spjd res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 210204076Spjd nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 211204076Spjd nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 212204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 213204076Spjd METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 214204076Spjd map = malloc(mapsize); 215204076Spjd if (map == NULL) { 216204076Spjd pjdlog_exitx(EX_TEMPFAIL, 217204076Spjd "Unable to allocate memory (%zu bytes) for activemap.", 218204076Spjd mapsize); 219204076Spjd } 220204076Spjd /* 221204076Spjd * When we work as primary and secondary is missing we will increase 222204076Spjd * localcnt in our metadata. When secondary is connected and synced 223204076Spjd * we make localcnt be equal to remotecnt, which means nodes are more 224204076Spjd * or less in sync. 225204076Spjd * Split-brain condition is when both nodes are not able to communicate 226204076Spjd * and are both configured as primary nodes. In turn, they can both 227204076Spjd * make incompatible changes to the data and we have to detect that. 228204076Spjd * Under split-brain condition we will increase our localcnt on first 229204076Spjd * write and remote node will increase its localcnt on first write. 230204076Spjd * When we connect we can see that primary's localcnt is greater than 231204076Spjd * our remotecnt (primary was modified while we weren't watching) and 232204076Spjd * our localcnt is greater than primary's remotecnt (we were modified 233204076Spjd * while primary wasn't watching). 234204076Spjd * There are many possible combinations which are all gathered below. 235204076Spjd * Don't pay too much attention to exact numbers, the more important 236204076Spjd * is to compare them. We compare secondary's local with primary's 237204076Spjd * remote and secondary's remote with primary's local. 238204076Spjd * Note that every case where primary's localcnt is smaller than 239204076Spjd * secondary's remotecnt and where secondary's localcnt is smaller than 240204076Spjd * primary's remotecnt should be impossible in practise. We will perform 241204076Spjd * full synchronization then. Those cases are marked with an asterisk. 242204076Spjd * Regular synchronization means that only extents marked as dirty are 243204076Spjd * synchronized (regular synchronization). 244204076Spjd * 245204076Spjd * SECONDARY METADATA PRIMARY METADATA 246204076Spjd * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 247204076Spjd * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 248204076Spjd * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 249204076Spjd * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 250204076Spjd * regular sync from secondary. 251204076Spjd * local=3 remote=3 local=3 remote=3 Regular sync just in case. 252204076Spjd * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 253204076Spjd * local=3 remote=3 local=4 remote=2 Split-brain condition. 254204076Spjd * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 255204076Spjd * regular sync from primary. 256204076Spjd * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 257204076Spjd */ 258204076Spjd if (res->hr_resuid == 0) { 259204076Spjd /* 260214284Spjd * Provider is used for the first time. If primary node done no 261214284Spjd * writes yet as well (we will find "virgin" argument) then 262214284Spjd * there is no need to synchronize anything. If primary node 263214284Spjd * done any writes already we have to synchronize everything. 264204076Spjd */ 265218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt == 0); 266204076Spjd res->hr_resuid = resuid; 267204076Spjd if (metadata_write(res) < 0) 268204076Spjd exit(EX_NOINPUT); 269214284Spjd if (nv_exists(nvin, "virgin")) { 270214284Spjd free(map); 271214284Spjd map = NULL; 272214284Spjd mapsize = 0; 273214284Spjd } else { 274214284Spjd memset(map, 0xff, mapsize); 275214284Spjd } 276220865Spjd nv_add_int8(nvout, 1, "virgin"); 277204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 278219830Spjd } else if (res->hr_resuid != resuid) { 279219830Spjd char errmsg[256]; 280219830Spjd 281226854Spjd free(map); 282219830Spjd (void)snprintf(errmsg, sizeof(errmsg), 283219830Spjd "Resource unique ID mismatch (primary=%ju, secondary=%ju).", 284219830Spjd (uintmax_t)resuid, (uintmax_t)res->hr_resuid); 285219830Spjd pjdlog_error("%s", errmsg); 286219830Spjd nv_add_string(nvout, errmsg, "errmsg"); 287219830Spjd if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) { 288219830Spjd pjdlog_exit(EX_TEMPFAIL, "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"); 332226854Spjd if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) { 333226854Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s", 334226854Spjd res->hr_remoteaddr); 335226854Spjd } 336226854Spjd nv_free(nvout); 337226854Spjd /* Exit on split-brain. */ 338226854Spjd event_send(res, EVENT_SPLITBRAIN); 339226854Spjd exit(EX_CONFIG); 340204076Spjd } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 341204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 342204076Spjd /* 343204076Spjd * This should never happen in practise, but we will perform 344204076Spjd * full synchronization. 345204076Spjd */ 346218138Spjd PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 347204076Spjd res->hr_primary_localcnt < res->hr_secondary_remotecnt); 348204076Spjd mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 349204076Spjd METADATA_SIZE, res->hr_extentsize, 350204076Spjd res->hr_local_sectorsize); 351204076Spjd memset(map, 0xff, mapsize); 352204076Spjd if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 353204076Spjd /* In this one of five cases sync from secondary. */ 354204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 355204076Spjd } else { 356204076Spjd /* For the rest four cases sync from primary. */ 357204076Spjd nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 358204076Spjd } 359204076Spjd pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 360204076Spjd (uintmax_t)res->hr_primary_localcnt, 361204076Spjd (uintmax_t)res->hr_primary_remotecnt, 362204076Spjd (uintmax_t)res->hr_secondary_localcnt, 363204076Spjd (uintmax_t)res->hr_secondary_remotecnt); 364204076Spjd } 365220007Spjd nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 366204076Spjd if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 367214276Spjd pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 368204076Spjd res->hr_remoteaddr); 369204076Spjd } 370214275Spjd if (map != NULL) 371214275Spjd free(map); 372209182Spjd nv_free(nvout); 373223181Strociny#ifdef notyet 374220271Spjd /* Setup direction. */ 375220271Spjd if (proto_recv(res->hr_remotein, NULL, 0) == -1) 376220271Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection direction"); 377223181Strociny#endif 378204076Spjd} 379204076Spjd 380204076Spjdvoid 381204076Spjdhastd_secondary(struct hast_resource *res, struct nv *nvin) 382204076Spjd{ 383213009Spjd sigset_t mask; 384204076Spjd pthread_t td; 385204076Spjd pid_t pid; 386219482Strociny int error, mode, debuglevel; 387204076Spjd 388204076Spjd /* 389204076Spjd * Create communication channel between parent and child. 390204076Spjd */ 391219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_ctrl) < 0) { 392204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 393204076Spjd pjdlog_exit(EX_OSERR, 394204076Spjd "Unable to create control sockets between parent and child"); 395204076Spjd } 396212038Spjd /* 397212038Spjd * Create communication channel between child and parent. 398212038Spjd */ 399219818Spjd if (proto_client(NULL, "socketpair://", &res->hr_event) < 0) { 400212038Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 401212038Spjd pjdlog_exit(EX_OSERR, 402212038Spjd "Unable to create event sockets between child and parent"); 403212038Spjd } 404204076Spjd 405204076Spjd pid = fork(); 406204076Spjd if (pid < 0) { 407204076Spjd KEEP_ERRNO((void)pidfile_remove(pfh)); 408204076Spjd pjdlog_exit(EX_OSERR, "Unable to fork"); 409204076Spjd } 410204076Spjd 411204076Spjd if (pid > 0) { 412204076Spjd /* This is parent. */ 413204076Spjd proto_close(res->hr_remotein); 414204076Spjd res->hr_remotein = NULL; 415204076Spjd proto_close(res->hr_remoteout); 416204076Spjd res->hr_remoteout = NULL; 417212038Spjd /* Declare that we are receiver. */ 418212038Spjd proto_recv(res->hr_event, NULL, 0); 419218043Spjd /* Declare that we are sender. */ 420218043Spjd proto_send(res->hr_ctrl, NULL, 0); 421204076Spjd res->hr_workerpid = pid; 422204076Spjd return; 423204076Spjd } 424211977Spjd 425211984Spjd gres = res; 426218043Spjd mode = pjdlog_mode_get(); 427219482Strociny debuglevel = pjdlog_debug_get(); 428211984Spjd 429218043Spjd /* Declare that we are sender. */ 430218043Spjd proto_send(res->hr_event, NULL, 0); 431218043Spjd /* Declare that we are receiver. */ 432218043Spjd proto_recv(res->hr_ctrl, NULL, 0); 433218043Spjd descriptors_cleanup(res); 434204076Spjd 435218045Spjd descriptors_assert(res, mode); 436218045Spjd 437218043Spjd pjdlog_init(mode); 438219482Strociny pjdlog_debug_set(debuglevel); 439218043Spjd pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 440220005Spjd setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role)); 441204076Spjd 442213009Spjd PJDLOG_VERIFY(sigemptyset(&mask) == 0); 443213009Spjd PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 444210880Spjd 445207371Spjd /* Error in setting timeout is not critical, but why should it fail? */ 446219721Strociny if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0) 447207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 448207371Spjd if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 449207371Spjd pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 450207371Spjd 451204076Spjd init_local(res); 452213007Spjd init_environment(); 453213007Spjd 454221899Spjd if (drop_privs(res) != 0) 455218049Spjd exit(EX_CONFIG); 456218214Spjd pjdlog_info("Privileges successfully dropped."); 457218049Spjd 458213007Spjd /* 459213007Spjd * Create the control thread before sending any event to the parent, 460213007Spjd * as we can deadlock when parent sends control request to worker, 461213007Spjd * but worker has no control thread started yet, so parent waits. 462213007Spjd * In the meantime worker sends an event to the parent, but parent 463213007Spjd * is unable to handle the event, because it waits for control 464213007Spjd * request response. 465213007Spjd */ 466213007Spjd error = pthread_create(&td, NULL, ctrl_thread, res); 467218138Spjd PJDLOG_ASSERT(error == 0); 468213007Spjd 469204076Spjd init_remote(res, nvin); 470212038Spjd event_send(res, EVENT_CONNECT); 471204076Spjd 472204076Spjd error = pthread_create(&td, NULL, recv_thread, res); 473218138Spjd PJDLOG_ASSERT(error == 0); 474204076Spjd error = pthread_create(&td, NULL, disk_thread, res); 475218138Spjd PJDLOG_ASSERT(error == 0); 476213007Spjd (void)send_thread(res); 477204076Spjd} 478204076Spjd 479204076Spjdstatic void 480204076Spjdreqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 481204076Spjd{ 482204076Spjd char msg[1024]; 483204076Spjd va_list ap; 484204076Spjd int len; 485204076Spjd 486204076Spjd va_start(ap, fmt); 487204076Spjd len = vsnprintf(msg, sizeof(msg), fmt, ap); 488204076Spjd va_end(ap); 489204076Spjd if ((size_t)len < sizeof(msg)) { 490204076Spjd switch (hio->hio_cmd) { 491204076Spjd case HIO_READ: 492204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 493204076Spjd "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 494204076Spjd (uintmax_t)hio->hio_length); 495204076Spjd break; 496204076Spjd case HIO_DELETE: 497204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 498204076Spjd "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 499204076Spjd (uintmax_t)hio->hio_length); 500204076Spjd break; 501204076Spjd case HIO_FLUSH: 502204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 503204076Spjd break; 504204076Spjd case HIO_WRITE: 505204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 506204076Spjd "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 507204076Spjd (uintmax_t)hio->hio_length); 508204076Spjd break; 509211882Spjd case HIO_KEEPALIVE: 510211882Spjd (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 511211882Spjd break; 512204076Spjd default: 513204076Spjd (void)snprintf(msg + len, sizeof(msg) - len, 514204076Spjd "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 515204076Spjd break; 516204076Spjd } 517204076Spjd } 518204076Spjd pjdlog_common(loglevel, debuglevel, error, "%s", msg); 519204076Spjd} 520204076Spjd 521204076Spjdstatic int 522226854Spjdrequnpack(struct hast_resource *res, struct hio *hio, struct nv *nv) 523204076Spjd{ 524204076Spjd 525226854Spjd hio->hio_cmd = nv_get_uint8(nv, "cmd"); 526204076Spjd if (hio->hio_cmd == 0) { 527204076Spjd pjdlog_error("Header contains no 'cmd' field."); 528204076Spjd hio->hio_error = EINVAL; 529204076Spjd goto end; 530204076Spjd } 531226854Spjd if (hio->hio_cmd != HIO_KEEPALIVE) { 532226854Spjd hio->hio_seq = nv_get_uint64(nv, "seq"); 533226854Spjd if (hio->hio_seq == 0) { 534226854Spjd pjdlog_error("Header contains no 'seq' field."); 535226854Spjd hio->hio_error = EINVAL; 536226854Spjd goto end; 537226854Spjd } 538226854Spjd } 539204076Spjd switch (hio->hio_cmd) { 540222164Spjd case HIO_FLUSH: 541211882Spjd case HIO_KEEPALIVE: 542211882Spjd break; 543204076Spjd case HIO_READ: 544204076Spjd case HIO_WRITE: 545204076Spjd case HIO_DELETE: 546226854Spjd hio->hio_offset = nv_get_uint64(nv, "offset"); 547226854Spjd if (nv_error(nv) != 0) { 548204076Spjd pjdlog_error("Header is missing 'offset' field."); 549204076Spjd hio->hio_error = EINVAL; 550204076Spjd goto end; 551204076Spjd } 552226854Spjd hio->hio_length = nv_get_uint64(nv, "length"); 553226854Spjd if (nv_error(nv) != 0) { 554204076Spjd pjdlog_error("Header is missing 'length' field."); 555204076Spjd hio->hio_error = EINVAL; 556204076Spjd goto end; 557204076Spjd } 558204076Spjd if (hio->hio_length == 0) { 559204076Spjd pjdlog_error("Data length is zero."); 560204076Spjd hio->hio_error = EINVAL; 561204076Spjd goto end; 562204076Spjd } 563204076Spjd if (hio->hio_length > MAXPHYS) { 564204076Spjd pjdlog_error("Data length is too large (%ju > %ju).", 565204076Spjd (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 566204076Spjd hio->hio_error = EINVAL; 567204076Spjd goto end; 568204076Spjd } 569204076Spjd if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 570204076Spjd pjdlog_error("Offset %ju is not multiple of sector size.", 571204076Spjd (uintmax_t)hio->hio_offset); 572204076Spjd hio->hio_error = EINVAL; 573204076Spjd goto end; 574204076Spjd } 575204076Spjd if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 576204076Spjd pjdlog_error("Length %ju is not multiple of sector size.", 577204076Spjd (uintmax_t)hio->hio_length); 578204076Spjd hio->hio_error = EINVAL; 579204076Spjd goto end; 580204076Spjd } 581204076Spjd if (hio->hio_offset + hio->hio_length > 582204076Spjd (uint64_t)res->hr_datasize) { 583204076Spjd pjdlog_error("Data offset is too large (%ju > %ju).", 584204076Spjd (uintmax_t)(hio->hio_offset + hio->hio_length), 585204076Spjd (uintmax_t)res->hr_datasize); 586204076Spjd hio->hio_error = EINVAL; 587204076Spjd goto end; 588204076Spjd } 589204076Spjd break; 590204076Spjd default: 591204076Spjd pjdlog_error("Header contains invalid 'cmd' (%hhu).", 592204076Spjd hio->hio_cmd); 593204076Spjd hio->hio_error = EINVAL; 594204076Spjd goto end; 595204076Spjd } 596204076Spjd hio->hio_error = 0; 597204076Spjdend: 598204076Spjd return (hio->hio_error); 599204076Spjd} 600204076Spjd 601212899Spjdstatic __dead2 void 602211984Spjdsecondary_exit(int exitcode, const char *fmt, ...) 603211984Spjd{ 604211984Spjd va_list ap; 605211984Spjd 606218138Spjd PJDLOG_ASSERT(exitcode != EX_OK); 607211984Spjd va_start(ap, fmt); 608211984Spjd pjdlogv_errno(LOG_ERR, fmt, ap); 609211984Spjd va_end(ap); 610212038Spjd event_send(gres, EVENT_DISCONNECT); 611211984Spjd exit(exitcode); 612211984Spjd} 613211984Spjd 614204076Spjd/* 615204076Spjd * Thread receives requests from the primary node. 616204076Spjd */ 617204076Spjdstatic void * 618204076Spjdrecv_thread(void *arg) 619204076Spjd{ 620204076Spjd struct hast_resource *res = arg; 621204076Spjd struct hio *hio; 622226854Spjd struct nv *nv; 623204076Spjd 624204076Spjd for (;;) { 625204076Spjd pjdlog_debug(2, "recv: Taking free request."); 626211877Spjd QUEUE_TAKE(free, hio); 627204076Spjd pjdlog_debug(2, "recv: (%p) Got request.", hio); 628226854Spjd if (hast_proto_recv_hdr(res->hr_remotein, &nv) < 0) { 629211984Spjd secondary_exit(EX_TEMPFAIL, 630204076Spjd "Unable to receive request header"); 631204076Spjd } 632226854Spjd if (requnpack(res, hio, nv) != 0) { 633226854Spjd nv_free(nv); 634211877Spjd pjdlog_debug(2, 635211877Spjd "recv: (%p) Moving request to the send queue.", 636211877Spjd hio); 637211877Spjd QUEUE_INSERT(send, hio); 638211877Spjd continue; 639211877Spjd } 640222228Spjd switch (hio->hio_cmd) { 641222228Spjd case HIO_READ: 642222228Spjd res->hr_stat_read++; 643222228Spjd break; 644222228Spjd case HIO_WRITE: 645222228Spjd res->hr_stat_write++; 646222228Spjd break; 647222228Spjd case HIO_DELETE: 648222228Spjd res->hr_stat_delete++; 649222228Spjd break; 650222228Spjd case HIO_FLUSH: 651222228Spjd res->hr_stat_flush++; 652222228Spjd break; 653226854Spjd case HIO_KEEPALIVE: 654226854Spjd break; 655226854Spjd default: 656226854Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 657226854Spjd hio->hio_cmd); 658222228Spjd } 659204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, 660204076Spjd "recv: (%p) Got request header: ", hio); 661211882Spjd if (hio->hio_cmd == HIO_KEEPALIVE) { 662226854Spjd nv_free(nv); 663211882Spjd pjdlog_debug(2, 664211882Spjd "recv: (%p) Moving request to the free queue.", 665211882Spjd hio); 666226854Spjd hio_clear(hio); 667211882Spjd QUEUE_INSERT(free, hio); 668211882Spjd continue; 669211882Spjd } else if (hio->hio_cmd == HIO_WRITE) { 670226854Spjd if (hast_proto_recv_data(res, res->hr_remotein, nv, 671226854Spjd hio->hio_data, MAXPHYS) < 0) { 672211984Spjd secondary_exit(EX_TEMPFAIL, 673212051Spjd "Unable to receive request data"); 674204076Spjd } 675204076Spjd } 676226854Spjd nv_free(nv); 677204076Spjd pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 678204076Spjd hio); 679211877Spjd QUEUE_INSERT(disk, hio); 680204076Spjd } 681204076Spjd /* NOTREACHED */ 682204076Spjd return (NULL); 683204076Spjd} 684204076Spjd 685204076Spjd/* 686204076Spjd * Thread reads from or writes to local component and also handles DELETE and 687204076Spjd * FLUSH requests. 688204076Spjd */ 689204076Spjdstatic void * 690204076Spjddisk_thread(void *arg) 691204076Spjd{ 692204076Spjd struct hast_resource *res = arg; 693204076Spjd struct hio *hio; 694204076Spjd ssize_t ret; 695225832Spjd bool clear_activemap, logerror; 696204076Spjd 697204076Spjd clear_activemap = true; 698204076Spjd 699204076Spjd for (;;) { 700204076Spjd pjdlog_debug(2, "disk: Taking request."); 701211877Spjd QUEUE_TAKE(disk, hio); 702204076Spjd while (clear_activemap) { 703204076Spjd unsigned char *map; 704204076Spjd size_t mapsize; 705204076Spjd 706204076Spjd /* 707204076Spjd * When first request is received, it means that primary 708204076Spjd * already received our activemap, merged it and stored 709204076Spjd * locally. We can now safely clear our activemap. 710204076Spjd */ 711204076Spjd mapsize = 712204076Spjd activemap_calc_ondisk_size(res->hr_local_mediasize - 713204076Spjd METADATA_SIZE, res->hr_extentsize, 714204076Spjd res->hr_local_sectorsize); 715204076Spjd map = calloc(1, mapsize); 716204076Spjd if (map == NULL) { 717204076Spjd pjdlog_warning("Unable to allocate memory to clear local activemap."); 718204076Spjd break; 719204076Spjd } 720204076Spjd if (pwrite(res->hr_localfd, map, mapsize, 721204076Spjd METADATA_SIZE) != (ssize_t)mapsize) { 722204076Spjd pjdlog_errno(LOG_WARNING, 723204076Spjd "Unable to store cleared activemap"); 724204076Spjd free(map); 725204076Spjd break; 726204076Spjd } 727204076Spjd free(map); 728204076Spjd clear_activemap = false; 729204076Spjd pjdlog_debug(1, "Local activemap cleared."); 730225831Spjd break; 731204076Spjd } 732204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 733225832Spjd logerror = true; 734204076Spjd /* Handle the actual request. */ 735204076Spjd switch (hio->hio_cmd) { 736204076Spjd case HIO_READ: 737204076Spjd ret = pread(res->hr_localfd, hio->hio_data, 738204076Spjd hio->hio_length, 739204076Spjd hio->hio_offset + res->hr_localoff); 740204076Spjd if (ret < 0) 741204076Spjd hio->hio_error = errno; 742204076Spjd else if (ret != (int64_t)hio->hio_length) 743204076Spjd hio->hio_error = EIO; 744204076Spjd else 745204076Spjd hio->hio_error = 0; 746204076Spjd break; 747204076Spjd case HIO_WRITE: 748204076Spjd ret = pwrite(res->hr_localfd, hio->hio_data, 749204076Spjd hio->hio_length, 750204076Spjd hio->hio_offset + res->hr_localoff); 751204076Spjd if (ret < 0) 752204076Spjd hio->hio_error = errno; 753204076Spjd else if (ret != (int64_t)hio->hio_length) 754204076Spjd hio->hio_error = EIO; 755204076Spjd else 756204076Spjd hio->hio_error = 0; 757204076Spjd break; 758204076Spjd case HIO_DELETE: 759204076Spjd ret = g_delete(res->hr_localfd, 760204076Spjd hio->hio_offset + res->hr_localoff, 761204076Spjd hio->hio_length); 762204076Spjd if (ret < 0) 763204076Spjd hio->hio_error = errno; 764204076Spjd else 765204076Spjd hio->hio_error = 0; 766204076Spjd break; 767204076Spjd case HIO_FLUSH: 768225832Spjd if (!res->hr_localflush) { 769225832Spjd ret = -1; 770225832Spjd hio->hio_error = EOPNOTSUPP; 771225832Spjd logerror = false; 772225832Spjd break; 773225832Spjd } 774204076Spjd ret = g_flush(res->hr_localfd); 775225832Spjd if (ret < 0) { 776225832Spjd if (errno == EOPNOTSUPP) 777225832Spjd res->hr_localflush = false; 778204076Spjd hio->hio_error = errno; 779225832Spjd } else { 780204076Spjd hio->hio_error = 0; 781225832Spjd } 782204076Spjd break; 783226854Spjd default: 784226854Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 785226854Spjd hio->hio_cmd); 786204076Spjd } 787225832Spjd if (logerror && hio->hio_error != 0) { 788204076Spjd reqlog(LOG_ERR, 0, hio->hio_error, hio, 789204076Spjd "Request failed: "); 790204076Spjd } 791204076Spjd pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 792204076Spjd hio); 793211877Spjd QUEUE_INSERT(send, hio); 794204076Spjd } 795204076Spjd /* NOTREACHED */ 796204076Spjd return (NULL); 797204076Spjd} 798204076Spjd 799204076Spjd/* 800204076Spjd * Thread sends requests back to primary node. 801204076Spjd */ 802204076Spjdstatic void * 803204076Spjdsend_thread(void *arg) 804204076Spjd{ 805204076Spjd struct hast_resource *res = arg; 806204076Spjd struct nv *nvout; 807204076Spjd struct hio *hio; 808204076Spjd void *data; 809204076Spjd size_t length; 810204076Spjd 811204076Spjd for (;;) { 812204076Spjd pjdlog_debug(2, "send: Taking request."); 813211877Spjd QUEUE_TAKE(send, hio); 814204076Spjd reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 815204076Spjd nvout = nv_alloc(); 816204076Spjd /* Copy sequence number. */ 817226854Spjd nv_add_uint64(nvout, hio->hio_seq, "seq"); 818204076Spjd switch (hio->hio_cmd) { 819204076Spjd case HIO_READ: 820204076Spjd if (hio->hio_error == 0) { 821204076Spjd data = hio->hio_data; 822204076Spjd length = hio->hio_length; 823204076Spjd break; 824204076Spjd } 825204076Spjd /* 826204076Spjd * We send no data in case of an error. 827204076Spjd */ 828204076Spjd /* FALLTHROUGH */ 829204076Spjd case HIO_DELETE: 830204076Spjd case HIO_FLUSH: 831204076Spjd case HIO_WRITE: 832204076Spjd data = NULL; 833204076Spjd length = 0; 834204076Spjd break; 835204076Spjd default: 836225782Spjd PJDLOG_ABORT("Unexpected command (cmd=%hhu).", 837225782Spjd hio->hio_cmd); 838204076Spjd } 839204076Spjd if (hio->hio_error != 0) 840204076Spjd nv_add_int16(nvout, hio->hio_error, "error"); 841204076Spjd if (hast_proto_send(res, res->hr_remoteout, nvout, data, 842204076Spjd length) < 0) { 843211984Spjd secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 844204076Spjd } 845204076Spjd nv_free(nvout); 846209185Spjd pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 847204076Spjd hio); 848226854Spjd hio_clear(hio); 849211877Spjd QUEUE_INSERT(free, hio); 850204076Spjd } 851204076Spjd /* NOTREACHED */ 852204076Spjd return (NULL); 853204076Spjd} 854